A numerical comparison of seven grids for polynomial interpolation on the interval (Q1963113)

The quest for an optimal set of grid points for polynomial interpolation on a real interval has been greatly complicated by the fact that there are many different measures of ``optimal''. In this article, six definitions of optimality are given: in the sense of Cauchy, Lebesgue, Fejér and Erdős, and furthermore the cardinal-function-minimizing optimality and the Vandermonde-function-minimizing optimality. The goal of Cauchy optimality is to minimize the classical function \(w(x)= \prod^n_{k=1} (x- x_k)\). The other measures depend upon the Lagrange polynomials here called cardinal functions. In this paper four of these optimality measures are computed for seven representative sets of grid points. The first grid consists of the roots of the Chebyshev polynomials. The second, third and fourth grids are Lobatto grids for Gegenbauer polynomials. The fifth and sixth sets of grid points will be the Lebesgue-optimal and Erdős-optimal grids numerically computed by \textit{Q. Chen} and \textit{I. Babuska} [Comput. Methods Appl. Mech. Eng. 128, No. 3-4, 405-417 (1995; Zbl 0862.65006); ibid. 147, No. 1, 89-94 (1996; Zbl 0877.65004); Commun. Numer. Methods Eng. 12, No. 8, 483-495 (1996; Zbl 0860.65001)], and the last grid is the expanded Chebyshev grid proposed by \textit{L. Brutman} [SIAM J. Numer. Anal. 15, 694-704 (1978; Zbl 0391.41002)]. It is shown that the grid which minimizes the Lebesgue constant is the best because it performs fairly well all of the considered four criteria.