Probability · arXiv:2610.40509 v1 · submitted 29 Oct 2026 · 31 references
Random Matching Between Distinct Distributions in Two Dimensions
The paper says
“Given two independent n-point samples drawn from possibly distinct probability distributions on a compact 2-dimensional surface, we minimize the total squared distance over all perfect bipartite matchings.”
more of the abstract
Under regularity assumptions, the expected matching cost minus n times the squared 2-Wasserstein distance between the sampling distributions grows logarithmically with n. We determine this difference up to an additive o(1) error, including the dependence of the logarithmic coefficient and constant term on geometry and sampling distributions, and establish the existence of a universal constant shared by these formulas. Our analysis expresses the expected matching cost in terms of the cost of transporting a heat-smoothed sample back to its original points. The reductions giving this relation hold in every dimension and for general costs under suitable assumptions, with consequences beyond the setting studied here. In two dimensions, the relation is exact up to an additive o(1) error, so determining the matching cost amounts to evaluating the smoothing cost with the same accuracy. We obtain this evaluation by subdividing the domain and exploiting cancellations between the sample and its own smoothing.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 30 Oct 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
Holds up, with conditions. The 2D conjecture proof is real and the smoothing-cost technique is genuinely new, but the distinct-distribution headline result on polygons rests on a boundary-regularity assumption the paper states rather than proves. This line is wrong
2 major objections · 5 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 31 references say.
The paper says, at Theorem 1, Corollary 1; Sections 5.3, 1.4; Appendices B.3.1, D.2
The review did not quote this passage.
Pith says: Objection 1 of 2 · major
The regularity hypothesis that T is a smooth diffeomorphism up to the boundary with H = DT uniformly positive definite is load-bearing for the central claim, and for convex domains it is assumed rather than derived (b54, b395). All terms of the formula (A_H, h_H, the density integral) and the reduction errors in Theorem 6 and Lemma D.1 depend on …
more of objection 1
… the global comparability lambda|x-z|^2 <= k(x,z), d_H(x,z)^2 <= Lambda|x-z|^2 established in Appendix D.2; if H degenerates at a corner, the identity k(x,z) = 2{Phi(x)-Phi(z)-nabla Phi(z) dot (x-z)} no longer implies that comparability, and the cubic replacement k = d_H^2 + O(|x-z|^3) used by Lemma D.1 fails. For polygonal domains with corner angles in (0,pi) — which Theorem 1 explicitly allows — second-boundary-value Monge-Ampere regularity is not guaranteed by smooth positive densities, and the manuscript itself documents the analogous corner singularity mechanism for the Neumann Poisson potential in Appendix B.3.1 (Hessian unbounded when omega > pi/2). The theorem is correct as a conditional statement, but Corollary 1's message that distinct distributions reduce the logarithmic coefficient, and the abstract's framing, are established only for such regular pairs. I recommend that the authors (i) state prominently, in the abstract or introduction, that for convex domains this regularity is an additional hypothesis that is automatic only on tori; (ii) discuss known sufficient conditions and obstructions for the second boundary value problem on cornered domains; and (iii) note that Example 5.2 satisfies all hypotheses for |epsilon| < 1 on the square, so the condition is non-vacuous for mu != nu.
This review says the paper's mathematics is wrong. Pith did not check that against the paper's source.
The paper says, at Section 1.5, Corollary 4; Appendix A.2
The review did not quote this passage.
Pith says: Objection 2 of 2 · major
The resolution of the Barthe-Bordenave conjecture depends on [17, Theorem 4.3], which is listed as 'to appear' with numbering following arXiv:2409.08612v2. Because this is a headline contribution, please state the publication status explicitly and verify in a short paragraph that the hypotheses of [17, Theorem 4.3] are satisfied by the expanding-cube model with a common Poisson sample size — in particular the concentration assumption with exponent alpha = 0 and the treatment of the symmetric boundary cost b_R^2.
more of objection 2
The verification of the concentration estimate is sketched in Appendix A.2, but the applicability of the theorem to the boundary-relaxed cost is the step that transfers the averaging identity to the Barthe-Bordenave constants, and it would benefit from being made explicit.
