Table of Contents
Fetching ...

Gradient flow structure for some nonlocal diffusion equations

Andrew Warren

TL;DR

This work casts a broad class of nonlocal diffusion equations as gradient flows of the relative entropy $\mathcal{H}(\cdot|\pi)$ with respect to a nonlocal Wasserstein-type metric built from a jump kernel $\eta$ and reference measure $\pi$. It develops a rigorous Eulerian variational framework (via curves of maximal slope and a Benamou–Brenier-type action) and proves a compactness theory, existence, and uniqueness of weak solutions, along with stability under perturbations of $(\eta,\pi)$ and a discrete-to-continuum convergence result. A chain rule and entropy-dissipation structure are established, yielding an a.e. gradient-flow interpretation and enabling an explicit nonlocal log-Sobolev inequality that gives exponential convergence to equilibrium under suitable conditions. The paper also connects this NLW gradient-flow perspective to nonlocal Dirichlet forms, and demonstrates a finite-volume discretization scheme that converges to the continuum gradient-flow solution. Overall, it extends the Otto calculus to nonlocal diffusion with singular kernels and provides a robust variational theory for well-posedness, stability, and convergence analyses.

Abstract

We study ``nonlocal diffusion equations'' of the form \[ \partial_{t}\frac{dρ_{t}}{dπ}(x)+\int_{X}\left(\frac{dρ_{t}}{dπ}(x)-\frac{dρ_{t}}{dπ}(y)\right)η(x,y)dπ(y)=0\qquad(\dagger) \] where $X$ is either $\mathbb{R}^{d}$ or $\mathbb{T}^{d}$, $π$ is a probability distribution on $X$, and $η(x,y)$ is a ``transition kernel'' which may be singular as $x\rightarrow y$. For a suitable notion of weak solutions which we discuss below, we show that solutions to these nonlocal diffusion equations can be interpreted as gradient flows of the relative entropy with respect to a certain nonlocal Wasserstein-type metric defined in terms of $η$ and $π$. These ``nonlocal Wasserstein metrics'' endow the space of probability measures on $X$ with a formal Riemannian structure, thereby providing for us a nonlocal analogue of the \emph{Otto calculus} originally developed in the context of the 2-Wasserstein metric. The class of equations $(\dagger)$ includes a family of ``nonlocal Fokker-Planck equations'', which are thus identified as nonlocal Wasserstein gradient flows of the relative entropy, analogously with the usual Fokker-Planck equation and the $W_{2}$ metric. The gradient flow structure we provide allows us to deduce: existence and uniqueness of solutions to ($\dagger$) in a suitable class of weak solutions; stability of solutions in the sense of evolutionary $Γ$-convergence, with respect to perturbations of initial condition, reference measure $π$, and transition kernel $η$; sufficient conditions for exponential convergence to equilibrium, in terms of a nonlocal analogue of the log-Sobolev inequality; as well as the consistency of a finite-volume-type spatial discretization scheme in the $\mathbb{T}^{d}$ case.

Gradient flow structure for some nonlocal diffusion equations

TL;DR

This work casts a broad class of nonlocal diffusion equations as gradient flows of the relative entropy with respect to a nonlocal Wasserstein-type metric built from a jump kernel and reference measure . It develops a rigorous Eulerian variational framework (via curves of maximal slope and a Benamou–Brenier-type action) and proves a compactness theory, existence, and uniqueness of weak solutions, along with stability under perturbations of and a discrete-to-continuum convergence result. A chain rule and entropy-dissipation structure are established, yielding an a.e. gradient-flow interpretation and enabling an explicit nonlocal log-Sobolev inequality that gives exponential convergence to equilibrium under suitable conditions. The paper also connects this NLW gradient-flow perspective to nonlocal Dirichlet forms, and demonstrates a finite-volume discretization scheme that converges to the continuum gradient-flow solution. Overall, it extends the Otto calculus to nonlocal diffusion with singular kernels and provides a robust variational theory for well-posedness, stability, and convergence analyses.

Abstract

We study ``nonlocal diffusion equations'' of the form where is either or , is a probability distribution on , and is a ``transition kernel'' which may be singular as . For a suitable notion of weak solutions which we discuss below, we show that solutions to these nonlocal diffusion equations can be interpreted as gradient flows of the relative entropy with respect to a certain nonlocal Wasserstein-type metric defined in terms of and . These ``nonlocal Wasserstein metrics'' endow the space of probability measures on with a formal Riemannian structure, thereby providing for us a nonlocal analogue of the \emph{Otto calculus} originally developed in the context of the 2-Wasserstein metric. The class of equations includes a family of ``nonlocal Fokker-Planck equations'', which are thus identified as nonlocal Wasserstein gradient flows of the relative entropy, analogously with the usual Fokker-Planck equation and the metric. The gradient flow structure we provide allows us to deduce: existence and uniqueness of solutions to () in a suitable class of weak solutions; stability of solutions in the sense of evolutionary -convergence, with respect to perturbations of initial condition, reference measure , and transition kernel ; sufficient conditions for exponential convergence to equilibrium, in terms of a nonlocal analogue of the log-Sobolev inequality; as well as the consistency of a finite-volume-type spatial discretization scheme in the case.
Paper Structure (18 sections, 28 theorems, 320 equations)

This paper contains 18 sections, 28 theorems, 320 equations.

Key Result

Proposition 10

The action $\mathcal{A}_{\theta,\eta}(\mu,\mathbf{v};\pi)$ is jointly convex in $(\mu,\mathbf{v},\pi)$, and is jointly sequentially lower semicontinuous in $\eta$ and $(\mu,\mathbf{v},\pi)$ with respect to: Moreover, $\mathcal{A}_{\theta,\eta}(\mu,\mathbf{v};\pi)$ is jointly topologically l.s.c. in $(\mu,\mathbf{v},\pi)$ for fixed $\eta$.

Theorems & Definitions (78)

  • Remark
  • Definition 2: Underlying graph
  • Definition 3: Locally finite signed measures on $G$
  • Definition 6: Nonlocal gradient and divergence
  • Remark 7
  • Definition 8: Action
  • Remark
  • Remark 9
  • Proposition 10
  • proof
  • ...and 68 more