Transferable Deep Learning Potential Reveals Intermediate-Range Ordering Effects in LiF-NaF-ZrF₄ Molten Salt

Area of Science:

• Materials Science
• Computational Chemistry
• Nuclear Engineering

Background:

• Multicomponent molten salts like LiF-NaF-ZrF4 are crucial for advanced clean energy systems due to their excellent thermophysical and transport properties.
• Understanding the complex, disordered structures of these molten salts and their relation to composition is challenging with traditional simulation methods.
• Previous simulations were limited in scale and accuracy, hindering the design of tailored molten salt compositions.

Purpose of the Study:

• To develop accurate and efficient machine learning potentials for simulating LiF-NaF-ZrF4 molten salts.
• To predict structures and properties beyond the first coordination shell in a wide range of compositions.
• To overcome the limitations of ab initio and classical models in capturing extended heterogeneous structures.

Main Methods:

• Utilized accurate, efficient, and transferable machine learning potentials (neural networks).
• Trained models on eutectic compositions (29% and 37% ZrF4) to simulate broader compositional ranges (11-40% ZrF4).
• Employed theoretical Raman spectroscopy calculations to validate simulated structures and analyze spectral features.

Main Results:

• Machine learning potentials accurately simulated LiF-NaF-ZrF4 structures across various compositions, showing different coordination chemistries.
• Simulations demonstrated remarkable agreement with experimental and theoretical Raman spectra.
• Identified a novel shift and flattening of the bending band around 250 cm-1 in higher ZrF4 content compositions, validating simulated extended structures.

Conclusions:

• Machine learning potentials provide a powerful tool for accurately predicting the structure and ionic diffusivities of LiF-NaF-ZrF4 molten salts at larger scales.
• This approach overcomes previous simulation limitations, enabling the design of tailored molten salt compositions for clean energy applications.
• The findings facilitate a deeper understanding of structure-property relationships in disordered molten salt systems.