Spectroscopic Confirmation of a Coma Cluster Progenitor at z ∼ 2.2

To increase the success rate of our spectroscopic observations, we focus on potential targets in the vicinity of the candidate protocluster that are likely emission-line galaxies (e.g., star-forming, starburst, or AGN). This is because detecting emission lines is easier and observationally more efficient than finding absorption features in the stellar continuum, which require longer integration times. Moreover, the strongest absorption features appear around the rest-frame 4000 Å and are then redshifted to the J band at the presumed redshift of the protocluster, a region populated by many atmospheric absorption and emission features.

Hence, as the primary targets in the vicinity of the overdensity, we rely on the narrowband selected Hα emitting candidates from the HiZELS survey (Sobral et al. 2013) in the COSMOS field at z ∼ 2.23. These are detected as excess color in the UKIRT/WFCAM and VLT/HAWK-I narrowband K filters (centered at λ ∼ 2.12–2.13 μm with an FWHM Δλ∼200–300 Å, corresponding to a redshift width of Δz ∼ 0.03–0.04 centered at z ∼ 2.23–2.24) relative to the broadband K filter. To minimize contamination from other emission lines, if available, a combination of double-line detections (in both narrowband K and H and/or K and J), broadband color–color selections (Z − K versus B − Z and B − R versus U − B), and photometric redshift cuts (1.7 < z_phot < 2.8) were also implemented. This primary target list is complete down to an Hα flux of ≳1 × 10⁻¹⁷ erg s⁻¹ cm⁻², rest-frame EW(Hα+[N ii]) ≥25 Å, observed SFR ≳3 M_⊙ yr⁻¹ (Chabrier initial mass function), and stellar mass limits of ≳10^9.7 M_⊙ (see Sobral et al. 2013, 2014 for details).

In our design of the multiobject spectroscopic masks, we also added filler objects. They are selected from either the latest COSMOS2015 K_s-band selected or the previous I-band selected photometric redshift catalogs (Ilbert et al. 2009; Laigle et al. 2016). The fillers are selected to be in the vicinity of the overdensity, classified as star-forming galaxies (to increase their detection rate) based on their rest-frame NUV − r versus r − J colors (Ilbert et al. 2013), with their photometric redshift in the range 1.7 < z_phot < 2.8. Given their selection, some of the fillers may belong to the potential protocluster as well.

The observations were conducted on 2018 December 8 and 2019 January 13–15 with Keck I/MOSFIRE NIR multiobject spectrograph under clear conditions with the average seeing of ∼05–06 in December and ∼03–04 in January. Given the expected redshift of the structure, we perform observations in both K (∼1.92–2.40 μm) and H (∼1.47–1.81 μm) bands to cover emission lines that can later be used to measure the SFR (Hα or Hβ), nebular extinction (Hβ and Hα), gas-phase metallicity ([N ii]λ6549, [N ii]λ6583, and Hα), electron density ([S ii]λλ6717,6731 doublet), source of ionization (BPT diagram), and ionization state of the gas ([O iii]λ4959, [O iii]λ5007, and Hβ) for galaxies.

We designed two masks in the vicinity of the protocluster candidate (Figure 1). They were designed in such a way to maximize the number of primary targets. The masks contained unique sources except for one source that would later be used to estimate systematics. In total, we placed 30 unique primary targets and 9 fillers on the masks.

A 2MASS star per mask was used to estimate the observing conditions, such as the seeing and the spatial profile of point sources. Using an ABBA dithering pattern, we observed each mask in each filter for a total exposure time of ∼72–96 minutes with a midpoint airmass of ∼1.0–1.3. Using sky lines, we estimate an FWHM observed spectral resolution of ∼4.5 Å and ∼6 Å in H and K bands, respectively, with the slit width of 07. These correspond to R ∼ 3600 and δz ∼ 0.0003.

We used the MOSFIRE DRP to reduce the data. The reduction involves flat-fielding, cosmic-ray removal, sky subtraction, and vacuum wavelength calibration on a slit-by-slit basis. The outputs are the 2D spectra and their uncertainties. We extract the 1D spectrum and its associated error using the optimal extraction algorithm of Horne (1986). This is done by weighted summing of fluxes in an optimized window around the 2D spectrum, where the weights incorporate both the flux uncertainties and the spatial extent of the 2D spectrum (spatial profile). To determine the optimized window, we use the spatial profile of each source. To extract the spatial profile, we collapse the 2D spectrum of each source along the wavelength direction in the vicinity of bright, high signal-to-noise ratio (S/N) features and then fit a Gaussian function to the profile. We choose the optimized window as ±3× the standard deviation of the spatial profile around its center. If determining the spatial profile fails because of, e.g., faint, low S/N spectrum, we instead rely on the spatial profile of our 2MASS star. In a few cases (e.g., nearby merging systems) where determining the optimized window is tricky, we instead extract the 1D spectra in a boxcar window wide enough to fully cover all the features (e.g., Figure 2 second example). Finally, for all the sources, we visually check the extraction window to make sure that the fluxes are fully measured. Figure 2 shows some example 2D and their extracted 1D spectra.

