A Multivariate Linear Regression Approach to Predict Ethene/1-Olefin Copolymerization Statistics Promoted by Group 4 Catalysts

Abstract

We report a combined multivariate linear regression (MLR) and density functional theory (DFT) approach for predicting the comonomer incorporation rate in the copolymerization of ethene with 1-olefins. The MLR model was trained to correlate the incorporation rate of a set of 19 experimental group 4 catalysts to steric and electronic features of the dichloride catalyst precursors. Although the assembled experimental data were produced in different laboratories and both propene and 1-hexene copolymerization results were considered, the trained MLR model results in a R2 value of 0.82 and a leave-one-out Q2 value of 0.72. The trained model was validated against a validation set comprising 3 catalysts from the literature and not included in the training set plus one catalyst synthesized by us. Except for one literature catalyst, data in the validation set were predicted with reasonable accuracy. Additionally, a catalyst synthesized by us, for which the MLR model predicted a comonomer incorporation of 4.0%, resulted in a 1-hexene experimental incorporation of 4.5–5%. The trained MLR model was used to predict the comonomer incorporation rate of 10 related zirconocenes having structural features similar to the 19 systems in the training set. We further explored the impact of the precatalyst structure on the comonomer incorporation rate by analyzing a set of 15 zirconocenes having steric and electronic features different from those in the training set. These predictions were validated by DFT calculations.