The Lyman Continuum Escape Fraction of Emission Line-selected z ∼ 2.5 Galaxies Is Less Than 15%

Star-forming galaxies (SFGs) likely reionize neutral hydrogen in the early universe (see the review in Loeb & Barkana 2001), when quasars are not sufficiently numerous to contribute significantly to the ionizing background (Ricci et al. 2017; cf. Giallongo et al. 2015). Verifying this assumption by directly measuring the ionizing output of Lyman continuum (LyC; λ < 912 Å) is impossible—the IGM effectively attenuates all LyC flux emitted along the line of sight to z > 6 redshift galaxies. Instead, the ionizing output of high-redshift SFGs must be constrained by surveys of low-redshift analogs, or indirectly (e.g., Jones et al. 2013). Criteria for pre-selecting LyC emitting candidates from among the class of all SFGs that produce LyC are crucial for such studies—large, blind, deep surveys are infeasible with the Hubble Space Telescope (HST), the only telescope currently capable of obtaining high-resolution LyC imaging and spectroscopy.

Previously, pre-selection was made on actively SF galaxies (Schaerer 2003; young, massive stars emit copious ionizing radiation, Q_H > 10⁴⁷ s⁻¹;). Studies with the HUT (Leitherer et al. 1995) and the HST SBC (e.g., Malkan et al. 2003; Siana et al. 2010) do not detect escaping LyC. Large archival studies of LyC emission from SFGs (Cowie et al. 2009) and Hα-selected emission line galaxies (Rutkowski et al. 2016) have generally reported non-detections, likely indicating that the strong star formation may be conducive to LyC escape, but does not guarantee it. Until recently, few LyC leakers were known; local starbursts Tol 0440-381, Tol 1247-232, Mrk 54, and Haro11 (e.g., Leitherer et al. 2016; Puschnig et al. 2016) and, at z ≳ 2, fewer than ∼10, UV-selected SFGs (e.g., de Barros et al. 2016; Mostardi et al. 2016; Smith et al. 2016).

Recently, five (of five galaxies targeted) z ≲ 0.3 compact (r_e ≲ 1 kpc) SFGs selected for their anomalously high nebular oxygen ratios (O₃₂ = [O iii]λ5007 Å/[O iiλλ3727,3729 Å] > 5) have been confirmed as LyC leakers (f_esc ∼ 5%–15%; Izotov et al. 2016a, 2016b), with an additional 2–3 compact O₃₂ galaxies predicted to be LyC leakers based upon their Lyα profiles (Verhamme et al. 2017). Furthermore, Naidu et al. (2017) applied an F275W–F336W color selection to identify three SFG LyC leakers at z ≲ 2 in the HDUV (PID: 13779; PI: P.Oesch), each with O₃₂ ≳ 3. The success rate of LyC detection in O₃₂ emitters makes this nebular-line diagnostic appealing for pre-selection. Here, we investigate that potential utility. In Section 2, we discuss the selection of ELGs using HST IR grism spectroscopic catalogs combined with rest-frame UV–optical imaging in the CANDELS fields. In Section 3, we present new measurements to the absolute escape fraction, f_esc, for these ELGs. We assume ΛCDM cosmology with Ω_m = 0.27, Ω_Λ = 0.73, and H₀ = 70 km s⁻¹ Mpc⁻¹ (Komatsu et al. 2011).

The HST WFC3/IR grism has been remarkably successful for surveying SFGs with strong emission lines at z ∼ 1–3, as the [O ii]λλ3726,3729 Å doublet, a well-calibrated signature of star formation (Kewley et al. 2004), can be detected with the G141 grism (λ_c ≃ 1.4 μm) in 2 ≲ z ≲ 3.5 SFGs. The G141 cannot resolve this doublet; thus, to avoid potential line misidentification of a candidate [O ii] emission line, a photometric redshift is critical. In the HST CANDELS fields, the broadband UV–optical SED for galaxies is well sampled, ensuring the robust identification of z ≳ 2 [O ii]-emitters. Unfortunately, rest-frame, broadband LyC imaging is not available across the entire survey footprint. Thus, in our search here for LyC emission from z ∼ 2.5 SFGs, we are limited to probing ∼40% of the area in GOODS-South and GOODS-North.

