A Detailed Study of Massive Galaxies in a Protocluster at z = 3.13

In this work, we make use of publicly available multiwavelength data including the deep optical ugriz images from the CFHTLS Deep Survey (Gwyn 2012) and the near-IR JHK_S bands from the WIRCam Deep Survey (WIRDS) (Bielby et al. 2012). We also use the Spitzer data from the Spitzer Wide-area InfraRed Extragalactic survey (SWIRE; Lonsdale et al. 2003) and the Spitzer Extragalactic Representative Volume Survey (SERVS; Mauduit et al. 2012). The former includes 5.8, 8.0, and 24 μm bands, while the latter taken as part of post-cryogenic IRAC observations includes 3.6 and 4.5 μm bands only, which are deeper than their SWIRE counterparts. The photometric depths of the CFHTLS and WIRDS data are measured from the sky fluctuations by placing 2'' diameter apertures in random image positions. The depths of the Spitzer data are measured in Vaccari (2015). Table 1 summarizes the data sensitivity and image quality in this paper.

We resample the Spitzer IR data to have the same pixel scale of 0.186'' as the optical CFHTLS and near-IR WIRDS data. To facilitate the comparison with the LAE overdensity found in the D1 field, all the images are trimmed to have the same dimension as the narrow-band o3 image used in Shi et al. (2019a) and identical masks, with an effective area of 0.32 deg².

We create a multiwavelength photometric catalog as follows. First, to accurately measure the photometry, we smooth the WIRDS images to match the broader point-spread functions (PSFs) of the CFHTLS data. To do so, the radial profile of the PSF in each image is approximated by a Moffat function with the measured seeing FWHM. A noiseless convolution kernel between the low- and high-resolution images is then derived using the Richardson–Lucy deconvolution algorithm (Richardson 1972). Each WIRDS image data is then convolved with its respective kernel to create a smoothed image that is PSF matched with the CFHTLS data. The LAE overdensity discovered in Shi et al. (2019a) lies near the edge the WIRDS images: 20% of the area has no K_S band coverage and an additional 10% has only partial coverage (<50% of the full exposure), which limits our comprehensive study of massive galaxies in this protocluster field. Therefore in this work we base our study on the Spitzer Infrared Array Camera (IRAC) 3.6 μm detection. At z = 3.13, the 3.6 μm mainly samples the rest-frame optical-NIR emission which enables the measurement of the stellar masses of galaxies.

Source detection and photometric measurements in the ugrizJHK_S bands are carried out by running the SExtractor software (Bertin & Arnouts 1996) in dual mode on the PSF-matched images with the i band data as the detection band. The SExtractor parameter MAG_AUTO is used to estimate the total magnitude, while colors are computed from fluxes within a fixed isophotal area (i.e., FLUX_ISO). As the images are PSF matched, aperture correction in all bands is assumed to be the difference between MAG_AUTO and MAG_ISO measured in the detection band.

As for the Spitzer images, since the PSFs of the IRAC and Multiband Imaging Photometer for Spitzer (MIPS) images are much broader (≈2'' and 6'', respectively), source blending on these images is a severe problem. In order to obtain accurate and unbiased measurement of fluxes and colors on the Spitzer images, we utilize the T-PHOT software (Merlin et al. 2015, 2016), which performs "template-fitting" photometry on the low-resolution image using the information of the high-resolution image and catalog. In our case, the i band image and catalog are used as the input priors of T-PHOT, while the low-resolution Spitzer images are analyzed to obtain precise photometry. We notice that the resultant Spitzer fluxes derived by T-PHOT do not strongly depend on the based priors. For a test purpose, we also do a similar analysis using r band as the prior, and obtain the corresponding Spitzer photometry. We crossmatch the r band-based catalog with that of the i band, finding that the 3.6 μm magnitude difference between the two has only a mean value of ∼0.03 with a standard deviation of 0.06. Therefore, we are assured that our T-PHOT photometry is robust and unbiased.

Finally, all photometric catalogs are merged together to create a multiwavelength catalog. In this work, to secure the measurement of the stellar masses of the galaxies, we focus on the sources with 3.6 μm magnitudes smaller than 23.04 (i.e., >5σ detection limit). In the end, 31,218 sources are selected in the final catalog.

We derive photometric redshift for each object in the catalog via the spectral energy distribution (SED) fitting technique using the CIGALE software (Noll et al. 2009; Boquien et al. 2019). Based on an energy balance principle (the energy emitted by dust in the mid- and far-IR exactly corresponds to the energy absorbed by dust in the UV-optical range), CIGALE builds composite stellar population models from various single stellar population models, star formation histories (SFHs), dust attenuation laws, etc. The model templates are then fitted to the observed fluxes of galaxies from the far-UV to the radio domain, and physical properties are estimated using a Bayesian analysis.

For the SED templates, we use the stellar population synthesis models of Bruzual & Charlot (2003), Calzetti et al. (2000) reddening law with E(B–V) values ranging from 0–2 in steps of 0.1 mag, the solar metallicity, and the Chabrier (2003) initial mass function. We use the delayed SFH (SFR ∝t×exp[−t/τ]) with a star-forming timescale τ ranging from 0.1–10 Gyr. The age of the main stellar population ranges from 100 Myr to 10 Gyr, with finer grids up to 2 Gyr, after which large grids are used in order to save computation time, as in this work we are only interested in selecting z ∼ 3 galaxies. Nebular emission is also included and dust emission is modeled by Dale et al. (2014). The input redshifts are set to be between 0.1 and 5.0 in steps of 0.1. In addition, for the 24 μm detected sources, we also include the AGN models from Fritz et al. (2006) to better constrain the dust emission and AGN contribution.

We compare our photometric redshift (photo-z) measurements with the spectroscopic redshifts from the VIMOS VLT Deep Survey (VVDS; le Fèvre et al. 2013) and VIMOS Ultra-Deep Survey (VUDS; le Fèvre et al. 2015). The precision of the photometric redshift is measured using the normalized median absolute deviation defined as ${\sigma }_{z}=1.48\times$ median($| {{\rm{\Delta }}}_{z}|$/(1+z_spec)), where ${{\rm{\Delta }}}_{z}={z}_{\mathrm{spec}}-{z}_{\mathrm{phot}}$. This scatter measurement corresponds to the rms of a Gaussian distribution and is not affected by catastrophic outliers (i.e., objects with $| {{\rm{\Delta }}}_{z}|$/(1+z_spec) >0.15) (Ilbert et al. 2006; Laigle et al. 2016).

We crossmatch our sample with the VVDS and VUDS catalogs and find 3,685 sources have spectroscopic redshifts. For all these sources, we obtain σ_z = 0.12. The number of catastrophic failures take up to 18% in these sources. The mean photometric redshift error derived by CIGALE is Δz ∼ 0.2, therefore, we select 532 galaxies with photo-z measurements of 2.9 < z_phot < 3.3 as potential protocluster galaxy candidates, among which 75 have spectroscopic redshifts, yielding σ_z = 0.06. The reason why σ_z becomes smaller for these protocluster galaxy candidates is that our SED modeling is tuned to select high-z galaxies as described previously. We visually inspect the 532 sources and remove those with potential contamination in the photometry, including those severely blended with nearby bright sources and near the borders of the images. We check the locations of the removed sources, confirming they are relatively randomly distributed we do not particularly remove the galaxies in the overdense regions due to the blending issue. We also remove possible M-dwarf stars by inspecting their spectra. In the end, 356 galaxies are selected as our photo-z protocluster galaxy candidates.

We fix the best-fit photo-z of the protocluster galaxy candidates, using the spectroscopic redshift when available, and refit their SEDs using CIGALE with the same configuration to determine their physical properties such as stellar mass, dust corrected star formation rate (SFR), and color excess of stellar continuum E(B–V), etc. The masses of the galaxies are best determined with an average error of 0.10 dex, while the errors of the SFRs are relatively larger, with an average value of 0.35 dex.

