pith. machine review for the scientific record. sign in

arxiv: 2605.08705 · v1 · submitted 2026-05-09 · 🧮 math.ST · math.PR· stat.ME· stat.ML· stat.TH

Recognition: no theorem link

Minimax Optimal Estimation of Transport-Growth Pairs in Unbalanced Optimal Transport

Donlapark Ponnoprat, Masaaki Imaizumi, Noboru Isobe

Pith reviewed 2026-05-12 01:09 UTC · model grok-4.3

classification 🧮 math.ST math.PRstat.MEstat.MLstat.TH
keywords unbalanced optimal transportminimax estimationtransport-growth pairMonge estimationstability reductionquadratic costKullback-Leibler penalty
0
0 comments X

The pith

Unbalanced optimal transport estimation targets transport-growth pairs with minimax optimal rates.

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

The paper argues that the appropriate target for statistical estimation in unbalanced optimal transport is not a transport map by itself but a transport-growth pair. Two estimators are developed for this pair, one based on optimal transport plans for general cases and another kernel-based for smooth densities. These estimators are shown to attain the minimax optimal error rate through a matching lower bound on the minimax risk. The proof relies on a value-based stability reduction that translates objective perturbations into separate transport and growth risks under a UOT gap condition.

Core claim

In unbalanced optimal transport with quadratic cost and Kullback-Leibler marginal penalties, the population target is a transport-growth pair, and estimators for this pair achieve the minimax optimal rate of convergence.

What carries the argument

The value-based stability reduction, which converts perturbations of the UOT objective into transport and growth risks through a UOT gap condition.

If this is right

  • The natural population target in unbalanced optimal transport is a transport-growth pair rather than a transport map alone.
  • An optimal transport plan-based estimator recovers the pair in general cases.
  • A kernel-based estimator recovers the pair when densities are smooth.
  • The estimators attain the minimax optimal rate, as confirmed by a matching lower bound.
  • The results supply a statistical foundation for Monge-type estimation in unbalanced optimal transport.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar value-based reductions might apply to unbalanced transport problems that use other divergence penalties.
  • In applications the gap condition would need to be checked or ensured before claiming the optimal rate.
  • The framework suggests studying minimax rates for transport-growth pairs under costs and penalties beyond the quadratic-KL case.

Load-bearing premise

The UOT gap condition holds for the underlying data distributions, allowing objective perturbations to separate into distinct transport and growth risks.

What would settle it

A data distribution where the UOT gap condition fails and the estimator error rate exceeds the derived lower bound would disprove the general claim.

Figures

Figures reproduced from arXiv: 2605.08705 by Donlapark Ponnoprat, Masaaki Imaizumi, Noboru Isobe.

Figure 1
Figure 1. Figure 1: (a) MSE of the four UOT estimators. Each plot shows the average over 10 seeds with one standard error. Top: The MSEs of estimating T0 and λ0 vs n. Bottom: Learning rates for T0 and λ0 vs. d. (b) Top: Incomplete 3D shapes. Middle: Complete 3D shapes predicted by the plan-based 1NN estimator. Bottom: Complete 3D shapes predicted by the plan-based kernel estimator. 5. Experiments 5.1. Simulation study. We sam… view at source ↗
read the original abstract

Unbalanced optimal transport (UOT) extends classical optimal transport to measures with different total masses, but statistical guarantees for Monge-type estimation remain limited. We study unbalanced transport with quadratic cost and Kullback-Leibler marginal penalties and argue that the natural population target is not a map alone, but a transport-growth pair. Consequently, we develop two estimators for the transport-growth pairs under several setups: an optimal transport plan-based estimator for a general case, and a kernel-based estimator for a case with smooth densities. We also show that an error of the estimator achieves the minimax optimal rate by deriving a matching lower bound of the minimax risk. Our main technical contribution is a value-based stability reduction that converts perturbations of the UOT objective into transport and growth risks through a UOT gap condition. These results provide a statistical foundation for Monge-type estimation in unbalanced optimal transport.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript studies minimax estimation of transport-growth pairs (rather than maps alone) in unbalanced optimal transport with quadratic cost and KL marginal penalties. It introduces an OT-plan-based estimator for the general case and a kernel-based estimator for smooth densities, then claims that the estimation error attains the minimax optimal rate by deriving a matching lower bound. The central technical device is a value-based stability reduction that converts perturbations of the UOT objective into separate transport and growth risks, provided a UOT gap condition holds.

