Singularities in the entropy of asymptotically large simple graphs (Q2342078)

Large networks or graphs whereby the density of certain subgraphs are controlled directly are analogues of microcanonical ensembles in statistical mechanics. This paper investigates the asymptotic entropy of large simple graphs, a quantity that relates, asymptotically in the number of vertices, the relative number of graphs with prescribed numbers of edges and triangles. Various large-scale quantities of interest may be obtained by differentiating the asymptotic entropy with respect to appropriate parameters, and phase transitions may be determined by examining the singularities of the derivatives. The authors present a rigorous proof of the non-differentiability of the entropy along the Erdős-Rényi curve and give a detailed discussion of this singular effect on the non-equivalence of ensembles. By looking into the maximizers of the entropy, they also characterize the precise modified bipartite structure of asymptotic graphs with edge density \(e=1/2\) and triangle density \(t\leq 1/8\).