Kolmogorov-Smirnov distance and discrepancies versus Wasserstein distances

Fabien Panloup (LAREMA); Gilles Pag\`es (LPSM (UMR\_8001))

arxiv: 2605.03528 · v1 · submitted 2026-05-05 · 🧮 math.PR

Kolmogorov-Smirnov distance and discrepancies versus Wasserstein distances

Gilles Pag\`es (LPSM (UMR\_8001)) , Fabien Panloup (LAREMA) This is my paper

Pith reviewed 2026-05-07 14:33 UTC · model grok-4.3

classification 🧮 math.PR

keywords Wasserstein distanceKolmogorov-Smirnov distancediscrepancyprobability measuresinequalitiesoptimal transportmomentsdensities

0 comments

The pith

The p-Wasserstein distance between measures on [0,1]^d is upper-bounded by powers of their uniform discrepancy, and on R^d by a power of their Kolmogorov-Smirnov distance times the sum of p-moments.

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

The paper establishes inequalities that compare the p-Wasserstein distance to distances built as suprema of measures assigned to axis-aligned boxes. For probability measures supported on the unit cube [0,1]^d, it derives sharp upper bounds on the Wasserstein distance in terms of powers of the uniform discrepancy. For measures on all of R^d, a parallel upper bound holds that multiplies a power of the Kolmogorov-Smirnov distance by the sum of the p-moments of the two measures. Reverse inequalities are shown when one measure possesses a density whose L^s norm is finite for some s greater than 1. These relations let supremum-based distances control the optimal-transport cost under explicit support and integrability conditions.

Core claim

We establish inequalities that compare the p-Wasserstein distance to distances which are built as suprema of box measures. More precisely, when the measures are supported on [0,1]^d, we obtain sharp upper-bounds of the p-Wasserstein distance by (powers of) the (uniform) discrepancy. When the two distributions are supported by the whole R^d, their p-Wasserstein distance is upper bounded by the product of a (power of) their Kolmogorov-Smirnov (KS) distance with the sum of their p-moments. Reverse inequalities are established when one of the two distributions has a density, depending on its L^s-integrability with respect to the Lebesgue measure for some s>1.

What carries the argument

Discrepancy and Kolmogorov-Smirnov distance, each defined as the supremum over absolute differences in measure of axis-aligned rectangular boxes, used to bound the p-Wasserstein distance from above.

If this is right

The p-Wasserstein distance between measures supported on [0,1]^d is sharply upper-bounded by powers of their uniform discrepancy.
On R^d the p-Wasserstein distance is upper-bounded by a power of the Kolmogorov-Smirnov distance multiplied by the sum of the p-moments.
Reverse inequalities bound the discrepancy or Kolmogorov-Smirnov distance from above by the Wasserstein distance whenever one measure has an L^s density for s>1.
The bounds hold uniformly in dimension d under the stated support and moment conditions.

Where Pith is reading between the lines

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

The comparison suggests that sequences with small discrepancy will also produce empirical measures that converge at a controlled rate in Wasserstein distance.
Similar supremum-over-box arguments could be adapted to bound other transportation costs or to treat signed measures.
In Monte Carlo or quadrature settings the bounds supply a route to certify Wasserstein error using only box-counting statistics.

Load-bearing premise

The measures must be supported on the compact set [0,1]^d for the discrepancy bounds or possess finite p-moments for the unbounded-domain case, plus an L^s density for the reverse inequalities.

What would settle it

Two measures on [0,1]^d whose p-Wasserstein distance exceeds every positive power of their discrepancy would disprove the claimed upper bound.

read the original abstract

We establish inequalities that compare the p-Wasserstein distance to distances which are built as suprema of box measures. More precisely, when the measures are supported on $[0,1]^d$, we obtain sharp upper-bounds of the $p$-Wasserstein distance by (powers of) the (uniform) discrepancy. As an application, we retrieve the Pro\''inov Theorem. When the two distributions are supported {by the whole} $R^d$, {their} $p$-Wasserstein distance is upper bounded by the product of a (power of) their Kolmogorov-Smirnov (KS) distance with the sum of their $p$-moments. Reverse inequalities are established when one of the two distributions has a density, depending on its ${\cal L}^s$-integrability with respect to the Lebesgue measure for some $s>1$.

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.

Desk Editor's Note private letter to a colleague

This paper gives explicit sharp upper bounds on p-Wasserstein distance via discrepancy on the cube and via KS times moments on R^d, plus reverses under L^s density.

read the letter

This paper sets up some direct comparisons between the p-Wasserstein distance and two other distances built from suprema over boxes: the discrepancy on the unit cube and the Kolmogorov-Smirnov distance on R^d. The headline results are sharp upper bounds in both settings, plus reverse inequalities when one measure has a density with L^s integrability for s>1. They recover the Proinov theorem as a corollary in the compact case and extend the bounds to higher dimensions and to unbounded domains with explicit moment factors. The conditions line up with what you need for the distances to make sense, and the stress-test note confirms there is no circularity or hidden failure in the hypotheses. The soft spots are limited. The reverse inequalities stop at the density assumption, so they do not give a full equivalence without extra structure. Sharpness is claimed, but it would be good to see how close the bounds get in examples. Still, nothing looks load-bearing wrong. This is the sort of paper that helps readers who need to translate between different metrics for convergence of measures. People in Monte Carlo simulation or statistical testing might use the explicit forms to get rates more easily. It is not going to change the field, but it gives clean tools. I would bring it to a reading group if the group works on probability metrics. It deserves a serious referee because the statements are precise and the approach is standard but the extensions look useful. Send it for review.

Referee Report

0 major / 3 minor

Summary. The paper establishes inequalities relating the p-Wasserstein distance to suprema-based distances such as the uniform discrepancy (on [0,1]^d) and the Kolmogorov-Smirnov distance (on R^d). It derives sharp upper bounds for the Wasserstein distance in terms of powers of these quantities under compact support or finite-moment conditions, retrieves the Proinov theorem as an application, and provides reverse inequalities when one measure admits an L^s density for s>1.

Significance. If the stated bounds hold with the claimed sharpness, the work usefully connects optimal transport distances to classical discrepancy and KS metrics, which may aid statistical applications involving empirical measures and approximation theory. Retrieval of the Proinov theorem serves as a consistency check. The hypotheses (compact support for discrepancy bounds, finite p-moments for the unbounded case, and L^s integrability for reverses) are standard and necessary, strengthening applicability without introducing ad-hoc restrictions.

minor comments (3)

Abstract: the phrasing 'supported {by the whole} R^d, {their} p-Wasserstein distance' contains apparent LaTeX or editing artifacts; rephrase for grammatical clarity as 'supported on the whole R^d, their p-Wasserstein distance'.
Abstract: 'retrieve the Pro''inov Theorem' uses an unusual double-prime; standardize to 'Proinov' or 'Proinov's' throughout.
The manuscript would benefit from an explicit statement (perhaps in the introduction) of the precise definition of the Kolmogorov-Smirnov distance used, to avoid any ambiguity with the classical one-dimensional version.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the careful review and the positive recommendation for minor revision. We are pleased that the significance of our results in linking Wasserstein distances with discrepancy and Kolmogorov-Smirnov distances is acknowledged, along with the consistency check provided by the Proinov theorem. Since the major comments section does not list any specific points, we provide no point-by-point responses. We will make any necessary minor revisions to the manuscript.

