Counting C_2 \wr S_4 fields with a power saving error term

Combine Lemma 2.1, Theorem 1.2, and Theorem 3.1.

Given an S_4-octic field L, we know that L contains a unique S_4-quartic field K. Moreover, L is the unique S_4-octic field satisfying K\subseteq L\subseteq\widetilde{K}.

Therefore, given an S_4-quartic field K, we can define

where L is the unique S_4-octic containing K. We then have

Trivially bounding this sum and using \operatorname{\mathcal{N}}_4(S_4,X)\ll X (see [MR2183288]) yields

as desired.

Using Lemma 5.1 we have

Using the parametrization given in [MR1918290], or alternatively Equation 3 in Section 3.2 of [MR4808549], we can bound the inner sum as

where S_2(K) denotes the 2-Selmer group of K and the notation \operatorname{N}(\mathfrak{b})=\square denotes that the absolute norm of the ideal \mathfrak{b} in K is a square in \mathbb{Z}. Using Proposition 3.2 of [MR1918290] and (Proof),

we find that

Moreover, we have

In the above, the a_n(K) are the coefficients in the Dirichlet series of the Dedekind zeta function \zeta_K(s) and \tau_4(n) is the number of ways to write n as a product of 4 positive integers. We have used the estimate a_n\leq \tau_4(n)\ll n^\varepsilon; see, for example, Exercise 1 on page 231 of [MR195803], and the result contained in [MR4188678]. Assembling all the pieces together and using partial summation on \operatorname{\mathcal{N}}_4(S_4,X)\ll X, we conclude that

As in Lemma 6.2 we write \operatorname{N}(\Delta_{L/K})=N_0 N_1 N_2 with N_1 being a 4-th power in \mathbb{Z}.

Moreover we write \Delta_K=3^a N_0 N_0' with (N_0',3)=1;

if p\mid N_0', then p^2\mid\Delta_K by the proof of Lemma 6.2. As we only seek an upper bound, we will use the fact that \operatorname{N}(\Delta_{L/K})\leq X implies N_0 N_1\leq X to avoid the wildly ramified primes in our count. Let Y>0 be a parameter to be determined later. We will condition on the value of \operatorname{rad}(N_0').

First, we treat the case where \operatorname{rad}(N_0')\leq Y. Observe that \Delta_K^2 N_0 N_1\leq |\Delta_L|\leq X and N_0=\Delta_K/(3^a N_0') implies N_1\leq 3^a N_0' X/\Delta_K^3. Moreover, in this case

N_0'\leq Y^2 (by Lemma 6.2, Part 3) and hence N_1\leq 3^a XY^2/\Delta_K^3. Making use of Lemma 6.3, we estimate:

Now we treat the cases \operatorname{rad}(N_0')>Y. In this case there exists a squarefree q>Y such that q^2\mid\Delta_K; namely, q=\operatorname{rad}(N_0'). This makes Lemma 6.4 applicable, which takes the form O(X^{1-\delta+\varepsilon}+X/Y^{1-\varepsilon}) with \delta=1/12. We estimate:

Summarizing, we have a bound of

where we have chosen Y=X^{1/12}. Using the values of \delta and \kappa we have available to us, we are forced to accept the second error term which has exponent less than 0.598.

We follow the proof for \operatorname{GL}_2(\mathbb{F}_3) but some details are different. We note that in the subsequent calculations, we use the following analog of Lemma 6.3, which is proved in the exact same way, except where v_p(N_1) = 4 is replaced by v_p(N_1) \geq 2:

We write \operatorname{N}(\Delta_{L/K})=N_0 N_1 N_2 where these quantities are defined the same way as in Section ?. The analog of Lemma 6.2 is not totally true. We have v_p(N_1)\geq 2 for all p\mid N_1, which is slightly weaker than Part 1 of the Lemma.

We condition on the value of \operatorname{rad}(D_1).

First, we consider the case where \operatorname{rad}(D_1)\leq Y. Notice that the condition |\Delta_L|\leq X implies N_1\leq X/(\Delta_K^2 N_0). Further, this implies

Indeed, observe that |\Delta_K|=D_0 D_1 D_2\leq 2^a 3^b D_0 Y^2, which implies D_0\geq \Delta_K/(2^a 3^bY^2).

