Universality Properties of Gaussian Quadrature, The Derivative Rule, and a Novel Approach to Stieltjes Inversion

Abstract

Stieltjes, Perron, and Markov in analysis of the moment problem, for absolutely continuous measures, constructed the underlying measure as the discontinuity across the cut of a Cauchy representation of an otherwise real-analytic function. Introduction of Gaussian quadratures suggests more direct approaches. The real, positive, and continuous weight function, rho(x), is constructed directly from interpolation of the zeros, x[k], and weights, w[k], of the polynomials orthogonal for that measure. This novel numerical approach rests on a simple theorem: namely, rho(x[k]), evaluated at the quadrature point x[k], is approximated by w[k]/(dx[k]/dk), with k now being assumed to be a continuous real variable. This derivative rule is shown to be closely related to a simple type of universal behavior, relating the clock-rule spacings of the x[k] to the derivatives dx[k]/dk which, once established, allows an immediate derivation of the derivative rule, via known and well understood clock-rule results for Nevai-Blumenthal class polynomials. This framework also leads to a statement of a universality for quadrature weights and weight functions. This is implicit in statements of clock-rule universality for the N-B class, but not made explicit. This new universality is illustrated for several classical orthogonal polynomials, and the non-classical Pollaczek polynomials. The derivative rule is then applied to the problem of numerical Stieltjes inversion, for both finite and unbounded intervals giving exponential convergence. Exponential convergence stands in stark contrast to the power law convergence of historically well-known histogram techniques A final section traces the origin of the derivative rule in developing novel approaches to the use of discrete spectral resolutions of Schr\"odinger operators.