HST GRISM CONFIRMATION OF TWO z ∼ 2 STRUCTURES FROM THE CLUSTERS AROUND RADIO-LOUD AGN (CARLA) SURVEY

The CARLA sample consists of Spitzer/IRAC channels 1 and 2 (3.6 and 4.5 μm bands, respectively) observations of 420 fields around powerful RLAGN obtained during a 400 hr Warm Spitzer Cycle 7 and 8 snapshot program. Each image covers an area of $5.2\times 5.2$ arcmin² with an original sampling of 1.22 arcsec pix⁻¹ reprocessed to 0.61 arcsec pix⁻¹. The data reach 95% completeness at 22.8 mag and 22.9 mag for the 3.6 μm and 4.5 μm bands, respectively. The program imaged the fields of 209 high-redshift radio galaxies (HzRGs) and 211 radio-loud quasars (RLQs), uniformly selected in radio-luminosity over the redshift range $1.3\lt z\lt 3.2$ (with rest-frame ${L}_{500\mathrm{MHz}}\geqslant {10}^{27.5}$ W Hz⁻¹). The HzRGs (type-2 RLAGN) were selected from the updated compendium of Miley & De Breuck (2008), and the RLQs (type-1 RLAGN) were selected from the Sloan Digital Sky Survey (SDSS, Schneider et al. 2010) and 2dF QSO Redshift Survey (2QZ, Croom et al. 2004). Galaxy cluster candidates were then identified as IRAC color-selected galaxy overdensities in the fields of the targeted RLAGN. We applied the color cut ${([3.6]\mbox{--}[4.5])}_{\mathrm{AB}}\gt -0.1\,\mathrm{mag}$, which reliably identifies $z\gt 1.3$ sources (Papovich 2008). Galaxy cluster candidates are defined as fields showing a $\gt 2\sigma$ overdensity of IRAC color-selected sources compared to the blank field surface density of similarly selected sources, as measured in the Spitzer UKIDSS Ultra-Deep Survey (SpUDS; P.I. Dunlop). We measured the density of IRAC color-selected sources within $1^{\prime}$ radius apertures centered on the RLAGN (∼500 kpc at $z\sim 2$). We identified 178 cluster candidates among the 420 fields. The galaxy density of these cluster candidates strongly peaks toward the position of the targeted RLAGN, suggesting that the clusters are reliable and that the RLAGN reside at the cluster cores. Detailed descriptions of the sample and initial scientific results are presented in Wylezalek et al. (2013, 2014).

Among the 178 CARLA cluster candidates, we selected the 20 richest fields as the most promising targets for additional study. These fields are $5.8\sigma$–$9.0\sigma$ overdense above the mean SpUDS density and the associated RLAGN at the center of each field cover the redshift range of $1.4\lt z\lt 2.8$. Ten of the twenty fields are associated with HzRGs and the other ten are associated with RLQs. Our team was awarded 40 HST orbits in a Cycle 22 WFC3 program to image and obtain spectroscopy of these 20 CARLA fields with the primary goal of spectroscopically confirming the cluster candidates (Program ID: 13740). Each field is visited twice using different orientations to mitigate contamination from overlapping spectra. The first field was observed in 2014 October and the whole program was completed in 2016 April. For each visit, our program obtains 0.5 ks $F140W$ direct imaging and 2 ks G141 slitless grism spectroscopy.

Each image covers a field of view of $2\times 2.3$ arcmin² at a sampling of 0.13 arcsec pix⁻¹. The G141 grism covers the wavelength range of $\lambda =(1.08\mbox{--}1.70)\,\mu {\rm{m}}$ with a throughput $\gt 10 \%$ and was chosen because it samples Hα at $0.65\lt z\lt 1.59$, [O iii] at $1.16\lt z\lt 2.40$, Hβ at $1.22\lt z\lt 2.50$, and [O ii] at $1.90\lt z\lt 3.56$, enabling us to identify strong galaxy features at the redshifts of our cluster candidates ($1.4\lt z\lt 2.8$) and potentially measure dust extinction via the Balmer decrement for $1.22\lt z\lt 1.59$.

2.3.1. HST Observations of CARLA J2039−2514 and CARLA J0800+4029

The field around MRC 2036−254 (HzRG at z = 1.997) and the field around B3 0756+406¹⁴ (RLQ at z = 2.004) were observed in 2014 October and November, respectively, partitioned in two single-orbit visits each, using two different orientations. Total exposure times on the field around MRC 2036−254 (B3 0756+406) reached $4023\,{\rm{s}}$ ($4123\,{\rm{s}}$) in grism mode and $1023\,{\rm{s}}$ ($1073\,{\rm{s}}$) in direct imaging mode. Table 1 lists the observation dates, exposure times, and orientation angles. We refer to the spectroscopically confirmed structures by their CARLA names, CARLA J2039−2514 and CARLA J0800+4029.

Table 1.  HST WFC3 Observations

Target	UT Date	Position Anglea (degree)	$F140W$/G141 Exp. Time (s)
MRC 2036−254	2014 Oct 14	−62	512/2012
	2014 Oct 15	−39	512/2012
B3 0756+406	2014 Nov 03	+125	537/2062
	2014 Nov 07	+162	537/2062

Note.

^aEast of north.

Download table as:  ASCIITypeset image

Each visit was divided into four dithered blocks of exposures, with the direct images taken just after the grism exposures to enable wavelength calibration of the spectra based on source position. We retrieved calibrated and flat-fielded individual exposures (FLT)¹⁵ from MAST,¹⁶  which uses the most up-to-date calibration files for WFC3. These were our primary data products before further reduction and extraction of the spectra.

2.3.2. B3 0756+406 Palomar Spectrum

Because the very broad lines of B3 0756+406 in our HST spectroscopy prevent us from measuring a reliable redshift (see Appendix D), and because of the relatively low signal-to-noise ratio of its SDSS spectrum, we obtained an optical spectrum of the source on UT 2015 February 17 using the Double Spectrograph on the Hale 200'' telescope at Palomar Observatory. The conditions were relatively clear, but not photometric, with $\sim 1\buildrel{\prime\prime}\over{.} 3$ seeing. We observed the target through a 15 slit for two 900 s exposures using the 600 ${\ell }{\mathrm{mm}}^{-1}$ grating on the blue arm of the spectrograph (${\lambda }_{\mathrm{blaze}}=4000$ Å), the 316 ${\ell }{\mathrm{mm}}^{-1}$ grating on the red arm of the spectrograph $({\lambda }_{\mathrm{blaze}}=7500\,\AA )$, and the 5500 Å dichroic. The data were processed using standard techniques within IRAF, and flux calibrated using standard stars from Massey & Gronwall (1990) observed on the same night. The optical spectrum (Figure 1) shows strong and broad emission lines from Lyα, C iv, C iii, and Mg ii at a redshift $z=2.004\pm 0.002$, slightly lower than the redshift z = 2.021 derived from the lower-quality SDSS spectrum. Narrow self-absorption is seen in all but the C iii line.

Zoom In Zoom Out Reset image size

Our grism data contain traces of several hundred objects (see Figure 2), including first and zeroth orders and even additional orders (−1, +2) in the case of bright objects. Therefore, in addition to standard reduction procedures (cosmic-ray rejection, sky subtraction, etc.) spectra need to be carefully extracted. We generally follow the steps presented in the WFC3 IR grism cookbook¹⁷ (v1.3), applying identical methods for both fields.

