Revealing the diffusion mechanism of Cs in amorphous and polycrystalline SiC by actively trained moment tensor potentials

The diffusion mechanism of Cs in high energy grain boundaries (HEGBs) of silicon carbide (SiC) remains unsolved due to the lack of reliable and computationally efficient for long time-scale diffusion simulations interatomic potentials. Constructing machine learning interatomic potentials (MLIPs) for complex HEGB structures is challenging as their sizes exceed ab initio computational capacity. Therefore, we proposed a workflow to develop moment tensor potentials (MTPs), which facilitates the extraction of unknown regions from HEGBs and enables efficient active training. Our developed MTPs allowed us to perform simulations of Cs diffusion in amorphous SiC and HEGBs. Simulations show that Cs diffuses following a cage-breaking mechanism. Radial distribution functions, Voronoi analysis and electronic structure calculations further elucidate local atomic environments and weak interactions governing Cs mobility. This work provides a generalizable workflow to train MLIPs for complex structures and atomic insights for modeling fission product behaviors.