Significance. If the UOT gap condition is verified to hold under the paper's assumptions and is shown to be non-vacuous, the matching upper and lower bounds would supply the first rigorous statistical guarantees for Monge-type estimation in UOT, extending classical OT results to settings with mass creation/destruction. The stability-reduction technique itself could be reusable beyond this model.

major comments (2)
  1. [Abstract / value-based stability reduction] Abstract and the section presenting the value-based stability reduction: the UOT gap condition is invoked to convert objective perturbations into separate transport and growth risks and thereby obtain the upper bound, yet the manuscript provides neither its precise mathematical statement nor a proof (or even a verification) that the condition holds for the data distributions under consideration or is non-vacuous. Because the upper bound (and therefore the claimed matching minimax rate) depends on this condition while the lower bound is presented as unconditional, the central optimality claim is at present conditional rather than unconditional.
  2. [Main minimax result] The paragraph stating the main result on minimax optimality: without an explicit derivation showing how the stability reduction produces the claimed rate once the gap condition is imposed, it is impossible to confirm that the upper bound indeed matches the lower bound or to assess the dependence on the gap parameter.
minor comments (1)
  1. [Abstract] The abstract is dense; a single sentence clarifying the achieved rate (e.g., in terms of sample size n and dimension) would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive report. The comments highlight important points about the presentation of the UOT gap condition and the explicit linkage in the minimax result. We address each major comment below and will revise the manuscript to incorporate the requested clarifications and derivations.

read point-by-point responses

  1. Referee: [Abstract / value-based stability reduction] Abstract and the section presenting the value-based stability reduction: the UOT gap condition is invoked to convert objective perturbations into separate transport and growth risks and thereby obtain the upper bound, yet the manuscript provides neither its precise mathematical statement nor a proof (or even a verification) that the condition holds for the data distributions under consideration or is non-vacuous. Because the upper bound (and therefore the claimed matching minimax rate) depends on this condition while the lower bound is presented as unconditional, the central optimality claim is at present conditional rather than unconditional.

    Authors: We agree that the UOT gap condition requires a precise statement and verification for the claims to be fully rigorous. The value-based stability reduction relies on this condition to decouple the transport and growth components of the risk from perturbations of the UOT objective. In the revised manuscript we will add the exact mathematical definition of the UOT gap condition (a quantitative lower bound on the difference between the UOT value and its linearized approximation) in the dedicated section. We will also include a short verification that the condition holds under the paper's standing assumptions (quadratic cost, KL penalties, and the moment or density bounds used for the estimators), together with a simple example showing that the gap is strictly positive for non-degenerate measures. We will further revise the statement of the main upper bound to make explicit that it holds conditionally on the gap condition, while the lower bound remains unconditional; this will accurately reflect the scope of the optimality result without overstating it. revision: yes

  2. Referee: [Main minimax result] The paragraph stating the main result on minimax optimality: without an explicit derivation showing how the stability reduction produces the claimed rate once the gap condition is imposed, it is impossible to confirm that the upper bound indeed matches the lower bound or to assess the dependence on the gap parameter.

