Free Access
Issue
A&A
Volume 561, January 2014
Article Number A89
Number of page(s) 17
Section Extragalactic astronomy
DOI https://doi.org/10.1051/0004-6361/201321519
Published online 08 January 2014

© ESO, 2014

1. Introduction

The study of galaxy formation and evolution is strongly related to the observational techniques used to detect galaxy populations at different epochs. In particular, the detection of the Lyα emission line in very distant galaxies has triggered many cosmological applications. The Lyα line is predicted to be the dominant spectral signature in young galaxies (Charlot & Fall 1993; Schaerer 2003), and combined with its rest-frame wavelength in the UV, one obtains a very efficient tool for detecting galaxies at 2 < z < 7 in the optical window from ground-based telescopes. Therefore, this potential has often been used to assemble large samples of star-forming galaxies at high redshift (e.g., Gronwall et al. 2007; Nilsson et al. 2009; Ouchi et al. 2010; Hayes et al. 2010; Cassata et al. 2011; Kashikawa et al. 2011). Lyα-based samples additionally offer the opportunity of constraining the stage of cosmic reionization using (i) the expected sharp drop in their number density as the ionization state of the intergalactic medium (IGM) changes (e.g., Malhotra & Rhoads 2006), or (ii) the shape of the Lyα line profile itself (Kashikawa et al. 2006; Dijkstra & Wyithe 2010).

Interpreting the Lyα-based observations hinges, however, on a complete understanding of the redshift evolution of the intrinsic properties of galaxies, the physical properties responsible for Lyα emission, and the way Lyα emitters (LAEs) are related to other galaxy populations. Given the different selection methods, it is still unclear how the LAE population is connected to the Lyman-Break Galaxies (LBGs) selected upon their rest-frame UV continuum. The possible overlap between these galaxy populations has been questioned in many studies recently but no definitive answer has emerged yet. Specifically, only a fraction of LBGs (Shapley et al. 2003) and Hα emitting galaxies (Hayes et al. 2010) show Lyα in emission, and LBGs appear to be more massive, dustier, and older (e.g., Gawiser et al. 2006; Erb et al. 2006). A reasonable scenario attributes these observed differences to the radiative transport of Lyα photons in the ISM (Schaerer & Verhamme 2008; Verhamme et al. 2008; Pentericci et al. 2010). Most observed trends of Lyα would be driven by variations in N(Hi) and the accompanying variation in the dust content that will dramatically enhance the suppression of the Lyα line as shown in Atek et al. (2009b). However, Kornei et al. (2010) find that Lyα emission is stronger in older and less dusty galaxies, and Finkelstein et al. (2009b) find that Lyα emission is stronger in massive and old galaxies. Although these results are still model-dependent, since they rely on SED fitting to infer the stellar population properties, they are apparently not compatible with the radiative transfer scenario.

Until recently we have been in the curious situation where we studied thousands of LAEs at redshift z ~ 2 and higher, whereas only a handful of Lyα galaxies are available in the nearby universe without any statistical bearing, since such studies are hampered by the difficulty of obtaining UV observations. In the past few years, the situation has changed significantly with the advent of the GALaxy evolution EXplorer (GALEX, Martin 2005), whose UV grism capabilities enabled the study of a large sample of UV-selected galaxies at z  ≲ 1 (Deharveng et al. 2008; Atek et al. 2009a; Finkelstein et al. 2009a; Scarlata et al. 2009; Cowie et al. 2010). The study of low-redshift LAEs offers the advantage of a much higher flux and the possibility of obtaining critical complementary informations at longer wavelengths in subsequent optical follow-up observations. In Atek et al. (2009a) we obtained optical spectroscopy for a subsample of LAEs detected by GALEX at a redshift of z ~ 0.3. Using optical emission lines, we were able to empirically measure the Lyα escape fraction fesc(Lyα) for the first time in a sample of LAEs and found a clear anticorelation with the dust extinction. Other studies derived the same quantity at different redshifts and find in general the same trend (Scarlata et al. 2009; Kornei et al. 2010; Hayes et al. 2010). Nevertheless, all these results show a large scatter around the fesc(Lyα)-dust relationship, which is presumably the result of a number of other physical parameters that may affect the Lyα escape fraction.

The next generation of telescopes, James Webb Space Telescope, (JWST) and Extremely Large Telescopes (ELTs) will heavily depend on Lyα as a detection and star-formation diagnostic tool in distant galaxies. Therefore, it is crucial that we understand how to retrieve the intrinsic Lyα emission from the observed quantity in order to properly derive the cosmological quantities. Consequently, the Lyα escape fraction represents a key parameter for interpreting current and future Lyα-based surveys. Using a new sample of z ~ 0.3 Lyα galaxies, we discuss the dependence of the Lyα escape fraction on the physical properties of the host galaxy. In Sect. 2, we present our spectroscopic follow-up of GALEX-detected LAEs and the sample of local galaxies included in the study. In Sect. 3 we show our emission-line measurements and the procedure we followed to derive the physical properties of our galaxies, i.e., the metallicity and extinction estimates and the AGN/starburst classification. In Sect. 4, we discuss the correlation of Lyα with various parameters, while Sect. 5 is devoted to the regulation of the Lyα escape fraction. Finally we compare the star formation rate (SFR) measurement based on Lyα with other indicators in Sect. 6 before a general summary in Sect. 7.

2. Observations

2.1. The GALEX sample

The galaxy sample used in the present study represents a subset of a Lyα emitters sample detected by Deharveng et al. (2008) from a GALEX slitless spectroscopic survey. Details about grism mode and the spectral extraction are given in Morrissey et al. (2007). The five deepest fields were used to extract all continuum spectra with a minimum signal-to-noise ratio (S/N) per resolution element of two in the far ultraviolet (FUV) channel (1350 − 1750 Å), giving a total area of 5.65 deg2. Lyα emitters are then visually selected on the basis of a potential Lyα emission feature, which approximately corresponds to a threshold of EWLyα ≳ 10 Å. This visual search yielded 96 Lyα galaxies in the redshift range z ~ 0.2−0.35, classified into three categories (1 = good, 2 = fair, and 3 = uncertain) according to the quality of their identification (see Table 2 of Deharveng et al. 2008).

2.2. Spectroscopic follow-up

We obtained optical spectra for 24 galaxies in the Chandra Deep Field South (CDFS) and ELAIS-S1, which have been first presented in Atek et al. (2009a). We selected only galaxies with good-quality (Q = 1 or 2) Lyα spectra in Deharveng et al. (2008). We used EFOSC2 on the NTT at ESO La Silla, under good observational conditions, with photometric sky and sub-arcsec seeing (0.5″−1″). We used Grism #13 to obtain a full wavelength coverage in the optical domain (3690−9320 Å) giving access to emission lines from [Oii] 3727Å to [Sii] 6717+6731 Å in the redshift range of our sample. A binning of 2 × 2 is used and corresponds to a pixel-scale of 0.24′′ px-1 and a spectral resolution of FWHM ~ 12 Å (for 1′′ slit). To avoid second-order contamination that affects the longer wavelength range, an order sorting filter has been mounted to cut off light blueward of 4200 Å. We first used a 5′′ longslit to perform a spectrophotometry of our targets to obtain aperture matched fluxes with respect to the GALEX Lyα measurements. Then, we observed our targets again in spectroscopic mode with 1′′ slit, which offers the higher spectral resolution required to separate Hα from Nii contamination.

The NTT spectra were reduced and calibrated using standard IRAF routines. The EFOSC2 images were flat-fielded using both normal and internal flats taken before each pointing. The two flats give similar results, particularly in terms of fringe residuals in the red part of the 2D spectra. Frames were combined for each object with cosmic ray rejection and useless segments of the images (blue-ward 4200 Å for instance) removed to prevent artificial discrepancies in the sensitivity function and potential errors in the flux calibration. Then, the aperture extraction of 1D spectra was performed with the DOSLIT task, where the dispersion solution is obtained from internal lamp calibration spectra to correct for telescope position variations. Finally, spectra were flux-calibrated using a mean sensitivity function determined by observations of standard stars (Feige110, HILT600, LTT1020, EG21) from the Oke (1990) catalog. A representative subset of the optical spectra is presented in Fig. 1.

thumbnail Fig. 1

Examples of the rest-frame optical spectra of our sample illustrating the quality of the data. The flux density units are the same for all spectra, although the scales are different. The target name is marked on top of each spectrum.

