{"id":"a5d6cd96-dcc8-42db-aff0-fe8f330ef5f6","arxiv_id":"2607.20123","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":11,"one_line_summary":"A prescription fitted to GAEA2023 reproduces the median HI-halo mass relation and its scatter using halo spin and concentration, enabling realistic 21-cm mock catalogs.","lead":"This paper uses the GAEA2023 semi-analytic galaxy-formation model to map how neutral hydrogen (HI) is distributed in dark-matter halos, and packages the result as a compact recipe that adds realistic HI to dark-matter-only catalogs. The recipe is aimed at making 21-cm intensity-mapping mocks for MeerKAT and SKA more realistic by connecting HI content to halo spin and assembly history.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unvalidated Vmax/Vvir concentration inversion may bias the concentration term and σ_int; a z=0 direct-NFW check would settle it.","rationale":"The paper is careful and transparent, and the reader's identification of the concentration inversion as the weakest assumption is correct. The inversion is not validated even where validation is possible (z=0), and it is demonstrably noisy at high z, exactly where the paper claims the concentration signal fades. However, this concern does not overturn the paper's central argument: the assembly dependence of the HI scatter is corroborated by the independent formation-time analysis in Appendix C.2, the z=0 results are unlikely to be qualitatively changed by moderate concentration errors, and the paper explicitly cautions that the inferred c_h is approximate for unrelaxed halos. The prescription remains useful as a practical tool, provided the concentration term is treated with due caution and the z50 alternative is available. The correct outcome is therefore to retain the ACCEPT verdict while noting that a straightforward z=0 validation would materially reduce the residual uncertainty.","tokens_in":36549,"tokens_out":14742,"duration_ms":151499,"concrete_test":"At z=0, using the available MSI/MSII particle data, fit NFW profiles to a sample of halos spanning 10^11–10^14 Msun and compare the fitted c_fit with the Eq. (4) inversion c_inv. Quantify the median offset, scatter, and dependence on halo relaxation state. Then refit the Table 3 scatter model at z=0 using c_fit in place of c_inv; if B_c and σ_int change by more than ~30% and ~0.05 dex respectively, the concentration-based prescription should be replaced by the z50-based alternative in Appendix C.2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claims — that concentration carries a significant part of the HI scatter and that the residual intrinsic dispersion is ~0.3 dex — rest on the concentration c_h recovered by inverting Eq. (4) from Vmax/Vvir, assuming relaxed NFW halos and restricting to the monotonic branch c_h >= 2.16. This inversion is most fragile precisely where the paper needs it at high redshift: by z~3 most halos have c_h ~ 3–5, near the shallow part of the Vmax/Vvir–c curve, so small errors in Vmax/Vvir (from mergers, triaxiality, substructure, or finite force resolution) map to large errors in c_h. Such errors would attenuate the fitted coefficient B_c and inflate the quoted σ_int, making the apparent high-z fading of the concentration signal at least partly an artifact of the proxy. The alternative formation-time analysis in Appendix C.2 shows that assembly information does persist to z~5, so the broad 'assembly matters' claim survives; but the catalog-level prescription in Table 3, which uses c_h as the assembly proxy, is not secure. Notably, at z=0 the particle data are available for direct NFW fits, yet the paper does not use them to validate Eq. (4). This is the most load-bearing assumption because the main scatter model in Eqs. (7)–(9) is specified in terms of c_h, and any bias in the inversion propagates directly into B_c and σ_int.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper characterizes the HI-halo connection using the GAEA2023 semi-analytic model applied to the Millennium-I and Millennium-II simulations, from z=0 to z~5. It presents the HI mass function, the median M_HI(M_h) relation, and an analysis of the scatter about that relation. The central claims are that the scatter at fixed halo mass is not random but correlates with halo assembly, that higher-spin and less-concentrated (later-forming) halos are HI-richer, and that a compact prescription based on halo mass, spin, and concentration, with a Gaussian residual of sigma_int ~0.3 dex, reproduces the full scatter and can be used to populate dark-matter-only catalogs for 21-cm mocks.","tokens_in":36984,"tokens_out":5744,"duration_ms":70402,"significance":"If the prescription is robust, it is a useful and practical contribution: it is compact, expressed entirely in terms of quantities available in N-body halo catalogs, and it directly