A REFINED ESTIMATE OF THE IONIZING EMISSIVITY FROM GALAXIES AT z ≃ 3: SPECTROSCOPIC FOLLOW-UP IN THE SSA22a FIELD^*

Our analyses make use of deep multiband imaging available in the SSA22a field. These data, which are described in detail in Nestor et al. (2011), include ground-based broad B-, V-, and R-band images and narrowband images with effective wavelengths at λ ∼ 3640 Å and λ ∼ 4980 Å (hereafter NB3640 and NB4980, respectively). The NB3640 and part of the NB4980 data were obtained using the Keck/LRIS imaging spectrograph, while the broadband imaging and the remainder of the NB4980 data were obtained with the Subaru/Suprime-Cam. Additionally, archival HST/ACS F814W imaging is available for 70% of the field.

The NB3640 filter has a central wavelength of 3635 Å and an FWHM of 100 Å. For sources at z ≃ 3.09, NB3640 samples the rest-frame spectral range λ ≃ 875–900 Å. It is opaque to wavelengths longward of the redshifted Lyman limit for sources above z ≃ 3.055 and therefore provides a clean probe of escaping LyC emission for such galaxies. The effective wavelength of the broad R-band filter (λ ≃ 6510 Å) corresponds to rest-frame λ ≃ 1600 Å at z ≃ 3.09. The NB3640−R color is accordingly a measure of the non-ionizing to ionizing UV flux-density ratio, F_UV/F_LyC, for galaxies with z ⩾ 3.055. The ACS/WFC F814W filter has an effective wavelength of λ ≃ 8090 Å and also probes the non-ionizing UV continuum at z ∼ 3.09. The footprint of our NB3640 image contains 109 color-selected LBG candidates, of which 28 (including 2 QSOs) have previous spectroscopically confirmed redshifts z ⩾ 3.055, as well as 41 LBG candidates without previous spectroscopic confirmation. The footprint also contains 110 LAE candidates identified by Nestor et al. (2011).

LAE candidates were selected using the B, V, and NB4980 data. The NB4980 filter covers redshifted H i Lyα λ1216 Å for galaxies with 3.05 ≲ z ≲ 3.12. We created a linear combination of the broad B and V images after scaling to a common photometric zero point, such that our so-called "BV" image has an effective wavelength of λ ≃ 4980 Å. The BV − NB4980 color is therefore a measure of Lyα equivalent width for galaxies with 3.05 ≲ z ≲ 3.12. The LAEs in the sample described by Nestor et al. (2011) were required to have NB4980 ⩽26 and BV − NB4980 ⩾ 0.7, which corresponds to an Lyα rest-frame equivalent width (REW) of ≳ 20 Å. We also subtracted the BV image from the NB4980 image to create a so-called "LyA" image, which was used to define the centroids of Lyα emission from the LAE candidates.

We performed spectroscopy in the SSA22a field using both shallow and deep observations. The purpose of the shallow observations was to obtain spectroscopic redshifts for LBG and LAE photometric candidates, while the deep observations were intended to provide detailed information for objects with NB3640 detections. Multi-object spectroscopy was performed using the blue arm of the LRIS dichroic spectrograph on Keck I (Oke et al. 1995; Steidel et al. 2004). Nine shallow slit masks were observed over the course of five observing runs in 2009 June, 2009 September, 2010 July, 2010 August, and 2011 May. A single deep mask was observed in 2010 August. Typical exposure times for the shallow masks were 5400 s (ranging from 3600 to 6270 s), while the deep mask was observed for 23,700 s. For most runs, the conditions were photometric, with seeing ranging from 05 to 07. In 2011 May, when additional data were collected for one of the shallow masks, conditions featured variable cloudiness and seeing ranging from 08 to 10.

The primary targets for shallow masks were LBG and LAE photometric candidates without previously determined spectroscopic redshifts. LBGs with spectroscopic redshifts were added as filler targets on the mask. A total of 50 LBGs and 114 LAEs were targeted on the shallow masks. Slits were centered on the detections in the NB4980 (i.e., Lyα) for LAEs, and in the R band (i.e., rest-frame UV continuum) for LBGs. Seven out of nine shallow masks were observed using a 300 line mm⁻¹ grism blazed at 5000 Å, along with the "d680" dichroic beam splitter, sending light with wavelengths bluer than ∼6800 Å to the blue arm of LRIS. The spectral resolution for these masks was R = 530. The two remaining shallow masks were observed at higher spectral resolution using a 600 line mm⁻¹ grism blazed at 4000 Å, one with the "d560" dichroic (splitting the incoming light beam at ∼5600 Å), and the other with the "d680" dichroic. The spectral resolution for these two masks was R = 1200.

The primary targets for the deep mask were LBGs and LAEs with NB3640 detections. A sample of 4 LBGs and 13 LAEs with NB3640 detections was targeted on the deep mask (along with one LBG and four LAEs lacking NB3640 detections, but added as filler). Slits for objects with NB3640 detections were centered on the detections in the NB3640 image (i.e., LyC for objects at z ⩾ 3.055), while the NB4980 and R-band detections were used, respectively, for centering the slits for the filler LAEs and LBG without NB3640 detections. The deep mask was observed using a 400 line mm⁻¹ grism blazed at 3400 Å, along with the "d680" dichroic beam splitter, sending light with wavelengths bluer than ∼6800 Å to the blue arm of LRIS. The spectral resolution for this mask was R = 700.

The data were primarily reduced using IRAF tasks, with scripts designed for cutting up the multi-object slit-mask images into individual slitlets, flat-fielding using spectra of the twilight sky, rejecting cosmic rays, subtracting the sky background, averaging individual exposures into final stacked two-dimensional spectra, extracting to one dimension, wavelength and flux calibrating, and shifting into the vacuum frame. These procedures are described in detail in Steidel et al. (2003). There were a couple of notable differences in the data reduction procedures used for this sample, relative to the typical LBG reduction strategy. First, we used a custom IDL script (N. Reddy 2010, private communication) to rectify the curved slitlets before performing any of the standard IRAF reduction tasks. Since the majority of our targets are LAEs with negligible continuum and a single emission line, we required slit rectification in order to extract a spectrum over a broad wavelength range at the location of the object indicated by the isolated bright Lyα feature. Likewise, for deep-mask spectra, the faintness of the continuum level in the LyC region precluded a robust trace without rectification. We also followed the procedures outlined in Shapley et al. (2006) for background subtraction of deep-mask spectra. Accordingly, to avoid potential over-subtraction of the background, the object continuum location was excluded from the estimate of the background fit at each dispersion point. We used the maximum possible number of pixels to fit the sky emission for each object. In practice, the widths of the sky regions on either side of the continuum location depended on the length of each slitlet and the position of the object along the slit.

