Limit results for maxima in non-stationary multivariate Gaussian sequences (Q1105266): Difference between revisions

Let \(\{Y_ k\), \(k\geq 1\}\) be a p-dimensional standard Gaussian sequence, so that \(Y_{ki}\) is standard Gaussian and suppose \(r_{i,j}(k,\ell)=cov(Y_{ki},Y_{\ell j})\) is the autocorrelation. Consider sequences \(\{O_ k=(O_{k1},...,O_{kp})\), \(k\geq 1\}\) such that \[ \lambda_ n=\min \{O_{ki},\quad k\leq n,\quad i\leq p\}\to \infty \quad as\quad n\to \infty \] and \(\sum_{k\leq n}\sum_{i\leq p}(1- \Phi (O_{ki}))\) is bounded in n where \(\Phi\) is the standard normal distribution. Let \(\delta_ m=\sup \{| r_{ij}(k,\ell)|:| \ell -k| \geq m\), i,j\(\leq p\}\). Then if \(\delta_ 1<1\) and \(\delta_ n \log n\to \infty\) it is shown that \[ P\{Y_ k\leq O_ k,\quad k\leq n\}-\prod^{n}_{k=1}P\{Y_ k\leq O_ k\}\to 0\quad as\quad n\to \infty. \] That is, \(\{Y_ k\}\) behaves asymptotically with respect to \(\{O_ k\}\) in the same way as an independent sequence. If \(\sup \{| r_{ij}(k,k)|\), \(i\neq j,k>1\}<1\), then the components of \(\{Y_ k\}\) also behave asymptotically independently with respect to \(\{O_ k\}\).