Accelerating Moment Tensor Potentials through post-training pruning

Computational Materials Science 2026 Open access

Read full text
Authors: Zijian Meng, Karim Zongo, M. Thoms, Christopher Maxwell, Edmanuel Torres, Ryan Eric Grant, Laurent Karim Béland
DOI: 10.1016/j.commatsci.2026.114725
Cited by: 0
References: 36

Abstract

Moment Tensor Potentials (MTPs) are machine-learning interatomic potentials whose basis functions are typically selected using a level-based scheme that is data-agnostic. We introduce a post-training, cost-aware pruning strategy that removes expensive basis functions with minimal loss of accuracy. Applied to nickel and silicon–oxygen systems, it yields models up to seven times faster than standard MTPs. The method requires no new data and remains fully compatible with current MTP implementations.

Fields of Study & Subjects

Quantum and electron transport phenomena Quantum many-body systems Machine Learning in Materials Science

Publication details

Journal / source
Computational Materials Science
Publisher
Elsevier BV
Volume
269
Pages
114725–114725
Publication date
23 April 2026
Language
EN