scientific article; zbMATH DE number 2187776
Let \(Z\geq 0\) be a random variable such that \(0<P(0\leq Z<x)<1\) for any \(x>0.\) For \(u\geq 0,\) put \(\Lambda (u)=Ee^{-uZ},\) \(m(u)=-(\log \Lambda (u))^{\prime },\) \(\sigma ^{2}(u)=(\log \Lambda (u))^{\prime \prime },\) \(Q(u)=-um(u)-\log \Lambda (u).\) For \(0<r\leq EZ,\) the author obtains the estimate \[ \frac{1}{2}\exp (-Q(h)-2a(1+\sqrt{1+2Q(h)/a}))\leq P(Z\leq r)\leq \exp (-Q(h)), \] where \(a=\sup_{u>0}u^{2}\sigma ^{2}(u)/Q(u)\) and \(h\) is the unique solution of the equation \(m(h)=r.\) A similar estimate for \(P(r-s<Z\leq r)\) is derived when \(s\leq r\leq EZ.\) These results are then applied with \(Z=\| X-\alpha\| ,\) where \(X\) is a centered Gaussian random vector taking values in a separable Hilbert space \(H\) and \(\alpha \in H.\)
- Small ball probabilities around random centers of Gaussian measures and applications to quantization
- A Gaussian small deviation inequality for convex functions
- On small deviations conjecture
- Remarks on a link between the Laplace transform and distribution function of a nonnegative random variable
- An optimal relation between small deviations and large deviations
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 Q5460976)