YOUNG GALAXY CANDIDATES IN THE HUBBLE FRONTIER FIELDS. I. A2744

Our understanding of the first billion years of cosmic time has increased significantly in recent years, thanks to the Hubble Space Telescope's (HST's) Wide-Field Camera 3/Infrared Channel (WFC3/IR; Kimble et al. 2008) as well as the Spitzer Space Telescope's Infrared Array Camera (IRAC; Fazio et al. 2004). Until recently, the Hubble Ultra Deep Field (Beckwith et al. 2006; Illingworth et al. 2013) has provided our deepest view of the universe, revealing a considerable number of galaxy candidates at z > 7, including one candidate at z ≃ 10 (Bouwens et al. 2011, 2012; Ellis et al. 2013; Oesch et al. 2013; Illingworth et al. 2013). The cosmic epoch of z ≃ 10 is important to study as it marks the dawn of galaxy formation and the beginning of reionization of the intergalactic medium. However, galaxies at that redshift are extremely faint, making it difficult to discover and study the abundant population of galaxies below L*, the knee of the luminosity function. Fortunately, the magnification boost afforded by gravitational lensing combined with HST's exquisite imaging capabilities in the near-infrared (NIR), provides an avenue for both discovering and characterizing the intrinsic properties of galaxies around z ≃ 12, when the universe was less than half a billion years old.

The Cluster Lensing And Supernova survey with Hubble (CLASH; Postman et al. 2012) carried out HST imaging of 25 galaxy clusters in 16 broad bands between 0.2–1.7 μm to a depth of AB magnitude ∼27 with a total of 20 orbits per cluster. The CLASH program has led to many interesting discoveries of magnified, intrinsically faint galaxies. Several hundred dropout galaxies have been uncovered in the range z ≃ 6–8, with a few notable examples at higher redshifts of z ≃ 9–11 (see Zheng et al. 2012b; Bouwens et al. 2014a; Coe et al. 2013; Bradley et al. 2014), helping to motivate dedicated deeper lensing surveys.

The Hubble Frontier Fields (HFF) is a new initiative now being carried out to observe the distant universe to an unprecedented depth, combining the power of deep HST imaging and gravitational lensing. In HST's Cycles 21 and 22, 560 orbits of Director's Discretionary Time have been allocated to observe four clusters. The observations are carried out with four WFC3/IR filters (F160W, F140W, F125W, F105W) and three ACS filters (Advanced Camera for Surveys; Ford et al. 1998, F814W, F606W, F435W). It is anticipated that 280 orbits will be allocated in Cycle 23 to observe two additional clusters. In addition, deep Spitzer and Chandra observations are planned for the six HFF fields. These coordinated observations will enable us to probe the star formation rate density at z ≳ 9, study the faint end of the galaxy population at z ≃ 3–8, and map the dark matter in these clusters in unprecedented detail via many multiple images of background sources (Hubble Deep Fields Initiative 2012 Science Working Group Report).¹⁶

We report the discovery of 24 candidate Lyman-break galaxies (LBGs) at z ≳ 7 in the field of A2744, based on the HFF observations and archival data (Table 1). The faintest sources detected are around AB magnitude 29. These objects, listed in Tables 2–4, have "secure" photometric redshifts greater than 7 and a negligible probability (<1%) of being at lower redshift.

Table 2. WFC3 and ACS Photometry^a of Candidates at z > 9

Name	Photometric	R.A.	Decl.	F160W	F140W	F125W	F105W	F814W	μb
	Redshift	(J2000)	(J2000)
JD2c	10.5 ± 0.7	3.597037	−30.412132	28.00 ± 0.15	29.00 ± 0.43	30.20 ± 1.33	29.52 ± 0.52	>31.00	$14.8 ^{+66} _{-7.8}$
JD3	$10.0^{+0.7}_{-1.0}$	3.603344	−30.390978	28.68 ± 0.16	29.26 ± 0.30	30.30 ± 0.78	>30.50	>30.50	$1.6 ^{+0.9} _{-0.1}$

Notes. ^aMagnitudes are isophotal, scaled by an aperture correction term derived in the F160W band. The errors and limiting magnitudes are 1σ. Photometric redshifts have been derived using BPZ, and the quoted uncertainties indicate the 68% confidence interval. ^bMagnification factor from Zitrin "NFW" model (see Section 4.1), where uncertainties are combined from two sources: (1) the model statistical uncertainties, which are relatively small; and more prominently, (2) the maximum differences with other six models, excluding one highest and one lowest values. ^cFor system JD1, see Zitrin et al. (2014).

