Draft version July 28, 2017Preprint typeset using LATEX style emulateapj v. 01/23/15

Ly PROFILE, DUST, AND PREDICTION OF Ly ESCAPE FRACTION IN GREEN PEA GALAXIES

Huan Yang1,2, Sangeeta Malhotra2,3, Max Gronke4, James E. Rhoads2,3, Claus Leitherer5, Aida Wofford6,Tianxing Jiang2, Mark Dijkstra4, V. Tilvi2, Junxian Wang1

Draft version July 28, 2017

ABSTRACT

We studied Lyman- (Ly) escape in a statistical sample of 43 Green Peas with HST/COS Lyspectra. Green Peas are nearby star-forming galaxies with strong [OIII]5007 emission lines. Oursample is four times larger than the previous sample and covers a much more complete range of GreenPea properties. We found that about 2/3 of Green Peas are strong Ly line emitters with rest-frameLy equivalent width > 20 A. The Ly profiles of Green Peas are diverse. The Ly escape fraction,defined as the ratio of observed Ly flux to intrinsic Ly flux, shows anti-correlations with a few Lykinematic features both the blue peak and red peak velocities, the peak separations, and FWHM ofthe red portion of the Ly profile. Using properties measured from SDSS optical spectra, we foundmany correlations Ly escape fraction generally increases at lower dust reddening, lower metallicity,lower stellar mass, and higher [OIII]/[OII] ratio. We fit their Ly profiles with the HI shell radiativetransfer model and found Ly escape fraction anti-correlates with the best-fit NHI . Finally, we fitan empirical linear relation to predict fLyesc from the dust extinction and Ly red peak velocity. Thestandard deviation of this relation is about 0.3 dex. This relation can be used to isolate the effect ofIGM scatterings from Ly escape and to probe the IGM optical depth along the line of sight of eachz > 7 Ly emission line galaxy in the JWST era.

1. INTRODUCTION

In young star forming galaxies, Lyman continuum(LyC) photons from hot stars ionize the surroundinghydrogen gas, and Ly photons come from the recom-bination of hydrogen gas. The Ly emission line is apowerful tool in discovering and studying high redshiftgalaxies. Thousands of high redshift Ly emission linegalaxies (LAE) have been found in the last two decades(e.g. Dey et al. 1998; Hu et al. 1998; Rhoads et al.2000; Ouchi et al. 2003; Gawiser et al. 2006; Wang et al.2009; Kashikawa et al. 2011; Erb et al. 2014; Matthee etal. 2014; Zheng et al. 2016). These high redshift LAEsgenerally have small size, low stellar mass, low dust ex-tinction, low metallicity, young age, and high specific starformation rate (sSFR) (e.g. Malhotra 2012; Bond et al.2010; Gawiser et al. 2007; Pirzkal et al. 2007; Finkel-stein et al. 2008). At 2 < z < 6, these LAEs are animportant population of star-forming galaxies, and theyconstitute an increasing fraction of Lyman break galax-ies across that range, reaching 60% of Lyman breakgalaxies (LBGs) at redshift z 6 (Stark et al. 2011).

A current frontier is searching for LAEs in the epochof Cosmic Reionization. As Ly photons propagate froma LAE to the observer, they pass through the intergalac-tic medium (IGM) and will be scattered away from theline of sight by HI in IGM. So Ly line can be used

1 CAS Key Laboratory for Research in Galaxies and Cosmol-ogy, Department of Astronomy, University of Science and Tech-nology of China; huan.y@asu.edu

2 Arizona State University, School of Earth and Space Explo-ration

3 NASA Goddard Space Flight Center4 Institute of Theoretical Astrophysics, University of Oslo,

Norway5 Space Telescope Science Institute6 National Autonomous University of Mexico, Institute of As-

tronomy

to probe reionization of IGM (e.g. Malhotra & Rhoads2004; Treu et al. 2012; Pentericci et al. 2014; Tilviet al. 2014; Matthee et al. 2015; Santos et al. 2016).These Ly based methods can effectively probe HI frac-tion in the later half of reionization. One major goalof JWST is to observe the Ly and rest-frame opti-cal lines spectra of z > 7 galaxies and probe reioniza-tion with Ly lines. However, the challenge is to iso-late the impact of IGM from other effects that may di-minish Ly. The Ly photons have to escape out ofthe galaxies before passing through the IGM and be-ing observed, i.e. (Observed Ly) = (Intrinsic Ly) (Ly escape fraction) (IGM Transmission). TheLy escape fraction describes how many Ly photons es-cape out of both interstellar medium (ISM) and circum-galactic medium (CGM) of a LAE. Thus, to use Lyreionization tests, we have to understand Ly escape andpredict Ly escape fraction from other properties.

Ly escape is also related to the LyC escape process.A large fraction ( 9/12) of known LyC leakers are LAEs(Leitet et al. 2013; Borthakur et al. 2014; Izotov etal. 2016; Leitherer et al. 2016; de Barros et al. 2016;Shapley et al. 2016). LAEs at the reionization epochmay be major contributors of ionizing photons. Ly lineprofiles may be used as a tool for detecting LyC leakers(Verhamme et al. 2015; Alexandroff et al. 2015; Dijkstraet al. 2016). Understanding Ly escape is very usefulfor the study of LyC escape.

As Ly is a resonance line, it has a high cross-sectionfor HI scattering. The emergent Ly emission has a com-plicatedly dependence on the amount of dust, the HI gascolumn density (NHI), the kinematics of HI gas, and thegeometric distribution of HI gas and dust (e.g. Neufeld1990; Charlot & Fall 1993; Ahn et al. 2001; Verhammeet al. 2006; Dijkstra et al. 2006; Laursen et al. 2013).The scattering of Ly photons can significantly modify

arX

iv:1

701.

0185

7v2

[as

tro-

ph.G

A]

27

Jul 2

017

2

the Ly line profile. LAEs usually show asymmetric or adouble-peaked Ly emission line profiles (e.g. Rhoads etal 2003; Kashikawa et al. 2011; Erb et al. 2014). There-fore the Ly line profile carries a lot of information aboutthe resonant scatterings and can be used to probe the HIgas properties.

To study Ly escape, it is ideal to have a large sam-ple of LAEs and measure high quality Ly line spec-trum, many optical emission lines, HI gas properties, andmultiple other galactic properties. So we can test whatproperties make Ly escape, and finally predict Ly es-cape fraction from those properties. At high redshift,however, absorption by the intergalactic Ly forest pre-vents reliable measurements of the blue portion of Lyemission lines. Other crucial observations are also im-practical, both because high-z LAEs are faint, and be-cause some features (notably rest-optical emission lines)are redshifted to obs > 2.4m, where presently avail-able instruments lack sensitivity. Therefore many studiesseek to solve the Ly escape problem by observing low-z galaxies with similar properties to high-z LAEs (e.g.Giavalisco et al. 1996; Kunth et al. 1998; Mas-Hesse etal. 2003; Deharveng et al. 2008; Finkelstein et al. 2009;Atek et al. 2009; Leitherer et al. 2011; Heckman et al.2011; Cowie et al. 2011; Wofford et al. 2013; Hayeset al. 2005, 2014; Ostlin et al. 2014; Rivera-Thorsenet al. 2015). However, low-z LAEs are rare and manynearby Ly emission line galaxies are older and moreevolved galaxies than typical high-z LAEs and may be adifferent population of Ly emitters. Perhaps the mostrelevant nearby analogs of high-z LAEs are Green Peagalaxies (Jaskot & Oey 2014; Henry et al. 2015; Yang etal. 2016a, hereafter Paper I).

Green Pea galaxies were discovered in the citizen sci-ence project Galaxy Zoo, in which public volunteersmorphologically classified millions of galaxies from theSloan Digital Sky Survey (SDSS). Green Peas are com-pact galaxies that are unresolved in SDSS images. Thegreen color is because the [OIII] doublet dominates theflux of SDSS r-band which is mapped to the greenchannel in the SDSSs false-color gri-band images (Lup-ton et al. 2004). They generally have small stellarmasses ( 10810M), low metallicities for their stellarmasses, high specific star formation rates (sSFR), andlarge [OIII]5007/[OII]3727 (hereafter [OIII]/[OII]) ra-tio (Cardamone et al. 2009; Amorin et al. 2010; Izotovet al. 2011). The UV spectra of 17 Green Peas gen-erally show strong Ly emission lines (Paper I; Jaskot& Oey 2014; Henry et al. 2015; Izotov et al. 2016;Verhamme et al. 2016). These studies have exploredthe relation of fLyesc and dust, metallicity, Ly profiles,and metal absorption lines with small samples of GreenPeas. Besides the small sample size, the previous sam-ples of Green Peas tend to be lower metallicity and lowerdust extinction than the whole Green Pea sample. In ourHST program, we observed an additional 20 Green Peasin order to have a statistical sample that spans a rangeof galaxy properties such as metallicity, dust extinction,and star-formation rate (SFR).

In this paper, we use HST/COS Ly spectra of GreenPeas to study the mechanism of Ly escape. In Section2, we show the sample and observations. In Section 3, wedescribe the measurement and properties of Ly equiva-

lent width and escape fraction. In Section 4, we show therelation between Ly escape and Ly kinematic features.In Section 5, we show the relation between Ly escapeand dust extinction, metallicity, stellar mass, morphol-ogy, and [OIII]/[OII] ratio. In Section 6, we fit the Lyprofiles with radiative transfer model. In Section 7, weshow an empirical relation to predict Ly escape fractionand discuss its applications on probing reionization.

2. SAMPLE AND OBSERVATIONS

2.1. The Sample

Since the strong [OIII]5007 line makes Green Peagalaxies have special optical broadband colors, we canselect a few thousand Green Pea candidates from theSDSS imaging survey (Yang et al. 2016 in-prep). InSDSS DR7, a sample of 251 Green Peas were observedas serendipitous spectroscopic targets (Cardamone et al.2009). A subset of 66 Green Peas have sufficient signal tonoise ratio (S/N) in both continuum and emission lines(H, H, and [OIII]5007) to study galactic propertiessuch as SFR, stellar mass, and metallicity (Cardamoneet al. 2009; Izotov et al. 2011). Galaxies with an ac-tive galaxies nucleus (AGN) (diagnosed by their broadBalmer emission lines or H/[NII] vs. [OIII]/H dia-gram) are excluded. In Paper I, we matched these 66Green Peas with the COS archive and studied Ly es-cape in a sample of 12 Green Peas with COS UV spec-tra. Compared to the larger Green Pea sample, these 12Green Peas tend to be lower metallicity and lower dustextinction (figure 1). To address the bias and expand thesample size, we took Ly spectra of 20 additional GreenPeas (PI S. Malhotra, GO 14201). These 20 galaxieswere selected based on their metallicity and H/H val-ues to supplement the previous sample, so that the totalsample can cover the whole range of metallicity and dustextinction of the parent sample. We use figure 1 to dothe selection first draw grids (shown in figure 1), thenpick one or two sources in each grid cell. Note that (a)empty cells are not used and (b) the non-empty cells arenot covered perfectly because in the proposal we usedgas metallicities measured in Izotov et al. (2011) whichare slightly different from the metallicities shown in fig-ure 1. After the selection, we compared the total samplewith the parent sample to make sure there is no obviousbiases.

