THE BRIGHT END OF THE z ∼ 9 AND z ∼ 10 UV LUMINOSITY FUNCTIONS USING ALL FIVE CANDELS FIELDS^∗

The first galaxies are thought to have formed in the first 300–400 Myr of the universe. Over the last few years, remarkable progress has been made in extending samples back to this time, with more than ∼700 probable galaxies identified at z ≳ 6.3 with the Hubble Space Telescope (HST; McLure et al. 2013; Bouwens et al. 2015; Finkelstein et al. 2015) and 20–30 candidate galaxies identified as far back as redshifts z ∼ 9–11 (Bouwens et al. 2011a, 2014a, 2015; Zheng et al. 2012; Coe et al. 2013; Ellis et al. 2013; McLure et al. 2013; Oesch et al. 2013, 2014, 2015; Zitrin et al. 2014; Ishigaki et al. 2015; McLeod et al. 2015).

At present and over the next year, considerable resources are being devoted to both the discovery and study of ultra-faint galaxies with HST from the new Frontier Fields initiative (e.g., Lotz et al. 2014; Coe et al. 2015).⁹ The goal of this initiative is to combine the power of gravitational lensing from galaxy clusters with very deep exposures with the Hubble and Spitzer Space Telescopes. Eight hundred forty orbits of HST observations are being invested in deep optical/ACS + near-IR/WFC3/IR observations of six galaxy clusters. Deep observations of a "blank" field outside the galaxy clusters are also being obtained in parallel with observations over the clusters.

Despite the considerable focus by the community on the Hubble Frontier Fields observations over galaxy clusters and deep fields (e.g., Atek et al. 2014, 2015; Zheng et al. 2014; Coe et al. 2015; Ishigaki et al. 2015; Oesch et al. 2015), it is also possible to uncover modest numbers of luminous z ∼ 9–10 galaxies over wide-field surveys, as first illustrated by Oesch et al. (2014) through the identification of six intrinsically luminous z ∼ 9–10 candidate galaxies over the GOODS-North and GOODS-South CANDELS fields (Grogin et al. 2011; Koekemoer et al. 2011). These sources allowed us to set some initial constraints on the rate at which UV-luminous galaxies evolve with cosmic time, and provided some constraints on the approximate shape of the UV luminosity functions (LFs) at z = 9–10.

Due to the inherent brightness of such sources, these sources are also valuable for efforts to measure the physical properties of galaxies at very early times. Measurements of the UV-continuum slopes (Oesch et al. 2014; Wilkins et al. 2016), Balmer-break amplitudes (Oesch et al. 2014), stellar masses (Oesch et al. 2014), and sizes (Holwerda et al. 2015; Shibuya et al. 2015) can all be achieved using very bright galaxies.

Despite the usefulness of bright z ∼ 9–10 galaxies for addressing many contemporary science questions, current samples of these objects remain quite small and constraints on their volume densities are poor. Of particular note is that the recent Oesch et al. (2014) sample over the GOODS-North and GOODS-South fields only contained 2 bright z ∼ 9 and 4 bright z ∼ 10 galaxies. With such small samples, current uncertainties on the volume density of bright z ∼ 9–10 galaxies are large indeed (≳0.3–0.4 dex). This is especially the case when one considers the impact of field-to-field variations ("cosmic variance"), which is as large as a factor of two across the CANDELS fields, e.g., see Figure 14 from Bouwens et al. (2015), and may be even larger for the brightest sources (Bowler et al. 2015; Roberts-Borsani et al. 2016). Clearly, we require many independent lines of sight to the z ∼ 9–10 universe to average over the large-scale structure. Unfortunately, the Frontier Fields Initiative will not significantly help with this issue for luminous sources, given the limited area covered by observations from this program.

Nevertheless, there is a huge quantity of HST and Spitzer data already available that can be used to construct larger samples of bright z ∼ 9–10 galaxies. The most significant of these data sets are the ∼500 arcmin² CANDELS UDS, COSMOS, and EGS fields which feature very deep optical, near-IR, and Spitzer/IRAC observations. These observations are very useful for the robust detection of bright z ∼ 9–10 candidates and also to confirm a blue color redward of the break, distinguishing such galaxies from dusty, red galaxies at z ∼ 1–3. While possessing great potential, the CANDELS UDS, COSMOS, and EGS search areas lack correspondingly deep observations at 1.05 μm, just blueward of the Lyman-break in candidate z ∼ 9–10 galaxies, which is important for confirming a spectral break at ∼1.2 μm and distinguishing these z ≳ 9 galaxy candidates from Balmer-break sources at z ∼ 1–3.

Fortunately, we can overcome the aforementioned limitations of the CANDELS UDS, COSMOS, and EGS data sets by leveraging essentially all of the existing observations over these fields (Table 1) to first identify the highest-probability z ∼ 9–10 candidates over these fields and then obtaining targeted follow-up observations of these candidates at 1.05 μm to determine which are likely at z > 8 (Figure 1). In cycle 21, we successfully proposed such a follow-up program of plausible candidate z ∼ 9–10 galaxies over the CANDELS UDS, COSMOS, and EGS fields. Observations from this program—which we call z9 (redshift 9)-CANDELS (Bouwens 2014: GO 13792)—are now complete and cover all 12 of the primary candidates from that program. Based on the information we obtained from our proposed follow-up observations and the selection criteria we used in identifying our initial sample of 12 candidate z ∼ 9–10 galaxies from these three CANDELS fields, we can derive constraints on the volume density of luminous z ∼ 9–10 galaxies. Searches over CANDELS-GOODS-North, CANDELS-GOODS-South, the ERS fields can be further used to improve the constraints we obtain on the bright end of the z ∼ 9–10 LFs.

Zoom In Zoom Out Reset image size

Table 1.  Observational Data Used to Identify^a the Bright z ∼ 9–10 Candidate Galaxies over the CANDELS UDS, COSMOS, and EGS Fields*

Two-part Search Strategy (Preselection + Follow-up: Sections 3, 4)
CANDELS UDS			CANDELS COSMOS			CANDELS EGS
	5σ			5σ			5σ
Filtera	Depthb	Source	Filtera	Depthb	Source	Filtera	Depthb	Source
V606	26.8	HST/ACS	V606	26.5	HST/ACS	V606	27.3	HST/ACS
I814	26.8	HST/ACS	I814	26.5	HST/ACS	I814	27.1	HST/ACS
J125	26.3	HST/WFC3	J125	26.1	HST/WFC3	J125	26.4	HST/WFC3
JH140	26.1	HST/WFC3	JH140	25.8	HST/WFC3	JH140	25.6	HST/WFC3
H160	26.5	HST/WFC3	H160	26.3	HST/WFC3	H160	26.6	HST/WFC3
u	25.8	CFHT/Megacam	u	27.7	CFHT/Megacam	u	27.4	CFHT/Megacam
B	28.0	Subaru/Suprime-Cam	B + g	28.4	Subaru/Suprime-Cam +	g	27.8	CFHT/Megacam
V + r	28.0	Subaru/Suprime-Cam			CFHT/Megacam	r	27.6	CFHT/Megacam
i	27.4	Subaru/Suprime-Cam	V + r	27.9	Subaru/Suprime-Cam +	i + y	27.4	CFHT/Megacam
z	26.3	Subaru/Suprime-Cam			CFHT/Megacam	z	26.0	CFHT/Megacam
Y	25.9	VLT/HAWKI/HUGS	i + y	27.7	Subaru/Suprime-Cam +	K	24.1	UKIRT/WIRCam
J1	25.6	Magellan/FOURSTAR			CFHT/Megacam	3.6 μm	25.4	Spitzer/S-CANDELS
J2	25.7	Magellan/FOURSTAR	z	26.4	Subaru/Suprime-Cam +	4.5 μm	25.3	Spitzer/S-CANDELS
J	25.4	UKIRT/WFCAM			CFHT/Megacam
J3	25.4	Magellan/FOURSTAR	Y	26.1	UltraVISTA
H	24.6	UKIRT/WFCAM	J1	25.6	Magellan/FOURSTAR
Hs	25.0	Magellan/FOURSTAR	J2	25.5	Magellan/FOURSTAR
Hl	24.8	Magellan/FOURSTAR	J	25.3	UltraVISTA
Ks	25.5	VLT/HAWKI/HUGS +	J3	25.3	Magellan/FOURSTAR
		UKIRT/WFCAM +	Hs	24.7	Magellan/FOURSTAR
		Magellan/FOURSTAR	H	25.0	UltraVISTA
3.6 μm	25.4	Spitzer/S-CANDELS	Hl	24.7	Magellan/FOURSTAR
4.5 μm	25.4	Spitzer/S-CANDELS	Ks	25.3	UltraVISTA +
					Magellan/FOURSTAR
			3.6 μm	25.3	Spitzer/S-CANDELS
			4.5 μm	25.3	Spitzer/S-CANDELS
Direct Search Strategy for z ≥ 8.4 Galaxies (Section 5)
CANDELS GOODS-South			ERS			CANDELS GOODS-North
B435	27.1–27.3	HST/ACS	B435	27.1	HST/ACS	B435	27.2–27.3	HST/ACS
V606	27.4–27.7	HST/ACS	V606	27.4	HST/ACS	V606	27.4	HST/ACS
${i}_{775}+$			${i}_{775}+$			${i}_{775}+$
I814	27.5–27.6	HST/ACS	I814	27.3	HST/ACS	I814	27.2–27.7	HST/ACS
z850	26.8–26.9	HST/ACS	z850	26.7	HST/ACS	z850	26.9–27.0	HST/ACS
Y105	26.4–27.0	HST/WFC3	Y098	26.5	HST/WFC3	Y105	26.5–26.8	HST/WFC3
J125	26.5–27.0	HST/WFC3	J125	27.0	HST/WFC3	J125	26.4–27.2	HST/WFC3
JH140	26.1	HST/WFC3	JH140	25.8	HST/WFC3	JH140	25.6	HST/WFC3
H160	26.5–27.0	HST/WFC3	H160	26.9	HST/WFC3	H160	26.5–27.1	HST/WFC3
Ks	26.5	VLT/HAWKI/HUGS +	Ks	26.5	VLT/HAWKI/HUGS +
		VLT/ISAAC +			VLT/ISAAC +
		PANIC +			PANIC +
		Magellan/FOURSTAR			Magellan/FOURSTAR
3.6 μm	25.8	Spitzer/S-CANDELS	3.6 μm	25.8	Spitzer/S-CANDELS	3.6 μm	25.8	Spitzer/S-CANDELS
4.5 μm	25.8	Spitzer/S-CANDELS	4.5 μm	25.8	Spitzer/S-CANDELS	4.5 μm	25.8	Spitzer/S-CANDELS

