Accelerating the search for global minima on potential energy surfaces using machine learning

Area of Science:

• Materials Science
• Computational Chemistry
• Surface Science

Background:

• Controlling molecule-surface interactions is crucial for applications like catalysis and gas sensing.
• Identifying stable adsorbate-surface structures requires finding the global minimum on potential energy surfaces.
• Current methods can be computationally intensive, necessitating faster approaches.

Purpose of the Study:

• To present a computational framework for accelerating the search for global minima on potential energy surfaces.
• To enable faster prediction of converged density functional theory (DFT) potential energies.
• To optimize the discovery of stable adsorbate-surface structures for chemical applications.

Main Methods:

• Developed a framework integrating Bayesian inference for predicting DFT potential energies.
• Utilized Bayesian optimization within the Bayesian Active Site Calculator.
• Applied global optimization methods to identify adsorption sites on material surfaces.

Main Results:

• Demonstrated the framework's performance on a hematite (Fe2O3) surface.
• Identified adsorption sites for hydrocarbons on the rutile TiO2 (110) surface.
• Showcased accelerated prediction of converged DFT energies with fewer self-consistent field iterations.

Conclusions:

• The presented framework significantly accelerates the search for stable adsorbate-surface structures.
• This approach enhances the efficiency of computational studies in catalysis and gas sensing.
• The method provides a powerful tool for discovering and characterizing molecule-surface interactions.