Research map / Probabilistic and scientific modeling / Differential equations and PDEs
LagrangeBench: A Lagrangian Fluid Mechanics Benchmarking Suite
Artur Toshev, Gianluca Galletti, Fabian Fritz, Stefan Adami, Nikolaus A. Adams
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 |
|---|---|---|---|
| train GNS-10-64 on 1% to 100% of the training data (ablation) | GNS-10-64 · All LagrangeBench datasets | position MSE | Improved |
| more training data (beyond ~10k frames) (ablation) | GNS-10-64 · 2D RPF and other datasets | position MSE | No clear effect |
| more training data (ablation) | GNS-10-64 · Dam break | position MSE | Improved |
| higher SEGNN tensor product order (L=2 vs L=1) (ablation) | SEGNN-10-64 · LagrangeBench datasets | accuracy vs runtime: 29.7 -> 86.0 | Mixed |
| use GNS instead of SEGNN on problems with boundaries (LDC, DAM) | baseline GNNs · 2D/3D LDC and DAM | MSE20 | Improved |
| use SEGNN (equivariant) instead of GNS on pure fluid problems (TGV, RPF) | baseline GNNs · TGV, 3D RPF | MSE20: 5.2e-3 | Improved |
| EGNN architecture instead of GNS or SEGNN | baseline GNNs · LagrangeBench datasets | training stability | Hurt |
| PaiNN extended to general vectorial inputs instead of GNS or SEGNN | baseline GNNs · LagrangeBench datasets | loss | Hurt |
| non-periodic displacement functions instead of periodic in EGNN (ablation) | EGNN-5-128 · inference time 2D (3.2K) and 3D (8.1K) | inference ms: 50.0 -> 3.49 (2D); 206 -> 15.1 (3D) | Improved |
| jaxmd_scan serialized neighbor list instead of jaxmd_vmap (ablation) | neighbor search · A6000 48GB memory limit | max particles that fit in memory: 1M -> 3.3M | Improved |
| matscipy CPU neighbor list instead of GPU JAX-MD (ablation) | SEGNN on 2D DAM · training speed | training time: +50% | Hurt |
| deeper/wider GNS (5-64 -> 10-64 -> 10-128) (ablation) | GNS · 2D inference (3.2K particles) | inference ms: 2.05 -> 3.89 -> 6.66 | Hurt |
Closest papers by meaning
- Neural SPH: Improved Neural Modeling of Lagrangian Fluid Dynamics (ICML 2024 · similarity 0.89)
- DeepLag: Discovering Deep Lagrangian Dynamics for Intuitive Fluid Prediction (NeurIPS 2024 · similarity 0.86)
- Learned Lagrangian Models of PDEs via Euler–Lagrange Residual Minimization (NeurIPS 2026 · similarity 0.86)
- Lagrangian Flow Networks for Conservation Laws (ICLR 2024 · similarity 0.86)
- Symmetric Basis Convolutions for Learning Lagrangian Fluid Mechanics (ICLR 2024 · similarity 0.86)
- Mosaic: A Benchmark Suite for Differentiable Physics Solvers (NeurIPS 2026 · similarity 0.85)
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.
