Hierarchy in Binary Liquid Phase Exfoliation of Graphitic Carbon Nitride: Dissecting the Dominance of One Solvent

This study targets an important question in the synthesis of graphitic carbon nitride (g-C₃N₄) nanosheets via liquid-phase exfoliation (LPE): how does a binary solvent mixture perform relative to its components? A machine learning model based on Extra Trees Regressor, when applied to 171 pure solvents and 14,535 binary solvents resulting from their arbitrary combination, reveals an interesting phenomenon where the LPE performance of the binary mixture can be dominated by one of its components. Quantitatively the free energy of exfoliation (ΔG_exf), defined as the free energy to separate a unit area of two stacked nanosheets, can have close values for the binary solvent and one of its components that has either a large ΔG_exf (low LPE efficacy) or a small ΔG_exf (high LPE efficacy). Such nonstandard performance is validated by umbrella sampling molecular dynamics (MD) simulations, and examined in detail for two representative binary solvents: one where <i>N</i>-Methyl-2-pyrrolidone (NMP) dominates in NMP:Cyclohexane mixture showing good LPE performance, and the other where methanol (MET) dominates in MET:Dichloromethane mixture showing poor LPE performance. Results from this study challenge existing, largely surface tension based, solvent selection criteria for LPE, and emphasize the importance of molecular details in the solvation layers around the nanosheets. Our numerical framework, integrating MD with machine learning, can be used to further explore strategies in the synthesis of 2D materials, such as optimizing solvent mixtures and introducing functional groups to g-C₃N₄.