Researchers at the Indian Institute of Science (IISc) in Bengaluru have used artificial intelligence and computer simulations to identify 9,389 chemical reactions taking place on a copper catalyst used to turn carbon dioxide (CO₂) into methanol and other useful products. The study was published in Nature Communications.
Standard models consider only a handful of reactions, because simulating them all would require huge computing power. The team began with 152 reactions calculated in detail and trained machine learning models to quickly estimate thousands more.
The difference was striking: the small model predicted the wrong product, while the expanded one matched the main products seen in the laboratory and predicted around 40 times more CO₂ conversion. The scientists also found that hydrogen can react as an intact molecule, without first splitting into atoms, contrary to what is usually taught.
The results were confirmed by experiments in laboratories in India and Singapore. The same method could help improve other important industrial processes, such as producing hydrogen from water.