Zoom In Zoom Out Reset image size

Table 1 lists the extracted redshifts for our spectroscopic sample in the two masks, as well as the coordinate, HiZELS ID (for primary sources), K_s magnitude, the COSMOS ID of each source (based on a match with the COSMOS2015 catalog with a 1'' radius), and whether a source is a primary target, a filler, a serendipitous detection, a potential merger, or a field galaxy. We report a secure redshift for galaxies that have at least two significant (S/N ≥ 3) emission lines. The reported redshift is the average redshift that we obtain based on the peak of all the available emission lines for each source (mostly Hα and [O iii]λ5007). For sources that show signs of mergers in their spectra and/or in their images (commented as "merger" in Table 1), the average redshift of different components is given.

To check for systematics in redshifts for objects on different masks, one object is observed twice (mask1-6* and mask2-14*). The extracted redshift difference for this source is ∼0.0003, similar to the resolution of δz ∼ 0.0003. Another primary object is also observed twice, with a serendipitous detection in the other mask (mask2-10* and mask1-91*). The extracted redshift difference for this source is zero. To check for systematics in obtaining redshifts in different bands (H and K), we compare redshifts obtained based on emission lines in each individual band (if available). The absolute difference is in the range Δz(HK) = 0.00009–0.00218 with a median value of 0.00029, similar to the redshift resolution of δz ∼ 0.0003. To further check the reliability of redshifts, for objects whose emission lines can be fitted with a single Gaussian function, we also determine redshift by fitting a Gaussian. In all cases, the extracted redshifts are within ∼0.0003 of what we originally determined.

Out of 30 unique primary targets (commented as "primary" in Table 1), 29 yield secure redshifts at z ∼ 2.23, showing the robustness of narrowband selection (when combined with further photometric information) in tracing the LSS at high redshift. This also shows that with modest spectroscopic observations (∼1–2 hr), true high-z clusters can be efficiently confirmed. We also find some fillers and serendipitous detections with spectroscopic redshifts in the vicinity of the protocluster.

In the vicinity of the protocluster (150.12 < R.A. (deg) < 150.28, +1.92 < decl. (deg) < +2.08, 2.21 < z < 2.25), we find 12 sources with spectroscopic redshift measurements from the zCOSMOS-deep survey (S. Lilly et al. 2020, in preparation, also see Lilly et al. 2009). We consider these as potential cluster members in addition to our observations. In Table 1, we denote these extra sources by the label "ancillary."

To select the protocluster members, we first determine the mean redshift and standard deviation of all unique galaxies (primary, filler, serendipitous, and ancillary). Sources that are within three standard deviations of the mean redshift are then used to determine the new mean redshift and standard deviation. We iteratively repeat this process until a final mean redshift (z_mean) and standard deviation (σ_z) is obtained. Only three galaxies (commented as "field" in Table 1) do not pass the selection criterion. With the remaining 47 galaxies (35 from our observation and 12 from ancillary data), we estimate the mean redshift, line-of-sight dispersion in redshift space, and line-of-sight velocity dispersion (σ_los = cσ_z/(1+z) where c is the speed of light) as z_mean = 2.23224 ± 0.00101, σ_z = 0.00696 ± 0.00074, and σ_los = 645 ± 69 km s⁻¹, respectively. The uncertainties are estimated using the bootstrap method with 10,000 resamples. If we only rely on the primary sources (29 galaxies), we obtain z_mean(primary) = 2.23321 ± 0.00113, σ_z(primary) = 0.00615 ± 0.00073, and σ_los(primary) = 570 ± 67 km s⁻¹, consistent with measurements using all the galaxies.

To investigate the role of a small sample size on the results, following Yuan et al. (2014), we randomly select only 10 galaxies from our 47 members and recalculate the velocity dispersion. We estimate the new bootstrapped velocity dispersion as σ_los(bootstrap) = 589 ± 149 km s⁻¹, consistent with what we found using the full sample, but with larger uncertainties. Figure 3 shows the redshift distribution, mean redshift, line-of-sight velocity distribution with respect to the mean redshift, and σ_los boundaries for our member galaxies.