    Authors: We acknowledge that the current paragraph is too terse. In the revision we will expand the main minimax theorem paragraph (and the surrounding discussion) with an explicit outline of the argument: first apply the value-based stability reduction to bound the sum of transport and growth risks by the UOT objective perturbation plus a term controlled by the gap parameter; then invoke the existing concentration or approximation bounds on the estimator to control the objective perturbation; finally combine with the matching lower bound to obtain the rate. The expanded text will also display the explicit dependence on the gap parameter (typically a multiplicative factor of the form 1/gap or gap^{-1/2} depending on the estimator). This step-by-step derivation will make the matching of upper and lower bounds transparent and allow readers to see the precise role of the gap. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on independent lower bound and explicit assumption

full rationale

The paper presents an upper bound on estimator error via a value-based stability reduction that invokes the UOT gap condition as a prerequisite, paired with a separately derived matching lower bound for the minimax rate. No equations or steps reduce the claimed rate to a quantity defined by the estimator itself, nor do they rename fitted inputs as predictions. The UOT gap condition functions as a stated assumption for the upper bound rather than a self-definitional construct or self-citation load-bearing premise. The lower bound is described as independent, and no self-citation chains or ansatz smuggling are indicated in the abstract or context. This keeps the central minimax optimality claim self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper relies on standard quadratic cost and KL divergence penalties from the unbalanced OT literature; no new free parameters or invented entities are introduced in the abstract. The UOT gap condition is presented as a technical assumption rather than a derived property.

axioms (1)
  • domain assumption Quadratic cost and Kullback-Leibler marginal penalties define a well-posed unbalanced OT problem whose population target is a transport-growth pair.
    Invoked in the first paragraph to justify studying pairs rather than maps alone.

pith-pipeline@v0.9.0 · 5467 in / 1296 out tokens · 47424 ms · 2026-05-12T01:09:37.281698+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 1 internal anchor

  1. [1]

    The MOSEK Python Fusion API manual

    MOSEK ApS. The MOSEK Python Fusion API manual. Version 11.0. , 2025

    work page 2025

  2. [2]

    The square root normal field distance and unbalanced optimal transport

    Martin Bauer, Emmanuel Hartman, and Eric Klassen. The square root normal field distance and unbalanced optimal transport. Applied Mathematics & Optimization , 85(3), may 2022

    work page 2022

  3. [3]

    Stability bounds for smooth optimal transport maps and their statistical implications

    Sivaraman Balakrishnan and Tudor Manole. Stability bounds for smooth optimal transport maps and their statistical implications. arXiv preprint arXiv:2502.12326 , 2025

    work page arXiv 2025

  4. [4]

    Scalable wasserstein gradient flow for generative modeling through unbalanced optimal transport

    Jaemoo Choi, Jaewoong Choi, and Myungjoo Kang. Scalable wasserstein gradient flow for generative modeling through unbalanced optimal transport. In Proceedings of the 41st International Conference on Machine Learning , ICML'24. JMLR.org, 2024

    work page 2024

  5. [5]

    Sur le transport de mesures p\'eriodiques

    Dario Cordero-Erausquin. Sur le transport de mesures p\'eriodiques. Comptes Rendus de l'Acad\'emie des Sciences - Series I - Mathematics , 329(3):199--202, 1999

    work page 1999

  6. [6]

    ShapeNet: An Information-Rich 3D Model Repository

    Angel X. Chang, Thomas A. Funkhouser, Leonidas J. Guibas, Pat Hanrahan, Qi - Xing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. Shapenet: An information-rich 3d model repository. CoRR , abs/1512.03012, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  7. [7]

    Unbalanced optimal transport through non-negative penalized linear regression

    Laetitia Chapel, R\' e mi Flamary, Haoran Wu, C\' e dric F\' e votte, and Gilles Gasso. Unbalanced optimal transport through non-negative penalized linear regression. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan, editors, Advances in Neural Information Processing Systems , volume 34, pages 23270--23282. Curran Associates, ...

    work page 2021

  8. [8]

    Youssuff Hussaini, Alfio Quarteroni, and Thomas A

    Claudio Canuto, M. Youssuff Hussaini, Alfio Quarteroni, and Thomas A. Zang. Spectral Methods: Fundamentals in Single Domains . Scientific Computation. Springer Berlin Heidelberg, 1 edition, 2006

    work page 2006

  9. [9]

    Unbalanced optimal transport: Dynamic and kantorovich formulations

    L \'e na \"i c Chizat, Gabriel Peyr \'e , Bernhard Schmitzer, and Fran c ois-Xavier Vialard. Unbalanced optimal transport: Dynamic and kantorovich formulations. Journal of Functional Analysis , 274(11):3090--3123, 2018

    work page 2018

  10. [10]

    CVXPY : A P ython-embedded modeling language for convex optimization

    Steven Diamond and Stephen Boyd. CVXPY : A P ython-embedded modeling language for convex optimization. Journal of Machine Learning Research , 17(83):1--5, 2016

    work page 2016

  11. [11]

    A tumor growth model of hele-shaw type as a gradient flow

    Di Marino, Simone and Chizat, Lénaïc . A tumor growth model of hele-shaw type as a gradient flow. ESAIM: COCV , 26:103, 2020

    work page 2020

  12. [12]

    Rates of estimation of optimal transport maps using plug-in estimators via barycentric projections

    Nabarun Deb, Promit Ghosal, and Bodhisattva Sen. Rates of estimation of optimal transport maps using plug-in estimators via barycentric projections. In Advances in Neural Information Processing Systems , volume 34, pages 29736--29753, 2021

    work page 2021

  13. [13]

    Optimal transport map estimation in general function spaces

    Vincent Divol, Jonathan Niles-Weed, and Aram-Alexandre Pooladian. Optimal transport map estimation in general function spaces. The Annals of Statistics , 53(3):963--988, 2025

    work page 2025

  14. [14]

    Orthogonal series density estimation

    Sam Efromovich. Orthogonal series density estimation. WIREs Computational Statistics , 2(4):467--476, 2010

    work page 2010

  15. [15]

    Alaya, Aur \'e lie Boisbunon, Stanislas Chambon, Laetitia Chapel, Adrien Corenflos, Kilian Fatras, Nemo Fournier, L \'e o Gautheron, Nathalie T.H

    R \'e mi Flamary, Nicolas Courty, Alexandre Gramfort, Mokhtar Z. Alaya, Aur \'e lie Boisbunon, Stanislas Chambon, Laetitia Chapel, Adrien Corenflos, Kilian Fatras, Nemo Fournier, L \'e o Gautheron, Nathalie T.H. Gayraud, Hicham Janati, Alain Rakotomamonjy, Ievgen Redko, Antoine Rolet, Antony Schutz, Vivien Seguy, Danica J. Sutherland, Romain Tavenard, Ale...

    work page 2021

  16. [16]

    On the rate of convergence in wasserstein distance of the empirical measure

    Nicolas Fournier and Arnaud Guillin. On the rate of convergence in wasserstein distance of the empirical measure. Probability Theory and Related Fields , 162(3--4):707--738, 2015

    work page 2015

  17. [17]

    Light unbalanced optimal transport

    Milena Gazdieva, Arip Asadulaev, Evgeny Burnaev, and Alexander Korotin. Light unbalanced optimal transport. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, editors, Advances in Neural Information Processing Systems , volume 37, pages 93907--93938. Curran Associates, Inc., 2024

    work page 2024

  18. [18]

    Regularity theory and geometry of unbalanced optimal transport

    Thomas Gallou \"e t, Roberta Ghezzi, and Fran c ois-Xavier Vialard. Regularity theory and geometry of unbalanced optimal transport. Journal of Functional Analysis , page 111042, 2025

    work page 2025

  19. [19]

    Mathematical Foundations of Infinite-Dimensional Statistical Models

    Evarist Gin \'e and Richard Nickl. Mathematical Foundations of Infinite-Dimensional Statistical Models . Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 2016

    work page 2016

  20. [20]

