Approximation of the distributions of sums of a random number of independent random variables by mixtures of Gaussian measures (Q1175849)

Let \(\xi_n\), \(n\geq1\), be independent random variables such that \(\sigma^2\xi_n<\infty\), \(n\geq1\). Let us put \(S_n=\xi_1+\ldots+\xi_n\), \(A_n=\hbox {E} S_ n\), \(B^2_n=\sigma^2S_n\). Let \(\nu\) be a positive integer-valued random variable independent of \(\xi_n\), \(n\ge 1\). Define \[ B^2=\hbox{E} B^2_\nu,\ G(x)=P(A_\nu<x)\text{ and } \Delta_n=\sup_x| P(S_n<x) - \Phi((x-A_n)/B_n)|, \] where \(\Phi\) denotes the standard normal distribution function. The main result gives the following Theorem: For every \(c>0\) \[ \sup_x| P(S_\nu<x)-G(x)*\Phi(c^{- 1}x)|\leq \hbox{E} \Delta_ \nu+1.6199c^{-2} \hbox{E} | B^2_\nu-c^2|. \] The same topic for sums of i.i.d. \(k\)-dimensional random variables is also considered.