The Gaussian approximation for multi-color generalized Friedman's urn model (Q1042973)

Consider the generalized Friedman's urn model and the process \((Y_n,N_n)\), where \(Y_n= (Y_{n1},\dots,Y_{nd})\) and \(Y_{nk}\) denotes the number of type \(k\) particles in the urn after \(n\) stages while \(N_n= (N_{n1},\dots,N_{nd})\) and \(N_{nk}\) stands for the number of times a type \(k\) particle was drawn in the first \(n\) stages. The process is defined recursively via addition rules matrices \(D_n\) describing (a random) rule of adding particles to the urn after the \(n\)th draw and defining generating matrices \(H_n\). If \(D_n\) are i.i.d. random matrices the model is said to be homogeneous. It is proved that both for homogeneous and non-homogeneous case, under some mild conditions on \(H_n\), the process \((Y_n,N_n)\) can be approximated almost surely by a Gaussian process which is a solution of a stochastic differential equation. This solution is shown to exist and to be unique. Next, the main results are used to derive a functional central limit theorem, the asymptotic normality and the law of the iterated logarithm for \(Y_n\) and \(N_n\). Asymptotic properties of the generalized randomized play-the-winner rule are obtained as well. The paper is related to \textit{Z. D. Bai}, \textit{F. Hu} and \textit{L.-X. Zhang} [Ann. Appl. Probab. 12, No. 4, 1149--1173 (2002; Zbl 1014.60025)] and to a series of articles by the same authors. The main results are proved by a different technique than applied in earlier papers.