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

Area of Science:

• Nuclear Engineering
• Materials Science
• Computational Chemistry

Background:

• Molten salt nuclear reactors (MSRs) require understanding solid fission product behavior in fuel salts.
• Experimental characterization of fission product formation, precipitation, and transport is challenging and costly.
• Molecular simulations offer a viable alternative for gaining insights into these complex processes.

Purpose of the Study:

• To investigate the behavior of ruthenium (Ru) and molybdenum (Mo) atoms in NaCl as a model system for MSR fuel salts.
• To leverage machine-learning potentials (MLPs) for efficient and accurate molecular simulations of fission products.
• To provide a computational framework for studying fission product behavior relevant to MSR design.

Main Methods:

• Utilized a pre-trained foundation machine-learning potential (MLP) for molecular dynamics simulations.
• Studied Ru and Mo atoms in NaCl, comparing MLP results with ab initio calculations for electronic structure and solvation.
• Performed long-duration simulations to observe nucleation of 20-atom nanoparticles and analyze their local structure and solvation.

Main Results:

• The MLP accurately reproduced ab initio results for isolated Ru and Mo atoms in NaCl.
• Observed nucleation of Ru and Mo atoms into 20-atom nanoparticles within the simulated salt.
• Characterized the potential of mean force for dimer formation, metal-metal coordination in clusters, and cluster solvation.

Conclusions:

• Machine-learning potentials provide a computationally efficient and accurate method for studying fission product behavior in MSR fuel salts.
• The simulation approach successfully captured key phenomena such as nanoparticle nucleation and solvation.
• This framework facilitates detailed analysis of fission product behavior, aiding in the design and safety of MSRs.