Warm-Starting PDE Solvers with Any-Dimensional Machine Learning

License: See LICENSE file · Updated 2026-09-30
My rating

This research targets general partial differential equation simulation across different spatial dimensions. It uses graph neural networks and other any-dimensional machine learning models to train a PDE learning-based solver in small dimensions, then directly apply it to higher-dimensional PDEs in a zero-shot fashion. The authors establish mathematical conditions for this transfer based on symmetries.

Tags: