Probability · arXiv:2611.16429 v1 · submitted 12 Nov 2026 · 28 references
On the continuity of martingale optimal transport in higher dimensions
The paper says
“We investigate qualitative and quantitative continuity of martingale optimal transport under perturbations of both marginals in dimensions d ≥ 2.”
more of the abstract
We assume that each pair of marginals admits a reference martingale coupling whose density relative to the product of its marginals is bounded above and away from zero, uniformly along the sequence. If the limiting target measure is not carried by a hyperplane, we prove continuity under W1-convergence for continuous costs of at most linear growth, assuming only finite first moments. For Lipschitz costs, a uniform p-moment bound on the targets, p > 1, yields an explicit stability estimate with exponent (1−1/p)/2; uniformly bounded target supports give exponent 1/2. The proof constructs a correction of couplings with small martingale bias while preserving both marginals. Truncated moment corrections allow the target measure to have unbounded support. The construction also shows that every martingale coupling of the limiting marginals can be approximated in W1 by martingale couplings of the approximating marginals. We also discuss counterexamples showing that neither the uniform density condition nor non-degeneracy of the limiting target measure can be omitted in general.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
The passes disagreed. Reader: holds up. Skeptic: holds up only with conditions. Referee: major revision.
Good paper with a genuine gap in Proposition 4.2: the bound s ≤ 1 does not hold for small κ, and the repair construction is not justified as written. This line is wrong
This review says the paper's mathematics is wrong. Pith did not check that against the paper's source.
1 major objection · 3 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 28 references say.
The paper says, at Proposition 4.2, Step 3 (lines after (4.10))
The review did not quote this passage.
Pith says: Objection 1 of 1 · major
The bound s ≤ 1 does not follow from the stated assumptions for all κ∈(0,1). Specifically, s := μ(E)/κ². From (4.7) we have μ(E) ≤ ‖b_π‖_{L¹}^{1/2} ≤ √(κ/4) = √κ/2, so s ≤ (√κ/2)/κ² = 1/(2κ^{3/2}). This exceeds 1 whenever κ < 2^{-2/3} ≈ 0.63, which is allowed by the paper's assumptions.
more of objection 1
The construction of π_0 requires s ≤ 1 to ensure that the mixture (1−s)π'_x + sπ̄_x − (1−s)ζ is a probability kernel; if s > 1 the kernel may not be well‑defined. This gap affects the repair step that compares V and P_R uniformly, and thus the central continuity result of Theorem 1.4. The authors must either correct the estimate, impose an additional lower bound on κ, or adjust the definition of β₀ to ensure s ≤ 1 for all admissible κ.
This review says the paper's mathematics is wrong. Pith did not check that against the paper's source.
3 minor objections
The paper says, at General
Pith says: minor objection 1 of 3
The paper is well‑structured and clearly written. The notation is consistent and the proofs are generally detailed.
The paper says, at Equation (4.1)
Pith says: minor objection 2 of 3
The symbol A is used both for a constant and for a set; this could be confusing. Consider renaming the constant.
show 1 more
The paper says, at Line after (2.6)
Pith says: minor objection 3 of 3
The definition of λ_T(ν) as the smallest eigenvalue of Σ_T(ν) is clear, but note that Σ_T(ν) may be singular for small T; the existence of T with λ_T(ν)>0 is guaranteed by Lemma 2.7(ii) under Assumption 1.3.
What would settle it
For κ=0.2, compute s from the proof: μ(E) ≤ ∥b_π∥^{1/2}_{L^1} ≤ √(0.2/4)=√0.05≈0.224, so s ≤ 0.224/0.04 = 5.6 >1. Check whether the kernel π_0 defined in Step 3 remains a valid (nonnegative) coupling under this s. If not, provide an alternative bound or a modification of the threshold that ensures s≤1 for all κ.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 13 Nov 2026
- Reader holds up
- Skeptic disagrees: says it holds up only with conditions Proof of Proposition 4.2 contains a gap: the bound s≤1 in Step 3 does not follow from the stated assumptions for small κ, potentially breaking the repair construction.
