betaFeedback

WorksMathematics

Inhomogeneous Fiber Refinement of Gallagher's Theorem

Abstract

Let s>0s>0, α,γ∈Rα,γ\in\mathbb{R}, and let ψ ⁣:N→R>0ψ\colon \mathbb{N}\to\mathbb{R}_{>0} be a non-increasing function. Consider the limsup set W_s(ψ,α,γ) = \left\{ β\in[0,1] \colon \norm{nβ-γ} < \frac{ψ(n)}{\norm{nα}^{\,s}} \ \text{ for infinitely many } n\in\mathbb{N} \right\}. Chow [Duke J. Math. 2018] settled the measure-theoretic result for s=1s=1. In this paper, we investigate the Lebesgue measure of Ws(ψ,α,γ)W_s(ψ,α,γ) for general s>0s>0.

Introduction

Metrical Diophantine approximation

A central theme in Diophantine approximation is to understand how well real numbers can be approximated by rationals. At the heart of metric number theory lies the study of the size—quantified via Lebesgue measure, Hausdorff dimension, Fourier dimension, etc.—of Diophantine sets defined by prescribed approximation conditions.

Given a function ψ:N→[0,∞)\psi:\mathbb{N}\to[0,\infty) and a number α∈[0,1]\alpha\in[0,1] (typically irrational), Diophantine approximation studies the existence of infinitely many pairs (m,n)∈Z×N(m,n)\in\mathbb{Z}\times\mathbb{N} satisfying

∣α−mn∣<ψ(n)n.(1)\left|\alpha-\frac{m}{n}\right|<\frac{\psi(n)}{n}. \tag*{(1)}

For the critical case ψ(n)=1/n\psi(n)=1/n, Dirichlet’s Theorem guarantees that (1) holds for all α∈[0,1]\alpha\in[0,1]. While this bound is asymptotically sharp for badly approximable numbers, Khintchine’s Theorem reveals that we can improve upon it for Lebesgue almost every α∈[0,1]\alpha\in[0,1].

Define the set of well-approximable numbers as

W(ψ,δ):={α∈[0,1]:∥nα−δ∥<ψ(n) for i.m. n∈N}.W(\psi,\delta):=\{\alpha\in[0,1]:\|n\alpha-\delta\|<\psi(n)\text{ for i.m. }n\in\mathbb{N}\}.

Here and in what follows “i.m.” stands for “infinitely many”, and ∥α∥\|\alpha\| denotes the distance from α\alpha to the nearest integer.

2020 Mathematics Subject Classification. 11J83 (primary); 11J70, 28A78 (secondary).

Key words and phrases. Diophantine approximation; Duffin–Schaeffer conjecture.

*Corresponding author.

This work is supported by National Key R&D Program of China (No. 2024YFA1013700) and NSFC (Nos. 12671113, 12331005 and 12601156), and the Fundamental Research Funds for the Central Universities.

Theorem 1.1 (Khintchine, [21]). Let ψ\psi be a monotonically decreasing function. Then the Lebesgue measure L(W(ψ,0))\mathcal{L}(W(\psi,0)) is given by

L(W(ψ,0))={0,if ∑n∈Nψ(n)<∞,1,if ∑n∈Nψ(n)=∞.\mathcal{L}(W(\psi,0)) = \begin{cases} 0, & \text{if } \sum_{n\in\mathbb{N}}\psi(n)<\infty,\\ 1, & \text{if } \sum_{n\in\mathbb{N}}\psi(n)=\infty. \end{cases}

Duffin and Schaeffer [11] later demonstrated that the monotonicity condition is essential: they constructed a ψ\psi supported on highly composite integers such that

∑n=1∞ψ(n)=∞,L(W(ψ,0))=0.\sum_{n=1}^{\infty}\psi(n)=\infty,\qquad\mathcal{L}(W(\psi,0))=0.

They conjectured that for almost all α\alpha, the inequality ∣nα−m∣<ψ(n)|n\alpha-m|<\psi(n) has infinitely many coprime solutions (n,m)(n,m) if and only if

∑n=1∞ϕ(n)nψ(n)=∞,(2)\sum_{n=1}^{\infty}\frac{\phi(n)}{n}\psi(n)=\infty, \tag*{(2)}

where ϕ\phi is Euler’s totient function. Following contributions by Gallagher [15], Erdős [13], Vaaler [34], and Pollington–Vaughan [27], the conjecture was resolved affirmatively by Koukoulopoulos and Maynard [22].

The inhomogeneous analogue of Khintchine’s theorem was established by Szűsz [32].

Theorem 1.2 (Szűsz, [32]). Let ψ:N→R+\psi:\mathbb{N}\to\mathbb{R}_{+} be monotonically decreasing, and let δ∈R\delta\in\mathbb{R} be fixed. Then

L(W(ψ,δ))={0,if ∑n∈Nψ(n)<∞,1,if ∑n∈Nψ(n)=∞.\mathcal{L}(W(\psi,\delta)) = \begin{cases} 0, & \text{if } \sum_{n\in\mathbb{N}}\psi(n)<\infty,\\ 1, & \text{if } \sum_{n\in\mathbb{N}}\psi(n)=\infty. \end{cases}

The non-monotonic case is more intricate: counterexamples to relaxing monotonicity were given in [6, 29]. Recently, The coprime-numerator assertion fails in general. Hauke-Treuer, Maynard and Pollington [19] constructed counterexamples for all nonzero rational shifts and certain Liouville shifts; independently, He and Liao [20] obtained counterexamples for all nonzero rational shifts and a residual set of real shifts.

An open problem in this direction is the so-called weak inhomogeneous Duffin–Schaeffer conjecture (Conjecture 1.22, [9]). It asserts that if the divergence condition (2) holds, then for almost all α∈[0,1]\alpha\in[0,1], the inequality ∥nα−δ∥<ψ(n)\|n\alpha-\delta\|<\psi(n) should have infinitely many solutions n∈Nn\in\mathbb{N}. Partial progress has been made in various aspects. For instance, it was solved for a wide class of non-monotonic functions in [5, 7, 9]. Results with extra divergence were established by Yu [35, 36]. More recently, OpenAI has announced a proof of the weak inhomogeneous Duffin–Schaeffer conjecture [24].

Theorem 1.3 (OpenAI, [24]). Let δ∈R\delta\in\mathbb{R}, and let ψ ⁣:N→[0,∞)\psi\colon\mathbb{N} \to[0,\infty). Then

L(W(ψ,δ))=1if∑n=1∞ϕ(n)nψ(n)=∞.\mathcal{L}(W(\psi,\delta)) = 1 \quad\text{if} \quad\sum_{n=1}^{\infty} \frac{\phi(n)}{n}\psi(n) = \infty.

Remark 1.4. While the divergence case is now fully resolved, a complete zero-one law for W(ψ,δ)W(\psi,\delta) remains unknown.

Multiplicative Diophantine approximation

Multiplicative Diophantine approximation is closely tied to Littlewood’s conjecture, which asserts that

lim inf⁡n→∞n∥nα∥∥nβ∥=0\liminf_{n\to\infty} n\|n\alpha\|\|n\beta\| = 0

for all α,β∈[0,1]\alpha,\beta\in[0,1]. Einsiedler, Katok, and Lindenstrauss [12] proved that the exceptional set has Hausdorff dimension zero. However, the conjecture remains open, even for natural cases like (α,β)=(2,3)(\alpha,\beta)=(\sqrt{2},\sqrt{3}). To date, the only non-trivial resolved instances are those where α\alpha and β\beta lie in the same cubic field [4, 26].

Turning to the metric theory, we define the set of multiplicative well-approximable numbers as

W×(ψ):={(α,β)∈[0,1]×[0,1]:∥nα∥∥nβ∥<ψ(n) for i.m. n∈N}.W^\times(\psi) := \{(\alpha,\beta)\in[0,1]\times[0,1] : \|n\alpha\|\|n\beta\| < \psi(n)\text{ for i.m. }n\in\mathbb{N}\}.

Theorem 1.5 (Gallagher, [16]). Let ψ ⁣:N→R+\psi\colon\mathbb{N}\to\mathbb{R}_{+} be monotonically decreasing. Then

L(W×(ψ))={0,if ∑n∈Nψ(n)log⁡n<∞,1,if ∑n∈Nψ(n)log⁡n=∞.\mathcal{L}(W^\times(\psi)) = \begin{cases} 0, & \text{if }\sum_{n\in\mathbb{N}}\psi(n)\log n<\infty,\\ 1, & \text{if }\sum_{n\in\mathbb{N}}\psi(n)\log n=\infty. \end{cases}

Without assuming monotonicity, Beresnevich, Haynes and Velani [2] proved a dichotomy for the Lebesgue measure of W×(ψ)W^\times(\psi) under additional assumptions. By removing these conditions, Frühwirth and Hauke [14] established the following: for almost all (α,β)∈[0,1]2(\alpha,\beta)\in[0,1]^2, there exist infinitely many nn such that ∣nα−m1∣∣nβ−m2∣<ψ(n)|n\alpha-m_1||n\beta-m_2|<\psi(n) with m1,m2m_1,m_2 coprime to nn, if and only if the series ∑nϕ(n)ψ(n)nlog⁡(nϕ(n)ψ(n))\sum_n\frac{\phi(n)\psi(n)}{n}\log\left(\frac{n}{\phi(n)\psi(n)}\right) diverges.

