Machine learning molecular dynamics simulations of coordination and diffusion behaviors in lithiated gallium electrodes.

Liquid gallium (Ga) has emerged as a promising anode material for flexible lithium-ion batteries owing to its exceptional fluidity, intrinsic self-healing capability, and high theoretical capacity. However, the understanding of structure and transport properties across diverse Li-Ga alloy (LGA) phases formed during lithiation remains limited. Here, we develop machine learning force fields (MLFFs) for four experimentally identified LGAs (Li3Ga14, Li2Ga7, LiGa, and Li2Ga) and perform large-scale molecular dynamics simulations to investigate local coordination and diffusion behaviors. Our simulations reveal a lithiation-induced evolution of Li local environments from Ga-dominated coordination shells in Li3Ga14 and Li2Ga7 to Li-rich networks in LiGa and Li2Ga. Polyhedral template matching further indicates that all four LGA phases remain predominantly disordered, while the fraction of short-range ordered motifs increases upon lithiation. Consistently, Li exhibits liquid-like mobility in Li3Ga14 and Li2Ga7, but strongly localized, solid-like dynamics in LiGa and Li2Ga. The Li diffusion coefficient in Li3Ga14 (4.46 × 10-11 m2 s-1) is nearly an order of magnitude higher than that in Li2Ga7 (3.14 × 10-12 m2 s-1), primarily due to the weaker interactions between Li and the surrounding Li/Ga in the former system. Finally, van Hove analysis and trajectory visualizations uncover intermittent residence-jump (hopping-like) dynamics in Li2Ga7 and Li2Ga. Overall, our findings clarify the structure-diffusion relationship across different LGAs and offer important theoretical insights into the structural evolution of Ga-based anodes during the lithiation process.