Artificial Intelligence‐Assisted Ultrafast High‐Throughput Screening of High‐Entropy Hydrogen Evolution Reaction Catalysts

This study overcomes high-entropy alloy (HEA) catalyst design challenges using AI-driven methods. Large language models analyzed 14 k papers to identify 10 key HER-active elements, narrowing candidates to 126 Pt-based HEAs. Genetic algorithms optimized synthesis via ultrafast thermal shocks, converging in 4 iterations (24 samples). This AI-autonomous framework reduces discovery time from millennia to hours, enabling efficient multi-catalyst design for energy applications.

Abstract

The development of high-entropy alloy (HEA) catalysts is hindered by the “combinatorial explosion” challenge inherent to their complex component design. This study presents an artificial intelligence-assisted high-throughput framework that synergizes large language models (LLMs) for literature mining and genetic algorithms (GAs) for iterative optimization to overcome this challenge. Here, LLMs analyzed 14 242 publications to identify 10 critical hydrogen evolution reaction (HER)-active elements (Fe, Co, Ni, Pt, etc.), narrowing the candidate pool to 126 Pt-based HEA combinations. GA-driven experiment optimizes this subset via ultrafast high-throughput material synthesis and screening using ultrafast high-temperature thermal shock technology, achieving convergence in 4 iterations (24 samples) for 60% reduction of the versus conventional GA approaches. The optimal IrCuNiPdPt/C catalyst exhibits the record-low HER overpotentials of 25.5 and 119 mV at 10 and 100 mA cm⁻², surpassing commercial Pt/C by 49% and 18%, respectively, which demonstrates 300-h stability with negligible decay. This work establishes a paradigm-shifting strategy bridging computational intelligence and autonomous experiment, that slashes the discovery time from millennia to hours, enabling rational design of multi-component catalysts for sustainable energy applications.