Geodesic convexity of the relative entropy in reversible Markov chains (Q368543)

The present work is motivated by the author's previous text [Nonlinearity 24, No. 4, 1329--1346 (2011; Zbl 1227.35161)], where the gradient structure for the Fokker-Planck equation has been generalized to encompass general reaction-diffusion systems with detailed balance. Here, finite-dimensional time-continuous Markov chains, obeying the detailed balance (reversibility) condition, are interpreted as gradient systems with the relative entropy as a driving functional. An important formal ingredient of the dynamical rule is the Onsager matrix, recognised as an inverse of the Riemannian metric tensor. The focus of this work is to provide conditions such that the relative entropy is geodesically \(\lambda\)-convex with respect to the Riemannian metric. This amounts to selecting optimal values of \(\lambda\) or analytic estimates that yield suitable bounds for \(\lambda\). General methods are developed and tested for several examples, including some discretizations of the one-dimensional Fokker-Planck equations. Links with the Wasserstein transport (and metric) are mentioned.