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper derives explicit inequalities relating the p-Wasserstein distance to discrepancy and Kolmogorov-Smirnov distances under standard assumptions (compact support on [0,1]^d or finite p-moments on R^d, plus L^s density for reverse bounds). These follow from direct control of box probabilities, tail truncation, and integrability, without any reduction of a claimed result to a fitted parameter, self-definition, or load-bearing self-citation. Retrieval of the Proinov theorem is presented as an application of a known external result rather than an internal loop. No equations or steps are shown to be equivalent to their inputs by construction; the central claims remain independent of the paper's own fitted quantities or prior author work.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper's results rely on standard definitions and properties of probability measures, Wasserstein distances, Kolmogorov-Smirnov distances, and discrepancies from prior literature in probability theory. No free parameters or new entities are postulated.

axioms (1)

standard math Standard definitions and properties of p-Wasserstein distances, Kolmogorov-Smirnov distances, and uniform discrepancies on probability measures.
The inequalities build upon these established concepts in probability theory.

pith-pipeline@v0.9.0 · 5456 in / 1177 out tokens · 210213 ms · 2026-05-07T14:33:29.048026+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Convergence rate of the occupation measure of classes of ergodic processes toward their invariant distribution in mean Wasserstein distance
math.PR 2026-05 unverdicted novelty 5.0

General criteria extend L^p-mean Wasserstein convergence rates of occupation measures to non-stationary or non-Markovian ergodic processes under conditional convergence to equilibrium, with applications to Brownian di...

Reference graph

Works this paper leans on

14 extracted references · cited by 1 Pith paper

[1]

Numerical methods for stochastic processes

Nicolas Bouleau and Dominique L\'epingle. Numerical methods for stochastic processes . Wiley Series in Probability and Mathematical Statistics: Applied Probability and Statistics. John Wiley & Sons, Inc., New York, 1994. A Wiley-Interscience Publication

1994
[2]

An estimate concerning the K olmogoroff limit distribution

Kai-Lai Chung. An estimate concerning the K olmogoroff limit distribution. Trans. Amer. Math. Soc. , 67:36--50, 1949

1949
[3]

Constructive quantization: approximation by empirical measures

Steffen Dereich, Michael Scheutzow, and Reik Schottstedt. Constructive quantization: approximation by empirical measures. Ann. Inst. Henri Poincar\'e Probab. Stat. , 49(4):1183--1203, 2013

2013
[4]

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. Probab. Theory Relat. Fields , 162(3-4):707--738, 2015

2015
[5]

Gaunt and Siqi Li

Robert E. Gaunt and Siqi Li. Bounding kolmogorov distances through wasserstein and related integral probability metrics. Journal of Mathematical Analysis and Applications , 522(1):126985, 2023

2023
[6]

Jack C. Kiefer. On large deviations of the empiric D . F . of vector chance variables and a law of the iterated logarithm. Pacific J. Math. , 11:649--660, 1961

1961
[7]

Uniform distribution of sequences

Lauwerens Kuipers and Harald Niederreiter. Uniform distribution of sequences . Pure and Applied Mathematics. Wiley-Interscience [John Wiley & Sons], New York-London-Sydney, 1974

1974
[8]

Marginal and functional quantization of stochastic processes , volume 105 of Probability Theory and Stochastic Modelling

Harald Luschgy and Gilles Pag\`es. Marginal and functional quantization of stochastic processes , volume 105 of Probability Theory and Stochastic Modelling . Springer, Cham, 2023

2023
[9]

E. L. Lehmann and Joseph P. Romano. Testing statistical hypotheses . Springer Texts in Statistics. Springer, Cham, fourth edition, [2021] 2021

2021
[10]

Random number generation and quasi- M onte C arlo methods , volume 63 of CBMS-NSF Regional Conference Series in Applied Mathematics

Harald Niederreiter. Random number generation and quasi- M onte C arlo methods , volume 63 of CBMS-NSF Regional Conference Series in Applied Mathematics . Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1992

1992
[11]

Numerical probability

Gilles Pag \`e s. Numerical probability. An introduction with applications to finance . Universitext. Cham: Springer, 2nd edition edition, 2026