Inhomogeneous multiplicative Diophantine approximation yields a natural yet nontrivial generalization of this theory. Fix δ,γ∈R\delta,\gamma\in\mathbb{R}, Shapira [31] proved that for almost every pair (α,β)(\alpha,\beta),

lim inf⁡n→∞n∥nα−δ∥∥nβ−γ∥=0.\liminf_{n\to\infty} n\|n\alpha-\delta\|\|n\beta-\gamma\|=0.

This was subsequently improved by Gorodnik and Vishe [17], who showed that there exists ε>0\varepsilon>0 such that almost all (α,β)∈[0,1]2(\alpha,\beta)\in[0,1]^2 satisfy

lim inf⁡n→∞(log⁡5n)εn∥nα−γ∥⋅∥nβ−δ∥=0,\liminf_{n\to\infty}(\log_5 n)^\varepsilon n\|n\alpha-\gamma\|\cdot\|n\beta-\delta\|=0,

where log⁡5\log_5 denotes the fifth iterate of the function x↦max⁡{1,log⁡x}x\mapsto\max\{1,\log x\}. While the above results support Littlewood’s conjecture, Adiceam, Nesharim, and Lunnon [1] provided a counter-indication by proving that its function field analogue is false.

Fibre results are especially relevant to Littlewood-type problems because almost-everywhere results in the whole plane can avoid the most arithmetic pairs. Recall that the set of badly approximable numbers is defined as

Bad⁡:={α∈[0,1]:lim inf⁡n→∞n∥nα∥>0}.\operatorname{Bad} := \left\{ \alpha\in[0,1] : \liminf_{n\to\infty} n\lVert n\alpha\rVert> 0 \right\}.

Borel–Cantelli lemma yields that Bad⁡\operatorname{Bad} has Lebesgue measure zero, while by the Jarník–Besicovitch theorem, it has full Hausdorff dimension. Pollington and Velani [28] proved that for every α∈Bad⁡\alpha\in\operatorname{Bad}, there exists a set G⊆Bad⁡G \subseteq\operatorname{Bad} with dim⁡HG=1\dim_{\mathrm{H}} G = 1 such that

lim inf⁡n→∞nlog⁡n∥nα∥∥nβ∥≤1for every β∈G.\liminf_{n\to\infty} n\log n\lVert n\alpha\rVert\lVert n\beta\rVert\le1 \qquad\text{for every } \beta\in G.

Chow and Zafeiropoulos [8] extended this result to the fully inhomogeneous setting and relaxed the assumption on α\alpha. Related developments include the Erdős–Vaaler-type fibre counting results of Yu [37] and the almost-everywhere multiplicative results on planar lines obtained by Chow and Yang [10].

Here we recall that an irrational real number α\alpha is Liouville if

lim inf⁡n→∞nw∥nα∥=0\liminf_{n\to\infty} n^w\lVert n\alpha\rVert= 0

for all w>0w > 0. The exponent of approximation of α∈[0,1]\alpha\in[0,1] is defined as

w(α):=sup⁡{w>0:∥nα∥<n−w for i.m. n∈N}.w(\alpha) := \sup\{w > 0 : \lVert n\alpha\rVert< n^{-w} \text{ for i.m. } n \in\mathbb{N}\}.

Note that, for an irrational α\alpha, by definition, w(α)<∞w(\alpha) < \infty if and only if α\alpha is not Liouville. Beresnevich, Haynes and Velani [2] proved that if α\alpha is irrational and non-Liouville, and assuming the Duffin–Schaeffer conjecture, then

lim inf⁡n→∞n(log⁡n)2∥nα−δ∥⋅∥nβ∥=0\liminf_{n\to\infty} n(\log n)^2\lVert n\alpha-\delta\rVert\cdot\lVert n\beta\rVert= 0

for almost all β∈[0,1]\beta\in[0,1]. Later, Chow [5] proved this result unconditionally (without assuming the Duffin–Schaeffer conjecture) and in a stronger, function-based form. By finding large generalised arithmetic progressions within inhomogeneous Bohr sets, he showed that if α\alpha is irrational and non-Liouville, and ψ\psi is nonincreasing, then

∑n=1∞ψ(n)log⁡n=∞⟹∥nα−δ∥∥nβ∥<ψ(n) for i.m. n,\sum_{n=1}^{\infty}\psi(n)\log n = \infty\quad\Longrightarrow\quad\lVert n\alpha-\delta\rVert\lVert n\beta\rVert< \psi(n) \text{ for i.m. } n,

for almost every β\beta. Furthermore, by combining this with Corollary 2.1 in [2], Chow established a complete zero-one law: if α∈Bad⁡\alpha\in\operatorname{Bad} and ψ ⁣:N→R≥0\psi\colon\mathbb{N}\to\mathbb{R}_{\ge0} is a nonincreasing function, then the Lebesgue measure of the set

{β∈[0,1]:∥nβ∥<ψ(n)∥nα∥ for i.m. n∈N}\left\{ \beta\in[0,1] : \lVert n\beta\rVert< \frac{\psi(n)}{\lVert n\alpha\rVert} \text{ for i.m. } n \in\mathbb{N} \right\}

is 0 or 1 according to the convergence or divergence of the series ∑nψ(n)log⁡n\sum_n \psi(n)\log n. The fully inhomogeneous extension was established in the recent work of Chow and Technau [9]. In particular, building upon Corollary 2.1 in [2], they proved that if α∈Bad\alpha\in\mathrm{Bad}, and ψ\psi is nonincreasing, then the Lebesgue measure of the set

{β∈[0,1]:∥nβ−γ∥<ψ(n)∥nα∥ for i.m. n∈N}\left\{\beta\in[0,1] : \|n\beta-\gamma\| < \frac{\psi(n)}{\|n\alpha\|} \text{ for i.m. } n \in\mathbb{N}\right\}

is 0 or 1 according to the convergence or divergence of the series ∑n∈Nψ(n)log⁡n\sum_{n\in\mathbb{N}}\psi(n)\log n.

In this paper, we further extend the inhomogeneous approximation problem by incorporating a weight into the approximation condition. More precisely, for an irrational number α\alpha, a shift γ∈R\gamma\in\mathbb{R}, a weight parameter s>0s > 0, and a nonincreasing function ψ:N→R>0\psi:\mathbb{N}\to\mathbb{R}_{>0}, we study the Lebesgue measure of the set

Ws(ψ,α,γ):={β∈[0,1]:∥nβ−γ∥<ψ(n)∥nα∥s for i.m. n∈N}.W_s(\psi,\alpha,\gamma):=\left\{\beta\in[0,1] : \|n\beta-\gamma\| < \frac{\psi(n)}{\|n\alpha\|^s} \text{ for i.m. } n \in\mathbb{N}\right\}.

Our main results read:

Theorem 1.6. Let 0<s<10 < s < 1, α∈[0,1]∖Q\alpha\in[0,1]\setminus\mathbb{Q} and γ∈R\gamma\in\mathbb{R}. Let ψ:N→R+\psi:\mathbb{N}\to\mathbb{R}_{+} be a nonincreasing function. Then

L(Ws(ψ,α,γ))={0,if ∑n∈Nψ(n)log⁡n<∞ and w(α)<1/s,1,if ∑n∈Nψ(n)log⁡n=∞.\mathcal{L}(W_s(\psi,\alpha,\gamma)) = \begin{cases} 0, & \text{if } \sum_{n\in\mathbb{N}}\psi(n)\log n < \infty\text{ and } w(\alpha) < 1/s,\\ 1, & \text{if } \sum_{n\in\mathbb{N}}\psi(n)\log n = \infty. \end{cases}

Theorem 1.7. Let s>1s > 1, α∈[0,1]∖Q\alpha\in[0,1]\setminus\mathbb{Q} and γ∈R\gamma\in\mathbb{R}. Let ψ:N→R+\psi:\mathbb{N}\to\mathbb{R}_{+} be a nonincreasing function.

(1) If for some ε0>0\varepsilon_0 > 0,

∑n=1∞ψ(n)nw(α)s+ε0−1<∞,\sum_{n=1}^{\infty}\psi(n)n^{w(\alpha)s+\varepsilon_0-1}<\infty,

then L(Ws(ψ,α,γ))=0\mathcal{L}(W_s(\psi,\alpha,\gamma))=0.

(2) If

∑n=1∞ψ(n)ns−1log⁡log⁡n=∞,\sum_{n=1}^{\infty}\psi(n)\frac{n^{s-1}}{\log\log n}=\infty,

then L(Ws(ψ,α,γ))=1\mathcal{L}(W_s(\psi,\alpha,\gamma))=1.

Let us make the following remarks regarding Theorems 1.6 and 1.7:

  • Khintchine’s Theorem and Dirichlet’s Theorem yields that L({α∈[0,1]:w(α)=1})=1\mathcal{L}(\{\alpha\in[0,1]:w(\alpha)=1\})=1. If s<1s<1, the condition w(α)=1<1/sw(\alpha)=1<1/s is automatically satisfied for almost every α\alpha.

  • Theorem 1.7 does not yield a complete zero-one law. The parameter ε0\varepsilon_0 originates from the limit superior in the definition of w(α)w(\alpha). If qk+1≪qkw(α)q_{k+1}\ll q_k^{w(\alpha)} holds eventually, this parameter can be eliminated. Namely, if ∑ψ(n)nw(α)s−1<∞\sum\psi(n)n^{w(\alpha)s-1}<\infty, then L(W(ψ,α,γ))=0\mathcal{L}(W(\psi,\alpha,\gamma))=0. The factor 1/log⁡log⁡n1/\log\log n in the divergence condition arises naturally

from Landau’s classical estimate for Euler’s totient function [23]: ϕ(n)/n≫1/log⁡log⁡n\phi(n)/n \gg1/\log\log n. This bound is optimal, as the limit inferior of ϕ(n)/(nlog⁡log⁡n)\phi(n)/(n\log\log n) is the positive constant e−γe^{-\gamma} (where γ≈0.5772\gamma\approx0.5772 is the Euler–Mascheroni constant).

Application. For a,b>0a,b>0 and a fixed irrational α\alpha, we define the weighted multiplicatively well-approximable fiber as

Wa,b×(ψ,α,γ):={β∈[0,1]:∥nα∥a∥nβ−γ∥b<ψ(n) for i.m. n∈N}.W_{a,b}^{\times}(\psi,\alpha,\gamma):=\{\beta\in[0,1]:\|n\alpha\|^{a}\|n\beta-\gamma\|^{b}<\psi(n)\text{ for i.m. }n\in\mathbb{N}\}.

By a direct substitution, this set is equivalent to the weighted inhomogeneous set:

Wa,b×(ψ,α,γ)=Wa/b(ψ1/b,α,γ).W_{a,b}^{\times}(\psi,\alpha,\gamma)=W_{a/b}(\psi^{1/b},\alpha,\gamma).

Organization. The paper is organized as follows. In § 2, we collect some known auxiliary results that will be used in our proofs. In Section 3, we estimate the reciprocal sums ∑1≤n≤N∥nα∥−s\sum_{1\le n\le N}\|n\alpha\|^{-s} for 0<s<10<s<1 and s>1s>1, which serve as a key ingredient in our measure estimates. Sections 4 and 5 are devoted to the proofs of Theorems 1.6 and 1.7, respectively.

Notation. We use Vinogradov’s notation: A≪BA\ll B means that A≤CBA\le CB for some constant C>0C>0. We write A≫BA\gg B if B≪AB\ll A, and A≍BA\asymp B if A≪B≪AA\ll B\ll A.

Preliminaries

In this section, we present some auxiliary results required for the proofs of our main theorems.

Let α∈[0,1]∖Q\alpha\in[0,1]\setminus\mathbb{Q}, and write

α=[a1,a2,…]\alpha=[a_1,a_2,\ldots]

for its continued-fraction expansion. Further, we write the finite truncations

[a1,a2,…,aj]:=pjqj[a_1,a_2,\ldots,a_j]:=\frac{p_j}{q_j}

to be the nn-th convergent to α\alpha. The following relations can be found in [30]:

qj+1=aj+1qj+qj−1,pj+1=aj+1pj+pj−1,q_{j+1}=a_{j+1}q_j+q_{j-1},\qquad p_{j+1}=a_{j+1}p_j+p_{j-1},

and

12qj+1<∣qjα−pj∣<1qj+1.\frac{1}{2q_{j+1}}<|q_j\alpha-p_j|<\frac{1}{q_{j+1}}.

We will often make use of the quantities

Dk=qkα−pk(k≥0).D_k=q_k\alpha-p_k\qquad(k\ge0).

Then

D0={α},Dk=(−1)k∥qkα∥(k≥1),D_0=\{\alpha\},\qquad D_k=(-1)^k\|q_k\alpha\|\qquad(k\ge1),

and

12≤qk+1∣Dk∣≤1(k≥0).\frac{1}{2}\le q_{k+1}|D_k|\le1\qquad(k\ge0).

The following lemma provides a useful criterion for computing the Diophantine exponent ω(α)\omega(\alpha) via the partial quotients of its continued fraction expansion.

Lemma 2.1 ([18]). Let α∈[0,1]\Q\alpha\in[0,1] \backslash\mathbb{Q}, and let {qk}k\{q_k\}_k be the sequence of its continued fraction denominators. Then

ω(α)=lim sup⁡k→∞log⁡qk+1log⁡qk.\omega(\alpha) = \limsup_{k \to\infty} \frac{\log q_{k+1}}{\log q_k}.

The following lemma presents the Ostrowski expansion (or Ostrowski numeration) of the integers.

Lemma 2.2 ([25]). Let α=[a1,a2,…]∈[0,1]\Q\alpha= [a_1, a_2, \ldots] \in[0,1] \backslash\mathbb{Q}. Then, for every n∈Nn \in\mathbb{N}, there is a unique integer K≥0K \ge0 such that

qK≤n<qK+1,q_K \le n < q_{K+1},

and a unique sequence {ck+1}k=0∞\{c_{k+1}\}_{k=0}^{\infty} of integers such that

n=∑k=0∞ck+1qk,n = \sum_{k=0}^{\infty} c_{k+1}q_k,

where

0≤c1<a1and0≤ck+1≤ak+1(∀k≥1),0 \le c_1 < a_1 \qquad\text{and} \qquad0 \le c_{k+1} \le a_{k+1} \quad(\forall k \ge1),
ck=0wheneverck+1=ak+1 with k≥1,c_k = 0 \qquad\text{whenever} \qquad c_{k+1} = a_{k+1} \text{ with } k \ge1,
ck+1=0(∀k>K).c_{k+1} = 0 \quad(\forall k > K).

Using the unique Ostrowski expansion n=∑k=0∞ck+1qkn = \sum_{k=0}^{\infty} c_{k+1}q_k provided by Lemma 2.2, let m≥0m \ge0 be the minimal integer such that cm+1≠0c_{m+1} \ne0. The following lemma expresses ∥nα∥\lVert n\alpha\rVert in terms of the corresponding continued-fraction errors DkD_k.

Lemma 2.3 ([3], Lemma 4.1). Let α∈[0,1]\Q\alpha\in[0,1] \backslash\mathbb{Q}, n∈Nn \in\mathbb{N}, and, with reference to Lemma 2.2, let mm be the smallest integer such that cm+1≠0c_{m+1} \ne0. If m≥2m \ge2 then

∥nα∥=∣∑k=m∞ck+1Dk∣=sgn⁡(Dm)⋅∑k=m∞ck+1Dk.(3)\lVert n\alpha\rVert= \left|\sum_{k=m}^{\infty} c_{k+1}D_k\right| = \operatorname{sgn}(D_m) \cdot\sum_{k=m}^{\infty} c_{k+1}D_k. \tag*{(3)}

Also if m=1m = 1 and {α}<1/2\{\alpha\} < 1/2, then we have (3). In all other cases, ∥nα∥≥∣D2∣\lVert n\alpha\rVert\ge|D_2|, which is a positive constant depending on α\alpha only.

Lemma 2.3 underpins the following two-sided estimate for ∥nα∥\lVert n\alpha\rVert.

Lemma 2.4 ([3], Lemma 4.2). Let α∈[0,1)\Q\alpha\in[0,1) \backslash\mathbb{Q} and n∈Nn \in\mathbb{N}, and let mm be the smallest integer such that cm+1≠0c_{m+1} \ne0. If ∥nα∥<∣D2∣\lVert n\alpha\rVert< |D_2| then m≥2m \ge2, and if m≥2m \ge2 then

(cm+1−1)∣Dm∣+(am+2−cm+2)∣Dm+1∣≤∥nα∥≤(cm+1+1)∣Dm∣.(c_{m+1} - 1)|D_m| + (a_{m+2} - c_{m+2})|D_{m+1}| \le\lVert n\alpha\rVert\le(c_{m+1} + 1)|D_m|.

:::`

Fix ℓ≥0\ell\ge0. Let d1,…,dℓ+1d_1,\ldots,d_{\ell+1} be non-negative integers satisfying

d1<a1,dk+1≤ak+1(1≤k≤ℓ),d_1 < a_1,\qquad d_{k+1} \le a_{k+1}\quad(1 \le k \le\ell),

and

dk=0whenever dk+1=ak+1,1≤k≤ℓ.d_k=0\quad\text{whenever }d_{k+1}=a_{k+1},\quad1\le k\le\ell.

