Robust prediction of force chains in jammed solids using graph neural networks.

Area of Science:

• Physics
• Materials Science
• Computational Science

Background:

• Force chains are critical load-bearing structures in jammed amorphous materials like granular media and foams.
• Predicting force chain formation is essential for understanding material properties but remains challenging.
• Current methods often require direct force measurements or visualizations, which are not always feasible.

Purpose of the Study:

• To develop a predictive model for force chain locations in jammed materials.
• To leverage graph neural networks (GNNs) for accurate force chain prediction.
• To assess the robustness of GNN predictions across various material parameters.

Main Methods:

• Utilized graph neural networks (GNNs) to analyze the undeformed structure of jammed materials.
• Trained and tested GNNs on both frictionless and frictional material models.
• Investigated the impact of parameters like packing fraction, friction, and system size on prediction accuracy.

Main Results:

• GNNs accurately predicted force chain locations solely from the material's initial structure.
• Prediction accuracy remained high despite variations in packing fraction, composition, deformation, friction, system size, and interaction potential.
• Identified key structural features influencing GNN prediction performance.

Conclusions:

• Graph neural networks offer a powerful, data-driven approach to predict force chain formation in disordered materials.
• The GNN methodology provides a robust and versatile tool for analyzing stress propagation in jammed systems.
• This approach is valuable for systems where direct force measurements are impractical, advancing the study of granular matter and other disordered systems.