Imprint of DESI fiber assignment on the anisotropic power spectrum of emission line galaxies

TL;DR

This paper examines how DESI's fiber assignment pattern affects the anisotropic power spectrum of emission line galaxies, identifies systematic biases, and proposes methods to mitigate these effects for accurate cosmological measurements.

Contribution

It introduces a coverage-based random catalog method and demonstrates that analyzing $P(k,bc)$ and removing low-b0 modes effectively reduces fiber assignment contamination.

Findings

Fiber assignment causes ~10% bias on large scales in power spectrum multipoles.

Coverage randoms reduce but do not fully eliminate systematic effects.

Removing the lowest b0 bin in $P(k,bc)$ minimizes contamination to a few percent.

Abstract

The Dark Energy Spectroscopic Instrument (DESI), a multiplexed fiber-fed spectrograph, is a Stage-IV ground-based dark energy experiment aiming to measure redshifts for 29 million Emission-Line Galaxies (ELG), 4 million Luminous Red Galaxies (LRG), and 2 million Quasi-Stellar Objects (QSO). The survey design includes a pattern of tiling on the sky and the locations of the fiber positioners in the focal plane of the telescope, with the observation strategy determined by a fiber assignment algorithm that optimizes the allocation of fibers to targets. This strategy allows a given region to be covered on average five times for a five-year survey, but with coverage varying between zero and twelve, which imprints a spatially-dependent pattern on the galaxy clustering. We investigate the systematic effects of the fiber assignment coverage on the anisotropic galaxy clustering of ELGs and show that, in the absence of any corrections, it leads to discrepancies of order ten percent on large scales for the power spectrum multipoles. We introduce a method where objects in a random catalog are assigned a coverage, and the mean density is separately computed for each coverage factor. We show that this method reduces, but does not eliminate the effect. We next investigate the angular dependence of the contaminated signal, arguing that it is mostly localized to purely transverse modes. We demonstrate that the cleanest way to remove the contaminating signal is to perform an analysis of the anisotropic power spectrum $P(k,\mu)$ and remove the lowest $\mu$ bin, leaving $\mu>0$ modes accurate at the few-percent level. Here, $\mu$ is the cosine of the angle between the line-of-sight and the direction of $\vec{k}$. We also investigate two alternative definitions of the random catalog and show they are comparable but less effective than the coverage randoms method.