The two-loop bispectrum of large-scale structure

TL;DR

This paper develops a two-loop perturbative model of the large-scale structure bispectrum within an effective field theory framework, improving accuracy for cosmological analyses in the weakly non-linear regime.

Contribution

It presents the first two-loop EFT-based bispectrum model, demonstrating UV independence and good agreement with simulations up to certain scales.

Findings

Two-loop terms become relevant at k≈0.07h/Mpc.

Percent-level agreement extends from 0.08 to 0.15h/Mpc.

EFT parameters effectively absorb UV sensitivities.

Abstract

The bispectrum is the leading non-Gaussian statistic in large-scale structure, carrying valuable information on cosmology that is complementary to the power spectrum. To access this information, we need to model the bispectrum in the weakly non-linear regime. In this work we present the first two-loop, i.e., next-to-next-to-leading order perturbative description of the bispectrum within an effective field theory (EFT) framework. Using an analytic expansion of the perturbative kernels up to $F_6$ we derive a renormalized bispectrum that is demonstrated to be independent of the UV cutoff. We show that the EFT parameters associated with the four independent second-order EFT operators known from the one-loop bispectrum are sufficient to absorb the UV sensitivity of the two-loop contributions in the double-hard region. In addition, we employ a simplified treatment of the single-hard region, introducing one extra EFT parameter at two-loop order. We compare our results to N-body simulations using the realization-based grid-PT method and find good agreement within the expected range, as well as consistent values for the EFT parameters. The two-loop terms start to become relevant at $k\approx 0.07h~\mathrm{Mpc}^{-1}$. The range of wavenumbers with percent-level agreement, independently of the shape, extends from $0.08h~\mathrm{Mpc}^{-1}$ to $0.15h~\mathrm{Mpc}^{-1}$ when going from one to two loops at $z=0$. In addition, we quantify the impact of using exact instead of Einstein-de-Sitter kernels for the one-loop bispectrum, and discuss in how far their impact can be absorbed into a shift of the EFT parameters.