betaFeedback

WorksMathematics

On metric multiplicative Diophantine approximation

Abstract

Gallagher's theorem (1962) gives a Khintchine-type dichotomy for the Lebesgue measure of the set of multiplicatively ψψ-approximable points in Rn\mathbb{R}^n, n≥2n \geq 2, for monotonic functions ψψ. We establish this dichotomy for arbitrary functions ψψ, proving a conjecture of Beresnevich, Haynes, and Velani (2013).

Introduction

Multiplicative Diophantine approximation concerns the set Gn(ψ)G_n(\psi) of α=(α1,…,αn)∈Rn\boldsymbol{\alpha}=(\alpha_1,\ldots,\alpha_n)\in\mathbb{R}^n such that there are infinitely many solutions to

∣α1−a1q∣⋅∣α2−a2q∣⋯∣αn−anq∣≤ψ(q)qn,(a1,…,an,q)∈Zn×N,(1)\left|\alpha_1-\frac{a_1}{q}\right|\cdot\left|\alpha_2-\frac{a_2}{q}\right|\cdots\left|\alpha_n-\frac{a_n}{q}\right|\le\frac{\psi(q)}{q^n},\qquad(a_1,\ldots,a_n,q)\in\mathbb{Z}^n\times\mathbb{N}, \tag*{(1)}

where ψ:N→[0,∞)\psi:\mathbb{N}\to[0,\infty) is a given function. The foremost open problem in this area is Littlewood’s conjecture, stating that

L:=⋂ε>0G2(ε/q)=R2.\mathcal{L}:=\bigcap_{\varepsilon>0}G_2(\varepsilon/q)=\mathbb{R}^2.

It is a simple matter to show that L\mathcal{L} has full Lebesgue measure, and much of the work toward Littlewood’s conjecture has resulted in further descriptions of the smallness of the exceptional set R2∖L\mathbb{R}^2\setminus\mathcal{L}. Most notably, Einsiedler, Katok, and Lindenstrauss [14] [2006] showed that R2∖L\mathbb{R}^2\setminus\mathcal{L} is a countable union of sets of zero box-counting dimension, improving vastly on the baseline knowledge that it is a null set.

In a related direction of research, one aims to determine the extent to which the claim in Littlewood’s conjecture can be strengthened, after allowing an exceptional set of Lebesgue measure zero. Gallagher proved the following theorem.

Gallagher’s Theorem ([16] [1962]). Let n≥2n\ge2 and let ψ:N→[0,∞)\psi:\mathbb{N}\to[0,\infty) be monotonically decreasing. Then

λn(Gn(ψ))={0if ∑ψ(q)(log⁡1ψ(q))n−1<∞,1if ∑ψ(q)(log⁡1ψ(q))n−1=∞,\lambda_n(G_n(\psi))= \begin{cases} 0 & \text{if }\sum\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1}<\infty,\\ 1 & \text{if }\sum\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1}=\infty, \end{cases}

where log⁡x:=max⁡{ln⁡x,1}\log x:=\max\{\ln x,1\}, and λn\lambda_n denotes the nn-dimensional Lebesgue measure on [0,1]n[0,1]^n.

Remark. In fact, the series appearing in [16] is ∑ψ(q)(log⁡q)n−1\sum\psi(q)(\log q)^{n-1}, but the above statement is equivalent, in the presence of the monotonicity assumption, see e.g. [5 Remark 4]. We remark that the above theorem is also true for n=1n = 1. In this case it is exactly Khintchine’s theorem [20] [1924], and hence, Gallagher’s theorem is the extension of Khintchine’s theorem to the multiplicative setting.

Gallagher’s theorem implies that Littlewood’s conjecture can be strengthened by a factor of (log⁡q)2(\log q)^2 for almost all points, but for almost no points by a factor of (log⁡q)2+δ(\log q)^{2+\delta}, that is,

λ2(⋂ε>0G2(εq(log⁡q)2))=1,whileλ2(⋂ε>0G2(εq(log⁡q)2+δ))=0,\lambda_2\left(\bigcap_{\varepsilon>0}G_2\left(\frac{\varepsilon}{q(\log q)^2}\right)\right)=1,\qquad\text{while}\qquad\lambda_2\left(\bigcap_{\varepsilon>0}G_2\left(\frac{\varepsilon}{q(\log q)^{2+\delta}}\right)\right)=0,

for every δ>0\delta> 0. Similarly, it implies that the nn-dimensional analogue of Littlewood’s conjecture, which is open for every n≥2n \ge2, holds almost everywhere in Rn\mathbb{R}^n with an extra factor of (log⁡q)n(\log q)^n, but almost nowhere with an extra factor of (log⁡q)n+δ(\log q)^{n+\delta}. Of course, Gallagher’s theorem says much more, giving a complete metric dichotomy for monotonic rates, in the style of Khintchine’s theorem [20] [1924]. Famously, the monotonicity assumption cannot be removed from Khintchine’s theorem [12] [1941]. In contrast, it has been conjectured by Beresnevich, Haynes, and Velani that Gallagher’s theorem holds for arbitrary functions [5] [2013, Conjecture 1]. In this paper, we prove that conjecture.

Theorem 1. Let n≥2n \ge2 and let ψ:N→[0,∞)\psi:\mathbb{N}\to[0,\infty) be arbitrary. Then

λn(Gn(ψ))={0if ∑ψ(q)(log⁡1ψ(q))n−1<∞,1if ∑ψ(q)(log⁡1ψ(q))n−1=∞.\lambda_n(G_n(\psi))= \begin{cases} 0 & \text{if }\sum\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1}<\infty,\\ 1 & \text{if }\sum\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1}=\infty. \end{cases}

Similar questions have been considered for other forms of approximation, such as simultaneous, dual, systems of linear forms, as well as inhomogeneous approximation and approximation with coprimality restrictions [2, 3, 4, 6, 9, 12, 15, 17, 18, 19, 22, 24, 25, 26]. The case of simultaneous approximation was handled by Gallagher in a follow-up paper [17] to the above cited [16]. He showed that the higher-dimensional Khintchine theorem [21] [1926] can be stated with arbitrary functions.

The main difficulty in the multiplicative setup comes from the unwieldy geometry of the sets involved. Specifically, the multiplicative problems present nonconvex sets whose intersections one must understand. (See Figure 1 to Figure 4.) By contrast, the other forms of approximation mentioned above present metric balls and thickened subspaces, whose intersections are more tractable. Historically, the literature has exhibited a trade-off between convexity and monotonicity. Beresnevich, Haynes, and Velani give an enlightening discussion of this trade-off and various instances of it:

Figure 1. On the left: An illustration of a set AqA_q with one of the translated copies of H‾q\overline{H}_q highlighted in purple. In this example, q=4q=4, and the highlighted set is H‾4+(14,12)\overline{H}_4+\left(\frac{1}{4},\frac{1}{2}\right). On the right: An illustration of the sets Aq′′A''_q appearing in Gallagher [16], coming from the condition that each coordinate of a\mathbf{a} is coprime to qq. Gallagher proved QIA for these sets, under a monotonicity assumption.

Figure 4. The contribution of the error terms. The top row shows the overlaps between an untruncated set H‾q\overline{H}_q and coordinatewise shifts of H‾r\overline{H}_r when there is a coincidence in the other coordinate. The middle row shows the same for the truncated sets HqH_q and HrH_r. Note the smaller purple regions of intersection for the truncated sets. The bottom row shows a comparison of the integrals of the measures of those overlaps.

“[T]he current body of metrical results for lim sup sets requires that either the approximating domains are convex or that the approximating function is monotonic.”

[5 Section 1.1]

Khintchine’s theorem for simultaneous approximation in higher dimensions [21] [1926] used both convexity and monotonicity. Later, Gallagher [17] showed that monotonicity could be removed from that theorem; his arguments relied heavily on the convexity of the sets that the simultaneous approximation setup provides. By contrast, the proof of the multiplicative, and hence non-convex, Gallagher theorem [16] uses monotonicity in a crucial way.

Beresnevich, Haynes, and Velani also point out that what is commonly called Gallagher’s theorem is actually a consequence of a more general statement established in [16]:

“In fact, Gallagher considers lim sup sets arising from more general domains but monotonicity plays a crucial role in his approach and seems to be unremovable.”

[5 Section 1.1]

Gallagher proved in [16] that instead of assuming convexity, a Khintchine-type behaviour under monotonicity holds for qα(mod1)∈Sqq\boldsymbol{\alpha}\pmod{1}\in S_q for infinitely many qq, as long as the sets (Sq)q∈N(S_q)_{q\in\mathbb{N}} satisfy

x=(x1,…,xn)∈Sqand∣xi′∣≤∣xi∣ (i=1,…,n)⟹x′=(x1′,…,xn′)∈Sq.(2)\mathbf{x}=(x_1,\ldots,x_n)\in S_q \quad\text{and}\quad|x_i'|\leq|x_i|\ (i=1,\ldots,n) \quad\Longrightarrow\quad\mathbf{x}'=(x_1',\ldots,x_n')\in S_q. \tag*{(2)}

The multiplicative approximation sets qHq‾q\overline{H_q} (see Figure 1) satisfy (2), and thus Gallagher’s multiplicative theorem is indeed an instance of this more general statement. A straightforward construction confirms the remark quoted above by showing that if one only assumes (2), then monotonicity is unremovable.

Theorem 2. For every n≥1n\geq1, there exists a sequence of sets SqS_q satisfying (2) such that ∑λ(Sq)\sum\lambda(S_q) diverges and such that the set WW of points α∈[0,1]n\boldsymbol{\alpha}\in[0,1]^n for which there are infinitely many solutions to

qα(mod1)∈Sq,q∈Nq\boldsymbol{\alpha}\pmod{1}\in S_q,\qquad q\in\mathbb{N}

has Lebesgue measure 0.

