Interpolation of intermolecular potentials using Gaussian processes.

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Molecular Modeling

Background:

• Accurate intermolecular potential energy surfaces (PES) are crucial for understanding molecular interactions.
• Generating high-quality PES often requires extensive computational resources and data.

Purpose of the Study:

• To develop an efficient procedure for constructing accurate PES from limited data.
• To improve interpolation techniques for molecular interaction energies.
• To enable accurate predictions of macroscopic properties from first principles.

Main Methods:

• Utilizing Latin hypercube design for geometrical configuration generation.
• Employing Gaussian processes with over-specified inverse molecular distances for interpolation.
• Incorporating symmetric covariance functions to ensure physical symmetries are respected.

Main Results:

• The proposed procedure accurately predicts independent test sets for various molecular systems (CO2, HF-HF, CH4-N2).
• Training with high-precision ab initio data allows parameter-free prediction of CO2-CO cross virial coefficients.
• The predicted cross virial coefficient shows excellent agreement with experimental data.

Conclusions:

• The developed method provides an accurate and efficient way to generate PES.
• This approach significantly reduces the data requirements for PES construction.
• The method has broad applicability to various molecular interactions and can lead to accurate macroscopic property predictions.