Download table as:  ASCIITypeset image

Table 3. WFC3 and ACS Photometry^a of Candidates at 8 ≲ z < 9

Name	Photometric	R.A.	Decl.	F160W	F140W	F125W	F105W	F814W	μb
	Redshift	(J2000)	(J2000)
YD1	$8.7^{+0.5}_{-0.2}$	3.603856	−30.381905	27.83 ± 0.08	27.86 ± 0.09	28.49 ± 0.17	29.85 ± 0.43	>31.00	$1.4 ^{+0.7} _{-0.1}$
YD2	8.3 ± 0.2	3.572515	−30.413267	28.12 ± 0.12	28.11 ± 0.13	28.11 ± 0.13	29.84 ± 0.48	>31.00	$1.6 ^{+0.4} _{-0.1}$
YD3	$8.8^{+0.4}_{-0.2}$	3.603858	−30.415842	28.33 ± 0.10	28.63 ± 0.15	28.66 ± 0.18	>30.50	>31.00	$2.9 ^{+0.5} _{-0.6}$
YD4c	8.5 ± 0.1	3.603864	−30.382265	26.42 ± 0.04	26.46 ± 0.04	26.82 ± 0.06	28.78 ± 0.27	>31.00	$1.4 ^{+0.7} _{-0.1}$
YD5	8.5 ± 0.3	3.579479	−30.386534	27.83 ± 0.10	27.66 ± 0.09	28.38 ± 0.18	29.24 ± 0.29	>31.00	$3.0 ^{+9.8} _{-0.9}$
YD6	8.3 ± 0.2	3.604005	−30.382309	26.95 ± 0.06	27.09 ± 0.07	27.36 ± 0.10	28.83 ± 0.26	>31.00	$1.4 ^{+0.7} _{-0.1}$
YD7c	8.3 ± 0.1	3.603397	−30.382256	26.17 ± 0.03	26.10 ± 0.03	26.38 ± 0.05	27.53 ± 0.09	>31.00	$1.4 ^{+0.7} _{-0.1}$
YD8c	8.1 ± 0.1	3.596096	−30.385832	26.65 ± 0.04	26.49 ± 0.04	26.68 ± 0.04	27.76 ± 0.09	>31.00	$1.9 ^{+5.6} _{-0.2}$
YD9	$8.3^{+0.2}_{-0.6}$	3.572902	−30.413658	28.49 ± 0.14	28.22 ± 0.12	28.68 ± 0.18	29.62 ± 0.32	>31.00	$1.6 ^{+0.5} _{-0.1}$
YD10c	8.3 ± 0.2	3.598108	−30.382393	27.70 ± 0.09	27.57 ± 0.09	27.89 ± 0.12	28.89 ± 0.23	>31.00	$1.5 ^{+2.3} _{-0.1}$
YD11	$8.0^{+0.3}_{-0.6}$	3.600947	−30.399149	28.96 ± 0.17	28.97 ± 0.19	28.89 ± 0.18	>30.50	>31.00	$3.4 ^{+0.9} _{-0.1}$

Notes. ^aMagnitudes are isophotal, scaled by an aperture correction term derived in the F160W band. The errors and limiting magnitudes are 1σ. Photometric redshifts have been derived using BPZ, and the quoted uncertainties indicate the 68% confidence interval. ^bMagnification factor from Zitrin "NFW" model (see Section 4.1), where uncertainties are combined from two sources: (1) the model statistical uncertainties, which are relatively small; and more prominently, (2) the maximum differences with other six models, excluding one highest and one lowest values. ^cAlso in Coe et al. (2014).

