The era of human-guided epoxide selection in CO₂ cycloadditions is over—AI tools now take the lead

While CO₂ capture alone presents technical and environmental limitations, converting it into value-added products under mild conditions is a promising alternative. Using DFT and machine learning, this study systematically analyzes a wide range of epoxides, including halogenated and glycidyl ethers, with ascorbic acid/TBAI as a biobased catalyst system. The role of substrate structure, electronic properties, and counter-cations is examined to understand and optimize the catalytic process. A team of scientists has addressed the urgent need for CO₂ valorization by exploring its conversion into cyclic carbonates via organocatalytic cycloaddition with epoxides. Their work was published in the journal Industrial Chemistry & Materials in May 2025.

“We aim to develop a predictive and sustainable methodology for cyclic carbonate synthesis from CO₂, offering mechanistic insights and substrate guidelines to guide future catalyst and material design: simply calculations and machine learning will substitute or reduce drastically nocive experiments” explains Albert Poater, a professor at the University de Girona. The valorization of CO₂ as a renewable carbon source has become a critical priority in addressing the challenges posed by rising atmospheric CO₂ levels and climate change. While carbon capture and storage (CCS) strategies have shown potential for mitigation, they are often limited by their high energy requirements, economic constraints, and dependence on energy-intensive or non-sustainable reagents. As such, carbon capture and utilization (CCU) has emerged as a more attractive and sustainable alternative, offering the possibility not only to remove CO₂ from the atmosphere but also to transform it into valuable chemical products.

Among the various CCU strategies, the nonreductive cycloaddition of CO₂ with epoxides stands out due to its mild operating conditions and broad applicability. This reaction leads to the formation of cyclic carbonates, which are versatile compounds used as aprotic polar solvents in lithium-ion batteries, as polymer precursors, fuel additives, and as intermediates in the production of pharmaceuticals, agrochemicals, and fine chemicals. Importantly, these transformations can be carried out using simple biobased catalysts, such as amino acids, polysaccharides, lignin, or ascorbic acid, in combination with nucleophilic promoters like tetrabutylammonium iodide (TBAI).

Ascorbic acid, in particular, has gained attention as an efficient and recoverable organocatalyst capable of promoting CO₂-epoxide cycloaddition under ambient conditions. Its biocompatibility, availability, and ability to function in biphasic systems make it an ideal candidate for sustainable catalysis. However, despite the growing interest in this reaction, most computational studies to date focus only on a limited number of epoxides, typically small and reactive ones like ethylene oxide, thus failing to capture the full range of epoxide reactivity and structural diversity.

To address this gap, a comprehensive study combining density functional theory (DFT) and machine learning (ML) was undertaken to investigate the influence of electronic and steric features across a wide variety of epoxide substrates. This approach identifies key molecular descriptors, such as the buried volume around oxygen atoms, that affect activation barriers and reaction efficiency. Ultimately, the aim is to enable predictive catalyst and substrate design for optimized CO₂ fixation, contributing to greener, more efficient chemical processes and advancing the goals of a circular carbon economy.

The research team conducted a comprehensive study on the CO₂ cycloaddition to epoxides, unveiling key thermodynamic and kinetic aspects of the reaction and the dual catalytic role of ascorbic acid and iodide. They confirmed that the cyclization step is the rate-determining step (rds), especially for ethylene oxide, with bulkier epoxides showing increased energy barriers due to steric hindrance. Through steric mapping and %V_Bur analysis developed by Cavallo and Poater, they revealed how substitution patterns influence accessibility to the transition state. Integrating these descriptors into a machine learning model, they demonstrated that even a single feature like %V_Bur can reliably predict reactivity trends. This work highlights the power of combining mechanistic insight with data-driven tools to guide sustainable catalyst and substrate design, reinforcing the potential of predictive catalysis in the valorization of CO₂.

Looking ahead, the research team hopes that their work might provide insights into the development of hierarchically porous supports anchored single-atom catalysts for various catalytic applications. “We next plan to demonstrate that combining mechanistic analysis with machine learning enables reliable prediction of reactivity in CO₂–epoxide cycloadditions, paving the way for sustainable and data-driven catalyst and substrate design,” said Poater.

The research team includes Thalía Ortiz-García, Layla El-Khchin, Miquel Solà and Albert Poater from Universitat de Girona; David Dalmau and Juan Vicente Alegre-Requena from Universidad de Zaragoza; Sergio Posada-Pérez from Vrije Universiteit Brussel; and Valerio D’ Elia from Vidyasirimedhi Institute of Science and Technology.

This research is funded by the Spanish Ministerio de Ciencia, Innovación y Universidades, the Generalitat de Catalunya, the European Union's Horizon 2020 research and innovation Maria Skłodowska-Curie Actions, Gobierno de Aragón-Fondo Social Europeo.

Industrial Chemistry & Materials is a peer-reviewed interdisciplinary academic journal published by Royal Society of Chemistry (RSC) with APCs currently waived. ICM publishes significant innovative research and major technological breakthroughs in all aspects of industrial chemistry and materials, especially the important innovation of the low-carbon chemical industry, energy, and functional materials. Check out the latest ICM news on the blog.