Pith. sign in

REVIEW 2 major objections 6 minor 25 references

Global $W^{2,p}$ Regularity in Optimal Transport

T0 review · 2 major / 6 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read Quadratic optimal transport potentials between merely continuous densities on convex domains are globally W^{2,p} for every p>1, with no boundary smoothness required.

desk verdict Solid extension of Collins–Tong: global W^{2,p} on plain convex domains with only continuous densities, via Cesàro-small scale errors and Borell reduction. read the letter →

arxiv 2607.24666 v1 pith:YJ3FORXC submitted 2026-07-27 math.AP

Shibing Chen , Yuanyuan Li , Xianduo Wang This is my paper · ORCID
classification math.AP MSC 35J9649Q2235B65
keywords optimaltransportMonge-AmpèreequationglobalW^{2p}estimatesconvexdomainscontinuousdensitiescenteredsectionsalmostmonotonicity
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 proves that the convex potentials that push one continuous density forward to another under quadratic cost have second derivatives in every L^p space, globally on the domains. The domains need only be bounded and convex; their boundaries can be rough. Earlier global W^{2,p} results demanded either Hölder densities or C^{1,1} boundaries (or restricted to the plane with constant densities). The advance is that plain continuity of the densities is enough. The argument tracks how mass concentrates in dyadic sections around each point. Continuity makes the average error over many scales small even when the total error may diverge, so most scales behave like the homogeneous case. At those good scales a compactness-plus-rigidity step forces sections to shrink like half-size balls; bad scales are controlled by a crude inclusion. The resulting sub-power control on section shape, combined with interior estimates and a covering argument, yields the global integrability of the Hessian.

What carries the argument

An almost-monotonicity formula for the normalized mass F_{x_0}(r) of centered sections: the ratio of F at two scales is controlled by the exponential of an integral of the moduli of continuity of the densities. The Cesàro mean of these additive errors over the first N dyadic scales is o(N), so only o(N) scales are bad; good scales force geometric decay of sections by blow-up rigidity (including Borell dimensional reduction when the limit measure is lower-dimensional).

What would settle it

Exhibit continuous positive densities on a pair of bounded convex domains for which the second derivatives of the optimal-transport potential fail to be in L^p for some p>1 near the boundary, or show that the Cesàro mean of the almost-monotonicity errors stays bounded away from zero along some sequence of base points.

Watch

Extended reading notes

Core claim

If two bounded open convex domains carry continuous densities bounded between positive constants and of equal mass, the dual convex potentials of quadratic optimal transport lie in W^{2,p} of their domains for every finite p. No boundary regularity is assumed.

Load-bearing premise

That continuous densities make the average error across successive length scales tend to zero, even though the total accumulated error need not stay finite.

Editorial extensions

If this is right

  • Global W^{2,p} holds for quadratic transport on any pair of bounded convex domains once the densities are merely continuous and non-degenerate.
  • Boundary smoothness assumptions used in earlier global Sobolev theory can be dropped entirely for the quadratic cost.
  • The same section-decay iteration supplies the geometric input needed for covering arguments that upgrade interior Hessian estimates to the whole domain.
  • When a blow-up limit is supported on a proper subspace, Borell’s theorem plus Brunn–Minkowski still force two-homogeneity, so the rigidity step survives dimensional collapse.

Reading between the lines

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

  • The average-error device may apply to other fully nonlinear equations whose monotonicity formulas carry continuous rather than Hölder coefficients.
  • Once global W^{2,p} is known, standard embedding and difference-quotient arguments become available for further boundary regularity questions under slightly stronger density assumptions.
  • The method suggests that any modulus of continuity whose dyadic averages vanish is admissible; quantitative moduli would yield quantitative section exponents.
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.

