On the quadratic barycentric transport problem

Nathael Gozlan (MAP5 - UMR 8145); Paul-Marie Samson (LAMA); Thibaut Le Gouic (I2M)

REVIEW 2 major objections 5 minor 28 references

On the quadratic barycentric transport problem

T0 review · 2 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The quadratic barycentric transport cost between two probability measures is equal to the minimum expected squared drift of a semimartingale with those endpoint laws.

desk verdict Genuinely new Benamou-Brenier formula for quadratic barycentric transport, with a load-bearing typo in the achievability proof that is easy to fix. read the letter →

arxiv 2509.04935 v1 pith:7K7NDTVU submitted 2025-09-05 math.FA math.PR

Nathael Gozlan (MAP5 - UMR 8145) , Thibaut Le Gouic (I2M) , Paul-Marie Samson (LAMA) This is my paper · ORCID

classification math.FAmath.PR MSC 60A1049J5560G42

keywords weakoptimaltransportquadraticbarycentricBenamou-BrenierformulaconvexordermartingalepavinggeodesicsGaussianmeasures

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper studies the quadratic barycentric transport cost between two probability measures—the smallest possible value of the expected squared conditional mean difference E‖E(Y−X|X)‖² over couplings (X,Y). It proves a Benamou–Brenier-type theorem for this cost: T2 equals the minimum expected integral of squared drift over processes X_t that start with law μ, end with law ν, and evolve as a drift plus a martingale. It also shows the minimizers are geodesics—the cost between any two times along an optimal path scales as the squared time difference—and that the initial drift is readable from the current position, which makes the optimal dynamics Markov whenever the martingale part is Markov. This turns a static optimization over couplings into a stochastic control problem and connects barycentric transport to semimartingale and martingale transport.

What carries the argument

The paper's load-bearing construction is the convex paving {D_f(z)} attached to a dual optimizer f of the barycentric problem. Each D_f(z) is the set of points y such that f is affine on the segment from z to y (and just past y); equivalently, it is the projection of the relative face of the epigraph of f. Lemma 4 and Proposition 5 show the paving partitions the domain into convex cells that are invariant under every martingale connecting μ̄ and ν: if M_0∼μ̄ and M_1∼ν, then M_t∈D_f(M_0) almost surely for all t. This invariance makes the drift u=X_0−∇ϕ(X_0) recoverable from X_t=(1−t)u+M_t: on each cell u lies in ∂f(m) for every m in the cell, and the map (m,u)↦(1−t)u+m is injective on the sub

What would settle it

Test the theorem on a pair μ,ν in P₂(R^d) with finite variance for which ν is purely atomic and d≥2, e.g. μ uniform on the vertices of a cube and ν uniform on a larger cube's vertices. Solve the finite-dimensional linear program for T2 from the decomposition (11), and independently solve a finely time-discretized version of the semimartingale problem in (20) by minimizing over v and martingale increments. If the two minima differ, Theorem 9(i) fails.

Watch

Extended reading notes

Core claim

The central claim is Theorem 9: for all μ,ν∈P₂(R^d), T2(ν|μ)=inf E∫₀¹‖v_t‖²dt, where the infimum ranges over progressively measurable drifts v and martingales M for which X_t=X_0+∫₀^t v_s ds+M_t−M_0, X_0∼μ, and X_1∼ν. Equality is attained by the explicit family X_t=(1−t)X_0+t∇ϕ(X_0)+M_t−∇ϕ(X_0), where ∇ϕ is the 1-Lipschitz gradient of a convex function pushing μ forward to μ̄, the backward projection of μ onto the set of measures dominated by ν in convex order, and M is any martingale with M_0=∇ϕ(X_0) and M_1∼ν. Along these processes the marginal laws satisfy T2(μ_t|μ_s)=(t−s)²T2(ν|μ), and if M is Markovian, X is Markovian.

