betaFeedback

WorksMathematics

Counting Nilpotent Galois extensions with prescribed genus number

Abstract

Let GG be a nontrivial finite nilpotent group. In this paper, we study the distribution of genus numbers of Galois extensions of \Q with Galois group isomorphic to GG, ordered by absolute discriminant. We show that the proportion of fields with any prescribed genus number tends to zero, and obtain a quantitative bound for this proportion. Moreover, if every element of order ℓ\ell in GG is central, where ℓ\ell is the smallest prime divisor of ∣G∣|G|, we determine the orders of magnitude of the average genus number and its higher moments.

Introduction

For a number field K/QK/\mathbb{Q}, its absolute genus field K∗K^* is the largest extension of KK obtained by adjoining an abelian extension of Q\mathbb{Q}, such that the resulting extension is contained in the narrow Hilbert class field of KK. The genus number of KK is gK=[K∗:K]g_K=[K^*:K]. One can ask how the genus number varies as KK varies in a family of number fields of fixed degree. In this paper, we consider this question for Galois extensions of Q\mathbb{Q} with a fixed non-trivial finite nilpotent Galois group GG. More precisely, how often does the genus number attain a prescribed value in this family and how large is it on average? We also study higher moments of genus numbers in these families.

Before addressing these questions, let us first review counting results on the number of Galois extensions of Q\mathbb{Q} with a prescribed Galois group. Shafarevich showed that every finite solvable group occurs as Galois group over Q\mathbb{Q} (see [14 Theorem 9.6.1]). In particular, if GG is nilpotent, there exists at least one Galois extension KK with Gal⁡(K/Q)\operatorname{Gal}(K/\mathbb{Q}) isomorphic to GG. One can ask how frequently they occur when ordered by their absolute discriminant. Let NG(X)N_G(X) be the number of such extensions with absolute discriminant at most XX. The strong form of Malle’s conjecture [9, 10] predicts that

NG(X)∼cGXa(G)(log⁡X)b(G,Q)−1N_G(X) \sim c_G X^{a(G)}(\log X)^{b(G,\mathbb{Q})-1}

for a positive constant cGc_G. Here a(G)=ℓ/((ℓ−1)∣G∣)a(G)=\ell/((\ell-1)|G|), where ℓ\ell is the smallest prime divisor of ∣G∣|G|, and b(G,Q)b(G,\mathbb{Q}) is the number of conjugacy classes of subgroups of order ℓ\ell in GG. The weak form predicts that for every ε>0\varepsilon>0,

Xa(G)≪GNG(X)≪G,εXa(G)+ε.X^{a(G)} \ll_G N_G(X) \ll_{G,\varepsilon} X^{a(G)+\varepsilon}.

The weak form was proved for nilpotent groups by Klüners-Malle [7]. More recently, Koymans–Pagano constructed a parametrization of nilpotent extensions in which the coordinates record ramification. Using this parametrization, they proved the strong form whenever every element of order ℓ\ell in GG is central; see [8 Theorems 1.1, 1.3 and 1.4]. In this case, b(G,Q)b(G,\mathbb{Q}) is the number of elements of order ℓ\ell divided by ℓ−1\ell-1. For an arbitrary finite nilpotent group, their result [8 Theorem 1.3, p. 312] gives

NG(X)≪GXa(G)(log⁡X)∣I(G)∣/(ℓ−1)−1,N_G(X) \ll_G X^{a(G)}(\log X)^{\lvert I(G)\rvert/(\ell-1)-1},

where I(G)I(G) is the set of elements of order ℓ\ell in GG.

We now recall some results on genus numbers in families of number fields of fixed degree. McGown-Tucker [11] studied cubic fields ordered by absolute discriminant, treating the totally real and complex families separately. They proved that approximately 96.23% of fields in either family have genus number one. More generally, the possible genus numbers are powers of 3 and each such power occurs for a positive proportion of cubic fields. They also showed that the average genus number tends to approximately 1.07854, in both signatures. However, the behaviour is different if one restricts to cyclic cubic fields. In this family, the proportion with genus number one tends to zero, whereas the average genus number tends to infinity. Note that cyclic cubic fields have density zero among all cubic fields. Recently, Yamada [Yam26] refined the cubic results of McGown-Tucker by obtaining secondary terms and sharper error estimates.

An analogous result for quintic fields was obtained by McGown-Thorne-Tucker [12]. When these fields are ordered by absolute discriminant, approximately 99.9935% have genus number one, with the same proportion in each signature. Nevertheless, every fixed power of 5 occurs as the genus number of a positive proportion of quintic fields. The limiting average is approximately 1.00026, again independently of the signature. These results also hold on restricting to quintic fields whose Galois closures have group S5S_5.

Kim [6] studied cyclic and dihedral fields of prime degree ℓ≥5\ell\ge5. His results are unconditional for cyclic fields, while the dihedral case requires additional information about class groups. Bose-McGown-Panpaliya-Welling-Williams [1] considered pure fields of odd prime degree, namely fields of the form Q(aℓ)\mathbb{Q}(\sqrt[\ell]{a}), ordered by absolute discriminant. Their Theorem 1.1 shows that the proportion with genus number one is asymptotic to a positive constant divided by log⁡X\log X. Their Theorem 1.2 gives an asymptotic formula for the average genus number, with order of growth (log⁡X)ℓ−1(\log X)^{\ell-1} and a positive leading constant. For abelian extensions, Frei–Loughran–Newton prove a general theorem over an arbitrary number field [3]. They obtain an asymptotic formula for the average genus number and show that each fixed genus value has density zero. Their extensions on the other hand are ordered by conductor. The first author and Pasupulati [2] studied further nonabelian families, including degree-3q3q fields whose Galois closures have group S3×Z/qZS_3 \times\mathbb{Z}/q\mathbb{Z}, where q≠3q \ne3 is prime, and quartic fields whose Galois closures have group D4D_4. Their Theorems 1.1–1.3 show that prescribed admissible genus numbers occur with positive proportions. They also obtain asymptotic formulas for averages and higher moments in the S3×Z/qZS_3 \times\mathbb{Z}/q\mathbb{Z} families. These different behaviours led them to formulate the following conjecture.

Conjecture 1.1 (Dixit–Pasupulati). Let GG be a nontrivial finite solvable group. Let NG(X)N_G(X) denote the number of Galois extensions L/QL/\mathbb{Q}, counted up to Q\mathbb{Q}-isomorphism, such that Gal⁡(L/Q)≃G\operatorname{Gal}(L/\mathbb{Q}) \simeq G and ∣Disc⁡(L)∣≤X\lvert\operatorname{Disc}(L)\rvert\le X. For a positive integer tt, let NG,t(X)N_{G,t}(X) count those extensions with absolute genus number gL=tg_L=t. Then, for every fixed t≥1t \ge1,

lim⁡X→∞NG,t(X)NG(X)=0.\lim_{X \to\infty} \frac{N_{G,t}(X)}{N_G(X)} = 0.

The above formulation is equivalent to [2 Conjecture 1.4], which is stated in terms of towers of abelian extensions. To see this in the regular representation, note that a Galois extension L/QL/\mathbb{Q} admits a tower

Q=K0⊂K1⊂⋯⊂Kr=L\mathbb{Q} = K_0 \subset K_1 \subset\cdots\subset K_r = L

with each Ki/Ki−1K_i/K_{i-1} abelian if and only if Gal⁡(L/Q)\operatorname{Gal}(L/\mathbb{Q}) is solvable. This follows from the Galois correspondence. The restriction to the regular representation is relevant here. The positive-density results for quartic D4D_4-fields concern non-Galois fields of degree four, whereas a Galois extension with group D4D_4 has degree eight.

Main results

We now state the main results. Let GG be a nontrivial finite nilpotent group, and let ℓ\ell be the smallest prime divisor of its order. Recall that NG(X)N_G(X) counts the Galois extensions L/QL/\mathbb{Q} with Gal⁡(L/Q)≃G\operatorname{Gal}(L/\mathbb{Q}) \simeq G and absolute discriminant at most XX. Let NG,t(X)N_{G,t}(X) count the number of these extensions with genus number tt. Let dℓ(G)d_\ell(G) be the largest integer dd for which GG has a quotient isomorphic to (Z/ℓZ)d(\mathbb{Z}/\ell\mathbb{Z})^d, and let νℓ(t)\nu_\ell(t) denote the exponent of ℓ\ell in the prime factorization of tt. Set RG(t)=dℓ(G)+νℓ(t)R_G(t) = d_\ell(G) + \nu_\ell(t). We prove the following.

Theorem 1.2. For every fixed positive integer tt and every X≥3X \ge3,

NG,t(X)≪G,tXa(G)log⁡X(1+log⁡log⁡X)RG(t)−1.(1)N_{G,t}(X) \ll_{G,t} \frac{X^{a(G)}}{\log X}(1+\log\log X)^{R_G(t)-1}. \tag*{(1)}

By the lower bound of Klüners-Malle, we have NG(X)≫GXa(G)N_G(X) \gg_G X^{a(G)}. Thus, we immediately deduce the following consequence.

Corollary 1.3. For every nontrivial finite nilpotent group GG and every fixed positive integer tt, the proportion of Galois GG-extensions of Q\mathbb{Q} with genus number tt, ordered by absolute discriminant, tends to zero.

This proves Conjecture 1.1 specialized to finite nilpotent groups. The proofs of these results are given in § 4.

Our next result concerns moments of genus numbers. For a real number r≥0r \ge0, define

MG,r(X):=∑L/Q GaloisGal⁡(L/Q)≃G∣Disc⁡(L)∣≤XgLr.M_{G,r}(X) := \sum_{\substack{L/\mathbb{Q}\ \mathrm{Galois}\\ \operatorname{Gal}(L/\mathbb{Q})\simeq G\\ |\operatorname{Disc}(L)|\le X}} g_L^r.

where the sum runs over Q\mathbb{Q}-isomorphism classes of extensions. Let I(G)I(G) denote the set of elements of order ℓ\ell in GG. If every element in I(G)I(G) is central, then b(G,Q)=∣I(G)∣/(ℓ−1)b(G,\mathbb{Q}) = |I(G)|/(\ell- 1). Under this assumption, we determine the order of growth of MG,r(X)M_{G,r}(X).

Theorem 1.4. Suppose that every element of order ℓ\ell in GG is central. For every fixed real r≥0r \ge0,

MG,r(X)≍G,rXa(G)(log⁡X)ℓrb(G,Q)−1.(2)M_{G,r}(X) \asymp_{G,r} X^{a(G)}(\log X)^{\ell r b(G,\mathbb{Q})-1}. \tag*{(2)}

Note that MG,0(X)=NG(X)M_{G,0}(X) = N_G(X). Applying the theorem with r=0r = 0 and dividing the estimates, we obtain for any $r \ge 0,

MG,r(X)NG(X)≍G,r(log⁡X)(ℓr−1)b(G,Q).\frac{M_{G,r}(X)}{N_G(X)} \asymp_{G,r} (\log X)^{(\ell r-1)b(G,\mathbb{Q})}.

In particular, the average genus number has order (log⁡X)(ℓ−1)b(G,Q)(\log X)^{(\ell-1)b(G,\mathbb{Q})}. For illustration, consider G=Z/ℓZG = \mathbb{Z}/\ell\mathbb{Z}. In this case, dℓ(G)=b(G,Q)=1d_\ell(G) = b(G,\mathbb{Q}) = 1, and hence for every fixed integer k≥0k \ge0,

NZ/ℓZ,ℓk(X)≪ℓ,kX1/(ℓ−1)log⁡X(1+log⁡log⁡X)k,N_{\mathbb{Z}/\ell\mathbb{Z},\ell^k}(X) \ll_{\ell,k} \frac{X^{1/(\ell-1)}}{\log X}(1+\log\log X)^k,

and

MZ/ℓZ,r(X)NZ/ℓZ(X)≍ℓ,r(log⁡X)ℓr−1.\frac{M_{\mathbb{Z}/\ell\mathbb{Z},r}(X)}{N_{\mathbb{Z}/\ell\mathbb{Z}}(X)} \asymp_{\ell,r} (\log X)^{\ell r-1}.

The quaternion group Q8Q_{8} provides a nonabelian example to which the moment theorem applies. On the other hand, the dihedral group D4D_{4} of order eight satisfies the hypothesis of Theorem 1.2, but not the centrality assumption in Theorem 1.4.

Methodology

The key ingredient in our approach is the relation between the genus number and the tamely ramified primes whose inertia groups have order ℓ\ell. For each such prime, there is an associated cyclic extension of degree ℓ\ell inside a cyclotomic field. Adjoining this extension to LL introduces no new ramification at finite places, so it gives a subextension of the genus field. The cyclotomic subfields coming from distinct primes are linearly disjoint. Consequently, these primes force a growing power of ℓ\ell to divide the genus number. In particular, fixing the genus number bounds the number of such primes. They are also the ramified primes with the smallest possible contribution to the discriminant. We use Koymans-Pagano parameterization to count extensions subject to this restriction. The solvable tuples parametrize epimorphisms bijectively and each field corresponds to exactly ∣Aut⁡(G)∣|\operatorname{Aut}(G)| epimorphisms.

Organization

The paper is organized as follows. In § 2, we collect the preliminary results on discriminants and genus fields. We construct cyclotomic extensions inside the genus field and use their linear disjointness to obtain lower bounds for the genus number. § 3 recalls the Koymans-Pagano coordinates, first for groups of prime-power order and then for a general nilpotent group by combining its Sylow factors. We explain how these coordinates capture tame inertia and the discriminant. We then prove the divisibility statement which bounds the number of primes with inertia of order ℓ\ell when the genus number is fixed. The section concludes with the estimates for integers with a bounded number of prime divisors that are needed in the counting argument. In § 4, we prove Theorem 1.2 and Corollary 1.3. § 5 is devoted to the weighted counting results. We first bound the genus number in terms of ramification indices and obtain a weighted upper bound for every nilpotent group. The argument treats the finitely many exceptional primes separately and applies Selberg-Delange estimates to the remaining primes. A central-twisting construction then gives a matching lower bound under the centrality assumption. Finally, we deduce the orders of magnitude of the averages and higher moments of genus numbers.

Declarations

Conflict of interest: There is no conflict of interest to report.

Availability of data and materials: No data was generated or analyzed in obtaining the results in this article.

Funding: No funding was received during the preparation of this article.

Use of artificial intelligence: The authors used ChatGPT 6.1 Sol as an exploratory tool to study related papers and to assist in correcting mistakes in the article.

Preliminaries

Let a nontrivial finite group GG act on itself by left multiplication. For g∈Gg \in G, let ind⁡(g)\operatorname{ind}(g) denote the permutation index of gg, namely

ind⁡(g)=∣G∣−#{orbits of ⟨g⟩ on G}.(3)\operatorname{ind}(g)=|G|-\#\{\text{orbits of }\langle g\rangle\text{ on }G\}. \tag*{(3)}

Suppose that g∈Gg \in G has order mm. Then every orbit of the cyclic subgroup ⟨g⟩\langle g\rangle has cardinality exactly mm. Consequently, the regular action of gg decomposes GG into exactly ∣G∣/m|G|/m cycles and it follows that

ind⁡(g)=∣G∣−∣G∣m=∣G∣(1−1m).(4)\operatorname{ind}(g)=|G|-\frac{|G|}{m}=|G|\left(1-\frac{1}{m}\right). \tag*{(4)}

Let ℓ\ell denote the smallest prime divisor of ∣G∣|G|. Every nonidentity element of GG has order at least ℓ\ell, and Cauchy’s theorem guarantees the existence of an element of order ℓ\ell. Consequently, we find that

min⁡1≠g∈Gind⁡(g)=∣G∣(1−1ℓ).(5)\min_{1\ne g\in G}\operatorname{ind}(g)=|G|\left(1-\frac{1}{\ell}\right). \tag*{(5)}

Definition 2.1. For g≠1g\ne1 set

eg:=ind⁡(g)=∣G∣(1−1ord⁡(g)).e_g:=\operatorname{ind}(g)=|G|\left(1-\frac{1}{\operatorname{ord}(g)}\right).

Also set

e0:=∣G∣(1−1ℓ)anda(G):=e0−1=ℓ(ℓ−1)∣G∣.e_0:=|G|\left(1-\frac{1}{\ell}\right)\qquad\text{and}\qquad a(G):=e_0^{-1}=\frac{\ell}{(\ell-1)|G|}.

We note that if ord⁡(g)>ℓ\operatorname{ord}(g)>\ell then it follows that eg>e0e_g>e_0, and consequently we have that a(G)eg>1a(G)e_g>1.

Lemma 2.2. Let L/KL/K be a finite Galois extension of number fields of degree nn. Let qq be a finite prime of KK which is tamely ramified in LL, and let ee be the common ramification index of the primes of LL above qq. Then

vq(DL/K)=n(1−1e),(6)v_q(\mathfrak{D}_{L/K})=n\left(1-\frac{1}{e}\right), \tag*{(6)}

where DL/K\mathfrak{D}_{L/K} is the relative discriminant ideal.

Proof. By [15 Chapter IV, §1, Proposition 4], the different exponent at every prime w∣qw\mid q is e−1e-1. Writing fwf_w for the residue degree at ww and taking the norm of the different, we obtain

vq(DL/K)=∑w∣qfw(e−1)=n(1−1e),v_q(\mathfrak{D}_{L/K})=\sum_{w\mid q}f_w(e-1)=n\left(1-\frac{1}{e}\right),

since ∑w∣qefw=n\sum_{w\mid q}ef_w=n. □

In particular, if L/QL/\mathbb{Q} has Galois group GG and g∈Gg\in G generates the tame inertia group at a rational prime qq, then eq(L/Q)=ord⁡(g)e_q(L/\mathbb{Q})=\operatorname{ord}(g). The exponent of qq in the discriminant is therefore

∣G∣(1−1ord⁡(g))=eg.|G|\left(1-\frac{1}{\operatorname{ord}(g)}\right)=e_g.

Fix an algebraic closure Q‾\overline{\mathbb{Q}} and write GQ=Gal⁡(Q‾/Q)G_{\mathbb{Q}}=\operatorname{Gal}(\overline{\mathbb{Q}}/\mathbb{Q}). For a finite group GG, let Epi⁡cont(GQ,G)\operatorname{Epi}_{\mathrm{cont}}(G_{\mathbb{Q}},G) denote the continuous group epimorphisms GQ↠GG_{\mathbb{Q}}\twoheadrightarrow G.

Lemma 2.3. Let F\mathcal{F} be any finite set of Galois extensions L/QL/\mathbb{Q} with Gal⁡(L/Q)≃G\operatorname{Gal}(L/\mathbb{Q})\simeq G, where membership in F\mathcal{F} depends only on the field LL and not on a chosen isomorphism Gal⁡(L/Q)≃G\operatorname{Gal}(L/\mathbb{Q})\simeq G. Then the number of epimorphisms ψ:GQ↠G\psi:G_{\mathbb{Q}}\twoheadrightarrow G whose kernel has fixed field belonging to F\mathcal{F} is ∣Aut⁡(G)∣∣F∣|\operatorname{Aut}(G)||\mathcal{F}|.

Proof. For a fixed field LL, the epimorphisms in question correspond exactly to isomorphisms Gal⁡(L/Q)≃G\operatorname{Gal}(L/\mathbb{Q})\simeq G. There are ∣Aut⁡(G)∣|\operatorname{Aut}(G)| such isomorphisms. Summing over L∈FL\in\mathcal{F} proves the assertion. □

Definition 2.4. Let K/QK/\mathbb{Q} be a number field. The absolute genus field K∗K^* is the maximal extension of KK satisfying:

(i) K∗/KK^*/K is unramified at every finite place of KK;

(ii) there exists a finite abelian extension F/QF/\mathbb{Q} such that K∗=KFK^*=KF.

The genus number is gK=[K∗:K]g_K=[K^*:K].

We allow ramification at the infinite places and this convention for the absolute genus field is used throughout. Every extension KF/KKF/K satisfying (i) and (ii) is abelian over KK: restriction identifies its Galois group with a subgroup of Gal⁡(F/Q)\operatorname{Gal}(F/\mathbb{Q}). It therefore lies in the narrow Hilbert class field of KK. Such extensions are closed under compositum, and their compositum is finite because the narrow Hilbert class field is finite. This gives a unique largest extension of the required form. In particular, K∗/KK^{*}/K is finite.

We also recall Abhyankar’s lemma (see [13 Chapter 5, Lemma 3]).

Lemma 2.5. Let FF be a nonarchimedean local field with residue characteristic pp. Let M/FM/F and N/FN/F be finite extensions inside a common algebraic closure of FF. Assume that M/FM/F is tamely ramified and that e(M/F)∣e(N/F)e(M/F) \mid e(N/F). Then MN/NMN/N is unramified.

Let qq be a rational prime and let d∣q−1d \mid q-1. Since

Gal⁡(Q(ζq)/Q)≃(Z/qZ)×\operatorname{Gal}(\mathbb{Q}(\zeta_q)/\mathbb{Q}) \simeq(\mathbb{Z}/q\mathbb{Z})^{\times}

is cyclic of order $q-1, there is a unique subfield Eq,d⊆Q(ζq)E_{q,d} \subseteq\mathbb{Q}(\zeta_q) with [Eq,d:Q]=d[E_{q,d}:\mathbb{Q}]=d. The extension Eq,d/QE_{q,d}/\mathbb{Q} is cyclic, ramified only at qq among finite primes, and totally tamely ramified at qq with ramification index dd.

Now let K/QK/\mathbb{Q} be Galois and suppose that qq is tamely ramified in KK. Let eq=eq(K/Q)e_q=e_q(K/\mathbb{Q}) be its ramification index and set dq:=gcd⁡(eq,q−1)d_q:=\gcd(e_q,q-1) and Eq:=Eq,dqE_q:=E_{q,d_q}.

Proposition 2.6. With the notation above, KEq/KKE_q/K is unramified at every finite place. Hence KEq⊆K∗KE_q \subseteq K^{*} and

[KEq:K]=[Eq:Eq∩K]∣gK.[KE_q:K]=[E_q:E_q\cap K]\mid g_K.

In particular,

gK≥[Eq:Eq∩K].g_K\geq[E_q:E_q\cap K].

Proof. It is clear that KEq/KKE_q/K is unramified at every prime of KK that does not lie above qq. Let vv be the unique prime of EqE_q above qq. For each prime w∣qw\mid q of KK, place the completions (Eq)v(E_q)_v and KwK_w inside a common algebraic closure of Qq\mathbb{Q}_q. The extension (Eq)v/Qq(E_q)_v/\mathbb{Q}_q is tamely ramified of ramification index dqd_q, and by definition

dq∣eq=e(Kw/Qq).d_q\mid e_q=e(K_w/\mathbb{Q}_q).

