betaFeedback

WorksMathematics

Open Homomorphisms between mm-step Solvable Galois Groups of Number Fields

Abstract

Let K,LK,L be number fields, and let m≥2m \geq 2 be an integer. In this paper, we prove that an open homomorphism ϕm+2:GKm+2→GLm+2ϕ_{m+2}: G_K^{m+2} \to G_L^{m+2} induces an open homomorphism ϕmϕ_m that arises uniquely from a field embedding τm:Lm↪Kmτ_m:L_m \hookrightarrow K_m. We first establish a sharper version of the mm-step solvable cyclotomic compatibility criterion by Mao and Saïdi: if ϕm+2:GKm+2→GLm+2ϕ_{m+2}:G_K^{m+2}\to G_L^{m+2} is compatible with the cyclotomic characters, then the induced map ϕm:GKm→GLmϕ_m:G_K^m \to G_L^m arises uniquely from a field embedding Lm↪KmL_m \hookrightarrow K_m. Then we apply the strategy of OpenAI's (together with Tsujimura's recent explanation of OpenAI's work) proposed proof of Uchida's conjecture to show that every open homomorphism ϕ4:GK4→GL4ϕ_4:G_K^4 \to G_L^4 is automatically compatible with the cyclotomic characters.

Conventions

Throughout this paper, we shall use the following conventions.

  • For a field kk, we shall denote by kˉ\bar{k} for its separable closure contained in some algebraic closure kalgk^{\mathrm{alg}} of $k.

  • Let GG be a profinite group, we shall denote by G[0]:=GG^{[0]} := G and G[i]:=[G[i−1],G[i−1]]G^{[i]} := [G^{[i-1]},G^{[i-1]}] for i≥1i \ge1. In this case, we shall denote by Gi:=G/G[i]G^{i} := G/G^{[i]} for i≥0i \ge0, we shall call this quotient the maximal ii-step solvable quotient of GG.

  • For an abelian group GG, and an integer n≥2n \ge2, we shall write

G[n]:=ker⁡(G→×nG)G[n] := \ker\left(G \xrightarrow{\times n} G\right)

if GG is viewed additively.

  • For a field kk, we shall denote by GkmG_k^m for the maximal mm-step solvable quotient of GkG_k for some m≥0m \ge0. In this case, we shall denote by km:=kˉGk[m]k_m := \bar{k}^{G_k^{[m]}} for the maximal mm-step solvable extension of kk. In particular, km=km−1abk_m = k_{m-1}^{\mathrm{ab}} for all m≥1m \ge1 (where k0:=kk_0 := k).

  • We shall denote by Primes\mathfrak{P}\mathrm{rimes} for the set of all rational primes.

  • Let k/Qk/\mathbb{Q} be an algebraic extension, we shall denote by Pk\mathcal{P}_k for the set of all primes of kk. Moreover, we shall denote by Pk<∞\mathcal{P}_k^{<\infty} for the set of all non-archimedean primes of kk and Pk∞\mathcal{P}_k^\infty for the set of all archiemdean primes of kk.

  • Let kk be a number field and let p∈Pk\mathfrak{p} \in\mathcal{P}_k be a prime. We shall denote by kpk_{\mathfrak{p}} for the completion of kk at p\mathfrak{p}.

  • Let kk be a number field, and let L/kL/k be a Galois extension, let v∈Pk<∞v \in\mathcal{P}_k^{<\infty} and w∈PL<∞w \in\mathcal{P}_L^{<\infty} lying above vv. We shall denote by Dw(L/k)D_w(L/k) (resp. Iw(L/k)I_w(L/k)) for the decomposition subgroup of Gal⁡(L/k)\operatorname{Gal}(L/k) at ww (resp. the

inertia subgroup of Gal⁡(L/k)\operatorname{Gal}(L/k) at ww). If there is no ambiguity, we shall simply write DwD_w (resp. IwI_w) for this decomposition group (or even DvD_v if we only need the conjugacy class).

  • Let kk be a number field or a pp-adic local field, we shall denote by χk\chi_k for the cyclotomic character χk:Gk→Z^×\chi_k : G_k \to\widehat{\mathbb{Z}}^{\times}. For some prime number ℓ\ell, we shall denote by χk,ℓ\chi_{k,\ell} for the ℓ\ell-adic cyclotomic character χk,ℓ:Gk→Zℓ×\chi_{k,\ell} : G_k \to\mathbb{Z}_{\ell}^{\times}

  • All homomorphisms between profinite groups in this paper are assumed to be continuous.

  • For simplicity, we shall denote by PP for the absolute Galois group of Q\mathbb{Q}.

Introduction

In [8], Uchida proposed an important conjecture in predicting that any open homomorphisms between solvably closed Galois groups of number fields should arises uniquely from the corresponding field embeddings. This conjecture was known as the hom-form of Grothendieck’s anabelian conjectures for number fields (where the isom-form is exactly the famous Neukirch-Uchida Theorem).

In 7th Oct 2026, OpenAI released a large amount of papers claiming solutions/partial solutions to many important conjectures in pure mathematics and theoretical computer science. In particular, one of them was the claiming a proof of Uchida’s conjecture mentioned above (c.f. [5]):

Theorem 1.1 (c.f. Theorem 1.1 in [5]). Let F1,F2F_1,F_2 be number fields, and let Ei/FiE_i/F_i be solvably closed Galois extensions for i∈{1,2}i \in\{1,2\}. Then every open homomorphism

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

arises uniquely from a field embedding β:E2↪E1\beta: E_2 \hookrightarrow E_1 such that g∘β=β∘α(g)g \circ\beta= \beta\circ\alpha(g) for all g∈Gal⁡(E1/F1)g \in\operatorname{Gal}(E_1/F_1).

Recently, Tsujimura also released a paper [7] contains a detailed explanation of the proof in [5], therefore we do not give a survey to the proof by OpenAI here.

On the other hand, one important variation of the Neukirch-Uchida theorem is the following finite-step solvable version:

Theorem 1.2 (c.f. Theorem 2 in [ST22]). Let m≥0m \ge0 be an integer and let K,LK,L be number fields. Let τm+3:GKm+3→∼GLm+3\tau_{m+3} : G_K^{m+3} \xrightarrow{\sim} G_L^{m+3} be an isomorphism.

(i) There exists an isomorphism σm:Km→∼Lm\sigma_m : K_m \xrightarrow{\sim} L_m such that τm(g)=σmgσm−1\tau_m(g) = \sigma_m g \sigma_m^{-1} for all g∈GKmg \in G_K^m where τm:GKm→∼GLm\tau_m : G_K^m \xrightarrow{\sim} G_L^m is the isomorphism induced by τm+3\tau_{m+3}. In particular, σm\sigma_m restricts to an isomorphism σ:K→∼L\sigma: K \xrightarrow{\sim} L.

(ii) Assume that m≥2m \ge2 (resp. m=1m = 1). Then the isomorphism σm:Km→∼Lm\sigma_m : K_m \xrightarrow{\sim} L_m (resp. σ:K→∼L\sigma: K \xrightarrow{\sim} L induced by σ1:K1→∼L1\sigma_1 : K_1 \xrightarrow{\sim} L_1) in (i) is uniquely determined by the property that τm(g)=σmgσm−1\tau_m(g) = \sigma_m g \sigma_m^{-1} for every g∈GKmg \in G_K^m.

Thus, it is natural to ask for an mm-step solvable version of Theorem 1.1. Previously, an mm-step solvable version of Theorem 1.1 with local conditions was obtained by Corato and Saïdi in [CS23]. The main result of this paper is the unconditional version of Theorem 1.1, and with one less step of solvability depth than the main result in [CS23]:

Theorem 1. Let m≥2m \ge2 be an integer and let K,LK, L be number fields. Let ϕm+2:GKm+2→GLm+2\phi_{m+2}: G_K^{m+2} \to G_L^{m+2} be an open homomorphism. Let ϕm:GKm→GLm\phi_m: G_K^m \to G_L^m be the open homomorphism induced by ϕm+2\phi_{m+2} (c.f. Proposition 2.2 in [CS23]). Then there exists a unique field embedding τm:Lm↪Km\tau_m: L_m \hookrightarrow K_m satisfying g∘τm=τm∘ϕm(g)g \circ\tau_m = \tau_m \circ\phi_m(g) for all g∈GKmg \in G_K^m.

Similar to the proof of Theorem 1.1, the proof of Theorem 1 uses the mm-step solvable version cyclotomic compatibility (c.f. the main Theorem in [2] and a previously proved mm-step solvable version Theorem 1.5 in [MS25]). In particular, we prove the following sharper mm-step solvable cyclotomic compatibility criterion than Theorem 1.5 in [MS25] to support the proof of Theorem 1:

Theorem 2. Let m≥2m \ge2 be an integer and let K,LK, L be number fields. Let ϕm+2:GKm+2→GLm+2\phi_{m+2}: G_K^{m+2} \to G_L^{m+2} be an open homomorphism. Then the followings are equivalent:

(i) χL∘ϕm+2=χK\chi_L \circ\phi_{m+2} = \chi_K;

(ii) There exists a field embedding τm:Lm↪Km\tau_m: L_m \hookrightarrow K_m satisfying

g∘τm=τm∘ϕm(g)g \circ\tau_m = \tau_m \circ\phi_m(g)

for all g∈GKmg \in G_K^m and ϕm\phi_m the open homomorphism induced by ϕm+2\phi_{m+2}. Furthermore, such τm\tau_m is unique and τm(L)⊂K\tau_m(L) \subset K.

