Ly alpha emitting galaxies as early stages in galaxy formation
We present optical spectroscopy of two samples of GALEX grism selected Ly emitters (LAEs): one at and the other at . We have also observed a comparison sample of galaxies in the same redshift intervals with the same UV magnitude distributions but with no detected Ly. We use the optical spectroscopy to eliminate active galactic nuclei (AGNs) and to obtain the optical emission-line properties of the samples. We compare the luminosities of the LAEs in the two redshift intervals and show that there is dramatic evolution in the maximum Ly luminosity over . Focusing on the samples alone, we show that there are tightly defined relations between all of the galaxy parameters and the rest-frame equivalent width (EW) of H. The higher EW(H) sources all have lower metallicities, bluer colors, smaller sizes, and less extinction, consistent with their being in the early stages of the galaxy formation process. We find that of the LAEs have EW(H) Å, and, conversely, that of galaxies with EW(H) Å are LAEs. We correct the broadband magnitudes for the emission-line contributions and use spectral synthesis fits to estimate the ages of the galaxies. We find a median age of yr for the LAE sample and yr for the UV-continuum sample without detected Ly. The median metallicity of the LAE sample is (O/H)=8.24, or about 0.4 dex lower than the UV-continuum sample.
Ly emission-line searches have been widely used to find high-redshift galaxies and, for the highest redshift galaxies, this line is the only spectroscopic signature that can be used to confirm the redshift of a galaxy selected on the basis of its color properties. However, Ly is a difficult line to interpret. Because the line is resonantly scattered by neutral hydrogen, determining its escape path and hence its dust destruction is an extremely complex problem, both theoretically (e.g., Neufeld 1991; Finkelstein et al. 2007) and observationally (e.g., Kunth et al. 2003; Schaerer & Verhamme 2008; Östlin et al. 2009). Thus, while we have empirical measurements that a significant fraction of UV-continuum selected samples have Ly lines with rest-frame equivalent widths above 20 Å over a wide range of redshifts from to (Shapley et al. 2003; Cowie et al. 2010, 2011; Stark et al. 2010), our understanding of what determines this fraction is still weak. In particular, we would like to know whether the presence of Ly emission is related to other properties of the galaxy, such as its metallicity, extinction, morphology, or kinematics, and how the Ly line escapes.
Due to the difficulty with observing in the UV, we currently have much more information on the Ly emitters (LAEs) and how their properties relate to those of other UV selected galaxies at these redshifts (e.g., Shapley et al. 2003; Reddy et al. 2010; Kornei et al 2010) than we do on the local samples. However, there has been considerable controversy in the interpretation of these high-redshift observations. The simplest interpretation is that the LAEs are younger, lower mass, and metal poor, representing early stages in galaxy evolution (e.g., Hu et al. 1998; Nilsson et al. 2007; Gawiser et al. 2007). However, other authors have argued that LAEs arise in relatively massive galaxies (e.g., Lai et al. 2008; Finkelstein et al. 2009c) with ages of around a Gyr, that young galaxies have weaker Ly than old galaxies (Shapley et al. 2001), and, more recently, that LAEs are older, less dusty, and in a later stage of galaxy evolution than sources with weaker Ly emission (Kornei et al. 2010).
The Ly signature can be produced by a range of galaxy types and masses, including even the most ultraluminous infrared galaxies (ULIRGs) (e.g., Chapman et al. 2005; Nilsson & Møller 2009), so some level of heterogeneity must be expected. However, all of the results that argue for the LAEs being predominantly old are based on spectral synthesis fitting, and most of the old ages inferred are almost certainly mis-estimates arising from the presence of very strong optical emission lines in the LAEs (e.g., Schaerer & deBarros 2009).
Until recently the only low-redshift Ly emitting sources that could be studied in detail were the local blue compact galaxies. However, these generally have much lower Ly luminosities than the high-redshift LAEs, and, while some of the blue compact galaxies have been studied in exquisite detail (e.g., Östlin et al. 2009) on an individual basis, it has not been easy to form large, uniformly selected samples that can be statistically analyzed. Thus, the recent determination that substantial samples of LAEs can be found (Deharveng et al. 2008) with the Galaxy Evolution Explorer (GALEX) (Martin et al. 2005) grism spectrographs has enabled a new approach to the subject (Atek et al. 2009a; Finkelstein et al. 2009a, 2009b; Scarlata et al. 2009; Cowie et al. 2010).
The low-redshift LAE samples have many advantages. The galaxies are bright and can be easily studied at other wavelengths, but perhaps even more importantly, they can be integrated into comprehensive studies of galaxies at the same redshifts to understand the selection biases intrinsic to the samples. Early papers on GALEX LAEs worked with relatively small samples, but the general conclusions are that low-redshift LAEs are somewhat heterogeneous yet more weighted to low metallicities and extinctions and more likely to be small, compact galaxies when compared to UV-continuum selected galaxies without detected Ly with the same luminosities in the same redshift interval.
Here we study larger and more optically spectroscopically complete samples of LAEs at and , together with comparison samples of UV-continuum selected galaxies without detected Ly in the same redshift intervals. In Section 2 we present our optical spectroscopy of all the samples obtained with the DEep Imaging Multi-Object Spectrograph (DEIMOS; Faber et al. 2003) on the Keck II 10 m telescope. In Section 3 we use the optical data to remove the small number of active galactic nuclei (AGNs) that were not previously identified from the UV spectra and then provide our final sample of candidate LAE galaxies in the two redshift intervals. We also use the spectra to measure the optical line fluxes and equivalent widths and to determine the metallicities of the galaxies. In Section 4 we determine how the Ly luminosity evolves with redshift. In Section 5 we consider the overall properties of the LAEs and the UV-continuum selected galaxies without detected Ly. Since the sample of LAEs is small (only eight objects) we only consider the sample in this section. We use the spectra to determine the emission-line contributions to the broadband fluxes and to show that these corrections must be included if spectral synthesis fitting is to give accurate ages for the youngest ( yr) galaxies. In Section 6 we interpret the results in terms of a constant star formation rate (SFR) model, comparing SFRs derived from the H, UV-continuum, and 20 cm observations and discussing the limits of validity for the UV-continuum determined SFRs. We also interpret the metallicity evolution. We summarize our results in Section 7. We use a standard = 70 km s Mpc, = 0.3, = 0.7 cosmology.
2 The Galex Sample
For the present study we use the LAE samples with rest-frame EW(Ly) Å from Cowie et al. (2010; their Tables 15 and 16), chosen from the nine blank high galactic latitude fields with the deepest GALEX grism spectroscopic observations (Morrisey et al. 2007). The area covered is just over 8 deg. The samples consist of sources whose grism UV spectra have a detectable single emission line, which we assume to be Ly. In order to eliminate sources that are clearly AGNs based on their UV spectra, we do not include any sources that have high-ionization lines, and we require the FWHM of the Ly lines to be less than 15 Å in the GALEX FUV spectra () or less than 30 Å in the GALEX NUV spectra (). In the redshift interval we also eliminate sources that do not show the break between the FUV and NUV bands. Such a break would be expected to be produced by the Lyman continuum edge. (See Section 4 of Cowie et al. 2010 for details on constructing the samples.) For the CDFS 00 field we have added additional sources from a deeper grism spectroscopic exposure than was used in Cowie et al. (2010). The full Ly selected sample in the CDFS 00 is given in the Appendix (i.e., this is an update of Table 4 in Cowie et al. 2010, but it also includes the optical redshifts, where available; see Section 2.2).
