Plate Nº 65 · recorded October 14, 2026
Chemistry & MaterialsReported finding
Machine Learning Uncovers 9,389 Hidden Reactions for Turning CO2 Into Fuel
IISc researchers used machine learning to expand a copper catalyst reaction network from 152 to 9,389 steps, lifting predicted CO2 conversion 40-fold and uncovering a molecular H2 pathway no one had modeled.
By James Calloway3 min read611 words
In brief
- Researchers mapped 9,389 reactions on a copper catalyst, up from an initial network of 152.
- Predicted CO2 conversion rose roughly 40-fold once the expanded reaction network was used.
- The original 152-reaction model wrongly identified formic acid as the main product instead of methanol.
- Hydrogen can transfer as intact H2 molecules, not only as separated atoms, in direct H2 steps that classical models had ignored.
- The study appeared in Nature Communications on October 13, 2026 (DOI: 10.1038/s41467-026-77080-4).
Scientists at the Indian Institute of Science (IISc) in Bangalore have identified 9,389 chemical reactions on a copper catalyst that could reshape efforts to turn carbon dioxide into fuel. The team, which reported its work in Nature Communications on October 13, 2026, used machine learning to expand a standard reaction database by more than 60-fold. Their expanded model predicted roughly 40 times more CO2 conversion than the original version.
The finding matters because CO2 hydrogenation — where CO2 reacts with hydrogen, helped along by a catalyst, to form fuels like methanol — is one of the most studied routes for recycling the greenhouse gas. Chemists usually model only the reactions they expect to matter. That shortcut keeps costs down but can hide important steps.
How does a small model go wrong?
The team started with 152 reactions derived from quantum mechanics simulations, computer calculations based on the physics of atoms and electrons. Most labs would consider that adequate. It was not. Fed the small network, the model pointed to formic acid, not methanol, as the chief product. It also badly underestimated how much CO2 was actually being converted in lab 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 CO2 gets converted," says corresponding author Ananth Govind Rajan, Associate Professor in the Department of Chemical Engineering at IISc. "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."
To scale up, the researchers trained machine learning models on the 152-reaction database to estimate energy barriers — the minimum energy a reaction needs to start — for other candidate reactions. They then ran automated tools over every possible single-step transformation involving 105 chemical species on the catalyst surface. The network jumped from 152 to 9,389 reactions. Predicted CO2 conversion rose about 40-fold. Methanol and carbon monoxide emerged as dominant products, matching what experiments had shown.
What does this change for catalyst design?
The expanded network surfaced a second, bigger surprise. In the bigger model, hydrogen could transfer to intermediate compounds as intact H2 molecules, not only as separated hydrogen atoms. Quantum mechanics calculations later confirmed this pathway, especially when the receiving molecule contained oxygen.
"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 at 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."
That points catalyst designers toward a previously overlooked lever: how strongly a catalyst holds onto H2. A material that grips molecular hydrogen more tightly might steer more CO2 down the methanol route. Possible in principle. Proven in a real reactor? Not yet.
Caveats the paper itself invites
The framework is computational. It has not yet produced a new, working material. Experimental checks were carried out by G. Valavarasu and Santhosh Kotni at Hindustan Petroleum Corporation Limited's Green Research and Development Centre, with Amol Amrute and colleagues at the Agency for Science, Technology and Research in Singapore. Those tests confirmed the model's predictions about a known copper catalyst, not a new one.
The team also tested only one catalyst system, and the H2 pathway needs further experimental proof. Even so, the same approach could feed studies of other industrial reactions, including nitrogen reduction and water splitting.
This is a sharper picture of an old reaction, not a finished reactor.
via dx.doi.org (Original)
More from James Calloway
Show full bio
Staff writer covering regulation, policy and their impact on general science news at SciBeat.
228 articles
Nearby plates
- Copper Catalyst Trick Balances Radical Chemistry for Hard Reactions
- Sulfur-Powered Material Splits Water Into Hydrogen Using Sunlight
- A hidden nickel-oxygen structure unlocks cheaper methane conversion
- 3D-Printed Bioreactor Turns Waste Methane Into Chemicals With 10x Efficiency
- AI Searches Millions of Enzymes to Find Pollution Fighters