Remark 1.3. An interesting corollary of Theorem 2 is the mm-step solvable isom-form. By assuming m≥2m \ge2, applying Theorem 1.26 together with Theorem 2 would prove that the m+3m+3 in Theorem 1.2 can be replaced by m+2m+2 (but we have to assume that m≥2m \ge2), which turns Theorem 1.2 from a 3-steps loss isom-form to a 2-steps loss isom-form.

Like strategy in [5], we prove that the cyclotomic compatibility is automatic and deduce Theorem 2, in particular, we prove

Theorem 3. Every open homomorphism β:GK→P4\beta: G_K \to P^4 satisfies χQ∘β=χK\chi_Q \circ\beta= \chi_K. Consequently every open homomorphism GK4→GL4G_K^4 \to G_L^4 is compatible with the cyclotomic characters.

Then Theorem 1 can be viewed as an immediate consequence of Theorem 3. The structure of this paper is organised as follows. In chapter 2, we analyse the homomorphic images of pro-ℓ\ell Sylow subgroups of decomposition groups. In chapter 3, we prove the mm-step solvable cyclotomic compatibility criterion namely Theorem 2. In chapter 4, we adapt the strategy in [5] and develop the logarithmic cyclotomic compatibility in the 4-step solvable case. In chapter 5, we deduce the full cyclotomic compatibility from the logarithmic cyclotomic compatibility proved in chapter 4, which gives a proof of Theorem 3, together with Theorem 2 gives Theorem 1.

Local Analysis of Decomposition Groups

We shall briefly review the parts of the local theory from [ST22] and [CS23] we need.

Let KK be a number field, and let ℓ∈Primes\ell\in\mathfrak{P}\mathrm{rimes} be a prime number. Let j≥2j \ge2 be an integer and let p∈PK<∞\mathfrak{p} \in\mathcal{P}^{<\infty}_{K} be a prime not lying above ℓ\ell and choose prolongations pj∈PKj<∞,p‾∈PK‾<∞\mathfrak{p}_{j} \in\mathcal{P}^{<\infty}_{K_{j}}, \overline{\mathfrak{p}} \in\mathcal{P}^{<\infty}_{\overline{K}} of p\mathfrak{p}. Then Proposition 1.1 in [ST22] implies that the natural surjection

Dp‾(K‾/K)→Dpj(Kj/K)D_{\overline{\mathfrak{p}}}(\overline{K}/K) \to D_{\mathfrak{p}_{j}}(K_{j}/K)

preserves the maximal tame quotient, so does the maximal unramified quotient. Furthermore, a pro-ℓ\ell Sylow S⊂DpjS \subset D_{\mathfrak{p}_{j}} is mapped isomorphically onto a pro-ℓ\ell Sylow subgroup of the maximal tame quotient of DpjD_{\mathfrak{p}_{j}}. In particular, we have an exact sequence

1→J→S→Sunr→11 \to J \to S \to S^{\mathrm{unr}} \to1

where both JJ and SunrS^{\mathrm{unr}} are free pro-ℓ\ell groups of rank 1. Furthermore, the centraliser of JJ in SS is JJ itself. In particular, if a homomorphism is injective on JJ, it is injective on SS.

Lemma 2.1. Let m≥4m \ge4 be an integer, and let α:GK→GLm\alpha: G_{K} \to G_{L}^{m} be an open homomorphism. If m≥jm \ge j, we shall write αj\alpha_{j} for the composite GK→αGLm→GLjG_{K} \xrightarrow{\alpha} G_{L}^{m} \to G_{L}^{j}.

Let ℓ\ell be a prime number and let p∤ℓ\mathfrak{p} \nmid\ell be a prime of KK and let P\mathfrak{P} be a prime of K‾\overline{K} above p\mathfrak{p}. Let S⊂DPS \subset D_{\mathfrak{P}} be a pro-ℓ\ell Sylow subgroup. Suppose that αn−1∣S\alpha_{n-1}|_{S} is injective. There are primes qm−1\mathfrak{q}_{m-1} of Lm−1L_{m-1} and qm−2\mathfrak{q}_{m-2} of Lm−2L_{m-2} not over ℓ\ell, such that

αm−1(S)⊂Dqm−1(Lm−1/L)andαm−2(DP)⊂Dqm−2(Lm−2/L).\alpha_{m-1}(S) \subset D_{\mathfrak{q}_{m-1}}(L_{m-1}/L) \quad\text{and} \quad\alpha_{m-2}(D_{\mathfrak{P}}) \subset D_{\mathfrak{q}_{m-2}}(L_{m-2}/L).

The prime qm−2\mathfrak{q}_{m-2} given by this construction is independent of SS for the chosen P\mathfrak{P}. The image of S∩IPS \cap I_{\mathfrak{P}} via αm−2\alpha_{m-2} is non-trivial and lies in Iqm−2(Lm−2/L)I_{\mathfrak{q}_{m-2}}(L_{m-2}/L).

Proof. It follows from Proposition 1.22 in [ST22] that the image of SS in GKm−1G_{K}^{m-1} is a (⋆ℓ)(\star_{\ell})-subgroup. Then it follows from Proposition 2.5 in [CS23] that the image of SS via GK→GKm−1→GKm−1G_{K} \to G_{K}^{m-1} \to G_{K}^{m-1} is also a (⋆ℓ)(\star_{\ell})-subgroup. Again by Proposition 1.22 in [ST22], we have αm−1(S)⊂Dqm−1(Lm−1/L)\alpha_{m-1}(S) \subset D_{\mathfrak{q}_{m-1}}(L_{m-1}/L). It follows from Proposition 2.9 in [CS23] that αm−2(DP)⊂Dqm−2(Lm−2/L)\alpha_{m-2}(D_{\mathfrak{P}}) \subset D_{\mathfrak{q}_{m-2}}(L_{m-2}/L), as well as the independence of the choice of SS. The final assertion follows from the fact that the inertia subgroup of a pro-ℓ\ell Sylow subgroup of the decomposition group can be characterised as the maximal normal abelian subgroup. □\square

We recall the definition of active primes introduced in [5].

Definition 2.2. Let p∈PK<∞\mathfrak{p} \in\mathcal{P}^{<\infty}_{K} be a prime (with suitable prolongation in K‾\overline{K}) not dividing ℓ\ell for an odd prime number ℓ\ell. We call that p\mathfrak{p} is active for β\beta if β2(J)\beta_{2}(J) is non-trivial. Here β2\beta_{2} is the composite GK→P4→P2G_{K} \to P^{4} \to P^{2}.

Next we shall obtain the residue exponent comparison in a similar way as in Lemma 2.6 in [5].

Lemma 2.3. Let p∈PK<∞\mathfrak{p} \in\mathcal{P}^{<\infty}_{K} be active for β:GK→P4\beta: G_{K} \to P^{4} with target prime q\mathfrak{q}. For g∈Dpg \in D_{\mathfrak{p}}, let b(g)∈Z^b(g) \in\widehat{\mathbb{Z}} be the residue exponent of gg at p\mathfrak{p} and let a(g)∈Z^a(g) \in\widehat{\mathbb{Z}} be the residue exponent of β2(g)\beta_{2}(g) at the chosen target prime of Q2\mathbb{Q}_{2}. Then

(Nv)b(g)=qa(g) in Zℓ×.(Nv)^{b(g)} = q^{a(g)} \text{ in } \mathbb{Z}_{\ell}^{\times}.

In particular, χQ,ℓ(β(g))=χK,ℓ(g)\chi_{\mathbb{Q},\ell}(\beta(g)) = \chi_{K,\ell}(g) on DvD_{v}.

Proof. Choose a topological generator τ∈J\tau\in J. Since p∤ℓ\mathfrak{p} \nmid\ell, we have J⟶∼Zℓ(1)J \overset{\sim}{\longrightarrow} \mathbb{Z}_{\ell}(1). It follows from Theorem 7.5.3 in [4] that gτg−1(τ(Nv)b(g))−1g\tau g^{-1}(\tau^{(Nv)^{b(g)}})^{-1} belongs to the wild inertia of DvD_{v}, a pro-pp group where p:=char⁡(p)p := \operatorname{char}(\mathfrak{p}). In the target tame quotient, the image of τ\tau is non-trivial in the pro-ℓ\ell tame inertia factor by Lemma 2.1. We shall denote by tt for this image of τ\tau. Then by considering the tame relation on the target side,

tqa(g)−(Nv)b(g)=1t^{q^{a(g)}-(Nv)^{b(g)}} = 1

which gives the desired equality. The cyclotomic assertion follows since both source and target primes have residue characteristic different from ℓ\ell.

The mm-step Solvable Version of Hoshi’s Cyclotomic Criterion

In this section, we give the proof of Theorem 2. We first recall some definitions given in [2].

  • Let KK be a number field, and let ℓ\ell be a prime number, we write

Aℓ(K):=GKab/ℓGKab.A_{\ell}(K) := G_{K}^{\mathrm{ab}}/\ell G_{K}^{\mathrm{ab}}.
  • Let p∈PK<∞\mathfrak{p} \in\mathcal{P}^{<\infty}_{K}, we write

FLℓ(p)⊂Aℓ(K)\mathrm{FL}_{\ell}(\mathfrak{p}) \subset A_{\ell}(K)

for the set of images of Frobenius liftings at p\mathfrak{p} in DpD_{\mathfrak{p}}, which is independent from the choice of prolongations of p\mathfrak{p} in the algebraic closure of KK.

  • Let LL be another number field, we say that an open homomorphism

φ:Aℓ(K)⟶Aℓ(L)\varphi: A_{\ell}(K) \longrightarrow A_{\ell}(L)

is Frobenius-preserving if for every degree 1 source prime over a cofinite set of rational primes, there is a target prime of the same residue characteristic and degree 1, with the image of FLℓ(p)\mathrm{FL}_{\ell}(\mathfrak{p}) contained in its set of liftings.

Now consider an open homomorphism ϕm+2:GKm+2→GLm+2\phi_{m+2}:G_K^{m+2}\to G_L^{m+2}, for each E/LE/L finite Galois and totally imaginary contained in LmL_m, let F/KF/K be the finite Galois extension of KK corresponding to the inverse image of Gal⁡(Lm/E)\operatorname{Gal}(L_m/E) by ϕm\phi_m and we shall denote by H⊂GKm+2H\subset G_K^{m+2} for this group. Furthermore, write H~\widetilde{H} for the inverse image of HH in GKG_K. In particular, we obtain a map

H~→∼GF→Gal⁡(Lm+1/E)→GEab\widetilde{H}\xrightarrow{\sim}G_F\to\operatorname{Gal}(L_{m+1}/E)\to G_E^{\mathrm{ab}}

which induces a continuous map

θE:A2(F)→A2(E).\theta_E:A_2(F)\to A_2(E).

Lemma 3.1. The map θE\theta_E is Frobenius-preserving.

Proof. Let u∈PF<∞u\in\mathcal{P}_F^{<\infty} be a degree 1 prime above an odd prime p∈Primesp\in\mathop{\mathrm{Primes}} unramified in F/QF/\mathbb{Q}. Let p∈PK<∞\mathfrak{p}\in\mathcal{P}_K^{<\infty} be u∣Ku|K and choose suitable prolongations of p\mathfrak{p}. Since Fu=Kp=QpF_u=K_{\mathfrak{p}}=\mathbb{Q}_p for pp the residue characteristic of uu, we have Du=DpD_u=D_{\mathfrak{p}}. Let SS be a pro-2 Sylow subgroup of DpD_{\mathfrak{p}}, and let J:=S∩IpJ:=S\cap I_{\mathfrak{p}}. In particular, JJ is the inertia subgroup of SS, isomorphic to the free pro-2 group of rank 1. Thus the image of JJ via the cyclotomic character is non-trivial. Hence the image of JJ via GF→Gal⁡(Lm+1/E)G_F\to\operatorname{Gal}(L_{m+1}/E) is non-trivial and has no involutions. Indeed, by Proposition 1.9 in [CS23], an involution of GLm+1G_L^{m+1} projects to the non-trivial element of an archimedean decomposition group in GLmG_L^m, hence acts non-trivially on i∈Ei\in E. Moreover, GF→Gal⁡(Lm+1/E)G_F\to\operatorname{Gal}(L_{m+1}/E) restricts to an injection to JJ, the arguments above Lemma 2.1 therefore implies that GF→Gal⁡(Lm+1/E)G_F\to\operatorname{Gal}(L_{m+1}/E) restricts to an injection to SS.

Now apply Lemma 2.1 to m+2m+2. At the upper prime qm+1\mathfrak{q}_{m+1} we have the containment of ϕm+1(S)\phi_{m+1}(S) and at its restriction to qm\mathfrak{q}_m the containment of ϕm(Dp)\phi_m(D_{\mathfrak{p}}). Set w:=qm∣Ew:=\mathfrak{q}_m|E. Since Dp⊂GFD_{\mathfrak{p}}\subset G_F, we have χK(Dp)⊂χE(Dw)\chi_K(D_{\mathfrak{p}})\subset\chi_E(D_w). By Lemma 3.1 (ii) in [2], the residue characteristic of ww is pp.

We shall write χE(p′)\chi_E^{(p')} for the prime-to-p′p' cyclotomic character of (−)(-). The image χE(p′)(Dw)\chi_E^{(p')}(D_w) is topologically generated by Nw=pfwNw=p^{f_w} where fwf_w is the inertia degree of ww. On the other hand, χE(p′)(Dw)\chi_E^{(p')}(D_w) contains pp. The map

Z^→∏q≠pZq×\widehat{\mathbb{Z}}\to\prod_{q\ne p}\mathbb{Z}_q^\times
x↦pxx\mapsto p^x

is injective. Thus p=(Nw)xp=(Nw)^x for some x∈Z^x\in\widehat{\mathbb{Z}}, which implies that 1=fwx1=f_wx in Z^\widehat{\mathbb{Z}} so fw=1f_w=1 (c.f. Lemma 3.1 (iii) in [2]).

Fix a Frobenius lifting g∈Dug\in D_u. The map from DuD_u to A2(F)×Z/2ZA_2(F)\times\mathbb{Z}/2\mathbb{Z} using the residue exponent mod 2 at the second component has pro-2 image. It follows from the profinite version of Sylow theorems that the restriction to SS of Du→A2(F)×Z/2ZD_u\to A_2(F)\times\mathbb{Z}/2\mathbb{Z} is also onto. Thus some s∈Ss\in S represents the same A2(F)A_2(F)-class as gg and has odd residue exponent. Furthermore, ϕm+1(s)\phi_{m+1}(s) is contained in the image of DwD_w. Choose a lifting tt of ϕm+1(s)\phi_{m+1}(s) in Dw⊂GED_w \subset G_E, we have χE(t)=χF(s)\chi_E(t)=\chi_F(s). The prime-to-pp cyclotomic character gives

paw(t)=pau(s) in ∏q≠pZq×p^{a_w(t)}=p^{a_u(s)}\ \text{in}\ \prod_{q\ne p}\mathbb{Z}_q^\times

where aw(t)a_w(t) and au(s)a_u(s) are the residue exponents of tt and ss respectively, and by the injectivity of Z^→∏q≠pZq×\widehat{\mathbb{Z}}\to\prod_{q\ne p}\mathbb{Z}_q^\times that aw(t)=au(s)a_w(t)=a_u(s). In particular, a=aw(t)a=a_w(t) is odd. Choose z∈Dwz\in D_w with residue exponent (a−1)/2∈Z^(a-1)/2\in\widehat{\mathbb{Z}}, then tz2tz^2 is a Frobenius lifting and has the same A2(E)A_2(E)-class as tt, which is precisely θE([g])\theta_E([g]). The inclusion of the Frobenius lifting sets follows for all uu above the indicated cofinite set, as required.

Now we can prove Theorem 2.

Theorem 3.2 (Theorem 2). Let m≥2m\ge2 be an integer and let K,LK,L be number fields. Let ϕm+2:GKm+2→GLm+2\phi_{m+2}:G_K^{m+2}\to G_L^{m+2} be an open homomorphism. Then the followings are equivalent:

(i) χL∘ϕm+2=χK\chi_L\circ\phi_{m+2}=\chi_K;

(ii) There exists a field embedding τm:Lm↪Km\tau_m:L_m\hookrightarrow K_m satisfying

g∘τm=τm∘ϕm(g)g\circ\tau_m=\tau_m\circ\phi_m(g)

for all g∈GKmg\in G_K^m and ϕm\phi_m the open homomorphism induced by ϕm+2\phi_{m+2}. Furthermore, such τm\tau_m is unique and τm(L)⊂K\tau_m(L)\subset K.

Proof. We first verify the implication (i) ⟹\Longrightarrow (ii). It follows from Corollary 2.8 in [2] that θE\theta_E gives a unique field embedding τE:E↪F\tau_E:E\hookrightarrow F. Let g∈GKg\in G_K, write a:=g∣Fa:=g|_F and b:=ϕm(g)∣Eb:=\phi_m(g)|_E and denote by ca,cbc_a,c_b for their conjugation actions. Since A2(−)A_2(-) is contravariant (c.f. Section 2 in [2]), the map attached to aθEb−1a\theta_Eb^{-1} is cbθEca−1=θEc_b\theta_Ec_a^{-1}=\theta_E, the faithfulness of A2(−)A_2(-) (c.f. Theorem 2.6 in [2]) gives aτE=τEba\tau_E=\tau_Eb.

