AI reveals a hidden chemical shortcut that could turn CO2 into fuel

Researchers at the Indian Institute of Science (IISc) used machine learning combined with quantum mechanics simulations to expand a copper-catalyst CO2 hydrogenation reaction network from 152 reactions to 9,389, according to findings published in Nature Communications. The smaller 152-reaction network wrongly predicted formic acid rather than methanol as the major product and underestimated CO2 conversion, while the expanded network's predictions matched experimental results and raised predicted CO2 conversion roughly 40-fold, the team reported. The work also identified a previously unexpected hydrogen reaction pathway that the authors say could inform more efficient catalyst design for converting CO2 into fuels such as methanol.

AI reveals a hidden chemical shortcut that could turn CO2 into fuel - Date: - October 13, 2026 - Source: - Indian Institute of Science IISc - Summary: - Scientists used artificial intelligence to uncover nearly 10,000 chemical reactions involved in transforming carbon dioxide into fuels and chemicals, revealing crucial pathways that traditional models had missed. The discovery dramatically improved predictions of CO2 conversion and uncovered a surprising hydrogen reaction that could inspire more efficient catalysts. - Share: Scientists have uncovered thousands of previously overlooked chemical reactions that could help improve the conversion of carbon dioxide CO<sub 2</sub into valuable fuels and industrial chemicals. By combining machine learning with advanced computer simulations, researchers at the Indian Institute of Science IISc identified 9,389 individual reactions occurring on a copper catalyst, revealing unexpected chemical pathways that conventional models had missed. The findings, published in Nature Communications , could help scientists develop more effective catalysts for transforming CO<sub 2</sub into useful products such as methanol. The research also uncovered a surprising way hydrogen can participate in these reactions, challenging established assumptions about how the process works. One promising approach to recycling carbon dioxide is hydrogenation, which uses a catalyst to help CO<sub 2</sub react with hydrogen and form chemicals such as methanol. However, the process involves an enormous number of individual reactions taking place across the catalyst's surface. Simulating every possible step using quantum mechanics would require immense computing power, so researchers typically focus on a limited selection of reactions believed to be the most important. This simplification can leave hundreds of potentially critical reactions out of the picture. "We began with a worry familiar to anyone who does mechanistic modeling: How do you know that your reaction network has not omitted the one step that matters?" says first author Anand Mohan Verma, who conducted the research as a CV Raman Postdoctoral Fellow at IISc and is now an Assistant Professor at the Motilal Nehru National Institute of Technology Allahabad MNNIT Allahabad . AI Reveals Thousands of Overlooked Chemical Reactions To overcome this limitation, the research team began by using quantum mechanics simulations to build a detailed database containing 152 chemical reactions. They then trained machine learning models to quickly estimate the energy barriers that additional reactions would need to overcome. Using automated computational tools, the scientists examined the possible reactions involving 105 chemical species present on the catalyst surface. They also determined which transformations could take place in a single reaction step. This approach expanded their original network from just 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 rather than methanol. It also predicted substantially less CO<sub 2</sub conversion than experiments indicated. "When we modeled the process using the 152 reactions considered initially, the network wrongly predicted formic acid, not methanol, as the major product, and underestimated how much CO<sub 2</sub gets converted. Only when we expanded the network to include thousands of additional, previously overlooked reactions did the predictions fall in line with what we and others see experimentally," explains corresponding author Ananth Govind Rajan, Associate Professor in the Department of Chemical Engineering, IISc. Expanded Model Predicts 40 Times More CO<sub 2</sub Conversion When the scientists incorporated the larger reaction network into a model of how quickly the chemical processes occur, the predicted amount of CO<sub 2</sub conversion increased by approximately 40 times compared with the original model. The expanded model also correctly identified methanol and carbon monoxide as major products, bringing its predictions into agreement with experimental findings. Experimental validation was carried out by G Valavarasu and Santhosh Kotni at Hindustan Petroleum Corporation Limited's Green Research and Development Centre, along with Amol Amrute and colleagues at the Agency for Science, Technology, and Research in Singapore. Ambedkar Dukkipati, Professor in the Department of Computer Science and Automation at IISc, also contributed to the machine learning models. A Surprising Hydrogen Pathway Challenges Conventional Chemistry Beyond improving the predictions, the much larger reaction network exposed an unexpected mechanism involving hydrogen. In several important reaction steps, hydrogen could be transferred to intermediate chemical compounds directly as molecular H2, rather than only through separate hydrogen atoms. Additional quantum mechanics calculations confirmed that this pathway can be especially favorable when hydrogen is transferred to intermediates containing oxygen. This finding challenges a long-standing assumption about how hydrogen participates in catalytic reactions and could provide valuable clues for designing better catalysts. "The idea that hydrogen can transfer as an intact molecule, without first splitting into atoms, runs against what most of us were taught," says co-author Shivam Chaturvedi, PhD student in the Department of Chemical Engineering, IISc. "This surfaced only because the network was large enough to allow for it, and the observation held up when we went back and computed those steps explicitly. This also suggests that catalysts that interact more strongly with H2 could potentially enhance pathways leading to methanol." The researchers believe this unexpected mechanism could guide the development of catalysts that interact more effectively with molecular hydrogen, potentially improving the chemical pathways responsible for producing methanol from CO<sub 2</sub . The broader computational framework may also prove useful well beyond carbon dioxide hydrogenation. By integrating quantum mechanics simulations, machine learning, automated identification of chemical reactions, and models of reaction rates, the approach could help researchers investigate other important industrial processes. Potential applications include CO<sub 2</sub reduction using different catalysts, nitrogen reduction, and water splitting. By revealing chemical steps that simpler models might miss, the technique offers scientists a more complete picture of complex catalytic reactions and new opportunities to improve processes used to produce fuels and chemicals. Story Source: Materials provided by Indian Institute of Science IISc . Note: Content may be edited for style and length. Journal Reference : 1. Anand M. Verma, Shivam Chaturvedi, Swastik Paul, Srinibas Nandi, Rahul Sheshanarayana, Kotni Santhosh, G. Valavarasu, Ambedkar Dukkipati, Chuandayani Gunawan Gwie, Pei Ying Moo, Chun Qi Joy Ng, Amol Amrute, Ananth Govind Rajan. Data-driven massive reaction networks reveal mechanistic pathways underlying catalytic CO2 hydrogenation . Nature Communications , 2026; DOI: 10.1038/s41467-026-77080-4 http://dx.doi.org/10.1038/s41467-026-77080-4 Cite This Page : ScienceDaily . Retrieved October 14, 2026 from www.sciencedaily.com