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 sampling distribution of the considered statistic follows a multivariate Gaussian distribution. In this work, we follow Sellentin & Heavens 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 ∼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
In modern cosmology, a primary objective is to quantify the precision with which future surveys, such as Euclid or DESI, can measure fundamental parameters like the matter density ($\Omega_m$) and neutrino mass ($M_\nu$). The Fisher Information Matrix (FIM) is the standard tool for these forecasts, prized for its ability to predict uncertainties without requiring actual observational data. However, the FIM framework rests on a critical assumption: the summary statistics employed must follow a multivariate Gaussian distribution across different realizations of the universe.
As gravity drives the formation of complex, non-linear structures on small scales, the underlying matter density field becomes highly non-Gaussian. While advanced statistics like the Bispectrum and Wavelet Scattering Transform (WST) are designed to capture this non-linear information, their own sampling distributions often violate the Gaussianity requirement. Failure to account for these non-Gaussianities can lead to "hallucinated" information, resulting in parameter constraints that are overly optimistic or physically inconsistent.
Methodology
To address this challenge, we propose a robust framework to identify and mitigate non-Gaussianities in cosmological statistics:
- Pairwise Gaussianity Testing: Following the approach of Sellentin & Heavens, we examine pairs of coefficients within a statistic to detect deviations from Gaussianity using kurtosis and skewness metrics.
- $\chi^2$ Distributional Analysis: We calculate the Mahalanobis distance for the samples. In a truly Gaussian distribution, these distances should follow a $\chi^2$ distribution.
- Iterative Gaussianization: For statistics failing these tests, we implement an iterative scheme to identify and remove the specific components (dimensions) that contribute most to non-Gaussianity until the remaining subset passes the tests.
- Fisher Comparison: We compare parameter constraints derived from the original "raw" statistics against those from the "Gaussianized" versions to quantify the bias introduced by the Gaussian assumption.
Data and Statistics
Our analysis utilizes the Quijote Simulations, a comprehensive suite of 44,100 N-body simulations. We use 15,000 simulations at the fiducial cosmology to estimate covariance matrices and sets of 500 simulations with varied parameters to calculate derivatives. We evaluate several key statistics: the Power Spectrum ($P_k$), Bispectrum ($B_k$), Marked Power Spectrum ($M_k$), and the Wavelet Scattering Transform (WST).
Results
Our findings demonstrate that ignoring non-Gaussianity in sampling distributions leads to significant constraint inflation. For the Power Spectrum, applying a log-transform ($\log P_k$) significantly improves Gaussianity on large scales, leading to more reliable forecasts. However, higher-order statistics are more vulnerable.
Key quantitative results include:
- Neutrino Mass ($M_\nu$): Constraints derived from the Bispectrum weakened by a factor of ~2 after Gaussianization.
- Marked Power Spectrum ($M_k$): Constraints on $\Omega_m$ and $\sigma_8$ roughly doubled in size, indicating that original estimates were 50% too optimistic.
- WST Performance: The WST exhibited significant non-Gaussianity; correcting for this altered predicted constraints by 56% to 90%.
Discussion and Conclusion
The tests presented here serve as a vital "sanity check" for cosmological forecasts. While our current Gaussianization method is conservative—potentially discarding real information to ensure validity—it highlights the risks of uncritical Fisher analysis. Future research will focus on optimal non-linear mappings, such as Normalizing Flows, to transform statistics into Gaussian space without losing dimensions. Additionally, we aim to apply these methodologies to observable tracers like galaxy distributions. To support the community, we have released a high-performance, GPU-accelerated library for computing these statistics.
Summary updated