THE LYMAN ALPHA REFERENCE SAMPLE. V. THE IMPACT OF NEUTRAL ISM KINEMATICS AND GEOMETRY ON Lyα ESCAPE^*

A full description and motivation of the sample is found in Östlin et al. (2014). Here, we provide a short summary.

Since the basic idea of the LARS is to investigate the fate of Lyα photons, which are mainly produced in the hot gas surrounding star-forming regions, the sample should consist of galaxies with strong star formation. To achieve this, a selection was made of UV-bright galaxies from the SDSS and GALEX catalogs. The galaxies were chosen to span UV-luminosities similar to those of Lyα emitters and Lyman Break Galaxies (LBGs) at redshifts ∼3. To ensure that the sample consists of actively star-forming galaxies, LARS galaxies were selected to have a Hα equivalent width of W(Hα) > 100 Å, a domain in which the galaxy luminosity is dominated by the young population of OB stars, and a high production of Lyα. Galaxies with FWHM(Hα) > 300 km s⁻¹, indicating an active galactic nucleus component, were rejected. The galaxies have redshifts within a range of 0.02 < z < 0.2; the lower side keeps Lyα well clear of the geocoronal Lyα line, while the higher side keeps Hα within the imaging bandpass of the Advanced Camera for Surveys (ACS) camera on the HST.

LARS overlaps with other COS campaigns. LARS 1, 6, 7, and 10 are observed for and included in the sample of Wofford et al. (2013) as KISSR 1578, 2019, 242, and 178, respectively. LARS 12, 13, and 14 are observed by Heckman et al. (2011). The COS data of LARS 7 have also been published in France et al. (2010) as KISSR 242.

LARS galaxies have been observed in several wavelength ranges and with various techniques: in UV continuum-, Lyα- and Hα imaging with the HST (Hayes et al. 2013, 2014; Östlin et al. 2014), in 21 cm H i emission with GBT single-dish and Karl I. Jansky VLA radio interferometric observations (Pardy et al. 2014), and Calar Alto PMAS IFU observations. An overview of key numbers describing the sample can be found in Table 1, including high precision redshifts derived from SDSS DR7 archival data.

As described in Section 2, the LARS sample overlaps with two other samples for which COS spectroscopy has been obtained in the appropriate settings. Here we rely upon the archival data, and GO program ID numbers are listed in column 2 of Table 2. Specifically, LARS targets 1, 6, 7, and 10 were observed during COS guaranteed time observations (ID: 11522, 12027, PI: Green, see Wofford et al. 2013), and LARS 12, 13, and 14 were observed under GO 11727 (Heckman et al. 2011).

Both sets of archival observations were obtained using the COS acquisition modes ACS/IMG or ACS/SEARCH, which successfully acquired and centered upon the brightest NUV point that fell within the Primary Science Aperture (PSA). For the FUV spectroscopy, somewhat different strategies have been adopted by the individual observers. GO 11522,12027 used one orbit per target, obtaining spectra in the G130M grating. These observers used two different central wavelength settings (CENWAVE = 1291 and 1318 throughout; column 3 of Table 2) to perform a spectroscopic dither that spanned the gap between the two segments of the detector. GO 12727 observations used two or four orbits per target and, because of the larger redshift of these galaxies, obtained spectra in both the G130M and G160M gratings. In each grating, two exposures were obtained using different FP-POS positions of the grating to perform a small spectral dither.

Our own observations (GO 12583; LARS 2, 3, 4, 5, 8, 9, 11) also used one orbit per target. We first used the ACS/SBC/F140LP and WFC3/UVIS/F336W, obtained as part of the imaging campaign (Östlin et al. 2014), to identify the regions of peak UV surface brightness. By knowing precisely the target coordinates for the COS observations, we acquired the targets with ACQ/IMG with the NUV imager. Again, targets were perfectly centered to within a couple of NUV pixels. For five of the seven targets we noted that multiple UV sources would fall within the PSA, and may degrade spectral resolution, so in these cases we imposed orient constraints (±20°) in order to align multiple sources with the cross-dispersion axis. We performed spectral observations with the G130M grating; since the targets vary in redshift between z = 0.028 and 0.084, we chose the CENWAVE settings (again see Table 2) to include Lyα and the largest possible number of low-ionization absorption lines, tuning the spectral gap between the detector segments to regions without features of interest. For each target we performed spectral dithers using three FP-POS settings, except for LARS 3, which was observed in the continuous viewing zone and used all four FP-POS positions.

For illustration, we show the position of the PSA (radius 1 25) on each target in Figure 1. In each case this found the brightest star cluster in the FUV. Images are those of Hayes et al. (2014), encoding continuum-subtracted Hα in red, FUV continuum in green, and continuum-subtracted Lyα in blue. The PSA is shown as a circle in each image. The lower right inset of each panel shows the corresponding COS/NUV acquisition frame, masked to match the aperture.

Zoom In Zoom Out Reset image size

Data reduction was performed with CALCOS version 2.15.6 (2011 November 03) pipeline tools. Because our absorption features do not fall close to geocoronal emission lines, we make no attempt to correct for O i lines near 1302 Å, as done in James et al. (2014), instead we collect continuum photons obtained at all Earth-limb angles. Finally data were resampled onto a uniform wavelength grid using the Δλ corresponding to the coarsest value, and stacked rejecting any pixel for which the data quality flag was >0. We check for systematic offsets in the wavelength calibration by measuring the centroids of the geocoronal emission lines, using the wings of Lyα and the O i λ 1302 line. For this, we produced additional combinations of the extracted spectra, which permitted the inclusion of detector pixels with low pulse height amplitude and gain sag, so as to fully include the geocoronal emission. In no case did we measure the wavelength of geocoronal emission features to be offset from the systemic values by more than the resolution; even accounting for switching between mirrors B and A in target acquisition (LARS 5 and 9 only) we observe no reason to doubt the wavelength calibration within about 20 km s⁻¹.

Figures 2 and 3 show the full spectra of all 14 galaxies in their rest frame. The transitions analyzed in this study are marked in blue; geocoronal and MW features are shown in orange. The spectra cover varying wavelength ranges; some spectra have been cut off where they did not contain any information included in this analysis. Galaxies with a global Lyα escape fraction >10% in Hayes et al. (2014) are marked with a ∗, while global Lyα absorbers are marked with a †.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We base the systemic redshift determination on nebular emission lines in the hot regions surrounding the starbursts. We assume that these lines originate from the same recombination processes as Lyα, and therefore can be expected to trace the intrinsic Lyα well. While no strong nebular emission lines other than Lyα are present in our COS spectra, such lines can be found in the SDSS archival spectra. Systemic redshifts were determined as described in Östlin et al. (2014) by the aid of the lines Hα, β, γ, and δ; [O i] λ6300, [O iii] λλ4959,5007, [N ii] λλ6548, 6583, and [S ii] λλ6716, 6731.

The COS pointings of this sample are very close to those of SDSS; we do not report the numbers here but for the subsample included in Wofford et al. (2013), the offsets have been measured to 0.10, 0.08, 0.38, and 062 for LARS 1, 6, 7, and 10, respectively. The aperture sizes are similar as well, 25 for COS versus 30 for SDSS, so the systemic redshift derived from SDSS spectroscopy is believed to describe the same regions from which the Lyα photons seen in the COS spectra originate. These redshifts were found by fitting each of the above mentioned lines in the SDSS data to a single Gaussian, inferring a redshift from each, and then taking a signal-to-noise ratio (S/N)-weighted average of these redshifts. SDSS-derived redshifts are listed in Table 1; other properties of the LARS galaxies measured and derived from SDSS spectroscopy can be found in Östlin et al. (2014). The computed redshifts all agree with those listed in the SDSS database to within ∼10 km s⁻¹.

In the case of LARS 1, 6, 7, and 10, (KISSR 1578, 2019, 242, and 178, respectively), the systemic redshifts were found to agree with those reached by Wofford et al. (2013) to within the precision reported there.

The COS has a resolution of R = 20,000 for a perfect point source, which corresponds to six detector pixels in the dispersion direction. We rebinned the data by a factor of six to reflect this, improving S/N without losing information.

