2/4/2025 | 9:10 AM - 10:30 AM | Regency B

Dissipative lagrangian neural networks for diffusion problems

Author(s)

Megna Shah | AFRL
Jeff Simmons | AFRL
Veera Sundararaghavan

Abstract

This work addresses forward and inverse simulation of diffusion using a novel neural network formulation of the Morse--Feshbach Lagrangian. The Morse-Feshbach Lagrangian models dissipative dynamics by doubling the number of dimensions of the system in order to create a ‘mirror’ latent representation that would counterbalance the dissipation of the observable system, making it a conservative system. We  train a network from simulated training data for dissipative systems and show that the approach is able to accurately capture dissipative dynamics. We demonstrate that both forward and backward diffusion can be solved using this approach and lay a pathway for use of this technique for generative modeling.