Systematic Atomic Structure Datasets for Machine Learning Potentials: Application to Defects in Magnesium

TL;DR

This paper introduces a systematic, physically motivated method for creating training datasets for machine learning interatomic potentials, enabling accurate modeling of magnesium defects without explicit defect training data.

Contribution

It proposes a novel approach to generate unbiased training sets through exploration of crystal space groups and deformations, improving transferability of potentials.

Findings

Potentials accurately reproduce magnesium polymorph properties.

Method achieves transferable potentials without defect-specific training data.

Exploration of training structures impacts potential accuracy.

Abstract

We present a physically motivated strategy for the construction of training sets for transferable machine learning interatomic potentials. It is based on a systematic exploration of all possible space groups in random crystal structures, together with deformations of cell shape, size, and atomic positions. The resulting potentials turn out to be unbiased and generically applicable to studies of bulk defects without including any defect structures in the training set or employing any additional Active Learning. Using this approach we construct transferable potentials for pure Magnesium that reproduce the properties of hexagonal closed packed (hcp) and body centered cubic (bcc) polymorphs very well. In the process we investigate how different types of training structures impact the properties and the predictive power of the resulting potential.