Download table as:  ASCIITypeset image

Table 4. WFC3 and ACS Photometry^a of Candidates at 7 ≲ z < 8

Name	Photometric	R.A.	Decl.	F160W	F140W	F125W	F105W	F814W	μb
	Redshift	(J2000)	(J2000)
ZD1	$7.4^{+0.3}_{-0.6}$	3.603582	−30.382442	28.52 ± 0.19	28.09 ± 0.14	28.47 ± 0.21	>30.50	>31.00	$1.4 ^{+0.7} _{-0.1}$
ZD2c,d	7.9 ± 0.1	3.604520	−30.380472	25.56 ± 0.03	25.73 ± 0.03	25.91 ± 0.04	26.83 ± 0.07	>30.00	$1.3 ^{+0.7} _{-0.1}$
ZD3c,d	7.7 ± 0.1	3.606477	−30.380993	26.45 ± 0.04	26.57 ± 0.05	26.64 ± 0.05	27.50 ± 0.08	>31.00	$1.3 ^{+1.0} _{-0.1}$
ZD4	7.8 ± 0.3	3.605263	−30.380606	27.97 ± 0.12	27.95 ± 0.13	28.29 ± 0.17	29.05 ± 0.26	>31.00	$1.3 ^{+0.7} _{-0.1}$
ZD5d	$7.6 ^{+0.1}_{-0.3}$	3.588985	−30.378662	27.55 ± 0.05	27.77 ± 0.07	27.52 ± 0.06	28.29 ± 0.09	>30.00	$1.6 ^{+0.9} _{-0.1}$
ZD6c	7.5 ± 0.1	3.606575	−30.380928	26.38 ± 0.04	26.45 ± 0.04	26.86 ± 0.06	27.39 ± 0.08	>31.00	$1.3 ^{+1.1} _{-0.1}$
ZD7	7.3 ± 0.2	3.592285	−30.409911	26.98 ± 0.06	26.99 ± 0.07	26.92 ± 0.06	27.39 ± 0.07	>30.50	$5.9 ^{+7.8} _{-3.0}$
ZD8	7.5 ± 0.3	3.579668	−30.398678	28.33 ± 0.12	28.11 ± 0.11	28.29 ± 0.14	28.89 ± 0.17	>31.00	$14.0 ^{+38} _{-6.8}$
ZD9c	7.0 ± 0.1	3.603208	−30.410368	26.48 ± 0.04	26.68 ± 0.06	26.71 ± 0.06	26.83 ± 0.05	>31.00	$3.4 ^{+7.8} _{-0.8}$
ZD10	7.0 ± 0.3	3.581282	−30.404207	28.75 ± 0.16	28.61 ± 0.16	28.78 ± 0.19	28.98 ± 0.17	>31.00	$9.8 ^{+15} _{-4.4}$
ZD11c	7.0 ± 0.1	3.585321	−30.397964	27.49 ± 0.04	27.45 ± 0.04	27.33 ± 0.04	27.55 ± 0.03	>31.00	$4.7 ^{+1.8} _{-3.1}$

Notes. ^aMagnitudes are isophotal, scaled by an aperture correction term derived in the F160W band. The errors and limiting magnitudes are 1σ. Photometric redshifts are BPZ with 1σ error. ^bMagnification factor from Zitrin "NFW" model (see Section 4.1), where uncertainties are combined from two sources: (1) the model statistical uncertainties, which are relatively small; and more prominently, (2) the maximum differences with other six models, excluding one highest and one lowest values. ^cReported by A14 and Laporte et al. (2014). ^dAlso in Coe et al. (2014).

Download table as:  ASCIITypeset image

We adopt a concordance cosmology with Ω_M = 0.3, $\Omega _\Lambda =0.7$, and h = H₀/100 km s⁻¹ Mpc⁻¹ = 0.7, and the AB magnitude system throughout.

A2744 (z = 0.308) is the first HFF target in HST's Cycle 21. It is one of the most actively merging galaxy clusters known (Merten et al. 2011), displaying a large critical curve of roughly 60'' × 30''. The six HFF clusters have been selected to maximize the lensing boost, which means that systems with highly complex mass distributions (e.g., clusters in the process of merging) have been selected (Torri et al. 2004; Redlich et al. 2012; Zitrin et al. 2013b). The HFF observations of A2744 (GO/DD 13495, PI: Lotz) were carried out between 2013 October 25 and 2014 July 1. Additional WFC3/IR images obtained in 2013 August and 2014 June–July (GO 13386, PI: Rodney) and ACS images of 2009 (GO 11689, PI: Dupke) are retrieved from the Mikulski Archive for Space Telescopes (MAST¹⁷) and used. Table 1 lists the exposure times and limiting magnitudes for all the imaging used in our analysis.