Lemma 2.5 therefore shows that Kw(Eq)v/KwK_w(E_q)_v/K_w is unramified. Hence KEq/KKE_q/K is unramified at every prime above qq as well. Thus KEq⊆K∗KE_q\subseteq K^{*}, and the asserted divisibility follows from the tower law. □\square

Lemma 2.7. Let q1,…,qrq_1,\ldots,q_r be distinct rational primes, and for each ii let Ei/QE_i/\mathbb{Q} be a nontrivial subextension of Q(ζqi)/Q\mathbb{Q}(\zeta_{q_i})/\mathbb{Q}. Then the fields E1,…,ErE_1,\ldots,E_r are linearly disjoint over Q\mathbb{Q}. In particular,

[E1⋯Er:Q]=∏i=1r[Ei:Q].[E_1\cdots E_r:\mathbb{Q}]=\prod_{i=1}^{r}[E_i:\mathbb{Q}].

Proof. We argue inductively. Set F=E1⋯Er−1F=E_1\cdots E_{r-1}. The extension F/QF/\mathbb{Q} is unramified at qrq_r, since each EiE_i with i<ri<r is ramified only at qiq_i. On the other hand, Er/QE_r/\mathbb{Q} is totally ramified at qrq_r. Let

D=Er∩F.D=E_r\cap F.

As a subfield of FF, the extension D/QD/\mathbb{Q} is unramified at qrq_r. As a subfield of the cyclic totally ramified extension Er/QE_r/\mathbb{Q}, it is totally ramified at qrq_r: the inertia group of Er/QE_r/\mathbb{Q} is the whole Galois group, and its image in every quotient is again the whole quotient. The only extension which

is simultaneously unramified and totally ramified at qrq_r is the trivial extension. Thus D=QD = \mathbb{Q}. Hence ErE_r and FF are linearly disjoint, and induction completes the proof. □

We write Gab:=G/[G,G]G^{\mathrm{ab}} := G/[G,G] for the abelianization of GG.

Proposition 2.8. Let K/QK/\mathbb{Q} be a finite Galois extension with group GG, and let TT be any finite set of rational primes tamely ramified in KK. Setting ET:=∏q∈TEqE_T := \prod_{q\in T} E_q, where the empty compositum is understood to be Q\mathbb{Q}, we have that KET/KKE_T/K is unramified at every finite place and

[KET:K]∣gK.[KE_T:K]\mid g_K.

Moreover,

gK≥[KET:K]=∏q∈Tdq[ET∩K:Q]≥1∣Gab∣∏q∈Tdq.(7)g_K \ge[KE_T:K] = \frac{\prod_{q\in T}d_q}{[E_T\cap K:\mathbb{Q}]} \ge\frac{1}{|G^{\mathrm{ab}}|}\prod_{q\in T}d_q. \tag*{(7)}

Proof. By Proposition 2.6, each KEq/KKE_q/K is unramified and therefore KET⊆K∗KE_T\subseteq K^*. Consequently, [KET:K]∣gK[KE_T:K]\mid g_K. After omitting the trivial factors Eq=QE_q=\mathbb{Q}, Lemma 2.7 implies that

[ET:Q]=∏q∈Tdq.[E_T:\mathbb{Q}]=\prod_{q\in T}d_q.

Therefore

[KET:K]=[ET:ET∩K]=∏q∈Tdq[ET∩K:Q].[KE_T:K]=[E_T:E_T\cap K]=\frac{\prod_{q\in T}d_q}{[E_T\cap K:\mathbb{Q}]}.

Finally ET∩KE_T\cap K is Galois and abelian over Q\mathbb{Q}, because it is the intersection of two Galois extensions and is contained in the abelian extension ET/QE_T/\mathbb{Q}. Hence Gal⁡(ET∩K/Q)\operatorname{Gal}(E_T\cap K/\mathbb{Q}) is an abelian quotient of GG, so

[ET∩K:Q]≤∣Gab∣.[E_T\cap K:\mathbb{Q}]\le|G^{\mathrm{ab}}|.

This proves (2.5). □

Definition 2.9. For a prime pp and a finite group GG, define

dp(G):=dim⁡FpHom⁡(G,Fp),d_p(G):=\dim_{\mathbb{F}_p}\operatorname{Hom}(G,\mathbb{F}_p),

where Fp\mathbb{F}_p is viewed as an additive group.

Note that pdp(G)p^{d_p(G)} is the order of the maximal elementary abelian pp-quotient of GG. This quotient is G/([G,G]Gp)G/([G,G]G^p), where Gp:=⟨gp:g∈G⟩G^p:=\langle g^p:g\in G\rangle. An elementary abelian pp-extension is a finite Galois extension with Galois group isomorphic to (Z/pZ)r(\mathbb{Z}/p\mathbb{Z})^r for some r≥0r\ge0.

Lemma 2.10. Let K/QK/\mathbb{Q} be Galois with group GG, and let E/QE/\mathbb{Q} be an elementary abelian pp-extension. Then

[E∩K:Q]≤pdp(G).[E\cap K:\mathbb{Q}]\le p^{d_p(G)}.

Proof. The field D=E∩KD=E\cap K is Galois over Q\mathbb{Q}. Since D⊆ED\subseteq E, its Galois group is an elementary abelian pp-group. Since D⊆KD\subseteq K, restriction identifies Gal⁡(D/Q)\operatorname{Gal}(D/\mathbb{Q}) with a quotient of GG. If [D:Q]=pr[D:\mathbb{Q}]=p^r, then GG therefore has a quotient (Z/pZ)r(\mathbb{Z}/p\mathbb{Z})^r, so r≤dp(G)r\le d_p(G) by Definition 2.9. This proves the claim. □

Proposition 2.11. Let K/QK/\mathbb{Q} be Galois with group GG. Let TT be a finite set of tamely ramified primes such that eq(K/Q)=ℓe_q(K/\mathbb{Q})=\ell and q≡1(modℓ)q\equiv1\pmod{\ell} for every q∈Tq\in T. Then

νℓ(gK)≥max⁡{0,∣T∣−dℓ(G)}.(8)\nu_\ell(g_K)\ge\max\{0,|T|-d_\ell(G)\}. \tag*{(8)}

Equivalently,

ℓmax⁡{0,∣T∣−dℓ(G)}∣gK.\ell^{\max\{0,|T|-d_\ell(G)\}}\mid g_K.

In particular,

gK≥ℓ∣T∣−dℓ(G).g_K \ge\ell^{|T|-d_\ell(G)}.

Proof. For q∈Tq \in T, one has

dq=gcd⁡(ℓ,q−1)=ℓ.d_q = \gcd(\ell,q-1)=\ell.

Thus Eq/QE_q/\mathbb{Q} is cyclic of degree ℓ\ell. By Lemma 2.7,

Gal⁡(ET/Q)≃(Z/ℓZ)∣T∣,\operatorname{Gal}(E_T/\mathbb{Q}) \simeq(\mathbb{Z}/\ell\mathbb{Z})^{|T|},

and consequently [ET:Q]=ℓ∣T∣[E_T:\mathbb{Q}]=\ell^{|T|}. Write [ET∩K:Q]=ℓh[E_T\cap K:\mathbb{Q}]=\ell^h. By Lemma 2.10, we have h≤dℓ(G)h\le d_\ell(G). Proposition 2.8 gives

[KET:K]=[ET:ET∩K]=ℓ∣T∣−h∣gK.[KE_T:K]=[E_T:E_T\cap K]=\ell^{|T|-h}\mid g_K.

It follows that

νℓ(gK)≥∣T∣−h≥∣T∣−dℓ(G).\nu_\ell(g_K)\ge|T|-h\ge|T|-d_\ell(G).

Combining this with νℓ(gK)≥0\nu_\ell(g_K)\ge0 proves (8). □

The congruence condition in Proposition 2.11 follows automatically from the other hypotheses, since ℓ\ell is the smallest prime divisor of ∣G∣|G|. Indeed, choose a prime of KK above qq, and denote its decomposition and inertia groups by DqD_q and IqI_q. Since Iq≃Z/ℓZI_q\simeq\mathbb{Z}/\ell\mathbb{Z}, the image of the conjugation action

Dq⟶Aut⁡(Iq)D_q \longrightarrow\operatorname{Aut}(I_q)

has order dividing both ∣G∣|G| and ℓ−1\ell-1. These integers are coprime, so this action is trivial. On the other hand, the tame Frobenius relation gives στσ−1=τq\sigma\tau\sigma^{-1}=\tau^q, where τ\tau generates IqI_q and σ\sigma lifts arithmetic Frobenius. Thus τq−1=1\tau^{q-1}=1, which gives q≡1(modℓ)q\equiv1\pmod{\ell}.

Koymans–Pagano coordinates and the genus–inertia dictionary

We recall the Koymans–Pagano parametrization [8 Section 2]. This will be needed in the proof of our fixed-genus counting results as well as our results on moments. The parameterization describes nilpotent extensions in terms of squarefree ideal coordinates which record the image of tame inertia. Throughout, we work over Q\mathbb{Q}. Let GG be a nontrivial finite nilpotent group. Write G=∏j=1cPjG=\prod_{j=1}^{c}P_j, where Pj=G(ℓj)P_j=G(\ell_j) is the Sylow ℓj\ell_j-subgroup and the ℓj\ell_j are the distinct prime divisors of ∣G∣|G|. We order these primes so that ℓ=ℓ1=min⁡{p:p∣∣G∣}\ell=\ell_1=\min\{p:p\mid|G|\}. For each jj, let GQpro−ℓjG_{\mathbb{Q}}^{\mathrm{pro}-\ell_j} denote the maximal pro-ℓj\ell_j quotient of GQG_{\mathbb{Q}}.

By [8 Section 2], there is a finite set of rational primes SGS_G, containing every prime divisor of ∣G∣|G|, for which the following choices can be made. Set

Ω~Q(ℓj):={q∉SG:q≡1(modℓj)}.\widetilde{\Omega}_{\mathbb{Q}}(\ell_j):=\{q\notin S_G:q\equiv1\pmod{\ell_j}\}.

For every q∈Ω~Q(ℓj)q\in\widetilde{\Omega}_{\mathbb{Q}}(\ell_j), fix a character

χj,q:GQ⟶Fℓj\chi_{j,q}:G_{\mathbb{Q}}\longrightarrow\mathbb{F}_{\ell_j}

which is ramified at qq and unramified at every finite prime outside SG∪{q}S_G\cup\{q\}. Also fix an Fℓj\mathbb{F}_{\ell_j}-basis χj,1,…,χj,tj\chi_{j,1},\ldots,\chi_{j,t_j} of the finite-dimensional space of characters unramified at every finite prime outside SGS_G, so that

{χj,1,…,χj,tj}∪{χj,q:q∈Ω~Q(ℓj)}\{\chi_{j,1},\ldots,\chi_{j,t_j}\}\cup\{\chi_{j,q}:q\in\widetilde{\Omega}_{\mathbb{Q}}(\ell_j)\}

is a basis of H1(GQ,Fℓj)H^1(G_{\mathbb{Q}},\mathbb{F}_{\ell_j}). Here Fℓj\mathbb{F}_{\ell_j} is regarded as a trivial Galois module, so these cohomology classes are continuous homomorphisms to its additive group. When j=1j=1, we write χq:=χ1,q\chi_q:=\chi_{1,q}.

