Rigorous justification for the combined message-passing and mean-field approximation

Establish a rigorous, analytic justification or proof for the accuracy of the combined approximation for the generating function of first return times of random walks on finite networks, specifically the approximation defined by F_i(z) = \widetilde{F}_i(z) + (z - z \widetilde{F}_i(1)) / (z + h_i - z h_i) with h_i = (2m - k_i \widetilde{F}_i'(1)) / (k_i - k_i \widetilde{F}_i(1)), where \widetilde{F}_i(z) is obtained via message passing on the r-neighborhood to capture local structure and short cycles. Demonstrate under what network conditions (e.g., sparsity, local tree-likeness, presence of short cycles) this approximation provably yields accurate first return time distributions and clarify the mechanism that ensures its effectiveness.

Background

The paper introduces a combined approximation for the first return time distribution of random walks on undirected, unweighted networks. The approach blends a local message-passing method (capturing short-time behavior and local topology, including an r-neighborhood to handle cycles) with a mean-field correction to enforce global constraints (F_i(1)=1 and F_i'(1)=2m/k_i) via a geometric-tail adjustment parameterized by h_i.

Although extensive numerical experiments show excellent agreement with ground truth across various sparse network families (random regular graphs, Poisson random graphs, and stochastic block models), the authors stress that the method is derived from intuitive arguments rather than established theory, and they do not provide an analytic proof of its validity. A rigorous explanation of why and when this combined approximation must work remains unresolved.