Symm-CGNN: Symmetry-Information-Enhanced Crystal Graph Neural Network for High-Symmetry Point Band Gap Prediction

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Accurate characterization of crystal optoelectronic properties requires precise band gap determination at high-symmetry points.
• Conventional Graph Neural Networks (GNNs) lack global symmetry information, limiting their ability to differentiate specific energy states.

Purpose of the Study:

• To develop a novel machine learning model that integrates local atomic environments with global symmetry information for enhanced crystal property prediction.
• To improve the accuracy of band gap predictions, particularly for high-symmetry points relevant to anisotropic optoelectronic properties.

Main Methods:

• Proposed the Symmetry-Information-Enhanced Crystal Graph Neural Network (Symm-CGNN) model.
• Incorporated global symmetry information (space groups, crystal systems, density, lattice constants) into the GNN architecture.
• Evaluated the model on comprehensive 3D (Materials Project) and 2D (2DMatpedia) datasets.

Main Results:

• Symm-CGNN demonstrated an 18% reduction in Mean Absolute Error (MAE) for high-symmetry band gap prediction compared to baseline Crystal Graph Convolutional Neural Networks (CGCNN).
• The model effectively bridges the gap between local atomic coordination and macroscopic crystal symmetry.

Conclusions:

• Symm-CGNN offers a robust and efficient machine learning approach for predicting crystal band gaps.
• This method facilitates high-throughput screening of materials with desirable anisotropic optoelectronic properties.