For the 356 photo-z galaxies, we also estimate their stellar mass completeness using an empirical method (Pozzetti et al. 2010; Ilbert et al. 2013; Laigle et al. 2016). For each galaxy, we compute the lowest stellar mass M_lim it would need to be detected at the given IRAC magnitude limit [3.6]_lim = 23.04:

$$\begin{eqnarray*}&&\mathrm{log}({M}_{\mathrm{lim}})=\mathrm{log}(M)-0.4({[3.6]}_{\mathrm{lim}}-[3.6]),\end{eqnarray*}$$

then the stellar mass completeness limit corresponds to the mass above which 90% of the galaxies lie. The resultant mass completeness limit is $\mathrm{log}({M}_{\mathrm{lim}})=9.9$ in our photo-z sample.

Finally, our photo-z galaxies lie around at z ≈ 3.1 where the K_S band photometry could be potentially contaminated by the [O iii]λλ4959,5007 nebular emission lines, which could possibly affect the stellar mass measurement. For example, Schenker et al. (2013) measured the rest-frame [O iii] equivalent widths (EWs) for a sample of 3.0 < z < 3.8 LBGs and determined an average value of 250 Å. At z ≈ 3.1, this leads to an overestimate of K_S band continuum flux density by 0.3 magnitude. However, they also noticed that if SED fitting is used to derive the physical properties, there is no significant change in the stellar mass when the [O iii] emission is corrected (stellar mass is only reduced by 3%). This is because the IRAC 3.6 and 4.5 μm data provide important information on the stellar component redward of the Balmer break and are not contaminated by nebular emissions at 3 < z < 4. Since our photo-z galaxies all have a secure 3.6 μm detection, the contamination from [O iii] is expected to be less severe. In addition, Malkan et al. (2017) noticed there is an anticorrelation between the stellar mass and [O iii] EW for LBGs at z ∼ 3: the higher the mass, the smaller the EW. According to their relation, our mass-selected sample at M_⋆ > 10^9.9 M_☉ has a typical EW of ∼100 Å, corresponding to a flux contamination of 0.13 mag. Thus, the influence of nebular emission on the derived physical properties such as stellar mass would be minimal, considering the robust 3.6 μm detection and high-mass galaxies our sample have.

One of the main focus of this paper is to study the diverse galaxy populations in this protocluster field, in order to better understand the environmental impacts on galaxy evolution. To do so, we classify our photo-z galaxies using a J − K_S versus [3.6]–[4.5] color–color diagram, which is similar to those in the literature (e.g., Labbé et al. 2005; Papovich et al. 2006; Nayyeri et al. 2014; Girelli et al. 2019; Shi et al. 2019b).

The left panel of Figure 1 shows our selection of different galaxy populations using the two-color diagram. Utilizing the EZGAL software (Mancone & Gonzalez 2012) with the stellar population synthesis models of Bruzual & Charlot (2003) and the Chabrier (2003) initial mass function, we compute the theoretical models of different SFHs and dust reddening. Three SFHs are considered: (1) an instantaneous burst; (2) an exponentially declined model (SFR ∝exp[−t/τ] with τ = 0.1 Gry; (3) an exponentially declined model with τ = 1.0 Gyr. The ages of galaxies at the protocluster redshift z = 3.13 are also indicated in the color tracks.

Zoom In Zoom Out Reset image size

Based on Figure 1, we classify galaxies with J − K_S > 1.7 and [3.6]–[4.5] < 0.36 as quiescent galaxies. The red J − K_S color imposes that a strong Balmer/4000 Å break fall between the J and K_S bands at the protocluster redshift. As can be seen in the figure, this criterion tends to select galaxies of relatively old ages (>0.4 Gyr for instantaneous burst SFH model and >0.6 Gyr for a declined SFH model of τ = 0.1) with the absence of dust. Meanwhile, the [3.6]–[4.5] color requires that the rest-frame optical-NIR continuum slope at λ = 8000–10000 Å be relatively flat, ensuring that the red J − K_S color is not due to dust reddening. It is noted that our selection of quiescent galaxies also fully incorporates the distant red galaxies (DRGs) criterion (J − K_S > 1.4) at 2 < z < 4 (Franx et al. 2003; van Dokkum et al. 2003). Moreover, Girelli et al. (2019) recently also used the same colors to select quiescent galaxies at 2 < z < 4 with very similar criteria as ours. Using our criteria, 81 galaxies are selected as quiescent galaxy candidates in our sample.

