arxiv: 2605.13522 · v1 · submitted 2026-05-13 · 🧮 math.ST · stat.TH

Recognition: 2 theorem links

· Lean Theorem

Dependence functions based on Chatterjee's rank correlation

Carsten Limbach

Pith reviewed 2026-05-14 18:15 UTC · model grok-4.3

classification 🧮 math.ST stat.TH

keywords Chatterjee's xi coefficientMarkov productdependence functionsrank correlationstochastic dependencegeometric interpretationdirected dependence

0 comments

The pith

The Markov product extends Chatterjee's ξ-coefficient to two dependence functions that measure both functional dependence and diagonal concentration in directed stochastic dependence.

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

The paper reinterprets Chatterjee's ξ-coefficient through the Markov product construction, in which a copy Y' of the response Y is generated conditionally independent of Y given the predictor vector X. It defines two new dependence functions, φ_{(Y,X)} and κ_{(Y,X)}, that build on this pair. These functions capture not only how well Y can be expressed as a function of X but also how tightly the joint distribution of (Y, Y') concentrates near the diagonal. A reader would care because the scalar ξ alone leaves the geometric and directional character of dependence unexamined.

Core claim

By constructing the Markov product (Y, Y') with Y' conditionally independent of Y given X, the dependence functions φ_{(Y,X)} and κ_{(Y,X)} furnish a geometric and distributional extension of Chatterjee's coefficient that quantifies directed stochastic dependence through both functional representation and concentration near the diagonal.

What carries the argument

The Markov product (Y, Y'), the pair in which Y' is a conditionally independent copy of Y given X; it supplies the geometric object whose concentration near the diagonal is measured by the new functions.

If this is right

The scalar ξ coefficient is replaced by functions that separately track functional dependence and diagonal concentration.
Directed dependence can be visualized through the geometry of the Markov product near the diagonal.
The framework applies directly to vector-valued predictors without additional modification.
Comparison of the two functions distinguishes cases where functional representation is strong but the product remains diffuse.

Where Pith is reading between the lines

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

The same construction could be applied to time-series pairs to detect lagged dependence without assuming linearity.
Nonparametric estimators of φ and κ from finite samples would allow routine use in exploratory data analysis.
The concentration measure might serve as a diagnostic for the strength of causal arrows inferred from observational data.

Load-bearing premise

The Markov product construction faithfully captures the relevant dependence structure without requiring further restrictions on the joint distribution of (Y, X).

What would settle it

A concrete counter-example in which Y is a deterministic function of X yet either dependence function fails to reach its upper bound of one would falsify the claimed extension.

Figures

Figures reproduced from arXiv: 2605.13522 by Carsten Limbach.

**Figure 1.** Figure 1: From left to right: the random vector (Y, X) associated with a Gaussian copula (2.8) with parameter 0.65, a Marshall–Olkin copula (2.9) with parameters (1, 1/3), and a jump copula with parameter 1 (see Example 2.10), based on a sample size of n = 2500. We have the following characterizations of independence and perfect dependence: ϕ(Y,X)(t) = 2t − t 2 for all t ∈ [0, 1] if and only if X ⊥ Y, whereas ϕ(Y,X)… view at source ↗

**Figure 2.** Figure 2: The dependence functions ϕ(Y,X) and κ for a random vector (Y, X) associated with a Gaussian copula (2.8) with parameter ρ = 0.65 (red), a Marshall–Olkin copula (2.9) with parameters (α, β) = (1, 1 3 ) (blue), and a jump copula (2.10) with parameter 1 (green). to perfect dependence of Y on X, that is, ϕ ⊥ (Y,X) (t) = 2t − t 2 , κ⊥ (Y,X) (t) = (1 − t) 3 for all t ∈ [0, 1], and ϕ pd (Y,X) (t) = 1, κ pd (Y,X) … view at source ↗

**Figure 3.** Figure 3: Representation of A1 4 for absolute distance. The final part of the paper is devoted to a consistent plug-in estimator for the newly introduced dependence functions ϕ(Y,X) and κ. This estimator enables a detailed analysis of the underlying dependence structure. In particular, it provides explicit insights into the conditional distribution of Y given X, including the identification of functional relationsh… view at source ↗

**Figure 4.** Figure 4: Example 2.7; First line: Scatterplot with n = 1000 of a random vector (Y, X) associated with a Fr´echet copula with parameters α = β = 0.5 and its Markov product. Second line: Visualization of ϕ(Y,X) and κ(Y,X) . Example 2.9 (Marshall–Olkin copula). [12] Let α, β ∈ [0, 1]. The Marshall–Olkin copula is defined by Cα,β(u, v) := min{u 1−α v, uv1−β }. It holds that Cα,β = Π, where Π denotes the independence co… view at source ↗