Referee Report

2 major / 6 minor

Summary. The paper proves that for quadratic optimal transport between arbitrary bounded convex domains Ω, Ω* ⊂ R^n with densities f, g continuous on the closures and satisfying 0 < λ ≤ f,g ≤ Λ and equal total mass, the dual convex potentials satisfy u ∈ W^{2,p}(Ω) and v ∈ W^{2,p}(Ω*) for every p > 1 (Theorem 1.1). This removes the Hölder assumption of Collins–Tong [12] and the C^{1,1} boundary assumption of Chen–Liu–Wang [9], and extends Savin–Yu [23] (planar, constant densities) to all dimensions and continuous densities. The proof has five steps: (i) an almost-monotonicity formula for F_{x_0}(r) = r^{-n}∫_{D_r} f with error C∫ϑ(s)/s ds built from the moduli of continuity (Theorem 3.1), proved by approximation from smooth uniformly convex domains; (ii) John-normalized blow-ups with simultaneous treatment of dual potentials (Sections 4–5); (iii) identification of degenerate limits via Borell's s-concave measure classification and a projected m-dimensional transport problem (Lemmas 5.2–5.4); (iv) an equality-case rigidity theorem for q-concave weights giving conic domains and 2-homogeneous potentials (Theorem 6.3), lifted to the original limit (Theorem 6.4); (v) a good/bad dyadic scale iteration in which the errors are not summable but have vanishing Cesàro mean, so only o(N) scales are bad, yielding sub-power section control (Propositions 7.4–7.5), from which W^{2,p} follows via Caffarelli's interior estimate and Savin's covering argument.

Significance. If correct, this is the optimal global Sobolev regularity result for optimal transport between convex domains: continuity of the densities is the natural sharp hypothesis (interior W^{2,p} for all p is already best possible under mere continuity), no boundary regularity is imposed, and the statement recovers C^{1,α} for all α < 1 by Sobolev embedding. The main technical contribution — replacing summable/Dini-type errors by errors whose dyadic Cesàro mean vanishes, so that bad scales have density zero — is cleanly isolated (Propositions 7.2–7.5) and is likely to be useful in other free-boundary and degenerate-elliptic iterations. The dimension-reduction step through Borell's theorem, with the projected target density's q-th root concave by Brunn–Minkowski and an exact mass identity preserved (5.22), is a well-executed treatment of the lower-dimensional blow-up alternative, and the equality-case rigidity (Theorem 6.3) is proved without assuming conic structure or homogeneity a priori. Appendix A supplies the needed localization of the Figalli–Jhaveri regularity theorem to locally finite measures. I verified the key non-summability point (Proposition 7.4: the log term is uniformly bou

