Prediction of the stable crystal structure for multinary (ternary or higher) compounds with unexplored compositions demands fast and accurate evaluation of free energies in exploring the vast configurational space. The machine-learning potential such as the neural network potential (NNP) is poised to meet this requirement but a dearth of information on the crystal structure poses a challenge in choosing training sets. Herein we propose constructing the training set from density functional theory (DFT)–based dynamical trajectories of liquid and quenched amorphous phases, which does not require any preceding information on material structures except for the chemical composition. To demonstrate suitability of the trained NNP in the crystal structure prediction, we compare NNP and DFT energies for , , , and over experimental phases as well as low-energy crystal structures that are generated theoretically. For every material, we find strong correlations between DFT and NNP energies, ensuring that the NNPs can properly rank energies among low-energy crystalline structures. We also find that the evolutionary search using the NNPs can identify low-energy metastable phases more efficiently than the DFT-based approach. By proposing a way to developing reliable machine-learning potentials for the crystal structure prediction, this work paves the way to identifying unexplored multinary phases efficiently.
20 August 2018 in The Journal of Physical Chemistry C
Owing to their simultaneous accuracy and computational efficiency, interatomic potentials machine-learned using first-principles calculation data are promising for investigating phenomena closely related to atomic motion in various energy materials. We have been working with one type of these potentials, high-dimensional (HD) neural network potentials (NNPs), and their applications, but we realized that our current understanding of HD NNPs, e.g. the meaning of the atomic energy mapping, remained insufficient, and that tuning their prediction performance for different target properties/phenomena often requires much trial and error. In this article, we illustrate the usefulness of NNPs through our studies on ion migration and thermal transport in energy and related materials. We also share our experiences with data sampling and training strategies and discuss the meaning of atomic energy mapping in HD NNPs.
29 June 2020 in The Journal of Physical Chemistry Letters