Performance and Cost of Machine Learning Interatomic Potentials

Our work on “Performance and Cost Assessment of Machine Learning Interatomic Potentials (ML-IAPs)” has been published in the Journal of Physical Chemistry A! Co-authored with the developers of four leading ML-IAPs, this work provides a rigorous assessment of ML-IAPs across several metrics – accuracy in energies and forces, materials properties and training and computing cost. This assessment was carried out using a diverse data set – bcc (Li, Mo) and fcc (Cu, Ni) metals and diamond group IV semiconductors (Si, Ge) – generated using high-throughput density functional theory (DFT) calculations. To facilitate the reuse and reproduction of our results, the code, data and optimized ML models in this work are published open-source on our mlearn Github repo. The code includes high-level Python interfaces for ML-IAPs development as well as LAMMPS material properties calculators.

Check out the publication at this link.