Modeling LiF and FLiBe Molten Salts with Robust Neural Network Interatomic Potential

Area of Science:

• Materials Science
• Computational Chemistry
• Nuclear Engineering

Background:

• Lithium-based molten salts, including lithium fluoride (LiF) and Flibe (66.6%LiF-33.3%BeF2), are critical for energy storage, advanced fission, and fusion reactors.
• Their complex structures and thermodynamic conditions pose challenges for traditional atomistic modeling.
• Accurate simulation is essential for optimizing their performance in demanding nuclear environments.

Purpose of the Study:

• To develop and validate a fast and accurate atomistic modeling method for lithium-based molten salts.
• To assess the capabilities of atom-centered neural network interatomic potentials (NNIPs) for simulating LiF and Flibe.
• To demonstrate the scalability and efficiency of NNIPs for large-scale molecular dynamics.

Main Methods:

• Development and application of atom-centered neural network interatomic potentials (NNIPs).
• Comparison of NNIP-based molecular dynamics (MD) with ab initio MD simulations.
• Validation against experimental and theoretical data for LiF and Flibe across various phases and conditions.

Main Results:

• NNIPs accurately reproduce ab initio interactions for LiF dimers, crystalline solids, and liquid phases.
• NNIPs effectively predict the structures and dynamics of Flibe under normal and high-temperature-pressure conditions.
• NNIP-based MD achieved over 3 orders of magnitude speedup compared to ab initio methods, with scalability to large system sizes and long timescales.

Conclusions:

• Atom-centered NNIPs offer a computationally efficient and accurate alternative to ab initio MD for molten salt simulations.
• This method significantly accelerates the study of nuclear materials, enabling exploration of complex systems.
• NNIPs provide a powerful tool for advancing the design and application of molten salts in energy and nuclear technologies.