The targets of this study are not point sources, so the effective resolution is somewhat lower than the instrument resolution; the amount depends on the extent and morphology of each target. We estimated the effective resolution for each target based on the reduced and cleaned UV continuum frames prepared by Hayes et al. (2014) in the following way:

We created a circular mask on the FUV frame corresponding to the COS aperture, rotated the part of the frame inside the mask to align with the COS acquisition image. Vignetting was taken into account by multiplying the flux in each aperture pixel with an interpolation of the aperture throughput grid in Figure 7 of Goudfrooij et al. (2010), and the flux was then summed along the cross-dispersion direction. The half-light width of the resulting profile was found by linearly interpolating on a grid 10 times finer than the actual pixel scale. The resulting width in pixel units was converted to arcseconds, which again can be translated to COS resolution elements by the conversion factor 0.171 arcsec/resel that was reported in the COS Instrument Handbook (Holland et al. 2014). The resulting widths are listed in column 6 of Table 2.

Key properties of the absorption lines selected for analysis in this work are tabularized in Table 3. Four Si ii absorption lines, one O i and one C ii feature are together assumed to trace the neutral medium. The Si ii lines all arise from the ground state in Si ii, allowing us to use their difference in line strength to evaluate the Si ii column density and neutral gas covering factor.

Three Si iv lines are presented and briefly analyzed; they are believed to trace the interface between the hot, ionized medium in the regions around the starburst, and the surrounding neutral medium (Grimes et al. 2009; Strickland & Heckman 2009). These lines may also contain a stellar component.

Finally, fluorescent emission lines connected to the fine structure splitting in the ground state of Si ii are included in the table because they hold some significance to the discussion of our results. These lines arise when electrons in the excited state of a transition with significant fine structure splitting of the ground state decay to the upper of these levels, which then quickly decays into the lower, fine structure level by the emission of a long-wavelength photon. Since the upper level of the ground state is generally not populated, they are seen in emission only as a peak somewhat redward of the main, resonant absorption feature—in case of the silicon lines treated here, the separation is ∼4 Å. These lines are marked with a * in Table 3.

After reduction and resampling, the spectra were continuum-normalized around the absorption features of interest. This normalizing was performed by hand, because the continuum is generally not sufficiently well-behaved to be modeled. For each absorption feature, the surrounding wavelength ranges showing values that were deemed to make up the continuum were selected by hand, after which these values were fitted to a simple linear function, by which the data were then normalized. Several of the spectra show a blending of the lines included in our analysis with MW and geocoronal features. Wavelength ranges affected by this were masked out and not included in the analysis.

The normalized lines and immediate surroundings were then interpolated linearly and evaluated on the coarsest grid in velocity space (i.e., the velocity step size of the line with lowest rest wavelength of each galaxy).

Figure 4 shows in the upper panel for each galaxy the normalized and resampled absorption profiles in velocity space of the Si ii transitions at λλ 1190, 1193, 1260, and 1334. Masked out wavelength ranges are not shown. The zero point value is the systemic velocity derived from the SDSS spectroscopy of the nebular lines of the H ii regions surrounding galaxies' star-forming regions. The profiles show a broad range of velocity shifts relative to systemic, as well as a variety of widths, depths, and shapes. The lower panels of Figure 4 show, for reference, the profile of Si ii λ 1304 (the only one of the transitions present in all 14 LARS galaxies), along with the profiles of O i λ 1302 and C ii λ 1334. Also, in thick black, is shown the average of these and the Si ii lines, or the subset hereof available in the given galaxy. Data shown in Figure 4 have been smoothed by a flat kernel that is five bins wide for clarity.

Zoom In Zoom Out Reset image size

Despite their different oscillator strengths, there is a tendency for the lines to have almost equal depth and shape. This is usually an indicator that the profiles are the result of absorption in an optically thick medium with a less-than-unity covering fraction; for a fully covering but partly transparent gas screen, the transitions of different strengths would show a corresponding difference in the depth of their absorption features (see Section 3.4).

Like the LIS lines, the Si iv transitions were resampled in velocity space, as described above, and averaged for comparison with the LIS lines. These are shown in Figure 5, with their propagated standard errors shown as a surrounding shaded region.

Zoom In Zoom Out Reset image size

The presence of residual flux in a given velocity interval within an absorption feature can indicate either that the gas at these velocities is not completely opaque, or that the gas is opaque, but only partly covering the background light source; or a combination hereof. In the presence of multiple lines arising from the same ground state, this can be disentangled by utilizing the differences in oscillator strengths fλ of said lines in the apparent optical depth method described by Savage & Sembach (1991), used by Pettini et al. (2002) and Quider et al. (2009). The basic idea behind this method is that the different lines, in the case of an optically thin medium, will show different absorption depths depending on their oscillator strengths, while an optically thick system with partial covering will yield identical profiles for the different transitions. An extension of the method is employed in this work, which allows us to compute the column density and covering fraction of these lines as a function of velocity offset from the systemic zero-point (Jones et al. 2013). For want of any suitable transitions of H i to directly measure the covering fraction of the neutral ISM, we use the presence of four Si ii lines within the wavelength range of our spectra. The method is based on the following.

For a given velocity range, the column density N and the covering fraction f_C can be calculated from the ratio of residual intensity to continuum level by the following relation:

$$\begin{eqnarray}&&\frac{I}{{{I}_{0}}}\;=\;1-{{f}_{C}}\left( 1-{{e}^{-\tau }} \right).\end{eqnarray} \tag{ 1 }$$

The optical depth τ is related to the column density and oscillator strength as:

$$\begin{eqnarray}&&\tau \;=\;f\lambda \;\frac{\pi {{e}^{2}}}{{{m}_{e}}c}N\;=\;f\lambda \;\frac{N}{3.768\times {{10}^{14}}},\end{eqnarray} \tag{ 2 }$$

where f is the oscillator strength, λ the wavelength in Å, and N the column density (in the given velocity range) in cm⁻². In the presence of multiple absorption lines with different values of f λ, arising from the same energy level in the same species, we can fit the data for N and f_C for each velocity bin. Our spectra cover the wavelengths of four lines arising from the ground state of Si ii with the wavelengths λλ 1190, 1193, 1260, and 1304.

Practically, this computation was performed by formulating I/I₀ as a function of f λ, and finding the values of N and f_C that minimize the squared residuals of subtracting the measured I/I₀ from the theoretical ones. The procedure for the minimization was to create a simple grid of (N, f_C) value pairs limited by the physically sensible, find the squared residuals in each point, and then find the combination that minimized this function. Confidence levels for the two parameters were found by projecting the $(\chi _{{\rm min} }^{2}+1)$ contour (see below) onto the N and f_C axes.

It is important to note at this point that the values of f_C(v) found in this way are not what one would usually call the "covering fraction" of the neutral gas. The f_C are the covering fractions of individual slices in velocity space of the gas, whereas the covering fraction in the usual sense of this term, from here on denoted F_C with a capital F, is the covering factor of all neutral gas regardless of velocity. Since the individual velocity-space slices generally do not occupy the same projected physical space, it is entirely possible to have f_C < 1 everywhere while still having a total F_C ≃ 1; the found values of f_C provide only a lower limit for F_C.

3.4.1. Fit Quality Indication

3.4.2. Averaged Absorption Profiles

Additionally, also in analogy with Jones et al. (2013), we have used a simple, complementary method to estimate the covering fractions by investigating saturated transitions. In the case where τ $\gg$ 1, Equation (2) reduces to:

$$\begin{eqnarray}&&I/{{I}_{0}}\;=\;1-{{f}_{C}}\end{eqnarray} \tag{ 3 }$$

In this case, we would expect all reasonably strong lines to show the same absorption profile, and said profile will be a good proxy for the covering fraction. The Si ii absorption profiles in Figure 4 suggest that the four Si ii lines are indeed saturated, as do similar plots of the lines of O i λ 1302 and C ii λ 1334.

To minimize random variations, an inverse variance weighted average of the low-ionized absorption profiles is used for further analysis; as usual, contaminated regions are masked out and not included in the final profile.

