Research map / Probabilistic and scientific modeling / Dynamical systems
Learning Mesh-Based Simulation with Graph Networks
Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, Peter Battaglia
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use mesh-space positions and pass messages along mesh edges rather than using particle-style fixed-radius connectivity (ablation) | MeshGraphNets on irregular FlagSimple meshes · FlagSimple | rollout stability and error | Improved |
| add world-space edges to mesh-space message passing (ablation) | MeshGraphNets · FlagDynamic and SphereDynamic | rollout RMSE: +51% and +92% RMSE without world edges | Improved |
| use relative positional encoding on edges instead of absolute positions as node features (ablation) | MeshGraphNets · Airfoil | rollout RMSE: main model vs $26.5$ | Improved |
| increase the number of graph message-passing blocks (ablation) | MeshGraphNets · evaluated systems | performance | Improved |
| add history beyond the shortest task-appropriate history (ablation) | MeshGraphNets · FlagDynamic | accuracy | Hurt |
| learn and apply adaptive remeshing during rollout (ablation) | MeshGraphNets · FlagDynamic and SphereDynamic | rollout performance | No clear effect |
| train the sizing model from estimated sizing targets rather than labeled sizing data (ablation) | MeshGraphNets learned remesher · FlagDynamic and SphereDynamic | rollout performance | No clear effect |
| use MeshGraphNets instead of GCN | AirfoilSteady · AirfoilSteady | RMSE: $0.159$ -> $0.116$ | Improved |
| generalize to steeper angles than training | MeshGraphNets · Airfoil | RMSE: $11.5$ -> $12.4$ | Hurt |
| generalize to higher inflow speeds than training | MeshGraphNets · Airfoil | RMSE: $11.5$ -> $13.1$ | Hurt |
Closest papers by meaning
- EvoMesh: Adaptive Physical Simulation with Hierarchical Graph Evolutions (ICML 2025 · similarity 0.90)
- Learning Distributions of Complex Fluid Simulations with Diffusion Graph Networks (ICLR 2025 · similarity 0.89)
- MeshMask: Physics-Based Simulations with Masked Graph Neural Networks (ICLR 2025 · similarity 0.89)
- MaNGO — Adaptable Graph Network Simulators via Meta-Learning (NeurIPS 2025 · similarity 0.88)
- PIORF: Physics-Informed Ollivier-Ricci Flow for Long–Range Interactions in Mesh Graph Neural Networks (ICLR 2025 · similarity 0.87)
- Diffusion-Based Hierarchical Graph Neural Networks for Simulating Nonlinear Solid Mechanics (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
