A benchmark dataset for Hydrogen Combustion.

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Machine Learning

Background:

• Generating reference data for deep learning (DL) models in reactive systems, especially combustion, is complex due to extreme conditions and radical species.
• Accurate potential energy surfaces are crucial for understanding combustion reaction dynamics.

Purpose of the Study:

• To extend intrinsic reaction coordinate (IRC) calculations with ab initio molecular dynamics (AIMD) and normal mode displacement to comprehensively map the potential energy surface for hydrogen combustion.
• To generate a large-scale reference dataset for training deep learning models to study hydrogen combustion reactions.

Main Methods:

• Utilized intrinsic reaction coordinate (IRC) calculations, ab initio molecular dynamics (AIMD) simulations, and normal mode displacement calculations.
• Employed a high-quality range-separated hybrid density functional, ωB97X-V, for accurate electronic structure calculations.
• Evaluated approximately 290,000 potential energies and 1,270,000 nuclear force vectors across 19 reaction channels.

Main Results:

• Constructed an extensive reference dataset covering 19 reaction channels for hydrogen combustion.
• The dataset includes crucial information such as transition state ensembles.
• Generated a substantial volume of data (∼290,000 potential energies, ∼1,270,000 force vectors) essential for deep learning model training.

Conclusions:

• The developed computational approach successfully generates comprehensive reference data for hydrogen combustion.
• This dataset will significantly aid the development and application of deep learning models in understanding complex combustion chemistry.
• The methodology provides a robust framework for creating reference data for other reactive systems.