PHOTOMETRIC CONSTRAINTS ON THE REDSHIFT OF z ∼ 10 CANDIDATE UDFj-39546284 FROM DEEPER WFC3/IR+ACS+IRAC OBSERVATIONS OVER THE HUDF^*

We analyze the full set of HST observations over the HUDF09/XDF, including data from the 192-orbit HUDF09 program (Bouwens et al. 2011b), CANDELS, the 128-orbit HUDF12 program, and several other sizeable programs.

255 orbits of HST WFC3/IR observations are now available over the HUDF09/XDF, including ∼100, ∼40, 30, and ∼85 orbits in the Y₁₀₅, J₁₂₅, JH₁₄₀, and H₁₆₀ bands, respectively. The biggest gains over the HUDF09 program came in the Y₁₀₅ and JH₁₄₀ bands. We reduced these observations in a similar manner as the original WFC3/IR data from the HUDF09 program (Bouwens et al. 2011b). Special care was taken to keep our reductions of the new CANDELS and HUDF12 observations separate from those of the original HUDF09 data, to enable us to evaluate the reality of sources from the original observations.

In addition, we now have new reductions of the optical observations over the HUDF from the XDF project that are ∼0.1–0.2 mag deeper than the original Beckwith et al. (2006) HUDF reductions due to our inclusion of all other HST data sets taken over the HUDF for the past 10 years, including the recent optical/ACS I₈₁₄ data.

To obtain photometric constraints on UDFj-39546284 redward of the H₁₆₀-band, we utilize the deep 120 hr Spitzer/IRAC (Fazio et al. 2004) observations in the [3.6] and [4.5] channels from the original GOODS IRAC program and the 262 hr IRAC Ultradeep Field program (IUDF10; PI: Labbé). These observations reach to 27.1 mag and 26.8 mag in the 3.6 μm and 4.5 μm bands, respectively (3σ; Labbé et al. 2012). We also utilize the very deep K_s-band observations over the HUDF (26.5 mag: 5σ), including data from VLT/HAWK-I (program 186.A-0898; PI: A. Fontana), VLT/ISAAC (program 73.A-0764, PI: I. Labbé and 168.A-0485, PI: C. Cesarsky), and PANIC (PI: I. Labbé). In co-adding the K_s-band observations, individual frames are weighted by the inverse variance expected for a point source (Labbé et al. 2003). Table 1 provides detailed information on the various K_s-band observations used for our deep reduction.

In summary, in addition to the HUDF09/HUDF12 WFC3/IR and the IUDF10 IRAC data sets (also used by Ellis et al. 2013, though Ellis et al. 2013 appear not to have accounted for the modest variations in the effective depth of the IRAC observations as a result of varying contamination from neighboring foreground sources), we utilize the deeper XDF optical/ACS data set on the HUDF, deep optical/ACS I₈₁₄ and deep K_s-band observations for our detailed study of UDFj-39546284.

We obtain flux measurements on the HST observations by running SExtractor (Bertin & Arnouts 1996) in dual-image mode, taking the detection image to be the H₁₆₀-band and using the point-spread function (PSF)-matched images for photometry. Colors are measured in small-scalable apertures (Kron 1980 factor of 1.2) and corrected to total by comparing the H₁₆₀-band flux in a larger-scalable aperture (Kron factor of 2.5) to that in the smaller-scalable aperture and then applying a correction to account for light on the wings of the PSF based on the tabulated encircled-energy distribution.

The K_s-band flux measurement was performed in a 065-diameter circular aperture and corrected to match our HST photometry by comparing the H₁₆₀-band flux (065-diameter apertures, after PSF-correction) to that found in our baseline scalable apertures.

IRAC photometry was performed utilizing software to model the light profiles of neighboring sources so that this light could be subtracted (Labbé et al. 2006, 2010a, 2010b, 2012). Clean photometry of the source is then performed (18-diameter circular apertures). A factor of ∼2.2 correction is made to the measured fluxes to account for light on the wings of the IRAC PSF.

The photometry we derive for UDFj-39546284 is presented in Table 1 and Figure 2. UDFj-39546284 again shows a very significant detection in the H₁₆₀-band and no significant detection in any other band. The fact that the source is detected in the new H₁₆₀-band observations at 5.3σ (05-diameter aperture) and 7.8σ (05-diameter aperture) in the total H₁₆₀-band stack demonstrates that this is definitely a real galaxy (Figure 1).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The present color measurements are consistent with those from Ellis et al. (2013), but our total H₁₆₀-band magnitude is ∼0.6 mag brighter than the 05-diameter aperture-magnitude measurement from Ellis et al. (2013). This is not unsurprising given the significant spatial extension of UDFj-39546284, our use of larger scalable apertures (more appropriate for this source), and our correction to total magnitudes.

The optical and near-IR observations blueward of the H₁₆₀-band are very deep and indicate a sharp fall-off in the spectrum at <1.6 μm (Figure 2). The amplitude of the flux decrement from the H₁₆₀-band flux measurement is a substantial >2.8 mag to a co-added Y₁₀₅ J₁₂₅ JH₁₄₀ bandpass, >3.3 mag to a co-added B₄₃₅ V₆₀₆ i₇₇₅ I₈₁₄ z₈₅₀ Y₁₀₅ J₁₂₅ JH₁₄₀ bandpass, and >2.2 mag to the JH₁₄₀-band. (The flux constraints from multiple bands were combined assuming a flat-spectrum (Fν) source.) Redward of the H₁₆₀-band, the IRAC and K_s-band observations are less deep, but place strong constraints on the general shape of the spectral energy distribution (SED).

The existence of a significant ∼8σ detection of UDFj-39546284 in the H₁₆₀-band, a strong break in the spectrum blueward of the H₁₆₀-band, and no prominent detection of the source redward of the H₁₆₀-band is suggestive of a z > 10 galaxy. Use of the photometric redshift code ZEBRA (Feldmann et al. 2006) yields a formal redshift estimate of 11.8 ± 0.3 for UDFj-39546284 (Figure 3). We find a similar result with the EAZY photometric redshift code (Brammer et al. 2008). The present estimates are somewhat lower than the Ellis et al. (2013) z = 11.9 estimate, likely due to our additional constraint on the luminosity of UDFj-39546284 from the deep K_s-band data.

Zoom In Zoom Out Reset image size

As in Oesch et al. (2012a) and Ellis et al. (2013), attempts to fit the source with a lower-redshift galaxy SED are not particularly successful. The best low-redshift fit, without a substantial emission-line contribution, is an evolved galaxy at z_lowz = 2.6. However, the high measured χ² value of this fit (χ²_lowz = 34.6) compared to the best-fit solution at z = 11.8 (χ² = 13.3) makes this conventional low-redshift solution untenable (but see below).

The new photometry also allows us to set useful constraints on the shape of the spectrum redward of the break. We fit the SED with a power law f_λ∝ f₀ λ^β, leaving the redshift, luminosity, and β as free parameters. We find 1σ and 2σ upper limits of –2.3 and –1.5, respectively, for β. These upper limits correspond to the maximum β's where Δχ² = χ² − χ²_min = 1 and 4, respectively. This demonstrates quite definitively that UDFj-39546284 is blue redward of the H₁₆₀-band and cannot be well fit by an old or dusty low-redshift SED.

UDFj-39546284 consists of a compact core, embedded in a more extended structure. The features to the upper left and lower right of the source (Figure 1) appear to extend some ∼04 in radius from the source (see also Ono et al. 2013). Using SExtractor, we measure a half-light radius of ∼017 for UDFj-39546284 in the deeper observations. This is larger than the ∼013-diameter half-light radius for the PSF. Correcting for the PSF, the half-light radius for this candidate is 013.

If we interpret this as a z ∼ 12 source, the implied ∼0.5 kpc (physical) half-light radius for the source is consistent with what one would expect applying a (1 + z)⁻¹ size scaling to the z ∼ 7–8 galaxy samples studied by Oesch et al. (2010) where the mean size for comparable-luminosity sources is 0.8 kpc (physical). A ∼(1 + z)⁻¹ size scaling has been found to describe the evolution of star-forming galaxies over a wide range in redshift (e.g., Buitrago et al. 2008; Oesch et al. 2010). This source is also potentially consistent with expectations if we interpret this as a z ∼ 2 galaxy. The measured half-light radius translates into a physical size of ∼1.1 kpc, at the top end of the range expected for extreme emission-line or star-forming galaxies at z ∼ 2 (van der Wel et al. 2012), after correcting for typical r∝ L^0.3 luminosity dependencies (e.g., de Jong & Lacey 2000).

While simply identifying UDFj-39546284 as a z ∼ 12 galaxy would seem appropriate (see also Ellis et al. 2013), it becomes problematic when both the apparent UV luminosity of this source and the total search volume in which the source was found are taken into consideration. If the source is at z ∼ 12, its intrinsic luminosity would be –20.3 (∼0.5× the luminosity of a z = 3 L* galaxy: Steidel et al. 1999). This is ∼4× higher than what Bouwens et al. (2011a) and Oesch et al. (2012a) inferred if the source was at z ∼ 10.4 (which seemed plausible before the JH₁₄₀ constraint was available). The much higher luminosity follows from the greater luminosity distance for UDFj-39546284 (factor of ∼1.4 change) and the fact that the source is only seen in the reddest one-third of the H₁₆₀-band (factor of 3 change).

