• Keine Ergebnisse gefunden

4.2 Projective varieties and the µ-invariant

4.2.1 Introduction

LetKdenote the composite of allZp-extensions ofK, i.e., Gal(K/K)∼=Zdp for somed∈N, and suppose thatd≥2. In the preceding section, we considered homomorphisms

πL: Zp[[Gal(K/K)]] //// Zp[[Gal(L/K)]]

for any fixedZp-extension Lof K. Baba˘ıcev more generally studied the set of all surjective homomorphisms

π: Λd //// Λ ,

where Λ = Λ1 =Zp[[T]] and Λd=Zp[[T1, . . . , Td]] (see Definition 2.16).

Let Γd, respectively, Γ, denote free abelian pro-p-groups of rank d, respec-tively, of rank 1. Then we have topological isomorphisms Γd∼=Zdp and Γ∼=Zp.

We will use some notation introduced in Section 2.1. Let ε(Γd) := {π: Γd //// Γ}

denote the set of all surjective Zp-module homomorphisms (i.e., continuous group homomorphisms) from Γd into Γ. In what follows, we will usually write the groups Γd and Γ multiplicatively, since in our applications, these groups will come up as Galois groups. Using the isomorphisms Γd ∼=Zdp and Γ∼=Zp, we will identifyε(Γd) with the set

ε(Zdp) := {π :Zdp //// Zp} that has been studied in Section 2.1.

We will identify two homomorphismsπ1, π2:Zdp −→Zp ifπ12u for some u ∈Zp. This will be important for the application to Zp-extensions (compare Remarks 2.6, (1) and Lemma 2.7), and makes it possible to obtain an isomor-phism between ε(Zdp) and the (d−1)-dimensional projective space Pd−1(Zp) overZpintroduced in Definition 2.1 (compare Proposition 2.5). Thereforeε(Γd) may be seen as a projective variety.

Using the isomorphism Gal(K/K)∼=Zdp, we may furthermore identifyε(Zdp) and

ε(Gal(K/K)) := {π: Gal(K/K) //// Zp} . This has been used in Lemma 2.7 in order to obtain a bijection

E(K) −→ ε(Gal(K/K)) ;

roughly speaking, each L∈ E(K) corresponds to the restriction map πL: Gal(K/K)Gal(L/K),

respectively.

Now let Γ and Γd be as above. Note that each homomorphism π ∈ ε(Γd) defines a homomorphism

π: Zp[[Γd]] //// Zp[[Γ]]

of the corresponding completed group rings. Let γ1, . . . , γd denote topological generators of Γd. Then Theorem 2.18 implies that there exists an isomorphism

ϕ: Zp[[Γd]] // Λd = Zp[[T1, . . . , Td]]

induced by the mapγi7→1 +Ti, 1≤i≤d.

Ifγ10, . . . , γd0 is another system of topological generators of Γd, then γj0 =

d

Y

i=1

γaii,j , 1≤j≤d ,

and A:= (ai,j)i,j ∈GLd(Zp) is an invertible matrix overZp. The map induced by γj0 7→1 +Tj, 1≤j≤d, yields another isomorphism

ϕ0 : Zp[[Γd]] // Λd , again using Theorem 2.18. The commutative diagram

Λd

α

Zp[[Γd]]

ϕ ;;

ϕ0 ##

Λd

4.2. PROJECTIVE VARIETIES AND THEµ-INVARIANT 143 defines an automorphismα: Λd

−→ Λd, given by the substitution

Tj 7→ Tj0 :=

d

Y

i=1

(1 +Ti)ai,j−1, 1≤j≤d .

Definition 4.6. A change of variables in Λd of the above shape is called admissible.

Lemma 4.7 (Baba˘ıcev). Let f(T1, . . . , Td) be a formal power series different from zero having coefficients in a commutative ring E of characteristic p6= 0.

Then there exists an admissible change of variables of the form X1 = (1 +T1)(1 +Td)a1 −1,

...

Xd−1 = (1 +Td−1)(1 +Td)ad−1 −1, Xd = Td,

witha1, . . . , ad−1 ∈N, under which f is carried to a series g(X1, . . . , Xd) such thatg(0, . . . ,0, Xd)6= 0. Actually, a1, . . . , ad−1 may be chosen as

a1 = . . . = ad−1 = pl, withl∈N sufficiently large.

Proof. See [Ba 76], Lemma 1. The proof given there in fact is an adaption of Lemmas 2 and 3 in [Bou 89], Chapter 7,§3, with the additional property that we want the changes of variables to be admissible.

Let π∈ε(Γd). If the topological generators of Γd are chosen such that the kernel ofπ : Γd−→Γ is generated by γ1, . . . , γd−1, and if δ :=π(γd), then δ is a topological generator of Γ. The induced homomorphismπ: Λd−→Λ is then given by

π(Ti) = π(γi−1) = π(γi)−π(1) = 1−1 = 0 for every 1≤i≤d−1, and

π(Td) = π(γd)−1 = δ−1 = T .

Iff ∈Zp[[Γd]]∼=Zp[[T1, . . . , Td]], then we simply have π(f) =f(0, . . . ,0, T).

We will now see that for a given π, we may always choose topological gen-erators of Γd such thatπ obtains this canonical form.

Remark 4.8. For every π ∈ ε(Γd), we may choose topological generators γ1, . . . , γdof Γdsuch that the kernel ofπ: Γd−→Γ is generated byγ1, . . . , γd−1. Proof. The kernel of π : Γd −→ Γ is a Zp-submodule of Γd and therefore is Zp-free. Its rank has to be strictly smaller thand, sinceπ is surjective, and in fact, ker(π) hasZp-rank equal tod−1, sinceπ induces an exact sequence

0 // Zrank(ker(π))

p // Zdp // Zp // 0 .

By the Principal Divisor Theorem (see [JS 06], Thm. VII.8.2), there exists a basisγ1, . . . , γdof Γd(i.e., a set of topological generators of this multiplicatively written group) such that ker(π) is generated topologically by γ1a1, . . . , γd−1ad−1, with a1, . . . , ad−1 ∈ Zp and a1 | a2 | . . . | ad−1. Let z := γd−1ad−1 ∈ ker(π).

Now assume that p |ad−1 inZp. Then we can write z =yp for some element y∈Γdthat does not lie in the kernel of π (since theγiai form a basis of ker(π), they are linearly independent). But then x := π(y) ∈ Zp is different from 1, and xp = (π(y))p = π(yp) = π(z) = 1, which contradicts the fact that Γ is torsion-free (as being a freeZp-module). Thereforepdoes not divide ad−1, i.e., a1, . . . , ad−1 ∈Zp, and ker(π) is generated byγ1, . . . , γd−1.

Definition 4.9. An element f ∈ Λd = Zp[[T1, . . . , Td]] is in Weierstraß normal form with respect to Td if

f = U·pm·(Tdk+ak−1Tdk−1+. . .+a0),

wherem∈N0,k∈N,U ∈Λdis a unit anda0, . . . , ak−1 ∈(p, T1, . . . , Td−1) are contained in the maximal ideal of the local ring Zp[[T1, . . . , Td−1]]. f is called regular in Td iff is in Weierstraß normal form with respect toTd andm= 0 in the corresponding representation.

Remarks 4.10.

(1) Iff is in Weierstraß normal form with respect toTd, thenf =U·pm·f˜(Td) with a distinguished polynomial

f˜(Td) ∈ (Zp[[T1, . . . , Td−1]])[Td] in the sense of Definition 1.11.

(2) Ifπ∈ε(Γd) is a homomorphism such that ker(π) is generated topologically byγ1, . . . , γd−1, thenδ :=π(γd) generates Γ. Iff ∈Λdis in Weierstraß nor-mal form with respect toTdin the variablesT1, . . . , Tdinduced byγ1, . . . , γd, then we can simply write

π(f) = π(U)·pm·(Tk+ak−1·Tk−1+. . .+a0),

with ai = π(ai) = ai(0, . . . ,0) ∈ p·Zp, 0 ≤ i ≤ d−1. In particular, π(f)6= 0, and p|π(f) if and only if m >0, i.e., if and only if p|f.