5 minor objections
The paper says, at Introduction, first paragraph
Pith says: minor objection 1 of 5
The sentence 'Drawn points independently from each of two probability distributions' should read 'Draw n points independently from each of two probability distributions'; a small copyediting pass is recommended throughout.
The paper says, at Reference [6]
Pith says: minor objection 2 of 5
Please verify the spelling of the first author's surname in reference [6] ('V. Armegioiu'); as printed it is an unusual form and a typographical error in a name would hinder retrieval.
show 3 more
The paper says, at Theorem 4 and Section 4.3
Pith says: minor objection 3 of 5
The equal-mass version of Theorem 4 is stated with 'a possibly different constant kappa*'; since Theorem 5 and Corollary 2 subsequently use kappa* without qualification, distinct symbols (e.g., kappa*_D and kappa*_E) would make it explicit which constant enters the semidiscrete universal constant beta_sd.
The paper says, at Section 1.2, discussion after Corollary 1
Pith says: minor objection 4 of 5
The sentence 'it determines the bias of W_2^2(muhat_n,nuhat_n) as an estimator of W_2^2(mu,nu) up to an o(1/n) error' is correct but easily misread as claiming the bias itself is o(1/n); consider writing 'determines the bias up to an additive o(1/n) error' and displaying the leading ((A_H/2pi) log n)/n term explicitly.
The paper says, at Example 5.2
Pith says: minor objection 5 of 5
For the square case, the example would be clearer if it stated explicitly that |epsilon| < 1 guarantees T is a diffeomorphism of Q onto itself and that q(Tx) = [1 + epsilon cos(2pi x_1)]^{-1} is strictly positive, so that the example indeed falls under the hypotheses of Theorem 1 with mu != nu.
What would settle it
Fix Omega=[0,1]^2, p=1, and q(x1,x2)=(1+eps x1 x2)/(1+eps/4) with small eps>0. Solve the second boundary value Monge-Ampère equation q(grad Phi) det D^2 Phi = 1, grad Phi(Omega)=Omega, either by a finite-element solver with meshes refined at the corners or by matched asymptotic expansions. Record the eigenvalues lambda_min, lambda_max of D^2 Phi along the side x1=0 approaching the corner (0,0), and the off-diagonal entry H12 = partial^2 Phi/partial x1 partial x2. Because a C^1 diffeomorphism of the square to itself must preserve sides, H12 must vanish on that side and the eigenvalues must stay bounded away from 0 and infinity.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 30 Oct 2026
- Reader holds up, with conditions
- Skeptic agrees
- Editor wrote the verdict above
- Referee major revision: 2 major objections, 5 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
Two things to know. First, the main 2D theorem holds up under reading: the smoothing-cost proxy is used at leading order rather than as a negligible error, the subdivision argument that evaluates it carries explicit error rates, and the constants are proved to exist, not fitted. Corollary 2 does prove both assertions of Conjecture 1, and the forward error is polynomial, O(n−1/180 log n). The averaging identity (Theorem 2) is a clean standalone contribution; Lemma A.1's bijection is clever and appears correct.
Second, the distinct-distribution extension is the part to be careful with. Theorem 1 is honestly conditional: on convex domains the author assumes the population optimal map is a smooth diffeomorphism up to the boundary with H = DT uniformly positive definite (b395), and for polygons with corners this is likely false in that strength for generic smooth densities — second-boundary-value Monge–Ampère regularity fails at corners and H can degenerate. So the stress-test concern lands, but as a scope fact, not an error: the paper states the assumption, tori are covered via [13], and the internal arguments all go through under the stated hypotheses. The practical consequence is that Corollary 1's strict reduction of the log coefficient is established for pairs satisfying the condition, and the square — the example readers will reach for first — is exactly where the assumption is unverified. The abstract and intro should say this louder than they do.
Smaller soft spots, in proportion. The Barthe–Bordenave equality depends on Huesmann–Goldman–Trevisan [17, Theorem 4.3], listed as "to appear" with arXiv numbering; this does not touch the core 2D chain, but the d ≥ 5 corollary is hostage to that paper. The universal constants β, β_sd, κ* are unevaluated — a limitation, not a flaw. And the appendix proofs are stated to have been completed with substantial ChatGPT assistance; the main text is clearly the author's, the disclosure is unusually honest, but a referee should actually re-derive the appendix arguments rather than skim them.
