Hidden in Plain Sight: A Massive, Dusty Starburst in a Galaxy Protocluster at z = 5.7 in the COSMOS Field

The ALMA data containing the [C ii] line for CRLE were first presented by Capak et al. (2015) as Cycle 1 observations of HZ10. A subsequent reprocessing of those data, together with a description of part of the Cycle 3 [N ii] observations, was described in detail by Pavesi et al. (2016).

The Cycle 1 observations of [C ii] were taken on 2013 November 16 in band 7 as part of a larger project (ID: 2012.1.00523.S; Capak et al. 2015). The pointing resulted in 56 minutes on-source time with 25 usable antennas. The primary beam attenuation factor at the position of CRLE is ∼3. The correlator was set up to target the expected frequency of the [C ii] line in HZ10 at 284.835 GHz and provide continuous coverage of the continuum emission in adjacent spectral windows (centered at ∼2, 12, and 14 GHz above) with channels of 15.6 MHz width in time division mode (TDM). The synthesized beam size is approximately 06 × 05 when adopting natural baseline weighting. We refer to Pavesi et al. (2016) for a complete description of the observations and imaging product.

Cycle 3 observations of [N ii] 205 μm targeting HZ10 and covering CRLE were taken on 2016 January 1 and 5 in band 6 as part of two separate programs (2015.1.00928.S and 2015.1.00388.S; PIs: Pavesi and Lu, respectively) with one track each in a compact configuration (max. baseline ∼300 m). The two sets of observations resulted in 50 and 35 minutes on source with ∼41–45 usable 12 m antennas under good weather conditions at 1.3 mm. The first set of observations was previously described by Pavesi et al. (2016). For the second set of observations, the nearby radio quasar J0948+0022 was observed regularly for amplitude and phase gain calibration, J1058+0133 was observed for bandpass calibration, and Callisto was used for flux calibration. We estimate the overall accuracy of the flux calibration to be within ∼10%. The correlator was set up to cover two spectral windows of 1.875 GHz bandwidth each at 15.6 MHz (∼20 km s⁻¹) resolution (dual polarization) in TDM in each sideband.

We used the Common Astronomy Software Application ( casa) version 4.5 for data reduction and analysis. We combined data from all observations and produced all images with the clean algorithm, using natural weighting for maximal sensitivity. Imaging these [N ii] data results in a synthesized beam size of 16 × 12 at the redshifted [N ii] frequency of CRLE and in the continuum map. The rms noise in the phase center is ∼0.14 mJy beam⁻¹ in a 44 km s⁻¹ wide channel. The final rms noise when averaging over the line-free spectral windows (i.e., over a total 7.1 GHz of bandwidth) is ∼18 μJy beam⁻¹. The primary beam attenuation factor at the position of CRLE is ∼1.8.

We observed the CO(2–1) transition line in CRLE (ν_rest =230.538 GHz, redshifted to ∼34.58 GHz at z ∼ 5.667), using NSF's Karl G. Jansky Very Large Array (VLA) in the Ka band (project ID: 17A-011, PI: Pavesi). Observations were carried out between 2017 March 4 and April 6, with 27 antennas in the most compact array configuration (D; max. baseline ∼950 m) under good to moderate weather conditions at 35 GHz (precipitable water vapor columns of 3–6 mm) for eight sessions. The pointing for these observations was centered on HZ10. The primary beam attenuation factor at the position of CRLE is ∼1.08. The combination of all data results in a total on-source time of 19.8 hr.

The nearby radio quasar J1041+0610 was observed regularly for amplitude and phase gain calibration. Also, 3C286 was observed for bandpass and flux calibration. We estimate the overall accuracy of the flux calibration to be within ∼10%, since the phase calibrator flux was measured to be constant within <5% across all sessions.

The VLA correlator was utilized in 8 bit mode for maximum sensitivity. In the first three sessions, the correlator was set up with two intermediate-frequency bands (IFs) of eight 128 MHz-wide spectral windows each, centering one IF on the redshifted CO(2–1) line frequency in HZ10 and the other contiguously below to provide additional continuum sensitivity. The lower IF was moved in the remaining five sessions by centering it on the CO(2–1) line in CRLE (which otherwise fell onto a subband gap). While these overlapping sidebands limit the simultaneous bandwidth and hence the continuum sensitivity, they provide uninterrupted coverage of the CO(2–1) line in both galaxies. The channels in all spectral windows were chosen to provide 1 MHz (∼9 km s⁻¹) resolution (dual polarization) in each IF, to obtain a simultaneous bandwidth of 2.048 GHz for the first three sessions and 1.349 GHz for the remaining five sessions. Because of overlapping spectral coverage, the measurements in the two IFs are not independent. Therefore, we never combine their data but rather only use the line IF for the analysis of CRLE.

We used casa version 4.7 for data reduction and analysis. We calibrated the visibilities using the scripted VLA pipeline, supplemented by manual flagging through inspection of standard visibility plots. We combined data from all observations and imaged them with the clean algorithm, using natural weighting for maximal sensitivity. The imaging of the CO(2–1) data results in a synthesized beam size of 27 × 23 at the redshifted CO(2–1) frequency and in the continuum map. The rms noise in the phase center is ∼45 μJy beam⁻¹ in a 35 km s⁻¹ wide channel. The final rms noise when averaging over the line-free 1.92 GHz of bandwidth is ∼2.7 μJy beam⁻¹.

