I WASAWA T HEORY

Romyar Sharifi
 Contents

Introduction 5

Chapter 1. Class groups and units 9

1.1. Notation and background 9
1.2. Regulators 11
1.3. Finite Galois extensions 13
1.4. Kummer theory 22
1.5. Leopoldt’s conjecture 26

Chapter 2. Module theory 35

2.1. Pseudo-isomorphisms 35
2.2. Power series rings 41
2.3. Completed group rings 44
2.4. Invariants of Λ-modules 48
2.5. Pontryagin duality 56
2.6. Iwasawa adjoints 58
2.7. The group ring of a cyclic p-group 63
2.8. Eigenspaces 65

Chapter 3. Iwasawa theory 71

3.1. Z p -extensions 71
3.2. Limits of class groups 73
3.3. The p-ramified Iwasawa module 78
3.4. CM fields 84
3.5. Kida’s formula 87

Chapter 4. Cyclotomic fields 93

4.1. Dirichlet L-functions 93
4.2. Bernoulli numbers 97
4.3. Cyclotomic units 103
4.4. Reflection theorems 105
3
4 CONTENTS

4.5. Stickelberger theory 108

4.6. Distributions 111
4.7. Sinnott’s theorem 114

Chapter 5. Kubota-Leopoldt p-adic L-functions 121

5.1. p-adic measures 121
5.2. p-adic L-functions 124
5.3. Iwasawa power series 129
5.4. Coleman theory 133

Chapter 6. The Iwasawa main conjecture 147

6.1. Semi-local units modulo cyclotomic units 147
6.2. The Ferrero-Washington theorem 150
6.3. The main conjecture over Q 154
6.4. The Euler system of cyclotomic units 156
6.5. The main conjecture via Euler systems 163
6.6. Geometry of modular curves 167

Appendix A. Duality in Galois cohomology 173

Bibliography 177
 Introduction

The class group ClF of a number field F is an object of central importance in number theory.
It is a finite abelian group, and its order hF is known as the class number. In general, the explicit
determination of hF , let alone the structure of ClF as a finite abelian group, can be a difficult and
computationally intensive task.
In the late 1950’s, Iwasawa initiated a study of the growth of class groups in certain towers of
number fields. Given a tower F = F0 ⊂ F1 ⊂ F2 ⊂ · · · of Galois extensions of F, one asks if there
is any regularity to the growth of hFn . The knowledge of this growth, in turn, can be used to say
something about the structure of ClFn as a finite abelian group. Iwasawa was concerned with towers
∼ Z p for some prime p, where F∞ = S Fn , known as Z p -extensions. He set
such that Gal(F∞ /F) = n
Γ = Gal(F∞ /F) and Γn = Gal(Fn /F), and let us suppose that Fn is chosen to be (cyclic) of degree pn
over F. For example, for odd p, the cyclotomic Z p -extension F∞ of F is the largest subextension of
F(µ p∞ )/F with pro-p Galois group.
The question of how hFn grows in the tower defined by a Z p -extension is quite difficult, in particular
as the order away from p of ClFn has little to do with the order away from p of ClFn+1 , other than the
fact that the latter order is a multiple of the former. On the other hand, if we concentrate on the order
(p)
hFn of the Sylow p-sugroup An of Fn , we have the following theorem of Iwasawa.

T HEOREM (Iwasawa). There exist nonnegative integers λ and µ and an integer ν such that

(p) n µ+ν
hFn = pnλ +p

for all sufficiently large n.

In the case that F∞ is the cyclotomic Z p -extension, Iwasawa conjectured that the invariant µ in the
theorem is 0. Ferrero and Washington later proved this result for abelian extensions of Q.
We have maps between the p-parts of class group in the tower in both directions jn : An → An+1 ,
which takes the class of an ideal a to the class of the ideal it generates, and Nn : An+1 → An , which
takes the class of an ideal to the class of its norm. Iwasawa considered the direct and inverse limits

A∞ = lim An and X∞ = lim An

−→ ←−
n n
5
6 INTRODUCTION

under the jn and Nn , respectively. As each An has the structure of a finite Z p [Γn ]-module through the
standard action of Γn on ideal classes, both X∞ and the Pontryagin dual A∨
∞ = Homcts (A∞ , Q p /Z p ) of
A∞ are finitely generated torsion modules over the competed Z p -group ring of Γ:

Z p JΓK = lim Z p [Γn ].

←−
n

The ring Λ = Z p JΓK is known as the Iwasawa algebra, and it has a very simple structure. In fact, a
choice of a topological generator γ of Γ gives rise to an isomorphism
∼
Z p JT K −
→ Λ, T 7→ γ − 1.

The following result on the structure of Λ-modules allowed Serre to rephrase the theorem of Iwa-
sawa.

T HEOREM (Serre). For any finitely generated torsion Λ-module M, there exists a homomorphism
of Λ-modules
s t
Λ/p` j Λ,
M M
M→ Λ/ fi (T )ki Λ ⊕
i=1 j=1
with finite kernel and cokernel, for some nonnegative integers s and t, irreducible fi (T ) ∈ Z p [T ] with
fi (T ) ≡ T deg fi mod p, and positive integers ki and ` j .

From Serre’s theorem, we are able to deduce several important invariants of a finitely generated
Λ-module M. For instance, in the notation of the theorem, let us set
s t
λ (M) = ∑ ki deg fi and µ(M) = ∑ ` j.
i=1 j=1

These are known as the λ and µ-invariants of M. Serre showed that these invariants for X∞ and A∨
∞ agree
with the λ and µ of Iwasawa’s theorem. An even more interesting invariant of M is its characteristic
ideal, given by !
s
charΛ M = pµ(M) ∏ fi (T )ki Λ,
i=1
which we shall consider in a specific case shortly.
It is worth remarking here that one usually thinks of X∞ as a Galois group. Recall that the Artin
reciprocity map provides an isomorphism between An and the Galois group of the Hilbert p-class field
S
Ln of Fn , which is to say the maximal unramified abelian p-extension of Fn . Setting L∞ = n Ln ,
we have a canonical isomorphism X∞ ∼= Gal(L∞ /F∞ ). The resulting action on Γ on Gal(L∞ /F∞ ) is a
conjugation action, given by a lift of Γ to a subsgroup of Gal(L∞ /F).
Let us focus now on the specific case that F = Q(µ p ), and let us take F∞ to be the cyclotomic
Z p -extension of F for an odd prime p. In this setting, Iwasawa proved that his µ = µ(X∞ ) is zero.
 INTRODUCTION 7

We define the Teichmüller character ω : ∆ → Z×

p by setting ω(δ ) for δ ∈ ∆ to be the unique
(p − 1)st root of unity in Z p such that
ω(δ )
δ (ζ p ) = ζ p
for any primitive pth root of unity ζ p .
As with Γ, the Galois group ∆ = Gal(F/Q) will act on X∞ . For any i, we may consider the
(i)
eigenspace X∞ of X∞ on which every δ ∈ ∆ acts through multiplication by ω i (δ ). We have the fol-
lowing theorem of Herbrand and Ribet.
(1−k)
T HEOREM (Herbrand-Ribet). Let k be an even with 2 ≤ k ≤ p − 3. Then X∞ 6= 0 if and only if
p divides the Bernoulli number Bk .

The interesting fact is that Bernoulli numbers and their generalizations appear as values of L-
functions. Kubota and Leopoldt showed how that the L-values of certain characters at negative integers
can be interpolated, in essence, by a function of Z p , denoted L p (χ, s) and known as a p-adic L-function.
Let us fix the particular generator γ of Γ such that γ(ζ ) = ζ 1+p for every p-power root of unity ζ ,
and in particular the isomorphism of Λ with Z p JT K. Iwasawa made the following conjecture on the
characteristic ideal of an eigenspace of X, which was later proven by Mazur and Wiles.

T HEOREM (Main conjecture of Iwasawa theory, Mazur-Wiles). Let k be an even integer. Then

charΛ X∞(1−k) = ( fk ),

where fk ((1 + p)s − 1) = L p (ω k , s) for all s ∈ Z p .

In fact, Mazur and Wiles proved a generalization of this to abelian extensions F of Q, and Wiles
proved a further generalization to abelian extensions of totally real fields. This line of proof was
primarily geometric in nature, and came by studying the action of the absolute Galois group of F
on the cohomology groups of modular curves. Rubin gave a proof of a rather different nature of a
main conjecture for abelian extensions of imaginary quadratic fields, following work of Kolyvagin and
Thaine, using a Galois cohomological tool known as an Euler system.
Let us end this introduction by mentioning the two of the major directions in which Iwasawa theory
has expanded over the years. As a first and obvious course of action, one can replace our limits of class
groups with more general objects. Via class field theory, we note that the Pontryagin dual X∞∨ may be
identified with the kernel of the map
 
ker H 1 (GF∞ ,S , Q p /Z p ) → H 1 (Iv , Q p /Z p ) ,
M

v∈S
where GF∞ ,S denotes the Galois group of the maximal extension of F∞ unramified outside S and S in
this case is the set of primes of F∞ lying over p, and where Iv is the inertia group at v ∈ S in the absolute
8 INTRODUCTION

Galois group of F∞ . That is, we have realized X∞∨ as what is known as a Selmer group. This generalizes
nicely.
By way of the most interesting example, let E be an elliptic curve over F with ordinary reduction
at p, and let E[p∞ ] denote its p-power torsion (over Q). The Selmer group of E over F∞ is exactly
 
Sel(E/F∞ ) = ker H 1 (GF∞ ,S , E[p∞ ]) → H 1 (Iv , E[p∞ ]) ,
M

v∈S
where S is now the set of primes of F∞ over p or any primes of bad reduction of E. In the case that
F = Q, there is a corresponding main conjecture for the structure of Sel(E/F∞ )∨ in terms of a p-adic
L-function of E. Great progress has been made on this particular main conjecture, due to successively
more recent work of Rubin (for CM curves), Kato, and Skinner and Urban.
In the second generalization, one allows the Galois group Γ of the tower to take a more general
form than Z p . The case that Γ is a p-adic Lie group, which is to say isomorphic to an open subgroup
of GLm (Z p ) for some m ≥ 1, has come under the greatest consideration. In this case, main conjectures
become more difficult to formulate, as the structure theory of Λ = Z p JΓK-modules is no longer simple.
Still, in the past decade, such main conjectures have been formulated using K-theory as one of several
tools. In the classical setting of limits of class groups, the corresponding main conjecture has been
proven by Kakde and Ritter-Weiss.1

1The reader should be aware that parts of these notes were hurriedly written and, at present, have not been proofread

or checked. Also, references and attributions of results have not been properly made. We hope to rectify these issues in
later versions. Comments pointing out errors are welcome.
 CHAPTER 1

Class groups and units

1.1. Notation and background

Throughout, we will let F be a number field. We recall a number of objects attached to F and finite
Galois extensions thereof and results regarding them.

N OTATION 1.1.1. To a number field, we attach the following objects:

• the ring of integers OF of F,
• the unit group OF× of F, which is to say the unit group of OF ,
• the ideal group IF of F, i.e., the group of nonzero finitely generated OF -submodules of F,
• the prinicipal ideal group PF of F, i.e., those OF -submodules (α) of F generated by a single
element α ∈ F × , and
• the class group ClF = IF /PF of F.

R EMARK 1.1.2. The class group ClF is a finite abelian group.

These objects fit into the following nice commutative diagram

PF l
B


1 / O× / F× / IF / ClF /0
F

in which the lower row is exact.

D EFINITION 1.1.3. The absolute norm Na of a nonzero ideal a of OF is the index Na = [OF : a].

N OTATION 1.1.4. The number of real places of F is denoted r1 (F), and the number of complex
places of F is denoted r2 (F).

R EMARK 1.1.5. Since each complex places consists of a pair of complex conjugate embeddings of
F, the degree formula tells us that r1 (F) + 2r2 (F) = [F : Q].

T HEOREM 1.1.6 (Dirichlet’s unit theorem). The unit group OF× is a finitely generated abelian group
of rank r1 (F) + r2 (F) − 1 with torsion subgroup the group µ(F) of roots of unity in F.

Next, we turn quickly to the zeta function of a number field.

9
10 1. CLASS GROUPS AND UNITS

D EFINITION 1.1.7. The Dedekind ζ -series ζF of a number field F is

1
ζF (s) = ∑ ,
a⊆OF (Na)s

for s ∈ C with real part Re s > 1, where the sum is taken over nonzero ideals of OF .

T HEOREM 1.1.8. The Dedekind ζ -series of F converges absolutely on s with Re s > 1. It has a
unique mermomorphic continuation to C which is holomorphic outside 1 and has a simple pole at
s = 1.

With this in hand, we define the Dedekind zeta function to be the meromorphic continuation of ζF
to C.

D EFINITION 1.1.9. The Dedekind zeta function ζF of a number field F is the meromorphic contin-
uation to C of the Dedekind ζ -series ζF .

The Dedekind zeta function has the following functional equation relating its values at s and 1 − s.

T HEOREM 1.1.10. Let

ΛF (s) = (2−r2 (F) π −[F:Q] |dF |1/2 )s Γ(s/2)r1 (F) Γ(s)r2 (F) ζF (s),

where Γ is the gamma function and dF denotes the discriminant of F. Then ΛF (s) is analytic on C and
satisfies
ΛF (s) = ΛF (1 − s).

For any Galois extension E/F and prime p of F, let ϕP denote a Frobenius at a prime P over p.
We have
ϕP (α) ≡ α Np mod P
for α ∈ OE . If E/F is unramified, its conjugacy class in Gal(E/F) depends only on p, and let us denote
it by [ϕp ]. If E/F is abelian and unramified, we denote the unique Frobenius more simply by ϕp .

T HEOREM 1.1.11 (Čebotarev density theorem). Let E/F be a finite Galois extension of number
fields with group G. Let X be the set of unramified primes in E/F. Let C be a conjugacy class in G.
Then
{p ∈ X | [ϕp ] ∈ C and Np ≤ N} |C|
lim = .
N→∞ {p ∈ X | Np ≤ N} |G|
We will denote the class of a fractional ideal a ∈ IF by [a] ∈ ClF . The class group has an other
description in terms of the Hilbert class field HF of F, which is to say the maximal unramified abelian
extension of F. We recall the following classical result of class field theory.
 1.2. REGULATORS 11

T HEOREM 1.1.12. The Artin map

φF : ClF → Gal(HF /F),

defined by φF ([p]) = ϕp for all primes p of F, is an isomorphism.

1.2. Regulators

Let F be a number field. We will shorten our notation for units slightly as follows.

N OTATION 1.2.1. We set EF = OF× .

D EFINITION 1.2.2. We say that a set of r units of F is independent if it generates a subgroup of EF

isomorphic to Zr .

We will use the following notation.

N OTATION 1.2.3. Set r = r1 (F) + r2 (F) − 1. Let σ1 , . . . , σr1 (F) : F ,→ R be the real embeddings
of F and σr1 (F)+1 , . . . , σr+1 : F ,→ C be representatives of the distinct complex conjugacy classes of
Lr1 (F)+r2 (F)
complex embeddings of F. Let V = i=1 Rσi and
( )
r+1 r+1
V0 = ∑ aiσi ∈ V | ∑ ai = 0 .
i=1 i=1

We define an R-linear homomorphism κ : F × ⊗Q R → V0 by

r+1
κ(α) = ∑ ci log |σi(α)|σi,
i=1
where 
1 if σi real
ci =
2 if σi complex.

The following is typically proven in the course of a proof of Dirichlet’s unit theorem. Let κ0 denote
the restriction of κ to a map κ0 : EF ⊗Q R → V0 , the image landing in V0 by the product formula.

P ROPOSITION 1.2.4. The map κ0 : EF ⊗Q R → V0 is an isomorphism.

D EFINITION 1.2.5. The regulator RF (α1 , α2 , . . . , αr ) of a set {α1 , α2 , . . . , αr } of r independent

units is | det R|, where R = R(α1 , α2 , . . . , αr ) is the r-by-r matrix with (i, j)-entry ci log |σi (α j )|.

R EMARK 1.2.6. Exactly one archimedean place is omitted in the definition of the regulator. For
any α ∈ EF , one has
r+1 r+1
∑ ci log |σi(α)| = log ∏ |σi(α)|ci = 0
i=1 i=1
12 1. CLASS GROUPS AND UNITS

by the product formula, so the rows of the matrix determining the regulator sum to the what would have
been the row corresponding to the embedding that is omitted. The choice of σi and their ordering are
then seen by the usual rules for the effect of row operations on determinants to not affect the absolute
value of the determinant of the matrix in question.

In particular, we have the following.

L EMMA 1.2.7. For a set {α1 , α2 , . . . , αr } of r = rankZ EF independent units, RF (α1 , α2 , . . . , αr ) is

the absolute value of the determinant of the linear transformation κ0 : EF ⊗Z R → V0 relative to the
1
basis of EF ⊗Z R given by the αi and the basis of V0 given by σ j − r+1 ∑r+1
k=1 σk for 1 ≤ j ≤ r.

D EFINITION 1.2.8. Let A and B be subgroups of an abelian group.

a. We say that A and B are commensurable if A and B are of finite index in A + B.
b. If A and B are commensurable, then we define the relative index of A in B by

(B : A) = [A + B : A] · [A + B : B]−1 .

The following is easily verified.

L EMMA 1.2.9. Let A and B be finitely generated subgroups of a vector space V over a subfield E
of C. If A and B are commensurable, then there exists an E-linear automorphism T of V such that
T (A) = B, and for any such T , we have (B : A) = | det(T )|.

L EMMA 1.2.10. Suppose that {α1 , α2 , . . . , αr } and {β1 , β2 , . . . , βr } are independent sets of r units
in F, where r = rankZ EF . Let

A = µ(F) · hα1 , α2 , . . . , αr i and B = µ(F) · hβ1 , β2 , . . . , βr i.

Then
RF (β1 , β2 , . . . , βr )
= (B : A).
RF (α1 , α2 , . . . , αr )
P ROOF. Let V = EF ⊗Z R. There exists an automorphism T of V carrying the image of A in V to
the image of B. Since both A and B contain µ(F), we have (B : A) = | det T |. On the other hand, κ0 is
an R-linear isomorphism, so
RF (β1 , β2 , . . . , βr ) | det(κ0 ◦ T )|
= = | det T |,
RF (α1 , α2 , . . . , αr ) | det κ0 |
with the determinant taken relative to the bases of Lemma 1.2.7. 

C OROLLARY 1.2.11. If {α1 , α2 , . . . , αr } and {β1 , β2 , . . . , βr } are independent sets in EF with im-
ages generating the same subgroup of EF /µ(F), then

RF (α1 , α2 , . . . , αr ) = RF (β1 , β2 , . . . , βr ).
 1.3. FINITE GALOIS EXTENSIONS 13

We may then make the following definitions.

D EFINITION 1.2.12. A fundamental set of units of a number field F is a set {α1 , . . . , αr } of r units
in EF such that
EF = µ(F) · hα1 , α2 , . . . , αr i.

D EFINITION 1.2.13. The regulator RF of a number field F is RF (α1 , α2 , . . . , αr ) for any fundamen-
tal set of units {α1 , α2 , . . . , αr } of F.

We also need the following notation.

N OTATION 1.2.14. Let wF denote the number of roots of unity in a number field F.

Now that we have defined the regulator, we can describe the residue at s = 1 of the Dedekind zeta
function.

T HEOREM 1.2.15 (Analytic class number formula). For a number field F, one has
2r1 (F) (2π)r2 (F) hF RF
lim (s − 1)ζF (s) = .
s→1 wF |dF |1/2

1.3. Finite Galois extensions

Suppose that E/F is a finite Galois extension, and let G = Gal(E/F). Then ClE becomes a G-
module via the action σ ([a]) = [σ a] for σ ∈ G and a ∈ IE , where

σ a = {σ (a) | a ∈ a} ∈ IE .

The Galois group Gal(HE /E) is a G-module too, but to see this requires a little bit of work.

P ROPOSITION 1.3.1. Let E be a finite Galois extension of F. Then HE /F is Galois.

P ROOF. Let H fE denote the Galois closure of HE as an extension of F. Let G = Gal(E/F). For


σ ∈ G, let σ̃ denote a lift of σ to Gal H fE /F . Note that the field σ̃ (HE ) is independent of the choice


of lift σ̃ of σ , as any element in Gal HfE /E necessarily preserves the subfield HE of H fE , as HE /E is
Galois.
Next, we claim that
fE =
H ∏ σ̃ (HE ).
σ ∈G

It suffices to show that ∏σ ∈G σ̃ (HE )/F is Galois by the minimality of H

fE as a Galois extension of F.


For this, note that for any δ ∈ Gal H fE /F , one has

δ σ̃ (HE ) = σ̃ 0 (HE ),
14 1. CLASS GROUPS AND UNITS

where σ 0 = δ |E σ ∈ G, by the independence of conjugates of HE from the choice of lift. This proves the
claim. It then follows that H˜E /E is abelian, since each σ̃ (HE )/E is. That is, any τ ∈ Gal(σ̃ (HE )/E)
has the property that σ̃ |−1
HE τ σ̃ |HE ∈ Gal(HE /E), and the latter group is abelian.


fE /E . If Iv
Now, let Iv be the inertia group at a prime v of E in the abelian extension Gal H
is nontrivial, then its image in some Gal(σ̃ (HE )/E) must be as well. Then σ̃ −1 Iv σ̃ has nontrivial
image in Gal(HE /E). Since the former group equals the inertia group Iσ −1 v and HE /E is unramified,
this image must be trivial. Therefore Iv = 0, and so H
fE /E is an unramified abelian extension of E
fE = HE , as desired.
containing HE . By the maximality of HE , we have H 

Consequently, if E/F is finite Galois with group G, then Gal(HE /E) becomes a G-module for the
conjugation action: σ ∈ G acts on τ ∈ Gal(HE /E) by sending it to σ τσ −1 . The following is then a
consequence of class field theory.

P ROPOSITION 1.3.2. For E/F finite Galois with Galois group G, then Artin map φE is G-equivariant
(i.e., a G-module homomorphism), which is to say that

φE ([σ a]) = σ φE ([a])σ −1

for all σ ∈ G and a ∈ IE .

D EFINITION 1.3.3. For E/F finite Galois, we define a map

jE/F : ClF → ClE , jE/F ([a]) = [aOE ]

and the norm map

" ! #
NE/F : ClE → ClF , NE/F ([a]) = ∏ σ (a) ∩ OF .
σ ∈G

Our goal in this section will be to study these maps.

L EMMA 1.3.4. Let p be a prime, and let AK denote the p-part of the class group of any number
field K. If the order of G is prime to p , then the maps

AF → AG
E and (AE )G → AF

defined by jE/F and NE/F , respectively, are isomorphisms.

P ROOF. Note that NE/F ◦ jE/F = |G|, so jE/F is injective on AF and the image of NE/F contains
AF . Let Z0 = Z[|G|−1 ], and note that AE is a Z0 [G]-module. Define an idempotent
1
εG = NG ∈ Z0 [G].
|G|
 1.3. FINITE GALOIS EXTENSIONS 15

Since εG AE = AG G
E , the group AE is both a submodule and a quotient of AE . Note that |G|εG = jE/F ◦
NE/F , which forces jE/F to have image AG G
E on AF . The map AF → AE induced by jE/F is therefore an
isomorphism. As AE is finite, both AG
E and (AE )G have the same order, so therefore now the same order
as AF . Since NE/F induces a surjective map (AE )G → AF , that map must also be an isomorphism. 

For instance, we have that A∆Q(µ p ) = 0 for any prime p, as p is prime to [Q(µ p ) : Q] = p − 1.

P ROPOSITION 1.3.5. Let E/F be a finite Galois extension of number fields with Galois group G.
There is a canonical exact sequence

0 → ker jE/F → H 1 (G, OE× ) → IEG /IF → ClG 1

E / jE/F (ClF ) → H (G, PE ).

P ROOF. By Hilbert’s Theorem 90, we have a commutative diagram

0 / O× / F× / PF /0
F


0 / O× / F× / PG / H 1 (G, O × ) / 0,
F E E

and it provides an isomorphism H 1 (G, OE× ) ∼

= PEG /PF . Noting this and applying the snake lemma to
the commutative diagram

0 / PF / IF / ClF /0

  
0 / PG / IEG / ClG / H 1 (G, PE ),
E E

we obtain the desired exact sequence. 

Note that the map ker jE/F → H 1 (G, OE× ) is given explicitly by taking [a] to a cocycle σ 7→ α σ −1 ,
where α ∈ E × satisfies (α) = aOE . The map H 1 (G, OE× ) → IEG /IF is given by taking a cocycle f to
the image of an element (α) ∈ IEG with f (σ ) = α σ −1 for all σ ∈ G.

D EFINITION 1.3.6. An ideal in IF with class in the kernel of jE/F is said to capitulate in the
extension E/F. The kernel of jE/F is known as the capitulation kernel.

L EMMA 1.3.7. Let E/F be a Galois extension of number fields with Galois group G. The cokernel
of NE/F is canonically isomorphic to the Galois group of the maximal unramified abelian subextension
of F inside E.

P ROOF. The norm map on ideal classes factors the the G-coinvariant group (ClE )G of ClF . We
consider the complex
NE/F
(ClE )G −−−→ ClF → coker NE/F → 0.
16 1. CLASS GROUPS AND UNITS

Using the Artin map, we may write the latter complex as

Gal(HE /E)G → Gal(HF /F) → coker NE/F → 0,

where the first map is restriction, and therefore has image Gal(HF /E ∩ HF ). It follows that

coker NE/F ∼
= Gal(HF ∩ E/F),

as desired. 

Lemma 1.3.7 has the following immediate corollary.

C OROLLARY 1.3.8. If E/F is totally ramified at any prime, then NE/F is surjective.

Now suppose that E/F is abelian. Let Iv denote the inertia group in G = Gal(E/F) at a prime v of
F, and let
M
ΣE/F : Iv → G
v
denote the map that is the product of the natural inclusions.

P ROPOSITION 1.3.9. Let E/F be an abelian extension of number fields with Galois group G. Then
there is an exact sequence
NE/F
ker ΣE/F → QE/F −−−→ ClF → coker ΣE/F → 0,

where QE/F is the quotient of ClE by its subgroup that is taken to the commutator subgroup of
Gal(HE /F) under the Artin map.

P ROOF. The exactness outside of (ClE )0 follows from Lemma 1.3.7, since coker ΣE/F ∼
= Gal(HF ∩
E/F) by definiton.
We define
ker ΣE/F → QE/F
as follows. Employing the Artin map, we have canonical isomorphisms

QE/F ∼
= Gal(HE /E)/[Gal(HE /F), Gal(HE /F)] ∼
= Gal(L/E),

where L is the maximal unramified extension of E that is abelian over F. Let Jv denote the inertia group
at a prime w over v in Gal(L/F). As L/E is unramified, Jv maps isomorphically to Iv under restriction.
We have a map
M
Jv → Gal(L/F)
v
L
given by the product of the canonial inclusions, and the map from v Iv is then given by the identifi-
cations Iv ∼
= Jv . Since ker ΣE/F lands in Gal(L/E) under this map, we have the desired map.
 1.3. FINITE GALOIS EXTENSIONS 17

It remains to check exactness at QE/F . Again, we use the Artin isomorphism to see that the kernel
of the map to ClF is precisely Gal(L/E · HF ). On the other hand, the image of ker ΣE/F in Gal(L/E) is
the intersection with Gal(L/E) of the subgroup of Gal(L/F) generated by its inertia groups. As L/F
is abelian and HF is the Hilbert class field of F, this is precisely Gal(L/E · HF ). 

We have the following interesting corollary.

C OROLLARY 1.3.10. Let E/F be a cyclic p-extension, and suppose that there is at most one prime
of F that ramifies in it. Then the map (ClE )G → ClF induced by NE/F is injective.
L
P ROOF. Since E/F is cyclic, the quotient QE/F of Proposition 1.3.9 equals (ClE )G . Since v∈S Iv
is either Iv at the unique ramified prime, or 0 if there is no ramified prime, the map ΣE/F is injective. 

In the case that |G| divides the order of p, we can give another nice consequence. First, we require
the following lemma.

L EMMA 1.3.11. Suppose that E/F is a finite Galois p-extension ramified at no more than one
prime of F. If E/F is nonabelian, suppose further that, if there is such a prime, that it is nonsplit in
the extension. Then if AF = 0, we have AE = 0 as well.

P ROOF. We begin with the case that G = Gal(E/F) is abelian. By Corollary 1.3.10, the map
(AE )G → AF is injective, and therefore (AE )G = 0. Thus, we have

AE /mG AE = (AE )G /p(AE )G = 0.

Noting Proposition 2.3.7, Nakayama’s lemma then tells us that AE = 0.

Since any finite p-group has a finite filtration with abelian (or even cyclic) graded quotients, the
result in general follows from the abelian case by recursion, noting that by assumption there is at most
one prime that ramifies in each intermediate extension. 

We illustrate the use of this with the following interesting example.

E XAMPLE 1.3.12. Let F = Q(µ p ) and E = F(p1/p ). Then this extension is totally ramified of
degree p at the unique prime above p in F, which is (1 − ζ p ) for a primitive pth root of unity ζ p .
Therefore, we have that (AE )G ∼
= AF via the norm map. For a prime p such that AF = 0, which is
known as a regular prime (e.g., all primes less than 37), Lemma 1.3.11 implies that AE = 0. When
p = 37, it turns out that AF = ClF ∼
= Z/37Z, and in fact we have that AE is isomorphic to Z/37Z as
well.

To go even further, it is useful to restrict to the case of a cyclic extension. We begin with the
following useful result.
18 1. CLASS GROUPS AND UNITS

P ROPOSITION 1.3.13. Let E/F be a cyclic extension of number fields. Then

NE/F E × = F × ∩ NEw /Fv Ew× ,

\

where v runs over all primes of F and w is some prime of E above v.

P ROOF. Let Gv denote the decomposition group in G at any w over v. Choose a set S of represen-
tatives of Gv \G. For a ∈ E × , we have
!
NE/F (a) = NEw /Fv ∏ σa ,
σ ∈S

so every global norm is a local norm everywhere.

Recall the following exact sequence for the Brauer group of E/F:
1
0 → H 2 (G, E × ) → H 2 (Gv , Ew× ) →
M
Z/Z,
v
|G|

where Gv denotes the decomposition group in G at any w over v. By the periodicity of Tate cohomology
of a cyclic group, this becomes
1
0 → Ĥ 0 (G, E × ) → Ĥ 0 (Gv , Ew× ) →
M
Z/Z.
v
|G|

In particular, we have an injection

F × /NE/F E × ,→ Fv× /NEw /Fv Ew× .

M

Therefore, if a ∈ F × is a local norm everywhere, it is a global norm, as desired. 

Note that an element a ∈ F × is automatically a local norm at any prime where E/F is unramified
and the valuation of a at that prime is trivial. Hence, there are actually only finitely many places to
check that a is a local norm to see that it is a global one.
We now derive a nine-term exact sequence that gives us information on the behavior of class groups
in cyclic extensions. A proof is possible by making use of Tate cohomology, as found in the appendix
to [HS], but we give a more explicit proof.

T HEOREM 1.3.14. Let E/F be a cyclic extension of number fields, and let G be its Galois group.
Let Iv denote the inertia group in G at a prime v of F, and let
M
ΣE/F : Iv → G
v
 1.3. FINITE GALOIS EXTENSIONS 19

denote the map that is the product of the natural inclusions. Then we have an exact sequence

0 → ker jE/F → H 1 (G, OE× ) → IEG /IF → ClG

E / jE/F (ClF )
NE/F
→ OF× /NE/F OE× → ker ΣE/F → (ClE )G −−−→ ClF → coker ΣE/F → 0.

Moreover, the group IEG /IF is noncanonically isomorphic to

L
v Iv .

P ROOF. By Proposition 1.3.5, we have an exact sequence

0 → ker jE/F → H 1 (G, OE× ) → IEG /IF → ClG 1

E / jE/F (ClF ) → H (G, PE )

including the first row. By Proposition 1.3.9, the final part of the sequence beginning with ker ΣE/F is
exact.
Let σ be a generator of G. Define

→ OF× /NE/F OE×

∂
ClG
E / jE/F (ClF ) −

as the map that takes image of an ideal class [a] ∈ ClG

E to the image of NE/F α, where α is any generator
of aσ −1 . To see that this is well-defined, note that if α is replaced by another generator α 0 , then α 0 = αu
with u ∈ OE× , and
NE/F α 0 · (NE/F α)−1 ∈ NE/F OE× .
Moreover, if a is replaced by an ideal a0 with the same class, then a0 = a · b for some b ∈ E × . We then
have
(ab)σ −1 = aσ −1 bσ −1 = (αbσ −1 ).
It follows that
NE/F (αbσ −1 ) = NE/F (α).
Finally, if b ∈ jE/F (ClF ), then bσ −1 = (1), so ∂ takes [b] to 1.
We check exactness at ClG G σ −1 = (1), so ∂ maps [b] to 1. On
E / jE/F (ClF ). If b ∈ IE , then again b
the other hand, if ∂ takes the image of [a] to NE/F α = 1, and therefore α = β σ −1 with σ ∈ G. We then
have
(aβ −1 )σ −1 = (1),
which means aβ −1 ∈ IEG . As [aβ −1 ] = [a], we have that the image of [a] is in the image of the map
from IEG /IF .
Next, we define
OF× /NE/F OE× → ker
M

Iv → G
v
by the direct sum of the local reciprocity maps ρEw /Fv . (We remark that Iv = 0 for all but finitely many
v, so the map ΣE/F makes sense.) Since the product of the reciprocity maps at all places on a global
20 1. CLASS GROUPS AND UNITS

element is trivial, the image of this map is indeed contained in ker ΣE/F Also, the map is well-defined
since every global norm is a local norm. Note that the image of ∂ is the set of NE/F α ∈ OF× with α ∈
E × . Again, such elements are local norms, and map to zero under each ρEw /Fv . Conversely, if c ∈ OF×
satisfies ρEw /Fv (c) = 1 for every v, then c ∈ NEw /Fv Ew× for all v, since c is a unit. By Theorem 1.3.13,
we have that c = NE/F α for some α ∈ F × with (α) = aσ −1 for some a ∈ IE and σ ∈ G. In other words,
c is the image of the image of the class of [a] under ∂ .
We check exactness at ker ΣE/F . Let L denote the maximal abelian extension of E that is abelian
over F. For c ∈ OF× , we have

∏ ρLw0 /Fv (c) = 1,

v

with w0 lying over w. Since ρLw0 /Fv (c)|E = ρEw /Fv (c), the image of c in Jv is ρLw0 /Fv (c), and the resulting
product in Gal(L/E) is trivial. On the other hand, suppose that σ ev ∈ Jv lifts some σv ∈ Iv and

∏ σev = 1.
v

Then there exist local units cv ∈ OF×v for each v with ρLw0 /Fv (cv ) = σ
ev . We take cv = 1 if σ
ev = 1. By
global class field theory, the idele c with cv = cv for each v is the product of the norm of an idele b of
L with an element c ∈ F × . Recall that

CF /NHF /F CHF ∼
= ClF ,

so we have that
F × NHF /F IHF = F × ∏ OF×v ,
v

where we take OFv = Fv if v is archimedean. Since L contains HF , the idele b may be taken to be a unit
at all places. But, as each cv is a local unit at all v and

(NL/F b · c)v = cv = cv

for all v, this means that c must be a unit at all places as well. That is, c ∈ OF× . As NL/F b is a local
norm from E everywhere, we have

ρEw /Fv (c) = ρEw /Fv (cv ) = σv

for every v, as desired.

Finally, recall that IE is the free abelian group generated by the prime ideals of OE . For an element
of IE to be fixed under G, every prime in its decomposition must appear with the same exponent as
its conjugates. That is, IEG is generated by the ∏gi=1 Pi , where P1 , . . . Pg are the primes of E lying
 1.3. FINITE GALOIS EXTENSIONS 21

over a prime p of F. Of course, (∏gi=1 Pi )ev = pOE , where ev is the ramification index of the place v
corresponding to p, so we have
IEG /IF ∼ Z/ev Z ∼
M M
= = Iv .
v v∈S


R EMARK 1.3.15. Every map but the map between the two rows is canonical in the exact sequence
of Theorem 1.3.14. The remaining map depends only upon a choice of generator of G. It can be made
canonical by considering instead the map
× ×
E / jE/F (ClF ) ⊗Z G → OF /NE/F OE
ClG

given on the image of a tensor of [a] ∈ ClG σ −1 = αO × and taking the image
E and σ ∈ G by writing a E
of NE/F α ∈ OF× in the quotient.

Next, we generalize the situation slightly.

N OTATION 1.3.16. For a set S of places in a number field F, we let S f denote its subset of finite
places and S∞ its subset of infinite places.

D EFINITION 1.3.17. Let S denote a set of places of F.

a. The S-class group ClF,S of F is the quotient of the class group by the subgroup generated by the
classes of the finite primes in S.
b. The Hilbert S-class field HF,S of F is the maximal unramified abelian extension of F in which
all primes in S split completely.
c. The ring of S-integers OF,S of F is

OF,S = {a ∈ F | vp (a) ≥ 0 for all p 6∈ S f },

where p is used to denote a finite prime of F and vp its additive valuation.

d. The S-ideal group IF,S is the group of nonzero fractional ideals in OF,S , and the S-principal ideal
group PF,S is the subgroup of principal fractional ideals.
×
e. The S-unit group in F is OF,S .

N OTATION 1.3.18. If S is a set of primes of F and let E/F is a finite extension, then we let SE
denote the set of places of E lying over those in S. For brevity, we denote OE,SE , ClF,SE , and so on
more succinctly by OE,S , ClE,S , and so on similarly. That is, we use S in the subscript to denote SE .
If E/F is algebraic, we may still speak of its S-integers OE,S as the union of S-integers in the finite
subextensions of F in E.
22 1. CLASS GROUPS AND UNITS

Let us fix a set of places of S for the rest of this section.

R EMARK 1.3.19. The Artin isomorphism φF induces an isomorphism

φF,S : ClF,S → Gal(HF,S /F).

We have an analogue of the exact sequence of Theorem 1.3.14 for S-class groups and S-units. The
proof is much as before, and is therefore omitted.

T HEOREM 1.3.20. Let E/F be a cyclic extension of number fields, and let G be its Galois group.
Let Iv (resp., Gv ) denote the inertia group (resp., decomposition group) in G at a prime v of F, and let
M M
ΣSE/F : Iv ⊕ Gv → G
v∈S
/ v∈S

denote the map that is the product of the natural inclusions. Then we have an exact sequence
jE/F
×
0 → ker(ClF,S −−→ ClE,S ) → H 1 (G, OE,S G
) → IE,S /IF,S → ClG
E,S / jE/F (ClF,S )
NE/F
× ×
→ OF,S /NE/F OE,S → ker ΣSE/F → (ClE,S )G −−−→ ClF,S → coker ΣSE/F → 0.

1.4. Kummer theory

For a set of S primes of F, we let S f denote the set of finite places of F in S, we let S∞ denote the
set of archimedean places, and for any n ≥ 1, we let Sn denote the set of primes of S above p for any
prime p dividing n. If E is an extension of F, we generally also use the symbol S to denote the set of
primes SE of E above those in S. We will let V denote the set of all primes of F, so we may speak of
V∞ and so forth. For brevity, let us set Vn∞ = Vn ∪V∞ .

D EFINITION 1.4.1. We say that an extension E of F is S-ramified if it is unramified outside of the

places in S.

L EMMA 1.4.2. There exists a maximal S-ramified extension FS of F, and it is Galois over F.

P ROOF. A union of S-ramified extensions is S-ramified, so the existence of FS is clear. If E is an

S-ramified finite degree extension of F, then so is any conjugate of E over F in an algebraic closure F
of F containing E, as the inertia degrees at conjugate primes above p in E and σ (E) are the same (and
similarly for real places). The product
∏ σ (E)
σ : E,→F
is Galois (in fact, it is the Galois closure of E in F) and also S-ramified as a compositum of S-ramified
extensions. Therefore, FS is a union of finite Galois subextensions, hence itself Galois. 
 1.4. KUMMER THEORY 23

D EFINITION 1.4.3. We use GF,S to denote the Galois group Gal(FS /F), i.e., the Galois group of
the maximal S-ramified extension FS of F.

Kummer theory in S-ramified extensions has as its basis the following proposition.

P ROPOSITION 1.4.4. Let S be a set of primes of F. We have a canonical isomorphism

∼
→ H 1 (GF,S , OF×S ,S ),
ClF,S −

given by taking an ideal class [a] to the cocycle that takes σ ∈ GF,S to α σ −1 , where α is a generator
of aOFS ,S .

P ROOF. To reduce clutter in the notation, let us set G = GF,S and Ω = FS . A similar argument to
that of the proof of Proposition 1.3.5 produces an isomorphism
G ∼ ×
PΩ,S → H 1 (G , OE,S
/PF,S − )

that takes (α) to σ 7→ α σ −1 . Again similarly to before, we have the commutative diagram

(1.4.1) 0 / PF,S / IF,S / ClF,S /0

  
0 / PG / IG / ClG
Ω,S Ω,S Ω,S

The lower row arises as a direct limit of like sequences for intermediate finite extensions E of F in
Ω. However, since S contains the finite primes that are ramified in any such extension E/F, the map
G is not merely an injection, but an isomorphism. Moreover, j
IF,S → IE,S Ω/F is the zero map, since Ω
contains HF by definition, and every ideal in IF becomes principal in HF . Hence, the snake lemma
provides an isomorphism
G
ClF,S → PΩ,S /PF,S

taking [a] to (α) where (α) = aOΩ,S . 

P ROPOSITION 1.4.5. Suppose that S contains Vn∞ . Then there is a canonical exact sequence
× ×n
1 → OF,S /OF,S → H 1 (GF,S , µn ) → ClF,S [n] → 0.

P ROOF. For any α ∈ OF×S ,S , the extension FS (α 1/n )/FS is unramified outside of S and therefore
trivial, as the only primes that can ramify in such a Kummer extension are the real places, those p with
vp (α) 6= 0, and those primes dividing n, all of which are contained in S. We then have that
n
1 → µn → OF×S ,S →
− OF×S ,S → 1
24 1. CLASS GROUPS AND UNITS

is exact, and the result follows immediately from the exact sequence
n n
H 0 (GF,S , OF×S ,S ) →
− H 0 (GF,S , OF×S ,S ) → H 1 (GF,S , µn ) → H 1 (GF,S , OF×S ,S ) →
− H 1 (GF,S , OF×S ,S ).

In the case that S = ∅, we have the following.

L EMMA 1.4.6. Fix n ≥ 1 such that F contains µn , and let Bn ≤ OF× be the subgroup

Bn = {a ∈ OF× | F(a1/n )/F is unramified}.

There is a canonical exact sequence

1 → Bn /OF×n → H 1 (GF,∅ , µn ) → ClF [n].

P ROOF. From the short exact sequence

n
1 → µn → OF×∅ →
− OF×n
∅
→ 1,

we obtain an exact sequence

n n
(1.4.2) OF× →
− OF×n
∅
∩ OF× → H 1 (GF,∅ , µn ) → H 1 (GF,∅ , OF×∅ ) →
− H 1 (GF,∅ , OF×n
∅
).
n
As the nth power map OF×∅ →
− OF×∅ takes values in OF×n
∅
, the kernel of the rightmost map in (1.4.2) is
contained in the kernel of
n
H 1 (GF,∅ , OF×∅ ) →
− H 1 (GF,∅ , OF×∅ ),

which is isomorphic to ClF [n] by Proposition 1.4.4. Noting that

Bn = F∅×n ∩ OF× = OF×n

∅
∩ OF× ,

equation (1.4.2) yields the result. 

D EFINITION 1.4.7. A number field F is said to be abelian if it is an abelian extension of Q.

D EFINITION 1.4.8. A number field F is said to be totally real if it has no complex places.

R EMARK 1.4.9. There exits a maximal totally real subfield F + of any number field F, as the
compositum of any two totally real fields is totally real.

D EFINITION 1.4.10. A number field F is CM if it has no real places and is a degree 2 extension of
F +.
 1.4. KUMMER THEORY 25

E XAMPLE 1.4.11. Let n ≥ 3. Then the cyclotomic field Q(µn ) is CM, and

Q(µn )+ = Q(ζn + ζn−1 ),

where ζn is a primitive nth root of unity. As a consequence of this and the Kronecker-Weber theorem,
every abelian field is either totally real or CM.

Fix a CM field F, and let τ be the nontrivial element of Gal(F/F + ). Given a Z[Gal(F/F + )]-
module A, we have submodules
A± = {a ∈ A | τ(a) = ±a}.
Note that
A+ ∩ A− = A[2].
and A/(A+ + A− ) is 2-torsion. If multiplication by 2 is invertible on A, then
a + τ(a) a − τ(a)
A∼
= A+ ⊕ A− , a 7→ + .
2 2
L EMMA 1.4.12. The groups OF× and (OF× )+ have the same Z-rank, and (OF× )− is the group µ(F)
of roots of unity in F.

P ROOF. The first statement is an immediate consequence of Dirichlet’s unit theorem. Since it
holds, (OF× )− consists only of elements of finite order, which is to say, roots of unity. Since every root
of unity ξ satisfies τ(ξ ) = ξ −1 , we have the result. 

We note that for an odd prime p, the map jF/F + provides a canonical identification of AF + with A+
F
by Lemma 1.3.4.

L EMMA 1.4.13. The map ClF + → Cl+

F induced by jF/F + has kernel of order dividing 2.

P ROOF. Note that if τ(x)x = 1 for x ∈ OF× , then x must be a root of unity. On the other hand,
the group of τ(y)y−1 with y ∈ OF× contains µ(F)2 = µ(F)τ−1 . Thus H 1 (G, OF× ) ∼ = Ĥ −1 (G, OF× ) is
isomorphic to a quotient of µ(F)/µ(F)2 . The result then follows from Proposition 1.3.5. 

R EMARK 1.4.14. If L and M are Galois extensions of a field K with L contained in M, then
Gal(L/K) acts on H i (Gal(M/L), A) for any Z p [Gal(M/K)]-module A. The action is induced by the
following action of τ ∈ Gal(M/K) on a cochain f ∈ Ci (Gal(M/L), A):

(τ · f )(σ1 , . . . , σi ) = τ · f (τ −1 σ1 τ, . . . , τ −1 σi τ).

On cohomology, this action factors through an action on Gal(L/K) since, on Gal(M/L), this action is
the conjugation action on cohomology, which is trivial.
26 1. CLASS GROUPS AND UNITS

For a finitely generated abelian group A, let us use r(A) to denote its rank and

r p (A) = dimF p A[p]

to denote its p-rank for a prime p.

T HEOREM 1.4.15. Let F be a CM field such that µ p ⊂ F for an odd prime p. We then have
− ×
r p (Cl+ +
F ) − δ ≤ r p (ClF ) ≤ r p (ClF ) + r(OF ),

where δ = 0 if F(µ(F)1/p )/F is ramified at p and 1 otherwise.

P ROOF. Note that

H 1 (GF,∅ , µ p ) ∼
= Hom(Gal(HF /F), µ p ),
and
Hom(Gal(HF /F), µ p )± ∼
= Hom(ClF , µ p )± ∼
= Hom(Cl∓
F , µ p ),

as τ acts on µ p by inversion. Combining this with Lemma 1.4.6, with B = B p as in said lemma, we
have
×p ±
r p (Cl∓ 1 ± ±
F ) = r p (H (GF,∅ , µ p ) ) ≤ r p (ClF ) + r p ((B/OF ) ).

By Lemma 1.4.12, we have that OF× = OF+ · µ(F) and

r p ((B/OF×p )− ) = r p (B ∩ µ(F)) = δ ,

while
r p ((B/OF×p )+ ) ≤ r p ((OF× /OF×p )+ ) = r((OF× )+ ) = r(OF× ).
The result follows. 