Let E′E' be another totally imaginary Galois extension of LL contained in LmL_m that contains EE and write F′F' for the extension of KK determined by E′E' as in the construction of FF. In this case, together with Lemma 3.1, we have the following commutative diagram

A2(F′)→θE′A2(E′)↓↓A2(F)→θEA2(E)\begin{CD} A_2(F') @>{\theta_{E'}}>> A_2(E') \\ @VVV @VVV \\ A_2(F) @>{\theta_E}>> A_2(E) \end{CD}

where all arrows are Frobenius-preserving. In particular, the composite A2(F′)→θE′A2(E′)→A2(E)A_2(F')\xrightarrow{\theta_{E'}}A_2(E')\to A_2(E) is also Frobenius-preserving, which gives a field embedding E↪F′E\hookrightarrow F'. There is another field embedding obtained by composing τE\tau_E with F↪F′F\hookrightarrow F'. Thus the two embeddings induce the same map A2(F′)→A2(E)A_2(F')\to A_2(E), by the faithfulness of A2(−)A_2(-). Since the fields EE are cofinal in Lm/LL_m/L, the compatible τE\tau_E induces the required τm:Lm↪Km\tau_m:L_m\hookrightarrow K_m. The equivariance implies that τm(L)⊂K\tau_m(L)\subset K by taking invariants. For the uniqueness of τm\tau_m, Proposition 2.1 in [MS25] gives the desired uniqueness. We give an alternative proof here. Consider two embeddings τm,τm′:Lm↪Km\tau_m,\tau'_m : L_m \hookrightarrow K_m. Then τm,τm′\tau_m,\tau'_m would induce the same map A2(K)→A2(L)A_2(K) \to A_2(L). Then by the faithfulness of A2(−)A_2(-), τm∣L=τm′∣L\tau_m|_L=\tau'_m|_L. Extend τm,τm′\tau_m,\tau'_m to t1,t2∈GQt_1,t_2 \in G_{\mathbb{Q}} respectively. Then t2−1t1∈GLt_2^{-1}t_1 \in G_L, and its restriction zz to LmL_m centralises φm(GKm)\varphi_m(G^m_K). We claim that the centraliser in GmmG^m_m of an open subgroup HH is trivial. Indeed, for every finite prime Q\mathfrak{Q} of LmL_m, H∩DQH \cap D_{\mathfrak{Q}} is open in DQD_{\mathfrak{Q}} and DzQD_{z\mathfrak{Q}}. It follows from Proposition 1.9 in [ST22] together with the commensurability between DQD_{\mathfrak{Q}} and DzQD_{z\mathfrak{Q}} that Q∣Lm−1\mathfrak{Q}|_{L_{m-1}} and zQ∣Lm−1z\mathfrak{Q}|_{L_{m-1}} are the same. Then Lemma 1.8 in [ST22] implies that z=1z=1 since zz fixes every finite prime. This proves the uniqueness. The implication (ii) ⟹\Longrightarrow (i) is immediate, this proves Theorem 2. □\square

As an immediate consequence, we obtain the following 2-steps loss version of Theorem 1.2 (c.f. Theorem 2 in [ST22]).

Corollary 3.3. Let K,LK,L be number fields, and let m≥2m \ge2 be an integer. Let φm+2:GKm+2→∼GLm+2\varphi_{m+2}:G_K^{m+2}\xrightarrow{\sim}G_L^{m+2} be an isomorphism. Write φm:GKm→∼GLm\varphi_m:G_K^m\xrightarrow{\sim}G_L^m for the induced isomorphism. Then there exists a unique field embedding σm:Lm→∼Km\sigma_m:L_m\xrightarrow{\sim}K_m satisfying σm∘φm(g)=g∘σm\sigma_m\circ\varphi_m(g)=g\circ\sigma_m for all g∈GKmg\in G_K^m and σm(L)⊂K\sigma_m(L)\subset K.

Proof. It follows from Theorem 1.26 in [ST22] that φm+2\varphi_{m+2} is compatible with the cyclotomic characters. Then Theorem 2 implies Corollary 3.2. □\square

The Logarithmic Cyclotomic Compatibility

In this section, we prove the logarithmic cyclotomic compatibility in the mm-step solvable case, we shall follow the strategy in [5] and [7], with suitable adaption in the finite-step solvable situation.

Fix ℓ∈Primes\ell\in\mathrm{Primes} and a number field AA, we shall write

Xℓ(A):=Hom⁡(GA,Qℓ)X_\ell(A):=\operatorname{Hom}(G_A,\mathbb{Q}_\ell)

for the vector space of all continuous characters ρ:GA→Qℓ\rho:G_A\to\mathbb{Q}_\ell. Those characters are unramified at every p∤ℓp\nmid\ell. If A/QA/\mathbb{Q} is Galois, then the action of Gal⁡(A/Q)\operatorname{Gal}(A/\mathbb{Q}) on Xℓ(A)X_\ell(A) is given by (gx)h=x(g~hg~−1)(gx)h=x(\widetilde{g}h\widetilde{g}^{-1}) where g~∈GQ\widetilde{g}\in G_{\mathbb{Q}} is a lifting of gg. We shall denote the averaging operator by

AJ:=∣J∣−1∑g∈JgA_J:=|J|^{-1}\sum_{g\in J}g

for any finite group JJ. For a prime number ℓ\ell, we shall denote by λA,ℓ\lambda_{A,\ell} for the ℓ\ell-adic logarithmic cyclotomic character, i.e. λA,ℓ=log⁡ℓχA,ℓ\lambda_{A,\ell}=\log_\ell\chi_{A,\ell}. In particular, we have Xℓ(Q)=Qℓλℓ,QX_\ell(\mathbb{Q})=\mathbb{Q}_\ell\lambda_{\ell,\mathbb{Q}}.

Lemma 4.1. Let KK be a number field, it holds that

r2(K)+1≤dim⁡Xℓ(K)≤[K:Q]r_2(K)+1\le\dim X_\ell(K)\le[K:\mathbb{Q}]

where r2(K)r_2(K) is the number of complex embeddings of KK and the dimension is the Qℓ\mathbb{Q}_\ell-dimension.

Proof. See section 3.1 in [5].

Lemma 4.2. Let KK be a number field, and let β:GK→P4\beta: G_K \to P^4 be an open homomorphism and fix an odd prime ℓ\ell. Then all but finitely many rational primes q≠ℓq \ne\ell are targets of active primes.

Proof. This is a finite-step solvable version of Proposition 1 in [8], and the arguments are almost identical. Suppose that infinitely many qq are missed. Put C0:=Q(ζℓ)⊂Q1C_0 := \mathbb{Q}(\zeta_\ell) \subset\mathbb{Q}_1. Let qiq_i be infinitely many distinct rational primes away from ℓ\ell. Those qiq_i’s are all unramified in C0C_0, hence the valuations at primes above qiq_i’s show that the qiq_i’s are linearly independent in C0×/(C0×)ℓC_0^\times/(C_0^\times)^\ell. Thus C∞:=C0(q11/ℓ,…,qi1/ℓ,…)C_\infty:= C_0(q_1^{1/\ell},\ldots,q_i^{1/\ell},\ldots) is an elementary abelian ℓ\ell-extension of infinite rank, ramified only at qiq_i’s and possibly at ℓ\ell. Moreover, C∞⊂Q2C_\infty\subset\mathbb{Q}_2. Denote by GK′G_{K'} the inverse image of the open subgroup of P4P^4 whose fixed subfield is C0C_0. Here K′/KK'/K is a finite extension. Then the image of GK′G_{K'} in Γ∞:=Gal⁡(C∞/C0)\Gamma_\infty:= \operatorname{Gal}(C_\infty/C_0) is open, so the pullbacked extension K∞′K'_\infty of K′K' has infinite elementary abelian rank.

At a prime of K′K' not dividing ℓ\ell. Such a prime is active for β∣GK′\beta|_{G_{K'}} and its target rational prime cannot be any of the missed primes. Indeed, activity is stable and compatible with taking open subgroups. On the other hand, the target rational prime cannot be ℓ\ell either. Thus, K∞′/KK'_\infty/K is unramified away from ℓ\ell, but this would imply that K∞′/KK'_\infty/K has finite rank, a contradiction.

Now we shall adapt the method Lemma 1.7 in [7] to derive the logarithmic cyclotomic compatibility in the finite-step solvable case.

Theorem 4.3. Let β:GK→P4\beta: G_K \to P^4 be open. For some integer n0≥1n_0 \ge1, and for sufficiently large odd prime ℓ≡1(modn0)\ell\equiv1 \pmod{n_0},

λQ,ℓ∘β=λK,ℓ.\lambda_{\mathbb{Q},\ell} \circ\beta= \lambda_{K,\ell}.