We process the HST data using APLUS (Zheng et al. 2012a), an automated pipeline modified from the APSIS package (Blakeslee et al. 2003) with an enhanced capability of processing WFC3 data and aligning them with the ACS data. We retrieve the calibrated images from the HST instrument pipelines, namely the flc images for ACS (corrected for the detector charge transfer efficiency) and flt images for WFC3/IR. Recently, we have updated APLUS so that images of individual exposures are aligned using DrizzlePac (Hack et al. 2013), achieving an astrometric precision of ∼0015 or better. Figure 1 displays a composite color image of the cluster field.

Zoom In Zoom Out Reset image size

Using APLUS, we align, resample, and combine all the available imaging in each filter to a common pixel scale of 0065, which is half of WFC3/IR's pixel scale and slightly larger than that of ACS. We then create detection images from the inverse-variance weighted sum of the WFC3/IR and ACS images, respectively, and run SExtractor (Bertin & Arnouts 1996) in dual-image mode. The z ≳ 7 candidates are first selected from the NIR catalog derived from the WFC3/IR detection image, with a threshold of 1.5 times the signal-to-noise value over a minimum of 4 pixels. We choose colors measured from isophotal magnitudes to select our high-redshift candidates (see Section 3), as they balance the need between depth and photometric precision (Ferguson & McGaugh 1995). The 5σ limiting magnitude in the WFC3/IR bands is ∼28.8 in a 04 diameter aperture (see Table 1), and ∼29.2 for the observed-frame optical ACS bands.

As part of the HFF campaign, deep Spitzer/IRAC images were obtained in 2013 September and 2014 January and February in Channels 1 and 2 at wavelengths 3.1–3.9 and 3.9–5.0 μm, respectively, using Director's Discretionary Time (Program 90257, PI: Soifer). The effective exposure time in each channel, including that of the archival data (Program 84; PI: Rieke) obtained in 2004, is ∼339 ks. The IRAC corrected Basic Calibrated Data (cBCD) images are processed with MOPEX (Makovoz & Khan 2005) and sampled to a final pixel scale of 06. In order to perform background matching of the individual cBCD frames we run all steps in the Overlap module, where we use the SExtractor background estimation with a window size of 25 pixels. To create the mosaic images we run all three outlier modules (i.e. Dual outlier, Mosaic Outlier, and Box Outlier) and we use the default interpolation, with the fine resolution parameter = 0. The estimated 1σ limiting magnitude is 27.3 for IRAC channel 1 (IRAC1, 3.6 μm) and 27.1 for channel 2 (IRAC2, 4.5 μm; see Table 1).

We search for LBGs using their distinct color around 0.1216(1  +  z) μm. For example, at z ≃ 7–8, the Lyman break is at ∼1 μm, between the F814W and F125W bands. Our selection criteria, in units of magnitude, are as follows:

F814W − F105W > 0.8

F105W − F125W < 0.6

F814W − F105W > 0.8 + (F105W − F125W).

These color cuts are similar to those utilized in previous work such as Oesch et al. (2010).

For z ≃ 8–9, the break is at ∼1.15 μm, between the F105W and F140W bands:

F105W − F140W > 0.8

F140W − F160W < 0.6

F105W − F140W > 0.8 + (F140W − F160W).

For z ≃ 10, the break is between the F125W and F160W bands: F125W − F160W > 0.8.

We require that a candidate must not be detected above 1σ in a summed image blueward of the selection bands defined above. For objects at z ≃ 7, this requires a non-detection in a summed image of the F606W and F435W bands, while for candidates at z ≳ 8 this requires a non-detection in the stacked optical detection image.

In addition to the color selection criteria described above, we also exclude candidates lying within 1 arcsec of the detector edges, in order to mitigate potentially spurious photometry. We also exclude candidates lying near stellar diffraction spikes, which are difficult to remove because HFF WFC3/IR exposures were obtained at the same position angle. Finally, we also identify and remove candidates with a color decrement of F160W − IRAC >3, as they are most likely extremely red objects at lower redshift (z ≃ 2).

