Graph Neural Network Driven Exploration of Non-Precious Metal Catalysts for Air-to-Ammonia Conversion.

Abstract

Efficient ammonia production directly from the air with minimal energy consumption remains one of the most challenging and ambitious scientific goals. NH₂OH has proven to be a promising stable intermediate in producing NH₃, with the direct generation of NH₃ from air achieved by coupling a continuous flow plasma reactor with an electrolyzer. However, the requirement of noble metal-doped Cu alloys as the cathode catalyst limits the scalability and cost-effectiveness of the coupled plasma-electrochemical system. In this work, graph neural networks, density functional theory calculations, and microkinetic modeling are combined to exhaustively explore the catalytic properties of all experimentally accessible alloy phases for NH₃ production, ultimately identifying the non-noble CuMnSb system as highly active for the conversion of air to NH₃. The experiments confirm an ammonia production rate of 28.47 mg h^-1 cm^-2 in a coupled plasma-electrolyser system. Such a finding confirms the future of machine learning and microkinetic theory in guiding the experimental exploration that transcends the constraints of conventional methods.