AI-Assisted Discovery of Catalytic Systems for Carbon Capture and Conversion

Abstract

The urgent need for efficient carbon capture and conversion technologies has outpaced traditional catalyst discovery methods. This study presents a multimodal artificial intelligence framework that accelerates the discovery of high-performance catalytic systems by integrating quantum chemistry simulations, robotic experimentation, and deep learning. Our approach combines graph neural networks for catalyst structure prediction, reinforcement learning for reaction pathway optimization, and automated high-throughput experimentation for validation. Testing across three catalytic platforms metal-organic frameworks (MOFs), transition metal complexes, and single-atom catalysts - revealed 17 novel high-performance candidates with CO2 conversion efficiencies exceeding 85% at ambient conditions. The AI models achieved 92% accuracy in predicting catalytic activity (RMSE = 0.18 eV) and reduced discovery time by 98% compared to conventional methods. Particularly promising was a nickel-graphene single-atom catalyst identified through this approach, demonstrating 94% Faradaic efficiency for CO2-to-ethylene conversion at -0.8 V vs RHE. This research establishes a paradigm for AI-driven catalyst discovery that simultaneously addresses activity, selectivity, and scalability challenges in carbon mitigation technologies.