Evaluating approaches for on-the-fly machine learning interatomic potential for activated mechanisms sampling with the activation-relaxation technique nouveau

TL;DR

This paper compares specific and general machine-learning potentials for modeling activated mechanisms in materials, finding that targeted on-the-fly potentials integrated with ARTn provide high accuracy and efficiency.

Contribution

It demonstrates that on-the-fly, system-specific ML potentials integrated with ARTn outperform general potentials in accuracy and cost-effectiveness for activated process simulations.

Findings

Targeted on-the-fly potentials yield higher precision in energy barriers.

System-specific ML potentials are more cost-effective than general ones.

The approach broadens the scope of high-accuracy ML potential applications.

Abstract

In the last few years, much efforts have gone into developing universal machine-learning potentials able to describe interactions for a wide range of structures and phases. Yet, as attention turns to more complex materials including alloys, disordered and heterogeneous systems, the challenge of providing reliable description for all possible environment become ever more costly. In this work, we evaluate the benefits of using specific versus general potentials for the study of activated mechanisms in solid-state materials. More specifically, we tests three machine-learning fitting approaches using the moment-tensor potential to reproduce a reference potential when exploring the energy landscape around a vacancy in Stillinger-Weber silicon crystal and silicon-germanium zincblende structure using the activation-relaxation technique nouveau (ARTn). We find that a a targeted on-the-fly approach specific and integrated to ARTn generates the highest precision on the energetic and geometry of activated barriers, while remaining cost-effective. This approach expands the type of problems that can be addressed with high-accuracy ML potentials.