Zoom In Zoom Out Reset image size

We combine the individual $F140W$ exposures using the aXe software (v2.2.4) to create deep drizzled direct images of the fields. From these images, we perform source extraction using SExtractor (Bertin & Arnouts 1996). In this step, we create two catalogs: a deep catalog cleaned by hand from spurious detections (the master catalog), and a shallow catalog with a more conservative magnitude limit. The latter allows us to subtract the sky background from the grism images in a later step.¹⁸ Since our targets were selected as CARLA overdense fields, we cross-correlate the color-selected IRAC $z\gt 1.3$ candidates with our HST master catalog, even though, in practice, we extract and analyze all HST sources (see details in Appendix A.3).

The master catalog is then projected back onto each individual direct exposure and the background level is subtracted from each grism exposure using up-to-date master sky background (v1.0) and grism mode configuration files (v2.5).¹⁹ Details of these two steps are presented in Appendix A.4.

We then extract individual spectra from each scaled grism frame based on source positions and sizes stored in the individual master catalogs, and stack the spectra of the same observation/orientation. The aXe task axecore determines contamination estimates from neighboring and/or overlapping objects. We use the "Gaussian" method of the qualitative contamination model as a first approximation, adopting a width scale factor of one instead of the standard factor of three, which we found makes the spectral traces in the spatial (cross-dispersion) direction too wide.²⁰ To obtain deep 2D first order spectra of constant dispersion in wavelength and linear spatial sampling, we run the tasks drzprep and axedrizzle. Note that we keep each orientation separate in order to better handle contamination issues. We obtain for each available orientation a deep 2D cutout of the first spectral order of each source as well as a deep 2D cutout containing the traces of its possible contaminants estimated from the "Gaussian" contamination model.

Finally, we manipulate the 2D cutouts to extract our own 1D spectra, flux calibrated using the G141 first order sensitivity function. The software identifies continuum and zeroth orders of possible contaminants on the 2D cutouts.

3.3.1. Imaging Noise Level

We measure the $F140W$ noise level from object-free regions based on the SExtractor segmentation map corresponding to our deep catalog. We use this map to mask all detected sources, and we measure the noise level by randomly placing 5000 non-overlapping 04 diameter apertures²¹ on the unmasked area covered by both orientations. For both fields, the noise values follow a Gaussian distribution with best-fit values given in Table 2. The images of both fields have $1\sigma$ noise levels of $\sim 2.5\times {10}^{-21}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}\,{\AA }^{-1}$, corresponding to ∼26.6 mag (AB) at $5\sigma$.

Table 2.  Noise Levels

Target	$F140W$		G141
	σa	mag${}_{\mathrm{AB}}$ (5σ)	σb
MRC 2036−254	2.7	26.5	⋯
Orient01	⋯	⋯	51
Orient02	⋯	⋯	60
B3 0756+406	2.4	26.7	⋯
Orient01	⋯	⋯	44
Orient02	⋯	⋯	59

Notes.

^aStandard deviation in flux density (10⁻²¹ $\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}\,{\AA }^{-1}$). ^bSame as ^a but at 15000 Å.

Download table as:  ASCIITypeset image

3.3.2. Spectroscopic Noise Level

3.3.3. Line Detection Limit

3.4.1. Emission Line Fitting

We determine the redshift of the sources and emission line fluxes using the python version of mpfit ²² , a Levenberg–Marquardt least-squares minimization fitting procedure. When required (e.g., for all cluster members), we fit the [O iii] doublet ($\lambda 4959,\lambda 5007$) with two Gaussians constrained to have the same width, same redshift, and a 1:3.2 flux ratio (Osterbrock 1989). Simultaneously, we fit the continuum using a third order polynomial and include a third Gaussian corresponding to Hβ, constrained to have the same redshift as [O iii]. We substitute [O iii] with, or add to the model, other Gaussians to account for Hα or [O ii] when required (e.g., for non-cluster members). Table 3 shows the redshifts and emission line fluxes of our cluster members, and Table 4 presents the redshifts of non-cluster members.

Table 3.  Confirmed Cluster Members

Field	Object	R.A. (hh:mm:ss.ss)	Decl. (dd:mm:ss.s)	$F140W$ (AB)	[3.6] (AB)	[4.5] (AB)	f[O iii]a	${f}_{{\rm{H}}\beta }$ a	z	Qualityb	Remarksc
CARLA J2039−2514	119	20:39:28.51	−25:14:31.2	22.35 ± 0.02	20.87 ± 0.02	20.88 ± 0.02	6.5 ± 2.9	⋯	1.987 ± 0.007	${{\rm{B}}}^{+}$	SF
	174	20:39:22.61	−25:14:16.0	22.87 ± 0.03	22.09 ± 0.06	22.13 ± 0.05	5.1 ± 1.5	⋯	2.000 ± 0.007	${{\rm{B}}}^{+}$	SF
	281	20:39:22.14	−25:15:08.6	23.98 ± 0.06	⋯	⋯	8.4 ± 3.2	⋯	2.002 ± 0.008	${{\rm{B}}}^{-}$	SF
	306	20:39:24.49	−25:14:30.7	20.94 ± 0.01	19.99 ± 0.01	19.79 ± 0.01	74.7 ± 3.8	7.3 ± 3.1	1.997 ± 0.004	A	HzRG
	306bd	20:39:24.52	−25:14:30.6	⋯	⋯	⋯	59.5 ± 2.7	17.7 ± 4.0	1.999 ± 0.004	A	AGN
	356	20:39:24.61	−25:14:34.2	24.41 ± 0.07	⋯	⋯	8.0 ± 1.9	⋯	1.999 ± 0.005	${{\rm{B}}}^{-}$	SF
	360	20:39:25.74	−25:14:07.9	23.92 ± 0.06	⋯	⋯	10.1 ± 2.8	⋯	2.006 ± 0.004	${{\rm{B}}}^{-}$	SF
	697	20:39:21.32	−25:13:55.6	23.80 ± 0.05	⋯	⋯	5.1 ± 1.8	⋯	2.001 ± 0.006	${{\rm{B}}}^{-}$	SF
	44300	20:39:24.57	−25:14:22.2	25.66 ± 0.16	⋯	⋯	3.5 ± 1.3	⋯	2.002 ± 0.005	${{\rm{B}}}^{-}$	SF
	90000	20:39:21.33	−25:14:25.3	26.35 ± 0.24	⋯	⋯	4.0 ± 1.5	⋯	1.995 ± 0.005	${{\rm{B}}}^{-}$	SF
CARLA J0800+4029	146	08:00:12.40	+40:30:40.4	24.89 ± 0.09	⋯	⋯	4.4 ± 1.8	⋯	1.998 ± 0.007	${{\rm{B}}}^{-}$	SF
	371	08:00:16.11	+40:29:55.6	19.66 ± 0.01	17.64 ± 0.01	17.22 ± 0.01	⋯	⋯	2.004 ± 0.002e	A	QSO
	372	08:00:15.94	+40:29:53.2	20.96 ± 0.01	Blend	Blend	22.8 ± 1.7	⋯	2.001 ± 0.004	A	SF
	401	08:00:18.54	+40:30:39.0	25.94 ± 0.24	⋯	⋯	5.4 ± 3.4	⋯	1.980 ± 0.009	${{\rm{B}}}^{-}$	SF
	436	08:00:14.76	+40:29:48.1	22.05 ± 0.01	20.33 ± 0.01	20.19 ± 0.01	5.6 ± 2.1	⋯	1.969 ± 0.007	${{\rm{B}}}^{+}$	SF
	443	08:00:17.29	+40:30:30.0	24.52 ± 0.07	⋯	⋯	7.6 ± 1.3	⋯	1.992 ± 0.004	${{\rm{B}}}^{-}$	SF
	542	08:00:11.95	+40:29:24.3	22.12 ± 0.01	20.83 ± 0.02	20.78 ± 0.02	18.4 ± 5.1	⋯	1.964 ± 0.007	${{\rm{B}}}^{+}$	SF
	591	08:00:11.61	+40:29:38.0	25.82 ± 0.18	⋯	⋯	3.4 ± 1.6	⋯	1.975 ± 0.007	${{\rm{B}}}^{-}$	SF
	749	08:00:16.12	+40:28:57.3	22.00 ± 0.02	20.30 ± 0.01	20.10 ± 0.01	27.1 ± 2.2	27.8 ± 3.5	2.001 ± 0.004	A	QSO
	767	08:00:18.96	+40:29:38.9	23.81 ± 0.05	23.84 ± 0.24	22.97 ± 0.09	4.2 ± 1.2	⋯	1.978 ± 0.004	${{\rm{B}}}^{+}$	SF

Notes.

