Abstract

Applications of machine learning to materials data

Atsuto Seko
Kyoto University

Our applications of machine learning to materials data will be shown. The first application is the constructution of machine learning interatomic potentials (MLIPs). MLIP based on a large dataset obtained by density functional theory (DFT) calculation has great potential for improving the accuracy and transferability effectively. Therefore the MLIP has also been increasingly applied to a wide range of materials regardless of their type of chemical bonding. In this study, accurate MLIPs of 31 elemental metals using linearized MLIP frameworks will be shown [1].

The second application is the estimation of compound-property relationship. The descriptors of a compound play an essential role in constructing a machine-learning model of its physical properties. In this study, we propose a procedure for generating a set of descriptors from simple elemental and structural representations [2]. First, it is applied to a large data set composed of the cohesive energy for about 18000 compounds computed by DFT calculation. As a result, we obtain a kernel ridge prediction model with a prediction error of 0.041 eV/atom, which is close to the “chemical accuracy” of 1 kcal/mol (0.043 eV/atom). The procedure is also applied to two smaller data sets, i.e., a data set of the lattice thermal conductivity for 110 compounds computed by DFT calculation and a data set of the experimental melting temperature for 248 compounds.

The third application is the prediction of the relevance of chemical compositions where stable crystals can be formed, i.e., chemically relevant compositions (CRCs) [3]. Herein we adopt recommender system approaches to estimate CRCs. This approach significantly accelerates the discovery of currently unknown CRCs that are not present in the training database.

[1] A. Seko et al., Phys. Rev. B 90, 024101 (2014), A. Seko et al., Phys. Rev. B 92, 054113 (2015). A. Takahashi, A. Seko and I. Tanaka, arXiv:1708.02741.
[2] A. Seko et al., Phys. Rev. B 95, 144110 (2017)
[3] A. Seko, H. Hayashi and I. Tanaka, submitted.