1.5. Leopoldt’s conjecture

For each place v of F, Let

n
×
v = lim
Fc Fv× /Fv×p ,
←−
n
and consider its subgroup
n
Uv = lim OF×v /OF×p .
←− v
n
×∼
v = Z p ⊕ Uv , and Uv is the group of p-power roots of unity in Fv if v does not lie
If v is finite, then Fc
over p while U is the group of 1-units if v lies over p. If v is infinite, then U = Fc× , and both groups
v v v
are trivial unless v is real and p = 2, in which case they are Z/2Z.
Let us set
n
EF = OF× ⊗Z Z p ∼
= lim O × /O ×p .
←− F F
n
 1.5. LEOPOLDT’S CONJECTURE 27

We may consider the natural map

ιF = (ιv )v∈Vp : EF → Uv .
M

v∈Vp

Clearly, the kernel of OF× → OF×v is trivial for v ∈ Vp . Yet, the problem may arise that there exist, for
instance, two units x, y ∈ OF× generating a rank two subgroup and a, b ∈ Z p such that ιv (x)a ιv (y)b = 1.
So, in theory, ιF could have a kernel. This brings us to Leopoldt’s conjecture.

C ONJECTURE 1.5.1 (Leopoldt). The map ιF : EF → v∈Vp Uv

L
is injective.

R EMARK 1.5.2. We could, equivalently, consider the map

ιF0 : EF → Uv ,
M

v∈Vp∞

×
that includes the archimdean places, setting Uv = Fc v for such v. The point is that Uv = 1 for
archimedean v unless p = 2 and v is real, in which case Uv ∼
= R× /R×2 .
We have that ker ιF0 ⊆ ker ιF by definition. On the other hand, we have (ker ιF )2 ⊆ ker ιF0 . In
particular, the two kernels have the same Z p -rank. Moreover, the 2-torsion in EF is µ2 , and −1 is not
in the kernel of ιv for any v, so ker ιF = 0 if and only if ker ιF0 = 0.

E XAMPLE 1.5.3. For F = Q, Leopoldt’s conjecture holds as EQ = 1 for p 6= 2 and EQ = µ2 for

p = 2.

Let S denote a finite set of primes of F containing Vp∞ . We wish to state several equivalent forms
of this conjecture. For this, we set
×
EF,S = OF,S ⊗Z Z p ,
and extend ιF to a map
×
ιF,S = (ιv )v∈S : EF,S →
M
v .
Fc
v∈S
Let XF,S denote the Galois group of the maximal abelian pro-p unramified outside S extension of F.
Let G
d ab denote the Galois group of the maximal abelian pro-p extension of F for each v. The following
v v
exact sequences will be useful.

T HEOREM 1.5.4. There are two exact sequences fitting into a commutative diagram

(1.5.1) 0 / ker ιF / EF / Lv∈S Uv / XF,S / AF /0

_ _ _

   
ιF,S ρF,S
0 / ker ιF,S / EF,S / Lv∈S Fc
v
× / XF,S / AF,S / 0,
28 1. CLASS GROUPS AND UNITS

where ρF,S is the product over v ∈ S of the composition of the p-completion of the local reciprocity
× ab ab
v → Gv with the natural map jv from Gv onto the decomposition group at v in XF,S , and
map ρv : Fc d d
where the maps XF,S → AF → AF,S are the natural quotient maps (under the indentifications given by
the Artin map).

P ROOF. In the horizontal sequences in the diagram (1.5.1), we note that im ρF,S (resp., the corre-
sponding map in the upper sequence) is the compositum of the decomposition groups (resp., inertia
groups) at all v ∈ S in XF,S . Being that XF,S already has trivial inertia groups at v 6∈ S, the quotient
coker ρF,S is therefore the Galois group of the maximal unramified abelian p-extension of F in which
all primes in S split completely (resp., maximal unramified abelian p-extension of F), and is therefore
canonically isomorphic to AF,S (resp., AF ) via Artin reciprocity.
For the upper horizontal sequence, the exactness at v∈S Uv will follow from the exactness at
L

c×
v∈S Fv in the lower horizontal sequence by noting that EF consists exactly of the elements of EF,S
L

v∈S Uv .
L
that have image under ιF,S lying in We are therefore reduced to proving the latter exactness.
n
Recall that H 1 (GF,S , µ pn ) is identified via Kummer theory with the quotient Bn /F ×p , where Bn
n
is the subgroup of x ∈ F × such that xOF,S = a p for some fractional ideal a of OF,S . In other words, we
have an exact sequence
×p n n
×
1 → OF,S /OF,S → Bn /F ×p → AF,S [pn ] → 0.

It then follows from the finiteness of AF,S that

n
lim Bn /F ×p ∼= EF,S ,
←−
n

We claim that there is an exact sequence

n n
Bn /F ×p → Fv× /Fv×p → XF,S /pn XF,S ,
M

v∈S

where the first map is induced by the localization maps and the second map is ρF,S taken modulo pn .
In that all of the terms of this sequence are finite, we can take the inverse limit as we vary n to obtain
an exact sequence
ιF,S M × ρF,S
(1.5.2) EF,S −−→ v −−→ XF,S ,
Fc
v∈S

finishing the verification of the exactness of the lower sequence.

Let us use ρn,v and jn,v to denote the modulo pn reductions of ρv and jv for any v. Any a ∈ Bn
has valuation a multiple of pn at v 6∈ S, so ρv (a) lies in the compositum of the inertia group and the
subgroup of pn th powers in G
d ab . In particular, we have that j (ρ (a)) = 1 for such v. Global class
v n,v n,v
 1.5. LEOPOLDT’S CONJECTURE 29

field theory then tells us that

∏ jv(ρv(a)) = ∏ jv(ρv(a)) = 1,
v∈S v

which tells us that (1.5.2) is a complex.

Let Mn be the maximal S-ramified abelian extension of F of exponent pn . Its Galois group
Gal(Mn /F) ∼
= XF,S /pn XF,S is the quotient of Gal(F ab /F) by the composition of all inertia groups
/ S and the pn th powers of all of the decomposition groups. By global class field theory,
at primes v ∈
we therefore have an isomorphism
IF ∼
pn
×
−
→ Gal(Mn /F).
/ Ov
F × · IF · ∏v∈S

where for simplicity of notation, we have set Ov = OFv . Similarly, if we let Ln0 be the maximal un-
ramified abelian extension of F of exponent pn in which every prime in S splits completely, so that
Gal(L0 /F) ∼
n = AF,S /pn AF,S , class field theory again provides an isomorphism
IF ∼
n → Gal(Ln0 /F).
−
F × · IFp · (∏v∈S Fv× × ∏v∈S ×
/ Ov )

We see, then, that we have isomorphisms

n n
IFp · (∏v∈S Fv× × ∏v∈S
/ Ov )
×
∼ F × · IFp · (∏v∈S Fv× × ∏v∈S
/ Ov ) ∼
×
n n = n = Gal(Mn /Ln0 ),
(F × ∩ IFp (∏v∈S Fv× × ∏v∈S × p ×
/ Ov )) · IF ∏v∈S Ov
× p
/ Ov
F · IF · ∏v∈S ×

where in the first step, we have used the second isomorphism theorem. Since
n
n
∼ IFp · (∏v∈S Fv× × ∏v∈S
/ Ov )
×
Fv× /Fv×p
M
= n
v∈S IFp ∏v∈S Ov×

and
!
pn
Bn = F × ∩ IF ∏ Fv× × ∏ Ov× ,
v∈S v∈S
/

we have an exact sequence

n
Bn → F × /F ×p → Gal(Mn /Ln0 ),
M

v∈S

where the maps agree with the maps in question, hence the result. 

R EMARK 1.5.5. Theorem 1.5.4 can also be derived using Poitou-Tate duality and Kummer theory.

P ROPOSITION 1.5.6. The kernel of ιF,S is contained in EF . In particular, Leopoldt’s conjecture is

equivalent to the injectivity of ιF,S .
30 1. CLASS GROUPS AND UNITS

P ROOF. Let α ∈ ker ιF,S . Then α may be written as

m
α = ∑ ai ⊗ ci ,
i=1
×
were ai ∈ OF,S and ci ∈ Z p for each 1 ≤ i ≤ m, for some m ≥ 0. For each v ∈ S f , we then have
m
∑ v(ai)ci = 0,
i=1

which means that the ci ∈ Z p are Z-linearly dependent if some v(ai ) 6= 0. If v(am ) 6= 0, without loss
of generality, then α v(am ) may be written as a sum of m − 1 tensors. Continuing in this way, we obtain
that some nonzero integer power of α is a Z p -linear combination of units at v. Since there are only
finitely many v ∈ S, we have α c ∈ EF for some c ∈ Z, which forces α ∈ EF . 

The following theorem is also a corollary of Theorem 1.5.4 and Proposition 1.5.6, which gives in
particular equivalent conditions for Leopoldt’s conjecture to hold (noting that EF is p-torsion free). Let
rankZ p A denote the Z p -rank of a finitely generated Z p -module A.

T HEOREM 1.5.7. The following are equivalent for a given δ ≥ 0:

i. rankZ p ker ιF = δ ,
ii. rankZ p im ιF = r1 (F) + r2 (F) − 1 − δ ,
iii. rankZ p ker ιF,S = δ , and
iv. rankZ p XF,S = r2 (F) + 1 + δ .

P ROOF. For v ∈ S, we have that


[F : Q ] if v ∈ V ,
v p p
rankZ p U
cv =
0 if v ∈ S −Vp .
We also have
rankZ p EF = r1 (F) + r2 (F) − 1
by Dirichlet’s unit theorem, and AF is finite. Note that

r1 (F) + 2r2 (F) = [F : Q] = ∑ [Fv : Q p].

v∈Vp

Hence, Proposition 1.5.6 and the exactness of the upper exact sequence in (1.5.1) yield the result. 

C OROLLARY 1.5.8. The Z p -module XF,S is finitely generated of Z p -rank independent of S contain-
ing Vp∞ .

The δ in Theorem 1.5.7 is known as the Leopoldt defect of F

 1.5. LEOPOLDT’S CONJECTURE 31

D EFINITION 1.5.9. The Leopoldt defect δ (F) is the Z p -rank of ker ιF .

Leopoldt’s conjecture for F is equivalent to the statement that the Leopoldt defect δ (F) is 0. We
may also phrase Leopoldt’s conjecture for F in terms of the nonvanishing of a p-adic regulator of F,
which replaces the always nonzero complex regulator.

D EFINITION 1.5.10. For a p-adic field E, the p-adic logarithm of E is the unique homomorphism
log p : E × → E such that log p (p) = 0 and such that for any x in the maximal ideal of OE , one has
∞ k
k−1 x
log p (1 + x) = ∑ (−1) .
k=1 k
R EMARK 1.5.11. The kernel of log p on a p-adic field E is µ(E).

N OTATION 1.5.12. We use C p to denote the completion of the algebraic closure of Q p with respect
to the unique extension of the p-adic absolute value on Q p . We have a p-adic absolute value

| · | p : C p → R≥0

with |p| p = p−1 .

R EMARK 1.5.13. The p-adic logarithm extends to a continuous homomorphism log p : C×

p → C p.

It turns out that C and C p are abstractly isomorphic (being algebraically closed of characteristic 0
and having the same cardinality), and we can fix an embedding ι : C → C p . Let d = [F : Q], and let
τi : F ,→ C p for 1 ≤ i ≤ r + 1 be the compositions τ = ι ◦ σi of the real and complex embeddings σi of
F previously chosen in Section 1.2.

D EFINITION 1.5.14. Let α1 , . . . , αr be r independent units in EF . The p-adic regulator R p (F) of

EF is the determinant of the r-by-r matrix R p (α1 , . . . , αr ) = (ci log p τi (α j ))i, j , where ci is 1 if σi is real
and 2 if σi is complex.

R EMARK 1.5.15. The p-adic regulator is well-defined up to sign, so as an element of C×

p /h−1i.

The following is immediate.

P ROPOSITION 1.5.16. Leopoldt’s conjecture for a number field F is equivalent to the statement the
nonvanishing of the p-adic regulator R p (F).
×
Baker proved that if α1 , . . . αn ∈ Q are such that 2πi, log α1 , . . . , log αn are Q-linearly independent,
then they are Q-linearly independent. Via Baker’s method, Brumer proved a p-adic analogue.

T HEOREM 1.5.17 (Brumer). Let β1 , . . . , βn be algebraic numbers that are also p-adic units, and
suppose that the p-adic logarithms log p β1 , . . . , log p βn are Q-linearly independent. Then these loga-
rithms are also Q-linearly independent.
32 1. CLASS GROUPS AND UNITS

Using this result, Brumer was able to prove Leopoldt’s conjecture for an abelian extensions of
number fields with r = 0. Note that the only fields with r = 0 are Q and the imaginary quadratic fields.
We need several preliminary results. We begin with the following result, only the first part of which is
needed at the moment.

P ROPOSITION 1.5.18. Let G be a finite abelian group and f : G → C be a function. Let Ĝ denote
the group of characters G → C× .
a. We have !
∏ ∑ χ(σ ) f (σ ) = det( f (σ τ −1 ))σ ,τ∈G .
χ∈Ĝ σ ∈G

In fact, the rank of ( f (σ τ −1 ))σ ,τ∈G is the number of χ ∈ Ĝ such that ∑σ ∈G χ(σ ) f (σ ) 6= 0.
b. We have !
∏ ∑ χ(σ ) f (σ ) = det( f (σ τ −1 ) − f (σ ))σ ,τ6=1 .
χ∈Ĝ σ ∈G
χ6=1

P ROOF. We compare two bases of the complex vector space V of functions G → C: the set of
characters Ĝ and the set of δ -functions

1 τ = σ,
δσ (τ) =
0 τ 6= σ
for σ ∈ G. Consider the linear transformation T : V → V given by

T (g)(τ) = ∑ f (σ )g(σ τ).

σ ∈G
Applied to g = χ, we obtain
T (χ) = ∑ f (σ )χ(σ )χ,
σ ∈G
so χ is an eigenvector with eigenvalue ∑σ ∈G χ(σ ) f (σ ). It follows that det T is the product of the latter
sums over all χ. On the other hand,

T (δσ )(ρ) = ∑ f (τ)δσ (ρτ) = ∑ f (τ)δσ τ −1 (ρ) = ∑ f (τ −1 σ )δτ (ρ)

τ∈G τ∈G τ∈G
so
T (δσ ) = ∑ f (σ τ −1 )δτ
τ∈G
so the (τ, σ )-entry of the matrix of T with respect to this basis is f (σ τ −1 ). In that the determinant and
rank of T are independent of the choice of basis, we have part a.
For part b, we consider the codimension 1 subspace W of V that consisting of the g : G → C with
∑σ ∈G g(σ ) = 0. One basis of these functions is given by Ĝ − {1}, and another is given by the functions
 1.5. LEOPOLDT’S CONJECTURE 33

δσ − |G|−1 for σ 6= 1. Also, we see immediately that T (W ) ⊆ W . The determinant of T |W with respect
to the character basis is clearly the left-hand side of the desired equality. On the other hand, noting that

∑ (δτ − |G|−1) = 0,
τ∈G

we have

T (δσ − |G|−1 ) = ∑ f (σ τ −1 )(δτ − |G|−1 ) = ∑ ( f (σ τ −1) − f (σ ))(δτ − |G|−1),

τ∈G τ∈G
τ6=1

which has the desired coefficients. 

We omit a proof of the following.

L EMMA 1.5.19. For a field K and a finite group G, let V and W be K[G]-modules of finite K-
dimension. Suppose that there is a field extension L of K such that V ⊗K L ∼
= W ⊗K L as L[G]-modules.
Then V ∼
= W as K[G]-modules.

P ROPOSITION 1.5.20. Let F be an abelian extension with Galois group G of either Q or an imag-
inary quadratic field. Then EF ⊗Z Q ∼
= IG ⊗Z Q as Q[G]-modules.

P ROOF. By Proposition 1.2.4, we have EF ⊗Z R ∼

= V0 , where V0 is as in Notation 1.2.3. That is
V0 is a hyperplane in the R-span of the archimedean places of F, in this case consisting of the formal
sums with coefficients summing to zero (since F is either totally real or purely imaginary). Since E
has just one archimedean place, all of the places of F are conjugate under the action determined by
precomposition of a representative by the inverse of an element of G. Fixing an embedding φ : F → C
then provides an isomorphism V0 ∼= IG ⊗Z R, so EF ⊗Z R ∼ = IG ⊗Z R. By Lemma 1.5.19, we then have
that EF ⊗Z Q ∼
= IG ⊗Z Q. 

T HEOREM 1.5.21 (Brumer). Leopoldt’s conjecture holds for all finite abelian extensions of Q and
all finite abelian extensions of any imaginary quadratic field.

P ROOF. By Proposition 1.5.20, we may pick α ∈ EF be such that {σ (α) | σ ∈ G − {1}} is an

independent set of r units of F. Let φ = ι ◦ φ : K → C p , and consider the function f : G → C p defined
by f (σ ) = log p φ (σ −1 α). Since

∏ σ −1α = ±1,
σ ∈G
we have

∑ f (σ ) = 0.
σ ∈G
34 1. CLASS GROUPS AND UNITS

If ∑σ ∈G χ(σ ) f (σ ) = 0 for some nontrivial character χ ∈ Ĝ, then

∑ (1 − χ(σ )) f (σ ) = 0.
σ ∈G−{1}

Since 1 − χ(σ ) ∈ Q̄. By Theorem 1.5.17, we then have that the quantities f (σ ) for σ ∈ G − {1} are
Q-linearly dependent, and hence Z-linearly dependent. That is, there exist elements kσ ∈ Z, not all
zero, such that
∏ (σ α)kσ ∈ µ(F).
σ ∈G−{1}
This, however, contradicts our choice of α.
Now choose an ordering of G and form the matrix ( f (σ τ −1 ))σ ,τ∈G , the (σ , τ)-entry of which is
log p (φ ◦ τ)(σ −1 α). It then follows from Proposition 1.5.18a that this matrix has rank r = |G| − 1,
and the σ = 1 row and τ = 1 column are linearly dependent on the others. If we remove them, the
resulting r-by-r minor is the p-adic regulator matrix attached to the basis σ −1 α with σ ∈ G − {1} and
the embeddings φ ◦ τ for τ ∈ G − {1}. Thus R p (F) 6= 0, so Leopoldt’s conjecture holds for F. 
 CHAPTER 2

Module theory

2.1. Pseudo-isomorphisms

D EFINITION 2.1.1. For an integral domain R, a pseudo-null R-module is an R-module M with

annihilator AnnR (M) of height at least 2.

D EFINITION 2.1.2. Let R be an integral domain. An R-module homomorphism f : A → B is a

pseudo-isomorphism if it has pseudo-null kernel and cokernel.

The existence of a pseudo-isomorphism from one object to another is not in general an equivalence
relation on the category of finitely generated R-modules, as it is not symmetric.

E XAMPLE 2.1.3. The quotient of F p [x, y] by the maximal ideal (x, y) of height 2 is pseudo-null.
However, as (x, y) is not principal, there is no pseudo-isomorphism F p [x, y] → (x, y).

Nevertheless, we can make the following definition, which does provide an equivalence relation.

D EFINITION 2.1.4. We say that two modules A and B over an integral domain R are pseudo-
isomorphic if Ap ∼
= Bp for all height one prime ideals p of R.

N OTATION 2.1.5. We write A ' B if A and B are pseudo-isomorphic modules over an integral
domain R.

If there exists a pseudo-isomorphism from one R-module to another, then they are pseudo-isomorphic.
Recall that a prime ideal p of R lies in the support of a R-module A if and only if Ap 6= 0.

L EMMA 2.1.6. Let f : A → B be a pseudo-isomorphism of R-modules, where R is an integral

domain. Then A and B are pseudo-isomorphic.

P ROOF. Since localization is an exact functor, for any height one prime ideal p of R, we have an
exact sequence
0 → (ker f )p → Ap → Bp → (coker f )p → 0.
Since any prime ideal q in the support of AnnR (ker f ) or AnnR (coker f ) has height at least 2, while Rp
has Krull dimension one, we have qRp = Rp . Since (ker f )p and (coker f )p then have no prime ideals
in their support, they are both zero. 
35
36 2. MODULE THEORY

L EMMA 2.1.7. Let A and B be modules over a commutative ring R with A finitely presented, and
let S be a multiplicative subset of R. Then we have a canonical isomorphism

HomS−1 R (S−1 A, S−1 B) ∼

= S−1 HomR (A, B).

P ROOF. As
S−1 A ∼
= S−1 R ⊗R A,
adjointness of Hom and ⊗ yields

(2.1.1) HomS−1 R (S−1 A, S−1 B) ∼

= HomR (A, S−1 B).

The result now follows from (2.1.1) and

(2.1.2) HomR (A, S−1 R ⊗R B) ∼

= S−1 R ⊗R HomR (A, B).

To see that (2.1.2) holds, note first that it holds if A = R. It then holds for every free R-module A of
finite rank, as finite direct sums commute with Hom in the first variable and direct sums commute with
tensor products. In general, choose a resolution

P1 → P0 → A

with P0 and P1 finitely generated free R-modules, and use the fact that the contravariant functors of A
in question are exact on right-exact sequences of R-modules, in particular as S−1 R is R-flat. 

We recall from the theory of primary decomposition that every ideal I in a noetherian ring is a
minimal finite intersection of primary ideals, and the minimal ideals among the finitely many associated
primes of I that are the radicals of these primary ideals are the isolated primes of I.

L EMMA 2.1.8. Any finitely generated torsion module over a noetherian ring R has only finitely
many height one prime ideals in its support.

P ROOF. A prime ideal p is in the support of a finitely generated R-module M if and only if and
only if it contains I = AnnR (M). Any height one prime ideal containing I is an isolated prime in its
primary decomposition, so there can be only finitely many. 

From now on in this section, we use Λ to denote an integrally closed noetherian domain. Note that
the localization of Λ at any height one prime is still a integrally closed noetherian domain, and it has a
unique nonzero prime, so it is a DVR.

L EMMA 2.1.9. Let A and B be torsion Λ-modules. Let X be the finite set of height one prime ideals
in the support of A or B. Set
[
S = Λ− p.
p∈X
 2.1. PSEUDO-ISOMORPHISMS 37

Let f : A → B be a Λ-module homomorphism. Then f is a pseudo-isomorphism if and only if the

localized map
S−1 f : S−1 A → S−1 B
is an isomorphism.

P ROOF. First, note that X is indeed finite by Lemma 2.1.8. Let X = {p1 , . . . , pr }, where the pi
are distinct. As localization is an exact functor, it suffices to show that a finitely generated torsion
Λ-module M with height one support in X is pseudo-null if and only if S−1 M = 0.
If M is pseudo-null, then its annihilator has height at least 2, so is not contained in any prime ideal
of height one. Thus, for each 1 ≤ i ≤ r, there exists an element yi ∈ AnnΛ (M), with yi ∈
/ pi . Since there
also exists an element xi ∈ pi with xi ∈
/ p j for all j 6= i, we have

x = y1 x2 x3 · · · xr + x1 y2 x3 · · · xr + · · · + x1 x2 · · · xr−1 yr ∈ S ∩ AnnΛ (M)

and so S−1 M = 0.
Conversely, suppose that S−1 M = 0. Then Mp = 0 for all p ∈ S, and hence for all height one prime
ideals p, which is to say that for each p, there exists s ∈ AnnΛ (M) with s ∈
/ p, from which it follows
that AnnΛ (M) 6⊆ p for any height one prime ideal p. Therefore, AnnΛ (M) has height at least 2. 

P ROPOSITION 2.1.10. Let A be a finitely generated, torsion Λ-module. Then A is pseudo-isomorphic

Ls ki
to a direct sum i=1 Λ/pi with pi a height one prime of Λ and ki ≥ 1 for all 1 ≤ i ≤ s and for s ≥ 0.
Moreover, this decomposition is unique up to ordering.

P ROOF. Let S be the complement of the union of the height one prime ideals in the support of A.
Then S−1 A is a torsion module over the principal ideal domain S−1 Λ, so we have an isomorphism
s
−1 ∼ M −1
g: S A−
→ S (Λ/pki i )
i=1
with the pi and ki as in the statement. By Lemma 2.1.7, there exists a Λ-module homomorphism f : A →
ki
with S−1 f = g. By Lemma 2.1.9, the map f is a pseudo-isomorphism. The uniqueness is
Ls
i=1 Λ/pi
clear from the uniqueness in the structure theorem for finitely generated S−1 Λ-modules. 

The following is now clear.

C OROLLARY 2.1.11. Two finitely generated, torsion Λ-modules A and B are pseudo-isomorphic if
and only if there exists a pseudo-isomorphism f : A → B.

R EMARK 2.1.12. A module M over an integral domain R is torsion if and only if M(0) = 0, which
is to say that its localization at 0 is trivial. In particular, the R-torsion submodule of such a module M
is the kernel of the localization map to M(0) .
38 2. MODULE THEORY

L EMMA 2.1.13. Let A be a finitely generated Λ-module, let T denote its Λ-torsion submodule, and
set Z = A/T . Then there is a pseudo-isomorphism

A → T ⊕ Z.

P ROOF. Supposing without loss of generality that T 6= 0, let S be the complement of the union of
the height one primes in the support of T . Then S−1 Λ is a principal ideal domain, and by the structure
theorem for finitely generated modules over principal ideal domains, we have a projection map

ρ 0 : S−1 A → S−1 T,

which realizes the S−1 Λ-torsion submodule S−1 T of S−1 A as a direct summand. (To see that S−1 T
is the S−1 Λ-torsion submodule of S−1 A, note that it is torsion and the quotient S−1 Z = S−1 A/S−1 T
is S−1 Λ-torsion-free, as the fact that Z → Z(0) is injective implies that S−1 Z → S−1 Z(0) = Z(0) is as
well.) In other words, if we let ν : A → Z be the quotient map and ν 0 denote its localization, then
(ρ 0 , ν 0 ) : S−1 A → S−1 T ⊕ S−1 Z is an isomorphism.
By Lemma 2.1.7, there exist ρ ∈ Hom(A, T ) and s ∈ S such that ρ = sρ 0 . We consider the map

(ρ, ν) : A → T ⊕ Z.

Its localization is an isomorphism as multiplication by s is an isomorphism on S−1 T . Since the kernel

and cokernel are a subgroup and a quotient of T , respectively, they are supported on S, and the triviality
of their localizations at S implies their pseudo-nullity. Thus, (ρ, ν) is a pseudo-isomorphism. 

N OTATION 2.1.14. For a Λ-module A, let us use

A∗ = HomΛ (A, Λ)

to denote its Λ-dual.

Note that Lemma 2.1.7 tells us that (A∗ )p ∼

= (Ap )∗ for any prime ideal p of Λ, the latter module
being defined as
(Ap )∗ = HomΛp (Ap , Λp ),
so we simply write A∗p . Let Q denote the quotient field of Λ.

L EMMA 2.1.15. Let Z be a finitely generated, torsion-free Λ-module. The map Z → Z ∗∗ is an

injective pseudo-isomorphism.

P ROOF. For any height one prime ideal p, the modules Zp and Zp∗∗ are free, being finitely generated
torsion-free modules over the principal ideal domain Λp . Moreover, the natural map Zp → Zp∗∗ is an
isomorphism, being identified with the map a finite rank free Λp -module to its Λp -double dual. That
is, Z → Z ∗∗ is a pseudo-isomorphism, which is injective as Z is torsion-free. 
 2.1. PSEUDO-ISOMORPHISMS 39

L EMMA 2.1.16. Let A be a finitely generated Λ-module. Inside A∗(0) , we have

A∗ = A∗p ,
\

p∈X1

where X1 denotes the set of height 1 primes of Λ

P ROOF. Since A∗ is torsion-free, it sits inside each A∗p , hence in the intersection. Let f ∈ A∗(0) lie in
A∗p for each p of height one. Then f : A → Λp for all p, so f has image in Λ =
T
p∈X1 Λp . It follows that
f ∈ A∗ , hence the result. 

D EFINITION 2.1.17. We say that a finitely generated Λ-module A is reflexive if the natural map
A → A∗∗ is an isomorphism.

Note that a reflexive Λ-module is necessarily torsion-free, since the dual of a finitely generated
Λ-module is torsion-free.

L EMMA 2.1.18. A finitely generated, torsion-free Λ-module Z is reflexive if and only if Z is the
intersection of the Zp over all height one prime ideals p of Λ.

P ROOF. We note that Lemma 2.1.16 implies that

Z ∗∗ = Zp∗∗ ,
\

p∈X1

and we recall that the natural map Zp → Zp∗∗ is an isomorphism. As the diagram

Z / Z ∗∗

o
 
T
p∈X1 Zp
∼
/ Tp∈X Zp∗∗
1

commutes, we have the result. 

We have the following immediate corollary of Lemmas 2.1.16 and 2.1.18.

C OROLLARY 2.1.19. Let A be a finitely generated Λ-module. Then A∗ is reflexive.

We recall that a noetherian local ring Ω is regular if its maximal ideal m is generated by the terms of
a regular sequence (xi )di=1 , which is to say that xi is not a zero divisor in Ω/(x1 , . . . , xi−1 ) for 1 ≤ i ≤ d.
In this case, Ω has Krull dimension d. Equivalently, a noetherian local ring Ω is regular of Krull
dimension d if its maximal ideal can be generated by d elements, or if dimΩ/m m/m2 = d.

P ROPOSITION 2.1.20. Let Ω be a regular local ring of Krull dimension 2. Then every finitely
generated, reflexive Ω-module is free.
40 2. MODULE THEORY

P ROOF. Let A be a finitely generated and reflexive Ω-module. As Ω is regular, it has a regular
sequence so a principal prime ideal p = ( f ). We first claim that A/pA is a free Ω/p-module. As Ω/p is
regular of Krull dimension 1, its maximal ideal is principal (and it is a domain), so it is a DVR. Thus,
it suffices to show that A/pA is torsion-free over Ω/p. For this, note that the exact sequence
f
0 → HomΩ (A∗ , Ω) −
→ HomΩ (A∗ , Ω) → HomΩ (A∗ , Ω/p)

implies that the map

A∗∗ /pA∗∗ → HomΩ (A∗ , Ω/p)

is injective. As
HomΩ (A∗ , Ω/p) ∼
= HomΩ/p (A∗ /pA∗ , Ω/p)
is Ω/p-torsion free, the module A∗∗ /pA∗∗ is Ω/p-torsion free. But A is reflexive, so A∗∗ /pA∗∗ ∼
= A/pA,
proving the claim.
Next, let s = dimΩ/m A/mA, where m is the maximal ideal of Ω. By Nakayama’s Lemma, there
exists a minimal Ω-generating set of A with s elements, which is to say a surjective map π : Ωs → A.
Since A/pA is Ω/p-free, and by what we have just said of rank s, the induced surjection (Ω/p)s → A/pA
is necessarily an isomorphism. Therefore, multiplication by f is surjective on ker π, and Nakayama’s
lemma then tells us that ker π = 0. 

R EMARK 2.1.21. Any regular local ring is a UFD by a theorem of Auslander and Buchsbaum. In
particular, regular local rings are integrally closed domains.

T HEOREM 2.1.22. Let Λ be a regular local ring of Krull dimension at most 2. Let A be a finitely
generated Λ-module. Then there exists a pseudo-isomorphism
s
Λ/pki i
M
A → Λr ⊕
i=1

for some r, s ≥ 0 and height one primes pi and integers ki ≥ 1 for 1 ≤ i ≤ s. Moreover, r and s are
unique, and the prime powers are unique up to ordering.

P ROOF. Suppose first that A is torsion-free. By Lemma 2.1.15, the map A → A∗∗ is an injective
pseudo-isomorphism, and by Proposition 2.1.20, we have that A∗∗ ∼
= Λr for some r. This is then the
unique r for which there exists a pseudo-isomorphism A → Λr , being that it is then the dimension of
A(0) over the quotient field of Λ.
The result for torsion modules is Proposition 2.1.10. We can combine the torsion-free and torsion
cases by applying Lemma 2.1.13 and the decompositions in each case. The uniqueness follows from
the uniqueness in the two cases. 
 2.2. POWER SERIES RINGS 41

2.2. Power series rings

Let O be a complete commutative local noetherian ring with maximal ideal m and finite residue
field of characteristic p. We study the ring Λ = OJT K, beginning with the following analogue of the
division algorithm.

P ROPOSITION 2.2.1 (Division algorithm). Let f , g ∈ Λ, and suppose that f ∈

/ mΛ. Let n be the
largest integer such that f ∈ mΛ + (T n ). Then we may write

g = qf +r

for a unique q ∈ Λ and r ∈ O[T ] with deg r < n.

P ROOF. Suppose without loss of generality that n > 0. Let u ∈ O × be the coefficient of T n in f .
Let a ∈ Λ and b ∈ O[T ] be such that
f = aT n + b,
where deg b < n. Note that b ∈ mΛ by choice of n, and since a − u lies in the maximal ideal of Λ, we
have a ∈ Λ× . Let q00 ∈ Λ and r0 ∈ O[T ] be such that

g = q00 T n + r0 ,

where deg r0 < n. Setting q0 = a−1 q00 , we have

g = q0 aT n + r0 ≡ q0 f + r0 mod mΛ.

Let g1 = g − q0 f − r0 ∈ mΛ, and repeat the process to obtain q1 ∈ mΛ and r1 ∈ mO[T ] with deg r1 < n
and
g1 ≡ q1 f + r1 mod m2 Λ.
Note then that
g ≡ (q0 + q1 ) f + (r0 + r1 ) mod m2 Λ.
Recursively, we may then construct

q = q0 + q1 + q2 + · · · ∈ Λ and r = r0 + r1 + r2 + · · · ∈ O[T ]

such that g = q f + r and deg r < n.

As for uniqueness, if g = q0 f + r0 with r0 ∈ O[T ] and deg r0 < n, then

(q − q0 ) f + (r − r0 ) = 0.

We then need only show that if c ∈ Λ and d ∈ O[T ] with deg d < n satisfy c f + d = 0, then c = d = 0.
Suppose that this is not the case, and let k ≥ 0 be such that c, d ∈ mk Λ but not both c and d are contained
in mk+1 Λ. We see that c f is congruent to a multiple of T n modulo mk+1 Λ, which forces d ∈ mk+1 Λ,
as deg d < n. But then c f ∈ mk+1 Λ, and since f ∈
/ mΛ, this forces c ∈ mk+1 Λ, a contradiction. 
42 2. MODULE THEORY

D EFINITION 2.2.2. A distinguished (or Weierstrass) polynomial f ∈ Λ is a polynomial with leading

coefficient 1 that satisfies
f (T ) ≡ T deg f mod mΛ.

T HEOREM 2.2.3 (Weierstrass preparation). Let g ∈ Λ with g ∈

/ mΛ. Then there exist a unique
distinguished polynomial f and unit u ∈ Λ× such that

g = uf.

P ROOF. We begin with existence. Let n be the maximal such that g ∈ mΛ + (T n ), and let u0 ∈ O ×
be the coefficient of T n in g. Using the division algorithm, write

T n = qg + r

for some unique q ∈ Λ and r ∈ O[T ] with deg r < n. Since all of the terms of g of degree less than n
lie in m, we have r ∈ mΛ. If we set f = T n − r, then f is a distinguished polynomial. Moreover, the
constant coefficient q0 of q satisfies q0 u0 ≡ 1 mod T , so is a unit in O, and therefore q ∈ Λ× . Letting
u = q−1 , we have g = u f , as desired. The uniqueness of f and u is forced by the uniqueness of q and
r. 

We then have the following corollaries of the Weierstrass preparation theorem.

C OROLLARY 2.2.4. Suppose that O is a PID. Then the ring Λ is a unique factorization domain.

P ROOF. Let π ∈ m be a generator. For g ∈ π n Λ−π n+1 Λ, we may apply the Weierstrass preparation
theorem to factor π −n g into a polynomial f times a unit, and then use the fact that O[T ] is a UFD to
factor f into a product of irreducible polynomials, each of which is a Weierstrass polynomial times a
unit. This gives the desired factorization of g as a product of a power of π, finitely many irreducible
Weierstrass polynomials, and a unit. Clearly any other factorization is equivalent (up to unit and
ordering) to such a factorization. 

R EMARK 2.2.5. In fact, it is more generally true that if O is a regular local ring, then so is Λ =
OJT K. For such O, the ring OJT1 , T2 , . . . , Tr K is then of course a UFD as well.

Suppose from now on that O is the valuation ring of a finite extension of Q p . Let π ∈ O be
a uniformizer. We may view O as sitting inside C p . Given f ∈ Λ and a ∈ C p with |a| p < 1, the
evaluation f (a) converges to an element of C p .

C OROLLARY 2.2.6. Let g ∈ Λ be nonzero. There exist only finitely many a ∈ C p with |a| p < 1 such
that g(a) = 0.
 2.2. POWER SERIES RINGS 43

P ROOF. By Weierstrass preparation, we have g = π µ u f with µ ≥ 0, u ∈ Λ× , and f a Weiserstrass

polynomial. As u is a unit, one cannot have a ∈ C p with |a| p < 1 such that u(a) = 0. Therefore,
g(a) = 0 if and only if f (a) = 0, and so the result follows from the fact that f is a polynomial. 

C OROLLARY 2.2.7. Let g ∈ O[T ], and let f be a distinguished polynomial in O[T ] with f dividing
g in Λ. Then g/ f ∈ O[T ].

P ROOF. Let n = deg f . Suppose α ∈ C p is a root of f . If |α| p > 1, then 0 = | f (α)| p = |α|np > 1,
which is impossible. If |α| p = 1, then

0 = f (α) ≡ α n mod mC p ,

where mC p denotes the maximal ideal of C p , which is again impossible. If |α| p < 1, then set q = g/ f ∈
Λ. Since q(α) converges, we have g(α) = 0. Let O 0 denote the valuation ring of the splitting field of
f . We divide g and f by T − α inside O 0 [T ] and repeat the process with the resulting polynomials,
which we denote f1 and g1 . After n iterations, we have obtain fn = 1, and since f is monic, gn is the
polynomial q ∈ O[T ]. 

Next, let us consider ideals in Λ.

D EFINITION 2.2.8. Two elements f , g ∈ Λ are said to be relatively prime if the only elements in Λ
that divide both f and g are units.

L EMMA 2.2.9. Suppose that f , g ∈ Λ are relatively prime. Then ( f , g) has finite index in Λ.

P ROOF. Suppose that h ∈ ( f , g) is a polynomial of minimal degree (which exists by Weierstrass

preparation), and suppose it is exactly divisible by a power π n of π. Assume first that h has positive
degree. Let h0 ∈ Λ be defined by h = π n h0 . Without loss of generality, suppose that h0 does not divide
f . The division algorithm produces q ∈ Λ and r ∈ O[T ] with deg r < deg h0 such that f = qh0 + r. Then
π n r ∈ ( f , g), which forces r = 0 by the minimality of the degree of h. But then h0 divides f , which is a
contradiction, so h must be of degree 0.
So now, suppose that n is minimal such that π n ∈ ( f , g). At least one of f and g is not divisible by
π: suppose it is f , and assume without loss of generality that f is a distinguished polynomial. We have
(π n , f ) ⊆ ( f , g), but
Λ/(π n , f ) ∼
= (O/π n O)[T ]/( f¯),
where f¯ is the image of f in O/π n O[T ], and the quotient ring is a finite ring by the division algorithm,
as O has finite residue field. 

P ROPOSITION 2.2.10. Every prime ideal of Λ is one of 0, (π, T ), (π), or ( f ), where f is an

irreducible distinguished polynomial.
44 2. MODULE THEORY

P ROOF. Suppose that p is a nonzero prime ideal in Λ with p 6= (π). By the primality of p, there
then exists a distinguished polynomial f in p that is irreducible and not divisible by π . So choose
such an f : if p = ( f ), we are done. Otherwise, there exists g ∈ p with g ∈
/ ( f ), and therefore by
Lemma 2.2.9, there exists π n ∈ ( f , g) for some n ≥ 1. Since p is prime, we then have π ∈ p, and since
f ≡ T deg f mod π, we have T deg f ∈ p. Again, primality of p then forces T ∈ p, and finally, p = (π, T )
by the maximality of (π, T ). 

R EMARK 2.2.11. We have that Λ/(π) ∼

= (O/m)JT K, while Λ/( f ) for a distinguished polynomial
f is free of rank deg f over O.

L EMMA 2.2.12. A finitely generated Λ-module is pseudo-null if and only if it is finite.

P ROOF. Suppose that M is a finitely generated, pseudo-null Λ-module. To say that AnnΛ (M) has
height at least 2 is to say that it contains two relatively prime elements, hence has finite index in Λ. On
the other hand, if M is a finite Λ-module, then
\
AnnΛ (M) = AnnΛ (m),
m∈M

and AnnΛ (m) must be of finite index in Λ, since m generates a finite Λ-module isomorphic to Λ/ AnnΛ (m).
It follows that AnnΛ (M) has finite index in Λ, and therefore has height 2. 

It follows that a pseudo-isomorphism of Λ-modules, for O a valuation ring in a finite extension of

Q p , is a Λ-module homomorphism with finite kernel and cokernel.

T HEOREM 2.2.13 (Structure theorem for finitely generated Λ-modules). Let M be a finitely gener-
ated Λ-module. Then there exists a pseudo-isomorphism
s t
Λ/( fiki ) ⊕
M M
r
M→Λ ⊕ Λ/(π l j )
i=1 j=1

for some r, s,t ≥ 0, ki ≥ 1 and fi a distinguished irreducible Λ-polynomial for 1 ≤ i ≤ s, and l j ≥ 1 for
1 ≤ j ≤ t. Moreover, these quantities are unique up to reordering.

P ROOF. This follows directly from Theorem 2.1.22 and the fact that the height one prime ideals in
Λ are (π) and the ideals ( f ) for f an irreducible distinguished polynomial. 

2.3. Completed group rings

For a profinite group G, we use U Eo G to denote that U is an open normal subgroup of G.

 2.3. COMPLETED GROUP RINGS 45

D EFINITION 2.3.1. Let G be a profinite group, and let O be a commutative ring. We define the
completed O-group ring of G to be the inverse limit

OJGK = lim O[G/U]

←−o UE G

with respect to the quotient maps O[G/V ] → O[G/U] for V ≤ U.

R EMARK 2.3.2. In the case that G is finite, we have OJGK = O[G], the usual group ring.

We shall study completed group rings only for certain very special classes of rings O and profinite
groups G. In particular, let us assume that O is local and complete with respect to a maximal ideal m,
which is to say that
O∼
= lim O/mn O.
←−
n

R EMARK 2.3.3. Since O is complete with respect to the maximal ideal m, we have

OJGK ∼
= lim (O/mn O)[G/U]
←−o UE G
n≥0

D EFINITION 2.3.4. The augmentation ideal IG of OJGK is equal to

− O),
ε
ker(OJGK →

where ε is the augmentation map, the inverse limit of the O-linear maps O[G/U] → O that take every
group element to 1.

R EMARK 2.3.5. The map ε is surjective, and therefore it induces an isomorphism

OJGK/IG ∼
=O

We require the following lemma.

L EMMA 2.3.6. Let k be a field of characteristic p, and let G be a finite abelian p-group. Then k[G]
is a local ring with maximal ideal the augmentation ideal in k[G].

P ROOF. Suppose that G ∼

= ri=1 Z/pni Z for some ni ≥ 1 and r ≥ 0, and let gi be the inverse image
L

of a generator of the ith component under this isomorphism. It is easy to see that
n1 n2 nr
k[G] ∼ p p
= k[X1 , X2 , . . . , Xr ]/(X1 − 1, X2 − 1, . . . , Xrp − 1)
ni n
under the map that takes gi to Xi . Moreover, Xip − 1 = (Xi − 1) p i for each i since k has characteristic
p. Setting Ti = Xi − 1 the resulting ring
n1 n2 nr
k[T1 , T2 , . . . , Tr ]/(T1p , T2p , . . . , Trp )
46 2. MODULE THEORY

is local with maximal ideal (T1 , T2 , . . . , Tr ). (This is well-known, but note that if f ∈
/ (T1 , T2 , . . . , Tr ),
then f has nontrivial constant coefficient, and we may construct an inverse by successive approxi-
mation, working modulo higher and higher total degrees.) The inverse image of this ideal under our
isomorphism is the augmentation ideal of k[G]. 

We now let O be a commutative noetherian local ring that is complete with the topology defined
by its maximal ideal m.

P ROPOSITION 2.3.7. Let O be a complete commutative noetherian local ring with finite residue
field characteristic p, and let G be a topologically finitely generated abelian pro-p group. Then the
algebra OJGK is a local ring with maximal ideal mOJGK + IG .

P ROOF. We note that

OJGK/(m + IG ) ∼
= O/m,
so m + IG is maximal. If M is any maximal ideal of OJGK, then we have an injection

O/(M ∩ O) → OJGK/M,

which forces O/(M ∩ O) to be a field, hence M ∩ O to be maximal in O, and therefore M ∩ O to be

equal to m.
Moreover, we have
OJGK/mOJGK ∼
= kJGK,
where k = O/m. This follows from the fact that mOJGK is an inverse limit of a countable inverse
system of modules m · (O/mn )[G/U] with surjective maps, as this implies that lim1 of the system
←−
vanishes. (Here, the countability of the system is guaranteed by the assumption of finite generation on
G.)
The problem is reduced to showing that the augmentation ideal of kJGK is its only maximal ideal.
As the quotient of kJGK by a maximal ideal surjects onto the quotient of k[G/U] by the image of that
maximal ideal for every open normal subgroup U of G, it suffices to demonstrate our claim in the case
of a finite abelian p-group G. However, that result is just Lemma 2.3.6. 

For any r ≥ 0, recall that

OJT1 , T2 , . . . , Tr K ∼
= lim O[T1 , T2 , . . . , Tr ]/(T1n , T2n , . . . , Trn ).
←−
n

The latter O-modules in the inverse limit are free of finite rank over O, and so can be given the m-adic
topology, and the inverse limit then defines a topology on the power series ring itself.
The following lemma will be of use to us.
 2.3. COMPLETED GROUP RINGS 47

L EMMA 2.3.8. Suppose that O is a complete commutative local noetherian ring with finite residue
field, and let m denote its maximal ideal. Let r ≥ 1. The following sets of ideals provide bases of open
neighborhoods of 0 that all define the same topology on the ring R = O[T1 , T2 , . . . , Tr ]:
i. {Is,t | s,t ≥ 1}, where Is,t = ms R + (T1t , T2t , . . . , Trt ),
ii. {Mn | n ≥ 1}, where M = mR + (T1 , T2 , . . . , Tr ),
iii. {Js,t | s,t ≥ 1}, where

Js,t = ms R + (ωt (T1 ), ωt (T2 ), . . . , ωt (Tr ))

and we define
n
ωn (T ) = (T + 1) p − 1

for any T and any n ≥ 0.

In particular, R is isomorphic to the inverse limit of the quotients modulo the ideals in any of these sets.

P ROOF. To show that two of the sets of ideals define the same topology is exactly to show that
every ideal in each of the two sets contains an ideal in the other set. Note that

(T1 , T2 , . . . , Tr )(t−1)r+1 ⊆ (T1t , T2t , . . . , Trt ).

We then see that

t
I1,t ⊇ M(t−1)r+1 and J1,t ⊇ M(p −1)r+1 ,

and from this we obtain that

t
s
Is,t ⊇ I1,t ⊇ Ms((t−1)r+1) and Js,t ⊇ J1,t
s
⊇ Ms((p −1)r+1) .

