Activation barriers in the glycolysis of PET as catalyzed by tertiary amines: Density functional calculations and predictions from interpretable machine learning

The efficient chemical recycling of poly(ethylene terephthalate) (PET) is a crucial part of any sustainable circular plastic economy. The depolymerization of PET can take place via a glycolysis reaction with tertiary amines as catalysts. We investigate this reaction using density functional calculations with the M06-2X functional, complemented by interpretable machine learning. A PET dimer containing C–O linkages is used as a representative model of the active region or the polymer. The concerted and the stepwise mechanisms of its glycolysis using trimethylamine as a catalyst are examined. The calculations reveal that the reaction proceeds preferentially via a stepwise mechanism with which the proton abstraction process being the rate-determining step. The structure–activity relationship shows that, in general, amines with side chains that are good electron donors enhance the reactivity. Deviations from this trend may be attributed to resonance or other specific structural factors. Based on the analysis of DFT-derived descriptors, the protonation free energy of the tertiary amines is identified as the most important single descriptor for predicting activation barriers. The Sure Independence Screening and Sparsifying Operator (SISSO) method is then employed to develop a few low-dimensional descriptors that enable a reliable prediction of activation barriers based solely on amine properties, thus eliminating the need for the computationally costly explicit transition state calculations.