Modeling nonlinear scales with COLA: preparing for LSST-Y1

TL;DR

This paper evaluates the effectiveness of the COLA method in modeling nonlinear scales for LSST-Y1 cosmic shear analysis, comparing it to high-resolution simulations and a benchmark emulator, to assess its accuracy and potential for extended cosmological models.

Contribution

It demonstrates that COLA emulators can accurately model nonlinear scales for LSST-Y1, with performance depending on the placement of high-resolution N-body samples, offering a promising computationally efficient alternative.

Findings

COLA emulators pass goodness-of-fit tests for LSST-Y1 data.

Performance depends on the placement of high-resolution N-body reference samples.

COLA emulators show promise for extended cosmological models beyond mbda CDM.

Abstract

Year 1 results of the Legacy Survey of Space and Time (LSST) will provide tighter constraints on small-scale cosmology, beyond the validity of linear perturbation theory. This heightens the demand for a computationally affordable prescription that can accurately capture nonlinearities in beyond-$\Lambda$CDM models. The COmoving Lagrangian Acceleration (COLA) method, a cost-effective \textit{N}-body technique, has been proposed as a viable alternative to high-resolution \textit{N}-body simulations for training emulators of the nonlinear matter power spectrum. In this study, we evaluate this approach by employing COLA emulators to conduct a cosmic shear analysis with LSST-Y1 simulated data across three different nonlinear scale cuts. We use the $w$CDM model, for which the \textsc{EuclidEmulator2} (\textsc{ee2}) exists as a benchmark, having been trained with high-resolution \textit{N}-body simulations. We primarily utilize COLA simulations with mass resolution $M_{\rm part}\approx 8 \times 10^{10} ~h^{-1} M_{\odot}$ and force resolution $\ell_{\rm force}=0.5 ~h^{-1}$Mpc, though we also test refined settings with $M_{\rm part}\approx 1 \times 10^{10} ~h^{-1}M_{\odot}$ and force resolution $\ell_{\rm force}=0.17 ~h^{-1}$Mpc. We find the performance of the COLA emulators is sensitive to the placement of high-resolution \textit{N}-body reference samples inside the prior, which only ensure agreement in their local vicinity. However, the COLA emulators pass stringent criteria in goodness-of-fit and parameter bias throughout the prior, when $\Lambda$CDM predictions of \textsc{ee2} are computed alongside every COLA emulator prediction, suggesting a promising approach for extended models.