Sur les déviations modérées des sommes de variables aléatoires vectorielles indépendantes de même loi. (On moderate deviations of sums of independent and identically distributed vector valued random variables) (Q1189608): Difference between revisions

Let \(X\) be a Borel random variable with values in a real separable Banach space \(B\) such that \(Ef(X)=0\) and \(Ef^ 2(X)<\infty\), \(\forall f\in B'\). Let \(\{X_ k,\;k\geq 1\}\) be a sequence of independent copies of \(X\) and set for each \(n\geq 1\), \(S_ n=X_ 1+\dots +X_ n\). Put \(I(x)=| x|^ 2/2\) or \(+\infty\) according as \(x\) belongs or not to reproducing kernel Hilbert space \((H,|\cdot|)\) associated to the covariance structure of the distribution of \(X\). Suppose that \(\{b_ n,\;n\geq 1\}\) is a sequence of positive numbers such that \(b_ n/\sqrt {n}\to\infty\) and \(b_ n/n\to 0\) as \(n\to \infty\). It is proved that for every closed set \(F\) of \(B\) \[ \limsup_{n\to \infty}nb^{-2}_ n\log P\{S_ n/b_ n\in F\}\leq -\inf_{x\in F}I(x) \] if and only if (i) \(S_ n/b_ n\to 0\), in probability as \(n\to\infty\), (ii) there exists \(M>0\) such that for any \(u>0\) \[ \limsup_{n\to\infty}nb^{-2}_ n\log(nP[\| X\| > ub_ n])\leq -u^ 2/M. \]