^aFlux in units of ${10}^{-17}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}$. ^bSee Section 4.1. ^cSF: star-forming galaxy, HzRG: high-redshift radio galaxy, AGN: active galactic nucleus, QSO: quasi-stellar object. ^dThis source is the companion to MRC 2036−254 (#306). ^eDetermined from the Palomar spectrum described in Section 2.3.2.

Download table as:  ASCIITypeset image

Table 4.  Confirmed Non-cluster Members

Field	Object	R.A. (hh:mm:ss.ss)	Decl. (dd:mm:ss.s)	$F140W$ (AB)	z	Qualitya	Line(s) fittedb
CARLA J2039−2514	54	20:39:28.65	−25:14:51.9	21.79 ± 0.01	1.219 ± 0.005	A	Hα
	83	20:39:28.83	−25:14:36.9	21.72 ± 0.01	0.984 ± 0.005	${{\rm{B}}}^{+}$	Hα
	133	20:39:21.92	−25:14:19.3	24.42 ± 0.08	2.289 ± 0.006	A	[O iii], [O ii]
	261	20:39:25.48	−25:13:35.9	24.34 ± 0.06	1.379 ± 0.004	A	[O iii], Hα
	265	20:39:25.05	−25:13:49.7	21.16 ± 0.01	0.813 ± 0.003	${{\rm{B}}}^{-}$	Hα
	298	20:39:24.13	−25:14:24.5	22.68 ± 0.02	0.944 ± 0.004	${{\rm{B}}}^{+}$	Hα
	382	20:39:25.40	−25:14:25.3	24.01 ± 0.09	1.452 ± 0.006	A	[O iii], Hα
	422	20:39:26.81	−25:14:06.4	22.71 ± 0.02	2.130 ± 0.004	A	[O iii], [O ii]
	429	20:39:25.05	−25:14:55.5	21.86 ± 0.02	1.908 ± 0.008	${{\rm{B}}}^{+}$	[O iii]
	496	20:39:24.63	−25:15:29.5	25.62 ± 0.17	1.934 ± 0.006	${{\rm{B}}}^{-}$	[O iii]
	564	20:39:28.53	−25:14:13.6	19.62 ± 0.01	0.714 ± 0.004	A	Hα
	564bc	20:39:28.53	−25:14:13.6	⋯	0.7	A	Hα (visual)
	565	20:39:28.52	−25:14:15.3	20.98 ± 0.01	0.709 ± 0.005	A	Hα
	570	20:39:26.13	−25:15:17.6	23.70 ± 0.05	1.330 ± 0.006	A	[O iii], Hα
	579	20:39:27.13	−25:14:57.8	23.70 ± 0.05	1.444 ± 0.004	A	[O iii], Hα
	581	20:39:25.77	−25:15:33.6	23.19 ± 0.03	1.381 ± 0.006	A	[O iii], Hα
	582	20:39:27.74	−25:14:42.8	24.52 ± 0.06	1.360 ± 0.004	A	[O iii], Hα
	588	20:39:27.93	−25:14:40.1	22.10 ± 0.02	1.472 ± 0.006	A	[O iii], Hα
	619	20:39:20.18	−25:14:56.0	22.91 ± 0.03	1.216 ± 0.004	${{\rm{B}}}^{+}$	Hα
	639	20:39:21.03	−25:14:18.1	22.39 ± 0.02	1.171 ± 0.004	A	Hα
	667	20:39:20.73	−25:14:19.5	21.57 ± 0.01	1.151 ± 0.003	${{\rm{B}}}^{-}$	Hα
	20700	20:39:29.21	−25:14:18.0	25.45 ± 0.15	1.732 ± 0.007	${{\rm{B}}}^{-}$	[O iii]
	55200	20:39:26.54	−25:14:04.2	25.64 ± 0.13	2.094 ± 0.004	${{\rm{B}}}^{-}$	[O iii]
CARLA J0800+4029	166	08:00:13.74	+40:30:53.6	23.88 ± 0.05	2.345 ± 0.005	A	[O iii], [O ii]
	198	08:00:14.20	+40:30:49.9	22.72 ± 0.02	1.927 ± 0.007	${{\rm{B}}}^{-}$	[O iii]
	204	08:00:10.62	+40:29:47.4	22.55 ± 0.02	1.745 ± 0.006	${{\rm{B}}}^{+}$	[O iii]
	294	08:00:18.39	+40:29:49.7	22.16 ± 0.01	0.753 ± 0.005	${{\rm{B}}}^{-}$	Hα
	313	08:00:17.38	+40:29:59.1	22.70 ± 0.02	2.171 ± 0.005	${{\rm{B}}}^{-}$	[O iii]
	349	08:00:12.42	+40:28:46.5	22.54 ± 0.02	2.234 ± 0.005	${{\rm{B}}}^{-}$	[O iii]
	355	08:00:13.63	+40:29:10.0	22.76 ± 0.02	1.916 ± 0.009	${{\rm{B}}}^{+}$	[O iii]
	357	08:00:18.59	+40:30:31.3	22.44 ± 0.02	1.905 ± 0.005	${{\rm{B}}}^{+}$	[O iii]
	387	08:00:14.23	+40:29:26.4	21.50 ± 0.01	1.923 ± 0.005	${{\rm{B}}}^{+}$	[O iii]
	554	08:00:13.72	+40:30:01.4	23.96 ± 0.07	1.876 ± 0.006	${{\rm{B}}}^{-}$	[O iii]
	584	08:00:15.62	+40:29:14.6	22.84 ± 0.02	0.799 ± 0.003	${{\rm{B}}}^{+}$	Hα
	587	08:00:16.78	+40:31:01.1	22.08 ± 0.01	1.258 ± 0.003	A	Hα
	673	08:00:16.41	+40:29:22.2	22.65 ± 0.02	1.743 ± 0.007	${{\rm{B}}}^{+}$	[O iii]
	697	08:00:19.04	+40:29:57.6	23.50 ± 0.04	2.092 ± 0.004	${{\rm{B}}}^{+}$	[O iii]
	746	08:00:19.47	+40:29:51.2	23.01 ± 0.03	1.873 ± 0.006	${{\rm{B}}}^{+}$	[O iii]
	776	08:00:19.20	+40:29:27.4	23.86 ± 0.06	1.439 ± 0.007	A	[O iii], Hα

Notes.

^aSee Section 4.1. ^bNote that, in some cases, additional lines have been identified but not fitted. ^cThis source is a second component to #564, blended in the $F140W$ imaging.

