Molecular Dynamics Study on Deep Learning Potential of the (LiF-YF₃)_eut.-Y₂O₃ Molten Salt System

In this study, a machine learning workflow via the Deep Potential Generator (DPGEN) was then used to train a molecular dynamics potential function for the (LiF-YF₃)_eut.-Y₂O₃ molten salt system. Radial distribution functions (RDFs) and angular distribution functions (ADFs) of expanded systems were calculated to analyze the dynamic evolution of cluster structures within the system. The results show that the deviation between the system density calculated using the trained potential function and the experimental value is no more than 3.0%. Within the temperature range of 1173-1353 K, the primary coordination numbers of F^- ions for Li-F and Y-F ion pairs in the (LiF-YF₃)_eut. system are 4-5 and 6-7, respectively. Consequently, the two relatively stable short-range dominant clusters, [YF₆]^3- and [YF₇]^4-, account for over 80% of the total clusters and belong to distorted octahedral structures. After saturated Y₂O₃ was dissolved in the (LiF-YF₃)_eut. system, local O^2- ions compete with F^- ions for binding to Li⁺ and Y³⁺, forming Li-O and Y-O coordinated ion pairs. Each O^2- ion replaces two F^- ions, ultimately generating [YOF_x]^n--type ionic clusters. The dominant forms of these clusters are [YOF₄]^3-or [YOF₅]^4-, which together account for 75% of the total clusters.