We detect [C ii], [N ii] 205 μm, and CO(2–1) line emission and the adjacent continuum toward CRLE at high significance (>8σ; Figure 1), which provides an unambiguous redshift identification. Although a foreground galaxy (z_phot ∼ 0.3) is located along the line of sight to CRLE, potentially causing gravitational lensing of its emission, in Appendix A we constrain the likely magnitude of this effect to be minor (<10% flux magnification and negligible spatial distortion). Therefore, we simply extract aperture spectra and fit double Gaussians to the spectral profiles to measure the intrinsic line properties (Figure 2). We report the line properties in Table 1, while the continuum fluxes are listed together with the archival photometric measurements in Table 2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 1.  Measured Line Properties of CRLE

	[C ii]	[N ii]	CO(2–1)
Component 1
νobs (GHz)	284.892 ± 0.013	219.0298a	34.5604 ± 0.0017
Speak (mJy)	16 ± 3	1.2 ± 0.2	0.49 ± 0.04
FWHM (km s−1)	280 ± 40	280a	290 ± 30
I (Jy km s−1)	4.5 ± 0.9	0.33 ± 0.07	0.14 ± 0.02
Component 2
νobs (GHz)	285.27 ± 0.07	219.3204a	34.613 ± 0.004
Speak (mJy)	9.6 ± 0.9	0.65 ± 0.13	0.25 ± 0.03
FWHM (km s−1)	610 ± 140	610a	420 ± 90
I (Jy km s−1)	6.3 ± 1.5	0.40 ± 0.13	0.11 ± 0.03
Total
Mean redshift	5.6666 ± 0.0008
I (Jy km s−1)	10.8 ± 1.2	0.73 ± 0.15	0.26 ± 0.02
L (107 L⊙)	930 ± 100	49 ± 10	2.7 ± 0.2
Deconvolved size (FWHM)	(063 ± 003) × (040 ± 005)	(098 ± 018)	⋯
Physical size (kpc2)	(3.7 ± 0.2) × (2.4 ± 0.3)	5.8 ± 1.1	⋯
L' (1010 K km s−1 pc2)	4.3 ± 0.5	0.49 ± 0.10	7.0 ± 0.5

Notes. We produced integrated line maps over the line FWHM and used casa to fit a 2D Gaussian model. Then, we extracted aperture spectra utilizing the FWHM of the spatial Gaussian model, correcting the total flux by a factor of 2 to account for the flux outside the aperture. The deconvolved emission size was measured from Gaussian fitting to the visibilities.

^aFixed to the parameter values derived from the [C ii] line.

Download table as:  ASCIITypeset image

We measure the deconvolved [C ii] spatial FWHM size from uv-plane modeling (see Appendix C) to be 046 ± 008, which corresponds to 2.7 ± 0.5 kpc at z = 5.667. Using an isotropic virial estimator (Engel et al. 2010) and assuming a [C ii] single-Gaussian fit line FWHM of 640 ± 60 km s⁻¹ (which is also compatible with the broad velocity component), we derive a dynamical mass of (1.5 ± 0.4) × 10¹¹ M_⊙.¹⁰

The [N ii] emission is only slightly resolved. Using the casa task uvmodelfit, we measure a deconvolved FWHM major axis size of 098 ± 018, corresponding to 5.8 ± 1.1 kpc. We use the same technique to fit the size of the [C ii] emission to provide an accurate comparison to the [N ii] emission size, obtaining a result that is compatible with our more sophisticated uv-plane modeling. The [N ii] line emission appears to be marginally more extended (formally by a factor of 2.1 ± 0.5) than the [C ii] emission, but higher resolution, and higher signal-to-noise [N ii] observations are necessary to confirm this finding. In particular, a manual inspection of the uv-radial profile of the [N ii] line visibilities appears compatible with the size of the [C ii] emission. Neither the CO line nor the adjacent continuum emission are resolved. We fit the size of the continuum emission at 158 and 205 μm in the uv plane, and they are compatible with the same deconvolved size of (039 ± 001) × (031 ± 002). This implies that the size of the [C ii]-emitting region is more extended than the size of the dust continuum by ∼(30% ± 20%) in linear dimensions, suggesting that the observed dust continuum may not represent the full extent of the star-forming gas distribution (see also, e.g., Riechers et al. 2014a; Chen et al. 2017).

We find a [C ii]/[N ii] line luminosity ratio of 19 ± 4 in CRLE, which is comparable to the line ratio in the z = 5.3 DSFG AzTEC-3 of 22 ± 8 (Pavesi et al. 2016). This line ratio is sensitive to the fraction of [C ii] emission coming from ionized gas, rather than photon-dominated regions (PDRs), because the ionization energy of nitrogen, in contrast to that of carbon, is higher than that of hydrogen. This makes it a tracer of ionized gas only, and the [C ii]/[N ii] ratio is approximately constant in ionized gas (e.g., Oberst et al. 2006; Pavesi et al. 2016). Assuming a line ratio of 3 ± 0.5 in the ionized gas (Díaz-Santos et al. 2017), we calculate a fraction of the [C ii] emission from PDRs of 84% ± 4% for CRLE and 86% ± 5% for AzTEC-3.