Load-bearing premise

Everything rests on the imported result that, on a sufficiently rich filtered probability space, a continuous martingale with prescribed marginals M_0=∇ϕ(X_0) and M_1∼ν can always be constructed from a Brownian motion; if that theorem needs extra moment or support conditions beyond finite variance, the equality can fail for some pairs.

Editorial extensions

If this is right

Barycentric optimal transport becomes a semimartingale transport problem: costs, couplings, and geodesics can be described by drift processes and martingale components rather than by two-point couplings.
The optimal interpolations are constant-speed geodesics for √T2: T2(μ_t|μ_s)=(t−s)²T2(ν|μ) for every pair of times along a process of the form (22).
The initial drift X_0−∇ϕ(X_0) is σ(X_t)-measurable for each t<1, so the optimal process is Markov whenever the martingale part is Markov; this is a stochastic analogue of non-crossing in McCann interpolation.
Every optimal T2 plan can be factored as the deterministic map ∇ϕ to the backward projection μ̄ followed by an arbitrary martingale coupling from μ̄ to ν; in the commuting Gaussian case μ̄ is Gaussian with covariance min(Σ_μ,Σ_ν) in a common eigenbasis and T2 has the closed form |m_ν−m_μ|²+Σ_i[σ_i(μ)−σ_i(ν)]²₊.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper, the dynamic formula suggests a numerical route to T2 that does not require dual potentials: discretize time and optimize over v and M with the martingale constraint imposed on conditional increments; the equality with T2 provides a built-in check on any discretization.
The martingale-invariant convex paving may transfer to entropic or regularized versions of barycentric transport, where the same cells could localize the effect of regularization on martingale constraints.
The correspondence between martingales from μ to a forward projection and martingales from μ̄ to ν hints at a full dictionary between backward and forward Wasserstein projections: optimizers of one problem may be transported to optimizers of the other by ∇ϕ and ∇ϕ*, beyond the Gaussian examples worked out here.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper does two real things: it proves a Benamou-Brenier type formula for T2 and shows optimal processes are geodesics with a Markov property. That is new and worth having. The main proof is modular and mostly rigorous. But the achievability half of Theorem 9 has a genuine gap: the stretched-Brownian kernel is built from the wrong first marginal.

The lower bound in (20) is a clean Jensen argument. The convex paving in Section 3 is a nice constructive structural result, and the Markov property (iv) follows from it. The paper also honestly flags that the Gaussian results duplicate Alfonsi-Jourdain, which is the right call.

The gap: in the proof of Theorem 9, they introduce a weak optimal transport problem with first marginal μ and constraint ∫y dp_x(y)=x, claiming "Since μ ≤c ν this set of kernels is non-empty." That is false in general; μ ≤c ν would be a martingale coupling, which is not assumed. What is true is μ̄ ≤c ν, where μ̄ = ∇ϕ# μ. The kernel should have first marginal μ̄, and then M1 ~ ν follows. As written, p* has first marginal μ, so M1 would have law μ̄p*, not necessarily ν, and p*_z is only defined for μ-a.e. z. This is repairable by replacing μ with μ̄ throughout the construction; the rest of the proof stands. It's a typo in a load-bearing place, but the fix is clear.

Minor issues: Proposition 14 omits a proof that is indeed identical; Proposition 26(ii) leaves details; the Gaussian section overlaps with [2] but they say so. No circularity—the main formula is proved independently.

Who it's for: people working in weak optimal transport, semimartingale transport, and martingale transport. A good referee can handle this. I'd send it to peer review; the gap is real but likely fixable. If the authors replace μ by μ̄ in that paragraph, the theorem goes through.

