Accelerating molecular dynamics simulations in V2O5

A standard method to calculate diffusion coefficients relies on molecular dynamics (MD) simulations performed at different temperatures. This method doesn’t require any a priori knowledge of the diffusion pathways, in contrast to methods such as the nudged-elastic band (NEB) method, where initial and final positions need to be known. However, a typical MD simulation has to run long enough for multiple diffusion events to be observed. This requires simulations performed over long time scales, especially at lower temperatures. 

To accelerate such MD simulations, we implemented a color-diffusion method within the Atomic Simulation Environment (ASE), in which an external biasing (“color”) field promotes the motion of the diffusing species. As a test case, we use a machine-learned foundation model (MACE) on vanadium pentoxide (V2O5), which is a promising multi-valent battery cathode material with a layered structure. 

For similar simulation times, we observe significantly more diffusion events, indicating that this method can indeed lead to a speedup of MD simulations.