CRLE exhibits approximately the same [C ii]/[N ii] ratio as AzTEC-3 (Pavesi et al. 2016). This ratio is compatible with the range observed in local (ultra)luminous infrared galaxies ((U)LIRGs). However, given the higher gas density and SFR surface density in high-redshift DSFGs relative to most local (U)LIRGs, one might expect a lower fraction of the [C ii] emission to come from the diffuse ionized gas, which is traced by [N ii], relative to atomic and molecular PDRs and therefore a higher [C ii]/[N ii] ratio (Pavesi et al. 2016). Therefore, our observed line ratio suggests that high-redshift DSFGs may host a significant reservoir of diffuse ionized gas, which may be more extended than the starbursting gas. The potentially larger size of the [N ii] emitting region in CRLE relative to [C ii] (although with significant uncertainty) would provide support for this interpretation if confirmed. Recent measurements of both [N ii] fine-structure lines at 205 μm (studied here) and 122 μm in local galaxies also support the interpretation of the [N ii] emission predominantly coming from diffuse ionized gas, rather than compact H ii regions (e.g., Herrera-Camus et al. 2016, 2018a, 2018b; Díaz-Santos et al. 2017).

A well-known decreasing trend of the [C ii]/L_IR ratio with infrared luminosity surface density (a proxy for FUV field strength) offers insight into the origin of the [C ii] emission. In particular, a decreasing trend is expected for PDR emission due to the saturation of [C ii] line emission at higher FUV field strengths (e.g., Stacey et al. 1991). This trend is clearly observed in a sample of local DSFGs from the Great Observatories All-sky LIRG Survey (GOALS; Díaz-Santos et al. 2017). However, at the calculated infrared luminosity surface density for CRLE and AzTEC-3, their observed [C ii]/L_IR ratios (∼3–4 × 10⁻⁴) lie significantly above the best fit to the relation for the GOALS sample (Figure 3). This may suggest an additional contribution besides PDRs to the [C ii] emission. If we assume that the PDR contribution to the [C ii] luminosity is uniquely determined by the intensity of the radiation field as traced by the L_IR surface density, a more significant contribution from diffuse ionized gas may be a reason for this excess. Furthermore, the fraction of [C ii] coming from PDR, rather than ionized gas (as traced by the [C ii]/[N ii] ratio), appears to be lower for CRLE and AzTEC-3 for their modeled FIR color, relative to the fitted relation for the GOALS sample (Figure 3). This quantifies the finding that the [C ii]/[N ii] ratio appears to be low for the more extreme (i.e., higher dust temperature; Figure 3) gas conditions observed in these z > 5 DSFGs relative to local starbursts (Pavesi et al. 2016). As such, it appears to point toward a larger contribution from diffuse ionized gas to the [C ii] emission than what is expected from local starbursts. This diffuse ionized gas may take the form of a gas halo, more extended than the star-forming region. We suggest that a potentially larger and more diffuse ionized gas component at z > 5 may be due to freshly accreted inflowing gas, which is expected to play an important role in powering starbursts in the first billion years of cosmic time. Metallicity is unlikely to play a significant role in affecting the [C ii]/[N ii] line ratio, as we constrain α_CO to be low, and the high dust mass for CRLE (see below) likely implies a solar or super-solar metallicity (Bolatto et al. 2013). Nonetheless, an important caveat to our conclusions is that variations in the carbon-to-nitrogen abundance ratio may affect our interpretation of the [N ii] line emission properties.

Zoom In Zoom Out Reset image size

CRLE is located behind a local foreground spiral galaxy with a photometric redshift of ∼0.35 and stellar mass of 3.3 × 10⁹ M_⊙ in the COSMOS2015 catalog (Laigle et al. 2016), which outshines its optical emission. We analyze the HST/WFC3 data targeting HZ10 (Barišić et al. 2017) and measure the NIR fluxes for the foreground galaxy in the Y, J, and H bands, potentially including a contribution from CRLE (Figure 4). We measure the fluxes in the HST images, adopting Kron fluxes as measured by SExtractor (Table 2; Bertin & Arnouts 1996). We then fit the foreground galaxy with Galfit (Peng et al. 2002) and find a faint residual flux to the west of its center. We fix the model to be an exponential light profile with an axis ratio of 0.5 and fit for center position, size, and flux. The residual flux in the H-band data corresponds to only ∼1%–4% of the total emission, which may be associated with CRLE or an imperfect fit to the central regions of the foreground galaxy (see Appendix B for further details). The foreground galaxy is therefore expected to dominate the total emission at optical-to-NIR wavelengths, at least up to observed-frame ∼2 μm. In order to constrain the contribution of CRLE to the total emission in the near-to-mid-IR, we use Cigale to carry out spectral energy distribution (SED) fitting of the optical and NIR emission (Noll et al. 2009; Serra et al. 2011). We fit the optical and NIR stellar emission with a delayed star formation history (with the ages of the oldest stars ranging from 250 Myr to 12 Gyr and an e-folding time ranging from 50 Myr to 8 Gyr), stellar population synthesis models with metallicity ranging from 1/50 Z_⊙ to Z_⊙ according to Bruzual & Charlot (2003), and a dust attenuation law according to Calzetti (2001). We determine a stellar mass of (3.0 ± 0.5) × 10⁹ M_⊙ and an SFR of ∼1.5 M_⊙ yr⁻¹ for the foreground galaxy, confirming the catalog stellar mass value. We use the best-fitting Cigale models to constrain the FIR luminosity of the foreground galaxy by adopting dust emission models according to Draine & Li (2007). We find approximate predictions for the Herschel/SPIRE fluxes of 3.7, 2.0, and 0.8 mJy in the 250, 350, and 500 μm bands, respectively. These are significantly lower than the measured fluxes in these bands. Therefore, we consider the FIR emission from the foreground galaxy to be subdominant to the emission from CRLE (Table 2).

