Why in the News
Researchers at the Indian Institute of Science (IISc), Bengaluru, have mapped 9,389 elementary chemical reactions involved in the making of fuel chemicals from carbon dioxide on a copper catalyst - the largest such data-driven analysis so far. The research was published online in Nature Communications on 17 September 2026 (open access); the institute's release and the main news coverage are of 6 October 2026 - keep the two dates separate. The co-first authors are Anand Mohan Verma (then a C.V. Raman postdoc at IISc, now an assistant professor at MNNIT Allahabad), Shivam Chaturvedi and Swastik Paul (IISc chemical engineering); the machine-learning part involved Prof. Ambedkar Dukkipati of the computer science department, and experimental validation involved Hindustan Petroleum's (HPCL) Green R&D Centre and Singapore's A*STAR. The essence of the finding: traditional small models misjudged the process; the large network not only matches experiments but also challenges a textbook assumption.
Background: The Stakes of Fuel and Methanol from CO2
Converting CO2 into fuels and chemicals such as methanol is a major scientific race towards a "circular carbon economy" - turning a greenhouse gas into raw material instead of waste. But in CO2 hydrogenation on a copper surface, hundreds to thousands of small molecule-level steps run together. Computing every possible step with quantum mechanics (such as DFT) is so expensive that researchers traditionally pick 100-200 reactions by estimation and experience and build a model. First author Verma says this is the field's old worry: what if the step that changes the whole result was left outside the chosen list? IISc's work attempts to erase this "blind spot" with machine learning and automated computation - and it is tied to experiments through industry partnership (HPCL), so it is not merely a computer-room exercise.
Understand methanol's importance separately: it is both a clean-burning fuel and a basic building block of the chemical industry - the base of hundreds of products such as plastics, solvents and formaldehyde. Today methanol is made mostly from natural gas or coal; if made from CO2 and green hydrogen, it is called green methanol - which is why its demand is being discussed in shipping fuel and energy storage. The present study strengthens the earliest understanding of that pipeline - what happens on the catalyst surface.
The Method: Quantum Mechanics + Machine Learning + Automated Computation
The team worked in three stages. First, a carefully built base database of 152 fundamental reactions was prepared from rigorous quantum-mechanical calculations. Then machine-learning models were trained that quickly estimate the "activation barriers" (the energy wall needed to start) of additional reactions - because pure quantum calculation of thousands of reactions would take decades. Finally, automated tools scanned possible single-step transformations among the 105 chemical species present on the catalyst surface. In this way the network swelled from 152 to 9,389 elementary reactions - according to the paper, human bias (we choose only what we consider "important") also came down.
Where Did the Small Model Miss?
The comparison is clear. The traditional model with only 152 reactions predicted that the main product would be formic acid and also showed low CO2 conversion - neither matches experiments. The detailed microkinetic model with 9,389 reactions matches experiments: the CO2 conversion rate is about 40 times higher, and the main products are methanol and carbon monoxide - exactly as seen in the laboratory. The paper itself admits that even networks considered "large, with 100+ reactions" turned out to be inadequate. The message is technical but firm: in complex catalysis, the size of the model can itself decide the quality of the conclusion.
A Question on a Textbook Assumption: Direct Transfer of Molecular H2
The most striking finding is qualitative. Chemistry teaching has held for decades that before hydrogenation on a metal surface, molecular hydrogen (H2) must break into separate atoms. The detailed network showed that in several important reactions the H2 molecule, without breaking, directly hands over hydrogen to intermediate compounds as a molecule - especially with oxygen-bearing intermediates. According to co-author Shivam Chaturvedi, this could emerge only because the network was so large that even non-traditional routes could be included; the selected steps were later confirmed by explicit quantum calculations. The paper calls it a "data-driven serendipitous discovery, which was verified afterwards". The practical signal: catalysts can be designed that form strong bonds with the H2 molecule, so that the routes leading to methanol gain priority.
