On theorems of Gelfond and Selberg concerning integral-valued entire functions (Q1764876): Difference between revisions

For each positive integer \(s\), the authors denote by \(\theta_{s}\) the positive number having the following property. On the one hand, any entire function \(g\) of exponential type \(<\theta_{s}\) satisfying \((d/dz)^{\sigma}g(m)\in\mathbb{Z}\) for any \(\sigma=0,\ldots , s-1\) and any \(m=0,1,2,\dots\) is a polynomial. On the other hand, conversely, for any \(\delta>0\) there exists an entire function \(g\), which is not a polynomial, whose first \(s\) derivatives at all the nonnegative integers are rational integers, and \(g\) has exponential type \(\leq \theta_{s}+\delta\). A well-known theorem of Pólya states that \(\theta_{2}=\log 2\). The authors point out that lower bounds for \(\theta_{s}\) have been obtained, especially by A. O.~Gel'fond and A.~Selberg, and that no reasonable upper bound has been established so far. Their aim is twofold. On the one hand they establish upper bounds for \(\theta_{s}\), including the estimate \(\theta_{s}\leq s\pi/3\). More precisely, a consequence of their results is that for any \(s\geq 1\), there exists a transcendental entire function, having an exponential type bounded by \(s\pi/3+\epsilon\) for any \(\epsilon>0\), whose \(s\) first derivatives at all integers (positive as well as negative) are integers. On the other hand they improve previously known lower bounds for \(\theta_{s}\). An interesting feature of their proof of these refined lower bounds is the use of the Rhin-Viola group-structure arithmetic method [\textit{G. Rhin} and \textit{C. Viola}, Acta Arith. 77, No. 1, 23--56 (1996; Zbl 0864.11037); ibid. 97, No. 3, 269--293 (2001; Zbl 1004.11042)], which enables them to get rid of the Selberg integral.