Graph Neural Networks accelerate physics simulations for magnets and fracture mechanics

Researchers have developed novel graph neural network (GNN) frameworks to accelerate complex physics simulations. One approach, SuperCond-GNN, uses GNNs as a surrogate model to predict voltage distribution in superconducting magnets, achieving a 4.3% mean absolute percentage error and offering scalable inference for design and monitoring. Another hybrid GNN-FEM framework tackles phase-field fracture simulations by integrating a GNN surrogate into a conventional finite element method, significantly reducing computational cost while maintaining accuracy and generalization across diverse problem settings. AI

IMPACT These GNN-based surrogate models demonstrate potential for significant speedups in complex physics simulations, enabling faster design exploration and analysis in fields like materials science and engineering.

RANK_REASON The cluster contains two academic papers detailing new research methodologies using graph neural networks for scientific simulations.

Read on arXiv cs.LG →

• arXiv
• GNN-FEM
• graph neural network
• Phase field models
• SciML
• HTS magnets
• Kirchhoff's current law
• SuperCond-GNN

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →