A Tutorial Review of Bayesian Optimization with Gaussian Processes to Accelerate Stationary Point Searches

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Accelerating searches for stationary points on potential energy surfaces is crucial for chemical and materials simulations.
• Traditional methods require numerous expensive electronic structure calculations.

Purpose of the Study:

• To present a unified Bayesian optimization framework for various stationary point searches.
• To demonstrate the efficiency and accuracy of the proposed surrogate model approach.

Main Methods:

• Gaussian process regression with derivative observations and inverse-distance kernels.
• Active learning strategies for efficient surrogate model updates.
• Unified six-step surrogate loop for minimization, saddle point, and path searches.

Main Results:

• The unified framework effectively accelerates searches for minima and saddle points.
• Significant reduction in electronic structure evaluations (up to an order of magnitude) is achievable.
• Optional extensions enhance scalability and robustness for production use.

Conclusions:

• Bayesian optimization provides a unified view for diverse stationary point searches.
• The developed framework offers a computationally efficient and accurate alternative to traditional methods.
• Accompanying code facilitates practical implementation and further research.