Localization of majorizing measures
On a metric space \((T,d)\) with finite diameter one can define certain quantities \(\Theta_N^M\) (\(-\infty<N<M\leq\infty\) integers), related with the set \(\mathcal P\) of probabilities on the Borel sets of \((T,d)\) by \(c_1\mathcal I_N^M \leq \Theta_N^M \leq c_2\mathcal I_N^{M+1}\), for some \(c_1,c_2>0\), where \[ \mathcal I_N^M:= \inf_{\mu\in\mathcal P} \sup_{t\in T} \int_{2q^{-M}}^{2q^{-N}} \sqrt{\log(1/\mu(B(t;\varepsilon))) d\varepsilon}. \] Several inequalities relating the values \(\Theta_N^M\) for different \(N\) and \(M\) allow to establish the following result concerning Gaussian processes \(X=\{X_t,\;t\in T\}\) on a subset \(T\) of a separable Hilbert space \(H\): If \[ \omega(t;\varepsilon):= \text{E}\Bigl[\sup_{s\in T,\;d(s,t)\leq \varepsilon} |X_s-X_t|\Bigr]\quad \text{and}\quad u(\varepsilon):= \text{E}\Bigl[\sup_{s,t\in T,\;d(s,t)\leq \varepsilon} |X_s-X_t|\Bigr] \] are the modulus of continuity of \(X\) at \(t\) and the modulus of uniform continuity, respectively, then for \(N\) large enough, \[ c_1\Theta_N^{\infty} \leq \sup_{t\in T} \omega(t;q^{-N}) + q^{-N}\sqrt{H(T,q^{-N})} \leq u(2q^{-N}) + q^{-N}\sqrt{H(T,q^{-N})} \leq c_2\Theta_N^{\infty} \;. \] Moreover, there is a constant \(c>0\) such that \(\Theta_N^{\infty}\leq c\sup_{M<\infty}\Theta_N^M\). In the above formulae, \(H\) is the metric entropy and \(\text{E}[\sup_{s\in T} Y_t]\) means \(\sup\{\text{E}[\sup_{s\in S} Y_s]:\;S\subseteq T,\;\text{}S\) finite For the entire collection see [Zbl 0995.00013].
This page was built for publication: Localization of majorizing measures
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4801541)
