On barycentric interpolation. II. Grünwald-Marcinkiewicz type theorems (Q519949): Difference between revisions

A classical and easy to compute method for approximating a continuous function, \(f\), on the unit interval by a polynomial of degree \(n\) (or fewer) consists of approximating it by the polynomial that interpolates to \(f\) on \(n\) selected points in \([0,1]\), and then adding more interpolating nodes while letting \(n \to \infty\). There are important theorems estimating the growth of the error. These often involve the smoothness of \(f\) and the placement of the nodes (such as, at the zeros of the Chebyshev polynomials). But, no matter how the interpolating nodes are chosen, there exist continuous functions for which the maximal error from \(f\) to the interpolating polynomial diverges to infinity. There are theorems that give lower bounds for the rapidity of divergence for the most egregious errors; and there are theorems that show that for the matrix of nodes such as the zeros of the \(n\)th Chebyshev polynomials for \(n = 1, 2, 3, \dots \), there are functions whose approximating interpolating polynomial do not converge anywhere -- to anything. This work reinforces this last result by showing that a similar divergence result is true even with the zeros of the Chebyshev polynomials replaced with the nodes for barycentric interpolation. For Part I see [\textit{P. Vértesi}, ibid. 147, No. 2, 396--407 (2015; Zbl 1363.41007)].