The 9th International Conference on Multiscale Materials Modeling

講演情報

Symposium

D. Data-Driven and Physics-Informed Materials Discovery and Design

[SY-D1] Symposium D-1

2018年11月1日(木) 09:45 〜 11:00 Room8

Chair: Daniel Urban(Fraunhofer IWM, Germany)

[SY-D1] Data-Driven Discovery of new materials

Invited

Isao Tanaka (Dept. Materials Science and Engineering, Kyoto Univ. , Japan)

Challenges for accelerated discovery of materials with the aid of data science have been well demonstrated. One of the approaches uses materials database generated by density functional theory (DFT) calculations. A large number of DFT calculations with the accuracy comparable to experiments can be used for high throughput screening (real screening). Another approach called virtual screening uses machine-learning technique 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. 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 crystals can be shown as an example of the application of the virtual screening technique [1]. Another approach relies only on inorganic crystal structure database (ICSD) that collects literature data obtained mostly by experiments. We have demonstrated that matrix- and tensor-based recommender systems are very powerful for discovery of currently unknown chemically relevant compositions (CRCs) of inorganic compounds from vast candidates [2]. A Tucker decomposition recommender system shows the best discovery rate of CRCs. For ternary and quaternary compositions, approximately 60 and 50 of the top 100 recommended compositions are found to be CRCs, respectively. The high discovery rate with neither DFT database nor other prior physical/chemical knowledge should be noteworthy.

[1] A. Seko et al, Phys. Rev. Lett. 115, 205901 (2015).

[2] A. Seko et al, Phys. Rev. Mater. 2, 013805 (2018).