Referee Report

2 major / 5 minor

Summary. The paper studies the quadratic barycentric transport cost T_2(ν|μ) between probability measures on R^d. Its main result, Theorem 9, asserts a Benamou–Brenier type representation: T_2(ν|μ) equals the infimum of E∫_0^1 ||v_t||^2 dt over semimartingales dX_t = v_t dt + dM_t with X_0∼μ, X_1∼ν, and M an F-martingale. It further claims that equality is attained by processes X_t = X_0 + t(∇φ(X_0)-X_0) + M_t - ∇φ(X_0), where ∇φ#μ is the backward projection μ̄ of μ onto {η≤_c ν}, M_0=∇φ(X_0), M_1∼ν; these optimal processes satisfy the geodesic identity T_2^{1/2}(μ_t|μ_s)=(t-s)T_2^{1/2}(ν|μ) and are Markov when M is Markovian. The proof is built on a dual-potential construction: Section 3 associates to a dual optimizer f̄ a convex paving {D_{f̄}(z)} that is invariant for all martingales from μ̄ to ν (Lemma 4, Proposition 5, Corollary 6). Section 5 connects martingales from μ to a forward projection ν̃ with martingales from μ̄ to ν via ∇φ. Section 6 gives explicit Gaussian formulas: the backward projection is Gaussian with covariance solving a matrix optimization, and a closed form in the commuting case.

Significance. If the main theorem is correct, the paper gives a genuine semimartingale-transport formulation of the barycentric cost, a geodesic structure analogous to McCann interpolation, and a Markov property for optimal processes. This is a valuable bridge between weak optimal transport, martingale transport, and the de March–Touzi convex paving theory. The paper is written in a modular way, with detailed proofs of Lemma 4 and Proposition 5, and it contains explicit, falsifiable statements in the Gaussian case (Propositions 15 and 18). The authors also honestly acknowledge the concurrent Gaussian results of Alfonsi–Jourdain. However, the proof of the achievability half of Theorem 9 contains a genuine error in the construction of the martingale M, as detailed below; the error appears repairable by replacing μ with μ̄, but it is load-bearing for the central claim in its current form.

major comments (2)

[Section 4, proof of Theorem 9, paragraph ‘In order to construct the martingale part…’] The weak optimal transport problem is formulated with first marginal μ: inf_p ∫ W_2^2(p_x,γ)dμ(x) subject to ∫ y dp_x(y)=x for μ-a.e. x and μp=ν. The text then says this set of kernels is non-empty because ‘μ ≤_c ν’. For arbitrary μ,ν∈P_2(R^d), μ≤_cν is false and the set is generally empty. What is needed is a martingale starting from μ̄=∇φ#μ, not from μ. The subsequent construction M_t=g_t(∇φ(X_0),B_t) uses the kernel p* at the point ∇φ(X_0); if p* is defined only μ-a.e., this is not defined μ̄-a.e., and even if extended, M_1 would have law μ̄p*, not necessarily ν. The correct fix is to replace μ by μ̄ throughout the weak transport formulation: inf_p ∫ W_2^2(p_z,γ)dμ̄(z) with ∫y dp_z(y)=z for μ̄-a.e. z and μ̄p=ν. Since μ̄≤_cν, the set is non-empty, and the same argument yields M_0=∇φ(X_0), M_1∼ν and a martingale M. This correction is necessary for the existence of the process (22) and h
[Section 5, Proposition 14 and Section 6] The proof of Proposition 14 is omitted with the statement ‘proof is identical and thus omitted’. Since Proposition 14 is used in the Gaussian section to identify C_{f̄}(z) as an affine subspace, the omission is not merely cosmetic. If the proof is truly identical to that of Propositions 11–12, a short indication of the dictionary between f̄ and ḡ should be supplied. In addition, Proposition 26(ii) says details are left to the reader; given that it is an adaptation of an external result, this is acceptable but should be made precise.

minor comments (5)