On the other hand, we have

Mn ⊇ In,n and Mn ⊃ Jn,n ,

where the latter containment uses that

pn  n 
p j 2 n
ωn (Ti ) = ∑ Ti ∈ (pn Ti , pn−1 Ti p , pn−2 Ti p , . . . , Ti p ) ⊂ (mn + mn−1 Ti + · · · + Tin )R ⊂ Mn .
j=1 j

Therefore, the topology defined by the powers of M agrees both with the topologies defined by the
ideals Is,t and by the ideals Js,t . The final remark follows from the first part, as the set of Is,t defines the
natural topology on the power series ring. 
48 2. MODULE THEORY

T HEOREM 2.3.9. Let O be a complete commutative local noetherian ring with finite residue field
of characteristic p. Suppose that G ∼
= Zr for some r, and let {γi | 1 ≤ i ≤ r} be a generating set of G.
p
Then there is a unique topological isomorphism
∼
OJGK −
→ OJT1 , T2 , . . . , Tr K

that takes γi − 1 to Ti .
n
P ROOF. Let Un be the open subgroup of G generated by {γip | 1 ≤ i ≤ r} for some n ≥ 0. We note
that
O[G/Un ] → O[T1 , T2 , . . . , Tr ]/(ωn (T1 ), ωn (T2 ), . . . , ωn (Tr ))
via the map that takes γi to Ti + 1. Moreover, note that ωm (Ti ) divides ωn (Ti ) for m ≤ n, and these
isomorphisms between group and polynomial rings are compatible with the canonical quotient maps
on both sides. Since the groups Un form a basis of open neighborhoods of 0 in G, we have

OJGK ∼
= lim O[T1 , T2 , . . . , Tr ]/(ωn (T1 ), ωn (T2 ), . . . , ωn (Tr ))
←−
n

On the other hand, we have

OJT1 , T2 , . . . , Tr K ∼
= lim O[T1 , T2 , . . . , Tr ]/(T1n , T2n . . . , Trn ).
←−
n

Since O ∼= lim O/ms as well, that the two inverse limits are isomorphic follows from the equality of
←−
the topologies defined by the sets of ideals in (i) and (iii) of Lemma 2.3.8. 

R EMARK 2.3.10. We remark that the theorem implies that OJZkp K is noetherian, as a power series
ring in finitely many variables over a noetherian ring is noetherian, and Lemma 2.3.8 implies that it is
complete with respect to its unique maximal ideal.

2.4. Invariants of Λ-modules

Let O be the valuation ring of a p-adic field, and let π be a uniformizer of the maximal ideal m
of O. Set Λ = OJT K. We can use the structure theorem to construct invariants attached to a finitely
generated Λ-module.

D EFINITION 2.4.1. Let M be a finitely generated Λ-module, pseudo-isomorphic to

s t
Λ/( fiki ) ⊕
M M
Λr ⊕ Λ/(π l j )
i=1 j=1

for some r, s,t ≥ 0, ki ≥ 1 and fi a distinguished irreducible Λ-polynomial for 1 ≤ i ≤ s, and l j ≥ 1 for
1 ≤ j ≤ t.
 2.4. INVARIANTS OF Λ-MODULES 49

i. The λ and µ-invariants of M are

s t
λ (M) = ∑ ki deg fi and µ(M) = ∑ l j,
i=1 j=1

respectively.
ii. The characteristic polynomial of M is
s
char(M) = π µ(M) ∏ fiki ,
i=1

and the characteristic ideal of M is the ideal charΛ (M) of Λ generated by char(M).

We remark that the characteristic polynomial is multiplicative in exact sequences, as follows from
the following lemma.

L EMMA 2.4.2. Let

ι π
0→A→
− B−
→C →0
be a short exact sequence of finitely generated, torsion Λ-modules. Then char(B) = char(A) char(C).

P ROOF. Let X be the set of height one prime ideals in the support of B, and let S = Λ −
S
p∈X p.
Identifying S−1 A, S−1 B, and S−1C with direct sums of quotients of S−1 Λ by height one prime ideals,
that the characteristic ideals of these modules are multiplicative in S−1 Λ is a standard result in the
theory of modules over a principal ideal domain. The lemma follows easily from this. 

We next consider the quotients of finitely generated, torsion Λ-modules. Recall that
n
ωn (T ) = (T + 1) p − 1

for any n ≥ 0.

R EMARK 2.4.3. Suppose Γ is a procyclic group isomorphic to Z p , and let γ ∈ Γ be a topological

∼
generator. Let Γn denote the quotient of Γ of order pn . Recall that we have an isomorphism OJΓK −
→Λ
n
that takes γ − 1 to T . Then γ p − 1 is taken to ωn , so we have that

O[Γn ] ∼
= Λ/(ωn ).

We then have that

Λ∼
= lim Λ/(ωn ).
←−
n
n
Moreover, the quotient M/ωn M of a Λ-module M is identified with the Γ p -coinvariant group MΓ pn of
M.
50 2. MODULE THEORY

L EMMA 2.4.4. If M is a finitely generated Λ-module, then the canonical maps

∼ ∼
M− → lim M/(π m , ωn )M
→ lim M/ωn M −
←− ←−
n m,n

are isomorphisms.

P ROOF. Since Λ is noetherian and M is finitely generated, there exists a presentation of M as a

Λ-module:
Λr → Λs → M → 0
for some r, s ≥ 0. Since tensor product is right exact and

Λ/(π m , ωn ) ⊗Λ M ∼
= M/(π m , ωn )M,

we have that
(Λ/(π m , ωn ))r → (Λ/(π m , ωn ))s → M/(π m , ωn )M → 0.
is exact as well. As the inverse limit is exact on finite groups, the resulting inverse limit

Λr → Λs → lim M/(π m , ωn )M → 0
←−
m,n

is exact, so there is a canonical isomorphism

∼
→ lim M/(π m , ωn )M.
M−
←−
m,n

Since the latter map factors as

M → lim M/ωn M → lim M/(π m , ωn )M,

←− ←−
n m,n

we are done if we can show the second of these maps is injective. By left exactness of the inverse limit,
this will follow from the injectivity of the maps

Mn → lim Mn /π m Mn ,
←−
m

where we have set Mn = M/ωn M. For this, note that Nakayama’s Lemma tells us that A =
T mM =
mπ n
0, since πA = A. 

R EMARK 2.4.5. The proof of Lemma 2.4.4 goes through with ωn replaced by any sequence fn of
distinguished polynomials with fm | fn for m ≤ n and fm 6= fn if m < n.

For n ≥ m, we set ωn,m = ωn /ωm . Let us also set ωn,−1 = ωn .

L EMMA 2.4.6. Let M be a finitely generated torsion Λ-module containing no elements of finite
order. Then there exists an integer n0 ≥ −1 such that ωn,n0 M = pn−n0 M for all n ≥ n0 .
 2.4. INVARIANTS OF Λ-MODULES 51

P ROOF. Since M has no p-torsion, we have µ(M) = 0. The structure theorem implies the existence
of a pseudo-isomorphism
s
M
φ: M→ Λ/( fi )
i=1
with fi distinguished, and which must be injective as, again, M has no p-torsion. As ∏si=1 fi annihilates
m
M, we have that T λ (M) annihilates M/πM. It follows that (T + 1) p acts as the identity on M/πM for
any m with pm ≥ λ (M). Fix such an m, and let n0 be an integer such that pn0 ≥ pm (e + 1), where e + 1
is the ramification index of π in O.
For θ ∈ EndΛ (M) given by the action of T + 1, the exact sequence
π
0 → EndΛ (M) −
→ EndΛ (M) → EndΛ (M/πM)

implies that
m
θ p − 1 ∈ π EndΛ (M).
For any n ≥ n0 , we then have
n m n−m n−m
θ p − 1 = ((θ p − 1) + 1) p − 1 ∈ (π p , pπ) EndΛ (M) = pπ EndΛ (M).
n
Let ψ ∈ EndΛ (M) with θ p = 1 + pπψ. Since
p−1
n
ωn+1,n = ∑ (T + 1)cp ,
c=0
we have that ωn+1,n acts on M as
p−1 p−1
∑ (1 + pπψ)c ∈ p + ∑ cpπψ + p2 EndΛ(M) ⊆ p + pπ EndΛ(M).
c=0 c=0

For M̄ = M/pπM, we therefore have that ωn+1,n · M̄ = p · M̄. This forces

ωn+1,n · M = p · M

by Nakayama’s lemma, which implies the result. 

We now have the following result on the orders of quotients of finitely generated, torsion Λ-
modules.

T HEOREM 2.4.7. Let M be a finitely generated, torsion Λ-module, and let n0 ≥ −1 be such that
char(M) and ωn,n0 are relatively prime for all nonnegative n ≥ n0 . Set λ = λ (M) and µ = µ(M). Let q
denote the order of the residue field k of O, and let e denote the ramification index of O over Z p . Then
there exists an integer ν ∈ Z such that
n µ+neλ +ν
|M/ωn,n0 M| = q p

for all sufficiently large n ≥ 0.

52 2. MODULE THEORY

P ROOF. Our proof consists of four steps. In the first, we treat the case of finite M. In the second, we
reduce to the case of direct sums of quotients of Λ. In the third, we treat the quotients of Λ by powers
of π, and in the fourth, we treat the quotients of Λ by distinguished polynomials. For simplicity of
notation, let us set ωn0 = ωn,n0 .
Step 1: Note first that if M is finite, then M/ωn0 M ∼
= M for n sufficiently large, as follows from
ν
Lemma 2.4.4, noting Remark 2.4.5. In this case, q is then just the order of M. To see that ν is an
integer and not just a rational number, note that M has a filtration {π i M | i ≥ 0} and the graded quotients
π i M/π i+1 M are finite-dimensional k-vector spaces, so of order a power of q. It follows that
∞
|M| = ∏ |π i M/π i+1 M|
i=0

is a power of q as well.
Step 2: In the general case, consider the map
s t
Λ/( fiki ) ⊕
M M
φ: M→N= Λ/(π l j )
i=1 j=1

constructed in Theorem 2.2.13. It has finite kernel and cokernel, and the induced maps

φn : M/ωn0 M → N/ωn0 N

fit into a commutative diagram

0 0 0

 ωn0  
ker φ / ker φ / ker φn

 ωn0  
M /M / M/ωn0 M /0
φ φ φn
 ωn0  
0 /N /N / N/ωn0 N /0

 ωn0  
coker φ / coker φ / coker φn /0

  
0 0 0

where the map ωn0 : N → N is injective since one cannot have ωn0 g ∈ ( fiki ) (or in (π l j )) for some i
(resp., j) unless g ∈ ( fiki ) (resp., (π l j )) as ωn0 is relatively prime to each fi by assumption (and to π by
definition). Now, for sufficiently large n, we have that multiplication by ωn0 is the zero map on ker φ
 2.4. INVARIANTS OF Λ-MODULES 53

and coker φ , as ker φ and coker φ are finite. Therefore, the snake lemma tells us that, for such n, we
have coker φ ∼= coker φn and an exact sequence

0 → ker φ → ker φn → coker φ → 0.

Defining η ≥ 0 by
| ker φn |
qη = | ker φ | = ,
| coker φn |
we have that
s t
|M/ωn0 M| = qη ∏ |Λ/(ωn0 , fiki )| · ∏ |Λ/(ωn0 , π l j )|
i=1 j=1

for the same sufficiently large n. This reduces the theorem to modules of the form M = Λ/(π l ) for
some l ≥ 1 or M = Λ/( f ) with f a (a power of an irreducible) distinguished polynomial relatively
prime to every ωn0 .
Step 3: Suppose now that M = Λ/(π l ) for some l ≥ 1. We then have

M/ωn0 M = Λ/(ωn0 , π l ) ∼
= (O/π l O)JT K/(ωn0 ),
n
Since ωn0 is a distinguished polynomial of degree pn − pn0 , the latter ring is isomorphic to (O/π l O) p
as an O-module. We therefore have that
n l−pn0 l
|M/ωn0 M| = q p .

Note that µ(M) = l, and we can take ν = −pn0 l for this M.

Step 4: Finally, suppose that M = Λ/( f ) for some distinguished polynomial f relatively prime to
every ωn0 . By Lemma 2.4.6, we have that there exists n1 ≥ n0 such that

ωn,n1 M = pn−n1 M

for all n ≥ n1 . We also have an exact sequence

ωn,n
0 → M/ωn0 1 M −−−→
1
M/ωn0 M → M/ωn,n1 M → 0,

and therefore we have

= M/pn−n1 M ∼
M/ωn,n1 M ∼ = (O/π e(n−n1 ) O)λ ,
the latter isomorphism being of O-modules. Defining ν ∈ Z by

qν = |M/ωn0 1 M| · q−n1 eλ ,

we then have
|M/ωn0 M| = qneλ +ν ,
as desired.
54 2. MODULE THEORY

We next wish to consider results which give us conditions that allow us to compute invariants of
Λ-modules from their quotients. For this, the following lemma is useful.

L EMMA 2.4.8. Let φ : M → N be a pseudo-isomorphism of Λ-modules, and let f ∈ Λ be a distin-

guished polynomial. Then the induced map φ f : M/ f M → N/ f N is also a pseudo-isomorphism, and
moreover, we have

| ker φ f | ≤ | ker φ || coker φ | and | coker φ f | ≤ | coker φ |.

Similarly, using A[ f ] to denote the kernel of f : A → A for any Λ-module A, the induced map f φ : M[ f ] →
N[ f ] is also a pseudo-isomoprhism, and we have

| ker f φ | ≤ | ker φ | and | coker f φ | ≤ | ker φ || coker φ |.

P ROOF. Consider first the diagram

/ M/ ker φ /N / coker φ /0
φ
0
f f f
  
/ M/ ker φ /N / coker φ / 0.
φ
0

The snake lemma then yields an exact sequence

(2.4.1) 0 → (M/ ker φ )[ f ] → N[ f ] → (coker φ )[ f ] → M/( f M + ker φ ) → N/ f N → coker φ f → 0.

The kernel of φ f has order at most the products of the orders of the kernels of the maps M/ f M →
M/( f M + ker φ ) and M/( f M + ker φ ) → N/ f N. The first clearly has order at most | ker φ |, and by
(2.4.1), the second has order at most | coker φ |. The statement on coker φ f is also clear from the exact
sequence.
As for f φ , the snake lemma applied to

/ ker φ /M / M/ ker φ /0
φ
0
f f f
  
/ ker φ /M / M/ ker φ / 0.
φ
0

yields exactness of

0 → (ker φ )[ f ] → M[ f ] → (M/ ker φ )[ f ] → ker φ / f ker φ ,

Together with (2.4.1), this implies that f φ has finite kernel contained in ker φ and finite cokernel of
order at most | ker φ | · | coker φ |. 
 2.4. INVARIANTS OF Λ-MODULES 55

D EFINITION 2.4.9. Let A be a finitely generated O-module. The π-rank rπ (A) of A is the dimension
of A/πA as a vector space over k = O/πO.

P ROPOSITION 2.4.10. Let M be a finitely generated, torsion Λ-module. Then µ(M) = 0 if and only
if the quantities rπ (M/ωn M) are bounded as n varies.

P ROOF. Consider a pseudo-isomorphism

s t
Λ/( fiki ) ⊕
M M
φ: M→N= Λ/(π l j ).
i=1 j=1

Then φωn : M/ωn M → N/ωn N has finite kernel and cokernel of bounded order by Lemma 2.4.8, so it
suffices to check the result for N. Note that the π-rank of Λ/( f , ωn ) for f a distinguished polynomial is
bounded by deg f , since Λ/( f ) ∼= O deg f as an O-module. On the other hand, Λ/(π l , ωn ) is isomorphic
n
to (O/π l ) p as an O-module, so has unbounded π-rank. 

Similarly, we have the following proposition for the λ -invariant.

P ROPOSITION 2.4.11. Let M be a finitely generated, torsion Λ-module. Then λ (M) is equal to the
following quantities:
i. rankO M and
ii. the maximal integer λ such that M has a quotient isomorphic to (O/π n O)λ as an O-module for
every n.

P ROOF. Consider a pseudo-isomorphism

s t
Λ/( fiki ) ⊕
M M
φ: M→N= Λ/(π l j ).
i=1 j=1

Let n > µ(M). Then Λ/(π l j ) has trivial O-rank and no quotient of the form O/π n O, since n > l j . On
the other hand, Λ/( fiki ) is isomorphic to O ki deg fi as an O-module by Remark 2.2.11, so has a quotient
of the form (O/π n O)m for exactly those m ≤ ki deg fi . Therefore, the result holds for N.
By definition, O-rank is not affected by pseudo-isomorphism, so λ (M) = rankO M. Moreover,
if λ = rankO M, then the quotient of M modulo its π-power torsion subgroup is a finitely gener-
ated torsion-free O-module of rank λ , hence is isomorphic to O λ and has a quotient isomorphic to
(O/π n O)m for exactly those m ≤ λ . 

Finally, for finitely generated Λ-modules which are not necessarily Λ-torsion, we have the follow-
ing result on Λ-ranks.
56 2. MODULE THEORY

P ROPOSITION 2.4.12. Let M be a finitely generated Λ-module. Then we have

rankΛ (M) = rankO (M/T M) − rankO (M[T ]).

Moreover, we have
rankO (M/ωn M) = pn rankΛ (M) + c
for some c ≥ 0 for all sufficiently large n.

P ROOF. Again consider a pseudo-isomorphism

s t
Λ/( fiki ) ⊕
M M
r
φ: M→N=Λ ⊕ Λ/(π l j ).
i=1 j=1

Then rankΛ (M) = rankΛ (N), and by Lemma 2.4.8, we have

rankO (M/T M) = rankO (N/T N) and rankO (M[T ]) = rankO (N[T ]),

or more strongly, that M[T ] → N[T ] is a pseudo-isomorphism.

Given this, the proof of the first part is reduced to case that M = N. Since Λ/T Λ has O-rank 1 and
Λ/( f ) for a distinguished polynomial f has

Λ/( f , T ) ∼
= O/ f (0)O,

we have that the O-rank of the latter module is nonzero, and then equal to 1, if and only if T divides
f . Finally, Λ/(T, π l ) ∼
= O/π l O for l ≥ 1 and so has trivial O-rank. It follows that rankO (N/T N) =
r + s, where s is the number of fi equal to T . As for N[T ], note that Λ[T ] = 0 and Λ/(π l )[T ] = 0,
while Λ/( f )[T ] is nonzero, and then of O-rank 1, if and only if T divides f . Therefore, we have
rankO N[T ] = s, and part a follows.
We note that rankOJωn K (M) = pn rankΛ (M), since Λ has rank pn over OJωn K. The first part applied
with T replaced by ωn then implies

rankO (M/ωn M) = pn rankΛ (M) + rankO (M[ωn ]).

It suffices then to show that rankO (M[ωn ]) is bounded in n. But this follows as ωn,m is relatively prime
to charΛ (M) for n sufficiently large for all m. 

2.5. Pontryagin duality

Let A be a locally compact, Hausdorff topological abelian group.

D EFINITION 2.5.1. The Pontryagin dual of A is defined to be the topological group

A∨ = Homcts (A, R/Z)

 2.5. PONTRYAGIN DUALITY 57

with the compact-open topology, which is to say, with basis of open sets of the form

B(K,U) = { f ∈ A∨ | f (K) ⊆ U},

where K ⊂ A is compact and U ⊂ R/Z is open.

Of course, if f : A → B is a continuous map of locally compact, Hausdorff abelian groups, then

there is a natural map f ∨ : B∨ → A∨ given by f ∨ (ϕ) = ϕ ◦ f .
The following is the key theorem regarding the Pontryagin dual, which we state without proof.

T HEOREM 2.5.2 (Pontryagin duality). Let L denote the category of locally compact, Hausdorff
topological abelian groups, let C denote the category of compact, Hausdorff topological abelian
groups, and let D denote the category of discrete topological abelian groups. Then the Pontryagin
dual provides a self-inverse contravariant functor from L to its itself. Moreover, it induces contravari-
ant equivalences of categories between C and D in both directions.

R EMARK 2.5.3. If A is a profinite or discrete torsion, then in fact

A∨ = Homcts (A, Q/Z),

while if A is pro-p or discrete p-torsion, then we have

A∨ = Homcts (A, Q p /Z p ).

Moreover, we note that if A is discrete, then every homomorphism from it is continuous. On the
other hand, if A is a finitely generated Z p -module, then every Z p -linear homomorphism is continuous,
so
A∨ = HomZ p (A, Q p /Z p ).

R EMARK 2.5.4. If A has the additional structure of a topological G-module for a profinite group
G, then A∨ has the continuous G-action given by

(g · f )(a) = f (g−1 a)

for g ∈ G, f ∈ A∨ and a ∈ A.

R EMARK 2.5.5. Pontryagin duality induces a nondegenerate continuous pairing

A × A∨ → Q p /Z p , (a, f ) 7→ f (a).

If A is also a topological G-module, then the latter pairing is G-equivariant.

Here is another interesting result.

P ROPOSITION 2.5.6.
58 2. MODULE THEORY

a. If A is a compact, Hausdorff topological Z p -module, then A is profinite.

b. If A is a discrete topological Z p -module, then A is Z p -torsion.

P ROOF. Let us start with part b. Since A is discrete, every element a ∈ A has pn a = 0 for some
n ≥ 0 by continuity of the action. As for part a, we note that the dual of a compact Z p -module is A a
discrete Z p -module, hence Z p -torsion. Then A∨ is the direct limit of the finite submodules generated
by any finite set of its elements, so A is the topologically the inverse limit of the Pontryagin duals of
those submodules, and therefore A is profinite. 

C OROLLARY 2.5.7. Every finite topological Z p -module has the discrete topology.

E XAMPLE 2.5.8. Since Z p is procyclic, a continuous homomorphism from it is determined by

where 1 is sent. Since Z p is a free pro-p group, we can send 1 to any element. Therefore, we have
Z∨p = Q p /Z p .

D EFINITION 2.5.9. We say an locally compact module over a profinite ring R is cofinitely generated
if its Pontryagin dual is a finitely generated right R-module.

2.6. Iwasawa adjoints

We continue to suppose that Λ = OJT K for a valuation ring O of a p-adic field with uniformizer π.
Let F denote the quotient field of O. We will be most interested in Pontryagin duals of Λ-modules.

D EFINITION 2.6.1. Let ι : Λ → Λ be the unique continuous O-linear ring homomorphism satisfy-
ing ι(T ) = (T + 1)−1 − 1.

We can convert the canonical right action on the Pontryagin dual of a Λ-module to a left action
using an involution, as follows.

P ROPOSITION 2.6.2. If M is a locally compact, Hausdorff topological Λ-module, then M ∨ is as

well, with respect to the action

(2.6.1) (λ · ϕ)(m) = ϕ(ι(λ )m)

for λ ∈ Λ, m ∈ M, and ϕ ∈ M ∨ .

Let s ≥ 0 be such that π s generates the different of O/Z p . Then the O-balanced pairing

(2.6.2) O × O → Z p, (x, y) 7→ TrF/Q p (π −s xy)

is perfect. For a locally compact, Hausdorff topological Λ-module M, we have a left Λ-module struc-
ture on HomO (M, F/O) as in (2.6.1), with ϕ now in HomO (M, F/O).
 2.6. IWASAWA ADJOINTS 59

P ROPOSITION 2.6.3. For every finitely or cofinitely generated O-module A, there exists an isomor-
phism
A∨ ∼
= HomO (A, F/O).
These can be chosen to be natural in A in a manner that is canonical up to the choice of uniformizer π
of O. Moreover, if A is a Λ-module, then the isomorphism is of Λ-modules.

P ROOF. The perfect pairing of (2.6.2) yields an isomorphism O ∼

= Hom(O, Z p ) and therefore the
composite O-module isomorphism

F/O ∼
= O ⊗Z p Q p /Z p ∼
= HomZ p (O, Z p ) ⊗Z p Q p /Z p ∼
= HomZ p (O, Q p /Z p ).

Since A is (co)finitely generated over Z p , we have the following Λ-module isomorphisms

A∨ ∼
= HomZ p (A, Q p /Z p ) ∼
= HomO (A, HomZ p (O, Q p /Z p )) ∼
= HomO (A, F/O),

and naturality is easily checked. 

We have the following analogue of Proposition 2.5.6.

P ROPOSITION 2.6.4.
a. Every compact Λ-module is an inverse limit of finite Λ-modules.
b. Every discrete Λ-module is a direct limit of finite Λ-modules.

P ROOF. By Pontryagin duality, it suffices to prove part b. For this, we again note that the continuity
of the Λ-action on a discrete module M ensures that, for any m ∈ M, the annihilator AnnΛ (m) is an
open ideal of Λ. But then M is the union of its finite Λ-submodules Λ · m for m ∈ M. 

Note that if M is a finitely generated Λ-module, we endow it with the topology under which
(π m , ωn )M forms a basis of open submodules of M.

D EFINITION 2.6.5. Let M be a finitely generated, torsion Λ-module, and set Mn = M/ωn,m M for
n ≥ m and some fixed m ≥ −1 with ωn,m relatively prime to char(M) for all n. Set

α(M) = lim Mn∨ ∼

= (lim M )∨ ,
←− −→ n
n n
where Mn → Mn+1 is induced by the map m 7→ ωn+1,n m on M. Then the Λ-module α(M) is called the
Iwasawa adjoint to M.

R EMARKS 2.6.6.
a. We leave it to the reader to check that the definition of α(M) does not depend on m.
b. If φ : M → N is a Λ-module homomorphism, where M and N are finitely generated and Λ-
torsion, then we obtain a natural Λ module homomorphism α(φ ) : α(N) → α(M).
60 2. MODULE THEORY

L EMMA 2.6.7. The contravariant functor α is left exact.

P ROOF. To see the exactness, note that

Mn ∼
= M ⊗Λ Λ/(ωn,m ),

the tensor product is right exact, the Pontryagin dual is an exact contravariant functor, and the inverse
limit is exact on finite abelian groups. 

L EMMA 2.6.8. If M is a finite Λ-module, then α(M) = 0.

P ROOF. Since M is finite, the map ωn,m : M → M is zero for n sufficiently large relative to a fixed
m. The result follows. 

L EMMA 2.6.9. If M is a finitely generated, torsion Λ-module with µ(M) = 0, then there are natural
isomorphisms
α(M) ∼
= HomZ p (M, Z p ) ∼
= HomO (M, O)
as Λ-modules. Here, Λ acts on both HomZ p (M, Z p ) and HomO (M, O) by

(λ · φ )(m) = φ (ι(λ )m).

P ROOF. Let N be the p-power torsion submodule of M. By Lemma 2.6.7 and Lemma 2.6.8, the
map α(M/N) → α(M) is an isomorphism, so we can and do suppose that M is p-torsion-free.
Since M is finitely generated over Z p , we have that for sufficiently large m and n ≥ m that ωn,m acts
on M by multiplication by pn−m by Lemma 2.4.6. Therefore, we see that

α(M) ∼
= lim (M/pn M)∨ ∼
= lim HomZ p (M/pn M, Z/pn Z) ∼
= HomZ p (M, Z p ).
←− ←−
n n
For the other isomorphism, we note that

HomZ p (M, Z p ) ∼
= HomZ p (M ⊗O O, Z p ) ∼
= HomO (M, Hom(O, Z p )) ∼
= HomO (M, O),

the latter isomorphism using the pairing of (2.6.2), and all of these isomorphisms are of Λ-modules. 

P ROPOSITION 2.6.10. Let φ : M → N be a pseudo-isomorphism of finitely generated, torsion Λ-

modules. Then the induced map α(φ ) : α(N) → α(M) an injective pseudo-isomorphism.

P ROOF. As the inverse limit is exact on finite modules, in order to show that α(φ ) is a pseudo-
isomorphism it suffices to show that the maps Nn∨ → Mn∨ have kernel and cokernel of bounded order.
By exactness of the Pontryagin dual, this reduces to proving that Mn → Nn has kernel and cokernel of
bounded order, which follows from Lemma 2.4.8.
Finally, by Lemma 2.6.7, we have that the sequence
α(φ )
0 → α(coker φ ) → α(N) −−−→ α(M)
 2.6. IWASAWA ADJOINTS 61

is exact. The injectivity of α(φ ) then follows from Corollary 2.6.8. 

D EFINITION 2.6.11. For a Λ-module M, we let M ι denote the Λ-module that is M as a set but on
which the Λ-action ·ι is
λ ·ι m = ι(λ )m

for λ ∈ Λ and m ∈ M.

L EMMA 2.6.12.
a. For any positive integer `, we have α(Λ/(π l )) ∼
= Λ/(π l ).
b. For any distinguished polynomial f , we have α(Λ/( f )) ∼
= Λ/(ι( f )).

P ROOF. For part a, set γ = T + 1 and let M = Λ/(π l ). Then any element in Mn = M/ωn M (taking
m = −1) may be uniquely written as
pn −1
f= ∑ ai γ i
i=0
n
modulo ωn = γ p −1, for some a i ∈ O/π l O for 0 ≤ i ≤ pn −1. Let us identify Mn∨ with HomO (Mn , F/O)
as in Proposition 2.6.3. We define a map

ψn : Mn → Mn∨

by setting
ai
ψn ( f )(γ i ) =
,
πl
and extending O-linearly. Then ψn is clearly an injective homomorphism, and it is also easily seen that
the ψn (γ i ) form a O-basis of Mn∨ , so ψn is surjective as well. Moreover, ψn is a map of Λ-modules as
ai−1
ψn (γ f )(γ i ) = = ψn ( f )(γ i−1 ) = (γ · ψn ( f ))(γ i ).
πl
The diagram
/ M∨
ψn+1
Mn+1 n+1
∨
ωn+1,n
 
/ Mn∨
ψn
Mn
commutes since
p−1
∨ nj
ωn+1,n (ψn+1 ( f ))(γ i ) = ψn+1 ( f )(ι(ωn+1,n )γ i ) = ∑ ψn+1( f )(γ i+p ) = ψn ( f )(γ i ).
j=0

In the inverse limit, we obtain α(Λ/(π l )) ∼

= Λ/(π l ).
62 2. MODULE THEORY

For part b, suppose that M = Λ/( f ) with f a distinguished polynomial of degree d. Let us define
ε : Λ → O by setting ε(g) equal to the coefficient of T d−1 in r, where r ∈ O[T ] is the unique polynomial
of degree less than d with g = q f + r for some q ∈ Λ. We then define

θ : Λ/( f ) → HomO (Λ/( f ), O)ι

by
θ (ḡ)(h̄) = ε(gh),
where ḡ, h̄ ∈ Λ/( f ) and g, h ∈ Λ are lifts of ḡ and h̄ respectively. This is clearly well-defined, and
moreover it is a Λ-module homomorphism, since

θ (λ ḡ)(h̄) = ε(λ gh) = θ (ḡ)(λ h̄) = (λ · θ (ḡ))(h̄).

If r ∈ O[T ] is nonzero of degree less k than d, then letting r̄ denote the image of r in Λ/( f ), we have

θ (r̄)(T d−1−k ) = ε(T d−1−k r) 6= 0,

/ ker θ , so θ is injective. A count of O-ranks now tells us that α has finite cokernel.
which means that r̄ ∈
In fact, θ is surjective, as for any 0 ≤ k ≤ d − 1 and g = ∑d−1 i
i=0 ai T , we have that
k d−1
T k g − ∑ ad− j T k− j f ≡ ∑ ai−k T i mod π,
j=1 i=k

since f is distinguished, and hence

θ (T k )(ḡ) ≡ ad−1−k mod π.

Since the functions φk ∈ HomO (Λ/( f ), O)ι with φk (g) = ad−1−k generate HomO (Λ/( f ), O)ι and agree
with the θ (T k ) modulo π, the θ (T k ) do as well by Nakayama’s lemma. In other words, θ is an
isomorphism Λ/( f ) → α(Λ/( f ))ι , and part b follows as (Λ/( f ))ι ∼
= Λ/(ι( f )). 

T HEOREM 2.6.13. Let M be a finitely generated, torsion Λ-module. Then α(M) is a finitely gener-
ated, torsion Λ-module that is pseudo-isomorphic to M ι . Moreover, α(M) contains no nontrivial finite
Λ-submodules.

P ROOF. Consider a pseudo-isomorphism

s t
Λ/( fiki ) ⊕
M M
θ: N= Λ/(π l j ) → M,
i=1 j=1

which exists by the structure theorem and Proposition 2.1.11. Note that α(θ ) : α(M) → α(N) is an
injective pseudo-isomorphism. If we can show that α(N) is pseudo-isomorphic to N ι , then clearly
α(M) will be pseudo-isomorphic to M ι , as pseudo-isomorphism is an equivalence relation on finitely
generated, torsion Λ-modules. Moreover, if α(N) has no nonzero finite Λ-submodules, then neither
 2.7. THE GROUP RING OF A CYCLIC p-GROUP 63

does α(M), being isomorphic to a submodule of α(N). By the additivity of the adjoint functor, it
then suffices to assume that M is a quotient of Λ by a height one prime ideal, but this is covered by
Lemma 2.6.12. 

2.7. The group ring of a cyclic p-group

Let us suppose that G is a cyclic group of order p. In this section, we wish to study the structure
theory of modules over Z p [G] that are finitely generated, free Z p -modules. From our study of modules
over Λ = Z p JT K (or representation theory over Q p ), we are easily able to classify such modules up to
pseudo-isomorphism.
Let NG ∈ Z p [G] denote the norm element, and let X = Z p [G]/NG , which is noncanonically isomor-
phic to the augmentation ideal IG via the map x 7→ (g − 1)x, for g ∈ G a generator.

L EMMA 2.7.1. Let A be a finitely generated Z p [G]-module, where G is cyclic of order p. Then
there are s,t ≥ 0 and a homomorphism

φ : A → X s ⊕ Ztp

with finite kernel and cokernel, and ker φ = 0 if and only if A is p-torsion free.

P ROOF. We remark that for a given generator g of G, we have an isomorphism

∼
ψ : Λ/(ω1 ) −
→ Z p [G]

determined by ψ(T ) = g − 1. Any element of A generates a cyclic Z p [G]-module, which may then be
viewed as a quotient of Λ/(ω1 ). Since ω1 = T · ω1,0 and ω1,0 is irreducible, we have Λ/(ω1 , f ) is finite
for a power series f ∈ Λ if f is not a unit times a product of a power of T and a power of ω1,0 . This
leaves three possibilities for nontrivial p-torsion free quotients of Λ/( f ), which are Λ/( f ) ∼
= Z p [G],
Λ/(T ) ∼
= Z p , and Λ/(ω1,0 ) ∼
= X, since ψ(ω1,0 ) = NG . Therefore, the structure theorem for finitely
generated Λ-modules tells us that A is pseudo-isomorphic to a direct sum of copies of the latter two
Z p [G]-modules, Z p and X. 

R EMARK 2.7.2. The Z p [G]-module Z p [G] is pseudo-isomorphic to X ⊕ Z p . Explicitly, letting ε

denote the augmentation map, we have

Z p [G] → X ⊕ Z p , a 7→ ((g − 1)a, ε(a))

X ⊕ Z p → Z p [G], (x, b) 7→ x + bNG ,

and both of these maps are injective with cokernel isomorphic to Z/pZ.

We now state the main result of the section.

64 2. MODULE THEORY

T HEOREM 2.7.3. Let A be a finitely generated Z p [G]-module that is p-torsion free. Then there is
an isomorphism
φ : A → Z p [G]r ⊕ X s ⊕ Ztp
of Z p [G]-modules for some r, s,t ≥ 0.

P ROOF. Since Z p [G] is Z p [G]-projective, we also have that if B ∼

= Z p [G]r is the maximal Z p [G]-
free quotient of A, then setting A0 = ker(A → B), we have an isomorphism

A∼
= A0 ⊕ Z p [G]r ,

where A0 has no free Z p [G]-quotient. We may therefore assume that A itself has no free Z p [G]-quotient.
Consider the sequence
0 → AG → A → IG A → 0.
Since A is p-torsion free, we must have (IG A)G = 0, since there is an injective pseudo-isomorphism

ψ : A → X s ⊕ Ztp

for some s,t ≥ 0, and (IG X)G = 0, while IG Z p = 0. In particular, we have that AG ∼
= Ztp , and there is
an injective pseudo-isomorphism from IG A to X u for the above u, since IG X ' X.
Let {x1 , . . . , xm } be a minimal generating set of IG A as a Z p [G]-module. We note that Z p [G]xi is
isomorphic to a finite index submodule of X, and it is therefore a power IGn X for some n. (Here, note
that pX ∈ IG X.) The map X → IGn given by x 7→ (g − 1)n x for a generator g ∈ G, is an isomorphism, so
in fact we have Z p [G]xi ∼
= X.
If y ∈ Z p [G]xi ∩ Z p [G]x j , then by minimality we clearly must have

y ∈ IG xi ∩ IG x j

since Z p [G]xi ∼
= X has IG xi as its unique maximal improper submodule. We then have xi0 ∈ Z p [G]xi and
x0j ∈ Z p [G]x j with
y = (g − 1)xi0 = (g − 1)x0j ,
which forces xi0 − x0j ∈ (IG A)G . In other words, we have xi0 = x0j , contradicting minimality. We therefore
have m = s and IG A ∼ = Is .
G
We now know that A fits in an exact sequence
π
→ X s → 0,
0 → Ztp → A −

which we claim splits. To see this, write X s = hx1 , . . . , xs i. Then zi = NG x̃i is an element of Ztp , and the
sequence splits if and only if zi ∈ pZtp for all i, since this means exactly that there exist yi ∈ Ztp with
zi = pyi and therefore NG (x̃i − yi ) = 0, which tells us that hx̃i − yi i ∼
= X. The Z p [G]-linear map taking
 2.8. EIGENSPACES 65

xi to x̃i − yi then determines the splitting. If not, we have that some x̃i generates a direct summand of A
isomorphic to Z p [G], since zi (for some i) may be taken as part of a basis {zi , w2 , . . . , wt } of Ztp , and

A = hx̃1 , . . . , x̃s , w2 , . . . , wt i ∼
= Z p [G] ⊕ hx̃1 , . . . , x̃i−1 , x̃i+1 , . . . , x̃s , w2 , . . . , wt i.

Since we have assumed that A has no Z p [G]-quotient, the latter cannot happen, so the sequence splits,
as desired. 

2.8. Eigenspaces

In this section, we suppose that ∆ is a finite abelian group. For a fixed prime p, we consider the
group
×
∆∗ = Hom(∆, Q p )
of p-adic characters of ∆. Let O denote the Z p -algebra generated by the roots of unity of order dividing
the exponent of ∆, and let E denote the quotient field of O. For χ ∈ ∆∗ , we let Oχ the Z p -algebra
generated by the values of χ, and let Eχ denote its fraction field. Cearly, the ring O contains Oχ .
What we shall call eigenspaces of a Z p [∆]-module shall in general, in fact, be quotients. Note that
χ ∈ ∆∗ induces a map χ̃ : O[∆] → O, which restricts to a map Z p [∆] → Oχ .

D EFINITION 2.8.1. Let A be an O[∆]-module, and let ψ ∈ ∆∗ . We define the ψ-eigenspace of A as

Aψ = A ⊗O[∆] O,

where the map O[∆] → O in the tensor product is χ̃.

R EMARK 2.8.2. If p - |∆|, then the canonical map A → Aψ induces an isomorphism

∼
→ Aψ .
{a ∈ A | δ a = ψ(δ )a for all δ ∈ ∆} −

It is the former module that might more typically be called an eigenspace. It can be interpreted as the
∆-invariant group of the twist A(ψ) of A that is A as an O-module but on which δ ∈ ∆ acts as ψ(δ )δ
does on A. Our eigenspace Aψ is instead the ∆-coinvariant group of A(ψ).

N OTATION 2.8.3. For ψ ∈ ∆∗ , set

1
eψ = ψ(δ )δ −1 ∈ E[∆].
|∆| δ∑
∈∆

Note that
σ eψ = ψ(σ )eψ
for every σ ∈ ∆, and in particular
O[∆]eψ = Oeψ
as an O[∆]-submodule of E[∆].
66 2. MODULE THEORY

P ROPOSITION 2.8.4. We have a canonical decomposition of rings and E[∆]-modules

E[∆] ∼
= ∏ Eeψ .
ψ∈∆∗

If p - |∆|, we similarly have a decomposition

O[∆] ∼
= ∏ Oeψ .
ψ∈∆∗

P ROOF. One need only remark that the eψ are mutually orthogonal idempotents that sum to 1, as
is a basic fact of character theory (in this case for a finitely generated abelian group). 

The following lemma is useful to note.

L EMMA 2.8.5. Let ψ ∈ ∆∗ . For any E[∆]-module A (or O[∆]-module A if p - |∆|), we have Aψ =
eψ A.

P ROOF. If a ∈ eψ A, then eψ a = a, as eψ is an idempotent. Conversely, if a ∈ A(ψ) , then

1
eψ a = ψ(δ )−1 δ a = a,
|∆| δ∑
∈∆
as δ a = ψ(δ )a. 

The following is a consequence of Proposition 2.8.4.

P ROPOSITION 2.8.6. For every E[∆]-module A, there is an internal direct sum decomposition

A∼
M
= Aψ .
ψ∈∆∗

If p - |∆|, then this decomposition holds for O[∆]-modules as well.

P ROOF. We have

A∼
= A ⊗O[∆] O[∆] ∼ Oeψ ∼ A ⊗O[∆] Oeψ ∼ eψ A ⊗O[∆] O[∆] ∼
M M M M
= A ⊗O[∆] = = = Aψ ,
ψ∈∆∗ ψ∈∆∗ ψ∈∆∗ ψ∈∆∗

with the second step being Proposition 2.8.4 and the last step following from Lemma 2.8.5. 

Eigenspaces of an O[∆]-module behave well under tensor products and homomorphism groups, as
seen in the following result.

L EMMA 2.8.7. Let A and B be O[∆]-modules with A = Aχ and B = Bψ for some χ, ψ ∈ ∆∗ . We

then have
A ⊗O B = (A ⊗O B)χψ
and
−1 ψ
HomO (A, B) = HomO (A, B)χ .
 2.8. EIGENSPACES 67

P ROOF. For a ∈ A and b ∈ B, we have

δ (a ⊗ b) = δ (a) ⊗ δ (b) = χ(δ )a ⊗ ψ(δ )b = χψ(δ ) · a ⊗ b.

For φ ∈ HomO (A, B), we have

(δ · φ )(a) = δ φ (δ −1 a) = ψ(δ )φ (χ(δ )−1 a) = ψ χ −1 (δ )φ (a).

We next consider a slightly different notion of eigenspaces, in this case for Z p [∆]-modules.

D EFINITION 2.8.8. Let A be a Z p [∆]-module, and let χ ∈ ∆∗ . The χ-eigenspace A(χ) of A is defined
as
A(χ) = A ⊗Z p [∆] Oχ ,
where the map Z p [∆] → Oχ is given by χ̃.

N OTATION 2.8.9. For χ ∈ ∆∗ , set

1
ẽχ = TrEχ /Q p (χ(δ ))δ −1 ∈ Z p [∆],
|∆| δ∑
∈∆
where TrEχ /Q p : Eχ → Q p denotes the trace map.

N OTATION 2.8.10. For a field E, let GE denote its absolute Galois group, which is to say the Galois
group of the extension of E given by a fixed separable closure.
×
D EFINITION 2.8.11. We say that two p-adic characters χ, ψ : ∆ → Q p are conjugate if there
exists σ ∈ GQ p such that χ = σ ◦ ψ.

R EMARK 2.8.12. If χ and ψ are conjugate, then Oχ = Oψ .

R EMARK 2.8.13. If A is also a Q p -vector space or p - |∆|, then the canonical map ẽχ A → A(χ)
is an isomorphism. Note that while A(χ) has an Oχ -module structure, the Z p [∆]-module ẽχ A is only
endowed with such a structure when a choice of character ψ in the conjugacy class of χ is made.

Let Σ denote the set of conjugacy classes in ∆∗ . We let [χ] denote the conjugacy class of χ ∈ ∆∗ .
We then have the following.

L EMMA 2.8.14. Let A be a Z p [∆]-module, and let χ ∈ ∆∗ . We have

A(χ) ⊗Oχ O ∼
= (A ⊗Z p O)χ .

If A is also a Q p -vector space or p - |∆|, then we also have

A(χ) ⊗Z p O ∼ (A ⊗Z p O)ψ
M
=
ψ∈[χ]
68 2. MODULE THEORY

P ROOF. For the first isomorphism, we merely note that

= A ⊗Z p [∆] eχ O ∼
= A ⊗Z p [∆] eχ Oχ ⊗Oχ O ∼
A(χ) ⊗Oχ O ∼ = (A ⊗Z p O)χ .

Let ∆χ = ∆/ ker χ, which is a cyclic group, generated by an element we call δχ . Note that ψ ∈ ∆∗
is conjugate to χ if and only if ψ factors through ∆χ and there exists σ ∈ GQ p such that ψ(δχ ) =
σ (χ(δχ )). Hence, the characters in [χ] are in one-to-one correspondence with the GQ p -conjugates of
χ(δχ ). Let ξ = χ(δχ ), and suppose that Φ ∈ Z p [X] is its minimal polynomial. We then have

O ⊗Z p Z p [ξ ] ∼
= O ⊗Z p Z p [X]/(Φ(X)) ∼ = ∏ O[X]/(X − ξ 0 ) ∼
= O[X]/(Φ(X)) ∼ = ∏ O,
ξ0 ξ0

where ξ 0 runs over the GQ p -conjugates of ξ , and the composite map takes 1 ⊗ ξ to ξ 0 in the ξ 0 -
coordinate. Reinterpreting this, we have

O ⊗Z p eχ Oχ ∼ eψ O
M
=
ψ∈[χ]

as O[∆]-modules, where the map takes 1 ⊗ eχ to eψ in the ψ-coordinate. (Note that eψ = σ eχ if

ψ = σ χ, if we let σ act on the coefficients of eχ .) Therefore, we may conclude that

A(χ) ⊗Z p O ∼
= A ⊗Z p [∆] eχ Oχ ⊗Z p O ∼ A ⊗Z p [∆] eψ O ∼ (A ⊗Z p O)ψ .
M M
= =
ψ∈[χ] ψ∈[χ]

P ROPOSITION 2.8.15. For every Q p [∆]-module A, and every Z p [∆]-module A if p - |∆|, there is a
direct sum decomposition
A∼ A(χ)
M
=
[χ]∈Σ

of Z p [∆]-modules, where the sum is over the conjugacy classes in Σ.

P ROOF. We define
A(χ)
M
Φ: A →
[χ]∈Σ
as the product of the surjective maps A → A ⊗Z p [∆] eχ Oχ that take a to a ⊗ eχ . We first show that Φ is
an isomorphism after tensoring with O. That is,

Φ ⊗ idO : A ⊗Z p O → A(χ) ⊗Z p O.
M

[χ]∈Σ

By Lemma 2.8.14, the right-hand side is isomorphic to

(A ⊗Z p O)ψ ∼ (A ⊗Z p O)ψ
M M M
=
[χ]∈Σ ψ∈[χ] ψ∈∆∗
 2.8. EIGENSPACES 69

under the map that takes (a ⊗ eχ ) ⊗ 1 to (a ⊗ 1) ⊗ eψ . The composite map is then the map that takes
a ⊗ 1 to (a ⊗ 1) ⊗ eψ , and this is an isomorphism by Proposition 2.8.6. Thus, we have that Φ ⊗ idO is
an isomorphism, and as O is a free Z p -module, we have that Φ is an isomorphism. 

Even if p | |∆|, we have a weaker direct sum decomposition of Z p [∆]-modules.

N OTATION 2.8.16. Let ϒ denote the set of maximal ideals of Z p [∆].

χ̃
→ O → F p . Thus, ϒ
R EMARK 2.8.17. Every m ∈ ϒ is the kernel of a composite map χ̃ : Z p [∆] −
may be identified with the set of equivalence classes of characters in ∆∗ under which two characters
are considered equivalent if the above compositions are GF p -conjugate. We write ψ ∈ m if ψ ∈ ∆∗ lies
in the equivalence class corresponding to m.

The proof of the following is left to the reader. Perhaps the easiest way to think of it is that each
Am is just A(ρ) for ρ a p-adic character of the prime-to-p part of the group ∆.

P ROPOSITION 2.8.18. For any Z p [∆]-module A, there is a canonical direct sum decomposition

A∼
M
= Am
m∈ϒ
(χ)
We have Am ∼ = A(χ) for χ ∈ m, and if p - |∆|, then m = [χ] and Am ∼
= A(χ) for any χ ∈ m. If A is a
Q p -vector space, then we have that
Am ∼ A(χ) .
M
=
[χ]⊂m
 CHAPTER 3

Iwasawa theory

Throughout this chapter, F will denote a fixed number field, and we let p be a prime.

3.1. Z p -extensions

D EFINITION 3.1.1. A Galois extension F∞ of F is said to be a Z p -extension if Gal(F∞ /F) ∼

= Z p.
n
Fix a Z p -extension F∞ of F, and set Γ = Gal(F∞ /F). The fixed field of Γ p is a number field Fn
with Gal(Fn /F) ∼= Z/pn Z. We set
n
Γn = Γ/Γ p = Gal(Fn /F).