**Figure 5.** Figure 5: First line: Two-dimensional Gaussian distribution with correlation coefficient [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

**Figure 6.** Figure 6: First line: Samples of size n = 1000 drawn from a Marshall–Olkin copula with parameter vector (α, β) = (1, 0.2) and its Markov product. Second line: Visualization of ϕ(Y,X) and κ(Y,X) for the same Marshall–Olkin copula. Denote by DLSL the class of all such functions. For δ ∈ DLSL, the associated lower semilinear copula Cδ : [0, 1]2 → [0, 1] is given by Cδ(x, y) =    y δ(x) x y ≤ x, x δ(y) y otherwis… view at source ↗

**Figure 7.** Figure 7: Top: Scatterplot of the jump copula Cm with sample size n = 1000 and its Markov product for m = 3. Bottom: Corresponding plots of ϕ(Y,X) and κ(Y,X) for Example 2.10. 3 Estimation In this section, we study the asymptotic behavior of the estimators ϕˆ (Y,X) and ˆκ(Y,X) . The general idea is closely related to [12]: the underlying Markov product distribution is approximated via a nearest-neighbor construction… view at source ↗

**Figure 8.** Figure 8: Top: Samples of size n = 10000 drawn from a lower semilinear copula with δ given by linear interpolation of fixed support points, together with its Markov product. Bottom: Visualization of ϕ(Y,X) and κ(Y,X) for Example 2.11. where V = FY (Y ) and V ′ = FY (Y ′ ). Proof. Define the empirical nearest-neighbor measure on [0, 1]2 by µn := 1 n Xn i=1 δ Ri n+1 , RN(i) n+1 . By Theorem 7 in [12], we have weak … view at source ↗

**Figure 9.** Figure 9: Wine-quality data for total sulfur dioxide and sulphates. The upper-left panel shows the original scatterplot, while the upper-right panel shows the corresponding Markov-product representation. The lower panels display the empirical ϕ(Y,X) - and κ(Y,X) -functions. Thus, ˆκ(Y,X)(1) indicates a moderate level of directed dependence, while ϕˆ (Y,X)(bn) quantifies the visible near-diagonal concentration in the… view at source ↗

**Figure 10.** Figure 10: Boxplots of the maximum deviation d∞(ϕ(Y,X) , ϕˆ (Y,X)) (left) and d∞(κ(Y,X) , κˆ(Y,X)) (right) across different sample sizes n. Each box is based on 500 independent repetitions. paper proves L 1 -consistency for ϕˆ (Y,X) and uniform consistency for ˆκ(Y,X) , rather than general uniform consistency for ϕˆ (Y,X) . Notably, the median deviation is systematically smaller for κ(Y,X) than for ϕ(Y,X) , whereas … view at source ↗

read the original abstract

We investigate a geometric and distributional reinterpretation of Chatterjee's $\xi$-coefficient, which measures functional dependence between a response variable $Y$ and a predictor vector $\mathbf{X}$. For this purpose, we analyze the Markov product $(Y,Y')$, where $Y'$ is a copy of $Y$ that is conditionally independent of $Y$ given $\mathbf{X}$. Based on this construction, we introduce and study two dependence functions, denoted by $\phi_{(Y,\mathbf{X})}$ and $\kappa_{(Y,\mathbf{X})}$. The proposed framework provides a geometric interpretation of the Markov product and extends Chatterjee's correlation coefficient to a richer and more interpretable object for the analysis of directed stochastic dependence. In particular, rather than only measuring how well $Y$ can be represented as a function of $\mathbf{X}$, the proposed dependence functions additionally quantify how strongly the corresponding Markov product is concentrated near the diagonal.

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

The paper defines two new dependence functions φ and κ from Chatterjee's ξ via the Markov product, adding a geometric concentration view.

read the letter

The core move here is straightforward: start with Chatterjee's ξ, form the Markov product (Y, Y') with Y' conditionally independent of Y given X, and define φ and κ directly from the joint law of that pair. This turns the original scalar coefficient into functions that also track how tightly the product measure hugs the diagonal. The geometric reading follows immediately from the conditional independence, without extra regularity conditions on the distributions. That part is clean and does not require fitting parameters or circular steps. The paper shows how these objects capture directed dependence beyond pure functional representation, which is a modest but natural extension. The construction itself is standard in probability, so the definitions land without hidden gaps. What the work does well is keep the focus on the joint law and avoid overcomplicating the setup. The claims in the abstract track with the Markov product properties, and the geometric interpretation is a direct consequence rather than an added assumption. Soft spots are limited. The extension feels incremental once you see the construction; similar ideas appear in other dependence measures that use auxiliary copies or conditionals, so the paper would benefit from explicit side-by-side comparisons to show where φ and κ differ in practice. Estimation details and finite-sample behavior are not highlighted, which leaves a practical gap for a statistics paper even if the population versions are well-defined. No load-bearing flaws show up in the setup or the stress-test notes. This is for readers already working with rank correlations and dependence coefficients who want a geometric angle on directed stochastic dependence. Someone extending Chatterjee's work or comparing multiple dependence measures would find it useful. It deserves peer review because the definitions are coherent, the math is grounded in standard tools, and referees can usefully push on comparisons and applications.

