Optimal estimates for the rate of strong Gaussian approximate in a Hilbert space (Q357237): Difference between revisions

This paper is a continuation of \textit{F. Götze} and \textit{A. Yu. Zaitsev's} paper [Sib. Math. J. 52, No. 4, 628--638 (2011); translation from Sib. Mat. Zh. 52, No. 4, 796--808 (2011; Zbl 1235.60028)] on the rate of approximation by the sequence of eigenvalues of the covariance operator of summands. More precisely, let \({\mathbf H}\) be the separable Hilbert space that consists of all sequences \(x= (x_1, x_2,\dots)\) for which \(\| x\|^2= x^2_1+ x^2_2+\cdots<\infty\). Let \({\mathbf Z}= (Z_1,Z_2,\dots)\) be a zero mean random vector, taking values in \({\mathbf H}\), such that coordinates of \({\mathbf Z}\) are uncorrelated and \(\sigma^2_m= EZ^2_m\downarrow 0\) \((m\geq 1)\). Independent copies of \({\mathbf Z}\) are to be constructed on the same probability space with a sequence of independent Gaussian random vectors. Then, the rate of approximation substantially depends on \(\sum^\infty_{n=m} \sigma^2_m\). In this paper, the author studies the sequences \(\{\sigma^2_m\}\) which provide estimates that are optimal in order.