arxiv: 2604.26023 · v1 · submitted 2026-04-28 · 🧮 math.OC · math.AP

Recognition: unknown

A PDE approach to Benamou--Brenier formula for the Schr\"odinger problem

Mattia Garatti , Luca Nenna , Simona Rota Nodari , Luca Tamanini

Authors on Pith no claims yet

Pith reviewed 2026-05-07 15:10 UTC · model grok-4.3

classification 🧮 math.OC math.AP

keywords Schrödinger problemBenamou-Brenier formulaoptimal transportsub-Gaussian measuresentropic interpolationPDE approachvelocity field

0 comments

The pith

The Benamou-Brenier formula for the Schrödinger problem holds for sub-Gaussian probability measures with unbounded support.

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

The paper extends the Benamou-Brenier dynamic formulation of the Schrödinger problem to sub-Gaussian marginals, which often have unbounded support. Existing proofs required compact supports to guarantee regularity of the Schrödinger potentials, but this work removes that restriction using fine Hessian estimates and entropic interpolation. The result establishes existence of a velocity field with polynomial growth that satisfies the necessary integrability conditions for the continuity equation. This justifies applying the dynamic formulation to Gaussian and mixture-of-Gaussians models common in applications.

Core claim

For sub-Gaussian probability measures, the Schrödinger problem admits a velocity field solving the continuity equation with polynomial growth, so the Benamou-Brenier formula holds without assuming compact support of the marginals.

What carries the argument

Fine Hessian estimates on the Schrödinger potentials obtained via the entropic interpolation, which guarantee the polynomial growth of the velocity field.

If this is right

The dynamic formulation can be applied directly to Gaussian and mixture-of-Gaussians models without compactification arguments.
Existence of an integrable velocity field follows from the polynomial growth control derived from the Hessian estimates.
The proof relies on PDE techniques applied to the Schrödinger potentials and the continuity equation.
Numerical schemes based on the dynamic formulation become justified for a wider class of unbounded-support problems.

Where Pith is reading between the lines

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

The extension may allow direct use of particle methods or gradient flows on unbounded domains without truncation.
Similar Hessian-control arguments could apply to other entropic transport problems with sub-Gaussian marginals.
It would be useful to check whether the polynomial growth can be sharpened to logarithmic growth for certain sub-Gaussian tails.

Load-bearing premise

The marginal measures are sub-Gaussian, which supplies the Hessian bounds needed for the velocity field to have polynomial growth.

What would settle it

A specific non-sub-Gaussian pair of marginals where the associated velocity field from the dynamic formulation fails to be square-integrable with respect to the entropic interpolation measure.

read the original abstract

We studied the Benamou--Brenier formulation of the Schr\"odinger problem, focusing on a gap between theoretical results and applications, that often involve measures with unbounded support. While the existing proof in the literature relies on the compactness of the marginals' supports to ensure the necessary regularity of the Schr\"odinger potentials, we extend the validity of the Benamou--Brenier formula to the larger class of sub-Gaussian probability measures. Exploiting fine estimates on the Hessian of the potentials and the entropic interpolation, we provide an almost self-contained proof that establishes the existence of a velocity field with the appropriate polynomial growth that ensures the right integrability. This result justifies the use of the dynamic formulation in more general settings, such as Gaussian and mixture-of-Gaussians models, important also for the applications.

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.

Desk Editor's Note private letter to a colleague

The paper extends the Benamou-Brenier formula to sub-Gaussian measures via Hessian estimates and entropic interpolation, but the bounds need checking for full generality.

read the letter

The paper's main point is that the Benamou-Brenier dynamic formulation of the Schrödinger problem holds for sub-Gaussian marginals, not just compactly supported ones. They achieve this by deriving fine Hessian estimates on the potentials and using entropic interpolation to control the growth of the velocity field, which ensures the necessary integrability. This is new compared to the compact support proofs in the literature, and it directly enables use with Gaussians and mixtures that come up in applications. The proof strategy looks reasonable and mostly self-contained as they describe. The focus on polynomial growth from the Hessian bounds is a sensible way to close the gap. A soft spot is whether sub-Gaussianity alone is enough to get those Hessian estimates without extra assumptions on the densities. The stress-test note points out that tail decay does not automatically give local C^2 control or uniform bounds, so the paper needs to show explicitly how the interpolation provides this for general sub-Gaussian cases. If it does, the result is fine; if it relies on unstated smoothness, the claim is narrower. This is useful for researchers in optimal transport who want to apply the dynamic Schrödinger formulation beyond compact supports. It is the kind of targeted technical improvement that deserves a serious referee to check the details of the estimates. I would recommend putting it through peer review.

