Robust Gaussian Process Regression Method for Efficient Tunneling Pathway Optimization: Application to Surface...

Area of Science:

• Computational chemistry
• Surface science
• Theoretical chemistry

Background:

• Simulating surface processes provides atomic-scale insights into heterogeneous catalysis, diffusion, and quantum tunneling.
• Optimization of reaction pathways, including semiclassical tunneling (instantons), is computationally intensive, especially with ab initio electronic structure calculations.
• Existing machine learning methods for reaction pathway optimization face numerical and efficiency challenges, particularly for condensed-phase reactions.

Purpose of the Study:

• To develop an improved, efficient, and numerically stable computational framework for simulating surface processes.
• To enhance the accuracy and applicability of machine learning in modeling chemical reactions, especially condensed-phase ones.
• To reduce the computational cost of ab initio instanton optimizations.

Main Methods:

• Developed an enhanced framework using Gaussian process regression for general transformed coordinates.
• Introduced a novel descriptor combining internal and Cartesian coordinates for modeling surface processes.
• Applied the framework to 11 instanton optimizations across three representative systems.

Main Results:

• The proposed method significantly improves efficiency and numerical stability compared to previous approaches.
• The new descriptor is suitable for modeling surface processes.
• Ab initio instanton optimization using the improved framework is computationally much cheaper, approaching the cost of classical transition-state theory calculations.

Conclusions:

• The enhanced framework offers a more cost-effective and stable approach for ab initio instanton optimizations.
• This advancement facilitates more extensive simulations of surface processes and chemical reactions.
• The method shows great potential for applications in heterogeneous catalysis and condensed-phase dynamics.