Graph Neural Network Guided Evolutionary Search of Grain Boundaries in 2D Materials

Area of Science:

• Materials Science
• Computational Materials Science
• Nanotechnology

Background:

• Grain boundaries (GBs) in two-dimensional (2D) materials significantly influence their physical, chemical, mechanical, electronic, and optical properties.
• Predicting realistic GB structures is essential for tailoring material properties but is challenging due to the vast search space.
• Current methods for discovering 2D material GBs are computationally intensive and complex.

Purpose of the Study:

• To develop an efficient and generalizable workflow for the discovery and design of novel 2D lateral grain boundary (GB) structures.
• To introduce a hybrid approach combining Graph Neural Networks (GNNs) with evolutionary algorithms for predicting 2D interface structures.
• To demonstrate the accuracy and efficiency of the proposed method using blue phosphorene (BP) as a model system.

Main Methods:

• A workflow combining Graph Neural Network (GNN) and an evolutionary algorithm was developed for 2D lateral interface discovery.
• The GNN model was trained using a machine learning bond order potential (Tersoff formalism) and density functional theory (DFT) data.
• The GNN was coupled with a multiobjective genetic algorithm (MOGA) to predict GB structures.

Main Results:

• The GNN model achieved high accuracy, predicting structural energy with under 0.5% mean absolute error using sparse training data (<2000 DFT labels).
• Systematic downsampling confirmed the model's ability to predict energies accurately even with limited training data.
• The coupled GNN-MOGA approach demonstrated strong accuracy in predicting 2D GB structures.

Conclusions:

• The developed GNN-based workflow efficiently predicts and designs novel 2D grain boundary structures.
• This material-agnostic method is anticipated to significantly accelerate the discovery of 2D GB structures and their applications.
• The approach offers a powerful tool for controlling material properties through precise interface engineering.