- Editor wrote the verdict above
- Referee major revision: 1 major objection, 3 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
The paper tackles an important open problem — continuity of MOT in higher dimensions — and does two things well. First, it introduces a clean penalization that replaces the usual minimax argument with a Lipschitz-stable surrogate PR. Second, it constructs a repair map that turns an approximate martingale coupling into an exact one, using the reference coupling as a reservoir. The resulting continuity theorem with explicit rates is the first of its kind for d ≥ 2 and is clearly the main event.
The structure is logical, the writing is precise, and the counterexamples are instructive. The reader's positive verdict is understandable.
But there is a real gap in Proposition 4.2, Step 3, exactly where the stress-test note lands. The paper sets s := μ(E)/κ² and claims s ≤ 1 using μ(E) ≤ ‖bπ‖^{1/2}_{L¹(µ)} ≤ β₀^{1/2} ≤ κ/2 (up to constants). The first inequality is (4.7) — fine. But the second step gives μ(E) ≤ √κ / 2, so s ≤ (√κ / 2) / κ² = 1/(2κ3/2). This exceeds 1 whenever κ < 2−2/3 ≈ 0.63, and κ ∈ (0,1) is allowed by the assumptions. The argument as written does not justify s ≤ 1 for small κ. The construction then mixes (1−s)π'ₓ + s̄πₓ − (1−s)ζ; if s > 1 the coefficient (1−s) is negative and the kernel may not be a probability measure. The whole repair step is compromised, and with it the uniform comparison between V and PR that drives the main theorem.
This is not a minor oversight — it is a load-bearing flaw in the proof as it stands. I suspect the result is still true and the gap can be closed by either restricting κ to a safe range (e.g., κ ≥ 1/2) or by a more careful handling of the small-κ regime, perhaps by choosing β₀ differently or by a separate argument for very small κ. The authors should address this.
Despite this, the paper is a serious piece of work. The main idea is sound, the tools are novel, and the counterexamples are valuable. It deserves a rigorous peer review, but the current version of Proposition 4.2 is not fully proved.
For a reading group: maybe — the gap is a good teaching moment. Would I cite it? Not in its present form, but I expect a corrected version to be citable. The editor should send this to a referee but flag the gap explicitly.
The strongest objection
In Proposition 4.2 (Step 3, lines around equation after (4.10)), the paper sets s := μ(E)/κ^2 and claims s≤1 under the condition ∥b_π∥_{L^1(μ)} ≤ β_0 ≤ κ/4. However, from (4.7) we have μ(E) ≤ ∥b_π∥^{1/2}_{L^1(μ)} ≤ √(κ/4) = √κ/2. Hence s ≤ (√κ/2)/κ^2 = 1/(2κ^{3/2}). This exceeds 1 whenever κ < 2^{-2/3} ≈ 0.63, which is allowed by the assumptions (κ∈(0,1) arbitrary). The claimed bound s≤1 is therefore unjustified for small κ. The construction of π_0 (the kernel that restores the second marginal) requires s≤1 to ensure (1−s)π'ₓ + s¯πₓ − (1−s)ζ is a probability kernel. If s>1, the mixture may not be well-defined. This gap affects the entire repair step that is needed to compare V and P_R uniformly, and thus the central continuity result of Theorem 1.4.
The reader's weakest_assumption focused on the strength of Assumption 1.1 itself. While that is a valid concern about applicability, I identified an internal gap in the proof of Proposition 4.2, a key lemma used to compare V and P_R uniformly. The error is not about the assumption's plausibility but about a non-sequitur in the derivation. If the gap cannot be closed, the uniform error bound e(R) in Corollary 4.3 may not hold for small κ, undermining the main theorem. The paper may be fixable by adjusting the condition on β_0 or by modifying the construction, but as written the proof is incomplete. Hence a conditional verdict is appropriate.
The referee's report
If correct, the result is an important contribution to martingale optimal transport in higher dimensions, where continuity was known to fail in general (Brückerhoff–Juillet 2022). The paper identifies a quantitative condition that rules out the hidden constraints causing discontinuity, and provides the first general continuity result and rates for d≥2. The repair construction is novel and the counterexamples are instructive. The significance is high.
