Modified log-Sobolev inequalities, Beckner inequalities and moment estimates (Q2076297)

In a probability space \((\mathcal{X},\mathcal{B},\mu)\) the Becker inequality with parameter \(p\in(1,2]\) holds if a constant \(\alpha_p>0\) exists such that \[ \alpha_p\,\mu(f^p)-\mu^p(f)\le\frac{p}{2}\,\mathcal{E}(f,f^{p-1})\,, \] where \(\mathcal{E}\) is a Dirichlet form and \(\mu(f)\) denotes the expectation fo \(f\). If the entropy is defined by \[ \mathrm{Ent}_{\mu}(f):=\mu(f\,\log f)-\mu(f)\,\log \mu(f)\,, \] the modified log-Sobolev inequality is satisfied if there is a constant \(\rho_0>0\) such that \[ \rho_0\,\mathrm{Ent}_{\mu}(f)\le\mathcal{E}(f,\log f)\,. \] The authors prove the following theorem (restated in this reviewer's words and in simplified form) Theorem. The following statements are equivalent:\par (a) The modified log-Sobolev inequality holds with some constant \(\rho_0>0\);\par (b) the Becker inequality holds for every \(p\in(1,2]\) with some \(\alpha_p\) bounded away from zero. Only one of the two implications is known in the literature. The authors also derive Sobolev-type moment estimations. Their results are illustrated by applications to estimates for concentration of measure.