betaFeedback

WorksMathematics

Open Homomorphisms of Global Solvably Closed Galois Groups

Abstract

We prove Uchida's conjecture on open homomorphisms of Galois groups. Every continuous open homomorphism between Galois groups of solvably closed Galois extensions of number fields is induced by a unique equivariant field embedding in the opposite direction. No restriction on the kernel or additional compatibility hypothesis is required.

Introduction

An algebraic extension of a number field is solvably closed if it has no nontrivial abelian algebraic extension. Algebraic closures and maximal prosolvable extensions are examples. The isomorphism theorems of Neukirch and Uchida show that Galois groups of such fields retain remarkably precise arithmetic information. The corresponding question for open homomorphisms asks whether this reconstruction also recovers field embeddings.

Let Ei/FiE_i/F_i be Galois extensions of number fields, with each EiE_i solvably closed, and write Gi=Gal⁡(Ei/Fi)G_i = \operatorname{Gal}(E_i/F_i). A field embedding j:E2↪E1j : E_2 \hookrightarrow E_1 induces a homomorphism α:G1→G2\alpha: G_1 \to G_2 if

g∘j=j∘α(g)(g∈G1).g \circ j = j \circ\alpha(g) \qquad(g \in G_1).

In particular, such an embedding sends F2F_2 into F1F_1. Uchida conjectured that every continuous homomorphism with open image arises in this way [15], p. 595. We prove the conjecture in its full form.

Theorem 1.1. Let F1,F2F_1, F_2 be number fields, and let Ei/FiE_i/F_i be possibly infinite solvably closed Galois extensions. Every continuous open homomorphism

α:Gal⁡(E1/F1)⟶Gal⁡(E2/F2)\alpha: \operatorname{Gal}(E_1/F_1) \longrightarrow\operatorname{Gal}(E_2/F_2)

is induced by a unique field embedding j:E2↪E1j : E_2 \hookrightarrow E_1.

The kernel is unrestricted. The proof establishes the necessary arithmetic compatibility from openness itself.

Historical context

The reconstruction of number fields from their Galois groups begins with Neukirch’s work on normal number fields and decomposition groups [8, 9]. The passage to arbitrary number fields and prescribed group isomorphisms involved Uchida, Iwasawa, and Ikeda; their contributions are recorded in the contemporary accounts [13, 10]. Uchida subsequently established the isomorphism theorem for arbitrary solvably closed Galois extensions of number fields [14].

In the paper formulating the homomorphism conjecture, Uchida proved it when the source number field is Q\mathbb{Q}, proved uniqueness in general, and established existence under a suitable local condition on decomposition groups [15], Theorem 1, Proposition 2, and Theorem 2. His local analysis also associates primes to one another whenever a chosen pro-ℓ\ell inertia group survives under the homomorphism, for an odd auxiliary prime ℓ\ell. For each fixed auxiliary prime ℓ\ell, this partial correspondence reaches all but finitely many target primes [15], Proposition 1.

A later reduction of Hoshi identifies an exact arithmetic condition for existence: an open homomorphism is induced by a field embedding if and only if it preserves the full cyclotomic character [3], Theorem 3.4. Hoshi’s subsequent work reformulates the unrestricted conjecture in terms of radicals and normality of the field fixed by a kernel [4], Theorem B. Related work on finite solvable quotients includes the isomorphism theorem of Saïdi and Tamagawa [12] and the cyclotomic-compatible homomorphism theorem of Mao and Saïdi [6]. Our proof uses the cyclotomic criterion at its endpoint. The main task is to derive that criterion from the partial local correspondences, even though their domains and exceptional sets initially depend on ℓ\ell.

The main ideas

Two arguments used below are useful beyond the final reduction. First, an additive ℓ\ell-adic character of the absolute Galois group of a number field is determined by its values on any infinite set of primes of absolute residue degree one. More precisely, a character vanishing at all those primes is zero. The logarithmic-rank input is a special case of the ℓ\ell-adic Waldschmidt–Masser theorem [16, 7, 2]. We give a complete proof using Laurent’s interpolation-determinant method [5] and a translate-and-degree zero estimate on a torus. Its variants for Frobenius averages and corestriction permit us to treat primes of arbitrary residue degree as well. These statements concern the actual unit-logarithm image and require no hypothesis on its rank.

Second, we refine the Kummer argument in Uchida’s proof of cofinite target coverage [15], Proposition 1. After logarithmic cyclotomic compatibility has been established, the number of target primes missed by the partial correspondence can be bounded independently of ℓ\ell. A missed prime supplies a Kummer class. On killing a finite cyclotomic twist, these classes lie in one character component of an SS-unit representation. The full extension degree may grow with ℓ\ell, but the multiplicity of that character is bounded by the number of orbits of SS, hence by the degree of the fixed source field. This uniformity allows us to choose ℓ\ell by simultaneous Kummer conditions and force enough corresponding primes to have the same rational residue characteristic.

Here is the route from these ingredients to Theorem 1.1. Put P=Gal⁡(Qs/Q)P=\operatorname{Gal}(\mathbb{Q}^{\mathrm{s}}/\mathbb{Q}), where Qs\mathbb{Q}^{\mathrm{s}} is the maximal prosolvable extension of Q\mathbb{Q}. For a number field FF, let GFG_F denote its absolute Galois group. It suffices to identify every continuous open homomorphism GF→PG_F\to P with ordinary restriction up to conjugation. Indeed, every solvably closed field contains Qs\mathbb{Q}^{\mathrm{s}}, and this identification will give the full cyclotomic compatibility required by Hoshi’s theorem.

We first restrict to a normal open subgroup of GQG_{\mathbb{Q}} and extend the map to the normalizer of its kernel. Write this extension as β:Gk→P\beta:G_k\to P. Following the regular-module constructions used by Uchida [14], we enlarge finite quotients of PP by elementary abelian kernels consisting of many copies of a regular representation. Each quotient R=Gal⁡(L/Q)R=\operatorname{Gal}(L/\mathbb{Q}) acts on the additive character space of GLG_L. Suitable characters distinguish a cyclic subgroup generated by a target Frobenius element from each of its proper subgroups. Comparing Frobenius averages proves that β\beta is surjective and preserves logarithmic cyclotomic characters for an infinite congruence class of auxiliary primes.

Next, pass to the preimage of PQ(i)=Gal⁡(Qs/Q(i))P_{\mathbb{Q}(i)}=\operatorname{Gal}(\mathbb{Q}^{\mathrm{s}}/\mathbb{Q}(i)). A sign character shows that, for almost every split target prime, only finitely many source primes can correspond to it as ℓ\ell varies. The uniform Kummer bound then forces a correspondence in the same residue characteristic for almost every such target prime. Finally, use rational primes split completely in suitable finite extensions. Their additive Frobenius evaluations identify the two character pullbacks, one from β\beta and one from ordinary restriction. A faithful representation in the character space identifies the group maps modulo every finite quotient; compactness supplies a single conjugation. Section 2 states the anabelian inputs and proves the local norm identity. Section 3 develops the character tests. Section 4 obtains surjectivity and logarithmic cyclotomic compatibility. Section 5 proves the uniform bound and matches residue characteristics. Section 6 identifies the map to PP and proves Theorem 1.1.

Galois groups and partial local correspondences

We collect the anabelian inputs and the local calculation that will connect Frobenius elements with cyclotomic characters. The substantial inputs from Uchida and Hoshi are stated explicitly; the remaining arguments use standard global and local class field theory, Kummer theory, and Chebotarev’s theorem.

Conventions and solvably closed fields

Fix an algebraic closure Q‾\overline{\mathbb{Q}} of Q\mathbb{Q}. For a number field M⊂Q‾M \subset\overline{\mathbb{Q}} put GM=Gal⁡(Q‾/M)G_M = \operatorname{Gal}(\overline{\mathbb{Q}}/M), and put

G=GQ,P=Gal⁡(Qs/Q),\mathcal{G} = G_{\mathbb{Q}}, \qquad P = \operatorname{Gal}(\mathbb{Q}^{\mathrm{s}}/\mathbb{Q}),

where Qs⊂Q‾\mathbb{Q}^{\mathrm{s}} \subset\overline{\mathbb{Q}} is the maximal prosolvable extension of Q\mathbb{Q}. For a number field M⊂QsM \subset\mathbb{Q}^{\mathrm{s}} write PM=Gal⁡(Qs/M)P_M = \operatorname{Gal}(\mathbb{Q}^{\mathrm{s}}/M). All homomorphisms between profinite groups and all characters in this paper are continuous. An open homomorphism is a homomorphism with open image: a continuous surjection between profinite groups is a quotient map and is open.

We use finite primes unless infinity is explicitly mentioned. For a finite prime vv, its absolute norm is NvNv, and its absolute residue degree is [κ(v):Fp][\kappa(v):\mathbb{F}_p], where pp is its rational residue characteristic. Frobenius is arithmetic. A chosen prolongation of vv determines a decomposition group DvD_v and inertia group IvI_v; a different prolongation conjugates both. Expressions involving several such groups always use specified compatible prolongations.

Lemma 2.1. Every solvably closed algebraic extension EE of Q\mathbb{Q} contains Qs\mathbb{Q}^{\mathrm{s}}. The field Qs\mathbb{Q}^{\mathrm{s}} is solvably closed. For every number field M⊂QsM \subset\mathbb{Q}^{\mathrm{s}}, the extension Qs/M\mathbb{Q}^{\mathrm{s}}/M is the maximal prosolvable extension of MM.

Proof. Since E(μn)/EE(\mu_n)/E is abelian, EE contains every root of unity. Adjoining an nnth root of an element of EE then gives a cyclic extension, so EE is closed under radicals. Finite solvable Galois extensions can be built by towers of radical extensions after adjoining roots of unity; hence Qs⊂E\mathbb{Q}^{\mathrm{s}} \subset E.

The same tower argument shows that Qs\mathbb{Q}^{\mathrm{s}} contains all roots of unity and is closed under radicals. Any nontrivial finite abelian extension would have a cyclic subextension of prime degree, which Kummer theory excludes. Any nontrivial abelian algebraic extension would have a nontrivial finite abelian subextension. Thus Qs\mathbb{Q}^{\mathrm{s}} is solvably closed.

Finally, a finite solvable Galois extension of MM lies in a finite solvable Galois extension of Q\mathbb{Q}: first take a solvable normal closure of M/QM/\mathbb{Q} inside Qs\mathbb{Q}^{\mathrm{s}}, and then the compositum of the finitely many conjugate extensions. The group of this compositum over that normal closure embeds in a product of solvable groups. This proves the last assertion. ∎

In particular, every additive ℓ\ell-adic character of GMG_M factors through PMP_M when M⊂QsM \subset\mathbb{Q}^{s}. For any solvably closed Galois extension E/FE/F, the action on roots of unity defines its cyclotomic character

χE/F:Gal⁡(E/F)⟶Z^×.\chi_{E/F} : \operatorname{Gal}(E/F) \longrightarrow\widehat{\mathbb{Z}}^{\times}.

We use χM,ℓ\chi_{M,\ell} for the usual ℓ\ell-adic character of GMG_M, and χℓ\chi_\ell for that of PP. The ℓ\ell-adic logarithm is normalized to vanish on the finite torsion subgroup of Zℓ×\mathbb{Z}_\ell^\times.

The anabelian inputs

The first input permits a homomorphism to be extended from an open subgroup to the normalizer of its kernel.

Theorem 2.2 (Uchida). Let Ei/FiE_i/F_i be solvably closed Galois extensions of number fields. Every isomorphism θ:Gal⁡(E1/F1)→∼Gal⁡(E2/F2)\theta: \operatorname{Gal}(E_1/F_1) \xrightarrow{\sim} \operatorname{Gal}(E_2/F_2) is induced by a unique field isomorphism j:E2→∼E1j : E_2 \xrightarrow{\sim} E_1, with gj=jθ(g)gj = j\theta(g) for all gg.

This is [14], Theorem. One consequence we shall use twice is the following.

Lemma 2.3. The centralizer in PP of any open subgroup of PP is trivial.

