Microenvironment‐controlled Li + /Mg 2+ separation in COFs : Explored using high‐throughput simulations and machine learning

Abstract Separating Li + from Mg 2+ in salt‐lake brines remains challenging due to their similar hydrated diameters. Herein, covalent organic frameworks were used as model membranes to systematically evaluate how steric hindrance, hydrophilicity, and membrane charge control ion transport. A dataset from 196 high‐throughput non‐equilibrium molecular dynamics simulations was analyzed using machine learning. Water and Cl − fluxes were governed mainly by membrane charge, whereas Li + transport depended on all three properties, and Mg 2+ flux was most sensitive to steric hindrance. A Gaussian process regression model accurately predicted fluxes and was coupled with mixed‐integer optimization to design an optimal membrane. No Mg 2+ permeation was detected in three validation trajectories of the optimized candidate, while water and Li + transport persisted. Shapley additive explanations revealed that Li + /Mg 2+ selectivity arises primarily from steric sieving at high hindrance, assisted by weak negative charge and low hydrophilicity, providing a data‐driven framework for covalent organic framework (COF) membrane optimization.