{"id":"3e115280-15f8-41e3-b05f-0eeb923dc1fc","arxiv_id":"1908.08128","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"High-redshift line intensity mapping could add about 11 percent to CMB-S4 delensing performance, but only if line-of-sight density modes survive foreground removal.","lead":"This paper forecasts that adding high-redshift line intensity mapping data to CMB-S4 delensing improves the removal of lensing B-modes by about 11 percent. The gain is real but conditional on preserving the lowest-frequency line-of-sight modes, which are exactly where foregrounds are strongest.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 11% delensing gain hinges on recovering k_parallel≈0 IM modes after foreground removal; the paper explicitly concedes this is unproven, so the headline should remain conditional.","rationale":"The reader identified the same load-bearing assumption, and the paper itself concedes it is unproven. I considered two alternative concerns: the sensitivity feasibility in Section III.C and line shot noise. The feasibility section only demonstrates partial saturation for [CII] (SNR 1e9 giving 6%, versus SNR 1e10 needed for saturation), which would reduce but not eliminate the headline. Shot noise would similarly degrade the correlation rather than nullify it. The foreground-mode assumption is more load-bearing because it is the difference between an 11% improvement and zero improvement, and it is a capability, not a derived result. The central calculation is otherwise coherent: the kernels, Limber integrals, multi-tracer combination, and residual-BB formula follow standard delensing formalism, and the paper is honest about the dominant caveat. The conditional verdict is therefore appropriate, with no change needed.","tokens_in":13286,"tokens_out":14304,"duration_ms":160125,"concrete_test":"Build an end-to-end simulation for one IM band (e.g., [CII] at 6<z<7): inject a fiducial clustering signal plus realistic smooth-spectrum foregrounds (Galactic synchrotron, CIB, CMB) into a frequency cube, add instrumental noise and beam/chromaticity, then apply the direct foreground subtraction described in §III.D.1 using external CMB maps. Form the 2D map in the bin and compute the recovered ρ_IMκ at ℓ<1000. Accept the 11% headline only if the recovered ρ_IMκ is within 10% of its no-foreground value; if it is substantially lower, re-run the forecast with the recovered ρ and rescale the claimed improvement accordingly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III.B and Figure 3 report the headline gain (α=4.9→5.43, 11%) under the assumption that the IM maps contain the line-of-sight projected density with only instrumental noise. This is mathematically equivalent to assuming that the k_parallel≈0 mode of each Δz=1 bin survives cleaning. Section III.D.1 explicitly states that smooth-spectrum foregrounds preferentially populate low-k_parallel modes, that simply excluding these modes would remove any correlation of the HI map with CMB lensing, and that if [CII] experiments cannot recover the line-of-sight density, their maps will not aid in delensing. The paper proposes direct subtraction for [CII] but supplies no simulation, end-to-end pipeline, or data demonstration that the k_parallel≈0 mode can be recovered with sufficient fidelity after foreground removal. Because a 2D projection of a 3D density field correlates with the CMB lensing kernel only through the radial mean, any foreground strategy that marginalizes over or high-pass filters smooth spectral components removes exactly the information the 11% number depends on. The failure mode is binary: if the modes are lost, the IM contribution to delensing is zero, not slightly degraded. The paper's own 'dominant concern' language is therefore the correct reading, and the headline result is conditional on an unproven capability.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper asks whether tomographic line intensity mapping (IM) surveys covering 2<z<10 can improve the delensing of degree-scale CMB B-modes beyond a baseline of CMB-S4 internal delensing plus CIB delensing. The author constructs a standard projected-tracer framework: Limber angular power spectra are computed from lensing kernels for the CMB, CIB, and top-hat IM redshift bins; instrumental noise is added; correlation coefficients are combined; and the residual BB power is converted to an improvement factor α for σ(r). The baseline α=4.9 improves to α=5.43 (about 11%) if the IM maps reach sufficiently high SNR, corresponding to σ(r) roughly 5e-4 to 4.5e-4. Integration-time estimates for CO/HI and [CII] surveys are given, and foreground effects are discussed, with interloper masking costing about 10% for [CII] and smooth-spectrum foregrounds identified as the dominant concern.","tokens_in":13569,"tokens_out":13118,"duration_ms":137541,"significance":"If the numerical result is taken at face value, the paper provides a useful planning result: high-redshift IM is unlikely to be a transformative delensing tool at CMB-S4 sensitivity, but it may be worth including as a byproduct tracer, and it becomes more valuable if internal or CIB delensing underperform. The calculation is transparent and reproducible from the stated equations; it is a forward model with no parameters fitted to the target result, and the saturation improvement is robust to the unknown line brightness because the bT amplitude cancels in the correlation coefficient at high SNR. The paper also deserves credit for clearly identifying the requirement to preserve k_parallel roughly equal to 0 modes, which is the actual obstacle rather than raw sensitivity.","major_comments":[{"comment":"The central 11% result (alpha=4.9 to 5.43) is computed in Section III B from IM maps with statistical noise only, i.e., it assumes the k_parallel roughly equal to 0 line-of-sight modes survive foreground cleaning. Section III D 1 states that smooth-spectrum foregrounds preferentially populate low-k_parallel modes, that excluding them removes any correlation with CMB lensing, and that if [CII] experiments cannot recover the line-of-sight density their maps will not aid in delensing. The paper therefore identifies a binary failure mode, but it does not quantify it or label Figure 3 and Table I as idealized upper bounds in the main text; the abstract's hedge is not repeated in Section III B or the Conclusions. I request a quantitative treatment of a k_parallel cutoff (e.g., recomputing rho and alpha after removing modes below a series of thresholds) or, at minimum, a prominent statement that the quoted improvements are ceilings contingent on an unproven foreground-removal capability.","section":"III D 1 / Figure 3"},{"comment":"The baseline alpha=4.9 against which the 11% improvement is measured depends on the Hall et al. CIB kernel, Eq. (3), yet the paper acknowledges in Section IV that the CIB kernel is a large source of error in CIB delensing. Because the marginal value of the IM tracers is defined relative to this baseline, the headline percentage is sensitive to CIB kernel assumptions; the paper's bracketing scenarios remove CIB entirely, but do not vary the kernel shape. A simple scan over zc, sigma_z, and beta (or a comparison with an alternative CIB model) would show whether the 11% number is stable, and would make the forecast more robust.","section":"II A / IV, Eq. (3)"}],"minor_comments":[{"comment":"The phrase 'unable to to correlate' contains a duplicated 'to'; please fix.","section":"Abstract"},{"comment":"The justification for assuming constant bT is relegated to a footnote; since this assumption directly affects the relative sensitivity of the IM bins, the test should be described in the main text or an appendix.","section":"III A, footnote [47]"},{"comment":"The statement that masked interloper voxels are 'uncorrelated with the structure that lenses the CMB' is too strong because lower-redshift matter also lenses the CMB; the correct point is that interlopers trace a different, largely disjoint lensing kernel, so the correlation with the target high-redshift signal is reduced rather than absent.","section":"III D 2"},{"comment":"The caveat 'Foreground mode loss is not included' appears only in figure captions; an explicit sentence in Section III B stating that the quoted improvements are upper bounds would make the status of the headline result unambiguous before the foreground discussion.","section":"III B / Figures 3 and 4"},{"comment":"The integration-time forecasts are quoted for single model line strengths from Refs. [28] and [54] without a range; a brief scaling of the required time with (bT)^-2 over the range of published model predictions would make the feasibility claims more transparent.","section":"III C"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the manuscript is a single-author forecast that is transparent about its main caveat, and the conditional language in the abstract is appropriate. The central issue is not internal inconsistency but the need to make the upper-bound status of the headline number explicit in the main text and to add a quantitative treatment of the foreground-mode-loss failure mode. My recommendation of major revision reflects these requests rather than a rejection of the paper's conditional logic."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"One-liner: a careful, well-scoped forecast that high-redshift line intensity mapping could improve CMB-S4 delensing by ~11% at best (α 4.9→5.43), but the headline is conditional on keeping the k_parallel≈0 modes after foreground removal—a capability the paper does not demonstrate and explicitly admits is unproven.\n\nWhat's new: the specific quantitative forecast for IM tracers over 2<z<10 against a CMB-S4 internal+CIB baseline, including the saturation analysis and feasibility estimates for planned CO, HI, and [CII] experiments. Prior work (Sigurdson & Cooray 2005) considered a different, much harder redshift range. The formalism is standard Limber projections; the calculation is a forward model, not a fit, and the saturation result is independent of the uncertain line amplitude. That makes the ceiling robust.\n\nWhere it's soft: the 11% is the best case. As Section III D 1 says, if you cannot measure the line-of-sight mean density (k_parallel≈0), the IM maps contribute nothing to delensing—not a degraded version, zero. That is binary, and there is no simulation or end-to-end pipeline showing smooth-spectrum foregrounds can be removed without killing those modes. So the headline should stay conditional. Two lesser concerns: the forecast has no error bars, and the CIB kernel uncertainty is not propagated, which matters when comparing 4.9 to 5.43. The constant-bT assumption is tested in a footnote with a varying line-strength model and shifts α by 1%, which is reassuring.\n\nThe math and citation pattern look solid. The paper is honest about its own limits—the 'dominant concern' language is accurate. This is a useful contribution for people planning CMB-S4 delensing strategies and IM survey designs. I'd send it to a referee who knows both internal delensing and IM foregrounds; the referee's job is to check the k_parallel argument and the noise curves, not to find a hidden error. I recommend engaging with it.","headline":"Solid forecast: IM could add ~11% to CMB-S4 delensing, but only if k_parallel≈0 survives foreground cleaning—a binary condition the paper honestly flags.","tokens_in":14085,"tokens_out":2809,"would_cite":true,"duration_ms":25726,"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":"Adding high-redshift line intensity mapping to CMB delensing improves tensor-to-scalar constraints by about 11 percent, if line-of-sight density modes survive foreground removal.","keywords":["cosmic microwave background B-modes","delensing","tensor-to-scalar ratio","line intensity mapping","cosmic infrared background","CMB lensing","epoch of reionization"],"falsifier":"Measure the cross-correlation between foreground-cleaned line intensity maps and a CMB lensing reconstruction at $\\ell<100$, comparing runs that keep and exclude modes with $k_\\parallel \\lesssim 0.02\\,h\\,\\mathrm{Mpc}^{-1}$; if the recovered line-of-sight modes show no significant correlation with lensing after foreground subtraction, the predicted 11% delensing improvement is not realized.","tokens_in":13105,"feed_emoji":"🔭","tokens_out":8322,"duration_ms":70795,"temperature":0.7,"pith_summary":"This paper asks whether maps of spectral-line emission from galaxies at redshifts 2 to 10, made by coarse-beam line intensity mapping, can sharpen upcoming searches for primordial gravitational waves. It finds that adding such maps to the planned combination of internal CMB delensing and cosmic infrared background delensing improves the delensing figure of merit by about 11 percent, lowering the projected tensor-to-scalar uncertainty from roughly $5\\times10^{-4}$ to $4.5\\times10^{-4}$. That gain requires retaining the line-of-sight density modes, which smooth-spectrum foreground cleaning tends to remove; without them the IM maps lose their correlation with the lensing field entirely. The paper also shows the needed map depths are plausible with next-generation instruments, though the line strengths remain uncertain.","feed_headline":"Line-intensity maps could improve CMB B-mode delensing by 11 percent","feed_subtitle":"Faint spectral-line maps could cut the projected tensor-to-scalar error by 11 percent — if line-of-sight modes survive.","key_machinery":"The engine is the effective correlation coefficient $\\rho_\\ell$ between the CMB lensing kernel and an optimally weighted combination of all available tracers, built from each tracer's redshift kernel and noise power spectrum. For IM maps the redshift kernel is $W(z)=b(z)\\,T(z)\\,dN/dz$, with redshift fixed by the observed line, and the noise follows a single-frequency CMB map model. The correlation is converted into a delensing improvement factor $\\alpha=\\sigma_0(r)/\\sigma_d(r)$ via the residual lensing B-mode power, so the paper's central number is the ratio of projected $\\sigma(r)$ with and without IM data. A critical feature is that delensing uses only the line-of-sight mean density, $k_\\parallel \\approx 0$, which is exactly the component that smooth-spectrum foregrounds dominate.","core_discovery":"The paper claims that at CMB-S4 sensitivity, the combination of internal CMB delensing and CIB delensing reaches an improvement factor $\\alpha=4.9$, removing about 88% of the lensing B-mode power; adding saturated tomographic line intensity maps spanning $2<z<10$ raises this to $\\alpha=5.43$, an 11% improvement, corresponding to a reduction in projected $\\sigma(r)$ from roughly $5\\times10^{-4}$ to $4.5\\times10^{-4}$. Saturation requires very high signal-to-noise in the maps, but the required depths are feasible for planned next-generation instruments if the emission lines are near their predicted strengths. The gain is contingent on recovering the line-of-sight density modes at $k_\\parallel\\approx 0$; without those modes the IM maps cannot correlate with the lensing kernel and provide no delensing benefit.","pith_inferences":["The $k_\\parallel=0$ requirement implies that IM experiments whose analysis pipelines avoid the foreground wedge, as is standard for power-spectrum measurements, will not automatically deliver delensing products; survey planning must include foreground subtraction and map-making that preserve line-of-sight density.","The paper cites mode-coupling techniques for reconstructing the lost $k_\\parallel\\approx 0$ modes, suggesting a testable extension: run foreground-cleaned IM simulations and check whether the reconstructed line-of-sight modes recover the expected correlation with a known lensing field.","Because the CIB redshift kernel is uncertain, the exact baseline $\\alpha=4.9$ is not fixed; precise IM measurements could help pin down the CIB kernel, which would shift the quoted percentage improvement.","A roughly 10% sharper $\\sigma(r)$ matters most in the pessimistic regime where $r$ is small and the only route to improved inflation constraints is more efficient delensing."],"forward_implications":["At saturated signal-to-noise, IM low ($2<z<6$) alone improves $\\alpha$ by 4%, IM high ($6<z<10$) by 7%, and both together by 11%, reaching $\\alpha=5.43$.","If smooth-spectrum foregrounds force exclusion of the $k_\\parallel\\approx 0$ modes, the IM maps do not correlate with the lensing kernel and add nothing to delensing.","IM delensing becomes more valuable if the baseline underperforms: without CIB the gain from both IM surveys is 19%, without internal delensing it is 37%, and with neither it is 104%.","Achieving saturation requires long integrations on the deep CMB patch, with estimates of thousands of hours for a Stage-2 HI survey and a few years for a large [CII] spectrometer array, assuming line strengths near current models.","Because the IM maps are byproducts of surveys built for other cosmology goals, the delensing gain comes at little extra cost, and additional tracers hedge against systematics in any single delensing map."],"supporting_citations":[{"why":"Sets the nominal CMB-S4 sensitivity at which lensing B-mode noise dominates and defines the $\\sigma(r)\\approx5\\times10^{-4}$ target.","marker":"[6]"},{"why":"Supplies the iterative EB estimator used to compute the internal lensing reconstruction noise spectrum for the baseline.","marker":"[11]"},{"why":"Provides the optimal multi-tracer combination formula and frames CIB delensing as a comparable external-tracer method.","marker":"[14]"},{"why":"Gives the baseline $\\alpha=4.9$ for internal plus CIB delensing and the comparison point for tomographic tracer delensing.","marker":"[18]"},{"why":"Defines the Stage-2 HI experiment parameters and line-temperature model used to forecast IM-low saturation time.","marker":"[28]"},{"why":"Supplies the CO line model and survey design used to estimate integration time for low-redshift IM delensing.","marker":"[34]"},{"why":"Provides the CIB redshift kernel model used in computing CIB correlation with the CMB lensing field.","marker":"[40]"},{"why":"Provides the [CII] line strength model used to forecast IM-high sensitivity and integration time.","marker":"[54]"},{"why":"Underpins the caveat that excluding low-$k_\\parallel$ modes removes the correlation between HI maps and CMB lensing.","marker":"[55]"},{"why":"Cited as the proposed mechanism for reconstructing the lost $k_\\parallel\\approx0$ modes from small-scale and large-scale mode coupling.","marker":"[62, 63]"}],"fun_headline_variants":["High-z line maps sharpen B-mode delensing by 11%","IM delensing gains 11% if low-k line-of-sight modes present","Line intensity mapping improves delensing 11%, if line-of-sight modes survive","11% better B-mode delensing with high-redshift line maps","High-z IM maps boost delensing 11% only with line-of-sight modes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central claim depends on removing smooth-spectrum foregrounds from the intensity maps without discarding the line-of-sight density modes at $k_\\parallel \\approx 0$; if those modes are cut or marginalized away, the maps no longer correlate with the CMB lensing kernel and the 11% improvement vanishes.","fun_headline_variants_meta":{"raw":{"variants":["High-z line maps sharpen B-mode delensing by 11%","IM delensing gains 11% if low-k line-of-sight modes present","Line intensity mapping improves delensing 11%, if line-of-sight modes survive","11% better B-mode delensing with high-redshift line maps","High-z IM maps boost delensing 11% only with line-of-sight modes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000964,"raw_usage":{"total_tokens":4084,"prompt_tokens":904,"completion_tokens":3180,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":3079}},"tokens_in":520,"tokens_out":3180,"duration_ms":22323,"temperature":1.0,"reasoning_tokens":3079,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:48:49.420443+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the cross-correlation between foreground-cleaned line intensity maps and a CMB lensing reconstruction at $\\ell<100$, comparing runs that keep and exclude modes with $k_\\parallel \\lesssim 0.02\\,h\\,\\mathrm{Mpc}^{-1}$; if the recovered line-of-sight modes show no significant correlation with lensing after foreground subtraction, the predicted 11% delensing improvement is not realized.","supporting_citations":[{"cited_title":"Future cosmic microwave background delensing with galaxy surveys","cited_arxiv_id":"1710.11038","evidence_quote":"Gives the baseline $\\alpha=4.9$ for internal plus CIB delensing and the comparison point for tomographic tracer delensing."},{"cited_title":"MKID development for SuperSpec: an on-chip, mm-wave, filter-bank spectrometer","cited_arxiv_id":"1211.1652","evidence_quote":"Provides the [CII] line strength model used to forecast IM-high sensitivity and integration time."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Underpins the caveat that excluding low-$k_\\parallel$ modes removes the correlation between HI maps and CMB lensing."}],"review_version":1}