Multiscale modeling and experimental validation of carbon dioxide absorption via superbase‐based deep eutectic solvents

Abstract To accelerate the development of high‐performance deep eutectic solvents (DESs) for CO 2 capture, a multiscale framework integrating machine learning (ML), experiments, and molecular simulations was established. A Mixture ID‐based grouped splitting strategy was used to reduce data leakage from repeated measurements of the same DES formulations under different conditions, and the optimized multilayer perceptron (MLP) model was then used for high‐throughput screening. Four superbase‐based DESs were constructed, and 1,5‐diazabicyclo[4.3.0]non‐5‐ene (DBN)‐monoethanolamine (MEA) exhibited the highest CO 2 absorption capacity of 0.2215 gCO 2 gDES −1 at ambient conditions. Fourier transform infrared spectroscopy (FT‐IR) and nuclear magnetic resonance (NMR) confirmed carbamate formation during CO 2 absorption, indicating a reaction‐assisted chemical absorption process. Molecular simulations revealed that basic sites on superbases promote CO 2 enrichment and regulate the local nonbonded environment, creating favorable pre‐reaction conditions for subsequent chemical absorption. This integrated strategy provides an efficient route for DES screening and mechanistic understanding for CO 2 capture.