AI Paper Atlas

Research map / Probabilistic and scientific modeling / Dynamical systems

Energy-Guided Transport for Projection-Free Physics-Informed Flow Matching

Giuliano Galadini, Aurelio Uncini, Danilo Comminiello

NeurIPS 2026 · Probabilistic and scientific modeling › Dynamical systems · approach: Physics-informed dynamics learning · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
Replace explicit projection (PCFM) with continuous energy guidanceEGT vs PCFM · Navier-Stokes 2D partial guidanceinference time: 93.97 s -> 0.71 s per sampleImproved
Use continuous energy guidance under full-PDE constraints instead of Jacobian projectionEGT vs PCFM · Heat and Navier-Stokes full PDEsuccess 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-04Improved
Apply the same terminal refinement to unguided FFM instead of continuous guidance (ablation)Vanilla FFM · Navier-Stokes 2D full PDEphysical 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 PDEstability-accuracy trade-offImproved
Use spatially correlated smooth-kernel prior instead of white noise (ablation)EGT and baselines · Heat full PDEsuccess rateMixed
Use a large guidance scale gamma (ablation)EGT · Navier-Stokes 2DODE stiffnessHurt
Apply projection to partial BC/IC constraintsPCFM · RD and Burgers partial guidancephysical residual: 1.11e-08 and 1.19e-08Improved
Apply PCFM projection on Navier-StokesPCFM · Navier-Stokes partial guidancesample MSE: 0.54 -> 0.76Hurt

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy