A FLUX-LIMITED SAMPLE OF z ∼ 1 Lyα EMITTING GALAXIES IN THE CHANDRA DEEP FIELD SOUTH^*,^†

Lyα is the only emission line that we can detect in the highest redshift (z > 6) galaxies, making it the only probe of the internal structure of these galaxies and one of the few diagnostic tools for studying the intergalactic gas. However, separating the internal and external effects will always be difficult at these redshifts. Thus, to determine the intrinsic Lyα properties of galaxies, we need to study lower redshift samples where the galaxy structure can be separated from the effects of the intergalactic medium.

Recently, substantial samples of z ∼ 0.2–0.4 Lyα emitters (LAEs) and a small number of z ∼ 1 LAEs have been found with the Galaxy Evolution Explorer (GALEX; Martin et al. 2005) grism spectrographs. The pipeline extractions for the grism spectrographs provide spectra to a fixed limiting near-ultraviolet (NUV) magnitude that can be searched for Lyα emission lines (Deharveng et al. 2008; Cowie et al. 2010). The procedure enables the selection of a substantial sample of sources that can be compared to the high-redshift LAEs. Since the selection process is based on finding Lyα in the UV-continuum-selected GALEX sources, it is most analogous to locating LAEs in the Lyman break galaxy population using spectroscopy (e.g., Shapley et al. 2003).

The advent of these low-redshift LAE catalogs has been very exciting, and many follow-up papers have been written (e.g., Finkelstein et al. 2009a, 2009b; Atek et al. 2009; Scarlata et al. 2009; Cowie et al. 2011). However, because the GALEX grism pipeline only includes objects whose UV continuum magnitudes are bright enough to generate measurable continuum spectra, only a fraction of the emission-line objects at these redshifts are detected. In particular, strong emitters with high equivalent widths (EWs) may have emission lines that are detectable but continuum magnitudes that fall below the continuum threshold.

Moreover, because the samples have both a continuum and an emission-line selection, it is not straightforward to compute the LAE luminosity functions (see the extensive discussions in Deharveng et al. 2008 and Cowie et al. 2010). Thus, ad hoc assumptions about the invariance of the EW distribution as a function of continuum magnitude need to be made, even though they may not be correct.

Fortunately, it is possible to detect directly the excluded emitters by re-extracting the GALEX data. In this paper, we show how to find "orphan" emission lines (e.g., Straughn et al. 2008; Keel et al. 2009) and how to generate a flux-limited Lyα emission-line sample. We refer to this approach as a data cube search. First, we extract a spectrum for every spatial pixel in the GALEX fields. Next, we form a three-dimensional data cube, analogous to that produced by an integral-field unit, with two spatial axes and a third wavelength axis. Finally, we search the wavelength slices in these data cubes for emitters.

An alternative procedure is to use the deep broadband GALEX images to develop a target list of galaxies fainter than those used in the pipeline and then to extract the spectra of these objects. However, since this approach still introduces a pre-selection (which must be included and modeled in any interpretation and may omit extreme EW emitters), we prefer to use the data cube approach.

In order to fully understand the LAEs that we find, we need a wealth of ancillary data. In particular, the key requirement is as extensive a set of redshift measurements as possible. Thus, for this paper, we focus on the intensively studied Chandra Deep Field South (CDFS; Giacconi et al. 2002) region, where we can use both existing optical spectroscopy to confirm the UV identifications of the Lyα line and deep X-ray data to identify active galactic nuclei (AGNs). In Section 2, we describe the rich ancillary data on the CDFS. In Section 3, we detail our data cube search methodology and use it to extract a sample of LAEs at z ∼ 1. We also use simulations to estimate the completeness with which we can detect emitters of a given flux. In Section 4, we present the properties of the 28 new z ∼ 1 LAEs that we found in our data cube search. We make use of the ancillary data to test the reliability of the LAEs and their UV redshifts and to get a first estimate of the AGN contamination. In Section 5, we present our discovery of a giant Lyα blob. We also compute the z ∼ 1 LAE luminosity function, which we compare with LAE luminosity functions at lower and higher redshifts. In Section 6, we summarize our results.