Zoom In Zoom Out Reset image size

We consider the centroid of the selected protocluster members as the protocluster center at R.A. = 150.197509 (deg) and decl. = +2.003213 (deg). The centroid is defined as the arithmetic mean of the Cartesian unit vectors representing the protocluster members. For a 2D Gaussian distribution, ∼40% of the weight of the distribution is within one standard deviation. Hence, we use the projected radius from the protocluster center that contains 40% of the members as a proxy for the typical radius of the core of the protocluster and estimate it to be R_proj = 0.75 ± 0.11 Mpc. Using only primary sources, we obtain R.A. (primary) = 150.208397 (deg), decl.(primary) = +2.000796 (deg), and R_proj(primary) = 0.65 ± 0.13 Mpc. Figure 4(A) shows the spatial distribution of the members. In Figure 4(B), they are color-coded by the line-of-sight velocities relative to the mean redshift of the protocluster. We find that 51(87)% of members are within 1(2) Mpc from the protocluster center.

Zoom In Zoom Out Reset image size

The match to the COSMOS2015 catalog shows that three of the members, mask1-1, mask1-15, and mask1-16 have Chandra X-ray detections (Elvis et al. 2009; Civano et al. 2016; Marchesi et al. 2016). This comprises 6.8% ± 3.7% (6.9% ± 4.9%) of the members (primary members), a factor of ∼4 larger than the overall fraction of X-ray detected Hα emitters in the HiZELS/COSMOS field at z = 2.23 (Calhau et al. 2017). All three have broad emission lines, indicative of their AGN nature and they are all Lyα emitters as well (Matthee et al. 2016; Sobral et al. 2017). The enhanced fraction of X-ray detected AGN in the protocluster relative to the field is in good agreement with Lehmer et al. (2013). mask1-16 also has a VLA 20 cm radio detection (Schinnerer et al. 2010). These indicate that highly rare and active systems, such as extreme X-ray sources and radio galaxies trace dense environments at high-z, further supporting the dense nature of the protocluster. A detailed analysis of the AGN fraction will be presented in a following paper.

One major difference between protoclusters and clusters, as discussed in, e.g., Diener et al. (2015) and Wang et al. (2016), is that protoclusters are not yet fully virialized. Hence, for such nonvirialized systems, the velocity dispersion is mainly an indicator of the dynamical state of the system rather than the halo mass. Therefore, any estimation of the dynamical mass based on the velocity dispersion for nonvirialized systems should be considered as order-of-magnitude estimates and should be used with caution.

If we assume that the protocluster is virialized (see Section 4.4) and σ_3d and R_proj are the total velocity dispersion and characteristic radius of the protocluster's core, then we can estimate its virial mass from the virial theorem as M_vir =${R}_{\mathrm{proj}}{\sigma }_{3{\rm{d}}}^{2}$/G, where G is the gravitational constant. Assuming a spherical symmetry, ${\sigma }_{3{\rm{d}}}^{2}$ = 3${\sigma }_{\mathrm{los}}^{2}$. Substituting R_proj and σ_los into the equation gives M_vir = (3${R}_{\mathrm{proj}}{\sigma }_{\mathrm{los}}^{2}$/G) = (2.2 ± 0.6) × ${10}^{14}{M}_{\odot }$. With primary sources, we obtain M_vir(primary) = (1.5 ± 0.5) × 10¹⁴M_⊙.

We can alternatively estimate the virial mass if we assume that the virial theorem applies to the protocluster and the halo of the protocluster is a spherical region within which the average density is 200ρ_c(z), where ρ_c(z) is the critical density of the universe at redshift of z (Navarro et al. 1997). Then, we can express the virial mass (M₂₀₀) of the protocluster in terms of its virial radius r₂₀₀ and the critical density as M₂₀₀ = $\tfrac{4\pi }{3}{r}_{200}^{3}200{\rho }_{c}(z)$. The critical density can be expressed in terms of the Hubble parameter (H(z)) as ρ_c(z) = 3H²(z)/(8πG). Assuming a spherical symmetry combined with the virial theorem implies r₂₀₀ = GM₂₀₀/(3${\sigma }_{\mathrm{los}}^{2}$). Therefore, we can express r₂₀₀ and M₂₀₀ as functions of σ_los and H(z) as r₂₀₀ = $\sqrt{3}{\sigma }_{\mathrm{los}}$/(10H(z)) and M₂₀₀ = ${(\sqrt{3}{\sigma }_{\mathrm{los}})}^{3}$/(10 GH(z)) (Carlberg et al. 1997). We estimate r₂₀₀ = 0.49 ± 0.05 Mpc and M₂₀₀ = (1.4 ± 0.5) × 10¹⁴M_⊙. Using only primary sources, r₂₀₀(primary) = 0.43 ± 0.05 Mpc and M₂₀₀(primary) = (1.0 ± 0.3) × 10¹⁴M_⊙. These are in good agreement with R_proj and M_vir found above.