We define A(d1,…,dℓ+1)A(d_1,\ldots,d_{\ell+1}) to be the set of all integers n∈Nn\in\mathbb{N} whose Ostrowski expansion begins with the fixed coefficients d1,…,dℓ+1d_1,\ldots,d_{\ell+1}. The corresponding truncated set is defined as

AN(d1,…,dℓ+1):=A(d1,…,dℓ+1)∩{1,…,N}.(4)A_N(d_1,\ldots,d_{\ell+1}):=A(d_1,\ldots,d_{\ell+1})\cap\{1,\ldots,N\}. \tag*{(4)}

The following lemma provides a precise bound on the cardinality of these truncated sets.

Lemma 2.5 ([3], Lemma 5.3). Let N≥1N\ge1 be an integer. Provided that AN(d1,…,dℓ+1)A_N(d_1,\ldots,d_{\ell+1}) is non-empty, i.e., #AN(d1,…,dℓ+1)≥1\# A_N(d_1,\ldots,d_{\ell+1})\ge1, we have

N3qℓ+1≤#AN(d1,…,dℓ+1)≤3Nqℓ+1+1.\frac{N}{3q_{\ell+1}}\le\# A_N(d_1,\ldots,d_{\ell+1})\le\frac{3N}{q_{\ell+1}}+1.

Estimates for reciprocal sums involving ∥nα∥−s\lVert n\alpha\rVert^{-s}

This section is devoted to estimating the reciprocal sum

∑1≤n≤N1∥nα∥s.\sum_{1\le n\le N}\frac{1}{\lVert n\alpha\rVert^s}.

The core idea originates from Beresnevich, Haynes and Velani [3]. For a given NN, let KK be the integer uniquely determined by

qK≤N<qK+1.q_K\le N<q_{K+1}.

Throughout this section, we assume N≥q3N\ge q_3. This choice of KK allows us to partition the integers 1≤n≤N1\le n\le N into two disjoint sets according to their residues modulo qKq_K: the regular class ΩN,1\Omega_{N,1} and the exceptional class ΩN,2\Omega_{N,2}, defined as

ΩN,1:={n∈N:1≤n≤N, n≢0(modqK) and n≢qK−1(modqK)},\Omega_{N,1}:=\{n\in\mathbb{N}:1\le n\le N,\ n\not\equiv0\pmod{q_K}\text{ and }n\not\equiv q_K-1\pmod{q_K}\},
ΩN,2:={n∈N:1≤n≤N, n≡0(modqK) or n≡qK−1(modqK)}.\Omega_{N,2}:=\{n\in\mathbb{N}:1\le n\le N,\ n\equiv0\pmod{q_K}\text{ or }n\equiv q_K-1\pmod{q_K}\}.

The exceptional class ΩN,2\Omega_{N,2} requires special attention, since it corresponds to particularly small values of ∥nα∥\lVert n\alpha\rVert.

Proposition 3.1. Let 0<s<10<s<1, α∈R∖Q\alpha\in\mathbb{R}\setminus\mathbb{Q} and N≥q3N\ge q_3. Let KK be the largest integer satisfying qK≤Nq_K\le N. Then

∑n∈ΩN,11∥nα∥s≍Nand∑n∈ΩN,21∥nα∥s≍N1−sqK+1sqK1−s.\sum_{n\in\Omega_{N,1}}\frac{1}{\lVert n\alpha\rVert^s}\asymp N \qquad\text{and}\qquad \sum_{n\in\Omega_{N,2}}\frac{1}{\lVert n\alpha\rVert^s}\asymp \frac{N^{1-s}q_{K+1}^s}{q_K^{1-s}}.

Proposition 3.2. Let s>1s>1, α∈R∖Q\alpha\in\mathbb{R}\setminus\mathbb{Q} and N≥q3N\ge q_3. Let KK be the largest integer satisfying qK≤Nq_K\le N. Then

∑n∈ΩN,11∥nα∥s≍NqKs−1and∑n∈ΩN,21∥nα∥s≍qK+1s.\sum_{n\in\Omega_{N,1}}\frac{1}{\lVert n\alpha\rVert^s}\asymp Nq_K^{s-1} \qquad\text{and}\qquad \sum_{n\in\Omega_{N,2}}\frac{1}{\lVert n\alpha\rVert^s}\asymp q_{K+1}^s.

The proofs of the two propositions are structured according to the partition of the integers into regular and exceptional classes. Specifically, Subsection 3.1 is devoted to the estimates for the regular class ΩN,1\Omega_{N,1}, while the contribution from the exceptional class ΩN,2\Omega_{N,2} will be treated in Subsection 3.2.

Estimates for the regular class

We begin by estimating the sum over the regular class ΩN,1\Omega_{N,1}. We split the sum into two parts: ∥nα∥<∣D2∣\lVert n\alpha\rVert<|D_2| and ∥nα∥≥∣D2∣\lVert n\alpha\rVert\ge|D_2|:

∑n∈ΩN,11∥nα∥s=∑n∈ΩN,1∥nα∥<∣D2∣1∥nα∥s+∑n∈ΩN,1∥nα∥≥∣D2∣1∥nα∥s.\sum_{n\in\Omega_{N,1}}\frac{1}{\lVert n\alpha\rVert^s} = \sum_{\substack{n\in\Omega_{N,1}\\\lVert n\alpha\rVert<|D_2|}}\frac{1}{\lVert n\alpha\rVert^s} + \sum_{\substack{n\in\Omega_{N,1}\\\lVert n\alpha\rVert\ge|D_2|}}\frac{1}{\lVert n\alpha\rVert^s}.

Upper bounds for ∑n∈ΩN,1∥nα∥−s\sum_{n\in\Omega_{N,1}}\lVert n\alpha\rVert^{-s}. For integers uu and aa satisfying 1≤u≤K+11\le u\le K+1 and 1≤a≤au+11\le a\le a_{u+1}, define

CN(u,a):={1≤n≤N:the Ostrowski coefficients {ck+1}k=0∞ of n satisfyc1=⋯=cu−1=0, and either(i) cu=0 and cu+1=a+1, or(ii) cu=1 and cu+1=au+1−a}.C_N(u,a):=\left\{1\le n\le N: \begin{aligned} &\text{the Ostrowski coefficients }\{c_{k+1}\}_{k=0}^{\infty}\text{ of }n\text{ satisfy}\\ &c_1=\cdots=c_{u-1}=0,\text{ and either}\\ &\text{(i) }c_u=0\text{ and }c_{u+1}=a+1,\text{ or}\\ &\text{(ii) }c_u=1\text{ and }c_{u+1}=a_{u+1}-a \end{aligned} \right\}.

Clearly, each CN(u,a)C_N(u,a) is a subset of {1,…,N}\{1,\ldots,N\} by definition. Furthermore, the uniqueness of the Ostrowski expansion immediately implies that these sets are pairwise disjoint; that is, for any two distinct pairs (u1,a)≠(u2,a′)(u_1,a)\ne(u_2,a'), we have CN(u1,a)∩CN(u2,a′)=∅C_N(u_1,a)\cap C_N(u_2,a')=\varnothing. We now focus on the integers nn for which ∥nα∥<∣D2∣\lVert n\alpha\rVert<|D_2|. As the following lemma shows, these integers are fully covered by this disjoint union.

Lemma 3.3.

{1≤n≤N:∥nα∥<∣D2∣}⊆⋃u=2K+1⋃a=1au+1CN(u,a).\{1\le n\le N:\lVert n\alpha\rVert<|D_2|\}\subseteq\bigcup_{u=2}^{K+1}\bigcup_{a=1}^{a_{u+1}}C_N(u,a).

Proof. Take nn in the set on the left, and let mm be the smallest index such that cm+1≠0c_{m+1}\ne0 in its Ostrowski expansion. By Lemma 2.4, we have m≥2m\ge2. Since n≤N<qK+1n\le N<q_{K+1}, we also have m≤Km\le K. Thus, 2≤m≤K2\le m\le K.

We now distinguish two cases based on the value of cm+1c_{m+1}:

  • Case 1: cm+1>1c_{m+1}>1. We set u=mu=m and a=cm+1−1a=c_{m+1}-1. Then 1≤a≤am+11\le a\le a_{m+1}, and it follows directly from the definition that n∈CN(u,a)n\in C_N(u,a).

  • Case 2: cm+1=1c_{m+1}=1. We set u=m+1u=m+1 and a=am+2−cm+2a=a_{m+2}-c_{m+2}. Since 0≤cm+2<am+20\le c_{m+2}<a_{m+2}, we have 1≤a≤am+21\le a\le a_{m+2}. Hence, n∈CN(m+1,a)n\in C_N(m+1,a).