Zoom In Zoom Out Reset image size

The Spitzer/IRAC 5.8 and 8.0 μm and MIPS 24 μm fluxes are not fit well by the available SED models (Figure 4). We interpret this to be an indication of significant contamination by emission from CRLE toward the longer wavelengths. Although the IRAC1 and IRAC2 fluxes are compatible with expectations, the higher IRAC3 and IRAC4 fluxes are difficult to reproduce with the adopted dust emission models (Draine & Li 2007). The best-fitting SED models were obtained by either including or excluding the MIPS 24 μm flux in the fitting (Figure 4). We find that the best-fitting models to the IRAC points overpredict the 24 μm flux when excluded, and that the best-fitting models that include the 24 μm measurement require strong PAH emission to fit the IRAC4 flux and underpredict the IRAC3 flux at 5.8 μm by at least a factor of ∼2. Even artificially reducing the measurement uncertainty on the IRAC3 flux does not improve the fit to this measurement. While it is unclear if the 24 μm flux is dominated by emission from CRLE or from the foreground galaxy, we interpret these results as suggesting that perhaps a significant fraction of the flux at 5.8 μm may be due to CRLE. A potential caveat to this conclusion may arise in case the foreground galaxy contains an AGN, which could contribute additional flux at these wavelengths, and is presently unconstrained. We use Cigale to derive approximate constraints to the stellar mass in CRLE. If all of the IRAC3 and IRAC4 emission were due to CRLE, the implied stellar mass would be of order ∼1–2 × 10¹¹ M_⊙. This likely provides an upper limit to the actual stellar mass of CRLE. A contribution of ∼25%–50% of the emission may be plausible, yielding a best stellar mass estimate in the range of ∼2.5–10 × 10¹⁰ M_⊙.

In order to model the FIR SED of CRLE as measured by Herschel, ALMA, and JCMT/SCUBA-2, we use a modified blackbody smoothly connected to a mid-IR power law. We use emcee (Foreman-Mackey et al. 2013) to explore the posterior probability distribution of the parameters shown in Figure 5. We employ uniform, nonconstraining priors for the parameters (i.e., flux normalization at rest frame 158 μm, dust temperature T_d, dust emissivity parameter β, rest-frame wavelength at which the optical depth becomes unity λ₀, and mid-IR power-law index α; Table 3). By integrating between 42.5 and 122.5 μm, we find a FIR luminosity of L_FIR = (1.55 ±0.05) × 10¹³ L_⊙. By integrating between 8 and 1000 μm, we derive an IR luminosity of L_IR = (3.2 ± 0.3) × 10¹³ L_⊙. Because the SED of CRLE is not constrained at mid-IR wavelengths, we follow the standard practice of estimating the SFR based on the FIR luminosity only, to provide a comparison to other z > 5 DSFGs (Riechers et al. 2014a). By adopting the standard conversion based on a Chabrier IMF (Carilli & Walter 2013), we infer an SFR of (1550 ± 50) M_⊙ yr⁻¹, with the caveat that the real SFR may be up to a factor of ∼2× higher. Given the high level of dust obscuration in CRLE, we cannot exclude the presence of an obscured AGN. However, we expect that an AGN would only introduce minor contributions to the rest-frame >42.5 μm luminosity given the high dust-obscured SFR of CRLE (see, e.g., Kirkpatrick et al. 2015). As an example, the dust-obscured AGN in the z = 4.05 DSFG GN20 may dominate the mid-IR luminosity, but its contribution to the total IR luminosity appears to be minor (Riechers et al. 2014b). Even for luminous high-redshift quasars, the contribution from AGN-heated hot dust to the FIR luminosity does not appear to exceed ∼20% (Leipski et al. 2013, 2014). By adopting standard (although uncertain) assumptions from Dunne et al. (2003), we find an estimated dust mass of (1.3 ± 0.3) × 10⁹ M_⊙.

