Learned surrogates for accelerated materials simulation
T. Ferreira · J. Okafor · Discovery Group
arXiv:2603.09918
First-principles simulation is the ground truth of materials science and far too slow to search. We train surrogate models that reproduce its predictions at a fraction of the cost, constrained architecturally to respect conservation laws.
The speed-up turns simulation from a verification step into a search primitive, letting us explore candidate materials at a scale experiment cannot approach.
This is a plain-language summary. The full manuscript and supplementary material are available on request.