Potential energy surfaces for high-energy N + O₂ collisions.

Area of Science:

• Computational Chemistry
• Chemical Dynamics
• Physical Chemistry

Background:

• Accurate potential energy surfaces (PES) are essential for simulating chemical reactions under extreme conditions, such as in shock waves.
• Modeling high-energy collisions between molecules and atoms, like oxygen (O2) and nitrogen (N), requires reliable PES for accurate dynamics calculations.

Purpose of the Study:

• To develop and compare two sets of potential energy surfaces (PES) for collisions between O2(3Σg−) and N(4S).
• To assess the fitting performance and computational efficiency of a neural network (NN) approach with permutationally invariant polynomials (PIPs) against a many-body (MB) method using PIPs and mixed-exponential-Gaussian bond order variables (MEGs).

Main Methods:

• Developed doublet, quartet, and sextet PES using the electronically adiabatic approximation.
• Employed two fitting strategies: 1) Neural Networks (NNs) with permutationally invariant polynomials (PIPs). 2) Least-squares many-body (MB) method with accurate diatomic potentials and PIP-MEG three-body terms.
• Utilized a consistent dataset for training and evaluating both fitting methods.

Main Results:

• The PIP-NN method demonstrated significantly superior fitting accuracy compared to the MB-PIP-MEG method, even when the latter used higher-order PIPs.
• Despite better accuracy, the PIP-NN PES required approximately five times more computational time for trajectory calculations than the MB-PIP-MEG PES.

Conclusions:

• The PIP-NN approach offers a highly accurate representation of the potential energy surfaces for O2-N collisions.
• A trade-off exists between the accuracy of the PIP-NN method and the computational efficiency of the MB-PIP-MEG method for dynamics simulations.