arXivarXiv:0905.1274v3 [math.NT] 15 Sep 2009 Par la mère apprenant que son fils est guéri, par...

arX

iv:0

905.

1274

v3 [

mat

h.N

T]

15

Sep

2009

Par la mere apprenant que son fils est gueri,par l’oiseau rappelant l’oiseau tombe du nid,par l’herbe qui a soif et recueille l’ondee,par le baiser perdu par l’amour redonne,et par le mendiant retrouvant sa monnaie:

Je vous salue, Marie1

To Theres and Seraina

THE T AND T ∗ COMPONENTS OF Λ - MODULES AND

LEOPOLDT’S CONJECTURE

PREDA MIHAILESCU

Abstract. The conjecture of Leopoldt states that the p - adic regula-tor of a number field does not vanish. It was proved for the abelian casein 1967 by Brumer, using Baker theory. A conjecture, due to Gross andKuz’min will be shown here to be in a deeper sense a dual of Leopoldt’sconjecture with respect to the Iwasawa involution. We prove both con-jectures for arbitrary number fields K. The main ingredients of the proofare the Leopoldt reflection, the structure of quasi - cyclic Zp[Gal(K/Q)]- modules of some of the most important Λ[Gal(K/Q)] - modules oc-curring (T acts on them like a constant in Zp), and the Iwasawa skewsymmetric pairing.

Contents

1. Introduction 21.1. General notations 41.2. Connection to Iwasawa theory 71.3. Structure 82. Groups, modules and fields 92.1. Idempotents, supports, annihilators and components in Zp[G] 92.2. Completions and local units 112.3. The growth of Λ - modules 122.4. Radicals and pairings 162.5. Reflection and duality 183. Classes as radicals and first applications 223.1. The symbols ΩE,n[a

1/p∞ ],ΩE [A1/p∞ ] and their radicals 23

3.2. Embedding adjoints 26

1Francis James: Priere. Music by Georges BrassensDate: Version 3.0 November 1, 2018.Key words and phrases. 11R23 Iwasawa Theory, 11R27 Units.

1

http://arxiv.org/abs/0905.1274v3

2 PREDA MIHAILESCU

3.3. The principal self-dual module of Iwasawa theory 283.4. Complex conjugation, explicite radicals and estimation of ranks 294. The filtration of M/K∞ and the Conjecture of Gross-Kuz’min 31

5. The filtration of ΩT ∗ , ΩT ∗ and the proof of the Main Theorem 336. Consequences 377. Appendix A: Basics 397.1. Annihilators, supports and components in the non-commutative

case 397.2. Arithmetic in Λ 447.3. Units 467.4. Λ - modules, radicals and duality 478. Appendix B: Facts and proofs for Section 3 548.1. The Iwasawa skew symmetric pairing 548.2. Complex multiplication and duality 579. Appendix C: Complements to sections 4 and 5 59References 62

1. Introduction

Let K/Q be a finite galois extension with group G. Dirichlet’s unit theo-rem states that, up to torsion made up by the roots of unity W (K) ⊂ K×,the units E = O(K)× are a free Z - module of Z - rank r1 + r2 − 1. Asusual, r1 and r2 are the numbers of real, resp. pairs of complex conju-gate embeddings K → C. Let p be a rational prime. We consider the setP = ℘ ⊂ O(K) : (p) ⊂ ℘ of distinct prime ideals above p and let

Kp = Kp(K) =∏

℘∈P

K℘ = K⊗Q Qp

be the product of all completions of K at primes above p. Let ι : K → Kpbe the diagonal embedding. We write ι℘(x) for the projection of ι(x) in thecompletion at ℘ ∈ P . If y ∈ Kp, then ι℘(y) is simply the component of y inK℘. If U ⊂ K×

p is the group of units, thus the product of local units at thesame completions, then E embeds diagonally via ι : E → U . Furthermoreone can use ι for inducing a galois structure on Kp (see §2.1).

Let E = ι(E) ⊂ U be the closure of ι(E); this is a Zp - module with

Zp-rk(E) ≤ Z-rk(E) = r1 + r2 − 1. The difference

D(K) = (Z-rk(E))− (Zp-rk(E))

is called the Leopoldt defect. The defect is positive if relations between theunits arise in the local closure, which are not present in the global case.Equivalently, if the p - adic regulator of K vanishes.

Leopoldt suggested in [21] that D(K) = 0 for all number fields K. Thisconjecture of Leopoldt was proved for abelian extensions by Brumer [7] in1967, using a result of Ax [5] and a local version of Baker’s linear forms in

LEOPOLDT’S CONJECTURE 3

logarithms [6]. It is still open for arbitrary non abelian extensions. Since1967 various attempts have been made for extending the results of [7] tonon abelian extensions, using class field theory, Diophantine approximationor both. The following very succinct list is intended to give an overviewof various approaches rather than a extensive list of results on Leopoldt’sconjecture. In [12], Greenberg notes for the first time a relation betweenthe Leopoldt Conjecture and a special case of the Greenberg Conjecture: heshows that Leopoldt’s Conjecture implies that B (see §1.1. for the defini-tions) is finite for totally real fields, i.e. the Greenberg Conjecture for holdsfor the T - part. The Conjectures of Leopoldt and Gross (see below) areequivalent to Greenberg’s Conjecture for the T and T ∗ - parts of A: see also§6 for more details on the connections between the Greenberg Conjectureand those treated in this paper.

The works of Federer [9] and and Federer and Gross [10] use Iwasawatheory and introduce the related Conjecture known as Gross’s Conjecture,which will be proved in Theorem 4 and again, together with the Proposition5. In an earlier paper, Kuz’min [18] had stated the ’Hypothesis 3’ assertingthat A′/((A′)T ) is finite. It is this version of the Gross Conjecture1, that weshall prove; since it is identical to Hypothesis 3, we shall speak of the Gross-Kuz’min Conjecture. Jaulent relates the Leopoldt and Gross Conjectures inthe wider context of a Conjecture which carries his name [16]; the Conjectureof Jaulent is beyond the scope of this paper. Emsalem, Kissilewski and Wales[8] use group representations and Baker theory for proving the Conjecture forsome small non abelian groups; this direction of research has been continuedby Emsalem or Emsalem and coauthors, in some further papers. Jaulentproves in [15] the Conjecture for some fields of small discriminants, using thephantom field Φ which we shall define below. The strongest result based onDiophantine approximation was achieved by Waldschmidt [24], who provedthat if r is the Z - rank of the units in the field K, then the Leopoldt defectverifies D(K) ≤ r/2.

It is easy to show that if K′/Q is a field such that Leopoldt’s Conjectureholds for some galois extension K/Q which contains K′, then it holds forK′. See for instance [19], the final remark on p. 108. Likewise, if theGross - Kuz’min Conjecture holds for K, then it holds for K′. We may thusconcentrate on galois extensions of Q and shall assume in the rest of thispaper that K/Q is galois and contains the p-th roots of unity; in particularK is complex. The Dirichlet number is r = r2 − 1 and the p - adic rankof E is rp = r − D(K). Furthermore, we assume that K is such that allthe primes above p are completely ramified in the Zp - cyclotomic extensionK∞/K and the Leopoldt defect is constant for all intermediate fields of thisextension. This can be assumed by choosing K sufficiently large, accordingto [25], Lemma 13.30; although the field F is assumed in the whole section

1See also, the ’generalized Gross Conjecture’, (iii) in [20], p. 854

4 PREDA MIHAILESCU

of Washington’s book to be totally real, the proof of 13.30 does not dependon this assumption, but see also Appendix A, proof of Lemma 13.30.

The main purpose of the paper is to prove:

Theorem 1. The Conjectures of Leopoldt and Gross - Kuz’min hold for allnumber fields K/Q.

The philosophy of this paper is reflected in the following targets:

A. Avoid µ - parts and pseudo-isomorphisms as much as possible, byconsidering adequate fields and their galois groups.

B. If X is a Λ - module, find explicite embeddings of abstract mod-ules like the dual X• and the adjoint α(X) and the double adjointα(α(X)), whenever possible in X itself. For this, we construct ex-plicite radicals for finite and infinite Kummer extensions and developinfinite Kummer extensions together with their radicals as projec-tive modules, from scratch. This is done in §2.4 and yields also asolution to the embedding problem.

C. Give a precise description of intermediate extensions of L/K∞ forarbitrary extensions L such that Gal(L/K∞) is a Λ - module. Thisis done in §2.3

D. Assume that the Leopoldt Conjecture is false; there is then a se-quence of fields, which emerge naturally from conditions imposed byreflection and class field theory, and which all have Zp - free groupsof rank D(K). These shall be called phantom fields. The neces-sary properties resulting from Iwasawa and class field theory shallbe followed from one consequence to another, eventually leading to acontradiction. This contradiction does indeed arise and it is caughtin Proposition 5.

In view of this program, some of the theoretical constructs go beyond theimmediate requirements for the proof of the Conjectures under investigation.Rather than introducing artificial reduction of generality, we chose to deferthe proof of technical and elementary facts to the Appendices which growaccordingly.

1.1. General notations. Let K∞/K be the cyclotomic Zp - extension of Kand Kn the intermediate fields of level n. The ground field is K, a complexgalois extension which contains the pk+1-th but not the pk+2-th roots ofunity. The constant k will be fixed in Definition 3 below, such that the Λ -modules related to K have some useful additional properties; we shall denotethen by κ the particular value chosen. We write K = K0 = K1 = K2 = . . . =Kk. As usual, we let τ be a topological generator of Γ = Gal(K∞/K) andT = τ − 1, Λ = Zp[[T ]]. We shall assume that the Leopoldt defect D(Kn) isconstant for all n ≥ k.

For all n ≥ 0 we let An be the p - Sylow subgroups of the class groupC(Kn) and A t he projective limit, a Λ - module. The groups A′

n, A′ are

defined as An, A, with respect to the class groups of p - integers: see also


[14], §4.3; the numeration respects the same rule as for the intermediatefields. The norms Nm,n = NKm/Kn for m > n ≥ k are supposed to besurjective as maps Am → An and A′

m → A′n. The product of all Zp -

extensions of K is M and ∆ = Gal(M/K); thus K∞ ⊂ M. We let furtherH,Ω be the maximal p - abelian extensions of K∞, which are unramified,respectively p - ramified. Note that we do not use the index ∞ for H andwill write instead H(K) = H0 for the Hilbert class field of K. For some fieldK we write E(K), E′(K) for the units respectively the p - units of K. Weshall consider the following additional subfields of Ω:

ΩE =⋃

n≥0

Kn

[E(Kn)

1/pn+1], ΩE′ =

⋃

n≥0

Kn

[E′(Kn)

1/pn+1],

so K∞ ⊂ ΩE ⊂ ΩE′ ⊂ Ω.If X is some infinite abelian group, we write X for its Zp - torsion

(which may include some torsion Λ - module of infinite p - rank) and forΩ ⊇ F ⊃ K∞, some infinite extension, we shall write

F = XGal(F/K∞)

for the fixed field of the Zp - torsion of galois group of this field. Thus

Gal(F/K∞) is a free Zp - module, possibly of infinite rank. This constructionalso cancels eventual µ - parts in our modules, so we need no assumptionson the vanishing of the µ - part of A. In general H 6= H and Ω = Ω iffµ(K) = 0, while ΩE = ΩE and ΩE′ = ΩE′ . We may thus use both simpleand barred notations for these fields.

The general notations from Iwasawa theory which we use here are:

p A rational prime,X The Zp - torsion of the abelian group X ,ζpn Primitive pn-th roots of unity with ζppn = ζpn−1 for all n > 0.,

µpn ζkpn , k ∈ N,K A galois extension of Q containing the p-th roots of unityK∞,Kn ∈ The cyclotomic Z− p - extension of K, resp. its n-th intermediate field,G Gal(K/Q),s The number of primes above p in K,Γ Gal(K∞/K) = Zpτ, τ a topological generator of ΓT τ − 1,∗ Iwasawa’s involution on Λ induced by T ∗ = (pk+1 − T )/(T + 1),Λ Zp[[T ]], Λn = Λ/(ωnΛ),

ωn (T + 1)pn−(k+1) − 1, (K×

n )ωn = 1,

Nm,n NKm/Kn = NKm/Kn ; Nn = NKn/K,K∞ ∪n≥0Kn : The cyclotomic Zp - extension of K,

6 PREDA MIHAILESCU

An = A(Kn) The p - part of the ideal class group of Kn,

A lim←−An,A′n = A′(Kn) The p - part of the ideal class group of the p - integers of Kn,

A′ lim←−A′n,

D(K) The Leopoldt defect of the field K ,B b = (bn)n∈N ∈ A : The classes bn contain products of ramified primes,Ω The maximal p - abelian p - ramified extension of K∞,

ΩE ∪∞n=0Kn[E(Kn)1/pn+1

] = K∞[E1/p∞ ],

ΩE′ ∪∞n=0Kn[E′(Kn)

1/pn+1] = K∞[E′1/p

∞

],M The product of all Zp extensions of K ,H The maximal p - abelian unramified extension of K∞.

Some galois groups repeatedly used, are

X = Gal(H/K∞), X = Gal(H/K∞)Y = Gal(Ω/K∞), Y = Gal(Ω/K∞)

H = Gal(Ω/H), H = Gal(Ω/H)Z = Gal(Ω/ΩE′), Z = Gal(Ω/ΩE′)YE = Gal(ΩE/K∞), YE′ = Gal(ΩE′/K∞)

(1)

At finite levels levels, we mostly write Fn/Kn for the maximal Kummersubextension of exponent pn+1 included in F. One exception may occurbecause of the ambiguous notation Ωn: it may be the maximal Kummersubextension of Ω over Kn but it may also be the maximal p - abelian p- ramified extension of Kn. We choose the first and denote the second,consistently, by Mn. The previous groups will then be indexed like in

Xn = Gal(Hn/Kn), YE,n = Gal(ΩE,n/K∞), etc.

If F (T ) is the characteristic polynomial of A and f |F is a distinguishedpolynomial, we let Ωf ⊂ Ω be the maximal subfield with galois group Yf =

Gal(Ωf/K∞) annihilated by f and free as a Zp - module. Thus Ωf , Ωfare subfields of Ω corresponding to some factors of the group Gal(Ω/K∞)of essential Λ - rank r2, much like finite subextensions of a Zp - extensioncorrespond to factors by the compact subgroups pn · Zp ⊂ Zp. The field

Ωf ⊂ Ω is defined like Ωf , with respect to H: the maximal subfield of Ω

with group Hf = Gal(Ωf/H) annihilated by f . Note that Gal(Ω/Ω) is Zp- torsion of infinite rank, so it is not annihilated by f , thus Ωf ⊂ Ω. The

fields Hf = Ωf ∩H, by definition. These definitions are given with proofs ofexistence in §3.

We assume that the primes above p are completely ramified in K∞/Kand ℘ ⊂ K is one such prime. We let π ∈ K be fixed by the decompositiongroup D℘ ⊂ G of ℘ and such that (π) = ℘ord(℘). With this, we fix Π =πσ : σ ∈ G/D℘ ⊂ K, thus s = |Π| is the number of primes above p in K,

like above; the choice of Π is not canonical, but ΩE[Π1/p∞ ] = ΩE′ , so Π is


a notational simplification. Here is a complementary list of notations whichwe introduce:

P σ℘ : σ ∈ G, and ℘ a prime of K above p,D℘ ⊂ G The decomposition group of ℘,C Coset representatives for G/D℘,

Π σπ : σ ∈ C, π ∈ K, (π) = ℘ord(℘)K(K) K⊗Q Qp, for global fields Kι℘ The projection from K(K) in the completion K℘,U(K) The units of K(K),E(K), E′(K) The units and p - units of some global field K,

U (1)(K) The one - units in U , Un = U (1)(Kn),U ′(K) The one units of absolute norm 1 in K(K), up to torsion,E(K) The completion of E(K) → U(K),

H Hϕ(A),HE H ∩ ΩE,

ΩE1 ∪∞n=0Kn[E(K)1/pn+1

] = K∞[E(K)1/p∞

] ⊂ ΩE,Ωf ∪L⊂Ω,(Gal(L/K∞)f=1 L,

Ωf ∪L⊂Ω,(Gal(L/H)f=1 L,

Yf Gal(Ωf/K∞), Yf = Gal(Ωf/H),Φ ΩE1 ∩H∞,Hf The maximal subfield of H with group fixed by f,Mn The product of all Zp extensions of Kn , M = M(K)∆n Gal(Mn/Kn), ∆n = Gal(Mn/K∞)ME ΩE ∩M.

The symbols K∞[E1/p∞ ],K∞[E′1/p∞

] are well defined, however symbols like

ΩE[A1/p∞ ] and even more so, K∞[A1/p∞ ] have lead to some ambiguity. The

use of radicals of classes will be developed in much detail in the third chapter,where all such symbols are either rigorously defined or discarded; the lastholds for K∞[A1/p∞ ] which has a possible non canonic definition, but isnot of great help and can thus be avoided. We then may write things likeΩE[a

1/p∞ ],HE[a1/p∞ ], a ∈ A or ΩE′ [B1/p∞ ], B ⊂ A′, etc.

1.2. Connection to Iwasawa theory. We shall take here an approachusing class field and Iwasawa theory. Let T → T ∗ be the Iwasawa involu-tion (see (10) for a definition); the main fields of our interest will be ΩTand ΩT ∗ , ΩT ∗ . Note that in this notation, ΩT = M, since by definitionGal(ΩT /K∞) is also Zp torsion - free.

In general, for arbitrary polynomials f , one may prove by using the Iwa-sawa skew symmetric pairing that Gal(Ωf/ΩE′) ∼ Gal((H∩Ωf∗)/K∞)•; thisand more consequences of Iwasawa’s work are proved in Theorem 3 and laythe ground for the proof of the final result. Although the general structureof Ωf and Ωf is beyond the scope of this paper, this and similar facts willbe exposed for the sake of completeness.

8 PREDA MIHAILESCU

Assuming that Leopoldt’s Conjecture is false for K, we shall show thatΩT ∗∩H∞ = Φ ⊂ ΩE is a non trivial extension with group of Zp - rank D(K):its existence is thus equivalent to the failing of Leopoldt’s Conjecture2. Thisand two other extensions with the same property are denoted by phantom- fields, for obvious reason: they encrypt a constant which should be zero.The Zp - ranks of the groups Yf can be calculated exactly by means of classfield theory and it turns out that Zp-rk(YT ∗) = r2 + s− 1, where s denoteslike usual the number of primes above p in K. We then use the tower (25)as a filter for the extension ΩT ∗/H: the sum of the ranks of the differentintersections must add up to the total of r2+s−1. Some direct investigationsof ranks show the following equalities:

Zp-rk(Gal((ΩT ∗ ∩ ΩE′)/K∞)

)= r2 + s− 1,

Zp-rk(Gal((ΩT ∗ ∩ ΩE′)/H∞)) = r2 + s− 1−D(K),

Zp-rk(Gal((ΩT ∗ · ΩE′)/ΩE′) = D(K).

(2)

The last equation in (2) is a consequence of the first two, based on theLemma 14. The proof of the second identity in (2) is given in Section 5and is the crucial step in proving the main result of this paper. The centralobservation of the proof is based on the remark that the field Φ is one withKummer radical annihilated by T and group annihilated by T ∗. This has asconsequence that a tower of extensions of the type Kn ⊂ Ln ⊂ L′

n, in which

Ln ⊂ Hf and L′n ⊂ Ωf , while L′

n/Kn is abelian, cannot exist. These towersare possible for f 6= T ∗, but the radicals of L′

n ·K2n come from K2n in thatcase, as shown in Example 2 in Appendix C. For f = T ∗, they must comefrom K, which leads to a contradiction and proves the Theorem 1.

1.3. Structure. The development of the material in the paper is rathersmooth and we introduce the basic concepts in more detail then directlynecessary; in part, proofs which are not of direct importance, are given inthe Appendices.

Since the Conjecture is related to T and T ∗ - parts, the idea is to follow allpossible extensions over K∞ or H with galois groups or radicals annihilatedby T and bring their relations and ranks in evidence. It becomes a pattern,that in all cases some phantom field with galois group of rank D(K) willemerge. In Chapter 2 and the related parts of Appendix A we treat thestructure of Zp[G] - modules and their isomorphy classes in the case when Gis non commutative, and develop the theory of local and global Minkowskiunits. In §2.3 we consider in detail the growth of Λ - modules, obtaining somegeneral results on Weierstrass modules – i.e. Λ - torsion modules of finiterank and Zp - torsion - free. These are then used as a base for presenting aprojective structure on radicals and defining the Leopoldt reflection for Λ[G]- modules. The action ofG on our modules is always taken into considerationin addition to the one of Λ.

2The field Φ was often noticed in the literature: for example in Jaulent’s recent paper[15], treating a special case of Leopoldt’s Conjecture


The Chapter 3 treats the extension Ω/ΩE and we define in this contextradicals as classes; this allows embedding adjoints and duals of submodulesof A in A itself and opens the way for the deep results on duality basedon the Iwasawa skew symmetric pairing, which are presented in §3.3. Thesection 3.4 gives the rank estimates for (2) and some new explicite radicalconstructions; these bring a new perspective on duality in the non CM case,when it appears as a useful counterpart for the lacking complex conjugation.In Chapter 4 we prove the Gross - Kuz’min Conjecture, using an approachwhich relays strongly on duality; the CM case is however quite simple. InChapter 5 we then prove the Proposition 5, from which the Main Theoremfollows. The final Chapter treats some consequences, like the fact that theproven conjectures are the special cases of the Greenberg Conjecture, forthe T and T ∗ parts of A′.

2. Groups, modules and fields

Most of the modules that we shall encounter are Zp[G] - modules, withG = Gal(K/Q) or G = Gn = Gal(Kn/Q). The first section of this Chapterwill treat the structure of such modules. Next we shall recall some propertiesof the completions of a field at primes above p, and consider local and globalMinkowski units in some detail.

Since most groups we encounter are finite or infinite p - groups, we definehere some notions that we repeatedly use:

Definition 1 (Group-related constants). Let X be an abelian p - groupwritten additively. If X is a Zp - torsion module, the exponent exp(X) =minpj : pjX = 0, and pi is an exponent for X iff i ≥ exp(X). Supposethat X has exponent pn and p - rank r. We say that X has sub-exponentpm ≤ pn for some m > 0, if there is a subgroup Y ⊂ X with p-rk(Y ) =p-rk(X) = r and for all y ∈ Y \ pY we have pm−1y 6= 0.

The quotient X/pX is an Fp - vector space with respect to which the p -rank is defined by p-rk(X) = dim(X/pX). If X is finite, then p-rk(X) =dim(X/pX) = dim(S(X)), where S(X) is the first socle of X. In general,the m-th socle is Sm(X) = Ker (pm : X → X). If X is infinite, theessential p - rank is ess. p-rk(X) = limm→∞ p-rk(pmX).

Let X be a Λ - module. By Nakayama’s Lemma, the number of elementsin a minimal set of generators xi ∈ X \ XM of X is the p - rank of Xω1

and does not depend on the choice of the xi. We denote this number by Λ -rank of X, so

Λ− rank(X) = p-rk(Xω1).

