{"id":"e6d5b4b9-7c66-46d3-a896-7167001543e0","arxiv_id":"2412.03078","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A simulation-based forecast shows that photometric redshift bin mismatch is a significant systematic for SPHEREx galaxy clustering measurements of f_NL, and that a scattering matrix correction removes the bias when the photo-z error model is known.","lead":"SPHEREx-like galaxy simulations show that photometric redshift errors scatter galaxies between tomographic bins, biasing estimates of the primordial non-Gaussianity parameter f_NL by up to 3 to 6 sigma and galaxy bias by up to 9 to 12 sigma. The authors show that a scattering matrix correction recovers unbiased values when the photo-z error distribution is known.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Closed-loop validation only: the scattering-matrix unbiasedness claim is untested for misspecified photo-z errors, so the central mitigation conclusion overreaches.","rationale":"I read the paper as making two claims: (1) redshift bin mismatch from photo-z errors produces 3–6 sigma f_NL biases and 9–12 sigma bias-parameter shifts in SPHEREx-like tomographic forecasts; (2) a scattering matrix correction removes this systematic and yields unbiased parameter recovery. Claim (1) is well supported by the 500-simulation comparisons: the same pipeline recovers f_NL before photo-z errors (Figure 9), and the observed power spectra shift substantially after scattering (Figure 6). Claim (2) is the load-bearing part for the paper's practical message, and it is validated only under exact knowledge of the photo-z error distribution. The reader's weakest-assumption statement identifies this same gap, and I agree with the conditional verdict: the forecast and method proposal are valuable, but the unbiased-recovery conclusion needs an explicit caveat and a misspecification test before it is used for real SPHEREx analysis.","tokens_in":14633,"tokens_out":7948,"duration_ms":81928,"concrete_test":"Rerun the Case-II mock suite for f_NL^true = 1, but generate photometric redshifts with a deliberately misspecified model: for example, draw from N(z_t, 1.1*sigma0(1+z)) while still constructing P with the fiducial sigma0, and separately add a 1% catastrophic-outlier population with z_p near 1.5. Apply the Section 4.3 scattering-matrix correction and re-estimate f_NL and the 15 bias parameters. If the corrected f_NL shifts by more than about 1 sigma, or if residual bias exceeds the uncorrected Case-I level, the unbiased-recovery result is a closed-loop artifact that requires an explicit photo-z calibration term before adoption.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central mitigation claim — that the scattering matrix formalism 'enables unbiased estimation' of f_NL and galaxy bias (Section 4.3, Figures 3 and 4) — rests entirely on a closed-loop test. The mock photometric redshifts are drawn from a Gaussian N(z_t, sigma0(1+z)) with the same sigma0(z) that is then used in Equation 8 and Equation 14 to construct the scattering matrix P; moreover, P is averaged over the same 500 realizations whose mean power spectra are corrected. In this setting, applying the inverse scattering matrix is algebraically guaranteed to recover the input up to noise, so the agreement with fiducial values is a consistency check rather than an empirical validation. The paper explicitly defers 'catastrophic redshift errors, photometric calibration errors' to future work (Section 5) and nowhere perturbs p(z_p|z_t) away from the truth. In real SPHEREx data, sigma0(z) and the full shape of the photo-z error distribution will be imperfectly known; a fractional error in P propagates through the inverse and can reintroduce bias at a level comparable to the 3–6 sigma offsets the method is designed to remove. Without an open-loop or misspecification test, the statement that the method is unbiased for SPHEREx is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses 500 log-normal GLASS simulations of SPHEREx-like galaxy density fields with Gaussian photometric redshift errors to study how redshift bin mismatch affects tomographic angular-power-spectrum forecasts of local primordial non-Gaussianity. The authors find that the standard convolution-based treatment of photometric redshifts leaves biases on f_NL of about 3-6 sigma and on galaxy linear halo bias of about 9-12 sigma, and they propose a scattering-matrix correction (Eqs. 12-14) that recovers the fiducial f_NL and bias within 1 sigma in their simulations. They also show that broader redshift bins reduce but do not remove the bias, and they recommend the scattering-matrix formalism for future tomographic analyses.","tokens_in":14805,"tokens_out":12442,"duration_ms":129316,"significance":"If the quantitative bias and the mitigation result are robust, this is a timely and important result for SPHEREx PNG science: it identifies a systematic that can masquerade as a several-sigma f_NL signal and offers a practical correction. The paper has clear strengths: the pipeline is validated on simulations without photometric redshift errors (Figure 9), the bias demonstration is carried out on 500 realizations generated with public codes, the covariance is estimated from the simulations, and the bin-width comparison (Figure 8) tests the robustness of the qualitative conclusion. The main caveat is