Combining Machine Learning and Spectroscopy to Model Reactive Atom + Diatom Collisions.

Area of Science:

• Chemical Physics
• Computational Chemistry
• Aerospace Engineering

Background:

• Predicting energy distributions in reactive atom + diatom collisions is vital for atmospheric re-entry simulations.
• Explicit dynamics studies (quasi-classical or quantum) are computationally impractical due to the vast number of accessible states.

Purpose of the Study:

• To develop and quantitatively test a machine-learned (ML) model for predicting product translational, vibrational, and rotational energy distributions.
• To assess the model's ability to retain atomistic details for coarse-grained simulations.

Main Methods:

• Developed an ML model utilizing translational energy and product vibrational states.
• Model incorporates a ro-vibrational coupled energy expression based on the Dunham expansion.
• Tested model against quasi-classical trajectory (QCT) simulations.

Main Results:

• ML models accurately reproduced final state distributions from QCT simulations with R² ≈ 0.98.
• Thermal rates predicted by ML models align with QCT simulation results.
• Demonstrated that ML retains essential atomistic details.

Conclusions:

• Machine learning provides a robust and accurate method for modeling complex chemical dynamics.
• The developed ML model is suitable for applications in coarse-grained simulations, aiding atmospheric re-entry research.
• ML is effective for integrating mixed computational and experimental data in physical sciences.