Ethylene-Oxygen Combustion: From Machine Learning Potential Function Construction to Molecular Dynamics Simulation.

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Materials Science

Background:

• Accurate modeling of complex fuel combustion is crucial but challenging.
• Machine learning potentials (MLPs) offer a promising approach for simulating large chemical systems.

Purpose of the Study:

• To construct and validate a reactive machine learning potential function (MLP) for the ethylene-oxygen (C2H4-O2) system.
• To investigate the combustion behavior of C2H4-O2 using the developed MLP.

Main Methods:

• Trained an MLP using DeePMD-kit with over 890,000 configurations, incorporating data from molecular dynamics (MD) simulations and key reaction structures.
• Assessed MLP accuracy and transferability via NVT ensemble MD simulations at 3000 K.
• Identified reaction species and pathways, reducing them to simplified networks.

Main Results:

• Achieved a low average fitting error of 0.014 eV for the C2H4-O2 MLP.
• MD simulations revealed 175 species and 633 reactions, simplified into networks forming H2O and CO2.
• Predicted reaction rates and networks showed good agreement with established combustion mechanisms.

Conclusions:

• The developed C2H4-O2 MLP accurately describes combustion energetics and kinetics.
• This work demonstrates the potential of MLPs for simulating combustion processes of complex fuels.
• Provides a foundation for future studies on more intricate fuel systems.