We summarize the final candidate LAE galaxy sample in the redshift interval in Table 1, sorted by the rest-frame EW(Ly). Through most of the paper we restrict our analysis to sources with rest-frame EW(Ly Å, so we list these sources in the main body of the table. However, for completeness, we also give the results for the lower EW(Ly) sources in a supplement to the table. For each source in Table 1 we give the GALEX name, the J2000 right ascension and declination, the redshift from the GALEX UV spectrum, the logarithm of the bolometric continuum luminosity, , above 1216 Å, the logarithm of the Ly luminosity, the rest-frame EW(Ly), the optical ground-based redshift, the rest-frame EW(H), the [NII]/H, [OIII]/H, and [OIII]/H ratios, and the logarithm of the H and H fluxes for the sources with optical continuum magnitudes in the Sloan Digital Sky Survey (SDSS) (see Section 2.1).
We summarize the final candidate LAE galaxy sample in the redshift interval in Table 2, also sorted by the rest-frame EW(Ly). There are 8 sources in this sample. We list the properties of these sources using the same format as in Table 1, except here we give the rest-frame EW(H) in place of the rest-frame EW(H), and we do not include the column for the H and H fluxes. We also do not split the sample into a main body of the table and a supplement to the table, since there are so few sources. Six of the eight sources satisfy the EW(Ly Å criterion.
Although the sample is small, it is important in that it demonstrates the presence of these LAEs, and it allows us to make a first estimate of the luminosity function at these redshifts. However, most of our analysis in this paper will focus on the sample. For each source with we computed the UV spectral index from the GALEX spectra using the Calzetti et al. (1994) prescription, which fits a power law to the spectrum in a set of wavelength windows chosen to minimize the effects of absorption and emission lines. Since we have only short wavelength data in the rest frame, we corrected our measured index with the offset of given by Meurer et al. (1999) to provide our final index. We did not compute the UV spectral index for the sources with , since at those redshifts many of the shortest wavelength windows in Calzetti et al. (1994) lie in the noisy regions between the GALEX FUV and NUV spectra.
2.1 Imaging Data
We compiled ancillary data from various archival sources. Just over of the area is covered by deep -band images obtained with the MegaPrime camera on the 3.6 m Canada France Hawaii Telescope (CFHT). We took the reduced images from the CADC pipeline reduction, which gives the 5 AB magnitude limits. These range from 26.5 to 27.5. We used these images to measure the -band magnitudes and the FWHM sizes of the galaxies at this wavelength. For the fields covered by the SDSS we also compiled the and SDSS model C magnitudes ( AB magnitude limits of 22.0, 22.2, 22.2, 21.3, and 20.5, respectively) from the DR6 release (Adelman-McCarthy et al. 2008). These cover roughly half of the observed targets. For the fields with publicly available 20 cm images (COSMOS 00: Schinnerer et al. 2007, of Jy beam; SIRTFFL 01: Condon et al. 2003, of Jy beam; HDFN 00: Morrison et al. 2010, ranges from Jy beam near the center to Jy beam at ), we measured the 20 cm fluxes from the images following the usual methods (e.g., Morrison et al. 2010). Finally, for the CDFS 00 targets covered by the Galaxy Evolution from Morphologies and SEDs (GEMS) survey (Rix et al. 2004), we compiled thumbnails of the galaxies in the F606W ( AB magnitude limit of 28.25) and F850LP ( AB magnitude limit of 27.10) filters from the Hubble Space Telescope (HST) observations with the Advanced Camera for Surveys (ACS).
2.2 Optical Spectroscopy
We obtained optical spectroscopy for a large fraction of the two GALEX LAE samples using the DEIMOS spectrograph on Keck II with the ZD600 /mm grating in observing runs throughout 2009 and 2010. The hr spectra have a resolution of 4.5 Å and a wavelength range of approximately 5000 Å centered at 6800 Å. More details of the observing and reduction procedures can be found in Cowie et al. (1996, 2010).
We observed 77 of the 91 LAEs in Table 1 (this includes the supplement) and all 8 LAEs in Table 2. We give the optical redshifts in Tables 1 and 2, where they can be compared with the UV redshifts from the GALEX data. For the 77 LAEs with optical redshifts, we confirmed the UV redshifts for 70. In Table 1 we show the optical redshifts in parentheses for the sources where we did not confirm the UV redshifts. We eliminated these sources from further consideration. We confirmed the UV redshifts for all 8 LAEs with the optical redshifts. We show the UV and optical spectra for these sources in Figures 2.2 and 2.2.
In order to understand how the GALEX LAEs relate to other UV-continuum selected sources with the same UV-continuum magnitude distributions, we also observed at optical wavelengths sources randomly chosen from the GALEX UV spectroscopic sample that did not have UV spectral line identifications. The latter is not a complete sample of all such sources in the field, but rather a randomly chosen subsample with an NUV magnitude selection function that is similar to that for the sources with strong Ly emission. Thus, it may be directly compared to the LAE sample.
We have obtained optical spectra for 450 of the 5642 objects in the total UV-continuum selected sample without detected Ly. 114 of these sources lie in the redshift interval , and 5 lie in the redshift interval . We will hereafter refer to these sources as our UV-continuum sample. These sources may be viewed as analogs of high-redshift Lyman break galaxies (LBGs) without strong Ly emission and can be combined with the LAE sample to understand how Ly galaxies are related to the more general UV-continuum selected population. In order to compare the numbers of UV-continuum sources with a given property with the LAE sample we need to correct for the fraction of UV-continuum sources that were observed. There is a weak dependence of the observed fraction on the NUV magnitude, ranging from observed at NUV=20 to observed at NUV=22. We use the inverse of this fraction as a function of the NUV magnitude to provide the weighting to the UV-continuum sample.
The new optical spectroscopic data presented in this paper more than doubles the optical spectroscopic data used in Cowie et al. (2010). There we observed with DEIMOS (including four literature redshifts) 34 of the 62 LAEs given in their Table 15 (this also includes EW(Ly Å sources) and 1 of the 4 LAEs given in their Table 16. We also observed with DEIMOS 124 UV-continuum selected sources without detected Ly, of which 46 were in the redshift interval and none were in the redshift interval .
2.3 Line Fluxes
Most of the spectra were obtained at a near-parallactic angle to minimize atmospheric refraction effects. However, in some cases the mask configuration did not allow for this. In addition, a small number of the spectra were obtained during periods of varying extinction. All of the spectra were relatively calibrated using the measured response from observed calibration stars. However, relative slit losses always pose a problem, and special care must be taken for the flux calibration.
For each spectrum we fitted all of the emission lines using the IDL MPFIT program of Markwardt (2009). We used simultaneous Gaussian fits to neighboring lines together with a linear fit to the continuum baseline. For weaker lines we held the full width constant, using the value measured in the stronger lines, and set the central wavelength to the measured redshift. We also measured the noise as a function of wavelength by fitting to random positions in neighboring portions of the spectrum and computing the dispersion in the results.
For the AGN classifications and the metallicity measurements we used only the ratios of the emission lines that are close in wavelength. These line flux ratios can be robustly measured from the spectra without even the relative flux calibration, since for these neighboring lines the DEIMOS response is essentially constant. For the primary lines used in our analysis we give the line ratios and their errors in Tables 1 and 2.
