On the dependence of galaxy assembly bias on the selection criteria, number density, and redshift of galaxy samples

TL;DR

This work quantifies galaxy assembly bias (GAB) in IllustrisTNG across galaxy selections, densities, and redshifts, revealing that GAB can modify clustering by up to $\sim$25% and cannot be captured by a single halo property. It decomposes GAB into halo assembly bias and occupancy variation, demonstrating that the interplay between these effects governs the net signal. The authors introduce a fast analytic framework to predict GAB from any halo-property–driven HAB and occupancy variation, validated against shuffling measurements with a high correlation ($r_p \approx 0.8$). These results underscore the necessity of multi-property or occupancy-based modeling for accurate nonlinear clustering predictions in cosmological analyses.

Abstract

One of the key factors influencing galaxy clustering in the nonlinear regime is galaxy assembly bias, which describes the dependence of galaxy clustering on halo properties beyond halo mass. We study this effect by analyzing galaxy samples selected according to stellar mass, luminosity, and broad-band colors from the IllustrisTNG hydrodynamical simulation. We find that galaxy assembly bias depends strongly upon the selection criteria, number density, and redshift of the galaxy sample, with this effect increasing or decreasing galaxy clustering by as much as 25%. Interestingly, no single secondary halo property fully captures the strength of galaxy assembly bias for any galaxy population. Therefore, empirical models predicting galaxy assembly bias as a function of a single halo property cannot reproduce predictions from hydrodynamical simulations. Finally, we investigate how galaxy assembly bias arises from the interplay between halo assembly bias -- the dependence of halo clustering on properties other than halo mass -- and occupancy variation -- the correlation between galaxy occupation and secondary halo properties. We provide a fast analytical expression to predict the level of galaxy assembly bias induced by any halo property in simulated galaxy catalogs without the need for computationally expensive shuffling techniques.

Figures (8)

• Figure 1: Rest-frame color-magnitude diagram of TNG-300 galaxies with $M_{\star}>10^{9}h^{-1}\mathrm{M}_{\odot}$. Blue and red colors display the results for galaxies with specific star formation rates greater and smaller than $\log_{10}({\rm sSFR} [{\rm yr}^{-1}]) = -11$, respectively. Contours denote deciles of the populations, with darker shaded areas indicating the most densely populated regions.
• Figure 2: Halo mass function (left panels) and stellar mass function (right panels) for the stellar mass, $r$-band, blue, and red samples. Each column corresponds to a different sample, while rows represent results at different redshifts. Colored lines show the mass functions for samples with distinct number densities. The black lines in the left panels display the halo mass function for $M_{\mathrm{h}} > 10^{10} \, h^{-1} \, \mathrm{M_\odot}$ halos, while the purple lines in the right panels depict the stellar mass function $M_\star > 10^{9} \, h^{-1} \, \mathrm{M_\odot}$ galaxies.
• Figure 3: Galaxy assembly bias for the stellar mass sample with number density $n = 0.003\ h^{3}\,\mathrm{Mpc}^{-3}$ at $z=0$. The blue symbols show the ratio of the clustering of this sample and of a modified version where galaxies are shuffled among halos of the same mass. The green, orange and red symbols show this ratio for modified versions of this sample where galaxies are shuffled among halos of the same mass and concentration, spin, and formation time, respectively. Horizontal lines display the average ratio on large scales, which corresponds to the level of galaxy assembly bias. Error bars and shaded areas indicate $1\sigma$ uncertainties.
• Figure 4: Measurements of galaxy assembly bias captured by each secondary property from the stellar mass, $r$-band, blue, and red galaxy samples (rows) across different number densities (columns) as a function of redshift. The y-axis displays our measurements of galaxy assembly bias, which we extract by averaging from $8$ to $25 h^{-1}\,\mathrm{Mpc}$ the ratio of galaxy clustering of the sample shown in the legend and that of a modified version where galaxies are shuffled among halos of the same mass. In all panels, for ease of comparison, symbols connected by dashed lines represent results for the stellar mass sample with $n = 0.003\, h^{3}\,\mathrm{Mpc}^{-3}$. The level of galaxy assembly bias varies significantly across samples, redshifts, and number density, increasing or decreasing galaxy clustering by as much as $\simeq24\%$.
• Figure 5: Halo assembly bias for concentration (left), spin (center), and formation time (right). In each panel, blue and green triangles display halo assembly bias for the 30% of halos with the highest and lowest value of the corresponding property, respectively. Solid lines show the best-fitting model to the measurements (see text).