Hostname: page-component-6dbcb7884d-64ff6 Total loading time: 0 Render date: 2025-02-14T01:08:33.962Z Has data issue: false hasContentIssue false

Average Analytic Ranks of Elliptic Curves over Number Fields

Published online by Cambridge University Press:  11 February 2025

Tristan Phillips*
Affiliation:
Departement of Mathematics, Dartmouth College, 29 North Main Street, Hanover, NH, 03755-3551, United States

Abstract

We give a conditional bound for the average analytic rank of elliptic curves over an arbitrary number field. In particular, under the assumptions that all elliptic curves over a number field K are modular and have L-functions which satisfy the Generalized Riemann Hypothesis, we show that the average analytic rank of isomorphism classes of elliptic curves over K is bounded above by $(9\deg (K)+1)/2$, when ordered by naive height. A key ingredient in the proof is giving asymptotics for the number of elliptic curves over an arbitrary number field with a prescribed local condition; these results are obtained by proving general results for counting points of bounded height on weighted projective stacks with a prescribed local condition, which may be of independent interest.

Type
Number Theory
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution, and reproduction in any medium, provided the original work is properly cited.
Copyright
© The Author(s), 2025. Published by Cambridge University Press

1. Introduction

1.1. Average analytic ranks of elliptic curves

Let E be an elliptic curve over a number field K. The Mordell–Weil theorem states that the set of K-rational points $E(K)$ of E forms a finitely generated abelian group $E(K)\cong E(K)_{\text {tor}}\oplus {\mathbb Z}^r$ , where $E(K)_{\text {tor}}$ is the finite torsion subgroup and $r\in {\mathbb Z}_{\geq 0}$ is the rank. The study of ranks of elliptic curves has become a central topic in number theory. Despite this attention, ranks remain a mystery in many ways.

Birch and Swinnerton-Dyer [Reference Birch and Swinnerton-DyerBSD65] famously conjectured that the rank of an elliptic curve equals its analytic rank (i.e., the order of vanishing of its L-function at $s=1$ ).

To get an understanding of ranks of elliptic curves in general, one can hope to determine the average rank of elliptic curves. As there are infinitely many elliptic curves, in order to make sense of this average, one must order elliptic curves in some way. Throughout this article, we order elliptic curves by their naive height, as defined at the end of Section 2. The average analytic rank of elliptic curves over the rational numbers was first bounded by Brumer [Reference BrumerBru92], who gave an upper bound of $2.3$ under the assumption of the Generalized Riemann Hypothesis for elliptic L-functions.Footnote 1 Under the same conditions, this bound was improved to $2$ by Heath-Brown [Reference Heath-BrownHB04], and then to $25/14\approx 1.8$ by Young [Reference YoungYou06]. In a remarkable series of papers [Reference Bhargava and ShankarBS15a, Reference Bhargava and ShankarBS15b, Reference Bhargava and ShankarBS13a, Reference Bhargava and ShankarBS13b], Bhargava and Shankar bounded the average size of $2$ -, $3$ -, $4$ - and $5$ -Selmer groups of elliptic curves, leading to an unconditional upper bound of $0.885$ for the average rank of elliptic curves over ${\mathbb Q}$ .

Surprisingly little is known about average ranks of elliptic curves over number fields beyond ${\mathbb Q}$ . In Shankar’s doctoral thesis [Reference ShankarSha13], he extends his work with Bhargava on bounding $2$ -Selmer groups to show that the average rank of elliptic curves over number fields is bounded above by $1.5$ . However, Shankar’s result counts reduced Weierstrass equations, which differs from counting isomorphism classes. In particular, counting reduced Weierstrass equations coincides with counting isomorphism classes of elliptic curves only for the finitely many number fields with unit group $\{\pm 1\}$ and class number $1$ – namely, ${\mathbb Q}$ and the imaginary quadratic extensions ${\mathbb Q}(\sqrt {-d})$ for $d\in \{2, 7, 11, 19, 43, 67, 163\}$ . Outside of these cases, there can be multiple reduced Weierstrass equations within the same isomorphism class of elliptic curves.

In this article, we prove a conditional bound for the average analytic rank of isomorphism classes of elliptic curves over an arbitrary number field. This appears to be the first known bound on average ranks of isomorphism classes of elliptic curves over arbitrary number fields, as well as the first bound for average analytic ranks of elliptic curves over number fields other than ${\mathbb Q}$ .

Theorem 1.1.1. Let K be a number field of degree d. Assume that all elliptic curves over K are modular and that their L-functions satisfy the Generalized Riemann Hypothesis. Then, the average analytic rank of isomorphism classes of elliptic curves over K, when ordered by naive height, is bounded above by $(9d+1)/2$ .

One of the main difficulties in extending methods for counting elliptic curves over the rational numbers to elliptic curves over more general number fields is that one may no longer have a bijection between reduced short Weierstrass models and isomorphism classes of elliptic curves. We overcome this difficulty by exploiting the geometry of the moduli stack of elliptic curves. In particular, the (compactified) moduli stack of elliptic curves is isomorphic to the weighted projective stack ${\mathcal P}(4,6)$ . From this perspective, questions about counting elliptic curves of bounded height turn into questions about counting points of bounded height on weighted projective stacks. Such questions can then be studied using various tools from Diophantine geometry.

One of the key ingredients in the proof of Theorem 1.1.1 is estimating the number of elliptic curves with prescribed local conditions over number fields. To state our results on counting elliptic curves, we introduce some notation. Let K be a number field with ring of integers ${\mathcal O}_K$ , discriminant $\Delta _K$ , class number $h_K$ , regulator $R_K$ , and which contains $\varpi _K$ roots of unity. Let $\text {Val}(K)$ denote the set of places of K, let $\text {Val}_0(K)$ denote the set of finite places, and let $\text {Val}_\infty (K)$ denote the set of infinite places. Let $\zeta _K$ denote the Dedekind zeta function of K. Let $H(\cdot )$ denote the Hurwitz-Kronecker class number (see Subsection 5.1 for the definition). For a prime ${\mathfrak p}\subset {\mathcal O}_K$ , above a rational prime $p\in {\mathbb Z}$ , we will write $\deg ({\mathfrak p})$ for the index of fields $[{\mathcal O}_K/{\mathfrak p}:{\mathbb Z}/p]$ , which we will refer to as the degree of ${\mathfrak p}$ .

Theorem 1.1.2. Let K be a degree d number field, and let ${\mathfrak p}$ be a prime ideal of ${\mathcal O}_K$ of norm q such that $2\nmid q$ and $3\nmid q$ . Let ${\mathscr L}$ be one of the local conditions listed in Table 1. Then the number of elliptic curves over K with naive height less than B, and which satisfy the local condition ${\mathscr L}$ at ${\mathfrak p}$ , is

$$\begin{align*}\kappa\kappa_{\mathscr L} B^{5/6}+O\,\left(\epsilon_{\mathscr L} B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align*}$$

where

$$\begin{align*}\kappa=\frac{h_K \left(2^{r_1+r_2}\pi^{r_2}\right)^{2}R_K10^{r_1+r_2-1}\gcd(2,\varpi_K)}{\varpi_{K} |\Delta_K|\zeta_K(10)}, \end{align*}$$

and where $\kappa _{\mathscr L}$ and $\epsilon _{\mathscr L}$ are as in Table 1.

Table 1 Local conditions.

We also give asymptotics for counting elliptic curves with a prescribed trace of Frobenius:

Theorem 1.1.3. Let K be a degree d number field, and let ${\mathfrak p}$ be a degree n prime ideal of ${\mathcal O}_K$ above a rational prime $p>3$ . Set $q=N_{K/{\mathbb Q}}({\mathfrak p})=p^n$ , the norm of ${\mathfrak p}$ . Let $a\in {\mathbb Z}$ be an integer satisfying $|a|\leq 2\sqrt {q}$ . Then, the number of elliptic curves over K with naive height less than B, good reduction at ${\mathfrak p}$ , and which have trace of Frobenius a at ${\mathfrak p}$ , is

$$\begin{align*}\kappa\kappa_{n,a} B^{5/6}+O\,\left(\epsilon_{n,a} B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align*}$$

where

$$\begin{align*}\kappa=\frac{h_K \left(2^{r_1+r_2}\pi^{r_2}\right)^{2}R_K10^{r_1+r_2-1}\gcd(2,\varpi_K)}{\varpi_{K} |\Delta_K|\zeta_K(10)}, \end{align*}$$

and where $\kappa _{n,a}$ and $\epsilon _{n,a}$ are as in Table 2.

Table 2 Constants for Theorem 1.1.3.

Theorem 1.1.2 and Theorem 1.1.3 generalize a result of Cho and Jeong [Reference Cho and JeongCJ23b, Theorem 1.4] for elliptic curves over ${\mathbb Q}$ to elliptic curves over arbitrary number fields. For more results on counting elliptic curves with prescribed local conditions, see the work of Cremona and Sadek [Reference Cremona and SadekCS23].

1.2. Counting points on weighted projective stacks

Our results for counting isomorphism classes of elliptic curves are special cases of a more general theorem for counting points of bounded height on weighted projective stacks.

In [Reference SchanuelSch79], Schanuel proved an asymptotic for the number of rational points of bounded height on projective spaces over number fields. This was generalized to weighted projective stacks by Deng [Reference DengDen98] (see also the work of Darda [Reference DardaDar21], which gives a proof using height zeta functions). In this paper, we extend these results to count points of bounded height which satisfy finitely many prescribed local conditions (see Theorem 4.0.5).

Remark 1.2.1. Let us briefly mention some additional results related to counting points of bounded height on weighted projective stacks. Recently, Bruin and Manterola Ayala [Reference AyalaMA21, Reference Bruin and AyalaBMA23] have generalized Deng’s result to count points whose images, with respect to a morphism of weighted projective stacks, are of bounded height. Using the geometric sieve, Bright, Browning and Loughran [Reference Bright, Browning and LoughranBBL16] have proven a generalization of Schanuel’s theorem which allows one to impose infinitely many local conditions on the points of projective space being counted. In [Reference PhillipsPhi22], the author combines these results to count points whose images, with respect to a morphism of weighted projective stacks, are of bounded height and satisfy prescribed local conditions. Such a result has applications to counting elliptic curves with certain prescribed level structures and local conditions. Building from this result, Cho, Jeong and Park [Reference Cho, Jeong and ParkCJP23] have given a conditional bound for average analytic ranks of isomorphism classes of elliptic curves over number fields with a prescribed level structure.

1.3. Asymptotic notation

Let $f:{\mathbb R}_{>0}\to {\mathbb R}$ and $g:{\mathbb R}_{>0}\to {\mathbb R}$ be real valued functions. Throughout the article, we will write $f=O(g)$ (or $f\ll g$ ) if there exist constants $C,D\in {\mathbb R}_{>0}$ such that for all $x\geq D$ , we have $|f(x)|\leq C\cdot g(x)$ . We refer to any such C as the implied constant. If the implied constant depends on a parameter t, then we will write $f=O_t(g)$ (or $f\ll _t g$ ).

1.4. Organization

In Section 2, we cover basic facts about weighted projective stacks and heights defined on weighted projective stacks. In Section 3, we prove a result for counting points of bounded height satisfying prescribed local conditions on affine spaces over number fields. In Section 4, we prove our main result for counting points of bounded height on weighted projective stacks (Theorem 4.0.5). In Section 5, we apply Theorem 4.0.5 to count elliptic curves with prescribed local conditions. This is where we prove Theorem 1.1.2 and Theorem 1.1.3. Finally, in Section 6, we use the explicit formula for L-functions, together with our results for counting elliptic curves with prescribed local conditions, to bound the average analytic rank of elliptic curves over number fields. This is where we prove Theorem 1.1.1.

2. Preliminaries on weighted projective stacks

In this section, we recall basic facts about weighted projective stacks with an emphasis on heights.

2.1. Heights on weighted projective stacks

Given an $(n+1)$ -tuple of positive integers ${\textbf {w}}=(w_0,\dots ,w_n)$ , the weighted projective stack ${\mathcal P}({\textbf {w}})$ is the quotient stack

$$\begin{align*}{\mathcal P}({\textbf{w}})\stackrel{\mathrm{def}}{=}[({\mathbb A}^{n+1}-\{0\})/\mathbb{G}_m], \end{align*}$$

where the multiplicative group scheme, $\mathbb {G}_m$ , acts on the punctured affine space, ${\mathbb A}^{n+1}-\{0\}$ , as follows:

$$ \begin{align*} \mathbb{G}_m\times({\mathbb A}^{n+1}-\{0\}) &\to ({\mathbb A}^{n+1}-\{0\})\\ (\lambda,(x_0,\dots,x_n)) &\mapsto \lambda\ast_{\textbf{w}} (x_0,\dots,x_n)\stackrel{\mathrm{def}}{=}(\lambda^{w_0}x_0,\dots, \lambda^{w_n}x_n). \end{align*} $$

In the special case when ${\textbf {w}}=(1,\dots ,1)$ , this recovers the usual projective space ${\mathbb P}^n$ .

The point $[(a_0,\dots ,a_n)]\in {\mathcal P}(w_0,\dots ,w_n)$ has stabilizer $\mu _m$ , where $m=\gcd (w_i : a_i\neq 0)$ . When K is a field of characteristic zero, we have that $\mu _m$ is finite and reduced over K. It follows that, over fields of characteristic zero, ${\mathcal P}(w_0,\dots ,w_n)$ is a Deligne–Mumford stack.

For any field F, let ${\mathcal P}({\textbf {w}})(F)$ denote the set of isomorphism classes of the groupoid of F-points of ${\mathcal P}({\textbf {w}})$ . More concretely, ${\mathcal P}({\textbf {w}})(F)$ is in canonical bijection with the quotient

$$\begin{align*}(F^{n+1}-\{0\})/F^\times, \end{align*}$$

where $F^\times $ acts on $F^{n+1}-\{0\}$ by the weighted action

$$ \begin{align*} F^\times \times (F^{n+1}-\{0\}) &\to (F^{n+1}-\{0\})\\ (\lambda, (a_0,\dots,a_n)) &\mapsto (\lambda^{w_0}a_0,\dots,\lambda^{w_n}a_n). \end{align*} $$

Throughout this article, unless otherwise stated, we assume all weighted projective stacks ${\mathcal P}({\textbf {w}})$ are defined over a number field K.

For each finite place $v\in \text {Val}_0(K)$ , let ${\mathfrak p}_v$ be the corresponding prime ideal, and let $\pi _v$ be a uniformizer for the completion $K_v$ of K at v. For $x=(x_0,\dots ,x_n)\in K^{n+1}-\{0\}$ , set $|x_i|_{{\textbf {w}},v}\stackrel {\mathrm{def}}{=}|\pi _v|_v^{\lfloor v(x_i)/w_i\rfloor }$ , and set

$$\begin{align*}|x|_{{{\textbf{w}}},v}\stackrel{\mathrm{def}}{=}\begin{cases} \max_i\{|x_i|_{{\textbf{w}},v}\} & \text{ if } v\in \text{Val}_0(K)\\ \max_i\{|x_i|_v^{1/w_i}\} & \text{ if } v\in \text{Val}_\infty(K). \end{cases} \end{align*}$$

Definition 2.1.1 (Height).

The (exponential) height of a point $x=[x_0:\cdots :x_n]\in {\mathcal P}({\textbf {w}})(K)$ is defined as

$$\begin{align*}\text{Ht}_{\textbf{w}}(x)\stackrel{\mathrm{def}}{=}\prod_{v\in \text{Val}(K)} |(x_0,\dots,x_n)|_{{\textbf{w}},v}. \end{align*}$$

It is straightforward to check that this height function does not depend on the choice of representative of x.

Definition 2.1.2 (Scaling ideal).

Define the scaling ideal ${\mathfrak I}_{{\textbf {w}}}(x)$ of $x=(x_0,\dots ,x_n)\in K^{n+1}-\{0\}$ to be the fractional ideal

$$\begin{align*}{\mathfrak I}_{{\textbf{w}}}(x)\stackrel{\mathrm{def}}{=}\prod_{v\in \text{Val}_0(K)} {\mathfrak p}_v^{\min_i\{\lfloor v(x_i)/w_i\rfloor\}}. \end{align*}$$

The scaling ideal ${\mathfrak I}_{{\textbf {w}}}(x)$ can be characterized as the intersection of all fractional ideals ${\mathfrak a}$ of ${\mathcal O}_K$ such that $x\in {\mathfrak a}^{w_0}\times \cdots \times {\mathfrak a}^{w_n}\subseteq K^{n+1}$ . It has the property that

$$\begin{align*}{\mathfrak I}_{\textbf{w}}(x)^{-1}=\{a\in K : a^{w_i}x_i\in {\mathcal O}_K \text{ for all } i\}. \end{align*}$$

The height can be written in terms of the scaling ideal as follows:

$$\begin{align*}\text{Ht}_{\textbf{w}}([x_0:\dots:x_n])=\frac{1}{N_{K/{\mathbb Q}}({\mathfrak I}_{\textbf{w}}(x))}\prod_{v\in \text{Val}_\infty(K)} \max_i\{|x_i|_v^{1/w_i}\}. \end{align*}$$

It is straightforward to check that this agrees with the height from Definition 2.1.1.

Remark 2.1.3. The height $\text {Ht}_{\textbf {w}}$ was first defined by Deng [Reference DengDen98]. In the case of projective spaces, ${\mathbb P}^n={\mathcal P}(1,\dots ,1)$ , this height corresponds to the usual Weil height. In more geometric terms, this height on weighted projective stacks can be viewed as the ‘stacky height’ associated to the tautological bundle of ${\mathcal P}({\textbf {w}})$ (see [Reference Ellenberg, Satriano and Zureick-BrownESZB23, §3.3] for this and much more about heights on stacks).

The (compactified) moduli stack of elliptic curves, ${\mathcal X}_{\text {GL}_2({\mathbb Z})}$ , is isomorphic to the weighted projective stack ${\mathcal P}(4,6)$ . This isomorphism can be given explicitly as

$$ \begin{align*} {\mathcal X}_{\text{GL}_2({\mathbb Z})} &\xrightarrow{\sim} {\mathcal P}(4,6)\\ y^2=x^3+Ax+B & \mapsto [A:B]. \end{align*} $$

Under this isomorphism, the tautological bundle on ${\mathcal P}(4,6)$ corresponds to the Hodge bundle on ${\mathcal X}_{\text {GL}_2({\mathbb Z})}$ . The usual naive height of an elliptic curve E over K with a reduced integral short Weierstrass equation $y^2=x^3+Ax+B$ is defined as

$$\begin{align*}\text{Ht}(E)\stackrel{\mathrm{def}}{=}\prod_{v\in \text{Val}_\infty(K)} \max\{|A|_v^3,|B|_v^2\}. \end{align*}$$

One can show that this naive height is the same as the stacky height on ${\mathcal X}_{\text {GL}_2({\mathbb Z})}$ with respect to the twelfth power of the Hodge bundle [Reference Ellenberg, Satriano and Zureick-BrownESZB23, §3.3]. For any short Weierstrass equation $y^2=x^3+ax+b$ of E, the naive height is given by

$$\begin{align*}\text{Ht}(E)\stackrel{\mathrm{def}}{=} \text{Ht}_{(4,6)}([a:b])^{12}. \end{align*}$$

3. Weighted geometry of numbers over number fields

Let K be a number field. In this section, we prove asymptotics for the number of K-rational points of bounded height with prescribed local conditions on affine spaces with respect to a weighting. For these asymptotics, we give a power savings error term which is independent of the local conditions.

3.1. O-minimal geometry

We briefly cover the basics of o-minimal geometry which will be needed in this article. A general reference for this subsection is [Reference van den DriesvdD98, §1]. Let $m_L$ denote Lebesgue measure on ${\mathbb R}^n$ .

Definition 3.1.1 (Semi-algebraic set).