Download table as:  ASCIITypeset image

3.4.2. Uncertainties

We have three proxies for identifying cluster members: emission lines, IRAC [3.6]–[4.5] colors, and the redshift priors of the targeted RLAGN. We assign three different quality redshifts; A, ${{\rm{B}}}^{+}$, and ${{\rm{B}}}^{-}$, defined as follows.

• 1.  
  Quality A. When several lines are identified, we assign a quality A to the source redshift, and the redshift is considered to be very secure.
• 2.  
  Quality ${B}^{+}$. When only one line is detected but the source has a robust IRAC counterpart, we use the mid-infrared color to determine the most likely line identification (e.g., [O ii], [O iii], or Hα) and assign a quality ${{\rm{B}}}^{+}$ to the redshift. Such sources are considered to have relatively secure redshifts.
• 3.  
  Quality ${B}^{-}$. When only one line is robustly detected but the source does not have an IRAC counterpart, the line identification is less secure and we assign a quality ${{\rm{B}}}^{-}$ to the redshift. Such redshifts are considered likely correct, albeit with the potential for some mis-identifications.

When only one line is detected, we assume that it is a strong line characteristic of star formation or AGN activity, given our observation depth. We discard the possibility of recovering a Lyα line because it would imply a redshift $\gt 8$, and therefore we only consider [O ii]$\lambda 3727$, [O iii]λλ4959, 5007, and Hα. We also use [O iii] or Hα non-detections to disentangle line identifications assuming a [O iii]:Hα ratio of unity; empirically, the line ratio has a dispersion of a factor of approximately two (e.g., Mehta et al. 2015). We do not use [O ii] non-detections because the line can be below our detection limit when the typically stronger [O iii] line is detected. In cases where a source is detected in our Spitzer data, we use its IRAC mid-infrared color to segregate $z\gt 1.3$ galaxies from foreground sources since, empirically, galaxies at $z\gt 1.3$ tend to have ${([3.6]\mbox{--}[4.5])}_{\mathrm{AB}}\gt -0.1\,\mathrm{mag}$ (Papovich 2008; see Galametz et al. (2012) for a comprehensive discussion of possible contaminants). Based on the 11 quality A redshifts from our spectroscopic sample with Spitzer detections, we find that eight out of nine sources with ${([3.6]\mbox{--}[4.5])}_{\mathrm{AB}}\gt -0.1\,\mathrm{mag}$ are at $z\gt 1.3$, while one out of two sources with bluer mid-infrared colors is at $z\lt 1.3$. This is consistent with Papovich (2008), who found that red mid-infrared colors effectively identify distant ($z\gt 1.3$) galaxies, but that distant galaxies do not exclusively have red mid-infrared colors. Cluster member sources are identified based on the [O iii] line, which is observed at ∼15000 Å at $z\sim 2$, the redshift of the targeted RLAGN. Based on our identification scheme, a single emission line observed around 15000 Å can also correspond to [O ii] at $z\sim 3$. However, based on the 79,609 HST grism redshifts from the 3D-HST field survey (Momcheva et al. 2015), there are 405 [O iii] and 63 [O ii] emitters above our detection limit in the wavelength range of (14840–15042) Å. With eight (CARLA J2039−2514) and seven (CARLA J0800+4029) quality B members (see Section 4.2 and Figure 5), we therefore expect that no more than one spectroscopically confirmed source per structure is likely [O ii] at $z\sim 3$ rather than [O iii] at the redshift of the RLAGN. We performed two independent redshift determinations (GN and DS), which yield consistent assessments. Table 3 shows the cluster member properties, and Figure 3 shows their spatial distributions. Table 4 lists the measured redshifts of other sources in the HST fields of view, Figure 4 shows the redshift distribution of identified sources in each field, and Figure 5 highlights the cluster redshift distributions. We measure our member properties independently for both orientations when possible, and determine the final physical parameters as the average of the results from the two orientations, with uncertainties added in quadrature.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Eisenhardt et al. (2008) defined a spectroscopically confirmed $z\gt 1$ cluster as a structure containing at least five galaxies within a physical radius of 2 Mpc whose spectroscopic redshifts are confined to within $\pm 2000(1+\langle {z}_{\mathrm{spec}})\,\mathrm{km}\,{{\rm{s}}}^{-1}$. From emission lines and IRAC colors (when available), we identify nine members in CARLA J2039−2514 at $\langle z\rangle =1.999\pm 0.002$ (median 2.000). This includes the RLAGN, which is kinematically complex, showing two components, potentially a dual AGN, separated by 05 (see Figure 3 inset). The standard deviation between the members is 0.005 in redshift. We identify 10 members in CARLA J0800+4029 at $\langle z\rangle =1.986\pm 0.002$ (median 1.986), with a 0.014 deviation between the members. In both cases, all confirmed members are located within $1^{\prime}$ radii, corresponding to projected radii of 500 kpc at the redshifts of the RLAGN (1.5 Mpc co-moving). As seen in Figure 5, members are confined within [−1200; +700] km s⁻¹ of their cluster mean redshift for CARLA J2039−2514, spanning 0.02 in redshift space (removing one outlier, we have [−500; +600] km s⁻¹ for eight of the nine confirmed members), and within [−2300; +1700] km s⁻¹ for CARLA J0800+4029, spanning 0.04 in redshift space.

While these two structures conform to the Eisenhardt et al. (2008) confirmation criteria, these criteria were initially developed for ground-based spectroscopic programs, and have the potential to misidentify clusters less massive structures such as protoclusters, groups, sheets, and filaments. While keeping these concerns in mind, in the following, we will sometimes refer to the confirmed structures as clusters. In particular, as discussed in Section 1, other hallmarks of a bona fide cluster include a significant population of evolved massive galaxies and a centrally concentrated distribution of galaxies. We show in Section 5.2.1 that CARLA J2039−2514 contains an overdensity of (spectroscopically unconfirmed) red galaxies with properties consistent with being passive galaxies at the redshift of the structure, and prior work by our team has demonstrated that the color-selected (i.e., red) candidate cluster members are centrally concentrated around the target RLAGN (Wylezalek et al. 2013). The overdensities reach $9.0\sigma$ and $7.8\sigma$ above the field value for the fields around MRC 2036−254 and B3 0756+406, respectively. Therefore, the balance of the evidence leans in favor of these structures being galaxy clusters/forming clusters, though a conservative approach refers to them simply as structures.

The overall redshift distribution of our two fields form a strong peak at $z\simeq 2$ (Figure 4). However, the coarse resolution of the WFC3 grism and low number statistics prevent us from inferring reliable velocity dispersions. The combined-quality A ∪ ${{\rm{B}}}^{+}$ redshifts of CARLA J2039−2514 and CARLA J0800+4029 have median redshifts of z = 1.999 and z = 1.990, respectively; the three quality A redshifts of CARLA J0800+4029 have a median redshift of z = 2.001.

