{"id":"147c113a-df05-4ce0-b7c1-613a4bc5810a","arxiv_id":"2412.03143","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"In 19 nearby star-forming galaxies, low-frequency radio turnovers are best explained by free-free absorption and ionisation losses in small, recent starburst regions, with no predictive global galaxy property.","lead":"Astronomers used new radio observations from 70 MHz to 17 GHz to model the shapes of 19 nearby star-forming galaxies' radio spectra. They find that low-frequency turnovers are not linked to galaxy-wide properties like mass, merger state, or orientation, pointing to dense gas in individual starburst regions as the cause.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The FFA attribution rests on Bayes factors computed with priors derived from the same data's posteriors; this circular prior invalidates the evidence comparison and can bias preference for FFA_ models.","rationale":"Read in good faith, the paper's goal is to infer the physical causes of low-frequency turnovers from radio SEDs. The strongest claim requires three conditions: (i) LFTOs are real; (ii) FFA/ionisation models are genuinely preferred over simpler power laws; and (iii) no global property predicts LFTO occurrence. Condition (iii) is supported by null results but only with a sample of 19 galaxies, as the reader noted; the more decisive problem is condition (ii). The evidence values in Table 4 are computed with priors derived from EMCEE posteriors of the same data. Since nested-sampling evidence is the integral of the likelihood over the prior, choosing the prior after seeing the data destroys the meaning of the Bayes factor and specifically collapses the usual penalty for adding a turnover parameter. The paper's Section 4.2.3 acknowledges dependence on prior volume but does not draw the consequence that the ln(Delta Z) values are not proper Bayesian evidences. This is an internal methodological inconsistency rather than a disagreement with consensus. The ATCA data reduction, modular model construction, and honest reporting of which sources fail to support LFTOs are real strengths, and the visual selection provides some independent support for the existence of low-frequency flattening. However, the causal interpretation in the abstract rises or falls on whether the FFA_ preferences survive a re-computation with fixed priors and full GLEAM covariance. If they do not, the conclusion must be recast as a hypothesis; if they do, the paper's provisional interpretation stands. The reader's conditional verdict remains appropriate, and no change to it is needed.","tokens_in":45964,"tokens_out":9308,"duration_ms":92837,"concrete_test":"Re-run DYNESTY model selection for the 11 LFTO galaxies (and the 8 controls for calibration) with fixed, independently specified priors: A/B/C/D log-uniform over a wide range fixed before seeing the data, alpha ~ U(-1.8, -0.2), nu_t,1 ~ U(10,300) MHz, nu_t,2/nu_b ~ U(300 MHz,17 GHz), and with the five GLEAM sub-bands combined using their known covariance rather than weighted averages. Record how many sources still prefer an FFA_ prefix model over PL by ln(Delta Z) > 3. If that count is materially below 6, the FFA attribution and the central LFTO interpretation fail; if it remains at least 6, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The weakest load-bearing point is not only the GLEAM photometry itself but the evidence comparison used to decide which SEDs have LFTOs. In Section 4.2.3, the DYNESTY prior for each model is set to the 1st-99th percentile of the same data's EMCEE posterior, with the justification that evidence depends on prior volume. This makes the reported ln(Delta Z) values in Table 4 data-dependent in a circular way: the Occam penalty from prior volume is effectively removed for any parameter whose posterior is well constrained. For FFA_PL versus PL, the extra turnover parameter nu_t is assigned a prior equal to its own posterior width, so the comparison is no longer a Bayes factor between the model and a fixed physical prior. The paper's six FFA_-preferred LFTO galaxies, and hence the central FFA/ionisation-loss interpretation, rest on this invalid evidence scale. The photometric issue flagged by the reader is real and additive, but fixing it alone would not validate the current model-selection test. In addition, ionisation losses are not included in the fitted model family, so the 'combination of FFA and ionisation losses' in the strongest claim is not directly tested; the immediate test is whether the FFA preference itself survives a valid evidence computation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents new ATCA radio continuum observations between 5.5 and 17 GHz for 19 nearby (z < 0.04) star-forming galaxies, combining them with GLEAM, SUMSS/NVSS, and RACS data to model rest-frame radio SEDs from 70 MHz to 17 GHz. The authors construct modular Bayesian models combining power-law synchrotron and free-free emission with free-free absorption (FFA) and synchrotron/inverse-Compton loss components, and use nested-sampling evidence values to select preferred models. They find that synchrotron-only power-law models are generally preferred, that six of the 11 initially selected LFTO galaxies favour FFA-prefix models, and that LFTO occurrence is decoupled from most global galaxy properties. They interpret LFTOs as arising from recent compact starburst regions rather than from global galaxy orientation or integrated ISM properties.","tokens_in":46362,"tokens_out":3285,"duration_ms":35285,"significance":"If the model-selection results are valid, this is a valuable dataset and analysis: it extends LFTO studies from luminous/ultraluminous infrared galaxies to a sample of nearby, lower-luminosity SFGs, and its finding that LFTOs do not correlate with global properties would be an important constraint on the physical origin of low-frequency radio curvature. The paper is generally careful in describing the sample selection, data processing, flux-density measurements, and the modular model family, and it presents tables of measured fluxes, fitted parameters, emission measures, and T-test results that will be useful to the community. However, the central model-selection step is compromised by a circular prior construction, and the paper's headline physical interpretation invokes ionisation losses that are not included in the fitted models. The significance of the results is therefore conditional on correcting the evidence computation and reframing the conclusions accordingly.","major_comments":[{"comment":"The model-selection evidence values are not valid Bayes factors because the DYNESTY priors are set to the 1st–99th percentiles of the EMCEE posteriors from the same data. This makes the prior volume data-dependent and removes the Occam penalty that is an essential part of an evidence comparison. For the FFA_PL versus PL comparison, the extra turnover parameter ν_t,1 is assigned a prior equal to its own posterior width, so the resulting ln(ΔZ) values in Table 4 are not comparisons against a fixed, physically motivated prior. The identification of six FFA_-preferred LFTO galaxies (Section 5.1) rests on this invalid evidence scale. The authors need to re-run the nested-sampling model selection with fixed priors independent of the posteriors (for example, using the priors described in Section 4.2.2), and report whether the preference for FFA_ prefix models survives. Without this fix, the central claim that a subset of the sample exhibits modelled LFTOs is not supported.","section":"§4.2.3 and Table 4"},{"comment":"The paper concludes that LFTOs are \"likely caused by a combination of FFA and ionisation losses,\" but ionisation losses are not part of the fitted model family. The models in Section 4.1 include FFA prefix components and synchrotron/inverse-Compton loss suffixes, and Section 6.1 explicitly states that ionisation losses were not included because their spectral shape is unconstrained. The data therefore do not directly test the ionisation-loss hypothesis; they can only test whether FFA-prefix models are preferred over the alternatives considered. The interpretive claim in the Abstract and Section 7 overreaches the modelling. The authors should either restrict the conclusion to FFA absorption or add an ionisation-loss model and perform the model comparison with that family included.","section":"Abstract and §6.1"},{"comment":"The LFTO sample selection and the fitted turnover frequencies depend on GLEAM low-frequency photometry, whose uncertainties are not propagated as conservatively as those of the other bands. Section 2.1 notes that 54 catalogue sources have negative sub-band flux densities, and Section 3.2.1 adds a 10% error in quadrature to non-GLEAM fluxes only, with no equivalent allowance for under-estimated GLEAM errors. If GLEAM flux uncertainties are larger than adopted, the preference for FFA_ prefix models and the derived ν_t,1 values could be artefacts of photometric noise. I request a robustness test: repeat the fitting and model selection with inflated GLEAM uncertainties (or with a 10% error added to GLEAM fluxes), and show that the identification of LFTO-containing sources and the fitted turnover frequencies are stable. This is a necessary control for a claim that hinges on low-frequency spectral curvature.","section":"§2.1 and §3.2.1"}],"minor_comments":[{"comment":"There are frequent formatting issues such as \"A TCA\" with a space, \"V oelk\" for Voelk, and inconsistent use of subscripts (e.g., αA vs α_ATCA). A thorough proofread for spacing and symbol consistency is needed.","section":"Throughout"},{"comment":"The caption for Figure 3 and footnote (a) of Table 5 state that PL_SIC α values are adjusted by +0.25; this adjustment should be stated in the text near the first use of the α_model–stellar mass correlation, and the unadjusted values should be reported somewhere for reproducibility.","section":"Table 5 and Figure 3"},{"comment":"The column header in Table 11 says \"ATCA Frequency (GHz)\" for the first column, but the columns appear to list frequencies and bandwidths; the table would be clearer with explicit sub-headers for each band's frequency and bandwidth.","section":"Table 11"},{"comment":"The discussion of the qFIR separation between LFTO and control samples would benefit from a statement of which individual sources drive the separation, given the small sample size and the outlier GLEAM J003652-333315 being removed from the analysis.","section":"Section 6.2"}],"recommendation":"major_revision","confidential_remarks":"The decisive issue is the circular prior construction in Section 4.2.3. If the authors re-run the evidence computation with fixed priors and the FFA preference does not survive, the paper's central claim collapses, in which case it should be rejected. If it does survive, the paper is publishable after the ionisation-loss overreach is removed and a GLEAM-uncertainty robustness test is added. The manuscript is within the scope of PASA and the data are valuable, but the model-selection step must be made statistically valid."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this if you care about radio SED modelling of star-forming galaxies. The genuinely new thing is the ATCA 5.5-17 GHz data for 19 nearby, lower-SFR galaxies, which extends the LFTO work from LIRGs/ULIRGs to more normal systems. The paper is honest about its limitations, and the modular model suite is a reasonable adaptation of Galvin et al. (2018). The null result that LFTO occurrence does not track global properties (stellar mass, SFR, inclination) is interesting and, as a comparison of two small samples, reasonably supported by the T-tests.\n\nThe soft spot is in the model selection. In Section 4.2.3, the DYNESTY priors are set to the 1st-99th percentile of the same data's EMCEE posteriors. That removes the Occam penalty the evidence comparison is supposed to supply. The Bayes factors in Table 4 are therefore not fair comparisons between models with and without an FFA turnover. The six FFA_-preferred LFTO galaxies, and thus the central claim that FFA is responsible for the curvature, rest on this invalid evidence scale. The GLEAM photometric uncertainties are a separate and additive concern; fixing them alone won't fix the prior problem.\n\nI also note that ionisation losses are not included in the fitted model family, so the conclusion that LFTOs are caused by \"a combination of FFA and ionisation losses\" is not directly tested. The immediate test is whether the FFA preference survives a valid evidence computation. Minor issues: Haro 11 is excluded from the statistics without a clear note, and the abstract says a sample of 20 while the table has 19.\n\nBottom line: the data are worth having, and the paper deserves a serious referee, but the evidence comparison needs to be redone with priors that are not derived from the same data's posteriors (or with a different model-selection approach). I'd recommend sending it to review with major revision required.","headline":"New ATCA data and a useful sample extension, but the evidence-based model selection is compromised by data-derived priors, so the FFA attributions need re-analysis.","tokens_in":46849,"tokens_out":2327,"would_cite":true,"duration_ms":22764,"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":"This paper argues that low-frequency turnovers in star-forming galaxies' radio spectra are produced locally in recent starburst regions, not by global galaxy properties.","keywords":["radio continuum: galaxies","galaxies: star formation","low-frequency turnover","free-free absorption","synchrotron losses","Bayesian SED modelling","GLEAM survey"],"falsifier":"Re-observe the same 19 galaxies at 50-300 MHz with deeper, few-percent-calibrated imaging and rerun the same model selection; if the turnovers and FFA-preferred models disappear or move between galaxies, the central interpretation fails.","tokens_in":45801,"feed_emoji":"📡","tokens_out":9215,"duration_ms":84521,"temperature":0.7,"pith_summary":"The paper sets out to establish what causes low-frequency turnovers (LFTOs) in the radio spectra of star-forming galaxies and whether those causes are visible in galaxy-wide properties. Using new 70 MHz to 17 GHz continuum observations of 19 nearby star-forming galaxies, 11 selected to have turnovers and eight matched controls, the authors fit modular Bayesian radio SED models. They find that LFTO and control galaxies are indistinguishable in stellar mass, star-formation rate, specific star-formation rate, and inclination, with only qFIR separating the two. The paper's conclusion is that LFTOs are produced by free-free absorption and ionisation losses in individual recent starburst regions with specific orientations and interstellar-medium densities, and that averaging over the whole galaxy hides the cause.","feed_headline":"Radio dips trace starburst regions, not galaxy-wide properties","feed_subtitle":"A 19-galaxy radio study finds low-frequency dips carry no global signature besides infrared excess.","key_machinery":"The load-bearing machinery is a modular Bayesian radio SED fitting framework. A base model (single power law, or synchrotron plus flat free-free emission) can be prefixed with free-free absorption at low frequencies, with the turnover frequency limited to 300 MHz or below, and suffixed with either a second free-free-absorbed component or synchrotron and inverse-Compton losses producing a high-frequency break. Model choice is made by nested-sampling evidence estimates after MCMC posterior sampling, and the fitted low-frequency turnover is converted to an emission measure using the standard free-free opacity formula. This modularity is what lets the paper attribute curvature to specific physical processes, and the emission-measure link is what connects the turnover to individual HII regions rather than the galactic disk.","core_discovery":"The central claim is that a low-frequency turnover in a star-forming galaxy's radio SED is a local phenomenon: it arises from free-free absorption and ionisation losses within individual recent starburst regions, not from anything special about the galaxy as a whole. The evidence is the absence of significant separation between the 11 LFTO and 8 control galaxies in most global astrophysical properties, the independence of turnovers from inclination, and the presence of only a qFIR excess (infrared brightness relative to 1.4 GHz radio) rather than a full set of distinguishing traits. The same modelling shows the fitted synchrotron spectral index steepens with stellar mass and galactic radius, flattens in mergers with elevated specific star-formation rates, and is uncorrelated with redshift, supporting synchrotron losses as the main high-frequency steepening agent in larger systems.","pith_inferences":["A testable prediction beyond the paper: if the qFIR excess reflects the roughly 10 Myr lag between IR and radio tracers, LFTO galaxies should show higher 40 GHz free-free to 1.4 GHz synchrotron ratios than control galaxies at the same SFR.","If LFTOs are local and transient, resolved spectral-index maps should show turnovers only in the most compact, highest-emission-measure HII regions, with turnover frequency scaling with emission measure; the current data are unresolved at GLEAM frequencies.","The absence of global correlations implies that stacking galaxies in deep low-frequency surveys will dilute turnovers, so the apparent rarity of LFTOs may be an observational resolution effect rather than a physical one."],"forward_implications":["If right, unresolved low-frequency surveys cannot use mass, SFR, or inclination to predict which star-forming galaxies will show turnovers; resolved observations of individual starburst regions are needed.","The steepening of the modelled synchrotron spectral index with stellar mass and galaxy size implies that radio K-corrections and 1.4 GHz SFR calibrations may need a mass- and frequency-dependent correction.","Merger-triggered starbursts inject fresh electrons, so merging systems should show flatter spectral indices and elevated sSFR; four of the five mergers in this sample were selected as LFTO galaxies.","The elevated qFIR of LFTO galaxies is consistent with a time lag between infrared and radio star-formation tracers, making LFTOs transient features of young starbursts rather than persistent galaxy states.","Simple single power-law models are preferred without high-frequency data, so adding sensitive observations above 17 GHz should reveal thermal emission and loss processes and steepen the recovered spectral indices."],"supporting_citations":[{"why":"Supplies the GLEAM-6dFGS parent catalogue and 200 MHz selection from which the LFTO and control samples are drawn.","marker":"Franzen et al. (2021)"},{"why":"Provides the GLEAM survey flux measurements and the low-frequency spectral-index method used to identify turnover candidates.","marker":"Hurley-Walker (2017)"},{"why":"Supplies the physical formalism for synchrotron and free-free emission, free-free absorption, and the canonical spectral index.","marker":"Condon (1992)"},{"why":"Its modular radio SED models and LIRG/ULIRG sample are the direct template the paper extends to lower-SFR galaxies.","marker":"Galvin et al. (2018)"},{"why":"Establishes free-free absorption as the explanation for low-frequency turnovers in luminous infrared galaxies.","marker":"Clemens et al. (2010)"},{"why":"Provides a comparison LIRG sample with FFA-modelled SEDs used in the spectral-index and qFIR comparisons.","marker":"Dey et al. (2022)"},{"why":"Adds the latest ULIRG SED sample and the spectral index versus mass/SFR relations the paper compares against.","marker":"Dey et al. (2024)"},{"why":"Gives the 1.4 GHz radio SFR calibration used to derive radio star-formation rates.","marker":"Molnár et al. (2021)"},{"why":"Quantifies how ionisation losses flatten the low-frequency synchrotron spectrum, supporting the LFTO interpretation.","marker":"Basu et al. (2015)"}],"fun_headline_variants":["Radio SED turnovers trace local starbursts, not global galaxy properties","Low-frequency dips: a local starburst signature in 19 galaxies","Galaxy radio dips are local, not global: 19-galaxy study","Turnovers in radio SEDs: local absorption, not global traits","Radio turnovers: starburst regions cause low-frequency dips"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the GLEAM 72-232 MHz flux densities are accurate enough for the fitted turnovers to be real; if low-frequency photometric errors are larger than the conservative 10 percent added in quadrature, the preference for free-free absorption models and the claimed decoupling from global properties could be noise.","fun_headline_variants_meta":{"raw":{"variants":["Radio SED turnovers trace local starbursts, not global galaxy properties","Low-frequency dips: a local starburst signature in 19 galaxies","Galaxy radio dips are local, not global: 19-galaxy study","Turnovers in radio SEDs: local absorption, not global traits","Radio turnovers: starburst regions cause low-frequency dips"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000269,"raw_usage":{"total_tokens":1690,"prompt_tokens":1081,"completion_tokens":609,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":697,"completion_tokens_details":{"reasoning_tokens":514}},"tokens_in":697,"tokens_out":609,"duration_ms":6024,"temperature":1.0,"reasoning_tokens":514,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T22:43:06.434754+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-observe the same 19 galaxies at 50-300 MHz with deeper, few-percent-calibrated imaging and rerun the same model selection; if the turnovers and FFA-preferred models disappear or move between galaxies, the central interpretation fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the GLEAM-6dFGS parent catalogue and 200 MHz selection from which the LFTO and control samples are drawn."},{"cited_title":"A Rescaled Subset of the Alternative Data Release 1 of the TIFR GMRT Sky Survey","cited_arxiv_id":"1703.06635","evidence_quote":"Provides the GLEAM survey flux measurements and the low-frequency spectral-index method used to identify turnover candidates."},{"cited_title":"J., Seymour , N., Marvil , J., et al","cited_arxiv_id":null,"evidence_quote":"Its modular radio SED models and LIRG/ULIRG sample are the direct template the paper extends to lower-SFR galaxies."},{"cited_title":"Radio-only and Radio-to-far-ultraviolet Spectral Energy Distribution Modeling of 14 ULIRGs: Insights into the Global Properties of Infrared Bright Galaxies","cited_arxiv_id":"2402.10786","evidence_quote":"Adds the latest ULIRG SED sample and the spectral index versus mass/SFR relations the paper compares against."}],"review_version":1}