A proto-cluster of massive quiescent galaxies at $z=4$

Galaxies in the local Universe exhibit a large diversity in their morphologies as characterized by the Hubble Sequence. Most massive galaxies tend to be elliptical and they show little sign of on-going star formation activities. They form a tight sequence on a color-magnitude diagram (red sequence; Bower et al. 1992) and the tightness indicates that they are a very homogeneous population. Their spectra indeed show that they are all dominated by old stars formed long time ago. Furthermore, detailed absorption line analyses suggest that they formed in an intense burst of star formation occurred in a short timescale, followed by passive evolution (i.e., simple aging without formation of new stars; Thomas et al. 2010). This is an intriguing observational fact because these galaxies are massive, and we expect massive galaxies to appear only at low redshifts in the current paradigm of hierarchical galaxy formation. These observations pose three major questions:

• •

  What physical process(es) drives the intense starburst in the early Universe?

• •

  What causes the rapid shutdown of the starburst activity?

• •

  What physical process(es) prevents subsequent formation of new stars for $>$10 Gyr?

An interesting observational approach to these questions is to go back in time and study progenitors of nearby massive elliptical galaxies in the early Universe. A lot of efforts have been put in the search for distant quiescent galaxies (i.e., galaxies with no star formation) and quiescent galaxies have been identified up to $z\sim 4$ and beyond (Schreiber et al., 2018a; Tanaka et al., 2019; Valentino et al., 2020; Kakimoto et al., 2023) thanks to the advent of sensitive near-infrared spectrographs on the ground. These galaxies are post-starburst galaxies; their star formation histories (SFHs) inferred from extensive fitting of their SEDs with stellar population synthesis models indicate that they all underwent an intense starburst in a recent past and then experienced rapid quenching. We are getting very close to the formation epoch of the nearby elliptical galaxies. However, as we show below, $z\sim 4$ is the spectroscopic sensitivity limit from the ground and it is extremely challenging to go further.

JWST has already started to identify quiescent galaxies at $z\gtrsim 3$ from its very first observations (Carnall et al., 2023a, b; Marchesini et al., 2023; Valentino et al., 2023), and it will surely revolutionize our understanding of the formation of massive quiescent galaxies. Due to its small field coverage, however, one key aspect will be missed in blank field JWST observations: environment. In the local Universe, we know galaxy properties are strongly dependent on their surrounding environment (Dressler, 1980). Quiescent early-type galaxies dominate the dense cluster core, while star-forming spiral galaxies are the main population in the low-density field. Also, we expect that galaxies form earlier in higher density peaks of the density fluctuation in the Universe; galaxy formation is accelerated in cluster regions. However, massive clusters are rare objects in the Universe especially at high redshifts. Thus, a wide-field observation is required to understand the role of environment in the formation of massive quiescent galaxies in the distant Universe.

There have been a lot of efforts to search for (proto-)clusters using star-forming galaxies as a tracer (e.g., Miller et al. 2018; Oteo et al. 2018; Toshikawa et al. 2018). However, much less attention has been paid to quiescent galaxies due primarily to difficulties in their identifications at high redshifts; they are faint and rare objects ($\sim 10^{-6}\rm\ Mpc^{-3}$; Schreiber et al. 2018a; Valentino et al. 2020; Marsan et al. 2022; Valentino et al. 2023), requiring deep imaging data over a wide area. Deep multi-wavelength data in fields like the Cosmic Evolution Survey (COSMOS) field make it possible to search for them in the early Universe. For example, Strazzullo et al. (2015) searched for concentrations of quiescent galaxies in the COSMOS field out to $z\sim 2.5$. Recently, Ito et al. (2023a) successfully confirmed a concentration of massive quiescent galaxies with deep near-IR spectroscopy at $z=2.7$. Kubo et al. (2015, 2021) identified a massive quiescent galaxy in a proto-cluster at $z=3.1$. The currently known high-$z$ proto-clusters with quiescent galaxies are still only a few and are limited to $z\lesssim 3$. In this paper, we focus on quiescent galaxies at $z=4$ based on deep spectroscopic observations and discuss an early structure populated by these quiescent galaxies to address the role of environment.

The paper is structured as follows. We describe our spectroscopic observations in Section 2, followed by spectroscopic analyses of a confirmed quiescent galaxy at $z=3.99$ in Section 3. We then focus on a concentration of several quiescent galaxies at the same redshift using joint photometric and spectroscopic fits to their observed spectral energy distributions (SEDs) in Section 4. Section 5 summarizes physical properties of the concentration such as the Butcher-Oemler effect (Butcher & Oemler, 1984) at $z=4$. Finally, we discuss our results in the context of massive galaxy formation and conclude the paper in Section 6. We adopt $\rm H_{0}=70\ km\ s^{-1}\ Mpc^{-1}$, $\rm\Omega_{M}=0.3$, and $\rm\Omega_{\Lambda}=0.7$. Magnitudes are in the AB system (Oke & Gunn, 1983).

We focus on the Subaru/XMM-Newton Deep Survey (SDXS), where very deep, multi-wavelength imaging data are available over a wide area. We have constructed a multi-band photometric catalog as summarized in Kubo et al. (2018). This is the same catalog that Tanaka et al. (2019) and Valentino et al. (2020) are based on. We apply the photometric redshift code from Tanaka (2015) to the catalog to infer photometric redshifts as well as physical properties of galaxies such as stellar mass and star formation rate (SFR). In brief, the code uses models from Bruzual & Charlot (2003) over a wide parameter range with template error functions to account for systematic uncertainties in the models. We adopt the Chabrier (2003) initial mass function.

The code is calibrated to relatively low-redshift objects with spectroscopy available at that time. Since then, VANDELS (Pentericci et al., 2018; Garilli et al., 2021) has carried out an extensive spectroscopic campaign focused on high-redshift galaxies in the field and released their catalog to the public. We compare our photometric redshifts with secure spectroscopic redshifts from VANDELS DR4 with zflag$=3$ or $4$ in Fig. 1. We note that the VANDELS targets are selected on the basis of photometric redshifts. There are a few different types of targets, but we are primarily using quiescent galaxies at $1<z<2$ and star-forming galaxies at $2.5<z<5$ here.

We emphasize that Fig. 1 is entirely a blind comparison; VANDELS spec-$z$’s were not available when we computed our photometric redshifts. As can be seen, the correspondence is excellent: an outlier rate defined as the fraction of objects with $|\delta z|/(1+z)>0.15$ (Tanaka et al., 2018) is 4.8% and the dispersion defined as $\sigma((\delta z)/(1+z))$ is 0.036. The outliers scatter down to low redshifts, but most of them are consistent with spec-$z$’s within the uncertainties.

As VANDELS does not include quiescent galaxies at high redshifts, we supplement it with high-$z$ quiescent galaxies from the literature (Schreiber et al. 2018a; Tanaka et al. 2019) as well as from this paper and show them as the red circles in Fig. 1. While the redshift distribution of the quiescent galaxies is very sparse, the photo-$z$ accuracy for the quiescent galaxies at $z\sim 4$ is similar to that of star-forming galaxies from VANDELS. The figure demonstrates that our photometric redshifts are accurate even for quiescent galaxies at high redshifts.

Motivated by the good accuracy, we search for an over-density of massive quiescent galaxies at high redshifts. We search for over-density regions by selecting quiescent galaxies in a redshift slice of $\delta z\pm 0.3$ and evaluate their over-density significance as a function of position. This redshift slice roughly corresponds to $\pm 1.5\sigma(z_{phot})$. Quiescent galaxies are defined as those with $1\sigma$ upper limit of specific SFR (sSFR) being smaller than $10^{-9.5}\rm Gyr^{-1}$, which corresponds to $\sim 1$ dex below the star formation main sequence, as we have done in our previous papers (Kubo et al., 2018; Valentino et al., 2020; Ito et al., 2022, 2023a). We identify a significant over-density of quiescent galaxies at $z\sim 4$ in SXDS as shown in Fig. 2; based on the kernel density estimate with Gaussian width of 2.4 arcmin (1 physical Mpc at $z=4$), the over-density significance is $\sim 20\sigma$. Tanaka et al. (2019) reported the discovery of the $z=4.01$ galaxy in the same field and at the same redshift as the over-density region. This is not a coincidence as we discuss below. The over-density region is approximately 5 arcmin ($\sim 2$ physical Mpc) on a side.

In this Section, we examine properties of the $z=3.99$ galaxy in detail and compare them with those of the $z=4.01$ galaxy presented in Tanaka et al. (2019) and Valentino et al. (2020). We closely follow the analysis of Tanaka et al. (2019) in this Section. We measure its physical size (Section 3.1), and discuss the dynamical properties (Section 3.2). We then infer its star formation histories (Section 3.3). As we show below, the two galaxies are both massive galaxies with suppressed SFRs, but there are interesting differences. The $z=3.99$ galaxy is in an over-density region (Section 5), while the $z=4.01$ galaxy is not. We compare their properties with their environmental differences in mind.

We turn our attention to the faint quiescent galaxies in the over-density region. While direct redshift measurements from the spectra were not possible for them due to their low-S/N (the median S/N per pixel is 0.8), the spectra trace the continuum shape around the 4000Å/Balmer break well. We here make an attempt to constrain their redshifts by using both the spectra and photometry. As our spectra are of low-S/N and there are residuals of the night sky lines, we first bin the spectra by 100Å by clipping the top and bottom 10% of the flux distribution in each bin. The clipping is intended to exclude the residual of the sky lines, which is non-Gaussian and often contaminates the mean flux. We then feed the binned spectra and photometry to a photometric redshift code (Tanaka, 2015) to constrain their redshifts.

As a proof of concept, the technique is first applied to the confirmed $z=3.99$ and $z=4.01$ galaxies. This “spectro-photometric” redshift technique successfully reproduces the original spectroscopic redshifts of these two galaxies as shown in Fig. 7. The redshift probability distribution functions are sharply peaked at the spectroscopic redshifts with very small uncertainties. These strong redshift constraints come from the continuum shape traced by the MOSFIRE spectra; the spectra cover the break feature and the location of the break gives a strong redshift constraint. We perform the same exercise for the spectroscopic objects presented in Ito et al. (2023a) and confirm that we can successfully reproduce the spectroscopic redshifts.

Motivated by the excellent redshift recovery, we apply the technique to the faint quiescent galaxy candidates for which no secure redshifts could be measured from MOSFIRE. Fig. 8 shows the result; all of them show a sharp redshift spike at $z=4$. There is one object with a significantly bimodal $P(z)$ distribution (SXDS2-16609), but the 2nd peak is consistent with $z=4$. Due to the location of the object in the MOSFIRE mask, we could not cover the blue part of the $K$-band for this object. That likely resulted in the wider $P(z)$. In any case, it is very likely that most of these faint objects are located at $z=4$. The inferred redshifts of the faint quiescent galaxies re summarized in Table 1.

The spectro-photometric fits also give fairly robust estimates of physical properties of galaxies such as stellar mass and SFR (Table 2). Fig. 9 plots SFR against stellar mass. The confirmed quiescent galaxies at $z=4$ are all below the star formation main sequence, and they tend to be massive galaxies. There are less quiescent galaxies and more star-forming galaxies at lower mass. We are complete to $\rm M_{*}\sim 3\times 10^{10}M_{\odot}$ and the decreasing quiescent galaxies towards lower mass is likely a real trend. We will elaborate on this point in Section 5.2.

We have presented the spectroscopic confirmation of a massive quiescent galaxy at $z=3.99$ as well as the discovery of a significant concentration of quiescent galaxies at $z=4$. Based on the deep spectroscopic observation with MOSFIRE, we have identified multiple Balmer absorption lines of the $z=3.99$ quiescent galaxy. Its stellar velocity dispersion is $\sigma=305\pm 103\rm\ km\ s^{-1}$ and is slightly higher than lower redshift counterparts with similar stellar masses, indicating a mild redshift evolution. Its physical size measured from the deep imaging data reveals its compact size and dense stellar density, consistent with other quiescent galaxies observed at similar redshifts in the literature.

Interestingly, the galaxy is located in an over-density region of quiescent galaxies at similar redshifts. While we could not directly measure their spectroscopic redshifts from their spectra, the high-accuracy spectro-photometric redshifts strongly suggests that they are at the same redshifts as the $z=3.99$ galaxy. All these galaxies exhibit a pronounced Balmer break, indicative of recent starburst activity followed by rapid quenching. Their UV emission is relatively weak and their SFRs inferred from their overall SEDs indeed suggest that they are quiescent. The concentration is embedded in a large-scale filament traced by the VANDELS spectroscopic galaxies, which adds further credits to the physical association. This is the first discovery of a concentration of quiescent galaxies at this high redshift and it serves as an unique laboratory to investigate the role of environment at $z=4$ for the first time.

The cluster galaxies form the early red sequence, suggesting that these galaxies form and quench at similar times, and the red sequence emerges during the first collapse of a cluster. In addition, the quiescent galaxy fraction is significantly higher than the field average. This may be expected because we define the over-density region around the quiescent galaxies, but we have extended the Butcher-Oemler effect to z=4 for the first time. The fact that the collapsing cluster exhibits a higher quiescent fraction suggests that we may be witnessing the environmental dependence of galaxy formation such that galaxies in higher density regions form earlier than the field galaxies. We aim to address this point further in our future paper.

Given the significant over-density, we expect that the cluster region will collapse to form the core of a massive cluster at lower redshifts. We make an attempt to roughly estimate the future halo mass of the system when they collapse. Using the halo occupation distribution modeling from Shuntov et al. (2022), we infer the halo mass of each quiescent galaxy from its stellar mass as summarized in Table 2. The individual halo mass ranges from 1 to $3\times 10^{12}\rm~{}M_{\odot}$. If we assume that they all collapse to form a single halo, the total mass exceeds $10^{13}\rm~{}M_{\odot}$, i.e., group-sized halo. We consider only the quiescent members, but there are star-forming galaxies with consistent photo-$z$’s, and thus the total halo mass here is a lower limit. We find that the virial radii of the galaxies derived from their halo mass do not overlap with each other, suggesting that they are not yet in the same halo at the time of the observation.

In order to further characterize the properties of the cluster, we have searched for similar systems in the Illustris TNG-300 simulation at $z=4.01$ (snapshot 21; Nelson et al. 2019). To be specific, we identify quiescent galaxies with $\rm M_{*}>10^{10}\rm M_{\odot}$ and $\log\rm sSFR<10^{-9.5}\ M_{\odot}\ yr^{-1}$ in the simulation and examine their spatial distribution. Stellar mass and SFR are measured within twice the stellar halo-mass radius, and SFR is averaged over 10 Myr. Interestingly, there is no over-density region of the quiescent galaxies in TNG-300; there are only 11 massive quiescent galaxies within the 300 comoving Mpc simulation box and the closest separation between them is as large as 8 physical Mpc (3d distance). In contrast, we have 5 quiescent galaxies within 1 physical Mpc radius (projected distance). While the 3d and projected distances are different, it is clear that the observed cluster is a much denser system. We note that we perform the same analysis but considering more massive galaxies with $>5\times 10^{11}\rm M_{\odot}$ only and confirmed there is still no counterpart in Illustris TNG-300.

We do not know yet if this is a failure of the simulation in reproducing quiescent galaxies at $z\sim 4$; the number density of massive quiescent galaxies at $3\lesssim z\lesssim 4$ is in broad agreement between observation and simulations (Valentino et al. 2023 and references therein), although it depends strongly on how quiescent galaxies are defined. It may be that our system is so rare that the simulation box is too small to reproduce it. In any case, massive halos at high redshifts have to be rare and it is crucial to increase the cluster sample to perform further comparisons with simulations.

We expect that our spectra of the faint cluster members will be able to strongly constrain their star formation histories as the spectra cover the Balmer/4000$\rm\AA$ break for most of them. All of the members show a sign of suppressed star formation as discussed earlier and it would be of interest to perform extensive SED fitting to infer their primary formation epoch and quenching timescale and compare them with those of the field galaxies from the literature. Also, we can measure the physical sizes of the members and see whether there is environmental dependence of galaxy sizes at these redshifts. The environmental dependence of physical sizes of galaxies at low redshifts is somewhat controversial in the literature (see Table 1 of Yoon et al. 2017, and Table A1 of Matharu et al. 2019), and it would be interesting to look at the $z=4$ system. We hope to address these points in a forthcoming paper. Also, detailed imaging/spectroscopic follow-up observations with JWST are an interesting avenue as they deliver secure spectroscopic redshifts, more accurate physical sizes, etc. ALMA observations to search for dust-obscured galaxies that went unnoticed in our work will be interesting, too. This work can be extended in multiple ways, and joint analyses will hopefully lead us to deeper insights into the early cluster formation and the quenching physics in the early Universe.