Hα, one of the most reliable SFR indicators, unfortunately falls outside of the grism wavelength range at the redshift of our clusters. We therefore assume [O iii]$\lambda 5007$/Hα = 1 and use the Kennicutt (1983) relation, $\mathrm{SFR}=L({\rm{H}}\alpha )/(1.12\times {10}^{41}\,\mathrm{erg}\,{{\rm{s}}}^{-1})\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$, to convert our [O iii]$\lambda 5007$ fluxes into SFRs. We note, however, that there is a large scatter in the [O iii]$\lambda 5007$/Hα ratio reported for $z\sim 2$ star-forming galaxies. Mehta et al. (2015) indicate a linear relation between [O iii]$\lambda 5007$ and Hα luminosities with [O iii]$\lambda 5007/{\rm{H}}\alpha \sim 2$ albeit with a large scatter. On the other hand, Juneau et al. (2014) suggests a [O iii]$\lambda 5007$/Hα ratio in the range of 0.5–1.5 for the redshift and mass of our confirmed cluster members. Using the mean [O iii]$\lambda 5007$ flux of our SF cluster members ($8\times {10}^{-17}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}$), but excluding the RLAGN whose [O iii] fluxes are likely enhanced by AGN photoionization, we compute a mean SFR of $20\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ for our SF cluster members, with no dust correction applied. Given the spread in literature values for the [O iii]$\lambda 5007$/Hα ratio, this is considered a very crude estimate of the SFR, only considered robust at the level of a factor of two.

To better constrain our SFR lower limit, we tentatively estimate the contribution of dust extinction to the SFR. One robust method is to use the ratio of the measured Balmer decrement (the flux ratio of Hα over Hβ). However, we deem it too uncertain to use [O iii]$\lambda 5007$ as a proxy for Hα to calculate Balmer decrements given (1) the scatter in the [O iii]$\lambda 5007$ to Hα ratio, (2) the 10 times shorter wavelength baseline between Hβ and [O iii]$\lambda 5007$ as compared to between Hβ and Hα, and (3) the low SNR for our measured fluxes, typically of the order of five. We therefore assume a constant A_V = 1 mag, a typical dust attenuation in the V-band for SF galaxies (Kewley et al. 2004). Other groups have applied similar approximations (e.g., Zeimann et al. 2012; Newman et al. 2014). In particular, the level of dust extinction is expected to correlate with stellar mass, though the improvement taking that effect into account is likely small compared to the uncertainties inherent to our oxygen-based SFRs. Using the Calzetti et al. (2000) extinction curves with ${R}_{V}=4.05$, we have ${A}_{[{\rm{O}}{\rm{III}}]}=0.96$ mag. Our SF cluster members, therefore, have dust-corrected SFRs in the range of $\sim (20\mbox{--}140)\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$, with a median (mean) cluster member SFR²³ of $\sim 35\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ ($\sim 50\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$). This leads to a lower limit on the total dust-corrected SFR of $\geqslant 400\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ for both of our clusters.

To estimate stellar masses, we scaled Spitzer/IRAC [3.6] and [4.5] magnitudes to Bruzual & Charlot (2003) stellar population synthesis models using a Chabrier (2003) initial mass function (IMF), single stellar population (SSP, i.e., single delta-burst population), and a z_f = 4.5 formation redshift. We estimate stellar masses in the range of $(0.1\mbox{--}1.9)\times {10}^{11}\,{M}_{\odot }$ for our CARLA cluster members (median $1.1\times {10}^{11}\,{M}_{\odot }$). For cluster members without CARLA counterparts, we assign upper limits of $\lt {10}^{10}\,{M}_{\odot }$ based on the IRAC depths. The RLAGN contributions to the spectral energy distributions (SEDs) likely contaminate the stellar component (e.g., Drouart et al. 2012) and overestimate the derived stellar masses. This is particularly true for type-1 AGNs like B3 0756+406, and typically less so for type-2 AGNs like MRC 2036−254. Without additional longer wavelength photometry to disentangle the stellar and AGN components in the SEDs (e.g., Seymour et al. 2007; De Breuck et al. 2010), we simply determine upper limits on the stellar masses of the RLAGN MRC 2036−254 and B3 0756+406, respectively $\lt 3\times {10}^{11}$ and $\lt 3\times {10}^{12}\,{M}_{\odot }$, and we determine a limiting stellar mass $\lt 2\times {10}^{11}\,{M}_{\odot }$ for the CARLA J0800+4029 object $\#749$ identified as a QSO.

We next make an estimate of the total masses of the two newly confirmed structures. The low resolution of the grism spectroscopy precludes a measurement of a velocity dispersion, and we currently lack the data to attempt other standard cluster mass measurements, such as weak lensing, diffuse X-ray emission, and SZ decrements. Therefore, we use the Spitzer imaging to estimate the total stellar mass in the systems, and then compare it to other, better studied clusters at comparable redshift.

Following the methodology discussed in detail in Wylezalek et al. (2014), we fit the CARLA J2039−2514 and CARLA J0800+4029 background-subtracted luminosity functions with a Schechter (1976) function, integrating over the range of $({m}^{\star }-2.5)$ to $({m}^{\star }+10)$. This allows us to estimate the total luminosity density of the two structures, which we convert to total stellar masses using Bruzual & Charlot (2003) solar metallicity SSP models with formation redshift z_f = 4.5. We assume a constant stellar mass to light ratio, determined for both the Chabrier (2003) and the Salpeter (1955) IMF. The Chabrier (2003) IMF provides total stellar masses of ${M}_{\star }=1.9\times {10}^{12}\,{M}_{\odot }$ for CARLA J2039−2514, and ${M}_{\star }=2.9\times {10}^{12}\,{M}_{\odot }$ for CARLA J0800+4029. We apply the same methodology to the well studied distant galaxy clusters ClG 0218.3−0510 ($z=1.62;$ Papovich et al. 2010; Tanaka et al. 2010) and IDCS J1426.5+3508 ($z=1.75;$ Stanford et al. 2012; Brodwin et al. 2016), finding total stellar masses of ${M}_{\star }=1.2\times {10}^{12}\,{M}_{\odot }$ for ClG 0218.3−0510 and ${M}_{\star }=1.9\times {10}^{12}\,{M}_{\odot }$ for IDCS J1426.5+3508. Our derived total stellar masses for the two new systems at z ∼ 2 are, therefore, similar to IDCS J1426.5+3508 and slightly higher than ClG 0218.3−0510. These total masses are all $\sim 50 \%$ larger if we instead adopt the Salpeter (1955) IMF.

Tanaka et al. (2010) report an X-ray derived mass ${M}_{200}=(5.7\pm 1.4)\times {10}^{13}\,{M}_{\odot }$ for ClG 0218.3−0510. Brodwin et al. (2012, 2016) report a total halo mass in the range of ${M}_{500}=(1.9\mbox{--}3.3)\times {10}^{14}\,{M}_{\odot }$ (${M}_{200}\sim 4\times {10}^{14}\,{M}_{\odot }$) for IDCS J1426.5+3508 from a variety of halo mass tracers, including SZ, X-rays, and strong gravitational lensing (see also Gonzalez et al. 2012; and Mo et al. 2016 for a weak lensing analysis), making IDCS J1426.5+3508 the most massive galaxy cluster currently known at $z\gt 1.5$. Galaxy clusters in this mass range typically have stellar to halo mass ratios of a few percent (e.g., Andreon 2012; Kravtsov et al. 2014), which is consistent with our Spitzer-derived stellar mass estimate above. We note, however, that these relations were derived using lower redshift clusters. Based on their estimated stellar masses, CARLA J2039−2514 and CARLA J0800+4029 have comparable stellar masses to IDCS J1426.5+3508, implying that their total halo masses are likely $\gt {10}^{14}\,{M}_{\odot }$. If confirmed by additional data, this would place both structures among the most massive galaxy clusters known at $z\gt 1.5$.

This section discusses cluster members of particular interest in detail. The full sample of cluster members is discussed in Appendix D. All spectra are shown in the same appendix.

