Fluctuations of subgraph counts in exponential random graphs
In the authors' words
In this paper, we study fluctuations of fixed subgraph counts in dense exponential random graphs on vertices. We compare each normalized subgraph count with the normalized edge count : their difference is in , whereas for every -Lipschitz function . We also obtain Kolmogorov and Wasserstein bounds, optimal for the edge count, with relative variance error of the same order. These estimates hold uniformly on compact subsets of the ferromagnetic subcritical and absolute-coefficient Dobrushin regions, and under small fixed signed perturbations of the former. Using this uniform control, we derive a local Gaussian likelihood expansion with rank-one leading Fisher information and two-sided moderate deviations with an explicit cubic correction. The proofs combine two applications of the normal Stein equation, covariance estimates from Glauber dynamics and exponential tilting.
Appeared: Wednesday, October 21. arXiv. Preprint, not yet peer-reviewed.
Authors' comment: 52 pages