This repository contains the official implementation of the paper PIBNet and its improvement accepted at the 1st Workshop on Differentiable Systems and Scientific Machine Learning @ EurIPS 2025.
Clone repository and install libraries in requirements.txt. Please refer to DGL installation page to install dgl.
sh scripts/gen_laplace_dirichlet.sh
sh scripts/gen_helmholtz_dirichlet.sh
sh scripts/gen_helmholtz_neumann.sh
sh scripts/train_pibnet.sh
sh scripts/test_pibnet.sh
sh scripts/train_improved_pibnet.sh
sh scripts/test_improved_pibnet.sh
We would like to thank the authors of Multiscale Neural Operators for Solving Time-Independent PDEs for their great work and for sharing their code and the authors of the NVIDIA PhysicsNeMo implementation of MeshGraphNet.