4.6.1. CARLA J2039−2514

4.6.2. CARLA J0800+4029

The two z = 2 structures reported in this paper likely represent some of the highest redshift clusters currently spectroscopically confirmed. They illustrate the high efficiency of our approach to target RLAGN with IRAC mapping and HST grism follow-up. Because the two fields are only a pilot study for our full sample of 20 cluster candidates, we next explore the efficiency of our grism spectroscopy and [3.6]–[4.5] selection.

5.1.1. Grism Efficiency

We assess in this section the outcome of our HST observations. In Figure 6, we show the flowchart of the classification of HST sources from our master catalog. The classification is shown for both fields, with the numbers on the left side of each box corresponding to the MRC 2036−254 field and the numbers on the right side corresponding to the B3 0756+406 field. We also display in parentheses numbers corresponding to the classification for secure Spitzer color-selected candidates that have a single HST counterpart. Overall, we determine redshifts for 6% of our exploitable HST sources (31/550 and 26/452, respectively, for the fields around MRC 2036−254 and B3 0756+406); where "exploitable" is defined to mean that $\gt 75 \%$ of the source continuum falls on the detector and contamination is less than 60% of the cutout length. We find that 2% of sources (9/550 and 10/452) are confirmed cluster members—i.e., for every three redshifts that we measure, we find one cluster member. This is consistent with probing a biased environment, and is not an instrumental bias. For example, the redshift distribution of the 3D-HST field survey (Momcheva et al. 2015) is roughly flat in the same redshift window ($0.7\lt z\lt 2.3$), based on their sample of $46,256$ grism redshifts in this window obtained over $626\,{\mathrm{arcmin}}^{2}$ with a similar two-orbit per field HST grism program. Of the exploitable HST sources, 71% (370/550 and 344/452) are not detected in the dispersed grism data. Moreover, of the sources with spectral detections, 66% (125/180 and 66/108) show continuum only, for which we could not measure a redshift because no emission lines were detected. These sources are likely a mixture of: (1) stars, (2) old and passive (quiescent) galaxies with little star formation, (3) galaxies at redshifts for which no strong features are covered by our grism observations, and (4) SF galaxies for which we do not detect emission lines at the depth of our data (SFR $\lesssim 20\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$, unless highly dust-obscured). Also, we have not been able to determine a redshift for 40% (24/55 and 16/42) of sources with detected emission line(s) (with or without continuum) because of the ambiguity of the nature of the line(s). This number improves by a factor of 3.5 when considering Spitzer color-selected sources since we have color information that helps identify ambiguous emission lines. Overall, we also lose 26% (195/745 and 170/622) of the source spectra because of full contamination of their spectral first orders (15%, 112/745 and 88/622) or because their traces fell outside of the G141 detector (12%, 83/745 and 82/622).

Zoom In Zoom Out Reset image size

5.1.2. IRAC Color-selection Efficiency

Williams et al. (2009; see also Labbé et al. 2005; Wuyts et al. 2007; Whitaker et al. 2011) have shown that it is possible to use rest-frame UVJ colors to separate the SF, dusty SF, and passive populations. To build our color–color diagrams (Figure 7), we use observed $z/i-F140W$ versus F140W – [3.6]. Following Mei et al. (2009) and using Mei's codes and the python version of EZGAL (Mancone & Gonzalez 2012), we transform Williams et al. (2009) color limits into our observed apparent colors. We use a Bruzual & Charlot (2003) SSP model with galaxy formation redshifts averaged between z_f = 3 and 8, and metallicities equal to 40% solar, solar, and 2.5 times solar; and we age the templates to z = 2. We average the color conversions using a set of formation redshift ranges from ${z}_{f}=3\mbox{--}8$ to ${z}_{f}=5\mbox{--}8$ increasing by steps of 0.1. This is a conservative simple model and assumes that passive galaxies are mostly located close to the passive boundary defined in Williams et al. (2009). A more detailed analysis of the CARLA cluster stellar population will be performed in future work.

Zoom In Zoom Out Reset image size

In Figure 7, we plot color–color diagrams for the two fields to investigate the presence of passive cluster members associated with CARLA J2039−2514 and CARLA J0800+4029, and in Figure 8 we plot their color–magnitude diagrams (CMDs). We use i-band data from Cooke et al. (2015) for CARLA J0800+4039 and a $4800\,{\rm{s}}$ z-band image obtained with VLT/ISAAC on UT 2002 July 17 for CARLA J2039−2514.²⁶ For both CMDs, we use HST/F140W data to bracket the D4000 break at z = 2. In the color–color diagrams, the passive candidates are located inside the upper left quadrant. The SF population is located below the horizontal line, whereas dusty SF galaxies lie on the right of the vertical boundary. We identify 14 passive galaxies for CARLA J2039−2514 (2 of them outside the plot) and 2 for CARLA J0800+4029.

Zoom In Zoom Out Reset image size

5.2.1. Density of Passive Candidates

5.2.2. Red Sequences and SF Populations

We compare our results to other high-redshift clusters spectroscopically confirmed using HST/WFC3 grism data. Gobat et al. (2013, hereafter G13), reported on a rich cluster at $z\simeq 2$. They first reported evidence of a fully established galaxy cluster at z = 2.07 (Gobat et al. 2011) from X-ray emission, ground-based spectroscopy, and ground- and space-based photometry, later revised by G13 to a slightly lower redshift using deep HST grism observations. The system consists of two unrelated aligned structures, with the background overdensity being "sparse and sheet-like." The G13 median redshift of the HST/WFC3 confirmed members of the foreground overdensity is z = 1.993, with a standard deviation of 0.012. With median redshifts of z = 2.000 (standard deviation 0.005) and z = 1.986 (standard deviation 0.014), CARLA J2039−2514 and CARLA J0800+4029, respectively, are among the most distant confirmed clusters currently known. G13 confirmed 22 members with 18 orbits and three orientations (12.5 hr on source) including five quiescent sources, and found no evidence for an already formed red sequence within $20^{\prime\prime}$ of the cluster core (Gobat et al. 2011). More recently, Newman et al. (2014), hereafter N14, confirmed another rich cluster at high redshift using HST/WFC3 slitless spectroscopy. With 14 orbits, they confirmed 19 members at z = 1.80, of which more than 75% are quiescent. They found a clear red sequence of observed mean color $\langle z-J\rangle =1.98\pm 0.02$, which includes 13 of their 15 quiescent cluster members. Zeimann et al. (2012, hereafter Z12), also confirmed both emission-line and quiescent sources in a cluster at z = 1.89 using six orbits of HST/WFC3 grism spectroscopy. Z12 showed that a significant fraction of early-type galaxies in the cluster field were consistent with forming a red sequence. These deep observations show that clusters can host a substantial fraction of quiescent galaxies even at early epochs. By design, our shallow two-orbit per field strategy only confirms SF members. In addition to the confirmed emission-line members, we find a large fraction (79%, 52/64, and 47/61) of Spitzer color-selected cluster candidates below our detection limit or showing continuum only. CARLA J2039−2514 and CARLA J0800+4029 are therefore relatively robust confirmations, and have a high potential for being richer structures than what our shallow HST observations allowed us to unveil. Furthermore, we note that our two-orbit strategy spectroscopically confirms approximately five cluster members per orbit, which is approximately four times more efficient than the $\gt 10$-orbit programs reported in G13 and N14.

