The asymptotic distribution of general interpolation arrays for exponential weights (Q1866484)

In this interesting paper, the author examines the relationship between the size of Lagrange interpolation points and Lebesgue constants of Lagrange interpolation. It was \textit{J. Szabados} [J. Approximation Theory 103, No. 2, 357--359 (2000; Zbl 0948.41001)] who initiated this study for Freud weights. The author considers more general even exponential weights. Let \(x_{1n}< x_{2n}<\cdots< x_{n+1,n}\) and \(\{\ell_{k,n}\}^{n+1}_{k=1}\) be the associated fundamental polynomials of Lagrange interpolation, so that \(\ell_{kn}\) is a polynomial of degree \(n\) satisfying \(\ell_{k,n}(x_{j,n})= \delta_{kj}\). Let \(I\) be a real interval and \(w: I\to(0,\infty)\). The Lebesgue constant associated with \(\{x_{jn}\}^{n+ 1}_{j= 1}\) and \(w\) is \[ \Lambda_n= \Biggl\| w\sum^{n+1}_{k=1} {|\ell_{kn}|\over (x_{kn})} \Biggr\|_{L_\infty(I)}. \] It is the norm of the Lagrange interpolation operator acting on a weighted space of continuous functions. For a class of even exponential weights \(w = e^{-Q}\), on a symmetric interval, the author establishes the estimate \[ \max_{1\leq j\leq n+1}| x_{jn}|\leq Ca_n\Biggl(1+ {\log\Lambda_n\over nT(a_n)}\Biggr)^{2/3}, \] where \(C\) is independent of \(n\). Here \(a_n\) is the \(n\)th-Mhaskar-Rakhmanov-Saff number of \(Q\) and \(T(x)= 1+ xQ''(x)/Q'(x)\). In the special case \(Q(x)= | x|^\alpha\), \(\alpha> 1\), \(T\) is constant, and \(a_n= C_1 n^{1/\alpha}\) for some \(C_1> 0\), and one recovers the Szabados result. The author presents the explicit form of this estimate for many other \(Q\), such as \(Q(x)= \exp_k(| x|^\alpha)\) on \((-\infty,\infty)\) or \(Q(x)= \exp_k((1- x^2)^{-\alpha})\) on \((-1, 1)\), where \(\exp_k\) is the \(k\)th iterated exponential.