pith. machine review for the scientific record. sign in

arxiv: 2605.03300 · v1 · submitted 2026-05-05 · 🧮 math.ST · stat.TH

Recognition: unknown

Smoothed estimation of Wasserstein barycenters

Changbo Zhu, Pengtao Li, Xiaohui Chen

Pith reviewed 2026-05-07 13:10 UTC · model grok-4.3

classification 🧮 math.ST stat.TH
keywords Wasserstein barycentersnonparametric estimationconvergence ratesSobolev smoothnesssemi-dual formulationdensity estimationoptimal transportstatistical rates
0
0 comments X

The pith

Smoothness allows nonparametric estimators of Wasserstein barycenters to achieve rates that improve with the degree of smoothness rather than suffering the full curse of dimensionality.

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

The paper aims to show that a smoothness-aware estimator, built from the semi-dual formulation and its Sobolev geometry, can recover both the barycenter functional and the barycenter itself at rates governed by smoothness. If this holds, the severe dimension-dependent slowdown seen in plain empirical barycenters can be avoided for sufficiently smooth data. The method works by first estimating densities nonparametrically and then optimizing inside the Sobolev structure induced by the dual problem. A reader would care because Wasserstein barycenters appear in distribution averaging tasks across statistics and machine learning, where high-dimensional data is common.

Core claim

Motivated by the semi-dual formulation of the barycenter problem and its associated Sobolev optimization geometry, we develop a smoothness-aware approach that combines density estimation with this geometric structure to estimate the population barycenter. We establish nonparametric convergence rates for estimating both the barycenter functional and its minimizer, demonstrating how smoothness can substantially improve statistical performance over existing empirical methods that exhibit a severe curse of dimensionality.

What carries the argument

The semi-dual formulation of the barycenter problem together with its Sobolev optimization geometry, which is used to guide the density estimation step.

If this is right

  • Convergence rates for the barycenter functional depend on the Sobolev smoothness index instead of dimension alone.
  • The barycenter minimizer itself can be recovered nonparametrically at smoothness-improved rates.
  • The estimator requires no extra restrictions on the support or tail decay of the input densities.
  • Exact Wasserstein barycenters become statistically feasible in higher dimensions once smoothness is present.

Where Pith is reading between the lines

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

  • The same semi-dual Sobolev device could be tested on related optimal-transport functionals such as entropic regularized barycenters.
  • In applications where data are approximately smooth, one could first verify the Sobolev index before trusting the faster rates.
  • Adaptive procedures that estimate the smoothness parameter from samples would extend the method to cases where the degree of smoothness is unknown.

Load-bearing premise

The population barycenter admits sufficient Sobolev smoothness and the densities involved allow consistent nonparametric estimation without additional assumptions on support or tail behavior.

What would settle it

Numerical experiments in which the true barycenter is known to be increasingly smooth but the observed error rates stay fixed at the non-smooth empirical rate would disprove the claimed improvement.

read the original abstract

This paper studies the statistical estimation of exact Wasserstein barycenters. Existing non-asymptotic results for empirical barycenters exhibit a severe curse of dimensionality. Motivated by the semi-dual formulation of the barycenter problem and its associated Sobolev optimization geometry, we develop a smoothness-aware approach that combines density estimation with Sobolev geometric structure to estimate the population barycenter. We establish nonparametric convergence rates for estimating both the barycenter functional and its minimizer, demonstrating how smoothness can substantially improve statistical performance.

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

1 major / 2 minor

Summary. The manuscript develops a smoothness-aware estimator for Wasserstein barycenters via the semi-dual formulation and its Sobolev optimization geometry. It combines nonparametric density estimation with this geometric structure to derive convergence rates for both the barycenter functional and the associated minimizer, claiming that higher smoothness yields substantially better statistical performance than empirical methods subject to the curse of dimensionality.

