Machine-learning-accelerated mechanistic exploration of interface modification in lithium metal anode

Although the electrode-electrolyte interface is a crucial electrochemical region, the comprehensive understanding of interface reactions is limited by the time and space scales of experimental tools. Theoretical simulations with this delicate interface also remain one of the most significant challenges for atomistic modeling, particularly for the stable long-timescale simulation of the interface. Here we introduce a novel scheme, hybrid ab initio molecular dynamics combined with machine learning potential (HAML), to accelerate the modeling of electrode-electrolyte interface reactions. We demonstrate its effectiveness in modeling the interfaces of Li metal with both liquid and solid-state electrolytes, capturing critical processes over extended time scales. Furthermore, we reveal the role of interface reaction kinetics in interface regulation through HAML simulations, combined with the similarity analysis method. It is demonstrated that element (Se, F, O) doping in the Li₆PS₅Cl system is an effective strategy for enhancing interface reaction kinetics, facilitating the formation of a more stable interface protective layer faster at room temperature. Moreover, moderate structural instability can positively contribute to interface stabilization. HAML offers a promising approach for addressing the challenge of designing stable interfaces while reducing computational costs. This work provides valuable insights for advancing the understanding and optimization of interface behaviors in Li metal batteries.