Open with DEXTER

2.3. IUE starburst sample

We have included 11 local starbursts that have been spectroscopically observed in the UV by the IUE satellite. The UV spectra were obtained with 20′′ × 10′′aperture, while the median size of the galaxies is about 25′′ in the optical. The spectra cover a wavelength range of 1200−3300 Å with a spectral resolution of 5 to 8 Å. The ground-based optical spectra were obtained at Cerro Tololo Inter-American Observatory (CTIO) using a 10′′-wide slit and a spectral resolution of 8 Å. The spectra were extracted in a 20′′-long aperture in order to match the IUE large aperture. We have re-analyzed the original UV spectra and the optical spectroscopic follow-up data presented in McQuade et al. (1995) and Storchi-Bergmann et al. (1995, see also Giavalisco et al. 1996, and measured the line fluxes again. The objects were chosen to be distant enough to separate the Lyα feature of the galaxy from geocoronal Lyα emission. The definition of a Lyα emitter/absorber, as well as of the Lyα flux measurement, could be ambiguous for P Cygni profiles or an emission blended with absorption. Therefore, in our measurement procedure, we only consider the net Lyα flux emerging from the galaxy. The line measurements were performed following the same procedure as used for our optical follow-up. We also show the 1D spectra in Fig. 2, which includes UV and optical coverage. We included in our analysis only net Lyα emitters, i.e. with EWLyα> 0. Furthermore, we included here our results from HST/ACS imaging of four nearby galaxies (Atek et al. 2008).

thumbnail Fig. 2

Example of rest-frame UV and optical spectra of the IUE sample. Most of the FUV spectra cover 1200−2000 Å wavelength range and the optical spectra the 3200−7000 Å range. The gap in-between is set to zero.

Open with DEXTER

thumbnail Fig. 3

BPT diagrams used to classify narrow emission-line galaxies. GALEX Lyα emitters are represented with black circles with corresponding error bars, together with an IUE sample of local starburst galaxies in blue squares. The left panel shows [Oiii] λ5007/Hβ versus [Nii] λ6584/Hα ratios. The dashed and solid curves are theoretical boundaries separating starbursts and AGN objects assuming an instantaneous burst (Dopita et al. 2000) and an extended star formation episode (Kewley et al. 2001), respectively. The dotted curve is the “Kauffmann line” (Kauffmann et al. 2003). The right panel represents the [Oiii] λ5007/Hβ versus [Sii] (λ6717 + 6732)/Hα diagnostic with the same theoretical models as in the right panel. The red dot-dashed line shows typical model uncertainties of ± 0.1 dex. The diagram includes all the 11 IUE local galaxies, among which is an object previously classified as an AGN (red square) that we decided to show as an example.

Open with DEXTER

3. Physical and spectral properties analysis

3.1. Emission line measurements

The spectra were analyzed using the SPLOT package in IRAF. The redshifts of the galaxies were derived from the position of several emission lines, and line measurements performed interactively on rest-frame spectra. When possible, we measured the flux and equivalent width of [Oii] 3727 Å, [Oiii] 4959, 5007 Å, Hδ, Hγ, Hα, Hβ and [Nii] 6548, 6584 Å. For most spectra, the Hα line (6563 Å) is blended with [Nii] lines (6548 and 6584 Å) even for the 1′′ slit observations. In this case a deblending routine was used within SPLOT measure individual fluxes in each line, then [Nii]/Hα line ratio was used to correct the spectrophotometric observations for [Nii] contamination. In cases where weak emission lines are only present in the higher resolution mode, photometry is deduced by scaling the flux with strong Balmer lines (such as Hα). Another concern is related to the contamination of Balmer emission lines by underlying stellar absorption. The equivalent width of stellar absorption is directly measured from higher order Balmer lines (typically Hγ and Hδ) provided they are in absorption. When these lines are in emission or undetectable, a typical average value of 2 Å, representative of star-forming galaxies, is adopted (Tresse et al. 1996; González Delgado et al. 1999). To determine uncertainties in the line fluxes, we used SPLOT package to run 1000 Monte Carlo simulations in which random Gaussian noise, based on the actual data noise, is added to a noise-free spectrum (our fitting model). Then, emission lines were fitted in each simulation. The computed MC errors essentially depend on the S/N quality of spectra. These Monte Carlo simulations were done automatically in SPLOT after each line fit.We then propagated the errors through calculating of all the quantities described above and the line ratios, extinction etc, computed hereafter.

3.2. Extinction

To measure the gas-phase dust extinction we use the Balmer ratio between Hα and Hβ. The reddening contribution of our Galaxy is negligible since all our objects are located at high Galactic latitude. The extinction coefficient C(Hβ) is then given by the relation: (1)where f(Hα) and f(Hβ) are the observed fluxes, and B the intrinsic Balmer ratio. We adopted a value of B = 2.86, assuming a case B recombination theory and a temperature of 104 K (Osterbrock 1989). The extinction curve values at Hα and Hβ wavelengths, noted S(Hα) and S(Hβ) respectively, were computed from the Cardelli et al. (1989) extinction law. The reddening EB − V is then simply computed using Eq. (1) and the relation E(B − V) = c/1.47. The extinction parameter AV is derived using a mean ratio of R = AV/E(B − V) = 3.2.

We see in Sects. 4.2 and 5 that some galaxies show negative E(B − V). Because we assumed an average correction for the stellar absorption, an overcorrection would lead to underestimating the Hα/Hβ ratio that can explain the few negative values as can be appreciated from the error bars on the E(B − V) values. In addition, a nebular reflection can also artificially increase the Hβ contribution. In all the following calculations, we assigned E(B − V) = 0 to all galaxies showing negative values. However, we show the actual measured values in the plots so the reader can appreciate the uncertainties.

3.3. AGN-starburst classification

To discuss the nature of the ionizing source in our galaxies, we rely on optical emission line ratios. We examine the [Oiii] λ5007/Hβ versus [Nii] λ6584/Hα diagnostic diagram, also known as the BPT diagram (Baldwin et al. 1981). It allows Hii regions photoionized by young stars to be distinguished from regions where photoionization is dominated by a harder radiation field, such as AGNs or low-ionization nuclear emission-line objects (LINERs). The underlying physics is that because photons from AGNs induce more heating than those of massive stars, they will favor the emission from collisionally excited lines with respect to recombination lines. We also use an additional diagram originally proposed by Veilleux & Osterbrock (1987) representing [Oiii] λ5007/Hβ versus [Sii] (λ6717 + 6732)/Hα.

Figure 3 shows the location of our galaxies in these diagrams using dereddened line ratios, although these values are nearly insensitive to dust extinction and to the associated uncertainties discussed above. A sample of IUE star-forming galaxies in the nearby universe are also shown for comparison. As expected, the points lie in a relatively narrow region highlighting the separation of starburst-like galaxies from other types. In both figures, we have plotted the theoretical curves based on photoionization grids, which leave star-forming regions to their lower left and AGN-type objects to the upper right. In the lefthand panel, the upper limit for Hii regions with an instantaneous zero-age star formation model (Dopita et al. 2000) and with an extended burst scenario (more than 4−5 Myr, Kewley et al. 2001) are shown. The BPT boundary lines are strongly dependent on the effective temperature of the ionizing source, which evolves with time and therefore depends strongly on the star-formation time scale (see for example Cerviño & Mas-Hesse 1994; Kewley et al. 2001). Using a large sample from the Sloan Digital Sky Survey (SDSS), Kauffmann et al. (2003) revised this upper limit downward. In the righthand panel we have also plotted the “Dopita and Kewley lines” for the second diagnostic method. However, these models are subject to different uncertainties related to the assumptions made on the chemical abundances, slope of the initial mass function (IMF), or the stellar atmosphere models. The errors on the starburst boundaries may be on the order of 0.1 dex (Kewley et al. 2001).

We can see that all z ~ 0.3 galaxies but one lie on or below the solid starburst division line. This is very clear in the left [Oiii]/Hβ versus [Nii]/Hα diagram but proves more ambiguous on the right [Oiii]/Hβ versus [Sii]/Hα plot, although this is still consistent with both model and observation uncertainties. The observational uncertainties are particularly important on the right diagram because both redshifted [Sii] λ6717,6732 and Hα lines are affected by a bright sky background at these wavelengths. Therefore, The BPT diagrams cast doubt on three candidates, as objects possibly excited by active nuclei, which we exclude from our analysis. This represents about ~12% of our sample, which is consistent with Cowie et al. (2010) and Scarlata et al. (2009), who find AGN fractions of 16% and 17%, respectively, in their z ~ 0.3 LAE samples. This appears higher than typical AGN fraction (~0−5%) found in high redshift LAE sample (Wang et al. 2004; Gawiser et al. 2006, 2007; Nilsson et al. 2009; Ouchi et al. 2008). Lacking optical spectroscopy, the high-redshift AGN fraction is determined from X-ray data, where the low-luminosity sources comparable to the low-z galaxies might be missed (Finkelstein et al. 2009a). Therefore, this apparent evolution of the AGN fraction might be selection effects due to the difficulty of identifying AGN in high-redshift observations. At low redshift, Finkelstein et al. (2009a) find a significantly higher fraction up to ~45% for their LAE sample. There is an overlap of 14 galaxies between the samples of Cowie et al. (2010) and Finkelstein et al. (2009a). Their AGN classification agree on nine of them, while the disagreement for the remaining three stems from either a discrepancy in the line measurements, i.e. the position in the BPT diagram, or the presence/absence of high-ionization lines. In particular, object GALEX1421+5239 has been classified as a star-forming galaxy by Cowie et al. (2011) based on the BPT diagram, while Finkelstein et al. (2009a) identifies this object (EGS14 in their catalog) as an AGN after the detection of high-ionization lines.

3.4. Oxygen abundance

The determination of the nebular metallicity can give important constraints on the chemical evolution and star formation histories of galaxies. The metal production and the chemical enrichment of the ISM is a direct result of stellar mass build-up and star formation feedback. In addition, the influence of the metal abundance on the evolution of the Lyα output in these galaxies can be investigated.

thumbnail Fig. 4

Comparison between the various oxygen-abundance-derived indicators: R23 = ([Oii] λ3727 + [Oiii] λλ 4959, 5007)/Hβ (Pagel et al. 1979), N2 = log ([Nii]λ6583/Hα) and O3N2 = log {([Oiii] λ5007/Hβ)/([Nii] λ 6583/Hα)} (Pettini & Pagel 2004). The z ~ 0.3 galaxies are shown with black circles and the local galaxies with blue squares with 1σ error bars. The solid line represents the 1:1 ratio with ±0.1 dex uncertainties (dashed lines). Some galaxies are not present in all the plots because the corresponding diagnostic lines were not detected with sufficient significance.

Open with DEXTER

To derive the nebular metallicity of our galaxies we use the optical emission line ratios and different metallicity calibrations according to the lines available in our spectra. For most of our galaxies, we measure the widely used abundance indicator, the R23 index, which is based on the ratio ([Oii] λ3727 + [Oiii] λλ 4959, 5007)/Hβ (Pagel et al. 1979; McGaugh 1991; Kobulnicky et al. 1999; Pilyugin 2003). The main problem with the relationship (O/H)–R23 is that there is a turnover around 12 + log(O/H) ~ 8.4 that makes this index double-valued. A given R23 value could correspond to high or low metallicity. Kewley & Ellison (2008) have shown that additional line ratios, such as [Nii]/[Oii] or [Nii]/Hα can be used to break this degeneracy. We used the [Nii]/[Oii] ratio to determine which of the upper and lower branches of the metallicity calibration to use.

Several studies have demonstrated that the R23 method generally overestimates the actual metallicity when compared to the direct method (e.g., Kennicutt et al. 2003; Bresolin et al. 2005; Yin et al. 2007). This discrepancy prevents us from having an accurate measurement with R23 for intermediate metallicities, i.e. around the turnover region. Therefore, we here turn to other metallicity indicators that give more consistent results. We first use the emission-line ratio log([Nii]λ6583/Hα), known as the N2 index (Denicoló et al. 2002; Pettini & Pagel 2004). The calibration of this indicator has been derived from the comparison of (O/H) measured using the “direct” method (via the electron temperature Te), and the N2 index in a large sample of extragalactic Hii regions. They obtain the relationship that translates N2 to oxygen abundance, 12 + log (O/H) = 8.90 + 0.57 × N2. Then, we consider the O3N2 index originally introduced by Alloin et al. (1979), which corresponds to the line ratio log{([Oiii] λ5007/Hβ)/([Nii] λ 6583/Hα)}. Pettini & Pagel (2004) used a subsample of their Hii regions to produce a calibration for this indicator and find a best fit with 12 + log (O/H) = 8.73 − 0.32 × O3N2. However this relationship is only valid for galaxies with O3N2 < 2, which is the case for our galaxies. Of our sample of 21 galaxies, we were able to perform these metallicity measurements for 20 of them (for the remaining one the quality of the spectrum is not sufficient to confidently measure emission line fluxes), and for all 10 IUE galaxies.

It is well known that the R23 and other indicators are sensitive to the ionization parameter, and applying the above calibration to our sample assumes no variation in the ionization state, which can lead to inaccurate metallicity estimates. A detailed discussion about the effects of the ionization parameter on the different metallicity indicators has been presented in several studies using photoionization models (see for example Cerviño & Mas-Hesse 1994; Kewley & Dopita 2002). When Z < 0.5 Z, i.e. 12 + log (O/H) ≲ 8.6, an increasing ionization parameter corresponds to an increasing value of the R23 index, and inversely for metallicities higher than 0.5 Z (see Fig. 5 in Kewley & Dopita 2002). For single-valued indicators, the index increases with increasing ionization parameter. However, for their sample of z ~ 0.3 galaxies, Cowie et al. (2011) determined the metallicity using the “direct” method thanks to the detection of the [Oiii]λ4363, and compared the results to the N2 indicator. They found a good correlation with small scatter indicating relatively small effects from the ionization parameter. Moreover, while the absolute value of the metallicity may suffer from uncertainties, the trends analyzed here should hold irrespective of the absolute values, given the narrow range of metallicities covered by the sample.

We show in Fig. 4 a comparison between the oxygen abundance derived from several indicators: R23, N2, and O3N2. It appears that N2 and O3N2 are in good agreement, although the values derived from O3N2 are systematically lower by 0.1 dex on average than those of the N2 method. On the other hand, although we do not have a direct measurement of the metallicity, we confirm that R23 index overestimates the oxygen abundance by a factor up to 0.4 dex when compared to the two other methods used here. The N2 index is used in the rest of this analysis when it is available, otherwise O3N2 is adopted.

4. The regulatory factors of the Lyα output in galaxies

4.1. Effects of metallicity

The first detections of the Lyα emission line in star-forming galaxies goes back to the early IUE observations of nearby Hii galaxies (e.g., Meier & Terlevich 1981; Hartmann et al. 1984; Deharveng et al. 1985). However, in the case of detection, the line was very weak, while in the remaining galaxies it was found in absorprtion. The most natural explanation was the attenuation of the Lyα photons by dust. The metallicity was often used as a dust indicator and compared with the Lyα strength. The study of small galaxy samples (Meier & Terlevich 1981; Terlevich et al. 1993; Charlot & Fall 1993) shows an anticorrelation between EWLyα and the metallicity, whereas the more comprehensive results of Giavalisco et al. (1996) only show a marginal trend.

Here we investigate the effects of the metallicity on Lyα emission in a large sample of galaxies. While the metallicity can be correlated with the Hα equivalent width as the result of metal enrichment with the age of the galaxy, this is not necessarily the case for the Lyα strength. Cowie et al. (2011) compared the metallicity of an LAE sample and UV-selected galaxies and find that galaxies without Lyα tend to have median metallicities 0.4 higher. Finkelstein et al. (2011b) also show that LAEs at low redshift have lower metallicities than similar-mass galaxies from the SDSS. One may expect a strong intrinsic Lyα emission for a low-metallicity galaxy but the observed Lyα output will also depend on other factors. Classical examples are IZw 18 and SBS 0335, the most metal-poor galaxies known, which show a strong absorption in Lyα (Kunth et al. 1998; Mas-Hesse et al. 2003; Atek et al. 2009b). In fact, the metallicity distributions of the LAEs and UV-continuum galaxies measured in Cowie et al. (2011) overlap significantly. Figure 5 shows how Lyα equivalent width is uncorrelated with the oxygen abundance. This means in particular that strong Lyα emitters are not systematically metal-poor galaxies.

thumbnail Fig. 5

Lyα equivalent width as a function of metallicity. The oxygen abundance is derived from the N2 index log([Nii]λ6583/Hα) using the equation of Pettini & Pagel (2004). The present sample is shown with black circles, Cowie et al. (2011) galaxies with magenta circles, and IUE galaxies with blue triangles.