For LAEs, redshifts were calculated from the observed centroid of the Lyα emission feature (rest-frame λ_Lyα = 1215.67 Å). For LBGs, emission redshifts were estimated from the observed centroid of Lyα, and absorption redshifts from the centroids of interstellar metal absorption features when present.

Finally, as described in B. Siana et al. (in preparation), near-IR spectra of several objects in the sample were obtained in 2011 August with NIRSPEC (McLean et al. 1998) on Keck II. These data were collected for a separate project; we refer to these data here as, in three cases, the near-IR spectra reveal emission features relevant to the interpretation of the NB3640 detections (see Sections 4.1.2 and 4.1.3).

In this section, we discuss the individual LBGs having NB3640 detections for which we have new data. We begin with the four systems which we retain as possible LyC-leaking galaxies. We next discuss the three LBGs for which we find evidence for the presence of a foreground interloper in our new data. We then discuss the special case of aug96M16 and conclude the section with a summary of our LBG NB3640 detection sample.

4.1.1. LBG Lyman-continuum Candidates

4.1.2. LBGs With Evidence for Contamination

4.1.3. Unconfirmed LBG Redshift

4.1.4. Summary of LBGs

As above, in this section we discuss the individual LAEs with NB3640 detections. We begin with an overview of the 17 systems which we retain as possible LyC-leaking galaxies. We next discuss the three LAEs for which we find evidence for the presence of a foreground interloper. We then discuss the six LAEs with NB3640 detections for which we were unable to confirm redshifts and are thus not included in our sample, and conclude the section with a summary of our LAE NB3640 detections.

4.2.1. LAE Lyman-continuum Candidates

4.2.2. LAEs With Evidence For Contamination

We find evidence for the presence of a foreground galaxy in the spectra of three of our LAEs, which we present below. Additionally, one of the LAE candidates from Nestor et al. (2011), LAE034, is at a low enough redshift that it experiences contamination of its NB3640 flux from non-ionizing UV radiation. We have thus removed LAE034 from our list of possible LyC-emitting galaxies as well as from the parent z ⩾ 3.055 LAE sample.

LAE003. LAE003 has the brightest NB3640 detection (m_NB3640 = 24.74) of any of our LBGs or LAEs in Nestor et al. (2011). The top-left panel of Figure 13 shows the F814W image indicating the relative positions of the non-ionizing UV flux, the deep- and shallow-mask slits, the Lyα flux, and the NB3640 flux. The middle panel shows the region of the two-dimensional spectrum corresponding to the expected location of the redshifted Lyα line. We identify the line detected at λ = 4980 as Lyα, indicating that LAE003 is at z = 3.097.

However, as can be seen in the lower panel, which shows a bluer region of the two-dimensional spectrum, there is another emission line spatially consistent with the LAE. If this line is also Lyα emission, it corresponds to a redshift of z = 1.763. The bluer line is very slightly offset to the south, consistent with the tail-like structure seen in the HST/ACS image. It is therefore possible that the z = 3.097 LAE003 is opaque below the Lyman limit and the NB3640 flux is non-ionizing continuum emerging from this lower-redshift galaxy. It is noteworthy that the NB3640 flux is not centered on the portion of the F814W flux that we associate with the lower-redshift interloper, and extends to the northwest. However, due to the clear presence of an interloper we take the conservative approach and remove LAE003 from our list of possible LyC-leaking galaxies. The one-dimensional extracted spectra are shown in the right panel of Figure 13.

LAE003 had previously been studied by Inoue et al. (2011) who determined a spectroscopic redshift of z = 3.100. Their spectrum did not extend blueward enough to detect the emission line at λ = 3359 Å, however. As the NB3640 (NB359 in Inoue et al. 2011) flux is spatially coincident with the R band and LyA flux for LAE003, these authors argue that the probability of foreground contamination is small and therefore the NB359 flux is indeed LyC. They interpret the implied very high ratio of ionizing to non-ionizing UV flux density of four objects, including LAE003, in terms of very young stellar populations with top-heavy initial mass functions. This interpretation is invalidated for LAE003 by the discovery of the low-redshift interloper in our deep-mask spectrum. The case of LAE003 highlights the importance of obtaining spectra extending as far to the blue as possible for properly interpreting possible LyC-leaking galaxies.

LAE019. We do not have HST imaging of LAE019. The top panel of Figure 14 shows our BV image with the deep- and shallow-mask slits indicated. The Lyα emission centroid is slightly offset to the north in both the imaging and two-dimensional deep-mask spectrum (middle panel), while the NB3640 flux centroid is slightly offset to the south in our imaging. The deep-mask spectrum of LAE019, shown in Figure 14, also exhibits a bluer emission feature at λ ≃ 3490 Å. If this feature is Lyα emission, it would indicate the presence of an interloper at z = 1.872. The bottom panel shows the two-dimensional spectrum in the λ ≃ 3490 Å region. This lower-redshift emission line is spatially consistent with the NB3640 flux, indicating that the NB3640 flux is likely due to the interloper. Therefore, we remove LAE019 from our list of possible LyC-leaking galaxies.

LAE025. We obtained spectra of LAE025 in both shallow and deep masks. The top panels of Figure 15 show the HST/ACS image, with the corresponding slit positions (shallow left; deep right). The inset in the top-left panel shows the central region without contours, after smoothing by a Gaussian kernel with FWHM =2.35 pixels (01). The bulk of the NB3640 flux is associated with the central clump. We detect flux in neither the F814W nor BV images at the location of the LyA flux centroid. The middle-left panel of Figure 15 shows the region of the two-dimensional shallow-mask spectrum corresponding to the expected location of the redshifted Lyα line. We detect a strong emission line spatially coincident with the LyA flux, which we identify as Lyα at z = 3.091. However, the non-ionizing UV flux detected in the HST/ACS image that is coincident with the NB3640 flux is just off of the slit. The middle-right panel shows the same spectral region of the deep-mask spectrum, for which the slit was centered on the NB3640 detection. No obvious emission at λ ≃ 4974 Å is seen, as might be expected if the region coincident with the NB3640 flux was also at z = 3.091. Based on possible detections of Lyα emission at λ ≃ 4942 Å and perhaps corresponding Si ii interstellar absorption, we tentatively assign a redshift of z = 3.065. This is a high enough redshift such that the NB3640 filter is still opaque to radiation longward of the Lyman limit. It should be noted, however, that the LAE at z = 3.091 for which we searched for corresponding NB3640 flux would be unrelated to the z = 3.065 object (the relative velocity between the two Lyα lines being ≃ 1800 km s⁻¹) associated with the NB3640 flux; their proximity on the sky would be coincidental.

Further complicating this system is the presence of an additional, much bluer line at λ ≃ 4035 Å, which is detected in both spectra (Figure 16). It is spatially consistent with the faint, low surface brightness flux detected in the F814W image to the north–northwest of the central clump, which falls in both slits. If identified as Lyα, it implies a redshift of z = 2.319.

