Improving constants in Esseen-Rozovskii type inequalities for identically distributed random terms (Q6043897)

Let \(X_1,X_2,\ldots,X_n\) be IID random variables with zero mean and unit variance. The authors consider the following Esseen and Rozovskii inequalities that provide upper bounds for the uniform distance \(\Delta_n\) between the distribution of the normalised version of \(X_1+\cdots+X_n\) and a standard Gaussian distribution: for each \(\varepsilon,\gamma>0\), \begin{align*} \Delta_n&\leq\frac{C_E(\varepsilon,\gamma)}{\sqrt{n}}\sup_{0<z\leq\varepsilon\sqrt{n}}\left\{\gamma|EX_1^3\mathbf{1}(|X_1|<z)|+zEX_1^2\mathbf{1}(|X_1|\geq z)\right\}\,,\\ \Delta_n&\leq\frac{C_R(\varepsilon,\gamma)}{\sqrt{n}}\left\{\gamma|EX_1^3\mathbf{1}(|X_1|<\varepsilon\sqrt{n})|+\sup_{0<z\leq\varepsilon\sqrt{n}}zEX_1^2\mathbf{1}(|X_1|\geq z)\right\}\,, \end{align*} where \(C_E(\varepsilon,\gamma)\) and \(C_R(\varepsilon,\gamma)\) are constants depending only on \(\varepsilon\) and \(\gamma\), and \(\varepsilon=\infty\) is permitted in the former inequality but not the latter. The main results of the present paper are upper bounds on \(C_E(\varepsilon,\gamma)\) and \(C_R(\varepsilon,\gamma)\). The authors show that both are monotonically decreasing in \(\gamma\), that \(C_E\) is also monotonically decreasing in \(\varepsilon\), that \(\inf_{\varepsilon,\gamma}C_E(\varepsilon,\gamma)\leq2.19\), and that \(\inf_{\varepsilon,\gamma}C_R(\varepsilon,\gamma)\leq2.25\). Numerical values for these constants and related quantities are also given for various choices of \(\varepsilon\) and \(\gamma\).