A New Genetic Algorithm Approach Applied to Atomic and Molecular Cluster Studies.

Area of Science:

• Computational Chemistry
• Materials Science
• Artificial Intelligence

Background:

• Predicting nanoparticle structures is crucial for designing materials with specific properties.
• Genetic algorithms (GAs) are powerful tools for structure prediction but can be improved.
• Operator efficiency significantly impacts the performance of GAs in optimization tasks.

Purpose of the Study:

• To develop and evaluate a novel procedure for improving genetic algorithms in nanoparticle structure prediction.
• To dynamically manage the creation rate of new individuals based on operator performance.
• To integrate and assess various optimization strategies within a GA framework.

Main Methods:

• A new strategy was implemented to manage the efficiency of 13 different genetic algorithm operators.
• Operators with higher performance in generating well-adapted offspring were favored.
• Tested on 26 and 55-atom clusters (Lennard-Jones potential), 18-atom carbon clusters (REBO potential), and polynitrogen systems (quantum methods).

Main Results:

• The proposed management strategy effectively avoids underperforming operators, maintaining high confidence in the overall method.
• The 'twist' operator demonstrated superior speed in generating new individuals compared to standard operators like Deaven and Ho cut-and-splice crossover.
• Operators adapted from basin-hopping methodologies also showed strong performance within the enhanced GA scheme.

Conclusions:

• The developed operator management strategy significantly enhances the efficiency of genetic algorithms for nanoparticle structure prediction.
• The 'twist' operator emerges as a highly effective tool for individual generation in this context.
• This approach offers a robust and efficient method for computational materials discovery.