Gaussian process models of potential energy surfaces with boundary optimization.

Area of Science:

• Computational chemistry
• Machine learning in materials science

Background:

• Modeling intermolecular potentials is crucial for understanding molecular interactions.
• Gaussian processes are effective but can be data-intensive.

Purpose of the Study:

• To develop a strategy for reducing the number of training points for Gaussian process-based intermolecular potential modeling.
• To maintain or improve accuracy while decreasing data requirements.

Main Methods:

• Implemented a boundary optimization technique combining an asymptotic function with Gaussian processes.
• Learned the crossover distance between the two models from training data.
• Tested on dimer systems: CO-Ne, HF-Ne, HF-Na+, CO2-Ne, and (CO2)2.

Main Results:

• Achieved significant reduction in training points, up to ~49%, compared to sequential learning.
• Demonstrated the effectiveness of boundary optimization across various molecular systems.
• Maintained accuracy despite reduced data input.

Conclusions:

• Boundary optimization offers a more efficient approach to modeling intermolecular potentials.
• The method is transferable to other statistical modeling problems.
• Reduces computational cost and data requirements in molecular modeling.