On a class of free Lévy laws related to a regression problem (Q2495359)

The family of classical Meixner laws describes the class of orthogonal polynomials \(p_n(x)\) with generating function of the form \(\sum_{n=0}^{\infty} t^n p_n(x) / n!\) \(=\) \(f(t) e^{x u(t)}\). It is well-known that such laws correspond to one of the six classical probability measures: Gaussian, Poisson, Gamma, Pascal (negative binomial), binomial, or a two-parameter hyperbolic secant density, and \textit{J. Wesołowski} [Probab. Math. Stat. 14, No.~1, 33--44 (1993; Zbl 0803.60073)] proved that the five infinite-divisible classical Meixner laws are the laws of stochastic processes with linear regressions and quadratic conditional variances when the conditional variances depend only on the increments of the process. \textit{N. Saitoh} and \textit{H. Yoshida} [Probab. Math. Stat. 21, No. 1, 159--170 (2001; Zbl 1020.46019)] and \textit{M. Anshelevich} [J. Funct. Anal. 201, No. 1, 228--261 (2003; Zbl 1033.46050)] introduced the free Meixner systems of polynomials, and the corresponding probability measures associated with the free Meixner system can be classified into six types: Wigner's semicircle, free Poisson, free Pascal (free negative binomial), free Gamma, a law called pure free Meixner, or a free binomial law. According to Anshelevich [op. cit., Zbl 1033.46050], the \(q\)-Meixner systems of polynomials interpolate between the classical case \(q=1\) and the free case \(q=0\). It was proved in [\textit{W. Bryc} and \textit{J. Wesołowski}, Probab. Theory Relat. Fields 131, No. 3, 415--441 (2005; Zbl 1118.60065)] that the interpolating \(q\)-deformed Meixner laws appear also as the laws of classical stochastic processes with linear regressions and quadratic conditional variances. In this paper, the authors succeed in relating free Meixner laws directly to free probability via regression properties, and they show that the free Meixner laws, arising as the distributions of orthogonal polynomials with constant-coefficient recursions, are the laws of the free pairs of random variables which have linear regressions and quadratic conditional variances when conditioned with respect to their sum. By employing the regression characterization, the authors show that the \(\boxplus\)-infinity divisible members of the free Meixner family are the distributions of the free Lévy processes with linear regressions and quadratic condition variances, which is a free counterpart of the classical result. Moreover, another parallel between the free and classical cases on the level of cumulants is also pointed out. Finally, a free analog of the classical characterization of Wishart matrices by the independence of the sum and the quotient is proved as well.