    Unbalanced kantorovich--rubinstein distance, plan, and barycenter on finite spaces: A statistical perspective

    Shayan Hundrieser, Florian Heinemann, Marcel Klatt, Marina Struleva, and Axel Munk. Unbalanced kantorovich--rubinstein distance, plan, and barycenter on finite spaces: A statistical perspective. Journal of Machine Learning Research , 26:37:1--37:70, 2025

    work page 2025

  21. [21]

    Minimax estimation of smooth optimal transport maps

    Jan-Christian H \"u tter and Philippe Rigollet. Minimax estimation of smooth optimal transport maps. The Annals of Statistics , 49(2):1166, 2021

    work page 2021

  22. [22]

    Optimal entropy-transport problems and a new hellinger--kantorovich distance between positive measures

    Matthias Liero, Alexander Mielke, and Giuseppe Savar \'e . Optimal entropy-transport problems and a new hellinger--kantorovich distance between positive measures. Inventiones mathematicae , 211(3):969--1117, 2018

    work page 2018

  23. [23]

    Liu and Jorge Nocedal

    Dong C. Liu and Jorge Nocedal. On the limited memory bfgs method for large scale optimization. Mathematical Programming , 45(1-3):503--528, August 1989

    work page 1989

  24. [24]

    Plugin estimation of smooth optimal transport maps

    Tudor Manole, Sivaraman Balakrishnan, Jonathan Niles-Weed, and Larry Wasserman. Plugin estimation of smooth optimal transport maps. The Annals of Statistics , 52(3):966--998, 2024

    work page 2024

  25. [25]

    Minimax estimation of discontinuous optimal transport maps: The semi-discrete case

    Aram-Alexandre Pooladian, Vincent Divol, and Jonathan Niles-Weed. Minimax estimation of discontinuous optimal transport maps: The semi-discrete case. In Proceedings of the 40th International Conference on Machine Learning , volume 202 of Proceedings of Machine Learning Research , pages 28128--28150, 2023

    work page 2023

  26. [26]

    Minimax rates of estimation for optimal transport map between infinite-dimensional spaces

    Donlapark Ponnoprat and Masaaki Imaizumi. Minimax rates of estimation for optimal transport map between infinite-dimensional spaces. arXiv preprint arXiv:2505.13570 , 2025

    work page arXiv 2025

  27. [27]

    On unbalanced optimal transport: An analysis of sinkhorn algorithm

    Khiem Pham, Khang Le, Nhat Ho, Tung Pham, and Hung Bui. On unbalanced optimal transport: An analysis of sinkhorn algorithm. In Proceedings of the 37th International Conference on Machine Learning , volume 119 of Proceedings of Machine Learning Research , pages 7673--7682, 2020

    work page 2020

  28. [28]

    arXiv preprint arXiv:2109.12004 , year =

    Aram-Alexandre Pooladian and Jonathan Niles-Weed. Entropic estimation of optimal transport maps. CoRR , abs/2109.12004, 2021

    work page arXiv 2021

  29. [29]

    Sharp convergence rates of empirical unbalanced optimal transport for spatio-temporal point processes

    Marina Struleva, Shayan Hundrieser, Dominic Schuhmacher, and Axel Munk. Sharp convergence rates of empirical unbalanced optimal transport for spatio-temporal point processes. arXiv preprint arXiv:2509.04225 , 2025

    work page arXiv 2025

  30. [30]

    Chapter 12 - unbalanced optimal transport, from theory to numerics

    Thibault S \'e journ \'e , Gabriel Peyr \'e , and Fran c ois-Xavier Vialard. Chapter 12 - unbalanced optimal transport, from theory to numerics. In Handbook of Numerical Analysis , volume 24, pages 407--471. Elsevier, 2023

    work page 2023

  31. [31]

