Some inequalities for Gaussian processes and applications (Q1114220): Difference between revisions

Let \((X_ 1,...,X_ n)\) and \((Y_ 1,...,Y_ n)\) be two Gaussian vectors such that E \(X^ 2_ i=E Y^ 2_ i\), \(1\leq i\leq n\), and E \(X_ iX_ j\leq E Y_ iY_ j\), \(1\leq i,j\leq k\); then the well-known lemma of \textit{D. Slepian} [Bell. Syst. Tech. J. 41, 463-501 (1962)] says that P(\(\cup_{i}[X_ i>\lambda])\geq P(\cup_{i}[Y_ i>\lambda])\) for any \(\lambda\geq 0\), i.e., the less correlated vector has a larger maximum. \textit{X. Fernique} [C. R. Acad. Sci., Paris, Sér. A 278, 363-365 (1974; Zbl 0274.60029)] gave an important variant of this inequality. In this paper the author generalizes these results by considering Gaussian matrices \((X_{ij})\), \((Y_{ij})\), \(1\leq i\leq n\), \(1\leq j\leq m\), and comparing P(\(\cap_{i}\cup_{j}[X_{ij}>\lambda_{ij}])\) and E(\(\min_{i}\max_{j} X_{ij})\) with corresponding expressions obtained by replacing \(X_{ij}\) with \(Y_{ij}\). The proofs are very similar to those of Slepian and Fernique, but the results are very useful. Among several interesting results, derived as simple corollaries, are some well-known ones, such as \textit{A. Dvoretzky}'s theorem [Proc. Int. Symp. Linear Spaces, Jerusalem 1960, 123-160 (1961; Zbl 0119.318)] on the existence of ``almost'' spherical sections for arbitrary convex bodies in \({\mathbb{R}}^ N\).