A (real) semi-algebraic subset of ${\mathbb R}^n$ is a finite union of sets of the form

$$\begin{align*}\{x\in {\mathbb R}^n: f_1(x)=\cdots=f_k(x)=0 \text{ and } g_1(x)>0,\dots,g_l(x)>0\}, \end{align*}$$

where $f_1,\dots ,f_k,g_1,\dots ,g_l\in {\mathbb R}[X_1,\dots ,X_n]$ .

Definition 3.1.2 (Structure).

A structure is a sequence, ${\mathcal S}=({\mathcal S}_n)_{n\in {\mathbb Z}_{>0}}$ , where each ${\mathcal S}_n$ is a set of subsets of ${\mathbb R}^n$ with the following properties:

  1. (i) If $A,B\in {\mathcal S}_n$ , then $A\cup B\in {\mathcal S}_n$ and ${\mathbb R}^n - A\in {\mathcal S}_n$ (i.e., ${\mathcal S}_n$ is a Boolean algebra).

  2. (ii) If $A\in {\mathcal S}_m$ and $B\in {\mathcal S}_n$ , then $A\times B\in {\mathcal S}_{n+m}$ .

  3. (iii) If $\pi :{\mathbb R}^n\to {\mathbb R}^m$ is the projection to m distinct coordinates and $A\in {\mathcal S}_n$ , then $\pi (A)\in {\mathcal S}_m$ .

  4. (iv) All real semi-algebraic subsets of ${\mathbb R}^n$ are in ${\mathcal S}_n$ .

A subset is definable in ${\mathcal S}$ if it is contained in some ${\mathcal S}_n$ . Let $D\subseteq S_n$ . A function $f\colon D\to {\mathbb R}^m$ is said to be definable in ${\mathcal S}$ if its graph, $\Gamma (f)\stackrel {\mathrm{def}}{=}\{(x,f(x)):x\in D\}\subseteq {\mathbb R}^{m+n}$ , is definable in ${\mathcal S}$ .

Note that the intersection of definable sets is definable by property (i).

Definition 3.1.3 (O-minimal structure).

An o-minimal structure is a structure in which the following additional property holds:

  1. (v) The boundary of each set in ${\mathcal S}_1$ is a finite set of points.

The class of semialgebraic sets is an example of an o-minimal structure (see, for example, [Reference van den DriesvdD98, §2]). The main structure we will use is ${\mathbb R}_{\exp }$ , which is defined to be the smallest structure in which the real exponential function, $\exp \colon {\mathbb R}\to {\mathbb R}$ , is definable. Observe that the function $\log (x)$ is definable in ${\mathbb R}_{\exp }$ since its graph,

$$\begin{align*}\Gamma(\log(x))=\{(x,\log(x)) : x\in {\mathbb R}_{>0}\}=\{(\exp(y), y) : y\in {\mathbb R}\}\subset {\mathbb R}^2, \end{align*}$$

is clearly definable in ${\mathbb R}_{\exp }$ .

Theorem 3.1.4 [Reference WilkieWil96].

The structure ${\mathbb R}_{\exp }$ is o-minimal.

From now on, we will call a subset of ${\mathbb R}^n$ definable if it is definable in some o-minimal structure.

3.2. Weighted geometry-of-numbers

The following proposition is a version of the Principle of Lipschitz, which gives estimates for the number of lattice points in a weighted homogeneous space.

Proposition 3.2.1. Let $R\subset {\mathbb R}^{n}$ be a bounded set definable in an o-minimal structure. Let $\Lambda \subset {\mathbb R}^n$ be a rank n lattice with successive minima $\lambda _1,\dots ,\lambda _n$ (with respect to the origin-centered unit ball in ${\mathbb R}^n$ ). Set

$$\begin{align*}R(B)=B\ast_{{\textbf{w}}}R=\{B\ast_{{\textbf{w}}}x : x\in R\}. \end{align*}$$

Then, for sufficiently largeFootnote 2 B,

$$\begin{align*}\#(\Lambda\cap R(B))=\frac{m_L(R)}{\det \Lambda} B^{|{\textbf{w}}|} + O\,\left(\lambda_n\det(\Lambda)^{-1} B^{|{\textbf{w}}|-w_{\min}}\right), \end{align*}$$

where the implied constant depends only on R and ${\textbf {w}}$ .

Proof. This will follow from a general version of the Principle of Lipschitz due to Barroero and Widmer [Reference Barroero and WidmerBW14, Theorem 1.3]. Since $m_L(R(B))=B^{|{\textbf {w}}|} m_L(R)$ , [Reference Barroero and WidmerBW14, Theorem 1.3] gives the desired leading term. Let $V_j(R(B))$ denote the sum of the j-dimensional volumes of the orthogonal projections of $R(B)$ onto each j-dimensional coordinate subspace of ${\mathbb R}^n$ . The error term given in [Reference Barroero and WidmerBW14, Theorem 1.3] is

(1) $$ \begin{align} O\,\left(1+\sum_{j=1}^{n-1} \frac{V_j(R(B))}{\lambda_1\cdots \lambda_j}\right), \end{align} $$

where the implied constant depends only on R. In our case, one observes that $V_i(R(B))=O(V_{j}(R(B)))$ for all $i\leq j$ . Moreover, $V_{n-1}(R(B))=O(\sum _{i\leq n} B^{|{\textbf {w}}|-w_i})=O(B^{|{\textbf {w}}|-w_{\min }})$ , where the implied constant depends only on R and ${\textbf {w}}$ . By dimension considerations, we see that if $B^{|{\textbf {w}}|-w_{\min }}=O(V_i(R(B)))$ , then we must have $i\geq n-1$ . From these observations, it follows that

$$\begin{align*}O\,\left(1+\sum_{j=1}^{n-1} \frac{V_j(R(B))}{\lambda_1\cdots \lambda_j}\right)=O\,\left(\frac{V_{n-1}(R(B))}{\lambda_1\cdots\lambda_{n-1}}\right). \end{align*}$$

By Minkowski’s second theorem [Reference Minkowski and ApproximationenMin07] (for a more modern reference see, for example, [Reference CasselsCas59, p. 203]),

$$\begin{align*}\lambda_1\cdots\lambda_n\geq \frac{2^n}{n!\cdot m_L(\mathscr{B}_n)}\det(\Lambda), \end{align*}$$

where $\mathscr {B}_n$ is the unit ball in ${\mathbb R}^n$ . From this, we obtain the desired expression for the error term,

$$\begin{align*}O\,\left(\frac{V_{n-1}(R(B))}{\lambda_1\cdots\lambda_{n-1}}\right)=O\,\left(\frac{\lambda_n B^{|{\textbf{w}}|-w_{\min}}}{\det(\Lambda)}\right).\\[-44pt] \end{align*}$$

Definition 3.2.2 (Boxes).

A ( $\mathbf {{\mathbb Z}_p^n}$ )-box is a subset ${\mathcal B}_p\subset {\mathbb Z}_p^n$ for which there exist closed balls

$$\begin{align*}{\mathcal B}_{p,j}=\{x\in {\mathbb Z}_p : |x-a_j|_p\leq b_{p,j}\}\subset {\mathbb Z}_p, \end{align*}$$

where $a_j\in {\mathbb Z}_p$ and $b_{p,j}\in \{p^{-k}: k\in {\mathbb Z}_{\geq 0}\}$ , such that ${\mathcal B}_p$ equals the Cartesian product $\prod _{j=1}^n {\mathcal B}_{p,j}$ . Let S be a finite set of primes. An S-box is a subset ${\mathcal B}\subset \prod _{p\in S} {\mathbb Z}_p^n$ for which there exist ${\mathbb Z}_p^n$ -boxes ${\mathcal B}_p$ such that

$$\begin{align*}{\mathcal B}=\prod_{p\in S} {\mathcal B}_{p}=\prod_{p\in S}\prod_{j=1}^n {\mathcal B}_{p,j}. \end{align*}$$

Lemma 3.2.3 (Box Lemma).

Let $\Omega _\infty \subset {\mathbb R}^n$ be a bounded subset definable in an o-minimal structure. Let S be a finite set of primes and ${\mathcal B}=\prod _{p\in S} {\mathcal B}_p$ an S-box. Then, for each $B\in {\mathbb R}_{>0}$ , we have

$$ \begin{align*} &\#\left\{x\in {\mathbb Z}^n \cap \left(B\ast_{\textbf{w}} \Omega_\infty\right): x\in \prod_{p\in S} \mathcal{B}_p \right\}\\ &\hspace{1cm}=\left(m_L(\Omega_\infty)\prod_{p\in S} m_p(\mathcal{B}_p)\right)B^{|{\textbf{w}}|} + O\,\left(\max_j\left\{\prod_{p\in S} b_{p,j}^{-1}\right\}\left(\prod_{p \in S} m_p(\mathcal{B}_p)\right)B^{|{\textbf{w}}|-w_{\min}} \right), \end{align*} $$

where the implied constant depends only on $\Omega _\infty $ and ${\textbf {w}}$ .

Proof. We have that $\mathcal {B}_p=\prod _{j=1}^n \mathcal {B}_{p,j}$ with $\mathcal {B}_{p,j}=\{x\in {\mathbb Z}_p : |x-a_j|_p\leq b_{p,j}\}$ , where $a_j\in {\mathbb Z}_p$ and $b_{p,j}\in \{p^{-k} : k\in {\mathbb Z}_{\geq 0}\}$ . Observe that for each $j\in \{1,\dots ,n\}$ , the set ${\mathbb Z}\cap \mathcal {B}_{p,j}$ is a translate of the sublattice $\langle b_{p,j}^{-1}\rangle \subseteq {\mathbb Z}$ . By the Chinese Remainder Theorem, the set ${\mathbb Z}\cap \prod _{p\in S} \mathcal {B}_{p,j}$ is a translate of the sublattice $\left \langle \prod _{p\in S} b_{p,j}^{-1}\right \rangle \subseteq {\mathbb Z}$ . Taking a Cartesian product, we find that the set

$$\begin{align*}\prod_{j=1}^n\left( {\mathbb Z}\cap \prod_{p\in S} \mathcal{B}_{p,j}\right)={\mathbb Z}^n\cap \prod_{p\in S} \mathcal{B}_p \end{align*}$$

is a translate of a sub-lattice of ${\mathbb Z}^n$ whose successive minima are $\prod _{p\in S} b_{p,j}^{-1}$ . In particular, the n-th successive minimum of the sublattice is

$$\begin{align*}\max_j\left\{\prod_{p\in S} b_{p,j}^{-1}\right\}, \end{align*}$$

and the determinant of the sublattice is

$$\begin{align*}\prod_{j=1}^n\prod_{p\in S} b_{p,j}^{-1}=\prod_{p\in S}m_p(\mathcal{B}_p)^{-1}. \end{align*}$$

The lemma then follows from Proposition 3.2.1.

Let $\Lambda $ be a free ${\mathbb Z}$ -module of finite rank n. Set $\Lambda _\infty =\Lambda \otimes _{\mathbb Z} {\mathbb R}$ and $\Lambda _p=\Lambda \otimes _{\mathbb Z} {\mathbb Z}_p$ . Equip $\Lambda _\infty $ and $\Lambda _p$ with Haar measures $m_\infty $ and $m_p$ , respectively, normalized so that $m_p(\Lambda _p)=1$ for all but finitely many primes p. As $\Lambda _p\cong {\mathbb Z}_p^n$ , we define a ( $\mathbf {\Lambda _p)}$ -box to be a subset of $\Lambda _p$ isomorphic to a $({\mathbb Z}_p^n)$ -box.

Lemma 3.2.4. Let $\Lambda $ be a free ${\mathbb Z}$ -module of finite rank n, let $\Omega _\infty \subset \Lambda _\infty $ be a bounded definable subset, and, for each prime p in a finite subset S, let $\Omega _p=\prod _j \{x\in {\mathbb Z}_p : |x-a_{p,j}|_p\leq \omega _{p,j}\}\subset \Lambda _p$ be a $(\Lambda _p)$ -box. Let ${\textbf {w}}=(w_1,\dots ,w_n)$ be an n-tuple of positive integers. Then,

$$ \begin{align*} &\#\{x\in \Lambda\cap \left(B \ast_{\textbf{w}} \Omega_\infty\right): x\in \Omega_p \text{ for all primes } p\in S\}\\ &\hspace{1cm}= \left(\frac{m_\infty(\Omega_\infty)}{m_\infty(\Lambda_\infty/\Lambda)}\prod_{p\in S} \frac{m_p(\Omega_p)}{m_p(\Lambda_p)}\right)B^{|{\textbf{w}}|} + O\,\left(\max_j\left\{\prod_{p\in S} \omega_{p,j}^{-1}\right\}\left(\prod_{p\in S} \frac{m_p(\Omega_p)}{m_p(\Lambda_p)}\right)B^{|{\textbf{w}}|-w_{\min}}\right), \end{align*} $$

where the implied constant depends only on $\Lambda $ , $\Omega _\infty $ , and ${\textbf {w}}$ .

Proof. Fix an isomorphism $\Lambda \cong {\mathbb Z}^n$ . The measures $m_\infty $ and $m_p$ on $\Lambda _\infty $ and $\Lambda _p$ induce measures on ${\mathbb R}^n$ and ${\mathbb Z}_p^n$ which differ from the usual Haar measures by $m_\infty (\Lambda _\infty /\Lambda )$ and $m_p(\Lambda _p)$ , respectively. It therefore suffices to prove the result in the case $\Lambda ={\mathbb Z}^n$ , and $m_\infty =m_L$ and $m_p$ are the usual Haar measures; but this case is precisely the Box Lemma (Lemma 3.2.3).

If K is a number field of degree d over ${\mathbb Q}$ with discriminant $\Delta _K$ , then its ring of integers ${\mathcal O}_K$ may naturally be viewed as a rank d lattice in $K_\infty \stackrel {\mathrm{def}}{=}{\mathcal O}_K\otimes _{{\mathbb Z}} {\mathbb R}=\prod _{v|\infty } K_v$ . This lattice has covolume $|\Delta _K|^{1/2}$ with respect to the usual Haar measure $m_\infty $ on $K_\infty $ (which differs from Lebesgue measure on $K_\infty \cong {\mathbb R}^{r_1+2r_2}$ by a factor of $2^{r_2}$ ) [Reference NeukirchNeu99, Chapter I Proposition 5.2]. More generally, any nonzero integral ideal ${\mathfrak a}\subseteq {\mathcal O}_K$ may be viewed as a lattice in $K_\infty $ with covolume $N_{K/{\mathbb Q}}({\mathfrak a})|\Delta _K|^{1/2}$ . For an n-tuple of positive integers ${\textbf {w}}=(w_1,\dots ,w_n)$ , define the lattice

$$\begin{align*}{\mathfrak a}^{\textbf{w}}\stackrel{\mathrm{def}}{=}{\mathfrak a}^{w_1}\times\cdots \times {\mathfrak a}^{w_n}\subset K_\infty^{n}, \end{align*}$$

where we view each ${\mathfrak a}^{w_i}$ as a subset of $K_\infty $ . This lattice has covolume $N_{K/{\mathbb Q}}({\mathfrak a})^{|{\textbf {w}}|}|\Delta _K|^{n/2}$ . For example, in the case that ${\mathfrak a}={\mathcal O}_K$ and $w_i=1$ for all i, one has the lattice ${\mathfrak a}^{\textbf {w}}={\mathcal O}_K^{n}$ of covolume $|\Delta _K|^{n/2}$ .

For any rational prime p, we have ${\mathcal O}_K\otimes _{{\mathbb Z}} {\mathbb Z}_p=\prod _{{\mathfrak p}|p} {\mathcal O}_{K,{\mathfrak p}}$ . Equip each ${\mathcal O}_{K,{\mathfrak p}}$ with the Haar measure $m_{\mathfrak p}$ , normalized so that $m_{\mathfrak p}({\mathcal O}_{K,{\mathfrak p}})=1$ . This measure induces a measure on ${\mathcal O}_{K,{\mathfrak p}}^n$ , which we will also denote by $m_{\mathfrak p}$ . Similarly, the measure $m_\infty $ on $K_\infty $ induces a measure on $K_\infty ^n$ , which we will denote by $m_\infty $ .

For $v\in \text {Val}_0(K)$ a finite place, let $q_v$ denote the size of the residue field ${\mathcal O}_{K}/{\mathfrak p}_v$ . Let ${\mathfrak a}_{\mathfrak p}$ denote the image of ${\mathfrak a}$ in ${\mathcal O}_{K,{\mathfrak p}}$ .

Definition 3.2.5 (Local condition).

An (affine) local condition at a finite place $v\in \text {Val}_0(K)$ will refer to a subset $\Omega _v\subseteq {\mathcal O}_{K,{\mathfrak p}_v}^n$ . We call a local condition $\Omega _v$ irreducible if it is an ( ${\mathcal O}_{K,{\mathfrak p}_v}^n$ )-box – that is, if there exist $a_{j}\in {\mathcal O}_{K,{\mathfrak p}_v}$ and $\omega _{j}\in \{q_v^{-k} : k\in {\mathbb Z}_{\geq 0}\}$ such that $\Omega _v=\prod _j \{ x\in {\mathcal O}_{K,{\mathfrak p}_v} : |x-a_j|_v\leq \omega _j\}\subset {\mathcal O}_{K,{\mathfrak p}_v}^n$ .

Remark 3.2.6. Though our results will mostly concern irreducible local conditions, by taking unions and complements, one can easily extend these results to many other local conditions.

We now apply Lemma 3.2.4 to the rank $nd$ lattice ${\mathfrak a}^{\textbf {w}}$ , and with the $(nd)$ -tuple of weights $\tilde {{\textbf {w}}}=(w_1,\dots ,w_1,w_2,\dots ,w_2,\dots ,w_n,\dots ,w_n)$ , obtained by repeating each of the weights in the n-tuple ${\textbf {w}}$ exactly d times. Observing that $|\tilde {{\textbf {w}}}|=d|{\textbf {w}}|$ and $\tilde {w}_{\min }=w_{\min }$ , we obtain the following proposition:

Proposition 3.2.7. Let $\Omega _\infty \subset K_\infty ^{n}$ be a bounded definable subset. Let S be a finite set of prime ideals of ${\mathcal O}_K$ . For each ${\mathfrak p}\in S$ , let $\Omega _{\mathfrak p}=\prod _j \{x\in {\mathcal O}_{K,{\mathfrak p}} : |x-a_{{\mathfrak p},j}|_{\mathfrak p}\leq \omega _{{\mathfrak p},j}\}\subset {\mathcal O}_{K,{\mathfrak p}}^n$ be an irreducible local condition. Then,

$$ \begin{align*} \#\{x\in {\mathfrak a}^{\textbf{w}}\cap \left(B \ast_{\textbf{w}} \Omega_\infty\right) : x\in \Omega_{\mathfrak p} \text{ for all primes } {\mathfrak p}\in S\}=\kappa B^{d|{\textbf{w}}|}+ O\,\left(\epsilon B^{d|{\textbf{w}}|-w_{\min}} \right), \end{align*} $$

where

$$ \begin{align*} \kappa&=\frac{m_\infty(\Omega_\infty)}{N_{K/{\mathbb Q}}({\mathfrak a})^{|{\textbf{w}}|}|\Delta_K|^{n/2}} \left(\prod_{{\mathfrak p}\in S} \frac{m_{\mathfrak p}(\Omega_{\mathfrak p}\cap {\mathfrak a}_{\mathfrak p}^{\textbf{w}})}{m_{\mathfrak p}({\mathfrak a}_{\mathfrak p}^{\textbf{w}})}\right),\\ \epsilon&=\max_j\left\{\prod_{{\mathfrak p}\in S} \omega_{{\mathfrak p},j}^{-1}\right\}\left(\prod_{{\mathfrak p}\in S} \frac{m_{\mathfrak p}(\Omega_{\mathfrak p}\cap {\mathfrak a}_{\mathfrak p}^{\textbf{w}})}{m_{\mathfrak p}({\mathfrak a}_{\mathfrak p}^{\textbf{w}})}\right), \end{align*} $$

and where the implied constant depends only on K, $\Omega _\infty $ and ${\textbf {w}}$ .

