General Duffin–Schaeffer-type counterexamples in diophantine approximation

Sam Chow and Manuel Hauke and Andrew Pollington and Felipe A. Ramírez · 2025

Read General Duffin–Schaeffer-type counterexamples in diophantine approximation by Sam Chow and Manuel Hauke and Andrew Pollington and Felipe A. Ramírez (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-16565.txt.

Introduction

Sam Chow, Manuel Hauke, Andrew Pollington, and Felipe Ramírez constructed counterexamples in Diophantine approximation, the study of how well numbers can be approximated by fractions. A classic result, Khintchine's theorem, needs the approximation function to shrink steadily; Duffin and Schaeffer showed it fails without that condition. The four authors generalize that counterexample: for any steadily shrinking function whose series diverges, they build examples by restricting which denominators are allowed. They also extend the work to the inhomogeneous setting, settling questions left open by Erdős, Vaaler, and Yu.

Catalog record

CREATORSam Chow and Manuel Hauke and Andrew Pollington and Felipe A. Ramírez
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-16565

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

General Duffin–Schaeffer-type counterexamples in diophantine approximation Sam Chow and Manuel Hauke and Andrew Pollington and Felipe A. Ramírez Abstract . Duffin and Schaeffer provided a famous counterexample to show that Khintchine’s theorem fails without monotonicity assumption. Given any monotonically decreasing approximation function with divergent series, we construct Duffin–Schaeffer-type counterexamples by restricting the denominator. We also extend these constructions to the inhomogeneous setting. Our results resolve some natural questions arising from the works of Erdős, Vaaler, and Yu. 2020 Mathematics Subject Classification 11J83 1. Introduction For \psi:\mathbb{N}\to[0,\infty) , the set of \psi -well approximable numbers in [0,1] is W(\psi)=\{{\alpha}\in[0,1]:\exists^{\infty}q\in\mathbb{N}\quad\|q{\alpha}\|<\psi(q)\}, where \|x\|=\min\{|x-a|:a\in\mathbb{Z}\} . Khintchine’s theorem [ 3 , Theorem 2.3] , the cornerstone of metric diophantine approximation, asserts that if \psi is non-increasing then (1.1) {\lambda}(W(\psi))=\begin{cases}0,&\text{if }\displaystyle\sum_{q=1}^{\infty}\psi(q)<\infty\\ \\ 1,&\text{if }\displaystyle\sum_{q=1}^{\infty}\psi(q)=\infty.\end{cases} Here and throughout, we write {\lambda} for Lebesgue measure on \mathbb{R} . Duffin and Schaeffer [ 5 ] famously constructed a counterexample to show that monotonicity cannot be relaxed in general. In the same article, they formulated the Duffin–Schaeffer conjecture on non-monotonic approximation with reduced fractions. Building upon the important steps made by Erdős [ 6 ] , Vaaler [ 13 ] and Pollington–Vaughan [ 11 ] , the conjecture was ingeniously resolved by Koukoulopoulos and Maynard [ 10 ] . Theorem 1.1 (Koukoulopoulos and Maynard) . Let \psi:\mathbb{N}\to[0,\infty) and W^{\prime}(\psi)=\{{\alpha}\in[0,1]:\exists^{\infty}(q,a)\in\mathbb{N}\times\mathbb{Z}\qquad|q{\alpha}-a|<\psi(q),\quad(a,q)=1\}. Then (1.2) {\lambda}(W^{\prime}(\psi))=\begin{cases}0,&\text{if }\displaystyle\sum_{q=1}^{\infty}\frac{\varphi(q)\psi(q)}{q}<\infty\\ \\ 1,&\text{if }\displaystyle\sum_{q=1}^{\infty}\frac{\varphi(q)\psi(q)}{q}=\infty.