Lyman Continuum Emission Escaping from Luminous Green Pea Galaxies at z=0.5

Compact starburst galaxies are thought to include many or most of the galaxies from which substantial Lyman continuum emission can escape into the intergalactic medium. Li and Malkan (2018) used SDSS photometry to find a population of such starburst galaxies at z~0.5. They were discovered by their extremely strong [OIII]4959+5007 emission lines, which produce a clearly detectable excess brightness in the i bandpass, compared with surrounding filters. We therefore used the HST/COS spectrograph to observe two of the newly discovered i-band excess galaxies around their Lyman limits. One has strongly detected continuum below its Lyman limit, corresponding to a relative escape fraction of ionizing photons of 20+/-2%. The other, which is less compact in UV imaging, has a 2-sigma upper limit to its Lyman escape fraction of<5%. Before the UV spectroscopy, the existing data could not distinguish these two galaxies. Although a sample of two is hardly sufficient for statistical analysis, it shows the possibility that some fraction of these strong [OIII] emitters as a class have ionizing photons escaping. The differences might be determined by the luck of our particular viewing geometry. Obtaining the HST spectroscopy, revealed that the Lyman-continuum emitting galaxy differs in having no central absorption in its prominent Ly{\alpha} emission line profile. The other target, with no escaping Lyman continuum, shows the more common double-peaked Ly{\alpha} emission.


Introduction-Finding the Population of Starburst Galaxies that
Reionized the Universe.
One of the most important events in the history of the Universe was the cosmic re-ionization. But our understanding of it remains limited, because we do not know where and when the ionizing photons originated. Although quasars are powerful enough sources to ionize local bubbles, they were too rare to have re-ionized the entire intergalactic medium (IGM) (Mitra et al. 2018;Parsa et al. 2018;Puchwein et al. 2019). The only remaining candidates which were sufficiently numerous are star-forming galaxies. Thus the question of re-ionization comes down to this: What ionizing radiation escapes from young star-forming galaxies?
Unfortunately, direct observations of ionizing photons escaping from galaxies are difficult, and when they have been done, they often yielded only upper limits. In particular, we first used HST's UV imaging sensitivity to search for Lyman limit continuum (LLC) leaking out of star-forming galaxies at z~1 (Malkan et al. 2003). Further FUV observations of large samples of emission-line galaxies have also set limits on the relative LLC escape fraction (compared to the expected intrinsic Lyman break strength) of <2% at z=1 (Rutkowski et al. 2016), <6% at z=2-2.5 (Rutkowski et al. 2017;Matthee et al. 2017), and <21% at z=3.08 (Siana et al. 2015;Japelj et al. 2017).
More recent works have begun to detect escaping LLC from galaxies at low redshifts (Izotov 2016a(Izotov ,b, 2018a and at higher redshifts (deBarros et al. 2016, Bian et al. 2017, Vanzella et al. 2018, Rivera-Thorsen et al. 2019. A recent spectroscopic survey of 124 z~3 galaxies has detected 15 of them (weakly) below the Lyman limit, specifically those with the strongest Lyman-α emission lines (Steidel et al. 2018). Converting the measurements into estimates of the intrinsic escaping Lyman continuum fluxes at high redshifts is complicated by the uncertainty in the corrections for IGM absorption.
Observations at lower redshift do not suffer absorption by intergalactic HI. However, until recently most EUV observations of local starbursts have also failed to detect escaping LLC. FUSE obtained one possible detection, of the dwarf starburst Haro 11 at f esc = 3.3% (Leitet et al. 2011). HST/COS has detected a 21% escape fraction in a starburst galaxy at z=0.23 (Borthakur et al. (2014), and two other dwarf starbursts with relative escape fractions of 21% (Leitherer et al. 2016). Corrected for internal dust extinction, these correspond to absolute escape fractions of 1--8%.
Simply targeting blue galaxies has not proved efficient-the key is to identify the subset of galaxies which is particularly likely to release ionizing photons. Recent work suggests that the best candidates are the most extreme young starbursts, which concentrate their energy release in very compact regions (Izotov 2018a,b). In the local Universe at least, these characteristics tend to be found in metal-deficient dwarf galaxies. One of their defining properties is very strong [OIII]-emission line doublets, which gives them the name 'green peas' (Cardamone et al. 2009;Izotov et al. 2011;Nakajima and Ouchi 2014 (Vanzella et al. 2020). Izotov et al. (2016aIzotov et al. ( , 2018b found that the latter is a necessary though not sufficient predictor, since high [OIII]/ [OII] correlates with stronger LLC escape, with substantial scatter. At z=3.2 another such green pea (EW([OIII])~1000Å) has a recent spectroscopic detection of its LLC which indicates a relative escape fraction of 64% (de Barros et al. 2016). A recent HST imaging study of the Lyman limits of 61 galaxies at z=3.1 (Fletcher et al., 2019)  measured, were 6--10 (Nakajima and Ouchi 2014). In the midst of these recent discoveries, we used a new search for green peas at redshift z~0.5 that were bright enough for FUV spectroscopic followup with the Hubble Space Telescope.