4. Counting points on weighted projective stacks

In this section, we prove our results for counting points of bounded height on weighted projective stacks.

Definition 4.0.1 (Local condition).

A (projective) local condition at a place $v\in \text {Val}(K)$ will refer to a subset $\Omega _v\subseteq \mathcal {P}({\textbf {w}})(K_v)$ .

For a projective local condition $\Omega _v\subseteq {\mathcal P}({\textbf {w}})(K_v)$ , we denote the affine cone of $\Omega _v$ by $\Omega _v^{\mathrm{aff}}\subseteq ({\mathbb A}^{n+1}-\{0\})(K_v)$ (i.e., $\Omega _v^{\mathrm{aff}}$ is the preimage of $\Omega $ with respect to the map $({\mathbb A}^{n+1}-\{0\})\to {\mathcal P}({\textbf {w}})$ ). For $v\in \text {Val}_0(K)$ , the set $\Omega _v^{\mathrm{aff}}\cap {\mathcal O}_{K,{\mathfrak p}_v}^{n+1}$ is an affine local condition, but not an irreducible affine local condition (see Definition 3.2.5).

Example 4.0.2. Let $v\in \text {Val}_0(K)$ be a finite place. Consider the projective local condition

$$\begin{align*}\Omega_v=\{[a:b]\in {\mathcal P}(1,1)(K_v):v(a)=0,\ v(b)=1\}. \end{align*}$$

Then,

$$\begin{align*}\Omega_v^{\mathrm{aff}}\cap {\mathcal O}_{K,v}^2=\{(\lambda a, \lambda b)\in {\mathcal O}_{K,v}^2 : v(a)=0,\ v(b)=1,\ \lambda\in K_v^\times\}. \end{align*}$$

Though this is not an irreducible affine local condition, it contains the irreducible affine local condition

$$\begin{align*}\Omega_{v,0}^{\mathrm{aff}}\stackrel{\mathrm{def}}{=}\{(a,b)\in {\mathcal O}_{K,v}^2 : v(a)=0,\ v(b)=1\}. \end{align*}$$

Letting $\pi _v$ be a uniformizer of ${\mathcal O}_{K,v}$ , we have

$$\begin{align*}\Omega_v^{\mathrm{aff}}\cap {\mathcal O}_{K,v}^2=\bigcup_{t\geq 0}\left(\pi_v^t\ast_{(1,1)} \Omega_{v,0}^{\mathrm{aff}}\right), \end{align*}$$

where $\pi _v^t \ast _{(1,1)} \Omega _{v,0}^{\mathrm{aff}}$ is an irreducible affine local condition for each t.

This example motivates the following definition:

Definition 4.0.3 (Irreducible projective local condition).

A projective local condition $\Omega _v\subseteq {\mathcal P}({\textbf {w}})(K_v)$ is said to be irreducible if there exists an irreducible affine local condition $\Omega _{v,0}^{\mathrm{aff}}$ such that

$$\begin{align*}\Omega_v^{\mathrm{aff}}\cap {\mathcal O}_{K,v}^{n+1}=\bigcup_{t\geq 0}\left(\pi_v^t\ast_{{\textbf{w}}} \Omega_{v,0}^{\mathrm{aff}}\right). \end{align*}$$

In analogy to the archimedean setting, for any ideal ${\mathfrak a}_v\subseteq {\mathcal O}_{K,v}$ , set

$$\begin{align*}{\mathfrak a}_v^{\textbf{w}}\stackrel{\mathrm{def}}{=} {\mathfrak a}_v^{w_0}\times\cdots \times {\mathfrak a}_v^{w_n}\subset K_v^{n+1}. \end{align*}$$

Proposition 4.0.4. Let $\Omega _v\subsetneq {\mathcal P}({\textbf {w}})(K_v)$ be a nontrivial irreducible projective local condition with

$$\begin{align*}\Omega^{\mathrm{aff}}_{v,0}=\prod_{j=0}^n\{x\in{\mathcal O}_{K,v} : |x-a_{j}|_v\leq \omega_j\}. \end{align*}$$

For $t\in {\mathbb Z}_{\geq 0}$ , set $\Omega ^{\mathrm{aff}}_{v,t}=\pi _v^t\ast _{\textbf {w}} \Omega ^{\mathrm{aff}}_{v,0}$ . Then we have the following:

  1. (a) There exists a j for which $|a_j|_v> \omega _j$ .

  2. (b) For $t\neq s$ , we have $\Omega ^{\mathrm{aff}}_{v,t}\cap \Omega ^{\mathrm{aff}}_{v,s}=\emptyset $ .

  3. (c) For $s,t\in {\mathbb Z}_{\geq 0}$ with $s<t$ , we have $\Omega _{v,s}^{\mathrm{aff}}\cap (\pi _v^{t})^{{\textbf {w}}}=\emptyset $ .

Proof. (a) We first prove the contrapositive of part (a). Suppose that $|a_j|_v\leq \omega _j$ for all j. If each $\omega _j=1$ , then $\Omega ^{\mathrm{aff}}_{v,0}={\mathcal O}_{K,v}^{n+1}=\Omega _{v}^{\mathrm{aff}}\cap {\mathcal O}_{K,v}^{n+1}$ , but then $\Omega _v={\mathcal P}({\textbf {w}})(K_v)$ would be the trivial projective local condition. Therefore, there must exist an index j for which $\omega _j\neq 1$ . By possibly reordering, we may assume $\omega _1\neq 1$ . As $|a_1|_v\leq \omega _1<1$ , we have that $|1-a_1|_v=1>\omega _1$ (by the ultrametric triangle inequality). Let $b\in {\mathcal O}_{K,v}-\{0\}$ be such that $|b-a_1|_v\leq \omega _1$ . Then $(b,0,\dots ,0)$ and $(1,0,\dots ,0)$ are points in the affine cone $K_v^{n+1}-\{0\}$ of ${\mathcal P}({\textbf {w}})(K_v)$ , each above the point $[1:0:\cdots :0]\in {\mathcal P}({\textbf {w}})(K_v)$ , but only one is contained in $\Omega _v^{\mathrm{aff}}$ . We conclude that $\Omega _v^{\mathrm{aff}}$ cannot be the entire subset of the affine cone above a projective local condition.

(b) We now show that if $|a_j|_v> \omega _j$ for some j, then for $t\neq s$ , we have $\Omega ^{\mathrm{aff}}_{v,t}\cap \Omega ^{\mathrm{aff}}_{v,s}=\emptyset $ . Observe that

$$\begin{align*}\Omega^{\mathrm{aff}}_{v,t}=\prod_{j=0}^n \{x\in {\mathcal O}_{K,v} : |x-\pi_v^{tw_j} a_j|_v\leq q_v^{-tw_j}\omega_j\} \end{align*}$$

is a product of v-adic balls centered at $\pi _v^{tw_j}a_j$ with radius $q_v^{-tw_j}\omega _j$ . Without loss of generality, we may assume $s< t$ . Using $|a_j|_v>\omega _j$ , we have that

$$\begin{align*}|\pi_v^{tw_j}a_j-\pi_v^{sw_j}a_j|_v=q_v^{-sw_j}\cdot |a_j|_v>q_v^{-sw_j} \omega_j>q_v^{-tw_j}\omega_j. \end{align*}$$

In particular, the distance between the centers of the balls

$$\begin{align*}\{x\in {\mathcal O}_{K,v} : |x-\pi_v^{tw_j} a_j|_v\leq q_v^{-tw_j}\omega_j\} \text{ and } \{x\in {\mathcal O}_{K,v} : |x-\pi_v^{sw_j} a_j|_v\leq q_v^{-sw_j}\omega_j\} \end{align*}$$

is larger than either of the radii, and thus, the balls must be disjoint (since we are in an ultrametric space).

(c) We will show that if $\Omega _{v,s}^{\mathrm{aff}}\cap (\pi _v^{t})^{{\textbf {w}}}$ is nonempty, then $\Omega _v$ is not a projective local condition. Let $y=(y_0,\dots ,y_n)\in \Omega _{v,s}^{\mathrm{aff}}\cap (\pi _v^{t})^{{\textbf {w}}}$ and let $y'=(y^{\prime }_0,\dots ,y^{\prime }_n)\in {\mathcal O}_{K,v}^{n+1}$ be such that $y=\pi _v^t\ast _{\textbf {w}} y'$ . It will suffice to show that $y'$ is not in $\Omega ^{\mathrm{aff}}_v$ .

By part (a), we may choose j to be such that $|a_j|_v>\omega _j$ . From this, together with the fact that $y_j\in \pi _v^{t w_j}$ satisfies $|y_j-\pi _v^{sw_j} a_j|_v\leq q_v^{-sw_j}\omega _j$ , implies that we must have $|y_j|_v=q_v^{-sw_j}|a_j|_v$ . Therefore, $|y^{\prime }_j|_v=q_v^{(t-s)w_j}|a_j|_v$ . Since $s<t$ , this implies that $|y^{\prime }_j|_v\neq |\pi _v^{rw_j}a_j|_v$ for any $r\in {\mathbb Z}_{\geq 0}$ , and thus, $y'$ is not contained in $\Omega ^{\mathrm{aff}}_v\cap {\mathcal O}_{K,v}^{n+1}=\bigcup _t \Omega ^{\mathrm{aff}}_{v,t}$ .

For any set of projective local conditions $(\Omega _v)_{v}$ , define the sets

$$ \begin{align*} \Omega&\stackrel{\mathrm{def}}{=}\{x\in {\mathcal P}({\textbf{w}})(K) : x\in \Omega_v \text{ for all } v\in \text{Val}(K)\} ,\\ \Omega_\infty&\stackrel{\mathrm{def}}{=}\{x\in {\mathcal P}({\textbf{w}})(K) : x\in \Omega_v \text{ for all } v\in \text{Val}_\infty(K)\},\\ \Omega_0&\stackrel{\mathrm{def}}{=}\{x\in {\mathcal P}({\textbf{w}})(K) : x\in \Omega_v \text{ for all } v\in \text{Val}_0(K)\}. \end{align*} $$

Set $\varpi _{K,{\textbf {w}}}\stackrel {\mathrm{def}}{=} \varpi _K/\gcd (\varpi _K,w_0,\dots ,w_n)$ , where we recall that $\varpi _K$ is the number of roots of unity in K.

Theorem 4.0.5. Let K be a degree d number field over ${\mathbb Q}$ . Let $S\subset \text {Val}_0(K)$ be a finite set of finite places. For each $v\in \text {Val}_\infty (K)\cup S$ , let $\Omega _v\subset {\mathcal P}({\textbf {w}})(K_v)$ be a projective local condition. Suppose that $\Omega ^{\mathrm{aff}}_\infty $ is a bounded definable subset of $K_\infty ^{n+1}$ and that $\Omega _v$ is irreducible for all $v\in S$ with

$$\begin{align*}\Omega_{v,0}^{\mathrm{aff}}=\prod_{j=0}^n\{x\in {\mathcal O}_{K,v}: |x-a_{v,j}|_v\leq \omega_{v,j}\}, \end{align*}$$

where $a_{v,j}\in {\mathcal O}_{K,v}$ and $\omega _{v,j}\in \{q_v^{-k}:k\in {\mathbb Z}_{\geq 0}\}$ (see Definitions 4.0.3 and 3.2.5). Then,

$$ \begin{align*} \#\{x\in {\mathcal P}({\textbf{w}})(K): \text{Ht}_{{\textbf{w}}}(x)\leq B, x\in \Omega\} = \kappa B^{|{\textbf{w}}|} + \begin{cases} O\,\left(\epsilon_\Omega B\log(B)\right) & {\substack{\text{ if } {\textbf{w}}=(1,1)\\ \text{and}\ K={\mathbb Q},}}\\ O\,\left(\epsilon_\Omega B^{\frac{d|{\textbf{w}}|-w_{\min}}{d}}\right) & \text{ else, } \end{cases} \end{align*} $$

where the leading coefficient is

$$\begin{align*}\kappa=\kappa_\Omega\frac{h_K \left(2^{r_1+r_2}\pi^{r_2}\right)^{n+1}R_K|{\textbf{w}}|^{r_1+r_2-1}}{\varpi_{K,{\textbf{w}}}|\Delta_K|^{(n+1)/2}\zeta_K(|{\textbf{w}}|)}, \end{align*}$$

where

$$\begin{align*}\kappa_\Omega=\prod_{v|\infty} \frac{m_v(\{x\in \Omega_v^{\mathrm{aff}} : \text{Ht}_{{\textbf{w}},v}(x)\leq 1\})}{m_v(\{x\in K_v^{n+1}: \text{Ht}_{{\textbf{w}},v}(x)\leq 1\})} \prod_{v\in S} m_v(\Omega_v^{\mathrm{aff}}\cap {\mathcal O}_{K,v}^{n+1}), \end{align*}$$

where the factor in the error term is

$$\begin{align*}\epsilon_{\Omega}=\sum_{t=0}^\infty \max\limits_j\left\{\prod\limits_{v\in S} \frac{N_{K/{\mathbb Q}}({\mathfrak p}_v)^{tw_j}}{\omega_{v,j}}\right\}\frac{1}{N_{K/{\mathbb Q}}({\mathfrak p}_v)^{t|{\textbf{w}}|}}\prod\limits_{v\in S} m_v(\Omega_{v,0}^{\mathrm{aff}}), \end{align*}$$

and where the implied constant is independent of the local conditions $\Omega _{v}$ with $v\in S$ .

Proof. Let ${\mathfrak c}_1,\dots ,{\mathfrak c}_h$ be a set of integral ideal representatives of the ideal class group of K. Then, we get the following partition of ${\mathcal P}({\textbf {w}})(K)$ into points whose scaling ideals are in the same ideal class:

$$\begin{align*}{\mathcal P}({\textbf{w}})(K)=\bigsqcup_{i=1}^h \{x\in {\mathcal P}({\textbf{w}})(K): [{\mathfrak I}_{\textbf{w}}(x)]=[{\mathfrak c}_i]\}. \end{align*}$$

This is well defined since the ideal class of a scaling ideal $[{\mathfrak I}_{\textbf {w}}(x)]$ does not depend on the representative of x [Reference DengDen98, Proposition 3.3].

For each ${\mathfrak c}\in \{{\mathfrak c}_1,\dots ,{\mathfrak c}_h\}$ , consider the counting function

$$\begin{align*}M(\Omega,{\mathfrak c},B)\stackrel{\mathrm{def}}{=}\#\{x\in {\mathcal P}({\textbf{w}})(K): \text{Ht}_{\textbf{w}}(x)\leq B,\ [{\mathfrak I}_{\textbf{w}}(x)]=[{\mathfrak c}],\ x\in \Omega\}. \end{align*}$$

Note that

$$\begin{align*}\#\{x\in {\mathcal P}({\textbf{w}})(K): \text{Ht}_{{\textbf{w}}}(x)\leq B, x\in \Omega\}=\sum_{i=1}^{h} M(\Omega,{\mathfrak c}_i,B). \end{align*}$$

Following Schanuel [Reference SchanuelSch79], we shall find an asymptotic for $M(\Omega ,{\mathfrak c},B)$ by counting integral points in affine space and then moding out by an action of the unit group, followed by an action of principal ideals.

Consider the (weighted) action of the unit group ${\mathcal O}_K^\times $ on $K^{n+1}-\{0\}$ given by

$$\begin{align*}u\ast_{{\textbf{w}}}(x_0,\dots,x_n)=(u^{w_0}x_0,\dots,u^{w_n}x_n), \end{align*}$$

and let $(K^{n+1}-\{0\})/{\mathcal O}_K^\times $ denote the corresponding set of orbits. Let $\Omega ^{\mathrm{aff}}$ be the affine cone of $\Omega $ . We may describe $M(\Omega ,{\mathfrak c},B)$ in terms of ${\mathcal O}_K^\times $ -orbits of an affine cone. In particular, there is a bijection between

$$\begin{align*}\{[x_0:\cdots:x_n]\in {\mathcal P}({\textbf{w}})(K): \text{Ht}_{\textbf{w}}(x)\leq B,\ [{\mathfrak I}_{\textbf{w}}(x)]=[{\mathfrak c}],\ x\in \Omega\} \end{align*}$$

and

$$\begin{align*}\left\{[(x_0,\dots,x_n)]\in (K^{n+1}-\{0\})/{\mathcal O}_K^\times : \frac{\text{Ht}_{{\textbf{w}},\infty}(x)}{N_{K/{\mathbb Q}}({\mathfrak c})}\leq B,\ {\mathfrak I}_{\textbf{w}}(x)={\mathfrak c},\ x\in \Omega^{\mathrm{aff}}\right\} \end{align*}$$

given by

$$\begin{align*}[x_0:\dots:x_n] \mapsto [(x_0,\dots,x_n)], \end{align*}$$

and therefore,

$$\begin{align*}M(\Omega,{\mathfrak c},B)=\#\left\{x\in (K^{n+1}-\{0\})/{\mathcal O}_K^\times : \frac{\text{Ht}_{{\textbf{w}},\infty}(x)}{N_{K/{\mathbb Q}}({\mathfrak c})}\leq B,\ {\mathfrak I}_{\textbf{w}}(x)={\mathfrak c},\ x\in \Omega^{\mathrm{aff}}\right\}. \end{align*}$$

Our general strategy will be to first find an asymptotic for the counting function

$$\begin{align*}M'(\Omega,{\mathfrak c},B)\stackrel{\mathrm{def}}{=}\#\left\{x\in (K^{n+1}-\{0\})/{\mathcal O}_K^\times : \frac{\text{Ht}_{{\textbf{w}},\infty}(x)}{N_{K/{\mathbb Q}}({\mathfrak c})}\leq B,\ {\mathfrak I}_{\textbf{w}}(x)\subseteq{\mathfrak c},\ x\in \Omega^{\mathrm{aff}}\right\} \end{align*}$$

and then use Möbius inversion to obtain an asymptotic formula for $M(\Omega ,{\mathfrak c},B)$ .

We are now going to construct a fundamental domain for the ( ${\textbf {w}}$ -weighted) action of the unit group ${\mathcal O}_K^{\times }$ on $K_\infty ^{n+1}-\{0\}$ . This will be done using Dirichlet’s Unit Theorem. Let $\varpi (K)$ denote the set of roots of unity in K.

Theorem 4.0.6 (Dirichlet’s Unit Theorem).

The image $\Lambda $ of the map

$$ \begin{align*} \lambda\colon {\mathcal O}_K^\times &\to {\mathbb R}^{r_1+r_2}\\ u &\mapsto (\log |u|_v)_{v\in \text{Val}_\infty(K)} \end{align*} $$

is a rank $r\stackrel {\mathrm{def}}{=} r_1+r_2-1$ lattice in the hyperplane H defined by $\sum _{v\in \text {Val}_\infty (K)} x_v=0$ and $\ker (\lambda )=\varpi (K)$ .

