Approximate high mode coupling potentials using Gaussian process regression and adaptive density guided sampling

Area of Science:

• Computational Chemistry
• Theoretical Chemistry
• Quantum Chemistry

Background:

• Accurate potential energy surfaces (PES) are crucial for understanding chemical reactions and molecular properties.
• Traditional PES construction methods can be computationally expensive, especially for higher-order interactions.

Purpose of the Study:

• To present a novel and efficient algorithm for constructing potential energy surfaces (PES).
• To leverage n-mode representation and Gaussian process regression for accurate higher-order mode potentials.
• To minimize computational cost while maintaining high accuracy in PES construction.

Main Methods:

• The algorithm utilizes the n-mode representation combined with an adaptive density-guided approach.
• Gaussian process regression is employed to construct approximate higher-order mode potentials.
• Delta-learning schemes integrating electronic structure methods at various levels of theory are explored.

Main Results:

• Approximate 2-mode potentials achieve errors around 8 cm^-1.
• Approximate 3-mode and 4-mode potentials exhibit errors below 1 cm^-1.
• The achieved accuracy surpasses contributions from missing higher-order electron excitations and relativistic effects.

Conclusions:

• The developed method offers a significant improvement over neglecting higher-order couplings in PES construction.
• This efficient approach enables the generation of highly accurate approximate potential energy surfaces.
• The findings have implications for advancing molecular simulations and theoretical chemical studies.