{"id":"24cf600e-282c-45f7-a540-ee3dc7dc370c","arxiv_id":"2412.10279","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A cross-correlation of DESI luminous red galaxies with Planck CMB lensing yields f_NL = 39 +40/-38 (and 24 +20/-21 combined with the galaxy autocorrelation), consistent with zero at about one sigma.","lead":"Astronomers measured a cosmological parameter, f_NL, that describes whether the early universe's density fluctuations were perfectly random, using a new cross-correlation of galaxies and CMB lensing. The result is consistent with zero, meaning no evidence for non-Gaussian initial conditions, and demonstrates that this technique can be used for future surveys.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Mock validation adds contamination only to LRG maps, never to Planck lensing; a shared foreground that biases both C_l^{kappa G} and C_l^{GG} is therefore untested, and if present it would shift the headline f_NL and the claim that the cross-correlation is systematics-robust.","rationale":"The reader's verdict identifies the correct weakest point. The central empirical claim (f_NL consistent with zero at ~1 sigma) is conditional on the measured cross-correlation being free of systematics shared between DESI LRG and Planck lensing. The mock validation explicitly assumes this away, and the mitigation pipeline cannot remove contamination from the lensing side. This is a genuine load-bearing assumption, not a disagreement with consensus: it is a gap between the validation setup and the data scenario the paper claims to be robust against. The test I propose would settle it by injecting a realistic common foreground into both maps and checking the f_NL shift against the error bar. If the shift is small, the current conclusion stands; if large, the robustness claim requires qualification. I do not see an internal inconsistency strong enough to reject the paper: the pipeline is carefully described, the data products are public, and the mocks include realistic LRG-side contamination. The choice of ell_min = 7 for CGG is data-informed and should be disclosed more prominently, but it does not affect the C_l^{kappa G}-only result, which is the centerpiece of the robustness claim. Hence the correct disposition is to keep the reader's conditional verdict until the correlated-systematics test is performed.","tokens_in":22771,"tokens_out":4385,"duration_ms":37998,"concrete_test":"Construct a new mock set (or reuse the 100 realizations) in which the same Galactic foreground template - e.g., the Planck 857 GHz map smoothed to the lensing resolution, or the extinction map used in SYSnet - is added coherently to both the LRG count map and the CMB lensing convergence map, with amplitude bracketed by the measured dust-LRG and dust-lensing cross-correlations in the real data. Rerun the full pipeline (SYSnet mitigation, pseudo-C_l estimation, covariance, MCMC) on these correlated-contamination mocks for f_NL = 0 and f_NL = 50. If the recovered f_NL shifts by more than about +/-20 for a contamination amplitude within the bracketed range, the assumption of no correlated systematics is falsified and the headline robustness claim should be weakened; if the shift is well below this, the concern is retired.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's robustness claim for C_l^{kappa G} rests on mock tests in Section 5 where, as the text states, 'we did not add any contamination to the CMB lensing mock fields. This was equivalent to assuming there is no correlation in systematics between the two probes.' SYSnet operates only on the LRG overdensity map (Section 4.1), so any systematic that enters the Planck lensing reconstruction and is spatially correlated with the LRG imaging templates (e.g., Galactic dust affecting both the photometric selection and the lensing reconstruction) is not removed, and no validation exercises this channel. The existing checks do not close the gap: varying ell_min (Fig. 12) only changes the scale range, NGC/SGC splits (Table 4) still share the same foreground physics, and the sigma8 and p-parameter tests (Figs 9-10) change the model, not the data systematics. A correlated dust-like component of order a few percent of the lensing signal on the largest scales could shift f_NL by tens, i.e., a substantial fraction of the quoted 1-sigma error, since the f_NL signal itself enters through the lowest multipoles. The central value f_NL = 39+40/-38 would then no longer be a clean statement about primordial non-Gaussianity.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper measures the local primordial non-Gaussianity parameter f_NL through the scale-dependent galaxy bias effect, using the angular cross-correlation between DESI DR9 luminous red galaxies (LRG) and Planck PR4 CMB lensing, and the LRG autocorrelation. The analysis pipeline consists of a SYSnet neural-network imaging systematics mitigation applied to the LRG maps, pseudo-C_ell estimation with NaMaster, a CAMB-based theoretical model including redshift-space distortions and lensing magnification with s=0.999, and MCMC inference over f_NL and a bias parameter (plus shot noise for the autocorrelation). The headline results are f_NL = 39+40/-38 (68% CL) from C_l^{kappa G} alone and f_NL = 24+20/-21 from the joint analysis with C_l^{GG}, with the autocorrelation using an ell_min=7 scale cut. The pipeline is tested on 100 correlated Gaussian mocks, including regressis-based imaging contamination.","tokens_in":23094,"tokens_out":4376,"duration_ms":45244,"significance":"If the constraints are unbiased, the paper provides an important demonstration that CMB lensing cross-correlations are a systematics-resistant avenue for f_NL measurements with photometric samples, with direct relevance to DESI and future surveys. The analysis is careful in many respects: end-to-end mock validation, multiple robustness tests (sigma8, p parameter, bias evolution, ell_min, NGC/SGC splits), and public release of data products are all strengths. However, the validation has a load-bearing gap: contamination is added only to the LRG mocks and not to the CMB lensing mocks, so correlated systematics between the two probes are explicitly assumed away. In addition, the scale cut for C_l^{GG} is motivated by mock tests but in the real data removes the bins showing the largest deviation, and the unblinded analysis leaves open the question of post-hoc selection. These issues do not invalidate the measurement, but they mean the 'robustness' claim currently outruns the evidence it is based on.","major_comments":[{"comment":"The mock validation explicitly assumes no correlated systematics between the two probes: the text states 'we did not add any contamination to the CMB lensing mock fields. This was equivalent to assuming there is no correlation in systematics between the two probes.' This assumption is load-bearing for the central claim that C_l^{kappa G} is robust to imaging systematics. SYSnet acts only on the LRG overdensity map (§4.1), so a systematic that enters both the Planck lensing reconstruction and the LRG selection (e.g., Galactic dust or extinction patterns) would not be removed, and the tests presented here do not exercise that channel. The NGC/SGC split and the ell_min variations do not close the gap because they do not change the shared foreground physics. I would ask for a specific null test, for example cross-correlating the mitigated LRG map with Planck lensing systematics maps or with lensing maps reconstructed from different frequency estimators; at minimum, the paper should state and quantify the residual foreground contamination that is assumed to be negligible for this measurement.","section":"§5, Fig. 7"},{"comment":"The ell_min=7 cut for C_l^{GG} was chosen after inspecting mock tests, but in the real data it removes the lowest-multipole bins that show the largest discrepancy from f_NL=0: the text reports approximately a 2-sigma deviation for ell_min=2 and a 1-sigma deviation for ell_min=7. Because the measurement is explicitly unblinded, it should be documented at which stage this scale cut was frozen, and the full results for ell_min=2 (including the C_l^{GG}-only and joint fits) should be reported as primary or secondary numbers. Without this, the reader cannot distinguish a pre-specified robustness cut from a posteriori tuning to reduce the impact of a statistical fluctuation.","section":"§4.2 and Fig. 12"},{"comment":"The mock validation shows systematic biases that are not negligible compared to the statistical uncertainties. For example, for the f_NL=50 contaminated and mitigated mocks, C_l^{kappa G} alone returns 21 +/- 32, which is approximately 0.9 sigma below the input; for the f_NL=-50 mocks, the joint constraint returns -30 +/- 27, about 1 sigma above the input. The statement that the input f_NL is always within 1 sigma is formally correct but weak for the paper's robustness claim, because a bias comparable to the quoted 1-sigma error is exactly what the measurement is trying to exclude. The paper should either add a systematic-error term to the covariance or the likelihood, apply a calibration based on the mocks, or explicitly soften the conclusion from 'robust' to 'consistent within the current statistical precision.'","section":"§5, Tables 1 and 2"}],"minor_comments":[{"comment":"There are typographical errors: 'louminous red galaxy' and 'line of slight' should be corrected.","section":"Introduction"},{"comment":"The caption reads 'after applying the δ < -30º cut'; this should be 'declination < -30 