For each $v\in \text {Val}_\infty (K)$ , define a map

$$ \begin{align*} \eta_v\colon K_v^{n+1}-\{0\} &\to {\mathbb R} \\ (x_0,\dots,x_n) &\mapsto \log\max_i |x_i|_v^{1/w_i}. \end{align*} $$

Combine these maps to obtain a single map

$$ \begin{align*} \eta\colon \prod_{v\in \text{Val}_\infty(K)} (K_v^{n+1}-\{0\}) &\to {\mathbb R}^{r_1+r_2}\\ x & \mapsto (\eta_v(x_v)). \end{align*} $$

Let H be the hyperplane in Dirichlet’s unit theorem, and let

$$ \begin{align*} \text{pr}\colon {\mathbb R}^{r_1+r_2} \to H \end{align*} $$

be the projection along the vector $(d_v)_{v\in \text {Val}_\infty (K)}$ , where $d_v=1$ if v is real and $d_v=2$ if v is complex. More explicitly,

$$\begin{align*}(\text{pr}(x))_v = x_v-\left(\frac{1}{d}\sum_{v'\in\text{Val}_\infty(K)} x_{v'}\right) d_v. \end{align*}$$

The reason for choosing this projection is to ensure that a certain expanding region (which we are about to construct) will be weighted homogeneous (see Lemma 4.0.7).

Let $\{u_1,\dots ,u_r\}$ be a basis for the image of ${\mathcal O}_K^\times $ in the hyperplane H, and let $\{\check {u}_1,\dots ,\check {u_r}\}$ be the dual basis. Then the set

$$\begin{align*}\tilde{{\mathcal F}}\stackrel{\mathrm{def}}{=}\{y\in H: 0\leq \check{u}_j(y)<1 \text{ for all }j\in\{1,\dots,r\}\} \end{align*}$$

is a fundamental domain for H modulo $\Lambda $ , and ${\mathcal F}\stackrel {\mathrm{def}}{=}(\text {pr}\circ \eta )^{-1}\tilde {{\mathcal F}}$ is a fundamental domain for the ( ${\textbf {w}}$ -weighted) action of ${\mathcal O}_K^\times $ on $\prod _{v\in \text {Val}_\infty (K)}(K_v^{n+1}-\{0\})$ .

Define the sets

$$\begin{align*}{\mathcal D}(B)\stackrel{\mathrm{def}}{=}\left\{x\in \prod_{v\in \text{Val}_\infty(K)} (K_v^{n+1}-\{0\}) : \text{Ht}_{{\textbf{w}},\infty}(x) \leq B\right\} \end{align*}$$

and ${\mathcal F}(B)\stackrel {\mathrm{def}}{=}{\mathcal F}\cap {\mathcal D}(B)$ . The sets ${\mathcal D}(B)$ are ${\mathcal O}_K^\times $ -stable, in the sense that if $u\in {\mathcal O}_K^{\times }$ and $x\in {\mathcal D}(B)$ , then $u\ast _{{\textbf {w}}} x\in {\mathcal D}(B)$ ; this can be seen by the following computation:

$$ \begin{align*} \text{Ht}_{{\textbf{w}},\infty}(u\ast_{\textbf{w}} x)&=\prod_{v|\infty} \max_i |u^{w_i}x_{v,i}|_v^{1/w_i}\\ &=\prod_{v|\infty} |u|_v \prod_{v|\infty}\max_i |x_{v,i}|_v^{1/w_i}\\ &=\prod_{v|\infty}\max_i |x_{v,i}|_v^{1/w_i}\\ &=\text{Ht}_{{\textbf{w}},\infty}(x). \end{align*} $$

Similarly, for any $t\in {\mathbb R}$ , we have that

$$\begin{align*}\text{Ht}_{{\textbf{w}},\infty}(t\ast_{\textbf{w}} x)=\prod_{v|\infty} \max_i |t^{w_i}x_{v,i}|_v^{1/w_i}=\prod_{v|\infty} |t|_v \prod_{v|\infty}\max_i |x_{v,i}|_v^{1/w_i}=|t|^{d}\text{Ht}_{{\textbf{w}},\infty}(x). \end{align*}$$

This shows that ${\mathcal D}(B)=B^{1/d}{\mathcal D}(1)$ for all $B>0$ .

However, ${\mathcal F}$ is stable under the weighted action of $t\in {\mathbb R}^\times $ , in the sense that $t\ast _{{\textbf {w}}} {\mathcal F}={\mathcal F}$ . To see this, note that for any $x\in \prod _{v|\infty } (K_v^{n+1}-\{0\})$ ,

$$\begin{align*}\eta(t\ast_{{\textbf{w}}} x)=(d_v)_{v|\infty}\log(|t|)+\eta(x). \end{align*}$$

Since the projection $\text {pr}$ is linear and annihilates the vector $(d_v)_{v|\infty }$ , we have that

$$\begin{align*}\text{pr}\circ\eta(t\ast_{{\textbf{w}}}x)=\text{pr}\circ\eta(x), \end{align*}$$

as desired. From our observations, we obtain the following lemma:

Lemma 4.0.7. The regions ${\mathcal F}(B)$ are weighted homogeneous, in the sense that ${\mathcal F}(B)=B^{1/d}\ast _{{\textbf {w}}} {\mathcal F}(1)$ for all $B>0$ .

We are now going to count lattice points in ${\mathcal F}(B)$ . In [Reference SchanuelSch79], this is done by using the classical Principle of Lipschitz [Reference DavenportDav51] (see also [Reference LangLan94, VI §2 Theorem 2]). One of the cruxes of Schanuel’s argument is verifying that a fundamental domain (analogous to our ${\mathcal F}(1)$ ) has Lipschitz parameterizable boundary, so that he can apply the Principle of Lipschitz. Though one can modify this part of Schanuel’s argument to work in our case, we will instead take a slightly different route, using an o-minimal version of the Principle of Lipschitz (Proposition 3.2.1). This allows one to give a more streamlined proof of this part of Schanuel’s argument, which may be useful in future generalizations.

Lemma 4.0.8. The set ${\mathcal F}(1)$ is bounded.

Proof. Let $\tilde {H}\subset \prod _{v|\infty } {\mathbb R}_v$ be the subset defined by $\sum _{v|\infty } x_v\leq 0$ . Note that ${\mathcal F}(1)=\eta ^{-1}(\tilde {H} \cap \text {pr}^{-1}(\tilde {{\mathcal F}}))$ . It follows from the definition of $\eta $ that, in order to show ${\mathcal F}(1)$ is bounded, it suffices to show that the set

$$\begin{align*}S\stackrel{\mathrm{def}}{=}\tilde{H} \cap \text{pr}^{-1}(\tilde{{\mathcal F}})\subset {\mathbb R}^{r_1+r_2} \end{align*}$$

is bounded above (i.e., there exist constants $c_1,\dots ,c_{r_1+r_2}$ such that for each

$$\begin{align*}x=(x_1,\dots,x_{r_1+r_2})\in S, \end{align*}$$

we have $x_i\leq c_i$ for all i). For this, note that any $x\in S$ can be written as

$$\begin{align*}x=\text{pr}(x) + (d_v)_v \frac{1}{d} \sum_{v'|\infty} x_{v'}, \end{align*}$$

and thus, the components of S are bounded above, noting that the first term, $\text {pr}(x)$ , has components bounded above, and the second term has negative components.

Lemma 4.0.9. The set ${\mathcal F}(1)$ is definable in ${\mathbb R}_{\exp {}}$ .

Proof. We make the following straightforward observations:

  • The product $\prod _{v|\infty } (K_v^{n+1}-\{0\})$ , viewed as a subset of ${\mathbb R}^{(r_1+2r_2)(n+1)}$ , is semi-algebraic since it is clearly the complement of a semi-algebraic set.

  • The set ${\mathcal D}(1)$ is semi-algebraic.

  • The set

    $$ \begin{align*} {\mathcal F}&=\{x\in \prod_{v|\infty} (K_v^{n+1}-\{0\}) : \text{pr}\circ\eta(x)\in \tilde{{\mathcal F}}\}\\ &=\left\{x\in \prod_{v|\infty} (K_v^{n+1}-\{0\}) : 0\leq \check{u}_j \left(\log(\max_i |x_i|_v^{1/w_i})-d_v\left(\frac{1}{d} \sum_{v|\infty}\log(\max_i |x_i|_{v'}^{1/w_i})\right)\right)_v <1 \forall j\right\} \end{align*} $$
    is definable in ${\mathbb R}_{\exp }$ since it can be described in terms of polynomials and $\log $ , and $\log $ is definable in ${\mathbb R}_{\exp }$ (as observed in Subsection 3.1).

It follows that the intersection ${\mathcal F}(1)={\mathcal F}\cap {\mathcal D}(1)$ is definable in ${\mathbb R}_{\exp }$ .

We now analyze $M'(\Omega ,{\mathfrak c},B)$ . Note that if ${\mathfrak I}_{\textbf {w}}(x)\subseteq {\mathfrak c}$ , then we must have that x is contained in the lattice

$$\begin{align*}{\mathfrak c}^{\textbf{w}}={\mathfrak c}^{w_0}\times\cdots\times {\mathfrak c}^{w_n}. \end{align*}$$

Therefore, we can write $M'(\Omega ,{\mathfrak c},B)$ as

$$\begin{align*}M'(\Omega,{\mathfrak c},B)=\#\left\{x\in ({\mathfrak c}^{\textbf{w}}-\{0\})/{\mathcal O}_K^\times : \frac{\text{Ht}_{{\textbf{w}},\infty}(x)}{N_{K/{\mathbb Q}}({\mathfrak c})}\leq B,\ x\in \Omega^{\mathrm{aff}}\right\}. \end{align*}$$

We now stratify $\Omega $ . For each $t\in {\mathbb Z}_{\geq 0}$ , set

$$\begin{align*}\Omega_{v,t}^{\mathrm{aff}}=\pi_v^t\ast_{\textbf{w}} \Omega_{v,0}^{\mathrm{aff}}=\prod_{j=0}^n \{x\in {\mathcal O}_{K,v} : |x- \pi_v^{w_j t} a_{v,j}|_v\leq q_v^{-w_j t}\omega_{v,j}\}. \end{align*}$$

For ${ {\textbf {t}}}=(t_v)_{v\in S} \in {\mathbb Z}_{\geq 0}^{\# S}$ , set

$$\begin{align*}\Omega_{{{\textbf{t}}}}^{\mathrm{aff}}\stackrel{\mathrm{def}}{=} \{x\in \Omega^{\mathrm{aff}} : x\in \Omega_{v,t_v}^{\mathrm{aff}} \text{ for all } v\in S\}. \end{align*}$$

Since each $\Omega _v$ is irreducible, we have

$$\begin{align*}\Omega^{\mathrm{aff}} \cap {\mathcal O}_{K,v}^{n+1} = \bigsqcup_{{{\textbf{t}}}\in {\mathbb Z}^{\# S}_{\geq 0}} \Omega_{{\textbf{t}}}^{\mathrm{aff}}. \end{align*}$$

Set

$$\begin{align*}\epsilon_{{{\textbf{t}}}}=\max_j\left\{\prod_{{\mathfrak p}\in S} q_v^{t_vw_j}\omega_{v,j}^{-1}\right\}\left(\prod_{{\mathfrak p}\in S} \frac{m_{\mathfrak p}(\Omega_{v,t_v}^{\mathrm{aff}}\cap {\mathfrak c}_{\mathfrak p}^{\textbf{w}})}{m_{\mathfrak p}({\mathfrak c}_{\mathfrak p}^{\textbf{w}})}\right). \end{align*}$$

Note that each $\varpi (K)$ -orbit (with respect to the ${\textbf {w}}$ -weighted action) of an element of $(K-\{0\})^{n+1}$ contains $\varpi _{K,{\textbf {w}}}$ elements. Thus,

(2) $$ \begin{align} M'(\Omega, {\mathfrak c}, B) =\sum_{{{\textbf{t}}}\in {\mathbb Z}^{\# S}_{\geq 0}} \left(\frac{\# \left\{ \Omega^{\mathrm{aff}}_{{{\textbf{t}}}} \cap {\mathcal F}\left(N_{K/{\mathbb Q}}({\mathfrak c}) B\right)\cap {\mathfrak c}^{\textbf{w}}\right\}}{\varpi_{K,{\textbf{w}}}} + O\,\left(\epsilon_{{\textbf{t}}} B^{|{\textbf{w}}|-w_{\min}/d}\right)\right), \end{align} $$

where each error term accounts for the points of $M'(\Omega ,{\mathfrak c}, B)$ contained in the intersection of $\Omega _{ {\textbf {t}}}^{\mathrm{aff}}$ and the subvariety of the affine cone of ${\mathcal P}({\textbf {w}})(K)$ consisting of points with at least one coordinate equal to zero; by Proposition 3.2.7, this error is at most $O\left (\epsilon _{ {\textbf {t}}} \left (B^{1/d}\right )^{d|{\textbf {w}}|-w_{\min }}\right )$ .

By Lemma 4.0.8 and Lemma 4.0.9, we may apply Proposition 3.2.7, with ${\mathfrak a}={\mathfrak c}$ and the local conditions $\Omega ^{\mathrm{aff}}_v\cap {\mathcal F}(1)$ for $v|\infty $ and $\Omega ^{\mathrm{aff}}_{v,t}$ for $v\in S$ . Doing so, we have

$$\begin{align*}\#\{\Omega_{{{\textbf{t}}}}^{\mathrm{aff}}\cap {\mathcal F}(N_{K/{\mathbb Q}}({\mathfrak c})B)\cap {\mathfrak c}^{{\textbf{w}}}\}=\frac{m_\infty(\Omega^{\mathrm{aff}}_\infty\cap {\mathcal F}(1))}{|\Delta_K|^{(n+1)/2}} \left(\prod_{{\mathfrak p}\in S} \frac{m_{\mathfrak p}(\Omega^{\mathrm{aff}}_{v,t_v}\cap{\mathfrak c}_{\mathfrak p}^{\textbf{w}})}{m_{\mathfrak p}({\mathfrak c}_{\mathfrak p}^{\textbf{w}})}\right)B^{|{\textbf{w}}|} + O\,\left(\epsilon_{{{\textbf{t}}}} B^{|{\textbf{w}}|-\frac{w_{\min}}{d}}\right). \end{align*}$$

Let $v({\mathfrak c})$ denote the v-adic valuation of the ideal ${\mathfrak c}\subseteq {\mathcal O}_K$ (i.e., the valuation of any (and all) embeddings of ${\mathfrak c}$ into ${\mathcal O}_{K,v}$ ). By Proposition 4.0.4(c), if $t< v({\mathfrak c})$ , then $m_{\mathfrak p}(\Omega ^{\mathrm{aff}}_{v,t}\cap {\mathfrak c}_{\mathfrak p}^{\textbf {w}})=0$ . However, when $t\geq v({\mathfrak c})$ , we compute

$$ \begin{align*} \frac{m_{\mathfrak p}(\Omega^{\mathrm{aff}}_{v,t}\cap{\mathfrak c}_{\mathfrak p}^{\textbf{w}})}{m_{\mathfrak p}({\mathfrak c}_{\mathfrak p}^{{\textbf{w}}})} &=\frac{1}{m_{\mathfrak p}({\mathfrak c}_{\mathfrak p}^{\textbf{w}})} m_{\mathfrak p}\left(\prod_{j=0}^n \{x\in {\mathfrak c}_{\mathfrak p}^{w_j} : |x-\pi_v^{tw_j}a_j|_v\leq q_v^{-tw_j}\omega_j\}\right)\\ &=\frac{1}{m_{\mathfrak p}({\mathfrak c}_{\mathfrak p}^{\textbf{w}})} m_{\mathfrak p}\left(\prod_{j=0}^n \{x\in {\mathcal O}_{K,{\mathfrak p}} : |\pi_v^{v({\mathfrak c})w_j}x-\pi_v^{tw_j}a_j|_v\leq q_v^{-tw_j}\omega_j\}\right)\\ &=m_{\mathfrak p}\left(\Omega^{\mathrm{aff}}_{v,t-v({\mathfrak c})}\right). \end{align*} $$

By Proposition 4.0.4, we know that the $\Omega _{v,t}^{\mathrm{aff}}$ are disjoint for distinct values of t. Therefore, summing over all $t\in {\mathbb Z}_{\geq 0}$ , it follows that

$$\begin{align*}\sum_{t\geq 0} \frac{m_{\mathfrak p}(\Omega^{\mathrm{aff}}_{v,t}\cap{\mathfrak c}_{\mathfrak p}^{\textbf{w}})}{m_{\mathfrak p}({\mathfrak c}_{\mathfrak p}^{{\textbf{w}}})}=\sum_{s\geq 0} m_{\mathfrak p}(\Omega_{v,s}^{\mathrm{aff}})=m_{\mathfrak p}(\Omega_v^{\mathrm{aff}}\cap {\mathcal O}_{K,v}^{n+1}), \end{align*}$$

and we obtain the asymptotic

(3) $$ \begin{align} M'(\Omega,{\mathfrak c},B)=\frac{m_\infty(\Omega^{\mathrm{aff}}_\infty\cap {\mathcal F}(1))}{\varpi_{K,{\textbf{w}}}|\Delta_K|^{(n+1)/2}} \left(\prod_{{\mathfrak p}\in S} m_{\mathfrak p}(\Omega^{\mathrm{aff}}_{v}\cap{\mathcal O}_{K,v}^{n+1})\right)B^{|{\textbf{w}}|}+ O\,\left(\epsilon_\Omega B^{|{\textbf{w}}|-w_{\min}/d}\right). \end{align} $$

Note that as $m_\infty $ is the product of measures $(m_v)_{v|\infty }$ , one has

$$\begin{align*}\frac{m_\infty(\Omega_\infty^{\mathrm{aff}}\cap {\mathcal F}(1))}{m_\infty({\mathcal F}(1))}=\prod_{v|\infty}\frac{m_v(\{x\in \Omega_v^{\mathrm{aff}} : \text{Ht}_v(x)\leq 1\})}{m_v(\{x\in K_v^{n+1} : \text{Ht}_v(x)\leq 1\})}, \end{align*}$$

and thus,

(4) $$ \begin{align} m_\infty(\Omega_\infty^{\mathrm{aff}}\cap {\mathcal F}(1))=m_\infty({\mathcal F}(1))\prod_{v|\infty}\frac{m_v(\{x\in \Omega_v^{\mathrm{aff}} : \text{Ht}_v(x)\leq 1\})}{m_v(\{x\in K_v^{n+1} : \text{Ht}_v(x)\leq 1\})}. \end{align} $$

Computing the volume of ${\mathcal F}(1)$ as in [Reference DengDen98, Proposition 5.3], we obtain

$$\begin{align*}m_\infty({\mathcal F}(1))=\left(2^{r_1+r_2}\pi^{r_2}\right)^{n+1}R_K|{\textbf{w}}|^{r_1+r_2-1}. \end{align*}$$

This differs by a factor of $2^{(n+1)r_2}$ from Deng’s result since we are using the usual Haar measure on $K_\infty $ , rather than the Lebesgue measure.

Observe that the leading cofficient of the asymptotic (3) equals $\zeta _K(|{\textbf {w}}|)\kappa /h_K$ , where $\kappa $ is as in the statement of Theorem 4.0.5.

Note that for $x\in {\mathfrak c}^{\textbf {w}}$ , ${\mathfrak I}_{\textbf {w}}(x)={\mathfrak a}{\mathfrak c}$ for some ideal ${\mathfrak a}\subseteq {\mathcal O}_K$ . Then, $\text {Ht}_{\textbf {w}}(x)=\text {Ht}_{{\textbf {w}},\infty }(x)/N_{K/{\mathbb Q}}({\mathfrak a}{\mathfrak c})$ . Therefore,

$$\begin{align*}M'(\Omega,{\mathfrak c},B)=\sum_{{\mathfrak a}\subseteq {\mathcal O}_K} M(\Omega,{\mathfrak a}{\mathfrak c}, B/N_{K/{\mathbb Q}}({\mathfrak a}))=\sum_{\substack{{\mathfrak a}\subseteq {\mathcal O}_K \\ N_{K/{\mathbb Q}}({\mathfrak a})\leq B}} M(\Omega,{\mathfrak a}{\mathfrak c}, B/N_{K/{\mathbb Q}}({\mathfrak a})), \end{align*}$$

where the inequality $N_{K/{\mathbb Q}}({\mathfrak a})\leq B$ comes from the fact that $M(\Omega ,{\mathfrak c},B)=0$ if $B<1$ , since all points have height greater than or equal to $1$ .

Since the ideals of ${\mathcal O}_K$ form a poset (ordered by inclusion), we may use Möbius inversion (see, for example, [Reference StanleySta97, §3.7] for details about Möbius inversion for posets). Applying Möbius inversion, and using our asymptotic (3) for $M'(\Omega ,{\mathfrak c},B)$ , we have that

$$ \begin{align*} M(\Omega,{\mathfrak c},B) &=\sum_{\substack{{\mathfrak a}\subseteq {\mathcal O}_K\\ N_{K/{\mathbb Q}}({\mathfrak a})\leq B}} \mu({\mathfrak a})\left(\frac{\zeta_K(|{\textbf{w}}|)\kappa}{h_K}\left(\frac{B}{N_{K/{\mathbb Q}}({\mathfrak a})}\right)^{|{\textbf{w}}|}+ O\,\left(\epsilon_\Omega\left(\frac{B}{N_{K/{\mathbb Q}}({\mathfrak a})}\right)^{|{\textbf{w}}|-w_{\min}/d}\right)\right)\\&=\frac{\zeta_K(|{\textbf{w}}|)\kappa}{h_K} B^{|{\textbf{w}}|}\left(\sum_{{\mathfrak a}\subseteq {\mathcal O}_K} \mu({\mathfrak a})\frac{1}{N_{K/{\mathbb Q}}({\mathfrak a})^{|{\textbf{w}}|}} - \sum_{\substack{{\mathfrak a}\subseteq {\mathcal O}_K\\ N_{K/{\mathbb Q}}({\mathfrak a})> B}}\mu({\mathfrak a})\frac{1}{N_{K/{\mathbb Q}}({\mathfrak a})^{|{\textbf{w}}|}}\right)\\& \quad +O\,\left(\epsilon_\Omega B^{|{\textbf{w}}|-w_{\min}/d} \sum_{\substack{{\mathfrak a}\subseteq {\mathcal O}_K\\ N_{K/{\mathbb Q}}({\mathfrak a})\leq B}}\frac{1}{N_{K/{\mathbb Q}}({\mathfrak a})^{|{\textbf{w}}|-w_{\min}/d}}\right)\\&=\frac{\zeta_K(|{\textbf{w}}|)\kappa}{h_K} B^{|{\textbf{w}}|}\left(\frac{1}{\zeta_K(|{\textbf{w}}|)} - O\,\left(B^{-|{\textbf{w}}|+1}\right)\right) + \begin{cases} O\,\left(\epsilon_\Omega B\log(B)\right) & {\substack{\text{ if } {\textbf{w}}=(1,1)\\ \text{and}\ K={\mathbb Q},}}\\ O\,\left(\epsilon_\Omega B^{|{\textbf{w}}|-w_{\min}/d}\right) & \text{ else, } \end{cases}\\&= \frac{\kappa}{h_K} B^{|{\textbf{w}}|} + \begin{cases} O\,\left(\epsilon_\Omega B\log(B)\right) & \text{ if } {\textbf{w}}=(1,1) \text{ and } K={\mathbb Q},\\ O\,\left(\epsilon_\Omega B^{|{\textbf{w}}|-w_{\min}/d}\right) & \text{ else. } \end{cases} \end{align*} $$

Summing over the $h_K$ ideal class representatives ${\mathfrak c}_i$ , we arrive at Theorem 4.0.5.

5. Counting elliptic curves

In this section, we apply Theorem 4.0.5 to the moduli stack of elliptic curves.

5.1. Counting elliptic curves with a prescribed local condition

An application of Theorem 4.0.5 gives the following result:

Corollary 5.1.1. Let ${\mathcal E}_K(B)$ denote the set of isomorphism classes of elliptic curves over K with height less than B. Then,

$$\begin{align*}\#{\mathcal E}_K(B)=\kappa B^{5/6}+O(B^{5/6-1/3d}), \end{align*}$$

where

$$\begin{align*}\kappa=\frac{h_K (2^{r_1+r_2}\pi^{r_2})^{2} R_K 10^{r_1+r_2-1}\gcd(2,\varpi_K)}{\varpi_K |\Delta_K| \zeta_K(10)}. \end{align*}$$

Proof. Recall that the moduli stack of elliptic curves, ${\mathcal X}_{\text {GL}_2({\mathbb Z})}$ , is isomorphic to the weighted projective stack ${\mathcal P}(4,6)$ . Also recall that the naive height on ${\mathcal X}_{\text {GL}_2({\mathbb Z})}$ corresponds to the twelfth power of the height on ${\mathcal P}(4,6)$ with respect to the tautological bundle (i.e., the height $\text {Ht}_{(4,6)}$ ). Therefore, the desired result follows from applying Theorem 4.0.5 to ${\mathcal P}(4,6)$ with trivial local conditions (i.e., $S=\emptyset $ ).

We now use Theorem 4.0.5 to count elliptic curves with a prescribed local condition.

Theorem 1.1.2.

Let K be a degree d number field, and let ${\mathfrak p}\subset {\mathcal O}_K$ be a prime ideal of norm q such that $2\nmid q$ and $3\nmid q$ . Let ${\mathscr L}$ be one of the local conditions listed in Table 1. Then the number of elliptic curves over K with naive height less than B and which satisfy the local condition ${\mathscr L}$ at ${\mathfrak p}$ is

$$\begin{align*}\kappa\kappa_{\mathscr L} B^{5/6}+O\,\left(\epsilon_{\mathscr L} B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align*}$$

where

$$\begin{align*}\kappa=\frac{h_K \left(2^{r_1+r_2}\pi^{r_2}\right)^{2}R_K10^{r_1+r_2-1}\gcd(2,\varpi_K)}{\varpi_{K} |\Delta_K|\zeta_K(10)}, \end{align*}$$

and where $\kappa _{\mathscr L}$ and $\epsilon _{\mathscr L}$ are as in Table 1.

Proof. Good reduction case. Note that the number of elliptic curves over ${\mathbb F}_q$ with good reduction is

$$\begin{align*}\#\{(a,b)\in {\mathbb F}_q\times{\mathbb F}_q : 4a^3+27b^2\not\equiv 0 \quad\pmod{q}\}=q^2-q. \end{align*}$$

For each pair $(a,b)$ in the above set, fix a lift in ${\mathcal O}_{K,v}^2$ , which, by an abuse of notation, we will also denote $(a,b)$ . Then, consider the irreducible local condition

$$\begin{align*}\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(a,b)\stackrel{\mathrm{def}}{=}\left\{(x_0,x_1)\in {\mathcal O}_{K,v}^{2} : |x_0-a|_v\leq \frac{1}{q},\ |x_1-b|_v\leq \frac{1}{q}\right\}. \end{align*}$$

We find the ${\mathfrak p}$ -adic measure

$$\begin{align*}m_{\mathfrak p}(\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(a,b))=\frac{1}{q^2}. \end{align*}$$

For $t\in {\mathbb Z}_{\geq 0}$ , the sets

$$\begin{align*}\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b)=\pi_v^t\ast_{(4,6)}\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(a,b)=\left\{(x_0,x_1)\in {\mathcal O}_{K,v}^{2} : |x_0-q^{4t}a|_v\leq \frac{1}{q^{4t+1}},\ |x_1-q^{6t}b|_v\leq \frac{1}{q^{6t+1}}\right\} \end{align*}$$

each have ${\mathfrak p}$ -adic measure

$$\begin{align*}m_{\mathfrak p}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b))=\frac{1}{q^{10t+2}}. \end{align*}$$

Applying Theorem 4.0.5 to each $\Omega _{{\mathfrak p},0}^{\mathrm{aff}}(a,b)$ , and summing over all the $q^2-q$ possible $(a,b)$ pairs, it follows that the number of elliptic curves of bounded height over K with good reduction at ${\mathfrak p}$ is

(5) $$ \begin{align} \kappa\kappa_{\mathscr L} B^{5/6}+O\,\left(\epsilon B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align} $$

where

$$\begin{align*}\kappa_{\mathscr L}=\sum_{(a,b)}\sum_{t=0}^\infty m_{{\mathfrak p}}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b))=(q^2-q)\sum_{t=0}^\infty \frac{1}{q^{10t+2}}=(q^2-q)\frac{1}{q^2}\frac{q^{10}}{q^{10}-1}= \frac{q-1}{q}\frac{q^{10}}{q^{10}-1}, \end{align*}$$

and

$$\begin{align*}\epsilon=\sum_{(a,b)}\sum_{t=0}^\infty\max\{q^{4t+1},q^{6t+1}\}m_{{\mathfrak p}}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b)) =(q^2-q)\sum_{t=0}^\infty \frac{q^{6t+1}}{q^{10t+2}} =\frac{q^5-q^4}{q^4-1}. \end{align*}$$

As the error term can be rewritten as $O(q B^{\frac {5}{6}-\frac {1}{3d}})$ , we may replace the coefficient $\epsilon $ in the asymptotic (5) with $\epsilon _{\mathscr L}=q$ .

Kodaira type III* case. By Tate’s algorithm [Reference TateTat75], an elliptic curve over $K_{\mathfrak p}$ given in short Weierstrass form, $E:y^2=x^3+ax+b$ , has type III* reduction if and only if $v_{\mathfrak p}(a)=3$ and $v_{\mathfrak p}(b)\geq 5$ . Observe that

$$\begin{align*}\#\{(a,b)\in {\mathcal O}_K/{\mathfrak p}^5\times {\mathcal O}_K/{\mathfrak p}^5 : v_{\mathfrak p}(a)=3,\ v_{\mathfrak p}(b)\geq 5\}=q(q-1). \end{align*}$$

For each pair $(a,b)$ in the above set, fix a lift in ${\mathcal O}_{K,v}^2$ , which, by an abuse of notation, we will also denote $(a,b)$ . Then, consider the irreducible local condition

$$\begin{align*}\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(a,b)\stackrel{\mathrm{def}}{=} \left\{(x_0,x_1)\in {\mathcal O}_{K,v}^{2} : |x_0-a|_v\leq \frac{1}{q^5},\ |x_1-b|_v\leq \frac{1}{q^5}\right\}. \end{align*}$$

We find that the ${\mathfrak p}$ -adic measure of this set is

$$\begin{align*}m_{\mathfrak p}(\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(a,b))=\frac{1}{q^{10}}. \end{align*}$$

The sets

$$\begin{align*}\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b)= \pi_v^t\ast_{(4,6)}\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(a,b)=\left\{(x_0,x_1)\in {\mathcal O}_{K,v}^{2} : |x_0-q^{4t}a|_v\leq \frac{1}{q^{4t+5}},\ |x_1-q^{6t}b|_v\leq \frac{1}{q^{6t+5}}\right\} \end{align*}$$

each have ${\mathfrak p}$ -adic measure

$$\begin{align*}m_{\mathfrak p}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b))=\frac{1}{q^{10t+10}}. \end{align*}$$

Applying Theorem 4.0.5 to each $\Omega _{{\mathfrak p},0}^{\mathrm{aff}}(a,b)$ , and summing over all pairs $(a,b)$ , yields the following asymptotic for the number of elliptic curves of bounded height over K with reduction of Kodaira type III* at ${\mathfrak p}$ :

(6) $$ \begin{align} \kappa\kappa_{\mathscr L} B^{5/6}+O\,\left(\epsilon B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align} $$

where

$$\begin{align*}\kappa_{\mathscr L} =\sum_{(a,b)}\sum_{t=0}^\infty m_{{\mathfrak p}}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b))=q(q-1)\sum_{t=0}^\infty \frac{1}{q^{10t+10}}=(q^2-q)\frac{1}{q^{10}}\frac{q^{10}}{q^{10}-1}=\frac{q-1}{q^9}\frac{q^{10}}{q^{10}-1}, \end{align*}$$

and

$$\begin{align*}\epsilon=\sum_{(a,b)}\sum_{t=0}^\infty\max\{q^{4t+5},q^{6t+5}\}m_{{\mathfrak p}}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(a,b)) =(q^2-q)\sum_{t=0}^\infty \frac{q^{6t+5}}{q^{10t+10}} =\frac{q-1}{q^{4}-1}. \end{align*}$$

As the error term can be rewritten as $O(q^{-3}B^{\frac {5}{6}-\frac {1}{3d}})$ , we may replace the coefficient $\epsilon $ in the asymptotic (6) with $\epsilon _{\mathscr L}=1/q^{3}$ .

The other cases can be proven similarly to those given above.

We now turn to estimating the number of elliptic curves with a prescribed trace of Frobenius. For $b,c\in {\mathbb F}_q$ , let $E_{b,c}$ denote the short Weierstrass model $y^2=x^3+bx+c$ , with disciminant $\Delta (b,c)=-16(4b^3+27c^2)$ and trace of Frobenius $a_q(E_{b,c})$ . We study the counting function

(7) $$ \begin{align} H(a,q)\stackrel{\mathrm{def}}{=}\#\{(b,c)\in {\mathbb F}_q\times {\mathbb F}_q : \Delta(b,c)\neq 0,\ a_q(E_{b,c})=a\}. \end{align} $$

This function is closely related to Hurwitz-Kronecker class numbers, which we now recall. Let K be an imaginary quadratic field with ring of integers ${\mathcal O}_K$ , and let ${\mathcal O}$ be an order in K with class number $h({\mathcal O})$ . The Hurwitz-Kronecker class number of ${\mathcal O}$ , denoted $H(\text {disc}({\mathcal O}))$ , is defined as

(8) $$ \begin{align} H(\text{disc}({\mathcal O}))\stackrel{\mathrm{def}}{=}\sum_{{\mathcal O}\subset {\mathcal O}' \subset {\mathcal O}_K} \frac{h({\mathcal O}')}{\# {\mathcal O}^{\prime\times}}. \end{align} $$

This definition of the Hurwitz-Kronecker class number agrees with the definition given in [Reference LenstraLen87] but is half as large as it is sometimes defined (such as in [Reference CoxCox89]). Let $q=p^n$ be a power of a prime $p>3$ . Then, a beautiful result, following largely from work of Deuring [Reference DeuringDeu41], expresses the number of elliptic curves over a finite field ${\mathbb F}_q$ with a prescribed trace of Frobenius in terms of Hurwitz-Kronecker class numbers:

(9) $$ \begin{align} \begin{aligned} H(a,q)=\begin{cases} (q-1) H(a^2-4q) & \text{ if } |a|<2\sqrt{q} \text{ and } p\nmid a\\ (q-1)H(-4p) & \text{ if } n \text{ is odd and } a=0\\ \frac{q-1}{12}\left(p+6-4\genfrac{(}{)}{}{}{-3}{p}-3\genfrac{(}{)}{}{}{-4}{p}\right) & \text{ if } n \text{ is even and } a^2=4q\\ (q-1)\left(1-\genfrac{(}{)}{}{}{-3}{p}\right) & \text{ if } n \text{ is even and } a^2=q\\ (q-1)\left(1-\genfrac{(}{)}{}{}{-4}{p}\right) & \text{ if } n \text{ is even and } a=0\\ 0 & \text{ otherwise.} \end{cases} \end{aligned} \end{align} $$

This formula can be derived from [Reference SchoofSch87, Theorem 4.6], which counts the number of isomorphism classes of elliptic curves over ${\mathbb F}_q$ with a prescribed trace of Frobenius. From this, one can deduce formula (9) by noting that the isomorphism class of each $E_{b,c}$ contains exactly $q-1$ elliptic curves. To see that each isomorphism class contains exactly $q-1$ elliptic curves, note that $E_{b,c}\cong E_{d,e}$ if and only if $(d,e)=(\lambda ^4b, \lambda ^6c)$ for some $\lambda \in {\mathbb F}_q^\times $ , and since $p>3$ , the pairs $(\lambda ^4b, \lambda ^6c)$ will be distinct for distinct $\lambda \in {\mathbb F}_q^\times $ . (See [Reference CoxCox89, Theorem 14.18] for a proof of formula (9) in the case that $q=p$ is a prime.)

Theorem 1.1.3.

Let K be a degree d number field, and let ${\mathfrak p}$ be a prime ideal of ${\mathcal O}_K$ of degree n above a rational prime $p>3$ . Set $q=p^n$ . Let $a\in {\mathbb Z}$ be an integer satisfying $|a|\leq 2\sqrt {q}$ . Then, the number of elliptic curves over K with naive height less than B, good reduction at ${\mathfrak p}$ , and which have trace of Frobenius a at ${\mathfrak p}$ is

$$\begin{align*}\kappa\kappa_{n,a} B^{5/6}+O\,\left(\epsilon_{n,a} B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align*}$$

where

$$\begin{align*}\kappa=\frac{h_K \left(2^{r_1+r_2}\pi^{r_2}\right)^{2}R_K10^{r_1+r_2-1}\gcd(2,\varpi_K)}{\varpi_{K} |\Delta_K|\zeta_K(10)}, \end{align*}$$

and where $\kappa _{n,a}$ and $\epsilon _{n,a}$ are as in Table 2.

Proof. For each of the $H(a,q)$ pairs $(b,c)$ in the set

$$\begin{align*}\{(b,c)\in {\mathbb F}_q\times {\mathbb F}_q : \Delta(b,c)\neq 0,\ a_q(E_{b,c})=a\}, \end{align*}$$

fix a lift in ${\mathcal O}_{K,v}^2$ , which, by an abuse of notation, we will also denote $(b,c)$ . Then consider the irreducible local condition

$$\begin{align*}\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(b,c)\stackrel{\mathrm{def}}{=}\left\{(x_0,x_1)\in {\mathcal O}_{K,v}^{2} : |x_0-b|_v\leq \frac{1}{q},\ |x_1-c|_v\leq \frac{1}{q}\right\}. \end{align*}$$

We find that the ${\mathfrak p}$ -adic measure of each of these local conditions is

$$\begin{align*}m_{\mathfrak p}(\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(b,c))=\frac{1}{q^2}. \end{align*}$$

More generally, for $t\in {\mathbb Z}_{\geq 0}$ , the sets

$$\begin{align*}\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(b,c)=\pi_v^t\ast_{(4,6)}\Omega_{{\mathfrak p},0}^{\mathrm{aff}}(b,c)=\left\{(x_0,x_1)\in {\mathcal O}_{K,v}^{2} : |x_0-q^{4t}b|_v\leq \frac{1}{q^{4t+1}},\ |x_1-q^{6t}c|_v\leq \frac{1}{q^{6t+1}}\right\} \end{align*}$$

each have ${\mathfrak p}$ -adic measure

$$\begin{align*}m_{\mathfrak p}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(b,c))=\frac{1}{q^{10t+2}}. \end{align*}$$

Applying Theorem 4.0.5 to each $\Omega _{{\mathfrak p},0}^{\mathrm{aff}}(b,c)$ , and summing over all the $H(a,q)$ possible $(b,c)$ pairs, it follows that the number of elliptic curves of bounded height over K with good reduction at ${\mathfrak p}$ and trace of Frobenius a at ${\mathfrak p}$ is

(10) $$ \begin{align} \kappa\kappa' B^{5/6}+O\,\left(\epsilon B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align} $$

where

$$\begin{align*}\kappa'=\sum_{(b,c)}\sum_{t=0}^\infty m_{{\mathfrak p}}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(b,c))=H(a,q)\sum_{t=0}^\infty \frac{1}{q^{10t+2}}=H(a,q)\frac{1}{q^2}\frac{q^{10}}{q^{10}-1}, \end{align*}$$

and

$$\begin{align*}\epsilon=\sum_{(b,c)}\sum_{t=0}^\infty\max\{q^{4t+1},q^{6t+1}\}m_{{\mathfrak p}}(\Omega_{{\mathfrak p},t}^{\mathrm{aff}}(b,c)) =H(a,q)\sum_{t=0}^\infty \frac{q^{6t+1}}{q^{10t+2}} =H(a,q)\frac{q^3}{q^4-1}. \end{align*}$$

From this, the desired asymptotics can now be derived using (9).