In addition to passive galaxies, objects with [3.6]–[4.5] > 0.36 are classified as dusty star-forming galaxies, because the majority of these galaxies have colors consistent with a continuous SFH with high dust reddening E(B–V)≥0.5. Normal star-forming galaxies are selected to be in the region of J − K_S < 1.7 and [3.6]–[4.5] < 0.36, as they are mostly consistent with continuous SFH models with mild dust obscuration. In the end, 65 galaxies are classified as dusty star-forming and 210 are selected as normal star-forming galaxies in the sample.

In the right panel of Figure 1, we plot all our photo-z galaxies in the color–color diagram. We also divide the sample into two categories: LBGs that satisfy the dropout selection criteria used in Shi et al. (2019a); non-LBGs that do not satisfy the LBG criteria. Among the 356 photo-z galaxies, 116 are LBGs, which account for 33% of the entire sample. The majority of the LBGs (77, account for 66%) lie in the region of normal star-forming galaxies, while 21 are distributed in the region of dusty star-forming galaxies and only 18 are classified as quiescent galaxy candidates. Our results agree with the general expectation that LBGs are usually star-forming galaxies with little or moderate dust obscuration (e.g., Giavalisco 2002).

We also crossmatch our photo-z galaxies with the LAEs in Shi et al. (2019a), finding three counterparts. These galaxies are all UV-bright sources that have luminosities log(L${}_{\mathrm{UV}}$) >28.5 erg s⁻¹ Hz⁻¹ at the rest frame of 1700 Å. In comparison, the entire LAE sample has an average UV luminosity of ∼28.0 erg s⁻¹ Hz⁻¹. Therefore, these galaxies are among the most UV-luminous LAEs in the field, which are also detected in 3.6 μm. In particular, one galaxy is located in the LAE overdensity found in Shi et al. (2019a). SED fitting suggests it has a stellar mass of ${10}^{10.5}\,{M}_{\odot }$ with an SFR of only $1\,{M}_{\odot }$ yr⁻¹ and belongs to the quiescent galaxy population. Having such a high mass and low SFR, this LAE may be a rare one and worth further investigating in future observation. The remaining LAEs are not detected in 3.6 μm and therefore not selected in the photo-z catalog.

To sum up, our photo-z sample includes a large fraction of massive galaxies that have been missed from the rest-frame UV-selected star-forming galaxies such as LBGs and LAEs. This highlights the importance of using rest-frame optical-NIR selection to study the high-mass end of the stellar mass function.

We investigate the physical properties of the different galaxy populations classified above. Figure 2 shows the SED-fitting results for a subsample of our photo-z candidates. We see that quiescent galaxies are distinguished by their prominent break between J and K_S, while dusty star-forming galaxies usually are redder beyond the NIR wavelength as indicated by their best-fit spectra.

Zoom In Zoom Out Reset image size

Figure 3 shows the stellar mass, SFR, and dust reddening E(B–V) distributions of different galaxy populations selected by our criteria. It can be seen that the stellar masses of quiescent and dusty galaxies are skewed toward the higher mass end: the median stellar masses of the quiescent and dusty galaxies are 10^10.59 M_⊙ and ${10}^{10.53}$ M_⊙, respectively, while it is only 10^10.25 M_⊙ for normal star-forming galaxies. The quiescent galaxy sample has a very low median SFR of 6 M_⊙ yr⁻¹ compared to the normal star-forming galaxy population, which has a median SFR of 37 M_⊙ yr⁻¹. The dusty star-forming galaxies are skewed toward the higher SFR end, with a median value of 55 M_☉yr⁻¹. As for the dust content, dusty star-forming galaxies have a higher dust extinction with a median E(B–V) = 0.26, while quiescent and normal star-forming galaxies are less obscured by dust with a median E(B–V) of 0.13 and 0.15, respectively. We also validate the difference of the three galaxy populations using the K-sample Anderson–Darling test (Scholz & Stephens 1987). This test is similar to the commonly used Kolmogorov–Smirnov (K–S) test but more sensitive and can deal with more than two samples. The Anderson–Darling test finds significant differences among the three populations: p-value < 0.001 in all cases for the mass, SFR, and E(B–V) distributions.

