REVIEW 3 major objections 3 minor 12 references
On a problem of optimal mixing
T0 review · 3 major / 3 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read This paper proves that simultaneous optimal transport with quadratic cost on Euclidean space has a deterministic optimal map whenever the density ratios are simple and continuous almost everywhere, so the Monge and Kantorovich solutions…
desk verdict The paper's discrete and equality results have merit, but the headline coincidence theorem is false—the n=1 perpendicular-segment counterexample kills it. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the constrained plan set $\Pi(\vec\mu,\nu)$, defined by the requirement that the conditional kernel $\pi_x$ sends each source measure $\mu_i$ to the same target $\nu$; this is what makes the transport simultaneous. The argument is carried by rewriting the constraint as $\int \frac{d\mu_i}{d\mu}\, d\pi_y = 1$ for $\nu$-almost every $y$, by a variational lemma showing optimal plans are $c$-monotone with respect to competitors that respect these constraints, and by the convex-analysis theorem that every cyclically monotone set lies in the subdifferential of a convex function. In the discrete case the central object is the simplex-valued map $\psi(x)=\frac1n(\frac{d\mu_1}{d\mu}(x),\ldots,\frac{d\mu_n}{d\mu}(x))$: optimal mixing points are exactly barycenters of minimal subsets, of size at most $n$, whose $\psi$-images contain the centre of the simplex.
What would settle it
Take $n=1$, $X=Y=\mathbb{R}^2$, quadratic cost, and let $\mu$ be normalized one-dimensional surface measure on the unit circle, with $\nu$ uniform on the unit disk. The only density ratio is $d\mu_1/d\mu=1$, which is simple and continuous $\lambda$-a.e., so the theorem would force the optimal plan onto a single-valued graph; solving the Kantorovich problem in this classical configuration shows the optimal plan's fibers over the circle are not singletons, contradicting the claimed graph support.
Extended reading notes
Core claim
On the paper's own terms, the central discovery is Theorem 5.8: in $\mathbb{R}^d$ with cost $\|x-y\|^2$, under the hypotheses that the density ratios $d\mu_i/d\mu$ are simple and continuous $\lambda$-a.e., the solution of the simultaneous Kantorovich problem is supported on the graph of a function and is unique with respect to Lebesgue measure. The same map therefore solves the Monge problem, so the minimum of the Kantorovich functional and the infimum of the Monge functional coincide, and the solution is shared. The proof divides the space into sets on which the density ratios are constant, shows the optimal plan is cyclically monotone on each such set, and then uses the fact that a convex function's subdifferential is single-valued almost everywhere to conclude the plan concentrates on a graph.
Load-bearing premise
The load-bearing premise is that "almost everywhere" with respect to Lebesgue measure is enough to control where the source measures sit: the sources must not put mass on the negligible set where a convex potential has more than one slope.
Editorial extensions
If this is right
- If Theorem 5.8 holds, simultaneous optimal transport in Euclidean space with simple density ratios is a deterministic problem: one function moves every $\mu_i$ onto the common target, and that function is the unique solution up to Lebesgue-null changes.
- The equality of Monge infimum and Kantorovich minimum on compact Souslin spaces means the map-based and plan-based formulations cannot disagree about the optimal cost for simultaneous transport.
- For the discrete problem with unfixed target, the characterization by barycenters of subsets of size at most $n$ turns the search for the optimal target into a finite linear program.
- The $c$-monotonicity of optimal constrained plans means that on each region of constant density ratio, the standard convex-potential machinery applies even though the global plan is constrained.
Reading between the lines
- The theorem's reliance on Lebesgue almost-everywhere differentiability suggests that for non-simple densities, simultaneous optimal transport should have a gradient-of-a-convex-function representation only when the source measures are absolutely continuous; singular source measures are where the graph conclusion should fail.
- The discrete barycenter characterization implies an algorithmic route for multidimensional optimal mixing: enumerate minimal subsets of size at most $n$ whose $\psi$-images surround the simplex centre, then solve a linear program over their barycenters.
- Because the constraint set is linear, the same reformulation connects the problem to other constrained transport problems, and the region-wise $c$-monotonicity may yield structural results for constrained plans beyond the simultaneous case.
- For $n=1$, the theorem should reproduce the classical one-dimensional monotone rearrangement; checking that special case against known singular examples would test whether the Lebesgue-a.e. argument really controls the full source mass.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies simultaneous optimal transportation, where several source measures μ_i are transported to a common target measure ν (fixed or variable) by a single plan or map, with cost c(x,y). It claims existence of optimal plans on completely regular spaces (Theorem 2.6), a structural reduction for discrete problems, equality of the Monge and Kantorovich values (Theorem 4.2), and a Euclidean result asserting that, when the density ratios dμ_i/dμ are simple and continuous λ-a.e., the optimal plan is supported on the graph of a map and is unique (Theorem 5.8). The arguments rely on Lyapunov's theorem, a variational lemma from [12], Rockafellar's theorem, and Prokhorov compactness.