To put this unusually high luminosity in context, we calculate the approximate volume in which we could have found the source. Utilizing the same techniques as in Oesch et al. (2013), we estimate a total search volume of ∼3 × 10³ Mpc³ (comoving) for UDFj-39546284 in the combined HUDF09/HUDF12/XDF data set.

The calculated selection volume and observed UV luminosity allow us to derive an approximate LF for star-forming galaxies at z = 12, assuming of course that UDFj-39546284 is indeed at z = 12. The result is shown in Figure 4 and is unique to this analysis. For context, the LFs inferred for star-forming galaxies at z ∼ 4–10 from Bouwens et al. (2007), Oesch et al. (2012b, 2013) are also presented. It is clear that UDFj-39546284 would be ∼10× more luminous than similarly-prevalent galaxies at z ∼ 10 and ∼20× more luminous than similarly-prevalent star-forming galaxies at z ∼ 12, extrapolating lower-redshift LF trends to z > 10.

Zoom In Zoom Out Reset image size

The evolution of the UV LF at early times, i.e., from z ∼ 10 to z ∼ 4, is sufficiently uniform that the discovery of a z ∼ 12 galaxy that is ∼20× more luminous than expected over such a small area is implausible and strongly argues for another explanation¹⁰.

The properties of UDFj-39546284 are puzzling and difficult to explain as either a low or high-redshift source if the bulk of the H₁₆₀-band flux is from continuum star light.

However, we can avoid this difficulty if the H₁₆₀-band light predominantly arises from line emission (see also Ellis et al. 2013). For example, in the lower-middle panel of Figure 3, we show one possible, though somewhat extreme, example, where we allow for arbitrary-EW line emission from an active galactic nucleus (AGN). The dust extinction is high (E(B − V) = 0.9), and the rest-frame EW of H_α is large (∼5000 Å). This fit has a χ²(=17.1) more similar to the z ∼ 12 solution, but for a redshift z ∼ 1.5. This is ad-hoc, but demonstrates what is needed.

Support for line emission contributing substantially to the flux in the H₁₆₀-band comes from the recent analysis of the deep WFC3/IR grism observations of UDFj-39546284 in a companion paper by Brammer et al. (2013). They find evidence of an emission-line feature at 1.6 μm that could provide most or all of the observed H₁₆₀-band flux for UDFj-39546284. The existence of such a feature is not unexpected given the difficulty in modeling the source as a pure-continuum galaxy at z ∼ 12 (because of its luminosity) or z ∼ 2–3 (χ² = 34.6; Figure 3).

Given the plausibility of line emission playing a role, the question arises as to the nature of the line emission. Brammer et al. (2013) argue that the emission-line contribution would be from an extreme emission-line galaxy (EELG), notably the [O iii]λ4959 + 5007 doublet at z ∼ 2.2. Such galaxies have been found in wide-area grism and imaging surveys with WFC3/IR (van der Wel et al. 2011; Atek et al. 2011; Brammer et al. 2012). Even more extreme examples are needed in the case of UDFj-39546284. Ellis et al. (2013) similarly suggested the source might be an EELG at z ∼ 2.4, but could not explain how such a source could produce the observed spectral break. Brammer et al. (2013) describe the discovery of an EELG at z ∼ 1.6 with an extremely-high [O iii] EW and relatively-red UV colors that would come close to satisfying the constraints if that source were at z ∼ 2.2.

Alternatively, the emission-line flux could be from Lyα. EWs of ∼200 Å are seen in star-forming galaxies in the z ∼ 4–6 universe (e.g., Stark et al. 2010) and would cause the source to be brighter by a factor of ∼4, resulting in a much more plausible intrinsic luminosity, i.e., ∼–18.8 mag. However, even with this luminosity, the source would still be at least >5× more luminous than one would expect for a comparably-prevalent z ∼ 12 galaxy. Attributing the emission to Lyα also seems implausible given the large amounts of neutral hydrogen expected in the z > 7 universe that would resonantly scatter Lyα photons. Lyα emission is found to be rare in z ≳ 7–8 galaxies, presumably due to an increasingly neutral intergalactic medium (Ono et al. 2012; Schenker et al. 2012; Pentericci et al. 2011; Caruana et al. 2012).

The existence of line emission is a plausible, though not proven, solution to the mystery regarding UDFj-39546284, with the evidence weighted toward a low-redshift solution with an extremely strong [O iii] feature.