Notes.

^aFor each source in our search fields, flux measurements are derived based on all of the observational data presented in this table. All of these measurements are used in deriving a redshift likelihood distribution for individual sources. ^bThe 5σ depths are estimated from the median 5σ uncertainties on the total flux measurements of sources found over our search fields with ${H}_{160,{AB}}$-band magnitudes of 26.0–26.5.

Download table as:  ASCIITypeset image

In this paper, we describe (1) the preselection we used to identify candidate z ∼ 9–10 galaxies from the CANDELS-UDS, COSMOS, and EGS fields for HST follow-up observations and (2) the results from this program. Our primary scientific objective is to obtain the best available constraints on the volume density of especially luminous z ∼ 9 and z ∼ 10 galaxies. Through such constraints, we have a direct measure of how fast (1) the bright end of the UV LF and (2) UV-luminous galaxies evolve. Through comparison with the volume density of fainter sources, the present search results also allow us to constrain the overall shape of the UV LF. Finally, we would expect our selection to allow us to considerably expand the overall sample of bright z ∼ 9 and z ∼ 10 galaxies available over the CANDELS fields. This has value both for the further characterization of the physical properties of z ≳ 9 galaxies and as bright sources to target with early James Webb Space Telescope observations. These bright samples will be further enhanced with bright z ∼ 9–10 galaxies from the BoRG${}_{[z910]}$ program (Trenti 2014).

Here, we present a brief plan for this paper. In Section 2, we include a description of the observational data that we use to identify high-probability z ∼ 9–10 galaxies over the CANDELS-UDS, COSMOS, and EGS fields. In Section 3, we describe our criteria for performing photometry and identifying high-probability z ∼ 9–10 galaxy candidates over the CANDELS-UDS, COSMOS, and EGS fields. In Section 4, we describe the results of the z9-CANDELS program where we use these observations to ascertain the likely nature of our selected z ∼ 9–10 candidate galaxies. In Section 5, we describe our search results for bright z ≳ 8.4 candidate galaxies over the CANDELS GOODS-North, GOODS-South fields, and ERS fields, extending previous work by Oesch et al. (2014: see also McLure et al. 2013 who also conducted such a search over the GOODS-South field). Finally, in Section 6, we make use of these search results to provide the first constraints on the bright end of the z ∼ 9 and z ∼ 10 LFs using a search over all five CANDELS fields.

For consistency with previous work, we quote results in terms of the luminosity ${L}_{z=3}^{* }$ derived by Steidel et al. (1999) at z ∼ 3, i.e., ${M}_{1700,{AB}}=-21.07$. We refer to the HST F435W, F606W, F775W, F814W, F850LP, F098M F105W, F125W, F140W, and F160W bands as B₄₃₅, V₆₀₆, i₇₇₅, I₈₁₄, z₈₅₀, Y₀₉₈, Y₁₀₅, J₁₂₅, JH₁₄₀, and H₁₆₀, respectively, for simplicity. Where necessary, we assume Ω₀ = 0.3, Ω_Λ = 0.7, and ${H}_{0}=70\,\mathrm{km}\,{{\rm{s}}}^{-1}\,{\mathrm{Mpc}}^{-1}$. All magnitudes are in the AB system (Oke & Gunn 1983).

In the present analysis, we conduct a search for bright z ∼ 9–10 candidate galaxies over the ∼450 arcmin² region within the CANDELS-UDS, COSMOS, and EGS fields with the deepest HST optical/ACS and near-IR/WFC/IR observations (∼75%–80% of the WFC3/IR area).

In conducting this search, we use the reductions of the HST observations described in Bouwens et al. (2015). Those reductions include all observations associated with the AEGIS, COSMOS, and CANDELS HST surveys and SNe follow-up programs, including the JH₁₄₀-band observations associated with the 3D-HST (Brammer et al. 2012) and AGHAST (Weiner et al. 2014) programs.

Beyond the HST observations themselves, perhaps the most valuable data set that we can leverage in our search for probable z ∼ 9–10 galaxies is the very deep Spitzer/IRAC S-CANDELS observations over the CANDELS fields (Ashby et al. 2015) which, when combined with Spitzer/IRAC SEDS observations (Ashby et al. 2013), reach 50 hr in depth (26.0 mag at 5σ: 2''-diameter apertures). Those observations provide us with constraints on the spectral slope of galaxies redward of the H₁₆₀ band which, when combined with evidence for a break across the J₁₂₅ and H₁₆₀ bands and a non-detection at optical wavelengths, is strongly suggestive of a z ∼ 9–10 galaxy.

In addition, we also make use of all significant, public ground-based observations over these fields, including optical observations from Subaru Suprime-Cam [CANDELS-COSMOS; CANDELS-UDS] and CFHT/Megacam [CANDELS-COSMOS; CANDELS-EGS], and deep near-IR observations from VISTA [CANDELS-COSMOS], UKIRT/WFCAM [CANDELS-UDS], VLT/HAWKI [CANDELS-UDS], Magellan/FOURSTAR [CANDELS-COSMOS; CANDELS-UDS], and CFHT/WIRCam [CANDELS-EGS]. The deep optical observations allow us to search for faint optical flux in the z ∼ 9–10 candidates identified over the CANDELS-UDS/COSMOS/EGS fields, while the near-IR observations allow us to test for the presence of a putative break at 1.2 μm, to verify that candidates show no flux blueward of the break, and to test for a flat UV-continuum redward of the break.

In our analysis of data over the five CANDELS fields and the ∼40 arcmin² ERS field (Windhorst et al. 2011), we use the Bouwens et al. (2015) reductions of the HST observations over all five CANDELS fields, the version 7 reduction of the deep CFHT legacy survey observations over the COSMOS and EGS fields,¹⁰ the public v2.0 reductions of the UltraVISTA observations (McCracken et al. 2012), the Cirasuolo et al. (2010) redutions of the deep Subaru Suprime-Cam observations over the UDS/SXDS field (Furusawa et al. 2008), the Bouwens et al. (2015) reductions of the HUGS HAWK-I observations (Fontana et al. 2014), the public reductions of the WIRCam deep survey K_s-band observations over the CANDELS EGS field (McCracken et al. 2010; Bielby et al. 2012), the v0.9.3/v0.95.5 reductions of the ZFOURGE COSMOS/UDS observations (I. Labbé et al. 2015, in preparation), the IUDF reductions of the Spitzer/IRAC observations over the GOODS-South and GOODS-North fields (Labbé et al. 2015), and the public reductions of the Spitzer SEDS and S-CANDELS programs (Ashby et al. 2013, 2015).

Table 1 provides a convenient summary of all of the observational data we use. Combining the flux measurements from the different data sets, the 5σ depths of these fields (derived from the median uncertainties on the total flux measurements) range from ∼28 mag at <0.8 μm, ∼26.0–26.5 mag at ∼0.9 μm, ∼26.0 mag at 1.05 μm, ∼26.6 mag at ∼1.2–1.6 μm, 24.1–25.5 mag at 2.3 μm, to 25.6–25.9 mag at 3.6 μm + 4.5 μm. We refer interested readers to Figure 3 from Bouwens et al. (2015) for a graphical representation of these depths as a function of wavelength.

3.1. Catalog Construction and Photometry

3.2. Selection of Bright z ∼ 9–10 Candidates over the CANDELS-UDS and CANDELS-COSMOS Fields

3.2.1. Selection Criteria

In searching for candidate galaxies at z ∼ 9–10, we suppose that these galaxies have colors and SEDs very similar to galaxies at slightly lower redshifts. Specifically, we would expect these sources to show a sharp spectral break at 1216 Å due to strong absorption from the neutral hydrogen forest, and to exhibit a blue UV-continuum redward of the break.

