Multi-scale studies on CO₂ capture: An in-depth investigation of amine-based deep eutectic solvents using machine learning and molecular simulations

Abstract

To address the urgent need for efficient and recyclable solvents in industrial carbon capture, this study designed and synthesized three alcohol amine-based deep eutectic solvents (DESs). Through a multi-scale strategy integrating experimental characterization, molecular simulation, and machine learning, the CO₂ capture mechanisms and optimization pathways of these solvents were elucidated. The results show that DES1, using tetrabutylammonium bromide (TBAB) as the hydrogen bond acceptor and monoethanolamine (MEA) as the hydrogen bond donor, exhibits the optimal performance: its CO₂ absorption capacity reaches 0.198 g/g at 30 °C, and the regeneration efficiency remains above 95 % after 7 cycles. The CatBoost machine learning model identified the minimum electrostatic potential (ESPmin) and the maximum atomic distance within molecules (Farthest_Distance) as core descriptors. Combined with density functional theory (DFT) and molecular dynamics simulations, it was confirmed that the electron-rich property of the amino nitrogen atom in DES1 and its electrostatic complementarity with CO₂ drive chemisorption, while the low viscosity significantly enhances mass transfer efficiency. In contrast, for DES3, the steric hindrance of the –CH₃ group in N-methyl diethanolamine (MDEA) suppresses reaction activity, leading to predominantly physical absorption. By integrating an “experiment-simulation-data-driven” strategy, this study clarifies the synergistic mechanism among electronic effects, steric hindrance, and mass transfer resistance, providing theoretical support and engineering screening tools for the targeted design of low-carbon solvents.