{"id":"6c3f5fbd-529d-4d41-83e0-74491bc6c152","arxiv_id":"2412.20415","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Simulations show that deprojection reduces beam-mismatch leakage to a negligible level for a third-generation ground-based CMB experiment, leaving lensing and tensor-to-scalar ratio measurements unbiased.","lead":"This study simulates a next-generation ground-based CMB experiment and tests whether deprojection can remove the signal distortions caused by mismatched detector beams. It finds that after deprojection and foreground cleaning, the remaining beam-systematic contamination is small enough not to bias CMB lensing measurements or the tensor-to-scalar ratio under the assumed sensitivity.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Noise-free deprojection is the load-bearing simplification: the paper does not propagate realistic TOD noise through the coefficient fits, so the claimed negligible residual is not yet demonstrated for S3 conditions.","rationale":"The reader's weakest assumption matches the most load-bearing simplification. The paper's positive results—undeprojected residual BB power 2–3 orders below input, lensing S/N close to the systematics-free value, and identical r upper limits—are all computed after deprojection coefficients have been estimated without S3 detector noise. Appendix A explicitly says the coefficient scatter is 'too optimistic' for this reason. This is not an internal inconsistency, and the paper deserves credit for stating the limitation, but the stated limitation directly affects the central claim: 'the residual leakage can be ignored in the following data analysis pipeline.' A noise-free fit is an idealized best case; real deprojection would estimate the same six modes from noise-dominated pair-difference TOD, and the resulting coefficient errors would leave additional T→P leakage. Uncorrelatedness of noise and templates controls bias of the coefficient estimates, not their variance. Averaging over ~106 time trunks per detector pair may reduce the effect, but the paper does not estimate this. The proposed test is a direct rerun with noise added before fitting, using the pipeline already built for the paper. If the test shows no shift beyond statistical uncertainty, the conditional verdict can be upgraded; if not, the conclusion should be restricted to the noise-free limit. No other concern (far sidelobes, template frequency mismatch, E-mode over-filtering) is as directly load-bearing, because the paper either excludes it explicitly or corrects for it internally. Therefore the reader's CONDITIONAL verdict is appropriate and should remain unchanged.","tokens_in":16489,"tokens_out":5129,"duration_ms":57587,"concrete_test":"Rerun the existing 301-realization pipeline with the nominal S3 white-noise simulations added to the pair-difference TOD before the template fitting in Section 2 (and, if available, a representative 1/f component after pair-differencing and polynomial filtering). Then propagate through deprojection, NILC/cILC, lensing, and the r MCMC. Compare the noise-degraded residual BB power after deprojection at 40<ell<200 with the systematics-free noise band and the r=0.01 spectrum. If the residual stays below ~10% of the r=0.01 signal and the r posterior shifts by <0.005, the concern is resolved; otherwise the noise-free claim does not generalize to S3.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central 'negligible residual' claim is established only for deprojection coefficients fitted to noise-free S3 TOD. Section 2 states 'Noise is not involved in the data since we expect that the noise is uncorrelated to the templates from Planck HFI maps,' and Appendix A concedes that the resulting scatter of recovered beam parameters is 'too optimistic since the noise was not added to the data.' Uncorrelatedness does not remove the problem: in a least-squares fit of the six leakage templates, additive detector noise contributes a zero-mean but nonzero-variance term to each fitted coefficient, and the residual after subtracting fitted templates inherits that variance. Since the headline conclusions are comparisons against S3 sensitivity (lensing S/N ~4.2, r upper limit 0.043), even a modest increase in residual B-mode power from noise-degraded deprojection could shift those numbers. The paper does not quantify how the per-chunk coefficient scatter averages down in map-making, so it is unknown whether the noise-free result survives realistic TOD noise.