Open with DEXTER

4.2. Effects of dust

As detailed in Sect. 3.2, we have derived the dust extinction in the gas phase from the Balmer decrement Hα/Hβ for all the galaxies studied here. Previous investigations concerning the effects of dust on the Lyα emission and, in particular, the correlation between the Lyα equivalent width and extinction, have interestingly led to different results and conclusions. Giavalisco et al. (1996) found a large scatter in the correlation between EWLyα and the color excess E(B − V) in a sample of 22 nearby starbursts. They concluded that, in addition to resonance scattering effects, Lyα escape may be affected by the geometry of the neutral phase of the ISM, such as patchy dusty regions or ionized holes. However, these results stand in contrast to high-redshift studies of Lyα emitters. Composite spectra of z ~ 3 LBGs (Shapley et al. 2003) showed a trend between EWLyα and E(B − V) where objects with strong Lyα emission also have steeper UV continua. Similar trends have been observed more recently (Vanzella et al. 2009; Pentericci et al. 2009; Kornei et al. 2010) where LBGs that exhibit Lyα in emission tend to have bluer UV slopes than those with Lyα in absorption. Following Shapley et al. (2003), Pentericci et al. (2009) separated their z ~ 4 LBG sample in bins of extinction and noted the same trend between the mean values of EWLyα versus stellar extinction. Again, all the observed correlations listed here are model-dependent and may be affected by several uncertainties. Differences in the inferred physical properties may arise from the different extinction laws adopted in those studies, such as Calzetti et al. (2000) vs. SMC (Prevot et al. 1984), which also depends on the age and the redshift of the galaxies (Reddy et al. 2006; Siana et al. 2009). More generally, the choice of a specific IMF and metallicity or the degeneracy between age and extinction may introduce systematic uncertainties. Finally, as discussed in Hayes et al. (2013), the IUE aperture at z ~ 0.01 probes a much smaller physical size than other studies at higher redshift. For lower dust extinction, Lyα is spatially more extended leading to aperture losses, hence an underestimation of the Lyα escape fraction.

thumbnail Fig. 6

Lyα equivalent width as a function of the gas-phase dust extinction. Black circles are the GALEX Lyα emitters, and green and red circles are the other GALEX samples of Scarlata et al. (2009) and Cowie et al. (2011), respectively. The blue triangles represent the IUE spectroscopic sample, and purple squares local starbursts from Atek et al. (2008).

Open with DEXTER

Figure 6 shows EWLyα as a function of the extinction. The plot includes a compilation of local starbursts and z ~ 0.3 GALEX-selected samples. The published values of Cowie et al. (2011) are in bins of flux, which explains the discrete positions of E(B − V) for this sample. The Lyα equivalent width shows a large scatter for E(B − V) ranging from no extinction to ~0.45. This is consistent with previous results of Lyα imaging observations of local starbursts (Atek et al. 2008; Östlin et al. 2009), of which data points are plotted in the figure. The dispersion of Lyα strength observed for a given value of extinction is indicative of the influence of other galaxy parameters besides pure dust attenuation, such as Hi column density that may increase the suppression of Lyα or neutral gas outflow that may ease the escape of Lyα photons. However, as previously shown by Schaerer & Verhamme (2008), Verhamme et al. (2008), and Atek et al. (2009b), the effects of superwinds might become insufficient in the case of high extinction, in which case we observe Lyα in absorption or with a low equivalent width. This can be observed in Fig. 6 where no high EWLyα values can be observed for large extinctions.

The absence of anti-correlation between EWLyα and E(B − V) can be attributed to the difference in the type of extinction probed. While the stellar extinction is responsible for continuum attenuation, the Lyα line flux is affected by the nebular extinction around the recombination regions. Therefore, the quantity of dust and, more importantly, the geometry of the dust distribution can be very different between the two phases. In that case, the total extinction we derive from integrated fluxes would not be representative of the “effective” extinction affecting each radiation, and the total E(B − V) could become meaningless for very high extinction. We show in Table 3 the extinction factor at λ = 1216 Å for different E(B − V) values and two extinction laws (Cardelli et al. 1989; Prevot et al. 1984). As we can see, for E(B − V) values above 0.3, we could hardly detect any UV continuum or Lyα emission, since the extinction factor becomes rapidly higher than 100. At E(B − V) ~ 1, photons in the UV domain should be completely absorbed, which contrasts with EWLyα values around 50 Å observed at such extinction levels. This suggests two mechanisms: (i) because of their resonant scattering, Lyα photons can spatially diffuse far from the ionized regions where the Balmer lines are produced (from which we measure the extinction) and will sample different extinction; (ii) in the optical domain we sample deeper regions and derive high average E(B − V) values, while in the UV, the observations are dominated by the contribution of the few stars/regions with lower values of extinction. Deeper regions would be completely obscured and would not contribute significantly in the UV. As a result, the average E(B − V) value in the UV (both Lyα and continuum) would be much lower than in the optical (see Otí-Floranes et al. 2012, for a detailed discussion about the differential extinction). In Sect. 5, we also investigate the effects of the differential extinction scenario on the Lyα escape fraction.

Table 3

Extinction factors at λ = 1216 Å, calculated for different values of E(B − V) using the Galactic extinction (Cardelli et al. 1989) and the SMC law (Prevot et al. 1984).

4.3. Lyα equivalent width as a function of UV luminosity

With the aim of understanding the different parameters governing the Lyα escape and observability, many studies, both observational and theoretical, have found an apparent variation of the Lyα strength with the UV luminosity (Shapley et al. 2003; Ando et al. 2006; Ouchi et al. 2008; Verhamme et al. 2008; Vanzella et al. 2009; Pentericci et al. 2009; Balestra et al. 2010; Stark et al. 2010; Schaerer et al. 2011; Shimizu et al. 2011). In particular, the data show a lack of high Lyα equivalent width in luminous LBGs. A well known trend is the decrease in EWLyα with SFRUV (Ando et al. 2004; Tapken et al. 2007; Verhamme et al. 2008). At higher redshift, Stark et al. (2010) find that galaxies with strong Lyα emitters have lower UV luminosities, and in general have a bluer UV slope compared to galaxies with weak Lyα emission. A similar plot is shown in Fig. 7. Although, the correlation presented in Verhamme et al. (2008) or Stark et al. (2010) cover a broader range in both SFRUV and EWLyα, our data do not show any such clear trend, but rather an EWLyα distribution independent of UV luminosity.

Together, Figs. 6 and 7 indicate that the difference in the type of extinction probed combined with complex transport of Lyα make the picture difficult to interpret. While the stellar extinction is responsible for continuum attenuation, the line flux is affected by the nebular extinction around the recombination regions combined to gas outflows. Therefore, the quantity of dust and, more importantly, the geometry of the dust distribution, can be very different between the two phases, hence between the continuum and the line behavior. The absence of strong Lyα emission at high SFR(UV) observed in high-z samples can be due to differences in star formation histories and time scales, rather than an increase in the dust content, because powerful instantaneous bursts that produce very high EWLyα are more likely at young ages, i.e. low SFRUV, and constant star formation episodes at later stages characterized by modest EWLyα at equilibrium.

thumbnail Fig. 7

EWLyα as a function of SFRUV. The color code is the same as in Fig. 6. The Lyα equivalent width is the observed rest-frame value, and the SFRUV values are calculated from FUV luminosity at 1530 Å not corrected for extinction. EWLyα uncertainties are 10% error bars, and SFRUV errors are derived from GALEX FUV photometry.

