A square bias transformation of probability distributions: some properties and applications (Q382185)

In earlier papers of other authors, the so-called \(X\)-size biased distribution of a random variable \(X^{(s)}\) [\textit{L. Goldstein} and \textit{Y. Rinott}, J. Appl. Probab. 33, No. 1, 1--17 (1996; Zbl 0845.60023)], based on the \(X\)-size biased transformation, and the \(X\)-zero biased distribution of a random variable \(X^{(z)}\) [\textit{L. Goldstein} and \textit{G. Reinert}, Ann. Appl. Probab. 7, No. 4, 935--952 (1997; Zbl 0903.60019)] were introduced. A third distribution, called \(X\)-square biased distribution of a random variable \(X^*\), is the subject of the present paper. All transformations are based on the characteristic function of a random variable \(X\). In the paper, several properties of the newly introduced random variable \(X^*\) are considered. The main theorem contains an upper and lower estimate of \(L_1(X; X^*)\) (in the \(L_1\)-metric for the random variables \(X\) and \(X^*\)) by help of the third absolute moment of \(X\). First estimates of this kind for \(L_1(X; X^{(z)})\) were proved independently by \textit{L. Goldstein} [Ann. Probab. 38, No. 4, 1672--1689 (2010; Zbl 1195.60034)] and \textit{I. S. Tyurin} [Dokl. Math. 80, No. 3, 840-843 (2009); translation from Dokl. Akad. Nauk, Ross. Akad. Nauk 429, No. 3, 312--316 (2009; Zbl 1201.60018) and \url{arxiv:0912.0726}]. The main theorem gives an opportunity to construct asymptotically exact (as \(t\to 0\)) bounds for characteristic functions (cf. Corollary 1 and Corollary 2 in the present paper).