Upper bounds for the spectral radius of the \(n \times n\) Hilbert matrix (Q816313)

The \(n\)-by-\(n\) Hilbert matrix has the entries \((i+j+1)^{-1}\), \(i,j=1,\ldots,n\). Hilbert has shown that its operator norm \(\| A_n\| \), which in this case coincides with the spectral radius, is bounded by \(2\pi\), and Schur gave the bound \(\pi\), optimal in the sense that there is no better bound which is independent of \(n\). There are several bounds dependent on \(n\), e.g. by \textit{H. Frazer} [J. Lond. Math. Soc. 21, 7--9 (1946; Zbl 0060.14903)], who showed \(\| A_n\| \leq n \sin(\pi/n)\). Later a better one by \textit{J. W. S. Cassels} [ibid. 23, 285--290 (1949; Zbl 0034.18501)] is described, namely \(\| A_n\| \leq 2 \arctan(\sqrt{2n})\). Here several improvements are given. Particularly this one: \(\| A_n\| \leq 2w_n \arcsin(1/w_n)\) where \(w_n= 2((n!)^2/(2n)!)^{1/2n}\), which is at least for large \(n\) better than the bound by Cassels. The method used to derive this bound is writing the Hilbert matrix as an integral operator with a positive kernel function and use then a Wielandt-type minmax principle for the spectral radius with special test functions.