2.1. Idempotents, supports, annihilators and components in Zp[G].In the context of Leopoldt’s Conjecture we are interested in ranks and notin torsion of modules over rings. It is thus a useful simplification to tensorthese modules with fields, so we introduce the following

10 PREDA MIHAILESCU

Definition 2. Let G be a finite group and A,B a Z -, respectively a Zp -module, which are torsion free. Let a ∈ A, b ∈ B. We denote

A = A⊗Z Q, a = a⊗ 1, B = B ⊗Zp Qp, b = a⊗ 1.

We note that Z-rk(A) = Q-rk(A) and Zp-rk(B) = Qp-rk(B).

The major advantage of this notation is that it will allow to consider, ina first step, annihilator modules as images of some idempotents in Qp[G]. IfX is some Zp[G] - module, we write

X⊤ = y ∈ Qp[G] : xy = 1,∀x ∈ X,

the annihilator module of X. The annihilator module of X will then beX⊤ = X⊤ ∩ Zp[G].

From class field theory, one has ([22], Chapter 5, Theorem 5.1):

Gal(M/H(K)) ∼= p - part of U (1)(K)/E(K).(3)

and the global Artin symbol extends to a covariant Qp[G] - isomorphism

ϕ : ˜U (1)(K)/E → ∆.

Alternatively, we may consider ϕ as a surjective Qp[G] - homomorphism

ϕ : U (1)(K) → ∆ with kernel E. The Artin map is an isomorphism ϕ :An → Gal(Hn/Kn) which extends in the projective limit to an isomorphismof Λ - modules, ϕ : A→ Gal(H/K∞).

It is known that there is a Minkowski unit δ ∈ E ([25], lemma 5.27), i.e.a unit such that

Z-rk(δZ[G]

)= r.

We shall use multiplicative notation, so all actions are from the right. IfA ⊂ Qp[G] is some module, then there is an idempotent α ∈ Qp[G] suchthat A = (α) = αQp[G]. This follows from the proof of Maschke’s Theorem[3], p. 116. The annihilator ideal of A is (1 − α)Qp[G] and conversely, A isthe annihilator of (1−α): in particular α · (1−α) = (1−α)α = 0. This is arephrasing of Maschke’s theorem, which makes explicit use of idempotents:(1−α)Qp[G] is a complement of A. Idempotents a ∈ Qp[G], can be regardedas linear maps of the Qp - vector space Qp[G] and as such we may define

rank(a) = dim(aQp[G]) = |G| − rank(1− a),(4)

We develop in the first subsection of the Appendix A the notions of sup-port and annihilator for non commutative group rings. This is important forthe understanding of extensions which are not CM, and will be developedalso in subsequent papers for a more in depth study of duality in numberfields.


2.2. Completions and local units. If K is a finite or infinite numberfield, we let K = K⊗Q Qp and may also write K(K) when the global fieldsneed precisation. The following is proved here for completeness. See alsoJaulent [16] §1.a (ii) for an alternative treatment of this construction.

Fact 1. Let K = Q[x], p, P and Kp be like in the introduction, suppose thatf ∈ Z[X] is a minimal polynomial of x and ι : Q → Qp is the naturalembedding. Then

Kp = Qp[X]/(ι(f))

is a galois algebra with group G = Gal(K/Q) and the embedding ι extends toan embedding K → Kp which commutes with the galois action. The imageι(K) ⊂ Kp is dense in the product topology.

The proof is direct and is given in Appendix A. The group under consid-eration is thus the multiplicative subgroup of ideles which are trivial at allplaces except for the primes above p. By the Chinese Remainder Theoremwe identify u ∈ U with (ι℘(u))℘∈P .

For arbitrary fields K, the units U (1)(K) are the products of U (1)(K℘) =u ∈ U : u ≡ 1 mod π for some uniformizor π of the completion K℘. For

K = Kn we simply write Un = U (1)(Kn). If K/Qp is a finite local extension,we also write Kn = K[µpn+1 ] and

U ′n = u ∈ U (1)(Kn) : NKn/K(u) = 1.

Then

Lemma 1. Let K/Qp be a local finite galois extension with K∩Qp[µp∞ ] =Qp[µpk+1 ]. Then the system (U ′

n)n∈N defined above is norm coherent and thenorm is surjective at all levels, that is

NKm/Kn(U ′

m) = U ′n, ∀m > n > 0.

Proof. This follows from class field theory: Gal(K/Qp) acts by conjugationon the groups Γn,m = Gal(Kn/Km), which are fixed under this action. Byclass field theory,

Γn,m ∼= K×m/NKn/Km

(K×n ).(5)

But Kn/Km is totally ramified, so we have

K×m/NKn/Km

(K×n ) = U (1)

m /NKn/Km(U (1)

n ),

and (5) implies that the norm residue group is Gal(K/Qp) - invariant, andit thus is a quotient of Zp. Since N(U ′

n) = 1 by definition, it follows thatthe restriction of the norm to U ′

n is indeed surjective.

We now show that there are local Minkowski units and describe theirrelation to the global ones. Serre proves in [23], §1.4, Proposition 3, in the

case when K/Qp is a local field, that the group U (1)(K) contains a cyclic

Zp[G] module of finite index, which is isomorphic to Zp[G]: thus U (1)(K) isquasi - cyclic.

12 PREDA MIHAILESCU

Using this result one obtains units generating Zp[G] - modules of finiteindex in U , which we call local Minkowski units for K. Let ℘ ∈ P be fixedand υ ∈ K℘ be a local Minkowski unit, according to Serre. Then we defineξ = ξ(υ) ∈ U and ρ ∈ U by:

ιτ℘(ξ) =

υ for τ = 1,

1 for τ ∈ G, τ 6= 1.(6)

ιτ℘(ρ) =

1 for τ = 1,

0 for τ ∈ G, τ 6= 1.(7)

Let D℘ be the decomposition group of ℘ and C = D℘\G be coset repre-sentatives. Then C acts on ξ and for σ ∈ C, the unit ξσ satisfies:

ιτ℘(ξ) =

υ for τ = σ,

1 for τ ∈ G, τ 6= σ.

We denote units the generators u ∈ U for U by local Minkowski units. Theprevious construction shows that such units exist. Since NKn/K(En) = 1,local units of norm one are interesting for the embedding En → Un; wedefine:

U ′ = u ∈ U (1) : NKp/Qp(u) = 1(8)

which is a quasi - cyclic Zp[G] submodule of U with U (1)/U ′ = U (1)(Zp) ∼=Zp. Therefore U ′ ∼= (1−NK/Q/|G|)Qp[G], the last being a two sided module

in Qp[G]. For any K we have E(K) ⊂ U ′ and therefore U (1)(Zp) is mappedinjectively in ∆ by the Artin map. Let ξ0 ∈ U ′

0 \ U ′p0 be a local Minkowski

unit for U ′0. From Lemma 1 we conclude that there is a norm coherent

system (ξn)n∈N, ξn ∈ U ′n in which ξn are local Minkowski units for U ′

n. Thequestion about norm coherent systems of global Minkowski units is relatedto the capitulation of the primes above p, which makes its investigation moreinvolved. We will not address it here.

By choosing a global Minkowski unit δ ∈ E, one can find a local oneξ ∈ U ′ such that

ξα = δ, with α2 = α ∈ Qp[G].(9)

This corresponds to a map φE : U ′ → ι(E), according to the general theoryon isomorphy classes of idempotents in Appendix A.

2.3. The growth of Λ - modules. Let Cn = Q[ζpn+1 ], n ≥ 0 be the

pn+1-th cyclotomic extension, Cn = Gal(Cn/Q). We identify C0 with itslift to Gal(Cn/Q) which generates a group of order p − 1 and let Bn =CC0n , the intermediate fields of the Zp - cyclotomic extension B/Q. Let

Γ = Gal(B/Q) ∼= Gal(C∞/C0) and τ ∈ Γ be a topological generator, actingby restriction on Cn and Bn.


We assumed thatK is a galois extension of Q with groupG, which containsthe Ck but not Ck+1. Let

K−1 =⋂

L⊂K : K=Ck·L

L ⊂ K

be the smallest galois extension which generates K by compositum with Ck

and G−1 = Gal(K−1/Q). Then Gal(Kn/Q) = Cn ⋉ G−1 and in particularG = Ck ⋉G−1. If K−1 and C1 are linearly disjoint, the product is a directone. We write Kn = K · Bn for n ≥ k and K∞ = B∞ · K, while Kj = K for0 ≤ j ≤ k.

Since Ck ⊂ K, the group Gal(K∞/K) ∼= Γpk. Our ground field will

be K, so we let Γ = Γpkand τ = τp

kbe a topological generator; Γn =

Gal(Bn/Bk) ∼= Γ/Γpk. We further assume that all the primes above p ramify

completely in K∞/K and the Leopoldt defect is constant in Kn, n ≥ k; sincethe Leopoldt defect is bounded (see [25], Lemma 13.30), this can be achievedby choosing k sufficiently large.

Like usual we let T = τ − 1 and Λ = Zp[[T ]] ∼= Zp[Γ]. For n = k + l > k

we let ωn = (T + 1)pl − 1 = γp

n+1 − 1 and Λn = Zp[Γn] ∼= Λ/(Λωn). Thenorms Nm,n = NKm,Kn = ωm/ωn for m > n ≥ 0 are sometimes denoted by

νn,m, but we use ν in connection with the group ∆ here. The notations T , Λare self-explained and may not be necessary. Let the cyclotomic characteract on Λ by k(τ) = (q + 1)τ , where q = pk+1. The Iwasawa involution isdefined by

T = τ − 1 7→ T ∗ =q − TT + 1

.(10)

Let (Mn)n∈N be a family of abelian p - groups, finite or infinite, whichare closed under the action of Γn and which form a projective system undera family of maps fm,n : Mm → Mn,m > n ≥ k, which shall mostly be thenorms Nm,n = NKm/Kn . All the Λ - modules we encounter are projectivelimits M = lim←−Mn of such systems. The structure of Λ - modules is well

known (e.g. [14] §1, [25], Chapter 7). Some of the most important groupswe shall encounter are (En)n∈N, (E

′n)n∈N, units and p - units of K; possible

projective systems on these units will be considered below. The groupsA = (An)n∈N, A

′ = (A′n)n∈N with An, A

′n the p - parts of the class groups of

the integers, resp. p - integers of Kn: these form classical projective systemswith limits A,A′.

If X is a Λ - module, it is customary to write X• for the module endowedwith action of Λ twisted by the Iwasawa involution, [14], §8.1. The notationX• will receive below a more general definition, which takes the action of Gand embedding in X into account. We may use the short - lived notationXI for the Iwasawa dual.

Taking the action of Gn = Gal(Kn/Q) under consideration, there is aninvolution of Zp[Gn] - modules, the Leopoldt reflection involution, definedas follows: let α =

∑σ∈G aσσ ∈ Zp[Gn], with aσ ∈ Λ. Then the Leopoldt

14 PREDA MIHAILESCU

reflection involution is an automorphism of Zp[Gn] defined by

α 7→ α′ =∑

σ∈Gn

aσ · χ(σ) · σ−1,(11)

with χ the Teichmuller (cyclotomic) character: ζχ(σ)pn+1 = σ(ζpn+1).

The following lemma generalizes in its first two points an observation ofFukuda [11]; it gives an overview of the asymptotic growth of torsion Λ -modules:

Proposition 1. Let the notations be like above, and k = 0 for ease ofnotation3. Let M = lim←−Mn be a torsion - Λ - module such that the mapsfm,n are all surjective for m > n ≥ 0. Then there is an n0 > 0 such that forall n ≥ n0 the following hold:

1. If |Mn| = |Mn+1|, then Mm =Mn for all m > n ≥ 0.2. If p-rk(Mn) = p-rk(Mn+1) then p-rk(Mm) = p-rk(Mn) = R for allm > n ≥ 0; furthermore there is a constant λ(M) ≤ R such that|Mn+1| − |Mn| = pλ for all n sufficiently large.

3. Suppose that the maps fm,n are the relative norms Nn+1,n, which areall surjective and let ι :Mn →Mn+1 be the natural embedding. Then

xp = ι(Nn+1,n(x)), ι(Mn) = pMn+1, for a all x ∈Mn+1.(12)

Furthermore, if M is like in point 2., and R = λ(M) then the mapι is injective.

The proof of this proposition is given in Appendix A. A typical applicationis the case when M ⊂ A is some Λ - submodule. The first two points areunsurprising, but the third gives useful additional information about thestationary growth of Λ - modules of finite rank. For such modules M , theZp - torsion is a canonical finite submodule, but the infinite part is primarilyjust a factor. We shall denote Λ - modules of finite p - rank and Zp - torsionfree by Weierstrass modules. Their structure is given by

Lemma 2. Let X be a finitely generated Weierstrass module written addi-

tively, with characteristic polynomial F and F =∏dj=1 fj be a decomposition

in pairwise coprime distinguished polynomial. Then Xj = XF/fj are canonicsubmodules with

X =d⊕

j=1

Xj.

For every distinguished polynomial f |F there is a canonic f - component

Xf = XF/f ⊂ X and if f and F/f are coprime, then Xf is a canonic directterm in X.

For Weierstrass modules we have the following consequence of Proposi-tion 1:

3This is allowed, since we shall not make explicit use of the roots of unity in Kn


Corollary 1. Let M ⊂ A be a Weierstrass Λ - submodule. Then for n > n0the capitulation kernel Ker (ι :Mn →Mn+1) is trivial and

xp = ι(Nn+1,n(x)), ∀ x ∈Mn+1.(13)

Furthermore, there is an integer z(x) which is independent on n and suchthat vp(ord(xn)) = n+1+z(x) for all n sufficiently large. As a map z : A→Z, we have z(xy) ≤ max(z(x), z(y)) and if c = maxvp(ord(x0))− (k′ +1) :x0 ∈ A(K), with k′ such that K ∩Ck′+1 = Ck′, then z(A) ∈ Z≤c. Finally,the map z : A→ A can be defined by

z(x) =

−∞ if x ∈ A

limn→∞ vp(ord(xn))− (n+ 1) otherwise.(14)

The next lemma shows that pure µ - modules have no capitulation. Theproofs of the Lemma 2, Corollary 1 and Lemma 3 are in the Appendix A.

Lemma 3. Let a ⊂ A be such that Λa ⊂ A is a Λ - module of unbounded p- rank. Then for sufficiently large n the capitulation kernel Ker (ιn,n+1) =1 and

vp (|Λan|) = µpn − ν, ν ≥ 0.

We are now in position to choose our base field K such as to avoid someunpleasant phenomena at finite levels:

Fact 2. Let K be a galois extension of Q containing the p-th roots of unityand Kn = K · Bn. Then there is an index n0 such that for all n ≥ n0 thefollowing facts are true:

1. The Leopoldt defect D(Kn) is constant, all the primes above p ramifycompletely and the norm An → An0 is surjective.

2. The p - ranks of all submodules B ⊂ A of finite rank are stable:p-rk(Bn) = p-rk(Bn0). Furthermore, there is a constant µ such that

|An| = |A

n0| · (pµ(pn−pn0).

and the constant pn0 annihilates A.3. The capitulation kernels ker(Bn → Bn+1) are trivial for all Weier-

strass submodules B ⊂ A and for all cyclic Λ - modules of unboundedp - rank.

4. The function z(x) in Corollary 1 is bounded by n0 and

xp = ιn,n+1(Nn+1,n(x)) for all x ∈ An+1.

Furthermore, for x = (xn)n∈N ∈ A and n such that xn 6= 1, the valuez(xn) = vp(ord(xn))− (n+ 1) = z(x) is constant.

This follows from the Corollary and Proposition above. Since A is afinitely generated Λ - module, one may choose a minimal set of generatorsai; for each module Λai there is a value mi for which the conditions aboveare fulfilled; set n0 to be larger or equal to the maximal value for all i.

16 PREDA MIHAILESCU

Definition 3 ( The ground field K ). We let κ be equal to the least of alln0 in Fact 2. Furthermore, we assume from now on that our base field K isKκ and thus enjoys all the properties 1.-4. above; henceforth κ replaces theprovisorious value k used since the introduction.

We define additionally An = x ∈ An : z(an) ≤ 0 and A = x ∈ A :z(a) ≤ 0; since z(an) = z(a) for all n in the present definition of K, it

follows that (A)n = An for all n.

The following lemma describes two important shift maps derived from thebound z(A) ≤ κ.

Lemma 4. Let ς : A → A : a 7→ apκand the restrictions An → An : an 7→

apκ

n be also denoted by ς. Then ς annihilates A and ς(An) ⊂ An.For d > 0 we let A(d) be the set of sequences a = (an)n∈N with an ∈ An

and such that a(d) := (am)m≥d is norm coherent and there is a b ∈ A suchthat a(d) = b(d); the elements (an)n<d = 1. Then A(d) is a Λ - module andfor d := 2κ, there is a map ιK : A→ A(d) such that

A ∋ a = (an)n∈N 7→ b = (bm)m∈N, bm =

1 if m < 2κ,

ιm−κ,m(am−κ) otherwise.

The kernels are Ker (ιK) = A and Ker (ς) = A. Furthermore, ς(A) is aWeierstrass module and ιK(A) has no finite Zp - torsion.

Proof. The map ς annihilates A since pκ is by definition an exponentthereof. It is a straight forward verification that A(d) is a Λ - module and ιKis an homomorphism of Λ - modules. By choice of κ, all ideals in classes ofAn capitulate in An+κ, so A ⊂ Ker (ιK). On the other hand if a ∈ A \ A,then Λa has unbounded p - rank and by Lemma 3 and the choice of κ, themap ιn,n+κ is injective on Λa, so a 6∈ Ker (ιK). Since p

κ annihilates the Zp- torsion, Ker (ς) = A and ς(A) is Zp - torsion free, so it is a Weierstrassmodule. The fact that ιK(A) has no finite Zp - torsion follows from Lemma3 and point 1. of Proposition 1.

It will make sense to extend the map ς also on local and global units, sofor arbitrary abelian p - groups X we let

ς : x 7→ xpκ, ∀x ∈ X.

2.4. Radicals and pairings. Let L ⊃ K∞ be some infinite abelian exten-sion and Ln ⊂ Kn be the maximal Kummer subextension of degree pn+1

contained in L. The radical rad(Ln/Kn) is often considered as a group

K×n ⊂ rad(Ln/Kn) ⊂ (K×

n )pn+1

, such that Ln = Kn[rad(Ln/Kn)1/pn+1

]. Theinjectivity condition Ln ⊂ Ln+1 translates into

rad(Ln/Kn)p ⊂ rad(Ln+1/Kn+1).

The relevance of the notion of subexponent for such extensions is given by:


Lemma 5. Let X = Gal(L/K∞) be a torsion Λ - module of p - rank rand free as a Zp - module. Suppose that Bn ⊂ K×

n are groups such that

L′n := Kn[B

1/pn+1

n ] ⊂ Ln and L′n+1 ⊃ Ln for all sufficiently large n. If

for all such n, the p-rk(Bn) = r and the subexponents of Bn diverge, then⋃n L

′n ∼ L.

Proof. The injective limit L′ =⋃n L

′n exists, from the hypothesis. Since the

subexponents diverge, we have Zp-rk(Gal(L′/K∞)) = lim←− p-rk(Gal(L′n/Kn)) =

r. Therefore Gal(L′/K∞) ⊂ Gal(L/K∞) is a Zp - free submodule of the samerank. It follows that L = L′.

The radicals rad(Ln/Kn) are endowed with an obvious structure of injec-tive systems. Can a projective system be loaded upon these radicals, maybeusing some modified definition ? The answer is affirmative for a wide rangeof cases, but the notion of radical introduced above is not adapted for sucha structure. We may use the alternative definition

RAD(Ln/Kn) = K×n 〈rad(Ln/Kn)

1/pn+1〉 / K×n ,(15)

where K×n 〈rad(Ln/Kn)

1/pn+1〉 ⊂ Ln is the multiplicative group spanned by

(rad(Ln/Kn))1/pn+1

over K×n ; this definition is frequently used, for instance

in cogalois theory, e.g. Albu’s monograph [1] and the specialized paper [2].The injectivity condition is now RAD(Ln/Kn) ⊂ RAD(Ln+1/Kn+1) and ifGal(Ln/Kn) is a Λ - module, so are both radicals defined above. A third ver-

sion of radicals, which we shall not explicitly use is rad((Ln/Kn)/(K×n )

pn+1).

All three definitions describe the same extension; the first rad(Ln/Kn) isuseful as a subgroup of K×

n . But it is large; the last one is the ’active part’of it, it has the advantage of being isomorphic to Gal(Ln/Kn) as a group(but only via twisted action as a Λ - module), but it is a factor. The radicalRAD(Ln/Kn) is isomorphic to Gal(Ln/Kn) and is not a factor; furthermoreit has the advantage of yielding projective systems.

We let now L′/L be an extension in the tower K∞ ⊂ L ⊂ L′ ⊂ Ω, suchthat both L and L′ are galois over K and Gal(L′/L) is torsion Λ - moduleof finite p - rank and Zp - torsion - free. The radicals RAD(L′

n/Ln) havethe nice property of being endowed with a structure of projective systems,as stated in the next proposition. The proof will be given in the AppendixA, since the result is not used in the proof of Theorem 1.

Proposition 2. Let L′/L be an extension in the tower K∞ ⊂ L ⊂ L′ ⊂ Ω,such that both L and L′ are galois over K and Gal(L′/L) is a torsion Λ - mod-ule. Then the radicals RAD(L′

n/Ln) form a projective system with respect tothe relative norms. The projective limit is RAD(L′/L) = lim←−RAD(L′

n/Ln).

18 PREDA MIHAILESCU

For L,L′ like in the proposition and m > n > n0 we4 have the compatiblesystems of Kummer pairings:

Gal(L′m/Lm) × RAD(L′

m/Lm) → µpm+1

↓ ↓ ↓Gal(L′

n/Ln) × RAD(L′n/Ln) → µpm+1 ,

(16)

in which the arrows are induced by the norms Nm,n. Let T (µ) = lim←−µpn+1

be the Tate module for the group µ of all the pn-th roots of unity ([14],§10). The projective limits in (16) are then Gal(L′/L),RAD(L′/L), T (µ),respectively. In view of RAD(Ln/Kn) = K×

n 〈rad(Ln/Kn)1/pn+1〉 / K×

n , forKummer extensions Ln/Kn of exponent pn+1, we may consider the limit

RAD(L/K) as a Tate module for the quotients rad(Ln/Kn)/(K×n )

pn+1. This

is done by Iwasawa in [14] §9.3, for the case when Kn ⊃ ΩE,n, a case weshall discuss below.