(3) We may apply the Weierstraß Preparation Theorem 1.14 in the ring of power series

Zp[[T1, . . . , Td]] ∼= (Zp[[T1, . . . , Td−1]])[[Td]], sinceZp[[T1, . . . , Td−1]] is a local ring with maximal ideal

Md−1 = (p, T1, . . . , Td−1)

which is complete with respect to theMd−1-adic topology; compare Propo-sition 2.17, (i). This implies that an elementf ∈Λdis regular with respect toTd if and only iff 6∈(p, T1, . . . , Td−1)⊆Λd.

4.2. PROJECTIVE VARIETIES AND THEµ-INVARIANT 145 Lemma 4.11 (Baba˘ıcev). Let f ∈Zp[[Γd]] be non-zero. LetU ⊆ε(Γd) denote the set of homomorphismsπ: Γd−→Γ such that

(1) we can choose topological generators γ1, . . . , γd of Γd such that ker(π) is generated byγ1, . . . , γd−1, and

(2) f is in Weierstraß normal form with respect to Td in the variables induced by γ1, . . . , γd via the map γi7→1 +Ti, 1≤i≤d.

Then U ⊆ ε(Γd) is open and dense in the topology defined on ε(Γd) via the bijectionε(Γd)−→ ε(Zdp)−→ Pd−1(Zp) (compare Remarks 2.6, (2)).

Proof. This is basically an application of Lemma 4.7 and Remark 4.8, see Propo-sition 1 in [Ba 76] for details.

Definition 4.12. Let M denote a finitely generated Λd-module. For every surjective homomorphismπ: ΛdΛ, we defineMπ :=M/(ker(π)·M); this is a Λ-module, where we identify Λ = Λd/ker(π).

Note that this corresponds to the notion XL used by Greenberg (compare the preceding section).

Theorem 4.13 (Baba˘ıcev).

(i) Let M denote a finitely generated Λd-module, and let m := rankΛd(M).

Then the subset

U := {π∈ε(Γd)|rankΛ(Mπ) =m} ⊆ ε(Γd) is open and dense in ε(Γd).

(ii) LetM denote a finitely generatedΛd-module, and assume that there exists a homomorphism π0 ∈ ε(Γd) such that Mπ0 is a finitely generated Zp -module. Then the setU ⊆ε(Γd) containing all π such that Mπ is finitely generated overZp is open and dense in ε(Γd).

We recall that the Λd-rank of a finitely generated Λd-module N may be defined via

rankΛd(N) := dimQ(N⊗ΛdQ),

where Q denotes the quotient field of Λd, and dimQ means the dimension as Q-vector space.

Proof. (i) This is Theorem 1 in [Ba 76]. Since Baba˘ıcev only gives a very brief proof, we will include here a proof giving full details.

Since M is a finitely generated Λd-module, there exists a surjection F // M // 0

for some free Λd-module F with basis f1, . . . , fl. Let R ⊆ F denote the kernel of this map. Then R is finitely generated over Λd, since F is Noetherian as being finitely generated over the Noetherian ring Λd, and rankΛd(M) =m if and only if rankΛd(R) =l−m. Indeed, the sequence

0 // R // F // M // 0

of Λd-modules is exact by construction, and therefore the following se-quence of Q-vector spaces also is exact:

0 // (R⊗Q) // (F ⊗Q) // (M⊗Q) // 0 . (?) Note that in general, tensoring a sequence of Λd-modules with a Λd-module N will be only right-exact (see [JS 06], p. 184 for an example over the ringZ). A Λd-moduleN is calledflat if tensoring withN is exact on both sides. In our situation,N =Q= Quot(Λd) is equal to the quotient field of Λd, and therefore flat by Corollary 3.6 in [AM 69], proving the exactness of the sequence (?). But the dimension of vector spaces is additive on exact sequences, and therefore

rankΛd(R) + rankΛd(M) = dimQ(R⊗Q) + dimQ(M⊗Q)

= dimQ(F ⊗Q)

= rankΛd(F) = l ,

proving that rankΛd(M) =m if and only if rankΛd(R) =l−m.

