IISc team needed 9,389 reactions to make a copper CO2 model predict methanol

"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?" said Anand Mohan Verma, the first author, who did the work as a postdoctoral fellow at the Indian Institute of Science ^[13]. The study, published in Nature Communications ^[4], tests that worry with a clean control: the group's own starting network of 152 reactions, each computed with quantum-mechanical simulations ^[5]. Modeled on its own, that network "wrongly predicted formic acid, not methanol, as the major product, and underestimated how much CO2 gets converted," said Ananth Govind Rajan, the corresponding author and an associate professor of chemical engineering at IISc ^[8].

The final network is about 62 times the size of the control ^[14], with 9,237 reactions added ^[15]. People hand-pick shortlists because running quantum mechanics on every possible reaction is prohibitively expensive ^[3]. This design spends the expensive calculations on a training set and lets machine learning estimate barriers for the rest ^[5][6]. The thing the phys.org account doesn't tell you is how good those estimates are, or how many of the added steps carry meaningful flux in the kinetic model.

That gap bears on the headline figure. The roughly 40-fold rise in CO2 conversion compares the large model with the small one ^[2]. It is not a laboratory measurement. Agreement with experiment is described in terms of the products observed ^[2]. "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," Govind Rajan said ^[1].

The best experiment in the paper is a small one. The enumerated network proposed that hydrogen can reach some intermediates as an intact H2 molecule, and the team went back to quantum mechanics to check ^[10]. "The idea that hydrogen can transfer as an intact molecule, without first splitting into atoms, runs against what most of us were taught," said Shivam Chaturvedi, a PhD student and co-author ^[11]. The cheap model raised the hypothesis and the expensive one tested it. "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," he said ^[12].

Catalyst design enters through that result. "This also suggests that catalysts that interact more strongly with H2 could potentially enhance pathways leading to methanol," Chaturvedi said ^[16]. It is a design hypothesis, and testing it would take a new catalyst and a reactor.

The evidence shows one curated 152-reaction network on copper getting the main product wrong ^[8]. I think one such failure justifies checking any shortlist mechanism against an enumerated network before ranking catalysts on it. It does not identify which published mechanisms are wrong. The authors say the same framework could be applied to CO2 reduction on other catalysts, nitrogen reduction and water splitting ^[17].