Impact of the glass transition temperature on Li-ion diffusion in amorphous alloy anodes via machine learning

Understanding Li diffusion in amorphous alloy anodes is critical for developing fast-charging batteries with improved rate performance and cycling stability. However, the complex structural evolution involving amorphous phases makes the atomic diffusion mechanism unclear. Here, we construct high-precision machine-learning potential for Li–Si/Ge/Sn/Pb systems to investigate Li diffusivity ( D Li ). Our simulations show that D Li increases with Li content and temperature, aligning with experimental trends. Conventional Arrhenius extrapolation from high-temperature data overestimates the D Li obtained from direct room-temperature simulations by 1–2 orders of magnitude, particularly in Li-poor phases. We identify the glass transition temperature ( T g ) as a key descriptor of the change in diffusion behavior. Above T g , the systems exhibit liquid-like dynamics with structural rearrangement and rapid Li motion. Below T g , they exhibit glass-like dynamics with slow structural relaxation and localized Li motion. We employed three dynamical analyses and found that Li-poor phases exhibit stronger Li localization and slower relaxation, whereas Li-rich phases exhibit broader displacements and faster relaxation, supporting a gradual transition from predominantly localized, cage-confined Li motion to more active and extended Li displacements across T g . These results establish a composition–local-structure–diffusion relationship, indicating that optimizing alloy composition to reduce T g may improve Li transport, offering theoretical guidance for developing next-generation battery anodes.