scientific article; zbMATH DE number 1323471
The distribution of a centered Gaussian random function \(X= (X_t)_{t\in T}\) is completely determined by its covariance structure \(({\mathbf E}X_sX_t)_{s,t\in T}\). It is therefore natural to analyze properties of \(X\) by using the geometry of \(T\) for the induced pseudo-metric \(d_X(s,t)= ({\mathbf E}(X_s- X_t)^2)^{1/2}\). The equivalence between the following statements was proved by \textit{X. Fernique} [in: École d'été Probab. Saint-Flour IV-1974. Lect. Notes Math. 480, 1-96 (1975; Zbl 0331.60025)] and \textit{M. Talagrand} [Acta Math. 159, No. 1/2, 99-149 (1987; Zbl 0712.60044) and Ann. Probab. 24, No. 3, 1049-1103 (1996; Zbl 0867.60017)]: (i) It holds \(\sup_{F\subseteq T_{\text{finite}}}{\mathbf E} \sup_{t\in F} X_t< \infty\). (ii) There exists a Borel probability measure \(\mu\) on \(T\) such that \[ {\mathcal I}\Theta_\mu:= \sup_{t\in T} \int^\infty_0 \sqrt{{1\over \mu(B(t,\varepsilon))}} d\varepsilon< \infty \] with \(B(t,\varepsilon)= \{s\in T; d_X(s,t)\leq\varepsilon\}\). (iii) There exists a Borel probability measure \(\mu\) on \(T\) such that \[ M\Theta_{{\mathcal A},\mu}:= \sup_{t\in T} \sum_{j> N_0} q^{-j} \sqrt{{1\over \mu(A_j(t))}}< \infty \] for a certain increasing sequence \({\mathcal A}= ({\mathcal A}_{N_0+ 1},{\mathcal A}_{N_0+ 2},\dots)\) of measurable partitions of \(T\) (\(A_j(t)\) is the set of \({\mathcal A}_j\) which contains \(t\), \(q\) is a sufficiently large fixed number). The measures in (ii) and (iii) are called majorizing measures. We introduce local variants \({\mathcal I}\Theta^{N_2}_{N_1,\mu}\) and \(M\Theta^{N_2}_{N_1,{\mathcal A},\mu}\) and some other related quantities (the classical quantities are included for \(N_1= N_0\) and \(N_2=\infty\)). We prove the equivalence between all these quantities in a pure geometric way (on an arbitrary metric space \((T,d)\)). For \(N_2<\infty\) we give inductively a construction for our local quantities and we show how the case \(N_2=\infty\) may be approximated by \(N_2<\infty\). In the last chapter, we consider relations between our local quantities (on \((T,d_X)\)) and the Gaussian random function \(X\).
- Majorizing measures and proportional subsets of bounded orthonormal systems
- Constructions of majorizing measures, Bernoulli processes and cotype
- Majorizing Measures and Ultrametric Spaces
- Localization of majorizing measures
- The majorizing measure approach to sample boundedness
- A theorem on majorizing measures
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 Q4255338)
