Low-rank approximation of Moment Tensor Potential enables reducing training set size without loss of accuracy
Abstract
In this study, we implement a low-rank approximation of Moment Tensor Potential (MTP) based on the tensor train (TT) decomposition. The implemented tensor-factorized MTP (TFMTP) and the original MTP model are actively trained via a MaxVol-based algorithm during molecular dynamics simulations of a four-component molten salt mixture, LiF-NaF-KF (FLiNaK), and geometry optimizations of a five-component equiatomic Mo-Nb-Ta-W-V random alloy. We demonstrate that under a 1.5-fold compression, TFMTP requires two times fewer configurations for fitting than the original MTP model, while maintaining an indistinguishable level of accuracy. These actively trained MTP and TFMTP models are further used to evaluate the density and viscosity of FLiNaK at temperatures ranging from 600 to 1200 K, as well as the elastic constants and bulk modulus of the Mo-Nb-Ta-W-V alloy at zero temperature. For both atomic systems, the differences in physical properties predicted by MTP and TFMTP are negligible.
Related
- Full-data accuracy with fewer labels for training and fine-tuning machine-learning force fields
- Accuracy, Reliability, and Efficiency of a MACE Machine-Learning Interatomic Potential for Aspirin: A Reproducible Benchmark for Molecular Simulation
- Many-body machine-learned potential for the hydrated proton at CCSD(T) accuracy with fast analytical gradient
- Implementing QESEM's High-Accuracy Error Mitigation on a Quantum Computer: a Water Potential Energy Surface Study
- Reducing Synthetic Complexity: A Fluorine-Free Benzodithiophene-Thienopyrrolodione Donor Polymer by Direct Arylation Polycondensation for Organic Photovoltaics
BibTeX
@misc{bondarenko2026lowrank,
title = {{Low-rank approximation of Moment Tensor Potential enables reducing training set size without loss of accuracy}},
author = {Anna Bondarenko and Nikita Rybin and Maxim Rakhuba and Ivan S. Novikov},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2609.07372},
eprint = {2609.07372},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.07372},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Low-rank approximation of Moment Tensor Potential enables reducing training set size without loss of accuracy
AU - Anna Bondarenko
AU - Nikita Rybin
AU - Maxim Rakhuba
AU - Ivan S. Novikov
PY - 2026
DA - 2026/10/20
PB - arXiv
DO - 10.48550/arxiv.2609.07372
UR - https://arxiv.org/abs/2609.07372
AB - In this study, we implement a low-rank approximation of Moment Tensor Potential (MTP) based on the tensor train (TT) decomposition. The implemented tensor-factorized MTP (TFMTP) and the original MTP model are actively trained via a MaxVol-based algorithm during molecular dynamics simulations of a four-component molten salt mixture, LiF-NaF-KF (FLiNaK), and geometry optimizations of a five-component equiatomic Mo-Nb-Ta-W-V random alloy. We demonstrate that under a 1.5-fold compression, TFMTP requires two times fewer configurations for fitting than the original MTP model, while maintaining an indistinguishable level of accuracy. These actively trained MTP and TFMTP models are further used to evaluate the density and viscosity of FLiNaK at temperatures ranging from 600 to 1200 K, as well as the elastic constants and bulk modulus of the Mo-Nb-Ta-W-V alloy at zero temperature. For both atomic systems, the differences in physical properties predicted by MTP and TFMTP are negligible.
ER -