Atomic-Level Features for Kinetic Monte Carlo Models of Complex Chemistry from Molecular Dynamics Simulations.

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Materials Science

Background:

• Traditional kinetic models struggle with predicting reactivity of unobserved molecules, limiting their application in complex chemical systems.
• Current models often rely on molecular descriptions, restricting predictions to species present in initial simulations.
• High computational costs of atomic interactions necessitate development of efficient kinetic modeling approaches.

Purpose of the Study:

• To introduce a novel approach for extracting reaction mechanisms and rates from molecular dynamics (MD) simulations using atomic-level features.
• To enable kinetic models to predict reactivity of unobserved molecules and explore novel reaction pathways.
• To develop more accurate, transferable, and data-efficient kinetic models for complex chemical systems.

Main Methods:

• Development of kinetic models parameterized using atomic-level features extracted from MD simulations.
• Application of the new approach to the complex chemical network of hydrocarbon pyrolysis.
• Comparative analysis of atomic-feature-based models against molecular-feature-based models for accuracy and transferability.

Main Results:

• Kinetic models built with atomic features successfully predicted reaction pathways not observed in the parameterizing MD simulations.
• Atomic features enabled construction of reaction mechanisms and rate estimation for unknown molecular species from elementary atomic events.
• Models demonstrated accuracy and transferability comparable to molecular-feature models but were more compact and required less data.
• Atomic features improved the description of large molecule formation, allowing simultaneous modeling of small molecules and condensed phases.

Conclusions:

• Kinetic models employing atomic features offer a significant advancement for predicting chemical evolution, especially for rare events and unobserved species.
• This approach enhances the capability to model complex chemical networks, including large molecules and condensed phases.
• Atomic-feature-based kinetic models provide a more robust, efficient, and transferable alternative to existing methods.