Machine learning-assisted screening of efficient ionic liquids for catalyzing CO2 cycloaddition reaction

The catalysis of CO2 cycloaddition reactions by ionic liquids holds significant promise in addressing environmental and chemical synthesis challenges. However, the design of effective ionic liquid catalysts is hindered by the sheer diversity of cation-anion combinations, leading to challenges in targeted catalyst development. This study addresses these issues by utilizing experimental data collected from literature on reactions involving epoxy compounds as substrates, conducted without solvents or co-catalysts, to establish a database for machine learning (ML). Simple descriptors derived from the ion pair structures of ionic liquids are employed as inputs for five ML classification algorithms to predict the yield of CO2 cycloaddition reactions. Subsequently, the highly accurate ML models are applied to forecast the catalytic performance of 1344 ionic liquids under ambient conditions. This approach identifies 13 cation structures and 8 anion structures that exhibit superior catalytic properties. Further refinement through density functional theory (DFT) calculations selects ion pair structures capable of catalyzing CO2 cycloaddition reactions at ambient temperature and pressure, demonstrating the efficacy of this method in guiding the design and development of ionic liquid catalysts for CO2 conversion reactions involving epoxy compounds.