Experimental Validation and Collaboration
The model did not remain on paper. G. Valvarasu and Santosh Kotni of HPCL's Green Research and Development Centre, and a team from Singapore's A*STAR, took part in the experimental checks; the paper also includes experimental measurements of CO2 conversion. Computational resources came from IISc's Supercomputer Education and Research Centre (SERC); funding support included HPCL, A*STAR, the Infosys Foundation (young researcher grant) and an Education Ministry PhD fellowship, and the open-access fee was paid through the Government of India's "One Nation One Subscription" (ONOS) initiative - this detail is recorded in the paper's funding section. The team says the same framework can be applied to other catalytic processes - CO2 reduction on other catalysts, nitrogen reduction (ammonia) and water splitting (hydrogen).
Limitations and Cautions
Several points should be kept measured. First, this is a computational (modelling) study - not a new catalyst material or a factory-scale process; it makes no claim about the cost or energy layout of methanol production. Second, the "40 times" figure is of the estimated conversion rate compared with the 152-reaction baseline model, not of an industrial reactor's efficiency gain. Third, uncertainty remains in machine-learning-estimated barriers - quantum verification was done for the important steps, but pure calculation of all 9,389 is not possible. Fourth, the model is specific: copper surface, CO2 hydrogenation; generalisation to other catalyst systems is a matter for future work. Fifth, the discovery of H2-molecule transfer is "verified post facto" - a strong signal, but independent laboratory confirmation is yet to spread through the literature. Sixth, keep the dates distinct: the paper went online on 17 September 2026, the IISc release/main coverage on 6 October - this is an interpretation of a month-old, not week-old, discovery; read today's publication as an explainer.
What to Watch Next?
Three things: (1) application of this framework to other catalysts and to ammonia/hydrogen processes; (2) experimental studies that directly demonstrate the H2-molecule path; and (3) pilot-scale tests by industry partners such as HPCL, where the model's suggestions enter real reactor design. From a policy view, this work shows the maturity of "computational chemistry + AI", where supercomputing infrastructure (SERC) and open-access policies such as ONOS take Indian fundamental research to international journals.
Reference Points for the Reader
Remember three things. First, a "catalyst" speeds up a reaction and is not itself consumed - copper is the traditional catalyst for CO2-to-methanol. Second, the "activation barrier" is the energy wall that decides how easily a reaction will happen; machine learning here estimates these very walls. Third, CO2 utilisation and carbon capture and storage (CCS) are different strategies - this study is on the utilisation side: turning the gas into chemicals, not burying it.
Mains Practice Question
How can artificial intelligence and computational chemistry accelerate India's clean-energy and carbon-utilisation ambitions? Discuss the bridge between fundamental research and industrial application with reference to IISc's 9,389-reaction network. (General Studies-III: Science and Technology/Environment; 250 words, 15 marks)
Sources and Limitations
- Primary research paper (Nature Communications, open access): Verma, Chaturvedi, Paul et al., "Data-driven massive reaction networks reveal mechanistic pathways underlying catalytic CO2 hydrogenation", 17 September 2026; DOI 10.1038/s41467-026-77080-4. Abstract, author affiliations and funding read.
- Crossref record: confirmation of online publication on 17 September 2026.
- IISc institutional page: 6 October 2026 - study summary and context (institutional release).
- The Times of India: 6 October 2026 - 152 to 9,389, 105 species, 40-times, Chaturvedi/Verma quotes, HPCL-A*STAR validation.
- India Today: 6 October 2026 - the discovery's summary and significance (news).
- EurekAlert press release: 6 October 2026 - institutional press material.
- Limitations: A computational study, no industrial-efficiency claim; "40 times" relative to the baseline model; ML-barrier uncertainty; copper-specific; the date difference between the paper (17 September) and the release (6 October); the full text is open access, but the main figures are taken from the abstract and coverage.