A large language models-guided grand canonical DFT framework for accelerating the discovery of efficient electrocatalysts

Electrocatalysis plays a critical role in enabling sustainable energy conversion. However, current electrocatalysts rely on either scarce precious metals or transition metals with limited efficiency, and their design strategies lack systematic frameworks for precise optimization guidance. Here, we propose an advanced collaborative framework that integrates large language models (LLMs) with grand-canonical density functional theory (GCE-DFT) calculations, thereby accelerating the discovery of high-performance and cost-effective electrocatalysts. As a validation, this framework was applied to the promising Fe-N-C oxygen reduction reaction (ORR) catalysts, demonstrating that asymmetric FeN4 coordination with mixed pyrrolic (pyrr-N) and pyridinic (pyri-N) nitrogen environments serves as a critical determinant of catalytic activity and stability. Furthermore, the integration of GCE-DFT calculations with in situ X-ray absorption spectroscopy demonstrated that the asymmetric FeN4 coordination in Fe-N-C eletrocatalysts induces symmetric electron density redistribution at the Fe active sites, thereby significantly enhancing both catalytic activity and stability. Based on these insights, we synthesized asymmetric N-coordinated FeN4 site in Fe-N-C electrocatalysts via templating-assisted pyrolysis for acidic ORR applications, achieving superior cycle stability (>90% activity retention after 30,000 cycles) while maintaining high activity. This breakthrough successfully addresses the long-standing trade-off riddle between stability and activity in acidic ORR systems. Overall, the proposed framework significantly accelerates the discovery of high-performance and cost-effective electrocatalysts by integrating data-driven artificial intelligence with atomistic-level simulations.