MP-ALOE: An r2SCAN dataset for universal machine learning interatomic potentials

TL;DR

MP-ALOE is a large, diverse dataset of nearly 1 million DFT calculations designed to train and evaluate machine learning interatomic potentials across a wide range of elements and structural conditions.

Contribution

This work introduces MP-ALOE, a comprehensive dataset created with active learning for training universal machine learning interatomic potentials.

Findings

ML potential trained on MP-ALOE performs well on thermochemical benchmarks.

Potential maintains physical soundness under extreme deformations.

Model demonstrates stability in molecular dynamics at high temperatures and pressures.

Abstract

We present MP-ALOE, a dataset of nearly 1 million DFT calculations using the accurate r2SCAN meta-generalized gradient approximation. Covering 89 elements, MP-ALOE was created using active learning and primarily consists of off-equilibrium structures. We benchmark a machine learning interatomic potential trained on MP-ALOE, and evaluate its performance on a series of benchmarks, including predicting the thermochemical properties of equilibrium structures; predicting forces of far-from-equilibrium structures; maintaining physical soundness under static extreme deformations; and molecular dynamic stability under extreme temperatures and pressures. MP-ALOE shows strong performance on all of these benchmarks, and is made public for the broader community to utilize.