Reinforcement learning-guided long-timescale simulation of defect diffusion in solids

Atomic diffusion in solids is an important process in various phenomena. However, atomistic simulations of diffusion processes are confronted with the timescale problem: the accessible simulation time is usually far shorter than that of experimental interests. In this work, we developed a long-timescale method using reinforcement learning that extends simulation capability to match the duration of experimental interest. As a testbed, we simulate hydrogen diffusion in pure metals and a medium entropy alloy, CrCoNi. The algorithm can derive hydrogen diffusivity reasonably consistent with previous experiments. The algorithm can also recover counter-intuitive H-V cooperative motion. We also demonstrate that our method can accelerate the sampling of low-energy configurations compared to the Metropolis-Hastings algorithm using hydrogen migration to copper (111) surface sites as an example.