EVIDENCE FOR UBIQUITOUS HIGH-EQUIVALENT-WIDTH NEBULAR EMISSION IN z ∼ 7 GALAXIES: TOWARD A CLEAN MEASUREMENT OF THE SPECIFIC STAR-FORMATION RATE USING A SAMPLE OF BRIGHT, MAGNIFIED GALAXIES

In selecting our small sample of bright, magnified z ∼ 7 galaxies, we make use of the deep HST observations available over the first 23 clusters in the CLASH multicycle treasury program (GO 12101: PI Postman), A1689 and A1703 (GO 11802; PI: Ford), the Bullet cluster (GO 11099; PI: Bradac), and 9 clusters from the Kneib et al. (GO 11591) program. The CLASH cluster fields are each covered with 20 orbit HST observations spread over 16 bands using the Advanced Camera for Surveys (ACS; B₄₃₅, g₄₇₅, V₆₀₆, r₆₂₅, i₇₇₅, I₈₁₄, and z₈₅₀), Wide Field Camera WFC3/UVIS (UV₂₂₅, UV₂₇₅, U₃₃₆ and U₃₉₀) and WFC3/IR instrument (Y₁₀₅, J₁₁₀, J₁₂₅, JH₁₄₀ and H₁₆₀). A1703 was covered with 22 orbits of ACS and WFC3/IR (B₄₃₅, g₄₇₅, V₆₀₆, r₆₂₅, i₇₇₅, z₈₅₀, J₁₂₅, H₁₆₀), A1689 was covered with 54 orbits (B₄₃₅, r₆₂₅, i₇₇₅, I₈₁₄, z₈₅₀, J₁₂₅, J₁₄₀, H₁₆₀), the Bullet cluster was covered with 16 orbits (V₆₀₆, i₇₇₅, z₈₅₀, J₁₁₀, H₁₆₀) and clusters in the Kneib et al. program were covered with 6 orbits (I₈₁₄, J₁₁₀, H₁₆₀). HST mosaics were produced using the Mosaicdrizzle pipeline (see Koekemoer et al. 2011 for further details), and individual bands in the deep imaging data reach 5σ depths of 26.4–27.7 mag (04-diameter aperture).

Deep Spitzer/IRAC observations of our fields in the [3.6] and [4.5] bands were provided by the ICLASH (GO 80168; Bouwens et al. 2011), the Ultra-Deep IRAC imaging of Massive Lensing Galaxy Clusters (GO 20439; PI: Egami) and the Spitzer IRAC Lensing Survey program (GO 60034; PI: Egami). The typical exposure time per cluster was 3.5–5 hr per band, allowing us to reach 26.5 mag at 1σ. Reductions of the IRAC observations used in this paper were performed with MOPEX (Makovoz & Khan 2005).

The photometry we obtain for sources in our cluster fields follows a similar procedure as described in Bouwens et al. (2012b). In short, we run the SExtractor software (Bertin & Arnouts 1996) in dual-image mode. The detection images are constructed from all bands redward of the Lyman break (i.e., Y₁₀₅, J₁₁₀, J₁₂₅, JH₁₄₀ and H₁₆₀). After point-spread function (PSF)-matching the observations to the H₁₆₀-band PSF, colors are measured in Kron-like apertures and total magnitudes derived from 06 diameter circular apertures.

Our initial source selection is based on the Lyman-break technique (Steidel et al. 1999), with the requirement that the source drops out in the I₈₁₄ band. Specifically, our requirements for z ∼ 6–7 sources are

$$(I_{814}-J_{110}>0.7)\,\wedge \,(J_{110}-{JH}_{140}<0.45).$$

For sources in the CLASH program, we require H₁₆₀ < 26 AB, while we select sources to the brighter magnitude limit H₁₆₀ < 25 AB in all other fields to ensure good photometric redshift constraints for all our sources. We also require sources to have either a nondetection in the V₆₀₆ band (<2σ) or to have a very strong Lyman break, i.e., V₆₀₆ − J₁₂₅ > 2.5. We require sources to be undetected in the optical χ² image (Bouwens et al. 2011) we construct from the observations blueward of the r₆₂₅ band. Finally, we require the SExtractor stellarity parameter (equal to 0 and 1 for extended and point sources, respectively) in the J₁₁₀ band be less than 0.92 to ensure that our selection is free of contamination by stars.

To identify those sources where we can obtain clean rest-frame optical stellar continuum, we also require that sources have a best-fit photometric redshift between z = 6.6 and 7.0, as determined by the photometric redshift software EAZY (Brammer et al. 2008). All available HST photometry (i.e., 16 bands for CLASH clusters) is used in the redshift determinations. No use of the Spitzer/IRAC photometry is made in the photometric redshift determination to avoid coupling the selection of our sources to the [3.6]–[4.5] colors we will later measure. We use templates of young stellar populations with no Lyα emission.

Strong Lyα emission can systematically influence the photometric redshift estimate. However, we emphasize that any potential sources from outside our desired redshift interval that could be in our sample because of uncertainties in the photometric redshift estimate would serve only to increase the flux in the [4.5] band and redden the [3.6]–[4.5] color (i.e., because of contamination in the [4.5] band of Hα at z < 6.6 and [O iii] + Hβ at z > 7.0). Correcting for this possible source of interlopers would result in higher EWs and sSFRs than in the case of no contamination. This reinforces the point we will make in Section 3 that the EWs we derive for the [O iii] + Hβ emission and the sSFRs are strong lower limits on the actual values.

Figure 1 shows the redshift range where we would expect the strongest emission lines, Hα, Hβ, [O iii] λλ4959, 5007 and [O ii] λ3727, to impact the [3.6] and [4.5] fluxes. The top panel indicates which lines fall in specific IRAC filters at a given redshift, while the bottom panel indicates the estimated [3.6]–[4.5] color offset caused by the various emission lines. We select sources in the redshift range z_phot = 6.6–7.0, where we know that both [O iii] and Hβ fall in [3.6], while [4.5] falls exactly between [O iii] and Hα where no significant emission lines are present (see for example Figure 2).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Photometry of sources in the available Spitzer/IRAC data over our fields is challenging because of the blending with nearby sources from the broad PSF. We therefore use the automated cleaning procedure described in Labbé et al. (2010a, 2010b). In short, we use the high-spatial-resolution HST images as a template with which to model the positions and flux profiles of the foreground sources. The flux profiles of individual sources are convolved to match the IRAC PSF and then simultaneously fit to all sources within a region of ∼13" around the source. Flux from all the foreground galaxies is subtracted and photometry is performed in 25 diameter circular apertures. We apply a factor of ∼2.0 × correction to account for the flux outside of the aperture, based on the radial light profile of the PSF. Figure 3 shows the cleaned IRAC images of our sample. Our photometric procedure fails when contaminating sources are either too close or too bright. Sources with badly subtracted neighbors are excluded. In total, clean photometry is obtained for 78% of the sources, resulting in seven sources in our final selection (excluding only one source behind RXJ1347 and one source behind MACS1206 from our sample).

Zoom In Zoom Out Reset image size

Our selection of sources in the redshift range z ∼ 6.6–7.0 provides us with the valuable opportunity to establish the typical EW of the nebular emission lines in z ≳ 6 sources through a comparison of the flux in [3.6] and [4.5]. For a given EW, we calculate the [3.6]−[4.5] color by assuming that the observed flux in one filter can be approximated by the following:

$$\begin{eqnarray} f_{\rm \nu,obs}&=& f_{\rm \nu,continuum}\cdot x_{\rm EW} \nonumber\\ x_{\rm EW} &=& \left(1+ \sum \limits _{{\rm lines},i}\frac{EW_{0,i}\cdot (1+z)\cdot R(\lambda _{{\rm obs},i})}{\lambda _{\rm obs,i}\int R(\lambda)/\lambda \, d\lambda }\right). \end{eqnarray} \tag{ 1 }$$

Here, f_{ν, continuum} is the intrinsic flux of the stellar continuum, while f_{ν, obs} is the observed flux of the filter. Furthermore, λ_{obs, i} is the observed wavelength of the line, and R(λ) is the response curve of the filter. We include all nebular emission lines listed in Anders & Fritze-v. Alvensleben (2003) and the hydrogen Balmer lines. We assume fixed flux ratios between the emission lines based on the tabulated values in Anders & Fritze-v. Alvensleben (2003) for 0.2 Z_☉ metallicity and assuming case B recombination. The observed [3.6]−[4.5] color then simplifies to the following:

$$\begin{eqnarray} &&[3.6]\hbox{--}[4.5]=([3.6]\hbox{--}[4.5])_{\rm continuum} -2.5\cdot \log \left(\frac{x_{\rm EW,3.6}}{x_{\rm EW,4.5}}\right).\nonumber\\ \end{eqnarray} \tag{ 2 }$$

LBGs at high redshift are expected to exhibit a flat optical stellar continuum based on stellar population synthesis models. In these models, young galaxies with typical ages between 50 and 200 Myr and low dust extinction, e.g., E(B − V) ∼ 0.1, will have a ([3.6]−[4.5])_continuum color of ∼0 ± 0.1 mag. However, extremely young (i.e., ∼3 × 10⁶ yr), dust-free galaxies can exhibit ([3.6]−[4.5])_continuum colors as blue as ∼  − 0.4. To be conservative, we adopt this for the color of the underlying stellar continuum and assume that any bluer [3.6]–[4.5] color arises from the impact of emission lines to establish robust lower limits.

In the bottom panel of Figure 1, the dotted line shows a prediction of the observed optical color due to emission lines, using Equations (1) and (2) and assuming a flat continuum, for a model of strongly increasing rest-frame emission line EWs as a function of redshift (dotted line), with EW₀([O iii] + Hβ)∝(1 + z)^1.8 Å, based on the evolution in EW₀(Hα) found by Fumagalli et al. (2012) for star-forming galaxies over the redshift range 0 ≲ z ≲ 2. The red points in Figure 1 show the observed colors for our sample. Most of our sources show quite blue [3.6]–[4.5] colors, and essentially all of them are bluer than expected based on a conservative model of constant rest-frame EW (solid black line: i.e., assuming no evolution from z ∼ 2 where EW₀([O iii] + Hβ) ∼ 140 Å, derived from the Hα EWs found by Erb et al. 2006). It is interesting that three of the sources from our sample have [3.6]–[4.5] colors even bluer than expected at z ∼ 6.7–6.8 for the model from Fumagalli et al. (2012) with EW₀(Hα)∝(1 + z)^1.8 Å. Four of the sources have [3.6]–[4.5] colors bluer than −0.8. Because we would expect galaxies to show such extreme [3.6]–[4.5] colors only in the narrow redshift range z ∼ 6.6–7.0, this assures us that our selection can effectively identify sources in the desired redshift range.

To obtain our best measurement of the [4.5] flux and hence stellar continuum light from z ∼ 7 galaxies, we construct a mean SED that can be directly compared with studies of the mean rest-frame optical properties of z ≳ 4 galaxies in the literature (e.g., González et al. 2012, 2014; Stark et al. 2013). We use a mean stack of the clean [3.6] and [4.5] images after dividing by the observed rest-frame UV luminosity (the geometric mean of the J₁₂₅, JH₁₄₀ and H₁₆₀ luminosities). We measure the flux in a 25 diameter aperture on the stacked image and apply an aperture correction measured from the PSF images (∼2.0 ×). Also, we use the mean of the individual measurements of the HST bands, after dividing by the rest-frame UV-luminosity. The mean SED of our stacked z ∼ 6.8 sample is shown in Figure 4. Errors are obtained through bootstrap resampling.

Zoom In Zoom Out Reset image size

We use the stacked detections in the IRAC bands to evaluate the mean contribution of the emission lines. From the mean [3.6]–[4.5] color, we estimate the [O iii] + Hβ EW by assuming that our entire sample is at z = 6.76, where we expect the most extreme colors because [4.5] is completely free of emission lines, whereas [3.6] is contaminated by both the [O iii] doublet and Hβ. In practice, this results in an underestimate of the intrinsic line strength because we know that the [O iii] lines start to drop out of [3.6] at z  ∼  6.9–7.0. Therefore, we expect a less extreme [3.6]–[4.5] color for a given mean EW at z ∼ 6.9–7.0 than at z ∼ 6.7–6.8. It is also possible that because of uncertainties in the photometric redshifts, sources outside of our target redshift range have been included in our selection; therefore, the measurement of the [4.5] flux is contaminated by either Hα (z < 6.6) or [O iii] (z > 7). This would also make the mean [3.6]–[4.5] color redder and accordingly make the emission lines appear to be less extreme.

