The clustering of galaxies in the SDSS-III Baryon Oscillation Spectroscopic Survey: Analysis of potential systematics

TL;DR

This paper systematically examines potential observational systematics in the SDSS-III BOSS DR9 CMASS/LOWZ galaxy clustering data to ensure robust cosmological constraints. It develops and validates a practical weighting scheme that mitigates imaging-related systematics, notably the strong correlation with local stellar density, and demonstrates that random redshift assignment from the data (shuffled $n(z)$) minimizes bias in clustering statistics. Using 600 mock catalogs, it quantifies covariances, tests radial selection models, and analyzes hemisphere and angular dependencies, finding BAO measurements stable under these corrections while large-scale fluctuations are largely consistent with cosmic variance. The study also reveals intriguing features at the largest scales (around $200\,h^{-1}$Mpc) whose origin remains unclear, underscoring the need for the complete BOSS dataset to confirm their nature and impact on cosmological inferences. Overall, the work provides a rigorous framework for handling systematics in large-volume galaxy surveys and supports the reliability of DR9-derived cosmological constraints when proper weighting and sampling choices are employed.

Abstract

We analyze the density field of galaxies observed by the Sloan Digital Sky Survey (SDSS)-III Baryon Oscillation Spectroscopic Survey (BOSS) included in the SDSS Data Release Nine (DR9). DR9 includes spectroscopic redshifts for over 400,000 galaxies spread over a footprint of 3,275 deg^2. We identify, characterize, and mitigate the impact of sources of systematic uncertainty on large-scale clustering measurements, both for angular moments of the redshift-space correlation function and the spherically averaged power spectrum, P(k), in order to ensure that robust cosmological constraints will be obtained from these data. A correlation between the projected density of stars and the higher redshift (0.43 < z < 0.7) galaxy sample (the `CMASS' sample) due to imaging systematics imparts a systematic error that is larger than the statistical error of the clustering measurements at scales s > 120h^-1Mpc or k < 0.01hMpc^-1. We find that these errors can be ameliorated by weighting galaxies based on their surface brightness and the local stellar density. We use mock galaxy catalogs that simulate the CMASS selection function to determine that randomly selecting galaxy redshifts in order to simulate the radial selection function of a random sample imparts the least systematic error on correlation function measurements and that this systematic error is negligible for the spherically averaged correlation function. The methods we recommend for the calculation of clustering measurements using the CMASS sample are adopted in companion papers that locate the position of the baryon acoustic oscillation feature (Anderson et al. 2012), constrain cosmological models using the full shape of the correlation function (Sanchez et al. 2012), and measure the rate of structure growth (Reid et al. 2012). (abridged)

Figures (29)

• Figure 1: The footprint of BOSS DR9 galaxies, projected into two dimensions using the McBryde-Thomas Flat Polar Quartic projection, is shaded in blue and red. Areas with CMASS data only are shaded blue. The CMASS and LOWZ footprints cover 3275 and 2208 deg$^2$, respectively. The grey area represents the final (planned) BOSS footprint.
• Figure 2: The distribution of CMASS targets for a selection of mask sectors. Circles represent the sky area covered by observing tiles and the number of overlapping tiles is indicated by the level of shading. Black dots indicate the positions of targets for which we obtained a 'good' redshift, as defined in Section \ref{['sec:zfail']}. Blue squares denote targets for which we did not allocate a fiber for spectroscopic observation because the target is within 62$^{\prime\prime}$ of another CMASS target ('close pair'). Green circles denote targets for which we did allocate a fiber that are not close pairs. Red triangles denote targets for which we allocated a fiber, but did not obtain a good redshift.
• Figure 3: The percentage of failed CMASS redshifts as a function of the position on the tile, averaged over 817 DR9 tiles. The lightest regions are 0% and the darkest regions are 12%. $\Delta \tilde{\alpha}$ is the distance along the right ascension direction and $\Delta \tilde{\delta}$ is the distance along the declination direction (both transformed so that the true angular separations are represented).
• Figure 4: Galaxy spatial co-moving number density assuming a flat $\Lambda$CDM cosmology with $\Omega_m=0.274$, for CMASS galaxies. The solid line is calculated for all galaxies, while the dashed line only includes those galaxies nearest to a redshift failure, renormalised to match the total density of the full sample. The error-bars assume Poissonian distribution for the number counts in each bin.
• Figure 5: The effect of including weights for redshift failures ($w_{zf}$, red) and un-observed close-pairs due to fiber-collisions ($w_{fc}$, blue) and their combination (black) on $\xi_{0}$ and $\xi_{2}$. The dashed lines display the $1\sigma$ statistical uncertainty expected from mock catalogs.