Clustering analysis of medium-band selected high-redshift galaxies

TL;DR

This paper analyzes the clustering of medium-band selected high-redshift galaxies (2.3 < z < 3.5) using IBIS imaging and DESI spectroscopy, interpreted with Halo Occupation Distribution (HOD) modeling and perturbation theory. It finds that the samples constitute an overlapping mixture of Ly$\alpha$ emitters (LAEs) and Lyman break galaxies (LBGs), with a real-space clustering length of $r_0 \simeq 3-4\,h^{-1}$Mpc and a linear bias of $b \simeq 1.8-2.5$, consistent with previous work. Mock catalogs from AbacusSummit support the interpretation and enable forecasts, including a Fisher-based projection that a future $\sim$3000 deg$^2$ IBIS survey with CMB lensing could constrain $\sigma_8(z)$ at the few-percent level. The results inform survey design and simulation requirements for high-redshift cosmology and demonstrate the potential of medium-band selections to map the early universe. Overall, the work lays a framework for leveraging medium-band high-$z$ tracers to advance 3D clustering studies and cosmological constraints at $z>2$.

Abstract

Next-generation large-scale structure spectroscopic surveys will probe cosmology at high redshifts $(2.3 < z < 3.5)$, relying on abundant galaxy tracers such as Ly$α$ emitters (LAEs) and Lyman break galaxies (LBGs). Medium-band photometry has emerged as a potential technique for efficiently selecting these high-redshift galaxies. In this work, we present clustering analysis of medium-band selected galaxies at high redshift, utilizing photometric data from the Intermediate Band Imaging Survey (IBIS) and spectroscopic data from the Dark Energy Spectroscopic Instrument (DESI). We interpret the clustering of such samples using both Halo Occupation Distribution (HOD) modeling and a perturbation theory description of large-scale structure. Our modeling indicates that the current target sample is composed from an overlapping mixture of LAEs and LBGs with emission lines. Despite differences in target selection, we find that the clustering properties are consistent with previous studies, with correlation lengths $r_0\simeq 3-4\,h^{-1}$Mpc and a linear bias of $b\sim1.8-2.5$. Finally, we discuss the simulation requirements implied by these measurements and demonstrate that the properties of the samples would make them excellent targets to enhance our understanding of the high-$z$ universe.

Figures (16)

• Figure 1: The spectroscopic redshift distribution of the high redshift samples in this work (both the total and wide samples) colored by the medium band used to select it (top), the inferred 3D number density (middle), and the corresponding filter curves (bottom), plotted against redshift assuming $\lambda=(1+z)1216$Å. The $g$ band is added on the bottom panel for comparison.
• Figure 2: The broadband magnitude distributions of the medium-band selected samples. We see the $g$, $r$ and $i$-band magnitudes are generally fainter than the MB flux cuts of 25 and 25.3 magnitudes, entirely as expected for samples with emission lines. In the lower-right panel we compare the $r$-band magnitude distribution of the galaxies that we check for LBG overlap in §\ref{['sec:overlap']} (M490 and M517 galaxies that overlap the footprint of CARS Hildebrandt09; green) to those that overlap with the LBGs selected by ref. Hildebrandt09 (red).
• Figure 3: Left: The distribution of selected targets on the sky, with colors corresponding to the filters in Figure \ref{['fig:dNdz']}. Center: The performance comparison between CLAUDS SExtractor LePhare photometric redshifts, which were used to inform the target selection, and the DESI spectroscopic redshifts ($\Delta\chi^2>30$ in template fitting software RedRockRedrock). Targets with EFFTIME>1HR, but no spec-$z$, are shown on the left end of the figure. The targets are again colored according to their selection bands. Right: The photo-$z$ distribution of objects with observation time $\texttt{EFFTIME}>\texttt{1HR}$, but no reliable spec-$z$. We isolate the objects that do not reach our fiducial EFFTIME in black outlines and find that the duration of observation does not have a qualitative impact on the photo-$z$ distribution of failures.
• Figure 4: The angular clustering measurements $w_\theta(R)=w(\theta=R/\chi_0)$ of each medium band, with the total and wide sample shown on the top and bottom panels, respectively. The data measurements, in black circles, are shown together with the predictions from the best-fit HOD model (blue squares) and 5 random realizations of the best-fit HOD (faint dashed lines). Note that these measurements and errorbars are not used to directly infer goodness-of-fit to the mock and determine the best-fit HOD. That is done using $w_p(R)$, shown in Figure \ref{['fig:wR']}. The errors are calculated using 256 pseudo-independent realizations of the HOD, by offsetting the observer and observing the simulation through the survey mask and redshift selection function. For display purposes the black points have been scaled by $1/2$ in the final column, since they exhibit large scatter.
• Figure 5: The pseudo-projected clustering, $w_p(R)$, for the $z=2.5$ and 3.0 $z$-bins for each clustering sample. The data (black circles) are compared against the best-fit HOD model (blue, Table \ref{['tab:HOD_params']}) yielding fits within $\approx1\sigma$. We also overlay the predictions from a power-law correlation function $\xi=(r_0/r)^\gamma$, where the orange, green, red, and purple lines correspond to $r_0 = 2,$ 3, 4, and 5$\,h^{-1}\,\text{Mpc}$. The errors are calculated using 256 pseudo-independent realizations, as in Figure \ref{['fig:wt']}.