Open with DEXTER

5. The Lyα escape fraction

In Atek et al. (2009a), we presented an empirical estimate of the Lyα escape fraction in a large sample of galaxies selected from GALEX spectroscopy (Deharveng et al. 2008). Here we derive fesc(Lyα) for two additional z ~ 0.3 galaxy samples of Scarlata et al. (2009) and Cowie et al. (2011). The Lyα escape fraction is calculated following the equation: (2)where f(Lyα) is the observed flux and f(Hα)C the extinction-corrected Hα flux. The recombination theory (case B) gives a an intrinsic ratio of f(Lyα)/f(Hα) = 8.7 (Brocklehurst 1971). The Cowie et al. data were obtained with a 1′′ slit, while Scarlata et al. used a 1.5′′ slit. These measurements are not matched well to the GALEX aperture and can suffer from systematic errors. Both groups used a larger aperture for only a few galaxies in their samples, and only to assess the importance of aperture effects. Scarlata et al. (2009) used a 5′′ slit for eight targets and compared the results with those made in the narrow slit, finding errors up to 25%. For the same reason, Cowie et al. (2011) compared their measurements in the narrow and wide slits with those of Scarlata et al. (2009) for a subset of galaxies that overlap, and they find general agreement for their line fluxes. However in some cases the differences were important reaching a factor of 2 or more. We note that those large aperture measurements were not used in their analysis. If this is the case for some of our galaxies, we will overestimate their fesc(Lyα) values.

thumbnail Fig. 8

Lyα equivalent width distribution for the three GALEX samples used in this work. It includes all sources with optical follow-up spectroscopy.

Open with DEXTER

Two effects can lead to inaccurate measurements of fesc(Lyα): (i) if the dust is not uniformly distributed across the galaxy, targeting only the central part of the galaxy would produce systematic error on E(B − V); (ii) the spatial extension of the Hα emission can go well beyond the size of the narrow slit, which results in an underestimate of the total Hα flux that should match the total Lyα flux from GALEX. To quantify these two effects on our results better, we compared our measurements in 1′′ and 5′′ slits. We found that the dust reddening E(B − V) tends to be overestimated by a factor of 1.5 in the narrow slit, which possibly means the dust amount is higher in the center of the galaxy. Regarding Hα emission, we first corrected both observed Hα fluxes using the E(B − V) calculated in their respective slits. We found that the intrinsic Hα flux is underestimated by 30% on average when using the narrow slit. We therefore applied these aperture correction factors to the samples of Scarlata et al. (2009) and Cowie et al. (2011). The effect of such a correction is to increase the intrinsic Hα flux, and thus to decrease the Lyα escape fraction. However, this does not have any significant impact on our results because log (fesc,Lyα) is revised by only ~0.1 on average, and the observed trends remain unchanged (e.g., the correlation between log (fesc,Lyα) and E(B − V)). Another difference is that, unlike the Scarlata et al. sample, Cowie et al. did not apply any stellar absorption to their emission lines. However, the median value for their Hα equivalent width is ~80 Å. A typical correction of 2 Å is then negligible and introduces much smaller errors than the slit loss problems discussed above. Finally, we show in Fig. 8 a comparison between the EWLyα distribution of our sample and those of Cowie et al. and Scarlata et al. We note that the distribution of our sample is shifted to higher EWLyα compared to that of Cowie et al. which peaks at small EWLyα (less than 20 Å). This is the result of our selection in the Deharveng et al. sample of only good quality Lyα lines which favors high EWLyα. This will also result in selecting relatively higher Lyα escape fraction as we see later.

We first show in Fig. 9 that the Lyα escape fraction is not correlated to the oxygen abundance. This is similar to what we have seen in Sect. 4.1. It indicates that the metal abundance is not an important regulatory factor of the Lyα escape.

thumbnail Fig. 9

Lyα escape fraction as a function of the gas-phase metallicity. The oxygen abundance is derived from the N2 index log([Nii]λ6583/Hα) using the equation of Pettini & Pagel (2004). The color code as the same as in Fig. 5.

Open with DEXTER

We present on the left side of Fig. 10 the relation between fesc(Lyα) and the gas-phase extinction. The color code for the different samples is presented in the legend. We fitted the log (fesc) – E(B − V) correlation using the MPFITEXY IDL routine (Williams et al. 2010), which is based on MPFIT procedure (Markwardt 2009). The routine fits the best straight line to the data taking the errors in both x and y directions into account. The best fit corresponds to the function fesc(Lyα) = CLyα × 10− 0.4 E(B − VkLyα where the origin CLyα and the slope kLyα are left as free parameters. Overall, the figure confirms the decreasing trend of fesc(Lyα) with increasing extinction observed in several studies (Atek et al. 2008, 2009a; Verhamme et al. 2008; Kornei et al. 2010; Hayes et al. 2010, 2011). However, when the origin is fixed at CLyα = 1, i.e. assuming fesc(Lyα) = 1 in the absence of dust, we derive an extinction coefficient of kLyα ~ 11.3 ± 0.7, which is slightly higher than what would be expected if dust extinction were the only factor affecting Lyα escape, k1216 = 9.9 assuming a Cardelli et al. (1989) extinction law.

thumbnail Fig. 10

Lyα escape fraction as a function of the nebular dust extinction. The figure shows the z ~ 0.3 Lyα emitters of this work with black circles. We also derived fesc(Lyα) for other galaxy samples: z ~ 0 IUE sample represented by blue triangles, z ~ 0.3 sample of Scarlata et al. (2009) with green circles, Cowie et al. (2011) sample with red circles, z ~ 0 galaxies of Atek et al. (2008) with purple squares. Note that the negative values of E(B − V) are shown in the plot but were assigned E(B − V) = 0 before fesc(Lyα) calculation or fitting the fescE(B − V) relationship. The left panel shows fesc(Lyα) as a function of E(B − V) assuming a Cardelli et al. (1989) extinction law. The solid black line denotes the best 2-parameter fit to the relationship with both the slope and the intercept as free parameters. The yellow region covers the 1σ uncertainties of the fit derived from MC simulations. The dashed red line is the expected attenuation law at the Lyα wavelength. In the right panel we plot the normalized escape fraction fesc,rel(Lyα) = log [fesc(Lyα)/fesc(cont)] as a function of extinction. The normalized escape fraction represents the deviation of fesc(Lyα) from the classical dust attenuation law at 1216 Å. The color code for the samples is the same as in the left panel.

Open with DEXTER

In the case of a two-parameter fit, we find kLyα ~ 6.67 ± 0.52 and CLyα = 0.22 ± 0.03. To estimate the errors on the fit, we performed Monte Carlo simulations by generating a sample of 1000 datasets by randomly varying fesc(Lyα) and E(B − V) within their respective uncertainties. Then, we used the same procedure to fit each dataset. The two-parameter fit shows that even when no dust is present, the Lyα emission is still attenuated by resonant scattering and the geometrical configuration of the ISM. Removing object GALEX1421+5239, which might be an AGN (cf. Sect. 3.3), results in a slightly higher extinction coefficient with kLyα ~ 6.73. Hayes et al. (2011) used the same prescription to describe the effects of dust on the Lyα escape fraction in a compilation of galaxies at redshift z ~ 2. However, the E(B − V) was derived from full SED fitting, and fesc(Lyα) was based either on the UV continuum or the Hα line corrected for extinction using the stellar E(B − V). After fitting the fesc(Lyα) – E(B − V) relation, they obtained kLyα ~ 13.8 and CLyα = 0.445, assuming a Calzetti et al. (2000) extinction law. Compared to these values, we obtain a smaller extinction coefficient kLyα and a lower intercept point at zero extinction. Here we draw attention to selection effects that can possibly contribute to the observed differences. The z ~ 0.3 LAEs detected by GALEX were selected based on a relatively strong Lyα emission, which is therefore biased toward high Lyα escape fractions. In the case of the z ~ 2 sample, galaxies were mostly selected upon their UV continuum or Hα emission. In fact, it has been shown in Hayes et al. (2010) that Lyα- selected galaxies have higher fesc(Lyα) values than Hα-selected ones.

While the dust dependence of fesc(Lyα) is clear, it does not follow classical dust attenuation prescriptions because of secondary parameters at play. In the right panel of Fig. 10, we illustrate the deviation of fesc(Lyα) from the value expected from E(B − V) under standard assumptions by plotting the normalized escape fraction defined as (3)where fesc(Lyα) is the Lyα escape fraction and fesc(cont1216) is the escape fraction of the UV continuum at 1216 Å. For reference, we show the case where the Lyα emission obeys to the same attenuation law as the UV continuum radiation. The values of fesc(Lyα) below this line can be explained by the the scattering of Lyα photons in the neutral gas, which increases their optical path, thus the extinction coefficient. The fesc(Lyα) values above the upper limit driven by the nebular dust extinction could be attributed to several physical processes or any combination thereof.

