Machine Learning Accelerated Diamine/Tertiary‐Amine Mixtures Design for CO ₂ Capture

ABSTRACT Chain diamines have gained attention in carbon capture recently for their high CO 2 absorption capacity and rate. However, how diamine structure regulates the activation barrier of CO 2 absorption remains unclear, and the large number of amine candidates hinders efficient screening of low‐energy absorbents. To resolve these issues, this study first used DFT to investigate the regulation mechanism of diamines on CO 2 absorption and clarify key reaction pathways and structure‐activity relationships. It was confirmed that diamines react with CO 2 via a zwitterion mechanism, while diamine/tertiary amine mixtures react with CO 2 through single‐step proton transfer. Diamines with more primary amine sites have lower barriers; methyl/ethyl substitution, carbon chain extension (on either amine), or hydroxyl substitution (on diamines) increases the proton transfer barrier. To address low screening efficiency from excessive candidates, an efficient framework integrating DFT and active learning was constructed. Using DFT‐calculated reaction barriers, a feature mapping with RDKit descriptors was built, and an active learning model was developed via 10 iterative rounds. The model achieved high prediction accuracy ( R 2 = 0.821) for the rate‐determining step's activation barrier. SHAP analysis identified the steric‐related first‐order molecular connectivity index (T_Chi1v) as the dominant feature. Finally, the optimal amine pair (AEEA + EDMA, activation barrier: 0.8 kcal·mol −1 ) was identified. This work clarifies the core mechanism via DFT, enables efficient candidate screening via active learning, and explains the optimal combination's performance through mechanistic tracing—providing an interpretable route for developing low‐energy, high‐efficiency mixed amine absorbents and advancing carbon capture technology.