Pith's reading of the paper (a machine's summary, not the authors')
The paper establishes continuity of the martingale optimal transport value under Wasserstein-1 perturbations of both marginals in dimensions two and higher. The key is a pair of conditions: a uniform density bound on a reference martingale coupling (Assumption 1.1), and a non-degeneracy condition that prevents the limit target from being carried by a hyperplane (Assumption 1.3). Under these, for continuous costs of at most linear growth, the value converges; for Lipschitz costs, an explicit rate of convergence is given. The proof constructs a correction mechanism that turns any coupling with small martingale bias into an exact martingale coupling at small cost, using the reference coupling as a reservoir. The results are sharp in the sense that each condition alone is insufficient, as shown by counterexamples.
What carries the argument. The central object is the repair map (Proposition 4.2), which turns a coupling with martingale bias bounded in L1(μ) into a true martingale coupling. The map proceeds in three steps: first replace kernels at points with large bias by the reference coupling; then restore the second marginal by mixing in a small amount of the reference coupling and subtracting a signed measure; finally cancel the remaining uniformly small bias by a tilting of the target measure along φ_T, a truncated version of y−m that remains dominated by ν. The reference coupling satisfying (Aκ) provides the room (κ ν ≤ π̄_x ≤ $κ^{{-1}}$ ν) needed to keep all intermediate signed perturbations non-negative.
Core claim. The central claim is that, for a sequence of marginal pairs satisfying the uniform density condition (Aκ) with a fixed κ and with the limit target not concentrated on a hyperplane, the martingale optimal transport value V(μ_n, ν_n) → V(μ, ν) whenever μ_n, ν_n converge in Wasserstein-1 to μ, ν with finite first moments. For Lipschitz costs and uniform p-th moment bounds on the targets (p>1), the error is bounded by L Δ_n + C $Δ_n^{{(1-1/p)/2}}$, where Δ_n is the sum of Wasserstein distances. The machinery behind the proof is a 'repair map' that takes any coupling with small L1 martingale bias and corrects it into an exact martingale coupling, with a controlled increase in transport cost; the correction uses the reference coupling to absorb removed mass and tilt the target distribution along the direction of the bias.
The premise it rests on. Assumption 1.1 (Aκ): for every pair of approximating marginals there exists a reference martingale coupling whose density with respect to the product of its marginals is bounded between κ and $κ^{{-1}}$ uniformly; this is a strong structural requirement that may fail in many natural settings.
If the paper is right (Pith's inferences, not the authors')
Martingale optimal transport is continuous under Wasserstein-1 convergence whenever a uniformly bounded density reference coupling exists and the limit target is non-degenerate, a stability guarantee that fails without these conditions.
Every martingale coupling of the limiting marginals can be approximated in W1 by martingale couplings of the approximating marginals, establishing a converse to the standard compactness result.
For Lipschitz costs, the convergence rate of the value is at most of order Δ_n^{(1-1/p)/2} under a p-th moment bound, or of order Δ_n^{1/2} under a uniform support bound.
The uniform density assumption (Aκ) forces the support of each first marginal μ_n to lie in a bounded set, and the non-degeneracy of ν ensures that the smallest eigenvalue of the truncated covariance matrix of ν_n stays bounded away from zero for large n.
The two conditions are jointly sufficient but individually necessary: Example 6.1 shows that (Aκ) alone without non-degeneracy still allows a jump in the limit, and a variant of [6] shows that non-degeneracy alone without (Aκ) does not guarantee continuity.
The repair map effectively demonstrates that, under the given conditions, the relaxed problem with an L1 penalty on the martingale bias can be uniformly approximated by the true martingale optimal transport problem; this suggests a general principle connecting constraint qualification to stability.
The truncated tilt construction using φ_T may be adaptable to other transport problems where the target has unbounded support but finite p-th moments, providing a way to handle unboundedness while keeping perturbations dominated.