For star-forming galaxies at redshifts z ∼ 8.4 and higher, the Lyman-break will already have redshifted a significant way through the J₁₂₅-band, yielding moderately red ${J}_{125}-{H}_{160}$ colors. As a result, the selection of all sources with red ${J}_{125}-{H}_{160}$ colors should allow us to identify the bulk of star-forming galaxies from z ∼ 8.7 to z ∼ 11 (particularly if those galaxies are not substantially dust obscured).

Here, we search for candidate z ≳ 8.4 galaxies over the CANDELS-UDS and COSMOS fields using a ${J}_{125}-{H}_{160}\gt 0.5$ criterion. Star-forming galaxies with a UV-continuum slope β of −1.6 (typical of luminous galaxies at z = 4–7) would have a ${J}_{125}-{H}_{160}$ color of ∼0.5 at z = 8.7, but the lower-redshift limit for our selection will depend on the intrinsic colors of individual galaxies and also can be affected by observational noise.

In addition to our ${J}_{125}-{H}_{160}$ criterion, we also require that sources be undetected (<2σ) in the V₆₀₆ or I₈₁₄ bands. Sources where the root mean square signal-to-noise ratio (S/N) in the V₆₀₆ and I₈₁₄ bands is greater than 1 are excluded. In addition to these non-detection requirements on the HST optical data, we also require that sources remain undetected (<2.5σ) in an inverse-variance-weighted mean stack of the ground-based optical data.

We also demand that sources show ${H}_{160}-[3.6]$ colors bluer than 1.4 mag to exclude intrinsically red or old z ∼ 2 galaxies from our samples, similar to the criteria applied by Oesch et al. (2014) or Bouwens et al. (2015). This particular color cut corresponds to a UV-continuum slope β of 0.0 (where ${f}_{\lambda }\propto {\lambda }^{\beta }$), which is approximately as red as bright galaxies are observed to be at z ∼ 6–8 (e.g., Wilkins et al. 2011; Bouwens et al. 2012, 2014b; Finkelstein et al. 2012; Rogers et al. 2014).

We require that all selected z ∼ 9–10 candidates show strong evidence of correspondence to real sources. We therefore require that (1) sources be detected in the H₁₆₀ band at >5σ significance, (2) the root mean square detection significance of sources in the JH₁₄₀- and H₁₆₀-band images be at least 6, and (3) the root mean square detection significance of sources in the JH₁₄₀, [3.6], [4.5], and K bands be at least 2σ.

Finally, in the last step, we compute the redshift likelihood distribution for each candidate source using the EAZY photometric redshift code (Brammer et al. 2008) based on the photometry we have available for sources, the standard EAZY_v1.0 template set, and a flat prior. We supplemented the standard EAZY_v1.0 template set with SED templates from the Galaxy Evolutionary Synthesis Models (Kotulla et al. 2009). Nebular continuum and emission lines were added to the later templates using the Anders & Fritze-v. Alvensleben (2003) prescription, a 0.2 Z_⊙ metallicity, and a rest-frame EW for Hα of 1300 Å (which appears to be appropriate for z ∼ 6–7; Smit et al. 2014, 2015; Roberts-Borsani et al. 2016).

The photometry used for constraining the likelihood distributions for individual sources included the HST ${V}_{606}{I}_{814}{J}_{125}{{JH}}_{140}{H}_{160}$ + Subaru-SuprimeCam BgVriz + CFHT/Megacam ugriyz + UltraVISTA YJHK_s + ZFOURGE ${J}_{1}{J}_{2}{J}_{3}{H}_{s}{H}_{l}$ + Spitzer/IRAC 3.6 μm + 4.5 μm S-CANDELS data sets for the CANDELS COSMOS field, HST ${V}_{606}{I}_{814}{J}_{125}{{JH}}_{140}{H}_{160}$ + Subaru-SuprimeCam BVriz + CFHT/Megacam u + UKIRT/WFCAM JHK_s + ZFOURGE ${J}_{1}{J}_{2}{J}_{3}{H}_{s}{H}_{l}$ + VLT/HAWKI/HUGS YK_s data sets for the CANDELS UDS field, and the HST ${V}_{606}{I}_{814}{J}_{125}{{JH}}_{140}{H}_{160}$ + CFHT/Megacam ugriyz + CFHT/WIRCam K_s + Spitzer/IRAC 3.6 μm+4.5 μm data sets for the CANDELS EGS field. The depths of these observations are provided in Table 1 and their areal coverage is illustrated in Figure 2.

Zoom In Zoom Out Reset image size

Sources that satisfied our aforementioned criteria, which showed a >50% probability of being at z > 8, and which could be confirmed to be a >90% likelihood candidate with a single orbit of follow-up observations (supposing sources are measured to have a flux of 0 ± 12 nJy in the Y₁₀₅ band), made it into our final preselection of candidate z ∼ 9–10 galaxies (to be targeted with follow-up observations). In computing the posterior probability that a source has a redshift of z > 8 or z < 8, we adopt a flat prior on the redshift.

Table 2 provides a convenient compilation of all of the selection criteria we employed in preselecting candidate z ∼ 9–10 galaxies to follow-up with targeted observations.

Table 2.  Selection Criteria Used in Assembling Our z ∼ 9 and z ∼ 10 Samples

Redshift	Selection Criteria
Sample	Preselection for Targeted HST Follow-up	After HST Follow-up
CANDELS-UDS + CANDELS-COSMOS
9	$({J}_{125}-{H}_{160}\gt 0.5)\wedge ({H}_{160}-[3.6]\lt 1.4)\wedge$ (S/N in both V606 and ${I}_{814}\lt 2$)∧	$(P(z\gt 8)\gt 0.9)\wedge (8.4\lt {z}_{\mathrm{phot}}\lt 9.5)$
	(rms S/N in V606 and ${I}_{814}\lt 1)\wedge$ (S/N(H160)$\gt 5)\wedge ({\chi }_{{{JH}}_{140}{H}_{160}}^{2})\gt 36)\wedge$
	(${\chi }_{K,[3.6],[4.5]}^{2}\gt 2)\wedge ({P}_{\mathrm{pre}}(z\gt 8)\gt 0.5)\wedge ({P}_{\mathrm{post}}(z\gt 8)\gt 0.9)$
10	idem	$(P(z\gt 8)\gt 0.9)\wedge (9.5\lt {z}_{\mathrm{phot}}\lt 11)$
CANDELS-EGS
9	$({J}_{125}-{H}_{160}\gt 0.5)\wedge ({H}_{160}-[3.6]\lt 1.4)\wedge$ (S/N in both V606 and ${I}_{814}\lt 2$)∧	$(P(z\gt 8)\gt 0.9)\wedge (8.4\lt {z}_{\mathrm{phot}}\lt 9.5)$
	(rms S/N in V606 and ${I}_{814}\lt 1)\wedge$ (S/N(H160)$\gt 5)\wedge ({\chi }_{{{JH}}_{140}{H}_{160}})\gt 6)\wedge$
	(${\chi }_{K,[3.6],[4.5]}^{2}\gt 2)\wedge ({P}_{\mathrm{post}}(z\gt 8)\gt 0.9)$
10	idem	$(P(z\gt 8)\gt 0.9)\wedge (9.5\lt {z}_{\mathrm{phot}}\lt 11)$
CANDELS-GOODS-North + CANDELS-GOODS-South + ERS
9	((Y-dropout) criterion from Bouwens+2015)$\,\vee \,({J}_{125}-{H}_{160}\gt 0.5))\wedge ({H}_{160}-[3.6]\lt 1.4)\wedge$
	(S/N in both V606 and ${I}_{814}\lt 2)\wedge ({\chi }_{{B}_{435}{V}_{606}{i}_{775}{I}_{814}{z}_{850}}^{2}\lt 4)\wedge$
	$(P(z\gt 8)\gt 0.8)\,\wedge \,$($8.4\lt {z}_{\mathrm{phot}}\lt 9.5$)
10	((Y-dropout) criterion from Bouwens+2015)$\,\vee \,({J}_{125}-{H}_{160}\gt 0.5))\wedge ({H}_{160}-[3.6]\lt 1.4)\wedge$
	(S/N in both V606 and ${I}_{814}\lt 2)\wedge ({\chi }_{{B}_{435}{V}_{606}{i}_{775}{I}_{814}{z}_{850}}^{2}\lt 4)\wedge$
	$(P(z\gt 8)\gt 0.8)\,\wedge \,$($9.5\lt {z}_{\mathrm{phot}}\lt 11.0$)

Note.

^aRedshift likelihood probability P(z) are computed using our flux meaurements in all photometric bands listed in Table 1. P_pre(z > 8) indicates the probability that a source has a redshift greater than 8 before acquiring any follow-up observations, while P_post(z > 8) indicates the probability that a source has a redshift greater than 8 after obtaining the 1-orbit of follow-up HST observations (assuming the follow-up observations yielded a measured flux of 0 ± 12 nJy in the Y₁₀₅-band filter).

Download table as:  ASCIITypeset image

3.2.2. UDS+COSMOS Results

Applying the selection criteria from the previous section to our source catalogs over the CANDELS-UDS and CANDELS-COSMOS fields, we found five sources which satisfied all of the criteria. A list of all 6 sources satisfying these criteria are included in Table 3 along with similar candidates from the CANDELS-EGS field.

