Majorization, exponential inequalities and almost sure behavior of vector-valued random variables (Q1176367)

Let \(\{X_ i\}\) be independent Banach valued random vectors with \(E\| X_ i\|<\infty\), let \(S_ n=\sum^ n_{i=1} X_ i\), let \(\{X_ i'\}\) be an independent copy of \(\{X_ i\}\), and \(R_ i\) independent Rademacher independent of the previous variables. It is shown that \(\| S_ n\|-E\| S_ n\|\) is dominated by \(\sum^ n_{i=1} R_ i(\| X_ i\|+\| X_ i'\|+2E\| X_ i\|)\), which is in turn dominated by \(\sum^ n_{i=1} R_ i\| X_ i\|\). This observation is used to derive moment estimates, exponential inequalities, strong laws of large numbers (an extension of Nagaev's strong law of large numbers is given) and bounded laws of the iterated logarithm for partial sums of Banach valued random vectors from the corresponding results for partial sums of independent real valued random variables. The reduction argument is related to arguments of \textit{V. V. Yurinskij} [Theory Probab. Appl. 19, 154-155 (1974); translation from Teor. Veroyatn. Primen. 19, 152-154 (1974; Zbl 0323.60029)]. A sample result, related to work of \textit{J. Kuelbs} [Illinois J. Math. 21, 784-800 (1977; Zbl 0392.60010)]: If \(\limsup_{n\to\infty} a^{-1}_ n\sum^ n_{i=1} R_ i\| X_ i\|<\infty\) a.s., then \(\limsup a^{-1}_ n\| S_ n\|<\infty\) a.s. if and only if the sequence \(\{a^{- 1}_ n\| S_ n\|: n\in N\}\) is stochastically bounded. The method in this paper is elegant but has of course a limitation: finer results on the law of large numbers, exponential inequalities, etc. depending on weak moments instead of moments of the norm are unattainable.