Machine Learning-Driven Exploration of Composition- and Temperature-Dependent Transport and Thermodynamic Properties in LiF-NaF-KF Molten Salts for Nuclear Applications

This study developed a high-precision deep potential (DP) model based on density functional theory (DFT) and the DP-GEN workflow to efficiently simulate the microscopic structures and thermophysical properties of LiF-NaF-KF molten salt systems with varying compositions. Through iterative optimization of the training data set using the DP-GEN active learning strategy, our DP model demonstrated excellent agreement with DFT calculations in predicting energies, forces, and stresses. Leveraging this model, we systematically investigated the local structures and properties of 22 FLiNaK molten salt compositions, including radial distribution functions (RDFs), coordination numbers (CNs), density (ρ), heat capacity (<i>C</i>_p), self-diffusion coefficients (SDCs), electrical conductivity, and shear viscosity. The analysis revealed that Li-F ion pairs exhibit the strongest localized coordination, with the coordination numbers of all cations increasing with higher LiF content. Density was found to be primarily governed by NaF concentration, showing a positive correlation with NaF content. Viscosity was significantly influenced by both temperature and composition, decreasing notably with increasing temperature - the viscosity of the eutectic composition decreased from 3.933 mPa·s at 873 K to 1.622 mPa·s at 1073 K. Higher KF content led to lower viscosity due to the weaker interactions of K-F ion pairs. Additionally, noneutectic compositions with high LiF content (e.g., 80% LiF) exhibited significantly superior <i>C</i>_p compared to the eutectic system. This work elucidates the regulatory effects of composition and temperature on the structure-property relationships of molten salts, expands the property database for noneutectic FLiNaK systems, and provides a theoretical foundation for the design and performance optimization of molten salt reactor materials.