A DEEP HUBBLE SPACE TELESCOPE AND KECK SEARCH FOR DEFINITIVE IDENTIFICATION OF LYMAN CONTINUUM EMITTERS AT cic>∼3.1^{* †}

Surveys are now identifying large numbers of galaxies at the reionization epoch ($z\sim 7$–8). Consequently, there has been much interest in how those galaxies reionized the hydrogen in the intergalactic medium (IGM). Several studies have determined that the number density of active galactic nuclei (AGNs) is not high enough to provide the required ionizing background (Inoue et al. 2006; Siana et al. 2008; Willott et al. 2010; Masters et al. 2012), though some have suggested that faint AGNs may contribute significantly (Glikman et al. 2011; Fontanot et al. 2012). Because massive stars are the only other significant source of ionizing photons, they are believed to be responsible for the reionization of intergalactic hydrogen.

Whether or not stars reionized the IGM and provided the ionizing background for 1–2 Gyr thereafter is contingent upon many uncertain parameters: the total star formation rate density, including the significant contribution from sources beyond our current detection limits; the intrinsic ionizing spectrum of the galaxies; the fraction of the ionizing radiation produced by these galaxies that escapes into the IGM; and the "clumping" of the hydrogen being ionized in the IGM at any epoch (and thus the recombination rate).

Many efforts, both theoretical and observational, are being made to determine the amount of undetected star formation (Alavi et al. 2014) and the clumping factor (Pawlik et al. 2009; Finlator et al. 2012). However, little is understood about the "escape fraction," f_esc, of ionizing radiation. Young stars form in dense molecular clouds. In order for a significant fraction of ionizing radiation to escape into the IGM, lines of sight (LOSs) with low neutral hydrogen column density must be cleared out or the stars must migrate out of the gas. Either process must occur within the short lifetime ($\lt 10$ Myr) of the O stars that produce the ionizing radiation.

It is not understood how the stars can be exposed on such short timescales, though several mechanisms have been proposed. These include supernova (SN) winds and their effects on the H i distribution in both the interstellar medium and the circumgalactic medium (CGM) (Dove et al. 2000; Clarke & Oey 2002; Fujita et al. 2003; Razoumov & Sommer-Larsen 2006), interactions/mergers with other galaxies (Gnedin et al. 2008), and runaway massive stars (Conroy & Kratter 2012). Each of these proposed mechanisms has observational signatures that would allow us to test which processes are responsible for high escape fractions and, thus, the reionization of the universe.

Regardless of the precise mechanism, the first goal is to measure the average escape fraction and any dependencies it might have on galaxy properties or environment. Over the past 15 years, many attempts have been made to directly detect escaping Lyman continuum (LyC) radiation from galaxies of varying types and at various redshifts (Leitherer et al. 1995; Deharveng et al. 2001; Steidel et al. 2001; Giallongo et al. 2002; Malkan et al. 2003; Grimes et al. 2007, 2009; Siana et al. 2007, 2010; Cowie et al. 2009; Iwata et al. 2009; Bridge et al. 2010; Vanzella et al. 2010b, 2012; Boutsia et al. 2011; Leitet et al. 2011, 2013; Nestor et al. 2011, 2013; Mostardi et al. 2013; Borthakur et al. 2014). These studies must be conducted at $z\lesssim 3.5$ because the IGM becomes optically thick to ionizing photons at higher redshift (Inoue & Iwata 2008; Prochaska et al. 2009). Also, galaxies in the nearby universe must be receding at sufficient velocities such that a significant fraction of the ionizing spectrum is redshifted to wavelengths greater than 912 Å to avoid absorption by Galactic H i (Hurwitz et al. 1997). The most sensitive limits have been obtained at $z\sim 1$ with ultraviolet imaging or spectroscopy with the Hubble Space Telescope (HST) and GALEX (Malkan et al. 2003; Siana et al. 2007; Cowie et al. 2009; Bridge et al. 2010; Siana et al. 2010) and at $z\sim 3$ with optical data from the ground (Shapley et al. 2006; Iwata et al. 2009; Nestor et al. 2011, 2013; Mostardi et al. 2013). The results of the studies at $z\sim 1$ and $z\sim 3$ are quite different. At $z\sim 1$, there are no detections from a sample of 68 galaxies (Siana et al. 2007, 2010; Bridge et al. 2010) and a null result from a stack of 626 galaxies (Cowie et al. 2009). At $z\sim 3$ however, roughly 10% of galaxies have been detected with large ionizing emissivities (Iwata et al. 2009; Nestor et al. 2011, 2013). Because the limits at $z\sim 1$ are typically far deeper than the surveys at $z\sim 3$, the large number of detections at $z\sim 3$ suggests an average escape fraction that is significantly higher at $z\sim 3$ (Siana et al. 2010).