Proof. An open subgroup is PMP_M for a number field M⊂QsM \subset\mathbb{Q}^{s}. An element centralizing PMP_M is a field automorphism of Qs\mathbb{Q}^{s} implementing the identity automorphism of PMP_M. The identity field map also implements it, so uniqueness in Theorem 2.2 gives the assertion. □

Here is the local input, in the precise setting in which we need it. For a homomorphism φ:GM→PN\varphi: G_M \to P_N, an odd prime ℓ\ell, and a prime v∤ℓv \nmid\ell of MM, say that vv is active at ℓ\ell if φ\varphi is nontrivial on a pro-ℓ\ell Sylow subgroup of IvI_v. The condition is independent of the chosen Sylow subgroup and prolongation, since such choices are conjugate.

Theorem 2.4 (Uchida’s partial local correspondence). Let MM be a number field, let N⊂QsN \subset\mathbb{Q}^{s} be a number field, and let φ:GM→PN\varphi: G_M \to P_N be open. Fix an odd prime ℓ\ell. Every prime v∤ℓv \nmid\ell active at ℓ\ell determines a unique prime w∤ℓw \nmid\ell of NN for which, with suitable prolongations,

φ(Dv)⊂Dw.\varphi(D_v) \subset D_w.

Moreover, φ\varphi is injective on a pro-ℓ\ell Sylow subgroup of IvI_v, and sends this subgroup into IwI_w. All but finitely many primes of NN arise from such active primes of MM.

This is the odd-prime case of the trichotomy in [15], §1, pp. 595–596, together with its Proposition 1. In that trichotomy, nontrivial pro-ℓ\ell inertia is case (ii); case (i) kills that inertia, and the torsion case (iii) requires ℓ=2\ell= 2. The open-image hypothesis, rather than surjectivity onto PNP_N, suffices. We will apply the theorem after restricting to smaller open subgroups as well. The cofinite target coverage in this theorem is for a fixed ℓ\ell; its exceptional set is not asserted to be uniform in ℓ\ell.

The uniqueness assertion concerns primes of the finite target field NN. When restricting a map to smaller open subgroups, we will intersect a containing target decomposition group with PN′P_{N'} for a finite extension N′/NN'/N inside Qs\mathbb{Q}^{s}, and then use uniqueness at the N′N'-level. Proposition 4.3 carries out this comparison with explicit source and target fields; no uniqueness of infinite prolongations is required.

The final input converts cyclotomic compatibility into the required field embedding.

Theorem 2.5 (Hoshi). Let Ei/FiE_i/F_i be solvably closed Galois extensions of number fields, and let α:Gal⁡(E1/F1)→Gal⁡(E2/F2)\alpha:\operatorname{Gal}(E_1/F_1)\to\operatorname{Gal}(E_2/F_2) be open. There exists an equivariant field embedding E2↪E1E_2\hookrightarrow E_1 if and only if

χE2/F2∘α=χE1/F1in Z^×.\chi_{E_2/F_2}\circ\alpha=\chi_{E_1/F_1}\quad\text{in }\widehat{\mathbb{Z}}^{\times}.

We use [3 Theorem 3.4]. Uniqueness of the embedding, when it exists, is [15 Proposition 2].

Tame inertia and norms

Decomposition groups in PMP_M are the full absolute Galois groups of the corresponding local fields. Here is a justification for using the usual local ramification structure. Fix a prime of Q‾\overline{\mathbb{Q}} above a prime of MM, and form inside a local algebraic closure the union of the completions of finite fields between MM and Qs\mathbb{Q}^{\mathrm{s}}. This union contains all roots of unity and is closed under radicals: for a fixed exponent, a nonzero local element can be approximated by a global element in the field defining that completion, and their ratio can be made a power. Thus the union has no cyclic extension. Every finite Galois group over a nonarchimedean local field is solvable, by its wild inertia, tame inertia, and unramified filtration. Every closed subgroup of its prosolvable absolute Galois group is prosolvable as well; a nontrivial such group has a nontrivial finite cyclic quotient after passing through a finite solvable quotient. The union is therefore the full local algebraic closure. The usual decomposition-group identification proves the claim.

For an active pair (v,w)(v,w) from Theorem 2.4, and g∈Dvg\in D_v, let b(g)∈Z^b(g)\in\widehat{\mathbb{Z}} be its source residue exponent and let a(g)∈Z^a(g)\in\widehat{\mathbb{Z}} be the target residue exponent of φ(g)\varphi(g). These are measured using NvNv- and NwNw-Frobenius, respectively. The target residue exponent is defined on DvD_v by composition; we do not assume that the map sends all source inertia into target inertia.

Lemma 2.6. For an active pair (v,w)(v,w) at an odd prime ℓ\ell,

(Nv)b(g)=(Nw)a(g)in Zℓ×(g∈Dv).(1)(Nv)^{b(g)}=(Nw)^{a(g)}\quad\text{in }\mathbb{Z}_{\ell}^{\times}\qquad(g\in D_v). \tag*{(1)}

Powers by profinite integers are interpreted by continuous powering in the compact group Zℓ×\mathbb{Z}_{\ell}^{\times}.

Proof. Let p≠ℓp\ne\ell be the residue characteristic of vv, and choose a topological generator τ\tau of a pro-ℓ\ell Sylow subgroup of IvI_v. Its image in source tame inertia generates the pro-ℓ\ell factor. The tame conjugation relation shows that

e=gτg−1τ−(Nv)b(g)e=g\tau g^{-1}\tau^{-(Nv)^{b(g)}}

belongs to source wild inertia, a pro-pp group. In the target tame quotient, the image of ee also lies in the pro-ℓ\ell factor of tame inertia: both φ(τ)\varphi(\tau) and its conjugate do. Its image is therefore trivial, because it belongs to a pro-pp subgroup and to a pro-ℓ\ell group. The nontrivial image of φ(τ)\varphi(\tau) in the torsion-free target pro-ℓ\ell tame factor then gives (1) by the target tame conjugation relation. The target residue characteristic is different from ℓ\ell by Theorem 2.4. □\square

Additive characters and Frobenius detection

We will compare homomorphisms of Galois groups by evaluating additive ℓ\ell-adic characters at Frobenius elements. The main fact needed for this comparison is that infinitely many primes of absolute residue degree one detect a character. We first prove the logarithm-rank statement behind this fact, then establish the versions involving averaging and corestriction. The section ends with a representation contained in the character space that will supply enough characters for the comparison.

A logarithm-rank lemma

For a finite extension K/QℓK/\mathbb{Q}_{\ell}, normalize its absolute value by ∣ℓ∣ℓ=ℓ−1|\ell|_{\ell}=\ell^{-1}. The ℓ\ell-adic logarithm on OK×\mathcal{O}_K^{\times} is the continuous homomorphism that agrees with the usual power series near 11; its kernel is the finite group of roots of unity in KK. Logarithms of points in a torus are taken coordinatewise.

For a number field MM, class field theory expresses an additive character of GMG_M as a linear form on M⊗QQℓM\otimes_{\mathbb{Q}}\mathbb{Q}_{\ell}. A zero Frobenius value at a prime v∤ℓv\nmid\ell forces this form to annihilate log⁡b\log b, where (b)=vh(b)=v^h for some h>0h>0. For primes of residue degree one chosen over distinct rational primes unramified in a Galois closure, the conjugate generators have disjoint ideal supports; the following lemma turns their multiplicative independence into full logarithmic rank.

This special rank estimate admits an interpolation-determinant proof. It is a special case of the ℓ\ell-adic Waldschmidt–Masser rank theorem [16, 7]. We retain a proof using the translate estimates of that theory and Laurent’s interpolation determinants [5]; see also [2] for Theorems 5.1 and 5.8.

Lemma 3.1 (Logarithm rank). Let n,m≥1n,m\geq1, let EE be a number field embedded in a finite extension K/QℓK/\mathbb{Q}_{\ell}, and let

ui=(ui1,…,uin)∈(E×)n,1≤i≤m,u_i=(u_{i1},\ldots,u_{in})\in(E^{\times})^n,\qquad1\leq i\leq m,

have algebraic-integer coordinates that are units in KK. Suppose that for every a=(a1,…,am)∈Zm∖{0}a=(a_1,\ldots,a_m)\in\mathbb{Z}^m\setminus\{0\}, the coordinates of ua:=∏i=1muiaiu^a:=\prod_{i=1}^{m}u_i^{a_i} are multiplicatively independent modulo roots of unity. If m>n(n−1)m>n(n-1), then

span⁡K{log⁡u1,…,log⁡um}=Kn.\operatorname{span}_K\{\log u_1,\ldots,\log u_m\}=K^n.

Proof. The multiplicative hypothesis implies that the cyclic subgroup generated by every uau^a with a≠0a\neq0 is Zariski dense in Gmn\mathbb{G}_m^n. Indeed, distinct Laurent monomials have distinct values at uau^a. If a Laurent polynomial vanished at all nonnegative powers of uau^a, a Vandermonde matrix applied to its finitely many monomials would force all its coefficients to vanish.

Suppose that the logarithms span a subspace of dimension r<nr<n. The case r=0r=0 contradicts the multiplicative hypothesis because the kernel of the logarithm on local units consists of roots of unity. Thus 1≤r<n1\leq r<n, and in particular n≥2n\geq2. We will obtain an algebraic determinant whose small ℓ\ell-adic absolute value is incompatible with its archimedean size.

First make all choices needed for the analytic estimate. Replacing all uiu_i by the same positive integer power preserves the hypotheses and the logarithmic rank. Such a power puts their coordinates in a neighborhood where exponential and logarithm are inverse. The OK\mathcal{O}_K-module generated by their logarithms is free of rank rr; choose a basis b1,…,br∈Knb_1,\ldots,b_r\in K^n. After a further common ℓ\ell-power, these basis vectors have sufficiently small coordinates that

ρ:=max⁡j,t∣bt,j∣ℓsatisfiesη:=ρℓ1/(ℓ−1)<1.\rho:=\max_{j,t}|b_{t,j}|_{\ell}\quad\text{satisfies}\quad\eta:=\rho\ell^{1/(\ell-1)}<1.

Here the basis is multiplied by the same power as the logarithms. Consequently there are zi∈OKrz_i\in\mathcal{O}_K^{r} such that

log⁡ui=∑t=1rzi,tbt.\log u_i=\sum_{t=1}^{r}z_{i,t}b_t.

From now on these powered points, the number field EE, the local field KK, the basis, and η\eta are fixed. In particular, none depends on the degree parameter introduced below.

For an integer vector α=(α1,…,αn)\alpha=(\alpha_1,\ldots,\alpha_n), the monomial XαX^\alpha on these points is represented by

fα(z)=exp⁡(∑j=1nαj∑t=1rbt,jzt).f_\alpha(z)=\exp\left(\sum_{j=1}^{n}\alpha_j\sum_{t=1}^{r}b_{t,j}z_t\right).

It is analytic on a polydisc of radius greater than 1. If fα(z)=∑ν∈Z≥0rcα,νzνf_\alpha(z)=\sum_{\nu\in\mathbb{Z}_{\geq0}^{r}}c_{\alpha,\nu}z^\nu, the exponential series and ∣k!∣ℓ−1≤ℓk/(ℓ−1)|k!|_\ell^{-1}\leq\ell^{k/(\ell-1)} give the uniform bound

∣cα,ν∣ℓ≤η∣ν∣,∣ν∣=ν1+⋯+νr.|c_{\alpha,\nu}|_\ell\leq\eta^{|\nu|},\qquad|\nu|=\nu_1+\cdots+\nu_r.

The bound is independent of α\alpha, since every integer has ℓ\ell-adic absolute value at most 1. Moreover, for every a∈Z≥0ma\in\mathbb{Z}_{\geq0}^{m}, the point z(a):=∑iaiziz(a):=\sum_i a_i z_i lies in OKr\mathcal{O}_K^r, and fα(z(a))=(ua)αf_\alpha(z(a))=(u^a)^\alpha.

We next show that sufficiently many of these evaluations have full algebraic rank. For an integer D≥1D\geq1, put A=(D+1)nA=(D+1)^n and choose an integer S≥1S\geq1 such that

(S+1)m>(nD)n,S=O(Dn/m).(S+1)^m>(nD)^n,\qquad S=O(D^{n/m}).