This paper is for the random-matching and empirical-optimal-transport crowd, and it will be cited there. It deserves a serious referee, and I would send it to review; the referee's time is best spent on Section 4 and Appendices C–D, plus a demand that the regularity assumption's scope be stated in the introduction.
The strongest objection
The most load-bearing premise is the regularity of the population optimal transport map in Theorem 1. The reduction chain (Theorem 6 + Lemma D.1 + Corollary 2) requires the population dual gap k to be comparable to squared distance in the induced metric and requires the induced metric to satisfy the surface assumptions of Section 3.2. On flat tori, smooth positive densities imply the needed regularity via [13], but for bounded convex domains the paper assumes, rather than derives, that T is a smooth diffeomorphism up to the boundary with H=DT uniformly positive definite (Section 5.3; sentences b54, b395). This matters because second-boundary-value Monge-Ampère regularity on polygonal domains with corners is known to fail for generic smooth positive densities: the map can lose C^1 up-to-boundary regularity, and D^2Phi can blow up or degenerate at a corner. If H degenerates, the identity k(x,z)=2{Phi(x)-Phi(z)-nabla-Phi(z) dot (x-z)} is still algebraic, but it no longer implies uniform comparability k comparable to d_H^2 up to the boundary, so Lemma D.1's replacement error O(n^{-1/4} log^{5/4}(en)) and the spectral treatment of (Omega,H) in Corollary 2 do not follow. The theorem remains logically correct as a conditional statement; the concern is scope: the distinct-distribution formula, and in particular Corollary 1's strict reduction of the logarithmic coefficient, is not established for generic smooth positive densities on polygonal domains, only for pairs whose optimal map satisfies an unverified boundary regularity condition.
I read the paper as a conditional proof: if the population optimal transport map is a smooth diffeomorphism with uniformly positive definite Jacobian, and if the long chain of reductions in Sections 2-5 and Appendices A-E is accepted, then Theorem 1 follows. I did not find an internal inconsistency in the reduction sequence; the averaging identity, the semidiscrete-to-smoothing reduction, the subdivision argument, and the distinct-to-same-distribution reduction are presented in enough detail that the main risk is not a detectable algebraic slip but the validity of the stated hypotheses in the intended setting. The reader's weakest_assumption identifies exactly this point: on flat tori the regularity is supplied by a published source, while on bounded convex domains it is assumed. My stress-test sharpens the concern: at a polygonal corner, a C^1 diffeomorphism with symmetric positive definite derivative must map the tangent cone to itself, which imposes a concrete boundary condition on H; Monge-Ampère regularity theory for such second boundary value problems on polygons does not guarantee that condition for generic smooth densities. Thus the theorem's applicability to natural distinct-distribution examples on the square is narrower than the surrounding narrative suggests. This does not refute the theorem, and the paper states the assumption explicitly, so no change of verdict is warranted. The conditional verdict is appropriate; the remaining verification concerns, including the unpublished citation [17] and the AI-assisted appendix proofs, reinforce rather than replace this assessment.
The referee's report
If the results hold, this is a major contribution to the 2D random matching problem: it determines the constant term in the expected matching cost, settling both assertions of Conjecture 1, and it provides the first sharp distinct-distribution extension with an explicit change of the logarithmic coefficient. The universal constants beta, beta_sd, and kappa* are proven to exist by a fourfold/ninefold scale-comparison argument rather than fitted to data, which is a genuine strength; the proof chain carries explicit polynomial error rates throughout (e.g., O(n^{-1/180} log n) in Theorem 1); and the paper makes falsifiable quantitative predictions, including the geometry-independent square-versus-torus difference 1/3 + log 2/pi. The transparency about hypotheses is also a strength: the conditional status of the main theorem is stated, and Example 5.2 exhibits a non-trivial mu != nu pair satisfying all the regularity hypotheses on the square. The principal caveat, developed in the major comments, concerns the scope of the regularity hypothesis for convex domains with corners; the proof is internally consistent, and I found no mathematical error in the reduction chain.
