Fitting Accurate Potentials

Dr Albert Bartok-Partay,  Leverhulme Early Career Research Fellow, Department of Engineering, University of Cambridge, Cambridge, UK presents:

In this presentation an approach to utilizing machine learning to fit accurate atomic potential energy surfaces based on ab initio data will be presented. The necessary ingredients will be discussed: representing atomic environments, fitting with Gaussian processes, and building databases of atomistic configurations. Finally results from experiments on silicon, tungsten, and water will be presented.