Supported Models

ALCHEMI Toolkit ships wrappers for several machine-learning interatomic potentials (MLIPs) and classical force fields. Every wrapper implements the BaseModelMixin interface and exposes a ModelCard that declares its capabilities and input requirements.

For a step-by-step guide on wrapping your own model, see the Models: Wrapping ML Interatomic Potentials.

Machine-Learned Potentials

Neural-network potentials that learn interatomic interactions from quantum mechanical reference data.

Wrapper	Energies	Forces	Stresses	Hessians	Dipoles	PBC	Needs PBC	Node Charges	Autograd Forces	Node Embed.	Graph Embed.	Neighbor Fmt.
MACEWrapper	✓	✓	✓	✗	✗	✓	✗	✗	✓	✓	✓	COO
DemoModelWrapper	✓	✓	✗	✗	✗	✗	✗	✗	✗	✗	✗	—

Physical / Classical Models

Analytical force fields and correction terms based on known physical functional forms.

Wrapper	Energies	Forces	Stresses	Hessians	Dipoles	PBC	Needs PBC	Node Charges	Autograd Forces	Node Embed.	Graph Embed.	Neighbor Fmt.
LennardJonesModelWrapper	✓	✓	✓	✗	✗	✓	✗	✗	✗	✗	✗	MATRIX
DFTD3ModelWrapper	✓	✓	✓	✗	✗	✓	✗	✗	✗	✗	✗	MATRIX
PMEModelWrapper	✓	✓	✓	✗	✗	✓	✓	✓	✗	✗	✗	MATRIX
EwaldModelWrapper	✓	✓	✓	✗	✗	✓	✓	✓	✗	✗	✗	MATRIX

Note

ComposableModelWrapper is excluded from the tables above because its capabilities are synthesized at runtime from the sub-models it composes (see ComposableModelWrapper). All tables are auto-generated from each wrapper’s ModelCard at documentation build time.

Foundation Models

Pre-trained checkpoints that can be loaded directly via list_foundation_models() and MACEWrapper.from_checkpoint(name).

Name	Description	Aliases	Wrapper
mace-mp-0b2-large	MACE-MP-0b2 large (128 channels, L=2 equivariance). Universal potential trained on Materials Project DFT data. Higher accuracy at greater computational cost.	mace-mp-large	MACEWrapper
mace-mp-0b2-medium	MACE-MP-0b2 medium (128 channels, L=1 equivariance). Universal potential trained on Materials Project DFT data. Recommended default for most materials simulation tasks.	mace-mp-medium, mace-mp, mace-mp-0b2	MACEWrapper
mace-mp-0b2-small	MACE-MP-0b2 small (128 channels, L=0 equivariance). Universal potential trained on Materials Project DFT data.	mace-mp-small, mace-mp-0b2-small	MACEWrapper
mace-mpa-0b3-medium	MACE-MPA-0b3 medium. Universal potential trained on the Materials Project Alexandria dataset with broader coverage of chemical space.	mace-mpa-medium, mace-mpa	MACEWrapper

References

If you use any of the model wrappers provided by ALCHEMI Toolkit, please cite the original publications for the underlying methods.

Model	Citation
MACE	Batatia, I. et al. “MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields.” Advances in Neural Information Processing Systems (NeurIPS), 2022. openreview.net/forum?id=YPpSngE-ZU
MACE-MP-0 (foundation)	Batatia, I. et al. “A foundation model for atomistic materials chemistry.” arXiv:2401.00096, 2023. doi:10.48550/arXiv.2401.00096
DFT-D3(BJ)	Grimme, S. et al. “A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu.” J. Chem. Phys. 132, 154104, 2010. doi:10.1063/1.3382344
	Grimme, S., Ehrlich, S. & Goerigk, L. “Effect of the damping function in dispersion corrected density functional theory.” J. Comput. Chem. 32, 1456–1465, 2011. doi:10.1002/jcc.21759
Lennard-Jones	Jones, J. E. “On the Determination of Molecular Fields.” Proc. R. Soc. Lond. A 106 (738), 463–477, 1924. doi:10.1098/rspa.1924.0082
Ewald Summation	Ewald, P. P. “Die Berechnung optischer und elektrostatischer Gitterpotentiale.” Ann. Phys. 369 (3), 253–287, 1921. doi:10.1002/andp.19213690304
Particle Mesh Ewald	Darden, T., York, D. & Pedersen, L. “Particle mesh Ewald: An N*log(N) method for Ewald sums in large systems.” J. Chem. Phys. 98 (12), 10089–10092, 1993. doi:10.1063/1.464397