Consider the matrix with rows indexed by α∈{0,…,D}n\alpha\in\{0,\ldots,D\}^n, columns indexed by a∈{0,…,nS}ma\in\{0,\ldots,nS\}^m, and entries (ua)α(u^a)^\alpha. We claim that its rank is AA.

Otherwise a nonzero polynomial PP of degree at most DD in each variable would vanish at every point uau^a in this grid. Over an algebraic closure of EE, define closed subsets of the torus by

Zj={x∈Gmn:P(uax)=0 for all a∈{0,…,jS}m},0≤j≤n.Z_j=\{x\in\mathbb{G}_m^n:P(u^a x)=0\text{ for all }a\in\{0,\ldots,jS\}^m\},\qquad0\leq j\leq n.

These sets are nested and nonempty, since 1∈Zn1\in Z_n, whereas dim⁡Z0≤n−1\dim Z_0\leq n-1. Hence some adjacent pair Zj,Zj+1Z_j,Z_{j+1} has the same dimension dd. Let VV be a dd-dimensional irreducible component of Zj+1Z_{j+1}. For every b∈{0,…,S}mb\in\{0,\ldots,S\}^m, the translate ubVu^bV is contained in ZjZ_j and has dimension dd, so it is an irreducible component of ZjZ_j. All these translates are distinct: an equality ubV=ub′Vu^bV=u^{b'}V for b≠b′b\ne b' would make VV stable under every power of ub−b′u^{b-b'}. The density proved above would then force V=GmnV=\mathbb{G}_m^n, contrary to V⊆Z0V\subseteq Z_0.

For completeness, a common zero set of polynomials of total degree at most δ≥1\delta\geq1 in An\mathbb{A}^n has at most δn\delta^n irreducible components. One can obtain this form of the affine Bézout bound by assigning a component YY the weight δdim⁡Ydeg⁡Y\delta^{\dim Y}\deg Y. Start with An\mathbb{A}^n, of weight δn\delta^n, and impose the equations successively. A component contained in the next hypersurface retains its weight; a component not contained in it is cut into components of dimension one less, whose total degree is at most δdeg⁡Y\delta\deg Y by Bézout. Thus the total weight does not increase, and removing redundant components can only decrease it. The same component bound holds after restriction to the torus. Applying it to ZjZ_j, whose equations have total degree at most nDnD, contradicts (3.2). This proves the rank claim.

Choose a nonzero A×AA\times A minor ΔD\Delta_D of the evaluation matrix. It is an algebraic integer in the fixed field EE. Expand its rows using the convergent power series for fαf_\alpha. A nonzero term must involve AA distinct monomials in the rr variables z1,…,zrz_1,\ldots,z_r, since repeated monomials give repeated rows in the corresponding evaluation determinant. The sum of the degrees of any AA distinct such monomials is at least crA1+1/rc_rA^{1+1/r} for all sufficiently large AA: the number of monomials of degree at most tt is (t+rr)=Or(tr)\binom{t+r}{r}=O_r(t^r), so at least half of the AA monomials have degree bounded below by a positive multiple of A1/rA^{1/r}. Their evaluations have absolute value at most 1. The nonarchimedean triangle inequality and (3.1), first for truncated series and then by convergence, therefore give a constant c>0c>0 such that

log⁡∣ΔD∣ℓ≤−cA1+1/r.\log|\Delta_D|_\ell\leq-cA^{1+1/r}.

At every archimedean embedding τ:E↪C\tau:E\hookrightarrow\mathbb{C}, an entry of this minor has absolute value at most exp⁡(CDS)\exp(CDS), with CC independent of DD. Expanding the determinant thus gives

log⁡∣τ(ΔD)∣≤C′ADS+Alog⁡A.\log|\tau(\Delta_D)|\leq C'ADS+A\log A.

Let λ\lambda be the place of EE induced by E↪KE \hookrightarrow K. Use absolute values at finite places extending the usual ones on Q\mathbb{Q}, and sum over all embeddings E↪CE \hookrightarrow\mathbb{C}, so that each complex place occurs twice. Writing pvp_v for the rational prime below a finite place vv, the product formula reads

0=∑τ:E↪Clog⁡∣τ(ΔD)∣+∑v finite[Ev:Qpv]log⁡∣ΔD∣v0 = \sum_{\tau:E\hookrightarrow\mathbb{C}} \log\lvert\tau(\Delta_D)\rvert+ \sum_{v\ \mathrm{finite}} [E_v:\mathbb{Q}_{p_v}]\log\lvert\Delta_D\rvert_v

Every term in the second sum is nonpositive because ΔD\Delta_D is an algebraic integer. Retaining its λ\lambda-term and using (3.3)-(3.4) would imply

c[Eλ:Qℓ]A1+1/r≤[E:Q](C′ADS+Alog⁡A).c[E_\lambda:\mathbb{Q}_\ell]A^{1+1/r} \leq[E:\mathbb{Q}](C'ADS+A\log A).

This is impossible as D→∞D\to\infty: the left side has order Dn+n/rD^{n+n/r}, whereas the right side is O(Dn+1+n/m+Dnlog⁡D)O(D^{n+1+n/m}+D^n\log D), and

nr≥nn−1>1+nm\frac{n}{r}\geq\frac{n}{n-1}>1+\frac{n}{m}

by m>n(n−1)m>n(n-1). This contradiction proves the lemma.

Detection, averaging, and corestriction

For a number field M⊂Q‾M\subset\overline{\mathbb{Q}}, write

Xℓ(M):=Hom⁡cont(GM,Qℓ),GM=Gal⁡(Q‾/M),X_\ell(M):=\operatorname{Hom}_{\mathrm{cont}}(G_M,\mathbb{Q}_\ell),\qquad G_M=\operatorname{Gal}(\overline{\mathbb{Q}}/M),

where Qℓ\mathbb{Q}_\ell is an additive group with trivial Galois action. Local class field theory shows that these characters are unramified at every finite prime v∤ℓv\nmid\ell: the local unit group has a pro-pp subgroup of finite index for p≠ℓp\ne\ell, and has no nonzero continuous homomorphism to Qℓ\mathbb{Q}_\ell. Thus x(Fr⁡v)x(\operatorname{Fr}_v) is defined for x∈Xℓ(M)x\in X_\ell(M) and v∤ℓv\nmid\ell.

If M/QM/\mathbb{Q} is Galois, R=Gal⁡(M/Q)R=\operatorname{Gal}(M/\mathbb{Q}) acts by (gx)(h)=x(g~−1hg~)(gx)(h)=x(\widetilde{g}^{-1}h\widetilde{g}), where g~∈GQ\widetilde{g}\in G_{\mathbb{Q}} lifts gg. The definition is independent of the lift. For a subgroup J⊆RJ\subseteq R, write

EJ=1∣J∣∑g∈JgE_J=\frac{1}{\lvert J\rvert}\sum_{g\in J}g

for the projector onto the JJ-invariants. We use res⁡L/M\operatorname{res}_{L/M} and cor⁡L/M\operatorname{cor}_{L/M} for restriction and corestriction on additive characters for a finite extension L/ML/M.

Proposition 3.2 (Frobenius detection). Let MM be a number field, let ℓ\ell be a rational prime, and let x∈Xℓ(M)x\in X_\ell(M).

(i) If x(Fr⁡v)=0x(\operatorname{Fr}_v)=0 at infinitely many primes v∤ℓv\nmid\ell of absolute residue degree one, then x=0x=0.

(ii) Suppose that M/QM/\mathbb{Q} is Galois, and fix δ∈Gal⁡(M/Q)\delta\in\operatorname{Gal}(M/\mathbb{Q}). If x(Fr⁡v)=0x(\operatorname{Fr}_v)=0 at infinitely many primes v∤ℓv\nmid\ell lying over rational primes unramified in MM whose Frobenius element at vv in Gal⁡(M/Q)\operatorname{Gal}(M/\mathbb{Q}) is δ\delta, then E⟨δ⟩x=0E_{\langle\delta\rangle}x=0.

(iii) If cor⁡M/Qx≠0\operatorname{cor}_{M/\mathbb{Q}}x\ne0, then x(Fr⁡v)=0x(\operatorname{Fr}_v)=0 at only finitely many primes v∤ℓv\nmid\ell, with no restriction on their residue degrees.

Proof. We first recall the class-field-theoretic description of the character space. Let VM=M⊗QQℓV_M=M\otimes_{\mathbb{Q}}\mathbb{Q}_\ell, and let UM⊆VMU_M\subseteq V_M be the Qℓ\mathbb{Q}_\ell-linear span of the logarithms of the global units, with logarithms taken at all places above ℓ\ell. Then

Xℓ(M)≃{λ∈Hom⁡Qℓ(VM,Qℓ):λ(UM)=0}.(2)X_\ell(M)\simeq\{\lambda\in\operatorname{Hom}_{\mathbb{Q}_\ell}(V_M,\mathbb{Q}_\ell):\lambda(U_M)=0\}. \tag*{(2)}

Indeed, pass to the inverse limit of the ray class sequences with moduli supported above ℓ\ell. Local logarithms identify the rationalized local unit groups with VMV_M; the global units impose exactly the subspace UMU_M. The ordinary class group and the factors at real places are finite, so contribute no additive Qℓ\mathbb{Q}_{\ell}-characters. This description uses the actual logarithmic image of the units and makes no assertion about its dimension.

For v∤ℓv \nmid\ell, choose hv>0h_v > 0 and bv∈OMb_v \in\mathcal{O}_M with (bv)=vhv(b_v)=v^{h_v}. Such a choice is possible because the ideal class group is finite. Principal reciprocity, with arithmetic Frobenius, gives for the linear form corresponding to xx the identity

λ(log⁡bv)=−hvx(Fr⁡v).(3)\lambda(\log b_v)=-h_vx(\operatorname{Fr}_v). \tag*{(3)}

Here bvb_v is a unit at every place above ℓ\ell; all other local unit contributions are killed by xx.

For (i), set n=[M:Q]n=[M:\mathbb{Q}] and choose m>n(n−1)m>n(n-1) of the zero primes, over distinct rational primes unramified in a Galois closure EE of MM. Let bib_i be the elements just chosen for these primes, and let

ui=(σ(bi))σ:M↪E∈(E×)n.u_i=(\sigma(b_i))_{\sigma:M\hookrightarrow E}\in(E^\times)^n.

All coordinates are algebraic integers and are units at every place of EE above ℓ\ell. We verify the multiplicative hypothesis of Lemma 3.1 by valuations.

Put G=Gal⁡(E/Q)G=\operatorname{Gal}(E/\mathbb{Q}) and H=Gal⁡(E/M)H=\operatorname{Gal}(E/M). For a selected degree-one prime viv_i and a prime pi\mathfrak{p}_i of EE above it, the decomposition group DiD_i is contained in HH. In fact, the rational prime is unramified in EE, and the degree-one condition says that [Di:Di∩H]=1[D_i:D_i\cap H]=1. The support of viOEv_i\mathcal{O}_E is indexed by H/DiH/D_i. Its translates corresponding to the embeddings M↪EM\hookrightarrow E, indexed by G/HG/H, have supports indexed by the disjoint sets gH/DigH/D_i. Distinct indices ii involve distinct rational primes. It follows that the mnmn numbers σ(bi)\sigma(b_i) have pairwise disjoint nonempty supports of their principal ideals. They are therefore multiplicatively independent modulo roots of unity. In particular, for any nonzero a∈Zma\in\mathbb{Z}^m, the nn coordinates of uau^a are multiplicatively independent modulo roots of unity.

Embed EE in a finite extension K/QℓK/\mathbb{Q}_{\ell}. Under the splitting

(M⊗QQℓ)⊗QℓK≃∏σ:M↪EK,(M\otimes_{\mathbb{Q}}\mathbb{Q}_{\ell})\otimes_{\mathbb{Q}_{\ell}}K\simeq\prod_{\sigma:M\hookrightarrow E}K,

the vector log⁡bi\log b_i becomes log⁡ui\log u_i. Lemma 3.1 says that these vectors span the right side. Equations (2) and (3) therefore force λ=0\lambda=0, and hence x=0x=0.

