IISc researchers use machine learning and computation to unveil mechanisms for CO2-to-fuel conversion

Researchers at India's prestigious Indian Institute of Science (IISc) have created a computational model that maps over 9,000 chemical reactions involved in transforming carbon dioxide (CO₂) into fuels and chemicals using hydrogenation over a copper catalyst. The study underscores the growing interest in CO₂ hydrogenation, where CO₂ reacts with hydrogen in the presence of a catalyst to produce valuable products such as methanol and carbon monoxide.

However, traditional modeling of these reactions, which involve numerous minute steps on the catalyst surface, poses significant computational challenges. Quantum mechanics simulations typically model only a limited number of likely reactions, potentially overlooking hundreds of critical processes.

To address this limitation, IISc scientists have introduced a data-driven computational framework that expands the initial network of 152 reactions, derived from quantum-mechanical simulations, to include thousands of additional reactions identified through machine learning. The researchers began by generating a comprehensive database of 152 reactions.

Subsequently, they employed machine learning models to rapidly predict activation energy barriers for additional reactions, as well as to predict all possible reactions involving 105 surface species and assess those occurring as single-step reactions. This approach resulted in an expanded network of 9,389 elementary reactions.

Initially, modeling the process with the smaller set of reactions led to inaccuracies, such as incorrectly predicting formic acid as the major product instead of methanol and underestimating CO₂ conversion rates. However, the inclusion of thousands of additional reactions in the expanded network resulted in predictions that aligned more closely with experimental observations.

The refined model predicted approximately a 40-fold increase in CO₂ conversion and accurately identified methanol and carbon monoxide as the primary products.

Furthermore, the larger network revealed a novel insight: hydrogen can transfer to reaction intermediates directly as a molecular H₂, rather than merely through individual hydrogen atoms. This finding, confirmed by quantum-mechanical calculations, runs counter to conventional understanding and could potentially enhance catalyst performance by strengthening interactions with H₂.

Co-author Shivam Chaturvedi noted that this pathway could be particularly favorable for hydrogen transfer to oxygen-containing intermediates, suggesting that catalysts interacting more strongly with H₂ might boost pathways leading to methanol production.