New [OIII]-Strong Emitting Green Pea Galaxies at z~0.5
The original "green pea" galaxies were discovered at redshifts around z~0.2, which shifts their strong [OIII]5007+4949 emission line doublet into the r bandpass (Cardamone et al. 2009). We have recently again used broadband photometry--pushing SDSS imaging to its limits--to extend the search for green peas out to z=0.6 (Li and Malkan 2018;LM18). Around this redshift, the [OIII] doublet is shifted into the i band. And if its equivalent width is higher than EW rest >200Å, it detectably brightens the entire broadband magnitude by several tenths of a magnitude relative to the flanking r and z photometric bands. There is of course a continuous distribution of [OIII] equivalent widths; we select this operational threshold because it is the starting point at which large low-resolution or photometric surveys work well.
LM18 had defined a photometric sample of 5169 extreme emission-line galaxies with strong i-band-excesses. These included contaminants, in which this excess is from strong Hα emission at z~0.15, and/or from an active galactic nucleus. Most of those are screened out by additional photometric cuts, including IR photometry from WISE. This narrowed the sample of photometric candidates down to 968. But spectroscopic confirmation was needed. With 7 nights of spectroscopy in 2018--19 with the Kast dual spectrograph on the Shane 3-m telescope at Lick Observatory, plus existing SDSS spectra, we have now confirmed that in a randomly selected sub-sample of 33 of these galaxies, the strong emission line in the i filter is, indeed, [OIII]5007+4959 from an extreme starburst, not an active galactic nucleus (Malkan and Chang 2020). Next, we discovered that these intermediateredshift [OIII] emitters are so extremely blue, that they are easily detected in the GALEX survey, in both its NUV and FUV filters. This means that their rest-frame continuum down to the Lyman limit is bright enough for spectroscopy with HST.

Target Galaxy Selection
If a galaxy's redshift is too low, its crucial Lyman limit wavelengths (LLC: λ 0 <912Å) lie in the dropping sensitivity of HST's UV-detecting instruments. If the redshift is high (z>2), the galaxies are extremely faint, and also suffer from an uncertain amount of intergalactic HI absorption. It turns out that a sensitive LLC search can be made at z~0.5.
Although the higher redshift makes a given galaxy fainter than the original "green pea" galaxies, this is compensated for by the fact that the COS spectral sensitivity of the G140L configuration is near its peak at an observing wavelength of 1250Å 1 .
We therefore selected from our new i-band excess sample the two green pea galaxies with the highest [OIII] luminosities, which are nonetheless representative of the entire class of z=0.5 green peas. Their higher redshifts (z=0.48, z=0.54) make J090334+174717 (denoted OIII-10 in LM18) and J020748+004735 (denoted OIII-17) easier targets than most previous lowredshift starbursts which have been searched for escaping LLC. For comparison, the detection of a relative LLC escape fraction of 21% in SDSS0921+4509 required 9 HST orbits (Borthakur et al. 2014), and comparable COS integrations were needed to weakly detect LLC in two of three other similar galaxies which were not green peas (Leitherer et al. 2016).

Spectroscopy
The rest-frame optical spectra of the two target starburst galaxies, from the Sloan Digital Sky Survey (SDSS, with spectrum ID's 0702-52178-416 for J020748+004735, and 5297-55913-0765 for J090334+174717, included in Figure   1), show absolutely no trace of any broad Hβ wings, [NeV] of any other indicator of Seyfert nuclear activity. J020748+004735 is mistakenly identified as a "QSO" in NED. That is a common Sloan misclassification because of the extremely blue broadband colors (Cardamone et al. 2009 All emission line fluxes are taken from tables published in LM18, after re-checking that they are correct measurements from the SDSS spectra.
Luminosity distances with the standard cosmology and given in the fifth column. Our two prime target galaxies listed in  (Ly et al. 2014(Ly et al. , 2016. Our two target galaxies have O/H abundances roughly a third of solar, and N/H is much more sub-solar. Finally, the redshifts of these targets are high enough to show--with SDSS optical spectroscopy--another characteristic associated with strong LLC leakage: clearly detectable MgII 2800Å emission (Henry et al. 2018).

Photometry.
An extensive range of additional observations are available to provide further diagnostics of these two galaxies. Table 2 presents broadband photometry in the UV, taken from GALEX 4 , in the optical, taken from the Sloan Digital Sky Survey 5 , and in the infrared from WISE 6 . The broadband spectral energy distributions of both targets are included as solid squares plotted in Figure 1. The second-to-reddest SDSS broadband filter is the i-band, which is enhanced relative to the surrounding r and z filters by the extremely strong [OIII]5007+4959 emission doublet flux. One of the targets, J090334+174717, also has SDSS spectroscopy extending into the near-infrared, providing measurements of additional key red emission lines.

<470
Both of our target galaxies have repeated solid GALEX detections in both the NUV and FUV channels. The observed NUV fluxes (bandpass runs from 1771-2831Å, with effective wavelength of 2270Å) were independently measured from multiple GALEX observations, entirely longward of the Lyman limits of these galaxies. The rest-frame 1450Å continuum fluxes measured by GALEX from PSF-fitting (neglecting any extended flux) are 10 and 32 µJy for the two target galaxies. We include WISE detections in bands W1, W2 and W3 (3.4, 4.6 and 12 microns), given in Table 2. The relatively red W1 -W2 colors of J090334+174717 and, especially, of J020748+004735 are produced by hot dust, as is seen in other extreme starburst galaxies with high equivalent widths of Hβ (Izotov et al. 2011).
Compared with the dwarf galaxies which have previously detected Lyman limit continuum leakage, the stellar masses of both of our target galaxies (~7*10 9 M⊙; LM18) are at the upper end of the distribution (Table 1). Their high rates of current star formation (~40 M⊙/year) are also exceptionally high, a factor of several times those of the most extreme green peas found at z~0.1 (Izotov et al. 2011;Jaskot and Oey 2013). Both of our targets have [OIII] luminosities exceeding 10 43 erg sec -1 , several times larger than the most luminous previously studied "green pea" galaxies at lower redshifts (LM18).

Ultraviolet Spectroscopy
In the Supplemental round of Cycle 26, we obtained 3 HST Guest Observing time to observe our two target galaxies with COS, for 3 orbits each. The spectra of the two galaxies were obtained in program GO-15471, in 2018 November and December 7 . First, short acquisition NUV imaging observations of one minute were obtained using the MIRROR A option (see Figure 5, in Section 3.3.5).
Then the remainder of the first orbit, and all of the remaining two orbits were devoted to long spectroscopic integrations with the G140L grating, through the PSA aperture, with the galaxy at Lifetime Position 4. The total spectroscopic integration times were about 8200 seconds per target, taken in ACCUM mode. The reduced, extracted 1-D spectra of the entire source were then binned combining groups of 7 pixels, giving 0.54Å final pixels, small enough to preserve the instrumental R~2000 resolution. The starlight continuum of each galaxy is very well detected longwards of their Lyman limits, including several strong absorption lines.

Lyman Limit Observations
We examined both the original individual 2-D and extracted, calibrated 1-D spectra for both target galaxies. In the full COS spectrum of J090334+174717 no flux is detected below the Lyman limit. However, significant flux is clearly detected beyond the Lyman limit of J020748+004735 (i.e. at observed wavelengths shorter than 1406Å). Before identifying this as genuine Lyman continuum leaking from the galaxy, it is imperative to rule out all other possible instrumental sources of spurious weak continuum signals that could artificially produce such an apparent flux. We already know that none of these problems occurred in the otherwise identical COS spectroscopic observation of J090334+174717.
We have accordingly made several tests of the reality of the Lyman limit continuum in J020748+004735. In particular, we were concerned that dark

Relative and Absolute Escape Fractions of Ionizing Photons
To determine the relative Lyman continuum escape fraction requires knowing the intrinsic flux drop from the red side to the blue side of the Lyman limit. The Lyman limit relative escape fraction can be defined as: This realistic model stellar atmosphere predicts that the intrinsic flux beyond the Lyman limit (900Å rest wavelength) is 0.3 times that of the 940Å continuum, and this is the standard assumed ratio in other Lyman limit studies (e.g., Reddy et al. 2016). Similar estimates would also have been obtained using almost any model of a population of recently formed stars. For example, LM18 fitted young stellar populations (age~70 Myr) to the optical spectra of both of these galaxies. As shown in Figure 1, these models predict a 900Å flux that is 0.28 times the 940Å continuum.

Lyman-α Emission lines
The COS spectra simultaneously cover the LLC, the FUV continuum redwards of the Lyman limit, and the complete Lyα line profiles. This removes uncertainties arising from calibration issues with separate observations, spatially extended emission, etc. Figure 4 shows zoomed in views of the Lyα emission line profiles from the two COS targets. The profile in J090334+174717 (right) has a central zero-velocity absorption, making it double-peaked. This profile is common in green pea galaxies (Henry et al. 2015), except that the blue peak is stronger than the red peak. There is no simple analytic form for these complex profiles; we fitted Gaussians merely to obtain a rough objective idea of the velocity spreads. J020748+004735 (left) is unusual in having no central dip, but instead a zero-velocity emission peak. If we attempt to characterize these complex profiles by a blue/red component separation velocity, as in Izotov et al. 2018b, it would be roughly 350 km/sec for J090334+174717, and <150 km/sec for J020748+004735. Gaussian in red has a rest-frame equivalent width equal to -21 Å and a full width half max of 5.3 Å. (1.3*10 -18 ) f + is the flux above the Lyman Limit, between rest wavelengths of 940-946Å. fis the flux below the Lyman Limit, between rest wavelengths of 880-912Å. In the last two columns, these fluxes have de-reddened for Milky Way extinction.
The Hβ emission line profile observed by SDSS is over-plotted on the same velocity scale. As expected for a small galaxy, it is spectrally unresolved, implying an intrinsic velocity FWHM of less than 150 km sec -1 .
This is similar to the narrow core of the Lyα line. However, Lyα has much broader wings (FWHM=900 km sec -1 ) which are presumably due to scattering in this optically thick resonance line. Comparing the observed Lyα fluxes (of 8.6 x 10 -15 and 2.2 x 10 -15 erg sec -1 cm -2 ) to the what is intrinsically predicted from the de-reddened Balmer line fluxes, assuming Case B recombination, gives Lyα escape fractions (Henry et al 2015) of 2.5% and 11% in J090334+174717 and J020748+004735, respectively. As will be discussed below, theoretical modelling has in fact predicted that larger Lyα escape fractions, and more symmetric centrally peaked Lyα emission line profiles, should be associated with higher Lyman continuum escape fractions (e.g. Dijkstra et al. 2016).

Spatial Distribution of UV Continuum from COS Imaging
Both target galaxies were spatially unresolved by GALEX, with its coarse angular resolution of 5 arcsec (FWHM). The SDSS images of J020748+004735 are virtually pointlike. J090334+174717 shows some faint extensions, which might indicate an interaction with a neighboring galaxy within a few Kpc. The features are confirmed in short-exposure NUV acquisition images from our HST visits which capture the bright central regions of the galaxies (see Figure 5). In particular, the UV surface density, and therefore the star formation surface density, is higher in J020748+004735. For both galaxies, all the detected UV flux was captured in the 2.5 arcsecond PSA in which the spectroscopy was obtained. The SDSS spectroscopic survey for Luminous Red Galaxies happened to observe a nearby galaxy with an absorption-line spectrum at a redshift of 0.542, SDSS J020748+004735, which may be in the same physical association in space.

Conclusions: Lyman Leakage from Green Pea Galaxies--Is It a Coin Flip?
The immediate goal of this observational study was to determine the incidence of escaping Lyman continuum photons from a newly discovered group of extremely luminous [OIII]-emitting galaxies at z~0.5. Having found that some but not all of them are LyC emitters, it is now interesting to compare their LyC emission with properties of other samples observed in the optical and UV which may have overlapping characteristics.