degrees' for clarity.","section":"Fig. 2 caption"},{"comment":"The statement that the multiplicative Monte Carlo correction is generally less than about 5% but reaches 10-12% on the largest scales is important for the low-ell f_NL signal; it would help to show the correction as a small figure or table, although the current text is acceptable.","section":"§3.2"},{"comment":"The covariance is computed with the analytic Gaussian covariance function using a smoothed version of the measured power spectra as input, and the paper says a mock-based covariance gives compatible results. It would be useful to state whether the smoothed input includes the residual systematics power or the mitigated power, since this choice can affect the error bars by tens of percent.","section":"§4.2"}],"recommendation":"major_revision","confidential_remarks":"This is a careful paper whose central measurement is likely sound, but the systematics-validation section does not fully support the strength of the robustness claims. The missing test of correlated systematics between Planck lensing and the LRG map is the most important issue; it is fixable with additional null tests or by a substantial reframing of the conclusions. I also think the unblinded nature of the analysis combined with the ell_min choice should be transparently addressed, because the paper is otherwise likely to be received as overstating its own robustness. The paper fits the scope of A&A well."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The new thing here is the first f_NL measurement from the DESI LRG and Planck lensing cross-correlation: f_NL = 39 +40/-38 from C_kappaG alone, and 24 +20/-21 when combined with the LRG autocorrelation. Both are consistent with zero, and the paper is careful and honest about what it does and does not control. Credit where due: the pipeline is tested on 100 correlated mocks for three input f_NL values, the contamination model uses regressis weights rather than something trivial, and the robustness tests (sigma8, p parameter, bias evolution, ell_min, NGC/SGC) are sensible. Data and chains are public, which is increasingly the norm but still worth noting.\n\nThe soft spots are real but not fatal. The analysis is unblinded, which the authors disclose. More important, the ell_min = 7 cut for CGG was chosen after seeing mock and data behavior, and it suppresses a ~2 sigma deviation in the autocorrelation; that should be flagged more prominently as a post-hoc choice. The larger concern is the one the stress-test note raises: mock contamination is added only to the LRG maps, never to the Planck lensing maps. The paper says this is equivalent to assuming no correlated systematics between the two probes. If dust or extinction affects both the lensing reconstruction and the LRG selection on large scales, SYSnet cannot remove it, and the cross-correlation inherits a bias that the validation never exercises. That does not invalidate the headline result, which is already within 1-2 sigma of zero, but it does weaken the paper's claim that C_kappaG is robustly more stable than C_GG against imaging systematics. The claim may be true, but the current tests do not fully support it.\n\nI would send this to a serious referee. It is a competent, useful measurement with a new data combination and public products, and the systematics discussion is worth airing. The referee should push on the correlated-systematics assumption and on the scale-cut selection, but I do not see a load-bearing flaw in the central f_NL = 0 result. For a reader working on f_NL from LSS or on CMB lensing cross-correlations, this is a solid data point; I would cite it. I would also bring it to a reading group, mostly to argue about the lensing-systematics channel.","headline":"A careful, honest fNL constraint from the DESI LRG x Planck lensing cross-correlation that lands at zero within ~1 sigma; the robustness claim has a real gap around correlated lensing-systematics, but the paper deserves review.","tokens_in":23951,"tokens_out":1413,"would_cite":true,"duration_ms":606157,"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":"Cross-correlating DESI luminous red galaxies with Planck CMB lensing keeps the local primordial non-Gaussianity parameter consistent with zero, and is more stable against imaging systematics than the galaxy autocorrelation.","keywords":["primordial non-Gaussianity","f_NL","scale-dependent galaxy bias","CMB lensing cross-correlation","DESI","luminous red galaxies","imaging systematics","angular power spectrum"],"falsifier":"Contaminate the mock CMB lensing maps with a template correlated with the LRG imaging systematics, for example dust extinction, and rerun the pipeline: if the recovered $f_{\\rm NL}$ shifts by more than the 68% error bar, the claimed