ON THE DETECTION OF IONIZING RADIATION ARISING FROM STAR-FORMING GALAXIES AT REDSHIFT z ∼ 3–4: LOOKING FOR ANALOGS OF "STELLAR RE-IONIZERS"

The origin of the ionizing radiation, whether from galaxies or active galactic nuclei (AGNs), responsible for the re-ionization of the universe at high redshift, z > 6, and for keeping it ionized at later epochs, is still poorly constrained. The contribution of quasars to the hydrogen ionizing background increases as we look back in time from z = 0 to z ∼ 2 as the peak of the quasar luminosity function is approached (e.g., Fanidakis et al. 2011). Beyond redshift 2 their contribution significantly decreases and the hydrogen photoionization rate is most probably maintained by stellar emission (e.g., Siana et al. 2008; Faucher-Giguère et al. 2009; Haardt & Madau 2012; but, see Fiore et al. 2012). The star-forming galaxies at z > 3 are therefore the leading candidates to provide the remaining ionizing photons. The severe intergalactic medium (IGM) absorption at z > 4.5 prevents us from observing directly the Lyman continuum (LyC) emission (Meiksin 2006; Inoue & Iwata 2008). This is particularly true during the re-ionization of the universe (z > 7). It is therefore essential to identify diagnostics of the LyC emitters at lower redshift, z ≃ 3–4 (at λ_rest > 1216 Å), and infer if galaxies with these characteristics are more common during the epoch of re-ionization (1 Gyr earlier or z > 7).

At z ∼ 3–4 the direct detection of LyC emission from galaxies is still difficult because ionizing radiation is severely attenuated by neutral gas and by dust in the interstellar and circumgalactic medium (ISM and CGM) of the source itself, as well as by the intervening IGM. As a result, the number of reported detections of star-forming galaxies at high redshift, typically selected as Lyman break galaxies (LBGs) or with equivalent criteria, with LyC emission is very small (Shapley et al. 2006; Iwata et al. 2009; Nestor et al. 2011; S06, I09, and N11, respectively, hereafter; Vanzella et al. 2010a, V10b hereafter; Boutsia et al. 2011). The issue is further complicated by the relatively high probability of finding faint low-redshift galaxies at very close angular separation, i.e., ∼1 arcsec or less, from a high-redshift galaxy with brighter apparent magnitude (Vanzella et al. 2010b, V10a hereafter). These interlopers, for which spectroscopic identification is not viable, can be wrongly interpreted as spots of LyC emission from the host galaxy if high angular resolution multi-band photometry is not available, as is often the case, to recognize their nature.

Recently, searches of LyC emitters with deep imaging below the Lyman limit have yielded samples of candidates at redshift z ∼ 3 where the region of emission of the candidate LyC ionization radiation is spatially offset from that of the non-ionizing rest far-UV light (called "nLyC" in what follows), namely, the main body of the galaxy typically observed between 1500 Å and 2000 Å. The displacement of these putative LyC-emitting "blobs" is generally less than 1 arcsec, but in some cases values as high as 2'' have been reported (I09; N11), corresponding to separation in the range of several kpc to a few tens of kpc. In other words, the candidate LyC emission would come from regions of the galaxies that are well separated from the center, which is where most of the LyC photons are produced, and in some cases so far from it to be classified as separate sources. If confirmed this would provide important empirical constraints on how LyC escapes from galaxies. Future observations with the new Hubble Space Telescope (HST)/Wide Field Camera 3 (WFC3) in the LyC rest frame at z ∼ 3 will help exploring this issue further. In the meantime, however, motivated by the potential importance if the above findings are confirmed, we use spatially resolved, multi-band photometry from the HST to constrain the nature of a sample of LyC emitter candidates that we have selected in the GOODS South, one of the fields targeted by the CANDELS project (Grogin et al. 2011; Koekemoer et al. 2011) from ultradeep U-band imaging below the Lyman limit of star-forming galaxies with known spectroscopic redshift z ⩾ 3.4. We also report on the only one LBG in our sample at z = 3.795 (Ion1, hereafter) with a clear LyC emission.

We have selected a sample of LyC emitter candidates from ultradeep Very Large Telescope (VLT)/VIMOS imaging in the U band of the GOODS South field (V10b). The U-band images, which are described in Nonino et al. (2009), were aimed to probe the spectral region below the Lyman limit of these galaxies for possible emission of ionizing radiation. The images are very deep, reaching 1σ flux upper limit of about 30 mag (AB) for an unresolved source within a circular aperture of 2'' diameter.

A relatively large number of sources, selected as LBGs, have secure redshift identification in the GOODS-S field (Vanzella et al. 2008, 2009; Popesso et al. 2009; Balestra et al. 2010). As described in V10b, we start from a spectroscopic sample of 122 B-band dropouts at 3.4 ⩽ z ⩽ 4.5 selected by Giavalisco et al. (2004) from the GOODS/Advanced Camera for Surveys (ACS) images for which we have robust redshifts, i.e., Quality Flag QF = A (Vanzella et al. 2009; Balestra et al. 2010). The lower limit of the redshift range is the lowest value such that the 912 Å Lyman limit is outside, redward of the system throughput in the U filter, while the upper is chosen because the rapid increase of opacity of the IGM produces too small transparency at higher redshift to make analysis of z > 4.5 galaxies useful (see V10b). Out of 122 B-band dropouts, 7 are detected in the Chandra 4 Ms images in the Chandra Deep Field South (CDF-S; Xue et al. 2011) and classified as AGNs. Among these 122 sources we have flagged as potential LyC emitter candidates the 32 of them which have flux detection at the 2σ level or larger within a circular aperture of 12 diameter in the U-band image. In this work, we discuss in detail these U-band emitters (named Uem hereafter).

As we noticed in V10b, in 28 of the 32 sources the U-band emission, i.e., the candidate LyC light, is spatially offset relative to the centroid of the rest-frame far-UV light of the LBG (the nLyC at wavelength around ≈1500 Å). A visual inspection of the HST/ACS images showed that in 9 of the 28 sources the U-band flux light can unambiguously be explained as coming from the outer isophotes of nearby sources, which most likely are foreground interlopers (e.g., see Figures 4 and 5 of V10b). We have eliminated these galaxies from the sample of LyC candidates. The remaining 19 candidates are listed in Table 1 and are analyzed here. They have counterparts in the ACS and WFC3 images which are close to the targeted host LBG with angular separations in the range 04–19. Image cutouts for these sources are shown in V10b, here we show in Figure 1 and the Appendix the more critical cases where the angular separation is smaller than 1''. All but one of the U-band sources are fainter, in AB magnitudes, than their ACS optical (rest UV) nLyC counterparts. Table 1 reports the ratio of F1500/F_LyC for the 19 objects.

Zoom In Zoom Out Reset image size

Table 1. Sample of Offset U-band Emission from LBGs in GOODS-S

GOODS ID	Δθ	zspec	zphota (Uem)	S/NB	Δ(B − V)obs	Δ(V − i)obs	$\left(\frac{F1500}{F_{{\rm LyC}}}\right)_{{\rm obs}}$	fbesc, rel
(Uem)	('')	LBG	GOODS,CANDELS	(Uem, LBG)	(LBG–Uem)	(LBG–Uem)		(%)
J033200.95−274443.4	1.45	3.485	1.10(0.06), −1.00	23,27	1.06	...	1.80	251
J033203.33−274524.6	1.90	3.472	2.34(0.15), 1.92(0.08)	97,26	1.13	...	1.20	376
J033204.91−274451.0⋆	0.70	3.404	1.98(0.23), −1.00	19,16	1.71	...	0.66	792
J033212.88−274833.1	1.50	4.292	1.44(0.09), 1.48(0.05)	30, <1	...	1.15	1.01	724
J033217.47−274141.6	1.20	3.780	0.52(0.04), 0.51(0.03)	20,4	1.00	...	6.85	76
J033217.41−275200.3	1.40	3.494	0.93(0.18), 0.67(0.24)	10,15	1.07	...	2.29	197
J033217.79−275050.4	1.60	3.477	1.55(0.03), 1.52(0.08)	8,13	1.58	...	1.77	255
J033218.85−275134.3	1.30	3.659	1.46(0.06), 1.18(0.05)	35,4	2.17	...	1.31	373
J033219.02−274148.6	1.10	3.764	1.90(0.14), 1.75(0.03)	20, <1	2.82	...	1.54	339
J033220.95−275021.8⋆	0.60	3.478	2.66(0.07), −1.00	14,6	1.22	...	3.08	146
J033222.79−274439.0	1.30	3.582	0.58(0.75), 0.39(0.02)	28,11	0.85	...	2.27	210
J033222.95−274727.8⋆	0.80	4.440	0.46(0.46), −1.00	4, <1	...	1.05	2.31	325
J033223.32−275155.9	1.10	3.470	0.07(0.07), −1.00	19,12	1.36	...	4.45	102
J033226.43−274124.5⋆	0.70	4.384	2.31(0.16), 2.32(0.20)	19,4	...	0.56	2.68	280
J033228.95−274232.9	1.40	3.585	2.72(0.18), 2.77(0.17)	12,11	1.05	...	2.42	197
J033234.33−274123.0	1.60	3.418	2.30(0.25), 2.33(0.31)	17,11	1.31	...	2.86	148
J033236.85−274557.6c	0.40	3.797	−1.00, −1.00	5,6	1.80	...	1.36	387
J033240.48−274431.7	1.40	4.120	2.39(0.07), 1.87(0.04)	36,2	...	0.54	1.24	505
J033253.57−275300.1	1.10	3.472	2.17(0.25), −1.00	9,16	1.06	...	1.66	272
(Ion1)	<01	3.795	3.64(0.04), 3.68(0.09)	10.3	...	...	15.00	35

Notes. (⋆) Sources shown in Figure 1. ^aThe photometric redshifts with their rms (GOODS and CANDELS, on left and right, respectively) without the inclusion of the U band are reported. ^bThese are the minimum f_{esc, rel} derived from Equation (2) adopting conservatively the maximum IGM transmission at the given redshift. 10,000 IGM transmissions have been calculated from simulations specifically computed at the spectroscopic redshift of the LBG. The intrinsic ratio has been assumed to be L1500/L_LyC = 3 (see the text for details). ^cGalaxy observed in the HUDF and shown in Figure 8. The angular separation between the compact blue emitter and the center of light of the LBG is 04.

Download table as:  ASCIITypeset image

The remaining 4 U-band detections of the initial sample of 32 are cospatial with the ACS and WFC3 images of the host sources, in the sense that the centroid of the U-band light falls within the errors (∼01) on the location derived from the optical HST/ACS images. As discussed in V10b (and reported in their Table 3), three out of four are detected in the 4 Ms X-ray Chandra images (0.5–8 keV), and also show typical signatures of AGNs in their optical spectra, such as high-ionization emission lines (e.g., C iv, N v). The forth source, which we call Ion1, is our only robust candidate stellar LyC emitter and we will describe it further in Section 6.

2.1. The Effect of the Intergalactic Attenuation

The method we are adopting here is based on an intermediate-band filter (FWHM = 350 Å, U band), and is similar to the typical narrowband (NB) imaging only in the cases where 3.4 < z < 3.5 (9 out of 19 LBGs), i.e., the Δλ between the Lyman limit of the galaxy and the red cutoff of the filter is minimal. In this case, the greater depth of our image compensates for the larger noise due to the broader bandpass. For example, if compared with the NB3640 narrowband filter used by N11 (FWHM = 100 Å), the depth of our U-band image nearly exactly compensates for the different filter setup.

Differences between the methods arise if we consider, as in our approach, a variable redshift. Indeed, the LyC region probed by our filter is redshift dependent, e.g., it is λ_rest < 909, 800, 741 Å at z = 3.4, 4.0, 4.4, respectively. We have investigated the effect of the IGM attenuation as a function of redshift by running Monte Carlo (MC) simulations as in V10b (where the IGM attenuation is extensively taken into account). To this end, as a reference, we adopt a magnitude of i₇₇₅ = 24 and a ratio F₁₅₀₀/F_LyC = 7 (V10b; Siana et al. 2007). We then apply 10,000 different IGM transmissions convolved with the U-band filter and add photometric noise to the estimated flux in the U band (see V10b). In these conditions, at z = 3.4 we retrieve 82⁺⁴ _{− 4}% of the sources in their LyC (U band) at S/N > 2 (88⁺³ _{− 4}% if i₇₇₅ = 22). As redshift increases, the fraction of recovered sources decreases because of the IGM attenuation (see Figure 2). The inner box of Figure 2 shows how the IGM affects the fraction of recovered sources as a function of redshift (normalized to the z = 3.4 case).

