arxiv: 2604.03810 · v1 · submitted 2026-04-04 · 📊 stat.ME

Recognition: 2 theorem links

· Lean Theorem

A test for normality based on self-similarity

Akin Anarat , Holger Schwender

Authors on Pith no claims yet

Pith reviewed 2026-05-13 17:14 UTC · model grok-4.3

classification 📊 stat.ME

keywords normality testself-similarityempirical characteristic functiongoodness-of-fitMonte Carlo calibrationstatistical test

0 comments

The pith

Only the normal distribution remains unchanged under repeated self-similarity transformations of its characteristic function, allowing a new test to detect non-normality by measuring shifts in the empirical version.

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

Testing for normality is a routine step before applying methods that assume normal data. The paper proposes the Self-Similarity Test for Normality (SSTN) that rests on the property that a suitably centered and scaled sum of i.i.d. variables has the same distribution as the original only when that distribution is normal. The procedure applies a self-similarity transformation to the standardized empirical characteristic function and tracks whether the functional form stays the same across successive iterations. Deviations from normality produce systematic changes that are summarized into a test statistic whose null distribution is obtained by Monte Carlo calibration for small samples and by an asymptotic approximation for larger ones. Simulations indicate that the resulting test performs at least as well as, and often better than, several established normality tests.

Core claim

The SSTN evaluates normality by applying a self-similarity transformation to the standardized empirical characteristic function and examining how the transformed functions change across successive applications. For the normal distribution, repeated applications preserve the functional form of the characteristic function, whereas deviations from normality manifest in systematic changes between consecutive transforms. These changes are aggregated into a test statistic, whose null distribution is obtained by Monte Carlo calibration, using a sample-size-specific calibration for small samples and an approximation of the asymptotic null distribution for larger ones.

What carries the argument

The self-similarity transformation applied to the standardized empirical characteristic function, which leaves the functional form invariant if and only if the underlying distribution is normal.

If this is right

The test supplies an alternative to moment-based or distribution-function-based procedures for checking normality.
Sample-size-specific Monte Carlo calibration produces accurate critical values for small n while an asymptotic approximation covers larger samples.
Comprehensive simulations show the SSTN is competitive with or superior to several established normality tests across a range of alternatives.
The procedure can be inserted directly into data-analysis pipelines that require a preliminary normality check before parametric methods are applied.

Where Pith is reading between the lines

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

The invariance property might be adapted to construct tests for other stable laws that satisfy analogous self-similarity relations.
Links could be explored with existing characteristic-function-based goodness-of-fit procedures to combine their strengths.
Performance in multivariate or serially dependent settings remains open for direct investigation.

Load-bearing premise

Systematic changes in the transformed empirical characteristic function reliably signal departures from normality and the Monte Carlo or asymptotic calibration accurately captures the null distribution of the test statistic.

What would settle it

Monte Carlo experiments in which the empirical rejection rate of the SSTN under true normality deviates substantially from the nominal level, or in which power against common alternatives such as t or chi-squared distributions falls below that of standard tests.

Figures

Figures reproduced from arXiv: 2604.03810 by Akin Anarat, Holger Schwender.

**Figure 1.** Figure 1: Heatmaps of empirical rejection rates under the null hypothesis (normal distribution) across sample sizes and tests, separated by the underlying parameter values. Since the D’Agostino–Pearson test was developed for sample sizes ≥ 20 it is not applicable for n = 10, so these cases are omitted. 0.242 0.636 0.917 0.999 1.000 1.000 0.256 0.663 0.931 1.000 1.000 1.000 0.216 0.551 0.900 1.000 1.000 1.000 0.069 … view at source ↗

**Figure 2.** Figure 2: Heatmaps of statistical power under the Gamma distribution, evaluated across sample sizes and tests and separated by the underlying parameter values. Cases with n = 10 for the D’Agostino–Pearson test are omitted [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗

read the original abstract

Testing for normality is a widely used procedure in statistics and data analysis, often applied prior to employing methods that rely on the assumption of normally distributed data. While several existing tests target distributional characteristics such as higher-order moments, others focus on functional aspects such as the distribution function. In this article, we propose an alternative idea by exploiting the self-similarity property of the normal distribution and introduce the Self-Similarity Test for Normality (SSTN). This procedure leverages the structural property that the distribution of a suitably centered and scaled sum of independent and identically distributed random variables with finite variance coincides with the original distribution if and only if that distribution is normal. The SSTN evaluates normality by applying a self-similarity transformation to the standardized empirical characteristic function and examining how the transformed functions change across successive applications. For the normal distribution, repeated applications preserve the functional form of the characteristic function, whereas deviations from normality manifest in systematic changes between consecutive transforms. These changes are aggregated into a test statistic, whose null distribution is obtained by Monte Carlo calibration, using a sample-size-specific calibration for small samples and an approximation of the asymptotic null distribution for larger ones. A comprehensive simulation study shows that the SSTN performs at least competitively and frequently superior to several well-established tests for normality.

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 SSTN test applies repeated self-similarity maps to the empirical characteristic function and aggregates the shifts into a statistic that simulations show competes with or beats standard normality tests.

read the letter

The main contribution here is a normality test built around the fact that only the normal distribution stays invariant under a certain centering-and-scaling self-similarity map applied to its characteristic function. The authors implement that map on the standardized empirical version, track how the function changes over successive applications, and turn the total change into a test statistic. Null critical values come from Monte Carlo for small samples and an asymptotic approximation for larger ones. That mechanism is not one of the usual moment or EDF-based procedures, so the construction itself is new. The simulation study is presented as showing the test is at least as good as, and often better than, several established competitors across a range of alternatives. That is the practical takeaway worth noting. The logic avoids circularity because the null distribution is generated externally rather than estimated from the same sample. The self-similarity property is a clean characterization, and mapping it to the empirical characteristic function is a reasonable step. One soft spot is that the abstract does not spell out the exact form of the transformation or the precise aggregation rule for the changes, which makes it harder to judge sensitivity to particular alternatives or very small n without the full details. The reported simulation evidence is the main support, and the stress test finds no internal contradiction in the argument. This is a paper for applied statisticians who already use normality checks and might want one more option with decent power. It is coherent enough and grounded enough to deserve peer review rather than a desk reject.

Referee Report

3 major / 3 minor

Summary. The paper introduces the Self-Similarity Test for Normality (SSTN), which exploits the fact that the normal distribution is the unique finite-variance law invariant under a self-similarity map applied to its characteristic function. The procedure standardizes the empirical characteristic function, applies the map iteratively, aggregates the successive differences into a test statistic, and obtains critical values by Monte Carlo simulation for small n together with an asymptotic approximation for large n. A simulation study is presented claiming that SSTN is at least competitive with, and often superior to, several established normality tests.

Significance. If the simulation results are robust, the SSTN supplies a new functional test grounded in a characterizing property of the normal law rather than moments or empirical distribution functions. The dual calibration strategy (Monte Carlo for small samples, asymptotic for large) is practically useful, and the approach may offer advantages against alternatives that alter the shape of the characteristic function in a self-similarity-sensitive way.

major comments (3)

[§3] §3 (Test statistic construction): the precise definition of the self-similarity map applied to the standardized empirical characteristic function is not stated explicitly, nor is a proof given that the map leaves the normal characteristic function exactly invariant while producing systematic changes for non-normal laws; without this, the mechanistic link between the property and the statistic remains informal.
[§4] §4 (Simulation study): the claim that SSTN is 'frequently superior' is supported only by tabulated rejection rates against a limited set of alternatives; no power curves versus sample size, no results for heavy-tailed or multimodal alternatives, and no comparison of computational cost are provided, so the scope of superiority cannot be assessed.
[§4.2] §4.2 (Asymptotic calibration): the paper invokes an asymptotic null distribution for large n but supplies neither a derivation of the limiting law nor a numerical check of the approximation error for the sample sizes where the switch from Monte Carlo to asymptotic occurs.

minor comments (3)

[§3] Notation for the iterated characteristic functions (e.g., φ_k) is introduced without a clear recursive formula or reference to the earlier definition of the self-similarity operator.
[§4] The simulation tables do not report standard errors of the estimated rejection probabilities, making it impossible to judge whether apparent differences between SSTN and competing tests are statistically meaningful.
[§1] A brief comparison with other characteristic-function-based normality tests (e.g., those using integrated squared differences) is missing from the introduction and discussion.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful reading and constructive suggestions. We address each major comment below and have prepared a revised manuscript that incorporates the requested clarifications and expansions.

read point-by-point responses

Referee: [§3] §3 (Test statistic construction): the precise definition of the self-similarity map applied to the standardized empirical characteristic function is not stated explicitly, nor is a proof given that the map leaves the normal characteristic function exactly invariant while producing systematic changes for non-normal laws; without this, the mechanistic link between the property and the statistic remains informal.

Authors: We agree that an explicit statement and proof strengthen the presentation. In the revised manuscript we have inserted the precise definition of the self-similarity map (the functional iteration applied to the standardized empirical characteristic function) directly into Section 3, together with a short appendix proof that the standard normal characteristic function is a fixed point of the map while non-normal finite-variance laws produce strictly positive increments under iteration. These additions make the link between the characterizing property and the test statistic fully rigorous. revision: yes
Referee: [§4] §4 (Simulation study): the claim that SSTN is 'frequently superior' is supported only by tabulated rejection rates against a limited set of alternatives; no power curves versus sample size, no results for heavy-tailed or multimodal alternatives, and no comparison of computational cost are provided, so the scope of superiority cannot be assessed.

Authors: We accept that the original simulation section was too concise. The revised version adds (i) power curves plotted against sample size for representative alternatives, (ii) additional heavy-tailed (Student-t with 3 and 5 df) and multimodal (two- and three-component Gaussian mixtures) cases, and (iii) a brief runtime comparison showing that the Monte-Carlo calibration step dominates cost but remains feasible up to n=1000. The original claim is retained only as “at least competitive and frequently superior on the alternatives examined,” now qualified by the expanded tables and figures. revision: partial
Referee: [§4.2] §4.2 (Asymptotic calibration): the paper invokes an asymptotic null distribution for large n but supplies neither a derivation of the limiting law nor a numerical check of the approximation error for the sample sizes where the switch from Monte Carlo to asymptotic occurs.

Authors: We have added both items. Appendix B now contains a derivation of the limiting null distribution obtained by applying a functional central-limit theorem to the iterated empirical characteristic function. In addition, we report a numerical study comparing Monte-Carlo and asymptotic critical values for n ranging from 200 to 1000; the relative error falls below 4 % for n ≥ 500, justifying the switch point chosen in the paper. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The SSTN constructs its test statistic directly from the known characterization that the normal law is the unique finite-variance distribution invariant under the self-similarity map on the characteristic function. This map is applied to the standardized empirical characteristic function, successive differences are aggregated into the statistic, and the null distribution is calibrated externally via Monte Carlo (sample-size-specific) or asymptotic approximation. No parameter is fitted to the same data and then relabeled as a prediction; the uniqueness property is invoked as an external mathematical fact rather than derived internally or smuggled via self-citation; the simulation study comparing power is independent of the construction. The derivation chain therefore remains self-contained and does not reduce to its inputs by definition.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on the classical characterization that the distribution of a suitably centered and scaled sum of i.i.d. random variables equals the original distribution if and only if the distribution is normal. No new free parameters or invented entities are introduced in the abstract; the test statistic is defined from the data and calibrated externally.

axioms (1)

domain assumption A random variable has a normal distribution if and only if the distribution of a suitably centered and scaled sum of i.i.d. copies coincides with the original distribution.
This is the classical self-similarity characterization invoked to motivate the transformation applied to the empirical characteristic function.

pith-pipeline@v0.9.0 · 5518 in / 1377 out tokens · 44898 ms · 2026-05-13T17:14:37.226954+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 echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

the normal distribution is the only self-similar distribution among all non-degenerate distributions with finite variance
IndisputableMonolith/Foundation/AlphaDerivationExplicit.lean phi_fixed_point echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

phi is the unique self-similar fixed point (forced)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages

[1]

C., Testing for Normality , CRC Press, Boca Raton, 2002, doi:10.1201/9780203910894

Thode, H. C., Testing for Normality , CRC Press, Boca Raton, 2002, doi:10.1201/9780203910894

work page doi:10.1201/9780203910894 2002
[2]

M., and Bera, A

Jarque, C. M., and Bera, A. K., Eﬃcient tests for normality, homoscedasticity and se- rial independence of regression residuals , Economics Letters, Vol. 6, No. 3, 1980, pp. 255–259, doi:10.1016/0165-1765(80)90024-5

work page doi:10.1016/0165-1765(80)90024-5 1980
[3]

D’Agostino, R. B. and Pearson, E. S., Tests for Departure from Normality. Empirical Results for t he Distributions of b2 and √ b1, Biometrika, 60(3):613–622, 1973, doi:10.2307/2335012

work page doi:10.2307/2335012 1973
[4]

S., and Wilk, M

Shapiro, S. S., and Wilk, M. B., An analysis of variance test for normality (complete sample s), Biometrika, Vol. 52, No. 3–4, 1965, pp. 591–611, doi:10.1093/biomet/52 .3-4.591

work page doi:10.1093/biomet/52 1965
[5]

W., and Darling, D

Anderson, T. W., and Darling, D. A., A Test of Goodness of Fit , Journal of the American Statistical Association, Vol. 49, No. 268, 1954, pp. 765–769, doi:10.23 07/2281537

work page 1954
[6]

W., On the Kolmogorov-Smirnov Test for Normality with Mean and V ariance Un- known, Journal of the American Statistical Association, Vol

Lilliefors, H. W., On the Kolmogorov-Smirnov Test for Normality with Mean and V ariance Un- known, Journal of the American Statistical Association, Vol. 62, No. 318, 1967, pp. 399–402, doi:10.1080/01621459.1967.10482916

work page doi:10.1080/01621459.1967.10482916 1967
[7]

Smirnov, N., Table for Estimating the Goodness of Fit of Empirical Distri butions, Annals of Mathematical Statistics, Vol. 19, No. 2, 1948, pp. 279–281, doi:10.1214/ aoms/1177730256

work page arXiv 1948
[8]

Stochastic Processes and Long Range Dependence, Springer Series in Operations Research and Financial Engin eering, Springer, Cham, 2016, doi:10.1007/978-3-319-45575-4 8

Samorodnitsky, G., Self-Similar Processes . Stochastic Processes and Long Range Dependence, Springer Series in Operations Research and Financial Engin eering, Springer, Cham, 2016, doi:10.1007/978-3-319-45575-4 8

work page doi:10.1007/978-3-319-45575-4 2016
[9]

Kallenberg, O., Foundations of Modern Probability , 3rd ed., Probability Theory and Stochastic Modelling, Springer, Cham, 2021, doi:10.1007/978-3-030-61871-1

work page doi:10.1007/978-3-030-61871-1 2021
[10]

G., A review of testing procedures based on the empirical charac teristic function , South African Statistical Journal, 50(1):1–14, 2016, doi:10.37 920/sasj.2016.50.1.1

Meintanis, S. G., A review of testing procedures based on the empirical charac teristic function , South African Statistical Journal, 50(1):1–14, 2016, doi:10.37 920/sasj.2016.50.1.1

work page 2016
[11]

Epps, T. W. and Pulley, L. B., A test for normality based on the empirical characteristic f unction, Biometrika, 70(3):723–726, 1983, doi:10.2307/2336512

work page doi:10.2307/2336512 1983
[12]

and Takeuchi, K., The studentized empirical characteristic function and its application to test for the shape of distribution , Biometrika, 68(1):55–65, 1981, doi:10.2307/2335805

Murota, K. and Takeuchi, K., The studentized empirical characteristic function and its application to test for the shape of distribution , Biometrika, 68(1):55–65, 1981, doi:10.2307/2335805

work page doi:10.2307/2335805 1981
[13]

V., Barrera D., Jim´ enez M

Alba M. V., Barrera D., Jim´ enez M. D. A homogeneity test based on empirical characteristic funct ions. Computational Statistics. 16(2):255–270, 2001. doi:10.1 007/s001800100064

work page 2001
[14]

Nolan, J. P., Univariate Stable Distributions: Models for Heavy Tailed D ata, Springer Series in Operations Research and Financial Engineering, Springer, Cham, 2020, doi:10.1007/978-3-030-52915-4

work page doi:10.1007/978-3-030-52915-4 2020
[15]

W., Tests for location-scale families based on the empirical ch aracteristic function , Metrika, 62:99–114, 2005, doi:10.1007/s001840400358

Epps, T. W., Tests for location-scale families based on the empirical ch aracteristic function , Metrika, 62:99–114, 2005, doi:10.1007/s001840400358

work page doi:10.1007/s001840400358 2005
[16]

Arnastauskait˙ e, J., Ruzgas, T., and Braˇ z˙ enas, M.,An Exhaustive Power Comparison of Normality Tests , Mathematics, Vol. 9, No. 7, 2021, Article 788, doi:10.3390/ math9070788

work page 2021
[17]

Delicado, P., Functional k-sample problem when data are density function s, Computational Statistics, 22:391–410, 2007, doi:10.1007/s00180-007-0047-y. 18 A TEST FOR NORMALITY BASED ON SELF-SIMILARITY Appendix 0.308 0.730 0.975 1.000 1.000 1.000 0.326 0.782 0.985 1.000 1.000 1.000 0.275 0.694 0.966 1.000 1.000 1.000 0.086 0.448 0.861 0.999 1.000 1.000 0.20...

work page doi:10.1007/s00180-007-0047-y 2007