Figure 6 shows the averaged absorption profile of LARS 1, with the computed covering fraction overlaid, along with the computed column densities on logarithmic scale and the pseudo-reduced χ² described in Section 3.4. It is apparent that the averaged line profile does indeed follow the computed covering fractions quite closely and thus is indeed a good proxy for the actual covering fraction. The other galaxies of the sample show similar behavior, as can be seen in Figure 8 through 10.

Zoom In Zoom Out Reset image size

The same figures also show that these averaged profiles have considerably less scatter than the computed values of f_C(v). Thus, after having shown that they do indeed follow the f_C(v) profiles well, we can benefit from adapting the averaged profiles as proxies for these in our further computations, although with some words of caution for a few of the galaxies (see the Appendix).

When describing the kinematics of the neutral gas, a good set of characteristic velocities is needed. The approach usually taken for simple line profiles is to approximate them to a simple or possibly skewed Gaussian or Voigt profile, and describe these by their centroid and FWHM or, for the Voigt profile, the broadening parameter b. However, our profiles are generally too complex and asymmetric to be well approximated by a single profile. Instead, we characterize the profiles by a set of velocities based on their integrated area as follows:

• Integrated central velocity (v_int). The velocity that has 50% of the integrated absorption on each side and so acts as a figurative balancing point for the profile. This is more accurate for irregular line shapes than the centroid of an approximated Gaussian or Voigt profile, which will be misleading for an asymmetrically shaped feature. Error bars are found through Monte Carlo simulation by perturbing the measured profile with random errors drawn from a Gaussian fit to the measured error distribution and repeating the above computations through 100 iterations, after which the standard deviation of the distribution of obtained values is reported as the uncertainty.
• Velocity width (W_90%). The distance on the velocity axis from 5 to 95% integrated absorption. This description of the line width is more robust to irregular line shapes than the more common FWHM, yet also robust to the uncertainty determining the exact border between what is considered a continuum and what is considered an absorption feature. It should, however, be kept in mind when comparing to other results that this width encompasses a larger fraction of the total velocity distribution than the FWHM. Error bars are found through the Monte Carlo method described above.
• FWHM. Standard FWHM in velocity space. This is found using the Python package statsmodels to fit the observed profile to a Local Linear Estimator (LLE) with a kernel width found through cross validation. Error bars are found through Monte Carlo simulation. The FWHM is not very sensitive to choice of kernel width.
• Velocity at 95% integrated obsorption (v_95%). The absolute value of the velocity that has 95% of the integrated absorption on its red side. This velocity gives a tracer of the far end of the velocity distribution relative to the systemic zero point.
• Velocity at minimum intensity (v_min). The velocity at which the intensity of the line has its minimum. In a perfectly symmetric absorption feature, this will be coincident with v_int.
• Lyα peak velocity ($v_{{\rm peak}}^{{\rm Ly}\alpha }$). For the galaxies that show Lyα emission in the COS aperture, we also report the velocity of the peak of this emission line. Like the FWHM, this is found fitting it to an LLE with a kernel width found by cross validation.

Figure 7 shows an illustration of the quantities mentioned above, except for FWHM, which was measured separately and is assumed to be well known. The averaged absorption profile is drawn in black steps, the absorbed area in shaded green. The velocities v_int, v_min, v_95%, and W_90% are shown along with the residual intensity at maximum absorption.

Zoom In Zoom Out Reset image size

Besides these quantities, we performed preliminary computations of the continuum-subtracted Lyα flux within the COS aperture to evaluate how well the global Lyα properties reflect the Lyα properties that arise as a direct consequence of the configuration and kinematic properties of the neutral gas.

An overview of the measurements described above is listed in Tables 4 and 5.

4.1.1. Low-ionized State

4.1.2. High-ionized State

Neutral gas covering in narrow velocity ranges were computed for all 14 LARS galaxies based on the method described in Section 3.4. The results are visualized in Figures 8–10, with black steps showing the averaged LIS profile and propagated one σ errorbands in shaded gray. Red dots indicate, on a reversed scale marked in red on the right, the computed neutral covering fractions for each velocity step. The pink shades show the confidence levels of the covering fractions, cut off at zero and one as subzero and above-one fractions are unphysical.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We find that while there is a certain scatter in the computed covering fractions, they generally coincide very well with (1 − I/I₀) of the averaged low-ionized absorption profile. In some galaxies, there is a tendency for the covering fraction plots to dip a bit below the stacked profiles, especially in the velocity ranges close to maximum absorption. These ranges are generally coincident with ranges in which either the Si ii λ 1304 transition profile lying visibly above the other lines, or the Si ii λ 1260 line profile lying significantly below the others. If these are due to polluting features or other systematics, this would cause the stacked profile to slightly misrepresent the actual physical system, and could also cause minor systematics in the fits for f_C(v). This effect introduces uncertainties about the presence of residual emission at maximum absorption for LARS 7 only; in general, the assumption of complete optical thickness seems well justified. From this point, therefore, we have adopted (1 − I/I₀) as a proxy for the velocity-binned covering fractions, and we shall report the maximum fractional depth of the profile as the maximum velocity-binned covering factor, f_C,max (see e.g., Table 4).

About half of these stacked profiles are saturated in absorption, while the rest show some residual intensity. Eight of these galaxies—LARS 1, 2, 5, 6, 7, 12, 13, and 14—show residual intensity at maximum absorption. All the strong Lyα emitters from (Hayes et al. 2014)—LARS 1, 2, 5, 7, and 14—are found within this group.

Three galaxies—LARS 6, 9, and 10—show deep, damped Lyα absorption features in the aperture. Their LIS lines have in common a relatively deep, static component of the neutral medium, as discussed in the introduction. This component is completely black at the line center for LARS 9 and 10, but not for LARS 6. Interestingly, only LARS 6 is also a global absorber in Lyα. This points to LARS 9 and 10 emitting the majority of their Lyα in extended halos, which is consistent with what is visible in Figure 1. LARS 4, which is a global absorber, shows some weak emission in the aperture, although not enough to outweigh what is absorbed in the outer regions.

The average LIS profiles and the Lyα profiles of LARS 6, 9, and 10 are shown for comparison in Figure 11. Especially LARS 9 and 10 show a quite striking contrast in width between the metal and the H i profiles.

Zoom In Zoom Out Reset image size

One major objective of this work is to compare detailed knowledge about conditions close to the hot, star-forming regions that give rise to the large majority of intrinsic Lyα photons obtained through SDSS and COS spectroscopy, with the strength and morphology of global Lyα output, which is believed to depend strongly on these central conditions. However, an important requirement for this to bear any physical significance is that the conditions of these sight lines be representative of general conditions in the galaxies studied. As a test, we computed the Lyα flux emitted into the aperture as a fraction of the intrinsic Lyα emission in the aperture, inferred from the above fluxes and Hα measured from SDSS spectroscopy (Östlin et al. 2014).

Radiative transfer of Lyα is, due to its strong resonance, a global phenomenon for a galaxy. Photons interacting along the LOS, seemingly absorbed, may eventually escape the galaxy, while photons originating from regions outside the aperture may scatter into it. We define the local emitted Lyα fraction, $f_{{\rm em},{\rm local}}^{{\rm Ly}\alpha }$ as the Lyα flux emitted into the aperture in units of the intrinsic Lyα flux within the aperture; this is listed in column 5 of Table 5. This is not exactly an escape fraction, but the closest comparable quantity that can be meaningfully defined in a local region of a galaxy.

Table 5.  Lyα Properties of the LARS Galaxies

ID	$W_{{\rm Ly}\alpha }^{{\rm glob}.}$	$v_{{\rm peak}}^{{\rm Ly}\alpha }$	$f_{{\rm esc}}^{{\rm Ly}\alpha }$	$f_{{\rm em},{\rm local}}^{{\rm Ly}\alpha }$
	(Å)	(km s−1)
(1)	(2)	(3)	(4)	(5)
LARS 01	33.0	125 ± 7	0.119	0.028
LARS 02	81.7	150 ± 7	0.521	0.093
LARS 03	16.3	347 ± 8	0.003	0.000
LARS 04	0.00	443 ± 7	0.000	0.003
LARS 05	35.9	166 ± 7	0.174	0.032
LARS 06	0.00	⋯	0.000	0.000
LARS 07	40.9	178 ± 7	0.100	0.037
LARS 08	22.3	114 ± 7	0.025	0.005
LARS 09	3.31	⋯	0.007	0.000
LARS 10	8.90	⋯	0.026	0.000
LARS 11	7.38	⋯	0.036	0.000
LARS 12	8.49	415 ± 8	0.009	0.006
LARS 13	6.06	276 ± 9	0.010	0.003
LARS 14	39.4	254 ± 8	0.163	0.119

Note. Column 2 shows global Lyα equivalent width; column 3 shows (where present) the Lyα peak velocity computed as described in Section 3.5; column 4 lists global, imaging-derived Lyα escape fractions; and column 5 lists fraction of intrinsic Lyα emitted into the COS aperture. Values of columns 2 and 4 were first reported in Hayes et al. (2014).

Download table as:  ASCIITypeset image

Figure 12 sees these locally emitted fractions plotted against the global Lyα escape fractions from Hayes et al. (2014). The data show a clear correlation, with LARS 2 and 14 as a outliers, with 14 lying just outside the 95% prediction band, while LARS 2 falls somewhere between the 68 and 95% curve (see discussion below). LARS 14 shows a noticeably stronger local emission relative to the global output than the remainder of the sample, while on the other hand LARS 2 has a stronger global and weaker local Lyα output than usual. The full sample shows a Pearson-r for these two quantities of ∼0.77. Evidently, the aperture regions are well representing the Lyα RT conditions in the galaxies of the LARS sample. This also means that in all subsequent figures showing relations involving $f_{{\rm esc}}^{{\rm Ly}\alpha }$, the in-aperture version would look essentially the same, except for LARS 2 and 14, which would have lower and higher local escape, respectively.

Zoom In Zoom Out Reset image size

4.4.1. Lyα Escape

Previous studies of nearby galaxies (Kunth et al. 1998; Wofford et al. 2013) conclude that an outflowing bulk motion of the neutral medium is necessary for Lyα to escape: Kunth et al. (1998) analyzed observations of eight galaxies and found velocity offsets of up to 200 km s⁻¹ between O i/Si ii absorption lines and emission from the hot, ionized gas in the emitting galaxies, while galaxies with a (near-) static ISM showed deep, damped absorption features in Lyα.

Our findings from the LARS sample are consistent with this picture (see Table 4). Like Wofford et al. (2013) we also stress, however, that other mechanisms must be in play; with a Pearson's r = −0.1 there is no statistical correlation between $f_{{\rm esc}}^{{\rm Ly}\alpha }$ and v_int in general, and we do see cases of galaxies with outflowing winds and yet almost no emission in Lyα, although no emission through a static medium is observed.

Outflowing winds must be present to allow for Lyα escape, but as can be seen in Figure 13, it is not sufficient to guarantee this: LARS 3, 8, 11, and 13 show significant outflow velocities, but still very low Lyα escape fractions. Evidently, other competing effects must have significance.

Zoom In Zoom Out Reset image size

4.4.2. Lyα Peak Velocity

From current models of Lyα radiative transfer (e.g., Verhamme et al. 2006), it is predicted that in an environment of a spherically symmetric outflowing medium of high H i column density, the emission feature will be dominated by a redshifted peak with a maximum around $v_{{\rm peak}}^{\alpha }\sim 2\times {{v}_{\operatorname{int}}}$, a component that is backscattered on the far, inner, receding surface of the expanding shell, whereas the component emitted directly toward the aperture in these models is predicted to be largely suppressed while traversing the medium. According to these models, we would expect to see a correlation between the outflow speed and Lyα emission peak velocity.

It is therefore a bit surprising that we find no such correlation in our sample. In Figure 14, these quantities are plotted against each other for the 10 spectra in LARS with Lyα emission. In the figure is also shown, in dotted dark gray, the least squares regression line weighted by the uncertainties in v_int. Uncertainties on $v_{{\rm peak}}^{{\rm Ly}\alpha }$ are negligible in comparison and were not included in the fit.

Zoom In Zoom Out Reset image size

The figure clearly shows that the expected correlation is not there. There is in fact a weak anticorrelation between these velocities; the direct opposite of the expected result, although with a Pearson-r ≈ 0.4 it is not a strong correlation and should be taken with a grain of salt. What we observe is more akin to the simpler model discussed in Mas-Hesse et al. (2003) of an emission feature overlaid by a multiplicative absorption profile.

One should be careful when interpreting the covering fractions presented in Section 3.4. The measured maximum velocity-binned covering fraction ${{f}_{C}}({{v}_{{\rm min} }})$ (see, e.g., Figure 7) constitutes only a lower limit to the total covering fraction F_C (Jones et al. 2013): a system at velocity other than v_min may cover a projected area with only partial or no overlap with the area covered by the gas at v_min. Such a system will add to the total covering fraction, increasing it beyond the minimum set by ${{f}_{C}}({{v}_{{\rm min} }})$—unless, of course, ${{f}_{C}}({{v}_{{\rm min} }})$ is unity.

Figure 15 shows a plot of the Lyα escape fraction vs. the maximum velocity-binned covering fraction f_C,max as described above. A negative residual intensity is unphysical and would be interpreted as zero. It is immediately clear that at a covering fraction of 1, very little Lyα radiation escapes. This physically corresponds to a full sheet of neutral gas screening off the emitting star clusters. All Lyα emitters of the sample have ${{f}_{C,{\rm max} }}\lesssim 0.9$.

Zoom In Zoom Out Reset image size

One quite surprising finding of this work came from comparing Figure 15 to Figure 5 in Hayes et al. (2014), panel 3 in the lower row, comparing W(Hα) to $f_{{\rm esc}}^{{\rm Ly}\alpha }$. The similarity is quite striking, pointing at a possible relation between EW(Hα) and f_C,max. Figure 16 shows that this is indeed the case, the two are—with all due reservations regarding sample size—strongly correlated. This is, to our knowledge, the first time such a correlation has been observed.

Zoom In Zoom Out Reset image size

4.7.1. Kinematics and Geometry

Our results suggest a picture of a neutral medium consisting of multiple dense clouds flowing predominantly outward at a range of velocities, although contributions redward of line center are also found in some of our sample galaxies. Considering the width and morphology of the line profiles, it seems clear that they consist of multiple contributions from separate subsystems that move outward at a range of velocities and therefore must be physically disjointed. These clumps are generally optically thick near the centers of the Si ii lines, but each may only partially cover the starburst region and hence leave residual flux even at the central wavelength of the absorption line for each subsystem. Together, the absorption features will blend into one line of complicated morphology, as observed. At each velocity, a part of the starburst is covered by a completely opaque cloud, while others are unblocked.

Consider a toy system of two clouds that each cover 50% of the starburst. If these move at the same velocity, they will block all light shifted into the line center in that frame. If, however, they have a velocity difference large enough to completely deblend the line contributions, the resulting profile would show two separate dips in flux density down to 50%, while at least 50% of the photons at any given wavelength would encounter no opacity at all when traversing the system.

The model is illustrated in Figure 17, which should be read as a radial beam cut out of a spherically symmetric system. The stars to the left signify the background sources, including the H ii regions in which the recombination lines like Lyα are emitted. Next to these, optically thick clumps of the neutral medium are shown as gray disks, moving outward at different velocities. The FUV continuum shining through the clumps is shown as green arrows and Lyα radiation is shown in blue, in accordance with the coloring scheme in Figure 1. Fractions of the FUV pass through the clouds, which each leave saturated absorption features in parts of the light. If the difference in velocities surpasses the line widths of the individual systems, the absorption feature of each clump will fall at different wavelengths in the spectrum, and the line will show residual flux everywhere, even though it consists of overlapping, saturated clumps that combined cover the source completely. The three green absorption profiles shown in the middle are the line imprints left by the individual clumps, and the larger green profile in the upper right shows the resulting profile when the light from the three different projected regions is integrated over the aperture: The simple, saturated contributions of each clump can, even when the clumps are together fully covering the background source, add up to a quite complex line profile with residual flux everywhere. In the lower right, in blue, is shown a typical P Cygni-shaped Lyα profile as it would often look from a system with unity combined covering fraction like this one.

Zoom In Zoom Out Reset image size

In Figure 18 we show for comparison a sketch of the "Picket Fence" model of Heckman et al. (2011). It is very similar to Figure 17, except the clumps are now only partly covering, leaving holes in the neutral medium allowing for direct escape of Lyα and hence also LyC. The individual clumps still leave saturated absorption lines in parts of the FUV light, which add up to a more complex line profile with residual intensity everywhere. Here, the residual intensity is only in part due to the velocity offsets between the clumps, while a contribution comes directly through holes in the neutral medium. The combined metal absorption feature of such a system is difficult to distinguish from the one seen in the F_C = 1 case above. The Lyα feature will, however, have a nonzero flux in the line center and, depending on the size of the holes, maybe even be dominated by the narrow, undistorted component directly leaking through. We therefore argue that a well resolved Lyα spectrum and precise systemic velocity zero points are necessary in order to distinguish between these two similar, but not identical cases.

Zoom In Zoom Out Reset image size

4.7.2. ISM Evolution

A set of properties of the neutral medium of the LARS galaxies measured by single-dish and interferometric 21 cm-band observations are presented in Pardy et al. (2014). We shall now briefly compare this global description of the neutral medium to our own description of the local properties along the sight lines to the most luminous star-forming knots.

The most directly comparable quantities are the line width of the neutral medium as measured directly, but globally, from 21 cm observations of atomic hydrogen, and as inferred using the low-ionization metal lines as a proxy. The numbers are shown in Table 4, which in column 8, FWHM_{Si ii}, shows the line widths as measured from the metal lines, and in column 9, FWHM_{H i}, shows the values from Pardy et al. (2014), in which FWHM_{H i} is reported as W₅₀.

The 21 cm measurements, which include large amounts of cool gas in the interstellar and circumgalactic medium, show significantly lower line widths than the COS measurements, which only target the neutral medium along the sight lines toward the brightest and hence most energetic star-forming knot in each galaxy. While there is no one-to-one relation, it is evident from Figure 19 that there is a strong correlation: Greater metal line widths within the aperture are also indicative of greater global line widths in H i. The 21 cm feature is dominated by the cool phase ISM, and its line width by large-scale motion, such as rotation. In contrast, the LIS lines are dominated by gas in the warm neutral medium in the central regions of our galaxies, which is heated and pushed by star-formation feedback. It is therefore not surprising that the 21 cm lines are narrower than the LIS lines; on the other hand, given that the 21 cm lines are global and the LIS lines such a local phenomenon, it is interesting and a bit surprising to see them correlate so well.

Zoom In Zoom Out Reset image size

It has been established both theoretically (Neufeld 1990; Verhamme et al. 2006) and observationally (e.g., Kunth et al. 1998; Mas-Hesse et al. 2003; Wofford et al. 2013) that outflows play an important role in regulating Lyα escape, with outflowing gas being Doppler shifted out of resonance with the Lyα photons emanating from the center as the main mechanism. It is still not clear, however, exactly how important this role is.

A number of surveys have consistently shown that stronger outflow velocities are correlated with higher Lyα escape fractions. Kunth et al. (1998) observed a sample of eight galaxies in the local universe and concluded that gas kinematics, rather than dust content, was the dominant factor governing Lyα escape. A similar conclusion was reached by Mas-Hesse et al. (2003) from a subsample of the galaxies from Kunth et al. (1998), which expands upon these results and proposes a unifying scenario of expanding superbubbles as explanation for the different profiles observed. In Wofford et al. (2013), a similar conclusion about the role of outflows was reached from a considerably larger sample of local star-forming galaxies, with better data.

Our findings are consistent with outflows being of key importance, although we observe this relation as an upper envelope rather than a correlation: while stronger outflow velocity allows for a higher Lyα escape fraction, we do observe galaxies (LARS 3, 8, 11, and 13) with high wind velocities and yet very low Lyα escape fractions. These galaxies have in-aperture dust contents in the upper half of the sample, which in its entirety has E(B − V) spanning from 0.047 ± 0.02 for LARS 6 to 0.688 ± 0.015 for LARS 3 (Östlin et al. 2014). In addition, their ISM span wide ranges in velocity, further making it harder for Lyα photons to be shifted out of resonance wavelengths.

Furthermore, as is discussed in Section 4.7, we suggest that the importance of the outflowing wind lies, not only in shifting Lyα photons out of resonance, but also in creating the Raleigh–Taylor instabilities that introduce some clumpiness and porosity that will carve out escape routes for Lyα photons through the neutral medium. We stress again that some fragmentation of the medium is sufficient for this to happen; no direct sight lines to the background sources are necessary and are indeed not likely to be present in our sample, with a possible exception in LARS 14.

For Lyα photons to escape from a galaxy they must traverse the neutral medium and undergo a number of scatterings, for each of which it suffers an increased chance of absorption in the meeting with a cosmic dust grain. In a medium of zero net velocity but an internal velocity distribution wider than the intrinsic line width of the Lyα line and with a nonzero dust content, any photons will undergo a large number of scatterings and a resulting increase in path length to travel out of the system. Increased path length means increased optical depth due to dust and thus a higher probability of absorption of the photon. It is practically impossible for radiation to escape such a system (Neufeld 1990).

A number of factors can drastically lower the absorption probability and thus raise the escape probability for a Lyα photon:

• H i Mass. The mass of neutral gas in the system governs how many times a photon is likely to scatter and therefore has a strong influence on escape probability. Lower mass may significantly raise the escape fraction of intrinsic Lyα.
• Neutral gas outflow. A bulk outflow of neutral gas will Doppler shift the atomic hydrogen out of resonance with the intrinsic Lyα, moving the redder parts of the line out of resonance, from where it is free to escape.
• Velocity range. A higher internal velocity range causes the hydrogen absorption profile to broaden, requiring a larger (Doppler) redshift of the intrinsic Lyα photons to grant them easy escape. Correspondingly, a lower velocity range will make it more likely for a photon to be redshifted sufficiently to escape afterward, which raises the escape fraction.
• Dust content. Interactions with dust grains are what destroys Lyα photons, so it is no surprise that higher dust content will result in lower escape fraction, as is the case for any other optical and UV photons.
• ISM perforation. As mentioned in Section 4.7, galaxies in LARS with any Lyα escape are all drawn from the part of the sample that has a maximum velocity-binned covering fraction f_c,max ≲ 0.9. This value depends strongly on the bin width. For a bin width less than or equal to the resolution of the spectrograph, bin width should not affect the velocity-binned covering fraction, but for larger bin widths, the covering fraction will grow monotonically with bin width because more velocities will be allowed to contribute to each bin.

The interplay between these competing effects can be intricate and unpredictable. In Figure 13, one can see LARS 2 far above and LARS 3, 8, 11, and 13 falling far below what would otherwise have been a neat correlation between outflow velocity and global Lyα escape fraction as predicted by Kunth et al. (1998) and concluded in Wofford et al. (2013). LARS 3, 8, 11, and 13 have the highest in-aperture dust contents of the sample (Östlin et al. 2014, Table 3). LARS 2, on the other hand, has the third lowest dust content in the sample, and is only surpassed by LARS 5 and 6. LARS 5 has a fairly high Lyα escape fraction despite a high velocity range; the low dust content and high outflow velocity help counter this. The medium of LARS 6 has a strong static component, paired with the highest line width of the sample. These conspire, despite the low dust content, to yield a very low escape fraction. LARS 2, on the other hand, shows the lowest W_90% of the sample. So for LARS 2, despite low outflow velocity, dust content and low velocity range conspire to aid in a larger Lyα escape fraction, whereas for LARS 3, 8, 11, and 13, the relatively high wind velocities are counteracted by a combination of high dust content and relatively high line widths to impede Lyα escape. These factors are certainly not the only ones to govern Lyα escape, but they serve to illustrate that the problem is multi-faceted.

The analysis of ISM covering and column density is interesting for the question of Lyman Continuum escape, which is important to solve the question of which sources were responsible for reionizing the universe. In the local universe, almost no leaking galaxies are observed, and the ones that are detected show very low escape fractions. Yet we know that around the time of reionization, the ionizing escape fraction must have been ≳0.2 (Robertson et al. 2013).

Heckman et al. (2011) analyzed a sample of 11 LBG analogs as possible leakers of Lyman Continuum radiation. They considered two different cases: one in which the UV sources are covered by a uniform shell with unit covering factor, and one in which they are covered by a number of optically thick clumps with openings between them. In the first model, any residual flux in the line would be the result of an optically thin medium. Since the optical depth at the Lyman Continuum is significantly larger than in the metal lines included in their analysis, such a residual flux would be necessary but not sufficient for allowing the escape of ionizing photons.

The second model they consider is the so-called picket fence model, in which the residual flux is interpreted as the result of a medium consisting of optically thick clouds that are only partly covering the background source. In a different implementation of the apparent optical depth method, they utilize the Si ii lines present in the COS spectra to determine which of these two cases is the more realistic. They conclude that where residual intensity is present in their spectra, the relative optical depths show that the picket fence model is more plausible, and thus conclude that galaxies in which this residual intensity is found are likely candidates for LyC escape. They point out that these objects all show anomalously low Hα emission, which is another indication that ionizing radiation is escaping. We note here that the latter finding seems to be in contradiction to the result of this work: that a lower f_C is quite strongly correlated with a higher Hα equivalent width. However, this might be a selection effect, as the two samples are selected from different criteria. Observation of a local example of a galaxy with a Picket Fence type ISM was recently confirmed by Borthakur et al. (2014).

Jones et al. (2013) also find that their galaxies, this time at redshifts 3 and 4, generally show residual flux in the lines, but they are more cautious, concluding that this implies any direct lines of sight (and thus escape) through the medium. They note as a caveat that line widths are in general much higher than the typical internal velocity dispersion of the gas, meaning that the line is a conglomerate of contributions from many separate subsystems. These systems generally do not occupy the same (projected) space, and thus a low measured covering fraction is still consistent with unit covering integrated across the line profile. However, a lower f_C would raise the probability of the medium being perforated to some degree, and they point out that generally, their galaxies at z = 4 show lower covering than at z = 3. Adding the fact that LyC escape is extremely rare in the local universe and at low z and extrapolating, this means that LyC escape could be even more common earlier and thus allow galaxies to be the main source of reionization of the early universe.

The high resolution of the COS spectra and the accurate systemic velocity zero points of LARS allow for a thorough analysis and modeling of the Lyα line profile morphology (I. Orlitova et al. 2015, in preparation). Visual inspection of the LARS Lyα profiles shows that, with the possible exception in LARS 2 and LARS 14 discussed above, we generally see no signs that lower covering factors coincide with emission at line center in Lyα, which is the signature of direct Lyα and LyC escape (e.g., Verhamme et al. 2014). We should mention that the optical depth in Lyα is higher than in LyC, so in cases where Lyα is quenched by an attenuated interclump medium, this could possibly still be penetrated by LyC. Only LARS 14 shows any signatures of column densities that could possibly be low enough for this to happen, and that shows no signs of LyC escape as inferred from the Lyα profile and the analysis of Heckman et al. (2011).

One possible caveat to our results is radiative transfer effects in the metal lines of the neutral medium, in which light is re-emitted into the absorption trough of the Si ii lines, as described by Prochaska et al. (2011) and Scarlata & Panagia (2015). The former work shows a more general model of how this effect can work, while the latter concerns itself more specifically with the Si ii lines within the COS spectral range.

In short, photons that are absorbed in the Si ii lines will be re-emitted in random directions, meaning that there is a probability that a photon will be scattered back in to the aperture of the spectrograph, partly filling up the absorption trough. Prochaska et al. (2011) showed that this effect, in certain circumstances, can cause a partially transparent, fully covering system to appear as an optically thick, partly covering system by causing transitions of a given species with different line strengths to appear as having equal depth. Among factors that could inhibit this effect they list (a) the aperture covering only a small fraction of the neutral medium of a galaxy, causing most of the re-emission to be blocked by the aperture, or (b) the presence of fine structure splitting of the ground state of the given transition. In the latter case, an absorbing electron can decay into the short-lived, upper fine structure level of the ground state, at a wavelength slightly higher than the absorbing transition to immediately after decay into the lower fine structure level by emitting an IR photon. The short lifetime of the upper fine structure layer ensures that no absorption happens at this wavelength, allowing a photon at this wavelength to leave the system without further atomic scattering events.

The strength of the latter effect is difficult to estimate, as it depends strongly on the typical number of scatterings a photon undergoes before escaping; for each scattering, there is a fixed fractional probability of re-emission into the fluorescent channel, and the typical number of scatterings is not well known, and could well depend strongly on, for example, geometry and kinematics in the system.

Scarlata & Panagia (2015) show in more detail how this affects the Si ii lines specifically. While the effect predicted in this work is more modest than the ones predicted in Prochaska's model, they do show that it is possible with an isotropic and fully covering outflow model to reproduce a stacked UV COS spectrum of 25 Lyα emitters, which appears to have the characteristic Si ii absorption signatures of an optically thick, partly covering system. The Si ii transitions involved here arise from a fine structure split ground state, allowing a fraction that is unique for each transition to escape through the fluorescent channel. Given the assumptions in their paper, this provides a direct way to estimate the strength of the refilling. The resonant and fluorescent emission happens in the same regions and under the same physical conditions, and at a fixed ratio per scattering event. Thus, if no fluorescent emission is present, neither will any significant resonant emission be. If fluorescent emission is present, although there is no noticeable absorption in the resonant channel, this can indicate that a denser neutral clump is present inside the aperture, but off the LOS. Jaskot & Oey (2014) present an interesting discussion of how this emission, in concert with the Lyα line shape, can reveal information about the geometry of an object, in that the presence of such fluorescent emission combined with the absence of any significant absorption in the accompanying resonant line, reveals the presence of significant amounts of scattering neutral gas inside the aperture but off the LOS. Strong absorption but weak or absent fluorescent emission reveals that scattering is present, but happening mostly outside the aperture. The presence of both absorption and fluorescent emission shows that absorbing gas is present, and most of the absorbing neutral gas is found inside the aperture. The strength and width of the Lyα profile can show additional information about the LOS column density of neutral hydrogen.

The suppression of the refilling effect due to small aperture would be strong for some LARS galaxies, but not for all; and Scarlata & Panagia (2015) show that even with a reasonably placed and sized slit, the effect could still be strong enough to emulate a system of opaque clumps. The strength of suppression by escape through the fluorescent channel would depend strongly on how many scattering events one photon would, on average, undergo before escaping the system, which is not known. Furthermore, it is unknown how strongly these effects depend on the assumption of spherical symmetry and Sobolev approximation, and how they would react to an anisotropic, clumpy medium, for example.

One way to possibly assess how strong this effect would be is by looking at the blue wing of the absorption features in question. The four Si ii transitions in question span more than a factor of 10 in line strength; if they line up neatly even at wavelengths where re-emission is expected to be weak, this would suggest that the absorbing systems are truly optically thick and only partially covering. Since re-emission in the fluorescent channel happens from the same atoms as resonant re-emission, the fluorescent lines could possibly be used to constrain whether resonant re-emission would happen in the full width of the absorption feature and hence whether the blue line wing is clear of this self-contamination. Preliminary measurements on the LARS galaxies suggest that the blue wings of our absorption profiles are not behaving differently from other parts of the absorption features.

At the current time, however, more research is needed before we know how strong an effect, if any, this has on the line shape and the interpretation of metal absorption lines; so at present, we shall just mention this as a possible caveat.

LARS 1 is one of the five strong global emitters in the sample, "strong" being defined by an integrated $f_{{\rm esc}}^{{\rm Ly}\alpha }\gt 10\%$ (see Hayes et al. 2014). All four Si ii line profiles fall inside the detector ranges and do not seem to be polluted by other features. The strength of the absorption differs somewhat, especially for λ 1304 Å, which has the lowest resonator strength of them all, indicating that this system might not be completely optically thick. Nonetheless, as can be seen from Figure 4, the fitted covering fractions are still close to the absorbed fraction, indicating that complete optical thickness is still a good approximation.

The profiles of the lines at λλ 1193, 1304 show smaller secondary spikes on the red side. The spike redward of Si ii 1193 is emission in the fluorescent Si ii* 1194 Å line, which shares the upper energy level with Si ii λ 1190.

