Modeling Exchange Reactions in Covalent Adaptable Networks with Machine Learning Force Fields

Recycling and reprocessing of conventional thermosetting polymers have received considerable attention in view of environmental protection and sustainable development. By incorporating specific functional groups capable of reversible exchange reactions into polymer networks, the covalent adaptable networks (CANs) can alter the topology arrangement and achieve stress relaxation. Studying the topology rearrangement using conventional empirical force fields is challenging since they have a fixed bond connectivity. Ab initio molecular dynamics is capable of describing the exchange reactions, but it cannot study the kinetics that exceeds the achievable time scale. To address these problems, we developed a machine-learning force field for the exchange reactions of polyimine CANs. We showed that the developed machine-learning force field can achieve DFT-level accuracy in energy and atomic force predictions. By combining the developed machine learning force field with enhanced sampling methods, we provided a description of the reaction mechanisms and quantified the corresponding free energy profiles, revealing how water molecules affect the amine-imine exchange reactions, which cannot be achieved by conventional empirical force fields with fixed bond connectivity. Additionally, we illustrated the exchange reaction-induced topology rearrangement and relaxation of local stress in polyimine CANs using the developed machine learning force field.