We use a standard H₀ = 70 km s⁻¹ Mpc⁻¹, Ω = 0.3, and Ω = 0.7 cosmology throughout. All magnitudes are in the AB magnitude system.

The four deepest NUV grism observations (CDFS-00, 353 ks; GROTH-00, 291 ks; NGPDWS-00, 165 ks; COSMOS-00, 140 ks) correspond to some of the most intensively studied regions in the sky. In this paper, we use the central 50' by 50' region (centered on right ascension 3^h30^m402 and declination −27°27'434 in J2000 coordinates) of the CDFS-00 for a pilot study to show the power of the data cube method. This region is extremely rich in ancillary information. For example, it contains the Hubble Space Telescope (HST) Advanced Camera for Surveys Great Observatories Origins Deep Survey-South (GOODS-S; Giavalisco et al. 2004) subregion, where there are more than 3000 spectroscopic redshifts in a 140 arcmin² area (Le Fèvre et al. 2005; Mignoli et al. 2005; Vanzella et al. 2008; Popesso et al. 2009; Balestra et al. 2010; Cooper et al. 2012; Cowie et al. 2012). In Figure 1, we show how >80% of the galaxies with B < 24 in the GOODS-S have spectroscopic identifications. In addition, multicolor observations from the HST programs GOODS-S (Giavalisco et al. 2004), Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS; Grogin et al. 2011; Koekemoer et al. 2011), and Galaxy Evolution from Morphologies and SEDs (GEMS; Rix et al. 2004) provide the galaxy morphologies and spectral energy distributions from the rest-frame far-UV (FUV) to the mid-infrared for a substantial fraction of the objects. Finally, the 4 Ms Chandra image of the CDFS (Luo et al. 2008; Xue et al. 2011) and the Extended CDFS (ECDFS; Lehmer et al. 2005; Virani et al. 2006) fields can be used to help eliminate AGNs. A number of optical spectroscopic surveys have specifically targeted the X-ray sources in these fields (Szokoly et al. 2004; Treister et al. 2009; Silverman et al. 2010).

Zoom In Zoom Out Reset image size

3.1. GALEX Spectral Images

3.2. Data Cube Construction

The slitless nature of the grism data results in spectral overlaps between neighboring objects. However, because each of the spectral images is taken at a different rotation angle, overlaps can be removed. Objects only overlap with a given neighbor at one orientation, and so in the multiple spectral images, the signal associated with a given object can be picked out from the occasional contamination. The grism data are often portrayed as a "propellor image," where images at many grism angles are added together, as shown in Figure 2. This type of image is extremely useful for visualizing our present GALEX grism extraction process. Each pixel (x, y) on the spectral image corresponds to a specified position in right ascension and declination. At a given orientation, the grism spectrum corresponding to that pixel runs at that angle relative to the central position. The zeroth order lies to the negative side, while the higher orders lie to the positive side. Thus, for intensively observed fields with large numbers of independent spectral images, we can form a four-dimensional data cube (x by y by the number of wavelength elements by the number of spectral images). The number of x and y pixels corresponds to the total number of spatial pixels in the grism image (here 2000 by 2000), and we hereafter refer to the number of spectral images as NN.

Zoom In Zoom Out Reset image size

A given wavelength corresponds to a circular annulus (large circle in Figure 2), the spokes of which correspond to the grism angles. A given emission line will occur at these specified positions in the suite of images with varying grism angles (e.g., Keel et al. 2009). If we construct a vector of values, only the object itself appears in all the elements. Other objects occur as random contamination in single elements of the vector. Thus, we can mask out pixels on the individual spectral images where we expect contamination from bright neighboring galaxies. We can further remove contamination from fainter objects by using a clipped average to determine the true signal. In doing so, we can collapse the four-dimensional data cube to a three-dimensional data cube (x by y by the number of wavelength elements). This final three-dimensional data cube is analogous to the data cube that would be obtained with a scanning imaging spectrometer or with an integral-field spectrograph where each spatial pixel has a spectrum associated with it. For most spatial pixels, there is no significant continuum flux or emission lines in the spectrum (the third axis). We may also view the three-dimensional data cube as a sequence of narrowband images, stepping in wavelength, each of which may be searched for emitters. It is in this way that we form our emission-line-selected sample.

