Correct estimate of the probability of false detection of the matched filter in weak-signal detection problems. III (Peak distribution method versus the Gumbel distribution method)

A. Biggs; N. Hayatsu; P. Andreani; R. Vio

arxiv: 1907.01465 · v1 · pith:YJLEW5ACnew · submitted 2019-07-02 · 🌌 astro-ph.IM

Correct estimate of the probability of false detection of the matched filter in weak-signal detection problems. III (Peak distribution method versus the Gumbel distribution method)

R. Vio , P. Andreani , A. Biggs , N. Hayatsu This is my paper

Pith reviewed 2026-05-25 10:46 UTC · model grok-4.3

classification 🌌 astro-ph.IM

keywords matched filterfalse alarm probabilityGaussian random fieldpeak distributionGumbel distributionweak signal detectionALMA maps

0 comments

The pith

The peak distribution method for false alarm probability in matched filtering is more flexible than the Gumbel method while producing nearly identical results.

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

This paper compares two parametric methods for estimating the probability of false alarm when applying matched filters to detect weak signals in a Gaussian random field. One method relies on the probability density function of peak amplitudes in a smooth isotropic field; the other uses the Gumbel distribution as the limiting form for extremes. The comparison is tested through simulations and real ALMA maps. Both approaches yield almost the same numerical values for the false alarm probability once the number of peaks is taken into account. The peak-amplitude method is presented as the more flexible of the two because it permits direct checks on whether the detection assumptions hold.

Core claim

Although the peak distribution method and the Gumbel distribution method produce almost identical results for calculating the probability of false detection, the peak distribution method is more flexible and allows checking the reliability of the detection procedure in matched filtering of weak signals in Gaussian random fields.

What carries the argument

Probability density function of peak amplitudes in a smooth isotropic Gaussian random field, used to compute the probability of false alarm while accounting for the total number of peaks.

If this is right

The number of peaks in the filtered field must be included in the calculation or the probability of false alarm is severely underestimated.
Parametric use of the peak-amplitude PDF exploits more information than non-parametric alternatives.
Both methods can be applied directly to astronomical maps such as those from ALMA.
The peak distribution approach supports additional reliability checks on the detection procedure that are not available with the Gumbel method.

Where Pith is reading between the lines

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

The added flexibility of the peak method could support iterative adjustment of detection thresholds when noise properties vary across a map.
The close agreement between the two methods suggests either could serve as a reference when testing new detection algorithms.
Extension to other domains that rely on peak finding in Gaussian noise, such as image processing or time-series analysis, may benefit from the same reliability checks.

Load-bearing premise

The noise is the realization of a smooth and isotropic Gaussian random field.

What would settle it

A dataset or simulation in which the observed distribution of peak amplitudes deviates from the expected PDF for a smooth isotropic Gaussian random field, causing the two methods to return materially different false-alarm probabilities.

Figures

Figures reproduced from arXiv: 1907.01465 by A. Biggs, N. Hayatsu, P. Andreani, R. Vio.

**Figure 2.** Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Figure 3: Left panel: histogram H(zmax) vs. the theoretical PDF υTh(zmax), mean PDF υMean(zmax) and the maximum likelihood PDF υML(zmax) of the value zmax of the highest peak of a zero-mean unit-variance GRF with autocorrelation given by a circular Gaussian with dispersion set to three pixels. The numerical experiment is based on the simulation of 5 × 103 GRF’s of size 500 × 500 pixels. υTh(zmax) has been computed u… view at source ↗

**Figure 4.** Figure 4: Cumulative distribution functions corresponding to the PDFs in [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

**Figure 5.** Figure 5: As in [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗

**Figure 6.** Figure 6: Check for the applicability of the proposed detection procedure for two different situations (see text). Left panel: Histogram H(z) vs. ψ(z) from Eq. (10) for a numerical realization of a zero-mean unit-variance Gaussian random field with autocorrelation given by a circular Gaussian with dispersion set to three pixels. Right panel: As in the left panel but with the dispersion of the Gaussian autocorrelatio… view at source ↗