Pith's reading of the paper (a machine's summary, not the authors')
Match n random points drawn from a law $\mu$ to n random points drawn from a possibly different law $\nu$ on a compact two-dimensional surface, with squared distance as the cost. The paper tries to show that the expected minimum matching cost exceeds n times the population transport cost $W_2^2(\mu,\nu)$ by a quantity determined up to additive $o(1)$: a term $(A_H/2\pi)\log n$, a universal constant times $A_H$, a spectral correction $2h_H$, and a log-density integral, where $H=DT$ is the Jacobian of the population optimal transport map sending $\mu$ to $\nu$. If true, the result fixes the sampling bias of the empirical 2-Wasserstein distance to $o(1/n)$, proves the spectral-universality conjecture for samples from a common distribution, and shows that distinct distributions strictly reduce the logarithmic coefficient. The route runs through four reductions: an averaging identity tying bipartite matching to semidiscrete transport with random Dirichlet masses, a semidiscrete-to-smoothing relation that replaces the matching cost by the cost of transporting a heat-smoothed sample back to its own points, a subdivision argument that evaluates that smoothing cost, and a reduction that rewrites the whole excess as a same-distribution matching problem in the metric induced by the optimal transport map.
What carries the argument. The load-bearing object is the smoothing cost: the minimum cost of transporting a heat-smoothed sample $P_t\nu_n$ back to the original point measure $\nu_n$. Four mechanisms carry the argument. The averaging identity (Theorem 2) equates the expected bipartite matching cost with $n$ times the expected semidiscrete transport cost from $\mu$ to $\sum_i W_i\delta_{Y_i}$ with Dirichlet$(1,\dots,1)$ weights, for general integrable costs in every dimension. The semidiscrete-to-smoothing reduction (Theorem 3) shows the semidiscrete mean cost equals the expected smoothing cost plus the explicit spectral term $\chi_n\{2F(t)-F(2t)\}$, up to a controlled error, in every dimension when the cost has a positive quadratic part; this is what makes strong heat smoothing admissible without losing the nonlinear transport contribution. Subdivision (Theorem 4 and Lemma 4.1) evaluates the smoothing cost by cutting the domain into cells and exploiting the cancellation between a sample and its own smoothing, which share the same large-scale mass excesses, leaving only $o(1)$ error and yielding one universal constant $\kappa_\star$ shared by square and torus. The distinct-to-same-distribution reduction (Theorem 6) pulls each target back through $T$ and writes the remaining cost as the population dual gap $k(x,z)=c(x,Tz)-a(x)-b(Tz)$, whose quadratic part defines the metric $H$; replacing $k$ by squared distance in that metric changes the expected cost by only $O(n^{-1/4}\log^{5/4}(en))$, so the same-distribution formula applies with $H$ as the geometry.
Core claim. On a flat torus or a bounded convex domain in $\mathbb{R}^2$, for strictly positive smooth densities $p,q$ whose population optimal transport map $T$ is a smooth diffeomorphism with uniformly positive definite Jacobian $H=DT$, Theorem 1 asserts $$\mathbb{E}M_n(\mu,\nu)-$nW_2^{2}$(\mu,\nu)=\frac{A_H}{2\pi}\log n + A_H\$\beta$ + 2h_H + \frac{1}{2\pi}\int_\$\Omega$ \log\rho_H\, dA_H + O_{\$\Omega$,p,q}($n^{{-1/180}}$\log n),$$ with $\beta$ a universal constant shared with the uniform square and torus problems. All inputs are computed once from the two laws: $dA_H=\sqrt{\det H}\,dx$ is the area of the metric induced by the population map, $A_H$ its total area, $\rho_H=p/\sqrt{\det H}$ the sampling density in that metric, and $h_H$ the spectral finite part of the corresponding weighted Laplacian. Corollary 1 states $0<A_H\le|\Omega|$, with equality if and only if $\mu=\nu$, so distinct laws shrink the logarithmic coefficient. Specializing to $\mu=\nu$ gives mean-cost formulas that prove both assertions of Conjecture 1, including the explicit square–torus difference $1/3+(\log 2)/\pi$; semidiscrete counterparts hold with the logarithmic, spectral, and density terms halved; and the same averaging identity settles the Barthe–Bordenave conjecture for quadratic cost in dimensions $d\ge 5$.
