Discovery of Ni^(I) Complexes for CO₂ Insertion Enabled by a Machine Learning-Computational-Selection Sequence

The fate of a catalyst or catalytic intermediate, i.e., its speciation, in situ is a critical aspect of the efficiency of a catalyst as well as the overall reactivity and selectivity of the catalyzed transformation. However, the precise factors that dictate catalyst speciation are rarely understood and trial-and-error approaches frequently prevail. To address this challenge and develop predictive tools to guide ligand selection for a desired metal speciation in a catalytically relevant context, we evaluated the feasibility of machine learning combined with computational activation barrier predictions to achieve CO₂ insertion at room temperature for the sterically least hindered (and most vulnerable) Ni^(I)-Ph complexes, which constitute key catalytic intermediates. Following an in depth computational rationalization on the origin of reactivity difference of Ni^(I) versus Ni^(II) toward CO₂ insertion, we subsequently pursued machine learning to identify ligands that favor the critical Ni^(I) oxidation state. To this end, a descriptor database was constructed in silico. Subsequent application of machine learning led to the prediction of numerous ligands that favor the more reactive Ni^(I)-Ph intermediate and oxidation state, which were subsequently filtered for candidates that also show desired room temperature reactivity through the calculation of activation barriers. Ultimately, a set of representative candidates was synthesized and experimentally tested for CO₂ insertion, confirming their reactivity and alignment with computational predictions. This work offers a blueprint for creating and analyzing virtual databases to predict ligands-including never synthesized ones-that control metal complex oxidation state, nuclearity, and reactivity.