Proof. Let M/QM/\mathbb{Q} be the Galois closure of KK over Q\mathbb{Q} and put n0:=[M:Q]n_0 := [M : \mathbb{Q}]. We shall denote by AA for the fixed field of the image of GMG_M via β\beta. Choose an odd prime ℓ≡1(modn0)\ell\equiv1 \pmod{n_0} unramified in AA. For integers b≥1b \ge1, put

Eb:=Q(ζℓb,ℓ1/b)andRb:=Q(ℓ1/b)E_b := \mathbb{Q}(\zeta_{\ell^b},\ell^{1/b}) \quad\text{and} \quad R_b := \mathbb{Q}(\ell^{1/b})

and we set R0:=QR_0 := \mathbb{Q}. Thus we have

Gal⁡(Eb/Q)≅Z/ℓbZ⋊(Z/ℓbZ)×\operatorname{Gal}(E_b/\mathbb{Q}) \cong\mathbb{Z}/\ell^b\mathbb{Z} \rtimes(\mathbb{Z}/\ell^b\mathbb{Z})^\times

and set V:=Z/ℓbZV := \mathbb{Z}/\ell^b\mathbb{Z} and C:=(Z/ℓbZ)×C := (\mathbb{Z}/\ell^b\mathbb{Z})^\times and we have EbC=RbE_b^C = R_b. Put c:=∣C∣=(ℓ−1)ℓb−1c := |C| = (\ell-1)\ell^{b-1}. The commutator subgroup of Gal⁡(Eb/Q)\operatorname{Gal}(E_b/\mathbb{Q}) is exactly VV. In particular, Eb⊂Q2E_b \subset\mathbb{Q}_2 and Ebab⊂Q3E_b^{\mathrm{ab}} \subset\mathbb{Q}_3. Furthermore, the EbE_b is unramified away from ℓ\ell by using its discriminant. Therefore, A∩EbA \cap E_b is unramified at every rational primes hence is Q\mathbb{Q}. Therefore, the composite GM→P4↠Gal⁡(Eb/Q)G_M \to P^4 \twoheadrightarrow\operatorname{Gal}(E_b/\mathbb{Q}) is onto for every bb.

The field RbR_b has signature (1,(ℓb−1)/2)(1,(\ell^b-1)/2), then it follows from Lemma 4.1 that

dim⁡Xℓ(Rb)≥(ℓb+1)/2>ℓb−1≥dim⁡Xℓ(Rb−1).\dim X_\ell(R_b) \ge(\ell^b+1)/2 > \ell^{b-1} \ge\dim X_\ell(R_{b-1}).

Take 0≠x∈ker⁡(Res⁡Rb/Rb−1)0 \ne x \in\ker(\operatorname{Res}_{R_b/R_{b-1}}) (where the restriction is on the character spaces Xℓ(−)X_\ell(-)). Let y:=Res⁡Eb/Rb(x)y := \operatorname{Res}_{E_b/R_b}(x), the vector yy is also non-zero and is fixed by CC. Then

Res⁡Eb/Rb−1Cor⁡Rb/Rb−1x=∑u∈V[ℓ]uy.\operatorname{Res}_{E_b/R_{b-1}}\operatorname{Cor}_{R_b/R_{b-1}}x=\sum_{u\in V[\ell]}uy.

Write W0:=Qℓ[V]yW_0 := \mathbb{Q}_\ell[V]y. We claim that W0W_0 is exactly the primitive irreducible VV-module Qℓ(ζℓb)\mathbb{Q}_\ell(\zeta_{\ell^b}). Indeed, the subgroup averaging operator on V[ℓ]V[\ell] annihilates the primitive part of Qℓ[V]\mathbb{Q}_\ell[V], which is a Qℓ(ζℓb)\mathbb{Q}_\ell(\zeta_{\ell^b})-factor of degree cc, and any non-zero cyclic submodule generated in that Qℓ(ζℓb)\mathbb{Q}_\ell(\zeta_{\ell^b})-factor is precisely Qℓ(ζℓb)\mathbb{Q}_\ell(\zeta_{\ell^b}). In the identification sending yy to 11, VV acts by multiplication and CC acts by Galois automorphism, thus the action is faithful and W0CW_0^C is one-dimensional.

Now denote by ρ\rho for the GMG_M-character λQ,ℓ∘β−λM,ℓ\lambda_{Q,\ell}\circ\beta-\lambda_{M,\ell}. The values of ρ\rho are in Zℓ\mathbb{Z}_\ell. For σ∈Gal⁡(M/Q)\sigma\in\operatorname{Gal}(M/\mathbb{Q}), let (σ⋅ρ)g=ρ(σ~gσ~−1)(\sigma\cdot\rho)g=\rho(\widetilde{\sigma}g\widetilde{\sigma}^{-1}) for g∈GMg\in G_M and any lifting σ~∈GQ\widetilde{\sigma}\in G_{\mathbb{Q}}. Notice that this action does not depend on the choice of the liftings σ~\widetilde{\sigma} since ρ\rho is an additive character (which factors through abelianisation). List the distinct characters in this finite orbit as ρ1,…,ρs\rho_1,\ldots,\rho_s. For each 1≤i<j≤s1\le i<j\le s, choose gij∈GMg_{ij}\in G_M with ρi(gij)≠ρj(gij)\rho_i(g_{ij})\ne\rho_j(g_{ij}). Define

b0:=1+max⁡({0}∪{ord⁡ℓ(ρi(gij)−ρj(gij)):1≤i<j≤s}).b_0:=1+\max\bigl(\{0\}\cup\{\operatorname{ord}_\ell(\rho_i(g_{ij})-\rho_j(g_{ij})):1\le i<j\le s\}\bigr).

For every b≥b0b\ge b_0, the reductions of ρi\rho_i and ρj\rho_j modulo ℓb\ell^b differ at gijg_{ij}. In particular, for each σ∈Gal⁡(M/Q)\sigma\in\operatorname{Gal}(M/\mathbb{Q}),

σ⋅ρ≡ρ(modℓb) on GM⟹σ⋅ρ=ρ.\sigma\cdot\rho\equiv\rho\pmod{\ell^b}\text{ on }G_M\quad\Longrightarrow\quad\sigma\cdot\rho=\rho.

We shall fix such an integer bb.

Choose a finite Galois extension F/QF/\mathbb{Q} containing the fixed field of the kernel of GM→V⋊CG_M\to V\rtimes C, in particular M⊂FM\subset F. Write Γ:=Gal⁡(F/Q)\Gamma:=\operatorname{Gal}(F/\mathbb{Q}) and H:=Gal⁡(F/M)H:=\operatorname{Gal}(F/M), write θ:H→V⋊C\theta:H\to V\rtimes C for the induced quotient. The restriction of β(GF)\beta(G_F) to EbabE_b^{\mathrm{ab}} give an open homomorphism GF→GEbabG_F\to G_{E_b}^{\mathrm{ab}} and the pullback

j:Xℓ(Eb)→Xℓ(F)j:X_\ell(E_b)\to X_\ell(F)

is injective and HH-equivariant. Write W:=j(W0)W:=j(W_0).

Fix a generator c∗c_* of CC. By Chebotarev density theorem, there are infinitely many rational primes q≠ℓq\ne\ell whose Frobenius conjugacy class in Gal⁡(Eb/Q)\operatorname{Gal}(E_b/\mathbb{Q}) is the class of c∗c_*. For each such qq, choose wq∗w_q^* above qq in EbE_b with Frobenius exactly c∗c_*. Since qq is unramified, Dwq∗(Eb/Q)=CD_{w_q^*}(E_b/\mathbb{Q})=C. Apply Lemma 4.2 to β∣GM\beta|_{G_M}, for all but finitely many such qq, there is a source prime vqv_q of MM with target rational prime qq.

For each remaining qq, choose a prolongation v~q\widetilde{v}_q of vqv_q to Q‾\overline{\mathbb{Q}}, a pro-ℓ\ell Sylow subgroup Sq⊂Dv~q⊂GMS_q\subset D_{\widetilde{v}_q}\subset G_M. Furthermore choose compatible primes m3,q0,m2,q0:=m3,q0∣Q2m^0_{3,q},m^0_{2,q}:=m^0_{3,q}|_{\mathbb{Q}_2} as in Lemma 2.1. Set wq0:=m2,q0∣Ebw_q^0:=m^0_{2,q}|_{E_b}. It follows from the transitivity of the Galois action on primes, we may choose hq∈GMh_q \in G_M with β2(hq)∣Ebwq0=wq∗\beta_2(h_q)|_{E_b}w_q^0 = w_q^*. Now put

v~q=hqv~q0,Sq=hqSq0hq−1,mj,q=βj(hq)mj,q0 for j=2,3.\widetilde{v}_q=h_q\widetilde{v}_q^0,\quad S_q=h_qS_q^0h_q^{-1},\quad m_{j,q}=\beta_j(h_q)m_{j,q}^0\text{ for }j=2,3.