We also supplement this sample with 11 additionalGreen Peas from published literature. In total, we have43 Green Peas from six HST programs 20 galaxiesfrom GO 14201 (PI S. Malhotra), 9 galaxies from GO12928 (PI A. Henry; Henry et al. 2015), 7 galaxies fromGO 11727 and GO 13017 (PI T. Heckman; Heckman etal. 2011; Alexandroff et al. 2015), 2 galaxies from GO13293 (PI A. Jaskot; Jaskot et al. 2014), and 5 galaxiesfrom GO 13744 (PI T. Thuan; Izotov et al. 2016). The7 galaxies in T. Heckmans program were originally se-lected as nearby Lyman-break analogs by their high FUVluminosity, high UV flux, and compact size. These 7galaxies can also be classified as Green Peas by their com-pact sizes in SDSS images and strong [OIII]5007 emis-sion lines in SDSS spectra. Their sizes and [OIII]5007equivalent width are similar to the Green Peas in Carda-mone et al. (2009). We dont find any obvious bias byincluding the Lyman-break analogs in the analysis. The

3

7 Green Peas in A. Jaskots program and T. Thuansprogram were selected as LyC leakers by their extreme[OIII]/[OII] ratios. In figure 1, we show the above sam-ples on the metallicity and dust extinction (H/H ratio)diagram. We can see the current sample is a representa-tive Green Pea sample.

2.2. Properties from SDSS Spectra

From SDSS optical spectra of Green Peas, we get manygalactic properties. We use the SDSS pipeline measure-ments of their H, H, [OIII]5007, and [OII]3727 emis-sion line fluxes and line width. We correct the measuredH and H fluxes for Milky Way extinction using theattenuation of Schlafly & Finkbeiner (2011) (obtainedfrom the NASA/IPAC Galactic Dust Reddening and Ex-tinction tool) and the Fitzpatrick (1999) extinction law.Then we calculate E(B-V) assuming the Calzetti et al.(2000) extinction law and an intrinsic H/H ratio of2.86 (if H/H< 2.86, we set E(B-V)=0), and correctthe observed emission line fluxes for dust extinction. Weuse the stellar mass measured from SDSS spectra by Izo-tov et al. (2011) for 37 galaxies and the stellar mass inMPA-JHU SDSS catalog for the other 6 galaxies (all areLyman-break analogs). Note that the methods used inIzotov et al. (2011) and MPA-JHU are different. Themasses here should be treated as very rough estimatesbecause it is very hard to get the masses of the under-lying old population for these young starburst galaxies.To measure the metallicity using Te method, we mea-sure the [OIII]4363 line flux in SDSS spectra by fit-ting a Gaussian function to the continuum subtracted[OIII]4363 line spectra. Then we calculate the metal-licity using [OIII]4363, [OIII]5007, and [OII]3727 linefluxes following the Te method described in Izotov etal. (2006) and Ly et al. (2014). We convert the ex-tinction corrected H luminosity to SFR using the for-mula SFR(M/yr) = LH(erg/s)1041.27 (Kennicutt& Evans 2012). The dust extinction, mass, metallicity,SFR, and emission lines properties of this sample areshown in Table 1 and Table 2.

2.3. HST/COS Observation

In our program GO14201, we used HST/COS to ob-serve 20 Green Peas with one orbit per target. First,the targets were imaged in the COS acquisition modeACQ/IMAGE with MIRRORA, from which we got highresolution near-UV (NUV) images. The targets werecentered accurately (error 0.05) in the 2.5 diame-ter Primary Science Aperture. Then the spectra weretaken with grating G160M to cover rest-frame wave-length ranges about 1100 1400 A. The other archivalGreen Peas in our sample were also observed in the sameCOS acquisition mode ACQ/IMAGE with MIRRORA,and their spectra were taken with grating G130M and/orG160M. The NUV acquisition images of this sample areshown in figure 2.

The spectral resolution of the above observation isabout FWHM20 km s1 for a point source (James et al.2014). The actual spectral resolution depends on sourceangular sizes. The half-light radius of the NUV emissionof Green Peas are about 10 pixels (dispersion 0.012 Apixel1) and it results in FWHM40 km s1 for the UVcontinuum spectra. As the Ly sizes of Green Peas are

1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

H / H

7.7

7.8

7.9

8.0

8.1

8.2

8.3

8.4

12+log(O

/H) (T

e method)

Fig. 1. The metallicity and dust extinction (H/H ratio) di-agram of our Green Pea sample. Red squares shows the 20 galax-ies with new HST observations (GO 14201, PI S. Malhotra). Theother samples include 9 Green Pea galaxies with low dust extinction(cyan circle, Paper I; Henry et al. 2015), 7 Lyman-break analoggalaxies (magenta pentagon, Heckman et al. 2011; Alexandroff etal. 2015), 2 Lyman-continuum leaker candidates (blue star, Jaskotet al. 2014), and 5 confirmed Lyman-continuum leakers (blue tri-angle, two blue triangles overlap; Izotov et al. 2016). The blackhollow circles show the other galaxies without HST UV spectra inthe sample of 66 Green Peas. Note that a few sources have verysmall Ha/Hb values. The reasons are not yet well understood,but could be 1) poor flat-field calibration or sky subtraction, 2)different gas conditions from the case-B assumption.

somewhat larger than the UV continuum sizes (Yang etal. 2016b), the spectral resolutions are worse for the Lyemission lines. We retrieved COS spectra of this samplefrom the HST MAST archive after they were processedthrough the standard COS pipeline.

3. Ly EQUIVALENT WIDTH AND ESCAPE FRACTION

3.1. Measurements of Ly flux, EW, and escapefraction

Most Green Peas in our sample show strong Ly emis-sion lines (figure 3). But about 1/3 Green Peas haverelatively weak Ly lines, where the Ly absorptions inunderlying continuum become non-negligible. Since wewant to measure Ly emission from the recombinationof interstellar HI gas, we need to subtract the underlyingcontinuum.

We first estimate a constant local continuum fromwavelength ranges near Ly where the spectra look flatand there are no obvious emission or absorption features.We calculate the local continuum f(continuum) as theaverage of the spectra in these continuum ranges.

For 33 Green Peas without damped Ly absorption(see Table 3), we subtract the local continuum andcalculate the Ly flux by integrating the spectra in wave-length range 12121221 A. Then we correct the Lyflux for underlying stellar absorption. The equivalentwidth of stellar Ly absorption mostly depends on thestar formation history and age of the stellar population(Pena-Guerrero & Leitherer 2013). By comparing theH EW of these Green Peas (about 300 900A) withmodel predictions of H EW in star-forming galaxies,we found that these Green Peas probably have instanta-neous starburst with a burst age of 4 5 Myr (Levesque

4

0 0.0004 0.0012 0.0028 0.0059 0.012 0.025 0.05 0.1 0.2 0.4

1333+6246 LyC 1559+0841 1219+1526 1514+3852 1503+3644 LyC 1442-0209 LyC 1133+6514

1249+1234 1009+2916 0815+2156 1424+4217 0926+4428 1152+3400 LyC 0021+0052

1122+6154 0925+1403 LyC 0911+1831 0917+3152 1137+3524 1025+3622 1440+4619

1429+0643 1054+5238 1428+1653 0303-0759 1244+0216 2237+1336 1454+4528

1018+4106 0751+1638 0822+2241 1339+1516 1543+3446 0938+5428 0927+1740

1457+2232 0749+3337 1032+2717 0805+0925 1205+2620 0055-0021 0339-0725

0747+2336

Fig. 2. The 3 3 NUV images of Green Peas from the COS target acquisitions. In all panels, the colors are in log-scale with thesame count-rates limits (from 0 to 0.4). These images are sorted by decreasing fLyesc from left to right, and from top to bottom. The labelshows the ID of each Green Pea. The five LyC leakers are marked with LyC. The green bar in each panel shows the physical scale of 1Kpc.

& Leitherer 2013). According to the model calculationsin Pena-Guerrero & Leitherer (2013), the stellar Ly ab-sorption EW is about 7 A. So we correct the Ly fluxesof these 33 Green Peas by an EW=7 A absorption.

In another 8 Green Peas, the spectra show damped Lyabsorption wings and weak residual Ly emission lines.The damped Ly absorption is caused by interstellar ab-sorption of the continuum and/or the Ly absorption ofthe underlying stellar atmosphere continuum spectra. Tomeasure flux of the residual Ly emission, we subtractLy line spectra by a constant absorbed continuum.The absorbed continuum is estimated as the average inthe wavelength range where the Ly emission line meetsthe absorbed continuum. Then we integrate the Ly linespectra to get Ly flux. Since the above absorption cor-

rection already includes stellar Ly absorption, we dontneed to correct the stellar absorption for these 8 GreenPeas. Note that in some cases, the stellar absorptionmight have a very narrow component which is not fullycorrected by this method.

In the remaining two Green Peas (GP03390725 andGP0747+2336), the Ly lines are too weak and we didntdetect Ly emission.

Then we correct the measured Ly fluxes for MilkyWay extinction using the Fitzpatrick (1999) extinc-tion law. The rest-frame EW(Ly) is calculated us-ing the Ly fluxes and the local continuum asEW(Ly)=flux(Ly)/f(continuum)/(1+redshift). TheLy escape fraction, fLyesc , is defined as the ratio of themeasured Ly flux to the intrinsic Ly flux. Assuming

5

case-B recombination, the intrinsic Ly flux is about 8.7times dust extinction corrected H flux (See Henry etal. 2015 for discussions about the factor 8.7). Thus thefLyesc is Ly(observed)/(8.7Hcorrected). The SDSSH spectra were taken with 3 diameter aperture whichmatches the COS 2.5 diameter aperture very well. Notethat many Ly galaxies have a very extended Ly halo(e.g. Ostlin et al. 2009; Hayes et al. 2013; Momose etal. 2014). For these Green Pea galaxies, their Ly toUV size ratios are about 24 (Yang et al. 2017). ThusCOS 2.5 aperture probably captured the majority ofLy emission of those Green Peas.

