Design of High-Performance MXene-Based 2D Membranes for Desalination via Machine Learning and Hybrid Optimization Algorithms

The global scarcity of freshwater resources has intensified the demand for efficient seawater desalination technologies. However, conventional approaches, such as reverse osmosis and nanofiltration, often suffer from high energy consumption and limited membrane performance. MXene, a promising two-dimensional (2D) material, offers unique structural and surface chemical properties that can enhance membrane-based separations. This study presents an integrated inverse design framework that combines machine learning (ML), optimization algorithms, and molecular dynamics (MD) simulations to accelerate the development of high-performance MXene membranes. A dual-target predictive model based on XGBoost was constructed to estimate the water flux and salt rejection, identifying surface charge density and pore area as key governing parameters. A range of intelligent optimization strategies was used to design the MXene membrane with a high desalination performance. Especially, the hybrid optimization methods showed clear advantages over baseline algorithms, delivering faster convergence and obtaining an optimized MXene membrane. MD simulations validated the predicted performance of optimized structures, showing good agreement with the prediction results of the ML model. The optimal configuration of Ti3C2O2─with a charge coefficient of 1.1 and a pore area of 82.72 Å2─achieved an excellent desalination performance with water permeability and salt rejection. This work demonstrates the power of integrating artificial intelligence and molecular modeling for the rational design of desalination membranes. The proposed framework provides a generalizable and scalable approach for advancing the intelligent development of 2D materials in water treatment applications.