Berry-Esseen bound for a sample sum from a finite set of independent random variables (Q702393)

Let \(\{ X_1 ,\dots, X_N \} \) be a set of independent random variables, let \(R = (R_1 ,\dots ,R_N )\) be a random vector independent of \(X_1 ,\dots ,X_N \), such that \(P(R = r) = 1/N!\) for any permutation \(r = (r_1 ,\dots ,r_N )\) of the numbers \(1,\dots ,N\), and put \(S_n = \sum_{j = 1}^n {X_{R_j } } \). The distribution of \(S_n /\sqrt {\operatorname{Var}S_n } \) can be approximated by a normal distribution under certain mild conditions. The paper estimates the rate of the normal approximation. Denote \(\mu _k = EX_k \), \(\beta _k = EX_k^2 \), \(\gamma _k = E\left| {X_k } \right| ^3 \), \(\alpha ^2 = N^{ - 1} \sum_{k = 1}^N {\mu _k^2 } \), \(p = {n/N}\), \(q = 1 - p\), \(b = 1 - p\alpha ^2 \). Theorem 1. Assume that \(\sum_{k = 1}^N {\mu _k } = 0\) and \(N^{ - 1} \sum_{k = 1}^N {\beta _k } = 1\). Then there exists an absolute constant \(C\) such that for all \(1 \leq n \leq N\), \[ \sup_x \left| {P(S_n /\sqrt {nb} \leq x) - \Phi (x)} \right| \leq C\min (\Delta _1 ,\Delta _2 + 1/\sqrt {nq} ). \] \(\Delta _1 \) and \(\Delta _2 \) are defined explicitly in terms of \(\mu _k \) and \(\gamma _k \).