In order to measure the absolute H and H line fluxes, we used the measured EW(H) and EW(H) in combination with the measured continuum fluxes at the appropriate wavelength. We first determined which SDSS filter the emission line lay in and then integrated through the transmission of this filter to obtain the averaged spectral flux. We then renormalized the spectrum to match the corresponding SDSS model C magnitude from the DR6 release (Adelman-McCarthy et al. 2008). We then determined the continuum flux and multiplied this by the observed-frame EW to obtain the corresponding line flux. This is an approximation, since it assumes that the measured EW is representative of the value averaged over the total light of the galaxy, including regions outside the slit. However, for the photometric cases we derive very similar values directly from the calibrated spectra, including calibrated spectra that we obtained using larger () slit widths for a small subset of the galaxies. This suggests the procedure is relatively robust.
However, the relative apertures used in measuring the GALEX Ly fluxes and the optical emission line fluxes may introduce systematic variations reflecting the relative geometries of the emission. Thus, the (Ly)/(H) ratios are more reliably obtained in large () aperture measurements, such as those used in Atek et al. (2009a). In the present paper we are primarily concerned with determining the optical continuum and emission-line properties that result in a galaxy showing a strong Ly emission line and not so much with the relative fluxes of the optical and Ly emission lines, making this less of a concern. In the discussion we will note where aperture effects might play a role.
In order to check the calibrations and aperture effects, we tested our measured fluxes against the independent measurements by Scarlata et al. (2009) for the subset of galaxies that overlap with our sample. The Scarlata et al. data were obtained with a larger slit. For eight cases they compared their results with those made in a slit, finding only small differences between the two measurements. Scarlata et al.’s procedure for measuring the fluxes relied on the relative accuracy of the spectrophotometry, using the magnitude from the SDSS data to set the absolute flux normalization. This differs from our procedure, where we normalized each region of the spectrum separately to the SDSS magnitudes. We compare the two results in Figure 2.3, where we show both the (Ly)/(H) (red squares) and the (H)/(H) ratios (blue diamonds). The overall agreement is very good, but in a small number of cases the disagreement is much larger than the statistical error in the (Ly)/(H) ratio. In the most extreme case the ratios differ by a multiplicative factor of just over two. We take this to be an estimate of the potential systematic errors.
|GALEX0333-2757||53.496414||-27.966471||0.3630||43.66||42.21||7112||0.3578(B)||261.||0.9560.015||5.87 0.11||0.417 0.37||,|
|GALEX0332-2810||53.001919||-28.182611||0.2795||43.12||41.76||559||0.2775||1103.||0.0710.028||3.04 0.14||0.480 0.41||,|
|GALEX1419+5315||214.794739||53.265835||0.2651||43.37||41.85||486||0.2637||111.||0.4030.035||1.84 0.31||-14.4, -15.3|
|GALEX1436+3459||219.100540||34.993530||0.2148||43.27||41.73||405||0.2131||601.||0.1210.018||3.76 0.13||0.687 0.34||-15.5,|
|GALEX0334-2752||53.711498||-27.876585||0.3371||43.54||42.11||395||0.3336(B)||6.1.||0.655 0.14||1.72 0.31||,|
|GALEX1436+3456||219.092041||34.942139||0.2716||43.62||41.91||273||0.2684||672.||0.1560.038||2.52 0.11||-14.8, -15.3|
|GALEX1418+5218||214.593506||52.306747||0.2392||43.19||41.44||264||0.2388||1702.||0.0380.012||4.410.064||0.213 0.14||-14.8, -15.3|
|GALEX1423+5244||215.842850||52.742474||0.2795||43.29||41.46||195||0.2771||392.||0.2970.037||0.853 0.11||-15.1, -15.6|
|GALEX1421+5224||215.477295||52.406723||0.3515||44.05||42.12||192||0.3454||26.5||0.5580.011||0.3260.068||0.421 0.17||-14.2, -14.9|
|GALEX1419+5223||214.811249||52.390808||0.2565||43.14||41.35||194||0.2541(A)||765.||0.1590.067||3.37 0.69||-14.1, -14.7|
|GALEX1418+5217||214.701202||52.299168||0.2421||43.31||41.42||184||0.2398||462.||0.0620.033||3.80 0.22||-15.1, -15.9|
|GALEX1717+5929||259.379578||59.487473||0.2997||43.91||41.99||173||0.2988||341.||0.4230.024||1.09 0.10||-14.3, -15.0|
|GALEX1420+5243||215.180359||52.718750||0.2508||43.45||41.63||172||0.2470||702.||4.83 0.10||-15.1, -15.6|
|GALEX1420+5250||215.186218||52.835056||0.2536||43.39||41.58||172||0.2519||662.||0.1680.038||2.38 0.11||-14.8, -15.1|
|GALEX1420+5250||215.186218||52.835056||0.2536||43.39||41.58||172||0.2519||662.||0.1680.038||2.38 0.11||-14.8, -15.1|
|GALEX1436+3535||219.243362||35.597694||0.2077||43.36||41.40||153||0.2055||141.||0.4530.029||0.531 0.16||-14.4, -15.1|
|GALEX1420+5313||215.237045||53.218750||0.7488||44.53||42.84||27 3||0.7442(A)||6||0.484 0.16|
3 AGN-Galaxy Discrimination
We might expect that sources classified as AGNs based on the presence of high-excitation lines or very wide lines in the UV spectra are truly AGNs. However, the converse is not true, which means our candidate LAE galaxies (Tables 1 and 2) may still have some degree of AGN contamination. In some cases the high-excitation lines fall at problematic wavelengths in the UV spectra, and in other cases these lines may be intrinsically weak. Optical spectroscopic data support these points: follow-ups of GALEX candidate LAE galaxy samples have shown that a significant fraction () have substantial AGN contributions (Finkelstein et al. 2009b; Scarlata et al. 2009; Cowie et al. 2010), though there is wide variation in the estimates of the precise degree of contamination reflecting the relatively small sample sizes and the differences in the classification procedures. (Finkelstein et al. 2009b find a higher AGN fraction in their sample than other groups, so we make a detailed comparison with their results in this section.)
We first searched our samples for optical spectra that showed broad emission lines. None of the spectra are quasars or Seyfert 1s, a consequence of this class of source being easily picked out with the UV spectra (Barger & Cowie 2010). However, three of the sources in the sample are intermediate-type Seyferts. In the optical redshift column (col. 8) of Table 1 we have labeled these sources as Sy1.5 or Sy1.8. We eliminate these sources from further consideration.
We next constructed the BPT diagram (Baldwin et al. 1981) for the sample. This diagram uses the [NII]/H and the [OIII]/H ratios to separate AGNs from star-forming galaxies. H is off the spectra in the sample, and we cannot run this test there. In addition, only sources with strong optical emission lines can be classified in this way. We restricted to galaxies that had either [NII] or H detected with a signal-to-noise above 5 and either [OIII] or H detected with a signal-to-noise above 5.
