> ## Content Index
> Fetch the complete content index at: https://research-pop.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Indian Institute of Science-led team maps nearly 9,400 steps in copper-catalyzed CO₂ hydrogenation
- URL: https://research-pop.com/indian-institute-of-science-led-team-maps-nearly-9-400-steps-in-copper-catalyzed-co2-hydrogenation/
- Published: 2026-09-18T13:38:59.000Z
- Updated: 2026-09-18T13:38:59.000Z
- Author: ResearchPOP
- Tags: Chemistry, Energy

![](https://storage.ghost.io/c/7c/5e/7c5e0911-739b-4a35-94eb-3e810bf78ba7/content/images/2026/09/Screenshot-2026-09-18-at-15.22.14.png)

**Source**: [https://doi.org/10.1038/s41467-026-77080-4](https://doi.org/10.1038/s41467-026-77080-4?ref=research-pop.com)

## At a glance

How large must a reaction network be to describe CO₂ hydrogenation on copper? A team led by researchers at the Indian Institute of Science calculated 152 elementary reactions using density functional theory, then combined machine learning with automated reaction enumeration to construct a network containing 9,389 elementary steps.

The expanded network changes the kinetic picture. In the authors’ microkinetic comparison, it predicts a CO₂ conversion rate about 40 times that obtained from a smaller, manually curated network and more closely follows the experimental trends in CO and methanol formation. The difference arises between two modeled networks using the same catalytic system. It shows how strongly pathway selection can influence a kinetic prediction.

Analysis of the larger network brings forward reaction classes that are easily omitted from compact mechanisms. These include hydrogen transfer between adsorbed species and hydrogenation involving molecular H₂. The researchers tested a Cu/SiO₂ catalyst to provide an experimental reference for the predicted conversion and product trends, connecting the computational workflow with catalytic measurements.

## Background

Hydrogenating CO₂ on copper can produce CO, methanol and other carbon-containing products. Even when the main products are known, the path from reactants to products can involve many adsorbed intermediates. Each intermediate may undergo hydrogenation, dehydrogenation, bond formation, bond cleavage or hydrogen transfer through several competing routes.

Microkinetic models describe this chemistry by linking elementary reactions and calculating how quickly material moves through the resulting network. Their conclusions depend on which reactions are included. A compact mechanism chosen from chemical intuition can provide a manageable starting point, but it may exclude routes that appear individually less familiar while becoming important when connected to the rest of the network.

Calculating every possible transition state directly with density functional theory is also expensive. The number of candidate reactions grows rapidly as more surface intermediates and carbon-containing species are introduced. The challenge is to retain the accuracy of first-principles calculations while extending the mechanism far beyond the number of steps that can be selected and calculated one by one.

## Research question

How does expanding a copper-catalysis model from a curated set of reactions to thousands of automatically generated elementary steps alter its mechanistic and kinetic description of CO₂ hydrogenation? The researchers also examine whether a machine-learning model trained on first-principles data can identify reaction classes that help reproduce experimental conversion and product trends.

## Inside the study

The team began by assembling a first-principles dataset for CO₂ reduction chemistry on copper. Density functional theory calculations supplied reaction energetics and activation barriers for 152 elementary reactions. The dataset covers a range of surface transformations and provides the reference information needed to train and evaluate the subsequent machine-learning model.

![](https://storage.ghost.io/c/7c/5e/7c5e0911-739b-4a35-94eb-3e810bf78ba7/content/images/2026/09/Screenshot-2026-09-18-at-15.22.22.png)

The researchers then developed a workflow that represents surface species and identifies chemically plausible transformations between them. Automated enumeration expands combinations of bond formation, bond cleavage and hydrogen transfer without requiring each pathway to be selected manually. Elementary-reaction identification converts those possibilities into individual steps that can enter a kinetic model.

This enumeration shows how quickly the mechanism grows. Even before extending the chemistry fully into larger C₃ and C₄ products, the number of possible intermediates and connections becomes too large for direct transition-state calculations of every step. The role of machine learning is to bridge that scale difference.

![](https://storage.ghost.io/c/7c/5e/7c5e0911-739b-4a35-94eb-3e810bf78ba7/content/images/2026/09/Screenshot-2026-09-18-at-15.22.32.png)

Trained on the DFT dataset, the machine-learning models estimate activation barriers for candidate reactions beyond the directly calculated set. Tests against held-out computational data examine whether the model can reproduce barriers that were not used during training. The resulting predictions allow thousands of reactions to be screened and incorporated into a microkinetic framework while retaining a connection to the underlying first-principles calculations.

![](https://storage.ghost.io/c/7c/5e/7c5e0911-739b-4a35-94eb-3e810bf78ba7/content/images/2026/09/Screenshot-2026-09-18-at-15.22.38.png)

The completed automatically generated network contains 9,389 elementary reactions. The authors compare it with a smaller network assembled through conventional pathway selection. Even a mechanism containing more than 100 reactions gives a substantially different result from the larger network, showing that a seemingly extensive curated mechanism can still leave out kinetically important connections.

