A REMARKABLY LUMINOUS GALAXY AT Z = 11.1 MEASURED WITH HUBBLE SPACE TELESCOPE GRISM SPECTROSCOPY

Galaxy samples at z ∼ 9–10 are now being assembled based on deep HST imaging data (Zheng et al. 2012; Coe et al. 2013; Ellis et al. 2013; Oesch et al. 2013; Bouwens et al. 2015a; McLeod et al. 2015). Such distant sources can be identified based on a continuum break in the J₁₂₅ band, i.e., at around 1.2 μm. In a recent analysis of the WFC3/IR imaging data from the public CANDELS survey (Grogin et al. 2011; Koekemoer et al. 2011), we identified six surprisingly bright sources (H₁₆₀ = 26.0–26.8 mag) with photometric redshifts z = 9.2–10.2 in the two GOODS fields (Oesch et al. 2014).

Interestingly, a large fraction of these luminous z > 9 galaxy candidates were detected individually in the rest-frame optical with Spitzer/IRAC (Fazio et al. 2004) data from the S-CANDELS survey (Ashby et al. 2015). This provides a sampling of the spectral energy distribution (SED) of these sources from the rest-frame UV to the rest-frame optical. The strong breaks measured in their ${J}_{125}-{H}_{160}$ colors, the complete non-detections in the optical images, and the blue colors from HST to IRAC significantly limit the contamination by known low-redshift SEDs and together point to a true high-redshift nature for all these sources. Stellar contamination could be ruled out based on colors and sizes. Furthermore, no evidence for contamination by an active galactic nucleus was found based on variability over a 1 year timescale and X-ray upper limits (Oesch et al. 2014). Nevertheless, spectroscopic confirmation is clearly required to verify the high-redshift solutions for these targets.

The brightness of these z ∼ 9–10 galaxy candidates puts them within reach of WFC3/IR grism continuum spectroscopy, opening up the possibility of obtaining a grism redshift based on a continuum break. The brightest of these candidates, GN-z11, also had the highest photometric redshift and was thus identified as the target for follow-up spectroscopy for our HST program GO-13871 (PI:Oesch).

GN-z11 lies in the CANDELS DEEP area in GOODS-North at (R.A., decl.) = (12:36:25.46, +62:14:31.4) and has ${H}_{160}=26.0\pm 0.1$. It was previously introduced as GN-z10-1 (Oesch et al. 2014) or as source 20253 in the 3D-HST photometric catalog (Skelton et al. 2014). GN-z11 also has the highest signal-to-noise ratio (S/N) in both IRAC bands (3.6 and 4.5 μm) among the sample of z ∼ 9–10 galaxies from Oesch et al. (2014). The H₁₆₀ profile of GN-z11 shows clear asymmetry with its isophotal area extending over nearly 06. A stellar source can therefore be excluded.

The depth of the data used to originally identify this candidate was 27.8 mag in H₁₆₀, 27.0 mag in IRAC channel 1, and 26.7 mag in channel 2 (all 5σ; see Oesch et al. 2014). The photometry of the target source is listed in Table 1. Compared to our discovery paper (Oesch et al. 2014), this includes new measurements in the two IRAC 3.6 and 4.5 μm filters as well as in the WFC3/IR ${{JH}}_{140}$ band: deeper data are now available in JH₁₄₀ as part of our grism pre-imaging (see next section), and the IRAC photometry was updated based on a new, independent reduction of all available IRAC imaging data in the Spitzer archive by our team (similar to Labbe et al. 2015).

Table 1.  Photometry of GN-z11

Filter	Flux Density [nJy]
B435	7 ± 9
V606	2 ± 7
i775	5 ± 10
I814	3 ± 7
z850	17 ± 11
Y105	−7 ± 9
J125	11 ± 8
JH140	64 ± 13a
H160	152 ± 10
K	137 ± 67
$\mathrm{IRAC}\quad 3.6\;\mu {\rm{m}}$	139 ± 21a
$\mathrm{IRAC}\quad 4.5\;\mu {\rm{m}}$	144 ± 27a

Note.

