Machine Learned Interatomic Potentials for Modeling Plasma Material Interactions

Molecular dynamics has been widely used to study many different material systems but the accuracy of the results is limited to the interatomic potential used. Recently, machine learning has been used to develop more quantum accurate potentials. Once such method, the Spectral Neighbor Analysis Potential (SNAP), uses a database of various structural configurations using quantum methods to generate a potential. This method has been applied to materials used as fusion reactor components, namely tungsten and beryllium. This talk will describe the potential fitting method as well as results relevant to studying the performance of materials in the extreme environment within a fusion reactor.