Mechanism-Oriented Reaction Descriptors for Dispersity Prediction in Nitroxide-Mediated Radical Polymerization Considering Mediator Stability

In recent years, there has been tremendous growth in data-driven approaches in the field of polymer science, mainly using large databases. However, these analyses do not often lead to mechanism-based interpretations, such as validating a hypothesis for a key reaction phenomenon. In this study, quantitative structure-property relationship (QSPR) models for dispersity (Đ) are constructed using reaction representations based on a putative reaction mechanism. These models are built on nitroxide-mediated radical polymerization reaction data sets collected from the literature. The proposed reaction representation consists of 17 descriptors based on the quasi-equilibrium reaction of forming dormant species, including the stability of mediator radicals. The tendency for the decomposition of the mediator radicals is quantified with a single descriptor. Excluding an unstable mediator radical based on this descriptor significantly increases the prediction accuracy for a poly(n-butyl acrylate) data set. Removing reactions proceeding on different reaction paths contributes to more accurate QSPR-based Đ prediction models, and also to clearer model interpretation in light of the mechanism assumption. Additionally, prediction is found to be possible for data sets from different research groups under similar experimental conditions, while no meaningful prediction is achieved for reactions conducted under different experimental conditions and monomer conversion.