\end{cases} 1.1. Main results Our work is principally motivated by the aforementioned results of Erdős and Vaaler. The latter asserts that \psi(q)\ll 1/q\quad\forall q\>\>\Longrightarrow\>\>\text{\eqref{DSconclusion}}. This raises the question of whether or not \psi(q)\ll 1/q\quad\forall q\>\>\Longrightarrow\>\>\text{\eqref{KhintchineConclusion}}. We answer this question negatively, showing a fortiori that if f:\mathbb{N}\to[0,\infty) is any non-increasing function such that (1.3) \lim_{q\to\infty}f(q)=0,\qquad\sum_{q=1}^{\infty}f(q)=\infty, then \psi(q)\leqslant f(q)\quad\forall q\>\>\not\Longrightarrow\>\>\text{\eqref{KhintchineConclusion}}. Theorem 1.2 . Let f:\mathbb{N}\to[0,\infty) be a non-increasing function satisfying ( 1.3 ). Then there exists S\subseteq\mathbb{N} such that \sum_{q\in S}f(q)=\infty,\qquad{\lambda}(W(f{\boldsymbol{1}}_{S}))=0. Though it may at this stage be clear to a handful of experts, we wish to point out that our constructions can be modified to obtain Duffin–Schaeffer-type counterexamples that are monotonic on a support that has full density at infinitely many scales. Writing \bar{\delta}(S)=\limsup_{N\to\infty}\frac{|S\cap[1,N]|}{N} for the upper density of S\subseteq\mathbb{N} , we establish the following refinement of Theorem 1.2 . Theorem 1.3 . Let f:\mathbb{N}\to[0,\infty) be a non-increasing function satisfying ( 1.3 ). Then there exists \psi:\mathbb{N}\to[0,\infty) , non-increasing on its support, such that (1.4) \psi\leqslant f,\qquad\sum_{q\in\mathbb{N}}\psi(q)=\infty,\qquad{\lambda}(W(\psi))=0,\qquad\bar{\delta}(\supp\psi)=1. Remark 1.4 . Note that, in Theorem 1.3 , it cannot be guaranteed that \psi will take the form {\boldsymbol{1}}_{S}f for some S\subset\mathbb{N} with full upper density. For example, if f(n)\gg 1/\varphi(n) , then Theorem 1.1 precludes the existence of such a set S . We explain this in § 2.2 . Note also that “upper density” in Theorem 1.3 cannot be replaced by “lower density”, since the Duffin–Schaeffer theorem [ 5 , Theorem III] — not to be confused with the Duffin–Schaeffer conjecture — implies that if \liminf_{N\to\infty}\frac{|S\cap[1,N]|}{N}>0 then \sum_{q\in S}\psi(q)=\infty\implies{\lambda}(W(\psi))=0. 1.2. Inhomogeneous results There has been considerable interest around inhomogeneous, non-monotonic approximation in recent years [ 1 , 2 , 4 , 7 , 8 , 9 , 12 , 14 ] . For \psi:\mathbb{N}\to[0,\infty) and {\gamma}\in\mathbb{R} , define W(\psi,{\gamma})=\{{\alpha}\in[0,1]:\exists^{\infty}q\in\mathbb{N}\quad\|q{\alpha}-{\gamma}\|<\psi(q)\}. The second motivation for our work comes from the aforementioned article of Yu [ 14 ] . A Liouville number is {\gamma}\in\mathbb{R}\setminus\mathbb{Q} such that if w\in\mathbb{R} then \|q{\gamma}\|<q^{-w} holds for infinitely many q\in\mathbb{N} . These form a set \mathcal{L} . Yu introduced the set \mathcal{T} of tamely Liouville numbers , which satisfies \mathbb{R}\setminus(\mathbb{Q}\cup\mathcal{L})\subsetneq\mathcal{T}\subsetneq\mathbb{R}\setminus\mathbb{Q}, and showed in [ 14 , Theorem 1.3] that (1.5) \psi(q)\ll\frac{1}{q(\log\log q)^{2}}\>\>\Longrightarrow\>\>\left(\sum_{q=1}^{\infty}\psi(q)=\infty\>\>\Longrightarrow\>\>{\lambda}(W(\psi,{\gamma}))=1\>\>\forall{\gamma}\in\mathcal{T}\right). This raises the question of whether the tame Liouville assumption can be replaced by the weaker assumption of irrationality. We answer this question negatively, showing a fortiori that if f:\mathbb{N}\to[0,\infty) is a non-increasing function satisfying ( 1.3 ), then \psi(q)\leqslant f(q)\quad\forall q\>\>\not\Longrightarrow\>\>\left(\sum_{q=1}^{\infty}\psi(q)=\infty\>\>\Longrightarrow\>\>{\lambda}(W(\psi,{\gamma}))=1\>\>\forall{\gamma}\in\mathbb{R}\setminus\mathbb{Q}\right). Theorem 1.5 . Let f:\mathbb{N}\to[0,\infty) be a non-increasing function satisfying ( 1.3 ). Then there exists a set S\subseteq\mathbb{N} and an uncountable set \Gamma\subseteq\mathbb{R} containing \mathbb{Q} such that if {\gamma}\in{\Gamma} then (1.6) \sum_{q\in S}f(q)=\infty,\qquad{\lambda}(W({\boldsymbol{1}}_{S}f,{\gamma}))=0. Far from being tamely Liouville in the sense of Yu, the set \Gamma in Theorem 1.5 consists of real numbers that are “wildly” Liouville in a sense that depends on the function f . In fact, to each f we are able to associate a notion of wildly Liouville that is sufficient for the existence of a set S=S_{\gamma} such that ( 1.6 ) holds (see Theorem 3.4 ). As a consequence, we have the following theorem for functions of the form (1.7) f_{k}(q)=\frac{1}{q(\log q)(\log\log q)\cdots\log^{(k)}q}, where \log^{(k)} denotes the k^{\mathrm{th}} iterate of x\mapsto\max\{1,\log x\} . Theorem 1.6 . Let k\geqslant 0 be an integer, and let \gamma\in\mathbb{R} . If \liminf_{q\to\infty}\lVert q\gamma\rVert\exp^{(k+3)}(q^{7000})<\infty, where \exp^{(k)} denotes the k^{\mathrm{th}} iterate of \exp , then there exists S\subseteq\mathbb{N} such that \sum_{q\in S}f_{k}(q)=\infty,\qquad\lambda(W({\boldsymbol{1}}_{S}f_{k},\gamma))=0. In particular, Yu’s result cannot be extended to real numbers \gamma for which \liminf_{q\to\infty}\lVert q\gamma\rVert\exp^{(4)}(q^{7000})<\infty. We remark that there is an enormous gap between this wild Liouville notion and Yu’s tame Liouville notion. While one can most probably sharpen the estimates applied in this article, it does not seem that the gap can be closed simply by optimising the proof of Theorem 1.6 . Our constructions have the novel feature of monotonicity on the support. In previous Duffin–Schaeffer-type constructions, the approximating function \psi has the weaker property that q\mapsto\psi(q)/q is monotonic. We conclude our introductory discussion with an open problem, to prove or disprove the following quantitative refinement of ( 1.5 ), that \psi(q)\ll 1/q\>\>\Longrightarrow\>\>\left(\sum_{q=1}^{\infty}\psi(q)=\infty\>\>\Longrightarrow\>\>{\lambda}(W(\psi,{\gamma}))=1\>\>\forall{\gamma}\in\mathcal{T}\right). 1.3. Methods Let us describe the proof of Theorem 1.2 in coarse terms. First of all, we require that {\lambda}\left(\bigcup_{q\in S^{\prime}}A_{q}\right)=o\left(\sum_{q\in S^{\prime}}{\lambda}(A_{q})\right), where S is a disjoint union of sets of the form S^{\prime} . Secondly, we require the series \sum_{q\in S}f(q) to diverge very slowly. To these ends, we take S^{\prime}=X\cdot Y:=\{xy:x\in X,\>y\in Y\}. The set Y is highly multiplicatively structured. We use some probability theory to show that a generic pair of elements of Y has a large common divisor. This enables us to deal with the first requirement. The set X is then chosen, consisting of much larger elements, so as to carefully secure the second requirement. For the inhomogeneous theory, say Theorem 1.5 , our shift {\gamma} is very close to some rational number with denominator b , in fact a sequence of these. The key extra ingredient, Lemma 3.2 , asserts that if \|b{\gamma}\|\leqslant b{\varepsilon} then A_{q}^{\gamma}({\varepsilon})\subseteq A_{bq}(2b{\varepsilon}) . This essentially reduces the problem to the homogeneous setting. Notation For {\varepsilon}>0,\gamma\in\mathbb{R} , and q\in\mathbb{N} , denote A_{q}^{\gamma}({\varepsilon})=\left\{\alpha\in[0,1]:\lVert q\alpha-\gamma\rVert<{\varepsilon}\right\}. For f:\mathbb{N}\to\mathbb{R}_{+} , we denote A_{q}^{\gamma}(f)=A_{q}^{\gamma}(f(q)) , and when it is clear from the context which function we are working with, we write A_{q}^{\gamma} for this set. Thus, W(f,\gamma)=\limsup_{q\to\infty}A_{q}^{\gamma}(f). When there is no superscript \gamma , the notation should be interpreted with \gamma=0 . We write \mathbb{P}=\{p_{1}<p_{2}<\ldots\} for the set of all primes. For m\in\mathbb{N} , we abbreviate \mathbb{Z}_{m}=\mathbb{Z}/m\mathbb{Z} . Funding AP was supported by the National Science Foundation. Rights For the purpose of open access, SC has applied a Creative Commons Attribution (CC-BY) licence to any Author Accepted Manuscript version arising from this submission. 2. Homogeneous counterexamples In this section, we prove Theorems 1.2 and 1.3 . 2.1. Proof of Theorem 1.2 2.1.1. Construction of counterexample blocks The

Read the complete text