Data-driven methods in materials modeling
Yury Lysogorskiy, ICAMS, Ruhr-University Bochum, Bochum, GermanyThomas Hammerschmidt, ICAMS, Ruhr-Universität Bochum, Bochum, DeutschlandRalf Drautz, ICAMS, Ruhr-Universität Bochum, Bochum,
The prediction of materials properties is critical for the development of many technologically relevant materials. At the atomistic scale many properties can be determined with high accuracy by density functional theory (DFT) calculations. However, the computational cost sets limits to high-throughput screening for new materials and to calculations of more complex but technologically relevant materials properties. These limitations may be overcome in part by simplified models that are orders of magnitude faster than DFT. The main challenge for setting up simplified models is their construction and validation. In this work we present a data-driven approach to analyze and set up simplified models for two applications. First, empirical interatomic potentials are validated against a high-throughput generated DFT reference database. Second, physically-inspired atomic descriptors are used in combination with machine learning for the prediction of formation energies and band gaps of transparent conductors.