Atomic-scale tracking of sodium metal-electrolyte reactions via adaptive machine learning force fields

Sodium (Na) metal anodes offer high capacity and low cost, making them promising for high-energy-density batteries, but their unstable solid electrolyte interphase (SEI) leads to dendrite growth and capacity loss, with formation mechanisms still unclear. While ab initio molecular dynamics (AIMD) can capture interfacial reactions, its limited ps-level timescale falls short of the ns–μs range required for SEI formation.

Recently, researchers Prof. Dawei Zhang and Prof. Xiang Chen developed an accelerated on-the-fly learning (AOFL) method by combining conventional on-the-fly learning force fields (COFL) with similarity structure screening to tackle these challenges, enabling the simulations to achieve real-time adaptation and high efficiency.

The team's research demonstrated that the AOFL approach achieves a 71% speedup over AIMD while retaining comparable accuracy, and the self-adaptive mechanism of AOFL approach enhances DFT sampling during reactive events, ensuring AIMD-level accuracy in capturing SEI formation.

Using the AOFL and AIMD methods, the team was able to identify key inorganic components of the SEI, including Na₂O and NaOH, which play a crucial role in influencing the stability of the interface.

The adaptive machine learning approach bridges the gap between quantum mechanical accuracy and empirical force-field efficiency. The advancement enables the simulation of reactive behaviors at extended timescales and in complex systems, thereby elucidating the formation mechanism of early-stage SEI at Na metal interfaces. These findings lay a solid foundation for understanding SEI formation and electrolyte design.

---

---