Universality and Fourth Moment Theorem for homogeneous sums. Orthogonal polynomials and apolarity [article]

Rosaria Simone
2017 arXiv   pre-print
The aim of the dissertation is threefold: the first two parts are devoted to explore the Fourth Moment Theorem and universality properties for homogeneous sums, while the last part approaches the classical theory of orthogonal polynomials via the invariant theory of binary forms. Specifically, Part I deals with the Lindberg method of influence functions in free probability spaces. By providing a general multidimensional invariance principle for homogeneous sums in freely independent variables,
more » ... class of universal laws for semicircular and free Poisson approximations will be derived that satisfies also the so-called Fourth Moment phenomenon. It is known that the fourth moment phenomenon for homogeneous sums applies, for instance, when X is Gaussian, Poisson, semicircular or free Poisson distributed. Is there a characterizing property enabling this phenomenon, or does it occur accidentally? The goal of Part II is the characterization of random variables X such that homogeneous sums based on i.i.d. copies of X verify the Fourth Moment Theorem, both for classical and free probability spaces. The main results will be the determination of a condition on the fourth cumulant of X that is sufficient for the Fourth Moment Theorem to hold. The optimality of such condition is also discussed. Part III has a purely algebraic flavour, and aims at recasting orthogonality of polynomials within the invariant theory of binary forms through the symbolic methods introduced by Kung and Rota. Then, by focusing on apolarity. a unifying representation for (generalized) sequences of orthogonal polynomials (in any number of variables) is achieved and explicit formulae for the moments of the random discriminants are derived. In this specific case of cumulants, a focus is presented within the combinatorial approach to stochastic integration in terms of diagonal measures.
arXiv:1705.03294v1 fatcat:qhwbeaepbncn7lywmayehi7ghi