Because the total counts per pixel in the UV continuumof this sample are small, we calculate the error spectrausing the Poisson noise of the total counts. The statis-tical errors of Ly fluxes are calculated from the errorspectra using the error propagation formula. The Lyflux, luminosity, EW(Ly), and fLyesc are shown in Ta-ble 3. A comparison of the fLyesc and EW(Ly) is shownin figure 4.

3.2. Ly EW distribution of Green Peas

With a large sample of Green Peas that cover the wholeranges of dust and metallicity, we now have a more reli-able estimation of the EW(Ly) distribution of GreenPeas than previous result. 41 out of 43 Green Peasshow Ly emission lines. 28 out of 43 GPs (65%) inour sample have rest-frame EW(Ly) > 20A and wouldbe classified as LAEs in a typical high-redshift narrow-band survey. We compared the EW(Ly) distribution ofthese 28 Green Peas to high redshift LAEs samples. Thehigh redshift LAEs samples include a sample of z = 2.8narrow-band selected LAEs (Zheng et al. 2016) and asample of spectroscopically confirmed LAEs at z=5.7or 6.5 (Kashikawa et al. 2011). To be consistent withthe methods used in high-z LAEs studies, we use theEW(Ly) of Green Peas without correction of the stellarLy absorption. We also add a GALEX selected z 0.3LAE sample to the comparison (Deharveng et al. 2008;Cowie et al. 2011; Finkelstein et al. 2009; Scarlata etal. 2009). Figure 5 shows the cumulative EW(Ly) frac-tion distributions of these four samples. These 28 GreenPeas have very similar EW(Ly) distribution to the high-redshift (z = 2.8) sample. So Green Peas in general arethe best nearby analogs of high-z LAEs.

4. Ly ESCAPE AND Ly PROFILES

4.1. Kinematic Features of Ly Profile

In the Ly escape process, Ly photons are resonantscattered by the HI gas. Depending on the column den-sity and bulk motion of HI gas, the resonant scatteringscan significantly modify the Ly profile. Therefore theLya profile carries a lot of information about the HI gasproperties. High-z LAEs usually show an asymmetric ora double-peaked Ly emission line profile (e.g. Rhoadset al 2003; Kashikawa et al. 2011; Erb et al. 2014). ForLAEs with detected optical emission lines and systemicredshifts, the peaks of Ly profiles are usually redshiftedwith respect to the systemic velocities (McLinden et al.2011, 2014; Chonis et al. 2013; Hashimoto et al. 2013;Song et al. 2014; Shibuya et al. 2014; Erb et al. 2014).The velocity offset of Ly emission line from the systemicvelocity is usually smaller in LAEs than in continuum

selected galaxies with weaker Ly emission lines or Lyabsorptions (Shapley et al. 2003).

Most Green Peas show double-peaked Ly profiles (fig-ure 3). For a typical double-peaked profile, we define thered peak as the peak in the Ly line profile occurringat velocity > 0, the blue peak as the Ly peak at ve-locity < 0, and the valley as the flux minimum betweenthe two peaks.

With a sample covering a large range of properties,we can see the Ly profiles are diverse. In figure 3,the 42 Green Peas are sorted by decreasing fLyesc fromtop left to bottom right. Three Green Peas with highfLyesc show single peak profiles where the peak veloci-ties are close to zero (GP1333+6246, GP14420209, andGP1249+1234). Many Green Peas with intermediatefLyesc generally show double-peaked profiles with muchstronger red peaks than blue peaks. On the other hand,many Green Peas with low fLyesc have a relatively largeratio of blue peak to red peak.

As in Paper I, we measure four kinematic features ofthe Ly profile: i) the blue peak velocity V(blue-peak);ii) the red peak velocity V(red-peak); iii) the peak sepa-ration V(red-peak)V(blue-peak); and iv) the full widthat half maximum (FWHM) of the red portion of Lyprofile, FWHM(red). The velocities are relative to thesystemic redshift derived from SDSS spectra. The mea-surements of these kinematic features are shown in Ta-ble 3. For some Green Peas, we dont measure theirvelocities because their Ly profiles are too noisy. In thenotes of Table 3, we explain the reason for each profilewithout velocity measurement. To measure the errorsof velocity peaks, we use a Monte-Carlo method to gen-erate 1000 fake spectra by adding Gaussian noise (withthe error spectra as the of Gaussian noise) to the ob-served spectra. Then we measure the peak velocities ofthese 1000 fake spectra and use the standard deviationsas the errors. In summary, we have measurements ofV(blue-peak) and the peak separation in 28 galaxies, andof V(red-peak) and FWHM(red) in 37 galaxies.

4.2. Relations between Ly escape and Ly kinematics

We show the relations between fLyesc and the kinematicfeatures of Ly profiles in figure 6. As fLyesc covers arange of about 3 dex, we show it in logarithmic scale.fLyesc shows anti-correlations with all four kinematic fea-tures V(blue-peak), V(red-peak), the peak separationV(red-peak)V(blue-peak), and the FWHM(red). Wecalculate the Spearman correlation coefficients of theserelations (shown in each panel of figure 6).

In Paper I, we found the fLyesc correlates strongly withV(blue-peak). Here we can see most Green Peas stillfollow the correlation, but there are a few Green Peaswith large scatter. So the overall correlation is worsethan in Paper I. These outliers suggest that the Ly bluepeak velocities are determined by multiple mechanisms.For example, one outlier (GP1454+4528, marked witha square and different color in figure 6) has a distinctprofile with the largest positive V(valley) (the velocityat the inter-peaks dip) and very strong blue portion Lyemission. Its V(blue-peak) and V(red-peak) clearly offsetfrom the trends. However, if we exchange the V(blue-peak) and V(red-peak), then it follows the trends verywell. There is probably strong gas inflows as well as gas

6

0.0

0.2

0.4

0.6

0.8

1.0 118.0%1333+6246

LyC

73.5%1559+0841 70.2%1219+1526 69.8%1514+3852 43.1%1503+3644

LyC

43.0%1442-0209

LyC

42.2%1133+6514

0.0

0.2

0.4

0.6

0.8

1.0 41.2%1249+1234 37.3%1009+2916 32.7%0815+2156 29.0%1424+4217 28.7%0926+4428 28.7%1152+3400

LyC

21.5%0021+0052

0.0

0.2

0.4

0.6

0.8

1.0 18.7%1122+6154 18.6%0925+1403

LyC

17.7%0911+1831 16.9%0917+3152 15.8%1137+3524 15.4%1025+3622 12.8%1440+4619

0.0

0.2

0.4

0.6

0.8

1.0 12.3%1429+0643 11.2%1054+5238 10.6%1428+1653 9.8%0303-0759 7.7%1244+0216 6.3%2237+1336 6.1%1454+4528

0.0

0.2

0.4

0.6

0.8

1.0 5.9%1018+4106 4.3%0751+1638 3.7%0822+2241 3.4%1339+1516 2.4%1543+3446 1.3%0938+5428 1.3%0927+1740

1000 0 1000

0.0

0.2

0.4

0.6

0.8

1.0 1.0%1457+2232

1000 0 1000

1.0%0749+3337

1000 0 1000

0.9%1032+2717

1000 0 1000

0.9%0805+0925

1000 0 1000

0.6%1205+2620

1000 0 1000

0.5%0055-0021

1000 0 1000

0339-0725

Velocity [km/s]

norm

aliz

ed f

lux

Fig. 3. Ly emission line spectra of Green Peas before subtracting continuum. These 42 galaxies are sorted by decreasing fLyesc from

left to right, and from top to bottom. The ID and fLyesc are given in each panel. The five LyC leakers are marked with LyC. The last onegalaxy (GP03390725) shows weak Ly absorption. One Green Pea (GP0747+2336) is not shown here, because its Ly spectra is verynoisy and no Ly emission or absorption lines are detected.

100 101 102

EW(Ly)

10-3

10-2

10-1

100

f esc(Ly

)

Fig. 4. Comparison of the fLyesc and EW(Ly) of Green Peas.

outflows in this galaxy. We excluded this object from thecalculation of correlation coefficients.

On the other hand, in Paper I, we found large scat-ter between fLyesc and V(red-peak) with 12 Green Peas.However, as the current sample covers a large range offLyesc and V(red-peak), f

Lyesc shows an anti-correlation

with V(red-peak). The relation between fLyesc and V(red-peak) in this Green Peas sample is very similar to the

0 20 40 60 80 100 120 140 160 180 200 220

EW(Ly) []

0.0

0.2

0.4

0.6

0.8

1.0

f (>

EW)

z=5.7 and 6.5 LAEz=2.8 LAEGALEX z=0.3 LAEGreen Peas

Fig. 5. Here we compare the rest-frame EW(Ly) distributionof Green Peas with different samples. The solid green line showsthe sample of 28 Green Peas with EW(Ly) > 20A. The blue dash-dot line shows the GALEX z 0.3 LAE sample (Cowie et al. 2011;Finkelstein et al. 2009; Scarlata et al. 2009). The magenta dashedline shows the z = 2.8 LAE sample from Zheng et al. (2016).The red dotted line shows the z = 5.7 and 6.5 LAE sample fromKashikawa et al. (2011).

relations between EW(Ly) and V(red-peak) in highredshift LAEs and LBGs, where the LAEs have high

7

5004003002001000

V(blue-peak) [km/s]

10-3

10-2

10-1

100

f esc

(Ly

)

r=0.47 P=1e-2

a)

0 100 200 300 400 500 600 700

V(red-peak) [km/s]

10-3

10-2

10-1

100

r=-0.57 P=3e-4

b)

200 300 400 500 600 700 800 900 10001100

V(red-peak)-V(blue-peak) [km/s]

10-3

10-2

10-1

100

f esc

(Ly

)

r=-0.66 P=2e-4

c)

100 150 200 250 300 350 400 450 500

FWHM(red) [km/s]

10-3

10-2

10-1

100

r=-0.63 P=4e-5

d)

Fig. 6. Relations between fLyesc and the kinematic features of

Ly profile: (a) fLyesc and the blue peak velocity of Ly profile,

V(blue-peak); (b) fLyesc and the red peak velocity of Ly pro-

file, V(red-peak); (c) fLyesc and the peak separation of Ly pro-

file; (d) fLyesc and the FWHM of the red portion of Ly profile,FWHM(red). The Spearman correlation coefficient and null proba-bility are shown. GP1454+4528 with possible gas inflows is markedby a square in different color in each panel.

EW(Ly) and small V(red-peak), while the LBGs havesmall EW(Ly) and large V(red-peak) (Shapley et al.2003; Hashimoto et al. 2013; Erb et al. 2014).

We also found that fLyesc anti-correlates withFWHM(red). We do a linear fit to this relation and getthe following function.

log(fLyesc ) = 0.545 (FWHM(red)/100km/s) + 0.563

The scatter of this relation is 0.43 dex in log(fLyesc ).

Since any high-z LAE with a spectrum will have a mea-sured FWHM for the red peak, it is easy to use thisrelation to infer the Ly escape fraction of high-z LAE.

Brief interpretations: The Ly profile depends on thecolumn density and the kinematics of HI gas. As the HIcolumn density increases, the numbers of scatterings forLy photons increase. The more scatterings generallyresult in larger offsets of peak velocities (V(blue-peak)and V(red-peak)) and broader line profile (FWHM(red)).Also, more scatterings increase the Ly photons pathlengths which makes the Ly radiation more susceptibleto dust extinction and consequently decreases the Lyescape fraction. Thus those anti-correlations mostly in-dicate that the fLyesc decreases as the column density ofHI gas increases.

5. Ly ESCAPE AND OTHER GALACTIC PROPERTIES

5.1. dust extinction, stellar mass, and metallicity

These Green Peas are very well studied galaxies andprovide a great opportunity to explore the dependence ofLy escape on other galactic properties. Previous studieshave found that fLyesc anti-correlates with dust extinction(Atek et al. 2014; Cowie et al. 2011; Paper I). However

0.0 0.1 0.2 0.3 0.4E(B-V)

10-3

10-2

10-1

100

f esc

(Ly

)

a)

r=-0.64 P=8e-6

7.7 7.8 7.9 8.0 8.1 8.2 8.312+log(O/H) (Te method)

10-3

10-2

10-1

100

f esc

(Ly

)

b)

r=-0.54 P=9e-4

7.5 8.0 8.5 9.0 9.5 10.0log(Mass (M)

10-3

10-2

10-1

100

f esc

(Ly

)

c)

r=-0.50 P=1e-3

7.5 8.0 8.5 9.0 9.5 10.0log(Mass (M)

7.6

7.7

7.8

7.9

8.0

8.1

8.2

8.3

8.4

12

+lo

g(O

/H)

(Te m

eth

od)

d)

2 1 0

Fig. 7. a) Relation between fLyesc and dust extinction E(B-V).The black dashed (blue dotted) line shows the expected Ly escapefraction if Ly is only absorbed by dust following the Calzetti etal. (2000) extinction law (the SMC extinction law). b) Relation

between fLyesc and the metallicity from Te method. c) Relation

between fLyesc and stellar mass. The Spearman correlation coef-ficient and null probability are shown in panel a), b), and c). d)The mass-metallicity relation of this sample. The color-bar shows

the value of log(fLyesc ). The dashed line shows the mass-metallicityrelation for SDSS galaxies in Amorin et al. (2010).

the relation between fLyesc and metallicity are unclear(Finkelstein et al. 2011; Atek et al. 2014; Hayes et al.2014; Paper I). Our sample covers the full ranges of dustextinction and metallicity of Green Peas. In figure 7, weshow the relations between fLyesc and E(B-V), metallicity,and stellar mass. The Spearman correlation coefficientsof these relations are shown figure 7.

The Green Peas with higher dust extinction tend tohave smaller fLyesc , confirming that dust extinction is animportant factor in Ly escape. In figure 7a, we alsoshow the expected Ly escape fractions if Ly is onlyabsorbed by dust following the Calzetti et al. (2000)extinction law (dashed line) or the SMC extinction law(Gordon et al. 2003) (dotted line). The SMC extinctionlaw is steeper in FUV than the Calzetti et al. (2000) ex-tinction law, so the extinction of Ly emission is largerfor SMC extinction law. Many Green Peas are below thedashed and dotted lines, because resonant scatterings in-crease the escape path length of Ly photons and thechances of being absorbed by dust. Interestingly, manyGreen Peas are above the relation for SMC extinctionlaw. If the dust extinction in Green Peas follows SMCextinction law, then it probably suggests resonant scat-terings in clumpy dust distributions decrease the dustextinction of Ly emission (Neufeld 1991; Hansen & Oh2006; Finkelstein et al. 2009; Scarlata et al. 2009; butalso see Laursen et al. 2013 showing that clumpy mediadoes not decrease the dust extinction of Ly for typicalconditions in LAEs).fLyesc also anti-correlates with metallicity and stellar

mass. In the fLyesc vs. metallicity diagram, only 37

8

galaxies with [OIII]4363 line S/N > 3 are shown. In fig-ure 7, we also show the mass-metallicity relation of GreenPeas and color the sample with fLyesc . The dashed lineshows the massmetallicity relation for SDSS galaxiesin Amorin et al. (2010), where the metallicity of SDSSgalaxies are calculated with the same effective temper-ature method. These Green Peas have lower metallici-ties than the massmetallicity relation of SDSS galaxies,similar to other emission line selected galaxies (Xia et al.2012; Ly et al. 2014; Song et al. 2014). These GreenPeas with lower metallicities and smaller masses haveless dust extinction. In addition, ionized gas outflowscan blow out the metal enriched gas and decrease themetallicity and dust extinction. At the same time, theionized gas outflows can make holes with low HI columndensities and help Ly escape.

5.2. Morphology and size of UV emission

We get the NUV image of each object from the COStarget acquisition (figure 2). So we also explore the rela-tion between Ly escape and the UV morphology. Thepixel scale of NUV image is 0.0235 0.0001 arcsec/pixel.The FWHM of point spread function is about 2 pixels or0.047. As we can see from the images, most Green Peasare very small and compact. Multiple clumps, tidal tails,and asymmetric shapes are common, which may suggestdwarf-dwarf mergers are common in Green Peas. In fig-ure 2, these images are sorted by decreasing fLyesc fromleft to right, and from top to bottom. The fLyesc does notshow an obvious relation with the morphology.

We then use GALFIT (Peng et al. 2010) to measurethe galaxy size. We fit the image with a single Sersicprofile component and get the half light radius of eachgalaxy. The half light radii are shown in Table 1. Therelation between fLyesc and the half light radius has verylarge scatter.

100 101

[OIII]/[OII]

100

101

102

EW

(Ly

)

r=0.52 P=4e-4

100 101

[OIII]/[OII]

10-3

10-2

10-1

100

f esc

(Ly

)

r=0.40 P=8e-3

Fig. 8. Left: Relation between EW(Ly) and [OIII]/[OII].

Right: Relation between fLyesc and [OIII]/[OII]. [OIII]/[OII]is defined as ([OIII]4959+[OIII]5007)/([OII]3726+[OII]3729).The Spearman correlation coefficient and null probability areshown in each panel.

5.3. [OIII]/[OII] ratio

Green Peas are selected to have large [OIII]/[OII] ra-tios. The [OIII]/[OII] ratio has been used to select LyCleaker candidates, and large [OIII]/[OII] may indicatethe existence of paths with low HI optical depth (Jaskot& Oey 2014; Izotov et al. 2016). In figure 8, we showthe relations of EW(Ly) vs. [OIII]/[OII] and fLyesc vs.[OIII]/[OII]. The Ly line strength generally increaseswith [OIII]/[OII], but the scatter is large.

6. Ly PROFILE FITTING

The Ly emission line profiles can usually be explainedby resonant scatterings of Ly photons by an outflowingHI gas shell (e.g. Ahn et al. 2001; Verhamme et al.2006; Dijkstra et al. 2006; Schaerer et al. 2011). To ex-tract more information from the Ly profiles and explorethe physical process of Ly escape, we fit the Ly pro-files with the outflowing HI shell radiative transfer model(Dijkstra et al. 2014; Gronke et al. 2015).

In the model, Ly photons were generated by a sourcefully surrounded by a spherical dusty HI gas shell whichscattered/absorbed the Ly photons. The intrinsic Lyline has a Gaussian profile with width . The shell isdescribed by four parameters: (i) outflow velocity vexp,(ii) HI column density NHI , (iii) temperature T (includ-ing turbulent motion as well as the true temperature),and (iv) dust optical depth d. Generally, these param-eters affect the Ly profile as follows: a larger outflowvelocity and a smaller NHI will decrease the red-peak ve-locity; a higher temperature will generally broaden theline profile; a larger dust optical depth will decrease theline strength. Then we find the best-fit model param-eters (, vexp, NHI , T, d) and calculate the errors ofparameters with Markov Chain Monte Carlo (MCMC)method. We refer the reader to Gronke et al. (2015) andPaper I for details of the model and the fitting method.

In Paper I, we showed the fitting results of 12 GreenPeas. The model fit nine profiles very well, but failedin the other three profiles. Here we show the fitting re-sults for another 23 Green Peas (out of the 31 additionalGreen Peas) with sufficient S/N in their Ly profiles.The model fit the observed profiles very well in manycases (figure 9). The best fit parameters are shown inTable 4. We discussed a few interesting fitting resultsbelow.

(1) HI column density: In Paper I, we found fLyesc anti-correlates with the best fit NHI for the 12 Green Peas.Here we show the relation between fLyesc and the best fitNHI in figure 10 for the combined sample of 35 GreenPeas. The result confirms the anti-correlation betweenfLyesc and NHI . For the three cases (GP1424+4217,GP1133+6514, and GP1219+1526, marked by large bluecircles) where the fitting procedure failed, we plot theNHI obtained by manually adjusting the model param-eters to match the observed depth of the valley andthe relative heights of blue and red peaks (see Section 6of Paper I). For GP1454+4528 (marked by a red square)with gas inflow, the fitting was bad. For the two galaxiesmarked by large cyan triangles, the best fit NHI are notconstrained. If the three galaxies marked by the squareand triangle are excluded, the Spearman correlation coef-ficient for the relation of fLyesc and NHI is r=-0.59 (P=4e-4). If all six galaxies marked by the large circle, squareand triangle are excluded, the Spearman correlation coef-ficient is r=-0.52 (P=4e-3). This result is consistent withstudies of high redshift LAEs that suggested LAEs havelower NHI than non-LAEs (e.g. Shibuya et al. 2014; Erbet al. 2014; Hashimoto et al. 2015). Therefore the lowcolumn density of HI gas is a key factor to make Lyescape.

(2) Intrinsic Ly line width: The intrinsic Ly lineGaussian width is about 23 times larger than the HGaussian width in many cases, as we discussed in Paper

9

0.0

0.2

0.4

0.6

0.8

1.0 1333+6246

LyC

1559+0841 1514+3852 1503+3644

LyC

1442-0209

LyC

1009+2916

0.0

0.2

0.4

0.6

0.8

1.0 1152+3400

LyC

0021+0052 1122+6154 0925+1403

LyC

0917+3152 1025+3622

0.0

0.2

0.4

0.6

0.8

1.0 1440+4619 1429+0643 1428+1653

1000 0 1000

2237+1336

1000 0 1000

1454+4528

1000 0 1000

1018+4106

1000 0 1000

0.0

0.2

0.4

0.6

0.8

1.0 0751+1638

1000 0 1000

0822+2241

1000 0 1000

1339+1516

1000 0 1000

0938+5428

1000 0 1000

0055-0021

Velocity [km/s]

norm

aliz

ed flux

Fig. 9. The observed Ly profiles (blue lines) and the best fit Ly profiles (red lines) for 23 Green Peas with good S/N in their Lyprofiles. In Paper I, we showed the radiative transfer model fitting results of another 12 Green Peas. These galaxies are sorted by decreasing

fLyesc from left to right, and from top to bottom.

16 17 18 19 20 21

log(NHI) (cm2 )

10-2

10-1

100

f esc(Ly

)

r=-0.59

P=4e-4

Fig. 10. Relation between fLyesc and the best fit NHIfrom radiative transfer model. Five known LyC leakers aremarked by large diamonds. For the three cases (GP1424+4217,GP1133+6514, and GP1219+1526, marked by large blue circles)where the fitting procedure failed, we plot the NHI obtained bymanually adjusting the model parameters to match the observeddepth of the valley and the relative heights of blue and red peaks(see Section 6 of Paper I). For GP1454+4528 (marked by a largered square) with gas inflow, the fitting is bad (see figure 8). Forthe two galaxies marked by large cyan triangles (GP1428+1653 andGP1122+6154), the best fit NHI are not constrained. The Spear-man correlation coefficient is calculated without the three galaxiesmarked by square and triangle.

I. In four cases, the best fit is narrow and comparable to

the H width because the best fit profile only has a singlepeak. The wide intrinsic Ly line profile can be dueto important radiative transfer effects that broaden Lyprofile near to the source, before the processes attributedto the outflowing HI shell.

(3) Outflow velocities: The best fit shell outflow ve-locities are mostly between 5 to 170 km s1 which aregenerally smaller than the outflow velocities measuredfrom the low-ionized UV absorption lines (Yang et al. in-prep). This may suggest the low-ionized absorption linestrace a different gas component from the HI gas. We alsonoticed that for six profiles with strong blue peaks, thebest fit shell outflow velocities are smaller than 20 kms1. In GP1454+4528, the outlier discussed in section4.1, the HI gas shell is inflowing with a best-fit velocityof 171 km s1.

(4) The three failed cases: In Paper I, the model failedin three profiles with positive velocities at the line val-ley. We later improved the model by adding a shift ofthe velocity zero point as a free parameter of the fitting.The improved model can fit these three profiles very well.But the shifts of velocity zero points are about 90 150km s1 which are too large to be due to the errors ofwavelength calibration. Those large shifts may be ex-plained by some additional radiative transfer effects be-fore the Ly photons meet the HI gas shell.

Although the shell model captures many real radiativetransfer effects and can fit the Ly profiles very well, weshould be cautious about the interpretation of the bestfit parameters. A simple shell model can mimic morecomplex real physical properties (Gronke et al. 2016).For example, a low NHI model can mimic a model in

10

which the gas is clumpy and the covering factor is low(Gronke & Dijkstra 2016). In this case, the best-fit NHIvalue is a simple approximation of the overall HI columndensities. Interestingly, the best fit NHI of the five LyCleakers are about 101720 cm2, larger than theNHI thatpermit LyC escape. It suggests that their LyC emissionprobably escape through some holes in the interstellarmedium with much lower NHI .

7. PREDICTING Ly ESCAPE FRACTION

As we said in the Introduction, one major reason forthe studies of Ly escape is to use Ly lines to probereionization. A fraction (fLyesc ) of intrinsic Ly pho-tons first escape out of an LAE, then they go throughthe IGM where they can be further scattered by HI,and the remaining photons can finally be observed asa Ly line. So the IGM transmission can be measuredfrom the observed Ly line flux if we know the intrin-sic Ly line flux and fLyesc , i.e. IGM Transmission =(Observed Ly)/(Intrinsic Ly fLyesc ). In the nearfuture, JWST will be able to measure the observed Lyline and derive the intrinsic Ly line from the observedH line for galaxies in the epoch of reionization. If theremaining factor, fLyesc , can be predicted from other ob-served galactic properties, then each Ly line can be usedas an IGM probe on its line of sight. With this sampleof Green Peas, we have found correlations between fLyescand Ly kinematic features, dust extinction, metallicity,stellar mass, and HI column density. So can we selecta few observable factors and fit an empirical relation topredict fLyesc ?

Physically, Ly escape depends on the properties ofdust and HI gas, so we should select the factors thatcan indicate the properties of dust and HI gas. Dustextinction is relatively easy to measure and could be auseful factor. The Ly kinematic features strongly de-pend on the column density and kinematics of HI gasand could be another useful factor. Among a few Lykinematic features, the Ly red-peak velocity is easierand more robust to measure than the blue-peak velocitywhich might be removed by absorption and the line widthwhich depends on the spectra resolution. The other threefactors metallicity, stellar mass, and HI column densityfrom fitting of Ly profile are difficult to measure andthe uncertainties are large. Furthermore, both dust ex-tinction and Ly V(red-peak) show relatively tight anti-correlations with fLyesc . So we fit an linear empirical re-lation to predict fLyesc from dust extinction and V(red-peak) of Ly profile.

In figure 11, we first show the relations of fLyesc , E(B-V), and V(red-peak). In the diagram of E(B-V) vs.V(red-peak), objects are color-coded by fLyesc . We cansee that (i) E(B-V) and V(red-peak) dont show a corre-lation; (ii) the Green Peas with lower dust extinction andsmaller V(red-peak) have larger fLyesc . In the diagramof fLyesc vs. E(B-V), objects are color-coded by V(red-peak). Those Green Peas with large V(red-peak) gen-erally have smaller fLyesc than the others with the sameE(B-V). Then we fit 37 Green Peas with both V(red-peak) and E(B-V) measurements. Two Green Peas,GP1454+4528 with gas inflow and GP0749+3337 withthe largest V(red-peak), are outliers of the fitting, so weremove these two objects. The final best-fit relation of

35 Green Peas is

log(fLyesc ) = a(E(BV )/0.1)+b(V (redpeak)/100)+c

, where (a = 0.437, b = 0.483, c = 0.464). Inthe bottom two panels of figure 11, we compare the ob-served and the predicted fLyesc and show the histogramof the differences, log(fLyesc )-log(predicted f

Lyesc ). The

standard deviation of this relation is 0.3 dex.Now we have a relation to predict fLyesc from dust

extinction and Ly V(red-peak). If JWST measuresthe observed Ly flux, observed H flux, dust ex-tinction, and Ly V(red-peak) of a z > 7 LAE,then we can infer the IGM transmission along thisline of sight using the formula IGM Transmission =(Observed Ly)/(Intrinsic Ly fLyesc ), where theIntrinsic Ly is calculated from dust extinction cor-rected H flux and fLyesc is calculated from the empiricalrelation.

The IGM measured by this method is the true IGMfar from the LAE, which is in contrast to the circum-galactic medium (CGM). The true IGM only affectsthe strength of Ly red peak by the damped absorptionfactor of e , where is the optical depth of the IGM HIgas along the line of sight, and its effect on the velocity ofthe narrow Ly red peak is negligible. Some simulationssuggested that the HI gas in the CGM can be very closeto the Ly photons in frequency, so the CGM HI gas canresonantly scatter and/or absorb Ly photons at V(red-peak) 6.5 (e.g. Hayes etal. 2011; Tilvi et al. 2014; Pentericci et al. 2014). Thiscould be due to small number statistics. But if this sig-nal is real, it suggests either (i) the true IGM opticaldepth increases rapidly or (ii) the optical depth of ISMand CGM increases rapidly. Using our empirical relation,we can measure the optical depth of the true IGM anddistinguish these two possibilities.

Some recent observations suggest that five z 7 galax-ies show very small velocity offsets about 20-150km s1

between Ly and [CII] emission lines (Pentericci et al.2016; Bradac et al. 2017). Those small V(red-peak) val-ues may indicate that the Ly escape fractions are highand the optical depths of ISM and CGM are small.

One caveat regards whether the empirical relation de-rived from low-z analogs is applicable to high-z LAEs.The properties of ISM and CGM likely evolve betweenthe low-z LAEs (Green Peas) and LAEs in the epoch ofreionization. However, since the physics of Ly resonantscattering is same in both low and high-z, increasing theHI gas column density in ISM probably doesnt changehow NHI affects Ly profile. So the empirical relation isvery likely applicable to z > 6 LAEs.

8. CONCLUSION

11

50 100 150 200 250 300 350 400 450V(red-peak) [km/s]

0.05

0.00

0.05

0.10

0.15

0.20

0.25E(B

-V)

color log(fesc(Ly))2 1 0

0.05 0.00 0.05 0.10 0.15 0.20 0.25E(B-V)

10-3

10-2

10-1

100

f esc

(Ly

)

color V(red-peak) 100 200 300

10-3 10-2 10-1 100

predicted fesc(Ly)

10-3

10-2

10-1

100

f esc

(Ly

)

0.8 0.4 0.0 0.4 0.8log(fesc(Ly)) log(predicted fesc(Ly))

0

2

4

6

8

10

12

Num

ber

Fig. 11. Top-left: the relation of E(B-V) vs. V(red-peak); The color-bar shows log(fLyesc ) value. Top-right: the relation of fLyesc

and E(B-V); The color-bar shows V(red-peak) value. Bottom-left: the comparison of observed and predicted fLyesc . Here the predicted

log(fLyesc )=a (E(BV )/0.1) + b (V (redpeak)/100) + c. Bottom-right: the histogram of the differences, log(fLyesc )-log(predicted fLyesc ).

We studied Ly escape in a statistical sample of GreenPeas with HST/COS Ly spectra. About 2/3 GreenPeas show strong Ly emission lines. Many Green Peasshow double-peaked Ly line profiles, but the Ly pro-files are diverse. These Green Peas have well measuredgalactic properties from SDSS optical spectra, so we in-vestigated the dependence of Ly escape on dust extinc-tion, metallicity, stellar mass, galaxy morphology, and[OIII]/[OII] ratio. We also fit their Ly profiles with theHI shell radiative transfer model. Finally, we derived anempirical relation to predict Ly escape fraction. Ourmajor conclusions are as follows:

1. With a statistical sample of 43 Green Peas thatcover the whole ranges of dust extinction andmetallicity properties of Green Peas, we foundabout 2/3 of Green Peas are strong Ly line emit-ters with distribution of EW(Ly) consistent withhigh-z LAEs. This confirmed that Green Peas gen-erally are the best analogs of high-z LAEs in thenearby universe.

2. The fLyesc shows anti-correlations with a few Lykinematic features the blue peak velocity, the

red peak velocity, the peak separation, and theFWHM(red) of Ly profile. These Ly kinematicfeatures are sensitive to the column density andthe kinematics of HI gas. As more scatterings inHI gas can make the Ly velocity offsets larger andthe Ly profile broader, these correlations stronglysuggest low NHI and fewer scatterings help Lyphotons escape.

3. With a large sample, we found many correlationsregarding the dependence of Ly escape on galacticproperties fLyesc generally increases at lower dustextinction, lower metallicity, lower stellar mass,and higher [OIII]/[OII] ratio. fLyesc does not havean obvious relation with the UV morphology ofGreen Peas.

4. The single shell radiative transfer model can repro-duce most Ly profiles of Green Peas. The best-fitNHI anti-correlates with f

Lyesc , indicating that low

NHI is key to Ly escape.

5. We fit an empirical linear relation between fLyesc ,dust extinction, and Ly red peak velocity. Thisrelation can be used to predict the fLyesc of LAEs

12

and isolate the effect of IGM scatterings from Lyescape. As JWST can measure the dust extinctionand Ly red peak velocity of some z > 7 LAEs,this relation makes it possible to measure the HIcolumn density of IGM along the line of sight ofeach LAE and to probe reionization with their Lylines.

We thank David Sobral, Edmund Christian Herenz,Kimihiko Nakajima, Alaina Henry, and the referee forvery helpful comments. The imaging and spectroscopydata are based on observations with the NASA / ESAHubble Space Telescope, obtained at the Space TelescopeScience Institute, which is operated by the Association

of Universities for Research in Astronomy (AURA), Inc.,under NASA contract NAS 5-26555. Some of the datapresented in this paper were obtained from the MikulskiArchive for Space Telescopes (MAST). STScI is operatedby the Association of Universities for Research in Astron-omy, Inc., under NASA contract NAS5-26555. Supportfor MAST for non-HST data is provided by the NASAOffice of Space Science via grant NNX09AF08G andby other grants and contracts. H.Y. acknowledges sup-port from China Scholarship Council. H.Y. and J.X.W.thanks supports from NSFC 11233002, 11421303, andCAS Frontier Science Key Research Program (QYZDJ-SSW-SLH006). This work has also been supported inpart by NSF grant AST-1518057; and by support forHST program #14201.

REFERENCES

Ahn, S.-H., Lee, H.-W., & Lee, H. M. 2001, ApJ, 554, 604Amorn, R. O., Perez-Montero, E., & Vlchez, J. M. 2010, ApJ,

715, L128Alexandroff, R. M., Heckman, T. M., Borthakur, S., Overzier, R.,

& Leitherer, C. 2015, ApJ, 810, 104Atek, H., Schaerer, D., & Kunth, D. 2009, A&A, 502, 791Atek, H., Kunth, D., Schaerer, D., et al. 2014, A&A, 561, A89Bond, N. A., Feldmeier, J. J., Matkovic, A., et al. 2010, ApJ, 716,

L200Borthakur, S., Heckman, T. M., Leitherer, C., & Overzier, R. A.

2014, Science, 346, 216Bradac, M., Garcia-Appadoo, D., Huang, K.-H., et al. 2017, ApJ,

836, L2Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682Cardamone, C., Schawinski, K., Sarzi, M., et al. 2009, MNRAS,

399, 1191Charlot, S., & Fall, S. M. 1993, ApJ, 415, 580Chonis, T. S., Blanc, G. A., Hill, G. J., et al. 2013, ApJ, 775, 99Cowie, L. L., Barger, A. J., & Hu, E. M. 2011, ApJ, 238, 136de Barros, S., Vanzella, E., Amorn, R., et al. 2016, A&A, 585,

A51Deharveng, J.-M., Small, T., Barlow, T. A., et al. 2008, ApJ, 680,

1072Dey, A., Spinrad, H., Stern, D., Graham, J. R., & Chaffee, F. H.

1998, ApJ, 498, L93Dijkstra, M., Haiman, Z., & Spaans, M. 2006, ApJ, 649, 14Dijkstra, M. 2014, PASA, 31, e040Dijkstra, M., Gronke, M., & Venkatesan, A. 2016, ApJ, 828, 71Erb, D. K., Steidel, C. C., Trainor, R., et al. 2014 ApJ, 795, 33Finkelstein, S. L., Rhoads, J. E., Malhotra, S., Grogin, N., &

Wang, J. 2008, ApJ, 678, 655Finkelstein, S. L., Cohen, S. H., Malhotra, S., et al. 2009, ApJ,

703, L162Finkelstein, S. L., Cohen, S. H., Moustakas, J., et al. 2011, ApJ,

733, 117Fitzpatrick, E. L. 1999, PASP, 111, 63Gawiser, E., van Dokkum, P. G., Gronwall, C., et al. 2006, ApJ,

642, L13Gawiser, E., Francke, H., Lai, K., et al. 2007, ApJ, 671, 278Giavalisco, M., Koratkar, A., & Calzetti, D. 1996, ApJ, 466, 831Gordon, K. D., Clayton, G. C., Misselt, K. A., Landolt, A. U., &

Wolff, M. J. 2003, ApJ, 594, 279Gronke, M., Bull, P., & Dijkstra, M. 2015, ApJ, 812, 123Gronke, M., Dijkstra, M., McCourt, M., & Oh, S. P. 2016,

arXiv:1611.01161Gronke, M., & Dijkstra, M. 2016, ApJ, 826, 14Hansen, M., & Oh, S. P. 2006, MNRAS, 367, 979Hashimoto, T., Ouchi, M., Shimasaku, K., et al. 2013, ApJ, 765,

70Hashimoto, T., Verhamme, A., Ouchi, M., et al. 2015, ApJ, 812,

157Hayes, M., Ostlin, G., Mas-Hesse, J. M., et al. 2005, A&A, 438, 71Hayes, M., Schaerer, D., Ostlin, G., et al. 2011, ApJ, 730, 8Hayes, M., Ostlin, G., Schaerer, D., et al. 2013, ApJ, 765, L27Hayes, M., Ostlin, G., Duval, F., et al. 2014, ApJ, 782, 6

Heckman, T. M., Borthakur, S., Overzier, R., et al. 2011, ApJ,730, 5

Henry, A., Scarlata, C., Martin, C. L., & Erb, D. 2015, ApJ, 809,19

Hu, E. M., Cowie, L. L., & McMahon, R. G. 1998, ApJ, 502, L99Izotov, Y. I., Stasinska, G., Meynet, G., Guseva, N. G., & Thuan,

T. X. 2006, A&A, 448, 955Izotov, Y. I., Guseva, N. G., & Thuan, T. 2011, ApJ, 728, 161Izotov, Y. I., Schaerer, D., Thuan, T. X., et al. 2016, MNRAS,

461, 3683James, B. L., Aloisi, A., Heckman, T., Sohn, S. T., & Wolfe,

M. A. 2014, ApJ, 795, 109Jaskot, A. E. & Oey, M. S. 2014, ApJ, 791, 19LKashikawa, N., Shimasaku, K., Matsuda, Y., et al. 2011, ApJ,

734, 119Kennicutt, R. C., & Evans, N. J. 2012, ARA&A, 50, 531Kunth, D., Mas-Hess, J. M., Terlevich, E., et al. 1998, A&A, 334,

11Laursen, P., Sommer-Larsen, J., & Razoumov, A. O. 2011, ApJ,

728, 52Laursen, P., Duval, F., & Ostlin, G. 2013, ApJ, 766, 124Leitet, E., Bergvall, N., Hayes, M., Linne, S., & Zackrisson, E.

2013, A&A, 553, A106Leitherer, C., Tremonti, C. A., Heckman, T. M., & Calzetti, D.

2011, AJ, 141, 37Leitherer, C., Hernandez, S., Lee, J. C., & Oey, M. S. 2016, ApJ,

823, 64Levesque, E. M., & Leitherer, C. 2013, ApJ, 779, 170Lupton, R., Blanton, M. R., Fekete, G., et al. 2004, PASP, 116,

133Ly, C., Malkan, M. A., Nagao, T., et al. 2014, ApJ, 780, 122Malhotra, S., & Rhoads, J. E. 2004, ApJ, 617, L5Malhotra, S., Rhoads, J. E., Finkelstein, S. L., et al. 2012, ApJ,

750, L36Mas-Hesse, J. M., Kunth, D., Tenorio-Tagle, G., et al. 2003, ApJ,

598, 858Matthee, J. J. A., Sobral, D., Swinbank, A. M., et al. 2014,

MNRAS, 440, 2375Matthee, J., Sobral, D., Santos, S., et al. 2015, MNRAS, 451, 400McLinden, E. M., Finkelstein, S. L., Rhoads, J. E., et al. 2011,

ApJ, 730, 136McLinden, E. M., Rhoads, J. E., Malhotra, S., et al. 2014,

MNRAS, 439, 446Momose, R., Ouchi, M., Nakajima, K., et al. 2014, MNRAS, 442,

110Neufeld, D. A. 1990, ApJ 350, 216Ostlin, G., Hayes, M., Kunth, D., et al. 2009, AJ, 138, 923Ostlin, G., Hayes, M., Duval, F., et al. 2014, ApJ, 797, 11Ouchi, M., Shimasaku, K., Furusawa, H., et al. 2003, ApJ, 582, 60Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2010, AJ,

139, 2097Pena-Guerrero, M. A., & Leitherer, C. 2013, AJ, 146, 158Pentericci, L., Vanzella, E., Fontana, A., et al. 2014, ApJ, 793, 113Pentericci, L., Carniani, S., Castellano, M., et al. 2016, ApJ, 829,

L11

http://arxiv.org/abs/1611.01161

13

Pirzkal, N., Malhotra, S., Rhoads, J. E., & Xu, C. 2007, ApJ,667, 49

Rhoads, J. E., Malhotra, S., Dey, A., et al. 2000, ApJ, 545, L85Rhoads, J. E., Dey, A., Malhotra, S., et al. 2003, AJ, 125, 1006Rivera-Thorsen, T. E., Hayes, M., Ostlin, G., et al. 2015, ApJ,

805, 14Santos, S., Sobral, D., & Matthee, J. 2016, MNRAS, 463, 1678Scarlata, C., Colbert, J., Teplitz, H. I., et al. 2009, ApJ, 705, 98LSchaerer, D., Hayes, M., Verhamme, A., & Teyssier, R. 2011,

A&A, 531, A12Schlafly, E. F. & Finkbeiner, D. F. 2011, ApJ, 737, 103Shapley, A. E., Steidel, C. C., Pettini, M., & Adelberger, K. L.

2003, ApJ, 588, 65Shapley, A. E., Steidel, C. C., Strom, A. L., et al. 2016, ApJ, 826,

L24Shibuya, T., Ouchi, M., Nakajima, K., et al. 2014, ApJ, 788, 74Song, M., Finkelstein, S. L., Gebhardt, K., et al. 2014, ApJ, 791, 3Stark, D. P., Ellis, R. S., & Ouchi, M. 2011, ApJ, 728, L2

Tilvi, V., Papovich, C., Finkelstein, S. L., et al. 2014, ApJ, 794, 5Treu, T., Trenti, M., Stiavelli, M., Auger, M. W., & Bradley,

L. D. 2012, ApJ, 747, 27Verhamme, A., Schaerer, D., & Maselli, A. 2006, A&A, 460, 397Verhamme, A., Orlitova, I., Schaerer, D., & Hayes, M. 2015,

A&A, 578, A7Verhamme, A., Orlitova, I., Schaerer, D., et al. 2016,

arXiv:1609.03477Wang, J.-X., Malhotra, S., Rhoads, J. E., Zhang, H.-T., &

Finkelstein, S. L. 2009, ApJ, 706, 762Wofford, A., Leitherer, C., & Salzer, J. 2013, ApJ, 765, 118Xia, L., Malhotra, S., Rhoads, J., et al. 2012, AJ, 144, 28Yang, H., Malhotra, S., Gronke, M., et al. 2016a, ApJ, 820, 130Yang, H., Malhotra, S., Rhoads, J. E., et al. 2017, ApJ, 838, 4Zheng, Z.-Y., Malhotra, S., Rhoads, J. E., et al. 2016, ApJS, 226,

23

http://arxiv.org/abs/1609.03477

14

TABLE 1The Sample

ID RA DEC z E(B-V)MW E(B-V) 12+log(O/H) log(M/M) SFR Re GO#(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

1333+6246a 13:33:03.94 +62:46:03.7 0.31812 0.017 0.000 7.72 8.50 1.4 0.72 137441559+0841 15:59:25.97 +08:41:19.1 0.29704 0.033 0.000 8.04 8.97 3.5 0.47 142011219+1526 12:19:03.98 +15:26:08.5 0.19560 0.022 0.000 7.81 8.35 13.0 0.33 129281514+3852 15:14:08.63 +38:52:07.3 0.33262 0.019 0.000 8.12 9.32 6.4 0.67 142011503+3644a 15:03:42.82 +36:44:50.8 0.35569 0.013 0.007 8.01 8.22 12.9 0.52 1374414420209a 14:42:31.37 02:09:52.8 0.29367 0.046 0.094 7.95 8.96 21.2 0.50 137441133+6514 11:33:03.80 +65:13:41.3 0.24140 0.009 0.040 7.95 9.30 6.4 0.82 129281249+1234 12:48:34.64 +12:34:02.9 0.26339 0.026 0.084 8.10 9.05 18.3 0.71 129281009+2916 10:09:18.99 +29:16:21.5 0.22192 0.019 0.000 7.92 7.87 3.7 0.46 142010815+2156 08:15:52.00 +21:56:23.6 0.14095 0.035 0.014 7.96 8.71 4.4 0.35 132931424+4217 14:24:05.73 +42:16:46.3 0.18479 0.009 0.028 8.02 8.34 19.2 0.48 129280926+4428 09:26:00.44 +44:27:36.5 0.18069 0.016 0.074 8.02 8.78 14.8 0.43 117271152+3400a 11:52:04.88 +34:00:49.8 0.34195 0.017 0.114 7.95 8.35 23.2 0.52 137440021+0052 00:21:01.02 +00:52:48.1 0.09836 0.021 0.038 8.14 9.30 13.7 0.44 130171122+6154 11:22:19.73 +61:54:45.4 0.20456 0.007 0.129 8.14 7.85 6.5 0.32 142010925+1403a 09:25:32.37 +14:03:13.0 0.30121 0.027 0.134 8.01 8.46 23.8 0.42 137440911+1831 09:11:13.34 +18:31:08.2 0.26220 0.024 0.168 7.96 9.75 26.8 0.57 129280917+3152 09:17:02.52 +31:52:20.5 0.30036 0.017 0.189 8.10 9.37 21.8 0.47 142011137+3524 11:37:22.14 +35:24:26.7 0.19439 0.016 0.043 8.12 9.56 19.5 0.72 129281025+3622 10:25:48.38 +36:22:58.4 0.12649 0.010 0.088 8.11 9.20 10.0 0.76 130171440+4619 14:40:09.94 +46:19:36.9 0.30076 0.012 0.148 8.13 9.62 38.0 0.72 142011429+0643 14:29:47.03 +06:43:34.9 0.17351 0.022 0.053 8.01 9.40 30.6 0.40 130171054+5238 10:53:30.83 +52:37:52.9 0.25264 0.013 0.069 8.08 9.77 27.3 0.62 129281428+1653 14:28:56.41 +16:53:39.4 0.18164 0.017 0.175 8.12 9.60 22.2 0.77 1301703030759 03:03:21.41 07:59:23.2 0.16488 0.085 0.000 7.87 9.15 8.9 0.56 129281244+0216 12:44:23.37 +02:15:40.4 0.23943 0.021 0.062 8.09 9.65 31.0 1.02 129282237+1336 22:37:35.05 +13:36:47.0 0.29350 0.049 0.126 8.11 9.45 30.7 1.08 142011454+4528 14:54:35.58 +45:28:56.3 0.26851 0.036 0.169 8.22 9.52 21.4 0.45 142011018+4106 10:18:03.24 +41:06:21.0 0.23705 0.012 0.094 7.93 9.32 10.4 0.78 142010751+1638 07:51:57.78 +16:38:13.2 0.26471 0.031 0.149 7.85 8.35 7.8 0.80 142010822+2241 08:22:47.66 +22:41:44.0 0.21619 0.039 0.195 8.11 8.43 41.6 0.68 142011339+1516 13:39:28.30 +15:16:42.1 0.19202 0.026 0.114 8.05 9.43 18.7 0.38 142011543+3446 15:43:01.22 +34:46:01.4 0.18733 0.025 0.000 7.96 8.05 2.6 0.77 142010938+5428 09:38:13.49 +54:28:25.0 0.10208 0.015 0.123 8.17 9.40 13.6 0.47 117270927+1740 09:27:28.67 +17:40:18.6 0.28831 0.026 0.180 8.06 9.26 18.2 0.94 142011457+2232 14:57:35.13 +22:32:01.7 0.14861 0.041 0.061 8.02 9.13 11.6 0.42 132930749+3337 07:49:36.77 +33:37:16.3 0.27318 0.048 0.203 8.18 9.49 62.3 1.47 142011032+2717 10:32:26.95 +27:17:55.2 0.19246 0.018 0.097 8.22 9.65 13.3 0.63 142010805+0925 08:05:18.04 +09:25:33.5 0.33034 0.018 0.402 7.98 9.36 22.9 0.81 142011205+2620 12:05:00.67 +26:20:47.7 0.34261 0.016 0.178 7.89 9.84 22.0 0.83 1420100550021 00:55:27.46 00:21:48.7 0.16745 0.022 0.217 8.18 9.70 30.4 0.46 1172703390725 03:39:47.79 07:25:41.2 0.26071 0.053 0.095 8.31 9.70 29.6 0.88 142010747+2336 07:47:58.00 +23:36:32.7 0.15524 0.051 0.085 8.02 9.06 5.9 0.59 14201

Note. Column Descriptions: (1) Object ID; (4) Redshifts are from SDSS optical spectra; (5) The Milky Way extinction E(B V )MW , basedon Schlafly & Finkbeiner (2011); (6) dust extinction; (7) metallicity; (8) stellar mass; (9) star formation rate in unit of M yr

1 derived from Hluminosity; (10) half light radius in unit of Kpc; (11) HST programs: GO14201 (PI S. Malhotra), GO13744 (PI T. Thuan; Izotov et al. 2016),GO13293 (PI A. Jaskot; Jaskot et al. 2014), GO12928 (PI A. Henry; Henry et al. 2015), GO11727 and GO13017 (PI T. Heckman; Heckman et

al. 2011; Alexandroff et al. 2015). These 43 galaxies are sorted by decreasing fLyesc from top to bottom. The machine readable table is availableonline.a These are confirmed LyC leakers from Izotov et al. (2016).

15

TABLE 2The line measurements from SDSS spectra

ID [OII]3727 [OIII]4363 H [OIII]4959 [OIII]5007 H EW(H) [OIII]/[OII](1) (2) (3) (4) (5) (6) (7) (8) (9)

1333+6246 1155 13.62.7 584 1292 3906 784 538 4.51559+0841 1694 9.41.3 12412 2292 6936 23211 288 5.51219+1526 4678 108.84.9 7769 16358 495325 220718 744 14.21514+3852 2705 9.23.1 1393 2482 7516 3276 232 3.71503+3644 2203 19.31.6 1842 3972 12037 53421 921 7.114420209 2485 32.22.4 2994 6475 196014 98813 858 9.01133+6514 2685 19.22.3 1963 3762 11387 5926 263 5.31249+1234 5758 28.82.4 3645 7405 224214 116912 717 4.61009+2916 1394 22.02.3 1734 3823 11569 47316 422 11.10815+2156 2935 56.43.3 4625 10657 322722 138313 717 14.01424+4217 112916 114.93.6 111911 246311 745933 333325 629 8.40926+4428 109014 56.33.5 7338 13187 399422 231418 437 4.41152+3400 2374 22.51.3 2283 4612 13977 7565 497 6.70021+0052 517229 127.86.7 290914 427512 1294936 885532 320 3.11122+6154 2575 11.61.3 2004 3693 111810 66710 495 4.90925+1403 2977 25.84.0 2824 5964 180611 96010 633 6.70911+1831 57610 15.53.2 3795 4423 13409 134314 348 2.50917+3152 3005 6.22.3 2103 2441 7394 7607 250 2.51137+3524 151917 51.12.7 94110 15637 473321 286521 434 3.91025+3622 181617 60.74.5 103810 174610 528931 331825 312 3.41440+4619 89511 19.53.0 4415 6374 192912 151314 325 2.41429+0643 224523 152.36.8 178515 350315 1061046 552437 686 5.81054+5238 106813 32.53.3 6617 9826 297417 206816 304 3.41428+1653 157417 19.63.0 7068 7334 222013 251120 261 1.503030759 4888 74.02.5 6567 13018 394123 196318 608 10.31244+0216 125212 64.13.3 8537 16818 509125 266518 667 4.92237+1336 7339 19.32.7 3764 5874 178011 129112 353 2.61454+4528 4988 9.23.2 2934 4013 12158 103311 277 2.61018+4106 2925 28.51.9 2633 5393 163310 8468 570 6.50751+1638 2166 10.73.6 1153 1872 5675 4017 299 2.80822+2241 106311 54.03.1 7816 15516 469919 288617 605 4.41339+1516 6028 61.43.2 6677 14856 449918 222257 523 8.41543+3446 1856 15.42.3 1945 3434 103711 4807 342 7.503390725 97812 14.72.3 5386 7334 222013 178615 345 2.60938+5428 330528 67.94.0 188715 262713 795739 631339 353 2.70927+1740 3286 17.72.1 1964 4193 12689 7079 707 4.01457+2232 7649 103.12.7 86812 209612 634936 275820 707 9.90749+3337 131214 20.32.4 6527 8115 245714 244727 361 1.91032+2717 8458 25.10.8 5205 9104 275713 168713 651 3.80805+0925 735 4.33.5 724 1233 3718 3336 353 4.01205+2620 3245 9.03.7 1653 1981 5994 5878 350 1.900550021 173811 29.03.0 9565 11973 362610 358712 249 2.10747+2336 3635 35.81.8 3545 7965 241116 116611 366 7.6

Note. Observed line fluxes from SDSS spectra in units of 1017 erg s1 cm2. The EW(H) is rest-frame H equivalent width. The[OIII]/[OII] ratio are extinction corrected using the Calzetti et al. (2000) extinction law. The machine readable table is available online.

16

TABLE 3Ly properties

ID Ly flux log(L(Ly) erg s1) EW(Ly) fLyesc V(blue-peak) V(red-peak) FWHM(red)1016 erg s1 cm2 A km s1 km s1 km s1

(1) (2) (3) (4) (5) (6) (7) (8)

1333+6246a 160.42.8 42.7 72.3 1.180 d 7917 245241559+0841 145.03.1 42.6 96.0 0.735 35524 16817 188291219+1526 1345.35.9 43.2 164.5 0.702 7613 17613 213181514+3852 180.84.1 42.8 60.0 0.698 e 15917 222581503+3644a 195.24.3 42.9 106.6 0.431 34943 11823 2292714420209a 504.55.6 43.1 134.9 0.430 d 13526 267251133+6514 208.01.9 42.6 42.3 0.422 6915 27122 234211249+1234 528.02.6 43.1 101.8 0.412 d 8327 364171009+2916 142.82.5 42.3 69.5 0.373 11645 14422 206260815+2156 401.21.4 42.3 82.2 0.327 12113 14413 216191424+4217 858.64.1 42.9 89.5 0.290 15032 22410 208160926+4428 636.82.3 42.8 47.8 0.287 16551 24417 327181152+3400a 248.64.6 43.0 74.5 0.287 14626 15844 235300021+0052 1523.59.7 42.6 32.8 0.215 41838 16412 253161122+6154 144.12.1 42.2 60.0 0.187 5621 19426 202250925+1403a 225.14.1 42.8 90.0 0.186 14524 13323 160250911+1831 315.72.1 42.8 56.5 0.177 27817 8112 207170917+3152 167.73.3 42.7 38.0 0.169 20942 10422 189251137+3524 381.13.4 42.6 40.4 0.158 35546 20122 285201025+3622 436.63.6 42.3 26.3 0.154 24839 21012 253151440+4619 214.23.6 42.8 33.8 0.128 d 6729 248301429+0643 607.12.9 42.7 42.7 0.123 25734 23119 324191054+5238 153.52.6 42.5 17.7 0.112 31488 19212 225251428+1653 311.92.2 42.5 29.1 0.106 36025 15020 2421803030759 99.62.1 41.9 14.2 0.098 31348 15326 270231244+0216 189.91.6 42.5 47.0 0.077 24014 24714 302252237+1336 51.42.6 42.1 15.3 0.063 d 14136 272351454+4528 72.32.1 42.2 30.0 0.061 5658 44418 202471018+4106 47.01.5 41.9 33.1 0.059 30644 20625 238320751+1638 13.91.3 41.5 15.8 0.043 e e e0822+2241 156.52.9 42.3 51.6 0.037 30465 21731 262391339+1516 82.51.9 41.9 44.7 0.034 35130 25642 435621543+3446b 10.60.8 41.0 5.4 0.024 e 26194 407900938+5428b 107.12.0 41.5 3.5 0.013 27920 39017 333430927+1740b 14.01.1 41.6 7.2 0.013 e 24291 4081501457+2232b 32.30.6 41.3 5.3 0.010 32937 40614 321540749+3337 9.21.7 41.3 8.9 0.010 42772 56872 4051141032+2717b 19.20.9 41.3 5.5 0.009 e e e0805+0925b 9.51.3 41.5 9.2 0.009 e e e1205+2620b 5.81.3 41.4 3.0 0.006 e e e00550021b 31.31.0 41.4 3.2 0.005 e 36845 4203903390725c 1.41.8 0747+2336c

Note. Column Descriptions: (1) Object ID; (2) Ly emission line flux; (3) Ly emission line luminosity; (4) equivalent width of Ly line; (5)Ly escape fraction; (6) Velocity of Ly blue peak; (7) Velocity of Ly red peak; (8) FWHM of the red portion of Ly profile. These 43 galaxies

are sorted by decreasing fLyesc from top to bottom. The machine readable table is available online.a These are confirmed LyC leakers from Izotov et al. (2016).b These Green Peas show damped Ly absorption wings in their Ly spectra.c No Ly emission line was detected.d Their Ly profiles dont have blue peaks.e Their Ly profiles are too noisy for measuring Ly kinematic features.

17

TABLE 4Ly profile Model Parameters

ID log(NHI cm2) Vexp log(T) d

(km s1) (K) km s1

(1) (2) (3) (4) (5) (6)

1333+6246 19.39+0.080.07 270+44 5.0

+0.20.1 0.71

+0.090.08 125

+22

1559+0841 19.40+0.070.07 90+34 3.0

+0.10.1 0.64

+0.080.06 203

+22

1514+3852 19.20+0.070.07 80+43 3.8

+0.10.1 0.01

+0.010.00 305

+77

1503+3644 16.81+0.080.08 140+44 5.4

+0.20.1 0.14

+0.070.05 266

+55

14420209 18.80+0.070.07 150+44 4.2

+0.20.2 0.01

+0.010.00 230

+22

1009+2916 19.60+0.070.07 30+44 3.4

+0.10.1 0.00

+0.000.00 201

+33

1152+3400 20.00+0.070.07 5+11 3.4

+0.20.1 0.01

+0.000.00 333

+66

0021+0052 19.59+0.070.07 130+43 5.0

+0.10.1 0.22

+0.010.01 100

+22

1122+6154 19.98+0.081.85 7+126 3.4

+0.20.2 0.00

+1.130.00 259

+45

0925+1403 19.79+0.070.08 8+11 3.0

+0.20.2 0.03

+0.000.00 229

+55

0917+3152 19.00+0.070.07 60+33 3.5

+0.30.2 0.01

+0.010.01 275

+88

1025+3622 19.40+0.070.07 110+34 4.1

+0.20.3 0.00

+0.010.00 228

+75

1440+4619 19.18+0.080.08 259+44 5.0

+0.10.1 0.23

+0.030.03 117

+22

1429+0643 20.39+0.080.08 15+22 3.4

+0.10.1 0.00

+0.000.00 392

+33

1428+1653 16.04+1.550.09 168+511 5.7

+0.20.4 0.10

+0.020.09 205

+204

2237+1336 19.88+0.150.13 258+4410 4.9

+0.21.1 1.14

+0.860.33 140

+128

1454+4528 16.44+0.360.09 -171+44 5.4

+0.10.1 4.78

+0.170.34 427

+108

1018+4106 19.60+0.070.07 49+44 3.8

+0.20.2 0.07

+0.040.04 268

+1211

0751+1638 19.39+0.070.08 121+1720 4.4

+0.31.3 0.17

+0.340.13 326

+3129

0822+2241 20.54+0.110.18 6+32 3.0

+0.10.2 0.01

+0.020.00 363

+1019

1339+1516 19.00+0.070.07 100+44 3.1

+0.60.2 4.86

+0.110.23 345

+55

0938+5428 20.60+0.080.08 15+22 4.6

+0.20.2 0.08

+0.010.01 64

+1510

00550021 20.40+0.070.07 60+34 4.2

+0.10.1 0.07

+0.020.02 322

+1717

Note. Column Descriptions: (2) HI column density of the outflowing HI shell; (3) outflowing velocity of the HI shell; (4) HI gas temperatureincluding turbulent motion as well as the true temperature; (5) dust optical depth; (6) 1 width of the Gaussian profile of the intrinsic Ly line.

These 23 galaxies are sorted by decreasing fLyesc from top to bottom.

ABSTRACT1 Introduction2 Sample and Observations2.1 The Sample2.2 Properties from SDSS Spectra2.3 HST/COS Observation

3 Ly Equivalent Width and Escape Fraction3.1 Measurements of Ly flux, EW, and escape fraction3.2 Ly EW distribution of Green Peas

4 Ly escape and Ly profiles4.1 Kinematic Features of Ly Profile4.2 Relations between Ly escape and Ly kinematics

5 Ly escape and other galactic properties5.1 dust extinction, stellar mass, and metallicity5.2 Morphology and size of UV emission5.3 [OIII]/[OII] ratio

6 Ly profile fitting7 Predicting Ly escape fraction8 Conclusion