LARS 1 shows strong Lyα emission in a classic P Cygni profile, the absorption part of which coincides neatly with the metal absorption feature, confirming that we are tracing the same gas. While the f_C(v) computed from the metal absorption lines shows a covering <1 for all velocities across the line feature, the Lyα feature shows no sign of direct sight lines to the H ii regions.

LARS 2 is another strong global Lyα emitter that locally shows a strong emission profile in the COS aperture. Like LARS 1, it shows a classic P Cygni-like emission profile in Lyα, the absorption part of which coincides neatly in velocities with the averaged metal line absorption profile.

The silicon absorption lines reveal a relatively narrow profile that is only slightly blueshifted from the systemic velocity. The four Si ii transitions line up very consistently around what resembles a symmetric Gaussian absorption feature with a shallow, secondary feature on the blue side, which is somewhat deeper for the λ 1260 transition. A quite strong fluorescent emission is present in these lines and in O i.

LARS 2 is peculiar because of its very high (=0.56) global Lyα escape fraction, which is more than twice that of its nearest competitor, the Green Pea type galaxy LARS 14. This is despite a high inferred metal line covering fraction and low-outflow velocity; factors that are normally quenching Lyα escape. The galaxy does however have low dust content and a narrow velocity range of the neutral medium, both of which ease the escape of Lyα.

It is interesting to note that the minimum of the absorption in the Lyα profile falls blueward of line center, leaving a significant emission at systemic velocity. This could possibly be a directly escaping, non-scattered component finding its way through a hole in the medium. Such a contribution would only make up a minor fraction of the total escaping Lyα, which generally shows the redshifts typical of having undergone radiative transfer, and thus do not answer the puzzle of how such a large global escape fraction can coexist with such a large local covering factor.

LARS 3 is a weak Lyα emitter, both globally and in the aperture. The Lyα line shows a weak P Cygni profile with a relatively broad blue absorption component, suggesting high H i column density.

All the Si ii transitions exhibit a deep, broad absorption profile, which is almost or completely saturated. The averaged profile is the widest in the sample, apart from LARS 6, which is centered at ∼−100 km s⁻¹ but with significant absorption out to ∼−450 km s⁻¹ on the blue side and ∼150 km s⁻¹ on the red side. The strong red wing means that the O i* λ 1305 Å fluorescent line stands out more conspicuously than on most other galaxies, giving rise to the large uncertainties on f_C in the red wing shown in Figure 8, panel 3, as well as artificially slightly raising the residual intensity in the red wing of the averaged profile. The latter effect is counteracted by the C ii λ 1334 line being deeper than the rest in the same velocity range.

A similar uncertainty is seen in the blue wing, which is probably due to the λ 1260 line being somewhat deeper than the others here. The reason for this is not known, as it follows the other lines very well at line center and redward. The effect of this on the averaged profile is likely modest, because the five other lines all line up quite well in this velocity range, but it could possibly give rise to the large uncertainties in f_C in the blue wing of the line. The 1260 line being deeper in the blue wing is a tendency that can also be seen in LARS 4, 9, 10, 12, and 14, and less markedly so in LARS 8, leading us to believe this could be contamination from an unidentified line at λ_rest ∼ 1259 Å.

LARS 4 is one of the two global Lyα absorbers of the sample. In the COS aperture it shows weak but distinct emission, but even in-aperture, Lyα absorption outweighs this. The line shape is very similar to that of LARS 3, although the blue absorption component is narrower.

The averaged LIS metal line is likewise significantly narrower than that of LARS 3, but like LARS 3 it is completely black at line center for all Si ii lines except at λ 1304, which does show a bit of residual intensity in the line center of LARS 4. This can partly be because this line, having the smallest oscillator strength, is not completely opaque, but at least part of it is likely due to emission in O i* λ 1305—fluorescent emission is unusually strong in this spectrum, as well as for Si ii* 1194. Along with the 1190 + 93 doublet, both C ii 1334 and O i 1302 are black at line center and follow the profile of the Si ii doublet well. Si ii 1260 again follows nicely in the red wing, but is somewhat deeper in the blue wing. This could be the reason for the discrepancy between the f_C(v) profile and the average line profile in the blue wing; an artificially lowered residual flux for Si ii 1260 will mean a higher inferred covering fraction due to its high value of fλ, while having little effect on the inferred column density. At the same time, only four lines are included in the f_C computations, while six are included in the average absorption profile, lending weaker significance to one outlier.

The average LIS metal profile of LARS 5 is broad, but relatively shallow, and shows a notable pair of minima at ∼−90 and ∼−235 km s⁻¹, probably representing at least two distinct subsystems.

Si ii λ 1304 is markedly shallower than the rest of the lines, possibly indicating that the medium is not completely opaque at this wavelength. However, the best fits at line center still agree very well with the hypothesis of full opacity and part covering, and the f_C(v) profile coincides remarkably well with the average line profile. Fluorescent emission at λλ 1194, 1305 is present but is the weakest of the galaxies so far mentioned. LARS 5 is analyzed in depth in F. Duval et al. (2015, in preparation).

This is the second of the two global Lyα absorbers, and locally in the aperture it shows a strong, damped absorption profile.

The metal absorption profile is the most complex of the LARS galaxies. It consists of three narrow components: one infalling at v ∼ 280 km s⁻¹, one outflowing at v ∼ −210 km s⁻¹, and one static. The static component is the strongest in metal absorption, but the Lyα absorption profile seems to be centered around the infalling component.

The outflowing component is interesting because the depths of the 1190 + 1193 doublet suggests that this component is optically thin. The transition at λ 1304 follows the shape of the 1190 + 1193 doublet very well, but it is significantly deeper than both of these, which is unexpected, considering that λ 1304 is by far the weaker of the three lines. This is due to to the center of the infalling component of the neighboring O i λ 1302 transition coinciding with the line center of the blue component of Si ii λ 1304, so that while looking like a single Gaussian absorption profile, it is actually the sum of two lines.

Pardy et al. (2014) show that LARS 6 is in fact interacting with a neighboring system. One possible explanation for the double-dip absorption profile could be that one of the dips actually comes from a tidal tail of gas from this interacting system extending into the aperture.

LARS 7 is another of the strong global Lyα emitters, and also shows relatively strong emission within the COS aperture. The emission profile is of P Cygni type with a narrow blue absorption component.

The strong emission in Lyα is peculiar considering that the metal absorption line, like that of LARS 2, is deep and centered close to zero velocity, which are factors that should act to suppress Lyα escape.

However, similar to LARS 2, the velocity distribution in the neutral gas is narrow, allowing for the red wing of Lyα emission to still escape. LARS 7 is also the faintest of the five strong global emitters, with a global Lyα escape fraction of 0.14. The in-aperture peak Lyα flux for LARS 7 is only around a third of that of LARS 2, whereas both its dust content (Hayes et al. 2014) and its line width are somewhat higher.

The Si ii lines of LARS 7 show some difference in depth. The λ 1304 feature is significantly shallower than the rest, some of which could be fluorescent emission from O i* 1305. The relative depths of the λλ 1190 + 1193 doublet and clear emission in the corresponding fluorescent transitions suggests that some re-emission effects are present. This is not straightforward to interpret, but it is at least worth a word of caution that the results of the analysis of this galaxy could be biased by these effects.

LARS 8 is noteworthy for its strongly outflowing neutral medium. The average metal absorption feature is black at line center, wide, and has the highest velocity of the sample.

Despite the strong outflow, the galaxy just barely qualifies as a Lyα emitter. Globally, it has a Lyα escape fraction of ∼ 2%. In the aperture, it displays a P Cygni profile that is weak in emission and with a broad absorption component suggesting a high H i column density in the outflowing medium.

The four Si ii lines agree well in depth, although the shape of the red wing is a bit ambiguous; the transitions at λλ 1193 and 1260 both show a wing reaching across zero velocity and into infalling velocities, while the two other lines show no absorption at zero velocity, but do show a minor feature on the red side. Some gas is present at line center, but it does not match the shape of the Lyα profile quite as neatly as many other lines. This discrepancy could be due to a high column density of the outflowing neutral hydrogen, meaning that the low Lyα flux at line center would be due to damping wings from the P Cygni absorption feature.

Other mechanisms that could possibly account for the low output is the fact that the neutral medium covers a large range of velocities, thus quenching escape at more different wavelengths, and the high dusts content of LARS 8, which is the third highest in the sample.

The computed f_C(v) show some scatter, but are generally in good agreement with the average LIS absorption profile, supporting that the absorbing system(s) are optically thick. Weak fluorescent emission is apparent at O i* 1305, while little or no fluorescent emission seems to be present at Si ii* 1194.

Globally a weak Lyα emitter, LARS 9 is the strongest absorber in the sample measured in the COS aperture. The Lyα profile is a broad, damped absorption feature, with significant absorption out to ± ∼ 4000 km s⁻¹, as illustrated in Figure 11.

The LIS average metal absorption profile consists of two components. A strong, saturated static or slowly infalling component centered at v ∼ 50 km s⁻¹, and a shallower component centered at ∼−220 km s⁻¹. This combination of no outflow, large velocity range and probably high column density (judging from the depth of the metal lines) efficiently suppresses Lyα escape.

The Si ii lines generally coincide well across the entire velocity range, except for Si ii λ 1260, which is significantly deeper on the blue side before merging with the other lines in the far blue wing. As mentioned above, a deep λ 1260 feature will lead to a fit that overestimates the covering fraction and underestimates column density, which is probably what gives rise to the difference between the absorption profile and computed covering fractions on the blue side of the f_C(v) profile.

It is unlikely that this apparent semi transparency is physical; Si ii 1260 is the strongest of the four lines and thus the least sensitive to changes in column density in the close to optically thick regime. Si ii 1304 would become shallower than the rest, then the 1190 + 1193 doublet. The fact that these three lines coincide so well suggests that some other effect is influencing the blue wing of Si ii 1260, probably a weak contaminating line.

Much of what was written of LARS 9 also applies to LARS 10, although on a more moderate scale. This galaxy is also a weak emitter globally and a strong absorber in the COS aperture, although not as strong as LARS 9. The metal absorption lines, while black at line center and almost static, are not as broad and not quite as strong as those of LARS 9. Instead of a slow inflow, the line is centered at a weakly outflowing velocity, but the line is still black at systemic velocity, all of which efficiently blocks Lyα escape.

Similar to LARS 9, the Si ii lines follow each other except for λ 1260, which is deeper in the blue wing than the rest. Combined with the fact that the transitions at λλ 1193 and 1304 are masked out in the blue wing due to contamination, this introduces some strong uncertainty in the computations of f_C(v) blueward of the main line. However, this should not affect any of the main conclusions of this work, which primarily rest on the properties at and around line center.

LARS 11 is, along with LARS 8, notable for having a strongly outflowing neutral medium while still having a low global Lyα escape fraction. In the COS aperture, Lyα emission is slightly more pronounced, forming a weak P Cygni profile. As for LARS 8, the width of the averaged LIS metal absorption profile is quite high, and the dust content of LARS 11 is the second highest of the sample, a combination that could possibly account for the missing Lyα output.

The Si ii 1190 + 1193 doublet of LARS 11 falls between the detectors at COS and could not be included in the analysis, meaning that the uncertainties on inferred f_C(v) profile are large, and the assumption of this profile coinciding with the averaged absorption profile must remain an assumption. This is further exacerbated by the spectrum being quite noisy and the estimated measuring errors being large. However, the Monte Carlo simulations show that at least the estimate of the center velocity is quite robust, and that the line persistently is completely opaque at line center.

The two Si ii lines present show an inflowing dip at around 250 km s⁻¹. However, for λ 1260 it could well be noise, and for λ 1304 it might be contamination. The feature has not been included in the analysis. This could of course have a significant impact, especially in v_int. However, given the large uncertainties of the inferred values for this galaxy, these new values would most likely be consistent with the currently given ones within 1σ.

None of the three Si iv absorption lines could be observed for LARS 11; two fall redward of the detector range, and one is blended with geocoronal Lyα.

Globally, LARS 12 is another weak emitter, and its output in the COS aperture is modest. It shows something that in Figure 10 could look like a skewed double-peak profile, but which, when viewing a broader range of wavelengths, turns out to be a quite classic, weak P Cygni absorption profile.

The average LIS profile is relatively shallow and wide. Across most of the profile, Si ii λ 1304 seems slightly shallower than the rest and λ 1260 slightly deeper, indicating that the absorbing systems might not be completely opaque. The best fits for f_C(v) also seem to fall a bit below the averaged profile, although the difference is small and full opacity still a good approximation.

The difference in line depth gets stronger on the red side, mainly due to O i* 1305 adding its contribution to the flux of Si ii 1304. This leads to uncertain and probably erratic values for f_C(v) being reached in this region.

LARS 13 is another globally weak Lyα emitter with a clear but modest P Cygni-like emission feature inside the COS aperture. The absorption component of this feature aligns particularly well with the averaged LIS absorption profile as seen on Figure 10. The averaged LIS absorption profile is relatively wide and everywhere well above zero, centered around v ∼ −100 km s⁻¹.

The Si ii absorption profiles again align quite well, except λ 1304, which is somewhat more shallow than the rest. Furthermore, this line and λ 1193 both have a small peak centered at zero velocity that is absent in the two other lines. These two lines are the ones most strongly affected by radiative transfer effects, but it is difficult to say exactly what creates this bump.

The transition at 1193 Å is slightly more shallow than its 1190 Å counterpart, which also indicates that some level of re-emission of light in the absorption troughs is present, although probably not much. The shallower 1304 Å line could be due to refilling the absorption line, or it could be due to the line actually being slightly transparent at this wavelength, since this is the weakest of the four transitions. This is reflected by the calculated values of f_C(v) consistently being slightly higher than the values inferred from the assumption of complete optical thickness. This effect is modest and not likely to differ by more than 5%, but it is difficult to say which cause is more important.

LARS 14 is the second strong outlier in the sample. It is a Green Pea type galaxy (Cardamone et al. 2009), a class of compact, hot galaxies with low neutral gas column density. The Lyα profile of LARS 14 shows strong emission in a double-peaked profile, the central absorption feature of which is remarkably shallow, with a minimum significantly above continuum level, indicating a low H i column density. This is further supported by the Si iv absorption profile being significantly deeper than the averaged LIS metal line, indicating an ionization degree that is unusually high for the sample.

The average LIS metal absorption profile consists of two shallow and narrow components. One is centered at v close to zero and just barely significant, but persistent across all six LIS lines. The other is somewhat stronger but still very shallow, with a minimum residual intensity of around 40%. It is centered at v ∼ 320 km s⁻¹. The central component sees its Si ii profiles lining up very well, although this, given the weakness of the feature and the measurement uncertainties, should be interpreted with caution. However, if this is in fact true, it indicates that the neutral medium at these velocities is optically thick but fragmented. The stronger, outflowing component shows a somewhat deeper feature at λ 1260 Å and a somewhat shallower feature at λ 1304 Å, which is consistent with a medium that is not fully opaque at these wavelengths, as is also apparent from the f_C(v) profile in Figure 10. This should, however, also be interpreted with caution, due to the uncertainties involved.

The uncertainties combined with the two shallow components of the profile also give a high uncertainty in the calculated dynamic velocities: When performing the Monte Carlo simulation, there is a relatively high probability that the uncertainties drawn can line up to shift the weight of the profile from one component to the other.

The Si ii profiles are peculiar in that they show very modest absorption and scattering along the LOS, but still strong fluorescent emission in O i* 1305 and Si ii* 1194. We suggest the explanation that significant scattering and re-emission in a relatively dense neutral medium is happening inside the aperture, but off the LOS to the central sources, as described in Jaskot & Oey (2014). We interpret this as further indication of a clumpy, fragmented neutral medium.

Green Peas have been suggested as candidates for Lyman Continuum leaking galaxies (e.g., Jaskot & Oey 2013; Nakajima & Ouchi 2014) due to their very H i column density. This would, in the case of LARS 14, be supported by the overall low values of f_C(v), with a maximum value of 0.4. However, Heckman et al. (2011) ruled out that this galaxy should leak ionizing radiation along our LOS, and we find no evidence against this claim.