The Shortest-Path distance on graphons
Cédric Simal, Julien Petit, Timoteo Carletti
TL;DR
This work defines an intrinsic, Varadhan-based distance on graphons by extending heat-kernel ideas from manifolds and graphs to the graphon setting, yielding an integer-valued distance that aligns with the usual shortest-path distance on step graphons. To obtain a finer geometry, it connects this Varadhan distance to the communicability distance, realized through the exponential of the adjacency operator and offering a Hilbert-space embedding via spectral data. The authors develop notions of connectedness for graphons, heat-kernel premetrics, and a computable pointwise distance d_W that coincides with finite-graph shortest-path distances in the step-graphon limit. They also extend the communicability embedding to graphons, enabling a Euclidean interpretation of graphon geometry through eigenfunctions. The framework lays groundwork for distance-based analysis of graphon limits and prompts future work on convergence, additional graph-analytic distances, and applications to random graph models.
Abstract
We define an analogue of the shortest-path distance for graphons. The proposed method is rooted on the extension to graphons of Varadhan's formula, a result that links the solution of the heat equation on a Riemannian manifold to its geodesic distance. The resulting metric is integer-valued, and for step graphons obtained from finite graphs it is essentially equivalent to the usual shortest-path distance. We further draw a link between the Varadhan distance and the communicability distance, that contains information from all paths, not just shortest-paths, and thus provides a finer distance on graphons along with a natural isometric embedding into a Hilbert space.
