Why so young? A curious connection between the broadband flux and neutral gas content of AGN

Elaine M. Sadler; Elizabeth K. Mahony; Emily F. Kerrison; Vanessa A. Moss

REVIEW 3 major objections 6 minor 21 references

Why so young? A curious connection between the broadband flux and neutral gas content of AGN

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read An untargeted HI absorption survey finds that 21 of its 33 detections are young, peaked-spectrum radio AGN, with chance odds of one in 33 million.

desk verdict A short, honest proceedings paper with a striking but load-bearing statistical claim: the 21/33 PS excess among HI absorbers is only as strong as the unmeasured parent-sample PS fraction. read the letter →

arxiv 2412.08002 v1 pith:JBALNPM5 submitted 2024-12-11 astro-ph.GA

Emily F. Kerrison , Elaine M. Sadler , Vanessa A. Moss , Elizabeth K. Mahony This is my paper · ORCID

classification astro-ph.GA

keywords radioAGNpeaked-spectrumsourcesHI21-cmabsorptionASKAP-FLASHsurveyneutralhydrogenyounggalaxiesbroadbandSEDclassificationintermediate-redshiftgas

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Using the 33 neutral-hydrogen (HI) 21-cm absorption systems found by the untargeted FLASH pilot surveys, this paper finds that 21 of them have broadband radio spectra which peak at low frequencies, the signature of young radio AGN triggered within roughly the last thousand years. Since peaked-spectrum sources are usually estimated to make up only 10-20% of the radio population, the observed fraction is statistically extreme: under the 20% baseline the probability of seeing 21 or more is about 3 in 100 million. The paper concludes that untargeted HI absorption does not see a representative sample of radio galaxies; it preferentially detects compact, young sources with estimated linear sizes between 0.05 and 2.2 kpc. If true, 21-cm absorption becomes a discovery tool for young AGN at intermediate redshifts, and the neutral gas must be arranged in screens whose sizes are matched to the compact radio emission.

What carries the argument

The argument is carried by the radio broadband spectral energy distribution of each AGN, assembled from flux-density measurements between 80 MHz and 100 GHz and modelled with RADIO SED, a Bayesian fitting framework designed to identify young, peaked-spectrum sources. The peaked-spectrum classification is the load-bearing step: a turnover in flux density over several decades of frequency marks a source as young (typically within about $10^3$ years of triggering) rather than old or blazar-variable. The paper then maps each rest-frame peak frequency to an estimated linear size through the empirical correlation between turnover frequency and source size, and uses the HI line's position relative to the SED peak to set lower or upper limits on that size depending on whether the absorber is classified as associated or intervening.

What would settle it

Compute the peaked-spectrum fraction for the full set of continuum sources in the same pilot fields above the survey's $S_{855.5\,\mathrm{MHz}} > 40$ mJy limit; if that parent fraction is already close to 60%, the 21-out-of-33 result is not an HI-specific over-representation. Alternatively, high-resolution radio imaging that resolves most of the 21 peaked sources into extended kiloparsec-scale structures would falsify the compact-source, sub-kiloparsec size interpretation.

Watch

Extended reading notes

Core claim

The paper's central claim is that the galaxies against which neutral hydrogen is detected in an untargeted survey are a specific subset of the radio-AGN population: young, compact, peaked-spectrum sources. Among the 33 strongest HI absorbers from the FLASH pilot surveys, 21 statistically favour a peaked broadband model, whereas the literature fraction of peaked-spectrum sources is 10-20%; with a 20% assumed fraction the binomial probability of 21 or more peaked detections is $3\times 10^{-8}$, and it falls further under the lower estimate. The authors further note that the rest-frame spectral peaks tend to sit just below the 1.4 GHz frequency of the HI line, and they convert peak frequencies into linear sizes of 0.05-2.2 kpc (mean $0.6\pm0.5$ kpc) using the empirical peak-size relation. These small sizes match the compact-source preference seen in targeted HI samples and support the idea of a size resonance between the absorbing gas screen and the young radio source.

Load-bearing premise

The argument assumes that the 10-20% literature fraction of peaked-spectrum sources is the correct baseline for the specific flux-limited, 855.5 MHz parent sample from which these absorbers were drawn; if compact sources are already over-represented in that parent sample, the reported excess disappears.

Editorial extensions

If this is right

The full FLASH survey should confirm the effect: as the absorber sample grows to hundreds or thousands, the peaked-spectrum fraction should remain well above the 10-20% typical of the general radio population.
Untargeted 21-cm absorption searches can be used as a discovery channel for young, compact radio AGN at $0.4 < z < 1$, a redshift range where both Ly-alpha emission and 21-cm emission are hard or impossible to observe from the ground.
For associated absorbers, the HI line's position in the SED sets a lower limit on the rest-frame peak frequency and therefore an upper limit on the radio source's linear size, giving size information without VLBI imaging.
The lack of a strong correlation between HI line width and estimated linear size means line width is not governed simply by source size, so future samples with secure spectroscopic redshifts will be needed to separate geometry from gas kinematics.
The mean estimated linear size of about 0.6 kpc matches the compact-source preference seen in targeted HI samples, which the paper interprets as evidence that the size match between the gas screen and the radio source, rather than target selection, drives absorption detectability.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A direct test of the paper's baseline assumption would be to run the same SED classification on all continuum sources in the pilot fields above the survey's flux limit; if the parent peaked-spectrum fraction is already high, the apparent preference for young sources among HI absorbers would shrink or vanish.
The resonance interpretation predicts that high-resolution VLBI imaging of the 21 peaked sources will show compact double or compact-symmetric-object morphologies on sub-kiloparsec scales, and that HI absorption detection rates should peak for radio sources whose linear size matches the gas-screen scale.
If the effect is general rather than a FLASH selection effect, other untargeted 21-cm absorption surveys should show the same excess of peaked-spectrum sources once they accumulate comparable numbers of detections.
Since the sample mixes associated and intervening systems, the linear-size distribution is only firmly interpretable for the associated subset; secure optical redshifts will be required before the sizes can be compared with models of jet-gas interactions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The thing to know about this paper is that it reports a real observational result — 21 of 33 HI absorbers from the untargeted FLASH pilot are classified as peaked-spectrum sources — and that the binomial probability of 3e-8 they quote is correct given their assumptions. But the assumption doing the work is that the parent FLASH continuum sample, defined at S_855.5 > 40 mJy, has the same 10–20% PS fraction as the literature samples. They never measure that fraction, even though the survey detected over 10^4 continuum sources. The stress-test note is right: if the true parent fraction were 40%, the p-value drops to per-mille; if it were above ~50%, the excess vanishes. That is a genuine soft spot in the paper's central claim, and it is not a minor one.

What the paper does well: it is the first statistical statement of this effect in an untargeted sample, the SED fitting via RADIO SED is a legitimate tool, and the authors are transparent about their caveats — they flag the preliminary machine-learning classifications, the small sample, and the need for larger samples. The section on linear sizes is appropriately hedged, and the 'resonance' interpretation is presented as a suggestion, not a conclusion. For a proceedings paper, the level of honesty is good.

The soft spots are, in order: (1) the unmeasured parent baseline, which is the load-bearing assumption; (2) the derived linear sizes use the same peak-frequency measurements that define sources as young and compact, so there is some circularity in the size distribution claim; (3) the associated/intervening classification, which affects the size upper limits, is preliminary for most sources. None of these are fatal to the observation that something interesting is going on, but they do mean the headline 'do not resemble the parent population' is not yet fully supported.

Who gets value from this: the AGN/HI absorption community, especially people planning FLASH and SKA surveys. It is a useful pointer that untargeted HI searches may be biased toward young, compact sources, but it needs a proper measurement of the parent PS fraction before that becomes a solid conclusion.

My recommendation: this deserves a serious referee, not a desk reject, because the claim is important and the data are real. But the referee should ask for the parent-sample baseline or a clear qualification that the 3e-8 is conditional on an assumed literature fraction. I would accept it with that condition.

Referee Report

3 major / 6 minor

Summary. The paper reports on the broadband radio spectral properties of 33 sources detected in HI absorption in the ASKAP-FLASH Pilot Surveys, an untargeted search. Using multi-frequency flux densities and the RADIO SED Bayesian fitting framework, the authors classify 21/33 sources as peaked-spectrum (PS) sources. They compare this fraction with a literature value of 10-20% for PS sources among all radio galaxies, compute a binomial probability of 3e-8, and conclude that HI-absorbing galaxies are a specific, young subset of the radio population. They further use the rest-frame peak frequencies to estimate linear sizes of 0.05-2.2 kpc and discuss an apparent 'resonance' between the absorbing screen size and the radio source size. The paper is a short conference proceedings contribution.

Significance. If the central claim is correct, the result is scientifically valuable: it would demonstrate that untargeted HI absorption surveys preferentially select young, compact, peaked-spectrum radio AGN rather than a representative sample of the radio galaxy population. This has consequences for interpreting absorption-selected samples and for planning future all-sky surveys such as the full FLASH survey. The paper also contributes a methodological template for combining broadband SED classification with HI absorption spectroscopy. The sample itself, drawn from an untargeted survey with a clearly described linefinder and significance threshold, is a useful resource. However, the headline statistical claim currently rests on an unmeasured control population, and several secondary inferences are partly circular or rely on preliminary classifications, so the strength of the conclusions exceeds what the present analysis supports.

major comments (3)

[Section 2] The binomial calculation Pr(N_PS >= 21) = 3e-8 assumes that the peaked-spectrum fraction in the parent FLASH continuum sample at S_855.5 MHz > 40 mJy is 20%, taken from O'Dea (1998) and Callingham et al. (2017). The paper does not measure this fraction in the FLASH parent sample, even though the pilot surveys detected over 10^4 continuum sources. The literature baseline was derived at different frequencies, flux limits, and from heterogeneous samples, so it is not established to apply to the 855.5 MHz, 40 mJy flux-limited parent population. If the true parent PS fraction were substantially higher (e.g., ~50%), the claimed excess would disappear. The central statement that these absorbers 'do not resemble the parent population' therefore requires either a direct measurement of the parent PS fraction in the FLASH continuum catalogue or a substantially hedged interpretation. As written, the headline over-representation claim is not yet fully supported.
[Section 3] The linear-size estimates are derived from the rest-frame peak frequency of each source using the Jeyakumar (2016) relation. These peak frequencies are the same measurements used to classify the sources as peaked-spectrum in Section 2, so the derived linear sizes are not an independent check of the claim that HI absorbers are preferentially compact; a source classified as PS will, by construction, tend to have a small derived size. The statement in Section 3 that the sample 'supports the idea of a resonance between the size of the absorbing screen and the radio source' is therefore partly circular. To make this argument convincing, the authors should either compare the derived sizes with a control sample of non-absorbing PS sources from the same survey or clearly state that the size distribution is a consequence of the selection and does not by itself confirm the resonance interpretation.
[Section 3 / Figure 1] The right panel of Figure 1 and the associated discussion separate sources into associated and intervening line-of-sight classifications based on 'preliminary' machine learning from Curran et al. (2016). The authors themselves flag these classifications as preliminary, yet the interpretation of the linear-size/linewidth relation, and the upper limits placed on intervening systems, depend on this split. Given that the sample spans only 33 sources and the classifications are uncertain, the conclusion that 'the width of the absorption line profile is not strongly correlated with the linear size' is not robustly supported. The analysis would benefit from a sensitivity check that treats the uncertain classifications as a source of systematic error, or from explicit statements of how misclassification would change the plotted trends.

minor comments (6)

[Header] The running title 'Neutral Hydrogen in an around galaxies in the SKA era' appears to contain a typo; it should likely read 'in and around galaxies'.
[Section 1] The text states '3,000 deg2 of sky'; the unit should be typeset as deg^2 or 'square degrees' for consistency.
[Section 1] The description of the Bayesian linefinder says it 'identified highly significant absorption lines in each 288 MHz-wide spectrum'; it would be clearer to state that it identified lines in the spectra, not in each spectrum in a way that implies a line in every spectrum.
[Section 2] The symbol 'S855.5 MHz' should be typeset with the frequency as a subscript (S_855.5 MHz) to avoid confusion with a multiplication.
[Section 3] The phrase 'restframe peak' is used in the text and Figure 1; this should be 'rest-frame peak' for consistency with standard terminology.
[Section 3] The sentence 'the location of the spectral peaks in these sources suggest that they all have linear sizes less than 2 kpc' has a subject-verb agreement issue ('location ... suggest' should be 'suggests').

Circularity Check

1 steps flagged · score 4.0 of 10

Secondary compact-size/resonance conclusion reduces to the PS classification by construction; the central PS over-representation claim remains independent.

self definitional [Section 3 (Connecting the dots), paragraph beginning 'Going one step further...']
"It is well known that the peak of a PS source correlates closely with its linear size (e.g. O’Dea and Baum 1997; Jeyakumar 2016), so we here combine our estimates of rest frame peak frequencies with the empirical relation from Jeyakumar (2016) (especially Figure 3 of that work) to determine an approximate linear size for the PS sources in our sample. ... A simpler observation with regards to this sample is that these sources all have estimated linear sizes between 0.05 – 2.2 kpc, with a mean size of 0.6 ± 0.5 kpc."

The linear sizes are not measured independently; they are computed from the rest-frame peak frequency via the Jeyakumar (2016) size–peak-frequency relation. The same peak frequency is the defining property of the PS classification in Section 2, where sources are called peaked because of the characteristic turnover. Thus the statement that all sources have linear sizes 0.05–2.2 kpc is a direct transform of the classification variable, not a new measurement. Claiming this 'supports' the resonance hypothesis is therefore partly by construction: selecting PS sources already selects compact sources through the adopted relation. The over-representation claim (binomial test) is independent, but the resonance confirmation is not an independent test.

full rationale

The central quantitative claim—21/33 PS sources versus a 10–20% literature baseline, with binomial p=3e-8—is not circular: the baseline is external (O'Dea 1998; Callingham et al. 2017), and the PS classification from RADIO SED is an independent measurement applied to the HI-selected sample. The unmeasured PS fraction in the FLASH continuum parent sample is a legitimate statistical criticism, but it concerns the validity of the external baseline, not circularity. Self-citations (RADIO SED, Curran et al. 2013, FLASH survey papers) are normal and not load-bearing in a circular sense here. The one genuine circular step is in Section 3: linear sizes are derived from rest-frame peak frequencies via the Jeyakumar (2016) relation, and those same peak frequencies define the PS classification. Therefore the statement that all sources have sizes 0.05–2.2 kpc is a transform of the classification variable, and using it as support for the Curran et al. (2013) resonance is partly by construction. The over-representation claim, however, stands independently of this step, so the overall circularity is partial and confined to the interpretive layer.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central statistical claim rests on the external baseline fraction and on the RADIO SED classifications. The size and resonance interpretation additionally assume the Jeyakumar (2016) relation and the association of HI lines with the radio sources. No entirely new entities or ad hoc parameters are introduced in this proceedings paper; the free parameters are the per-source SED fits imported from the companion RADIO SED work.

free parameters (1)

Per-source SED model parameters (peak frequency, peak flux, spectral indices)
The RADIO SED Bayesian fitting (Kerrison et al. 2024) fits these parameters to each source's broadband flux density measurements; the classification as peaked-spectrum and the derived linear sizes depend on these fits. Values are not tabulated in this proceedings paper.

assumptions (4)

domain assumption The RADIO SED Bayesian framework correctly identifies peaked-spectrum sources in this sample.
Section 2: 'The resulting SED of each source was then run through RADIO SED...' and Section 3 uses the resulting classifications. The framework is described in Kerrison et al. 2024 (same author group), and is not independently validated here.
domain assumption The literature peaked-spectrum fraction of 10-20% (O'Dea 1998; Callingham et al. 2017) applies to the FLASH continuum parent sample with S_855.5 MHz > 40 mJy.
Section 2: 'Literature estimates suggest these comprise only 10-20% of all radio sources'. The paper uses this fraction as the baseline for the binomial test without measuring the fraction in the FLASH parent sample.
domain assumption The empirical peak-frequency versus linear-size relation of Jeyakumar (2016) is valid for these sources.
Section 3: 'combine our estimates of rest frame peak frequencies with the empirical relation from Jeyakumar (2016) to determine an approximate linear size'. The relation is taken from the literature and not verified for this sample.
domain assumption For sources classified as associated, the HI line and the radio source are at the same redshift, so the rest-frame peak frequency can be computed.
Section 3: 'assuming all cases of absorption are associated with the radio source'; the classification between associated and intervening uses machine learning (Curran et al. 2016) that the authors describe as preliminary.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Why so young? A curious connection between the broadband flux and neutral gas content of AGN." pith.science (2026). https://pith.science/paper/JBALNPM5

@misc{pith2026241208002,
  author       = {Pith},
  title        = {Pith review of: Why so young? A curious connection between the broadband flux and neutral gas content of AGN},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JBALNPM5}},
  note         = {Machine review of arXiv:2412.08002}
}

read the original abstract

We present here a study of the broadband spectral properties of 33 sources detected in HI absorption as part of the ASKAP-FLASH Pilot Surveys. We outline our approach to spectral classification and discuss the correlation seen between spectral shape and the detection of HI absorption. We further consider the implications of the observed correlation on the spatial distribution of the neutral gas, and on the jet-gas interactions. Our results are evaluated in the context of the forthcoming, full ASKAP-FLASH survey and other large, untargeted searches of the radio sky.

Figures

Figures reproduced from arXiv: 2412.08002 by the authors.

**Figure 1.** Left: the normalised SEDs of all 21 sources in our sample classified as peaked spectrum using RADIOSED. The vertical dashed line indicates the median position of all HI lines relative to the broadband SEDs, with the shaded band indicating the spread in this position using the interquartile range. Right: linewidth of the HI line profiles plotted against estimates of the linear size for each source, derived from its … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 20 canonical work pages

[1]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...

work page
[2]

1990, Monthly Notices of the Royal Astronomical Society , 246, 263--272

Subrahmanyan, R., Narasimha, D., Pramesh Rao, A., & Swarup, G. 1990, Monthly Notices of the Royal Astronomical Society , 246, 263--272

work page 1990
[3]

O'Dea, C. P. & Baum, S. A. 1997, The Astronomical Journal , 113, 148

work page 1997
[4]

N., de Bruyn, A

Chengalur, J. N., de Bruyn, A. G., & Narasimha, D. 1999, Astronomy and Astrophysics , 343, L79--L82

work page 1999
[5]

2000, A & AS , 143(1), 23--32

Ochsenbein, F., Bauer, P., Marcout, J., et al. 2000, A & AS , 143(1), 23--32

work page 2000
[6]

R., Sadler, E

Allison, J. R., Sadler, E. M., & Whiting, M. T. 2012, PASA , 29(3), 221--228

work page 2012
[7]

J., Allison, J

Curran, S. J., Allison, J. R., Glowacki, M., et al. 2013, Monthly Notices of the Royal Astronomical Society , 431(4), 3408--3413

work page 2013
[8]

& Dickinson, M

Madau, P. & Dickinson, M. 2014, ARA & A , 52, 415--486

work page 2014

Show all 21 references

[9]

R., Sadler, E

Allison, J. R., Sadler, E. M., Moss, V. A., et al. 2015, MNRAS , 453(2), 1249--1267

2015
[10]

J., Duchesne, S

Curran, S. J., Duchesne, S. W., Divoli, A., & Allison, J. R. 2016, MNRAS , 462(4), 4197--4207

2016
[11]

The MeerKAT Absorption Line Survey (MALS)

Gupta, N., Srianand, R., Baan, W., et al. The MeerKAT Absorption Line Survey (MALS) . In Proceedings of MeerKAT Science: On the Pathway to the SKA — PoS(MeerKAT2016) 2016,, Trieste, Italy. Sissa Medialab

2016
[12]

R., Ekers, R

Callingham, J. R., Ekers, R. D., Gaensler, B. M., et al. 2017, ApJ , 836(2), 174

2017
[13]

R., Moss, V

Allison, J. R., Moss, V. A., Macquart, J.-P., et al. 2017, Monthly Notices of the Royal Astronomical Society , 465(4), 4450--4467

2017
[14]

M., Moss, V

Sadler, E. M., Moss, V. A., Allison, J. R., et al. 2020, MNRAS , 499(3), 4293--4311

2020
[15]

O’Dea, C. P. & Saikia, D. J. 2021, A & ARv , 29(1), 1--109

2021
[16]

R., Sadler, E

Allison, J. R., Sadler, E. M., Amaral, A. D., et al. 2022, PASA , 39(1), 010--1

2022
[17]

G., Mobasher, B., et al

Pacifici, C., Iyer, K. G., Mobasher, B., et al. 2023, The Astrophysical Journal , 944(2), 141

2023
[18]

F., Allison, J

Kerrison, E. F., Allison, J. R., Moss, V. A., et al. 2024, Monthly Notices of the Royal Astronomical Society , 533(4), 4248--4267

2024
[19]

M., Mahony, E

Yoon, H., Sadler, E. M., Mahony, E. K., et al. 2024, arXiv:2408.06626

2024 arXiv
[20]

2016, Monthly Notices of the Royal Astronomical Society , 458(4), 3786--3794

Jeyakumar, S. 2016, Monthly Notices of the Royal Astronomical Society , 458(4), 3786--3794

2016
[21]

O'Dea, C. P. 1998, PASA , 110(747), 493--532

1998

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...

work page

[2] [2]

1990, Monthly Notices of the Royal Astronomical Society , 246, 263--272

Subrahmanyan, R., Narasimha, D., Pramesh Rao, A., & Swarup, G. 1990, Monthly Notices of the Royal Astronomical Society , 246, 263--272

work page 1990

[3] [3]

O'Dea, C. P. & Baum, S. A. 1997, The Astronomical Journal , 113, 148

work page 1997

[4] [4]

N., de Bruyn, A

Chengalur, J. N., de Bruyn, A. G., & Narasimha, D. 1999, Astronomy and Astrophysics , 343, L79--L82

work page 1999

[5] [5]

2000, A & AS , 143(1), 23--32

Ochsenbein, F., Bauer, P., Marcout, J., et al. 2000, A & AS , 143(1), 23--32

work page 2000

[6] [6]

R., Sadler, E

Allison, J. R., Sadler, E. M., & Whiting, M. T. 2012, PASA , 29(3), 221--228

work page 2012

[7] [7]

J., Allison, J

Curran, S. J., Allison, J. R., Glowacki, M., et al. 2013, Monthly Notices of the Royal Astronomical Society , 431(4), 3408--3413

work page 2013

[8] [8]

& Dickinson, M

Madau, P. & Dickinson, M. 2014, ARA & A , 52, 415--486

work page 2014