Scientists have uncovered thousands of previously overlooked chemical reactions that could significantly improve the conversion of carbon dioxide into valuable green fuels and essential industrial chemicals. By combining modern machine learning algorithms with advanced computer simulations, researchers at the Indian Institute of Science have identified 9,389 individual reactions occurring on a copper catalyst, revealing unexpected chemical pathways that conventional models had completely missed.
The findings, published in the journal Nature Communications, could help scientists develop much more effective catalysts for transforming captured carbon dioxide into useful products such as methanol. Beyond improving yield predictions, the research also uncovered a surprising way hydrogen can participate directly in these reactions, challenging long-standing assumptions about how the industrial conversion process actually works at a molecular level.
Recycling carbon dioxide is considered a vital technological frontier in the global effort to mitigate climate change and transition away from a fossil-fuel-dependent economy. One of the most promising approaches to achieving this is catalytic hydrogenation, a chemical process that uses a specialized catalyst to help carbon dioxide react with hydrogen gas to form compounds such as methanol. Methanol is not only a widely used industrial solvent and chemical feedstock, but it also holds immense potential as a clean-burning fuel and a medium for hydrogen storage.
However, designing efficient catalysts for this transformation is exceptionally difficult because the underlying chemistry involves an enormous number of individual reaction steps taking place simultaneously across the microscopic surface of the catalyst. Simulating every single possible quantum mechanical step requires immense computing power, forcing researchers to rely on simplified models. Typically, scientists focus only on a carefully curated selection of reactions that they intuitively believe are the most important. While this reduction makes the simulations manageable, it introduces a major blind spot, potentially leaving hundreds or even thousands of potentially critical chemical pathways entirely 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 the Indian Institute of Science and is now an Assistant Professor at the Motilal Nehru National Institute of Technology Allahabad.
AI Reveals Thousands of Overlooked Chemical Reactions
To overcome the severe limitations of traditional modeling, the research team adopted an innovative computational strategy. They began by using rigorous quantum mechanics simulations to build a highly detailed baseline database containing 152 fundamental chemical reactions. They then trained sophisticated machine learning models to rapidly and accurately estimate the energy barriers that additional, unmapped reactions would need to overcome to proceed.
Using automated computational tools developed for the project, the scientists systematically examined the vast landscape of possible reactions involving 105 distinct chemical species present on the copper catalyst surface. They also carefully determined which specific molecular transformations could realistically take place within a single reaction step. This powerful automated approach dramatically expanded their original network from a modest 152 reactions to a staggering 9,389 interconnected pathways.
The difference in predictive accuracy between the small and large networks proved to be monumental. When the researchers relied solely on their initial, smaller network, their mathematical model produced flawed results, incorrectly identifying formic acid as the primary reaction product rather than methanol. Furthermore, the limited model predicted substantially lower overall carbon dioxide conversion rates than what had consistently been observed in physical laboratory experiments.
"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 carbon dioxide gets converted," explains corresponding author Ananth Govind Rajan, an Associate Professor in the Department of Chemical Engineering at the Indian Institute of Science. "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."
Expanded Model Predicts 40 Times More Carbon Dioxide Conversion
When the scientists incorporated the massive new reaction network into a comprehensive kinetic model describing how quickly the chemical processes occur, the mathematical predictions shifted dramatically. The projected amount of carbon dioxide conversion increased by approximately 40 times compared with the original, simplified model.
Crucially, the expanded computational model correctly identified methanol and carbon monoxide as the major reaction products, bringing its theoretical predictions into complete agreement with established experimental findings. This validation bridges a long-standing gap between theoretical chemistry models and real-world observations in laboratory and industrial settings.
Experimental validation for the theoretical work was carried out by G Valavarasu and Santhosh Kotni at Hindustan Petroleum Corporation Limited’s Green Research and Development Centre, working alongside Amol Amrute and their colleagues at the Agency for Science, Technology, and Research in Singapore. Additionally, Ambedkar Dukkipati, a Professor in the Department of Computer Science and Automation at the Indian Institute of Science, contributed significantly to the development and fine-tuning of the machine learning models.
A Surprising Hydrogen Pathway Challenges Conventional Chemistry
Beyond dramatically improving the accuracy of conversion predictions, the vastly larger reaction network exposed a completely unexpected mechanism involving how hydrogen behaves during the catalytic cycle.
In several of the most influential reaction steps identified by the algorithm, hydrogen was found to be transferred to intermediate chemical compounds directly as intact molecular hydrogen, rather than only through the traditional pathway of breaking apart into separate hydrogen atoms first. Additional, rigorous quantum mechanics calculations confirmed that this intact transfer pathway can be energetically favorable, particularly when hydrogen is transferred directly to chemical intermediates containing oxygen atoms.
This surprising discovery directly challenges long-standing chemical assumptions about how hydrogen participates in catalytic transformations on metal surfaces. The insight could provide invaluable clues for industrial chemists seeking to design better, highly targeted 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, a PhD student in the Department of Chemical Engineering at the Indian Institute of Science. "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 molecular hydrogen could potentially enhance pathways leading to methanol."
The research team believes this unexpected molecular mechanism could guide the future development of novel catalysts designed to interact more effectively with molecular hydrogen, potentially optimizing the complex chemical pathways responsible for turning greenhouse gases into useful fuels.
Furthermore, the broader computational framework developed by the team holds immense promise well beyond the specific domain of carbon dioxide hydrogenation. By seamlessly integrating quantum mechanics simulations, machine learning algorithms, automated identification of chemical reactions, and advanced kinetic rate models, the methodology offers a versatile toolkit for investigating a wide array of other complex industrial chemical processes.
Potential future applications of the technique include exploring carbon dioxide reduction using alternative metal catalysts, improving nitrogen reduction methods for sustainable ammonia synthesis, and optimizing water-splitting reactions for clean hydrogen production. By revealing hidden chemical steps that simpler, traditional models inevitably miss, the new computational approach provides scientists with a much more complete and reliable picture of complex catalytic reactions, opening up exciting new opportunities to improve the chemical technologies that power modern industry.
