1. Universal Machine Learning Interatomic Potentials are Ready for Phonons
- Author
-
Loew, Antoine, Sun, Dewen, Wang, Hai-Chen, Botti, Silvana, and Marques, Miguel A. L.
- Subjects
Condensed Matter - Materials Science ,Physics - Computational Physics - Abstract
There has been an ongoing race for the past couple of years to develop the best universal machine learning interatomic potential. This rapid growth has driven researchers to create increasingly accurate models for predicting energy, forces, and stresses, combining innovative architectures with big data. Here, these models are benchmarked for their performance in predicting phonon properties, which are critical for understanding the vibrational and thermal behavior of materials. Our analysis is based on around 10 000 ab initio phonon calculations, enabling us to assess performance across a range of phonon-related parameters while testing the universal applicability of these models. The results reveal that some models are already capable of predicting phonon properties with a high level of accuracy. However, others still exhibit substantial inaccuracies, even if they excel in the prediction of the energy and the forces for materials close to dynamical equilibrium. These findings highlight the importance of considering phonon-related properties in the development of universal machine learning interatomic potentials.
- Published
- 2024