Learning Design Rules and Generating Reactive Training Data for Energy Efficient and Deconstructable Polymerizations Using High Throughput Virtual Screening Workflows
Lauren Chua, Massachusetts Institute of Technology
While thermoset plastics have revolutionized high-performance engineering applications such as aviation and wind energy, these materials require copious amounts of energy to produce and persist in landfills for centuries. To enable more sustainable production of thermosets, strategies like frontal ring opening metathesis polymerization (FROMP) and degradable co-monomer incorporation are being investigated. Dicyclopentadiene (DCPD) is one of the most successful monomers capable of FROMP, and can be paired with silyl-ether co-monomers to impart chemically-triggered degradation. These strategies for more sustainable thermoset production, however, are poorly understood on the mechanistic scale, leading to challenges in design optimization and phenomenological understanding. Additionally, while DCPD has been studied extensively with respect to the multi-target property optimization required for FROMP, there are few other chemistries in the vicinity of the monomer’s chemical space that have been explored similarly. In this work, we have produced computational workflows to expand the chemical landscape around DCPD to learn design rules for FROMP while also generating high-quality training data for machine learned interatomic potentials (MLIPs). Critical to this data generation is a high-throughput pipeline for quantum chemical calculations on reactions, employing automated techniques for transition state finding. MLIPs trained on these datasets could help elucidate the various mechanisms at play in these energy efficient and deconstructable polymerizations at accuracies and speeds otherwise unapproachable.