Zoom In Zoom Out Reset image size

It is worth noting that the majority of the sources analyzed here are at z < 3.8 (15 out of 19). Moreover, two examples at relatively high redshift have recently been reported: an LBG at z = 3.8 (Ion1, also discussed in this work) and an AGN at z = 4.0 with i₇₇₅ = 26.09, for which an LyC emission at λ_rest < 830 Å and λ_rest < 800 Å is detected at signal-to-noise ratio (S/N) of 5.2 and 3.3 by our deep U-band imaging, respectively (V10b). This suggests that despite the low average IGM transmission at z ≳ 3.8, its stochastic behavior makes the variance to be relatively large. Indeed, at fixed redshift, the distribution of the IGM transmissions (U-band-convolved) is asymmetric with an extended tail toward high values (Inoue & Iwata 2008). The reason is that the LyC absorption by the IGM is very stochastic because it is related to the presence of relatively rare Lyman limit systems (LLSs) or damped Lyα system (DLA), having ${\rm NH\,\mathsc{i}}>10^{17} {\rm \,cm}^{-2}$. With an LLS, the transmission is suddenly cut down at the corresponding wavelength. Conversely, without an LLS (or DLA system) near the source, we can expect a significant transmission even far below the source Lyman limit (see Inoue & Iwata 2008).

Having this in mind, we decided to keep the whole sample up to z ∼ 4.4 in the analysis performed in the present work. If we could establish that the U-band-selected sources observed in proximity of their LBGs were at the same redshift of their companions, then we would safely conclude that they are LyC emitters. We investigate the nature of their redshift in Section 4. In the following section, we briefly recall the issue of the contamination by lower redshift sources randomly placed at small angular separations from the higher redshift LBG.

The likelihood that a foreground interloper located in the vicinity of an LBG is responsible for the detection in the U band increases with the redshift of the LBG, with the angular separation and the quality and depth of the observations. Siana et al. (2007), I09, and N11 calculated analytically the probability that candidate LyC emission from an LBG is due to an interloper, given the characteristics of the data (quality and depth). V10a performed the same calculations and ran MC simulations to quantify this effect.

The displacement of the U-band light relative to that at redder wavelengths, as measured from the light centroid in the U band and ACS images, for the sample 19 galaxies discussed here, is in the range 0.4 arcsec < Δθ < 1.9 arcsec; in all cases there always is a counterpart to the U-band source in the HST/ACS and WFC3 images. We calculated that the number of sources observed in the two annular bins with radii 0''–1'' and 1''–2'' from the LBG centroid in the ACS z₈₅₀-band images is equal, within the errors, to the expectations for foreground galaxies at increasing separations. That is, given the number counts and assuming a uniform distribution, the fraction of intercepted foreground sources increases with the area of the annulus considered.

One of the parameters adopted in V10a was the seeing (the point-spread function (PSF) of the images), strictly related to the possibility to deblend close sources and related to the probability of superposition, which increases for the worst seeing conditions. The fact that the LyC emission could be intrinsically offset from the main galaxy (as reported in N11 and I09) further complicates the interpretation, if the redshift of these potential emitters is not known. This is still true if high spatial resolution images are available and the sources are well separated, i.e., in the absence of the redshift information an intrinsically offset LyC emission is fully compatible with the emission of a lower redshift object.

On the one hand, the effects of this foreground contamination can be corrected statistically. On the other hand, it is worth investigating carefully each LyC candidate, since any consideration about the mechanisms that allow the LyC photons to escape primarily depends on the reliability of the LyC detection. Therefore, we now turn to the discussion of the observational evidence that will help us to constrain the redshift, and hence the nature, of these sources in our sample (Section 4).

4.1. The Escape Fractions

In this section, we calculate the absolute and relative LyC escape fractions (defined below) of our potential LyC emitters assuming that they are at the same redshift of the LBG. Since f_esc and (f_{esc, rel}) have to obey to clear limits, this in turn puts constraints on other quantities, in particular the redshift. In order to do that, we have to first define the relation among various quantities and calculate the escape fraction where spatially offset LyC and nLyC emissions are present.

4.1.1. Escape Fraction from a Morphologically Resolved Lyman Continuum

Following Siana et al. (2007) the observed flux ratio between the 1500 Å and the LyC is affected by several factors and is expressed as

$$\begin{eqnarray} \left(\frac{F1500}{F_{{\rm LyC}}}\right)_{{\rm OBS}} &=& \left(\frac{L1500}{L_{{\rm LyC}}}\right)_{{\rm INT}} \times 10^{-0.4(A1500-A_{{\rm LyC}})} \nonumber \\ && \times e^{\tau _{{\rm H\,\mathsc{i},IGM}}({\rm LyC})} \times e^{\tau _{{\rm H\,\mathsc{i},ISM}}({\rm LyC})}, \end{eqnarray} \tag{ 1 }$$

where LyC is the wavelength at which the Lyman continuum is observed, (L1500/L_LyC)_INT is the intrinsic luminosity density ratio, (A₁₅₀₀ − A_LyC) is the differential dust attenuation (in magnitudes), $\tau _{{\rm H\,\mathsc{i},IGM}}({\rm LyC})$ is the line-of-sight opacity of the IGM for LyC photons (and transmission is $T_{{\rm LyC}}^{{\rm IGM}}=e^{-\tau ^{\rm IGM}_{{\rm LyC}}}$), and $\tau _{{\rm H\,\mathsc{i},ISM}}({\rm LyC})$ is the optical depth of the LyC absorption from H i within the observed galaxy's ISM (whose transmission is defined as $T_{{\rm ISM}}^{{\rm H\,\mathsc{i}}}=e^{-\tau _{{\rm H\,\mathsc{i},ISM}}({\rm LyC})}$).

The relative fraction of escaping LyC photons relative to the fraction of escaping nLyC (1500 Å) photons is obtained rearranging the above equation:

$$\begin{equation} f_{\rm esc,rel} = \frac{(L1500/L_{{\rm LyC}})_{\rm int}}{(F1500/F_{{\rm LyC}})_{\rm obs}} \exp \big(\tau ^{\rm IGM}_{{\rm LyC}}\big), \end{equation} \tag{ 2 }$$

it compares the observed flux density ratio (corrected for the IGM opacity) with models of the ultraviolet spectral energy distribution (SED) of star-forming galaxies. If the dust attenuation A₁₅₀₀ is known, f_{esc, rel} can be converted to f_esc as $f_\mathrm{esc} = 10^{-0.4A_{1500}} f_\mathrm{esc,rel}$ (e.g., Inoue et al. 2005; Siana et al. 2007). Again, from the above equations, f_esc can be written as

$$\begin{equation} f_{\rm esc} = \hbox{exp} [{-}\tau _{{\rm H\,\mathsc{i},ISM}}({\rm LyC})] \times 10^{-0.4(A_{{\rm LyC}})}, \end{equation} \tag{ 3 }$$

the two factors on the right side have values in the range [0–1]. Clearly, their product, i.e., f_esc, cannot be greater than 1.