In fact, the sets we construct for the proof of Theorem 2 are even convex. From a geometric perspective, removing monotonicity might require the consideration of further properties of the sets SqS_q, perhaps a bound on the ratio between the diameter (of reasonably truncated subsets of SqS_q) and the volume: Theorem 2 is based on very thin rectangles that reduce the problem to the one-dimensional setting, where monotonicity is known to be necessary. As we will see in the proof of Theorem 1, the hyperbolic shapes for multiplicative approximation are just “higher-dimensional enough” to allow the removal of monotonicity. The upshot is that the multiplicative problem (1) strikes a rare balance that allows us to dispense with both convexity and monotonicity for Theorem 1.

Organization of the paper

The remainder of the paper is organized as follows. In Section 2 we present the proof strategy in the context of previous methods, highlighting how our proof differs from previous approaches. In Section 3 we collect notation, and in Section 4 we state the established tools from the literature that are used in the proof of Theorem 1. In Section 5 we make elementary calculations that are directly used in the subsequent technical sections. In Section 6 we provide the technical heart of the paper by calculating upper bounds on second moments, and in Section 7 we conclude the proof of Theorem 1, as well as the (particularly short) proof of Theorem 2.

Acknowledgments

MHT was funded by the Austrian Science Fund (FWF) project 10.55776/ESP5134624. Parts of this project were carried out while MHT visited FAR. MHT would like to thank Wesleyan University, as well as the whole Ramírez family (Yeni, Luna, Joaquín and Felipe) for their hospitality. More precisely, he thanks Yeni for the excellent Sancocho de Cola, Luna for teaching him how to do Origami, and Joaquín and Felipe for showing him where to find Asteroids, Waldo, and other things that really matter in life.

The proof strategy

In this section, we outline the strategy of the proof of Theorem 1, focusing on the case where n=2n=2. Note that G2(ψ)=lim sup⁡q→∞Aq\mathcal{G}_2(\psi)=\limsup_{q\to\infty} A_q where

Aq=⋃a∈Zq2(H‾q+aq).A_q=\bigcup_{\mathbf{a}\in\mathbb{Z}_q^2}\left(\overline{H}_q+\frac{\mathbf{a}}{q}\right).

with

H‾q+aq={(α,β)∈[0,1]2:∣α−a1q∣∣β−a2q∣≤ψ(q)q2,∣α−a1q∣≤12q,∣β−a2q∣≤12q}.\overline{H}_q+\frac{\mathbf{a}}{q}=\left\{(\alpha,\beta)\in[0,1]^2:\left|\alpha-\frac{a_1}{q}\right|\left|\beta-\frac{a_2}{q}\right|\leq\frac{\psi(q)}{q^2},\left|\alpha-\frac{a_1}{q}\right|\leq\frac{1}{2q},\left|\beta-\frac{a_2}{q}\right|\leq\frac{1}{2q}\right\}.

See Figure 1 for an illustration of AqA_q and H‾q\overline{H}_q.

A straightforward calculation (Lemma 5.3) shows that λ(Aq)≍ψ(q)log⁡1ψ(q)\lambda(A_q)\asymp\psi(q)\log\frac{1}{\psi(q)} and, therefore, the convergence part of Theorem 1 follows by an application of the first Borel–Cantelli lemma. The divergence part of Theorem 1 would follow from the second Borel–Cantelli lemma, if the sets (Aq)q∈N(A_q)_{q\in\mathbb{N}} were stochastically independent. However, they are far from independent. Instead, we aim to show that they contain subsets Aq∗⊆AqA_q^*\subseteq A_q such that ∑λ(Aq∗)\sum\lambda(A_q^*) diverges and such that the (Aq∗)q(A_q^*)_q exhibit a weaker form of independence, called quasi-independence on average (QIA). Specifically, it would suffice to prove that there exists an absolute constant C>0C>0, and subsets Aq∗⊆AqA_q^*\subseteq A_q such that for infinitely many disjoint ranges [X,Y][X,Y] with ∑X≤q≤Yλ(Aq∗)∈[1,2]\sum_{X\leq q\leq Y}\lambda(A_q^*)\in[1,2], we have

∑q,r∈[X,Y]λ(Aq∗∩Ar∗)≤C∑q,r∈[X,Y]λ(Aq∗)λ(Ar∗)≤4C.(3)\sum_{q,r\in[X,Y]}\lambda(A_q^*\cap A_r^*)\leq C\sum_{q,r\in[X,Y]}\lambda(A_q^*)\lambda(A_r^*)\leq4C. \tag*{(3)}

for then the Chung–Erdős Lemma (see § 5) would imply that in these ranges we had

λ(⋃X≤q≤YAq)≥λ(⋃X≤q≤YAq∗)≥(4C)−1,\lambda\left(\bigcup_{X\leq q\leq Y}A_q\right)\geq\lambda\left(\bigcup_{X\leq q\leq Y}A_q^*\right)\geq(4C)^{-1},

and λ(G2(ψ))≥(4C)−1\lambda(\mathcal{G}_2(\psi))\geq(4C)^{-1} would follow. A zero-one law (Theorem BHV, see § 5 and [5 Theorem 1]) would then give Theorem 1.

The most straightforward approach when aiming for (3) would be to choose Aq∗=AqA_q^*=A_q. Computing λ(Aq∩Ar)\lambda(A_q\cap A_r) amounts to summing the volume of (H‾q+aq)∩(H‾r+br)\left(\overline{H}_q+\frac{\mathbf{a}}{q}\right)\cap\left(\overline{H}_r+\frac{\mathbf{b}}{r}\right) over all values of (a,b)∈Zq2×Zr2(\mathbf{a},\mathbf{b})\in\mathbb{Z}_q^2\times\mathbb{Z}_r^2. Clearly, the volume of (H‾q+aq)∩(H‾r+br)\left(\overline{H}_q+\frac{\mathbf{a}}{q}\right)\cap\left(\overline{H}_r+\frac{\mathbf{b}}{r}\right) is determined by aq−br\frac{\mathbf{a}}{q}-\frac{\mathbf{b}}{r}. Hence, writing

w^(x)=w^q,r(x):=λ((H‾q+x)∩H‾r),\widehat{w}(\mathbf{x})=\widehat{w}_{q,r}(\mathbf{x}):=\lambda\left((\overline{H}_q+\mathbf{x})\cap\overline{H}_r\right),

one has

λ(Aq∩Ar)=∑(a,b)∈Zq2×Zr2w^(aq−br)=gcd⁡(q,r)2∑c∈Zlcm⁡(q,r)2w^(clcm⁡(q,r)),(4)\lambda(A_q\cap A_r)=\sum_{(\mathbf{a},\mathbf{b})\in\mathbb{Z}_q^2\times\mathbb{Z}_r^2}\widehat{w}\left(\frac{\mathbf{a}}{q}-\frac{\mathbf{b}}{r}\right)=\gcd(q,r)^2\sum_{\mathbf{c}\in\mathbb{Z}_{\operatorname{lcm}(q,r)}^2}\widehat{w}\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right), \tag*{(4)}

by an application of the Chinese Remainder Theorem. Since w^(x)\widehat{w}(\mathbf{x}) is monotonically decreasing in ∣xi∣|x_i| for i=1,2i=1,2 (here, the property (2) is decisive), we may bound the sum with an integral,

∑c∈Z2w^(clcm⁡(q,r))≤lcm⁡(q,r)2∫R2w^(x) dx+∑c∈Z2c1=0 or c2=0w^(clcm⁡(q,r)).(5)\sum_{\mathbf{c}\in\mathbb{Z}^2}\widehat{w}\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right)\leq\operatorname{lcm}(q,r)^2\int_{\mathbb{R}^2}\widehat{w}(\mathbf{x})\,\mathrm{d}\mathbf{x}+\sum_{\substack{\mathbf{c}\in\mathbb{Z}^2\\c_1=0\text{ or }c_2=0}}\widehat{w}\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right). \tag*{(5)}

Straightforward calculations reveal that

∫R2w^(x) dx≍1q21r2λ(Aq)λ(Ar),(6)\int_{\mathbb{R}^2}\widehat{w}(\mathbf{x})\,\mathrm{d}\mathbf{x}\asymp\frac{1}{q^2}\frac{1}{r^2}\lambda(A_q)\lambda(A_r), \tag*{(6)}

which, combined with (4), (5), and (6), shows

λ(Aq∩Ar)≪λ(Aq)λ(Ar)+gcd⁡(q,r)2∑c∈Z2c1=0 or c2=0w^(clcm⁡(q,r)).(7)\lambda(A_q\cap A_r)\ll\lambda(A_q)\lambda(A_r)+\gcd(q,r)^2\sum_{\substack{\mathbf{c}\in\mathbb{Z}^2\\c_1=0\text{ or }c_2=0}}\widehat{w}\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right). \tag*{(7)}

Evidently, the only trouble can come from pairs (a,b)(\mathbf{a},\mathbf{b}) where

a1q=b1rora2q=b2r.(8)\frac{a_1}{q}=\frac{b_1}{r}\quad\text{or}\quad\frac{a_2}{q}=\frac{b_2}{r}. \tag*{(8)}

However, calculations show that the contribution from these points can indeed destroy the desired QIA property (3) for the choice Aq∗=AqA_q^*=A_q.

Gallagher’s idea [16] was to show (3) for Aq′′⊆AqA_q'' \subseteq A_q instead, defined by

Aq′′:=⋃a=(a1,a2)∈Zq2,gcd⁡(a1,q)=1,gcd⁡(a2,q)=1(Hq‾+aq).A_q'' := \bigcup_{\substack{\mathbf{a}=(a_1,a_2)\in\mathbb{Z}_q^2,\\ \gcd(a_1,q)=1,\\ \gcd(a_2,q)=1}} \left(\overline{H_q}+\frac{\mathbf{a}}{q}\right).

