Decoding polyethylene formation in Cr/PNP catalyzed ethylene oligomerization via experimentally guided machine learning

Abstract

Polyethylene formation remains a critical side reaction in ethylene selective oligomerization, lowering α-olefin yields and lacking reliable predictive strategies. Here, an automated workflow incorporating structural parsing and indexing was developed to construct a Cr/PNP catalysts library, extract comprehensive molecular descriptors, and integrate them with machine learning models trained on experimentally measured PE values. A combined classification-regression strategy enabled accurate identification of low-PE catalysts and quantitative prediction of PE contents, establishing a broadly applicable framework for catalyst design and side-product control in homogeneous catalysis. Analysis of SHAP feature importance and DFT calculations revealed that the intrinsic properties of the ligand primarily determine whether it produces low or high PE. In contrast, optimizing experimental conditions plays a pivotal role in further reducing PE in low-PE ligands.