2.1 Flux, luminosity, and equivalent width

Lyα is mainly produced by recombinations in photoionised nebulae, where under standard Case B assumptions 68% of ionising photons absorbed by hydrogen are reprocessed into Lyα in the following radiative cascade (See Dijkstra Reference Dijkstra2014 for an intuitive explanation). For continuously star-forming galaxies, with constant star formation rate (SFR), the Lyα EW is about 80Å. For very young systems, the EW peaks around 300Å (Charlot & Fall Reference Charlot and Fall1993), while very high EWs that exceed 1000 Å may in principle be expected for very low-metallicity, population iii stellar systems (Schaerer Reference Schaerer2003; Raiter, Schaerer, & Fosbury Reference Raiter, Schaerer and Fosbury2010). Naturally if the SFR is declining, the intrinsic Lyα EW may take any value less than these, and it is worth noting that for a simple stellar population (SSP), W _Lyα exceeds 20 Å only during the first 6 Myr (Leitherer et al. Reference Leitherer1999).

Measuring the flux and EW of most emission lines is straightforward. However for Lyα, this is not necessarily so and these quantities may depend upon both methodology and definition. As a resonant transition, both nebular Lyα and continuum radiation with wavelength λ = 1216 Å may be absorbed by Hi. Depending upon the column density, this absorption may reach EWs of several tens of Å, which is a substantial fraction of the nebular flux. Fluxes measured in a given aperture may or may not be reduced by this amount. Narrow-band imaging, for example, needs to be continuum-subtracted and therefore measures the sum of nebular emission and absorption. Spectroscopic observations, on the other hand, may enable the observer to isolate the components and separate nebular emission from ISM absorption, should this be the goal of the measurement. However, even in spectroscopic mode isolating the emission-only flux depends upon the spectral resolution, and the separation will be much easier with high-resolution slit spectrographs than low-resolution survey telescopes.

Not only is Lyα absorbed in the ISM but also, depending on temperature and the properties of their winds, in the atmospheres of stars. For the hottest O stars Lyα EWs may be negligible, but as the population ages or the SFR declines, the dominant source of UV continuum will shift to progressively cooler stars. Valls-Gabaud (Reference Valls-Gabaud1993) showed that Lyα measurements from some local galaxies may be subject to significant uncertainties from stellar absorption, and recent models by Peña-Guerrero & Leitherer (Reference Peña-Guerrero and Leitherer2013) demonstrate that stellar Lyα absorption may reach EWs of −30 Å. For example, the effect of stellar absorption may also vary from O-star-dominated systems where the nebular EW is high and the stellar feature is negligible, to later B-star systems where the reverse is true. The stellar feature may therefore shorten the timescale over which an episode of star formation remains bright in Lyα, and the effect may even be seen on resolved scales within a galaxy.

2.2 Lyα/Balmer ratios and escape fraction

EWs have the advantage that only a short range in wavelength is needed to make the measurement, over which the effects of interstellar dust (reddening curve, total extinction) should have a negligible effect. As discussed above, the evolutionary phase of star formation dominates the intrinsic EW. Lyα/Balmer line ratios, however, may also be used as a measure of the strength of Lyα, and because the intrinsic ratios are limited to a narrow range of values, have other advantages. For example the Hα line (λ = 6564.61 Å) is a well-known, calibrated tracer of current star formation activity in nearby galaxies (e.g. Kennicutt Reference Kennicutt1983). For Case B nebulae at temperatures in the range 5 000–20 000 K and electron density in the range n _e = 10²–10⁴ cm⁻³, the Lyα/Hα ratio ranges between 8.1 and 11.6 (Hummer & Storey Reference Hummer and Storey1987). Thus, deviations from the intrinsic line ratios encode information about the Hi scattering and dust absorption. While different authors do tend to adopt slightly different values, the range of permitted values is relatively narrow. For this review, we will adopt the value of Lyα/Hα = 8.7, which for T = 10⁴ K gas corresponds to n_e ≈ 350 cm⁻³ s, and as a convention can be traced back to at least Hu, Cowie, & McMahon (Reference Hu, Cowie and McMahon1998).

Frequently, we make reference to the Lyα escape fraction (f ^Lyα_esc), in an effort to find a quantity that most closely reflects the combined impact that gas and dust have on suppressing the emission line. We define f ^Lyα_esc as the ratio of the emitted Lyα luminosity to that intrinsically produced, but naturally a number of assumptions can enter our estimates of the intrinsic value. The most robust estimates of f ^Lyα_esc will naturally come from comparing Lyα with other hydrogen recombination lines, where in practice Hα is the strongest and easiest to observe. Assuming that Hα is unobscured, f ^Lyα_esc will simply be the observed Lyα/Hα ratio divided by its intrinsic case B value (8.7 as mentioned above). Of course, Hα can also be significantly absorbed by dust, and in local ‘normal’ galaxies suffers about 1 mag of extinction on average (Kennicutt & Kent Reference Kennicutt and Kent1983). Obviously, redder hydrogen lines would be better as they suffer less extinction but also become systematically weaker in flux, and become harder to observe in the infrared. Radio recombination lines would be ideal, suffering no extinction at all, but are even more challenging to observe beyond the very local universe. Thus often the best route to f ^Lyα_esc is to dust-correct Hα using the Hβ line, which should recover all the star formation down to moderate optical depths (Hayes et al. Reference Hayes, Östlin, Mas-Hesse, Kunth, Leitherer and Petrosian2005; Atek et al. Reference Atek, Kunth, Schaerer, Hayes, Deharveng, Östlin and Mas-Hesse2009a). However, when Hα becomes very optically thick, in very dusty star-forming galaxies (e.g. Martin et al. Reference Martin, Dijkstra, Henry, Soto, Danforth and Wong2015), even dust-corrected Hα traces only a small fraction of the true ionised gas, making the inferred escape fraction prone to strong biases. In such instances, one may do better by comparing the Lyα-derived SFR with that estimated from dust emission in the FIR, under the assumption that systematic errors on the SFR calibrations are smaller than the fraction of Hα that is recoverable. In the highest optical depth regimes, this is probably true.

3.1 The first Lyα spectra of active galactic nuclei

Even before IUE, some extragalactic objects were observed with sounding rocket experiments, which were able to launch small UV telescopes above enough of the atmosphere to observe in the UV. Davidsen, Hartig, & Fastie (Reference Davidsen, Hartig and Fastie1977) targeted the first known quasar, 3C 273, with a UV spectro-photometer onboard a sounding rocket and provided the first measurement of Lyα and Balmer emission lines from any astrophysical body. Combining the measured Lyα flux with optical measurements revealed a Lyα/Hβ ratio of 4, a measurement that then sat in stark contrast to the value of ≈ 40 that was expected for nebular recombinations and a Lyα enhancement from collisions. In 1977, Davidsen et al. discussed the order-of-magnitude Lyα deficit with the following statement:

While this review is not focused upon QSOs, replacing ‘QSO’ with ‘galaxy’ in this statement comes precisely to the point. Today Lyα observations play a major role in understanding the interstellar and circumgalactic media of galaxies at effectively all redshifts.

The 1978 launch of IUE opened up the rest-frame UV to systematic study. IUE had a single 20 arcsec × 10 arcsec entrance aperture and Short and Long-Wavelength Prime channels (SWP and LWP, respectively) that could provide R ≈ 250 spectroscopy between 1150 and 3000 Å. First observations were again turned to QSOs and various Seyfert galaxies, and immediately showed Lyα to be weaker than expected for recombination theory, with Lyα/Hβ and Hα/Hβ values almost never falling along characteristic reddening curves, and Lyα always being preferentially suppressed (Oke & Zimmerman Reference Oke and Zimmerman1979; Wu, Boggess, & Gull Reference Wu, Boggess and Gull1980; Lacy et al. Reference Lacy1982). Interpretations varied, with suggestions that multiple scatterings of Lyα could be responsible, that broad-line regions experience a wide variation in their extinction laws, ionisation/excitation from already-excited states, and that there may be no representative intrinsic spectral shape for Seyfert galaxies (Wu et al. Reference Wu, Boggess and Gull1980). Indeed, Lacy et al. (Reference Lacy1982) concluded that the low observed Lyα fluxes were likely the result of a combination of reddening, high densities, and high Hi optical depth. All these effects conspire in the same direction but as no single dominant quantities could be identified, it started to become clear that Lyα transfer in true astrophysical objects is a complicated multi-parametric process.

