Combining Deep Learning Neural Networks with Genetic Algorithms to Map Nanocluster Configuration Spaces with Quantum...

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Exploring configuration spaces of bimetallic nanoclusters is computationally intensive.
• Accurate prediction of stable nanocluster structures requires quantum mechanical precision.
• Existing methods struggle with the vastness of configuration spaces.

Purpose of the Study:

• To develop an efficient and accurate method for exploring configuration spaces of bimetallic nanoclusters.
• To combine genetic algorithms with neural networks trained on DFT.
• To provide a computationally practical solution for searching complex energy landscapes.

Main Methods:

• Utilizing genetic algorithms coupled with neural networks.
• Training neural networks on density functional theory (DFT) data.
• Implementing a machine learning algorithm to learn DFT potentials and generate/relax structures.
• Applying the method to bimetallic gold-palladium (AuPd) nanoclusters of sizes 15, 20, and 25 atoms.

Main Results:

• The combined approach efficiently maps the configuration space and converges to energy minima.
• The machine learning algorithm demonstrates increasing performance and unbiased structure generation.
• Demonstrated computational efficiency for AuPd nanoclusters.
• Identified physical insights like geometric gold surface segregation and stoichiometric gold minimization.

Conclusions:

• The methodology offers an optimizable and computationally practical solution for exploring vast configuration spaces.
• The approach is scalable and applicable to various systems beyond nanoclusters (bulk, surfaces, adsorption).
• The study provides valuable insights into the stability and configuration of bimetallic nanoclusters.