AI Insight
The authors introduce Neural Discrete Equilibrium (NeurDE), a scientific machine learning framework that integrates a neural network into a kinetic (Boltzmann-based) solver to predict the behavior of nonlinear conservation laws over long time horizons. Rather than replacing the full numerical solver, NeurDE uses the neural network only to learn the local equilibrium closure, while the solver retains responsibility for transport, relaxation, and conservation enforcement. Tested across six conserved physical systems, including subsonic, transonic, and supersonic shock scenarios, NeurDE outperforms state-of-the-art neural operators and large pretrained foundation models that are up to one million times larger, while also improving upon the baseline numerical method.
Why it matters
Accurately and efficiently simulating nonlinear conservation laws is critical in aerospace engineering, fluid dynamics, and climate modeling; NeurDE offers a compact, physically faithful alternative to large general-purpose models, potentially reducing computational costs while maintaining stability and accuracy in industrial and scientific simulations.
arXiv:2501.06933v3 Announce Type: replace-cross
Abstract: Nonlinear conservation laws govern a broad class of important physical systems in science and industry and are central to scientific machine learning (SciML). Large general-purpose models offer speed, but replacing the numerical and physical structure of solvers often compromises stability, accuracy, and physical faithfulness. Here, we aim to balance the general inductive bias of conservation with the flexibility and speed of neural networks through a conservation-aware SciML backbone, which we call Neural Discrete Equilibrium (NeurDE). NeurDE places machine learning inside a kinetic solver by learning the local equilibrium closure of a Boltzmann formulation. The kinetic solver still performs transport, relaxation, moment recovery, and conservation; the neural network provides only the nonlinear equilibrium target. We test NeurDE on $6$ conserved systems, including three very challenging subsonic, transonic, and supersonic shock systems. NeurDE outperforms state-of-the-art SciML methods, including neural operators and pretrained SciML foundation models that are $10^4$ and $10^6$ times larger, respectively. Most notably, NeurDE improves upon the numerical method from which it is derived. NeurDE therefore provides a compact target for scientific machine learning in conservative simulation: learn the equilibrium law toward which the system relaxes, not the evolution law itself.
Source: Neural equilibria for long-term prediction of nonlinear conservation laws