Our HST photometry is measured within an isophotal aperture, but aperture-corrected to a total flux using the ${mag} \_{auto} - {mag}\_{iso}$ difference in the F160W band. In a few cases where source blending is significant, we visually inspect the images and choose an aperture that is larger than the source's FWHM, but not so large as to be affected by nearby sources, and use the corresponding aperture magnitude in place of ${mag}\_{auto}$. We also verify that our aperture colors (and therefore our list of high-redshift candidates) are not affected by image artifacts if we use the publicly released HST mosaics based on the Mosaicdrizzle pipeline (Koekemoer et al. 2013).¹⁸

The IRAC images of our candidates suffer from crowding due to the instrument's large point spread function (PSF, FWHM ≃ 16), such that simple aperture photometry might result in inaccurate fluxes due to contamination from nearby sources. To address this issue, we use a deblending technique whereby contaminating neighbors are subtracted using GALFIT (Peng et al. 2010) by performing a fit to the objects of interest and all their close neighbors simultaneously in a ∼10'' × 10'' fitting window around the source of interest (Overzier et al. 2009; Zheng et al. 2012b). The IRAC PSF is determined from the same image by stacking 10 bright, isolated point sources. Positions and radial profiles of neighboring sources in this region are derived from the higher resolution HST F160W-band mosaic, while the initial input magnitudes are obtained by running SExtractor on the IRAC images. During the fitting process, all input parameters are allowed to vary, except for the positions of the objects of interest. This process is similar to that of Labbé et al. (2006) and Labbé et al. (2010). Eight of our candidates are so heavily blended by nearby bright sources with complex radial profiles that GALFIT fails to satisfactorily model contaminating emission from bright sources, leading to an over- or under-subtraction of fluxes and hence a large χ² value for the fit. As a result, the fluxes for these eight candidates are not measurable with reasonable uncertainties. Our analysis yields a total of 16 sources for which photometry or upper limits from GALFIT are possible (see Table 5).

We carry out extensive tests to check the reliability of our IRAC photometry for each source. First, we place a simulated point source of magnitude 25 near the candidate and run GALFIT with different fitting windows and background levels until the expected magnitude of each simulated source is recovered (with a magnitude difference <0.1 mag compared to the input value, consistent with the photometric errors). We then proceed to fit the flux of each candidate, using the fitting window and background level on the image that recovered the brightness of the simulated source. We repeat these tests at three different positions for the simulated source to verify our measurement of the source magnitude. To account for the uncertainties in estimating the background, we choose the average magnitude of three measurements as the source magnitude, which are reported in Table 5.

In Figures 2–4 we show cutout images of all the candidates.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

4.1. Gravitational Lensing Models

4.2. Photometric Redshifts

4.3. Multiple Systems

To help corroborate the high-redshift nature of our 24 candidates, we search for potential counter images near the locations predicted by the gravitational lensing model. Among the candidates that are inside or near the z = 7 critical curves, JD1 (Zitrin et al. 2014) and ZD7 are the only two cases where multiple images are found. For the others, the predicted counter images are either behind bright foreground galaxies, or too faint to be confirmed. Two of the counter images predicted for source ZD10 are near that of ZD11, which have been discussed by A14 (system 5).

The triple system ZD7 is shown in Figure 5 and Table 6. Image A is an arclet made of two components that are separated only by ∼1.5 kpc in the source plane, assuming a magnification of six. This is similar to a case in A1689 where a pair of LBGs at z ≃ 7.6 may be merging (Bradley et al. 2008). Image B is behind a bright foreground galaxy, which we subtract before carrying out photometry. We make use of the pure geometric scaling induced by strong lensing to estimate a purely geometric distance for this triply imaged case. The Zitrin "NFW" lensing model described in Section 4.1 is based on eleven sets of multiply lensed galaxies between 2 < z < 4, including a spectroscopic redshift of system 6 at z = 2.019 (Richard et al. 2014 ²¹). This spectroscopic redshift provides a normalization of the model so that the deflection field induced in the lens plane, $\boldsymbol {\alpha _L}(\boldsymbol {\theta })$, can be scaled to any redshift via the lensing source distance ratio f(z) = d_ls(z)/d_s(z) to provide the observed deflection field $\boldsymbol {\alpha }(\boldsymbol {\theta })={d_{ls}(z)/d_s(z)}\boldsymbol {\alpha _L}(\boldsymbol {\theta })$. Hence only a simple scaling of the relative lensing distance ratios is required to relate deflections at any given redshift to the lensing distance of the normalization used to calibrate the lens model, which in our case is f(z)/f(z = 2.019). We find that this factor is ∼1.12 for the triple system which minimizes the location of the observed images relative to that generated by the model, and this corresponds to a best BPZ estimate of z ≃ 7.4. A possible faint fourth image D is noted between images A and B, close to another bright galaxy (Figure 5).

