Combining perturbation theories with halo models for the matter bispectrum

TL;DR

This paper develops a unified model combining perturbation theories and halo models to accurately predict the matter bispectrum across all scales, matching simulations and improving over existing methods.

Contribution

It introduces a Lagrangian framework that decomposes the bispectrum into halo contributions with counterterms, enhancing accuracy and physical consistency across scales.

Findings

Good agreement with simulations on large and small scales

Achieves ~10% accuracy on nonlinear scales

Reaches ~1% accuracy on weakly nonlinear scales

Abstract

We investigate how unified models should be built to be able to predict the matter-density bispectrum (and power spectrum) from very large to small scales and that are at the same time consistent with perturbation theory at low $k$ and with halo models at high $k$. We use a Lagrangian framework to decompose the bispectrum into "3-halo", "2-halo", and "1-halo" contributions, related to "perturbative" and "non-perturbative" terms. We describe a simple implementation of this approach and present a detailed comparison with numerical simulations. We show that the 1-halo and 2-halo contributions contain counterterms that ensure their decay at low $k$, as required by physical constraints, and allow a better match to simulations. Contrary to the power spectrum, the standard 1-loop perturbation theory can be used for the perturbative 3-halo contribution because it does not grow too fast at high $k$. Moreover, it is much simpler and more accurate than two resummation schemes investigated in this paper. We obtain a good agreement with numerical simulations on both large and small scales, but the transition scales are poorly described by the simplest implementation. This cannot be amended by simple modifications to the halo parameters, but we show how it can be corrected for the power spectrum and the bispectrum through a simple interpolation scheme that is restricted to this intermediate regime. Then, we reach an accuracy on the order of 10% on mildly and highly nonlinear scales, while an accuracy on the order of 1% is obtained on larger weakly nonlinear scales. This also holds for the real-space two-point correlation function.