Machine-learning-assisted discovery of a stable Li3As2 intermediate phase in the Li-As binary system and its electrochemical implications

Abstract

Alloy-type anodes offer high capacity, but deep lithiation usually causes severe volume expansion and structural instability. Identifying a thermodynamically stable intermediate phase with a moderate lithiation potential may offer a viable route to mitigating this dilemma. Here, we propose a hierarchical computational workflow that combines global structure search with a machine-learning interatomic potential to systematically explore phase evolution in the Li-rich region of the Li-As binary system. Fine-tuning a pretrained potential on a dataset of approximately 3500 configurations labeled by density functional theory (DFT) yields an energy root-mean-square error (RMSE) of 25.3 meV/atom on an independent test set. Structure screening with the fine-tuned (FT) potential, followed by DFT validation, reveals a thermodynamically stable intermediate phase, C2/c-Li₃As₂, lying on the convex hull between LiAs and Li₃As. Phonon and electronic-structure calculations show that this phase is dynamically stable at 0 K and metallic. Electrochemical thermodynamic analysis indicates an equilibrium potential of about 0.95 V (vs. Li/Li + ) for the LiAs ⇌ Li₃As₂ two-phase reaction. When lithiation is limited to Li₃As₂, the theoretical capacity reaches 536.6 mAhg^-1 with a volume expansion of about 68.6%. Further climbing-image nudged elastic band (CI-NEB) calculations and molecular dynamics (MD) simulations show a low Li + migration barrier. These results identify C2/c-Li₃As₂ as a promising intermediate phase for shallow-lithiation strategies and highlight its potential for fast-charging anodes.