On fine properties of mixtures with respect to concentration of measure and Sobolev type inequalities (Q974769): Difference between revisions

The paper is devoted to the properties of mixtures in the context of Sobolev type functional inequalities. Sections 2 and 3 are devoted to the estimates for the log-Laplace transform \(\alpha_\mu(\lambda):\mathbb{R} \to [0,\infty]\), \(\alpha_\mu(\lambda):=\log \sup_{\|f\|_{Lip} \leq 1} E e^{\lambda(f-E_\mu f)}\), and the related transform \(\beta_\mu(r):=\sup_{\|f\|_{Lip} \leq 1} \mu(|f-E_\mu f|\geq r)\). Here the mixture \(\mu\) on \(\mathbb{R}^n\) is such that \(E_\mu |f|<\infty\) for every Lipschitz function \(f\). In Section 4, the Poincaré and logarithmic Sobolev inequalities for two component mixtures are investigated. A decomposition of variance and entropy allows to reduce the problem to the Poincaré and logarithmic Sobolev inequalities for each component, to discrete inequalities for the Bernoulli mixing law \(p\delta_1 + q\delta_0\), and to the control of a mean-difference term. The Poincaré constant of the two-component mixture can remain bounded as \(\min(p, q)\to 0\), while the logarithmic Sobolev constant may blow up at the speed \(-\log(\min(p, q))\). The results are illustrated by various examples.