The mean observed [3.6]–[4.5] color for our sample is −0.9 ± 0.3 (error obtained through bootstrap resampling). In the most conservative estimate, we assume that the underlying stellar continuum exhibits a [3.6]–[4.5] color of ∼−0.4; therefore, the [O iii] and Hβ are responsible for a color of [3.6]–[4.5] ∼ −0.5, which would give a robust lower limit of EW₀([O iii] + Hβ) ≳ 637 Å for the mean z ∼ 7 galaxy distribution. The [O iii] + Hβ EW we estimate here is equivalent to EW₀(Hα + [N ii])  ≳  495 Å, adopting the tabulated values from Anders & Fritze-v. Alvensleben (2003) for 0.2 Z_☉ metallicity and assuming case B recombination.

The mean observed [3.6]–[4.5] color for our four bluest sources is −1.4  ±  0.4. If we assume again a very blue underlying continuum (i.e., −0.4), line emission would be responsible for a color of [3.6]–[4.5] ∼ −1.0, consistent with a robust lower limit of EW₀([O iii] + Hβ) ≳ 1582 Å for these sources. The individual lower limits for these blue sources are listed in Table 2. Although the four bluest sources in our sample are, given their extreme colors, almost certainly at a redshift z ∼ 6.7–6.8, it is unclear whether the three other sources are less extreme because of a lower [O iii] + Hβ EW or simply because they lie in a different redshift range (i.e., z ≳ 6.9 or close to z ∼ 6.6), where for a given EW we expect somewhat redder colors.

To obtain a good estimate of the EW, we need detailed knowledge of the redshift distribution and the underlying stellar continuum color. Given our lack of deep spectroscopy for our sample, we make some reasonable assumptions to obtain a model estimate of the rest-frame EW of our sample. We assume that the redshift probability distribution of our sources is given by the sum of the probability distributions obtained from our photometric redshift code, corrected for the fact that galaxies are more difficult to observe at higher redshift scaling roughly as (dlog ϕ/dz) = 0.3 (Bouwens et al. 2012a) and assuming no sources are outside our desired redshift range z ∼ 6.6–7.0. We use this redshift probability distribution in our sample to estimate the expected [3.6]–[4.5] color distribution, for a given mean EW₀([O iii] + Hβ) and 0.3 dex scatter around the mean. We assume the underlying continuum color is −0.25, as would be expected for a galaxy age of ∼10 Myr. We randomly draw sources from the distribution and calculate the 68% likelihood of finding a mean 〈[3.6]–[4.5]〉 color given a total of seven observed sources and the observed photometric errors. On the basis of this modeling, the observed [3.6]–[4.5] ∼ −0.9 mag color is consistent with a possible EW₀([O iii] + Hβ) EW of ${\sim } 1806_{-863}^{+1826}$ Å. This is equivalent to EW₀(H$\alpha + [\rm N\,{\scriptsize{II}}])\sim 1323^{+1338}_{-632}$ Å, adopting the same conversion factor from Anders & Fritze-v. Alvensleben (2003) assumed earlier.

The aforementioned modeling we perform indicates that the true EW₀([O iii] + Hβ) may be ∼2–3 × larger than our robust lower limit of 637 Å. Figure 5 compares our results with other determinations from the literature (Erb et al. 2006; Shim et al. 2011; Fumagalli et al. 2012; Stark et al. 2013; Labbé et al. 2013). The solid and dashed lines in Figure 5 show the expected evolution of the Hα + [N ii] EWs extrapolating the evolution found in Fumagalli et al. (2012) at z ∼ 0–2. We note, however, that a direct comparison is difficult to make because the Fumagalli et al. (2012) relation was derived for galaxies in the mass range M_* = 10¹⁰–10^10.5 M_☉, while we are probing galaxies in the mass range 10⁹–10^9.5 M_☉. As a reference, we show the possible evolution of galaxies M_* = 10⁹–10^9.5 M_☉ (top dashed line), using the same scaling with redshift EW₀(Hα + [N ii])∝(1 + z)^1.8 Å but extrapolating the normalization to lower masses, based on the mass trend in the SDSS-DR7 data derived in Fumagalli et al. (2012).

Zoom In Zoom Out Reset image size

In general, the EWs we infer are in good agreement with extrapolations from previous results at lower redshift. However, our results are based on a UV-selected sample, which could yield different results from a mass-complete sample. It is clear nonetheless that our EWs estimates strongly support the high EWs used by Stark et al. (2013) and González et al. (2014) in correcting the SEDs of z ∼ 5–7 samples to derive higher values of the sSFRs.

The redshift range where we select galaxies is the only redshift window at z ≳ 5 where we can probe the rest-frame stellar continuum light in an uncontaminated fashion, using the [4.5] IRAC band (see Figure 1). This allows us to estimate the sSFR with minimal contamination from emission lines.

Because of the low S/N of the uncontaminated [4.5] band, we can set only weak constraints on the sSFRs of the individual galaxies (see Table 2). To optimize the S/N, we perform stellar population modeling of our stacked photometry, leaving out the [3.6] measurement. We performed the modeling with FAST (Kriek et al. 2009), using the Bruzual & Charlot (2003, hereafter BC03) stellar populations synthesis models. We use a Salpeter (1955) IMF with limits 0.1–100 M_☉ and a Calzetti et al. (2000) dust law. We consider ages between 10 Myr and the age of the universe at z ∼ 6.8 and dust extinction between A_V = 0–2. We assume a constant star-formation history and subsolar metallicity (0.2 Z_☉). For the mean stack, we fix the redshift to the median of the photometric redshifts, at z = 6.77.

Given this freedom of parameters, the mean SED is best described by a fairly young galaxy (age ≲ 100 Myr) and reasonable dust (A_V ∼ 0.7) to fit both the small Balmer break (H₁₆₀ − [4.5] ∼ 0.2) and moderately red UV-continuum slope (β ∼ −1.9), resulting in a notably high sSFR of $\rm 52^{+50}_{-41}\,Gyr^{-1}$ (see the SED in Figure 4). However, the interpretation of this result is not straightforward. First, we have only the [4.5] band in the rest-frame optical to break the age-dust degeneracy. Given our modest sample, a range of models can still fit the data well. We obtain a possibly more insightful answer when we fix the dust to the expected value derived from the typical spread of UV-continuum slopes and the Meurer et al. (1999) law (e.g., Bouwens et al. 2012b) similar to the assumptions made in Bouwens et al. (2012b), Stark et al. (2013), González et al. (2014), and Labbé et al. (2013). This results in a dust content of A_V = 0.38 ± 0.16 using the latest numbers from Bouwens et al. (2013). In turn, we fix the dust content to A_V = [0.22, 0.38, 0.54]. The fit for A_V = 0.38 is shown with the thin red line in Figure 4. We obtain a sSFR of $\rm 7^{+7}_{-3}\,Gyr^{-1}$. Given that the reduced χ² with the lowest dust content A_V = 0.22 is only ∼1.3 times higher than for our best fit (χ² =1.79 versus χ² =1.33), we assume a conservative lower limit of the sSFR of ≳ 4 Gyr⁻¹.

Alternatively, we can estimate the sSFR of our z ∼ 7 sample from the [O iii] + Hβ EWs we infer, by converting to Hα EW assuming same line ratios from Anders & Fritze-v. Alvensleben (2003) as described in Section 3.2. We use the Kennicutt (1998) relation to convert Hα luminosity to SFR, and we use BC03 models (assuming no dust) to convert the rest-frame optical continuum light to stellar mass. Using these assumptions, we obtain a lower limit on the sSFR for our seven source samples and the bluest four sources of ∼14 Gyr⁻¹ and ∼130 Gyr⁻¹, respectively, on the basis of the robust lower limits on the [O iii] + Hβ EW derived in Section 3.2.

Comparing with direct constraints at z ∼ 2, we estimate ≳ 2 × evolution in the sSFR over this redshift range (see Figure 6), in good agreement with estimates at z ∼ 7 on the basis of a few spectroscopically confirmed sources (Ono et al. 2012; Tilvi et al. 2013) and extrapolating the Hα EWs from lower redshifts (Yabe et al. 2009; Stark et al. 2013; González et al. 2014). Our derived constraint is also in agreement with theoretical models that predict the sSFR to follow the specific infall rate of baryonic matter (e.g., Neistein & Dekel 2008).

Zoom In Zoom Out Reset image size