Independently of the understing of RAD(L′/L) as a Tate module, there isthus a projective - projective Kummer pairing at infinity, given by the projec-tive limit of this diagram. We shall denote this pairing, in analogy with theIwasawa skew symmetric pairing5 also by [ , ] : Gal(L′/L)×RAD(L′/L)→T (µ). The pairing [ , ] enjoys all the properties of the Kummer pairings atfinite levels:

Theorem 2. The pairing [, ] : Gal(L′/L) × RAD(L′/L) → T (µ) definedabove is bilinear, non degenerate and for all (σ, b) ∈ Gal(L′/L)×RAD(L′/L)and g ∈ Gal(K∞/Q) we have

[σ, b]g = [σg, bg] .(17)

Proof. The proof follows from Proposition 2 and the compatibility in (16).The fact that the properties of Kummer pairing are conserved in the pro-jective limit can be verified for the three types of galois groups Gal(L′/L)considered in the proof of Proposition 2 below.

2.5. Reflection and duality. We may define Λ[G] as the Zp - group ringof Γ⋉G0 and

Λ[G] =

x =

∑

g∈G

ag ⋉ g, ag ∈ Λ

.

Although Λ and G interact in general, the development of x above is unique:if W =

∏wi=1 τ

eigi with ei > 0 and gi ∈ Gal(K/Cκ, then the word has aunique reduction to a product W1 = τ eg: indeed, there is an fw such that

4The existence of n0 is given in the proof of the Proposition. It is likely that in the rangeof fields considered, the value of n0 is uniformly bounded by κ or some larger value, but wedo not investigate this fact here. The obstruction is the free Λ - module in Gal(ΩE/K∞);but one may always restrict to considering at most a finite family of Λ-torsion modules ata time, for which n0 exists

5The pairings are related but not identical and the relation will investigated in depthfurther on


τ fwgw−1 = gw−1τew , since G∞ = Γ⋉G0, with G0 = Gal(K/Cκ). Replacing

gw−1 by gw−1 · gw and ew−1 by ew−1 + fw, we obtain W =∏w−1i=1 τ eigi a

word of smaller length. The claim follows by induction on w. For α ∈ Λand g ∈ G0, we have

(αg)′ = α∗ · g′.Furthermore, the word reduction above is compatible with the Leopoldtinvolution, which makes it into an automorphism of Λ[G]. However, Λ[G]is in general not a group ring, since αg 6= gα for all pairs (α, g). It is one,if K−1 and B are linearly disjoint. The reflection automorphism of Λ[G] isdefined by:

∑

g∈G

ag ⋉ g 7→∑

g∈G

a∗gχ(g)⋉ g−1, ag ∈ Λ.(18)

We extend the definition of X• to Λ[G] - modules as follows

Definition 4 (Duals). Let X be an abelian p - group which is a compact Γ- module on which G acts, making it into a Λ[G] - module. We define the

dual module X by

Γ×X → X : (γ, x) 7→ γ x = χ(γ)γ−1x,

G0 ×X → X : (g, x) 7→ g x = χ(g)g−1x.

The module Λ[G] acts on X via Leopoldt’s involution on X. Let Y ⊂ Xbe a submodule and suppose that there is a pseudo-isomorphic embeddingY • ⊂ X. If Y such that Y ∼ Y • = Y • ∩X, then Y • is the embedded dualof Y in X. Not all submodules have embedded duals.

We say that X is self - dual, if there is and embedded dual X• ⊆ X suchthat:

X• ⊆ X and X• ∩X ∼ X ∼ X• ∼ X.The product of all submodules Y ⊂ X which have an embedded dual Y • ⊂ Xis the self - dual part ofX, a canonic submodule, up to pseudo-isomorphisms.

The following example explains the difference between duals, Iwasawaduals and embedded duals.

Example 1. Let ς(A) have characteristic polynomial F and f |F be a distin-guished polynomial. Since ς(A) is a Weierstrass module, there is a canonicmaximal submodule Af ⊂ ς(A) annihilated by f ; however, already its Iwa-

sawa dual denoted above by AIf is only embedded in ς(A) iff f∗|F . The fact

that the same holds for the Leopoldt dual Af is a consequence of Theorem2, which will be stated explicitly below.

The p-th cyclotomic field K = C0 yields a concrete case. Let εk, k =0, 1, . . . , p−2 denote like usual the orthogonal idempotents of Zp[Gal(C0/Q)]([25], §6.3) and suppose that a ∈ εkA with k odd is a class with f ∈ Zp[T ] thecharacteristic polynomial of Λa. Then the Iwasawa dual (Λa)I is annihilatedby (εk, f

∗), and it is embedded iff f∗ divides the characteristic polynomial of

20 PREDA MIHAILESCU

εkA. The Leopoldt dual is more natural in the sense that Λa is annihilatedby ((1− εp−k), f∗), thus duality acts also on the group of th base field. Suchmodules only arise, if the Greenberg Conjecture is false. The Conjecture forthis simple field is in fact equivalent to the fact that the self - dual part of Ais trivial.

Finally, if Λa is a Zp - torsion module of infinite p - rank, then it is pseudo- isomorphic to its Iwasawa dual, but this does not imply the existence of anembedded Leopoldt dual.

The main property of reflection is the following: let L′/L/K be a Kum-mer extension of K ∈ Kn : n ∈ N ∪ ∞, such that L,L′ are galois overQ, so Gal(L′/L) is a Λ[G] - module which has exponent pn+1, if K = Kn.If K = K∞, then we assume that Gal(L′/L) is a Λ - torsion module. Forarbitrary b ∈ RAD(L′/L), σ ∈ Gal(L′/L) and α ∈ Λ[G] we have:

〈σ, bα〉 = 〈σα′

, b〉, if K = Kn and [σ, bα] =[σα

′

, b], otherwise.(19)

This is easily verified at finite levels, from the properties of the Kummer

pairing. For g ∈ G we have 〈ag, ν〉 = (〈ag, ν〉)g−1χ(g): by definition ofthe cyclotomic character, g−1χ(g) acts like identity on the roots of unity

and the galois equivariance of the Kummer pairing yields (〈ag, ν〉)g−1χ(g) =(〈agg−1

, νg−1〉)χ(g)

while the bilinearity of the Kummer pairing implies 〈ag, ν〉 =〈a, νg′〉, with g′ = g−1χ(g). It is an important property of extensions L/K∞

which are galois over K, that their multiplicative groups are large self - dualΛ[G] - modules. In particular, the Leopoldt duals of radicals are alwaysembedded, so at infinity we have

RAD(L′/L)• = Gal(L′/L).(20)

This follows from Theorem 2 by taking projective limits in (19). Neverthe-less, by acting with the Leopoldt involution on this identity, we obtain in

general only RAD(L′/L) = Gal(L′/L), since the galois group may not havean embedded dual, even relative to Gal(Ω/L).

The properties of the Kummer pairing at infinity and reflection will be-come fruitful in the next section, where we prove isomorphisms betweensubgroups of A and the Kummer radicals of extensions L/ΩE.

A module X is a Λ[G] - module iff it is a Zp - module closed under theaction of Γ⋉G0. We are mainly interested here in T and T ∗; the associatedΛ[G] - modules will be particularly simple. We investigate at the end someproperties of these particular modules.

If Y is a Λ[G] - module fixed by T , then it is isomorphic to a Zp[G] moduleon which τ acts trivially; if Y is fixed by T ∗, it is still isomorphic to a Zp[G]- module, but τ acts via the cyclotomic character, according to (10), whileG acts via G′, namely (g, y) ∈ G × Y 7→ χ(g)g−1y. In particular, the dualof Λ[G]T , the submodule fixed by T , is always embedded and

Λ[G]T ∼= Zp[G], (Λ[G]T )• ∼= Zp[G′].


If Y is cyclic as a Zp[G] module and Y is embedded in a larger cyclic Λ[G] -

module X, then there is an embedding Y · Y → X with image X0, a cyclicZp[G] - module. The following Lemma gives details about generators in thecase of such cyclic Λ[G] - modules.

Lemma 6. Let K∞ ⊂ L ⊂ L′ ⊂ Ω be a tower with L,L′ galois over Kand X = Gal(L′/L) be a cyclic Λ[G] - module with XT = 1. ThenRAD(L′/L)T

∗

= 1 and RAD(L′/L)• ∼= Gal(L′/L) is a Zp[G′] - moduleon which τ acts by the cyclotomic character.

If X is any cyclic Λ[G] - module annihilated by T and x ∈ X a generator,then there is a subset Gx ⊂ G such that xg : g ∈ Gx is a base for X as

a Zp[G] - module. Moreover, support and annihilator of X and X are dualto each other:

(X⊥)• = (X ′)⊥, (X⊤)• = (X ′)⊤.

The proof is given in Appendix A. The following application is importantfor fields which are not CM. We shall show in (4) that

Zp-rk (Gal(ΩE ∩M/K∞)) = r2 and

rad(ΩE ∩M/K∞) = ∪∞n=κ+1EN∗

nn ·Epn+1

n .

Consider now the group V − := Gal(ΩE∩M/K∞) as a Zp[G] - module. Thenwe may define its canonic supports and annihilators by Definition 8. They

induce a decomposition Qp[G] = Qp[G]e+⊕Qp[G]

e− with Qp[G]e+ = (V −)⊤

and Qp[G]e− = (V −)⊥. We let αe+, αe− ∈ Qp[G] be the respective canonic

idempotents. From the above, we see that

(Qp[G]e+)⊤ = (Qp[G]

e+)• = Qp[G]e−, and(21)

(Qp[G]e+)⊥ = (Qp[G]

e−)• = (Qp[G]e−)⊤.(22)

Furthermore, we have the fundamental property that

|Ee−n | <∞, ∀n ≥ 0.(23)

Moreover, is X is an arbitrary quasi - cyclic Zp[G] - module, then thereis a decomposition in components induced by αe+ and αe−:

X = Xe+ ⊕ Xe−.(24)

We prove the above statements in:

Corollary 2. There is a canonic decomposition of Qp[G] in isomorphyclasses Re+, Re− of submodules of Qp - rank r2, such that Re− = (Re+)•

and Qp[G] = Re+⊕Re−. This induces a canonic decomposition of arbitraryquasi - cyclic Zp[G] - modules X according to (24). The module X = En, isessentially equal to its e+-component, according to (23).

22 PREDA MIHAILESCU

3. Classes as radicals and first applications

In this section we consider the Kummer radicals of extensions L′/L inwhich L ∈ ΩE ,ΩE′ and show their relations to the class group A. Weinvestigate Kummer pairings and radicals of various relative extensions be-tween K∞ and Ω. For this we shall keep the fundamental tower

K∞ ⊂ ΩE ⊂ ΩE′ ⊂ Ω ⊂ Ω(25)

in sight, together with the intersections of H and H with the fields of tower.This gives also valuable insights on the structure of Λ - submodules of Aand A′. If F ⊂ Ω is any extension, it decomposes with respect to the towerinto

K∞ ⊂ F ∩ ΩE ⊂ F ∩ ΩE′ ⊂ F ∩ Ω ⊂ F.

We may also write FE = F ∩ ΩE ,FE′ = F ∩ ΩE′ ,F = F ∩ Ω, F = F. Theextension F/FE will be, with a slight abuse of language, equal to F ·ΩE/ΩE;the groups are most of the times isomorphic; likewise for F/FE′ and F/F. Weshall discuss whether this rule applies or there are deviations, in particularfor every new type of fields F which we encounter.

Let n > κ; we define the intermediate levels of (25) as follows: ΩE′,n isthe largest subextension of exponent pn+1 with

ΩE′,n ⊂ ΩGal(ΩE′/K∞)ωn

E′ .

Thus ΩE′,n/Kn is a Kummer extension of maximal exponent, while the

extension ΩGal(ΩE′/K∞)ωn

E′ is infinite. The fields Ωn, Ωn are defined similarly,

with respect to Ω and Ω: if µ = 0, then Ωn = Ωn.Although we restrict our attention in this paper mainly to the barred

fields, in this Chapter we derive some facts on the structure of A which de-serve a treatment in full generality. In particular, most of the general resultsare not directly necessary for the proof of Leopoldt’s Conjecture. Since atthis point the distinction between T, T ∗ - parts and other Λ - modules issomewhat artificial, we choose to treat the general case in the text ratherthan in an Appendix: most phenomena are identical with the case of theT and T ∗ - parts. For F like above, we let also Fn ⊂ FGal(F/K∞)ωn be thelargest subextension with group of exponent pn+1. At finite levels, the tower(25) and the related filtration of Fn are thus

Kn ⊂ ΩE,n ⊂ ΩE′,n ⊂ Ωn ⊂ Ωn,(26)

Kn ⊂ FE,n ⊂ FE′,n ⊂ F ⊂ F = F,

together with the respective intersections with Hn and Hn, which are alsodefined as the maximal Kummer extensions contained in the Hilbert classfield of Kn, respectively its intersection with H.


3.1. The symbols ΩE,n[a1/p∞ ],ΩE [A

1/p∞ ] and their radicals. If L =

ΩE , then the radicals rad(L′n/Ln) are defined up to units and it will be

possible to relate them to subgroups of An. Consider a norm coherent

sequence a = (an)n∈N ∈ A. The notations ΩE[a1/pn+1

n ] and ΩE [A1/p∞ ] are

meaningful, but require certain precision. We define them in full detail, inthe way that we shall use them, below. In conformity with Definition 3, welet

A := a ∈ A : z(a) ≤ 0 ⊃ Apκ = ς(A),(27)

Then A is a subgroup6 and even a Λ - module. Assume first that ord(an) ≤pn+1 for all n and let B ∈ an and β ∈ Kn with (β) = Bpn+1

, a principal

ideal. The Kummer extension L = ΩE,n[β1/pn+1

] is uniquely determined by

an and we shall denote it by ΩE,n[a1/pn+1

]. Indeed, let B′ = (x)B ∈ an bean ideal from the same class (which needs not differ from B) and (β′) =

(B′)pn+1

= (xpn+1

) · (β); then there is a unit e with β′ = eβxpn+1

and

ΩE′,n[(β′)1/p

n+1] = ΩE′,n[(ex

pn+1β)1/p

n+1] = ΩE′,n[β

1/pn+1] = L.

The field L = ΩE,n[a1/pn+1

n ] is thus well defined for z(a) ≤ 0. In this case, wenote that for any cn ∈ An which generates the same cyclic subgroup in An we

also have L = ΩE,n[c1/pn+1

n ]. The converse holds too, by Kummer theory, so

it might be more accurate then to write ΩE,n[(Zpan)1/pn+1

]. We keep though

notation simple along the usual lines and observe that ΩE,n[a1/pn+1

n ] is well

defined for all an ∈ An. Since A is a Λ-module, the notation ΩE,n[B1/pn+1

n ]

is also defined for all subgroups or submodules B ⊂ A. Let now a ∈ A \A;

we show that Ln := ΩE,n[a1/pn+1

n ] form an injective system. This followsfrom (12) and the injectivity of ι in Proposition 1:

Ln+1 ⊃ ΩE,n+1

[(apn+1

)1/pn+2]= ΩE,n+1[ι(an)

1/pn+1] = ΩE,n+1 · Ln.

Consequently there is a field L = ∪∞n=0Ln ⊂ Ω, which we shall denote by

ΩE [a1/p∞ ].

The next lemma shows that A is inoffensive in our context:

Lemma 7.

∪∞n=0 ΩE,n[A1/pn+1

n ] = ΩE .

Proof. Let a = (an)n∈N ∈ A and Ln = ΩE,n[a1/pn+1

n ]. For any Bn ∈ an ∈ Anand (βn) = Bpn+1

, by point 3. of Proposition 1 and the choice of κ, theideal Bn · O(Kn+κ) = (αn+κ) is principal. Therefore βn = αqn+κ · e for someq ≥ p and a unit e ∈ En+M , so

ΩE,n+κ[a1/pn+1

n ] = ΩE,n+κ · Ln = ΩE,n ⊂ ΩE .

This holds for all an, so the claim follows.

6This should not be mistaken for Iwasawa’s notation X = X•

24 PREDA MIHAILESCU

Remark 1. Since ΩE[A1/pn+1

n ] = ΩE[ιK(A = ΩE, the extensions ΩE,n[A1/pn+1

n ]are volatile in a particular way: they arise at level n and vanish before leveln + κ; nevertheless, at each finite level there is one such extension of con-stant size. Therefore the p - rank of the galois groups Gal(Ωn/ΩE,n) at finite

levels are larger than the Zp - rank of Gal(Ω/ΩE).

We also observe that the extensions ΩE [a1/p∞ ] and ΩE,n[a

1/pn+1

n ] are welldefined for all a = (an)n∈N ∈ Acirc, also when Λa has infinite p - rank.

The p - ramified extensions which are at the center of our attention, arecharacterized by the following common fact from Kummer theory:

Fact 3. Let L = ∪∞n=1Ln with Ln/Kn Kummer extensions of exponent divid-ing pn+1 and Ln+1 ⊃ Ln. If L/K∞ is p - ramified, then there are Kummer

radicals Bn ∈ K×n with (K×

n )pn+1 ⊂ Bn, such that

1. Ln = Kn[B1/pnn ].

2. For each an ∈ Bn there is an ideal B ⊂ O(Kn) and an ideal p whichis divisible only by primes above p, such that (a) = p · Bpn . Inparticular, an may be a unit.

3. If L ⊂M, then aT∗

n ∈ (K×m)

pm.

Proof. Point 1 is a consequence of Ln being Kummer extensions. Since Lnis p - ramified, we deduce point 2. Finally, if L ⊂ M, it is by definitionabelian over K. Therefore, if α ∈ Gal(Lm/Km) is a generator, then αT = 1and Kummer pairing yields

〈a, αT 〉 = 〈aT ∗

, α〉 = 1,

which confirms point 3, the Kummer pairing being non - degenerate.

In view of Fact 3, almost every cyclic extension L/ΩE,n,L ⊂ Ωn arises in

ΩE,n[A1/pn+1

n ]. The almost is related to the fact that in general An ( An;the difference arises from the sequences a ∈ A with 0 < z(a) < κ, but the

good news is that z(a) is uniformly bounded. We needed An in the definition

of A1/pn+1

n in order to have compatible group structures. However, if a =(an)n∈N ∈ A has z(a) > 0, it also gives raise to extensions in Ωn defined

in a similar way to the ones above. Let B ∈ an; then Bpn+1+z(a)= (β) is

a principal ideal and z(a) is the least positive integer for which this holds.

Arguing like before, we see that L = ΩE,n[β1/pn+1

] is an extension thatdoes not depend on β or B but only on the class an. One might write it

as ’ΩE,n[apn+1+z(a)

pn+1

n ]’; it is a radical extension related to the group An: the

canonic definition above corresponds to ’ΩE,n[apn+1

pn+1

n ]’ and only applies whenz(a) ≤ 0. It is useful to keep this ugly notation for a short time, in order toclarify the point where we need – and develop – a canonic work around forthe case when z(a) 6= −∞.


The next lemma shows that all cyclic extensions of ΩE,n arise from roots

of An:

Lemma 8. Let n ≥ κ and L ⊂ Ωn be a cyclic extension of ΩE,n. Then

there is a class an ∈ An such that L = ΩE,n[apn+1+z(a)

pn+1

n ].

Proof. Let L = ΩE,n[β1/pn+1

] for some β ∈ ΩE,n. Then it follows from Fact

3 that (β) =(Bpord(B)

)pm· π with π ∈ E′

n and 0 ≤ m ≤ n + 1; if an = [B]

is the class of B, it follows that L = ΩE,n[apn+1+z(an)+m

pn+1

n ].

We have thus a one - to - one map between the RAD(Ωn/ΩE,n) and frac-tional powers of An. As a map υ : An → RAD(Ωn/ΩE,n) it is however nota group homomorphism, since, for m = 0 in the above proof, we may have

υ(an) = apn+1+z(a)

pn+1

n and υ(bn) = bpn+1+z(b)

pn+1

n , but υ(anbn) = (anbn)pn+1+z(ab)

pn+1 ,and since z is not a morphism, neither can υ be. We have thus a canonichomomorphism from a subgroup of An to a subgroup of the radical of Ωnand a map to the full radical, which is not a homomorphism.

The work around is based on the uniform bound on z(a) and uses the fact

that ς(An) ⊂ An, and thus apn+κ+1

pn+1 =((ap

κ)pn+1)1/pn+1

in ’ugly’ notation,

corresponds to the canonical root ς(an)1/pn+1

. From Lemma 8 we have

ΩSκ(Gal(Ωn/ΩE,n))n = ΩE,n[ς(A)

1/pn+1

n ] ⊂ ΩE,n[A1/pn+1

],(28)

Kn+κ · Ωn ⊂ Ωn+κ[A1/pn+κ

n+κ ].

where Sκ is the κ-th socle in Definition 1; we have an exact sequence

1→ ς(An)→ RAD(Ωn/ΩE,n)→ Cn → 1,

where Cn is a group of exponent bounded by pκ. At infinity, we obtain

Ω = ∪nΩn = ∪nKn+κ · Ωn = K∞ · ∪nΩn = ∪nK∞[A1/pn+1

n ]

= ∪nK∞[ς(An)1/pn+1

] = ∪nK∞[ιK(An)1/pn+1+κ

]

We have the following back - doors to the canonic radicals: ς annihi-lates the full Zp - torsion, which may be more than we wish, but ιK onlyannihilates An, thus the volatile part of Ωn/ΩE,n. We have

Gal(Ωn/Ωn) = Gal(Ωn[An]/Ωn)

∼= Gal(ΩE,n+κ[ιK(An)1/pn+κ+1

]/ΩE,n+κ),(29)

Gal(Ωn/ΩE,n) ∼= Gal(ΩE,n+κ[ιK(An)1/pn+κ+1

]/ΩE,n+κ).

Finally, since ς(A)n ⊂ An while ιK(An) ⊂ An+κ and the radical A1/pn+1is

canonical for all n, we obtain the following consistent definition of radicalsof class groups:

26 PREDA MIHAILESCU

Definition 5 ( Class - group Radicals). Notations being like above, the

symbols ΩE,n[ς(an)1/pn+1

] and ΩE,n+κ[ιK(An)1/pn+κ+1

] are well defined forall an ∈ An and

RAD(Ωn/ΩE,n)pκ ⊂ ς(An)

1/pn+1 ⊂ RAD(Ωn/ΩE,n),

RAD((Ωn ·Kn+κ)/ΩE,n+κ

)= ιK(An)

1/pn+κ+1 ⊂ A1/pn+κ+1

n+κ .

Furthermore, ΩE [ς(A)1/p∞ ] = ∪nΩE,n[ς(An)1/p

n+1] and

ΩE [ιK(A)1/p∞ ] = ∪n>2κΩE,n[ιK(An−κ)

1/pn+1]

are well defined notations too, and

Ω = ΩE [ς(A)1/p∞ ], Ω = ΩE[ιK(A)

1/p∞ ].(30)

Herewith, we define the Class group radicals by

C-RAD(Ω/ΩE) = ιK(A)1/p∞ , C-RAD(Ω/ΩE) = ς(A)1/p

∞

,

C-RAD(Ωn/ΩE′,n) = ιK(An−κ)1/pn+1

, C-RAD(Ωn−κ/ΩE,n−κ) = ς(An)1/pn+1

.