As hqh_q fixes MM, v~q\widetilde{v}_q still lies above vqv_q. Conjugating the two containments in Lemma 2.1 gives

β3(Sq)⊂Dm3,q(Q3/Q) and β2(Dv~q)⊂Dm2,q(Q2/Q),\beta_3(S_q)\subset D_{m_{3,q}}(\mathbb{Q}_3/\mathbb{Q})\text{ and }\beta_2(D_{\widetilde{v}_q})\subset D_{m_{2,q}}(\mathbb{Q}_2/\mathbb{Q}),

and by construction, we have m3,q∣Q=m2,qm_{3,q}|_{\mathbb{Q}}=m_{2,q} and wq′:=m2,q∣Eb=wq∗w_q':=m_{2,q}|_{E_b}=w_q^*. Set vq′:=v~q∣Fv_q':=\widetilde{v}_q|_F and p0(q):=vq′∣Q=vq∣Qp_0(q):=v_q'|_{\mathbb{Q}}=v_q|_{\mathbb{Q}}. The primes vqv_q are all distinct since each has a unique target rational prime qq. Thus q↦p0(q)q\mapsto p_0(q) has finite fibres. After excluding the qq for which p0(q)p_0(q) ramifies in FF, choose an infinite set TT of remaining qq on which p0(q)p_0(q) is injective. For q∈Tq\in T, let δq:=Frob⁡vq′/p0(q)∈Γ\delta_q:=\operatorname{Frob}_{v_q'/p_0(q)}\in\Gamma be the Frobenius at the specified prime vq′v_q'. Since Γ\Gamma is finite, there exists δ∈Γ\delta\in\Gamma such that {q∈T:δq=δ}\{q\in T:\delta_q=\delta\} is infinite. In this case, we set Δ:=⟨δ⟩\Delta:=\langle\delta\rangle. Also, we shall denote by v,v′,w′,p0v,v',w',p_0 for vq,vq′,wq′,p0(q)v_q,v_q',w_q',p_0(q) respectively.

Write ff for the inertia degree of vv over p0p_0, i.e. ff is the order of δ∣M\delta|_M. A Frobenius lifting at vv maps to δf∈H\delta^f\in H. It follows from the containment β2(Dv~q)⊂Dm2,q(Q2/Q)\beta_2(D_{\widetilde{v}_q})\subset D_{m_{2,q}}(\mathbb{Q}_2/\mathbb{Q}) that θ(δf)∈C\theta(\delta^f)\in C. Then Lemma 2.3 identifies its cyclotomic image with p0f(modℓb)p_0^f\pmod{\ell^b}. On CC the component is the isomorphism with (Z/ℓbZ)×(\mathbb{Z}/\ell^b\mathbb{Z})^\times, so θ(δf)\theta(\delta^f) is determined by its value. Since f∣cf\mid c, it follows from p0c≡1(modℓb)p_0^c\equiv1\pmod{\ell^b} that θ(δc)=1\theta(\delta^c)=1. Therefore, δc\delta^c acts trivially on WW.

For a selected v′v', let Dv′⊂GFD_{v'}\subset G_F be its decomposition group and Iv′I_{v'} for the inertia subgroup for the chosen prolongation v~q\widetilde{v}_q. And choose a Frobenius lifting g∈Dv′g\in D_{v'} of v′v'. In the procyclic group ⟨g⟩\langle g\rangle, take its pro-ℓ\ell component gℓg_\ell. Any additive (logarithmic) ℓ\ell-adic character takes the same value of gℓg_\ell as on gg since the prime-to-ℓ\ell part has zero image. We have gℓ∈Dv′⊂Dvg_\ell\in D_{v'}\subset D_v. Take a pro-ℓ\ell Sylow of DvD_v containing gℓg_\ell. By Lemma 2.1, the image of such pro-ℓ\ell Sylow is contained a target 3-step solvable decomposition subgroup, whose associated prime restricts to Q2\mathbb{Q}_2 is the one attached to vv. Thus the image of gℓg_\ell fixes EbE_b, which gives a local element at w′w' with residue exponent a∈Za\in\mathbb{Z}. Then Lemma 2.3 gives

aℓlog⁡ℓNw′=log⁡ℓNv′a_\ell\log_\ell Nw'=\log_\ell Nv'

where aℓa_\ell is the ℓ\ell-adic part of aa. The logarithms on the both sides are non-zero since p0,q≠ℓp_0,q\ne\ell. Then for every x0∈W0x_0\in W_0, we have

j(x0)(Frob⁡v′)=log⁡ℓNv′log⁡ℓNw′x0(Frob⁡w′).j(x_0)(\operatorname{Frob}_{v'})=\frac{\log_\ell Nv'}{\log_\ell Nw'}x_0(\operatorname{Frob}_{w'}).

The evaluation at Frobenius x0↦x0(Frob⁡w′)x_0\mapsto x_0(\operatorname{Frob}_{w'}) on W0W_0 is CC-invariant. Indeed, a lifting γ∈C\gamma\in C in the decomposition group in GQG_{\mathbb{Q}} at the chosen prolongation acts trivially by conjugation on the unramified Galois group over Qq\mathbb{Q}_q. Hence (γ⋅x0)(Frob⁡w′)=x0(Frob⁡w′)(\gamma\cdot x_0)(\operatorname{Frob}_{w'})=x_0(\operatorname{Frob}_{w'}) and (ACx0)(Frob⁡w′)=x0(Frob⁡w′)(\mathcal{A}_C x_0)(\operatorname{Frob}_{w'})=x_0(\operatorname{Frob}_{w'}). If ACx0=0\mathcal{A}_C x_0=0, then Proposition 3.2 (ii) in [5] gives AΔj(x0)=0\mathcal{A}_{\Delta}j(x_0)=0. Conversely, choose 0≠y∈W0C0 \ne y \in W_0^C. It descends to the non-zero character c−1Cor⁡Eb/EbCy0c^{-1}\operatorname{Cor}_{E_b/E_b^C}y_0 on GEbCG_{E_b^C}, whose restriction to GEbG_{E_b} is y0y_0 by CC-invariance. The restrictions of the chosen w′w' to EbCE_b^C have degree 1 and relative inertia degree cc in EbE_b. Then Proposition 3.2 (i) in [5], y(Frob⁡w′)≠0y(\operatorname{Frob}_{w'}) \ne0 for all but finitely many chosen w′w'. If ACx0≠0A_Cx_0 \ne0, we have x0(Frob⁡v′)≠0x_0(\operatorname{Frob}_{v'}) \ne0. Similarly, (AΔj(x0))(Frob⁡v′)=j(x0)(Frob⁡v′)(A_{\Delta}j(x_0))(\operatorname{Frob}_{v'}) = j(x_0)(\operatorname{Frob}_{v'}), since Dv′(F/Q)=ΔD_{v'}(F/\mathbb{Q}) = \Delta and its liftings act trivially by conjugation on the local unramified quotient. Hence AΔj(x0)≠0A_{\Delta}j(x_0) \ne0. And thus,

j(ker⁡(AC∣W0))=ker⁡(AΔ∣W).j(\ker(A_C|_{W_0})) = \ker(A_{\Delta}|_W).

Since j:W0→Wj: W_0 \to W is an isomorphism, and AC∣W0A_C|_{W_0} projects onto W0CW_0^C, AΔ∣W:W→Xℓ(F)A_{\Delta}|_W: W \to X_\ell(F) has rank 1.

Now let D:=[H,H]D := [H,H] which is normal in Γ\Gamma with θ(D)=V\theta(D) = V. Since WW decomposes into primitive eigenlines, choose an eigenvector uu with AΔu≠0A_{\Delta}u \ne0. The rank one image of AΔ∣WA_{\Delta}|_W is defined over Qℓ\mathbb{Q}_\ell, rescale uu so that z=AΔu∈Xℓ(F)z = A_{\Delta}u \in X_\ell(F). Let Z:=Qℓ[D]zZ := \mathbb{Q}_\ell[D]z, which is stable under δ\delta, because δ\delta fixes zz and normalises DD. The scalar extension of ZZ is contained in the span of δju\delta^j u for 0≤j<c0 \le j < c. Indeed, each is a DD-eigenvector and δcu=u\delta^c u = u. Every character appearing in this span has order ℓb\ell^b. Since Z≠0Z \ne0 and representations of the finite group DD in characteristic 0 are semisimple, some element of DD has a primitive ℓb\ell^b-th root eigenvalue on ZZ. The degree of its minimal polynomial over Qℓ\mathbb{Q}_\ell is of degree cc, so dim⁡Z≥c\dim Z \ge c. The span above has dimension at most cc, equality gives uu in the scalar extension of Z∩WZ \cap W. Therefore Z∩W≠0Z \cap W \ne0. The irreducible of WW over Qℓ[D]\mathbb{Q}_\ell[D] and dimension cc gives Z=WZ = W.

The faithfulness of the action of Gal⁡(Eb/Q)\operatorname{Gal}(E_b/\mathbb{Q}) on WW shows that δ\delta normalises ker⁡(θ)\ker(\theta) in HH (i.e. ker⁡(θ)=ker⁡(H→GL⁡(W))\ker(\theta) = \ker(H \to\operatorname{GL}(W))). Therefore, it induces an automorphism on V⋊CV \rtimes C. The subgroup VV is the derived subgroup hence is characteristic, the induced action on C=(V⋊C)/VC = (V \rtimes C)/V is trivial since C→Aut⁡(V)C \to\operatorname{Aut}(V) is injective and Aut⁡(V)\operatorname{Aut}(V) is abelian. Thus conjugation by a lifting δ\delta preserves χQℓ∘β\chi_{\mathbb{Q}_\ell} \circ\beta (mod ℓb\ell^b) on GMG_M. Thus

(δ∣M)⋅ρ≡ρ(modℓb)(\delta|_M)\cdot\rho\equiv\rho\pmod{\ell^b}

on GMG_M, and by the choice of bb, we have (δ∣M)⋅ρ=ρ(\delta|_M)\cdot\rho= \rho.

