Accelerating Moment Tensor Potentials through post-training pruning
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.
Authors
- Ryan E. Grant (ORCID: https://orcid.org/0000-0002-0163-3892)
- Karim Zongo (ORCID: https://orcid.org/0009-0000-3813-0757)
- Laurent Karim Béland (ORCID: https://orcid.org/0000-0001-5332-7128)
- M. Thoms
Publication Details
- Journal
- Computational Materials Science
- Published
- 2026-04-23
- DOI
- https://doi.org/10.1016/j.commatsci.2026.114725
- Primary Topic
- Quantum and electron transport phenomena
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Mitacs
- Atomic Energy of Canada Limited
- University Network of Excellence in Nuclear Engineering
- Alliance de recherche numérique du Canada
- Ministry of Colleges and Universities
- Natural Sciences and Engineering Research Council of Canada