We created a pipeline to convert the NUV and FUV spectral images into four-dimensional data cubes. We chose a wavelength step of 5 Å for the NUV output cube in order to oversample the roughly 20 Å wavelength resolution. We chose the wavelength range (1930–2730 Å) to cover the 25% response range of the NUV grism. This gives 160 wavelength elements.

In forming the four-dimensional data cubes (x by y by 160 by NN), we wanted to maintain the statistical independence of the pixels. Thus, in assigning to each (x, y, λ) a vector (the fourth axis), we need to be concerned with the fact that we are oversampling the spectral images in wavelength. To avoid averaging, we decided to assign to each (x, y, λ) position in a given spectral image (recall that each position will have NN values since there are NN spectral images) the nearest neighbor pixel using the grism angle from the GALEX pipeline. Note that the GALEX pipeline computes the grism angle by fitting to the spectra of bright objects in the field. To maintain the statistical independence of the pixels, we will only assign each pixel in a given spectral image once, so since we are oversampling the spectral images in wavelength, this means that some of the pixels in the fourth axis are blank. These blank pixels are ignored hereafter.

At this point, we collapsed the fourth axis by forming the median, comparing each value in the fourth axis with the median, and eliminating points that deviate from the median by more than 5σ. We formed an exposure time-weighted average (which allows for the range in exposure times of the individual images) from the remaining points to give the final signal at this (x, y, λ) pixel in the three-dimensional data cube.

We used an identical procedure for the FUV processing, except here we chose a wavelength step of 2.5 Å over the wavelength range 1345–1745 Å, giving 160 wavelength elements. Also, because of the low background photon detection rate per pixel in the FUV, and because contamination is much less of a problem in the FUV, we formed a simple average rather than a median for comparison with each value in the fourth axis. We eliminated points that deviated from the average by more than 5σ. Then, as we did for the NUV processing, we formed an exposure time-weighted average of the clipped vector to obtain the final signal at this (x, y, λ) pixel in the three-dimensional data cube.

For the present emission-line search, we are not interested in the objects with brighter continua whose spectra are already included in the existing GALEX pipeline data products. We also want to minimize any contamination from these brighter galaxies and stars in the final three-dimensional data cube. We therefore masked out the spectral region in each GALEX spectral image that corresponded to a pixel with a surface brightness in excess of 25.8 mag arcsec⁻² in the corresponding NUV image of the field smoothed with a 2 pixel boxcar smoothing. This removes all objects with NUV magnitudes brighter than 20.4 and no objects with NUV magnitudes fainter than 21.25. This faint magnitude bound is considerably brighter than the magnitude limit of the pipeline sample (22), so we only eliminate objects that already have spectra and hence already have been searched for Lyα lines. (The remaining objects that are already present in the pipeline sample are eliminated later in the process; see Section 3.3.) None of our results are sensitive to the precise choice of the masking surface brightness threshold.

In Figure 3 (left panel), we show a typical portion of a wavelength "slice" (a 40 Å region obtained by summing eight of the 5 Å steps; see Section 3.3) in the final data cube compared to (right panel) the corresponding NUV continuum image. The positions in the final data cube that correspond to the regions that were masked appear white in the wavelength slice image (see left panel). 10901 spatial pixels are eliminated by the masking, or just under 0.3% of the spatial area covered by the three-dimensional data cube.

Zoom In Zoom Out Reset image size

3.3. Data Cube Search

With the three-dimensional data cube in hand, we can now carry out a search for emission-line objects. We first divided each 5 Å wavelength slice in the data cube by the grism effective area at that wavelength to take out response variations. We used the values given in http://galexgi.gsfc.nasa.gov/docs/galex/Documents. We then formed forty 20 Å (in order to roughly match the resolution element) narrowband slices from the NUV data cube. For each slice, we subtracted the average of slices on either side of the primary slice, N. We used slices N−5 to N−3 and slices N+3 to N+5 to form this average. The background subtraction removes most of the residual structure and most of the continuum objects in the field, as we see by comparing the left panel of Figure 3 with the upper-right panel of Figure 4. This will retain any emitter whose width is less than about 240 Å. Even broadline AGNs satisfy this criterion, so we should not be eliminating any objects by this process.