We estimate:

In the case where \operatorname{rad}(N_0')>Y, we have

Then we choose Y=X^{1/16} and the final estimate is

As before, we are forced to accept the second error term due to the values of \delta, \kappa that we have available.

Suppose L\in\mathcal{F}. We have that G\colonequals \operatorname{Gal}(\widetilde{L}/\mathbb{Q}) embeds into C_2 \wr S_4. See Theorem 1.8 of [MR4858206] or Proposition 3.2 of [MR4594565].

Using SageMath, one finds that there are 32 subgroups of C_2 \wr S_4 that are transitive as subgroups of S_8. Of these 32 subgroups, those with S_4 as a quotient are precisely the groups in \mathcal{G}.

Conversely, suppose that G \in \mathcal{G}. Then H_L \colonequals \text{Gal}(\widetilde{L}/L) is an index 8 subgroup of G containing no nontrivial normal subgroups of G. One can check that each such subgroup of G is contained in an index 4 subgroup H_K such that the quotient G/M is isomorphic to S_4, where M is the normal core of H_K.

Thus, the subfield of \widetilde{L} fixed by H_K is a quartic S_4-field contained in L, so L\in\mathcal{F}.

We begin by writing

By Theorem 2 of [MR4808549], the inner sum equals

Here we have used Theorem 1.1 of [MR4155220], which states that

Moreover, using the parametrization given in [MR1918290], or alternatively Equation 3 in Section 3.2 of [MR4808549], we also have the weaker estimate

In the above, S_2(K) is the 2-Selmer group of K, which can be bounded using Proposition 3.2 of [MR1918290]. Moreover, we have used \sum_{\operatorname{N}(\mathfrak{a}) \leq Y}1\leq|\Delta_K|^\varepsilon Y; this is not hard to derive and appears as Equation (21) in [MR4808549].

Let 0<Z<X^{1/2} be a parameter. Splitting the sum in (Proof), we have

Using (Proof), the first piece of (?) equals

From [MR2183288] one knows \operatorname{\mathcal{N}}_4(S_4,X)\ll X, and therefore partial summation yields

Note we have also used the estimate \zeta^*_K(1)\ll |\Delta_K|^\varepsilon. Therefore we have

For the second piece of (?), we use (Proof) to obtain

Summarizing, the error term is

Choosing Z=X^{2/5}, we have

where C is as in Theorem 1.1.

Let L be such an extension. One verifies that any embedding of C_2^3 \rtimes S_4 in S_8 is contained in A_8. If f(x) \in \mathbb{Q}[x] is the minimal polynomial of a primitive element \alpha of L/\mathbb{Q}, then it is well-known that the previous statement implies that \operatorname{Disc}(f) is a square. But \operatorname{Disc}(f) is also the discriminant of the lattice \mathbb{Z}[\alpha], which differs from \Delta_L by a square. Hence \Delta_L is also a square.

(1) The set \{1,2,3,4,6,8\} is precisely the set of orders of the cyclic subgroups of G.

(2) Suppose that e(\widetilde{L}/p) = 2. Then the inertia groups I(\widetilde{\mathfrak{P}}/p) are all isomorphic to C_2. Since the indices e(\mathfrak{p}/p) are not all the same, if all the inertia groups were the same, we would reach a contradiction by restricting to \widetilde{L}/K. Hence, the inertia groups I(\widetilde{\mathfrak{P}}/p) form a size 12 conjugacy class. Thus, the normalizer N of each inertia group I(\widetilde{\mathfrak{P}}/p) has index 12 in G. Each corresponding decomposition group D(\widetilde{\mathfrak{P}}/p) is contained in N, so |D(\widetilde{\mathfrak{P}}/p)| \leq 4. We have thus proved f(\widetilde{L}/p) \in \{1,2\}.

(3) The ramification index e(\widetilde{L}/p) cannot be 8, as then 4 would have to divide the degree of the extension \widetilde{L}/L. If e(\widetilde{L}/p) = 6, then all the inertia groups I(\widetilde{\mathfrak{P}}/p) must have intersection isomorphic to either C_3 or C_6 with \text{Gal}(\widetilde{L}/L). However, we have that \text{Gal}(\widetilde{L}/L) \cong S_3 in G and every copy of S_3 only intersects one copy of C_6 nontrivially; this copy cannot constitute a conjugacy class as no C_6 is normal in G. Finally, one can verify that every intersection of a copy of C_4 with a copy of S_3 is trivial in G, which excludes the case e(\widetilde{L}/p) = 4 since ramification happens in \widetilde{L}/L.

