Moment bounds and central limit theorem for functions of Gaussian vectors (Q5953869)

The author introduces and proves bounds for moments of functions of Gaussian vectors, which generalizes earlier results by \textit{M. S. Taqqu} [Z. Wahrscheinlichkeitstheorie Verw. Geb. 40, 203-238 (1977; Zbl 0358.60048)] and \textit{Arcones} [Ann. Probab. 15, No. 4, 2243 (1994)]. Also the author proves a central limit theorem for weighted sums of functions of Gaussian vectors. It is adapted to deal with non-stationary processes which arise in some statistical applications. The author presents some statistical applications which can deal with long range dependent Gaussian processes. The main theorem is the following: Let \(\Phi\) be a function defined on \(\mathbf{d}\) such that \(\|\Phi \|< \infty\) and with Hermite rank at least \(\tau.\) Let \(X\) be a Gaussian vector with covariance matrix \(\Gamma\) such that the spectral radius, say \(\rho,\) of \(\Gamma - I_{a}\) satisfies \(\rho < 1/3 - \varepsilon\) for some \(0 < \varepsilon <1/3.\) Then there exists a constant \(c(\varepsilon, \tau, a)\) which depends only on \(\varepsilon, \tau\) and \(a\) such that NEWLINE\[NEWLINE |\mathbb{E}[\Phi (X)] |\leq c(\varepsilon, \tau, a)\|\Phi \|\rho^{\tau/2},NEWLINE\]NEWLINE where NEWLINE\[NEWLINE \|\varphi \|=\{ \mathbb{E}^{0}[\varphi^{2}(X)]\}^{1/2}= \Biggl\{(2\pi)^{-a/2}\int_{\mathbb{R}^{d}} \varphi^{2}e^{-(1/2)x^{\text{T}}x} dx \Biggr\}^{1/2}. NEWLINE\]NEWLINE Here \(x^{\text{T}}\) is the transpose of \(x\) and \(\mathbb{E}^{0}\) denotes the expectation under the multivariate standard normal distribution. The method of moments and the necessity of its use are described and the use of the bounds is obtained in the main theorem. A very simple sufficient condition for asymptotic normality of weighted sums of functions of Gaussian vectors is obtained.