For example, in the case that a is an integer satisfying $|a|< 2\sqrt {q}$ and $p\nmid a$ , we have that $H(a,q)=(q-1)H(a^2-4q)$ . Thus,

$$\begin{align*}\kappa'=\frac{H(a,q)}{q^2}\frac{q^{10}}{q^{10}-1}=\frac{(q-1)H(a^2-4q)}{q^2}\frac{q^{10}}{q^{10}-1}=\kappa_{n,a} \end{align*}$$

and

$$\begin{align*}\epsilon=(q-1)H(a^2-4q)\frac{q^3}{q^4-1}. \end{align*}$$

As the error term can be rewritten as $O\left (H(a^2-4q)B^{\frac {5}{6}-\frac {1}{3d}}\right )$ , we may replace the coefficient $\epsilon $ in the asymptotic (10) with $\epsilon _{n,a}=H(a^2-4q)$ .

6. Average analytic ranks

In this section, we use our results from the previous section on counting elliptic curves with prescribed local conditions in order to give a bound for the average analytic rank of elliptic curves over number fields. Our strategy will mainly follow that of Cho and Jeong [Reference Cho and JeongCJ23b, Reference Cho and JeongCJ23a] and is related to Brumer’s strategy (see Remark 6.1.3).

6.1. L-functions of elliptic curves

In this subsection, we recall basic facts about L-functions of elliptic curves. Our exposition mostly follows [Reference MillerMil02, Appendix A].

Let E be an elliptic curve of discriminant $\Delta _E$ over a number field K. For each finite place $v\in \text {Val}_0(K)$ , let $E_v$ denote the reduction of E at v, let $q_v$ be the order of the residue field $k_v$ of K at v, let $N_v=\# E_v({\mathbb F}_{q_v})$ , and let $a_v=q_v+1-N_v$ . If E has bad reduction at v, set

$$\begin{align*}b_v\stackrel{\mathrm{def}}{=} \begin{cases} 1 & \text{ if }E\text{ has split multiplicative reduction at }v,\\ -1 & \text{ if }E\text{ has nonsplit multiplicative reduction at }v,\\ 0 & \text{ if }E\text{ has additive reduction at }v. \end{cases} \end{align*}$$

For each $v\in \text {Val}_0(K)$ , define a polynomial $L_v(E/K,T)$ by

$$\begin{align*}L_v(E/K,T)\stackrel{\mathrm{def}}{=}\begin{cases} 1-a_v T+ q_v T^2 & \text{ if }E\text{ has good reduction at }v,\\ 1-b_vT & \text{ if }E\text{ has bad reduction at }v. \end{cases} \end{align*}$$

The Hasse–Weil $\mathbf {L}$ -function $L(E/K,s)$ of E over K is defined as the Euler product

$$\begin{align*}L(E/K,s)\stackrel{\mathrm{def}}{=} \prod_{v\in \text{Val}_0(K)} L_v(E/K, q_v^{-s})^{-1}. \end{align*}$$

The logarithmic derivative of $L(E/K,s)$ is

$$\begin{align*}\frac{L'}{L}(E/K,s)=-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)>0}} \frac{d}{ds}\left(\log(1-b_v q_v^{-s})\right)-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)=0}} \frac{d}{ds}\left(\log(1-a_v q_v^{-s}+q_v^{1-2s})\right). \end{align*}$$

The first sum can be rewritten as

$$\begin{align*}-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)>0}} \frac{d}{ds}\left(\log(1-b_v q_v^{-s})\right)=\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)>0}} \frac{d}{ds}\left(\sum_{k=1}^\infty \frac{b_v^k}{k q_v^{ks}}\right)=-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)>0}} \sum_{k=1}^\infty \frac{b_v^k \log(q_v)}{q_v^{ks}}. \end{align*}$$

For the second sum, factor $1-a_v T+q_v T^2$ as $(1-\alpha _v T)(1-\beta _v T)$ , where $\alpha _v+\beta _v=a_v$ and $\alpha _v\beta _v=q_v$ . Then,

$$\begin{align*}-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)=0}} \frac{d}{ds}\left(\log(1-a_v q_v^{-s}+q_v^{1-2s})\right)=-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)=0}} \sum_{k=1}^\infty \frac{(\alpha_v^k+\beta_v^k) \log(q_v)}{q_v^{ks}}. \end{align*}$$

We thus obtain the following expression for the logarithmic derivative of $L(E/K,s)$ :

$$ \begin{align*} \frac{L'}{L}(E/K,s)=-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)>0}} \sum_{k=1}^\infty \frac{b_v^k \log(q_v)}{q_v^{ks}}-\sum_{\substack{v\in \text{Val}_0(K)\\ v(\Delta_E)=0}} \sum_{k=1}^\infty \frac{(\alpha_v^k+\beta_v^k) \log(q_v)}{q_v^{ks}}. \end{align*} $$

Define the von Mangoldt function on integral ideals ${\mathfrak a}\subseteq {\mathcal O}_K$ as

$$\begin{align*}\Lambda_K({\mathfrak a})\stackrel{\mathrm{def}}{=} \begin{cases} \log(q_v) & \text{ if } {\mathfrak a}={\mathfrak p}_v^k \text{ for some } k,\\ 0 & \text{ otherwise.} \end{cases} \end{align*}$$

For prime powers, set

$$\begin{align*}\widehat{a_E}({\mathfrak p}_v^k)=\begin{cases} \alpha_v^k+\beta_v^k & \text{ if }E\text{ has good reduction at }{\mathfrak p}_v,\\ b_v^k & \text{ if }E\text{ has bad reduction at }{\mathfrak p}_v, \end{cases} \end{align*}$$

and for ${\mathfrak a}$ not a power of a prime, set $\widehat {a_E}({\mathfrak a})=0$ . Then our expression for the logarithmic derivative of $L(E/K,s)$ becomes

(11) $$ \begin{align} \frac{L'}{L}(E/K,s)=-\sum_{{\mathfrak a}\subseteq {\mathcal O}_K} \frac{\widehat{a_E}({\mathfrak a})\Lambda_K({\mathfrak a})}{N_{K/{\mathbb Q}}({\mathfrak a})^s}, \end{align} $$

where the sum is over the nonzero integral ideals of ${\mathcal O}_K$ .

Let $\Gamma (s)\stackrel {\mathrm{def}}{=}\int _0^\infty t^{s-1}e^{-t} dt$ be the usual gamma function. Let $\Gamma _K(s)$ be the gamma factor

$$\begin{align*}\Gamma_K(s)\stackrel{\mathrm{def}}{=}\left((2\pi)^{-s} \Gamma(s)\right)^{[K:{\mathbb Q}]}. \end{align*}$$

Let $\mathfrak {f}_{E/K}$ be the conductor of E over K, and define a constant

$$\begin{align*}A_{E/K}\stackrel{\mathrm{def}}{=} N_{K/{\mathbb Q}}(\mathfrak{f}_{E/K}) \Delta_K^2. \end{align*}$$

Define the completed Hasse–Weil $\mathbf {L}$ -function, $\Lambda (E/K, s)$ , as

$$\begin{align*}\Lambda(E/K,s)\stackrel{\mathrm{def}}{=} A_{E/K}^{s/2}\Gamma_K(s) L(E/K,s). \end{align*}$$

We now assume that our elliptic curve E is modular over K, so that the L-function $L(E/K,s)$ is automorphic. This implies that the Hasse–Weil conjecture holds true for E, and thus,

$$\begin{align*}\Lambda(E/K,s)=\epsilon(E) \Lambda(E/K, 2-s), \end{align*}$$

where $\epsilon (E)\in \{\pm 1\}$ is the root number. We see that the logarithmic derivative of $\Lambda (E/K, s)$ is

(12) $$ \begin{align} \frac{\Lambda'}{\Lambda}(E/K, s)=\frac{\log(A_{E/K})}{2}+\frac{\Gamma_K'}{\Gamma_K}(s)+\frac{L'}{L}(E/K,s)=-\frac{\Lambda'}{\Lambda}(E/K,2-s). \end{align} $$

We now state the explicit formula for L-functions of elliptic curves over number fields:

Proposition 6.1.1 (Explicit Formula).

Let $\phi \colon {\mathbb C}\to {\mathbb C}$ be a partial function that is even and analytic in the strip $|\operatorname {Im}(s)|\leq \frac {1}{2}+\epsilon $ for some $\epsilon>0$ , and assume that on this strip $|\phi (s)|\ll (1+|s|)^{-(1+\delta )}$ for some $\delta>0$ as $|\text {Re}(s)|\to \infty $ . For $x\in {\mathbb R}$ , assume that $\phi (x)\in {\mathbb R}$ , and set

$$\begin{align*}\widehat{\phi}(t)\stackrel{\mathrm{def}}{=}\int_{\mathbb R} \phi(x) e^{-2\pi i tx} dx, \end{align*}$$

the Fourier transform. Let E be a modular elliptic curve over a number field K. Suppose that the L-function $L(E/K,s)$ associated to E satisfies the Generalized Riemann Hypothesis, so that all nontrivial zeros $\rho _j$ of $L(E/K,s)$ are of the form

$$\begin{align*}\rho_j=1+i\gamma_j \text{ for some } \gamma_j\in {\mathbb R}. \end{align*}$$

Let $r(E)\stackrel {\mathrm{def}}{=}\text {ord}_{s=1}(L(E/K,s))$ be the analytic rank of E. Then, we have

$$ \begin{align*} &r(E)\phi(0)+\sum_{\rho_j\neq 1} \phi(\gamma_j)\\ &\hspace{5mm} =\frac{\widehat{\phi}(0)}{2\pi}\log(A_{E/K})+\frac{1}{\pi}\int_{{\mathbb R}}\frac{\Gamma_K'}{\Gamma_K}(1+iy) \phi(y) dy -\frac{1}{\pi} \sum_{{\mathfrak a}\subseteq {\mathcal O}_K} \frac{ \widehat{a_E}({\mathfrak a}) \Lambda_K({\mathfrak a})}{N_{K/{\mathbb Q}}({\mathfrak a})}\cdot\widehat{\phi}\left(\frac{\log(N_{K/{\mathbb Q}}({\mathfrak a}))}{2\pi}\right), \end{align*} $$

where the sum on the left-hand side is over the zeros $\rho _j\neq 1$ with relevant multiplicity.

Proposition 6.1.1 is a slight generalization of [Reference Iwaniec and KowalskiIK04, Theorem 5.12] (see also [Reference MestreMes86] and [Reference MillerMil02, Appendix A] for the specific case of L-functions of elliptic curves). In [Reference Iwaniec and KowalskiIK04], the condition on the test function $\phi $ is that it is a Schwartz function. However, as observed in [Reference Goldston and GonekGG07, Lemma 1], the proof of [Reference Iwaniec and KowalskiIK04, Theorem 5.12] can be adapted to work for the set of functions $\phi $ satisfying the conditions in Proposition 6.1.1. We will use the following modified version of the explicit formula:

Corollary 6.1.2 (Modified Explicit Formula).

Maintaining the notation from Proposition 6.1.1 and letting X be a parameter, we have that

(13) $$ \begin{align} \begin{split} &r(E)\phi(0)+\sum_{\gamma_j\neq 0} \phi\left(\gamma_j\frac{\log(X)}{2\pi}\right) \\ &\hspace{1mm}= \frac{\widehat{\phi}(0) \log(A_{E/K})}{\log(X)} -\frac{2}{\log(X)} \sum_{{\mathfrak a}\subseteq {\mathcal O}_K} \frac{ \widehat{a_E}({\mathfrak a}) \Lambda_K({\mathfrak a})}{N_{K/{\mathbb Q}}({\mathfrak a})}\cdot\widehat{\phi}\left(\frac{\log(N_{K/{\mathbb Q}}({\mathfrak a}))}{\log(X)}\right)+O\,\left(\frac{1}{\log(X)}\right). \end{split} \end{align} $$

Proof. From the explicit formula (Proposition 6.1.1), it follows that

(14) $$ \begin{align} r(E)\phi(0)+\sum_{\gamma_j\neq 0} \phi\left(\gamma_j\frac{\log(X)}{2\pi}\right) &= \frac{\widehat{\phi}(0) \log(A_{E/K})}{\log(X)}+\frac{1}{\pi}\int_{{\mathbb R}}\frac{\Gamma_K'}{\Gamma_K}(1+iy) \phi\left(y\frac{\log(X)}{2\pi}\right) dy \nonumber\\& \quad -\frac{2}{\log(X)} \sum_{{\mathfrak a}\subseteq {\mathcal O}_K} \frac{ \widehat{a_E}({\mathfrak a}) \Lambda_K({\mathfrak a})}{N_{K/{\mathbb Q}}({\mathfrak a})}\cdot\widehat{\phi}\left(\frac{\log(N_{K/{\mathbb Q}}({\mathfrak a}))}{\log(X)}\right). \end{align} $$

A classical estimate for the logarithmic derivative of the gamma function is

$$\begin{align*}\left| \frac{\Gamma'}{\Gamma}(1+iy)\right|=O(\log(|y|+2)). \end{align*}$$

From this, and the fact that $\phi $ is absolutely integrable (since $|\phi (s)|\ll (1+|s|)^{-(1+\delta )}$ ), it follows that

$$\begin{align*}\int_{{\mathbb R}}\frac{\Gamma_K'}{\Gamma_K}(1+iy) \phi\left(y\frac{\log(X)}{2\pi}\right) dy=O\,\left(\frac{1}{\log(X)}\right). \end{align*}$$

Substituting this into (14) gives the modified explicit formula.

We now use the modified explicit formula to obtain an expression that bounds the analytic rank. Let E be an elliptic curve with minimal discriminant ${\mathfrak D}_{E/K}$ and of height less than or equal to X. For each such E, we have that $N_{K/{\mathbb Q}}({\mathfrak D}_{E/K})\ll X$ , and therefore,

$$\begin{align*}\log\left(N_{K/{\mathbb Q}}({\mathfrak D}_{E/K})\right)\leq \log(X)+O(1). \end{align*}$$

But since the conductor ${\mathfrak f}_{E/K}$ divides the minimal discriminant ${\mathfrak D}_{E/K}$ , one has $N_{K/{\mathbb Q}}({\mathfrak f}_{E/K})\leq N_{K/{\mathbb Q}}({\mathfrak D}_{E/K})$ , and thus,

$$\begin{align*}\log(N_{K/{\mathbb Q}}({\mathfrak f}_{E/K}))/\log(X)\leq 1+O(1/\log X). \end{align*}$$

It follows that $\log (A_{E/K})/\log (X)\leq 1+O(1/\log X)$ . By Corollary 6.1.2, this observation gives the inequality

$$ \begin{align*} &r(E)\phi(0)+\sum_{\gamma_j\neq 0} \phi\left(\gamma_j\frac{\log(X)}{2\pi}\right)\\ &\hspace{1cm}\leq \widehat{\phi}(0)-\frac{2}{\log(X)} \sum_{{\mathfrak a}\subseteq {\mathcal O}_K} \frac{ \widehat{a_E}({\mathfrak a}) \Lambda_K({\mathfrak a})}{N_{K/{\mathbb Q}}({\mathfrak a})}\cdot\widehat{\phi}\left(\frac{\log(N_{K/{\mathbb Q}}({\mathfrak a}))}{\log(X)}\right) +O\,\left(\frac{1}{\log X}\right). \end{align*} $$

We further simplify by rewriting the sum on the right-hand side as

$$\begin{align*}\sum_{{\mathfrak a}\subseteq {\mathcal O}_K} \frac{ \widehat{a_E}({\mathfrak a}) \Lambda_K({\mathfrak a})}{N_{K/{\mathbb Q}}({\mathfrak a})}\cdot\widehat{\phi}\left(\frac{\log(N_{K/{\mathbb Q}}({\mathfrak a}))}{\log(X)}\right) =\sum_{v\in \text{Val}_0(K)} \sum_{k\geq 1} \frac{ \widehat{a_E}({\mathfrak p}_v^k) \log(q_v)}{q_v^k}\cdot\widehat{\phi}\left(\frac{k \log(q_v)}{\log(X)}\right). \end{align*}$$

By the Weil Conjectures for Elliptic curves, $|\alpha _v|=|\beta _v|=\sqrt {q_v}$ (see, for example, [Reference SilvermanSil09, §V Theorem 2.3.1(a)]). From this, and the fact that $|b_v|\leq 1$ , we have $|\widehat {a_E}({\mathfrak p}_v^k)|\leq 2q_v^{k/2}$ . Using this, we bound the $k\geq 3$ part of the sum:

$$ \begin{align*} \sum_{v\in \text{Val}_0(K)} \sum_{k\geq 3} \frac{ \widehat{a_E}({\mathfrak p}_v^k) \log(q_v)}{q_v^k}\cdot\widehat{\phi}\left(\frac{k \log(q_v)}{\log(X)}\right) & \leq \sum_{v\in \text{Val}_0(K)} \sum_{k\geq 3} \frac{ 2 \log(q_v)}{q_v^{k/2}} ||\widehat{\phi}||_\infty\\ &\leq \sum_{v\in \text{Val}_0(K)} \frac{ 2 \log(q_v)}{q_v^{3/2} (1-q_v^{-1/2})} ||\widehat{\phi}||_\infty\\ &=O(1), \end{align*} $$

where $||f||_\infty $ is the usual $L^\infty $ -norm defined as

$$\begin{align*}||f||_\infty\stackrel{\mathrm{def}}{=}\inf\left\{a\geq 0 : m_L(\{x : |f(x)|>a\})=0\right\}, \end{align*}$$

where $m_L$ is the Lebesgue measure on ${\mathbb R}$ . Also note that for any finite set of places $S\subset \text {Val}_0(K)$ , we clearly have

$$\begin{align*}\sum_{v\in S} \sum_{k=1}^2 \frac{ \widehat{a_E}({\mathfrak p}_v^k) \log(q_v)}{q_v^k}\cdot\widehat{\phi}\left(\frac{k \log(q_v)}{\log(X)}\right)=O(1). \end{align*}$$

Therefore, setting

$$\begin{align*}U_{k}(E,\phi, X)\stackrel{\mathrm{def}}{=} \sum_{\substack{v\in \text{Val}_0(K) \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\widehat{a_E}({\mathfrak p}_v^k)\log(q_v)}{q_v^k} \cdot \widehat{\phi}\left( \frac{k\log(q_v)}{\log(X)}\right), \end{align*}$$

we have

$$\begin{align*}r(E)\phi(0)+\sum_{\gamma_j\neq 0} \phi\left(\gamma_j\frac{\log(X)}{2\pi}\right) \leq \widehat{\phi}(0)-\frac{2}{\log(X)} ( U_1(E,\phi ,X)+U_2(E,\phi,X)) +O\,\left(\frac{1}{\log X}\right). \end{align*}$$

We now average over isomorphism classes of elliptic curves. For each elliptic curve E, let $\rho _{E,j}=1+i\gamma _{E,j}$ be the nontrivial zeros of the L-function $L(E/K,s)$ . Let ${\mathcal E}_K(B)$ denote the set of isomorphism classes of elliptic curves over K with height bounded by B. Setting

$$ \begin{align*} S_k(\phi,B) &\stackrel{\mathrm{def}}{=} \frac{2}{\log(B) \#{\mathcal E}_{K}(B)} \sum_{E\in {\mathcal E}_{K}(B)} U_k(E,\phi,B)\\ &=\frac{2}{\log(B) \#{\mathcal E}_{K}(B)} \sum_{\substack{v\in \text{Val}_0(K) \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v^k} \cdot \widehat{\phi}\left( \frac{k\log(q_v)}{\log(B)}\right)\sum_{E\in{\mathcal E}_{K}(B)} \widehat{a_E}({\mathfrak p}_v^k), \end{align*} $$

we have that

(15) $$ \begin{align} \begin{split} &\phi(0) \frac{1}{\#\mathcal{E}_{K}(B)} \sum_{E\in {\mathcal E}_{K}(B)} r(E)+\frac{1}{\# {\mathcal E}_{K}(B)} \sum_{E\in{\mathcal E}_{K}(B)}\sum_{\gamma_{E,j}\neq 0}\phi\left(\gamma_{E,j}\frac{\log(B)}{2\pi}\right)\\& \quad \leq \widehat{\phi}(0)-S_1(\phi,B)-S_2(\phi,B)+O\,\left(\frac{1}{\log(B)}\right). \end{split} \end{align} $$

Remark 6.1.3. So far, we have followed the same strategy as Brumer [Reference BrumerBru92]. The main difficulty in estimating the $S_k(\phi ,B)$ is estimating the sums of the $\widehat {a_E}({\mathfrak p}_v^k)$ . Brumer does this by relating the sums to certain exponential sums. Following Cho and Jeong [Reference Cho and JeongCJ23b, Reference Cho and JeongCJ23a], we will take a slightly different approach, instead estimating the sums using our estimates for the number of elliptic curves with a prescribed trace of Frobenius (Theorem 1.1.3).