major comments (2)
  1. [§7, Proof of Theorem 1.1 (final sentence)] The entire conclusion of the paper is discharged in one sentence: 'The global W^{2,p} estimate follows from Proposition 7.5, Caffarelli's interior W^{2,p} estimate [3], and a covering argument of Savin [22].' This step is load-bearing and is not quite a direct citation: Savin's covering argument in [22] is formulated for sections of the solution contained in its domain, whereas here the relevant centered sections S^c_h[u](x_0) may contain points outside Ω (as the paper itself notes in §2), and the input (7.12) is a two-sided Euclidean shape estimate for centered sections with an h^ε loss. The adaptation of Savin's Vitali-type covering to boundary-cut centered sections exists in the literature (e.g., [9, §7] for C^{1,1} domains and [12] for convex domains), but the manuscript should say so explicitly and indicate precisely which lemma is being invoked and why its hypotheses (engulfing, de
  2. [Eq. (3.3) (and Step (i) of the introduction), Proposition 7.4] As printed, the integral limits in the central estimate are reversed. Theorem 3.1 states F_{x_0}(r_1) ≤ F_{x_0}(r_2) exp(C∫_{r_1}^{r_2} ϑ(s)/s ds) for 0 < r_2 < r_1; with the standard convention this exponential is < 1, which contradicts (3.4) (F' ≤ CϑF/r integrates to F(r_1) ≤ F(r_2) exp(C∫_{r_2}^{r_1} ϑ/s ds)). The same reversal appears in Proposition 7.4, where e_{x_0}(a,b) = C∫_b^a ϑ(s)/s ds for a < b is negative as printed, contradicting the immediately following assertion that e_{x_0} is nonnegative, and again in the display ∑d_k = log(F(r_N)/F(r_0)) + 2C∫_{r_0}^{r_N} in its proof. The intended direction is unambiguous from (3.4) and from the role of a_{x_0}, e_{x_0} in Propositions 7.2–7.3, so this is evidently typographical rather than mathematical; but since these displays define the monotone quantity and the error on which the whole iteration rests, the conventions should be co
minor comments (6)
  1. [§6, Theorem 6.4 proof] The exclusion of m = 0 is carried out twice: Lemma 5.3 already proves m ≥ 1, yet the proof of Theorem 6.4 repeats the argument ('The case m = 0 can be ruled easily...'). One of the two can be replaced by a cross-reference.
  2. [Appendix A, Proposition A.1] The result is labeled 'Sketch of proof' but is used in a load-bearing way in Sections 5–6 (strict convexity, C^1 regularity, and the homeomorphism property for locally finite transports). The sketch is detailed and the modifications relative to [14] are clearly indicated, but the authors should either upgrade it to a complete proof or state explicitly that no step beyond those indicated requires the finite-mass hypothesis of [14].
  3. [§1, Theorem 1.1] It would help the reader to state the dependence of the W^{2,p} constants (on n, p, λ, Λ, the moduli of continuity, and the inner/outer radii of the domains), and to add a short remark that W^{2,p} for all p is the sharp scale under mere continuity (no C^{1,1} or uniform modulus on D^2u is possible), together with the C^{1,α} (all α < 1) corollary by Sobolev embedding.
  4. [§2, below (2.3)] Typo: 'The centered sub-level set of heighthatx0' should read 'of height h at x_0'. There are also several spacing artifacts around displayed equations that suggest a compilation issue worth checking in the source file.
  5. [§6, Lemma 6.2(iii)] 'After possibly decreasing the local exponent' — please clarify that α' may depend on the compact subset of C (resp. C*), consistent with the compact-set-dependent α in the preamble of the lemma.
  6. [References] Several key inputs (Propositions 2.3, 2.5, 2.6; Theorem 3.1) are cited from the preprint Collins–Tong [12] (arXiv:2507.05395). Since the present paper's Sections 3 and 6 closely parallel that work, a brief remark in the introduction delineating precisely what is taken from [12] and what is new here (the non-summable error treatment, the q-concave equality case without conic structure, the localized Appendix A) would sharpen the novelty statement.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: pure a-priori estimate derived from external assumptions via monotonicity, blow-up, and rigidity; citations are independent lemmas, not self-restatements of the target.

full rationale

Theorem 1.1 is a global Sobolev bound under stated hypotheses (bounded convex domains; continuous densities bounded away from 0 and ∞). The derivation chain is self-contained against those inputs: (i) almost-monotonicity of F_{x_0} with additive continuous-modulus error (Thm 3.1), proved by smooth computation plus approximation; (ii)–(iii) normalized blow-ups and Borell dimension reduction (Lemmas 5.1–5.4); (iv) equality-case rigidity giving half-section decay on good scales (Thms 6.3–6.4); (v) Cesàro o(N) bad-scale count (Prop 7.4) yielding sub-power section control (Prop 7.5), then Caffarelli interior W^{2,p} plus Savin’s covering. Prior work (Caffarelli, Collins–Tong, Chen–Liu–Wang, Borell, Figalli–Jhaveri, Savin) enters only as lemmas under hypotheses that do not include the target conclusion; there are no fitted parameters, no prediction-from-fit loops, and no uniqueness theorem that merely renames the claim. Self-citations are incremental (weaker density/boundary assumptions) and not load-bearing by definition. Score 0 is the honest finding.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

The result is a conditional regularity theorem in the classical Caffarelli framework. Load-bearing external inputs are convexity of domains, continuity and uniform positivity/boundedness of densities, equal mass, and the quadratic cost. Analytic machinery (Brenier maps, Caffarelli C^{1,α} and interior W^{2,p}, John ellipsoids, Borell s-concavity, Brunn–Minkowski, Figalli–Jhaveri local regularity, Savin covering) is taken from the literature. No free parameters are fitted; no new physical entities are postulated.

