Neural Network Based Molecular Dynamics Simulations of Sputtering Processes

The interaction of plasma-facing materials (PFM) under the influence of energetic particles is one of the most important issues in a fusion reactor. Erosion reduces the thickness and stability of the PFM and contaminates the plasma which leads to radiative losses. In this paper we model this interaction by employing high-performance computing with a high-dimensional neural network potential energy function. As PFM we selected the technologically most important beryllium surfaces, subjected to energetic deuterium particles. By simulating this system in a non-cumulative scenario and varying several environmental parameters, we can calculate most relevant material properties.