D EFINITION 3.1.2. The absolute Galois group of a field E is the Galois group GE = Gal(E sep /E)
of a separable closure E sep of E over E.

For a field E of characteristic not p, let µ p∞ denote the group of p-power roots of unity in a separable
closure E sep of E.

D EFINITION 3.1.3. For a field E of characteristic not p, the p-adic cyclotomic character is the map
χ : GE → Z×
p defined by σ (ζ ) = ζ
χ(σ ) for all ζ ∈ µ ∞ .
p

m =ζ
Let us fix a primitive nth root of unity in Q for each n ≥ 1, subject to the condition that ζnm n
for all m ≥ 1.

R EMARK 3.1.4. For a number field F, the p-adic cyclotomic character χ : GF → Z×

p induces an
injection of Gal(F(µ p∞ )/F) onto an open subgroup of Z×
p . It is an isomorphism if F = Q.

It is easy to see that 

(1 + pZ ) × µ (Z ) p odd
×∼ p p−1 p
Zp = .
(1 + 4Z ) × h−1i p = 2,
2
Let q = p if p is odd and q = 4 if p = 2. Every element of 1 + qZ p is a p-adic power of some
topological generator u of 1 + qZ p , such as 1 + q, which is to say that the map that takes a ∈ Z p to ua
is an isomorphism from Z p to 1 + qZ p . We therefore have

Z × Z/(p − 1)Z p odd
×∼ p
(3.1.1) Zp =
Z × Z/2Z p = 2.
2

71
72 3. IWASAWA THEORY

As a consequence, we have the following result.

L EMMA 3.1.5. Any open subgroup of Z×

p has a unique quotient isomorphic to Z p for any p.

P ROOF. That the quotient of Z×

p by its group of torsion elements is isomorphic to Z p follows from
(3.1.1). We then need only remark that any open subgroup of Z p has the form pn Z p for some n ≥ 0, so
is itself isomorphic to Z p . 

Together, Remark 3.1.4 and Lemma 3.1.5 allow us to make the following definition.

D EFINITION 3.1.6. The cyclotomic Z p -extension Fcyc of F is the unique subfield of F(µ p∞ ) that is
a Z p -extension of F.

In fact, if F is totally real, then its cyclotomic Z p -extension Fcyc will lie in the maximal totally real
subfield of F(µ p∞ ) (and therefore will equal it in the case that p = 2).
Next, we study ramification in Z p -extensions.

P ROPOSITION 3.1.7. Suppose that v is a place of F not over p. Then v is unramified in any Z p -
extension F∞ /F.

P ROOF. The inertia subgroup of v in Γ = Gal(F∞ /F) is a closed subgroup of Γ and therefore equal
n
to Γ p for some n ≥ 0, unless it is trivial. In the case that v is archimedean, only the latter case is
possible as an inertia group at v has order at most 2 in general. In general, in the former case, Fn is its
fixed field, and the completion of Fn at a prime over v has a tamely, totally ramified Z p -extension that
is the completion of F∞ . On the other hand, the completion of Fn being a characteristic zero local field,
such an extension does not exist. 

L EMMA 3.1.8. There exists a prime v over p in F and an n ≥ 0 such that F∞ /Fn is totally ramified
at v.

P ROOF. By Proposition 3.1.7, no prime not over p ramifies in F∞ /F, so if no primes over p ramify,
then F∞ /F would be unramified everywhere. However, the Hilbert class field of F is of finite degree, so
this is not possible. That is, there exists a v over p such that that the inertia group at v in Γ is nontrivial,
n
hence equal to some Γ p . 

In the case of the cyclotomic Z p -extension, we can say more.

P ROPOSITION 3.1.9. Let Fcyc be the cyclotomic Z p -extension of F. No finite prime splits com-
pletely in Fcyc /F, and every prime over p is totally ramified in Fcyc /Fn for some n ≥ 0.
 3.2. LIMITS OF CLASS GROUPS 73

P ROOF. If v split completely in Fcyc /F, then it would also have to split completely in the extension
F(µ p∞ )/F(µq ), since F(µ p∞ ) = F∞ (µq ), where q = p for p odd and q = 4 for p = 2. But this means
that Fv (µ p∞ ) = Fv (µq ), which is to say that Fv (µq ) contains µ p∞ , which is impossible.
On the other hand, we know that Qcyc /Q is totally ramified at p, so the resulting local extension
Q p,cyc /Q p is totally ramified as well. But then the completion of Fcyc at a prime above v is simply
the compositum Fv · Q p,cyc , and therefore its intersection with the maximal unramified extension of
Q p must be of finite degree over Q p . In particular, Fv · Q p,cyc /Fv has an infinite inertia group, which
p n
therefore must have the form Γcyc for some n ≥ 0, where Γcyc = Gal(Fcyc /F). 

We note the following interesting corollary.

C OROLLARY 3.1.10. Let v be a prime of Fcyc not lying above p. Suppose that E/Fcyc is a pro-p
extension in which it does not ramify. Then v splits completely in E/Fcyc .

P ROOF. Since Fv,cyc /Fv is an unramified Z p -extension by Propositions 3.1.7 and 3.1.9, it is the
maximal unramified pro-p extension of Fv . It follows that for any prime w of E lying over v, we must
have Ew = Fv,cyc , since the Galois closure of Ew /Fv is a pro-p extension of Fv containing Fv,cyc . 

Finally, we consider the maximal number of independent Z p -extensions of F, which is to say the
Z p -rank of the Galois group of the maximal abelian Vp -ramified extension of F.

P ROPOSITION 3.1.11. Let F̃ denote the compositum of all Z p -extensions of F. Then Gal(F̃/F) ∼
=
Zrp2 +1+δ , where δ is the Leopoldt defect of F.

P ROOF. This is a consequence of Theorem 1.5.7, since Proposition 3.1.7 tells us that the Z p -rank
of the maximal abelian Vp∞ -ramified extension of F is the Z p -rank of Gal(F̃/F). 

3.2. Limits of class groups

Let F∞ be a Z p -extension of F with Γ = Gal(F∞ /F). We define Fn and Γn as before.

D EFINITION 3.2.1. We refer to Λ = Z p JΓK as the Iwasawa algebra of the extension F∞ /F.

D EFINITION 3.2.2. A Λ-module, or module over the Iwasawa algebra, is also called an Iwasawa
module.

By definition, we have Λ = lim Z p [Γn ]. Note that any Z p [Γn ]-module is automatically an Iwasawa
←−
module, with Λ acting through the quotient map πn : Λ → Z p [Γn ]. Therefore, given an inverse (resp.,
direct) system of Z p [Γn ]-modules Mn with respect to maps that are Λ-module homomorphisms, the
inverse (resp., direct) has the structure of a Λ-module.
74 3. IWASAWA THEORY

D EFINITION 3.2.3. Let F∞ /F be a Z p -extension. For n ≥ m ≥ 0, let us set

Nn,m = NFn /Fm : An → Am and jn,m = jFn /Fm : Am → An .

With respect to the systems defined by these maps, we set

X∞ = lim An and A∞ = lim An .

←− −→
n n
T ERMINOLOGY 3.2.4. The direct limit limn ClFn contains A∞ as its p-part and is called the class
−→
group of F∞ .

The maps Nn,m and jn,m are Z p [Γn ]-module homomorphisms, and so both X∞ and A∞ have canonical
structures of Λ-modules.
Recall that the Artin map sets up an isomorphism between An and Gal(Hn /Fn ), where Hn is the
p-Hilbert class field of Fn . Under this identification, the norm map Nn,m becomes the map on Galois
groups that is restriction. We then have the following.

R EMARK 3.2.5. Let E be an algebraic extension of F, and for a set of primes S of F, let SE be
the set of primes of E lying above those in S. We will say that more simply that an extension of E is
S-ramified if it is SE -ramified.

P ROPOSITION 3.2.6. Let S be a set of primes of F. Let Ln denote the maximal S-ramified abelian
pro-p extension of Fn for n ≥ 0 or n = ∞. Then the inverse limit of restriction maps

Gal(L∞ /F∞ ) → lim Gal(Ln /Fn )

←−
n
is an isomorphism of Λ-modules.

P ROOF. Since Ln /Fn is an S-ramified abelian pro-p extension, so is Ln F∞ /F∞ . Therefore, Ln ⊆ L∞ .

= L∞ . Let x ∈ L∞ . Then F∞ (x)/F∞ is an S-ramified abelian p-extension. Let y be
S
We claim that n Ln
a field generator of the Galois closure of F(x) as an extension of F. To show that x ∈ Ln for some n, it
therefore suffices to show that y ∈ Ln . Let m be such that Fn (y) ∩ F∞ = Fm . Then Fm (y) ∩ F∞ = Fm as
well, and the restriction map
Gal(F∞ (y)/F∞ ) → Gal(Fm (y)/Fm )
is surjective, so Fm (y)/Fm is abelian.
Since L∞ /F∞ is S-ramified, and F∞ /F is Vp -ramified, we have that Fm (y)/Fm is S ∩Vp -ramified. If
v is a place over p in Fm that is not in SFm , then since F∞ (y)/F∞ is unramified over v, the same must be
true of Fn (y)/Fn for some n, and therefore y ∈ L∞ .
It now follows that the inverse limit of restriction maps
∼
Gal(L∞ /F∞ ) −
→ lim Gal(Ln /F∞ ∩ Ln )
←−
n
 3.2. LIMITS OF CLASS GROUPS 75

n (F∞ ∩ Ln ) = F∞
S S
is an isomorphism, and since = n Fn , we have that
∼
lim Gal(Ln /F∞ ∩ Ln ) −
→ lim Gal(Ln /Fn )
←− ←−
n n
is an isomorphism as well, as desired. 

C OROLLARY 3.2.7. The inverse limit of Artin maps provides a canonical identification between
X∞ and the Galois group of the maximal unramified abelian pro-p extension of F∞ .

T ERMINOLOGY 3.2.8. We call the Λ-module X∞ the unramified Iwasawa module.

R EMARK 3.2.9. If K is an algebraic extension of Q, we may speak of its primes as the valuations
on K extending the valuations of Q. To say that an extension L of K is unramified at a prime v is
exactly to say that every extension of v to a prime w of L is unramified in the sense that the extension
Lw /Kv of completions is unramified, which is to say Galois with group restricting isomorphically to
the Galois group of the corresponding extension of residue fields. (If v is archimedean, this just means
that Lw = Kv .)

More generally, we make the following definition.

D EFINITION 3.2.10. Let S be a set of primes of F. The S-ramified Iwasawa module over F∞ is the
Galois group X∞,S of the maximal S-ramified abelian pro-p extension of F∞ .

Let us choose a topological generator γ of Γ, which defines a unique continuous, Z p -linear iso-
∼
morphism Λ −
→ Z p JT K that takes γ − 1 to T . Therefore, we may speak of characteristic ideals of Λ as
elements of Z p JT K. We have the following result on the structure of X∞ .

P ROPOSITION 3.2.11. The Λ-module X∞ is finitely generated and torsion.

P ROOF. For n ≥ m, set Σn,m = ΣFn /Fm in the notation of Theorem 1.3.14, which also provides exact
sequences fitting into commutative diagrams
Nn0 ,m
ker Σn0 ,m / (An0 ) pm / Am / coker Σn0 ,m /0
Γ n0
Nn0 ,n
  Nn,m 
ker Σn,m / (An ) pm / Am / coker Σn,m /0
Γ n

(m) m
of Z p [Γm ]-modules for n0 ≥ n. Let Iv denote the inertia group at v in Γ p , which can only be nontrivial
for v ∈ Vp which do not split completely in F∞ /F, and let
(m) m
Σ(m) :
M
Iv → Γp
v∈Vp (Fm )
76 3. IWASAWA THEORY

be the natural map given by inclusion and product. In the inverse limit over n, we obtain exact se-
quences

(3.2.1) ker Σ(m) → (X∞ )Γ pm → Am → coker Σ(m) → 0.

Note that ker Σ(m) is finitely generated over Z p and Am is finite. By Nakayama’s Lemma, we see that
X∞ is a finitely generated Λ-module. Moreover, we see that (X∞ )Γ pm is of bounded Z p -rank for all
m. Were X∞ to have nontrivial Λ-rank, then since there would exist a pseudo-isomorphism from X∞
to the direct sum M of Λr and a torsion module, the ranks of (X∞ )Γ pm would necessarily have been
unbounded, since Λ pm ∼= Z p [Γm ], and
Γ

(X∞ )Γ pm → MΓ pm

has finite cokernel. 

R EMARK 3.2.12. If there exists a unique prime above p in F, and it is unsplit in Fm , then (3.2.1)
implies that the map (X∞ )Γ pm → Am is an injection. If, moreover, p is totally ramified in F∞ , then
(X∞ )Γ pm → Am is an isomorphism for every m.

We have the following theorem of Iwasawa that was mentioned in the introduction.

T HEOREM 3.2.13 (Iwasawa). Let λ = λ (X∞ ) and µ = µ(X∞ ). Then there exists ν ∈ Z such that
n µ+nλ +ν
|An | = p p

for all sufficiently large n.

P ROOF. Let Nn : X∞ → An be the inverse limit of norm maps Nn0 ,n for n0 ≥ n. Let us use Yn to
denote the kernel of Nn , which is a Λ-submodule of X∞ that is pseudo-isomorphic to X∞ .
Fix m sufficiently large such that every prime over p that ramifies in F∞ /Fm is totally ramified. In
particular, we have that Nm is surjective. We consider n ≥ m. Let Sn be the set of primes (over p) in Fn
that ramify in F∞ , and hence are totally ramified, and note that |Sn | = |Sm | 6= 0 by Lemma 3.1.8. Then
n n
the inertia group at v ∈ Sn in Γ p is Γ p itself.
Let Ln be the maximal unramified abelian pro-p extension of Fn , and let L∞ be the maximal
unramified abelian pro-p extension of F∞ . We have X∞ ∼
= Gal(L∞ /F∞ ) and An ∼= Gal(Ln /Fn ), so
Yn ∼= Gal(L∞ /Ln F∞ ). Let E be the maximal unramified p-extension of Fn in L∞ . Since any v ∈ Sn
is totally ramified in F∞ /Fn , we have E ∩ F∞ = Fn . Since EF∞ /F∞ is abelian, this tells us that E/Fn
is abelian as well. Thus, E is equal to the maximal unramified abelian p-extension of Fn in L∞ . Con-
(n)
sequently, Gal(L∞ /Ln ) is topologically generated by the inertia groups Jv in Gal(L∞ /Fn ) for primes
v ∈ Sn , and Yn is the intersection of the latter group with Gal(L∞ /F∞ ), i.e., it consists of those elements
which restrict trivially to Γ.
 3.2. LIMITS OF CLASS GROUPS 77

In other words (for n = m), we have that Ym is topologically generated as a pro-p group by elements
(m) (m)
g = σ τ −1 ∈ Gal(L∞ /F∞ ), where σ ∈ Jv and τ ∈ Jw for primes v, w ∈ Sm are such that σ and τ both
m
restrict to γ p for a fixed topological generator γ of Γ. We can compute the action of the element
p n−m −1 m
ωn,m = ∑i=0 γ p i on g as follows:
pn−m −1
n−m n−m n−m n−m
ωn,m · g = ∏ τ i gτ −i = (gτ) p τ −p = σp τ −p .
i=0
n−m n−m
As the elements σ p τ −p topologically generate Yn , this implies that ωn,mYm = Yn .
Since An = X∞ /Yn , we conclude that

|An | = |X∞ /Ym | · |Ym /ωn,mYm |

for all n ≥ m. Since Ym is pseudo-isomorphic to X∞ , we have λ = λ (Ym ) and µ = µ(Ym ). Since |X∞ /Ym |
is a constant power of p, Theorem 2.4.7 yields the result. 

Finally, we compare X∞ and A∞ .

P ROPOSITION 3.2.14. The Λ-modules α(X∞ ) and A∨ ∨

∞ are pseudo-isomorphic, and in particular A∞
is finitely generated and Λ-torsion. Moreover, A∨
∞ has no nonzero finite Λ-submodules.

P ROOF. As in the proof of Theorem 3.2.13, we let Yn denote the kernel of the inverse limit of norm
maps Nn : X∞ → An for each n ≥ 0. We showed that there exists m ≥ 0 sufficiently large so that Nn is
surjective and ωn,mYm = Yn for all n ≥ m. We consider a directed system of short exact sequences with
morphisms as in the following diagram

0 / Ym /ωn,mYm / X∞ /ωn,mYm / X∞ /Ym /0

ωn0 ,n ωn0 ,n ωn0 ,n
  
0 / Ym /ωn0 ,mYm / X∞ /ωn0 ,mYm / X∞ /Ym /0

for n0 ≥ n ≥ m. Since X∞ /Ym is a finite Λ-module, in the direct limit we obtain isomorphisms
∼ ∼
α(Ym )∨ = lim Ym /ωn,mYm −
→ lim X∞ /ωn,mYm −
→ lim An = A∞ .
−→ −→ −→
n n n

Since Ym injects into X∞ with finite cokernel, Proposition 2.6.10 yields that the natural map α(X∞ ) →
α(Ym ) is an injective pseudo-isomorphism. Since A∨ ∼ = α(Ym ), the final statement follows from Theo-
∞
rem 2.6.13. 

Again noting Theorem 2.6.13, we have the following corollary.

C OROLLARY 3.2.15. The Λ-module A∨ ι

∞ is is pseudo-isomorphic to X∞ , and in particular, X∞ and
A∨
∞ have the same λ and µ-invariants.
78 3. IWASAWA THEORY

We end with a still open conjecture of Iwasawa, which is known in the case of abelian fields by
work of Ferrero-Washington: see Theorem 6.2.1.

C ONJECTURE 3.2.16 (Iwasawa’s µ-conjecture). If F∞ is the cyclotomic Z p -extension Fcyc of F,

then µ(X∞ ) = 0.

We will also have cause to study two modules related to X∞ and A∞ .

D EFINITION 3.2.17. Let F∞ /F be a Z p -extension. For S = Vp , let us set A0n = AFn ,S . We then define

X∞0 = lim A0n and A0∞ = lim A0n

←− −→
n n
with respect to the maps Nn,m and jn,m on these groups.

D EFINITION 3.2.18. We call X∞0 the completely split Iwasawa module, while A0∞ is the p-part of
the p-class group limn ClFn ,Vp of F∞ .
−→
We summarize without proof the results for X∞ and A∞ that also hold for X∞0 and A0∞ by much the
same arguments.

P ROPOSITION 3.2.19. The Λ-module X∞0 is finitely generated and torsion. It is canonically isomor-
phic via an inverse limit of Artin maps to the Galois group of the maximal unramified abelian pro-p
extension of F∞ in which every prime over p splits completely. Moreover, (X∞0 )ι is pseudo-isomorphic
to (A0∞ )∨ , and the latter module has no nonzero finite Λ-submodules.

For the cyclotomic Z p -extension, we note that we could just have well have chosen any set of
primes containing Vp in defining X∞0 .

P ROPOSITION 3.2.20. Let F∞ /F be the cyclotomic Z p -extension. Then the natural maps

X∞0 → lim AFn ,S and A0∞ → lim AFn ,S

←− −→
n n
are respectively an isomorphism and a surjective pseudo-isomorphism for any finite set S of primes of
F containing Vp .

3.3. The p-ramified Iwasawa module

In this section, we focus for simplicity on the case that S = Vp∞ , though there is no theoretical
obstruction to considering a larger finite set. We make the following definition.

D EFINITION 3.3.1. Let F∞ /F be a Z p -extension. Let Xn = XFn ,Vp∞ for n ≥ 0, and let

X∞ ∼
= lim X
←− n
n
be the Vp∞ -ramified Iwasawa module, which we refer to as the p-ramified Iwasawa module.
 3.3. THE p-RAMIFIED IWASAWA MODULE 79

Consider the following weakening of the Leopoldt conjecture.

C ONJECTURE 3.3.2 (Weak Leopoldt conjecture). Let F∞ /F be a Z p -extension. Then the Leopoldt
defects δ (Fn ) are bounded in n ≥ 0.

We will abbreviate δ (Fn ) by δn .

The weak Leopoldt conjecture has the following consequence for the p-ramified Iwasawa module.

T HEOREM 3.3.3. Let F∞ /F be a Z p -extension for which the weak Leopoldt conjecture holds. Then

rankΛ X∞ = r2 (F).

P ROOF. Let M∞ be such that X∞ = Gal(M∞ /F∞ ), and define Mn for n ≥ 0 by

Xn = Gal(Mn /Fn ).

We then have that (X∞ )Γn ∼

= Gal(Mn /F∞ ), and therefore we have an exact sequence

0 → (X∞ )Γn → Xn → Γn → 0.

Since any archimedean place splits completely in a Z p -extension, we have r2 (Fn ) = pn r2 (F) and
hence rankZ p Xn = pn r2 (F) + 1 + δn . It follows that rankZ p (X∞ )Γn = pn r2 (F) + δn for all n. Since δn
is bounded in n, the result then follows from Proposition 2.4.12. 

We prove the weak Leopoldt conjecture in the case of the cyclotomic Z p -extension. Let En = EFn
for each n ≥ 0, and let Un,v be the p-completion of OF×n,v for any prime v of Fn .

T HEOREM 3.3.4. Suppose that F∞ /F is the cyclotomic Z p -extension of F. Then the weak Leopoldt
conjecture holds for F∞ /F. In fact, if F contains µ2p , then δn ≤ λ (X∞0 ) for every n.

P ROOF. Assume first that F contains µ2p . Let r = rankZ p En , and choose units such that α1 , . . . , αr ∈
OF×n generate En modulo its torsion subgroup as a Z p -module and such that the images of αδn +1 , . . . , αr
under
ιn : En → Un,v
M

v∈Vp (Fn )
generate ιn (En ) modulo its torsion subgroup.
Let pk be the exponent of the p-power torsion in ιn (En ). Then, for each 1 ≤ i ≤ δn , there exist
ai j ∈ Z p for each δn + 1 ≤ j ≤ r such that
r
a
ιn (αi ) ∏ ιn (α j i j )
j=δn +1

has trivial pk th power. Fix l ≥ 1. For every i and j as above, choose bi j ∈ Z such that

bi j ≡ ai j mod pl Z p ,
80 3. IWASAWA THEORY

and then set

r
b
βi = αi ∏ αj ij.
j=δn +1
k k+l
It follows that ιn (βi ) p ∈ ιn (En ) p for each i.
Since α1 , . . . , αr form a Z p -linear basis of the maximal p-torsion-free quotient of En , the images
l
of the elements β1 , . . . , βδn in Fn× /Fn×p generate a subgroup isomorphic to (Z/pl Z)δn . By Kummer
theory, the group Fn× ∩ F∞×p is exactly µ p∞ Fn×p , and since the closed subgroup of En generated by
l
β1 , . . . , βδn is p-torsion-free, the images of these elements generate a subgroup of F∞× /F∞×p that is also
isomorphic to (Z/pl Z)δn .
Now consider
1/pl 1/pl

K = F∞ β1 , . . . , βδn ,
k l+k 1/pl
and note that Gal(K/F∞ ) is isomorphic to (Z/pl Z)δn . Since ιn (βi ) p ∈ ιn (En ) p and βi is a pl+k th
pk
root of βi , we have that every prime of v over p splits completely in this extension. Since Gal(K/F∞ )
is already a quotient of X∞ , it is then a quotient of X∞0 . In other words, we have surjections

X∞0 → (Z/pl Z)δn

for every l. Since X∞0 is Λ-torsion, Proposition 2.2.13 tells us that δn ≤ λ (X∞0 ).
Now, if F does not contain µ2p , we still have δ (Fn ) ≤ δ (Fn (µ2p )), and since the latter numbers are
bounded, we have the result. 

We next study sequences into which X∞ fits. For this, we need to define several more Λ-modules.

D EFINITION 3.3.5. Let F∞ be the cyclotomic Z p -extension of F, and let S = Vp∞ . We let

E∞ = lim EFn and E∞0 = lim EFn ,S .

←− ←−
n n

where the inverse limits are taken under norm maps. Letting Un,v denote the pro-p-completion of OF×n,v
for v ∈ S(Fn ), we set
×
U∞,v = lim Un,v and F∞,v = lim F
d ,
←− ←− n,v
n n

for v ∈ S(F∞ ), with the inverse limits taken with respect to the local norm maps. Set

U∞ = ∏ U∞,v and F∞ = ∏ F∞,v .

v∈S(F∞ ) v∈S(F∞ )

Let ι∞ and ι∞0 denote the canonical maps

ι∞ : E∞ → U∞ and ι∞0 : E∞0 → F∞ .

 3.3. THE p-RAMIFIED IWASAWA MODULE 81

P ROPOSITION 3.3.6. Let F∞ be the cyclotomic Z p -extension of F. Assume, moreover, that p is odd
or F has no real places. We have a map of canonical exact sequences of Λ-modules

0 / ker ι∞ / E∞ ι∞
/ U∞ / X∞ / X∞ /0

  ι∞0  
0 / ker ι∞0 / E∞0 / F∞ / X∞ / X∞0 / 0.

P ROOF. This is simply the inverse limit of the sequences of Theorem 1.5.4 for the fields Fn , which
remains exact as the modules in question are profinite. The assumptions are simply to insure that the
number of terms in the direct sum of local unit or multiplicative groups is finite: otherwise, one need
merely replace the direct sums by inverse limits of direct sums at the finite level. 

D EFINITION 3.3.7. We set

A0∞ = lim AFn ,Vp∞ .
−→
n

R EMARK 3.3.8. An element γ ∈ Γ acts on H 1 (GF∞ ,S , µ p∞ ) through its action on cocycles: i.e., for
a cocycle f , γ ∈ Γ, and σ ∈ GF,S we have

(γ · f )(σ ) = γ · f (γ̃ −1 σ γ̃),

where γ̃ is any lift of γ to GF,S . Giving this cohomology group the discrete topology, with respect to
which it is p-power torsion, we have that Γ acts continuously and Z p -linearly, and hence we obtain a
Λ-action.

Kummer theory allows us to prove the following proposition.

P ROPOSITION 3.3.9. Let F∞ /F be the cyclotomic Z p -extension, and let S = Vp∞ . There is canonical
map of exact sequences

1 / O × ⊗Z Q p /Z p / H 1 (GF ,S , µ p∞ ) / A∞ /0
F∞ _ ∞

 
1 / O × ⊗Z Q p /Z p / H 1 (GF ,S , µ p∞ ) / A0∞ /0
F∞ ,S ∞

of Λ-modules.

P ROOF. Recall from Theorem 1.4.5 that we have exact sequences

m
1 → OF×n ,S /OF×p
n ,S
→ H 1 (GFn ,S , µ pm ) → A0n [pm ] → 0,

where A0n = AFn ,S . The direct limit as n heads towards infinity yields
m
1 → OF×∞ ,S /OF×p
∞ ,S
→ H 1 (GF∞ ,S , µ pm ) → A0∞ [pm ] → 0.
82 3. IWASAWA THEORY

For any abelian group B, with respect to the maps B/pm B → B/pm+1 B induced by multiplication by
p, we have
 1 
lim B/pm B ∼= B ⊗Z lim m Z/Z ∼ = B ⊗Z Q p /Z p ,
−→ −→ p
m m
where the maps are the natural inclusion maps on the right-hand side of the middle term. Applying this
to B = OF×∞ ,S and noting that
A0∞ = lim A0∞ [pm ]
−→
m
since A0∞ is p-power torsion, we have the lower exact sequence. (Note that this did not require F∞ to be
the cyclotomic Z p -extension).
Now, note that H 1 (GF∞ ,S , µ p∞ ) is isomorphic via Kummer theory to the direct limit of the groups
m
Bn,m /Fn×p , where Bn,m is the subgroup of x ∈ Fn× such that xOFn ,S = a p for some fractional ideal a
m

of OFn ,S . We then have maps

m
Bn,m /Fn×p → A0n [pm ], x 7→ [a]0 ,

where [a]0 denotes the class of a in A0n , of which the map

θ 0 : H 1 (GF∞ ,S , µ p∞ ) → A0∞

is the direct limit. Given any x ∈ Bn,m , note that there exists n0 > n independent of x such that
m
xOFn0 = b p for some fractional ideal b of OFn0 since every prime over p is totally ramified in F∞ /Ft for
sufficiently large t. We then have a map
m
Bn,m /Fn×p → An0 [pm ], x 7→ [b].

In this way, we obtain in the direct limit a map

θ : H 1 (GF∞ ,S , µ p∞ ) → A∞

which is θ 0 after composing with the natural projection A∞ → A0∞ , which implies that the diagram in
the statement of the proposition commutes.
We need only verify exactness in the upper sequence in the statement of the proposition. The kernel
of θ is identified by Kummer theory with exactly those

x ⊗ p−m ∈ F∞× ⊗ Q p /Z p

such that xOF∞ is the pm th power of a principal ideal (z), which is to say that
m
x ⊗ p−m = xz−p ⊗ p−m ∈ OF×∞ ⊗Z Q p /Z p .

Moreover, if b ∈ A∞ , then b ∈ A∞ [pm ] for some m ≥ 1, and then b is the image of some element
m
bn ∈ An [pm ] by definition of the direct limit. We have then that bnp = xOFn for some x ∈ Fn× , and so
θ (x ⊗ p−m ) = [b]. Hence, θ is surjective. 
 3.3. THE p-RAMIFIED IWASAWA MODULE 83

C OROLLARY 3.3.10. Let F∞ /F be the cyclotomic Z p -extension, and let S = Vp∞ . Then there is a
canonical exact sequence

1 → OF×∞ ⊗Z Q p /Z p → OF×∞ ,S ⊗Z Q p /Z p −
→ A∞ → A0∞ → 0
θ

of Λ-modules.

P ROOF. This follows from Proposition 3.3.9 via the snake lemma. 

D EFINITION 3.3.11. Let M be a Z p [GE ]-module for a field E of characteristic not p. For i ∈ Z, the
ith Tate twist of M is the Z p [GE ]-module M(i) that is M as a Z p -module, but on which GE acts via the
new action ·i given by
σ ·i m = χ(σ )i σ m
for σ ∈ GE and m ∈ M.

E XAMPLE 3.3.12. Given a field E of characteristic not p, a choice of compatible system ζ pn of

primitive pn th roots of unity in a separable closure, i.e., such that ζ ppn+1 = ζ pn for each n ≥ 1, gives rise
to isomorphisms
∼
Z p (1) −
→ lim µ pn , a 7→ (ζ pan )n ,
←−
n
∼ a
Q p /Z p (1) −
→ µ p∞ , 7→ ζ pan
pn
of Z p [GE ]-modules.

R EMARK 3.3.13. The Tate twist Z p (i) for i ∈ Z may be viewed as a Λ-module that is isomorphic
to Z p as a Z p -module, and on which γ ∈ Γ = Gal(F∞ /F) acts by χ(γ)i , where χ : Γ → Z×
p is the
homomorphism induced by the cyclotomic character. More generally, if B is any Λ-module, then we
may speak of its Tate twist B(i) ∼
= B ⊗Z Z p (i), which is a new Λ-module that is B with a modified
p

action of Γ given by γ ·i b = χ(γ)i γb.

C OROLLARY 3.3.14. Suppose that µ2p ⊂ F and F∞ /F is the cyclotomic Z p -extension. Then we
have an exact sequence

0 → A∨ ×
∞ (1) → X∞ → HomZ p (OF∞ ⊗Z Z p , Z p (1)) → 0

of finitely generated Λ-modules.

P ROOF. By assumption, we have µ p∞ ⊂ F∞ . Hence, we have

H 1 (GF∞ ,S , µ p∞ ) ∼
= Homcts (X∞ , µ p∞ ) ∼
= X∨
∞ (1).
84 3. IWASAWA THEORY

Taking the Tate twist of the Pontryagin dual of the sequence of Proposition 3.3.9, we obtain an exact
sequence
0 → A∨ ×
∞ (1) → X∞ → HomZ p (OF∞ ⊗Z Q p /Z p , µ p∞ ) → 0.

The result now follows from the following calculation for an abelian group B:

HomZ p (B ⊗Z Q p /Z p , Q p /Z p ) ∼
= HomZ p (B ⊗Z Z p ⊗Z p Q p /Z p , Q p /Z p )
∼
= HomZ p (B ⊗Z Z p , HomZ p (Q p /Z p , Q p /Z p ))
∼
= HomZ p (B ⊗Z Z p , Z p ),

where in the second-to-last step we have used the adjointness of Hom and ⊗. 

3.4. CM fields

In this subsection, we shall consider the behavior of inverse and direct limits of p-parts of class
groups in the cyclotomic Z p -extension F∞ of F in the case that F is a CM field. We remark that F∞ is
itself a CM field, and for the most part, we could take F∞ to be any CM Z p -extension of F in this section
(though conjecturally, as we shall see later, there no others). We assume that p is odd throughout this
subsection.

P ROPOSITION 3.4.1. The natural maps

jn− : A− −
n → A∞

are injective for all n. Moreover, the natural maps

Nn− : X∞− → A−
n

are all surjective.

P ROOF. The second statement is easy, since the cokernel of Nn is isomorphic as a Λ-module the
n
maximal unramified quotient of Γ p , and Γ has trivial minus part.
− −
For the first statement, it suffices to show that jn+1,n : A−
n → An+1 is injective for each n. Let G =
Gal(Fn+1 /Fn ), and let On denote the ring of integers of Fn . As the maps in Proposition 1.3.5 are easily
− × −
seen to be Galois equivariant, we have that ker jn+1,n is isomorphic to a submodule of H 1 (G, On+1 ) .
Let µ(Fn ) denote the group of p-power roots of unity in Fn for each n. The exact sequence
× ×
1 → µ(Fn+1 ) → On+1 → On+1 /µ(Fn+1 ) → 1

of Z p [Gal(Fn /Fn+ )]-modules, gives rise to a long exact sequence in Tate cohomology
× × ×
· · · → Ĥ 0 (G, On+1 /µ(Fn+1 )) → H 1 (G, µ(Fn+1 )) → H 1 (G, On+1 ) → H 1 (G, On+1 /µ(Fn+1 )) → · · · ,

also of Z p [Gal(Fn /Fn+ )]-modules, so it remains exact after taking minus parts.
 3.4. CM FIELDS 85

Now, for any Z p [Gal(Fn+1 /Fn+ )]-module A, we have a canonical isomorphism

∼
→ H 1 (G, A)− ⊗Z p G ∼
Ĥ −1 (G, A)− − = H 1 (G, A)−

of Z p [Gal(Fn /Fn+ )]-modules, as Gal(Fn /Fn+ ) acts trivially on G (as it acts by lifting and conjugating).
× ×
Since Ĥ i (G, On+1 /µ(Fn+1 ))− is a p-group that is a subquotient of (On+1 /µ(Fn+1 ))− for i = 0, −1,
and the latter group has trivial p-part, we have that there is an isomorphism
∼
Ĥ −1 (G, µ(Fn+1 ))− − × −
→ H 1 (G, On+1 ) .

Note that µ(Fn+1 ) p = µ(Fn ). The map NG : µ(Fn+1 ) → µ(Fn+1 ) induced by the norm element is given
by raising to the pth power so has ker(NG ) = µ p (F), while IG µ(Fn+1 ) = µ p (F), so we have

Ĥ −1 (G, µ(Fn+1 ))− = 0,

finishing the proof. 

We also have the following fact regarding X∞− .

P ROPOSITION 3.4.2. The Λ-module X∞− has no nonzero finite Λ-submodules.

P ROOF. Let M be a finite Λ-submodule of X∞− . Since M is finite, there exists m ≥ 0 such that
m
M → MΓ pm is an isomorphism, which is to say that Γ p acts trivially on M. Let x ∈ M, and suppose
that x is an element of order p in M. Set xn = Nn (x). Then xn 6= 0 for sufficiently large n, which we
may take be at least m. For such an n, note that jn+1,n (xn ) 6= 0 by Proposition 3.4.1. We also have

jn+1,n (xn ) = jn+1,n (Nn+1,n (xn+1 )) = pxn+1

n
by the triviality of the action of Γ p on M. In particular, pxn+1 6= 0, whch forces px 6= 0, contradicting
the existence of x. Hence M = 0. 

Note that µ(X∞ ) = µ(X∞+ ) + µ(X∞− ) and λ (X∞ ) = λ (X∞+ ) + λ (X∞− ), since X∞ ∼
= X∞− ⊕ X∞+ .

P ROPOSITION 3.4.3. Suppose that µ p ⊂ F. Then µ(X∞ ) = 0 if and only if µ(X∞− ) = 0.

P ROOF. If µ(X∞− ) = 0, then Lemma 2.4.10 tells us that the p-ranks of the (X∞ )−
Γp
n are bounded in

n. Since Nn− is surjective, the p-ranks of the A−

n are then bounded as well. By the reflection theorem,
the p-ranks of the A+ + −
n are then bounded, as r p (An ) ≤ r p (An ) + 1. In turn, this implies that the p-ranks
of the (X∞ )+
Γp
+
n are bounded (since the kernel to An has p-rank less than or equal to the number of

ramified primes minus 1 in F∞ /Fn , and this number is bounded in n). Again applying Lemma 2.4.10,
we have that µ(X∞+ ) = 0. 

C ONJECTURE 3.4.4 (Greenberg). The Iwasawa module X∞+ is finite.

86 3. IWASAWA THEORY

R EMARK 3.4.5. Greenberg’s conjecture means exactly that λ (X∞+ ) = µ(X∞+ ) = 0. Therefore, under
the assumption of Iwasawa’s µ-conjecture, Greenberg’s conjecture is equivalent to the statement that
λ (X∞+ ) = 0.

P ROPOSITION 3.4.6. Greenberg’s conjecture holds if and only if A+

∞ = 0.

P ROOF. This is an immediate consequence of Corollary 3.2.15, since (A+ ∨

∞ ) has no finite Λ-
submodules and hence can be finite if and only if it is zero. 

P ROPOSITION 3.4.7. Suppose that µ p ⊂ F. We have an isomorphism

∼
(A− ∨
→ X+
∞ ) (1) − ∞

and an exact sequence

∨ − ×
0 → (A+
∞ ) (1) → X∞ → HomZ p (OF∞ ⊗Z Z p , Z p (1)) → 0.

In particular, Greenberg’s conjecture implies that

X− ∼ ×
∞ = HomZ p (OF∞ ⊗Z Z p , Z p (1)).

P ROOF. Dirichlet’s unit theorem tells us that

OF×∞ ⊗Z Z p ∼
= (OF×+ ⊗Z Z p ) × µ p∞
∞

as Z p [Gal(F/F + )]-modules. We have

HomZ p (OF×∞ ⊗Z Z p , Z p (1))− = HomZ p (OF×∞ ⊗Z Z p , Z p (1)),

and
HomZ p (OF×∞ ⊗Z Z p , Z p (1))+ ∼
= HomZ p (µ p∞ , Z p (1)) = 0.
The first statement is then a consequence of Corollary 3.3.14, and the second is then a consequence of
Proposition 3.4.6. 

C OROLLARY 3.4.8. The finitely generated, Λ-torsion modules (X∞− )ι (1) and X+
∞ are pseudo-
isomorphic.

P ROOF. This is an immediate consequence of Proposition 3.4.7 and Corollary 3.2.15. 

To obtain even finer information, we can pass to eigenspaces.

×
C OROLLARY 3.4.9. Let F be totally real, let χ : GF → Z p be a finite odd character of prime-to-p
order, and let E be an abelian extension of F of degree prime to p containing Fχ (µ p ). Considering
Iwasawa modules for the cyclotomic Z p -extension E∞ /E, we have
−1 )
X(ω χ ∼ (χ) ∨
= (A∞ ) (1) ' (X∞(χ) )ι (1)
∞
 3.5. KIDA’S FORMULA 87

as Λ[Gal(E/F)]-modules.

R EMARK 3.4.10. In Corollary 3.4.9, the Iwasawa modules in question, X∞ , A∞ , and X∞ , have an
action of Gal(E/F) that commutes with the Λ-action, since

Gal(E∞ /F) ∼
= Gal(F∞ /F) × Gal(E/F)

in that E∞ /F is abelian and Gal(E/F) has prime-to-p order.

3.5. Kida’s formula

Suppose that F is a number field and E is a cyclic extension with Galois group G. The exact
sequence of Theorem 1.3.14 is not quite canonical as written, since one of the maps depends on a
choice of generator of G, but it becomes canonical when written in the form

0 → ker jE/F ⊗Z G → Ĥ −1 (G, OE× ) → IEG /IF ⊗Z G → ClG

E / jE/F (ClF ) ⊗Z G
NE/F
→ OF× /NE/F OE× → ker ΣE/F → (ClE )G −−−→ ClF → coker ΣE/F → 0,

which is to say that the map

× ×
E / jE/F (ClF ) ⊗Z G → OF /NE/F OE
ClG

of Remark 1.3.15 is canonical, noting that there is a canonical isomorphism

H 1 (G, A) ⊗Z G ∼
= Ĥ −1 (G, A)

for any Z[G]-module A. Moreover, if E is Galois over F0 ⊂ F, the maps in the above sequence are all
Gal(F/F0 )-equivariant.
Suppose now that we consider the cyclotomic Z p -extensions F∞ /F and E∞ /F. Then we may
consider the inverse limit of the above exact sequences for the extensions En /Fn , and we obtain the
following result, in which we distinguish Iwasawa modules over F∞ and E∞ by writing them in the
notation of functions of the base field; e.g., X∞ (E) is the Galois group of the maximal unramified
abelian pro-p extension of E∞ .

T HEOREM 3.5.1. Let E/F be a cyclic of prime power order Galois extension of number fields with
G = Gal(E/F). Let F∞ denote the cyclotomic extension of F, and let E∞ = EF∞ be the cyclotomic
Z p -extension of E. We suppose that E ∩ F∞ = F, so we have G ∼
= Gal(E∞ /F∞ ). Let

j∞ : X∞ (F) → X∞ (E)G

denote the direct limit of the maps jEn /Fn . For v ∈ V (F∞ ), let Iv denote the inertia group of v in G. Let
M
Σ∞ : Iv → G
v∈V (F∞ )
88 3. IWASAWA THEORY

denote the product of the inclusion maps. We then have a canonical exact sequence of Λ-modules:

0 → ker j∞ ⊗Z G → Ĥ −1 (G, E∞ (E)) → lim IEGn /IFn ⊗Z G → coker j∞ ⊗Z G

←−
n

→ Ĥ (G, E∞ (E)) → ker Σ∞ → X∞ (E)G → X∞ (F) → coker Σ∞ → 0.

0

If F is a CM field and p is odd, then E is also CM, and the sequence of Theorem 3.5.1 is
Gal(F/F + )-equivariant. Taking minus parts, we are able to obtain the following.

L EMMA 3.5.2. Let E/F be a Galois extension of CM fields with G = Gal(E/F) ∼

= Z/pZ, for p
odd. Suppose that µ(X∞− (F)) = 0. Let δ = 1 if µ p ⊂ F and 0 otherwise. Let T denote the set of
primes of F∞+ that split in F∞ /F∞+ , ramify in E∞ /F∞ , and do not lie over p. Then the Herbrand quotient
h(X∞− (E)) exists and equals pδ −|T | . Moroever, µ(X∞− (E)) = 0.

P ROOF. Note that G = G+ and E∞− (E) = Z p (1)δ , so

Ĥ i (G, E∞ (E))− ∼
= Ĥ i (G, E∞− (E)) ∼
= Ĥ i (G, Z p (1))δ .

One checks immediately that Ĥ 0 (G, Z p (1)) ∼

= µ p and Ĥ −1 (G, Z p (1)) = 0. In particular j∞ is injective
on minus parts.
We remark first that IEGn /IFn is generated by the classes of the ramified primes of En that are ramified
over Fn , and is a direct sum of copies of Z/pZ, one for each such prime. Now, a norm compatible
sequence of nontrivial images of primes in the IEGn /IFn as n varies must consist of primes above p, for
a prime ideal not over p in En is inert in En+1 /En for large enough n, and then therefore is not a norm
from the extension. On the other hand, those above p are totally ramified in En+1 /En for large enough
n, so do form part of a unique norm compatible sequence. We therefore have that

lim IEGn /IFn ∼

M
= Iv .
←−
n v∈Vp (F∞ )

Since G is of order p, we have either Iv = G or Iv = 0 if v is a prime of F∞ . We note that Iv− = 0 if

u ∈ V + does not split in F∞ /F + and v lies above u while (Iv ⊕ Iv0 )− ∼
F∞ ∞ = Iu if u splits into v and v0 . Noting
also that G− = 0, we obtain an exact sequence

G → (X∞− (E))G / j∞ (X∞− (F)) → µ pδ → G → X∞− (E)G → X∞− (F) → 0,

M M
(3.5.1) 0→
v∈S p v∈S

where S denotes the set of primes of F∞+ that split in F∞ /F∞+ and ramify in E∞ /F∞ , and S p ⊆ S is the
subset of primes over p.
The exact sequence (3.5.1) tells us that µ((X∞− (E))G ) = 0, since µ(X∞− (F)) = 0. But if AG is
finitely generated over Z p , then A is finitely generated over Z p [G], and hence over Z p since G is finite.
Therefore, we have µ(X∞− (E)) = 0.
 3.5. KIDA’S FORMULA 89

− : X − (F) → X − (E)G is injective and N − : X − (E) → X − (F) is surjective, we have

Since j∞ ∞ ∞ ∞ ∞ ∞

−∼ X∞− (E)G
coker j∞ = NG X∞− (E)
= Ĥ 0 (G, X∞− (E)),
N
ker N∞− ∼
= ker(X∞− (E)G −→ X∞− (E)G ) = Ĥ −1 (G, X∞− (E)).
G

Therefore, we have
| coker j∞−| − −
h(X∞− (E)) = −
= p|S p |+δ −|S | = pδ −|T | .
| ker N∞ |


We are now ready to prove Kida’s formula. Kida’s formula may be thought of as an analogue of
the Riemann-Hurwitz formula, which describes the growth of genus of Riemann surfaces in branched
covers.

T HEOREM 3.5.3 (Kida). Let p be an odd prime, and let E/F be a finite p-extension of CM-number
fields. Let E∞ (resp., F∞ ) be the cyclotomic Z p -extension of E (resp., F), and suppose that E ∩ F∞ = F.
Assume that µ(X∞− (F)) = 0. Then µ(X∞− (E)) = 0, and we have

λ (X∞− (E)) − δ = [E : F](λ (X∞− (F)) − δ ) + ∑ (|Iw | − 1),

w∈QE

where δ = 1 if µ p ⊂ F and 0 otherwise,

QE = {w ∈ V (E∞+ ) −Vp (E∞+ ) | w splits in E∞ /E∞+ },

and Iw is the ramification group of w in Gal(E∞+ /F∞+ ).

P ROOF. First, we reduce the result to cyclic groups of order p by induction on the order of G =
Gal(E + /F + ) ∼
∞ ∞ = Gal(E/F). Let K be an intermediate field in E/F, let G0 = Gal(K/F), and let G00 =
Gal(E/K) (which can be taken to be of order p). For v ∈ VK∞+ , let Iv0 denote the ramification group of v
in G0 , and for w ∈ VE∞+ , let Iw00 denote the ramification group of w in G00 . The statement on µ-invariants
then follows immediately by induction and Lemma 3.5.2. Then, by induction, we have

λ (X∞− (E)) − δ = [E : K](λ (X∞− (K)) − δ ) + ∑ (|Iw00 | − 1)

w∈QE
 
= [E : K] [K : F](λ (X∞− (F)) − δ ) + ∑ (|Iv0 | − 1) + ∑ (|Iw00 | − 1)
v∈QK w∈QE

= [E : F](λ (X∞− (F)) − δ ) + [E : K] ∑ (|Iv0 | − 1) + ∑ (|Iw00 | − 1).

v∈QK w∈QE
90 3. IWASAWA THEORY

For any v ∈ QK and w ∈ QE lying above v, Corollary 3.1.10 tells us that [G00 : Iw00 ] is the number of
primes of QE lying above v. We then have

[E : K] ∑ (|Iv0 | − 1) = |G00| ∑ [G00 : Iw00 ]−1 (|Iv0 | − 1)

v∈QK w∈QE

= ∑ (|Iw | − |Iw00 |) = ∑ (|Iw | − 1) − ∑ (|Iw00 | − 1),

w∈QE w∈QE w∈QE

finishing the inductive step.

Now, we are reduced to the case that [E : F] = p. Note that in this case, a prime w ∈ QE is either
totally ramified (of degree p) or completely split in E/F, so

∑ (|Iw | − 1) = ∑ (p − 1) = (p − 1)|T |,
w∈QE v∈T

where T is, as in Lemma 3.5.2, the set of primes of QF that ramify in E∞+ /F∞+ . By Proposition 3.4.2
and the fact that µ(X∞− (E)) = 0, we have that X∞− (E) is free of finite rank over Z p . It is also a Z p [G]-
module, and therefore
X∞− (E) ∼
= Z p [G]r ⊕ X s ⊕ Ztp
for some r, s,t. It follows immediately that

λ (X∞− (E)) = pr + (p − 1)s + t = p(r + t) + (p − 1)(s − t).

We compute, under these isomorphisms

X∞− (E)G ∼
= (NG )r ⊕ Ztp and NG X∞− (E) ∼
= (NG )r ⊕ (pZ p )t
X∞− (E)[NG ] = X r ⊕ X s and IG X∞− (E) = X r ⊕ (IG X)s ,

so
|Ĥ 0 (G, X∞− (E))|
h(X∞− (E)) = −1 −
= pt−s .
|Ĥ (G, X∞ (E))|
By Lemma 3.5.2, we therefore have that

s − t = |T | − δ .

One sees immediately from Theorem 3.5.1 that the inverse limit of norm maps

X∞− (E)G → X∞− (F)

is a pseudo-isomorphism. We then have that

λ (X∞− (F)) = λ (X∞− (E)G ) = rankZ p (X∞− (E)G ) = r + t.

It follows that

λ (X∞− (E)) − δ = pλ (X∞− (F)) + (p − 1)(|T | − δ ) − δ = p(λ (X∞− (F)) − δ ) + (p − 1)|T |,

 3.5. KIDA’S FORMULA 91

finishing the proof. 

 CHAPTER 4

Cyclotomic fields

4.1. Dirichlet L-functions

In this section, we summarize, largely without proof, various results regarding L-functions of
Dirichlet characters.

D EFINITION 4.1.1. A multiplicative function χ : Z → C is called a Dirichlet character if it is

periodic of some period n ≥ 1 and χ(a) 6= 0 for a ∈ Z if and only if (a, n) = 1. The integer n is called
the modulus of χ.

E XAMPLE 4.1.2. There is a unique Dirichlet character 1 which has value 1 at every a ∈ Z, and it
is known as the trivial character.

D EFINITION 4.1.3.
a. The conductor f χ of a Dirichlet character χ is the smallest integer f dividing its period such that
there exists a Dirichlet character ψ of modulus f with χ(a) = ψ(a) for all a ∈ Z with (a, n) = 1.
b. We say that a Dirichlet character is primitive if its conductor equals its modulus.

D EFINITION 4.1.4. We say that a Dirichlet character χ is even (resp., odd) if χ(−1) = 1 (resp.,
χ(−1) = −1.)

Every character φ : (Z/nZ)× → C× gives rise to a Dirichlet character χ : Z → C of period n with

χ(a) = φ (a (mod n)) for a ∈ Z with (a, n) = 1. The resulting character χ has conductor f , where f is
minimal such that φ factors through (Z/ f Z)× .

D EFINITION 4.1.5. Let φ : (Z/nZ)× → C× , and suppose that the induced Dirichlet character has
conductor f . The primitive Dirichlet character attached to φ is the primitive Dirichlet character of
conductor f that satisfies φ (a) = χ(a0 ) for a ∈ Z, (a, f ) = 1, where a0 ∈ Z is any integer with a0 ≡
a mod f and (a0 , n) = 1.

Let F/Q be an abelian field, and let n ≥ 1 be such that F ⊆ Q(µn ). The cyclotomic character then
allows us to identify Gal(F/Q) with a quotient of (Z/nZ)× .
93
94 4. CYCLOTOMIC FIELDS

N OTATION 4.1.6. The set X(F) of primitive Dirichlet characters of F ⊆ Q(µN ) consists of the
primitive characters of conductor dividing n attached to characters of (Z/nZ)× that factor through
Gal(F/Q).

R EMARK 4.1.7. A Dirichlet character χ ∈ X(F) is even if and only if the associated character on
Gal(F/Q) is even.

To any Dirichlet character, we can attach an L-series.

D EFINITION 4.1.8. Let χ be a Dirichlet character. The Dirichlet L-series attached to χ is the
complex-valued function on s ∈ C with Re s > 1 defined by
∞
χ(n)
L(χ, s) = ∑ s
.
n=1 n

E XAMPLE 4.1.9. For χ = 1, one has L(1, s) = ζ (s), the Riemann ζ -function.

We note that Dirichlet L-series have Euler product expansions.

P ROPOSITION 4.1.10. One has

1
L(χ, s) = ∏ −s
p prime 1 − χ(p)p

for all s ∈ C with Re s > 1.

T HEOREM 4.1.11. The L-series L(χ, s) has a meromorphic continuation to all of C that is analytic
if f χ > 1, while ζ (s) is holomorphic aside from a simple pole at s = 1 with residue 1.

D EFINITION 4.1.12. The Dirichlet L-function L(χ, s) of a Dirichlet character χ is the meromorphic
continuation of the L-series L(χ, s) to C.

D EFINITION 4.1.13. The Γ-function is the unique meromorphic function on C that satisfies
Z ∞
Γ(s) = t s−1 e−t dt
0
for all s ∈ C with Re s > 0 and
Γ(s + 1) = sΓ(s)
for all s for which it is defined.

R EMARK 4.1.14. The Γ-function has poles, which are all simple, at exactly the nonpositive inte-
gers. It also satisfies Γ(n) = (n − 1)! for any positive integer n.

D EFINITION 4.1.15. The Gauss sum attached to a Dirichlet character χ of modulus n is the value
n
τ(χ) = ∑ χ(a)e2πia/n.
a=1
 4.1. DIRICHLET L-FUNCTIONS 95

D EFINITION 4.1.16. For a Dirichlet character χ, we let χ̄ denote its complex conjugate, which
satisfies χ̄(a) = χ(a) for all a ∈ Z.

We mention a couple of basic lemmas regarding Gauss sums that will be of use.

L EMMA 4.1.17. Let χ be a primitive Dirichlet character. Then we have

fχ
χ(b)τ(χ̄) = ∑ χ̄(a)e2πiab/ fχ
a=1

for all b ∈ Z.

P ROOF. If χ(b) = 0, then setting d = (b, f χ ) and m = d −1 f χ , we have

fχ m d
∑ χ̄(a)e2πiab/ fχ = ∑ ∑ χ̄(a + mc)e2πiab/ fχ ,
a=1 a=1 c=1

and
d
∑ χ̄(a + mc) = 0
c=1
for all a. If χ(b) 6= 0, then
fχ
χ(b)τ(χ̄) = ∑ χ̄(ab−1)e2πia/ fχ ,
a=1
which gives the desired equality upon reordering the sum. 

L EMMA 4.1.18. For a primitive Dirichlet character χ, we have

1/2
|τ(χ)| = f χ .

P ROOF. Note that τ(χ) = χ(−1)τ(χ̄). We then have

fχ
|τ(χ)| = χ(−1) ∑ χ(a)τ(χ̄)e2πia/ f χ ,
a=1

and by Lemma 4.1.17, this equals

fχ fχ
! fχ fχ
χ(−1) ∑ ∑ χ̄(b)e2πiab/ fχ e2πia/ f χ = χ(−1) ∑ χ̄(b) ∑ e2πia(b+1)/ f χ .
a=1 b=1 b=1 a=1

The latter sum of exponentials is zero unless b = f χ − 1, in which case it is f χ . Hence,

|τ(χ)| = |χ(−1)|2 f χ = f χ .


96 4. CYCLOTOMIC FIELDS

D EFINITION 4.1.19. For a primitive Dirichlet character χ, we set

 s/2  
τ(χ) fχ s + δχ
δχ = (1 − χ(−1))/2, εχ = δ p , and Λ(χ, s) = Γ L(χ, s),
i χ fχ π 2

T HEOREM 4.1.20. Let χ be a primitive Dirichlet character. Then the L-functions of χ and χ̄ satisfy
the functional equation
Λ(χ, s) = εχ Λ(χ̄, 1 − s)

for all s ∈ C.

We give the relationship between Dirichlet L-functions and the Dedekind zeta function of an abelian
field.

P ROPOSITION 4.1.21. Let F be an abelian field. Then

ζF (s) = ∏ L(χ, s).

χ∈X(F)

P ROOF. It suffices to check this on s with Re s > 1 by uniqueness of the meromorphic continua-
tions. In turn, it suffices to check that for each prime p, we have

(4.1.1) ∏ (1 − (Np)−s ) = ∏ (1 − χ(p)p−s ).

p∈Vp (F) χ∈X(F)

As F/Q is Galois, we have Np = p− f s , where f is the common residue degree of the primes over p
in F, so the lefthand side is just (1 − p− f s )g , where g = |Vp (F)|. Note that χ(p) = 0 if p ramified
in the fixed field of the kernel of χ. Thus, the product reduces to χ ∈ X(E), where E is the maximal
subextension of F/Q that is unramified at p. Viewing χ ∈ X(E) as a Galois character, so χ(p) is the
value of χ on the Frobenius at p, which is a generator of a cyclic subgroup of order f in Gal(E/Q).
Since f g = [E : Q], there are g characters χ such that χ( f ) = ζ fi for a fixed primitive f th root of unity
ζ f and given integer i with 0 ≤ i ≤ f − 1. The righthand side of (4.1.1) is then simply
f −1
∏ (1 − ζ f p−s)g = (1 − p− f s)g,
i=0

as required. 

C OROLLARY 4.1.22. Let χ be a Dirichlet character with associated primitive character nontrivial.
Then L(χ, 1) 6= 0.

P ROOF. Since ζF (s) has a simple pole at s = 1, as does L(χ0 , s), for χ0 the trivial character of
modulus [F : Q], while L(χ, s) is analytic for χ 6= χ0 , this is a direct result of Proposition 4.1.21. 
 4.2. BERNOULLI NUMBERS 97

4.2. Bernoulli numbers

t
D EFINITION 4.2.1. For n ≥ 0, the nth Bernoulli number Bn is the value of the nth derivative of et −1
at 0.

In other words, Bn is the rational number appearing in the Taylor expansion

t ∞
tn
= B
∑ n n!
et − 1 n=0

E XAMPLE 4.2.2. We have

et − 1 ∞
tn 1 1
=∑ = 1 + t + t2 + · · · ,
t n=0 (n + 1)! 2 6

so B0 = 1, B1 = − 21 , and B2 = 1
6 after inverting the series.

R EMARK 4.2.3. Note that

−t tet t
−t
= t
= t + t,
e −1 e −1 e −1
so
t 1
+ t
−1 2 et
is an even function, and therefore we have Bn = 0 for all odd n ≥ 2.

We shall require generalizations of these numbers attached to Dirichlet characters.

D EFINITION 4.2.4. Let χ be a primitive Dirichlet character, and let m be any multiple of f χ . Then
the generalized Bernoulli number Bn,χ is the algebraic number appearing in the series expansions
m
teat ∞
tn
∑ χ(a) emt − 1 = ∑ Bn,χ n! .
a=1 n=0

R EMARK 4.2.5. The independence from m in the definition of Bn,χ is easily seen to boil down to
the fact that
r−1
xi 1
∑ xr − 1 = x − 1 ,
i=0

taking r = m/ f χ and x = e fχ t .

1
R EMARK 4.2.6. We have Bn,1 = Bn for all n ≥ 2, but B1,1 = 2 = −B1 .

R EMARK 4.2.7. We have that Bn,χ = 0 for n 6≡ δχ mod 2, aside from B1,1 .

We also have Bernoulli polynomials.

98 4. CYCLOTOMIC FIELDS

D EFINITION 4.2.8. The nth Bernoulli polynomial Bn (X) ∈ Q[X] is the polynomial appearing in the
series expansion
teXt ∞
tn
= ∑ n n! .
B (X)
et − 1 n=0

E XAMPLE 4.2.9. We have B0 (X) = 1 and B1 (X) = X − 12 .

L EMMA 4.2.10. Let χ be a primitive Dirichlet character, and let m be a multiple of f χ . We have
m
Bn,χ = mn−1 ∑ χ(a)Bn( ma )
a=1

for n ≥ 1.

P ROOF. We have
m  a  tn m  a  (mt)n m
∞ ∞
teat
∑ mn−1 ∑ χ(a)Bn m n!
= ∑ χ(a)m−1 ∑ Bn m n!
= ∑ χ(a) emt − 1
.
n=0 a=1 a=1 n=0 a=1

C OROLLARY 4.2.11. Let χ be a primitive, nontrivial Dirichlet character of conductor dividing m.

Then we have
1 m
B1,χ = ∑ χ(a)a.
m a=1

P ROOF. We compute easily that B1 (x) = x − 1/2. The result then follows from Lemma 4.2.10 and
the fact that the sum over all χ(a) for 1 ≤ a ≤ m is zero, since χ is nontrivial. 

D EFINITION 4.2.12. A value of L(χ, s) at s ∈ Z is known as an L-value, or as a special value of the

L-function L(χ, s).

The following proposition gives a relationship between L-values and generalized Bernoulli num-
bers.

P ROPOSITION 4.2.13. Let χ be a primitive Dirichlet character. Then we have

Bn,χ
L(χ, 1 − n) = −
n
for all positive integers n.

P ROOF. Let x ∈ R with 0 < x ≤ 1, and consider the complex function

te(1−x)t ∞
tn
f (t) = = ∑ B n (1 − x) .
et − 1 n=0 n!
 4.2. BERNOULLI NUMBERS 99

For s ∈ C, set Z
g(s) = lim f (t)t s−2 dt,
ε→0+ γε
where the path γε consists of the horizontal infinite path along the real axis to ε, following by a coun-
terclockwise traversal around the circle Cε of radius ε, followed by the horizontal infinite path from ε
along the positive real axis. Here, t s−2 = e(s−2) logt , where we take the branch of the logarithm given
by the positive real axis. Then
 Z ∞ Z 
2πis s−2 s−2
g(s) = lim (e − 1) f (t)t dt + f (t)t dt .
ε→0+ ε Cε

If Re s > 1, the second term vanishes in the limit, and this simplifies to
Z ∞ ∞ Z ∞ ∞
−1
(e2πis
− 1) g(s) = f (t)t s−2
dt = ∑ t s−1 e−(x+k)t dt = ∑ (x + k)−sΓ(s) = Γ(s)ζ (s, x),
0 k=0 0 k=0

where we set ζ (s, x) = ∑∞ −s

k=0 (x + k) . The latter function can be meromorphically continued to all of
C which is again analytic away from s = 1. We therefore have

g(s) = (e2πis − 1)Γ(s)ζ (s, x)

for all s ∈ C − {1}.

For s = 1 − n, we obtain
Bn (1 − x)
Z
2πis
lim (e − 1)Γ(s)ζ (s, x) = lim f (t)t −1−n dt = 2πi ·
s→1−n ε→0+ Cε n!
by Cauchy’s integral formula. We have
(−1)n−1
lim (e2πis − 1)Γ(s) = 2πi lim sΓ(s) = 2πi ,
s→1−n s→1−n (n − 1)!
so we obtain
Bn (1 − x) Bn (x)
ζ (1 − n, x) = (−1)n−1 =− .
n n
Finally, setting f = f χ , we need only note that
f f
n−1 1 Bn,χ
L(χ, 1 − n) = ∑ χ(a) f ζ (1 − n, a
f ) = −n ∑ χ(a) f n−1Bn( af ) = − n
.
a=1 a=1

T HEOREM 4.2.14. Let χ be a nontrivial primitive Dirichlet character. We have


πiτ(χ)
 f χ B1,χ̄ if χ is odd,



L(χ, 1) = f
τ(χ) χ 2πia/ f χ
− f ∑ χ̄(a) log |1 − e | if χ is even .



χ a=1
100 4. CYCLOTOMIC FIELDS

P ROOF. If χ is odd, then the functional equation and the fact that Γ(1/2) = π 1/2 imply that
πiτ(χ) πiτ(χ)
L(χ, 1) = − L(χ̄, 0) = B1,χ̄ .
fχ fχ
Now let χ be even, and set f = f χ . By Lemma 4.1.17, we then have
f f
∞
χ(n) ∞
1 χ̄(a)e2πian/ f 1
L(χ, 1) = ∑ n = ∑ τ(χ̄) ∑ n
= −
τ( χ̄) ∑ χ̄(a) log (1 − e2πia/ f ).
n=1 n=1 a=1 a=1

By Lemma 4.1.18 (and Lemma 4.1.17), we have that τ(χ̄)τ(χ) = f , and the evenness of χ̄ plus the
fact that the sum is taken over all a mod f tell us that we may replace log (1 − e2πia/ f ) with
1
log |1 − e2πia/ f | = (log(1 − e2πia/ f ) + log(1 − e2πi( f −a)/ f )).
2


Combining the analytic class number formula with Proposition 4.1.21 and Theorem 4.1.11, we
obtain the following, which we will at times also refer to as the analytic class number formula.

C OROLLARY 4.2.15. Let F be an abelian field. Then we have

2r1 (F) (2π)r2 (F) hF RF
∏ L(χ, 1) = wF |dF |1/2
.
χ∈X(F)
χ6=1

We note the following.

L EMMA 4.2.16. Let F be a CM field. Set QF = [EF : µ(F)EF+ ]. Then QF ∈ {1, 2} and
QF
[EF : EF+ ] = 2 wF .

P ROOF. Let τ be the generator of Gal(F/F + ). For α ∈ EF , we have |α 1−τ | = 1 under any complex
embedding of F, so α 1−τ ∈ µ(F). Consider the commutative diagram

1 1 1
  
1 / h−1i / µ(F) / µ(F)2 /1

  
1 / E+ / EF 1−τ
/ µ(F)
F

  
1 / E + /h−1i / EF /µ(F) 1−τ
/ µ(F)/µ(F)2
F

  
1 1 1
 4.2. BERNOULLI NUMBERS 101

The snake lemma tells us that the cokernels K of the two maps τ − 1 are isomorphic. The lower two
rows yield
wF 2
[EF : EF+ ] = and [EF : µ(F)EF+ ] = ,
|K| |K|
and the result follows. 

We remark that for cyclotomic fields, QF is computable.

L EMMA 4.2.17. Let F = Q(µm ) for some m ≥ 1 with m 6≡ 2 mod 4. Then


1 m is a prime power
QF =
2 otherwise.

P ROOF. Let τ be the generator of Gal(F/F + ). Note that

1−τ
QF = 2|coker(EF −−→ µ(F))|−1

by the proof of Lemma 4.2.16. If m is not a prime power, then 1 − ζm is a unit, and (1 − ζm )1−τ = −ζm ,
which generates µ(F). Thus QF = 2 in this case. Conversely, if α 1−τ = −ζm generates µ(F) for some
α ∈ EF , we would have α −1 (1 − ζm ) ∈ F + . If m were a power of a prime p, then α −1 (1 − ζm ) would
generate the unique prime over p in F. Since this prime is ramified in F/F + , its generator cannot lie
in F + . This forces QF to be 1 if m is a prime power. 

N OTATION 4.2.18. For a CM field F, we set R+

F = RF +

L EMMA 4.2.19. Let F be a CM field. Then

wF
RF = 2r2 (F)−2 +
+ RF .
[EF : EF ]
P ROOF. Let
r = r2 (F) − 1 = rank EF = rank EF+ .
Suppose that α1 , α2 , . . . , αr ∈ EF+ satsify

h−1, α1 , α2 , . . . , αr i = EF+ .

Then
µ(F) · hα1 , α2 , . . . , αr i = µ(F)EF+ ,
which has index 2[EF : EF+ ]/wF in EF , so Lemma 1.2.10 tells us that
wF
RF = RF (α1 , α2 , . . . , αr ).
2[EF : EF+ ]
102 4. CYCLOTOMIC FIELDS

On the other hand, note that each ci in Definition 1.2.5 is 2 for F but 1 for F + , so

RF (α1 , α2 , . . . , αr ) = 2r R+
F,

as desired. 

Corollary 4.2.15 implies the following.

T HEOREM 4.2.20. Suppose that F is a CM abelian field. Then

f
!
−B1,χ 1 −1 χ
h−
F = 2[EF : EF+ ] ∏ 2 and h+
F = + ∏ ∑ χ(a) log |1 − e2πia/ f χ
| .
χ∈X(F) RF χ∈X(F)
2 a=1
χ odd χ6=1 even

P ROOF. Let E be an arbitrary abelian field. We remark that for χ ∈ X(E), the quantity f χ is the
conductor of the corresponding character (Z/ f χ Z)× → C× . Therefore, the conductor-discriminant
fomula tells us that

(4.2.1) |dE | = ∏ fχ .
χ∈X(E)

Moreover, a comparison of the functional equations of the Dirichlet L-functions and the Artin L-
functions yields that
∏ εχ = 1,
χ∈X(E)
so

(4.2.2) ∏ τ(χ) = ir2 (E) |dE |1/2 .

χ∈X(E)

Taking the quotient of the analytic class number formula for F by that for F + and applying Theo-
rem 4.2.14, we obtain
πiτ(χ) π r2 (F) RF /R+
(4.2.3) ∏ f χ 1,χ̄ |dF /d + |1/2 wF /wFF+ h−F .
B =
χ∈X(F) F
χ odd

Applying (4.2.1) and (4.2.2) for E = F and E = F + , we see that

πiτ(χ) (−π)r2 (F)
∏ fχ = ,
χ∈X(F) |dF /dF + |1/2
χ odd

and Lemma 4.2.19 tells us that

RF /R+
F
= 2r2 (F)−1 [EF : EF+ ],
wF /wF +
since wF + = 2. Equation (4.2.3) is then immediately reduced to the desired form.
 4.3. CYCLOTOMIC UNITS 103

On the other hand, the analytic class number formula for F + and Theorem 4.2.14,
f
! +
−τ(χ) χ 2πia/ f χ 2r1 (F ) h+
F RF
+
∏ ∑ χ̄(a) log |1 − e
f χ a=1
| =
2|dF + |1/2
,
χ∈X(F)
χ6=1 even

Applying (4.2.1) and (4.2.2) and noting that replacing χ̄(a) by χ(a) in the resulting sum makes no
difference in the result, we obtain the formula for h+
F. 

4.3. Cyclotomic units

The product appearing in the formula for h+

F in Theorem 4.2.20 may appear itself something like a
regulator. This is essentially the case.

D EFINITION 4.3.1. If F is an abelian field contained in Q(µm ) for m ≥ 1, we let S = Vm∞ and
define the group of cyclotomic S-units CF,S of F to be the subgroup

CF,S = h1 − ζma | 1 ≤ a < mi ∩ F ×

×
of OF,S , where ζm is a primitive mth root of unity. The group of cyclotomic units of F is then defined
as the intersection CF = EF ∩CF,S .

R EMARK 4.3.2. The definition of CF is independent of the multiple m of the conductor of F + .

We have the following result of Hasse, which is due to Kummer in the case of Q(µ p ) for a prime
p. We will prove a generalization of this result to arbitrary cyclotomic fields in Theorem 4.7.1.

T HEOREM 4.3.3 (Hasse). Let F = Q(µ pn ) for an odd prime p and n ≥ 1. Then we have

h+ + +
F = [EF : CF ].

P ROOF. The set  

a/2 −a/2
 ζ pn − ζ pn n

ξa = 1/2 −1/2
1 < a < p /2, (a, p) = 1
 ζ pn − ζ pn 

forms an independent set of generators of CF+ . Let us let Rcyc denote the regulator of the latter set.
Then Rcyc is the absolute value of the determinant of the matrix with rows and columns indexed by the
integers a prime to p with 1 < a < pn /2 with entries in the row and column corresponding to (a, b)
given by log |σa (ξb )|, where σa (ζ pn ) = ζ pan . Now

log |σa (ξb )| = log |1 − ζ pabn | − log |1 − ζ pan |.

104 4. CYCLOTOMIC FIELDS

Proposition 1.5.18 applied to the group Gal(F + /Q) yields

n
pn −1
!
p /2−1
1
Rcyc = ∏+ ∑ χ(b) log |1 − ζ pbn | = ∏+ 2 ∑ χ(c) log |1 − ζ pcn | .
χ∈X(F ) b=1 χ∈X(F ) c=1
χ6=1 (b,p)=1 χ6=1 (c,p)=1

As χ has conductor dividing pn and

k−1
c+ jk
1 − ζnc = ∏ (1 − ζnk )
j=0

for n, k ≥ 1 and c 6≡ 0 mod n, we have

pn −1 f χ −1
∑ χ(c) log |1 − ζ pcn | = ∑ χ(c) log |1 − ζ fcχ |,
c=1 c=1
(c,p)=1 (c,p)=1

the middle step by Theorem 4.2.14. By Theorem 4.2.20, it then follows that Rcyc = h+ +
F RF . On the
other hand, we have Rcyc = R+ + +
F [EF : CF ] by Lemma 1.2.10. 

A standard choice of primitive mth roots of unity for m ≥ 1, viewing Q as a subset of C, is to take
n/m
ζm = e2πi/m for m ≥ 1. This choice has the advantage that ζn = ζm for m dividing n. Let us make
such a choice. We first remark that the elements 1 − ζm for m divisible by two distinct primes are in
fact units.

L EMMA 4.3.4. If m is divisible by two distinct primes, then 1 − ζm ∈ CQ(µm ) .

P ROOF. For a positive integer d, let Φd denote the dth cyclotomic polynomial. We have
m
Φm (1) = ∏ (1 − ζmi ),
i=1
(i,m)=1

so it suffices to show that Φm (1) = ±1. We have

xm − 1
= ∏ Φd (x).
x−1 d|m
d>1
Plugging in x = 1, we obtain
m= ∏ Φd (1).
d|m
d>1
Note Φ pk (1) = pk for any power pk of a prime p. Expressing m = ∏gi=1 pki i as a product of powers of
distinct primes pi , we then also have
g
m = ∏ Φ pki (1).
i=1 i

Since each Φd (1) is an integer, it follows that Φm (1) = ±1, as desired. 

 4.4. REFLECTION THEOREMS 105

Next, we note the following the compatibility of the elements 1 − ζm under norms.

L EMMA 4.3.5. For m ≥ 1 and a prime `, we have


1 − ζm
 if ` | m,
NQ(µm` )/Q(µm ) (1 − ζm` ) = `−1 1 − ζm
−ζm
 −1 if ` - m.
1 − ζm`
P ROOF. Note that
`
∏(1 − ζm`ζ`i) = 1 − ζm.
i=1
If ` divides m, then the left-hand side runs over the conjugates of 1 − ζm` under Gal(Q(µm` )/Q(µm )),
so the product equals the norm.
If ` does not divide m, then let a, b ∈ Z with a` + bm = 1. We then have ζma ζ`b = ζm` , so the
conjugates of ζm` have the form ζm` ζ`i with i 6≡ −b mod `. Note that b ≡ m−1 mod `, so moving this
term from the product to the other side, we have
1 − ζm −1 1 − ζm
NQ(µm `)/Q(µm ) (1 − ζm` ) = −`−1
= −ζm` −1 .
1 − ζm 1 − ζm`


4.4. Reflection theorems

We now refine Theorem 1.4.15 by working with eigenspaces. Start with a totally real field F. Let
×
χ : GF → Q

be a character with finite image. Any embedding ϕ of F in C fixes an element cϕ ∈ GF that is the
restriction of complex conjugation in Gal(C/R), since F is taken to a subfield of R under the embed-
ding. All such complex conjugations in GF arise in this way, and they form [F : Q] distinct conjugacy
classes in GQ for the real embeddings of F in F = Q. In Gab
F , these complex conjugations restrict to
exactly [F : Q] distinct elements, with the elements of the same class restricting to the same element.
×
D EFINITION 4.4.1. We say that a character χ : GF → Q of a totally real field F is totally even if
χ is trivial on all complex conjugations and totally odd if χ is nontrivial on all complex conjugations.
If F = Q, we say more simply that χ is even or odd in the respective cases.

We let Fχ denote the extension of F that is the fixed field of the kernel of χ, which will itself be
totally real if χ is totally even and CM if χ is totally odd. If F = Q, these are the only cases.
We now suppose that χ has order prime to a given odd prime p. We fix an embedding ι p : Q → Q p ,
× ×
which allows us to view χ as a character with values in Q p , and hence in Z p .
106 4. CYCLOTOMIC FIELDS

One key character of interest to us is the Teichmüller character

ω : GF → Q×
p

which has image contained in µ p−1 (Z p ) and is defined by the equality

σ (ζ ) = ζ ω(σ )

for any σ ∈ GF and ζ ∈ µ p . Note that the Teichmüller character is an odd character on GF .
×
T HEOREM 4.4.2 (Leopoldt’s Spiegelungsatz). Let F be a totally real field, and let χ : GF → Q p
be a totally odd character of finite order prime to p. Let E be an abelian extension of F of degree prime
to p that contains Fχ (µ p ). Then we have
(ω χ −1 ) (ω χ −1 ) −1 )
) + r p ((OE× /OE×p )(ω χ
(χ)
r p (AE ) − δχ ≤ r p (AE ) ≤ r p (AE ),

where δχ is 0 unless χ = ω and the extension E(µ(F)1/p )/E is unramified, in which case it is 1.

P ROOF. Let ∆ = Gal(E/F). Let O be the ring generated over Z p by the character values of ∆. Let
k denote the residue field of O, and let kψ denote the residue field of Oψ , the ring of values of ψ, for
any ψ ∈ ∆∗ . As O is unramified over Z p , we have [O : Oψ ] = [k : kψ ].
For a Z p [∆]-module B, we let BO = B ⊗Z p O. We remark that Lemma 2.8.7 implies that

r p (BO ) = [k : kψ ]r p (B(ψ) ).
ψ

Note also that we have

ψ ψ
r p (BO ) = [k : F p ] dimk ((B/pB)O ),
so

r p (B(ψ) ) = [kψ : F p ]−1 dimk ((B/pB)O ).

ψ
(4.4.1)

Since Oχ = Oω χ −1 and since δχ = 0 unless χ = ω, in which case kχ = F p , equation (4.4.1) tells us

that the desired inequalities are equivalent to
ω χ −1 ω χ −1 ω χ −1
) + dimk ((OE× /OE×p )O
χ
dimk (AO ) − δχ ≤ dimk (AO ) ≤ dimk (AO ),

where we have set A = AE /pAE to shorten notation.

Note that we have the following isomorphisms of groups

HomZ p (AE , µ p )O ∼
= HomZ p (AE , (µ p )O ) ∼
= HomO ((AE )O , (µ p )O )

the first step following from the freeness of O over Z p and the second from the adjointness of Hom
and ⊗. Moreover, Lemma 2.8.7 implies that
−1
HomO ((AE )O , (µ p )O )ψ ∼
ωψ
= HomO ((AE )O , (µ p )O )
 4.4. REFLECTION THEOREMS 107

for any ψ ∈ ∆∗ . Recalling Lemma 1.4.6, we then have an exact sequence

ωψ −1
0 → ((B ∩ OE× )/OE×p )O → HomO ((AE )O
ψ ψ
, (µ p )O ) → (AE )O [p],

where B is the set of elements of E × that have pth roots that generate unramified extensions of E.
Since (µ p )O is a one-dimensional k-vector space, we have

ωψ −1 −1
, (µ p )O ) ∼
ωψ
HomO ((AE )O = Homk (AO , k),

ωψ −1
which as the k-dual of a k-vector space has dimension equal to dimk (AO ).
In the case that ψ = χ, we then have that

ω χ −1 (χ) (χ)
) ≤ dimk (((B ∩ µ(E))/µ(E) p )O ) + dimk (AO ) = δχ + dimk (AE ),
χ
dimk (AO

since the p-power roots of unity in E have trivial χ-eigenspace unless [χ] = [ω], which happens if and
only if χ = ω, as ω takes its values in Z p . On the other hand, if we take ψ = ω χ −1 , then we have

ω χ −1 ω χ −1
dimk (AO ) ≤ dimk ((OE× /OE×p )O
χ
) + dimk (AO ),

finishing the proof. 

In the special case that F = Q and E = Q(µ p ), we remark that δω = 0, as Q(µ p2 )/F is ramified at
the unique prime over p. Moreover, we have the following.

L EMMA 4.4.3. Let k be an even integer. Then


× (ω k ) ∼
Z
p k 6≡ 0 mod (p − 1)
(OQ(µ p ) ⊗Z Z p ) = .
0 k ≡ 0 mod (p − 1)

C OROLLARY 4.4.4. For any even integer k, we have

(ω k ) (ω 1−k ) (ω k )
r p (AQ(µ p ) ) ≤ r p (AQ(µ p ) ) ≤ r p (AQ(µ p ) ) + 1.

(ω) (1)
C OROLLARY 4.4.5. We have AQ(µ p ) = AQ(µ p ) = 0.

P ROOF. We know that

(1) Gal(Q(µ p )/Q) ∼
AQ(µ p ) = AQ(µ p ) = AQ = 0.

(1) (ω)
As Theorem 4.4.2 and Lemma 4.4.3 tell us that r p (AQ(µ p ) ) = r p (AQ(µ p ) ), so we are done. 
108 4. CYCLOTOMIC FIELDS

4.5. Stickelberger theory

Let us fix an integer m ≥ 1 and a primitive mth root of unity ζm throughout this section.

D EFINITION 4.5.1. Let F = Q(µm ), and let G = Gal(F/Q).

a. For a ∈ Z with (a, m) = 1, let σa ∈ G be such that σa (ζm ) = ζma . The Stickelberger element θF
is the element of Q[G] given by
m
1
θF = ∑ aσa−1 .
m a=1
(a,m)=1

b. The Stickelberger ideal of F is the ideal IF = Z[G]θF ∩ Z[G] of Z[G].

L EMMA 4.5.2. Let J denote the ideal of Z[G] generated by elements of the form σb − b for b ∈ Z
with (b, m) = 1. Then J = {x ∈ Z[G] | xθF ∈ Z[G]}.

P ROOF. Let us use hαi to denote the fractional part of α ∈ Q. We note

m m  
a −1 ab
σb θF = ∑ σb σa = ∑ σa−1 .
a=1 m a=1 m
(a,m)=1 (a,m)=1

m i − h m ib ∈ Z, we have (σb − b)θF ∈ IF for all b ∈ Z prime to m, and hence JθF ⊆ Z[G].
Since h ab a

Now take x = ∑b eb σb with xθF ∈ Z[G]. Writing this out, we have

m m  !
ab
∑ ∑ eb m σa−1 ∈ Z[G],
a=1 b=1
(a,m)=1 (b,m)=1

which implies that

m
∑ eb b ∈ mZ.
b=1
(b,m)=1
But note that m = (m + 1) − σ1 ∈ J, so mZ ⊂ J. We then have
m m
x= ∑ eb (σb − b) + ∑ eb b ∈ J,
b=1 b=1
(b,m)=1 (b,m)=1

finishing the proof. 

D EFINITION 4.5.3. Let q be a power of a prime ` and χ : F× ×

q → C be a character, which we
extend to a function χ : Fq → C by χ(0) = 0. The Gauss sum attached to χ is

g(χ) = − ∑× χ(α)e2πi Tr(α)/`

α∈Fq

where Tr = TrFq /F` is the trace map.

 4.5. STICKELBERGER THEORY 109

L EMMA 4.5.4. Let q be a power of a prime ` prime to m. Let χ : F×

q → µm be a character, so
g(χ) ∈ Q(µ`m ). Let b ∈ Z be relatively prime to m, and let σb ∈ Gal(Q(µ`m )/Q(µ` )) be the unique lift
of σb ∈ G. Then
g(χ)σb −b ∈ Q(µm ).
In particular, we have g(χ)m ∈ Q(µm ).

P ROOF. For τ ∈ Gal(Q(µ`m )/Q(µm )) with τ(ζ` ) = ζ`c , we have

g(χ)τ = − ∑ χ(α)e2πi Tr(cα)/` = χ(c)−1 g(χ).

α∈Fq

On the other hand, we have g(χ)σb = g(χ b ) as σb fixes µ` , so we see that

(g(χ)σb −b )τ = g(χ b )τ g(χ)−bτ = χ b (c)−1 χ(c)−b g(χ)σb −b = g(χ)σb −b ,

as desired. 

L EMMA 4.5.5. Let q be a power of a prime ` and χ : F× ×

q → C be a character. Then

g(χ)g(χ̄) = χ(−1)`.

We state Stickelberger’s theorem for F = Q(µm ). A similar result holds for abelian fields in general.

T HEOREM 4.5.6 (Stickelberger). Let F = Q(µm ), set G = Gal(F/Q). Then the Stickelberger ideal
of F annihilates the class group: IF · ClF = 0.

P ROOF. Fix C ∈ ClF , and let l be a prime ideal representing C in OF that lies above a completely
split prime ` of Q. Note that ` ≡ 1 mod m, and let c ∈ Z be a primitive root modulo `. Let χ : F×
` →
C× denote the character with χ(c) = e2πi/m . There is unique prime L of Q(µ`m ) lying above l, and
L`−1 = l · Z[µ`m ]. We use vL to denote the additive valuation attached to L. For b ∈ Z prime to m, and
σb ∈ Gal(Q(µ`m )/Q(µ` )) the unique lift of σb ∈ Gal(F/Q), we set

tb = vσ −1 L (g(χ)).
b

By Lemma 4.5.5, we have that g(χ) | (`), so tb ≤ ` − 1, and by Lemma 4.5.4, we have in the smaller
field F that
vσ −1 l (g(χ)`−1 ) = tb .
b

In other words, we have the factorization

m
g(χ)`−1 OF = ∏ (σb−1 l)tb ,
b=1
(b,m)=1
110 4. CYCLOTOMIC FIELDS

so
m
∑ tb σb−1
b=1
(b,m)=1
annihilates the class of l.
Now take τ ∈ Gal(F(µ` )/F) given by τ(ζ` ) = ζ`c . Then since every prime over ` is totally ramified
F(µ` )/F, we have that τ is in the inertia group of all such primes. Note that

vσ −1 L (ζ` − 1) = 1
b

for all b. We calculate

g(χ) g(χ)τ χ(c)−1 g(χ)
≡ ≡ mod σb−1 L.
(ζ` − 1)tb (ζ`c − 1)tb ctb (ζ` − 1)tb
This forces e2πi/m ≡ c−tb mod σb−1 L and therefore modulo σb−1 l, since both sides of the latter congru-
ence lie in F. In other words, we have

e2πib/m ≡ c−tb mod l.

On the other hand, there exists some a prime to m such that

e2πi/m ≡ c−(`−1)a/m mod l.

We therefore have that

(` − 1)ab
tb ≡ mod (` − 1),
m
forcing  
ab
tb = (` − 1) .
m
It follows that
m  
ab
(` − 1) ∑ σb−1 = (` − 1)σa θF
b=1 m
(n,m)=1
annihilates C.
Now suppose x ∈ Z[G] is such that xθF ∈ IF . We then have
−1
(g(χ)σa x )`−1 OF = l(`−1)xθF .
−1 x
By Lemmas 4.5.2 and 4.5.4, we have g(χ)σa ∈ F. Therefore, the identity
−1
(g(χ)σa x )OF = lxθF

actually holds, and so we see that xθF annihilates C. So, IF annihilates C, and we are done. 

This has an interesting application for the field Q(µ p ).

 4.6. DISTRIBUTIONS 111

T HEOREM 4.5.7 (Herbrand). Let p be an odd prime, and set F = Q(µ p ). Let j 6≡ 1 mod p − 1 be
(ω j ) (ω)
an odd integer, and suppose that AF 6= 0. Then B1,ω − j ∈ pZ p . Moreover, we have AF = 0.

P ROOF. By Stickelberger’s theorem, we have that IF · AF = 0. In particular, we have that I j =

(ω j )
eω j IF annihilates AF , where eω j ∈ Z p [G] is the idempotent attached to ω j . Note that for, b prime
to p, we have

1 p−1 − j
eω j (σb − b)θF = (ω j (b) − b) ∑ aω (a)eω j = (ω j (b) − b)B1,ω − j eω j ,
p a=1

where we have applied Corollary 4.2.11 in the last step. It follows that (ω j (b) − b)B1,ω − j annihilates
(ω j )
AF for all b prime to p. Choosing b to be a primitive root of 1, we have that ω j (b) 6≡ b mod p, so if
(ω j )
AF is nontrivial, then B1,ω − j must be divisible by p. For j = 1, we note that
p
(ω j (1 + p) − (1 + p))B1,ω −1 = −pB1,ω −1 = − ∑ ω(a)−1 a ≡ 1 mod p,
i=1

(ω)
so we get that 1 annihliates AF , hence the result. 

As with the plus part, the minus part of the class number of a cyclotomic field of prime power roots
of unit can be interpreted as an index, as in the following result of Iwasawa. The proof is deferred to
its generalization to arbitrary cyclotomic fields in Theorem 4.7.1.

T HEOREM 4.5.8 (Iwasawa). Let F = Q(µ pn ) for a prime p and n ≥ 1. Then

h− − −
F = [Z[G] : IF ].

4.6. Distributions

D EFINITION 4.6.1. Let {Xi | i ∈ I} be a collection of finite sets, were I is a directed set under ≤,
and let πi j : Xi → X j for i ≥ j be a collection surjective maps. Let A be an abelian group. An A-valued
distribution on the collection (Xi , πi j ) is a set of maps ψi : Xi → A for i ∈ I that satisfy the distribution
relation
ψ j (x) = ∑ ψi (y)
y∈πi−1
j (x)

for all j ≤ i and x ∈ Xi .

R EMARK 4.6.2. Given a collection (Xi , πi j ) as above, we may consider the inverse limit

X = lim Xi .
←−
i∈I
112 4. CYCLOTOMIC FIELDS

Let πi : X → Xi be the map induced by the system. Let Step(X, A) denote the set of A-valued step
functions on X. Supposing now that A is a ring, a distribution {ψi : Xi → A | i ∈ I} on the collection
(Xi , πi j ) (or more simply, on X) gives rise to an A-module homomorphism

ψ̃ : Step(X, A) → A

as follows. If χY denotes the characteristic function of a compact-open subset Y of X, then we let

ψ̃(χπ −1 (x) ) = ψi (x)

i

for any i ∈ I and x ∈ Xi . We take ψ̃ as the A-linear extension of this map to the group of all step
functions. The distribution relation insures that it is well-defined. Conversely, given an A-module
homomorphism ψ̃ : Step(X, A) → A, we may define ψi (x) to be ψ̃(χπ −1 (x)), and the ψi provide a
i
distribution on X.

E XAMPLE 4.6.3. Let I be the set of positive integers, ordered in the usual manner. Let Xi = Z/pi Z,
and let πi j for j ≤ i be the reduction modulo p j map. Let a ∈ Z p . Define

1 if x ≡ a mod pi ,
ψi (x) =
0 otherwise.

Then {ψi | i ≥ 0} is an R-valued distribution for any ring R, called the δ -distribution at a. The corre-
sponding functional δa satisfies δa ( f ) = f (a), where f ∈ Step(Z p , R) is any congruence function.

Let us focus on a specific case of interest.

D EFINITION 4.6.4. Let A be an abelian group, and let D be a divisible abelian group with finitely
topologically generated Pontryagin dual.
a. By an A-valued distribution on D, we mean a function ψ : D → A with the property that

(4.6.1) ψ(d) = ∑ ψ(c)

c∈D
nc=d

for all positive integers n and d ∈ D.

b. By an A-valued punctured distribution on D, we mean a function ψ : D − {0} → A satisfying
the distribution relation (4.6.1) for all positive integers n and d ∈ D − {0}.

R EMARK 4.6.5. For an abelian group A and a torsion divisible abelian group D, the A-valued distri-
butions on D are in one-to-one correspondence with the A-valued distributions {ψn | n ≥ 1} on the col-
lection of n-torsion subgroups D[n] in D for n ≥ 1, together with the transition maps πn,m : D[n] → D[m]
for m dividing n given by multiplication by mn . That is, ψ and the maps ψn take the same values on the
 4.6. DISTRIBUTIONS 113

elements of 1n Z/Z. If A is a ring, then the maps ψ also give rise to a functional ψ̃ : Step(limn D[n], A) →
←−
A, as noted above.

R EMARK 4.6.6. Punctured distributions on D do not quite give rise to distributions on the sets
n
D[n] − {0}, since multiplication by m does not preserve these sets.

E XAMPLE 4.6.7. Let I be the set of positive integers, ordered by divisibility. Fix k ≥ 0, and for
0 ≤ a < n with n ≥ 1, set
a a
(k)
ψn = nk−1 Bk .
n n
For m dividing n, we have
n/m−1   n−1  
(k)
a
k−1
a
k−1 a + jm (k) b
ψn =m Bk = ∑ n Bk = ∑ ψn .
m m j=0 n b=0 n
b≡a mod m

Thus, we can safely make the following definition.

D EFINITION 4.6.8. For k ≥ 0, the kth Bernoulli distribution ψ (k) is the Q-valued distribution on
Q/Z defined by
a D a E
(k) k−1
ψ = n Bk ,
n n
where hαi denotes the smallest nonegative rational number representing α ∈ Q/Z.

We also mention the following example of something close to a distribution.

E XAMPLE 4.6.9. Define ψ : Q/Z − {0} → Q(µ∞ )× by ψ( ni ) = 1 − ζni . If m | n and i 6≡ 0 mod m,

we have
  n/m−1 n−1  
i j
ψ = 1 − ζmi = ∏ (1 − ζni+km ) = ∏ ψ ,
m k=0 j=0 n
j≡i mod m

so ψ satisfies the distribution relations under multiplication. Thus ψ is a punctured distribution on

Q/Z.

We will be interested in the following resulting distribution.

N OTATION 4.6.10. Let ψcyc be the R-valued punctured distribution on Q/Z given by
1
ψcyc (α) = − log |1 − e2πihαi |
2
for α ∈ Q/Z.
114 4. CYCLOTOMIC FIELDS

R EMARK 4.6.11. Note that an A-valued (punctured) distribution ψ on Q/Z gives rise to an A-
valued map ψ̃ on (nontrivial) Dirichlet characters χ in that Dirichlet characters are step functions on Ẑ
(that are zero at zero). In particular, if χ has modulus dividing m, then
m−1 a
ψ̃(χ) = ∑ χ(a)ψ .
a=0 m

E XAMPLE 4.6.12. By Lemma 4.2.10, we have

ψ (n) (χ) = Bn,χ

for a primitive Dirichlet character χ. In particular, ψ (n) (χ) = 0 if n 6≡ χ(−1) mod 2, unless n = 1 and
χ = 1. Similarly, ψcyc (χ) = 0 unless χ is even.

4.7. Sinnott’s theorem

In this section, we fix m > 1 with m 6≡ 2 mod 4. We set F = Q(µm ) and G = Gal(F/Q). The goal
of this section is to prove the following generalization of the results of Hasse and Iwasawa for F, which
is due to Sinnott.1

T HEOREM 4.7.1 (Sinnott). Let F = Q(µm ) for m > 1 with m 6≡ 2 mod 4. Then we have
− − b −
F and [Z[G] : IF ] = 2 hF ,
[EF+ : CF+ ] = 2a h+

where
 
0 if g = 1 0 if g = 1
a= and b =
2g−2 + 1 − g if g ≥ 2 2g−2 − 1 if g ≥ 2,
for g the number of primes dividing m.

N OTATION 4.7.2. For χ ∈ Ĝ, we have the idempotent

m
1
eχ = ∑ χ(a)σa−1 ∈ C[G].
ϕ(m) a=1
(a,m)=1

We also have idempotents

1 ± σ−1
e± = ∈ Q[G].
2
The following is essentially immediate from the definitions.

L EMMA 4.7.3. We have e± A = 12 Z[G]± inside Q[G]. In particular, we see that

[e± Z[G] : Z[G]± ] = 2ϕ(m)/2 .