Zoom In Zoom Out Reset image size

Table 6. Photometry^a of Multiple System

Name	R.A.	Decl.	F125W	F160W − F125W	F140W − F125W	F105W − F125W	F814W − F125W	μ
	(J2000)	(J2000)
ZD7A1	3.592410	−30.409897	27.33 ± 0.08	−0.08 ± 0.10	−0.05 ± 0.10	0.38 ± 0.11	>4.0	$5.7 ^{+7.5} _{-2.7}$
ZD7A2	3.592160	−30.409925	28.24 ± 0.12	−0.06 ± 0.16	0.09 ± 0.18	0.36 ± 0.18	>3.0	$6.4 ^{+13.2} _{-3.4}$
ZD7B	3.588430	−30.410340	27.7 ± 0.1	−0.3 ± 0.1	0.0 ± 0.1	0.6 ± 0.3	>2.0	$32.5 ^{+31} _{-32}$
ZD7C	3.600940	−30.400824	28.58 ± 0.15	0.25 ± 0.22	−0.21 ± 0.22	0.81 ± 0.28	>2.0	$2.8 ^{+1.5} _{-0.1}$

Note. ^aSee notes in Table 4.

Download table as:  ASCIITypeset image

5.1. Individual Candidates

5.2. Physical Properties

In addition to reporting on the discovery of our high-redshift candidates, we can also begin to characterize their physical properties. We defer a more detailed analysis of the full sample to a forthcoming paper. Here, we focus on the 10 objects at z > 7 with the highest-confidence photometric redshifts which have well-measured photometry in at least one IRAC channel. Photometry redward of the Balmer break is particularly important for placing meaningful constraints on the stellar mass and age of the stellar populations in these distant objects.

To infer the physical properties of these galaxies we use iSEDfit, but adopt a more restricted set of priors than the ones used to estimate photometric redshifts (see Section 4.2). Specifically, we adopt uniform priors on galaxy age t ∈ [10, 750] Myr, τ ∈ [10, 1000] Myr, stellar metallicity Z/Z_☉ ∈ [0.04, 1.0], and we assume no dust attenuation (e.g., Bouwens et al. 2010). Recall that the age of the universe at z = 7–8 in our adopted concordance cosmology is 750–630 Myr.

We find that our z > 7 candidates have demagnified stellar masses of around 10⁹ M_☉, and SFRs of approximately 4 M_☉ yr⁻¹. These results imply an average doubling time of around 500 Myr, which is comparable to the age of the universe at z ≃ 8.²² The ages of the galaxies in our sample are less well constrained given the uncertainties in our IRAC photometry; nevertheless, we find a median SFR-weighted age for the sample of ≲ 430 Myr (95% confidence interval), corresponding to a typical formation redshift of z ≲ 19. Figure 6 presents the SEDs of four representative galaxies in our sample, sorted by decreasing redshift, as well as the maximum likelihood fits derived using iSEDfit.

Zoom In Zoom Out Reset image size

5.3. Source Clustering

An apparent concentration of candidates northeast of the cluster center is shown in Figure 1. Nine objects at z ≃ 7–8 are found within a region of 20''. Since the average magnification in that area is not high (μ ≃ 1.4), this apparent overdensity is likely intrinsic rather than being due to lensing. In addition, Figure 7 shows a small region "Quintet" where five candidates are located within approximately 2'' of one another: objects YD1, YD4, YD6, YD7, and ZD1. These objects have similar estimated photometric redshifts of between 7.9 and 8.6, and their projected separations in the source plane are within ∼8 kpc. Given the uncertainties in our photometric redshifts, it is therefore possible that these sources are physically associated.