stability of $C_\\ell^{\\kappa G}$ fails. A simpler observational check is to cross-correlate the Planck lensing map with the dust template itself at the same multipoles; a significant signal would mean the cross-correlation is susceptible to the unmodeled correlation.","tokens_in":22578,"feed_emoji":"🌌","tokens_out":13858,"duration_ms":98176,"temperature":0.7,"pith_summary":"The paper sets out to measure the local primordial non-Gaussianity parameter $f_{\\rm NL}$ from the angular cross-correlation of about nine million DESI luminous red galaxies with Planck CMB lensing, exploiting the scale-dependent galaxy bias that a nonzero $f_{\\rm NL}$ would imprint on the largest scales. Using the cross-correlation alone it finds $f_{\\rm NL} = 39^{+40}_{-38}$ at 68% confidence, and combining with the LRG autocorrelation (with a scale cut to limit systematics) gives $f_{\\rm NL} = 24^{+20}_{-21}$. Both results are consistent with zero primordial non-Gaussianity, meaning the data do not require non-Gaussian initial conditions. The paper's broader methodological claim is that the cross-correlation is less sensitive to imaging systematics than the autocorrelation, and that with proper neural-network mitigation this channel can help future surveys reach competitive $f_{\\rm NL}$ constraints. A reader should care because a robust $f_{\\rm NL}$ measurement is a direct test of the simplest inflationary models, which predict Gaussian initial conditions.","feed_headline":"DESI–Planck cross-correlation keeps f_NL consistent with zero","feed_subtitle":"The cross-correlation resists imaging systematics better than the autocorrelation, pointing to tighter future PNG bounds","key_machinery":"The load-bearing mechanism is the scale-dependent galaxy bias induced by local primordial non-Gaussianity: parametrize the primordial potential as $\\Phi = \\phi + f_{\\rm NL}(\\phi^2 - \\langle\\phi^2\\rangle)$; a nonzero $f_{\\rm NL}$ adds a contribution $\\Delta b(k,z) = 2(b_g - p) f_{\\rm NL} \\delta_{\\rm crit} / \\alpha(k)$ to the bias, which grows as $1/k^2$ and is therefore largest at the low multipoles where the cross-correlation is measured. The pipeline estimates $C_\\ell^{\\kappa G}$ and $C_\\ell^{GG}$ from masked maps using pseudo-$C_\\ell$ (NaMaster), models them with CAMB including lensing magnification and redshift-space distortions, and infers $f_{\\rm NL}$ and the linear bias $b_0$ via an MCMC likelihood. A neural-network systematics mitigator (SYSnet) removes imaging contaminants from the LRG map before computing the spectra, and mock tests justify cutting the first multipole bin of $C_\\ell^{GG}$ while keeping all multipoles of $C_\\ell^{\\kappa G}$.","core_discovery":"The central empirical claim is that the local primordial non-Gaussianity parameter is consistent with zero: $f_{\\rm NL} = 39^{+40}_{-38}$ at 68% confidence from the LRG--CMB lensing cross-correlation $C_\\ell^{\\kappa G}$ alone, and $f_{\\rm NL} = 24^{+20}_{-21}$ when combined with the LRG autocorrelation $C_\\ell^{GG}$, with both within about one $\\sigma$ of zero. The supporting methodological claim is that $C_\\ell^{\\kappa G}$ is more stable against imaging systematics than $C_\\ell^{GG}$, because the cross-correlation is less affected by the excess large-scale power that contaminates the density maps; the paper demonstrates this stability by running the full pipeline on contaminated mocks and by showing that the $f_{\\rm NL}$ constraint is robust to the choice of $\\sigma_8$, bias evolution, and the PNG response parameter $p$. The analysis also reports a model-independent constraint on the product $b_\\phi f_{\\rm NL} = 146^{+154}_{-142}$, which remains compatible with Gaussian initial conditions.","pith_inferences":["Because the mock validation contaminated only the LRG maps, the pipeline never tests for systematics correlated between Planck lensing and the LRG density; a mock test that adds the same foreground template to both maps would directly probe this blind spot.","The persistently positive best-fit values, such as $+39$ from $C_\\ell^{\\kappa G}$, could be a residual of the slight overfitting seen in mock tests, and might shift toward zero with a better-calibrated mitigation.","Since the observable is really the product $b_\\phi f_{\\rm NL}$, a simulation-calibrated $b_\\phi$ beyond the universality assumption would convert the reported $b_\\phi f_{\\rm NL} = 146^{+154}_{-142}$ directly into a cleaner $f_{\\rm NL}$ bound.","With future CMB lensing maps of lower noise and DESI's full spectroscopic sample, the same cross-correlation strategy could approach the forecast $\\sigma(f_{\\rm NL})\\sim 5$, making it a competitive probe of inflaton dynamics."],"forward_implications":["Future $f_{\\rm NL}$ measurements can exploit CMB lensing cross-correlations as a systematics-resistant observable, either alone or combined with galaxy autocorrelations.","Combining $C_\\ell^{\\kappa G}$ with $C_\\ell^{GG}$ tightens the uncertainty from $\\sigma(f_{\\rm NL})\\sim 40$ to $\\sim 20$, even with a conservative scale cut on the autocorrelation.","The lowest multipole bin of the LRG autocorrelation ($\\ell = 2$ to $6$) remains contaminated after mitigation, so scale cuts there are necessary for unbiased joint constraints.","The $f_{\\rm NL}$ estimate is stable under changes in $\\sigma_8$, the galaxy bias evolution, and the assumed response parameter $p$, supporting the robustness of the cross-correlation channel."],"supporting_citations":[{"why":"Introduces the scale-dependent bias effect that translates f_NL into a large-scale galaxy clustering signal.","marker":"Dalal et al. (2008)"},{"why":"Defines the local f_NL parametrization of the primordial potential used in the model.","marker":"Komatsu & Spergel (2001)"},{"why":"Performed the previous f_NL analysis of this LRG sample from autocorrelation and supplies the SYSnet mitigation recipes.","marker":"Rezaie et al. (2024)"},{"why":"Prior DESI quasar-CMB lensing cross-correlation measurement; provides the multiplicative lensing normalization correction applied here.","marker":"Krolewski et al. (2024)"},{"why":"Provides the Planck PR4 CMB lensing maps used as the second probe.","marker":"Carron et al. (2022)"},{"why":"Supplies the NaMaster code for pseudo-C_l power spectra and Gaussian covariances.","marker":"Alonso et al. (2019)"},{"why":"Determines the fiducial galaxy bias evolution b_g = b0 D(z)^{-1} assumed in the analysis.","marker":"Zhou et al. (2021)"},{"why":"Gives the alternate PNG response parameter p = 0.55 used in the robustness test.","marker":"Barreira (2020)"}],"fun_headline_variants":["DESI–Planck lensing keeps primordial non-Gaussianity at zero","No primordial non-Gaussianity from DESI–Planck cross-correlation","Cross-correlation tames systematics for tighter f_NL limits","DESI–Planck cross-correlation gives f_NL consistent with zero","f_NL stays zero when DESI galaxies meet Planck lensing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes there are no systematics correlated between the Planck CMB lensing map and the DESI LRG map; the mock validation adds contamination only to the LRG maps and leaves the lensing maps clean, so any real-world correlation between the two probes would bias the cross-correlation without being tested.","fun_headline_variants_meta":{"raw":{"variants":["DESI–Planck lensing keeps primordial non-Gaussianity at zero","No primordial non-Gaussianity from DESI–Planck cross-correlation","Cross-correlation tames systematics for tighter f_NL limits","DESI–Planck cross-correlation gives f_NL consistent with zero","f_NL stays zero when DESI galaxies meet Planck lensing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00029,"raw_usage":{"total_tokens":1796,"prompt_tokens":1145,"completion_tokens":651,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":761,"completion_tokens_details":{"reasoning_tokens":554}},"tokens_in":761,"tokens_out":651,"duration_ms":6025,"temperature":1.0,"reasoning_tokens":554,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T15:59:48.650921+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Contaminate the mock CMB lensing maps with a template correlated with the LRG imaging systematics, for example dust extinction, and rerun the pipeline: if the recovered $f_{\\rm NL}$ shifts by more than the 68% error bar, the claimed stability of $C_\\ell^{\\kappa G}$ fails. A simpler observational check is to cross-correlate the Planck lensing map with the dust template itself at the same multipoles; a significant signal would mean the cross-correlation is susceptible to the unmodeled correlation.","supporting_citations":[{"cited_title":"& Spergel, D","cited_arxiv_id":null,"evidence_quote":"Defines the local f_NL parametrization of the primordial potential used in the model."},{"cited_title":"J., Seo, H.-J., et al","cited_arxiv_id":null,"evidence_quote":"Performed the previous f_NL analysis of this LRG sample from autocorrelation and supplies the SYSnet mitigation recipes."},{"cited_title":"J., Ferraro, S., et al","cited_arxiv_id":null,"evidence_quote":"Prior DESI quasar-CMB lensing cross-correlation measurement; provides the multiplicative lensing normalization correction applied here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Determines the fiducial galaxy bias evolution b_g = b0 D(z)^{-1} assumed in the analysis."},{"cited_title":"2020, Journal of Cosmology and Astroparticle Physics, 2020, 031–031","cited_arxiv_id":null,"evidence_quote":"Gives the alternate PNG response parameter p = 0.55 used in the robustness test."}],"review_version":1}