Machine Learning-Driven Molecular Dynamics Study of LiF-SmF₃ Molten Salts

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• Lithium fluoride-samarium trifluoride (LiF-SmF3) systems are essential for Sm-Fe alloy production through molten salt electrolysis.
• Understanding their properties and local structures is vital for process optimization.
• Experimental measurements of high-temperature fluoride melts are time-consuming and expensive.

Purpose of the Study:

• To develop an accurate machine learning potential model for LiF-SmF3 systems.
• To efficiently obtain properties and structural features of these molten salt systems.
• To optimize the electrolytic preparation process of Sm-Fe alloys.

Main Methods:

• Development of a machine learning potential model for LiF-SmF3 systems.
• Validation of the model's accuracy against density functional calculations.
• Application of the model in molecular dynamics simulations (1100-1350 K).

Main Results:

• The machine learning model achieved high accuracy (96.92%) with low energy and force deviations.
• Calculated density and viscosity showed minimal deviations (1.03%-2.77% and 3.36%-4.58%) from experimental data.
• Ion self-diffusion coefficients and structural features (radial distribution function) were successfully computed.

Conclusions:

• The developed machine learning potential model provides an efficient and accurate approach for studying LiF-SmF3 systems.
• This computational method significantly reduces the need for extensive experimental measurements.
• The findings contribute to optimizing Sm-Fe alloy production through improved understanding of the molten salt medium.