LATIS: Galaxy-Environment Relations at Cosmic Noon and the Role of Sample Selection

TL;DR

This study probes galaxy environment at cosmic noon ($z\sim2.5$) by combining Ly$\alpha$ forest tomography from LATIS with UV-selected spectroscopic galaxies and a heterogeneous COSMOS compilation, all interpreted through forward-modeling with IllustrisTNG300-1. The authors quantify how spectroscopic selection biases reshape environmental trends in stellar mass and specific star formation rate (sSFR), finding that mass-complete simulations predict rising $M_*$ and declining sSFR in overdense regions, while UV-selected samples (and photometric redshifts) erase or suppress these signals, and even produce mild reversals in some cases. Through a parametric sSFR model and multiple mock realizations, they show that selection effects primarily drive the observed trends, with quiescent/dusty populations being missed in UV-selected surveys; photometric redshift uncertainties further wash out genuine correlations. The work cautions against inferring physical environmental effects from incomplete samples at $z\sim2$–3 and argues for deeper, more representative spectroscopic surveys to robustly characterize environmental influences. Overall, the paper demonstrates the importance of modeling selection biases when interpreting environment–galaxy relationships during peak galaxy assembly.

Abstract

We investigate the environmental dependence of galaxy properties at $z\sim2.5$ using the Ly$α$ Tomography IMACS Survey (LATIS), which provides high-resolution three-dimensional maps of intergalactic medium (IGM) overdensity via Ly$α$ forest tomography. Our analysis focuses on a UV-selected spectroscopic sample of 2185 galaxies from LATIS and a complementary set of 1157 galaxies from heterogeneous spectroscopic surveys in the COSMOS field. We compare these datasets to forward-modeled mock catalogs constructed from the IllustrisTNG300-1 simulation, incorporating realistic selection functions to match both LATIS and the literature sample. While the mass-complete simulation predicts strong environmental trends--more massive and quiescent galaxies preferentially occupy overdense regions--we find that such trends are significantly weaker or absent in the observed samples. The LATIS galaxies show no measurable correlation between specific star formation rate (sSFR) and IGM overdensity, a result reproduced by LATIS-like mock catalogs, confirming that UV selection systematically excludes passive and dusty galaxies in dense environments. The literature compilation, despite improved high-mass coverage, remains incomplete and affected by similar biases. We also analyze a mass-complete photometric sample from the COSMOS-Web catalog at $z\sim2.5$ and find no detectable sSFR-environment relation, a null result that our simulations indicate can be explained by photometric redshift uncertainties. In particular, we find no evidence for a reversal of the sSFR-density relation at cosmic noon. These results demonstrate that observed correlations can be heavily shaped by selection effects, and caution against inferring physical trends from incomplete spectroscopic samples. Deeper, more representative spectroscopic surveys are needed to robustly characterize environmental effects at this epoch.

Figures (9)

• Figure 1: Comparison of the star-forming main sequence at $z \sim 2.5$ between simulations and observations. The heatmap shows the number density of TNG300-1 galaxies in the $\log(\mathrm{SFR})$–$\log(\rm M_*/M_\odot)$ plane. The light gray curve represents the empirical main sequence derived from COSMOS2020 star-forming galaxies. Blue points show the median SFR in stellar mass bins for TNG300-1 galaxies with $\mathrm{sSFR} > 10^{-10.5}\,\mathrm{yr}^{-1}$, after applying a +0.2 dex correction to the simulated SFRs. Dashed black contours show the LATIS spectroscopic galaxy sample, with the innermost and outermost contours approximating the 1$\sigma$ and 2$\sigma$ loci of the distribution.
• Figure 2: Selection function maps for the LATIS (top row) and literature compilation (bottom row) spectroscopic samples. Each panel shows the median relative selection probability projected onto a two-dimensional space: stellar mass–SFR (left), $\Delta_F$–stellar mass (center), and $\Delta_F$–SFR (right). Darker regions indicate populations that are underrepresented due to selection biases. Both LATIS and the literature compilation systematically under-sample low-SFR galaxies, with LATIS further biased against massive galaxies, as expected for a UV-selected survey.
• Figure 3: Stellar mass as a function of IGM overdensity $\Delta_F$ in the mass-complete TNG300-1 simulation ($M_* \geq 10^9\,M_\odot$). Left: Average stellar mass in bins of halo mass $M_H$. Right: Same, in bins of sSFR. In both cases, galaxies in overdense regions are significantly more massive on average.
• Figure 4: Stellar mass as a function of $\Delta_F$ in the LATIS (left) and literature (right) spectroscopic samples (orange points), compared to forward-modeled mock samples from TNG300-1 (green). Black dashed lines show the observed average trend. Spearman correlation coefficients and $p$-values are shown in each panel. LATIS exhibits a nearly flat $M_*$–$\Delta_F$ relation, whereas the literature sample shows a stronger correlation; selection-matched TNG mocks reproduce both, indicating that the observed amplitudes are primarily driven by selection effects.
• Figure 5: Environmental dependence of the stellar mass function (SMF) in the TNG300-1 simulation. Each panel shows the SMF across eight bins of IGM overdensity $\Delta_F$, with equal numbers of galaxies per bin. The left panel shows the mass-complete simulation, where overdense regions contain a higher number of massive galaxies than underdense regions. The center and right panels show LATIS-like and literature-like mock samples, respectively, where spectroscopic selection effects alter the observed environmental trends. Error bars for the mass-complete SMF represent Poisson uncertainties. For the LATIS-like and literature-like samples, uncertainties reflect the standard deviation over 500 bootstrap resamplings used to generate the selection-matched mocks. In the mass-complete TNG300-1 sample, overdense regions have a more top-heavy SMF; this intrinsic dependence is mostly suppressed in LATIS-like mocks and partially retained in literature-like mocks.