Nudged elastic band calculations accelerated with Gaussian process regression

Area of Science:

• Computational chemistry
• Materials science
• Statistical mechanics

Background:

• Minimum energy paths are crucial for understanding atomic and spin rearrangements in thermalized systems.
• The nudged elastic band method is commonly used but computationally intensive, especially with ab initio or DFT calculations.

Purpose of the Study:

• To reduce the computational effort required for minimum energy path calculations.
• To improve the efficiency of transition path calculations in thermalized systems.

Main Methods:

• Utilizing Gaussian process regression to generate and refine an approximate energy surface.
• Integrating pre-calculated Hessian matrices for enhanced stability in harmonic transition state theory.
• Employing uncertainty estimates from the Gaussian process model to guide calculations.

Main Results:

• Reduced energy and force evaluations by an order of magnitude compared to traditional methods.
• Achieved a further reduction of needed energy and force evaluations by half by focusing on uncertain regions of the path.
• Demonstrated methodology on the Müller-Brown potential and a 13-transition benchmark for heptamer island rearrangements.

Conclusions:

• Gaussian process regression offers a significant computational speedup for minimum energy path calculations.
• The method enhances efficiency and accuracy for determining transition paths in complex systems.
• This approach has broad applicability in computational studies of chemical and physical transformations.