There are several possible explanations for the seemingly discrepant results at $z\sim 1$ and $z\sim 3$. First, the two samples were selected in different ways. In particular, most of the galaxies in the $z\sim 3$ samples were selected as Lyα emitters. Because both the Lyα and LyC escape are dependent on the distribution of neutral hydrogen, it is certainly possible that an Lyα selection would also tend to find LyC emitters. However, Nestor et al. (2011) saw lower Lyα fluxes for galaxies with LyC detections, suggesting that Lyα emission is not necessarily positively correlated with high LyC escape fractions. Second, the role of environment may be important. The large samples at $z\sim 3$ have been selected within large overdensities to increase observing efficiency, but the increased star formation in these dense environments might affect the local ionizing background and the neutral hydrogen columns in the CGM. Finally, the $z\sim 1$ and $z\sim 3$ samples may be in different phases of their evolution. The samples have similar star formation rates and stellar masses (Siana et al. 2010). However, because the $z\sim 1$ samples were selected to be very luminous, they lie well above the star-forming main sequence at $z\sim 1.3$ (Whitaker et al. 2012) and are likely in a transient starburst phase. It is possible that the beginning stages of such a starburst may not be conducive to high LyC escape fractions (for example, because gas fractions are particularly high or because many are merging systems). Finally, it may be possible that other galaxy properties that have not been measured well—their sizes or CGM, for example—can influence the LyC escape fraction.

Many uncertainties remain in understanding ionizing photon escape and the differences in the $z\sim 1$ and $z\sim 3$ studies. Foremost among these is the possibility of foreground contamination of the $z\sim 3$ LyC detected galaxies. In some cases, it is possible that a faint star-forming galaxy lies along the LOS to the target galaxy, but at significantly lower redshift. In this case a candidate detection in the rest-frame LyC of the background galaxy could simply be non-ionizing emission from the foreground galaxy. At $z\sim 1$, the LOS is much smaller and the flux densities required to probe significant escape fractions are not as low. Therefore, there are far fewer foreground objects along the LOS. In addition, the $z\sim 1$ studies have been done with the HST and its high spatial resolution allows for easier identification of foreground objects at small angular separation (or impact parameter).

Some studies have used deep U-band (which probes below the Lyman limit above $z\gtrsim 3$) number counts to determine the probability of foreground contamination at $z\sim 3$ (Siana et al. 2007; Vanzella et al. 2010b; Nestor et al. 2011; Mostardi et al. 2013) and find that a significant fraction of LyC detections are in fact attributed to foreground contamination. This contamination fraction is a strong function of the depth of the LyC images (since the number density of foreground objects rises steeply at fainter magnitudes; e.g., Alavi et al. 2014), as well as the FWHM of the point-spread function (PSF) (as the PSF widens, contaminants at larger impact parameters cannot be recognized).

These statistical corrections for foreground contamination are useful when determining what fraction of galaxies have detectable LyC escape, or for constraining the total ionizing emissivity of galaxies. However, it is useful to know for certain which galaxies truly have high escape fractions, as they need to be studied in greater detail to understand the mechanisms for escape.

There are two ways to mitigate the foreground contamination problem. First, deep spectroscopy of candidate LyC emitters may provide evidence of the foreground object if emission lines are detected at other redshifts. However, the foreground object may lack strong emission lines or the lines may be shifted out of the wavelength range covered by the spectrum. Other evidence for a foreground galaxy may be seen in absorption. Any foreground contaminant would lie at a very small impact parameter ($\lt 10\;{\rm kpc}$), and Steidel et al. (2010) have shown that galaxies generally exhibit strong metal-line absorption at these distances. A high signal-to-noise ratio (S/N) detection of the continuum of the background galaxy is required to identify foreground absorption and is very difficult to obtain for galaxies fainter than R = 24.5.

Another way to minimize the frequency of foreground contamination is to increase the spatial resolution of the LyC imaging, as the contamination rate is proportional to the area of the seeing disk. If one resolves the background galaxies (with half-light radii of ∼02) instead of spreading the light over the typical seeing disk size (06–10), the contamination rate goes down by a factor of ∼9–25 and becomes almost negligible (Vanzella et al. 2012).

In addition to simply removing the possibility of foreground contamination, higher-resolution images may allow us to resolve the LyC-emitting regions of galaxies. If SN winds clear low column density sight lines, the LyC may escape primarily from the regions with highest star formation surface density. However, if galaxy interactions or runaway stars are the primary mechanism, the LyC emission will be less concentrated than the non-ionizing UV continuum.

We have set out to determine definitively whether candidate LyC emitters have foreground contamination or indeed have high ionizing emissivities and, if the latter, which mechanisms allow high escape fractions. To do this, we have followed up some of the brightest LyC-emitting candidates selected from ground-based observations by Nestor et al. (2011) in SSA22, a field with a large overdensity of galaxies at z = 3.09 (Steidel et al. 1998). Because of the large number of galaxies at the same redshift, this field has been used to efficiently find candidate LyC emitters with a narrow band filter just below the observed wavelength of the Lyman limit (Iwata et al. 2009; Nestor et al. 2011, 2013).

First, we obtained deep HST imaging of the rest-frame LyC of these galaxies to determine from which galaxy (or regions of the galaxy) the candidate LyC is escaping. Second, we obtained deep Keck near-IR spectroscopy to detect the rest-frame optical emission lines (primarily [O iii] λ5007) at high spatial resolution (∼05) from the regions emitting apparent LyC emission to verify the redshifts of the sources. The HST imaging allows us to identify possible contaminating faint foreground galaxies, and the Keck spectroscopy allows us to identify the redshifts of all candidate LyC emitters. Together, these observations allow us to detect foreground contaminants and, therefore, identify the true LyC emitters.

In Section 2, we discuss the HST observations. In Section 3, we discuss the Keck near-IR spectroscopic observations. In Section 4, we discuss each of the galaxies individually. Finally, in Section 5, we put the findings in context with other recent work.

2.1. Target Selection

Our target galaxies are all bright, $L\gtrsim 0.5{{L}^{*}}$ LBGs at $z\gt 3.06$ in the SSA22 field and were originally cataloged in Steidel et al. (2003). The LyC region of these galaxies at $z\gt 3.06$ was imaged with a blue narrowband filter with the Low Resolution Imaging Spectrograph (LRIS) on Keck I (NB 3640; see Figure 1 for the narrowband filter curves) by Nestor et al. (2011). The galaxies were selected by their Lyα emission at $z\sim 3.09$ with a narrowband filter (NB 4980) or via a Lyman break selection using broadband filters (Steidel et al. 2003). The HST WFC3/UVIS field of view ($2\buildrel{\,\prime}\over{.} 7\times 2\buildrel{\,\prime}\over{.} 7$) is considerably smaller than that of Keck/LRIS ($5\buildrel{\,\prime}\over{.} 5\times 7\buildrel{\,\prime}\over{.} 6$), and a very deep exposure is required to detect the rest-frame LyC of these galaxies with HST. Therefore, we chose to obtain a single, deep HST LyC image near the NW corner of the LRIS image that has several bright candidate LyC emitters. The bright galaxies lying in the single WFC3/UVIS pointing are MD32, C49, MD46, and D17. There is an additional candidate, aug96M16, that does not lie in this HST field, but we obtained Keck/NIRSPEC spectroscopy of that source as well (see Section 3).

Zoom In Zoom Out Reset image size

2.2. Hubble Observations

2.3. Data Reduction and Analysis

The HST data were reduced using multidrizzle, the standard PyRAF routines provided by the Space Telescope Science Institute. The flattened frames were cleaned of cosmic rays and stacked onto images with matching WCS and aligned within a fraction of a pixel. Because sub-pixel dither patterns were used, we sample the PSF on a finer scale than the native pixel scales of 004 for WFC3/UVIS, 005 for ACS/WFC, and 013 for WFC3/IR. Therefore, we drizzled the data onto images with pixel sizes of 002, 004, and 008 for WFC/UVIS, ACS/WFC, and WFC3/IR, respectively. Using stars in the field, we measure PSF FWHMs of 007, 009, and 018 in the WFC3/UVIS, ACS/WFC, and WFC/IR channels.

The output variance maps from multidrizzle were used to determine the depths of the images. These depths agree well with the pixel-to-pixel variations in the images once they are corrected for correlated noise via the methods in Casertano et al. (2000). The $3\sigma$ depths are listed in Table 1.

We combined the F110W and F160W images (weighting by the inverse variance in each image pixel) to produce a deep detection image. We ran SourceExtractor (Bertin & Arnouts 1996) to identify sources above the background fluctuations. The isophotes from the deep detection image were then mapped onto the finer pixel scales of the WFC3/UVIS and ACS/WFC images. Because the images were drizzled to pixel scales of 002, 004, and 008 (in WFC3/UVIS, ACS/WFC, and WFC3/IR, respectively), each pixel in the WFC3/IR image has either four ACS/WFC pixels or 16 WFC3/UVIS pixels in it. Therefore, the isophotes from the detection image can be mapped directly to the finer pixel images. Finally, we used our own code to determine isophotal fluxes in the same isophotes in all filters. The images of the targets are displayed in Figure 2, and fluxes are listed in Table 2.

Zoom In Zoom Out Reset image size

Table 2.  Galaxy Properties

Galaxy	zspec	exp($-{{\tau }_{{\rm IGM}}}$)a	$f_{{\rm esc},{\rm rel}}^{{\rm LyC}}$ b c	F336Wb d	F606Wd	F814Wd	F110Wd	F160Wd
C49	3.153	0.158	$\lt 0.088$	$\lt 0.037$	5.74 ± 0.16	6.37 ± 0.10	6.69 ± 0.14	8.13 ± 0.24
C49_NW	2.029	...	...	0.40 ± 0.026	1.73 ± 0.19	1.28 ± 0.07	1.89 ± 0.10	3.28 ± 0.17
MD46	3.091	0.0183	$\lt 0.066$	$\lt 0.060$	9.31 ± 0.25	11.9 ± 0.16	11.1 ± 0.18	14.4 ± 0.33
MD46_NW	...	...	3.62e	0.40 ± 0.022	0.87 ± 0.09	1.45 ± 0.06	1.47 ± 0.10	1.23 ± 0.18
MD32	3.102	0.178	$\lt 0.451$	$\lt 0.025$	0.41 ± 0.11	0.75 ± 0.07	0.78 ± 0.07	1.43 ± 0.12
MD32_NE	2.885? 1.964?	...	...	0.56 ± 0.038	1.74 ± 0.16	2.22 ± 0.10	3.28 ± 0.11	5.41 ± 0.20
aug96M16_E	3.095	...	...	...	...	2.73 ± 0.17	...	6.31 ± 0.45
aug96M16	...	...	...	...	...	3.87 ± 0.24	...	9.04 ± 0.49
aug96M16_W	3.291	...	...	...	...	3.22 ± 0.14	...	4.06 ± 0.32
D17	3.069	0.194	$\lt 0.081$	$\lt 0.031$	4.10 ± 0.14	4.78 ± 0.20	4.61 ± 0.13	6.37 ± 0.22

Notes.