Referee Report

2 major / 2 minor

Summary. The paper reinterprets Chatterjee's ξ-coefficient geometrically via the Markov product (Y, Y'), where Y' is conditionally independent of Y given X. It defines two new dependence functions φ_{(Y,X)} and κ_{(Y,X)} that quantify the concentration of the product measure near the diagonal, thereby extending the original coefficient from a measure of functional dependence to a richer descriptor of directed stochastic dependence.

Significance. If the constructions and properties hold, the work supplies a parameter-free extension of a popular rank correlation that adds geometric content to the Markov product. The direct derivation from the joint law of (Y, Y') without auxiliary fitting or regularity assumptions on the conditional distributions is a clear strength and could support new diagnostic tools in nonparametric dependence analysis.

major comments (2)

The abstract asserts that φ and κ extend ξ while quantifying diagonal concentration, yet the provided text supplies neither the explicit integral or expectation definitions of these functions nor the derivation showing how they recover ξ as a special case; without these steps the central claim remains unverified.
The geometric interpretation is said to follow from concentration of the product measure near the diagonal, but the manuscript must demonstrate that this concentration is strictly stronger than the original ξ and is not an immediate restatement of the conditional-independence construction already used to define ξ.

minor comments (2)

Notation for the dependence functions should be introduced with a clear display equation immediately after the Markov-product definition rather than only in the abstract.
The paper should include a short table or proposition comparing the range and invariance properties of φ, κ, and ξ under monotone transformations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. We address the major comments point by point below, indicating the revisions we will implement.

read point-by-point responses

Referee: The abstract asserts that φ and κ extend ξ while quantifying diagonal concentration, yet the provided text supplies neither the explicit integral or expectation definitions of these functions nor the derivation showing how they recover ξ as a special case; without these steps the central claim remains unverified.

Authors: We agree that the abstract and introductory presentation would benefit from greater explicitness. The functions φ_{(Y,X)} and κ_{(Y,X)} are defined in Section 2 via expectations over the Markov product (Y,Y'), specifically as φ_{(Y,X)}(t) = P(|Y - Y'| ≤ t | X) integrated appropriately against the marginal of X, and κ as a normalized version. The recovery of ξ occurs as the integral of φ over [0,1] equaling 1 - ξ or the appropriate functional. In the revised manuscript we will insert the explicit integral/expectation formulas directly into the abstract and add a short derivation paragraph in the introduction that recovers ξ as the t=0 or integrated case. revision: yes
Referee: The geometric interpretation is said to follow from concentration of the product measure near the diagonal, but the manuscript must demonstrate that this concentration is strictly stronger than the original ξ and is not an immediate restatement of the conditional-independence construction already used to define ξ.

Authors: We will strengthen the geometric argument. Although the Markov product is the same object used for ξ, φ and κ retain the full law of the distance |Y-Y'| rather than collapsing it to a single scalar. We will add a dedicated subsection containing (i) a proposition showing that the map from the conditional law to the concentration function is injective on a strictly larger class than the functional-dependence case captured by ξ, and (ii) two explicit counter-examples (one discrete, one continuous) in which ξ is identical but the functions φ and κ differ, thereby establishing that the geometric descriptor is strictly richer. revision: yes

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central construction rests on the Markov product with conditional independence; no free parameters, invented entities, or additional axioms beyond standard probability are visible in the abstract.

axioms (1)

domain assumption Existence of a conditionally independent copy Y' given X
Invoked to form the Markov product (Y, Y') that underlies both dependence functions.

pith-pipeline@v0.9.0 · 5445 in / 1149 out tokens · 45860 ms · 2026-05-14T18:15:36.484894+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
ϕ_{(Y,X)}(t) := P(|F_Y(Y)−F_Y(Y')|≤t); κ_{(Y,X)}(t) := 1−3∫_0^t(1−ϕ(s))ds; concentration of Markov product near diagonal
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear
geometric interpretation via band A_t around diagonal after PIT

Reference graph

Works this paper leans on

17 extracted references · 14 canonical work pages · 1 internal anchor

[1]