that the correction is validated only under precisely the Gaussian photometric-redshift error model used to generate the simulations, so the 'unbiased estimation' claim is currently a closed-loop consistency check rather than an empirical validation against misspecified photo-z errors.","major_comments":[{"comment":"The central claim that the scattering-matrix formalism 'enables unbiased estimation' of f_NL and galaxy bias is supported only by a closed-loop test: the same Gaussian photometric-redshift error distribution used to scatter galaxies in the simulations is used to construct P and to correct the average spectra. A misspecification test -- for example, perturbing sigma0(z) by 10-20 percent, using a non-Gaussian p(zp|zt), or adding a small catastrophic-outlier population -- is needed before the method can be presented as unbiased for real SPHEREx data. Section 5 explicitly defers catastrophic redshift errors and photometric calibration errors to future work, but the abstract and Section 4.3 state the unbiased result without this qualification.","section":"Section 3.2, Section 4.3, Eq. (14)"},{"comment":"The normalization and indexing of the scattering matrix are not consistent as written. Equation (12) requires P_{xi} to be the probability that a galaxy in true-redshift bin x is observed in photometric bin i, normalized by the true-redshift bin population, but Eq. (14) defines P_{ij} as the fraction of galaxies in photometric bin i whose true redshift lies in bin j, normalized by the photometric-redshift bin population. These differ by the ratio of the photometric and true redshift bin populations; as written, the relation C^{gg,ph} = P^T C^{gg,tr} P does not follow from Eq. (14). Please state the intended index convention and provide the correct normalization for the scattering matrix.","section":"Eqs. (12)-(14)"},{"comment":"The likelihood in Eq. (10) uses the sample covariance K of the individual power-spectrum estimates, while the data vector d_l is the average over 500 realizations. The covariance of the averaged data vector is K/500, so the reported posterior widths and the quoted '3-6 sigma' and '9-12 sigma' significances are not correctly normalized for the quantity being fit. Either divide K by 500 when fitting the mean spectrum, or state explicitly that K is intended as the per-survey covariance for a single SPHEREx-like realization and that the quoted shifts are offsets of the mean signal relative to that single-survey error.","section":"Section 3.3, Eqs. (6) and (10)"}],"minor_comments":[{"comment":"The text says photometric redshifts were generated 'by drawing a positive random value from Gaussian distribution N(zt, sigma0(1+z))'; since a Gaussian draw can be negative, please specify whether negative draws are rejected and re-drawn or whether the distribution is truncated.","section":"Section 3"},{"comment":"The theory vector t_l(theta) is not defined explicitly; please state that it is the angular power spectrum computed from Eqs. (2)-(4), including the scattering-matrix model when that approach is used.","section":"Section 3.3, Eq. (10)"},{"comment":"The caption says the upper and lower panels correspond to Case-I and Case-II, but the figure appears as a 2x3 grid; please label each subplot with the case and the fiducial f_NL value so the panels can be read unambiguously.","section":"Figure 3"},{"comment":"The sentence 'We refer the readers to C24 for a more detailed explanation on the scattering matrix formalism' leaves the key new derivation mostly in a companion paper; since the formalism is the central result, at least the step that applies P to either the data or the theory (inversion versus forward modeling) should be given in this paper.","section":"Section 4.3"},{"comment":"The phrase 'redshift mismatch of galaxies' should be 'redshift bin mismatch of galaxies' for consistency with the body, and 'forecasts on PNG' could be rephrased as 'forecasts for PNG constraints'.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The bias demonstration is solid and within the journal's scope; the main issue is that the mitigation claim rests on a closed-loop validation. If the authors add misspecification tests or appropriately qualify the unbiasedness claim, the paper would be suitable for publication. The covariance normalization in the likelihood should also be corrected or clarified before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper is worth reading for anyone forecasting f_NL with SPHEREx or similar photometric surveys. The genuinely new piece is the quantitative forecast: 500 lognormal simulations, two redshift-accuracy configurations, and clear sigma-level biases from photo-z bin mismatch (3-6 sigma on f_NL, 9-12 sigma on bias). The pipeline validation before adding photo-z errors is well done, and the direct with/without comparison in Figure 6 makes the systematic obvious. The convolution-based scattering matrix is a practical alternative to the earlier deconvolution approach.\n\nThe main soft spot is exactly what the stress-test note says: the scattering-matrix correction is validated only in a closed loop. The mocks are generated from a Gaussian photo-z error distribution with a known sigma0(z), and the scattering matrix is built from that same sigma0(z) and averaged over the same 500 realizations whose power spectra are corrected. Inverting P in that setting is algebraically guaranteed to recover the input up to noise. So the statement that the formalism 'enables unbiased estimation' for SPHEREx overreaches. The paper explicitly defers catastrophic outliers and calibration errors, but doesn't test even a slightly misspecified sigma0 or a non-Gaussian error distribution. A simple open-loop test would strengthen the mitigation claim a lot.\n\nThere are also a few implementation details left vague: how shot noise enters the power spectra and the scattering matrix normalization aren't fully spelled out, and no code is released. Those are fixable in revision rather than fatal.\n\nThe central systematic claim holds up: the bias exists, it's large for SPHEREx's proposed bins, and the paper demonstrates it convincingly. The mitigation claim is promising but not yet established for real data. I'd send this to peer review and ask the authors either to add misspecification tests or to soften the unbiased language to 'unbiased under the exactly known photo-z error model.' For someone planning SPHEREx PNG analyses, this is a relevant and timely forecast.","headline":"SPHEREx-specific forecast of photo-z bin mismatch bias is solid, but the scattering-matrix 'unbiased' claim rests only on a closed-loop test and needs softening.","tokens_in":15401,"tokens_out":2150,"would_cite":true,"duration_ms":21640,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Redshift bin mismatch in SPHEREx clustering can mimic 3-6 sigma non-Gaussianity; a scattering-matrix correction removes the bias.","keywords":["primordial non-Gaussianity","local non-Gaussianity parameter f_NL","photometric redshifts","redshift bin mismatch","tomographic angular power spectrum","SPHEREx","galaxy bias","scattering matrix"],"falsifier":"Run the same 500-realization pipeline with a photometric-redshift error model deliberately different from the one used to scramble the galaxies—for instance, inflating the error width by 20% or adding a small catastrophic-outlier tail—and check whether the scattering-matrix-corrected $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ posterior still contains the fiducial value within $1\\sigma$; if it does not, the unbiased-recovery claim is limited to exactly known redshift errors.","tokens_in":14359,"feed_emoji":"🔭","tokens_out":14030,"duration_ms":112828,"temperature":0.7,"pith_summary":"The paper argues that a specific, easily overlooked systematic—galaxies assigned to the wrong redshift bin because of photometric redshift errors—can masquerade as genuine primordial non-Gaussianity in tomographic clustering measurements from SPHEREx. Using 500 simulated SPHEREx-like galaxy density fields, it shows that this bin mismatch shifts the inferred local non-Gaussianity parameter $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ by 3–6$\\sigma$ and the galaxy linear bias by 9–12$\\sigma$ when all five SPHEREx redshift-accuracy bins are used. The paper then proposes a scattering matrix that tracks the fraction of galaxies migrating between bins and corrects the measured angular power spectra, recovering unbiased estimates of both $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ and bias within $1\\sigma$. If correct, this establishes that redshift bin mismatch must be accounted for before SPHEREx data can be used to discriminate between single- and multi-field inflation models.","feed_headline":"Redshift-bin mismatch can fake up to 6-sigma f_NL signals","feed_subtitle":"A scattering-matrix correction removes the bias, restoring unbiased f_NL and galaxy-bias forecasts.","key_machinery":"The load-bearing object is the scattering matrix $P_{ij}$, the fraction of galaxies that migrate from true redshift bin $i$ to photometric bin $j$ because of photometric-redshift errors. It connects the observed photometric angular power spectrum to the true one through $C_{ij}^{\\mathrm{gg,ph}}(\\ell)=\\sum_{x,y}P_{xi}P_{yj}C_{xy}^{\\mathrm{gg,tr}}(\\ell)$, so the observed spectrum is a quadratic mixture of true spectra from different bins. A simple convolution of redshift distributions cannot capture this mixing, because the mixing weights enter quadratically. The paper computes $P$ by convolving the observed photometric redshift distribution with the photometric-redshift error distribution, avoiding the regularization problems of deconvolution, and shows that using this matrix to correct the measured power spectra removes the parameter biases.","core_discovery":"The central discovery is that photo-z-induced redshift bin mismatch is not a small correction for SPHEREx: it is large enough to create apparent tensions of 1–3$\\sigma$ on $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ and up to 9$\\sigma$ on galaxy bias when only the three high-accuracy redshift bins are used (Case-I), and up to 6$\\sigma$ on $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ and 12$\\sigma$ on bias when all five accuracy bins are used (Case-II). These shifts persist even when the true redshift distribution is estimated directly from the simulated galaxy catalogue, showing they come from the diffusion of galaxies across bin boundaries rather than from a bad estimate of $dN/dz$. The paper further claims that computing the scattering matrix from the observed photometric redshift distribution and the photometric-redshift error distribution, and using it to correct the power spectra, recovers $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ and the full galaxy-bias evolution within $1\\sigma$ for both configurations and for fiducial values $f_{\\mathrm{NL}}^{\\mathrm{loc}}=1,10,100$.","pith_inferences":["The same quadratic-mixing logic should extend to other photometric surveys with comparable redshift scatter, with the bias amplitude set by the off-diagonal weight of each survey's scattering matrix; the paper does not quantify this transfer.","Because the scattering matrix links bins, cross-bin angular power spectra carry additional information about the migration fractions; a joint auto-plus-cross analysis could sharpen the correction or expose misspecification of the photometric-redshift model.","For real SPHEREx data, a practical route is to calibrate the scattering matrix with a spectroscopic subsample and then marginalize over its uncertainty, converting the systematic into a nuisance parameter rather than assuming the error model is exact."],"forward_implications":["If the scattering-matrix correction is applied, SPHEREx can recover unbiased $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ and galaxy-bias estimates from tomographic auto-spectra even when low-accuracy redshift bins are added to increase the galaxy sample.","Without the correction, apparent 3–6$\\sigma$ shifts in $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ could be misread as evidence for multi-field inflation, and 9–12$\\sigma$ shifts in galaxy bias would corrupt any clustering-based cosmology from the same data.","Widening tomographic bins reduces the mismatch bias but sacrifices sensitivity to redshift evolution; the scattering matrix removes the bias regardless of bin width, so narrow bins can be kept.","The failure of convolution and deconvolution estimates of the true redshift distribution to cure the bias implies that future tomographic analyses must forward-model or invert the bin-to-bin scattering rather than only correcting the redshift distribution."],"supporting_citations":[{"why":"Introduces the scattering-matrix formalism for correcting tomographic photo-z bin mismatch that this paper adapts and validates for SPHEREx.","marker":"Shekhar Saraf & Bielewicz 2024"},{"why":"Supplies the relation linking photometric and true angular power spectra through bin-to-bin galaxy scattering that the correction is built on.","marker":"Zhang et al. 2010"},{"why":"Supplies the SPHEREx survey specifications, redshift-accuracy bins, number densities, and galaxy bias evolution used as fiducial simulation inputs.","marker":"Doré et al. 2014"},{"why":"Supplies the simulation code used to generate the lognormal galaxy density fields on which the forecasts are based.","marker":"Tessore et al. 2023"},{"why":"Provides the scale-dependent halo-bias model through which the local non-Gaussianity parameter enters the angular power spectrum.","marker":"Slosar et al. 2008"},{"why":"Provides the companion derivation of scale-dependent bias used to set the fiducial non-Gaussianity signal in the simulations.","marker":"Dalal et al. 2008"},{"why":"Supplies the current best constraint on the local non-Gaussianity parameter, the precision target that makes these systematics consequential.","marker":"Planck Collaboration et al. 2020"}],"fun_headline_variants":["Redshift-bin mix-ups fake 6-sigma f_NL for SPHEREx","Scattering matrix restores SPHEREx f_NL forecasts","Photo-z bin mismatch causes 12-sigma bias shifts","SPHEREx: Redshift errors create false f_NL signals"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The correction is only demonstrated for the case where the photometric-redshift error model used in the analysis exactly matches the true error distribution that scrambled the galaxies, so a real survey with mismodeled or catastrophic redshift outliers is not covered by this validation.","fun_headline_variants_meta":{"raw":{"variants":["Redshift-bin mix-ups fake 6-sigma f_NL for SPHEREx","Scattering matrix restores SPHEREx f_NL forecasts","Photo-z bin mismatch causes 12-sigma bias shifts","SPHEREx: Redshift errors create false f_NL signals"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000325,"raw_usage":{"total_tokens":1868,"prompt_tokens":1037,"completion_tokens":831,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":653,"completion_tokens_details":{"reasoning_tokens":755}},"tokens_in":653,"tokens_out":831,"duration_ms":7538,"temperature":1.0,"reasoning_tokens":755,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T22:48:11.463692+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same 500-realization pipeline with a photometric-redshift error model deliberately different from the one used to scramble the galaxies—for instance, inflating the error width by 20% or adding a small catastrophic-outlier tail—and check whether the scattering-matrix-corrected $f_{\\mathrm{NL}}^{\\mathrm{loc}}$ posterior still contains the fiducial value within $1\\sigma$; if it does not, the unbiased-recovery claim is limited to exactly known redshift errors.","supporting_citations":[{"cited_title":"2024, , 687, A150, 10.1051/0004-6361/202348732","cited_arxiv_id":null,"evidence_quote":"Introduces the scattering-matrix formalism for correcting tomographic photo-z bin mismatch that this paper adapts and validates for SPHEREx."}],"review_version":1}