[Section 4, after (22)] In the bullet list defining the optimal process, the notation ‘µ := argmin_{η≤_cν} W_2^2(µ,η)’ reuses μ for the backward projection. It should read μ̄ = argmin…, and later W_2^2(μ,μ̄) rather than W_2^2(μ,μ).
[Section 2, Lemma 1] In the second bullet, ‘if g is an optimizer of (8), then f=P_2g is an optimizer of (8)’ should refer to problem (7).
[Remark 10] In the displayed chain, the term ‘(t−s)^2 T_2^{1/2}(ν|μ)’ should be ‘(t−s)^2 T_2(ν|μ)’, since the expectation involving E||∇φ(X_0)−X_0||^2 equals T_2(ν|μ), not its square root.
[Section 6, Proposition 18] The phrase ‘let P be an orthonormal matrix P’ is redundant; also the expression D_μ := PΣ_μP^T is used before P is fully specified. Minor rewording would help.
[Throughout] The paper contains a number of small typographical slips (e.g. ‘Item 2.’ for ‘Item (ii)’, the use of ar µ versus µ in the backward projection, and the notation µ_s for the backward projection of µ_s in Remark 10). These do not affect the mathematics but should be cleaned up.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central semimartingale formula is proved independently, though the achievability proof contains a marginal typo that is a correctness gap, not a circular reduction.

full rationale

The main equality (20) is not circular. The lower bound T2(ν|μ) ≤ inf E∫||v_t||²dt is a direct Jensen/Cauchy-Schwarz argument valid for all admissible processes and does not presuppose the target formula. The achievability half constructs X_t = X0 + t(∇φ(X0)-X0) + M_t - ∇φ(X0), with ∇φ#μ = μ̄ the backward projection; the existence of μ̄, ∇φ, and the identity T2(ν|μ)=W2²(μ,μ̄) are imported from the prior theorem [16, Thm 1.2] (which has one overlapping author) and from [21]. Those theorems do not assume the semimartingale formula (20); they are parameter-free statements with independent proofs, so this is independent support rather than a self-referential reduction. The martingale M is obtained from [6, Thm 2.2], an external construction by other authors, not from the paper's own inputs. There is, however, a genuine gap in the submitted proof: the weak transport problem is written with first marginal μ and constraint μp=ν, but the condition actually needed is μ̄≤_cν, and the sentence 'Since μ ≤_cν this set of kernels is non-empty' is false in general. As written, p*_x is only defined μ-a.e., while M_t = g_t(∇φ(X0),B_t) requires p*_z for z~μ̄; consequently Law(M1) is not shown to equal ν. This is a correctness/typo issue in the achievability argument, repairable by replacing μ with μ̄ throughout the weak transport construction, but it is not a circularity. The Gaussian results in Section 6 are independently derived and explicitly acknowledge overlap with [2]; the self-citations to [16] and [17] are background results, not answers smuggled into the derivation. No step fits the patterns of self-definition, fitted-input-called-prediction, or renaming a known result; the central claim has independent content.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

No free parameters or invented entities. The paper rests on established theorems from optimal transport, convex analysis, and stochastic processes, listed as axioms. The most consequential imported result is the stretched Brownian motion existence from [6].

assumptions (6)

standard math Strassen's theorem: α ≤_c β if and only if there is a martingale kernel q with αq = β (Strassen [26]).
Ensures the set M(\bar μ,ν) of martingale kernels is non-empty; used in Section 2 after (11).
domain assumption Dual existence and projection representation: T2(ν|μ) = inf_{η≤_cν} W2²(μ,η), with a unique backward projection \bar μ and an optimizer \bar f of (7) ([16, Theorem 6.1] and [16, Theorem 1.2]).
This is the foundation for defining φ, \bar μ, and the paving; Section 2.
domain assumption Existence and uniqueness of the stretched Brownian motion minimizer ([6, Theorem 2.2]) and its Markov property ([6, Corollary 2.5]).
Used in the proof of Theorem 9 to construct the martingale M achieving equality in (20) and to obtain the Markov property.
standard math Relative-face facts for closed convex sets: the family {rf_a(A)} partitions A, and parts of [13, Proposition 3.1] (a convexity property of relative faces) hold.
Used in Lemma 4 and in the proof of Proposition 5(i) to show q_z is supported on the relative face D_f(z).
standard math Brenier's theorem on existence and uniqueness of optimal transport maps for W2: optimal maps are gradients of convex functions.
Used to derive the structure of ∇φ and the Gaussian transport maps in Section 6.
standard math Cuesta-Albertos, Matrán-Bea, Tuero-Diaz lower bound W2(η,ν) ≥ W2(N(m_η,Σ_η), N(m_ν,Σ_ν)) (Lemma 17, [12]).
Used to prove the Gaussian formula Proposition 15.

