Assessing the large-scale angular clustering of UNIONS Lyman Break Galaxies via cross-correlations

Alan W. McConnachie; Calum Murray; Christophe Y\`eche; Constantin Payerne; Hendrik Hildebrandt; Kenneth C. Chambers; Martin Kilbinger; Scott Chapman; Thomas de Boer; William d'Assignies Doumerg

UNIONS Lyman-break galaxies work for cosmology through cross-correlations, not auto-clustering, at z around 2.5.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 14:24 UTC pith:XH23DCJO

load-bearing objection Solid first large-area UNIONS LBG cross-correlation result: auto fails, cross works, systematics mostly add variance not bias. the 2 major comments →

arxiv 2607.09943 v2 pith:XH23DCJO submitted 2026-07-10 astro-ph.CO

Assessing the large-scale angular clustering of UNIONS Lyman Break Galaxies via cross-correlations

Constantin Payerne , Christophe Y\`eche , William d'Assignies Doumerg , Hendrik Hildebrandt , Martin Kilbinger , Calum Murray , Thomas de Boer , Kenneth C. Chambers

show 2 more authors

Scott Chapman Alan W. McConnachie

This is my paper

classification astro-ph.CO

keywords Lyman-break galaxiesangular power spectrumCMB lensingcross-correlationimaging systematicsUNIONShigh-redshift clusteringprimordial non-Gaussianity

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper asks whether Lyman-break galaxies selected from UNIONS multi-band photometry can serve as cosmological tracers of large-scale structure above redshift 2. Direct angular clustering of the galaxies themselves is ruined on large scales by survey imaging variations (depth, seeing, dust, stars), even after linear and non-linear corrections. Cross-correlating the same galaxies with independent maps that do not share those systematics—Planck CMB lensing and two quasar samples—recovers a signal whose amplitude matches simple theory. Residual imaging problems mainly inflate the error bars rather than bias the mean signal. That result turns a faint, systematics-limited photometric sample into a usable high-redshift tracer for future growth and non-Gaussianity studies.

Core claim

Spatially varying imaging systematics prevent a usable LBG auto-angular power spectrum on large scales, but the LBG–CMB lensing cross-power spectrum (and LBG–quasar cross-spectra) can be measured more robustly, with amplitude consistent with theoretical predictions; residual systematics appear mainly as excess variance without significant bias in the recovered signal, establishing UNIONS LBGs as reliable tracers for cross-correlation cosmology at z ~ 2.5.

What carries the argument

Cross-power spectra between the UNIONS LBG overdensity and external tracers (Planck PR4 CMB lensing, DESI DR1 quasars, Quaia), after linear de-projection or Random-Forest weighting of the LBG field; the cross-spectrum is largely insensitive to imaging systematics that dominate the auto-spectrum.

Load-bearing premise

The redshift distribution measured on a tiny overlapping field, together with a simple rescaled bias model, is assumed good enough to judge that the full-sky cross-spectrum amplitude matches theory.

What would settle it

A spectroscopic or clustering-redshift n(z) for the full UNIONS LBG sample that shifts the expected LBG–CMB lensing cross-power amplitude well outside the measured error bars would falsify the claimed consistency.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Solid first large-area UNIONS LBG cross-correlation result: auto fails, cross works, systematics mostly add variance not bias.

read the letter

This is a careful feasibility paper that does what it claims. The new result is the first wide-area angular cross-power measurement of UNIONS u-dropout LBGs with Planck PR4 lensing and with DESI/Quaia quasars, plus a clean demonstration that residual imaging systematics (depth, seeing, extinction, stars) wreck the auto-spectrum even after namaster linear and regressis RF corrections, while the cross-spectrum stays usable and consistent with a simple fiducial model.

What they do well is the mock campaign. Controlled linear and non-linear contamination (Appendix C), then data-driven RF weights applied to mocks (Fig. 5), show that the cross recovers the input amplitude; the main damage is excess variance, especially at low ell. The Jackknife-versus-Gaussian comparison (Fig. 7, Table 1) makes the same point on the real data and quantifies how badly that variance will hurt f_NL. Footprint cleaning is partly by eye and the first multipole bin is retained with a warning, but neither is load-bearing. The math is standard NaMaster pseudo-C_ell, the citations are appropriate, and the circularity burden is low: the theory curves use previously published n(z) and bias models that are independent of this measurement.

The soft spot the reader flagged is real but secondary. The XMM-LSS photometric n(z) plus a free b0 rescaling of Wilson & White is only a few square degrees; if the full-footprint redshift kernel or bias is substantially different, the absolute amplitude comparison loses force. That does not undercut the differential claim that systematics primarily inflate variance rather than bias the cross. No public weight maps or catalogs yet, which is typical for this stage but limits immediate reuse.

This is for people building high-z photometric samples or planning Stage-IV cross-correlation pipelines. It is not a cosmology constraint paper; it is a methods-and-validation paper that shows the sample is usable for cross-correlations at z~2.5. I would send it to peer review. The central robustness claim holds under the tests they actually ran.

