On the accuracy of the Gaussian approximation (Q845500): Difference between revisions

Let system of \(n\) zero-mean independent random variables \(X_j\) have variances \(\sigma^2_j\) and finite third absolute moments \(\beta_j\). Let \(\sigma^2= \sum \sigma^2_j\), \(\varepsilon_n= \sum \beta_j/\sigma^2\), \(S_n= \sum X_j/n\) and \(N\) be the standard Gaussian variable. It holds: \(\sup|P\,(S_n\leq x)- P\,(N\leq x)|< C\varepsilon_n\). An unimprovable estimate for the proximity in the mean metric between a probability distribution and its zero-bias transformation is presented in the paper. New estimates for the rate of the convergency in Lyapunov's theorem are obtained. The constant \(C\leq 0.6379\) and especially \(C\leq 0.5894\) in the case of identically distributed summands.