On the real accuracy of approximation in the central limit theorem (Q642086)

Let \(X_1,X_2,\dots\) be i.i.d. random variables with mean zero and unit variance and let \(F_n\) denotes the distribution function of \((X_1+\cdots+ X_n)/\sqrt{n}\). A question how many summands one should take to approximate \(F_n\) by the standard normal distribution function with a given precision (\(10^{-3}\) in the paper) is studied. Begining from the Berry-Esséen estimate through some recent author's estimates up to new author's asymptotic expansions in CLT involving Chebyshev-Hermite polynomials and assuming finite sixth moments it is shown that for the same distribution this number of summands varies from hundreds of thousands through several thousands up to a few tens.