From X-ray emission and richness, G13 estimated a mass of ${M}_{200}\sim 5\times {10}^{13}\,{M}_{\odot }$ for their cluster. N14 reported a massive cluster with ${M}_{200}=(2\mbox{--}3)\times {10}^{14}\,{M}_{\odot }$, including five very massive members whose stellar masses are in the range of $(4\mbox{--}10)\times {10}^{11}\,{M}_{\odot }$. We do not find such high masses for our confirmed cluster members (with the exception of the RLAGN); our 7 IRAC-detected sources have a median stellar mass of $1.1\times {10}^{11}\,{M}_{\odot }$, and the remaining 12 non-IRAC detected members have masses $\lt {10}^{10}\,{M}_{\odot }$. Z12 derived SFRs from the [O ii] and Hβ lines and find SFRs in the range of $(20\mbox{--}40)\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$. Based on the admittedly crude [O iii] SFR indicator, we find similar SFRs for 9/16 of our SF cluster members. We also find higher SFRs, in the range of $(40\mbox{--}140)\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$, for a significant number of SF cluster members (7/16). Note that our line detection limit imposes an SFR lower limit of $\gt 20\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ (similar to Z12 but based on a different line). According to the SFR/stellar-mass relation in Rodighiero et al. (2011) for $1.5\lt z\lt 2.5$ galaxies, our low-mass cluster members (typically all the non-IRAC detected members, with masses $\lesssim {10}^{10}\,{M}_{\odot }$) are above the star-forming main-sequence toward starburst galaxies. Our detection limits prevent us from confirming main-sequence galaxies at these masses and redshifts.

In Section 4.3, we showed that confirmed emission line members of both structures imply total cluster SFRs of at least $\sim 400\,{M}_{\odot }{\mathrm{yr}}^{-1}$ within ∼500 kpc of the cluster centers, which are assumed to be coincident with the target RLAGN. Based on Spitzer 24 μm imaging of a sample of clusters from the Spitzer/IRAC Shallow Cluster Survey, Brodwin et al. (2013) showed a steeply increasing SFR in cluster cores out to z = 1.50, with an average SFR of several hundred ${M}_{\odot }\,{\mathrm{yr}}^{-1}$ found in the cores of the highest redshift clusters in that study. Alberts et al. (2014, 2016) find similar results based on longer wavelength Herschel data of a similar cluster sample. The results found here at $z\sim 2$ for CARLA J2039−2514 and CARLA J0800+4029 are consistent with those studies, though higher fidelity SFR indicators for these newly confirmed structures would be highly preferable to the current [O iii]-based values.

Although G13 found no evidence for an already formed red sequence within $20^{\prime\prime}$ of the cluster core, they determined, from their best subsample of (96) cluster candidates comprising 14 spectroscopic members and 82 photo-z candidates (expected to include 50 interlopers), that ∼(60–80)% of candidates within $\sim 20^{\prime\prime}$ of the cluster core are passive regardless of mass, compared to ∼20%, ∼40%, and ∼60% in the field for $\mathrm{log}M/{M}_{\odot }\gt 10,10.5,$ and 11, respectively (Strazzullo et al. 2013). We additionally note that the structure reported in Spitler et al. (2012) and Yuan et al. (2014), respectively, discovered at z = 2.2 from photometric redshifts and later spectroscopically confirmed at $\langle z\rangle =2.095$ with 57 members, comprises several $1^{\prime}$ radius overdense groups covering a $12^{\prime} \times 12^{\prime}$ area in COSMOS (Scoville et al. 2007). This structure has a slightly enhanced number of red galaxies for two groups compared to the field, with ${N}_{\mathrm{red}}=0.5\pm 0.2\times {N}_{\mathrm{tot}}$ compared to 0.2 ± 0.03 in the field. Our results are also indicative of the presence of passive CARLA candidates associated with CARLA J2039−2514, while CARLA J0800+4029 exhibits a number of passive candidates similar to the field in our comparison. A careful analysis accurately evaluating the passive fraction of galaxies in our structures will be addressed in E. A. Cooke et al. (2016, in preparation).

Three different kinds of calibrated data can be retrieved from MAST: "calibrated, flat-fielded individual exposures" (FLT), "calibrated, cosmic-ray-rejected, combined images" (CRJ), and "calibrated, geometrically corrected, dither-combined images" (DRZ; WFC3 Data Handbook, Rajan et al. 2010). According to the handbook, the Astrodrizzle package of the aXe software for slitless spectroscopy data extraction (Kümmel et al. 2009) supersedes the CRJ and DRZ data preparations of the standard WFC3 calibration program (calwf3). Therefore, we only retrieved FLT files, which are used as input by aXe.

We first used the aXe (v2.4.4) astrodrizzle task to combine the eight $F140W$ dithered direct exposures of each field (four from each orientation). This step creates a deep (1023–1073 s) drizzled direct image with bad pixels and cosmic rays rejected. The combined direct images of each field are shown in Figure 3, with the positions of confirmed cluster members indicated.

In order to associate sources and spectra and to wavelength calibrate the spectra, we first create a catalog of sources in each field using SExtractor (Bertin & Arnouts 1996) on the drizzled image. We create a deep catalog using SExtractor detection parameters selected to identify even the faintest sources ($\mathrm{DETECT}\_\mathrm{MINAREA}=4$, $\mathrm{DETECT}\_\mathrm{THRESH}=2$, and $\mathrm{ANALYSIS}\_\mathrm{THRESH}=2$). Lower values only add spurious detections (e.g., noise clumps, stellar diffraction spikes, dismemberment of extended sources) while higher values miss a significant number of faint sources. This deep catalog, gathering information on source positions, magnitudes, sizes, and orientations, is then cleaned by hand from spurious detections. We also generate a more conservative catalog using $\mathrm{DETECT}\_\mathrm{MINAREA}=10$, $\mathrm{DETECT}\_\mathrm{THRESH}=3$, and $\mathrm{ANALYSIS}\_\mathrm{THRESH}=3$ to restrict the number of detections for the grism sky background subtraction (see Appendix A.4).

We cross-correlate our HST and CARLA/Spitzer catalogs for two reasons. First, we update the nature of the CARLA sources themselves. With nearly 10 times better spatial sampling than Spitzer, HST provides morphological information and identifies some IRAC sources as spurious detections or blends, potentially leading to misfigured IRAC colors and erroneous CARLA selection. Second, we wish to investigate the spectra of all CARLA sources in our grism data. To the cleaned deep catalog, we add Spitzer CARLA candidates too red to be detected in the HST direct imaging. Even though these undetected sources may not show any continuum, they can potentially show emission lines in the grism data. This step required a slight astrometric correction ($\sim 0\buildrel{\prime\prime}\over{.} 2$) to the IRAC positions. We based our correction on secure CARLA sources identified in the first step. At this point, we now have our cleaned, complete and final catalog of sources for spectroscopic analysis, referred to as the master catalog.

Exposure blocks (pairs of direct and grism observations) are slightly offset from one another. Therefore, we cannot directly associate the master catalog sources to their spectra. Using the Astrodrizzle aXe task iolprep, we then project back the master catalog to each individual direct image and obtain individual catalogs corresponding to each grism exposure, required for stacking the grism data of each observation.

Prior to stacking the grism data, we remove the overall background level of the G141 detector, which varies significantly along its surface ($\sim (0.9\mbox{--}2.4)\,{e}^{-}\,{{\rm{s}}}^{-1}$). The aXe task axeprep scales a master sky background on spectral-free regions. We use the shallow catalog, projected back to individual frames, as our master catalog is too deep for aXe to produce enough sky-free regions.