**Figure 7.** Figure 7: Interferometric 1075 × 1075 pixels ALMA map used for testing the detection performances of PAM and GDM. Six point sources have been detected by both methods with a high level of confidence. They are labeled with a number in order of decreasing intensity. Article number, page 13 of 15 [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗

**Figure 8.** Figure 8: Checks of the conditions of applicability of the detection procedure for the interferometric ALMA map (see text). Top-left panel: Histogram of the values of the pixels vs. the standard Gaussian PDF φ(x). Top-right panel: Histogram of the peak values vs. the theoretical PDF ψ(z) given by Eq. (10). Bottom-left panel: Slices along the X and the Y directions of the sample autocorrelation function vs. the corre… view at source ↗

**Figure 9.** Figure 9: Sub-maps corresponding to the areas of the point sources detected in the map in [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗

read the original abstract

The matched filter (MF) represents one of the main tools to detect signals from known sources embedded in the noise. In the Gaussian case the noise is assumed to be the realization of a Gaussian random field (GRF). The most important property of the MF, the maximization of the probability of detection subject to a constant probability of false detection or false alarm (PFA), makes it one of the most popular techniques. However, the MF technique relies upon the a priori knowledge of the number and the position of the searched signals in the GRF which usually are not available. A typical way out is to assume that the position of a signal coincides with one of the peaks in the matched filtered data. A detection is claimed when the probability that a given peak is due only to the noise (i.e. the PFA) is smaller than a prefixed threshold. In this case the probability density function (PDF) of the amplitudes has to be used for the computation of the PFA, which is different from the Gaussian. Moreover, the probability that a detection is false depends on the number of peaks present in the filtered GRF, the greater the number of peaks in a GRF, the higher the probability of peaks due to the noise that exceed the detection threshold. If not taken into account, the PFA can be severely underestimated. Many solutions proposed to this problem are non-parametric hence not able to exploit all the available information. This limitation has been overcome by means of two efficient parametric approaches, one based on the PDF of the peak amplitudes of a smooth and isotropic GRF whereas the other uses the Gumbel distribution (the asymptotic PDF of the corresponding extreme). Simulations and ALMA maps show that, although the two methods produce almost identical results, the first is more flexible and allows us to check the reliability of the detection procedure.

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 peak and Gumbel methods produce nearly identical PFA numbers on the tested simulations and ALMA maps, but the paper supplies no numbers, no robustness checks, and no evidence that the claimed flexibility survives real noise deviations from the ideal GRF model.

read the letter

The paper's main point is a head-to-head comparison showing that the peak-amplitude PDF method and the Gumbel extreme-value method give almost the same false-alarm probabilities on both simulated Gaussian random fields and real ALMA maps, while the peak method is presented as more flexible for checking detection reliability. That direct numerical comparison on actual telescope data is the only concrete new element here; the rest recycles the two parametric approaches the authors have already published in earlier parts of the series. They deserve credit for moving beyond pure theory and including real maps rather than stopping at simulations. The execution, however, stays thin. The abstract states the two methods agree but gives no quantitative measure of agreement, no error analysis, and no description of how the methods were coded or how many peaks or maps were used. The flexibility advantage rests entirely on the assumption that the filtered noise remains a smooth, isotropic Gaussian random field whose peak PDF is known exactly. The stress-test note correctly flags that no test is shown of how the computed PFA drifts when that assumption is mildly violated, as often happens with real interferometer data. Without those checks the reliability-check procedure the authors advertise cannot be trusted to be reliable. This is a narrow, incremental methods note aimed at the small group already using matched filters for weak-source detection in radio astronomy. Most readers will find nothing new enough to justify time on it. The thinking is clear and the citations look standard for the subfield, but the evidential gaps are large enough that the paper does not merit sending out for peer review.

Referee Report

2 major / 1 minor

Summary. The manuscript compares two parametric methods for estimating the probability of false alarm (PFA) when using the matched filter to detect weak signals in a Gaussian random field (GRF): the peak-amplitude PDF method derived for a smooth isotropic GRF, and the Gumbel distribution as the asymptotic extreme-value PDF. It claims that simulations and ALMA maps demonstrate nearly identical results between the methods, while the peak method is more flexible and permits checks on detection reliability. The GRF assumption is stated explicitly as the foundation for the peak method.

Significance. If the quantitative agreement holds and the peak PDF remains valid under realistic conditions, the work would supply a practical, information-preserving approach to PFA estimation that accounts for peak multiplicity, which is relevant for astronomical source detection in interferometric data. The parametric nature of both methods is a potential strength relative to non-parametric alternatives.

major comments (2)

[Abstract] Abstract: the assertion that 'simulations and ALMA maps show that, although the two methods produce almost identical results, the first is more flexible' is unsupported by any quantitative metrics, error analysis, implementation details, or description of how agreement or flexibility was measured, preventing evaluation of the central claim.
[Abstract] Abstract: the claimed flexibility of the peak distribution method for reliability checks rests on the untested assumption that the exact peak PDF for a smooth isotropic GRF remains valid for filtered ALMA noise; no quantitative assessment is provided of PFA error growth under controlled violations of smoothness, isotropy, or stationarity, which directly affects trustworthiness of the reliability-check procedure.

minor comments (1)

[Abstract] The abstract would benefit from specifying the number of simulations performed, the properties of the ALMA maps, and the precise definition of 'almost identical results'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We address each point below, proposing targeted revisions to better support the claims while preserving the manuscript's scope.

read point-by-point responses

Referee: [Abstract] Abstract: the assertion that 'simulations and ALMA maps show that, although the two methods produce almost identical results, the first is more flexible' is unsupported by any quantitative metrics, error analysis, implementation details, or description of how agreement or flexibility was measured, preventing evaluation of the central claim.

Authors: The abstract serves as a concise summary; the quantitative comparisons (direct PFA value agreement between methods), error considerations, and implementation (peak PDF parameterization versus Gumbel fitting) are detailed in the main text via the simulation setups and ALMA map applications. To address the concern, we will revise the abstract to include a brief qualifier referencing the observed agreement level and the specific flexibility (ability to perform per-peak reliability checks using the exact PDF). revision: yes
Referee: [Abstract] Abstract: the claimed flexibility of the peak distribution method for reliability checks rests on the untested assumption that the exact peak PDF for a smooth isotropic GRF remains valid for filtered ALMA noise; no quantitative assessment is provided of PFA error growth under controlled violations of smoothness, isotropy, or stationarity, which directly affects trustworthiness of the reliability-check procedure.

Authors: The manuscript explicitly grounds the peak PDF in the smooth isotropic GRF assumption and presents ALMA maps as a practical test case where results align with simulations. A systematic quantification of PFA error growth under controlled violations of those assumptions is not performed, as the work centers on method comparison under the stated conditions rather than robustness testing. We will revise the abstract and add a clarifying sentence in the discussion to restate the assumptions and note the lack of such sensitivity analysis. revision: partial

Circularity Check

0 steps flagged

No significant circularity; methods rest on external GRF peak statistics and Gumbel asymptotics

full rationale

The derivation chain invokes the known PDF of peak amplitudes in a smooth isotropic Gaussian random field and the Gumbel extreme-value distribution. These are standard results in the literature on random fields and are not derived or fitted within the paper. The comparison of the two PFA methods is performed via simulations and ALMA maps, which constitute external benchmarks rather than self-referential fits. No equations reduce a claimed prediction to a parameter defined by the same paper, and no load-bearing uniqueness theorem or ansatz is imported via self-citation in the provided text. The GRF assumption is stated explicitly but is an external modeling choice, not a circular definition.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that the noise field is smooth and isotropic, allowing the peak-amplitude PDF to be used directly; no free parameters or invented entities are mentioned in the abstract.

axioms (1)

domain assumption The noise is the realization of a smooth and isotropic Gaussian random field
Invoked to justify use of the peak amplitude PDF for PFA computation

pith-pipeline@v0.9.0 · 5894 in / 1201 out tokens · 49150 ms · 2026-05-25T10:46:18.337172+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages · 1 internal anchor

[1]

2016, Spatial Point Patterns (New York: CRC Press)

Baddeley, A., Rubak, E., & Turner, R. 2016, Spatial Point Patterns (New York: CRC Press)

work page 2016
[2]

Benjamini, Y ., & Hochberg, Y ., 1995, J. R. Stat. Soc. B, 57, 289

work page 1995
[3]

2004, Extreme Val- ues and Related Models with Applications in Engineering and Science (San Francisco: John Wiley & Sons)

Castillo, E., Hadi, A.S., Balakrishnan, N., & Sarabia, J.M. 2004, Extreme Val- ues and Related Models with Applications in Engineering and Science (San Francisco: John Wiley & Sons)

work page 2004
[4]

2011, MNRAS, 414, 2436

Colombi, S., Davis, O., Devriendt, J., Prunet„ J., & Silk, J. 2011, MNRAS, 414, 2436

work page 2011
[5]

On the Explicit Height Distribution and Expected Number of Local Maxima of Isotropic Gaussian Random Fields

Cheng, D., & Schwartzman, A. 2015b, arXiv:1503.01328 [math.PR]

work page internal anchor Pith review Pith/arXiv arXiv
[6]

2018, Bernoulli 4B(24), 3422

Cheng, D., & Schwartzman, A. 2018, Bernoulli 4B(24), 3422

work page 2018
[7]

et al., 2017, MNRAS, 466, 861

Dunlop, J.S., McLure, R.J., Biggs, A.D. et al., 2017, MNRAS, 466, 861

work page 2017
[8]

2013, Introduction to Mathematical Statistics (New York: Pearson)

Hogg, R.V ., McKean, J.W., & Craig, A.T. 2013, Introduction to Mathematical Statistics (New York: Pearson)

work page 2013
[9]

2002, AJ, 123, 1086

Hopkins, A.M., Miller, C.J., & Connolly, A.J. 2002, AJ, 123, 1086

work page 2002
[10]

2005, Journal of Statistical Physics, 119, 777

Majumdar, S.N., & Comtet, A. 2005, Journal of Statistical Physics, 119, 777

work page 2005
[11]

1998, Fundamentals of Statistical Signal Processing: Detection The- ory (London: Prentice Hall)

Kay, S.M. 1998, Fundamentals of Statistical Signal Processing: Detection The- ory (London: Prentice Hall)

work page 1998
[12]

Lehmann, E.L., & Romano, J.P., 2005, Testing Statistical Hypotheses (New: York: Springer)

work page 2005
[13]

Miller, J.C., Genovese, C., Nichol, R.C. et al. 2001, AJ, 122, 3492

work page 2001
[14]

Neyman, J., & Pearson, E.S., 1933, Phil. Trans. Roy. Soc. Ser. A231, 289

work page 1933
[15]

Pavesi, R., Sharon, C.E., Riechers, D.A. et al. 2018, Ap.J., 864, 49P

work page 2018
[16]

2004, A&A, 414, 17

Vio, R., Andreani, P., & Wamsteker, W. 2004, A&A, 414, 17

work page 2004
[17]

2016, A&A, 589, A20 (V A16)

Vio, R., & Andreani, P. 2016, A&A, 589, A20 (V A16)

work page 2016
[18]

2018, A&A, 604, A115 (VV A17)

Vio, R., Vergès, C., & Andreani, P. 2018, A&A, 604, A115 (VV A17)

work page 2018
[19]

34854corr Table 1

Vitells, O., & Gross, E., 2001, Astroparticle Physics, 35, 230 Article number, page 5 of 15 A&A proofs: manuscript no. 34854corr Table 1. Identiﬁcation of the sources on the map in Fig. 7. Source number identiﬁcation coordinates RA DEC Source 1 UDF2 3h32m43.53s -27d46 ′39.25′′ Source 2 UDF3 3h32m38.55s -27d46 ′34.57′′ Source 3 UDF5 3h32m36.96s -27d47 ′27....

work page 2001

[1] [1]

2016, Spatial Point Patterns (New York: CRC Press)

Baddeley, A., Rubak, E., & Turner, R. 2016, Spatial Point Patterns (New York: CRC Press)

work page 2016

[2] [2]

Benjamini, Y ., & Hochberg, Y ., 1995, J. R. Stat. Soc. B, 57, 289

work page 1995

[3] [3]

2004, Extreme Val- ues and Related Models with Applications in Engineering and Science (San Francisco: John Wiley & Sons)

Castillo, E., Hadi, A.S., Balakrishnan, N., & Sarabia, J.M. 2004, Extreme Val- ues and Related Models with Applications in Engineering and Science (San Francisco: John Wiley & Sons)

work page 2004

[4] [4]

2011, MNRAS, 414, 2436

Colombi, S., Davis, O., Devriendt, J., Prunet„ J., & Silk, J. 2011, MNRAS, 414, 2436

work page 2011

[5] [5]

On the Explicit Height Distribution and Expected Number of Local Maxima of Isotropic Gaussian Random Fields

Cheng, D., & Schwartzman, A. 2015b, arXiv:1503.01328 [math.PR]

work page internal anchor Pith review Pith/arXiv arXiv

[6] [6]

2018, Bernoulli 4B(24), 3422

Cheng, D., & Schwartzman, A. 2018, Bernoulli 4B(24), 3422

work page 2018

[7] [7]

et al., 2017, MNRAS, 466, 861

Dunlop, J.S., McLure, R.J., Biggs, A.D. et al., 2017, MNRAS, 466, 861

work page 2017

[8] [8]

2013, Introduction to Mathematical Statistics (New York: Pearson)

Hogg, R.V ., McKean, J.W., & Craig, A.T. 2013, Introduction to Mathematical Statistics (New York: Pearson)

work page 2013

[9] [9]

2002, AJ, 123, 1086

Hopkins, A.M., Miller, C.J., & Connolly, A.J. 2002, AJ, 123, 1086

work page 2002

[10] [10]

2005, Journal of Statistical Physics, 119, 777

Majumdar, S.N., & Comtet, A. 2005, Journal of Statistical Physics, 119, 777

work page 2005

[11] [11]

1998, Fundamentals of Statistical Signal Processing: Detection The- ory (London: Prentice Hall)

Kay, S.M. 1998, Fundamentals of Statistical Signal Processing: Detection The- ory (London: Prentice Hall)

work page 1998

[12] [12]

Lehmann, E.L., & Romano, J.P., 2005, Testing Statistical Hypotheses (New: York: Springer)

work page 2005

[13] [13]

Miller, J.C., Genovese, C., Nichol, R.C. et al. 2001, AJ, 122, 3492

work page 2001

[14] [14]

Neyman, J., & Pearson, E.S., 1933, Phil. Trans. Roy. Soc. Ser. A231, 289

work page 1933

[15] [15]

Pavesi, R., Sharon, C.E., Riechers, D.A. et al. 2018, Ap.J., 864, 49P

work page 2018

[16] [16]

2004, A&A, 414, 17

Vio, R., Andreani, P., & Wamsteker, W. 2004, A&A, 414, 17

work page 2004

[17] [17]

2016, A&A, 589, A20 (V A16)

Vio, R., & Andreani, P. 2016, A&A, 589, A20 (V A16)

work page 2016

[18] [18]

2018, A&A, 604, A115 (VV A17)

Vio, R., Vergès, C., & Andreani, P. 2018, A&A, 604, A115 (VV A17)

work page 2018

[19] [19]

34854corr Table 1

Vitells, O., & Gross, E., 2001, Astroparticle Physics, 35, 230 Article number, page 5 of 15 A&A proofs: manuscript no. 34854corr Table 1. Identiﬁcation of the sources on the map in Fig. 7. Source number identiﬁcation coordinates RA DEC Source 1 UDF2 3h32m43.53s -27d46 ′39.25′′ Source 2 UDF3 3h32m38.55s -27d46 ′34.57′′ Source 3 UDF5 3h32m36.96s -27d47 ′27....

work page 2001