Significance. If the rates are rigorously established under appropriate conditions, the work provides a concrete way to leverage smoothness to improve upon the dimensional curse in Wasserstein barycenter estimation. The semi-dual Sobolev approach is a natural and potentially powerful idea that could extend to other optimal transport functionals.

major comments (1)
  1. [Assumptions (likely §2 or §3)] The central nonparametric rates rest on the population barycenter admitting sufficient Sobolev smoothness and the densities permitting consistent estimation. However, no explicit moment bounds or support restrictions appear to be stated. Wasserstein geometry is sensitive to tails, so density estimation errors at infinity can produce unbounded discrepancies in the dual potentials or the barycenter itself; without such conditions the claimed smoothness improvement may fail to materialize. This assumption is load-bearing for the main theorems.
minor comments (2)
  1. [Abstract] The abstract states that rates are established but gives no explicit form or dependence on the smoothness index; a one-sentence summary of the main rate (e.g., in terms of Sobolev order s and dimension d) would improve readability.
  2. [Section 2 (semi-dual formulation)] Notation for the semi-dual potentials and the Sobolev norm should be introduced with a short reminder of the relevant embedding or duality result to aid readers unfamiliar with the geometric setup.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and the constructive major comment. We agree that explicit conditions on moments and support are necessary for the stability of Wasserstein distances and the semi-dual formulation, and we will incorporate them into the revised manuscript.

read point-by-point responses

  1. Referee: The central nonparametric rates rest on the population barycenter admitting sufficient Sobolev smoothness and the densities permitting consistent estimation. However, no explicit moment bounds or support restrictions appear to be stated. Wasserstein geometry is sensitive to tails, so density estimation errors at infinity can produce unbounded discrepancies in the dual potentials or the barycenter itself; without such conditions the claimed smoothness improvement may fail to materialize. This assumption is load-bearing for the main theorems.

    Authors: We agree that the current statement of assumptions is incomplete on this point. In the revised version we will add the following explicit conditions (new Assumption 2.3): the measures μ_i are supported on a common compact set K ⊂ ℝ^d with diameter at most R < ∞, or alternatively possess uniform finite moments of order 2+δ for some δ>0. Under these conditions the dual potentials remain uniformly bounded (by standard OT stability results), density estimation errors remain controlled in the relevant Sobolev norms, and the convergence rates for both the barycenter functional and the minimizer continue to hold with constants that depend explicitly on R or the moment bound. We will also add a short remark explaining why the tail control is needed for the semi-dual Sobolev geometry to be well-behaved. These additions do not alter the main theorems but make the load-bearing assumptions fully explicit. revision: yes

Circularity Check

0 steps flagged

No circularity: rates derived from external Sobolev theory and density estimation

full rationale

The paper combines density estimation with Sobolev geometric structure in the semi-dual formulation to obtain nonparametric convergence rates for the barycenter functional and minimizer. No load-bearing step reduces by construction to fitted inputs, self-definitions, or unverified self-citations; the derivation relies on standard external results in nonparametric statistics and Sobolev spaces. The central claims remain independent of the target rates.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only abstract available; no explicit free parameters, axioms, or invented entities listed. Typical assumptions for such work (Sobolev regularity, density estimation consistency) are implicit but unstated here.

pith-pipeline@v0.9.0 · 5366 in / 933 out tokens · 50810 ms · 2026-05-07T13:10:31.400126+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

52 extracted references · 7 canonical work pages

  1. [1]

    2026 , booktitle =

    Kim, Kaheon and Zhou, Bohan and Zhu, Changbo and Chen, Xiaohui , title =. 2026 , booktitle =

    2026

  2. [2]

    , isbn =

    Koltchinskii, V. , isbn =. Oracle Inequalities in Empirical Risk Minimization and Sparse Recovery Problems:. 2011 , bdsk-url-1 =

    2011

  3. [3]

    Probability theory and related fields , volume=

    Convergence rates for empirical barycenters in metric spaces: curvature, convexity and extendable geodesics , author=. Probability theory and related fields , volume=. 2020 , publisher=

    2020

  4. [4]

    Annales de l'Institut Henri Poincaré, Probabilités et Statistiques , number =

    Jérémie Bigot and Raúl Gouet and Thierry Klein and Alfredo López , title =. Annales de l'Institut Henri Poincaré, Probabilités et Statistiques , number =. 2017 , doi =

    2017

  5. [5]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =

    Zhu, Changbo and Müller, Hans-Georg , title = ". Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =

  6. [6]

    Journal of the American Statistical Association , volume=

    Wasserstein regression , author=. Journal of the American Statistical Association , volume=. 2023 , publisher=

    2023

  7. [7]

    2008 , publisher=

    Gradient flows: in metric spaces and in the space of probability measures , author=. 2008 , publisher=

    2008

  8. [8]

    2025 , booktitle =

    Kim, Kaheon and Yao, Rentian and Zhu, Changbo and Chen, Xiaohui , month = jul, title =. 2025 , booktitle =

    2025

  9. [9]

    Biometrika , pages =

    Zhu, Changbo and Müller, Hans-Georg , title =. Biometrika , pages =. 2025 , month =

