Unlocking Adsorption Potential: Machine-Learning-Assisted Design of Doped Magnesium Oxide Structures for CO2 Capture

Magnesium oxide (MgO) is widely regarded as a promising CO₂ capture material due to its stability, low cost, and high theoretical adsorption capacity. However, the actual adsorption performance of intrinsic MgO is hindered by the limited exposure of alkaline sites. Doping modification has been demonstrated as an effective method to improve its adsorption capacity. Unfortunately, detailed configurations of these doped MgO structures and their correlation with adsorption energy remain little explored and are unclear. This study systematically constructs a total of 142 doped MgO structures (Element–Concentration–Adsorption sites, ECA) involving 17 elements across the periodic table. Through a synergistic approach combining density functional theory (DFT) and machine learning, optimal doping elements and configurations are identified. Nonmetals (N, P, S, Cl) and metalloids (Si, Al) are identified as the most effective dopants, enhancing CO₂ adsorption energy by up to 720% compared to undoped MgO. Most of the dopant atoms exhibit site-dependent performance, with oxygen adsorption sites generally favorable. Key electronic descriptors, such as s- and d-orbital electron differences of the dopant atom (NE-s-dA) and band gap variations (BGD-d), strongly correlate with adsorption energy. Machine-learning models further validate these findings, highlighting the critical role of dopant-induced charge redistribution and orbital interactions. This work provides a direct guide for designing high-performance MgO-based adsorbents, emphasizing the efficacy of nonmetal and metalloid doping for efficient CO₂ capture.