For (ii), put C=⟨δ⟩C=\langle\delta\rangle, F=MCF=M^C, and y=ECxy=E_Cx. Since δ\delta fixes every indicated prime vv, averaging preserves its Frobenius value, so y(Fr⁡v)=0y(\operatorname{Fr}_v)=0. The restriction–corestriction identity gives

res⁡M/Fcor⁡M/Fy=∑g∈Cgy=∣C∣y.\operatorname{res}_{M/F}\operatorname{cor}_{M/F}y=\sum_{g\in C}gy=|C|y.

Thus yy is the restriction of z=∣C∣−1cor⁡M/Fy∈Xℓ(F)z=|C|^{-1}\operatorname{cor}_{M/F}y\in X_\ell(F). At an indicated prime, the decomposition group of M/QM/\mathbb{Q} is exactly CC. The prime u=v∩Fu=v\cap F consequently has absolute residue degree one, and

0=y(Fr⁡v)=[κ(v):κ(u)]z(Fr⁡u).0=y(\operatorname{Fr}_v)=[\kappa(v):\kappa(u)]z(\operatorname{Fr}_u).

Infinitely many distinct uu occur because M/FM/F is finite. Part (i) gives z=0z=0, and therefore y=0y=0.

For (iii), suppose that infinitely many zero primes exist. Let E/QE/\mathbb{Q} be a finite Galois closure of MM, put R=Gal⁡(E/Q)R=\operatorname{Gal}(E/\mathbb{Q}), and let y=res⁡E/Mxy=\operatorname{res}_{E/M}x. Discarding finitely many primes, choose primes of EE above these zero primes whose rational primes are unramified in EE. Restriction multiplies a Frobenius value by the relative residue degree, so they are zero primes for yy. An infinite subset has one fixed Frobenius element δ∈R\delta\in R. By (ii), E⟨δ⟩y=0E_{\langle\delta\rangle}y=0, and hence ERy=0E_Ry=0. It follows that

res⁡E/Qcor⁡E/Qy=∣R∣ERy=0.\operatorname{res}_{E/\mathbb{Q}}\operatorname{cor}_{E/\mathbb{Q}}y=|R|E_Ry=0.

Restriction on additive characters across a finite extension is injective: a character that vanishes on a subgroup of finite index has finite image in the torsion-free group Qℓ\mathbb{Q}_{\ell}, and so is zero. On the other hand, transitivity and the corestriction–restriction identity give

cor⁡E/Qy=cor⁡M/Qcor⁡E/Mres⁡E/Mx=[E:M]cor⁡M/Qx≠0,\operatorname{cor}_{E/\mathbb{Q}} y = \operatorname{cor}_{M/\mathbb{Q}} \operatorname{cor}_{E/M} \operatorname{res}_{E/M} x = [E : M]\operatorname{cor}_{M/\mathbb{Q}} x \ne0,

a contradiction.

A signed representation in the character space

The class field description also controls the supply of characters when the number field varies. The regular representation gives upper bounds for dimensions of subgroup invariants, while the signed representation below gives lower bounds and ensures a faithful action on the character space.

Proposition 3.3. Let M/QM/\mathbb{Q} be a finite, totally imaginary Galois extension, put R=Gal⁡(M/Q)R = \operatorname{Gal}(M/\mathbb{Q}), and let c∈Rc \in R be a complex conjugation. For every rational prime ℓ\ell, there are RR-equivariant embeddings

WM:=Ind⁡⟨c⟩R(sign⁡)↪Xℓ(M)↪Qℓ[R],(4)W_M := \operatorname{Ind}_{\langle c\rangle}^{R}(\operatorname{sign}) \xhookrightarrow{} X_{\ell}(M) \xhookrightarrow{}\mathbb{Q}_{\ell}[R], \tag*{(4)}

where sign⁡(c)=−1\operatorname{sign}(c) = -1. In particular, the action of RR on Xℓ(M)X_{\ell}(M) is faithful.

Proof. By the normal basis theorem, VM=M⊗QQℓV_M = M \otimes_{\mathbb{Q}} \mathbb{Q}_{\ell} is the regular representation of RR, as is its dual. The naturality of (2) gives the second embedding.

The equivariant Dirichlet unit theorem [1 Section 3] identifies the representation OM×⊗ZQℓ\mathcal{O}_M^{\times} \otimes_{\mathbb{Z}} \mathbb{Q}_{\ell} with the permutation representation on the archimedean places minus its trivial summand. Since these places are indexed by R/⟨c⟩R/\langle c\rangle, writing Ualg=OM×⊗ZQℓU^{\mathrm{alg}} = \mathcal{O}_M^{\times} \otimes_{\mathbb{Z}} \mathbb{Q}_{\ell} gives

Qℓ⊕Ualg≃Ind⁡⟨c⟩R1.\mathbb{Q}_{\ell} \oplus U^{\mathrm{alg}} \simeq\operatorname{Ind}_{\langle c\rangle}^{R} 1.

The character identity follows first over the real numbers from the usual logarithmic unit embedding, and hence over Qℓ\mathbb{Q}_{\ell} by semisimplicity in characteristic zero. In particular, UalgU^{\mathrm{alg}} is self-dual. Its ℓ\ell-adic logarithmic image UMU_M is a quotient of UalgU^{\mathrm{alg}}; no injectivity of the logarithm on this tensor product is needed.

The decomposition of the regular representation induced from the two characters of ⟨c⟩\langle c\rangle is

Qℓ[R]≃Ind⁡⟨c⟩R1⊕WM≃Qℓ⊕Ualg⊕WM.\mathbb{Q}_{\ell}[R] \simeq\operatorname{Ind}_{\langle c\rangle}^{R} 1 \oplus W_M \simeq\mathbb{Q}_{\ell} \oplus U^{\mathrm{alg}} \oplus W_M.

Since UM∨U_M^{\vee} is a subrepresentation of (Ualg)∨≃Ualg(U^{\mathrm{alg}})^{\vee} \simeq U^{\mathrm{alg}}, semisimplicity and Xℓ(M)≃(VM/UM)∨X_{\ell}(M) \simeq(V_M/U_M)^{\vee} show that Xℓ(M)X_{\ell}(M) contains Qℓ⊕WM\mathbb{Q}_{\ell} \oplus W_M. This proves the first embedding in (4).

Finally, realize WMW_M as the direct sum of the signed lines indexed by the left cosets of ⟨c⟩\langle c\rangle. An element acting trivially must fix the line of the identity coset, and therefore belongs to ⟨c⟩\langle c\rangle. The nonidentity element cc acts by −1-1 on that line. Thus only the identity acts trivially on WMW_M, proving faithfulness.

Normalizer extension and Frobenius generation

Let FF be a number field and let

α0:GF⟶P\alpha_0 : G_F \longrightarrow P

be a continuous open homomorphism. Our eventual aim is to identify this map with ordinary restriction, up to conjugation in PP. We first enlarge an open part of its domain in a canonical way. For the enlarged map we will prove surjectivity and equality of cyclotomic logarithms at an infinite set of auxiliary primes. The main step is to show that images of local Frobenius elements generate prescribed cyclic subgroups in suitable finite quotients of PP.

Choose an open normal subgroup U⊴GU \trianglelefteq G contained in GFG_F, and put

A=α0(U),N=ker⁡(α0∣U),T=NG(N).A = \alpha_0(U), \qquad N = \ker(\alpha_0|_U), \qquad T = N_G(N).

The subgroup AA is open in PP. Since U⊂TU \subset T, the normalizer TT is open and closed in GG; write T=GkT = G_k.

Lemma 4.1 (Extension to the normalizer). There is a unique continuous open homomorphism

β:T⟶P\beta:T \longrightarrow P

whose restriction to UU is α0∣U\alpha_0|_U.

Proof. Conjugation by t∈Tt \in T induces a continuous automorphism of U/N=AU/N=A. By Theorem 2.2, applied to Qs/(Qs)A\mathbb{Q}^{\mathrm{s}}/(\mathbb{Q}^{\mathrm{s}})^A, this automorphism is conjugation by a unique element β(t)\beta(t) of PP. Indeed, the implementing field automorphism of Qs\mathbb{Q}^{\mathrm{s}} fixes Q\mathbb{Q}. Uniqueness shows both that β\beta is a homomorphism and that β(u)=α0(u)\beta(u)=\alpha_0(u) for u∈Uu \in U. It also shows that any extension of α0∣U\alpha_0|_U to TT must equal β\beta: its values must induce these same conjugations on AA.

Continuity on the open subgroup UU implies continuity on TT. Moreover, β(T)\beta(T) contains AA and is therefore open in PP. A continuous homomorphism of profinite groups is open onto its image, so β\beta is open.

We keep UU, AA, NN, TT, kk, and β\beta fixed for the rest of the argument. List the distinct conjugates of NN other than NN itself as

Ni=tiNti−1(1≤i≤r),ti∉T,Hi=β(Ni).N_i=t_iNt_i^{-1}\quad(1\le i\le r),\qquad t_i\notin T,\qquad H_i=\beta(N_i).

There are finitely many because UU normalizes NN. Normality of UU in GG implies that each NiN_i lies in UU and is normal in UU, so HiH_i is normal in AA. Each HiH_i is nontrivial. Otherwise tiNti−1⊂Nt_iNt_i^{-1}\subset N, and iterating this inclusion would give a descending chain returning to NN: some power of tit_i lies in UU and hence normalizes NN. All inclusions would be equalities, contrary to ti∉Tt_i\notin T. We allow r=0r=0, with sums indexed by ii then equal to zero.

A finite quotient with enough characters

The next lemma constructs characters that distinguish a proper subgroup of a cyclic group while avoiding the invariant spaces attached to the HiH_i. The latter condition will prevent cancellation by conjugates of NN when we average a pulled-back character on the source. For a finite subgroup JJ acting on a characteristic-zero vector space, we use the averaging operator EJ=∣J∣−1∑g∈JgE_J=|J|^{-1}\sum_{g\in J}g.

Lemma 4.2 (Quotient amplification). Fix an odd prime ℓ\ell and a finite quotient R0R_0 of PP. There is a finite Galois extension L/QL/\mathbb{Q} in Qs\mathbb{Q}^{\mathrm{s}}, totally imaginary, such that the quotient R=Gal⁡(L/Q)R=\operatorname{Gal}(L/\mathbb{Q}) dominates R0R_0 and has the following property. For every γ∈R\gamma\in R, with C=⟨γ⟩C=\langle\gamma\rangle, and every proper subgroup D⪇CD\lneq C, there is x∈Xℓ(L)x\in X_\ell(L) satisfying

x∈Xℓ(L)D,ECx=0,x∉∑i=1rXℓ(L)Hi.(5)x\in X_\ell(L)^D,\qquad E_Cx=0,\qquad x\notin\sum_{i=1}^{r}X_\ell(L)^{H_i}. \tag*{(5)}

Here each HiH_i acts through its image in RR.

Proof. Choose a quotient R1=Gal⁡(L1/Q)R_1 = \operatorname{Gal}(L_1/\mathbb{Q}) dominating R0R_0 such that every HiH_i has nontrivial image in R1R_1, the kernel of P→R1P \to R_1 is contained in AA, and L1L_1 contains Q(i)\mathbb{Q}(i). These are finitely many compatible conditions on an open normal subgroup of PP. Put n=∣R1∣n = |R_1|.

We will enlarge R1R_1 while keeping element orders bounded by 2n2n. At the same time, the images of the HiH_i and the conjugacy class of complex conjugation will grow. Growth of the HiH_i makes the sum of their invariant spaces small. Growth of the conjugacy class makes the signed representation supply enough DD-invariant vectors killed by ECE_C to escape that sum.