1This section is roughly written at present and might be safely skipped for now, aside from the statement of the theorem.
 4.7. SINNOTT’S THEOREM 115

N OTATION 4.7.4. For any Z[G]-module A, set A0 = ker(NG : A → A).

R EMARK 4.7.5. For a Z[G]-module A, we note that e− (1 − e1 )A = e− A.

N OTATION 4.7.6. For each prime p dividing m, set

λp = ∑ (1 − χ̄(p))eχ ∈ Q[G].
χ∈Ĝ

For each positive integer f dividing m, set G f = Gal(Q(µm )/Q(µ f )). Let U denote the Z[G]-module
generated by the elements
u f = NG f ∏ λ p ∈ Q[G]
p| f
for positive integers f dividing m, where the product is taken over primes dividing f .

We briefly sketch a proof of the following proposition.

P ROPOSITION 4.7.7. Let g be the number of primes dividing m. Then we have the following
equalities: 
± ±
1 if g = 1
(e Z[G] : e U) =
22g−2 if g ≥ 2.

P ROOF. If g = 1, then U is generated by NG and λ p for the unique prime p dividing m. We have
u1 = NG = |G|e1 and u p = λ p = 1 − e1 . Then

[e1 Z[G] + Z[G] : U] = |G| = [e1 Z[G] + Z[G] : Z[G]],

so [Z[G] : U] = 1. Moreover, note that e− e1 = 0, and from this it is easily seen that e− Z[G] = e−U,
and as a result, [e+ Z[G] : e+U] = 1 as well.
For g ≥ 2, we indicate only a few details of the proof. One uses the fact that U is the product over
primes p dividing m of the modules U p generated by NIp and λ p , where I p < Gal(F/Q) is the inertia
group at p, to see that (Z[G] : U) = 1. On the other hand,

(Z[G] : U) = (e+ Z[G] : e+U)(Z[G]− : U − ).

One checks that the order of

Ĥ −1 (Gal(F/F + ),U) = U − /(σ−1 − 1)U ∼

= e− Z[G]/e−U
g−1
is 22 . We then have
g−1
(e+ Z[G] : e+U)(e− Z[G] : e−U) = 22 ,
and the proof is finished upon showing that (e− Z[G] : e−U) = 2 2g−2 , which we omit. 

Recall that IG denotes the augmentation ideal in Z[G].

116 4. CYCLOTOMIC FIELDS

C OROLLARY 4.7.8. Let g be the number of primes dividing m. Then


ϕ(m)−1 if g = 1
+ +
(e IG : e U0 ) =
22g−2 ϕ(m)−1 if g ≥ 2,

P ROOF. The quotient e+ Z[G]/e+ IG is isomorphic to Z via the augmentation map, while e+U/e+U0
is generated by the class of u1 = NG , and the image of NG ∈ e+ Z[G] under the augmentation map is
|G| = ϕ(M). It follows that

(e± Z[G] : e±U) = ϕ(M)(e+ IG : e+U0 ),

and we apply Proposition 4.7.7. 

N OTATION 4.7.9. For a punctured C-valued distribution ψ on Q/Z, let Tψ be the subgroup of C[G]
generated by the elements
m  
bc
ηψ (c) = ∑ ψ σb−1
b=1 m
(b,m)=1
for positive integers c with c 6≡ 0 mod m.

R EMARK 4.7.10. The group Tψ is a Z[G]-module, as σa ηψ (c) = ηψ (ac) for a prime to m. As a

Z[G]-module, it is then generated by the elements ηψ (d) for d positive dividing m.

At times, we will view the elements of Ĝ also as primitive Dirichlet characters.

P ROPOSITION 4.7.11. Let ψ be a punctured C-valued distribution on Q/Z. Then

(1 − e1 )Tψ = ωψ U,

where
ωψ = ∑ ψ(χ̄)eχ ∈ C[G].
χ∈Ĝ−{1}

m
P ROOF. For d ≥ 1 dividing m, set f = d. Let χ be a nontrivial character of Ĝ. Then eχ ηψ (d)
vanishes if f does not divide the conductor f χ of χ, and if f | f χ , then
!
m  
b ϕ(m)
eχ ηψ (d) = eχ ∑ ψ χ̄(b) = eχ (1 − χ̄(p)) ψ(χ̄).
b=1 f ϕ( f ) ∏ p| f
(b,m)=1

Noting that eχ ωψ = eχ ψ(χ̄), that eχ λ p = eχ (1 − χ̄(p)), and that


eχ ϕ(m) if f χ | f
ϕ( f )
eχ NG f =
0 otherwise,
 4.7. SINNOTT’S THEOREM 117

we conclude that
eχ ηψ (d) = eχ ωψ u f .
This holds for all χ 6= 1, and we also ahve that e1 ωψ = 0, so we obtain (1 − e1 )ηψ (d) = ωψ u f . In that
this holds for all d, the result follows. 

L EMMA 4.7.12. Let ψ be a punctured C-valued distribution on Q/Z. In the notation of Proposi-
tion 4.7.11, if ψ(χ) = 0 for all nontrivial χ ∈ Ĝ with χ(−1) = ∓1, then

(e±U0 : (1 − e1 )Tψ ) = ∏ ψ(χ) .

χ∈Ĝ−{1}
χ(−1)=±1

P ROOF. By our condition on χ, the element ωψ of Proposition 4.7.11 is

ωψ = ∑ ψ(χ)eχ .
χ∈Ĝ−{1}
χ(−1)=±1

Then ωψ ∈ (1 − e1 )e± C[G] by assumption on ψ, and Proposition 4.7.11 implies that

(1 − e1 )Tψ = ωψ U = (1 − e1 )e± ωψ U0 .

Note that e1 λ p = 0 for any prime p dividing m, so e1 u f = 0 if f is a positive divisor of m other than
1. Since the u f generate U as a Z[G]-module and u1 = NG , we therefore have U = U0 + NG Z. It follows
that (1 − e1 )U = U0 . Multiplication by ωψ determines an C-linear endomorphism of (1 − e1 )e± C[G]
that takes e±U0 onto (1 − e1 )Tψ . The idempotent eχ for nontrivial χ ∈ Ĝ with χ(−1) = ±1 is an
eigenvector of this endomorphism with eigenvalue ψ(χ). The determinant is of course the product of
these eigenvalues. The result then follows by Lemma 1.2.9. 

R EMARK 4.7.13. For any Z[G]-module A that is free over Z, we have A0 = A ∩ (1 − e1 )A, since
e1 A0 = 0 and the kernel of NG is the image of e1 on A ⊗Z Q.

E XAMPLE 4.7.14. The R-vector space V spanned by the elements of G has V0 equal to the elements
with coefficients summing to 0. For S the set of primes above m in F, we have T = Tψcyc = ρ(CF,S ) is
contained in V , and note that T0 = ρ(CF ) by the product formula.

L EMMA 4.7.15. For ψ = ψcyc and T = Tψcyc , we have

[(1 − e1 )T : T0 ] = 2−g ϕ(m).

P ROOF. Note that

(1 − e1 )T /T0 ∼ = (e1 T + T )/T ∼

= ((1 − e1 )T + T )/T ∼ = e1 T /T G .
118 4. CYCLOTOMIC FIELDS

We have
1 1 NG
e1 T = NG ρ(CF,S ) = ρ(CF,S ).
ϕ(m) ϕ(m)
k
Note that |(1 − ζ f )NG | = 1 if f is not a prime power, and (1 − ζ pk )NG = pϕ(m)/ϕ(p ) . It follows that

1 1
e1 T = ∑ log p · NG Z,
2 p|m ϕ(pk p )

where k p ≥ 1 is the additive p-adic valuation of m.

Next, note that α ∈ CF,S satisfies j(α) ∈ T G if and only if j(α σ −1 ) = 0 for all σ ∈ G, which is
equivalent to α σ −1 ∈ µ(F), which is in turn equivalent to α 1+τ ∈ Q× , with τ complex conjugation.
Let

P = {α ∈ CF,S | α 1+ j ∈ Q× },

and note that T G = ρ(P) = 21 ρ(P1+τ ). For an odd prime p dividing m, set

(p−1)/2
αp = ∏ (1 − ζ pa ),
a=1

and set α2 = 1 − ζ4 if m is even. Then each α p for p dividing m lies in P, so P1+τ contains the group
H generated by all primes dividing m. Since P1+τ is a subgroup of the positive rationals, the quotient
P1+τ /H is torsion-free, and on the other hand (P1+ j )ϕ(m) = (P1+ j )NG ⊆ H, which forces P1+τ = H.
Thus
1
TG = ∑ log p · NGZ.
4 p|m

It follows that
G ϕ(pk ) ϕ(m)
[(1 − e1 )T : T0 ] = [e1 T : T ] = ∏ = g .
p|m
2 2

L EMMA 4.7.16. Let ρ : EF,S → V + denote the Z[G]-module homomorphism

1
ρ(α) = − log |σ (α)|σ −1 .
2 σ∑
∈G

Then
+ R+
(e IG : ρ(EF )) = F .
QF
 4.7. SINNOTT’S THEOREM 119

P ROOF. Let X = (1 − e1 )e+V , in which ρ(EF ) forms a lattice of full rank r = ϕ(m)
2 − 1. The lattice
e+ IG has a basis e+ (1 − σa−1 ) for 1 < a < m
2 with (a, m) = 1. Fix a complex embedding of F, hence an
absolute value. For an independent system of units α1 , . . . , αr ∈ EF+ generating EF /µF , we have
d m2 e−1 d m2 e−1
ρ(αi ) = − ∑ log |σa (αi )|σa−1 = ∑ log |σa (αi )|e+ (1 − σa−1 ).
a=1 a=2
(a,m)=1 (a,m)=1

2r +
Since the matrix with entries log |ηiσa | has determinant 2−r RF by definition and RF = QF RF , we are
done. 

L EMMA 4.7.17. We have

[e− Z[G]θF : IF− ] = wF .
.

P ROOF. Let ΘF = Z[G]θF for brevity. Since (σa − a)θF ∈ IF for all a ∈ Z, we have that

ΘF = IF + θF Z,

and therefore ΘF /IF ∼

= Z/mZ as m is minimal with mθF integral. From the fact that hαi+i1 − αi = 1
/ Z, one see that e+ θF = 12 NG Z. Since (σ2 − 2)θF ∈ IF and e+ (σ2 − 2) = − 12 NG , we then have
for α ∈
F = IF , which in turn implies that
that e+ ΘF = e+ IF and therefore (Z[G]θ )+ +

−∼ ∼
Θ−
F /IF = ΘF /IF = Z/mZ.

If m is even, then σm/2 θF = 21 NG = e+ θF . Therefore, we have e+ ΘF ⊆ ΘF , and in turn this implies

that e− ΘF ⊆ ΘF . In other words, we have [e− ΘF : Θ−
F ] = 1.
If m is odd, then e− σa θF = σa θF − 12 NG , so

e− ΘF + ΘF = 21 NG Z + θF ,

and therefore
e− ΘF /Θ− ∼1 1
F = 2 NG Z/(ΘF ∩ 2 NG Z).

Note that NG = (1 + j)θF ∈ ΘF but 12 NG ∈

/ ΘF since mΘF ⊂ Z[G] and m is odd. Therefore, we have
[e− ΘF : Θ−
F ] = 2.
For arbitrary m, we conclude that
wF
[e− Z[G]θF : IF− ] = [e− ΘF : Θ− − −
F ][ΘF : IF ] = · m = wF .
m


We are now ready to prove Sinnott’s theorem.

120 4. CYCLOTOMIC FIELDS

P ROOF OF T HEOREM 4.7.1. First consider ψ = ψcyc , and set T = Tψ . Since T0 = ρ(CF ), we may
write our index as a product

[EF+ : CF+ ] = [ρ(EF ) : ρ(CF )] = (ρ(EF ) : e+ IG )(e+ IG : e+U0 )(e+U0 : (1 − e1 )T )((1 − e1 )T : T0 ).

The latter four relative indices are computed by Lemma 4.7.16, Lemma 4.7.12, Corollary 4.7.8, and
Lemma 4.7.15, respectively. Plugging in, we obtain

g−1
QF (22 )1/2 ϕ(m) a 1
[EF+ : CF+ ] = · · ∏ ψcyc (χ) · = 2 ∏ ψcyc (χ) = 2a h+
F,
R+
F ϕ(m)
χ∈Ĝ−{1}
2g R+
F χ∈Ĝ−{1}
χ even χ even

where the last equality follows from Theorem 4.2.20.

Next, consider ψ = ψ (1) , the first Bernoulli distribution, which by definition has Tψ (1) = e− Z[G]θF .
We write the index in question as a product as follows:

[Z[G]− : IF− ] = (Z[G]− : e− Z[G])(e− Z[G] : e−U)(e−U : e− Z[G]θF )(e− Z[G]θF : IF− ).

The latter four relative indices are computed by Lemmas 4.7.3, Proposition 4.7.7, 4.7.12, and 4.7.17,
g−2
respectively. Noting also that 2b QF = 22 if g ≥ 2 and 2b QF = 1 if g = 1, we obtain

−B1,χ
[Z[G]− : IF− ] = 2−ϕ(m)/2 · 2b QF · ∏ ψ (1) (χ) · wF = 2b · 2[EF : EF+ ] ∏ = 2b h−
F
2
χ∈Ĝ χ∈Ĝ
χ odd χ odd

where the second equality uses that QF wF = 2[EF : EF+ ] by Lemma 4.2.16, and the final equality
follows from Theorem 4.2.20. 
 CHAPTER 5

Kubota-Leopoldt p-adic L-functions

5.1. p-adic measures

In this section, we study C p -valued distributions.

D EFINITION 5.1.1. We say that a C p -valued distribution {ψi }i∈I on an inverse system of finite sets
Xi is bounded if there exists a constant B ∈ R≥0 such that |ψi (x)| ≤ B for all x ∈ Xi for all i ∈ I, where
| · | is the unique extension of the p-adic valuation on Q p to C p .

R EMARK 5.1.2. To say that {ψi } is bounded is the same as saying the corresponding functional ψ
on step functions on the profinite space X = limi Xi satisfies
←−
|ψ(χ)| ≤ Bkχk,

where kχk = supx∈X |χ(x)| (which is actually a maximum, as X is compact).

N OTATION 5.1.3. For a topological subring O of C p , let C(X, O) denote the space of continuous
functions from X to O, endowed with the compact-open topology.

R EMARK 5.1.4. The set Step(X, C p ) is dense in C(X, C p ).

D EFINITION 5.1.5. For a topological subring O of C p and a profinite space X, an O-valued mea-
sure on X is a bounded linear functional

µ : C(X, O) → O.

We write Z
gdµ
X
for the value µ(g).

R EMARK 5.1.6. Measures on X are in one-to-one correspondence with bounded distributions, since
Step(X, C p ) is dense in C(X, C p ).

E XAMPLE 5.1.7. The δ -distribution at x ∈ X gives rise to the Dirac measure

Z
gdδx = g(x).
X
We briefly discuss measures on Z p .
121
122 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

R EMARK 5.1.8. Let g : Z p → C p be a continuous function, and let µ be a C p -valued measure on

Z p with corresponding distribution {µn }. Then
Z pn −1
gdµ = lim ∑ gn (a)µn (a).
Zp n→∞
a=0

Let O denote the valuation ring of a finite extension of Q p .

P ROPOSITION 5.1.9. There is a canonical bijection between O-valued measures µ on Z p and

elements λ of OJZ p K, seen explicitly as follows. Write λ ∈ OJZ p K as λ = (λn )n with λn ∈ O[Z/pn Z].
Then µ is the measure associated to the distribution {µn }n≥1 with µn : Z/pn Z → O corresponds to λ
if and only if
pn −1
λn = ∑ µn (a)[a]n
a=0
where [a]n ∈ O[Z/pn Z] is the group element attached to a.

P ROOF. Clearly, the data of the λn determine the µn and conversely. One need only see that f
is well-defined if and only if the µn satisfy the distribution relations. But, from the definitions, the
element λn+1 maps to λn if and only if
pn −1
µn (a) = ∑ µn+1 (a + pn b),
b=0
as required. 

R EMARK 5.1.10. We have µ(χa+pn Z p ) = µn (a), so knowing each µn determines µ on step func-
tions explicitly.

R EMARK 5.1.11. Since OJT K ∼

= OJZ p K via the continuous O-linear isomorphism taking T + 1 to
the group element of 1, we have a canonical bijection between O-valued measures on Z p and power
series in OJT K.

C OROLLARY 5.1.12. The power series f attached to an O-valued measure µ on Z p is given by

∞ Z   
x
f (T ) = ∑ dµ(x) T i ∈ OJT K.
i=0 Z p i

P ROOF. Let fn ∈ OJT K/(ωn ) be the image of f . By Proposition 5.1.9, the measure µ attached to
f given by the distribution {µn }n≥1 is related to fn through the formula
pn −1 ∞ pn −1  
a a
fn (T ) = ∑ µn (a)(T + 1) = ∑ ∑ µn (a)T i ,
a=0 i=0 a=0 i
so the inverse limit f of the fn satisfies the desired equation. 
 5.1. p-ADIC MEASURES 123

C OROLLARY 5.1.13. Let f be the power series attached to an O-valued measure µ on Z p . If t ∈ m,

where m is the maximal ideal of O, the value f (t) may be calculated by
Z
f (t) = (1 + t)x dµ(x),
Zp

where µ is the measure corresponding to f .

T HEOREM 5.1.14 (Mahler). We have

(   )
∞
x
C(Z p , O) = ∑ ci i | ci ∈ O, ci → 0 ,
i=0

and the representation of g ∈ C(Z p , O) as a sum as in the latter set is unique.

P ROOF. Suppose that there is a sequence (ci )i≥1 of elements of O that converges to 0. Since each
| xi | is bounded by 1 on Z p , any g = ∑∞ x



i=0 ci i with ci → 0 is the uniform limit of its continuous partial
sums, hence continuous.
Consider the difference operator ∇ on g ∈ C(Z p , O) defined by ∇(g)(x) = g(x + 1) − g(x). Then
       
x x+1 x x
∇ = − = ,
i i i i−1
so if g has the form in the theorem, then ∇i (g)(0) = ci . In other words, the representation of g as a sum
is unique if it exists.
We now show existence. For this, it suffices to consider Z p -valued functions by choice of a ba-
sis and projection. We have a Z p -linear map from the set of sequences in Z p that converge to 0 to
x

C(Z p , Z p ) given by (ci )i≥0 7→ ∑∞
i=0 ci i . It suffices to show that this map is surjective. This can
be derived via recursion from the claim that the set of eventually zero sequences in F p surjects onto
C(Z p , F p ) via the reduction modulo p of this map.
Note that
C(Z p , F p ) = lim Maps(Z/pn Z, F p ),
−→
n
x
For 0 ≤ i ≤ p − 1, the map x 7→ i mod p lies in Maps(Z/pn Z, F p ), since
n

n n n
(1 + T )x+p ≡ (1 + T )x (1 + T p ) ≡ (1 + T )x mod (p, T p )Z p JT K.

Thus, our map restricts to a map

{(ci )0≤i<pn | ci ∈ F p } → Maps(Z/pn Z, F p )

that is injective by our earlier uniqueness argument using ∇ and surjective by equality of F p -dimensions.
This proves the desired surjectivity. 

The following is a matter of switching the order of a sum and an integral.

124 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

i ∈ C(Z p , O) and µ
x

C OROLLARY 5.1.15. For g = ∑∞
i=0 ci the measure attached to
∞
f = ai T i ∈ OJT K,
∑
i=0
we have Z ∞
gdµ = ∑ ai ci .
Zp i=0

Typically, we are more interested in measures on Z×

p , or the units in a slightly larger ring. Let us
recall that 1 + qZ p , where q = p if p is odd and q = 4 for p = 2, is isomorphic to Z p via the map that
takes ua to a for any a ∈ Z p , where u is a fixed topological generator of 1 + qZ p , such as 1 + q. In this
way, measures on 1 + qZ p are made to correspond to measures on Z p .

D EFINITION 5.1.16. For an O-valued measure ν on 1 + qZ p , let µ be the O-valued measure on Z p

defined by Z Z
x
g(u )dµ(x) = gdν.
Zp 1+qZ p
The power series in OJT K attached to ν is the power series corresponding to µ by Proposition 5.1.9.

L EMMA 5.1.17. The power series f attached to an O-valued measure ν on 1 + qZ p satisfies

Z
f (us − 1) = xs dν(x)
1+qZ p

for s ∈ Z p , and f is uniquely determined by this formula.

P ROOF. Let µ be the measure on Z p corresponding to ν and f . Set t = us − 1 for some s ∈ Z p .

Then Corollary 5.1.13 tells us that
Z Z
s sx
f (u − 1) = u dµ(x) = xs dν(x).
Zp 1+qZ p

We leave the last simple statement to the reader. 

R EMARK 5.1.18. We can also attach a measure on Z p to a measure on Z×

p , by extension by zero.
Similarly, we can restrict measures on Z p to the latter multiplicative subgroups.

5.2. p-adic L-functions

D EFINITION 5.2.1. Let p be a prime number, and let m ≥ 1 be prime to p. Set

Z p,m = lim (Z/mpn Z).

←−
n
Then
∼
Z p,m −
→ Z p × Z/mZ,
and we let c p denote the first coordinate of the image of c ∈ Z p,m .
 5.2. p-ADIC L-FUNCTIONS 125

Note that
Z×
p,m = lim (Z/mpn Z)× ∼
= Z× ×
p × (Z/mZ) ,
←−
n
and, setting q = p for p odd and q = 4 for p = 2, we also have
∼
Z× → (1 + qZ p ) × (Z/qmZ)× .
p,m −

D EFINITION 5.2.2. For c ∈ Z× ×

p,m , we let c p denote its image in Z p and hci p denote its image in
1 + qZ p .

Note also that Z×

p,m is canonically isomorphic to the Galois group of Q(µmp∞ )/Q. We will typically
be interested in measures on Z×
p,m .
Let us set ∆ = (Z/qmZ)× . We have

Z p JZ× ∼ ∼
p,m K = Z p [∆]J1 + qZ p K = Z p [∆]JT K,

the latter isomorphism taking the group element u to T + 1.

Let O be the valuation ring of a finite extension of Q p . By the same discussion as before, replacing
O by the group ring O[∆], we have the following.

L EMMA 5.2.3. There is a canonical bijection between O-valued measures on Z×

p,m and elements
of OJT K[∆].

Explicitly, the power series f ∈ OJT K[∆] attached to an O-valued measure ν on Z×

p,m satisfies
Z
s
f (u − 1) = ∑ xs dν(σ x) · σ
σ ∈∆ 1+qZ p

If ν arises from a distribution ψ = (ψn ) on the groups (Z/mpn Z)× , then f is then given by the com-
patible system of elements
pn m
∑ ψn (a)[a]n ∈ O[(Z/pn mZ)× ],
a=1
(a,mp)=1
where [a]n denotes the group element of a.
The Bernoulli distribution will be the key to our definition of p-adic L-functions, but it is not
necessarily integral. Therefore, we introduce the following modification.

D EFINITION 5.2.4. Set N = pn m, let c ∈ Z×

p,m , and take k ≥ 1. For x ∈ Z/NZ, we view
x
N as an
1
element of N Z/Z and define
(k) 1 D x E
En (x) = N k−1 Bk
k N
and
(k) (k) (k)
En,c (x) = En (x) − ckp En (c−1 x).
126 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

(k) (N) (k) (k)

Note that En = 1k ψk , so the En and En,c form distributions on Q/Z.
(k)
P ROPOSITION 5.2.5. For N = pn m with n ≥ 1 and k ≥ 1, we have En,c (x) ∈ Z p and
(k) (1)
En,c (x) ≡ xk−1 En,c (x) mod pn Z p

for all x ∈ Z/NZ.

P ROOF. We have
teXt
  ∞ i
1 1 2 X i
= 1 − t + t + · · · ∑ t,
et − 1 2 6 i=0 i!
from which we see that the kth Bernoulli polynomial has the form
k
Bk (X) = X k − X k−1 + k f (X)
2
k−1 k−2
with f ∈ Q[X] of degree k − 2 (or 0 if k = 1), the leading term of f (X) being 6 X . Let ek ≥ 0 be
minimal such that pek f (X) ∈ Z p JXK
Let a ∈ Z with 0 ≤ a < N lift x ∈ Z/NZ, and let b ∈ Z with 0 ≤ b < N and y ∈ Z p be such that
c−1
p a b
N = N + y. We then have

b j ≡ c− j j − j+1 j−1
p a − jNc p a y mod N 2

for j ≥ 1, which yields

    
(k) 1 1 k k k−1 k 1 k k k−1 k−1 1
En,c (x) ≡ a − a − cp b − b ≡a c p y + (c p − 1) mod pn−ek Z p .
k N 2 N 2 2
(1)
Since this holds for all k, we have in particular that En,c (x) is in Z p for all n, noting that c p ≡ 1 mod
2Z p , and that
(k) (1)
En,c (x) ≡ xk−1 En,c (x) mod pn−ek Z p
(k)
is integral for sufficiently large n as well. By the distribution relation for the En,c , this integrality then
0
holds for all n. If we choose n ≥ e0k , where e0k is minimal such that pek ( f (X) − k−1
6 X
k−2 ) ∈ Z [X], then
p
we can refine the above to to
(k) (1) k−1
En,c (x) ≡ xk−1 En,c (x) + Nxk−2 (1 − c2p ) mod pn Z p
6
Since c2p ≡ 1 mod 6Z p , this reduces to
(k) (1)
En,c (x) ≡ xk−1 En,c (x) mod pn Z p ,

and the congruence then follows for arbitrary n by the distribution relation. 
(k) (k)
R EMARK 5.2.6. Together the En,c form a Z p -valued measure Ec on Z p,m , hence on Z×
p,m as well
by restriction. We can integrate the resulting measure against functions on Z×
p,m that arise as limits of
Dirichlet characters of conductor dividing pn m for some n.
 5.2. p-ADIC L-FUNCTIONS 127

(k) (k)
D EFINITION 5.2.7. We let Ec denote the measure defined by the En,c .

p,m , O), we have

R EMARK 5.2.8. Given g ∈ C(Z×
Z Z
(k) (1)
g(x)dEc (x) = g(x)xk−1
p dEc (x)
Z×
p,m Z×
p,m

for every k ≥ 1.

R EMARK 5.2.9. When χ : Z p,m → O × is a continuous multiplicative function, we have

Bk,χ
Z
(k)
χ(x)dEc (x) = (1 − χ(c)ckp ) .
Z p,m k
Note that if χ has finite order, so is the map attached to a primitive Dirichlet character of conductor
(k)
dividing mpn for some n, then χEc defines an O-valued measure on Z p,m , with volume given by the
(k)
above formula. Here, the χEn,c are really only defined for pn m a multiple of the conductor of χ, but
the ith terms of the distribution for i less than the minimal such n can be defined by the distribution
relations.

D EFINITION 5.2.10. Let ν be an O-valued measure on Z×

p,m . We define its p-adic Mellin transform
to be the O-valued function M p (ν) on Z p given by
Z
M p (ν)(s) = hxisp x−1
p dν(x).
Z×
p,m

R EMARK 5.2.11. When p is odd, x p = hxi p ω(x) for any x ∈ Z×

p,m , where ω is the Teichmüller
character, which factors through (Z/pZ)× . For p = 2, we simply define ω : Z×
2,m → µ2 (Z2 ) by the
above formula.

R EMARK 5.2.12. If ν is an O-valued measure on Z×

p,m , then so is ψν for any Dirichlet character
ψ of conductor dividing pn m for some n. In particular, we have
Z
M p (ν)(s) = hxis−1 d(ω −1 ν).
Z×
p,m

D EFINITION 5.2.13. Let χ : Z× ×

p,m → C p be a finite-order character. We define the Kubota-Leopoldt
p-adic L-function of χ to be the C p -valued function on Z p given by
−1 (1)
L p (χ, s) = −(1 − χ(c)hci1−s
p ) M p (χEc )(1 − s)

for s ∈ Z p and c ∈ Z×
p,m such that χ(c) 6= 1 if χ 6= 1.

Rewriting this, we have

Z
−1 (1)
−(1 − χ(c)hci1−s
p )L p (χ, s) = χ(x)hxi1−s
p x p dEc (x)
Z×
p,m
(5.2.1) Z
(1)
= hxi−s −1
p d(χω Ec )(x).
Z×
p,m
128 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

R EMARK 5.2.14. The factor (1 − χ(c)hci1−s −1 in the definition of L (χ, s) removes the depen-
p ) p
dence of the definition of the p-adic L-function on the value c. Note that such a factor (without the
(1)
inverse) was used in defining Ec in the first place.

A finite order character χ : Z× ×

p,m → C takes values in Q and may be viewed as a p-adic character
through a choice of embedding of Q in Q p , we have the following.

P ROPOSITION 5.2.15. Let χ be a primitive Dirichlet character of conductor pn m for some n ≥ 0,

and let χ also denote the resulting character χ : Z× ×
p,m → C p , fixing a place over p in Q. For k ≥ 1, we
have
Bk,χω −k
L p (χ, 1 − k) = −(1 − χω −k (p)pk−1 ) = (1 − χω −k (p)pk−1 )L(χω −k , 1 − k).
k

P ROOF. Set χk = χω −k . We note that

Z Z
(1) (k)
(1 − χk (c)ckp )L p (χ, 1 − k) = χk (x)xk−1
p dEc (x) = χk (x)dEc (x),
Z×
p,m Z×
p,m

and we split the latter integral into a difference of an integral over Z p,m by an integral over pZ p,m ,
given that χk is trivial on elements of Z p,m not prime to m. By Remark 5.2.9, the former is

Bk,χk
Z
(k)
χk (x)dEc (x) = (1 − χk (c)ckp ) .
Z p,m k

(k) (k)
Since En,c (pb) = En−1,c (b) for b ∈ Z/mpn Z, the latter is

Z mpn−1
(k) (k)
χk (x)dEc (x) = ∑ χk (pa)En,c (pa)
pZ p,m a=1
mpn−1
k−1 (k)
= χk (p)p ∑ χk (a)En−1,c (a)
a=1
Z
(k)
= χk (p)pk−1 χk (x)dEc (x).
Z p,m

Taking the difference of the two terms, we have the result. 

C OROLLARY 5.2.16. The p-adic L-function of χ is independent of the choice of c in its definition.

P ROOF. The function L p (χ, s) is continuous, and its values at the dense subset of Z p consisting of
the nonnegative integers are independent of c by Proposition 5.2.15. 
 5.3. IWASAWA POWER SERIES 129

5.3. Iwasawa power series

D EFINITION 5.3.1. A finite order p-adic character χ on Z×

p,m is a of the first kind if χ is trivial on
1 + qZ p and of the second kind if it is trivial on ∆.

In general, a finite order p-adic character χ on Z×

p,m is a unique product χ = χt χw of a p-adic
character χt of the first kind and a p-adic character χw of the second kind. We use the subscripts “t”
and “w” to indicate “tame” and “wild”, respectively, though the terminology is technically incorrect if
p = 2. If we view χ as corresponding to a primitive Dirichlet character of conductor mpn for n ≥ 0,
then χ is of the first kind if and only if it has conductor dividing mq, and χ is of the second kind if and
only if it has p-power order and conductor pn for some n ≥ 1 (and then necessarily at least 2 if p = 2).
If χ is of the second kind, it is necessarily even.

N OTATION 5.3.2. For any Dirichlet character χ, let Λχ = Oχ JT K, where Oχ is the Z p -algebra
generated by the values of χ, fixing a choice of a embedding Q ,→ Q p . Let Kχ denote the quotient field
of Oχ , and let Q(Λχ ) denote the quotient field of Λχ , which contains Kχ JT K.

P ROPOSITION 5.3.3. Let χ be a primitive even Dirichlet character of conductor m or mq. There
exists a unique element Fχ ∈ Q(Λχ ) such that

Fχ (ξ us − 1) = L p (χρ, s)

for all s ∈ Z p and ξ of p-power order, where ρ is of the second kind satisfying ρ(u) = ξ −1 .

P ROOF. It follows from (5.2.1) that

Z 
(1)
−(1 − χρ(c)hci1−s
p )L p (χρ, s) = ∑ hxi−s
p ρ(x)dEc (x) χω −1 (σ ).
σ ∈∆ 1+qZ p

Let a ∈ Z p be such that hci p = ua , and set

hχ,c (T ) = χ(c)hci p (1 + T )−a − 1.

Then
hχ,c (ξ us − 1) = χ(c)hci p ξ −a u−sa − 1 = −(1 − χρ(c)hci1−s
p ).

Similarly, if we let f χ,c ∈ Oχ JT K be such that

f χ,c (us − 1) = −(1 − χ(c)hci1−s

p )L p (χ, s)

for all s ∈ Z p , then

f χ,c (ξ us − 1) = −(1 − χρ(c)hci1−s
p )L p (χρ, s).
f χ,c
Thus Fχ = hχ,c has the desired property. In that the integral power series f χ,c satisfies f χ,c ((1 + q)s −
1) = hχ,c ((1 + q)s − 1)L p (χ, s) for all s ∈ Z p , it is unique, and therefore so is Fχ . 
130 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

R EMARK 5.3.4. In the notation of Proposition 5.3.3, we have If χ 6= 1, then we may take c ∈ ∆ ⊂
Z×
p,m to be such that χ(c) 6= 1, so hχ,c = χ(c) − 1, and (χ(c) − 1)Fχ ∈ Λχ . If χ = 1, then we may take
c = u ∈ 1 + qZ p , so hχ,u = u(T + 1)−1 − 1.

D EFINITION 5.3.5. Let Fn = Q(µmpn ), and let Gn = Gal(Fn /Q). For any b ∈ Z×
p,m , set
mpn
(k) (k)
Θn (b) = ∑ En (ab)σa−1 .
a=1
(a,mp)=1
(k) (k) (k)
We set Θn = Θn (1) and refer to Θn as the kth higher Stickelberger element for Fn .
(k) (k)
Since the En form a distribution, the Θn give a compatible system in the inverse limit. Set
∼
G∞ = Gal(F∞ /Q). We have a continuous isomorphism Z×
p,m −
→ G∞ via a 7→ σa , under which 1 + qZ p
is idenitifed with Γ = Gal(F∞ /F) for F = Q(µmq ), and ∆ = (Z/mqZ)× is identified with the torsion
subgroup of G∞ , which we also denote by ∆. For locally compact Z p -algebra R, we then have an
identification
RJG∞ K = R[∆]JΓK ∼
= R[∆]JT K
of topological rings, where T = γ − 1 for γ = σu .

N OTATION 5.3.6. Let

(k)
Θ(k)
∞ (b) = (Θn (b))n ∈ Q p [∆]JT K,
(k) (k)
and set Θ∞ = Θ∞ (1).

R EMARK 5.3.7. Since

mpn mpn
(k) (k)
(1 − ckp σc−1 ) ∑ En (a)σa−1 = ∑ En,c (a)σa−1 ∈ Z p [Gn ],
a=1 a=1
(a,mp)=1 (a,mp)=1
(k)
we have (1 − ckp σc−1 )Θ∞ ∈ Z p [∆]JT K. Aside from the use of σa−1 in place of σa , the latter is the power
(k)
series corresponding to the measure given by the En,c on Z×
p,m .

N OTATION 5.3.8. For any nontrivial primitive even Dirichlet character χ of conductor m or mq, let
^
f χ = −ω χ −1 (Θ(1)
∞ ),

^
where ω χ −1 : Q p [∆]JT K → Kχ JT K is the unique continuous Q p JT K-linear map that restricts to χ1−1 =
ω χ −1 on ∆. Set
f1 = (1 − u(1 + T )−1 )ω(Θ
e (1)
∞ ).

D EFINITION 5.3.9. For any primitive even Dirichlet character of conductor m or mq, the power
series f χ is called the Iwasawa power series of χ.
 5.3. IWASAWA POWER SERIES 131

R EMARK 5.3.10. For any nontrivial χ, the image of f χ in Gal(Fn /Q) is

mpn   mpn
a 1 1
− ∑ n
− χω −1 (a)σhai
−1
=− n ∑ aχω −1 (a)σhai
−1
,
a=1 mp 2 p mp a=1
p

(a,mp)=1 (a,mp)=1

where the second equality is by the nontriviality of χ1 . Then − f χ becomes identified with
 
ω^ χ −1 (θFn ) ∈ Q p JΓK,
n

where ω χ −1 is defined in the obvious fashion. In other words, f χ is the negative of the ω χ −1 -
specialization of the inverse limit of Stickelberger elements of the fields Fn .

For χ nontrivial, the power series f χ agrees with Fχ defined above.

L EMMA 5.3.11. For any primitive even Dirichlet character χ of conductor m or mq, we have
f χ = Fχ . For χ = 1, we have f1 = h1 F1 , where h1 = 1 − u(1 + T )−1 .

P ROOF. We have
 
mp Z
(1)
−(1 − χ(c)) f χ (us − 1) = ω
^ χ −1  x−s dEc (x)σa−1 
 
∑ 1+qZ p
a=1
(a,mp)=1
Z
(1)
= χ1 (x)hxi−s
p dEc (x)
Z×
p,m
Z
−1 (1)
= χ(x)hxi1−s
p x p dEc (x)
Z×
p,m

= −(1 − χ(c))L p (χ, s).

It follows that f χ = Fχ . The case that χ = 1 is similar and left to the reader. 

We now prove the integrality of the Iwasawa power series f χ for odd p.

P ROPOSITION 5.3.12. Let χ be a primitive even Dirichlet character of conductor m or mq. Then
1
2 fχ ∈ Λχ .

P ROOF. We prove this in the case that p is odd. For χ = 1, this is immediate from Lemma 5.3.11
and Remark 5.3.4. For χ nontrivial, we are already done if χ is not of p-power order, as χ(c) − 1 can
be chosen to be a unit. In particular, we may suppose that m 6= 1, so is divisible by a prime ` 6= p. We
claim that
−1 (1)
Θ(1)
∞ − ` Θ∞ (`) ∈ Z p JG∞ K.
132 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

To see this, note that

mpn   
(1) (1) a `a
Θn − `−1 Θn (`) = ∑ − `−1 σa−1 ∈ Z p [Gn ]
a=1 mpn mpn
(a,mp)=1

for all n, since ` ∈ Z× a `a 0

p and ` mpn − h mpn i ∈ Z by definition. Setting m =
m
`, for ∆˜ ⊂ {1, . . . , mpn } a
set of representatives of ∆ viewed inside Z/mpn Z and a fixed b ∈ Z p , we have that the coefficient of
(1)
γ −b ∈ Oχ [Gn ] of ω
^ χ −1 Θn (`) is
m0 pn
ab `−1
 
`ab −1
∑ ωχ (a) = ∑ 0 pn ∑
ω χ −1 (a + im0 pn ),
a∈∆˜
mpn a=1 m i=0
a∈∆˜

and the latter sum is 0 since ` divides the conductor of ω χ −1 . (Note that one value of a + im0 pn in the
(1)
sum will not be prime to m if ` - m0 , but ω χ −1 (a+im0 pn ) = 0 for this value.) Thus, ω
^ χ −1 (Θ∞ (`)) = 0,
so
−1 (1)
^
fχ = ω χ −1 (Θ(1)
∞ − ` Θ∞ (`)) ∈ Λχ .

Putting this all together, we have the following.

T HEOREM 5.3.13. Let χ be a primitive even p-adic Dirichlet character of the first kind. There
exists a unique element f χ ∈ Λχ such that if χ is nontrivial, we have

f χ (ξ us − 1) = L p (χρ, s),

and if χ = 1, then for h1 = u(1 + T )−1 − 1, we have

f1 (ξ us − 1) = h1 (ξ us − 1)L p (ρ, s)

for all s ∈ Z p and ξ of p-power order, where ρ is of the second kind satisfying ρ(u) = ξ −1 .

Recall that X∞ denotes the unramified Iwasawa module over F∞ . The interpretation of f χ in terms
of Stickelberger elements also gives the following.

P ROPOSITION 5.3.14. For any primitive Dirichlet character χ of conductor m or mq, the Iwasawa
(ω χ −1 )
power series 12 f χ ∈ Λχ annihilates X∞ .
(1) (1)
P ROOF. We again suppose that p is odd. Recall that Θ∞ − `−1 Θ∞ (`) is integral, and
^
ωψ −1 (Θ(1) (`)) = 0
∞

for every nontrivial even character ψ of conductor m or mp. Write χ = νρ where ν has order prime to
(1)
p and ρ has p-power order. By varying ρ over its GQ p -conjugates, this implies that eων −1 Θ∞ (`) = 0,
 5.4. COLEMAN THEORY 133

(1)
where eων −1 is the idempotent for ων −1 on the prime-to-p part of ∆. Then eων −1 Θ∞ ∈ Oν JG∞ K. By
Remark 5.3.10, it annihilates eων −1 X∞ . By projection, we then have that f χ annihilates the quotient
(ω χ −1 )
X∞ . 

C OROLLARY 5.3.15. Suppose that p is odd. For any even k ≥ 2 not divisible by p − 1 and every
j ≥ 1, we have
B j,ω k− j
B1,ω k−1 ≡ mod p.
j
In particular, we have
Bk
B1,ω k−1 ≡ mod p.
k
B j ,ω k− j
P ROOF. We have L p (0, ω k ) = −B1,ω k−1 and L p (1 − j, ω k ) = − j , so this follows from the fact
that ω k 6= 1, then L p (χ, s) = f χ (us − 1), and u1− j − 1 ≡ 0 mod p. 

C OROLLARY 5.3.16. Suppose that p is odd and j ≡ k mod pn−1 (p − 1) are even positive integers
not divisible by p − 1. Then
Bj Bk
(1 − p j−1 ) ≡ (1 − pk−1 ) mod pn .
j k
B
P ROOF. We have L p (1 − j, ω j ) = −(1 − p j−1 ) jj . As

ω j (x)hxi pj−1 = x pj hxi−1

p
j
and x p ≡ xkp mod pn , we have the result so long as

1 − ω j (c)hci pj = 1 − c pj

can be taken to be a unit, which occurs if j 6≡ 0 mod p − 1. 

5.4. Coleman theory

Let E be an unramified extension of Q p with valuation ring O. Let q denote the order of the residue
field of O. Let En = E(µ pn+1 ), and let On denote its valuation ring, for n ≥ 0. Fix a sequence (ζ pn )n of
primitive pn th roots of unity in En such that ζ ppn+1 = ζ pn for each n ≥ 1. Let Λ = OJT K.

N OTATION 5.4.1. Let [p] denote the continuous Z p -linear endomorphism of Λ given on f ∈ Λ by

[p]( f )(T ) = f ((1 + T ) p − 1).

L EMMA 5.4.2. The image of [p] is equal to the set of all f ∈ Λ such that

f (ζ pi (1 + T ) − 1) = f (T )

for all i ∈ Z.
134 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

P ROOF. We need only show that every f with the above property is in the image of [p], which
is to say that it can be expanded in a power series in P = [p](T ) = (1 + T ) p − 1. For this, suppose
inductively that we have written f as
n−1
f= ∑ aiPi + Pn fn
i=0

with ai ∈ O for some n ≥ 0. Then fn also has the property that fn (ζ pi (1 + T ) − 1) = fn (T ) for all i.
Taking T = 0, we see that fn (ζ pi − 1) = fn (0) for all i, and therefore

fn − fn (0) = P fn+1

for some fn+1 ∈ Λ having the desired property, and we set an = fn (0). We then have f = ∑∞ i
i=0 ai P in
the limit. 

P ROPOSITION 5.4.3. There exist unique maps N : Λ → Λ and S : Λ → pΛ satisfying

p−1 p−1
([p] ◦ N )( f )(T ) = ∏ f (ζ pi (1 + T ) − 1) and ([p] ◦ S )( f )(T ) = ∑ f (ζ pi (1 + T ) − 1)
i=0 i=0

for all f ∈ Λ.

P ROOF. For f ∈ Λ, consider

p−1
g(T ) = ∏ f (ζ pi (1 + T ) − 1),
i=0

which is clearly in Λ as its coefficients are fixed by Gal(Q(µ p )/Q). We have g(T ) = g(ζ pi (1 + T ) − 1)
for all i ∈ Z, so by Lemma 5.4.2, we have g = [p](N ( f )) for some N ( f ) ∈ Λ, which is unique by the
injectivity of [p].
If we take
p−1
h= ∑ f (ζ pi (1 + T ) − 1) ∈ Λ
i=0
then as
f (ζ pi (1 + T ) − 1) ≡ f (T ) mod (1 − ζ p ),
for each i, we have h(T ) ∈ pΛ. As in the case of N , we have h(T ) = h(ζ pi (1 + T ) − 1) for all i, so
h = [p](S ( f )) for a unique S ( f ) ∈ pΛ. 

D EFINITION 5.4.4. Coleman’s norm operator N : Λ → Λ and Coleman’s trace operator S : Λ →

Λ are the maps characterized by Proposition 5.4.3.

L EMMA 5.4.5. If f ∈ Λ and n ≥ 1, then f ≡ 1 mod pn if and only if [p]( f ) ≡ 1 mod pn .

 5.4. COLEMAN THEORY 135

P ROOF. We consider the nontrivial direction. Let m ≥ 0 be maximal with f ≡ 1 mod pm , and let
k ≥ 0 be maximal such that
f ≡ 1 + apm T k mod (pm+1 , T k+1 )

for some nonzero a ∈ O × . Since [p](T ) ≡ T p mod p, we have

[p]( f ) ≡ 1 + pm aT pk mod (pm+1 , T k+1 ).

So, if [p]( f ) ≡ 1 mod pn , then n ≤ m. 

S
Let ϕ denote the unique Frobenius element in Gal(E∞ /Q p ), where E∞ = n En , which we also let
act on Λ through its action on coefficients.

P ROPOSITION 5.4.6. If f ∈ Λ× , then N ( f ) ≡ ϕ( f ) mod p. If f ≡ 1 mod pn for some positive

integer n, then N ( f ) ≡ 1 mod pn+1 .

P ROOF. Take f ∈ Λ× , and suppose that f ≡ 1 mod pk for some k ≥ 0. We then have

f (ζ pi (1 + T ) − 1) ≡ f (T ) mod pk (1 − ζ p )

for each i ∈ Z, and our assumption on f implies that

p−1
([p] ◦ N )( f ) = ∏ f (ζ pi (1 + T ) − 1) ≡ f (T ) p mod pk+1.
i=1

If k ≥ 1, then f p ≡ 1 mod pk+1 , so Lemma 5.4.5 tells us that N ( f ) ≡ 1 mod pk+1 as well.
If k = 0, then we can at least say that f (T ) p ≡ ϕ( f )(T p ) ≡ [p]( f )(T ) mod p, so
N (f)
 
[p] ≡ 1 mod p,
ϕ( f )
and therefore Lemma 5.4.5 tells us that N ( f ) ≡ ϕ( f ) mod p. 

C OROLLARY 5.4.7. Suppose that f ∈ Λ× . For n ≥ m, we have

N n (ϕ −n ( f )) ≡ N m (ϕ −m ( f )) mod pm+1 .

P ROOF. By repeated application of Proposition 5.4.6 with k = 0, we have

N n−m
( f ) ≡ ϕ n−m ( f ) mod p,
N n−m ( f )
and again by Proposition 5.4.6, the congruence follows by applying N m ◦ ϕ −n to ϕ n−m ( f ) . 

C OROLLARY 5.4.8. Suppose that f ∈ Λ× . Then g = lim N i (ϕ −i ( f )) exists, and N (g) = ϕ(g).
i→∞
136 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

P ROOF. By Corollary 5.4.7, the limit g in question exists, and we have

N (g) = lim N i+1

(ϕ −i ( f )) = ϕ lim N i+1
(ϕ −(i+1) ( f )) = ϕ(g).
i→∞ i→∞

T HEOREM 5.4.9 (Coleman). Suppose that u = (un )n≥0 forms a norm compatible sequence of units
with un ∈ On× . Then there exists a unique f ∈ Λ× such that f (ζ pn − 1) = ϕ n (un ) for all n ≥ 0, and it
has the property that N ( f ) = ϕ( f ).

