Logarithmic Sobolev inequality for symmetric forms (Q1585604)

The author considers a logarithmic Sobolev inequality on a measurable probability space \((E, \mathcal E, \pi)\) whose diagonal is measurable, i.e. \(\{(x, x)\mid x \in E\} \in \mathcal E \times \mathcal E\). Let \(L^2(\pi)\) denote the usual real \(L^2\)-space. For a symmetric form \(D(f, g)\) with domain \(\mathcal D(D)\) on \(L^2(\pi)\), the logarithmic Sobolev inequality means \[ \pi(f^2 \log f^2) \leq c^{-1} D(f,f),\quad f\in \mathcal D(D),\quad \|f\|= 1 \] for some constant \(c\), where \(\pi(f) = \int f d\pi\). The largest possible \(c\) is denoted \(\sigma\) and is called the logarithmic Sobolev constant. The author gives a explicit lower bound on \(\sigma\) for the form \[ D(f, g) = \tfrac 12 \int [f(x) - f(y)][g(x)- g(y)] J(dx,dy) \] with \(J\) nonnegative and symmetric and \(\mathcal D(D):= \{f \in L^2(\pi): D(f, f) < \infty \;\}\). The bounds are far too technical to give here, but are generally sharper than the results of \textit{F. Y. Wang} [Proc. Am. Math. Soc. 128, No. 12, 3675-3682 (2000; Zbl 0961.60079)] and allow him to recover the results of Wang [loc. cit.] for an ergodic birth-death process.