Universal fundamental relations for polymer-solvent phase behavior

The ability to accurately predict polymer phase behavior is imperative to addressing many core questions in polymer physics. By allowing thermodynamic model parameters to be vectors instead of scalars (analogous to moving from Hildebrand to Hansen solubility parameters) and employing first principles physics-informed mixing models, we have demonstrated the ability accurately predict large infinite dilution activity coefficient data sets for small molecules. These predictions can extrapolate to the prediction of entire composition-dependent phase diagrams even without any composition-dependent experimental data. Further, they use only single component parameters, simplifying the dimensionality of the parameterization problem and allowing sparse data to be used to predict mixture properties for which no measurements have been performed. Herein, this model is extended from small molecule mixtures to polymer-solvent mixturesThe model extension is achieved by first compiling property data for a set of polymers as well as polymer-solvent mixture data and adding it to the graph network. Then, additional layers, including a polymer molar mass adjustment, are added to the Gibbs free energy model to capture polymer-solvent mixture behavior that is not present in small molecule mixtures with a particular focus on dealing with challenges polymers present due to local composition correlations and polydispersity.