Interpretable machining learning assisted insights into bifunctional squaramide catalyzed ring-opening polymerization of lactide†

Abstract

The structural understanding of catalysts is essential for achieving efficient and selective polymerization. In this study, we designed a series of bifunctional catalysts based on squaramide, carboxylates, and alkali cations for the ring-opening polymerization (ROP) of lactide. These catalysts exhibited controlled polymerization behavior with narrow dispersity (Đ_M = 1.08–1.12). Kinetic evaluations revealed a linear relationship between the catalyst's chain length and activity for short CH₂ chains (X = 1–4). However, as the CH₂ segments lengthened, an “odd-even” effect on the kinetics was found, suggesting that the chain length alternately enhances or diminishes catalytic activity. The catalytic activity was significantly influenced by the counter cation (Li⁺, Na⁺, K⁺, and Cs⁺) of carboxylate, with larger radius cations showing higher rate constants (k_{obs Cs⁺} > k_{obs K⁺} > k_{obs Na⁺} > k_{obs Li⁺}). Computational studies demonstrated that this correlation resulted from varying binding energies. Moreover, the k_obs value of the catalyst can be tuned by adding different ratios of the crown ether. An interpretable machine learning method was introduced to link physical properties and activities, guiding the further design of effective catalysts for ROP.