Selective gas adsorption using graphitic carbon nitride: Exploring the role of molecular descriptors by artificial intelligence frameworks

Artificial Intelligence (AI) frameworks estimate the adsorption energies of crucial pollutants like CO₂, O₂, NO, NO₂, SO₂F₂, HCHO, and CO on Graphitic Carbon Nitride (g-C₃N₄) surfaces. The predictive capabilities of two AI-based models, namely, Artificial Neural Network (ANN) and ANN coupled with Grey Wolf Optimization (ANN-GWO), are assessed for this purpose. The frameworks are built over 232 data points of adsorption energy collected from Density Function Theory calculations (DFT). Further, molecular descriptors with two-dimensional and three-dimensional descriptors over molecular surfaces are created, serving as structural input for the AI frameworks. Both models, ANN and ANN-GWO, excel in estimating adsorption energies for polar gases such as CO₂, achieving prediction errors around 10⁻⁷, while nonpolar gases like NO₂ and HCHO exhibited larger deviations due to electron cloud diffusion. This emphasizes the critical role of molecular polarity in gas-surface interactions. This study underlines the significance of selecting appropriate molecular descriptors for reliable estimation of adsorption characteristics, offering a robust, computationally efficient alternative to conventional methods. The proposed frameworks provide valuable insights into pollutant-gas interactions, paving the way for advancements in material design for environmental applications.