Significance. If the main claims were correct, the paper would provide an attractive extension of classical optimal transport: simultaneous transport with variable target marginal, a linear-programming reduction on finite sets, and a Monge–Kantorovich coincidence theorem for simple density ratios. The discrete reduction in Section 3 is potentially useful, and the Lyapunov-based viewpoint is natural. However, the central Euclidean theorem is false as stated, and the proofs of the existence theorem and of the Monge–Kantorovich equality contain substantial gaps. The advertised coincidence of Monge and Kantorovich solutions in Euclidean space therefore does not follow from the manuscript as written.
major comments (3)
- [Section 5, Theorem 5.8] Theorem 5.8 is false as stated. Take n=1, X=Y=R^2, μ uniform on [0,1]×{0}, and ν uniform on {0}×[0,1]. Then dμ_1/dμ ≡ 1, so the hypotheses hold. For every π ∈ Π(μ,ν), the quadratic cost ∫||x−y||² dπ equals ∫x² dμ + ∫y² dν, which is independent of π; hence the product measure μ×ν is optimal. But μ×ν is not supported on the graph of any function. The proof's step that the subdifferential of the convex function is a singleton λ-a.e. only excludes a Lebesgue-null set, and the statement contains no hypothesis that the source marginal μ gives zero mass to that exceptional set. In the n=1 case the theorem would imply that every atomless measure has a deterministic optimal map for the quadratic cost, which is false for singular marginals. The theorem and the abstract's claim of coincidence in Euclidean space need an assumption such as μ ≪ λ, or an explicit condition that μ charges no Rockafellar exceptional set.
- [Section 2, proof of Theorem 2.6] The compactness proof for Π(→μ,ν) is not supported as written. To prove closedness, the text invokes Proposition 4.3.17 of [5] to pass from π_n → π to (dμ_i/dμ)π_n → (dμ_i/dμ)π. This requires more than μ-a.s. continuity of the densities; the densities are not assumed bounded, and multiplication by an unbounded μ-a.e. continuous function need not preserve weak convergence. In addition, Prokhorov's theorem is invoked for completely regular spaces, although the standard form of that theorem requires complete metric or otherwise Prokhorov spaces. The existence theorem 2.6 therefore needs either a Polish-space hypothesis, bounded densities, or a different argument for compactness.
- [Section 4, proof of Theorem 4.2] The construction of the map T in the proof of equality (4.1) is incomplete. After partitioning each A_i into sets X^i_k with vector measure (μ_1(X^i_k),...,μ_n(X^i_k)) equal to (∫ g_k dμ_1,...,∫ g_k dμ_n), the proof asserts that a map T^i_k defined on X^i_k can 'translate the constraints of measures μ_1,...,μ_n to the constraint of measure ν on B_k multiplied by μ_j(X^i_k)'. No such map is constructed. Since the total masses μ_j(X^i_k) generally vary with j, a single map on X^i_k cannot push all μ_j forward to a common target measure on B_k; a further Lyapunov-type splitting inside X^i_k is required. As it stands, the equality of the Monge infimum and Kantorovich minimum is not established.
minor comments (3)
- [Equation (1.1)] The notation Π(µ,ν) is used for the constrained simultaneous-transport plans, while later in the same paragraph Π(µ,ν) denotes the usual set of couplings with fixed marginals; this overloaded notation should be disambiguated.
- [Section 5, proof of Theorem 5.8] The sentence 'The sets {A_k} can be considered closed, since the functions dμ_i/dμ are continuous λ-a.e.' is not justified; a function that is λ-a.e. equal to a continuous function need not have closed level sets. This point is secondary to the failure of the theorem, but it should be corrected in any revision.
- [Proposition 2.8] The tightness argument for the marginal measures ν_n implicitly assumes that the relevant second moments are uniformly bounded; if μ does not have finite second moment, the statement that mass escaping to infinity makes the cost arbitrarily large needs additional justification or a growth condition on μ.
