A note on numerical rate of convergence estimates in central limit theorem (Q1175859)

Let \(\{X_n; n\ge 1\}\) be a sequence of i.i.d. r.v.s with mean zero and variance \(1\) and \(F_n\) be the d.f. of the r.v. \((X_1+\ldots+X_n)/\sqrt n\). Denote by \(G\) the d.f. of the standard normal law, and by \(\nu_i\), \(i=0\div 3\), the absolute pseudomoments. For the uniform metric \(\rho\) put \(\rho(n) = \rho(F_n,G)\). The main result of the paper says that \[ \begin{alignedat}{2} \rho(n)&\leq C_ 0(\nu^{r_ 1}_ 0\lor\nu_ 3)/\sqrt n,&\quad C_ 0&=1.8,\\ \rho(n)&\leq C_ 1(\nu^{r_ 2}_ 1\lor\nu_ 3)/\sqrt n,&\quad C_ 1&=4.2,\\ \rho(n)&\leq C_ 2(\nu^{r_ 3}_ 2\lor\nu_ 3)/\sqrt n,&\quad C_ 2&=13.5,\\ \rho(n)&\leq C_ 3(\nu^{r_ 4}_ 3\lor\nu_ 3)/\sqrt n,&\quad C_ 3&=35,\end{alignedat} \] where \(n\geq1\), \(r_i=1-(i-n)^ +/i\), \(i=1\div4\). The author also tabulates the estimates \(\rho(n)\), obtained by a slightly different technique.