Within these regions, we select [O ii]-emitters identified by the 3D-HST G141 grism survey. We selected ELGs requiring (1) (S/N)_{[O ii]} > 3 and (2) within the redshift range 2.38 < z < 2.9, where z ≡ "z_best" measured by Momcheva et al. (2016) using both grism and broadband photometry. The lower-redshift limit of this sample is fixed to ensure the broadband (WFC3/UVIS F275W) imaging is strictly sensitive to rest-frame LyC emission. In total, we select 208 [O ii]-emitters (74, 109, and 25 in ERS, GOODS-N, and UVUDF, respectively), with a mean (median) SFR = 8.0 (3.9) M_⊙ yr⁻¹ and stellar mass M_⋆ ≃ 10^9.9(10^9.6), respectively. Included in this sample are 13 ELGs that were also considered in the unpublished LyC survey in the ERS field presented by Smith et al. (2016).

We note that within a narrow redshift range the G141 grism is simultaneously sensitive to [O ii] and [O iii]. Thus, we select a second, independent sample of SFGs requiring: (1) 2.25 < z < 2.31; (2) (S/N)_{[O iii]} > 3; and (3) (S/N)_Hβ > 1.5.

Zoom In Zoom Out Reset image size

This redshift range implies that the sample's LyC photometry could include a contribution from non-ionizing emission in the bandpass. As illustrated in Figure 2, the F275W throughput, T, ≳1% at λ < 3086 Å (910 Å at z = 2.39). Only in the case of zero attenuation by intervening neutral gas and dust (i.e., f_esc ≡ 100%) will the contribution redward of the Lyman edge to the F275W photometry be negligible (<0.5%). The contribution by non-ionizing photons to the measured ionizing flux will introduce a systematic uncertainty to f_esc strictly less than unity measured using the broadband method we apply in Section 3. Any candidate LyC leakers identified in this sample must be considered tentative pending spectroscopic follow-up.

Zoom In Zoom Out Reset image size

We identify 41 O₃₂ emitters (22 and 19 in the GOODS-N and ERS, respectively; Figure 1). For the measurement of O₃₂, we require [O iii]λ5007, which we derive from [O iii]λλ4959,5007 Å reported in 3D-HST catalogs, applying a uniform correction that assumes an intrinsic ratio of λ5007/λ4959 = 2.98 (Storey & Zeippen 2000) to correct for the contribution of [O iii]λ4959. Of these O₃₂ emitters, 13 are identified with O₃₂ > 5. By comparison with the [O ii]-emitter sample, these ELGs have a similarly high mean (median) SFR = 6.8 (4.1) M_⊙ yr⁻¹ and moderate stellar mass M_⋆ ≃ 10^10.0(10^9.7).

In the following analysis of both samples, we use publicly available F275W imaging mosaicked by the individual survey teams (see Table 1). In GOODS-North, the HDUV team has prepared public mosaics "v0.5" combining five of eight HDUV pointings with CANDELS-Deep data. We note that the WFC3/UVIS is susceptible to significant (∼50% losses) charge transfer inefficiencies. To mitigate this, UVUDF and GOODS-North imaging programs (referenced in column 4 of Table 1) included a ∼10e⁻ post-flash to minimize charge losses. In the case of the ERS, these data were among the first data obtained with the then newly installed WFC3 and were minimally affected by the CTE (see Smith et al. 2016 for details). In preparation for analysis, we extracted 12 × 12 postage stamps from the science and associated rms maps centered on each ELG. For uniformity, all stamps were rebinned to a common pixel frame of 009 pix⁻¹, the coarsest scale for which mosaics are available.

Table 1.  Archival F275W imaging

Field	Survey	Areaa	Survey Depth (3σ)b	References	NO ii	N${}_{{{\rm{O}}}_{32}}$	N${}_{{{\rm{O}}}_{32}\gt 5}$
GOODS-North	CANDELS-Deep	120	27.8	Koekemoer et al. (2011)	52	7	4
	HDUVc	70	27.9	d	57	15	4
GOODS-South	UVUDF	4.5	28.2	Rafelski et al. (2015)	25	⋯	⋯
	ERS	60	26.5	Windhorst et al. (2011)	74	19	5

Notes.

^aApproximate area in square arcminutes. ^bWe report published point-source completeness limits [AB mag] but estimate the depth from the sky variance for the HDUV GOODS-N public mosaic. ^cFor HDUV, mosaics are available at http://www.astro.yale.edu/hduv/DATA/v0.5/. ^dHST Program 13872.

Download table as:  ASCIITypeset image

Broadband imaging surveys readily make differential measurements of the ionizing (LyC) to non-ionizing (UV, measured at λ_rest ≃ 1500 Å) luminosity from galaxies. At high redshift, LyC from young stars within galaxies will be attenuated by neutral gas and dust in the ISM, as well as by neutral H i in the IGM along the line of sight. This partly motivates a definition of the relative escape fraction, following Steidel et al. (2001):

$$\begin{eqnarray}&&{f}_{\mathrm{esc},\mathrm{rel}}=\displaystyle \frac{{({L}_{\mathrm{UV}}/{L}_{\mathrm{LyC}})}_{\mathrm{int}}}{{({L}_{\mathrm{UV}}/{L}_{\mathrm{LyC}})}_{\mathrm{obs}}}\cdot \exp [{\tau }_{\mathrm{IGM}}],\end{eqnarray} \tag{ 1 }$$

τ_IGM is the (redshift-dependent) IGM attenuation of LyC by neutral H i, typically modeled on measurements from absorption line surveys toward high-redshift bright quasars (e.g., Fardal et al. 1998). The intrinsic UV-to-LyC ratio must be modeled for each galaxy individually, but typically ranges between 2 and 10 for SFGs with ages less than ∼10⁸yr. If the magnitude of extinction due to dust in the ISM can be estimated from the SED, then the absolute escape fraction can be directly related to f_esc,rel as

$$\begin{eqnarray}&&{f}_{\mathrm{esc}}={f}_{\mathrm{esc},\mathrm{rel}}\times \exp [-{\tau }_{\mathrm{UV},\mathrm{dust}}].\end{eqnarray} \tag{ 2 }$$

We measured ionizing and non-ionizing photometry in the F275W and F606W postage stamps, respectively, using Source Extractor (Bertin & Arnouts 1996) in dual image mode, with the F606W as the detection image.¹⁰

The median F275W S/N for all ELGs is consistent with a statistical non-detection ($\langle {SNR}\rangle =0.12$), as measured within each ELG's corresponding F606W aperture defined in source extraction. A visual inspection of all F275W stamps confirms that no individual ELGs are LyC leakers, including those [O ii]-emitters in the Smith et al. (2016) sample. Note that for the median F606W (rest-frame UV) continuum m = 25 AB of this sample, the surveys limit f_esc,rel ≲ 2% in the deepest (UVUDF) and f_esc,rel ≲ 12% in the shallowest (ERS) mosaics. Smith et al. report a detection of f_esc = 0.14% for the sample that overlaps with the [O ii]-emitter sample. Within the HDUV field, no LyC leakers have been previously identified.¹¹

To measure f_esc, we apply the stacking procedure defined in Siana et al. (2010), summing over all pixels in the F275W and F606W stamps associated with the F606W-defined segmentation map. Furthermore, we sum in quadrature the associated errors from the error maps, applying a (small) correction for correlated noise introduced in the rebinning of the error maps where necessary (Casertano et al. 2001). This stacking yields no statistically significant detections of LyC leakers. A visual inspection of the associated stacked LyC frames (combined using IRAF imcombine; Figure 3) reveals no perceptible LyC flux within an aperture defined by the non-ionizing UV image stack. Here, we have cleaned all LyC stamps before stacking, using the segmentation maps, to replace pixels not associated with the ELG with randomly assigned pixel values consistent with the sky background measured within each stamp. In Table 2, we report f_esc for each stack as upper limits.

Zoom In Zoom Out Reset image size

Table 2.  Measured UV/LyC Flux Ratios: Upper Limits to f_esc

Selection	Nobjs	Δ(z)	Observed fν,LyCa	Observed fν,UV	IGM corr. UV/LyC	${f}_{\mathrm{esc},\mathrm{rel}}^{\mathrm{LyC}}$ b	${f}_{\mathrm{esc}}^{\mathrm{LyC}}$
[O ii]	208	2.38 < z < 2.9	0.45 ± 0.27 (1.6σ)	9.12	>41.49	<7.0%	<5.6%
All O32	41	2.25 < z < 2.31	0.16 ± 0.140 (1.1σ)	4.53	>38.44	<7.8%	<6.7%
O32 > 5	13	2.25 < z < 2.31	−0.02 ± 0.090 (−0.16σ)	1.15	>15.81	<18.9%	<14.0%

Notes.

^aFlux densities reported here in μJy. ^bWe assume (L_UV/L_LyC)_int = 3; italicized entries indicate non-detections and should be interpreted as limits.

