Skip to content
DraconiticResearch
Search Sources API
197,768 studies31 sourcesUpdated 16 min ago
PreprintCC BY 4.0arXiv

Finite-Step Geometry of Marginal-Preserving Velocity Fields in Generative Flows

  • Ewsey Obzherin
  • Valentin Khrulkov
Open at sourcePDFDOICite

Abstract

Generative flows can admit different velocity fields that induce the same continuous marginal evolution but behave differently after numerical discretization. We study this finite-step non-equivalence while keeping the marginal path and numerical method fixed. Projected modified equations characterize the component of local integration error that changes the marginal law and give sufficient conditions for higher marginal convergence order. For affine fixed-shape distributions, we characterize exact finite-step reachability and derive spectral obstructions and sharp nearest-reachable Wasserstein errors. For diffusion probability-flow ODEs, we construct skew-score perturbations that preserve the exact marginal path under exact scores and quantify their leakage under score error. Controlled Gaussian and mixture experiments recover the predicted order, reachability, and obstruction phenomena.

Related

  1. Curvature measures and volume preserving flows in h-convex geometry
    Xinxiu Zhang
    arXivPreprintarXiv licenceFull textGeometry & topologyarXiv
  2. A structure preserving finite volume scheme for the multidimensional model in cardiac electrophysiology
    Jad Dabaghi, Fatima Khochman, Raafat Talhouk +1
    HALPreprintPublisher termsFull textNumerical analysis & scientific computingHAL
  3. A Structure-Preserving and Thermodynamically Compatible Cell-Centered Lagrangian Finite Volume Scheme for Continuum Mechanics
    Walter Boscheri, Michael Dumbser, Raphaël Loubère +1
    SIAM Journal on Numerical AnalysisPeer-reviewedLicence not statedFull text2 citedGeneral mathematicsHAL
  4. A Deterministic Game-Theoretic Approach to Convexity-Preserving Property and Finite-Time Extinction of Level Sets for Geometric Equations
    Yousuke Saga, Kazuhiro Takimoto
    arXivPreprintarXiv licenceFull textAnalysis & PDEsarXiv
  5. Convex Regression at Scale: Geometry-Preserving Constraint Sparsification
    Anish Senapati, Jose Blanchet
    arXivPreprintCC BY 4.0Full textOptimisation & controlStatisticsarXiv

Cite

BibTeXRIS
BibTeX
@misc{obzherin2026finitestep,
  title = {{Finite-Step Geometry of Marginal-Preserving Velocity Fields in Generative Flows}},
  author = {Ewsey Obzherin and Valentin Khrulkov},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.37936},
  eprint = {2610.37936},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.37936},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - Finite-Step Geometry of Marginal-Preserving Velocity Fields in Generative Flows
AU  - Ewsey Obzherin
AU  - Valentin Khrulkov
PY  - 2026
DA  - 2026/10/27
PB  - arXiv
DO  - 10.48550/arxiv.2610.37936
UR  - https://arxiv.org/abs/2610.37936
AB  - Generative flows can admit different velocity fields that induce the same continuous marginal evolution but behave differently after numerical discretization. We study this finite-step non-equivalence while keeping the marginal path and numerical method fixed. Projected modified equations characterize the component of local integration error that changes the marginal law and give sufficient conditions for higher marginal convergence order. For affine fixed-shape distributions, we characterize exact finite-step reachability and derive spectral obstructions and sharp nearest-reachable Wasserstein errors. For diffusion probability-flow ODEs, we construct skew-score perturbations that preserve the exact marginal path under exact scores and quantify their leakage under score error. Controlled Gaussian and mixture experiments recover the predicted order, reachability, and obstruction phenomena.
ER  -