scientific article; zbMATH DE number 219291

From MaRDI portal
Publication:4695441





[For the entire collection see Zbl 0747.00049.] For independent \(R\times H\) valued pairs \(\{Y_ i,{\mathcal Y}_ i\}\), where \(H\) is a real separable Hilbert space, and a bounded r.v. \(D\) define \[ L_ n={1\over\sqrt n}\sum^ n_{i=1}Y_ i+{D\over\sqrt n}\left\|{1\over\sqrt n}\sum^ n_{i=1}{\mathcal Y}_ i\right\|^ 2_ H. \] Suppose \(EY_ i=0\), \(EY^ 2_ i=1\) and assume there are \(0<a\leq 1/2\), \(b>0\) such that \(E \exp\{b| Y|^ a\}+E\exp\{b\|{\mathcal Y}\|^ a\}<\infty\). Then there are constants \(A>0\), \(c\geq 0\) such that \[ | P(L_ n>r)/P(G>r)-1|\leq c(1+r)^ 3/\sqrt n \] for all \(0\leq r\leq An^{a/(4-2a)}\), where \(G\) denotes the standard normal \(N(0,1)\) r.v. Estimate \[ \sup_ r| P(L_ n>r)- P(G>r)|\leq c/\sqrt n \] is given under the assumption of finiteness of third moments. Large deviation estimates for linear combinations of real i.i.d. random variables are given.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4695441)