For our applications, we shall use the test function

$$\begin{align*}\phi(y)=\begin{cases} \left(\frac{\sin(\pi \nu y)}{2\pi y}\right)^2 & \text{ if } y\neq 0,\\ \nu^2/4 & \text{ if }y=0, \end{cases} \end{align*}$$

whose Fourier transform is

$$\begin{align*}\widehat{\phi}(t)=\begin{cases} \frac{1}{2}\left(\frac{\nu}{2}-\frac{|t|}{2}\right) &\text{ for } |t|\leq \nu\\ 0 &\text{ for } |t|>\nu. \end{cases} \end{align*}$$

Observe that this test function satisfies the conditions of Proposition 6.1.1 (despite not being a Schwartz function).

Our strategy will be to show that

$$\begin{align*}-S_1(\phi,B)-S_2(\phi,B)=\frac{\phi(0)}{2}+o(1). \end{align*}$$

This will imply the following upper bound for the average analytic rank of elliptic curves over the number field K:

$$\begin{align*}\frac{\widehat{\phi}(0)}{\phi(0)}+\frac{1}{2} = \frac{1}{\nu}+\frac{1}{2}. \end{align*}$$

6.2. Class number estimates

In this subsection, we give some estimates for the counting functions $H(a,q)$ , which were defined in Subsection 5.1 equation (7). These results will be used in the next subsection.

Proposition 6.2.1. For any $a\in {\mathbb Z}$ , we have

$$\begin{align*}H(a,q)\ll_{\epsilon} q^{3/2+\epsilon}, \end{align*}$$

as a function of q for any $\epsilon>0$ .

Proof. By formula (9) for $H(a,q)$ , it suffices to show $H(a^2-4q)\ll _{\epsilon } q^{1/2+\epsilon }$ and $H(-4p)\ll _{\epsilon } q^{1/2+\epsilon }$ , and for this, it will suffice to show that for all possible n, we have $H(n)\ll _{\epsilon } |n|^{1/2+\epsilon }$ . As n must be the discriminant of an order in an imaginary quadratic field, it must be negative, non-square and congruent to $0$ or $1$ modulo $4$ .

Let $\text {sqf}(n)$ denote the squarefree part of n (e.g., $\text {sqf}(12)=3$ ), and let $\text {sq}(n)$ be such that $n=\text {sqf}(n)\cdot \text {sq}(n)^2$ (e.g., $\text {sq}(12)=2$ ). For non-square negative integers D that are congruent to $0$ or $1$ modulo $4$ , let ${\mathcal O}_{D}$ denote the imaginary quadratic order of discriminant D. We can rewrite equation (8) as

$$\begin{align*}H(n)=\sum_{\substack{d|\text{sq}(n)\\ n/d^2\equiv 0,1\quad\pmod{4}}} \frac{h({\mathcal O}_{n/d^2})}{\#{\mathcal O}_{n/d^2}^{\times}}. \end{align*}$$

Define a quadratic character

$$ \begin{align*} \chi_D:{\mathbb Z}&\to {\mathbb C}^{\times}\\ m & \mapsto \genfrac{(}{)}{}{}{D}{m}. \end{align*} $$

Observe that

(16) $$ \begin{align} \lim_{s\to 1}(s-1)\zeta_{{\mathcal O}_D}(s) = \sum_{m=1}^\infty \frac{\chi_D(m)}{m} =\sum_{m=1}^\infty \chi_D(m) \int_m^{\infty}\frac{1}{x^2}dx. \end{align} $$

For any positive integer $r\in {\mathbb Z}_{\geq 1}$ , the Pólya–Vinogradov inequality gives

$$\begin{align*}\left|\sum_{m=1}^r \chi_D(m)\right|\leq 2\sqrt{|D|} \log(|D|). \end{align*}$$

In particular, the functions $f_r(x)=\sum _{m=1}^r \chi _D(m)/x^2$ are dominated by $2\sqrt {|D|}\log (|D|)/x^{2}$ on ${\mathbb R}_{\geq 1}$ . Therefore, by the dominated convergence theorem, we may switch the sum and integral in (16), so that

$$ \begin{align*} \sum_{m=1}^\infty \chi_D(m) \int_m^{\infty}\frac{1}{x^2}dx=\int_1^{\infty} \left(\sum_{m=1}^x \chi_D(m)\right) \frac{1}{x^2} dx. \end{align*} $$

Splitting this integral and again applying the Pólya–Vinogradov inequality, we find that for any $\epsilon>0$ ,

$$ \begin{align*} \lim_{s\to 1}(s-1)\zeta_{{\mathcal O}_D}(s) &= \int_1^{\sqrt{|D|}} \left(\sum_{m=1}^x \chi_D(m)\right) \frac{1}{x^2} dx + \int_{\sqrt{|D|}}^{\infty} \left(\sum_{m=1}^x \chi_D(m)\right) \frac{1}{x^2} dx \\ &\ll \int_1^{\sqrt{|D|}} \frac{x}{x^2} dx + \int_{\sqrt{|D|}}^{\infty} \frac{\sqrt{|D|}\log(|D|)}{x^2} dx \\ &\ll \log(|D|) + \log(|D|) \\ &\ll |D|^{\epsilon}. \end{align*} $$

From this and the analytic class number formula [Reference Jordan and PoonenJP20, Theorem 1.1], we have the following upper bound for class numbers of orders in imaginary quadratic fields:

$$\begin{align*}h({\mathcal O}_D)\ll |D|^{1/2+\epsilon}. \end{align*}$$

Let $\tau (n)$ denote the number of divisors of n, and recall that $\tau (n)\ll _{\epsilon } |n|^{\epsilon }$ for any $\epsilon>0$ . Combining the above observations, we have

$$\begin{align*}H(n)\leq \sum_{\substack{d|\text{sq}(n)\\ n/d^2\equiv 0,1\quad\pmod{4}}} h({\mathcal O}_{n/d^2})\ll \tau(\text{sq}(n)) |n|^{1/2+\epsilon}\ll_{\epsilon} |n|^{1/2+\epsilon}.\\[-58pt] \end{align*}$$

The next proposition gives estimates for certain sums involving $H(a,q)$ .

Proposition 6.2.2. Let q be a power of a prime $p>3$ . For any $\epsilon>0$ , we have the following estimates:

$$ \begin{align*} &\sum_{|a|\leq 2\sqrt{q}} H(a,q)= q^2-q,\\ &\sum_{|a|\leq 2\sqrt{q}} aH(a,q)=0,\\ &\sum_{|a|\leq 2\sqrt{q}} a^2 H(a,q)=q^3+O\,\left(q^{\frac{5}{2}+\epsilon}\right). \end{align*} $$

Proof. The first sum: We have that

$$\begin{align*}\sum_{|a|\leq 2\sqrt{q}} H(a,q)=\#\{(b,c)\in {\mathbb F}_q\times{\mathbb F}_q : \Delta(b,c)\neq 0\}=q^2-q. \end{align*}$$

The second sum: Pairing positive and negative terms, we have that

$$\begin{align*}\sum_{|a|\leq 2\sqrt{q}} a H(a,q)=\sum_{0<a \leq 2\sqrt{q}} \left(a H(a,q)-a H(a,q)\right)= 0, \end{align*}$$

where we have used that $H(a,q)=H(-a,q)$ .

The third sum: This sum is the most subtle. As observed in Subsection 5.1, each isomorphism class of elliptic curves over ${\mathbb F}_q$ contains exactly $q-1$ elliptic curves, which can be thought of as the size of the orbit of an elliptic curve E with respect to the weighted ${\mathbb G}_m({\mathbb F}_q)$ action, with weights $4$ and $6$ . Let ${\mathcal E}_{{\mathbb F}_q}$ denote the set of isomorphism classes of elliptic curves over ${\mathbb F}_q$ . By the orbit stabilizer theorem,

$$\begin{align*}H(a,q)=\sum_{\substack{E\in {\mathcal E}_{{\mathbb F}_q}\\ a_q(E)=a}}\frac{q-1}{\#\text{Aut}_{{\mathbb F}_q}(E)}. \end{align*}$$

Therefore, we can rewrite the third sum as

$$\begin{align*}\sum_{|a|\leq 2\sqrt{q}} a^2 H(a,q) =(q-1)\sum_{|a|\leq 2\sqrt{q}}\sum_{{\substack{E\in {\mathcal E}_{{\mathbb F}_q}\\ a_q(E)=a}}} \frac{a^2}{\#\text{Aut}_{{\mathbb F}_q}(E)} =(q-1)\sum_{E\in {\mathcal E}_{{\mathbb F}_q}} \frac{a_q(E)^2}{\#\text{Aut}_{{\mathbb F}_q}(E)}. \end{align*}$$

The final sum can be estimated using work of Birch [Reference BirchBir68] and Ihara [Reference IharaIha67], as we now explain. Write $q=p^r$ . Define an indicator function

$$\begin{align*}\delta_{2{\mathbb Z}}(r)=\begin{cases} 1 & \text{ if }r\text{ is even}\\ 0 & \text{ if }r\text{ is odd}. \end{cases} \end{align*}$$

Let $T_k(m)$ be the m-th Hecke operator on the space of weight k cusp forms of level 1, and let $\text {Tr}(T_k(m))$ denote the trace of $T_k(m)$ .

The following result is due to Birch (in the $r=1$ case) [Reference BirchBir68] and Ihara (for $r>1$ ) [Reference IharaIha67] (see [Reference Kaplan and PetrowKP17, Theorem 2]):

Theorem 6.2.3 (Birch-Ihara).

With notation as above, we have that

$$ \begin{align*} \frac{1}{q}\sum_{E\in {\mathcal E}_{{\mathbb F}_q}} \frac{a_q(E)^2}{\#\text{Aut}_{{\mathbb F}_q}(E)} \end{align*} $$

equals

$$ \begin{align*} &\frac{1}{q}\left(p^3\text{Tr}(T_4(p^{r-2}))-\text{Tr}(T_4(q)) +\frac{1}{2}\left( p^3 \sum_{i=0}^{r-2} \min\{p^i,p^{r-2-i}\}^3-\sum_{i=0}^{r} \min\{p^i, p^{r-i}\}^3\right)\right)\\& \quad + p \text{Tr}(T_2(p^{r-2})) - \text{Tr}(T_2(q)) + \frac{1}{2}\left(p \sum_{i=0}^{r-2} \min\{p^i,p^{r-2-i}\} - \sum_{i=0}^r \min\{p^i,p^{r-i}\} \right) \\& \quad + \sum_{i=0}^r p^i-p\sum_{i=0}^{r-2}p^i. \end{align*} $$

We now show that for any $\epsilon>0$ ,

$$ \begin{align*} \sum_{E\in {\mathcal E}_{{\mathbb F}_q}} \frac{a_q(E)^2}{\#\text{Aut}_{{\mathbb F}_q}(E)} = q^2 + O(q^{3/2+\epsilon}). \end{align*} $$

Using the Ramanujan–Peterson–Deligne bound for Hecke eigenvalues [Reference DeligneDel74], one can show that for any prime power $p^s$ and any $\epsilon>0$ ,

$$\begin{align*}\text{Tr}(T_k(p^s))\ll_k p^{s(k-1)/2}s\ll_k p^{s((k-1)/2+\epsilon)} \end{align*}$$

(see, for example, the introduction of [Reference PetrowPet18]). In particular,

$$ \begin{align*} p^3\text{Tr}(T_4(p^{r-2}))-\text{Tr}(T_4(q)) \ll p^3 p^{(r-2)(3/2+\epsilon)} + q^{3/2+\epsilon} \ll q^{3/2+\epsilon} \end{align*} $$

and

$$ \begin{align*} p \text{Tr}(T_2(p^{r-2})) - \text{Tr}(T_2(q))\ll p p^{(r-2)(1/2+\epsilon)}+ q^{1/2+\epsilon} \ll q^{1/2+\epsilon}. \end{align*} $$

We now estimate the sums in Theorem 6.2.3. As a function of q, the sum $\sum _{i=0}^{r} \min \{p^i, p^{r-i}\}^3$ has on the order of $\log (q)$ terms, and each term is bounded above by $\left (p^{r/2}\right )^3=q^{3/2}$ . Therefore,

$$ \begin{align*} \sum_{i=0}^{r} \min\{p^i, p^{r-i}\}^3\ll q^{3/2}\log(q)\ll q^{3/2+\epsilon}. \end{align*} $$

Similar reasoning shows that

$$ \begin{align*} p^3 \sum_{i=0}^{r-2} \min\{p^i,p^{r-2-i}\}^3\ll p^3\log(q)p^{3(r-2)/2}\ll q^{3/2+\epsilon}. \end{align*} $$

We also compute

$$ \begin{align*} p \sum_{i=0}^{r-2} \min\{p^i,p^{r-2-i}\} - \sum_{i=0}^r \min\{p^i,p^{r-i}\} =-\sum_{i\in\{0,r\}} \min\{p^i,p^{r-i}\} =-2, \end{align*} $$

and

$$ \begin{align*} \sum_{i=0}^r p^i-p\sum_{i=0}^{r-2}p^i=\sum_{i\in \{0,r\}} p^i = 1 + q. \end{align*} $$

Combining the above estimates, we find that

$$ \begin{align*} \frac{1}{q}\sum_{E\in {\mathcal E}_{{\mathbb F}_q}} \frac{a_q(E)^2}{\#\text{Aut}_{{\mathbb F}_q}(E)} &= {\mathcal O}\left(\frac{q^{3/2+\epsilon}}{q}\right) + O(q^{1/2+\epsilon})-\frac{2}{2}+(1+q)= q + O(q^{1/2+\epsilon}). \end{align*} $$

From this, we obtain

$$\begin{align*}\sum_{|a|\leq 2\sqrt{q}} a^2 H(a,q)=(q-1)\left(q^2 + O(q^{3/2+\epsilon})\right)=q^3+O(q^{5/2+\epsilon}).\\[-51pt] \end{align*}$$

6.3. Estimating $S_1(\phi ,B)$ and $S_2(\phi ,B)$

Lemma 6.3.1. Let ${\mathfrak p}\subset {\mathcal O}_K$ be a prime ideal of norm $N_{K/{\mathbb Q}}({\mathfrak p})=q$ such that $2\nmid q$ and $3\nmid q$ . Then, for any $\epsilon>0$ , we have the following upper bound:

$$\begin{align*}\sum_{E\in {\mathcal E}_K(B)} \widehat{a_E}({\mathfrak p})\ll_{\epsilon} \frac{1}{q} B^{5/6}+q^{3/2+\epsilon} B^{5/6-1/3d}. \end{align*}$$

Proof. Let ${\mathcal E}_K^{\text {mult}}(B)$ denote the set of elliptic curves over K of naive height bounded by B and multiplicative reduction at ${\mathfrak p}$ . Then, we have

$$\begin{align*}\sum_{E\in {\mathcal E}_K(B)}\widehat{a_E}({\mathfrak p})=\sum_{|a|\leq 2\sqrt{q}} \sum_{\substack{E\in{\mathcal E}_K(B)\\ a_{\mathfrak p}(E)=a}} \widehat{a_E}({\mathfrak p}) + \sum_{E\in {\mathcal E}_K^{\text{mult}}(B)} \widehat{a_E}({\mathfrak p}). \end{align*}$$

By Theorem 1.1.2, we have that

(17) $$ \begin{align} \left|\sum_{E\in {\mathcal E}_K^{\text{mult}}(B)} \widehat{a_E}({\mathfrak p})\right|\ll \#{\mathcal E}_K^{\text{mult}}(B) \ll \frac{1}{q}B^{5/6}+B^{5/6-1/3d}. \end{align} $$

By Proposition 6.2.2, Proposition 6.2.1 and the proof of Theorem 1.1.3, we have that

$$ \begin{align*} \sum_{|a|\leq 2\sqrt{q}} \sum_{\substack{E\in{\mathcal E}_K(B)\\ a_{\mathfrak p}(E)=a}} \widehat{a_E}({\mathfrak p}) &=\sum_{|a|\leq 2\sqrt{q}} a\left(\kappa\frac{H(a,q)}{q^2}\frac{q^{10}}{q^{10}-1}B^{5/6}+O\,\left(q^{-1}H(a,q)B^{5/6-1/3d}\right)\right)\\ &=0+\sum_{|a|\leq 2\sqrt{q}} O\,\left(a q^{-1}H(a,q)B^{5/6-1/3d}\right)\\ &\ll \max_{|a|\leq 2\sqrt{q}}\{H(a,q)\} B^{5/6-1/3d}\\ &\ll_{\epsilon} q^{3/2+\epsilon} B^{5/6-1/3d}. \end{align*} $$

This, together with the bound (17), implies the lemma.

Lemma 6.3.2. Let ${\mathfrak p}\subset {\mathcal O}_K$ be a prime ideal with norm $N_{K/{\mathbb Q}}({\mathfrak p})=q$ such that $2\nmid q$ and $3\nmid q$ . Then, we have the following estimate:

$$\begin{align*}\sum_{E\in {\mathcal E}_K(B)} \widehat{a_E}({\mathfrak p}^2)=-\kappa q B^{5/6}+O\,\left(q^{1/2+\epsilon}B^{\frac{5}{6}}+q^2 B^{\frac{5}{6}-\frac{1}{3d}}\right), \end{align*}$$

where the implied constant does not depend on q.