It has recently been argued by N11 that due to the details of the radiative transfer and sources of the corresponding photons, any Lyα emission and escaping LyC flux will not be necessarily cospatial with either each other or with the bulk of the rest-frame UV flux in a given galaxy. This is true when considering Lyα emission, which typically arises from backscattering of moving hydrogen gas that can be spatially decoupled from the ionizing sources. The observed light centroids (barycenter) of LyC and nLyC can be displaced in the case in which the LyC arises from a sub-region of a larger area, however, the local emission is not spatially shifted, i.e., if the ionization radiation is measured in some sub-region, then the nLyC radiation is expected to be detected too (typically with brighter magnitude). The crucial point here is that the two quantities (F1500)_obs and (F_LyC)_obs must be measured in the same spatial (i.e., physical) region, where the ionizing and non-ionizing radiations arise.

Conversely, recent works have performed measures with the aim to include both the fluxes (F1500)_obs and (F_LyC)_obs by enlarging the apertures (I09) or by deriving total fluxes from the images regardless of the misalignment (e.g., SExtractor MAG-AUTO in U and R bands, as in N11), resulting in measures within different aperture sizes and/or shapes. In this way, the measured observed ratio (F1500/F_LyC)_obs, and consequently the f_esc quantity, is strongly biased. In particular the resulting f_esc would be severely underestimated because the correct (F1500)_obs value would be smaller. Figure 3 shows an illustrative example in the GOODS-S field (also discussed below and reported in Table 1) in which the observed ratio has been calculated with and without taking into account the offset emission, top and bottom panels, respectively. Under the same assumptions (intrinsic luminosity ratio and IGM transmission), the former method (top panel) produces an f_esc that is ∼10 times lower than the latter (bottom panel).

Zoom In Zoom Out Reset image size

The consequence is that if the estimated f_esc exceeds the value of 100%, then some other quantity has to be revised. It is even more significant if this happens under conservative assumptions of (L1500/L_LyC)_int, dust, and IGM attenuation.

If, from one hand, the f_esc has to be less than 100%, the constraints can be even more stringent if f_{esc, rel} has to be less than 100%. Indeed from Equations (1) and (2), it turns out that

$$\begin{equation} 10^{-0.4(A_{1500}-A_{{\rm LyC}})} \times f_{\rm esc,rel} = \hbox{exp} [{-}\tau _{{\rm H\,\mathsc{i},ISM}}({\rm LyC})]. \end{equation} \tag{ 4 }$$

In order to have the transmission of the ISM correctly in the range [0–1] (right side), the f_{esc, rel} has to be less than $10^{0.4(A_{1500}-A_{{\rm LyC}})}$, which in turns is a quantity less than 1 (see Siana et al. 2007, their Figure 2). Therefore, also the relative escape fraction has to be less than 1.

4.1.2. Anomalous Escape Fractions in Our Sample

Following the discussion of the previous section, we have estimated f_{esc, rel} for our sample in the correct way, i.e., in same spatial regions where the U-band emission is observed, assuming that the U-band sources are at the same redshift as the LBG, i.e., they are LyC emitters. Following Equation (2), we calculate the f_{esc, rel} by assuming an intrinsic ratio (L1500/L_LyC)_int = 3 (Shapley et al. 2006) and the maximum IGM transmission at the redshift of the LBG.⁹ The intrinsic ratio is also justified by the fact that the sources show a UV spectral slope (from their (i − z) color) similar or redder than their LBGs. Therefore, the adopted ratio is a conservative assumption.

The f_{esc, rel} values are reported in the last column of Table 1. All but one U-band LyC candidates have f_{esc, rel} larger than 100%, which would imply a less dust attenuation for shorter wavelengths, i.e., A_LyC < A_nLyC, in contrast to any dust extinction law. This is even more evident if we adopt the ratio (L1500/L_LyC)_int = 7 (e.g., Siana et al. 2007; V10b), that increases the f_{esc, rel} values of Table 1 of a factor 7/3. The one case with f_{esc, rel} < 100% has inconsistent photometric redshift and UV colors as described below.

4.2. The UV Colors

The B − V color of galaxies at the redshift considered here largely depends on the cosmic opacity of the IGM (see Madau 1995), specifically due to the effects of the Lyα forest. Given the spatial correlation scale of the IGM, if the U-band companions are at the same redshift as their associated LBGs, their B − V colors should therefore be similar.

Figure 4 shows the difference of the observed (B − V) color Δ(B − V) between the LBG and their companion LyC candidates (empty circles, Δ(B − V) = (B − V)_LBG − (B − V)_Uem) as a function of the transverse separation (physical) calculated at the mean redshift of the Lyα forest. The same is shown for the four 4.0 < z < 4.5 galaxies for which the Δ(V − i) has been adopted, more suitable than the (B − V) color in probing the IGM decrement. For all, the typical error in the color is ≲0.1 mag, because these sources are detected at more than 10σ in each band. As the figure illustrates, the LyC candidates are much bluer than their companion LBGs, implying either a very high fraction of escaping ionizing radiation and/or a high transmission of the IGM, or more likely as we are about to argue that their redshift is significantly lower than that of the LBG, in agreement with the previous result and the photometric redshift analysis (reported in the next section). This is seen through MC simulations in which we have simulated f_esc = 1 for the LyC candidates in the UV rest-frame continuum probed by the B₄₃₅ band and we have conservatively assumed that the IGM transmission at λ < 912 Å is the same as that at 912 Å < λ < 1215 Å (for the simulations we used the IGM transmission by Inoue & Iwata 2008 and Inoue et al. 2011, and the SB99 UV templates with age 100 Myr, continuous star formation and metallicity z = 0.004, and Salpeter initial mass function). We can then correct the observed (B − V) color of the LyC emitters and recalculate the Δ(B − V), which we plot in Figure 4 as filled circles (the case in which the f_esc = 1). The corrected color differences still show that the LyC candidates are much bluer than their putative LBG companions. An even much more conservative correction can be made by assuming unitary IGM transmission blueward of the Lyα line, in all the cases and recomputing the color differences (filled triangles), which still shows that the candidates would be much bluer than the LBG.

Zoom In Zoom Out Reset image size

The observed median (B − V) color and its 68% percentile interval of the LyC candidates is 0.24^+0.40 _{− 0.22}, significantly bluer than that of the LBG sample, which is 1.85^+0.95 _{− 0.58}, in contrast with what would be expected if their redshift were the same as that of the LBG and their stellar populations were similar (e.g., Meiksin 2006).