^aNote that the earlier JH₁₄₀ data in which GN-z11 was first detected was less deep and gave a flux measurement of 104 ± 47 nJy. For other previous photometry measurements see Table 3 in (Oesch et al. 2014).

Download table as:  ASCIITypeset image

The primary data analyzed in this paper are 12 orbit deep G141 grism spectra from our HST program GO-13871 (PI: Oesch). These spectra were taken at two different orients in two epochs of six orbits each on 2015 February 11 and April 3 (see Figure 1). The data acquisition and observation planning followed the successful 3D-HST grism program (Brammer et al. 2012; Momcheva et al. 2015). Together with each G141 grism exposure, a short 200 s pre-image with the JH₁₄₀ filter was taken to determine the zeroth order of the grism spectra. The JH₁₄₀ images were placed at the beginning and end of each orbit in order to minimize the impact of variable sky background on the grism exposure due to the bright Earth limb and He 1.083 μm line emission from the upper atmosphere (Brammer et al. 2014). A four-point dither pattern was used to improve the sampling of the point-spread function and to overcome cosmetic defects of the detector.

Zoom In Zoom Out Reset image size

The grism data were reduced using the grism reduction pipeline developed by the 3D-HST team. The main reduction steps are explained in detail by Momcheva et al. (2015). In particular, the flat-fielded and global background-subtracted grism images are interlaced to produce 2D spectra for sources at a spectral sampling of ∼23 Å, i.e., about one-quarter of the native resolution of the G141 grism. The final 2D spectrum is a weighted stack of the data from individual visits. In particular, we down-weight pixels which are affected by neighbor-contamination using

$$\begin{eqnarray}&&w={[{(2\ast {f}_{\mathrm{contam}})}^{2}+{\sigma }^{2}]}^{-1}\end{eqnarray} \tag{ 1 }$$

where f_contam is the contamination model flux in a particular pixel and σ is the per-pixel rms taken from the WFC3/IR noise model (see Section 3.4.3 from Rajan et al. 2011). We have tested that σ is accurately characterized as demonstrated by the pixel flux distribution function in the 2D residual spectra (see Figure 11).

A local background is subtracted from the 2D frame which is a simple second order polynomial estimated from the contamination-cleaned pixels above and below the trace of the target source. 1D spectra are then computed using optimal extraction on the final 2D frames, weighting the flux by the morphology of the source as measured in the WFC3/IR imaging (Horne 1986).

The GOODS-N field, which contains our candidate GN-z11, has previously been covered by 2 times 2-orbit deep grism data from "A Grism H-Alpha SpecTroscopic survey" (AGHAST; GO-11600; PI:Wiener). However, the spectrum of our target GN-z11 is significantly contaminated in the AGHAST observations by several nearby galaxies given their orientation (see Figure 2). Furthermore, the observations were severely affected by variable sky background (Brammer et al. 2014). The final data used in this paper therefore do not include the AGHAST spectra, which were not optimized for GN-z11 that was discovered 3–4 years after they were taken. We confirmed that including the AGHAST data would not affect our results, however, because of the down-weighting of heavily contaminated pixels in our stacking procedure.

Zoom In Zoom Out Reset image size

One major challenge in slitless grism spectroscopy is the systematic contamination of the target spectrum by light from nearby galaxies. Our spectra were therefore taken at two orientations with the grism dispersion offset by 32° and tailored to show the least possible contamination while being schedulable (see Figure 2). The optimal orientations were determined through extensive simulations of the location of all orders from −1st to +3rd of all sources in the field detected in the extensive ancillary imaging data sets. Nevertheless, some contaminating flux can never be avoided and needs to be modeled. Following the same techniques as developed for the 3D-HST survey, this was done by fitting all the neighboring galaxies' SEDs (i.e., the EAZY template fits to the Skelton et al. photometric catalog) and using their morphologies to create 2D spectral models for all of them, which were then subtracted from the 2D spectrum of our target GN-z11 spectrum.