3.2 Star-forming galaxies

When IUE was first turned to star-forming galaxies, the results directly mirrored those obtained in both nearby QSOs and also those being reported from high-z blind narrow-band and spectroscopic surveys (see Pritchet Reference Pritchet1994, for a review, and the forthcoming PASA review in this series by S. Malhotra): Lyα was either absent or unexpectedly weak in all galaxies.

With the intent of studying the analogues of primeval galaxies at high-z, Meier & Terlevich (Reference Meier and Terlevich1981) found Lyα in emission from just one of three nearby Hii-selected galaxies, and in that single case with a flux well below that expected for the nebular dust content (Figure 1). This led them to conclude that under normal conditions, Lyα emission would be a rare phenomenon. It was determined that if a normal prescription for dust attenuation were used to explain the Lyα/Hβ ratio, this would greatly over-predict the Hα/Hβ ratio compared to observation. Similar conclusions were reached by Hartmann, Huchra, & Geller (Reference Hartmann, Huchra and Geller1984), who furthered the discussion, showing how a mixed medium of Hi and dust could preferentially suppress Lyα, and that static Hi columns of density above 10¹⁹ cm⁻² would be needed to reconcile the line ratios. Further, they raised the issue that, where Hi 21-cm data are available, most blue compact galaxies (BCGs) are found to sit inside extended Hi halos of sufficient column density to reproduce the measured fluxes. All signs pointed towards the fact Lyα emission would not be the good observational beacon to identify primeval galaxies in the early universe that Partridge & Peebles (Reference Partridge and Peebles1967) had predicted.

Figure 1. The first Lyα spectra of Hii galaxies. Main transitions and cosmic ray hits are labelled. The upper panel shows Mrk 701 (C 0842+163), with a redshift of z = 0.0522 and a metallicity of Z ≈ 0.4 Z _⊙. At the wavelength of λ = 1278 Å, the galaxy shows only a Lyα absorption feature. The lower panel shows a very different object – C 1543+0907, with a redshift of z = 0.0366 and metallicity of 1/10 the solar value. At the wavelength of λ = 1260 Å, the bright high-contrast Lyα emission line is obvious. Reproduced by permission of the AAS, from Meier & Terlevich (Reference Meier and Terlevich1981).

While the influence of Hi on Lyα visibility was starting to be seen empirically, some correlation with the dust abundance should still be expected, albeit with a large spread. After subsequent data acquisition, the anticorrelation between Lyα EW and gas-phase metallicity (Z) was discovered (Hartmann et al. Reference Hartmann, Huchra, Geller, O’Brien and Wilson1988; Calzetti & Kinney Reference Calzetti and Kinney1992; Terlevich et al. Reference Terlevich, Diaz, Terlevich and Vargas1993; Charlot & Fall Reference Charlot and Fall1993), seemingly confirming the prediction. However the spread on the relation was particularly large, and it was clear that something beyond variations in the extinction law (e.g. Valls-Gabaud Reference Valls-Gabaud1993) were behind the wide range of line ratios, with ISM geometry and holes being the most often-cited scenarios.

Giavalisco, Koratkar, & Calzetti (Reference Giavalisco, Koratkar and Calzetti1996) performed a full reanalysis of the IUE archival data, resulting in several changes. First, the IUE data reduction software reached a higher level of maturity, and new spectral extraction and cosmic-ray removal tools were implemented: this reduced the measured W _Lyα significantly in about half the sample, and completely removed the weak Lyα feature that had been reported in some galaxies. Secondly, these authors performed proper spatial matching between the IUE and apertures used for optical line spectroscopy. When the homogenised reanalysis was complete, both the W _Lyα and Lyα/Balmer line ratios showed no significant correlation with nebular dust attenuation or the UV continuum colour (β), and only a weak but significant correlation with nebular oxygen abundance. Results concerning the Lyα/Hα line ratios can be seen Figure 2.

Figure 2. The observed ratio of Lyα/Hα fluxes, shown on a linear scale against nebular metallicity and the Hα/Hβ line ratio for the full sample of star-forming galaxies observed with the IUE, and reprocessed by Giavalisco et al. (Reference Giavalisco, Koratkar and Calzetti1996). Negative Lyα/Hα ratios are the result of the Lyα feature being dominated by ISM absorption. The reference coding is MT81=Meier & Terlevich (Reference Meier and Terlevich1981), H84=Hartmann et al. (Reference Hartmann, Huchra and Geller1984), H88=Hartmann et al. (Reference Hartmann, Huchra, Geller, O’Brien and Wilson1988), C94=Calzetti, Kinney, & Storchi-Bergmann (Reference Calzetti, Kinney and Storchi-Bergmann1994), CK92=Calzetti & Kinney (Reference Calzetti and Kinney1992), D86=Deharveng, Joubert, & Kunth (Reference Deharveng, Joubert, Kunth, Kunth, Thuan, Thanh Van, Lequeux and Audouze (SAO/NASA ADS)1986), T93=Terlevich et al. (Reference Terlevich, Diaz, Terlevich and Vargas1993). The dotted black line shows the effect of the Calzetti et al. (Reference Calzetti, Armus, Bohlin, Kinney, Koornneef and Storchi-Bergmann2000) dust attenuation curve, assuming intrinsic ratios of Lyα/Hα=8.7 and Hα/Hβ=2.86 and no scattering. Most of the galaxies lie far below this curve.

Regarding these correlations, it is not clear why Lyα throughput should be more strongly correlated with oxygen abundance than nebular attenuation. Metals alone do not absorb Lyα, which can only happen by interaction with dust grains. Thus, if the W _Lyα–Z anti-correlation results from a positive correlation between metal and dust abundance, then a tighter correlation between Lyα/Hα and E _{B − V} would be expected. This correlation is completely absent. Furthermore, when the dust reddening measured from the Balmer decrement is used to correct the observed Lyα for extinction, the Lyα flux does not reach the expected case B recombination value in a single galaxy in the IUE sample. This demonstrates that either some preferential attenuation of Lyα must be at play in every galaxy, or/and that dust attenuation laws, when applied as a screen of absorbing material, are not representative.

When interpreting these analyses of the IUE samples, it is important to keep in mind the selection functions by which the individual samples were established. When effective telescope areas are small and exposure times need to be long, the result is small samples. Some of the blue compact dwarf (BCD) and Hii galaxy studies were designed to study the analogues to ‘primeval’ galaxies that are undergoing their first phase of star formation, and the samples did not include galaxies with strong star-bursting nuclei (Hartmann et al. Reference Hartmann, Huchra and Geller1984). Yet the primeval stages of galaxy formation, prior to the initial dust and metal production, are expected to be short-lived because the first generations of supernovae will enrich the local ISM on timescales of just a few Myr. Consequently, the bulk of the galaxy population we can observe in the high-redshift universe is likely to be significantly more metal-enriched than that of primeval galaxies (e.g. Pettini et al. Reference Pettini, Rix, Steidel, Adelberger, Hunt and Shapley2002; Shapley et al. Reference Shapley, Steidel, Pettini and Adelberger2003), unless observations catch galaxies in very narrow time window. Thus while providing very interesting astrophysical laboratories, the samples are biased and not necessarily in a direction tuned to the real importance of Lyα: probing the faint population of normal galaxies at z > 2. The IUE samples contain few galaxies that can be considered the local analogues of high-z Lyman Break Galaxies (LBG), LAEs, or primeval galaxy building blocks.

4.1 HST finds deep Lyα absorption

GHRS observations of i Zw 18 revealed a profile showing only damped Lyα absorption and no hint of emission (Kunth et al. Reference Kunth, Lequeux, Sargent and Viallefond1994). The Lorentzian wings of the absorption profile can be traced out to at least 6000 km s⁻¹ (Figure 3; Mas-Hesse et al. Reference Mas-Hesse, Kunth, Tenorio-Tagle, Leitherer, Terlevich and Terlevich2003), implying a Hi column density above several 10²¹ cm⁻². Furthermore, measurements of the low ionisation stage (LIS) metal lines that form in the neutral ISM (e.g. Oi λ1302, Siii λ1304) show that the Hi gas is static with respect to the Hii regions, and at the measured column density Lyα radiation at line-centre will have to traverse 10⁷ optical depths in order to escape if the Hi is homogeneous.