Now Res⁡F/Mρ\operatorname{Res}_{F/M}\rho is fixed by Δ\Delta and vanishes on the chosen Frob⁡v′\operatorname{Frob}_{v'} by Lemma 2.3. Then by Proposition 3.2 (ii) in [5], its average is 0 and since it is invariant, the character itself must be 0. Restricting on additive characters is injective from GKG_K to GMG_M and from GMG_M to GFG_F, which proves the desired equality (where the congruence progression has infinitely many primes and only a fixed finite set was excluded, which does not depend on the choice of bb):

λQℓ∘β=λK,ℓ.\lambda_{\mathbb{Q}_\ell} \circ\beta= \lambda_{K,\ell}.

Remark 4.4. The proof of Theorem 4.3 can be viewed as the mm-step solvable reinterpretation of section 4 in [5] together with Lemma 1.7 in [Tsu26.

In particular, at this moment, it is not clear that what should be an mm-step solvable version of Lemma 4.1 in [5], therefore, Tsujimura’s arguments in Lemma 1.7 in [7] plays an important role in the proof of Theorem 4.3, in order to avoid an mm-step solvable version of Lemma 4.1 in [5].

From Logarithmic Cyclotomic Compatibility to Cyclotomic Compatibility

In this section, we shall prove the cyclotomic compatibility using Theorem 4.3 and hence prove Theorem 3 and thus deduce Theorem 2 from Theorem 3. We shall follow the strategy in [5] to deduce the cyclotomic compatibility from Theorem 4.3. For the rest of this section, we shall use the followings

  • Fix an open homomorphism β:GK→P4\beta: G_K \to\mathcal{P}^4.

  • A set L\mathcal{L} of rational primes, satisfying the congruence condition in the statement of Theorem 4.3.

  • B:=Q(i)B := \mathbb{Q}(i) and FBF_B for the finite extension of KK obtained by forming the fixed subfield of K‾\overline{K} of the inverse image of Gal⁡(Q4/B)\operatorname{Gal}(\mathbb{Q}_4/B) by β\beta. In this case [FB:K]≤2[F_B : K] \le2.

  • d:=[FB:Q]d := [F_B : \mathbb{Q}].

  • For a given ℓ∈L\ell\in\mathcal{L}, and an active prime vv of FBF_B, let q∈PQ2<∞q \in\mathcal{P}_{\mathbb{Q}_2}^{<\infty} be the prime corresponding vv and w:=q∣Bw := q|_B. In particular, by Lemma 2.1, the image of DvD_v via β2\beta_2 is contained in Dq(Q2/B)D_q(\mathbb{Q}_2/B); and a pro-ℓ\ell inertia of DvD_v is contained in Iq(Q2/B)I_q(\mathbb{Q}_2/B). In this case, we call (v,w)(v,w) an active pair at ℓ\ell, here ww is unique by Proposition 1.3 in [ST22].

  • For each w∈PB<∞w \in\mathcal{P}_B^{<\infty}, put Vw:={v∈PFB<∞:(v,w)V_w := \{v \in\mathcal{P}_{F_B}^{<\infty} : (v,w) is active for some ℓ∈L}\ell\in\mathcal{L}\}.

Lemma 5.1. For all but finitely many degree 1 prime w∈PB<∞w \in\mathcal{P}_B^{<\infty}, VwV_w is finite.

Proof. We shall adapt the proof of Lemma 5.1 in [5] to prove this lemma. Fix r∈Lr \in\mathcal{L}. Since B1=Bab⊂Q2B_1 = B^{\mathrm{ab}} \subset\mathbb{Q}_2, restricting β\beta to GFBG_{F_B} gives the pullback

jB:Xr(B)→Xr(FB).j_B : X_r(B) \to X_r(F_B).

The composite

cB:=Cor⁡FB/Q∘jB:Xr(B)→Xr(Q)c_B := \operatorname{Cor}_{F_B/\mathbb{Q}} \circ j_B : X_r(B) \to X_r(\mathbb{Q})

has non-zero kernel since dim⁡Xr(B)≥2\dim X_r(B) \ge2 and dim⁡Xr(Q)=1\dim X_r(\mathbb{Q}) = 1. Let 0≠ψ∈ker⁡(cB)0 \ne\psi\in\ker(c_B), and set x0:=λB,rx_0 := \lambda_{B,r}. Theorem 4.3 gives

jBx0=λFB,rj_Bx_0 = \lambda_{F_B,r}

hence cB(x0)=dλQ,r≠0c_B(x_0) = d\lambda_{\mathbb{Q},r} \ne0. It follows from Proposition 3.2 (i) in [5] that ψ(Frob⁡w)≠0\psi(\operatorname{Frob}_w) \ne0 for all but finitely many degree one primes w∤rw \nmid r. For such ww over a rational prime qq, write t:=ψ(Frob⁡w)/log⁡rq≠0t := \psi(\operatorname{Frob}_w)/\log_r q \ne0. The character ψ−tx0\psi- tx_0 is unramified at ww and vanishes on the decomposition group of ww. Then it follows from Lemma 2.1 that jB(ψ−tx0)j_B(\psi- tx_0) vanishes on the decomposition group of vv for every v∈Vwv \in V_w. Then the desired finiteness follows from Proposition 3.2 (iii) in [5].

Lemma 5.2. Let kk be a number field and let S⊂PkS \subset\mathcal{P}_k containing all archimedean primes and primes above ℓ\ell for some rational prime ℓ\ell. Then

H1(Gk,Qℓ(1))unramified outside S→∼Ok,S×⊗ZQℓ.H^1(G_k,\mathbb{Q}_{\ell}(1))_{\mathrm{unramified\ outside}\ S} \xrightarrow{\sim} \mathcal{O}_{k,S}^{\times} \otimes_{\mathbb{Z}} \mathbb{Q}_{\ell}.

Proof. See Lemma 5.2 in [5].

Proposition 5.3. For every ℓ∈L\ell\in\mathcal{L}, at most 2d2d primes w∤ℓw \nmid\ell of BB fail to appear in an active pair at ℓ\ell, if including the primes above ℓ\ell, the number is at most 2d+22d+2.

Proof. The same proof of Proposition 5.3 in [5] also works in our case with suitable finite-step solvable adaption. We shall only explain the finite-step solvable adaption here, the determination of the bounds is identical so we omit the identical details here.

Put H:=Gal⁡(Q‾2/B)H := \operatorname{Gal}(\overline{\mathbb{Q}}_2/B). For a∈B×a \in B^{\times}, the radicals a1/ℓna^{1/\ell^n} lie in Q2\mathbb{Q}_2 by Kummer theory. Hence the Kummer classes associated to the missing primes are defined on HH and inflate to the usual Kummer classes in H1(GB,Qℓ(1))H^1(G_B,\mathbb{Q}_{\ell}(1)) (c.f. Lemma 5.2). Again by Lemma 5.2, pullback along β2∣GFB:GFB→H\beta_2|_{G_{F_B}}:G_{F_B}\to H is injective on H1(H,Qℓ(1))H^1(H,\mathbb{Q}_{\ell}(1)).

Theorem 4.3 gives ϵ:=(χQ,ℓ∘β∣GFB)χFB,ℓ−1\epsilon:= (\chi_{\mathbb{Q},\ell}\circ\beta|_{G_{F_B}})\chi_{F_B,\ell}^{-1} with image in μℓ−1\mu_{\ell-1}, which also gives a finite cyclic extension F′:=Q‾ker⁡(ϵ)F' := \overline{\mathbb{Q}}^{\ker(\epsilon)} of FBF_B. The unramifiedness of the Kummer class on GF′G_{F'} follows from the 5th bullet point at the beginning of this section. Together with Lemma 2.1, we may then apply the determination of the bounds used in the proof of Proposition 5.3 in [5] to obtain the desired bound.

Lemma 5.4. Let q1,…,qsq_1,\ldots,q_s be distinct rational primes, and for each jj, let Pj\mathcal{P}_j be a finite set of rational primes not containing qjq_j. For any integers n0,d≥1n_0,d \ge1, there are infinitely many odd primes ℓ≡1(modn0)\ell\equiv1 \pmod{n_0} distinct from every qjq_j and every prime in ⋃jPj\bigcup_j \mathcal{P}_j outside any prescribed finite set of primes such that

pf(modℓ)∉⟨qj(modℓ)⟩p^f \pmod{\ell} \notin\langle q_j \pmod{\ell}\rangle

where 1≤j≤s1 \le j \le s, p∈Pjp \in\mathcal{P}_j and 1≤f≤d1 \le f \le d.

Proof. See Lemma 5.4 in [5] for the proof.

Proposition 5.5. All but finitely many degree 1 primes ww of BB occur in an active pair at some ℓ∈L\ell\in\mathcal{L} with a source prime of the same residue characteristic.

Proof. See Proposition 5.5 in [5] for the proof.