Download table as:  ASCIITypeset image

In rest-frame UV morphology, these galaxies are compact. For reference, an aperture defined to include 90% of the segmentation pixels common to all galaxies has an area of ∼0.3 square arcseconds (physical radius, r ≲ 4 kpc), in good agreement with the measurements of z ∼ 2 galaxy sizes from Shibuya et al. (2015). Many galaxies (>70%) do show faint irregular UV features. Though we used the segmentation map (defined by the rest-frame UV morphology) to define pixels to include in the stacking of each galaxy, in principle these asymmetric low surface brightness features could be lost to the sky when stacked.

We measure f_esc,rel, correcting each galaxy for IGM attenuation using the correction factor from the piecewise parameterization of the redshift distribution and column density of intergalactic absorbers (see Haardt & Madau 2012). For reference, exp[τ_IGM] = 1.72 (2.56), at z = 2.29 (2.56), the median redshift of the O₃₂ ([O ii])-selected samples. Assuming (L_UV/L_LyC)_int = 3, appropriate for a young (∼10⁷ yr), solar-metallicity stellar population (Rutkowski et al. 2016), we measure f_esc,rel ≲ 7.0%, 7.8%, and 18.9% (1σ) for the [O ii]-, all O₃₂-, and high O₃₂-selected samples. We measure f_esc correcting for dust attenuation for each galaxy individually. We measure τ_UV,dust assuming a Calzetti et al. (2000) reddening law (R_V = 4.05) and calculating stellar E(B–V) from the best-fit A_V measured from the broadband SED by Skelton et al. (2014).

We report f_esc < 5.6% for the [O ii]-selected sample. Note that the upper limit on f_esc measured for random sub-samples of [O ii]-emitters drawn exclusively from the individual (unbinned) mosaics scale approximately as ${N}^{-1/2}$, as expected from purely Poisson statistics. Thus, in future work, using a re-reduction of all available F275W imaging in the CANDELS fields to improve the size of the [O ii]-selected sample emitters, we will test for variations in f_esc in sub-samples selected on, e.g., UV luminosity or inclination.

For the full O₃₂-selected sample, we measure f_esc < 6.7%; for the O₃₂ > 5 sample, f_esc < 14%, consistent with the expectation for Poissonian statistics if the extinction correction is appropriately re-normalized to reflect the higher average extinction reported for the O₃₂ > 5 sample. The mean IGM transmission for the O₃₂- and [O ii]-selected samples differs by a factor of ∼1.5, the O₃₂-emitters are intrinsically more luminous (∼3×), and the possibility of a non-negligible contribution from non-ionizing flux in the F275 bandpass (see Section 2) makes a direct comparison of f_esc upper limits for these samples more difficult.

If these SFGs are analogs to the high-redshift sources of reionization, the measured upper limits can be informative. First, the 1σ upper limit to f_esc measured for [O ii]-emitter sample is inconsistent with the threshold of f_esc ≳ 13% required if high-redshift SFGs reionize the universe (see Robertson et al. 2015) compatible with the independent constraints on the ionization history of the IGM from the CMB (the electron scattering opacity, τ_es; see Planck Collaboration et al. 2016) and QSO absorption line studies (Mesinger & Haiman 2007). This tension is alleviated considering the 3σ f_esc upper limit and noting that dwarf galaxies less massive than these ELGs (with median M ≃ 10^9.5–10 M_⊙) are expected to contribute most significantly to reionization (Wise et al. 2014; Robertson et al. 2015).

Note Rutkowski et al. (2016) measured, for (Hα-selected) z ∼ 1 SFGs, f_esc < 4% (3σ). Selecting more distant SFGs using the same grism spectroscopy here, we are more sensitive to intrinsically brighter line luminosities, ∼4 brighter at z ∼ 2.5 than z ∼ 1, though intrinsically we can expect [O ii]/Hα ≲ 1 (≃0.5 at z ≃ 0.1; Mouhcine et al. 2005). As such, the average SFR for the [O ii]-selected sample is ∼2× that of the Hα sample in previous work, though the median SFR is measured for 3D-HST sources from the broadband SED in contrast to, e.g., Rutkowski et al. (2016), which used the extinction-corrected Hα luminosity. Thus, we caution any strict interpretation of the f_esc upper limits derived here for z ≃ 2.5 SFGs and previous work at z ≃ 1 as evidence for an evolution in f_esc.

Our observations are not deep enough to detect f_esc ∼ 10%, typical of the low-redshift LyC emitters (Izotov et al. 2016b), which have comparably high nebular emission line ratios or similar star formation rate surface densities, Σ_SFR.¹² Our upper limit on f_esc derived for the high O₃₂ ELGs is marginally consistent (f_esc,rel (3σ) ≲ 0.57) with the detection of LyC in a similar galaxy (ion2; ${f}_{\mathrm{esc},\mathrm{rel}}={0.64}_{-0.1}^{1.1}$) studied by de Barros et al. (2016), though difficult to reconcile with the high f_esc > 50% observed for the z = 3.2 strong O₃₂ (≳10%) emitter observed by Vanzella et al. (2016).

We call attention to f_esc measured for the small number of SFGs identified with high O₃₂ emission and the implication for the contribution of their high-redshift analogs to reionization. Generally, reionization proceeds when a sufficient ionizing background can be maintained by either a large number of relatively inefficient LyC leakers or relatively fewer emitters that efficiently source LyC. The number of ionizing background photons in a cosmological volume is proportional to f_esc × n_SFG, where n_SFG is the volume density of SF galaxies and f_esc ≃ 10% is necessary for reionization. However, not all SF galaxies are LyC leakers. In fact, it is well established that the general population of SFGs have an escape fraction ≪10% (Siana et al. 2010; Grazian et al. 2016), and only the extreme O₃₂ galaxies appear to meet the requisite f_esc (Izotov et al. 2016b). If f_leak is the fraction of SF galaxies that are LyC leakers, then the previous relationship for the number of ionizing photons, N_ion, can be rewritten as N_ion ∝ f_esc × f_leak × n_SFG. In the WFC3 spectroscopic parallel survey (WISP; Atek et al. 2010), sensitive to both [O ii] and [O iii] emission at 1.4 ≲ z ≲ 2.3, 50% of the cataloged galaxies are detected in both oxygen lines (Ross et al. 2016). Only ∼4% of these sources are O₃₂ > 5 emitters. With the upper limits presented here, assuming f_leak ∼ 4% and that this fraction does not evolve substantially to z ∼ 6, such extreme objects would not support reionization. High-redshift (z > 3) SFGs do exhibit, on average, an enhanced ionization state relative to low-redshift SFGs (e.g., Stanway et al. 2014), inferred from the [O iii]/Hβ ratio. Recently, Faisst (2016) modeled this increased ionization state with redshift to predict the evolution of the escape fraction evolution with redshift of O₃₂-emitters and found such galaxies to be nearly sufficient to reionize the universe at z ∼ 6. Clearly, direct measurement of the median escape fraction for strong emitters (O₃₂ > 5) with HST at z < 3 is critical. This, in combination with the direct measure of the evolution of the number density of such extreme O₃₂ galaxies toward the epoch of reionization (z > 7), a key result for JWST, will ultimately determine whether such sources may reionize the universe.

We have combined archival high-resolution HST UV imaging in the rest-frame LyC for z ∼ 2.5 galaxies in the CANDELS deep fields, selected on the presence of nebular oxygen emission lines in the 3D-HST IR grism spectra. We do not detect LyC escaping from [O ii]- or O₃₂-selected emitters individually.

We stack the individual non-detections, and measure for each stack upper limits to the absolute escape fraction less than 5.6%, 6.7%, and 14% (1σ), respectively. Our limits on f_esc (3σ) for such relatively massive galaxies do not rule out the possibility that SFGs are able to sustain reionization. However, whether at z ≳ 2, strong star formation and high O₃₂ ratios alone are indicative of significant LyC escape remains uncertain. Furthermore, we note that at z ∼ 2 the class of galaxies with extreme O₃₂ ratios remain exceedingly rare. In order for galaxies to be able to sustain reionization, SFGs must evolve substantially from z ∼ 6 to present, such that at high redshift most have such highly ionized ISM conditions indicated by the high O₃₂ ratio. Such galaxies will be prime targets for JWST at z > 3, and future grism surveys and further constraints on LyC emission from lower-redshift O₃₂-selected ELGs will be important for calibrating the evolution of LyC toward the epoch of reionization. Deep HST surveys of large volumes at intermediate redshift will be necessary to obtain the large sample sizes of strong O₃₂-emitters necessary to determine whether LyC escape is linked to these observable parameters such that their contribution can be meaningfully extrapolated to the epoch of reionization probed by JWST.

This research was supported by NASA NNX13AI55G, HST-AR-12821.01, and HST-GO-13352 using observations from NASA/ESA HST, operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. M.H. acknowledges the support of the Swedish Research Council (Vetenskapsrådet), the Swedish National Space Board (SNSB), and the Knut and Alice Wallenberg Foundation. This work was funded by NASA JWST Interdisciplinary Scientist grants to R.A.W., NAG5-12460 and NNX14AN10G from GSFC, and HST-AR-13877.001-A and HST-AR-14591.001-A.