We also have Ωn = Ωn[ιK(An−κ)1/pκ ] for all n > 2κ and Ω = Ω[ιK(A)

1/pκ ].By the choice of indeces, we see that all the class field radicals C-RAD(L′/L)are canonically isomorphic, at finite and infinite levels with the plain radicalsRAD(L′/L) in all the situations covered by the definition. This yields a solid

interpretation to the notation F[A1/p∞ ] used by Lang in [22] §6.2, at least

for the case then L′ ⊃ ΩE,n or L′ ⊃ ΩE . The notation K∞[A1/p∞ ] must

be understood as an arbitrary choice of βn ∈ Kn such that (βn) = Bpn+1

n ,with B, βn like above: this is in general not a canonic choice, but there areexceptions. For instance, if K is CM and we restrict to the minus part Ω−,i.e. the fixed field of Z+, then, for odd p, the radicals (A−)1/p

∞

are welldefined over K∞.

The radicals ς(A) are Zp - torsion destructive, while ιK(A) conserves theµ - part. According to the interest or disinterest for the µ - part, one maychoose one or the other, both being canonic. By the above, they are notdistinct kinds of radicals, they are distinguished by the choice between Ωnand Ωn, resp. Ω or Ω as upper field.

3.2. Embedding adjoints. Let B ⊂ A be the projective limits of idealclasses an ∈ An containing products of ramified primes above p, followingGreenberg’s notation [12] and D = A/(AT ·B). Then the simplest result inour direction is

ΩE′ = ∪∞n=1ΩE,n[Π1/pn ] = ΩE,n[B

1/p∞ ].

The inclusion ΩE′ ⊇ ΩE,n[B1/p∞ ] is clear. The other direction follows from

the definition of the p-units E′. We may thus proceed with the investigationof extensions over ΩE′,n – but will encounter ΩE′/ΩE once again, whendiscussing unramified extensions.


Since the p - units are powers in ΩE′,n, it follows that

Ωn = ΩE′,n[ς(A′n)

1/pn+1], Ω = ΩE′ [ς(A′)1/p

∞

].(31)

We shall fix a Λ - module A† ⊂ A′ which is a complement of the Zp -

torsion: A′ = A†⊕(A′). Let F (T ) ∈ Zp[T ] be the characteristic polynomial

of A′ – which is also the one of A† and f |F be a distinguished polynomial.Since ς(A′) is a Weierstrass module, by Lemma 2 there is a canonic sub-

module ς(A′)f ⊂ ς(A′) with ς(A′)f ∼=(A′/(A′)f

)pκ. This shows a relation

to the double adjoint of A′/(A′)f ); adjoints were introduced by Iwasawa in[13] in relation to Kummer pairing. Later, in [14] §1.3, he follows Tate andthe definition is free of concern for an explicite embedding of the adjointα(X),X ⊂ A in A. We do not need adjoints here, but are interested in theobservation that the main property of α(X) ∼ X•, namely that of being acanonical module free of finite Zp - torsion, is close to our ς(A′). Indeed, ifX ⊂ A′ has finite p - rank, we see that

ς(X) ∼= (α α(X))pκ(X) ∼= (α(α(X)))p

κ ∼= α(α(X)).

For arbitrary X, we have ιK(X) ∼= α(α(X)).Restricted to RAD(Ω/ΩE′), it follows that ς(A′)f ∼= α(α(A′/(A′)f )), so

we may regard ς(A′)f as an explicite embedding of α(α(A′/(A′)f )) in A′.

Furthermore, for any choice of A†, there is a unique Weierstrass submoduleA′f ⊂ A† such that ς(A′

f ) = ς(A′)f , that is, a pκ-th root of ς(A′)f in

A†: there is no such canonic root, but the possible choices are all reducedto the one choice of A†. We may write with a slight abuse of language

ΩE′ [ς(A′)1/p∞

f ] = ΩE′ [(A′f )

1/p∞ ].

We now construct subfields of Ω and Ω with group fixed by f . Since

∪n(fY(1 + pnZp)) = fY and ∪n (fH(1 + pnZp)) = fH,

the subgroups fY ⊂ Y and fH ⊂ H are compact and there are fixed fields

Ωf := ΩYf and Ωf := Ω

Hf , which are the maximal subfields of Ω with galois

groups over K∞, resp. H, annihilated by f .

Definition 6 ( The fields Ωf , Ωf and the modules Af , ς(Af ) ). Let F ∈Zp[T ] be the characteristic polynomial of A and f |F . The fields Ωf , Ωf aredefined by

ΩYf , and Ωf := Ω

Hf .

The intermediate fields Ωn,f , Ωn,f are the maximal Kummer extensions of

exponent pn+1 over Kn, respectively Hn, which are included in Ωn, resp. Ωf .

For f(T ) 6= T , we define ς(A)f = x ∈ ς(A) : xf = 1 and we have in

fact ς(A)f ⊂ ς(A′); furthermore Af ⊂ A† is the unique submodule in the

fixed Weierstrass module A†, such that ς(Af ) = ς(A)f . In this case we haveς(A′)f = ς(A′

f ), A′f = Af .

28 PREDA MIHAILESCU

For f(T ) = T we let ς(A′)T = x ∈ ς(A′) : xT = 1 and ς(A)T =ς(B) · ς(A′)T . The modules AT , A

′T are defined by intersection with A† ∩A′

and A†, like before.The galois groups are

Gal(Ωf/K∞) = Yf Gal(Ωf/K∞) = Hf ,

Gal(Ωn,f/Kn) = Yf,n Gal(Ωf,n/Hn) = Hf,n

We have ΩT = M; the further fields of interest in this paper are ΩT ∗ andΩT ∗ together with the modules A′

T , AT ∗, resp. ς(A′)T , ς(A)T ∗ .

3.3. The principal self-dual module of Iwasawa theory. We come toconsider the intersections of Hwith Ωf . By definition, the group Gal(H/K∞) ∼=A/A ∼= ς(A) is a Weierstrass module, and the Artin map induces an iso-morphism ϕ : ς(A) → Gal(H/K∞). For f |F as above we let Hf = H ∩ Ωfand HE = H ∩ ΩE = H ∩ΩE′ ; the last equality is a consequence of the factthat ΩE′/ΩE is totally ramified. Let AE ⊂ ς(A) be the group such thatϕ(AE) fixes HE in H. Note that the Artin map ϕ only determines a non

canonic counterpart A†E ⊂ A†, which explains the use of ϕ.

Let a, b ∈ ς(A); ramified and unramified extension have the following

product rule: if ΩE [a1/p∞ ] and Ω[b1/p

∞

] are unramified, then so is ΩE [(ab)1/p∞ ].

If only one is ramified and one is unramified, then ΩE [(ab)1/p∞ ] is ramified.

If both are ramified, nothing can yet be said about ΩE[(ab)1/p∞ ]. There is

a maximal unramified subextension Ω/ΩE , and we denote by Anr ⊂ ς(A)

its class - radical: the maximal submodule such that ΩE [A1/p∞nr ] is totally

unramified. Since RAD((ΩE ∩ H)/K∞) is a Λ - cyclic Weierstrass moduleby definition of ΩE and H, it follows that

ς(A)/ς(AE) ∼= Gal((ΩE ∩H)/K∞)

is also a Λ - cyclic Weierstrass module. We let Ar ⊂ ς(A) be a radicalfor Ω/ΩE · H. Then Ar ∼= ς(A)/Anr, yet the choice of Ar is in general notcanonic; but see also the proof of Theorem 3. It is not difficult to verify that

the notation HE[A1/p∞nr ] is well defined and the field diagram

ΩE ΩE [A1/p∞

nr ]

H HE [A1/p∞nr ].

(32)

commutes. Indeed, let a = (an)n∈N ∈ Anr. At finite levels, if ΩE,n[α1/pn+1

n ] =

ΩE,n[β1/pn+1

n ] with (αn) = (βn) = Bpn+1and [B] ∈ an, then the extension is

unramified and we find that Kn[αn/βn] ⊂ HE,n, which shows thatHE[A1/p∞nr ]

is well defined. By definition of AE we have

ϕ(AE) ∼= Gal(Ωnr/ΩE) = Gal(ΩE [A1/p∞nr ]/ΩE) ∼= Anr,


the last isomorphism being by (20). Since the Artin map is covariant, itfollows by comparing the first and last terms above that Anr and AE haveembedded duals in A, which are dual to each other: A•

nr = AE and A•E =

Anr. Using Iwasawa’s skew symmetric pairing we have the stronger result:

Theorem 3. Notation being like above,

1. The groups AE = Anr, being thus self - dual; Gal(Ωnr/ΩE) ∼= AE isa self - dual group, isomorphic to Anr.

2. For f |F and ς(A′)f defined in Definition 6 we have

ΩE′ ·Hf = ΩE′ [(ς(A′)f∗ ∩Anr)1/p∞

],

ΩE′ · Ωf = (H · ΩE′)[(ς(A′)f∗ ∩Ar)1/p∞

].

3. The ramified extension Ω/(ΩE ·H) is related to HE by:

Gal(Ω/(ΩE′ ·H))• ∼= Gal(HE/K∞) ∼= ς(A)/AE ∼= Ar.

4. Let FE = χA/AE ∈ Zp[T ] be the characteristic polynomial; if f isa Weierstrass polynomial such that f |FE and (f, F/f) = 1, then

ΩE′ [ς(A)1/p∞

f ]/ΩE′ is totally ramified.

5. In particular, ΩE′ [ς(A)1/p∞

T ∗ ]/ΩE′ and ΩE′ [ς(A′)1/p∞

T ]/ΩE′ are totallyramified.

We review in Appendix B the Iwasawa skew symmetric pairing and givea proof of the theorem by using this pairing.

3.4. Complex conjugation, explicite radicals and estimation of ranks.

If K is a CM field and ⊂ Gal(K∞/Q) is the complex conjugation, then forany Λ - module X there is a canonic splitting

X = X+ ⊕ X− = X(1+)/2 ⊕ X(1−)2,

and the denominator can be omitted for p > 2. Accordingly, if L/K∞ is a

CM extension, then L+ = LGal(L/K∞)− ,L− = LGal(L/K∞)+ . An importantproperty of complex conjugation is

Fact 4. If K is CM, then the capitulation kernels Ker (A−n → A−) = 1

for all n.

We gather some technical results on explicite radicals. The core questionis the following: let f |F and Ωn,f ⊂ Ωn be the maximal Kummer extensionsof Kn of exponent pn+1 included in Ωf . How can the radicals rad(Ωn,f/Kn)be expressed explicitly, in the filtration (26) ? The answer is in general

indicated by the following observation: let pc(n) be the least power of p inthe ideal (f∗(T ), ωn(T )) and suppose explicitly that

f∗(T ) · un(T ) + ωn(T )v(T ) = pc(n)wn(T ), un, vn, wn ∈ Λ.

Then un+1(T ) ≡ un(T ) mod ωn(T ). If Rn = rad(Ω/K∞) then, essentially,

rad(Ωn,f/Kn) = R(ωn,un)n , where some additional work has to be done about

the powers of p. For estimating ranks of totally ramified extensions, we are

30 PREDA MIHAILESCU

also interested in Uf and its intermediate groups Un,f , where U := ∪nUn.The idea is common for both questions and the technical details are relatedto the issue of eliminating eventual p - powers. In our case, the interestingvalue is f(T ) = T . The results are used, among other, for determining Zp -ranks of the interesting extensions in (2). Some of the results of this sectionwill be proved in Appendix B. The Lemma 14 in Appendix A is a centralinstrument of proof. It has the following important consequence, proved inAppendix B:

Proposition 3. For each n > κ we have

p-rk(HT ∗,n) = r2 + s− 1,(33)

and in the limit,

Zp-rk(HT ) = r2 + s− 1.(34)

Recall that K is a complex galois extension, so r1 = 0. The secondapplication of Lemma 14 is a constructive proof of the known result (35);the constructive approach yields additional information about the radicals,which brings insights on duality in Zp[G].

Proposition 4. Let En = EN∗

n,κn Ep

n+1

n and E = ∪∞n=κ+1E1/pn+1

n /En. Then

p-rk(E) = r2 and M ∩ ΩE = K∞[E] = ∪Kn[E1/pn+1

n ; moreover

Zp-rk(Gal((M ∩ ΩE)/K∞)

)= r2.(35)

When K is a CM field, by class field theory ([22], Chapter 5, Theorem5.1) and since E− is finite, being equal to the group of roots of unity, (35)specializes to

M− ⊂ Ω−E .(36)

In both cases, Leopoldt’s Conjecture is equivalent to

M ⊂ ΩE .(37)

We see from (35) and (36) that, independently on the truth of Leopoldt’sConjecture, the intersection M∩ΩE is a canonical subfield of M. In the CMcase, it coincides with M−, while the global Artin symbol ϕ is bijective as amap U(K)−/(U(K))→ Gal(M−/K∞).

In the case when Leopoldt’s Conjecture is false, Zp-rk(Gal(M/K∞)) =r2 +D(K) and it follows from (35) that Zp-rk(Gal(M/ΩE)) = D(K). There

is a submodule A∗ ⊂ AT ∗ of rank D(K) such that M ·ΩE = ΩE[ς(A∗)1/p∞ ]:

indeed, the radical of RAD((M·ΩE)/ΩE) is a class field radical by definition,and duality implies that it is annihilated by T ∗. We shall write Φr = ΩE ·M,a phantom extension with group over ΩE having the rank equal to theLeopoldt defect. Let Φ = HT ∗ and D = RAD(Φ/K∞); then DT = 1.Now D is a Weierstrass module, and from Fact 3 and the finiteness of A0

we deduce that ς(D) ⊂ E(K). It is not difficult to see that limn p-rk(ι(E) ∩Up

n) = D(K), but the following lemma gives a construction of the radicals

rad(Φn/Kn):


Lemma 9. Assume that the Leopoldt defect r = D(K) > 0. For every n > 0there is a Z - submodule Dn ⊂ E such that (Dn ·Ep)/Ep has p - rank r and

ι(Dn) ⊂ Upn+1

. Furthermore, Dn+k ⊂ Dn ·Epn+k+1

and (Dn ·Epn+1

)/Epn+1

is a group with exponent and sub-exponent pn+1. In particular, the exten-

sions Kn[D1/pn+1

n ]/Kn are unramified and form an injective sequence with

limit ∪nKn[D1/pn+1

n ] = Φ.

It follows that the field Φ = HT ∗ has group of rank D(K): it is the firstphantom field mentioned in the introduction. Furthermore, since Gal(H/K∞)is a Weierstrass - module, we have AT ∗

∼= Gal(Φ/K∞) so A∗ = AT ∗ , havingthe same rank.

The last radicals which we need to consider for the sequel are rad(ΩT ∗,n/Kn).The following result concerns these radicals

Lemma 10. Let K be like above, let n ≥ 3κ and Ln ⊂ ΩE,n be a cyclicsubextension with group annihilated by T ∗ and [Ln : Kn] = pm ≤ pn+1,

m ≥ κ. If Ln = Kn[e1/pmn ], then the unit en ∈ E(Kn) verifies

en = cn · wpm−(κ+1)

n , cn ∈ E′(K) \ (E′(K))p, wn ∈ E2κ · (E′n)T .(38)

Proof. It follows by duality from the definition of Ln, en, that eTn = dpm ∈

Epm

n and thus Nn(d)pm = 1, so Nn(d) = ξ ∈ µpκ+1 ⊂ K. By Hilbert 90,

it follows that dpκ+1

= wTn , wn ∈ Kn and thus(en/w

pm−(κ+1)

n

)T= 1, so

en = cn · wpm−(κ+1)

n with cn ∈ K. Furthermore, wTn ∈ En and by Lemma16, it follows that wn = d1d

T2 with d1 ∈ E2κ, d2 ∈ E′

n(K). In particular,cn ∈ E′(K) and since en 6∈ Epn, we must have cn 6∈ (E′(K))p, which completesthe proof.

As a consequence concerning the filtration of ΩT ∗, it follows that ΩE′ ∩ΩT ∗ = K∞[(E′

1)1/p∞ ]. We shall write from now on ΩE1 = K∞[E(K)1/p

∞

]

and ΩE′

1= K∞[E′1/p

∞

1 ] = ΩE1 [Π1/p∞ ]. It thus follows from Lemma 10 that:

ΩT ∗ ∩ ΩE′ = ΩE1 [Π1/p∞ ] = ΩE′

1.(39)

4. The filtration of M/K∞ and the Conjecture of

Gross-Kuz’min

We recall from Theorem 3 the main result of interest for the sequel:

Lemma 11. The fields ΩE′ [ς(A)1/p∞

T ∗ ],ΩE′ [ς(A′)1/p∞

T ] are totally ramified

over ΩE′ and HT ⊂ME.

Let Rn = E(K)pn−κ ·Nn(En) for n ≥ κ; then Rn ⊃ Rn+1 and the quotients

En := E(K)/Rn form a projective system with limit E := lim←−E(K)/Rn.Then

Lemma 12. Let E := lim←−E(K)/Rn, like above. There is an injective map

of Zp[G] - modules υ : ς(A′)T → ς(E) with ς(E) ∼= ς(A′)T .

32 PREDA MIHAILESCU

Proof. We give an explicite construction of the map υ. Let a′1,a′2, . . . ,a

′k ∈

A′T be a minimal set of generators. Then z(a′i) ≥ 1−κ; let ai = (a′i)

p−(1+κ+z(a′i) ,so z(ai) = −(1 + κ) for all i. We shall prove that for any a = (an)n∈N ∈ai : i = 1, 2, . . . , k and each n > 0, there is a unique en ∈ E(K) dependingon an and such that en 6∈ Nn(En); furthermore en+1 ≡ en mod Nn(En).

Let n > 2κ,An ∈ an and (αn) = Apn−κn ; since aTn = 1, there is a νn ∈ K×

n

with ATn = (νn) and αTn = dnν

pn−κn , dn ∈ En. Let c := Nn(νn) ∈ E(K); then

Nn(c) = cpn−κ

= Nn(νn)pn−κ = Nn(en)

−1; thus by Hilbert 90, en = c · uTand αTn ≡ en mod

((K×)T · (K×)p

n−κ). In this form, one can verify that

en does not change upon choosing different representants for the principal

ideal Apn−κ

n or other ideals from the class an. Assuming that en ∈ Nn(En),then Nn(δν) = en · e−1

n = 1 and there is a unit δ ∈ En such that ν = δ · xT .It follows that (An/(x)) = (1) and An is an ambig ideal, thus a product oframified primes and ideals from K which capitulated in Kn. Since n > 2κ allideals from K belonging to classes in coherent sequences of bounded orderare principal in Kn, but not an: it remains that an ∈ Bn and αn is a p -unit,which contradicts the assumption a ∈ A′. Therefore en ∈ Nn(En) iff an isthe principal class and there is thus an injective map υn : A′

T,n → En which

extends an 7→ en by Z/(pn−κ · Z)-linearity.In order to show that en ∈ En form a projective system, let m > n > κ

and Qm ∈ am be a prime that splits completely over Q and αm, νm, embe defined with respect to Qm. Then we have seen that em = Nm(νm)

−1

independently of the choice of the generator νm of ATm. Since (Nm,n(νm)) =Nm,n(Qm)

T = QTn = (νn), it follows from the uniqueness of em, en, that

em ≡ en mod Nn(En). Note also that υn(axn) = Nn(ν

−xn ) = υn(an)

x, bothfor x ∈ N and for x ∈ G. Thus the maps υn can extend in the projective limitto an injective morphism of Zp[G] - modules υ : A′

T → E. The initial choiceof a deserves some remark: in order to achieve ord(an) = pn−κ uniformly onall an – which simplifies the notation – we assumed that a = aq for some p- power q and a ∈ A′

T ⊂ (A′T )p. We may define νn, en with respect an, and

the map υ extends naturally to a; it follows that υ(a) = υ(a)q. Thus υ isdefined on a minimal set of generators for A′

T and extends to an injectivehomomorphism of Zp[G] - modules, as claimed.

Conversely, let e = (en)n∈N ∈ E and n > κ. By the Hasse Norm Theoremfor cyclic extensions, there is a xn ∈ Kn with Nn(xn) = en and, consideringthe prime ideal decomposition of xn, we find that there is some ideal Xnsuch that (xn) = XTn . The definitions above show that en = υn ([Xn]);let xn = [Xn] and x = (xn)n∈N ∈ A. Obviously xT = 1; furthermore,ord(x) =∞ since otherwise ord(Xn) ≤ κ and one deduces that eκn ∈ Nn(En)for all n and thus ς(e) = 1. The restriction of υ to ς(A′)T is thus bijective,which completes the proof.

We are prepared for the


Theorem 4. The Conjecture of Gross - Kuz’min holds for K.

Proof of the Conjecture of Gross-Kuz’min. For completing the argument, weconsider first the simpler CM case. Since HT ⊂ ΩE , by Lemma 11, it followsby reflection, in the CM case, that A′

T ⊂ A−. However, E(K)− = µpκ is atorsion group, and the previous lemma implies that (A′

T )− = A′

T = 1 inthis case, which completes the proof. Using the Lemma 18, this argumentextends to the non - CM case.

We give an alternative proof for the non CM case, using Lemma 12.Formally, En are submodules of factors of E(K); the extensions Ω′

n =

Kn[ς(E)1/pn+1

n ] are well defined, since E(K)pn+1 ⊂ Rn. Let Ω′ = ∪nΩ′

n =K∞[ς(E1/p∞ ] and Y ′ = Gal(Ω′/K∞). By reflection we have (Y ′)T

∗

= 1,and thus Ω′ ⊂ ΩT ∗ ∩ΩE while Ω′′ := ΩE[ς(A

′)1/p∞

T ] ⊂ ΩT ∗ too. The radicalsof these extensions are isomorphic and υ induces by duality an isomorphism

Gal(Ω′/K∞) ∼= Gal(ΩE [ς(A′)1/p∞

T ]/ΩE).

We know little about ΩT ∗ but HT ∗ = Gal(ΩT ∗/H) is a quasi - cyclicZp[G] module of essential Zp - rank r2 + s − 1, by Proposition 3; also

ΩE[ς(A′)1/p∞

T ]/ΩE is totally unramified by Lemma 11. The quasi - cyclicity

of YT ∗ implies that Ω′ ∩ ΩT ∗ = K∞. It follows that Ω′ ⊂ ΩT ∗ ∩H = Φ. Wehave the following diagram of fields:

K∞ Ω′ Φ

ΩE ΩE[ς(A′)1/p∞

T ] ΩT ∗ · ΩE.

(40)

in which the field extensions on the right hand side have isomorphic groupsand radicals, while both lower and upper lines are Zp[G] - quasi cyclic ex-tensions. Since Gal(Φ/K∞) ∼= AT ∗ , it follows that ς(A′

T ) → ς(AT ∗)•. Butthis contradicts the Lemma 11, thus completing the proof of the Gross -Kuz’min Conjecture for non CM fields.

We might try dualization for proving the Conjecture of Leopoldt. The

natural dual of E(K) appears to be ∪nEN∗

nn . The condition Nn(νn) 6∈ Nn(En)

in the proof of Lemma 12 becomes νN∗

nn 6∈ EN

∗

nn . However, in turns out in

this case that xn := νN∗

nn is nothing else but ς(αn), up to units. Using thus

xn to define a new map υ′ is thus not fruitful, since the image of this mapis essentially ς(AT ∗) itself. The approach for the Leopoldt Conjecture willtherefore take an other path.

