A transferable PINN-based method for quantum graphs with unseen structure

Paper i proceeding, 2025

This study introduces a transferable approach for solving partial differential equations (PDEs) on metric graphs, often called quantum graphs, employing Physics-Informed Neural Networks (PINNs). Unlike traditional solvers constrained by specific graph structures, our method utilizes a Neumann-Neumann domain decomposition technique, offering adaptability across various network topologies. By incorporating edge-wise surrogates, this approach achieves experimental results comparable to those obtained with FEM across diverse network configurations.