Zoom In Zoom Out Reset image size

Numerous studies have indicated a correlation between stellar mass (M_star) and the SFR for star-forming galaxies, which is the so-called star-forming main sequence (MS) (e.g., Daddi et al. 2007; Elbaz et al. 2007; Noeske et al. 2007; Rodighiero et al. 2011; Reddy et al. 2012; Speagle et al. 2014; Salmon et al. 2015; Santini et al. 2017). In Figure 4, we show the locations of our photo-z galaxies on the SFR–M_star plane. In the figure we also show the MS relation from both observation and simulation at z ∼ 3. On the one hand, most of our star-forming galaxies and dusty star-forming galaxies are located close to the MS from both observation (Speagle et al. 2014) and simulation (Dutton et al. 2010). On the other hand, the majority of the quiescent galaxy candidates lie below the MS relation: among the 81 candidates, only seven lie above the Dutton et al. (2010) relation, while only two lie above the Speagle et al. (2014) relation. Therefore, our quiescent galaxy candidates are indeed quenched systems with little ongoing star formation activities, compared to the star-forming galaxy populations. This further justifies our selection criteria for different galaxy populations.

Zoom In Zoom Out Reset image size

Having identified the high-density protocluster regions in Section 4, we investigate the impacts of local environment of the protocluster on the physical properties of galaxies in this section.

The main challenge in studying the environmental effects on protocluster galaxies is the lack of spectroscopic redshifts. The large redshift dispersion of our photo-z galaxies (Δz ∼ 0.4) prohibits the precise determination of the galaxy membership and a robust mapping of the genuine protocluster region. Thus, follow-up spectroscopic observations on this protocluster are urgently needed.

With the above caveats in mind, we compare the physical properties of protocluster galaxy candidates with those in the field. To this end, we divide our photo-z sample into two subsamples: the "overdensity" sample within the 1.3Σ iso-density contour lines enclosed by boxes A and B shown in Figure 5 and the "field" sample that is simply all the 356 photo-z galaxies in the entire field. As discussed in Section 4, since this paper focuses on the z = 3.13 confirmed protoclusters in A and B, also in lack of spectroscopic information elsewhere, we do not consider other apparent "overdense" regions. Because we cannot rule out the possibility that other overdense regions are genuine structures at other redshift, we choose the whole survey field (including A and B) as the general field. This definition of the field should represent the average galaxy populations at z ∼ 3. The overdensity sample is further divided into two groups (regions A and B, respectively). Our photo-z galaxies are all selected from the same set of photometric data and their properties are determined using the same method, therefore, no selection effect needs to be accounted for.

Forty-nine galaxies are located within the high-density regions (31 in region A and 18 in region B). Table 2 lists the median physical properties obtained of each subsample. The errors correspond to the median absolute deviations that are less affected by outliers.

Table 2.  Physical Properties of the Photo-z Galaxies in Different Environments

Region	N	log(Mass)	SFR	E(B–V)	log (sSFR)
		(${M}_{\odot }$)	(${M}_{\odot }\ {{yr}}^{-1}$)		(${\mathrm{yr}}^{-1}$)
A	31	10.39 ± 0.46	44 ± 44	0.17 ± 0.13	−8.9 ± 0.7
B	18	10.38 ± 0.33	65 ± 35	0.22 ± 0.12	−8.4 ± 0.4
A + B	49	10.39 ± 0.40	51 ± 53	0.18 ± 0.12	−8.7 ± 0.7
Field	356	10.37 ± 0.30	29 ± 39	0.17 ± 0.13	−8.9 ± 0.8

