Title: Material Design and Synthesis from First Principle Calculations and Machine Learning ~Fundamentals towards Application~ 

Abstract: Material informatics is introduced in terms of new methods for searching new materials. In particular, prediction of desired material synthesis and design is proposed on the basis of first principle calculations and machine learning [1]. Material big data is constructed based on density functional theory where every possible element combination is considered and then used as training sets for machine learning. The predicted material properties for common materials are successfully matched with experimental data. In addition, material combinations based on desired material properties are also able to be predicted. Thus, the proposed work flow becomes the bridge between the material database and designing materials. The approach enables efficient material mining from material big data and could potentially reveal undiscovered desired materials.

The constructed material database is then applied to searching for new functional materials. In particular, the following discovered materials are briefly introduced:

1.Iron and hydrogen generally do not bond under ordinary conditions. However, it was discovered that small Fe clusters are able to absorb large amounts of hydrogen. In addition, such clusters are found to be stable over graphene. Predicted hydrogenated Fe/graphene system is then experimentally performed for confirmation where synthesis and hydrogenation of the Fe/graphene system was achieved [2].

2.It was also discovered that the Fe/graphene system has high reactivities. Further density functional theory calculations predict that NO and H2 over FeO clusters results in the formation of NH3 and H2O without activation barriers [3]. Additionally, the Fe/graphene material is predicted to be able to lower the dehydrogenation temperature of MgH2. The catalytic effect of Fe/graphene towards MgH2 is also confirmed in experiment [4].

3.Lastly, newly two dimensional materials are explored. In particular, calculations predict that two dimensional tin, stanene, is found to be reactive against COX, SOX, and NOX gases. In particular, stanene is able to trap and dissociate those gases at low temperature [5]. Moreover, two dimensional Sn, SnSb, InSb, and InSn are found to be energetically stable. Two dimensional SnSb particularly has an outstanding hydrophobic effect while two dimensional InSb has a high antioxidant effect [6].

Thus, utilizing material big data and machine learning can accelerate material discovery.

 

Bio: Dr Keisuke Takahashi is a Researcher at the National Institute for Material Science, Ibaraki, Japan and also holds a Research Fellowship with the Japan Society for the Promotion of Science, Hokkaido University Japan. He obtained his Doctor of Philosophy in Engineering, Material Science and Engineering from Hokkaido University in 2014, Master of Science in Advanced Engineering Materials from Chalmers University of Technology, Gothenburg in 2011 and prior to that, a Bachelor of Science in Material Science and Engineering in 2008 from the University of Arizona, College of Engineering in Tucson, Arizona. His research interests include Computational Material Science (Density Functional Theory, Solidification Modelling), Material Informatics (Big Data, Machine Learning), Two Dimensional Materials, Heterogeneous Catalyst and Atomic Clusters.