Zoom In Zoom Out Reset image size

The dust SED shape for CRLE appears to be intermediate between that of AzTEC-3 (z ∼ 5.3) and ADFS-27 (z ∼ 5.7; Figure 5; Riechers et al. 2014a, 2017). We also compare the measured FIR SED of CRLE to model templates of selected low- and high-redshift starbursts spanning a range of physical conditions. A comparison to the template for local starbursts Arp 220 and M82 shows that CRLE resembles M82 more closely, suggesting a comparatively low dust optical depth¹¹ (Silva et al. 1998). A comparison to the ultraluminous z = 6.34 DSFG HFLS3 shows that CRLE appears to have a significantly lower dust temperature (Riechers et al. 2013; Ivison et al. 2016), as evidenced by the longer wavelength of the peak emission, while at the same time displaying a warmer dust temperature than is observed in the lower-redshift ALESS sources (Swinbank et al. 2014). The SED of CRLE appears to closely resemble that of GN20, an extended z ∼ 4 DSFG (Figure 5; e.g., Carilli et al. 2010; Magdis et al. 2011; Hodge et al. 2012, 2015). Our reference SED templates suggest a potential redshift trend toward hotter dust in higher-redshift DSFGs, as evidenced by the shorter-wavelength SED peak in CRLE relative to the ALESS sample average (mostly in the range z ∼ 2–3; Danielson et al. 2017) and the longer wavelength in comparison to HFLS3 (z = 6.34; Riechers et al. 2013; Faisst et al. 2017). Heating from a warmer cosmic microwave background (CMB) may partly contribute to this tentative dust temperature trend (e.g., da Cunha et al. 2013). However, selection effects may also be partly responsible for this trend, since most of these DSFGs were selected at fixed observed-frame wavelengths.

CRLE is detected at 7σ significance in 3 GHz continuum emission (24.3 ± 3.8 μJy; Smolčić et al. 2017b). The emission is not resolved at a synthesized beam size of 075, and it is aligned with the dust continuum emission (Figure 4). CRLE is also potentially weakly detected at 1.4 GHz at the 2.5σ level (Schinnerer et al. 2010). We conservatively adopt a 3σ limit of <84 μJy at 1.4 GHz, which is consistent with a spectral index of −0.7. Sensitive low-radio-frequency observations also provide constraining nondetections (K. Tisanić et al. 2018, in preparation). We derive 3σ upper limits of 320 μJy at 325 MHz and 150 μJy at 610 MHz. These imply constraints to the effective radio spectral index to the observed-frame 3 GHz of >−1.16 and >−1.14 for the 325 and 610 MHz limits, respectively.

Adopting the redshift-dependent FIR–radio correlation measured by Delhaize et al. (2017), we can convert our measured 3 GHz flux and the constraints to the 1.4 GHz flux to FIR luminosities. The FIR–radio correlation has been calibrated at a rest-frame frequency of 1.4 GHz, which corresponds to observed-frame 210 MHz for CRLE. This comparison therefore requires significant extrapolation from our measurement at 3 GHz, and it is sensitive to the spectral index. Adopting a radio power-law index of −0.7, as in Delhaize et al. (2017), we use 3 GHz flux and a 1.4 GHz flux upper limit to derive FIR luminosities of 4.0 × 10¹² and <8.0 × 10¹² L_⊙, respectively. This radio-inferred FIR luminosity is significantly lower than our direct measurement. We could reconcile these estimates with an effective spectral index of −1.2 between rest-frame 20 and 1.4 GHz, but this would be in slight tension with our 325 and 610 MHz upper limits. Alternatively, Molnár et al. (2018) suggested that the FIR–radio correlation may not evolve in star-forming, disk-dominated galaxies. If we assume no redshift evolution, the 3 GHz flux and 1.4 GHz flux upper limit would imply FIR luminosities of ∼2.7 × 10¹³ and <5.5 × 10¹³ L_⊙, in agreement with our direct measurement.

However, the observed radio emission may not be well described by a single power-law spectral index down to rest-frame 1.4 GHz. In particular, the analysis by Tabatabaei et al. (2017) suggests that approximately half of the radio flux at this frequency may be due to thermal free–free emission. Under this assumption, and adopting the relationship between thermal radio emission and SFR by Murphy et al. (2011) for an H ii region electron temperature of T_e = 10⁴ K, we would infer an SFR ∼ 4000 M_⊙ yr⁻¹ from the 3 GHz continuum flux, with large uncertainties due to assuming a thermal versus nonthermal fraction.

