pith. machine review for the scientific record. sign in

arxiv: 2605.00980 · v1 · submitted 2026-05-01 · 🌌 astro-ph.CO · astro-ph.IM· stat.ME

Recognition: unknown

Coverage is not enough: Frequentist tests of simulation-based inference for primordial non-Gaussianity

Toka Alokda , Cristiano Porciani , Alexander Eggemeier
Authors on Pith no claims yet

Pith reviewed 2026-05-09 18:27 UTC · model grok-4.3

classification 🌌 astro-ph.CO astro-ph.IMstat.ME
keywords primordial non-Gaussianitysimulation-based inferencecoverage testspower spectrumbispectrumwavelet scattering transformf_NLdark matter halos
0
0 comments X

The pith

Simulation-based inference for primordial non-Gaussianity passes coverage tests yet produces underconfident posteriors with weaker constraints than individual statistics.

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

The paper tests whether coverage diagnostics suffice to validate simulation-based inference when extracting the primordial non-Gaussianity parameter f_NL from simulated dark matter halo fields. Using contrastive neural ratio estimation, the authors compare SBI posteriors to likelihood-based inference that assumes a Gaussian likelihood for the power spectrum, bispectrum, and wavelet scattering transform coefficients. They find that SBI and likelihood-based results agree on average for means and skewness across 1000 realizations, but variance shows only moderate consistency and kurtosis differs more strongly, indicating problems in the posterior tails. These mismatches already appear for the power spectrum alone and grow largest for the combined power spectrum plus bispectrum case, where SBI posteriors are frequently underconfident and return looser bounds on f_NL than either statistic yields by itself. The work also shows that wavelet scattering transform coefficients tighten constraints even when restricted to large scales.

Core claim

Coverage tests under the prior predictive distribution only verify calibration in an averaged sense and leave unchecked the posterior shape at any fixed value of f_NL. When SBI is applied to halo-field simulations, it passes these tests but still yields posteriors whose variance and especially kurtosis deviate from likelihood-based results, producing underconfident constraints that are weaker than those from the power spectrum or bispectrum alone in the combined case.

What carries the argument

Contrastive neural ratio estimation (CNRE) for simulation-based inference, validated against likelihood-based inference on power spectrum, bispectrum, and wavelet scattering transform summaries of the halo field.

If this is right

  • SBI posteriors for the combined power spectrum and bispectrum are often underconfident and yield weaker f_NL constraints than either statistic separately.
  • Discrepancies in posterior variance and kurtosis already appear when SBI is applied to the power spectrum alone.
  • Wavelet scattering transform coefficients provide tighter f_NL bounds than power spectrum plus bispectrum even when limited to large scales.
  • Posterior means and skewness agree well between SBI and likelihood-based inference across realizations.

Where Pith is reading between the lines

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

  • Cosmologists should supplement coverage checks with tests that fix the parameter value and verify actual interval coverage on individual realizations.
  • The same validation approach could be applied to SBI for other cosmological parameters where analytic likelihoods are unavailable.
  • If SBI remains underconfident for higher-order statistics, new training or calibration strategies that target posterior tails may be required.

Load-bearing premise

Observed differences between SBI and likelihood-based posteriors reflect shortcomings in SBI calibration rather than inaccuracies in the Gaussian likelihood assumption used for likelihood-based inference.

What would settle it

A set of simulations at one fixed true value of f_NL in which the fraction of cases where the SBI 68 percent credible interval contains the true value deviates substantially from 68 percent would show that coverage has not guaranteed correct posterior behavior at that parameter.

Figures

Figures reproduced from arXiv: 2605.00980 by Alexander Eggemeier, Cristiano Porciani, Toka Alokda.

