A Gaussian correlation inequality and its applications to the existence of small ball constant. (Q2574597)

The author presents a contribution to the Gaussian correlation inequality in the following form: Let \(Y=(X_1,X_2,\dots ,X_n)\) be an \(n\)-dimensional random vector with Gaussian distribution and zero mean. Let \(1\leq k<n\) and write \({Y}_1=(X_1,\dots ,X_k)\) and \({Y}_2=(X_{k+1},\dots ,X_n)\). Then \[ P(\| {Y}\| _\infty \leq x) \leq (1/\varrho ) P(\| {Y}_1\| _\infty \leq x) P(\| {Y}_2\| _\infty \leq x), \] \[ P(\| {Y}\| _\infty \leq x) \geq \max \{\varrho , 2^{-\min \{k,n-k\}}\} P(\| {Y}_1\| _\infty \leq x)P(\| {Y}_2\| _\infty \leq x), \] where \(\varrho \) is given in terms of covariance matrices of \(Y\), \(Y_1\) and \(Y_2\). As a consequence it is deduced that for a fractional Brownian motion \(\{X(t)\mid 0\leq t\leq 1\}\) of the order \(\alpha \in (0,1)\) the limit \(\lim _{x\to 0} x\ln P(\sup \{X(t)\mid 0\leq t\leq 1\}\leq x^\alpha)\) exists.