how reviews work

0 comments

Cite this review

Pith. "Pith review of On the quadratic barycentric transport problem." pith.science (2026). https://pith.science/paper/7K7NDTVU

@misc{pith2026250904935,
  author       = {Pith},
  title        = {Pith review of: On the quadratic barycentric transport problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7K7NDTVU}},
  note         = {Machine review of arXiv:2509.04935}
}

read the original abstract

We investigate the structure of optimal transport plans, dual optimizers, and geodesic paths for the quadratic barycentric transport problem.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 26 canonical work pages

[2]

W asserstein p rojections in the convex order: regularity and characteriz ation in the quadratic Gaussian case

Aur´ elien Alfonsi and Benjamin Jourdain. W asserstein p rojections in the convex order: regularity and characteriz ation in the quadratic Gaussian case. Preprint, arXiv:2506.23981 [ math.PR] (2025), 2025

arXiv 2025
[1]

Sampling of probability measures in the convex order by W asserstein projection

Aur´ elien Alfonsi, Jacopo Corbetta, and Benjamin Jourd ain. Sampling of probability measures in the convex order by W asserstein projection. Ann. Inst. Henri Poincar´ e, Probab. Stat., 56(3):1706–1729, 2020

work page 2020
[3]

Alibert, G

J.-J. Alibert, G. Bouchitt´ e, and T. Champion. A new clas s of costs for optimal transport planning. Eur. J. Appl. Math. , 30(6):1229–1263, 2019

work page 2019
[4]

Backhoﬀ-Veraguas, M

J. Backhoﬀ-Veraguas, M. Beiglb¨ ock, and G. Pammer. Exis tence, duality, and cyclical monotonicity for weak transpo rt costs. Calc. Var. Partial Diﬀer. Equ. , 58(6):28, 2019. Id/No 203

work page 2019
[5]

Backhoﬀ-Veraguas and G

J. Backhoﬀ-Veraguas and G. Pammer. Applications of weak transport theory. Bernoulli, 28(1):370–394, 2022

work page 2022
[6]

Martingale benamou–brenier

Julio Backhoﬀ-Veraguas, Mathias Beiglb¨ ock, Martin Hu esmann, and Sigrid K¨ allblad. Martingale benamou–brenier . The Annals of Probability , 48(5):2258–2289, 2020

work page 2020
[7]

The Fundamental Theorem of W eak Optimal Transport

Mathias Beiglb¨ ock, Gudmund Pammer, Lorenz Riess, and S tefan Schrott. The Fundamental Theorem of W eak Optimal Transport. Preprint, arXiv:2501.16316 [math.PR] (2025), 2025

arXiv 2025
[8]

A computational ﬂu id mechanics solution to the Monge-Kantorovich mass transf er problem

Jean-David Benamou and Yann Brenier. A computational ﬂu id mechanics solution to the Monge-Kantorovich mass transf er problem. Numer. Math. , 84(3):375–393, 2000

work page 2000

Show all 28 references

[9]

D´ ecomposition polaire et r´ earrangement monotone des champs de vecteurs

Yann Brenier. D´ ecomposition polaire et r´ earrangement monotone des champs de vecteurs. (Polar decomposition and in- creasing rearrangement of vector ﬁelds). C. R. Acad. Sci., Paris, S´ er. I , 305:805–808, 1987