For q∉SGq \notin S_G, let Iq(ℓj)I_q(\ell_j) denote the image of inertia in GQpro−ℓjG_{\mathbb{Q}}^{\mathrm{pro}-\ell_j}, after fixing compatible local embeddings. If q∈Ω~Q(ℓj)q \in\widetilde{\Omega}_{\mathbb{Q}}(\ell_j), choose a topological generator σj,q\sigma_{j,q} of Iq(ℓj)I_q(\ell_j) with χj,q(σj,q)=1\chi_{j,q}(\sigma_{j,q})=1. Otherwise Iq(ℓj)I_q(\ell_j) is trivial, and we set σj,q=1\sigma_{j,q}=1. The latter assertion concerns the image of inertia in the pro-ℓj\ell_j Galois quotient. The Frobenius relation with a tame inertia generator forces this image to be trivial when q≢1(modℓj)q \not\equiv1 \pmod{\ell_j}.

Let Sj′S'_j denote the set of squarefree positive integers supported entirely on Ω~Q(ℓj)\widetilde{\Omega}_{\mathbb{Q}}(\ell_j), and set

Sj:={0,1}[tj]×Sj′S_j := \{0,1\}^{[t_j]} \times S'_j

where [tj]:={1,…,tj}[t_j]:=\{1,\ldots,t_j\}.

Two elements w=(T,b)w=(T,b) and w′=(T′,b′)w'=(T',b') are called coprime if bb and b′b' are relatively prime and there is no i∈[tj]i\in[t_j] for which Ti=Ti′=1T_i=T'_i=1. We write Prim⁡(SjPj∖{1})\operatorname{Prim}(S_j^{P_j\setminus\{1\}}) for the set of tuples (wh,j)h∈Pj∖{1}(w_{h,j})_{h\in P_j\setminus\{1\}} whose coordinates are pairwise coprime.

Projection onto the Sylow factors gives rise to the following identification:

Epi⁡cont(GQ,G)≃∏j=1cEpi⁡cont(GQpro−ℓj,Pj);\operatorname{Epi}_{\mathrm{cont}}(G_{\mathbb{Q}},G)\simeq\prod_{j=1}^{c}\operatorname{Epi}_{\mathrm{cont}}(G_{\mathbb{Q}}^{\mathrm{pro}-\ell_j},P_j);

see [8 Proposition 2.11]. We use the common exceptional set SG∪{∞}S_G\cup\{\infty\} in every Sylow construction. Enlarging the original exceptional sets only enlarges the finite exceptional character spaces.

Koymans–Pagano bundle the Sylow coordinates into a tuple (vg(1),vg(2))g∈G∖{1}(v_g(1),v_g(2))_{g\in G\setminus\{1\}} as follows. Set w1,j:=(0,1)w_{1,j}:=(0,1). For q∉SGq\notin S_G, let gj(q)=hg_j(q)=h if q∣wh,j(2)q\mid w_{h,j}(2), and let gj(q)=1g_j(q)=1 if qq divides none of the wh,j(2)w_{h,j}(2). For g=(g1,…,gc)≠1g=(g_1,\ldots,g_c)\ne1, set

vg(1):=(wg1,1(1),…,wgc,c(1))andvg(2):=∏q∉SG(g1(q),…,gc(q))=gq.v_g(1):=(w_{g_1,1}(1),\ldots,w_{g_c,c}(1))\quad\text{and}\quad v_g(2):=\prod_{\substack{q\notin S_G\\(g_1(q),\ldots,g_c(q))=g}}q.

The vg(2)v_g(2) are squarefree, pairwise coprime, and supported outside SGS_G. The first components take values in fixed finite sets. Let S′S' be the set of squarefree positive integers supported outside SGS_G, and set

S:={0,1}[t1+⋯+tc]×S′.S:=\{0,1\}^{[t_1+\cdots+t_c]}\times S'.

We denote the image of the product of the Sylow primitive spaces under this bundling by Prim⁡(SG∖{1})\operatorname{Prim}(S^{G\setminus\{1\}}). Fix the construction choices in [8 Section 2]. The solvable primitive tuples, denoted by Prim⁡(SG∖{1})solv\operatorname{Prim}(S^{G\setminus\{1\}})^{\mathrm{solv}} are those that give rise to epimorphisms onto GG.

Theorem 3.1 (Koymans–Pagano parametrization). There is a bijection

PG:Prim⁡(SG∖{1})solv⟶∼Epi⁡cont(GQ,G).(9)P_G:\operatorname{Prim}(S^{G\setminus\{1\}})^{\mathrm{solv}}\overset{\sim}{\longrightarrow}\operatorname{Epi}_{\mathrm{cont}}(G_{\mathbb{Q}},G). \tag*{(9)}

Proof. This follows from [8 Propositions 2.8 and 2.11] and the bundling described above. □

There is also a restriction on which primes may occur in a given coordinate. If q∣vg(2)q\mid v_g(2) and ℓj∣ord⁡(g)\ell_j\mid\operatorname{ord}(g), then the ℓj\ell_j-part of inertia is nontrivial. Hence q∈Ω~Q(ℓj)q\in\widetilde{\Omega}_{\mathbb{Q}}(\ell_j) and therefore q≡1(modℓj)q\equiv1\pmod{\ell_j}. In particular, whenever ord⁡(g)=ℓ\operatorname{ord}(g)=\ell and q∣vg(2)q\mid v_g(2), one has q≡1(modℓ)q\equiv1\pmod{\ell}.

Proposition 3.2. Let vv be a solvable bundled tuple corresponding to a GG-extension L/QL/\mathbb{Q}, and let q∉SGq\notin S_G. If q∣vg(2)q\mid v_g(2), then a tame inertia generator at qq maps to gg. Consequently, the ramification index satisfies eq(L/Q)=ord⁡(g)e_q(L/\mathbb{Q})=\operatorname{ord}(g). If qq divides none of the vg(2)v_g(2), then qq is unramified in LL.

Proof. This is [8 Proposition 2.12]. □

For a positive integer dd and a finite set SS of rational primes, write

free⁡S(d):=∏q∉Sqvq(d)\operatorname{free}_{S}(d) := \prod_{q\notin S} q^{v_q(d)}

for the largest divisor of dd supported outside SS.

Proposition 3.3. For a solvable bundled tuple corresponding to L/QL/\mathbb{Q},

free⁡SG∣Disc⁡(L)∣=∏g∈G∖{1}vg(2)eg,(10)\operatorname{free}_{S_G}|\operatorname{Disc}(L)| = \prod_{g\in G\setminus\{1\}} v_g(2)^{e_g}, \tag*{(10)}

where we recall that eg=∣G∣(1−1ord⁡(g))e_g = |G|\left(1-\frac{1}{\operatorname{ord}(g)}\right). Consequently, if ∣Disc⁡(L)∣≤X|\operatorname{Disc}(L)| \le X, then

∏g≠1vg(2)eg≤X.(11)\prod_{g\ne1} v_g(2)^{e_g} \le X. \tag*{(11)}

Proof. The result follows from [8 Proposition 2.13]. □

Now let

I(G):={g∈G∖{1}:ord⁡(g)=ℓ}.I(G) := \{g\in G\setminus\{1\}:\operatorname{ord}(g)=\ell\}.

By § 2, the elements of I(G)I(G) are exactly the nonidentity elements of minimal index, and

eh=e0=a(G)−1(h∈I(G)),a(G)eg>1(ord⁡(g)>ℓ).e_h=e_0=a(G)^{-1}\qquad(h\in I(G)),\qquad a(G)e_g>1\qquad(\operatorname{ord}(g)>\ell).

Proposition 3.4. Let L/QL/\mathbb{Q} be a nilpotent Galois extension with group GG, represented by a solvable bundled Koymans–Pagano tuple. Let q∉SGq\notin S_G and suppose that q∣vg(2)q\mid v_g(2). Then

(i) the tame inertia group at qq has order ord⁡(g)\operatorname{ord}(g);

(ii) the exponent of qq in Disc⁡(L)\operatorname{Disc}(L) is

∣G∣(1−ord⁡(g)−1);|G|\left(1-\operatorname{ord}(g)^{-1}\right);

(iii) the degree of the tame cyclotomic extension associated to qq in Ishida’s construction [5] is

gcd⁡(ord⁡(g),q−1).\gcd(\operatorname{ord}(g),q-1).

In particular, if g∈I(G)g\in I(G), then this degree is ℓ\ell.

Proof. Parts (i) and (ii) follow from Propositions 3.2 and 3.3. For a tamely ramified prime qq, the cyclotomic construction preceding Proposition 2.6 associates the unique subfield of Q(ζq)\mathbb{Q}(\zeta_q) of degree gcd⁡(eq(L/Q),q−1)\gcd(e_q(L/\mathbb{Q}),q-1). By (i), one has eq(L/Q)=ord⁡(g)e_q(L/\mathbb{Q})=\operatorname{ord}(g), which proves (iii). Finally, if g∈I(G)g\in I(G), then ord⁡(g)=ℓ\operatorname{ord}(g)=\ell. Moreover, q∣vg(2)q\mid v_g(2) implies that qq belongs to the clean set supporting the ℓ\ell-coordinates, and hence q≡1(modℓ)q\equiv1\pmod{\ell}. Therefore

gcd⁡(ord⁡(g),q−1)=gcd⁡(ℓ,q−1)=ℓ.\gcd(\operatorname{ord}(g),q-1)=\gcd(\ell,q-1)=\ell.

□

The number of primes appearing in the minimal coordinates admits an intrinsic description. For a solvable tuple vv, set M(v):=∏h∈I(G)vh(2)M(v):=\prod_{h\in I(G)}v_h(2). Since the integers vh(2)v_h(2) are squarefree and pairwise coprime, M(v)M(v) is squarefree and ω(M(v))=∑h∈I(G)ω(vh(2))\omega(M(v))=\sum_{h\in I(G)}\omega(v_h(2)), where ω(n)\omega(n) denotes the number of distinct prime divisors of a squarefree integer n. By Proposition 3.2, this quantity depends only on the extension and not on the chosen coordinates. We therefore define

ωmin⁡,SG(L):=#{q∉SG:q ramifies in L, eq(L/Q)=ℓ}.\omega_{\min,S_G}(L):=\#\{q\notin S_G:q\text{ ramifies in }L,\ e_q(L/\mathbb{Q})=\ell\}.

Thus ωmin⁡,SG(L)=ω(M(v))\omega_{\min,S_G}(L)=\omega(M(v)). In words, ωmin⁡,SG(L)\omega_{\min,S_G}(L) counts the clean ramified primes whose inertia group has the smallest possible nontrivial order. These are exactly the primes which occur in the minimal-index coordinates.

Theorem 3.5. Let L/QL/\mathbb{Q} be a nilpotent Galois extension with group GG, and set r:=ωmin⁡,SG(L)r := \omega_{\min,S_G}(L). Then

νℓ(gL)≥max⁡{0,r−dℓ(G)}.(12)\nu_{\ell}(g_L) \ge\max\{0,r-d_{\ell}(G)\}. \tag*{(12)}

In particular, if r>dℓ(G)r>d_{\ell}(G), then ℓr−dℓ(G)∣gL\ell^{r-d_{\ell}(G)} \mid g_L, and hence gL≥ℓr−dℓ(G)g_L \ge\ell^{r-d_{\ell}(G)}.

Proof. Let TT denote the set of primes counted by ωmin⁡,SG(L)\omega_{\min,S_G}(L), so that ∣T∣=r|T|=r. These primes are tamely ramified in LL, and Proposition 3.4 gives

eq(L/Q)=ℓandq≡1(modℓ)(q∈T).e_q(L/\mathbb{Q})=\ell\quad\text{and}\quad q\equiv1\pmod{\ell}\quad(q\in T).