Table 3.  z ∼ 9–10 Candidate Galaxies over the CANDELS UDS, COSMOS, and EGS Program Targeted with Our z9-CANDELS Follow-up Program

ID	R.A.	decl.	H160,AB	${z}_{\mathrm{phot},\mathrm{pre}}$ a	Ppre(z > 8)a	${z}_{\mathrm{phot},\mathrm{post}}$ b	Ppost (z > 8)b
z = 9–10 Candidates Preselected for Targeted Follow-Up Observations with HST
COS910-0	10:00:43.16	02:25:10.5	26.2 ± 0.1	9.1	0.72	7.8	0.47
COS910-1	10:00:30.34	02:23:01.6	26.4 ± 0.2	9.0	0.95	9.0	0.99
COS910-2	10:00:14.91	02:12:10.8	26.3 ± 0.2	9.3	0.74	9.3	0.37
COS910-3	10:00:27.98	02:11:49.5	25.9 ± 0.1	9.2	0.63	2.3	0.27
UDS910-0	02:17:55.50	−05:11:41.3	26.4 ± 0.2	8.8	0.72	1.7	0.09
UDS910-1	02:17:21.96	−05:08:14.7	26.6 ± 0.2	8.7	0.74	8.6	0.74
EGS910-0	14:20:23.47	53:01:30.5	26.2 ± 0.1	9.1	0.67	9.1	0.92
EGS910-1	14:20:21.54	52:57:58.4	26.6 ± 0.1	8.9	0.19d	0.4	0.02
EGS910-2	14:20:44.31	52:58:54.4	26.7 ± 0.2	9.6	0.69	9.6	0.71
EGS910-3	14:19:45.28	52:54:42.5	26.4 ± 0.2	8.9	0.64	9.0	0.97
EGS910-4	14:19:23.59	52:49:23.4	26.2 ± 0.2	9.2	0.10d	1.0	0.02
EGS910-5	14:19:11.08	52:46:25.7	25.8 ± 0.1	9.2	0.28	1.8	0.11
z ∼ 9–10 Galaxy Candidates Targeted at No Additional Cost (Not Preselected)c
EGS910-6	14:19:13.84	52:50:44.7	26.6 ± 0.2	9.3	0.40	7.0	0.00
EGS910-7	14:20:23.72	53:01:38.3	26.0 ± 0.1	⋯	⋯	2.4	0.18

Notes.

^aBest-fit z > 4 redshift and integrated z > 8 likelihood for source derived from our HST+Spitzer/IRAC+ground-based photometry (Table 1) before obtaining observations from our z9-CANDELS follow-up program. ^bBest-fit redshift and integrated z > 8 likelihood for source derived from our HST+Spitzer/IRAC+ground-based photometry (Table 1) after obtaining observations from our z9-CANDELS follow-up program. ^cThese sources could be fit within the same WFC3/IR tiles as our primary targets, and hence required no additional HST time to investigate. ^dOver the CANDELS EGS field, we selected sources which, if they showed a null detection in the Y₁₀₅ band in a 1-orbit integration, could be confirmed with >90% probability to lie at z > 8. While these two sources initially only showed a modest probability for being at z > 8, their SEDs were nevertheless consistent with lying at z > 8 (particularly if a null detection at 1.05 μm could be confirmed).

Download table as:  ASCIITypeset image

The observed spectral energy distributions for these five candidate z ∼ 9–10 galaxies are presented in Appendix A (Figure 11), along with SED fits to a model z > 8 galaxy and a model z < 3 galaxy. Also shown in this figure is the redshift likelihood distribution (solid black line) based on the photometry we have available for each candidate in the ∼20 different wavelength channels (HST + Spitzer/IRAC + ground-based observations). In addition, this figure presents the redshift likelihood distribution we would expect assuming that these candidates are not detected in the single orbit of follow-up Y₁₀₅-band observations from the z9-CANDELS program.

Postage stamp images of these six candidates are also presented in Appendix A (Figure 10). As should be obvious from this figure, all six of the present z ∼ 9–10 candidates show clear detections in the H₁₆₀ band, as well as significant ∼2–3σ detections in the J₁₂₅-band and JH₁₄₀-band observations (where available), as well as in the S-CANDELS Spitzer/IRAC data.

All six of these candidates bear a remarkable similarity to the first samples of particularly luminous z ∼ 9–10 galaxies identified by Oesch et al. (2014) in terms of their very blue ${H}_{160}-[3.6]$ colors (see also Wilkins et al. 2016), red [3.6]–[4.5] colors, and observed sizes (Holwerda et al. 2015).

3.3. Selection of z ∼ 9–10 Candidates Over the CANDELS-EGS

HST observations are now available over all 12 candidate z ∼ 9–10 galaxies targeted by our z9-CANDELS program. These observations allow us to make a fairly definitive assessment of the nature of these candidate z ∼ 9–10 galaxies based on the flux we measure for these candidates at 1.05 μm. One orbit of Y₁₀₅-band observations has already been obtained for eight candidates targeted by our program COS910-0, COS910-1, COS910-3, UDS910-0, UDS910-1, EGS910-0, EGS910-1, EGS910-3, and EGS910-4. Slightly shallower observations (i.e., 1/3 and 2/3 of an orbit) in the Y₁₀₅ band were acquired for the candidates EGS910-5 and COS910-2 due to the greater brightness of the former candidate and the utility of an additional 1/3 orbit JH₁₄₀-band observations to investigate the nature of the potential z ∼ 10 candidate galaxies EGS910-2 and COS910-3.

Y₁₀₅-band images for these candidates are presented in either Figure 3 or Figure 12 from Appendix B, in conjunction with images of these candidates at other wavelengths. Figure 4 and Figure 13 from Appendix B show the observed SEDs for the targetted z ∼ 9–10 candidates in the CANDELS program.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The present observations photometrically confirm 5 of the first 12 z ∼ 9–10 candidates targeted by our program. Two of these five confirmations are only partial confirmations (EGS910-2 and UDS910-1: more observations are needed for these candidates to be >90% secure). Detailed remarks on the confirmed z ∼ 9–10 candidates can be found here.

• COS910-1: COS910-1 is not detected (<1σ) in the Y₁₀₅-band follow-up observations at 1.05 μm. A detailed fit to its SED suggests that it is actually a star-forming galaxy at z = 9.1, with <0.7% probability of it corresponding to a z < 8 galaxy.
• UDS910-1: UDS910-1 is not detected at (<1σ) in the Y₁₀₅-band follow-up observations we obtained at 1.05 μm. Rederiving the redshift likelihood distribution using the new flux information in the Y₁₀₅ band, we compute a best-fit photometric redshift of 8.6 with 4% and 24% probabilities of corresponding to a z < 7 and z < 8 source, respectively.
• EGS910-0: EGS910-0 is not detected (<1σ) in the Y₁₀₅-band follow-up observations at 1.05 μm. Rederiving the redshift likelihood distribution using the new flux information in the Y₁₀₅ band, we compute a best-fit photometric redshift of 9.1 with only a 4% probability of corresponding to a z < 8 source.
• EGS910-2: Follow-up of EGS910-2 in the JH₁₄₀ band shows a clear 2.6σ detection of the source which is in excellent agreement with the expected flux given a model redshift of z ∼ 9.6 for the source. Nevertheless, the source is sufficiently faint that the redshift likelihood distribution shows a 29% likelihood of the source being at z < 8. Deeper follow-up observations at 1.05 μm will be required to rule out the z < 8 solution.
• EGS910-3: EGS910-3 shows no detection (<1σ) in the Y₁₀₅-band follow-up observations we obtained at 1.05 μm. Rederiving the redshift likelihood distribution using the new flux information in the Y₁₀₅ band, we compute a best-fit photometric redshift of 9.0 with a 3% probability of corresponding to a z < 8 source.

The three confirmed z ≳ 8.4 candidates—and two partially confirmed candidates—from our z9-CANDELS program are compiled for convenience in Table 4. We will also include in this table some additional candidates we identify in Section 5 (also as identified by Oesch et al. 2014 and Bouwens et al. 2015).

Table 4.  Photometrically Confirmed z ∼ 9–10 Galaxies over the CANDELS Fields

ID	R.A.	decl.	${H}_{160,{AB}}$	zphota	P(z > 8)	References
z ∼ 9 Sample
Two-part Search Strategy (Preselection + Follow-up: Sections 3, 4):
COS910-1	10:00:30.34	02:23:01.6	26.4 ± 0.2	${9.0}_{-0.5}^{+0.4}$	0.99
EGS910-0	14:20:23.47	53:01:30.5	26.2 ± 0.1	${9.1}_{-0.4}^{+0.3}$	0.92
EGS910-3	14:19:45.28	52:54:42.5	26.4 ± 0.2	${9.0}_{-0.7}^{+0.5}$	0.97
UDS910-1b	02:17:21.96	−05:08:14.7	26.6 ± 0.2	${8.6}_{-0.5}^{+0.6}$	0.74
Direct Search Strategy for z ≥ 8.4 Galaxies (Section 5):
GS-z9-1	03:32:32.05	−27:50:41.7	26.6 ± 0.2	9.3 ± 0.5	0.9992	(1), (2)
GS-z9-2	03:32:37.79	−27:42:34.4	26.9	${8.9}_{-0.3}^{+0.3}$	0.83	(2)
GS-z9-3	03:32:34.99	−27:49:21.6	26.9	${8.8}_{-0.3}^{+0.3}$	0.95	(2), (3)
GS-z9-4	03:33:07.58	−27:50:55.0	26.8	${8.4}_{-0.3}^{+0.2}$	0.97	(2), (3)
GS-z9-5	03:32:39.96	−27:42:01.9	26.4	${8.7}_{-0.7}^{+0.8}$	0.55	(2)
GN-z9-1	12:36:52.25	62:18:42.4	26.6 ± 0.1	9.2 ± 0.3	>0.9999	(1), (2)
z ∼ 10 Sample
Two-part Search Strategy (Preselection + Follow-up: Sections 3, 4):
EGS910-2b	14:20:44.31	52:58:54.4	26.7 ± 0.2	${9.6}_{-0.5}^{+0.5}$	0.71
Direct Search Strategy for z ≥ 8.4 Galaxies (Section 5):
GN-z10-1c	12:36:25.46	62:14:31.4	26.0 ± 0.1	11.1 ± 0.1	>0.9999	(1), (2), (4), (5)
GN-z10-2	12:37:22.74	62:14:22.4	26.8 ± 0.1	9.9 ± 0.3	0.9994	(1), (2)
GN-z10-3	12:36:04.09	62:14:29.6	26.8 ± 0.2	9.5 ± 0.4	0.9981	(1), (2)
GS-z10-1	03:32:26.97	−27:46:28.3	26.9 ± 0.2	9.9 ± 0.5	0.9988	(1), (2)

Notes.