Referee Report

2 major / 1 minor

Summary. The paper extends the Benamou-Brenier dynamic formulation of the Schrödinger problem from compactly supported marginals to the larger class of sub-Gaussian probability measures. It does so by deriving fine Hessian estimates on the Schrödinger potentials, combined with properties of entropic interpolation, to construct a velocity field with at most polynomial growth that satisfies the required integrability conditions in the dynamic formulation.

Significance. If the Hessian estimates and resulting integrability hold for general sub-Gaussian measures, the result would justify applying the dynamic formulation to important unbounded-support models such as Gaussians and Gaussian mixtures, closing a gap between existing theory and common applications.

major comments (2)

[Proof strategy (abstract and §3)] The central extension rests on obtaining Hessian bounds for the Schrödinger potentials that guarantee polynomial growth of the velocity field. The abstract and proof sketch indicate these bounds are derived from sub-Gaussian tail assumptions via entropic interpolation, but it is not clear whether sub-Gaussianity alone yields the necessary local C^2 regularity, uniform second-derivative control, and applicability of maximum principles without additional smoothness or positivity assumptions on the densities.
[Existence of velocity field (likely §4)] The integrability argument for the velocity field (ensuring it lies in the appropriate L^2 space for the dynamic formulation) appears to follow from the polynomial growth claim. However, the transition from sub-Gaussian marginals to the precise growth rate is load-bearing and requires explicit verification that no hidden compactness or support restriction is reintroduced.

minor comments (1)

[Introduction] Notation for the Schrödinger potentials and the entropic interpolation should be introduced with explicit references to the static and dynamic formulations to improve readability for readers outside the immediate subfield.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments. We address the two major points below, clarifying the proof structure and indicating the revisions we will make to improve readability and explicitness.

read point-by-point responses

Referee: [Proof strategy (abstract and §3)] The central extension rests on obtaining Hessian bounds for the Schrödinger potentials that guarantee polynomial growth of the velocity field. The abstract and proof sketch indicate these bounds are derived from sub-Gaussian tail assumptions via entropic interpolation, but it is not clear whether sub-Gaussianity alone yields the necessary local C^2 regularity, uniform second-derivative control, and applicability of maximum principles without additional smoothness or positivity assumptions on the densities.

Authors: The derivation in Section 3 begins from the sub-Gaussian tail decay of the marginals, which (under the standard setup of the Schrödinger problem) guarantees the existence of smooth positive densities. The entropic interpolation between these marginals inherits sufficient smoothness, allowing the Schrödinger potentials to be C^2 locally. Uniform second-derivative control is obtained by applying the maximum principle to the Hessian components on large balls whose radii tend to infinity; the sub-Gaussian tails provide the necessary decay to pass to the limit without compactness of the supports. We will revise the abstract to mention the density regularity inherited from sub-Gaussianity and insert a short preliminary paragraph in §3 that states the precise smoothness and positivity assumptions together with the maximum-principle argument. revision: yes
Referee: [Existence of velocity field (likely §4)] The integrability argument for the velocity field (ensuring it lies in the appropriate L^2 space for the dynamic formulation) appears to follow from the polynomial growth claim. However, the transition from sub-Gaussian marginals to the precise growth rate is load-bearing and requires explicit verification that no hidden compactness or support restriction is reintroduced.

Authors: Section 4 constructs the velocity field explicitly from the gradient of the potential along the entropic interpolation. The polynomial growth rate (of order |x|^k with k determined by the sub-Gaussian parameter) is combined with the uniform-in-time sub-Gaussian moment bounds on the interpolation measure to verify the L^2 integrability directly via integration by parts and moment estimates. No compactness or support restriction is used; the argument relies only on the tail decay. We will add a self-contained lemma in §4 that spells out this integrability calculation step by step. revision: yes

Circularity Check

0 steps flagged

No circularity: self-contained PDE proof via Hessian estimates and entropic interpolation

full rationale

The paper derives the extension of the Benamou-Brenier formula to sub-Gaussian measures through an almost self-contained argument that obtains fine Hessian bounds on the Schrödinger potentials from sub-Gaussian tail assumptions and then applies entropic interpolation to construct a velocity field of polynomial growth. No load-bearing step reduces by definition or construction to a fitted input, a self-citation chain, or a renamed known result; the central existence claim follows directly from the stated PDE estimates and interpolation properties without circular reduction to the paper's own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on abstract only; the proof rests on domain assumptions about sub-Gaussian tail decay enabling Hessian control, with no free parameters or invented entities mentioned.

axioms (1)

domain assumption Sub-Gaussian measures admit sufficiently strong Hessian estimates on the Schrödinger potentials to produce a velocity field of polynomial growth
Invoked to close the integrability argument for the dynamic formulation

pith-pipeline@v0.9.0 · 5450 in / 1162 out tokens · 39999 ms · 2026-05-07T15:10:07.308003+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

9 extracted references · 4 canonical work pages

[1]

Large deviations and gra- dient flows

[Ada+13] S. Adams, N. Dirr, M. Peletier, and J. Zimmer. “Large deviations and gra- dient flows”. In:Philosophical Transactions of the Royal Society A: Mathe- matical, Physical and Engineering Sciences371.2005 (2013), p. 20120341. [AGS05] L. Ambrosio, N. Gigli, and G. Savar´ e.Gradient flows: in metric spaces and in the space of probability measures. Springer,

2005
[2]

A computational fluid mechanics solution to the Monge-Kantorovich mass transfer problem

[BB00] J.-D. Benamou and Y. Brenier. “A computational fluid mechanics solution to the Monge-Kantorovich mass transfer problem”. In:Numerische Mathematik 84.3 (2000), pp. 375–393. [Ben+15] J.-D. Benamou, G. Carlier, M. Cuturi, L. Nenna, and G. Peyr´ e. “Iterative Bregman projections for regularized transportation problems”. In:SIAM Journal on Scientific Co...

2000
[3]

Entropy minimization, DAD problems, and doubly stochastic kernels

[BLN94] J. M. Borwein, A. S. Lewis, and R. D. Nussbaum. “Entropy minimization, DAD problems, and doubly stochastic kernels”. In:Journal of Functional Analysis123.2 (1994), pp. 264–307. [Chi+20] L. Chizat, P. Roussillon, F. L´ eger, F.-X. Vialard, and G. Peyr´ e. “Faster Wasserstein distance estimation with the Sinkhorn divergence”. In:Advances in neural i...

work page arXiv 1994
[4]

Weak semiconvexity estimates for Schr¨ odinger potentials and logarithmic Sobolev inequality for Schr¨ odinger bridges

1995, pp. 95–116. [Con24] G. Conforti. “Weak semiconvexity estimates for Schr¨ odinger potentials and logarithmic Sobolev inequality for Schr¨ odinger bridges”. In:Probability The- ory and Related Fields189.3 (2024), pp. 1045–1071. [CPT23] G. Carlier, P. Pegon, and L. Tamanini. “Convergence rate of general entropic optimal transport costs”. In:Calculus of...

1995
[5]

A formula for the time derivative of the entropic cost and applications

[CT21] G. Conforti and L. Tamanini. “A formula for the time derivative of the entropic cost and applications”. In:Journal of Functional Analysis280.11 (2021), p. 108964. [Cut13] M. Cuturi. “Sinkhorn distances: Lightspeed computation of optimal trans- port”. In:Advances in neural information processing systems26 (2013). [EMR15] M. Erbar, J. Maas, and D. R....

work page doi:10.1214/ecp.v20-4315 2021
[6]

Benamou–Brenier and duality formulas for the entropic cost onRCD ∗(K, N) spaces

2017, pp. 569–600. [GT20] N. Gigli and L. Tamanini. “Benamou–Brenier and duality formulas for the entropic cost onRCD ∗(K, N) spaces”. In:Probability Theory and Related Fields176.1 (2020), pp. 1–34. [GT21] N. Gigli and L. Tamanini. “Second order differentiation formula on RCD* (K, N) spaces”. In:Journal of the European Mathematical Society23.5 (2021), pp....

work page doi:10.3934/dcds.2014.34 2017
[7]

Entropy-regularized 2-Wasserstein distance between Gaussian measures

[MGM22] A. Mallasto, A. Gerolin, and H. Q. Minh. “Entropy-regularized 2-Wasserstein distance between Gaussian measures”. In:Information Geometry5.1 (2022), pp. 289–323. [Mik04] T. Mikami. “Monge’s problem with a quadratic cost by the zero-noise limit of h-path processes”. In:Probability theory and related fields129.2 (2004), pp. 245–260. [MT08] T. Mikami ...

work page arXiv 2022
[8]

A relationship between arbitrary positive matrices and doubly stochastic matrices

1932, pp. 269–310. [Sin64] R. Sinkhorn. “A relationship between arbitrary positive matrices and doubly stochastic matrices”. In:The annals of mathematical statistics35.2 (1964), pp. 876–879. [Sin67] R. Sinkhorn. “Diagonal equivalence to matrices with prescribed row and col- umn sums”. In:The American Mathematical Monthly74.4 (1967), pp. 402–

1932
[9]

Concerning nonnegative matrices and dou- bly stochastic matrices

[SK67] R. Sinkhorn and P. Knopp. “Concerning nonnegative matrices and dou- bly stochastic matrices”. In:Pacific Journal of Mathematics21.2 (1967), pp. 343–348. [Vil08] C. Villani.Optimal transport: old and new. Vol

1967