1987
[10]

Polar factorization and monotone rearra ngement of vector-valued functions

Yann Brenier. Polar factorization and monotone rearra ngement of vector-valued functions. Commun. Pure Appl. Math. , 44(4):375–417, 1991

1991
[11]

A no vel notion of barycenter for probability distributions bas ed on optimal weak mass transport

Elsa Cazelles, Felipe Tobar, and Joaquin Fontbona. A no vel notion of barycenter for probability distributions bas ed on optimal weak mass transport. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. W ortman Vaughan, editors, Advances in Neural Information Process...
[12]

On lower bounds for the L2-W asserstein metric in a Hilbert space

Juan Antonio Cuesta-Albertos, Carlos Matr´ an-Bea, and Araceli Tuero-Diaz. On lower bounds for the L2-W asserstein metric in a Hilbert space. Journal of Theoretical Probability , 9(2):263–283, 1996

1996
[13]

Irreducible convex pa ving for decomposition of multidimensional martingale tra nsport plans

Hadrien De March and Nizar Touzi. Irreducible convex pa ving for decomposition of multidimensional martingale tra nsport plans. Ann. Probab., 47(3):1726–1774, 2019

2019
[14]

A pro of of the Caﬀarelli contraction theorem via entropic regula r- ization

Max Fathi, Nathael Gozlan, and Maxime Prod’homme. A pro of of the Caﬀarelli contraction theorem via entropic regula r- ization. Calc. Var. Partial Diﬀer. Equ. , 59(3):18, 2020. Id/No 96

2020
[15]

Gallou¨ et, Andrea Natale, and Gabriele Todes chi

Thomas O. Gallou¨ et, Andrea Natale, and Gabriele Todes chi. Metric extrapolation in the W asserstein space. Calc. Var. Partial Diﬀer. Equ. , 64(5):35, 2025. Id/No 147

2025
[16]

On a mixture of bren ier and strassen theorems

Nathael Gozlan and Nicolas Juillet. On a mixture of bren ier and strassen theorems. Proceedings of the London Mathematical Society, 120(3):434–463, 2020

2020
[17]

Hamilton Jacobi equations on metric spaces and transport en tropy inequalities

Nathael Gozlan, Cyril Roberto, and Paul-Marie Samson. Hamilton Jacobi equations on metric spaces and transport en tropy inequalities. Rev. Mat. Iberoam. , 30(1):133–163, 2014

2014
[18]

Characterization of a class of weak transport-entropy inequalities on the line

Nathael Gozlan, Cyril Roberto, Paul-Marie Samson, Yan Shu, and Prasad Tetali. Characterization of a class of weak transport-entropy inequalities on the line. Ann. Inst. Henri Poincar´ e, Probab. Stat., 54(3):1667–1693, 2018

2018
[19]

Kantorovich duality for general transport c osts and applications

Nathael Gozlan, Cyril Roberto, Paul-Marie Samson, and Prasad Tetali. Kantorovich duality for general transport c osts and applications. Journal of Functional Analysis , 273(11):3327–3405, 2017

2017
[20]

Inequalities for the gaussian measure a nd an application to wiener space

Gilles Harg´ e. Inequalities for the gaussian measure a nd an application to wiener space. Comptes Rendus de l’Acad´ emie des Sciences-Series I-Mathematics , 333(8):791–794, 2001. ON THE QUADRATIC BARYCENTRIC TRANSPORT PROBLEM 19

2001
[21]

Backward and forward W asserstein projections in stochastic order

Young-Heon Kim and Yuanlong Ruan. Backward and forward W asserstein projections in stochastic order. J. Funct. Anal. , 286(2):63, 2024. Id/No 110201

2024
[22]

Robert J. McCann. A convexity principle for interactin g gases. Adv. Math. , 128(1):153–179, 1997

1997
[23]

Optimal control for absolutely continu ous stochastic processes and the mass transportation probl em

Toshio Mikami. Optimal control for absolutely continu ous stochastic processes and the mass transportation probl em. Elec- tron. Commun. Probab. , 7:199–213, 2002. Id/No 20

2002
[24]

Duality theorem for the stochastic optimal control problem

Toshio Mikami and Mich` ele Thieullen. Duality theorem for the stochastic optimal control problem. Stochastic Processes Appl., 116(12):1815–1835, 2006

2006
[25]

Convex analysis:(pms-28)

Ralph Tyrell Rockafellar. Convex analysis:(pms-28). 2015

2015
[26]

Strassen

V. Strassen. The existence of probability measures wit h given marginals. Ann. Math. Stat. , 36:423–439, 1965

1965
[27]

W asserstein geometry of Gaussian measu res

Asuka Takatsu. W asserstein geometry of Gaussian measu res. Osaka Journal of Mathematics , 48(4):1005 – 1026, 2011

2011
[28]

Optimal transportation und er controlled stochastic dynamics

Xiaolu Tan and Nizar Touzi. Optimal transportation und er controlled stochastic dynamics. Ann. Probab., 41(5):3201–3240, 2013. N. G. : Universit ´e Paris Cit ´e, CNRS, MAP5, F-75006 Paris Email address : nathael.gozlan@u-paris.fr T. Le G. : Centrale Marseille, I2M, UMR 7373, C...

2013

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [2]

W asserstein p rojections in the convex order: regularity and characteriz ation in the quadratic Gaussian case

Aur´ elien Alfonsi and Benjamin Jourdain. W asserstein p rojections in the convex order: regularity and characteriz ation in the quadratic Gaussian case. Preprint, arXiv:2506.23981 [ math.PR] (2025), 2025

arXiv 2025

[2] [1]

Sampling of probability measures in the convex order by W asserstein projection

Aur´ elien Alfonsi, Jacopo Corbetta, and Benjamin Jourd ain. Sampling of probability measures in the convex order by W asserstein projection. Ann. Inst. Henri Poincar´ e, Probab. Stat., 56(3):1706–1729, 2020

work page 2020

[3] [3]

Alibert, G

J.-J. Alibert, G. Bouchitt´ e, and T. Champion. A new clas s of costs for optimal transport planning. Eur. J. Appl. Math. , 30(6):1229–1263, 2019

work page 2019

[4] [4]

Backhoﬀ-Veraguas, M

J. Backhoﬀ-Veraguas, M. Beiglb¨ ock, and G. Pammer. Exis tence, duality, and cyclical monotonicity for weak transpo rt costs. Calc. Var. Partial Diﬀer. Equ. , 58(6):28, 2019. Id/No 203

work page 2019

[5] [5]

Backhoﬀ-Veraguas and G

J. Backhoﬀ-Veraguas and G. Pammer. Applications of weak transport theory. Bernoulli, 28(1):370–394, 2022

work page 2022

[6] [6]

Martingale benamou–brenier

Julio Backhoﬀ-Veraguas, Mathias Beiglb¨ ock, Martin Hu esmann, and Sigrid K¨ allblad. Martingale benamou–brenier . The Annals of Probability , 48(5):2258–2289, 2020

work page 2020

[7] [7]

The Fundamental Theorem of W eak Optimal Transport

Mathias Beiglb¨ ock, Gudmund Pammer, Lorenz Riess, and S tefan Schrott. The Fundamental Theorem of W eak Optimal Transport. Preprint, arXiv:2501.16316 [math.PR] (2025), 2025

arXiv 2025

[8] [8]

A computational ﬂu id mechanics solution to the Monge-Kantorovich mass transf er problem

Jean-David Benamou and Yann Brenier. A computational ﬂu id mechanics solution to the Monge-Kantorovich mass transf er problem. Numer. Math. , 84(3):375–393, 2000

work page 2000