^a1σ uncertainties are computed based on the z > 4 likelihood distributions. ^bThis candidate could only be partially confirmed, given the limited orbit allocation to our HST program. ^cThis source is now spectroscopically confirmed to lie at z = 11.1 (Oesch et al. 2016), but broadly lies within our z ∼ 10 selection window.

References. (1) Oesch et al. (2014), (2) Bouwens et al. (2015), (3) McLure et al. (2013), (4) Oesch et al. (2016), (5) Bouwens et al. (2010).

Download table as:  ASCIITypeset image

Our overall confirmation rate is 42% (5/12) for sources preselected by our criteria. We achieve an even higher 56% success rate targeting those sources from our selection which are high-probability (>50%) z > 8 galaxies before our follow-up observations. While imperfect, this program is very efficient, supplementing some 270 orbits of HST time and hundreds of hours of Spitzer time with only 11 orbits of additional HST time. By contrast, the CANDELS + ERS programs over the GOODS-North + South cost some ∼500 orbits, and we identified only 9 candidates in those data, or 0.02 z ∼ 9–10 candidates per invested orbit.

Detailed remarks on the z ∼ 9–10 candidate galaxies that were not confirmed by our follow-up program are included in Appendix A.

We can obtain the best constraints on the volume density of bright z ∼ 9–10 galaxy candidates by not simply considering a search over the CANDELS UDS, COSMOS, and EGS fields as we did in the previous sections, but also considering a search for similar sources over the GOODS-North and GOODS-South fields.

5.1. Criteria for Identifying z ∼ 8.5–9.0 Galaxies

The purpose of the present section is to obtain a complete census of the bright z ∼ 8.5–11 galaxy candidates over the CANDELS GOODS-North+GOODS-South + ERS fields.

In Oesch et al. (2014) and Bouwens et al. (2015), we had already conducted a significant search for galaxies in this redshift range by looking for sources with red ${J}_{125}-{H}_{160}\gt 0.5$ colors and blue ${H}_{160}-[3.6]\lt 1.4$ colors. However, such a selection is only sensitive to galaxies with redshifts z ≳ 9 and can suffer significant incompleteness at z < 9.

Here, we extend the search from Oesch et al. (2014) and Bouwens et al. (2015) to also consider sources with redshifts z ≳ 8.4. We select these sources by considering all those sources which satisfy the z ∼ 8 color–color criteria of Bouwens et al. (2015), deriving photometric redshifts for all such sources using the EAZY photometric redshift code (Brammer et al. 2008) and including those sources where the most likely redshift is greater than 8.4.

The photometry we consider in deriving the redshift likelihood contours are the Bouwens et al. (2015) reductions of the HST B₄₃₅V₆₀₆i₇₇₅I₈₁₄z₈₅₀Y₀₉₈Y₁₀₅J₁₂₅JH₁₄₀H₁₆₀ data, the Labbé et al. (2015) reductions of essentially all Spitzer/IRAC observations over the GOODS-North and South fields, and the Bouwens et al. (2015) reductions of the HUGS HAWK-I K_s-band observations.

Briefly, the Bouwens et al. (2015) selection criteria for identifying $z\sim 8$ sources is

$$\begin{eqnarray*}&&({Y}_{105}-{J}_{125}\gt 0.45)\wedge ({J}_{125}-{H}_{160}\lt 0.5)\wedge \\ &&\quad ({Y}_{105}-{J}_{125}\gt 0.75({J}_{125}-{H}_{160})+0.525)\end{eqnarray*}$$

for sources over the CANDELS GOODS-North + GOODS-South fields and

$$\begin{eqnarray*}&&({Y}_{098}-{J}_{125}\gt 1.3)\wedge ({J}_{125}-{H}_{160}\lt 0.5)\wedge \\ &&\quad ({Y}_{098}-{J}_{125}\gt 0.75({J}_{125}-{H}_{160})+1.3)\end{eqnarray*}$$

for sources over the ∼40 arcmin² ERS field. Sources are required to be detected at 6σ in a χ² stack of the H₁₆₀-band or ${{JH}}_{140}+{H}_{160}$ band observations redward of the break (in a fixed 036-diameter aperture).

To ensure that contamination is kept to a minimum, an optical "χ²" is computed for each candidate source (Bouwens et al. 2011b) based on the flux in the ${B}_{435}{V}_{606}{i}_{775}{I}_{814}{z}_{850}$-band observations. ${\chi }_{\mathrm{opt}}^{2}$ is taken to equal ${{\rm{\Sigma }}}_{i}\mathrm{SGN}({f}_{i}){({f}_{i}/{\sigma }_{i})}^{2}$, where f_i is the flux in band i in a consistent aperture, σ_i is the uncertainty in this flux, and SGN(f_i) is equal to 1 if f_i > 0 and −1 if f_i < 0. Any candidate with a measured χ_opt in excess of 4 is excluded from our selections.

We only search for z ≳ 8.4 sources over the GOODS-North and GOODS-South fields brightward of ${H}_{160,{AB}}=27$ mag to ensure that we have strong constraints on the nature of the selected sources to the limit of our search. The effective depth of our z ∼ 9–10 search over the CANDELS-UDS, COSMOS, and EGS fields is also approximately ∼27 mag, and so the effective depth of our search is similar across all five CANDELS fields that we use.

5.2. Selection Results

Using the selection criteria from the previous section, we identify three high-probability (>90% confidence) and one moderate probability (∼50% confidence) z ∼ 8.4–9.0 galaxies over the ERS, CANDELS GOODS-South, and CANDELS GOODS-South fields.

The H₁₆₀-band magnitudes of the z ∼ 8.4–9.0 galaxies we have selected range from 26.4 and 26.9, similar to those found for our z ∼ 9–10 sample over the CANDELS-UDS, COSMOS, and EGS fields. We have included the four new z ∼ 8.4–9.0 candidates in Table 4 along with other high-probability z ∼ 8.4–11 candidates identified here. Fits to the observed SEDs for our new z ∼ 8.4–9.0 candidates over these fields are shown in Figure 5. Postage stamp images of the candidates are provided in Figure 6.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

5.3. Criteria for Identifying z ≳ 9.0 Galaxies

From previous work (Wyithe et al. 2011; Barone-Nugent et al. 2015; Fialkov & Loeb 2015; Mason et al. 2015), it is well known that gravitational lensing from foreground galaxies can have a particularly significant effect in enhancing the surface density of bright z ≥ 6 galaxies on the sky. This is especially true for the brightest sources due to the intrinsic rarity and the large path length available for lensing by foreground sources.

Given this phenomenon, it has become increasingly common for researchers searching for the brightest z ∼ 6–10 galaxies to look for possible evidence of lensing amplification (Bowler et al. 2014, 2015; Oesch et al. 2014; Zitrin et al. 2015; Roberts-Borsani et al. 2016). While there are a number of cases where such magnification boosts may be present (e.g., Barone-Nugent et al. 2015; Roberts-Borsani et al. 2016), the fraction of lensed sources among bright samples still does not appear to be particularly high (Bowler et al. 2015).

As in the above work, we explicitly check our compilation of bright z ∼ 9–10 galaxy candidates from these fields for evidence of gravitational lensing. For convenience, we use the Skelton et al. (2014) catalogs providing radii and stellar mass estimates for all sources over the CANDELS areas that we have searched. The Skelton et al. (2014) catalogs use the diverse multi-wavelength data over the CANDELS fields, including HST optical, near-infrared, Spitzer/IRAC, and ground-based observations, to provide flux measurements of a wide wavelength range, and then use these flux measurements to estimate the redshifts and stellar masses.

As in Roberts-Borsani et al. (2016), we model galaxies in our bright z ∼ 9–10 sample as singular isothermal spheres and use the measured half-light radius and inferred stellar mass to derive velocity dispersion estimates for individual galaxies in these samples. We found only two examples of galaxies whose measured fluxes appear likely to be slightly boosted (>0.1 mag) by lensing amplification.