Zoom In Zoom Out Reset image size

From the left panel of Figure 3, we can see residual regions around the small number of very bright sources that were masked, and these can cause contamination. Thus, in each of the 40 slices, we masked out additional spatial regions around these bright sources. For sources with NUV magnitudes brighter than 14, we masked a 4' radius region around the source, and for sources with NUV magnitudes brighter than 17, we masked a 15 radius region.

We searched each slice for objects with surface brightnesses of at least 2 × 10⁻⁶ photons cm⁻² s⁻¹ per 15 spatial pixel using a 4 pixel spatially smoothed version of the slice. The smoothing roughly matches the 53 spatial resolution in the GALEX NUV band. The surface brightness cut was chosen empirically to provide the maximum depth while not generating an unmanageably large number of spurious sources that have to be eliminated by hand. Bright emitters may appear in more than one slice, and we formed our preliminary catalog by merging all of the emitters found in neighboring slices. As a consequence of the background subtraction process, we only cover a wavelength range of 2030–2630 Å, which corresponds to a redshift range for Lyα of z = 0.67–1.16. At this stage, any objects already in the pipeline sample were removed.

We created the one-dimensional spectra directly from the three-dimensional data cube using a 4 × 4 spatial pixel box together with a local background subtraction. We flux calibrated by matching to the NUV magnitude of the object. We visually inspected each of the narrowband slices and the one-dimensional spectrum of each selected object and eliminated objects that were artifacts, such as remaining edge effects from the brighter objects or objects with stellar spectra, to form our primary catalog. The initial selection yielded 161 objects, of which we retained 28.

For each of the 28 objects, we also created a two-dimensional spectral image from which we could extract final one-dimensional spectra. For doing the extractions, we created our own version of the GALEX pipeline. The only difference between our version of the pipeline and the standard pipeline is that we used a profile-weighted spectral extraction to form the one-dimensional spectra (Horne 1986). As we illustrate in Figure 5, we obtained some slight improvements in signal to noise from the profile weighting (about a factor of 1.3), which is helpful in dealing with weak features, such as the faint Lyα emission lines. (Note that the source shown in Figure 5 was from the pipeline sample rather than from our data cube sample.)

Zoom In Zoom Out Reset image size

In Figure 6, we show the two-dimensional NUV spectral images of our 28 new z ∼ 1 LAEs, ordered according to Table 1 (see Section 4), along with their FUV and NUV continuum images (shown to the far-left and left of each spectral image, respectively). In Figure 7, we show the one-dimensional spectra that we extracted from these. We performed the final flux calibration using the given GALEX spectral response.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 1. Emission-line Sample

