Styrene Polymerization with Group III Metallocenes: A Generative and Interpretable Catalyst Design Strategy Combining Density Functional Theory and Machine Learning

The rational design of group III metallocene catalysts for styrene polymerization remains a complex challenge due to intertwined steric and electronic effects that govern monomer coordination and insertion. A generative and interpretable machine learning (ML) framework is presented, grounded in density functional theory (DFT) calculations, to predict and optimize the coordination energy (Δ H coord ) and activation barrier (Δ H act ) of the first styrene insertion step across 32 scandium and yttrium metallocenes. Key catalyst features—including Sterimol parameters, Tolman cone angle, metal‐centroid distance, buried volume, and natural metal charge—are extracted and used as descriptors in various regression models. The best‐performing ML models, notably support vector regressor‐bagging and random forest regressors, achieve sub‐2 kcal mol −1 mean absolute error for Δ H coord and 2.1 kcal mol −1 for Δ H act , exceeding the accuracy of DFT itself in some cases. SHapley Additive exPlanations analyses reveal steric parameters and metal acidity as primary drivers of catalyst performance. Using descriptor‐space optimization and generative mapping to ligand architectures, candidate ligands predicted to achieve target insertion barriers are identified, with validated DFT agreement. This integrative strategy highlights a path toward data‐driven, interpretable, and computationally efficient discovery of tailored polymerization catalysts. Through deliberate construction of both the descriptor space and the underlying chemical space, the approach remains effective in data‐light regimes, establishing a generalizable strategy for inverse design across homogeneous catalytic reactions.