In the microkinetic simulations, the large network predicts a CO₂ conversion rate about 40 times that of the smaller network. It also follows the experimental direction of CO and methanol production more closely. The larger prediction does not come from changing the catalyst or operating conditions. It emerges because the expanded mechanism gives adsorbed species additional routes through which they can react and reach the observed products.

![](https://storage.ghost.io/c/7c/5e/7c5e0911-739b-4a35-94eb-3e810bf78ba7/content/images/2026/09/Screenshot-2026-09-18-at-15.22.45.png)

Pathway analysis identifies hydrogen movement as a central difference. Conventional mechanisms often represent hydrogenation as the sequential addition of adsorbed hydrogen atoms to one intermediate. The expanded network also contains hydrogen transfer between different surface species, allowing one adsorbate to donate hydrogen as another accepts it. These inter-species transfers create connections between reaction families that may remain separated in a smaller mechanism.

The model also highlights hydrogenation by molecular H₂. Rather than requiring H₂ to dissociate completely before every hydrogenation event, some routes allow the molecule to participate directly in transforming a surface intermediate. The authors examined these machine-learning-derived pathways after their identification and used further analysis to support their chemical relevance.

![](https://storage.ghost.io/c/7c/5e/7c5e0911-739b-4a35-94eb-3e810bf78ba7/content/images/2026/09/Screenshot-2026-09-18-at-15.22.48.png)

To connect the model with catalytic behaviour, the group prepared, characterized and tested a Cu/SiO₂ catalyst for CO₂ hydrogenation. Measurements of CO₂ conversion and product selectivity provided experimental trends against which the networks could be compared. The large network better reproduces the relative behaviour of CO and methanol, supporting the inclusion of the additional hydrogen-transfer and molecular-H₂ pathways in the kinetic description.

The comparison also shows the different roles of computation and experiment in the study. The experiments establish the overall catalytic response of the copper material. The network analysis then examines which combinations of elementary steps can generate the same direction of conversion and selectivity. Together, they provide a route from observed products back to a broader set of possible surface transformations.

## Takeaways and outlook

The study presents CO₂ hydrogenation as a reaction-network problem as much as a catalyst problem. Starting from 152 DFT-calculated reactions, the workflow expands the mechanism to 9,389 elementary steps and incorporates those steps into microkinetic modeling. The resulting change in predicted conversion shows why network size and composition can determine the outcome of a kinetic analysis.

The expanded model also shifts attention toward hydrogen-transfer chemistry. Interactions between different adsorbed species and hydrogenation involving molecular H₂ provide routes that are less visible in a compact, sequential mechanism. Their emergence from the larger network gives the authors a mechanistic explanation for the improved agreement with experimental CO and methanol trends.

More broadly, the workflow combines chemical rules, first-principles calculations, machine-learning barrier prediction, automated enumeration, microkinetic analysis and catalyst testing. This structure can be adapted to other catalytic reactions in which the number of plausible surface transformations grows beyond what can be explored manually. As computational datasets and reaction representations expand, the same approach may help reveal productive pathways that remain outside conventional curated mechanisms.

## About the researchers

Anand M. Verma (Indian Institute of Science and Motilal Nehru National Institute of Technology Allahabad) is the first-listed author. Ananth Govind Rajan (Indian Institute of Science) is the corresponding author.

The other authors are Shivam Chaturvedi, Swastik Paul, Srinibas Nandi and Ambedkar Dukkipati (Indian Institute of Science); Rahul Sheshanarayana (Indian Institute of Science and Cornell University); Kotni Santhosh and G. Valavarasu (Hindustan Petroleum Corporation); and Chuandayani Gunawan Gwie, Pei Ying Moo, Chun Qi Joy Ng and Amol Amrute (A\*STAR).

## Original research

Anand M. Verma, Shivam Chaturvedi, Swastik Paul, Srinibas Nandi, Rahul Sheshanarayana, Kotni Santhosh, G. Valavarasu, Ambedkar Dukkipati, Chuandayani Gunawan Gwie, Pei Ying Moo, Chun Qi Joy Ng, Amol Amrute and Ananth Govind Rajan. “Data-driven massive reaction networks reveal mechanistic pathways underlying catalytic CO₂ hydrogenation.” *Nature Communications*, published online 17 September 2026\. DOI: 10.1038/s41467-026-77080-4\. [Journal article](https://www.nature.com/articles/s41467-026-77080-4?ref=research-pop.com)

---

## Research POP Notes

This article reflects the independent interpretation of the Research POP team and does not represent the views of the authors, their institutions or the journal. If you identify any inaccuracies or have concerns regarding the content, figures or attribution, please contact us at [team.researchpop@gmail.com](mailto:team.researchpop@gmail.com). We will review the matter promptly and make corrections or remove the relevant material where appropriate.