Rate of convergence in a theorem of Heyde (Q449020): Difference between revisions

Let \(\{X,X_{n},n\geq 1\}\) be a sequence of i.i.d. random variables with \(\operatorname{E}X=0\) and \(0<\operatorname{E}X^2=\sigma^2<\infty\), set \(S_{n}=X_1+\dotsb+X_{n}\), and put \(\lambda(\varepsilon)=\sum_{n\geq 1}\operatorname{P}(|S_{n}|\geq n\varepsilon)\), \(\varepsilon>0\). A classical result, due to \textit{P. L. Hsu} and \textit{H. Robbins} [Proc. Nat. Acad. Sci. USA 33, 25--31 (1947; Zbl 0030.20101)], asserts that \(\lambda(\varepsilon)<\infty\) for all \(\varepsilon>0\). Later, \textit{C. C. Heyde} [J. Appl. Probab. 12, 173--175 (1975; Zbl 0305.60008)] proved that \(\lim_{\varepsilon\to 0}\varepsilon^2 \lambda(\varepsilon)=\sigma^2\), and \textit{O. I. Klesov} [Theory Probab. Math. Stat. 49, 83--87 (1994); translation from Teor. Jmovirn. Mat. Stat. 49, 119--125 (1993; Zbl 0861.60054)], under the additional assumption \(\operatorname{E}|X|^3<\infty\), showed that \(\varepsilon^2\lambda(\varepsilon)-\sigma^2=o(\varepsilon^{1/2})\) as \(\varepsilon \to 0\). The present authors obtain rates of convergence of \(\varepsilon^2 \lambda (\varepsilon)\) to \(\sigma^2\), as \(\varepsilon \to 0\), under only \(\operatorname{E}X^2<\infty\), or under \(\operatorname{E}X^2<\infty\) and the assumption that \(\operatorname{E}[X^2 I|X|>t]=O(t^{-\delta})\) for some \(\delta>0\) as \(t\to \infty\). Reviewer's remarks: The presentation of the paper is not satisfactory, and the English is rather bad. After each of the Theorems A, B and C, several lines are wrongly typeset in italics.