5. The filtration of ΩT ∗ , ΩT ∗ and the proof of the Main

Theorem

We have shown by class field theory in Lemma 3 that Zp-rk(Gal(ΩT ∗/HT ∗)) =

r2+s−1. We now traverse the filtration (25) from upside down: ΩE′ ·ΩT ∗ =

34 PREDA MIHAILESCU

ΩE′[ς(A′)1/p∞

T ] = ΩE′ in view of the proof of the Conjecture of Gross-Kuz’min. Note that we only need the first step of the proof for this,namely the argument based on self-duality of Anr, Lemma 11. We thushave ΩT ∗ , ΩT ∗ ⊂ ΩE′ . From (39) we gather that ΩT ∗ ⊂ ΩE′

1. Conversely,

Gal(ΩE′

1/K∞)T

∗

= 1, so we have

ΩE′

1= ΩT ∗.

It is a straight forward verification that

Zp-rk(Gal(ΩE′

1/K∞)) = p-rk(Π)+p-rk(E(K)) = s+r2−1 = Zp-rk(Gal(ΩT ∗/Φ)).

Since ΩT ∗ ∩ H = Φ and the groups YT ∗ and YT ∗ are torsion - free bydefinition, a rank computation yields for the intersection Ωi = ΩT ∗ ∩ ΩT ∗

that

Zp-rk(Gal(Ωi/K∞)) = r2 + s− 1−D(K)

while ΩI ⊂ ΩE′

1since both fields are included in ΩE′

1. The radical rad(ΩI/K∞) ∼=

E′1/rad(Φ). However, since Zp-rk(YT ∗) = r2+s−1 too, it remains that there

is a totally ramified extension Φ∗/Φ with L ⊂ ΩE and Gal(L/Φ)T∗

= 1,while Zp-rk(Gal(Φ∗/Φ)) = D(K). This is the last phantom extension wehave to consider; we shall show by a detailed investigation at finite levels,that it cannot exist, and this implies the Main Theorem.

Like usual, we denote by ΩT ∗,n the maximal Kummer extension of Kn ·(Hn ∩ ΩT ∗) of exponent pn+1 included in ΩT ∗ . Since it is a subfield of Ωn,

it is important to observe that Gal(ΩT ∗,n/Kn

)is abelian and it may have

larger exponent than pn+1. We shall give a proof of the Theorem 1, byshowing that there are no abelian extensions of Kn which are ramified andcontain Φn. This implies that Φ∗ cannot exist and D(K) = 0, thus theLeopoldt Conjecture.

The extension (Ωf∗ ∩ ΩE)/K∞ is defined for every distinguished polyno-

mial; if f does not divide the characteristic polynomial FE(T ) = char(Gal(H∩ΩE)), then Ωf∗ = Ωf∗ . The polynomial of interest in relation with Leopoldt’sConjecture is f(T ) = T and we know that T |FE(T ) iff D(K) > 0. We shallderive a contradiction in Proposition 5 below, which implies D(K) = 0: we

show that Gal(ΩT ∗/HT ∗) cannot have the rank proved by class field theoryin Lemma 3, a fact based on Lemma 10 and the resulting particularity off(T ) = T , which makes that

rad((ΩT ∗,n ·K2n)/K2n) ⊂ E′1 · (K×

2n)p2n+1

.

Since this is the central point of the proof of Leopoldt’s Conjecture, itmay be useful, before proceeding to the technical proof of the Proposition 5below, to compare to cases f |FE , with f(T ) 6= T , which are cases which dooccur. We illustrate in Appendix C on the example of K = Q[ζp], the factthat for some polynomials f(T ) 6= T , one has in general extensions Kn ⊂Fn ⊂ Ln such that Fn/Kn is unramified, Ln/Fn is p - ramified and Ln/Kn


abelian. Furthermore both groups Gal(Fn/Kn),Gal(Ln/Fn) are annihilatedby f(T ∗). As predicted, the difference consists in the fact that in the generalcase, the radicals

rad((Ωf∗,n ·K2n)/K2n) ⊂ E′2n · (K×

2n)p2n+1

.

and the effective part rad((Ωf∗,n · K2n)/K2n)/(K×2n)

p2n+1 6⊂ K×n , modulo

p2n+1 powers. It turns out that Ωf∗ 6⊂ Ωf∗ but Ωf∗ ⊂ Ω(f∗)2 . At finite levels,

every extension Ln ⊂ Ωf∗ \ Ωf∗,n, which is cyclic over Kn of degree pn+1 <

[Ln : Kn] ≤ p2(n+1) arises as subfield in the product Fn · F′n of subfields

Fn ⊂ Ωf∗,n and F′n = Ω(f∗)2,n; however, only the Fn form an injective

sequence! This general phenomenon is analyzed in detail in Example 2.

❩❩❩❩❩❩❩❩❩❩

❩❩❩❩❩❩❩❩❩❩

H∞ Ω

ΩE1

K

Ωr

K∞

H∞

Q

Ωr0

Φ

Fig. 1: Overview of the main extensions of K∞.The values accross lines are Zp - ranks of galois groups

s

r2 − 1−D(K)

Zp-rk(B)

D(K) ΩE′ ∩ ΩT ∗

ΩT ∗

Trivial extension, L. 8ΩE ∩ ΩT ∗

Turning back to f(T ) = T we prove:

Proposition 5. Notations being like above,

Zp-rk(Gal((Ω ∩ ΩE′)/H)

)= r2 − 1 + s−D(K).(41)

Proof. We have seen in Lemma 10 that rad(ΩT ∗,n/Kn) stems from E′(K).On the other hand, Lemma 3 implies that the maximal p - abelian exten-sion ΩT ∗,n/Kn, which is totally ramified at p above Hn, has group YT ∗,n =

Gal(ΩT ∗,n/Hn) with p-rk(YT ∗,n) = r2 + s − 1 and sub-exponent at least

p(n−k)/2. If ΩT ∗,n ⊂ ΩT ∗,n, since there are D(K) independent unramifiedextensions of Kn in ΩT ∗,n, namely Φn ⊂ ΩT ∗,n, the claim already follows. If

the claim is false, we must thus have ΩT ∗,n 6⊂ ΩT ∗,n.

36 PREDA MIHAILESCU

The previous example shows that the fields in ΩT ∗,n may be extensionsof ΩT ∗,n with group over Kn which is annihilated by (T ∗)2 rather than T ∗.

It suffices to consider subfields of ΩE : we shall show that

ΩT ∗,n ∩ ΩE,n = ΩT ∗,n ∩ ΩE1,n,

and the galois group has p - rank r2−1−D(K) over Φn; it follows then that

Ω′n = ΩT ∗,n ∩ ΩE1,n[Π

1/p∞ ] has a group of p - rank r2 − 1 + s − D(K) overHn and

Zp-rk(Gal(Ω′/HT )) = r2 − 1 + s−D(K).

Let δ ∈ E(K) be a Minkowski unit with annihilator θ ∈ Qp[G], let q′ be

the p - power dividing |G| and n be sufficiently large. For M > 4(n + 1),let αM , θM ∈ Z[G] be approximants to the pM -th order of q′α, q′θ ∈ Zp[G]as in the Lemma 9, so α generates the annihilator ideal of Gal(Φ/K∞).

We suppose that M is such that E(K)θM ⊂ U(K)p4n. Using the approxi-

mants above, we have Φn ⊂ ΩθME1,n= Kn[δ

θM/pn+1], an abelian unramified

extension with group of p - rank D(K). Therefore ΩθME,n/Φn must contain

D(K) independent cyclic extensions of sub-exponent p(n−k)/2 and with groupannihilated by T ∗: this follows from Lemma 14.

We restrict ourselves for simplicity to one maximal cyclic extension Ln/Kn

with Ln ⊂ ΩθMn,T ∗ ∩ ΩE: as observed above, if the claim of the proposition

was false, such an extension must exist. Let Ln∩Φn = Fn = Kn[d1/pn+1

] forsome d ∈ DM , the radical of Φn. Then p

m := [Ln/Fn] ≥ pn/2−κ,m ≤ n+ 1and Gal(Ln/Fn)T

∗

= 1. Furthermore, Ln is cyclic over Kn. We shall showthat the two conditions are incompatible. Assuming that such an extensionexists, then L′

n = K2n · Ln is an abelian extension of Kn and Kummer overK2n.

Let e ∈ E2n generate its radical, so L′n = K2n[e

1/pn+1+m]. Since K2n ·

Fn = K2n[d1/pn+1

] ⊂ L′n, we may assume by Kummer theory that e =

d · upn+1for some u ∈ E2n. Since (Gal(L′

n/K2n[d1/pn+1

])T∗

= 1, we must

have by duality also that (dupn+1

)T = (upn+1

)T ≡ (dupn+1

)pmmod Ep

n+m+1

2n

so upn+1T ∈ dp

m · Epn+m+1

2n . Moreover, L′n/Kn is abelian, so (dup

n+1)ω

∗

n ∈Ep

n+m+1

2n . We now apply (50) to u and d independently:

upn+1ω∗

n ≡ utωnpn+1 ≡ dpmtNn ≡ dtpm+n−κmod Ep

n+m+1

2n , t ∈ Λ×

while dω∗

n = dcpn+1

, c ∈ Z×p . Combining the two, we obtain

(dupn+1

)ω∗

n ≡ d(c+tNnpm−(κ+1))pn+1mod Ep

n+m+1

2n .

Therefore L′n/Kn can not be abelian for m ≥ (n − k)/2 > (κ + 1), i.e.

n ≥ 3(κ + 1).For n sufficiently large, there is thus no p - ramified extension Ln/Fn of

degree pm,m > (n − κ)/2, with group annihilated by T ∗ and with Ln/Kn

abelian. Since this holds for all extensions above Φn, it follows that D(K)


independent cyclic unramified extensions in Φn have no cyclic continuationsover Kn that are p - ramified over Φn

7.It follows that p-rk(Gal(Ωn,T ∗/Hn)) = r2 − 1 − D(K), which completes

the proof of (41).

The main Theorem 1 follows from the Proposition, as explained above:since ΩT ∗ ∩ ΩE′ has group of rank r2 + s − 1−D(K) while ΩT ∗ ⊂ ΩE′ and

by Lemma 3, Zp-rk(Gal(ΩT ∗/K∞)) = r2 + s − 1, it follows that D(K) = 0.The Example 2 is continued by an illustration of the main steps of the prooffor the case K = Q[ζp], which was known from the Theorem of Baker andBrumer.

PPPPPP

❳❳PPPPPPP

K∞

KQ

D(K)

F

Fn

Φn

ΩnHn

LnKn

Kn+mL = Kn+m[e

1/pn+m ]

L∞

F∞

Fig. 2: The unamified (marked: - ) and ramified(marked =) extensions at finite levels.

Ωn,T ∗

6. Consequences

The results in the previous section give a complete picture of the T andT ∗ parts of the class groups and p - abelian extensions in the cyclotomic Zp- extension of arbitrary galois fields.

The following Conjecture is a natural generalization of the GreenbergConjecture to arbitrary fields:

7Note also that we might consider F,L as extensions of Ωr,n = Kn[Π1/pn+1

], since thisextension is independent of the Leopoldt Conjecture for all choices of Π. In this case, wemay even assume d1 ∈ E(K)θM , so Ln would also be unramified.

38 PREDA MIHAILESCU

Conjecture 1. Let K be a number field, K∞ its cyclotomic Zp - extension

and H = Hϕ(A)∞ . Then

H ⊂ ΩE .(42)

Note that Gal(H/K∞) is a Λ - torsion module by definition, so (42) hasno information on µ(K). The Conjecture µ(K) = 0 for the cyclotomic Zp -extension of arbitrary number field is thus independent of the (42).

For the case when K is totally real, we may adjoin roots of unity to K andfind that H ∩ ΩE = K∞, since Gal(ΩE/K∞)1+ = 1: thus the Conjecture1 plus µ = 0 is equivalent to Greenberg for CM fields K. If K is abelian,the Greenberg Conjecture is equivalent to (42), since µ = 0 was proved byFerrero and Washington [25], §7.5.

If f(T ) divides the characteristic polynomial of Gal(H/K∞), we say thatGreenberg’s Conjecture holds for the f(T ) - part of A, if Hf ⊂ ΩE, with Hf

in Definition 6. With this we have proved:

Theorem 5. Let K be a complex galois extension. Then Greenberg’s Con-jecture holds for the T and T ∗ parts of A and A/AT

∗

is finite.

Proof. The Gross - Kuz’min Conjecture says that A′T = 1 and AT ∗ = 1

is Leopoldt’s Conjecture. The last also implies that M ⊂ ΩE and thereforeAT ∼ B ∼ Gal((HT ∩ ΩE)/K∞); but HT ⊂ M ⊂ ΩE, which confirms alsothe Greenberg Conjecture for the T - part.

Note that B is not necessarily finite for complex fields K. In the casewhen K is CM we give below a precise description of AT , since in this caseAT ∼= B−, canonically.

Proposition 6. Let K/Q be a CM galois extension and Kn,K∞, An, A bedefined as previously. Let ℘ ⊂ O(K+) be any prime above p and let

g′ =

0 if ℘ is unsplit in K/K+,

g(℘) = [K+ : Q]|D℘|

otherwise;

here D℘ ⊂ Gal(K+/Q) is the decomposition group of ℘. Then the moduleB− is a free Zp - module of rank g′.

Proof. Since ess. p-rk(A/AT ) = Zp-rk(B), it suffices to consider primes℘ ⊂ K which ramify in ideals ℘n ⊂ Kn with diverging orders in the idealclass group. We know that A/(AT

∗

) is finite and since K is a CM field,this implies that B+ is finite and ess. p-rk(B) = Zp-rk(B−), where B− isa torsion-free module. It follows in particular that B is finite if the primesabove p are unsplit in K/K+, which explains that g′ = 0 in this case.

Now if ℘ splits inK/K+, we have Zp-rk(B−) = Zp-rk ((Gal(M− ∩HT )/K∞),so it suffices to evaluate the second rank. Let L ⊂M−∩HT ) be a Zp - exten-sion of K∞. Then Lτ℘/Kτ℘[µ∞] is either a trivial extension or the uniqueunramified Zp - extension of Kτ℘: its group is in particular G - invariant.For each τ ∈ G there is exactly one such extension, and since for each


ν ∈ Gal(M−/K∞) we have ν1+ = 1, we see from the definition of g′ thatthere are at most g′ independent extensions L with this property. In orderto verify that there are exactly g′ extensions L, we observe that M− = M asa consequence of Leopoldt’s Conjecture and thus Zp-rk(Gal(M−/K∞)) = r2and Zp-rk(Gal(Mτ℘/Kτ℘[µ∞] = [Kτ℘ : Qp]. In particular, there is a G -

invariant local subextension for each τ ; the count now follows from ν1+τ

for ντ ∈ ∆ an automorphism which generates Gal(Mτ℘/Kτ℘). ThereforeZp-rk(B) = g′, which completes the proof.

In [16], Jaulent generalizes this expression to fields K which do not containthe p-th root of unity.

❩❩❩❩❩❩❩❩❩❩❩❩

Ω

Ωr

MHT

H∞

H

Q K

K∞[A1/p∞ ]

K∞[(A′)1/p∞

]

Fig. 3: The a posteriori field diagram, for D(K) = 0

HE ΩE

7. Appendix A: Basics

The first section of this Appendix is a technical extension of §2.1 anddeals with the difficult problem of defining annihilators for Zp[G] - modulesin the non commutative case.

7.1. Annihilators, supports and components in the non-commutative

case. If G is a finite group and X a Zp[G] - module, then X is quasi - cyclic

if there is some x ∈ X \ Xp such that [X : xZp[G]] < ∞, or, equivalently,

X = xQp[G]. Then x is called a generator for X; if x ∈ Xp, we maydenote it by weak generator for X and such generators may occur whenconsidering norm coherent sequences of modules. The module is cyclic , ifthe index is 1. Let q = pvp(|G|). For any idempotent α ∈ Qp[G], we haveqα ∈ Zp[G]. The Theorem of Maschke [3], p. 116 asserts that each submod-ule A ⊂ Qp[G] is a direct sum of irreducible modules. Furthermore, one seesfrom the proof of the theorem, that there is an idempotent α ∈ Qp[G] such

40 PREDA MIHAILESCU

that A = (α) = αQp[G]. Modules and their idempotents are too numerous,so we will identify isomorpy classes of modules. It is then known that A isisomorphic to an algebra of column - matrices over Qp; more precisely, wehave the following standard fact ([3], Chapters 5 and 6).

Fact 5. Let G be a finite group. Then Qp[G] has a canonic decomposition asa sum of bilateral submodules Qp[G] =

∑ψ 1ψQp[G], where ψ are the central

irreducible idempotent of Qp[G], associated to the irreducible characters ψ.If ψ is any irreducible character and fψ is the dimension of the associatedrepresentation, then there is canonic decomposition

1ψQp[G] = ⊕fψi=1Aψ,i,

Qp[G] =⊕

ψ 1ψ ⊕fψi=1 Aψ,i

(43)

with Aψ,i being the modules of column fψ × fψ matrices having 0 in all butthe i-th column. Furthermore, every submodule A ⊂ Qp[G] is isomorphicto the direct sum of a collection of Aψ,i’s. In particular, there are exactly

2f isomorphy classes of modules in Qp[G], where f =∑

ψ fψ. Since Qp is

not algebraically close, the Parceval formula |G| =∑ψ f2ψ may not hold over

Qp.

Using this fact, we shall develop the notions of isomorphy classes of Qp[G]-modules, their supports, annihilators, generating idempotents and compo-nents. This is, among other, of importance for dealing with fields which arenot CM and defining a canonic orthogonal pairing on U(K), with respectto which the Leopoldt involution can be used in the cases of interest, inorder to define ’plus’ and ’minus’ parts of cyclic Zp[G] - modules, whichare reminiscent of complex conjugation. The proof of the Gross - Kuz’minConjecture for non CM fields is the only place where we explicitly need thisconstruction. The reader who wishes to proceed quickly to the proof of themain Theorem may jump this and related parts and follow the line of theproof for the simpler CM case, in which we may distinguish between plusand minus parts of modules by means of the complex conjugation.

The relation between modules and idempotents is explained in detail inthe following

Proposition 7. Let R ⊂ Qp[G] be some non trivial submodule. If α, β ∈Qp[G] are two idempotents, they both generate the same module R = αQp[G] =βQp[G] iff there is a nilpotent ν ∈ Qp[G] with

ν2 = 0, να = 0, and αν = α.(44)

Furthermore, if R′ ⊂ Qp[G] is an other submodule, then R′ ∼= R iff thereis a unit u ∈ Qp[G]

× and an idempotent α ∈ Qp[G] such that

R = αQp[G], R′ = α′ = Qp[G] with α′ = u · α · u−1.(45)

Proof. Let α, β be like in the hypothesis and ν = α− β. Since β ∈ αQp[G],there is a u ∈ Qp[G] with β = αu and since α ∈ βQp[G], it follows that u is


a unit. Using the idempotent property, we have

βα = ββu−1 = βu−1 = α, αβ = ααu = αu = β, thus

ν2 = (α− β)2 = α− αβ − βα+ β = 0,

να = (α− β)α = α− βα = 0, αν = α− αβ = α− β = ν.

Conversely, if R = (α)Qp[G] and ν has the properties in (44), then β =α − ν = α − αν = α(1 − ν) and u = (1 − ν) = (1 + ν)−1 ∈ Qp[G]

×, soβQp[G] = R. We also have

β2 = α(1− ν) · α(1− ν) = α · α(1− ν)− να(1− ν) = α(1− ν)− 0 = β,

which confirms the first claim.Let α ∈ Qp[G] be an idempotent andR = αQp[G]; for any u ∈ Qp[G]

×, β :=uαu−1 is an idempotent and we claim that R′ = βQp[G] ∼= R. Indeedthe map ψ0 : Qp[G] → Qp[G], x 7→ ux is an automorphism of Qp[G] andinduces an isomorphism R → R′ : αy 7→ βψ0(y). Conversely, if R =αQp[G] ∼= R′ = βQp[G] and ψ : R → R′, then ψ(α) = β · v; furthermore,βv ∈ (R′)×, since ψ is surjective. Let u ∈ Qp[G]

× be such that βu = βvand define α′ = uαu−1. Let α′ = uαu−1; by definition of ψ, it follows thatψ(1−α) = βu(1−α) = 0 = β(1−α′), so β ∈ α′Zp[G], with α′ = uαu−1 andby comparing ranks, the two modules must be isomorphic. The propositionimplies that β − α′ verifies (44) and there is a further unit u′ = u(1 − ν)(with ν = 1 also possible), such that β = u′α(u′)−1, which completes theproof.

As a consequence,

Corollary 3. Let A = (α)Qp[G] ⊂ Qp[G] be an irreducible module withα ∈ I0. Then A ∼= Aψ,i for some central irreducible idempotent ψ and1 ≤ i ≤ fψ, and there is a canonical representant α(c) of the image of α inI/ ∼=, such that αQp[G] = Aψ,i. In general, every isomorphy class α ∈ I/ ∼=has exactly one canonic representant α(c) with

α(c)Qp[G] = ⊕(ψ,i)∈IAψ,i, for someI ⊂ (ψ, i) : ψ irred. centr. idpt. and 1 ≤ i ≤ fψ.

(46)

Proof. See [3], Chapter 5, Theorem 6 for the isomorphism of A with anelementary module of column matrices. The existence and unicity of thecanonic representant follows from the map between isomorphy classes ofidempotents and classes of isomorphic modules. Finally, the statement forgeneral modules follows from Maschke’s theorem and the decomposition inirreducible modules.

In view of the above proposition and its corollary, we define the followingrelations among idempotents:

42 PREDA MIHAILESCU

Definition 7 (Classes of idempotents). Let α, β ∈ Zp[G] be two idempotents.Then

α ≡ β iff ν := α− β verifies (44),

α ∼= β iff there is a u ∈ Zp[G]× with β = uαu−1.

We say that α is congruent to β if α ≡ β and they are isomorphic, if α ∼= β.Let I0 ⊂ Zp[G] be the set of all idempotents. Then I := I0/ ≡ is in one - to -one correspondence with the submodules of Zp[G], while I/ ∼= parameterizesthe isomorphism classes of submodules of Zp[G]. By abuse of language, weshall identify α ∈ I0 with its class in I. The canonic representant of α ∈ Iis α(c) and it verifies (46)

For α′ = uαu−1 we have 1−α′ = u(1−α)u−1; therefore complementaritycan be defined also for isomorphy classes of idempotents. Let I ⊂ Qp[G]be the set of all idempotents and I/ ∼= the set of isomorphy classes ofidempotents and denote the class of α ∈ I by α. Then we may build

complements on isomorphy classes by α⊤ = (1− α) and the notation 1− αlays at hand for this complement; in general we may write 1−a for a ⊂ I anisomorphisms class. For intersections and direct sums, the definition needsadditional clauses, in order to assure compatibility under ’conjugation’ withunits.

