L\'{e}vy Score Function and Score-Based Particle Algorithm for Nonlinear L\'{e}vy--Fokker--Planck Equations

TL;DR

This paper introduces a novel Lévy score function and a score-based particle algorithm for nonlinear Lévy--Fokker--Planck equations, enabling better modeling and sampling of jump processes with discontinuities.

Contribution

It derives the Lévy score function for jump processes, develops a training algorithm, and proposes a self-consistent particle sampling method with error analysis.

Findings

Effective sampling of Lévy processes demonstrated in biological applications.

Error bounds established for the divergence between numerical and true densities.

Numerical results show improved modeling of systems with jumps.

Abstract

The score function for the diffusion process, also known as the gradient of the log-density, is a basic concept to characterize the probability flow with important applications in the score-based diffusion generative modelling and the simulation of It\^{o} stochastic differential equations. However, neither the probability flow nor the corresponding score function for the diffusion-jump process are known. This paper delivers mathematical derivation, numerical algorithm, and error analysis focusing on the corresponding score function in non-Gaussian systems with jumps and discontinuities represented by the nonlinear L\'{e}vy--Fokker--Planck equations. We propose the L\'{e}vy score function for such stochastic equations, which features a nonlocal double-integral term, and we develop its training algorithm by minimizing the proposed loss function from samples. Based on the equivalence of the probability flow with deterministic dynamics, we develop a self-consistent score-based transport particle algorithm to sample the interactive L\'{e}vy stochastic process at discrete time grid points. We provide error bound for the Kullback--Leibler divergence between the numerical and true probability density functions by overcoming the nonlocal challenges in the L\'{e}vy score. The full error analysis with the Monte Carlo error and the time discretization error is furthermore established. To show the usefulness and efficiency of our approach, numerical examples from applications in biology and finance are tested.