Quantification of high dimensional non-Gaussianities and its implication to Fisher analysis in cosmology

Core Francisco Park, Erwan Allys, Francisco Villaescusa-Navarro, Douglas P. Finkbeiner

Abstract

It is well known that the power spectrum is not able to fully characterize the statistical properties of non-Gaussian density fields. Recently, many different statistics have been proposed to extract information from non-Gaussian cosmological fields that perform better than the power spectrum. The Fisher matrix formalism is commonly used to quantify the accuracy with which a given statistic can constrain the value of the cosmological parameters. However, these calculations typically rely on the assumption that the likelihood of the considered statistic follows a multivariate Gaussian distribution. In this work we follow Sellentin & Heavens (2017) and use two different statistical tests to identify non-Gaussianities in different statistics such as the power spectrum, bispectrum, marked power spectrum, and wavelet scattering transform (WST). We remove the non-Gaussian components of the different statistics and perform Fisher matrix calculations with the Gaussianized statistics using Quijote simulations. We show that constraints on the parameters can change by a factor of $\sim 2$ in some cases. We show with simple examples how statistics that do not follow a multivariate Gaussian distribution can achieve artificially tight bounds on the cosmological parameters when using the Fisher matrix formalism. We think that the non-Gaussian tests used in this work represent a powerful tool to quantify the robustness of Fisher matrix calculations and their underlying assumptions. We release the code used to compute the power spectra, bispectra, and WST that can be run on both CPUs and GPUs.

Introduction

Fisher matrix analysis is the standard tool for predicting how well future cosmological surveys, such as DESI and Euclid, will constrain the universe's fundamental parameters. However, this formalism relies on a critical, often-untested assumption: that the summary statistics used follow a multivariate Gaussian distribution. As researchers move toward "non-Gaussian" statistics like the Bispectrum ($Bk$) and Wavelet Scattering Transform (WST) to capture information missed by the standard Power Spectrum ($Pk$), this assumption becomes increasingly suspect. The late-time universe is highly non-linear, and if the statistics themselves are non-Gaussian, Fisher analysis can produce "artificially tight" bounds, leading to over-optimistic scientific forecasts.

Methods and Data

The study utilizes the Quijote simulations, a massive suite of 44,100 N-body simulations. To detect and mitigate non-Gaussianity, the authors implement a two-pronged approach:

  1. Pairwise Gaussianity Test: Checks if pairs of dimensions within a statistic exhibit non-Gaussian relations.
  2. $\chi^2$ Distributional Test: Evaluates the overall shape of the likelihood peak by comparing the distribution of Mahalanobis distances to the expected $\chi^2$ distribution.

A "Gaussianization" procedure is then applied, iteratively removing dimensions that most severely violate Gaussianity until the remaining subset passes the tests. Fisher calculations are performed on both original and "Gaussianized" versions of $Pk$, $\log(Pk)$, $Mk$, $Bk$, and WST.

Covariance matrices for various cosmological statistics ($Pk$, $\log(Pk)$, $Mk$, $Bk$, $WST$). The complex structures illustrate the high-dimensional correlations analyzed in this work.

Key Results: Constraint Inflation

The analysis reveals that standard Fisher analysis can overestimate parameter precision by a factor of ~2 in some cases.

  • The Information Paradox: A key finding is that concatenating a statistic with a local transformation of itself (e.g., $Pk \oplus \log(Pk)$) can lead to tighter Fisher constraints. Since a local transformation cannot add new cosmological information, this "improvement" is a clear indicator that the Gaussian likelihood assumption has failed.
  • Bispectrum ($Bk$): This was the most sensitive probe. After Gaussianization, constraints on parameters like neutrino mass ($M_\nu$) were found to be affected by up to 170%, suggesting that previous optimism for the Bispectrum might be overstated.
  • WST Robustness: The Wavelet Scattering Transform proved relatively stable, with corrections around 50%, suggesting its components may be more naturally Gaussian.
Side-by-side comparison of Fisher constraints before (a) and after (b) Gaussianization. The expansion of the ellipses in (b) demonstrates how non-Gaussianities can lead to artificially tight parameter bounds.

Conclusion

The non-Gaussian tests presented here serve as a powerful diagnostic tool to quantify the robustness of Fisher matrix calculations. While the "Gaussianization" method is conservative and may lose some real information, it highlights the need for more robust modeling, such as Likelihood-Free Inference (LFI), when dealing with highly non-linear cosmological fields. To facilitate further research, the authors have released high-performance code for computing $Pk$, $Bk$, and WST on both CPUs and GPUs.

Summary updated

Abstract

It is well known that the power spectrum is not able to fully characterize the statistical properties of non-Gaussian density fields. Recently, many different statistics have been proposed to extract information from non-Gaussian cosmological fields that perform better than the power spectrum. The Fisher matrix formalism is commonly used to quantify the accuracy with which a given statistic can constrain the value of the cosmological parameters. However, these calculations typically rely on the assumption that the likelihood of the considered statistic follows a multivariate Gaussian distribution. In this work we follow Sellentin & Heavens (2017) and use two different statistical tests to identify non-Gaussianities in different statistics such as the power spectrum, bispectrum, marked power spectrum, and wavelet scatering transform (WST). We remove the non-Gaussian components of the different statistics and perform Fisher matrix calculations with the \textit{Gaussianized} statistics using Quijote simulations. We show that constraints on the parameters can change by a factor of $\sim 2$ in some cases. We show with simple examples how statistics that do not follow a multivariate Gaussian distribution can achieve artificially tight bounds on the cosmological parameters when using the Fisher matrix formalism. We think that the non-Gaussian tests used in this work represent a powerful tool to quantify the robustness of Fisher matrix calculations and their underlying assumptions. We release the code used to compute the power spectra, bispectra, and WST that can be run on both CPUs and GPUs.
BibTeX
@article{park2023quantification,
  author = {Core Francisco Park and Erwan Allys and Francisco Villaescusa-Navarro and Douglas P. Finkbeiner},
  title = {{Quantification of high dimensional non-Gaussianities and its implication to Fisher analysis in cosmology}},
  journal = {The Astrophysical Journal},
  year = {2023},
  doi = {10.3847/1538-4357/acbe3b},
}
Figure