The BPT diagrams for the (a) LAE and (b) UV-continuum samples having strong optical emission lines defined in this way are shown in Figures 3(a) and 3(b), respectively. The dotted curve traces the Kewley et al. (2001) division between galaxies whose extreme UV ionizing radiation field is dominated by an AGN () and those dominated by star formation. It is a theoretical curve based on photoionization models for giant HII regions and a range of stellar population synthesis codes. The dashed curve traces the Kauffmann et al. (2003) division between AGNs and star formers based on an empirical separation of SDSS galaxies. As mentioned above, we have eliminated the intermediate-type Seyferts in Table 1 from Figure 3(a), but the BPT diagram suggests that a further four sources are also AGNs. We mark these with open black squares in Figure 3(a) and label them as “B” in the optical redshift column (col. 8) of Table 1. Another four sources have weak [OIII] and H and do not appear on the diagram, but they have strong ([NII]/H ratios, which suggests that they are also AGNs. We also label these sources as “B” in column 8 of Table 1. Most of the remaining sources in Figure 3(a) clearly lie along the star-forming galaxy track, as do all the UV-continuum sources in Figure 3(b), though there are a small number of sources that lie in positions where they may have mixed star formation and AGN contributions. None of the results in the paper are changed if we exclude these objects.
There remain 8 sources with optical spectra in the LAE sample where the [OIII] and H lines are too weak to place them on the BPT diagram and where the [NII]/H ratios are not high. These galaxies are labeled as “A” in the optical redshift column (col. 8) of Table 1 for their weak optical emission lines. In one case the optical spectrum is poor, but in the remaining cases the sources either have more normal [NII]/H ratios or are absorbers. It is possible that these also contain AGNs. The percentage of AGNs in the sample is if we include only the classified AGNs, but that percentage would rise to if the 8 weak emission-line sources were also AGNs.
Finkelstein et al. (2009b) optically classified 23 sources from the Deharveng et al. (2008) sample in the Groth strip. Fourteen of these overlap with sources in our Table 1. The remaining 9 do not, either because the source lies outside the region of the field we used or outside the redshift range we used, or because we found the Ly identification to be dubious, or because we classified the source as an AGN based on the UV spectrum. Of the 14 sources in common, the classifications agree for 11 (3 AGNs and 8 star formers). We could not classify GALEX1420+5306, the first of the remaining 3 overlapped sources, as it has very weak emission lines. Finkelstein et al. (2009b) found this source, which they called EGS1, to be an AGN based on its [NII]/H ratio. As discussed in Cowie et al. (2010), we believe the second of the remaining 3 overlapped sources, GALEX1417+5224 (EGS2), is a high-excitation, very low metallicity galaxy rather than an AGN. Finally, we find that the third source, GALEX1421+5239 (EGS14), lies on the star-forming galaxy track in the BPT diagram. We do not detect high-ionization lines. This is in contrast to Finkelstein et al. (2009b), who classified the source as an AGN on the basis of both its position on the BPT diagram and the presence of high-ionization lines in its spectrum.
All eight of the LAEs in Table 2 have been optically observed (Figures 2.2 and 2.2), and in all cases the Ly identification is confirmed with the optical redshift. Seven of these have strong emission lines, and none of them shows [NeV] emission or broad MgII emission that might classify the source as an AGN. The remaining source (GALEX1420+5313; see Figure 2.2(e)) has a post-starburst spectrum. (This galaxy is also identified in the DR3 release of the DEEP2 survey; Davis et al. 2007.) We label it as “A” in column 8 of Table 2 for its weak optical emission lines. It is possible that the Ly emission is AGN-dominated in this galaxy. We therefore exclude this source from the LAE sample.
Throughout our subsequent analysis we restrict to the LAEs that are confirmed as star formers from the optical spectra. We eliminate all sources with AGN signatures in either the optical or UV (including those sources identified as AGNs based on the BPT diagram), and we also eliminate the optically unobserved galaxies, the unconfirmed galaxies, and the galaxies with weak optical emission lines. This leaves a sample of 40 LAEs with rest-frane EW(Ly Å in the interval and a sample of 5 LAEs with EW(Ly Å in the interval.
In Figure 3 we compare the average UV spectrum of the 40 LAEs (red spectrum) with that of the UV-continuum sample in the same redshift interval (blue spectrum). In order to make the most direct comparison, we eliminate weak emission-line galaxies with rest-frame EW(H) Å in the UV-continuum sample, since we have eliminated these sources from the LAE sample. We also eliminate two objects which are classified as BPT AGN in the UV-continuum sample. This leaves 101 UV-continuum galaxies. However, including the remaining galaxies makes no difference to the results.
The NUV normalizations are almost identical since the samples are chosen with the same distribution of NUV magnitudes. The shapes are also nearly indistinguishable above Å. By selection the LAE sample has strong emission, and the average spectrum for this sample has a rest-frame EW(Ly Å. While a a stacking procedure is not the best way to analyze objects with a mixture of Ly emission and absorption, it at least allows us to see that the UV-continuum sample has Ly emission and that its stacked spectrum is flatter than that of the LAE sample below Å. Fitting the Ly emission, we find a rest-frame EW(Ly Å for the emission feature. This is comparable to, but slightly higher than, the median value of EW(Ly Å found for the LBG population by Kornei et al. (2010). The LBG sample includes the LAEs as a subsample; thus, if anything, our sample, which excludes the LAEs, should be lower.
4 LAE Luminosity evolution
A rest-frame EW(Ly Å is normally used to define the high-redshift LAE population (e.g., Hu et al. 1998), so in this section we apply that criterion. In Figure 4 we plot Ly luminosity versus redshift and compare it with the measured luminosities of high-redshift LAE samples taken from Shimasaku et al. (2006), Ouchi et al. (2008), and Hu et al. (2010), all of which were chosen with the same EW selection criterion.
The primary conclusion that we draw from Figure 4 is that there is a dramatic change in the properties of the LAEs between and . There are no sources in the sample with luminosities close to those of the brightest LAEs at high redshifts, but such sources are present by . This is not simply a volume effect: Cowie et al. (2010) determined the LAE luminosity function in the redshift interval and showed that the for a Schechter (1976) function fit with (red dashed line) is a factor of 8 fainter than that measured by Gronwall et al. (2007) at and more than an order of magnitude fainter than that measured by Hu et al. (2010) at (blue solid line).
We have computed the Ly luminosity function for the observed sample using the technique (Felten 1976) with the accessible volumes calculated from the areas versus NUV magnitudes in the sample. We show the luminosity function with the open black squares in Figure 4. As discussed extensively in Deharveng et al. (2008) and Cowie et al. (2010), this Ly luminosity function corresponds to sources with NUV and must be corrected for LAEs at fainter continuum magnitudes. In the redshift interval most sources with Ly luminosities in the erg s range lie at NUV and the corrections are small (Cowie et al. 2010). In contrast, many of the LAEs wth Ly luminosities erg s at are fainter than NUV and the correction is substantial. Following the procedures outlined in Deharveng et al. (2008) and Cowie et al. (2010), we have computed the correction by assuming the NUV counts derived from the luminosity functions of Arnouts et al. (2005) and the shape of the equivalent width distribution derived in Cowie et al. (2010) for the sample. (This shape is almost identical to that seen at by Shapley et al. 2003.) We show the corrected points with the solid squares in Figure 4.
