The Little 4f Electron that Could:An Active Learning Approach to Model Cerium at Extreme Conditions

Cerium has a wide range of unique crystallographic phases including a pressure induced iso-structural phase transition and a liquid-liquid phase transition. Cerium also has a wide range of stable hydride concentrations that can significantly vary in mechanical properties. However, modeling these materials under a single electronic structure method that captures all the necessary physics has an extreme computational cost associated. Therefore, we utilize a data-driven approach to develop a machine learned, classical MD forcefield that captures these processes. An ensemble based, active learning approach is used to design the training set in a way that minimizes the necessary data and computational cost. Physics-informed choices in the initial bootstrapping allowed for process optimization. The resulting model was used to run shock simulations in polycrystalline Cerium at the iso-structural phase transition point, as well as shocks in Cerium Hydride for different hydrogen concentrations.