arxiv: 2605.13326 · v1 · submitted 2026-05-13 · 📊 stat.ME

Recognition: unknown

A Note on the Folding Test of Unimodality: limitation and improved alternative

Colombe Becquart , Aurore Archimbaud , Anne M. Ruiz , Zaineb Smida

Authors on Pith no claims yet

Pith reviewed 2026-05-14 18:08 UTC · model grok-4.3

classification 📊 stat.ME

keywords unimodality testfolding testmixture distributionsmultimodality detectionDirac mixturesGaussian mixturesstatistical tests

0 comments

The pith

The folding test of unimodality misclassifies certain multimodal mixtures as unimodal, but a double-folding version corrects the error.

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

The paper identifies a limitation in the Folding Test of Unimodality where it fails to detect multimodality in specific univariate mixture distributions, such as certain Dirac and Gaussian mixtures. These failures are fully characterized, showing systematic misclassification of multimodal data as unimodal. A new double-folding procedure is introduced that applies the folding operation twice to capture additional information about the distribution's modality. This leads to the Double Folding Test of Unimodality, which corrects the original test's errors and demonstrates improved power for multimodality detection in simulation studies.

Core claim

The Folding Test of Unimodality (FTU) can systematically fail on Dirac mixtures and Gaussian mixtures by classifying them as unimodal despite clear multimodality; the double-folding procedure provides a corrected test that avoids these failures.

What carries the argument

The double-folding procedure, which performs two successive folding operations to obtain complementary modality information.

If this is right

FTU produces false unimodal results for Dirac mixtures with specific component separations.
Certain Gaussian mixtures also cause FTU to misclassify multimodality.
The double-folding test eliminates these specific misclassifications.
Simulations confirm higher detection power for multimodality under the new procedure.

Where Pith is reading between the lines

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

The approach could be tested on mixtures beyond Diracs and Gaussians to check if failures are limited to those families.
Double folding might be combined with other transformation-based tests to improve robustness.
In practice, the new test could reduce errors in applications like clustering where unimodality assumptions matter.

Load-bearing premise

That applying the folding operation twice captures complementary information about modality without introducing new biases or reducing detection power.

What would settle it

A multimodal Gaussian mixture with parameters that trigger FTU failure but where the double-folding test still returns a unimodal classification would show the new test does not fully resolve the problem.

Figures

Figures reproduced from arXiv: 2605.13326 by Anne M. Ruiz, Aurore Archimbaud, Colombe Becquart, Zaineb Smida.

**Figure 1.** Figure 1: 2-Gaussian mixture with ε1 = 0.3, µ1 = −2.8, ε2 = 0.7, µ2 = 1.2, and varying σ 2 : (a) values of Φ ∗ (X) (black) as a function of σ 2 , (b) probability density functions and associated exact pivots s ∗ , (c) cumulative distribution functions. 3.2 Motivation through the 3-Dirac mixture case Let us consider the standardized model (2) with k = 3 and ordered locations. Throughout this section, when the minimiz… view at source ↗

**Figure 2.** Figure 2: 3-Gaussian mixtures under different configurations, together with the corresponding exact and approximate [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗

read the original abstract

This note addresses a key limitation of the Folding Test of Unimodality (FTU). In specific univariate mixture settings, the folding-based criterion can systematically fail, misclassifying clearly multimodal distributions as unimodal. We fully characterize these failures for Dirac mixtures and extend the analysis to Gaussian mixtures. We then introduce a double-folding procedure that captures complementary information, leading to a new test, the Double Folding Test of Unimodality. It resolves the FTU failures and improves multimodality detection power in simulations.

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 note usefully pins down when the folding test fails on certain mixtures and offers a double-folding alternative that improves power in the simulations they ran, but the new test rests on limited theory and finite checks.

read the letter

The main takeaway is that the Folding Test of Unimodality has a documented blind spot on some Dirac and Gaussian mixtures, where it calls clearly multimodal data unimodal, and the authors characterize those failures before proposing a double-folding procedure to fix them. The characterization for Dirac mixtures looks solid and the extension to Gaussians is a reasonable next step. The double-folding idea is simple and the simulations show it catches the cases the original test misses while lifting overall detection power. That is the concrete advance here. The paper does a service by spelling out exactly where the original criterion breaks rather than just claiming a new method. The simulations appear to be set up to target the known failure regimes, which makes the reported gains credible for those settings. On the soft side, the argument for the new test stays simulation-driven. There is no general consistency proof or power analysis that covers distributions outside the mixtures they examined, so it is unclear whether double-folding introduces fresh false-unimodal cases in heavy-tailed or asymmetric laws. The abstract and stress-test note both flag this gap, and nothing in the provided material closes it. The work is aimed at statisticians who use or develop unimodality tests for mixture data. A reader already familiar with the folding test will get immediate value from the failure characterization and the practical alternative. It is narrow enough that it will not interest a broad audience, but the limitation it documents is real and worth recording. I would send this to peer review. The core observation is worth referee scrutiny even if the authors need to add more theory or wider simulations before acceptance.