In either case, we obtain 2≤u≤K+12 \le u \le K+1 and 1≤a≤au+11 \le a \le a_{u+1}. This proves that nn belongs to the disjoint union on the right. □\square

By Lemma 3.3, any integer nn satisfying ∥nα∥<∣D2∣\lVert n\alpha\rVert< |D_2| must belong to some CN(u,a)C_N(u,a) with 2≤u≤K+12 \le u \le K+1 and 1≤a≤au+11 \le a \le a_{u+1}. In particular, if n∈ΩN,1n \in\Omega_{N,1}, the range of uu can be further restricted. The following lemma shows that the boundary cases u∈{K,K+1}u \in\{K,K+1\} always fall into the exceptional class ΩN,2\Omega_{N,2}. Consequently, for n∈ΩN,1n \in\Omega_{N,1} with ∥nα∥<∣D2∣\lVert n\alpha\rVert< |D_2|, the associated parameters must satisfy 2≤u≤K−12 \le u \le K-1 and 1≤a≤au+11 \le a \le a_{u+1}.

Lemma 3.4. If n∈⋃u=KK+1⋃a=1au+1CN(u,a)n \in\bigcup_{u=K}^{K+1}\bigcup_{a=1}^{a_{u+1}} C_N(u,a), then

n≡0(modqK)orn≡qK−1(modqK).n \equiv0 \pmod{q_K}\quad\text{or}\quad n \equiv q_{K-1} \pmod{q_K}.

Proof. If n∈CN(K,a)n \in C_N(K,a), then n≤N<qK+1n \le N < q_{K+1}, and its Ostrowski expansion can only be (a+1)qK(a+1)q_K or qK−1+(aK+1−a)qKq_{K-1}+(a_{K+1}-a)q_K. Thus, n≡0(modqK)n \equiv0 \pmod{q_K} or n≡qK−1(modqK)n \equiv q_{K-1} \pmod{q_K}, respectively.

If n∈CN(K+1,a)n \in C_N(K+1,a), its Ostrowski expansion can only be n=qK+(aK+2−a)qK+1n=q_K+(a_{K+2}-a)q_{K+1}. Since n≤N<qK+1n \le N < q_{K+1}, this forces n=qKn=q_K, giving n≡0(modqK)n \equiv0 \pmod{q_K}. □\square

Lemma 3.5. We have

∑n∈ΩN,1∥nα∥<∣D2∣1∥nα∥s≪{N,if 0<s<1,NqKs−1,if s>1.\sum_{\substack{n\in\Omega_{N,1}\\ \lVert n\alpha\rVert<|D_2|}} \frac{1}{\lVert n\alpha\rVert^s} \ll \begin{cases} N, & \text{if } 0<s<1,\\ Nq_K^{s-1}, & \text{if } s>1. \end{cases}

Proof. First, by Lemmas 3.3 and 3.4, we obtain

∑n∈ΩN,1∥nα∥<∣D2∣1∥nα∥s≤∑u=2K−1∑a=1au+1∑n∈CN(u,a)1∥nα∥s.\sum_{\substack{n\in\Omega_{N,1}\\ \lVert n\alpha\rVert<|D_2|}} \frac{1}{\lVert n\alpha\rVert^s} \le\sum_{u=2}^{K-1}\sum_{a=1}^{a_{u+1}}\sum_{n\in C_N(u,a)}\frac{1}{\lVert n\alpha\rVert^s}.

For n∈CN(u,a)n \in C_N(u,a), Lemma 2.4 yields

∥nα∥≥a∣Du∣≥a2qu+1.\lVert n\alpha\rVert\ge a|D_u| \ge\frac{a}{2q_{u+1}}.

Furthermore, the set CN(u,a)C_N(u,a) can be explicitly decomposed into a disjoint union of two truncated sets:

CN(u,a)=AN(0,…,0⏟u times,a+1)⊔AN(0,…,0⏟u−1 times,1,au+1−a),C_N(u,a)=A_N(\underbrace{0,\ldots,0}_{u\ \text{times}},a+1)\sqcup A_N(\underbrace{0,\ldots,0}_{u-1\ \text{times}},1,a_{u+1}-a),

where AN(⋅)A_N(\cdot) defined in (4). Applying Lemma 2.5, we obtain

#CN(u,a)≤2(3Nqu+1+1)≤8Nqu+1,u≤K−1.\#C_N(u,a)\le2\left(\frac{3N}{q_{u+1}}+1\right)\le\frac{8N}{q_{u+1}},\qquad u\le K-1.

Hence, for 0<s<10 < s < 1, we have

∑n∈ΩN,1∥nα∥<∣D2∣1∥nα∥s≪N∑u=2K−1qu+1squ+1∑a=1au+11as≪N∑u=2K−1au+11−squ+11−s≪N.\sum_{\substack{n\in\Omega_{N,1}\\ \lVert n\alpha\rVert<\lvert D_2\rvert}} \frac{1}{\lVert n\alpha\rVert^s} \ll N\sum_{u=2}^{K-1}\frac{q_{u+1}^s}{q_{u+1}}\sum_{a=1}^{a_{u+1}}\frac{1}{a^s}\ll N\sum_{u=2}^{K-1}\frac{a_{u+1}^{1-s}}{q_{u+1}^{1-s}}\ll N.

For s>1s > 1, we have

∑n∈ΩN,1∥nα∥<∣D2∣1∥nα∥s≪N∑u=2K−1qu+1squ+1∑a=1au+11as≪N∑u=2K−1qu+1s−1≪NqKs−1.\sum_{\substack{n\in\Omega_{N,1}\\ \lVert n\alpha\rVert<\lvert D_2\rvert}} \frac{1}{\lVert n\alpha\rVert^s} \ll N\sum_{u=2}^{K-1}\frac{q_{u+1}^s}{q_{u+1}}\sum_{a=1}^{a_{u+1}}\frac{1}{a^s}\ll N\sum_{u=2}^{K-1}q_{u+1}^{s-1}\ll Nq_K^{s-1}.

Lemma 3.6. For any s>0s > 0, we have

∑1≤n≤N∥nα∥≥∣D2∣1∥nα∥s≤2q3sN.\sum_{\substack{1\le n\le N\\ \lVert n\alpha\rVert\ge\lvert D_2\rvert}} \frac{1}{\lVert n\alpha\rVert^s}\le2q_3^sN.

Proof. A direct calculation yields

∑1≤n≤N∥nα∥≥∣D2∣1∥nα∥s≤N∣D2∣s≤2q3sN.\sum_{\substack{1\le n\le N\\ \lVert n\alpha\rVert\ge\lvert D_2\rvert}} \frac{1}{\lVert n\alpha\rVert^s}\le\frac{N}{\lvert D_2\rvert^s}\le2q_3^sN.

Combining Lemmas 3.5 and 3.6, we obtain

∑n∈ΩN,11∥nα∥s≪{N+q3sN,if 0<s<1,NqKs−1+q3sN,if s>1.(5)\sum_{n\in\Omega_{N,1}}\frac{1}{\lVert n\alpha\rVert^s}\ll \begin{cases} N+q_3^sN, & \text{if } 0<s<1,\\ Nq_K^{s-1}+q_3^sN, & \text{if } s>1. \end{cases} \tag*{(5)}

Lower bounds for ∑n∈ΩN,1∥nα∥−s\sum_{n\in\Omega_{N,1}}\lVert n\alpha\rVert^{-s}. For mm, a∈Na\in\mathbb{N}, recall that

AN(0,…,0⏟m times,a)={1≤n≤N:the Ostrowski expansion of n satisfiesck+1=0 for all k<m,cm+1=a}.A_N(\underbrace{0,\ldots,0}_{m\text{ times}},a)=\left\{1\le n\le N:\begin{array}{l} \text{the Ostrowski expansion of }n\text{ satisfies}\\ c_{k+1}=0\text{ for all }k<m,\quad c_{m+1}=a \end{array}\right\}.

To simplify notation, we write A(0m,a)A(0^m,a) for AN(0,…,0,a)A_N(0,\ldots,0,a). Since 1≤n≤N<qK+11\le n\le N<q_{K+1}, we have m≤Km\le K. Note that if m=Km=K, then any n∈A(0m,a)n\in A(0^m,a) must be of the form n=aqKn=aq_K, which yields n≡0(modqK)n\equiv0\pmod{q_K}. Thus, we restrict our attention to m≤K−1m\le K-1.

Furthermore, it is easily checked that if n∈A(0m,a)n\in A(0^m,a) satisfies one of the following conditions:

  • 1≤a≤am+11\le a\le a_{m+1} and m≤K−2m\le K-2, or

  • 2≤a≤am+12\le a\le a_{m+1} and m=K−1m=K-1,

then n∈ΩN,1n \in\Omega_{N,1}.

Hence, for 0<s<10 < s < 1, we obtain

∑n∈ΩN,11∥nα∥s≥∑m=2K−2∑a=1am+1∑n∈A(0m,a)1∥nα∥s+∑a=2aK∑n∈A(0K−1,a)1∥nα∥s≫∑m=2K−2∑a=1am+1Nqm+11as∣Dm∣s+∑a=2aKNqK1as∣DK−1∣s≫N(∑m=2K−2∑a=1am+11qm+11−sas+∑a=2aK1qK1−sas)=N(∑m=2K−1∑a=1am+11qm+11−sas−1qK1−s)≫N∑m=2K−1am+11−sqm+11−s−NqK1−s≫N(∑m=2K−11qm1−s−1qK1−s)≫N∑m=2K1qm1−s≫N,(6)\begin{aligned} \sum_{n\in\Omega_{N,1}}\frac{1}{\|n\alpha\|^s} &\ge\sum_{m=2}^{K-2}\sum_{a=1}^{a_{m+1}}\sum_{n\in A(0^m,a)}\frac{1}{\|n\alpha\|^s} +\sum_{a=2}^{a_K}\sum_{n\in A(0^{K-1},a)}\frac{1}{\|n\alpha\|^s}\\ &\gg\sum_{m=2}^{K-2}\sum_{a=1}^{a_{m+1}}\frac{N}{q_{m+1}}\frac{1}{a^s|D_m|^s} +\sum_{a=2}^{a_K}\frac{N}{q_K}\frac{1}{a^s|D_{K-1}|^s}\\ &\gg N\left(\sum_{m=2}^{K-2}\sum_{a=1}^{a_{m+1}}\frac{1}{q_{m+1}^{1-s}a^s} +\sum_{a=2}^{a_K}\frac{1}{q_K^{1-s}a^s}\right)\\ &=N\left(\sum_{m=2}^{K-1}\sum_{a=1}^{a_{m+1}}\frac{1}{q_{m+1}^{1-s}a^s} -\frac{1}{q_K^{1-s}}\right)\\ &\gg N\sum_{m=2}^{K-1}\frac{a_{m+1}^{1-s}}{q_{m+1}^{1-s}}-\frac{N}{q_K^{1-s}}\\ &\gg N\left(\sum_{m=2}^{K-1}\frac{1}{q_m^{1-s}}-\frac{1}{q_K^{1-s}}\right)\\ &\gg N\sum_{m=2}^{K}\frac{1}{q_m^{1-s}}\gg N, \tag*{(6)} \end{aligned}

where the second inequality follows from Lemmas 2.4 and 2.5.

For s>1s > 1, we have

∑n∈ΩN,11∥nα∥s≥∑m=2K−2∑a=1am+1∑n∈A(0m,a)1∥nα∥s+∑a=2aK∑n∈A(0K−1,a)1∥nα∥s≫∑a=2aKNqK1as∣DK−1∣s≍NqKs−1.(7)\begin{aligned} \sum_{n\in\Omega_{N,1}}\frac{1}{\|n\alpha\|^s} &\ge\sum_{m=2}^{K-2}\sum_{a=1}^{a_{m+1}}\sum_{n\in A(0^m,a)}\frac{1}{\|n\alpha\|^s} +\sum_{a=2}^{a_K}\sum_{n\in A(0^{K-1},a)}\frac{1}{\|n\alpha\|^s}\\ &\gg\sum_{a=2}^{a_K}\frac{N}{q_K}\frac{1}{a^s|D_{K-1}|^s} \asymp Nq_K^{s-1}. \tag*{(7)} \end{aligned}

Combining the upper bounds (5) with the lower bounds (6) and (7), we obtain

∑n∈ΩN,11∥nα∥s≍{N,if 0<s<1,NqKs−1,if s>1,(8)\sum_{n\in\Omega_{N,1}}\frac{1}{\|n\alpha\|^s} \asymp \begin{cases} N, & \text{if } 0<s<1,\\ Nq_K^{s-1}, & \text{if } s>1, \end{cases} \tag*{(8)}

which completes the proof for the regular classes.

Estimates for the exceptional classes

We now treat the exceptional classes n≡0,qK−1(modqK)n\equiv0,q_{K-1}\pmod{q_K}. We begin with the class n≡0(modqK)n\equiv0\pmod{q_K}.

First, note that N≥q3N\ge q_3 implies K≥3K\ge3. Let 1≤n≤N1\le n\le N satisfy n≡0(modqK)n\equiv0\pmod{q_K}. Then

n=aqKwith1≤a≤NqK.n=aq_K\quad\text{with}\quad1\le a\le\frac{N}{q_K}.

Further, since N<qK+1=aK+1qK+qK−1N<q_{K+1}=a_{K+1}q_K+q_{K-1}, we have

a≤NqK<aK+1+qK−1qK<aK+1+1.a\le\frac{N}{q_K}<a_{K+1}+\frac{q_{K-1}}{q_K}<a_{K+1}+1.

which implies a≤aK+1a\le a_{K+1}. Thus, n=aqKn=aq_K is the Ostrowski expansion of nn. Since the only non-zero Ostrowski coefficient is cK+1=ac_{K+1}=a, it follows from Lemma 2.3 that

∥nα∥=a∣DK∣≍aqK+1.\lVert n\alpha\rVert=a|D_K|\asymp\frac{a}{q_{K+1}}.

For 0<s<10<s<1, this gives

∑1≤n≤Nn≡0(modqK)1∥nα∥s≍qK+1s∑1≤a≤N/qK1as≍N1−sqK+1sqK1−s.(9)\sum_{\substack{1\le n\le N\\ n\equiv0\pmod{q_K}}}\frac{1}{\lVert n\alpha\rVert^s}\asymp q_{K+1}^s\sum_{1\le a\le N/q_K}\frac{1}{a^s}\asymp\frac{N^{1-s}q_{K+1}^s}{q_K^{1-s}}. \tag*{(9)}

For s>1s>1, we similarly obtain

∑1≤n≤Nn≡0(modqK)1∥nα∥s≲qK+1s∑1≤a≤N/qK1as≲qK+1s.(10)\sum_{\substack{1\le n\le N\\ n\equiv0\pmod{q_K}}}\frac{1}{\lVert n\alpha\rVert^s}\lesssim q_{K+1}^s\sum_{1\le a\le N/q_K}\frac{1}{a^s}\lesssim q_{K+1}^s. \tag*{(10)}

For the shifted residue class n≡qK−1(modqK)n\equiv q_{K-1}\pmod{q_K}, we write nn as

n=qK−1+(aK+1−b)qKwithaK+1−N−qK−1qK≤b≤aK+1n=q_{K-1}+(a_{K+1}-b)q_K\quad\text{with}\quad a_{K+1}-\frac{N-q_{K-1}}{q_K}\le b\le a_{K+1}

for some integer bb. Since N<qK+1=aK+1qK+qK−1N<q_{K+1}=a_{K+1}q_K+q_{K-1}, we have

aK+1−N−qK−1qK>0,and hence b≥1.a_{K+1}-\frac{N-q_{K-1}}{q_K}>0,\qquad\text{and hence }b\ge1.

Since K≥3K\ge3, we have K−1≥2K-1\ge2. Applying Lemma 2.3 yields

∥nα∥=∣DK−1∣−(aK+1−b)∣DK∣=∣DK+1∣+b∣DK∣≍b∣DK∣≍bqK+1.\begin{aligned} \lVert n\alpha\rVert&=|D_{K-1}|-(a_{K+1}-b)|D_K|\\ &=|D_{K+1}|+b|D_K|\asymp b|D_K|\asymp\frac{b}{q_{K+1}}. \end{aligned}

Thus, for 0<s<10<s<1,

∑1≤n≤Nn≡qK−1(modqK)1∥nα∥s≪∑b∈ZaK+1−(N−qK−1)/qK≤b≤aK+1qK+1sbs≪qK+1s∑1≤b≤N/qK1bs≪qK+1sN1−sqK1−s.(11)\sum_{\substack{1\le n\le N\\ n\equiv q_{K-1}\pmod{q_K}}}\frac{1}{\lVert n\alpha\rVert^s}\ll\sum_{\substack{b\in\mathbb{Z}\\ a_{K+1}-(N-q_{K-1})/q_K\le b\le a_{K+1}}}\frac{q_{K+1}^s}{b^s}\ll q_{K+1}^s\sum_{1\le b\le N/q_K}\frac{1}{b^s}\ll\frac{q_{K+1}^sN^{1-s}}{q_K^{1-s}}. \tag*{(11)}

and for s>1s>1, we similarly obtain

∑1≤n≤Nn≡qK−1(modqK)1∥nα∥s≪∑b∈ZaK+1−(N−qK−1)/qK≤b≤aK+1qK+1sbs≪qK+1s.(12)\sum_{\substack{1\le n\le N\\ n\equiv q_{K-1}\pmod{q_K}}}\frac{1}{\lVert n\alpha\rVert^s}\ll\sum_{\substack{b\in\mathbb{Z}\\ a_{K+1}-(N-q_{K-1})/q_K\le b\le a_{K+1}}}\frac{q_{K+1}^s}{b^s}\ll q_{K+1}^s. \tag*{(12)}

Combining (9) and (11) for 0<s<10 < s < 1, and (10) and (12) for s>1s > 1, we conclude that

∑n∈ΩN,21∥nα∥s≲{N1−sqK+1sqK1−s,if 0<s<1,qK+1s,if s>1.(13)\sum_{n\in\Omega_{N,2}} \frac{1}{\lVert n\alpha\rVert^{s}} \lesssim\begin{cases} \dfrac{N^{1-s}q_{K+1}^{s}}{q_{K}^{1-s}}, & \text{if } 0 < s < 1, \\ q_{K+1}^{s}, & \text{if } s > 1. \end{cases} \tag*{(13)}

Finally, together with (8) and (13), this completes the proofs of Propositions 3.1 and 3.2. In what follows, we obtain the following asymptotic formula for the full reciprocal sum:

∑1≤n≤N1∥nα∥s≲{N+N1−sqK+1sqK1−s,if 0<s<1,NqKs−1+qK+1s,if s>1.(14)\sum_{1\le n\le N} \frac{1}{\lVert n\alpha\rVert^{s}} \lesssim\begin{cases} N+\dfrac{N^{1-s}q_{K+1}^{s}}{q_{K}^{1-s}}, & \text{if } 0 < s < 1, \\ Nq_{K}^{s-1}+q_{K+1}^{s}, & \text{if } s > 1. \end{cases} \tag*{(14)}

Proof of Theorem 1.6

For clarity, the proof is divided into the following two parts:

  • the convergence part for 0<s<10 < s < 1,

  • the divergence part for 0<s<10 < s < 1.

Convergence part (0<s<10 < s < 1). Choose a sufficiently small ε0>0\varepsilon_{0} > 0 such that w‾(α)+ε0<1/s\overline{w}(\alpha) + \varepsilon_{0} < 1/s. Then, by Lemma 2.1, for all sufficiently large KK, we have

qK+1≤qKw‾(α)+ε0≤Nw‾(α)+ε0<N1/s.q_{K+1} \le q_{K}^{\overline{w}(\alpha)+\varepsilon_{0}} \le N^{\overline{w}(\alpha)+\varepsilon_{0}} < N^{1/s}.

Combining this with (14), we obtain

∑1≤n≤N1∥nα∥s≲N.\sum_{1\le n\le N} \frac{1}{\lVert n\alpha\rVert^{s}} \lesssim N.

Consequently, applying Abel’s summation formula yields

∑1≤n≤Nψ(n)∥nα∥s=∑n=1N−1(ψ(n)−ψ(n+1))∑m=1n1∥mα∥s+ψ(N)∑m=1N1∥mα∥s\sum_{1\le n\le N}\frac{\psi(n)}{\lVert n\alpha\rVert^{s}}=\sum_{n=1}^{N-1}(\psi(n)-\psi(n+1))\sum_{m=1}^{n}\frac{1}{\lVert m\alpha\rVert^{s}}+\psi(N)\sum_{m=1}^{N}\frac{1}{\lVert m\alpha\rVert^{s}}
≲∑n=1N−1(ψ(n)−ψ(n+1))n+ψ(N)N=∑1≤n<Nψ(n).\begin{aligned} \lesssim\sum_{n=1}^{N-1}(\psi(n)-\psi(n+1))n+\psi(N)N \\ &=\sum_{1\le n<N}\psi(n). \end{aligned}

If ∑ψ(n)<∞\sum\psi(n)<\infty, the first Borel–Cantelli lemma yields L(Ws(ψ,α,γ))=0\mathcal{L}(W_{s}(\psi,\alpha,\gamma))=0, completing the proof of the convergence part.

Divergence part (0<s<10 < s < 1). Since ψ(n)≤ψ(n)/∥nα∥s\psi(n)\le\psi(n)/\lVert n\alpha\rVert^{s}, we have the inclusion W(ψ,γ)⊂Ws(ψ,α,γ)W(\psi,\gamma)\subset W_{s}(\psi,\alpha,\gamma). By Theorem 1.2, the divergence condition ∑n=1∞ψ(n)=∞\sum_{n=1}^{\infty}\psi(n)=\infty ensures that W(ψ,γ)W(\psi,\gamma) has full measure. It follows immediately that L(Ws(ψ,α,γ))=1\mathcal{L}(W_{s}(\psi,\alpha,\gamma))=1.

Proof of Theorem 1.7

The proof proceeds in two steps:

  • the convergence part for s>1s > 1,

  • the divergence part for s>1s > 1.

Convergence part (s>1s > 1). For ε>0\varepsilon> 0, combining Lemma 2.1 with qK≤N<qK+1q_K \le N < q_{K+1}, we deduce that for all sufficiently large KK,

qK+1≤qKw(α)+ε/s≤Nw(α)+ε/s.q_{K+1} \le q_K^{w(\alpha)+\varepsilon/s} \le N^{w(\alpha)+\varepsilon/s}.

Substituting this bound into the estimate of (14), we obtain

∑1≤n≤N1∥nα∥s≪qK+1s≤Nw(α)s+ε.\sum_{1\le n\le N}\frac{1}{\lVert n\alpha\rVert^s}\ll q_{K+1}^s\le N^{w(\alpha)s+\varepsilon}.

Combining this bound with Abel’s summation formula, we deduce that

∑1≤n≤Nψ(n)∥nα∥s=∑n=1N−1(ψ(n)−ψ(n+1))∑m=1n1∥mα∥s+ψ(N)∑m=1N1∥mα∥s≪∑n=1N−1(ψ(n)−ψ(n+1))nw(α)s+ε+ψ(N)Nw(α)s+ε≍∑1≤n≤Nψ(n)nw(α)s+ε−1.\begin{aligned} \sum_{1\le n\le N}\frac{\psi(n)}{\lVert n\alpha\rVert^s} &=\sum_{n=1}^{N-1}(\psi(n)-\psi(n+1))\sum_{m=1}^{n}\frac{1}{\lVert m\alpha\rVert^s}+\psi(N)\sum_{m=1}^{N}\frac{1}{\lVert m\alpha\rVert^s}\\ &\ll\sum_{n=1}^{N-1}(\psi(n)-\psi(n+1))n^{w(\alpha)s+\varepsilon}+\psi(N)N^{w(\alpha)s+\varepsilon}\\ &\asymp\sum_{1\le n\le N}\psi(n)n^{w(\alpha)s+\varepsilon-1}. \end{aligned}

Therefore, if for some 0<ε0<10 < \varepsilon_0 < 1 we have

∑n=1∞ψ(n)nw(α)s+ε0−1<∞,\sum_{n=1}^{\infty}\psi(n)n^{w(\alpha)s+\varepsilon_0-1}<\infty,

then the first Borel–Cantelli lemma gives

L(Ws(ψ,α,γ))=0.\mathcal{L}(W_s(\psi,\alpha,\gamma))=0.

Divergence part (s>1s > 1). According to Theorem 1.3, it suffices to establish the following estimate:

∑1≤n≤Nϕ(n)nψ(n)∥nα∥s≫∑1≤n≤Nψ(n)ns−1log⁡log⁡n.\sum_{1\le n\le N}\frac{\phi(n)}{n}\frac{\psi(n)}{\lVert n\alpha\rVert^s}\gg\sum_{1\le n\le N}\frac{\psi(n)n^{s-1}}{\log\log n}.

We first establish two-sided bounds for the unweighted sum.

Lemma 5.1. Let Cα>0C_\alpha>0 be a constant depending only on α\alpha. We have

ϕ(qK)qKqK+1s≤∑1≤n≤Nϕ(n)n∥nα∥s≤CαqK+1s.\frac{\phi(q_K)}{q_K}q_{K+1}^s\le\sum_{1\le n\le N}\frac{\phi(n)}{n\lVert n\alpha\rVert^s}\le C_\alpha q_{K+1}^s.

Proof. The lower bound follows by choosing n=qKn=q_K. The upper bound is a direct consequence of (14).

Lemma 5.2. For N≥q3N \ge q_3, we have

∑1≤n≤Nϕ(n)n∥nα∥s≫Nslog⁡log⁡N.\sum_{1\le n\le N} \frac{\phi(n)}{n\lVert n\alpha\rVert^s} \gg\frac{N^s}{\log\log N}.

Proof. Combining the lower bound in Lemma 5.1 with the classical estimate [23 pp. 216–219]: ϕ(n)/n≫1/log⁡log⁡n\phi(n)/n \gg1/\log\log n, we deduce that

∑1≤n≤Nϕ(n)n∥nα∥s≥ϕ(qK)qKqK+1s≫Nslog⁡log⁡qK≫Nslog⁡log⁡N.\sum_{1\le n\le N} \frac{\phi(n)}{n\lVert n\alpha\rVert^s} \ge\frac{\phi(q_K)}{q_K}q_{K+1}^s \gg\frac{N^s}{\log\log q_K} \gg\frac{N^s}{\log\log N}.

Finally, applying Abel’s summation formula and Lemma 5.2, we obtain the desired weighted lower bound.

Lemma 5.3.

∑1≤n≤Nϕ(n)nψ(n)∥nα∥s≫∑1≤n≤Nψ(n)ns−1log⁡log⁡n.\sum_{1\le n\le N} \frac{\phi(n)}{n}\frac{\psi(n)}{\lVert n\alpha\rVert^s} \gg\sum_{1\le n\le N}\frac{\psi(n)n^{s-1}}{\log\log n}.

Proof. Since ψ\psi is nonincreasing, Abel’s summation formula yields

∑1≤n≤Nϕ(n)nψ(n)∥nα∥s=∑n=1N−1(ψ(n)−ψ(n+1))∑m=1nϕ(m)m∥mα∥s+ψ(N)∑m=1Nϕ(m)m∥mα∥s\sum_{1\le n\le N}\frac{\phi(n)}{n}\frac{\psi(n)}{\lVert n\alpha\rVert^s} =\sum_{n=1}^{N-1}\bigl(\psi(n)-\psi(n+1)\bigr)\sum_{m=1}^{n}\frac{\phi(m)}{m\lVert m\alpha\rVert^s} +\psi(N)\sum_{m=1}^{N}\frac{\phi(m)}{m\lVert m\alpha\rVert^s}
≫∑n=1N−1(ψ(n)−ψ(n+1))nslog⁡log⁡n+ψ(N)Nslog⁡log⁡N\gg\sum_{n=1}^{N-1}\bigl(\psi(n)-\psi(n+1)\bigr)\frac{n^s}{\log\log n} +\psi(N)\frac{N^s}{\log\log N}
≍∑1≤n≤Nψ(n)ns−1log⁡log⁡n.\asymp\sum_{1\le n\le N}\frac{\psi(n)n^{s-1}}{\log\log n}.

Combining Theorem 1.3 and Lemma 5.3, we complete the proof of the divergence part.

References

School of Mathematics and Statistics, Wuhan University of Technology

Email address: [email protected]

School of Mathematics and Statistics, Huazhong University of Science and Technology

Email address: [email protected]

School of Mathematics and Statistics, Wuhan University of Technology

Email address: [email protected]

References

  1. [1]F. Adiceam, E. Nesharim and F. Lunnon. On the \(t\)-adic Littlewood conjecture. Duke Math. J. 170 (2021), no. 10, 2371–2419.DOI
  2. [2]V. Beresnevich, A. Haynes and S. Velani. Multiplicative zero-one laws and metric number theory. Acta Arith. 160 (2013), no. 2, 101–114.arxiv.org/abs/1012.0675
  3. [3]V. Beresnevich, A. Haynes, and S. Velani. Sums of reciprocals of fractional parts and multiplicative Diophantine approximation. Mem. Amer. Math. Soc. 263 (2020), no. 1276.arxiv.org/abs/1511.06862
  4. [4]J. Cassels and H. Swinnerton-Dyer. On the product of three homogeneous linear forms and indefinite ternary quadratic forms. Philos. Trans. Roy. Soc. London Ser. A 248 (1955), 73–96.DOI
  5. [5]S. Chow. Box sets and multiplicative Diophantine approximation. Duke Math. J. 167 (2018), no. 9, 1623–1642.DOI
  6. [6]S. Chow, M. Hauke, A. Pollington and F. A. Ramírez. General Duffin–Schaeffer-type counterexamples in Diophantine approximation. Mathematika, to appear, preprint 2025, arXiv:2504.16565.arxiv.org/abs/2504.16565
  7. [7]S. Chow and N. Technau. Higher-rank Bohr sets and multiplicative Diophantine approximation. Compos. Math. 155 (2019), no. 11, 2214–2233.DOI
  8. [8]S. Chow and A. Zafeiropoulos. Fully-inhomogeneous multiplicative Diophantine approximation of badly approximable numbers. Int. Math. Res. Not. IMRN (2022), no. 15, 11637–11670.arxiv.org/abs/2103.03605
  9. [9]S. Chow and N. Technau. Littlewood and Duffin–Schaeffer-Type Problems in Diophantine Approximation. Mem. Amer. Math. Soc. 296 (2024), no. 1475.
  10. [10]S. Chow and L. Yang. Effective equidistribution for multiplicative Diophantine approximation on lines. Invent. Math. 235 (2024), no. 3, 973–1007.arxiv.org/abs/1902.06081
  11. [11]R. Duffin and A. Schaeffer. Khintchine’s problem in metric Diophantine approximation. Duke Math. J. 8 (1941), 243–255.DOI
  12. [12]M. Einsiedler, A. Katok and E. Lindenstrauss. Invariant measures and the set of exceptions to Littlewood’s conjecture. Ann. of Math. (2) 164 (2006), 513–560.DOI
  13. [13]P. Erdős, On the distribution of the convergents of almost all real numbers, J. Number Theory 2 (1970), 425–441.DOI
  14. [14]L. Fruihwirth and M. Hauke. The Duffin-Schaeffer Conjecture for multiplicative Diophantine approximation. (2024), arXiv: 2403.11257.
  15. [15]P. Gallagher, Approximation by reduced fractions. J. Math. Soc. Japan 13 (1961), 342–345.DOI
  16. [16]P. Gallagher. Metric simultaneous Diophantine approximation. J. London Math. Soc. 37 (1962), 387–390.DOI
  17. [17]A. Gorodnik and P. Vishe. Diophantine approximation for products of linear maps—logarithmic improvements. Trans. Amer. Math. Soc. 370 (2018), 487–507.
  18. [18]G. Harman. Metric Number Theory. London Mathematical Society Monographs. New Series, vol. 18, Clarendon Press, Oxford University Press, New York, 1998.
  19. [19]M. Hauke-Treuer, J. Maynard, and A. Pollington. Counterexamples to the inhomogeneous Duffin–Schaeffer conjecture. arXiv:2609.30921v1, 2026.arxiv.org/abs/2609.30921
  20. [20]Y. He and L. Liao. Counterexamples to the inhomogeneous Duffin–Schaeffer conjecture for a residual set of shifts. arXiv:2609.30870v1, 2026.arxiv.org/abs/2609.30870
  21. [21]A. Khintchine. Einige Sätze über Kettenbrüche, mit Anwendungen auf die Theorie der Diophantischen Approximationen. Math. Ann. 92 (1924), 115–125.
  22. [22]D. Koukoulopoulos and J. Maynard. On the Duffin–Schaeffer conjecture. Ann. of Math. (2) 192 (2020), 251–307.arxiv.org/abs/1907.04593
  23. [23]E. Landau. Handbuch der Lehre von der Verteilung der Primzahlen. Vol. 1, Teubner, Leipzig, 1909.
  24. [24]OpenAI. The Weak Inhomogeneous Duffin–Schaeffer Conjecture. OpenAI Math Release preprint, OAI: The-weak-inhomogeneous-Duffin-Schaeffer-conjecture-September-25-2026, 2026.DOI
  25. [25]A. Ostrowski. Bemerkungen zur Theorie der Diophantischen Approximationen. Abh. Math. Semin. Univ. Hambg., 1:77–98, 1922.
  26. [26]L. G. Peck. Simultaneous rational approximations to algebraic numbers. Bull. Amer. Math. Soc. 67 (1961), 197–201.DOI
  27. [27]A. D. Pollington and R. C. Vaughan, The k-dimensional Duffin and Schaeffer conjecture. Mathematika 37 (1990), 190–200.
  28. [28]A. D. Pollington and S. L. Velani. On a problem in simultaneous Diophantine approximation: Littlewood’s conjecture. Acta Math. 185 (2000), 287–306.DOI
  29. [29]F. A. Ramírez. Counterexamples, covering systems, and zero-one laws for inhomogeneous approximation. Int. J. Number Theory 13 (2017), 633–654.
  30. [30]A. M. Rockett and P. Szűsz. Continued fractions. World Scientific Publishing Co., Inc., River Edge, NJ, 1992.
  31. [31]U. Shapira. A solution to a problem of Cassels and Diophantine properties of cubic numbers. Ann. of Math. (2) 173 (2011), 543–557.arxiv.org/abs/0810.4289
  32. [32]P. Süsz. Über die metrische Theorie der Diophantischen Approximation. Acta Math. Hungar. 9 (1958), 177–193.
  33. [34]J. D. Vaaler, On the metric theory of Diophantine approximation, Pacific J. Math. 76 (1978), 527–539.DOI
  34. [35]H. Yu. A Fourier-analytic approach to inhomogeneous Diophantine approximation. Acta Arith. 190 (2019), 263–292.arxiv.org/abs/1704.04691
  35. [36]H. Yu. On the metric theory of inhomogeneous Diophantine approximation: an Erdős-Vaaler type result. J. Number Theory 224 (2021), 243–273.DOI
  36. [37]H. Yu. On the metric theory of multiplicative Diophantine approximation. J. Anal. Math. 149 (2023), no. 2, 763–800.arxiv.org/abs/2010.09004

Paper details

vlnc:W_bn6n9s7a
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents