Machine learned force fields for molecular dynamics simulation of fusion materials

The development of fusion reactors depends on a combination of the insight from multiple disciplines at a wide range at length and timescales. An important part of any future fusion reactor are materials that are exposed to extreme radiation and temperature conditions as plasma facing components. While a significant amount of research has focused on refractory alloys, another class of promising materials are ultra-high temperature ceramics as promising candidates. We investigate TiB2 as a ceramic material for plasma-materail interaction simulations using molecular dynamics. To model microstructure evolution and surface erosion, we perform a combination of first principles static and molecular dynamics (MD) simulations and derive machine learned (ML) force fields with the goal to achieve DFT like accuracy in classical MD simulations. We will present our ML force fields for TiB2 and Deuterium and compare the accuracy and performance of classical MD simulations using these potentials to first principles DFT simulations and comparison against temperature dependent erosion experiments.