Place: DPG Spring Meeting, Berlin, Germany
Alvin Noe Ladines
Interatomic potentials provide a means to simulate extended length and time scales that are outside the reach of ab-initio calculations. However, the transferability of interatomic potentials to atomic environments which were not included in the parameterization of the potential is often unknown and a (re-)parameterization usually very time consuming. We suggest a data-driven approach to overcome these problems for the development of tight-binding (TB) based bond-order-potentials (BOP). The parameterization is mapped to a multi-dimensional optimization problem, which is initialized by a comprehensive database of TB parameters across the periodic table. Using a 2d structure map we systematically sample the possible atomic environments during the parameterization process. This also allows us to predict and analyse the transferability to structures which were not included in the parameterization process. We demonstrate that our method accelerates the development of BOPs with a transparent parameterization process and at the same time provides a direct and quantitative measurement of transferability.