Geometric Guidance Integrated with Directed Electrostatics Strategy within a Graph Neural Network Approach for...

Area of Science:

• Computational chemistry and materials science
• Nanotechnology and materials engineering
• Artificial intelligence in scientific discovery

Background:

• Predicting stable nanocluster configurations is crucial for materials science.
• Existing methods struggle with large atomic clusters and specific symmetries.
• Graph Neural Networks offer a powerful framework for complex structural predictions.

Purpose of the Study:

• To develop an efficient computational method for predicting stable nanocluster structures.
• To integrate geometric and electronic strategies for enhanced prediction accuracy.
• To identify new symmetric isomers of medium to large nanoclusters.

Main Methods:

• Introduction of the Geometric-DESIGNN method, integrating Geometric Guidance and Directed Electrostatics Strategy.
• Utilizing a Graph Neural Network framework for predicting nanocluster configurations on potential energy surfaces.
• Alternating geometric and DESIGNN building strategies for shell-by-shell cluster construction.

Main Results:

• Successful prediction of stable configurations for Mg_n clusters up to n < 561.
• Identification of new symmetric isomers for Mg_n clusters (n < 150) by constraining point-group symmetry.
• Construction of stable Mg_n nanoclusters for specific large sizes (n = 332, 338, 561).

Conclusions:

• Geometric-DESIGNN is an efficient tool for accelerated nanocluster structure prediction.
• The method effectively merges geometric and electronic strategies for accurate structural modeling.
• This approach facilitates the discovery of novel atomic cluster structures with desired symmetries.