Letr1, . . . , rq denote generators of R⊆F. There exist elementsai,j ∈Λd such that

ri =

l

X

j=1

aijfj , 1≤i≤q .

Now the condition rankΛd(R) =l−m is equivalent to the fact that there exists a non-vanishing minor of the matrix (aij)i,j of orderl−m, whereas every minor of order greater thanl−mis zero. Letf ∈Λddenote the non-trivial minor of (aij)i,j. By Lemma 4.11, there exists an open and dense subsetU ⊆ε(Γd) such that for everyπ ∈U,π(f) is in Weierstraß normal form, and in particular non-zero. We will show that rankΛd(Mπ) =mfor π∈U, proving (i).

We have a surjection

Fπ =F/(ker(π)·F) ψ // Mπ =M/(ker(π)·M) // 0

induced by the surjective Λd-module homomorphism F −→ψ M −→ 0, which maps ker(π)·F into ker(π)·M. The map Rπ −→ϕ Fπ induced by 0 −→ R −→ϕ F perhaps is not injective, so we divide out the kernel Xπ

and define a Λ-module ˜Rπ :=Rπ/Xπ. Then the sequence

0 //π ϕ˜ // Fπ ψ // Mπ // 0 (??) is exact, where the induced injective map ˜ϕ : ˜Rπ −→ Fπ is defined via r+ ker(ϕ) 7→ ϕ(r).

Indeed, it remains to show that ker(ψ)⊆im( ˜ϕ). Letf ∈F be such that ψ(f + ker(π)·F)∈ker(π)·M. Write

ψ(f) = X

i

αi·mi

4.2. PROJECTIVE VARIETIES AND THEµ-INVARIANT 147 We know from the first part of the proof that there exists a non-vanishing minorf ∈Λdof the matrix (aij)i,j of orderr. The setU ⊆ε(Γd) has been π(f)6= 0, this matrix has a non-vanishing minor of order r, proving that rankΛ(Rπ) ≥ rankΛ(ϕ(Rπ)) ≥ r. Let J ⊆ {1, . . . , q} denote the set of indices corresponding to the submatrix of (aij) whose determinant is the minor f.

withβj ∈ker(π)⊆Λd,j= 1, . . . , l. But then 0 =

l

X

j=1

(X

i∈I

λiaij−βj)·fj

and thereforeP

i∈Iλiaijj ∈ker(π) for all 1≤j ≤l, recalling that the fj are Λd-linear independent.

If I was equal to J, then we would obtain a non-trivial vanishing linear combination of the rows of (aij)i,j∈J in Λ = Λd/ker(π). But this would contradict the fact that

π(f) = det((aij)i,j∈J) 6= 0. Therefore, rankΛ( ˜Rπ)≥ |J|=r.

On the other hand, if rankΛ( ˜Rπ) was strictly larger than r, then there would exist a non-vanishing minor g of the matrix (aij) of order greater thanr, since ˜Rπ is generated by the cosets ofr1, . . . , rq. Sinceπ: Λd−→Λ is surjective, we could lift g to a non-vanishing minor of (aij) of order greater than r, in contradiction to the fact that rankΛd(R) =r.

(ii) Now suppose thatMπ0 is a finitely generatedZp-module. ThenMπ0 is a torsion Λ-module. Greenberg has shown that this happens only if there exists an annihilator f ∈Λdof M such that f 6∈ker(π0) (see Lemma 4.2, (i)). Furthermore, we may assume that f is not divisible by p. Indeed, µ(Mπ0) = 0 by Proposition 1.31, (iii). Therefore the characteristic poly-nomial g(T) ∈ Zp[T]⊆ Λ of Mπ0 is not divisible by p. g(T) annihilates the elementary Λ-module EMπ

0. Since the finite kernel of the pseudo-isomorphismMπ0

−→ EMπ0 may be annihilated by an appropriate power ofT, by Nakayama’s Lemma (compare Remark 3.49), we may augmentg in order to obtain an annihilatorg of Mπ0 that is still not divisible by p.

