On asymptotic normality of estimates for correlation functions of stationary Gaussian processes in the space of continuous functions (Q4940452)

For a Gaussian stationary process \(X(t)\) with correlation function \(B(h)\) and spectral density \(f(\lambda)\) such that \(\|f\|_{L_2}<\infty\) the empirical correlogram \(\hat B_T(h)=T^{-1}\int_0^T X(t)X(t+h)dt\) and the empirical process \(Y_T(h)=\sqrt{T}(\hat B_T(h)-B(h))\) are considered as elements of \(C[0,A]\). The pseudometric \(\sqrt{\sigma}\) on \([0,A]\) is defined by the formula \[ \sqrt{\sigma}(h_1,h_2) = [\int_{-\infty}^{+\infty} f^2(\lambda)\sin^{2}(\lambda |h_2-h_1|/2) d\lambda]^{1/4}. \] Theorem: If \(\|f\|_{L_2}<\infty\) and \(\int_0 H_{\sqrt{\sigma}}(\varepsilon) d\varepsilon<\infty,\) then \(Y_T\) converges weakly in \(C[0,A]\) to a Gaussian process \(Y\) with zero mean and \[ EY(h_1)Y(h_2)=4\pi \int_{-\infty}^{+\infty} f^2(\lambda) \cos(\lambda h_1)\cos(\lambda h_2)d\lambda, \] where \(H_{\sqrt{\sigma}}(\varepsilon)\) is the \(\varepsilon\)-entropy of \([0,A]\) with respect to \(\sqrt{\sigma}\). Corollary: If \(\|f\|_{L_2}<\infty\) and \(|f(\lambda)|\) decreases monotonically as \(\lambda\to\infty,\) then the statement of the theorem holds true.