Physicochemical interpretation of CO2 adsorption mechanisms in N-Doped mesoporous carbons via multilayer perceptron neural network

Developing efficient CO 2 capture technologies requires both advanced adsorbents and robust predictive models to decode their complex adsorption mechanisms. This study evaluates nitrogen-doped mesoporous carbons (MCNs) synthesized via a solvent-free mechanochemical route using a physics-informed Multilayer Perceptron (MLP) neural network. To overcome the limitations of traditional empirical models, our MLP architecture replaces standard mathematical activation functions with statistical physics-derived theoretical expressions. This approach enables highly accurate predictions of CO 2 adsorption isotherms R 2 >0.9998 while breaking the “black-box” nature of artificial intelligence to allow direct physicochemical interpretation. The MLP successfully decoupled the capture mechanisms, revealing that while undoped carbon relies on parallel multi-anchorage in micropores (64% Type I contribution), nitrogen doping structurally shifts the preference toward high-affinity monomolecular adsorption (74.4% Type I). Ultimately, this work validates physics-informed MLPs as robust, interpretable tools for the mechanistic evaluation and inverse design of high-performance carbon mitigation materials.