Pareto-Archive Multiplayer Monte Carlo Tree Search for Multi-Objective Inverse Design of Ionic Liquids for CO2 Capture

Abstract We present a Pareto-archived framework for inverse design of ionic liquids (ILs) for CO2 capture using multiplayer Monte Carlo tree search and recurrent neural networks. The mass-based CO2 Henry constant, inverse CO2/N2 selectivity, and temperature-dependent Henry-constant ratio are retained as separate minimization objectives, with COSMO-SAC evaluation and local nondominated archives guiding molecular generation. Across 12 ion-pair systems and 2800 evaluated structures, the within-system Pareto union contains 115 candidates spanning three anion classes, whereas scalar absorption–selectivity–desorption index (ASDI) ranking retains only dicyanamide-based ILs in its global top 115. Three pyridinium–dicyanamide candidates have predicted mass-based CO2 Henry constants of 102.2–105.8 bar, CO2/N2 selectivities of 11.00–11.06, and temperature-dependence descriptors of 0.597–0.612. Feature analysis identifies substituent trends, while supplementary ASKCOS assessment supports synthetic prioritization. Together, these analyses provide a shortlist of computational targets for further synthetic and physical-property assessment.