Toward Reaction Vessel Mimicry: Machine Learning-Assisted Automated Exploration of Alkene Polymerization and Its...

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Engineering

Background:

• Accurate reaction kinetics and product prediction are crucial for simulating chemical reactions.
• Semiempirical methods offer speed but lack accuracy, while DFT methods are accurate but computationally expensive.
• Existing methods struggle with efficient exploration and reliable prediction for complex reaction systems.

Purpose of the Study:

• To demonstrate the advantages of iterative training of delta-learning neural network potentials (ΔNNP) for automated reaction path exploration.
• To achieve DFT-level accuracy in predicting reaction kinetics and products using a computationally efficient approach.
• To establish a robust framework for mimicking complex chemical reactions in realistic reaction vessels.

Main Methods:

• Iterative training of a delta-learning neural network potential (ΔNNP) to learn the energy difference between DFT and semiempirical methods.
• Utilizing ethylene polymerization catalyzed by the [ZrCp2CH3]+ catalyst as a model system.
• Applying the trained ΔNNP to explore reaction path networks involving varying numbers of ethylene molecules and incorporating cocatalyst effects.

Main Results:

• The ΔNNP achieved DFT-level accuracy for reaction kinetics and product prediction in ethylene polymerization.
• The model successfully captured key elementary steps including initiation, propagation, and termination.
• The framework demonstrated adaptability to propylene polymerization and ZrCp2-mediated chemistry like metallacycle formation with minimal additional training.

Conclusions:

• Iterative ΔNNP training provides a powerful and efficient approach for automated reaction path exploration and product prediction.
• This framework significantly advances the ability to realistically mimic complex chemical reactions and polymer growth processes.
• The method offers a scalable solution for studying reactions with repeated analogous elementary steps, paving the way for more accurate simulations.