On continuity of Chatterjee's rank correlation and related dependence measures

J. Ansari and S. Fuchs (2025). On continuity of Chatterjee’s rank correlation and related dependence measures.Bernoulli, to appear. arXiv:2503.11390.https://arxiv.org/abs/2503.11390

work page internal anchor Pith review Pith/arXiv arXiv 2025
[2]

Ansari and S

J. Ansari and S. Fuchs (2025). An ordering for the strength of functional dependence. arXiv:2511.06498.https://arxiv.org/abs/2511.06498

work page arXiv 2025
[3]

N. Deb, P. Ghosal, and B. Sen (2020). Measuring association on topological spaces using kernels and geometric graphs. arXiv:2010.01768.https://arxiv.org/abs/2010.01768

work page arXiv 2020
[4]

Tran and F

L. Tran and F. Han (2024). On a rank-based Azadkia–Chatterjee correlation coefficient. arXiv:2412.02668.https://arxiv.org/abs/2412.02668

work page arXiv 2024
[5]

Dalitz, J

C. Dalitz, J. Arning, and S. Goebbels (2024). A simple bias reduction for Chatterjee’s correlation. J. Stat. Theory Pract.18, Article 51. DOI:https://doi.org/10.1007/s42519-024-00399-y

work page doi:10.1007/s42519-024-00399-y 2024
[6]

Lin and F

Z. Lin and F. Han (2023). On boosting the power of Chatterjee’s rank correlation.Biometrika 110(2), 283–299. DOI:https://doi.org/10.1093/biomet/asac048. 19

work page doi:10.1093/biomet/asac048 2023
[7]

H. Shi, M. Drton, and F. Han (2024). On Azadkia–Chatterjee’s conditional dependence coefficient. Bernoulli30(2), 851–877. DOI:https://doi.org/10.3150/22-BEJ1529

work page doi:10.3150/22-bej1529 2024
[8]

Ansari and S

J. Ansari and S. Fuchs (2022). A direct extension of Azadkia & Chatterjee’s rank correlation to multi-response vectors. arXiv:2212.01621.https://arxiv.org/abs/2212.01621

work page arXiv 2022
[9]

Huang, Z

W. Huang, Z. Li, and Y. Wang (2025). A multivariate extension of Azadkia–Chatterjee’s rank coefficient. arXiv:2512.07443.https://arxiv.org/abs/2512.07443

work page arXiv 2025
[10]

Azadkia and S

M. Azadkia and S. Chatterjee (2021). A simple measure of conditional dependence.Ann. Stat. 49(6), 3070–3102. DOI:https://doi.org/10.1214/21-AOS2073

work page doi:10.1214/21-aos2073 2021
[11]

Fuchs and C

S. Fuchs and C. Limbach (2026). A dimension reduction for extreme types of directed dependence. Dependence Modeling14(1), Article 20250016. DOI:https://doi.org/10.1515/demo-2025-0016

work page doi:10.1515/demo-2025-0016 2026
[12]

Fuchs (2024)

S. Fuchs (2024). Quantifying directed dependence via dimension reduction.J. Multivar. Anal.201, Article 105266. DOI:https://doi.org/10.1016/j.jmva.2023.105266

work page doi:10.1016/j.jmva.2023.105266 2024
[13]

Durante and C

F. Durante and C. Sempi (2015).Principles of Copula Theory. Boca Raton, FL: Chapman and Hall/CRC

2015
[14]

R. B. Nelsen (2006).An Introduction to Copulas. 2nd ed., Springer

2006
[15]

Fuchs, C

S. Fuchs, C. Limbach, and F. Sch¨ urrer (2026). On exact regions between measures of concordance and Chatterjee’s rank correlation for lower semilinear copulas.Int. J. Approx. Reason.189, Article 109588. DOI:https://doi.org/10.1016/j.ijar.2025.109588

work page doi:10.1016/j.ijar.2025.109588 2026
[16]

Wine Quality dataset – Classification

Taweilo (n.d.). Wine Quality dataset – Classification. Kaggle dataset.https://www.kaggle.com/ datasets/taweilo/wine-quality-dataset-balanced-classification
[17]

Cortez, A

P. Cortez, A. Cerdeira, F. Almeida, T. Matos, and J. Reis (2009). Modeling wine preferences by data mining from physicochemical properties.Decision Support Systems47(4), 547–553. DOI: https://doi.org/10.1016/j.dss.2009.05.016. 20 A Proofs and additional results Proof of Theorem 1.1 Proof.By results in the literature [1], we have ξ(Y,X) = 6 Z R P(Y≥y, Y ′ ...

work page doi:10.1016/j.dss.2009.05.016 2009