    2025

  10. [10]

    Mathematical Methods of Operations Research , volume=

    Discrete Wasserstein barycenters: Optimal transport for discrete data , author=. Mathematical Methods of Operations Research , volume=. 2016 , publisher=

    2016

  11. [11]

    International Conference on Artificial Intelligence and Statistics , pages=

    Regularity as regularization: Smooth and strongly convex brenier potentials in optimal transport , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2020 , organization=

    2020

  12. [12]

    Annals of mathematics , volume=

    Boundary regularity of maps with convex potentials--II , author=. Annals of mathematics , volume=. 1996 , publisher=

    1996

  13. [13]

    2008 , publisher=

    Optimal transport: old and new , author=. 2008 , publisher=

    2008

  14. [14]

    Nonparametric Density Estimation &amp; Convergence Rates for GANs under Besov IPM Losses , url =

    Uppal, Ananya and Singh, Shashank and Poczos, Barnabas , booktitle =. Nonparametric Density Estimation &amp; Convergence Rates for GANs under Besov IPM Losses , url =

  15. [15]

    Journal of Machine Learning Research , year =

    Tengyuan Liang , title =. Journal of Machine Learning Research , year =

  16. [16]

    Nonparametric density estimation under adversarial losses , volume =

    Shashank Singh and Ananya Uppal and Boyue Li and Chun-Liang Li and Manzil Zaheer and Barnabás Póczos , booktitle =. Nonparametric density estimation under adversarial losses , volume =

  17. [17]

    Probability Theory and Related Fields , volume=

    Quantitative stability of barycenters in the Wasserstein space , author=. Probability Theory and Related Fields , volume=. 2024 , publisher=

    2024

  18. [18]

    Journal of the European Mathematical Society , volume=

    Fast convergence of empirical barycenters in Alexandrov spaces and the Wasserstein space , author=. Journal of the European Mathematical Society , volume=

  19. [19]

    2015 , publisher=

    Optimal transport for applied mathematicians , author=. 2015 , publisher=

    2015

  20. [20]

    arXiv preprint arXiv:2502.02726 , year=

    Sample complexity and weak limits of nonsmooth multimarginal Schrödinger system with application to optimal transport barycenter , author=. arXiv preprint arXiv:2502.02726 , year=

    work page arXiv

  21. [21]

    Chizat , author=

    Doubly Regularized Entropic Wasserstein Barycenter: L. Chizat , author=. Foundations of Computational Mathematics , pages=. 2025 , publisher=

    2025

  22. [22]

    SIAM Journal on Mathematical Analysis , volume=

    Penalization of barycenters in the Wasserstein space , author=. SIAM Journal on Mathematical Analysis , volume=. 2019 , publisher=

    2019

  23. [23]

    The Annals of Statistics , volume=

    Minimax estimation of smooth densities in Wasserstein distance , author=. The Annals of Statistics , volume=. 2022 , publisher=

    2022

  24. [24]

    ESAIM: Control, Optimisation and Calculus of Variations , volume=

    Comparison between W2 distance and H-1 norm, and localization of wasserstein distance , author=. ESAIM: Control, Optimisation and Calculus of Variations , volume=. 2018 , publisher=

    2018

  25. [25]

    2012 , publisher=

    Wavelets, approximation, and statistical applications , author=. 2012 , publisher=

    2012

  26. [26]

    , journal=

    Srivastava, Sanvesh and Li, Cheng and Dunson, David B. , journal=. Scalable

  27. [27]

    Yubo Zhuang and Xiaohui Chen and Yun Yang , booktitle =

  28. [28]

    Scale Space and Variational Methods in Computer Vision: Third International Conference (SSVM-2011) , pages=

    Wasserstein barycenter and its application to texture mixing , author=. Scale Space and Variational Methods in Computer Vision: Third International Conference (SSVM-2011) , pages=

    2011

  29. [29]

    ACM Transactions on Graphics , author =

    Solomon, Justin and de Goes, Fernando and Peyr\'. Convolutional. ACM Transactions on Graphs , keywords =. 2015 , bdsk-url-1 =. doi:10.1145/2766963 , issn =

    work page doi:10.1145/2766963 2015

  30. [30]

    SIAM Journal on Mathematical Analysis , volume =

    Carlier, Guillaume and Eichinger, Katharina and Kroshnin, Alexey , title =. SIAM Journal on Mathematical Analysis , volume =. 2021 , doi =. https://doi.org/10.1137/20M1387262 , abstract =

    work page doi:10.1137/20m1387262 2021

  31. [31]

