RIGNO: A Graph-based framework for robust and accurate operator learning for PDEs on arbitrary domains
Research paper by Sepehr Mousavi and collaborators. arXiv record dated 2025-10-14. Research area: Machine learning.
Resource information
- Authors
- Sepehr Mousavi, Shizheng Wen, Levi Lingsch, Maximilian Herde, Bogdan Raonić, Siddhartha Mishra
- Research area
- Machine learning
- Application domain
- General AI research
- AI methods
- Neural networks
- Research topics
- cs.LG
- arXiv record date
- 2025-10-14
- Publication
- arXiv
- Upstream last modified
- Oct 15, 2025
- License
- CC BY 4.0
- Last record update
- Nov 4, 2026
Abstract
Learning the solution operators of PDEs on arbitrary domains is challenging due to the diversity of possible domain shapes, in addition to the often intricate underlying physics. We propose an end-to-end graph neural network (GNN) based neural operator to learn PDE solution operators from data on point clouds in arbitrary domains. Our multi-scale model maps data between input/output point clouds by passing it through a downsampled regional mesh. The approach includes novel elements aimed at ensuring spatio-temporal resolution invariance. Our model, termed RIGNO, is tested on a challenging suite of benchmarks composed of various time-dependent and steady PDEs defined on a diverse set of domains. We demonstrate that RIGNO is significantly more accurate than neural operator baselines and robustly generalizes to unseen resolutions both in space and in time. Our code is publicly available at github.com/camlab-ethz/rigno.