Machine Learning Models for Predicting Zirconocene Properties and Barriers.

Area of Science:

• Organometallic Chemistry
• Catalysis
• Computational Chemistry

Background:

• Zirconocene complexes are tunable catalysts for polyethylene production.
• Modifying the aromatic ligand framework offers significant control over catalyst properties.
• Developing predictive models for catalyst performance is crucial for efficient catalyst design.

Purpose of the Study:

• To develop machine learning (ML) models for predicting properties of zirconocene catalysts.
• To assess the performance of various ML algorithms and featurization methods for predicting electronic properties and reaction barriers.
• To identify key factors influencing the catalytic activity of zirconocenes in ethylene polymerization.

Main Methods:

• Generated a large library of over 700 DFT-calculated zirconocene systems.
• Developed and compared multiple ML models (e.g., using fingerprints, Coulomb matrices, SOAPs, persistence images) for predicting HOMO-LUMO gaps.
• Investigated ML models based on quantum-chemical descriptors (QCDs) for predicting ethylene migratory insertion barrier heights.
• Analyzed feature importance to understand fundamental principles governing catalyst reactivity.

Main Results:

• Highly accurate ML models were achieved for predicting HOMO-LUMO gaps, with performance dependent on algorithm and featurization.
• Machine learning models based on structural connectivity performed poorly for predicting barrier heights.
• Robust ML models for barrier heights were developed using QCDs, with those from π-coordination complexes showing superior accuracy.
• A Hammett-type principle emerged, indicating QCDs from specific structures better describe transition states.

Conclusions:

• ML models can accurately predict electronic properties of zirconocene precatalysts.
• Predicting reaction barriers requires different approaches, particularly utilizing QCDs.
• The choice of structure for harvesting QCDs significantly impacts model performance for barrier heights.
• Feature importance analysis provides fundamental insights into zirconocene catalyst reactivity.