Difficulty of Selecting the Lyman Leakers from Optical Spectroscopy
The two green pea galaxies we observed are so similar that--without UV Thus a high ratio of O32 > 6 appears to be a necessary, but not sufficient, condition for f esc,rel to exceed 5%. This is consistent with the detections presented by Izotov et al. (2016a and2018b), and also at higher redshifts by  (Feltre et al 2018). This could also make EW 0 (Mg II) a predictor of escaping LyC photons. That hypothesis is consistent with our observations, but of course more data will be required to determine how reliable this prediction can be.

Predicting the Lyman leakers from COS observations
Although the previous observations did not give us any strong differences that allowed us to predict which of our starburst galaxies would be a detected Lyman continuum leaker, our UV observations with HST provide some possible hints: a) Intense current star formation. The detected Lyman leaker--J020748+004735--is somewhat bluer. Its far-UV flux within the COS PSA is ~one magnitude brighter than that of J090334+174717. This is reflected in the somewhat higher specific star formation rate we infer: the SFR normalized by the stellar mass of J020748+004735 is 2.4 times higher than that of J090334+174717. b) Compact Star formation. The detected Lyman leakers tend to have very compact UV continuum emission (e.g., Marchi et al. 2018;Izotov et al. 2018b). Both of our target galaxies are small, but the one with detectable Lyman continuum is more compact than the other one, which may be undergoing a merger. and Venkatesan 2017), as we observe, and is also seen at high redshift (Vanzella et al. 2020).
Although Lyman continuum leakage appears to be correlated with very strong [OIII] emission and these other observables, none are perfect predictors--there is still scatter which remains unexplained. There are many possible sources of stochasticity in the ISM of these galaxies. One plausible explanation is that although the class of galaxies with extremely strong [O III] emission includes many of the best producers of escaping ionizing photons, their escape may be anisotropic, as considered by Zackrisson, Inoue and Jensen (2013) and Fletcher et al. (2019), for example. As modelled by theoretical studies (e.g. Behrens et al. 2014), the escape might occur along ionized channels, but their geometry (bipolar, "picket fence", random holes, etc.) is not known. As long as our particular viewing angle of a star-burst geometry cannot be predicted in advance, the f esc that we observe may include an element of luck.
Our observed sample of z~0.5 green pea galaxies has only two UV spectra, but should be representative of the entire class of luminous strong [OIII]-emitters at z=0.5 surveyed by LM18. If we did not have any UV observations, we might guess a highly uncertain ~50% success rate of detecting leaking Lyman limit continuum from these z=0.5 green pea galaxies.
Nonetheless, this is too small a number on which to base a firm statistical conclusion about the probability of escaping ionizing continuum from the entire class.

Implications for Reionization at High Redshifts
There is growing evidence that the green pea phenomenon studied here (very strong [OIII] line emission) was more widespread in galaxies at higher redshifts. It is extremely rare in the low-z universe (Cardamone et al. 2009), but the ``knee" in the [OIII] line luminosity function brightens by 2.5 times from z=0.41 to z=0.83 (Ly et al. 2007). And the evolution continues strongly to higher redshifts. At z~1.5, ultra-strong [OIII] emitters are ~10 times more numerous than at z~0.1 (Maseda et al. 2018).
Similarly, Atek et al. (2011) used Hubble Space Telescope infrared grism spectroscopy to discover ~a hundred [OIII] emitting galaxies at z~2 (with rest equivalent widths from 200 to 400A), in only 180 square arcminutes. Atek et al. (2014) then quadrupled this galaxy sample, and demonstrated that the overwhelming majority of them are extreme starbursts. By z=3.5, stacked spectral energy distributions (SEDs) of thousands of U-dropout galaxies show that as the stellar mass of the average Lyman break galaxy drops from 10 9 to 10 8 Suns, the rest EW([OIII]5007+4959) rises from 400 to >2000Å, showing that 'green peas' are ubiquitous at z>3 (Malkan et al. 2017). Such high-ionization starbursts are evidently even more universal at z>6, as indicated by their IRAC-band excess flux due to extremely strong [OIII] emission in the broadband infrared filters (Smit et al. 2015). Thus, an extrapolation of the results in this work and others to high redshifts suggests that they are the best candidates for re-ionizing the universe.