The premise it rests on. The load-bearing premise is that the population optimal transport map $T$ is a smooth diffeomorphism, up to the boundary, with uniformly positive definite Jacobian $H=DT$; on convex domains this regularity is assumed rather than derived, and if $H$ degenerates at the boundary the identity for the population dual gap and its replacement by squared distance in the induced metric no longer carry the $o(1)$ error that Theorem 1 needs.
If the paper is right (Pith's inferences, not the authors')
The bias of $W_2^2(\hat\mu_n,\hat\nu_n)$ as an estimator of $W_2^2(\mu,\nu)$ is pinned down to $o(1/n)$, upgrading earlier $O(\log n)$ bounds to a complete first-order asymptotic.
For samples from a common distribution, both assertions of Conjecture 1 hold: geometry enters only through the spectral finite part $h_\Omega$, the remainders converge to finite constants, and unit-area comparisons are explicit, e.g. $C^n_Q-C^n_{T^2}\to 1/3+(\log 2)/\pi$.
Distinct laws strictly reduce the logarithmic coefficient from $|\Omega|/(2\pi)$ to $A_H/(2\pi)$ whenever $\mu\neq\nu$ (Corollary 1), even after subtracting the population cost $nW_2^2(\mu,\nu)$.
The universal constant $\beta$ has to be computed only once, from uniform sampling on the unit square, and then applies to every pair of distributions satisfying the regularity assumptions.
The semidiscrete mean cost obeys the same formula with $\beta^{sd}$ and with the logarithmic, spectral, and density corrections halved; for power costs $|x-y|^r$ with $r>1$, a fixed nonzero translate between the laws yields logarithmic excess cost with an explicit induced metric (Example 5.1).
Read as an effective-area statement, Theorem 1 says matching into a different law behaves like same-law matching on a deformed surface of area $A_H$; a testable consequence is that for $\mu$ close to $\nu$ the coefficient deficit $|\Omega|-A_H$ should scale linearly with $W_2^2(\mu,\nu)$ (Example 5.2 exhibits $A_H=1-\varepsilon^2/16+O(\varepsilon^4)$ alongside $W_2^2(\mu,\nu)=\varepsilon^2/(8\pi^2
The cancellation between a sample and its own heat smoothing looks like a mechanism rather than a squared-cost artefact, so the subdivision strategy plausibly transfers to other fluctuation-dominated matching problems, such as capacitated matching or matching on graphs, where logarithmic laws are expected but no formula is known.
The error $O(n^{-1/180}\log n)$ arises from balancing repair width against cell size in the subdivision step, so the true remainder for fixed smooth data is probably much smaller; a high-precision simulation of the square–torus difference would indicate how far the stated rate is from the best possible one.
The paper leaves open which leading orders occur for unsquared distance cost when the dual gap degenerates (its quadratic part vanishes along the displacement direction); a plausible next step is to regularize that degeneracy inside the same distinct-to-same-distribution reduction, which would decide whether intermediate orders between $\log n$ and $\sqrt{n\log n}$ are realized.
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
Given two independent n-point samples drawn from possibly distinct probability distributions on a compact 2-dimensional surface, we minimize the total squared distance over all perfect bipartite matchings. Under regularity assumptions, the expected matching cost minus n times the squared 2-Wasserstein distance between the sampling distributions grows logarithmically with n. We determine this difference up to an additive o(1) error, including the dependence of the logarithmic coefficient and constant term on geometry and sampling distributions, and establish the existence of a universal constant shared by these formulas. Our analysis expresses the expected matching cost in terms of the cost of transporting a heat-smoothed sample back to its original points. The reductions giving this relation hold in every dimension and for general costs under suitable assumptions, with consequences beyond the setting studied here. In two dimensions, the relation is exact up to an additive o(1) error, so determining the matching cost amounts to evaluating the smoothing cost with the same accuracy. We obtain this evaluation by subdividing the domain and exploiting cancellations between the sample and its own smoothing.