2026
[12]

Pro\"inov

Petko D. Pro\"inov. Discrepancy and integration of continuous functions. J. Approx. Theory , 52(2):121--131, 1988

1988
[13]

Topics in optimal transportation , volume 58 of Graduate Studies in Mathematics

C\'edric Villani. Topics in optimal transportation , volume 58 of Graduate Studies in Mathematics . American Mathematical Society, Providence, RI, 2003

2003
[14]

Optimal transport , volume 338 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]

C\' e dric Villani. Optimal transport , volume 338 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences] . Springer-Verlag, Berlin, 2009. Old and new

2009

[1] [1]

Numerical methods for stochastic processes

Nicolas Bouleau and Dominique L\'epingle. Numerical methods for stochastic processes . Wiley Series in Probability and Mathematical Statistics: Applied Probability and Statistics. John Wiley & Sons, Inc., New York, 1994. A Wiley-Interscience Publication

1994

[2] [2]

An estimate concerning the K olmogoroff limit distribution

Kai-Lai Chung. An estimate concerning the K olmogoroff limit distribution. Trans. Amer. Math. Soc. , 67:36--50, 1949

1949

[3] [3]

Constructive quantization: approximation by empirical measures

Steffen Dereich, Michael Scheutzow, and Reik Schottstedt. Constructive quantization: approximation by empirical measures. Ann. Inst. Henri Poincar\'e Probab. Stat. , 49(4):1183--1203, 2013

2013

[4] [4]

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. Probab. Theory Relat. Fields , 162(3-4):707--738, 2015

2015

[5] [5]

Gaunt and Siqi Li

Robert E. Gaunt and Siqi Li. Bounding kolmogorov distances through wasserstein and related integral probability metrics. Journal of Mathematical Analysis and Applications , 522(1):126985, 2023

2023

[6] [6]

Jack C. Kiefer. On large deviations of the empiric D . F . of vector chance variables and a law of the iterated logarithm. Pacific J. Math. , 11:649--660, 1961

1961

[7] [7]

Uniform distribution of sequences

Lauwerens Kuipers and Harald Niederreiter. Uniform distribution of sequences . Pure and Applied Mathematics. Wiley-Interscience [John Wiley & Sons], New York-London-Sydney, 1974

1974

[8] [8]

Marginal and functional quantization of stochastic processes , volume 105 of Probability Theory and Stochastic Modelling

Harald Luschgy and Gilles Pag\`es. Marginal and functional quantization of stochastic processes , volume 105 of Probability Theory and Stochastic Modelling . Springer, Cham, 2023

2023

[9] [9]

E. L. Lehmann and Joseph P. Romano. Testing statistical hypotheses . Springer Texts in Statistics. Springer, Cham, fourth edition, [2021] 2021

2021

[10] [10]

Random number generation and quasi- M onte C arlo methods , volume 63 of CBMS-NSF Regional Conference Series in Applied Mathematics

Harald Niederreiter. Random number generation and quasi- M onte C arlo methods , volume 63 of CBMS-NSF Regional Conference Series in Applied Mathematics . Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1992

1992

[11] [11]

Numerical probability

Gilles Pag \`e s. Numerical probability. An introduction with applications to finance . Universitext. Cham: Springer, 2nd edition edition, 2026

2026

[12] [12]

Pro\"inov

Petko D. Pro\"inov. Discrepancy and integration of continuous functions. J. Approx. Theory , 52(2):121--131, 1988

1988

[13] [13]

Topics in optimal transportation , volume 58 of Graduate Studies in Mathematics

C\'edric Villani. Topics in optimal transportation , volume 58 of Graduate Studies in Mathematics . American Mathematical Society, Providence, RI, 2003

2003

[14] [14]

Optimal transport , volume 338 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]

C\' e dric Villani. Optimal transport , volume 338 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences] . Springer-Verlag, Berlin, 2009. Old and new

2009