{"id":"265fb45f-d9ff-4601-b656-580a6238f578","arxiv_id":"1909.02242","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A semi-analytic model predicts the redshift evolution of the HI mass function, cosmic HI density, halo HI content, bias, and shot noise, and provides HOD fitting functions for 21 cm intensity mapping.","lead":"This paper uses the GAEA semi-analytic galaxy model on the Millennium simulations to map how atomic hydrogen is distributed in galaxies and dark matter halos from redshift zero to five. It provides fitting functions and clustering predictions for future 21 cm intensity mapping surveys such as SKA.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 21 cm power-spectrum predictions for z>2 inherit an unresolved ΩHI tension: the model's HI density lies below DLA-based data, and the paper's diffuse-IGM explanation (~20%) is quantitatively insufficient; since P21cm scales as xHI², the headline SKA-aimed amplitude is conditional on an…","rationale":"The reader's CONDITIONAL verdict is appropriate, and this pass leaves it unchanged. I weighed other candidate concerns and set them aside. The z=1 parameters of Eq. 2 (log10 Mmin = -1.3, log10 Mbreak = 8.30) look pathological, but this is a fitting artifact in a secondary deliverable and the paper's integral check via Eq. 4 is insensitive to shape errors. The acknowledged 0.5-dex overproduction of the stellar mass function below 10^7.5 M_sun is softened by the fact that the low-mass HIMF is the tuned target, so the HI content there is constrained by construction. The WMAP1-versus-Planck cosmology shift is explicitly argued to be subdominant with supporting references. The paper earns credit where it has external support: the bHI values and the z=0 dip agree with Illustris-based Villaescusa-Navarro et al. (2018) and with Anderson et al. (2018), and the shot-noise check via Eq. 4 is consistent within a factor of two. The one concern that directly threatens the abstract's promise of predictions 'to be tested against data from future radio telescopes such as SKA' is the high-redshift amplitude chain. The model's ΩHI is below DLA measurements; the paper admits the disagreement is unresolved; the diffuse-IGM explanation is cited at ~20%, which is numerically insufficient to close a factor-of-several gap; and P21cm scales as xHI². Even a perfectly correct bias and shape prediction would then yield a 21 cm amplitude low by roughly the square of the ΩHI discrepancy. The paper itself notes that DLA and emission measurements may not trace the same quantity, so the tension could in principle be spurious—but that is exactly why the concern is unresolved rather than refuted. The concrete test quantifies the gap and shows how much the SKA-aimed amplitude moves under a DLA-calibrated ΩHI; a shift of more than ~2x at z ≥ 2 would require this condition to be stated explicitly before the predictions are used in forecasts. The reader's weakest_assumption identified the same territory (subgrid validity plus the Section 4 DLA tension); I narrow it to the ΩHI-to-P21cm amplitude chain and add the quantitative insufficiency of the 20% diffuse-IGM attribution, hence partial agreement.","tokens_in":32235,"tokens_out":11415,"duration_ms":112849,"concrete_test":"Compare the GAEA-predicted ΩHI(z) at z = 2-5 directly with the DLA-based compilation of Crighton et al. (2015) shown in Figure 4, computing the ratio per redshift and testing whether adding a 20% diffuse-IGM component closes the gap. Then recompute P21cm(k = 0.2 h/Mpc, z) from Eq. 9 using the DLA-calibrated ΩHI instead of the model value. If the resulting amplitude shifts by more than a factor ~2 at any z ≥ 2, the SKA-aimed predictions are materially conditional on the unresolved ΩHI tension; if not, the concern does not land.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The weakest link in the central argument is the chain from the model's high-redshift HI density to the paper's headline IM deliverable. Section 4 and Figure 4 show that GAEA's ρHI(z) declines with redshift and falls below the DLA-based compilation of Crighton et al. (2015); the paper states the result is 'in tension with observational data based on DLAs' and, in Section 7.1, declares a full understanding 'beyond the scopes of this work.' It offers three candidate fixes: resolution, low-mass halos, and diffuse IGM HI. The paper argues resolution is not the main driver, and cites Villaescusa-Navarro et al. (2018) that at z = 5 only about 80% of HI is inside galaxies, i.e., the diffuse component is only ~20%. If the model–data gap in Figure 4 is a factor of two or more (the paper gives no number), then the 20% diffuse component cannot close it, so the tension is effectively unmodelled. This is load-bearing because the headline 21 cm prediction, Eq. 9 (P21cm = Tbar_b² xHI² P_RS_HI), scales the amplitude with the square of the model's ΩHI. At z ≳ 2 the P21cm amplitude SKA will test is therefore offset by roughly the square of the ΩHI discrepancy even if the bias, shot-noise, and shape predictions are exactly right. The paper frames P21cm as 'a fundamental prediction to be tested,' but for z ≳ 2 that prediction is anchored to an admitted, untested assumption rather than to the DLA data displayed in the same paper.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a semi-analytic model (GAEA) run on the Millennium and Millennium-II N-body simulations to characterize the atomic hydrogen (HI) content of the post-reionization universe from z=0 to z=5. It reports the HI mass function, the cosmic HI density, the HI-to-halo-mass relation including a dependence on halo assembly history, the HI bias, the shot noise, and the redshift-space 21 cm power spectrum. It also provides fitting functions for the HI-halo mass relation for use in HOD-based mock catalogs. The z=0 HI mass function is reproduced by construction, since the star formation efficiency is tuned to it; the predicted cosmic HI density declines with redshift and lies below DLA-based measurements at z>2, a tension the authors attribute partly to diffuse IGM HI but leave largely unresolved. The 21 cm power spectrum amplitude inherits this tension through its dependence on the HI density squared.","tokens_in":32613,"tokens_out":6081,"duration_ms":62271,"significance":"If the model predictions hold, this is a useful contribution: it provides new fitting functions for MHI(Mh) with assembly-bias dependence, a detailed decomposition of HI clustering into central/satellite and red/blue populations, and internally consistent shot-noise and bias estimates that agree reasonably with hydrodynamical simulations such as Illustris. The internal checks are reassuring: the shot noise matches the analytic expectation in Eq. (3), and the redshift-space power spectrum reproduces the Kaiser limit at large scales. The main deliverable is a falsifiable prediction for the 21 cm intensity-mapping signal, but its high-redshift amplitude is conditional on an unresolved discrepancy in the cosmic HI density, so the significance of that specific prediction is limited until the tension is quantified and addressed.","major_comments":[{"comment":"The high-redshift ΩHI tension is load-bearing for the headline 21 cm prediction but is left unquantified. Section 4 and Figure 4 show that the model's ρHI(z) falls below the DLA-based data of Crighton et al. (2015), and Section 7.1 attributes the gap only in small part to resolution and suggests diffuse IGM HI at the ~20% level. If the model–data gap is a factor of two or more, as the figure qualitatively suggests, a 20% diffuse component cannot close it, and the predicted P21cm amplitude, which scales as xHI² in Eq. (9), is offset by roughly the square of the discrepancy for z≳2. Please quantify the gap, discuss the implied amplitude correction, and either provide a physically motivated fix or explicitly restrict the 'prediction to be tested' to the redshift range where ρHI is secure.","section":"§4, §7.1, Eq. (9)"},{"comment":"The agreement of the z=0 HI mass function with HIPASS/ALFALFA in Figure 1 is a calibration target, not an independent test, because the star formation efficiency parameter is explicitly tuned to reproduce it (Section 2). The text acknowledges this, but the abstract and conclusions still present the z=0 agreement as evidence that the model 'reproduces well the HI distribution measured in the local Universe.' Please rephrase to distinguish calibrated outputs from genuine predictions, such as the assembly-bias dependence, bias evolution, and shot-noise levels.","section":"§2 and §3"},{"comment":"The proposed fitting formula is a central deliverable for HOD mock construction, but no goodness-of-fit statistic is reported, and several best-fit values are unphysical (e.g., negative a2 at z=5; log10(Mmin) = -1.3 at z=1, which effectively removes the low-mass cutoff). Please provide a quantitative measure of the fit quality for each redshift, discuss parameter degeneracies, and state the applicable mass range over which the formula should not be used.","section":"§5.1, Eq. (2) and Table 2"}],"minor_comments":[{"comment":"Typo: 'thee contribution' should be 'the contribution' in the paragraph describing the power-spectrum calculation.","section":"§6"},{"comment":"Typo: 'fucntion' should be 'function' in the first paragraph.","section":"§6.3"},{"comment":"Typo: 'thespatial' should be 'the spatial' in the caption.","section":"Figure 9 caption"},{"comment":"The claim that resolution is not the main driver of the ρHI tension is supported only by a qualitative argument. A quantitative convergence test (e.g., comparing MII results with a higher-resolution run or a resolvable halo-mass cut) would strengthen this claim.","section":"§7.1"},{"comment":"Please clarify that Eq. (8) is the angle-averaged Kaiser limit for the monopole of the redshift-space power spectrum, and that the comparison in Figure 19 uses the spherically averaged power spectrum.","section":"§6.7, Eq. (8)"},{"comment":"The fitting formula in Table B3 uses γ=0.3 while Table 2 uses γ=0.5. Please explain the choice and whether the assembly-bias fits are sensitive to this parameter.","section":"Appendix B, Table B3"}],"recommendation":"major_revision","confidential_remarks":"The paper is transparent about the ΩHI tension, but the framing of the 21 cm prediction as 'a fundamental prediction to be tested' is stronger than the evidence warrants; the amplitude at z>2 depends on an unvalidated assumption. The calibration of the z=0 HI mass function is acknowledged, but the abstract should be adjusted to avoid presenting it as an independent validation. I recommend major revision rather than rejection because the clustering predictions, bias evolution, and fitting functions are largely independent of the absolute ΩHI normalization and are of interest to the intensity-mapping community."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Solid, clearly written SAM analysis. What's actually new: fitting functions for the GAEA HI–halo mass relation with separate centrals/satellites and assembly-age bins, plus HI radial profiles. These are directly usable for HOD mocks, and the assembly-bias split is a genuine improvement over the simple mass-only fits most IM forecasts use. The clustering work is careful: shot noise matches Equation 3, the Kaiser limit is reproduced at large scales, and the bias evolution agrees with Illustris to within a factor of two on shot noise. Those results are the paper's core, and they hold up.\n\nThe soft spots are real but mostly admitted. The z=0 HI mass function is a calibration target, so Figure 1 is not an independent test. The bigger issue is the high-redshift HI density falling below DLA data. Since P21cm scales with the square of ΩHI, the amplitude of the headline 21 cm predictions for z>2 is conditional on that tension. The paper says so in Section 6.7, so it isn't hidden, but the suggested diffuse-IGM fraction (~20%) probably can't close a factor-of-two gap—and the authors never state the gap size. A referee should ask for that number and for the P21cm curves renormalized to the DLA-based ΩHI as well as to the model.\n\nMinor but worth noting: they keep galaxies in the low-stellar-mass regime where the SAM overproduces the stellar mass function by ~0.5 dex; that could influence the low-mass shot noise and bias, though the convergence checks mitigate the concern. No public code or catalogs, which limits how quickly the fitting functions get adopted.\n\nThe stress-test note is on target but not fatal: it makes the high-z amplitude conditional, not the shape or the bias. This paper deserves serious peer review; the main requested revision should be a quantitative treatment of the ΩHI tension and its propagation into P21cm.","headline":"Useful HOD-ready fitting functions and a careful clustering dissection from GAEA; the high-z 21 cm amplitudes carry an admitted but unquantified ΩHI caveat.","tokens_in":33170,"tokens_out":3070,"would_cite":true,"duration_ms":30510,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A semi-analytic galaxy formation model provides a comprehensive description of atomic hydrogen from z=0 to z=5, predicting the HI bias, shot noise, and 21cm power spectrum for intensity mapping experiments.","keywords":["atomic hydrogen","21 cm intensity mapping","HI mass function","HI bias","shot noise","semi-analytic galaxy formation model","halo occupation distribution"],"falsifier":"Measure the 21cm power spectrum and cosmic HI density at $z\\approx2$–$4$ with a future intensity mapping survey; if the measured HI bias or shot noise deviates strongly from these predictions, or if the missing HI at $z>2$ turns out to reside in galaxies rather than diffuse intergalactic gas, the central assumption would be wrong.","tokens_in":32012,"feed_emoji":"📡","tokens_out":6361,"duration_ms":61761,"temperature":0.7,"pith_summary":"This paper uses a semi-analytic galaxy formation model, GAEA, run on two large N-body simulations, to provide a coherent description of atomic hydrogen (HI) from the local universe to $z=5$. The authors show that the model reproduces the local HI mass function, and use it to give new fitting formulas for how much HI a dark matter halo of a given mass contains, including how that relation depends on when the halo assembled. They then predict the HI bias and shot noise relevant for 21cm intensity mapping, finding that the bias grows with redshift and that shot noise will not hinder BAO-scale measurements. The final output is a predicted 21cm power spectrum in redshift space, ready to be tested by next-generation radio telescopes.","feed_headline":"HI bias rises with redshift out to z=5","feed_subtitle":"GAEA model predicts 21cm power spectrum ready for SKA to test.","key_machinery":"The load-bearing object is the HI halo mass function $M_{\\rm HI}(M_h)$, the total HI mass in a dark matter halo of mass $M_h$, which the paper parametrizes with a fitting formula that combines a power-law rise with an exponential low-mass cutoff and a high-mass flattening. It is derived from the GAEA semi-analytic model, whose key physical ingredients are a pressure-based split of cold gas into atomic and molecular hydrogen, star formation proportional to molecular hydrogen surface density, and AGN feedback that suppresses gas cooling in massive halos. The fitting functions let users build HI mocks with halo occupation distribution (HOD) techniques.","core_discovery":"On its own terms, the paper establishes that the GAEA semi-analytic model delivers a realistic and comprehensive description of neutral hydrogen in the post-reionization universe. The predicted HI mass function matches HIPASS and ALFALFA in the local universe without an excess of low-mass HI galaxies. The central new results are: the HI-halo mass relation requires a low-mass cutoff and a high-mass flattening (the latter from AGN feedback); its scatter is driven primarily by halo assembly history; the HI bias increases with redshift; and shot noise is small enough that BAO-scale intensity mapping is feasible. The paper closes with a redshift-space 21cm power spectrum for SKA to test.","pith_inferences":["If assembly history truly drives the scatter in $M_{\\rm HI}(M_h)$, then standard HOD mocks that assign HI only from halo mass will systematically misestimate small-scale clustering; adding a formation-time or assembly-bias parameter should improve them.","The paper's explanation for the $z>2$ HI density deficit, diffuse IGM gas outside halos, implies that future auto-power 21cm measurements at high redshift must account for a diffuse component, not just galaxies.","The predicted $z=0$ dip in HI bias at $k\\sim 1\\,h\\,\\mathrm{Mpc}^{-1}$ offers a sharp test: cross-correlating 21cm maps with optical galaxy samples split by color should reproduce the spoon shape if the model's satellite physics is right.","Comparing these mocks with the full shape of the 21cm power spectrum from SKA pathfinders at $z\\sim 0.8$ could help break the degeneracy between HI density and bias."],"forward_implications":["The new $M_{\\rm HI}(M_h)$ fitting functions allow fast generation of HI mocks with HOD techniques from $z=0$ to $z=5$, useful for forecasting intensity mapping surveys.","The HI bias increases with redshift, so the 21cm signal grows stronger at high redshift, helping future intensity mapping experiments.","Shot noise is low enough that BAO-scale intensity mapping is feasible, as quantified by $nP_{0.2}$ values well above one.","The predicted dip in the $z=0$ HI bias at $k\\sim 1\\,h\\,\\mathrm{Mpc}^{-1}$, linked to HI-poor red satellites, can be tested with cross-correlations between 21cm maps and galaxy surveys.","The redshift-space 21cm power spectrum predictions provide a direct target for SKA and its pathfinders to confirm or rule out."],"supporting_citations":[{"why":"Provides the local HI mass function from HIPASS used to tune the star formation efficiency.","marker":"Zwaan et al. (2005)"},{"why":"Supplies the ALFALFA local HI mass function as a second observational constraint.","marker":"Martin et al. (2010)"},{"why":"Gives the pressure-based H2/HI relation used to partition cold gas into atomic and molecular hydrogen.","marker":"Blitz & Rosolowsky (2006)"},{"why":"Introduces the GAEA model with an H2-based star formation law that is central to the paper's HI predictions.","marker":"Xie et al. (2017)"},{"why":"Provides the Millennium simulation, which supplies the large-volume dark matter merger trees.","marker":"Springel et al. (2005)"},{"why":"Provides the Millennium II simulation, which gives better mass resolution for low-mass halos.","marker":"Boylan-Kolchin et al. (2009)"},{"why":"Serves as the main comparison for HI bias and shot noise from hydrodynamical simulations.","marker":"Villaescusa-Navarro et al. (2018)"},{"why":"Offers the parametrization that is extended here with a low-mass cutoff, and a comparison for shot noise.","marker":"Baugh et al. (2019)"},{"why":"Reports the observed dip in 21cm cross-power spectrum that motivates the red/blue galaxy analysis.","marker":"Anderson et al. (2018)"},{"why":"Supplies the halo mass function used to compute the shot noise in the halo model.","marker":"Sheth & Tormen (1999)"}],"fun_headline_variants":["GAEA model matches local HI surveys, predicts z=5 bias","Shot noise won't block 21cm intensity mapping","GAEA model: HI bias rises, shot noise low for SKA","Redshift-space 21cm power spectrum from GAEA for SKA"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model's recipes for star formation and the atomic/molecular gas split, tuned to local observations, are assumed to stay valid at high redshift and in low-mass halos down to about $10^{10}$ solar masses, despite the paper showing that this leads to a lower cosmic HI density at $z>2$ than DLA measurements.","fun_headline_variants_meta":{"raw":{"variants":["GAEA model matches local HI surveys, predicts z=5 bias","Shot noise won't block 21cm intensity mapping","GAEA model: HI bias rises, shot noise low for SKA","Redshift-space 21cm power spectrum from GAEA for SKA"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000584,"raw_usage":{"total_tokens":2745,"prompt_tokens":939,"completion_tokens":1806,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":555,"completion_tokens_details":{"reasoning_tokens":1732}},"tokens_in":555,"tokens_out":1806,"duration_ms":14795,"temperature":1.0,"reasoning_tokens":1732,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:55:51.557486+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the 21cm power spectrum and cosmic HI density at $z\\approx2$–$4$ with a future intensity mapping survey; if the measured HI bias or shot noise deviates strongly from these predictions, or if the missing HI at $z>2$ turns out to reside in galaxies rather than diffuse intergalactic gas, the central assumption would be wrong.","supporting_citations":[],"review_version":1}