assumptions (7)
  • domain assumption Ω, Ω* bounded open convex; f∈C(Ω̄), g∈C(Ω̄*); 0<λ≤f,g≤Λ; equal total mass (display (1.1))
    Hypothesis of Theorem 1.1; convexity is essential for Caffarelli boundary C^{1,α} and section geometry.
  • standard math Brenier’s theorem: unique (a.e.) optimal map is the gradient of a convex potential (and dual)
    Cited [2,19]; used to define u,v and the Monge–Ampère structure.
  • standard math Caffarelli global C^{1,α} for continuous positive densities on bounded convex domains
    Prop 2.1(i) via [6,7,16]; supplies section diameter decay diam D_r ≤ C r^σ used to feed moduli of continuity into the almost-monotonicity error.
  • standard math Caffarelli interior W^{2,p} estimates for the Monge–Ampère equation
    Invoked in the conclusion with Savin’s covering [3,22] to upgrade section shape control to global W^{2,p}.
  • standard math Borell’s classification of s-concave measures (density of 1/(n-m)-concave type on the affine hull)
    Lemma 5.2 citing [1,15]; load-bearing for dimensional reduction when blow-up source measure is lower-dimensional.
  • standard math Local strict convexity/C^1/homeomorphism for locally finite doubling OT maps (Figalli–Jhaveri type)
    Appendix Prop A.1, localized from [14]; used for infinite-mass limit transports in §§5–6.
  • ad hoc to paper Almost-monotonicity of F_{x_0}(r) with error controlled by ∫(ω_f+ω_g)(Cs^σ)/s ds, without Dini summability
    Theorem 3.1, adapted from Collins–Tong smooth computation but proved here for continuous densities via approximation; this is the paper’s main new analytic axiom for the iteration.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Global $W^{2,p}$ Regularity in Optimal Transport." pith.science (2026). https://pith.science/paper/YJ3FORXC

@misc{pith2026260724666,
  author       = {Pith},
  title        = {Pith review of: Global $W^2,p$ Regularity in Optimal Transport},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YJ3FORXC}},
  note         = {Machine review of arXiv:2607.24666}
}
abstract

In this paper we establish global $W^{2,p}$ estimates for the convex potential functions of quadratic optimal transport between bounded convex domains. The source and target densities are assumed only to be continuous and bounded away from zero and infinity, and no regularity is imposed on the domain boundaries.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 2 linked inside Pith

  1. [12]

    T. C. Collins and F. Tong,Boundary regularity of optimal transport maps on convex domains, arXiv:2507.05395 (2025)

    arXiv 2025

  2. [9]

    S. Chen, J. Liu, and X.-J. Wang,Global regularity for the Monge–Amp` ere equation with natural boundary condition, Ann. of Math. (2)194(2021), no. 3, 745–793. GLOBALW 2,p REGULARITY IN OPTIMAL TRANSPORT 31

    2021

  3. [23]

    Savin and H

    O. Savin and H. Yu,Regularity of optimal transport between planar convex domains, Duke Math. J.169(2020), no. 7, 1305–1327

    2020

  4. [3]

    L. A. Caffarelli,Interior W 2,p estimates for solutions of Monge–Amp` ere equations, Ann. of Math. (2)131 (1990), 135–150

    1990

  5. [22]

    Savin,Global W 2,p estimates for the Monge–Amp` ere equation, Proc

    O. Savin,Global W 2,p estimates for the Monge–Amp` ere equation, Proc. Amer. Math. Soc.141(2013), no. 10, 3573–3578

    2013

  6. [1]

    Borell,Convex set functions ind-space, Period

    C. Borell,Convex set functions ind-space, Period. Math. Hungar.6(1975), 111–136

    1975

  7. [2]

    Brenier,Polar factorization and monotone rearrangement of vector-valued functions, Comm

    Y. Brenier,Polar factorization and monotone rearrangement of vector-valued functions, Comm. Pure Appl. Math.44(1991), no. 4, 375–417

    1991

  8. [4]

