Hydrogen, Oxygen, and Lead Adsorbates on Al₁₃Co₄(100): Accurate Potential Energy Surfaces at Low Computational Cost...

Area of Science:

• Materials Science
• Computational Chemistry
• Surface Science

Background:

• Intermetallic compounds are crucial for surface interactions like catalysis.
• Adsorption energy maps are vital for understanding surface properties but computationally expensive.
• Current methods require significant computational resources for exploring adsorption energy landscapes.

Purpose of the Study:

• To develop an efficient Machine Learning (ML) method for predicting adsorption energies on intermetallic surfaces.
• To reduce the computational cost associated with building adsorption energy maps.
• To accelerate the discovery of novel functional materials through improved surface property prediction.

Main Methods:

• Utilized a Machine Learning (ML) scheme incorporating Density Functional Theory (DFT) calculations.
• Employed Farthest Point Sampling (FPS) to select a minimal set of sites for DFT estimates.
• Applied the FPS-ML method to the Al13Co4(100) surface with H, O, and Pb adsorbates.

Main Results:

• The FPS-ML method demonstrated superior performance compared to interpolation and MACE-ML, particularly with fewer data points (n<36).
• Achieved higher correlation (Pearson R-factor up to 0.90 for Pb) between predicted and DFT adsorption energies.
• Reported lower unbiased root-mean-square error (ubRMSE) for FPS-ML predictions across various adsorbates.

Conclusions:

• The proposed FPS-ML methodology offers an efficient and accurate approach for predicting adsorption energies.
• This method can significantly reduce computational expenses in materials science research.
• The findings are extendable to diverse systems, potentially driving the discovery of new functional materials.