^aDefined in Equation (2). Here we use the average transmission of 500 simulated sight lines. ^b"<" denotes $1\sigma$ limits. ^cDefined in Equation (1). ^dFlux density in units of 10⁻³⁰ erg s⁻¹ cm⁻² Hz⁻¹. ^eFor this calculation, MD46_NW was assumed to have the same redshift as MD46, z = 3.091, and the same value of transmission through the IGM. An ${{f}_{{\rm esc},{\rm rel}}}\gt 1$ means that the measured flux is larger than assumed. This could be caused by a very recent burst in star formation (Domínguez et al. 2014). The number may be artificially high if LOS transmission is higher than average.

Download table as:  ASCIITypeset image

In addition to the fluxes, the estimated relative escape fractions (or the $1\sigma$ limits) are listed as well. The relative escape fraction, $f_{{\rm esc},{\rm rel}}^{{\rm LyC}}$ is defined as in Siana et al. (2007):

$$\begin{eqnarray}&&f_{{\rm esc},{\rm rel}}^{{\rm LyC}}=\frac{{{({{f}_{1500}}/{{f}_{{\rm LyC}}})}_{{\rm stel}}}}{{{({{f}_{1500}}/{{f}_{{\rm LyC}}})}_{{\rm obs}}}}\ {\rm exp} ({{\tau }_{{\rm HI},{\rm IGM}}}),\end{eqnarray} \tag{ 1 }$$

where f₁₅₀₀ and f_LyC are the flux densities, in ${{f}_{\nu }}$, at 1500 Å (F814W fluxes) and in the LyC (F336W fluxes, ∼820 Å for these observations), respectively. The LyC flux density is assumed to be flat (constant) in ${{f}_{\nu }}$ before IGM attenuation, and the intrinsic amplitude of the Lyman break in the stellar SED is assumed to be ${{({{f}_{1500}}/{{f}_{{\rm LyC}}})}_{{\rm stel}}}=6.0$ (Siana et al. 2007). We note that this ratio can be affected by short-term variations in the star formation history. Specifically, after a rise in the star formation rate, the LyC will be significantly enhanced compared to the 1500 Å continuum. Conversely, shortly after a downturn in the star formation rate, the LyC flux can quickly decrease, whereas the 1500 Å flux will be largely unaffected. Using numerical simulations of galaxies with realistic feedback prescriptions that broadly reproduce the spread in the star-forming main sequence at these epochs (Whitaker et al. 2012), Domínguez et al. (2014) show that galaxies with ${{M}^{*}}\gt {{10}^{9}}\;{{M}_{\odot }}$ have a ∼0.09 dex (or ∼23%) spread in the ${{({{f}_{1500}}/{{f}_{{\rm LyC}}})}_{{\rm stel}}}$ ratio.

In addition, the ${{({{f}_{1500}}/{{f}_{{\rm LyC}}})}_{{\rm stel}}}$ ratio may be significantly lower (by a factor of ∼2) when including the effects of the binarity and rotation of massive stars, as the number of WR stars is increased and the main-sequence lifetime and effective temperature are increased (Eldridge & Stanway 2009).

The IGM transmission, ${\rm exp} (-{{\tau }_{{\rm IGM},{\rm HI}}})$, is integrated along the entire F336W filter bandpass,

$$\begin{eqnarray}&&{\rm exp} (-{{\tau }_{{\rm IGM},{\rm HI}}})=\frac{\int {\rm exp} (-{{\tau }_{\lambda }})\ {{T}_{\lambda }}d\lambda }{\int {{T}_{\lambda }}d\lambda },\end{eqnarray} \tag{ 2 }$$

where the IGM opacity, τ, and the filter transmission, T, are both functions of wavelength. Using a similar technique as in Nestor et al. (2013) to simulate the IGM LOSs, the value for the average IGM transmission through the F336W filter curve at z = 3.09 was estimated to be 0.184, though the value can vary wildly among sight lines (16th and 84th percentile values are 0.011 and 0.365, respectively). The most opaque sight lines can be ruled out if the NB 3640 detections are real. The IGM transmission varies slowly with redshift. In Table 2 we list the average transmission value at each redshift and use that value to compute $f_{{\rm esc},{\rm rel}}^{{\rm LyC}}.$

Near-IR spectra were obtained for four of the targets using the NIRSPEC instrument (McLean et al. 1998) on the Keck II telescope. We observed with NIRSPEC on 2011 August 18 and 19. Conditions were photometric, with 05 seeing in the K band. The exposure times were 6 × 900 s for C49 and MD32, 8 × 900 s for MD46, and 4 × 900 s for aug96M16. All targets were observed with a 0 76 × 42'' long slit. As explained in Section 4, many of the targets either are elongated or have several detected clumps (or separate galaxies) within a 1 2 radius. The position angle of the NIRSPEC slit (listed in Table 3) was chosen in the direction of elongation or to collect spectra of multiple objects. Since the primary spectral features of interest were [O iii] λ5007 and Hβ at $z\sim 3.06$–3.30, we used the N6 filter, a broad $H+K$ filter centered at 1.925 μm with a bandwidth of 0.75 μm. The spectral resolution of our NIRSPEC observations was ∼15 Å, as determined from sky lines measured in the N6 filter. This resolution corresponds to ${\Delta }\upsilon \;\sim$ 230 km s⁻¹ (R ∼ 1300). Due to the faint nature of these objects in the $K$ band, we acquired each target using blind offsets from a bright star in the surrounding field. We returned to the offset star between each integration of the science target to recenter and dither along the slit. In all cases we dithered back and forth between two positions near the center of the slit.

Table 3.  NIRSPEC Measurements

Object	Slit PAa	zemb	Hα Fluxc	Hα FWHMd	[O iii] Fluxc	[O iii] FWHMd	${{\sigma }_{v}}([{\rm O}\;{\rm III}])$ e	Hβ Fluxc
C49f	−38	3.1532	...	...	7.5 ± 0.25	17 ± 0.7	54.7$_{-9}^{+8}$	<1.4
C49_NW	−38	2.0286	1.4 ± 0.16	14 ± 2.6	...	...	...	...
MD46	−34	3.0857	...	...	5.6 ± 0.15	16 ± 0.5	27.1$_{-15}^{+10}$	1.4 ± 0.15
MD46_NWf	−34	...	...	...	<0.40	...	...	...
MD32f	37	3.0968	...	...	1.2 ± 0.14	14 ± 1.4	...	<0.36
MD32_NE	37	2.8844g	...	...	2.3 ± 0.19	15 ± 1.7	...	...
aug96m16_Wf	90	3.2899	...	...	3.0 ± 0.24	15 ± 1.8	<50.48	<0.82
aug96m16_E	90	3.0938	...	...	2.9 ± 0.34	18 ± 2.1	60.2$_{-31}^{+22}$	0.8 ± 0.25
aug96m16f	90	...	...	...	<1.2	...	...	...

Notes.

^aSlit position angle in degrees east of north. ^bRest-frame optical nebular emission-line redshift measured from [O iii] λ5007 or Hα. ^cEmission-line flux and error in units of 10⁻¹⁷ ergs s⁻¹ cm⁻². ^dObserved FWHM and error in units of Å. The FWHM is a raw value, uncorrected for instrumental broadening. ^eNebular velocity dispersion measured in units of km s⁻¹. ^fUpper limits for [O iii] and Hβ fluxes are quoted at the 3σ level. For aug96M16 and MD46_NW, limits were calculated by assuming the same redshift, respectively, as aug96M16_E (z = 3.0938) and MD46 (z = 3.0857). ^gRedshift estimated assuming that the single emission line detected corresponds to redshifted [O iii] $\lambda 5007$. If the emission line is instead redshifted Hα, the redshift MD32_NE is z = 1.9635.

Download table as:  ASCIITypeset image

3.1. NIRSPEC Data Reduction

Data reduction was performed using a method similar to that described in Liu et al. (2008), where the sky background was subtracted from the two-dimensional unrectified science images using an optimal method (Kelson 2003; G. D. Becker 2006, private communication). The sky lines were fit with a low-order polynomial, and a b-spline fit was used in the dispersion direction. After background subtraction, cosmic rays and bad pixels were removed from each exposure. The individual images were then rotated, cut out along the slit, and rectified in two dimensions to take out the curvature in both the wavelength and spatial directions. The final rectified two-dimensional exposures for each object were then registered and combined into one spectrum.

Each of our targets had previously been spectroscopically confirmed as a $z\sim 3$ galaxy based on optical (i.e., rest-frame UV) spectroscopy. Our goal was to determine if the specific regions associated with apparent LyC emission were also at $z\gt 3.06$, high enough redshift that the NB 3640 and F336W filters provided a clean probe of the LyC spectral region. Accordingly, in addition to confirming the redshift for the main target, we also searched the NIRSPEC slit for additional emission-line features offset from the main region of non-ionizing UV flux, and perhaps originating from the same regions as the NB 3640 or F336W flux.

In the spectra of C49, aug96M16, and MD32, we discovered evidence for an additional emission-line spectrum coincident with or very near the location of NB 3640/F336W emission (referred to, respectively, as C49_NW, aug96M16_E & aug96M16_W, and MD32_NE), while MD46 yielded only a single detected spectrum centered on the location of the non-ionizing UV continuum.

One-dimensional spectra were extracted from the two-dimensional reduced image at the location of any detected emission lines. The corresponding 1σ error spectra were also extracted. The average aperture size along the slit was 13, with a range of 11–20. Each one-dimensional spectrum was then flux-calibrated using A-type stars according to the method described in Shapley et al. (2005) and Erb et al. (2003) and placed in a vacuum, heliocentric frame. The final flux-calibrated, one-dimensional spectra are plotted in Figure 3.

Zoom In Zoom Out Reset image size

3.2. NIRSPEC Measurements

Because there are only five galaxies in our sample and each has different properties, we have chosen to discuss the objects individually. We first discuss the HST imaging, as the resolved images informed the strategy for the near-IR spectroscopy (slit orientation and required seeing for each galaxy). In all cases one must keep in mind that the Keck narrowband filter (NB 3640) used to identify the LyC emission has a width corresponding to ∼25 physical Mpc, significantly smaller than the mean free path of a photon at the Lyman limit at $z\sim 3.09$ (∼100 Mpc; Prochaska et al. 2009; O'Meara et al. 2013). The HST F336W filter, which we have used to detect the LyC of these galaxies, has a similar red cutoff (near the Lyman limit) but is a factor of ∼5.5 times wider, with the additional transmission blueward of the NB 3640 filter (Figure 1). At the bluer wavelengths probed with the F336W filter, but not probed by the NB 3640 filter, the average LOS through the IGM toward a z = 3.09 galaxy will have significantly less transmission. Thus, the typical galaxy at z = 3.09 should appear significantly fainter (1.0 mag) in the F336W filter than in the Keck NB 3640 filter (see Section 5 for more details).

4.1. C49

The galaxy C49 was first identified as an LyC emitter in deep rest-frame ultraviolet spectroscopy from Keck (Shapley et al. 2006), and again with narrowband imaging just below the Lyman limit (Iwata et al. 2009; Nestor et al. 2011). At ground-based resolution (0 8), there was already a suggestion in Figure 3 of Nestor et al. (2011) that the LyC flux was only emitted from the NW corner of the non-ionizing UV-emitting area of C49.

The HST imaging resolves C49 into two sources separated by 0 65 (5.0 proper kpc at z = 3.153) in all filters. The fact that the two clumps are distinct at rest-frame optical wavelengths suggests that these are two separate galaxies rather than two UV-bright star-forming regions of the same galaxy. Because the galaxy to the SE is brighter and has a previously known redshift (z = 3.153), we are referring to it with the original C49 label and the nearby galaxy as C49_NW. Shapley et al. (2006) had previously obtained a deep optical spectrum (rest-frame UV) of this system, but the two sources were unresolved and most of the light from both objects fell within the 1 2 wide slit. Because C49 is five times brighter than C49_NW in the optical, it is the galaxy with the previously reported redshift.

The F336W (LyC) image shows strong emission from C49_NW, with no detectable emission from C49. We derive a $1\sigma$ limit of ${{f}_{{\rm esc},{\rm rel}}}\lt 0.088$ ($1\sigma$) for C49. If C49_NW is at the same redshift as C49, or at any other redshift above z = 3.06, then the F336W flux would be a direct detection of LyC emission.

The NIRSPEC slit was oriented to lie along both objects (${\rm PA}=-38{}^\circ$). The seeing was sufficient (0 5) to marginally resolve [O iii] lines from both components if they were both at the same redshift. The 2D spectrum is shown in Figure 4. The strong [O iii] $\lambda \lambda 4959,5007$ emission lines are clearly detected at the redshift of C49. There is no indication of fainter [O iii] emission at the same redshift from the companion. Instead, an emission line is detected at shorter wavelength ($\lambda =1.988$ μm) ∼0 5 from C49 in the direction of C49_NW. The two most likely identifications of this line are [O iii] λ5007 or Hα, indicating redshifts of z = 2.9698 or z = 2.0286, respectively.

Zoom In Zoom Out Reset image size

Given that C49_NW is likely at one of these two redshifts, we re-examined the deep LRIS spectrum of C49 (Shapley et al. 2006) to look for absorption features from a foreground source. As mentioned above, the impact parameter is ∼5 kpc, and strong absorption from both the interstellar medium and CGM of C49_NW would be expected (Steidel et al. 2010). In Figure 5, we plot the deep Keck/LRIS spectrum. There are several strong absorption features associated with z = 2.0286 in the Lyα forest (Si ii $\lambda 1260$, O i/Si ii $\lambda \lambda 1302,1304$, C ii $\lambda 1334$, Si ii $\lambda 1526$) and at the red end of the spectrum (Fe ii $\lambda \lambda 2343,2374,2382$). Therefore, we conclude that the emission line in the NIRSPEC spectrum is Hα and the flux detected at $\lambda \lesssim 3780$ Å in both Shapley et al. (2006) and Nestor et al. (2011) is not LyC emission. The foreground contamination was not originally recognized with the LRIS spectrum alone because all of the strong UV absorption features are in the Lyα forest of C49 (except for the faint Fe ii lines at the red end of the spectrum where the sensitivity is poor). The additional detection of an emission line in the rest-frame optical allowed us to identify the absorption features as associated with C49_NW and not with foreground Lyα lines.

Zoom In Zoom Out Reset image size

4.2. MD46

4.3. MD32

4.4. aug96M16

4.5. D17

There have been several surveys for LyC emitters at this redshift, some detecting few or no LyC emitters (Boutsia et al. 2011; Vanzella et al. 2012), some claiming dozens of candidates (Iwata et al. 2009; Nestor et al. 2011, 2013; Mostardi et al. 2013). It can be confusing to synthesize these results into a unifying picture. As we have seen in this paper, it is critical to carefully consider contamination and its effects on each survey when trying to compare results.

The rate of contamination from foreground galaxies is quite low at these wavelengths. As an example, the odds of a foreground galaxy with ${{U}_{{\rm AB}}}\lt 27.5$ lying within 10 of a target galaxy at $z\sim 3$ is roughly 8.5% given the completeness-corrected surface density tabulated in Vanzella et al. (2010b) and taken from Nonino et al. (2009). But detection of LyC emitters is also rare. Obviously, if the detection rate is near the contamination rate, then most or all of the candidate LyC emitters will be spurious. It is therefore critical that any survey for LyC-emitting galaxies accurately calculates the contamination rate and has a significantly higher detection rate if it is to claim a bona fide LyC detection. We list here several factors that can affect the contamination rates and detection rates in these surveys. First, the factors affecting the contamination rate, many of which have been outlined before by other authors (Siana et al. 2007; Vanzella et al. 2010b):

• 1.  
  Depth: With increasing depth, fainter and more numerous foreground galaxies (e.g., Alavi et al. 2014) can be detected.
• 2.  
  Spatial resolution: At lower resolution, foreground galaxies at larger angular separation from the target galaxy can contaminate the photometry.
• 3.  
  Redshift: At higher redshift, because the line-of-sight to the target galaxy is much larger, the volume of the contaminating cylinder is larger and the number of contaminants will therefore be larger. This is part of the reason that contamination concerns are much smaller for the $z\sim 1$ samples of Siana et al. (2007, 2010) and Bridge et al. (2010).

Generally, one can observe the surface density of objects in the field at the wavelengths of the LyC observations and determine the rate at which an interloper would lie in the seeing disk. The example above, because it is very deep and uses a fairly large, 1'' radius aperture, results in an 8.5% contamination rate. Of course, one also has to consider the flux distribution of expected interlopers as well.

Below are the factors affecting the detection rate of galaxies:

• 1.  
  Depth: Deeper data will be sensitive to lower escape fractions, increasing the detection rate. Deeper data are also sensitive to escaping LyC from less luminous galaxies.
• 2.  
  Redshift and IGM opacity: This factor is often overlooked. When looking at rest-frame wavelengths near the Lyman limit (e.g., ${{\lambda }_{{\rm rest}}}=850-910$ Å) at $z\sim 3$, the IGM is relatively transparent. At shorter rest-frame wavelengths, the IGM transmission is significantly lower. The reason for this difference is that the mean-free-path of an LyC photon at $z\sim 3$ is ${{l}_{{\rm mfp}}}\sim 100$ Mpc (Prochaska et al. 2009). As a photon travels that distance, it redshifts by about 10%. Therefore, a photon with ${{\lambda }_{{\rm rest}}}=829$Å at $z\sim 3$ has to travel 1l_mfp to redshift beyond the Lyman limit so that it is no longer subject to bound-free absorption from hydrogen. A photon of ${{\lambda }_{{\rm rest}}}=760$ Å has to travel 2l_mfp. (Note, we are ignoring the much smaller effects of decreasing cross section with decreasing wavelength and the changing mean free path with redshift to illustrate the simple point.) In summary, surveys that probe the LyC of a large number of galaxies at an optimal redshift for the filter (e.g., Nestor et al. 2011, at $z\sim 3.09$) will likely have a higher detection rate than a survey of field galaxies at a variety of redshifts. Of course, if the observations are done spectroscopically, then one can analyze many galaxies at a variety of redshifts with good sensitivity near the Lyman limit.
• 3.  
  Filter width: Because the mean free path of an LyC photon in the IGM at z ∼ 3 corresponds to observed 330 Å (see above), the filter width can affect the level of signal detected. Filters significantly smaller than the mean free path (and near the Lyman limit) will probe regions of the LyC with high transmission. Broadband filters, however, will probe wavelengths with little or no transmission at the blue end of the filter, as the typical filter (e.g., the Johnson U filter) is twice the width of the mean free path. In fact, a wide filter will often be adding noise at the short-wavelength end, but no signal. Therefore, if two images reach the same depth in AB magnitudes but one uses a narrow band (near the Lyman limit) and the other a broad band, the narrow band will be more sensitive to escaping LyC.

All of these factors must be carefully considered when comparing candidate detection rates and contamination rates. If the detection rate is not significantly higher than the expected contamination rate, then no detection should be claimed.

Nestor et al. (2011) carefully calculate the contamination rate using the surface density of objects at the same wavelength as the LyC filter, and using a contamination radius (12) equal to the radius used to match the target catalog to the LyC filter catalog. When doing so, they determine a contamination rate that would account for about half of the detected brighter Lyman-break-selected galaxies, and ∼16% of the fainter, Lyα-selected galaxies. Our results in this paper are consistent with the high contamination rate for the brighter LBGs.

For comparison, Iwata et al. (2009) covered a much wider area, but 0.6 mag shallower in depth. Because of this, they were sensitive only to the escaping LyC in more luminous galaxies. They detect a very low fraction of their candidates, 8.6% (17 of 198), and estimate that only ∼2.37% (∼4.7 detections) could be explained by foreground contamination. However, the contamination rate estimate was not precise. They used published number counts at 3000 Å, not the wavelength of their observations (∼3590 Å). As number counts rise quickly from the near-UV to optical, the counts turn out to be significantly different. Also, the counts they use are incomplete and are therefore too low. If instead we use the completeness-corrected cumulative number counts at ∼3750 Å from Nonino et al. (2009) and Vanzella et al. (2010b), then we get a much larger foreground source density ($\sim 3.5\times {{10}^{5}}$ deg⁻² instead of $1.0\times {{10}^{5}}$ deg⁻²). This estimate increases the expected contamination by a factor of 3.5, and the expected number of contaminants is ∼16.6, consistent with the 17 detected sources. It is true that, because the number counts stated above are from slightly redder wavelengths, the foreground source counts are slightly higher than at ∼3590 Å. However, the difference is likely small because of the small wavelength change, and even if the contamination rate were increased by only a factor of 2.5 (to 5.9%, or 11.7 contaminants), instead of 3.5, from their original estimate, the significance of 17 detections would only be a 1.4σ deviation from the expectation value. Of course, there could be real LyC sources in the sample, but it cannot be claimed with high confidence that the sample is anything but foreground sources.

Boutsia et al. (2011) were careful to only analyze galaxies at redshifts where the Lyman limit was close to the red edge of transmission in the filter ($3.27\lt z\lt 3.35$, when the transmission edge corresponds to $z\sim 3.27$). However, their sample was limited to relatively luminous galaxies (10 of 11 galaxies have $L\gt {{L}^{*}}$), where the average escape fraction appears to be quite low. In addition, the sample is small, and it is very possible, given the rare LyC detections of luminous galaxies, that had they sampled significantly larger samples, they would have detected some galaxies. Nestor et al. only detected 20% of luminous LBGs with deeper data ($1\sigma =29.1$ AB in the narrow-band filter, which is more sensitive to LyC versus 29.3 AB in the broadband filter), so a null result for Boutsia et al. (2011) is not unexpected.

Vanzella et al. (2010a) had large samples and were careful about contamination. However, the sample is very bright (so a low percentage of detections is already expected), and many of the galaxies are at much higher redshift than the most sensitive redshifts. Specifically, less than half of the non-AGNs (58 of 128) are at $z\lt 3.6$, such that the longest wavelength of transmission in the filter (∼3900 Å) is sensitive to the galaxies' LyC emission within one mean free path of the ${{}_{{\rm mfp}}}(z=3.6)\sim 50$ Mpc (Prochaska et al. 2009). Thus, most of these galaxies would not be detectable with significant escape fractions. In addition, the escape fraction limits for the galaxies at $z\sim 3.5$–3.6 are significantly less sensitive than those at $z\sim 3.4$, the lowest redshift probed by their filter. Therefore, it is perhaps not surprising that Vanzella et al. (2010a) do not have a significant number of detections (after using high-resolution HST resolution to minimize foreground contamination).

In summary, all of these surveys at $z\sim 3$ are consistent with the interpretation that high escape fractions in bright, $L\gtrsim 0.5{{L}^{*}}$, galaxies are rare. Our results in this paper are consistent with this scenario. In addition, Nestor et al. (2013) suggest that it is more common to have high escape fractions from faint Lyα emitters. In a subsequent paper, we will investigate escape fractions and foreground contamination in the fainter sample.

In total, there were six bright, $z\sim 3.1$, LBGs with candidate LyC emission identified via detection in the Keck/LRIS NB 3640 filter (Nestor et al. 2011). Here we have presented follow-up HST and Keck/NIRSPEC observations of five of the six candidates. In each case, the candidate LyC emission is associated with one of multiple clumps or, in the case of D17, offset from the primary galaxy but not obviously associated with another galaxy. Therefore, spectroscopic follow-up was required to confirm the redshifts of the clumps that were emitting the candidate LyC emission.

One of the five target galaxies (aug96m16) was found to have been misidentified as a galaxy at $z\gt 3.06$ from the Keck/LRIS optical spectrum. The actual redshift of the galaxy detected in the NB 3640 filter is unknown, so it should not have been included in the original sample as an LyC-emitting candidate. Therefore, there are four galaxies with $z\gt 3.06$ with candidate LyC emission. The Keck/NIRSPEC spectroscopy revealed that two of the remaining four candidates were found to be contaminated by galaxies at lower redshifts (C49_NW and MD32_NE). A redshift for the component to the NW of MD46, MD46_NW, could not be confirmed. Nestor et al. (2013) suggest that extended Lyα emission in the direction of MD46_NW indicates that the galaxy is likely at the same redshift as the brighter counterpart, MD46. However, the NIRSPEC limit on [O iii] λ5007 is quite low, suggesting that it must have a significantly lower ${{L}_{[{\rm O}\,{\rm III}]}}/{{L}_{{\rm UV}}}$ ratio than MD46 if MD46_NW is indeed at the same redshift as MD46. This remains the most promising LyC-emitting candidate of the bright galaxies in this paper. Finally, the purported LyC emission from D17 is offset by ∼1''. There is not a counterpart galaxy at the location of the emission down to F814W ∼27.3 (AB). The low-S/N emission seen from the ground is not seen in the F336W image; however, varying IGM opacities may account for this difference.

In summary, we find that the contamination rate of this bright, $L\gt 0.5{{L}^{*}}$, sample of galaxies is at least 50% (2 of 4), and we do not confirm any definitive LyC emitters in the sample. This is consistent with the contamination rates estimated in Nestor et al. (2011, 2013). Nestor et al. (2013) find significantly lower contamination rates in their faint, Lyα-selected sample, suggesting that these types of galaxies have far higher average escape fractions. In Section 5, we show that the other $z\sim 3$ surveys for LyC are generally consistent with this picture, once varying depth, sample luminosity, and contamination rate are taken into account. Because the fainter, Lyα-selected samples are expected to have higher escape fractions and lower contamination rates, in a subsequent paper we will present results of additional HST imaging and Keck spectroscopy of this promising population.

Support for program 11636 was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. The authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain.

Facilities: HST (WFC3, ACS), Keck:I (LRIS), Keck:II (NIRSPEC)