Download table as:  ASCIITypeset image

In terms of stellar mass, we do not find obvious differences between different subsamples. A two-sample K–S test cannot distinguish between region A and/or B with the field, as well as between A and B (p-value >0.8 in all cases), which is also confirmed in Table 2. As for the SFR, on one hand, there is no significant difference between A and the field (p = 0.2), while the K–S test indicates there is a significant difference between B and the field (p = 0.003). However, when we compare only A with B, the difference fades away, with a p-value of 0.3. If A + B is compared with the field, K–S test suggests the probability that they come from the same underlying distribution is <1% (p = 0.004). In Figure 7, the photo-z protocluster galaxies (A + B) and field galaxies are shown on the SFR–M_star plane. It can be seen that although the stellar masses of the two groups are similar in distribution, the SFRs of the protocluster galaxies are skewed toward higher values than their field counterparts. The enhancement of the SFRs can be further seen in the right panel of Figure 8, where we show the distributions of the specific star formation rate (sSFR, defined as SFR/M_star) for the two groups. The K–S test implies a strong distinction between B and the field (p = 0.02), while no significant difference between A and the field is observed (p = 0.6). This leads to a moderate difference between the overall overdensity with the field (p = 0.07), but the K–S test cannot reject the null hypothesis that A and B come from the same distribution (p = 0.2). These K–S tests appear to suggest that galaxies within the protocluster regions (especially in region B) are forming stars more actively than the general field. In Table 2, we can also see that the SFR in the protocluster regions A + B is enhanced by ∼76% as compared to the field. This elevation is even higher for region B (∼124%) than that for region A (∼52%).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As for the dust extinction, no significant difference between regions A and/or B with the field is observed (p > 0.2). Figure 8 (left panel) shows the distribution of the dust attenuation parameter E(B–V) between the protocluster galaxies and field galaxies. Both Table 2 and the histogram imply that the overdense protocluster regions have dust content similar to that of the general field.

The enhancement of the SFRs in the protocluster regions is consistent with our previous work (Shi et al. 2019a) where we found that the Lyα luminosity and UV luminosity of the protocluster galaxies have higher median values than the field. In Shi et al. (2019a), since we did not have the mass measurement of the LAEs, we could not rule out the possibility that the trend is due to a deficit of low-mass galaxies in the protocluster environment. This work goes a step further by measuring the stellar masses, confirming that the enhancement of the SFRs of the protocluster galaxies is not due to the lack of low-mass galaxies but rather is due to an overall boost of star formation efficiency in the overdense protocluster environment. Furthermore, our results appear to suggest that protocluster galaxies in region B have even higher star formation efficiency, compared to those in region A. We will discuss this later in Section 5.2.

Recently, Shimakawa et al. (2018) studied a protocluster at z = 2.5 using HAEs to trace the large-scale structure. They found that HAEs in the densest regions of the protocluster have enhanced SFRs and dust extinctions at high confidence level, indicating a rapid mass assembly of star-forming galaxies in the protocluster regions. At a similar redshift, Wang et al. (2018) investigated the molecular gas properties of a distant X-ray cluster (Wang et al. 2016), finding that the star formation efficiency (indicated by the ratio between the SFR and gas mass) is elevated in the cluster region in comparison with the field. They argued that the galaxies in the central regions of this cluster will consume all the gas and become quiescent in a short timescale. The enhancement of star formation activity in our protocluster field (both measured from LAEs and photo-z galaxies) is consistent with the above studies. We argue that galaxies in our protocluster regions are also experiencing an accelerated mass assembly, likely consuming their gas rapidly and becoming quiescent in a short time period.

It is known that protoclusters often host extremely dusty star-forming galaxies such as submillimeter galaxies (SMGs) with SFRs exceeding 1000 solar masses per year (e.g., Casey 2016; Kato et al. 2016; Miller et al. 2018; Oteo et al. 2018; Cheng et al. 2019). These objects are very luminous in the submillimeter wavelength and are heavily dust obscured, and are generally invisible in rest-frame UV-NIR wavelengths. It is possible that some SMGs exist in our protocluster that are missed by our selection. An extensive study of this galaxy population requires future submillimeter observations in this field. If these dusty starbursts systems are confirmed to be preferably concentrated in our protocluster regions, the enhancement of star formation activities in the dense environments would be even higher.

