Stochastic processes and special functions: On the probabilistic origin of some positive kernels associated with classical orthogonal polynomials

Abstract

We describe and investigate a class of Markovian models based on a form of “dynamic occupancy problem” originating in statistical mechanics. The most fundamental of these gives rise to a transition-probability matrix over ( N + 1) discrete states, which proves to have the Hahn polynomials as eigenvectors. The structure of this matrix, which is a convolution of two negative hypergeometric distributions, leads to a factorization into finite-difference sumoperators having forms analogous to the Erdelyi-Kober operators for the continuous variable. These make possible the exact solution of the corresponding eigenvalue problem and hence the spectral representation of the transition matrix. By taking suitable limits, further families of Markov processes can be generated having other classical polynomials as eigenvectors; these, like the polynomials, inherit their properties from the original Hahn system. The Meixner, Jacobi and Laguerre systems arise in this way, having their origin in variants of the basic model. In the last of these cases, the spectral resolution of the continuous transition kernel proves to be identical with Erdelyi's (1938) bilinear formula, which is thus both generalized and given a physical interpretation. Various symmetry and “duality” properties are explored and a number of interesting formulas are obtained as by-products. The use of statistical models to generate kernels, which are thereby guaranteed to be both positive and positive definite, appears to be mathematically fruitful, while the models themselves seem likely to have application to a variety of topics in applied probability.