• EGS910-3: There is a foreground galaxy at z ∼ 1.9 with an estimated stellar mass of a 10^10.32-M_⊙ that lies within 1.9 arcsec of this source. Based on the velocity dispersion we estimate for this source, ∼220 km s⁻¹, we compute a magnification boost of 0.25 mag for this source.
• GN-z10-2: This source is estimated to be boosted by 0.11 mag by a 10^10.64 M_⊙ galaxy with a spectroscopic redshift of z = 1.02 (Barger et al. 2008) that lies within 40 of the targeted source. This source was previously flagged by Oesch et al. (2014) as being slightly lensed.

7.1. Constraints on the UV LFs at z = 9 and z = 10

In this section, we use the combined sample of z ∼ 9–10 candidates over the CANDELS-UDS, CANDELS-COSMOS, and CANDELS-EGS fields and similar z ∼ 9–10 candidates over the CANDELS GOODS-North and GOODS-South fields (Section 5) to quantify the UV LFs at z ∼ 9 and z ∼ 10. Table 4 provides a compilation of the relevant sources for our determination of the LF.

As in our recent paper on the z ∼ 4–10 LFs, we use the results from these simulations to derive the selection volumes needed to relate the UV LF function ϕ(M) to the observed surface density of sources on the sky. Formally, we write the UV LF in stepwise format ϕ_j as ${\rm{\Sigma }}{\phi }_{j}W(M-{M}_{j})$, where j is an index running over the magnitude bins, where M_j corresponds to the absolute magnitude at the center of each bin, where

$$\begin{eqnarray}W(x)=0, & x\lt -0.4\\ 1, & -0.4\lt x\lt 0.4\\ 0, & x\gt 0.4,\end{eqnarray} \tag{ 1 }$$

and where x gives the position within a magnitude bin. We take the width of the magnitude bins to be 0.8 mag (e.g., versus the 0.5 mag used by Bouwens et al. 2015), given the limited number of bright z ∼ 9 and z ∼ 10 galaxies.

We then look for the derived LF ϕ_j that yields the observed surface density of z ∼ 9 and z ∼ 10 galaxies on the sky with maximum probability ${ \mathcal L }$. As in Bouwens et al. (2015), the likelihood ${ \mathcal L }$ is computed as

$$\begin{eqnarray}&&{ \mathcal L }={{\rm{\Pi }}}_{\mathrm{field}}{{\rm{\Pi }}}_{i}p({m}_{i}),\end{eqnarray} \tag{ 2 }$$

where the above products run over the different search fields and magnitude interval i used in the LF determinations, and where p(m_i) is the probability of identifying a certain number of sources in magnitude interval i in a given search field.

For simplicity (and given the small numbers in each of our samples: see the discussion in Section 4 of Bouwens et al. 2008), we ignore field-to-field variance in deriving the LF results and compute the likelihood that our survey fields show a certain number of sources assuming Poissonian statistics. We therefore compute p(m_i) as follows:

$$\begin{eqnarray}&&p({m}_{i})={e}^{-{N}_{\exp ,i}}\displaystyle \frac{{({N}_{\exp ,i})}^{{N}_{\mathrm{obs},i,j}}}{({N}_{\mathrm{obs},i,j})!},\end{eqnarray} \tag{ 3 }$$

where ${N}_{\mathrm{obs},i}$ is the observed number of sources in search field and magnitude interval i, and ${N}_{\exp ,i}$ is the expected number of sources in a search field and magnitude interval i. The expected number of sources in a search field ${N}_{\mathrm{expected},i}$ is computed as

$$\begin{eqnarray}&&{N}_{\mathrm{expected},i}={{\rm{\Sigma }}}_{j}{\phi }_{j}{V}_{i,j},\end{eqnarray} \tag{ 4 }$$

where ${V}_{i,j}$ is the effective volume over which one could expect to find a source of absolute magnitude j in the observed magnitude interval i.

The selection volumes ${V}_{i,j}$ are estimated using an almost identical procedure to that in Bouwens et al. (2015). Specifically, we constructed catalogs with mock sources spanning the entire redshift range z ∼ 7.5 to z ∼ 12. To ensure that sources had reasonable sizes and morphologies, we randomly selected similar luminosity z ∼ 4 galaxies from the Hubble Ultra Deep Field (Beckwith et al. 2006; Illingworth et al. 2013) to use as a template for modeling the two-dimensional pixel-by-pixel profiles of individual sources. The sizes of the model sources were assumed to scale with redshift as ${(1+z)}^{-1.2}$ to match the size scaling observed for sources with fixed luminosity from z ∼ 10 to z ∼ 2 (Oesch et al. 2010; Ono et al. 2013; Holwerda et al. 2015; Bouwens et al. 2015; Kawamata et al. 2015; Shibuya et al. 2015). The UV-continuum slopes of sources were assumed to have a mean value of −1.8, which is consistent with that measured at high luminosities at z ∼ 5–8 (Bouwens et al. 2012; Finkelstein et al. 2012; Willott et al. 2013; Bouwens 2014; Rogers et al. 2014), with a dispersion of 0.3 (Bouwens et al. 2012; Castellano et al. 2012).

We generate simulated images of each source in all HST, ground-based, and Spitzer/IRAC wavelength channels. Artificial images of individual sources in the ground-based and Spitzer/IRAC channels are produced by convolving the simulated HST images with the PSF-matching kernels we derive with mophongo (Labbé et al. 2013). These images are then added to sections of the CANDELS UDS, COSMOS, EGS, GOODS-North, and GOODS-South, and ERS fields, catalogs are constructed, and sources are selected using exactly the same procedures as we apply to the real observations. We include both the criteria used for our preselection and our confirmation criteria (i.e., $P(z\gt 8)\gt 0.9$) in computing the selection volume. We implement these criteria in a manner identical to how they are applied to the observations.

For example, to be included in our selection volume estimates, simulated sources are preselected using the criteria we describe in Section 3.2.1 or 3.3.1. For simulated sources within the CANDELS-UDS and CANDELS-COSMOS data sets, this means that their cumulative probability of lying at z > 8 must be greater than 50% before the addition of any Y₁₀₅-band data. In addition, simulated sources (over the CANDELS-UDS/COSMOS/EGS fields) must show a probability >90% of lying at z > 8 after the inclusion of the flux constraint (0 ± 12 nJy: nominally the flux constraint one would obtain for a z ∼ 9–10 galaxy based on a single orbit of Y₁₀₅-band observations) and must have a measured ${J}_{125}-{H}_{160}$ color >0.5 mag. Our simulation results make it clear how important the preselection can be. While increasing the efficiency of our search results significantly, preselection can also introduce a modest amount of incompleteness into the z ∼ 9–10 samples we identify from CANDELS, particularly at z < 9 (where it is ∼40% from the preselection step alone).

In addition to considering the selection of sources from CANDELS-UDS, CANDELS-COSMOS, and CANDELS-UDS fields, we also consider the selection of z ∼ 9 and z ∼ 10 galaxies from the CANDELS GOODS-North, CANDELS GOODS-South, and ERS fields.

In computing the number of confirmed z = 9–10 sources from our program, we assume all of the sources in Table 4 are bona fide z ∼ 9–10 galaxies and that there is no contamination in our selection. This would appear to be a good assumption, given that the typical z ∼ 9–10 candidate formally prefers a z > 8 solution at 99% likelihood. We do not include EGS910-2 and UDS910-1 in our LF calculation since they do not meet our formal criteria for inclusion (but nevertheless appear to be probable z ≳ 8.5 galaxies). We suppose that all of the candidates from our follow-up program that were not explicitly confirmed by that program lie at z < 8.4 (all but one of these candidates was detected at ≥2σ in the follow-up Y₁₀₅-band observations and are therefore unlikely z > 8.4). The z ∼ 9 candidate GS-z9-5 is modeled as a 0.5 z ∼ 9 galaxy (i.e., ∼50% probability of contamination) given that its computed P(z > 8) was only 0.66 (Figure 5). We ignore the impact of possible lensing amplification on one source in our selection (EGS910-3) given the size of the magnification factor (0.25 mag) and the fact that source volume and magnification factor trade off in such a way as to have little impact on the derived LF.

Our z ∼ 9 and z ∼ 10 LF results are presented in Figure 7 and tabulated in Table 5. In computing the uncertainties on the z ∼ 9 and z ∼ 10 UV LFs, we also include the expected large-scale structure uncertainties, using the results from the cosmic variance calculator of Trenti & Stiavelli (2008) and the observed comoving volume density. For context, the earlier LF results of McLure et al. (2013), Oesch et al. (2013, 2014), Bouwens et al. (2014a, 2015), and McLeod et al. (2015) are presented in Figure 7.

Zoom In Zoom Out Reset image size

Table 5.  Binned Determination of the Rest-frame UV LF at z ∼ 9 and z ∼ 10

${M}_{1600,{AB}}$ a	ϕk (10−3 Mpc−3 mag−1)
z ∼ 9 galaxies
−21.94	<0.0024b
−21.14	${0.0044}_{-0.0024}^{+0.0042}$
−20.34	${0.0322}_{-0.0138}^{+0.0217}$
z ∼ 10 galaxies
−22.05	<0.0017b
−21.25	${0.0009}_{-0.0007}^{+0.0021}$
−20.45	${0.0180}_{-0.0098}^{+0.0174}$

Notes.