In Section 4, we find that the photo-z galaxies form two overdensities A and B in the field, which are co-spatial with our previously identified LBG and LAE overdensities, respectively (Shi et al. 2019a). Given the large end-to-end size of these two overdensities (∼40 Mpc) that is almost twice the size of the largest protocluster in Chiang et al. (2013) at z ∼ 3 (∼20 Mpc), we assume these two overdensities trace separate structures in the following discussion.

In Shi et al. (2019a), we argued that the spatial segregation of regions A and B is possibly due to different formation times of underlying dark matter halos. The former structure formed earlier than the latter, thus is traced by older, more massive LBGs, while the latter is traced by a younger LAE population. In this work, our galaxy selection criteria suggest that 42 ± 12% of the galaxies in region A are massive quiescent and/or dusty galaxies (similar to the field of 41 ± 3%), compared to 22 ± 11% in region B (Section 4). Meanwhile, the normal star-forming galaxies are strongly correlated with the LAEs as shown in Figure 6. Therefore, it turns out that the overdensity B largely coincides with the LAE overdensity (see Figure 5). It is also noticed that there are two 24 μm detected objects (orange diamonds in Figure 5) located in region A, including one brightest cluster galaxy candidate discovered in Shi et al. (2019a), while no source is detected at 24 μm in region B. In addition, in Section 5.1 we see that galaxies in region B appear to have a higher star formation efficiency than those in region A. Taken together, these results suggest that region A is a more evolved structure, which is mainly traced by old and/or dusty galaxy populations. In comparison, region B formed at a later stage and is dominated by a younger galaxy population, which consists mostly of normal star-forming galaxies such as LAEs. The elevated star formation activities in region B suggests that its galaxy constituents are rapidly building their masses. In comparison, region A appears to be a more settled structure that has already passed the peak of its star formation.

Finally, we consider the scenario that regions A and B are two protoclusters that embedded in a primordial supercluster. In the local and nearby universe, a supercluster typically consists of a group of galaxy clusters, which forms the largest structures residing in the filaments of the cosmos (e.g., Abell 1958; Chon et al. 2013; Tully et al. 2014). Although the definition of superclusters is not precise, the size of a supercluster can range from several tens of megaparsecs to more than 100 Mpc (Chon et al. 2013). At high redshift (z > 2), several potential primordial superclusters have been reported (e.g., Ouchi et al. 2005; Dey et al. 2016; Topping et al. 2016; Cucciati et al. 2018; Toshikawa et al. 2020). Especially, in the same CFHTLS D1 field, Toshikawa et al. (2020) recently found evidence of the presence of a primeval supercluster at z ∼ 4.9 within a volume of ∼33 × 12 × 64 Mpc³. Based on follow-up spectroscopic observations of the LBGs selected in Toshikawa et al. (2016), they identified three overdense structures with a redshift separation Δz ∼ 0.05 between each density peak. They argued that these structures will evolve independently and become part of a supercluster by z = 0. These studies suggested that primordial superclusters appear at high redshift in parallel with the formation of its cluster/group components.

In this work, the end-to-end distance between the two structures is ∼40 Mpc, comparable to that of the nearby superclusters while too large for a typical protocluster (∼20 Mpc). The total mass of the two structures is $\sim {10}^{15}{M}_{\odot }$ each, as estimated in Shi et al. (2019a). The mass of the LAE structure was calculated using the observed galaxy overdensity and its enclosed volume, while for the mass of the LBG structure we used simulation to infer its intrinsic overdensity along with the assumed volume to get an estimate. Region A has five spectroscopically confirmed LBGs at z = 3.13 identified in Toshikawa et al. (2016) (green triangles in Figure 5), while region B is dominated by a large population of LAEs at z = 3.132 ± 0.023. Considering their similar mass and redshift but large transverse separation, it is likely that regions A and B will grow independently into two separate massive clusters as part of a supercluster by z = 0. Only future spectroscopy in these regions can elucidate the true underlying large-scale structure.

Above we have selected the protocluster galaxy candidates using the available optical-IR (OIR) data. Apart from these OIR sources, dense protocluster environments are often found to host powerful radio galaxies or X-ray luminous AGNs (e.g., Venemans et al. 2005; Overzier et al. 2006; Miley & de Breuck 2008; Digby-North et al. 2010; Hayashi et al. 2012; Kubo et al. 2013; Cooke et al. 2014; Krishnan et al. 2017), thus it is interesting to search for these rare sources in our protocluster to look for a sign of enhanced AGN activities.

First, we crossmatch our photo-z sources with the new X-ray Multi-Mirror Mission (XMM)-Newton point-source catalog from from the XMM-SERVS survey (Chen et al. 2018). Their catalog has 5242 sources detected in the soft (0.5–2 keV), hard (2–10 keV), and full (0.5–10 keV) bands, which reaches a flux limit of 1.7 × 10⁻¹⁵, 1.3 × 10⁻¹⁴, and 6.5 × 10⁻¹⁵ erg cm⁻² s⁻¹, respectively. Using a matching radius of 1.5'', no counterpart in our photo-z sample is found. It is possible that some faint X-ray sources in our sample are simply missed by their detection, as the X-ray sources in the famous SSA22 protocluster from the Chandra catalog (Lehmer et al. 2009) have an average value of 1.6 × 10⁻¹⁵ erg cm⁻² s⁻¹ for the full band, which lies well below the sensitivity of the XMM-SERVS survey. In order to further investigate the potential X-ray signals in our protocluster field, future deep X-ray surveys are needed.

Second, we also search for radio counterparts in our photo-z sample, using the publicly available radio catalog obtained from the Very Large Array at 1.4 GHz covering our field (Bondi et al. 2003). The catalog contains radio sources down to a 5σ depth of ∼0.08 mJy. We find four counterparts in our photo-z sample within a 1.5'' search radius whose total flux densities are in the range of 0.09–3.30 mJy. However, none of these sources is in the overdense protocluster region (A or B). Therefore, our protoclusters may generally lack luminous (≳0.1 mJy) radio sources.

Last, we further investigate the BCG candidate found in our previous study. In Shi et al. (2019a), we discovered an ultra-massive galaxy G411155 that lies very close to the spectroscopic sources in region A (see Figure 5). G411155 is the brightest source in our LBG catalog and also the reddest. Our preliminary SED-fitting result using the Bielby et al. (2012) and Lonsdale et al. (2003) catalogs suggested that this galaxy is dominated by a dust-obscured AGN and is in a phase of intense star formation.

In this work, using our improved PSF-matched photometry from optical to IR, we revisit the physical properties of G411155. Our photometry shows that this galaxy has a K_S magnitude of 21.07, consistent with that of the Bielby et al. (2012) catalog. We obtain J − K_S = 2.15 for this galaxy from our photometry, which is larger than that of Bielby et al. (2012) (J − K_S = 1.92). This extremely red color places G411155 further into the category of hyper extremely red objects (HEROs) (J − K_S > 2.1) (Totani et al. 2001), which are thought to be primordial elliptical galaxies that are still in the phase of a dusty starburst. Our updated SED fitting on this source yields a stellar mass of $1.0\times {10}^{11}\,{M}_{\odot }$ with an SFR of $\sim 123\,{M}_{\odot }$ yr⁻¹. The age of G411155 is ∼500 Myr, which is older than previous estimate (∼200 Myr). SED fitting also suggests that 80% of its IR luminosity is dominated by a dust-obscured AGN. In the entire field of this work (1,156 arcmin²), G411155 is the only object that meets the HERO selection criterion with a mass $\geqslant {10}^{11}{M}_{\odot }$ and an SFR $\gt 100\,{M}_{\odot }$ yr⁻¹. We conclude that G411155 is a rare source in the protocluster region and even in the field. It is likely that we are witnessing the formation of a BCG in the protocluster region A. Future follow-up spectroscopic observations using telescopes such as Keck/Multi-Object Spectrometer for Infra-Red Exploration (MOSFIRE) or the James Webb Space Telescope (JWST) are needed to confirm this BCG and provide us with more information of its properties.