Restricted-Variance Constrained, Reaction Path, and Transition State Molecular Optimizations Using Gradient-Enhanced...

Area of Science:

• Computational Chemistry
• Machine Learning in Chemistry

Background:

• Standard surrogate models are commonly used for molecular equilibrium geometry optimization.
• Gaussian process regression (GPR) has emerged as a promising alternative.
• Gradient-enhanced Kriging (GEK) with specific optimizations shows competitive performance.

Purpose of the Study:

• To extend the GEK approach for constrained optimizations and transition state location.
• To benchmark the performance of the extended GEK method for chemical reactions.
• To compare the GEK method against standard techniques in challenging computational chemistry tasks.

Main Methods:

• Utilized gradient-enhanced Kriging (GEK) incorporating internal coordinates and restricted-variance optimization.
• Developed efficient hyperparameter estimation for GEK.
• Applied the method to constrained optimizations and transition state searches for various reactions.

Main Results:

• The extended GEK method demonstrated performance comparable to or better than standard methods.
• GEK showed superior efficiency and robustness in locating transition states.
• The method proved effective for both isolated constrained optimizations and reaction path computations.

Conclusions:

• The enhanced GEK approach is a highly efficient and robust tool for molecular geometry optimization, especially for constrained problems and transition state searches.
• This machine learning-based method offers significant advantages over traditional techniques in computational chemistry.
• The findings support the broader adoption of GPR in complex chemical modeling.