pith. sign in

arxiv: 2606.04089 · v1 · pith:BH3XP7HSnew · submitted 2026-06-02 · 🌌 astro-ph.CO · astro-ph.IM

A multi-eigenbasis approach to covariance matrix denoising for cosmological inference

Wynne Turner This is my paper

Pith reviewed 2026-06-28 08:20 UTC · model grok-4.3

classification 🌌 astro-ph.CO astro-ph.IM
keywords covariance matrix estimationLyα forestcosmological inferenceeigenbasis projectionmock simulationsDESIcorrelation matrix denoising
0
0 comments X

The pith

A multi-eigenbasis method projects noisy covariance matrices onto mock-derived bases and corrects residuals to recover accurate estimates for Lyα forest cosmological inference.

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

The paper addresses the problem of estimating invertible covariance matrices when the number of independent samples falls short of the data vector dimension, which prevents reliable parameter inference in surveys like the 3D Lyα forest. It proposes first projecting the noisy measured covariance onto the eigenbasis obtained from a mock reference to produce an initial denoised matrix. A second step constructs a weighted correction by projecting the remaining residual onto an eigenbasis learned from a mock-trained classifier. Validation on held-out mock covariances shows gains over the existing off-diagonal smoothing procedure both in matrix reconstruction quality and in how closely recovered cosmological parameter posteriors match those from the true covariance.

Core claim

The measured noisy covariance is first projected onto the eigenbasis of a mock-based reference to yield an initial denoised estimate; a weighted residual correction is then formed by projecting the noisy residual onto a second eigenbasis derived from a mock-trained classifier, and this combined reconstruction improves matrix-level metrics and cosmological parameter posterior recovery relative to smoothing when tested on withheld mock covariances.

What carries the argument

The multi-eigenbasis denoising procedure that performs an initial projection of the noisy covariance onto a mock-reference eigenbasis followed by a residual correction onto a classifier-derived eigenbasis.

If this is right

  • The resulting covariance matrix becomes invertible and yields well-calibrated uncertainties in likelihood analyses.
  • Cosmological parameter constraints from the denoised matrix align more closely with those from the true covariance than constraints from the smoothed matrix.
  • The method reduces bias in recovered off-diagonal correlations compared with simple smoothing for the same data vector size.
  • It enables joint use of auto- and cross-correlation measurements in the 3D Lyα forest without requiring an impractically large number of realizations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The two-stage basis approach could be tested on other large-scale structure analyses that face similar sample-size limits.
  • If the classifier eigenbasis encodes features shared across different mock suites, the method might transfer to real data with minimal retraining.
  • Combining more than two reference eigenbases might further reduce residual noise in even higher-dimensional covariance matrices.

Load-bearing premise

The eigenbases extracted from the mock simulations accurately capture the true correlation structure that exists in the real observational data.

What would settle it

On held-out mocks, the parameter posteriors obtained with the denoised covariance deviate from those obtained with the covariance measured from many independent realizations by more than the reported improvement margin.

Figures

Figures reproduced from arXiv: 2606.04089 by Wynne Turner.

