On a majorization inequality for sums of independent random vectors (Q1314685): Difference between revisions

The author gives an improvement of the basic result of \textit{E. Berger} [Ann. Probab. 19, No. 3, 1206-1226 (1991; Zbl 0757.60002)]. Let \(X_ 1, \dots, X_ n\) be independent random vectors in a separable Banach space \((B, \| \cdot \|)\) with \(E \| X_ i \| < \infty\), \(i = 1, \dots, n\), and set \(S_ n = X_ 1 + \cdots + X_ n\). Let \(e_ 1, \dots, e_ n\) be independent random variables with \(P(e_ i = 1) = P(e_ i = -1) = 1/2\), \(i = 1, \dots, n\), independent also of \(X_ 1, \dots, X_ n\). For any convex function \(g:\mathbb{R} \to \mathbb{R}\) and any \(n\)- tuple \((a_ 1, \dots, a_ n) \in B^ n\), one has \[ E \Bigl[ g \bigl( \| S_ n \| - E \| S_ n \| \bigr) \Bigr] \leq E \left[ g \left( 2 \sum^ n_{i = 1} e_ i \| X_ i - a_ i \| \right) \right]. \]