First, the geometry of the ISM can change the behavior of Lyα photons with respect to dust attenuation. In the case of a multiphase ISM, where the neutral gas and dust reside in small clouds within an ionized inter-cloud medium, Lyα photons scatter off of the surface of the clouds to escape more easily than non-resonant photons that enter the clouds to encounter dust grains (Neufeld 1991; Hansen & Peng Oh 2006; Finkelstein et al. 2008). Therefore, the prefered escape of Lyα increases with increasing extinction, which is observed in our case. Finkelstein et al. (2011a) investigated the effects of ISM geometry on the Lyα escape in a sample of 12 LAEs. Five of their galaxies were consistent with an enhancement of the observed EWLyα. Laursen et al. (2013) and Duval et al. (2014) recently discussed the scenario of multi-phase ISM scenario to explain the unusually high Lyα equivalent widths observed in high-z LAEs. Their models show that very specific and restrictive conditions are needed to enhance the intrinsic EW(Lyα), such as high extinction and low expansion velocity.

Always in the context of a peculiar ISM geometry, Scarlata et al. (2009) discuss the possibility of a different effective attenuation resulting from a clumpy ISM (Natta & Panagia 1984), which does not take the classical eτ form but will depend on the number of clumps. This can reproduce the Lyα enhancement observed in the objects above the dashed line. Given the increase in with increasing extinction, it is possible that the two mechanisms (enhancement of Lyα by scattering and a peculiar extinction law) are in play, as they both require high extinction to significantly increase the escape of Lyα. If we go beyond the simplistic spherical geometry, more realistic configurations of the ISM could also favor Lyα transmission. Ionized cones along the observer’s sightline can primarily transmit Lyα photons via channeling, i.e. scattering of Lyα photons on the Hi outskirts along the cone. Moreover, the Lyα strength also depends on the inclination of the galaxy. Because Lyα photons follow the path of least opacity they will tend to escape face-on. This is not necessarily the case for Hα photons, which do not undergo any important angular redistribution (Verhamme et al. 2012).

More generally, as we have shown in Sect. 4.2, when the extinction is high enough (typically E(BV) > 0.3), the differential extinction sampling would naturally produce values that are higher than one. In the Hii regions affected by dust, fesc(Lyα) will be very low, while in the dust-free regions, where the Lyα emission dominates the rest of Balmer lines, will be higher than unity, since the total E(BV) derived does not apply to those regions. In summary, if one expects starburst galaxies to have a gradation of extinction in different parts of the Hii regions, from regions completely obscured in the UV to regions almost devoid of dust, the convolution of all regions would give an evolution of with average E(B − V) similar to the diagram shown in Fig. 10.

thumbnail Fig. 11

Comparison between SFR indicators. The SFR values are calculated using Kennicutt (1998) calibration. The left panel shows SFR(Hα) versus SFR(UV) both corrected for dust extinction derived from the Balmer decrement. The dashed line denotes the line of equality, while the dotted lines have a factor-of-2 deviation. The right panel presents the observed SFR(Lyα) as a function of SFR(UV), with no correction for dust extinction. The two dashed lines mark the 1:1 and 1:10 ratios for SFRLyα:SFRUV. The symbols are the same as in Fig. 10.

Open with DEXTER

Another possible source of the Lyα excess is collisional excitation. If the electron temperature is high enough (typically higher than ~2 × 104 K), collisions with thermal electrons lead to excitation and then radiative de-excitation. This process becomes non-negligible compared to the recombination source in the case of high ISM temperatures. Furthermore, it has been suggested that a weak component of Lyα emission can come from ionization by the plasma of very hot X-ray emitting filaments, as Otí-Floranes et al. (2012) observed a spatial correlation between the soft X-ray emission and the diffuse Lyα component in the galaxy Haro 2. Given the low values of fesc(Lyα) that show a significant Lyα enhancement, i.e. fesc(Lyα) < 5% for (Lyα) > 1, one needs a relatively small contribution from the other emission processes to reproduce the deviation from expected attenuation by dust.

6. Lyα as a star formation rate indicator

In view of the important role that the Lyα line plays in the exploration of the distant universe with the upcoming new generation of telescopes, such as the JWST and the ELTs, it is essential to know to what extent the observed Lyα flux is representative of the intrinsic emission. The key capability for detecting galaxies at z > 7 will be the rest-frame UV emission and, in the case of faint galaxies, the strongest emission line Lyα. Typical high-redshift emission-line surveys rely on the strength of the Lyα line, which is usually observed on top of a faint continuum. However, interpretation of this line is hampered by many uncertainties related to complex transmission through the ISM and the IGM.

In Fig. 11, we first compare the SFR derived from Hα with the one based on the UV. Both quantities were corrected for dust extinction, were the stellar extinction affecting the UV continuum was assumed to be a factor of 2 lower than the gas phase extinction derived from the Hα/Hβ ratio (Calzetti et al. 2000). The Hα emission is the result of the recombination of hydrogen atoms that have been previously ionized by the radiation of hot OB stars that have short lifetimes. It therefore traces the quasi instantaneous SFR on the scale of a few Myr. The UV (non ionizing) radiation is also emitted by less massive galaxies with longer life times – typically a few Gyr – and is indicative of the averaged SFR over the galaxy lifetime. We can see that the two SFR indicators are in broad agreement with a significant dispersion. Small differences are actually expected due to different star formation histories traced by these two emissions (cf. Kennicutt 1998; Schaerer 2003).

To assess the reliability of Lyα as tracer of star formation, we now compare the SFR(Lyα) with SFR(UV). The result is presented in the righthand panel of Fig. 11, which shows the observed SFRs derived using the Kennicutt (1998) calibrations, showing the ratios SFRUV/SFRLyα = 1 and 10. We see that the Lyα indicator clearly underestimates the SFR when compared with the UV indicator. This is consistent with the discrepancy found in most of high-z observations (e.g., Yamada et al. 2005; Taniguchi et al. 2005; Gronwall et al. 2007; Tapken et al. 2007; Guaita et al. 2010). We note that the correction of SFR(Lyα) with fesc(Lyα) gives a good estimate of the true SFR since it becomes equivalent to plotting SFR(Hα), given the definition of fesc(Lyα) (cf. Eq. (2)). However, obtaining the Lyα escape fraction for high-z galaxies remains very challenging.

We now compare the deviation of SFR(Lyα) from the true SFR (which is assumed to be SFR(UV)) as a function of SFR(UV). The idea is to know when the observed Lyα flux can be used as an SFR measurement. We plot in Fig. 12 the logarithm of the observed ratio SFRLyα/SFRUV as a function of SFRUV. It is important here to compare our results to the high-z studies available in the literature. For this reason, we included in the plot of Fig. 12 additional samples described in the caption from z ~ 2 to z ~ 6. The yellow region represents the range of values of SFRUV/SFRLyα that would be retrieved for different star formation histories, when the standard calibration from Kennicutt (1998) is used, (This calibration is based on the case of an evolved starburst already in its equilibrium phase. For a young burst or constant star formation that has not reached equilibrium yet (age < 100 Myr), SFR(Lyα) is higher than SFR(UV), and a ratio of SFRLyα/SFRUV ~ 4 can be observed (Schaerer 2003; Verhamme et al. 2008), which is represented by the upper part of the yellow region. For older galaxies (>1 Gyr) with a constant star formation history, SFR(UV) becomes higher than SFR(Lyα), which explains the lower part of the yellow region.

thumbnail Fig. 12

SFR(Lyα) to SFR(UV) ratio as a function of SFR(UV). The color code for the local sample is the same as in previous figures. Several high-z samples are also included: green and blue diamonds (Gronwall et al. 2007; Guaita et al. 2010); yellow, orange, and red stars (Curtis-Lake et al. 2012; Taniguchi et al. 2005; Jiang et al. 2013). The SFRs are observed values not corrected for dust extinction.The dashed line corresponds to the SFR ratio of unity. The yellow region denotes the SFR ratio values that would be derived for different star formation histories (see text for details).