Let X be a Qp[G] - module; for each x ∈ X we write x⊤ = a ∈ Qp[G] :xa = 1 for its annihilator module. Defining the annihilator of X in generalis more delicate, but we may use the canonic idempotents; we restrict to thecase when X is a cyclic Qp[G] - module of rank r ≤ |G| as a Qp - module.We write X0 ⊂ X for the set of all generators of X as a Qp[G] - module

and X = xpZ : x ∈ X0. By abuse of notation, we identify elements x ∈ Xwith their classes xp

Z ∈ X. If X is a quasi - cyclic Zp[G] - module, then

X = x ∈ X \ Xp : xpZ ∈ X; therefore X contains uniquely determined

elements of X in this case. We would expect from a proper definition ofthe annihilator module X⊤ ⊂ Qp[G], that it is a space of dimension |G| − rover Qp. This is in general not the case, if we use the naive definition

X⊤ = a ∈ Qp[G] : xa = 1, for all x ∈ X. If γ ∈ X is any generator, then

γ⊤ verifies naturally the rank condition, but it depends on the choice of γ.If G is commutative, then the annihilator X⊤ is well defined and the supportis

X⊥ = ⊕1χ:1χX 6=1 1χX,

where 1χ are central idempotents belonging to irreducible characters χ; thisyields an embedding X → Zp[G].

In the non commutative case, for any generator γ′ there is a unit u ∈Zp[G] such that γ′ = γu; therefore the annihilator modules γ⊤ ∼= (γ′)⊤ assubmodules of Qp[G]. There is an isomorphy class IX ∈ I/ ∼= and a map

φX : IX → X such that φX(α)⊤ = αQp[G]; in view of (44), φX(α) = φX(β)

iff α ∼= β. Furthermore, for each α ∈ Iα, we have (φX(α)⊤)⊤ = (1 −


α)Qp[G] and thus X ∼= αZp[G] for all α ∈ IX . Moreover, there is a canoniccanonic annihilator idempotent α(c) ∈ I, which corresponds to a uniquely

determined generator x ∈ X, such that x⊤ = α(c)Qp[G]. We shall say that

X⊤ = α(c)Qp[G] is the canonic annihilator of X, and write also α(c) = α⊤.We note that it is defined via a well determined choice of a generator x ∈X. More generally, one may view annihilators also a isomorphy classes ofidempotents together with a map to the set of generators of X. The supportis then X⊥ = (X⊤)⊤ and we write α⊥ for its canonic generator. One verifiesthat for each elementary module A = Aψ,i ⊂ Qp[G] we have AX 6= 0 ⇔AX⊥ 6= 0. By definition, we have X⊤⊕X⊥. Like the annihilators, supportscan also be viewed as isomorphy classes of modules together with a map tothe generators of X.

Finally, for α(c) the canonic idempotent of some module A ⊂ Qp[G], wedefine the α(c) - component of X by letting β be a canonic idempotent

generating α(c)Qp[G] ∩X⊥ and

Xα = β ·X, X = Xα ⊕X1−α.(47)

The following computation explains the decomposition of X in complemen-tary components: let x ∈ X be the generator such that x⊥ = α⊥ and supposethat A ⊂ X⊥. Since A is a sum of elementary modules, it has a complementA′ which is also sum of elementary modules and we let B = A′ ∩X⊥ and βbe a canonic idempotent generating B. Suppose that Xα ∩ X1−α 6= 0 andlet t ∈ Qp[G] be such that t = xαu = xβv ∈ Xα∩X1−α. Then αu+βv ∈ X⊤

and by choice of α, β, it follows that u, v = 0 and thus t = 0.The next definition synthesizes the above constructions:

Definition 8 (Annihilators and Supports). Let X be a cyclic (right) Qp[G] -

module. Then there are two canonic idempotents α⊤ and α⊥ and a uniquelydetermined x ∈ X such that x⊤ = α⊤Qp[G] and x

⊥ = α⊥Qp[G]. Further-

more, if IX ⊂ I is the isomorphy class of α⊤, there is a map φX : IXX andthe two are related by

φX(β)⊤ = βQp[G], ∀β ∈ IX .(48)

If X is a cyclic Zp[G], then the pair (IX , φX) is defined with respect to Xand the relation (48) becomes

φX(α)⊤ = αQp[G] ∩ Z[G], ∀α ∈ IX .

Finally, if A ⊂ X⊥ is an elementary module with generator α, then thereis a well defined A component Xα with support X⊥

α = A and its comple-ment in X is X1−α. Components may be defined as isomorphy classes ofsubmodules, using the map φX above.

Since supports and annihilators come with canonic choices of generatorsof X, one may induce maps between two cyclic Qp[G] - modules havingthe same support and annihilators. Furthermore, a decomposition of Qp[G]in components induces decompositions of all cyclic Qp[G] - modules. In

44 PREDA MIHAILESCU

the case of Zp[G] - modules X, one defines the annihilators, supports and

components with respect to X and then takes intersections with Zp[G].In the case of quasi - cyclic Zp[G] - modules X, there are always p -

powers involved in the passage from classes xpZ ∈ X to x ∈ X; of course,

these powers depend strongly upon X, its deviation from a cyclic Zp[G] andvp(|G|). The next lemma describes the deviation of quasi - cyclic modulesfrom cyclicity.

Lemma 13. Let X be a quasi - cyclic module and x′ a generator. Thenthere is a further generator x depending on x′ and such that Xq ⊂ xZp[G].In particular, if vp(|G|) = 1, then the all quasi - cyclic modules are cyclic.

Proof. Let α ∈ I generate the annihilator x⊤ and (1−α)Qp[G] = ⊕iβiQp[G]

be a decomposition of x⊥ in a direct sum of irreducible modules, generatedby the idempotents βi. Let x

′i = (x′)qβi ∈ X and let xi ∈ X \Xp such that

x′i = xpei

i for some ei ≥ 0. We obviously have X = ⊕ixQp[G]i and X = X∩X.

Furthermore, in every component, xqQp[G]i ⊂ X. Letting x =

∏i xi, we see

that xqβiZp[G] ⊂ X and Xqβi ∩X ⊂ xqβiZp[G], since βiZp[G] are irreducible.

If q = 1, it follows that X = xZp[G], so X is cyclic.

As an important consequence, when p is not coprime to |G|, the decom-position of X in disjoint components works only up to a power of p.

7.2. Arithmetic in Λ. We gather in this section several elementary com-putations related to the arithmetic in Λ. Since the numbering of the in-termediate fields starts with K0 = Kκ (see §2.3 Definition 3 for the precise

definition of κ), we have ωn = (T + 1)pn−κ − 1. Then

Nn := NKn/K0= ωn/T and Nm,n = NKm/Kn = ωm/ωn,(49)

for m > n ≥ κ. The action of the Iwasawa involution on ωn is seen from(T + 1)p

n−κω∗n + ωn = (pκ − 1)p

n−κ − 1; it follows that:

ωn + tω∗n = pn−κc, t ∈ Λ×

n , c ∈ Z×p .(50)

We shall use the following observation of B. Angles [4], Lemma 2.1, (2): letm = κ+ l and l′ = [l/2]. Then

ωm(T ) = TNm ∈(pl

′

, T pl′+1).

We may thus choose a, b ∈ Λm with a ∈ Λ×m such that

N∗m = apl

′

+ bNl′+1.(51)

For the Fp - modulesRn = Λn/pΛn we have the isomorphismRn ∼= Fp[T ]/(T )pn−κ

,so the image of T is a nilpotent in these modules. Furthermore, T ∼=cT ∗ mod pΛ, c ∈ Λ∗. These facts will be used below for the study of mod-ules of infinite p - rang. A further set of identities which we use repeat-edly is the following: let ϑ generate Gal(Kn+1/Kn) and θ = ϑ − 1. Then


Gal(Kn+1/Kn) ∼= Fp[X]/Xp via ϑ − 1 7→ X; some simple arithmetic yieldsthe following useful expansions:

Nn+1,n = p+ θ ·(

p∑

k=2

(p

k

)θk−1

)

= p+ θh(θ) = p+ θ ·((

p

2

)+ h1(θ)

)(52)

= p · h3(θ) + θp−1, h1(θ), h2(θ) ∈ Fp[θ], h3(θ) ∈ (Fp[θ])×.

The following Lemma is a general result on duality:

Lemma 14. Let X be a Λ - module, Xn = XΓn be the submodule fixed

by ωn and Xn,T ∗ = x ∈ Xn : xT∗ ≡ 1 mod Xpn−κ

n .; we assume thatNm,n : Xm → Xn is surjective. If µn ⊂ X is the maximal submodule of

exponent pn−κ which is annihilated by T ∗ modulo Xpn−κn , then

Xn,T ∗/Xpn−κn = (XN∗

nn · µn ·Xpn−κ

n )/(Xpn−κn ).(53)

Moreover, if X is quasi - cyclic of Λ - rank r as a Λ - module, then for n > κ,Xn,T ∗/µn are quasi - cyclic finite Zp[G] - modules of p - rank r and subexpo-

nent p(n−κ)/2. They form an injective system with limit XT ∗ = lim−→Xn,T ∗/µnwhich is a quasi - cyclic Zp[G] - module of rank r. Furthermore, if ιn,m :

Xn → Xm for m > n, then ιn,m(Xn,T ∗) ⊂ Xpm−n

m,T ∗ .

Proof. Apply (50): if x ∈ XN∗

nn , then xT

∗

= yω∗

n = yc′pn−κ , c′ ∈ Λ×. Since

µn ⊂ Xn,T ∗ by definition, this shows the inclusion ’⊃’. For the other direc-

tion, consider x ∈ Xn with xT∗

= ypn−κ

; then xT∗N∗

n = xc′pn−κ = yN

∗

n·pn−κ

and thus(x/(y(c

′)−1N∗

n))pn−κ

= 1, so x = y(c′)−1N∗

n · ξ, with ξpn−κ = 1; from

xT∗ ∈ Xpn−κ it follows that ξ ∈ µn. This implies also ’⊂’ and completes the

proof of (53).We now prove the properties of XT ∗ . Let Gn = Gal(Kn/Q), Xn =

Xn,T ∗/µn and g = |G|. An element α ∈ Z[Gn] acting onXn has the followingdual development in the group ring:

α =

g−1∑

i=0

Ai(T∗) · τi, τi ∈ Gal(K/Q),(54)

where the Ai ∈ Z[x] are polynomials of degree deg(Ai) < ppn. Let α0 =∑g−1

i=0 Ai(0)τi. We show that p-rk(Xn,T ∗/µn) = r.We shall construct a subset D′ ⊂ Xm such that (Xp

mD′)/(Xpm) is an

Fp - space of maximal rank r. Let δn ∈ Xn be a generator for the Λ -cyclic modules Xn and δ0 = Nn(δn), a power of a generator for X0; consider

H ⊂ G\1, a maximal subset such that δR[H]0 ⊂ X0 is a free Zp - module of

rank r−1. LetDn = δσn : σ ∈ H∪1 be a system of relative generators forXn/X0 andDn ⊂ Xn be their Zp - span; the identity automorphism accounts

46 PREDA MIHAILESCU

for the pre-image N−1Kn/K

(1) ⊂ Xn of 1 in Xn. The system Dn has p - rank r;

considerDN∗

nn ⊆ XN∗

nn . From (54) we deduce that p-rk(D

N∗

nn /(D

N∗

nn )p

n−κ) = r;

a fortiori, p-rk(Xn) ≤ r: it may happen that DN∗

nn ⊂ Xpn−κ

n . We show thatthis is not the case for sufficiently large n, and the two ranks are then equal.

Let thus Tn = XN∗

nn /D

N∗

nn be the torsion and t1, t2, . . . , ty ∈ Tn be a

minimal system of generators with y ≤ r and decreasing orders in the torsiongroup Tn, so ord(t1) ≥ ord(t2) ≥ . . . ≥ ty. We shall identify the ti with a

set of representatives in Xn and let d′i = tord(ti)i ∈ D

N∗

nn , i = 1, 2, . . . , y. Then

d′i are Zp[G] - independent; we may choose d′j ∈ DN∗

nn , y < j ≤ r such that

XN∗

nn = Span(ti, d

′j)R, 1≤i≤y<j≤r.

The set F = ti : 1 ≤ y ∪ d′j : y < j ≤ x2 is then a set of Zp[G]

- generators for XN∗

nn and this shows that X

N∗

nn has the p -rank r. By con-

struction, (Span(F ) · Xpn)/X

pn has also the rank r as an Fp - vector space

and thus,

p-rk((XN∗

nn ·Xpn−κ

n )/Xpn−κ

n

)= p-rk(XN∗

nn ) = r.

We finally show that the exponents of Xn := XN∗

nn /X

pn−κN∗

nn are diverging.

For this we use (51). Let x ∈ Xn \ Span(F )p, so x = zN∗

n , z ∈ Xn \Xpn, and

set l′ =[n−κ2

]. The formula (51), in which we choose a to be a unit, implies

that x 6∈ Xpl′+1

n and therefore x generates a cyclic group of order at leastpl

′

. Since (Span(F )Xpn)/X

pn has rank r, it follows that there is a subgroup

Wn ⊂ Xn with Wn∼= (Cpl′ )

r, thus Wn has p-rank r and subexponent l′ =[n−κ2

]. Finally we have to show that Xn form a projective system. Let

xn+1 ∈ Xn+1 and xn = Nn+1,n(xn+1). In view of (50), we have N∗n+1,n ≡

c1Nn+1,n mod pn−κ with c1 ∈ Λ×, and thus

N∗n+1 = N∗

n+1,n ·N∗n ≡ c1Nn+1,n ·N∗

n mod pn−κ.

If follows that xN∗

n+1

n+1 ≡ xc1N∗

nn mod Xpn−κ

n and thus Xn,T ∗ form an injec-tive system. One verifies with the same methods that ιn,n+1(Xn,T ∗) =Nn+1,n(Xn+1,T ∗)p, where Nn+1,n is regarded as an endomorphism of Xn+1;this induces also a projective structure which is best understood in terms of

“radicals” X1/pn−κ

n,T ∗ .

7.3. Units.

Proof of Fact 1. Let e, f, g denote as usual, the ramification index, the de-gree of the residual fields and the splitting index of the primes above p.The polynomial ι(f(X)) is separable over Qp and splits in g polynomials ofdegree ef . Thus Kp = Qp[X]/(ι(f)) is the product of g isomorphic local,unramified extensions of degree ef . Each completion K℘

∼= K is a ramifiedextension of degree e of the unramified extension K/Qp of degree f .

It follows from the Chinese Remainder Theorem that ι : Q → Qp extendsto an embedding ι : K → Qp[X]/(ι(f)) and that the image of K is dense in


Kp. By continuity, the galois action of G extends to Kp and commutes withthe embedding.

Indeed, for any t ∈ Kp there is an h ∈ Qp[X] of degree deg(h) < deg(f),such that t = h(ι(α)). Since Q is dense in Qp, there are polynomials hn ∈Q[X] which approximate h to the power pn, so limn→∞ ι(hn) = h; settingtn = hn(α) ∈ K, we also have ι(tn) = ι(hn(α)) → h(ι(α)) = t; hence K isdense in Kp. For any σ ∈ H, we define σ(t) = h(ι(σ(α)). This action is welldefined and commutes with the embedding, since for t ∈ K we have

ι(σ(t)) = ι(h(σ(α))) = h(ι(σ(α)) = σ(ι(t))).

7.4. Λ - modules, radicals and duality. The following elementary lemmawill be used for the proof of Proposition 1:

Lemma 15. Let A and B be finitely generated abelian p−groups denotedadditively, an let N : B → A, ι : A→ B two Zp - linear maps such that:

1. N is surjective;2. The p−ranks of A and B are both equal to r and |B|/|A| = pr.3. N(ι(a)) = pa,∀a ∈ A and ι is rank preserving;

Then ι is injective, ι(A) = pB and ord(x) = p · ord(Nx) for all x ∈ B.

Proof. We start by noting that for any finite abelian p - group A of p - rankr and any pair αi, βi; i = 1, 2, . . . , r of minimal systems of generators thereis a matrix E ∈ Mat(r,Zp) which is invertible over Zp, such that

~β = E~α.(55)

This can be verified either by tensoring with Qp, or directly by extendingthe map αi 7→ βi linearly to A and, since (βi)

ri=1 is also a minimal system of

generators, deducing that the map is invertible, thus regular. It representsa unimodular change of base in the vector space A⊗Zp Qp.

The maps ι and N induce maps

ι : A/pA→ B/pB, N : B/pB → A/pA.

From 1, we see N is surjective and since, by 2., it is a map between finitesets of the same cardinality, it is actually an isomorphism. But 3. impliesthat N ι : A/pA→ A/pA is the trivial map and since N is an isomorphism,ι must be the trivial map, hence ι(A) ⊂ pB.

Assume now that ι is rank preserving and let bi, i = 1, 2, . . . , r be aminimal set of generators of B: thus the images bi of bi in B/pB form anFp - base of this algebra. Let ai = N(bi); since p-rk(B/pB) = p-rk(A/pA),the set (ai)i also forms a minimal set of generators for A. We claim that|B/ι(A)| = pr.

Pending the proof of this equality, we show that ι(A) = pB and ι isinjective. Indeed, we have the equality of p- ranks:

|B/pB| = |A/pA| = |B/ι(A)| = pr,

48 PREDA MIHAILESCU

implying that |pB| = |ι(A)|; since ι(A) ⊂ pB and the p - ranks are equal,the two groups are equal, which is the first claim. The second claim will beproved after showing that |B/ι(A)| = pr. Since |B|/|A| = pr, it follows that|A| = |ι(A)|, so ι is injective.

Let S(X) denote the socle of the finite abelian p - group X. There isthe obvious inclusion S(ι(A)) ⊂ S(B) ⊂ B and since ι is rank preserv-ing, p-rk(A) = p-rk(S(A)) = p-rk(B) = p-rk(S(B)) = p-rk(S(ι(A))), thusS(B) = S(ι(A)). Let (ai)

ri=1 be a minimal set of generators for A and

a′i = ι(ai) ∈ B, i = 1, 2, . . . , r; the (a′i)ri=1 form a minimal set of generators

for ι(A) ⊂ B. We choose in B two systems of generators in relation to a′iand the matrix E will map these systems according to (55).

First, let bi ∈ B be such that peibi = a′i and ei > 0 is maximal among allpossible choices of bi. From the equality of socles and p - ranks, one verifiesthat the set (bi)

ri=1 spans B as a Zp - module; moreover, ι(A) ⊂ Bp implies

ei ≥ 1. On the other hand, the norm being surjective, there is a minimalset of generators b′i ∈ B, i = 1, 2, . . . , r such that N(b′i) = ai. Since bi, b

′i

span the same finite Zp - module B, (55) in which ~α = ~b and ~β = ~b′ defines

a matrix with ~b = E · ~b′. On the other hand,

ι(~a) = ~a′ = Diag(pei)~b = Diag(peii )E · ~b′,The linear map N : B → A acts component-wise on vectors ~x ∈ Br.

Therefore,

N~b = ~Nbi = N(E~b′) = N

(∏

j

b′P

j ei,jj )ri=1

=

∏

j

(Nb′j)P

j ei,j

r

i=1

=

∏

j

(aj)P

j ei,j

r

i=1

= E(~a).

We obtain thus two expressions for N~a′ as follows

~Na′ = p~a = pI · ~a= N

(Diag(pei)~b

)= Diag(pei) ·N(~b) = Diag(pei) ·E~a, and thus

~a = Diag(pei−1) · E~aThe aj form a minimal system of generators and E is regular over Zp; there-

fore ~(α) := (αj)rj=1 = E~a is also minimal system of generators of A and the

last identity above becomes

~a = Diag(pei−1) · ~α.If ei > 1 for some i, then the right hand side is not a generating system of Awhile the left side is: it follows that ei = 1 for all i. Therefore |B/ι(A)| = pR

and we have shown above that this implies the injectivity of ι.


Finally, let x ∈ B and q = ord(Nx) ≥ p. Then qN(x) = 1 = N(qx), andsince qx ∈ ι(A), it follows that N(qx) = pqx = 1 and thus pq annihilatesx. Conversely, if ord(x) = pq, then pqx = 1 = N(qx) = qN(x), andord(Nx) = q. Thus ord(x) = p · ord(Nx) for all x ∈ B with ord(x) > p. Iford(x) = p, then x ∈ S(B) = S(ι(A) ⊂ ι(A) and Nx = px = 1, so the lastclaim holds in general.

This small exercise in linear algebra avoids a deeper investigation of Mas a sum of irreducible Λ - modules, and the afferent pseudo-isomorphismswhich may arise. We can non give the

Proof of Proposition 1 . We assume without restriction of generality that0 is the least integer n for which |Mn| = |Mn+1|, for point 1., respectivelyp-rk(Mn) = p-rk(Mn+1), for point 2. Let

Y = x = (xn)n∈N ∈M : fn,0(xn) = 1 for all n ≥ 0 ⊂M.

One verifies from the definition that Y is a Λ - submodule. For n > 0 themap fn,0 is surjective with Ker (fn,0) = x ∈ Mn : fn,0(x) = 1 = ν0,nYand we thus have a commutative diagram in which Mn → M0 is inducedby the map fn,0 while the horizontal isomorphism are deduced from theprevious remark:

Mn∼= M/ν0,nY

↓ ↓M0

∼= M/Y.(56)

For the first point we assume |M1| = |M0|. Then M1 → M0 is an isomor-phism; therefore ν0,1Y = Y . Since8 ν0,1 ∈ (p, T ) ⊂ Λ, the unique maximalideal, and since Y is finitely generated over Λ, it follows from Nakayama’slemma that Y = 0. Consequently M ∼= M0 is finite and Mn

∼= B0∼= M for

all n ≥ 0. This proves the assertion 1.Suppose now that p-rkM1 = p-rkM0. Then M1/pM1

∼= M0/pM0 andthus M/(ν0,1Y + pM) ∼=M/(Y + pM) and ν0,1Y + pM = Y + pM . LettingZ = (Y + pM)/pM , we have

ν0,1Z = (ν0,1Y + pM)/pM = (Y + pM)/pM = Z.

By Nakayama’s lemma we conclude that Z = 0 and Y ⊂ pM . Thereforep-rk(Mn) = p-rk(M/ν0,nY ) = p-rk(M/(ν0,nY + pM)) = p-rk(M/pM) =Zp-rk(M) for all n ≥ 0. By Iwasawa’s formula, for n sufficiently large wehave

|Mn| = pµpn+λn+ν ,

and since the rank stabilizes, we see that µ(M) = 0 and |Mn+1|−|Mn| = pλ.This proves assertion 2.

8We follow here Fukuda [11] and use Iwasawa’s notation νn,m = Nm,n = ωm/ωn form > n ≥ 0.