We compare with the Ly luminosity functions at (blue curve), (red curve), (green solid curve), and (green dashed curve) using the Schechter (1976) function fits given in Cowie et al. (2010), Gronwall et al. (2007), and Hu et al. (2010). Even allowing for the small numbers in the present sample and the substantial corrections, we can see that there is a dramatic evolution in the luminosity function between and . The data do not justify a Schechter (1976) function fit, but we show the Gronwall et al. (2007) luminosity function with reduced by a factor of 2.5 as the black dashed curve. This function (, , and ) provides a reasonable fit to the data, though it is by no means unique. It would imply a reduction by a factor of in the Ly luminosity density between and , which is much smaller than the dramatic factor of 50 reduction that Cowie et al. (2010) estimate between and . This suggests that much of the evolution occcurs in the range, as is seen in many other populations.
5 What are the LAEs?
We now turn to the samples alone and what they can tell us about how LAE galaxies are drawn from the more general UV-continuum selected galaxy population. In this section we will occasionally include the LAEs with EW(Ly Å from the Table 1 supplement when this is appropriate. We parameterize the data with the rest-frame EW(H). Using the EW(H) allows us to intercompare easily the properties of the Ly and UV-continuum samples. In addition, the EW(H) is a rough measure of the age of a galaxy or, more precisely, its specific SFR (SSFR), since the conversion from SSFR to age is dependent on the star formation history of the source (e.g., Leitherer et al. 1999). To zeroth order we may also expect the EW(H) to be independent of extinction. In contrast, the complex escape path of the Ly photons relative to the continuum photons combined with the effects of extinction may result in substantial changes in the EW(Ly). These have been observed by some as decreases (Shapley et al. 2003; Pentericci et al. 2009) but not by others (Atek et al. 2008; Finkelstein et al. 2009c). We note, however, that even the EW(H) may have some dependence on extinction, since the stars producing the ionizing photons may lie in different spatial regions of the galaxy than those producing the optical continuum photons.
While in the absence of dust we may expect the flux of Ly to be related to that of H by the case B ratio (8.7 for a temperature of 10,000 K; Brocklehurst 1971), the relation between the EW(H) and the EW(Ly) is more complex because of the relative evolution of the continuum fluxes. The EW(Ly) declines rapidly after a single burst of star formation, as does the EW(H). However, for a constant SFR model, the EW(Ly) instead becomes roughly constant, since the UV-continuum flux and the Ly line strength are governed by ongoing massive star formation (Charlot & Fall 1993; Schaerer & Verhamme 2008). In contrast, the EW(H) continues to decline, even for a constant SFR model, as the older stars build up in the galaxy. For a galaxy with a large old population in place, the onset of a new burst of star formation will give a high EW(Ly) while leaving the EW(H) relatively weak. It is this ongoing build-up of the older stars that makes the EW(H) a function of the age of the source.
In Figure 5.1 we show the rest-frame EW(Ly) versus the rest-frame EW(H) for the LAE galaxies (red squares), including those with rest-frame EW(Ly) between 15 and 20 Å from the supplement to Table 1. We also show the spread of the EW(H) for the UV-continuum sample (blue diamonds). While there is a large scatter, there is a broad general trend for the LAEs to have higher EW(H). Specifically, we find that in the EW(Ly Å and EW(Ly)= Å samples, the median EW(H) is 113 Å and 71 Å, respectively, while in the UV-continuum sample, it is 38 Å, where the quantities in parentheses are the confidence range. A Mann-Whitney rank-sum test gives less than a probability that the distribution of EW(H) is the same in the LAE sample as in the UV-continuum sample. For the 12 highest EW((Ly) objects with rest-frame EW(Ly Å the rank sum test give less than a probability that the distribution of EW(H) is the same as that for those with with weaker EWs between 15 and 40 Å. This suggests that the LAEs are preferentially drawn from the younger galaxies in the sample.
These results differ from those of Östlin et al. (2009), who suggested, based on their observations of local blue compact galaxies, that there might be an anti-correlation between EW(Ly) and EW(H). As Östlin et al. noted, their small sample is highly selected and their result is not statistical. The Östlin et al. sample does emphasize, however, that there are galaxies such as SBS 0335-052 that have a very high EW(H)=1434 Å but do not have strong Ly emission. We do not detect any sources this extreme in our UV-continuum sample, where only four sources even have EW(H) Å, so they are clearly rare.
We can take this further by comparing other properties of the galaxies with the EW(H). In Figure 11 we plot the following versus the rest-frame EW(H): (a) the SDSS colors, (b) the FWHM of the galaxies in the CFHT -band images, (c) the UV spectral index, (d) the flux ratio (H)/(H), (e) the flux ratio (Ly)(H), and (f) the logarithm of the flux ratio ([NII])/(H). Panel (b) is restricted to the subsample covered by the -band CFHT images, while panels (a), (d), and (e) correspond to the subsample covered by the SDSS observations. Panels (c) and (f) contain the full sample. In panel (a) we have used the spectra to correct for the contributions of emission lines to the magnitude (see Section 5.2), since these can make significant contributions that are a function of the EW(H). We now consider each of these relations in detail.
5.2 Colors and Ages
The colors of Figure 11(a) follow a single well-defined track (see also Overzier et al. 2008, who show a similar figure for their low-redshift analogs to the LBGs). We compare the data with a constant star formation model (solid curve) computed with the STARBURST99 code (Leitherer et al. 1999). Here the EW(H) decreases as the age increases so time increases to the left in the plot. The predicted evolution of the color versus the EW(H) up to an age of yr matches to the emission-line corrected colors. By contrast, a single starburst (dashed curve) does not match, since the EW(H) drops rapidly with time, while the colors remain blue.
We can use the full color information to infer ages and extinctions by spectral energy distribution (SED) fitting. For sources with strong optical emission lines we emphasize that it is critical to remove the substantial line contributions. The correction is a function of the EW(H) and hence introduces a systematic bias into any comparison of LAEs with UV-continuum selected galaxies. For strong-line sources SED fitting to the uncorrected data substantially overestimates the ages. This then causes the extinctions to be underestimated and the masses to be overestimated.
We illustrate the problem in Figure 5.2, where we show the broadband fluxes before (black open squares) and after (red diamonds) line correction for three LAE galaxies spanning a wide range in the EW(H). The strong [OIII] and H line contributions raise the SED at wavelengths above Å. In the fitting procedure this is spuriously interpreted as a strong Å break, which results in a substantial age overestimate. The strongest emitters (Figures 5.2(a) and 5.2(b)) are, in fact, very young but are misinterpreted as having ages of 0.45 and 0.9 Gyr when fits to the uncorrected data are used.
We have fitted ages and extinctions for all of the galaxies covered by the SDSS in both the LAE and UV-continuum samples. We used the Bruzual & Charlot (2003) models with solar metallicities and a range of exponentially declining SFR models with e-folding times from 1 Gyr to 20 Gyr. We combined these with a Calzetti et al. (2000) extinction law. For each star formation history we determined the age and extinction that produced a minimized fit using the measured errors in the broadband fluxes, together with a error to allow for systematic uncertainties.