We use the regular-module auxiliary extension construction appearing in Uchida’s proof [14], pp. 360–362, with quantitative bounds on its invariant spaces. More precisely, we construct a further quotient with kernel consisting of mm copies of the regular F2[R1]\mathbb{F}_2[R_1]-module, where mm will be chosen below. Choose mm distinct rational primes that split completely in L1L_1. For each of them, select one prime of L1L_1 above it. Approximation gives elements a1,…,am∈L1×a_1,\ldots,a_m \in L_1^\times such that aja_j has valuation one at its selected prime and valuation zero at every other prime in these mm sets. The mnmn elements

g(aj),g∈R1,1≤j≤m,g(a_j), \qquad g \in R_1,\quad1 \leq j \leq m,

have independent square classes: their valuations at the selected sets of primes form a permutation of the identity matrix modulo two. Adjoin their square roots and call the resulting field LL. Kummer theory gives

1⟶V⟶R=Gal⁡(L/Q)⟶R1⟶1,V≃F2[R1]m(6)1 \longrightarrow V \longrightarrow R = \operatorname{Gal}(L/\mathbb{Q}) \longrightarrow R_1 \longrightarrow1,\qquad V \simeq\mathbb{F}_2[R_1]^m \tag*{(6)}

as R1R_1-modules. The dual module supplied by Kummer theory is again the regular permutation module. The field LL is Galois over Q\mathbb{Q} because all conjugates of the aja_j were included; its group is solvable, so L⊂QsL \subset\mathbb{Q}^s. No splitting of (6) is required.

Every element of RR has order at most 2n2n, since its nnth power lies in the exponent-two group VV. Write ARA_R and Hi,RH_{i,R} for the images of AA and HiH_i in RR. The choice of R1R_1 gives V⊂ARV \subset A_R. Choose hi∈Hi,Rh_i \in H_{i,R} with nonidentity image gi∈R1g_i \in R_1, of order di>1d_i > 1. As Hi,RH_{i,R} is normal in ARA_R,

(gi−1)V=[hi,V]⊂Hi,R.(g_i-1)V=[h_i,V]\subset H_{i,R}.

On mm regular modules the rank of gi−1g_i-1 is mn(1−1/di)≥mn/2mn(1-1/d_i)\geq mn/2. Consequently

∣Hi,R∣≥2mn/2.|H_{i,R}| \geq2^{mn/2}.

If c∈Rc \in R is a complex conjugation, its image in R1R_1 is a nontrivial involution because L1L_1 is totally imaginary. Conjugation by VV alone therefore gives at least 2mn/22^{mn/2} distinct conjugates of cc; thus

∣cR∣≥2mn/2.|c^R| \geq2^{mn/2}.

Set η=2−mn/2\eta= 2^{-mn/2} and write X=Xℓ(L)X=X_\ell(L). By Proposition 3.3, XX embeds in the regular Qℓ[R]\mathbb{Q}_\ell[R]-representation and contains

W=Ind⁡⟨c⟩R(sign⁡).W=\operatorname{Ind}_{\langle c\rangle}^{R}(\operatorname{sign}).

The regular-representation bound and (4.3) give

dim⁡(∑i=1rXHi)≤∑i=1r∣R∣∣Hi,R∣≤rη∣R∣.(7)\dim\left(\sum_{i=1}^{r}X^{H_i}\right)\leq\sum_{i=1}^{r}\frac{|R|}{|H_{i,R}|}\leq r\eta|R|. \tag*{(7)}

The character of WW takes the value ∣R∣/2|R|/2 at the identity, the value −∣R∣/(2∣cR∣)-|R|/(2|c^R|) on cRc^R, and zero elsewhere. Hence, for every subgroup J⊂RJ\subset R,

∣dim⁡WJ−∣R∣2∣J∣∣≤η∣R∣2.\left|\dim W^J-\frac{|R|}{2|J|}\right|\leq\frac{\eta|R|}{2}.

For D<C=⟨γ⟩D < C = \langle\gamma\rangle, the map EC:WD→WCE_C : W^D \to W^C is surjective. Since ∣D∣≤∣C∣/2|D| \le|C|/2 and ∣C∣≤2n|C| \le2n, (4.6) implies

dim⁡ker⁡(EC∣WD)=dim⁡WD−dim⁡WC≥∣R∣2(1∣D∣−1∣C∣)−η∣R∣≥∣R∣(14n−η).\begin{aligned} \dim\ker(E_C|_{W^D}) &= \dim W^D - \dim W^C \\ &\ge\frac{|R|}{2}\left(\frac{1}{|D|} - \frac{1}{|C|}\right) - \eta|R| \\ &\ge|R|\left(\frac{1}{4n} - \eta\right). \end{aligned}

Choose the integer m≥1m \ge1 so large that

(r+1)2−mn/2<14n.(r+1)2^{-mn/2} < \frac{1}{4n}.

This single choice works for all γ\gamma and all proper D<CD < C. The last dimension is then greater than (7), so its kernel contains a vector outside the indicated sum. This vector gives (5). If CC is trivial, there is no proper subgroup to consider.

Generation by local Frobenius images

We now use these characters to strengthen the local correspondence. Containment of a source decomposition group in a target decomposition group places its Frobenius image in a cyclic subgroup of a finite quotient. The next proposition shows that, after enlarging the quotient, infinitely many such images generate the entire cyclic subgroup.

Proposition 4.3 (Frobenius generation). Fix an odd prime ℓ\ell, a finite quotient R0R_0 of PP, and γ0∈R0\gamma_0 \in R_0. There are a quotient

P↠R=Gal⁡(L/Q)↠R0P \twoheadrightarrow R = \operatorname{Gal}(L/\mathbb{Q}) \twoheadrightarrow R_0

and a lift γ∈R\gamma\in R of γ0\gamma_0 with the following property. There are infinitely many ℓ\ell-active primes vv of kk, over distinct rational primes p≠ℓp \ne\ell, for which compatible prolongations can be chosen so that:

  1. the corresponding target rational prime q≠ℓq \ne\ell is unramified in LL, and the target prolongation restricts to a prime w′w' of LL whose Frobenius in RR is γ\gamma;

  1. the image in RR of an arithmetic Frobenius lift at vv generates ⟨γ⟩\langle\gamma\rangle.

Proof. Choose RR by Lemma 4.2 and choose any lift γ\gamma of γ0\gamma_0. Put C=⟨γ⟩C = \langle\gamma\rangle. Take an open normal subgroup U′⊴GU' \trianglelefteq G contained in U∩β−1(PL)U \cap\beta^{-1}(P_L); for example, take the core of this open subgroup in GG. Let F′/QF'/\mathbb{Q} be the finite Galois extension with GF′=U′G_{F'} = U'. The map

β∣U′:GF′⟶PL\left.\beta\right|_{U'} : G_{F'} \longrightarrow P_L

is open.

Chebotarev’s theorem supplies infinitely many primes w′w' of LL, over distinct rational primes q≠ℓq \ne\ell unramified in LL, with Frobenius exactly γ\gamma. Theorem 2.4 applied to β∣U′\left.\beta\right|_{U'} pairs all but finitely many of them with ℓ\ell-active primes v′v' of F′F', away from ℓ\ell. Uniqueness of the target prime for an active source prime implies that infinitely many v′v' occur. Since only finitely many primes of F′F' lie over a given rational prime, we may pass to a subset with distinct source rational primes pp, all unramified in F′F'. Passing to another infinite subset, we may also assume that their Frobenius elements in Gal⁡(F′/Q)\operatorname{Gal}(F'/\mathbb{Q}) are a fixed element δ\delta. Fix a source prolongation above each v′v', and use it also for the prime v=v′∣kv = v'|_k. The original correspondence at v′v' can be made with this prolongation: changing the source prolongation conjugates the target group by an element of β(U′)⊂PL\beta(U') \subset P_L, which preserves the finite prime w′w'. Activity passes from v′v' to vv, since the relevant inertia subgroup for U′U' is contained in that for TT. Apply Theorem 2.4 at vv using this same source prolongation. If w1w_1 is the restriction to LL of its target prolongation, then

β(Dv′)⊂β(Dv)∩PL⊂Dw1.\beta(D_{v'}) \subset\beta(D_v) \cap P_L \subset D_{w_1}.

Uniqueness of the target prime for v′v' at the level of LL gives w1=w′w_1 = w'. Consequently the corresponding target rational prime is qq, and the image of DvD_v in RR lies in CC. These changes of prolongations preserve v′v' and w′w', so they preserve the previously selected finite Frobenius elements δ\delta and γ\gamma.

Put H=T/U′⊂G/U′H = T/U' \subset G/U' and let ff be the residue degree of vv over pp. Figure 1 records the primes involved. Here δ=Fr⁡(v′/p)\delta= \operatorname{Fr}(v'/p) and γ=Fr⁡(w′/q)\gamma= \operatorname{Fr}(w'/q) belong to the finite Galois groups over Q\mathbb{Q}. The symbols Fr⁡v′\operatorname{Fr}_{v'} and Fr⁡w′\operatorname{Fr}_{w'} used below instead denote local arithmetic Frobenius over F′F' and LL, respectively.

Figure 1. Vertical arrows restrict primes; dashed arrows are the partial local correspondences. The residue degree of v/pv/p is ff. No equality between pp and qq is assumed at this stage.

In the decomposition group ⟨δ⟩⊂Gal⁡(F′/Q)\langle\delta\rangle\subset\operatorname{Gal}(F'/\mathbb{Q}) we have

⟨δ⟩∩H=⟨δf⟩,f=min⁡{j≥1:δj∈H}.(8)\langle\delta\rangle\cap H = \langle\delta^f\rangle,\qquad f = \min\{j \ge1 : \delta^j \in H\}. \tag*{(8)}

Indeed, the residue degree is the orbit size of the distinguished coset in the action of ⟨δ⟩\langle\delta\rangle on Gal⁡(F′/Q)/H\operatorname{Gal}(F'/\mathbb{Q})/H. This description does not require HH to be normal. In particular ff is constant on the chosen infinite set. The homomorphism β\beta induces H→RH \to R, since β(U′)⊂PL\beta(U') \subset P_L; define

d=β(δf)∈R.d = \beta(\delta^f) \in R.

This notation is independent of the lift of δf\delta^f to TT. A local lift of rational Frobenius to the ffth power is a Frobenius lift at vv, so the preceding containment gives d∈Cd \in C.

We claim that D=⟨d⟩D = \langle d\rangle equals CC. Suppose that D<CD < C, and choose xx as in (5). Pullback along β∣U′\beta|_{U'} gives a linear map

j:Xℓ(L)⟶Xℓ(F′),y=j(x).j : X_\ell(L) \longrightarrow X_\ell(F'),\qquad y = j(x).

The map jj is injective: its defining homomorphism has open image in PLP_L, and an additive Qℓ\mathbb{Q}_\ell-valued character vanishing on an open subgroup vanishes everywhere. Frobenius evaluation at w′w' is invariant under γ\gamma, because γ\gamma fixes w′w' and acts trivially by conjugation on its unramified local quotient. Therefore

x(Fr⁡w′)=(ECx)(Fr⁡w′)=0.x(\operatorname{Fr}_{w'}) = (E_Cx)(\operatorname{Fr}_{w'}) = 0.

The character xx is unramified at w′w', so this equality says that it vanishes on the entire decomposition group there. It follows that y(Fr⁡v′)=0y(\operatorname{Fr}_{v'})=0 for every prime in our infinite set. Proposition 3.2(ii) now gives

E⟨δ⟩y=0.E_{\langle\delta\rangle}y=0.

To see why this contradicts the choice of xx, write a=∣⟨δ⟩∣a=|\langle\delta\rangle| and Y=j(Xℓ(L))Y=j(X_{\ell}(L)). Pullback is TT-equivariant, where the action on the target character space is through β\beta. In particular yy is fixed by NN. In the sum defining (4.8), the powers δj\delta^j belonging to HH are exactly those for which f∣jf\mid j, by (4.7). Each such power acts as the identity on yy, because xx is DD-invariant. If f∤jf\nmid j, choose a lift tt of δj\delta^j to GG. Then t∉Tt\notin T, and δjy\delta^j y is fixed by tNt−1tNt^{-1}. Multiplying (4.8) by aa thus yields

0=afy+∑0≤j<af∤jδjy,y∈∑i=1rXℓ(F′)Ni.0=\frac{a}{f}y+\sum_{\substack{0\leq j<a\\ f\nmid j}}\delta^j y,\qquad y\in\sum_{i=1}^{r}X_{\ell}(F')^{N_i}.