50 PREDA MIHAILESCU

For point 3. we need Lemma 15. We have seen in the proof of the Lemma,that ι(Mn) ⊂ pMn+1 in general, while in the case when ι is rank preservingwe have equality. The kernel Ker (ι) is the capitulation kernel and has thusexponent p. Therefore, ι is rank preserving iff the subexponent of Mn (seeDefinition 1) is larger than p. Suppose thus that ι is rank preserving andι(Mn) = pMn+1 and let N ′ = ι N .

For proving xp = ι(N(x)) = N ′(x), we make explicite use of the structureof Γ. Let t = ωn = (T + 1)p

n − 1 and

N ′ = NKn+1/Kn = p+ t · v = p+ t(

(p

2

)+ tw)), v, w ∈ Z[t],

as follows from the Newton development of N = (t+1)p−1t (see also 52). By

definition, t annihilates ι(Mn) and x0 := xp ∈ ι(Mn) from our assumption.Thus (xt)p = xt0 = 1 and xt ∈ S(Mn+1) = S(ι(Mn)) ⊂ ι(Mn), as shown inthe proof of the Lemma 15. In view of the above development then

N ′x = xp · (xt)(p2)+tv = xp,

which completes the proof.Suppose now that ι is not rank preserving, so the subexponent is p for

all n. Then M = M ′ ⊕M ′′, with M ′′ a module of exponent p and M ′ ofsubexponent larger than p. The result holds for M ′ by the above, while forM ′′ we have xp = 1 for all x ∈ Mn+1. Consequently, Mn+1 is an Fp[T ] -module of finite rank annihilated by the image of ωn+1 in Fp[T ], which is

ωn+1 = (T + 1)pn+1 − 1 mod pZp[T ] = T p

n+1. The rank being finite, we see

that xt = 1 too, for sufficiently large n. Consequently N ′(x) = xp for all Mof finite rank and n sufficiently large.

The first point is concerned with pure capitulation. Indeed, we deducefrom N ′(x) = xp and |Mn| = |Mn+1| that p-rk( Ker (ι)) = p-rk(Mn) = R,so the whole socle of Mn must capitulate. In the second case, assuming inaddition that R = λ(M) we have vp(|Mn|) = µpn + λn + ν and since theranks are bounded, µ = 0. From vp(|Mn+1|) − vp(|Mn|) = λ and |Mn+1| =p|ι(Mn)|, so |Mn+1| − |ι(Mn)| = pR. In the case when R = λ, it follows that|ι(Mn)| = |Mn| so the two groups are isomorphic, ι is injective and M is aWeierstrass module. This completes the proof.

We see that the growth of infinite Λ - modules of finite rank is verysimple, at levels beyond the rank stabilization. The whole taxonomy ofthese modules is established by the growth prior to this stabilization and thismay be very complex. Even the question whether capitulation in Weierstrassmodules can be excluded also before rank stabilization is not easy to answer.We therefore chose to avoid this question by adequate choice of the basefield. It is however remarkable that the weaker condition that ι be rankpreserving already determines regular conditions of growth: the relationord(x)/ord(ι(N(x))) = p depends only on this condition and we may onlyhave ord(x) > p · ord(ι(N(x))) if ι(N(x)) = 1, for x ∈ Mn+1 \Mp

n+1 with


ι(N(x)) = 1. Some additional work using the objects in the proof of Lemma15 shows that in this case ord(x) ≤ p2.Remark 2. Like any phenomenon that is not impossible, this does in facthappen, as shown by the following example due to Kraft and Schoof [17]. In

the cyclotomic Z3 - extension of K0 = Q[√5529], for the transition A1 → A2

there is an a ∈ A2 of order p2 with N ′(a) = 1. The groups A1, A2 have p- ranks 2 and are of types (9, 3) and (9, 9). There is total capitulation, andthus ι∗ is not rank preserving; indeed, p-rk

(ι∗1,2(A1)

)= 1 < p-rk(A1).

This phenomenon is not in contradiction with (12), since there we assumen to be sufficiently large – in particular, beyond total capitulation of thefinite part of Mn. There, the regularity is recovered. We now consider theresults on Weierstrass modules:

Proof of Lemma 2. Let Y ⊂ X be a Λ - submodule. Since X is free as a Zp- module and Zp has no finite subgroups, Y is also Weierstrass. Thereforefinite intersections of Weierstrass modules are trivial. This implies the directsum decomposition X = ⊕jXj in the claim.

Proof of Corollary 1. Since M is a Weierstrass module, we can apply point3. in Proposition 1, which implies the claim (13). Let n > n0 be the leastinteger such that xn 6= 1 and define z(x) = vp(ord(xn)) − (n + 1). Then(13) and the injectivity of ι implies vp(ord(xm) − (m + 1) = z(x) for allm > n. It makes therefore sense to use the notation z(an) for an 6= 1. Theultrametric inequality z(xy) ≤ max(z(x), z(y)) follows from ord(xnyn) ≤max(ord(xn), ord(yn)), and from this, the surjectivity of the norm impliesthe bound c. Indeed, if ai = (ai,n)n∈N ∈M : i = 1, 2, . . . r is a minimal setof generators of M , then by Nakayama’s lemma we have ai,0 6= 1 and thusz(ai) ≤ c by the definition of c. Since the ai generate M , the ultrametricinequality implies z(x) ≤ c for all x ∈ M . Note that c is constant also forall intermediate fields Kn: replacing K by a larger intermediate extensiondoes not increase the value of the constant.

Finally we show that z extends to a map defined on A by (14). If x ∈ Ahas infinite order, then the module Λxp

dis Zp - torsion free, for sufficiently

large d, so it is Weierstrass. Thus z(xpd) is well defined and the limit

limn→∞ vp(ord(xn))−(n+1) exists, which completes the proof. As a map onA, one may regard p−z as a degenerated ultrametric on A, with kernel A.For any an ∈ An with an 6= 1 and any lifts a ∈ A which project on an at then-th level, the value z(an) = vp(ord(an))− (n+ 1) = z(a) is constant.

Finally we consider the µ - part, giving the

Proof of Lemma 3 . Let a = (an)n∈N ∈ A be such that Λb has unboundedp - rank and let λ, µ, ν be the Iwasawa constants of this module; then λ = 0and, for n > n0,

vp (|Λan|) = µpn + ν,(57)

52 PREDA MIHAILESCU

Let b = apµ−1

. Since Λa ∼ Λ/(pµ), it follows that the module B := Λb ∼Λ/(p) has also unbounded p - rank. We claim that Ker (Bn → Bn+1) = 1for all n > n0. Since ord(b) = p, it follows that B is an Fp[[T ]] - module andfrom

ωn ≡ T pnmod pFp[[T ]],(58)

it follows that p-rk(Bn) ≤ pn. For n > n0 we have |Bn| = pp-rk(Bn), so

ν ≤ 0. Comparing with (57), in which µ = 1, we see that aTj

n , j ≤ pn + νform a base for the Fp - vector space Bn, for all n >n 0. We assume that

there is capitulation for n > n0, so ιn,n+1(bT kn ) = 1 for some k < pn + ν; let

N = Nn+1,n ≡ T pn+1(p−1) mod pFp[T ],

where the congruence follows from (58). Since the norm is surjective in A,we have an = N(an+1); let An+1 ∈ an+1 be a prime that splits completely

above Q and An = N(An+1) = ATpn(p−1)

n+1 ∈ an. Consequently, ATpn(p−1)+k

n+1 isa principal ideal of Kn+1 and thus

pn+1 + ν = p-rk(Bn+1) ≤ (p− 1)pn + k,

and k ≥ pn + ν. But p-rk(Bn) = pn + ν, so in this case aTk

n = 1 anyhow:there is no capitulation for n > n0.

We now proceed to radicals and prove first an auxiliary lemma:

Lemma 16. Let n > 3κ and y ∈ K×n be such that yT ∈ E′

n, and if c ∈ Kdivides y, then c ∈ E′(K). Moreover, y = d · eT with d ∈ E2κ, e ∈ E′

n.

Proof. Since yT is a p - unit, it follows that (y) is an ambig ideal fixed byτ . Let A ⊂ Kn with [A] = an ∈ An and an lifting to a sequence a ∈ Abe any ambig ideal fixed by Γ. If a has finite order, A capitulated in K2κ,since pκ is an exponent for A, and thus A = (α), α ∈ K2κ. In particular,if (A, p) = (1), then this must be the case. Let (y) = B · P be a splittingsuch that (B, p) = (1) and P is a product of primes above p. By theabove, B = (β), β ∈ K2κ and since B is principal, so must be P. ThusP is a p - unit; let ℘m ⊂ Km be the primes above K ⊃ ℘ ⊃ (p) andam = [℘m]. Recall that C = G/D℘ is a set of s coset representatives ofthe quotient of G by the decomposition group of some prime ℘ like above.Then P = ℘θn with θ =

∑σ∈C aσσ ∈ Z[C]; if a = (am) has infinite order,

then ord(an) > pn−κ and ℘ord(an)n = ℘ord(a0) = (π0). We may split in

θ = θs + θl with θs =∑

σ∈C:vp(aσ)<vp(ord(an)aσσ and herewith, P = Ps ·Pl,

where Ps := ℘θsn and Pl = ℘θln . Now by definition of Pl, it is a principalideal (πl) ⊂ K, and it remains that so must be Ps. The classes [℘θsm ] havebounded order for n → ∞ and thus Ps = (πs) with πs ∈ K2κ. Altogether,(y) = (β)(πs)(πl) and y = e · β · πs · πl, with e ∈ En. But (β · πs)T ∈ E2κ

and πTl = 1, so it remains that yT = d · eT , d ∈ E2κ, which completes theproof.


We proceed to the

Proof of Proposition 2. Let L′,L be like in the hypothesis. We shall givethe proof for elementary Λ - modules of one of the three types: finite Zp- torsion, infinite Zp - torsion and Weierstrass modules. Since every Λ -torsion group Gal(L′/L) in the given tower is the product of modules ofthese types, the general case follows.

Assume that Gal(L′/L) is a Weierstrass module and Bn := RAD(L′n/Ln).

We have shown that Bn are Λ - modules of bounded rank, at all finite levelsand Bn ⊂ Bn+1 ⊂ ∪nBn ⊂ L′. The condition Bn ⊂ Bn+1 is equivalent toL′nLn+1 ⊂ Ln+1.Our task is to prove that the norm Nn+1,n : Bn+1 → Bn is surjective, so

we cannot apply the Proposition 1 directly, but we may apply it to the dualgalois groups. Indeed, since L,L′ are galois over K, it follows for instancefrom Iwasawa, [14] §3.1, that Gal(L/K∞),Gal(L′/K∞) are Λ - modules.Then so is L = Gal(L′/L) as a factor of Λ - modules, and by assumptionit has finite p - rank. The Galois groups Ln = Gal(L′

n/Ln) form an infiniteprojective system with limit L = Gal(L′/L), under the norm maps. Letιn,n+1 be any lift Ln → Ln+1. One can choose compatible chains of lifts suchthat for n < m < m′, we have ιn,m ιm,m′ = ιn,m′ . The group Gal(Km/Kn)acts on Lm by conjugation and it fixes ιn,m(Ln), independently of the liftchosen. Therefore the condition Nn+1,nιn,n+1 = p as endomorphisms of Lnis fulfilled and we may apply Proposition 1 to L. It follows that for n > n0,ιn,n+1(Ln) = Lpn+1 and thus Gal(L′

n+1/(Ln+1 · L′n)) is an Fp[T ] - module of

rank r = p-rk(L) for n > n0. Furthermore, ιn,n+1 is injective, so

|Gal((Ln+1L′n)/Ln+1) = |Ln|.(59)

Let o ∈ Gal(Kn+1/Kn) be a generator; setting t = o − 1, we have N =p + tv, as in (52). By choosing n0 large enough, we may assume that forn > n0, the element t ≡ T p

nmod pΛ annihilates Gal(L′

n+1/(Ln+1 · L′n)).

By duality, Bn+1 := RAD(L′n+1/(Ln+1 · L′

n)) is also a group of exponent p

annihilated by t∗; however T ∗ ≡ T mod p, so t is also an annihilator of Bn+1

and therefore, taking norms, it follows that Nn+1,n(Bn+1) ⊂ RAD((Ln+1 ·L′n)/Ln+1); the inclusion ’⊃’ follows from the fact that Ln+1[Nn+1,n(Bn+1)]

is the maximal subfield of L′n+1 which is fixed by o. It remains to show

that RAD((Ln+1 · L′n)/Ln+1) ∼= RAD(L′

n/Ln); this follows from (59). Thiscompletes the proof for this case. If L is a finite Zp - torsion module, thenby the first point of Proposition 1, for n > n0 we have ιn,n+1(Ln) = Lpn+1and we deduce like before, that Nn+1,nBn+1 = Bn.

Finally, suppose that L is an infinite Zp - torsion module. We may restrictourselves, without loss of generality, to the case when L = Λl is cyclicof exponent pµ and ιn,n+1 is injective for n > n0: this discards finite Zp- torsion, which can be achieved by taking complements. Moreover, weassume that µ = 1, in order to avoid modules over Z/(pµ · Z). The generalcase follows by induction on µ and the number of generators of L.

54 PREDA MIHAILESCU

Applying Lemma 3 and its proof, we see that Ln are Fp[T ] - mod-

ules of finite rank pn − ν and Nn+1,nLn+1 = Lop−1

n+1 . Note that in this

case the exponent p is replaced by op−1 for obvious reasons. Since o∗ ≡o mod p, as observed in the previous section, it follows by duality that

Nn+1,nBn+1 = Bop−1

n+1 . Since ιn,n+1 is injective, it follows by comparingranks that Nn+1,nBn+1 = ιn,n+1Bn. This completes the proof.

We now pass to reflection, giving the

Proof of Lemma 6. We have seen that the dual of X is a Z[G] - module onwhich the action of τ is determined by τ∗ = 1. The claim follows afterproving the existence of α. Let x ∈ X be a generator. The existence ofthe set Gx is a consequence of the Steinitz lemma in linear algebra. Letθ = α⊤ ∈ Qp[G] and α = 1 − θ; we claim α satisfies the condition. Indeed

X⊥ = αQp[G] is the minimal submodule of Qp[G] annihilated by θQp[G], so

x⊤ ∼= ˜X⊤, which confirms the claim.

We indicate here the general proof for the stabilization of the Leopoldtdefect.

Proof of Lemma 13.30 . From Proposition 4, Zp-rk((Gal(ΩE ∩Mn)/Kn) =r2 for arbitrary base fields K containing p-th roots of unity. The proof is

constructive, since it amounts to Mn∩ΩE = ∪m>nK∞[EN∗

m,n/pm

m ]. Therefore(Mn ·ΩE)/ΩE is an extension of Zp - rank D(Kn). At the same time, Mn ⊂ Ωand the last is a field attached to a fixed base field K. Since Gal((Mn ·ΩE)/ΩE) is a free Zp - group, it follows that this rank is upperbounded byλ(K). Plainly D(Kn) ≤ λ(K) for all n, which completes the proof. The sameidea is used in Washington’s proof.

8. Appendix B: Facts and proofs for Section 3

We start with the exposition of the Iwasawa skew symmetric pairing whichallows the use of duality in the extension Ω/ΩE′ .

8.1. The Iwasawa skew symmetric pairing. In this section we recallIwasawa’s skew symmetric pairing [14], §§9-11 and prove the Theorem 3.For the ease of the reader we give a small table translating from our notationto the one used by Iwasawa in his seminal paper.

Ω → M H → LΩE → N ΩE′ → N ′

Gal(Ω/ΩE′) = Z → X Gal(Ω/(ΩE′ ·H′)) → Y

This translation is pseudo-isomorphic; more precisely, Iwasawa also allowsfor some finite Zp - torsion parts, which we do not consider in our notation.He carefully detaches only the µ part of the galois group Y. This detail ishowever not important for the arguments below. We define here H′ ⊂ H tobe the maximal subfield of H which splits all primes above p. In Iwasawa’snotation, H′ = L′.


We recall the fundamental properties of Iwasawa’s skew symmetric pair-ing. It is a map [ , ] : X × X → Zp, written additively, with kernelω−1(Y ) = ∪nω−1

n (Y ), i.e. [X,x] = 0 ⇔ x ∈ ω−1(Y ): this important prop-erty is based on Lemma 14, §9. The pre-image ω−1(Y ) defined in [14], §9.3corresponds to the totally ramified subextensions of Ω/ΩE′ . The pairing isrelated to the Kummer pairing as follows: at finite levels n > 0, n ∈ N, Iwa-sawa associates to x ∈ X a class c(x, n) ∈ A′

n, such that ΩE [c(x, n)1/pn+1

] isunramified. Then the finite level pairing is induced from Kummer pairingby:

[x, x′]n = 〈x, c(x′, n)〉The skew symmetric pairing is the projective limit of these pairings, switch-ing to additive notation. The kernel of the map c : X → A′ is exactlyω−1(Y ), and it induces the kernel of the pairing. The magnificent work ofIwasawa consists in proving that the pairing is skew symmetric. In particu-lar, [x, x′] 6= 0 iff [x′, x] 6= 0. As a consequence, we can show

Lemma 17. Notations being like above, let f ∈ Zp[T ] be a distinguishedpolynomial which divides the characteristic polynomial of A′ and a ∈ Af ,

the adjoint embedding in Definition 6. Suppose that ΩE′ [ς(a)1/p∞

]/ΩE′ isan unramified Zp - extension which splits the primes above p. Then there is

a b ∈ Af∗ such that ΩE′ [ς(b)1/p∞

]/ΩE′ is unramified, [a, b] = [b, a] and a, bgenerate reciprocally the galois groups of the above fields, via Artin symbol.

Proof. Let a = (an)n∈N ∈ Af and L = ΩE′ [a1/p∞

]. Since L/ΩE′ splits theprimes above p, we may apply the Iwasawa skew symmetric pairing. Itfollows that there is a b = (bn)n∈N ∈ Af∗ , such that

[an, bn]n = [bn, an]n = ζpm(n) , pm(n) = ord(bn)→∞.(60)

By definition of the skew symmetric pairing, there are αn, βn ∈ ΩE,n and

ideals A ∈ an,B ∈ bn such that (αn) = Aord(an), (βn) = Bord(bn) such that

ΩE[α1/pn+1

n ] = ΩE[a1/pn+1

n ], ΩE[β1/pn+1

n ] = ΩE[b1/pn+1

n ], and

[an, bn]n = 〈ϕ(an), βn〉; [bn, an]n = 〈ϕ(bn), αn〉.Since 〈ϕ(an), βn〉 = ζpm(n) , the above relation implies 〈ϕ(bn), αn〉 = ζpm(n) .

In particular, b 6∈ Ker (c) and the extension ΩE′ [b1/p∞

] must be unramified.Furthermore, it follows from (60) that ϕ(b) ·Zp = Gal(L/ΩE′ and ϕ(a) ·Zp =Gal

(ΩE′ [b1/p

∞

]/ΩE′

). This completes the proof.

With this, we can give the

Proof of Theorem 3. It follows from the Lemma 17 that the radical Anr isselfdual, and since AE = A•

nr, it follows that AE = Anr and ΩE′[ς(A)E ] isthe maximal unramified extension of ΩE′ in Ω. This proves point 1. Notethat we cannot make any statement about Ω/Ω: there may be ramified µ- extensions, unramified ones or both; however, these extensions have finiteexponent.

56 PREDA MIHAILESCU

For f |F like in the hypothesis of the Proposition, the radical ς(A′)f splits

into a ramified and an unramified part: the first generates a subfield of Hf ,

the second of Ωf . The non canonical choice of Ar has here the same role

as A†: the intersection ς(A′)f ∩Ar is a well defined radical which generates

the extension ΩE′ · Ωf over ΩE′ [A1/p∞nr ] = H · ΩE′ . This is point 2.

Point 3 is directly verified from the points above: by considering radicals,we have AE = Anr and thus. The definition of AE yields (ς(A)/AE)

• ∼=Gal(HE/K∞)• while A•

E = AE and A•r∼= Gal(Ω/(ΩE′ ·H)) explains the last

two isomorphisms. It follows that Ar ∼= ς(A/AE).For point 4, let f |F be such that Hf ⊂ ΩE, then we use the fact that

rad(ΩE/K∞) is Λ quasi - cyclic and thus Gal((Hf ∩ΩE)/K∞) is a cyclic Λ -torsion Weierstrass module. If the f - part of ς(A) is cyclic or f is coprimeto F/FE , then ς(A)f ∼= Gal((Hf ∩ΩE)/K∞) is well defined. Applying point

3, it follows that ΩE′ [ς(A)1/p∞

f ]/ΩE′ is totally ramified. Note that in this

case ς(A)f is a canonic component of Ar. The obstruction to a canonicdefinition of Ar arises thus only from f - components which are not cyclicΛ - modules.

In particular, for f = T ∗ and f = T , which are both prime polynomialssuch that the f - primary part Ωf is cyclic, the point 4. applies and it follows

that ΩE′ [ς(A)1/p∞

T ∗ ] and ΩE′ [ς(A′)1/p∞

T ] are both totally ramified, which ispoint 5. This completes the proof of Theorem 3.

The skew symmetric property of the Iwasawa pairing is proved by clas-sical reciprocity. It is intimately related to the Leopoldt reflection, as thefollowing simple example illustrates. Suppose that f is a prime polynomialwith f(T ) 6∈ T, T ∗ and ς(A) ∼= A/Af ⊕A/Af ⊕Af∗ . We may also assumethat K is CM and thus f∗ annihilates A+. The f - part of A is A−

f and

has rank 2, but there is a canonic (up to finite torsion) cyclic submodule

A0 = Λa ⊂ Af such that ΩE′ ·H+= ΩE′ [ς(A)

1/p∞

0 ], while reflection requires

that ΩE′ [ς(A)1/p∞

f∗ ]/ΩE′ be unramified. Therefore Anr = A0 · Af∗ . We cer-

tainly have Af∗ ⊂ AE , since (A/AE)1+ = 1; but is A•

0∼= Af∗ under

Kummer pairing? This is a direct consequence of the Iwasawa skew sym-metric pairing. Showing that A0 ⊂ AE by means of the Leopoldt reflectionis the difficult part, since it arises canonically as a radical, but non-canonicalas a factor. However, if A0 6⊂ AE , since AE is a Weierstrass module and fis assumed to be prime, it follows that A0 ∩AE = 1; comparing ranks, wefind that Gal(Hf/K∞) ∼= ϕ(A0)|Hf . We now invoke the Artin symbol, by

which

RAD(HE/K∞) ∼= RAD(ΩE/K∞)/RAD(ΩE/HE)

∼= U+∞/E∞

∼= Gal(Ω/(H · ΩE′)) ∼= A•r .

It follows that A•r∼= Gal(HE/K∞)• and Ar ∼= A/AE ∼= A0. But then

(ΩE′ ·H)[A1/p∞

0 ] is unramified, in contradiction with the choice of A0. This


proof scheme is easily generalized in the CM case; for the non CM casewe need the Corollary 2 below. One can prove thus by mere Leopoldtreflection, that AE = A•

nr, which then implies all the other statements in theTheorem 3. In particular, one may view the Kummer pairing as a pairingAE × A•

E → Zp; it induces a symmetric quadratic form Q : a → [a, a•].Deducing a map c : AE → AE such that

[a, c(a)•] = [c(a)•, a] = [c(a)−1, a•],

which induces Iwasawa’s skew symmetric pairing, requires some more workwhich we skip here, since we may use directly the results of [14].