It is well known that the EW(H) can be used as an age indicator (e.g., Leitherer et al. 1999; Bruzual & Charlot 2003). We plot the galaxy ages determined from the SED fits versus the EW(H) in Figure 5.2. In Figure 5.2(a) we show the spurious ages that would be derived from a simple fit to the data without the emission-line corrections. In Figure 5.2(b) we show the ages derived from the corrected broadband fluxes. The inferred ages are somewhat degenerate with the assumed exponential star formation history, so we also show the range of ages corresponding to the range of SFR models to give a measure of the modeling uncertainties. We can see from Figure 5.2(a) that nearly all of the sources with strong EW(H) are assigned ages of about a Gyr. Thus, without the line correction, we would consider there to be no difference in age for a wide range of EW(H). However, we can see from Figure 5.2(b) that when the line correction is made, we obtain a smooth evolution in age with the strongest EW(H) sources having the lowest ages of about 10 Myr. The high EW(H) sources are quite well described by a constant SFR model (black solid curve in Figure 5.2(b)), but at lower EW(H), the sources tend to fall below this curve, presumably as the SFR begins to decline. The data are quite well represented by a single power law fit
where the age T is in Myr and the EW(H) is in Å. This is shown by the green dashed line in Figure 5.2(b).
In Figure 5.2 we plot the surface densities of galaxies in our LAE and UV-continuum samples versus the EW(H). It is worth emphasizing that in both the LAE and UV-continuum samples the galaxies are chosen solely on the basis of their NUV magnitudes; that is, they have NUV, lie in the redshift interval , and have the same NUV magnitude distribution. The only difference between the two samples is the presence or absence of strong Ly emission. Thus, while there might be a relation between the NUV magnitude and the EW(H), this will operate in the same way in both samples and will not affect any comparison.
For the LAE sample (red shaded histogram) we plot the sum of the inverse areas over which a galaxy with a given NUV magnitude could be observed in the GALEX fields (Cowie et al. 2010). (We also corrected the LAE density for the of sources that were not spectroscopically observed.) For the UV-continuum sample (blue histogram) we also weight the areas with the fraction of sources that we observed as a function of NUV magnitude. The black solid curve shows the shape computed from a model where we assume a constant rate of formation of galaxies and that the time spent in each equivalent width bin is determined from Equation 1. The normalization is proportional to the birthrate of galaxies, and we have matched it to the observed surface density. We can see that this model matches well to the shape at high EW(H), so the surface density of galaxies versus the EW(H) is also consistent with a constant production rate of new galaxies that then evolve with a constant SFR.
We can also see from the figure that Ly emission is rare in sources with EW(H) Å (about of galaxies) but common in higher EW(H) sources where the blue and red histograms become comparable. of the EW(H) Å galaxies and of the EW(H) Å galaxies have Ly emission with the EW(Ly Å. The uncertainties reflect the small number of UV-continuum sources with high EW(H). Finally, we can see that a large fraction of the LAEs () are drawn from the high EW(H) Å population. This again shows that the LAEs are drawn primarily from the youngest galaxies. The remaining of LAEs with EW(H) Å may have geometries, kinematics, or orientations that are unusually conducive to Ly escape. However, some may be objects where the Ly emission is produced by AGN activity, but the AGN signatures in the optical are swamped by the galaxy contributions.
In the following sections we will use the EW(H Å as a rough cut above which Ly emission is more common.
Figure 11(b) shows that there is also a size evolution as a function of EW(H). The higher EW(H) sources are generally unresolved at the resolution of the CFHT MegaPrime -band images, while the lower EW(H) sources are generally extended. However, unlike in the other correlations, there is a significant difference between the LAEs and the UV-continuum sources at the same EW(H). From Figure 11(b) we can see that at the same EW(H), sources with Ly emission are significantly smaller than those without; that is, where Ly is strong is in the most compact sources at a given EW(H). A rank-sum test shows that there is a probability that the LAEs at EW(H Å are physically as large as the UV-continuum sources with the same EW(H) selection. (We have chosen the EW(H) to provide a significant number of UV-continuum sources since there are only 4 UV-continuum sources with EW(H Å as can be seen in Figure 11(b)). The median FWHM for these LAEs is , comparable to the resolution in the CFHT MegaPrime images, while that for the UV-continuum sources with the same EW(H) selection is .
The ground-based -band imaging is inadequate to resolve the sizes for most of the LAEs. Thus, although the optical morphologies may differ from those in the UV, here we use HST-based images to obtain rest-frame optical sizes. In Figure 5.3 we show the combined F606W and F850W HST ACS images for the sources that lie in the GEMS survey field. In Figure 5.3 we show the contours from the combined F606W and F850W images for the four LAEs. The FWHM in these higher resolution images is 0.5 kpc for the three sources with EW(Ly Å, though the one source with EW(Ly Å is more extended with a FWHM of about 2 kpc. These sizes are very similar to the continuum sizes of LAEs in the rest-frame UV at (Bond et al. 2009, 2010; Cowie et al. 2011) though again we wmphasize that it would be desirable to make the comparison at the same rest-frame wavelength.
We can measure the dust extinction from the galaxies in several ways: from the UV spectral slopes, from the SED fits, and from the Balmer ratios. We may intercompare these measurements and also use each of them to test the effects of extinction on the Ly line.
In Figure 11(c) the UV spectral index, which is often used as a measure of the stellar extinction, shows a general trend towards less negative values as we move to lower EW(H), which would be expected if the extinction is rising as the galaxies age. However, there is a good deal of scatter, and at least part of the change is caused by evolution in the intrinsic spectrum. The evolution of the reddening may be more clearly seen in Figure 11(d): the Balmer ratio shows a systematic decrease as the EW(H) decreases, indicating an increase in the nebular extinction. The derived E(B-V) is similar whether we use the Cardelli et al. (1989) extinction law or the Calzetti et al. (2000) extinction law. With the Calzetti et al. reddening we obtain a median E(B-V) of 0.23 (0.15, 0.32) from the Balmer ratio for sources with EW(H Å and 0.33 (0.30, 0.38) for sources with EW(H Å, where the quantities in parentheses are the range. Interestingly, in both Figures 11(c) [UV spectral index versus EW(H)] and 11(d) [Balmer ratio versus EW(H)] there appears to be little differentiation between LAEs and UV-continuum galaxies at the same EW(H), suggesting that extinction is not the primary reason for Ly being suppressed at a given EW(H). This is a well-known result which has been found by many authors (e.g., Giavalisco et al. 1996; Mas-Hesse et al. 2003; Atek et al. 2009b; Finkelstein et al. 2009a).
As has been pointed out by Atek et al. (2009a) and Scarlata et al. (2009) based on the GALEX LAE samples (see also Hayes et al. (2010) for a discussion of a higher redshift sample), there is a relation between (Ly(H) and (H(H) for LAE galaxies but with a wide dispersion. In Figure 5.4(a) we show such a plot for our LAE galaxies (red squares) and UV-continuum galaxies (blue diamonds) with EW(H Å, where the H limit is chosen to include nearly all the LAEs (see Figure 5.2). The upper limits on the Ly fluxes for the UV-continuum galaxies are computed for an observed-frame EW(Ly Å. We emphasize again that, unlike Atek et al. (2009a), our H measurements are not matched to the GALEX aperture, so our (Ly(H) ratio may have additional systematic scatter. In addition, the relative flux calibration and uncertainties in the underlying absorption correction for the H line may result in scatter in (H(H) and occasional unphysical values (negative extinction) in this ratio.