    Reconstructing growth and dynamic trajectories from single-cell transcriptomics data

    Yutong Sha, Yuchi Qiu, Peijie Zhou, and Qing Nie. Reconstructing growth and dynamic trajectories from single-cell transcriptomics data. Nature Machine Intelligence , 6(1):25--39, November 2023

    work page 2023

  32. [32]

    A relaxation viewpoint to unbalanced optimal transport: Duality, optimality and monge formulation

    Giuseppe Savar \'e and Giacomo Enrico Sodini. A relaxation viewpoint to unbalanced optimal transport: Duality, optimality and monge formulation. Journal de Math \'e matiques Pures et Appliqu \'e es , 188:114--178, 2024

    work page 2024

  33. [33]

    Geoffrey Schiebinger, Jian Shu, Marcin Tabaka, Brian Cleary, Vidya Subramanian, Aryeh Solomon, Joshua Gould, Siyan Liu, Stacie Lin, Peter Berube, Lia Lee, Jenny Chen, Justin Brumbaugh, Philippe Rigollet, Konrad Hochedlinger, Rudolf Jaenisch, Aviv Regev, and Eric S. Lander. Optimal-transport analysis of single-cell gene expression identifies developmental ...

    work page 2019

  34. [34]

    The discrete cosine transform

    Gilbert Strang. The discrete cosine transform. SIAM Review , 41(1):135--147, 1999

    work page 1999

  35. [35]

    Faster unbalanced optimal transport: Translation invariant sinkhorn and 1-d frank-wolfe

    Thibault S \'e journ \'e , Fran c ois-Xavier Vialard, and Gabriel Peyr \'e . Faster unbalanced optimal transport: Translation invariant sinkhorn and 1-d frank-wolfe. In International Conference on Artificial Intelligence and Statistics , pages 4995--5021. PMLR, 2022

    work page 2022

  36. [36]

    Tsybakov

    Alexandre B. Tsybakov. Introduction to Nonparametric Estimation . Springer Series in Statistics. Springer, 2009

    work page 2009

  37. [37]

    Stability and upper bounds for statistical estimation of unbalanced transport potentials

    Adrien Vacher and Fran c ois-Xavier Vialard. Stability and upper bounds for statistical estimation of unbalanced transport potentials. arXiv preprint arXiv:2203.09143 , 2022

    work page arXiv 2022

  38. [38]

    Semi-dual unbalanced quadratic optimal transport: fast statistical rates and convergent algorithm

    Adrien Vacher and Fran c ois-Xavier Vialard. Semi-dual unbalanced quadratic optimal transport: fast statistical rates and convergent algorithm. In International Conference on Machine Learning , pages 34734--34758. PMLR, 2023

    work page 2023

  39. [39]

    Sharp asymptotic and finite-sample rates of convergence of empirical measures in wasserstein distance

    Jonathan Weed and Francis Bach. Sharp asymptotic and finite-sample rates of convergence of empirical measures in wasserstein distance. Bernoulli , 25(4A):2620--2648, 2019

    work page 2019

  40. [40]

    Pcn: Point completion network

    Wentao Yuan, Tejas Khot, David Held, Christoph Mertz, and Martial Hebert. Pcn: Point completion network. In 3D Vision (3DV), 2018 International Conference on , 2018

    work page 2018

  41. [41]

    Yang and Caroline Uhler

    Karren D. Yang and Caroline Uhler. Scalable unbalanced optimal transport using generative adversarial networks. In International Conference on Learning Representations , 2019

    work page 2019

  42. [42]

    Learning Gaussian mixtures using the Wasserstein–Fisher–Rao gradient flow

    Yuling Yan, Kaizheng Wang, and Philippe Rigollet. Learning Gaussian mixtures using the Wasserstein–Fisher–Rao gradient flow . The Annals of Statistics , 52(4):1774 -- 1795, 2024

    work page 2024