We also explored a more extreme possibility by calculating at a given redshift and f_esc = 1 the expected colors arising from population III stars (Pop III). The Pop III stars of Inoue (2011) template (with zero metallicity), convolved with the U-band filter and the IGM transmissions (1000 lines of sight), are partially able to reproduce the observed (B − V) colors only in the case of high transparency of the IGM. Moreover, the presence of Pop III stars produces a very steep UV spectral slope (with a color (i − z) = −0.2), that is in contrast to what observed in our sample, i.e., 〈(i − z)〉 = 0.1 ± 0.2.¹⁰ Therefore, we exclude the presence of Pop III stars.

Regardless of theoretical expectations, in all cases that we considered the U-band companions have (B − V) colors that are bluer by more than 1 mag (see Δ(B − V) in Table 1) than their corresponding LBGs. If placed at the same redshift, the line of sight to each LyC candidate/LBG pair would be probing the IGM at transverse separations smaller than 20 kpc (physical) at the mean redshift of the Lyman forest. This is much smaller than the transverse cross-correlation function observed in QSO pairs separated of several arcminutes, for which the coherence length in the IGM is at least of the order of 500 h⁻¹ kpc proper (e.g., Fang et al. 1996; D'Odorico et al. 1998; Rauch et al. 2005; Cappetta et al. 2010). This implies that the IGM attenuation due to the forest should be the same for the LyC candidates and their associated LBGs if they were at the same redshift. Since the (i − z) colors of the Uem sources are similar or even redder than those of the LBGs, implying comparable or even redder ultraviolet spectral slope (similar stellar populations and obscuration properties), their (B − V) colors should also be similar, since the IGM attenuation would in practice be the same.

Thus, we interpret these simulations as evidence that the LyC candidates are not regions of the LBGs where ionizing radiation is escaping from but rather relatively unobscured star-forming galaxies at significantly lower redshift than the LBGs.

4.3. Photometric Redshift and SED Fitting

To further investigate the nature of the U-band sources, namely, whether the regions of the host LBGs from where LyC ionizing radiation is escaping or sources located at different, very likely lower, redshifts, we have fit the spatially resolved CANDELS multi-band HST (BVizJYH), ground-based (U and K) and Spitzer/IRAC photometry to templates and spectral population synthesis models to derive their photometric redshifts and the parameters of their stellar populations (stellar mass, star formation rate (SFR), dust obscuration, and age).

The images in the above bands have different angular resolution, and to robustly measure the photometry of our sources we have used the TFIT software package, developed by the GOODS and CANDELS teams (Laidler et al. 2007), which uses positional priors and PSF information to measure apparent magnitudes in matched apertures. For each source, TFIT uses the spatial position and morphology from the HST high-resolution images (we used both the ACS z band and the WFC3/IR H band) to construct a template image, which includes nearby sources. While these are fully resolved at the HST angular resolution, they might be marginally or even heavily blended in the other images. This template is then fit to the images of the object in all other low-resolution bands after convolution with the appropriate PSF. During the fitting procedure, the fluxes of the object in low-resolution bands are left as free parameters. The best-fit fluxes are considered as the fluxes of the object in low-resolution bands. These procedures can be simultaneously done for several objects which are close enough to each other in the sky. Experiments on both simulated and real images show that TFIT is able to measure accurate isophotal photometry of objects to the limiting sensitivity of the image (Laidler et al. 2007).

We derive the photometric redshift and the physical properties of the LyC emitter candidates by fitting the observed SEDs to stellar population synthesis models. Models used to measure photo-z's are extracted from the library of PEGASE 2.0 (Fioc & Rocca-Volmerange 1997). Instead of using the redshift with the minimum χ², we integrate the probability distribution function of redshift (zPDF) and derive the likelihood-weighted average redshift. When the zPDF has two or more peaks, we only integrate the main peak that has the largest power.

Since the U-band-selected LyC candidates have counterparts in the HST z and H bands in each case, we have generated two sets of fits, one based on positional priors from the ACS z-band images, which sample the rest far-UV SED of the galaxies at λ ≲ 1800 Å, and the other on priors from the WFC3 H-band images (λ ≲ 3600 Å). The first (GOODS) set uses photometry in the U and the ACS BViz bands, in the VLT/ISAAC JHK images of GOODS South described by Giavalisco et al. (2004) and Retzlaff et al. (2010), and in the four GOODS Spitzer/IRAC bands. The second (CANDELS) set uses the WFC3/IR YJH near-IR photometry in place of the ground-based one. The z-band positional priors might better reflect the spatial location of the LyC-emitting regions, since essentially the same stars that power it also power most of the light in the nLyC far-UV spectrum. The z-band images also have better angular resolution than the H-band ones. The second set offers an independent set of measures, which takes advantage of the generally more sensitive and higher-resolution WFC3 data; however, it lacks the K-band data.

Since the potential emission in the LyC spectral range is not taken into account in the models, and thus the flux in the U band could skew the photometric redshift calculation toward lower values, for both photometric sets we have run the fit with and without the U-band photometry. Figure 5 shows the redshift probability function derived from the GOODS photometric set; Figure 6 shows the one from the CANDELS set. In all cases the solid curve refers to the calculation made using the U-band photometry, the dashed curve without it. In all cases the photometric redshifts are calculated as the weighted average of zPDF as

$$\begin{equation} z_{{\rm phot}} = {\int _0^\infty z\, P(z)\, dz \over \int _0^\infty P(z)\, dz }. \end{equation} \tag{ 5 }$$

A blank panel in Figure 6 means that the source is outside of the region covered by the CANDELS observations.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The general result from our analysis is that the photometric redshifts of the LyC emitter candidates are systematically much smaller than the spectroscopic redshifts of their host LBGs. Exceptions include the cases of the LyC candidates GDS J033204.91−274451.0, GDS J033220.95−275021.8, GDS J033222.95−274727.8, and GDS J033223.32−275155.9 (see Figures 5 and 6), where the GOODS photometric redshift is much lower than the spectroscopic redshift of the LBG, but the CANDELS photometric redshift is not. These are the cases in which the separation between the LBG and the Uem is ≲1'' and are not individually detected in the CANDELS H-band-based catalog (they are marked with red crosses in Figure 6).

• 1.  
  In the first case, GDS J033204.91−274451.0, the LyC emitter candidate is resolved as a different source from the LBG in the ACS z-band image but not in the H-band one. The photometric redshift from the GOODS photometry yields a significantly lower redshift than the LBG's spectroscopic one, with some dependence of the redshift probability function on the exclusion of the U band from the fit, but the resulting photometric redshift is in any case much lower than the LBG's. Since the LyC candidate and the LBG are not resolved in the H band, the CANDELS photometric redshift is essentially that of the LBG itself. If the U-band photometry is not included in the fit, the spectroscopic and photometric redshifts are in very good agreement. Including the U-band data, however, yields a photometric redshift somewhat lower than the spectroscopic one (z_spec = 3.404, z_phot = 3.351), which is not surprising since this is the LBG with the lowest redshift in our sample and thus some flux detected in the bluest band might easily skew the fit toward lower redshift values.
• 2.  
  The other three sources (GDS J033222.95−274727.8, GDS J033220.95−275021.8, GDS J033223.32−275155.9) are also the cases in which the LyC candidate and the LBG are classified as individual sources in the z band but not in the H band. In both cases, the GOODS photometric redshift is much lower than the LBG's spectroscopic one. In the former, since the LBG redshift is at the high end of our sample (z_spec = 4.440), adding the U-band data to the derivation of the photometric redshift does not make a difference. In the other two, neglecting the U-band data results in very good agreement between the spectroscopic and photometric data, while including it lowers the photometric redshift solution.

The LyC emitter candidate with coordinates R.A. = 03:32:36.85, decl. = −27:45:57.6, associated with galaxy HUDF J033236.83−274558.0 is discussed in detail in the Appendix. It is the fainter among the cases discussed in this work with an U mag of 28.6. A new observed feature in the J band (at 10σ significance) is consistent with relatively strong line emission of a low-redshift galaxy, lower than its companion LBG.

A general characteristic of the fits is that including the U-band photometry in the derivation of the photometric redshift of the LyC emitter candidates does not make a substantial difference in the results and that there is very good quantitative agreement between the GOODS and CANDELS photometric redshift, pointing to the robustness of our analysis.

Finally, a further check has been performed by calculating the photometric redshifts without the inclusion of the U and B₄₃₅ bands, which makes the test independent from the single B − V color analysis described in Section 4.2. Even though the efficiency of the photometric redshift prediction is slightly reduced, the resulting z_phot best-fit values still favor low-redshift solutions for all sources, except for GDS J033223.32−275155.9 that increases from z_phot = 0.07 ± 0.02 to 3.60 ± 0.2. In this case, however, the (B − V) = 0.65 color, the relative escape fraction, and the Δ(B − V) measurements are in contrast with this high-redshift solution.

4.4. Conclusion From f_esc, UV Colors, and Photometric Redshift Analysis

In this section, we review all known (to us) galaxies at high redshift (z ≳ 3) for which emission of LyC ionizing radiation has been reported or suspected.

• 1.  
  Direct detection of LyC ionizing radiation has been reported by Shapley et al. (2006, hereafter S06) in deep spectra of two LBGs at z ∼ 3 (U-band dropouts) in the SSA22 field, dubbed D3 and C49 in their nomenclature, out of a total of 14 sources observed with comparable sensitivity. It is worth noting that I09 and N11 found a highly significant null detection for the object SSA22a-D3, which is the brighter of the two objects for which S06 claimed an ionizing flux detection.
• 2.  
  Iwata et al. (2009) used deep NB imaging to image the rest-frame SED blueward of the 912 Å Lyman limit of members of a well-known overdensity of galaxies at z ∼ 3.1, also in the SSA22 field (Steidel et al. 1998). They reported 17 detections, 7 from galaxies selected as LBGs and 10 as Lyman Alpha Emitters (LAEs).
• 3.  
  Nestor et al. (2011) used the same technique to observe galaxies in the same SSA22 field, but extended the sensitivity of previous observations by ≈0.6 mag. They reported 34 detections, out of 156 sources, which include 6 LBGs and 28 LAEs.
• 4.  
  In the paper V10b, we used a technique similar, but not identical, to the one by I09 and N11. Instead of using an NB, we carried out ultradeep U-band imaging to search for candidate LyC emitters among galaxies at 3.4 ≲ z ≲ 4.5, namely, such that the redshifted 912 Å ionization edge of the targeted sources is to the red and completely outside of the filter's bandpass. We found 1 candidate out of 102 LBGs (described in Section 6).

For most of the NB-selected candidates and LBGs, the region where the LyC ionization radiation originates is found to be spatially offset from that of the non-ionizing far-UV light, namely, the main body of the galaxy. The displacement is generally less than 1 arcsec, but in some cases values as high as 2'' have been reported. We performed a visual inspection of the images of the galaxies by I09 and N11, and consulted Tables 4 and 5 of N11, and found that ≳ 70% of the presently identified LyC emitter candidates exhibit a spatial offset between the LyC emission and the "non-ionizing" far-UV image of the galaxies. Since an offset emission is also consistent with a lower redshift interloper, a key step is to measure the statistical occurrence of these cases (as we did in V10a). In particular, with the availability of large-area HST UV and near-IR surveys with WFC3, such as CANDELS, the prospect for substantial progress in this regard is good. The other important point is that if the offset LyC emission is real, then care must be exercised when calculating the fraction of escaping ionizing radiation f_esc in each case, i.e., whether it is from the galaxy as a whole or only from a close companion, to avoid bias in the derivation of average properties. For example it is worth mentioning that the offset possible LyC detection in S06, SSA22-C49, has a reported observed ratio of (F1500/F_LyC)_OBS = 16.4 ± 6.1 (N11), in which the flux F1500 used in the calculation is that of the main galaxy. Presumably, a calculation restricted to the local region where the putative LyC emission is seen would produces a much smaller value, i.e., a much higher f_{esc, rel}.

Finally, it must be kept in mind that part of the candidate LyC emitters reported above may have the contribution to f_esc from an AGN component that is difficult to identify without multi-wavelength and spectroscopic surveys.

Therefore, the amount of stellar LyC emission from high-z star-forming galaxies is still uncertain. Very few sources have spectroscopic redshift confirmation: (1) those (five) reported by Inoue et al. (2011) that need extreme stellar populations to justify the flux ratios observed, (2) the LBG reported in S06 (SSA22-C49) that shows an offset (05) U-band emission whose redshift is not conclusively known, and finally (3) the one we present in the next section and discovered in the GOODS-S field (Ion1).

The direct detection of LyC emitters at high redshift (as high as possible) allows us to characterize their properties at λ > 1215.7 Å and tries to identify similar sources and/or look if they are more common during the re-ionization epoch. For this reason it is important to identify secure LyC emissions. We recall one of the most promising stellar ionizers at redshift 3.795 we have identified in the GOODS-S field and currently the highest redshift known so far (GDS J033216.64−274253.3, Ion1) and is useful to revisit it in the context of this work by comparing its appearance with the offset Uem sources discussed above.

