Iterative reconstruction excursions for Baryon Acoustic Oscillations and beyond

TL;DR

This paper evaluates iterative reconstruction techniques for BAO feature recovery in galaxy surveys, finding limited improvements over standard methods but enhanced stability with small smoothing kernels, especially in noisy, sparse fields.

Contribution

It systematically assesses iterative reconstruction methods for BAO and broadband features, highlighting practical surrogate approaches and stability benefits in noisy, sparse galaxy fields.

Findings

Iterative steps do not significantly outperform optimized standard reconstruction.

Small smoothing kernels become more stable with iterative reconstruction.

Debiasing is crucial for displacement field reconstruction in galaxy surveys.

Abstract

The density field reconstruction technique has been widely used for recovering the Baryon Acoustic Oscillation (BAO) feature in galaxy surveys that has been degraded due to nonlinearities. Recent studies advocated adopting iterative steps to improve the recovery much beyond that of the standard technique. In this paper, we investigate the performance of a few selected iterative reconstruction techniques focusing on the BAO and the broadband-shape of the two-point clustering. We include redshift-space distortions, halo bias, and shot noise and inspect the components of the reconstructed field in Fourier space and in configuration space using both density field-based reconstruction and displacement field-based reconstruction. We find that the displacement field reconstruction becomes quickly challenging in the presence of non-negligible shot noise and therefore present surrogate methods that can be practically applied to a much more sparse field such as galaxies. For a galaxy field, implementing a debiasing step to remove the Lagrangian bias appears crucial for the displacement field reconstruction. We show that the iterative reconstruction does not substantially improve the BAO feature beyond an aggressively optimized standard reconstruction with a small smoothing kernel. However, we find taking iterative steps allows us to use a small smoothing kernel more `stably', i.e., without causing a substantial deviation from the linear power spectrum on large scales. In one specific example we studied, we find that a deviation of 13\% in $P( k \sim 0.1h/Mpc)$ with an aggressive standard reconstruction can reduce to 3-4\% with iterative steps.