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.
@article{arxiv.2510.19737,
title = {Accelerating Moment Tensor Potentials through Post-Training Pruning},
author = {Zijian Meng and Karim Zongo and Matthew Thoms and Ryan Eric Grant and Laurent Karim Béland},
journal= {arXiv preprint arXiv:2510.19737},
year = {2025}
}
Comments
16 pages, 5 figures Software available from https://github.com/RichardZJM/MTP_basis_optimization