Figure 1
Figure 1. Figure 1: — Several correlation and residual matrices for visualization purposes. The correlation matrix on the left is computed from a stack of ∼ 100,000 correlation function measurements from Saclay mocks and is taken as the “truth” for this suite. Top row, left to right: the correlation matrix measured from ∼ 1000 sub-samples of a single mock (representative of a survey measurement); the smoothed version produced… view at source ↗
Figure 2
Figure 2. Figure 2: — Pseudo-eigenvalue spectra and classifier feature weights. Top: pseudo-eigenvalues λ˜ k as a function of eigenmode index k. Solid lines show the median and shaded regions show the 16th–84th and 2.5th–97.5th percentile intervals across the classifier training set for the LyαCoLoRe (blue) and Saclay (orange) mocks. The pseudo-eigenvalues of the DESI DR1 Lyα forest covariance (gray dashed line) are consisten… view at source ↗
Figure 3
Figure 3. Figure 3: — The Frobenius norm (Equation 16, lower axis) and MSE (Equation 17, upper axis) evaluated on the mock testing set at each pipeline stage relative to the truth: the measured noisy correlation matrix (gray), the initial denoised estimate (purple), and the final denoised estimate (blue). The smoothed matrix from the current approach (orange) is also shown for comparison. Points show the median and error bars… view at source ↗
Figure 4
Figure 4. Figure 4: — [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: — Normalized parameter bias distributions (µmethod − µtruth)/σtruth for a selection of parameters: αp (isotropic BAO scale parameter), ϕf (full-shape AP scale parameter), αs (isotropic broadband scale parameter), and fσ8. Distributions are computed from maximum likelihood fits on the mock testing set, where σtruth is the uncertainty inferred from the fit using the true covariance and µ is the mean. A value… view at source ↗
Figure 6
Figure 6. Figure 6: — Uncertainty ratio distributions σmethod/σtruth for the same parameters and covariance methods as in [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: — Posteriors for a full-shape analysis on a Saclay mock in the testing set sampled with PolyChord. I compare results using the true covariance measured from a stack of 100 Saclay mocks with ∼ 1000 sub-samples per mock (black contours), the smoothed covariance used in the current approach (orange dashed contours), and the final denoised covariance from this work (blue filled contours). Cross-hairs indicate … view at source ↗
Figure 8
Figure 8. Figure 8: — Posteriors for the DESI DR1 Lyα full-shape analysis sampled with PolyChord using the standard smoothed covariance (orange dashed contours) and the final denoised covariance from this work (blue filled contours). The two results are in excellent agreement for quantities derived from the BAO and AP scale parameters. A ∼ 0.5σ shift in fσ8 is present, but this parameter was not demonstrated to be unbiased in… view at source ↗
Figure 9
Figure 9. Figure 9: — True correlation matrices for each mock suite projected onto the default reference eigenbasis (Eq. 4). Left: the projection of the Saclay correlation matrix computed from 100 mocks and ∼ 1000 sub-samples each. Right: the projection of the LyαCoLoRe correlation matrix computed from 200 mocks and ∼ 1000 sub-samples each. In both cases the diagonal-to-total power ratio (Eq. A1) P ≈ 0.96, demonstrating that … view at source ↗
Figure 10
Figure 10. Figure 10: — Distributions of the normalized KL divergence DKL/n (Eq. 18) for the final denoised covariances on the mock testing set, using the default mixed initial eigenbasis (blue filled) and the LyαCoLoRe-only eigenbasis (black unfilled). The two distributions are in close agreement, demonstrating that the results are robust to the choice of initial reference matrix [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: — Gaussian posteriors of cosmological parameters from a full-shape fit on a Saclay mock in the classifier testing set. The results compare the default mixed initial eigenbasis (gray contours; also shown in [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
read the original abstract

Accurate covariance matrix estimation is crucial to cosmological analyses, enabling unbiased parameter inference with well-calibrated uncertainties. Obtaining a reliable estimate generally requires far more independent samples than the dimension of the data vector, which is not always feasible. This challenge is especially relevant for the 3D Ly$\alpha$ forest analysis, which measures the Ly$\alpha$ auto-correlation and its cross-correlation with quasars in bins of comoving separation to jointly constrain cosmological parameters. The consequence is a very large data vector, and the data-driven covariance measured from sub-samples is non-invertible. The current approach applies a smoothing procedure to the off-diagonals of the correlation matrix to establish invertibility, but this does not fully capture the true correlation structure. In this work, I present a novel multi-eigenbasis denoising method for the data-driven covariance matrix, developed in the context of the 3D Ly$\alpha$ forest analysis and conditioned on DESI DR1 mock simulations. The measured noisy covariance is first projected onto the eigenbasis of a mock-based reference, yielding an initial denoised estimate. A weighted residual correction is then constructed by projecting the noisy residual onto a second eigenbasis derived from a mock-trained classifier, capturing correlation structure not recovered in the initial reconstruction. I validate the method on mock covariances withheld from classifier training and find significant improvements over the current smoothing-based approach in matrix-level reconstruction metrics and in the recovery of cosmological parameter posteriors when compared to those obtained from the true covariance measured from many mock realizations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a multi-eigenbasis denoising method for covariance matrices in 3D Lyα forest analyses. The noisy data-driven covariance is first projected onto the eigenbasis of a mock-derived reference covariance; a weighted residual is then projected onto a second eigenbasis obtained from a mock-trained classifier to capture additional correlation structure. Validation on mock covariances withheld from classifier training reports improvements over the standard smoothing procedure both in matrix reconstruction metrics and in the recovery of cosmological parameter posteriors relative to the covariance measured from a large number of mocks.

Significance. If the reported gains hold when the method is applied to real observations, the approach would address a practical bottleneck in high-dimensional cosmological analyses where the sample covariance is singular or noisy. The explicit use of withheld mocks for validation and the comparison against both smoothing and the “true” mock covariance constitute a clear, reproducible test of the internal consistency of the procedure.

major comments (2)
  1. [§4] §4 (validation): all quantitative improvements in matrix metrics and posterior recovery are demonstrated exclusively on mock covariances drawn from the same DESI DR1 mock suite used to construct both eigenbases. No test is presented that quantifies degradation when the mock correlation structure is deliberately mismatched to the test covariance (e.g., by rescaling off-diagonal elements or altering the input cosmology), which is the central assumption required for application to real data.
  2. [§3.2] §3.2 (residual correction): the scalar weight multiplying the classifier-derived residual correction is tuned on the training mocks. The manuscript does not show that the optimal weight remains stable under variations in mock realism or that the final cosmological posteriors are insensitive to reasonable changes in this weight; because the weight is the only free parameter, its robustness is load-bearing for the claim of improved inference.
minor comments (2)
  1. [§2] The notation for the two eigenbases (reference vs. classifier) is introduced without a compact symbol table; a short table in §2 or §3 would improve readability.
  2. [Figure 3] Figure 3 caption does not state the number of withheld mocks used for the posterior comparison; adding this number would allow direct assessment of statistical significance of the reported improvement.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed and constructive report. We address the major comments below and propose revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§4] §4 (validation): all quantitative improvements in matrix metrics and posterior recovery are demonstrated exclusively on mock covariances drawn from the same DESI DR1 mock suite used to construct both eigenbases. No test is presented that quantifies degradation when the mock correlation structure is deliberately mismatched to the test covariance (e.g., by rescaling off-diagonal elements or altering the input cosmology), which is the central assumption required for application to real data.

    Authors: We agree that the lack of tests on mismatched mocks is a limitation for claiming applicability to real data. The current results validate the method when the test covariance shares the same underlying structure as the training mocks. To address this concern, we will add a new analysis in the revised §4, where we deliberately mismatch the test covariances by altering the input cosmology or rescaling off-diagonal correlations, and quantify the performance relative to the true covariance and the smoothing method. revision: yes

  2. Referee: [§3.2] §3.2 (residual correction): the scalar weight multiplying the classifier-derived residual correction is tuned on the training mocks. The manuscript does not show that the optimal weight remains stable under variations in mock realism or that the final cosmological posteriors are insensitive to reasonable changes in this weight; because the weight is the only free parameter, its robustness is load-bearing for the claim of improved inference.

    Authors: The scalar weight is the sole tunable parameter and was optimized on the training set. We will revise the manuscript to include a sensitivity analysis, varying the weight around its optimal value and assessing the impact on the recovered cosmological posteriors. Additionally, we will test the stability using different training mock subsets to demonstrate robustness. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method is an empirical denoising procedure validated on held-out mocks.

full rationale

The paper defines a multi-eigenbasis procedure that projects a noisy data-driven covariance first onto a mock-derived reference eigenbasis and then applies a residual correction from a second mock-trained classifier eigenbasis. Validation metrics and posterior recovery are computed on withheld mock covariances and compared to the ensemble 'true' covariance obtained from many independent mock realizations. No equation or step in the provided description reduces the claimed improvement to a fitted parameter or self-definition by construction; the mock bases are external inputs, the test set is withheld, and the comparison target is the direct ensemble average rather than any quantity derived from the denoising itself. No self-citations, uniqueness theorems, or ansatzes are invoked as load-bearing elements. The derivation chain is therefore self-contained as a simulation-conditioned algorithm whose performance claims rest on explicit out-of-sample comparison rather than tautological reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim depends on the assumption that the provided mock simulations faithfully represent the statistical properties of the real Lyα forest data; the method introduces no new physical entities but does rely on the classifier learning genuine additional correlation structure from the mocks.

free parameters (1)
  • weight for residual correction
    The abstract states that a weighted residual correction is constructed; the weight is a tunable parameter whose value is not derived from first principles and must be chosen to optimize performance.
axioms (1)
  • domain assumption Mock simulations accurately reproduce the true covariance structure of the observational data
    The entire denoising pipeline is conditioned on DESI DR1 mock simulations for both the reference eigenbasis and the classifier training; the validation uses withheld mocks from the same suite.

pith-pipeline@v0.9.1-grok · 5798 in / 1520 out tokens · 34471 ms · 2026-06-28T08:20:39.329178+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 40 canonical work pages · 3 internal anchors

  1. [1]

    G., Aguilar, J., Ahlen, S., et al

    Adame, A. G., Aguilar, J., Ahlen, S., et al. 2025, J. Cosmology Astropart. Phys., 2025, 124, doi: 10.1088/1475-7516/2025/01/124

    work page doi:10.1088/1475-7516/2025/01/124 2025

  2. [2]

    and Paczynski, B

    Alcock, C., & Paczynski, B. 1979, Nature, 281, 358, doi: 10.1038/281358a0

    work page doi:10.1038/281358a0 1979

  3. [3]

    E., Busca, N

    Bautista, J. E., Busca, N. G., Guy, J., et al. 2017, A&A, 603, A12, doi: 10.1051/0004-6361/201730533

    work page doi:10.1051/0004-6361/201730533 2017

  4. [4]

    G., et al

    Blomqvist, M., du Mas des Bourboux, H., Busca, N. G., et al. 2019, A&A, 629, A86, doi: 10.1051/0004-6361/201935641

    work page doi:10.1051/0004-6361/201935641 2019

  5. [5]

    2025, arXiv e-prints, arXiv:2506.15262

    Busca, N., Rich, J., Bautista, J., et al. 2025, arXiv e-prints, arXiv:2506.15262. https://arxiv.org/abs/2506.15262

    arXiv 2025

  6. [6]

    G., Delubac, T., Rich, J., et al

    Busca, N. G., Delubac, T., Rich, J., et al. 2013, A&A, 552, A96, doi: 10.1051/0004-6361/201220724

    work page doi:10.1051/0004-6361/201220724 2013

  7. [7]

    K., Chaves-Montero, J., et al

    Casas, L., Herrera-Alcantar, H. K., Chaves-Montero, J., et al. 2025, arXiv e-prints, arXiv:2503.14741, doi: 10.48550/arXiv.2503.14741

    work page doi:10.48550/arxiv.2503.14741 2025

  8. [8]

    2021, MNRAS, 506, 5439, doi: 10.1093/mnras/stab1999

    Cuceu, A., Font-Ribera, A., Joachimi, B., & Nadathur, S. 2021, MNRAS, 506, 5439, doi: 10.1093/mnras/stab1999

    work page doi:10.1093/mnras/stab1999 2021

  9. [9]

    2023, MNRAS, 523, 3773, doi: 10.1093/mnras/stad1546

    Cuceu, A., Font-Ribera, A., Martini, P., et al. 2023, MNRAS, 523, 3773, doi: 10.1093/mnras/stad1546

    work page doi:10.1093/mnras/stad1546 2023

  10. [10]

    K., Gordon, C., et al

    Cuceu, A., Herrera-Alcantar, H. K., Gordon, C., et al. 2025a, arXiv e-prints, arXiv:2509.15308, doi: 10.48550/arXiv.2509.15308 —. 2025b, J. Cosmology Astropart. Phys., 2025, 148, doi: 10.1088/1475-7516/2025/01/148

    work page doi:10.48550/arxiv.2509.15308 2025

  11. [11]

    The Baryon Oscillation Spectroscopic Survey of SDSS-III

    Dawson, K. S., Schlegel, D. J., Ahn, C. P., et al. 2013, AJ, 145, 10, doi: 10.1088/0004-6256/145/1/10

    work page internal anchor Pith review doi:10.1088/0004-6256/145/1/10 2013

  12. [12]

    S., Kneib, J.-P., Percival, W

    Dawson, K. S., Kneib, J.-P., Percival, W. J., et al. 2016, AJ, 151, 44, doi: 10.3847/0004-6256/151/2/44 de Sainte Agathe, V., Balland, C., du Mas des Bourboux, H., et al. 2019, A&A, 629, A85, doi: 10.1051/0004-6361/201935638 de Santi, N. S. M., & Abramo, L. R. 2022, J. Cosmology Astropart. Phys., 2022, 013, doi: 10.1088/1475-7516/2022/09/013

    work page doi:10.3847/0004-6256/151/2/44 2016

  13. [13]

    The DESI Experiment Part I: Science,Targeting, and Survey Design

    Delubac, T., Bautista, J. E., Busca, N. G., et al. 2015, A&A, 574, A59, doi: 10.1051/0004-6361/201423969 DESI Collaboration, Adame, A. G., Aguilar, J., et al. 2016, The DESI Experiment Part I: Science,Targeting, and Survey Design, doi: 10.48550/arXiv.1611.00036 —. 2022, AJ, 164, 207, doi: 10.3847/1538-3881/ac882b —. 2024, arXiv e-prints, arXiv:2404.03002,...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004-6361/201423969 2015

  14. [14]

    Dodelson, S., & Schneider, M. D. 2013, Phys. Rev. D, 88, 063537, doi: 10.1103/PhysRevD.88.063537 du Mas des Bourboux, H., Le Goff, J.-M., Blomqvist, M., et al. 2017, A&A, 608, A130, doi: 10.1051/0004-6361/201731731 du Mas des Bourboux, H., Rich, J., Font-Ribera, A., et al. 2020, ApJ, 901, 153, doi: 10.3847/1538-4357/abb085

    work page doi:10.1103/physrevd.88.063537 2013

  15. [15]

    L., Rich, J., et al

    Etourneau, T., Goff, J.-M. L., Rich, J., et al. 2024, J. Cosmology Astropart. Phys., 2024, 077, doi: 10.1088/1475-7516/2024/05/077

    work page doi:10.1088/1475-7516/2024/05/077 2024

  16. [16]

    Denoising clustering covariance matrices with Rotational Invariant Estimators

    Farina, A., Guidi, M., Veropalumbo, A., & Guida, C. 2026, arXiv e-prints, arXiv:2604.13851, doi: 10.48550/arXiv.2604.13851

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.13851 2026

  17. [17]

    Farr, J., Font-Ribera, A., Bourboux, H. d. M. d., et al. 2020, J. Cosmology Astropart. Phys., 2020, 068, doi: 10.1088/1475-7516/2020/03/068

    work page doi:10.1088/1475-7516/2020/03/068 2020

  18. [18]

    Font-Ribera, A., Arnau, E., Miralda-Escud´ e, J., et al. 2013, J. Cosmology Astropart. Phys., 2013, 018, doi: 10.1088/1475-7516/2013/05/018

    work page doi:10.1088/1475-7516/2013/05/018 2013

  19. [19]

    Font-Ribera, A., Kirkby, D., Busca, N., et al. 2014, J. Cosmology Astropart. Phys., 2014, 027, doi: 10.1088/1475-7516/2014/05/027

    work page doi:10.1088/1475-7516/2014/05/027 2014

  20. [20]

    2024, MNRAS, 528, 2667, doi: 10.1093/mnras/stae092 G´ orski, K

    Font-Ribera, A. 2024, MNRAS, 528, 2667, doi: 10.1093/mnras/stae092 G´ orski, K. M., Hivon, E., Banday, A. J., et al. 2005, ApJ, 622, 759, doi: 10.1086/427976

    work page doi:10.1093/mnras/stae092 2024

  21. [21]

    , Simon, P

    Hartlap, J., Simon, P., & Schneider, P. 2007, A&A, 464, 399, doi: 10.1051/0004-6361:20066170

    work page doi:10.1051/0004-6361:20066170 2007

  22. [22]

    K., Mu˜ noz-Guti´ errez, A., Tan, T., et al

    Herrera-Alcantar, H. K., Mu˜ noz-Guti´ errez, A., Tan, T., et al. 2025, J. Cosmology Astropart. Phys., 2025, 141, doi: 10.1088/1475-7516/2025/01/141

    work page doi:10.1088/1475-7516/2025/01/141 2025

  23. [23]

    2017, MNRAS, 466, L83, doi: 10.1093/mnrasl/slw240 Kara¸ caylı, N

    Joachimi, B. 2017, MNRAS, 466, L83, doi: 10.1093/mnrasl/slw240 Kara¸ caylı, N. G., Cuceu, A., Aguilar, J., et al. 2026, arXiv e-prints, arXiv:2603.04281, doi: 10.48550/arXiv.2603.04281

    work page doi:10.1093/mnrasl/slw240 2017

  24. [24]

    Kirkby, D., Margala, D., Slosar, A., et al. 2013, J. Cosmology Astropart. Phys., 2013, 024, doi: 10.1088/1475-7516/2013/03/024

    work page doi:10.1088/1475-7516/2013/03/024 2013

  25. [25]

    Kullback, S., & Leibler, R. A. 1951, Ann. Math. Stat., 22, 79

    1951

  26. [26]

    2016, Ann

    Lam, C. 2016, Ann. Stat., 44, 928, doi: 10.1214/15-AOS1393

    work page doi:10.1214/15-aos1393 2016

  27. [27]

    2012, Ann

    Ledoit, O., & Wolf, M. 2012, Ann. Stat., 40, 1024, doi: 10.1214/12-aos989

    work page doi:10.1214/12-aos989 2012

  28. [28]

    2021, MNRAS, 502, 3510, doi: 10.1093/mnras/stab177

    Liu, B., & Bordoloi, R. 2021, MNRAS, 502, 3510, doi: 10.1093/mnras/stab177

    work page doi:10.1093/mnras/stab177 2021

  29. [29]

    J., Wang, M

    Looijmans, M. J., Wang, M. S., & Beutler, F. 2025, MNRAS, 537, 21, doi: 10.1093/mnras/stae2786 14

    work page doi:10.1093/mnras/stae2786 2025

  30. [30]

    2021, simqso: Simulated quasar spectra generator, Astrophysics Source Code Library, record ascl:2106.008

    McGreer, I., Moustakas, J., & Schindler, J. 2021, simqso: Simulated quasar spectra generator, Astrophysics Source Code Library, record ascl:2106.008. http://ascl.net/2106.008

    2021

  31. [31]

    J., Friedrich, O., Sellentin, E., & Heavens, A

    Percival, W. J., Friedrich, O., Sellentin, E., & Heavens, A. 2022, MNRAS, 510, 3207, doi: 10.1093/mnras/stab3540

    work page doi:10.1093/mnras/stab3540 2022

  32. [32]

    J., Ross, A

    Percival, W. J., Ross, A. J., S´ anchez, A. G., et al. 2014, MNRAS, 439, 2531, doi: 10.1093/mnras/stu112

    work page doi:10.1093/mnras/stu112 2014

  33. [33]

    Philcox, O. H. E., Ivanov, M. M., Zaldarriaga, M., Simonovi´ c, M., & Schmittfull, M. 2021, Phys. Rev. D, 103, 043508, doi: 10.1103/PhysRevD.103.043508

    work page doi:10.1103/physrevd.103.043508 2021

  34. [34]

    2025, A&A, 698, A292, doi: 10.1051/0004-6361/202453377 Planck Collaboration, Ade, P

    Pistis, F., Fumagalli, M., Fossati, M., et al. 2025, A&A, 698, A292, doi: 10.1051/0004-6361/202453377 Planck Collaboration, Ade, P. A. R., Aghanim, N., et al. 2016, A&A, 594, A13, doi: 10.1051/0004-6361/201525830 Planck Collaboration, Aghanim, N., Akrami, Y., et al. 2020, A&A, 641, A6, doi: 10.1051/0004-6361/201833910

    work page doi:10.1051/0004-6361/202453377 2025

  35. [35]

    , keywords =

    Pope, A. C., & Szapudi, I. 2008, MNRAS, 389, 766, doi: 10.1111/j.1365-2966.2008.13561.x Pˆ aris, I., Petitjean, P., Rollinde, E., et al. 2011, A&A, 530, A50, doi: 10.1051/0004-6361/201016233 Ram´ ırez-P´ erez, C., Sanchez, J., Alonso, D., & Font-Ribera, A. 2022, J. Cosmology Astropart. Phys., 2022, 002, doi: 10.1088/1475-7516/2022/05/002 Ram´ ırez-P´ erez...

    work page doi:10.1111/j.1365-2966.2008.13561.x 2008

  36. [36]

    Sellentin, E., & Heavens, A. F. 2016, MNRAS, 456, L132, doi: 10.1093/mnrasl/slv190

    work page doi:10.1093/mnrasl/slv190 2016

  37. [37]

    M., et al

    Slosar, A., Font-Ribera, A., Pieri, M. M., et al. 2011, J. Cosmology Astropart. Phys., 2011, 001, doi: 10.1088/1475-7516/2011/09/001

    work page doi:10.1088/1475-7516/2011/09/001 2011

  38. [38]

    Slosar, A., Irˇ siˇ c, V., Kirkby, D., et al. 2013, J. Cosmology Astropart. Phys., 2013, 026, doi: 10.1088/1475-7516/2013/04/026

    work page doi:10.1088/1475-7516/2013/04/026 2013

  39. [39]

    M., & Lubin, D

    Suzuki, N., Tytler, D., Kirkman, D., O’Meara, J. M., & Lubin, D. 2005, ApJ, 618, 592, doi: 10.1086/426062

    work page doi:10.1086/426062 2005

  40. [40]

    2014, MNRAS, 442, 2728, doi: 10.1093/mnras/stu996

    Taylor, A., & Joachimi, B. 2014, MNRAS, 442, 2728, doi: 10.1093/mnras/stu996

    work page doi:10.1093/mnras/stu996 2014

  41. [41]

    2013, MNRAS, 432, 1928, doi: 10.1093/mnras/stt270

    Taylor, A., Joachimi, B., & Kitching, T. 2013, MNRAS, 432, 1928, doi: 10.1093/mnras/stt270

    work page doi:10.1093/mnras/stt270 2013

  42. [42]

    G., et al

    Turner, W., Martini, P., Kara¸ caylı, N. G., et al. 2024, ApJ, 976, 143, doi: 10.3847/1538-4357/ad8239

    work page doi:10.3847/1538-4357/ad8239 2024

  43. [43]

    Turner, W., Cuceu, A., Martini, P., et al. 2026, J. Cosmology Astropart. Phys., 2026, 010, doi: 10.1088/1475-7516/2026/05/010 APPENDIX SENSITIVITY TO CHOICE OF REFERENCE MATRIX As discussed in Section 3, I derive the initial eigenbasis from a reference correlation matrix based on a combina- tion of Saclay and LyαCoLoRe mocks. I assume that the true correl...

    work page doi:10.1088/1475-7516/2026/05/010 2026