Two kinds of contamination models can be generated using the aXe task axecore: quantitative or qualitative. For each source, the quantitative contamination model produces a 2D cutout of the first order grism spectrum, and shows how many potential contaminating spectral traces from other sources fall in the cutout. However, this model is purely geometrical and includes all spectral orders from all sources regardless of their likelihood of actually being detected. This leads to large over-estimation of the contamination and typically no contaminant-free regions. On the other hand, the qualitative contamination model computes the shapes and flux intensities of the spectral traces, based on position, magnitude, and size of the sources. One can choose between the "Gaussian" and the "fluxcube" methods (described in the aXe User Manual (v2.3), Kümmel et al. 2011). Briefly, the "Gaussian" method approximates the shapes of sources as Gaussians and requires magnitudes from at least one band to estimate flux intensities (more bands provide better estimates). The "fluxcube" method creates more realistic spectral trace shapes as it models the morphologies of the sources from at least one HST image and a segmentation map (again, more images from multiple bands provide better estimates). Gobat et al. (2013) report that the "fluxcube" method is often a poor approximation of the spectrum, even with several multi-band HST images available. With only $F140W$ available to us, we use the "Gaussian" method to locate traces and zeroth orders of possible contaminants.

• 1.  
  #119: We identify contamination above 15200 Å and below 11500 Å in the first observation, and up to 15200 Å in the second observation, albeit slightly offset from the center of the 2D extraction. This Spitzer-selected CARLA source has a mid-infrared color consistent with the emission line seen in both orientations as being [O iii]. We measure a redshift of $z=1.987\pm 0.007$, and we do not detect Hβ or other emission lines. Hence, we assign a quality ${{\rm{B}}}^{+}$ to the redshift based on the emission line and Spitzer color.
• 2.  
  #174: The contamination model for this Spitzer-selected CARLA source does not identify strong spectroscopic contamination. However, the two emission-like features seen at 14100 and 14600 Å in the first observation are not observed in the second observation. This suggests unidentified contamination at the level of our contamination contours. This could be due to pixel noise, a non-extracted source in the field, a source outside the field of view, or a shallow emission-line contaminant. However, the emission line at 15000 Å is consistent between both orientations and the source's Spitzer color is consistent with [O iii] at $z\sim 2$. We measure a redshift of $z=2.000\pm 0.007$ from the [O iii] emission line. Because of the contamination concerns, the line flux listed in Table 3 is from the second observation. We do not detect other lines, and therefore, assign a quality ${{\rm{B}}}^{+}$ to this source.
• 3.  
  #281: Another source ($\#280$) is located $1^{\prime\prime}$ from $\#281$. We detect the same emission lines in the $\#280$ spectra as in $\#281$, however, spatially offset above and below the continuum for the former (depending on the observation). Therefore, we associate the emission lines to $\#281$, and we do not exclude $\#280$ as a quiescent companion to $\#281$, even though it might be a foreground or background source. The contamination model indicates slight contamination around 15000 Å. The contamination is likely the origin of the broad shape of the line in the second observation, not seen in the first one, and might enhance the true flux of the line, but we still use both orientations to determine the source properties. We measure $z=2.002\pm 0.008$ (quality ${{\rm{B}}}^{-}$).
• 4.  
  #356: Another source ($\#355$) is located $1^{\prime\prime}$ from $\#356$. For similar reasons as for $\#281$, we associate the emission lines to $\#356$, and do not exclude $\#355$ as a quiescent companion to $\#356$. We measure $z=1.999\pm 0.005$ (quality ${{\rm{B}}}^{-}$) from the second observation, as the first one is strongly contaminated by a bright source.
• 5.  
  #360: We detect contamination above 15300 Å in the first observation, and below 14000 Å in the second. However, our model does not indicate contamination around the well-detected line at 15050 Å, which we identify as [O iii]. We do not identify any other emission lines, and we consider the emission-like feature seen in the second observation at the expected position of Hβ to be due to contamination since it is not present in the first observation. We measure a redshift of $z=2.006\pm 0.004$ (quality ${{\rm{B}}}^{-}$).
• 6.  
  #697: The contamination model indicates shallow contamination in the first observation, and spatially offset contamination in the second. We use both observations and determine a quality ${{\rm{B}}}^{-}$ redshift of $z=2.001\pm 0.006$, based on the detection of a single emission line in both orientations.
• 7.  
  #44300: From the second observation, free from contamination, we detect a single emission line, which we associate with [O iii] at a redshift of $z=2.002\pm 0.005$ (quality ${{\rm{B}}}^{-}$).
• 8.  
  #90000: The first orientation is free from contamination within our 1D extraction region. The second observation is contaminated longward of the emission line, which affects our spectral modeling. Therefore, we only use the first observation to measure the line parameters in Table 3, providing a quality ${{\rm{B}}}^{-}$ redshift of $z=1.995\pm 0.005$.

• 1.  
  #146: The second observation is free from contaminant. The first observation is contaminated up to 15700 Å, but the contamination is spatially offset by ∼05 from the source. Therefore, we carefully extract the 1D spectrum from the 2D cutout to avoid the contaminant, allowing us to use both observations. We measure a redshift of $z=1.998\pm 0.007$ (quality ${{\rm{B}}}^{-}$).
• 2.  
  #401: The second orientation of this source is visually contaminated below 15000 Å, but we do not have a contamination model associated with this source because it was extracted alone after the main source extraction. We measure a redshift of $z=1.980\pm 0.009$ (quality ${{\rm{B}}}^{-}$).
• 3.  
  #436: This source has a Spitzer color consistent with $z\gt 1.3$. Contamination is detected offset from the source continuum in the 2D cutouts, without overlap. A single line is detected in both observations, providing a quality ${{\rm{B}}}^{+}$ redshift of $z=1.969\pm 0.007$.
• 4.  
  #443: Above 15500 Å, the second observation is contaminated by a zeroth order image of a bright source. We identify an emission line at ∼14980 Å in both observations, but the line is potentially contaminated in the second observation by the mentioned zeroth order. Therefore, we only use the first observation, and measure a redshift of 1.992 ± 0.004 (quality ${{\rm{B}}}^{-}$). The fitting procedure tentatively detects Hβ below our detection limit, and therefore is not taken into account.
• 5.  
  #542: The second observation shows the same emission line as the first observation, but broadened in the dispersion direction (and slightly in the spatial direction). This could be due to the spatial extent of the source, which is more elongated in the dispersion direction of the second observation, or due to an undetected contaminant in the second observation. The mid-infrared Spitzer color is consistent with [O iii] at $z\sim 2$, but no other emission lines are detected. We use the two observations to measure its redshift and line flux despite the elongated spectral shape of the second observation (note that only using the first observation does not significantly change the results). We measure a redshift of $z=1.964\pm 0.007$ (quality ${{\rm{B}}}^{+}$).
• 6.  
  #591: Given our contamination model, both observations are free from contaminants. However, we identify in the direct imaging a source located $\sim 2^{\prime\prime}$ from #591 and likely contaminating part of the spectral 2D cutouts. Therefore, we carefully extract the 1D spectrum from the 2D cutout to avoid potentially contaminated regions, allowing us to use both observations. We measure a redshift of $z=1.975\pm 0.007$ (quality ${{\rm{B}}}^{-}$).
• 7.  
  #767: This source was detected in our IRAC images and has a Spitzer color consistent with [O iii] at $z\sim 2$. However, it was not selected as a CARLA candidate because it did not pass our CARLA flux cut. The second observation is contaminated below 14900 Å, with the contamination also somewhat affecting the emission line. The first observation is not contaminated around the emission line at 14910 Å, but is likely contaminated at shorter wavelengths. Using the first observation, we determine a redshift of $z=1.978\pm 0.004$, of quality ${{\rm{B}}}^{+}$. The observed feature at the expected location of Hβ is likely due to contamination, just at our flux detection limit.