{"id":"55d465fc-8bac-44b7-b545-27271197260e","arxiv_id":"1908.05316","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":11,"one_line_summary":"Stacking FIRST maps at SDSS quasar positions lets the authors infer the 1.4-GHz quasar luminosity function about two orders of magnitude below the survey limit, revealing a faint-end break near L ~ 10^25 W/Hz.","lead":"By stacking radio maps at the positions of tens of thousands of optically selected quasars, this paper reconstructs the quasar radio luminosity function down to roughly one hundredth of the usual detection limit. The faint-end shape shows a flattening and a steepening that may trace where host-galaxy star formation begins to dominate the radio emission.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The faint-end flattening and steepening is not securely intrinsic: the SDSS optical magnitude limit can imprint it, as the authors themselves note, so the central RLF shape and star-formation coincidence need an optical-radio selection test.","rationale":"The reader's weakest_assumption identifies the same load-bearing point: the SDSS optical flux limit can imprint the faint-end RLF shape. I agree and find the paper's own text supports this. The simulation tests in Section 4 validate the Bayesian stacking machinery for a sample cut in radio luminosity, but they do not validate the recovery of an intrinsic RLF from an optically flux-limited sample with a correlated optical-radio distribution. The test the authors ran (raising the optical magnitude limit makes the turnover more prominent) actually moves in the direction expected if the selection is the cause of the turnover. Consequently, the abstract's flattening and steepening luminosities, and the claimed coincidence with star-forming galaxy domination, are not yet established as intrinsic. This does not make the paper unsound: the method is sensible, the low-redshift comparison with Kellermann et al. is encouraging, and the authors are explicit about the caveat. It does mean the central physical claim is conditional, exactly as the reader concluded. No separate internal inconsistency in the likelihood or evidence computation was identified; the concern is the mapping from the selected sample to the intrinsic RLF. The concrete test above would settle whether the headline shape survives a realistic optical selection function.","tokens_in":25826,"tokens_out":8781,"duration_ms":95399,"concrete_test":"Forward-model the selection: draw a mock SDSS quasar sample from a bivariate optical-radio luminosity function obeying the White et al. (2017) correlation with its ~1 dex scatter and an input RLF with no flattening or steepening break (e.g. single or double power law over the full range); apply the SDSS i<19.1 limit and the per-bin Mi cut in Eq. 1; add Gaussian FIRST noise with sigma=150 uJy; then run the same bayestack pipeline with Models A-C. If the recovered RLF reproduces the flattening near log10[L1.4/W Hz^-1] ~ 25.5 and steepening below ~24.8, the observed features are selection artifacts and the star-formation coincidence is unsupported. If the recovered RLF remains featureless, the optical-limit concern is resolved in favour of the paper's interpretation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is the sub-mJy RLF shape: flattening below log10[L1.4/W Hz^-1] ~ 25.5 and steepening below ~24.8, with the luminosity coincidence motivating star formation. For that claim to stand, the observed break must reflect the radio luminosities of optically selected quasars rather than the SDSS optical flux limit. The paper does not establish this. Eq. 1 imposes a hard Mi cut in each redshift bin, and the parent catalogue is i<19.1 limited; White et al. (2017) find a correlation between optical and radio luminosity with roughly an order of magnitude scatter. That correlation plus the magnitude cut can differentially remove low-optical-luminosity quasars, whose radio luminosities populate exactly the regime where the flattening and steepening appear. The authors' Fig. 8 places the White et al. optical-radio limit just above the turnover, and Sec. 6.1 states that 'at least some of the flattening is due to incompleteness introduced by the optical magnitude limit of the parent sample.' Their robustness test - raising the optical limit makes the turnover more prominent - is consistent with selection imprinting the shape rather than a physical break. The SKADS validation does not remove this concern because the simulated samples are cut in radio luminosity, not selected by an optical flux limit correlated with radio luminosity. The abstract states the flattening and steepening luminosities without uncertainties, and the star-formation interpretation rests on the luminosity coincidence. The central claim is therefore conditional on a bivariate optical-radio selection model that is not implemented.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper applies a Bayesian stacking technique (bayestack) to 1.4-GHz FIRST flux densities extracted at the positions of SDSS DR7 quasars, fitting parametric models for the radio luminosity function (RLF) in seven redshift bins up to z=2.15. The authors claim to reconstruct the optically selected quasar RLF down to roughly two orders of magnitude below the FIRST 1 mJy detection threshold, and report that the bright-end RLF flattens below log10[L1.4/W Hz^-1] ~ 25.5 and steepens below ~24.8, where radio-quiet quasars and a possible star-formation contribution emerge. The method is tested on SKADS simulations and the lowest-redshift RLF is compared with deeper VLA data from Kellermann et al. (2016) and Condon et al. (2013).","tokens_in":26235,"tokens_out":4074,"duration_ms":42902,"significance":"If the reconstruction is reliable, this is a valuable technique for measuring the faint radio emission of optically selected quasars using wide, shallow surveys, and the deep low-z agreement with Kellermann et al. (2016) is an encouraging validation. The authors are careful to test the pipeline on SKADS simulations with realistic FIRST noise and to compare with independent data at low redshift; these are genuine strengths. However, the central physical claims about the RLF shape below the detection threshold rest on parametric model assumptions and on a subtle treatment of optical selection, and the paper itself acknowledges that at least part of the flattening may be caused by the optical magnitude limit. Because the claimed star-formation coincidence depends on the shape of the reconstructed faint end, the significance of that interpretation is not yet established.","major_comments":[{"comment":"The claimed flattening at log10[L1.4/W Hz^-1] ~ 25.5 and the steepening below ~24.8 are not independent measurements but are properties of the winning double power-law model (Model B, Eq. 13): the turnover locations are essentially the break luminosities L*1 and L*2 of the fitted functions. The model-selection evidence in Table 4 only ranks Models A, B, and C relative to each other; it does not demonstrate that a turnover is required by the data, for example by testing against a model without a low-luminosity break or by computing an absolute goodness of fit. As a result, the abstract's statement of these transition luminosities, without associated uncertainties and without an explicit statement that they are posterior model parameters, is stronger than the analysis supports.","section":"§5, Table 4, Eqs. 12–14"},{"comment":"The optical selection can imprint the very features the paper interprets physically. The sample is limited by i<19.1 and each redshift bin is truncated at a maximum absolute magnitude (Eq. 1); if optical and radio luminosities are correlated (as in White et al. 2017, with roughly an order of magnitude scatter), this cut differentially removes low-optical-luminosity quasars whose radio luminosities populate the regime where the flattening and steepening appear. The authors themselves state in Sec. 6.1 that 'at least some of the flattening is due to incompleteness introduced by the optical magnitude limit of the parent sample' and that they 'cannot rule out' an optical-selection origin for the bright-end flattening. Their robustness test, raising the optical limit and finding that the turnover becomes more prominent, is fully consistent with selection imprinting the shape rather than with a physical break. The SKADS validation does not address this issue because the simulated samples are cut in radio luminosity, not by an optical flux limit correlated with radio luminosity. To support the physical interpretation, the paper needs a forward-model test that injects an optical-radio correlation and an optical flux limit into the simulations and quantifies how much flattening/steepening is produced.","section":"§2.1 (Eq. 1) and §6.1"},{"comment":"The White et al. (2007) bias correction (Eq. 2) was derived for sources above or near the detection threshold where noise can be neglected, and the authors correctly note this in the text. Applying this correction per source inside the likelihood, via SF = max{S/1.4, S - 0.25 mJy}, to noise-dominated sub-threshold flux densities is an extrapolation that is not validated by the SKADS simulations, which set S = SF and therefore do not include the clean or snapshot biases. Because the faint end of the RLF is precisely where this correction is applied, a mis-modeled bias could artificially produce or modify the apparent steepening below log10[L1.4/W Hz^-1] ~ 24.8. The paper should either justify the inverse correction for sub-threshold sources with simulations that include these biases or demonstrate that the results are robust to the assumed bias model.","section":"§3.2, Eq. 2 and §4"},{"comment":"The parameters that most directly control the low-luminosity shape are poorly constrained. Table 5 shows that the boundary parameters Lmin1, Lmax1, Lmin2, and Lmax2 are largely unconstrained, with many posterior intervals spanning several decades, and the appendix states that the faint-end slope beta2 is not well constrained. The authors argue that these 'have very little impact on the actual observed numbers,' but the claimed values of the flattening and steepening luminosities, and the integrated radio-quiet fractions in Table 1, are derived from these posterior components. The paper should quantify how the uncertainty in these boundary and slope parameters propagates into the claimed turnover locations and into the conclusion that the RLF 'peaks' and then 'drops rather abruptly,' rather than only presenting the MAP reconstruction.","section":"§5.1, Table 5, and Appendix Fig. A0"}],"minor_comments":[{"comment":"The description of the Bayes factor contains a typo: 'ln[ZB−ZA]' should be 'ln(ZB/ZA)'.","section":"§3.1"},{"comment":"Several references are incomplete in the bibliography: Chen et al. (2017) is listed only as an arXiv preprint, Gürkan et al. (2018) has no journal or arXiv identifier, and Richards (2006) is listed as 'ArXiv Astrophysics e-prints.'","section":"References"},{"comment":"The phrase 'the sources are volume-limited in the optical (i.e no brightness cutoff)' is misleading because the sample does have an absolute-magnitude cut (Eq. 1); the intended meaning is that there is no radio brightness cutoff, but the wording should be clarified.","section":"§5.1"},{"comment":"The blue dotted and red dashed lines in Fig. 8 are described as 'an estimate of the radio-luminosity limit that corresponds to the optical limit' based on White et al. (2017); the caption should explicitly state that these are not measured limits but model-dependent extrapolations of the optical-radio correlation, given the scatter in that relation.","section":"Fig. 8 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a potentially useful method and an honest discussion of its limitations, but the abstract and conclusions currently present the RLF shape below 1 mJy as a robust observational result despite the absence of a selection-function test and the admitted optical-incompleteness degeneracy. In revision, the authors should either add a forward-model test that includes an optical-radio correlation and the optical flux limit, or substantially weaken the physical interpretation in the abstract and Section 5. I would also encourage them to provide the posterior constraints on the claimed turnover luminosities rather than point estimates."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nYou should know two things about this one. First, it is a competent application of an existing stacked-Bayesian fitting method (bayestack) to a new sample: the sub-mJy 1.4-GHz RLF of optically selected SDSS quasars, across seven redshift bins down to two orders of magnitude below FIRST's threshold. Second, the headline shape—the flattening below log L ~25.5 and the steepening below ~24.8—is less secure than the abstract implies, because the authors themselves show in Sec 6.1 that part of it can be an artifact of the optical flux limit.\n\nWhat is genuinely new is the measurement itself: applying the method to uniformly selected SDSS DR7 quasars, and getting a lowest-redshift bin that agrees with the deeper Kellermann et al. data. That agreement is the strongest evidence the pipeline works. The SKADS tests are also useful, though they do not settle the selection question.\n\nThe main soft spot is exactly the one flagged in the stress test: the faint-end shape is an output of the fitted double power-law (Model B), not an independent measurement. Changing the functional family would move the features. The authors are honest about this—they write that 'at least some of the flattening is due to incompleteness introduced by the optical magnitude limit'—and their robustness test, where raising the optical threshold makes the turnover more prominent, goes the same direction. The SKADS validation cuts in radio luminosity, not through an optical-radio correlation, so it cannot rule out this imprint. On top of that, the boundary parameters Lmin/Lmax are largely unconstrained, the White et al. (2007) bias correction is extrapolated to undetected sources, and the abstract's transition luminosities come without uncertainties. No code or parameter files are released.\n\nNone of this is fatal. The paper is a legitimate step forward, and the authors themselves point to the proper fix—a bivariate optical-radio luminosity function—which is what a referee should push them toward. I'd send it to review. A serious referee can handle the caveats.","headline":"Competent application of bayestack to a new sample, but the sub-mJy RLF shape is conditional on optical selection effects the authors themselves flag.","tokens_in":26853,"tokens_out":6300,"would_cite":true,"duration_ms":56916,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Bayesian stacking of FIRST pixels at SDSS quasar positions measures the quasar radio luminosity function to about 100 times fainter than the survey limit, exposing a flattening then steepening where radio-quiet quasars emerge.","keywords":["radio luminosity function","quasars","Bayesian stacking","FIRST survey","SDSS","radio-quiet quasars","sub-mJy radio sources","1.4 GHz continuum"],"falsifier":"Run the same Bayesian stacking fit on a quasar sample with a fainter optical magnitude limit or with relaxed absolute-magnitude cuts over the same redshift bins: if the flattening near $\\log_{10}[L_{1.4}/\\mathrm{W\\,Hz}^{-1}] \\approx 25.5$ and the steepening near $\\approx 24.8$ stay fixed in luminosity, the shape is physical, whereas if they shift or disappear the optical selection created them.","tokens_in":25598,"feed_emoji":"📡","tokens_out":7964,"duration_ms":70974,"temperature":0.7,"pith_summary":"This paper claims that the 1.4-GHz radio luminosity function (RLF) of optically selected quasars can be measured down to about two orders of magnitude below the 1 mJy detection limit of the FIRST survey by fitting parametric models to the flux densities extracted from FIRST maps at SDSS quasar positions using a Bayesian stacking approach. Reconstructed over seven redshift bins out to $z = 2.15$, the RLF of radio-loud quasars flattens below $\\log_{10}[L_{1.4}/\\mathrm{W\\,Hz}^{-1}] \\approx 25.5$ and steepens again below $\\approx 24.8$, where radio-quiet quasars emerge. The steepening luminosity coincides with the luminosity where star-forming galaxies are expected to begin dominating radio source counts, suggesting that host-galaxy star formation may contribute substantially to faint quasar radio emission. The authors explicitly note that at least part of the flattening could be an artifact of the SDSS optical magnitude limit rather than a physical break, and this is the key uncertainty in the interpretation.","feed_headline":"Stacking recovers quasar radio luminosity 100x below survey limit","feed_subtitle":"Bayesian fit to FIRST pixels at SDSS quasar positions maps the faint end where radio-quiet quasars emerge.","key_machinery":"The central mechanism is a forward-modeled Poisson likelihood for binned FIRST pixel flux densities. A parametric radio luminosity function is converted to a source-count model, the FIRST clean and snapshot biases are applied, and the predicted counts are convolved with Gaussian noise before being compared to the histogram of extracted pixel values at SDSS quasar positions. This lets sources buried in the 150-$\\mu$Jy noise constrain the luminosity function rather than being averaged into a single stacked flux. The faint-end models tested are a power law, a double power law, and a log-normal power law, with the double power law winning by Bayesian evidence in all redshift bins.","core_discovery":"The central claim is that a full Bayesian stacking analysis of FIRST survey pixels can recover the 1.4-GHz radio luminosity function of optically selected quasars two orders of magnitude below the 1 mJy detection threshold, and that the recovered RLF has a distinctive faint-end shape. In every redshift bin from $0.2 < z < 2.15$, the data prefer a model with a double power-law for both the luminous and the faint populations. The bright-end function rises steeply toward lower luminosity, flattens near $\\log_{10}[L_{1.4}/\\mathrm{W\\,Hz}^{-1}] \\approx 25.5$, and then steepens again below $\\approx 24.8$, the regime where the authors associate the population with radio-quiet quasars. The agreement of the low-redshift reconstruction with deep JVLA observations of the same quasars, together with the coincidence between the steepening luminosity and the expected crossover to star-forming galaxy dominance, is presented as evidence that host-galaxy star formation may contribute to the radio emission of quasars; the paper explicitly notes that at least part of the flattening could instead be imposed by the SDSS optical magnitude limit.","pith_inferences":["Applying the same forward-model stacking to a parent quasar sample with a fainter optical magnitude limit would directly test whether the flattening luminosity tracks the optical limit or stays fixed, distinguishing a selection artifact from a true physical break.","A bivariate optical-radio luminosity function, a route the authors mention, would remove the need for hard absolute-magnitude cuts and could separate accretion-driven from star-formation-driven radio emission more cleanly than the RLF alone.","High-resolution imaging that resolves the host-galaxy scale would allow the extended, star-formation-related radio component to be separated from the AGN core, giving a direct test of the star-formation interpretation.","The same noise-dominated pixel fitting could measure luminosity functions of other faint populations in current and future wide surveys, since it extracts shape information from sources buried below the detection threshold without needing deep images."],"forward_implications":["A wide, shallow survey like FIRST is sufficient to constrain the faint end of the optically selected quasar RLF down to radio luminosities around $10^{22}\\,\\mathrm{W\\,Hz}^{-1}$, roughly two orders of magnitude below the nominal detection threshold.","The double power-law description of the faint population is preferred over a power law or a log-normal in all seven redshift bins, so the faint-end shape is not confined to a single redshift range.","The flattening near $\\log_{10}[L_{1.4}/\\mathrm{W\\,Hz}^{-1}] \\approx 25.5$ and the steepening near $\\approx 24.8$ recur in every bin, indicating that the same population change is present at all redshifts studied.","The steepening luminosity coincides with where star-forming galaxies are expected to dominate radio source counts, so host-galaxy star formation may contribute significantly to radio-quiet quasar emission.","In the lowest redshift bin the sub-mJy RLF matches deep JVLA observations of individual quasars, validating the stacking reconstruction against direct detections."],"supporting_citations":[{"why":"Supplies the fully Bayesian stacking framework that the paper modifies and extends to fit radio luminosity functions.","marker":"Zwart et al. (2015b)"},{"why":"Established likelihood-based fitting of source-count models to stacked flux-density measurements.","marker":"Mitchell-Wynne et al. (2014)"},{"why":"Pioneered fitting luminosity-function models to stacked sources, defining the approach used here.","marker":"Roseboom & Best (2014)"},{"why":"Provides the FIRST clean and snapshot bias correction incorporated into the forward model.","marker":"White et al. (2007)"},{"why":"Gives the optical-radio luminosity relation used to estimate where the SDSS optical limit could make the RLF incomplete.","marker":"White et al. (2017)"},{"why":"Deep JVLA observations of the same quasars supply the independent low-redshift RLF against which the reconstruction is checked.","marker":"Kellermann et al. (2016)"},{"why":"The NVSS-based RLF is the literature comparison for the lowest redshift bin, with resolution differences explaining the discrepancies.","marker":"Condon et al. (2013)"},{"why":"SKADS simulated skies provide the test data used to validate the reconstruction before application to real data.","marker":"Wilman et al. (2008, 2010)"}],"fun_headline_variants":["Bayesian stack digs 100x deeper into quasar radio sky","Quasar radio function mapped 100x below FIRST limit","Faint quasar radio glow recovered by stacking","Stacking unveils quasar radio faint end, hints at star formation","Radio-quiet quasars emerge from stacking analysis"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the SDSS optical magnitude limit, after the per-bin absolute magnitude cut, does not itself imprint the observed flattening and steepening on the radio luminosity function; the paper concedes that at least some of the flattening could be due to optical incompleteness.","fun_headline_variants_meta":{"raw":{"variants":["Bayesian stack digs 100x deeper into quasar radio sky","Quasar radio function mapped 100x below FIRST limit","Faint quasar radio glow recovered by stacking","Stacking unveils quasar radio faint end, hints at star formation","Radio-quiet quasars emerge from stacking analysis"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000905,"raw_usage":{"total_tokens":3979,"prompt_tokens":1119,"completion_tokens":2860,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":735,"completion_tokens_details":{"reasoning_tokens":2778}},"tokens_in":735,"tokens_out":2860,"duration_ms":20985,"temperature":1.0,"reasoning_tokens":2778,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:18:12.364929+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same Bayesian stacking fit on a quasar sample with a fainter optical magnitude limit or with relaxed absolute-magnitude cuts over the same redshift bins: if the flattening near $\\log_{10}[L_{1.4}/\\mathrm{W\\,Hz}^{-1}] \\approx 25.5$ and the steepening near $\\approx 24.8$ stay fixed in luminosity, the shape is physical, whereas if they shift or disappear the optical selection created them.","supporting_citations":[{"cited_title":"G., Afonso J., Jarvis M","cited_arxiv_id":null,"evidence_quote":"Established likelihood-based fitting of source-count models to stacked flux-density measurements."}],"review_version":1}