Circularity Check
No circularity: Theorem 5.8's flaw is a missing absolute-continuity hypothesis, not a derivation that reduces to its own inputs.
full rationale
The paper contains no circular derivation. The existence argument in Theorem 2.6 uses Prokhorov compactness and continuity of the density ratios; the duality theorem is quoted from Wang and Zhang [11], a different research group; the variational lemma is quoted from Zaev [12]; Theorem 4.2 is an independent Lyapunov-based proof in the spirit of Bogachev–Kalinin–Popova and Pratelli; and Theorem 5.8 combines Rockafellar's theorem with the λ-a.e. single-valuedness of subdifferentials. None of these ingredients is the conclusion being proved. There are no fitted parameters, no calibration to data, and no self-citation used as the sole justification of a load-bearing assumption. The serious defect in Theorem 5.8 is a missing hypothesis, not circularity: the proof passes from 'the subdifferential of the convex function λ-a.e. consists of one element' to 'π is supported on a graph,' although the source marginal μ = (1/n)Σ μ_i may charge the λ-null exceptional set, and for n=1 the theorem would assert a deterministic optimal map for every atomless μ, which is false for singular marginals. That is an overclaim in a non-circular proof, so the circularity score is 0.
Assumptions & free parameters
assumptions (6)
- standard math Lyapunov's theorem for atomless vector measures
- domain assumption Prokhorov's theorem: tightness implies weak compactness
- domain assumption Variational lemma from [12]
- standard math Rockafellar's theorem: cyclically monotone sets are contained in the subdifferential of a convex function
- standard math The subdifferential of a convex function is single-valued Lebesgue-a.e.
- standard math Caratheodory's theorem
Cite this review
Pith. "Pith review of On a problem of optimal mixing." pith.science (2026). https://pith.science/paper/PIBB462Q
@misc{pith2026241116651,
author = {Pith},
title = {Pith review of: On a problem of optimal mixing},
year = {2026},
howpublished = {\url{https://pith.science/paper/PIBB462Q}},
note = {Machine review of arXiv:2411.16651}
}
read the original abstract
We consider the simultaneous optimal transportation of measures, where the target marginal is not necessarily fixed. For this problem, we prove the existence of a solution for completely regular spaces and investigate the structure of the discrete problem. We establish a connection between the Monge problem and the Kantorovich problem by showing that their functionals are equal and that the solutions coincide in Euclidean space.
Reference graph
Works this paper leans on
-
[12]
D. Zaev. On the Monge–Kantorovich problem with additional linear co nstraints. In: Mathematical Notes 98 (2015), pp. 725–741. 11
work page 2015
-
[5]
V. I. Bogachev. Weak Convergence of Measures. American Mathematical Society, (2018)
work page 2018
-
[1]
M. Beiglb¨ ock and N. Juillet. On a problem of optimal transport under marginal martingale constraints. In: The Annals of Probability, 44(1), p. 42–106, 2016
work page 2016
-
[2]
V. I. Bogachev, and A. N. Kalinin and S. N. Popova. On the Equality of Values in the Monge and Kantorovich Problems , Journal of Mathematical Sciences, 238(4), (2019), pp. 377–389
work page 2019
-
[3]
V. I. Bogachev. Measure Theory. Springer Berlin Heidelberg, (2007)
work page 2007
-
[4]
V. I. Bogachev. The Kantorovich problem of optimal transportation of measur es: new directions of research . In: Uspekhi Matematicheskikh Nauk 77.5(467), (2022), pp. 3–52
work page 2022
-
[6]
The Monge–Kantorov ich problem: achieve- ments, connections, and perspectives
V. I. Bogachev and A. V. Kolesnikov. “The Monge–Kantorov ich problem: achieve- ments, connections, and perspectives”. In: Uspekhi Matema ticheskikh Nauk 67.5, (2012), pp. 3–110
work page 2012
-
[7]
On semi-discrete sub-partitions of vector-valued measures
S. Gover and G. Wolansky On semi-discrete sub-partitions of vector-valued mea- sures., arXiv:2007.03982, (2020)
work page Pith review arXiv 2020
Show all 12 references
-
[8]
Gozlan et al
N. Gozlan et al. Kantorovich duality for general transport costs and applic ations. In: Journal of Functional Analysis 273.11 (2017), pp. 3327– 3405
2017
-
[9]
Lyapunov
A. Lyapunov. On completely additive vector functions. In: I. Izv. Russian Acad. Sci. (1940)
1940
-
[10]
A. M. Pratelli. On the equality between Monge’s infimum and Kantorovich’s mi n- imum in optimal mass transportation. In: Annales De L Institut Henri Poincare probabilites Et Statistiques 43 (2007), pp. 1–13
2007
-
[11]
Wang and Z
R. Wang and Z. Zhang. Simultaneous Optimal Transport, arXiv:2201.03483, (2023)
2023 arXiv
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.