Real and virtual screening for materials discovery through first principles calculations

Isao Tanaka
Kyoto University

Recently, challenges for accelerated discovery of materials with the aid of data science have been well demonstrated. One of the approaches uses materials database that is generated by first principles density functional theory (DFT) calculations. Thanks to recent progress of computational power and technique, a large number of DFT calculations can be made with the accuracy comparable to experiments, which can be used for “real screening”. Needless to say, however, a DFT calculation requires an initial crystal structure, although atomic positions can be locally optimized. In order to construct a database by DFT calculations, we need to have a priori knowledge of their structures as in ICSD (Inorganic Crystal Structure Database). Alternatively, we need to restrict our search space within prototype structures. Here we will show an example to discover novel Sn(II)-based oxide compounds suitable for daylight-driven photocatalysis using prototype structures registered in ICSD and real screening[1].

Another approach called “virtual screening” uses machine-learning techniques to select predictors for making a model to estimate the target property. The whole library can then be screened. Verification process is generally required to examine the predictive power of the model. Models and the quality of the screening can be improved iteratively through Bayesian optimization process. The virtual screening is useful when real screening based upon the DFT data is not practical, i.e. when the computational cost for the descriptors is too high to cover the whole library within the practical time frame. This is the same if one needs to explore too large space to cover exhaustively. Discovery of new low lattice thermal conductivity (LTC) crystals through the virtual screening technique[2] will be shown as an example. We have established our own dataset of LTC computed by the first principles anharmonic force constant method as implemented in phonopy[3], an open source package for first principles phonon calculations. Candidates found by the virtual screening are validated by first principles LTC calculations.

[1] H. Hayashi, S. Katayama, T. Komura, Y. Hinuma, T. Yokoyama, F. Oba, I. Tanaka, Adv. Sci. (2016) in press.
[2] A. Seko, A. Togo, H. Hayashi, K. Tsuda, L. Chaput, and I. Tanaka, Phys. Rev. Lett. 115, 205901 (2015).
[3] A. Togo, and I. Tanaka, Scr. Mater., 108, 1-5 (2015)

Presentation (PDF File)

Back to Workshop I: Machine Learning Meets Many-Particle Problems