Revealing Li Staging Process in Graphite via a Genetic Algorithm Coupled with a Machine-Learning Interatomic Potential.

Graphite remains the dominant anode material in Li-ion batteries, yet a complete understanding of its Li intercalation mechanism, commonly described as staging, is still elusive. This difficulty arises from two fundamental challenges. First, the strong coupling between graphite stacking and Li ordering obscures the identification of consistent, state of charge (SOC)-dependent stable LixC structures, preventing a unified interpretation of the intercalation pathway. Second, identifying thermodynamically preferred LixC structures requires exploring a vast space of Li ordering and graphite stacking combinations, a task that is computationally prohibitive using density functional theory (DFT) alone. To overcome these limitations, we introduce a decoupling strategy that separates the Li intercalation process into local and global aspects. We first determine the relative stability of AA and AB stacking as a function of SOC, establishing critical SOC thresholds. These stacking conditions then guide the search for LixC structures to determine thermodynamically preferred Li ordering. Next, we employ a genetic algorithm (GA) coupled with a high-fidelity machine-learning interatomic potential (MLIP) trained on dispersion-corrected DFT data to sample Li configurations beyond DFT-accessible scales. This combined framework reveals a van der Waals (vdW)-driven unified mechanism in which the shifting balance among C-C, Li-C, and Li-Li interactions with increasing SOC governs the coupled evolution of stacking and Li ordering. Beyond resolving longstanding ambiguities in graphite staging process, this framework provides a rigorous foundation for predicting the performance and durability of graphite anodes.