Bounds for weighted Lebesgue functions for exponential weights (Q5946615)

In this interesting paper, the author obtains matching upper and lower bounds for the Lebesgue function of Lagrange interpolation, associated with zeros of orthogonal polynomials for exponential weights. Let \(I= (c,d)\) be a real interval, with \(-\infty\leq c< 0< d\leq\infty\). Let \(Q: I\to [0,\infty)\) be convex, and \(W= \exp(-Q)\). We call \(W^2\) an exponential weight. If \(W^2\) has all finite power moments, we may define orthonormal polynomials \(\{p_n\}\) satisfying \[ \int_I p_n p_m W^2= \delta_{mn}. \] Let us denote the zeros of \(p_n\) by \[ c< x_{nn}< x_{n-1,n}< x_{n-2,n}<\cdots< x_{1n}< d, \] and let \(\{\ell_{jn}\}^n_{j=1}\) denote the fundamental polynomials of Lagrange interpolation at \(\{x_{jn}\}^n_{j=1}\), so that \(\ell_{jn}(x_{kn})= \delta_{jk}\). The (weighted) Lebesgue function of Lagrange interpolation is \[ \Lambda_n(x)= W(x) \sum^n_{j=1} |\ell_{jn}(x)|W^{-1}(x_{jn}). \] It is the norm of the Lagrange interpolation process as a linear functional at \(x\), when the interpolated functions \(f\) are measured in terms of the norm \(\|fW\|_{L_\infty(\mathbb{R})}\). For a wide class of exponential weights, the author obtains matching upper and lower bounds for \(\Lambda_n\). In particular, he shows that given \(\varepsilon> 0\), \[ \Lambda_n(x)\leq C\log n,\quad x\in (x_{nn}(1- \varepsilon), x_{1n}(1-\varepsilon)). \] More precisely, for this range of \(x\), \[ \Lambda_n(x)\sim (\log n)|p_n W|(x)[|(x- a_{-n})(a_n- x)|^{1/4}]+ |p_n W|(x)\sqrt{a_n+|a_{-n}|}+ 1, \] where \(a_{\pm n}\) are the Mhaskar-Rakhmanov-Saff numbers for \(Q\) and \(\sim\) denotes that the ratio of the two sides is bounded above and below by positive constants independent of \(n\) and \(x\). The author also establishes the exact rate at which \(\Lambda_n(x)\) grows when \(x\) is close to \(x_{nn}\) or \(x_{1n}\), and hence determines the growth of the sup-norm of \(\Lambda_n\) over \(I\). This paper is of essential interest to those working in Lagrange interpolation or orthogonal polynomials.