Strengthening of the Bourgain-Kontorovich theorem on small values of Hausdorff dimension (Q2160241)

Let \(\mathbf{A} \subseteq \mathbb{N}\) be a finite alphabet and let \(\mathcal{D}_{\mathbf{A}}\) be the set of all natural numbers \(M\) with the property that there is a reduced rational number \(a/M\) such that all partial quotients in its continued fraction expansion belong to \(\mathbf{A}\). \textit{S. K. Zaremba} [in: Appl. Number Theory Numer. Analysis, Proc. Sympos. Univ. Montreal 1971, 39--119 (1972; Zbl 0246.65009)] famously conjectured that there is a constant \(A \in \mathbb{N}\) such that for \(\mathbf{A} = \{1,\ldots,A\}\), we have \(\mathcal{D}_{\mathbf{A}} = \mathbb{N}\), also conjecturing that \(A = 5\) is sufficient. \textit{J. Bourgain} and \textit{A. Kontorovich} [Ann. Math. (2) 180, No. 1, 137--196 (2014; Zbl 1370.11083)] made significant progress towards the solution, showing that \(\frac{1}{N}\lvert \mathcal{D}_{\mathbf{A}} \cap [1,N]\rvert \to 1\) as \(N \to \infty\), provided that \(A \geq 50\). This result follows from several statements proven in a more general setting: Let \(\mathcal{C}_{\mathbf{A}}\) denote the set of all \(x \in (0,1)\) such that all partial quotients in its continued fraction belong to \(\mathbf{A}\) and let \(\Delta_{\mathbf{A}}\) denote the Hausdorff dimension of \(\mathcal{C}_{\mathbf{A}}\). Among other results, Bourgain and Kontorovich showed that for any \(\mathbf{A}\) that satisifies \(\Delta_{\mathbf{A}} > 1/2\), and any \(\varepsilon > 0\), \[ \lvert \mathcal{D}_{\mathbf{A}} \cap [1,N]\rvert \gg_{\varepsilon} N^{\Delta_{\mathbf{A}} + (2\Delta_{\mathbf{A}} -1)\frac{1 - \Delta_{\mathbf{A}}}{5 - \Delta_{\mathbf{A}}} - \varepsilon}. \] This estimate was improved by the author in an earlier paper [Sb. Math. 212, No. 7, 921--964 (2021); translation from Mat. Sb. 212, No. 7, 39--83 (2021; Zbl 1481.11073)] for a range of \(\Delta_{\mathbf{A}}\) close to \(1/2\): If \[ 1/2 < \Delta_{\mathbf{A}} < (\sqrt{40} -4)/3 \approx 0.7748, \tag{1} \] then \[ \lvert \mathcal{D}_{\mathbf{A}} \cap [1,N]\rvert \gg N^{\Delta_{\mathbf{A}} + (2\Delta_{\mathbf{A}} -1) \min\left\{ \frac{2\Delta_{\mathbf{A}} +1}{7 - \Delta_{\mathbf{A}}}, \frac{\Delta_{\mathbf{A}}+1}{7-4\Delta_{\mathbf{A}}} \right\}}.\tag{2} \] The main result of the current paper (Theorem 1.4) is to improve the exponent in (2). Under the same assumption (1), the author shows that \[ \lvert \mathcal{D}_{\mathbf{A}} \cap [1,N]\rvert \gg N^{\Delta_{\mathbf{A}} + (2\Delta_{\mathbf{A}} -1)\frac{\Delta_{\mathbf{A}}+2}{\Delta_{\mathbf{A}}+6}}.\tag{3} \] The author compares the ``non-trivial'' parts of the exponents of (2) and (3) (that is, exponent minus \(\Delta_{\mathbf{A}}\)) and shows that the improvement for the non-trivial part of the exponent is up to \(28\)\%. Since \textit{D. Hensley} [J. Number Theory 58, No. 1, 9--45 (1996; Zbl 0858.11039)] conjectured that Zaremba's conjecture already holds with \(A = 2\), the case \(\mathbf{A} = \{1,2\}\) is of particular interest. It is known that \(\Delta_{\{1,2\}} \approx 0.531\) and applying (3) gives the estimate \[ \lvert \mathcal{D}_{\{1,2\}} \cap [1,N]\rvert \gg N^{\Delta_{\{1,2\}} + 0.0241\ldots}. \] This improves the current best known bound \[ \lvert \mathcal{D}_{\{1,2\}} \cap [1,N]\rvert \gg N^{\Delta_{\{1,2\}} + 0.0195\ldots}, \] which is obtained by applying (2).