Having access to deep grism spectra at multiple independent orients also allows us to further refine the 2D spectral model of neighboring galaxies by including actually measured emission line fluxes. We thus extracted spectra of all neighboring sources at both orientations and fit their spectra as outlined by Momcheva et al. (2015). The resulting best fits were then used directly as contamination models. This produces significantly cleaner 2D spectra compared to assuming simple continuum emission templates for contamination modeling. The resulting contamination of all neighboring sources in our data as well as in the AGHAST spectra are shown in Figure 2. The quality of this quantitative contamination model is also discussed in the next sections and shown in Figure 3.

Zoom In Zoom Out Reset image size

Based on our previous photometric redshift measurement for GN-z11 (${z}_{\mathrm{phot}}=10.2$), we expected to detect a continuum break at 1.36 ± 0.05 μm. This can clearly be ruled out. However, the grism data are consistent with an even higher redshift solution. Interpreting the observed break as the 1216 Å break, which is expected for high-redshift galaxies based on absorption from the neutral inter-galactic hydrogen along the line of sight, we obtain a very good fit to the spectrum with a reduced χ² = 1.2 (see Figures 4 and 5). The best-fit redshift is ${z}_{\mathrm{grism}}={11.09}_{-0.12}^{+0.08}$, corresponding to a cosmic time of only ∼400 Myr after the Big Bang. The grism redshift and its uncertainty are derived from an MCMC fit to the 2D spectrum which also includes the morphological information of the source as well as the photometry, adopting identical techniques as used for the 3D-HST survey redshifts (see Momcheva et al. 2015).

Zoom In Zoom Out Reset image size

This high-redshift solution also reproduces the expected count rate based on an H = 26 mag continuum source, as well as the overall morphology of the 2D grism spectrum. This is demonstrated in Figure 3, where we show the original data, the neighbor subtracted 2D spectrum as well as the residual after subtracting out the z = 11.09 model with the correct H-band magnitude. Note the drop of the flux longward of 1.65 μm due to the reduced sensitivity of the grism. This is an important constraint, because it shows that the detected flux originates from the source itself and is not due to residual neighbor contamination.

Figure 11 further shows that the pixel distribution of the residual 2D frame is in excellent agreement with the expectations from pure Gaussian noise. This demonstrates that our contamination subtraction and flux uncertainty estimates were derived appropriately and that the resulting values are accurate.

Despite the difference from the previous photometric redshift estimate, the measured grism redshift is consistent with the photometry of this source (see Figure 6). While our previous photometric redshift estimate was ${z}_{\mathrm{phot}}=10.2\pm 0.4$, the redshift likelihood function contained a significant tail to z > 11. The updated and deeper JH₁₄₀ photometry subsequently resulted in a shift of the peak by Δ_z = 0.2 to a higher redshift. The z = 11.09 solution is within 1σ of the now better measured ${{JH}}_{140}-{H}_{160}$ color, which is the main driver for the photometric redshift estimate, as shown in Figure 6. The grism data significantly tighten the redshift likelihood function (bottom panel Figure 6) in addition to excluding lower redshift solutions.

Zoom In Zoom Out Reset image size

In the next two sections we also show that we can safely exclude all plausible lower redshift solutions. The new grism redshift confirms that this source lies well beyond the peak epoch of cosmic reionization (${z}_{\mathrm{reion}}=8.8;\;$ Planck Collaboration et al. 2015) and makes it the most distant known galaxy. This includes sources with photometric redshift measurements, apart from a highly debated source in the HUDF/XDF field, which likely lies at z ∼ 2 but has a potential z ∼ 12 solution (see Bouwens et al. 2011, 2013; Brammer et al. 2013; Ellis et al. 2013).

The principal goal of our grism program was to unequivocally exclude a lower redshift solution for the source GN-z11. While GN-z11 shows a very strong continuum break with ${J}_{125}-{H}_{160}\gt 2.4$ (2σ), without a spectrum, we could not exclude contamination by a source with very extreme emission lines with line ratios reproducing a seemingly flat continuum longward of 1.4 μm (Oesch et al. 2014).

The previous AGHAST spectra already provided some evidence against strong emission line contamination (Oesch et al. 2014), and we also obtained Keck/MOSFIRE spectroscopy to further strengthen this conclusion (see the appendix). However, the additional 12 orbits of G141 grism data now conclusively rule out that GN-z11 is such a lower redshift source. Assuming that all the H-band flux came from one emission line, we would have detected this line at >10σ. Even when assuming a more realistic case where the emission line flux is distributed over a combination of lines (e.g., Hβ + [O iii]), we can confidently invalidate such a solution. The lower left panel in Figure 4 compares the measured grism spectrum with that expected for the best-fit lower redshift solution we had previously identified (Oesch et al. 2014). A strong line emitter SED is clearly inconsistent with the data. Apart from the emission lines, which we do not detect, this model also predicts weak continuum flux across the whole wavelength range. At <1.47 μm, this is higher than the observed mean, while at >1.47 μm the expected flux is too low compared to the observations. Overall the likelihood of a z ∼ 2 extreme emission line SED based on our grism data is less than 10⁻⁶ and can be ruled out.

Note that in very similar grism observations for a source triply imaged by a CLASH foreground cluster, emission line contamination could also be excluded (Pirzkal et al. 2015). We thus have no indication currently that any of the recent z ∼ 9–11 galaxy candidates identified with HST is a lower redshift strong emission line contaminant (but see, e.g., Brammer et al. 2013, for a possible z ∼ 12 candidate).

Another potential source of contamination for very high redshift galaxy samples are dusty z ∼ 2–3 sources with strong 4000 Å or Balmer breaks (Hayes et al. 2012; Oesch et al. 2012). However, the fact that the IRAC data for GN-z11 show that it has a very blue continuum longward of 1.6 μm, together with the very red color in the WFC3/IR photometry, already rules out such a solution (see SED plot in Figure 4). Nevertheless, we additionally explore what constraints the grism spectrum alone can set on such a solution.

The expected flux for such a red galaxy increases gradually across the wavelength range covered by the G141 grism, unlike what we observe in the data (lower right in Figure 4). Compared to our best-fit solution (see next section) we measure a Δχ² = 15 when comparing the data with the expected grism flux. Apart from the extremely large discrepancy with the IRAC photometry, we can thus exclude this solution at 98.9% confidence based on the spectrum alone.

Similar conclusions can be drawn from the break strength alone (see e.g., Spinrad et al. 1998). Assuming that the observed break at 1.47 μm corresponds to 4000 Å at z = 2.7, a galaxy with a maximally old SED (single burst at z = 15) would show a flux ratio of $(1-{f}_{\nu }^{\mathrm{short}}/{f}_{\nu }^{\mathrm{long}})\lt 0.63$ when averaged over 560 Å bins. This is based on simple Bruzual & Charlot (2003) models without any dust. As mentioned earlier, the observed spectrum has a break of $(1-{f}_{\nu }^{\mathrm{short}}/{f}_{\nu }^{\mathrm{long}})\gt 0.68$ at 2σ, thus indicating again that we can marginally rule out a 4000 Å break based on the spectrum alone without even including the photometric constraints.

Despite being the most distant known galaxy, GN-z11 is relatively bright and reliably detected in both IRAC 3.6 and 4.5 μm bands from the S-CANDELS survey (Ashby et al. 2015). This provides a sampling of its rest-frame UV SED and even partially covers the rest-frame optical wavelengths in the IRAC 4.5 μm band (see Figure 6).

The photometry of GN-z11 is consistent with a SED of $\mathrm{log}\;M/{M}_{\odot }\sim 9$ using standard templates (Bruzual & Charlot 2003, see appendix). The UV continuum is relatively blue with a UV spectral slope β = −2.5 ± 0.2 as derived from a powerlaw fit to the H₁₆₀, K, and [3.6] fluxes only, indicating very little dust extinction (see also Wilkins et al. 2016). Together with the absence of a strong Balmer break, this is consistent with a young stellar age of this galaxy. The best fit age is only 40 Myr (<110 Myr at 1σ). GN-z11 thus formed its stars relatively rapidly. The inferred star formation rate is 24 ± 10 M_⊙ yr⁻¹. All the inferred physical parameters for GN-z11 are summarized in Table 2. Overall, our results show that galaxy build-up was well underway at ∼400 Myr after the Big Bang.

