Almost sure central limit theorems on the Wiener space (Q988674): Difference between revisions

Let \(\{X_n\}_{n\geq 1}\) be a sequence of real independent identically distributed random variables with \(EX_n=0\) and \(EX^2_n= 1\). We assume that all random variables are defined on some probability space \((\Omega,{\mathcal F}, P)\). Let us define \(S_n= (X_1+\cdots+ X_n)/\sqrt{N}\). \textit{P. Lévy} [Théorie de l'addition des variables aléatoires, Paris: Gauthier-Villars (1937; JFM 63.0490.04)] stated without proof the following almost sure central limit theorem: \[ \lim_{n\to\infty} {1\over\ln n} \sum^n_{k=1} {1\over k} I_{\{S_k\leq x\}}= {1\over 2\pi} \int^x_{-\infty} e^{-u^2/2}\,du, \quad x\in\mathbb{R},\text{ almost surely}. \] The authors give a short history of this theorem, and, in particular, list all mathematicians who contributed to this topic. \textit{I. A. lbragimov} and \textit{M. A. Lifshits} [Theory Probab. Appl. 44, No.~2, 254--272 (1999); translation from Teor. Veroyatn. Primen. 44, No.~2, 328--350 (1999; Zbl 0970.60032)] provided sufficient conditions for a more general result. Let \(\{G_n\}_{n\geq 1}\) be a sequence of real random variables converging in distribution to a random variable \(G_\infty\). For any \(n= 1,2,\dots\) and \(t\in\mathbb{R}\), we set \[ \Delta_n(t)= {1\over\ln n} \sum^n_{k=1} {1\over k}\left(e^{itG_k}- E(e^{itG_\infty})\right). \] If, for all \(t> 0\), \[ \sup_{|t|\leq r} \sum^\infty_{n=1} {E|\Delta_n(t)|^2\over n\ln n}< \infty, \] then, almost surely, for all continuous and bounded functions \(\varphi: \mathbb{R}\to\mathbb{R}\), we have \[ \lim_{n\to\infty} {1\over n} \sum^n_{k=1} {1\over k} \varphi(G_k)= E\varphi(G_\infty).\tag{1} \] One of the main results of the paper is a generalization of the Ibrgaimov and Lifshits theorem for cylindrical random variables \(G_n\) of special form defined in terms of a Gaussian field with \(G_\infty\) distributed according to the standard normal distribution. It is rather hard to formulate their result since many special details of cylindrical measures and Malliavin calculus are involved. As an application of this theorem, the authors prove: Theorem. Let \(\{X_n\}_{n>-\infty}\) be a centered stationary Gaussian sequence with unit variance such that \[ \sum^\infty_{r=-\infty} |E(X_0 X_r)|< \infty. \] Let \(G_\infty\) be a standard normal random variable. Let \(f:\mathbb{R}\to \mathbb{R}\) be a symmetric real function of class \(C^2\) such that \(E|f''(G_\infty)|^4<\infty\). For any \(n= 1,2,\dots\), let \[ G_n= {1\over \sigma_n\sqrt{n}} \sum^n_{k=1} (f(X_k)- Ef(X_k)), \] where \(\sigma_n\) is the normalizing positive constant such that \(E|G_n|^2= 1\). Then (1) holds. The authors provide some other applications of their theorem, in particular, they prove (1) for \(G_n= B^H_n/n^H\), where \(B^H\) is a fractional Brownian motion with Hurst index \(H\in(0, 1)\) and \(G_\infty\) is a standard normal random variable. They also prove similar assertions for sums of Hermite polynomials of increments of a fractional Brownian motion.