From Static Pathways to Dynamic Mechanisms: A Committor-Based Data-Driven Approach to Chemical Reactions.

Area of Science:

• Computational Chemistry
• Chemical Dynamics
• Reaction Mechanisms

Background:

• Dynamic effects are crucial for understanding chemical reactions.
• Existing computational methods often rely on static approximations.
• Accurate modeling of reaction pathways requires dynamic considerations.

Purpose of the Study:

• To develop a committor-based workflow integrating artificial neural networks and machine learning potentials.
• To accurately capture dynamic effects in chemical reaction pathways.
• To uncover complex reaction mechanisms and energy landscapes.

Main Methods:

• Developed a Path-Committor-Consistent Artificial Neural Network (PCCANN).
• Integrated PCCANN with an iteratively trained Message Passing Atomic Convolutional Encoder (MACE) potential at hybrid DFT level.
• Applied the workflow to S_NAr reactions and protonated alcohol isomerization.

Main Results:

• Investigated an S_NAr reaction, finding a concerted mechanism with a lower dynamic barrier than static methods.
• Mapped the free-energy landscape for protonated isobutanol isomerization, revealing three competing pathways.
• Identified novel metastable intermediates in stepwise reaction routes.

Conclusions:

• The synergistic PCCANN-MACE protocol accurately models complex reaction dynamics.
• The workflow reveals mechanistic diversity and uncovers previously undescribed reaction pathways.
• This approach serves as a proof-of-concept for committor-based discovery in chemical dynamics.