The Spitzer Archival Cluster Survey (SACS) is a comprehensive search for distant galaxy clusters in all Spitzer/IRAC extragalactic pointings available in the mission archive (A. Rettura et al. 2020, in preparation). Using the algorithm described in Rettura et al. (2014), high-redshift clusters are identified as overdensities in the mid-infrared data combined with shallow all-sky optical data. We find a match in their catalog (at a similar redshift), cluster SACS-COSMOS-J100052 + 020018, separated by only ∼12 from our protocluster. The position of their candidate is shown with a multiplication sign in Figure 4. This provides further confirmation for the existence of the detected structure as Rettura et al. use a completely independent approach in finding high-z protoclusters. Based on a relation calibrated in Rettura et al. (2018; see their Equation (6)), they use the Spitzer 4.5 μm richness of their clusters to infer their dynamical mass. This candidate cluster has an estimated mass, log(M₅₀₀/M_⊙) = 14.06 ± 0.25, consistent with our estimate based on the velocity dispersion.

Using simulated clusters, Munari et al. (2013) suggest a scaling relation as M₂₀₀/10¹⁵M_⊙ = (σ_1D/A_1D)^−α/h(z), where A_1D and α are two parameters, σ_1D is the 1D velocity dispersion, and h(z) = H(z)/H₀. According to their Figure 3, A_1D ∼ 1185 ± 30 km s⁻¹ and α ∼ 0.38 ± 0.01 at z = 2 using galaxies as a tracer for the total mass of clusters. With this scaling relation, we obtain M₂₀₀(scaling) = (3.8 ± 0.2) × 10¹⁴M_⊙, a factor of ∼3 larger than M₂₀₀ we found before but within the same order of magnitude.

We note again that we have made a number of assumptions, such as virialization and the spherical symmetry in estimating the dynamical quantities. These assumptions may not be entirely correct, particularly for protoclusters at high redshift as they are likely still forming (see Section 4.4). Therefore, these should be considered as order-of-magnitude estimates of the protocluster mass. In Table 2, we summarize the protocluster CC2.2 characteristics using all the members and primary sources only.

Is the protocluster relaxed and fully virialized by the time of observation (z ∼ 2.23)? The redshift distribution is not symmetrically Gaussian (skewness = −0.5262, although the difference from a normal distribution is at <1.6σ significance level) and the line-of-sight velocities with respect to the mean redshift are not fully symmetric (Figure 3), indicating that the structure is still in the assembly process. As shown in Figure 1, the presence of other potential overdensities and filamentary-like structures in the vicinity of the protocluster further suggests that the structure is likely not relaxed at z ∼ 2.23 and still coalescing.

We estimate the dynamical timescale (τ_dyn) of the protocluster. The protocluster could be virialized at z ∼ 2.23 if at least one dynamical timescale (in practice, a few) has elapsed since its formation. We estimate τ_dyn ∼ r_3d/σ_3d where r_3d is the characteristic radius of the protocluster and σ_3d is its total velocity dispersion. If we assume r_3d ∼ R_proj and the spherical symmetry and use the estimated line-of-sight velocity dispersion and R_proj from Section 4.1, we obtain τ_dyn ∼ 0.75 Mpc/($\sqrt{3}\times 645$ km s⁻¹) ∼ 0.6 Gyr. Therefore, if the protocluster was initially formed prior to z ∼ 2.8, it would have had sufficient time to get virialized by the time of observation. Estimating the formation epoch of the protocluster is not straightforward. However, the average age of the stellar populations of its member galaxies, particularly the quiescent systems can place robust constraints on its formation time. By selection, quiescent galaxies are currently missing in our spectroscopic observation. However, future deep follow-up spectroscopic observations of potential passive galaxies in the protocluster can put stringent constraints on its formation epoch.

Is the protocluster relaxed by now (z = 0)? To answer this, we investigate the evolution of the protocluster overdensity in the linear regime of a spherical collapse model and compare it with the critical collapse threshold of δ_c = 1.69.¹² Within a redshift slice of Δz ∼ 0.03 (width of the narrowband filter) and a projected 2 Mpc radius circle placed at the center of the protocluster, we find 35 Hα emitters (the original sample from which the primary targets were selected for spectroscopy). The average number of Hα emitters in the same volume is ∼4.6 (corrected for the effective area of the survey and the enhancement due to the overdensity). Therefore, using narrowband selected Hα emitters, the galaxy number density enhancement is δ_g = $\tfrac{35-4.6}{4.6}$ = 6.6.

Following Steidel et al. (2005), δ_g is related to the mass density enhancement (δ_m) via 1 + bδ_m = C(1+δ_g), where b is the clustering bias and C is a correction term due to the redshift space distortions and is calculated using C = 1 + f − f(1+δ_m)^1/3, where f = Ω_m(z)^0.6. Using f(z = 2.23) = 0.96 and the clustering bias of b = 2.4 for the Hα emitters at z ∼ 2.23 (Geach et al. 2012), we obtain δ_m(z = 2.23) ∼ 1.61. In a spherical collapse model (Mo & White 1996), this is related to a linear matter enhancement of δ_L(z = 2.23) ∼ 0.73 and is expected to grow to δ_L(z = 0) ∼ 1.9 by z = 0. This exceeds the collapse threshold of δ_c = 1.69. Therefore, the protocluster is expected to fully collapse and virialize by now (z = 0). In fact, the linear matter enhancement reaches the collapse threshold at z ∼ 0.1, indicating that the protocluster should have been virialized since the past ∼1.0–1.5 Gyr. The collapse threshold at any redshift is approximated as δ_c(z) ≃ 1.69D(z = 0)/D(z), where D(z) is the linear growth function (Percival 2005). At the redshift of the protocluster, δ_c(z = 2.23) ∼ 4.3. This is larger than δ_L(z = 2.23), further indicating that the structure is likely not virialized at z = 2.23.

We estimate the virialized mass of the protocluster at present M_dyn(z = 0) through M_dyn(z = 0) = ρ_m(V_obs/C)(1 + δ_m), where ρ_m is the mean comoving density, V_obs is the observed comoving volume of the structure, and C is the correction term introduced above (Steidel et al. 1998). The Hα emitter candidates are dominated by those selected in the UKIRT/WFCAM narrowband K filter (Sobral et al. 2013). Assuming a tophat shape for the filter corresponds to a redshift width of Δz ∼ 0.032 centered at z ∼ 2.23. The corresponding comoving radial width (Δχ) is then ∼42 Mpc. This leads to the comoving V_obs ∼ 5500 Mpc³ for a cylinder of width Δχ and a projected physical radius of 2 Mpc at z ∼ 2.23. Given δ_m(z = 2.23) ∼ 1.61 and C = 0.64, we estimate M_dyn(z = 0) ∼ 9.2 × 10¹⁴M_⊙. Therefore, the protocluster is likely the progenitor of a Coma-type cluster at z = 0. Simulations of Chiang et al. (2013) show that at z > 2, the progenitors of a Coma-type cluster traced by SFR >1 M_⊙ yr⁻¹ galaxies are expected to have a galaxy density enhancement of δ_g ∼ ${5.5}_{-0.8}^{1.5}$ probed over 15³ ∼ 3500 Mpc³ comoving volumes. These values are in rough agreement with our measurements, indicating that our protocluster is expected to evolve into a ∼${10}^{15}{M}_{\odot }$ Coma-type cluster at z = 0.

The comoving volume associated with Hα emitter candidates in the HiZELS/COSMOS field is ∼5.48 × 10⁵ Mpc³ (Sobral et al. 2013). Given the detection of one protocluster in this volume, we estimate a comoving space and mass density of ∼1.8 × 10⁻⁶ Mpc⁻³ and ∼(1.8–3.6) × 10⁸M_⊙ Mpc⁻³ for a M_dyn ∼ (1–2) × 10¹⁴M_⊙ protocluster at z ∼ 2. However, we note that Poisson uncertainties are as large as the reported values. With the Poisson uncertainty, the space density of the protocluster is ≲3.6 × 10⁻⁶ Mpc⁻³. The halo mass function of Bocquet et al. (2016) predicts a space density of ∼1–2 × 10⁻⁷ Mpc⁻³ for a M₂₀₀ ∼ 10¹⁴M_⊙ halo at z = 2, a factor of ∼10 smaller than our estimate, but consistent with it given the large Poisson uncertainty in our measurement. Moreover, for a sample of similarly selected Hα emitters in the UDS (Sobral et al. 2013) and Boötes (Matthee et al. 2017) fields at z ∼ 2.2 with comoving volumes of ∼2.24 × 10⁵ Mpc³ and ∼2.7 × 10⁵ Mpc³, respectively, no overdensity of Hα emitters is found. This increases the effective volume and subsequently decreases the space density of our protocluster, making our measurement more consistent with the halo mass function predictions.