P ROOF. We choose arbitrary fn ∈ Λ× that satisfy fn (ζ pn − 1) = ϕ n (un ) for each n, and we set
gn = N n (ϕ −n ( f2n )). As (gn )n is a sequence in a compact set Λ, it has a limit point, which we call f .
We claim that this f has the desired property.
For any n ≥ m, we have
pn−m −1
m m m−n
ϕ (um ) = ϕ (NEn /Em (un )) = ϕ ∏ fn (ζ pi n−m ζ pn − 1)
i=0
n−m m−n
= (N ϕ fn )([p]n−m (ζ pn − 1))
n−m m−n
= (N ϕ fn )(ζ pm − 1).

Since 2n − m ≥ n, Corollary 5.4.7, tells us that

N 2n−m m−2n
ϕ f2n ≡ N n ϕ −n f2n mod pn+1 ,

so
ϕ m (um ) = N 2n−m m−2n
ϕ f2n (ζ pm − 1) ≡ gn (ζ pm − 1) mod pn+1 .
This forces f (ζ pm − 1) = ϕ m (um ) by taking the limit over the subsequence of (gn )n converging to f .
The power series f is unique, as its difference with any other such power series would have infin-
itely many zeros in the maximal ideal of O∞ . Note that

N ( f )(ζ pn − 1) = N ( f )([p](ζ pn+1 − 1)) = ([p] ◦ N )( f )(ζ pn+1 − 1)

p−1
= ∏ f (ζ pi n+1 − 1) = NFn+1/Fn ϕ n+1(un+1) = ϕ n+1(un) = ϕ( f )(ζ pn − 1)
i=0

for all n, which similarly forces N ( f ) = ϕ( f ). 

N OTATION 5.4.10. Let Γ̃ = Gal(E∞ /E). Let U∞ = limn On× under norm maps.
←−
We let σ ∈ Γ̃ act on f ∈ OJT K by

(σ f )(T ) = f ((1 + T )χ(σ ) − 1),

 5.4. COLEMAN THEORY 137

where χ : Γ̃ → Z× ∼
p denotes the p-adic cyclotomic character. The group Gal(E∞ /Q p ) = hϕi × Γ̃ then
acts on OJT K through the action of powers of Frobenius on coefficients and the action of Γ̃ described
above.

N OTATION 5.4.11. Set

M = { f ∈ Λ× | N ( f ) = ϕ( f )}.

D EFINITION 5.4.12. The Coleman power series attached to u = (un )n ∈ U∞ is the unique f ∈ M
such that f (ζ pn − 1) = ϕ n (un ) for all n ≥ 0.

C OROLLARY 5.4.13. The map U∞ → M that takes a norm compatible sequence to its associated
Coleman power series is a continuous Gal(E∞ /Q p )-equivariant isomorphism.

P ROOF. That the map is an injective homomorphism is a consequence of uniqueness of the power
series f attached to u by Theorem 5.4.9, and its image is in M by said theorem.
For any f ∈ M , if we set un = ϕ −n ( f (ζ pn − 1)), then
p−1
ϕ n (un ) = f (ζ pn − 1) = ϕ −1 N ( f )(ζ pn − 1) = ϕ −1 ∏ f (ζ pn+1 ζ pi n − 1) = ϕ n (NFn+1 /Fn un+1 ).
i=0

Thus f is the power series attached to (un )n ∈ U∞ . Continuity follows from the construction of the map
and is easily checked. 

L EMMA 5.4.14. For all f ∈ Λ, we have

S ([p]( f )) = p f .

P ROOF. By definition, we have that

p−1
([p] ◦ S ◦ [p]( f ))(T ) = ∑ f ([p](ζ pi (1 + T ) − 1)) = p f ([p](T )) = [p](p f )(T ).
i=0

The result then follows by injectivity of [p]. 

N OTATION 5.4.15. Let

ΛS =0 = { f ∈ Λ | S ( f ) = 0} and ΛS =pϕ = { f ∈ Λ | S ( f ) = pϕ f }.

P ROPOSITION 5.4.16. The sequence

1−[p]ϕ f 7→TrE/Q p f (0)
0 → Z p → ΛS =pϕ −−−−→ ΛS =0 −−−−−−−−−→ Z p → 0

is exact.
138 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

P ROOF. Any constant a ∈ Z p satisfies pϕa = pa = (S ◦ [p])(a) = S (a), so sits inside ΛS =pϕ . If
f ∈ ΛS =pϕ , then
S ((1 − [p]ϕ)( f )) = pϕ( f ) − pϕ( f ) = 0
by Lemma 5.4.14, so (1 − [p]ϕ)( f ) ∈ ΛS =0 . Thus, the sequence is well-defined.
Note that (1 − [p]ϕ)(a) = a − ϕ(a) = 0 for a ∈ Z p and (1 − [p]ϕ)( f )(0) = f (0) − ϕ( f (0)) for
f ∈ Λ, which is carried to 0 under TrE/Q p . Thus, the sequence is a complex.
Injectivity of the first map is obvious, so we consider exactness at ΛS =pϕ . If f ∈ ΛS =pϕ satisfies
[p]ϕ( f ) = f , then f (0) ∈ Z p , and we may replace f by g = p−m ( f − f (0)) ∈ ΛS =pϕ for m ≥ 0
maximal, supposing g 6= 0. We then have

g ≡ bT i mod (p, T i+1 )

for some b ∈ O × and i ≥ 1. But this congruence forces ϕ(g)(T p ) ≡ 0 mod (p, T i+1 ), a contradiction.
Thus, we have f = f (0) ∈ Z p .
We next consider exactness at ΛS =0 . Suppose that g ∈ Λ with TrE/Q p g(0) = 0. Then g(0) =
(1 − [p]ϕ)(b) for some b ∈ O by Hilbert’s theorem 90, and

(1 − [p]ϕ)(aT i ) ≡ aT i mod (pT i , T i+1 )

for all a ∈ O and i ≥ 1, so we can find a sequence in the image of 1 − [p]ϕ that converges to g
recursively, and thus g = (1 − p[ϕ])( f ) for some f ∈ Λ. If moreover g ∈ ΛS =0 , then

S ( f ) = S (g) + S ([p]ϕ( f )) = S ([p]ϕ( f )) = pϕ( f ).

Let ξ ∈ µq−1 (E) satisfy TrE/Q p ξ = 1. Note that ξ (1 + T ) ∈ ΛS =0 , since

p−1
[p] ◦ S (ξ (1 + T )) = ∑ ζ pi ξ (1 + T ) = 0,
i=0

and [p] is injective. Thus, the final map is surjective. 

N OTATION 5.4.17.
a. Define D : Λ → Λ on f ∈ Λ by D( f ) = (1 + T ) f 0 (T ).
b. Define log : Λ× → EJT K to be the homomorphism satisfying
∞
(−1)i−1 f i
log(1 + f ) = ∑
i=1 i
for f ∈ (p, T ) and log(ξ ) = 0 for ξ any root of unity in O.
(T )0
c. Define D log : Λ× → Λ on f ∈ Λ× by D log( f ) = (1 + T ) ff (T ).

R EMARK 5.4.18. Note that D log = D ◦ log. We also consider Dk log = Dk−1 ◦ D log for k ≥ 1.
 5.4. COLEMAN THEORY 139

L EMMA 5.4.19. For any f ∈ Λ× , the quantity

fp
 
1
log
p [p]ϕ( f )

lies in Λ.

P ROOF. We have

ϕ( f )((1 + T ) p − 1) ≡ ϕ( f )(T p ) ≡ f p (T ) mod p,

fp
so [p]ϕ( f ) = 1 + pg for some g ∈ Λ. We have

1 ∞
(−1)i−1 pi−1 gi
log(1 + pg) = ∑ ,
p i=1 i

and the latter quantity clearly lies in Λ, since i ≥ v p (i) + 1 for all i ≥ 1. 

N OTATION 5.4.20. Define L : Λ× → Λ on f ∈ Λ× by

1
L ( f ) = log f − log([p]ϕ( f )).
p

P ROPOSITION 5.4.21. We have a commutative square

L
M / ΛS =0
D log D
 1−[p]ϕ

ΛS =pϕ / ΛS =0

of continuous Gal(E∞ /Q p )-equivariant Z p -linear homomorphisms.

P ROOF. For any g ∈ Λ, we have

p−1
([p] ◦ S )(D(g)) = ∑ ζ pi (1 + T )g0(ζ pi (1 + T ) − 1)
i=0

(5.4.1) p−1
= ∑ D(g(ζ pi (1 + T ) − 1))
i=0

= D(([p] ◦ S )(g)),

employing the chain rule in the second equality. Since [p] is an injective endomorphism, it follows that
D(ΛS =0 ) ⊆ ΛS =0 .
140 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

For h ∈ M , that D log h ∈ ΛS =pϕ follows from the string of equalities

[p](S (D log h)) = D(([p] ◦ S )(log h))

= D log(([p] ◦ N )(h))
= D log(ϕ[p](h))
= [p](pϕ(D log h)),

the first using (5.4.1), the second using that log is a homomorphism, the third using h ∈ M , and the
fourth by definition of the endomorphism [p]. We also have L (h) ∈ ΛS =0 , since
p−1
[p] ◦ S (L (h)) = ∑ L (h)(ζ pi (1 + T ) − 1)
i=0
p−1
= ∑ log(h)(ζ pi (1 + T ) − 1) − log([p]ϕ(h)) = 0,
i=0
p−1 i
the last step using [p](T ) = ∏i=0 (ζ p (1 + T ) − 1).
Next, for any h ∈ Λ× , we have by the chain rule that
(h ◦ [p])0 (T ) h0 ([p](T ))
D log([p] ◦ h) = (1 + T ) = (1 + T ) p = p[p](D log h).
h ◦ [p](T ) h([p](T ))
It follows from this that the diagram commutes. Galois-equivariance of log and then L is clear, and
Galois-equivariance of D and then D log follows from the chain rule (for the Γ̃-action). 

L EMMA 5.4.22. For every n ≥ 1, we have

 
1 n1+T 1+T
S [p](T ) = Tn .
p T T
p−1 i
P ROOF. Set P = [p](T ), and note that P = ∏i=0 (ζ p (1 + T ) − 1), and apply D log to both sides. We
then have
1 + P p−1 ζ pi (1 + T )
p = ∑ i .
P i=0 ζ p (1 + T ) − 1
We have
   p−1 ζ i (1 + T )  
n 1+T p n1+P n1+T
[p] S P · n
=P ∑ i = pP = [p] pT ,
T i=0 ζ p (1 + T ) + 1 P T
which yields the result. 

N OTATION 5.4.23. Let Ω = Fq JT K, and define ∂ : Ω× → T Ω by

f 0 (T )
∂(f) = T
f (T )
for f ∈ Ω.
 5.4. COLEMAN THEORY 141

L EMMA 5.4.24. We have

T Ω = ∂ (Ω× ) + { f p | f ∈ T Ω}.

P ROOF. We claim that

( )
∞
∂ (Ω× ) = ∑ ai T i ∈ Ω | a pi = aip for all i ≥ 1 .
i=1

Let us denote the latter set by S. For this, we note that any u ∈ Ω× may be written uniquely as an
infinite product
∞
u = c ∏ (1 − bn T n )
n=1
with c ∈ F×
q and bn ∈ Fq for n ≥ 1. Note that
−nbn T n ∞
∂ (1 − bn T n ) = = −n ∑ binT ni ∈ S,
1 + bn T n i=1

so ∂ (u) ∈ S as ∂ is a continuous homomorphism. Conversely, any element of S may be written as an

infinite sum of terms of the form −n ∑∞ i ni
i=1 a T for some a ∈ Fq and n prime to p (by using it to specify
i
the coefficient of T n and thereby of the T np for i ≥ 1), and such an element is equal to ∂ (1 − aT n ).
Thus, we have the claim.
Note that we can pick the coefficients ai with p - i of a power series in S arbitrarily. The fact that
∂ (Ω× ) = S implies the result since the pth powers of elements of T Ω are exactly the power series in
T p Fq JT p K, for which we can pick the coefficients ai with p | i arbitrarily. 

P ROPOSITION 5.4.25. The map D log : M → ΛS =pϕ is surjective and has kernel the group µq−1
of roots of unity of O of prime-to-p order.

P ROOF. Clearly, the kernel of D log on Λ× is O × . Note that a ∈ O × ∩ M if and only if a p = ϕ(a).
It is easy to see that no element of 1 + pO can have this property, while every element of µq−1 does.
Thus, the kernel is as stated.
We claim first that it suffices to check the surjectivity of D log modulo p. Let Ω = Fq JT K, and note
that the formula for D log makes sense on Ω× . Let f ∈ ΛS =pϕ . Suppose by induction that there exists
hk ∈ M such that D log(hk ) ≡ f mod pk Λ. Then set
1
f0 = ( f − D log(hk )) ∈ ΛS =pϕ ,
pk
and choose h0 ∈ M such that Dh0 ≡ f 0 mod pΛ. Setting hk+1 = hk + pk h0 , we then have

D log(hk+1 ) ≡ f mod pk+1 Λ.

If we set h = limk→∞ hk , then D log(h) = f . Thus, we have the claim.

142 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

Next, we note that the reduction modulo p map M → Ω× is surjective. This is straightforward: if
f¯ ∈ Ω× , then choose any lift f of it to Λ× and consider g = limk→∞ ϕ −k N k f , which also lifts f¯ but
now lies in M . To see that D log is surjective, it is then enough to see that the image Φ of ΛS =pϕ
1+T ×
under reduction modulo p is contained in T ∂ (Ω ).
Let v ∈ Φ. By Lemma 5.4.24, we have that
T
v = ∂ (u) + f p
1+T
for some u ∈ Ω× and f ∈ T Ω. Note that 1p ϕ −1 S : Λ → Λ reduces to an operator s : Ω → Ω that fixes
1+T 1+T p 1+T p 1+T
both v and T ∂ (u), so fixes T f . On the other hand, Lemma 5.4.22 tells us that s( T f ) = T f ,
since [p](ϕ( f )) = f p in Ω. But for 1+T 1+T p
T f = T f to hold for f ∈ T Ω, we must have f = 0. Therefore,
v = 1+T
T ∂ (u), finishing the proof. 

C OROLLARY 5.4.26. The map D : ΛS =0 → ΛS =0 is a bijection.

P ROOF. The kernel of D on Λ is Z p , but S (a) = pa for all a ∈ Z p , so D is an injection. Since

D log is a surjection by Proposition 5.4.25, the exact sequence of Proposition 5.4.16 reduces us to
the claim that the composite map g 7→ TrE/Q p Dg(0) is surjective. For a ∈ O, one may observe that
S (a(1 + T )) = 0, so a(1 + T ) ∈ ΛS =0 , and D(a(1 + T ))(0) = a. The corollary now follows by the
surjectivity of trace in unramified extensions. 

We now have the following consequence of what we have proven.

P ROPOSITION 5.4.27. The diagram

(ξ ,a)7→ξ (1+T )a L f 7→TrE/Q p f 0 (0)
0 → µq−1 × Z p (1) −−−−−−−−−→ M −→ ΛS =0 −−−−−−−−−→ Z p (1) → 0

is an exact sequence in the category of compact abelian groups with continuous Gal(E∞ /Q p )-actions.

P ROOF. We use Proposition 5.4.21, Proposition 5.4.25, and Corollary 5.4.26 to replace the middle
terms in the exact sequence of Proposition 5.4.16. The fact that D log : M → ΛS =pϕ has kernel µq−1
is taken care of by adding it to the first term to preserve exactness. Note for this that the Coleman
power series attached to the norm compatible sequence (ζ pan )n for a ∈ Z p is exactly (1 + T )a , and the
map is clearly Gal(E∞ /Q p )-equivariant. Also, note that D f (0) = f 0 (0), so the last map is as stated,
and
f ((1 + T )χ(a) − 1)0 |T =0 = χ(a) f 0 (0),
so it is also Gal(E∞ /Q p )-equivariant. 

Recall that an O-valued measure on Z p is identified with an element of OJZ p K which is isomorphic
to Λ under the continuous O-linear map that takes the group element of [i] to (1 + T )i .
 5.4. COLEMAN THEORY 143

N OTATION 5.4.28. We define an operator Φ : Λ → Λ on f ∈ Λ by

1
Φ( f ) = f − [p](S ( f )).
p
By definition, f ∈ Λ satisfies Φ( f ) = f if and only if f ∈ ΛS =0 .

P ROPOSITION 5.4.29. A measure µ on Z p is the extension by zero of a measure on Z×

p if and only
if the power series attached to µ lies in ΛS =0 . In other words, the continuous O-linear isomorphism
∼ ∼
OJZ p K −
→ Λ sending the group element 1 to T + 1 restricts to an isomorphism OJZ× → ΛS =0 of
pK −
topological O-modules.
n
P ROOF. Let f be the power series attached to µ, and let f¯n ∈ O[T ]/((1 + T ) p − 1) denote its
image. For the distribution (µn )n attached to µ, we have that
pn −1
fn = ∑ µn (k)(1 + T )k ∈ O[T ]
k=0

lifts f¯n . Here, µn (k) = for χk+pn Z p the chararacteristic function of k + pn Z p . So, µ is
R
Z p χk+pn Z p dµ
the extension by zero of a measure on Z× n
p if and only if µn (k) = 0 for all 0 ≤ k ≤ p − 1 with p | k. We
claim this occurs if and only if Φ( f ) = f , which will finish the proof.
Note that for any a ∈ O, we have

1 p−1 a(1 + T )k if p - k
Φ(a(1 + T )k ) = a(1 + T )k − ∑ aζ pik (1 + T )k =
p i=0 0 if p | k.
Then
pn −1
Φ( fn ) = ∑ µn (k)(1 + T )k ∈ O[T ],
k=0
p-k
and it is clear that
n
Φ( fn ) ≡ fn mod ((1 + T ) p − 1)
if and only if µn (k) = 0 for all p | k. This holds for all n if and only if µ is the extension by zero of a
measure on Z×
p. 
∼
Note that OJZ× → ΛS =0 is Gal(E∞ /Q p )-equivariant, using the action of σ ∈ Gal(E∞ /E) by
pK −
multiplication by the group element of χ(σ ) on OJZ×
pK

D EFINITION 5.4.30. The Coleman map Col : U∞ → OJZ×

p K is the map that takes u ∈ U∞ to the
element of OJZ×
p K corresponding to L ( f ), where f is the Coleman power series attached to u.

Set ζ = (ζ pn )n ∈ U∞ . For ξ ∈ µq−1 , let ξ˜ ∈ U∞ be the unique norm compatible sequence of elements
of µq−1 with norm ξ ∈ O × .
144 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

T HEOREM 5.4.31. There is an exact sequence

R
(ξ ,a)7→ξ˜ ζ a λ 7→TrE/Q p Z p xdλ (x)
Col
0→ µq−1 × Z p (1) −−−−−−→ U∞ −−→ OJZ×
p K −−−−−−−−−−−−→ Z p (1) → 0

of continuous Gal(E∞ /Q p )-equivariant homomorphisms.

P ROOF. The Coleman map is the composite

∼ L ∼
→ M −→ ΛS =0 −
Col : U∞ − → OJZ×
pK

of the Coleman power series isomorphism with L and the isomorphism of Proposition 5.4.29. We use
this to replace the middle part of the exact sequence of Proposition 5.4.27 with Col. That the first map
is then as stated is immediate. That the final map is as stated comes from the fact that for f ∈ ΛS =0
corresponding to λ ∈ OJZ×
p K (which yields a measure on Z p by extension by zero) and the distribution
(λn )n , we have
pn −1 pn −1 Z
0 k−1
f (0) = lim ∑ kλn (k)(1 + T ) |T =0 = lim ∑ kλn (k) = xdλ (x).
n→∞ n→∞ Zp
k=0 k=0

L EMMA 5.4.32. Let µ be an O-valued measure on Z p , and let f ∈ Λ be the corresponding power
series. For all k ≥ 0, we have Z
xk dµ(x) = (Dk f )(0).
Zp

P ROOF. We have a linear functional defined by

Z
L(g) = xg(x)dµ(x)
Zp

for all g ∈ C(Z p , C p ). We then have

|L(g)| ≤ max |g(a)|
a∈Z p

for all g, so L is bounded and thus gives a measure µ1 , with a corresponding power series h ∈ Λ.
n
R x

We claim that h = D f . To see this, write f = ∑∞
n=0 cn T ∈ Λ, where cn = Z p n dµ(x). Note that
∞
Df = ∑ (ncn + (n + 1)cn+1)T n.
n=0

Write h = ∑∞ n
n=0 en T . Then  
x
Z
en = x dµ(x).
Zp n
x
x
x

Since x n = (n + 1) n+1 + n n , we have en = (n + 1)cn+1 + ncn and therefore the claim.
 5.4. COLEMAN THEORY 145

Now, to prove the lemma, it suffices (by repeated application of the claim) to show that
Z Z
xk dµ = dµk ,
Zp Zp

where µk is the measure corresponding to Dk f . We can see this by induction, it being a consequence
of the claim for k = 1. That is, if we know if for all measures with k − 1 in place of k, then
Z Z
xk−1 dµ1 = dµk ,
Zp Zp

since Dk−1 (D f ) = Dk f . But by the claim, we have

Z Z
k−1
x dµ1 = xk dµ,
Zp Zp

so we are done. 

D EFINITION 5.4.33. For k ≥ 1, the kth Coates-Wiles homomorphism δk : U∞ → O takes u ∈ U∞ to

Dk log( f )(0), where f is the Coleman power series attached to u.
∼
→ Z×
L EMMA 5.4.34. Let χ : Γ̃ − p be the p-adic cyclotomic character. Then

δk (σ (u)) = χ(σ )k δk (u)

for all u ∈ U∞ and σ ∈ Γ̃.

P ROOF. Note that for any g ∈ EJT K and a ∈ Z p , we have

D(g((1 + T )a − 1)) = a(1 + T )a g0 ((1 + T )a − 1) = a(Dg)((1 + T )a − 1).

So, by recursion we see that

(5.4.2) Dk (g((1 + T )a − 1)) = ak (Dg)((1 + T )a − 1).

We can apply this with g = log f for f the Coleman power series attached to u ∈ U∞ and a = χ(σ ) for
σ ∈ Γ̃. For this, note that the Coleman power series attached to σ (u) is σ ( f )(T ) = f ((1 + T )χ(σ ) − 1).
Therefore, we have
Dk (log σ ( f )) = χ(σ )k (Dk log f )(σ (T )),
and plugging in 0, we get the desired formula. 

P ROPOSITION 5.4.35. For u ∈ U∞ , let λu = Col(u), which we view as an O-valued measure on Z×

p.
We then have Z
xk dλu = (1 − pk−1 ϕ)δk (u)
Z×
p

for all k ≥ 1.
146 5. KUBOTA-LEOPOLDT p-ADIC L-FUNCTIONS

P ROOF. Let f ∈ Λ× denote the power series attached to u, and note that
Z
xk dλu = (Dk L ( f ))(0)
Z×
p

= Dk log( f )(0) − p−1 Dk log(ϕ( f ) ◦ [p])(0)

= δk (u) − pk−1 Dk log ϕ( f )(0)
= (1 − pk−1 ϕ)δk (u),

the second-to-last step following from (5.4.2). 

 CHAPTER 6

The Iwasawa main conjecture

6.1. Semi-local units modulo cyclotomic units

(χ)
Let F = Q(µmp ), where p - m and m 6≡ 2 mod 4. Let us first observe that we have a map Col : U∞ →
Λχ . Note that each place of Q(µm ) over p is totally ramified in F∞ . Let Vp denote the set of places over
p in any intermediate field. Recall that

U∞ ∼ Uv,∞ .
M
=
v∈Vp

Let O be the valuation ring of a place w of Q(µm ) over p. (Note that such places are totally ramified
in F = Q(µmp ), and even in F∞ .) Then O is free of rank one over Z p [∆ p ], for ∆ p the decomposition
group at p in ∆ = Gal(F/Q). Giving
O
M

v∈Vp
the structure of a Z p [Gal(Q(µm )/Q)]-module by allowing ∆ to permute the factors, it is then is free
of rank one as a Z p [Gal(Q(µm )/Q)]-module, and we can think of 1 in the term for w as a generator.
Similarly, then, we obtain
Λ̃ = Z p JZ× ∼
p,m K = ∏ OJZ×p K,
v∈Vp

and we can use this to define a Coleman map Col : U∞ → Λ̃ as the direct sum of the Coleman maps at
the places over p.

P ROPOSITION 6.1.1. There is a homomorphism Col : U∞ → Λ̃ induced by the Coleman maps at

the places of F∞ above p, fitting in an exact sequence
Col
0 → Z p [∆/∆ p ](1) → U∞ −−→ Λ̃ → Z p [∆/∆ p ](1) → 0

of Λ̃-modules.

C OROLLARY 6.1.2. Let χ be a nontrivial Dirichlet p-adic character of conductor dividing mp.
(χ)
Then the Coleman map induces a map U∞ → Λχ which is an isomorphism if and only if χω −1 (p) −
1 ∈ Oχ× .

We aim to prove the following theorem that, at least in the case of F = Q(µ p ), was proven by
Iwasawa.
147
148 6. THE IWASAWA MAIN CONJECTURE

T HEOREM 6.1.3. Let χ be a nontrivial, even primitive Dirichlet character of conductor m or mp,
where p is an odd prime and m is a positive integer prime to p. Let F = Q(µmp ), and let F∞ be its
cyclotomic Z p -extension. Then there is an exact sequence

0 → Aχ (1) → U∞(χ) /C∞(χ) → Λχ /(g0χ ) → 0,

where Aχ = Oχ /(χω −1 (p) − 1), and where g0χ = T −δ gχ for δ = 1 if χω −1 (p) = 1 and δ = 0 other-
wise, where gχ ∈ Λχ satisfies
gχ (us − 1) = L p (χ, 1 − s)

for all s ∈ Z p .

C OROLLARY 6.1.4. With the notation of Theorem 6.1.3, suppose that χω −1 (p) − 1 ∈ Oχ× . Then

U∞(χ) /C∞(χ) ∼
= Λχ /(gχ ).

P ROPOSITION 6.1.5. Suppose that p - ϕ(m). Let χ be a nontrivial, even Dirichlet p-adic character
(χ)
of conductor dividing mp, and let f be the prime-to-p part of its conductor. The Λχ -module C∞ is
generated by the image of the norm compatible sequence (1 − ζ f pn )n .

P ROOF. Since only primes over p ramify in F∞ /F, the norm compatible sequences of elements
of Fn× are all norm compatible sequences of p-units. The group of norm compatible sequences of
cyclotomic p-units in F∞ /F is generated as a Λ̃-module by (1 − ζd pn )n for d dividing m. Such a
sequence is of true units if f 6= 1. From this, it is easy to see that C∞ is similarly generated by the
(1 − ζd pn )n and ((1 − ζ pn )σc −1 )n , where c ∈ Z is a primitive root modulo p. The elements eχ (1 − ζd pn )n
vanish unless f is a multiple of d. On the other hand, eχ (1 − ζd pn ) for d a multiple of f equals eχ times
the norm for Q(ζd pn )/Q(ζ f pn ) of the elements (1 − ζd pn ). That is, eχ (1 − ζd pn )n is a Λ̃-multiple of
eχ (1 − ζ f pn )n for f dividing d. 

Let us now focus on the case that F = Q(µ p ), for which we suppose that E = Q p . Note that
U∞ = U∞ × µ p−1 . Consider the cyclotomic unit
−c/2 c/2
ζ pn − ζ pn
un,c = −1/2 1/2
ζ pn − ζ pn

for c prime to p, and let uc = (un,c )n ∈ U∞ . Let ζ˜p,c = Col(un,c ).

P ROPOSITION 6.1.6. For k ≥ 1, we have

Z
xk d ζ˜p,c (x) = (1 − ck )(1 − pk−1 )ζ (1 − k) = (1 − ck )L p (ω k , 1 − k).
Z×
p
 6.1. SEMI-LOCAL UNITS MODULO CYCLOTOMIC UNITS 149

P ROOF. We employ the Coleman power series

(1 + T )−c/2 − (1 + T )c/2
f (T ) =
(1 + T )−1/2 − (1 + T )1/2
and the change of variables T = et − 1. By Lemma 5.4.32, we have that
dk
δk (uc ) = log f (et − 1) |t=0 .
dt k
We have
   
d 1 1 1 c 1 1
log f (et − 1) = − − −
dt 2 e−t − 1 et − 1 2 e−ct − 1 ect − 1
∞ ∞
Bk B
= ∑ 2 · k! ((−t)k−1
− t k−1
+ c((ct)k−1
− (−ct)k−1
)) = ∑ k!k (ck − 1)t k−1,
k=0 k=0
so
Bk
δk (uc ) = (ck − 1) = (1 − ck )ζ (1 − k).
k
The result then follows from Proposition 5.4.35. 

We then have the following.

(0)
C OROLLARY 6.1.7. The Z p -valued measure Ec on Z×
p satisfying
Z Z
(0) (1)
×
h(x)dEc (x) = x−1 h(x)dEc (x)
Zp Z×
p

for all h ∈ C(Z p , C p ) is equal to ζ˜p,c .

S KETCH OF PROOF OF T HEOREM 6.1.3 FOR Q(µ p ). Note that C∞ is topologically generated by
the elements uc , and in particular it is generated as a Z p JZ× K ∼
= Λ[∆]-module, where ∆ = Gal(F∞ /Q∞ ),
p
by uc for any integer c that is a primitive root modulo p. Proposition 6.1.6 tells us that

Col(C∞ ) = Z p JZ× ˜
p Kζ p,c .

Note that ζ˜p = (1 − σc )−1 ζ˜p,c is independent of c, though it is not quite integral, though it becomes
integral up application of any element in the augmentation ideal I of Λ[∆]. If k 6≡ 0 mod p − 1, then
1 − ck ∈ Z×
p . The “equivariant” version of Iwasawa’s theorem is then proven: it reads

U∞ /C∞ ∼
= Λ[∆]/I ζ˜p .

Recall that a character ω k of ∆ defines an in this case surjective homomorphism ω fk : Λ[∆] → Λ of

Λ-algebras. For even k, the image ωfk (ζ˜ ) is nonzero, and it is integral if and only if k 6≡ 0 mod p − 1.
p
A simple check yields using Corollary 6.1.7 yields that the power series corresponding to ω fk (ζ˜ ) (or
p
T times it if k ≡ 0 mod p − 1) is gω k , so we have Iwasawa’s theorem. 
150 6. THE IWASAWA MAIN CONJECTURE

6.2. The Ferrero-Washington theorem

T HEOREM 6.2.1 (Ferrero-Washington). Let F be a finite abelian extension of Q, and let F∞ be its
cyclotomic Z p -extension for a prime p. Then µ(X∞ ) = 0.

The following is immediate from the theorem and Proposition 3.4.2.

C OROLLARY 6.2.2. Let F be an abelian extension of Q, and let F∞ be its cyclotomic Z p -extension
for an odd prime p. Then the p-torsion subgroup of X∞− is zero.

In this section, we prove the Ferrero-Washington theorem in the case of F = Q(µ p ) for an odd
prime p. We follow their original proof in this case.

N OTATION 6.2.3. For a ∈ Z p and a nonnegative integer m, let [a]m ∈ Z denote the unique integer
with 0 ≤ a < pm+1 to which a is congruent modulo pm+1 . Let δ0 (a) = [a]0 and δm (a) = p−m ([a]m −
[a]m−1 ) if m ≥ 1.

We may think of δm (a) as the coefficient of pm in the usual p-adic expansion of a.

P ROPOSITION 6.2.4. The µ-invariant of X∞ is nonzero if and only if there exists an even integer
k 6≡ 0 mod p − 1 such that
∑ δm (aξ )ξ k−1 ≡ 0 mod p
ξ ∈µ p−1 (Z p )

for all m ≥ 0 and all a ∈ Z p .

(ω) (ω 1−k )
P ROOF. Since X∞ is trivial, we need only show that the µ-invariant µk of X∞ is zero for every
(ω 1−k )
even k with 2 ≤ k ≤ p − 3. Since fω k annihilates X∞ , it suffices to show that fω k is not in pZ p JT K.
For b ∈ Z p , let 1 ≤ im (b) ≤ pm be such that hbi p ≡ (1 + p)im (b) mod pm+1 . The expression for fω k
given by Remark 5.3.10 reduces to
pm+1
1 m −i (b)
fω k ≡− m
p ∑ bω k−1(b)(T + 1) p m
mod ωm .
b=1
p-b
m m
Since ωm ≡ T p mod p, the congruence holds modulo (p, T p ) as well. To say that µk is nonzero
is then equivalent to saying that every coefficient of a power of T + 1 in each such expansion as we
vary m is zero. Let Tm (a) denote the set of positive integers b < pm+1 that are prime to p and satisfy
im (b) ≡ im (a) mod pm . By what we have just said, we have µk > 0 if and only if

∑ bω k−1 (b) ≡ 0 mod pm+1

b∈Tm (a)
 6.2. THE FERRERO-WASHINGTON THEOREM 151

for all a ∈ Z p and m ≥ 0. Note that im (b) ≡ im (a) mod pm+1 if and only if there exists ξ ∈ µ p−1 (Z p )
such that [b]m = [ξ a]m . For given a ≡ 1 mod p and ξ there is exactly one 0 < b < pm with p - b having
this property. Since ω(b) = ω(ξ ), we then have µk > 0 if and only if

∑ [aξ ]m ξ k−1 ≡ 0 mod pm+1 .

ξ ∈µ p−1 (Z p )

As δ0 (aξ ) = [aξ ]0 and δm (aξ ) = p−m ([aξ ]m − [aξ ]m−1 ) for all m ≥ 1, the result follows from the
equivalence of µk > 0 with the latter congruences. 

D EFINITION 6.2.5. A sequence (bi )i≥1 of tuples in [0, 1)r is uniformly distributed if for every
product U ⊆ (0, 1)r of open intervals in (0, 1), the volume of U is proportional by a positive real
number, independent of U, to the density of the bi in U, which is to say the limit of N1 |{i ≤ N | bi ∈ U}|
as N → ∞.

D EFINITION 6.2.6. For r ≥ 1, we say that (a1 , . . . , ar ) ∈ Zrp is normal if the sequence of tuples

(p−m [a1 ]m−1 , . . . , p−m [ar ]m−1 )

with m ≥ 1 is uniformly distributed in [0, 1)r .

We omit the proof of the following.

T HEOREM 6.2.7 (Weyl). A sequence (bi,1 , . . . , bi,r )i≥1 of tuples in [0, 1)r is uniformly distributed if
and only if for every tuple (t1 , . . . ,tr ) ∈ Zr − {0}, we have
1 N 2πi ∑rj=1 bi, j t j
lim ∑ e = 0.
N→∞ N
i=1

P ROPOSITION 6.2.8. For r ≥ 1, let b1 , . . . , br ∈ Z p be such that b1 , . . . , br are Q-linearly indepen-

dent. Then the complement of the set of a ∈ Z p with (ab1 , . . . , abr ) normal has Haar measure zero.

P ROOF. Let t = (t1 , . . .tr ) ∈ Zr − {0}, and let c = ∑rj=1 b j t j , which is nonzero by the linear inde-
pendence of the b j . For a ∈ Z p , we have
r r
[ac]m−1 ≡ ac ≡ ∑ ab jt j ≡ ∑ [ab j ]m−1t j mod pm.
j=1 j=1

Therefore, that (ab1 , . . . , abr ) is normal is equivalent by the criterion of Weyl to the statement that
1 N 2πip−m [ac]m−1
lim ∑e =0
N→∞ N
m=1
for all t. We claim that this holds outside a set of a of measure zero for each t. Since there are only
/ pZr , as the convergence to zero of
countably many t, this implies the result. We also suppose that t ∈
limit in question is unaffected by dividing by a power of p.
152 6. THE IWASAWA MAIN CONJECTURE

Set
1 N 2πip−m [ac]m−1
pN (a) = ∑e ,
N m=1
and note that
1 1 N 1
Z Z
−n [ac] −m [ac]
2
|pN (a)| da = + 2 ∑ e2πi(p n−1 −p m−1 ) da = ,
Zp N N m,n=1 Zp N
m6=n

each integral in the latter sum being zero, being a multiple of a sum over all pn th (resp., pm th) roots of
unity if n > m (resp., m > n). It follows that
∞ Z ∞
1 π2
∑ |pM2 (a)|2 da = ∑ 2 6 =
M=1 Z p M=1 M

is finite, which forces limM→∞ pM2 (a) = 0 outside of a set of measure zero. Note also that for any
N ∈ Z with M 2 ≤ N < (M + 1)2 , we have
2M 2
|pN (a)| < |pM2 (a)| + ≤ |pM2 (a)| + ,
N M
as pN (a) − pM2 (a) is a sum of N − M 2 roots of unity. Thus (pN (a))N has limit 0 outside of the same
measure zero set. 
p−1
P ROPOSITION 6.2.9. Set s = 2 and r = ϕ(p − 1). Let b1 , . . . , bs ∈ Z p be such that (b1 , . . . , br ) is
normal, bi b−1
1 ∈/ Z for all 2 ≤ i ≤ s, and
r
bi = ∑ ci, j b j
j=1

for some ci, j ∈ Z for all r < i ≤ s. Then there exist nonnegative integers m and n such that δn (b j ) =
δm (b j ) for all 2 ≤ j ≤ s, while δn (b1 ) = 1 and δm (b1 ) = 0.

P ROOF. Take x1 = 1p , and let x2 , . . . , xr ∈ (0, 1) be such that the x1 , . . . , xr are Q-linearly indepen-
dent. For r < i ≤ s, set * +
r
xi = ∑ ci, j x j .
j=1
If xi ∈ Q for such an i, then ci, j = 0 for 2 ≤ j ≤ r by the assumed linear indepedence, so xi = c1, j x1 .
But this would imply that bi b−1
1 ∈ Z, contradicting our hypotheses. Thus, the xi for 2 ≤ i ≤ s are all
irrational.
Let y1 ∈ (0, 1p ), set yi = xi for 2 ≤ i ≤ r, and set yi = h∑rj=1 ci, j y j i for r < i ≤ s. Suppose that x1 − y1
is sufficiently small so that for each 2 ≤ i ≤ s, we have an 0 ≤ a < p such that xi , yi ∈ ( ap , a+1
p ). Since
(b1 , . . . , br ) is normal, there exists an m ≥ 0 such that

|p−m−1 [bi ]m − yi | < ε

 6.2. THE FERRERO-WASHINGTON THEOREM 153

for a given choice of ε > 0. for all 1 ≤ i ≤ r. For r < i ≤ s, we have

!
r
p−m−1 [bi ]m − ∑ ci, j [b j ]m ∈ Z,
j=1

so
r r
|p−m−1 [bi ]m − yi | ≤ ∑ |ci, j ||p−m−1[b j ]m − yi| < ∑ |ci, j | · ε,
j=1 i=1

noting that both terms are in (0, 1) in the middle step. We may take ε small enough that small enough
p−m−1 [bi ]m lies in the same open interval ( ap , a+1
p ) as yi for all 1 ≤ i ≤ s. We then have

p−1 δm (bi ) < p−m [bi ]m < p−1 (δm (bi ) + 1),

so δm (bi ) = bpyi c, and we note that bpyi c = bpxi c for i ≥ 2. Since y1 < 1p , we have δm (b1 ) = 0.
1 2
Now repeat the argument, but this time replace y1 with z1 where p < z1 < p and z1 − x1 is small
enough. We then again obtain an n ≥ 0 such that δn (bi ) = bpxi c, this time for all i, noting that bpx1 c = 1.
Thus δn (bi ) = δm (bi ) for all i ≥ 2, while δn (b1 ) = 1 > 0 = δm (b1 ). 
p−1
P ROOF OF T HEOREM 6.2.1 FOR F = Q(µ p ) WITH p ODD . Set s = 2 and r = ϕ(p − 1). Let
ξ be a primitive (p − 1)th root of unity. Note that ξ s+1 = −ξ , so for a ∈ Z p we have δm (−aξ ) =
p − 1 − δm (aξ ) so long as m ≥ 1 + v p (a). It follows that
p−1 s s
(6.2.1) ∑ δm(aξ )ξ i(k−1) = 2 ∑ δm(aξ )ξ i(k−1) − (p − 1) ∑ ξ i(k−1)
i=1 i=1 i=1

for all a ∈ Z p and even integers k. The ξ i with 1 ≤ i ≤ r are linearly independent: in fact, they form
a Z-basis of Z[µ p−1 ] ⊂ Z p . Let a ∈ Z p be such that (aξ , aξ 2 , . . . , aξ r ) is normal, and set bi = aξ i
for each 1 ≤ i ≤ s. Then the conditions of Proposition 6.2.9 are satisfied for the bi , so we can find
nonnegative integers m and n as in its statement.
Suppose that µ(X∞− ) > 0. By Proposition 6.2.4, there exists an even 2 ≤ k ≤ p − 1 such that
p−1
∑ δl (aξ )ξ i(k−1) ≡ 0 mod p
i=1

for all l ≥ 0. Applying (6.2.1), we then have that

!
s s
2ξ k−1 = 2 ∑ δn(aξ )ξ i(k−1) − ∑ δm(aξ )ξ i(k−1)
i=1 i=1
p−1 p−1
= ∑ δn(aξ )ξ i(k−1) − ∑ δm(aξ )ξ i(k−1) ≡ 0 mod p,
i=1 i=1

providing the desired contradiction. 

154 6. THE IWASAWA MAIN CONJECTURE

6.3. The main conjecture over Q

In its most classical form, the main conjecture of Iwasawa theory, or Iwasawa main conjecture,
states that the characteristic ideals of odd eigenspaces of X∞ are generated by the power series interpo-
lating corresponding p-adic L-functions in the case that F is an abelian field and F∞ is its cyclotomic
Z p -extension. We refer to this as the main conjecture over the rationals, since it deals with fields cut
out by abelian characters of the absolute Galois group over Q. Its formulation in print is due to Green-
berg. While the main conjecture was actually proven by Mazur and Wiles in 1984, we shall label it as
a conjecture here in order to discuss its equivalent forms. We discuss its proof in later sections.

C ONJECTURE 6.3.1 (The Iwasawa Main Conjecture). Let p be an odd prime. Let χ be a nontrivial,
even finite order p-adic character of GQ of conductor not divisible by p2 , and let F be the the fixed
field of the kernel of χ. For the cyclotomic Z p -extension F∞ of F, we have
−1 )
charΛχ X∞(ω χ = ( f χ ),

where f χ ∈ Λχ satisfies
f χ ((1 + p)s − 1) = L p (χ, s)
for all s ∈ Z p .

We can reformulate the main conjecture in terms of the p-ramified Iwasawa module.

P ROPOSITION 6.3.2. The Iwasawa main conjecture is equivalent to the statement that
(χ)
charΛχ X∞ = (gχ ),

where gχ ∈ Λχ satisfies
gχ ((1 + p)1−s − 1) = L p (χ, s)
for all s ∈ Z p .

P ROOF. By Corollary 3.4.9, we have a pseudo-isomorphism

(χ) −1 )
X∞ ' (X∞(ω χ )ι (1),

and pseudo-isomorphic modules have the same characteristic ideal. We then have

gχ (T ) = f χ (u(1 + T )−1 − 1),

and the result follows. 

We can also reformulate the main conjecture as a comparison between global units modulo cyclo-
tomic units and the plus part of the Iwasawa module. This formulation eschews the use of L-functions.
 6.3. THE MAIN CONJECTURE OVER Q 155

T HEOREM 6.3.3. The Iwasawa main conjecture is equivalent to the statement that

charΛχ (E∞(χ) /C∞(χ) ) = charΛχ (X∞(χ) ).

P ROOF. From the first exact sequence of Proposition 3.3.6, we obtain an exact sequence

0 → E∞(χ) /C∞(χ) → U∞(χ) /C∞(χ) → X∞

(χ)
→ X∞(χ) → 0.

Iwasawa’s theorem tells us that the characteristic ideal of the second term has characteristic ideal (gχ ).
Since the alternating product of characteristic ideals of Iwasawa modules in an exact sequence of finite
length is 1, we have that
charΛχ (E∞(χ) /C∞(χ) ) = charΛχ (X∞(χ) )
(χ)
if and only if charΛχ X∞ = (gχ ). The latter statement is an equivalent form of the main conjecture by
Proposition 6.3.2. 

Mazur and Wiles proved the following interesting consequence of the main conjecture.

T HEOREM 6.3.4 (Mazur-Wiles). Let p, F, χ, and Oχ be as in the Iwasawa main conjecture, and
suppose that χ has prime-to-p order. We then have
(ω χ −1 )
|AF | = |B1,χω −1 |−1
χ ,

where | · |χ denotes the normalized multiplicative valuation on the unramified extension Oχ of Z p .

In particular, the converse to Herbrand’s theorem (due to Ribet) holds.

We also note that any one divisibility of characteristic ideals in the main conjecture for all χ of
the Galois group of a given totally real abelian field implies the other. This is a consequence of the
following result, which can be derived using the analytic class number formula (for instance, using
Sinnott’s work).

P ROPOSITION 6.3.5. Let F be an abelian, CM extension of Q of conductor not divisible by p2 ,

and let G = Gal(F + /Q). Let f = ∏χ∈Ĝ f χ ∈ Z p JT K, and let µ( f ) = µ(Λ/( f )) and λ ( f ) = λ (Λ/( f )).
Then

µ(X∞− ) = µ( f ) and λ (X∞− ) = λ ( f ).

As a final note, we treat the powers of the variable T itself that appear in the ideals of the main
conjecture.
(ω χ −1 )
P ROPOSITION 6.3.6. We have that T | charΛ X∞ if and only if χω −1 (p) = 1.
156 6. THE IWASAWA MAIN CONJECTURE

P ROOF. Let NF∞ /F : E∞ → EF be projection to the first term of a norm compatible sequence. Con-
sider the exact sequence
 

EF /NF∞ /F E∞ → ker 
M
Γv → Γ → (X∞ )Γ → AF
v∈Vp (F)

(ω χ −1 )
that exists by Theorem 1.3.14. We take ω χ −1 -eigenspaces. Since AF is finite, EF = 0, and
Γ (ω χ −1 ) = 0, we have that
 (ω χ −1 )
−1 )
(X∞(ω χ
M
)Γ '  Γv  ,
v∈Vp (F)

and the latter isomorphic to Z p or 0 depending on whether χω −1 (p) = 1 or not. 

T HEOREM 6.3.7 (Ferrero-Greenberg). We have T 2 - f χ , and T | f χ if and only if χω −1 (p) = 1.

(ω χ −1 )
We can see from this (and Sinnott’s work, for instance) that T 2 - charΛ X∞ for all χ as well, so
(ω χ −1 )
the same power of T divides both f χ and charΛ X∞ .

6.4. The Euler system of cyclotomic units

Let m > 1 be a positive integer, and let F = Q(µm )+ . Let ∆ = Gal(F/Q). Consider the set P of
nontrivial products of distinct prime numbers that split completely in F, which is to say are congruent
to ±1 modulo m. For any r ∈ P, we set Fr = F(µr ) for brevity, and we let Gr = Gal(Fr /F), which is
isomorphic to Gal(Q(µr )/Q) by restriction. For ` | r, we view G` as the subgroup Gal(Fr /Fr/` ) of Gr .
With this identification, if we let Nr ∈ Z[Gr ] be the norm element, we then have

Nr = ∏ N` ,
`|r

the product being (implicitly) taken over primes. Fix a generator σ` of G` for each prime ` ∈ P, and
let ϕ` denote the Frobenius in Gr for any r ∈ P not divisible by `.

D EFINITION 6.4.1. For r ∈ P, the rth derivative element is

Dr = ∏ D` ∈ Z[Gr ]
`|r

where for a prime ` ∈ P, we set

`−2
D` = ∑ iσ`i.
i=1

The `th derivative element has the following key property.

 6.4. THE EULER SYSTEM OF CYCLOTOMIC UNITS 157

L EMMA 6.4.2. For ` ∈ P, we have

(σ` − 1)D` = ` − 1 − N` .

