Autonomous Search for Polymers with High Thermal Conductivity Using a Rapid Green–Kubo Estimation

Abstract

A rapid Green–Kubo (GK) estimation method was developed herein to evaluate the thermal conductivity of linear amorphous polymers by using equilibrium molecular dynamics simulations. Statistical errors of heat flux correlations in the GK relation were greatly reduced by neglecting intermolecular contributions. This accelerates evaluation of thermal conductivity 100 times faster than the conventional GK scheme. Our method was applied for autonomous search for new polyimides using the Monte Carlo tree search algorithm. Approximately 1000 all-atom molecular dynamics evaluations resulted in the highest thermal conductivity of 0.25 W/m·K. The importance of chemical fragments for high thermal conductivity was quantified via Shapley value analysis for a regression model built by using the data. Furthermore, the chain-conformation dependence of thermal conductivity was investigated by using bond vector correlations and earth mover’s distances. The correlation functions along the polymer chains were found to be a good descriptor of polymer thermal conductivity.