Gaussian Process Model for Collision Dynamics of Complex Molecules.

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Chemical Physics

Background:

• Scattering calculations are crucial for understanding molecular interactions.
• Accurate potential energy surfaces (PES) are vital for reliable simulations.
• Exploring the multidimensional dependence of scattering observables is computationally intensive.

Purpose of the Study:

• To develop an efficient computational method combining Gaussian process models with scattering calculations.
• To establish a multidimensional mapping of scattering observables against controllable and PES parameters.
• To demonstrate the utility of this model for inverse scattering and error reduction in molecular dynamics.

Main Methods:

• Utilizing Gaussian process models trained on a limited number of scattering calculations (approx. 100-200).
• Applying classical trajectory calculations to generate a ten-dimensional hypersurface for Ar-C$_{6}$H$_{6}$ collisions.
• Integrating both classical and quantum calculations to train the predictive model.

Main Results:

• A ten-dimensional hypersurface was generated for Ar-C$_{6}$H$_{6}$ collisions using 200 classical trajectories.
• The model accurately predicts collision lifetimes based on collision energy, internal temperature, and eight PES parameters.
• The trained model accurately describes quantum scattering cross sections, including near resonances.

Conclusions:

• Gaussian process models offer an efficient route to multidimensional mapping of scattering observables.
• This method facilitates inverse scattering problem solutions and sensitivity analysis of PES parameters.
• The approach enhances molecular dynamics by enabling efficient averaging and error reduction over PES variations.