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 · 2026-10-30
The authors' main claim, as the classifier picked it out 0.68
“We determine this difference up to an additive $o(1)$ error, including the dependence of the logarithmic coefficient and constant term on geometry and sampling distributions, and establish the existence of a universal constant shared by these formulas.”
How the main claim is stated: asserted 1.00
What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.61 · human participants 0.02 · controlled comparison 0.09 · randomized assignment 0.03 · states quantitative result 0.90 · compares with prior results 0.21 · 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 |
|---|---|---|---|---|
| Given two independent $n$-point samples drawn from possibly distinct probability distributions on a compact 2-dimensional surface, we minimize the total squared distance over all perfect bipartite matchings. | background 0.58 | 0.55 | 0.04 | asserted 0.77 |
| Under regularity assumptions, the expected matching cost minus $n$ times the squared 2-Wasserstein distance between the sampling distributions grows logarithmically with $n$. | result 1.00 | 0.79 | 0.56 | asserted 0.99 |
| We determine this difference up to an additive $o(1)$ error, including the dependence of the logarithmic coefficient and constant term on geometry and sampling distributions, and establish the existence of a universal constant shared by these formulas. | result 0.80 | 0.96 | 0.60 | asserted 1.00 |
| Our analysis expresses the expected matching cost in terms of the cost of transporting a heat-smoothed sample back to its original points. | method 0.77 | 0.90 | 0.09 | asserted 1.00 |
| The reductions giving this relation hold in every dimension and for general costs under suitable assumptions, with consequences beyond the setting studied here. | interpretation 0.65 | 0.77 | 0.06 | asserted 0.98 |
| In two dimensions, the relation is exact up to an additive $o(1)$ error, so determining the matching cost amounts to evaluating the smoothing cost with the same accuracy. | interpretation 0.53 | 0.85 | 0.18 | asserted 1.00 |
| We obtain this evaluation by subdividing the domain and exploiting cancellations between the sample and its own smoothing. | method 1.00 | 0.94 | 0.07 | asserted 1.00 |
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, 2026-10-30: 1 request, 5,642 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 (1)
References (31)
- [13] D. Cordero-Erausquin. Sur le transport de mesures périodiques.Comptes Rendus de l’Académie des Sciences, Série I, Mathématique329 (1999), 199–202 on Pith
- [1] M. Ajtai, J. Komlós, and G. Tusnády. On optimal matchings.Combinatorica4 (1984), 259–264 on Pith
- [2] L. Ambrosio and F. Glaudo. Finer estimates on the2-dimensional matching problem.Journal de l’École polytechnique—Mathématiques6 (2019), 737–765 on Pith
- [3] L. Ambrosio, F. Glaudo, and D. Trevisan. On the optimal map in the2-dimensional random matching problem.Discrete and Continuous Dynamical Systems39 (2019), 7291–7308 on Pith
- [4] L. Ambrosio, M. Goldman, and D. Trevisan. On the quadratic random matching problem in two-dimensional domains.Electronic Journal of Probability27 (2022), article 54 on Pith
- [5] L. Ambrosio, F. Stra, and D. Trevisan. A PDE approach to a 2-dimensional matching problem. Probability Theory and Related Fields173 (2019), 433–477 on Pith
- [6] V. Armegioiu, M. Goldman, F. Grotto, and D. Trevisan. Exact asymptotics for the 2D Euclidean random matching problem. arXiv:2609.08717, 2026 on PithPith's reviewarXiv
- [7] F. Barthe and C. Bordenave. Combinatorial optimization over two random point sets. In Séminaire de Probabilités XLV, Lecture Notes in Mathematics 2078, Springer, 2013, pp. 483– 535 on Pith
- [8] D. Benedetto, E. Caglioti, S. Caracciolo, M. D’Achille, G. Sicuro, and A. Sportiello. Random assignment problems on 2d manifolds.Journal of Statistical Physics183 (2021), article 34 on Pith
