Research map / Probabilistic and scientific modeling / Variational and generative sampling
Physics vs Distributions: Pareto Optimal Flow Matching with Physics Constraints
Giacomo Baldan, Qiang Liu, Alberto Guardone, Nils Thuerey
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 with PBFM (residual loss, ConFIG, unrolling, stochastic sampler) instead of plain FM-OT | flow matching DiT · Darcy flow | residual MSE: 4.159 -> 0.838 | Improved |
| train with PBFM instead of plain FM-OT | flow matching DiT · Darcy flow | Wasserstein distance x1e2: 0.059 -> 0.138 | Hurt |
| train with PBFM instead of plain FM-OT | flow matching DiT · Kolmogorov flow | residual MSE x1e1: 2.314 -> 1.362 | Improved |
| train with PBFM instead of plain FM-OT | flow matching DiT · Dynamic stall | residual MSE x1e6: 11.02 -> 0.339 | Improved |
| use stochastic sampler (t*=0.2) instead of deterministic sampler (ablation) | PBFM · Darcy flow | Wasserstein distance x1e2: 1.470 -> 0.138 | Improved |
| increase noise resampling threshold t* from 0.2 to 1.0 (ablation) | PBFM stochastic sampler · Darcy flow | residual vs WD: RE 0.838 -> 0.632; WD 0.138 -> 0.316 | Mixed |
| add ConFIG conflict-free gradients alone (no unrolling) (ablation) | PBFM training · Darcy flow | residual MAE | Improved |
| add unrolling steps during training (ablation) | PBFM · Darcy flow | residual MAE | Improved |
| increase sigma_min from 0 to 1e-4/1e-3/1e-2 (ablation) | PBFM · Dynamic stall | residual and WD: RE 0.339 -> 0.473; WD 1.814 -> 3.246 (1e-3); MMSE 1.490 -> 1.200 (1e-2) | Mixed |
| increase unrolling steps from 1 to 4 (ablation) | PBFM training cost · Darcy flow | training memory GB: 12.62 -> 44.15 GB | Hurt |
| use PBFM instead of PCFM | inference · Dynamic stall | inference time ms: 3906 -> 60.47 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DiffusionPDE — DiffusionPDE: Generative PDE-Solving under Partial Observation (NeurIPS 2024) · better on 3 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.94)
- Chance-constrained Flow Matching for High-Fidelity Constraint-aware Generation (NeurIPS 2026 · similarity 0.89)
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.87)
- Flow Matching with General Discrete Paths: A Kinetic-Optimal Perspective (ICLR 2025 · similarity 0.87)
- Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling (NeurIPS 2025 · similarity 0.87)
- Physics-Informed Diffusion Models (ICLR 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.
