Conceptual and practical bases for the high accuracy of machine learning interatomic potential

TL;DR

This paper provides conceptual and practical insights into why machine learning interatomic potentials (MLIPs) achieve high accuracy, and demonstrates the development of the most accurate MLIP for elemental titanium using advanced descriptors.

Contribution

It offers a theoretical understanding of MLIP accuracy and introduces a highly precise MLIP for titanium using a linearized framework and angular-dependent descriptors.

Findings

High-accuracy MLIP for titanium developed

Conceptual basis for MLIP accuracy established

Use of angular-dependent descriptors enhances precision

Abstract

Machine learning interatomic potentials (MLIPs) based on a large dataset obtained by density functional theory (DFT) calculation have been developed recently. This study gives both conceptual and practical bases for the high accuracy of MLIPs, although MLIPs have been considered to be simply an accurate black-box description of atomic energy. We also construct the most accurate MLIP of the elemental Ti ever reported using a linearized MLIP framework and many angular-dependent descriptors, which also corresponds to a generalization of the modified embedded atom method (MEAM) potential.