Computational Design and Modeling of Vitrimers

Abstract

Vitrimers are an emerging class of sustainable polymers that consist of dynamic covalent adaptive networks. The ability of their polymer chains to attach and detach from one another through bond exchange reactions provides them with self-healing capabilities. However, the widespread application of vitrimers across various industries is currently limited by a lack of understanding of their molecular structures at the atomistic scale, primarily due to two key aspects. First, the limited choice of available monomers constrains current vitrimer chemistries, thereby restricting their thermomechanical properties. Second, existing experimental and simulation methods fall short in capturing the intricate bond exchange reactions and subsequent network rearrangement, which in turn limits our understanding of the viscoelastic properties of vitrimers. To address these challenges, this dissertation presents an integrated molecular dynamics (MD) - machine learning (ML) framework for the design and investigation of vitrimers at the atomistic scale. For training the ML models, a diverse vitrimer dataset of one million chemistries is curated. The glass transition temperature (Tg) for 8,424 of these chemistries is then calculated using highthroughput MD simulations calibrated by a Gaussian process model. A novel variational autoencoder model, which employs dual graph encoders and a latent dimension overlapping scheme, is developed and trained on this MD data. This design allows for the individual representation of multi-component vitrimers. High accuracy and efficiency of this framework are demonstrated by its ability to discover novel vitrimers with desirable Tg values that are outside the training regime. To experimentally validate the effectiveness of the framework, vitrimer chemistries are generated with a target Tg of 323 K. By incorporating chemical intuition, a novel vitrimer with a Tg of 311–317 K is successfully synthesized, which experimentally demonstrated both healability and flowability. The proposed framework offers a promising tool for polymer chemists to design and synthesize novel, sustainable polymers for various applications. Leveraging the same MD data, we train and benchmark seven ML models with six different feature representations for the prediction of Tg. By averaging the predicted Tg values from different models, an ensemble approach is shown to outperform individual models, enabling accurate and efficient property prediction on unlabeled datasets. Two novel vitrimers are identified and synthesized, exhibiting experimentally validated higher Tg than existing bifunctional transesterification vitrimers, along with demonstrated healability. This work demonstrates the potential of using MD data to train predictive ML models in the absence of sufficient experimental data, thereby enabling the discovery of novel, synthesizable polymer chemistries with a wide range of desirable properties. To realistically capture the complete reaction pathways of bond exchange reactions in vitrimers, we extend the Accelerated Reactive Molecular Dynamics (ReaxFF) technique. This extended framework enables a more accurate representation of vitrimer viscoelastic behavior at the molecular level. Bayesian optimization is used to select force field parameters within the Accelerated ReaxFF framework. Additionally, an empirical function is proposed to model temperature dependency, which allows for the control of reaction probabilities under varying temperatures. The extended framework is then used to simulate the non-isothermal creep behavior of vitrimers under various applied stress levels, heating rates, and numbers of reactions. The simulation results show good agreement with experimental findings in the literature, which validates the framework’s robustness. Using the Accelerated ReaxFF framework, we investigate the interplay between vitrimer chemistry, Tg, and healability. By quantifying the mobility of reactive atoms through mean square displacement calculations, we provide atomistic insights into experimental observations which offer guidance for designing vitrimers with better healability.