Figure 3. The Lyα spectra of first dwarf galaxies to be observed in the far UV with HST. The left panel shows i Zw 18, first observed with GHRS by Kunth et al. (Reference Kunth, Lequeux, Sargent and Viallefond1994). The geocoronal Lyα line has been masked out and is shown by a line set to zero flux at velocities of −500 to −2000 km s⁻¹. No intrinsic Lyα emission is seen, and damping wings are visible that extend to at least 6 000 km s⁻¹. Figure is taken from Mas-Hesse et al. (Reference Mas-Hesse, Kunth, Tenorio-Tagle, Leitherer, Terlevich and Terlevich2003). The right panel shows Haro 2, first observed by Lequeux et al. (Reference Lequeux, Kunth, Mas-Hesse and Sargent1995). Haro 2 has a much higher dust abundance than i Zw 18, but shows a strong Lyα emission line. Furthermore, the line profile is P Cygni-like, which indicates the intrinsically produced frequencies have been redistributed by scattering in an expanding neutral medium. Data are replotted from Mas-Hesse et al. (Reference Mas-Hesse, Kunth, Tenorio-Tagle, Leitherer, Terlevich and Terlevich2003).

Subsequent observations of i Zw 18 with the Space Telescope Imaging Spectrograph (STIS) enabled detailed, spatially resolved, and empirically well-constrained studies of Lyα radiative transport, which indeed shows that this profile can be reproduced, including spatial variation in the damping wings, using column densities of N _HI ~ 3 × 10²¹ cm⁻², and E _{B −V}= 0.05 (Atek, Schaerer, & Kunth Reference Atek, Schaerer and Kunth2009). This is fully consistent with the directly observed values on both N _Hi and E _{B −V}. Moreover, the same transport simulations predict that for static gas with this Hi column density, almost all the Lyα radiation is absorbed by the small amount of available dust (escape fraction the order of 10⁻⁴ to 10⁻³), as many scattering events increase the probability of absorption.

With the HST servicing mission 4, the Cosmic Origins Spectrograph (COS) was installed on HST. COS has a similarly sized entrance window to GHRS and similar grating specifications, but can observe much larger wavelength range in a single observation, and is many times more sensitive. Recently obtained COS observations of i Zw 18 enable us to go much deeper than was previously possible, and a COS/G130M exposure of 29 ks actually does reveal a small bump of Lyα emission, hidden at the bottom of the absorption profile (Figure 4; see also Lebouteiller et al. Reference Lebouteiller, Heap, Hubeny and Kunth2013; James et al. Reference James, Aloisi, Heckman, Sohn and Wolfe2014). The Lyα emission feature lies in the centre of the absorption trough, and the flux amounts to just ≈ 4 × 10⁻¹⁶ erg s⁻¹ cm⁻². Recalling at this point that HST spectrographs measure only very local properties, we would need to make a large aperture correction to estimate the total Lyα output. If we assume an exponential Lyα surface brightness profile we would need a scale length of 80 arcsec in order for 100% of the Lyα to be emitted (integrating to infinity). While this is not ruled out – Hi is extends over at least a square arcmin (van Zee et al. Reference van Zee, Westpfahl, Haynes and Salzer1998) – this scale length is 40 times the UV effective radius, which is an extreme extension of Lyα compared with other nearby objects (Section 5). The alternative is that a large fraction of the Lyα photons are absorbed by dust after numerous scattering events, as suggested by Atek, Schaerer, & Kunth (Reference Atek, Schaerer and Kunth2009).

Figure 4. Very deep Lyα spectrum i Zw 18 obtained with the Cosmic Origins Spectrograph. While previously referred to as a Lyα absorbing galaxy, i Zw 18 shows a very small but significant bump of Lyα emission. This bump is redshifted from the expected systemic velocity (the dot–dashed vertical line) derived from the Hα line. The overlaid red line shows the profile of Lyα transferring through a static shell of atomic hydrogen with a column density of log(N _HI/cm⁻²) = 21.1, taken from the Schaerer et al. (Reference Schaerer, Hayes, Verhamme and Teyssier2011) grid of transfer models. The strong emission feature is the geocoronal Lyα emission line. The inset shows the wider spectral profile.

In i Zw 18, the Lyα bump is also offset from the systemic velocity (measured from Hα) by 350 km s⁻¹ while the neutral gas shows bulk velocities of −10 km s⁻¹ (measured from Siii). The bulk motion is insufficient by an order of magnitude to kinematically redshift the Lyα by such a velocity, so other physical processes must be at play. The red line in Figure 4 shows a radiative transport model produced by the grid of Schaerer et al. (Reference Schaerer, Hayes, Verhamme and Teyssier2011), for a completely static and dust-free Hi shell with a column density log(N _HI/cm⁻²) = 21.1, which is within 0.25 dex of previous estimates based upon modelling just the absorption feature. A plausible interpretation is that the bump is the red half of a double-peaked profile, which results from wing scattering events that shift photons in frequency many Doppler widths into the redistribution profile. If so, and the scattering medium is completely static, there would be a corresponding blue peak at –350 km s⁻¹; this however would be hidden below the geocoronal Lyα line, that swamps any intrinsic emission.

Observations of similar dwarf galaxies show the deep Lyα absorption seen in i Zw 18 is not unique. Two of the other most metal-deficient galaxies known, SBS 0335–052 and Tol 65 (Thuan & Izotov Reference Thuan and Izotov1997), have metallicities just a factor of 2 higher than i Zw 18. Both also show broad Lyα absorption profiles with EWs of −20 to −30 Å, no hints of Lyα emission, and clear damping wings that imply Hi column densities above 2 × 10²¹ cm⁻². Similar deep absorption is visible in the COS spectrum of low-metallicity dwarf SBS 1415+437 (James et al. Reference James, Aloisi, Heckman, Sohn and Wolfe2014; 12 + log(O/H) ≈ 7.6), which also shows a similarly redshifted bump of Lyα in emission. These galaxies are very rare in the local universe, and one may question whether far-reaching conclusions may be drawn from them. Nevertheless, such objects may become more abundant at higher redshifts, and any complete theoretical picture of Lyα must also include them.

4.2 Galaxy winds and atomic gas kinematics

In contrast to the strongly absorbing dwarfs, the second BCG with a published HST Lyα spectrum, Haro 2, was found to emit a strong Lyα line with W _Lyα ≈ 7 Å (emission part only, Lequeux et al. Reference Lequeux, Kunth, Mas-Hesse and Sargent1995, right panel of Figure 3). This is particularly curious because Haro 2 is an order of magnitude more dust- and metal-rich than i Zw 18. Comparing with the absorbing BCDs discussed in Section 4.1, the total column density of hydrogen along the line-of-sight is roughly the same, but with two important differences. First, much less of that hydrogen column density is contributed by the neutral phase, although the measured column of N _Hi ≳ 10²⁰ cm⁻² would still be sufficient to cause damped absorption. Secondly, the absorption centroid of Lyα is blueshifted by ≈ 200 km s⁻¹ relative to the systemic frame of rest of the galaxy. The result of this first resolved observation of a Lyα emission line is an asymmetric profile that comprises a blue absorption component and a red emission peak, similar to the P Cygni profile. Even though the wing of the absorption profile is quite extended, the velocity offset is sufficient to shift the neutral medium partially out of resonance with Lyα, and enable some of the Lyα radiation to escape.