Using the arguments from the proof of Lemma 4.2, (i), it follows thatM is a torsion Λd-module, and that there exists an annihilator f ∈Λdof M such that

f ≡ gl mod (ker(π0)),

where l denotes the number of generators of the finitely generated Λd -module M. In particular, p-f.

Since there exists a surjective homomorphism (Λd/(f))l −→ M, it will suffice to prove assertion (ii) for the moduleN := Λd/(f). By Lemma 4.11 and Remarks 4.10, (2), there exists an open and dense subsetU ⊆ε(Γd) such that for every π∈U, the imageπ(f) =u·f˜is the product of a unit u∈Zp[[T]]= Λ and a distinguished polynomial ˜f ∈Zp[T]. Therefore

Nπ = Λ/(π(f)) = Λ/( ˜f) ∼= Zdeg( ˜p f)

is finitely generated overZp for everyπ ∈U.

Using this theorem, Baba˘ıcev proved his first result concerning the Iwasawa µ-invariant in Zp-extensions of K. In order to apply the theory developed so

4.2. PROJECTIVE VARIETIES AND THEµ-INVARIANT 149 far, we let Γ = Γ1 := Zp and Γd := Gal(K/K) ∼= Zdp, where K denotes the composite of allZp-extensions ofK, as usual. Then the study of surjectiveZp -module homomorphisms π ∈ε(Γd) corresponds to the study of Zp-extensions ofK (compare Lemma 2.7).

Definition 4.14. Let c∈N0. Define Eµ>c(K) to be the set ofZp-extensions L/K satisfying µ(L/K) > c. Furthermore, let E0(K) denote the set of Zp -extensionsL/K such that µ(L/K) = 0.

Theorem 4.15(Baba˘ıcev). If there exists a Zp-extensionL∈ E0(K)such that only finitely many primes ofLlie overp, then the subsetE0(K)ofE(K)is open and dense.

Proof. This is Theorem 4 in [Ba 76]. Whereas the proof given there uses coho-mology theory, we will use more elementary arguments. LetY := Gal(H(K)/K) denote the Galois group of the maximal p-abelian unramified extension of K. Then Y is a finitely generated torsion Λd-module (compare Theorem 1 in [Gr 73]). Furthermore, if π denotes the surjective homomorphism corre-sponding to L/K via Lemma 2.7, then our assumptions on L imply that Yπ := Y /(ker(π)·Y) is a finitely generated Zp-module (compare Lemma 4.3, (i)).

Theorem 4.13, (ii) implies that there exists an open and dense subsetU of ε(Gal(K/K)) such that Yπ is a finitely generated Zp-module for every π∈U.

We will now make use of the following fact.

Lemma 4.16.For everyZp-extensionL/K, and corresponding homomorphism π∈ε(Gal(K/K)), we have an exact sequence

Yπ // Xπ // Gal((H(L)∩K)/L)

of Λ-modules, where Xπ := Gal(H(L)/L) denotes the Galois group of the max-imalp-abelian unramified extensionH(L) of L, and Yπ is defined as above.

Proof. Let F denote the subfield of H(K) fixed by ker(π) ·Y ⊆ Y. Thus, Gal(H(K)/F) = ker(π)·Y and Gal(F/K)∼=Yπ =Y /(ker(π)·Y). Assume that we have chosen a set of topological generatorsγ1, . . . , γdof Gal(K/K) such that the kernel ofπ∈ε(Gal(K/K)) is generated by γ1, . . . , γd−1.

Claim 4.17. The maximal p-abelian unramified extension H(L) of L is con-tained in F.