Referee Report

3 major / 2 minor

Summary. The manuscript identifies a limitation in the Folding Test of Unimodality (FTU), where the folding-based criterion systematically misclassifies certain clearly multimodal distributions (specific Dirac mixtures and Gaussian mixtures) as unimodal. It fully characterizes these failures for Dirac mixtures, extends the analysis to Gaussian mixtures, and introduces a double-folding procedure that yields the Double Folding Test of Unimodality (DFTU). The new test is claimed to resolve the FTU failures while improving multimodality detection power, as shown in simulations.

Significance. If the characterization and simulation results hold, the note provides a useful practical warning about when FTU should not be applied in mixture settings and offers a simple alternative procedure. The explicit failure characterization for Dirac mixtures adds concrete diagnostic value for users of unimodality tests; the simulation evidence for improved power is a modest but positive contribution to the methodological literature on distribution modality testing.

major comments (3)

[Section introducing the double-folding procedure and DFTU] The central claim that DFTU 'resolves the FTU failures' rests on the double-folding operator capturing complementary modality information without new biases. No general consistency or power guarantee is supplied beyond the characterized mixtures; if double-folding can produce false-unimodal cases for heavy-tailed or asymmetric laws outside the simulated regimes, the resolution claim does not hold universally.
[Analysis of Gaussian mixtures] The extension of the failure characterization from Dirac mixtures to Gaussian mixtures is described as an 'extension' rather than a full parallel characterization. The specific parameter regimes (e.g., component separation, weights, variances) under which FTU fails for Gaussians must be stated with explicit conditions and proofs to support the systematic-failure assertion.
[Simulation study] Power improvement is demonstrated only via finite simulations. Without an accompanying asymptotic power analysis or consistency proof for DFTU, the claim that it improves multimodality detection remains simulation-dependent and does not yet fully substantiate the recommendation of DFTU as a general replacement.

minor comments (2)

[Abstract] The abstract states that DFTU 'improves multimodality detection power in simulations' but omits the number of Monte Carlo replications, sample sizes, and specific mixture configurations used; adding these details would improve reproducibility and clarity.
[Methodological sections] Notation for the folding operator and the double-folding operator should be introduced with explicit mathematical definitions early in the text to avoid ambiguity when comparing FTU and DFTU statistics.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address each major comment point by point below, indicating the revisions we will incorporate.

read point-by-point responses

Referee: [Section introducing the double-folding procedure and DFTU] The central claim that DFTU 'resolves the FTU failures' rests on the double-folding operator capturing complementary modality information without new biases. No general consistency or power guarantee is supplied beyond the characterized mixtures; if double-folding can produce false-unimodal cases for heavy-tailed or asymmetric laws outside the simulated regimes, the resolution claim does not hold universally.

Authors: We agree that the scope of our claims must be stated precisely. The manuscript characterizes specific failures of FTU for Dirac mixtures and extends the analysis to Gaussian mixtures, with DFTU shown to resolve those cases in both the analytic characterization and the simulations. We do not claim or prove universal consistency or absence of new biases for all distributions. In the revision we will update the abstract, introduction, and conclusion to explicitly limit the resolution claim to the characterized mixtures and the simulated regimes, and we will add a brief discussion of potential limitations for heavy-tailed or asymmetric distributions outside those regimes. revision: yes
Referee: [Analysis of Gaussian mixtures] The extension of the failure characterization from Dirac mixtures to Gaussian mixtures is described as an 'extension' rather than a full parallel characterization. The specific parameter regimes (e.g., component separation, weights, variances) under which FTU fails for Gaussians must be stated with explicit conditions and proofs to support the systematic-failure assertion.

Authors: We accept that greater precision is needed. The current text presents the Gaussian case as an extension of the Dirac analysis. In the revised manuscript we will add explicit conditions on the mixture parameters (minimum component separation relative to variances and weights) under which FTU systematically fails for two-component Gaussian mixtures, together with the corresponding derivations. revision: yes
Referee: [Simulation study] Power improvement is demonstrated only via finite simulations. Without an accompanying asymptotic power analysis or consistency proof for DFTU, the claim that it improves multimodality detection remains simulation-dependent and does not yet fully substantiate the recommendation of DFTU as a general replacement.

Authors: We acknowledge that the power comparison is based on finite-sample simulations and that a full asymptotic analysis lies beyond the scope of this short note. In the revision we will expand the simulation study to include additional heavy-tailed and asymmetric distributions, report results for a wider range of sample sizes, and moderate the language in the abstract and conclusion to present DFTU as an improved practical alternative for the settings studied rather than a general replacement. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper first directly characterizes the failure modes of the existing FTU on specific Dirac mixtures (fully enumerated) and extends the analysis to Gaussian mixtures via explicit construction. It then defines the double-folding operator and the resulting test statistic without any self-referential definitions, without renaming fitted quantities as predictions, and without load-bearing self-citations. The improvement claim is supported only by the mixture-specific characterization plus simulation results; no step reduces by construction to its own inputs. The derivation chain therefore remains independent of the target result.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the characterization of FTU failures in mixture models and the proposal of double folding as an improvement, with no additional free parameters or invented entities beyond the new test statistic.

axioms (1)

standard math The folding operation preserves certain properties of the distribution
The test builds on the mathematical definition of folding around the median.

pith-pipeline@v0.9.0 · 5391 in / 1189 out tokens · 72179 ms · 2026-05-14T18:08:03.080870+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

[1]

The annals of Statistics , pages=

The dip test of unimodality , author=. The annals of Statistics , pages=. 1985 , publisher=

work page 1985
[2]

Are your data gathered? , abstract =

Siffer, Alban and Fouque, Pierre-Alain and Termier, Alexandre and Largouët, Christine , year =. Are your data gathered? , abstract =. Proceedings of the 24th

work page
[3]

New statistical methods for data mining, contributions to anomaly detection and unimodality testing , copyright =

Siffer, Alban , collaborator =. New statistical methods for data mining, contributions to anomaly detection and unimodality testing , copyright =. 2019 , keywords =

work page 2019
[4]

arXiv preprint arXiv:2311.16614 , author =

A. arXiv preprint arXiv:2311.16614 , author =

work page arXiv
[5]

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

Using. Journal of the Royal Statistical Society Series B: Statistical Methodology , author =. 1981 , pages =

work page 1981
[6]

Pattern Recognition , author =

The. Pattern Recognition , author =. 2022 , keywords =

work page 2022
[7]

Journal of the American Statistical Association , author =

Alternatives to the. Journal of the American Statistical Association , author =. 1993 , keywords =

work page 1993
[8]

Izvestiya Nauchno-Issledovatel’skogo Instituta Matematiki i Mekhaniki , author =

On unimodal distributions , volume =. Izvestiya Nauchno-Issledovatel’skogo Instituta Matematiki i Mekhaniki , author =. 1938 , pages =

work page 1938
[9]

Biometrika , volume=

Group sequential methods in the design and analysis of clinical trials , author=. Biometrika , volume=. 1977 , publisher=

work page 1977
[10]

Ralph Urlus , title =

work page
[11]

Alban Siffer , title =

work page
[12]

Technometrics , volume=

The folded normal distribution , author=. Technometrics , volume=. 1961 , publisher=

work page 1961
[13]

and Haberland, Matt and Reddy, Tyler and Cournapeau, David and Burovski, Evgeni and Peterson, Pearu and Weckesser, Warren and Bright, Jonathan and

Virtanen, Pauli and Gommers, Ralf and Oliphant, Travis E. and Haberland, Matt and Reddy, Tyler and Cournapeau, David and Burovski, Evgeni and Peterson, Pearu and Weckesser, Warren and Bright, Jonathan and. Nature Methods , year =

work page
[14]

Journal of Statistical Planning and Inference , author =

A flexible fixed-sequence testing method for hierarchically ordered correlated multiple endpoints in clinical trials , volume =. Journal of Statistical Planning and Inference , author =. 2008 , keywords =. doi:10.1016/j.jspi.2007.06.009 , abstract =

work page doi:10.1016/j.jspi.2007.06.009 2008
[15]

and Hubert, L

Arabie, P. and Hubert, L. J. , editor =. Cluster. Advanced methods in marketing research , publisher =. 1994 , pages =

work page 1994
[16]

Journal of Educational Psychology , author =

Analysis of a complex of statistical variables into principal components , volume =. Journal of Educational Psychology , author =. 1933 , pages =

work page 1933
[17]

Joliffe, I. T. , year =. Principal

work page
[18]

and Ruiz, A

Caussinus, H. and Ruiz, A. , editor =. Interesting. 1990 , pages =. doi:10.1007/978-3-642-50096-1_19 , booktitle =

work page doi:10.1007/978-3-642-50096-1_19 1990
[19]

Journal of the Royal Statistical Society: Series B (Statistical Methodology) , author =

Invariant co-ordinate selection , volume =. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , author =. 2009 , pages =. doi:10.1111/j.1467-9868.2009.00706.x , abstract =

work page doi:10.1111/j.1467-9868.2009.00706.x 2009
[20]

Econometrics and Statistics , author =

Tandem clustering with invariant coordinate selection , volume =. Econometrics and Statistics , author =. 2024 , keywords =. doi:10.1016/j.ecosta.2024.03.002 , abstract =

work page doi:10.1016/j.ecosta.2024.03.002 2024
[21]

2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

Efficient unimodality test in clustering by signature testing , author=. 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2014 , organization=

work page 2014
[22]

Advances in neural information processing systems , volume=

Dip-means: an incremental clustering method for estimating the number of clusters , author=. Advances in neural information processing systems , volume=

work page
[23]

Pattern Recognition , volume=

To cluster, or not to cluster: An analysis of clusterability methods , author=. Pattern Recognition , volume=. 2019 , publisher=

work page 2019
[24]

Multimodal projection pursuit using the dip statistic , author=

work page