Referee Report

2 major / 5 minor

Summary. The paper evaluates UNIONS u-dropout LBGs (r~23.5–24.2) as high-z (z~2.5) large-scale structure tracers. It shows that spatially varying imaging systematics (PSF depth, seeing, extinction, stellar density) severely contaminate the LBG auto-angular power spectrum on large scales (ℓ≲300), even after linear namaster deprojection or non-linear regressis Random Forest weighting, leaving residual power far above fiducial models. Cross-correlations with Planck PR4 CMB lensing and with DESI DR1/Quaia quasars are far more robust: mocks with controlled linear/non-linear contamination and data-driven weights recover the input C_ℓ^κg without significant bias, while residual systematics mainly inflate variance (Jackknife/Gaussian ratio ~3–15). The measured LBG–κ and LBG–QSO spectra have amplitudes consistent with a fiducial model (XMM-LSS n(z), Wilson & White bias rescaled by free b0≈0.65, magnification+RSD), establishing the sample for cross-correlation cosmology and illustrating the excess-variance impact on f_loc_NL constraints.

Significance. If the robustness claim holds, the work provides a practical path for Stage-IV photometric LBG samples (UNIONS, LSST, CSST) to contribute to high-z growth and local PNG measurements via cross-correlations, where auto-spectra are compromised by depth/seeing variations near survey limits. Strengths include explicit controlled contamination–decontamination tests (Appendix C, linear α=1 and non-linear α=3 models plus data-driven RF weights applied to mocks), direct Jackknife-versus-Gaussian covariance comparison (Fig. 7, Table 1), and transparent demonstration that systematics primarily add variance rather than bias the cross-signal. These elements make the feasibility result falsifiable and useful for survey strategy and analysis pipelines.

major comments (2)

§5.1 and Fig. 6 (left): The claim that the measured C_ℓ^κg amplitude is “consistent with theoretical predictions” rests on the XMM-LSS photometric n(z) (~2 deg²) plus the Wilson & White (2019) bias prescription rescaled by a free b0≈0.65 taken from prior clustering-redshift work. Because the u-dropout kernel is broad and the full-footprint selection is magnitude-limited near survey depth, spatial variations in completeness could shift the effective n(z) or bias. The absolute-amplitude comparison is therefore only weakly diagnostic; a full-footprint clustering-redshift validation (or at least a quantitative assessment of n(z) variation) is needed before the consistency statement can be regarded as load-bearing support for sample reliability.
§4.1: Footprint cleaning removes Dec>75° and the Ra∈[150,180], Dec∈[20,45] region “by hand” / “by eye” after inspecting PSF/seeing maps. While motivated by visible patterns, the procedure is not fully reproducible and changes the effective area from ~2800 to ~2400 deg². The impact of these cuts on both the recovered cross-spectrum amplitude and the Jackknife covariance should be quantified (e.g., by repeating the analysis with and without the manual masks) so that residual systematics can be cleanly separated from selection effects.

minor comments (5)

Appendix A title and text contain the typo “agular” (should be “angular”).
Fig. 4 caption and surrounding text note that neither angular mode removal nor deprojection noise bias has been applied; a short quantitative statement of their expected size relative to the residual power would help the reader judge the residual contamination.
§5.1 retains the first multipole bin (0<ℓ<30) “for completeness” while correctly warning that the mode-coupling matrix is poorly conditioned. Given that f_loc_NL sensitivity is largest on these scales, either drop the bin from all fits or show explicitly that its inclusion/exclusion does not change the Table 1 posteriors beyond the already-large Jackknife errors.
Several figure panels (e.g., Fig. 3, Fig. 17) would benefit from a common vertical scale or an explicit statement of the mean density used for the overdensity, so that the size of the residual trends after correction can be compared directly.
The manuscript header carries a future date (“Version July 14, 2026”); this should be corrected for the published version.

Circularity Check

1 steps flagged

Mild self-citation of prior n(z)/bias models for amplitude comparison; central robustness claim (cross recovers input, systematics add variance not bias) is independently supported by mocks and data.

specific steps

self citation load bearing [Section 5.1 (theoretical curves overplotted on Fig. 6 left; also Table 1 and Appendix A.1)]
"the LBG linear bias b(z) modeled following Wilson & White (2019), with an overall rescaling b′(z)=b0b(z), where b0=0.65 was found to provide a good description of the LBG bias inferred from clustering measurements for the same LBG selection (see Payerne et al. 2025a)"

The claim of amplitude consistency with “theoretical predictions” relies on a bias amplitude fixed from the authors’ own prior work on the identical LBG selection. While that prior measurement used independent data (clustering redshifts), the present paper’s absolute comparison is therefore partially self-referential; the free refit of b0 in the f_NL analysis recovers a compatible value, confirming the mild circularity is not forced but still present for the consistency statement.

full rationale

The paper's derivation chain for its strongest claim—that LBG–CMB lensing (and LBG–QSO) cross-spectra are robust to imaging systematics while the auto-spectrum is not—is self-contained and non-circular. Controlled contamination mocks (linear/non-linear analytic models in Appendix C; data-driven RF weights from regressis applied to mocks in Fig. 5) demonstrate recovery of the input C_ℓ^κg without bias, with residuals appearing only as excess variance. Jackknife vs Gaussian covariance comparisons and the observed data points in Fig. 6 further support this differentially. The absolute amplitude comparison to a “fiducial theoretical prediction” does invoke n(z) measured on XMM-LSS plus a Wilson & White bias prescription rescaled by b0 = 0.65 taken from the authors’ prior clustering-redshift analysis (Payerne et al. 2025a). This is a minor self-citation, but it is not load-bearing: the same paper freely refits b0 (Table 1) and obtains a compatible value, the mocks themselves use an independent fiducial, and the no-bias conclusion does not require the absolute normalization. No equation forces a measured quantity to equal a fitted input by construction, no uniqueness theorem is imported, and no ansatz is smuggled. Score 2 reflects only the non-central self-citation; the core result stands independently.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

The central claim rests on standard projected clustering formalism, publicly available external maps, and two free amplitude parameters (overall bias rescaling and f_NL) that are fitted rather than predicted. No new physical entities are introduced; the main modeling choices are conventional domain assumptions about linear bias, magnification, and the form of imaging contamination.

free parameters (2)

b0 (overall bias rescaling of Wilson & White 2019 prescription) = 0.6–0.8
Fitted to the cross-spectrum amplitude; best-fit values ~0.6–0.8 appear in Table 1 and are used to claim consistency with theory.
f_loc_NL = order ±350 (JK)
Fitted in the simplified PNG exercise of Section 5.2; the large Jackknife errors are themselves a result of residual systematics.

axioms (3)

domain assumption Linear bias plus scale-dependent PNG correction of the form Δb ∝ f_NL / k² is an adequate description of LBG clustering on the scales ℓ < 300.
Invoked throughout Sections 2 and 5 and Appendix A; standard in the literature but not re-derived here.
domain assumption Imaging systematics that affect the LBG density map are uncorrelated with the Planck CMB lensing map and with the DESI/Quaia quasar density maps.
Core justification for preferring cross-spectra (Section 4 and 5); supported by mock tests but remains an assumption about the external datasets.
domain assumption The photometric redshift distribution measured on the ~2 deg² XMM-LSS field is representative of the full ~2400 deg² cleaned UNIONS footprint.
Used to generate the theoretical C_ℓ curves in Figure 6; validated only partially by earlier clustering-redshift work on a subset of the footprint.

pith-pipeline@v1.1.0-grok45 · 26613 in / 2564 out tokens · 29974 ms · 2026-07-14T14:24:29.918293+00:00 · methodology

0 comments

read the original abstract

Lyman-break galaxies (LBGs), selected via the strong spectral break blueward of the Lyman limit, are powerful tracers of large-scale structure at redshifts $z>2$. In this work, we assess the feasibility of using LBGs selected from the Ultraviolet Near Infrared Optical Northern Survey (UNIONS) multi-band photometric catalog as cosmological probes of the high-redshift Universe using two-point statistics. We demonstrate that spatially varying imaging systematics, driven by variations in PSF depth, seeing across the UNIONS footprint, limit robust measurements of the LBG auto-angular power spectrum on large scales, even after correcting the LBG field with linear or non-linear mitigation techniques. This study shows that clustering analyses of faint galaxy samples close to survey depth are challenging. We therefore turn to cross-correlation measurements with external tracers, in particular the \textit{Planck} CMB lensing convergence and quasars from DESI DR1 and \textit{Quaia}, which are less sensitive to the angular imaging systematics. Using both data and mock catalogues, we demonstrate that the LBG--CMB lensing cross-power spectrum can be measured more robustly than the auto-spectrum, with an amplitude consistent with theoretical predictions. Residual systematics primarily manifest as excess variance at large angular scales, without introducing a significant bias in the recovered signal. Taken together, these results establish UNIONS-selected LBGs as reliable tracers for cross-correlation cosmology at $z\sim 2.5$, and highlight cross-correlation techniques as a powerful and robust avenue for extracting cosmological information from photometric high-redshift galaxy samples in the presence of complex imaging systematics.

Figures

Figures reproduced from arXiv: 2607.09943 by Alan W. McConnachie, Calum Murray, Christophe Y\`eche, Constantin Payerne, Hendrik Hildebrandt, Kenneth C. Chambers, Martin Kilbinger, Scott Chapman, Thomas de Boer, William d'Assignies Doumerg.

Assessing the large-scale angular clustering of UNIONS Lyman Break Galaxies via cross-correlations

Core claim

What carries the argument

Load-bearing premise

What would settle it

discussion (0)