Zoom In Zoom Out Reset image size

With the data at hand it is difficult to either confirm or refute the claim that the detected NB3640 flux is ionizing continuum. Nonetheless, considering (1) there is evidence for a nearby foreground object, and (2) our identification of the redshift of the blob associated with the NB3640 emission is tentative, we remove LAE025 from our list of possible LyC detections.

4.2.3. Unconfirmed LAE Candidates with NB3640 Detections

4.2.4. Summary: LAEs

The centroids of the NB3640 detections in our samples are generally offset from the corresponding centroids of the non-ionizing UV emission. Many of these NB3640 detections, despite their relatively close proximity on the sky to LBGs and LAEs, have clear associations with unrelated neighboring sources and thus have already been rejected as possible LyC emission by Nestor et al. (2011). We have now also removed from our samples those galaxies with unconfirmed redshifts, and have rejected NB3640 detections displaying evidence for contamination in their LRIS or NIRSPEC spectra. In Figure 17 we show the resulting surface density of the total (open histogram) and rejected (hatched region) number of NB3640 detections, as a function of offset, for both the LBG and LAE samples. For the LAEs, we present surface densities computed using the offsets of the NB3640 detections from both the UV continuum (i.e., R band or BV for LAE081 which is undetected in R) and LyA-band centroids. The dashed line in Figure 17 indicates the global surface density of sources in the range of NB3640 magnitudes spanned by our detections, ρ_S = 0.024 arcsec⁻² over 25 ⩽ m_NB3640 ⩽ 27.25, which corresponds to the expected level of contamination. Both the LBG and LAE samples have, at relatively small (≲ 1'') offsets, an excess of NB3640 detection surface density above the expected foreground level. In particular, the excess for the LAEs using R-band offsets is strikingly significant. In order to quantify the number NB3640 detections that may be due to foreground interlopers, in addition to those already rejected, we performed a Monte Carlo simulation using the observed surface densities of sources and predicted levels of contamination.

Zoom In Zoom Out Reset image size

To account for the increased probability that NB3640 detections at larger offsets are contaminants, we considered each individual detection in turn, computing the contamination probability based on offset in the following manner. For the k^th detection, we constructed an annulus, having area A^k, encompassing the detection and centered on the non-ionizing UV flux centroid. We applied this same annulus to each of the N galaxies in the sample (N = 41 or 91 for the LBG or LAE samples, respectively) and determined the number of previously rejected detections, n^k_rej, and putative LyC detections, n^k_LyC, contained within the N annuli. We then constructed the probability distribution for the number of expected random foreground interlopers, n^k_fore, in the regions spanned by the annuli:

$$\begin{equation} P\left(n_{{\rm fore}}^k\right) = \left(\begin{array}{@{} c @{}}N \\ n_{{\rm fore}}\end{array}\right)\,p^{n_{{\rm fore}}}\,(1-p)^{(N-n_{{\rm fore}})}, \end{equation} \tag{ 1 }$$

where p = A^k × ρ_S is the probability that a given LBG or LAE has a random foreground contaminant in A^k (see, e.g., Vanzella et al. 2010; Nestor et al. 2011). In each realization of the Monte Carlo simulation, we randomly chose a value for n^k_fore from P(n^k_fore). Of these n^k_fore interlopers, n^k_rej had already been accounted for. Thus the number of additional interlopers predicted to lie within the N annuli is n^k_fore − n_rej^k. The NB3640 detection in question was flagged as an interloper if n^k_fore − n_rej^k ⩾ n^k_LyC, and was retained if n^k_fore − n_rej^k ⩽ 0. Otherwise, we randomly determined if the NB3640 detection was to be flagged as an interloper based on a probability =(n^k_fore − n_rej^k)/n^k_LyC. We then proceeded to the next (i.e., k^th + 1) detection and repeated the entire process. Once each possible LyC detection in the sample had been considered, we recorded the total predicted number of additional interlopers. The simulation was repeated for a total of 1000 iterations to determine the expected number and uncertainty in the average number of uncontaminated NB3640 detections.

The distribution of the number of NB3640 detections flagged as interlopers in each realization of the simulation is shown in Figure 18 for the LBG and LAE samples. The widths of the distributions depend slightly on the size of annular apertures used in the simulations due to NB3640 detections stochastically entering and exiting the apertures when their sizes were varied. The variability in the distribution widths was small and did not depend systematically on the aperture sizes. We used a circle of radius equal to the seeing FWHM in the NB3640 image (08) for detections with offsets ⩽04. For detections at larger offsets, the radii of the annuli were set such that all values of A^k were equal. Our simulation suggests that 2.6 ± 1.2 of the six possible LBGs with LyC detections are contaminated by foreground sources. Together with the three LBGs showing evidence for contamination in their spectra discussed in Section 4.1.2, the resulting contamination rate is 62% ± 13% (i.e., 5.6 ± 1.2 of 9). The contamination-corrected detection rate for the sample as a whole is 8% ± 3% (3.4 ± 1.2 of 41).

Zoom In Zoom Out Reset image size

Using R-band offsets, our simulation suggests that 3.8 ± 1.3 of the 17 possible LAEs with LyC detections are contaminated. Together with the three LAEs with contamination discussed in Section 4.2.2, the resulting contamination rate for the LAE sample is 34% ± 7% (6.8 ± 1.3 of 20) and the contamination-corrected detection rate is 15% ± 1% (13.2 ± 1.3 of 91). If we instead use LyA offsets, the predicted number of additional contaminants becomes 6.4 ± 1.9, the contamination rate becomes 47% ± 10% (9.4 ± 1.9 of 20), and the detection rate becomes 12% ± 2% (10.6 ± 1.9 of 91).

In addition to estimating the number of foreground interlopers, our simulation computes the (contamination-corrected) average LyC and R magnitudes and uncertainties. The uncertainties include sample variance computed by first randomly reassigning individual magnitudes based on the measured magnitude and error, assuming Gaussian magnitude uncertainties determined from our photometric simulations, and then bootstrap resampling each data set. When computing the sample-average LyC magnitudes, we alternately assumed that NB4630 non-detections and detections flagged as interlopers had flux levels equal to zero or the maximum average magnitude consistent with the (1σ) limits set by our stacking analysis. The resulting contaminated-corrected sample-average NB3640−R colors are listed in Table 4.