Figure 1
Figure 1. Figure 1: Comparison of posterior properties obtained with LBI ( view at source ↗
Figure 2
Figure 2. Figure 2: Summary statistics (mean, standard deviation, 68% and view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of LBI (blue), SBIsbc (orange), and SBIloss (green) posteriors for six test realizations with fNL = 0 using P + B. These cases correspond to normalized 2-Wasserstein dis￾tances between the LBI and SBIsbc posteriors close to the 90th percentile of the distribution in view at source ↗
Figure 3
Figure 3. Figure 3: CDF of the KLD and 2-Wasserstein distance between view at source ↗
read the original abstract

(Abridged) Simulation-based inference (SBI) has emerged as a powerful framework for extracting cosmological information from complex, non-linear data where analytical likelihoods are unavailable. Its reliability is commonly assessed using coverage-based diagnostics under the prior predictive distribution, which probe calibration only in an averaged sense and do not constrain posterior behavior at fixed parameter value, the regime relevant for practical inference. We investigate these limitations in the context of primordial non-Gaussianity, parameterized by $f_\mathrm{NL}$, using simulations of the dark matter halo field. We compare SBI based on contrastive neural ratio estimation (CNRE) with likelihood-based inference (LBI) using the power spectrum, bispectrum, and wavelet scattering transform (WST) coefficients across 1000 realizations. SBI and LBI agree well on posterior means and skewness, while the variance agrees on average but shows weaker realization-by-realization consistency. Larger differences arise in the kurtosis, indicating discrepancies in the posterior tails. These effects are already present for the power spectrum - where the Gaussian likelihood assumed in LBI is best justified - and are most pronounced for the combined power spectrum and bispectrum, where SBI posteriors are often underconfident and can yield weaker constraints than either statistic individually, despite passing coverage tests. WST coefficients further tighten constraints on $f_\mathrm{NL}$, even when restricted to large scales. Our results highlight both the potential of higher-order statistics and the need for validation strategies that probe the posterior shape beyond standard coverage diagnostics.

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 argues that coverage diagnostics for simulation-based inference (SBI) are insufficient because they average over the prior predictive distribution and do not test posterior calibration at fixed parameter values. In the context of primordial non-Gaussianity parameterized by f_NL, the authors perform an empirical comparison of SBI (using contrastive neural ratio estimation) against likelihood-based inference (LBI) assuming Gaussian likelihoods on power spectrum, bispectrum, and wavelet scattering transform coefficients. Using 1000 halo-field realizations, they report good agreement on posterior means and skewness, average agreement but weak realization-by-realization consistency on variance, and larger discrepancies in kurtosis. They conclude that SBI posteriors are frequently underconfident—especially for the combined power spectrum plus bispectrum case—yielding weaker constraints than the individual statistics despite passing coverage tests, while WST coefficients tighten constraints even on large scales.

Significance. If the central empirical findings hold, the work provides a concrete demonstration that standard coverage tests can miss underconfidence in SBI posteriors for cosmological inference with higher-order statistics. The use of 1000 realizations allows direct comparison of posterior moments (mean, variance, skewness, kurtosis) between SBI and LBI on the same data, which is a strength. The result also illustrates the constraining power of WST coefficients for f_NL. The findings motivate development of more stringent validation methods that probe posterior shape at fixed parameters rather than relying solely on coverage.

major comments (2)
  1. The central claim that SBI posteriors are underconfident (and can be weaker than individual statistics) rests on interpreting SBI–LBI discrepancies as evidence of SBI shortcomings. However, this attribution assumes the Gaussian likelihood in LBI is sufficiently accurate to serve as reference. Even for the power spectrum alone—where the paper states the Gaussian assumption is “best justified”—the discrete halo field, finite survey volume, and possible mode coupling can induce non-Gaussian features in the estimator distribution. The manuscript should quantify how much residual non-Gaussianity remains in the LBI baseline (e.g., via higher-moment checks or exact tests on controlled Gaussian fields) before the SBI underconfidence conclusion can be considered load-bearing.
  2. The statement that SBI “can yield weaker constraints than either statistic individually” (abstract) for the combined PS+bispectrum case requires explicit support. The manuscript should report, for a substantial fraction of the 1000 realizations, the fraction in which the SBI posterior variance exceeds both the PS-only and bispectrum-only LBI variances, together with the typical magnitude of the difference. Without this per-realization breakdown, the claim that combined SBI is weaker remains qualitative.
minor comments (2)
  1. Clarify the precise definition of “underconfident” used for the SBI posteriors (e.g., whether it is defined via variance ratio, credible-interval coverage at fixed f_NL, or another metric) and state it explicitly in the methods section.
  2. The wavelet scattering transform results are promising; a brief comparison of the number of coefficients retained versus the information gain relative to PS+bispectrum would help readers assess computational trade-offs.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their insightful and constructive comments, which have prompted us to strengthen the presentation of our results and caveats. We respond to each major comment below.

read point-by-point responses

  1. Referee: The central claim that SBI posteriors are underconfident (and can be weaker than individual statistics) rests on interpreting SBI–LBI discrepancies as evidence of SBI shortcomings. However, this attribution assumes the Gaussian likelihood in LBI is sufficiently accurate to serve as reference. Even for the power spectrum alone—where the paper states the Gaussian assumption is “best justified”—the discrete halo field, finite survey volume, and possible mode coupling can induce non-Gaussian features in the estimator distribution. The manuscript should quantify how much residual non-Gaussianity remains in the LBI baseline (e.g., via higher-moment checks or exact tests on controlled Gaussian fields) before the SBI underconfidence conclusion can be considered load-bearing.

    Authors: We agree that the accuracy of the Gaussian LBI baseline must be carefully assessed before attributing discrepancies primarily to SBI. Although the Gaussian assumption is best justified for the power spectrum, residual non-Gaussianity from the discrete halo field and finite volume is possible. In the revised manuscript we will add explicit higher-moment checks (skewness and kurtosis) on the LBI estimator distributions for the power spectrum to quantify deviations from Gaussianity. We will also expand the discussion to include stronger caveats on the limitations of the LBI reference. However, performing exact tests on controlled Gaussian fields would require a new suite of simulations that is outside the scope of the present work; we therefore cannot provide that specific quantification. We maintain that the fact that discrepancies already appear in the power-spectrum-only case (where the Gaussian assumption is strongest) still supports our broader point about coverage tests, but we will make the limitations more prominent. revision: partial

  2. Referee: The statement that SBI “can yield weaker constraints than either statistic individually” (abstract) for the combined PS+bispectrum case requires explicit support. The manuscript should report, for a substantial fraction of the 1000 realizations, the fraction in which the SBI posterior variance exceeds both the PS-only and bispectrum-only LBI variances, together with the typical magnitude of the difference. Without this per-realization breakdown, the claim that combined SBI is weaker remains qualitative.

    Authors: We thank the referee for this request, which will make the claim quantitative. Using our existing set of 1000 realizations, we will add a new panel or supplementary table in the revised manuscript that reports (i) the fraction of realizations in which the SBI combined (PS+bispectrum) posterior variance exceeds both the PS-only and bispectrum-only LBI variances, and (ii) the typical magnitude of the difference (e.g., median and interquartile range of the variance ratios). This analysis is straightforward with the data already in hand and will be included. revision: yes

Circularity Check

0 steps flagged

Empirical comparison of SBI and LBI on shared simulations contains no circular derivation

full rationale

The paper conducts a direct numerical comparison of CNRE-based SBI posteriors against LBI posteriors (using power spectrum, bispectrum, and WST summaries) on the same 1000 halo-field realizations for f_NL. All reported quantities—posterior means, variances, skewness, kurtosis, and coverage diagnostics—are computed from the outputs of these two pipelines applied to the simulations. No equation or result is obtained by fitting a parameter to a subset of the data and then relabeling it as a prediction, nor does any central claim reduce to a self-citation chain or an ansatz imported from the authors' prior work. The methods themselves are standard or externally referenced, and the conclusion that coverage tests are insufficient follows from the observed empirical discrepancies rather than from any definitional loop within the paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger reflects assumptions implied by the described methods. No free parameters or new entities are introduced. The main domain assumption is the validity of the Gaussian likelihood for LBI on the chosen statistics.

