Dynamically training machine-learning-based force fields for strongly anharmonic materials

doi:10.1016/j.cpc.2026.110423

Dynamically training machine-learning-based force fields for strongly anharmonic materials

2026-09-22

0

OA

AI

DOI：10.1016/j.cpc.2026.110423

Original

Original request for help

Save

Abstract

En 中文

• Dynamic training leads to reliable force fields even for anharmonic materials. • Reduce redundant training data by using a trajectory average of the Bayesian error. • The fitting error in Bayesian regression provides a lower bound for the actual error. • Convergence of the dynamic training can thus be measured by the fitting error.

Keywords:

Molecular dynamics

Dynamic training

Machine-learning

Bayesian linear regression

Anharmonicity

Dynamically training machine-learning-based force fields for strongly anharmonic materials

Abstract

Journal

Organization

Cited Papers