    Machine Learning and Knowledge Discovery in Databases , year=

    Karimi, Hamed and Nutini, Julie and Schmidt, Mark , editor=. Machine Learning and Knowledge Discovery in Databases , year=

  32. [32]

    The Annals of Statistics , volume=

    On the sample complexity of entropic optimal transport , author=. The Annals of Statistics , volume=. 2025 , publisher=

    2025

  33. [33]

    The Annals of Statistics , volume=

    Improved learning theory for kernel distribution regression with two-stage sampling , author=. The Annals of Statistics , volume=. 2025 , publisher=

    2025

  34. [34]

    Springer Series in Statistics

    Introduction to Nonparametric Estimation. Springer Series in Statistics. Springer, New York , author=

  35. [35]

    The Annals of Statistics , volume=

    Minimax estimation of smooth optimal transport maps , author=. The Annals of Statistics , volume=

  36. [36]

    International Conference on Artificial Intelligence and Statistics , pages=

    Nearly tight convergence bounds for semi-discrete entropic optimal transport , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2022 , organization=

    2022

  37. [37]

    arXiv preprint arXiv:2407.18163 , volume=

    Statistical optimal transport , author=. arXiv preprint arXiv:2407.18163 , volume=. 2024 , publisher=

    work page arXiv 2024

  38. [38]

    Publications math

    Partial regularity for optimal transport maps , author=. Publications math

  39. [39]

    2017 , issn =

    Young-Heon Kim and Brendan Pass , keywords =. Advances in Mathematics , volume =. 2017 , issn =. doi:https://doi.org/10.1016/j.aim.2016.11.026 , url =

    work page doi:10.1016/j.aim.2016.11.026 2017

  40. [40]

    Applied Mathematics Letters , volume=

    Convexity of the support of the displacement interpolation: Counterexamples , author=. Applied Mathematics Letters , volume=. 2016 , publisher=

    2016

  41. [41]

    Conference on Learning Theory , pages=

    Gradient descent algorithms for Bures-Wasserstein barycenters , author=. Conference on Learning Theory , pages=. 2020 , organization=

    2020

  42. [42]

    Barycenters in the

    Agueh, Martial and Carlier, Guillaume , date-added =. Barycenters in the. SIAM Journal on Mathematical Analysis , number =

  43. [43]

    International Conference on Information Processing in Medical Imaging , pages=

    Fast optimal transport averaging of neuroimaging data , author=. International Conference on Information Processing in Medical Imaging , pages=. 2015 , organization=

    2015

  44. [44]

    arXiv preprint arXiv:2505.21274 , year=

    Sample complexity of optimal transport barycenters with discrete support , author=. arXiv preprint arXiv:2505.21274 , year=

    work page arXiv

  45. [45]

    Journal of Scientific Computing , volume=

    Efficient and exact multimarginal optimal transport with pairwise costs , author=. Journal of Scientific Computing , volume=. 2024 , publisher=

    2024

  46. [46]

    Distribution’s template estimate with Wasserstein metrics , author=

  47. [47]

    SIAM Journal on Scientific Computing , volume=

    Geodesic PCA versus log-PCA of histograms in the Wasserstein space , author=. SIAM Journal on Scientific Computing , volume=. 2018 , publisher=

    2018

  48. [48]

    Bernoulli , volume=

    Wide consensus aggregation in the Wasserstein space. Bernoulli , volume=. 2018 , publisher=. doi:10.3150/17-BEJ957 , url=

    work page doi:10.3150/17-bej957 2018

  49. [49]

    Advances in Neural Information Processing Systems , volume=

    Generalizing point embeddings using the wasserstein space of elliptical distributions , author=. Advances in Neural Information Processing Systems , volume=

  50. [50]

    Mathematische Nachrichten , volume=

    On a formula for the L2 Wasserstein metric between measures on Euclidean and Hilbert spaces , author=. Mathematische Nachrichten , volume=. 1990 , publisher=

    1990

  51. [51]

    SIAM Journal on Mathematics of Data Science , volume=

    Wasserstein barycenters are NP-hard to compute , author=. SIAM Journal on Mathematics of Data Science , volume=. 2022 , publisher=

    2022

  52. [52]

    Journal of Machine Learning Research , volume=

    Wasserstein barycenters can be computed in polynomial time in fixed dimension , author=. Journal of Machine Learning Research , volume=