- [9] N. Berline, E. Getzler, and M. Vergne.Heat Kernels and Dirac Operators. Springer, 1992 on Pith
- [10] S. G. Bobkov and M. Ledoux. A simple Fourier analytic proof of the AKT optimal matching theorem.The Annals of Applied Probability31(6) (2021), 2567–2584. 22 on Pith
- [11] S. Caracciolo, C. Lucibello, G. Parisi, and G. Sicuro. Scaling hypothesis for the Euclidean bipartite matching problem.Physical Review E90 (2014), 012118 on Pith
- [12] S. Caracciolo and G. Sicuro. Quadratic stochastic Euclidean bipartite matching problem. Physical Review Letters115 (2015), 230601 on Pith
- [14] D. Daners. Heat kernel estimates for operators with boundary conditions.Mathematische Nachrichten217 (2000), 13–41 on Pith
- [15] M. Goldman, M. Huesmann, and F. Otto. Almost sharp rates of convergence for the average cost and displacement in the optimal matching problem. InPartial Differential Equations: Waves, Nonlinearities and Nonlocalities, Abel Symposia 18, Springer, 2025, pp. 93–109 on Pith
- [16] M. Goldman and D. Trevisan. Convergence of asymptotic costs for random Euclidean matching problems.Probability and Mathematical Physics2 (2021), 341–362 on Pith
- [17] M. Huesmann, M. Goldman, and D. Trevisan. Asymptotics for random quadratic transportation costs.Communications on Pure and Applied Mathematics, to appear. Result numbering follows arXiv:2409.08612v2 on PithPith's reviewarXiv
- [18] J. Kitagawa, Q. Mérigot, and B. Thibert. Convergence of a Newton algorithm for semi-discrete optimal transport.Journal of the European Mathematical Society21(9) (2019), 2603–2651 on Pith
- [19] A. V. Kolesnikov. Hessian metrics,CD(K,N )-spaces, and optimal transportation of log-concave measures.Discrete and Continuous Dynamical Systems34(4) (2014), 1511–1532 on Pith
- [20] T. Manole, S. Balakrishnan, J. Niles-Weed, and L. Wasserman. Plugin estimation of smooth optimal transport maps.The Annals of Statistics52(3) (2024), 966–998 on Pith
show 11 more
- [21] T. Manole and J. Niles-Weed. Sharp convergence rates for empirical optimal transport with smooth costs.The Annals of Applied Probability34(1B) (2024), 1108–1135 on Pith
- [22] M. Murugan. Heat kernel for reflected diffusion and extension property on uniform domains. Probability Theory and Related Fields190 (2024), 543–599 on Pith
- [23] L. Saloff-Coste. Parabolic Harnack inequality for divergence form second order differential operators.Potential Analysis4 (1995), 429–467 on Pith
- [24] K. Treleaven, M. Pavone, and E. Frazzoli. Asymptotically optimal algorithms for one-to-one pickup and delivery problems with applications to transportation systems.IEEE Transactions on Automatic Control58(9) (2013), 2261–2276 on Pith
- [25] R. Peyre. Comparison betweenW2 distance and ˙H−1 norm, and localization of Wasserstein distance.ESAIM: Control, Optimisation and Calculus of Variations24 (2018), 1489–1501 on Pith
- [26] C. Villani.Optimal Transport: Old and New. Grundlehren der mathematischen Wissenschaften 338, Springer, 2009 on Pith
- [27] T.-K. L. Wong and J. Yang. Pseudo-Riemannian geometry encodes information geometry in optimal transport.Information Geometry5 (2022), 131–159. 23 on Pith
- [28] M. Dauge. Neumann and mixed problems on curvilinear polyhedra.Integral Equations and Operator Theory15 (1992), 227–261 on Pith
- [29] G. M. Lieberman.Second Order Parabolic Differential Equations. World Scientific, 1996 on Pith
- [30] M. Nursultanov, J. Rowlett, and D. A. Sher. The heat kernel on curvilinear polygonal domains in surfaces.Annales mathématiques du Québec49 (2025), 1–61 on Pith
- [31] E. M. Stein.Singular Integrals and Differentiability Properties of Functions. Princeton University Press, 1970. A The averaging identity The bijection below completes the proof of the averaging identity in Section 2. We then apply that identity to the Barthe–Bordenave conjectu... on Pith
Cited by 1 Pith paper
Reviewed papers on Pith that cite this one. The full citation record
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