The LIS lines intrinsic to Haro 2 were also found to be blueshifted with respect to the systemic velocity, and by the same velocity as measured from the Lyα absorption (≈ 200 km s⁻¹). The peak of the Lyα emission, however, is instead redshifted by 350 km s⁻¹, or roughly twice the blueshift of the neutral medium. This led Lequeux et al. (Reference Lequeux, Kunth, Mas-Hesse and Sargent1995) to suggest that much of the Lyα is able to avoid Hi absorption in Haro 2 because it does not see the atomic gas as static, and the redshifted Lyα emission supports a picture in which the Lyα that is emitted is ‘backscattered’ from a receding shell of Hi gas (Verhamme et al. Reference Verhamme, Schaerer, Atek and Tapken2008). Furthermore in Haro 2, diffuse soft X-ray emission covers and extends beyond the UV-bright, star-forming regions, which is produced by the mechanical energy released by the star formation episode (Otí-Floranes et al. Reference Otí-Floranes, Mas-Hesse, Jiménez-Bailón, Schaerer, Hayes, Östlin, Atek and Kunth2012). This X-ray emission is spatially consistent with an extension of the Lyα emission in the 2D spectral image. In contrast, i Zw 18, which shows only Lyα absorption and static low-ionisation absorption lines, is undetected at soft X-ray energies (Ott, Walter, & Brinks Reference Ott, Walter and Brinks2005). The conventional picture for galaxy outflows is that cold gas is accelerated by expanding hotter gas (e.g. Strickland et al. Reference Strickland, Heckman, Colbert, Hoopes and Weaver2004), which supports scenario where feedback-driven outflows promote the emission of Lyα.

This early picture easily generalised in a sample of eight local BCGs observed by Kunth et al. (Reference Kunth, Mas-Hesse, Terlevich, Terlevich, Lequeux and Fall1998), four of which show net Lyα in emission and four in absorption. For those with net absorption, their Oi λ1302 and Siii λ1304 absorption lines lie close to the systemic velocity (within 25 km s⁻¹), while the other four show outflowing gas with centroid velocities shifted by 60–180 km s⁻¹ (see also Leitherer et al. Reference Leitherer, Chandar, Tremonti, Wofford and Schaerer2013). This correlation does not necessarily imply a causal relationship and the fact that Lyα is seen to be locally emitted where outflows are strong could also be explained also by orientation: transport models show more Lyα to be emitted perpendicular to galaxy disks purely because of the Hi distribution (Verhamme et al. Reference Verhamme, Dubois, Blaizot, Garel, Bacon, Devriendt, Guiderdoni and Slyz2012; Laursen, Duval, & Östlin Reference Laursen, Duval and Östlin2013), and winds are also stronger in the polar direction because the pressure is also lower (e.g. Bland & Tully Reference Bland and Tully1988; Veilleux & Rupke Reference Veilleux and Rupke2002). More compelling evidence for a causal association comes from the fact that all the Lyα emission lines in the sample show P Cygni-like asymmetric profiles, indicating that photons are interacting directly in the outflowing medium. For this, it is much harder to argue for a non-causal relation.

Galaxies in the Kunth et al. (Reference Kunth, Mas-Hesse, Terlevich, Terlevich, Lequeux and Fall1998) sample were originally chosen to span a range of metallicities and dust contents, but both of these quantities were found to be secondary in governing Lyα emission/absorption when compared to the presence/absence of outflowing neutral gas.

COS Lyα observations of local galaxies are ongoing, but already the instrument has far outdone the GHRS in terms of numbers. Using larger samples of both FUV-selected (Heckman et al. Reference Heckman2011) and Hα-selected objects (Wofford, Leitherer, & Salzer Reference Wofford, Leitherer and Salzer2013), this picture of kinematic regulation easily has strengthened. For galaxies in the ‘Lyman-break analogue’ (LBA) samples of Heckman et al. (Reference Heckman2011), P Cygni emission is ubiquitous while for the Hα-galaxies, redshifted Lyα peaks and LIS lines blueshifted by around 100 km s⁻¹ are found among the emitters, while absorption lines (including Lyα) consistent with zero velocity shift at 68% confidence are exhibited by the absorbers.

Similar results are found among the Lyman alpha Reference Sample (LARS; Section 5.2; Rivera-Thorsen et al. Reference Rivera-Thorsen2015), which are summarised in the lower panels of Figure 5. The left-most panel shows an example where the ISM in front of the brightest nuclear star cluster (where the COS aperture is positioned) is static, and from where broad damped Lyα absorption is also observed. From other regions of the galaxy, however, Lyα emission is recovered and the galaxy becomes a weak LAE (W _Lyα ≈ 10 Å) in apertures that encompass the galaxy. The central panel shows an example where the neutral ISM is outflowing along the line-of-sight by around 250 km s⁻¹, the Hi absorption is similarly blueshifted, and a weak Lyα emission feature is visible within the pointing of the COS. The right panel instead shows a galaxy where the atomic gas is outflowing at higher velocity still, and a very bright Lyα emission line is visible with W _Lyα ≈ 40 Å (80 Å when including extended emission).

Figure 5. Example images and Lyα spectra of local galaxies. Upper panels show colour composite images, encoding Hα in red, FUV continuum in green, and Lyα in blue. Dotted white lines indicate the size and position of the COS aperture. Lower panels show the corresponding spectra around Lyα (dark blue) and Siiiλ1260 Å (red). Objects are selected to illustrate the variety of Lyα and LIS profiles. Note the different velocity scales, which extend to ± 7000 km s⁻¹ in the left-most frame but just ± 1500 km s⁻¹ in the other two. The left object, LARS 09, shows a completely absorbed Siii profile that is centred at zero velocity. This absorbing gas removes 300 km s⁻¹ (Full Width at Half Maximum [FWHM]) in the Siii line but completely damps the Lyα line, resulting in a FWHM that is 20 times broader. The centre plot, LARS 08, shows a similarly broad and saturated Siii profile, but one that is offset in velocity by ~ −250 km s⁻¹. The blue wing of the Lyα absorption profile shows a similar shape and is offset in a similar way to the left panel, but in this case a redshifted Lyα line is emitted. The right panel shows an example of a strong Lyα-emitter, in which the Siii absorption is blueshifted by ≈ 350 km s⁻¹, is far from saturated, and a very bright Lyα emission line is seen.

Obviously, if all neutral gas can be cleared from zero velocity, then Lyα should escape unhindered. For the column densities probed using the Siii lines, this appears to be the case in the right panel of Figure 5, where only more tenuous gas appears to remain at Δv ≈ 0 to absorb Lyα. Yet removal of cold gas from Δv ≈ 0 not a requirement for Lyα emission, and Rivera-Thorsen et al. (Reference Rivera-Thorsen2015) present several examples of galaxies that show strong Siii absorption at Δv = 0, but also significant Lyα emission (their Figure 8). However, while there is clearly gas that does not have a velocity shift, the centroids of the absorption profiles are offset – usually by ≲ −50 km s⁻¹ – which demonstrates that there is fast-moving gas that can Doppler shift Lyα out of resonance with the static material. Additional support for this comes from the fact that the peak of the Lyα profile is redshifted in every case.

Figure 6 summarises the situation, by showing the average velocity shift of low-ionisation absorption lines compared with W _Lyα measured in the same aperture (diameter of 1.9 and 2.5 arcsec with GHRS and COS, respectively). Outflow velocities span a range from −250 to + 50 km s⁻¹ (net inflow), with a median value of −50 km s⁻¹. No galaxy with low-ionisation gas moving in the velocity range −50 to + 50 km s⁻¹ shows significant Lyα emission: all but two of these objects show either net absorption or emit Lyα with a total EW below 3 Å. However, galaxies with faster outflowing gas show a wide range of W _Lyα, which reaches up to 40 Å, with an average near 10 Å. We can state with confidence that at least on small scales Lyα emission is correlated with feedback of sufficient magnitude, that acts to accelerate the highest density neutral ISM along the line of sight. Note, however, that this does not mean it is the case that all galaxies with a strong outflow are LAEs – clearly there are galaxies with Δv ≈ −150 km s⁻¹ and W _Lyα below 2 Å. Furthermore, while the correlation exists, the causal mechanism by which feedback affects the transport has not necessarily been established. For example, whether feedback is simply shifting the Lyα out of resonance by scattering from bulk-flowing gas, or the instigation of fluid instabilities that disrupt the ISM.

Figure 6. The relationship between W _Lyα and outflow velocity of the atomic gas. All measurements are made within spectroscopic apertures with HST/GHRS and COS. Where the Lyα absorption component dominates over emission, W _Lyα is set to zero. Δv is calculated from the average of velocity centroids of all observed LIS lines compared with the intrinsic velocity of the nebulae measured from optical emission lines. A cut of Δv = −50 km s⁻¹ divides the sample in half by Δv, segregating galaxies with fast outflows from those with weakly outflowing, static, or inflowing neutral media. Grey points with errorbars represent the average and standard deviation of galaxies of these two sub-samples.

4.3 Dissecting the neutral medium in detail

In addition to gas kinematics, the plethora of resonance absorption lines in the UV also provide a proxy for the fractional covering of the cold Hi medium (e.g. Savage & Sembach Reference Savage and Sembach1996; Pettini et al. Reference Pettini, Rix, Steidel, Adelberger, Hunt and Shapley2002). The principle is simple: strong resonance transitions are assumed to be saturated at normal metallicities and column densities, and thus if the absorbing line does not drop to the level of zero intensity then the observation hints that there may be multiple clouds inside the spectroscopic aperture that do not fully cover the stellar sources of continuum radiation that lie beneath. Thus if there are direct sightlines between the observer and the ionised regions, Lyα may escape unimpeded, and importantly, without frequency shift. Indeed, if scattering could be completely mitigated and dust confined to the cold gas phase, f ^Lyα_esc should be at least 1 −f _c, where f _c is the Hi covering fraction.

These methods have been used to place indirect limits on the escape of ionising radiation from starburst galaxies (Grimes et al. Reference Grimes2009; Heckman et al. Reference Heckman2011), and have recently been verified by direct observations in the ionising continuum with HST/COS (Borthakur et al. Reference Borthakur, Heckman, Leitherer and Overzier2014). Similar tests, verified against Lyα emission, have been conducted at high-z (e.g. Jones et al. Reference Jones, Ellis, Schenker and Stark2013) and low-z COS studies focussed on Lyα recently been presented. For the majority of local UV-selected galaxies, the depth of the normally saturated Siii lines indicates a covering fraction close to unity. However, there is some deviation from this: a weak trend is seen for galaxies with f ^Lyα_esc above 0.1 to be drawn from systems with Siiiλ1260 Å absorption lines that are not saturated (Rivera-Thorsen et al. Reference Rivera-Thorsen2015). Very well exposed continuum observations are needed to solve for covering fraction, but solutions include the possibility of f _c < 1 for the thickest neutral gas columns in Lyα-emitting galaxies but not in the case of absorbing systems.

Many of the resonant UV transitions, including the Siii discussed above, have an associated fluorescent transition at longer wavelength, denoted with a * (e.g. Cii*), that provide additional diagnostics of the atomic medium (Prochaska, Kasen, & Rubin Reference Prochaska, Kasen and Rubin2011; Rubin et al. Reference Rubin, Prochaska, Ménard, Murray, Kasen, Koo and Phillips2011; Jaskot & Oey Reference Jaskot and Oey2014; Scarlata & Panagia Reference Scarlata and Panagia2015). Since the absorption lines are resonant, they may be partially filled by scattered radiation, similarly to Lyα. However, the fluorescent transition associated with each line has a roughly similar Einstein A coefficient to the resonant de-excitation, which implies that roughly half of absorbed photons should be emitted in the longer wavelength * line at every absorption event. In a symmetric, energy-conserving system without losses, absorption along the sightline must be balanced by isotropic fluorescent emission.

Jaskot & Oey (Reference Jaskot and Oey2014) recently presented spectra of two particularly interesting bright LAEs, with W _Lyα between 70 and 150 Å. Both of these objects are well detected in the stellar continuum, but absorption lines of Cii λ1334 and Siii λ1260 Å lines are barely visible. However, the fluorescent counterpart of each transition is clearly seen in emission, suggesting that the ISM is indeed partly covered or shows a low Hi column density in these galaxies. A similar spectrum is that of LARS 14, illustrated in the lower-right panel of 5 (Rivera-Thorsen et al. Reference Rivera-Thorsen2015), which shows a bright Lyα line with a blue peak, incomplete Siii absorption, and a fluorescent emission line (seen at relative velocity of +1 200 km s⁻¹).

Further information may be inferred from the profiles of Lyα. While Lyα and Siii λ1260 have cross sections of the same order of magnitude and become optically think at similar column density, the metal abundances imply that Lyα may be absorbed by gas that is not visible to metal absorption. Particularly, in the galaxies of Jaskot & Oey (Reference Jaskot and Oey2014) and LARS 14, the Lyα lines do not resemble the strongly asymmetric absorption+emission of P Cygni profiles that occur at high column densities of completely covered gas. Instead, they show double-peaked profiles with narrow absorption at Δv = 0. This indicates that there must be absorbing Hi that is not Doppler shifted, but that is also not of sufficient column density and/or metallicity to be seen in metal absorption lines. This implies N _Hi between 10¹⁵ and 10¹⁸ cm⁻² for normal ranges of metal abundance.

5.1 Resolved analyses: A case study of Haro 11

Local luminous BCG Haro 11 is an LBA (e.g. Grimes et al. Reference Grimes2007), and emits Lyα with a total EW of 15 Å. As shown in Figure 7, it comprises three main star-forming knots (labelled in the FUV image), all of which are bright in the UV and Hα, but only one of these condensations emits Lyα (Hayes et al. Reference Hayes, Östlin, Atek, Kunth, Mas-Hesse, Leitherer, Jiménez-Bailón and Adamo2007). Following this breakdown of the galaxy, knots A and B (the west-most two) are by far the brightest in Hα and must produce the bulk of the Lyα radiation, but both absorb Lyα. In contrast, it is only the single easterly knot C, which is the faintest of the three in Hα but brightest in the UV, that locally emits its Lyα. Haro 11 also emits a halo of Lyα emission, centred around knot C, which can most easily be explained if Lyα is re-radiated after scattering in the surrounding neutral gas. Diffuse emission also surrounds the two easterly knots, but at the positions of the clusters absorption outweighs the emission, giving a negative overall flux. While the Lyα surface brightness of the halo is low, it is also very much larger than the UV continuum-bright regions, and in total contributes ≈ 90% of the total Lyα flux (Hayes et al. Reference Hayes, Östlin, Atek, Kunth, Mas-Hesse, Leitherer, Jiménez-Bailón and Adamo2007). Results inferred from small-aperture spectroscopic observations will be a strong function of both the size and placement of the aperture.

Figure 7. HST/ACS imaging of luminous blue compact galaxy Haro 11 (Kunth et al. Reference Kunth, Leitherer, Mas-Hesse, Östlin and Petrosian2003; Hayes et al. Reference Hayes, Östlin, Atek, Kunth, Mas-Hesse, Leitherer, Jiménez-Bailón and Adamo2007). Upper Left: far UV continuum; Upper Right: continuum-subtracted Hα; Lower Left: continuum-subtracted Lyα. The Lower Right panel shows a composite of the other three images, with Hα in red, FUV in green, and Lyα in blue. The physical scale in kpc is shown on the axes. In the FUV image, the three main star-forming condensations (knots A, B, and C) are labelled in the nomenclature of Vader et al. (Reference Vader, Frogel, Terndrup and Heisler1993); the Lyα-emitting clump with no detection at any other wavelength (Knot D in Kunth et al. Reference Kunth, Leitherer, Mas-Hesse, Östlin and Petrosian2003) is labelled in the Lyα image.

In Haro 11, Lyα produced in knots A and B may still be emitted, and all the observation can say is that more radiation from the stellar continuum is absorbed locally than the sum of directly emitted Lyα and any Lyα scattered into the line of sight. Thus for a given pixel, we still cannot say whether Lyα is scattered and absorbed by dust locally, or whether it propagates some kpc and contributes to the halo emission. Remarkably in this three-region decomposition of Haro 11, is that the strongly emitting knot C is also the dustiest, and shows E _{B −V} ≈ 0.4 magnitude, while knot A in particular is far less extinguished (Atek et al. Reference Atek, Kunth, Hayes, Östlin and Mas-Hesse2008). Under the simplistic assumption that dust plays a dominant role in regulating Lyα visibility, this would be unexpected, although results from these Lyα absorbing knots are reminiscent of the dwarf galaxies discussed in Section 4.

Similar phenomena were noted throughout the sample of nearby galaxies first observed with ACS. Specifically ESO 338–IG04 was also found to exhibit a diffuse Lyα halo that dominates the Lyα output (Hayes et al. Reference Hayes, Östlin, Mas-Hesse, Kunth, Leitherer and Petrosian2005), and in a small sample of six local starbursts, global Lyα emission is invariably associated with large-scale halo emission (Atek et al. Reference Atek, Kunth, Hayes, Östlin and Mas-Hesse2008; Östlin et al. Reference Östlin, Hayes, Kunth, Mas-Hesse, Leitherer, Petrosian and Atek2009).

