Research map / Probabilistic and scientific modeling / Dynamical systems
Energy-Guided Transport for Projection-Free Physics-Informed Flow Matching
Giuliano Galadini, Aurelio Uncini, Danilo Comminiello
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes an energy-guided transport approach for physics-informed flow matching that enforces physical constraints at inference without projection (based only on the abstract).
Problem
Flow Matching generative models for physical systems are agnostic to governing equations and often violate physical constraints during inference; existing zero-shot constraint enforcement relies on rigid algebraic projections.
Approach
Replaces rigid algebraic projection steps with an energy-guided transport mechanism that steers flow matching sampling toward physically consistent outputs, avoiding explicit projection.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace explicit projection (PCFM) with continuous energy guidance | EGT vs PCFM · Navier-Stokes 2D partial guidance | inference time: 93.97 s -> 0.71 s per sample | Improved |
| Use continuous energy guidance under full-PDE constraints instead of Jacobian projection | EGT vs PCFM · Heat and Navier-Stokes full PDE | success rate: 0% -> 100% | Improved |
| Add terminal energy-descent refinement (Hybrid EGT) (ablation) | EGT · Burgers full PDE (white noise) | physical residual: 1.92e-03 -> 1.49e-04 | Improved |
| Apply the same terminal refinement to unguided FFM instead of continuous guidance (ablation) | Vanilla FFM · Navier-Stokes 2D full PDE | physical MAE: 7.50e-3 -> 3.96e-3 (vs 1.60e-3 for pure EGT) | Mixed |
| Reduce ODE steps N from 200 to about 50 (ablation) | EGT · Burgers full PDE | stability-accuracy trade-off | Improved |
| Use spatially correlated smooth-kernel prior instead of white noise (ablation) | EGT and baselines · Heat full PDE | success rate | Mixed |
| Use a large guidance scale gamma (ablation) | EGT · Navier-Stokes 2D | ODE stiffness | Hurt |
| Apply projection to partial BC/IC constraints | PCFM · RD and Burgers partial guidance | physical residual: 1.11e-08 and 1.19e-08 | Improved |
| Apply PCFM projection on Navier-Stokes | PCFM · Navier-Stokes partial guidance | sample MSE: 0.54 -> 0.76 | Hurt |
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 2 of 2 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.
Experimental setup
Builds on: Flow Matching
Closest papers by meaning
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.93)
- Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling (NeurIPS 2025 · similarity 0.87)
- Training Free Guided Flow-Matching with Optimal Control (ICLR 2025 · similarity 0.87)
- Gradient-Free Generation for Hard-Constrained Systems (ICLR 2025 · similarity 0.87)
- Chance-constrained Flow Matching for High-Fidelity Constraint-aware Generation (NeurIPS 2026 · similarity 0.87)
- Generating Physical Dynamics under Priors (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.