6.1. Observed Photometric and Spectroscopic Properties of Ion1

This source was initially reported in V10b, and here we briefly summarize and report on new observational constraints, and compare it with the offset Uem sources discussed above.

SED. The source has not been detected in the F225W, F275W, and F336W channels of HST/WFC3 observations in the GOODS-S (down to 26.3, 26.4, 26.1 at 5σ for point-like objects; Windhorst et al. 2011). It lies in the B-band dropout selection scheme, with (B − V) = 1.70 and (V − z) = 0.40 (V09), and has a blue UV spectral slope, having an (i − z) = −0.015, and a β = −2.09 ± 0.16 derived from Castellano et al. (2012). It is worth noting that this source has the bluest (B − V) color among the available 17 LBGs with 3.7 < z < 3.9 and comparable UV slope. This is consistent with the fact that the B₄₃₅ band is probing the rest-frame interval λ ≲ 1020 Å, in which the LyC is contributing to the observed flux (an S/N of 10 is measured in this band) and reduces the (B − V) color. The probability that zphot > 3.4 is ≃100%. It has not been detected in the new 4 Ms Chandra X-ray observations, for which we set a 1σ upper limit of L_X < 3 × 10⁴² erg s⁻¹, nor in the 24 μm Spitzer/MIPS observations (Santini et al. 2009). The SED is fully comparable to that of a star-forming galaxy, in particular from the SED fitting we obtain: SFR ≃ 50 M_☉ yr⁻¹, stellar mass of 2.3 × 10⁹ M_☉, E(B − V) ≲ 0.1 (see Figure 7).

Zoom In Zoom Out Reset image size

Optical spectrum. Optical spectra from Keck/DEIMOS and VLT/VIMOS have been obtained, although the latter is too shallow and the S/N prevents us from adding information. From the Keck spectrum it is clear that the Lyα line is not in emission, the continuum break is evident together with the [Si iv]1393.8–1402.8 and [C iv]1548.2–1550.8 absorption lines, at redshift 3.795. Also [C ii]1335.1 in absorption seems to be present. The other UV absorption lines like [O i]1302.2, [Si ii]1260.4, and [Si ii]1526.7 are not detected (see Figure 7).

UV and B-band rest-frame morphology. At 1700 Å/1900 Å and B-band rest-frame wavelengths, the galaxy shows a resolved and compact shape, with a half-light radius of 0.9 kpc (physical size) in both the z₈₅₀ and H bands (the latter from CANDELS), respectively (see Figure 7).

The hypothesis that it is a faint AGN (L_X < 3 × 10⁴² erg s⁻¹) is still open. However, the spectral features in the optical spectrum and the shape of the SED from the U band to the far-infrared (Spitzer/MIPS 24 μm) are fully compatible with a star-forming and relatively low-mass galaxy. In the following we assume that it is a star-forming galaxy.

The LyC and the nLyC emissions (namely, U and i₇₇₅ bands) are spatially aligned within the errors, Δθ < 01. As discussed in V10b, the probability of a superposition by a lower redshift source within a circle of 01 radius centered on Ion1 is less than 0.1% (adopting the U-band number counts of Nonino et al. 2009 down to the observed U-band magnitude). Moreover, Ion1 is isolated and there is no obvious trace of a close companion in the high spatial resolution HST/B₄₃₅ image.

The LyC detection at λ < 830 Å implies a relatively transparent line of sight free from LLSs. From Equation (2), we derive a minimum f_{esc, rel} = 82% assuming (L1500/L_LyC)_int = 7 and an IGM transmission of 0.575 (maximum value for our U-band filter of the 10,000 realizations at z = 3.8). Given its E(B − V) ≲ 0.1 and adopting the Calzetti extinction law, we have f_esc > 56% (Calzetti et al. 2000). Assuming an intrinsic value for (L1500/L_LyC)_int = 3, the escape fractions are f_{esc, rel} > 35% and f_esc > 24%.

6.2. How Does Ion1 Compare With the Rest of the LBG Population ?

The observation of the LyC in distant galaxies is a difficult task because several attenuations occur: in the galaxy itself and CGM (by dust and neutral hydrogen gas) and along the intergalactic travel (Lyα forest, LLSs, and DLAs). A further complication is related to the possible presence of foreground (lower-z) superimposed sources that can mimic the LyC emission of the background galaxy. It is therefore crucial to establish whether they are genuinely associated with the redshift of the main LBG.

In this work, we have discussed three diagnostics that can be used to constrain the redshift of these offset and faint sources. To this end, the deep and high angular resolution multi-wavelength images available from the GOODS and CANDELS surveys have been exploited. In particular, starting from the sample of U-band emitters identified in V10b, we (1) calculated accurate photometric redshifts, (2) considered the IGM radial and transverse absorption (simulating UV colors and comparing those of LBGs and their U-band emitters), and (3) provided new constraints on their nature by calculating the f_esc quantity under conservative assumption. These three analyses suggest that none of offset U-band sources are actual sources of LyC emission from the LBGs at z > 3.4, but are instead foreground interlopers. This strengthens the results of V10b, i.e., the median f_esc quantity is very small for L > L* galaxies, or assuming a bimodal distribution for the f_esc, the higher values are rare.

At present, the offset candidates reported in literature by S06, I09, and N11 need spectroscopic redshift confirmation and are difficult to access if deep multi-wavelength high angular resolution imaging is not available.

Moreover, we note that in order to avoid biased measures, the calculation of the f_esc quantity in the case of spatially offset emission must be performed carefully considering quantities arising from the same physical region. We argue that the measurements reported by S06, I09, and N11 are affected by this problem and need further investigations.

The current situation is far from clear. The number of bona fide LyC detections at z > 3 is very small and not statistically significant.

We have discovered one good candidate (named Ion1) that is currently the highest redshift galaxy known with direct LyC detection. Its LyC emission at λ < 830 Å (probed by the U band) is aligned with the source detected in the UV nLyC, i.e., no offset is observed with this resolution. Apart from the three cases reported by Inoue et al. (2011) with aligned LyC emission that need population III stars to explain their observed F1500/F_LyC flux ratios (even smaller than 1), the Ion1 emission is easily explainable with standard stellar population and an f_esc > 25%. If the AGN component is negligible, as seems to be the case from the X-ray and spectroscopic data, it would be the most promising stellar ionizer that may resemble those responsible for H i re-ionization.

It is worth noting that a high value of the escape fraction would correspond to faint nebular emission (e.g., Robertson et al. 2010). In this case the Lyα line is not in emission, despite the fact that the source shows a blue UV spectral slope (β = −2.09), relatively weak interstellar absorption lines, and compact morphology, all characteristics that positively correlate with the strength of the Lyα emission line (e.g., Shapley et al. 2003; V09; Pentericci et al. 2010; Balestra et al. 2010). It is not clear if the absence of Lyα is connected to the high value of the f_esc or due to other effects like the dust attenuation, that however, would be small, giving the blue UV slope.

The derived stellar mass is 2.3 × 10⁹ M_☉ and seems to fit with recent model predictions such that f_esc decreases steeply with increasing halo mass (Yajima et al. 2011; Razoumov & Sommer-Larsen 2010). However, the estimated SFR ≃ 50 M_☉ yr⁻¹ is in contrast with the predictions of the same authors. Conversely, the SFR agrees with the predictions of Gnedin et al. (2008), in which the f_esc positively correlates with the SFR, but the observed mass appears too low. A measure of the wind velocity would be also important in this respect since galactic outflows can form holes in the ISM that allow ionizing radiation to escape directly to the IGM (Fujita et al. 2003; Heckman et al. 2011; Overzier et al. 2009). It is worth noting, however, that a statistical significant sample of LyC emitters would be necessary to perform more reliable comparison with the model predictions.

We also note that Ion1 is not an LAE, therefore it is important to complement the LAE surveys with sources like this. Clearly, sources like Ion1 placed at z > 7 would be extremely difficult to detect spectroscopically with present facilities. However, since at z > 7 the direct measure of LyC is not feasible, sources like Ion1 are the only viable ways, we have to investigate the mechanism that allows the ionization radiation to escape, i.e., to address the interplay between the f_esc quantity and the non-ionizing UV features like the strength of the nebular emission lines (e.g., oxygen, Balmer, and Lyman lines), UV slopes, the nature of the stellar populations, geometry, winds, etc.

We thank the anonymous referee for useful comments. We acknowledge financial contribution from the agreement ASI-INAF 1/009/10/0 and from the PRIN MIUR 2009 "The Intergalactic Medium as a probe of the growth of cosmic structures." We thank A. K. Inoue for providing us the transmissions of the IGM and the spectral template of Pop III stars.

Figure 8 shows the HST/ACS images of the offset U-band LyC emitter candidate that we have identified at 04 from LBG HUDF J033236.83−274558.0 at z = 3.797 (its celestial coordinates are R.A. = 03:32:36.85, decl. = −27:45:57.6). This pair has already been discussed by V10a, who argued that the LyC candidate is most likely a low-redshift interloper. Here, we add another piece of evidence in support of this interpretation. We have complemented the ACS optical photometry, which takes advantage of the HUDF images (Beckwith et al. 2006), with the new WFC3/IR data obtained as part of the CANDELS project (Koekemoer et al. 2011; Grogin et al. 2011) in the Y (F105W), J (F125W), and H (F160W) bands. The SED of the Uem from 0.5 to 1.6 μm is blue, essentially consistent with a flat spectrum, i.e., f_ν∝ν⁰, except in the J band where the source is ≃0.43 mag brighter (at 10σ), namely, U = 28.63 ± 0.2, B₄₃₅ = 28.59 ± 0.05, V₆₀₆ = 28.31 ± 0.03, i₇₇₅ = 28.31 ± 0.03, z₈₅₀ = 28.23 ± 0.05, Y = 28.31 ± 0.07, J = 27.88 ± 0.04, and H = 28.32 ± 0.06. Using the HST photometry alone, with or without the U band, the photometric redshift remains unchanged, i.e., z_phot ≃ 3.8.

Zoom In Zoom Out Reset image size

At the redshift of the LBG, z = 3.797, the J band probes the rest-frame spectral range λλ ∼ 2295, 2918 Å  where there are no strong emission lines, except for the possible [Mg ii] 2797,2803 P Cygni feature, which, however, has not been observed to yield any net emission in star-forming galaxies at redshift up to z ∼ 2 (e.g., Weiner et al. 2009; Rubin et al. 2010; Giavalisco et al. 2011). The simplest and, in our view, most convincing interpretation is that the U-band source is at lower redshift and that the J-band flux is boosted by an emission line (or more than one), most likely [O ii] or Hβ and [O iii].

All photometric redshift solutions lie at z < 3. The required rest-frame equivalent width of a single line to boost the J-band flux of 0.4 mag is 670 Å if Hα at z ≃ 0.9 (L_Hα = 1.5 × 10⁴⁰ erg s⁻¹), or 383 Å if [O ii]3727 at z ≃ 2.5 ($L_{[{\rm O}\,\mathsc{ ii}]}=1.9 \times 10^{41} {\rm \,erg\,s^{-1}}$).

Also the complex of the three lines Hβ, [O iii] 4959, and [O iii] 5007 can fall in the suitable wavelength interval. In the case of Hα, we would detect [O ii]3727, Hβ, and [O iii] 4959–5007 in the spectrum of the LBG, that is not the case. Similarly, in the case of Hβ and [O iii] complex, the [O ii]3727 line would be present in the spectrum of the LBG at ∼9700 Å. However, in this latter case it may be too faint to be detected. Therefore, we exclude the Hα possibility only and keep the other two configurations.

Given the response curve of the WFC3/Y/J/H filters, in order to have the above complex of three lines or the single [O ii]3727 line in the J filter only (green shaded region of Figure 8), the redshift of the source would be 1.45 < z < 1.8 or 2.25 < z < 2.7, respectively.

As mentioned above, the required rest-frame equivalent widths are high (larger than 300 Å); however, we note that ultrastrong emission line (compact) galaxies with these features have been recently detected (Kakazu et al. 2007; Hu et al. 2009; Izotov et al. 2011 and more recently van der Wel et al. 2011). Our source would extend their findings to a more extreme limit of luminosity, i.e., a rest frame absolute B₄₃₅-band magnitude of −16.86. Going a bit more into the details, we note that the SED shows a significant decrease of flux in the U and B₄₃₅ bands (see Figure 8), that would not be consistent with the redshift z ∼ 1.6, which corresponds to λ > 1200 Å. Conversely, in the case of [O ii]3727 at z ∼ 2.5, these bands are affected by the IGM and the Lyman limit attenuation. Indeed the best photometric redshift solution is zphot ∼ 2.5. Therefore, we argue that the compact source described here is at z < 3 and is a probable strong oxygen emitter (similar to those recently identified in the CANDELS survey by van der Wel et al. 2011), that mimics an LyC emission of the background LBG.