Dirac masses determination with orthogonal polynomials and \(\epsilon\)- algorithm. Application to totally monotonic sequences (Q1201284)

Suppose a distribution \(c\) (or a measure \(\mu\)) has support on \([-1,1]\) and consists of an absolutely continuous part and a discrete part with Dirac masses at points \(a_ 1,\dots,a_ q\in[-1,1]\). The author proposes some techniques to locate the position of these Dirac masses and the size of their contribution. A first technique uses Chebyshev polynomials and consists of evaluating \(\sum^ n_{k=0}c(T_ k)T_ k(a)/\sum^ n_{k=0}T^ 2_ k(a)\). This function (for large \(n\)) typically has peaks at the points \(a_ i\) and the size of the peaks determines the size of the contribution of the Dirac measure. A second technique uses Christoffel functions for the orthogonal polynomials associated to \(c\), and these Christoffel functions also have peaks at the point \(a_ i\). If only one Dirac measure is present, then it is shown that the \(\varepsilon\)-algorithm of Wynn also gives the desired size of the Dirac measure. Some numerical examples are given and an application to the calculation of the limit of a totally monotonic sequence.