To investigate the attenuation of escaping LyC flux by absorption from the IGM, we used the observed column density and frequency distributions of the Lyα forest over the relevant redshift range, 1.7 ⩽ z ⩽ z_source, to model 500 random sightlines through the IGM for each LBG and LAE redshift. We then computed the attenuation of the continuum flux in the NB3640 filter due to the randomly generated neutral clouds. In this manner we determined the fraction of flux transmitted, t^j, for each of the 500 sightlines. The process was identical to that described in Nestor et al. (2011) except that we made use of updated estimates of the IGM opacity (Rudie et al. 2013) and a more precise treatment of the higher-order Lyman absorption lines from each cloud. As the galaxies in our sample lie either in or behind the SSA22a protocluster, it may be expected that the IGM in the proximate foreground of our sources is either more opaque due to the mass overdensity of the region or less opaque due to the overdensity of ionizing emissivity. In either scenario, the difference in average IGM opacity should manifest in differences in rest-frame UV colors that span Lyα emission and/or the Lyman limit. We compared the U − G and G − R colors for our sources at 3.06 ⩽ z ⩽ 3.12 with those of field LBGs in the same redshift range and found no statistical differences in their distributions, suggesting that any such effect is small in relation to our other uncertainties.

For each of the N redshifts in a sample, the expectation value for the fraction of transmitted flux, <t(z) >, is simply the average of the 500 simulated t^j(z) values. For a single galaxy at z ≃ 3.09, the distribution of t is very broad, ranging from ≈0%–60% (see, e.g., Figure 8 of Nestor et al. 2011). The uncertainty in t(z) is thus a major uncertainty in our estimate of the contribution to _LyC from each individual source. The LBG and LAE sample-average LyC flux values, however, can be corrected with much less uncertainty. We define the sample-average transmission as

$$\begin{equation} \bar{t}_{{\rm sample}} \equiv \frac{\sum _{i=1}^N t(z_i) F_i}{\sum _{i=1}^N F_i}, \end{equation} \tag{ 2 }$$

where F_i are the N individual intrinsic (i.e., prior to any IGM absorption) NB3640 fluxes, and t(z_i) are the actual IGM transmission values. As before, N = 41 or 91 for the LBG or LAE samples, respectively. Although both the individual F_i and t(z_i) values are unknown, as the two sets are independent the expectation value for $\bar{t}_{{\rm sample}}$ is simply the average of the N values of <t(z) >, and is thus independent of the values of F_i and t(z_i). The uncertainty in $\bar{t}_{{\rm sample}}$, $\sigma _{\bar{t}}$, does depend on the t(z) and F distributions, however. While we have an accurate model for the probability distributions for the t(z_i) values, the F_i values are poorly constrained. In Nestor et al. (2011), we estimated $\sigma _{\bar{t}}$ by randomly drawing the N values of t(z_i) from the corresponding t^j(z_i) values to compute $\bar{t}_{{\rm sample}}$. We computed $\bar{t}_{{\rm sample}}$ in this manner 1000 times and equated $\sigma _{\bar{t}}$ to the standard deviation of the resulting $\bar{t}_{{\rm sample}}$ values. That procedure is equivalent to assuming the N values of F_i are all equal, and will underestimate $\sigma _{\bar{t}}$ for more realistic distributions of F.

To improve upon our estimation of $\sigma _{\bar{t}}$, we assumed an exponentially decreasing function for the intrinsic flux probability distributions: p(F)∝e^−F/β. This choice of parameterization has the advantage that the results are only mildly sensitive to the e-folding parameter, β. To determine the best choice for β, we convolved p(F) with the t(z) distributions for a range of β values. We then used the resulting attenuated-flux probability distributions to compute the likelihood of our contamination-corrected LBG and LAE NB3640 data sets, retaining the value of β that maximized these likelihoods. To determine $\sigma _{\bar{t}}$, we randomly selected the N F_i values from the maximum likelihood exponential distributions. For each flux value, we also randomly selected t(z_i) from one of the sets of t^j(z_i) values. These F_i and t(z_i) values were used with Equation (2) to determine $\bar{t}_{{\rm sample}}$. This process was repeated 1000 times, and $\sigma _{\bar{t}}$ was set equal to the standard deviation in the 1000 simulated $\bar{t}_{{\rm sample}}$ values.

We found that an exponential form for p(F), when convolved with our modeled IGM transmission values, resulted in a qualitatively acceptable match to our observed NB3640 fluxes. However, as the actual F distributions are only poorly constrained by the data, it is likely that other functional forms are also able to match the observations and may result in different estimates of $\sigma _{\bar{t}}$. In particular, if the true p(F) distributions contain a spike at zero flux (as suggested by our non-detection of NB360 flux in the stacked images of the non-detections), we would still be underestimating $\sigma _{\bar{t}}$. Nonetheless, as our exponential model represents an improvement over the past method which effectively assumed p(F) = constant, and the error budget in _LyC is not dominated by the uncertainty in $\bar{t}_{{\rm sample}}$, we adopt the $\sigma _{\bar{t}}$ values determined with an exponential p(F).

For our LBG sample, we found $\bar{t}_{{\rm sample}} = 0.298 \pm 0.040$, and for our LAE sample we found $\bar{t}_{{\rm sample}} = 0.320 \pm 0.027$. We used these values to correct the sample-average NB3640 magnitudes and NB3640−R colors. These colors can be expressed in terms of the sample-average UV-to-LyC flux-density ratios, η ≡ 〈F_UV/F_LyC〉. After applying the contamination and IGM corrections, we find η_LBG = 18.0^+34.8_−7.4 for our LBG sample, and η_LAE = 3.7^+2.5_−1.1 for our LAE sample. These values are consistent with, though larger than, those estimated by Nestor et al. (2011): η_LBG = 11.3^+10.3_−5.4 for LBGs and η_LAE = 2.2^+0.9_−0.6 for LAEs. The relative uncertainties in our updated estimates of η are larger than those in our previous estimates, due to an improved treatment of the errors. The updated samples also contribute to the increased relative uncertainties, as the newly confirmed LBGs have, on average, larger R-band photometric uncertainties than those in the previous sample, and the spectroscopically confirmed LAE sample used here is smaller than the full photometrically selected LAE sample. Nonetheless, by refining our samples and including empirical evidence for the presence or absence of foreground contamination in individual systems, our revised values represent improved estimates of η for LBGs and LAEs at z ∼ 3. We list the colors and η values in Table 4 for our raw LBG and LAE samples, for those samples after application of the contamination corrections and after correcting for both contamination and IGM absorption. The contamination- and IGM-corrected values are used below to estimate _LyC.

One of the primary goals of this paper is to determine the global luminosity space density of ionizing radiation, _LyC, contributed by star-forming galaxies at z ∼ 3. As the SSA22 protocluster represents an over-dense volume of the universe, computing _LyC directly from our data would overestimate the space density of ionizing flux. However, the UV-continuum luminosity functions of LBGs and LAEs have been measured over relatively representative volumes, at rest-frame effective wavelengths close to those of our R-band data (λ ∼ 1600 Å), by Reddy et al. (2008) and Ouchi et al. (2008), respectively. Thus, we can use the values determined for the sample-average UV-to-LyC flux-density ratios together with the corresponding luminosity functions to determine _LyC:

$$\begin{equation} \epsilon _{{\rm LyC}} = \int \frac{1}{\eta }\, L\, \Phi \, dL, \end{equation} \tag{ 3 }$$

where L refers to the non-ionizing UV continuum luminosity and η may be a function of L. When determined in this fashion, the value computed for _LyC will depend on the choices of the shape and normalization for the luminosity function, and the range in luminosity over which Equation (3) is integrated.

Reddy et al. (2008) give Schechter function parameters for the λ ∼ 1700 Å UV-continuum luminosity function of LBGs at z ∼ 3 of ϕ*_LBG = 1.66 × 10⁻³ Mpc⁻³, M*_LBG = −20.84, and α_LBG = −1.57. For LAEs at z ∼ 3.1, Ouchi et al. (2008) determine λ ∼ 1600 Å UV-continuum luminosity function parameters of ϕ*_LAE = 0.56 × 10⁻³ Mpc⁻³, M*_LAE = −19.8, and α_LAE = −1.6. However, their luminosity function is determined for LAEs having REW ≳ 64 Å, while our sample has REW ≳ 20 Å. From the REW distribution determined by Gronwall et al. (2007), we estimate that LAEs represented by our sample are a factor of ≈1.8 more common at z ∼ 3 than those described by Ouchi et al. (2008). We therefore adopt ϕ*_LAE = 1.01 × 10⁻³ Mpc⁻³ for our LAE sample.⁵

The absolute magnitude distributions for our LBG and LAE samples are shown in Figure 19. Spectroscopic confirmation of color-selected LBGs in the SSA22a field was limited to ${\cal R} \le 25.5$. Consequently, the bulk of our LBG sample is brighter than M_AB = −20. In contrast, our LAE sample, which includes galaxies down to our photometric limit of R = 27.3 (or M_AB ≃ −18.3 at z ≃ 3.09), is dominated by galaxies fainter than M_AB = −20. Thus, we first compute the contributions to _LyC from LBGs and LAEs separately, and only from sources within the UV-continuum luminosity ranges over which we determined the values of η: M_AB ⩽ −20.0 or our LBG sample, and −20 < M_AB ⩽ −18.3 for our LAE sample. The absolute magnitude limits M_AB = −20.0 and −18.3 correspond to L_min = 0.46 L*_LBG and 0.1 L*_LBG, respectively. With these magnitude ranges, the parameters for the respective luminosity functions discussed above, and η_LBG = 18.0^+34.8_−7.4 for LBGs and η_LAE = 3.7^+2.5_−1.1 for LAEs as determined in Section 5, we find ^LBG_LyC=5.2^+3.6_−3.4 × 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³ for M_AB ⩽ −20.0 and ^LAE_LyC=6.7^+3.6_−3.4 × 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³ for −20 < M_AB ⩽ −18.3.

Zoom In Zoom Out Reset image size

We now turn our attention to the total contribution to _LyC from all star-forming galaxies brighter than M_AB = −18.3 at z ∼ 3. First, however, we must address the principle source of the large (factor of ∼5) difference between the values determined for η in our LBG and LAE samples. As the LBG and LAE samples are dominated by galaxies with relatively bright and faint UV-continuum luminosities, respectively, it may be that η has a strong dependence on galaxy luminosity. In this scenario, which we refer to as our luminosity-dependent η model, we identify η_LBG with all star-forming galaxies having M_AB ⩽ −20.0 and η_LAE with all star-forming galaxies having −20 < M_AB ⩽ −18.3. Here we make the approximation that the Reddy et al. (2008) luminosity function, which is determined from color-selected galaxies, is a complete census of all star-forming galaxies with M_AB ⩽ −18.3; i.e., that there are no sources brighter than this limit that contribute to _LyC at z ∼ 3 with UV colors such that they would not be selected as LBGs. With these assumptions we can integrate Equation (3) piecewise:

$$\begin{eqnarray} \epsilon _{{\rm LyC}}^{{\rm lum.-dep.}} &=& \frac{1}{\eta _{{\rm LAE}}}\, \int _{0.1\,L^*}^{0.46\,L^*} L\, \Phi _{{\rm LBG}}\, dL \nonumber\\ &&+ \frac{1}{\eta _{{\rm LBG}}}\, \int _{0.46\,L^*}^{\infty } L\, \Phi _{{\rm LBG}}\, dL, \end{eqnarray} \tag{ 4 }$$

finding ^lum.-dep._LyC = 32.2^+12.0_−11.4 erg s⁻¹ Hz⁻¹ Mpc⁻³, with contributions from sources fainter and brighter than M_AB = −20 of 27.0^+11.4_−10.9 × 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³ and 5.2^+3.6_−3.4 × 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³, respectively. This value of ^lum.-dep._LyC is ≈60% of that estimated in Nestor et al. (2011), primarily due to our refined values for η_LBG and η_LAE.

Alternatively, the large difference in η between the LBG and LAE samples may be driven by some property of star-forming galaxies associated with LAEs. In this scenario, which we refer to as our LAE-dependent η model, we can estimate the total value of _LyC by assuming η_LAE holds for all LAEs over the full range M_AB ⩽ −18.3, and η_LBG is representative of all LBGs with M_AB ⩽ −18.3 that are not also LAEs. The fraction of LBGs that are not LAEs is determined from the LBG and LAE space densities, which can be computed by integrating the respective LBG and LAE luminosity functions, $\rho = \int _{L_{{\rm min}}}^{\infty } \Phi \, dL$.⁶ Setting L_min = 0.1 L*_LBG, we estimate that 23% of LBGs are also LAEs with REW ≳ 20 Å, which is consistent with past results at z ∼ 3 (Steidel et al. 2000; Shapley et al. 2003; Kornei et al. 2010). Here we have made the approximation that LAEs comprise a sub-sample of LBGs; i.e., that all LAEs at z ∼ 3 with M_AB ⩽ −18.3 would meet the color-selection criteria used by Reddy et al. (2008). Equation (3) then becomes

$$\begin{equation} \epsilon _{{\rm LyC}}^{{\rm LAE-dep.}} = \frac{0.77}{\eta _{{\rm LBG}}}\, \int _{0.1\,L^*}^{\infty } L\, \Phi _{{\rm LBG}}\, dL + \frac{1}{\eta _{{\rm LAE}}}\, \int _{0.1\,L^*}^{\infty } L\, \Phi _{{\rm LAE}}\, dL, \end{equation} \tag{ 5 }$$

resulting in ^LAE-dep._LyC = 16.8^+6.9_−6.5 × 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³, with contributions from non-LAE and LAE galaxies of 8.2^+5.8_−5.5 × 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³ and 8.6^+3.7_−3.5 × 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³, respectively. This value of ^LAE-dep._LyC is similar to that estimated for just LBGs with L ⩾ 0.1 L* in Nestor et al. (2011), and ≈30% of the total value of _LyC estimated by Nestor et al. (2011). This difference is due to the combination of our improved measurements of η_LBG and η_LAE and the reduced contribution from sources with low values of η in our LAE-dependent model.

It is difficult, given the present data, to distinguish which of the luminosity-dependent or LAE-dependent η models is more appropriate. In Nestor et al. (2011), we investigated differences in the average Lyα emission strengths of systems with NB3640 detections compared to those without detections by creating stacks of LRIS spectra of our LBGs, and by comparing the distributions of photometrically estimated Lyα REWs of our LAEs. For both LBGs and LAEs, we found that systems with NB3640 detections tend to have weaker Lyα emission relative to those not detected in NB3640. We repeated these tests using our refined LBG and LAE samples and excluding systems showing evidence for contamination in their spectra, and found trends consistent with our previous results. Thus, it is unlikely that Lyα emission strength has a direct influence on the average value of η. However, the LyC properties of z ∼ 3 galaxies may be dependent on some other property or combination of properties, such as star formation surface density, stellar population age, metallicity, etc., that is more typical of the LAEs in our sample compared to our LBGs, independent of UV-continuum luminosity. For example, in a sample of local LBG analogs, Heckman et al. (2011) present evidence for small covering factors for optically thick neutral gas in the most compact galaxies in their sample. Indeed, in the available HST/ACS F814W imaging, the LAEs in our sample do appear, qualitatively, more compact on average than the LBGs. Nonetheless, as our LBG and LAE samples are largely distinct in magnitude range, future work including fainter LBGs and larger LAE samples is needed to clearly differentiate between the two proposed scenarios.

We summarize the various contributions to _LyC in Table 5. The uncertainties in the values of _LyC are dominated by the uncertainties in our estimates η, which are large compared to the errors in the first moments of the luminosity functions. The uncertainties in _LyC do not, however, include systematic errors arising from our choice of luminosity functions. For example, Reddy & Steidel (2009) find a steeper faint-end slope at z ∼ 3 in previous works (see the discussion in Reddy & Steidel 2009). Furthermore, in estimating η we have only accounted for the contribution from galaxies brighter than 0.1 L*, i.e., the luminosity range over which we have empirical estimates of η. We note that, for the Reddy et al. (2008) luminosity function, galaxies fainter than 0.1 L* contribute ∼40% of the luminosity density at 1600 Å.

Table 5. Contributions to the Ionizing Backgrounds

	LFa	ηb	Magnitude Rangec	LyCd
(i)	LBG	18.0+34.8−7.4	MAB ⩽ −20.0	5.2+3.6−3.4
(ii)	LAE	3.7+2.5−1.1	−20 < MAB ⩽ −18.3	6.7+2.8−2.7
(iii)	LBG	3.7+2.5−1.1	−20 < MAB ⩽ −18.3	27.0+11.4−10.9
(iv)	LBG	18.0+34.8−7.4	MAB ⩽ −18.3	10.7+7.5−7.1
(v)	LAE	3.7+2.5−1.1	MAB ⩽ −18.3	8.6+3.7−3.5
	Total (lum.-dep.)e	...	MAB ⩽ −18.3	32.2+12.0−11.4
	Total (LAE-dep.)f	...	MAB ⩽ −18.3	16.8+6.9−6.5

Notes. ^aLuminosity function parameters used in Equations (3)–(5): LBG from Reddy et al. (2008); or LAE from Ouchi et al. (2008) scaled to include LAEs having REW ⩾20 Å. ^bSample average flux-density ratio used in Equations (3)–(5). ^cMagnitude range over which the first moment of the luminosity function is determined. M_AB = −20.0 and −18.3 correspond to 0.46 L*_LBG and 0.1 L*_LBG, respectively. ^dComoving specific emissivity of ionizing radiation in units of 10²⁴ erg s⁻¹ Hz⁻¹ Mpc⁻³. ^eTotals for our luminosity-dependent η model, determined by summing rows (i) and (iii). ^fTotal for our LAE-dependent η model, determined by summing 0.77 × row (iv) and row (v).

Download table as:  ASCIITypeset image

Given the spectral shape of the ionizing continuum flux and the mean free path to ionizing radiation in the IGM, λ_mfp, the value of _LyC implies a corresponding (proper) hydrogen photoionization rate in the IGM, $\Gamma _{{\rm H\,\scriptsize{I}}}$. If we assume a power-law form for the LyC flux density such that f_ν∝ν^−α, then

$$\begin{equation} \Gamma _{{\rm H\,\scriptsize{I}}} = \frac{(1+z)^3\,\sigma _{{\rm H\,\scriptsize{I}}}\, \lambda _{\mathrm{mfp}}}{h\,(3 + \alpha)}\, \epsilon _{{\rm LyC}}^{all,\,L\, \ge \, 0.1\,L^*}, \end{equation} \tag{ 6 }$$

where h is Planck's constant and $\sigma _{{\rm H\,\scriptsize{I}}} = 6.3 \times 10^{-18}$ cm⁻² is the atomic hydrogen photoionization cross section at the Lyman limit. Following Nestor et al. (2011), we adopt α = 3 and λ_mfp = 75.6 Mpc. For the two estimates of _LyC determined above, we find $\Gamma _{{\rm H\,\scriptsize{I}}}^{{\rm lum.-dep.}} = 2.7^{+1.0}_{-0.9} \times 10^{-12}$ s⁻¹ in our luminosity-dependent η model and $\Gamma _{{\rm H\,\scriptsize{I}}}^{{\rm LAE-dep.}} = 1.4\pm 0.6 \times 10^{-12}$ s⁻¹ in our LAE-dependent η model.⁷ We list these values of $\Gamma _{{\rm H\,\scriptsize{I}}}$ in Table 6, together with estimates of $\Gamma _{{\rm H\,\scriptsize{I}}}$ in the z ∼ 3.1 Lyα forest by Meiksin & White (2004), Bolton & Haehnelt (2007), and Faucher-Giguère et al. (2008). The values we obtain for $\Gamma _{{\rm H\,\scriptsize{I}}}$ are larger than the values reported in the literature by a factor of ≈2–4, although they are consistent given the uncertainties with the literature values at ∼2σ and ∼1σ, for the luminosity- and LAE-dependent η assumptions, respectively.

It is often of interest to estimate the fraction of LyC radiation produced by star formation that escapes into the IGM, f^LyC_esc. This value can be useful for constraining models of star formation histories, star formation feedback in the interstellar medium (ISM), etc. Additionally, f^LyC_esc can provide a useful parameterization in reionization models. The value of f^LyC_esc can be estimated empirically by

$$\begin{equation} f_{{\rm esc}}^{\mathrm{LyC}} = \frac{\eta ^{\mathrm{stars}}}{\eta }\, f_{{\rm esc}}^{\mathrm{UV}}, \end{equation} \tag{ 7 }$$

where η^stars is the intrinsic ratio of UV-to-LyC luminosity densities in star-forming regions and f^UV_esc is the escape fraction of non-ionizing UV radiation. Reddy et al. (2008) have estimated f^UV_esc ≈ 20% in LBGs at z ∼ 3. LAEs, however, exhibit bluer average UV continua and smaller UV attenuation, with f^UV_esc ≈ 30% (e.g., Ouchi et al. 2008; Kornei et al. 2010; Blanc et al. 2011). The term η^stars/η, which is often referred to as the relative escape fraction, is a measure of the attenuation of the UV flux relative to that of the LyC flux. In this work we have estimated η for samples of LBGs and LAEs. The value η^stars, however, must be obtained from spectral synthesis models, and depends on the ages of the stellar populations. Even for a given star formation history, various models produce differing values for η^stars. To illustrate the dependencies of the inferred value of f^LyC_esc on different models and parameters, we determined η^stars using the stellar population synthesis models of Bruzual & Charlot (2003, BC03), as well as a recent set of model-integrated spectral energy distributions produced by The Binary Population and Spectral Synthesis code (BPASS; Eldridge & Stanway 2009), which includes the effects of massive binary stars and nebular emission. For each of the models we adopted a constant star formation rate and two different metallicities (Z = 0.004 and 0.020) and measured η^stars, using our NB3640 and R filter passbands, at three ages (T = 10⁶, 10⁷, and 10⁸ yr). The model values we obtain for η^stars range from ≃ 1.3 for the T = 10⁶ yr, Z = 0.004 population in the BPASS models, to ≃ 6.4 for the T = 10⁸ yr, Z = 0.02 population in the BC03 models. The complete set of modeled η^stars values is given in Table 7.

For a given combination of η and f^UV_esc, the modeled values of η^stars imply a corresponding f^LyC_esc. We show these f^LyC_esc values in Table 7 for each value of η^stars, assuming η_LBG = 18.0 for LBGs and η_LAE = 3.7 for LAEs, and adopting, in turn, f^UV_esc =0.2, f^UV_esc =0.3, and the limiting dust-free case where f^UV_esc =1. The f^LyC_esc values vary by factors of ∼5–7 for a given η and f^UV_esc. They vary from 1% for LBGs in the youngest f^UV_esc =0.2 BPASS model, to ∼50% for LAEs in the oldest f^UV_esc =0.3 BC03 models, and reach greater than unity for several of the dust-free LAE models. For a given f^UV_esc and η, f^LyC_esc varies most with stellar population age, although the choice of either the BC03 or the BPASS model also affects the inferred value of f^LyC_esc by as much as a factor of ∼2. Over the range investigated, metallicity only has a small effect on f^LyC_esc. There are other model inputs that we have not varied, such as the stellar initial mass function or star formation history, that are also likely to affect η^stars and therefore the determination of f^LyC_esc. With these caveats in mind, LBGs at z ∼ 3 have a median age of ∼300 Myr (Kornei et al. 2010) with metallicities approaching ∼ solar (Pettini et al. 2001; Shapley et al. 2004; Maiolino et al. 2008) and f^UV_esc ≃ 0.2, suggesting f^{LyC, LBG}_esc ≃ 5%–7%. High-redshift LAEs, in contrast, have typical ages ∼20 Myr (Gawiser et al. 2007), low metallicity (Finkelstein et al. 2011; Nakajima et al. 2012), and f^UV_esc ≃ 0.3, suggesting f^{LyC, LAE}_esc ≃ 10%–30%. In general, we caution against the naive use of LyC escape fractions in reionization models without consideration of the values of η^stars and f^UV_esc used in their determination.

In Table 4 we also list the values of η for only galaxies with LyC detections. Our sample of LBGs with LyC detections has η = 1.7^+1.7_−0.8 ≈ η^stars for most of our modeled η^stars values, with the exception of the older BC03 models, implying an ∼ unity relative escape fraction in LBGs with LyC detections. Note that, as LyC radiation is more susceptible than non-ionizing UV radiation to attenuation by gas and dust, η ≈ η^stars implies little attenuation and therefore a large value of f^UV_esc. For the LAEs with NB3640 detections, our IGM-corrected value η = 0.4^+0.2_−0.1 is inconsistent with all of our modeled values of η^stars at more than 3σ (note that it is unphysical to have η > η^stars). We can bring our LAE detections-only value for η in line with the model predictions if we assume that we can only detect LyC in our LAE sample along fortuitously clear IGM sightlines, and therefore can neglect the IGM correction, resulting in η = 1.4^+0.7_−0.4. This scenario, however, is very unlikely.⁸ Deciphering the correct interpretation of these surprisingly low values of η is a primary goal of our ongoing work, which includes HST/UVIS LyC and UV imaging of many of our NB3640 detections. Better multi-wavelength constraints on the stellar populations of NB3640-detected LAEs will also be a key component of determining their underlying nature.

In Section 5.1 we found an LyC detection rate of ∼8% in our LBG sample, and ∼12%–15% in our LAE sample. While it is not clear if the LyC properties of a "typical" z ∼ 3 LBG or LAE are similar to that of the average system, these rates do indicate the solid angle over which LyC radiation escapes, at a level above our detection limit, averaged over all galaxies in each sample. In Nestor et al. (2011), we proposed a "blow out" model in which feedback from regions of dense star formation clears portions of the ISM of gas and dust. When viewed along favorable sightlines, such regions appear to have large escape fractions, as we find in our LBGs and LAEs with LyC detections. Galaxies that either have failed to sufficiently remove their ISM over a significant solid angle or are viewed along unfavorable sightlines, will appear to have negligible escape fractions. The strong LyC-flux upper limits in systems without NB3640 detections, derived from our stacking analysis, are consistent with this picture. Additionally, in the subsample of our LAEs having HST imaging,⁹ we find no significant difference in the distributions of sizes or surface brightnesses for sources with and without NB3640 detections. The similarity of these properties between NB3640 detected and non-detected LAEs is consistent with a scenario in which viewing angle is a significant factor in the ability to detect escaping LyC.

We show the HST/ACS imaging and one-dimensional shallow-mask spectrum of LAE010 in the top panel of Figure 7. An emission line is clearly detected at λ = 4979 Å, indicating a redshift of z = 3.096. The shallow-mask slit also covers the position of the NB3640 detection. No additional emission features are detected in the spectrum. Thus, we retain LAE010 as a possible LyC-leaking galaxy

We show the HST/ACS imaging and one-dimensional shallow-mask spectrum of LAE016 in the middle panel of Figure 7. An emission line is clearly detected at λ = 4973 Å, indicating a redshift of z = 3.091. The NB3640 detection is weak and offset by 08, but covered by the slit. As no additional emission features are detected in the spectrum, we retain LAE016 as a possible LyC-leaking galaxy

The bottom panel of Figure 7 shows the HST/ACS image of LAE018 as well as the shallow- and deep-mask spectra. An emission line at λ = 4976 Å is detected in both spectra, indicating z = 3.093. The LyA flux is centered on a compact source to the northeast, while the NB3640 flux extends over more diffuse non-ionizing UV flux to the south and west. Both slits cover the NB3640 flux. However, no other emission features were detected in either spectrum. Thus, we retain LAE018 as a possible LyC-leaking galaxy.

As we do not have HST imaging of LAE021, we show our BV image and shallow-mask spectrum in the top panel of Figure 8. An emission line is clearly detected at λ = 4948 Å, indicating a redshift of z = 3.070. No other emission features are detected and we retain LAE021 as a possible LyC-leaking galaxy. We note, however, that the NB3640 flux is offset by 14 from the LyA flux, and is not covered by the shallow-mask slit. As with all of our retained NB3640 detections, we account for the possibility the NB3640 flux is due to a foreground interloper in our Monte Carlo simulation.

We obtained spectra of LAE028 on both shallow and deep masks. The middle panel of Figure 8 shows the HST/ACS image. There is an offset of ≃ 09 between the LyA and NB3640 flux centroids, each of which is centered on an individual clump. An emission line is detected at λ ≃ 4973 Å in both spectra indicating a redshift of z = 3.088. No additional emission or absorption features are detected in either spectra. We therefore retain LAE028 as a possible LyC-leaking galaxy.

We obtained spectra of LAE038 in the deep mask as well as two of the shallow masks. The HST/ACS image, the deep-mask spectrum, and one of the shallow-mask spectra are shown in the bottom panel of Figure 8. An emission line at λ ≃ 4983 Å, indicating z = 3.099, is detected in each of the spectra. In all three of the spectra, the Lyα emission line is offset from the continuum by ≃ 05. The directions and magnitude of the offsets are consistent with the continuum emanating from the location of NB3640 flux and the Lyα emission from the LyA flux. No additional lines are detected in any of the three spectra. We therefore retain LAE038 as a possible LyC-leaking galaxy.

We show the BV image and shallow-mask spectrum of LAE039 in the top panel of Figure 9. We identify an emission line at λ = 4978 Å, indicating z = 3.095. No additional lines are detected. We therefore retain LAE039 as a possible LyC-leaking galaxy.

We obtained spectra of LAE041 in the deep mask as well as two of the shallow masks. The HST/ACS image, the deep-mask spectrum, and one of the shallow-mask spectra are shown in the middle panel of Figure 9. We identify an emission line at λ ≃ 4983 Å, indicating z = 3.066. No additional lines are detected in any of the three spectra. We therefore retain LAE041 as a possible LyC-leaking galaxy.

We show our BV image and shallow-mask spectrum of LAE046 in the bottom panel of Figure 9. An emission line is detected at λ = 4984 Å, indicating a redshift of z = 3.100. No other emission features are detected and we retain LAE046 as a possible LyC-leaking galaxy. We note, however, that the NB3640 flux is offset by 18 from the LyA flux, and is not covered by the shallow-mask slit. As with all of our retained NB3640 detections, we account for the possibility that the NB3640 flux is due to a foreground interloper in our Monte Carlo simulation.

We show the HST/ACS image and shallow-mask spectrum of LAE048 in the top panel of Figure 10. An emission line is detected at λ = 4977 Å, indicating a redshift of z = 3.094. No other emission features are detected and we retain LAE048 as a possible LyC-leaking galaxy. We note, however, that the NB3640 flux is offset by 17 from the LyA flux, and is not covered by the shallow-mask slit. As with all of our retained NB3640 detections, we account for the possibility that the NB3640 flux is due to a foreground interloper in our Monte Carlo simulation.

We show the HST/ACS imaging and shallow-mask spectrum of LAE051 in the middle panel of Figure 10. An emission line is clearly detected at λ = 4977 Å, indicating a redshift of z = 3.094. The shallow-mask slit also covers the position of the NB3640 detection. No additional emission features are detected in the spectrum. Thus, we retain LAE051 as a possible LyC-leaking galaxy.

We obtained spectra of LAE053 in one of the shallow masks and the deep mask. The bottom panel of Figure 10 shows the HST/ACS image and extracted spectra. There is an offset of ≃ 09 between the LyA and NB3640 flux centroids, each of which is centered on an individual clump. Each clump and corresponding LyA or NB3640 flux is covered by only one of the masks. A relatively broad emission line is detected at λ ≃ 4973 Å in the shallow-mask spectrum, indicating a redshift of z = 3.088. An absorption feature is detected in the deep mask at λ ≃ 4967 Å indicating Lyα in absorption at z = 3.085. This absorption feature suggests that the NB3640 source is also at z ≃ 3.09. No other emission or absorption features are detected in either spectra. We therefore retain LAE053 as a possible LyC-leaking galaxy.

We show the HST/ACS imaging and the combined shallow-mask spectrum of LAE064 in the top panel of Figure 11. An emission line is clearly detected at λ = 4994 Å, indicating a redshift of z = 3.108. Both slits also cover the position of the NB3640 detection, which is offset by 10 from the LyA flux. The extraction apertures for the spectra shown were centered on the Lyα line, which was offset from faint continuum emission consistent in magnitude (∼1'') and direction from the position of the NB3640 flux. No additional emission features are detected in either (one- or two-dimensional) spectra. Thus, we retain LAE064 as a possible LyC-leaking galaxy.

We show our BV image and shallow-mask spectrum of LAE069 in the middle panel of Figure 11. An emission line is detected at λ = 4953 Å, indicating a redshift of z = 3.074. No other emission features are detected and we retain LAE069 as a possible LyC-leaking galaxy. We note, however, that the NB3640 flux is offset from the LyA flux by 09, and is not covered by the shallow-mask slit.

The bottom panel of Figure 11 shows the HST/ACS image of LAE074 as well as the shallow- and deep-mask spectra. An emission line at λ = 4990 Å is detected in the shallow-mask spectrum, indicating z = 3.093. While the Lyα emission line is not detected in the deep-mask spectrum, the slit covers only part of the diffuse LyA flux. No other emission features were detected in either spectrum. Thus, we retain LAE074 as a possible LyC-leaking galaxy.

The top of Figure 12 shows the HST/ACS image of LAE081 as well as the shallow- and deep-mask spectra. An emission line at λ = 4989 Å is detected in both spectra, indicating z = 3.104. Both slits cover the bulk of the LyA and NB3640 flux. No other emission features were detected in either spectrum. Thus, we retain LAE081 as a possible LyC-leaking galaxy.

The bottom of Figure 12 shows the HST/ACS image of LAE083 as well as the shallow-mask spectrum. An emission line at λ = 4942 Å is detected, indicating z = 3.065. The slit also covers the bulk of the NB3640 flux. No other emission features were detected. Thus, we retain LAE083 as a possible LyC-leaking galaxy.