{"id":"01e6f4a4-6a9d-4636-a2b3-a99b06e10b99","arxiv_id":"1908.03024","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"SKA-I is forecast to give the tightest dark-energy constraints of the three single-dish HI intensity mapping surveys, with combining Planck tightening w0 and wa further.","lead":"This paper forecasts how tightly three radio telescopes designed to map cosmic hydrogen, FAST, BINGO and SKA-I, could measure dark energy and other cosmological parameters after one year of observing. It is a planning tool that ranks the telescopes and shows how much extra constraining power comes from combining each with Planck.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Planck-combined forecasts drop two parameters instead of marginalizing over them, so the quoted combined errors and gains are likely over-tight.","rationale":"The reader flagged foreground cleaning and the unstated HI bias, both of which are real limitations. The bias omission affects the amplitude of all forecasts and is a legitimate condition for acceptance. However, the more crisp, internally testable flaw is the Planck-combination step. The paper's own text (Section 3.3) says Neff and Σmν 'were omitted' when combining with Planck. In a Fisher forecast, removing a parameter from the information matrix is not the same as marginalizing over it; it imposes infinite prior weight on the fiducial value. The combined columns of Table 3 therefore report errors that are not comparable to the marginalized errors in the same table for the IM-only rows or for Planck alone. A standard Schur-complement exercise (or simply embedding the Planck Fisher into the 9x9 IM Fisher and inverting) would settle whether the combined numbers move materially. The independent ranking of the three telescopes may survive, which is why the verdict remains CONDITIONAL rather than REJECT, but the exact quoted Planck gains in the Abstract are not yet supported. The bias concern is also important and should be addressed by stating b_HI(z); I would add that to the conditions as well.","tokens_in":23833,"tokens_out":13058,"duration_ms":143472,"concrete_test":"Recompute the combined forecast for SKA-I+Planck by embedding the 7x7 Planck Fisher matrix into the full 9x9 IM Fisher matrix (zero rows/columns for Neff and Σmν), inverting, and marginalizing to the 7x7 (w0, wa, ...) block. Compare the resulting marginalized errors with Table 3's SKA-I+Planck row. If the w0 error moves upward by more than 10% (e.g., from 0.0678 toward the naive ~0.10 combination), the quoted Planck gains are overstated. Repeating for BINGO and FAST will show whether the ranking survives.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Sections 3.3 and 4 state that the Planck Fisher matrix is added 'entry-wise' to the IM Fisher matrices for 7 parameters, and that Neff and Σmν are 'omitted' to match the Planck chains. In standard Fisher-matrix algebra, deleting a parameter from the Fisher matrix is equivalent to fixing it (a delta-function prior), not to marginalizing over it. The correct procedure is to add the 7x7 Planck Fisher into the 9x9 IM Fisher (zero blocks for the two omitted parameters), invert the combined 9x9 matrix, and then read off the 7x7 marginal covariance. Because the paper instead appears to combine the 7x7 sub-blocks directly (Table 3 shows '−−' for Neff and Ωνh2 in all combined columns), the quoted combined errors (e.g., SKA-I+Planck (w0, wa)=(0.0678,0.2679)) are conditional errors with Neff and Σmν held fixed. These are systematically tighter than the marginalized errors quoted for the IM-only forecasts (Section 4.1). The headline improvements relative to Planck alone and the cross-experiment comparisons therefore mix marginalized and conditional quantities, so the quantitative claims in the Abstract and Figures 3, 11-13 are not yet established as stated.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents Fisher-matrix forecasts of cosmological parameter constraints from single-dish HI intensity mapping with FAST, BINGO, and SKA-I. The signal is modeled through a tomographic angular power spectrum (Eqs. 3.1-3.4), the noise is treated as a diagonal thermal-noise covariance (Eq. 3.5), and a 9-parameter cosmological model is considered. The main quantitative claims are the marginalized 1σ errors (w0,wa) = (0.9293,3.5792), (0.4083,1.5878), and (0.3158,0.4622) for BINGO, FAST, and SKA-I, respectively, implying that SKA-I gives the strongest dark-energy constraints; combining with Planck tightens these to (0.0832,0.3520), (0.0791,0.3313), and (0.0678,0.2679). The paper also reports a trade-off between SKA-I and FAST across the other cosmological parameters and tests the robustness of the dark-energy results to discarding large angular scales (ℓ<10).","tokens_in":24143,"tokens_out":9161,"duration_ms":97647,"significance":"If the numbers are correct, the paper provides a useful, up-to-date comparison of three representative single-dish HI intensity mapping experiments and a transparent Fisher-matrix methodology that can be adapted to other surveys. Its strengths are the explicit signal and noise formulas, the use of narrow frequency channelizations, the comparison of old and updated SKA-I dish numbers, and the ℓ≥10 robustness test. The relative ranking of the three experiments for the dark-energy equation of state is plausible and likely robust to the issues discussed below. However, the absolute error forecasts rest on idealized assumptions (perfect foreground cleaning, diagonal noise, and an unspecified HI bias), and the Planck combination is not marginalized consistently. The absolute numbers and the quoted improvement percentages relative to Planck alone are therefore not all established as stated.","major_comments":[{"comment":"The Planck combination as implemented is not a marginalization over the full 9-parameter space. The text states that the Planck Fisher matrix is added 'entry-wise' for Ωb h², Ωc h², w0, wa, ln(10¹⁰As), H0, and ns, while Neff and Σmν/94.07 eV are omitted. Deleting rows and columns of a Fisher matrix before inversion is equivalent to fixing those parameters at their fiducial values, not to marginalizing over them. The IM-only columns in Table 3 are marginalized errors from the 9×9 Fisher matrix, whereas the combined columns are conditional errors from a 7×7 matrix. The correct procedure is to add the 9×9 Planck Fisher matrix (with zero blocks for the two parameters on which Planck has no information) to the 9×9 IM Fisher matrix, invert the full sum, and then extract the 7×7 marginal covariance. Until this is done, the quoted combined errors and the improvement percentages relative to Planck alone mix marginalized and conditional quantities, and the headline combined numbers in the Abstract and in Figures 3, 11, and 13 should be regarded as preliminary.","section":"Section 3.3, Table 3"},{"comment":"The signal amplitude in the forecast depends on the HI density contrast δn, but the HI clustering bias b_HI is never specified. In intensity mapping the HI density contrast is normally written δn = b_HI δm, and without a value (or a redshift-dependent model) for b_HI the amplitude of Cℓ, and therefore the absolute 1σ errors in Tables 3 and 4, is not uniquely determined. Ω_HI is fixed to 0.62×10⁻³, but b_HI is a separate and equally important multiplier. If the authors implicitly set b_HI = 1, this should be stated and justified. This issue does not necessarily change the relative ranking, but it prevents the quoted absolute error bars from being reproduced.","section":"Section 3.1, Eq. (3.4)"},{"comment":"The forecast assumes that foreground emission is cleaned perfectly and that the noise covariance is diagonal. Because the dark-energy information in this analysis comes substantially from large angular scales (ℓ≥2) and foreground residuals are expected to be largest there, the absolute constraints are optimistic. The ℓ≥10 test in Table 4 is a useful sensitivity check and shows that the ranking is stable, but it also shows a non-negligible loss of constraining power: for SKA-I alone, σ(w0) degrades from 0.3158 to 0.4059 and σ(wa) from 0.4622 to 0.5735. The headline numbers still include ℓ=2 and should be presented with a clear caveat that they are idealized upper limits under perfect foreground removal.","section":"Section 4.2 and Eq. (3.12)"}],"minor_comments":[{"comment":"The dish count for SKA-I is inconsistent: Table 2 and Section 2.3 use Nant = 133, while the text near Figure 8 and the final paragraph of Section 4.1 state that 'in this forecast we assume that the SKA-I project is an integrated 190 15-metre single-dishes in autocorrelated mode.' Please clarify which configuration was used for the Table 2 forecasts and, if necessary, recompute the affected numbers.","section":"Section 2.3 vs Section 4.1"},{"comment":"The FAST improvement over BINGO for w0 is quoted as 56.04% in the Abstract and 56.06% in Section 4.1; the latter follows from the stated numbers, so the Abstract should be corrected.","section":"Abstract vs Section 4.1"},{"comment":"The caption describes the columns as 'covariance matrices' but the entries are 1σ errors; please use consistent terminology such as '1σ marginalized errors'.","section":"Table 3 caption"}],"recommendation":"major_revision","confidential_remarks":"The Planck marginalization issue is the key technical concern; I would ask the authors to rerun the combined forecasts by inverting the full 9×9 Fisher sum and then marginalizing. The unspecified HI bias is also a blocking issue for reproducibility of the absolute numbers. I do not see a fundamental novelty problem; the comparative forecast is within the journal's scope and would be acceptable once the technical points are fixed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper's main qualitative answer—SKA-I single-dish beats FAST, which beats BINGO, for w0–wa—survives contact with the text and with the ℓ≥10 test. The Planck-combined numbers, however, are not what they are labeled. The stress-test note is right: the authors add the Planck Fisher matrix 'entry-wise' for seven parameters and simply omit Neff and Σmν. In a 9×9 Fisher forecast, dropping a row/column is a delta-function prior, not a marginalization. The correct move is to embed the Planck prior in the 9×9 problem (zero blocks for the omitted parameters, or the full Planck covariance), invert, then marginalize. As written, the SKA-I+Planck (w0, wa) = (0.0678, 0.2679) and the 'improvement relative to Planck alone' percentages mix conditional quantities with the marginalized IM-only numbers. The combined claims in the abstract and Section 4 need to be redone or relabeled.\n\nWhat is genuinely useful here is the updated, like-for-like comparison: 133-dish SKA-I specs, FAST at 10,000 deg2 and 300-m aperture, BINGO at 3,000 deg2; 10 MHz and 1 MHz channelization; and the ℓ≥10 robustness check. That last test matters because large-scale foreground cleaning is the usual Achilles' heel for IM forecasts; discarding ℓ<10 weakens constraints but does not change the ranking. The FAST/SKA-I trade-off—FAST better on ns and small-scale parameters, SKA-I better on wa and Ωνh2—is physically sensible and is the kind of result survey planning can use. The authors are also honest about the diagonal-noise, perfect-foreground assumption.\n\nSoft spots, in proportion. The unstated HI bias is a real gap: Eq. (3.4) uses δn but the paper never says what b_HI is, and the absolute Fisher errors scale with the signal amplitude. This is less serious for the relative ranking, which is the paper's main claim. Foreground residuals and 1/f noise are deferred, which is fine for a forecast but should be labeled as optimistic. No code is provided; for a Fisher forecast with standard formulas that is an inconvenience, not a sin. The self-citation to Li & Ma and Xu, Ma & Weltman is fine.\n\nBottom line: this is a competent planning-level forecast with one load-bearing methodological mistake in the Planck-combined section. Send it to a serious referee; require them to fix the 7×7/9×9 combination and restate the combined errors as properly marginalized (or clearly labeled conditional). The IM-only comparison is worth keeping.","headline":"The IM-only ranking (SKA-I > FAST > BINGO) is probably right; the Planck-combined numbers are over-tight because two parameters were dropped instead of marginalized.","tokens_in":24643,"tokens_out":5004,"would_cite":false,"duration_ms":55012,"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 paper forecasts that SKA-I single-dish HI intensity mapping gives the tightest dark-energy constraints among BINGO, FAST, and SKA-I, with one-year 1σ errors (w0, wa) = (0.3158, 0.4622).","keywords":["HI intensity mapping","21 cm cosmology","Fisher matrix forecast","dark energy equation of state","BINGO telescope","FAST telescope","SKA-I telescope","tomographic angular power spectrum"],"falsifier":"Measure the off-diagonal foreground-residual noise matrix in the first year of real single-dish HI data: if $N_\\ell$ has significant correlations between frequency channels at $\\ell<10$, the headline SKA-I errors $(0.3158, 0.4622)$ move toward the paper's own $\\ell\\geq10$ values $(0.4059, 0.5735)$, and the forecast is optimistic. A second check is to measure $\\Omega_{\\rm HI} b_{\\rm HI}$ directly from the cross-correlation of an HI intensity map with an optical galaxy survey; a value different from the fiducial $0.62\\times10^{-3}$ rescales the signal amplitude in Eq. (3.2) and with it all the quoted constraints.","tokens_in":23651,"feed_emoji":"📡","tokens_out":13336,"duration_ms":121340,"temperature":0.7,"pith_summary":"This paper forecasts what three radio telescopes—BINGO, FAST, and SKA-I in single-dish mode—would learn about cosmology from one year of neutral-hydrogen (HI) intensity-mapping observations. Its central result is a ranking: SKA-I would give the tightest dark-energy equation-of-state constraints, with marginalized $1\\sigma$ errors $(w_0, w_a) = (0.3158, 0.4622)$, followed by FAST and then BINGO; relative to BINGO these are improvements of 66.02% on $w_0$ and 87.09% on $w_a$. Adding Planck data tightens SKA-I to $(0.0678, 0.2679)$, a 37.22% and 30.33% improvement over Planck alone. The forecast also finds a parameter-by-parameter trade-off: FAST wins on small-scale-sensitive parameters such as $n_s$, $\\Omega_c h^2$, $H_0$, and $\\ln(10^{10}A_s)$, while SKA-I wins on $w_0$, $w_a$, $\\Omega_\\nu h^2$, and $N_{\\rm eff}$. The authors intend these numbers as a benchmark for the relative capabilities of the next-generation HI intensity-mapping surveys.","feed_headline":"SKA-I wins the dark-energy forecast over FAST and BINGO","feed_subtitle":"One-year HI intensity mapping with SKA-I reaches (w0, wa) errors of (0.316, 0.462), up to 87 percent better than BINGO.","key_machinery":"The load-bearing object is the tomographic HI angular power spectrum $C_\\ell^{ij}$ of Eq. (3.1), computed from a transfer function that includes redshift-space distortions and is multiplied by a Gaussian beam window. It is compared with the single-dish thermal noise power spectrum $N_\\ell^{ij} = \\delta^{ij} T_{\\rm sys}^2 S_{\\rm survey}/(N_{\\rm ant} N_{\\rm feed} t_{\\rm TOT}\\Delta\\nu)$ via the Fisher matrix $F_{\\alpha\\beta} = f_{\\rm sky}\\sum_\\ell (2\\ell+1)/2\\,\\mathrm{tr}[C_{\\ell,\\alpha}\\Sigma_\\ell C_{\\ell,\\beta}\\Sigma_\\ell]$ with $\\Sigma_\\ell=(C_\\ell+N_\\ell)^{-1}$ and diagonal noise. This converts telescope specifications (dish diameter, receiver temperature, frequency range, survey area, channel width) into predicted marginalized $1\\sigma$ errors for nine cosmological parameters, and lets the paper rank the three experiments and their Planck combinations.","core_discovery":"The paper claims that SKA-I, used as an array of single dishes in autocorrelation mode, will impose the most stringent dark-energy equation-of-state constraints of the three HI intensity-mapping experiments considered, reaching marginalized $1\\sigma$ errors $(w_0, w_a) = (0.3158, 0.4622)$ after one year of observation. FAST improves on BINGO by 56.04% on $w_0$ and 55.64% on $w_a$, and SKA-I improves on BINGO by 66.02% and 87.09%. When each experiment's Fisher matrix is added to Planck's, SKA-I + Planck gives $(0.0678, 0.2679)$, which is 18.51% and 23.89% tighter than BINGO + Planck and 37.22% and 30.33% tighter than Planck alone. The paper further claims that across the nine cosmological parameters no single experiment dominates: FAST's larger dish gives it the edge on small angular scales ($n_s$, $\\Omega_c h^2$, $H_0$, $\\ln(10^{10}A_s)$), while SKA-I's wider frequency coverage (350–1050 MHz) gives it the edge on $w_0$, $w_a$, $\\Omega_\\nu h^2$, and $N_{\\rm eff}$.","pith_inferences":["The paper's own $\\ell\\geq10$ table suggests the headline $(w_0, w_a)$ errors are best-case numbers: if foreground residuals survive at $\\ell<10$, SKA-I's $w_0$ error grows from 0.3158 to 0.4059, so the 66% and 87% improvements over BINGO should be read as contingent on clean large-scale modes.","Because the signal amplitude entering Eq. (3.2) is set by $\\Omega_{\\rm HI}$ and an unspecified HI clustering bias, a direct measurement of $\\Omega_{\\rm HI} b_{\\rm HI}$ from HI-galaxy cross-correlation would convert these relative rankings into calibrated absolute predictions.","A natural next step, not taken here, is to combine FAST and SKA-I Fisher matrices: their complementary strengths (small angular scales versus wide frequency range) suggest joint constraints could beat either experiment alone, especially on $w_a$ and $n_s$.","The 1 MHz channelization test points to a computational and modelling cost: real receivers will have correlated noise across adjacent narrow channels, so the diagonal-noise approximation that carries this forecast will need to be relaxed in the analysis pipeline."],"forward_implications":["SKA-I single-dish HI intensity mapping, run for one year, is predicted to give marginalized $1\\sigma$ errors $(w_0, w_a) = (0.3158, 0.4622)$, about 66% tighter on $w_0$ and 87% tighter on $w_a$ than BINGO.","Combining SKA-I with Planck tightens these to $(0.0678, 0.2679)$, a 37.22% improvement over Planck alone on $w_0$ and 30.33% on $w_a$.","FAST is predicted to lead on small-scale-sensitive parameters ($n_s$, $\\Omega_c h^2$, $H_0$, $\\ln(10^{10}A_s)$), while SKA-I leads on $w_0$, $w_a$, $\\Omega_\\nu h^2$, and $N_{\\rm eff}$; neither survey dominates the full parameter set.","Using 1 MHz rather than 10 MHz frequency channels substantially improves constraints by preserving redshift-space-distortion information along the line of sight.","Removing the foreground-contaminated modes $\\ell<10$ inflates SKA-I's $w_0$ error from 0.3158 to 0.4059, so the quality of foreground cleaning directly controls whether the headline precision is reached."],"supporting_citations":[{"why":"Supplies the diagonal-noise simplification and the telescope-parameter formulas used for the noise matrix.","marker":"Li & Ma (2017)"},{"why":"Provides the forecast methodology and the central-redshift bin approximation that the paper extends to nine cosmological parameters.","marker":"Bull et al. (2015b)"},{"why":"Gives the updated 133-dish SKA-I configuration and the frequency-dependent system-temperature model used for SKA-I noise.","marker":"Square Kilometre Array Cosmology Science Working Group et al. (2018)"},{"why":"Supplies the fiducial cosmology and the Planck TT+TE+EE+lensing Fisher matrix added to each HI experiment for combined constraints.","marker":"Planck Collaboration et al. (2016)"},{"why":"Provides the transfer-function and redshift-space-distortion formalism for the tomographic angular power spectrum, including its bandwidth dependence.","marker":"Hall et al. (2013a)"},{"why":"Gives the HI mean brightness-temperature formula that sets the signal amplitude in the angular power spectrum.","marker":"Chang et al. (2008)"},{"why":"Supplies the fiducial Ω_HI = 0.62×10^-3 used in the brightness-temperature calculation.","marker":"Switzer et al. (2013)"},{"why":"A prior comparative forecast of FAST versus BINGO that the paper checks its ranking against.","marker":"Bigot-Sazy et al. (2016)"},{"why":"Defines BINGO's dish, feed-horn, and frequency specifications used as inputs to the signal and noise calculations.","marker":"Battye et al. (2016)"}],"fun_headline_variants":["SKA-I tops HI intensity mapping for dark energy precision","SKA-I outdoes FAST and BINGO in dark energy forecasts","Dark energy constraints: SKA-I leads, 87% better than BINGO","HI mapping forecast: SKA-I wins over FAST and BINGO","SKA-I bests FAST, BINGO in dark energy constraints"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Every quoted error bar assumes foregrounds are cleaned perfectly so that the noise between different frequency channels is uncorrelated and diagonal, and it further assumes a fixed HI signal amplitude even though the clustering bias multiplying that amplitude is never specified in the paper.","fun_headline_variants_meta":{"raw":{"variants":["SKA-I tops HI intensity mapping for dark energy precision","SKA-I outdoes FAST and BINGO in dark energy forecasts","Dark energy constraints: SKA-I leads, 87% better than BINGO","HI mapping forecast: SKA-I wins over FAST and BINGO","SKA-I bests FAST, BINGO in dark energy constraints"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000621,"raw_usage":{"total_tokens":2974,"prompt_tokens":1134,"completion_tokens":1840,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":750,"completion_tokens_details":{"reasoning_tokens":1744}},"tokens_in":750,"tokens_out":1840,"duration_ms":13249,"temperature":1.0,"reasoning_tokens":1744,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:27:33.565298+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the off-diagonal foreground-residual noise matrix in the first year of real single-dish HI data: if $N_\\ell$ has significant correlations between frequency channels at $\\ell<10$, the headline SKA-I errors $(0.3158, 0.4622)$ move toward the paper's own $\\ell\\geq10$ values $(0.4059, 0.5735)$, and the forecast is optimistic. A second check is to measure $\\Omega_{\\rm HI} b_{\\rm HI}$ directly from the cross-correlation of an HI intensity map with an optical galaxy survey; a value different from the fiducial $0.62\\times10^{-3}$ rescales the signal amplitude in Eq. (3.2) and with it all the quoted constraints.","supporting_citations":[{"cited_title":"R., Masui, K","cited_arxiv_id":null,"evidence_quote":"Supplies the fiducial Ω_HI = 0.62×10^-3 used in the brightness-temperature calculation."},{"cited_title":"A., et al","cited_arxiv_id":null,"evidence_quote":"A prior comparative forecast of FAST versus BINGO that the paper checks its ranking against."},{"cited_title":"2016, arXiv:1610.06 826 iii, v","cited_arxiv_id":null,"evidence_quote":"Defines BINGO's dish, feed-horn, and frequency specifications used as inputs to the signal and noise calculations."}],"review_version":1}