The Worpitzky-Pringsheim theorem on continued fractions (Q5948858): Difference between revisions

It has been an open question for a long time to determine when the limit circle case occurs in the Sleszynski - Pringsheim theorem. That is, if \(|b_n|\geq |a_n|+1\) for all \(n\), and thus \[ T_n(\mathbb D)\subseteq T_{n-1}(\mathbb D)\subseteq\dots\subseteq \mathbb D \] where \(\mathbb D\) is the unit disk and \[ T_n(z)={a_1\over b_1} {\quad\atop +} {a_2\over b_2} {\quad\atop +\dots +} {a_n\over b_n+z}, \] when will the radius \(r_n\) of \(T_n(\mathbb D)\) converge to a positive limit \(\widetilde r\)? This is obviously so if all \(a_n=a\) and \(b_n=b\) where the linear fractional transformation \(t(z)=a/(b+z)\) has two fixed points in the closure of \(\mathbb D\). In that case the repulsive fixed point is on the boundary of \(\mathbb D\), and \(\mathbb D\) contains a circle invariant under \(t\). It is also probably so if \(a_n\to a\) and \(b_n\to b\) fast enough, but this has not been proved. What is known is that \(|T_n^{-1}(\infty)|\to 1\) is necessary. In this paper Beardon brings attention to an old and hidden result by W. J. Thron: If \(\partial T_n(\mathbb D)\cap\partial\mathbb D\neq\emptyset\) for all \(n\) (\(\partial A\) denotes the boundary of a set \(A\)), then \[ \widetilde r=1/\left(2\prod_{k=1}^\infty |T_k^{-1}(\infty)|\right). \] He also gives a much simpler and more transparent proof of this result. The disadvantage is that it gives \(\widetilde r\) in terms of \(T_n^{-1}(\infty)\) instead of the quantities \(a_n\) and \(b_n\). Nevertheless, it gives insight in the old problem and a good starting point for further investigations. The main part of Beardon's paper is devoted to the limit point case where \(\widetilde r=0\). He finds useful bounds for the radii \(r_n\) of \(T_n(\mathbb D)\) which in turn give truncation error bounds for the continued fraction. His isometric circles of \(t_n(z)=(a_n/b_n+z)\) and its inverse \(t_n^{-1}\). In particular he proves that if \[ |b_n|\geq 1+\max\{|a_n|, \sqrt{|a_n|}\}\quad\text{for all }n, \] then \[ r_n\leq 1/\left(1+2\sum_{k=1}^n|a_k|^{-1/2}\right)\quad\text{for all }n. \]