Table 2.  Summary of Measurements for GN-z11

R.A.	12:36:25.46
$\mathrm{Decl}.$	+62:14:31.4
$\mathrm{Redshift}\quad {z}_{\mathrm{grism}}$	${11.09}_{-0.12}^{+0.08}$ a
$\mathrm{UV}\quad \mathrm{Luminosity}\quad {M}_{{UV}}$	−22.1 ± 0.2
Half–Light Radiusb	$0.6\pm 0.3\;\mathrm{kpc}$
$\mathrm{log}\;{M}_{\mathrm{gal}}/{M}_{\odot }$ c	9.0 ± 0.4
$\mathrm{log}\;\mathrm{age}\;{\mathrm{yr}}^{-1}$ c	7.6 ± 0.4
$\mathrm{SFR}$	$24\pm 10\quad {M}_{\odot }\quad {\mathrm{yr}}^{-1}$
${A}_{\mathrm{UV}}$	$\lt 0.2\quad \mathrm{mag}$
$\mathrm{UV}\quad \mathrm{slope}\quad \beta \quad \quad ({f}_{\lambda }\propto {\lambda }^{\beta })$	−2.5 ± 0.2d

Notes.

^aAge of the universe at $z=11.09$ using our cosmology: 402 Myr. ^bFrom Holwerda et al. (2015). ^cUncertainties are likely underestimated, since our photometry only partially covers the rest-frame optical for GN-z11. ^dSee also Wilkins et al. (2016).

Download table as:  ASCIITypeset image

The spectrum of GN-z11 indicates that its continuum break lieswithin the H₁₆₀ filter (which covers ∼1.4–1.7 μm; see Figure 6). The rest-frame UV continuum flux of this galaxy is therefore ∼0.4 mag brighter than inferred from the H₁₆₀ magnitude. The estimated absolute magnitude is M_UV = −22.1 ± 0.2, which is roughly a magnitude brighter (i.e., a factor 3×) than the characteristic luminosity of the UV luminosity function at z ∼ 7–8 (Finkelstein et al. 2015; Bouwens et al. 2015b). With ${z}_{\mathrm{grism}}=11.09$, the galaxy GN-z11 is thus surprisingly bright and distant (see Figure 5). While one single detection of a galaxy this bright is not very constraining given the large Poissonian uncertainties, it is interesting to estimate how many such galaxies we could have expected based on (1) the currently best estimates of the UV LF at z > 8 and (2) based on theoretical models and simulations.

Our target was found in a search of the GOODS fields, which amount to ∼160 arcmin². However, in a subsequent search of the three remaining CANDELS fields no similar sources were found with likely redshifts at z ≳ 10 (Bouwens et al. 2015a). We therefore use the full 750 arcmin² of the CANDELS fields with matching WFC3/IR and Advanced Camera for Surveys (ACS) imaging for a volume estimate, which amounts to 1.2 × 10⁶ Mpc³ (assuming Δz = 1).

Using the simple trends in the Schechter parameters of the UV LFs measured UV at lower redshift (z ∼ 4–8) and extrapolating these to z = 11, we can get an empirical estimate of the number density of very bright galaxies at z ∼ 11. This amounts to 0.06 (Bouwens et al. 2015b) or 0.002 (Finkelstein et al. 2015) expected galaxies brighter than ${M}_{\mathrm{UV}}=-22.1$ in our survey corresponding to less than 0.3 per surveyed square degree. Similarly, recent empirical models (Mashian et al. 2015; Mason et al. 2015; Trac et al. 2015) predict only 0.002−0.03 galaxies as bright as GN-z11 in our survey or 0.01−0.2 per deg². All the assumed LF parameters together with the resulting estimates of the number of expected bright galaxies N_exp are listed in Table 3.

Table 3.  Assumed LFs for z ∼ 10–11 Number Density Estimates

Reference	$\phi \ast /{10}^{-5}$	$M\ast$	α	Nexp
	(Mpc−3)	(mag)		(<−22.1)
Bouwens et al. (2015b)	1.65	−20.97	−2.38	0.06
Finkelstein et al. (2015)	0.96	−20.55	−2.90	0.002
Mashian et al. (2015)	0.25	−21.20	−2.20	0.03
Mason et al. (2015)	0.30	−21.05	−2.61	0.01
Trac et al. (2015)	5.00	−20.18	−2.22	0.002

Note. The parameters $\phi \ast$, $M\ast$, and α represent the three parameters of the Schechter UV LF taken from the different papers.

Download table as:  ASCIITypeset image

The above estimates illustrate that our discovery of the unexpectedly luminous galaxy GN-z11 may challenge our current understanding of galaxy build-up at z > 8. A possible solution is that the UV LF does not follow a Schechter function form at the very bright end as has been suggested by some authors at z ∼ 7 (Bowler et al. 2014), motivated by inefficient feedback in the very early universe. However, current evidence for this is still weak (see discussion in Bouwens et al. 2015b). Larger area studies will be required in the future (such as the planned WFIRST High Latitude Survey; Spergel et al. 2015) surveying several square degrees to determine the bright end of the UV LF to resolve this puzzle.

In the main part of our analysis, we use a weighted sum to combine the 2D grism data from the six different visits to the final 12-orbit data. In particular, our weight includes a term to down-weight pixels which are affected by contamination (see Equation (1)). To ensure the detected signal is not the result of our stacking procedure, we additionally computed a simple median-stacked spectrum. This is shown in Figure 8. The median stack clearly still shows the continuum break, albeit it is somewhat noisier overall, since it does not include any contamination-based weighting. Nevertheless, the median flux is consistent with the expected count rate for an H₁₆₀ = 26 mag source at ${z}_{\mathrm{grism}}=11.09$.

Zoom In Zoom Out Reset image size

The 1D spectrum in the main body of the paper is based on an optimal extraction taking into account the asymmetric morphology of GN-z11. Figure 9 shows different 1D extractions as a function of position relative to the peak of the trace of GN-z11. These are simple sums extending over 018 (i.e., 3 pixels) in the spatial direction. These 1D spectra show consistency with the expected count rate from our H-band morphological model of GN-z11. The figure also shows that the negative dip at ∼1.6 μm stems from negative pixels slightly above the peak trace of GN-z11. We ensured that these negative pixels are not the result of any cosmic rays or inaccurate persistence or contamination subtraction. The dip extending over 3 rebinned pixels is consistent with simple Gaussian fluctuations based on our noise model.

Zoom In Zoom Out Reset image size

We also tested whether a break is seen when further splitting up the data into our two independent epochs. The S/N of the continuum detection in the final 12-orbit stack shown in the main body of the paper is already relatively low, but this split-data test is a good cross-check on the viability and consistency of the result. These data of the individual epoch each consist of 6 orbits but at two different orientations, resulting in different contamination levels as a function of wavelength. Figure 10 shows both the 2D and the corresponding 1D spectra for both epochs. When rebinned to 560 Å a continuum break is seen in both epochs separately, consistent with our best-fit redshift ${z}_{\mathrm{grism}}=11.09$. However, the S/N at 93 Å resolution in these spectra is small and the detailed differences should not be over-interpreted. While some pixels off the trace in Figure 10 show residual flux in the epoch-split data, these pixels are heavily contaminated by neighboring galaxies and are downweighted in the final stack. In Figure 11 we show the final stacked pixel flux distribution within 06 of the trace, which is perfectly consistent with a Gaussian. This demonstrates that our neighbor subtraction model and our pixel rms estimates are both accurate.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size