Title: Criteria for entropic curvature on graph spaces

Abstract: This paper presents local criteria for lower bounds on entropic curvature of graph spaces along Schr\"odinger bridges at zero temperature, according to the definition given by the second named author in [31], in the continuity of the work by C. L\'eonard [19] inspired by the Lott-Sturm-Villani theory. A graph space is defined as a quadruple $(\mathcal{X},d,L,m)$ where $\mathcal{X}$ is the set of vertices, $d$ is the combinatorial distance and $m$ is a reversible reference measure with respect to a generator $L$ of a Markov semi-group on $\mathcal{X}$. The criteria are given by local optimization problems on balls of radius two, depending only on the generator $L$ and on the discrete structure of these balls. General tensorization properties of the criteria are presented for the study of the Cartesian product of graphs. This approach is robust since it applies to a wide range of graph spaces and also for any measure $m$, including measures with interaction potential like Ising models. We introduce a very large class of structured graphs, for which the local criteria give non-negative entropic curvature for the uniform measure. A Bonnet-Myers type of theorem also ensures that any such graph with positive entropic curvature is finite. On any graph space, positive entropic curvature provides transport-entropy inequalities with so-called weak optimal transport costs, as well as Poincar\'e or modified log-Sobolev inequalities for the renormalized probability measure $\mu = m/m(\mathcal{X})$. These inequalities are related to refined concentration properties of the measure $\mu$, speed of convergence of semi-groups to the measure $\mu$. We also present examples of graphs with negative curvature. Some comparisons of our results with other notions of curvature are established, such as Bakry-Emery curvature conditions [7,8], Ollivier or Lin-Lu-Yau's curvature [28,20].