MLIP-Compatible Recipes¶

Several popular datasets used to train machine-learned interatomic potentials (MLIPs) were built from large collections of DFT calculations run with a specific, fixed set of computational settings. To let users generate additional data that is consistent with one of these datasets (e.g. to extend a dataset or build an internally-consistent benchmark without introducing methodological drift), quacc ships a set of recipes that reproduce those settings.

Supported Datasets¶

A Representative Example¶

from ase.build import bulk
from quacc.recipes.vasp.fairchem import omat_static_job

# Make an Atoms object of a bulk Cu structure
atoms = bulk("Cu")

# Run a static calculation with OMat24-compatible VASP settings
result = omat_static_job(atoms)
print(result)

As with all quacc recipes, you can override any of the default calculator settings by passing additional keyword arguments:

result = omat_static_job(atoms, encut=600)