It remains to return this membership to the target character space. The finite group U/U′U/U' acts on Xℓ(F′)X_{\ell}(F'), and its subspace YY is stable. Semisimplicity supplies a UU-equivariant projection

π:Xℓ(F′)⟶Y.\pi:X_{\ell}(F')\longrightarrow Y.

Every NiN_i lies in UU, so π\pi carries NiN_i-invariants into YNiY^{N_i}. Equivariance and injectivity of jj give

YNi=j(Xℓ(L)Hi).Y^{N_i}=j\left(X_{\ell}(L)^{H_i}\right).

Applying π\pi to the displayed membership and then applying j−1j^{-1} contradicts the last condition in (4.1). Only UU-equivariance of the projection is used; no normality of TT in GG is needed. Thus D=CD=C, proving the proposition.

Surjectivity and cyclotomic logarithms

The generation statement now gives the two properties needed in the next stage. The first removes any restriction on the target image. For the second, a congruence condition on ℓ\ell forces the source primes supplied by the proposition to have residue degree one, so the character-detection theorem applies.

Theorem 4.4. The extension β:Gk→P\beta:G_k\to P is surjective. Put

m0=[k:Q]!,L={ℓ:ℓ is an odd prime and ℓ≡1(modm0)}.m_0=[k:\mathbb{Q}]!,\qquad\mathcal{L}=\{\ell:\ell\text{ is an odd prime and }\ell\equiv1\pmod{m_0}\}.

The set L\mathcal{L} is infinite, and for every ℓ∈L\ell\in\mathcal{L},

log⁡χℓ∘β=log⁡χk,ℓas additive characters on Gk.(9)\log\chi_{\ell}\circ\beta=\log\chi_{k,\ell}\quad\text{as additive characters on }G_k. \tag*{(9)}

Here χℓ\chi_{\ell} is the ℓ\ell-adic cyclotomic character on PP, and χk,ℓ\chi_{k,\ell} is the ordinary cyclotomic character on GkG_k.

Proof. Let R0R_0 be any finite quotient of PP and let γ0∈R0\gamma_0\in R_0. Proposition 4.3 produces an element dd in the image of β\beta in a further quotient RR such that ⟨d⟩=⟨γ⟩\langle d\rangle=\langle\gamma\rangle, where γ\gamma lifts γ0\gamma_0. Thus γ\gamma, and hence γ0\gamma_0, belongs to the image. Since β(T)\beta(T) is compact, surjectivity onto every finite quotient implies β(T)=P\beta(T)=P.

Dirichlet’s theorem on primes in arithmetic progressions shows that L\mathcal{L} is infinite. Fix ℓ∈L\ell\in\mathcal{L}, take

R0=Gal⁡(Q(μℓ)/Q)≃Fℓ×.R_0=\operatorname{Gal}(\mathbb{Q}(\mu_{\ell})/\mathbb{Q})\simeq\mathbb{F}_{\ell}^{\times}.

and choose γ0\gamma_0 to generate this cyclic group. The proposition gives infinitely many active primes vv of kk over distinct rational primes p≠ℓp \ne\ell, for which the image of a Frobenius lift σv\sigma_v acts on μℓ\mu_\ell by a generator. Let qq be the corresponding target rational prime and let f=[κ(v):Fp]f=[\kappa(v):\mathbb{F}_p], so Nv=pfNv=p^f. If a(σv)∈Z^a(\sigma_v)\in\widehat{\mathbb{Z}} is the target residue exponent, Lemma 2.6 gives

pf=qa(σv)in Zℓ×.(10)p^f=q^{a(\sigma_v)}\quad\text{in }\mathbb{Z}_\ell^\times. \tag*{(10)}

Its reduction modulo ℓ\ell says that pfp^f generates Fℓ×\mathbb{F}_\ell^\times. In particular gcd⁡(f,ℓ−1)=1\gcd(f,\ell-1)=1. But 1≤f≤[k:Q]1\le f\le[k:\mathbb{Q}], and every prime divisor of any f>1f>1 in this range divides m0m_0, hence divides ℓ−1\ell-1. Consequently f=1f=1.

Taking logarithms in (10) gives

(log⁡χℓ∘β)(σv)=log⁡p=log⁡χk,ℓ(σv).(\log\chi_\ell\circ\beta)(\sigma_v)=\log p=\log\chi_{k,\ell}(\sigma_v).

The difference is an element of Xℓ(k)X_\ell(k) vanishing at infinitely many primes of absolute residue degree one. Proposition 3.2(i) proves (4.9).

Matching residue characteristics

We now show that almost every prime of Q(i)\mathbb{Q}(i) above a split rational prime has an active partner of the same residue characteristic. The auxiliary prime defining activity may vary with the pair. Two finiteness statements make this possible: almost every such target prime has only finitely many possible partners, and the number of target primes missed at a fixed auxiliary prime admits a uniform bound.

Keep the surjection β:Gk→P\beta:G_k\to P and the set L\mathcal{L} from Theorem 4.4, and put

B=Q(i),GK=β−1(PB),βB=β∣GK:GK↠PB.B=\mathbb{Q}(i),\qquad G_K=\beta^{-1}(P_B),\qquad\beta_B=\beta|_{G_K}:G_K\twoheadrightarrow P_B.

Thus K/kK/k is quadratic. An active pair at ℓ\ell in this section means a source prime vv of KK active at ℓ\ell for βB\beta_B, together with its corresponding prime ww of BB from Theorem 2.4. In particular, both primes lie away from ℓ\ell. For a prime ww of BB, define

Vw={v:(v,w) is an active pair at some ℓ∈L}.\mathcal{V}_w=\{v:(v,w)\text{ is an active pair at some }\ell\in\mathcal{L}\}.

Finite fibers and a uniform bound on missed primes

Lemma 5.1. For all but finitely many primes ww of BB above rational primes split in BB, the set Vw\mathcal{V}_w is finite.

Proof. Fix r∈Lr\in\mathcal{L}. Proposition 3.3 supplies a nonzero character ψ∈Xr(B)\psi\in X_r(B) on which Gal⁡(B/Q)\operatorname{Gal}(B/\mathbb{Q}) acts by its nontrivial character. By Proposition 3.2(i), ψ(Fr⁡w)\psi(\operatorname{Fr}_w) is nonzero at all but finitely many degree-one primes w∤rw\nmid r. Fix such a ww above a split rational prime q≠rq\ne r, and put

x0=(log⁡χr)∣PB,z=ψ(Fr⁡w)log⁡rq.x_0=(\log\chi_r)|_{P_B},\qquad z=\frac{\psi(\operatorname{Fr}_w)}{\log_r q}.

The denominator is nonzero: the kernel of the logarithm on Zr×\mathbb{Z}_r^\times consists of roots of unity, and the positive rational integer qq is not a root of unity. Thus z≠0z\ne0, and ψ−zx0\psi-zx_0 vanishes on the decomposition group at ww, since it is unramified there and vanishes on Frobenius.

Consider its pullback

y=βB∗(ψ−zx0)∈Xr(K).(11)y=\beta_B^*(\psi-zx_0)\in X_r(K). \tag*{(11)}

The pullback of ψ\psi transforms by sign under Gk/GKG_k/G_K. Its corestriction to kk is therefore zero: restricting that corestriction back to KK gives the sum of the two translates, and restriction is injective for additive characters with characteristic-zero coefficients. Theorem 4.4 identifies βB∗x0\beta_B^*x_0 with log⁡χK,r\log\chi_{K,r}. Consequently

cor⁡K/Q(y)=−z[K:Q]log⁡χQ,r≠0.\operatorname{cor}_{K/\mathbb{Q}}(y)=-z[K:\mathbb{Q}]\log\chi_{\mathbb{Q},r}\ne0.

For every v∈Vwv\in\mathcal{V}_w, away from rr, decomposition containment gives y(Fr⁡v)=0y(\operatorname{Fr}_v)=0, regardless of which auxiliary prime makes the pair active. Proposition 3.2(iii) permits only finitely many such vv. There are also only finitely many primes of KK above rr, proving the assertion.

For ℓ∈L\ell\in\mathcal{L}, let Mℓ\mathcal{M}_{\ell} be the set of primes of BB which occur in no active pair at ℓ\ell. This includes the primes above ℓ\ell.

We seek a bound for #Mℓ\#\mathcal{M}_{\ell} depending only on [K:Q][K:\mathbb{Q}]. The proof will place independent classes attached to missed primes in one character space of an SS-unit group. We first record the standard cohomological identification of that group, including the completion step; see also [11].

Lemma 5.2 (Unramified Kummer classes). Let FF be a number field, let ℓ\ell be a rational prime, and let SS be a finite set of places of FF containing all places above ℓ\ell and all archimedean places. Then Kummer theory induces a natural isomorphism

Hcont⁡1(GF,Qℓ(1))unramified outside ⁡S≃OF,S×⊗ZQℓ.(12)H^1_{\operatorname{cont}}(G_F,\mathbb{Q}_{\ell}(1))_{\operatorname{unramified\ outside\ }S}\simeq\mathcal{O}_{F,S}^{\times}\otimes_{\mathbb{Z}}\mathbb{Q}_{\ell}. \tag*{(12)}

It is equivariant for every group of field automorphisms of FF preserving SS, with the natural action on cohomology combining conjugation on GFG_F and the action on the Tate twist.

Proof. Integral Kummer theory gives

Hcont⁡1(GF,Zℓ(1))≃lim←⁡nF×/(F×)ℓn.H^1_{\operatorname{cont}}(G_F,\mathbb{Z}_{\ell}(1))\simeq\varprojlim_n F^{\times}/(F^{\times})^{\ell^n}.

The inverse-limit assertion follows from the finite-level Kummer isomorphisms; the degree-zero groups are finite and satisfy the Mittag–Leffler condition. Every continuous rational cocycle has bounded image, since GFG_F is compact, and hence becomes integral after multiplication by one power of ℓ\ell. The same observation for coboundaries gives rationalization of this integral cohomology. At a prime outside SS, inertia acts trivially on Zℓ(1)\mathbb{Z}_{\ell}(1), and its Kummer evaluation is the valuation times the tame ℓ\ell-adic character. These integral inertia homomorphisms are torsion-free. Thus unramifiedness after rationalization is exactly the vanishing of all integral valuation coordinates outside SS.

For completeness, put US=OF,S×U_S=\mathcal{O}_{F,S}^{\times} and let

Hn={[a]∈F×/(F×)ℓn:ord⁡v(a)≡0(modℓn) for every v∉S}.H_n=\{[a]\in F^{\times}/(F^{\times})^{\ell^n}:\operatorname{ord}_v(a)\equiv0\pmod{\ell^n}\text{ for every }v\notin S\}.

Writing Cl⁡S(F)\operatorname{Cl}_S(F) for the SS-ideal class group, there is an exact sequence

0⟶US/USℓn⟶Hn⟶Cl⁡S(F)[ℓn]⟶0.(13)0\longrightarrow U_S/U_S^{\ell^n}\longrightarrow H_n\longrightarrow\operatorname{Cl}_S(F)[\ell^n]\longrightarrow0. \tag*{(13)}

The last map sends [a][a] to the ideal class of div⁡S(a)/ℓn\operatorname{div}_S(a)/\ell^n, where div⁡S\operatorname{div}_S records the valuations outside SS. The transition from level n+1n+1 to level nn induces multiplication by ℓ\ell on these obstruction classes. Since Cl⁡S(F)\operatorname{Cl}_S(F) is finite, their inverse limit is zero. It follows that the kernel of all outside-SS valuations in the completed multiplicative group is precisely lim⁡⟵nUS/USℓn\lim_{\longleftarrow n}U_S/U_S^{\ell^n}. The argument takes place in the full completed multiplicative group; it does not assume finite valuation support for its elements. Finally USU_S is finitely generated, so rationalizing its completion gives US⊗QℓU_S\otimes\mathbb{Q}_{\ell}. This proves (5.1). All the constructions commute with field automorphisms preserving SS, giving the asserted equivariance. ∎

Uchida’s proof of target-side finiteness associates Kummer classes to missed primes and bounds them by a unit rank [15 Proposition 1]. We refine that argument by keeping the classes in one character space of a cyclic extension. The multiplicity of this character has a bound independent of ℓ\ell, even when the extension degree grows.

