Published in Nature Communications, the study could help scientists develop more effective catalysts for converting CO₂ into useful products such as methanol. The research also revealed surprising ways in which hydrogen participates in these reactions, challenging established assumptions about how the process works.

One promising approach to recycling carbon dioxide is hydrogenation. This process uses catalysts to react CO₂ with hydrogen and produce chemicals such as methanol. However, it involves a vast number of individual reactions occurring across the catalyst’s surface.

Because simulating every possible step using quantum mechanics requires enormous computational power, researchers typically focus on a limited selection of reactions considered to be the most important. This simplification can leave hundreds of potentially significant reactions out of the model.

“We started with a concern that is familiar to anyone who does mechanistic modelling: How do we know that our reaction networks aren’t skipping one critical step?” said lead author Anand Mohan Verma. He conducted the research as a CV Raman Postdoctoral Fellow at IISc and is currently an Assistant Professor at Motilal Nehru National Institute of Technology Allahabad (MNNIT Allahabad).

Machine learning uncovers thousands of overlooked chemical reactions

To overcome this limitation, the research team first used quantum mechanical simulations to build a detailed database containing 152 chemical reactions. They then trained a machine learning model to quickly estimate the energy barriers associated with additional reactions.

The scientists used automated computational tools to investigate reactions involving 105 different species present on the catalyst surface. They also determined which transformations could occur in a single reaction step. This approach expanded the original network from 152 reactions to 9,389.

The difference was significant. When the researchers relied on the smaller network, their model incorrectly identified formic acid as the main product instead of methanol. It also predicted substantially lower CO₂ conversion than was observed in experiments.

“When we first modeled the process using the 152 reactions we considered, the network incorrectly predicted formic acid rather than methanol as the main product and underestimated the amount of carbon dioxide that would be converted,” explained corresponding author Ananth Govind Rajan, associate professor in IISc’s Department of Chemical Engineering. “It was only when we expanded the network to include thousands of additional reactions that were previously overlooked that our predictions matched what we and others observed experimentally.”

Expanded model predicts 40 times more CO₂ conversion

When the scientists incorporated the larger reaction network into models of the rates of the chemical processes, the predicted amount of CO₂ conversion increased by approximately 40 times compared with the previous model.

The expanded model also correctly identified methanol and carbon monoxide as the major products, and its predictions matched experimental results.

Experimental validation was carried out by G Valavarasu and Santhosh Kotni of the Hindustan Petroleum Corporation Green Research and Development Center, and Amol Amrute of the Singapore Agency for Science, Technology and Research. Ambedkar Dukkipati, a professor in IISc’s Department of Computer Science and Automation, also contributed to the machine learning model.

Unexpected hydrogen pathway challenges conventional chemistry

Beyond improving the model’s predictions, the larger reaction network revealed an unexpected mechanism involving hydrogen.

In some important reaction steps, hydrogen can be transferred directly to intermediate compounds as intact H₂ molecules rather than only through individual hydrogen atoms. Additional quantum mechanical calculations confirmed that this route is particularly advantageous when hydrogen is transferred to an oxygen-containing intermediate.

The discovery challenges long-held assumptions about how hydrogen participates in catalytic reactions and may provide valuable clues for designing more effective catalysts.

“The idea that hydrogen can travel as an intact molecule without first splitting into atoms goes against what most of us have been taught,” said co-author Shivam Chaturvedi, a PhD student in IISc’s Department of Chemical Engineering. “This problem surfaced because the network was large enough to allow it, and the observation held when we went back and explicitly calculated these steps. This also suggests that catalysts that interact more strongly with H₂ could potentially enhance the pathway to methanol.”

The researchers believe this unexpected mechanism could support the development of catalysts that interact more effectively with hydrogen molecules, potentially improving the chemical pathway for producing methanol from CO₂.

A broader framework for studying catalytic reactions

The broader computational framework could prove useful beyond carbon dioxide hydrogenation. By integrating quantum mechanical simulations, machine learning, automated reaction identification and kinetic models, the approach could help researchers investigate other important industrial processes.

Potential applications include CO₂ reduction, nitrogen reduction and water splitting using various catalysts. By revealing chemical steps that may be missed by simpler models, the technology gives scientists a more complete picture of complex catalytic reactions and new opportunities to improve the processes used to make fuels and industrial chemicals.