pith. sign in

arxiv: 2506.12755 · v2 · submitted 2025-06-15 · 🧮 math.PR

Stochastic intrinsic gradient flows on the Wasserstein space

Panpan Ren , Michael R\"ockner , Feng-Yu Wang , Simon Wittmann This is my paper

Pith reviewed 2026-05-19 09:56 UTC · model grok-4.3

classification 🧮 math.PR
keywords Wasserstein spacestochastic gradient flowsDirichlet formsintrinsic gradientmartingale solutionsGaussian reference measuresporous media equationsenergy functionals
0
0 comments X

The pith

Gaussian-based measures and Dirichlet forms allow construction of martingale solutions to stochastic intrinsic gradient flows on the 2-Wasserstein space.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper constructs stochastic versions of gradient flows for energy functionals on the space of probability measures with finite second moments. These functionals take the integral form of F applied to position and density, covering cases such as entropy and Lyapunov functions for porous media equations. The approach begins with Gaussian reference measures on this space that support symmetric Markov processes. Perturbations of these measures by a factor proportional to the exponential of the negative energy are then quantized using Dirichlet form methods to produce the desired stochastic dynamics. The result is a process that incorporates both the intrinsic gradient of the energy and additive noise from the reference process.

Core claim

The intrinsic gradient DW_F of the energy is defined for almost every measure with respect to the Gaussian reference Lambda, and suitable choices of Lambda make the resulting distorted process a martingale solution to the equation d mu_t equals minus DW_F(mu_t) dt plus d R_t.

What carries the argument

The intrinsic gradient DW_F(mu) defined Lambda-almost everywhere, obtained by perturbing Gaussian reference measures via Dirichlet forms with density proportional to e to the minus W_F.

If this is right

  • The construction directly covers the entropy functional and more general Lyapunov functions for generalized porous media equations.
  • Stochastic quantization is achieved for these energy functionals through the perturbed Markov processes.
  • The resulting dynamics combine deterministic descent along the intrinsic gradient with stochastic increments from the reference process R_t.
  • The method supplies a probabilistic representation of solutions to the associated deterministic gradient flow equations in the presence of noise.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar Dirichlet-form constructions could be attempted with non-Gaussian reference measures to handle energies with different growth or support properties.
  • The martingale solutions might be used to study long-time behavior or ergodicity questions for stochastic versions of porous media equations.
  • Numerical approximations of the reference processes could provide simulation schemes for sampling measures weighted by such energies.

Load-bearing premise

The functions F and its partial derivative with respect to the second argument are locally Lipschitz on R^d times the positive reals.

What would settle it

A concrete counterexample energy functional satisfying the local Lipschitz condition for which no Gaussian reference measure Lambda makes the constructed process satisfy the martingale solution property for the stochastic gradient equation.

read the original abstract

We construct stochastic gradient flows on the $2$-Wasserstein space $\mathcal P_2$ over $\mathbb R^d$ for energy functionals of the type $W_F(\rho d x)=\int_{\mathbb R^d}F(x,\rho(x))d x$. The functions $F$ and $\partial_2 F$ are assumed to be locally Lipschitz on $\mathbb R^d\times (0,\infty)$. This includes the relevant examples of $W_F$ as the entropy functional or more generally the Lyapunov function of generalized porous media equations. First we define a class of Gaussian-based measures $\Lambda$ on $\mathcal P_2$ together with a corresponding class of symmetric Markov processes ${(R_t)}_{t\geq 0}$. Next, using Dirichlet form techniques we perform stochastic quantization for the perturbations of these objects which result from multiplying such a measure $\Lambda$ by a density proportional to $e^{-W_F}$. Finally we show that the intrinsic gradient $DW_F(\mu)$ is defined for $\Lambda$-a.e. $\mu$ and that the Gaussian-based reference measure $\Lambda$ can be chosen in such way that the distorted process ${(\mu_t)}_{t\geq 0}$ is a martingale solution for the equation $d\mu_t=-DW_F(\mu_t) d t+d R_t$, $t\geq 0$.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper constructs stochastic intrinsic gradient flows on the 2-Wasserstein space P_2(R^d) for energy functionals W_F(ρ dx) = ∫ F(x, ρ(x)) dx where F and ∂_2 F are locally Lipschitz. It introduces a class of Gaussian-based reference measures Λ on P_2 together with symmetric Markov processes (R_t), perturbs the associated Dirichlet form by the density e^{-W_F}, and claims that the resulting distorted process (μ_t) is a martingale solution to the equation dμ_t = -D W_F(μ_t) dt + d R_t for Λ-a.e. starting points.

Significance. If the technical gaps are closed, the result supplies a Dirichlet-form-based stochastic quantization procedure for Wasserstein gradient flows that applies to entropy and generalized porous-media energies. The use of Gaussian-based Λ to begin from a known regular form is a concrete strength; the construction yields a martingale solution rather than a strong solution, which is appropriate for the non-locally-compact setting.

major comments (1)
  1. [Section 3 (perturbation of the Dirichlet form) and Section 4 (martingale-solution property)] The central existence claim for the distorted process (μ_t) as a Markov process satisfying the martingale problem relies on the perturbed Dirichlet form being quasi-regular (or at least having a suitable core with full capacity). No explicit verification of quasi-regularity on the non-locally-compact space P_2 is supplied after the perturbation by e^{-W_F}; the appeal to Ma-Röckner theory therefore remains formal. This step is load-bearing for the existence of (μ_t) and for the identification of D W_F as the intrinsic gradient.
minor comments (2)
  1. [Introduction / Section 2] The notation for the intrinsic gradient D W_F(μ) is introduced without an explicit formula or domain statement; a displayed expression relating it to the first variation of W_F would improve readability.
  2. [Section 2] The precise choice of the Gaussian-based measure Λ (e.g., the covariance parameter or the centering) is described only qualitatively; an explicit parametrization would make the dependence on the reference process (R_t) easier to track.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. The primary concern regarding the lack of explicit verification of quasi-regularity for the perturbed Dirichlet form is well-taken, and we address it directly below. We will incorporate the necessary additions in the revised version.

read point-by-point responses

  1. Referee: [Section 3 (perturbation of the Dirichlet form) and Section 4 (martingale-solution property)] The central existence claim for the distorted process (μ_t) as a Markov process satisfying the martingale problem relies on the perturbed Dirichlet form being quasi-regular (or at least having a suitable core with full capacity). No explicit verification of quasi-regularity on the non-locally-compact space P_2 is supplied after the perturbation by e^{-W_F}; the appeal to Ma-Röckner theory therefore remains formal. This step is load-bearing for the existence of (μ_t) and for the identification of D W_F as the intrinsic gradient.

    Authors: We agree that the current manuscript appeals to the general Ma-Röckner theory without supplying a self-contained verification of quasi-regularity for the perturbed form on the non-locally-compact space P_2. In the revised version we will insert a new subsection (approximately 1.5 pages) in Section 3 that explicitly checks the conditions. The argument proceeds in two steps: first, the Gaussian-based reference measures Λ are constructed so that the unperturbed Dirichlet form is quasi-regular with a core of cylinder functions having full capacity; second, the local Lipschitz assumptions on F and ∂_2 F ensure that the density e^{-W_F} belongs to the Kato class associated with the reference form and that the perturbation does not destroy the quasi-regularity or the existence of a suitable core. These estimates rely on the explicit Gaussian structure of Λ and on the fact that W_F is locally bounded on sets of positive capacity. With this addition the existence of the associated Markov process (μ_t) and the identification of DW_F as the intrinsic gradient become fully rigorous. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper constructs the stochastic intrinsic gradient flow by first introducing a class of Gaussian-based reference measures Λ on P_2 together with symmetric Markov processes (R_t), then perturbing the associated Dirichlet form by the density proportional to e^{-W_F} under the local Lipschitz assumption on F and ∂_2 F. It subsequently verifies that the intrinsic gradient DW_F(μ) exists for Λ-almost every μ and that a suitable choice of Λ yields a distorted process (μ_t) satisfying the martingale problem for the target equation dμ_t = -DW_F(μ_t) dt + dR_t. This proceeds via standard Dirichlet-form perturbation techniques applied to an externally chosen reference object; no step defines the target martingale solution in terms of itself, renames a fitted quantity as a prediction, or reduces the central claim to a load-bearing self-citation whose content is unverified outside the paper. The derivation remains self-contained against external benchmarks from Dirichlet form theory and Wasserstein geometry.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The paper rests on background results from stochastic analysis and optimal transport; the main additions are the specific class of reference measures and the quantization step.

axioms (2)
  • domain assumption Existence and symmetry of Markov processes associated with the Gaussian-based measures Lambda on P2
    Invoked to define the reference noise process R_t before tilting by the energy.
  • domain assumption Local Lipschitz continuity of F and its second partial derivative
    Stated in the abstract as the condition ensuring W_F and the perturbations are well-defined.
invented entities (1)
  • Class of Gaussian-based measures Lambda on P2 no independent evidence
    purpose: Reference measures for stochastic quantization and definition of the driving noise
    Newly introduced class to enable the Dirichlet-form construction on the Wasserstein space.

pith-pipeline@v0.9.0 · 5776 in / 1603 out tokens · 41149 ms · 2026-05-19T09:56:07.268196+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    Differential geometry of Poisson spaces.C

    Albeverio, S., Kondratiev, Y., R¨ ockner, M. Differential geometry of Poisson spaces.C. R. Acad. Sci. Paris S´ er. I Math.323, 1129-1134 (1996)

    work page 1996

  2. [2]

    Analysis and geometry on Configuration Spaces

    Albeverio, S., Kondratiev, Y., R¨ ockner, M. Analysis and geometry on Configuration Spaces. J. Funct. Anal. 154, 444-500 (1998)

    work page 1998

  3. [3]

    Classical Dirichlet forms on topological vector spaces − Closability and a Cameron–Martin formula

    Albeverio, S., R¨ ockner, M. Classical Dirichlet forms on topological vector spaces − Closability and a Cameron–Martin formula. J. Funct. Anal. 88, 395-436 (1990)

    work page 1990

  4. [4]

    Stochastic differential equations in infinite dimensions: solutions via Dirichlet forms

    Albeverio, S., R¨ ockner, M. Stochastic differential equations in infinite dimensions: solutions via Dirichlet forms. Probab. Th. Rel. Fields 89, 347-386 (1991)

    work page 1991

  5. [5]

    Gradient flows in metric spaces and in the space of probability measures

    Ambrosio, L., Gigli, N., Savar´ e, G. Gradient flows in metric spaces and in the space of probability measures. Birkh¨ auser Verlag, Basel, 2005

    work page 2005

  6. [6]

    Bismut formula for Lions derivative of distribution-path dependent SDEs

    Bao, J., Ren, P., Wang, F. Bismut formula for Lions derivative of distribution-path dependent SDEs. J. Differ. Equ. 282, 285-329 (2021)

    work page 2021

  7. [7]

    The evolution to equilibrium of solutions to nonlinear Fokker–Planck equation

    Barbu, V., R¨ ockner, M. The evolution to equilibrium of solutions to nonlinear Fokker–Planck equation. Indiana Univ. Math. J. 72, 89–131 (2023)

    work page 2023

  8. [8]

    Uniqueness for nonlinear Fokker–Planck equations and for McKean–Vlasov SDEs: The degenerate case

    Barbu, V., R¨ ockner, M. Uniqueness for nonlinear Fokker–Planck equations and for McKean–Vlasov SDEs: The degenerate case. J. Funct. Anal. 285, 109980 (2023). 29

    work page 2023

  9. [9]

    Nonlinear Fokker–Planck Flows and their Probabilistic Counterparts

    Barbu, V., R¨ ockner, M. Nonlinear Fokker–Planck Flows and their Probabilistic Counterparts. Springer, 2024

    work page 2024

  10. [10]

    Potential Anal

    Baur, B., Grothaus, M., Construction and Strong Feller Property of Distorted Elliptic Diffusion with Reflecting Boundary. Potential Anal. 40, 391–425 (2014)

    work page 2014

  11. [11]

    Gaussian measures

    Bogachev, V. Gaussian measures. In: Mathematical Surveys and Monographs , American Mathematical Society, 1998

    work page 1998

  12. [12]

    Hirsch, F

    Bouleau, N. Hirsch, F. Dirichlet forms and analysis on Wiener space. Walter de Gruyter & Co., Berlin, 1991

    work page 1991

  13. [13]

    Dello Schiavo, L., A Rademacher-type theorem on L2-Wasserstein spaces over closed Riemannian mani- folds. J. Funct. Anal. 278, 108397 (2020)

    work page 2020

  14. [14]

    Entropic and gradient flow formulations for non-linear diffusion

    Dirr, N., Stamatakis, M., Zimmer, J. Entropic and gradient flow formulations for non-linear diffusion. J. Math. Phys.. 57, 081505 (2016)

    work page 2016

  15. [15]

    M., Real Analysis and Probability

    Dudley, R. M., Real Analysis and Probability. In: Cambridge Studies in Advanced Mathematics . Cam- bridge University Press, 2nd edition, 2002

    work page 2002

  16. [16]

    Girsanov-type transformations of local Dirichlet forms: an analytic approach

    Eberle, A. Girsanov-type transformations of local Dirichlet forms: an analytic approach. Osaka J. Math. 33, 497-531 (1996)

    work page 1996

  17. [17]

    Erbar, M., The heat equation on manifolds as a gradient flow in the Wasserstein space. Ann. Inst. Henri Poincar´ e - Probab. Stat.46, 1-23 (2010)

    work page 2010

  18. [18]

    Glaudo, F

    Figalli, A. Glaudo, F. An Invitation to Optimal Transport, Wasserstein Distances and Gradient Flows. EMS textbooks in mathematics, EMS Press, 2021

    work page 2021

  19. [19]

    J., Absolute continuity of symmetric diffusions

    Fitzsimmons, P. J., Absolute continuity of symmetric diffusions. Ann. Probab. 25, 230-258 (1997)

    work page 1997

  20. [20]

    Jordan, R., Kinderlehrer, D., Otto, F., The variational formulation of the Fokker–Planck equation. SIAM J. Math. Anal. 29, 1-17 (1998)

    work page 1998

  21. [21]

    Takeda, M

    Fukushima, M., Oshima, Y. Takeda, M. Dirichlet forms and symmetric Markov processes. Walter de Gruyter & Co., Berlin, 2011

    work page 2011

  22. [22]

    Heat Flow on Time-Dependent Metric Measure Spaces and Super-Ricci Flows

    Kopfer E., Sturm, K.-T. Heat Flow on Time-Dependent Metric Measure Spaces and Super-Ricci Flows. Commun. Pure Appl. Math. 71, 2500-2608 (2018)

    work page 2018

  23. [23]

    Introduction to the theory of (non-symmetric) Dirichlet forms

    Ma, Z., R¨ ockner, M. Introduction to the theory of (non-symmetric) Dirichlet forms. Springer-Verlag, Berlin, 1992

    work page 1992

  24. [24]

    Gradient flows on Wasserstein spaces over compact Alexandrov spaces

    Ohta, S.-I. Gradient flows on Wasserstein spaces over compact Alexandrov spaces. Amer. J. Math. 131, 475-516 (2009)

    work page 2009

  25. [25]

    Heat flow on Finsler manifolds

    Ohta, S.-I., Sturm, K.-T. Heat flow on Finsler manifolds. Commun. Pure Appl. Math. 62, 1386-1433 (2009)

    work page 2009

  26. [26]

    Integral Varadhan formula for non-linear heat flow

    Ohta, S.-I., Suzuki, K. Integral Varadhan formula for non-linear heat flow. J. Funct. Anal. 289, 110983 (2025)

    work page 2025

  27. [27]

    Nonlinear Fokker–Planck–Kolmogorov equations as gradient flows on the space of probability measures

    Rehmeier, M., R¨ ockner, M. Nonlinear Fokker–Planck–Kolmogorov equations as gradient flows on the space of probability measures. Preprint, arXiv:2306.09530 (2023)

    work page arXiv 2023

  28. [28]

    Spectral gap for measure-valued diffusion processes

    Ren, P., Wang, F. Spectral gap for measure-valued diffusion processes. J. Math. Anal. Appl. 483, 123624 (2020)

    work page 2020

  29. [29]

    Ornstein–Uhlenbeck type processes on Wasserstein spaces

    Ren, P., Wang, F. Ornstein–Uhlenbeck type processes on Wasserstein spaces. Stoch. Proc. Appl. 172, 104339 (2024)

    work page 2024

  30. [30]

    Linearization of non-linear Fokker-Planck equations and applications

    Ren, P., Wang, F., R¨ ockner, M. Linearization of non-linear Fokker-Planck equations and applications. J. Differential Equations 332, 1-37 (2022). 30

    work page 2022

  31. [31]

    Ren, P., Wang, F., Wittmann, S., Diffusion Processes on p-Wasserstein Space over Banach Space, Preprint, arXiv:2402.15130 (2024)

    work page arXiv 2024

  32. [32]

    Dirichlet forms on infinite-dimensional ‘manifold-like’ state paces: A survey of recent results and some prospects for the future

    R¨ ockner, M. Dirichlet forms on infinite-dimensional ‘manifold-like’ state paces: A survey of recent results and some prospects for the future. In: Probability Towards 2000, 287-306, Springer, New York, 1998

    work page 2000

  33. [33]

    Uniqueness of generalized Schr¨ odinger operators and applications.J

    R¨ ockner, M., Zhang, T. Uniqueness of generalized Schr¨ odinger operators and applications.J. Funct. Anal. 105, 187-231 (1992)

    work page 1992

  34. [34]

    Gradient flows and diffusion semigroups in metric spaces under lower curvature bounds.C

    Savar´ e, G. Gradient flows and diffusion semigroups in metric spaces under lower curvature bounds.C. R. Acad. Sci. Paris S´ er. I Math.345, 151-154 (2007)

    work page 2007

  35. [35]

    Transformations of local Dirichlet forms by supermartingale multiplicative functionals

    Takeda, M. Transformations of local Dirichlet forms by supermartingale multiplicative functionals. In: Ma, Z., R¨ ockner, M., Yan, J., (ed.) Dirichlet Forms and Stochastic Processes , 363-374, De Gruyter, Berlin, 1995

    work page 1995

  36. [36]

    Optimal Transport: Old and New

    Villani, C. Optimal Transport: Old and New. Springer-Verlag, Berlin, 2009. 31

    work page 2009