WorksPhysics and Astronomy
Quantifying the Dominance and Emergence of Brightest Cluster Galaxies in SPT and Planck Clusters: A Novel Application of the Gini Index
Abstract
We introduce the Gini index (G) as a metric for studying the emergence of brightest cluster galaxies (BCGs) by quantifying the disparity in the light distribution of galaxy cluster cores. The Gini index collapses the skewness of the bright-end luminosity function to a single statistic, and provides a measure for the light dominance of the BCG over other cluster members that is more robust than the magnitude gap. We argue that the Gini index is far less sensitive to short-term changes in the cluster dynamical state than the magnitude gap and solves its known degeneracies by taking the entire cluster membership into account, making G a more sensitive metric to long-term BCG assembly. We apply the magnitude gap and Gini index to a progenitor-selected sample of 92 South Pole Telescope- and Planck-selected cluster cores at 0 < z < 1 with corresponding Hubble Space Telescope F110W and F200LP and Sloan Digital Sky Survey g and r photometry. Based on both parametric and empirical tests, we find no statistical evidence for evolution in G or over 0 < z < 1, and find that BCGs have not grown significantly (2) faster than the cluster population for z < 1 at 3 confidence. This implies that BCGs achieve dominance and establish defining properties early (z > 1) in massive clusters. In general, we find that the Gini index is a valuable and relatively straightforward measurement, with well-controlled systematics, that could be applied to much larger samples to study BCG dominance and emergence.
Introduction
Galaxy clusters contain a rich array of co-evolving components, including hundreds to thousands of (mostly) red galaxies, a hot intracluster medium (ICM), diffuse intracluster light (ICL), and a massive dark matter halo. The processes driving the coevolution of these components are complicated, but many are connected to the evolution and growth of the brightest cluster galaxy (BCG). BCGs are a special population of galaxies with nearly uniform properties [128, 84, 35, 133, 81, 43, 117, 120]. They are the most massive galaxies in the Universe (), with ancient stellar populations of mean ages Gyr [35, 127]. They are found in the centers of massive () galaxy clusters, and often host radio-bright AGN [133, 12]. One defining property of BCGs is their luminosity dominance over the other cluster members, with magnitudes typically 1–2 mags brighter than the next brightest member [42, 135], and little intrinsic scatter in luminosity across the population [112, 114, 66, 105]. This is a testament to the extreme hierarchical growth experienced by BCGs throughout their histories.
It is still relatively unknown which channels – major or minor mergers, gas rich or poor mergers, cooling flows, or in-situ star formation – contribute most to the assembly of BCG stellar mass at different redshifts. While a subject of some uncertainty [29, 35, 136, 30, 124, 75, 137, 106, 28], it is generally accepted that BCGs experience a factor of 2 increase in stellar mass between and [e.g. 78, 79]. This question has been harder to address at , where representative samples of clusters have been previously limited – though recent studies suggest that massive galaxies and potential BCG progenitors may be assembling and quenched as early as [e.g. 126]. One open question is when the BCG designation even becomes relevant: at what redshift do BCGs emerge as dominant, massive galaxies in their host clusters? BCG emergence is an indication that cluster-specific growth channels are operating efficiently to assemble massive galaxies. The redshift and timescale over which BCGs achieve dominance should help constrain the mechanisms of growth, and may coincide with the establishment of other BCG or cluster properties.
Historically, BCG dominance has been measured using the magnitude gap [e.g. 8], (Equation 1), defined as the difference in absolute magnitude between the BCG and nth brightest cluster galaxy, where typically n = 2, 3, or 4 [70, 34, 58, 52, 53, 54].
For application to stellar masses, we adopt the modified definition of the magnitude gap as the “mass gap” in Equation (2), where is the stellar mass of the nth brightest galaxy. In this analysis, we apply the magnitude gap () to observed clusters and the mass gap () to simulated clusters. The behavior of the indicators is effectively identical, and we will refer to or the “magnitude gap” for simplicity in general discussion.
In isolation, the magnitude gap would be a measure of cluster age, with larger values of correlating with older systems and earlier formation times [41, 33, 34, 58] (A. M. von Benda-Beckmann et al. 2008).
However, the magnitude gap has several known limitations. Its calculation relies on the correct identification of the BCG and other massive galaxies. does not take the relative dominance of the most massive galaxies over M* (the satellite population) into account, leading to degeneracy between relatively less evolved systems with and those with several massive galaxies magnitudes brighter than . As a result, is far more susceptible to changes in cluster dynamical state [120], driven by transient phenomena like cluster mergers. As found by A. A. Dariush et al. [34] and A. M. von Benda-Beckmann et al. (2008), the magnitude gap criterion for identifying fossil groups diminishes within fractions of the Hubble time. The magnitude gap struggles to capture the true dominant end-state of the BCG. A classic example of this is the Coma cluster, which has effectively no magnitude gap despite being highly evolved.
We introduce the Gini index (G), a widely-used indicator of economic uniformity/disparity, as an alternative indicator of BCG dominance over other cluster members. Our analysis is organized as follows. In § 2, we introduce the Gini index, describe its calculation and extrema, and compare it to and for TNG300-1 clusters. We then apply the Gini index and to a sample of 92 SZ-selected cluster cores with optical counterparts between (§ 3) and membership selection described in § 4. Finally, we test the evolution of and (Section 5) and discuss implications for BCG emergence (Section 6).
We use km/s/Mpc and flat CDM cosmology with and , except when working with IllustrisTNG TNG300-1 simulation data in § 2.2, where we use Planck 2015 cosmological parameters for consistency. We use AB magnitudes throughout and the E. E. Salpeter [110] initial mass function (IMF).
The Gini index
Introduced as a sociological indicator by C. Gini [50], the Gini index is a statistic used widely in economics, particularly as a measure of wealth inequality among human populations. In this work, we apply the Gini index to the galaxy population in clusters, where it will serve as a measure of light (or stellar mass) inequality among member galaxies. Just as a low Gini index in economics refers to a population with a relatively uniform distribution of wealth, a low Gini index in this work will refer to a galaxy cluster for which the member galaxies are all similarly luminous. Thus, like the magnitude gap, the Gini index can quantify the dominance of the BCG over the other cluster members. As a point of emphasis, this is not equivalent to previous applications of the Gini index to quantify the morphological smoothness of individual galaxies (e.g. [1, 85, 83, 46, 96]). Nor is this equivalent to applications quantifying the smoothness of the ICM in the X-ray (e.g. [49]).
Mathematically, the Gini index is calculated (bar a normalization factor) by taking twice the difference of two integrated distributions: the observed cumulative distribution of the total luminosity contributed by each galaxy (the Lorenz curve), and the theoretical cumulative distribution when all galaxies are equally luminous. As light is concentrated in fewer, more massive galaxies in a sample, the area under the observed cumulative distribution decreases and the Gini index grows. Another, perhaps more intuitive, way to think about is as a non-parametric measure of the shape or skewness of the luminosity function. For a sample of galaxies with luminosities , the Gini index is calculated by Equation (3). The definition is equivalent for stellar masses :
Typically left out of formalizations for economics, the normalization factor is used when is applied to smaller, finite samples of objects to remove dependence on sample size (e.g. [92]).
We note that the Gini index is insensitive to the choice of magnitude system, as long as it is unbiased. When represents a consistent fraction of for each galaxy, this fraction cancels in Equation (3).
Maximum and Minimum Gini index
The Gini index is a purely positive quantity with a theoretical range between (all of the galaxies have the same luminosity) and (one galaxy has all of the luminosity). Fossil groups ([102, 101, 131, 91, 69, 70]), containing a single, evolved BCG-like galaxy and X-ray emitting ICM in a cluster-sized dark matter halo, are astrophysical systems that approach the idealized limit. On the other hand, the minimum Gini index would be achieved in systems for which all of the member galaxies are of a similar magnitude, such as in a compact group like Stephan’s Quintet ([63, 64, 14, 82]). In reality, the majority of groups and clusters have a mix of bright and faint galaxies, and will lie somewhere in between these extremes.
The Gini index is particularly sensitive to BCG assembly and increases in BCG dominance. This interpretation is most evident in the calculation of the Gini index in Figure 1. When we consider a relatively constant presence of satellites with low probability of merging in massive clusters, the Gini index increases when bright galaxies merge, driving a more sharply-peaked Lorenz curve. Thus, while explicit BCG identification is not required to calculate the Gini index, is theoretically a metric sensitive to how close the cluster is to concentrating the flux of its galaxy population into a single BCG.

Figure 1. An illustration of the calculation of the Gini index. The Lorenz curves for the idealized and cases are plotted, as well as those for two galaxy clusters with (dashes) and (dots). The Gini index is calculated by doubling the area between the 1:1 line of perfect equality and Lorenz curve and then applying the finite sample normalization factor, .
The field population, where in-situ galaxy evolution dominates over merger-based growth, is likely the most uniform population of galaxies available and, thus, likely represents the realistic minimum . The measurement of is only weakly dependent on the total number of galaxies included, but strongly dependent on the limiting magnitude applied to the distribution, as we demonstrate in Figure 2. For the limiting magnitudes we are considering in this work (), assuming a luminosity function measured in the field (M. R. Blanton et al. [15]), we find a median .

Figure 2. The evolution of the Gini index with a varying limiting magnitude () and a varying flat cluster richness () cut, applied here to the M. R. Blanton et al. [15] SDSS r-band Schechter function. Given the application to the luminosity function, this figure displays the distribution of the minimum Gini index, (equivalently ) compared to the limiting magnitude and richness, but the trends apply generally. We show the median with 68% and 95% confidence intervals, calculated by sampling above and measure for 10,000 bootstrapping iterations. When limiting richness, we fix . While the Gini index is largely independent of Schechter model parameters, and thus band and redshift, it is strongly dependent on the limiting magnitude truncating the luminosity function. When is fixed, is approximately flat at statistically-representative samples of . Below , decreases slightly and errors on become larger. We caution against making a flat richness cut (e.g., always calculating using the 20 brightest members) on observed clusters, as this indirectly and imprecisely limits .
As mentioned above, the Gini index scales weakly with the number of galaxies included when is fixed, as shown in Figure 2. is relatively flat with relatively small errors when galaxies are contributing to the calculation. On the other hand, the limiting magnitude (or how many faint galaxies we include) has a strong effect on , as shown in Figure 2, which makes intuitive sense. Thus, we will use a fixed limiting magnitude as a general requirement in subsequent sections. We advise against fixing when calculating for observed clusters, as this imprecisely varies . (see § 6.2 and Appendix § 9). We note that when we write or “richness” here, we are referring to the actual number of galaxies used in the calculation of , not the inherent cluster richness ().
Behavior of the Gini index
To further understand the behavior of the Gini index when compared to the magnitude gap, we consider two simple processes that can alter the values of G, , and in galaxy clusters. One is the merger of galaxies, particularly major mergers that lead to the assembly of the BCG or several BCG-like galaxies. In this scenario, G will generally increase as stellar mass is concentrated into fewer galaxies and the relative brightness disparity amongst members increases. While mergers between bright galaxies can cause individual terms in Equation (3) between bright galaxies to decrease, the corresponding increase in terms between bright and and fainter satellite galaxies prevent large decreases in the index.
is far more sensitive to how mass is concentrated – the merger of two BCGs can drive from a value of to in a dynamical time. Systems with two massive proto-BCGs and less massive satellites can have identical measurements of to clusters with many satellite galaxies and no obvious BCG. This known degeneracy has led to wider adoption of by the community, where this problem is less pronounced due to the likelihood that the th brightest galaxy is more representative of the satellite population ([34, 52, 53, 54]). The Gini index is effectively an integral of the generalized magnitude gap over all .
A second process that can drive the values of G, , and is the mergers of galaxy clusters (A. M. von Benda-Beckmann et al. 2008). Consider the following idealized scenario: a rich cluster merges with another rich cluster of equal mass and an identical population of member galaxies which relax together within the new host cluster potential. In this idealized scenario, the identical cluster merger drives to zero due to the presence of two identical BCGs. will similarly decrease, though not to zero but rather to the pre-merger value of due to the re-indexing of the BCG and fourth brightest galaxy in the cluster post-merger. However, due to its definition, the Gini index remains unchanged before and after the cluster merger. The terms in Equation (3) for duplicated identical galaxies will cancel. For non-identical galaxies, the cumulative sum of remaining terms will double the initial value of the sum, but this factor will cancel out upon the doubling of . Though this scenario is fictional, it illustrates why the Gini index is well-suited for measuring the long-term emergence and assembly of the BCG, since it is the most agnostic to short-lived phenomena (like the cluster’s dynamical state) and the indexing of individual member galaxies.
To further illustrate their behavior, we apply the Gini index and mass gaps (Equation (2)) to the two most massive clusters () in the IllustrisTNG TNG300-1 simulation FoF group catalog at ([95, 98, 121, 94, 93, 87]). These are the only two clusters available in the group catalog with overlapping with our sample selection function in Figure 5. We trace the membership of these clusters iteratively using the merger history of the BCG (via [107] SubLink merger trees), which is identified as the most massive subhalo in groups 0 and 1 at . The BCG is traced to its corresponding main progenitor in previous simulation snapshots ( Myr), which we use to identify the group (cluster) in which it is embedded and the group’s 20 most massive member galaxies at each time-step. Our Clusters 0 and 1 are thus understood as the progenitors of groups 0 and 1 in the redshift zero TNG300-1 group catalog, obtained by following the main progenitor branch of their redshift zero BCGs backwards in time for 10 Gyr.

Figure 5. Samples and selection function. HST-SPT (with uncertainties on ) and SDSS-Planck are plotted over the SPT-SZ 2500d, SPT-ECS, and L. Salvati et al. [111] Planck PSZ2 samples. Open HST-SPT markers represent clusters that were imaged by HST, but not used in the final sample (see further discussion in § 3.2). Our selection function is linear and has a slope of consistent with expected cluster growth rates from simulations (O. Fakhouri et al. [45]). Progenitors of approximately Perseus to Coma mass clusters are selected between two intercepts at : and .
The results of Figure 3 support the results of the above thought experiments. G, , and display an overall positively increasing trend with time, generally expected of cluster and BCG co-assembly ([75, 137]). However, the fractional scatter and its evolution are quite different between G, , and , as shown in Figure 4.

Figure 3. The Gini index and mass gaps and applied to TNG300-1 Clusters 0 and 1 for 10 Gyr of assembly history traced by following the main progenitor branch of the BCG. We select and apply , , and to the 20 brightest galaxies in each cluster at each timestep. The plotted markers are the mean and 1-sigma standard deviations of each indicator in 2 Gyr time bins spanning , with horizontal error bars denoting the width of each bin. Each indicator shows the increasing trend expected from structural assembly and dynamical relaxation through time as BCGs grow more dominant, although is nearly consistent with flat within its mean errors.

Figure 4. Fractional scatter about mean values of G, , and calculated in Figure 3 for TNG300-1 Clusters 0 and 1. is the 1-sigma standard deviation on each indicator, , within a 2 Gyr time period (denoted by the horizontal error bar on each marker). for all times, which agrees with our understanding that sensitivity to the clusters dynamical state and degeneracies due to galaxy indexing drive more scatter in the magnitude gaps than the Gini index.
The fractional scatter (over a 2 Gyr period) on the Gini index is smaller than that of both and for all time. The fractional scatter on is smaller than that of for all time. This agrees with our analytical predictions that is the most sensitive to dynamical phenomena, while G is the least sensitive. The reduction in fractional scatter is one of the primary benefits of using G over . By taking the entire satellite population into account (above some limiting magnitude), G reduces the fractional scatter on by over a factor of 2.5 at each timestep in Figure 4.
Comparing Figure 3 and Figure 4, the intrinsic scatter on is O(1). This means that the predicted evolution of between and of mag is at the same level as the moment-to-moment scatter. Phrased another way, the magnitude gap is too sensitive to the recent dynamical history of the cluster to adequately capture the long-term evolution in this illustrative example. has a nearly unchanging average scatter for all times later than 6 Gyr in the example clusters, likely due to its sensitivity to cluster accretion and galaxy indexing.
Conversely, the scatter in G and decreases nearly monotonically in Figure 4. The errors on the mean values of , in particular, decrease by between as reaches a nearly fixed mean value after 8 Gyr in Figure 3. This is similar to the understood assembly history of BCGs, which should theoretically involve increasing to a large, nearly fixed value as the BCG emerges and dominates the luminosity distribution of its host cluster (e.g. [130]). This behavior indicates that is capable of capturing increases in BCG dominance, given the evolution of at early times caused by BCG growth at a rate faster than its surrounding
cluster. It is possible that evolution in the moment-to-moment scatter in and could also be used to indicate BCG emergence, though a larger dedicated study of simulated clusters would be required to confirm this.
Overall, we argue that short-term dynamical processes and degeneracies due to mass concentration drive scatter in that can fully obscure evolution when applied to two massive TNG300-1 clusters. By taking a larger portion of the satellite population into account, is relatively agnostic to these effects, reducing fractional scatter measured in the more degenerate indicator by a factor . We argue that of the three metrics, from their performance in this illustrative example, G is the best-suited for measuring BCG assembly and emergence.
Sample Selection and Data Analysis
The HST-SPT Cluster Sample
HST-SPT is a sample of 76 SZ-selected, massive galaxy clusters () spanning , selected in an unbiased way for follow-up from the SPT-SZ 2500 degree survey (SPT-SZ 2500d) ([18, 16]) and SPTpol Extended Cluster Survey (SPT-ECS) ([17]). All 76 clusters in this sample have shallow HST UVIS+IR imaging in the F110W and F200LP bands (found in MAST at 10.17909/7x0k-0323). The combined high resolution photometry and progenitor-motivated selection of HST-SPT is well-suited for evolutionary studies of BCGs. The data can also be leveraged for studies of strong lensing statistics and the concentration-mass relation, or studies of AGN activity and feedback (e.g. M. S. Calzadilla et al. [22]).
The initial HST-SPT sample (HST 15307; PI Gladders) was selected through a redshift-dependent mass cut on ( from S. Bocquet et al. [18]), applied to cluster samples from the SPT-SZ 2500d and SPT-ECS surveys. The slope of the mass cut was based on O. Fakhouri et al. [45], which predicts an approximate doubling of from for massive halos (). At high redshifts, this selection function was flattened to account for the paucity of known low-mass, high-redshift clusters at the time. This yielded a sample of hundreds of candidate clusters, from which 122 clusters were ultimately observed as part of the HST-SNAP program.
For this work, we exclude the highest redshift clusters, opting for a cleaner selection function including clusters at with descendant () masses between and . This reduced sample of 76 clusters allows us to study a coevolving cluster population over Gyr, as shown in Figure 5. Further rationale for these more stringent cuts is described in § 3.2.
For the majority of HST-SPT, we use cluster redshifts published in S. Bocquet et al. [18] (for SPT-SZ 2500d clusters) and L. E. Bleem et al. [17] (for SPT-ECS clusters). For a few SPT-SZ 2500d clusters, the S. Bocquet et al. [18] redshift was inconsistent with the observed red sequence (RS) upon inspection. In these cases, we opted to use the L. E. Bleem et al. [16] redshifts instead that preceded the S. Bocquet et al. [18] values. We use the L. E. Bleem et al. [16] values of for SPT-CLJ0643-4535, for SPT-CLJ2108-4445 and for SPT-CLJ2148-4843.
HST Observations
The Gini index and magnitude gap are indices sensitive to the luminosity distribution of bright galaxies. In observing the HST-SPT sample with the Hubble Space Telescope, we opted to maximize the sensitivity and spatial resolution with which the clusters are observed. To reduce exposure times (as required for SNAP programs), while ensuring sensitivity to a wide range of galaxy SEDs in the optical and near-IR, we chose to use the broadest possible filters on WFC3: F200LP (UVIS) and F110W (IR). F200LP spans from the UV to the NIR, while F110W spans the red optical to the bluer NIR, particularly the I and J bands. F200LP provides greater or equal depth to similar filters, F606W and F814W, for all reasonable galaxy SEDs, and does not require post-flash to reduce CTI when orbit packing, maximizing exposure time and observation depth. We additionally configured the orbits to use three different exposure times– short, medium, or long– observing higher redshift clusters for longer times to ensure sufficient depth across the sample. To optimize PSF reconstruction and observational efficiency, we opted for a 3-point line dither in UVIS and a 4-point box dither in the IR. This approach also ensures the effective management of detector defects and cosmic rays.
There is a subset of clusters for which we have HST imaging that are not used in the final HST-SPT sample. As described above, we tightened our selection function from that originally proposed, and thus some observed high- and low-mass clusters have been eliminated from the HST-SPT sample. Additionally, we found that our HST SNAP observations were too shallow to probe deeply into the luminosity function for clusters with , which introduced an unphysical decline in richness () at high-. This was especially concerning given the dependence of on limiting magnitude, as our inability to observe fainter galaxies in these high- clusters imparted an effective magnitude limit different than that of the rest of the sample. Thus, to ensure completeness and statistical comparability, we limited HST-SPT to clusters with , where we can consistently reach depths of . Finally, we removed a selection of clusters from HST-SPT due to foreground contamination in their optical images. SPT-CLJ0641-5950, SPT-CLJ2040-5342, and SPT-CLJ2050-4213 have bright foreground stars or diffraction spikes co-occurrent or obscuring the identified BCG, causing unreliable photometry in the cluster cores.
In total, we observed 122 SPT cluster targets with the HST-SNAP program from 2017 to 2019, 76 of which make up the finalized HST-SPT sample used in this work. On average, our clusters are observed for integration times of s in F110W and s in F200LP. Even at high redshifts with relatively short exposures, our broadband observations are sensitive enough to capture the faint end of the red sequence down to M* + 2 mag (M ) at .
HST Data Reduction
Prior to combining the calibrated single images produced by the MAST archive, we apply two corrections. The first addresses a visible mismatch in sky levels in the F200LP data between the two WFC3-UVIS chips. We rectify this issue by subtracting the background sky on each chip, with the sky estimated by fitting an iterated Gaussian model, sigma-clipped on the high end, to the pixel values. This correction has no effect on photometry.
Secondly, in the IR/F110W data, we correct for the well-known ‘IR blobs’ [100] with additional flat fielding. To do so we acquired several hundred additional F110W images of similar integration time from the archive (from program GO-15163), which we then combined with most of the F110W data described here using IRAF’s imcombine task, tuned to provide robust rejection of the faint wings of object outliers in this large image stack. We then used Galfit [97] to build a model - using a combination of Sérsic [116] profiles - of each identified IR blob across the stacked image. The composite full image of all the fitted blob models is then used to flat-field these features in the F110W data.
Data reduction of the WFC3 imaging with the F200LP and F110W filter - tuned as above - then follows the standard procedures established by STScI, employing the Python3 package Drizzlepac [55, 67]. For astrometric matching of images, sub-exposures taken with the same filter were aligned to the same reference grid, using tweakreg operating on tuned catalogs extracted using Source Extractor [11]; the two filters were also astrometrically matched. These exposures were subsequently combined using astrodrizzle, using a Gaussian kernel, with nominal cosmic ray rejection, a final pixfrac = 1.0 and 0.8 (for the F110W and F200LP, respectively) and pixel scale of /pixel.
GALAPAGOS Modeling
The calculation of the Gini index requires luminosity estimates for each member galaxy, which requires (i) source identification, (ii) source deblending, and (iii) estimating the total luminosity at large radius. This analysis benefits from the high angular resolution and depth of the HST data, which allows individual galaxies to be identified and separated more cleanly.
To convert our HST imaging to photometric catalogs, we utilize the GALAPAGOS-2 software [7, 60, 61]. The finer details of how this software works can be found in the above references, but we will summarize again here. This software package uses Source Extractor [11] to identify and deblend sources, which provides rough estimates of the total luminosity, size, and shape of each source. GALAPAGOS then identifies which sources are in overlapping areas, based on their apparent sizes, and groups all overlapping sources into cutouts to be jointly fit. All sources of similar brightness in a cutout are simultaneously fit using GALFIT [97], beginning with the brightest sources. These models are then held fixed while fainter sources in the cutout are fit, until all sources in a cutout have been modeled. For this analysis, we assume a single Sérsic [116] profile for each galaxy, with initial guesses for each parameter inferred from the values from Source Extractor. For HST-SPT BCGs, the GALFIT F110W Sérsic magnitudes are, on average, 0.98 0.4 mags brighter than those returned by Source Extractor.
We use a single Sérsic profile throughout, rather than a more flexible double Sérsic model. The faintest galaxies at high redshift are well-fit by a single component, but will often result in spurious fits when model flexibility is increased, as the additional degrees of freedom are used to fit nearby galaxies or artifacts. Allowing for additional model flexibility only at low redshift would likely lead to bias (low- galaxies would have a more robust extrapolation of their light profile to infinity) that mimics evolution in G and M. For example, the luminosities of double and single-component BCGs vary by about 0.2 mag systematically [42], which can drive moderate changes in and .
An additional feature of GALAPAGOS-2 is that multi-band photometry is fit simultaneously using GALFITM [61]. Utilizing this feature, we fit data in the F110W and F200LP bandpasses at the same time, requiring that the position, ellipticity, orientation, scale radius, and Sérsic index, n, are constant between the two fits, with only the total luminosity varying. This choice enforces a flat color gradient for each galaxy, but ensures that the significantly-shallower F200LP fits do not have the freedom to settle on anomalous values. We utilize the library of PSFs available from STScI1 for both WFC3-UVIS and WFC3-IR, rotating the model PSF to match our observation before using it as a convolution kernel for the model fitting.
| Sample Name | # Clusters | Redshift Range | Description |
| HST-SNAP | 122 | SPT-SZ 2500d and SPT-ECS sample clusters with HST-SNAP program imaging in F200LP and F110W. | |
| HST-SPT | 76 | HST-SNAP clusters within selection function (Figure 5) and . Three clusters within selection function removed from final sample due to foreground obscuration of the cluster core. | |
| SDSS-Planck | 16 | Planck PSZ2 clusters from L. Salvati et al. [111] within selection function (Figure 5) in SDSS footprint with . Six clusters removed from final sample due to unreliable BCG photometry or issues with optical cluster identification. |
Table 1. **Note—**Summary of cluster samples described in § 3.1 - § 3.5
For each field, we visually inspect the best-fit model image after the fitting procedure is complete. In most cases, we make a second pass through GALAPAGOS after manually masking spurious sources (e.g., diffraction spikes), which considerably improves the fitting process. We repeat this procedure until all galaxies are well-modeled, and there are no obvious artifacts being included in the modeling process.
Ultimately, this process yielded structural parameters for tens of thousands of galaxies. We use the two-band integrated brightnesses from this process as inputs for our color-magnitude plots and for the calculation of G and .
SDSS-Planck Low-Redshift Sample
We use the same selection function applied to HST-SPT to select a low-redshift comparison sample () from the Planck PSZ2 cluster catalog (P. A. R. Ade et al. [2]). While the SPT beam radius of is optimized for SZ cluster detection at high redshift, high pass filtering used in the CMB maps to reduce atmospheric contamination limits the telescope’s sensitivity to large angular scales, and thus low- clusters. Planck’s beam is better suited for cluster detection at low , given the larger angular size of low- clusters on the sky. To fairly compare between SPT and Planck clusters, we adopt the de-biased marginalized masses for Planck PSZ2 from L. Salvati et al. [111], given their agreement with S. Bocquet et al. [18] masses for clusters observed by both SPT and Planck. For the corresponding optical data, we select Planck PSZ2 clusters in the Sloan Digital Sky Survey (SDSS) footprint, bar PSZ2 G127.71-69.55, which
does not have an obvious optical overdensity of galaxies within of its SZ-center.
We note that the SDSS-Planck and HST-SPT samples are well-matched in terms of their relative angular resolution. The SDSS pixel scale is , which corresponds to kpc over the range of the SDSS-Planck sample (). Over the range of HST-SPT (), the WFC3-IR pixel scale of corresponds to kpc.
We query the SDSS Galaxy table (SDSS Collaboration et al. [115]) for all objects within a radius around the published PSZ2 SZ center in the and bands. Previous studies of SDSS galaxies have used combinations of - (e.g. G. Gavazzi et al. [48]) or - (e.g. I. K. Baldry et al. [6]) to capture the 4000 Å break. We find, for our population of quiescent galaxies between , that - produced the tightest observed red sequence and distinguishable blue cloud feature when combined with our red sequence modeling (see § 4.1). This limits contamination when making our membership selection via a color-cut, which we use for SDSS-Planck despite the availability of spectroscopic redshifts to best complement HST-SPT. We use extinction-corrected model-Mags when making our color-cut, due to their optimization for measuring galaxy colors. Extinction-corrected Petrosian magnitudes, which are optimized for accurate total flux measurements, are used for all luminosity-sensitive methods, including cluster center identification (§ 4.3), making limiting magnitude cuts (§ 4.4), and calculating G and . We opt for Petrosian magnitudes over modelMags to avoid photometric issues related to poor surface brightness fitting in crowded and blended cluster cores.
We note that Petrosian magnitudes, though relatively unbiased (M. A. Strauss et al. [125]), display monotonic magnitude deficits for more concentrated galaxies when compared with magnitudes from well-fit Sérsic profiles ( mags dimmer for , for example) [57]. This leads to a moderate possible decrease in , but is a relatively random effect that only systemically affects the SDSS-Planck sample. To understand the impact of the different magnitude systems on our analysis, we artificially brighten the SDSS-Planck Petrosian magnitudes by 0.2 mag to mimic Sérsic magnitudes. The HST-SPT sample is left unchanged and the analysis is repeated. We show in Table 2 and Table 3 that the results of this injection test are consistent within to those from our main sample. We also note that our result is robust when only the unaffected HST-SPT sample is considered in § 5.1 and § 5.2.
| Indicator | All BCGs? | Center Type | |||
| SDSS-Planck + HST-SPT | |||||
| G | Yes | BCG | |||
| G | Yes | Algorithm | |||
| G | No | Algorithm | |||
| Yes | BCG | ||||
| Yes | Algorithm | ||||
| No | Algorithm | ||||
| HST-SPT | |||||
| G | Yes | BCG | |||
| G | Yes | Algorithm | |||
| G | No | Algorithm | |||
| Yes | BCG | ||||
| Yes | Algorithm | ||||
| No | Algorithm | ||||
| SDSS-Planck Injection Test | |||||
| G | Yes | BCG | |||
| G | Yes | Algorithm | |||
| G | No | Algorithm | |||
| Yes | BCG | ||||
| Yes | Algorithm | ||||
| No | Algorithm |
Table 2. Note—Summary of parameters fit to the G and (“Indicator”) data in § 5.1 for the full SDSS-Planck + HST-SPT sample, the HST-SPT sample, and the SDSS-Planck (with magnitude offset) + HST-SPT sample (“SDSS-Planck Injection Test”) discussed in § 3.5. Columns 2-3 specify the choice of BCG or light-weighted (“Algorithm”) cluster centers (“Center Type”) and if non-RS BCGs are included (“Yes”/“No”) in the calculation of G and (“All BCGs?”). The fit slope (), intercept () and intrinsic scatter () are given in columns 4-6 with normally-distributed errors.
| All BCGs? | Center Type | Full Sample | Full Sample | HST-SPT | HST-SPT | High-Low z | High-Low z | Injection Test | Injection Test |
| Yes | BCG | 0.99 | 0.81 | 0.99 | 0.81 | 0.99 | 0.90 | 1.0 | 0.81 |
| Yes | Algorithm | 0.89 | 0.71 | 0.69 | 0.80 | 0.69 | 0.90 | 0.89 | 0.71 |
| No | Algorithm | 0.81 | 0.59 | 0.32 | 0.57 | 0.90 | 0.67 | 0.81 | 0.59 |
Table 3. Empirical Evolution Results
We make a minimal selection of quality cuts based on photometric flags when querying SDSS. We select objects flagged with BINNED1, BINNED2, or BINNED4 detections, but otherwise do not reject objects flagged as SATURATED or with mismatched BLENDED, DE-BLENDED, and CHILD flags, as these cuts were shown to preferentially eliminate bright galaxies in some clusters in early phases of the project. We choose to optimize for completeness of member galaxies detection over pristine member photometry given the sensitivity of G and to the collective luminosity distribution of cluster galaxies. To eliminate foreground stars, we limit SDSS-Planck to clusters at least 20 degrees above the galactic plane. Additionally, we visually inspect each cluster and cut sources corresponding to bright foreground stars and diffraction spikes, along with any galaxies obscured by these features, as we did for the HST-SPT sample.
We removed five clusters from the SDSS-Planck sample at this stage due to photometric issues with their identified BCGs. PSZ2 G033.81+77.18 has a BCG with no photometry in SDSS and PSZ2 G067.52+34.75 has a BCG obscured by a bright foreground star. Additionally, PSZ2 G077.90-26.63, PSZ2 G115.25-72.07, and PSZ2 G217.09+40.15 have BCGs that were treated as point sources when deblended, and are flagged as DE-BLENDED_AS_PSF. In total, we have sufficient data for 16 galaxy clusters in the SDSS-Planck sample (see Figure 5).
Measuring the Gini index
To measure the Gini index of HST-SPT and Planck-SDSS clusters, we need to accomplish the following goals. First, we identify the member galaxies for each cluster. Lacking access to spectroscopic redshift information for individual galaxies, we rely on a red sequence (RS) selection for membership. Next, we identify the positions of the cluster cores about which our measurement will be centered. Finally, we apply cuts on cluster-centric radius and absolute magnitude to produce statistically comparable samples of galaxies across all redshifts. Below, we describe our methods in reference to the HST-SPT sample in detail, but note that we repeat the identical processes for the SDSS-Planck anchor sample, unless explicitly stated otherwise.
Red Sequence Modeling, K-Corrections, and Membership Selection
We use a color cut to constrain the membership of each sample cluster. Cluster galaxies are known to preferentially reside on the red sequence (RS) in color-magnitude space [51]. When modeled linearly, the intercept of the red sequence scales strongly with redshift, while its slope is fixed by the stellar mass – stellar metallicity relationship (stellar MZR) [72]. For our purposes, this method of membership selection is reliable without redshift information for individual galaxies. We work with color offsets, which are the difference between the observed F200LP-F110W color and simulated red sequence colors: . Positive color offsets correspond to galaxies with colors redder than the simulated red sequence, while negative color offsets correspond to observed colors bluer than the simulated red sequence. Color offsets remove redshift dependence from the red sequence feature and allow for red sequence stacking across the entire HST-SPT sample. We find that the stacked red sequence feature is clearly visible even in the broad F110W and F200LP bands (Figure 6). In color offset space, the stacked red sequence feature should flatten if our simulated red sequences sufficiently model the slope of observed red sequences.

Figure 6. The left panel shows the HST-SPT stacked sample plotted in color-offset (observed - simulated red sequence colors) and absolute magnitude space. The right panel shows the FSPS simulated red sequences for the HST-SPT sample. The y-intercepts of the simulated red sequences scales strongly with redshift, while the slope remains relatively constant, fixed by our mass-metallicity relationship (Equation (4)). An example color cut is shown, which uses a red sequence width around the measured sample mean . Despite our use of broad filters, our red-sequence modeling produces a well-defined red sequence and blue cloud for HST-SPT.
We simulate individual cluster red sequences by modeling the colors of typical elliptical galaxies at each cluster redshift. To achieve this aim, we produce simple stellar populations (SSPs) using the Flexible Stellar Population Synthesis library (FSPS) [32, 31, 68]. We assume that the old population is formed in a single burst at with a Salpeter IMF, characteristic of typical S0s for quiescent cluster galaxies [35, 127]. The SSPs are then “observed” at the cluster redshift in the same bandpasses as our observations. This process is repeated for a broad range of possible galaxy stellar masses and thus stellar metallicity inputs, allowing for the measurement of the red sequence over the dynamic range of the observed sample magnitudes. We adopt the linear form of the N. Leethochawalit et al. [77] quiescent-galaxy stellar MZR for metallicity inputs to FSPS to generate a realistic slope for our simulated red sequences. This relationship is consistent with other relationships in the literature over our sample redshift range. When we applied the MZR as published to HST-SPT, our stacked red sequence feature was not flat, indicating that our simulated red sequences were not fully capturing the slope of the observed red sequences. To correct this, we empirically adjusted the slope and intercept of the MZR to flatten the stacked red sequence and better fit our observed data. We note that even with these adjustments, our empirical stellar MZR (Equation (4)) is still consistent with other published studies of quiescent galaxies ([47, 27, 77, 71, 129, 13]). For consistency, we use the adjusted MZR when modeling SDSS-Planck red sequences as well.
Finally, we use our FSPS-modeled SSPs to calculate appropriate K-corrections for the samples, both to account for the reddening of galaxy SEDs within the F110W- and r-bands and their differing bandpasses. We simulate F110W- and r-band magnitudes for our Salpeter-IMF SSP across , calculating the change in apparent magnitude measured as a function of redshift relative to an arbitrary normalization point. These corrections are added in addition to the distance modulus when converting between apparent and absolute magnitudes. We use the resulting rest-frame corrected absolute magnitudes for the remainder of the work, which are now band-agnostic and allow and to be compared directly.
Stacking the color-magnitude diagram for galaxies after these corrections are applied reveals well-defined red sequence and blue cloud features (Figure 6). We fit a sum of three Gaussians to the distribution of HST-SPT color offsets to capture the red sequence, blue cloud, and background components.
The galaxy population shown in Figure 6 contains cluster galaxies, field galaxies, and fore-background contamination. Both the blue cloud and red sequence features are contaminated, to different extent, by interlopers. In § 5.4 we explore the impact of contamination of the red sequence on our measurements of . The blue cloud cluster galaxy population is not relevant to the main goals of this work and thus we do not particularly focus on estimating its contamination by interlopers.
The HST-SPT color offset red sequence has a mean color offset of mag and a width of mag. The width of the HST-SPT RS is primarily driven by our use of broad filters and small differences between the model and true SFH of individual galaxies. When making the color-cut for individual HST-SPT clusters, we allow to re-center within mag of its predicted color if a greater overdensity of galaxies exists within of the shifted mean. This accounts primarily for uncertainties in cluster redshift, which may be a secondary driver of the HST-SPT RS width, and other secondary systematics likely related to our assumed SFHs breaking down at higher sample redshifts. We did not find it necessary to repeat this re-centering of for the SDSS-Planck sample of nearby clusters in the narrow SDSS - and -bands, where modeling the red sequence is more straightforward and redshifts are generally more secure. For the SDSS-Planck sample, we measured mag and a width of mag.
We apply a color cut to the HST-SPT sample in color offset space, where we treat all galaxies in each HST field as if they are located at the cluster redshift. Galaxies with color errors mag and with color offsets that fall within of the re-centered are taken to be cluster members (see § 4.5 for more details about cut ranges). Given our redshift-dependent red sequence modeling, our flat color cut on color-offset is equivalent to making a sloped cut in color-magnitude space, with the slope matching the predicted red sequence slope. The probability of extreme foreground and background contamination of a cluster’s membership is low, given the nearly monotonic variance of the RS with and the narrow fields of view considered ( for WFC3).
BCG Identification
We visually inspect every HST-SPT cluster and identify one BCG. In our high-resolution HST images, we look for the co-occurrence of strong lensing features and a large and bright galaxy or systems of several bright galaxies at the peak of the optical galaxy density. Most of our identified BCGs have colors that are consistent with their cluster’s color cut (72/76 for HST-SPT and 14/16 for SDSS-Planck) and are selected as members regardless. However, we force the small number of BCGs outside the red sequence to be included as members of their respective clusters in some of our calculations of and , as described in subsequent sections.
Four of the added BCGs are those with color-offsets bluer than . BCGs can naturally fall outside the red sequence due to ICM-cooling processes or galaxy-galaxy interactions that incite star formation, or contamination from lensing ([25, 28]). The BCG of SPT-CLJ2250-4808 is the bluest outlier with , but features a clumpy blue morphology characteristic of ongoing star formation. The remaining outlier BCGs with reddened color offsets may suggest small uncertainties in the cluster redshifts.
Cluster Core Identification
We calculate a light-weighted average position to locate each cluster core on the sky, where the density of red-sequence galaxies is the highest. These “centers-of-light” are analogous to calculating the center-of-mass for galaxy luminosities, and are robust against BCG misidentification. Starting at the peak of the SZ detection, we calculate the F110W luminosity-weighted average position of red-sequence member galaxies within a radius circle, assuming that all galaxies that survive the red-sequence cut are located at the cluster redshift.
The radius encompasses most of the WFC3 field of view, and provides a coarse estimate of the center-of-light. We repeat this process iteratively five times, to allow the light-weighted position to be made more accurate through gradient ascent as it algorithmically approaches the peak of the F110W luminosity density. Typically, this coarse-averaging is needed if the center of the SZ detection is significantly offset from the HST pointing.
We iteratively repeat the centering process within and radii to fine-tune the center-of-light. This process typically places the finalized algorithm center within of the identified BCG upon inspection, though this can vary in cases of mergers or when clusters have several bright galaxies. The centers-of-light can be calculated without explicit identification of a BCG, and are one position about which the Gini index is calculated. For comparison, we additionally calculate and about the BCG coordinates — another common identifier of the cluster center ([80, 76, 86, 39]).
We choose a range of physical apertures from around the identified center-of-light to define the cluster core (see § 4.5 for more details about cut ranges). This range is based on the physical field of view of the lowest redshift clusters in our sample, and ensures statistically similar galaxy populations are being selected across the full sample, given the relative universality of the luminosity function with (P. Popesso et al. 2005a,b; A. Moretti et al. 2015; Z. L. Wen & J. L. Han 2015) and the dependence of the quiescent fraction and spatial concentration of galaxies on [56, 62]. This process is repeated in full for the SDSS-Planck low- sample. However, to account for the lower mean redshift and thus larger mean angular size of SDSS-Planck clusters on the sky, we scale the angular radii over which we perform our coarse, intermediate, and fine-tuned averaging by a factor of three. This factor allows averaging to be performed over approximately the same physical areas for HST-SPT and SDSS-Planck clusters.
Limiting Magnitude Cuts
We control for the number of galaxies included in the calculation of the Gini index by either imposing a limiting absolute magnitude or a fixed number of galaxies (via a flat, fixed richness cut discussed in § 2.1, not to be confused with intrinsic physical richness, ) across all redshifts. To determine an appropriate limiting magnitude, we fit a single-component Schechter function to the HST-SPT red sequence. We measure the characteristic magnitude for the HST-SPT red sequence. We apply a limiting magnitude cut of , the lower limit of which is determined by the depth of the HST-SPT cluster observations (see § 4.5 for more details about cut ranges). While the observations of lower redshift clusters reach deeper into the luminosity function, this choice of limiting magnitude ensures a statistically fair comparison across all redshifts.
Error Estimation
Typically, when using a bootstrapping procedure to estimate uncertainties on the Gini index or magnitude gap, we would resample within the uncertainties on the absolute magnitudes used to calculate the indicators. However, the errors on our absolute magnitudes, which are , are not the largest sources of uncertainty on estimates of and for each cluster. Rather, the median values of these indicators are far more influenced by our choices of cuts used to define the population of member galaxies that they are applied to.
We define reasonable ranges through which the membership, size, and limiting magnitude cuts could plausibly vary. Our color cut allows for around each re-centered , our size cut allows for , and our magnitude limit allows for . We calculate or 1000 times for each cluster, allowing each cut quantity to vary randomly within the specified ranges. We then plot the median values and 68% confidence intervals from the distributions of and formed by this modified bootstrapping.
In addition to allowing cut values to vary, we alternate whether all (non-RS+RS) BCGs or just BCGs within our color cut (RS BCGs) are used in the calculation of and (§ 4.2), and whether BCG or algorithm-derived cluster centers are used (§ 4.3). The permutations of these choices make up the columns 1-3 in Figure 7, Figure 8, and Figure 11, and are designed to illustrate the effect of bias from our choice of cuts in our results.

Figure 7. The Gini index (upper) and Magnitude Gap (lower) vs. redshift of the HST-SPT and SDSS-Planck samples for three cluster center types and BCG color-cut permutations: BCG centers with All BCGs (non-RS+RS) (panel 1), light-weighted algorithm centers with all BCGs (non-RS+RS) (panel 2), and light-weighted algorithm centers with only RS BCGs (panel 3). We plot the best fit line (solid) with 16th and 84th percentile error regions (shaded) and fit intrinsic scatter (dashed) for each distribution. The redshift distribution of the Gini index is relatively similar across the cut permutations, with a weakly-decreasing trend consistent with non-evolution within 1. features a fit fractional scatter that is times greater than that of , and is fit with a wider variety of linear profiles that are consistent with non-evolution within 1. Our application of linear regression and the KS-test conclude that and are unevolving for , though the result is noisier.

Figure 8. Cluster richness (N) vs. redshift for HST-SPT and SDSS-Planck. We plot the median richness in the redshift bins in black, with standard errors on the median approximated by . Horizontal error bars are used to represent the width of the redshift bin where the median is calculated. The decrease in N appears to be driven by proportionality with our decreasing progenitor selection and the decreasing red galaxy fraction with cluster redshift (Butcher-Oemler effect), which is most pronounced for . The difference in mean N in the lowest-z bins is driven by data availability and suggests the HST-SPT and SDSS-Planck samples are well-matched. While the declining RS-galaxy fraction causes N to approach values in the bin that drive G to decrease slightly (see Figure 2), this does not drive significant evolution of G in Figure 7.

Figure 11. Reproduction of Figure 7 and Figure 8 for a constant richness cut of to indirectly limit depth. This cut drives notable non-physical scatter in , both for individual clusters and globally with an average increase in fit scatter when compared to that from the limiting magnitude cut. and are still non-evolving within .
The Redshift Evolution of the Gini index and Magnitude Gap
After selecting our populations of core member galaxies, we calculate the Gini index and magnitude gap for each cluster in the HST-SPT and SDSS-Planck samples. We retrieve three iterations of and vs. redshift, based on the variation of center types and included BCGs made in Section 4.5. We present two tests of the redshift evolution of and : one using linear regression and the other using the KS statistic.
Parametric Evidence of non-Evolution
In Figure 7 we show the primary results from this work. At a glance, a few things are obvious from this figure. First, there is no clear evolution in either or over the redshift range spanned by this sample. Second, the Gini index produces considerably less fractional scatter than the magnitude gap, as we predicted above. And, finally, our choices of cluster center and whether or not to include non-RS BCGs do not significantly impact the finding of non-evolution, though they can drive significant scatter in or for individual clusters (as discussed below). To make each of these assertions more quantitative, we turn to parametric modeling of the evolution.
We fit a linear model of the form with intrinsic scatter () to and using the LtsFit package [24]. This method allows for more robust linear regression in cases where scatter, such as that caused by variation in cluster assembly histories, is intrinsic to the distribution. We fit about a pivot of the full sample mean redshift and reject outliers with the FAST Least Trimmed Squares technique [108]. When fitting only the HST-SPT sample, we use . The resulting fit slope, intercept, and intrinsic scatter parameters with normally-distributed errors are reported in Table 2. These fits, and their uncertainties, are shown in Figure 7. As expected when fitting about , the error regions are largest at the highest and lowest sample redshifts and minimized at the pivot redshift.
As an indicator, the Gini index produces far less intrinsic scatter across the full sample than . When applied to HST-SPT and SDSS-Planck, we find median fit values of () and mag (). This is consistent with our understanding of the Gini index as an indicator that is less sensitive to cluster dynamical state and individual galaxy indexing, and more sensitive to the dominance of massive galaxies over .
The sensitivity of to the definition of the BCG and 2nd-brightest member galaxy increases its sensitivity to choices that can redefine them for individual clusters, such as the inclusion of non-RS BCGs, and location and size of the cluster core. thus produces larger and more variant error regions on individual outlier points than across the panels of Figure 7.
The response of between panels is more predictable, such as the decrease in with the removal of non-RS BCGs for a few individual clusters between the 2nd and 3rd panel of Figure 7, due to the more equal resulting luminosity distribution of the remaining RS galaxies when non-RS BCGs are not included in the calculation of . appears to be less sensitive to the choice of cluster center than , likely due to its use of the entire galaxy population in the core. Regardless, the parameters fit to the distributions of and are relatively insensitive to choices made in the analysis, and are consistent across the panels of Figure 7 within .
The results of this fitting procedure are consistent with the conclusion that G and do not evolve between . The slopes fit to G() and in Table 2 are consistent with non-evolution within . This is further highlighted by the fact that the error region (grey shading) on the mean fits is consistent with a horizontal line in Figure 7, further supporting the null evolution hypothesis for and .
Empirical Evidence of non-Evolution
As an independent test of the redshift evolution of and , we apply the two sample Kolmogorov–Smirnov (KS) test [119, 65] to redshift-binned and data. In our evolution-focused context, the rejection of the null hypothesis (p < 0.05) implies the chosen indicator evolves with redshift, as the underlying distribution driving the data changes with redshift. Unlike the parametric test of evolution, where we model and with Gaussian scatter about a linear functional form, the KS test is model independent.
We partition HST-SPT and Planck-SDSS into three sets of sub-samples listed in Table 3 to test for evolution. The “Full Sample” comparison involves splitting the sample in half about , while “HST-SPT” divides and compares only the HST-SPT sample (excluding SDSS-Planck data) about . “High-Low z” compares the lowest quartile () of the combined HST-SPT and Planck-SDSS sample to the highest quartile (). Finally, we apply the KS test to the combined HST-SPT and SDSS-Planck sample with an added magnitude offset (“Injection Test”, see § 3.5), split across . When applying the KS test, we find that no significant evolution in or is detected across all sub-samples.
Richness Evolution & Systematics
Several possible sources of systematic errors exist within the process of obtaining our results. One possible source is our combination of different cluster samples with different observations. Additionally, the membership, size, and luminosity cuts made on the cluster cores can impact the galaxy populations and number of member galaxies probed. We thus examine the distribution of number of member galaxies, , used in our calculations of and for evidence of non-physical evolution driven by underlying systematics.
Generally, is expected to scale with [4, 3, 26]. In the BCG-centered cores, the median in the redshift bins and decreases by a factor of 1.6 from to , similar to HST-SPT’s selected factor of 1.4 decrease in between . Given that we select for membership via a color-cut on the red sequence, this effect is coupled with the Butcher–Oemler effect [21]: a decreasing fraction of red sequence galaxies in rich cluster cores with [e.g. 99]. We hypothesize that the Butcher–Oemler effect leads to the slightly enhanced decline in richness over that expected from the selection function at high-. This effect is most prominent between , where the red sequence is the sparsest.
This effect could also be caused, in part, by increased sensitivity in our membership selection to our assumed SFHs in § 4.1. At higher redshifts, where is closer to , the younger ages of the simulated SSPs used in our red sequence modeling may cause more variation between the observed and simulated red sequences, producing a sparser stacked red sequence feature as these differences drive scatter. We note that the decline in richness for the majority of HST-SPT at does not appear to be driven by our HST observation depth, which caused a discontinuous decrease to for all clusters .
The redshift distribution of cluster richness, , in Figure 8 suggests the SDSS-Planck and HST-SPT samples are well-matched. The median falls slightly for the SDSS-Planck sample from the lowest- HST-SPT value, and scatter towards lower increases when comparing SDSS-Planck to low- HST-SPT clusters. However, this appears to be an effect of the distribution within our selection function, rather than an effect related to our K-corrections or de-biasing between the SZ surveys. In Figure 5, SDSS-Planck clusters are more concentrated near the lower threshold than HST-SPT clusters within , accounting for the lower reached by SDSS-Planck in Figure 8. We attempted to incorporate more massive Planck clusters in the sample near the upper threshold, but found no overlap within SDSS or DES for this mass range.
The evolution of richness across our sample appears to be driven by understood phenomena. On average, over the full sample, is above the threshold at which decreases slightly with richness (Figure 2), limiting the ability of richness to drive evolution in or its scatter. In the bin, where approaches the sensitivity threshold for due to the lower fraction of RS-galaxies in cluster cores, the expected decrease in ( from a drop in in Figure 2) from this effect does not produce significant non-physical evolution in our result. That is, a lower number of galaxies in the highest-redshift bin ought to lead to stronger evolution in , which we do not observe. Thus, we expect these results to be robust to any evolution in the red sequence richness over our sample.
**Note—**Summary of p-values returned from the application of the KS-test to G () and () data in § 5.2. Columns 1-2 specify the choice of BCG or light-weighted (“Algorithm”) cluster centers (“Center Type”) and if non-RS BCGs are included (“Yes”/“No”) in the calculation of G and (“All BCGs?”). The KS-test is applied to the distributions of four sub-samples: “Full Sample” (Column 3) compares the combined HST-SPT and SDSS-Planck sample split across , “HST-SPT” (Column 4) compares the HST-SPT sample split across , “High-Low z” (Column 5) compares the lowest () and highest () z-quartiles of the combined HST-SPT and SDSS-Planck sample, and “Injection Test” (Column 6) compares the combined HST-SPT and SDSS-Planck sample (with magnitude offset, discussed in § 3.5) split across .
Robustness of Results Against Membership Contamination
We discuss in § 4.1 that our membership selection is not optimized to eliminate contamination from interlopers. For simplicity, we treat foreground and background interloper galaxies, as well as galaxies outside of clusters at , as field galaxies. As we show in Figure 10, the populations of bright galaxies in the field and cluster cores produce significantly different values of G. Additionally, as discussed in § 5.3, there is evidence that our stacked red sequence feature becomes sparser at higher redshifts, particularly due to the inherent decreased fraction of red sequence galaxies at high-z, and the possible breakdown of assumptions folded into our membership selection. We explore the impact of field contamination on our membership selection further, given the ability of field galaxies to significantly decrease G, particularly at high-z where the red sequence signal to noise is weaker.

Figure 10. The Gini index of cluster cores vs. the galaxy field. We reproduce panel 1 of Figure 7, where is calculated for red sequence galaxies in BCG-centered cores of the HST-SPT and SDSS-Planck samples with all BCGs (non-RS+RS) included. We additionally calculate the median Gini index (blue lines) with 16th and 84th percentile C.I.s for several published luminosity functions (M. R. Blanton et al. [15] r-band, E. F. Bell et al. [9] r-band, and S. M. Faber et al. [44] for all galaxies), representing the full galaxy field. We sample each luminosity function above the median cut applied to our observed data (), and find that and thus the shape of the full luminosity function does not evolve strongly over . and is non-evolving for the cluster cores across , implying that BCGs are dominant and growing with the red sequence for .
To quantify the impact of field contamination on G, we draw samples of galaxies from two distinct field and cluster distributions. We form the cluster distribution by combining K-corrected absolute magnitudes from SDSS-Planck and HST-SPT galaxies that survive the most stringent color (width), median size () and magnitude () cuts implemented in § 4.5. “Field” samples are drawn from the M. R. Blanton et al. [15] luminosity function. We draw samples of varying percentages of field and cluster galaxies and calculate G for each, forming the black line in Figure 9.

Figure 9. Gini index vs. Fraction of Contamination (), where G (Median – black line, shading – 68% and 95% C.I.s after 10,000 bootstrapping iterations) is calculated with its 1 and 2 errors for samples of galaxies drawn from a “cluster” distribution of bright, red sequence core galaxies from HST-SPT and Planck-SDSS and a “field” distribution given by the M. R. Blanton et al. [15] luminosity function. Here, corresponds to the value of G calculated for a purely “cluster”-drawn sample, and corresponds to the value of G calculated for a purely “field”-drawn sample. The inset demonstrates the three-component Gaussian fitting of the color offset distribution used to estimate for the full HST-SPT sample, as well as HST-SPT split across three z-bins: , , . While our measured is relatively high and increases with redshift, G is robust at the levels of contamination estimated in our cluster membership selection.
We then estimate the fraction of field and blue cloud galaxies selected by our red sequence cut. We fit a three-component Gaussian profile to the distribution of color offsets for the full HST-SPT sample, as well as the , , and redshift bins separately. We show how the distribution of color offsets for the full HST-SPT sample was fit for the red sequence, blue cloud, and field contamination in the inset panel of Figure 9. The contamination fraction is given by the ratio of the areas under the field and blue cloud distributions to the area under the total color-offset distribution within the red sequence cut. We assume that the contamination fraction will be highest for the HST-SPT sample, given our use of broad bands anderiwa the much higher cluster redshifts spanned. Thus, we do not repeat this process for our low- SDSS-Planck anchor sample.
We find that the contamination fraction is relatively high, with a value of for the full sample. The contamination fraction increases with redshift (: , : , : ), which we expect as the red sequence feature becomes more sparse at high-. However, the high contamination fraction for HST-SPT does not significantly impact , with only a 6% decrease in when compared to the non-contaminated case in the most affected bin (), which falls within the fractional scatter fit in § 5.1. While our method of membership selection allows for high levels of possible field contamination, we show that our calculation of is relatively robust against this effect. This is likely caused by the relatively lower of the field – most interloper galaxies sampled are likely similar in brightness to low-mass satellites, which do not significantly drive .
Discussion
Implications of a Non-evolving Gini index
Our parametric and empirical tests of evolution in § 5 suggest that the Gini index and magnitude gap are non-evolving in the regime. While transient, dynamical phenomena can drive significant scatter in , the Gini index is more sensitive to massive galaxy assembly and the dominance of the BCG over the general cluster population.
Our primary results have two major implications. One is that the luminosity distribution of red sequence galaxies in cluster cores is more skewed towards higher luminosities than the full luminosity function of the field. In Figure 10, we compare the Gini index for our BCG-centered cores (panel 1 of Figure 7) with the Gini index calculated from large field surveys based on the published luminosity functions ([15, 9, 44]), assuming a consistent limiting magnitude. The Gini index of our sample clusters is a factor of two larger than the field Gini index for . At in the cluster core, this is driven by the presence of the BCG and other massive galaxies – which are not present in the field – that skew the luminosity distribution of red sequence galaxies in cluster cores brighter. This result is consistent with studies of the cluster luminosity function that show either an increase in or require a second component at the bright end, due to the excess of bright, red galaxies in clusters over the field ([38, 134, 89, 74]).
The increased fraction of faint galaxies in clusters over the field ([74]) should also drive to increase, though this population () is not targeted in our analysis due to our conservative choice of limiting magnitude. The enhanced Gini index in clusters is a testament to the unique growth channels of massive galaxies and the environmental quenching of faint galaxies in the cluster environment that produce a higher ratio of faint to bright galaxies than the field. is relatively unevolving over , which additionally implies that the shape of the full field luminosity function at the bright end does not evolve strongly during this epoch.
The second major result is that our measurement of non-evolution in over is consistent with BCGs emerging early in their host clusters, before , as BCGs transitioning from non-dominant to dominant, massive galaxies should result in the general increase of as mergers of bright () galaxies in the core concentrate light into fewer members. The non-evolution of implies that the concentration of light is not drastically changing over . Clusters are not drastically closer to concentrating their light into fewer galaxies (per our interpretation of in § 2.1) through mechanisms related to the assembly of BCGs. The growth channels that concentrate light into the (eventual) BCG and skew the bright end of the cluster luminosity function must be operating before , consistent with the interpretation that BCGs grow slowly and steadily with their host clusters over the past Gyr.
The narrative that the BCG is established early is consistent with several other studies of its defining properties, ([5, 35, 122, 113, 106, 28, 23, 40, 130]), and other studies of massive galaxy assembly in the cluster environment ([71]).
One interesting caveat of this result is that we know that BCGs are actively growing during this period, nearly doubling in stellar mass between ([78]). Thus, across our sample, BCGs are actively accreting stellar mass, and it might be reasonable to expect that this growth alone should produce an evolutionary trend in between the highest and lowest redshift clusters. We explore the effect of BCG growth on during this epoch with a thought experiment conducted with our TNG300-1 clusters.
Doubling the stellar mass of the BCG at the 6 Gyr timestep to estimate the C. Lidman et al. [78] predicted growth leads to a modest increase of at when the satellite population is left unchanged. However, coincides with enrichment of the red sequence (N. Martinet et al. [88]), with studies predicting a doubling of the RS stellar mass during this period, driven by satellite galaxies (E. F. Bell et al. [10]; G. De Lucia et al. [36, 37]; M. J. I. Brown et al. [19]; S. M. Faber et al. [44]; M. J. I. Brown et al. [20]; G. Rudnick et al. [109]; J. P. Stott et al. [123]). Simply doubling the number of satellite galaxies in TNG300-1 clusters to mimic a richer RS above with the predicted BCG growth rates leads to at , which does not account for the mass growth of individual RS galaxies that would further decrease . Thus, the non-evolution of does not imply that the BCG and red sequence are stagnant over . Rather, our result is consistent with the interpretation that the growth of the BCG and enrichment and growth of the red sequence occur together at similar rates. The early assembly of satellite galaxies may further support shared histories of growth between the red sequence and BCGs (G. Khullar et al. [71]). Both a stagnant and co-growing BCG+red sequence population can lead to non-evolution in over at the level detected by our analysis, supporting the interpretation that relative BCG dominance is not changing over , and was established before this epoch.
Do we ever expect to observe evolution in BCG dominance through ? At some point, the cluster population may resemble the field population more closely, which would require the two tracks in Figure 10 converging. However, it has been shown that an excess of bright galaxies exists in protoclusters when compared to the field [59, 73, 118], that likely would increase above . The answer likely lies at high redshifts, possibly up to the era of protoclusters. Quantifying BCG emergence may require determining a more realistic for clusters, given evidence of massive galaxy assembly before virialization. Future studies at high- will require a more careful treatment of cluster membership, as the red sequence continues to become less populated at , to avoid biasing the Gini index low and artificially recovering evolution. Regardless, our analysis, particularly our parametric test of evolution, should be sensitive to a slope from evolution of within (via the slope uncertainties in Table 2).
Application and Benefits of the Gini index
The Gini index provides a non-parametric measure of the light dominance of the BCG over the satellite population in galaxy clusters. More broadly, the Gini index provides a model-independent probe of the shape of the luminosity function, by essentially quantifying the bright-end skewness of the underlying luminosity function in a single parameter. Comparing measurements of allows for a study of the deviation of the luminosity function from a chosen standard, without any consideration of functional form. By applying different magnitude limits and membership selection, one could imagine repurposing the Gini index to quantify the evolution in the faint upturn of the luminosity function, without assuming an underlying functional form.
When applying the Gini index to our sample, we found it necessary to apply K-corrections to account for SED-aging across and band differences between HST-SPT and SDSS-Planck when calculating absolute magnitudes. It was most straightforward to work with fully K-, band-, and distance-corrected absolute magnitudes when designing and applying cuts to our samples. When comparing across different works, such corrections, as well as matching magnitude and radial cuts, are critical to facilitate a statistically fair comparison.
We strongly recommend making cuts on absolute magnitude as opposed to fixing the number of included sources via some “brightest galaxies” approach, given the insensitivity of to (see Figure 2). Forcing a fixed number of galaxies to be included in the calculation across a sample effectively applies a varying magnitude cut to each cluster given by the magnitude of the th brightest galaxy, driving non-physical scatter in (for a visualization of this effect, please see Appendix § 9, where we reproduce Figure 7 and Figure 8 for a flat richness cut).
Conclusion
We present the Gini index as an alternate indicator of brightest cluster galaxy dominance and emergence in galaxy clusters. The Gini index is less sensitive to the cluster dynamical state and member galaxy indexing than the magnitude gap, and accounts for the relative brightness of massive galaxies over M*, providing a more complete picture of structural assembly with fewer degeneracies. When applied to observed cluster cores, produces times less fractional scatter than the more commonly used “magnitude gap”, .
We calculate the Gini index, , for each cluster in a sample of 92 Planck- and SPT-selected galaxy clusters, lying on a common evolutionary track spanning to . These estimates are based on optical data from SDSS (low-) and HST (high-). Over the redshift range , we find no statistically significant evidence for evolution in or . In addition, we find that (red sequence, cluster core) (all galaxies) for all . This has two major implications: (i) We find no evidence for evolution in , and rule out that BCGs have grown significantly () faster than the cluster population between and at confidence. Our result is consistent with BCGs in massive clusters emerging from the galaxy luminosity distribution of their host clusters early, establishing their relative dominance over red sequence satellite galaxies before . This means that the use of BCGs as proxies for massive host cluster properties up to should be robust; and (ii) We find no evidence for evolution in during a known period of BCG stellar mass growth. The enrichment of the red sequence and mass growth of red sequence galaxies can suppress evolution in if they occur at a comparable rate to the growth of the BCG. In other words, given the measured constraints on the evolution of , we can conclude that BCGs established their dominance at . We find no evidence that they have been growing faster than other member galaxies since .
The Gini index generally probes deviations in the shape of the luminosity function, allowing for versatile application to different scientific contexts. is highly sensitive to the choice of limiting magnitude when considering a galaxy population, and is not easily comparable between studies with different choices unless the underlying shape of the luminosity function is fully known. This non-parametric approach to quantifying BCG evolution can be easily and rapidly applied to large samples. Additionally, the calculation of does not require the identification of a BCG, and we show that our results are consistent across permutations of BCG-dependent or -independent cut choices. G is less sensitive than the often-used magnitude gap to short-term evolution (e.g., cluster-cluster mergers) and is more sensitive to the long term assembly history of the galaxy population.
Acknowledgments
This research is based on observations made with the NASA/ESA Hubble Space Telescope obtained from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5–26555. Support for US investigators in program HST-GO-15307 was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. K. Fosdick and M. McDonald also acknowledge support from the National Science Foundation via AAG grant number AST-2510234. Argonne National Laboratory’s work was supported by the U.S. Department of Energy, Office of High Energy Physics, under contract DE-AC02-06CH11357. For the purpose of open access, the authors have applied a Creative Commons Attribution (CC BY) license to any Author Accepted Manuscript version arising.
Flat Richness Cuts
We advise against making a flat richness cut (like that discussed in § 2.1 as an indirect method of indirectly limiting depth) when applying to observed galaxy clusters, and provide this appendix as an illustration of this recommendation. As discussed in sections 2.1 and 6.2, the Gini index is sensitive to which limiting magnitude the luminosity function is truncated. Applying a flat richness cut of to our sample of clusters, such as that shown in row 3 of Figure 11, is effectively equivalent to limiting depth with individual values of determined by the magnitude of the th brightest galaxy. This cut drives notable non-physical scatter in , with an average increase in fit scatter and larger variation in for individual clusters. Here, and are still non-evolving within .
| Indicator | All BCGs? | Center Type | m | b | |
| Yes | BCG | ||||
| Yes | Algorithm | ||||
| No | Algorithm | ||||
| Yes | BCG | ||||
| Yes | Algorithm | ||||
| No | Algorithm |
Table 4. Limiting Richness Parametric Evolution Results
**Note—**Summary of parameters fit to the and (“Indicator”) data for the flat richness cut. Columns 2-3 specify the choice of BCG or light-weighted (“Algorithm”) cluster centers (“Center Type”) and if non-RS BCGs are included (“Yes”/“No”) in the calculation of and (“All BCGs?”). The fit slope (), intercept () and intrinsic scatter () are given in columns 4-6 with normally-distributed errors.
| All BCGs? | Center Type | Full Sample | HST-SPT | High-Low z | |||
| All BCGs? | Center Type | ||||||
| Yes | BCG | 0.40 | 0.67 | 0.46 | 0.91 | 0.14 | 0.99 |
| Yes | Algorithm | 0.54 | 0.71 | 0.57 | 0.80 | 0.26 | 0.90 |
| No | Algorithm | 0.50 | 0.59 | 0.83 | 0.57 | 0.14 | 0.69 |
Table 5. Limiting Richness Empirical Evolution Results. **Note—**Summary of p-values returned from the application of the KS-test to G () and M () data for the flat richness cut. Columns 1-2 specify the choice of BCG or light-weighted (“Algorithm”) cluster centers (“Center Type”) and if non-RS BCGs are included (“Yes”/“No”) in the calculation of G and M (“All BCGs?”). The KS-test is applied to the distributions of three sub-samples: “Full Sample” (Column 3) compares the combined HST-SPT and SDSS-Planck sample split across , “HST-SPT” (Column 4) compares the HST-SPT sample split across , and “High-Low z” (Column 5) compares the lowest () and highest () z-quartiles of the combined HST-SPT and SDSS-Planck sample.
References
- [1]Abraham, R. G., van den Bergh, S., & Nair, P. 2003, The Astrophysical Journal, 588, 218–229, doi: 10.1086/373919DOI
- [2]Ade, P. A. R., Aghanim, N., Arnaud, M., et al. 2016, Astronomy & Astrophysics, 594, A27, doi: 10.1051/0004-6361/201525823DOI
- [3]Andreon, S., & Bergé, J. 2012, A&A, 547, A117, doi: 10.1051/0004-6361/201220115DOI
- [4]Andreon, S., & Hurn, M. A. 2010, Monthly Notices of the Royal Astronomical Society, 404, 1922, doi: 10.1111/j.1365-2966.2010.16406.xDOI
- [5]Aragon-Salamanca, A., Baugh, C. M., & Kauffmann, G. 1998, Monthly Notices of the Royal Astronomical Society, 297, 427–434, doi: 10.1046/j.1365-8711.1998.01495.xDOI
- [6]Baldry, I. K., Glazebrook, K., Brinkmann, J., et al. 2004, The Astrophysical Journal, 600, 681–694, doi: 10.1086/380092DOI
- [7]Barden, M., Häußler, B., Peng, C. Y., McIntosh, D. H., & Guo, Y. 2012, MNRAS, 422, 449, doi: 10.1111/j.1365-2966.2012.20619.xDOI
- [8]Bautz, L. P., & Morgan, W. W. 1970, ApJL, 162, L149, doi: 10.1086/180643DOI
- [9]Bell, E. F., McIntosh, D. H., Katz, N., & Weinberg, M. D. 2003, The Astrophysical Journal Supplement Series, 149, 289–312, doi: 10.1086/378847DOI
- [10]Bell, E. F., Wolf, C., Meisenheimer, K., et al. 2004, ApJ, 608, 752, doi: 10.1086/420778DOI
- [11]Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393, doi: 10.1051/aas:1996164DOI
- [12]Best, P. N., von der Linden, A., Kauffmann, G., Heckman, T. M., & Kaiser, C. R. 2007, MNRAS, 379, 894, doi: 10.1111/j.1365-2966.2007.11937.xDOI
- [13]Bevacqua, D., Saracco, P., Boecker, A., et al. 2024, Astronomy & Astrophysics, 690, A150, doi: 10.1051/0004-6361/202348831
- [14]Bitsakis, T., Charmandaris, V., da Cunha, E., et al. 2011, A&A, 533, A142, doi: 10.1051/0004-6361/201117355DOI
- [15]Blanton, M. R., Hogg, D. W., Bahcall, N. A., et al. 2003, The Astrophysical Journal, 592, 819–838, doi: 10.1086/375776DOI
- [16]Bleem, L. E., Stalder, B., de Haan, T., et al. 2015, The Astrophysical Journal Supplement Series, 216, 27, doi: 10.1088/0067-0049/216/2/27DOI
- [17]Bleem, L. E., Bocquet, S., Stalder, B., et al. 2020, The Astrophysical Journal Supplement Series, 247, 25, doi: 10.3847/1538-4365/ab6993DOI
- [18]Bocquet, S., Dietrich, J. P., Schrabbach, T., et al. 2019, The Astrophysical Journal, 878, 55, doi: 10.3847/1538-4357/abf10
- [19]Brown, M. J. I., Dey, A., Jannuzi, B. T., et al. 2007, The Astrophysical Journal, 654, 858–877, doi: 10.1086/509652DOI
- [20]Brown, M. J. I., Zheng, Z., White, M., et al. 2008, ApJ, 682, 937, doi: 10.1086/589538DOI
- [21]Butcher, H., & Oemler, Jr., A. 1984, ApJ, 285, 426, doi: 10.1086/162519DOI
- [22]Calzadilla, M. S., Bleem, L. E., McDonald, M., et al. 2023, The Astrophysical Journal, 947, 44, doi: 10.3847/1538-4357/acc6c2DOI
- [23]Calzadilla, M. S., McDonald, M., Benson, B. A., et al. 2024, ApJ, 976, 169, doi: 10.3847/1538-4357/ad8916DOI
- [24]Cappellari, M., Scott, N., Alatalo, K., et al. 2013, Monthly Notices of the Royal Astronomical Society, 432, 1709–1741, doi: 10.1093/mnras/stt562DOI
- [25]Cerulo, P., Orellana, G. A., & Covone, G. 2019, MNRAS, 487, 3759, doi: 10.1093/mnras/stz1495DOI
- [26]Chiu, I.-N., Ghirardini, V., Liu, A., et al. 2022, A&A, 661, A11, doi: 10.1051/0004-6361/202141755DOI
- [27]Choi, J., Conroy, C., Moustakas, J., et al. 2014, The Astrophysical Journal, 792, 95, doi: 10.1088/0004-637X/792/2/95DOI
- [28]Chu, A., Durret, F., & Márquez, I. 2021, A&A, 649, A42, doi: 10.1051/0004-6361/202040245DOI
- [29]Collins, C. A., & Mann, R. G. 1998, MNRAS, 297, 128, doi: 10.1046/j.1365-8711.1998.01482.xDOI
- [30]Collins, C. A., Stott, J. P., Hilton, M., et al. 2009, Nature, 458, 603–606, doi: 10.1038/nature07865DOI
- [31]Conroy, C., & Gunn, J. E. 2010, ApJ, 712, 833, doi: 10.1088/0004-637X/712/2/833DOI
- [32]Conroy, C., Gunn, J. E., & White, M. 2009, ApJ, 699, 486, doi: 10.1088/0004-637X/699/1/486DOI
- [33]Dariush, A., Khosroshahi, H. G., Ponman, T. J., et al. 2007, MNRAS, 382, 433, doi: 10.1111/j.1365-2966.2007.12385.xDOI
- [34]Dariush, A. A., Raychaudhury, S., Ponman, T. J., et al. 2010, Monthly Notices of the Royal Astronomical Society, doi: 10.1111/j.1365-2966.2010.16569.xDOI
- [35]De Lucia, G., & Blaizot, J. 2007, Monthly Notices of the Royal Astronomical Society, 375, 2–14, doi: 10.1111/j.1365-2966.2006.11287.xDOI
- [36]De Lucia, G., Pogg ianti, B. M., Aragón-Salamanca, A., et al. 2004, Proceedings of the International Astronomical Union, 2004, doi: 10.1017/S1743921304001024DOI
- [37]De Lucia, G., Poggianti, B. M., Aragon-Salamanca, A., et al. 2007, Monthly Notices of the Royal Astronomical Society, 374, 809–822, doi: 10.1111/j.1365-2966.2006.11199.xDOI
- [38]De Propris, R., Col less, M., Driver, S. P., et al. 2003, Monthly Notices of the Royal Astronomical Society, 342, 725–737, doi: 10.1046/j.1365-8711.2003.06510.xDOI
- [39]De Propris, R., West, M. J., Andrade-Santos, F., et al. 2021, Monthly Notices of the Royal Astronomical Society, 500, 310, doi: 10.1093/mnras/staa3286DOI
- [40]Ding, J., Dalal, R., Sunayama, T., et al. 2024, Monthly Notices of the Royal Astronomical Society, 536, 572–591, doi: 10.1093/mnras/stae2601DOI
- [41]D’Onghia, E., Sommer-Larsen, J., Romeo, A. D., et al. 2005, ApJL, 630, L109, doi: 10.1086/491651DOI
- [42]Donzelli, C. J., Muriel, H., & Madrid, J. P. 2011, The Astrophysical Journal Supplement Series, 195, 15, doi: 10.1088/0067-0049/195/2/15DOI
- [43]Díaz-Giménez, E., Mamon, G. A., Pacheco, M., Mendes de Oliveira, C., & Alonso, M. V. 2012, Monthly Notices of the Royal Astronomical Society, 426, 296, doi: 10.1111/j.1365-2966.2012.21705.x
- [44]Faber, S. M., Willmer, C. N. A., Wolf, C., et al. 2007, The Astrophysical Journal, 665, 265–294, doi: 10.1086/519294DOI
- [45]Fakhouri, O., Ma, C.-P., & Boylan-Kolchin, M. 2010, Monthly Notices of the Royal Astronomical Society, 406, 2267–2278, doi: 10.1111/j.1365-2966.2010.16859.xDOI
- [46]Florian, M. K., Li, N., & Gladders, M. D. 2016, The Astrophysical Journal, 832, 168
- [47]Gallazzi, A., Bell, E. F., Zibetti, S., Brinchmann, J., & Kelson, D. D. 2014, The Astrophysical Journal, 788, 72, doi: 10.1088/0004-637X/788/1/72DOI
- [48]Gavazzi, G., Fumagalli, M., Cattaneo, A., & Boselli, A. 2010, Astronomy and Astrophysics, 517, A73, doi: 10.1051/0004-6361/201014153DOI
- [49]Ghirardini, V., Bahar, Y. E., Bulbul, E., et al. 2022, A&A, 661, A12, doi: 10.1051/0004-6361/202141639DOI
- [50]Gini, C. 1912, Variabilità e mutabilità: contributo allo studio delle distribuzioni e delle relazioni statistiche. [Fasc. I.], Studi economico-giuridici pubblicati per cura della facoltà di Giurisprudenza della R. Università di Cagliari (Tipogr. di P. Cuppini). https://books.google.com/books?id=fqjaBPMxB9kC
- [51]Gladders, M. D., & Yee, H. K. C. 2000, AJ, 120, 2148, doi: 10.1086/301557DOI
- [52]Golden-Marx, J. B., & Miller, C. J. 2018, The Astrophysical Journal, 860, 2, doi: 10.3847/1538-4357/aac2bd
- [53]Golden-Marx, J. B., Miller, C. J., Zhang, Y., et al. 2022, The Astrophysical Journal, 928, 28, doi: 10.3847/1538-4357/ac4cb4
- [54]Golden-Marx, J. B., Zhang, Y., Ogando, R. L. C., et al. 2025, MNRAS, 538, 622, doi: 10.1093/mnras/staf277
- [55]Gonzaga, S., Hack, W., Fruchter, A., & Mack, J. 2012, The DrizzlePac Handbook (Space Telescope Science Institute)
- [56]Goto, T., Yagi, M., Tanaka, M., & Okamura, S. 2004, MNRAS, 348, 515, doi: 10.1111/j.1365-2966.2004.07324.xDOI
- [57]Graham, A. W., Driver, S. P., Petrosian, V., et al. 2005, AJ, 130, 1535, doi: 10.1086/444475DOI
- [58]Harrison, C. D., Miller, C. J., Richards, J. W., et al. 2012, ApJ, 752, 12, doi: 10.1088/0004-637X/752/1/12DOI
- [59]Hatch, N. A., Overzier, R. A., Kurk, J. D., et al. 2009, Monthly Notices of the Royal Astronomical Society, 395, 114–125, doi: 10.1111/j.1365-2966.2009.14525.xDOI
- [60]Häußler, B., Bamford, S. P., Vika, M., et al. 2013, MNRAS, 430, 330, doi: 10.1093/mnras/sts633
- [61]Häußler, B., Vika, M., Bamford, S. P., et al. 2022, A&A, 664, A92, doi: 10.1051/0004-6361/202142935
- [62]Hennig, C., Mohr, J. J., Zenteno, A., et al. 2017, Monthly Notices of the Royal Astronomical Society, stx175, doi: 10.1093/mnras/stx175DOI
- [63]Hickson, P. 1982, ApJ, 255, 382, doi: 10.1086/159838DOI
- [64]Hickson, P., Kindl, E., & Auman, J. R. 1989, ApJS, 70, 687, doi: 10.1086/191354DOI
- [65]Hodges, J. L. 1958, Arkiv for Matematik, 3, 469, doi: 10.1007/BF02589501DOI
- [66]Hoessel, J. G., & Schneider, D. P. 1985, AJ, 90, 1648, doi: 10.1086/113871DOI
- [67]Hoffmann, S. L., Mack, J., Avila, R., et al. 2021, in American Astronomical Society Meeting Abstracts, Vol. 53, American Astronomical Society Meeting Abstracts, 216.02
- [68]Johnson, B., Foreman-Mackey, D., Sick, J., et al. 2024,, v0.4.7 Zenodo, doi: 10.5281/zenodo.12447779
- [69]Jones, L. R., Ponman, T. J., & Forbes, D. A. 2000, MNRAS, 312, 139, doi: 10.1046/j.1365-8711.2000.03118.xDOI
- [70]Jones, L. R., Ponman, T. J., Horton, A., et al. 2003, MNRAS, 343, 627, doi: 10.1046/j.1365-8711.2003.06702.xDOI
- [71]Khullar, G., Bayliss, M. B., Gladders, M. D., et al. 2022, The Astrophysical Journal, 934, 177, doi: 10.3847/1538-4357/ac7c0c
- [72]Kodama, T., & Arimoto, N. 1997, A&A, 320, 41, doi: 10.48550/arXiv.astro-ph/9609160DOI
- [73]Koyama, Y., Kodama, T., Tadaki, K.-i., et al. 2012, Monthly Notices of the Royal Astronomical Society, 428, 1551–1564, doi: 10.1093/mnras/sts133DOI
- [74]Lan, T.-W., Ménard, B., & Mo, H. 2016, Monthly Notices of the Royal Astronomical Society, 459, 3998–4019, doi: 10.1093/mnras/stw898DOI
- [75]Laporte, C. F. P., White, S. D. M., Naab, T., & Gao, L. 2013, MNRAS, 435, 901, doi: 10.1093/mnras/stt912DOI
- [76]Lauer, T. R., Postman, M., Strauss, M. A., Graves, G. J., & Chisari, N. E. 2014, The Astrophysical Journal, 797, 82, doi: 10.1088/0004-637X/797/2/82
- [77]Leethochawalit, N., Kirby, E. N., Moran, S. M., Ellis, R. S., & Treu, T. 2018, The Astrophysical Journal, 856, 15, doi: 10.3847/1538-4357/aab26a
- [78]Lidman, C., Suherli, J., Muzzin, A., et al. 2012, Monthly Notices of the Royal Astronomical Society, 427, doi: 10.1111/j.1365-2966.2012.21984.xDOI
- [79]Lin, Y.-T., Brodwin, M., Gonzalez, A. H., et al. 2013, ApJ, 771, 61, doi: 10.1088/0004-637X/771/1/61DOI
- [80]Lin, Y.-T., & Mohr, J. J. 2004, ApJ, 617, 879, doi: 10.1086/425412DOI
- [81]Lin, Y.-T., Ostriker, J. P., & Miller, C. J. 2010, ApJ, 715, 1486, doi: 10.1088/0004-637X/715/2/1486DOI
- [82]Lisenfeld, U., Alatalo, K., Zucker, C., et al. 2017, Astronomy & Astrophysics, 607, A110, doi: 10.1051/0004-6361/201730898
- [83]Lisker, T. 2008, ApJS, 179, 319, doi: 10.1086/591795DOI
- [84]Loh, Y.-S., & Strauss, M. A. 2006, MNRAS, 366, 373, doi: 10.1111/j.1365-2966.2005.09714.xDOI
- [85]Lotz, J. M., Primack, J., & Madau, P. 2004, AJ, 128, 163, doi: 10.1086/421849DOI
- [86]Mann, A. W., & Ebeling, H. 2012, MNRAS, 420, 2120, doi: 10.1111/j.1365-2966.2011.20170.xDOI
- [87]Marinacci, F., Vogelsberger, M., Pakmor, R., et al. 2018, Monthly Notices of the Royal Astronomical Society, doi: 10.1093/mnras/sty2206
- [88]Martinet, N., Durret, F., Guennou, L., et al. 2015, Astronomy & Astrophysics, 575, A116, doi: 10.1051/0004-6361/201427396
- [89]McNaught-Roberts, T., Norberg, P., Baugh, C., et al. 2014, Monthly Notices of the Royal Astronomical Society, 445, 2125–2145, doi: 10.1093/mnras/stu1886
- [91]Mulchaey, J. S., & Zabludoff, A. I. 1999, ApJ, 514, 133, doi: 10.1086/306952DOI
- [93]Naiman, J. P., Pillepich, A., Springel, V., et al. 2018, Monthly Notices of the Royal Astronomical Society, 477, 1206–1224, doi: 10.1093/mnras/sty618
- [94]Nelson, D., Pillepich, A., Springel, V., et al. 2017, Monthly Notices of the Royal Astronomical Society, 475, 624–647, doi: 10.1093/mnras/stx3040
- [95]Nelson, D., Springel, V., Pillepich, A., et al. 2021, https://arxiv.org/abs/1812.05609
- [96]Nersesian, A., Zibetti, S., D’Eugenio, F., & Baes, M. 2023, A&A, 673, A63, doi: 10.1051/0004-6361/202345962DOI
- [97]Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, AJ, 124, 266, doi: 10.1086/340952DOI
- [98]Pillepich, A., Nelson, D., Hernquist, L., et al. 2017, Monthly Notices of the Royal Astronomical Society, 475, 648–675, doi: 10.1093/mnras/stx3112
- [99]Pintos-Castro, I., Yee, H. K. C., Muzzin, A., Old, L., & Wilson, G. 2019, ApJ, 876, 40, doi: 10.3847/1538-4357/ab14ee
- [100]Pirzkal, N., Viana, A., & Rajan, A. 2010, Instrument Science Report WFC3 2010-06, 16 pages
- [101]Ponman, T. J., Allan, D. J., Jones, L. R., et al. 1994, Nature, 369, 462, doi: 10.1038/369462a0DOI
- [102]Ponman, T. J., & Bertram, D. 1993, Nature, 363, 51, doi: 10.1038/363051a0DOI
- [105]Postman, M., & Lauer, T. R. 1995, ApJ, 440, 28, doi: 10.1086/175245DOI
- [106]Ragone-Figueroa, C., Granato, G. L., Ferraro, M. E., et al. 2018, MNRAS, 479, 1125, doi: 10.1093/mnras/sty1639
- [107]Rodriguez-Gomez, V., Genel, S., Vogelsberger, M., et al. 2015, Monthly Notices of the Royal Astronomical Society, 449, 49–64, doi: 10.1093/mnras/stv264
- [108]Rousseeuw, P. J., & Van Driessen, K. 2006, Data Mining and Knowledge Discovery, 12, 29, doi: 10.1007/s10618-005-0024-4DOI
- [109]Rudnick, G., von der Linden, A., Pelló, R., et al. 2009, ApJ, 700, 1559, doi: 10.1088/0004-637X/700/2/1559DOI
- [110]Salpeter, E. E. 1955, ApJ, 121, 161, doi: 10.1086/145971DOI
- [111]Salvati, L., Saro, A., Bocquet, S., et al. 2022, The Astrophysical Journal, 934, 129, doi: 10.3847/1538-4357/ac7ab4DOI
- [112]Sandage, A. 1972, ApJ, 178, 1, doi: 10.1086/151763DOI
- [113]Santos, J. S., Tozzi, P., Rosati, P., Nonino, M., & Giovannini, G. 2012, Astronomy & Astrophysics, 539, A105, doi: 10.1051/0004-6361/201118162DOI
- [114]Schneider, D. P., Gunn, J. E., & Hoessel, J. G. 1983, ApJ, 264, 337, doi: 10.1086/160602DOI
- [115]SDSS Collaboration, Adamane Pallathadka, G., Aghakhanloo, M., et al. 2025, arXiv e-prints, arXiv:2507.07093, doi: 10.48550/arXiv.2507.07093DOI
- [116]Sérsic, J. L. 1963, Boletin de la Asociacion Argentina de Astronomia La Plata Argentina, 6, 41
- [117]Shen, S., Yang, X., Mo, H., van den Bosch, F., & More, S. 2014, The Astrophysical Journal, 782, 23, doi: 10.1088/0004-637X/782/1/23DOI
- [118]Shimakawa, R., Kodama, T., Hayashi, M., et al. 2018, MNRAS, 473, 1977, doi: 10.1093/mnras/stx2494DOI
- [119]Smirnoff, N. 1939, Bull. Math. Univ. Moscou, 2, 16
- [120]Solanes, J. M., Perea, J. D., Darriba, L., et al. 2016, MNRAS, 461, 321, doi: 10.1093/mnras/stw1278DOI
- [121]Springel, V., Pakmor, R., Pillepich, A., et al. 2017, Monthly Notices of the Royal Astronomical Society, 475, 676–698, doi: 10.1093/mnras/stx3304DOI
- [122]Stott, J. P., Collins, C. A., Burke, C., Hamilton-Morris, V., & Smith, G. P. 2011, Monthly Notices of the Royal Astronomical Society, 414, 445–457, doi: 10.1111/j.1365-2966.2011.18404.xDOI
- [123]Stott, J. P., Pimblet, K. A., Edge, A. C., Smith, G. P., & Wardlow, J. L. 2009, Monthly Notices of the Royal Astronomical Society, 394, 2098–2108, doi: 10.1111/j.1365-2966.2009.14477.xDOI
- [124]Stott, J. P., Collins, C. A., Sahlén, M., et al. 2010, ApJ, 718, 23, doi: 10.1088/0004-637X/718/1/23DOI
- [125]Strauss, M. A., Weinberg, D. H., Lupton, R. H., et al. 2002, AJ, 124, 1810, doi: 10.1086/342343DOI
- [126]Strazzullo, V., Pannella, M., Mohr, J. J., et al. 2019, A&A, 622, A117, doi: 10.1051/0004-6361/201833944DOI
- [127]Thomas, D., Maraston, C., Schawinski, K., Sarzi, M., & Silk, J. 2010, MNRAS, 404, 1775, doi: 10.1111/j.1365-2966.2010.16427.xDOI
- [128]Tremaine, S. D., & Richstone, D. O. 1977, ApJ, 212, 311, doi: 10.1086/155049DOI
- [129]Vaughan, S. P., Barone, T. M., Croom, S. M., et al. 2022, Monthly Notices of the Royal Astronomical Society, 516, 2971–2987, doi: 10.1093/mnras/stac2304DOI
- [130]Vicentin, M. C., Strauss, M. A., Sodré, Jr., L., et al. 2026, ApJ, 1000, 103, doi: 10.3847/1538-4357/ae45fcDOI
- [131]Vikhlinin, A., McNamara, B. R., Hornstrup, A., et al. 1999, The Astrophysical Journal, 520, L1–L4, doi: 10.1086/312134DOI
- [133]Von Der Linden, A., Best, P. N., Kauffmann, G., & White, S. D. M. 2007, Monthly Notices of the Royal Astronomical Society, 379, 867–893, doi: 10.1111/j.1365-2966.2007.11940.xDOI
- [134]Vulcani, B., Poggianti, B. M., Oemler, A., et al. 2013, A&A, 550, A58, doi: 10.1051/0004-6361/201118388DOI
- [135]Wen, Z. L., & Han, J. L. 2015, MNRAS, 448, 2, doi: 10.1093/mnras/stu2722DOI
- [136]Whiley, I. M., Aragón-Salamanca, A., De Lucia, G., et al. 2008, MNRAS, 387, 1253, doi: 10.1111/j.1365-2966.2008.13324.xDOI
- [137]Zhang, Y., Miller, C., McKay, T., et al. 2016, ApJ, 816, 98, doi: 10.3847/0004-637X/816/2/98DOI