Weighted maximum over minimum modulus of polynomials, applied to ray sequences of Padé approximants (Q1601822)

It is known that there are functions analytic in the unit disk for which the \((n,n)\) Padé approximants \([n,n]\) do not converge in capacity or measure in any open subset of the disk. The author shows that, nevertheless, a positive result can also be given in this setting: If \(f\) is meromorphic in \(|z|<1\) without pole at 0 and if \(0<\delta <1/8\), then there exists \(\varepsilon= \varepsilon (\delta)\in (0,1)\) and \(n_0\) such that, for \(n\geq n_0\), \[ m_1\Biggl(\Bigl \{r\in[0,1]: \max_{|x|=r} \bigl|f-[n,n] \bigr|(z) \leq\varepsilon^n \Bigr\}\Biggr)\geq 1/8- \delta, \] where \(m_1\) denotes the linear Lebesgue measure. A similar estimate from below holds for \([\lambda n,n]\) and \([n,\lambda n]\) approximants for positive integers \(\lambda\) and with 1/8 replaced by a larger value (depending on \(\lambda)\). For the proof, the author provides a lower bound for the Lebesgue measure of the set \[ \left\{r \in [0,1]: r^{an}{\max_{|t|=1} \bigl|P(t)\bigr|\over \min_{|t|=r} \bigl|P(t)\bigr |}\leq\varepsilon^n \right\}, \] where \(a\geq 0\) and \(\varepsilon >0\). This estimate turns out to be sharp!