Open with DEXTER

Clearly, the SFR measured from Lyα is, most of the time and at all redshifts, lower than the true SFR. We can observe a general decline of the SFRLyα/SFRUV with increasing SFRUV. As discussed for Fig. 7 describing the decrease in EWLyα as a function of SFRUV, this can be interpreted as a decrease in fesc(Lyα) with increasing UV luminosity, mass, and possibly dust content, but also to the natural decline resulting from ploting 1/x versus x. This trend is predicted by the models, since Garel et al. (2012) find a lower fesc(Lyα) in higher SFR galaxies at z ~ 3−5. An equivalent result is the observed decline of the fraction of LBGs that show Lyα emission (defined as the Lyα fraction xLyα) with increasing UV luminosity (e.g. Stark et al. 2010; Schaerer et al. 2011; Schenker et al. 2012). Quantitatively, the studies find xLyα ~ 50% for MUV = −19 and xLyα ~ 10% for MUV = −21 at 3 < z < 6.

More importantly, we note that this relationship in the different samples is shifted as the redshift increases, from z ~ 0 to z ≳ 6 passing by z ~ 2−3. This can be seen as the result of an increasing Lyα escape fraction as a function of redshift. Indeed, the ratio of SFRLyα over SFRUV represents a proxy for the Lyα escape fraction, assuming constant star formation history. By comparing the observed Lyα luminosity function with the intrinsic one at different redshifts, a clear evolution of fesc(Lyα) with redshift can be seen in Hayes et al. (2011). The study relies on a compilation of Lyα LFs extracted from the literature between z ~ 0 and 8, and integrated over homogeneous limits to obtain a Lyα luminosity density, which is then compared to the intrinsic Lyα luminosity density to obtain the sample-averaged volumetric Lyα escape fraction. The intrinsic Lyα luminosity density is obtained from dust-corrected Hα at z ~ 2.3 and from UV continuum at z > 2.3 where Hα observations are not available. The resulting Lyα escape fraction increases from ≲1% at z ~ 0 to ~40% at z ~ 6, comparable to the redshift range presented in Fig. 12. Similarly, Stark et al. (2010) show that the prevalence of Lyα emitters among LBGs increases with redshift over 3 < z < 6. This trend seems to continue down to z ~ 0.3, where Cowie et al. (2010) show that only 5% of their GALEX-NUV-selected galaxies are Lyα emitters, while they represent ~20% of the z ~ 3 LBG sample of Shapley et al. (2003).

This trend is mainly driven by the decrease in the dust content of galaxies with redshift. The measured E(B − V) in the galaxy samples compiled in Hayes et al. (2011) evolves clearly with redshift. Furthermore, Hayes et al. (2011) combined the fesc(Lyα)–E(B − V) relation with their measured Lyα and UV SFR densities to predict the evolution of dust content with redshift. Independently of the measured E(B − V), they found a clear decrease in dust with increasing redshift up to z ~ 6. Such an evolution has been demonstrated in other studies that measure the UV slope of LBGs at 2 < z < 7 (e.g., Hathi et al. 2008; Bouwens et al. 2009).

Considering the different techniques used to detect high-z galaxies, one should be aware of possible selection biases that could affect this kind of analysis. First, since the evolution of the SFR ratio is attributed to the evolution of fesc(Lyα), we should investigate whether this ratio is affected by other important factors. It has been shown, for instance, that the prevalence of high-EW sources increases with redshift (Ouchi et al. 2008; Atek et al. 2011; Shim et al. 2011), which could produce the observed redshift evolution. Because most of the high-z LAEs are elected in narrow-band surveys, they could miss the low-EW objects that lie below the detection threshold, whereas they will be detected in lower redshift studies that use spectroscopic selection, for instance. However, the typical threshold used by narrow-band surveys is EWLyα = 20 Å, and the distributions of EWLyα at high-z suggest that a maximum of 20% of such galaxies can be missed. Another effect is that we are comparing galaxy samples with different UV luminosities, while there is evidence that Lyα is stronger in lower luminosity galaxies as explained above (Ando et al. 2006; Verhamme et al. 2008; Pentericci et al. 2009; Stark et al. 2010; Schaerer et al. 2011). This is what we see in Fig. 12 where the SFR ratio decreases with increasing SFR(UV) for a given redshift slice. The result is consistent with both theoretical and observational evidence, where galaxies with higher UV luminosities are more obscured by dust and tend to have higher metallicities, masses, and a larger reservoir of neutral gas

(Reddy et al. 2006; Pirzkal et al. 2007; Gawiser et al. 2007; Overzier et al. 2008; Verhamme et al. 2008; Lai et al. 2008), all of which contribute to lowering the Lyα escape fraction and therefore the SFR ratio. The region at the center of the plot where the different samples tend to overlap is most likely symptomatic of the general dispersion of fesc(Lyα) because of the multiparameter nature of the resonant scattering process (Verhamme et al. 2006; Atek et al. 2008, 2009a; Hayes et al. 2010). We here refer the reader to Hayes et al. (2011) for a detailed discussion about the other factors that can affect the Lyα/UV ratio.

As pointed out in Sect. 5, the galaxy sample of the present study is clearly biased toward strong Lyα emitters as opposed to high-z LBGs for instance. This favors galaxies that show high Lyα escape fractions, which could explain the high number of galaxies with recombination ratios above the level expected from dust attenuation when compared to other selection methods (Kornei et al. 2010; Hayes et al. 2011). If anything, this selection bias should work against the anti-correlation observed in Fig. 12 between the SFR ratio and SFR(UV). Another selection effect is that at higher redshift, galaxies will have higher UV luminosity, which makes the increase in SFR ratio as a function of redshift not very obvious at a given UV (or SFR) luminosity. Ideally, we would assemble samples of galaxies at different redshifts within the same UV luminosity bin. Nevertheless, as the redshift increases, the SFR(UV) increases, but so does fesc(Lyα). For this reason we are able to clearly see mean SFRLyα/SFRUV getting closer to unity at high redshift.

7. Conclusion

We analyzed a large sample Lyα emitters at redshift z ~ 0 and z ~ 0.3. We combined our spectroscopic follow-up of z ~ 0.3 galaxies, which was originally detected by Deharveng et al. (2008) with GALEX, with various data obtained from the literature to investigate the influence of several physical parameters on the escape of Lyα emission. After estimating the AGN contamination, we measured the oxygen abundance and the gas-phase extinction, using optical emission line ratios. We also derived the Lyα escape fraction, fesc(Lyα), using the Hα flux and the nebular extinction. We summarize here the main conclusions we draw from comparing these parameters with the Lyα and UV properties.

  • The Lyα escape fraction or equivalent width does not show a correlation with metallicity. While one may expect metal-poor galaxies to show strong Lyα emission as observed in Cowie et al. (2011), our results show that this is not always the case, and they agree with the observations of local blue compact galaxies, such as IZw 18 and SBS 0335.

  • Looking at the Lyα equivalent width as a function of extinction, we do not find the trend commonly observed in high-redshift samples (e.g. Shapley et al. 2003; Pentericci et al. 2009). We explain the absence of correlation to the decoupling of the attenuation affecting the UV continuum and the multiparameter process responsible for the attenuation the Lyα line. In addition, high-redshift studies use a model-dependent extinction that refers to the stellar extinction, which can be very different from the gas-phase dust encountered by Lyα photons.

  • The Lyα escape fraction presents a clear correlation with the dust extinction, confirming earlier results (e.g. Atek et al. 2009a; Kornei et al. 2010; Hayes et al. 2011), albeit with a large dispersion indicating that dust attenuation is not the only regulatory factor of Lyα emission. A two-parameter fit of fesc(Lyα)− E(B − V) yields an extinction coefficient of kLyα ~ 6.67 ± 0.52 and an intercept point of CLyα = 0.22 ± 0.03. We see that even when no dust is present, the point of fesc(Lyα) = 1 is only an upper limit and that Lyα emission can be attenuated by the resonant scattering into neutral gas.

