Gradient flow structure, well-posedness and asymptotic behavior of Fokker-Planck equation on locally finite graphs
Cong Wang
TL;DR
This work extends the gradient-flow interpretation of the Fokker-Planck equation from finite graphs to infinite, locally uniformly finite graphs by constructing a 2-Wasserstein-type metric on the probability density space. The FP equation is shown to be the gradient flow of the free energy $\\mathcal{F}(\\bm{\\rho}) = \sum_i \\pi_i\\Psi_i\\rho_i + \\sum_i \\pi_i\\rho_i\\log\\rho_i$ on the infinite-dimensional manifold $(\\mathcal{P}_0^*(G),\\mathcal{W}_2)$, with the Gibbs density $\\rho_i^* = \frac{1}{K}e^{-\\Psi_i}$ serving as the global attractor. The authors establish the existence and uniqueness of global solutions within $\\mathcal{P}_0^*(G)$ and prove convergence to the Gibbs state in $\\ell^{r}(V,\\bm{\\pi})$ norms for $r\\in[2,\\infty]$, providing the first rigorous Wasserstein-type metric framework and FP analysis on infinite graphs. These results lay groundwork for understanding diffusion and equilibration on large or infinite networks under a variational, metric-measure-theoretic lens.
Abstract
This paper investigates the gradient flow structure, well-posedness, and asymptotic behavior of the Fokker-Planck equation defined on locally uniformly finite graphs, which is highly non-trivial compared with the finite case. We first construct a 2-Wasserstein-type metric and gradient flow equation in the probability density space associated with the underlying graphs. Then, we prove the global existence of solution to the Fokker-Planck equation using a novel approach that differs significantly from the methods applied in the finite case. We also demonstrate that the solution converges to the Gibbs distribution in the $\ell^{r}(V,\bmπ)$ norm with $r\in [2,\infty]$, by using the indicator set partitioning method. To the best of our knowledge, this work seems the first result on the study of Wasserstein-type metrics and the Fokker-Planck equation in probability density spaces defined on infinite graphs.
