Designing high-performance catalysts is essential for cleaner energy technologies, but the behavior of multi-element, modern catalyst materials are difficult to predict. In this study, researchers at Tohoku University and international collaborators developed a collaborative framework that combines large language models with lab experiments to accelerate the discovery of high-entropy alloy catalysts for the oxygen reduction reaction, a key process in fuel cells.

The team developed a domain-specific AI assistant for high-entropy alloy (HEA) electrocatalysis called ChatHEA. ChatHEA helped extract knowledge from scientific literature, enumerate promising element combinations, guide experimental planning, and analyze catalytic activity data. Using this framework, 100 five-element high-entropy alloy catalysts were synthesized and evaluated through high-throughput experimentation -which saves time by testing multiple reactions at the same time.

The analysis revealed that catalytic activity is not simply determined by individual elements, but by synergistic interactions among element systems such as Fe-Co-Cu, Fe-Co-Ni, Pt-Ir, and Pt-Pd. Among the screened catalysts, FeCoCuPtIr showed excellent oxygen reduction activity and durability, outperforming commercial Pt/C in both electrochemical tests and fuel-cell device evaluation. The FeCoCuPtIr-based fuel cell achieved a peak power density of 0.789 W cm⁻²

"The U.S. Department of Energy sets standards for minimum activity levels that fuel cells should ideally operate at, and we are happy to report that our fuel cell exceeded the 2025 activity target," says Distinguished Professor Hao Li (Advanced Institute for Materials Research (WPI-AIMR))

Further theoretical calculations and pH-dependent microkinetic modeling showed that multi-element synergy optimizes the electronic structure of active sites and improves the adsorption strength of key reaction intermediates. This work provides not only a promising fuel-cell catalyst, but also a general AI-driven strategy for discovering complex materials more efficiently.

"ChatHEA, was not used only as a prediction tool," says Li. "Instead, it supported the full research workflow, including literature knowledge extraction, element-combination design, experimental planning, data processing, and mechanistic analysis."

This research introduces an AI-guided approach to accelerate the discovery of advanced catalysts. This research may contribute to cleaner energy technologies, including hydrogen fuel cells for vehicles, backup power systems, and future low-carbon energy infrastructure. More efficient catalysts could help reduce the amount of precious metals needed and support the development of more affordable and sustainable energy devices.

The findings were published in National Science Review on March 14, 2026.