Proof. Let ${\mathcal E}_K^{good}(B)$ (resp. ${\mathcal E}_K^{\text {mult}}(B)$ ) denote the set of elliptic curves over K with good reduction (resp. multiplicative reduction) at ${\mathfrak p}$ and height bounded by B. In this case, we have that

$$\begin{align*}\sum_{E\in {\mathcal E}_K(B)} \widehat{a_E}({\mathfrak p}^2)=\sum_{|a|\leq 2\sqrt{q}} \sum_{\substack{E\in{\mathcal E}_K^{good}(B)\\ a_{\mathfrak p}(E)=a}} \widehat{a_E}({\mathfrak p}^2) + \sum_{E\in {\mathcal E}_K^{\text{mult}}(B)} \widehat{a_E}({\mathfrak p}^2). \end{align*}$$

Since $\widehat {a_E}({\mathfrak p}^2)=\widehat {a_E}({\mathfrak p})^2=1$ for all $E\in {\mathcal E}_K^{\text {mult}}(B)$ , we have, by Theorem 1.1.2, that

(18) $$ \begin{align} \left|\sum_{E\in {\mathcal E}_K^{\text{mult}}(B)} \widehat{a_E}({\mathfrak p}^2)\right|=\#{\mathcal E}_K^{\text{mult}}(B)=\kappa \frac{q-1}{q^2}\frac{q^{10}}{q^{10}-1} B^{5/6}+O(B^{5/6-1/3d})=O\,\left(B^{5/6}\right). \end{align} $$

In the case that E has good reduction at ${\mathfrak p}$ , a straightforward computation shows that $\widehat {a_E}({\mathfrak p}^2)$ equals $a_{\mathfrak p}(E)^2-2q$ . Therefore, by Proposition 6.2.2 and the proof of Theorem 1.1.3, we have that

$$ \begin{align*} \sum_{|a|\leq 2\sqrt{q}} \sum_{\substack{E\in{\mathcal E}_K^{good}(B)\\ a_{\mathfrak p}(E)=a}} \widehat{a_E}({\mathfrak p}^2) &=\sum_{|a|\leq 2\sqrt{q}} \sum_{\substack{E\in{\mathcal E}_K^{good}(B)\\ a_{\mathfrak p}(E)=a}} (a_{\mathfrak p}(E)^2-2q)\\&=\sum_{|a|\leq 2\sqrt{q}} (a^2-2q)\left(\kappa\frac{H(a,q)}{q^2}\frac{q^{10}}{q^{10}-1}B^{5/6}+O\,\left(q^{-1}H(a,q)B^{5/6-1/3d}\right)\right)\\&=\frac{\kappa q^{8}}{q^{10}-1}\left(\sum_{|a|\leq 2\sqrt{q}} a^2 H(a,q)-2q\sum_{|a|\leq 2\sqrt{q}} H(a,q)\right) B^{\frac{5}{6}}+O\,\left(q^2 B^{\frac{5}{6}-\frac{1}{3d}}\right)\\&=\frac{-\kappa q^{11}}{q^{10}-1} B^{5/6}+O\,\left(q^{1/2+\epsilon}B^{\frac{5}{6}}+q^2 B^{\frac{5}{6}-\frac{1}{3d}}\right). \end{align*} $$

Observe that

$$\begin{align*}\frac{q^{10}}{q^{10}-1}=\sum_{k=0}^{\infty} q^{-10k}=1+O(q^{-10}). \end{align*}$$

Therefore,

$$\begin{align*}\sum_{|a|\leq 2\sqrt{q}} \sum_{\substack{E\in{\mathcal E}_K^{good}(B)\\ a_{\mathfrak p}(E)=a}}\widehat{a_E}({\mathfrak p}^2)=-\kappa q B^{5/6}+O\,\left(q^{1/2+\epsilon}B^{\frac{5}{6}}+q^2 B^{\frac{5}{6}-\frac{1}{3d}}\right). \end{align*}$$

This, together with the bound (18), implies the lemma.

6.4. Bounding the average analytic rank of elliptic curves

We now prove our main result.

Theorem 1.1.1.

Let K be a number field of degree d. Assume that all elliptic curves over K are modular and that their L-functions satisfy the Riemann-Hypothesis. Then the average analytic rank of isomorphism classes of elliptic curves over K, when ordered by naive height, is bounded above by $(9d+1)/2$ .

Proof. By Lemma 6.3.1 and the observation that $\widehat {\phi }$ is supported on the interval $[-\nu ,\nu ]$ , we have that

$$ \begin{align*} S_1(\phi,B)&=\frac{2}{\log(B) \#{\mathcal E}_{K}(B)} \sum_{\substack{ v\in \text{Val}_0(K)\\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v} \cdot \widehat{\phi}\left( \frac{\log(q_v)}{\log(B)}\right)\sum_{E\in{\mathcal E}_{K}(B)} \widehat{a_E}({\mathfrak p}_v)\\ &\ll_{\epsilon} \frac{2}{\log(B) \#{\mathcal E}_{K}(B)} \sum_{\substack{v\in \text{Val}_0(K)\\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v} \cdot \widehat{\phi}\left( \frac{\log(q_v)}{\log(B)}\right)\left(\frac{1}{q_v} B^{5/6}+q_v^{3/2+\epsilon}B^{5/6-1/3d}\right)\\ &\ll \frac{2}{\log(B) \#{\mathcal E}_{K}(B)} \sum_{\substack{v\in \text{Val}_0(K) \\ q_v\leq B^\nu \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v} \left(\frac{1}{q_v} B^{5/6}+q_v^{3/2+\epsilon}B^{5/6-1/3d}\right). \end{align*} $$

By Corollary 5.1.1, we know that $\#{\mathcal E}_K(B)=\kappa B^{5/6}+O(B^{5/6-1/3d})$ . Using this and the prime ideal theorem, we have that

(19) $$ \begin{align} S_1(\phi,B) \ll_{\epsilon} \frac{2}{\log(B)} \sum_{\substack{v\in \text{Val}_0(K) \\ q_v\leq B^\nu \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v} \left(\frac{1}{q_v}+q_v^{3/2+\epsilon}B^{-1/3d}\right) \ll\frac{1+B^{(3/2+\epsilon)\nu-1/3d}}{\log(B)}. \end{align} $$

By Lemma 6.3.2, we have

$$ \begin{align*} S_2(\phi,B) &=\frac{2}{\log(B) \#{\mathcal E}_{K}(B)} \sum_{\substack{v\in \text{Val}_0(K) \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v^2} \cdot \widehat{\phi}\left(\frac{2\log(q_v)}{\log(B)}\right)\sum_{E\in{\mathcal E}_{K}(B)} \widehat{a_E}({\mathfrak p}_v^2)\\&=\frac{2}{\log(B) \#{\mathcal E}_{K}(B)} \sum_{\substack{v\in \text{Val}_0(K) \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v^2} \cdot \widehat{\phi}\left(\frac{2\log(q_v)}{\log(B)}\right)\left(-\kappa q_v B^{\frac{5}{6}}+O(q_v^{\frac{1}{2}+\epsilon}B^{\frac{5}{6}}+q_v^2 B^{\frac{5}{6}-\frac{1}{3d}})\right). \end{align*} $$

Assuming $\epsilon <1/4$ and using $\#{\mathcal E}_K(B)=\kappa B^{5/6}+O(B^{5/6-1/3d})$ , we have

$$ \begin{align*} S_2(\phi,B) &=\frac{2}{\log(B)} \sum_{\substack{v\in \text{Val}_0(K) \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v^2} \cdot \widehat{\phi}\left( \frac{2\log(q_v)}{\log(B)}\right)\left(-q_v +O(q_v^{1/2+\epsilon}+q_v^2 B^{-1/3d})\right)\\&=\frac{-2}{\log(B)} \sum_{\substack{v\in \text{Val}_0(K) \\ 2\nmid q_v,\ 3\nmid q_v}} \frac{\log(q_v)}{q_v} \cdot \widehat{\phi}\left( \frac{2\log(q_v)}{\log(B)}\right) +O\,\left(\frac{1}{\log{B}}\sum_{\substack{v\in \text{Val}_0(K) \\ q_v\leq B^{\nu/2}}}\left(\frac{1}{q_v^{3/2-\epsilon}}+\log(q_v)B^{-1/3d}\right)\right)\\ &=\frac{-1}{\log(B)} \sum_{\substack{ v\in \text{Val}_0(K) \\ q_v\leq B^{\nu/2} \\ 2\nmid q_v,\ 3\nmid q_v}} \left(\frac{\nu \log(q_v)}{2q_v} - \frac{\log(q_v)^2}{q_v\log(B)}\right) +O\,\left(\frac{1+B^{\nu/2-1/3d}}{\log(B)}\right). \end{align*} $$

Arguments using the prime ideal theorem show that

$$\begin{align*}\sum_{\substack{ v\in \text{Val}_0(K) \\ q_v\leq X}} \frac{\log(q_v)}{q_v}=\log(X)+O(1) \end{align*}$$

and

$$\begin{align*}\sum_{\substack{ v\in \text{Val}_0(K) \\ q_v\leq X}} \frac{\log(q_v)^2}{q_v}=\frac{\log(X)^2}{2}+O(\log\log(X)). \end{align*}$$

From this, it follows that

$$ \begin{align*} &\sum_{\substack{ v\in \text{Val}_0(K) \\ q_v\leq B^{\nu/2} \\ 2\nmid q_v,\ 3\nmid q_v}} \left(\frac{\nu \log(q_v)}{2q_v} - \frac{\log(q_v)^2}{q_v\log(B)}\right)\\& \quad = \frac{\nu}{2} \left(\log(B^{\nu/2})+O(1)\right)-\frac{1}{\log(B)}\left(\frac{\log(B^{\nu/2})^2}{2}+O(\log\log(B^{\nu/2}))\right) \\& \quad =\left(\frac{\nu^2}{4}\log(B)+O(1)\right)-\left(\frac{\nu^2}{8}\log(B)+O(1)\right)\\& \quad =\frac{\nu^2}{8}\log(B)+O(1). \end{align*} $$

Therefore,

$$\begin{align*}S_2(\phi,B)=\frac{-\phi(0)}{2}+O\,\left(\frac{1+B^{\nu/2-1/3d}}{\log(B)}\right). \end{align*}$$

Taking $\nu =(3d(3/2+\epsilon ))^{-1}$ and combining the above estimate for $S_2$ with our bound (19) for $S_1$ , we obtain

$$\begin{align*}-S_1(\phi,B)-S_2(\phi,B)=\frac{\phi(0)}{2}+o(1). \end{align*}$$

Substituting this into (15) and taking the limit superior as B goes to infinity gives the following upper bound for the average analytic rank of elliptic curves over K:

$$\begin{align*}\frac{1}{\nu}+\frac{1}{2}=3d(3/2+\epsilon)+\frac{1}{2}, \end{align*}$$

for any $0<\epsilon <1/4$ . Since the limit superior exists, and since $\epsilon $ can be taken arbitrarily small, we obtain the bound $(9d+1)/2$ .

Acknowledgements

The author thanks Brandon Alberts, Peter Bruin, Peter Cho, Jordan Ellenberg, Ben Kane, Irati Manterola Ayala, Steven Miller, Grant Molnar, Filip Najman, Junyeong Park, John Voight and David Zureick-Brown for helpful conversations and comments on this work. Special thanks to Bryden Cais and Doug Ulmer for their encouragement throughout this project. The author is especially grateful to Keunyoung Jeong for providing thoughtful feedback and corrections. Finally, a huge thank you to the anonymous referee for thoroughly reading this article multiple times and pointing out countless corrections and suggestions that greatly improved the accuracy and readability of the article.

Competing interest

The authors have no competing interest to declare.

Funding statement

During the completion of this work, the author was supported by the National Science Foundation, via grant DMS-2303011.

Footnotes

1 Brumer also assumed that all elliptic curves over ${\mathbb Q}$ were modular, which was unknown at the time, but later confirmed in a series of papers by Andrew Wiles and his students.

2 Here, just how large sufficiently large is depends on the lattice $\Lambda $ .

References

Bright, M. J., Browning, T. D. and Loughran, D., ‘Failures of weak approximation in families’, Compos. Math. 152(7) (2016), 14351475.CrossRefGoogle Scholar
Birch, B. J., ‘How the number of points of an elliptic curve over a fixed prime field varies’, J. London Math. Soc. 43 (1968), 5760.CrossRefGoogle Scholar
Bruin, P. and Ayala, I. M., ‘Counting rational points on weighted projective spaces over number fields’, Preprint, 2023, arXiv:2302.10967.Google Scholar
Brumer, A., ‘The average rank of elliptic curves. I’, Invent. Math. 109(3) (1992), 445472.CrossRefGoogle Scholar
Bhargava, M. and Shankar, A., ‘The average number of elements in the 4-selmer groups of elliptic curves is 7’, Preprint, 2023, arXiv:1312.7333.Google Scholar
Bhargava, M. and Shankar, A., ‘The average size of the 5-selmer group of elliptic curves is 6, and the average rank is less than 1’, Preprint, 2013, arXiv:1312.7859.Google Scholar
Bhargava, M. and Shankar, A., ‘Binary quartic forms having bounded invariants, and the boundedness of the average rank of elliptic curves’, Ann. of Math. (2) 181(1) (2015), 191242.CrossRefGoogle Scholar
Bhargava, M. and Shankar, A., ‘Ternary cubic forms having bounded invariants, and the existence of a positive proportion of elliptic curves having rank 0’, Ann. of Math. (2) 181(2) (2015), 587621.CrossRefGoogle Scholar
Birch, B. J. and Swinnerton-Dyer, H. P. F., ‘Notes on elliptic curves. II’, J. Reine Angew. Math. 218 (1965), 79108.CrossRefGoogle Scholar
Barroero, F. and Widmer, M., ‘Counting lattice points and O-minimal structures’, Int. Math. Res. Not. IMRN (18) (2014), 49324957.CrossRefGoogle Scholar
Cassels, J. W. S., An Introduction to the Geometry of Numbers (Die Grundlehren der mathematischen Wissenschaften in Einzeldarstellungen mit besonderer Berücksichtigung der Anwendungsgebiete) vol. 99 (Springer-Verlag, Berlin-Göttingen-Heidelberg, 1959).CrossRefGoogle Scholar
Cho, P. J. and Jeong, K., ‘The average analytic rank of elliptic curves with prescribed torsion’, J. Lond. Math. Soc. (2) 107(2) (2023), 616657.CrossRefGoogle Scholar
Cho, P. J. and Jeong, K., ‘On the distribution of analytic ranks of elliptic curves’, Math. Z. 305(3) (2023), Paper No. 42, 20 pp.CrossRefGoogle Scholar
Cho, P. J., Jeong, K. and Park, J., ‘The average analytic rank of elliptic curves with prescribed level structure’, Preprint, 2023, arXiv:2312.05817.Google Scholar
Cox, D. A., Primes of the Form ${x}^2+n{y}^2$ . (A Wiley-Interscience Publication) (John Wiley & Sons, Inc., New York, 1989). Fermat, class field theory and complex multiplication.Google Scholar
Cremona, J. E. and Sadek, M., ‘Local and global densities for Weierstrass models of elliptic curves’, Math. Res. Lett. 30(2) (2023), 413461.CrossRefGoogle Scholar
Darda, R., ‘Rational points of bounded height on weighted projective stacks’, PhD thesis, Université Paris Cité, 2021.Google Scholar
Davenport, H., ‘On a principle of Lipschitz’, J. London Math. Soc. 26 (1951), 179183.CrossRefGoogle Scholar
Deligne, P., ‘La conjecture de Weil. I’, Inst. Hautes Études Sci. Publ. Math. (43) (1974), 273307.CrossRefGoogle Scholar
Deng, A.-W., ‘Rational points on weighted projective spaces’, Preprint, 1998, arXiv:9812082.Google Scholar
Deuring, M., ‘Die Typen der Multiplikatorenringe elliptischer Funktionenkörper’, Abh. Math. Sem. Hansischen Univ. 14 (1941), 197272.CrossRefGoogle Scholar
Ellenberg, J. S., Satriano, M. and Zureick-Brown, D., ‘Heights on stacks and a generalized Batyrev-Manin-Malle conjecture’, Forum Math. Sigma 11 (2023), Paper No. e14, 54 pp.CrossRefGoogle Scholar
Goldston, D. A. and Gonek, S. M., ‘A note on $S(t)$ and the zeros of the Riemann zeta-function’, Bull. Lond. Math. Soc. 39(3) (2007), 482486.CrossRefGoogle Scholar
Heath-Brown, D. R., ‘The average analytic rank of elliptic curves’, Duke Math. J. 122(3) (2004), 591623.CrossRefGoogle Scholar
Ihara, Y., ‘Hecke Polynomials as congruence $\zeta$ functions in elliptic modular case’, Ann. of Math. (2) 85 (1967), 267295.CrossRefGoogle Scholar
Iwaniec, H. and Kowalski, E., Analytic Number Theory (American Mathematical Society Colloquium Publications) vol. 53 (American Mathematical Society, Providence, RI, 2004).Google Scholar
Jordan, B. W. and Poonen, B., ‘The analytic class number formula for 1-dimensional affine schemes’, Bull. Lond. Math. Soc. 52(5) (2020), 793806.CrossRefGoogle Scholar
Kaplan, N. and Petrow, I., ‘Elliptic curves over a finite field and the trace formula’, Proc. Lond. Math. Soc. (3) 115(6) (2017), 13171372.CrossRefGoogle Scholar
Lang, S., Algebraic Number Theory (Graduate Texts in Mathematics) vol. 110, second edn. (Springer-Verlag, New York, 1994).CrossRefGoogle Scholar
Lenstra, H. W. Jr.Factoring integers with elliptic curves’, Ann. of Math. (2) 126(3) (1987), 649673.CrossRefGoogle Scholar
Ayala, I. M., ‘Counting rational points on weighted projective spaces over number fields’, Master’s thesis, University of Zurich, 2021.Google Scholar
Mestre, J.-F., ‘Formules explicites et minorations de conducteurs de variétés algébriques’, Compos. Math. 58(2) (1986), 209232.Google Scholar
Miller, S. J., ‘1- and 2-level densities for families of elliptic curves: Evidence for the underlying group symmetries’, PhD thesis, Princeton University, ProQuest LLC, Ann Arbor, MI, 2002.Google Scholar
Minkowski, H., Approximationen, Diophantische. Eine Einführung in die Zahlentheorie. Mathematische Vorlesungen an der Universität Göttingen. II. Leipzig: B. G. Teubner. viii, 235 S. Mit 82 Fig., 1907.Google Scholar
Neukirch, J., Algebraic Number Theory (Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]) vol. 322 (Springer-Verlag, Berlin, 1999). Translated from the 1992 German original and with a note by Norbert Schappacher, With a foreword by G. Harder.CrossRefGoogle Scholar
Petrow, I., ‘Bounds for traces of Hecke operators and applications to modular and elliptic curves over a finite field’, Algebra Number Theory 12(10) (2018), 24712498.CrossRefGoogle Scholar
Phillips, T., ‘Rational points of bounded height on some genus zero modular curves over number fields’, Preprint, 2022, arXiv:2201.10624.Google Scholar
Schanuel, S. H., ‘Heights in number fields’, Bull. Soc. Math. France 107(4) (1979), 433449.CrossRefGoogle Scholar
Schoof, R., ‘Nonsingular plane cubic curves over finite fields’, J. Combin. Theory Ser. A 46(2) (1987), 183211.CrossRefGoogle Scholar
Shankar, A., ‘The average rank of elliptic curves over number fields’, PhD thesis, Princeton University, 2013.Google Scholar
Silverman, J. H., The Arithmetic of Elliptic Curves (Graduate Texts in Mathematics) vol. 106, second edn. (Springer, Dordrecht, 2009).CrossRefGoogle Scholar
Stanley, R. P., Enumerative Combinatorics. Vol. 1 (Cambridge Studies in Advanced Mathematics) vol. 49 (Cambridge University Press, Cambridge, 1997). With a foreword by Gian-Carlo Rota, Corrected reprint of the 1986 original.Google Scholar
Tate, J., ‘Algorithm for determining the type of a singular fiber in an elliptic pencil’, in Modular Functions of One Variable, IV (Proc. Internat. Summer School, Univ. Antwerp, Antwerp, 1972) (Lecture Notes in Math.) vol. 476 (1975), 3352.Google Scholar
van den Dries, L., Tame Topology and o-minimal Structures (London Mathematical Society Lecture Note Series) vol. 248 (Cambridge University Press, Cambridge, 1998).CrossRefGoogle Scholar
Wilkie, A. J., ‘Model completeness results for expansions of the ordered field of real numbers by restricted Pfaffian functions and the exponential function’, J. Amer. Math. Soc. 9(4) (1996), 10511094.CrossRefGoogle Scholar
Young, M. P., ‘Low-lying zeros of families of elliptic curves’, J. Amer. Math. Soc. 19(1) (2006), 205250.CrossRefGoogle Scholar
Figure 0

Table 1 Local conditions.

Figure 1

Table 2 Constants for Theorem 1.1.3.