The counterexamples suggest that the true topological barrier to continuity in higher dimensions is the simultaneous presence of a 'thin' limiting target and the lack of a reference coupling with uniform density: either condition alone introduces a rigidity that may disappear in the limit.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
We investigate qualitative and quantitative continuity of martingale optimal transport under perturbations of both marginals in dimensions d ≥ 2. We assume that each pair of marginals admits a reference martingale coupling whose density relative to the product of its marginals is bounded above and away from zero, uniformly along the sequence. If the limiting target measure is not carried by a hyperplane, we prove continuity under W1-convergence for continuous costs of at most linear growth, assuming only finite first moments. For Lipschitz costs, a uniform p-moment bound on the targets, p > 1, yields an explicit stability estimate with exponent (1−1/p)/2; uniformly bounded target supports give exponent 1/2. The proof constructs a correction of couplings with small martingale bias while preserving both marginals. Truncated moment corrections allow the target measure to have unbounded support. The construction also shows that every martingale coupling of the limiting marginals can be approximated in W1 by martingale couplings of the approximating marginals. We also discuss counterexamples showing that neither the uniform density condition nor non-degeneracy of the limiting target measure can be omitted in general.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-11-13
The authors' main claim, as the classifier picked it out 0.97
“If the limiting target measure is not carried by a hyperplane, we prove continuity under $W_1$-convergence for continuous costs of at most linear growth, assuming only finite first moments.”
How the main claim is stated: asserted 1.00
What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.94 · human participants 0.01 · controlled comparison 0.02 · randomized assignment 0.01 · states quantitative result 0.94 · compares with prior results 0.12 · mentions code or data availability 0.02
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| We investigate qualitative and quantitative continuity of martingale optimal transport under perturbations of both marginals in dimensions $d\geq2$. | contribution 0.98 | 0.89 | 0.12 | asserted 0.65 |
| We assume that each pair of marginals admits a reference martingale coupling whose density relative to the product of its marginals is bounded above and away from zero, uniformly along the sequence. | method 0.55 | 0.37 | 0.03 | asserted 0.91 |
| If the limiting target measure is not carried by a hyperplane, we prove continuity under $W_1$-convergence for continuous costs of at most linear growth, assuming only finite first moments. | result 1.00 | 0.94 | 0.51 | asserted 1.00 |
| For Lipschitz costs, a uniform $p$-moment bound on the targets, $p>1$, yields an explicit stability estimate with exponent $(1-1/p)/2$; uniformly bounded target supports give exponent $1/2$. | result 1.00 | 0.92 | 0.18 | asserted 1.00 |
| The proof constructs a correction of couplings with small martingale bias while preserving both marginals. | method 1.00 | 0.95 | 0.05 | asserted 1.00 |
| Truncated moment corrections allow the target measure to have unbounded support. | method 0.96 | 0.87 | 0.03 | asserted 0.99 |
| The construction also shows that every martingale coupling of the limiting marginals can be approximated in $W_1$ by martingale couplings of the approximating marginals. | result 0.97 | 0.94 | 0.16 | asserted 1.00 |
| We also discuss counterexamples showing that neither the uniform density condition nor non-degeneracy of the limiting target measure can be omitted in general. | limitation 0.62 | 0.92 | 0.04 | asserted 0.99 |
What kind of paper the classifier reports this is: theoretical result 1.00 · evidence from proof or derivation 1.00
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 6,330 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (3)
Figures are the authors', reproduced from the paper.
References (28)
- [1] J. Backhoff-Veraguas and G. Pammer. Stability of martingale optimal transport and weak optimal trans- port.Ann. Appl. Probab., 32(1):721–752, 2022 on Pith
- [2] M. Beiglböck, P. Henry-Labordère, and F. Penkner. Model-independent bounds for option prices—a mass transport approach.Finance Stoch., 17(3):477–501, 2013 on Pith
