Lyα EMISSION FROM GREEN PEAS: THE ROLE OF CIRCUMGALACTIC GAS DENSITY, COVERING, AND KINEMATICS^*

The Lyα emission line is a heavily used diagnostic in studies of high-redshift galaxies and the intergalactic medium (IGM). At redshifts z ≳ 2, this feature is shifted into readily observable optical wavelengths, leading many studies to rely on Lyα for both discovery and spectroscopic redshift confirmation of galaxies. Lyα-based surveys are now extending to z ∼ 6 and beyond, where they are identifying the building blocks of present-day galaxies and the sources that are most important for metal enrichment and reionization of the IGM (Martin et al. 2008; Ouchi et al. 2010; Hu et al. 2010; Dressler et al. 2011, 2015; Kashikawa et al. 2011; Henry et al. 2012; Rhoads et al. 2012). In addition, because Lyα photons scatter resonantly in neutral hydrogen, they offer a wealth of information about this gas. Spectroscopic line profiles can constrain gas content and kinematics (Erb et al. 2014; Verhamme et al. 2015; Martin et al. 2015), and Lyα imaging can illuminate the gaseous halos around galaxies (Steidel et al. 2011; Zheng et al. 2011; Hayes et al. 2013; Momose et al. 2014). In the epoch of reionization, weaker or less frequent Lyα emission may indicate a significant neutral hydrogen fraction in the IGM (Ota et al. 2010; Pentericci et al. 2011; Schenker et al. 2012; Treu et al. 2012, 2013), an increase in the escape fraction of ionizing photons, or both (Dijkstra et al. 2014).

Despite the fact that a wide range of studies rely on Lyα emission, we do not understand how to interpret the feature. The Lyα luminosity, equivalent width, W_Lyα, and Lyα/Hα flux ratio of galaxies are seen to vary widely, and are difficult to predict from other galaxy properties like star formation rate (SFR) and dust extinction (Giavalisco et al. 1996; Scarlata et al. 2009). Qualitatively, it is believed that the Lyα emission line is modified by resonant scattering in the interstellar medium (ISM) and circumgalactic medium (CGM). This process increases the optical depth of Lyα photons, making them more susceptible to dust absorption, and sometimes turns the emission profile into absorption. Alternatively, in the absence of dust, the Lyα photons may simply diffuse to large radii and be missed by poor sensitivity to low surface brightness emission (Steidel et al. 2011; Hayes et al. 2013). Besides dust and geometry of the H i gas, the kinematics of galaxy outflows may also play a role, since Lyα photons may escape more easily when scattered by neutral hydrogen that is out of resonance with the ISM (Kunth et al. 1998; Wofford et al. 2013). Nevertheless, our poor quantitative understanding of Lyα output is cause for concern when studies try to draw robust conclusions about galaxies and the IGM from this feature.

Efforts to understand Lyα escape from high-redshift galaxies have generally suffered from the limitations associated with studying faint, distant sources (e.g., Shapley et al. 2003; Erb et al. 2010, 2014; Jones et al. 2012, 2013). Fortunately, many of these difficulties can be overcome when studying nearby galaxies. If low-redshift galaxies can be found where the physical conditions are similar to high-redshift star-forming galaxies, it is possible to learn more about the astrophysics that regulates Lyα output. Despite this realization, most investigations to date have struggled to identify local galaxies that are similar to the low-mass, low-metallicity, high specific SFR galaxies that dominate the cosmic star formation at early times (e.g., Alavi et al. 2014; Stark et al. 2014). Instead, present samples of nearby galaxies with Lyα observations are, on average, more massive, dustier, metal-rich, and have lower specific SFRs than the bulk of high-redshift galaxies (Giavalisco et al. 1996; Scarlata et al. 2009; Cowie et al. 2011; Heckman et al. 2011; Hayes et al. 2013, 2014; Wofford et al. 2013; Martin et al. 2015). As a possible consequence of this selection, Lyα luminosities in low-redshift samples—when measured—are around an order of magnitude lower than the luminosities of most Lyα-selected galaxies observed at z > 3 (Gronwall et al. 2007; Hu et al. 2010; Ouchi et al. 2010; Dressler et al. 2011, 2015; Henry et al. 2012).

The recent discovery of extremely high equivalent width emission line galaxies (the "Green Peas") within the Sloan Digital Sky Survey (SDSS) offers a new avenue to investigate Lyα emission from high-redshift analogs (Cardamone et al. 2009). With masses reaching below 10⁹ M_⊙, metallicities lower than 12 + log(O/H) ∼ 8.2 (from direct T_e measurements; Izotov et al. 2011), and Hα equivalent widths (W_Hα) exceeding hundreds of Å, these local galaxies may be more representative of high-redshift Lyman alpha emitters (LAEs; W_Lyα ≳ 20 Å) and reionization epoch galaxies than other local samples. As a comparison, inferred emission line contamination of Spitzer/IRAC photometry provides tentative evidence for similar W_Hα in star-forming galaxies at z ∼ 4–7 (Shim et al. 2011; Labbé et al. 2013; Stark et al. 2013; Smit et al. 2014). Ultimately, the James Webb Space Telescope will clarify which segment of the high-redshift population is most analogous to the Green Peas. At present, however, quantifying the Lyα output from these nearby objects is a critical benchmark for future comparisons.

In this paper we present a UV spectroscopic study of Green Peas using the Cosmic Origins Spectrograph (COS) on the Hubble Space Telescope (HST). For a sample of ten galaxies at 0.16 < z < 0.26, we measure the Lyα emission that escapes the galaxies from within a few kpc (corresponding to the COS aperture). At the same time, by observing absorption lines in both hydrogen and metals, these data allow us to quantify the role of galactic outflows, ISM, and CGM gas. This paper is organized as follows. In Section 2 we describe our sample selection and the galaxy properties derived from GALEX and SDSS; in Section 3 we describe the COS observations and resultant data and in Section 4 we present the Lyα line profiles and measurements; Section 5 explores how the stellar population properties and dust of the Green Peas compare to other nearby samples that have been observed in Lyα. In Sections 6 and 7 we show how the strength of the Lyα line varies with interstellar absorption equivalent width and kinematics, and discuss the velocity structure of the Lyα line. Finally, Section 8 offers a comprehensive interpretation of the data and our conclusions are summarized in Section 9.

Throughout this paper we use a Chabrier (2003) initial mass function (IMF), and a ΛCDM cosmology with ${{\rm{\Omega }}}_{M}=0.3$, ${{\rm{\Omega }}}_{{\rm{\Lambda }}}=0.7$, and H₀ = 70 km s⁻¹ Mpc⁻¹. We adopt the equivalent width sign convention used in literature focused on high-redshift galaxies: positive equivalent widths indicate emission while negative values are used to signify absorption. All COS, GALEX, and SDSS data are corrected for Milky Way extinction using attenuation measured by Schlafly & Finkbeiner (2011) and the Fitzpatrick (1999) extinction law. The foreground extinction values were obtained from the NASA Extragalactic Database,⁶ and are listed in Table 1.

Table 1.  Green Pea Sample and Observations

ID	R.A.	Decl.	z	$E(B-V)$MW	Rest Wavelength Coverage	G130M Exposure	G160M Exposure
	(J2000)	(J2000)		(mag)	(Å)	(s)	(s)
0303–0759	03 03 21.41	−07 59 23.2	0.164887	0.0877	975–1515	2190	3829
1244+0216	12 44 23.37	02 15 40.4	0.239420	0.0213	945–1430	2042	6507
1054+5238	10 53 30.80	52 37 52.9	0.252645	0.0132	910–1425	824	2736
1137+3524	11 37 22.14	35 24 26.7	0.194396	0.0161	965–1505	1264	2340
0911+1831	09 11 13.34	18 31 08.2	0.262236	0.0248	900–1435	2074	6530
0926+4427	09 26 00.44	44 27 36.5	0.180698	0.0165	970–1505	5640	6180
1424+4217	14 24 05.72	42 16 46.3	0.184800	0.0094	965–1220	1209	0a
1133+6514	11 33 03.80	65 13 41.4	0.241400	0.0097	945–1430	1232	4589
1249+1234	12 48 34.63	12 34 02.9	0.263403	0.0252	900–1425	1644	6372
1219+1526	12 19 03.98	15 26 08.5	0.195614	0.0239	965–1505	716	2304

Note.

^aFailed observation.

Download table as:  ASCIITypeset image

The present sample of Green Peas were drawn from the catalog presented by Cardamone et al. (2009). To systematically identify objects originally discovered by the Galaxy Zoo (Lintott et al. 2008), Cardamone et al. defined a color selection for objects with strong [O iii] emission in the SDSS r-band. Following this selection, objects with low signal-to-noise ratio (S/N) SDSS spectra were removed and active galactic nuclei (AGNs) were rejected on the basis of broad emission lines or their optical emission line ratios (Hα/[N ii] versus [O iii]/Hβ; Baldwin et al. 1981). The remaining sample consists of 80 galaxies with rest-frame [O iii] λ5007 equivalent widths of hundreds to over 1000 Å.

We selected Green Peas that were bright enough in the FUV to enable continuum detection with COS. From the 80 star-forming Green Peas in Cardamone et al. (2009), we considered objects with GALEX photometry (GR6) and m_FUV ≤ 20 AB. We also required z < 0.27, so that the FUV spectra would cover the Si iv λλ1393, 1403 lines. The resulting sample of ten galaxies is listed in Table 1. We verified that these galaxies have stellar masses, SFRs, metallicities, and [O iii] equivalent widths consistent with the parent sample of Green Peas (Cardamone et al. 2009; Izotov et al. 2011). However, because of the FUV magnitude limit, the average UV luminosity in the present sample is around 1.6 times higher than the average (GALEX detected) Green Peas in Cardamone et al. (2009).

In Table 2 we compile properties of the Green Peas, drawn from the literature or derived from GALEX and SDSS. We describe these quantities below. In most cases, the high S/N of these data implies insignificant statistical errors on the measurements. Therefore we do not list these uncertainties.

Table 2.  Green Pea Properties from GALEX+SDSS

ID	WHα	Hα/Hβ	$E{(B-V)}_{\mathrm{gas}}$	log(LHα/erg s−1)	SFR	Log(M/M☉)	12 + log(O/H)	MFUV	β	RUV
	(Å)		(mag)		(M☉ yr−1 )			(mag)		(kpc)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
0303–0759	670	2.78	0.00	42.24	7.6	8.89	7.86	−20.35	−2.23	0.8
1244+0216	840	3.10	0.07	42.78	26.2	9.39	8.17	−20.32	−1.70	2.6
1054+5238	400	3.15	0.08	42.71	22.4	9.51	8.10	−21.31	−1.94	1.3
1137+3524	580	3.08	0.06	42.58	16.8	9.30	8.16	−20.56	−1.78	1.8
0911+1831	420	3.50	0.17	42.68	21.1	9.49	8.00	−20.56	−1.82	1.1
0926+4427	610	3.20	0.10	42.49	13.6	8.52	8.01	−20.58	−1.98	1.0
1424+4217	1100	3.01	0.04	42.57	16.5	8.08	8.04	−20.40	−1.90	1.0
1133+6514	300	2.90	0.01	42.02	4.6	9.04	7.97	−20.40	−1.93	1.9
1249+1234	670	3.09	0.07	42.50	13.8	8.79	8.11	−20.25	−1.82	1.8
1219+1526	1270	2.87	0.00	42.43	11.9	8.09	7.89	−19.94	−1.65	0.7

Note. Quantities are derived from the SDSS and GALEX, or taken from Izotov et al. (2011). Measurements errors are dominated by systematics in these high S/N data, so statistical errors are not given. Emission line measurements include both broad and narrow components, as described in Section 2. Column descriptions: (1) object ID; (2) rest-frame Hα equivalent width, in Å, measured from SDSS spectrum; (3) flux ratio, F(Hα)/F(Hβ) measured from SDSS spectrum; (4) nebular gas extinction inferred from the Hα/Hβ flux ratio, assuming the intrinsic ratio of 2.86 as described in the text; (5) Hα luminosity measured from the SDSS spectrum; (6) SFR estimated from L_Hα using the Kennicutt (1998) calibration, divided by 1.8 to convert from a Salpeter to Chabrier (2003) IMF; (7) stellar mass from Izotov et al. (2011), also converted from a Salpeter to Chabrier (2003) IMF; (8) direct, T_e([O iii])-based metallicities taken from Izotov et al. (2011); (9) FUV absolute magnitude (AB), calculated from GALEX photometry, interpolated to provide an estimated rest-frame 1500 Å luminosity; (10) UV slope, defined as F_λ ∝ λ^β, calculated from GALEX data; (11) NUV Petrosian radius (see text in Section 2), calculated from the COS acquisition image.

Download table as:  ASCIITypeset image

First, we chose to remeasure the emission lines in the SDSS spectra rather than use catalog measurements. One of the Green Peas in our sample (0303–0759) contains an unphysical Balmer decrement (an Hα/Hβ flux ratio of 2.25) in the MPA-JHU DR7⁷ catalogs, which is the result of an incorrect Hα flux. This failure may be attributable to difficulty in fitting the nearly blended Hα+ [N ii] lines with single Gaussians, when the lines clearly show evidence of broad wings extending hundreds of km s⁻¹ (also noted by Amorin et al. 2012 and Jaskot & Oey 2014). Therefore, for each of the Green Peas in this paper, we refit the emission line spectra with two kinematic components. We defined a kinematic model with a narrow component centered at the systemic redshift (v = 0), and a broad component that is allowed to be offset from v = 0. For 25 of the strongest lines in the spectrum (Hα through H9, [S ii], [N ii], [O iii], [O ii], [Ne iii], and 7 He i lines), the amplitudes of the broad and narrow components were allowed to vary, but their velocity widths and centroids were required to be the same from line to line. Hence, the resultant model has 54 free parameters (redshift, Doppler shift of the broad component, velocity widths of both the broad and narrow component, and two Gaussian amplitudes for each line). This model was fit to the continuum-subtracted spectrum provided by the SDSS, thereby measuring redshifts (Table 1), emission line fluxes (for the combined broad and narrow kinematic components), and the kinematics of the ionized gas. Equivalent widths are evaluated by taking the ratio of the emission line flux to the median continuum flux density in a 100 Å window centered on each line. The kinematic measures derived from our fits are similar for all of the Green Peas, showing a narrow component at the systemic redshift with a FWHM around 200 km s⁻¹, and a broad component, also centered near v ∼ 0, with a FWHM around 500 km s⁻¹. For the isolated Hβ, line, the broad component contributes 15%–35% of the total line flux in these galaxies.

Table 2 lists the Hα equivalent width and the Hα to Hβ flux ratio. From this latter quantity, we derived dust extinction assuming a Calzetti et al. (2000) extinction curve and an intrinsic ratio of F(Hα)/F(Hβ) = 2.86. This intrinsic ratio is appropriate for electron temperature T_e = 10⁴ K and density n_e = 10² cm⁻²; increasing T_e two-fold decreases the intrinsic Balmer decrement to 2.75 and increases the extinction, $E{(B-V)}_{\mathrm{gas}}$, by approximately 0.03 mag. Dust-corrected Hαluminosities are given, and converted to SFR using the calibration given in Kennicutt (1998; adjusted to a Chabrier 2003 IMF).

The stellar masses of the Green Peas are taken from Izotov et al. (2011). These authors note that, for emission lines of high equivalent width, the corresponding bound–free continuum emission can be comparable to the stellar continuum. Therefore previous estimates of stellar mass in these galaxies may be too high; to correct for this mis-estimation, Izotov et al. derived stellar masses from fits to the SDSS spectra, using models that include nebular continuum. For consistency, we have decreased these masses by a factor of 1.8 to convert them from a Salpeter to Chabrier (2003) IMF. Oxygen abundances are also taken from Izotov et al. (2011), where they were derived from [O iii] electron temperature ([O iii] λ4363). All 10 of the Green Peas in this paper show detections in this line, confirming their low metallicities.

The ultraviolet continuum luminosity and slope were derived from GALEX photometry (FUV: λ ∼ 1340–1790 Å, NUV: λ ∼ 1770–2830 Å). The FUV fluxes were corrected for the contribution from Lyα emission of the Green Peas (described in Section 4), and both bands were corrected for Galactic foreground extinction (as noted in Section 1). From the corrected FUV–NUV color, we calculated the UV power-law slope, β (F_λ ∝ λ^β).⁸ The absolute FUV magnitude was calculated by interpolating between the FUV and NUV bands to obtain the luminosity, L_ν, at 1500 Å.

In the final column of Table 2 we list R_UV, the size of the galaxy measured from the NUV acquisition images. For the purposes of comparison in Section 5, we choose a Petrosian radius, r_P20 as adopted by Hayes et al. (2014). This radius defines the circular isophote where the local surface brightness is 20% of the internal surface brightness.

3.1. Observations

The observations in this paper are part of two COS programs. One of the galaxies that met our selection (0926+4427) was also classified as a Lyman Break Analog (LBA; Heckman et al. 2011), and was previously observed with COS (GO 11727; PI T. Heckman). The remaining nine Green Peas were observed as part of GO 12928 (PI A. Henry).

For all 10 of the galaxies, the target acquisition was accomplished with NUV imaging, configured with the Primary Science Aperture and Mirror A. As detailed in the COS Instrument Handbook, an initial image is obtained and analyzed to find the peak NUV flux. Next, HST is moved to place the peak flux in the center of the COS aperture, and a second image is taken to verify the shift. The target acquisition images of the Green Peas are shown in Figure 1. These data indicate relatively compact emission, implying that most of the UV continuum emission from these galaxies falls within the central 1'' diameter of the COS aperture (shown as circles) where vignetting is minimal. (Likewise, if optical line emission closely follows the UV continuum, we expect negligible aperture losses in the 3'' SDSS fiber spectroscopy.)

Zoom In Zoom Out Reset image size

The spectra were obtained in the FUV, using both the G130M and G160M configurations (in 9/10); with both gratings the rest-frame wavelength coverage spans from approximately 950–1450 Å, or somewhat shorter/longer depending on the redshifts. The rest-frame wavelengths covered for each galaxy are listed in Table 1, along with the exposure time for each grating. For 1424+4217, the second orbit containing the G160M observations failed; it did not qualify for a repeat observation because the program was more than 90% complete. In this case, the observations cover the Green Pea Lyα emission only for velocities v ≲ 750 km s⁻¹. As recommended by the COS Instrument Handbook, all spectra were taken at four FP-POS settings in each grating in order to mitigate the effects of fixed pattern noise. In addition, the central wavelengths for each grating were chosen to avoid placing strong absorption lines between the A and B segments. For one galaxy (1054+5328) this choice was not possible in G130M, so two central wavelengths were used. Since CALCOS does not combine data taken at different central wavelengths, we combined them by simply taking the mean (except in the gaps, where we adopt the values from the spectrum that has coverage). The errors were derived from taking the Poisson noise on the total counts, as we will describe below in Section 3.3.

All data from GO 12928 were taken at the COS lifetime position two; they were downloaded from MAST on 2013 August 23, implying that they were processed with CALCOS version 2.19.7. The spectra from GO 11727 were taken in lifetime position one, and were downloaded on 2012 November 2 (CALCOS version 2.18.5).

The full COS spectra for all 10 Green Peas are shown in the Appendix.

3.2. Estimating Spectral Resolution

3.3. Binning and Noise

3.4. Velocity Precision

4.1. The Prevalence of Strong, Double-peaked Lyα Emission

Figure 2 shows the Lyα spectra of the Green Peas in the present sample, displayed in order of increasing Lyα escape fraction (estimated below). Their appearance is remarkable, for a few reasons. First, all ten of the galaxies show Lyα in emission. This result is not trivial; typical star-forming galaxies at all redshifts show a range of Lyα strength, ranging from pure (even damped) absorption to P-Cygni emission plus absorption, to pure emission (Pettini et al. 2002; Shapley et al. 2003; Wofford et al. 2013). On the other hand, the Green Peas are not typical of nearby star-forming galaxies, so the Lyα emission seen in all ten galaxies is not entirely surprising. Nearby, optical emission lines of high equivalent width, like those in the Green Peas, have been suggested to favor Lyα emission (Cowie et al. 2011). Furthermore, in more distant galaxies at 4 < z < 6, the increasing incidence and strength of Lyα emission (Stark et al. 2011) may go hand in hand with optical emission lines of high equivalent width (inferred from contamination to broad-band photometry; Shim et al. 2011; Labbé et al. 2013; Stark et al. 2013). The detection of Lyα emission in all ten Green Peas is consistent with these claims.

Zoom In Zoom Out Reset image size

The second noteworthy feature about the spectra in Figure 2 is the prevalence of the double-peaked line shape. Of the Green Peas that we observed with COS, 9/10 share the spectral morphology with both redshifted and blueshifted emission peaks. Again, this line shape is not typical of normal nearby star-forming galaxies. Wofford et al. (2013) find only one doubled-peaked emitter in their sample of 20 Hα-selected objects around z ∼ 0.03. And Martin et al. (2015), in a study of 8 nearby ULIRGs, show that Lyα can exhibit complicated kinematic profiles with broad blueshifted emission in more dusty environments. In contrast, at higher redshifts (z ∼ 2–3), Kulas et al. (2012) report that 30% of UV continuum-selected galaxies with Lyα emission show multiple-peaked profiles. Comparison to the Green Peas, however, reveals the importance of spectral resolution in this measurement. At the 200–500 km s⁻¹ resolution used by Kulas et al. (2012), many of the Green Pea spectra would be observed as single-peaked lines. Indeed, spectroscopy with 120 km s⁻¹ resolution showed double-peaked line profiles in 3/3 LAEs targeted by Chonis et al. (2013). Nevertheless, the Lyα line profiles of the Green Peas are notably different than the other low-redshift samples, where W_Hα is lower and dust content is higher (Wofford et al. 2013; Martin et al. 2015; Rivera-Thorsen et al. 2015). As indicated by Lyα radiative transfer models, the blue peak should appear when the H i column density is low, plausibly due to anisotropies in the gas distribution (Behrens et al. 2014; Verhamme et al. 2015; Zheng & Wallace 2014).

In order to better understand how the unusual conditions in the Green Peas are influencing their Lyα escape, we next provide quantitative measurements of the Lyα lines and show how these compare to nearby samples. Then, in the remainder of the paper we will use the UV absorption lines to explore how the conditions in the ISM and CGM affect the Lyα emission. Radiative transfer models of the Lyα spectral line profiles will follow in I. Orlitová et al. (2015, in preparation).

4.2. Lyα Emission Line Measurements

The Lyα emission line measurements are presented in Table 3. The line flux is measured by directly integrating the emission line spectra out to the velocity where the continuum is met. We adopt a conservative 20% uncertainty on the continuum flux near Lyα, since the broad N v 1240 Å stellar wind feature makes continuum normalization difficult. This error makes little difference for the Lyα flux uncertainties, but it dominates the error budget for measurements of equivalent width. For 1424+4217, where the G160M spectrum was unavailable, we use the G130M observations even though they only cover Lyα for v ≲ 750 km s⁻¹. Since the other galaxies show only a small contribution at these velocities, we do not correct Lyα measurement for missed flux.

Table 3.  Lyα Measurements from COS Spectroscopy of Green Pea Galaxies

ID	${F}_{\mathrm{Ly}\alpha }$	${L}_{\mathrm{Ly}\alpha }$	Lyα/Hα	${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$	${W}_{\mathrm{Ly}\alpha }$	${W}_{\mathrm{Ly}\alpha }^{\mathrm{red}}$	${W}_{\mathrm{Ly}\alpha }^{\mathrm{blue}}$		vredpeak	vbluepeak	vmaxred	vmaxblue
	(10−14 erg s−1 cm−2)	(1042 erg s−1)			(Å)				(km s−1)
0303–0759	1.1 ± 0.2	0.8	0.5	0.05	9 ± 2	9 ± 2	−0.1 ± 0.3		170	−290	900	−400
1244+0216	2.0 ± 0.1	3.4	0.7	0.07	48 ± 10	36 ± 7	12 ± 2		250	−280	1100	−1000
1054+5238	1.7 ± 0.2	3.1	0.8	0.07	12 ± 3	12 ± 2	1 ± 1		160	−250	700	−700
1137+3524	3.8 ± 0.2	4.0	1.2	0.12	35 ± 7	33 ± 7	3 ± 1		150	−400	1100	−800
0911+1831	3.3 ± 0.1	6.8	2.3	0.16	59 ± 12	48 ± 10	8 ± 2		90	−280	1100	−800
0926+4427	6.0 ± 0.3	5.4	2.3	0.20	40 ± 8	36 ± 7	5 ± 1		160	−250	1200	−800
1424+4217	8.5 ± 0.2	8.0	2.4	0.25	95 ± 19	57 ± 12	35 ± 7		230	−150	...	−850
1133+6514	2.1 ± 0.1	3.6	3.5	0.40	36 ± 7	25 ± 5	11 ± 2		230	−100	1000	−600
1249+1234	5.4 ± 0.1	11.3	4.4	0.41	98 ± 20	74 ± 15	16 ± 3		70	...	1300	−700
1219+1526	13.7 ± 0.2	14.7	5.5	0.62	164 ± 33	118 ± 24	41 ± 8		140	−100	1300	−950

Note. Lyα measurements from the COS spectra of the Green Peas give the fraction of Lyα that escapes from within the few kpc probed by the spectroscopic aperture. The Lyα fluxes and equivalent widths are calculated by directly integrating the line profile to the velocity where it meets the continuum, and W_Lyα^red and W_Lyα^blue are calculated for v > 0 and v < 0, respectively. The fluxes and equivalent widths assume a conservative 20% error on the continuum flux. Additionally, Lyα kinematics give the velocities marking the red and blue peaks, as well as the maximal blue and red velocities where the emission reaches the continuum. The former velocities are good to better than 50 km s⁻¹ in all cases except for the red peak of 0926+4427, which appears impacted by foreground absorption. The maximal velocities are less certain, with typical errors around 200 km s⁻¹.

Download table as:  ASCIITypeset image

Although the compact sizes of the Green Peas suggests minimal aperture losses in the continuum spectra, the Lyα emission may be more extended. At the redshifts of our sample, the unvignetted portion of the COS aperture shown in Figure 1 (1'' diameter) corresponds to 2.8–4.0 kpc; the full COS aperture is 2.5 times larger. As a comparison, in the nearby Lyman Alpha Reference Sample (LARS), which comprises 14 galaxies with HST imaging of Lyα, all but two objects show emission extending to at least 10 kpc (Hayes et al. 2014; Östlin et al. 2014). Based on their curves of growth, a 3–4 kpc diameter aperture would capture only one third to one half of the Lyα flux. Indeed, aperture losses are confirmed for three galaxies in our sample. One Green Pea, 0926+4427, is also identified as LARS 14, and two (1133+6514 and 1219+1526) are serendipitously covered by GALEX grism observations. For the former, the COS aperture captured 40% of the total large-aperture luminosity estimated from LARS imaging (${L}_{\mathrm{Ly}\alpha }^{\mathrm{total}}\sim 1.4\times {10}^{43}$ erg s⁻¹). The latter two cases show that 60% and 75% of the GALEX grism flux is detected by COS. Because the physical extent of the Lyα emission may vary from galaxy to galaxy, and 7/10 of the Green Peas have no large-aperture measurements, we do not correct the COS measurements. Ultimately, though, the COS measurements are interesting because they tell us about the Lyα that is observed within the central few kpc. These quantities are important for comparing to high-redshift spectroscopic studies, where slits may subtend physical scales similar to the COS aperture.

Besides the line fluxes, Table 3 lists the Lyα equivalent width, luminosity, and the Lyα escape fraction, ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. The latter quantity is defined as the ratio of the observed Lyα luminosity to the intrinsic Lyα luminosity, ${L}_{\mathrm{Ly}\alpha }^{\mathrm{obs}}/{L}_{\mathrm{Ly}\alpha }^{\mathrm{int}}$ (see also Hayes et al. 2014). The intrinsic Lyα luminosity is inferred from the dust-corrected Hα luminosity times the intrinsic case-B ratio of ${L}_{\mathrm{Ly}\alpha }/{L}_{{\rm{H}}\alpha }\sim 8.7.$ ¹¹ We also include measurements of the red- and blue-side ${W}_{\mathrm{Ly}\alpha }$ for the 9/10 double-peaked Lyα lines. These quantities are calculated by directly integrating the emission profiles for v > 0 and v < 0 separately. In two cases (0303–0759 and 1054+5238) the equivalent width of the blue peak is consistent with zero; this finding is not an indication that the blue peak is undetected. Rather, the blue emission is weak, and there is net absorption around zero velocity.

Finally, Table 3 also lists kinematic signatures from the Lyα lines. The velocities of the blue and red peaks are given, and the broad wings of the lines are quantified by calculating the velocity where F_λ reaches the continuum.

In this section, we explore the role of dust and stars in regulating the Lyα output of the Green Peas. In this context, it is useful to compare the Green Peas to nearby galaxies that have also been observed in Lyα. We choose two samples with published Lyα measurements: LARS (Hayes et al. 2013, 2014; Östlin et al. 2014; Rivera-Thorsen et al. 2015) and nearby galaxies identified as LAEs (W_Lyα > 15–20 Å) from GALEX grism surveys (Deharveng et al. 2008; Scarlata et al. 2009; Cowie et al. 2011). We acknowledge that the selection effects and aperture size likely influence this comparison. On one hand, the GALEX LAEs were Lyα selected via slitless grism spectroscopy with a large 5'' FWHM point-spread function. In this configuration, we expect that most of the Lyα emission should be included in the measurements, but the sample is biased toward rare, strong Lyα-emitting galaxies. On the other hand, the measurements reported for the Hα-selected LARS galaxies are integrated inside Petrosian apertures that are defined using the $\mathrm{Ly}\alpha$ plus continuum images. These apertures, which range in diameter from 2.6 to 32 kpc, do not capture all of the Lyα emission detected at large radii in the LARS images. For the low-mass, low-dust, high ${W}_{{\rm{H}}\alpha }$ end of the LARS sample, the typical 4 kpc diameter aperture is not too different than the COS aperture at z ∼ 0.2. Indeed, for LARS 14/0926+4427, the Petrosian aperture Lyα flux from the LARS image is comparable to (83% of) the flux included in the COS spectrum. However, as we noted in Section 4, these measurements account for less than half of the total Lyα flux in the LARS image. Nevertheless, we conclude that the LARS Petrosian aperture measurements and COS/Green Pea spectra are still useful, because they give a sense of how much Lyα is escaping from the central regions of the galaxies. Finally, we acknowledge that Lyα from nearby COS-observed galaxies has also been presented by Wofford et al. (2013), but we exclude this sample from comparison because their close proximity (z ∼ 0.03) implies that the COS aperture subtends only 0.6 kpc (unvignetted). Indeed, their COS NUV continuum images show much greater spatial extent compared to the Green Peas, suggesting that the Lyα measurements are not easily compared to the other samples considered here.