^aDerived at a rest-frame wavelength of 1600 Å. ^b1σ upper limit.

Download table as:  ASCIITypeset image

Given the limited number of z ∼ 9–10 candidates in our samples and some arbitrariness in the choice of bin centers (and width) for our stepwise z ∼ 9–10 LF results, it is conceivable that our z ∼ 9–10 LF results could depend on how we bin our sample. We therefore also model the bright end of the LF as a power law (motivated by the results of, e.g., Bowler et al. 2014 and Bouwens et al. 2015). By marginalizing over both the normalization and power-law slope of the model LFs, we derive constraints (68% confidence levels) on the volume density of the z ∼ 9–10 galaxy candidate for a given M_UV. These results are presented in Figure 7 and shown in Table 6.

Table 6.  68% Confidence Regions on the Volume Density of Galaxies at z ∼ 9 and z ∼ 10 vs. M_UV

	Volume Density (10−3 Mpc−3 mag−1)
${M}_{1600,{AB}}$ a	Lower Boundb	Upper Boundb
z ∼ 9 galaxies
−21.84	0.0006	0.0024
−21.64	0.0010	0.0032
−21.44	0.0016	0.0044
−21.24	0.0024	0.0060
−21.04	0.0039	0.0084
−20.84	0.0059	0.0120
−20.64	0.0087	0.0174
−20.44	0.0126	0.0262
−20.24	0.0176	0.0402
−20.04	0.0250	0.0621
z ∼ 10 galaxies
−21.95	0.0004	0.0018
−21.75	0.0006	0.0022
−21.55	0.0009	0.0028
−21.35	0.0012	0.0036
−21.15	0.0017	0.0048
−20.95	0.0023	0.0066
−20.75	0.0030	0.0093
−20.55	0.0039	0.0135
−20.35	0.0049	0.0200
−20.15	0.0060	0.0299

Notes.

^aDerived at a rest-frame wavelength of 1600 Å. ^b68% Confidence Region.

Download table as:  ASCIITypeset image

To guide expectations, in Figure 7 we also include the LFs we derive by extrapolating the LF results from Bouwens et al. (2015) to z ∼ 9 and z ∼ 10.2 (the mean redshift of our z ∼ 9 and z ∼ 10 derived from our selection volume simulations). In deriving a fitting formula from the Bouwens et al. (2015) results, we only consider evolution over the range from z ∼ 8 to z ∼ 5 where the UV LF evolves in a relatively smooth manner. This results in the Schechter parameters depending on redshift in the following manner: ${M}_{\mathrm{UV}}^{* }=(-20.97\pm 0.10)\ +(0.17\pm 0.10)(z-6)$, ${\phi }^{* }=({0.45}_{-0.08}^{+0.10}){10}^{(-0.21\pm 0.09)(z-6)}{10}^{-3}\,{\mathrm{Mpc}}^{-3}$, $\alpha =(-1.91\pm 0.05)+(-0.13\pm 0.05)(z-6)$. This fitting formula implies M* = −20.45, ϕ* = 0.10 × 10⁻³ Mpc⁻³, and α = −2.3 at z ∼ 9 and M* = −20.28, ϕ* = 0.059 × 10⁻³ Mpc⁻³, and α = −2.46 at z ∼ 10.2. The best-fit evolutionary scenario from Bowler et al. (2014) is very similar to what we present above (where it was found that ${{dM}}^{* }/{dz}\sim 0.2$).

Relative to these extrapolations of the z = 4–8 results to z > 8, our present LFs are typically 1.5× lower in the mean, which is consistent with slightly faster evolution at z > 8. The results from the above fitting formula are also featured in Figure 8 where we combine the new LF results with previous results from the literature at z ∼ 4, z ∼ 5, z ∼ 6, z ∼ 7, and z ∼ 8.

Zoom In Zoom Out Reset image size

The present estimates of the volume density of particularly bright z ∼ 9 galaxies are the first available in the literature and have comparable uncertainties to the bright end of the present LF determinations at z ∼ 10 (and as derived earlier: Oesch et al. 2014; Bouwens et al. 2015).

7.2. Evolution of the UV Luminosity Density for Luminous Galaxies from z ∼ 10 to z ∼ 4

With the present constraints on the volume density of bright galaxies at z ∼ 9 and z ∼ 10, we can examine the evolution of the luminosity density of galaxies in the rest-frame UV with cosmic time.

Integrating the light in our binned representation of the bright end of the z ∼ 9 and z ∼ 10 LFs to −20 mag (approximately the faint-end limit of the present bright search), we derive a total UV luminosity density ρ_UV of ${10}^{{24.21}_{-0.19}^{+0.13}}$ erg s⁻¹ Hz⁻¹ Mpc⁻³ at z ∼ 9 and ${10}^{{23.94}_{-0.28}^{+0.18}}$ erg s⁻¹ Hz⁻¹ Mpc⁻³ at z ∼ 10. These inferred luminosity densities are ${5}_{-2}^{+3}\times$ and ${8}_{-3}^{+9}\times$ lower, respectively, than those found by Bouwens et al. (2015) at z ∼ 8.

In Figure 9, we compare the derived luminosity density ρ_UV at z ∼ 9 and z ∼ 10 with the UV luminosity density derived  by Bouwens et al. (2015) from z ∼ 4 to z ∼ 8 to the same limiting magnitude. Also shown in this figure is the best-fit evolution (68% confidence intervals) derived for the UV luminosity density based on the z ∼ 4–8 results. Our new luminosity density results ρ_UV at z ∼ 9 and z ∼ 10 are ${2.6}_{-0.9}^{+1.5}\times$ and ${2.2}_{-1.1}^{+2.0}\times$ lower than the extrapolated luminosity densities at z ∼ 9.0 and z ∼ 10.2.

Zoom In Zoom Out Reset image size

The present results are therefore consistent with a more rapid evolution at z > 8 than between z ∼ 8 and z ∼ 4, as first noted by Oesch et al. (2012), with a best-fit $d\,{\mathrm{log}}_{10}\,{\rho }_{\mathrm{UV}}/{dz}$ evolution of −0.45 ± 0.07 versus the −0.29 ± 0.02 evolution observed from z ∼ 8 to z ∼ 4. Interestingly, the evolution we derive here is completely consistent with the $d\,{\mathrm{log}}_{10}\,{\rho }_{\mathrm{UV}}/{dz}=-{0.54}_{-0.36}^{+0.19}$ scaling previously derived by Oesch et al. (2012). It is also consistent with the ${(1+z)}^{-10.9}$ evolutionary scalings considered in Oesch et al. (2014: see also Oesch et al. 2013).

Other authors (Ishigaki et al. 2015; McLeod et al. 2015; Oesch et al. 2015) have remarked that there is less evidence for faster evolution in the UV luminosity density at z > 8 integrated to lower luminosities, particularly when incorporating new results from the full Hubble Frontier Fields program (Lotz et al. 2014; Coe et al. 2015). If the HUDF/XDF is systematically underdense in z ∼ 9–10 galaxies or if the evolution of the LF proceeds more smoothly for lower-luminosity galaxies, then one might expect such a result. A much more definitive exploration of this issue will be possible if we  consider the much larger number of faint sources at z ∼ 9–10 expected from the full Frontier Fields program (Coe et al. 2015).

In this paper, we provide improved constraints on the volume density of especially luminous (${H}_{160,{AB}}\lt 27$) z ∼ 9 galaxies by making use of a search over all 5 CANDELS fields (750 arcmin² area in total). We also rederive the bright end of the $z\sim 10$ LF and extend it slightly fainter, taking advantage of the additional search power provided by the S-CANDELS data (Ashby et al. 2015) over the CANDELS-UDS, COSMOS, and EGS fields.

To obtain these constraints, we extend the earlier sample of 6 bright z ∼ 9 and z ∼ 10 galaxies identified by Oesch et al. (2015) to make use of the search area over all 5 CANDELS fields, including an additional ∼450 arcmin² area over the CANDELS-UDS, CANDELS-COSMOS, and CANDELS-EGS fields. We also expand the Oesch et al. (2014) search over the GOODS-South and GOODS-North to obtain a more complete selection of galaxies over the redshift range z ∼ 8.4–9.0 (see Section 5).

To identify a robust sample of bright z ∼ 9 and z ∼ 10 galaxies over this new area, we employed a two-part strategy for selecting sources. To begin, we selected those sources that showed the highest probability of corresponding to z ∼ 9 and z ∼ 10 galaxies based on the existing observations. Such identifications were possible given the depth of the HST and ground-based imaging observations which placed constraints on both the sharpness of a possible spectral break at 1.2 μm and the spectral slope blueward of the break.

The second step was to obtain deep observations on each of these candidates at 1.05 μm to test the nature of these candidates. These follow-up observations were obtained with the 11-orbit HST program z9-CANDELS (Bouwens 2014: GO 13792).

Using the new HST observations from the z9-CANDELS program, we find that 5 out of the 12 z ∼ 9–10 candidates that we followed up with our program appear to be bona fide bright z ∼ 9–10 galaxies (Figure 4).

Combining our new samples with previous samples of luminous z ∼ 9–10 candidates over CANDELS GOODS-North and GOODS-South (Oesch et al. 2014), and also including four probable z ∼ 8.4–9.0 galaxies we identified over the GOODS-S field (finding no additional bright sources over the GOODS-N field), we identify a total sample of 15 bright z ∼ 9–10 candidates over the five CANDELS fields (Table 4). This is ≳2× larger than the sample of luminous z ∼ 9–10 galaxies we identified earlier in Oesch et al. (2014).

