Compressed magnetic Moment Tensor Potentials via low-rank matrix and tensor factorizations

We propose a parameter-reduced version of magnetic Moment Tensor Potential (mMTP) based on various matrix and tensor decompositions. The resulting compressed magnetic machine-learning potential reduces the number of parameters by a factor of approximately 1.5-3 without compromising predictive performance on the validation set. We evaluate the performance of the compressed potentials for magnetic, structural, and vibrational properties, as well as in molecular dynamics simulations of Fe-Al and CrN systems. We demonstrate that the simulation results obtained with the compressed mMTP are numerically consistent with those of the original uncompressed model and are in agreement with density functional theory calculations and experimental data.