A straightforward approach for quantifying Lyα escape is to directly compare Lyα flux to the Hα and Hβ emission that probes the nebular gas and dust. In Figure 3, we plot the Lyα to Hα flux ratio against the Hα to Hβ flux ratio for the Green Peas, LARS galaxies, and GALEX LAEs. The latter sample comprises 45 galaxies taken from Scarlata et al. (2009) and Cowie et al. (2011), where slit-loss corrections facilitate the comparison between Lyα and Hα fluxes. Additionally, we show predictions from case-B recombination theory, assuming extinction laws from Cardelli et al. (1989) and Calzetti et al. (2000). In the absence of resonant scattering, the flux ratios should follow these lines. We also show the clumpy dust models of Scarlata et al. (2009) which form loops when the limit of high clump optical depth returns the observed line ratios to the intrinsic case-B ratios. While these models are needed to explain some of the high Lyα /Hα ratios in dusty galaxies, they do not explain line ratios of the Green Peas, which have low Lyα to Hα for their dust content. Although the Green Peas show little to no dust, their Lyα to Hα flux ratios span a factor of 10. Two of the Green Peas, 1249+1234 and 0911+1831, fall close to the Lyα to Hα ratio that is predicted for their dust extinction, but the remaining eight show ratios that are too low to be explained by dust extinction alone.

Zoom In Zoom Out Reset image size

To clarify how the Green Peas sample a different range of galaxy properties than LARS and the GALEX LAEs, Figure 4 shows histograms in stellar mass, SFR, M_UV, UV slope, β, ${W}_{{\rm{H}}\alpha }$, $E{(B-V)}_{\mathrm{gas}}$, ${L}_{\mathrm{Ly}\alpha }$, the UV isophotal Petrosian radius, R_UV, as well as W_Lyα, and ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. The derivation of these properties is outlined in Section 2 for the Green Peas and Hayes et al. (2013, 2014) for LARS. For the GALEX LAEs, we take the M_UV, ${W}_{{\rm{H}}\alpha }$, ${L}_{\mathrm{Ly}\alpha }$, and W_Lyα from Tables 1 and 2 in Cowie et al. (2011; 44 galaxies). The SFR and $E{(B-V)}_{\mathrm{gas}}$ are calculated from the slit-loss corrected line fluxes noted above (from Scarlata et al. 2009 and Cowie et al. 2011; 45 galaxies), assuming a Calzetti et al. (2000) extinction law, an intrinsic Hα/Hβ ratio of 2.86, and the Kennicutt (1998) SFR calibration (converted to a Chabrier 2003 IMF). GALEX LAEs identified as AGNs have been excluded.

Zoom In Zoom Out Reset image size

Figure 4 shows that the Green Peas occupy a region of parameter space that is poorly sampled by other studies. Their stellar masses overlap with the low-mass end of the LARS galaxies, but extend to masses an order of magnitude smaller. At the same time, the Green Peas' SFRs and UV luminosities are, on average, higher than the LARS galaxies and GALEX LAEs. They are uniformly low in dust, with $E{(B-V)}_{\mathrm{gas}}\lt 0.2$ and β ∼ −2.0. Finally, Figure 4 shows that the sizes of the Green Peas are similar to the more compact half of the LARS galaxies. These different Green Pea properties seem to impact the Lyα output: the Lyα luminosities are an order of magnitude larger than are observed for LARS and the GALEX LAEs. In fact, among these three samples, only the Green Peas have Lyα luminosities in the range of most high-redshift LAEs: ${L}_{\mathrm{Ly}\alpha }\gtrsim {10}^{42.5}$ erg s⁻¹ (Ouchi et al. 2010). Moreover, W_Lyα and ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ show a broad range of values for the Green Peas, while the LARS galaxies are more peaked at low values. (The GALEX LAEs, by definition, exclude the low values of W_Lyα and ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$). In fact, Figure 5 shows that in the combined set of LARS and the Green Peas, W_Hα is strongly correlated with W_Lyα. The Spearman rank correlation coefficient¹² is 0.64 and a spurious correlation is rejected with a probability of 8.1 × 10⁻⁴.

Zoom In Zoom Out Reset image size

This observation that the low masses, high SFRs, and low dust content of the Green Peas may favor Lyα emission is consistent with trends reported in Cowie et al. (2011). These authors compared GALEX LAEs to a UV-continuum-selected control sample and found that the LAEs had bluer colors, more compact sizes, lower metallicities, and higher ${W}_{{\rm{H}}\alpha }$ than their non-emitting counterparts. Similarly, Cowie et al. found Lyα emission to be more common in samples with higher W_Hα, and also reported a weak correlation between W_Hα and W_Lyα. Here, by actually comparing ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ instead of W_Lyα for the optical emission line-selected samples, we see that this correlation probably originates from increased Lyα escape, rather than young stellar populations with intrinsically high W_Lyα. At the same time, however, we detect no statistically significant correlation between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and W_Hα for the GALEX LAEs.

Regardless, a relation between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and W_Hα (or any of the other properties highlighted in Figure 4) does not explain the physical mechanism regulating Lyα escape. Rather, it suggests that gas properties associated with the youth of a stellar population increase Lyα escape. In the sections that follow, we use the ultraviolet interstellar absorption lines to further investigate the role of this gas.

The most common approach to studying the impact of outflows on Lyα is to observe low-ionization interstellar (LIS) metal lines. With ionization potentials less than 13.6 eV, ions such as Si ii and C ii can trace both neutral and ionized hydrogen. As such, these transitions are typically used to quantify the H i gas that scatters Lyα photons. In Figures 6–8, we show the COS spectra covering Si ii λλ1190, 1193, λ1260, and C ii λ1334. The spectra have been normalized by linear fits to the local continuum, and zero velocity is marked by a dashed vertical line. Similarly, dotted lines mark the expected locations of fluorescent fine-structure emission lines, Si ii* and C ii*. These lines form when electrons excited by absorption in the Si ii λλ1190, 1193, λ1260, and C ii λ1334 transitions subsequently decay to the excited ground state (followed by a fine-structure transition to the ground state). We exclude from our analysis the O i and Si ii lines at λ1302, λ1304, because their small wavelength separation and contribution from fluorescent and resonant re-emission (O i* λ1304) complicates the interpretation (see spectra shown in the Appendix). We list the rest-frame vacuum wavelengths, ionization potentials, absorption oscillator strengths, f_lu, and emission coefficients, A_ul, for these lines in Table 4. Additionally, we note that contamination to these ISM features by stellar absorption is negligible. In the top left panel of Figures 6–8, we show a 50 Myr old, continuous star-forming model spectrum, with Z = 0.002 (Leitherer et al. 2014). Similar stellar absorption is seen across a wide range of young, metal-poor stellar population properties, where UV spectra are dominated by O stars.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 4.  Observed UV Absorption and Emission Lines

Ion	Eion	Vacuum Wavelength	flu	Aul	${E}_{\mathrm{low}}-{E}_{\mathrm{up}}$
	(eV)	(Å)		(s−1)	(eV)
H i Ly	0	937.80	0.0078	1.64 × 106	0.00–13.22070331
H i Lyδ	⋯	949.74	0.014	4.12 × 106	0.00–13.0545011
H i Lyγ	⋯	972.54	0.029	1.28 × 107	0.00–12.7485388
H i Lyβ	⋯	1025.72	0.08	5.57 × 107	0.00–0 12.0875046
H i Lyα	⋯	1215.67	0.46	4.69 × 108	0.00–10.1988353
O i	0	1302.17	0.05	3.41 × 108	0.00–9.5213634
O i*	⋯	1304.86	0.05	2.03 × 108	0.0196224–9.5213634
		1306.03	0.05	6.76 × 108	0.0281416–9.5213634
Si ii	8.15	1190.42	0.277	6.53 × 108	0.00–10.415200
		1193.29	0.575	2.69 × 109	0.00–10.390117
		1260.42	1.22	2.57 × 109	0.00–9.836720
		1304.37	0.09	3.64 × 108	0.00–9.505292
Si ii*	⋯	1194.50	0.737	3.45 × 109	0.035613–10.415200
		1197.39	0.150	1.40 × 109	0.035613–10.390117
		1264.74	1.09	3.04 × 109	0.035613–9.838768
		1309.28	0.08	6.23 × 108	0.035613–9.505292
C ii	11.26	1334.53	0.129	2.42 × 108	0.00–9.290460
C ii*	⋯	1335.71	0.115	2.88 × 108	0.007863–9.290148
Si iii	16.34	1206.50	1.67	2.55 × 109	0.00–10.276357
Si iv	33.49	1393.76	0.513	8.80 × 108	0.00–8.895697
		1402.77	0.255	8.63 × 108	0.00–8.838528

Note. Atomic line data are given for the transitions considered in this paper. Values are taken from the NIST Atomic Spectra Database.

Download table as:  ASCIITypeset image

Weaker equivalent width of these LIS absorption lines has been associated with increased W_Lyα in stacked spectra of galaxies at z ∼ 3–4 (Shapley et al. 2003; Jones et al. 2012). Qualitatively, the Green Peas lend some support to this scenario: the two strongest absorbers 1244+0216 and 1137+3524 are among the lower ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ Green Peas, and the weakest absorbers (1133+6514 and 1219+1526) are among those with the highest ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. To quantify this trend, we measure equivalent widths of the LIS metal absorption lines by directly integrating the normalized spectra out to the velocity where the absorption meets the continuum. The uncertainties in equivalent width are determined by propagating the error vector over the same velocity range, including a systematic 10% uncertainty on the continuum normalization. For undetected lines, we approximate the upper limit by taking the equivalent width of the marginally detected lines. For the LIS lines this threshold is around 0.5 (0.4) Å in the observed (rest) frame. These measured equivalent widths are listed in Table 5.

In Figure 9, we compare W_Lyα to the equivalent widths of Si ii λ1260 and C iiλ 1334 for both the Green Peas and the stacked z ∼ 3 Lyman break galaxy (LBG) sample from Shapley et al. (2003). In this plot, the Green Peas are consistent with the z ∼ 3 LBGs, although they extend to higher W_Lyα and weaker W_LIS. Nevertheless, we do not detect a significant correlation; the Spearman correlation coefficient (ρ) and probability of the null hypothesis (P) are given in the upper left of each panel in Figure 9. However, since W_Lyα is only a rough proxy for Lyα escape, we recast this diagram using ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ in the right panels of Figure 9. Here, the significance of the correlation is improved: the Spearman coefficient (which, using ranks, is not sensitive to the non-detections in this case) implies that the correlation is robust at 91% confidence for Si ii and 97% confidence for C ii. Although these trends are only marginally significant, the consistency between C ii and Si ii supports a real correlation. Moreover, these trends work in the same direction as the z ∼ 3–4 measurements from stacked spectra. It appears that conditions that create weaker LIS absorption lines also favor greater ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. We will return to this trend and discuss its possible meaning in Section 8, where we offer a more comprehensive interpretation of the data.

Zoom In Zoom Out Reset image size

Despite heavy reliance on the LIS metal lines, they remain indirect probes of the neutral hydrogen that is scattering the Lyα emission. Fortunately, the COS spectra of the Green Peas allow a more direct look at the H i gas, with at least one Lyman series line (besides Lyα) observed in each of the 10 galaxies. Figure 10 highlights these features, showing either Lyβ or Lyγ for each Green Pea. Whenever more than one Lyman series line is observed (7/10 galaxies) we find that their absorption profiles are consistent. The equivalent widths of these lines (Table 6) are measured in the same manner as for the metal LIS lines, although we instead adopt a 20% uncertainty on the continuum placement because normalization is more challenging at these wavelengths. Additionally, the same stellar absorption model used for the metal lines is shown for Lyβ and Lyγ. In contrast to the metals, the stellar H i absorption is significant around zero velocity. Nevertheless, the stellar absorption does not explain the blueshifted and highly opaque absorption, indicating a significant contribution from outflowing interstellar gas. We conclude that these lines are still useful for probing the outflowing gas at moderate to high (blueshifted) velocities.

Zoom In Zoom Out Reset image size

Table 6.  Equivalent Widths of H i Absorption Lines

Object ID	Lyβ	Lyγ	Lyδ	Ly
	(Å)
0303–0759	−2.6 ± 0.6	...	...	...
1244+0216	−2.6 ± 0.7	...	−2.1 ± 0.6	...
1054+5238	−2.8 ± 0.7	...	−2.5 ± 0.6	1.9 ± 0.5
1137+3524	−2.5 ± 0.8	−2.5 ± 0.6	...	...
0911+1831	−1.8 ± 0.7	...	...	1.2 ± 0.5
0926+4427	−2.4 ± 0.5	...	...	...
1424+4217	...	−1.9 ± 0.5	...	...
1133+6514	−2.9 ± 0.7	...	−2.0 ± 0.7	...
1249+1234	−2.2 ± 0.7	−2.1 ± 0.6	...	2.2 ± 0.8
1219+1526	−2.0 ± 0.7	−2.6 ± 0.7	...	...

Note. Rest-frame equivalent widths of H i Lyman series absorption lines discussed in this paper. Lyman series equivalent widths contain a contribution of around 0.6–0.8 Å from stellar absorption.

Download table as:  ASCIITypeset image

The H i absorption lines show some remarkable differences from the LIS metal lines. First, the equivalent widths of the Lyman series lines show no variation with ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ or W_Lyα. Second, and in contrast to the spectra shown in Figures 6–8, the hydrogen absorption lines show little to no residual intensity at modest blueshifted velocities (a few hundred km s⁻¹). This finding indicates that nearly 100% of the stellar light is covered by H i absorbing gas at these velocities. Moreover, the high opacity and the similarity between Lyβ, Lyγ, Lyδ, and Ly equivalent widths (when more than one line is detected) indicate that these lines are saturated but not damped. Hence, the neutral hydrogen column density is poorly constrained: ${N}_{{\rm{H}}\;{\rm{I}}}\sim {10}^{16}-{10}^{18}$ cm⁻². Remarkably, even the low end of this range leads to high Lyα optical depth at line center, even if it is optically thin to hydrogen-ionizing Lyman continuum (LyC) photons (τ_Lyα > 10²–10³; Verhamme et al. 2006, 2015).

Finally, we show higher ionization Si iii λ1206.5 and Si iv λ1394, λ1403 in Figures 11 and 12. The equivalent widths of these lines are also listed in Table 5. Unlike the LIS metal lines, the equivalent widths of these higher ionization lines do not appear to change across the sample. This conclusion is most apparent for the Si iii lines, all of which have good S/N and W ∼ −1.3 Å. In this aspect, the Green Peas are also consistent with the z ∼ 3 LBGs: Shapley et al. (2003) report that, despite the relation between W_Lyα and W_LIS, ${W}_{\mathrm{Si}\;{\rm{IV}}}$ is the same in each of their four W_Lyα-defined stacks.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The kinematics of the CGM may play an important role in regulating Lyα escape from galaxies. When the photons scatter in outflowing gas, they can Doppler shift out of resonance with the bulk of the ISM and escape more easily. Hence, we may expect that a large velocity gradient in the CGM promotes strong Lyα emission. Early studies find evidence for this scenario. In a sample of eight local starburst galaxies, Kunth et al. (1998) find that half of their sample with damped Lyα absorption showed no evidence of outflowing gas in their LIS lines. Yet the other half of their sample that showed Lyα emission exhibited outflows with velocities around 200 km s⁻¹. Similarly, Wofford et al. (2013) report similar findings from COS observations of 20 nearby galaxies. In their sample, seven Lyα-emitting galaxies have v_LIS ∼ −100 km s⁻¹, whereas 10 galaxies with damped Lyα profiles have v_LIS ∼ −20 km s⁻¹. And most recently, Martin et al. (2015) show that Lyα escape from ULIRGs is enhanced when the wings of the Lyα and [O iii] lines reach greater blueshifted velocities.

In this section, we will test whether the Lyα escape is aided by outflowing gas in the Green Peas. To quantify the outflow kinematics, we will use metal and hydrogen absorption lines, and also explore the kinematic measures from the Lyα emission.

7.1. Gas Kinematics in Absorption

Figures 6–8 illustrate the kinematics of the LIS ions. When the lines are detected, they are blueshifted with wings extending several hundred km s⁻¹. To measure the gas kinematics, we estimate characteristic outflow velocities, v_c. Since many of the lines have non-Gaussian shapes, we avoid fitting centroid velocities, and instead calculate the equivalent width-weighted velocity:

$$\begin{eqnarray}&&{v}_{c}=\displaystyle \frac{\int v(1-{f}_{\mathrm{norm}})d\lambda }{\int (1-{f}_{\mathrm{norm}})d\lambda },\end{eqnarray} \tag{ 1 }$$

where f_norm is the normalized flux, v represents velocity at each wavelength, and the denominator is easily recognized as the equivalent width. The resultant characteristic velocities are listed in Table 7. The errors on these velocities are calculated using 1000 Monte Carlo realizations where the spectrum is perturbed according to its error vector and a 10% continuum normalization uncertainty. We also quantify the maximum outflow velocity by determining where the absorption trough reaches the continuum. These velocities are listed in Table 8; the uncertainties are calculated using a Monte Carlo simulation in the same manner as for the characteristic outflow velocities. For completeness, and to obtain some kinematic information for 1133+6514 and 1219+1526 which have no LIS lines, we also list the characteristic and maximum outflow velocities for the higher ionization Si iii and Si iv lines. And likewise, maximum outflow velocities for the Lyman series lines are listed in Table 9 (where we have again included a 20% continuum normalization uncertainty). However, because of the stellar component and the breadth of the H i absorption lines, we do not measure characteristic outflow velocities for these features.

Table 7.  Characteristic Outflow Velocities from ISM Absorption Lines

Object ID	v1190	v1193	v1260	v1334	v1206	v1393	v1403
0303–0759	...a	...a	−180 ± 60	−220 ± 60	−240 ± 50	−240 ± 60	−230 ± 80
1244+0216	−100 ± 20	−50 ± 20	−110 ± 20	−80 ± 20	−70 ± 20	−230 ± 70	−100 ± 50
1054+5238	−140 ± 100	−120 ± 80	−280 ± 70	...b	−200 ± 40	−180 ± 80	−160 ± 70
1137+3524	−140 ± 30	−130 ± 30	−190 ± 30	−150 ± 40	−170 ± 40	−140 ± 30	−200 ± 40
0911+1831	−200 ± 80	−300 ± 90	−360 ± 70	−130 ± 90	−400 ± 90	−290 ± 50	−140 ± 100
0926+4427	...a	...a	−330 ± 170	−230 ± 100	−320 ± 60	−320 ± 70	−270 ± 70
1424+4217	−260 ± 80	−210 ± 90	...c	...c	−280 ± 60	...c	...c
1133+6514	...a	...a	...a	...a	−270 ± 50d	...a	...a
1249+1234	... e	... e	−150 ± 60	−210 ± 60	−250 ± 60	−210 ± 50	−250 ± 80
1219+1526	...a	...a	...a	...a	−420 ± 90	−370 ± 80	...a

Notes. Equivalent width-weighted velocities (in km s⁻¹) are given for Si ii λλ1190.4, 1193.3, λ1260.4, C ii λ1334.5, Si iii λ1206.5, and Si iv λλ1393.8, 1402.7.

^aLine is marginally detected or completely undetected, so reliable kinematic information is unavailable. ^bLine is contaminated by Milky Way absorption. ^cLine is not covered by the present spectra. ^dThe blue wing of the line may extend more than −1200 km s⁻¹, but uncertainties in continuum normalization make this extent unclear (see Figure 11). We report the velocity from a Gaussian fit to the main absorption component of this line. ^eSi ii λλ1190.4, 1193.3 lines are blended.

Download table as:  ASCIITypeset image

Table 8.  Maximum Velocities of Metal Absorption Lines

Object ID	v1190max	v1193max	v1260	v1334max	v1206max	v1393max	v1403max
0303–0759	...a	...a	...b	−560 ± 50	−710 ± 100	−580 ± 120	−440 ± 30
1244+0216	−410 ± 60	−330 ± 70	−550 ± 90	−350 ± 60	−320 ± 80	−630 ± 90	−400 ± 70
1054+5238	−440 ± 60	−420 ± 100	−760 ± 150	...b	−690 ± 100	−650 ± 110	−490 ± 70
1137+3524	−440 ± 90	−400 ± 60	−590 ± 130	−450 ± 40	−510 ± 110	−500 ± 100	−630 ± 160
0911+1831	−410 ± 110	−600 ± 50	−800 ± 200	−600 ≲ v ≲ −300c	−930 ± 240	−700 ± 160	...
0926+4427	...a	...a	−580 ± 130	−580 ± 90	−880 ± 100	−930 ± 90	−680 ± 60
1424+4217	−630 ± 130	−300 ± 70	...d	...d	−560 ± 70	...d	...d
1133+6514	...a	...a	...a	...a	∼−600e	...a	...a
1249+1234	−500 ± 50	...f	−390 ± 70	−420 ± 90	−540 ± 50	−470 ± 120	−400 ± 110
1219+1526	...a	...a	...a	...a	−960 ± 160	−670 ± 220	...a

Notes. Maximum velocities (in km s⁻¹) are given for the same lines as in Table 7.

^aLine is marginally detected or completely undetected, so reliable kinematic information is unavailable. ^bLine is contaminated by Milky Way absorption. ^cThis absorption profile appears divided into two parts, possibly from a noise spike that crosses unity in the normalized spectrum around −300 km s⁻¹. The maximum velocity in the more blueshifted portion of the profile is around −600 km s⁻¹. ^dLine is not covered by the present spectra. ^eThe blue wing of the line may extend more than −1200 km s⁻¹, but uncertainties in continuum normalization make this extent unclear (see Figure 11). Most of the absorption, however, lies within −600 km s⁻¹. ^fThe continuum is not reached before the absorption from neighboring Si ii λ1190.4 is reached. Therefore the maximum outflow velocity from this line is unreliable.

Download table as:  ASCIITypeset image

Table 9.  Maximum Velocities of Lyman Series Absorption Lines

Object ID	${v}_{\mathrm{Ly}\beta }^{\mathrm{max}}$	${v}_{\mathrm{Ly}\gamma }^{\mathrm{max}}$	${v}_{\mathrm{Ly}\delta }^{\mathrm{max}}$	${v}_{\mathrm{Ly}\epsilon }^{\mathrm{max}}$
0303–0759	−800 ≲ v ≲ −200a	...	...	...
1244+0216	−660 ± 70	...	−530 ± 40	...
1054+5238	−800 ± 60	...	−670 ± 50	...
1137+3524	...	−750 ± 90	...	...
0911+1831	−760 ± 180	−1200 ± 270	...	−730 ± 130
0926+4427	∼−700 ± 100b	...	...	...
1424+4217	...	−700 ± 110	...	...
1133+6514	−720 ± 60	−800 ± 170	−820 ± 160	...
1249+1234	−620 ± 100	−690 ± 90	...	−760 ± 170
1219+1526	−660 ± 70	−740 ± 30	...	...

Notes. Maximum velocities (in km s⁻¹) are given for the Lyman series lines. Lines that are not listed are either contaminated by Milky Way absorption, or fall beyond the blue-wavelength cutoff of our COS spectrum or near the gap between the FUV segments. Unlike the metal lines, we do not fail to detect absorption from H i gas.

^aBlue wing of line impacted by Milly Way absorption. ^bThe normalized continuum does not reach unity before nearby Milky Way absorption sets in (see Figure 10). Visual inspection of the spectrum suggests an approximate maximum velocity of around −700 ± 100 km s⁻¹.

Download table as:  ASCIITypeset image

Figure 13 shows how ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ depends on these kinematic features for the H i, LIS metals, and higher ionization lines. The top two panels focus on v_c, with different symbols indicating different transitions. The upper left panel shows that we detect no significant trend between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and outflow velocity measured from the LIS metals. However, we must note that two of the three highest ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ galaxies cannot be included in this plot and correlation test, because they are undetected in all LIS absorption lines (1133+6514 and 1219+1526). Their addition in the upper right panel of Figure 13, where we show the higher ionization states of Si, hints at a correlation. For Si iii, which is detected at good S/N in all ten Green Peas, the Spearman correlation coefficient rules out a spurious relation at 98% confidence. But this trend is only significant with the addition of the two high ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ Green Peas. In the bottom panels, where we show the maximal velocities, the metals tell a similar story: the Green Pea 1219+1526 (the highest ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ object) has greater maximal outflow velocity than most when Si iii is considered. But otherwise, we detect no trend. Finally, in the bottom left panel of Figure 13 we add maximal outflow velocities reached by H i Lyman series lines (red points). These lines consistently reach −700 to −800 km s⁻¹, and show no trend with ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. In summary, since the only possible correlation detected in Figure 13 relates ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ with higher ionization lines that do not trace H i, we conclude that these data give no compelling evidence for a scenario where Lyα escape is enhanced by scattering in outflowing H i gas.

Zoom In Zoom Out Reset image size

Remarkably, Figure 13 shows that the maximal outflow velocities measured by the H i Lyman series lines are systematically bluer than the maximal velocities of the LIS metal lines. This result demonstrates that the H i absorption is markedly more sensitive to low-density, high-velocity gas that cannot be detected in the LIS metal absorption lines. Yet this H i likely plays an important role in scattering Lyα photons; we will explore this topic more in Section 7.2 where we compare the Lyman series absorption and Lyα emission kinematics.

7.2. The Velocity Structure of Lyα Emission

The Lyα emission line profiles give another probe of the outflowing gas kinematics, as the broad emission is (at least in part) generated by resonant scattering in the gas around the galaxy. Table 3 gives kinematic measures of the Lyα lines, including the velocities of the red and blue peaks, and the maximal velocities of the red and blue wings of the lines. Figure 14 shows how ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ changes with these Lyα kinematic measures. Statistically significant correlations appear in some, but not all, of these quantities. Each panel in Figure 14 lists the Spearman correlation coefficient, ρ, and the probability, P, that the correlation arises by chance. For this analysis, we exclude the Green Pea with single-peaked Lyα emission (1249+1234) from the statistics and plots involving the blue peak and the peak separation, but it is included in the other diagnostics. Similarly, the maximal red velocity for 1424+4217 is a lower limit (the spectrum was truncated due to a failed observation), and this galaxy is excluded from the relevant statistics.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 14 shows that the separation between the Lyα emission peaks becomes smaller when ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ is larger (top right panel), and that this trend is driven primarily by a shift in the blue emission peak (top left panel). At the same time, the galaxies with higher ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ may show somewhat lower velocities for their red Lyα peaks, but the correlation is not significant. Compared to the blue emission peaks, the red Lyα emission peaks also inhabit a smaller range of velocities. These findings are consistent with studies of high-redshift galaxies (z ∼ 2–3), where the kinematics of the red Lyα emission peak has been studied by multiple groups. Notably, Hashimoto et al. (2013) and Shibuya et al. (2014) measure a red peak velocity, ${\rm{\Delta }}{v}_{\mathrm{Ly}\alpha }\sim 200$ km s⁻¹, for 10–20 LAEs with W_Lyα > 50 Å, compared with 450 km s⁻¹ measured for UV-continuum-selected LBGs (Steidel et al. 2010). In this sense, the Green Peas are more similar to the high-redshift LAEs than to the LBGs. Indeed, for a larger sample of 158 galaxies, Erb et al. (2014) report a significant anticorrelation between W_Lyα and ${\rm{\Delta }}{v}_{\mathrm{Ly}\alpha }$, such that Lyα is stronger when it emerges closer to the systemic velocity. Our trend of increasing ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ with peak separation fits naturally with this scenario; increased emission near the systemic velocity serves to shift the emission peaks closer together. A visual inspection of Figure 2 supports this idea: where the Lyα emission is strong, there is net emission around v ∼ 0, whereas net absorption appears around the systemic velocity when ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ is low. Here, for the first time, we show that the kinematics of the blue Lyα emission peak shows even greater variation than the red peak.

This tight correlation between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and peak separation suggests that Lyα escape is determined by the neutral hydrogen column density. In short, when the column density is low, the Lyα photons can escape nearer to the systemic velocity, with less scattering in the expanding circumgalactic envelope. Variations in dust, on the other hand, preserve the shape of the profiles while decreasing the strength of the emission (Verhamme et al. 2006, 2008, 2015; Behrens et al. 2014). Indeed, the H i density has already been used to explain the trends with the red Lyα peak velocity, ${\rm{\Delta }}{v}_{\mathrm{Ly}\alpha }$, seen in z ∼ 2–3 galaxies (discussed above; Chonis et al. 2013; Hashimoto et al. 2013; Erb et al. 2014; Shibuya et al. 2014). Moreover, recent radiative transfer models from Verhamme et al. (2015) show that small separations between peaks, ${\rm{\Delta }}v\lt 300$ km s⁻¹, imply ${N}_{{\rm{H}}\;{\rm{I}}}\lesssim {10}^{18}$ cm⁻², whereas higher separations, $300\lt {\rm{\Delta }}v\lt 600$ km s⁻¹, arise in galaxies with 10¹⁸ ≲ N_{H i} ≲ 10²⁰ cm⁻².

However, the Lyα radiative transfer models models of Verhamme et al. (2015) do not produce all the features in our data. First, unlike the Green Peas, increasing N_{H i} in the models shifts the red peak to higher velocities while holding the blue peak at a nearly fixed velocity. Second, in modeled profiles, the high-velocity wings of the Lyα reach only to ±a few hundred km s⁻¹ in contrast to the several hundred km s⁻¹ observed in the Green Peas. This discrepancy may be attributable to an assumed intrinsic Lyα profile that neglects the broad wings observed in the optical emission lines. Indeed, Martin et al. (2015) successfully produce broad Lyα wings on eight ULIRGs by modeling the Lyα emission as a superposition of intrinsic (broad + narrow) and scattered components. And the third difference between the data and models of Verhamme et al. is that the double-peaked emission profiles appear mostly when the shell expansion velocity is low, v < 100 km s⁻¹. Yet the Green Peas show −300 ≲ v_c ≲ −100 km s⁻¹ in their LIS lines. Nevertheless, Lyα radiative transfer modeling offers some intriguing lines of investigation. As Verhamme et al. (2015) already noted, the galaxies with small peak separations may have low enough column densities to be optically thin to the hydrogen-ionizing LyC photons. The Green Peas with the most closely spaced peaks, 1219+1526 and 1133+6514, make excellent candidates for follow-up observations aimed at direct detection of the LyC. Remarkably, among the present sample of Green Peas, 1133+6514 shows lower W_Hα and W_Lyα despite its high ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. These conditions are also consistent with LyC leakage.

In addition to the peaks of the Lyα emission, we explore the kinematics probed in the wings of the lines. The bottom panels of Figure 14 show how the maximal velocity reached in the Lyα lines relates to ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. In these diagnostics, the typical 200 km s⁻¹ uncertainty on the maximal velocities makes it difficult to draw robust conclusions. Although only the center panel (illustrating the maximal velocity in the red wing of the line) shows a significant relation, these plots hint at a scenario where broader Lyα emission wings may be associated with increased Lyα escape. This finding is qualitatively consistent with the sample of eight ULIRGs studied by Martin et al. (2015), where increased Lyα escape was associated with greater blueshifted Lyα and [O iii] velocities. In this case, the close association of the blueshifted Lyα and [O iii] λ5007 kinematics suggested an in situ production of Lyα photons in the high-velocity gas (rather than scattering). As demonstrated by Martin et al. (2015), this emission would result from cooling of the hot galactic wind. However, with the resolution of the SDSS spectra, it remains difficult to test whether the Lyα kinematics in the wings corresponds closely with the [O iii] and Hα kinematics. Ultimately, more work is needed to obtain higher spectral resolution observations of the nebular gas, so that we may compare the line profiles in greater detail. We defer this analysis to a future study.

In order to more fully understand what the Lyα kinematics are measuring, we also compare the spectra to H i absorption profiles measured by the Lyβ or Lyγ lines in Figure 15. This comparison shows that the maximal velocities of the Lyα and H i absorption are well matched (except for possibly 1244+0216), extending to around −700 km s⁻¹ in most cases. The presence of H i absorption over the same range of velocities as the blueshifted Lyα emission is noteworthy; it demonstrates clearly that the gas which scatters Lyα photons exists in an envelope spanning a range of velocities. Shell models, where the cool gas exists at a single velocity, cannot describe absorption spanning several hundred km s⁻¹, even though they often provide satisfactory fits to Lyα emission profiles (Verhamme et al. 2008). Previous studies (e.g., Pettini et al. 2002; Kulas et al. 2012) have already noted this shortcoming of the shell model, drawing a similar conclusion from the velocity gradient seen in metal lines. Here, for the first time, we are able to demonstrate the need for an envelope of H i gas, without the need for metals as a proxy. This added constraint, it turns out, is important since the LIS and Lyman series lines show somewhat different kinematics in their maximal outflow velocities. As we showed in Figure 13, the LIS metal lines reach maximal velocities around −400 to −600 km s⁻¹, whereas the Lyman series lines extend to between −600 and −800 km s⁻¹. We interpret this difference as an indication that the LIS metal absorption lines are insensitive to the gas that constitutes the wings of the Lyman series lines. Nevertheless, the excellent correspondence between the blueshifted Lyα emission and H i absorption velocities suggests that this high-velocity, low-density gas is still important for scattering Lyα photons and creating the profiles that we observe.

Stacking analyses of UV-luminous galaxies at z ∼ 2–4 have also explored the origin of the blue peak emission by comparing it to LIS lines (Steidel et al. 2010; Jones et al. 2012). These studies have found that the velocity of the blue peak most closely corresponds with the velocities in the blue wing of the LIS absorption lines, around −500 to −600 km s⁻¹. In other words, the blue Lyα emission is strongest not where the apparent optical depth of the LIS lines is greatest, but instead where it is decreasing. The Green Peas show that these characteristics of stacked spectra are not uniform among galaxies. In Figure 16, we summarize the kinematic measurements from H i absorption, LIS metal absorption (Si ii and C ii), and blueshifted Lyα emission. We show the velocity range probed by each set of features, by computing the error-weighted mean and its uncertainty. The Green Peas are sorted by ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$, with a comparison from stacked spectra of LBGs at z ∼ 2–4 shown at bottom (from Steidel et al. 2010 and Jones et al. 2012). While the LIS absorption velocities are similar between the high-redshift samples and Green Peas, the latter show blueshifted Lyα emission emerging closer to the systemic velocity and v_c. In fact, while the lower ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ Green Peas show the blue peak emission falling between the LIS v_c and v_max, this trend may break down as ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ increases. Both 0926+4427 and 1424+4217 show that the blueshifted Lyα peak probably emerges at lower blueshifted velocities than v_c. Moreover, if the LIS metals that are too weak to detect in 1133+6514 and 1219+1526 still follow the kinematics of the more highly ionized gas, we might expect this trend to be stronger. These galaxies show v_c (Si iii) of −270 and −420 km s⁻¹ respectively, compared to their Lyα v_peak^blue around −100 km s⁻¹.

Zoom In Zoom Out Reset image size

In summary, we conclude that while the wings of the Lyα emission lines emerge over the same range of velocities probed by H i absorption, the peak emission shows little relation to any of the outflow kinematic measures. This finding is consistent with the conclusions from previous high-redshift studies, where it is argued that the emergent Lyα emission is more strongly affected by the optical depth of gas near the systemic velocity than by outflow kinematics (Steidel et al. 2010; Law et al. 2012; Chonis et al. 2013; Erb et al. 2014).

In our analysis so far, we have explored the role of ISM and CGM gas in regulating the amount of Lyα emission that we observe from the Green Peas. In this section we will discuss the physical implications of our measurements. First, however, we summarize the findings from the previous sections.

• 1.  
  The "Green Pea" classification identifies objects with prominent Lyα emission and little to no Lyα absorption. 9 of the 10 galaxies in our sample have Lyα profiles that show both blue and red peaks, and unlike other low-redshift samples, their Lyα luminosity and W_Lyα reach the range probed by high-redshift Lyα surveys (e.g., Henry et al. 2010, 2012; Ouchi et al. 2010).
• 2.  
  Comparison to other nearby populations with published Lyα measurements (Scarlata et al. 2009; Cowie et al. 2011; Hayes et al. 2014; Östlin et al. 2014; Martin et al. 2015) showed that the Green Peas have (on average) lower masses, higher SFRs and specific SFRs, brighter M_UV, bluer UV slopes, higher W_Hα, lower dust extinction, and smaller sizes.
• 3.  
  Despite similarly low dust extinction in the Green Pea sample, we measure Lyα escape fractions (within the COS aperture) that span a factor of 10: f_esc ^Lyα = 0.05–0.62.
• 4.  
  We detect LIS absorption in 8/10 Green Peas, and confirm previous findings that weaker W_LIS is associated with stronger Lyα emission (e.g., Shapley et al. 2003) albeit, extending prior measurements to higher W_Lyα and weaker W_LIS. Although the correlation between W_LIS and W_Lyα is not significant in our data, a tentative correlation is found when we use ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ directly, instead of W_Lyα as a proxy. Furthermore, the robustness of this relation is also supported by detection in two independent measures: Si ii 1260 Å and C ii 1334 Å. At the same time, we show that the equivalent width of Si iii λ1206 does not vary among the present sample.
• 5.  
  2 of the 10 Green Peas show no LIS metal detections, and these are found among those with the highest Lyα escape fraction, f_esc ^Lyα ≥ 0.40, within the COS aperture.
• 6.  
  Absorption in the H i Lyman series (β, γ, δ, and/or ) is detected in all ten Green Peas. Unlike the LIS metal lines, the H i absorption appears uniform in kinematics and equivalent width across the sample. Furthermore, and in contrast to the LIS metal lines, the Lyman series lines are unambiguously saturated, with little to no residual intensity at line center.
• 7.  
  Kinematic measurements from Si ii and C ii show no correlation between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and outflow velocities (v_c or v_max) in the LIS metal lines, although two of the three highest ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ Green Peas cannot be included due to the absence of detectable LIS absorption. The data do show a marginal correlation between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and the outflow velocities (v_c and v_max) for Si iii, but because this line does not trace neutral hydrogen, this trend is not compelling evidence that scattering in the cool, neutral phase of the outflow is aiding Lyα escape.
• 8.  
  The Lyα escape fraction, ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$, shows a significant anticorrelation with the velocity separation between the blue and red peaks, which is driven by shifts in the blue peak velocity. This trend is consistent with a changing neutral hydrogen column density, where lower H i density allows more Lyα to emerge near the systemic velocity.
• 9.  
  The velocity of the blue Lyα emission spans the same range of velocities as the H i absorption lines, demonstrating (for the first time with H i) that the Lyα emission arises from an envelope of gas spanning several hundred km s⁻¹. At the same time, the maximal outflow velocities reached by the Lyman series lines are consistently bluer than the maximum outflow velocities of the LIS metal lines; the good agreement between the velocities probed by blue Lyα emission and Lyβ/Lyγ absorption implies that, compared to LIS metals, these lines are a better proxy for the gas that scatters Lyα photons.

These observations allow us to take a closer look at how the interstellar and circumgalactic gas regulates the escape of Lyα photons from the central few kpc of the Green Peas (as probed by the COS aperture). In past studies, a few different Lyα escape mechanisms have been considered as possibly important. In the remainder of this section we take a comprehensive look at the present Green Pea data, and explore which Lyα escape models are consistent with our results.

Galaxy Outflow Kinematics—Galactic outflows are one mechanism for increasing the transmission of Lyα photons. In this case, the Lyα photons can scatter in H i gas, which is Doppler shifted with respect to the ISM. The scattered photons are then out of resonance with the ISM and escape more easily. Previous studies have noted this effect: Kunth et al. (1998), Wofford et al. (2013), and Rivera-Thorsen et al. (2015) show that Lyα-emitting galaxies have outflow velocity v ∼ −100 to −200 km s⁻¹, while Lyα-absorbing galaxies show no outflows (LIS absorption is near v ∼ 0). On one hand, the Green Peas are consistent with these previous measurements: all LIS lines are measured with v_c ∼ −100 to −200 km s⁻¹, and all the Green Peas show Lyα emission. On the other hand, despite similar outflow velocities, the Green Peas show a factor of eight difference in the Lyα escape fraction ($0.05\leqslant {f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }\leqslant 0.41$). Taken together with the previous studies, these data suggest that outflows may be necessary to permit Lyα escape, but they are insufficient to explain the wide range of Lyα emission strengths. Ultimately, more work is needed to clarify the role of outflows, and confirm that the relation between outflow velocity and Lyα escape is not secondary to some other physical characteristic like the ionization state or geometry of outflowing gas.

Neutral Hydrogen Covering—Studies using stacked, composite spectra of high-redshift galaxies (3 ≲ z ≲ 4) have suggested that Lyα escape is regulated by the covering fraction in neutral hydrogen, f_c (Shapley et al. 2003; Jones et al. 2012, 2013). The motivation for this model is the correlation between W_LIS and W_Lyα. Since the LIS metal lines appear saturated in composite spectra, but the absorption is not black, the interpretation is that this absorbing gas does not cover the entire galaxy. Weaker W_LIS suggests lower f_c in H i gas, resulting in greater ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$.

While our data confirm the correlation between W_LIS and W_Lyα (see Section 6), we cannot attribute this relation to non-uniform covering in the H i. By directly probing the H i Lyman absorption series, rather than using metals as a proxy, we showed that the lines are saturated and opaque (or nearly so) in all ten Green Peas (see Figure 10). Hence, we conclude that low covering of H i cannot explain the variations in ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ seen in our sample.

Besides affecting our interpretation of Lyα escape, the addition of the Lyman series lines requires that we re-evaluate our picture of how the CGM is structured around high-redshift galaxies. Previous analyses of UV spectra support a model where saturated LIS metal absorption arises from cool, dense clouds of gas entrained in a more highly ionized outflow (Shapley et al. 2003; Heckman et al. 2011). For the Green Peas, the LIS metal lines are difficult to interpret; because of their low S/N, and the probable contribution from emission filling (Prochaska et al. 2011; Scarlata & Panagia 2015), we cannot determine whether the lines are optically thin or optically thick. On one hand, if the LIS metals from the Green Peas are optically thick, then the data require a density-dependent covering fraction, f_c. This scenario requires a pervasive low-density, low-ionization component surrounding higher density clumps; only the latter give rise to detectable metal absorption. On the other hand, the data could also be described by a homogeneous low-density envelope of gas, which gives rise to optically thin LIS metal absorption. The Si ii lines that we observe become optically thick at line center for N_{Si ii} ∼ 0.9–3.8 × 10¹³ cm⁻²; assuming solar abundance [Si/H] = −4.5 (Asplund et al. 2009), Z = 0.1 Z_⊙, and 100% of the Si exists in Si ii, the implied H i columns would be around 10¹⁸–10¹⁹ cm^–2. These are consistent with our observations of the Lyman absorption lines, which we inferred were on the flat part of the curve of growth.

A difference between the spatial distribution of detected metals and H i has been noted before. Quider et al. (2010) inferred the existence of a pervasive H i component in the lensed "Cosmic Eye" at z ∼ 3. Similar to the Green Peas, the spectrum of this galaxy showed partial covering in metal lines, yet complete covering in damped Lyα absorption. The present COS data demonstrate that the "Cosmic Eye" is not alone in this characteristic. These observations suggest that a widespread, low-density H i component may be common.

Spatially Extended Lyα Emission—In Section 4 we noted that extended Lyα emission, falling outside the COS aperture, may be significant for this sample. Three galaxies have serendipitous Lyα measurements that should approximate a "total" flux. The LARS imaging showed that the COS aperture captured 40% of the flux of 0926+4427 (Hayes et al. 2014), and GALEX grism spectra show that we detected 60% and 75% of the flux for 1133+6514 and 1219+1526. Intriguingly, for this small subsample, the fraction of Lyα flux detected by COS increases with ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. However, an aperture correction alone still does not account for all of the Lyα photons produced in H ii regions; the small but significant amount of dust measured in Section 2 is probably still important, especially when coupled with resonant scattering. Systematic uncertainties on the absolute dust correction for Lyα preclude a full and accurate accounting of the flux.

The emission and absorption line profiles in our UV spectra give another means for testing the significance of aperture effects. First, in a qualitative sense, aperture losses can mimic some of the features observed in our Lyα spectral profiles. For a spherically expanding shell, the outermost regions are expanding in the plane of the sky, with projected velocity v ∼ 0 (see Figure 2 in Scarlata & Panagia 2015). Consequently, including a greater contribution from these regions would increase the contribution from Lyα emission near the systemic velocity, while also increasing ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$. At the same time, a greater contribution from spatially extended, scattered emission would also increase the amount emission from metals. This emission would both "fill in" the resonant Si and C absorption lines in our data and produce fluorescent Si ii* and C ii* emission. In the absence of an aperture, and for a spherical geometry, we expect the equivalent widths of emission and absorption to sum to zero: ${W}_{\mathrm{abs}}+{W}_{\mathrm{em}}=0$ (Prochaska et al. 2011). But if the finite COS aperture misses some of the scattering CGM gas, W_em will be decreased (Scarlata & Panagia 2015). We tested whether C ii and Si ii transitions in the present sample were consistent with ${W}_{\mathrm{abs}}+{W}_{\mathrm{em}}=0$, but found that this measurement is extremely sensitive to the precise placement of the continuum. Because of the 10% normalization uncertainty appropriate for our data, we cannot determine whether scattered Si ii and C ii emission is missing from the COS aperture.

Ultimately, direct detection of extended Lyα emission, similar to the approach adopted for the LARS galaxies (Hayes et al. 2013, 2014; Östlin et al. 2014), is needed. Such data would clarify how much total Lyα emission emerges from the Green Peas, while determining its spatial extent and distribution, and whether any regions show Lyα absorption.

Neutral Hydrogen Density—Another galaxy property that should impact Lyα output is the neutral hydrogen column density. With a greater density of H i, the Lyα photons will undergo increased scattering. In this case, they are more susceptible to absorption by dust, and must diffuse further in frequency from the line core before escaping.

Variations in the H i density offer the most promising explanation for the range of ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ that we detect in the Green Peas. In Section 7.2 (see Figure 14), we showed that the ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ correlates tightly with the velocity separation between the blue and red Lyα emission peaks. This correlation is driven mostly by a shift in the blue emission peak, which appears closer to the systemic velocity when ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ is high. Since we see no correlation between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and the outflow kinematics discussed in Section 7.1, the kinematic trends seen in the Lyα lines more probably arise from a sequence in H i density. In fact, Verhamme et al. (2015) have already attributed small peak separations to low H i column densities. In this model, the peaks are closer together when low densities allow the Lyα photons to escape close to the systemic velocity. Hence, in the present sample, we attribute the increase in ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ to a decrease in the neutral hydrogen density. It is plausible that decreased dust contents correspond with lower H i column density, but more sensitive dust measurements are needed to test this effect. Finally, we note that variations in H i density may explain the correlation between ${f}_{\mathrm{esc}}^{\mathrm{Ly}\alpha }$ and W_LIS that we showed in Section 6; if the metal absorption arises from optically thin gas in a homogeneous envelope, the lower columns of neutral H i would also imply weaker LIS absorption.

In this paper, we have presented a COS/FUV spectroscopic study of ten Green Pea galaxies at z ∼ 0.2. We have focused on understanding how Lyα photons escape from the ISM and CGM. The Lyα emission line is frequently used in high-redshift studies, not only to discover objects, but also to infer crude constraints on the properties of galaxies and the IGM. Yet our limited understanding of how Lyα photons escape from galaxies hampers our ability to fully interpret high-redshift observations (Dijkstra et al. 2014; Dressler et al. 2015). The detection of Lyα emisison in all 10 of the Green Peas suggests that in the lowest mass, least dusty, highest SFR galaxies this feature may appear in the majority of cases.

Beyond simple detection of Lyα, the UV absorption line data provide important constraints on the ISM and CGM gas that regulate Lyα output. These data show that while cool outflows traced by LIS metal absorption may play a role in permitting Lyα escape, they do not explain the widely varying Lyα strengths observed in the Green Peas. Moreover, aided by the low redshifts of these galaxies and the absence of Lyα forest absorption, for the first time we are able to directly analyze H i absorption in the Lyman series. These transitions show clear differences from the LIS metal lines that are usually used as a proxy for H i; most importantly, the H i covering fraction, f_c, is uniformly high for gas with N_{H i} > 10¹⁶ cm⁻². The Lyα photons do not escape the Green Peas through holes that are completely devoid of CGM/ISM gas. Instead, the kinematic variations in the Lyα emission profiles are strongly suggestive that H i column density is the primary characteristic regulating Lyα escape in these galaxies. The ubiquity of double-peaked profiles and the small velocity separation between the peaks strongly suggest that a low column density of H i gas is a typical characteristic among Green Pea galaxies.

This study demonstrates that nearby, high-redshift analog galaxies are useful local laboratories, where a wealth of high-fidelity data can provide challenges for frequently adopted models. By observing the "Green Pea" galaxies, we have provided a fresh look at objects which, to our knowledge, are excellent analogs for young galaxies in an early universe. Yet much work remains before we can be certain that the conclusions drawn here are applicable at moderate to high redshifts. Certainly, spectroscopy with the James Webb Space Telescope will clarify the overlap between high-z galaxies and their nearby counterparts. Ultimately, this study serves as a strong motivation for future investigations. An even greater understanding of these galaxies could be achieved with maps of the spatially extended Lyα emission, as well as sensitive measures of the small amount of dust present in these systems.

We acknowledge Marc Rafelski, Sanchayeeta Borthakur, Tucker Jones, Amber Straughn, and Jonathan Gardner for helpful discussions. A.H. is supported by HST GO 12928 and an appointment to the NASA Postdoctoral Program at the Goddard Space Flight Center, administered by Oak Ridge Associated Universities through a contract with NASA. C.L.M. acknowledges partial support from NSF AST-1109288. D.K.E. acknowledges support from NSF CAREER grant AST-1255591. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. A.H. and C.S. also acknowledge travel support and gracious hosting from the Nordic Institute for Theoretical Physics during their program, Lyα as an Astrophysical Tool.

The full spectra for all 10 Green Peas are shown in Figures 17–26.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size