Metric Diophantine approximation on the middle-third Cantor set (Q294192): Difference between revisions

In [Bull. Aust. Math. Soc. 29, 101--108 (1984; Zbl 0517.10001)], \textit{K. Mahler} asked how closely irrational elements of the middle-third Cantor set \(K\) can be approximated by rationals in \(K\) as well as by rationals outside \(K\). The problem has attracted a great deal of attention in recent years. In the present paper, the authors consider the problem from the metrical point of view in terms of irrationality exponents. Define for a real number \(\xi\), the irrationality exponent \(\mu(\xi)\) as the supremum of all \(\mu\) for which the inequality \[ | \xi - p/q | < q^{-\mu} \] has infinitely many solutions \(p/q \in {\mathbb Q}\). Let \({\mathcal M}(\mu)\) denote the set of real numbers \(\xi\) with \(\mu(\xi) \geq \mu\). The classical Jarník-Besicovitch Theorem states that this set has Hausdorff dimension \(2/\mu\). The object of study in the present paper is the Hausdorff dimension of the set \({\mathcal M} \cap K\). Letting \(\kappa = \log 2/ \log 3\) denote the Hausdorff dimension of \(K\), the authors conjecture that \[ \dim({\mathcal M}(\mu) \cap K) = \max\{2/\mu + \kappa - 1, \kappa/\mu\}. \] A proof of this conjecture would involve obtaining detailed information about the distribution of rational numbers nearby \(K\), which is currently unavailable. Nonetheless, the authors provide compelling evidence towards this conjecture. The essence of the conjecture is in the range \([2, (2-\kappa)/(1-\kappa)]\), where the maximum above is attained by \(2/\mu + \kappa - 1\). This is the range in which the approximating rationals outside of \(K\) are conjectured to make a significant contribution to the Hausdorff dimension of the set. First, some heuristic arguments are presented in favour of the conjecture. It is then shown that for Lebesgue-almost all \(\alpha\) and any \(\mu \in [2, 2/(1-\kappa)]\), the Hausdorff dimension of \({\mathcal M}(\mu) \cap (K + \alpha)\) is bounded from above by the conjectured value \(2/\mu + \kappa - 1\). Subsequently, the authors replace approximation by rationals with approximation by random points whose distributional properties are similar to those of rationals. It is shown that with this modification, the conjecture is true with probability one. This part of the paper constitutes the main results of the paper. Along with the study of the exponents \(\mu(\xi)\), the exponents \(\nu_b(\xi)\) are also considered. These are defined for a fixed integer \(b \geq 2\) as the supremum of all \(\nu\) for which the inequality \[ \| b^j \xi \| < b^{-\nu j} \] is satisfied for infinitely many \(j \in {\mathbb N}\). In this setting, analogous conjectures are made for the Hausdorff dimension of the corresponding intersections; and the probabilistic model is shown to provide evidence for these. Along the way, various results of independent interest about metric Diophantine approximation are obtained, and a number of additional open problems are suggested.