Machine learning with bond information for local structure optimizations in surface science.

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Adsorption systems involve complex interactions at multiple scales, from molecular to substrate levels.
• Optimizing these systems is crucial for applications in catalysis, separation, and energy storage.
• Current machine learning methods for optimization can be computationally intensive.

Purpose of the Study:

• To enhance machine learning-based optimization of adsorption systems.
• To improve the efficiency and accuracy of finding local minima in molecular-substrate interactions.
• To demonstrate the benefits of incorporating bond characteristics into optimization algorithms.

Main Methods:

• Development of an anisotropic kernel within the Gaussian process regression framework.
• Explicit modeling of bond characteristics at different scales (within substrate, molecule, and between them).
• Implementation of a limited memory approach for computational efficiency.

Main Results:

• The anisotropic kernel significantly improves the performance of machine learning optimization for adsorption systems.
• The method demonstrates a speed-up of up to a factor of two compared to standard optimization techniques.
• A limited memory approach further reduces computational resources and calculation costs.

Conclusions:

• Explicitly modeling bond characteristics with an anisotropic kernel enhances machine learning optimization for adsorption systems.
• The proposed method offers a computationally efficient and effective approach for exploring local minima in atomic systems.
• This work paves the way for more advanced and efficient computational studies in materials science and chemistry.