Exhaustive state-to-state cross sections for reactive molecular collisions from importance sampling simulation and a...

Area of Science:

• Chemical Kinetics
• Computational Chemistry
• Fluid Dynamics

Background:

• High-temperature gas flows exhibit inherent nonequilibrium energy and state distributions.
• Modeling these complex systems is computationally intensive due to numerous accessible states and transitions.
• Accurate modeling is crucial for understanding phenomena like hypersonic gas flow.

Purpose of the Study:

• To develop and validate neural networks (NNs) for quantitatively describing nonequilibrium reactive gas flows.
• To integrate NN-based models into mesoscale simulation methods like Direct Simulation Monte Carlo (DSMC).
• To enable more realistic simulations of hypersonic gas flow regimes.

Main Methods:

• Quasiclassical trajectory (QCT) simulations were used to compute state-to-state cross sections for the N + NO reaction.
• Neural networks (NNs) were trained on sparsely sampled, noisy QCT data.
• The trained NNs were used to predict reaction rates and state distributions.

Main Results:

• NNs accurately predicted independently generated reference data.
• State-specific and total reaction rates from NNs quantitatively agreed with QCT simulations.
• NN-predicted vibrational and rotational energy distributions matched explicit simulations.

Conclusions:

• NNs trained on physical data offer a computationally efficient alternative to explicit microscopic calculations.
• This methodology significantly advances the realistic modeling of nonequilibrium ensembles in simulations.
• The approach is particularly beneficial for mesoscale simulations in hypersonic regimes.