Molecular Dynamics Modeling of Plasma Material Interactions using Machine Learned Interatomic Potentials

Multiple materials, namely tungsten and beryllium, will be present in future fusion reactors as plasma facing components. Experiments of beryllium implantation in tungsten indicate the formation of W-Be intermetallics on the surface, which can affect the performance of the tungsten divertor. Molecular dynamics (MD) can provide insight into physical processes related to experimental observations. However, MD is limited by the accuracy of the interatomic potential used. Recently, machine learning methods are being used to develop more accurate, quantum-informed potentials.
In this work, we have developed a machine learned Spectral Neighbor Analysis Potential (SNAP) for W-Be and used this potential to study beryllium implantation in tungsten. An amorphous W-Be layer forms within the first 50 ns that is limited to the near surface region and an exchange mechanism that allows tungsten to migrate from the substrate into the amorphous layer was observed. Ordered structures similar to expected intermetallic configurations were seen within the amorphous layer. These results show the early stages of a surface W-Be layer that can provide further insight into the formation of the intermetallics observed in experiments.