The covariance of squeezed bispectrum configurations

Matteo Biagetti, Lina Castiblanco, Jorge Noreña, E. SefusattiJournal of Cosmology and Astroparticle Physics 2022 · DOI · arXiv

Abstract

Accurate modeling of the bispectrum covariance is essential for extracting cosmological information from galaxy surveys. This paper investigates the halo bispectrum covariance using a large suite of N-body simulations, specifically focusing on "squeezed" configurations where one side of the triangle is much shorter than the others. We find that the standard diagonal Gaussian approximation fails significantly in this regime due to large correlations among squeezed configurations and between the bispectrum and the long-wavelength power spectrum. By incorporating non-Gaussian terms via a response function formalism, we develop a model that aligns closely with numerical results. Finally, we demonstrate that neglecting these non-Gaussian correlations leads to an over-optimistic estimation of constraints on primordial non-Gaussianity ($f_{NL}$) by a factor of approximately two.

Introduction

The bispectrum—the Fourier transform of the three-point correlation function—is a primary tool for probing the non-linear growth of structure and the physics of the early universe. It is particularly sensitive to primordial non-Gaussianity (PNG), which provides a window into the inflationary epoch. However, the utility of the bispectrum is limited by the complexity of its covariance matrix. Because the bispectrum involves a high-dimensional set of triangle configurations, calculating a full covariance matrix from N-body simulations is computationally prohibitive. Consequently, many analyses rely on the "Gaussian approximation," which assumes that different triangle configurations are independent and that the covariance matrix is diagonal. This study demonstrates that this assumption is fundamentally flawed for the very configurations most sensitive to PNG: the squeezed limit.

The Failure of the Gaussian Approximation

In the squeezed limit, one wavevector ($k_1$) is much smaller than the other two ($k_2 \approx k_3$). These configurations are highly sensitive to the presence of long-wavelength modes. The authors find that squeezed triangles sharing the same long-wavelength mode are not independent; they exhibit strong correlations that create significant off-diagonal elements in the covariance matrix.

Furthermore, there is a substantial cross-covariance between the squeezed bispectrum and the long-wavelength power spectrum. When these probes share large-scale modes, they cannot be treated as independent. In these cases, the non-Gaussian terms in the covariance can be an order of magnitude larger than the Gaussian terms, even within the quasi-linear regime where perturbation theory is typically expected to perform well.

Methodology and Theoretical Framework

To address these discrepancies, the authors propose a refined analytical model based on several key components:

  1. Response Function Formalism: The model uses response functions to simplify the complex higher-order statistics (specifically the 4-point trispectrum and 6-point pentaspectrum) that contribute to the bispectrum covariance. In the squeezed limit, these can be expressed efficiently in terms of the bispectrum itself.
  2. Refined Mode Counting: Standard approximations for the number of "fundamental triangles" in a bin often fail for "open" or "flattened" triangles. The authors derive new analytical expressions that reduce errors in mode counting from over 100% to less than 15%.
  3. Analytical Inversion: To make the non-diagonal covariance matrix practical for likelihood analysis, the authors utilize a generalization of the Sherman-Morrison formula. This allows for the analytical inversion of the matrix, providing a computationally efficient alternative to numerical inversion.

Numerical Validation

The model was validated using two primary datasets:

  • Quijote Simulations: A suite of 2,377 N-body realizations used to establish the "ground truth" for the halo bispectrum covariance.
  • Eos Simulations: Realizations with non-zero primordial non-Gaussianity ($f_{NL} = \pm 250$) used to calculate the derivatives required for Fisher matrix forecasts.

The proposed model matches the simulation results within ~20% for squeezed configurations. This represents a massive improvement over the Gaussian approximation, which can deviate by 100% or more in the same regime.

Impact on Primordial Non-Gaussianity

The most significant consequence of this research concerns the predicted precision of future surveys. By applying the refined covariance model to a Fisher matrix forecast, the authors found that constraints on local primordial non-Gaussianity ($f_{NL}$) are degraded by a factor of ~2 compared to the Gaussian approximation. This suggests that previous forecasts and analyses using diagonal covariances may have significantly overestimated their sensitivity to inflationary physics.

Conclusion

This work highlights the necessity of moving beyond the diagonal Gaussian approximation in bispectrum analysis. By accounting for non-Gaussian correlations in squeezed configurations, researchers can avoid biased or over-optimistic cosmological constraints. While this study focused on dark matter halos in real space, future work must extend these models to redshift space and incorporate the effects of survey window functions to fully prepare for upcoming data from missions like Euclid and DESI.

Summary updated

Abstract

Abstract We measure the halo bispectrum covariance in a large set of N-body simulations and compare it with theoretical expectations. We find a large correlation among (even mildly) squeezed halo bispectrum configurations. A similarly large correlation can be found between squeezed triangles and the long-wavelength halo power spectrum. This shows that the diagonal Gaussian contribution fails to describe, even approximately, the full covariance in these cases. We compare our numerical estimate with a model that includes, in addition to the Gaussian one, only the non-Gaussian terms that are large for squeezed configurations. We find that accounting for these large terms in the modeling greatly improves the agreement of the full covariance with simulations. We apply these results to a simple Fisher matrix forecast, and find that constraints on primordial non-Gaussianity are degraded by a factor of ∼ 2 when a non-Gaussian covariance is assumed instead of the diagonal, Gaussian approximation.
BibTeX
@article{biagetti2022covariance,
  author = {Matteo Biagetti and Lina Castiblanco and Jorge Noreña and E. Sefusatti},
  title = {{The covariance of squeezed bispectrum configurations}},
  journal = {Journal of Cosmology and Astroparticle Physics},
  year = {2022},
  doi = {10.1088/1475-7516/2022/09/009},
}
Figure