Multiple Wiener-Itô integrals. With applications to limit theorems (Q388754)

This book is a slightly augmented, second edition, coming 33 years after the first one [Zbl 0451.60002]. Notably, the author has added some explanations of basic notions, and made the book more self-contained. The central purpose of this book is to describe, explain and establish a generic convergence result for the laws of conveniently normalized partial sums of strongly correlated stationary Gaussian random variables or fields. Such a result does not pertain to any central limit theorem class; it is of a different nature. The main theorem of the book, about which the whole text is written, the so-named Theorem 8.2, appeared first in an article by \textit{R. L. Dobrushin} and the author [Z. Wahrscheinlichkeitstheor. Verw. Geb. 50, 27--52 (1979; Zbl 0397.60034)]. It is about a centred stationary Gaussian field \(\{X_n: n\in\mathbb{Z}^d\}\) having a correlation function \(r_n:= \operatorname{E}[X_j x_{j+n}]\) given by \(r_0=1\) and with \(r_n= |n|^{-\alpha} a({n\over|n|}) L(|n|)\) for \(n\in\mathbb{Z}^d\setminus\{0\}\), where \(0<\alpha< d\), \(a(\cdot)\) is a symmetrical continuous function on \(\mathbb{S}^{d-1}\), and \(L(\cdot)\) is a locally bounded slowly varying function on \([1,\infty[\). Denote by \(G\) the spectral measure \[ r_n= \int_{[-\pi,\pi[^d} e^{in\cdot x}G(dx),\quad n\in\mathbb{Z}^d. \] Consider then an integer \(k< d/\alpha\), denote by \(H_k\) the \(k\)th Hermite polynomial with leading coefficient 1, set \(\xi_n:= H_k(X_n)\), and set \[ Z^n_n:= N^{k\alpha/2-d}\times L(N)^{-k/2}\times \sum_{j\in C^N_n} \xi_j \] for \(n\in\mathbb{Z}^d\) and \(N\in\mathbb{N}^*\). Here, \(C^N_n\) denotes the hypercube of \(\mathbb{Z}^d\) based at \(n\) and having volume \(N^d\). Then the main Theorem 8.2 of the book asserts that the finite-dimensional distributions of the normalized partial sums \(Z^N_n\) converge, as \(N\to\infty\), towards those of a random field \[ Z^*_n:= \int_{\mathbb{R}^{d}} \varphi_n(x_1+\cdots+ x_k)\,Z_{G_0}(dx_1)\cdots Z_{G_0} (dx_k). \] Here, \(\varphi\) denotes the Fourier transform of \(\mathbf{1}_{C^1_n}\), and \(Z_{G_0}\) is the random spectral measure associated with the spectral measure \(G_0\), this is \(\alpha\)-homogenous: \(G_0(A)= t^{-\alpha} G_0(tA)\), and is obtained as the limit of \(G_N(A):= {N^\alpha\over L(N)} G(A/N)\), for any bounded Borelian subset \(A\) of \(\mathbb{R}^d\). To establish this theorem, once the existence of \(G_0\) is proved, a canonical representation \[ \xi_n= \int_{\mathbb{R}^{dk}} e^{in(x_1+\cdots+ x_k)} Z_G(dx_1)\cdots Z_G(dx_k) \] is deduced from the Itō formula relating to Wiener-Itō integrals. This is then easily handled, to deduce that \(Z^N_n\) has the law of \[ \int f_N(y_1,\dots, y_k) \varphi_n(y_1+\cdots+ y_k)\, Z_{G_N}(dx_1)\cdots Z_{G_N}(dx_k), \] where the function \(f_N(y_1,\dots, y_k)\) is explicit and goes to 1, uniformly in any compact set, as \(N\to\infty\). It remains to justify that the convergences \(f_N\to 1\) and \(G_N\to G_0\) can be inserted in the above integral; which is a technical and somewhat delicate part. There is a continuous version of this theorem, which is, however, not stated precisely in the book. Theorem 8.5 regarding the case \(k\geq d/\alpha\) is also stated and commented in the book, but only with hints of proof. In order to yield an almost self-contained treatment, the author presents all needed material in details, within the first chapters: stationary fields, spectral measures, random spectral measures, Bochner theorem, Wick polynomials, multiple Wiener-Itō integrals and their Itō formula, Feynman diagrams, and self-similar fields. Part of this material is also presented in a version with continuous index action, using the Schwartz space as set of test-functions.