    L. A. Caffarelli,Some regularity properties of solutions of Monge–Amp` ere equations, Comm. Pure Appl. Math. 44(1991), 965–969

    1991

Show all 25 references
  1. [5]

    L. A. Caffarelli,The regularity of mappings with a convex potential, J. Amer. Math. Soc.5(1992), no. 1, 99–104

    1992

  2. [6]

    L. A. Caffarelli,Boundary regularity of maps with convex potentials, Comm. Pure Appl. Math.45(1992), 1141–1151

    1992

  3. [7]

    L. A. Caffarelli,Boundary regularity of maps with convex potentials—II, Ann. of Math. (2)144(1996), 453–496

    1996

  4. [8]

    L. A. Caffarelli and R. J. McCann,Free boundaries in optimal transport and Monge–Amp` ere obstacle problems, Ann. of Math. (2)171(2010), no. 2, 673–730

    2010

  5. [10]

    S. Chen, J. Liu, and X.-J. Wang, C 2,α regularity of free boundaries in optimal transportation, Comm. Pure Appl. Math.77(2024), no. 1, 731–794

    2024

  6. [11]

    S. Chen, Y. Li, and J. Liu,Optimal (partial) transport to non-convex polygonal domains, preprint, arXiv:2311.15655, 2023

    2023 arXiv

  7. [13]

    Figalli,The Monge–Amp` ere Equation and Its Applications, Zurich Lectures in Advanced Mathematics, European Mathematical Society, Z¨ urich, 2017

    A. Figalli,The Monge–Amp` ere Equation and Its Applications, Zurich Lectures in Advanced Mathematics, European Mathematical Society, Z¨ urich, 2017

    2017

  8. [14]

    Figalli and Y

    A. Figalli and Y. Jhaveri,Regularity properties of monotone measure-preserving maps, Adv. Nonlinear Stud.23 (2023), no. 1, 20220057

    2023

  9. [15]

    Fradelizi and O

    M. Fradelizi and O. Gu´ edon,The extreme points of subsets of s-concave probabilities and a geometric localization theorem, Discrete Comput. Geom.31(2004), 327–335

    2004

  10. [16]

    Jhaveri and O

    Y. Jhaveri and O. Savin,On the regularity of optimal transports between degenerate densities, Arch. Ration. Mech. Anal.245(2022), no. 2, 819–861

    2022

  11. [17]

    Loeper,On the regularity of solutions of optimal transportation problems, Acta Math.202(2009), no

    G. Loeper,On the regularity of solutions of optimal transportation problems, Acta Math.202(2009), no. 2, 241–283, doi:10.1007/s11511-009-0037-8

    2009 doi

  12. [18]

    X.-N. Ma, N. S. Trudinger, and X.-J. Wang,Regularity of potential functions of the optimal transportation problem, Arch. Ration. Mech. Anal.177(2005), no. 2, 151–183

    2005

  13. [19]

    R. J. McCann,Existence and uniqueness of monotone measure-preserving maps, Duke Math. J.80(1995), no. 2, 309–323

    1995

  14. [20]

    Mooney and A

    C. Mooney and A. Rakshit,Sobolev regularity for optimal transport maps of nonconvex planar domains, SIAM J. Math. Anal.56(2024), no. 4, 4742–4758

    2024

  15. [21]

    R. T. Rockafellar,Convex Analysis, Princeton Math. Ser. 28, Princeton University Press, Princeton, 1970

    1970

  16. [24]

    Villani,Topics in Optimal Transportation, Grad

    C. Villani,Topics in Optimal Transportation, Grad. Stud. Math., vol. 58, Amer. Math. Soc., Providence, RI, 2003

    2003

  17. [25]

    Villani,Optimal Transport: Old and New, Grundlehren Math

    C. Villani,Optimal Transport: Old and New, Grundlehren Math. Wiss., vol. 338, Springer-Verlag, Berlin, 2009. School of Mathematical Sciences, University of Science and Technology of China, Hefei, Anhui 230026, China Email address:chenshib@ustc.edu.cn Institute for Theoretical ...

    2009

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.