Proposition 5.3. For every ℓ∈L\ell\in\mathcal{L},

#{w∈Mℓ:w∤ℓ}≤2[K:Q],#Mℓ≤2[K:Q]+2.\#\{w \in\mathcal{M}_{\ell}: w \nmid\ell\} \le2[K:\mathbb{Q}], \qquad\#\mathcal{M}_{\ell} \le2[K:\mathbb{Q}] + 2.

Proof. Fix ℓ∈L\ell\in\mathcal{L} and any finite set J⊂MℓJ \subset\mathcal{M}_{\ell} of primes away from ℓ\ell. For each w∈Jw \in J, choose aw∈B×a_w \in B^\times and hw>0h_w > 0 with

(aw)=whw.(a_w)=w^{h_w}.

Its compatible Kummer cocycles define a class

κw∈Hcont1(PB,Qℓ(1)).\kappa_w \in H^1_{\mathrm{cont}}(P_B,\mathbb{Q}_{\ell}(1)).

They factor through PBP_B because Qs\mathbb{Q}^{s} contains the roots of unity and all the required radicals. Valuations at the distinct primes ww show that these classes are linearly independent. Each is unramified at every finite prime other than ww and the primes above ℓ\ell.

Write

ρ=χℓ∘βB,ϵ=ρχK,ℓ−1.\rho=\chi_{\ell}\circ\beta_B,\qquad\epsilon=\rho\chi_{K,\ell}^{-1}.

By Theorem 4.4, ϵ\epsilon takes values in μℓ−1⊂Qℓ×\mu_{\ell-1}\subset\mathbb{Q}_{\ell}^{\times}. Let K′/KK'/K be the cyclic extension fixed by its kernel, and let Δ=Gal⁡(K′/K)\Delta=\operatorname{Gal}(K'/K). Pullback along the surjection βB\beta_B is injective on H1H^1, with pulled-back coefficient module Qℓ(ρ)\mathbb{Q}_{\ell}(\rho). Restriction to GK′G_{K'} remains injective: its kernel is a first cohomology group of the finite group Δ\Delta, which vanishes in characteristic zero. We therefore obtain independent classes

cw∈Hcont1(GK′,Qℓ(1))(w∈J).c_w\in H^1_{\mathrm{cont}}(G_{K'},\mathbb{Q}_{\ell}(1))\qquad(w\in J).

Under the natural Δ\Delta-action on this cohomology, they all lie in the ϵ−1\epsilon^{-1} eigenspace. Indeed the original coefficient module is Qℓ(1)⊗ϵ\mathbb{Q}_{\ell}(1)\otimes\epsilon, so invariance for its twisted action becomes the eigenvalue ϵ−1\epsilon^{-1} after restriction.

We claim that every cwc_w is unramified away from ℓ\ell. Suppose that it is ramified at a prime u∤ℓu\nmid\ell of K′K', and let vv be the prime below it in KK. The cyclotomic module is trivial on IuI_u, so the restricted cocycle is a nonzero continuous homomorphism from IuI_u to the additive group Qℓ(1)\mathbb{Q}_{\ell}(1). Choose a pro-ℓ\ell Sylow subgroup Ju⊂IuJ_u\subset I_u on which it is nonzero, and a pro-ℓ\ell Sylow subgroup Jv⊂IvJ_v\subset I_v containing JuJ_u. Then βB(Jv)≠1\beta_B(J_v)\ne1, so vv is active at ℓ\ell. Theorem 2.4 sends JvJ_v into inertia at a corresponding prime t∤ℓt\nmid\ell of BB. Since tt occurs in an active pair, t∉Jt\notin J, whereas κw\kappa_w is unramified at tt. Inertia at tt acts trivially on the coefficient module, so the Kummer cocycle itself vanishes on that inertia group. Its pullback therefore vanishes on JuJ_u, a contradiction. This proves the claim.

Let SS consist of the archimedean places of K′K' and its primes above ℓ\ell. By Lemma 5.2, the classes cwc_w give independent elements of the ϵ−1\epsilon^{-1} eigenspace in OK′,S×⊗Qℓ\mathcal{O}^{\times}_{K',S}\otimes\mathbb{Q}_{\ell}. It remains to bound this eigenspace. The equivariant Dirichlet SS-unit theorem [1 Section 3] gives the character of the full SS-unit module as

[OK′,S×⊗Qℓ]=[Qℓ[S]]−[Qℓ].[\mathcal{O}^{\times}_{K',S}\otimes\mathbb{Q}_{\ell}]=[\mathbb{Q}_{\ell}[S]]-[\mathbb{Q}_{\ell}].

Each Δ\Delta-orbit in SS contributes at most one copy of the one-dimensional character ϵ−1\epsilon^{-1} to the permutation representation: for a stabilizer H⊂ΔH\subset\Delta, its multiplicity in Qℓ[Δ/H]\mathbb{Q}_{\ell}[\Delta/H] is one if it is trivial on HH, and zero otherwise. There are

#(S/Δ)=r1(K)+r2(K)+#{v of K:v∣ℓ}≤2[K:Q]\#(S/\Delta)=r_1(K)+r_2(K)+\#\{v\text{ of }K:v\mid\ell\}\le2[K:\mathbb{Q}]

such orbits. Subtracting the trivial representation cannot increase this multiplicity. Hence the independent classes cwc_w lie in a space of dimension at most 2[K:Q]2[K:\mathbb{Q}], and #J≤2[K:Q]\#J \le2[K:\mathbb{Q}]. Since JJ was arbitrary, this bounds all missed primes away from ℓ\ell. There are at most two primes of BB above ℓ\ell, giving the second bound.

The bound depends on KK, but neither on ℓ\ell nor on the degree of the cyclic extension K′/KK'/K. We can therefore choose one auxiliary prime that excludes every possible partner of several specified target primes, and contradict this bound. The following elementary Kummer construction provides that choice.

Simultaneous separation by power residues

Lemma 5.4. Let mm, d≥1d \ge1, let q1,…,qnq_1,\ldots,q_n be distinct rational primes, and for each jj let Pj\mathcal{P}_j be a finite set of rational primes not containing qjq_j. Given any finite set of rational primes to avoid, there are infinitely many odd primes ℓ≡1(modm)\ell\equiv1 \pmod m outside that set and all the displayed prime sets such that

pf mod ℓ∉⟨qj mod ℓ⟩(1≤j≤n, p∈Pj, 1≤f≤d).p^f \bmod\ell\notin\langle q_j \bmod\ell\rangle\qquad(1 \le j \le n,\ p \in\mathcal{P}_j,\ 1 \le f \le d).

The sets Pj\mathcal{P}_j may overlap, and may contain qiq_i for i≠ji \ne j.

Proof. Choose distinct auxiliary primes sjs_j larger than mm, dd, and every prime in the lists, and put

C=Q(ζm∏jsj).C = \mathbb{Q}\left(\zeta_{m\prod_j s_j}\right).

For each jj, the classes of the distinct rational primes in {qj}∪Pj\{q_j\}\cup\mathcal{P}_j are independent in C×/(C×)sjC^\times/(C^\times)^{s_j}. Indeed, at any prime in those lists, the ramification index in C/QC/\mathbb{Q} divides [Q(ζm):Q][\mathbb{Q}(\zeta_m):\mathbb{Q}]: the other cyclotomic conductors are prime to that prime. This index is prime to sjs_j, and valuations at the separate rational primes give the asserted independence.

Kummer theory therefore identifies the Galois group over CC of

Ej=C(qj1/sj, p1/sj (p∈Pj))E_j = C\left(q_j^{1/s_j},\ p^{1/s_j}\ (p \in\mathcal{P}_j)\right)

with an elementary abelian sjs_j-group having one independent coordinate for each listed prime. Choose σj∈Gal⁡(Ej/C)\sigma_j \in\operatorname{Gal}(E_j/C) which fixes qj1/sjq_j^{1/s_j} and acts nontrivially on every p1/sjp^{1/s_j} with p∈Pjp \in\mathcal{P}_j. The extensions Ej/CE_j/C are linearly disjoint, since their degrees are powers of distinct primes. Thus these automorphisms combine to an element σ∈Gal⁡(E/C)\sigma\in\operatorname{Gal}(E/C), where EE is their compositum.

The extension E/QE/\mathbb{Q} is Galois. Moreover, conjugation in Gal⁡(E/Q)\operatorname{Gal}(E/\mathbb{Q}) preserves whether each of the specified radical coordinates of σ\sigma is trivial: a conjugation only raises its root-of-unity multiplier to a power prime to the corresponding sjs_j. Chebotarev’s theorem supplies infinitely many rational primes ℓ\ell with Frobenius in the conjugacy class of σ\sigma, avoiding the prescribed finite set, all listed primes, and 22. They split completely in CC, so ℓ≡1(modm)\ell\equiv1 \pmod m and ℓ≡1(modsj)\ell\equiv1 \pmod{s_j}. At these primes, qjq_j is an sjs_jth power modulo ℓ\ell and every p∈Pjp \in\mathcal{P}_j is not.

The quotient of Fℓ×\mathbb{F}_\ell^\times by its sjs_jth powers has prime order sjs_j. A nontrivial class stays nontrivial on raising it to any exponent 1≤f≤d<sj1 \le f \le d < s_j. Every power of qjq_j, however, is an sjs_jth power. This proves the displayed separation. In particular, overlaps between different lists impose no conflict: their conditions belong to linearly disjoint Kummer extensions with different prime exponents.

Proposition 5.5. For all but finitely many primes ww of BB above rational primes qq split in BB, there exist a prime vv of KK above the same qq and an auxiliary prime ℓ∈L\ell\in\mathcal{L} for which (v,w)(v,w) is an active pair for βB\beta_B.

Proof. Suppose that there are infinitely many exceptions. Lemma 5.1 allows us to choose

w1,…,wn,n>2[K:Q]+2,w_1,\ldots,w_n,\qquad n > 2[K:\mathbb{Q}]+2,

among them, above distinct split rational primes q1,…,qnq_1,\ldots,q_n, with every Vwj\mathcal{V}_{w_j} finite. Let PjP_j be the set of rational residue characteristics of the primes in Vwj\mathcal{V}_{w_j}. The failure of the asserted conclusion means qj∉Pjq_j \notin P_j.

Apply Lemma 5.4 with m=m0m=m_0 and d=[K:Q]d=[K:\mathbb{Q}]. It gives ℓ∈L\ell\in\mathcal{L}, away from all the primes involved, such that

Nv mod ℓ∉⟨qj mod ℓ⟩(v∈Vwj, 1≤j≤n),(14)Nv \bmod\ell\notin\langle q_j \bmod\ell\rangle\qquad(v\in\mathcal{V}_{w_j},\ 1\leq j\leq n), \tag*{(14)}

because Nv=pfNv=p^f with p∈Pjp\in P_j and 1≤f≤[K:Q]1\leq f\leq[K:\mathbb{Q}]. If wjw_j occurred in an active pair at this ℓ\ell, its source prime would lie in Vwj\mathcal{V}_{w_j}. Applying Lemma 2.6 to a lift of source Frobenius and reducing modulo ℓ\ell would give

Nv mod ℓ∈⟨Nwj mod ℓ⟩=⟨qj mod ℓ⟩,Nv \bmod\ell\in\langle Nw_j \bmod\ell\rangle=\langle q_j \bmod\ell\rangle,

contrary to (5.3). All nn primes wjw_j therefore belong to Mℓ\mathcal{M}_\ell, contradicting Proposition 5.3. □\square

Identifying the homomorphism

We now turn the prime correspondence into equality of group homomorphisms. The argument uses one fixed coefficient prime rr, although the auxiliary prime witnessing activity is allowed to vary. This distinction is what makes the correspondences constructed in Section 5 sufficient.

Retain the map β:T=Gk→P\beta:T=G_k\to P, the infinite set L\mathcal{L}, and the fields B=Q(i)B=\mathbb{Q}(i) and KK from Sections 4–5. Thus GK=β−1(PB)G_K=\beta^{-1}(P_B), and Theorem 4.4 gives

log⁡χr∘β=log⁡χk,r(r∈L).(15)\log\chi_r\circ\beta=\log\chi_{k,r}\qquad(r\in\mathcal{L}). \tag*{(15)}

Write σ:T→P\sigma:T\to P for ordinary restriction.