targets intensity-mapping mock construction. The paper is careful in several respects: it discloses that the z=0 HIMF is partly a calibration target, it tests the sensitivity of the median fit to the MSI/MSII stitching scale with bootstrap uncertainties, and it provides an alternative formation-time-based scatter model in Appendix C.2. The redshift-dependent median fit and the comparison with existing prescriptions and observations are valuable. However, the main scatter model rests on a halo-concentration proxy that is never validated against direct NFW fits, and the 'reproduces the full scatter' claim is, by the authors' own statement, partly by construction. These issues are load-bearing for the quantitative claims but appear fixable.","major_comments":[{"comment":"The central scatter model, Eqs. (7)-(9), is specified in terms of c_h recovered by inverting the Vmax/Vvir-c relation under the assumptions of relaxed halos and an NFW profile, restricted to the monotonic branch c_h>=2.16. This inversion is most fragile at high redshift, where typical c~3-5 lie near the shallow minimum of Eq. (4). Small Vmax/Vvir errors from mergers, triaxiality, substructure, or resolution can then produce large c_h errors, which would attenuate B_c and inflate sigma_int. The paper explicitly states in footnote 2 that particle data are available for the z=0 snapshot but does not use them to validate Eq. (4). Because the high-z fading of the concentration signal and the redshift growth of sigma_int are among the main quantitative results, the authors should provide a direct NFW-fit comparison at z=0 and, if possible, at one or two higher redshifts, and propagate the unce","section":"Sect. 5.2, Eq. (4); Table 3"},{"comment":"The statement that the prescription 'reproduces the full scatter' is circular as presented: the scatter model is fitted to the GAEA2023 simulation and then compared with the same simulation. The authors acknowledge this in the text ('By construction the two components reproduce the overall HI scatter'), but the central claim in the abstract and Sect. 7 is stronger. Since the paper's purpose is to provide a predictive prescription for mocks, the authors should either reframe the claim as a self-consistency check or, preferably, add an out-of-sample test: e.g., fit on MSII and test on MSI (or a split sample in mass/redshift), or compare the resulting 21-cm clustering/shot noise against observations or an independent model. At minimum, the mass-dependent performance should be quantified with formal variance-reduction statistics, rather than the r2_lambda and r2_c values quoted in the text,","section":"Sect. 6.2, Eqs. (7)-(9); Fig. 10"}],"minor_comments":[{"comment":"The table lists A_lambda, B_c, standardization constants, and sigma_int without uncertainties. Given the bootstrap treatment used for the median fit, the scatter-fit parameters should carry similar error estimates; this is particularly important for sigma_int and B_c because of the concentration-proxy issue.","section":"Table 3"},{"comment":"The symbols r2_lambda and r2_c are introduced in the text but never formally defined. Please define them as variance reductions relative to the scatter about the median relation, and state over which halo-mass range and redshift they are evaluated.","section":"Sect. 6.2"},{"comment":"For z>=3.1, M_break is railed against its prior and a_2 is unconstrained. This is noted and reasonable, but the table should mark these entries more prominently as unphysical, or move them to a separate table, to avoid misuse by readers constructing mocks at high redshift.","section":"Table 2"},{"comment":"The caption of Fig. C.2 says 'Scatter fit (λ+ z50)' while the legend refers to 'Median fit'; please harmonize the labels. Also clarify how many halos have undefined z50 and how the sample restriction affects the comparison with Table 3.","section":"Appendix C.2 / Fig. C.2"},{"comment":"The sentence about the convolution with 0.25 dex is clear, but consider adding a sentence in the figure captions of Figs. 1-3 and A.1 stating that the observational-resolution convolution is applied only in that comparison, to avoid readers interpreting the model scatter in terms of this convolution.","section":"Section 3.1"}],"recommendation":"major_revision","confidential_remarks":"This is a solid paper in scope for the journal and the main scientific narrative is likely correct. The key concern is fixable: the concentration proxy needs validation at z=0 with direct NFW fits, and the scatter-model self-consistency claim should be reframed or strengthened with an out-of-sample check. If the authors address these points, the paper should be acceptable. The Appendix C.2 formation-time model indicates that the assembly-dependence claim is robust even if the concentration proxy is noisy at high redshift."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Kamran et al. does exactly what it says: it characterizes the HI-halo mass relation in GAEA2023 and packages it as a prescription for populating dark-matter halos. The median relation and fitting form are inherited from Spinelli et al./Baugh et al., but the systematic scatter analysis — spin, concentration, formation time, redshift evolution, residual scatter — is the new part, and it is done carefully. The z=0 comparison with ALFALFA/HIPASS, the conditional HIMF against Jones et al., and the comparison with other models are all transparent and honest.\n\nThe paper is also candid about its own limitations. It openly states that the z=0 HIMF is a calibration target, so agreeing with it is partly circular. It flags that the scatter model is a fit to the same simulation it is compared against. It reports that high-redshift fit parameters for the median relation are unconstrained. I appreciate that — the text does not oversell.\n\nThe most substantive concern is the concentration inversion. Eq. (4) recovers c_h from Vmax/Vvir assuming NFW and relaxed halos. At z~3, where most halos sit at c_h~3-5 close to the minimum of that curve, small errors in the input map to large errors in c_h. That will attenuate B_c and inflate sigma_int, so the fading concentration signal at high z is at least partly an artifact of the proxy. The paper acknowledges this, but it does not do the obvious z=0 check: direct NFW fits are possible for the z=0 snapshot and would validate or calibrate Eq. (4) without waiting for higher-z particle data. That feels like a missed opportunity.\n\nThat said, the concern does not sink the paper. The spin term is robust at all redshifts, and the formation-time model in Appendix C.2 (Table C.3) captures assembly information out to z=5 with nearly constant sigma_int ~0.3 dex. So the broader 'assembly matters' claim survives, and the user of the prescription has a better tool for high z. I would trust the z=0 concentration model and use the z50 version beyond z~2.\n\nThis is a useful, solid paper for the 21-cm mock community. It is not a breakthrough and it does not resolve a long-standing question, but it is a genuine step up from the usual single-scatter-number prescriptions. The math and the fitting procedures are straightforward and reproducible; the tabulated parameters make implementation trivial.\n\nMy recommendation: send it to peer review. It deserves a serious referee. I would suggest the referee push for the z=0 direct NFW validation (or an explicit statement of why it cannot be done) and for a clearer caveat that the sigma_int values at high z are partly proxy-limited. The paper will be better for it.","headline":"A careful, honest fitting paper that gives 21-cm mock builders a practical prescription — the concentration-based scatter model has a real but acknowledged weak spot at high z, and the z50 alternative should be used there.","tokens_in":37516,"tokens_out":3736,"would_cite":true,"duration_ms":37285,"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":"The hydrogen content of a dark-matter halo is set as much by its assembly history as by its mass.","keywords":["HI–halo mass relation","21-cm intensity mapping","halo assembly bias","halo spin","halo concentration","semi-analytic galaxy formation","neutral hydrogen","mock catalogs"],"falsifier":"Fit NFW profiles to the z=0 snapshot particle data and compare the directly measured concentrations to the Vmax/Vvir-inverted values used here; if the two disagree systematically at the ~0.1–0.2 dex level for halos near 10^12 solar masses, the quoted intrinsic scatter and the concentration term are compromised.","tokens_in":36420,"feed_emoji":"📡","tokens_out":4431,"duration_ms":47175,"temperature":0.7,"pith_summary":"This paper shows that the neutral-hydrogen (HI) content of a dark-matter halo is not a single function of halo mass but depends systematically on how the halo assembled. At fixed mass, faster-spinning, later-forming, and less-concentrated halos are systematically HI-richer, and this assembly-driven variation accounts for a large part of the observed scatter. The authors distill the median relation and the scatter into a compact analytic prescription built from halo-catalog quantities (mass, spin, concentration), leaving only a residual intrinsic dispersion of about 0.3 dex. Because 21-cm intensity-mapping surveys require large dark-matter-only catalogs populated with HI, a physically grounded prescription with realistic scatter is a practical tool for building mock observations and interpreting current and upcoming surveys.","feed_headline":"Halo spin and assembly set which halos keep hydrogen","feed_subtitle":"A compact model maps neutral hydrogen onto dark-matter halos, including assembly-driven scatter that matters for 21-cm surveys.","key_machinery":"The scatter plane: for each halo, the deviation of its HI content from the median relation at its mass is modeled as a linear function of standardized halo spin and concentration (Δ = C + A_λ x_λ + B_c x_c). This plane, combined with a Gaussian intrinsic scatter term, is what carries the argument that the scatter is predictable from catalog-level halo properties rather than being irreducible noise.","core_discovery":"The central discovery is that the substantial scatter (about 0.5 dex) in the HI–halo mass relation is not random: at fixed halo mass, higher-spin, later-forming, and less-concentrated halos hold significantly more HI. Spin is the dominant secondary property along the gas-rich rising branch, while concentration (or equivalently formation time) matters most near and above the quenching scale, and a standardized linear plane in spin and concentration, supplemented by a Gaussian residual of about 0.3 dex, reproduces the full scatter of the model from z=0 to z=5. This assembly dependence is encoded in a five-parameter median fit plus a two-predictor scatter model, expressed entirely in quantities","pith_inferences":["If the assembly-driven scatter is real, HI-selected samples should exhibit assembly bias in clustering: HI-rich halos of a given mass are likely more clustered if they assembled later, a prediction that could be tested by cross-correlating 21-cm intensity maps with galaxy surveys.","The same scatter model could be propagated into the predicted 21-cm auto-power spectrum, especially the shot-noise term, which the paper does not explicitly compute but whose amplitude depends directly on the halo-to-halo HI variance.","Because concentration becomes unreliable as a predictor at z≳3 due to the Vmax/Vvir inversion degeneracy, a practical hybrid prescription would use concentration at low redshift and formation time at high redshift, a choice the paper's own comparison supports.","The residual intrinsic scatter of about 0.3 dex may partly arise from baryonic feedback processes not captured by halo properties; if so, it represents a fundamental floor for any halo-based HI prescription, motivating future work that incorporates baryon-sensitive features."],"forward_implications":["The prescription allows dark-matter-only N-body catalogs to be populated with HI, including a halo-property-dependent scatter, enabling more realistic 21-cm mock catalogs for intensity-mapping surveys.","Accounting for the assembly dependence of HI content changes the predicted clustering and shot noise of the 21-cm field, because HI is preferentially found in later-forming, higher-spin halos.","At high redshift (z≳3), the median HI–halo mass relation becomes a nearly single power law, and the spin-plus-formation-time scatter model remains predictive, whereas concentration loses discriminating power.","The scatter peaks near the quenching scale (~10^12 solar masses), where the HI distribution is bimodal; a symmetric Gaussian residual under-represents the gas-poor tail, indicating limits of the simple scatter prescription."],"fun_headline_variants":["Spin and assembly dictate which halos hold hydrogen","Halo spin and assembly explain 21-cm scatter","Assembly, not chance, sets halo hydrogen content","Halo spin and formation time predict HI richness","HI scatter tied to halo spin and assembly"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The concentration used to predict HI scatter is recovered by inverting the Vmax/Vvir relation assuming an NFW profile and restricting to the monotonic branch c_h ≥ 2.16; if that inversion is biased, particularly for unrelaxed halos and at high redshift where concentrations cluster near the degeneracy minimum, the scatter model's decomposition is unreliable.","fun_headline_variants_meta":{"raw":{"variants":["Spin and assembly dictate which halos hold hydrogen","Halo spin and assembly explain 21-cm scatter","Assembly, not chance, sets halo hydrogen content","Halo spin and formation time predict HI richness","HI scatter tied to halo spin and assembly"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000563,"raw_usage":{"total_tokens":2591,"prompt_tokens":907,"completion_tokens":1684,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":651,"completion_tokens_details":{"reasoning_tokens":1612}},"tokens_in":651,"tokens_out":1684,"duration_ms":12720,"temperature":1.0,"reasoning_tokens":1612,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T10:41:59.325311+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fit NFW profiles to the z=0 snapshot particle data and compare the directly measured concentrations to the Vmax/Vvir-inverted values used here; if the two disagree systematically at the ~0.1–0.2 dex level for halos near 10^12 solar masses, the quoted intrinsic scatter and the concentration term are compromised.","supporting_citations":[],"review_version":1}