Proposition 2.11 therefore yields

νℓ(gL)≥max⁡{0,r−dℓ(G)}.\nu_{\ell}(g_L) \ge\max\{0,r-d_{\ell}(G)\}.

The remaining assertions follow immediately. □

Thus fixing the genus number imposes a uniform bound on the number of prime divisors which may occur among all the minimal coordinates.

Corollary 3.6. For t≥1t\ge1, set RG(t):=dℓ(G)+νℓ(t)R_G(t):=d_{\ell}(G)+\nu_{\ell}(t). If gL=tg_L=t, then ωmin⁡,SG(L)≤RG(t)\omega_{\min,S_G}(L)\le R_G(t). Equivalently, every Koymans–Pagano tuple representing LL satisfies

ω(∏h∈I(G)νh(2))≤RG(t).\omega\left(\prod_{h\in I(G)} \nu_h(2)\right)\le R_G(t).

Proof. Set r=ωmin⁡,SG(L)r=\omega_{\min,S_G}(L). By Theorem 3.5, max⁡{0,r−dℓ(G)}≤νℓ(gL)=νℓ(t)\max\{0,r-d_{\ell}(G)\}\le\nu_{\ell}(g_L)=\nu_{\ell}(t). Thus r≤dℓ(G)+νℓ(t)=RG(t)r\le d_{\ell}(G)+\nu_{\ell}(t)=R_G(t). □

Notice that RG(t)R_G(t) is always positive. Indeed, the Sylow ℓ\ell-subgroup of GG is nontrivial and, as every nontrivial finite ℓ\ell-group has a quotient of order ℓ\ell, projection to this Sylow factor followed by such a quotient gives a nonzero homomorphism G→FℓG\to\mathbb{F}_{\ell}. Thus dℓ(G)≥1d_{\ell}(G)\ge1, and consequently RG(t)≥1R_G(t)\ge1 for every t≥1t\ge1.

For an integer r≥0r\ge0 and Y≥1Y\ge1, let Ar(Y):=#{n≤Y:μ2(n)=1, ω(n)=r}A_r(Y):=\#\{n\le Y:\mu^2(n)=1,\ \omega(n)=r\}. Here μ\mu is the Möbius function, so the condition μ2(n)=1\mu^2(n)=1 is equivalent to nn being squarefree, while ω(n)\omega(n) denotes the number of distinct prime divisors of nn. For our application, we need the following upper bound.

Lemma 3.7. Fix integers M≥1M\ge1 and R≥1R\ge1, and define

BM,R(Y):=∑n≤Yμ2(n)=1ω(n)≤RMω(n).B_{M,R}(Y):=\sum_{\substack{n\le Y\\ \mu^2(n)=1\\ \omega(n)\le R}}M^{\omega(n)}.

Then

BM,R(Y)≪M,RYlog⁡Y(1+log⁡log⁡Y)R−1(13)B_{M,R}(Y)\ll_{M,R}\frac{Y}{\log Y}(1+\log\log Y)^{R-1} \tag*{(13)}

for Y≥3Y\ge3. For all Y≥1Y\ge1 one also has the bound BM,R(Y)≪M,RYB_{M,R}(Y)\ll_{M,R}Y.

Proof. We first prove that, for every fixed integer r≥1r\ge1,

Ar(Y)≪rYlog⁡Y(1+log⁡log⁡Y)r−1,(14)A_r(Y)\ll_r\frac{Y}{\log Y}(1+\log\log Y)^{r-1}, \tag*{(14)}

for Y≥3Y\ge3. For r=1r=1, the integers counted by A1(Y)A_1(Y) are precisely the primes, so this follows from Chebyshev’s estimate π(Y)≪Y/log⁡Y\pi(Y)\ll Y/\log Y. Suppose that r≥2r\ge2 and that the assertion holds

for r−1r-1. Every squarefree integer n≤Yn \le Y with exactly rr prime divisors has a smallest prime divisor p≤Y1/rp \le Y^{1/r}. Writing n=pmn=pm, the integer mm is squarefree and has exactly r−1r-1 prime divisors. Consequently,

Ar(Y)≤∑p≤Y1/rAr−1(Y/p).A_r(Y) \le\sum_{p \le Y^{1/r}} A_{r-1}(Y/p).

where the sum is over primes. This is an upper bound because the sum also allows integers mm divisible by pp or having a prime divisor smaller than pp. Assume first that Y≥9Y \ge9. For every prime in this sum,

Y/p≥Y1−1/r≥Y≥3and thuslog⁡(Y/p)≥12log⁡Y.Y/p \ge Y^{1-1/r} \ge\sqrt{Y} \ge3 \quad\text{and thus} \quad\log(Y/p) \ge\frac{1}{2}\log Y.

The induction hypothesis therefore gives

Ar(Y)≪rYlog⁡Y(1+log⁡log⁡Y)r−2∑p≤Y1/r1p.A_r(Y) \ll_r \frac{Y}{\log Y}(1+\log\log Y)^{r-2}\sum_{p \le Y^{1/r}}\frac{1}{p}.

Chebyshev’s estimate and partial summation imply

∑p≤Y1p=π(Y)Y+∫2Yπ(t)t2 dt≪1+log⁡log⁡Y.\sum_{p \le Y}\frac{1}{p}=\frac{\pi(Y)}{Y}+\int_2^Y\frac{\pi(t)}{t^2}\,\mathrm{d}t \ll1+\log\log Y.

Since the prime sum over p≤Y1/rp \le Y^{1/r} is no larger, we obtain the required bound for Ar(Y)A_r(Y). Enlarging the implied constant covers 3≤Y<93 \le Y < 9 and completes the induction.

We now group the terms defining BM,R(Y)B_{M,R}(Y) according to their number of prime divisors as

BM,R(Y)=∑r=0RMrAr(Y)=1+∑r=1RMrAr(Y).B_{M,R}(Y)=\sum_{r=0}^{R}M^r A_r(Y)=1+\sum_{r=1}^{R}M^r A_r(Y).

The initial term is 1 because the only positive integer with no prime divisors is n=1n=1. Applying the bound (14) gives

BM,R(Y)≪M,R1+Ylog⁡Y∑r=1RMr(1+log⁡log⁡Y)r−1.B_{M,R}(Y)\ll_{M,R}1+\frac{Y}{\log Y}\sum_{r=1}^{R}M^r(1+\log\log Y)^{r-1}.

For Y≥3Y \ge3, we have 1+log⁡log⁡Y≥11+\log\log Y \ge1. Hence

∑r=1RMr(1+log⁡log⁡Y)r−1≤(∑r=1RMr)(1+log⁡log⁡Y)R−1.\sum_{r=1}^{R}M^r(1+\log\log Y)^{r-1}\le\left(\sum_{r=1}^{R}M^r\right)(1+\log\log Y)^{R-1}.

The finite sum depends only on MM and RR. Moreover, Y/log⁡Y≥1Y/\log Y \ge1, so the initial constant term satisfies the same upper bound. This proves (13). Finally, for every integer occurring in the defining sum, ω(n)≤R\omega(n)\le R and therefore Mω(n)≤MRM^{\omega(n)}\le M^R. Since there are at most ⌊Y⌋\lfloor Y\rfloor positive integers up to YY,

BM,R(Y)≤MR⌊Y⌋≤MRY≪M,RY,B_{M,R}(Y)\le M^R\lfloor Y\rfloor\le M^R Y\ll_{M,R}Y,

for Y≥1Y \ge1.

In the applications in the next section we take M=∣I(G)∣M=|I(G)| and R=RG(t)R=R_G(t).

Counting nilpotent extensions of fixed genus

In this section, we prove Theorem 1.2. Recall that GG is a fixed non-trivial finite nilpotent group, ℓ\ell is the smallest prime divisor of ∣G∣|G|,

I(G)={g∈G:ord⁡(g)=ℓ}andH(G)=I(G)∪{1}.I(G)=\{g\in G:\operatorname{ord}(g)=\ell\}\quad\text{and}\quad H(G)=I(G)\cup\{1\}.

We also write

J(G):=G∖H(G)={g∈G:ord⁡(g)>ℓ}.J(G):=G\setminus H(G)=\{g\in G:\operatorname{ord}(g)>\ell\}.

For h∈I(G)h\in I(G) one has eh=e0=a(G)−1e_h=e_0=a(G)^{-1}, whereas a(G)eg>1a(G)e_g>1 for every g∈J(G)g\in J(G).

For a real number X≥1X\geq1 and a positive integer tt, set

EG,t(X):={ψ∈Epicont⁡(GQ,G):∣Disc⁡(Lψ)∣≤X, gLψ=t}.(15)\mathcal{E}_{G,t}(X):=\left\{\psi\in\operatorname{Epicont}(G_{\mathbb{Q}},G):|\operatorname{Disc}(L_\psi)|\leq X,\ g_{L_\psi}=t\right\}. \tag*{(15)}

where LψL_\psi denotes the fixed field of ker⁡(ψ)\operatorname{ker}(\psi).

Lemma 4.1. For every X≥1X\geq1 and every positive integer tt,

∣EG,t(X)∣=∣Aut⁡(G)∣NG,t(X).(16)|\mathcal{E}_{G,t}(X)|=|\operatorname{Aut}(G)|N_{G,t}(X). \tag*{(16)}

Proof. Apply Lemma 2.3 to the fields counted by NG,t(X)N_{G,t}(X).

Fix tt and set

R:=RG(t)=dℓ(G)+νℓ(t).(17)R:=R_G(t)=d_\ell(G)+\nu_\ell(t). \tag*{(17)}

By Corollary 3.6, every ψ∈EG,t(X)\psi\in\mathcal{E}_{G,t}(X) has

ωmin⁡,SG(Lψ)≤R.(18)\omega_{\min,S_G}(L_\psi)\leq R. \tag*{(18)}

We translate this restriction into the Koymans–Pagano coordinates introduced in the previous section.

Let v=(vg)g≠1v=(v_g)_{g\neq1} be the bundled primitive tuple corresponding to ψ\psi. Write vg=(vg(1),vg(2))v_g=(v_g(1),v_g(2)) as in § 3. Since the first coordinates vg(1)v_g(1) take values in fixed finite sets, the total number of possibilities for (vg(1))g≠1(v_g(1))_{g\neq1} is bounded in terms of GG only. They will therefore be absorbed into the implied constants below.

For the squarefree coordinates define

m(v):=∏h∈I(G)vh(2)andD(v):=∏g∈J(G)vg(2)eg.(19)m(v):=\prod_{h\in I(G)}v_h(2)\quad\text{and}\quad D(v):=\prod_{g\in J(G)}v_g(2)^{e_g}. \tag*{(19)}

The vg(2)v_g(2) are squarefree positive integers, and they are pairwise coprime. Hence m(v)m(v) is squarefree and

ω(m(v))=∑h∈I(G)ω(vh(2))=ωmin⁡,SG(Lψ)≤R.(20)\omega(m(v))=\sum_{h\in I(G)}\omega(v_h(2))=\omega_{\min,S_G}(L_\psi)\leq R. \tag*{(20)}

Moreover, Proposition 3.3 gives

m(v)e0D(v)=free⁡SG∣Disc⁡(Lψ)∣≤∣Disc⁡(Lψ)∣≤X.(21)m(v)^{e_0}D(v)=\operatorname{free}_{S_G}|\operatorname{Disc}(L_\psi)|\leq|\operatorname{Disc}(L_\psi)|\leq X. \tag*{(21)}

Since e0=a(G)−1e_0=a(G)^{-1}, it follows that

m(v)≤(XD(v))a(G).(22)m(v)\leq\left(\frac{X}{D(v)}\right)^{a(G)}. \tag*{(22)}

For an upper bound, we enlarge the parameter set. For g∈J(G)g\in J(G), replace the squarefree integer vg(2)v_g(2) by an arbitrary positive integer ngn_g, and discard all coprimality, congruence and solvability conditions involving these coordinates. Set

n=(ng)g∈J(G),andD(n):=∏g∈J(G)ngeg.(23)\mathbf{n}=(n_g)_{g\in J(G)},\quad\text{and}\quad D(\mathbf{n}):=\prod_{g\in J(G)}n_g^{e_g}. \tag*{(23)}

In our arguments, for minimal coordinates we retain only the information that their product mm is squarefree and that ω(m)≤R\omega(m) \le R. Once such an mm is fixed, each prime divisor of mm may be assigned to one of the ∣I(G)∣|I(G)| coordinates indexed by I(G)I(G). Thus the number of possible ordered minimal-coordinate tuples with product mm is at most ∣I(G)∣ω(m)|I(G)|^{\omega(m)}. Dropping the restrictions q≡1(modℓ)q \equiv1 \pmod{\ell} and the Koymans–Pagano solvability conditions can only increase the count.

Recall from Lemma 3.7 that

BM,R(Y)=∑m≤Yμ2(m)=1ω(m)≤RMω(m).B_{M,R}(Y) = \sum_{\substack{m \le Y\\ \mu^2(m)=1\\ \omega(m)\le R}} M^{\omega(m)}.

Lemma 4.2. One has

∣EG,t(X)∣≪G∑nD(n)≤XB∣I(G)∣,R((XD(n))a(G)).(24)|\mathcal{E}_{G,t}(X)| \ll_G \sum_{\substack{\mathbf{n}\\ D(\mathbf{n})\le X}} B_{|I(G)|,R}\left(\left(\frac{X}{D(\mathbf{n})}\right)^{a(G)}\right). \tag*{(24)}

Proof. By the Koymans–Pagano parametrization, it is enough to count the corresponding solvable bundled primitive tuples. The first coordinates have OG(1)O_G(1) possible values and may therefore be absorbed into the implied constant.

Fix the nonminimal coordinates n\mathbf{n}. The discriminant inequality gives

m(v)e0D(n)≤X.m(\mathbf{v})^{e_0}D(\mathbf{n}) \le X.

Thus D(n)≤XD(\mathbf{n}) \le X and, since e0=a(G)−1e_0=a(G)^{-1},

m(v)≤(XD(n))a(G).m(\mathbf{v}) \le\left(\frac{X}{D(\mathbf{n})}\right)^{a(G)}.

Furthermore, m(v)m(\mathbf{v}) is squarefree and ω(m(v))≤R\omega(m(\mathbf{v})) \le R.

For a fixed squarefree integer mm, the ordered minimal-coordinate tuples with product mm are obtained by assigning each prime divisor of mm to one of the ∣I(G)∣|I(G)| minimal coordinates. Their number is therefore at most ∣I(G)∣ω(m)|I(G)|^{\omega(m)}. Hence, for fixed n\mathbf{n}, the number of possible minimal-coordinate tuples is at most

∑m≤(X/D(n))a(G)μ2(m)=1ω(m)≤R∣I(G)∣ω(m)=B∣I(G)∣,R((XD(n))a(G)).\sum_{\substack{m\le(X/D(\mathbf{n}))^{a(G)}\\ \mu^2(m)=1\\ \omega(m)\le R}} |I(G)|^{\omega(m)} = B_{|I(G)|,R}\left(\left(\frac{X}{D(\mathbf{n})}\right)^{a(G)}\right).

Summing over n\mathbf{n} proves the result. □

Set a=a(G)a=a(G). If J(G)=∅J(G)=\varnothing, then (24) reduces to B∣I(G)∣,R(Xa)B_{|I(G)|,R}(X^a), and Theorem 1.2 follows immediately from Lemma 3.7. Assume therefore that J(G)≠∅J(G)\ne\varnothing, and set rG:=∣J(G)∣r_G:=|J(G)|. For g∈J(G)g\in J(G) define λg:=aeg\lambda_g:=ae_g. By Definition 2.1, one has λg>1\lambda_g>1. Set

κG:=min⁡g∈J(G)aeg−1rGeg>0.(25)\kappa_G:=\min_{g\in J(G)}\frac{ae_g-1}{r_Ge_g}>0. \tag*{(25)}

Choose an integer BG≥1B_G\ge1 such that

BGκG≥2,(26)B_G\kappa_G\ge2, \tag*{(26)}

and set ZX:=(log⁡X)BGZ_X:=(\log X)^{B_G}. We take XX sufficiently large that ZX<XZ_X<X. For D(n)≤XD(\mathbf{n})\le X, set

Yn:=(XD(n))a.Y_{\mathbf{n}}:=\left(\frac{X}{D(\mathbf{n})}\right)^a.

Suppose first that D(n)≤ZXD(\mathbf{n}) \le Z_X. For XX sufficiently large in terms of GG,

log⁡Yn=alog⁡X−alog⁡D(n)≥a2log⁡X.\log Y_{\mathbf{n}}=a\log X-a\log D(\mathbf{n})\ge\frac{a}{2}\log X.

In particular, Yn≥3Y_{\mathbf{n}}\ge3 once XX is sufficiently large. Also 1+log⁡log⁡Yn≪G1+log⁡log⁡X1+\log\log Y_{\mathbf{n}}\ll_G 1+\log\log X. Hence Lemma 3.7 gives

B∣I(G)∣,R(Yn)≪G,tXalog⁡X(1+log⁡log⁡X)R−1D(n)−a.(27)B_{|I(G)|,R}(Y_{\mathbf{n}})\ll_{G,t}\frac{X^a}{\log X}(1+\log\log X)^{R-1}D(\mathbf{n})^{-a}. \tag*{(27)}

Summing over D(n)≤ZXD(\mathbf{n})\le Z_X and then dropping this restriction, we obtain

∑D(n)≤ZXB∣I(G)∣,R(Yn)≪G,tXalog⁡X(1+log⁡log⁡X)R−1∏g∈J(G)ζ(aeg).(28)\sum_{D(\mathbf{n})\le Z_X}B_{|I(G)|,R}(Y_{\mathbf{n}}) \ll_{G,t}\frac{X^a}{\log X}(1+\log\log X)^{R-1}\prod_{g\in J(G)}\zeta(ae_g). \tag*{(28)}

The product on the right is finite because aeg>1ae_g>1 for every g∈J(G)g\in J(G).

It remains to estimate the terms in (24) for which D(n)>ZXD(\mathbf{n})>Z_X. The crude estimate in Lemma 3.7 gives

∑ZX<D(n)≤XB∣I(G)∣,R(Yn)≪G,tXa∑D(n)>ZXD(n)−a.(29)\sum_{Z_X<D(\mathbf{n})\le X}B_{|I(G)|,R}(Y_{\mathbf{n}})\ll_{G,t}X^a\sum_{D(\mathbf{n})>Z_X}D(\mathbf{n})^{-a}. \tag*{(29)}

If D(n)>ZXD(\mathbf{n})>Z_X, then for some g∈J(G)g\in J(G), ngeg>ZX1/rGn_g^{e_g}>Z_X^{1/r_G}. Consequently

∑D(n)>ZXD(n)−a≤∑g∈J(G)(∏h∈J(G)h≠gζ(aeh))∑n>ZX1/(rGeg)n−aeg.(30)\sum_{D(\mathbf{n})>Z_X}D(\mathbf{n})^{-a}\le\sum_{g\in J(G)}\left(\prod_{\substack{h\in J(G)\\h\ne g}}\zeta(ae_h)\right)\sum_{n>Z_X^{1/(r_Ge_g)}}n^{-ae_g}. \tag*{(30)}

For N≥1N\ge1, comparison with an integral gives

∑n>Nn−aeg≤N−aeg+∫N∞u−aeg du≪GN1−aeg,\sum_{n>N}n^{-ae_g}\le N^{-ae_g}+\int_N^\infty u^{-ae_g}\,du\ll_G N^{1-ae_g},

since aeg>1ae_g>1. Taking N=ZX1/(rGeg)N=Z_X^{1/(r_Ge_g)}, the exponent of ZXZ_X in this bound is

1−aegrGeg≤−κG.\frac{1-ae_g}{r_Ge_g}\le-\kappa_G.

It follows from (25) that

∑D(n)>ZXD(n)−a≪GZX−κG≪G(log⁡X)−2.(31)\sum_{D(\mathbf{n})>Z_X}D(\mathbf{n})^{-a}\ll_G Z_X^{-\kappa_G}\ll_G(\log X)^{-2}. \tag*{(31)}

Combining (29) and (31), the tail is OG,t(Xa(log⁡X)−2)O_{G,t}(X^a(\log X)^{-2}). Since R≥1R\ge1, this is smaller than the right-hand side of Theorem 1.2.

Proof of Theorem 1.2. If J(G)=∅J(G)=\varnothing, apply Lemma 3.7 directly to (24). If J(G)≠∅J(G)\ne\varnothing, combine (24), (28) and (31). In both cases,

∣EG,t(X)∣≪G,tXa(G)log⁡X(1+log⁡log⁡X)RG(t)−1.|\mathcal{E}_{G,t}(X)|\ll_{G,t}\frac{X^{a(G)}}{\log X}(1+\log\log X)^{R_G(t)-1}.

Theorem 1.2 now follows from Lemma 4.1. □

Corollary 4.3. For every fixed positive integer tt,

NG,t(X)NG(X)≪G,t(1+log⁡log⁡X)RG(t)−1log⁡X.\frac{N_{G,t}(X)}{N_G(X)} \ll_{G,t} \frac{(1+\log\log X)^{R_G(t)-1}}{\log X}.

If, in addition, I(G)⊆Z(G)I(G) \subseteq Z(G), then

NG,t(X)NG(X)≪G,t(1+log⁡log⁡X)RG(t)−1(log⁡X)b(G,Q).\frac{N_{G,t}(X)}{N_G(X)} \ll_{G,t} \frac{(1+\log\log X)^{R_G(t)-1}}{(\log X)^{b(G,\mathbb{Q})}}.

Proof. The weak Malle lower bound of Klüners–Malle gives NG(X)≫GXa(G)N_G(X) \gg_G X^{a(G)}; see [7 Theorem 6.3]. Dividing the estimate of Theorem 1.2 by this lower bound proves the first assertion. If I(G)⊆Z(G)I(G) \subseteq Z(G), Koymans–Pagano prove NG(X)∼cGXa(G)(log⁡X)b(G,Q)−1N_G(X) \sim c_G X^{a(G)}(\log X)^{b(G,\mathbb{Q})-1} with cG>0c_G > 0; see [8 Theorem 1.1]. Division by this asymptotic gives the second assertion. □

We now record the consequence of Corollary 4.3 for Conjecture 1.4 of Dixit–Pasupulati [2]. Following their notation, let

M(G,n,X):={K/Q:[K:Q]=n, Gal⁡(K~/Q)≅G, ∣Disc⁡(K)∣≤X},\mathcal{M}(G,n,X) := \{K/\mathbb{Q} : [K:\mathbb{Q}] = n,\ \operatorname{Gal}(\widetilde{K}/\mathbb{Q}) \cong G,\ |\operatorname{Disc}(K)| \le X\},

where K~\widetilde{K} is the Galois closure of K/QK/\mathbb{Q} and fields are counted up to Q\mathbb{Q}-isomorphism. Let M′(G,n,X)\mathcal{M}'(G,n,X) be the subset consisting of fields which admit a tower from Q\mathbb{Q} in which every successive extension is abelian.