P ROOF. We have
`−2 `−1 `−1 `−1
σ` D` = ∑ iσ`i+1 = ∑ (i − 1)σ`i = ∑ iσ`i − ∑ σ`i = (D` + ` − 1) − N`.
i=1 i=1 i=1 i=1

Fix a primitive mth root a unity ζm and a primitive `th root of unity ζ` for each ` ∈ P. For r ∈ P,
set ζr = ∏`|r ζ` . Let
αr = (ζm ζr − 1)(ζm−1 ζr − 1) ∈ Fr ,

which is a cyclotomic unit if r 6= 1 or m is composite. It has two key properties: the first is that

αr ≡ αr/` mod L

for every prime L of Fr over `. The second is the so-called Euler system relation found in the following
lemma. Note that we use additive notation for the multiplicative action of the group ring.

L EMMA 6.4.3. We have N` αr = (ϕ` − 1)αr/` .

P ROOF. Set s = `r . We have

`−1
ζm` ζs` − 1
N` (ζm ζr − 1) = ∏ (ζm ζ`i ζs − 1) = = (ϕ` − 1)(ζm ζs − 1),
i=1 ζm ζs − 1

and replacing ζm with ζm−1 , we have the lemma. 

Fix an odd positive integer n, and let Pn denote the subset of elements of P that are products of
primes that are 1 modulo n.

L EMMA 6.4.4. If r ∈ Pn , then Dr αr ∈ (Fr× /Fr×n )Gr .

P ROOF. We prove this by induction on the number of primes dividing r, the case that the number
is zero, i.e., r = 1, being clear. If r = `s for some prime ` and s in Pn , then

(σ` − 1)Dr αr = (` − 1 − N` )Ds αr = (` − 1)Ds αr + (1 − ϕ` )Ds αs

by the Euler system relation. The latter of course agrees with (1 − ϕ` )Ds αs modulo (Fr× )`−1 . Now,
by induction we have Ds αs ∈ Fs×n , and since ` ∈ Pn , this tells us that (σ` − 1)Dr αr ∈ Fr×n . Since this
holds for all `, we have proven the lemma. 
158 6. THE IWASAWA MAIN CONJECTURE

Note that µn ∩ F = {1} since F is totally real and n is odd, and this and the fact that n and r are
relatively prime tell us that µn ∩ F(µr ) = {1}. We therefore have that µn has trivial GFr -invariants, so
the sequence of base terms in the Hochschild-Serre spectral sequence yields an isomorphism
∼
(Fr× /Fr×n )Gr −
→ F × /F ×n

inverse to the inflation map H 1 (GF , µn ) → H 1 (GFr , µn )Gr . Let κr ∈ F × /F ×n denote the image of Dr αr
under this map.

T ERMINOLOGY 6.4.5. The element κr is called the Kolyvagin derivative of αr .

R EMARK 6.4.6. Note that for any y ∈ F`× , the element (1 − σ` )y = y

σ` y is necessarily a unit at
primes over `. As ` splits completely in F and all primes over it are totally ramified in F` /F, it makes
sense to take the image of (σ` − 1)y in

(OF /`OF )× ∼
= ∏(OF /lOF )× ∼
= ∏(OF` /LOF` )× .
l|` L|`

Let κ̃r denote a lift of κr to F × . Write

Dr αr = κ̃r βrn

for some βr ∈ Fr× .

L EMMA 6.4.7. The fractional ideal βr OFr is invariant under Gr .

P ROOF. For σ ∈ Gr , the element (σ − 1)βr is an nth root of (σ − 1)Dr αr , since κ̃r ∈ F. In partic-
ular, (σ − 1)βr is a unit for all σ ∈ Gr , and the result follows from this. 

Let I` denote the subgroup of the ideal group IF of F generated by the prime ideals l in OF dividing
L
a rational prime `. Then IF = ` I` , where the direct sum is taken over all primes.

L EMMA 6.4.8. If r ∈ Pn and ` is prime with ` - r, then we may choose κ̃r so that βr ∈ Fr× is a unit
at all primes over `.

P ROOF. Note that the choice of κ̃r is canonical up to an element of F ×n , so βr is similarly-well

determined exactly up to an element of F × . Since no prime over ` ramifies in Fr /F, we have that the
Gr -fixed part of the summand of IFr generated by primes over ` is I` . By Lemma 6.4.7, we can find
a ∈ F × such that aβr is a unit at all primes over `, as required. 

For a ∈ F × /F ×n , we let [a]` to denote the image of aOF in I` /nI` under the canonical projection.
 6.4. THE EULER SYSTEM OF CYCLOTOMIC UNITS 159

L EMMA 6.4.9. Let ` ∈ Pn . Then there exists a unique ∆-equivariant surjection

Π` : (OF /`OF )× → I` /nI`

such that
Π` ((1 − σ` )x) = [N` x]`
for all x ∈ F`× /F`×n .

P ROOF. Since F` /F is tamely ramified at each prime dividing `, the ∆-equivariant map
1−σ
p` : F`× /F`×n −−−→
`
(OF /`OF )×

that exists by Remark 6.4.6 is surjective. Similarly, the ∆-equivariant map q` : F`× /F`×n → I` /nI` given
by q` (x) = [N` x]` is surjective as all primes dividing ` in F are totally ramified in F` .
For x ∈ F`× /F`×n , we have p` (x) = 0 if and only if the order ` − 1 of the residue field of each prime
L over ` in F` divides the valuation vL (x), which of course implies that ` − 1 divides vl (N` (x)) for
each prime l of F over L. Since ` ∈ Pn , we then have [N` (x)]` = 0. Consequently, the map q` factors
through the map p` , producing the unique map Π` . 

Let
π` : {a ∈ F × /F ×n | [a]` = 0} → I` /nI`
be the map that takes an element a to the value of Π` on the image of a in (OF /`OF )× .

R EMARK 6.4.10. From the proof of Lemma 6.4.9, we have that x ∈ ker π` if and only if x is an nth
power modulo l for all prime l dividing `.

P ROPOSITION 6.4.11. For any r ∈ Pn and prime `, we have


π (κ ) if ` | r
` r/`
[κr ]` =
0 if ` - r.

P ROOF. If ` - r, then we saw in Lemma 6.4.8 that βr may be chosen to be a unit at all primes over
`, in which case κ̃r will also be a unit at `, and therefore [κr ]` = 0.
If ` | r, then write r = `s. We choose βs to be a unit at primes over `. Since βrn is a unit times an
element of F × , we have that vL (βrn ) is a multiple of the ramification index ` − 1 for each prime L of
Fr over `. Since such primes are unramified over F` , we can find ν ∈ F`× such that βr ν (`−1)/n is a unit
at all primes over `. Since N` ν and ν `−1 have the valuation at each L over ` and βr−n OFr = κ̃r OFr , we
therefore have [N` ν]` = [κr ]` .
Fix a prime L over ` in Fr . Since L is ramified over F, we have

(1 − σ` )ν (`−1)/n ≡ (σ` − 1)βr mod L.

160 6. THE IWASAWA MAIN CONJECTURE

Since κ̃r , κ̃s ∈ F, we have

(σ` − 1)βrn = (σ` − 1)Dr αr = (` − 1 − N` )Ds αr

= (` − 1)Ds αr − (ϕ` − 1)Ds αs = (` − 1)Ds αr − (ϕ` − 1)βsn ,

the third equality by the Euler system relation. Since αr ≡ αs mod L, and

(ϕ` − 1)βs ≡ (` − 1)βs mod L

by definition of the Frobenius, we have

(`−1)/n  (`−1)/n
Ds αr Ds αs (`−1)/n
(σ` − 1)βr ≡ ≡ ≡ κ̃s mod L.
(ϕ` − 1)βs βsn
`−1
In other words, the elements (1 − σ` )ν and κ̃s differ by an n th root of unity modulo primes over ` in
F` . We then have that Π` ((1 − σ` )ν) = π` (κs ). By Lemma 6.4.9, we have the result. 

Now suppose that p is an odd prime, and let n = pk for some k ≥ 1. The following theorem
guarantees the existence of enough primes for our application.

P ROPOSITION 6.4.12. Given an ideal class c ∈ AF , a finite Z[∆]-submodule M of F × /F ×n , and a

Galois-equivariant map θ : M → Z/nZ[∆], there exist infinitely many primes l ∈ c that lie over some
prime ` ∈ Pn such that M has trivial image in I` /nI` and there exists a unit u ∈ (Z/nZ)× such that

π` (x) = uθ (x)l mod nI`

for all x ∈ M.

P ROOF. Let E = F(µn ) and H be the p-Hilbert class field of F. The inertia group at any prime over
√
p in Gal(E/F) has index at most 2, so H ∩ E = F as p is odd. Note that Gal(E( n M)/E) injects into
Hom(M, µn ) by Kummer theory. The element ρ ∈ Gal(E/F) corresponding to complex conjugation
acts as 1 on M and as −1 on µm , so ρ acts as −1 on Hom(M, µn ). It also acts as 1 on Gal(HE/E), so
√ √
E( n M) ∩ HE = E. Since H ∩ E = F, it follows that E( n M) ∩ H = F.
Since µn ∩ F = {1}, we have Ĥ 0 (Gal(E/F), µn ) = 0 and therefore H 1 (Gal(E/F), µn ) = 0 as
Gal(E/F) is cyclic. The natural map F × /F ×n → E × /E ×n is therefore an injection, and we see that
the injection
√
n
Gal(E( M)/E) → Hom(M, µn )

is in fact an isomorphism.
Fix a primitive nth root of unity ζn , and define a homomorphism ι : (Z/nZ)[∆] → µn on group
elements by ι(1) = ζn and ι(δ ) = 1 for δ 6= 1. The homomorphism ι ◦ θ : M → µn corresponds to an
 6.4. THE EULER SYSTEM OF CYCLOTOMIC UNITS 161
√
element τ ∈ Gal(E( n M)/E) satisfying
√
τ( n x)
√n
= ι ◦ θ (x)
x
for all x ∈ M.
By what we have shown, restriction maps define an isomorphism
√ √
Gal(HE( M)/F) ∼
n n
= Gal(H/F) × Gal(E/F) × Gal(E( M)/E).
√
So, we may choose σ ∈ Gal(HE( n M)/F) such that σ |E( √n M) = τ and σ |H corresponds to c ∈ AF via
the Artin isomorphism. By the Čebotarev density theorem, there exist infinitely many primes ` that
√
are unramified in E( n M) and for which the Frobenius ϕ` at ` has the same conjugacy class as σ in
√
Gal(HE( n M)/Q).
Now fix such a prime `, and let l be a prime lying over it. Here then are its most easily derived
properties. Since σ |F = 1, the prime l has degree 1, or in other words ` ∈ P. Since σ |H corresponds
to c, we have l ∈ c. Since σ |E = 1, the prime l splits in E/F, so ` ∈ Pn . Since ` is unramified in the
√
Galois extension E( n M) of Q, we have [x]` = 0 for all x ∈ M.
The component of π` (x) ∈ I` /nI` as l is trivial if and only if x is an nth power modulo l, as in
Remark 6.4.10. On the other hand, θ (x)l ∈ I` /nI` is trivial if and only if ι ◦ θ (x) = 1, so if and only if
√ √
τ| fixes n x, and then if and only if ϕl fixes n x, and then finally if and only if x is an nth power modulo
l. Thus, there exists a u ∈ (Z/nZ)× such that the l-component of π` (x) and uθ (x)l agree for all x ∈ M.
The map
M
π` (x) − uθ (x)l : M → (Z/nZ)l ⊂ I` /nI`
l0 |`
l6=l0
is ∆-equivariant as the difference of ∆-equivariant maps, so its image is Z/nZ[∆]-stable, but its image
also lies in a subgroup of I` /nI` containing no nontrivial Z/nZ[∆]-submodule, as ∆ acts transitively on
the primes of F over `. It follows that π` (x) = uθ (x)l for all x ∈ M. 

Recall that EF = EF ⊗Z Z p , and set CF = CF ⊗Z Z p .

L EMMA 6.4.13. Suppose that χ : ∆ → Oχ× is a p-adic character of ∆. Extending χ to a primitive

(χ)
Dirichlet character, if 1 − χ(`) ∈ Oχ× for all ` | m, then eχ (1 − ζm ) generates CF as an Oχ -module.
m/d
P ROOF. Let ζd = ζm for d dividing m. The group CF is the intersection with EF of the Z p [∆]-
module generated by the elements 1 − ζd for d dividing m. Since the norm from Q(ζd ) to Q(ζe ) for
e dividing d of the element 1 − ζd is 1 − ζe so long as every prime dividing d also divides e, we can
reduce this generating set to the set of 1 − ζd with (d, md ) = 1. In general, if `1 , . . . , `k are the primes
dividing d but not e, then the norm of 1 − ζd is the application of (1 − ϕ`−1
1
) · · · (1 − ϕ`−1
k
) to 1 − ζe .
162 6. THE IWASAWA MAIN CONJECTURE

Projecting to the χ-isotypical quotient, we have that it becomes the multiple of the image of 1 − ζe by
(1 − χ(`1 )−1 ) · · · (1 − χ(`k )−1 ), which is a unit by assumption. 

We may now bound the orders of eigenspaces of even eigenspaces of p-parts of class groups. The
proof of the following result using Euler systems is due to Kolyvagin. We suppose that m is divisible
by 4 if it is even.

T HEOREM 6.4.14. Suppose that p - |∆|, and let χ be a primitive finite order p-adic character of ∆.
(χ)
Then the order of AF divides the order of (EF /CF )(χ) .

P ROOF. Let O be the Z p -algebra generated by the image of χ, and let f be its residue degree. Let
(χ)
aχ = |AF |1/ f and qχ = |(EF /CF )(χ) |1/ f , and set n = aχ qχ . Let c1 , . . . , cq be ideal classes generating
(χ)
AF as an O-module.
(χ)
Set δr = eχ κr for r ∈ Pn . Primitivity and the fact that p - |∆| imply, by Lemma 6.4.13, that CF is
free of rank 1 over Oχ , generated by δ1 = eχ α1 = eχ (ζm − 1)2 . Then qχ is the maximal integer t0 such
that δ1 ∈ (F ×t0 /F ×n )(χ) .
Let 1 ≤ i ≤ g, and suppose that for each 1 ≤ j < i, we have found primes l j ∈ c j lying over primes
j
` j ∈ Pn such that for r j = ∏h=1 `h and t j ≤ n the largest power of p such that

δr j ∈ (F ×t j /F ×n )(χ) ,

one has t j | t j−1 and

t j−1
(6.4.1) c j ∈ O(c1 , . . . , c j−1 ).
tj
We look for li with the same properties.
Let Mi be the O-submodule of F × /F ×n generated by δri−1 . Define

θi : Mi → ((Z/nZ)[∆])(χ) , θi (δri−1 ) = ti−1 eχ .

By Proposition 6.4.12, there exists a prime li ∈ ci over some `i ∈ Pn and satisfying [δri−1 ]`i = 0 and

π`i (δri−1 ) = uiti−1 eχ li

for some ui ∈ (Z/nZ)× . Now let ri = ∏ij=1 ` j and ti ≤ n be the largest power of p such that δri ∈
F ×ti /F ×n .
We have by Proposition 6.4.11 that

(6.4.2) [δri ]`i = π`i (δri−1 ) = uiti−1 eχ li

in I`i /nI`i . Since δri ∈ F ×ti /F ×n , this forces ti | ti−1 . In particular, ti divides t0 = qχ , and therefore
n
aχ = qχ divides tni .
 6.5. THE MAIN CONJECTURE VIA EULER SYSTEMS 163

Proposition 6.4.11 also tells us that [δri ]` = 0 unless ` | ri . Thus δ`i has a ti th root in F × and nonzero
valuation modulo n only at primes dividing `1 , . . . , `i . It follows that t1i [δri ]`i has trivial image in the
(χ)
quotient of AF by the O-span of the classes c1 , . . . , ci−1 of l1 , . . . , li−1 . Moreover, we have by (6.4.2)
that
1 ti−1 n
[δri ]`i ≡ ui eχ li mod I`i ,
ti ti ti
ti−1
This implies that ti ci ∈ O(c1 , . . . , ci−1 ), completing the recursion.
Multiplying together (6.4.1) for 1 ≤ j ≤ g gives that aχ divides
g
ti−1 qχ
∏ = ,
i=1 ti tg
which clearly divides qχ . 

6.5. The main conjecture via Euler systems

Let p be an odd prime. Let m be a positive integer not divisible by p and divisible by 4 if m is even.
p n−1
Set F = Q(µmp ), and let Fn = Q(µmpn ) for n ≥ 1 and F∞ = ∞
S
n=1 Fn . Let Γ(n) = Γ = Gal(F∞ /Fn )
and Γn = Gal(Fn /F) ∼
= Z/p Z. Let χ : (Z/mpZ) → O be an even character of order prime to p,
n−1 × ×

where O = Oχ , which we also view as a primitive O-valued Dirichlet character. Set Λ = OJΓK and
Λn = O[Γn ]. We make a usual choice of identification of Λ with OJT K.

L EMMA 6.5.1.
(χ) (χ)
a. The restriction map (X∞ )Γ(n) → Xn is an isomorphism.
(χ) (χ)
b. The restriction map (X∞ )Γ(n) → Xn has trivial kernel unless χ(p) = 1 and χ 6= 1, in which
case it is isomorphic to O. It has trivial cokernel unless χ = 1, in which case it is a finite quotient of
Γn that is zero for sufficiently large n.
(χ) (χ)
c. The inverse limit of norm maps (U∞ )Γ(n) → Un is an injection unless χω −1 (p) = 1 and a
surjection unless χ(p) = 1.
(χ) (χ)
d. The inverse limit of norm maps (C∞ )Γ(n) → Cn is an injection which is an isomorphism if
χ(p) 6= 1.

Note that if M is a finitely generated Λ-module such that MΓ is finite, then it is torsion and T
does not divide its characteristic ideal, so M Γ is finite as well, and in particular M Γ is contained in the
maximal finite submodule Mfin of M.

P ROPOSITION 6.5.2. Suppose that χ(p) 6= 1 and χω −1 (p) 6= 1. Then there exists an open ideal a of
(χ) (χ)
Λ that annihilates both the kernel and cokernel of the inverse limit of norm maps Nn : (E∞ )Γn → En .
164 6. THE IWASAWA MAIN CONJECTURE

P ROOF. Consider the following two commutative diagrams with top rows arising from taking the
Γ(n)-homology of Proposition 3.3.6 (noting Theorem 3.3.4) and the bottom rows coming from Theo-
rem 1.5.4 (noting Theorem 1.5.21):
(χ)
(X∞ )Γ(n) / (U∞(χ) /E∞(χ) )Γ(n) / (X∞
(χ)
)Γ(n) / (X∞(χ) )Γ(n) /0

πn o o
  
0 / Un /En(χ)
(χ) / Xn(χ) / An(χ) /0

and
(χ) (χ)
(U∞ /E∞ )Γ(n) / (E∞(χ) )Γ(n) / (U∞(χ) )Γ(n) / (U∞(χ) /E∞(χ) )Γ(n) /0

Nn o πn
  
0 / En(χ) / Un(χ) / Un /En(χ)
(χ) / 0.

By Lemma 6.5.1 and our assumption that χ(p) 6= 1, we have that the rightmost two vertical arrows in
the first diagram and the middle vertical arrow in the second diagram are isomorphisms. In particular,
we have exact sequences
π (χ) (χ)
(6.5.1) (X∞(χ) )Γ(n) → (U∞(χ) /E∞(χ) )Γ(n) −→
n
Un /En → 0.

and
N (χ)
(6.5.2) (U∞(χ) /E∞(χ) )Γ(n) → (E∞(χ) )Γ(n) −→
n
En → ker πn → 0.
(χ) (χ) (χ)
Since (X∞ )Γ(n) ∼
= An is finite, so is (X∞ )Γ(n) , and being contained in the maximal finite Λ-
(χ)
submodule of Xn , it has order bounded independent of n. By (6.5.1) and (6.5.2), we then have that
(χ) (χ)
coker Nn ∼
= ker πn also has bounded order. The group Xn ∼ = (X∞ )Γ(n) is finite by our assumption on
(χ) (χ)
χ and the theorem of Ferrero-Greenberg, so its subgroup Un /En is finite as well. Equation (6.5.1)
(χ) (χ) (χ) (χ)
then also yields that (U∞ /E∞ ) is finite, so ker Nn ∼
Γ(n) = (U∞ /E∞ )Γ(n) is finite of order bounded
(χ) (χ)
in n by the order of the maximal finite submodule of U∞ /E∞ (which is in fact trivial). Letting a be
(χ) (χ) (χ)
the annihilator of (X∞ )fin ⊕ (U∞ /E∞ )fin , we are done. 
(χ)
Let hχ denote a characteristic power series of X∞ , and let jχ denote a characteristic power series
(χ) (χ)
of E∞ /C∞ .

P ROPOSITION 6.5.3. Suppose that χ(p) 6= 1. Then there exists an open ideal a of Λ such that for
(χ)
all λ ∈ a and n ≥ 1, there exists a Λn -module homomorphism θn,λ : En → Λn such that

θn,λ (C∞(χ) ) = λ jχ Λn .
 6.5. THE MAIN CONJECTURE VIA EULER SYSTEMS 165

(χ) ∼
P ROOF. Under our assumption that χ(p) 6= 1, we have U∞ = Λ via the product of Coleman maps
(χ)
by Corollary 6.1.2. Consequently, its submodule E∞ is torsion-free of rank one. In particular, there
(χ) (χ)
exists an injective pseudo-isomorphism θ : E∞ → Λ. By Proposition 6.1.5, the Λ-module C∞ is
cyclic, so
θ (C∞(χ) ) = charΛχ (Λχ /θ (C∞(χ) )) = charΛχ (E∞(χ) /C∞(χ) ) = ( jχ ).
(χ)
Now let a be as in Proposition 6.5.2. Let θn : (E∞ )Γ(n) → Λn be the map induced by θ . For any
(χ)
λ ∈ a and u ∈ En , we let θn,λ (u) be θn (v) for any v with Nn (v) = λ u, which exists since λ annihilates
coker Nn and is unique since θn is necessarily trivial on the finite kernel of Nn . The result then follows
by definition of θn,λ . 
(χ)
L EMMA 6.5.4. Suppose that χ(p) 6= 1. Let fi with 1 ≤ i ≤ g be such that X∞ ' ∏gi=1 Λ/( fi ). Then
(χ)
there exists an open ideal b of Λ such that, for each n ≥ 1, there exist elements c1 , . . . , cg of An such
(χ)
that the annihilator Ann(ci ) of each ci as an element of the Λn -module An /Λn (c1 , . . . , ci−1 ) satisfies
b Ann(ci ) ⊆ fi Λn .

P ROOF. By the given pseudo-isomorphism, there exists an exact sequence

g
Λ/( fi ) → X∞(χ) → Q → 0
M
0→
i=1
(χ) (χ)
with Q finite. Taking Γn -homology and noting that (X∞ )Γ(n) ∼
= An by assumption, we obtain an
exact sequence
g
Γ(n)
M (χ)
(6.5.3) Q → Λn / fi Λn → An → QΓ(n) → 0.
i=1

Let b be the Λ-annihilator of Q. Let ci be the image of the generator ei of the ith summand Λn / fi Λn in
(χ)
An .
If x ∈ Λn is such that x · ci ∈ Λn (c1 , . . . , cn ), then we have xei is in the sum of the image of QΓ(n)
Lg
by (6.5.3). Since b annhilates Q, any λ ∈ b satisfies λ xei ∈
L
and j6=i Λn / f j Λn in j=1 Λn / f j Λn
but λ xei is clearly in the ith summand, so λ xei = 0. In other words, b Ann(ci ) ⊆
L
j6=i Λn / f j Λn ,
fi Λn . 

Let us now work over Fn+ .

L EMMA 6.5.5. Let r ∈ Pm for a power m of p, let ` be a prime divisor of r, and let q be a prime of
r (χ)
Fn+ over `. Let B be the subgroup of A+
n generated by the primes dividing ` . Let c = eχ [q] ∈ An . Let
δr = eχ κr , and let M = Λn δr ⊆ (F × /F ×m )(χ) . Suppose that we can choose
(χ)
m ≥ |An | · |(I` /mI` )(χ) /Λn [δr ]` |.
166 6. THE IWASAWA MAIN CONJECTURE

Let I ⊆ Λn denote the annihilator of the image of c in A+

n /B. Suppose also that λ , f ∈ Λn are such that

λ I ⊂ f Λn , and Λn / f Λn is finite.

Then there exists a Λn -module homomorphism

θ : M → (Λn /mΛn )(χ)

such that for

η : F × /F ×n → Λn /mΛn , η(x)q = π` (x),

we have

θ (δr ) = λ η(δr ).

(χ)
P ROOF. By assumption, we have m · An = 0. Define η̃ : Fn× → Λn by η̃(x)q = (x)` , where (x)`
denotes the image of (x) in I` so that η̃ lifts η. Let δ̃r ∈ Fn× be a lift of δr . Then (δ̃r ) is a multiple of m
at primes not dividing r, so its image in A+
n /B is η(δr )q, but also zero as the image of a principal ideal.
Thus η(δr ) ∈ I, and so λ η̃(δ̃r ) ∈ f Λn by assumption. Since Λn / f Λn is finite, we may set

λ η̃(δ̃r )
α= ∈ Λn .
f

We define θ as in the lemma as the unique Λn -module homomorphism with θ (δr ) = α, if it exists.
(χ)
If a ∈ Λn is such that aδr = 0, then [aδr ]` = 0. For h = |An |, we have

m
· (I` /mI` )(χ) ⊆ Λn [δr ]` ,
h

so we must have a ∈ hΛn . Writing aδ̃r = xm for some x ∈ Fn× , we have eχ [x]` = [ m1 aδ̃r ]` , and (x) has
(χ) 1
valuation a multiple of h at primes not dividing r. Since h · An = 0, the element m (aδ̃r )` has trivial
(χ)
image in An /B. In other words, m1 η̃(aδ̃r ) · c ∈ B. Then

aα f = aλ η̃(δ̃r ) ∈ m f Λn ,

so θ is 0 on aδr . Therefore, θ is well-defined. 

We now come to our proof of a divisibility in the main conjecture. For now, the proof is omitted.

(χ) (χ) (χ)

T HEOREM 6.5.6. If χ(p) 6= 1 and χω −1 (p) 6= 1, then char(X∞ ) divides char(E∞ /C∞ ).
 6.6. GEOMETRY OF MODULAR CURVES 167

6.6. Geometry of modular curves

The original approach of Mazur and Wiles to the main conjecture was a heavily involved study of
Galois actions on the cohomology of modular curves, inspired by the work of Ribet in his proof of the
converse to Herbrand’s theorem, which looked at the Galois representations attached to a newform sat-
isfying a mod p congruence with an Eisenstein series. The work of Wiles was a significant refinement,
and in some sense simplification, of the work of Mazur-Wiles that employed Hida theory and Galois
representations constructed out of pseudo-representations to complete the proof of the more general
main conjecture over totally real extensions of Q. Back in the setting of the main conjecture over Q,
a further simplification of Wiles’ work can be found in the work of Masami Ohta (for primes p ≥ 5).
In this setting, the Galois representations that Wiles constructs are quotients of inverse limits of coho-
mology groups of modular curves, so one can study cohomology directly. It is this approach that we
will attempt to roughly sketch here. For this, we will have to assume substantially more background
than earlier in these notes, so we will try to focus on ideas to compensate for this.
For a given level N ≥ 4, the modular curve X1 (N) may be defined as a scheme over Z. Over,
Z[ N1 ], it is a compactification of the fine moduli scheme Y1 (N) that represents the functor that to a
Z[ N1 ]-scheme S associates the set of pairs (E, P), where E is an elliptic curve over S and P is a point
of order N generating a subgroup scheme of E/S isomorphic to (Z/NZ)/S . If we consider the base
change X1 (N) of X1 (N) to Q, then its p-adic étale cohomology group TN = Hét1 (X1 (N), Q p (1)) has a
continuous action of GQ that is unramified outside of the primes over N and ∞.
There is also an action on cohomology of Hecke operators given by correspondences. To describe
this, we remark that a choice of embedding Q ,→ C gives rise to an isomorphism

∼
TN −
→ H 1 (X1 (N)(C), Z p )

of Q p -vector spaces, where the right-hand side is singular cohomology. This isomorphism commutes
with the actions of Hecke operators, so we can describe them on the right side. Recall that X1 (N)(C)
is a quotient of the union H∗ of the upper-half place H and Q ∪ {∞} by the congruence subgroup

( ! )
a b
Γ1 (N) = ∈ SL2 (Z) | (c, d) ≡ (0, 1) mod N .
c d

For a prime `, set

( ! )
a b
Γ1 (N, `) = ∈ Γ1 (N) | c ≡ 0 mod ` .
c d
168 6. THE IWASAWA MAIN CONJECTURE

Consider the diagram

Γ1 (N, `)\H∗
ψ` π`

w '
X1 (N)(C) X1 (N)(C)

where ψ` is induced by multiplication by ` on H∗ and π` is induced by the identity. This gives rise to
two correspondences on H 1 (X1 (N)(C), Z p ) which are in a sense dual: we take the dual correspondence
T ∗ (`) given by pullback by ψ` followed by pushforward by π` . (The usual Hecke correspondence T (`)
is given instead by (ψ` )∗ π`∗ .) We also have dual diamond operators h ji∗ for j ∈ (Z/NZ)× (inverse to


the usual ones) that are the automorphisms induced by the maps on Y1 (N) given by ac db ∈ SL2 (Z)
with d ≡ j−1 mod N. We let h(N) denote the Hecke algebra of endomorphisms of H 1 (X1 (N)(C), Z p )
generated by these dual correspondences and diamond operators. Back on étale cohomology, the
Galois and Hecke actions commute.
If N | M, then we have trace maps Tr : TM → TN given on singular cohomology by summing over
Γ1 (N)/Γ1 (M)-conjugates (upon pullback to TM via the injective map induced by the identity on H).
One key reason for our use of dual Hecke operators is that the trace map commutes with their actions.
In particular, if we consider a tower of modular curves X1 (N pn ) for a fixed N ≥ 1 not divisible by p and
n ≥ 1, then we have an inverse limit of cohomology groups limn Tmpn under trace maps. Of particular
←−
interest to us is the T ∗ (p)-ordinary part T = limn Tmp
ord of H: it is the maximal direct summand of H
n
←−
on which T ∗ (p) acts invertibly. The ordinary part h∗ = limn h(mpn )ord inverse limit of Hecke algebras
←−
acting on T . This Hecke algebra h∗ is known as Hida’s ordinary (dual, cuspidal) Z p -Hecke algebra of
tame level m.
One of the key properties of Hida’s ordinary Hecke algebra h is it nicely encapsulates the structure
of ordinary parts of cuspidal Hecke algebras of all weights and levels. The Hecke algebra is free of
finite rank over the Iwasawa algebra Λ = Z p JT K, where T = h1 + pi∗ − 1. If for k ≥ 2 and n ≥ 1, the
ordinary part of the weight k, level N pn Hecke algebra that acts on H 1 (X1 (mpn )(C), Symk−1 (Z2p ))ord ,
n n (k−2)
is isomorphic to h/((1 + T ) p − (1 + p) p ). Moreover, the latter cohomology group is isomorphic
n n (k−2)
to the quotient of the free of finite rank Λ-module T by the action of (1 + T ) p − (1 + p) p .
It is perhaps more typical to speak of Hida’s Hecke algebra as acting on the space of ordinary
Λ-adic cusp forms via the usual (not dual) action of Hecke operators. (The algebras of usual and
dual Hecke algebras are isomorphic via the map that takes a Hecke operator to the corresponding
dual operator.) For this, one has the theory of Λ-adic modular forms, which are q-expansions with
coefficients in Λ that specialize upon plugging in (1 + p)k−2 − 1 for T to weight k cusp forms for each
(or, equivalently, all but finitely many) k ≥ 2. For an eigenform to be T (p)-ordinary means that its
pth Fourier coefficient is a unit. Again, we have the same sort of good control when we specialize
 6.6. GEOMETRY OF MODULAR CURVES 169

at various weights and levels. Let us denote the h-module of Λ-adic cusp forms by S . Hida proved
that the pairing h × S → Λ of Λ-modules that takes (T, f ) to the q-coefficient of T f is perfect, so
h∼= HomΛ (S , Λ) and S ∼= HomΛ (h, Λ). Moreover, S ⊗Λ Q, where Q is the quotient field of Λ, is
free of rank one over h ⊗Λ Q.
One sees that T fits in an exact sequence of Z p JGQ p K-modules of the form

0 → Tsub → T → Tquo → 0,

where Tquo has unramified action and is noncanonically isomorphic to the space of ordinary Λ-adic
cusp forms via an isomorphism that switches dual and usual Hecke actions. The key point here is
that for the Galois representation T to be ordinary for T ∗ (p) means also to be ordinary in the sense
of p-adic Hodge theory, which insures that it has a filtration of the above form. The Hecke operator
T ∗ (p) acts as the Frobenius ϕ p on Tquo . The characteristic polynomial of the Frobenius ϕ` for ` -
mp acting on the rank two module T ⊗Λ Q is an h ⊗Λ Q-representation with T ∗ (p)-action given
by x2 − T ∗ (`)x + `h`i∗ . One might roughly think of T as encapsulating all of the p-adic Galois
representations attached to ordinary cusp forms of tame level (dividing) m at once.
A version of Poincaré duality, modified to be compatible with the inverse limit, sets up a perfect
pairing of Λ-modules ( , ) : T × T → Λ such that (T x, y) = (x, Ty) for x, y ∈ T and T ∈ h, and this
induces a perfect pairing Tsub × Tquo → Λ. From this and the duality between Hida’s Hecke algebra
and ordinary Λ-adic cusp forms, we see that Tsub ∼
= h. We remark that we may lift Tquo ⊗Λ Q to
a subspace of T ⊗Λ Q complementary to Tsub ⊗Λ Q. We would preferably lift Tquo itself, but it is
not clear one can do this if θ ω −1 (p) = 1. However, we can get away with something close in all
eigenspaces using the action of a chosen element ν of the inertia group I p at p with ν(ζ pn ) = ζ p1+p
n for
all n. Set u = (1 + T )(1 + p) and Λ0 = Λ[(u − 1)−1 ]. We declare T + to be the h ⊗Λ Λ0 -submodule
fixed by ν. This clearly works, as the determinant in Q × of the action of ν is u, but ν acts trivially on
the quotient Tquo ⊗Λ Λ0 . We set T − = Tsub ⊗Λ Λ0 .
By picking an ordered basis of T ⊗Λ Q from T − and T + , respectively, we see that the Galois
representation !
a b
ρ : GQ → GL2 (h ⊗Λ Q), ρ=
c d
is upper-triangular on GQ p and has the form
!
det ρ b
ρ|Ip =
0 1
on the inertia subgroup I p . We are particularly interested in the map c.
Let I denote the ideal of h generated by all T ∗ (`) − 1 − `h`i∗ for primes ` - mp and T ∗ (`) − 1 for
primes ` | mp, and fix an even p-adic character θ of (Z/mpZ)× of conductor m or mp. Set χ = θ ω 2 .
170 6. THE IWASAWA MAIN CONJECTURE

The image Iθ of I in h(θ ) (which corresponds to the θ −1 -eigenspace of the usual non-cuspidal Hecke
algebra acting on the space Λ-adic cuspidal modular forms) is the image of the ideal of the Hecke
algebra acting on Λ-adic modular forms that is the annihilator of the Λ-adic Eisenstein series
∞
1 0
Gθ −1 = gχ + ∑ ∑ dθ −1 (d)hκ(d)iqn ,
2 n=1 d|n
(d,mp)=1

where κ(d) is the projection of d ∈ Z× 0 0

p,m into 1 + pZ p . Here gχ = gχ if θ ω(p) 6= 1 and gχ = (T −
p)−1 gχ otherwise.
The quotient (h/I)(θ ) measures, in a sense, the failure of the above Eisenstein series Gθ −1 to be a
cusp form. This Eisenstein series induces map from Hida’s full modular Hecke algebra H acting on
the space of Λ-adic modular forms to Λθ −1 , taking T (`) to the corresponding Fourier coefficient, and
its kernel is the Eisenstein ideal in the θ −1 -eigenspace of this Hecke algebra. On the dual cuspidal
Hecke algebra h(θ ) , this yields a surjection (h/I)(θ ) → Λθ /ι(g0χ ) since Gθ −1 becomes a cusp form
when reduced modulo its constant term. In fact, this surjection is an isomorphism for θ 6= ω 2 , though
we shall not require it in our proof.
/ (Λχ [T −1 ])× . Note that T divides f χ if and only if χω −1 (p) = θ ω(p) = 1.
Now suppose that f χ ∈
(Recall that f χ ((1 + p)s − 1) = L p (χ, s) for all s ∈ Z p .). By the result of Ferrero and Greenberg, T
exactly divides f χ in the “exceptional” case that χω −1 (p) = 1, and T - f χ for non-exceptional χ.
We shall be interested in the θ -eigenspaces (under the action of diamond operators) of our Galois
representation ρ that is defined by T (θ ) ⊗Λ Q, so we view ρ as taking values in GL2 (h(θ ) ⊗Λ Q) by
projection.

L EMMA 6.6.1. For σ , τ ∈ GQ , the elements a(σ ) − det ρ(σ ), d(σ ) − 1, and b(σ )c(τ) of h(θ ) ⊗Λ Q
are all contained in Iθ ⊂ h(θ ) .

P ROOF. Note that h = Endh (h) = Endh (S ), so a and d take values in h, and moreover b(σ )c(τ) ∈
h for all σ , τ ∈ GQ since compositions of elements in Homh (S , h) and Homh (h, S ) lie in one of the
aforementioned endomorphism groups.
It suffices to show the containments in question on Frobenius elements ϕ` (or their “geometric”
inverses) at ` | N p by the Čebotarev density theorem. One has that

a(ϕ`−1 ) + d(ϕ`−1 ) = `−1 T (`) = `−1 h`iT ∗ (`) ≡ 1 + `−1 h`i mod Iθ .

Since det ρ(ϕ`−1 ) = `−1 h`i for all `, we therefore have

a(σ ) + d(σ ) ≡ 1 + det ρ(σ ) mod Iθ

 6.6. GEOMETRY OF MODULAR CURVES 171

for all σ ∈ GQ . The element ν used to lift Tquo satisfies

!
u 0
ρ(ν) = ,
0 1
where u = (1 + p)(T + 1). Taking the trace of ρ(νσ ), we see that

ua(σ ) + d(σ ) ≡ 1 + u det ρ(σ ) mod Iθ ,

again for all σ . It follows that a(σ ) − det ρ(σ ) ∈ Iθ and d(σ ) − 1 ∈ Iθ .
Now consider σ , τ ∈ GQ and note that a(σ τ) = a(σ )a(τ) + b(σ )c(τ). Thus we have

b(σ )c(τ) = (a(σ τ) − det ρ(σ τ)) − (a(σ )a(τ) − det ρ(σ ) · det ρ(τ)) ∈ Iθ .

Let B (resp., C) denote the h⊗Λ Λ0 -submodules of h(θ ) ⊗Λ Q generated by the elements b(σ ) (resp.,
c(σ )) with σ ∈ GQ . The h-module BC of sums of products is an ideal of h(θ ) contained in Iθ .

L EMMA 6.6.2. The ideal BC of h(θ ) ⊗Λ Λ0 is a faithful h(θ ) -module.

P ROOF. The map δ : GQ → (h/BC)× induced by σ 7→ d(σ ) is a homomorphism that is unramified

outside of the primes over m. It is then at most tamely ramified at these primes, so by class field
theory the map factors through a quotient of ∏`|m Z× ×
` . Since the pro-abelian group (h/BC) has finite
prime-to-p part and the group ∏`|m Z×
` has finite p-part, we see that the image of δ is finite
For ` - mp, we have

`−1 h`i(T ∗ (`) − 1 − `h`i∗ ) = a(ϕ`−1 ) + d(ϕ`−1 ) − det ρ(ϕ`−1 ) − 1

= −(a(ϕ`−1 ) − 1)(d(ϕ`−1 ) − 1) + b(ϕ`−1 )c(ϕ`−1 ),

By the Čebotarev density theorem, we can find infinitely many primes ` - mp such that d(ϕ`−1 ) − 1 ∈
BC. For such an `, we have then T ∗ (`) − 1 − `h`i∗ ∈ BC. This element is not a zero divisor in h(θ )
(as it does not annihilate any ordinary Λ-adic cuspidal eigenform with character θ −1 , which we do not
verify here), so the annihilator of BC in h(θ ) is trivial. 

We have the following corollary.

C OROLLARY 6.6.3. The h(θ ) ⊗Λ Λ0 -modules B and C are faithful.

Let F = Q(µmp ), and let F∞ be its cyclotomic Z p -extension.

P ROPOSITION 6.6.4. The map c̄ : GQ → C/Iθ C induced by c restricts to a homomorphism on GF∞

(ω χ −1 )
with the same image as c̄ and which factors through X∞
172 6. THE IWASAWA MAIN CONJECTURE

P ROOF. For σ , τ ∈ GQ , we have that

c(σ τ) = a(τ)c(σ ) + c(τ)d(σ ) ≡ det ρ(τ)c(σ ) + c(τ) mod Iθ .

Since det ρ factors through Gal(F∞ /Q), we see that c̄ is a homomorphism, and it factors through X∞
since c|Ip = 0.
j
For σ j ∈ Gal(F∞ /Q) with σ j (ζmpn ) = ζmpn for all n, where j ∈ Z×
p,m , we have

det ρ(σ j ) = j p h ji∗ = ωθ ( j)κ( j)hκ( j)i∗ .

In particular, for j ∈ (Z/mpZ)× and τ ∈ GF∞ , we have

c̄(σ j τσ −1 −1 −1 −1
j ) = det ρ(σ j ) c̄(τ) = (ωθ ) ( j)c̄(τ) = ω χ ( j)c̄(τ).

Finally, letting σ ∈ GQ , the commutator [ν, σ ] lies in GF∞ , and we have

c̄([ν, σ ]) = (u−1 − 1)c̄(σ ).

Since u−1 − 1 is a unit in Λ0 , we are done. 

Using the fact that C is a faithful h(θ ) -module and the theory of Fitting ideals, one can show that
the characteristic ideal of C/Iθ C as a module over the algebra Λθ of diamond operators is divisible
(ω χ −1 )
by g0ω 2 θ −1 . Since X∞ maps surjectively to C/Iθ C via c̄, we obtain the following theorem (upon
application of the theorem of Ferrero and Greenberg to deal with exceptional zeros).
(ω χ −1 )
T HEOREM 6.6.5. The ideal ( f χ ) divides charΛχ X∞ .
 APPENDIX A

Duality in Galois cohomology

Fix a prime p. For a field E, let GE denote its absolute Galois group, i.e., the Galois group of its
separable closure E sep as an extension of E. Let µ p∞ denote the group of all p-power roots of unity in
E sep .
Now let E be a nonarchimedean local field of characteristic not equal to p. Recall that its Brauer
group Br(E) = H 2 (GE,S , (E sep )× ) is canonically isomorphic to Q p /Z p by class field theory. This has
the following corollary.

L EMMA A.0.1. We have an isomorphism

∼
H 2 (GE , Q p /Z p (1)) −
→ Q p /Z p .

P ROOF. Of course, we may replace Q p /Z p (1) by µ p∞ in the statement, which in fact will make the
isomorphism canonical. First, we remark that, since direct limits are exact, we have an isomorphism

H 2 (GE , µ p∞ ) ∼
= lim H 2 (GE , µ pn ).
−→
n
Kummer theory sets up an exact sequence
pn
0 → H 2 (GE , µ pn ) → Br(E) −→ Br(E),

the left exactness following from Hilbert’s theorem 90. Since the pn torsion in Br(E) is canonically
isomorphic to 1/pn Z/Z and Q p /Z p is the direct limit of the latter groups, we have the result. 

R EMARK A.0.2. If T is a finite Z p [GE ]-module, then H i (GE , T ) is finite for every i.

T HEOREM A.0.3 (Tate duality). Let T be a finite Z p [GE ]-module. Then for i ∈ Z the cup product

H i (GE , T ) × H 2−i (GE , T ∨ (1)) → H 2 (GE , Q p /Z p (1))

is nondegenerate, inducing an isomorphism

H i (GE , T ) ∼
= H 2−i (GE , T ∨ (1))∨ .

R EMARK A.0.4. In fact, we have, more generally, such a duality for compact Z p -modules T with
continuous GE -actions. Here, we must use continuous cohomology, i.e., the cohomology groups of
the complex of continuous GE -cochains with values in T . We will in general denote such cohomology
groups using the same notation as the usual profinite cohomology groups.
173
174 A. DUALITY IN GALOIS COHOMOLOGY

Finally, let F denote a global field of characteristic not equal to p, and let S be a finite set of primes
of F.

D EFINITION A.0.5. Let T be a finite Z p [GF,S ]-module. For i ∈ {1, 2}, the ith Shafarevich-Tate
group of T is  
i i ∑ Resv M i
X (GF,S , T ) = ker H (GF,S , T ) −−−→ H (GFv , T ) ,
v∈S
where the map Resv is the compostion of restriction to a decomposition group at v ∈ S in GF,S with
inflation to the absolute Galois GFv .

The duality theorem is then as following

T HEOREM A.0.6 (Poitou-Tate duality). Let T be a finite Z p [GF,S ]-module. For i ∈ {1, 2}, we have
isomorphisms
∼
→ X3−i (GF,S , T ∨ (1))∨ .
Xi (GF,S , T ) −

For v ∈ S∞ and any i ∈ Z, we will use H i (GFv , T ) to denote the ith Tate cohomology group of T , by
abuse of notation.

R EMARK A.0.7. For v ∈ S∞ , the cup product induces isomorphisms

∼
→ H 2−i (GFv , T ∨ (1))∨
H i (GFv , T ) −

for all i ∈ Z.

Combining Poitou-Tate duality with Tate duality, we obtain the following nine-term exact se-
quence.

T HEOREM A.0.8 (Poitou-Tate sequence). For a finite Z p [GF,S ]-module T , we have an exact se-
quence

0 / H 0 (GF,S , T ) / Lv∈S H 0 (GF , T ) / H 2 (GF,S , T ∨ (1))∨

v

/ H 1 (GF,S , T ) / Lv∈S H 1 (GF , T ) / H 1 (GF,S , T ∨ (1))∨

v

/ H 2 (GF,S , T ) / Lv∈S H 2 (GF , T ) / H 0 (GF,S , T ∨ (1))∨ / 0.

v

P ROOF. We first define the maps in question. The maps

Resv : H i (GF,S , T ) → H i (GFv , T )

 A. DUALITY IN GALOIS COHOMOLOGY 175

are the compositions of the restriction maps from GF,S to a decomposition group above v ∈ S with
inflation to the absolute Galois group GFv . The maps

H i (GFv , T ) → H 2−i (GF,S , T ∨ (1))∨

are the compositions of the maps

H i (GFv , T ) → H 2−i (GFv , T ∨ (1))∨

of Tate duality with the Pontryagin duals of the maps Resv with for the module T ∨ (1). Finally the
maps
H 3−i (GF,S , T ∨ (1))∨ → H 1 (GF,S , T )
are defined to be the natural maps that factor through the Poitou-Tate isomorphisms

X3−i (GF,S , T ∨ (1))∨ → Xi (GF,S , T ).

We briefly sketch the proof of exactness. Exactness at the first and last stages follows from injectiv-
ity of restriction on zeroth cohomology groups. Exactness at the local stages follows from global class
field theory, which tells us that the image of H i (GF,S , T ) is the orthogonal complement of the image of
H 2−i (GF,S , T ∨ (1)) under the sum of local cup products. (We omit the argument, but see [NSW, Section
8.6].) Finally, exactness at the other four global stages follows directly from Poitou-Tate duality. 

R EMARK A.0.9. As with Tate duality, we have Poitou-Tate duality and the Poitou-Tate sequence
more generally for compact Z p -modules T with continuous GF,S -actions.
 Bibliography

[HS] Y. Hachimori and R. Sharifi, On the failure of pseudo-nulllity of Iwasawa modules, J. Alg. Geom. 14 (2005),
567–591.
[NSW] J. Neukirch, A. Schmidt and K. Wingberg, Cohomology of Number Fields, Second Edition, Grundlehren der
mathematischen Wissenschaften 323, Springer, 2008.

177