{"id":"422066a4-b2c6-46ae-acfe-974851a69f08","arxiv_id":"2607.26180","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":11,"one_line_summary":"Radio recombination lines contaminate the post-reionization 21 cm power spectrum at roughly 10^-4 of the signal near BAO scales, shifting BAO extrema by a few x 10^-6 to 10^-5—well below the cosmic-variance limit.","lead":"This paper estimates how much radio recombination lines from hydrogen and carbon contaminate the 21 cm hydrogen signal used to map the large-scale structure of the universe. It finds that realistic models keep the contamination far below the level that could bias baryon acoustic oscillation measurements in upcoming 21 cm intensity mapping surveys.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central claim rests on calibrating RRL optical depths at z<0.02 and extrapolating to z>0.5 where the contaminating lines originate; §5 admits high-z HII regions could be denser/more covering, but no physical upper bound is derived, so the >10x safety margin is an assertion.","rationale":"The reader's weakest assumption correctly identifies the load-bearing extrapolation: the empirical RRL optical-depth calibration is anchored at z<0.02, while the contamination relevant for BAO at z21=0.5-2 originates from lines emitted at z_RRL≈0.5-2. The paper's own conclusions acknowledge that high-redshift galaxies may be more compact and have denser HII regions, both of which would raise emission, but the claimed safety margin is not quantified. I found no internal inconsistency in the power-spectrum formalism; the derivation of the cross-power spectrum, window functions, and BAO-shift estimate are self-consistent. The issue is genuinely empirical. Because the most dangerous scaling is linear in f τ_RRL, an order-of-magnitude increase in the high-z optical-depth normalization could move the predicted BAO shift from ~10^-5 to ~10^-4, and a factor of ~30 would approach the cosmic-variance floor. The suggested test directly targets this scaling by adopting observationally motivated high-z HII-region properties and recomputing the headline figure. The current paper is a careful, well-posed calculation with honest caveats, and the CONDITIONAL verdict is appropriate; my stress test does not change that verdict, but it sharpens the specific condition that would invalidate the central claim.","tokens_in":21442,"tokens_out":8356,"duration_ms":87840,"concrete_test":"Compute the contamination and BAO shifts using a high-z HII-region evolution model: adopt n_e(z) = 10^3 cm^-3 × ((1+z)/2)^α with α=1.5-2, f(z) rising from 0.3 at z=0 to 0.6 at z=2, and EM = n_e^2 L with L=1 pc. Propagate these through eqs. (2.10)-(2.15) and the power-spectrum pipeline to recompute fig. 7. If the median |Δk/k0| at z21=1 exceeds ~3×10^-4, or if stacked contamination exceeds 10^-3 P21 at k~0.1 h/Mpc, the conclusion that RRLs are negligible for BAO cosmology is not robust. A complementary observational check: use the two z>0.5 RRL detections (Emig+19, Emig+23) to place a 95% upper limit on fτ_RRL at n~170 at z~1; if this limit is more than an order of magnitude above Model C/D, the current safety margin is unconstrained.","verdict_should_be":"UNCHANGED","load_bearing_attack":"At the heart of the conclusion are eqs. (2.10)-(2.12): the RRL brightness temperature and all derived power spectra scale linearly in f τ_RRL ΔV (and the 21 cm-RRL cross term is linear in T_RRL). The calibration in fig. 2 uses almost exclusively z<0.02 starbursts (refs. 2,32,40,44,61), with only two z>0.5 detections. Yet fig. 5 shows the contamination at k=0.1 h/Mpc for z21=0.5-2 is dominated by RRLs with n~130-250, i.e., emitted at z_RRL≈0.5-2, not at the calibrated z≈0. The models bracket local HII-region parameters, but high-z ISM is expected to have higher pressure, more compact HII regions, and larger covering fractions. Since τ_RRL ∝ EM T_e^{-5/2} b_n β_n (eq. 2.14), and EM ∝ n_e^2 L, the observed factor ~5-10 increase in n_e from z=0 to z~2 (Davies et al. 2021) with fixed physical size raises EM by 25-100, pushing f τ_RRL well above Model D. Because the dominant cross-power term scales linearly with T_RRL, the BAO shift scales roughly linearly with f τ_RRL: a factor ~30 increase moves the median |Δk/k0| from ~3×10^-5 to ~10^-3, at the cosmic-variance floor. The §5 statement that it 'would be challenging to devise a model that exceeds our predictions by more than an order of magnitude' is not a derived bound; the two z>0.5 detections do not constrain population scatter or the high-EM tail. The central claim is therefore only as secure as the untested assumption that high-z HII regions are no more extreme than the local bracketing models.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a formalism for estimating radio recombination line (RRL) contamination of post-reionization 21 cm intensity mapping power spectra. It derives a brightness-temperature expression for individual RRLs following Petrovic & Oh (2011) with a documented correction, calibrates five physically motivated models (A–D for hydrogen, E for carbon) against extragalactic RRL observations, and computes the RRL auto-power, 21 cm–RRL cross-power, and RRL–RRL cross-power spectra, including finite-bandwidth and beam-window effects. For the fiducial Model C, contamination near BAO scales is ~10^-4 P21 and the inferred shift in BAO extrema is a few x 10^-6 to 10^-5, below the cosmic-variance floor. The paper concludes that RRLs are a negligible systematic for post-reionization 21 cm cosmology and that the PO11 fixed-optical-depth model substantially overestimates the contamination.","tokens_in":21947,"tokens_out":14006,"duration_ms":134878,"significance":"If correct, this is a valuable negative result for a key science driver of 21 cm intensity mapping: it would remove RRLs as a concern for BAO measurements and correct an alarming extrapolation of the PO11 model. The paper's strengths include an explicit power-spectrum formalism with a verified shot-noise limiting case, a documented correction to the PO11 brightness-temperature relation, and empirical calibration to a compilation of RRL observations rather than tuning to the BAO-shift target. The central claim, however, rests on an extrapolation from low-redshift (z<0.02) RRL observations to the higher-redshift lines that dominate the contamination; this extrapolation is acknowledged as a limitation but is not bounded quantitatively.","major_comments":[{"comment":"The empirical calibration in Fig. 2 uses almost exclusively z<0.02 starbursts (refs. 2,32,40,44,61), with only two z>0.5 detections, while Fig. 5 shows that the dominant contamination at k=0.1 h/Mpc for z21=0.5–2 comes from RRLs with n~130–250, emitted at z_RRL≈0.5–2. Since T_RRL and all derived power spectra are linear in f τ_RRL ΔV (eqs. 2.10–2.12) and τ_RRL ∝ EM T_e^-5/2 (eq. 2.14), the high-redshift extrapolation is load-bearing. Observed electron densities increase by factors of 5–10 from z=0 to z~2 (Davies et al. 2021); with fixed physical size this raises EM by factors of 25–100. A factor ~30 increase in T_RRL would move the median BAO shift from ~3×10^-5 to ~10^-3, at the cosmic-variance floor. The §5 statement that 'it would be challenging to devise a model that exceeds our predictions by more than an order of magnitude' is not derived. The free-free factor e^{-τ_ff} in eq. (2.1","section":"§5, eqs. (2.10)–(2.14), Fig. 2, Fig. 5"},{"comment":"The abstract claims the shift in the BAO extrema is 'at least two orders of magnitude smaller than relevant for percent-precision cosmology,' but the quantitative basis is the median shift of a few×10^-6 to a few×10^-5 in Fig. 7. The bars in Fig. 7 extend to a few×10^-4 at some redshifts, which is only about one order below the cosmic-variance floor of ~10^-3, not two. Since the central message is a safety margin, the claim should be based on the envelope over models and redshifts, not the median, or the wording should be adjusted to match the plotted maximum values.","section":"§4.2, Fig. 7, abstract"}],"minor_comments":[{"comment":"The abstract and Conclusions disagree on the size of the safety margin ('at least two orders of magnitude' vs. 'lower ... by about an order of magnitude'). Please harmonize these statements with the quantitative results in Fig. 7.","section":"Abstract vs §5"},{"comment":"The typesetting of these equations is ambiguous (e.g., multiple (1+z) factors in numerators and denominators). Please clarify the derivation so the redshift scaling can be checked step by step.","section":"Eqs. (2.5), (2.10), (2.11)"},{"comment":"The caption refers to 'cross markers' while the text mentions 'star symbols' and 'dashed segments'; please make the marker definitions consistent and explicit.","section":"Fig. 2 caption and text"},{"comment":"The paper rightly cautions that the carbon departure coefficients from [46] may be unreliable. Since carbon is subdominant this is not a blocking issue, but the caution currently appears only in the text near eq. (2.14); it would be clearer if also noted in the Table 1 caption.","section":"§2, Model E"},{"comment":"The perturbative Newton's-method expression is clear, but the text should state explicitly which k-range is used for the median and maximum reported in Fig. 7, since the limits (0.05<k<0.3 h/Mpc) affect the quoted safety margin.","section":"§4.2, eq. (4.4)"}],"recommendation":"major_revision","confidential_remarks":"This is a competent and potentially important paper. The main issue is that the central 'safe' conclusion depends on a high-redshift extrapolation that is acknowledged but not bounded. I would support publication once the authors either provide a quantitative physical upper bound (e.g., using free-free saturation or high-z ISM constraints) or explicitly reframe the conclusions as conditional on the local calibration. I do not see grounds for rejection; the formalism and the empirical comparison are sound within their stated assumptions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a credible, careful forecast that radio recombination lines are not going to bias post-reionization 21 cm BAO measurements at a level worth worrying about. The power-spectrum formalism is new, the empirical calibration is a real step beyond PO11's fixed-optical-depth model, and the main conclusion holds up for the models considered. The one genuine soft spot is the extrapolation from z<0.02 RRL calibrators to the z~0.5-2 emission that actually drives the contamination; the authors acknowledge it, but they don't close it.\n\nWhat's new: they correct the PO11 brightness-temperature coefficient, construct physically motivated HII-region models spanning a wide range of densities and emission measures, calibrate them to the available extragalactic RRL detections, and generalize the Lidz & Taylor interloper formalism to include 21-RRL and RRL-RRL cross-power spectra. The window-function treatment looks right and the shot-noise limiting case checks out. The BAO-shift calculation is straightforward and the numbers--median fractional shifts around 10^-5, at least two orders of magnitude below the cosmic-variance floor--follow from their models. The figures are thorough, and the limitations section is honest.\n\nThe soft spot is exactly what you'd expect: the empirical calibration in fig. 2 is almost entirely z<0.02 starbursts, with only two detections at z>0.5, and the extrapolation to high redshift leans on the assumption that high-z HII regions are no more extreme than the local bracketing models. The authors include Model D as an extreme case, and they note that higher ISM pressure and covering fractions could push things up, but they do not derive a physical upper bound. I take the stress-test arithmetic seriously--a factor ~30 in fτ could move the BAO shift to ~10^-3--but I don't think it sinks the paper. Model D already sits above most plausible extrapolations, and the safety margin is roughly two orders of magnitude, not a hair. The more immediate issue is reproducibility: the hydrogen departure coefficients are from private communication and the code isn't released. For a forecast meant to be cited, that should be fixed.\n\nThe paper is for anyone designing or interpreting CHIME, PUMA, SKA1-MID, or similar 21 cm intensity mapping efforts, and for theorists working on line interlopers. It deserves a serious referee. I'd send it out, with the request that the authors publish the departure-coefficient tables and code, and either justify or soften the claim that no model can exceed their predictions by more than an order of magnitude--ideally by giving an explicit upper bound on high-z covering fraction and emission measure. With those changes, I'd be glad to cite it.","headline":"Solid forecast that RRLs are a negligible systematic for post-reionization 21 cm BAO surveys; the main caveat is an acknowledged but unclosed extrapolation from local calibrators.","tokens_in":22493,"tokens_out":5514,"would_cite":true,"duration_ms":53907,"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":"Radio recombination lines contaminate the 21 cm intensity-mapping power spectrum at only about 10^-4 of the signal on BAO scales, shifting inferred BAO positions by a few parts in 10^5 or less.","keywords":["radio recombination lines","21 cm intensity mapping","baryon acoustic oscillations","power spectrum contamination","H II regions","post-reionization cosmology","interloper lines"],"falsifier":"Measure the RRL luminosity per unit star formation (equivalently f τ ΔV at n~170) in star-forming galaxies at z>1 with a deep low-frequency radio survey. If stacked emission from high-redshift H II regions exceeds the bracketed local values by more than an order of magnitude, or if a 21 cm intensity-mapping survey detects oscillatory contamination at k~0.1 h/Mpc above ~10^-4 of the 21 cm power, the paper's conclusion fails.","tokens_in":21269,"feed_emoji":"📡","tokens_out":5843,"duration_ms":50147,"temperature":0.7,"pith_summary":"This paper asks whether radio recombination lines—spectral lines from hydrogen atoms recombining at very high principal quantum numbers—contaminate the 21 cm intensity-mapping signal that future surveys will use to measure the baryon acoustic oscillation (BAO) scale. The authors build empirically calibrated models of the RRL emission from H II regions, replacing an earlier fixed-optical-depth model that had suggested severe contamination at low redshift. They find that RRL contamination of the 21 cm power spectrum is highly oscillatory in wavenumber but small: near BAO scales it reaches only about 10^-4 of the 21 cm power, and it shifts the inferred BAO peak and trough positions by a few parts in 10^6 to 10^5, at least two orders of magnitude below the cosmic-variance limit. Carbon RRLs are subdominant. If correct, RRLs are not the systematic that limits post-reionization 21 cm cosmology.","feed_headline":"Radio recombination lines shift 21 cm BAO scale by ~10^-5","feed_subtitle":"Empirically calibrated models put the contamination below a part in 10,000 of the 21 cm power on BAO scales.","key_machinery":"The central object is the radio recombination line (RRL)—an electronic transition between very high principal quantum numbers (n>86) in hydrogen and carbon atoms—which forms a picket fence of spectral lines that redshift into a 21 cm observing band. The argument is carried by a brightness-temperature model parameterized by the frequency-integrated RRL optical depth times the covering fraction of the galaxy's radio continuum (f τ ΔV), calibrated to published H II region observations. On top of that, the paper builds a power-spectrum formalism that includes the RRL auto-spectrum plus 21 cm-RRL and RRL-RRL cross-spectra, each carrying a line-of-sight displacement phase exp(-ik∥ Δx) and a survey","core_discovery":"The paper's central claim is that radio recombination lines do not jeopardize the BAO program in post-reionization 21 cm intensity mapping. Using models whose optical depths are tied to observed RRL measurements from H II regions—rather than a fixed upper-bound optical depth used in earlier work—the authors show that the total RRL contamination (auto-power plus 21 cm-RRL and RRL-RRL cross-power) is about 10^-4 of the 21 cm auto-power at k~0.1 h/Mpc at all redshifts from 0 to 6. The contamination is dominated by cross terms between lines emitted close in space to the 21 cm emitters, producing rapid oscillations in k, but the resulting shift in the BAO extrema is a few×10^-6 to a few×10^-5, we","pith_inferences":["The empirical calibration rests almost entirely on low-redshift starburst galaxies; if high-redshift H II regions are much denser or cover more of the galaxy's radio continuum, RRL contamination could rise by an order of magnitude and approach the cosmic-variance floor, which the paper acknowledges but does not bound physically.","The predicted oscillatory signature in k is a fingerprint: a future 21 cm survey that sees unexplained wiggles at the predicted frequencies could test this model directly, while a null detection would not strongly constrain it since the predicted amplitude sits below current sensitivity.","The same cross-term formalism could be reused to assess line confusion in other line-intensity mapping efforts where closely spaced spectral lines overlap in the observing band.","If RRLs are this benign, the next interloper worry for 21 cm BAO surveys shifts to continuum emission such as OH masers and foreground subtraction, which the paper notes can affect only broadband amplitude rather than BAO positions."],"forward_implications":["RRL contamination near BAO scales (k~0.1 h/Mpc) is about 10^-4 of the 21 cm power at all redshifts, so it cannot by itself bias upcoming 21 cm BAO measurements.","The shift in inferred BAO extrema is typically 10^-6 to 10^-5, and at most a few×10^-4, at least an order of magnitude below the cosmic-variance limit for percent-precision cosmology.","The earlier fixed-optical-depth RRL model overestimates contamination by orders of magnitude at z<2; physically calibrated models are needed for reliable forecasts.","Carbon RRLs are subdominant to hydrogen RRLs in post-reionization 21 cm intensity mapping.","Because the contamination is oscillatory, it could in principle be detected and removed using cross-correlations with other large-scale-structure tracers, as the paper notes in its conclusions."],"fun_headline_variants":["Radio lines can't budge 21 cm BAO scale","RRL contamination 10^-4 of 21 cm power on BAO","Empirical RRL models slash BAO contamination","21 cm BAO robust against radio recombination lines","RRL noise too small to shake 21 cm BAO"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The models are calibrated almost entirely on nearby starburst galaxies; if distant galaxies' star-forming gas is much denser or covers more of the galaxy's radio emission, the contamination could rise an order of magnitude and approach the cosmic-variance floor.","fun_headline_variants_meta":{"raw":{"variants":["Radio lines can't budge 21 cm BAO scale","RRL contamination 10^-4 of 21 cm power on BAO","Empirical RRL models slash BAO contamination","21 cm BAO robust against radio recombination lines","RRL noise too small to shake 21 cm BAO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00046,"raw_usage":{"total_tokens":2158,"prompt_tokens":780,"completion_tokens":1378,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":524,"completion_tokens_details":{"reasoning_tokens":1295}},"tokens_in":524,"tokens_out":1378,"duration_ms":12439,"temperature":1.0,"reasoning_tokens":1295,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T00:33:41.705729+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the RRL luminosity per unit star formation (equivalently f τ ΔV at n~170) in star-forming galaxies at z>1 with a deep low-frequency radio survey. If stacked emission from high-redshift H II regions exceeds the bracketed local values by more than an order of magnitude, or if a 21 cm intensity-mapping survey detects oscillatory contamination at k~0.1 h/Mpc above ~10^-4 of the 21 cm power, the paper's conclusion fails.","supporting_citations":[],"review_version":1}