Nebulae produce Lyα radiation intrinsically at an intensity of 8.7 times that of Hα (Section 2), but when we contrast the local surface brightnesses of the two lines a very wide range of line ratios is found. These are illustrated in Figure 8, which contrasts the Hα and Lyα surface brightness in Haro 11 pixel-by-pixel. We now discuss various regions of the diagram:

1. (a) Lyα can be emitted with fluxes similar to those expected from recombination. In Figure 8 above log Lyα surface brightness of ≈ −12.8, a region of proportionality is seen between Lyα and Hα that falls right on top of the expectation value for Case B. These pixels correspond to knot C in Figure 7. Since this region of points is well defined and rather narrow, it can most easily be understood as Lyα photons leaving the galaxy with little interaction with surrounding Hi. Lyα photons emitted in these regions were most likely produced here.

2. (b) Lyα can be emitted with fluxes below those expected from recombination. The preponderance of points in Figure 8 lie below the case B line. This can be the result of two factors: Lyα can be absorbed by dust, decreasing the Lyα/Hα ratio (just as the Balmer decrement increases with dust in nebulae), or Lyα can be scattered out of the line of sight by Hi, i.e. dust and Hi scattering act to move points down from the red line. Recall, also, that in this logarithmic plot Lyα absorption cannot be visualised and more pixels are to be found at negative values of Lyα.

3. (c) Lyα can be emitted with fluxes above those expected from recombination. Toward the upper left region of the diagram, Lyα/Hα exceeds the value of 8.7 expected for Case B. In this example, some pixels are 10 times brighter in Lyα than expected. This happens only at lower Hα surface brightness, and here Lyα is spatially redistributed from elsewhere and emitted after scattering in the neutral ISM, resulting in the halo phenomenon discussed above and in the following section. Thus some of the Lyα that shows Lyα/Hα ≲ Case B (point b) must be scattered and not simply attenuated by dust.

Figure 8. Comparison of Lyα and Hα surface photometry for Haro 11. The abscissa shows the logarithmic Hα surface brightness, measured in individual pixels in the HST images, while the ordinate axis shows the corresponding surface brightness in Lyα, i.e. contours show the density of points when comparing pixel values in the frames from Figure 7. The red line shows the case B recombination line ratio of Lyα=8.7×Hα; intrinsically astrophysical nebulae produce Lyα and Hα radiation that follows this line. The plot is logarithmic, so only Lyα emitting regions can be visualised. Histograms above and to the right show the overall the distribution of light emitted as a function of surface brightness.

5.2 Extended halos

Early imaging observations showed that in galaxies with net Lyα emission, the dominant fraction comes from a component of large-scale extended emission that surrounds the star-forming regions. Indeed, upwards of 50% of the Lyα is typically emitted in halo regions that extended at least 10 kpc from the UV-bright clusters (Hayes et al. Reference Hayes, Östlin, Mas-Hesse, Kunth, Leitherer and Petrosian2005; Atek et al. Reference Atek, Kunth, Hayes, Östlin and Mas-Hesse2008), and in some galaxies this is the only Lyα that emerges. The first large-scale Lyα imaging survey of local starbursts – LARS (Östlin et al. Reference Östlin2014) – has shown that extended halos are near ubiquitous in Lyα-emitting galaxies (Hayes et al. Reference Hayes2013). Moreover, LARS has enabled the first systematic survey of the sizes of these halos and the comparison with other wavelengths.

Continuum-subtracted Lyα images show a wide range of morphologies. Typically, they do qualitatively resemble those of the UV and Hα, but are more extended, and envelop the galaxies. On average, halos have twice the linear size in Lyα that the galaxy does in either the UV stellar continuum or Hα (Figure 9, Hayes et al. Reference Hayes2013). These estimates are made using the depth- and redshift-independent Petrosian radii, which are found to be below 15 kpc in Lyα for LARS galaxies, with a median value of 5 kpc. Furthermore, the extension of the Lyα surface (ξ_Lyα, the ratio of Lyα radius to Hα radius) is not an independent quantity, and is correlated with a number of measured properties: notably ξ_Lyα is anti-correlated with dust abundance, as demonstrated by the lower panel of Figure 9, and Lyα extension is also found to be larger at lower metallicity and stellar mass. Radiative transfer simulations (using Verhamme et al. Reference Verhamme, Dubois, Blaizot, Garel, Bacon, Devriendt, Guiderdoni and Slyz2012) show that this effect cannot be reproduced simply varying the dust content and what gives rise to these extended halos is currently unclear. Pardy et al. (Reference Pardy2014) have shown that higher ξ_Lyα is produced by galaxies with narrower 21-cm line-widths, but not the total mass in Hi (which neglecting mergers correlates with the line-width), possibly indicating that lower mass galaxies with less complex large-scale morphologies are the ones in which Lyα scatters to the largest relative distances.

Figure 9. Extension of Lyα halos in the LARS sample; figure is taken from Hayes et al. (Reference Hayes2013). The upper panel shows the Petrosian radius measured in Lyα plotted against that measured in Hα. The two galaxies that globally absorb Lyα are set to negative size. The dashed line shows the 1-to-1 line, and Lyα is on average twice the linear size of Hα. The lower panel shows the relative extension of Lyα compared to Hα, ξ_Lyα. This example shows how the extension is anticorrelated with the UV continuum slope, β, which is commonly used as a proxy for the dust content in galaxies.

6.1 Evolution of Lyman alpha galaxies into the nearby universe

IUE demonstrated that Lyα emission is rare in the nearby universe but GALEX could determine how rare. The first Lyα luminosity functions (LFs) at z ≈ 0.3 showed that Lyα-emitting galaxies have become both fainter and less abundant than at high-z. Figure 11 shows the LFs measured at z ~ 0.3 and 1 (Cowie et al. Reference Cowie, Barger and Hu2010; Wold et al. Reference Wold, Barger and Cowie2014, respectively), together with recent measurements for z = 2.1 (Ciardullo et al. Reference Ciardullo2012).

Figure 11. Left: The evolution of the Lyα luminosity functions with redshift. Turquoise line shows the z ≈ 0.3 LF of Cowie, Barger, & Hu (Reference Cowie, Barger and Hu2010) and purple shows the z ≈ 1 LF of Wold, Barger, & Cowie (Reference Wold, Barger and Cowie2014), measured by GALEX FUV and NUV, respectively. The dark red line shows the LF of Ciardullo et al. (Reference Ciardullo2012), measured at z = 2.1 using narrow-band filters. Significant evolution is seen between the distant universe and the local one; grey arrows show the suspected modes of evolution in both luminosity and density (see the text for details). The dotted line shows the Hα LF at z = 0.2 of Tresse & Maddox (Reference Tresse and Maddox1998). Right: the cumulative Lyα equivalent width distribution at the same redshifts; data are taken from the same surveys, with the z ≈ 2.1 distributions from Guaita et al. (Reference Guaita2010, Reference Guaita2011).

It should be noted that, while these LFs are the best that can be done with GALEX, the samples are not large: 119 at z ~ 0.3 and 141 at z ~ 1. However, because of GALEX’s wide field-of-view and the large continuous wavelength range provided by slitless spectroscopy, the cosmic volumes probed are in fact rather large, and covering several 10⁶ comoving Mpc³. Thus while the individual parameters in the Schechter function may not be very tightly constrained, the small number of galaxies is certainly because of the relative paucity of LAEs at lower redshifts.

As discussed in Section 6, the two GALEX samples cover different luminosity ranges (Figure 10). Moreover, the dynamic range of the surveys is not large, and even for the less luminous z ≲ 0.4 sample it is not possible to calculate the faint-end-slope (α) of the LF. Thus the LFs presented in Figure 11 assume α, basing the assumption on measured values at z = 2–3 for the z ~ 1 sample, and the α for local Hα-emitters for the z ≲ 0.4 sample. Note also that the faint-end slope of the Lyα LF is also not very tightly constrained at high redshift (reasons outlined in Dressler et al. Reference Dressler, Henry, Martin, Sawicki, McCarthy and Villaneuva2015).

At z ≈ 0.3 and to the UV limits mentioned in Section 6, the shape of the Lyα LF closely resembles that of Hα and Hβ, although is lower in normalisation: Lyα-emitting galaxies make up about ≈ 15% of the local Hα-selected galaxy population (Deharveng et al. Reference Deharveng2008; Hu et al. Reference Hu, Cowie, Kakazu and Barger2009) and 1/20 the FUV counts at the same z (Cowie et al. Reference Cowie, Barger and Hu2010). This equates to a volume-averaged escape fraction of below 1% (Hayes et al. Reference Hayes, Schaerer, Östlin, Mas-Hesse, Atek and Kunth2011), although this f ^Lyα_esc will be slightly higher when considering possible emission from galaxies not formally classed as LAEs (i.e. galaxies that emit weaker Lyα with 0 ⩽ W _Lyα < 20 Å). At z ~ 2, the average f ^Lyα_esc is ≈ 5%, so the average Lyα output of the whole cosmic volume decreases 5-fold. Note that this decrease in the emitted fraction of Lyα happens on top of the decrease in the cosmic SFR density, which also drops by a factor of 5–10 over the same change in redshift (Madau & Dickinson Reference Madau and Dickinson2014), implying the Lyα luminosity density of the local universe is far below that of z ≳ 2.

A large fraction of this evolution takes place in the 4.3 Gyr that elapses between redshifts of 1 and 0.3 (Wold et al. Reference Wold, Barger and Cowie2014). At the higher redshift of 1, luminous LAEs are certainly in place, with luminosities equivalent to L ^⋆ at z ≈ 2 −3, and the evolution between z ≈ 0.3 and 1 can be well described by a simple factor of 10 increase of L ^⋆ in the Schechter (Reference Schechter1976) function. Over this redshift range, the space density of LAEs does not appear to change but the galaxies simply scale up in luminosity. Between z = 1 and 2 (3 Gyr), ϕ^⋆ increases by an order of magnitude to produce the LFs observed in ground-based surveys. Arrows in Figure 11 show how the evolution of the LF manifests as a drop in density, followed by a dimming. Between the peak in the cosmic SFRD at z ~ 2–3 and z = 1, the universe first acts to turn off a fraction of the LAEs, whereafter between z = 1 and the nearby universe the abundance is constant but the galaxies get fainter in line with the evolution in both the UV and Hα.

Interestingly, this evolution is not strongly reflected in the EW distribution of LAEs, which does not evolve as dramatically. The EW distribution of Guaita et al. (Reference Guaita2011) at z ≈ 2.1 agrees well with the distribution at z ~ 1 (shown in the right panel of Figure 11) despite the fact that the overall Lyα-emitting fraction has decreased by a factor near 5 (Cowie et al. Reference Cowie, Barger and Hu2010 contrasted with Shapley et al. Reference Shapley, Steidel, Pettini and Adelberger2003). Thus whatever process is turning LAEs off, it does not affect the shape of the remaining EW distribution (i.e. galaxies with W _Lyα ≳ 20 Å). Of course, the EW distribution of the overall population changes significantly, as many galaxies drop below the canonical 20 Å limit; the higher EW tail of the distribution remains largely constant. Note that it is not necessarily fair to conclude strong evolution in the EW distribution to z ~ 0.3 from Figure 11, as the GALEX FUV observations are first continuum selected. This will lead to a fraction of continuum-faint objects with high EWs being missed, which will extend the tail of the distribution. However, because the bulk of the luminosity comes from low-EW galaxies, the LFs are will be largely unaffected.

In light of the extended Lyα halos discussed in Section 5.2, we may ask whether much Lyα also extends beyond the spectroscopic extraction apertures of GALEX. These spectra are extracted using an optimal model of the point spread function (PSF, Morrissey et al. Reference Morrissey2007), which has FWHM of ≈ 5 arcsec. At z = 0.3 (z = 1), this aperture corresponds to a spatial scale of 22 (40) kpc, and thus one-dimensional spectra are summed over scales that exceed this. At z ~ 0.3, the aperture is five times the median Lyα Petrosian radius in LARS, and while we do not known precisely how the Lyα profiles behave at larger radii, it is likely that in most cases GALEX captures the majority of the total flux. At z = 1, Barger et al. (Reference Barger, Cowie and Wold2012) discovered a giant Lyα blob, extended over an 18-arcsec diameter, but also measure such objects to be extremely rare (one in the whole volume). Moreover, aperture sizes (≈ 5 arcsec at z = 0.3) are equivalent to a 3-arcsec diameter aperture at z = 2, so if halo extension does significantly affect the recovered Lyα flux, it is likely by a similar factor as in high-z observations.

6.2 The properties of nearby Lyman alpha galaxies

Unlike most high-z studies, both the GALEX and HST samples are sufficiently close that many of their properties may be systematically measured. I assemble some of the key data obtained from these telescopes in Figure 12, and in this section discuss what we have learned about galaxies that emit, and do not emit, Lyα.

Figure 12. Scatter plots to the Left show the observed global Lyα/Hα ratio measured by GALEX (pink squares) and HST (dark blue circles). For the GALEX samples, data are assembled from Cowie et al. (Reference Cowie, Barger and Hu2011), Scarlata et al. (Reference Scarlata2009), Finkelstein et al. (Reference Finkelstein, Cohen, Moustakas, Malhotra, Rhoads and Papovich2011), and Atek et al. (Reference Atek, Kunth, Schaerer, Mas-Hesse, Hayes, Östlin and Kneib2014); for the HST samples data are taken from Östlin et al. (Reference Östlin, Hayes, Kunth, Mas-Hesse, Leitherer, Petrosian and Atek2009, Reference Östlin2014) and Hayes et al. (Reference Hayes2013, Reference Hayes2014). Note that because of the differing quantities plotted on the abscissa, the figures do not necessarily include the same number of points. Furthermore, since these points are chosen to have Lyα in emission, the non-emitting population cannot be visualised. Upper Left: the absolute UV magnitude in the AB system. Upper Right: the Hα equivalent width. Lower Left: the Hα/Hβ ratio, where the black line shows the effect of dust attenuation (Calzetti et al. Reference Calzetti, Armus, Bohlin, Kinney, Koornneef and Storchi-Bergmann2000 law), assuming Lyα/Hα=8.7 and Hα/Hβ=2.86. Lower Right: the [Nii]λ6584 Å/Hα ratio (=N2 index). Histograms to the Right compare the frequency of GALEX-LAEs (pink filled histograms) and non-Lyα-emitting galaxies (green outlined histogram), as a function of W _Hα (upper) and N2 index (lower).

8.1 What is the atomic gas distribution in star-forming galaxies?

The spatial distribution of emitted and absorbed Lyα must reflect some set of properties of the Hi gas. In very nearby galaxies, 21-cm observations have already revealed the structure of the atomic gas where, for example, the Hi Nearby Galaxy Survey (THINGS, Walter et al. Reference Walter, Brinks, de Blok, Bigiel, Kennicutt, Thornley and Leroy2008) and VLA-ANGST (Ott et al. Reference Ott2012) surveys have found an atomic medium that is largely inhomogeneous and clumpy. It is clear that if Lyα photons were injected in these galaxies, they would experience more scattering in some regions than others. Examining at least these 21-cm observations, we see that the Hi does not much resemble the shells and slabs in which many radiation transfer calculations are done. In order to empirically determine the effects that Hi and its distribution have on Lyα emission requires resolved observations of individual targets in both Hi and Lyα.

Currently, we may contrast average Lyα surface brightness profiles (e.g. Hayes et al. Reference Hayes2014) with those of the intensity at 21 cm (e.g. Bigiel & Blitz Reference Bigiel and Blitz2012). These samples show both Lyα and 21-cm intensity profiles that are best fit with Sérsic profiles that are close to exponential (n ~ 1), while UV/optical wavelengths in the LARS sample show significantly higher Sérsic indices (n ≳ 3). Does the Lyα surface brightness trace the gas column density? While curious, this may be entirely coincidental as the sample selection is very different in the two cases. Such observations established in the same galaxies would be enormously instructive in interpreting the Lyα halos of both galaxies at low and high redshifts.

