Modified log-Sobolev inequalities and isoperimetry (Q931300)

Summary: We find sufficient conditions for a probability measure \(\mu\) to satisfy an inequality of the type \[ \int\limits_{\mathbb{R}^{d}} f^{2} F \left(\frac{f^{2}}{\int_{\mathbb{R}^{d}} f^{2} d \mu}\right) d \mu \leq C \int\limits_{\mathbb{R}^{d}} f^{2} c^{\ast} \left(\frac{|\nabla f|}{|f|}\right) d \mu + B \int\limits_{\mathbb{R}^{d}} f^{2} d \mu, \] where \(F\) is concave and \(c\) (a cost function) is convex. We show that under broad assumptions on \(c\) and \(F\) the above inequality holds if for some \(\delta >0\) and \(\varepsilon >0\) one has \[ \int\limits_{0}^{\varepsilon} \Phi \left(\delta c \left[\frac{tF(1/t)}{\mathcal{I}_{\mu} (t)}\right]\right) dt< \infty, \] where \(\mathcal{I}_{\mu}\) is the isoperimetric function of \(\mu\) and \(\Phi = (y F(y)-y)^{\ast}\). In the particular case when \[ \mathcal{I}_{\mu} (t) \geq k t \varphi^{1-1/\alpha} (1/t), \] where \(\varphi\) is a concave function growing not faster than log, \(k>0,\, 1 < \alpha \leq 2\) and \(t \leq 1/2\), we establish a family of tight inequalities interpolating between the \(F\)-Sobolev and modified inequalities of log-Sobolev type. A basic example is given by convex measures satisfying certain integrability assumptions.