A fast transferable method for predicting the glass transition from chemical structure, applied to poly(aryl ether ketone)s

We present a new method that successfully predicts the glass transition temperature T_g of polymers based on their monomer structure. The model combines ideas from Group Additive Properties (GAP) and Quantitative Structure Property Relationship (QSPR) methods, where GAP (or Group Contributions) assumes that sub-monomer fragments contribute additively to T_g , and QSPR links T_g to molecular descriptors of the fragments. We apply our method to a dataset of 146 linear poly(aryl ether ketone) (PAEK) homo- and co-polymers, resulting in a median root-mean-square error of 8 K (out-of-sample). The QSPR-GAP method offers a much faster prediction than typical QSPR methods, since the descriptor calculations are performed on smaller sub-monomer motifs. This method allows predictions for polymers based outside the data sample, which resolves the main limitation of the GAP approach. Using a genetic algorithm, we show that only two molecular descriptors are sufficient to predict T_g for PAEK polymers.