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents an end-to-end forecast of beam-mismatch deprojection for a third-generation ground-based CMB experiment (S3). It simulates 300 realizations with 7,000 polarized detectors at 95/150 GHz, adds stochastic beam mismatches (gain, beam width, pointing, and plus/cross ellipticity) to the time-ordered data, and applies the deprojection technique by fitting six leakage templates derived from Planck 100/143 GHz maps. The residual maps and power spectra are then propagated through NILC and cILC foreground cleaning, lensing reconstruction, and an MCMC estimate of the tensor-to-scalar ratio r. The authors find that the residual beam-systematic leakage after deprojection is negligible: the lensing signal-to-noise ratio is about 4.2 and the 95% upper limit on r is <0.043, nearly identical with and without beam systematics. They conclude that the residual leakage can be ignored in the S3 data analysis pipeline.","tokens_in":16695,"tokens_out":6910,"duration_ms":73485,"significance":"The paper is a detailed and careful forecast that exercises a full mock-observation and analysis pipeline, including scan strategy, pair differencing, component separation with NILC and cILC, lensing reconstruction, and r estimation. Its strengths are the explicit internal consistency checks: recovered beam parameters correlate strongly with inputs (Appendix A), the residual BB spectrum after deprojection is 2-3 orders of magnitude below the input spectrum (Section 4.2), and the r posterior is essentially unchanged by systematics (Section 4.5). If the deprojection implementation survives a realistic treatment of detector noise, the paper would provide a useful validation of a standard mitigation technique for S3-era experiments. At present, however, the central 'negligible residual' claim rests on a deprojection fit performed on noiseless TOD and on a self-consistent six-parameter mismatch model, so the headline result should be regarded as conditional on those choices.","major_comments":[{"comment":"The deprojection coefficients are fitted to noiseless TOD, and Appendix A concedes that the resulting scatter is 'too optimistic since the noise was not added to the data.' This is load-bearing for the central claim that the residual beam-systematic leakage can be ignored. In a least-squares fit of the six leakage templates to a pair-difference TOD containing additive detector noise, the fitted coefficients acquire a noise-induced variance proportional to (A^T A)^{-1} σ^2 even when the noise is uncorrelated with the templates; subtracting the fitted templates then leaves a residual term proportional to the template amplitude times the coefficient error. The paper does not quantify this coefficient noise or show how it averages down when the per-chunk fits are coadded in the map-making step of Section 4.1. Because the headline comparisons are made against S3 sensitivity (lensing S/N about 4.2, r upper limit 0.043), the conclusion that the residual leakage can be ignored is not yet demonstrated for realistic TOD noise. I request either simulations that include noise in the deprojection fit or an analytic estimate of the resulting residual B-mode power.","section":"Sect. 2 ('Noise is not involved...') and Appendix A"},{"comment":"The correction for the differential plus-ellipticity bias uses the mean recovered value, ⟨δp_recov⟩=7.2×10^{-3}, estimated from the paper's own systematics-free and noise-free simulations. If detector noise is added to the TOD, the distribution of fitted δp broadens, and any bias-correction scheme must be recomputed with matching noise so that the correction is unbiased for the actual analysis. The current comparison is self-consistent, but the specific numerical value of the correction and the resulting E-mode filtering residual may not transfer to a noise-realistic analysis. The revision should either re-derive the bias correction in the presence of detector noise or argue analytically why the noise-free value remains applicable.","section":"Sect. 4.1, Fig. 1 and Appendix A"},{"comment":"The injected beam mismatches are drawn from exactly the six-parameter family whose derivatives are used as deprojection templates, so the simulation demonstrates that the deprojection can remove contaminants of the same functional form; it does not independently validate the template basis against more general beam errors. Real beams can contain higher-order aberrations, frequency-dependent structure, or near-sidelobe features not representable by the six templates. The final sentence of Section 5 (\"the residual leakage can be ignored in the following data analysis pipeline\") is broader than what the simulation tests. I recommend qualifying the conclusion to the modeled six-parameter family, or adding at least one out-of-family mismatch (for example, a higher-order Hermite-Gauss mode or an asymmetric sidelobe) to test robustness.","section":"Sect. 2, Table 1, and Sect. 5"}],"minor_comments":[{"comment":"The text groups gain, bandpass, pointing, beam-width, and ellipticity differences under the term 'beam mismatch,' but only the spatial beam parameters in Table 1 are actually simulated and deprojected. Please clarify which of the enumerated error types are included in the mock and which are left for future work.","section":"Sect. 2"},{"comment":"The text refers to '301 simulation sets mentioned above,' while Section 2 states that the mock dataset consists of 300 simulated sky maps. Please clarify the consistency between these numbers.","section":"Sect. 4.4"},{"comment":"The caption states that the power spectra are shown with ℓmax = 1500, whereas the text in Sections 3.1 and 3.2 describes analysis with ℓmax = 2000. Please clarify whether the plotted spectra are truncated for display or whether the analysis was actually performed to a lower maximum multipole.","section":"Fig. 6 caption"},{"comment":"The deprojection residual is described as '2~3 orders of magnitude lower' than the input CMB-plus-foreground spectrum. Because this comparison is made for a single realization, quoting a one-realization estimate would be clearer; a small ensemble-averaged residual spectrum would better support the claim.","section":"Sect. 4.2, Fig. 3"}],"recommendation":"major_revision","confidential_remarks":"The reader's stress-test concern is legitimate and lands directly on the paper's central claim: the deprojection is performed on noiseless TOD, and the authors themselves concede in Appendix A that the resulting scatter is too optimistic. The paper is otherwise well constructed and the internal checks are consistent. A revision that adds a noise-inclusive deprojection test, or an analytic bound on the noise-induced residual, would substantially strengthen the conclusions. The self-consistency of the six-parameter mismatch model is a secondary but real scope limitation that should be acknowledged more explicitly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: this is a useful, incremental forecast that deprojection removes the T→P leakage from elliptical-Gaussian beam mismatches to a level that is negligible for an S3-like experiment, in their simulations. The main caveat, which the authors acknowledge, is that the deprojection coefficients are fit to noise-free TOD; the residual leakage is therefore probably underestimated, and the paper does not quantify the margin.\n\nWhat's new: Han et al. (2023b) used the same foreground-cleaning and lensing pipelines without beam systematics. The new piece is the injection of beam mismatches into the S3 TOD, the deprojection step, and the demonstration that the downstream results—foreground-cleaned T/E/B spectra, lensing S/N, and the r posterior—are essentially unchanged. That end-to-end demonstration is genuinely useful for pipeline planning.\n\nWhat's done well: The internal checks are consistent and reassuring. Appendix A shows the recovered beam parameters track the inputs strongly. After deprojection, the residual BB power is two to three orders of magnitude below the input BB at the frequencies shown. The most persuasive single result is the r posterior: the 95% upper limit is 0.043 with and without systematics, which is direct evidence that the systematic residual does not shift the main science target.\n\nSoft spots: (1) The noise-free fit is the load-bearing simplification. Least-squares fitting of six templates to pair-difference TOD with realistic white noise yields fitted coefficients with zero-mean but nonzero-variance errors; subtracting the best-fit templates leaves a residual that is the projection of the noise onto the template space. That residual averages down across chunks and detectors, but the paper does not quantify how much. The authors say in Appendix A that the scatter of recovered coefficients is 'too optimistic since the noise was not added to the data,' which is exactly the concern. (2) The injected mismatches have the same functional form as the deprojection templates, so this is a self-consistency test. It does not cover non-elliptical beam errors or far sidelobes, which the paper explicitly leaves out. (3) The residual spectra in Figs. 3 and 6 appear to be single realizations; a 300-simulation ensemble has been produced, so showing the spread of the residual power would make the 'negligible' claim far more robust.\n\nWho it's for: instrumental systematicists and pipeline developers for S3-class CMB experiments, particularly those considering deprojection for beam-mismatch control. It is a reasonable forecast, not a definitive proof.\n\nRecommendation: It deserves a serious referee. The referee should ask the authors to quantify the effect of noise on the deprojection fit, and to show the simulation-to-simulation scatter of the residual spectra, before the 'negligible residual' conclusion is taken as a general statement about S3.","headline":"A careful, internally consistent forecast that deprojection clears beam-mismatch leakage for S3, but noise-free fits make the residual estimate optimistic and the margin unquantified.","tokens_in":17237,"tokens_out":4415,"would_cite":true,"duration_ms":43118,"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":"Deprojection removes the beam-mismatch leakage that would otherwise bias next-generation CMB polarization measurements.","keywords":["cosmic microwave background","beam mismatch","deprojection","T-to-P leakage","tensor-to-scalar ratio","CMB lensing reconstruction","internal linear combination","foreground cleaning"],"falsifier":"Run the same deprojection and foreground-cleaning pipeline while adding white noise (and, where possible, 1/f noise) to the time-ordered data during the template fit, then compare the residual BB power and the inferred r bias against the systematics-free case. If the residual beam-systematic power rises above the noise floor or shifts r by more than the reported uncertainty, the paper's \"negligible residual\" conclusion fails.","tokens_in":16259,"feed_emoji":"📡","tokens_out":4330,"duration_ms":38961,"temperature":0.7,"pith_summary":"This paper forecasts whether the \"deprojection\" technique can remove the temperature-to-polarization leakage caused by mismatched detector beams in a third-generation ground-based cosmic microwave background experiment (S3). Using mock time-ordered data with beam mismatches at the level measured by BICEP2, the authors find that subtracting six beam-mismatch templates from the pair-differenced detector streams recovers the input power spectra. After foreground cleaning with the NILC and cILC methods, the residual systematics are too small to bias measurements of the T, E, and B modes, the CMB lensing potential, or the tensor-to-scalar ratio under S3 sensitivity. The paper's practical conclusion is that residual beam-mismatch leakage can be ignored in the S3 analysis pipeline.","feed_headline":"Deprojection wipes out beam-mismatch leakage for next CMB experiments","feed_subtitle":"Mock S3 data show residual systematics don't bias TEB maps, lensing, or the tensor-to-scalar ratio.","key_machinery":"The load-bearing object is the set of six leakage templates derived from the beam-smoothed temperature map $\\tilde{T}$ and its first and second spatial derivatives: $\\delta g\\,\\tilde{T}$, $\\delta\\sigma(\\nabla_x^2+\\nabla_y^2)\\tilde{T}$, $\\delta x\\,\\nabla_x\\tilde{T}$, $\\delta y\\,\\nabla_y\\tilde{T}$, $\\delta p(\\nabla_x^2-\\nabla_y^2)\\tilde{T}$, and $\\delta c\\,2\\nabla_x\\nabla_y\\tilde{T}$. Because map-making is linear, the total T-to-P leakage from small beam mismatches is a linear combination of these templates, so deprojection fits the six coefficients in time-ordered data and subtracts the fitted leakage. The fitted coefficients are then averaged over time chunks per detector pair and the maps are passed through NILC for T and E modes and cILC for B-mode foreground cleaning, with the residual evaluated by comparing systematics-added versus systematics-free maps.","core_discovery":"The central claim is that deprojection makes beam-mismatch systematics negligible for a third-generation ground-based CMB experiment. The authors model the differential beam of each detector pair with six parameters (gain, two pointing shifts, beamwidth, plus-ellipticity, and cross-ellipticity), generate mock S3 time-ordered data with those mismatches, and fit the six corresponding leakage templates to remove the T-to-P leakage before map-making. After propagating the deprojected maps through the full analysis chain, the residual contamination is shown to be far below the noise uncertainty for TT, EE, TE, and BB power spectra, and the reconstructed lensing potential and r constraints agree with the systematics-free case, giving a 95% upper limit r < 0.043 in both cases. This establishes deprojection as sufficient to prevent beam mismatch from biasing the science goals of S3.","pith_inferences":["If detector noise is added to the time-ordered data during the template fit, the recovered per-detector deprojection coefficients will scatter more, and whether the residual stays negligible will likely depend on how the noise is filtered before fitting.","The paper only treats T-to-P leakage; polarization-angle miscalibration (E-to-B leakage) and far sidelobes are explicitly left out, so the benign conclusion should not be read as covering all beam-related systematics.","The recovered bias on plus-ellipticity, about $7.3\\times 10^{-3}$, comes from the cosmological TE correlation; real data would likely need a systematics-free simulation set to calibrate this bias before science analysis.","A natural extension would be to test deprojection robustness with anisotropic or sidelobe-dominated beams rather than the elliptical-Gaussian differential beam model."],"forward_implications":["S3 experiments can adopt deprojection in their map-making pipeline and ignore residual beam-mismatch leakage in later analysis.","Residual systematics after deprojection do not bias NILC- and cILC-cleaned TT, EE, TE, or BB band powers at S3 sensitivity.","CMB lensing reconstruction from polarization retains essentially the same signal-to-noise ratio (about 4.2) as the systematics-free case.","The tensor-to-scalar ratio posterior is unchanged by beam systematics after deprojection, with a 95% upper limit of r < 0.043."],"supporting_citations":[{"why":"Supplies the six-parameter beam-mismatch model, the leakage template formalism, and the measured beam-parameter levels used to set the mock systematics.","marker":"BICEP2 Collaboration et al. 2015"},{"why":"Provides the result that small beam mismatches produce leakage equal to the beam-smoothed temperature map and its first and second derivatives.","marker":"Hu et al. 2003"},{"why":"Introduces the deprojection technique used to filter the leakage templates from time-ordered data.","marker":"Sheehy 2019"},{"why":"Establishes the NILC/cILC foreground-cleaning and lensing-reconstruction pipeline whose systematics-free signal-to-noise (about 4.5) serves as the baseline for comparison.","marker":"Han et al. 2023b"},{"why":"Provides the constrained ILC method used for B-mode foreground cleaning.","marker":"Remazeilles et al. 2021"},{"why":"Supplies the quadratic lensing estimator formalism used to reconstruct the lensing potential from polarization.","marker":"Carron & Lewis 2017"}],"fun_headline_variants":["Deprojection erases beam mismatch for CMB experiments","Deprojection eliminates beam leakage in CMB data","Beam mismatch no problem with deprojection for S3","Deprojection cleans beam systematics for next-gen CMB"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The deprojection coefficients are fitted to mock time-ordered data with no detector noise, so the recovered per-detector beam parameters are optimistic; if realistic noise degrades the template fit, the residual T-to-P leakage after deprojection would be larger than reported.","fun_headline_variants_meta":{"raw":{"variants":["Deprojection erases beam mismatch for CMB experiments","Deprojection eliminates beam leakage in CMB data","Beam mismatch no problem with deprojection for S3","Deprojection cleans beam systematics for next-gen CMB"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000564,"raw_usage":{"total_tokens":2685,"prompt_tokens":962,"completion_tokens":1723,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":578,"completion_tokens_details":{"reasoning_tokens":1655}},"tokens_in":578,"tokens_out":1723,"duration_ms":14151,"temperature":1.0,"reasoning_tokens":1655,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:23:37.347831+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same deprojection and foreground-cleaning pipeline while adding white noise (and, where possible, 1/f noise) to the time-ordered data during the template fit, then compare the residual BB power and the inferred r bias against the systematics-free case. If the residual beam-systematic power rises above the noise floor or shifts r by more than the reported uncertainty, the paper's \"negligible residual\" conclusion fails.","supporting_citations":[{"cited_title":"Deprojecting beam systematics for next-generation CMB B-mode searches","cited_arxiv_id":"1911.03547","evidence_quote":"Introduces the deprojection technique used to filter the leakage templates from time-ordered data."},{"cited_title":"2021, MNRAS, 503, 2478","cited_arxiv_id":null,"evidence_quote":"Provides the constrained ILC method used for B-mode foreground cleaning."},{"cited_title":"& Lewis, A","cited_arxiv_id":null,"evidence_quote":"Supplies the quadratic lensing estimator formalism used to reconstruct the lensing potential from polarization."}],"review_version":1}