No label defined (Q4940249)

The author considers a homogeneous and isotropic square continuous random field in \(n\)-dimensional Euclidean space \(E^{n}\) with zero mean, correlation function \(B(u)\) and the spectral function \(\Phi(\lambda)\), \(B(u)=\int\limits_0^{\infty}Y_{n} (\lambda u) d\Phi(\lambda)\), where \(Y_{n}(u)=a_{n} J_{(n-2)/2}(u)/u^{(u-2)/2}\), \(J_{\nu}(u)\) is a Bessel function of the first kind of order \(\nu\), \(a_{n}=2^{(n-2)/2} \Gamma(n/2)\), \(\Phi'(\lambda)=\lambda^{n-1}g(\lambda).\) The unbiased estimator of the correlation function \[ {\hat B}_{R}(u)=|V_{R}|^{-1} \int_{V_{R}(0)}\xi(x)\eta_{u}(x) dx \] is considered. \(\eta_{u}(x)=|S_{u}|^{-1} \int_{S_{u}(x)}\xi(x)m_{u}^{n}(dx)\), \(V_{R}(x)\) and \(S_{R}(x)\) are the \(n\)-dimensional ball and the sphere with radius \(R\) and center at the point \(x\), resp., \(m_{u}^{n}(dx)\) is the Lebesgue measure on the sphere, and \(|\cdot|\) is the Lebesgue measure of a ball resp. a sphere. The author finds conditions for convergence of the finite-dimensional distributions of the process \({\tilde\gamma}_{R}(u)=\sqrt{|V_{R}|} [{\hat B}_{R}(u)-B(u)]\) to the finite dimensional distributions of a Gaussian process \(\gamma(u)\) with known correlation function. For example, this convergence holds true if for \(\delta>0\) \[ (A)\qquad \int\limits_0^{\infty}\lambda^{n-1} g^{2+\delta}(\lambda) d\lambda<\infty. \] The author finds conditions of weak convergence of \({\tilde\gamma}_{R}(u)\) in the space \(C_{[0,\infty]}^0.\) The following assertion is proved. Theorem. Let \(\int\limits_0^{\infty}\lambda^{n+\beta} g^2(\lambda) d\lambda<\infty\) for \(\beta>0\) and condition (A) be satisfied. Then the processes \({\tilde\gamma}_{R}(u)\) and \(\gamma(u)\) generate measures in \(C_{[0,\infty]}^0\) and the measures generated by \({\tilde\gamma}_{R}(u)\) converge weakly to the measure generated by \(\gamma(u).\)