- [3] M. Beiglböck, B. Jourdain, W. Margheriti, and G. Pammer. Approximation of martingale couplings on the line in the adapted weak topology.Probab. Theory Related Fields, 183:359–413, 2022 on Pith
- [4] M. Beiglböck and N. Juillet. On a problem of optimal transport under marginal martingale constraints. Ann. Probab., 44(1):42–106, 2016 on Pith
- [5] M. Beiglböck, M. Nutz, and N. Touzi. Complete duality for martingale optimal transport on the line. Ann. Probab., 45(5):3038–3074, 2017 on Pith
- [6] M. Brückerhoff and N. Juillet. Instability of martingale optimal transport in dimensiond≥2.Electron. Commun. Probab., 27:Paper No. 24, 1–10, 2022 on Pith
- [7] G. Carlier, H. Malamut, and M. Sylvestre. Weak optimal transport with moment constraints: constraint qualification, dual attainment and entropic regularization. arXiv preprint arXiv:2511.16211, 2025 on PitharXiv
- [8] F. Chen, G. Conforti, Z. Ren, and X. Wang. Convergence of sinkhorn’s algorithm for entropic martingale optimal transport problem.Mathematics of Operations Research, 2026 on Pith
- [9] H. Chen, R. Xu, and X. Zhang. Entropic martingale optimal transport with Wasserstein relaxed marginal constraints. arXiv preprint arXiv:2611.08743, 2026 on PithPith's reviewarXiv
- [10] H. De March. Entropic approximation for multi-dimensional martingale optimal transport. arXiv preprint arXiv:1812.11104, 2018 on PithPith's reviewarXiv
- [11] H. De March and N. Touzi. Irreducible convex paving for decomposition of multidimensional martingale transport plans.Ann. Probab., 47(3):1726–1774, 2019 on Pith
- [12] S. Eckstein and M. Nutz. Quantitative stability of regularized optimal transport and convergence of Sinkhorn’s algorithm.SIAM J. Math. Anal., 54(6):5922–5948, 2022 on Pith
- [13] A. Galichon, P. Henry-Labordère, and N. Touzi. A stochastic control approach to no-arbitrage bounds given marginals, with an application to lookback options.Ann. Appl. Probab., 24(1):312–336, 2014 on Pith
- [14] N. Ghoussoub, Y.-H. Kim, and T. Lim. Structure of optimal martingale transport plans in general di- mensions.Ann. Probab., 47(1):109–164, 2019. ON THE CONTINUITY OF MARTINGALE OPTIMAL TRANSPORT IN HIGHER DIMENSIONS 29 on Pith
- [15] G. Guo and J. Obłój. Computational methods for martingale optimal transport problems.Ann. Appl. Probab., 29(6):3311–3347, 2019 on Pith
- [16] P. Henry-Labordère. From (martingale) Schrödinger bridges to a new class of stochastic volatility models. arXiv preprint arXiv:1904.04554, 2019 on PithPith's reviewarXiv
- [17] M. M. Kay. Multi-dimensional martingales from mutual information. arXiv preprint arXiv:2512.16544, 2025 on PitharXiv
- [18] A. Kazeykina, Z. Ren, and H. Wang. Exponential convergence of Sinkhorn algorithm for entropy martin- gale optimal transport. arXiv preprint arXiv:2609.18850, 2026 on PithPith's reviewarXiv
- [19] M. Nutz and J. Wiesel. On the martingale Schrödinger bridge between two distributions.arXiv preprint arXiv:2401.05209, 2024 on PithPith's reviewarXiv
- [20] M. Nutz, J. Wiesel, and L. Zhao. Limits of semistatic trading strategies.Math. Finance, 33(1):185–205, 2023 on Pith
show 8 more
- [21] M. Nutz, J. Wiesel, and L. Zhao. Martingale Schrödinger bridges and optimal semistatic portfolios.Fi- nance Stoch., 27(1):233–254, 2023 on Pith
- [22] J. Obłój and P. Siorpaes. Structure of martingale transports in finite dimensions. arXiv preprint arXiv:1702.08433, 2017 on PithPith's reviewarXiv
- [23] R. T. Rockafellar.Convex analysis, volume 28 ofPrinceton Mathematical Series. Princeton University Press, Princeton, NJ, 1970 on Pith
- [24] W. Rudin.Real and complex analysis. McGraw-Hill, New York, third edition, 1987 on Pith
- [25] M. Sion. On general minimax theorems.Pacific J. Math., 8:171–176, 1958 on Pith
- [26] V. Strassen. The existence of probability measures with given marginals.Ann. Math. Statist., 36:423–439, 1965 on Pith
- [27] C. Villani.Optimal transport: Old and new, volume 338 ofGrundlehren der mathematischen Wis- senschaften. Springer-Verlag, Berlin, 2009 on Pith
- [28] J. Wiesel. Continuity of the martingale optimal transport problem on the real line.Ann. Appl. Probab., 33:4645–4692, 2023. (Francesco Mattesini)Department of Mathematics, CIT School, Technical University of Munich, & Munich Center for Machine Learning (MCML), Germany Email add... on Pith
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment