{"id":"9d43227a-4bac-40d5-ae7a-c309492c2c19","arxiv_id":"2507.08048","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Deep MWA imaging of a candidate Epoch of Reionization field finds 2,576 radio sources and shows PCA foreground subtraction leaves residual power more than an order of magnitude above the predicted 21 cm signal.","lead":"Astronomers imaged a patch of sky with the Murchison Widefield Array radio telescope to test whether it is quiet enough to search for the faint signal from the universe's first stars and galaxies.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The PCA-3 residual power spectrum is never compared to a noise floor, so the residual may be noise-dominated; the conclusion that PCA foreground subtraction is insufficient is therefore unsupported.","rationale":"The reader's CONDITIONAL verdict is appropriate, but the specific weakest_assumption they identified (foregrounds occupying ≤3 spectral modes and the EoR signal surviving PCA) is not the most load-bearing issue. The most load-bearing concern is that the paper never demonstrates that the PCA-3 residual is foreground power rather than instrumental noise. Without a noise floor, Figures 14 and 15 only show that the residual exceeds the cosmic signal, which is trivially expected for a 4.4-hour observation at these frequencies. The paper's Table 2 already suggests the noise floor is far above the EoR signal, so the comparison to 21cmFAST alone cannot support the conclusion that 'further improvements in data reduction and foreground subtraction techniques are necessary.' That conclusion requires either a noise power spectrum showing the residual is above it, or an explicit statement that the residual is foreground-dominated. Since this is an easily fixable omission rather than a fatal flaw in the source counts or field characterization, the verdict remains CONDITIONAL; the stress-test does not change the reader's verdict but identifies a different, more fundamental condition that must be checked.","tokens_in":23477,"tokens_out":9735,"duration_ms":105124,"concrete_test":"Compute a noise-only angular power spectrum from the same data and overplot it on Figures 14 and 15: for example, difference two independent halves of the observations, run both halves through the identical PCA-3 pipeline, and take C_l of the difference map divided by 2; or, alternatively, generate Monte Carlo noise realizations using the measured per-pixel RMS and the known beam and pass them through the same PCA-3 procedure. If the PCA-3 residual C_l is statistically consistent with this noise C_l, the residual is noise-dominated and the paper's central claim about foreground subtraction insufficiency does not follow.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 5.2 (Figures 14 and 15) compares the angular power spectrum of the PCA-3 residual map only against the 21cmFAST cosmological signal. No noise power spectrum is shown, so the residual cannot be attributed to residual foregrounds. This matters because the paper's central conclusion — that standard PCA foreground subtraction is insufficient for deep EoR imaging — depends on the residual being foreground-dominated. The paper's own Table 2 gives a thermal noise of 0.51 mJy/beam at 216 MHz and a measured RMS of 1.80 mJy/beam; after converting to brightness temperature and accounting for the beam, the expected noise C_l lies many orders of magnitude above the EoR signal. If the PCA-3 residual C_l is consistent with this noise floor, then the residual contains no measurable foreground power and the correct conclusion is that the observation is sensitivity-limited, not that foreground subtraction is inadequate. Equation 10 propagates only cosmic variance, not instrumental noise, and no noise realization or noise-only power spectrum is provided anywhere in Section 5.2.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents MWA Phase II extended-array observations of a candidate SKA EoR field (G0044; center RA 8h, Dec +5 deg), producing deep images at 88 and 216 MHz with 4.43 h total integration. The authors construct a 2,576-source catalogue at 216 MHz, measure source counts and completeness, estimate thermal and confusion noise, and use PCA to remove foregrounds from four 7.68 MHz sub-band images per band. They compute angular power spectra of the PCA residuals and compare them to 21cmFAST predictions, finding that the residuals remain more than an order of magnitude above the expected EoR signal, and conclude that standard PCA foreground subtraction is insufficient for deep EoR imaging in this field.","tokens_in":23665,"tokens_out":4363,"duration_ms":46560,"significance":"The observational products -- source catalogue, completeness simulations, noise estimates, and field characterization -- are useful inputs for SKA1-Low field selection and for testing calibration and imaging pipelines. The source counts and confusion-noise estimates, if carefully propagated with uncertainties, are of interest to the low-frequency community. The PCA residual power-spectrum comparison, however, is currently the weakest part of the paper: the absence of a noise power spectrum means the central conclusion about foreground subtraction is not yet supported. With the addition of a noise comparison and a more careful treatment of the PCA-mode test, the paper could become a solid observational reference for this candidate EoR field.","major_comments":[{"comment":"The residual angular power spectra are compared only to the 21cmFAST cosmological signal; no instrumental noise power spectrum or noise-only realization is shown. With the stated thermal noise of 0.51 mJy/beam at 216 MHz and a measured RMS of 1.80 mJy/beam (Table 2), the expected noise C_l can lie many orders of magnitude above the EoR signal, so the observed factor-of-ten excess may be entirely consistent with the observation being sensitivity-limited rather than foreground-limited. The conclusion that 'standard PCA foreground subtraction is insufficient' therefore requires a demonstration that the residual power is above the noise floor; please add a noise power spectrum (analytic or from a noise-only simulation) and error bars that include the noise contribution.","section":"5.2 (Figs. 14-15)"},{"comment":"The statement that 'nearly 100% of the radio sources in observation have been extracted by PCA-3' is by construction and does not validate foreground removal. With four sub-band images, removing three principal components leaves a one-dimensional residual subspace, so any source whose spectral signature is correlated across sub-bands will be assigned to the PCA-reconstructed maps. The source-matching statistic therefore only shows that the reconstructed images retain the positions of detected sources, not that the residual map is free of foreground contamination or that the cosmological signal survives the projection. A test using injected simulated signals (for example, adding a mock EoR signal at the image level and checking its recovery rate) is needed before this claim can be assessed.","section":"5.1, Table 9"},{"comment":"The SKA1-Low confusion-noise numbers are obtained by integrating the differential source-count fit (Eq. 1) that was derived from the same MWA image. While the text notes this at the end of Section 3.3.3, the quantitative statement that 'confusion noise will be the primary factor' for SKA1-Low should be accompanied by a propagation of the fit uncertainties (k and gamma) and by an explicit statement that the prediction is an extrapolation of the MWA counts, not an independent estimate. This matters because S_lim = 5 sigma is also set from the MWA RMS, and the resulting confusion noise may carry a large systematic uncertainty.","section":"3.3.3, Eq. (3)"}],"minor_comments":[{"comment":"The variance expression uses only cosmic variance and f_sky; please add a noise term or note explicitly that the plotted error bars are cosmic-variance-only, otherwise the error bars in Figures 14 and 15 are incomplete.","section":"5.2, Eq. (10)"},{"comment":"The text says the integration times correspond to stacking 5, 10, 40, and all snapshots, but the figure legends label them as 10, 20, 80, and 338/266 minutes; please make the integration-time notation consistent.","section":"3.1 and Figs. 5-6"},{"comment":"The column headers repeat the frequency labels in a way that is difficult to parse; please reformat the header so that it is clear which columns correspond to measured RMS, thermal noise, and confusion noise at 88 MHz and 216 MHz.","section":"Table 2"},{"comment":"Completeness corrections are listed as '——' for several high-flux bins; please either define this symbol (presumably unity) or list the actual values.","section":"Table 8"},{"comment":"The phrasing 'nearly all resolved radio sources can be successfully removed using PCA' overstates what is demonstrated; suggest replacing 'removed' with 'identified' or 'reconstructed in the PCA foreground model.'","section":"Abstract and Section 6"},{"comment":"There are a few citation style inconsistencies, for example '(Offringa et al. 2015)' appears parenthetically in one place while 'Sokolowski et al. 2017' does not; please ensure uniform citation formatting.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"This is essentially an observational data paper with a foreground-subtraction science claim. The data products are likely useful to the MWA/SKA community, and the paper is within the scope of MNRAS. The main risk is that the central conclusion is currently unsupported by the lack of a noise power spectrum; this is fixable in revision. I see no citation or novelty concerns beyond the usual need to frame the SKA confusion-noise estimates as extrapolations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The useful core is the deep MWA imaging of the G0044 field at 88 and 216 MHz, the source catalogue (2,576 sources at 216 MHz, 90% complete at 10.4 mJy), and the completeness and confusion-noise estimates. That is genuine progress for SKA EoR field planning: one candidate quiet field gets real measurements instead of extrapolation. The reduction pipeline is standard and careful—AEGEAN source finding, BANE noise maps, injected-source completeness simulations, flux and astrometric checks against GLEAM and NVSS. The catalogue is published with the paper, which makes the results independently usable. I trust the source counts and the noise numbers.\n\nThe soft spots are in the PCA foreground-removal section. The residual angular power spectra after removing three PCA components are compared only to the 21cmFAST EoR signal. No noise power spectrum is shown. With the measured RMS of 1.8 mJy/beam at 216 MHz and the thermal noise estimate of 0.5 mJy, the noise C_ell almost certainly sits well above the EoR signal. So the finding that the residual is an order of magnitude above theory could simply mean the observation is sensitivity-limited, not that foreground subtraction is insufficient. That needs to be checked and stated. Related: the PCA uses only four 7.68-MHz sub-bands, so three components is a very small model; the paper doesn't test how many spectral modes the foregrounds actually occupy.\n\nThe confusion-noise forecasts for SKA are obtained by integrating the source-count fit derived from this same field, so they are not independent predictions; they are an estimate of what this field's source population implies. That is acceptable if flagged, but the paper should say it more plainly. Also, Table 9's near-100% source match after PCA-3 is essentially by construction, since the PCA components are computed from the same maps; it is a sanity check, not external validation.\n\nOverall: the observational catalogue and noise characterization deserve to be published and would be useful to anyone planning deep low-frequency observations. The PCA analysis is the weakest link. A revision that adds a noise power spectrum, specifies the 21cmFAST parameters, and rephrases the residual interpretation would make the paper solid. I would accept it for review.","headline":"Useful new deep MWA catalogue and noise estimates for a quiet EoR field; the PCA residual claim needs a noise floor before it can be taken seriously.","tokens_in":24290,"tokens_out":4164,"would_cite":true,"duration_ms":45094,"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":"This paper reports that after removing the three strongest PCA foreground components from MWA sub-band images of a candidate Epoch of Reionization field, the residual angular power spectrum remains more than an order of magnitude above…","keywords":["epoch of reionization","21 cm cosmology","foreground subtraction","principal component analysis","angular power spectrum","Murchison Widefield Array","radio source catalogue","SKA pathfinder"],"falsifier":"Run PCA-3 on the same four sub-band images after injecting a known synthetic 21 cm signal into a realistic foreground simulation: if the recovered residual power still sits an order of magnitude above the injected signal, the gap is caused by the subtraction procedure itself; if it tracks the injected signal, the observed excess is due to real unmodelled foregrounds. A simpler check is to repeat the subtraction with one, two, and four PCA components and see whether the residual power drops monotonically toward the 21cmFAST curve or saturates above it.","tokens_in":23232,"feed_emoji":"📡","tokens_out":9399,"duration_ms":93065,"temperature":0.7,"pith_summary":"This paper tests whether a sky field selected as 'quiet' for future Square Kilometre Array Epoch of Reionization imaging can be cleaned of radio foregrounds by Principal Component Analysis. From 4.43 hours of Murchison Widefield Array Phase II observations at 88 and 216 MHz, it builds deep images, detects up to 2,576 radio sources within 5 degrees of the field centre, and removes the three strongest PCA components from the four 7.68 MHz sub-band images. The central result is that the angular power spectrum of the residual maps remains more than an order of magnitude above the theoretical 21 cm CD/EoR signal predicted by 21cmFAST at every sub-band. The paper concludes that standard PCA foreground subtraction, together with this data reduction path, is not yet sufficient for deep imaging of reionization structures in this field.","feed_headline":"Foreground cleanup still hides the 21 cm EoR signal by 10x","feed_subtitle":"Deep MWA imaging of a candidate SKA field shows standard PCA subtraction cannot reach reionization structure.","key_machinery":"The mechanism that carries the argument is a three-component PCA foreground removal applied to four sub-band images, followed by an angular power-spectrum comparison. Principal Component Analysis is a linear projection onto the directions of largest variance across frequency; because foregrounds are spectrally smooth, the first components are expected to contain the foregrounds while the 21 cm signal, which varies rapidly with frequency, is expected to survive. The paper computes the brightness-temperature maps of the four 7.68 MHz sub-bands, subtracts the first three PCA components (denoted PCA-3), and measures the angular power spectrum $C_\\ell$ of the residuals via spherical-harmonic decomposition. The target for comparison is the CD/EoR angular power spectrum extracted from the 21cmFAST light-cone simulation, rescaled to the same sky coverage and pixel size.","core_discovery":"On the paper's own terms, the discovery is that the G0044 field—selected as a low-brightness, low-variance candidate—still leaves a foreground residual that swamps the cosmological signal after aggressive PCA cleaning. After removing the first three principal components, over 98% of the resolved radio sources are captured by the subtracted components, and the residual angular power spectrum $C_\\ell$ drops in all $\\ell$-modes relative to the input temperature maps. Yet at both redshift windows ($z \\approx 13$\\textendash$18$ at 88 MHz and $z \\approx 5$\\textendash$6$ at 216 MHz) the residual power sits more than an order of magnitude above the theoretically predicted 21 cm signal from the 21cmFAST simulation. The paper reads this as evidence that the foregrounds in this field are not fully described by three spectral modes across a 7.68 MHz band, and that further improvements in data reduction and foreground subtraction are required before the field can deliver tomographic EoR images.","pith_inferences":["The three-mode PCA limit may be an artefact of using only four frequency channels: a wider band or finer frequency sampling would allow more foreground modes to be modelled and could close part of the gap.","The local minima seen in the residual spectra (around $\\ell \\approx 178$ at 88 MHz and $\\ell \\approx 116$ at 216 MHz) suggest a characteristic angular scale of residual diffuse foregrounds; if that scale is real, it may mark a window where EoR extraction is less contaminated.","A decisive test of whether the excess is method-inherent would be to apply an independent foreground filter, such as a Gaussian process or a trained denoiser, to the same four sub-band images and check whether the residual power approaches the 21cmFAST prediction.","The field's 'quietness' in source counts does not guarantee spectral cleanliness: source-count quietness and foreground-mode simplicity are different properties, and future field selection may need to rank candidates by spectral mode occupation rather than by source density."],"forward_implications":["The G0044 field cannot yet be used for deep EoR imaging; the residual foreground power exceeds the expected 21 cm signal by more than an order of magnitude after PCA-3 subtraction.","The 216 MHz deep image, with 2,576 detected sources, 90% completeness at 10.4 mJy and an average RMS of 1.80 mJy, provides a foreground source catalogue that can feed future calibration and sky-model construction.","Confusion noise, not thermal noise, will be the limiting noise floor for SKA1-Low deep imaging of fields similar to G0044, given the estimated confusion levels of 1.47 mJy at 88 MHz and 0.17 mJy at 216 MHz for the MWA.","Because nearly all resolved sources are removed by the first three PCA components, the remaining excess power must come from diffuse or unresolved foreground structure, or from signal loss in the projection, rather than from bright point sources."],"supporting_citations":[{"why":"Supplies the field-selection criteria and the candidate 'quiet' field that G0044 covers.","marker":"Zheng et al. (2020)"},{"why":"Provides the data-reduction pipeline used to calibrate, image, and stack the MWA snapshots.","marker":"Duchesne et al. (2020)"},{"why":"Provides the GLEAM catalogue used for the sky model, flux scale, and source-count comparison.","marker":"Hurley-Walker et al. (2017)"},{"why":"Supplies the 21cmFAST light-cone simulation whose angular power spectrum is the theoretical target the residuals must beat.","marker":"Mesinger, Greig & Sobacchi (2016)"},{"why":"Establishes the spectral-smoothness rationale for treating foregrounds as low-rank and removable by PCA.","marker":"de Oliveira-Costa et al. (2008)"},{"why":"Supplies the PCA foreground-removal methodology that the paper applies to the sub-band images.","marker":"Makinen et al. (2021)"},{"why":"Supplies the completeness-simulation and catalogue-validation methodology that the paper follows.","marker":"Lynch et al. (2021)"}],"fun_headline_variants":["MWA deep field: residual foregrounds still 10x above EoR signal","PCA cleanup leaves foregrounds 10x too bright for 21 cm EoR detection","Residual radio foregrounds thwart MWA EoR attempt by 10x","Deep MWA imaging: foregrounds still hide 21 cm signal by 10x","MWA's best field still leaves 10x too much foreground for EoR"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that radio foregrounds occupy no more than three spectral components across the 7.68 MHz sub-band, so subtracting the three largest PCA components removes foregrounds without projecting away the 21 cm signal; with only four frequency channels this premise is not tested in the paper.","fun_headline_variants_meta":{"raw":{"variants":["MWA deep field: residual foregrounds still 10x above EoR signal","PCA cleanup leaves foregrounds 10x too bright for 21 cm EoR detection","Residual radio foregrounds thwart MWA EoR attempt by 10x","Deep MWA imaging: foregrounds still hide 21 cm signal by 10x","MWA's best field still leaves 10x too much foreground for EoR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000486,"raw_usage":{"total_tokens":2464,"prompt_tokens":1083,"completion_tokens":1381,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":699,"completion_tokens_details":{"reasoning_tokens":1270}},"tokens_in":699,"tokens_out":1381,"duration_ms":10771,"temperature":1.0,"reasoning_tokens":1270,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:40:27.013849+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run PCA-3 on the same four sub-band images after injecting a known synthetic 21 cm signal into a realistic foreground simulation: if the recovered residual power still sits an order of magnitude above the injected signal, the gap is caused by the subtraction procedure itself; if it tracks the injected signal, the observed excess is due to real unmodelled foregrounds. A simpler check is to repeat the subtraction with one, two, and four PCA components and see whether the residual power drops monotonically toward the 21cmFAST curve or saturates above it.","supporting_citations":[{"cited_title":"W., Li W., 2020, , 499, 3434","cited_arxiv_id":null,"evidence_quote":"Supplies the field-selection criteria and the candidate 'quiet' field that G0044 covers."},{"cited_title":"W., Johnston-Hollitt M., Zhu Z., Wayth R","cited_arxiv_id":null,"evidence_quote":"Provides the data-reduction pipeline used to calibrate, image, and stack the MWA snapshots."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the GLEAM catalogue used for the sky model, flux scale, and source-count comparison."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the 21cmFAST light-cone simulation whose angular power spectrum is the theoretical target the residuals must beat."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the spectral-smoothness rationale for treating foregrounds as low-rank and removable by PCA."},{"cited_title":"L., Lancaster L., Villaescusa-Navarro F., Melchior P., Ho S., Perreault-Levasseur L., Spergel D","cited_arxiv_id":null,"evidence_quote":"Supplies the PCA foreground-removal methodology that the paper applies to the sub-band images."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the completeness-simulation and catalogue-validation methodology that the paper follows."}],"review_version":1}