It is unfortunate that resolutions attainable in 21 cm and in the UV/optical are not well matched; sampling small physical scales at 21 cm requires very local galaxies, while the redshift requirements to observe Lyα imply targets must lie beyond several tens of Mpc. Progress has been made by Cannon et al. (Reference Cannon, Skillman, Kunth, Leitherer, Mas-Hesse, Östlin and Petrosian2004) and Pardy et al. (Reference Pardy2014), but with synthesised beam sizes of ≈ 15 arcsec in the best case (usually much coarser), such Hi observations can place only two resolution elements inside the linear size of the ACS/SBC camera. However, VLA in configurations B and A can provide resolutions down to around 4 and 1.3 arcsec, respectively, albeit with a substantial increase in observing time. Observational programmes at higher resolution (VLA C and B configurations) are ongoing, but it is clear that large steps forward may be taken if Hi observations are pushed to the highest spatial resolutions, and such observations are now essential.

8.2 How is the Lyα spectral profile built?

The total integrated Lyα profile of a galaxy is built by emission from different regions, likely with differing kinematics, orientation, and with different contributions to the total Lyα. The spectral profiles of Lyα measured in small apertures tend to be P Cygni, with radiation absorbed from the blue and re-radiated in the red. In these small apertures, Lyα escape fractions are also low. However, imaging tells us that these photons are often not expunged from the system, but scattered back into the line of sight at different position, where large-aperture photometry measures significantly higher f ^Lyα_esc. It is an unfortunately common feature of absorption spectroscopy that we rarely know the line-of-sight distance between the emitting sources and the absorbing gas.

Ideally, we would like to know where the frequency redistribution of Lyα occurs, and to what extent the halo emission shares the spectral profile of the more central regions where the starburst is located. For example, if frequency redistribution occurs close to ionising clusters and Lyα is singly scattered at large radii, the profile may indeed be similar over large distances. However, if photons also get caught for many scatterings in halo gas then a new kinematic structure may be encoded in the Lyα, or double-peaked profiles may be seen.

To resolve this, we may take high-resolution spectra of the diffuse Lyα-emitting regions in galaxy halos. These observations would be best supported by aperture-matched optical spectroscopy of Hα or Hβ, to tightly constrain the rest velocity distribution of Lyα, and ideally also the highest possible resolution 21-cm observations (previous section). The best tool for this would be HST/STIS with narrow slits, which could provide resolutions of around 20 km s⁻¹, but again at the cost of long integrations in faint regions. Mas-Hesse et al. (Reference Mas-Hesse, Kunth, Tenorio-Tagle, Leitherer, Terlevich and Terlevich2003) have shown that in IRAS 08339+6517, the blue wing of the Lyα profile sets on at a similar wavelength over at least 10 kpc, suggesting the star-forming regions are surrounded by a Hi medium that may be rather homogeneous in both space and velocity. However, 21-cm observations find gas at much larger radii, where Lyα has not yet been spectroscopically detected; whether the profile bends smoothly and traces the edge of a bubble, and how this effect may vary in different galaxies are all currently unknown.

8.3 Ionisation state of the interstellar medium

Section 6.2 discusses the effect of a large number of galaxy properties on the emission of Lyα. Many have been studied over time, but largely overlooked has been the effect of ionisation state of the ISM. Specifically ionising radiation from stars may not only produce recombination nebulae but also heat the diffuse warm medium. Thus as well as producing Lyα, the further propagation of LyC radiation may increase the ionisation levels of the diffuse gas, lowering the optical depth of the Hii regions to Lyman radiation (e.g. Pellegrini et al. Reference Pellegrini, Oey, Winkler, Points, Smith, Jaskot and Zastrow2012). Given that the ISM of galaxies may be very inhomogeneous, and that Lyα may be absorbed close the nebulae in which it forms, the propagation of ionising radiation could potentially have a large impact upon the first stages of Lyα transfer.

The ionisation parameter (the number of hydrogen-ionising photons per atom) governs the excitation of the gas, which is usually quantified observationally by the excitation parameter (an emission line ratio that contrasts high- and low-ionisation species). This is done most effectively by taking ratios of p ² and p ³ ions: Zastrow et al. (Reference Zastrow, Oey, Veilleux, McDonald and Martin2011, Reference Zastrow, Oey, Veilleux and McDonald2013) performed ‘ionisation parameter mapping’ of several local starbursts using the [Siii]λ9069/[Sii]λ6716 ratio, to identify highly ionised cones that could signpost LyC emission. Similar integrated measurements of ‘Green Pea’ galaxies found high [Oiii]λ5007/[Oii]λ3727 ratios (Jaskot & Oey Reference Jaskot and Oey2013) that implies high ionisation states, and indeed these galaxies have been found to be high-EW LAEs (Jaskot & Oey Reference Jaskot and Oey2014; Henry et al. Reference Henry, Scarlata, Martin and Erb2015).

Currently, we do not know how these highly ionised regions affect Lyα on small scales in the ISM. Resolved Lyα imaging has now been obtained for over 50 galaxies with ACS, but as yet the ionisation structure has only been mapped in one of them. Bik et al. (Reference Bik, Östlin, Hayes, Adamo, Melinder and Amram2015) used VLT/MUSE observations of [Sii] and [Oiii] to map the ionisation parameter and gas kinematics in ESO 338–IG04, and compare with the Lyα imaging from Hayes et al. (Reference Hayes, Östlin, Mas-Hesse, Kunth, Leitherer and Petrosian2005). Two outflowing and highly ionised conical regions are revealed, that approximately align with the brightest Lyα regions. This would be consistent with a scenario in which Lyα may propagate with less scattering through these highly ionised regions, but the observations of one system cannot establish a causal relation, and a large sample of similar observations needs to be obtained. Pertinent observational questions include whether Lyα emission is systematically enhanced in regions of high ionisation, or whether chimneys through the ISM may feed brighter regions of Lyα emission in extended halos.

8.4 Can we predict Lyα observables from other information?

Given the importance of Lyα observations at both low and high redshift (see Section 1.1), a vital question becomes whether we can predict the Lyα escape fraction, EW, or line profile from a given set of quantities. That is, if we are given a dust content, a characteristic velocity for outflowing atomic gas, etc., do our predictions for Lyα emission/absorption match reality?

Given the absence of strong correlations involving Lyα (for example in Figure 12), one may be tempted to conclude that our understanding is not this sophisticated. However if, on the other hand, the Lyα that we observe is governed by geometrical considerations such as viewing angle, then averaged over many galaxies the answer may be more encouraging. For example it has been shown that transfer models inside homogeneous shells can reproduce very wide ranges of line profiles (Schaerer et al. Reference Schaerer, Hayes, Verhamme and Teyssier2011), similar to those that are observed globally at high-z; furthermore when coupled with semi-analytical models of galaxy formation, such models are able to reproduce the broad features of the Lyα LF over a wide range of redshift (Garel et al. Reference Garel, Blaizot, Guiderdoni, Schaerer, Verhamme and Hayes2012). This may imply that our more general picture of Lyα transport is correct.

It is not clear whether current samples are sufficiently large, or span a high enough dynamic range in luminosity/SFR for such an exploration. However, in the coming years, larger databases of global properties will become available; GALEX LAEs will remain at around 100 objects; HST spectroscopic samples already exceed this while imaging observations remain somewhat smaller. This should permit statistical studies using linear discriminant analyses, to determine how combinations of properties produce the observed Lyα characteristics. Many such possibilities may be envisaged if the signal is strong enough and the samples are sufficiently large.

More empirically motivated simulations can be performed using the wealth of data available for low-z galaxies as input. In such an approach, nature sets up the ISM instead of computers. For example, transport calculations have been run in synthetic galaxies output by hydrodynamical simulations (Laursen, Sommer-Larsen, & Andersen Reference Laursen, Sommer-Larsen and Andersen2009; Verhamme et al. Reference Verhamme, Dubois, Blaizot, Garel, Bacon, Devriendt, Guiderdoni and Slyz2012; Yajima et al. Reference Yajima, Li, Zhu, Abel, Gronwall and Ciardullo2014), although as yet there has been no attempt to construct realistic input conditions of a galaxy based upon observation. Such an experiment is not easy, especially without detailed knowledge of the Hi, but as that becomes available the more the study becomes a possibility. With maps of the ionising stellar population, nebular gas and ionisation structure, Hi distribution including large- and small-scale kinematics, it will become possible to generate sets of model galaxies that are based upon real systems for which Lyα observations have been obtained. Transfer simulations in such models will then recover the Lyα morphology and spectral profile, that can be tested against observation.