Proof. Since ker(π)⊆Λdis generated by T1, . . . , Td−1, the subfield F of H(K) is fixed by< T1, . . . , Td−1>·Y.

Note thatH(K) actually is Galois over L. Since

< T1, . . . , Td−1>·Y ⊆ Gal(H(K)/L)

is a closed subgroup, it follows thatF ⊆H(K) is also Galois overL. We claim thatF is the maximal subextension that is abelian overL; this relies on the fact that< T1, . . . , Td−1>·Y corresponds to the topological commutator subgroup of Gal(H(K)/L), as we will see in Chapter 5 (compare Lemma 5.19).

Indeed, Gal(K/L) acts on Gal(H(K)/K) =Y via conjugation. Since γi·σ·γi−1·σ−1(x) = ((γi−1)·σ)(x) = (Ti·σ)(x) = x

for everyσ ∈Y,x∈F,i= 1, . . . , d−1, it follows thatγi·σ·γi−1(x) =σ(x) for each σ ∈Y,x∈F, i.e., γi·σ·γi−1 =σ for every σ ∈Gal(F/K), proving that F/L is abelian.

Conversely, if M ⊆H(K) is abelian overL, then γi·σ·γ−1i =σ for every σ ∈ Gal(M/K) and every i ∈ {1, . . . , d−1}, since Gal(K/L) is generated by γ1, . . . , γd−1. But then M ⊆H(K)<T1,...,Td−1>·Y =F.

Since H(L) is abelian over L, and H(L) ⊆ H(K) (compare Proposition 1.34), it is now immediate that H(L)⊆F.

Now let Xπ := Gal(H(L)/L). Then H(L) := H(L) ·K ⊆ F. We let Xπ := Gal(H(L)/K) and summarise our situation in the following diagram:

H(K)

F

ker(π)·Y

K

Y

Yπ

Xπ

H(L)

L Xπ H(L)

K

Since it is possible that K∩H(L) %L, we may not conclude that Xπ ∼=Xπ. However, since H(L)⊆F, we have a surjective map

Yπ //// Xπ , σ // σ+ Gal(F/H(L)) =: σ .

Furthermore, since Xπ = Gal(H(L)/K) ∼= Gal(H(L)/(K∩H(L))), we obtain a map Xπ −→ Xπ, induced by restriction to H(L). Note that this latter map will not be surjective wheneverH(L)∩K6=L. However, the cokernel will yield us the desired exact sequence:

First note that the composite

i:Yπ // Xπ // Xπ , σ // σ // σ|H(L) ,

is well-defined since every element in Gal(F/H(L)) fixes H(L)⊆H(L). Let j:Xπ = Gal(H(L)/L) // Gal((H(L)∩K)/L), τ // τ|(H(L)K) ,

4.2. PROJECTIVE VARIETIES AND THEµ-INVARIANT 151 denote the canonical restriction map. Then the sequence

Yπ i // Xπ

j // Gal((H(L)∩K)/L)

is exact. First of all, it is clear that the image of i is contained in ker(j), sinceσ|(H(L)K) = id∈ Gal((H(L)∩K)/L) for every σ ∈ Gal(H(L)/K). On the other hand, if τ|(H(L)K) = id for some τ ∈ Xπ = Gal(H(L)/L), then τ ∈Gal(H(L)/(H(L)∩K)), and therefore τ is the restriction toH(L) of some element in Gal(H(L)/K) = Xπ. Since Yπ Xπ is surjective, we obtain that τ =i(σ) for a suitable σ.

This shows that Xπ is a finitely generated Zp-module if Yπ is finitely gen-erated over Zp, since rankZp(Gal((K ∩ H(L))/L)) ≤ d−1. In particular, Xπ is finitely generated over Zp and therefore Λ-torsion for every π ∈U, i.e., U ⊆ E(X) in the notation of Section 4.1. The assertion of Theorem 4.15 now follows from Corollary 4.5, (ii) and Proposition 1.31, (iii).