Zoom In Zoom Out Reset image size

To estimate the uncertainties in IRAC photometry for these sources whose separations are close to the IRAC PSF size, we place 500 sets of simulated sources of different brightness and compared the input and fitted fluxes. In more than 68% of the cases, the fitted results are within 0.5 mag of the inputs for three brighter sources in the F160W band, while for the faintest source the output flux decreases with the F160W-IRAC color. Due to the source confusion, our simulations suggest that the de-blending with GALFIT works better for the brighter objects, while it is biased to fainter objects in the sense that their fluxes are likely to be overestimated. This is because that GALFIT often crashes under a very faint flux level. We therefore conclude that we are able to deblend these sources with reasonable accuracy, except for the faint source ZD1 with large uncertainties.

Overdensities in the high-redshift domain have been previously reported. For example, Trenti et al. (2012) identified four candidates with z ≃ 8 within 70'' in the Field BoRG58. Bradley et al. (2012) found seven LBGs at z ≃ 7 in the WFC3/IR field (∼120'' × 130'') of A1703. However, our finding of nine LBGs at 7 < z < 9 within 20'' (∼80 kpc in the source plane) is unique, suggesting that the cosmic variance in source density (Trenti & Stiavelli 2008) is more significant than anticipated. It appears that cosmic variance increases with redshift (Moster et al. 2011; Bouwens et al. 2014b), as evidenced by different results from surveys. This trend might explain both the large number of clustered z ≃ 8 sources and the deficiency of z ≳ 9 sources and illustrate the need for observations in more fields.

We find 24 LBG candidates at 7 ≲ z ≲ 10.5 in the HFF imaging of A2744, reaching an intrinsic magnitude of ∼32. One source at z ≃ 7.4 is lensed into three images. Significant clustering is observed on the intrinsic scale of 10–100 kpc. Thanks to gravitational lensing, we are able to carry out Spitzer/IRAC photometry for 16 of the sources. SED fitting to the brightest candidates suggests stellar masses of ≃ 10⁹ M_☉, star formation rates of ≃ 4 M_☉ per year, and a typical formation redshift of z ≃ 19.

The redshift distribution of our sample is not a smoothly declining function towards higher redshift. In particular, our redshift distribution does not extend smoothly beyond z ≃ 9, clustering notwithstanding, Considering the effect of cosmic variance, the number density in our sample is consistent with that derived from other studies, e.g., Bouwens et al. (2014b). Given the level of clustering that we see in A2744, it will be important to average over more HFF, and to perform a luminosity function analysis so that the redshift dependence can be better related to galaxy mass.

The work presented in this paper is based on observations made with the NASA/ESA Hubble Space Telescope, and has been supported by award AR-13279 from the Space Telescope Science Institute (STScI), which is operated by the Association of Universities for Research in Astronomy, Inc. under NASA contract NAS 5-26555. It is also based on data obtained with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. This work utilizes gravitational lensing models produced by PIs Bradac, Ebeling, Merten & Zitrin, Sharon, and Williams, funded as part of the HST Frontier Fields program conducted by STScI. These models were calibrated using arcs identified in archival HST imaging by Merten et al. (2011), spectroscopic redshifts of arcs obtained using the VLT/FORS2 spectrograph (Richard et al. 2014), and VLT and Subaru/Suprimecam imaging of the A2744 field (Cypriano et al. 2004; Okabe & Umetsu 2008; Okabe et al. 2010a, 2010b). X.S. acknowledges support from FP7-SPACE-2012-ASTRODEEP-312725, and NSFC grants 11103017 and 11233002. Support for A.Z. is provided by NASA through Hubble Fellowship grant HST-HF-51334.01-A awarded by STScI. F.E.B. acknowledges support from Basal-CATA PFB-06/2007, CONICYT-Chile grants (FONDECYT 1141218, ALMA-CONICYT 31100004, Gemini-CONICYT 32120003, Anillo ACT1101), and the Millennium Institute of Astrophysics (Project IC120009). J.M.D. acknowledges support from the Spanish consolider project CAD2010-00064 and AYA2012-39475-C02-01. We thank M. Meneghitti for helpful comments.