Electronegativity Informed Graph Neural Networks for Superconducting Temperature Prediction with Generative Crystal...

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Accurate prediction of superconducting critical temperature (Tc) is crucial for discovering new superconducting materials.
• Data-driven approaches are increasingly important in materials discovery but face challenges in predictive accuracy.

Purpose of the Study:

• To develop and optimize a graph neural network (GNN) framework informed by electronegativity for predicting Tc.
• To compare the performance of different electronegativity descriptors within GNN models.
• To integrate the optimized GNN with crystal generative models for high-Tc material screening.

Main Methods:

• Development of an electronegativity (EN) informed GNN framework, comparing modified MEGNet and CGCNN models.
• Systematic evaluation of node-, global-, and edge-level EN descriptors, with a focus on nonlinear radial basis function encoding of edge EN differences.
• Bayesian hyperparameter optimization (HPO) for model refinement.
• Screening of generated crystal structures using CDVAE and CrystaLLM, followed by first-principles validation.

Main Results:

• The mCGCNN-EΔEN-rbf-HPO model achieved a test RMSE of 8.02 K and R² of 0.824, outperforming baseline models.
• Nonlinear radial basis function encoding of edge electronegativity differences demonstrated superior performance.
• Screening of 33,234 generated structures identified promising high-Tc candidates.
• First-principles analysis linked local Cu-O coordination and electronic states near the Fermi level to higher Tc.

Conclusions:

• The developed EN-informed GNN provides an accurate and transferable method for predicting Tc.
• The workflow effectively combines descriptor design, GNN prediction, crystal generation, and first-principles validation for accelerated materials discovery.
• This approach establishes a physically motivated pathway for identifying novel high-Tc superconducting materials.