Proposition 5.6. For every N≥1N \ge1, put EN:=BQ(ζN)⊂Q1E_N := \mathrm{BQ}(\zeta_N) \subset\mathbb{Q}_1. Let πN:P4→RN:=Gal⁡(EN/Q)\pi_N : P^4 \to R_N := \operatorname{Gal}(E_N/\mathbb{Q}). Then for every g∈GKg \in G_K, we have

πN(β(g))=πN(g∣Q4).\pi_N(\beta(g)) = \pi_N(g|_{\mathbb{Q}_4}).

Proof. We adapt the proof of Proposition 6.1 in [5]. Put βN:=πN∘β\beta_N := \pi_N \circ\beta. Fix r∈Lr \in\mathcal{L}, choose a finite Galois extension M~/Q\widetilde{M}/\mathbb{Q} containing FBF_B, ENE_N and the fixed subfield Q‾ker⁡(βN)\overline{\mathbb{Q}}^{\ker(\beta_N)}. In particular, GM~⊂GFBG_{\widetilde{M}} \subset G_{F_B} and β(GM~)\beta(G_{\widetilde{M}}) fixes ENE_N. Since ENab⊂Q2E_N^{\mathrm{ab}} \subset\mathbb{Q}_2, there are two injective pullbacks

jβ,jnat:Xr(EN)↪Xr(M~)j_\beta,j_{\mathrm{nat}} : X_r(E_N) \hookrightarrow X_r(\widetilde{M})

where jβj_\beta is induced by β∣GM~\beta|_{G_{\widetilde{M}}} restricted to ENabE_N^{\mathrm{ab}} and jnatj_{\mathrm{nat}} is induced by restricting Galois automorphisms.

Choose infinitely many rational primes q≠rq \ne r splitting completely in M~\widetilde{M}, omitting the finite exceptional set in Proposition 5.5. For each of such qq, choose a prime ww above qq of BB, and an active pair (v,w)(v,w) at some prime ℓ∈L\ell\in\mathcal{L} with vv a prime of FBF_B above qq. Let v′v' be the induced prime of M~\widetilde{M} above vv and let uu be the induced prime of ENE_N on the other side. Both v′v' and uu have degree 1 since qq splits completely in M~\widetilde{M}. If g∈Dv′g \in D_{v'} is a Frobenius lifting, take a Frobenius lifting of β2(g)\beta_2(g) in the corresponding decomposition group, it holds that

arlog⁡rq=λQ,r(β(g))=λM~,r(g)=log⁡rqa_r\log_r q = \lambda_{\mathbb{Q},r}(\beta(g)) = \lambda_{\widetilde{M},r}(g) = \log_r q

where ara_r is the rr-adic component of the residue exponent of β2(g)\beta_2(g). Since log⁡rq≠0\log_r q \ne0, ar=1a_r = 1.

Since the Qr\mathbb{Q}_r-characters are unramified away from rr, for all x∈Xr(EN)x \in X_r(E_N), we have

jβ(x)(Frob⁡v′)=x(Frob⁡u).j_\beta(x)(\operatorname{Frob}_{v'}) = x(\operatorname{Frob}_u).

On the other hand,

jnat(x)(Frob⁡v′)=x(Frob⁡u0)j_{\mathrm{nat}}(x)(\operatorname{Frob}_{v'}) = x(\operatorname{Frob}_{u_0})

where u0=v′∣ENu_0 = v'|_{E_N}. Choose a∈RNa \in R_N with a(u)=u0a(u) = u_0, the conjugation action (c.f. at the beginning of section 4) gives

x(Frob⁡u)=(ax)(Frob⁡u0).x(\operatorname{Frob}_u) = (ax)(\operatorname{Frob}_{u_0}).

Since RNR_N is finite, we may keep infinitely many qq with one fixed point of aa. For each xx, the difference jβ(x)−jnat(ax)j_\beta(x)-j_{\mathrm{nat}}(ax) vanishes at infinitely many degree 1 Frobenii, so is zero by Proposition 3.2 (i) in [5]. Thus,

jβ=jnat∘a.j_\beta= j_{\mathrm{nat}} \circ a.

The GKG_K-action on Xr(M~)X_r(\widetilde{M}) is well-defined since M~/Q\widetilde{M}/\mathbb{Q} is Galois, we obtain the followings by conjugation and pullback

gjβ(x)=jβ(βN(g)x), and gjnat(x)=jnat(σN(g)x)g j_\beta(x) = j_\beta(\beta_N(g)x), \text{ and } g j_{\mathrm{nat}}(x) = j_{\mathrm{nat}}(\sigma_N(g)x)

where σN\sigma_N denotes the composite GK→g↦g∣Q4P4→RNG_K \xrightarrow{g\mapsto g|_{\mathbb{Q}_4}} P^4 \to R_N. Substitute the identification jβ=jnat∘aj_\beta=j_{\mathrm{nat}}\circ a gives

aβN(g)x=σN(g)axa\beta_N(g)x=\sigma_N(g)ax

for all xx. Since the action of RNR_N on Xr(EN)X_r(E_N) is faithful by Proposition 3.3 in [5], aβN(g)=σN(g)aa\beta_N(g)=\sigma_N(g)a. On the other hand, RNR_N is abelian gives βN(g)=σN(g)\beta_N(g)=\sigma_N(g), as required. □

Now we obtain the 4-step solvable version of the cyclotomic compatibility.

Theorem 5.7 (Theorem 3). Every open homomorphism β:GK→P4\beta:G_K\to P^4 satisfies χQ∘β=χK\chi_{\mathbb{Q}}\circ\beta=\chi_K. Consequently every open homomorphism GK4→GL4G_K^4\to G_L^4 is compatible with the cyclotomic characters.

Proof. We apply Proposition 5.6 for every positive NN, this gives χQ∘β=χK\chi_{\mathbb{Q}}\circ\beta=\chi_K. Now consider the composite

GK4→GL4→P4G_K^4\to G_L^4\to P^4

where GK4→GL4G_K^4\to G_L^4 is an arbitrary open homomorphism, GL4→P4G_L^4\to P^4 is given by η↦η∣Q4\eta\mapsto\eta|_{\mathbb{Q}_4}. Since GL4→P4G_L^4\to P^4 is defined to be restriction, hence preserves the cyclotomic character, thus we have the following commutative diagram

GK↓GK4⟶GL4↘↓↙P4↓Z^×\begin{aligned} &G_K \\ &\downarrow\\ &G_K^4 \longrightarrow G_L^4 \\ &\searrow\quad\downarrow\quad\swarrow\\ &\qquad P^4 \\ &\qquad\downarrow\\ &\qquad\widehat{\mathbb{Z}}^\times \end{aligned}

and in particular any open homomorphism GK4→GL4G_K^4\to G_L^4 preserves the cyclotomic characters, as required. □

As a consequence, we obtain the mm-step solvable version of Uchida’s hom conjecture for number fields at a 2-solvable depth loss:

Theorem 5.8 (Theorem 1). Let m≥2m\ge2 be an integer and let K,LK,L be number fields. Let φm+2:GKm+2→GLm+2\varphi_{m+2}:G_K^{m+2}\to G_L^{m+2} be an open homomorphism. Let φm:GKm→GLm\varphi_m:G_K^m\to G_L^m be the open homomorphism induced by φm+2\varphi_{m+2} (c.f. Proposition 2.2 in [CS23]). Then there exists a unique field embedding τm:Lm↪Km\tau_m:L_m\hookrightarrow K_m satisfying g∘τm=τm∘φm(g)g\circ\tau_m=\tau_m\circ\varphi_m(g) for all g∈GKmg\in G_K^m.

Proof. Since m+2≥4m+2\ge4, Theorem 5.7 implies that φm+2:GKm+2→GLm+2\varphi_{m+2}:G_K^{m+2}\to G_L^{m+2} is automatically compatible with the cyclotomic characters, then Theorem 3.2 gives the desired assertion. □

Remark 5.9. We need 2 extra steps above mm to use (⋆ℓ)(\star_{\ell}) condition, which is minimal to prove Theorem 3.2. On the other hand, when dealing with β:GK→P4\beta: G_K \to P^4, we need the two extra steps to obtain the local containment at metabelian level, where the metabelian level is the minimal level we can maintain the notation of active pairs.

References

  1. [2]Yuichiro Hoshi. Homomorphisms of global solvably closed Galois groups compatible with cyclotomic characters. Tohoku Mathematical Journal, 77(1):17 – 32, 2025.DOI
  2. [4]Jürgen Neukirch, Alexander Schmidt, and Kay Wingberg. Cohomology of Number Fields. Springer, Berlin, 2008.DOI
  3. [5]OpenAI. Open Homomorphisms of Global Solvably Closed Galois Groups. OpenAI Math Release preprint OAI:Open-Homomorphisms-of-Global-Solvably-Closed-Galois-Groups-October-5-2026, 2026.DOI
  4. [7]Shota Tsujimura. A higher dimensional generalization of uchida’s hom conjecture. Preprint, 2026.
  5. [8]Kôji Uchida. Homomorphisms of Galois groups of solvably closed Galois extensions. Journal of the Mathematical Society of Japan, 33(4): 595 – 604, 1981.DOI

Paper details

vlnc:W_tdk7w2t8
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 7

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents