Accelerating Moment Tensor Potentials through Post-Training Pruning

This repository provides data for Accelerating Moment Tensor Potentials through Post-Training Pruning. It contains two systems: Nickel and Silicon–Oxygen. The pruning code is available from https://github.com/RichardZJM/MTP_basis_optimization. Each system has a folder containing five subfolders: pareto: The resultant Pareto front from the pruning process for both MOEA/D and NSGA-II. training_accuracy_cost: The cost and training errors of both the pruned and original potentials. All costs were measured on a single AMD EPYC 9654 core: Nickel with 2048 FCC atoms; Silicon–Oxygen with 1944 α-quartz atoms. level_pots: Potential files of the original level-based potentials, including active learning data. pruned_pots: Potential files of the pruned potentials, labeled 0–11 (instead of A–L), including active learning data. properties: Physical properties of the potentials and corresponding DFT and experimental reference data. Nickel: Physical properties were obtained using the Nickel potential benchmarking suite from Thoms et al. [1]. Pruned potentials are labeled 0–11 instead of A–L. Silicon–Oxygen: Simulation procedure and conditions are available in the supporting information of the manuscript. Pruned potentials are labeled 0–11 instead of A–L. Original level-based potentials are labeled 0–11 instead of 6–28. The training datasets are not included. The nickel dataset can be obtained from [2,3], and the silicon–oxygen dataset is available from [4]. References: [1] Thoms, M.; Sun, H.; B´eland, L. K. Benchmarking 34 OpenKIM Nickel Potentials with an Emphasis on Surfaces and Extended Defects. 2025; https://arxiv.org/abs/2510.18033. [2] Andolina, C. M.; Saidi, W. A. Highly transferable atomistic machine-learning potentials from curated and compact datasets across the periodic table. Digital Discovery 2023, 2, 1070–1077. [3] Vita, J. A.; Fuemmeler, E. G.; Gupta, A.; Wolfe, G. P.; Tao, A. Q.; Elliott, R. S.; Martiniani, S.; Tadmor, E. B. ColabFit exchange: Open-access datasets for data-driven interatomic potentials. The Journal of Chemical Physics 2023, 159. [4] Zongo, K.; Sun, H.; Ouellet-Plamondon, C.; Béland, L. K. A unified moment tensor potential for silicon, oxygen, and silica. npj Computational Materials 2024, 10, 218.