Modeling fission product nucleation in molten NaCl using universal machine-learning potentials

The design of molten salt nuclear reactors (MSRs) requires a detailed understanding of the formation, precipitation, and transport of solid fission products in the fuel salt. However, the experimental characterization of these processes remains challenging and expensive. Molecular simulations can be leveraged to obtain insight into these systems. In particular, machine-learning potentials (MLPs) offer a promising path to obtain relevant insight at an acceptable computational cost. Here, we use a pre-trained foundation MLP to study Ru and Mo atoms in NaCl as a model system for solid fission products in MSR fuel salts. We find the MLP shows good agreement with ab initio calculations in terms of electronic structure and solvation of isolated atoms in the salt, and allows for sufficiently long simulations to observe their nucleation into 20-atom nanoparticles and study the local structure around them. We report the potential of mean force of dimer formation, the evolution of metal-metal coordination upon growth of larger clusters, and the solvation of these clusters in NaCl. This approach provides a practical framework for studying fission-product behavior in molten salts relevant to MSR design.