Research map / Probabilistic and scientific modeling / Variational and generative sampling
Physically-Informed Flow Matching with Graph Neural Networks for Complex Fluid Dynamics
Xiaozhuang Song, Tianshu Yu
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 |
|---|---|---|---|
| add the direct branch alongside the curl branch (ablation) | PIFM-GN · Ellipse (T=1) | R2: 0.975 -> 0.993 | Improved |
| add the curl-based potential branch alongside the direct branch (ablation) | PIFM-GN · Ellipse (T=1) | R2: 0.988 -> 0.993 | Improved |
| add RBF distance embeddings on edges (ablation) | PIFM-GN · Ellipse (T=1) | R2: 0.990 -> 0.993 | Improved |
| add curl branch on simple geometry (latent model) (ablation) | PIFM-GN (L) · Ellipse (T=1) | R2: 0.996 -> 0.996 | No clear effect |
| use flow matching instead of diffusion for single-step sampling | PIFM-GN · Wing (T=1) | R2: -2.522 -> 0.986 | Improved |
| use latent space (pretrained VGAE) for flow matching | PIFM-GN · Wing (T=1) CPU time | CPU seconds: 7.499 -> 1.085 | Improved |
| use latent variant instead of direct graph model | PIFM-GN · EllipseFlow (T=1) | R2: 0.974 -> 0.987 | Improved |
| increase integration steps from 10 to 50 | PIFM-GN (L) · EllipseFlow | R2: 0.996 -> 0.993 | Hurt |
| set curl-branch weight lambda to 0.25 instead of 0 or 1 (ablation) | PIFM-GN · Wing | R2 | Improved |
| add divergence-free physics constraint for OOD generalization | PIFM-GN (L) · Ellipse OOD (T=1) | R2: R2 above 0.989 on all OOD variants | Improved |
| increase training set size (ablation) | PIFM-GN · Ellipse (T=50) | W2 graph | Improved |
Closest papers by meaning
- Learning Distributions of Complex Fluid Simulations with Diffusion Graph Networks (ICLR 2025 · similarity 0.92)
- PIORF: Physics-Informed Ollivier-Ricci Flow for Long–Range Interactions in Mesh Graph Neural Networks (ICLR 2025 · similarity 0.89)
- MeshMask: Physics-Based Simulations with Masked Graph Neural Networks (ICLR 2025 · similarity 0.87)
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.87)
- Prometheus: Out-of-distribution Fluid Dynamics Modeling with Disentangled Graph ODE (ICML 2024 · similarity 0.87)
- From Zero to Turbulence: Generative Modeling for 3D Flow Simulation (ICLR 2024 · similarity 0.86)
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.