8.2. Complex multiplication and duality.

Proof of Fact 4 . Suppose that for an n ≥ 0, the map ιn,n+1(A−n ) is not

injective and let Q ∈ a−n ∈ A−n be a prime of order p which capitulates

in Kn+1. Let (β) = Qp, β ∈ Kn and O(Kn+1)Q = (b). Then there is

a unit e ∈ E(Kn+1) with β = ebp and β1− = (e/e) · bp(1−); but then

e/e = ζc is a root of unity and after eventually multiplying β by ζ−c/2 we

find (β/bp)1− = 1. Since Kn+1 = Kn[ζ1/ppn+1 ], it follows by Kummer theory

that b1− = β(1−)/p ∈ Kn+1 and β1− = ζc′

pn+1 · βp1 . This implies that Q is

principal and an = 1, so there is no capitulation.

We use now the Lemma 14 in the previous Appendix for proving theProposition 3 and Proposition 4.

Proof of Proposition 3. We use class field theory; let Ω′n denote here the

maximal p - ramified p - abelian extension of Kn: thus Ωn ⊂ Ω′n is the

maximal Kummer extension contained in Ω′n. Then ΩT ∗,n ⊂ Ω′

n is themaximal subextension with group over Hn annihilated by T ∗: by class field

theory, writing Vn = U(1)n /En, we have

YT ∗,n = Gal(ΩT ∗,n/Hn) ∼= Vn/VT ∗

n .

We apply Lemma 14 to Vn. Now U∞ = lim←−U(1)n is a Λ - module with

a cyclic part of Λ rank r = 2r2 while E∞ = lim←−U(1)n has the essential Λ

- rank r2. Since the Leopoldt defect is bounded – and in our hypothesisstable from K upwards – the limit U∞/E∞ has the essential Λ - rank r2 andp-rk(Vn) ≥ r2pn−κ.

We consider the torsion part of Vn,T ∗; this is µpn+1 for En, while Un has a p- torsion part Tn ∼= (Cpn+1)s, where s is the number of ramified primes above

p: this is thus a group of exponent and sub-exponent pn+1. The torsion isgenerated by the roots of unity in the various completions at primes abovep. Indeed, let ρ = ρ(ζpn+1 − 1) + 1 ∈ Un, with ρ defined in (7). Thenι℘(ρ) = ζpn+1 while ισj℘(ρ) = 1 for j > 1; since U℘[ζpn+1 ] = Un,℘, we see

that Γ fixes Un,℘ and ρT∗

= 1. The torsion is thus Tn = ρZp[C] ∼= (Cpn+1)s

and annihilated by T ∗: it follows that, in the notation of Lemma 14, we have

58 PREDA MIHAILESCU

Tn = µ(Un). Consequently, the torsion group µn(Vn) = Tn/µpn+1∼= Cs−1

pn+1 ; it

is annihilated by T ∗. Let Ω′n,T ∗ = Ω

ϕ(µn(Vn))n,T ∗ be the fixed field of the torsion

part. Its galois group is isomorphic, by Lemma 14, to Yn := VN∗

nn /V

pn−κN∗

nn .

It is thus Zp[G] quasi - cyclic of p - rank r2 and subexponent pm withm ≥

[n−κ2

].

We have shown in Lemma 14 that ιn,m(Yn) = Y pm−n

m , which implies that

Ωn,T ∗,E · Km =(Ωm,T ∗,E

)ϕ(Y pm−n

m ). In particular, the fields Ω′

n,T ∗ form an

injective sequence of fields with groups of constant p - rank and in the limitwe have

Ω′T ∗ = Ω

ϕ(µ∞(V )T ∗ .

Putting the two pieces together, HT ∗ ⊂ Ω′T ∗ ⊂ ΩT ∗ and Gal(Ω′

T ∗/HT ∗)

is Zp[G] quasi - cyclic of p - rank r2 while Gal(ΩT ∗/Ω′T ∗) = ϕ(µ∞(V )) is

a group on which the G/D℘ acts transitively and which is annihilated bythe norm of K, having p - rank s − 1. It is interesting to observe thatK∞[p1/p

∞

] ⊂ Ω′T ∗ is the subfield with group fixed by G; one may ’shift’

the structure by this field, which naturally belongs to the p - unit fieldK∞[Π1/p∞ ] – this would correspond to the denominator µn(En) in µn(Vn).

We would then have ΩT ∗ = ΩT ∗,E [Π1/p∞ ], for a new ΩT ∗,E with group of

rank r2 − 1. The rank computation is the same and the difference consistsin the side to which K∞[p1/p

∞

] is counted.

We proceed with the simpler

Proof of Proposition 4 . The radical RAD((ΩE ∩M)/K∞) ⊂ E1/p∞ is anni-hilated by T ∗. At finite levels we may apply the Lemma 14 to the finite Λ -

modules Vn := En/(Epn+1

n ). The torsion part is µn(Vn) = µpn+1 , the p - rootsof unity of K. They correspond to the extension K∞; hence, the contribution

of the torsion vanishes at infinity and we have ΩE∩M = ∪nKn[(EN∗

nn )1/p

n+1].

By Lemma 14, the fields Kn[(EN∗

nn )1/p

n+1] build an injective sequence with

groups YT,n which are Zp[G] - quasi cyclic, of constant p - rank r2 and di-verging subexponents. Therefore YT = Gal(ΩE∩M) is a quasi - cyclic Zp[G]- module of rank r2, free as a Zp - module.

As a useful consequence, we have at least the following generalization ofFact 4:

Lemma 18. Suppose that the field K is not CM. Let B ⊂ A be a Λ[G] - sub-

module such that Be+ = 1. Then the capitulation kernel Ker (ς(Bn) →ς(B)) = 1.Proof. Like in the proof of Fact 4, we choose an ∈ Bn and Q ∈ an, (β) =Qp = (b) with b ∈ Kn+1, so β = ebp. Now we use annihilation of E by Re−,

according to (23): the annihilator module ˜ς(e)⊤ ⊂ Re− is disjoint from Re+,

which annihilates an. There is then an α ∈ Fp[G] such that aαn 6∈ ς(B)pn


but epκα ∈ Epκ+1

n+1 and thus βpκ·α ∈ Kpκ+1

n+1 . The same Kummer argument as

before implies that βpκ·α = xp

κ+1 ∈ Kpκ+1

n and thus ((Q/(x))α)pκ+1

= (1),so Qpκ·α is principal, in contradiction with our assumption.

We proceed now to the proof of the simple Lemma 9

Proof of Lemma 9 . Let δ ∈ E be a Minkowski unit and θ ∈ Zp[G] suchthat θ/|G| ∈ Qp[G] is an idempotent which generates the annihilator ideal

δ⊤ ⊂ Qp[G]. Let θ = θm + pm+1rm, with θ ≡ θm mod pm+1Zp[G], so θm arethe rational approximants of θ to the pm-th order. LetH ⊂ G be a minimalsubset such that θZp[G] = θZp[H]. We first define D′

n = Span(δθn+1σ)σ∈H ,

where Span denotes here the Z - span. Then D′n ⊂ Up

n+1by construction.

However the condition that (Dn ·Ep)/Ep has p - rank r may not be fulfilled,so we shall need to perform some change of generators. This will be doneby combining D′

n with radicals from D′n+j for j > 0; the approach is similar

to the one in the proof of Lemma 14.The set S1 = ((θ1Zp[H]) · (pZp[G]))/(pZp[G]) is finite and D′

1∼= δS1 mod

(D′1)p. Let i(x) : E → N be the p - index, so i(x) = k ⇔ x ∈ Epk \ Epk+1

;there is then a finite k = max(i(δs) : s ∈ S1). If k = 0, then we may

define Dn = D′n. Otherwise, let r′1 < r be the p - rank of (D′

1Epk)/Ep

kand

r1 = r − r′1. Let

d′j ∈ D′1, ej ∈ E : d′j = ep

k

j , j = 1, 2, . . . , r1

be a system of Z - independent units and let tj ∈ Z[G] be such that d′j =

δθ1·tj . Then we define

dj,n = δθ1+ktj/pk.

By construction, we see that dj,n ∈ E \ Ep and dj,n ∈ Upn+1

. Let D1,n =Span(dj,n)

r1j=1. We proceed by induction as follows: let H1 ⊂ H be a maxi-

mal subset such that δθ1Z[H] andD1,1 are Z - independent, thus |H1| = r−r1.Let S2 be defined with respect toH1 by S2 = ((θ1Zp[H1])·(pZp[G]))/(pZp[G])and k1 = max(i(δs) : s ∈ S2). If k1 = 0, then we let Dn = D1,n · δθnH1 . Thesystems Dn ⊂ E fulfill the required properties by construction. If k1 6= 0,we proceed like in the previous step and since k1 < k, the procedure willeventually end for a value kh = 0. Thus, we obtain systems of units Dn ⊂ Ewith p-rk ((DnE

p)/Ep) = r, Dn ⊂ Upn+1

and Dn+1 ⊂ Dn · Epn+1

. The

sub-exponent pn+1 for Dn/Dpn+1

n follows from the fact that

p-rk(Dn/Dpn+1

n ) = p-rk(Dn ·Epn+1

/Epn+1

),

which holds by construction.

9. Appendix C: Complements to sections 4 and 5

We give an example of an extension Ωf∗/Hf∗ with a polynomial f(T ) 6= T :

60 PREDA MIHAILESCU

Example 2. Let K = Q[ζ] be the p-th cyclotomic extension. Then s = 1and r2 = (p− 1)/2. Thus Zp-rk(Gal(ΩT ∗/K∞)) = r2 and

ΩT ∗ = ΩE1 [Π1/p∞ ].

Suppose now that p is such that Vandiver’s Conjecture holds and the irreg-ularity index is 1. Let then A = A− = Λa and assume additionally, that theminimal polynomial of a is linear, namely f(T ∗) = T ∗ + cp, c ∈ Z×

p . Thisis a situation which occurs often. The cyclotomic units Cn = En = O(Kn)and the local units U ′

n are norm coherent and the norm is surjective on bothsystems of units; let εk be the orthogonal idempotent with εp−kA 6= 1 andχ ∈ Z[G] approximate εk, the reflected idempotent, to order pM , for somelarge M ; in particular, k is even. There is for n ≤M a system of local andglobal Minkowski units ξn ∈ Un, ηn ∈ R ∩Kn such that

ξχf(T )n = ηχn · xpM

n , xn ∈ Un.(61)

In particular ξχn , ηχn generate one dimensional Λn/p

MΛn - modules. Let nbe fixed with 2n < M ; by choice of f , the classes in An have order pn+1, sothere is a cyclic unramified extension Fn/Kn of degree pn+1. By the proofof Lemma 14, class field theory requires that there also be a p - ramifiedextension Ln/Hn of degree pm with pn+1 ≥ pm ≥ p[n/2] and galois group inthe εp−k component of Gal(Ωn/Hn), annihilated by f(T ∗).

The Lemma concerns in fact only the polynomial f(T ) = T , but the casewhen f is an arbitrary polynomial is proved similarly. In general, if f(T ) isa polynomial of degree d, there exist for n sufficiently large gn, hn ∈ Λ suchthat

gn · f + hn · ωn = pn+1.

It follows that (Ugnn · Upn+1

n )/Upn+1

n has p - rank d · (2r2) and is annihi-lated by f . Defining f like above and Ωf by Definition 6, it follows thatp-rk

(εp−kGal(Ωn,f2/Kn)

)= 2 = deg(f2). In our example, the ramified ex-

tension must be a cyclic extension of Fn and L′n = Kn+mLn is a Kummer

cyclic extension which is abelian over Kn and F′n = Fn ·Kn+m ⊂ L′

n.

Let L′n = Kn+m[e

1/pn+m+1] and ν ∈ Gal(L′

n/Kn+m) be a generator. Then

νpn+1

is a generator for the ramified extension L′n/F

′n; by hypothesis we must

have νpn+1·f(T ∗) = 1. Furthermore, ν generates by restriction Gal(F′

n/Kn+m)

and the hypothesis implies that νf(T∗) fixes F′

n, thus νf(T∗) ∈ νp

n+1. As-

sembling the two conditions, we deduce that νf(T∗)2 = 1. It follows that

Ln ⊂ εp−kΩn,f2, a p - abelian, p - ramified extension of p - rank 2, whereidempotents act on fields by acting on galois groups fixing these fields:

εp−kΩn,f2 = Ω(1−εp−k)Gal(Ωn/Kn)n .

We now consider Kummer radicals. Reflection implies for e that

ef(T )2 ∈ Epn+m+1

n+m .(62)


Furthermore, since L′n/Kn is abelian, we have the condition

eω∗

n ∈ Epn+m+1

n+m .(63)

Additionally, Fn is Kummer over Kn, so there are e0 ∈ En and u ∈ En+mwith

e = e0 · upn+1

, e0 ∈ Upn+1

n , ef(T )0 ∈ Epn+1

n .(64)

The three conditions must have a solution, as required by class field theory.Let e = ηλn+m, with λ ∈ Λ. Then (64) yields λ = Nn+m,na(T ) + pn+1b(T )for some a(T ), b(T ) ∈ Λ \ pΛ such that

a(T ) · f(T ) ∈ (ωn, pn+1)Λ

Nn+m,n · a(T ) · f2(T ) + pn+1b(T ) · f2(T ) ∈ (ωn+m, pn+m+1)Λ

Nn+m,n · a(T ) · ω∗n + pn+1b(T ) · ω∗

n ∈ (ωn+m, pn+m+1)Λ

Nn+m,n · a(T ) · f(T ) + pn+1b(T ) · f(T ) 6∈ (ωn+m, pn+2)Λ.

The last condition stems from ηn+m = ξf(T )n+m, which is (61), and implies that

L′/F′n is ramified. A solution arises by using (50) and the general fact that,

for coprime polynomials f, g ∈ Zp[T ], the ideal (f, g) is of finite index in Λand there is a linear combination uf + vg = ps with ps ≥ max(f(0), g(0)).Let gnf + xnωn = pn+1. The first condition implies that a(T ) is a multipleof gn, say a(T ) = gn(T )a

′(T ). The second and the last conditions becomethen

a′(T ) + b(T )f(T ) ∈ Λ \ (p, ωn+m)Λ,(65)

a′(T )f(T ) + b(T )f2(T ) ∈ (ωn+m, pm)Λ,

while the third becomes, via (50),

gn(T )a′(T ) + b(T ) · unωn ∈ (ωn+m, p

m)Λ.(66)

Finally the resulting system can be solved as follows: first find a couplea′1(T ), b1(T ) ∈ Λ \ pΛ with minimal valuations and such that the condition(66) is fulfilled. Set a′(T ) = a2(T ) · a1(T ) and b(T ) = b1(T ) · ps · a2(T ) andsolve (65) with respect to a2(T ) and s. A possible solution arises by settings = 0 and g′(T ) ∈ (p, ωn+m)Λ, such that g′(T )f(T ) + y(T )ωn+m ∈ pmΛ.Then let λ′ = gn(T )a1(T ) + b1(T )f(T ), which is the right hand side in thefirst condition of (65). We may assume that λ′ 6∈ pΛ, since both terms arenot p - multiples and if the sum is, one may always add a multiple of pm tob1(T ), achieving the required result. Thus we solve

a2(T )λ′ ∈ (g′(T ), pm)Λ.

Then neither e0 nor u are p - powers and the resulting e verifies all therequired conditions, including the fact that L′/F′

n is ramified.After having shown the existence of the extension towers Kn ⊂ Fn ⊂ Ln,

it is interesting to consider the picture at infinity. We have shown thatthe galois groups Gal(Ωn/Hn) are norm coherent. The extensions Fn forman injective system, so let F =

⋃n Fn. Since εp−kΩ(f∗)2 has group of Zp -

62 PREDA MIHAILESCU

rank 2, there is a Zp - extension K∞ ⊂ F ⊂ εp−kΩ(f∗)2 which is linearly

disjoint from F and with galois group annihilated by f2 but not by f . SinceGal(Ln/Fn) form a projective system, it follows that F ·Ln are injective and

L =⋃n Ln ·F is a Zp - extension of F with Gal(L/F)f(T

∗) = 1, as required.Furthermore, Ln ⊂ Ωn,f2 for all n, so it follows that L ⊂ Ω(f∗)2 . AlthoughLn/Kn are cyclic for all n, the system (K∞ · Ln)n∈N is not injective. Thiscan also be verified from the explicite construction above. For f(T ) 6= T onethus observes that in the case when Hf 6= K∞ there is a p - abelian andtotally p - ramified extension L/Hf with group annihilated by f . In this caseL ⊂ Ω(f∗)2 \ Ωf .

The same arguments require that Ω′(f∗)2/H with Gal

(Ω′(f∗)2/H

)εp−k(f∗)2=

1 has Zp - rank 2. The rank loss propagates and there must be a Zp - subex-tension L′ ⊂ Ω′

(f∗)2 with L′ ⊂ Ω(f∗)3 \ Ω(f∗)2 . Since Ω contains a free Λ -

submodule, the rank loss is absorbed at infinity.

Next we illustrate the arguments of the proof of Proposition 5 on theexample of the p - cyclotomic tower, used in the previous example. Thus,K = Q[ζ]. In this example, we may assume that Vandiver’s Conjecture holdsfor p, so the units E(Kn) are cyclotomic and Nm,n(Em) = En. Let εk =1p−1

∑σ∈G ω

k(σ)σ−1 be the orthogonal idempotents of Zp[G] and assume

that Leopoldt’s Conjecture is false. Then there is an even number p − ksuch that εp−kE = 1; the construction of Φ shows that εkA/(A

T ∗

) is

infinite. Let χ ∈ Z[G] approximate εp−k to the pM -th power for a large

M , so ηχ ∈ U(K)pM, with η a real cyclotomic unit generating E(K) as a

Zp[G] - module. Let M/4 > n > 0 and Φn = Kn[ηχ/pn+1

], an unramifiedextension. Proposition 5 shows that there is no totally ramified extensionLn/Φn of degree pn/2 ≤ pm = [Ln : Φn] ≤ pn+1, such that Ln/Kn is abelian– here the Leopoldt case differs from Example 2. It follows from Lemma 3

that ΩE′ [A1/p∞

T ]/ΩE′ must be totally ramified; however the Iwasawa skewsymmetric pairing shows that this extension is unramified. Therefore it mustbe trivial and the Leopoldt defect vanishes.

Acknowledgments: The third version of this manuscript contains sev-eral improvements of redaction and corrections in the lemma 10; lemma 3was removed. Several colleagues helped directly or indirectly to the redac-tion of the present version, through their comments and discussion. In orderto avoid ambiguity concerning the endorsement of the results included here,a complete list of those who have my full gratitude for their contributionalong the work at this paper will be added after a peer review of this man-uscript is completed.

References

[1] T. Albu. Cogalois theory. Number 252 in Monographs and textbooks in pure andapplied mathematics. Marcel Dekker Inc., 2003.


[2] T. Albu. Field theoretic cogalois theory via abstract cogalois theory. J. Pure Appl.Algebra, 208(1):101–106, 2007.

[3] J. Alperin and R. Bell. Groups and Representations, volume 162 of Graduate Textsin Mathematics. Springer, 1995.

[4] B. Angles. On the p - adic Leopoldt transformation of a power series. Acta Arith-metica, 134(4):349–368, 2008.

[5] J. Ax. On the units of an algebraic number field. Illinois Journal of Mathematics,9:584–589, 1965.

[6] A. Baker. Linear forms in the logarithms of algebraic numbers I, II, III. Mathematika,13, 14:204–216; 102–107, 220–228, 1966, 67.

[7] A. Brumer. On the units of algebraic number fields. Mathematika, 14:121–124, 1967.

[8] M. Emsalem, H. Kisilevsky, and D. Wales. Independance lineaire sur Q de logarithmesp - adiques de nombres algebriques et rang p - adique du groupe des unites d’un corpsde nombres. Journal of Number Theory, 19:384–391, 1984.

[9] L. Federer. The nonvanishing of Gross’ p - adic regulator Galois cohomologically.In Journees Arithmetiques de Besancon (1985), volume 147-48 of Asterisque, pages71–77, 1987.

[10] L. Federer and B. Gross. Regulators and Iwasawa modules. Invent. Math., 62(3):443–457, 1981.

[11] T. Fukuda. Remarks on Zp-extensions of number fields. Proc. Japan Acad. Ser. AMath. Sci., 70(8):264–266, 1994.

[12] R. Greenberg. On the Iwasawa invariants of totally real fields. American Journal ofMathematics, 98:263–284, 1976.

[13] K. Iwasawa. On γ - finite modules. Ann. Math. Second Series, 70:290 – 312, 1959.[14] K. Iwasawa. On Zℓ - extensions of number fields. Ann. Math. Second Series, 98:247

– 326, 1973.[15] J. Jaulent. Note sur la conjecture de Leopoldt.

http://front.math.ucdavis.edu/0712.2995, 2007.[16] J. Jaulent. Sur les conjectures de Leopoldt et Gross. In Journees Arithmetiques de

Besancon (1985), volume 147-48 of Asterisque, pages 107–120, 1987.[17] J. Kraft and R. Schoof. Computing Iwasawa modules of real quadratic number fields.

Compositio Mathematica, 97:135–155, 1995.[18] L. Kuz’min. The Tate module for algebraic number fields. Math. USSR Izvestija,

6(2):263–321, 1972.[19] M. Laurent. Rang p - adique d’unites et action de groupes. J. reine angew. Math.,

399:81–108, 1989.[20] M. Le Floc’h, A. Movahhedi, and T. Nguyen Quang Do. On Capitulation Cokernels

in Iwasawa Theory. American Journal of Mathematics, 127(5):851–877, 2005.[21] H. Leopoldt. Zur Artihmetik in Abelschen Zahlkorper. J. Reine Angew. Math, 209:54–

71, 1962.[22] S. Lang. Cyclotomic fields I and II, volume 121 of Graduate Texts in Mathematics.

Springer, combined Second Edition edition, 1990.[23] J. Serre. Local Class Field Theory. In Cassels and Frohlich, editors, Algebraic Number

Theory, pages 129–161. Academic Press, 1967.[24] M. Waldschmidt. Transcendence et exponentielles en plusieures variables. Inventiones

Mathematicae, 63, 1981.[25] L. Washington. Introduction to Cyclotomic Fields, volume 83 of Graduate Texts in

Mathematics. Springer, 1996.

(P. Mihailescu) Mathematisches Institut der Universitat Gottingen

E-mail address, P. Mihailescu: [email protected]

arXivarXiv:0905.1274v3 [math.NT] 15 Sep 2009 Par la mère apprenant que son fils est guéri, par...

Documents

Transcript of arXivarXiv:0905.1274v3 [math.NT] 15 Sep 2009 Par la mère apprenant que son fils est guéri, par...