  • We also introduce the normalized fesc(Lyα), which corresponds to the deviation of the observed fesc(Lyα) from what is expected from the case of pure dust attenuation. At low extinction, Lyα appears more attenuated than the UV continuum, most likely because of the scatting process. At higher extinction, fesc(Lyα) show values above the predictions for standard dust attenuation. This can be the results of various mechanisms, including an enhancement of Lyα thanks to the ISM geometry (Neufeld 1991; Laursen et al. 2013; Verhamme et al. 2012), a different extinction law (Scarlata et al. 2009), a significant contribution from collisional excitation, or ionization by a hot plasma (Otí-Floranes et al. 2012).

  • To assess the reliability of the Lyα emission line as a star formation indicator, we compared SFR(Lyα) and SFR(UV), which have the advantage of being independent of the extinction. Overall, Lyα tend to underestimate the SFR by a factor up to 10. We observe that the SFRLyα/SFRUV ratio decreases with increasing SFRUV, which can be interpreted as a decrease in fesc(Lyα) as a function of UV luminosity as observed in several studies. We also note that this trend is shifted with increasing redshift because of the redshift dependence of the Lyα escape fraction.

Acknowledgments

We thank the anonymous referee for useful suggestions that improved the clarity of the paper, Anne Verhamme for interesting discussions, and Len Cowie for providing us with his emission line measurements. H.A. and D.K. acknowledge support from the Centre National d’Etudes Spatiales (CNES). H.A. and J.P.K. are supported by the European Research Council (ERC) advanced grant “Light on the Dark” (LIDA). J.M.M.H. is funded by Spanish MINECO grants AYA2010-21887-C04-02 (ESTALLIDOS) and AYA2012-39362-C02-01. This work is based on observations made with ESO Telescopes at La Silla Observatories under program ID 082.B-0392.

References

Online material

thumbnail Fig. 13

Same as Fig. 1.

Open with DEXTER

Table 1

Target information and emission line measurements for the GALEX sample.

Table 2

Same as Table 1 for the IUE sample.

All Tables

Table 3

Extinction factors at λ = 1216 Å, calculated for different values of E(B − V) using the Galactic extinction (Cardelli et al. 1989) and the SMC law (Prevot et al. 1984).

Table 1

Target information and emission line measurements for the GALEX sample.

Table 2

Same as Table 1 for the IUE sample.

All Figures

thumbnail Fig. 1

Examples of the rest-frame optical spectra of our sample illustrating the quality of the data. The flux density units are the same for all spectra, although the scales are different. The target name is marked on top of each spectrum.

Open with DEXTER
In the text
thumbnail Fig. 2

Example of rest-frame UV and optical spectra of the IUE sample. Most of the FUV spectra cover 1200−2000 Å wavelength range and the optical spectra the 3200−7000 Å range. The gap in-between is set to zero.

Open with DEXTER
In the text
thumbnail Fig. 3

BPT diagrams used to classify narrow emission-line galaxies. GALEX Lyα emitters are represented with black circles with corresponding error bars, together with an IUE sample of local starburst galaxies in blue squares. The left panel shows [Oiii] λ5007/Hβ versus [Nii] λ6584/Hα ratios. The dashed and solid curves are theoretical boundaries separating starbursts and AGN objects assuming an instantaneous burst (Dopita et al. 2000) and an extended star formation episode (Kewley et al. 2001), respectively. The dotted curve is the “Kauffmann line” (Kauffmann et al. 2003). The right panel represents the [Oiii] λ5007/Hβ versus [Sii] (λ6717 + 6732)/Hα diagnostic with the same theoretical models as in the right panel. The red dot-dashed line shows typical model uncertainties of ± 0.1 dex. The diagram includes all the 11 IUE local galaxies, among which is an object previously classified as an AGN (red square) that we decided to show as an example.

Open with DEXTER
In the text
thumbnail Fig. 4

Comparison between the various oxygen-abundance-derived indicators: R23 = ([Oii] λ3727 + [Oiii] λλ 4959, 5007)/Hβ (Pagel et al. 1979), N2 = log ([Nii]λ6583/Hα) and O3N2 = log {([Oiii] λ5007/Hβ)/([Nii] λ 6583/Hα)} (Pettini & Pagel 2004). The z ~ 0.3 galaxies are shown with black circles and the local galaxies with blue squares with 1σ error bars. The solid line represents the 1:1 ratio with ±0.1 dex uncertainties (dashed lines). Some galaxies are not present in all the plots because the corresponding diagnostic lines were not detected with sufficient significance.

Open with DEXTER
In the text
thumbnail Fig. 5

Lyα equivalent width as a function of metallicity. The oxygen abundance is derived from the N2 index log([Nii]λ6583/Hα) using the equation of Pettini & Pagel (2004). The present sample is shown with black circles, Cowie et al. (2011) galaxies with magenta circles, and IUE galaxies with blue triangles.

Open with DEXTER
In the text
thumbnail Fig. 6

Lyα equivalent width as a function of the gas-phase dust extinction. Black circles are the GALEX Lyα emitters, and green and red circles are the other GALEX samples of Scarlata et al. (2009) and Cowie et al. (2011), respectively. The blue triangles represent the IUE spectroscopic sample, and purple squares local starbursts from Atek et al. (2008).

Open with DEXTER
In the text
thumbnail Fig. 7

EWLyα as a function of SFRUV. The color code is the same as in Fig. 6. The Lyα equivalent width is the observed rest-frame value, and the SFRUV values are calculated from FUV luminosity at 1530 Å not corrected for extinction. EWLyα uncertainties are 10% error bars, and SFRUV errors are derived from GALEX FUV photometry.

Open with DEXTER
In the text
thumbnail Fig. 8

Lyα equivalent width distribution for the three GALEX samples used in this work. It includes all sources with optical follow-up spectroscopy.

Open with DEXTER
In the text
thumbnail Fig. 9

Lyα escape fraction as a function of the gas-phase metallicity. The oxygen abundance is derived from the N2 index log([Nii]λ6583/Hα) using the equation of Pettini & Pagel (2004). The color code as the same as in Fig. 5.

Open with DEXTER
In the text
thumbnail Fig. 10

Lyα escape fraction as a function of the nebular dust extinction. The figure shows the z ~ 0.3 Lyα emitters of this work with black circles. We also derived fesc(Lyα) for other galaxy samples: z ~ 0 IUE sample represented by blue triangles, z ~ 0.3 sample of Scarlata et al. (2009) with green circles, Cowie et al. (2011) sample with red circles, z ~ 0 galaxies of Atek et al. (2008) with purple squares. Note that the negative values of E(B − V) are shown in the plot but were assigned E(B − V) = 0 before fesc(Lyα) calculation or fitting the fescE(B − V) relationship. The left panel shows fesc(Lyα) as a function of E(B − V) assuming a Cardelli et al. (1989) extinction law. The solid black line denotes the best 2-parameter fit to the relationship with both the slope and the intercept as free parameters. The yellow region covers the 1σ uncertainties of the fit derived from MC simulations. The dashed red line is the expected attenuation law at the Lyα wavelength. In the right panel we plot the normalized escape fraction fesc,rel(Lyα) = log [fesc(Lyα)/fesc(cont)] as a function of extinction. The normalized escape fraction represents the deviation of fesc(Lyα) from the classical dust attenuation law at 1216 Å. The color code for the samples is the same as in the left panel.

Open with DEXTER
In the text
thumbnail Fig. 11

Comparison between SFR indicators. The SFR values are calculated using Kennicutt (1998) calibration. The left panel shows SFR(Hα) versus SFR(UV) both corrected for dust extinction derived from the Balmer decrement. The dashed line denotes the line of equality, while the dotted lines have a factor-of-2 deviation. The right panel presents the observed SFR(Lyα) as a function of SFR(UV), with no correction for dust extinction. The two dashed lines mark the 1:1 and 1:10 ratios for SFRLyα:SFRUV. The symbols are the same as in Fig. 10.

Open with DEXTER
In the text
thumbnail Fig. 12

SFR(Lyα) to SFR(UV) ratio as a function of SFR(UV). The color code for the local sample is the same as in previous figures. Several high-z samples are also included: green and blue diamonds (Gronwall et al. 2007; Guaita et al. 2010); yellow, orange, and red stars (Curtis-Lake et al. 2012; Taniguchi et al. 2005; Jiang et al. 2013). The SFRs are observed values not corrected for dust extinction.The dashed line corresponds to the SFR ratio of unity. The yellow region denotes the SFR ratio values that would be derived for different star formation histories (see text for details).

Open with DEXTER
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.