Persistence of WFR guarantees beyond the log-concave regime

Determine whether the theoretical convergence and structural guarantees established for Wasserstein–Fisher–Rao gradient flows persist for target distributions beyond the strongly log-concave regime, including weakly log-concave, non-convex, or multimodal distributions.

Background

The paper establishes uniform preservation of strong log-concavity and exponential convergence for Wasserstein–Fisher–Rao gradient flows under explicit curvature assumptions on strongly log-concave target and initial distributions. These assumptions restrict the theory to a comparatively narrow class of sampling problems.

The authors explicitly identify extending the analysis to weakly log-concave, non-convex, and multimodal targets as an unresolved issue. This is motivated by the fact that WFR dynamics are intended to improve sampling in challenging landscapes, where the persistence of the paper’s theoretical guarantees is not established.