(See Figure 1.) For q≠rq \ne r, the coprimality conditions rule out the existence of contributions from pairs (a,b)(\mathbf{a},\mathbf{b}) satisfying (8). However, one now has λ(Aq′′)≍(φ(q)q)2λ(Aq)\lambda(A_q'') \asymp\left(\frac{\varphi(q)}{q}\right)^2\lambda(A_q) (where φ\varphi denotes the Euler totient function), and if ψ\psi is supported only on numbers qq with many small prime factors, this can cause the series ∑λ(Aq′′)\sum\lambda(A_q'') to converge. It is for this reason that Gallagher’s theorem has a monotonicity assumption on ψ\psi: the weights φ(q)/q\varphi(q)/q are bounded away from 0 on average, so employing partial summation shows that for monotonic ψ\psi we have

∑q≤Qλ(Aq′′)≍∑q≤Qλ(Aq),(9)\sum_{q\le Q}\lambda(A_q'')\asymp\sum_{q\le Q}\lambda(A_q), \tag*{(9)}

and the series ∑λ(Aq′′)\sum\lambda(A_q'') diverges. Now, having removed the contributions from (8) the bound (7) becomes

λ(Aq′′∩Ar′′)≪λ(Aq)λ(Ar),\lambda(A_q''\cap A_r'')\ll\lambda(A_q)\lambda(A_r),

and (9) is used again to show that one has

λ(Aq)λ(Ar)≪λ(Aq′′)λ(Ar′′),\lambda(A_q)\lambda(A_r)\ll\lambda(A_q'')\lambda(A_r''),

on average over long ranges of qq, rr. This implies (3) which, as discussed, is sufficient for full measure of the limsup set. This proves Gallagher’s theorem, but having crucially used the monotonicity assumption.

In fact, it was shown by Frühwirth and the first author [15] that a coordinate-wise coprimality condition in (1) also leads to Khintchine-type behaviour, without assuming monotonicity: denoting the underlying sets by Aq′A_q', it was shown that ∑q∈Nλ(Aq′)=∞\sum_{q\in\mathbb{N}}\lambda(A_q')=\infty implies λ(lim sup⁡q→∞Aq′)=1\lambda(\limsup_{q\to\infty}A_q')=1 for arbitrary ψ\psi. This resolved a conjecture of Beresnevich, Haynes, and Velani [5 Conjecture 2], and incidentally clarified an overlooked detail in the problem, namely, that Aq′A_q' has “arms” that extend past the arms of Aq′′A_q'' (see Figure 2), and result in a different divergence condition from the conjectured one appearing in [5]. As in the previously described argument for Gallagher’s theorem, it was necessary to restrict to subsets of the sets arising naturally for the respective approximation sets. Instead of working directly with Aq′A_q', the arguments in [15] used fiberwise-smooth bump functions 0≤fq≤1Aq′0\le f_q\le\mathbf{1}_{A_q'} that could heuristically be thought of as producing truncated sets Aq∗⊆Aq′A_q^*\subseteq A_q' similar to the ones Figure 2 depicts.1

The sets $A'_q$ and $A^*_{q,*}$ appearing in Frühwirth–Hauke

Figure 2. The sets Aq′A'_q and Aq,∗∗A^*_{q,*} appearing in Frühwirth–Hauke [15]. Note the long arms on Aq′A'_q in comparison to the short arms of Aq′′A''_q in Figure 1, and truncations in Aq,∗∗A^*_{q,*}.

Returning to the original question, which assumes only the divergence of ∑qλ(Aq)\sum_q\lambda(A_q), it is clear that restricting to Aq′′A_q'' will not help, because without monotonicity, there is no guarantee that the series ∑λ(Aq′′)\sum\lambda(A_q'') diverges. We will see, however, that instead of pairwise coprimality gcd⁡(a1,q)=gcd⁡(a2,q)=1\gcd(a_1,q)=\gcd(a_2,q)=1, we can use jointly coprime vectors, i.e.,

A~q:=⋃a=(a1,a2)∈Zq2,gcd⁡(a1,a2,q)=1(Hq‾+aq).\widetilde{A}_q := \bigcup_{\substack{\mathbf{a}=(a_1,a_2)\in\mathbb{Z}_q^2,\\ \gcd(a_1,a_2,q)=1}} \left(\overline{H_q}+\frac{\mathbf{a}}{q}\right).

(See Figure 3.) Unlike coprimality in dimension one or the pairwise coprimality assumption employed by Gallagher, the joint coprimality condition gcd⁡(a1,a2,q)=1\gcd(a_1,a_2,q)=1 holds for a positive proportion of a∈Zq2\mathbf{a}\in\mathbb{Z}_q^2 for every qq. Therefore, the divergence of ∑qλ(Aq)\sum_q\lambda(A_q) implies the divergence of ∑qλ(A~q)\sum_q\lambda(\widetilde{A}_q) without having to assume monotonicity. For q≠rq\ne r, the condition gcd⁡(a1,a2,q)=1\gcd(a_1,a_2,q)=1 rules out coincidences where both equalities in (8) hold, that is,

Sets $\widetilde{A}_q$ and their truncations

Figure 3. On the left: The sets A~q\widetilde{A}_q defined using gcd⁡(a1,a2,q)=1\gcd(a_1,a_2,q)=1. On the right: Their truncations Aq∗A_q^*.

a1q=b1rANDa2q=b2r.\frac{a_1}{q}=\frac{b_1}{r}\quad\text{AND}\quad\frac{a_2}{q}=\frac{b_2}{r}.

but cannot avoid contributions from coordinate-wise coincidences

a1q=b1r and a2q≠b2rORa1q≠b1r and a2q=b2r.\frac{a_1}{q}=\frac{b_1}{r}\ \text{and}\ \frac{a_2}{q}\ne\frac{b_2}{r}\quad\text{OR}\quad\frac{a_1}{q}\ne\frac{b_1}{r}\ \text{and}\ \frac{a_2}{q}=\frac{b_2}{r}.

Avoiding (i) (but not (ii)), allows us to refine (5) by bounding

∑c∈Z2∖{0}c1=0 or c2=0w^(clcm⁡(q,r))≪lcm⁡(q,r)∫Rw^(x,0) dx.(10)\sum_{\substack{\mathbf{c}\in\mathbb{Z}^2\setminus\{\mathbf{0}\}\\ c_1=0\ \text{or}\ c_2=0}} \widehat{w}\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right) \ll\operatorname{lcm}(q,r)\int_{\mathbb{R}}\widehat{w}(x,0)\,\mathrm{d}x. \tag*{(10)}

again using monotonicity and symmetry of w^\widehat{w}. The arguments that led to (7) now give

λ(A~q∩A~r)≪λ(Ar)λ(Aq)+gcd⁡(q,r)2lcm⁡(q,r)∫Rw^(x,0) dx≪λ(A~q)λ(A~r)+qrgcd⁡(q,r)∫Rw^(x,0) dx⏟:=E^q,r.\begin{aligned} \lambda(\widetilde{A}_q\cap\widetilde{A}_r)\ll\lambda(A_r)\lambda(A_q)+\gcd(q,r)^2\operatorname{lcm}(q,r)\int_{\mathbb{R}}\widehat{w}(x,0)\,\mathrm{d}x \\ &\ll\lambda(\widetilde{A}_q)\lambda(\widetilde{A}_r)+qr\gcd(q,r)\underbrace{\int_{\mathbb{R}}\widehat{w}(x,0)\,\mathrm{d}x}_{:=\widehat{E}_{q,r}}. \end{aligned}

Evidently, we must study the error terms E^q,r\widehat{E}_{q,r}. If one could show that

∑q,r∈[X,Y]E^q,r≪(∑q∈[X,Y]λ(Aq))2≍1,(11)\sum_{q,r\in[X,Y]}\widehat{E}_{q,r}\ll\left(\sum_{q\in[X,Y]}\lambda(A_q)\right)^2\asymp1, \tag*{(11)}

then the sufficient QIA (3) for (A~q)q(\widetilde{A}_q)_q would follow.

We note that Gallagher himself used the same joint primitivity condition to remove monotonicity from Khintchine’s Theorem [21] in the question of simultaneous approximation [17]. There, the underlying approximation sets are squares, hence the function ww behaves much more regularly. In fact, a direct application of the above method with w^(x,0)≤∥w^∥∞\widehat{w}(x,0)\leq\lVert\widehat{w}\rVert_\infty allows us to establish (11) for the analogous sets. This highlights the convenience of working with sets having nice geometric properties. Khintchine’s theorem is about balls in a norm-induced metric; they are convex and have bounded eccentricity.

For the actual sets (A~q)q(\widetilde{A}_q)_q and the underlying hyperbolic regions H‾q\overline{H}_q, this crude approach fails. In order to study (11) in more detail in an idealized setup, we pose the following Model problem: Consider a set M⊆[X,2X]\mathcal{M}\subseteq[X,2X] with #M=N\#\mathcal{M}=N, where XX is arbitrarily large in terms of NN (in other words, assume Y=2XY=2X in the above). Further, assume that supp⁡ψ=M\operatorname{supp}\psi=\mathcal{M}, ψ\psi is constant on its support, and the total measure is comparable to 11, i.e.,

∑q∈Mψ(q)log⁡1ψ(q)=1,\sum_{q\in\mathcal{M}}\psi(q)\log\frac{1}{\psi(q)}=1,

which implies

ψ(q)≍1Nlog⁡N,q∈M.\psi(q)\asymp\frac{1}{N\log N},\qquad q\in\mathcal{M}.

The Model problem is to show (11) in this simplified scenario. Now it is particularly easy to compute that

E^q,r≍gcd⁡(q,r)min⁡{ψ(q)q,ψ(r)r}≍1Nlog⁡Ngcd⁡(q,r)qr.\widehat{E}_{q,r}\asymp\gcd(q,r)\min\left\{\frac{\psi(q)}{q},\frac{\psi(r)}{r}\right\}\asymp\frac{1}{N\log N}\frac{\gcd(q,r)}{\sqrt{qr}}.

