Gaussian Process Regression for Minimum Energy Path Optimization and Transition State Search.

Area of Science:

• Computational chemistry
• Materials science
• Chemical physics

Background:

• Determining reaction pathways and transition states is crucial for understanding chemical reactions and material properties.
• Traditional methods like the nudged elastic band (NEB) can be computationally expensive, requiring numerous energy and gradient evaluations.
• Efficient algorithms are needed to accelerate the exploration of potential energy surfaces.

Purpose of the Study:

• To introduce and detail a novel gradient-based algorithm for calculating minimum energy paths (MEPs).
• To assess the performance of the new algorithm against the established nudged elastic band (NEB) method.
• To evaluate the impact of different initial path guesses on the algorithm's efficiency.

Main Methods:

• Implementation of a gradient-based optimization algorithm.
• Utilizing Gaussian process regression (GPR) for efficient energy and gradient approximations.
• Comparative performance analysis against the nudged elastic band (NEB) method across 27 diverse test systems.
• Evaluation of three distinct strategies for providing an initial guess of the minimum energy path.

Main Results:

• The developed GPR-based algorithm significantly reduces the number of required energy and gradient evaluations.
• The algorithm demonstrates competitive or superior performance compared to the NEB method in the tested systems.
• The choice of initial path guess influences the efficiency but the algorithm remains effective across different starting points.
• Fast subsequent searches for transition states are enabled by the MEP calculation.

Conclusions:

• The novel GPR-based gradient algorithm offers a more computationally efficient approach to finding minimum energy paths.
• This method accelerates the characterization of reaction mechanisms and transition states in chemical and materials systems.
• The algorithm provides a valuable tool for researchers seeking to reduce computational costs in exploring potential energy landscapes.