Interpolation and Extrapolation of Global Potential Energy Surfaces for Polyatomic Systems by Gaussian Processes with...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Machine Learning in Chemistry

Background:

• Gaussian process (GP) regression is a powerful tool for constructing global potential energy surfaces (PES) for polyatomic molecules.
• Increasing the number of energy points improves GP model accuracy but also increases computational difficulty.
• Existing methods face challenges in balancing accuracy and computational cost for complex molecular systems.

Purpose of the Study:

• To develop a method for enhancing the accuracy of global PES without increasing the number of energy points.
• To demonstrate the effectiveness of iteratively increasing GP kernel complexity for improved accuracy.
• To explore the capability of GP models with composite kernels for physical extrapolation of PES.

Main Methods:

• Iterative increase in the complexity of Gaussian process kernels.
• Optimization of training point distributions for GP models.
• Construction of a six-dimensional PES for H3O+ using ab initio data.

Main Results:

• GP models with composite kernels achieve higher accuracy for a given number of potential energy points.
• The approach allows for significant accuracy improvements by varying training point distributions.
• Accurate global PES extrapolation to higher energies was achieved using lower-energy training data.

Conclusions:

• Composite GP kernels maximize accuracy for a given distribution of potential energy points.
• The developed method offers a way to improve PES accuracy efficiently.
• This approach facilitates accurate physical extrapolation of potential energy surfaces.