MACE
An equivariant machine-learning interatomic-potential framework for training models and using pretrained molecular or materials force fields.
Where it fits
Fast energies, forces and dynamics within a model’s validated chemical domain.
Limits and practical considerations
Pretrained models differ in elements, charge/spin treatment and reference theory; validate before transfer.
Licensing and access
MIT-licensed code; model weights have separate licences, including MIT and ASL variants. Check each model.