(1) Let p be a prime dividing N_1.

By definition, p ramifies in L but is unramified in K. Then p must also be unramified in \widetilde{K}, a degree 24 subfield of \widetilde{L}, and so we deduce that e(\widetilde{L}/p) = 2. The restriction of any of the inertia groups I(\widetilde{\mathfrak{P}}/p) to the field extension \widetilde{L}/\widetilde{K} must have order 2 since e(\widetilde{\mathfrak{P}}/p) = 2 and

p is unramified in \widetilde{K}. We conclude that the inertia groups I(\widetilde{\mathfrak{P}}/p) are all equal to \text{Gal}(\widetilde{L}/\widetilde{K}) since it is the unique normal subgroup of order 2 in \text{GL}_2(\mathbb{F}_3). Hence

e(\widetilde{\mathfrak{P}}_1/\mathfrak{P}_1) = e(\widetilde{\mathfrak{P}}_2/\mathfrak{P}_2) for any primes \widetilde{\mathfrak{P}}_1, \widetilde{\mathfrak{P}}_2 of \widetilde{L} (where \mathfrak{P}_1,\mathfrak{P}_2 are the primes of L lying underneath). It follows that the ramification indices e(\mathfrak{P}/p) are the same

for all primes \mathfrak{P} of L over p. Thus, since p is unramified in K, we find that all primes of K above p ramify in L/K.

(2) Let p be a prime that divides N_0. Since v_p(\Delta_L)\leq 7, from the relation

we know v_p(\Delta_K)\leq 3. We treat cases based on the value of v_p(\Delta_K), which must be positive since p ramifies in K. In the case where v_p(\Delta_K)=3, we deduce v_p(N_0) = 1 and the conclusion follows immediately. The cases where v_p(\Delta_K)=1,2 remain.

Factor (p) = \mathfrak{p}_1^{e_1} \mathfrak{p}_2^{e_2} \cdots \mathfrak{p}_m^{e_m} in K and write f_i \colonequals f(\mathfrak{p}_i/p). We will denote the splitting type of p in K by \sigma_K(p) \colonequals (f_1^{e_1} \dots f_m^{e_m}). Let \delta_i denote the indicator function for whether the prime \mathfrak{p}_i ramifies in L/K. We have that

We would like to show that the first sum is at most the value of the second sum. We note that it suffices to show that \delta_i = 1 implies e_i>1; that is, for any prime \mathfrak{p} in K above p, if \mathfrak{p} ramifies in L/K, then \mathfrak{p} ramifies in K/\mathbb{Q}.

Now suppose that v_p(\Delta_K) = 2. Then we must have \sigma_K(p)\in\{(1^2 1^2), (1^3 1), (2^2)\}. In all cases, we find v_p(N_0)\leq v_p(\Delta_K) from (?).

For the final case, we suppose that v_p(\Delta_K) = 1. Since p is tamely ramified in L/\mathbb{Q}, the inertia group of p is generated by a permutation of index v_p(\Delta_L). As the indices of all elements in \text{GL}_2(\mathbb{F}_3) are contained in \{3,4,6,7\}, we have via (?) that v_p(N_0) \in \{1,2,4,5\}. On the other hand, since p is ramified in K, it follows from (?) that v_p(N_0)\leq 3. Hence it suffices to show that v_p(N_0)\neq 2.

Suppose towards a contradiction that v_p(N_0) = 2.

As v_p(\Delta_K) = 1 we must have \sigma_K(p)\in\{(1^22), (1^211)\}. In the case that \sigma_K(p) = (1^22), we may assume that the degree 2 prime \mathfrak{p}_2 in K above p

ramifies in L, as otherwise the conclusion holds by (?). We also observe that only this prime ramifies in L/K, as otherwise v_p(N_0) > 2. Thus \sigma_L(p) \in \{(1^21^2 2^2), (2^2 2^2)\}. By part (3) of Lemma  6.1, we know that e(\widetilde{L}/p) = f(\widetilde{L}/p) = 2. Since the inertia groups I(\widetilde{\mathfrak{P}}/p) \cong C_2 are not all the same, they constitute a conjugacy class of size 12. Since all the ramification indices of primes in L over p are the same, the size of all the intersections I(\widetilde{\mathfrak{P}}/p) \cap S_3 must be the same for some copy of S_3 \leq G. However one checks that the 12 non-normal copies of C_2 in G intersect any copy of S_3 in G with orders both 1 and 2, a contradiction.

Moving on, we now assume \sigma_K(p) = (1^211). As we are in the situation where two primes above p ramify in L/K, we have

\sigma_L(p)\in\{(1^4 1^2 2), (1^4 1^2 11), (1^21^21^21^2), (2^21^21^2)\}. The first two may be resolved easily by observing that 4 does not divide the degree of the extension \widetilde{L}/L. The last may be resolved with the same logic as for the case of \sigma_K(p) = (1^2 2). Thus we assume that \sigma_L(p) = (1^21^21^21^2). If f(\widetilde{L}/p) = 2, then the same logic applies, so we can further assume that f(\widetilde{L}/p) = 1 by part (3) of Lemma 6.1. We observe that every non-normal copy of C_2 in G must intersect some copy of S_3 trivially, since the ramification indices e(\mathfrak{P}/p) are all equal to 2. As before, this does not happen and so we have achieved a contradiction.

(3) Suppose that v_p(\Delta_K) = 3. Then the relation \sum_{i} (e_i-1) f_i = 3 holds simultaneously with \sum_{i} e_i f_i = 4 in K. We deduce that p totally ramifies in K, so we can write (p) = \mathfrak{p}^4 with \mathfrak{p} \subseteq \mathcal{O}_K. Since p is tamely ramified, the inertia groups I(\widetilde{\mathfrak{P}}/p) are cyclic of order a multiple of 4. The only such subgroups of G are C_4, C_8. We know that every C_4, C_8 in G intersects any copy of D_6 \cong \text{Gal}(\widetilde{L}/K) as C_2, so we deduce e(\widetilde{L}/\mathfrak{p}) = 2. Then e(\widetilde{L}/p) = 8 and I(\widetilde{\mathfrak{P}}/p) \cong C_8 for every prime \widetilde{\mathfrak{P}} in L lying above p. We also know that C_8 \cap S_3 = \{1\} for every copy of C_8, S_3 in G (note that S_3 \cong \text{Gal}(\widetilde{L}/L)). This guarantees that for any prime \mathfrak{P} in L above \mathfrak{p}, we have e(\widetilde{L}/\mathfrak{P}) = 1. It follows that \mathfrak{p} ramifies in L/K, and so p \mid \operatorname{N}(\Delta_{L/K}).

The type of estimate we make here essentially appears in [MR2904935]. In addition to the primes dividing N_1, such a field L is potentially ramified at the primes dividing 6\Delta_K, but no others. Therefore, using the same idea as (Proof), the number of such extensions is bounded by a constant times \left|\operatorname{Cl}(K)[2]\right| times the number of squarefree ideals \mathfrak{a} in K supported at primes \mathfrak{p} dividing N_0 N_1 N_2. The number of squarefree \mathfrak{a} supported at primes dividing N_0 N_2 is bounded by 2^{4\omega(6\Delta_K)}. By Lemma 6.2, we must have that N_1 is a 4-th power, and thus the number of N_1\leq Y is bounded by \sum_{m\leq Y^{1/4}}a_{m^4}(K). Here the values a_n(K)\leq \tau(n)^4 are the coefficients of the Dedekind zeta function of K, which represent the number of ideals in K with norm n. Therefore

In [BCT], the analogous result for S_5-quintic fields is proved. We mimic the same proof, but use the parametrization for quartic fields given in [MR2183288].