However, there are known ensembles M⊆N\mathcal{M}\subseteq\mathbb{N} of cardinality NN where

∑q,r∈Mgcd⁡(q,r)qr=ω(Nlog⁡N),\sum_{q,r\in\mathcal{M}}\frac{\gcd(q,r)}{\sqrt{qr}}=\omega(N\log N),

thus (11) fails. Note that this is not due to coarseness of estimates, but a second moment that is really too large.

In order to overcome the issue, we further prune the sets (A~q)q(\widetilde{A}_q)_q, but this time at a local level: for fixed a=(a1,a2)\mathbf{a}=(a_1,a_2), instead of considering the whole hyperbolic region

H‾q+aq={(α,β)∈R2:∣α−a1q∣⋅∣β−a2q∣≤ψ(q)q2,∣α−a1q∣≤12q,∣β−a2q∣≤12q},\overline{H}_q+\frac{\mathbf{a}}{q}=\left\{(\alpha,\beta)\in\mathbb{R}^2:\left|\alpha-\frac{a_1}{q}\right|\cdot\left|\beta-\frac{a_2}{q}\right|\leq\frac{\psi(q)}{q^2},\left|\alpha-\frac{a_1}{q}\right|\leq\frac{1}{2q},\left|\beta-\frac{a_2}{q}\right|\leq\frac{1}{2q}\right\},

we consider the truncated region (see Figure 3)

Hq+aq={(α,β)∈R2:∣α−a1q∣⋅∣β−a2q∣≤ψ(q)q2,∣α−a1q∣≤ψ(q)1/32q,∣β−a2q∣≤ψ(q)1/32q}.H_q+\frac{\mathbf{a}}{q}=\left\{(\alpha,\beta)\in\mathbb{R}^2:\left|\alpha-\frac{a_1}{q}\right|\cdot\left|\beta-\frac{a_2}{q}\right|\leq\frac{\psi(q)}{q^2},\left|\alpha-\frac{a_1}{q}\right|\leq\frac{\psi(q)^{1/3}}{2q},\left|\beta-\frac{a_2}{q}\right|\leq\frac{\psi(q)^{1/3}}{2q}\right\}.

Note that this truncation satisfies λ(Hq)≍λ(H‾q)\lambda(H_q)\asymp\lambda(\overline{H}_q) (see Lemma 5.3), which consequently implies that for

Aq∗:=⋃a∈Zq2gcd⁡(a1,a2,q)=1(Hq+aq),A_q^*:=\bigcup_{\substack{\mathbf{a}\in\mathbb{Z}_q^2\\ \gcd(a_1,a_2,q)=1}}\left(H_q+\frac{\mathbf{a}}{q}\right),

we still have

λ(Aq∗)≍λ(A~q)≍λ(Aq).\lambda(A_q^*)\asymp\lambda(\widetilde{A}_q)\asymp\lambda(A_q).

This pruning results in a new function

w(x)=wq,r(x):=λ((Hq+x)∩Hr),w(\mathbf{x})=w_{q,r}(\mathbf{x}):=\lambda\left((H_q+\mathbf{x})\cap H_r\right),

that has substantially smaller values along the coordinate axes than the original w~\widetilde{w} did. (See Figure 4.) The error term to control is now

Eq,r:=qrgcd⁡(q,r)∫0∞w(0,x) dx.E_{q,r}:=qr\gcd(q,r)\int_0^\infty w(0,x)\,\mathrm{d}x.

A slightly longer yet elementary calculation (Lemma 6.2) shows that

Eq,r≪gcd⁡(q,r)qrmin⁡{ψ(q)7/6,ψ(r)7/6}≍1(Nlog⁡N)7/6gcd⁡(q,r)qr.E_{q,r}\ll\frac{\gcd(q,r)}{\sqrt{qr}}\min\left\{\psi(q)^{7/6},\psi(r)^{7/6}\right\}\asymp\frac{1}{(N\log N)^{7/6}}\frac{\gcd(q,r)}{\sqrt{qr}}.

By a classical GCD sum bound, see Theorem DH,

∑q,r∈Mgcd⁡(q,r)qr≪N1+ε,\sum_{q,r\in\mathcal{M}}\frac{\gcd(q,r)}{\sqrt{qr}}\ll N^{1+\varepsilon},

which implies that

∑q,r∈MEq,r≪1(Nlog⁡N)7/6∑q,r∈Mgcd⁡(q,r)qr≪NεN1/6≪1.\sum_{q,r\in\mathcal{M}}E_{q,r}\ll\frac{1}{(N\log N)^{7/6}}\sum_{q,r\in\mathcal{M}}\frac{\gcd(q,r)}{\sqrt{qr}}\ll\frac{N^\varepsilon}{N^{1/6}}\ll1.

This established (3) in the Model problem, where the support was restricted to dyadic ranges with constant function values on its support.

Going from the Model problem to arbitrary sets M\mathcal{M} with varying values of ψ\psi is mainly a technical challenge that must be undertaken in order to correctly evaluate, or at least efficiently bound, an error term Eq,rE_{q,r} that now depends non-trivially on qq and rr as well as ψ(q)\psi(q) and ψ(r)\psi(r). However, an application of Cauchy–Schwarz reduces this situation essentially to the Model problem. The passage to the higher-dimensional case n≥3n \ge3 also works in a reasonably straightforward manner: instead of primitive vectors, we define the sets A~q\widetilde{A}_q using vectors (a1,…,an)(a_1,\ldots,a_n) where gcd⁡(ai,aj,q)=1\operatorname{gcd}(a_i,a_j,q)=1 for all i≠ji \ne j, a condition which is again satisfied by a positive proportion of all vectors a∈Zqn\mathbf{a} \in\mathbb{Z}_q^n.

Now the nn-dimensional version of the error term E^q,r\widehat{E}_{q,r} contains an (n−1)(n-1)-dimensional integral ∫Rn−1w^(0,x)dx\int_{\mathbb{R}^{n-1}}\widehat{w}(0,\mathbf{x})\mathrm{d}\mathbf{x} whose value is too large for QIA and, again, the issue is fixed by restricting to polynomially truncated sets Aq∗⊆A~qA_q^* \subseteq\widetilde{A}_q. (See Lemma 6.2.)

Notation

Throughout, ψ:N→[0,∞)\psi:\mathbb{N}\to[0,\infty) is an arbitrary function. The quantity ψ(q)(log⁡1ψ(q))n−1\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1} is of special interest, so we reiterate the conventions that log⁡x=max⁡{1,ln⁡x}\log x=\max\{1,\ln x\} and 0⋅log⁡(1/0)n−1=00\cdot\log(1/0)^{n-1}=0.

We study the inequality

∣α1−a1q∣⋅∣α2−a2q∣⋯∣αn−anq∣≤ψ(q)q(a1,…,an,q)∈Zn×N,(12)\left|\alpha_1-\frac{a_1}{q}\right|\cdot\left|\alpha_2-\frac{a_2}{q}\right|\cdots\left|\alpha_n-\frac{a_n}{q}\right|\le\frac{\psi(q)}{q}\qquad(a_1,\ldots,a_n,q)\in\mathbb{Z}^n\times\mathbb{N}, \tag*{(12)}

where α=(α1,…,αn)∈Rn\boldsymbol{\alpha}=(\alpha_1,\ldots,\alpha_n)\in\mathbb{R}^n and, in particular, the set

Gn(ψ)={α∈[0,1]n:(12) holds for infinitely many (a,q)∈Zn×N}G_n(\psi)=\{\boldsymbol{\alpha}\in[0,1]^n:(12)\text{ holds for infinitely many }(\mathbf{a},q)\in\mathbb{Z}^n\times\mathbb{N}\}.

Note that we now adopt the convention of viewing GnG_n as a subset of [0,1]n[0,1]^n. No generality is lost, since its defining condition is Zn\mathbb{Z}^n-periodic. We define

Gq:=Gq(n)(ψ)={(x1,…,xn)∈[−12,12]n:∏i∣xi∣≤ψ(q)},G_q:=G_q^{(n)}(\psi)=\left\{(x_1,\ldots,x_n)\in\left[-\frac{1}{2},\frac{1}{2}\right]^n:\prod_i|x_i|\le\psi(q)\right\},
Hq:=Hq(n)(ψ)=1q(Gq(n)(ψ)∩[−ψ(q)1n+12,ψ(q)1n+12]n),H_q:=H_q^{(n)}(\psi)=\frac{1}{q}\left(G_q^{(n)}(\psi)\cap\left[-\frac{\psi(q)^{\frac{1}{n+1}}}{2},\frac{\psi(q)^{\frac{1}{n+1}}}{2}\right]^n\right),
Aq:=Aq(n)(ψ)={α∈[0,1]n:(12) holds for some a∈Zn},A_q:=A_q^{(n)}(\psi)=\{\boldsymbol{\alpha}\in[0,1]^n:(12)\text{ holds for some }\mathbf{a}\in\mathbb{Z}^n\},
Aq∗:=Aq∗,(n)(ψ)=⋃a∈Zqngcd⁡(ai,aj,q)=1 (i≠j)(aq+Hq),A_q^*:=A_q^{*,(n)}(\psi)=\bigcup_{\substack{\mathbf{a}\in\mathbb{Z}_q^n\\ \gcd(a_i,a_j,q)=1\ (i\ne j)}}\left(\frac{\mathbf{a}}{q}+H_q\right),

and note that lim sup⁡Aq∗⊆lim sup⁡Aq=Gn(ψ)\limsup A_q^*\subseteq\limsup A_q=G_n(\psi).

We use λn\lambda_n to denote Lebesgue measure on [0,1]n[0,1]^n, often dropping the subscript when the context makes it clear, and write Zn:=Z/nZ\mathbb{Z}_n:=\mathbb{Z}/n\mathbb{Z}. We use the classical OO-notation as well as the Vinogradov notation ≪\ll for f≪g⟺f=O(g)⟺∃c>0:f(x)≤cg(x)f\ll g\Longleftrightarrow f=O(g)\Longleftrightarrow\exists c>0:f(x)\le cg(x), and write f≍gf\asymp g if both f=O(g)f=O(g) and g=O(f)g=O(f).

Tools from the literature

The following theorem is due to Dyer–Harman [13 Theorem 1], see also Harman’s monograph [18 Chapter 3, Theorem 3.2].

Theorem DH. Let ε>0\varepsilon>0. Then

∑q,r∈Mgcd⁡(q,r)qrψ(q)ψ(r)≪ε∣M∣ε∑q∈Mψ(q)\sum_{q,r\in\mathcal{M}}\frac{\gcd(q,r)}{\sqrt{qr}}\sqrt{\psi(q)\psi(r)}\ll_\varepsilon|\mathcal{M}|^\varepsilon\sum_{q\in\mathcal{M}}\psi(q)

for all finite sets M⊆N\mathcal{M}\subseteq\mathbb{N} and any ψ:M→[0,1]\psi:\mathcal{M}\to[0,1].

Remark. In fact sharper estimates are known than the simple ∣M∣ε|\mathcal{M}|^\varepsilon, see Aistleitner–Berkes–Seip [1], Lewko–Radziwiłł [23], Bondarenko–Seip [7], or de la Bretèche–Tenenbaum [11] for more recent developments. For our purposes here, the bound stated above suffices.

The following zero-one law is due to Beresnevich–Haynes–Velani [5 Theorem 1].

Theorem BHV. Let n≥1n \ge1 and ψ:N→[0,∞)\psi: \mathbb{N} \to[0,\infty). Then λn(Gn(ψ))∈{0,1}\lambda_n(\mathcal{G}_n(\psi)) \in\{0,1\}.

The following result can be found in [10].

Chung–Erdős Lemma. For measurable sets A1,A2,…,AN⊆(X,μ)A_1,A_2,\ldots,A_N \subseteq(X,\mu) in an arbitrary probability space,

μ(⋃n=1NAn)≥(∑n=1Nμ(An))2∑m,n=1Nμ(Am∩An).\mu\left(\bigcup_{n=1}^{N} A_n\right) \ge\frac{\left(\sum_{n=1}^{N}\mu(A_n)\right)^2}{\sum_{m,n=1}^{N}\mu(A_m\cap A_n)}.

Preliminary lemmas

In this section we gather elementary calculations that go into the proof of Theorem 1. We start with the following lemma that concerns intersections of sets S⊆RnS \subseteq\mathbb{R}^n having the property that

x=(x1,…,xn)∈S  and  ∣xi′∣≤∣xi∣ (i=1,…,n)⟹x′=(x1′,…,xn′)∈S.(13)\mathbf{x}=(x_1,\ldots,x_n)\in S\ \text{ and }\ |x_i'|\leq|x_i|\ (i=1,\ldots,n)\quad\Longrightarrow\quad\mathbf{x}'=(x_1',\ldots,x_n')\in S. \tag*{(13)}

It is a direct adaptation of [16 Lemma 1] to our setup, and it only differs from Gallagher’s lemma in that our sets are symmetric over 00, while Gallagher’s lie in the positive orthant. This is purely an expository choice; the two points of view are equivalent.

Lemma 5.1 ([16 Lemma 1]). If A,B⊆RnA,B\subseteq\mathbb{R}^n satisfy (13), then λ(A∩(B+t))\lambda(A\cap(B+\mathbf{t})) is a decreasing function of ∣ti∣|t_i| (1≤i≤n1\leq i\leq n) when the other components of t\mathbf{t} are held fixed.

Proof. Write t=(s,t)\mathbf{t}=(\mathbf{s},t) where s=(t1,…,tn−1)\mathbf{s}=(t_1,\ldots,t_{n-1}) and t=tnt=t_n. Then

λ(A∩(B+t))=∫λ1(Ar∩(B+(s,t))r) dr=∫λ1(Ar∩(Br−s+t)) dr.(14)\begin{aligned} \lambda(A\cap(B+\mathbf{t}))=\int\lambda_1\left(A^{\mathbf{r}}\cap(B+(\mathbf{s},t))^{\mathbf{r}}\right)\,\mathrm{d}\mathbf{r} \\ &=\int\lambda_1\left(A^{\mathbf{r}}\cap(B^{\mathbf{r}-\mathbf{s}}+t)\right)\,\mathrm{d}\mathbf{r}. \tag*{(14)} \end{aligned}

where

Ar={t∈R:(r,t)∈A}andBr−s={t∈R:(r−s,t)∈B}.A^{\mathbf{r}}=\{t\in\mathbb{R}:(\mathbf{r},t)\in A\} \qquad\text{and}\qquad B^{\mathbf{r}-\mathbf{s}}=\{t\in\mathbb{R}:(\mathbf{r}-\mathbf{s},t)\in B\}.

By (13), the sets ArA^{\mathbf{r}} and Br−sB^{\mathbf{r}-\mathbf{s}} are intervals centered at 00. Denoting their radii a:=a(r)a:=a(\mathbf{r}) and b:=b(r)b:=b(\mathbf{r}), respectively, we have

λ1(Ar∩(Br−s+t))={min⁡(2a,2b)if ∣t∣<∣a−b∣,a+b−∣t∣if ∣a−b∣≤∣t∣<a+b,0if ∣t∣≥a+b,\lambda_1\left(A^{\mathbf{r}}\cap(B^{\mathbf{r}-\mathbf{s}}+t)\right)= \begin{cases} \min(2a,2b) & \text{if } |t|<|a-b|,\\ a+b-|t| & \text{if } |a-b|\leq|t|<a+b,\\ 0 & \text{if } |t|\geq a+b, \end{cases}

(interpreting ∣∞−∞∣=∞|\infty-\infty|=\infty). This function is decreasing in ∣t∣|t| so, by (14), the measure λ(A∩(B+t))\lambda(A\cap(B+\mathbf{t})) is decreasing in ∣tn∣|t_n|, while holding all other coordinates fixed. The same argument can be carried out for tit_i, i=1,…,ni=1,\ldots,n. □

Next, we show that having pairwise joint coprimality is a generic condition.

Lemma 5.2. Let n≥2n \ge2, and let

Z‾qn:={a=(a1,…,an)∈Zqn:gcd⁡(ai,aj,q)=1 for all 1≤i<j≤n}.\overline{\mathbb{Z}}_q^n := \{\mathbf{a}=(a_1,\ldots,a_n)\in\mathbb{Z}_q^n:\gcd(a_i,a_j,q)=1\text{ for all }1\le i<j\le n\}.

Then we have #Z‾qn≫nqn\#\overline{\mathbb{Z}}_q^n\gg_n q^n.

Proof. Note that q↦#Z‾qnq\mapsto\#\overline{\mathbb{Z}}_q^n is multiplicative, so we can focus on q=pkq=p^k, with pp prime. Here we have

#Z‾qn=#{a=(a1,…,an)∈Zqn:p∤ai ∀ 1≤i≤n}+∑i=1n#{a=(a1,…,an)∈Zqn:p∣ai, p∤aj ∀j≠i, 1≤j≤n}\#\overline{\mathbb{Z}}_q^n=\#\{\mathbf{a}=(a_1,\ldots,a_n)\in\mathbb{Z}_q^n:p\nmid a_i\ \forall\,1\le i\le n\} +\sum_{i=1}^n\#\{\mathbf{a}=(a_1,\ldots,a_n)\in\mathbb{Z}_q^n:p\mid a_i,\ p\nmid a_j\ \forall j\ne i,\ 1\le j\le n\}
=qn((1−1p)n+np(1−1p)n−1)=qn(1−On(1p2)).=q^n\left(\left(1-\frac{1}{p}\right)^n+\frac{n}{p}\left(1-\frac{1}{p}\right)^{n-1}\right)=q^n\left(1-O_n\left(\frac{1}{p^2}\right)\right).

By the multiplicative property, we can lift this to q=∏p∣qpνpq=\prod_{p\mid q}p^{\nu_p}, obtaining

#Z‾qn=qn∏p∣q(1−On(1p2))≫nqn.\#\overline{\mathbb{Z}}_q^n=q^n\prod_{p\mid q}\left(1-O_n\left(\frac{1}{p^2}\right)\right)\gg_n q^n.

Lemma 5.3. Let ψ:N→[0,∞)\psi:\mathbb{N}\to[0,\infty) and n≥2n\ge2. For all ψ(q)≤2−n\psi(q)\le2^{-n}, we have

λ(Gq)≍qnλ(Hq)≍λ(Aq)≍λ(Aq∗)≍ψ(q)(log⁡1ψ(q))n−1,\lambda(G_q)\asymp q^n\lambda(H_q)\asymp\lambda(A_q)\asymp\lambda(A_q^*)\asymp\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1},

with implied constants depending only on nn. For ψ(q)≥2−n\psi(q)\ge2^{-n}, we have λ(Gq)=λ(Aq)=1\lambda(G_q)=\lambda(A_q)=1.

