A remark on the growth of the denominators of convergents (Q2494364): Difference between revisions

Consider for each real, irrational number \(x\) the convergents \(p_n/q_n \in {\mathbb{Q}}\) of the continued fraction expansion of \(x\). Define the quantities \[ \beta_*(x) = \liminf_{n \rightarrow \infty}{ { \log q_n } \over {n}} \text{ and } \beta^*(x) = \limsup_{n \rightarrow \infty}{ { \log q_n } \over {n}}. \] \textit{P. Lévy} showed that for almost all real numbers \(x\), \(\beta_*(x) = \beta^*(x) = {\pi^2 \over{12 \log 2}}\) [Bull. Soc. Math. Fr. 57, 178--194 (1929; JFM 55.0916.02)]. It is also well known that \(\beta_*(x) \geq \log {{\sqrt{5} +1} \over {2}}\) for any irrational number \(x\). The present paper is concerned with the sets, \[ E(\lambda_*, \lambda^*) = \big\{x \in [0,1) : \beta_*(x) = \lambda_* \text{ and } \beta^*(x) = \lambda^* \big\}, \] where \(\log {{\sqrt{5} +1} \over {2}} \leq \lambda_* \leq \lambda^* < \infty\). It is shown that the Hausdorff dimension of \(E(\lambda_*, \lambda^*)\) is bounded from below by \({{\lambda_* - \log {{\sqrt{5}+1} \over {2}}} \over {2 \lambda^*}}\). While this result is not optimal, as is easily seen by letting \(\lambda_* = \lambda^* = {\pi^2 \over{12 \log 2}}\) and comparing with Lévy's result, it improves substantially on a previous result by \textit{C. Baxa} [Acta Math. Hung. 83, No. 1--2, 125--130 (1999; Zbl 0928.11007)], who showed that the sets \(E(\lambda_*, \lambda^*)\) are uncountable. To prove the result, the author uses the classical recurrence formulas for the convergents of an irrational number to construct a particularly nice subset of \(E(\lambda_*, \lambda^*)\). Subsequently, a probability measure supported on this subset is constructed, and the main result follows by the mass distribution principle.