Corollary 4.4. Conjecture 1.4 of Dixit–Pasupulati holds for every nontrivial finite nilpotent group in the regular representation. More precisely, for every nontrivial finite nilpotent group GG and every fixed positive integer tt,

#{K∈M′(G,∣G∣,X):gK=t}#M′(G,∣G∣,X)≪G,t(1+log⁡log⁡X)RG(t)−1log⁡X.\frac{\#\{K \in\mathcal{M}'(G,|G|,X) : g_K = t\}}{\#\mathcal{M}'(G,|G|,X)} \ll_{G,t} \frac{(1+\log\log X)^{R_G(t)-1}}{\log X}.

In particular, this ratio tends to zero as X→∞X \to\infty.

Proof. If K∈M(G,∣G∣,X)K \in\mathcal{M}(G,|G|,X) and K~\widetilde{K} is its Galois closure, then

[K:Q]=∣G∣=[K~:Q],[K:\mathbb{Q}] = |G| = [\widetilde{K}:\mathbb{Q}],

so K=K~K = \widetilde{K}. Thus the fields in M(G,∣G∣,X)\mathcal{M}(G,|G|,X) are precisely the Galois GG-extensions counted by NG(X)N_G(X). Since GG is nilpotent, choose a subnormal series

1=Gm◃Gm−1◃⋯◃G0=G1 = G_m \triangleleft G_{m-1} \triangleleft\cdots\triangleleft G_0 = G

with cyclic prime-order quotients. The corresponding fixed fields form a tower

Q=KG0⊂KG1⊂⋯⊂KGm=K,\mathbb{Q} = K^{G_0} \subset K^{G_1} \subset\cdots\subset K^{G_m} = K,

and Gal⁡(KGi+1/KGi)≅Gi/Gi+1\operatorname{Gal}(K^{G_{i+1}}/K^{G_i}) \cong G_i/G_{i+1} is abelian for every ii. Thus every such extension belongs to M′(G,∣G∣,X)\mathcal{M}'(G,|G|,X). Hence

M′(G,∣G∣,X)=M(G,∣G∣,X),\mathcal{M}'(G,|G|,X) = \mathcal{M}(G,|G|,X),

and the ratio in the statement is NG,t(X)/NG(X)N_{G,t}(X)/N_G(X). The result now follows from Corollary 4.3. □

:::

Bounds for genus moments

In this section, we prove Theorem 1.4. We first obtain an upper bound for the moments for an arbitrary finite nilpotent group. Under the centrality assumption, we construct a family of extensions which gives a matching lower bound.

Throughout this section, GG is a nontrivial finite nilpotent group, ℓ\ell is the smallest prime divisor of ∣G∣|G|, and r≥0r \ge0 is fixed. All estimates in XX are understood as X→∞X \to\infty. Set

a:=a(G),M:=∣I(G)∣,i:=i(G,Q)=Mℓ−1,u:=ℓr.a := a(G), \qquad M := |I(G)|, \qquad i := i(G,\mathbb{Q}) = \frac{M}{\ell- 1}, \qquad u := \ell^r.

Recall that i=b(G,Q)i = b(G,\mathbb{Q}) when I(G)⊆Z(G)I(G) \subseteq Z(G).

Proposition 5.1. Let K/QK/\mathbb{Q} be a finite Galois extension. Then

gK≤∏q∣Disc⁡(K)eq(K/Q).(32)g_K \le\prod_{q \mid\operatorname{Disc}(K)} e_q(K/\mathbb{Q}). \tag*{(32)}

Proof. This follows from the narrow genus-number formula [4 Theorem 7], applied over Q\mathbb{Q}. □

To estimate the moments, we compare the genus number with quantities determined by ramification. The primes with inertia of order ℓ\ell give a lower bound, while the product of the ramification indices gives an upper bound. For a Galois GG-extension L/QL/\mathbb{Q}, define

Ur(L):=ℓrωmin⁡,SG(L),(33)U_r(L) := \ell^{r\omega_{\min,S_G}(L)}, \tag*{(33)}
Vr(L):=∏q∉SGq ramified in Leq(L/Q)r.(34)V_r(L) := \prod_{\substack{q \notin S_G \\ q\ \mathrm{ramified\ in}\ L}} e_q(L/\mathbb{Q})^r. \tag*{(34)}

If v=(vg)g≠1v = (v_g)_{g \ne1} is its Koymans–Pagano tuple, then Proposition 3.2 gives

Vr(L)=∏g≠1ord⁡(g)rω(vg(2)).(35)V_r(L) = \prod_{g \ne1} \operatorname{ord}(g)^{r\omega(v_g(2))}. \tag*{(35)}

By Theorem 3.5 and Proposition 5.1,

ℓ−rdℓ(G)Ur(L)≤gLr≤∣G∣r∣SG∣Vr(L).(36)\ell^{-r d_\ell(G)} U_r(L) \le g_L^r \le|G|^{r|S_G|} V_r(L). \tag*{(36)}

For the upper bound, we used ep(L/Q)≤∣G∣e_p(L/\mathbb{Q}) \le|G| at each p∈SGp \in S_G. Thus, it suffices to obtain an upper bound for the sum of Vr(L)V_r(L) and a lower bound for the sum of Ur(L)U_r(L).

We shall use the following consequence of the Selberg–Delange method to estimate the contribution from primes occurring in the minimal coordinates.

Lemma 5.2. Let TT be a finite set of rational primes and let z>0z > 0. Put

Bℓ,T,z(Y):=∑m≤Yμ2(m)=1q∣m⇒q∉T, q≡1 (mod⁡ ℓ)zω(m).B_{\ell,T,z}(Y) := \sum_{\substack{m \le Y \\ \mu^2(m)=1 \\ q \mid m \Rightarrow q \notin T,\ q \equiv1\ (\operatorname{mod}\ \ell)}} z^{\omega(m)}.

Then

Bℓ,T,z(Y)≍ℓ,T,zY(log⁡Y)z/(ℓ−1)−1(Y→∞).(37)B_{\ell,T,z}(Y) \asymp_{\ell,T,z} Y(\log Y)^{z/(\ell-1)-1} \qquad(Y \to\infty). \tag*{(37)}

In particular, for every Y≥1Y \ge1,

Bℓ,T,z(Y)≪ℓ,T,zY(1+log⁡Y)z/(ℓ−1)−1.(38)B_{\ell,T,z}(Y) \ll_{\ell,T,z} Y(1+\log Y)^{z/(\ell-1)-1}. \tag*{(38)}

Proof. The associated Dirichlet series is

∏q∉Tq≡1 (mod ℓ)(1+zq−w),ℜ(w)>1.\prod_{\substack{q\notin T\\ q\equiv1\ (\mathrm{mod}\ \ell)}}(1+zq^{-w}),\qquad\Re(w)>1.

Apply [8 Theorem 4.1] to Q(ζℓ)/Q\mathbb{Q}(\zeta_\ell)/\mathbb{Q}, the identity Frobenius class, and the excluded-prime set T∪{ℓ}T\cup\{\ell\}. The Selberg–Delange parameter is z/(ℓ−1)z/(\ell-1), and the leading constant is positive. This proves (37); enlarging the implied constant for bounded YY gives (38).

Theorem 5.3. For every nontrivial finite nilpotent group GG,

∑L/Q GaloisGal⁡(L/Q)≃G∣Disc⁡(L)∣≤XVr(L)≪G,rXa(log⁡X)ui−1.(39)\sum_{\substack{L/\mathbb{Q}\ \mathrm{Galois}\\ \operatorname{Gal}(L/\mathbb{Q})\simeq G\\ |\operatorname{Disc}(L)|\leq X}}V_r(L)\ll_{G,r}X^a(\log X)^{u_i-1}. \tag*{(39)}

Consequently,

MG,r(X)≪G,rXa(log⁡X)ui−1.(40)M_{G,r}(X)\ll_{G,r}X^a(\log X)^{u_i-1}. \tag*{(40)}

Proof. We follow the counting argument of § 4, keeping track of the weight Vr(L)V_r(L). Set J:=J(G)=G∖(I(G)∪{1})J:=J(G)=G\setminus(I(G)\cup\{1\}). For the nonminimal coordinates n=(ng)g∈J\mathbf{n}=(n_g)_{g\in J}, put

D(n):=∏g∈Jngeg,c(n):=∏g∈Jord⁡(g)rω(ng).D(\mathbf{n}):=\prod_{g\in J}n_g^{e_g},\qquad c(\mathbf{n}):=\prod_{g\in J}\operatorname{ord}(g)^{r\omega(n_g)}.

The first components vg(1)v_g(1) of the Koymans–Pagano tuple have OG(1)O_G(1) possible values. We retain squarefreeness and pairwise coprimality among the ngn_g, and discard the other restrictions on the nonminimal coordinates.

The product mm of the minimal coordinates is squarefree, supported on primes q∉SGq\notin S_G with q≡1(modℓ)q\equiv1\pmod{\ell}, and satisfies

m1/aD(n)≤X.m^{1/a}D(\mathbf{n})\leq X.

Each prime dividing mm may be assigned to one of the MM coordinates indexed by I(G)I(G), and each assignment contributes the weight uu. Discarding the solvability conditions and the coprimality conditions between minimal and nonminimal coordinates therefore gives

∑L/Q GaloisGal⁡(L/Q)≃G∣Disc⁡(L)∣≤XVr(L)≪G∑n∗D(n)≤Xc(n)Bℓ,SG,uM((X/D(n))a),\sum_{\substack{L/\mathbb{Q}\ \mathrm{Galois}\\ \operatorname{Gal}(L/\mathbb{Q})\simeq G\\ |\operatorname{Disc}(L)|\leq X}}V_r(L) \ll_G \sum_{\substack{\mathbf{n}^{*}\\D(\mathbf{n})\leq X}} c(\mathbf{n})B_{\ell,S_G,uM}\left((X/D(\mathbf{n}))^a\right),

where the asterisk indicates squarefree, pairwise coprime coordinates. The factor ∣Aut⁡(G)∣|\operatorname{Aut}(G)| relating epimorphisms to fields is absorbed into the implied constant.

Put κ:=ui−1\kappa:=ui-1. Since M≥ℓ−1M\geq\ell-1 and u≥1u\geq1, we have κ≥0\kappa\geq0. If D(n)≤XD(\mathbf{n})\leq X and Y=(X/D(n))aY=(X/D(\mathbf{n}))^a, then

Y≥1,(1+log⁡Y)κ≪G,r(log⁡X)κ(X≥3).Y\geq1,\qquad(1+\log Y)^\kappa\ll_{G,r}(\log X)^\kappa\qquad(X\geq3).

Applying Lemma 5.2 and extending the sum to all squarefree, pairwise coprime tuples n\mathbf{n}, with no restriction on their prime divisors, we obtain

∑L/Q GaloisGal⁡(L/Q)≃G∣Disc⁡(L)∣≤XVr(L)≪G,rXa(log⁡X)κ∑n∗c(n)D(n)a=Xa(log⁡X)κ∏p(1+∑g∈Jord⁡(g)rpaeg).\begin{aligned} \sum_{\substack{L/\mathbb{Q}\ \mathrm{Galois}\\ \operatorname{Gal}(L/\mathbb{Q})\simeq G\\ |\operatorname{Disc}(L)|\le X}} V_r(L) \ll_{G,r} X^a(\log X)^\kappa\sum_{\mathbf{n}}^{*} \frac{c(\mathbf{n})}{D(\mathbf{n})^a} \\ &= X^a(\log X)^\kappa\prod_p \left(1+\sum_{g\in J}\frac{\operatorname{ord}(g)^r}{p^{ae_g}}\right). \end{aligned}

The Euler product converges because aeg>1ae_g>1 for every g∈Jg\in J. If JJ is empty, it is interpreted as 1. This proves (39), and (36) gives (40). □

Proposition 5.4. Suppose that I(G)⊆Z(G)I(G)\subseteq Z(G). Then

∑L/Q GaloisGal⁡(L/Q)≃G∣Disc⁡(L)∣≤XUr(L)≫G,rXa(log⁡X)ub(G,Q)−1.(41)\sum_{\substack{L/\mathbb{Q}\ \mathrm{Galois}\\ \operatorname{Gal}(L/\mathbb{Q})\simeq G\\ |\operatorname{Disc}(L)|\le X}} U_r(L) \gg_{G,r} X^a(\log X)^{ub(G,\mathbb{Q})-1}. \tag*{(41)}

Proof. For the lower bound, we fix one GG-extension and twist its defining epimorphism by characters taking values in the central subgroup H=I(G)∪{1}H=I(G)\cup\{1\}. These twists allow us to introduce ramification of order ℓ\ell at additional primes.

The set H:=I(G)∪{1}H:=I(G)\cup\{1\} is a central elementary abelian ℓ\ell-subgroup: for h,h′∈Hh,h'\in H, one has (hh′)ℓ=hℓ(h′)ℓ=1(hh')^\ell=h^\ell(h')^\ell=1. Write HH additively, and let ι:H↪G\iota:H\hookrightarrow G denote the inclusion.

Fix a GG-extension L0/QL_0/\mathbb{Q} once for each GG, and let ψ0:GQ↠G\psi_0:G_{\mathbb{Q}}\twoheadrightarrow G be an associated epimorphism. Such an extension exists by Shafarevich’s theorem; see [7 Theorem 6.1]. Let TT be the union of SGS_G and the finite primes ramified in L0L_0. Thus TT is fixed throughout the construction. For each prime q∉Tq\notin T with q≡1(modℓ)q\equiv1\pmod{\ell}, choose a nonzero character

χq:GQ⟶Fℓ\chi_q:G_{\mathbb{Q}}\longrightarrow\mathbb{F}_{\ell}

corresponding to the degree-ℓ\ell subfield of Q(ζq)\mathbb{Q}(\zeta_q). It is ramified at qq and unramified at every other finite prime.

Let mm be squarefree and supported on these primes. For each q∣mq\mid m, choose hq∈H∖{0}=I(G)h_q\in H\setminus\{0\}=I(G), and define

ξ:=∑q∣mhqχq,ψξ(σ):=ι(ξ(σ))ψ0(σ).\xi:=\sum_{q\mid m}h_q\chi_q,\qquad\psi_\xi(\sigma):=\iota(\xi(\sigma))\psi_0(\sigma).

Centrality of HH implies that ψξ\psi_\xi is a homomorphism. At every prime in TT, its restriction to inertia agrees with that of ψ0\psi_0. At a prime q∣mq\mid m, its inertia image is ⟨hq⟩\langle h_q\rangle, of order ℓ\ell.

We show that ψξ\psi_\xi is surjective. Put Jξ:=im⁡(ψξ)J_\xi:=\operatorname{im}(\psi_\xi). Its image in G/HG/H is surjective, so JξH=GJ_\xi H=G. Since HH is central, JξJ_\xi is normal in GG. The composite

GQ→ψ0G⟶G/JξG_{\mathbb{Q}}\xrightarrow{\psi_0}G\longrightarrow G/J_\xi

is unramified at every finite prime. Indeed, at primes in TT, the inertia image of ψ0\psi_0 equals that of ψξ\psi_\xi and is contained in JξJ_\xi; outside TT, the map ψ0\psi_0 is unramified. Minkowski’s discriminant theorem implies that this quotient is trivial. Thus Jξ=GJ_\xi=G.

Let LξL_\xi be the field defined by ψξ\psi_\xi. Outside TT, it is ramified precisely at the primes dividing mm, each with tame inertia of order ℓ\ell. The discriminant contribution of these primes is therefore m1/am^{1/a}. For p∈Tp\in T, let E=(Lξ)wE=(L_\xi)_w for a prime w∣pw\mid p. Since Lξ/QL_\xi/\mathbb{Q} is Galois,

vp(∣Disc⁡(Lξ)∣)=∣G∣[E:Qp]vp(Disc⁡(E/Qp)).v_p\bigl(|\operatorname{Disc}(L_\xi)|\bigr)=\frac{|G|}{[E:\mathbb{Q}_p]}v_p\bigl(\operatorname{Disc}(E/\mathbb{Q}_p)\bigr).

There are only finitely many extensions of Qp\mathbb{Q}_p of degree at most ∣G∣|G|. Consequently, there is a constant CG,T≥1C_{G,T} \ge1, independent of ξ\xi, such that

∣Disc⁡(Lξ)∣≤CG,Tm1/a.|\operatorname{Disc}(L_\xi)| \le C_{G,T}m^{1/a}.

Also, since T⊇SGT \supseteq S_G and u≥1u \ge1,

Ur(Lξ)≥uω(m).U_r(L_\xi) \ge u^{\omega(m)}.

We now count the epimorphisms obtained in this way. For q∣mq \mid m, fix an inertia subgroup Iq⊆GQI_q \subseteq G_{\mathbb{Q}}. Since ψ0\psi_0 is unramified at qq, we have

ψξ(τ)=ι(hqχq(τ))(τ∈Iq).\psi_\xi(\tau)=\iota(h_q\chi_q(\tau)) \qquad(\tau\in I_q).

The restriction of χq\chi_q to IqI_q is surjective and has been fixed in advance. Thus, different choices of hqh_q give different restrictions to inertia, even when they generate the same subgroup of HH. Different choices of mm give different ramified primes outside TT. Consequently, distinct choices of mm and (hq)q∣m(h_q)_{q\mid m} give distinct epimorphisms. For each mm there are Mω(m)M^{\omega(m)} such choices. Every field arises from at most ∣Aut⁡(G)∣|\operatorname{Aut}(G)| of these epimorphisms. Hence

∑L/Q GaloisGal⁡(L/Q)≃G∣Disc⁡(L)∣≤XUr(L)≥1∣Aut⁡(G)∣Bℓ,T,uM((X/CG,T)a)\sum_{\substack{L/\mathbb{Q}\ \mathrm{Galois}\\ \operatorname{Gal}(L/\mathbb{Q})\simeq G\\ |\operatorname{Disc}(L)|\le X}} U_r(L) \ge \frac{1}{|\operatorname{Aut}(G)|}B_{\ell,T,u}M\left((X/C_{G,T})^a\right)
≫G,rXa(G)(log⁡X)uM/(ℓ−1)−1,\gg_{G,r}X^{a(G)}(\log X)^{uM/(\ell-1)-1},

by Lemma 5.2. Since M/(ℓ−1)=b(G,Q)M/(\ell-1)=b(G,\mathbb{Q}), this proves (41).

Combining the upper and lower bounds, we obtain the following result, which proves Theorem 1.4.

Theorem 5.5. Suppose that I(G)⊆Z(G)I(G)\subseteq Z(G). For every fixed real r≥0r\ge0,

MG,r(X)≍G,rXa(G)(log⁡X)ℓrb(G,Q)−1.(42)M_{G,r}(X)\asymp_{G,r}X^{a(G)}(\log X)^{\ell r b(G,\mathbb{Q})-1}. \tag*{(42)}

Moreover,

MG,r(X)NG(X)≍G,r(log⁡X)(ℓr−1)b(G,Q).(43)\frac{M_{G,r}(X)}{N_G(X)}\asymp_{G,r}(\log X)^{(\ell r-1)b(G,\mathbb{Q})}. \tag*{(43)}

Proof. By (36), the required bounds follow from Theorem 5.3 and Proposition 5.4, since i=b(G,Q)i=b(G,\mathbb{Q}) under the centrality assumption. Taking r=0r=0 gives

NG(X)=MG,0(X)≍GXa(G)(log⁡X)b(G,Q)−1.N_G(X)=M_{G,0}(X)\asymp_G X^{a(G)}(\log X)^{b(G,\mathbb{Q})-1}.

Dividing the two estimates proves (43).

Department of Mathematics, Institute of Mathematical Sciences (HBNI), CIT Campus, IV Cross Road, Chennai, India 600113

Email address: [email protected]

Chennai Mathematical Institute, H1, SIPCOT IT Park, Kelambakkam, Siruseri, Tamil Nadu 603103, India

Email address: [email protected]

References

  1. [1]Sambhabi Bose, Kevin J. McGown, Ishan Panpaliya, Natalie Welling, and Laney Williams. The average genus number for pure fields of prime degree. arXiv:2507.14317, 2025.arxiv.org/abs/2507.14317
  2. [2]Anup B. Dixit and Sunil Kumar Pasupulati. Statistics of the genus number of S₃ × ℤ/qℤ and D₄-fields. arXiv:2605.04792, 2026.
  3. [3]Christopher Frei, Daniel Loughran, and Rachel Newton. Distribution of genus numbers of abelian number fields. Journal of the London Mathematical Society, 107(6):2197–2217, 2023.DOI
  4. [4]Stanley J. Gurak. Ideal-theoretic characterization of the relative genus field. Journal für die reine und angewandte Mathematik, 296:119–124, 1977.DOI
  5. [5]Makoto Ishida. The Genus Fields of Algebraic Number Fields, volume 555 of Lecture Notes in Mathematics. Springer-Verlag, Berlin–Heidelberg, 1976.
  6. [6]Henry H. Kim. Genus numbers of cyclic and dihedral extensions of prime degree. Acta Arithmetica, 192(3):289–300, 2020.DOI
  7. [7]Jürgen Klüners and Gunter Malle. Counting nilpotent Galois extensions. Journal für die reine und angewandte Mathematik, 572:1–26, 2004.DOI
  8. [8]Peter Koymans and Carlo Pagano. On Malle’s conjecture for nilpotent groups. Transactions of the American Mathematical Society, Series B, 10:310–354, 2023.DOI
  9. [9]Gunter Malle. On the distribution of Galois groups. Journal of Number Theory, 92:315–329, 2002.DOI
  10. [10]Gunter Malle. On the distribution of Galois groups. II. Experimental Mathematics, 13(2):129–135, 2004.DOI
  11. [11]Kevin J. McGown and Amanda Tucker. Statistics of genus numbers of cubic fields. Annales de l’Institut Fourier, 73(4):1365–1383, 2023.
  12. [12]Kevin J. McGown, Frank Thorne, and Amanda Tucker. Counting quintic fields with genus number one. Mathematical Research Letters, 30(2):577–588, 2023.
  13. [13]W. Narkiewicz, Elementary and analytic theory of algebraic numbers, Springer, 2004.
  14. [14]Jürgen Neukirch, Alexander Schmidt, and Kay Wingberg. Cohomology of Number Fields, volume 323 of Grundlehren der mathematischen Wissenschaften. Springer, Berlin, 2 edition, 2008.DOI
  15. [15]Jean-Pierre Serre. Local Fields, volume 67 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1979. Translated from the French by Marvin Jay Greenberg.

Paper details

vlnc:W_yb8swe9y
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

GPT-6.1 Sol (bare)

View statement
Looking for related papers…
Contents