Prediction of high-temperature polymer dielectrics using a Bayesian molecular design model

Machine learning has shown its great potential in the accelerated discovery of advanced materials in the field of computational molecular design. High-temperature polymer dielectrics are urgently required with the emerging applications of energy-storage dielectric film capacitors under high-temperature conditions. Here, we demonstrate the successful prediction of polymers with a high dielectric constant (ɛ) and high glass transition temperature (Tg) using a Bayesian molecular design model. The model is trained on a joint data set containing 382 computed ɛ values using density functional perturbation theory and experimentally measured Tg values of ∼7000 polymers to build relative quantitative structure–property relationships and identify the promising polymers with specific desired range of dielectric constant and glass transition temperature. From the hypothetical polymer candidates, ten promising polymers are proposed based on their predicted properties and synthetic accessibility score for high-temperature dielectric film capacitors’ application. Moreover, 250k novel polymer structures are generated with the model to support future polymer informatics research. This work contributes to the successful prediction of high-temperature polymer dielectrics using machine learning models.