Low-J CO line luminosities are expected to provide a reliable estimate of the molecular gas mass (e.g., Bolatto et al. 2013). Here we assume a brightness temperature ratio of R₂₁ = 1 between the CO J = 2–1 and 1–0 lines, due to the moderately high inferred dust temperature of CRLE (Table 3).¹² Another effect that may be relevant to the interpretation of the observed CO line flux is contrast against the warmer CMB and the additional gas heating this provides at z > 5 (da Cunha et al. 2013). While the effect this may introduce is difficult to estimate without additional CO excitation constraints, da Cunha et al. (2013) suggested that for warm gas kinetic temperatures and moderately high gas densities, we may expect the observed CO line flux to be suppressed by ∼0.5–0.8 at this redshift. We do not attempt to estimate a correction factor for this effect but include it in the definition of α_CO instead (i.e., a larger CMB correction would imply a higher α_CO). By assuming the dynamical mass to be completely accounted for by molecular gas, we can derive a conservative upper limit on the CO conversion factor of α_CO < 2.1 ± 0.6 M_⊙ (K km s⁻¹ pc²)⁻¹. This conversion factor is conservative, because it does not include the stellar contribution to the dynamical mass. Our stellar mass estimate is too uncertain to provide a useful constraint in this context. Such low conversion factors are typically only observed in highly metal-enriched galaxies, which appears to suggest that metallicity is not a major source of uncertainty in the interpretation of line ratios (Bolatto et al. 2013). Furthermore, from the dust mass estimates presented in Section 4.2, we can adopt a gas-to-dust ratio in order to derive an independent estimate of α_CO. The gas-to-dust ratio may be assumed to follow the same dependence on metallicity as measured at lower redshift. Here we follow Magdis et al. (2011) by adopting a super-solar metallicity, expressed as 12 + log(O/H), between 8.8 and 9.2. This implies a gas-to-dust mass ratio of 35–75, which is consistent with an average value for DSFGs of ∼50 (Santini et al. 2010). This range implies a range of α_CO from 0.65 ± 0.16 to 1.4 ± 0.3 M_⊙ (K km s⁻¹ pc²)⁻¹. These ranges are consistent with expectations from local ULIRGs, for which a conversion factor of α_CO ∼ 0.8 M_⊙ (K km s⁻¹ pc²)⁻¹ was estimated (Downes & Solomon 1998). In the following, we adopt a conversion factor of α_CO = 1 M_⊙ (K km s⁻¹ pc²)⁻¹, which provides a conservative estimate of the molecular gas mass in CRLE.¹³ We thus derive a total molecular gas mass of M_gas = (7.0 ± 0.5) × 10¹⁰ M_⊙, corresponding to ∼50% ± 15% of the dynamical mass estimate. This gas mass estimate is similar to the gas masses of other z > 5 DSFGs, such as AzTEC-3 (∼5.3 × 10¹⁰ M_⊙) and HFLS3 (∼4.5 × 10¹⁰ M_⊙; Riechers et al. 2010, 2013; Cooray et al. 2014). Our inferred α_CO is compatible with model predictions by Vallini et al. (2018) at z ∼ 6, which suggest α_CO = (1.5 ± 0.9) M_⊙ (K km s⁻¹ pc²)⁻¹, despite the subsolar metallicity of their simulated galaxy. They interpret this low conversion factor as the result of the high density and turbulence in the molecular gas of their simulated galaxy. The measured [C ii]/CO(2–1) line ratio in CRLE is also compatible with their model predictions, although we caution that their simulated galaxy targets the "normal" star-forming population with an SFR over an order of magnitude lower than CRLE (Vallini et al. 2018).

We also use single-wavelength dust continuum emission measurements on the Rayleigh–Jeans tail to explore alternative methods to estimate the ISM mass of CRLE. Using the the rest-frame 205 μm and 1.3 mm observations, we follow Scoville et al. (2016, 2017a) to estimate total ISM masses of 1.1 × 10¹² and 8.0 × 10¹¹ M_⊙, respectively. This method was calibrated at a rest-frame wavelength of 850 μm; hence, the latter estimate is expected to be more reliable, since the extrapolation is smaller. We also use rest-frame 500 and 250 μm fluxes in CRLE to infer a gas mass following Groves et al. (2015). The rest-frame 500 and 250 μm fluxes are estimated from our best-fitting modified blackbody model and imply gas masses of 1.3 × 10¹² and 2.4 × 10¹¹ M_⊙, respectively. Therefore, these dust-based methods would imply very large ISM masses, exceeding our dynamical mass estimate and suggesting that hotter dust temperatures may affect these techniques, making them less reliable for this type of high-redshift galaxy (e.g., Scoville et al. 2015).

Adopting the relationship between CO luminosity and SFR (i.e., the star formation law) for local starburst galaxies (Kennicutt 1998; Carilli & Walter 2013), we can convert the measured ${L}_{\mathrm{CO}}^{{\prime} }$ to an IR luminosity of L_IR = 1.3 × 10¹³ L_⊙ with large uncertainties (the scatter of this relationship is a factor of ∼2.5). This is consistent with our measured IR luminosity, although apparently at the low end of the confidence interval. By adopting the fiducial values for the SFR and the molecular gas mass, we derive a gas depletion timescale of 45 ± 4 Myr, which is comparable to estimates for other z > 5 DSFGs, such as AzTEC-3 (∼50 Myr) and HFLS3 (∼36 Myr) (Riechers et al. 2010, 2013, 2014a).

As shown in Section 4.2, the dust SED in CRLE appears to be intermediate between those of other z > 5 DSFGs, such as AzTEC-3 and ADFS-27 (Figure 5; Riechers et al. 2010, 2014a, 2017). A possible interpretation of these differences in the dust properties (e.g., temperature or optical depth) among these galaxies involves the merger stage. In particular, AzTEC-3 appears to be a dispersion-dominated starburst with an exceptionally compact central density profile that is suggestive of a late-stage merger after coalescence (D. Riechers et al. 2018, in preparation). On the other hand, ADFS-27 is an interacting galaxy pair, where the individual galaxies are separated by just 9 kpc and therefore may represent an early merger stage (Riechers et al. 2017). In Appendix C, we carry out dynamical modeling of the [C ii] emission, which suggests that a simple rotating disk model does not provide a good fit to CRLE. This conclusion is also supported by the asymmetric profile visible in the line spectra (Figure 2). We find evidence for a second dynamical component, which may suggest an overall dispersion-dominated system. This evidence appears to suggest an ongoing merger in CRLE with a component separation of 2.0 ± 0.4 kpc (from the [C ii] emission modeling). Therefore, we tentatively interpret the intermediate SED in CRLE as a reflection of its merger stage. However, the SED in CRLE also resembles that in GN20, which shows a coherently rotating massive gas disk and no evidence for a galaxy merger (Hodge et al. 2012).

The close association in the sky and redshift space to HZ10 (z = 5.654, 13'' distance, corresponding to only 77 kpc in projection and ∼580 km s⁻¹ of relative radial velocity) suggests that CRLE may be located in an overdense galaxy environment. We follow the procedure described in Smolčić et al. (2017a) to evaluate a potential galaxy overdensity around CRLE. First, we analyze the small-scale overdensity around the DSFG in the COSMOS2015 catalog, making use of the photometric redshift information, as detailed in Smolčić et al. (2017a). Then, we apply a similar technique to also investigate the galaxy overdensity in the LAE catalog at z ∼ 5.7, which may offer a higher accuracy in the redshift range selected (Murayama et al. 2007).

We do not apply the magnitude cut ${i}^{+}\lt 25.5$ of Smolčić et al. (2017a), because the ${i}^{+}$ magnitude of HZ10 is 26.45, which we know to have a redshift in the correct range. Based on the magnitude of HZ10, we also expect that even massive galaxies at this high redshift are likely to have an i⁺ magnitude below the adopted threshold of Smolčić et al. (2017a). On the other hand, our choice to include fainter galaxies may limit the photometric redshift accuracy. We adopt their choice of redshift binning, which implies a photo-z range of Δz = 0.64 for the catalog "slice." Following Smolčić et al. (2017a), we define a galaxy overdensity parameter by ${\delta }_{g}(r)=\tfrac{{N}_{r}}{{{\rm{\Sigma }}}_{\mathrm{bg}}\,\pi {r}^{2}}-1$, where N_r is the number of galaxies in the catalog within a radius r of the DSFG (including the DSFG itself), and Σ_bg is the mean galaxy surface density in the redshift "slice." We take into account masked regions when evaluating the surface area. We evaluate the overdensity parameter around CRLE at radii of 025 (to capture the overdensity due to HZ10 alone) and 05–3' in steps of 05 in both the photometric redshift and LAE catalogs (Figure 6).

Zoom In Zoom Out Reset image size

In order to evaluate the significance of the measured overdensity parameter values, we follow the procedure by Smolčić et al. (2017a), producing 10 mock random catalogs adopting the same masked regions with the same number of galaxies as the real ones and measuring the overdensity around 1000 randomly placed centers. The rate of chance occurrence of the actual N_r profile (equivalently, of the overdensity parameter) can then be evaluated. We mark significantly overdense radii with squares in Figure 6, adopting the same 5% false-positive rate as Smolčić et al. (2017a).

In the COSMOS2015 photometric redshift catalog, we find that HZ10, due to its very close separation, constitutes a significant overdensity of >50, and that at a radius of 2', there is only a 1% probability of the observed overdensity (seven galaxies; corresponding to δ_g ∼ 2–3) being produced by chance. The chance probability of at least one neighbor within 05 is only 12% (the nearest, HZ10, is at 0216). This chance probability is comparable to that of having at least two neighbors within 105 (i.e., the separation to the next galaxy) or at least three neighbors within 15 (the following galaxy). In the LAE catalog, the overdensity is even more significant, due to the lower contamination from galaxies at incorrect redshifts. We find a significant overdensity at all radii up to 25, with HZ10 already implying an overdensity of >500 and a second LAE within 15 constituting an overdensity of ∼30. The overdensity in the LAE catalog at a radius of 5', including a total of seven galaxies, is also >4.

We also explore the second technique utilized by Smolčić et al. (2017a) in order to study the larger-scale overdensity. However, we are limited to scales of 4'–5' in the LAE catalog and 3' in the photometric redshift catalog, since CRLE is close to the southern edge of the catalog. This method utilizes a Voronoi tessellation analysis (VTA) in order to identify overdense regions and then determines an overdensity center by evaluating a barycenter for the "region." It then evaluates the significance of the overdensity parameter value as a function of radius around this newly found center. The VTA analysis in the case of CRLE indicates that, in both the photometric redshift and LAE catalogs, CRLE and HZ10 are in an overdense region (i.e., above the 80th percentile of a randomized galaxy density distribution) due to the close proximity to the next galaxy neighbors. The overdensity center is evaluated to be approximately the midpoint between CRLE and HZ10. Because the overdensity center is close to CRLE, the overdensity evaluation for this method reproduces similar results to those of our previous analysis.

Four galaxies (all part of the LAE catalog) within 5' have a spectroscopic redshift that place them in the overdensity (5.665, 5.674, 5.688, and HZ10 at 5.659 from Keck/DEIMOS; P. Capak et al. 2018, in preparation). The other two LAEs within this radius have no known spectroscopic redshifts.¹⁴ Of the four LAEs with spectroscopic redshift confirmation, two (including HZ10) are also part of the photometric redshift sample. No other photometric redshift candidates within 5' have a spectroscopic redshift. If we expand the search radius to 17', corresponding to 40 cMpc in the spectroscopic redshift catalog, we find four more galaxies with spectroscopic redshifts between 5.5 and 6, which may potentially be part of the same overdensity (with spectroscopic redshifts of 5.682, 5.728, 5.663, and 5.742).

Here we compare the galaxy overdensity around CRLE, with the cases of AzTEC-3 and other lower-redshift DSFGs (e.g., Walter et al. 2012; Oteo et al. 2018; Smolčić et al. 2017a), with LAE overdensities at z > 5 (Higuchi et al. 2018) and in relation to theoretical expectations (e.g., Chiang et al. 2013, 2017).

The closest analog to the galaxy overdensity around CRLE is the protocluster around AzTEC-3 at z = 5.3, which is also located in the COSMOS field (Riechers et al. 2010, 2014a; Capak et al. 2011). The close proximity of the "normal" LAE/LBG HZ10 to CRLE (∼77 kpc away) is comparable to the close separation between the luminous DSFG AzTEC-3 and the "normal" galaxy LBG-1 (∼95 kpc away; Riechers et al. 2014a). Capak et al. (2011) reported a rich galaxy overdensity around AzTEC-3 with an overdensity parameter of ≳11 within a 2 cMpc radius. We also detect a comparable overdensity of LAEs of δ_g ≳ 10 within a radius of ∼5 cMpc of CRLE. Because of the higher redshift of CRLE (z ∼ 5.667) than AzTEC-3 (z ∼ 5.298), the catalogs contain significantly fewer galaxies. The overdensity is therefore associated with a smaller number of galaxies and hence is subject to larger uncertainties from random fluctuations. The "normal," "companion" galaxies LBG-1 and HZ10 are similar in terms of their stellar masses and UV luminosities, but the FIR luminosity in HZ10 is almost an order of magnitude higher, suggesting that the ISM in HZ10 may by substantially more enriched (Capak et al. 2015; Pavesi et al. 2016). Furthermore, the [C ii]/[N ii] ratio appears to suggest that the metallicity of HZ10 is compatible with local and high-redshift starburst galaxies, while it suggests a lower metallicity for LBG-1 (Pavesi et al. 2016). The overdense environment experienced by HZ10 may be partly responsible for the particularly enriched ISM state in this "normal" galaxy. In particular, it is possible that metal-enriched material that was ejected by CRLE may have been accreted by HZ10. Also, it is likely that the local dark matter overdensity, as suggested by the galaxy overdensity, may be responsible for the early galaxy growth by providing abundant gas fueling to the central regions. However, LBG-1 seems to be located in a similar environment and does not display a comparable level of enrichment. Capak et al. (2011) estimated that the dynamical time for LBG-1 to reach AzTEC-3 may be of order ∼0.5 Gyr, providing several dynamical times for a merger to occur by z ∼ 2. A similar estimate applies to HZ10 and its likely eventual merger with CRLE, which may produce a central cluster galaxy.

Smolčić et al. (2017a) analyzed galaxy overdensities around previously known DSFGs in the COSMOS field at z ∼ 1–5.3 and found an incidence of approximately ∼50% when using methods similar to the ones employed here. They tentatively found a higher occurrence of DSFGs occupying overdense environments at z > 3 than at z < 3. This would be compatible with our finding of a high galaxy overdensity including CRLE at z ∼ 5.7. A higher incidence of overdensities associated with the highest-redshift DSFGs would be consistent with the idea that these massive galaxies may be associated with the highest peaks of the density field, tracing the most massive dark matter halos at early cosmic epochs (e.g., Springel et al. 2005; Li et al. 2007; Overzier et al. 2009; Capak et al. 2011).

Recent observations with the Hyper-Suprime-Cam on the Subaru telescope have yielded a deeper and wider catalog of LAEs in COSMOS at z ∼ 5.7 than available for our analysis (Ouchi et al. 2018; Shibuya et al. 2018). These authors reported numerous LAE overdensities at this redshift, showing that a significant fraction of star-forming galaxies at this epoch may be part of protocluster environments (Higuchi et al. 2018). In particular, these authors report an overdensity, HSC-z6PCC5, in the COSMOS field, with an overdensity parameter of δ ∼ 10. The reported overdensity center is located 44 cMpc away from CRLE and at a redshift of z = 5.686, corresponding to only 9 cMpc of radial separation. It is therefore possible that CRLE and its associated small-scale galaxy overdensity may also be associated with this significantly larger-scale early cosmic structure.

Recent theoretical work suggests that protoclusters may have dominated star formation in the first 2 billion yr of cosmic time (Chiang et al. 2013, 2017). This is due to the fact that the fraction of the cosmic volume occupied by all future (proto)clusters increases by nearly three orders of magnitude from z = 0 to 7. More importantly, most models suggest that early galaxy formation may be dominated by the central regions of the most massive overdensities and that star formation may evolve inside-out to galaxies in lower-density environments (Chiang et al. 2017). These may be crucial predictions of structure formation in the early universe. A quantification of the fraction of star formation in different environments as a function of cosmic time may be an important cosmological probe in the era of wide-area surveys. The physical processes associated with the "central" galaxy forming in a protocluster, which may be a DSFG at least during part of its life, may strongly affect the evolution of such protoclusters, both by enriching the intracluster medium and by providing energy input through winds and radiation.