Proposition 6.1. There is an element a∈Pa\in P such that β=Ad⁡(a)∘σ\beta=\operatorname{Ad}(a)\circ\sigma on TT.

Proof. Fix a finite Galois extension L/QL/\mathbb{Q} inside Qs\mathbb{Q}^{\mathrm{s}} containing BB, and put R=Gal⁡(L/Q)R=\operatorname{Gal}(L/\mathbb{Q}). Choose a finite Galois extension F′/QF'/\mathbb{Q} containing KK and LL such that β(GF′)⊂PL\beta(G_{F'})\subset P_L. This is possible by taking the core in GG of the appropriate open subgroup of TT. Fix r∈Lr\in\mathcal{L}.

The maps β\beta and σ\sigma, restricted to GF′G_{F'}, give two pullbacks

iβ,iσ:Xr(L)⟶Xr(F′).(16)i_\beta,i_\sigma:X_r(L)\longrightarrow X_r(F'). \tag*{(16)}

Both are injective: their underlying homomorphisms have open image, and an additive Qr\mathbb{Q}_r-valued character vanishing on a finite-index subgroup is zero. For a∈Pa\in P let σa=Ad⁡(a)∘σ\sigma_a=\operatorname{Ad}(a)\circ\sigma and let iai_a be its pullback. Since L/QL/\mathbb{Q} is Galois, σa(GF′)⊂PL\sigma_a(G_{F'})\subset P_L. The map iai_a depends only on aa modulo PLP_L, because inner automorphisms of PLP_L act trivially on additive characters.

Frobenius evaluations. Let q≠rq\ne r range over rational primes split completely in F′F'. By Proposition 5.5, for all but finitely many such qq there is a prime vv of KK above qq whose decomposition group maps into a decomposition group at a prime ww of BB above the same qq. Choose a prime v′v' of F′F' above vv and compatible prolongations. Then

β(Dv′)⊂Du⊂PL\beta(D_{v'})\subset D_u\subset P_L

for a prime uu of LL above ww. Every prime here has absolute residue degree one, because qq splits completely in F′F'.

Choose a lift hh of arithmetic Frobenius in Dv′D_{v'}. Let aq∈Z^a_q \in\widehat{\mathbb{Z}} be the residue exponent of β(h)\beta(h) at uu. Equation (15) gives

(aq)rlog⁡rq=log⁡rq.(a_q)_r \log_r q = \log_r q.

The number log⁡rq\log_r q is nonzero: q∈Zr×q \in\mathbb{Z}_r^\times is a positive rational integer larger than one and is not a root of unity. Consequently (aq)r=1(a_q)_r = 1. Every x∈Xr(L)x \in X_r(L) is unramified at uu, so

iβ(x)(Fr⁡v′)=x(Fr⁡u).(17)i_\beta(x)(\operatorname{Fr}_{v'}) = x(\operatorname{Fr}_u). \tag*{(17)}

This calculation uses only decomposition containment from the active correspondence. It does not require that its auxiliary prime be rr.

Under ordinary restriction, the same Frobenius maps to Frobenius at another prime of LL above qq. An element of RR carries this prime to uu. Choose a lift a∈Pa \in P of that element. Then

iβ(x)(Fr⁡v′)=ia(x)(Fr⁡v′)i_\beta(x)(\operatorname{Fr}_{v'}) = i_a(x)(\operatorname{Fr}_{v'})

for every x∈Xr(L)x \in X_r(L).

The choice of aa modulo PLP_L may initially depend on qq. Since RR is finite, an infinite set of distinct qq uses one fixed choice. For each xx, the difference iβ(x)−ia(x)i_\beta(x)-i_a(x) therefore vanishes at infinitely many degree-one primes of F′F'. Proposition 3.2 gives

iβ=ia.i_\beta= i_a.

From characters to the finite group. The field F′/QF'/\mathbb{Q} is Galois, so TT acts on Xr(F′)X_r(F') by conjugation. The two pullbacks intertwine this action with the respective actions on Xr(L)X_r(L) induced by β\beta and σa\sigma_a. If ii denotes their common injective value in (6.4), then for t∈Tt \in T and x∈Xr(L)x \in X_r(L),

i(β(t)⋅x)=t⋅i(x)=i(σa(t)⋅x).i\bigl(\beta(t)\mathbin{\cdot}x\bigr)=t\mathbin{\cdot}i(x)=i\bigl(\sigma_a(t)\mathbin{\cdot}x\bigr).

The field LL is totally imaginary because it contains BB. By Proposition 3.3, RR acts faithfully on Xr(L)X_r(L). Hence

β(t) mod PL=σa(t) mod PL(t∈T).\beta(t)\bmod P_L=\sigma_a(t)\bmod P_L \qquad(t\in T).

One conjugation for all finite quotients. For each such LL, let

CL={a∈P: β(t)=Ad⁡(a)(σ(t))(modPL) for every t∈T}.C_L=\{a\in P:\ \beta(t)=\operatorname{Ad}(a)(\sigma(t))\pmod{P_L}\text{ for every }t\in T\}.

We have proved that CLC_L is nonempty. It is closed, and indeed clopen, since the defining condition depends only on aa modulo PLP_L. If LL contains finitely many fields L1,…,LsL_1,\ldots,L_s, then CL⊂⋂jCLjC_L\subset\bigcap_j C_{L_j}. Composita thus give the finite-intersection property. Compactness of PP supplies an element in every CLC_L. The extensions LL containing BB are cofinal among finite Galois subextensions of Qs/Q\mathbb{Q}^{\mathrm{s}}/\mathbb{Q}, so this element gives the asserted equality in PP. □

We remove the normalizer extension and return to the original source.

Theorem 6.2. Let FF be a number field in Q‾\overline{\mathbb{Q}}. Every continuous open homomorphism φ:GF→P\varphi:G_F\to P is conjugate in PP to ordinary restriction.

Proof. As in Lemma 4.1, choose an open normal subgroup UU of G\mathcal{G} contained in GFG_F, and extend φ∣U\varphi|_U to the normalizer of its kernel. Proposition 6.1 shows that, after one conjugation of φ\varphi, it agrees on UU with ordinary restriction σ:GF→P\sigma:G_F\to P.

For g∈GFg\in G_F and u∈Uu\in U, normality of UU gives

φ(g)σ(u)φ(g)−1=φ(gug−1)=σ(gug−1)=σ(g)σ(u)σ(g)−1.\varphi(g)\sigma(u)\varphi(g)^{-1}=\varphi(gug^{-1})=\sigma(gug^{-1})=\sigma(g)\sigma(u)\sigma(g)^{-1}.

Thus σ(g)−1φ(g)\sigma(g)^{-1}\varphi(g) centralizes σ(U)=φ(U)\sigma(U)=\varphi(U), an open subgroup of PP. Lemma 2.3 makes this element trivial. The equality holds for every g∈GFg\in G_F. □

Proof of Theorem 1.1. Choose copies of the given fields inside Q‾\overline{\mathbb{Q}}. By Lemma 2.1, each EiE_i contains Qs\mathbb{Q}^{s}. Form the composite

GF1⟶G1→αG2⟶P(18)G_{F_1} \longrightarrow G_1 \xrightarrow{\alpha} G_2 \longrightarrow P \tag*{(18)}

where the outer arrows are restrictions. The first is surjective. The last has open image, since its composite with GF2↠G2G_{F_2} \twoheadrightarrow G_2 is ordinary restriction from an open subgroup of GG to PP. Continuous surjective homomorphisms between profinite groups are open, so (18) is open.

By Theorem 6.2, this composite is conjugate to ordinary restriction. Cyclotomic characters are unchanged by conjugation. All roots of unity lie in Qs\mathbb{Q}^{s}, so the full cyclotomic characters factor through the groups in (18). Surjectivity of the first arrow therefore yields

χE2/F2∘α=χE1/F1in Z^×.\chi_{E_2/F_2} \circ\alpha= \chi_{E_1/F_1} \quad\text{in } \widehat{\mathbb{Z}}^{\times}.

Theorem 2.5 gives an equivariant embedding E2↪E1E_2 \hookrightarrow E_1. Its uniqueness is Uchida’s theorem cited after Theorem 2.5.

References

  1. [1]Hyman Bass, Generators and relations for cyclotomic units, Nagoya Math. J. 27 (1966), no. 2, 401–407. doi:10.1017/S0027763000026258.DOI
  2. [2]Samit Dasgupta, Ranks of matrices of logarithms of algebraic numbers. I: The theorems of Baker and Waldschmidt–Masser, Essential Number Theory 2 (2023), no. 1, 93–138. doi:10.2140/ent.2023.2.93.DOI
  3. [3]Yuichiro Hoshi, Homomorphisms of global solvably closed Galois groups compatible with cyclotomic characters, Tohoku Math. J. (2) 77 (2025), no. 1, 17–32. doi:10.2748/tmj.20230414.DOI
  4. [4]Yuichiro Hoshi, A note on open homomorphisms between global solvably closed Galois groups, Kodai Math. J. 49 (2026), no. 1, 63–79. doi:10.2996/kmj49104.DOI
  5. [5]Michel Laurent, Sur quelques résultats récents de transcendance, Astérisque 198–199–200 (1991), 209–230. doi:10.24033/ast.93.DOI
  6. [6]Yu Mao and Mohamed Saïdi, Open homomorphisms between m-step solvable Galois groups compatible with the cyclotomic characters, preprint, 2025. arXiv:2512.13312.arxiv.org/abs/2512.13312
  7. [7]D. W. Masser, On polynomials and exponential polynomials in several complex variables, Invent. Math. 63 (1981), no. 1, 81–95. doi:10.1007/BF01389194.DOI
  8. [8]Jürgen Neukirch, Kennzeichnung der p-adischen und der endlichen algebraischen Zahlkörper, Invent. Math. 6 (1969), 296–314. doi:10.1007/BF01425420.DOI
  9. [9]Jürgen Neukirch, Kennzeichnung der endlich-algebraischen Zahlkörper durch die Galoisgruppe der maximal auflösbaren Erweiterungen, J. reine angew. Math. 238 (1969), 135–147. doi:10.1515/crll.1969.238.135.DOI
  10. [10]Jürgen Neukirch, Über die absoluten Galoisgruppen algebraischer Zahlkörper, Astérisque 41–42 (1977), 67–79. https://www.numdam.org/item/AST_1977__41-42__67_0/.numdam.org/item/AST_1977__41-42__67_0
  11. [11]Jürgen Neukirch, Algebraic Number Theory, translated by Norbert Schappacher, Grundlehren der mathematischen Wissenschaften, vol. 322, Springer-Verlag, Berlin, 1999. doi:10.1007/978-3-662-03983-0.DOI
  12. [12]Mohamed Saïdi and Akio Tamagawa, The m-step solvable anabelian geometry of number fields, J. reine angew. Math. 789 (2022), 153–186. doi:10.1515/crelle-2022-0025.DOI
  13. [13]Kōji Uchida, Isomorphisms of Galois groups, J. Math. Soc. Japan 28 (1976), no. 4, 617–620. doi:10.2969/jmsj/02840617.DOI
  14. [14]Kōji Uchida, Isomorphisms of Galois groups of solvably closed Galois extensions, Tohoku Math. J. (2) 31 (1979), no. 3, 359–362. doi:10.2748/tmj/1178229803.DOI
  15. [15]Kōji Uchida, Homomorphisms of Galois groups of solvably closed Galois extensions, J. Math. Soc. Japan 33 (1981), no. 4, 595–604. doi:10.2969/jmsj/03340595.DOI
  16. [16]Michel Waldschmidt, Transcendance et exponentielles en plusieurs variables, Invent. Math. 63 (1981), no. 1, 97–127. doi:10.1007/BF01389195.DOI

Paper details

vlnc:W_8n4en8sm
Version
v1 · latest
Published
License
LicenseRef-OpenAI-public
Usage

views · 36

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

unreleased internal OpenAI model

View statement
Looking for related papers…
Contents