We use these larger samples of z ∼ 9–10 galaxies to derive improved constraints on the bright end of the z ∼ 9 and z ∼ 10 LFs. We achieve these results by simulating both stages in the selection process (for the CANDELS-UDS, COSMOS, and EGS fields)—or a single stage in the case of the CANDELS GOODS-North, GOODS-South, or ERS fields—to arrive at statistically robust conclusions.

As one would expect with any follow-up program of limited size, the z9-CANDELS program is not able to observe all potential z ∼ 9 and z ∼ 10 galaxies over the CANDELS-UDS, COSMOS, and EGS fields¹¹ , and therefore likely suffers from moderate incompleteness, particularly for galaxies at z < 9, due to our exclusively preselecting follow-up sources with redder ${J}_{125}-{H}_{160}\gt 0.5$ colors than galaxies which exist in this range. Some incompleteness would also result from the detection of spurious flux in the optical bands (∼25% effect). Nevertheless, we remark that all such effects are included in our selection volume simulations and our focus on ${J}_{125}-{H}_{160}\gt 0.5$ galaxies means that our search is most complete at z > 9.

We would expect these results to be improved, particularly toward fainter magnitudes, if even deeper observations over the CANDELS-UDS, CANDELS-COSMOS, and CANDELS-EGS fields could be obtained. Deeper observations in the F105W band should increase the size of our bright (${H}_{160,{AB}}\lesssim 27$) z ∼ 9–10 samples over the CANDELS-WIDE fields by improving the completeness of our search results at z < 9.¹² Meanwhile, deeper observations in various redder bands (e.g., ${{JH}}_{140,{AB}}$) would allow us to extend these searches to even fainter magnitudes (i.e., to >26.5 mag). We would also expect gains from analysis of the 500-orbit, 500 arcmin² BoRG${}_{[z910]}$ program (Trenti 2014).

We thank Jim Dunlop and Steve Finkelstein for valuable conversations. We thank Steve Willner for providing us with feedback on an earlier draft of this manuscript. This work is based in part on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. We acknowledge the support of NASA grant NAG5-7697, NASA grant HST-GO-11563, and a NWO vrij competitie grant 600.065.140.11N211.

We identified 12 candidate z = 9–10 sources over the CANDELS-UDS, CANDELS-COSMOS, and CANDELS-EGS fields that we explicitly targeted for follow-up observations.

Postage stamp images of these 12 candidates are presented in Figure 10. As should be obvious from this figure, all six of the present z ∼ 9–10 candidates show clear detections in the H₁₆₀ band as well as significant ∼2–3σ detections in the J₁₂₅-band and JH₁₄₀-band observations (where available), as well as in the S-CANDELS Spitzer/IRAC data.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The observed photometry to those candidate sources are presented in Figure 11 along with SED fits to a model z > 8 galaxy and a model z < 3 galaxy. Also shown in this figure is the redshift likelihood distribution (solid black line) based on the photometry we have available for each candidate in the ∼20 different wavelength channels (HST + Spitzer/IRAC + ground-based observations). In addition, this figure presents the redshift likelihood distribution we would expect, assuming that these candidates are not detected in the single orbit of follow-up Y₁₀₅-band observations from the z9-CANDELS program.

All 12 candidates show ${H}_{160}-[3.6]$ colors (see also Wilkins et al. 2016), red [3.6]–[4.5] colors, and observed sizes (Holwerda et al. 2015; Shibuya et al. 2015) similar to the first samples of particularly luminous z ∼ 9–10 galaxies identified by Oesch et al. (2014).

A fraction of the candidate z ∼ 9–10 galaxies targeted by our follow-up observations from the z9-CANDELS program appear not to be at z > 8. In Figure 12, we present postage stamps for 9 such sources (7 explicitly targeted by our program and 2 lower-probability z ∼ 9–10 candidates which were incidentally targeted). In almost all of the candidates which are not confirmed by our follow-up observations, the sources show a ≥2σ detection in the Y₁₀₅ band. Figure 13 shows the best-fit SED models for the tentative z ∼ 9–10 galaxies that were not confirmed by observations from our follow-up program.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Detailed comments on specific candidate z ∼ 9–10 galaxy that were targeted for follow-up with our z9-CANDELS program can be found below.

• COS910-0: Follow-up of COS910-0 in the Y₁₀₅ band shows a 3σ detection at 1.05 μm. A detailed fit to its SED suggests that it is actually a star-forming galaxy at z = 7.8. Its inclusion in our original sample of z > 8.4 candidate galaxies occurred due to its measured ${J}_{125}-{H}_{160}$ color, which was likely redder than reality due to the impact of noise.
• COS910-2: COS910-2 is detected at 2.2σ in the Y₁₀₅-band follow-up observations. Such a detection is not expected if the galaxy is actually at z ≳ 9, and so the source must have a redshift of z ≲ 8. It may correspond to either a z ∼ 0.5 or a z ∼ 8 galaxy.
• COS910-3: Follow-up of COS910-2 in the Y₁₀₅ band and the JH₁₄₀ band (1/3 of an orbit) shows a 1σ detection at 1.05 μm and 2σ detection in the JH₁₄₀ band. Perhaps most importantly, the deeper photometry over the source in the JH₁₄₀ band confirms that the source has a ${{JH}}_{140}-{H}_{160}$ color of 1.25 mag. While this is consistent with the source having a redshift of z ≳ 11, one would not expect to detect the source at 2σ in the J₁₂₅ band in this case. These results suggest that the apparent break in the spectrum at 1.5 μm is not especially sharp and that this source shows faint but detectable flux to much bluer wavelengths.
• UDS910-0: Follow-up of UDS910-0 in the Y₁₀₅ band shows a 4σ detection at 1.05 μm. Therefore, this source cannot plausibly correspond to a z ≳ 8.4 galaxy. The best-fit redshift we compute for the source is z ∼ 7. Like COS910-0, its inclusion in our sample of z ∼ 9–10 galaxies likely occurred as a result of noise in the measured ${J}_{125}-{H}_{160}$ color.
• EGS910-1: Follow-up of EGS910-1 in the Y₁₀₅ band shows a 3σ detection at 1.05 μm, consistent with a redshift of z < 8.
• EGS910-4: Follow-up of EGS910-4 in the Y₁₀₅ band shows a 3σ detection at 1.05 μm, consistent with a redshift of z < 8.
• EGS910-5: Follow-up of EGS910-5 in the Y₁₀₅ band shows a 1.7σ detection at 1.05 μm, strongly suggesting that it is not at z > 8.
• EGS910-6: EGS910-6 is detected at 4.5σ in our Y₁₀₅-band follow-up observations at 1.05 μm, providing clear evidence that it is at z < 8. A z < 8 solution is preferred at 99.8% confidence (Figure 13).
• EGS910-7: Follow-up of EGS910-7 in the Y₁₀₅ band only shows a 1.1σ detection at 1.05 μm. However, the ${H}_{160}-[3.6]$ color is sufficiently red that it seems more consistent with a lower-redshift galaxy than a z > 8 galaxy. The nature of this source is unclear.

When putting together a compilation of the most promising z ∼ 9–10 candidates to target with follow-up HST observations, we experimented with a variety of different procedures to identify possible z ∼ 9–10 galaxy candidates. As a result, we identified a larger number of possible z = 9–10 galaxy candidates than we could thoroughly follow-up with the 11 orbits allocated to our HST program.

While the redshift likelihood distributions we derived for these candidates suggested that the vast majority of them likely corresponded to sources at z < 8, it is nevertheless possible that several of these candidates might have redshifts in excess of z ∼ 8.

We provide a short list of the "lower-probability" z ∼ 9–10 galaxy candidates we identified over the CANDELS COSMOS, UDS, and EGS fields in Table 7. These sources did not meet our criteria for preselection, but nevertheless may correspond to z ∼ 9–10 galaxies. Figure 14 shows postage stamp cut-outs of these sources, while Figure 15 illustrates fits to their spectral energy distributions and our derived redshift likelihood contours for these sources.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 7.  Possible z ∼ 9–10 Galaxies over the CANDELS UDS, COSMOS, and EGS Program that Did Not Satisfy Our Criteria for Preselection and Hence Were Not Targeted for Follow-up Observations

ID	R.A.	decl.	${H}_{160,{AB}}$	zphot	P(z > 8)a
COS910-4	10:00:15.52	02:17:01.5	25.6 ± 0.1	9.3	0.18
UDS910-2b	02:17:13.08	−05:15:55.4	26.6 ± 0.2	10.2	0.68
UDS910-3	02:17:52.38	−05:15:06.3	26.9 ± 0.2	9.4	0.28
UDS910-4	02:17:14.61	−05:15:15.7	26.6 ± 0.2	9.1	0.50

Notes.

^aRedshift likelihood is computed using the flux measurements for these sources in all photometric bands shown in Table 1. ^bWhile the source UDS910-2 nominally has >50% probability of lying at z > 8, it was not targeted with our follow-up program z9-CANDELS because it could not be confirmed to be >90% probability candidate with the addition of a single orbit of HST observations.

Download table as:  ASCIITypeset image