axioms (1)
  • domain assumption Likelihood-based inference assumes a Gaussian likelihood for the power spectrum, bispectrum, and wavelet scattering transform coefficients.
    Stated as the basis for the LBI comparison in the abstract.

pith-pipeline@v0.9.0 · 5587 in / 1436 out tokens · 41592 ms · 2026-05-09T18:27:35.015991+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

54 extracted references · 32 canonical work pages

  1. [1]

    International Conference on Machine Learning , year=

    Automatic Posterior Transformation for Likelihood-Free Inference , author=. International Conference on Machine Learning , year=

  2. [2]

    Proceedings of the National Academy of Sciences , volume =

    Kyle Cranmer and Johann Brehmer and Gilles Louppe , title =. Proceedings of the National Academy of Sciences , volume =. 2020 , doi =

    2020

  3. [3]

    Proceedings of the 37th International Conference on Machine Learning , articleno =

    Durkan, Conor and Murray, Iain and Papamakarios, George , title =. Proceedings of the 37th International Conference on Machine Learning , articleno =. 2020 , publisher =

    2020

  4. [4]

    Advances in Neural Information Processing Systems , editor=

    Contrastive Neural Ratio Estimation , author=. Advances in Neural Information Processing Systems , editor=. 2022 , url=

    2022

  5. [5]

    Advances in Neural Information Processing Systems , editor=

    Towards Reliable Simulation-Based Inference with Balanced Neural Ratio Estimation , author=. Advances in Neural Information Processing Systems , editor=. 2022 , url=

    2022

  6. [6]

    , keywords =

    Percival, Will J and Friedrich, Oliver and Sellentin, Elena and Heavens, Alan , year=. Matching Bayesian and frequentist coverage probabilities when using an approximate data covariance matrix , volume=. MNRAS , publisher=. doi:10.1093/mnras/stab3540 , number=

    work page doi:10.1093/mnras/stab3540

  7. [7]

    2020 , eprint=

    Validating Bayesian Inference Algorithms with Simulation-Based Calibration , author=. 2020 , eprint=

    2020

  8. [8]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    Sampling-Based Accuracy Testing of Posterior Estimators for General Inference , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  9. [9]

    Towards an optimal estimation of cosmological parameters with the wavelet scattering transform , volume=

    Valogiannis, Georgios and Dvorkin, Cora , year=. Towards an optimal estimation of cosmological parameters with the wavelet scattering transform , volume=. Phys. reviews D. , publisher=. doi:10.1103/physrevd.105.103534 , number=

    work page doi:10.1103/physrevd.105.103534

  10. [10]

    Going beyond the galaxy power spectrum: An analysis of BOSS data with wavelet scattering transforms , volume=

    Valogiannis, Georgios and Dvorkin, Cora , year=. Going beyond the galaxy power spectrum: An analysis of BOSS data with wavelet scattering transforms , volume=. Phys. reviews D. , publisher=. doi:10.1103/physrevd.106.103509 , number=

    work page doi:10.1103/physrevd.106.103509

  11. [11]

    2024 , month =

    Peron, Matteo and Jung, Gabriel and Liguori, Michele and Pietroni, Massimo , title =. 2024 , month =. doi:10.1088/1475-7516/2024/07/021 , url =

    work page doi:10.1088/1475-7516/2024/07/021 2024

  12. [12]

    , journal=

    Bruna, Joan and Mallat, S. , journal=. 2013 , volume=. doi:10.1109/TPAMI.2012.230 , url =

    work page doi:10.1109/tpami.2012.230 2013

  13. [13]

    Solid Harmonic Wavelet Scattering: Predicting Quantum Molecular Energy from Invariant Descriptors of 3D Electronic Densities , url =

    Eickenberg, Michael and Exarchakis, Georgios and Hirn, Matthew and Mallat, Stephane , booktitle =. Solid Harmonic Wavelet Scattering: Predicting Quantum Molecular Energy from Invariant Descriptors of 3D Electronic Densities , url =

  14. [14]

    Planck 2018 results. I. Overview and the cosmological legacy of Planck. , keywords =. doi:10.1051/0004-6361/201833880 , archivePrefix =. 1807.06205 , primaryClass =

    work page doi:10.1051/0004-6361/201833880 2018

  15. [15]

    Coulton, William R and Villaescusa-Navarro, Francisco and Jamieson, Drew and Baldi, Marco and Jung, Gabriel and Karagiannis, Dionysios and Liguori, Michele and Verde, Licia and Wandelt, Benjamin D. , year=. Quijote-PNG: Simulations of Primordial Non-Gaussianity and the Information Content of the Matter Field Power Spectrum and Bispectrum , volume=. The As...

    work page doi:10.3847/1538-4357/aca8a7

  16. [16]

    The Astrophysical Journal Supplement Series , volume =

    Villaescusa-Navarro, Francisco and others , title =. The Astrophysical Journal Supplement Series , volume =. 2020 , doi =

    2020

  17. [17]

    Communications in Statistics-Theory and Methods , volume=

    A class of invariant and consistent tests for multivariate normality , author=. Communications in Statistics-Theory and Methods , volume=. 1990 , publisher=

    1990

  18. [18]

    Proceedings of the National Academy of Sciences , year=

    Averting a Crisis in Simulation-Based Inference , author=. Proceedings of the National Academy of Sciences , year=

  19. [19]

    Advances in Neural Information Processing Systems , volume=

    Benchmarking Simulation-Based Inference , author=. Advances in Neural Information Processing Systems , volume=

  20. [20]

    Astronomy & Astrophysics , author =

    Hartlap, J. and Simon, P. and Schneider, P. , year=. Why your model parameter confidences might be too optimistic. Unbiased estimation of the inverse covariance matrix , volume=. , publisher=. doi:10.1051/0004-6361:20066170 , number=

    work page doi:10.1051/0004-6361:20066170

  21. [21]

    Dodelson, Scott and Schneider, Michael D. , year=. The effect of covariance estimator error on cosmological parameter constraints , volume=. Physical Review D , publisher=. doi:10.1103/physrevd.88.063537 , number=

    work page doi:10.1103/physrevd.88.063537

  22. [22]

    Imprints of primordial non-Gaussianities on large-scale structure: Scale-dependent bias and abundance of virialized objects , volume=

    Dalal, Neal and Doré, Olivier and Huterer, Dragan and Shirokov, Alexander , year=. Imprints of primordial non-Gaussianities on large-scale structure: Scale-dependent bias and abundance of virialized objects , volume=. Physical Review D , publisher=. doi:10.1103/physrevd.77.123514 , number=

    work page doi:10.1103/physrevd.77.123514

  23. [23]

    Communications on Pure and Applied Mathematics , volume =

    Mallat, Stéphane , title =. Communications on Pure and Applied Mathematics , volume =. doi:https://doi.org/10.1002/cpa.21413 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1002/cpa.21413 , year =

    work page doi:10.1002/cpa.21413

  24. [24]

    Constraints on local primordial non-Gaussianity from large scale structure , volume=

    Slosar, Anže and Hirata, Christopher and Seljak, Uroš and Ho, Shirley and Padmanabhan, Nikhil , year=. Constraints on local primordial non-Gaussianity from large scale structure , volume=. Journal of Cosmology and Astroparticle Physics , publisher=. doi:10.1088/1475-7516/2008/08/031 , number=

    work page doi:10.1088/1475-7516/2008/08/031 2008

  25. [25]

    Field-level simulation-based inference of galaxy clustering with convolutional neural networks , author =. Phys. Rev. D , volume =. 2024 , month =. doi:10.1103/PhysRevD.109.083536 , url =

    work page doi:10.1103/physrevd.109.083536 2024

  26. [26]

    and Crocce, M

    Sefusatti, E. and Crocce, M. and Scoccimarro, R. and Couchman, H. M. P. , year=. Accurate estimators of correlation functions in Fourier space , volume=. , publisher=. doi:10.1093/mnras/stw1229 , number=

    work page doi:10.1093/mnras/stw1229

  27. [27]

    2011 , month =

    Tobias Baldauf and Uroš Seljak and Leonardo Senatore , title =. 2011 , month =. doi:10.1088/1475-7516/2011/04/006 , url =

    work page doi:10.1088/1475-7516/2011/04/006 2011

  28. [28]

    The Three point correlation function of the cosmic microwave background in inflationary models

    Gangui, Alejandro and Lucchin, Francesco and Matarrese, Sabino and Mollerach, Silvia , year=. The three-point correlation function of the cosmic microwave background in inflationary models , volume=. doi:10.1086/174421 , journal=

    work page doi:10.1086/174421

  29. [29]

    Akiba, S

    Akiba, Takuya and Sano, Shotaro and Yanase, Toshihiko and Ohta, Takeru and Koyama, Masanori , title =. 2019 , isbn =. doi:10.1145/3292500.3330701 , booktitle =

    work page doi:10.1145/3292500.3330701 2019

  30. [30]

    2025 , eprint=

    How many simulations do we need for simulation-based inference in cosmology? , author=. 2025 , eprint=

    2025

  31. [31]

    Planck 2018 results. IX. Constraints on primordial non-Gaussianity. , keywords =. doi:10.1051/0004-6361/201935891 , archivePrefix =. 1905.05697 , primaryClass =

    work page doi:10.1051/0004-6361/201935891 2018

  32. [32]

    Gradshteyn, I. S. and Ryzhik, I. M. , biburl =

  33. [33]

    2025 , eprint=

    Simulation-Based Inference: A Practical Guide , author=. 2025 , eprint=

    2025

  34. [34]

    Jan Boelts and Michael Deistler and Manuel Gloeckler and Álvaro Tejero-Cantero and Jan-Matthis Lueckmann and Guy Moss and Peter Steinbach and Thomas Moreau and Fabio Muratore and Julia Linhart and Conor Durkan and Julius Vetter and Benjamin Kurt Miller and Maternus Herold and Abolfazl Ziaeemehr and Matthijs Pals and Theo Gruner and Sebastian Bischoff and ...

    work page doi:10.21105/joss.07754 2025

  35. [35]

    K. V. Mardia , journal =. Measures of Multivariate Skewness and Kurtosis with Applications , urldate =

  36. [36]

    Székely and Maria L

    Gábor J. Székely and Maria L. Rizzo , keywords =. A new test for multivariate normality , journal =. 2005 , issn =. doi:https://doi.org/10.1016/j.jmva.2003.12.002 , url =

    work page doi:10.1016/j.jmva.2003.12.002 2005

  37. [37]

    TEST , volume =

    Ebner, Bruno and Henze, Norbert , title =. TEST , volume =. 2020 , doi =

    2020

  38. [38]

    2019 , eprint=

    Testing multivariate normality by zeros of the harmonic oscillator in characteristic function spaces , author=. 2019 , eprint=

    2019

  39. [39]

    2019 , eprint=

    A new test of multivariate normality by a double estimation in a characterizing PDE , author=. 2019 , eprint=

    2019

  40. [40]

    The shape of non-Gaussianities , volume=

    Babich, Daniel and Creminelli, Paolo and Zaldarriaga, Matias , year=. The shape of non-Gaussianities , volume=. Journal of Cosmology and Astroparticle Physics , publisher=. doi:10.1088/1475-7516/2004/08/009 , number=

    work page doi:10.1088/1475-7516/2004/08/009 2004

  41. [41]

    The Effect of Primordial Non-Gaussianity on Halo Bias , volume=

    Matarrese, Sabino and Verde, Licia , year=. The Effect of Primordial Non-Gaussianity on Halo Bias , volume=. The Astrophysical Journal , publisher=. doi:10.1086/587840 , number=

    work page doi:10.1086/587840

  42. [42]

    2003 , publisher=

    An Introduction to Multivariate Statistical Analysis , author=. 2003 , publisher=

    2003

  43. [43]

    Homer, O

    Homer, J. and Friedrich, O. and Gruen, D. , year=. Simulation-based inference has its own Dodelson–Schneider effect (but it knows that it does) , volume=. doi:10.1051/0004-6361/202453339 , journal=

    work page doi:10.1051/0004-6361/202453339

  44. [44]

    , keywords =

    Structure formation from non-Gaussian initial conditions: Multivariate biasing, statistics, and comparison with N-body simulations. , keywords =. doi:10.1103/PhysRevD.81.063530 , archivePrefix =. 0911.0017 , primaryClass =

    work page doi:10.1103/physrevd.81.063530

  45. [45]

    X., & Bloom, J

    Constraining primordial non-Gaussianity with future galaxy surveys. , keywords =. doi:10.1111/j.1365-2966.2012.20604.x , archivePrefix =. 1109.0958 , primaryClass =

    work page doi:10.1111/j.1365-2966.2012.20604.x 2012

  46. [46]

    , year = 2014, volume =

    Percival, Will J. and Ross, Ashley J. and Sánchez, Ariel G. and Samushia, Lado and Burden, Angela and Crittenden, Robert and Cuesta, Antonio J. and Magana, Mariana Vargas and Manera, Marc and Beutler, Florian and Chuang, Chia-Hsun and Eisenstein, Daniel J. and Ho, Shirley and McBride, Cameron K. and Montesano, Francesco and Padmanabhan, Nikhil and Reid, B...

    work page doi:10.1093/mnras/stu112

  47. [47]

    , volume =

    Friedrich, Oliver and Eifler, Tim , title =. , volume =. 2017 , month =. doi:10.1093/mnras/stx2566 , url =

    work page doi:10.1093/mnras/stx2566 2017

  48. [48]

    1982 , publisher=

    Aspects of Multivariate Statistical Theory , author=. 1982 , publisher=

    1982

  49. [49]

    Biometrika , volume =

    Wishart, John , title =. Biometrika , volume =. 1928 , month =. doi:10.1093/biomet/20A.1-2.32 , url =

    work page doi:10.1093/biomet/20a.1-2.32 1928

  50. [50]

    , title =

    Sellentin, Elena and Heavens, Alan F. , title =. , volume =. 2016 , month =. doi:10.1093/mnras/stw2697 , url =

    work page doi:10.1093/mnras/stw2697 2016

  51. [51]

    Optimal constraints on local primordial non-Gaussianity from the two-point statistics of large-scale structure , author =. Phys. Rev. D , volume =. 2011 , month =. doi:10.1103/PhysRevD.84.083509 , url =

    work page doi:10.1103/physrevd.84.083509 2011

  52. [52]

    Akitsu, M

    Cosmology inference with perturbative forward modeling at the field level: a comparison with joint power spectrum and bispectrum analyses. arXiv e-prints , keywords =. doi:10.48550/arXiv.2509.09673 , archivePrefix =. 2509.09673 , primaryClass =

    work page doi:10.48550/arxiv.2509.09673

  53. [53]

    , keywords =

    How Much Information Can Be Extracted from Galaxy Clustering at the Field Level?. , keywords =. doi:10.1103/PhysRevLett.133.221006 , archivePrefix =. 2403.03220 , primaryClass =

    work page doi:10.1103/physrevlett.133.221006

  54. [54]

    Hirn and Edouard Oyallon and Sixin Zhang and Carmine Cella and Michael Eickenberg , title =

    Mathieu Andreux and Tomás Angles and Georgios Exarchakis and Roberto Leonarduzzi and Gaspar Rochette and Louis Thiry and John Zarka and Stéphane Mallat and Joakim Andén and Eugene Belilovsky and Joan Bruna and Vincent Lostanlen and Muawiz Chaudhary and Matthew J. Hirn and Edouard Oyallon and Sixin Zhang and Carmine Cella and Michael Eickenberg , title ...