A Performance and Cost Assessment of Machine Learning Interatomic Potentials.

By Yunxing Zuo, Chi Chen, Xiang-Guo Li, Z. Deng, Yiming Chen, J. Behler, Gábor Csányi, Alexander V. Shapeev, Aidan P. Thompson, Mitchell A. Wood and 1 more. Journal of Physical Chemistry A, 2019

807 citations recorded by the source corpus.

Machine learning of the quantitative relationship between local environment descriptors and the potential energy surface of a system of atoms has emerged as a new frontier in the development of interatomic potentials (IAPs). Here, we present a comprehensive evaluation of ML-IAPs based on four local environment descriptors --- atom-centered symmetry functions (ACSF), smooth overlap of atomic positions (SOAP), the Spectral Neighbor Analysis Potential (SNAP) bispectrum components, and moment tensors --- using a diverse data set generated using high-throughput density functional theory (DFT) calculations. The data set comprising bcc (Li, Mo) and fcc (Cu, Ni) metals and diamond group IV semiconductors (Si, Ge) is chosen to span a range of crystal structures and bonding. All descriptors studied show excellent performance in predicting energies and forces far surpassing that of classical IAPs, as well as predicting properties such as elastic constants and phonon dispersion curves. We observe a general trade-off between accuracy and the degrees of freedom of each model, and consequently computational cost. We will discuss these trade-offs in the context of model selection for molecular dynamics and other applications.

Subfields: Materials Chemistry, Physical and Theoretical Chemistry

View Paper: https://www.semanticscholar.org/paper/60446c372811c25acc1e47b044e8e7458f0a4986

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Materials Chemistry

In Physical and Theoretical Chemistry

Citations in the corpus

Machine Learning for High-entropy Alloys: Progress, Challenges and Opportunities (2022, methodology): “To this end, a series of ML AIMs have been proposed [61, 64], including HDNNP, GAP, SNAP, MTP, LRP, Bayesian CE.”
A universal graph deep learning interatomic potential for the periodic table (2022, methodology): “These so-called ML-IAPs typically express the PES as a function of local environment descriptors such as the interatomic distances and angles, or atomic densities, and have been demonstrated to substantially outperform classical IAPs across a broad range of chemistries 9 .”
Scaling deep learning for materials discovery (2023, methodology): “Although neural network models offer flexibility that allows them to achieve state-of-the-art performance on a wide range of problems, they may not generalize to data outside the training distribution.”
Exceptional piezoelectricity, high thermal conductivity and stiffness and promising photocatalysis in two-dimensional MoSi2N4 family confirmed by first-principles (2020, methodology): “To this aim we train moment tensor potentials (MTPs)[58], which have been proven as accurate and computationally efficient models for describing the atomic interactions [59–61].”
Emerging materials intelligence ecosystems propelled by machine learning (2020, background): “Interestingly, many of the ML-based potentials report low errors, on the order of 10 meV per atom and 0.3 eV Å −1 for energies and forces, respectively, at least for structures that are ‘close’ to the training set 126,152 .”
Machine-Learning Interatomic Potentials for Materials Science (2021, background): “In the recent tests of several ML potentials [24], the MTP potentials have shown the optimal combination of accuracy and computational efficiency.”
Deep Potentials for Materials Science (2022, background): “A comprehensive comparison of the major ML potentials in terms of accuracy and efficiency can be found in a recent paper [43].”
The MLIP package: moment tensor potentials with MPI and active learning (2020, methodology): “Probably because of this MTP together with GAP showed excellent accuracy in recent cheminformatics benchmark test [54] and interatomic potential test [82]; in the latter MTP showed also a very good balance between accuracy and computational efficiency when compared against other machine-learning pot”