Proof. The last claim follows from the fact that [−1/2,1/2]n⊆Gq[-1/2,1/2]^n\subseteq G_q when ψ(q)≥2−n\psi(q)\ge2^{-n}. Hence we may assume for the rest of the proof that ψ(q)≤2−n\psi(q)\le2^{-n}. We prove the claim for GqG_q by induction on nn. When n=2n=2,

λ(Gq(2))=4ψ(q)+4∫2ψ(q)1/2ψ(q)y dy≍ψ(q)log⁡1ψ(q),\lambda(G_q^{(2)})=4\psi(q)+4\int_{2\psi(q)}^{1/2}\frac{\psi(q)}{y}\,dy\asymp\psi(q)\log\frac{1}{\psi(q)},

establishing the base case. Now let n≥3n\ge3 and assume that the claim holds up to n−1n-1. Writing (x,y)∈Rn−1×R(\mathbf{x},y)\in\mathbb{R}^{n-1}\times\mathbb{R}, we have

λ(Gq)=∫λn−1(Gqy) dy,\lambda(G_q)=\int\lambda_{n-1}(G_q^y)\,dy,

where

Gqy={x∈Rn−1:(x,y)∈Gq}G_q^y=\{\mathbf{x}\in\mathbb{R}^{n-1}:(\mathbf{x},y)\in G_q\}
={x∈[−12,12]n−1:∣x1∣⋯∣xn−1∣≤ψ(q)∣y∣}=\left\{\mathbf{x}\in\left[-\frac{1}{2},\frac{1}{2}\right]^{n-1}:|x_1|\cdots|x_{n-1}|\le\frac{\psi(q)}{|y|}\right\}
=Gq(n−1)(ψ(q)∣y∣).=G_q^{(n-1)}\left(\frac{\psi(q)}{|y|}\right).

with ψy=ψ/∣y∣\psi_y=\psi/|y|. Note that if ∣y∣≤2n−1ψ(q)|y|\le2^{n-1}\psi(q), then Gqy=[−1/2,1/2]n−1G_q^y=[-1/2,1/2]^{n-1}. Consequently,

λ(Gq)=2nψ(q)+2∫2n−1ψ(q)1/2λn−1(Gq(n−1)(ψy)) dy(15)\lambda(G_q)=2^n\psi(q)+2\int_{2^{n-1}\psi(q)}^{1/2}\lambda_{n-1}\left(G_q^{(n-1)}(\psi_y)\right)\,dy \tag*{(15)}
≍ψ(q)+∫2n−1ψ(q)1/2ψ(q)y(log⁡yψ(q))n−2 dy\asymp\psi(q)+\int_{2^{n-1}\psi(q)}^{1/2}\frac{\psi(q)}{y}\left(\log\frac{y}{\psi(q)}\right)^{n-2}\,dy
≍ψ(q)(log⁡1ψ(q))n−1,\asymp\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1},

which establishes the claim for λ(Gq)\lambda(G_q). Notice that Aq=T−1(Gq)A_q=T^{-1}(G_q) where T:Tn→TnT:\mathbb{T}^n\to\mathbb{T}^n is the mapping x↦qx(mod1)\mathbf{x}\mapsto q\mathbf{x}\pmod1. Since TT is Lebesgue measure-preserving, the result for AqA_q holds.

For HqH_q, notice that

qψ(q)−1n+1Hq={x∈[−12,12]n:∣x1∣⋯∣xn∣≤ψ(q)1n+1}=Gq(ψ1n+1).(16)q\psi(q)^{-\frac{1}{n+1}}H_q=\left\{\mathbf{x}\in\left[-\frac{1}{2},\frac{1}{2}\right]^n:|x_1|\cdots|x_n|\le\psi(q)^{\frac{1}{n+1}}\right\}=G_q\left(\psi^{\frac{1}{n+1}}\right). \tag*{(16)}

It follows that

λ(qψ(q)−1n+1Hq)≍ψ(q)1n+1(log⁡1ψ(q)1n+1)n−1,\lambda\left(q\psi(q)^{-\frac{1}{n+1}}H_q\right)\asymp\psi(q)^{\frac{1}{n+1}}\left(\log\frac{1}{\psi(q)^{\frac{1}{n+1}}}\right)^{n-1},

which implies that

qnλ(Hq)≍ψ(q)(log⁡1ψ(q))n−1,q^n\lambda(H_q)\asymp\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1},

establishing the claim for qnλ(Hq)q^n\lambda(H_q).

For Aq∗A_q^*, recall that Aq∗A_q^* is a disjoint union of #Z‾qn\#\overline{\mathbb Z}_q^n translates of HqH_q. Thus, by Lemma 5.2,

λ(Aq∗)=#Z‾qnλ(Hq)≍nqnλ(Hq),\lambda(A_q^*)=\#\overline{\mathbb Z}_q^n\lambda(H_q)\asymp_n q^n\lambda(H_q),

which implies the claim for Aq∗A_q^* and completes the proof. □

The overlap estimate

Proposition 6.1. Let q≠rq\ne r, ψ(q),ψ(r)≤2−n\psi(q),\psi(r)\le2^{-n}. Then

λ(Aq∗∩Ar∗)≪λ(Aq∗)λ(Ar∗)+gcd⁡(q,r)ψ(q)ψ(r)qrmin⁡(ψ(q),ψ(r))1/6.\lambda(A_q^*\cap A_r^*)\ll\lambda(A_q^*)\lambda(A_r^*)+\gcd(q,r)\frac{\sqrt{\psi(q)\psi(r)}}{\sqrt{qr}}\min(\psi(q),\psi(r))^{1/6}.

Proof. Define the weight function w=wq,r(n):Rn→R≥0w=w_{q,r}^{(n)}:\mathbb{R}^n\to\mathbb{R}_{\ge0} by

w(x):=λ(Hq∩(Hr+x)).w(\mathbf{x}):=\lambda\left(H_q\cap(H_r+\mathbf{x})\right).

Then

λ(Aq∗∩Ar∗)=∑a∈Z‾qn∑b∈Z‾rnλ((Hq+aq)∩(Hr+br))=∑c∈Znrep⁡(c)w(clcm⁡(q,r)),\lambda(A_q^*\cap A_r^*)=\sum_{\mathbf{a}\in\overline{\mathbb{Z}}_q^n}\sum_{\mathbf{b}\in\overline{\mathbb{Z}}_r^n}\lambda\left(\left(H_q+\frac{\mathbf{a}}{q}\right)\cap\left(H_r+\frac{\mathbf{b}}{r}\right)\right)=\sum_{\mathbf{c}\in\mathbb{Z}^n}\operatorname{rep}(\mathbf{c})w\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right),

where

rep⁡(c)=#{(a,b)∈Z‾qn×Z‾rn:aq−br=clcm⁡(q,r)}.\operatorname{rep}(\mathbf{c})=\#\left\{(\mathbf{a},\mathbf{b})\in\overline{\mathbb{Z}}_q^n\times\overline{\mathbb{Z}}_r^n:\frac{\mathbf{a}}{q}-\frac{\mathbf{b}}{r}=\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right\}.

The Chinese Remainder Theorem gives rep⁡(c)≤gcd⁡(q,r)n\operatorname{rep}(\mathbf{c})\leq\gcd(q,r)^n, and by the definition of Z‾q\overline{\mathbb{Z}}_q, we have rep⁡(c)=0\operatorname{rep}(\mathbf{c})=0 if c\mathbf{c} has at least two vanishing entries. Consequently,

λ(Aq∗∩Ar∗)gcd⁡(q,r)n≪∑c∈Znci>0,i=1,…,nw(clcm⁡(q,r))+∑i=1n∑c∈Znci=0,cj>0,j≠iw(clcm⁡(q,r))=∑c∈Znci>0,i=1,…,nw(clcm⁡(q,r))+n∑c∈Znc1=0,cj>0,j≥2w(clcm⁡(q,r)),\begin{aligned} \frac{\lambda\left(A_q^*\cap A_r^*\right)}{\gcd(q,r)^n}\ll \sum_{\substack{\mathbf{c}\in\mathbb{Z}^n\\c_i>0,\,i=1,\ldots,n}} w\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right) +\sum_{i=1}^n\sum_{\substack{\mathbf{c}\in\mathbb{Z}^n\\c_i=0,\,c_j>0,\,j\neq i}} w\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right) \\ &= \sum_{\substack{\mathbf{c}\in\mathbb{Z}^n\\c_i>0,\,i=1,\ldots,n}} w\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right) +n\sum_{\substack{\mathbf{c}\in\mathbb{Z}^n\\c_1=0,\,c_j>0,\,j\geq2}} w\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right), \end{aligned}

using the symmetry of ww. Note that HqH_q is an intersection of two sets that both satisfy (13), hence HqH_q also satisfies (13). Therefore, Lemma 5.1 shows that ww is a decreasing function in each coordinate ∣xi∣|x_i|. Using this property, we obtain

∑c∈Znci>0,i=1,…,nw(clcm⁡(q,r))≤lcm⁡(q,r)n∫Rnw(x) dx,\sum_{\substack{\mathbf{c}\in\mathbb{Z}^n\\c_i>0,\,i=1,\ldots,n}} w\left(\frac{\mathbf{c}}{\operatorname{lcm}(q,r)}\right) \leq\operatorname{lcm}(q,r)^n\int_{\mathbb{R}^n}w(\mathbf{x})\,d\mathbf{x},

as well as

∑c∈Zn−1cj>0,1≤j≤n−1w((0,c)lcm⁡(q,r))≤lcm⁡(q,r)n−1∫Rn−1w(0,x) dx.\sum_{\substack{\mathbf{c}\in\mathbb{Z}^{n-1}\\c_j>0,\,1\leq j\leq n-1}} w\left(\frac{(0,\mathbf{c})}{\operatorname{lcm}(q,r)}\right) \leq\operatorname{lcm}(q,r)^{n-1}\int_{\mathbb{R}^{n-1}}w(0,\mathbf{x})\,d\mathbf{x}.

Note that by Lemma 5.3,

∫Rnw(x) dx=∫Rnλ(Hq∩(Hr+x)) dx=λ(Hq)λ(Hr)≍λ(Aq∗)λ(Ar∗)qnrn,\int_{\mathbb{R}^n}w(\mathbf{x})\,d\mathbf{x} =\int_{\mathbb{R}^n}\lambda(H_q\cap(H_r+\mathbf{x}))\,d\mathbf{x} =\lambda(H_q)\lambda(H_r)\asymp\frac{\lambda(A_q^*)\lambda(A_r^*)}{q^n r^n},

hence

λ(Aq∗∩Ar∗)≪λ(Aq∗)λ(Ar∗)+qn−1rn−1gcd⁡(q,r)∫Rn−1w(0,x) dx.\lambda(A_q^*\cap A_r^*)\ll\lambda(A_q^*)\lambda(A_r^*)+q^{n-1}r^{n-1}\gcd(q,r)\int_{\mathbb{R}^{n-1}}w(0,\mathbf{x})\,d\mathbf{x}.

The proposition follows by applying the following lemma. □

Lemma 6.2. Let n≥2n\geq2, q,r∈Nq,r\in\mathbb{N}, 0≤ψ(q),ψ(r)≤2−n0\leq\psi(q),\psi(r)\leq2^{-n}. Then we have

qn−1rn−1gcd⁡(q,r)∫Rn−1w(0,x) dx≪gcd⁡(q,r)ψ(q)ψ(r)qrmin⁡(ψ(q),ψ(r))1/6.q^{n-1}r^{n-1}\gcd(q,r)\int_{\mathbb{R}^{n-1}}w(0,\mathbf{x})\,d\mathbf{x} \ll\gcd(q,r)\frac{\sqrt{\psi(q)\psi(r)}}{\sqrt{qr}}\min(\psi(q),\psi(r))^{1/6}.

Proof. Observe that with HqyH_q^y defined analogous to GqyG_q^y in the proof of Lemma 5.3,

∫Rn−1w(0,x) dx=∫Rn−1λn(Hq∩(Hr+(0,x))) dx=∫Rn−1∫Rλn−1(Hqy∩(Hry+x)) dy dx\int_{\mathbb{R}^{n-1}}w(0,\mathbf{x})\,d\mathbf{x} =\int_{\mathbb{R}^{n-1}}\lambda_n(H_q\cap(H_r+(0,\mathbf{x})))\,d\mathbf{x} =\int_{\mathbb{R}^{n-1}}\int_{\mathbb{R}}\lambda_{n-1}(H_q^y\cap(H_r^y+\mathbf{x}))\,dy\,d\mathbf{x}
=∫R∫Rn−1λn−1(Hqy∩(Hry+x)) dx dy=∫Rλn−1(Hqy)λn−1(Hry) dy=\int_{\mathbb{R}}\int_{\mathbb{R}^{n-1}}\lambda_{n-1}(H_q^y\cap(H_r^y+\mathbf{x}))\,d\mathbf{x}\,dy =\int_{\mathbb{R}}\lambda_{n-1}(H_q^y)\lambda_{n-1}(H_r^y)\,dy
≤(∫Rλn−1(Hqy)2 dy)1/2(∫Rλn−1(Hry)2 dy)1/2,(17)\leq\left(\int_{\mathbb{R}}\lambda_{n-1}(H_q^y)^2\,dy\right)^{1/2} \left(\int_{\mathbb{R}}\lambda_{n-1}(H_r^y)^2\,dy\right)^{1/2}, \tag*{(17)}

by the Cauchy–Schwarz inequality. Now,

∫Rλn−1(Hqy)2 dy=q−2(n−1)ψ(q)2(n−1)n+1∫Rλn−1(qψ(q)−1n+1Hq)qψ(q)−1n+1y2 dy=q−2(n−1)−1ψ(q)2(n−1)+1n+1∫Rλn−1(qψ(q)−1n+1Hqu)2 du=q−2n+1ψ(q)2n−1n+1∫Rλn−1(Gq(ψ(q)1n+1u))2 du\begin{aligned} \int_{\mathbb{R}} \lambda_{n-1}\left(H_q^y\right)^2\,\mathrm{d}y = q^{-2(n-1)}\psi(q)^{\frac{2(n-1)}{n+1}} \int_{\mathbb{R}} \lambda_{n-1}\left(q\psi(q)^{-\frac{1}{n+1}}H_q\right)^{q\psi(q)^{-\frac{1}{n+1}}y}{}^2\,\mathrm{d}y \\ &= q^{-2(n-1)-1}\psi(q)^{\frac{2(n-1)+1}{n+1}} \int_{\mathbb{R}} \lambda_{n-1}\left(q\psi(q)^{-\frac{1}{n+1}}H_q^u\right)^2\,\mathrm{d}u \\ &= q^{-2n+1}\psi(q)^{\frac{2n-1}{n+1}} \int_{\mathbb{R}} \lambda_{n-1}\left(G_q\left(\psi(q)^{\frac{1}{n+1}}u\right)\right)^2\,\mathrm{d}u \end{aligned}

by (16). Continuing as in (15), we bound the above by

≪q−2n+1ψ(q)2n−1n+1(2n−1ψ(q)1n+1+∫2n−1ψ(q)1n+1∞(ψ(q)1/(n+1)u(log⁡uψ(q)1/(n+1))n−2)2 du)≪q−2n+1ψ(q)2n−1n+1(ψ(q)1n+1+ψ(q)1n+1∫1∞(1v(log⁡v)n−2)2 dv)≪q−2n+1ψ(q)ψ(q)n−1n+1≪q−2n+1ψ(q)ψ(q)13.\begin{aligned} \ll q^{-2n+1}\psi(q)^{\frac{2n-1}{n+1}}\left(2^{n-1}\psi(q)^{\frac{1}{n+1}}+\int_{2^{n-1}\psi(q)^{\frac{1}{n+1}}}^{\infty}\left(\frac{\psi(q)^{1/(n+1)}}{u}\left(\log\frac{u}{\psi(q)^{1/(n+1)}}\right)^{n-2}\right)^2\,\mathrm{d}u\right) \\ &\ll q^{-2n+1}\psi(q)^{\frac{2n-1}{n+1}}\left(\psi(q)^{\frac{1}{n+1}}+\psi(q)^{\frac{1}{n+1}}\int_1^\infty\left(\frac{1}{v}(\log v)^{n-2}\right)^2\,\mathrm{d}v\right) \\ &\ll q^{-2n+1}\psi(q)\psi(q)^{\frac{n-1}{n+1}}\ll q^{-2n+1}\psi(q)\psi(q)^{\frac{1}{3}}. \end{aligned}

Similarly, of course,

∫Rλn−1(Hry)2 dy≪r−2n+1ψ(r)ψ(r)13.\int_{\mathbb{R}} \lambda_{n-1}\left(H_r^y\right)^2\,\mathrm{d}y\ll r^{-2n+1}\psi(r)\psi(r)^{\frac{1}{3}}.

Putting these into (17) gives

qn−1rn−1gcd⁡(q,r)∫Rn−1w(0,x) dx≪gcd⁡(q,r)ψ(q)ψ(r)qrmin⁡(ψ(q),ψ(r))16,q^{n-1}r^{n-1}\gcd(q,r)\int_{\mathbb{R}^{n-1}} w(0,\mathbf{x})\,\mathrm{d}\mathbf{x}\ll\gcd(q,r)\frac{\sqrt{\psi(q)\psi(r)}}{\sqrt{qr}}\min(\psi(q),\psi(r))^{\frac{1}{6}},

proving the lemma. □

Remark. The proof shows that in fact in place of 16\frac{1}{6}, the exponent n−12n+2=12+o(1)\frac{n-1}{2n+2}=\frac{1}{2}+o(1) is admissible in the above. This follows the recurring theme in metric Diophantine approximation that higher-dimensional problems are easier to prove because of the smaller overlaps of their concomitant sets.

Proof of Theorems 1 and 2

Proof of Theorem 2

Proof. Let Sq=[−ψ(q),ψ(q)]×[−12,12]n−1S_q=[-\psi(q),\psi(q)]\times[-\frac{1}{2},\frac{1}{2}]^{n-1}, where ψ\psi is a function such that ∑ψ(q)=∞\sum\psi(q)=\infty and such that almost no real number is ψ\psi-approximable. That is, ψ\psi is a Duffin–Schaeffer-type counterexample, such as was first constructed in [12].

Since SqS_q is a product of symmetric intervals, it satisfies (13). Notice that λn(Sq)=2ψ(q)\lambda_n(S_q)=2\psi(q), hence ∑λn(Sq)\sum\lambda_n(S_q) diverges. Finally, notice that

W=W1(ψ)×[0,1]n−1,\mathcal{W}=\mathcal{W}_1(\psi)\times[0,1]^{n-1},

where W1(ψ)\mathcal{W}_1(\psi) denotes the set of ψ\psi-approximable numbers. Therefore λn(W)=0\lambda_n(\mathcal{W})=0 by Fubini’s theorem. □

Proof of Theorem 1

The convergence part of Theorem 1 is standard: by Lemma 5.3, the convergence assumption in the theorem statement implies that ∑q≥1λ(Aq)\sum_{q\geq1}\lambda(A_q) converges, so, by the Borel–Cantelli lemma, the limsup set Gn(ψ)G_n(\psi) has zero measure. For the divergence part of Theorem 1, assume

∑q=1∞ψ(q)(log⁡1ψ(q))n−1=∞.\sum_{q=1}^{\infty}\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1}=\infty.

Since Gn(ψ)G_n(\psi) contains the simultaneously ψ1/n\psi^{1/n}-approximable points, the result would follow as a consequence of [17 Theorem 1] if ∑ψ(q)\sum\psi(q) diverged. We may therefore assume ∑ψ(q)\sum\psi(q) converges and, in particular, that ψ(q)→0\psi(q)\to0. This implies that

lim⁡q→∞ψ(q)(log⁡1ψ(q))n−1=0.\lim_{q\to\infty}\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1}=0.

We may now find arbitrarily large X<YX<Y such that

1≤∑X≤q≤Yψ(q)(log⁡1ψ(q))n−1≤2,(18)1\leq\sum_{X\leq q\leq Y}\psi(q)\left(\log\frac{1}{\psi(q)}\right)^{n-1}\leq2, \tag*{(18)}

with max⁡X≤q≤Yψ(q)≤2−n\max_{X\leq q\leq Y}\psi(q)\leq2^{-n} and in particular,

∑X≤q≤Yλ(Aq∗)≍1(19)\sum_{X\leq q\leq Y}\lambda(A_q^*)\asymp1 \tag*{(19)}

by Lemma 5.3. We have

∑X≤q,r≤Yλ(Aq∗∩Ar∗)≪∑X≤q≤Yλ(Aq∗)+∑X≤r<q≤Yλ(Aq∗∩Ar∗)≪(∑X≤q≤Yλ(Aq∗))2+∑X≤r<q≤Ygcd⁡(q,r)qrψ(q)ψ(r)min⁡(ψ(q),ψ(r))1/6,\begin{aligned} \sum_{X\leq q,r\leq Y}\lambda(A_q^*\cap A_r^*) &\ll\sum_{X\leq q\leq Y}\lambda(A_q^*)+\sum_{X\leq r<q\leq Y}\lambda(A_q^*\cap A_r^*)\\ &\ll\left(\sum_{X\leq q\leq Y}\lambda(A_q^*)\right)^2+\sum_{X\leq r<q\leq Y}\frac{\gcd(q,r)}{\sqrt{qr}}\sqrt{\psi(q)\psi(r)}\min(\psi(q),\psi(r))^{1/6}, \end{aligned}

by Lemma 6.1. Now

∑X≤r<q≤Ygcd⁡(q,r)qrψ(q)ψ(r)min⁡(ψ(q),ψ(r))1/6≤∑j≥1∑X≤r<q≤Y2−j≤min⁡(ψ(q),ψ(r))≤2−j+1gcd⁡(q,r)qrψ(q)ψ(r)min⁡(ψ(q),ψ(r))1/6≪∑j≥12−j/6∑X≤r<q≤Yψ(q),ψ(r)≥2−jgcd⁡(q,r)qrψ(q)ψ(r)≪∑j≥12−j/6(#{q∈[X,Y]:ψ(q)≥2−j})ε∑q∈[X,Y]ψ(q).\begin{aligned} \sum_{X\leq r<q\leq Y}\frac{\gcd(q,r)}{\sqrt{qr}}\sqrt{\psi(q)\psi(r)}\min(\psi(q),\psi(r))^{1/6} &\leq\sum_{j\geq1}\sum_{\substack{X\leq r<q\leq Y\\2^{-j}\leq\min(\psi(q),\psi(r))\leq2^{-j+1}}}\frac{\gcd(q,r)}{\sqrt{qr}}\sqrt{\psi(q)\psi(r)}\min(\psi(q),\psi(r))^{1/6}\\ &\ll\sum_{j\geq1}2^{-j/6}\sum_{\substack{X\leq r<q\leq Y\\\psi(q),\psi(r)\geq2^{-j}}}\frac{\gcd(q,r)}{\sqrt{qr}}\sqrt{\psi(q)\psi(r)}\\ &\ll\sum_{j\geq1}2^{-j/6}\left(\#\{q\in[X,Y]:\psi(q)\geq2^{-j}\}\right)^\varepsilon\sum_{q\in[X,Y]}\psi(q). \end{aligned}

by Theorem DH (§ 5). From (18), we obtain

#{q∈[X,Y]:ψ(q)≥2−j}≪2j,\#\{q \in[X,Y] : \psi(q) \ge2^{-j}\} \ll2^j,

therefore, for ε<16\varepsilon< \frac{1}{6},

∑X≤q,r≤Yλ(Aq∗∩Ar∗)≪(∑X≤q≤Yλ(Aq∗))2+(∑j≥12−j(1/6−ε))∑q∈[X,Y]ψ(q)≪(∑X≤q≤Yλ(Aq∗))2+∑q∈[X,Y]ψ(q)≪(∑X≤q≤Yλ(Aq∗))2.\begin{aligned} \sum_{X\le q,r\le Y} \lambda(A_q^* \cap A_r^*) &\ll\left(\sum_{X\le q\le Y} \lambda(A_q^*)\right)^2 + \left(\sum_{j\ge1}2^{-j(1/6-\varepsilon)}\right)\sum_{q\in[X,Y]}\psi(q)\\ &\ll\left(\sum_{X\le q\le Y} \lambda(A_q^*)\right)^2 + \sum_{q\in[X,Y]}\psi(q) \ll\left(\sum_{X\le q\le Y} \lambda(A_q^*)\right)^2. \end{aligned}

It follows by the Chung–Erdős Lemma (§ 5) that

λ(⋃X≤q≤YAq∗)≫1,\lambda\left(\bigcup_{X\le q\le Y} A_q^*\right) \gg1,

and since there are arbitrarily large X<YX<Y satisfying (19), this shows that

λ(lim sup⁡q→∞Aq∗)>0,\lambda\left(\limsup_{q\to\infty} A_q^*\right)>0,

hence λ(Gn(ψ))>0\lambda(G_n(\psi))>0. By Theorem BHV (§ 5), this implies λ(Gn(ψ))=1\lambda(G_n(\psi))=1. □

AI declaration

We used ChatGPT-6 Astra to generate TikZ code for Figures 1 to 4, which we then adjusted. We did not use AI for any of the mathematical work or the writing.

References

  1. [1]C. Aistleitner, I. Berkes, and K. Seip. GCD sums from Poisson integrals and systems of dilated functions. J. Eur. Math. Soc. (JEMS), 17(6):1517–1546, 2015. 11arxiv.org/abs/1210.0741
  2. [2]D. Allen, M. Hauke-Treuer, and F. A. Ramírez. The inhomogeneous Khintchine Theorem in dimension two, 2026. Preprint, arXiv:2605.19582. 2arxiv.org/abs/2605.19582
  3. [3]D. Allen and F. A. Ramírez. Independence inheritance and Diophantine approximation for systems of linear forms. Int. Math. Res. Not. IMRN, (2):1760–1794, 2023. 2arxiv.org/abs/2109.03929
  4. [4]D. Allen and F. A. Ramírez. Inhomogeneous approximation for systems of linear forms with primitivity constraints. Math. Z., 309(1):Paper No. 10, 27, 2025. 2arxiv.org/abs/2305.16098
  5. [5]V. Beresnevich, A. Haynes, and S. Velani. Multiplicative zero-one laws and metric number theory. Acta Arith., 160(2):101–114, 2013. 2, 3, 4, 6, 12arxiv.org/abs/1012.0675
  6. [6]V. Beresnevich and S. Velani. Classical metric Diophantine approximation revisited: the Khintchine-Groshev theorem. Int. Math. Res. Not. IMRN, (1):69–86, 2010. 2arxiv.org/abs/0811.0809
  7. [7]A. Bondarenko and K. Seip. GCD sums and complete sets of square-free numbers. Bull. Lond. Math. Soc., 47(1):29–41, 2015. 11DOI
  8. [9]S. Chow and N. Technau. Littlewood and Duffin-Schaeffer-type problems in Diophantine approximation. Mem. Amer. Math. Soc., 296(1475):v+74, 2024. 2
  9. [10]K. L. Chung and P. Erdős. On the application of the Borel-Cantelli lemma. Trans. Amer. Math. Soc., 72:179–186, 1952. 12
  10. [11]R. de la Bretèche and G. Tenenbaum. Sommes de Gál et applications. Proc. London Math. Soc. (3), 119(1):104–134, 2019. 11DOI
  11. [12]R. J. Duffin and A. C. Schaeffer. Khintchine’s problem in metric Diophantine approximation. Duke Math. J., 8:243–255, 1941. 2, 16DOI
  12. [13]T. Dyer and G. Harman. Sums involving common divisors. J. London Math. Soc. (2), 34(1):1–11, 1986. 11DOI
  13. [14]M. Einsiedler, A. Katok, and E. Lindenstrauss. Invariant measures and the set of exceptions to Littlewood’s conjecture. Ann. of Math. (2), 164(2):513–560, 2006. 1DOI
  14. [15]L. Frühwirth and M. Hauke. The Duffin–Schaeffer conjecture for multiplicative Diophantine approximation, 2024. Preprint, arXiv:2403.11257. 2, 6, 7arxiv.org/abs/2403.11257
  15. [16]P. Gallagher. Metric simultaneous diophantine approximation. J. London Math. Soc., 37:387–390, 1962. 1, 2, 3, 5, 6, 12DOI
  16. [17]P. X. Gallagher. Metric simultaneous diophantine approximation. II. Mathematika, 12:123–127, 1965. 2, 8, 17DOI
  17. [18]G. Harman. Metric number theory, volume 18 of London Mathematical Society Monographs. New Series. The Clarendon Press Oxford University Press, New York, 1998. 2, 11
  18. [19]M. Hauke and F. A. Ramirez. The Duffin–Schaeffer conjecture with a moving target. Discrete Analysis, to appear. arXiv:2407.05344. 2arxiv.org/abs/2407.05344
  19. [20]A. Khintchine. Einige Sätze über Kettenbrüche, mit Anwendungen auf die Theorie der Diophantischen Approximationen. Math. Ann., 92(1-2):115–125, 1924. 2
  20. [21]A. Khintchine. Zur metrischen Theorie der diophantischen Approximationen. Math. Z., 24(1):706–714, 1926. 2, 8DOI
  21. [22]D. Koukoulopoulos and J. Maynard. On the Duffin–Schaeffer conjecture. Ann. of Math. (2), 192(1):251–307, 2020. 2
  22. [23]M. Lewko and M. Radziwiłł. Refinements of Gal’s theorem and applications. Adv. Math., 305:280–297, 2017. 11arxiv.org/abs/1408.2334
  23. [24]A. D. Pollington and R. C. Vaughan. The k-dimensional Duffin and Schaeffer conjecture. Mathematika, 37(2):190–200, 1990. 2
  24. [25]F. A. Ramírez. The Duffin–Schaeffer conjecture for systems of linear forms. Journal of the London Mathematical Society, 109(5):e12909, 2024. 2arxiv.org/abs/2211.09996
  25. [26]V. G. Sprindžuk. Metric theory of Diophantine approximations. Scripta Series in Mathematics. V. H. Winston & Sons, Washington, D.C.; John Wiley & Sons, New York-Toronto, Ont.-London, 1979. Translated from the Russian and edited by Richard A. Silverman, With a foreword by Donald J. Newman. 2

Paper details

vlnc:W_k3dsatqc
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents