{"id":"ebccd408-64b3-4fc8-a7cb-615165dd2de6","arxiv_id":"2509.10271","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"About 25% of luminous quasars at z<0.76 are hosted by post-starburst galaxies, a 28-fold excess over mass-matched inactive galaxies, implying mergers trigger a significant fraction of quasar activity.","lead":"This paper applies a machine-learning spectral decomposition technique, MFICA, to split Sloan Digital Sky Survey quasar spectra into the black hole's glow and the host galaxy's starlight. It finds that about a quarter of quasars live in recently star-bursting, rapidly quenching galaxies, suggesting mergers feed many bright quasars.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Mock validation uses quasar spectra inside the MFICA subspace, leaving the AF weight—the sole basis of the post-starburst fraction—untested against real quasar diversity; the 25% claim needs an out-of-subspace test.","rationale":"The paper is a careful observational study: it constructs a mass-matched control sample, performs mock decompositions, and reports explicit robustness checks on stellar-mass uncertainties and redshift restrictions. The stress-test concern is not that the authors are careless, but that the one load-bearing premise—the uniqueness of the separation of the small AF component from real quasar spectral diversity—is not tested by the current mock recipe. The reader's weakest-assumption analysis identified exactly this issue, and the manuscript itself flags related limitations: Section 3.2 notes that the classification boundaries are drawn visually and are somewhat arbitrary; Section 4.3 acknowledges the known degeneracy between the OB component and AGN continuum; Section 5.3 shows that stellar-mass uncertainties can alter the relative star-forming/green-valley/quiescent proportions. None of these checks, however, varies the quasar basis or injects real quasar residuals, which is the regime most likely to bias the AF weight. The concern is unresolved rather than demonstrated: the mocks do recover f_AF and f_K with small offsets, and the post-starburst excess is large enough that a modest systematic contamination would not necessarily erase it. Because the issue is real but not yet shown to break the claim, the conditional verdict is appropriate. I see no internal inconsistency that would justify rejection, so I recommend no change to the reader's CONDITIONAL verdict.","tokens_in":33608,"tokens_out":6954,"duration_ms":73112,"concrete_test":"Build mocks whose quasar component includes out-of-subspace diversity: take real SDSS DR7 quasar spectra in the training redshift range, fit them with the 10 MFICA QSO components plus galaxy components, and store the fit residuals. Then construct mock spectra by adding known galaxy composite spectra (SF, SB, PSB, GV, red) at f_gal between 0.2 and 0.5 to (a) MFICA QSO reconstructions alone and (b) the same reconstructions plus the stored real-quasar residuals, with SDSS noise added as in Section 4.2(iv). Run the Section 4.1 decomposition on both sets and compare recovered f_AF values and the fraction classified as post-starburst. If including real residuals shifts the recovered PSB fraction by more than the quoted uncertainties, the 25% claim needs a systematic error term; if the shift is negligible, the mock validation is adequate.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central post-starburst excess in Section 7(iii) rests entirely on the recovered AF-component weight, but the mock validation does not certify that weight against real quasar spectral diversity. Section 4.2(ii) builds every mock quasar as an MFICA reconstruction of the quasar training sample, and Section 2.4.2 removes the average ~2 per cent AF contribution from those training spectra before generating the QSO components. Any real quasar spectral structure that projects onto the AF subspace—broad Balmer features, Fe II pseudo-continuum, associated/BAL absorption, or continuum-shape diversity—is therefore absent from the mocks, while in real data it can be partially absorbed by the positive AF component. Because the post-starburst classification boundary in Section 3.2 is defined directly in f_AF-f_K space, even a small systematic AF contamination can move hosts across the boundary and change the quoted 24.9±0.6 per cent. The 28±1 excess carries only the formal covariance-based error from Section 4.1, with no systematic term for this degeneracy. The robustness discussion in Section 5.3 varies the host-galaxy subtraction and stellar-mass estimation, but it does not vary the quasar basis or test what happens when real quasar residuals are present. The uniqueness of the linear decomposition for real quasar spectra is therefore the load-bearing premise, and it is currently supported only by mocks that lie inside the quasar-component subspace.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces Mean-Field Independent Component Analysis (MFICA) to decompose SDSS DR7 quasar spectra into quasar and host-galaxy components, using the galaxy component fractions f_K, f_AF, and f_OB to classify host galaxies by recent star-formation history. After validating the decomposition with mock spectra, the authors apply it to 3376 quasars with f_gal >= 0.2 and S/N >= 10, compare the resulting host types with a stellar-mass-matched control sample, and report that roughly 53 per cent of quasar hosts are star-forming or starburst, about 5 per cent are quiescent, and approximately 25 per cent are post-starburst, corresponding to a 28 +/- 1 excess relative to controls. The paper interprets the post-starburst excess as evidence that 30-50 per cent of quasars originate from merger-induced starbursts, with the remainder secularly fed.","tokens_in":33897,"tokens_out":4986,"duration_ms":45152,"significance":"If the decomposition is unbiased, this is one of the first large spectroscopic censuses of the stellar populations of unobscured quasar hosts, and the mass-matched control comparison makes the claims substantially stronger than a simple host-type census. The claimed 28 +/- 1 post-starburst excess is a striking result that, if robust, materially strengthens the case for merger-triggered quasar activity in a subpopulation of quasars. The paper has clear strengths: the mock construction uses galaxy composites that are out of the MFICA galaxy subspace, the classification criteria are stated explicitly, the covariance-based error propagation is careful, and the analysis is reproducible with public SDSS data and published templates. The central quantitative claims rest on the recovered AF-component weight and on visually defined classification boundaries, and both of these ingredients currently lack direct robustness tests that bear on the headline percentages.","major_comments":[{"comment":"Section 4.2(ii) constructs every mock quasar as an MFICA reconstruction of the quasar training sample, so all mock quasars lie exactly inside the quasar-component subspace. The validation therefore does not test whether the small AF component (average roughly 2 per cent of quasar light; Section 2.4.2) can be separated from real quasar spectral diversity that falls outside that subspace, such as broad Balmer emission residuals, Fe II pseudo-continuum, BAL or associated absorption, or continuum-shape variations. Because the post-starburst classification in Section 3.2 is defined directly in f_AF-f_K space, a systematic AF contamination of even a few per cent could move hosts across the PSB boundary and change the quoted 24.9 +/- 0.6 per cent and the 28 +/- 1 excess, whose quoted errors are formal covariance errors only. Please add an out-of-subspace test, for example by adding quasar residual spectra (PCA components or galaxy-subtracted real quasar residuals) to the mock quasar inputs and quantifying the induced bias in f_AF.","section":"4.2(ii), 5.2, 7(iii)"},{"comment":"The classification boundaries in Fig. 5 are defined visually and are described as somewhat arbitrary, yet they directly determine all percentages in Table 1 and the central excess quoted in Section 7(iii). The paper does not quantify how the post-starburst fraction or the 28 +/- 1 excess responds to plausible shifts in these boundaries. Please provide a boundary-sensitivity analysis, for example by perturbing each boundary by the typical scatter of the galaxy classifications or by the width of the post-starburst locus, and report the resulting range on the headline percentages.","section":"3.2, Table 1, 7(iii)"},{"comment":"Section 5 states that the sample is limited to quasars with chi-squared_nu >= 2.0, while the caption of Fig. 10 and the text around it use chi-squared_nu < 2.0; the sentence describing the lim it is internally inconsistent. Please correct the inequality and justify the chosen threshold, since this cut determines the sample of 3376 quasars on which all science results are based.","section":"5, Fig. 10"}],"minor_comments":[{"comment":"The summation limits in Eq. (3) and Eq. (5) appear to be off by one relative to the stated number of components (seven galaxy components and ten quasar components); please make the indexing and component counts consistent.","section":"Eq. (3), Eq. (5)"},{"comment":"The text says the quasar mocks 'represent the full diversity of spectra seen across the quasar population', but since they are MFICA reconstructions they only represent diversity inside the quasar-component subspace; please temper this wording to match what is actually tested.","section":"4.2(ii)"},{"comment":"The statement that the decomposition was also tested using real galaxy spectra rather than high-S/N composites is not accompanied by any quantitative result or figure; please provide the recovery accuracy for that test.","section":"4.3"},{"comment":"The claim that the post-starburst fraction is 'quantitatively robust' to the host-subtraction prescription would be easier to evaluate if the paper reported the PSB fractions obtained in the two variant decompositions rather than only the resulting stellar-mass changes.","section":"5.3"},{"comment":"There is a typo in item (ii): 'quiscent/red' should read 'quiescent/red'.","section":"7(ii)"},{"comment":"Figure 14 error bars are described as the maximum and minimum percentages from 10,000 resamples, while Table 1 reports standard deviations; please state explicitly which quantity is displayed and why the two differ.","section":"Fig. 14, Table 1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a good fit for MNRAS and the science question is timely. My recommendation is driven by two load-bearing gaps: the mock validation does not exercise quasar spectra outside the MFICA subspace, which is exactly the regime where the AF weight could be contaminated, and the visual classification boundaries directly set the headline percentages without a sensitivity test. Both are fixable within the scope of the manuscript. I would also ask the authors to resolve the chi-squared cut inconsistency before acceptance. No concerns about citation practice or novelty disclosure."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know about this one. First, it is the first statistical census of quasar host stellar populations done with MFICA decomposition, and it includes a mass-matched control sample analysed with the same method—that alone is a real step beyond earlier quasar-host work. Second, the headline result—about 25 per cent of quasars hosted by post-starburst galaxies, a 28±1 excess over control—is plausible but not as solid as the error bars suggest, because the validation mocks never test the one quantity the classification rests on.\n\nThe paper does a lot well. The MFICA components are derived carefully from galaxy training samples, and the galaxy-component validation reproduces H-delta_A and D_n4000 from MPA-JHU without bias. The mock decomposition tests S/N and host fraction limits, and the authors use a random forest for stellar masses rather than a too-simple SPS fit, explaining why. They also check the effect of removing host-galaxy light from the quasar template and confirm the post-starburst excess is robust to that choice and to a z<0.3 restriction. That is honest, reproducible work in the best sense.\n\nThe soft spots are real but not fatal. The stress-test concern is correct: every mock quasar is an MFICA reconstruction of the training sample, so the AF weight—the sole basis for post-starburst classification—is never tested against real quasar spectral diversity that might project onto the AF subspace. Broad Balmer features or Fe II pseudo-continuum could be partially absorbed by the AF component and push hosts across the boundary. The 28±1 carries only covariance error, no systematic term for this. I would want an out-of-subspace test before trusting 24.9±0.6 rather than, say, 20 or 30 per cent.\n\nTwo smaller issues: the classification boundaries in Fig. 5 are visually set and directly set the quoted percentages; the paper acknowledges this but attaches no systematic error. And the chi-squared cut is stated as χ2ν ≥2.0 in Section 5 but χ2ν <2.0 in the Fig. 10 caption—one of them is wrong. No code is released, which limits reproducibility, though templates are available.\n\nWho is this for? Anyone working on AGN feeding, quasar host galaxies, or post-starburst populations. The method paper is worth a serious referee; the central claim will survive or fall on the AF degeneracy, and a good referee can ask for that test. I would send it to review.","headline":"First MFICA census of quasar host stellar populations, with a plausible 25% post-starburst claim that needs an out-of-subspace validation before the exact number is trusted.","tokens_in":34450,"tokens_out":2078,"would_cite":true,"duration_ms":17501,"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":"Around a quarter of all quasars are hosted by post-starburst galaxies—28 times the rate in mass-matched inactive galaxies—implying that 30–50% of quasars are triggered by gas-rich major mergers.","keywords":["quasar host galaxies","post-starburst galaxies","independent component analysis","spectral decomposition","stellar populations","galaxy mergers","AGN feeding","Sloan Digital Sky Survey"],"falsifier":"Compare the A-star component weight for quasars whose host galaxies can be observed independently—spatially resolved integral-field data, or hosts bright enough to dominate the spectrum—against Balmer absorption measured directly in the host light; a host that is demonstrably not post-starburst but carries a large A-star weight would expose the 25 per cent fraction as an artifact. A cheaper computational version injects real quasar spectra (not MFICA reconstructions) with synthetic host galaxies of known type and checks whether residual quasar diversity is misattributed to the A-star component.","tokens_in":33395,"feed_emoji":"🌌","tokens_out":8674,"duration_ms":64548,"temperature":0.7,"pith_summary":"This paper introduces a blind source-separation technique, Mean-Field Independent Component Analysis (MFICA), to pry host-galaxy light out of optical quasar spectra and read off each host's recent star-formation history. Applied to roughly 3,400 Sloan Digital Sky Survey quasars at $0.16 \\leq z \\leq 0.76$, the decomposition yields a census: about 53 per cent of quasar hosts are star-forming, 17 per cent sit in the green valley, 5 per cent are quiescent, and roughly 25 per cent are post-starburst galaxies—hosts that underwent a starburst and then stopped forming stars within the past billion years. The post-starburst fraction is $28\\pm1$ times higher than in a stellar-mass-matched control sample of inactive galaxies, and $7.3\\pm0.4$ times higher per star-forming galaxy. The authors interpret this excess as evidence that 30–50 per cent of quasars are preceded by gas-rich major mergers, with the remaining majority fed by secular processes.","feed_headline":"Quasars are 28 times likelier to live in post-starburst galaxies","feed_subtitle":"New spectral decomposition ties one in four quasars to a recent starburst and rapid quenching.","key_machinery":"The central object is Mean-Field Independent Component Analysis (MFICA), a blind source-separation method that decomposes spectra into statistically independent, positivity-constrained components without the whitening step that distorts PCA-based templates. The paper trains MFICA on separate galaxy and quasar samples to build a small template set: three stellar-population components (K for old K-giant-like light, AF for intermediate-age A/F stars marked by strong Balmer absorption, OB for young stars and active star formation) plus emission-line components, and seven quasar components with corrective terms. A quasar spectrum is fit as a single linear combination of galaxy and quasar components, and the fractional weights $f_K$, $f_{AF}$, and $f_{OB}$ assign the host to starburst, star-forming, post-starburst, green-valley, or quiescent categories. The AF component is the load-bearing feature: its weight traces the post-starburst population on which the central excess claim rests.","core_discovery":"On the paper's own terms, the central discovery is the post-starburst excess: roughly 25 per cent of quasars in the redshift range $0.16 \\leq z \\leq 0.76$ are hosted by post-starburst galaxies, versus less than 1 per cent of stellar-mass-matched inactive galaxies, a $28\\pm1$-fold excess. Because quasar hosts are massive galaxies (median fibre stellar mass $10^{10.8}\\,M_\\odot$, total near $10^{11.3}\\,M_\\odot$) and massive post-starbursts are predominantly merger products, the paper argues that 30–50 per cent of quasars follow a merger-triggered sequence of starburst, rapid quenching, and black hole growth. The remaining majority—star-forming hosts on the main sequence—are interpreted as a separate, secular feeding mode in which black hole growth keeps step with star formation through a shared gas supply. The paper frames this as two parallel feeding modes for supermassive black holes, and notes that the post-starburst excess is quantitatively robust to the stellar-mass prescription while the exact star-forming-versus-quiescent split is more sensitive to it.","pith_inferences":["A duty-cycle argument could turn the $28\\pm1$ excess into a causal timescale: if post-starburst hosts stay recognizable for roughly 0.5–1 Gyr while quasar episodes last about $10^7$ yr, random coincidence would predict a far smaller overlap, so the observed excess implies quasars are actively triggered during or immediately after the post-starburst window.","The AF component might partially absorb quasar-side spectral diversity that falls outside the trained quasar subspace, since real quasar continua may contain weak Balmer features; this is testable by checking whether the AF weight tracks host properties rather than quasar luminosity or redshift.","The same decomposition could be applied to Type-2 AGN hosts and to higher-redshift surveys with matched rest-frame coverage, providing a consistency check on whether the post-starburst excess grows or shrinks with AGN luminosity.","Extending the decomposition redward to include features like the Mg I triplet could catch older post-starburst populations, testing whether the true merger-triggered fraction lies at the bottom or top of the paper's 30–50 per cent range."],"forward_implications":["Quasars are not a random subset of massive galaxies: the mass-matched control sample is 74 per cent quiescent while only ~5 per cent of quasar hosts are, so young stellar populations and luminous black hole accretion are causally linked, directly or indirectly.","A quarter of quasars show evidence of a recent starburst followed by rapid quenching, connecting luminous AGN activity to a short-lived host phase and constraining when feedback shuts off star formation.","If most massive post-starburst galaxies are merger remnants, then 30–50 per cent of quasars are merger-induced, in line with morphological merger fractions for luminous AGNs (~0.37–0.41) cited in the paper.","The majority of quasars (~53 per cent) live in normally star-forming hosts, implying that secular feeding—bars, minor mergers, disk instabilities—carries the bulk of black hole growth.","The post-starburst excess is quantitatively robust to the exact host stellar-mass calibration, whereas the precise star-forming/quiescent split is sensitive to it."],"supporting_citations":[{"why":"Supplies the mean-field variational ICA algorithm that MFICA is built on.","marker":"Højen-Sørensen et al. (2002)"},{"why":"Contributes the adapted MFICA code, the positivity priors, and the galaxy template generation approach.","marker":"Allen et al. (2013)"},{"why":"Established MFICA reconstruction of quasar spectra, the basis for the paper's quasar components.","marker":"Rankine et al. (2020)"},{"why":"Supplies the SDSS DR7 quasar catalogue from which the quasar sample is drawn.","marker":"Schneider et al. (2010)"},{"why":"Provides the PCA galaxy classification used to build a balanced training sample and the emission-line-free components used to measure H-delta strength.","marker":"Wild et al. (2007)"},{"why":"Supplies the MPA-JHU stellar masses, specific star-formation rates, and the inactive galaxies used for mass matching.","marker":"Brinchmann et al. (2004)"},{"why":"Documents that roughly 90 per cent of massive post-starburst galaxies show merger signatures, the bridge from the post-starburst excess to the merger fraction.","marker":"Ellison et al. (2024)"},{"why":"Quantifies how rare post-starburst galaxies are in the field at z~0.7, anchoring the control comparison.","marker":"Rowlands et al. (2018)"}],"fun_headline_variants":["Quasars 28x more likely in post-starburst galaxies","Merger-triggered starbursts fuel up to half of quasars","Post-starburst excess reveals merger path to quasar ignition","Quasar hosts: one in four is a post-starburst galaxy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the tiny A-star component (about 2 per cent of quasar light) records genuine host-galaxy starlight, because the mock spectra used for validation were built from the method's own quasar components and so never test whether real quasar spectral diversity outside that component set leaks into the A-star signal.","fun_headline_variants_meta":{"raw":{"variants":["Quasars 28x more likely in post-starburst galaxies","Merger-triggered starbursts fuel up to half of quasars","Post-starburst excess reveals merger path to quasar ignition","Quasar hosts: one in four is a post-starburst galaxy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000933,"raw_usage":{"total_tokens":4084,"prompt_tokens":1127,"completion_tokens":2957,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":743,"completion_tokens_details":{"reasoning_tokens":2877}},"tokens_in":743,"tokens_out":2957,"duration_ms":389771,"temperature":1.0,"reasoning_tokens":2877,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:56:07.277310+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the A-star component weight for quasars whose host galaxies can be observed independently—spatially resolved integral-field data, or hosts bright enough to dominate the spectrum—against Balmer absorption measured directly in the host light; a host that is demonstrably not post-starburst but carries a large A-star weight would expose the 25 per cent fraction as an artifact. A cheaper computational version injects real quasar spectra (not MFICA reconstructions) with synthetic host galaxies of known type and checks whether residual quasar diversity is misattributed to the A-star component.","supporting_citations":[],"review_version":2}