The inclusion of the UV-continuum sample shows that the LAEs correspond to the maximum values of (Ly(H) at a given (H(H), as would be expected from the initial selection in the Ly line. At any given (H(H) there is a range in (Ly(H) stretching up to roughly the value set by simple extinction of the Ly line if it followed the same escape path as the H line. The LAEs primarily lie along this relation. The line followed by the LAEs is consistent either with the Calzetti et al. (2000) reddening law (black solid line) or with uniform screen models, such as those of Cardelli et al. (1989; black dashed) or Fitzpatrick et al. (1999; black dotted), all of which give nearly identical predictions. There is considerable scatter about the relation, but we suspect that this is a consequence of systematic errors in the flux calibration discussed above, rather than an indication of the need for more complex dust models, as suggested by Scarlata et al. (2009). We can also see that the data are not consistent with the Ly photons in the LAE galaxies having a widely different E(B-V) path from the H photons. Following Scarlata et al., we have plotted the relation when the Ly path is twice that of the Balmer photons (lower green line) and also when it is half that of the Balmer photons (upper green line). It is clear that these are a much poorer fit to the LAE observations.
In contrast, if we plot the UV spectral index for the same sample versus (H)/(H) (Figure 5.4(b)), then we see that the Calzetti et al. (2000) reddening law gives a substantially different result than the Cardelli et al. (1989) or Fitzpatrick et al. (1999) uniform screen models, reflecting their different shapes at UV wavelengths. Most of the data appear to follow the models of Cardelli et al. or Fitzpatrick et al. rather than the reddening law of Calzetti et al., though there are a handful of outlying points. This suggests that for these galaxies the stellar extinction is better reperesented as a uniform screen rather than a patchy distribution. This is the case for both the LAEs and the UV-continuum sources.
Perhaps the most interesting plot in Figure 11 is that of N2=([NII])/(H)) versus EW(H) (panel f). N2 is a widely used metallicity indicator with low values corresponding to low metallicities. The fact that both the LAE and UV-continuum galaxies follow a surprisingly tight track in this diagram with N2 rising as EW(H) decreases appears to show that we are seeing a smooth metal build-up with age in both classes of galaxies.
In order to translate N2 to a metallicity in the galaxies we use the Pettini & Pagel (2004) local relationship, (O/H)N2, which was determined by comparing to direct measures of the O abundance over the range N2= to . Extrapolating the Pettini & Pagel relation to these galaxies requires assuming that there is no change in the ionization parameter, which may well be incorrect. However, at Cowie & Barger (2008) found a narrow range of ionization parameters () and a similar relation to that of Pettini & Pagel. Cowie & Barger (2008) also showed that other line diagnostics gave similar metal-luminosity relations to that derived from N2 in the redshift interval .
We can also test the Pettini & Pagel (2004) relation with the present data. Ten of the sources have [OIII] detected at above the level. These are marked with green diamonds in Figure 3(a). None of the UV-continuum sample at has detected [OIII] at the level. All of the [OIII] galaxies have low values of N2, except for GALEX 1240+6233, which is classified as an AGN based on the BPT diagram. Conversely, most of the low-N2 galaxies are detected in [OIII]. For the sources that are not AGNs and where we have detected [OIII] we used the ‘direct’ or method to determine the metallicity (e.g., Seaton 1975; Pagel et al. 1992; Pilyugin & Thuan 2005; Izotov et al. 2006). To derive [O III] and the oxygen abundances, we used the Izotov et al. (2006) formulae, which were developed with the latest atomic data and photoionization models. We compare the derived O abundances with N2 in Figure 5.5 (red squares), where we also show similar measurements from a sample of ultra-strong emission-line galaxies taken from Hu et al. (2009; green diamonds). The O abundance versus N2 is broadly consistent with the Pettini & Pagel relation (black line), though there is a good deal of scatter, probably reflecting the variation in the ionization parameter.
We can see immediately from Figure 11(f) that the LAEs in the redshift interval have lower values of N2 than the UV-continuum sources (see also Figure 3). This effect was previously noted by Cowie et al. (2010), though the present larger sample substantially increases the statistical significance of the result. This result may be more clearly seen in Figure 5.5, where we show the N2 distributions for the LAE (red shaded histogram) and UV-continuum (blue histogram) samples. While the N2 distributions overlap, the distribution for the LAEs clearly extends to lower values, and the median N2 is lower. A rank-sum test gives only a probability that the two GALEX samples are similar. Using the Pettini & Pagel (2004) conversion gives a median (O/H)=8.24 (8.17, 8.35) for the LAEs, where the quantities in parentheses are the range. This is about 0.4 dex lower than that of the UV-continuum sample, though the median value for the UV-continuum sample is more poorly determined, since at the higher values of N2 the conversion breaks down as the N2 value saturates.
We can now see from Figure 11(f) that the difference in N2 distributions is primarily a reflection of the evolutionary state of the galaxy. That is, the metallicity is building up with age, and the LAEs are preferentially drawn from the younger galaxies. UV-continuum galaxies have similar metallicities to LAEs with the same EW(H). However, the LAE sample is much more weighted to high EW(H) galaxies, which are younger and more metal poor.
6 The Physical Properties of the Galaxy Samples
In order to convert the observed quantities into physical properties of the galaxies, we will assume a constant SFR model with a Salpeter (1955) initial mass function (IMF). As we have seen, this appears to provide a self-consistent fit to most of the observed properties. For simplicity, we assume that the IMF extends smoothly from 0.1 M to 100 M, though this can be easily converted into other favored IMFs with flatter slopes below M by a simple renormalization of the SFR.
In Figure 6 we show the H luminosities of the LAE and UV-continuum samples corrected for extinction using the Balmer ratio and a Fitzpatrick et al. (1999) extinction law. The median corrections for the LAE and UV-continuum samples are 1.7 and 2.1, respectively.
For our assumed constant SFR model the conversion from H luminosity to SFR is roughly invariant over the range of EW(H) and is given by the Kennicutt (1998) relation:
On the right-hand axis of Figure 6 we give the SFR in M yr corresponding to a given H luminosity. Individual galaxies have SFRs that range from about 1 M yr up to about 30 M yr. The lower bound is roughly set by the NUV magnitude limit for the selection, but the upper bound on the SFR is a measure of the maximum SFRs seen in UV-continuum selected galaxies at these redshifts. On the top axis we give the logarithm of the age in years corresponding to a given EW(H). The LAEs are primarily drawn from sources younger than Gyr.
We can see from Figure 6 that there is no evolution in the distribution of the SFRs as a function of the EW(H). This is consistent with our constant SFR assumption. There is also no difference in the distribution of the SFRs for the LAE and UV-continuum galaxies. If the highest SFR galaxies maintain their rates for a subtantial fraction of the local age of the universe, then the final stages will have stellar masses in excess of M.
We can test our H luminosity-determined SFRs against those inferred from 20 cm observations (where available; see Section 2.1) and those determined from the UV continuum. In Figure 6 we plot rest-frame 20 cm power calculated assuming a radio spectrum of versus extinction-corrected H luminosity. We see a nearly linear relation, though the number of measured sources is fairly small. We compare the data with two conversions of the radio power to SFR: Condon (1992; solid line) and Bell (2003; dashed line). The two relations appear to bracket the data points. (Prior to the H extinction correction the points are more scattered and fall to the left of the expected relations.)
There is one significantly deviant point, GALEX0959+0151 in COSMOS 00, where the H luminosity is very high compared to the radio power. This source has the highest EW(H) in the figure ( Å) and an age of 9 Myr from SED fitting (Figure 5.2(b)). This suggests that the radio power is low because the supernovae that generate the radio emission have yet to occur. Unfortunately, the other sources in the sample with very high rest-frame EW(H) are not in the fields with radio data, so we cannot test this further at present.
