Combining Evolutionary Algorithms with Clustering toward Rational Global Structure Optimization at the Atomic Scale.

Area of Science:

• Computational Materials Science
• Materials Informatics
• Machine Learning in Chemistry

Background:

• Accurate atomic-scale structure prediction is crucial for understanding material properties.
• Existing global optimization techniques often generate excessive intermediate data, hindering efficiency.
• Analyzing this data during optimization could enable more rational search algorithms.

Purpose of the Study:

• To develop a rational algorithm for global structure optimization using machine learning.
• To enhance the efficiency of evolutionary algorithms (EA) by incorporating clustering techniques.

Main Methods:

• Combined an evolutionary algorithm (EA) with clustering, a machine learning technique.
• Clustered the configuration space of intermediate structures into geometrically similar regions.
• Enabled the EA to suppress irrelevant regions and favor promising ones during optimization.

Main Results:

• Demonstrated significantly faster global minimum searches for organic molecules and oxide surfaces.
• The clustering-enhanced EA effectively favored stable structures in unexplored regions.
• Showcased the potential for adaptive global optimization based on accumulated data.

Conclusions:

• The developed clustering-enhanced EA offers a more efficient approach to global structure optimization.
• This method represents a significant step towards adaptive optimization techniques.
• Facilitates faster and more rational discovery of atomic-scale material structures.