Atomic Structure Optimization with Machine-Learning Enabled Interpolation between Chemical Elements

Area of Science:

• Computational materials science
• Chemical physics
• Machine learning applications

Background:

• Global optimization of atomic structures is crucial for discovering new materials with desired properties.
• Traditional methods often struggle with the high dimensionality and complexity of the energy landscape.
• Efficiently predicting stable atomic configurations remains a significant challenge.

Purpose of the Study:

• To introduce a novel computational method for the global optimization of structure and ordering in atomic systems.
• To leverage machine learning and Bayesian optimization for enhanced efficiency in materials discovery.
• To validate the method's performance across diverse material systems.

Main Methods:

• Development of a computational framework incorporating interpolation between chemical elements.
• Utilization of a machine-learning structural fingerprint to represent atomic configurations.
• Application of Bayesian optimization with Gaussian processes for efficient search.
• Testing on Au-Cu bulk systems, Cu-Ni surfaces with CO adsorption, and Cu-Ni clusters.

Main Results:

• The method consistently identifies low-energy atomic structures across various systems.
• Identified structures are highly likely to represent the global energy minima.
• Achieved significant reduction in required energy and force calculations (3-75) for systems of 23-66 atoms.

Conclusions:

• The proposed computational method offers an efficient and reliable approach for global structure optimization.
• This machine learning-driven strategy accelerates the discovery of stable atomic configurations.
• The method shows broad applicability for complex atomic systems in materials science.