SFRs for high-redshift galaxies are generally calculated from the observed UV luminosity together with the stellar extinction correction determined from the UV continuum slope (Meurer et al. 1999). It is therefore interesting to commpare the SFRs derived using this method with those computed from the H luminosities in the present low redshift sample and to estimate the limits of its validity. In Figure 6 we show the ratio of the SFR derived from the stellar extinction corrected UV luminosity with that computed from the nebular extinction corrected H luminosity as a function of EW(H). We have computed the UV SFR from the luminosity, extinction corrected using the value derived from the UV spectral index, and converted using the value at yr computed from the STARBURST99 model,
With this calibration the UV SFRs are closely matched to the H SFRs with an average ratio of 0.9 for sources with EW(H Å. This small absolute difference may be a simple consequence of the relative optical and UV calibrations. As can be seen in Figure 6, the UV calibration underestimates the SFR at higher EW(H) and overestimates it at lower EW(H).
The reasons for the deviations reflect the limit of validity of the underlying assumptions. At low EW(H) the derivation of the UV continuum extinction from the UV spectral slope breaks down. The Meurer et al. (1999) correction is only applicable to starbursting galaxies, since it depends on the intrinsic spectral index being approximately fixed, as is the case for these sources (see Figure 11(c)). At low EW(H) (less than about 60 Å) the intrinsic UV spectral indices are shallower, and the inappropriate use of the relationship overestimates the SFR, as can be seen in Figure 6. A second problem with the UV continuum method is that the UV continuum flux evolves with time, even in the constant SFR models. In Figure 6 we show (green curve) the evolution of evaluated at 1600 Å () as a function of EW(H). (The normalization is arbitrary and depends on the SFR.) The UV luminosity increases as a function of age up to yr or an EW(H Å. Thus, the UV continuum calculation combined with the UV spectral index extinction correction should only be used over the EW(H Å range, or, equivalently, for rest-frame line-corrected galaxy colors of (Figure 11(a)).
In Figure 6 we plot the metallicities of the LAE (red solid squares) and UV-continuum (blue diamonds) samples versus the EW(H). If the gas reservoir were fixed, the constant SFR model would predict that the metallicities should rise linearly with time in the early stages. This is clearly much steeper than is seen in the present samples, which can be approximated by a model where the metallicity rises as (black curve: this is only shown over the range of validity of the N2 diagnostic which is only appropriate for low metallicity and saturates at near solar metallicity). Thus, the galaxies must have ongoing accretion of gas.
In Figure 6 we also compare the present samples with a fainter continuum sample from the GOODS-N observations of Barger et al. (2008) (black triangles) and with a fainter sample selected to have very high EW(H) from Hu et al. (2010) (green circles), as well as with the local blue compact galaxy sample of Östlin et al. (2009) (open squares). As might be expected, the fainter samples have systematically lower metallicities than the present data. However, the effect is small compared to the EW(H) dependence, suggesting that the ages of the galaxies primarily determine the metallicities.
Using the SFRs derived from the extinction-corrected H luminosities and the masses derived from the constant SFR fits to the line-corrected SEDs, we show our final result in Figure 6. In Figure 6(a) we show the distribution of the SFRs. The two samples are nearly identical with a median of M yr. In contrast, as is shown in Figure 6(b), the LAEs, reflecting their younger ages, are systematically lower in mass than the UV-continuum sources. The median mass for the LAEs is M, while that for the UV-continuum sample is M.
We analyzed a substantial sample of UV continuum-selected galaxies with and without Ly emission lines to try to understand how LAEs are drawn from the general population and how they evolve with redshift. To do this, we obtained extensive optical spectroscopy of GALEX grism selected samples (both LAE samples and comparison samples with the same UV magnitude distributions but no detected Ly) in two redshift intervals: and . We used the optical spectroscopy to eliminate AGNs and to obtain the optical emission-line properties of the samples.
We confirmed that the GALEX selected LAEs are considerably fainter and much rarer than the high-redshift LAEs (Deharveng et al. 2008; Cowie et al. 2010). Cowie et al. (2010) showed that the in a Schechter (1976) function fit to the LAE luminosity function is almost an order of magnitude fainter than the in a fit at and that only about of UV-continuum selected galaxies have rest-frame EW(Ly Å. Here we showed that there are also LAEs that can be found in the GALEX spectra at and that these are similar in luminosity to the most luminous high-redshift galaxies. Thus, we conclude that most of the observed evolution occurs over the redshift interval . This appears to be a simple downsizing effect, with the presence of the higher Ly luminosity sources at corresponding to higher SFR sources initiating their star formation at this redshift.
We showed that at , SED fits to strong emission-line galaxies significantly overestimate the ages and masses and underestimate the extinctions if we do not correct for the emission-lines. We found that the SED fits to the uncorrected broadband fluxes give ages of about a Gyr. However, the inferred ages drop substantially when we correct the galaxy broadband magnitudes using the observed spectra to remove the emission-line contributions. All the galaxies with ages much less than a Gyr have strong emission lines and must be corrected for the line contributions. Spectral synthesis fitting shows a smooth evolution of the rest-frame EW(H) with age when the emission-line contributions are removed from the broadband fluxes.
We found that at , all sources, regardless of the strength of the Ly emission line, follow a single, well-defined sequence as a function of the rest-frame EW(H). Higher EW(H) sources all have lower metallicities, bluer colors, smaller sizes, and less extinction. The number distribution of galaxies versus the EW(H) is consistent with a constant formation rate of new galaxies. The bulk (%) of the LAEs lie at high EW(H Å, and % of all UV-continuum selected galaxies with EW(H Å are LAEs. We conclude that the low-redshift LAEs are primarily drawn from a population of young galaxies that have recently initiated star formation.
It appears that LAEs represent an early stage in a starburst when the star-forming gas is still relatively pristine and the primary star-forming region is small. It also appears that there is a time sequence, with the Ly emission line dying away and the metallicity of the gas rising as the galaxy evolves.
Appendix A Appendix
Since the analysis of Cowie et al. (2010), much deeper GALEX grism spectroscopic observations were released for the CDFS 00 field. We have obtained the one- and two-dimensional GALEX spectra from MAST and analyzed them to search for Ly emission using the same procedures as in Cowie et al. (2010). We find a sample of 100 Ly selected sources within a radius field. We give the properties of this sample in Table A1. For each source we give the number, the GALEX name, the J2000 right ascension and declination in decimal degrees, the NUV and FUV magnitudes, the redshift inferred from the Ly emission line in the GALEX UV spectrum, the line width in km s together with the error, whether the galaxy is classified as an AGN based on the presence of high-excitation lines in the UV spectrum, and, finally, the optical redshift, if available. The optical redshifts are primarily taken from our DEIMOS spectroscopy, but we also include redshifts from Balestra et al. (2010) and Vanzella et al. (2008), which we note with a colon after the redshift. Where a source lies within the X-ray observations of the Extended Chandra Deep Field-South (ECDF-S) and is (is not) detected in X-rays (Lehmer et al. 2005), we give the logarithm of the rest-frame keV luminosity (an “E”) in parentheses in the optical redshift column (col. 10).