Score-based neural ordinary differential equations for reaction-diffusions
Organizations: Department of Mathematics, University of South Carolina, Columbia, SC 29208, USA
Abstract
A diffusion equation can be rewritten as a transport equation driven by the score function of the evolving density. Conventional probability-flow formulations, which move particles deterministically with this score-driven velocity, typically conserve total mass. In this work, we develop a Lagrangian method for reaction-diffusion equations in which each particle carries a position and a physical weight. The characteristic ODEs incorporate source terms into the evolution of the log-density and its gradient, while particle weights track changes in mass. We combine this representation with a learned velocity field and weighted particle training to approximate these dynamics. For non-increasing reaction rates, a relative-entropy inequality supports the mass-normalized self-consistency residual as the training objective. This objective can control the error of the induced density when the growth rate is bounded. The algorithm divides the time interval into short windows. In each window, a frozen velocity generates the particle data for a regression of the network velocity. We draw the particles from period-adapted initial sampling distributions and weight them in the loss partly by their predicted future mass, because the reaction can make regions with little initial mass important later. We numerically investigate the method on the Ornstein--Uhlenbeck (OU) equation with a linear source term, Fisher--KPP equations, and Wasserstein--Fisher--Rao gradient flows.
Figures & tables
| Problem | (%) | (%) | (%) | ||
|---|---|---|---|---|---|
| OU with linear source | 2 | 0.145–0.152 | 0.56 | 0.0094 | 0.05 |
| Fisher–KPP | 1 | 0.10–0.18 | 0.27 | 0.026 | 0.11 |
| Fisher–KPP | 3 | 0.11–0.12 | 0.17 | 0.030 | 0.08 |
| Sampling distribution | Weights | (%) | (%) | |
|---|---|---|---|---|
| period-adapted | adaptive | 3.6–4.2 | 0.74 | 0.013 |
| single | adaptive | 6.3–20.4 | 1.2 | 0.014 |
| period-adapted | physical | 7.9–10.3 | 3.6 | 0.049 |
| single | physical | 10.8–25.6 | 4.2 | 0.063 |
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
| Problem | Predicted density | Defensive density |
|---|---|---|
| OU | solution without source | invariant density of the drift |
| Fisher–KPP | heat flow followed by the logistic map, | |
| WFR | mixture of the exact WFR gradient flows from the Gaussian towards each single target component | the target |
| Component | Setting |
|---|---|
| Velocity | (Appendix C.1 ); : 3 hidden layers of width 256, SiLU, double precision, output with |
| Integration | Heun, ; windows of width (100 windows for , 200 for ) |
| Periods | , , |
| Sampling | three period-adapted sampling distributions with 8192 particles each (32768 for 3D Fisher–KPP); pilot of 262144 initial points; second round of 131072 per distribution; at most 4 Gaussian components per mode (8 for 3D Fisher–KPP); , , |
| Loss | Huber (threshold 1), , adaptive weights ( 32 ) |
| Optimizer | Adam [ 27 ] , learning rate , minibatch 2048 (1024 for the regularizer term), gradient-norm clip 10; Adam is reset and the network warm-started at each window |