Liquid-Vapor Phase Equilibrium in Molten Aluminum Chloride (AlCl₃) Enabled by Machine Learning Interatomic Potentials

Area of Science:

• Materials Science and Engineering
• Computational Chemistry
• Chemical Engineering

Background:

• Molten salts are vital for clean energy technologies, but experimental determination of their thermophysical properties is challenging.
• Accurate knowledge of properties like vapor pressure is critical for safe operation of molten salt systems.
• Molecular simulations offer a scalable alternative to experimental methods for predicting these properties.

Purpose of the Study:

• To develop and validate machine learning interatomic potentials (MLIP) for simulating aluminum chloride (AlCl3) molten salt.
• To predict thermophysical properties, including temperature-surface tension correlations, liquid-vapor phase diagrams, and viscosities.
• To assess the performance of different MLIP architectures and training strategies for molten salt systems.

Main Methods:

• Development of two MLIP architectures: Kernel-based potential and neural network interatomic potential (NNIP).
• Training MLIPs using ab initio molecular dynamics (AIMD) data, including low-density configurations and PBE-D3 functional.
• Conducting two-phase coexistence simulations to determine phase diagrams and critical properties.
• Validation against experimental data, including Raman spectra and neutron structure factor, for molten salt structure.

Main Results:

• The NNIP accurately predicted critical temperature and density for AlCl3 within 3% and 7% of experimental values, respectively.
• Accurate correlations for temperature-surface tension and temperature-viscosity were established.
• MLIPs trained with PBE-D3 functional showed good agreement with experimental molten salt structure (Al2Cl6 dimers) and superior density-temperature correlation.

Conclusions:

• MLIPs, particularly NNIPs trained with appropriate data, are effective tools for predicting molten salt properties.
• Including low-density configurations in training is crucial for accurately representing wide phase-space behavior.
• This simulation approach can accelerate the screening of molten salts for nuclear reactors and improve safety assessments.