For a G_{\mathbb{Z}}-invariant set S \subseteq V_{\mathbb{Z}}, let \operatorname{\mathcal{N}}(S,X) denote the number of absolutely irreducible G_{\mathbb{Z}}-orbits of S having discriminant bounded by X. For a suitable box H (which we determine later), with characteristic function \Phi = \Phi_H, we can sum through the three nondegenerate orbits of G_{\mathbb{R}} on V_{\mathbb{R}} to obtain

where M_i = M_i(H) is a constant depending only on i and H. Even if S \subseteq V_{\mathbb{Z}} is not invariant under the action of G_{\mathbb{Z}}, we define \operatorname{\mathcal{N}}(S,X) to be the quantity in Equation (?). Following Lemma 7 of [BCT], we fix any of the 12 coordinates of V_{\mathbb{Z}} as y and let \mathcal{Y} \subseteq \mathbb{A}_{\mathbb{Z}}^{12} be defined by

which is a subscheme of \mathbb{A}_{\mathbb{Z}}^{12} of codimension 2.

We know by the discussion in Section 4.2 of [bhargava2014geometricsievedensitysquarefree] that for v \in V_{\mathbb{Z}} absolutely irreducible, if p^2 \mid \operatorname{Disc}(v) then \operatorname{Disc} is strongly a multiple of p^2 at v. Moreover, by Lemma 3.6 of [bhargava2014geometricsievedensitysquarefree], we know that this implies v \allowbreak\mkern5mu({\operator@font mod}\,\,p) \in \mathcal{Y}(\mathbb{F}_p). Thus, it suffices for our purposes to bound the quantity \operatorname{\mathcal{N}}(S_Z(\mathcal{Y}), X), where we define S_Z(\mathcal{Y}) to be the intersection of the set

with the set of all absolutely irreducible elements of V_{\mathbb{Z}}. By Lemma 11 of [MR2183288], the contribution from v = (A,B) with a_{11} = 0 is O(X^{11/12}), so

Then we have by Equation 7 of [MR2183288] that

Now let

Then the compact set

contains KN'H. Equation (?) then becomes

where \Phi' denotes the characteristic function for H'. We recall that \lambda a(s)^{-1}(A,B) \in H' only if (A,B) \in B_0, where B_0 is the box defined in Equation 10 of [MR2183288]. Thus, we have

where \Phi_0 is the characteristic function for B_0. Each side length of B_0 can be written in the form r_i' \cdot \frac{120}{\lambda}, and so we can invoke Theorem 13 of [BCT] to bound the integrand with n = 12, k = 2, r_i = r_i'/\lambda, and B a centrally symmetric hypercube of side length 120. We note that the product of the r_i' is equal to 1, so T = \lambda^{-12}, which gives

Here we have used r_i' \gg s_1^{-1}s_2^{-4} s_3^{-2}, which follows from the inequalities defining B_0 and the fact that the s_i are absolutely bounded from below. Plugging this back into the integral gives

Thus \operatorname{\mathcal{N}}(S_Z(\mathcal{Y}), X) \ll X^{11/12 + \epsilon} + X/Z^{1-\varepsilon},

as desired.

(1) Since v_p(\Delta_K) \leq 3 and \text{rad}(D_0) = \text{rad}(N_0), the first inequality is immediate. If v_p(N_0) \in \{1,2,3\}, the second inequality is proved. To this end, we note that v_p(N_0) \neq 4 since the indices of all permutations in \text{8T40} are \{2,4,8\}.

(2) It is enough to show that if v_p(\Delta_K) \notin \{0,2\}, then v_p(\operatorname{N}(\Delta_{L/K})) > 0. If v_p(\Delta_K) = 3, then the above lemma gives us the desired inequality. So now assume that v_p(\Delta_K) = 1. This implies that there is one ramified prime in K lying over p having inertia degree 1; that is, \sigma_K(p) \in \{ (1^2 11), (1^2 2)\}. Suppose towards a contradiction that v_p(\operatorname{N}(\Delta_{L/K})) = 0. Then v_p(\Delta_L) = 2 and the inertia groups I(\widetilde{\mathfrak{P}}/p) are generated by a permutation \tau of index 2 in 8\text{T}40. All such permutations have order two, so e(\widetilde{L}/p) = 2. However, for every such \langle \tau \rangle, one can check that for any D \leq G containing \langle \tau \rangle as a normal subgroup, the intersection of D with any C_2 \times S_4 \cong \text{Gal}(\widetilde{L}/K) is nontrivial. We immediately conclude that for any prime ideal \mathfrak{p} in K above p, \mathfrak{p} ramifies in \widetilde{L}/K. However, this is a contradiction, as e(\widetilde{L}/p) = 2 and there is a prime ideal of K above p that is already ramified.