Name	R.A.	Decl.	NUV	FUV	zgalex	f	log L	EW	UV	log f	zground	Opt
	(J2000.0)	(J2000.0)	(AB)	(AB)		(Lyα)	(Lyα)	(Lyα)	Type	(2–8 keV)		Typea
						(erg cm−2 s−1)	(erg s−1)	(Å)		(erg cm−2 s−1)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)
GALEX033112-281517	52.800373	−28.254972	23.19	27.56	1.160	77 ± 3	43.16	109 ± 2	...	...	...
GALEX033359-275326	53.499123	−27.890778	22.14	24.11	1.144	24 ± 3	43.45	82 ± 4	AGN	...	...
GALEX033301-273227	53.255127	−27.540943	23.64	−27.39	1.137	20 ± 3	42.89	92 ± 6	...	...	...
GALEX033146-272942	52.942501	−27.495249	22.81	25.06	1.133	15 ± 3	42.71	28 ± 8	...	...	...
GALEX033230-280804	53.125668	−28.134527	22.42	23.76	1.126	18 ± 3	43.11	50 ± 6	AGN	...	...
GALEX033348-280216	53.453625	−28.037998	23.49	25.57	1.125	10 ± 3	42.93	89 ± 5	...	...	...
GALEX033120-274917	52.837460	−27.821417	23.00	−29.55	1.116	13 ± 2	42.76	38 ± 5	...	...	...
GALEX033255-281320	53.231544	−28.222471	22.51	28.98	1.084	14 ± 3	42.80	28 ± 5	...	...	...
GALEX033200-274319	53.001377	−27.722055	22.45	24.24	1.042	10 ± 3	43.11	61 ± 7	...	−14.2	1.037b	BLAGN
GALEX033228-273614	53.118168	−27.604000	23.69	25.77	1.033	41 ± 2	42.59	58 ± 6	...	ECDFS	...
GALEX033113-274949	52.807587	−27.830278	22.54	−28.61	1.032	16 ± 2	42.70	26 ± 5	...	...	...
GALEX033202-280320	53.010712	−28.055723	23.55	25.24	1.017	26 ± 2	42.89	104 ± 3	AGN	−13.7	1.014c	Sy2
GALEX033111-275506	52.798542	−27.918472	24.00	−27.28	0.996	15 ± 2	42.62	89 ± 8	...	...	...
GALEX033336-274224	53.402752	−27.706667	23.04	26.32	0.978	28 ± 2	43.10	116 ± 9	AGN	−14.7	...
GALEX033246-274154d	53.195625	−27.698416	23.41	25.82	0.977	14 ± 2	42.86	93 ± 3	...	ECDFS	...
GALEX033204-273725	53.017002	−27.623806	22.87	24.31	0.976	8 ± 3	43.14	107 ± 8	AGN	−13.5	0.970e	BLAGN
GALEX033335-273934	53.398125	−27.659584	22.62	23.31	0.949	9 ± 2	42.85	47 ± 6	...	−14.2	...
GALEX033329-280127	53.373249	−28.024277	23.23	25.71	0.941	23 ± 2	42.62	48 ± 1	...	ECDFS	...
GALEX033206-281408	53.026459	−28.235723	21.88	25.76	0.913	10 ± 2	43.22	60 ± 3	...	...	...
GALEX033044-280237	52.686874	−28.043751	23.33	27.41	0.872	12 ± 2	42.57	56 ± 7	...	...	0.855	NELG
GALEX033251-280809	53.214336	−28.135973	22.76	26.13	0.836	39 ± 2	42.67	44 ± 0	...	...	0.832	NELG
GALEX033045-274506	52.689747	−27.751862	23.21	25.77	0.830	7 ± 2	42.64	65 ± 2	...	...	...
GALEX033314-274834	53.310581	−27.809526	24.24	25.30	0.830	11 ± 2	42.49	119 ± 3	...	ECDFS	...
GALEX033057-273316	52.739708	−27.554583	22.49	24.02	0.795	20 ± 2	42.72	43 ± 6	...	...	0.790	NELG
GALEX033124-275625	52.851665	−27.940474	22.78	25.39	0.775	9 ± 2	42.62	48 ± 9	...	ECDFS	0.776f	NELG
GALEX033235-274059	53.148876	−27.683195	23.74	25.95	0.746	7 ± 2	42.70	150 ± 9	...	ECDFS	0.735g	NELG
GALEX033146-274846	52.942459	−27.812805	23.18	25.48	0.742	19 ± 2	42.78	107 ± 3	...	ECDFS	...
GALEX033131-273429	52.880333	−27.574804	22.57	25.54	0.695	8 ± 1	43.21	191 ± 8	...	−14.1	0.688h	Sy2

Notes. ^aBLAGN: broadline AGN, i.e., presence of at least one emission line having FWHM > 2000 km s⁻¹; Sy2: Seyfert 2, i.e., [Ne v] emission; NELG: narrow emission-line galaxy, i.e., at least one emission line and no BLAGN signatures or [Ne v] emission. ^bSzokoly et al. (2004). ^cTreister et al. (2009). ^dGiant Lyα blob; see Figures 14 and 15. ^eTreister et al. (2009); Szokoly et al. (2004) give z = 0.977. ^fCooper et al. (2012). ^gVanzella et al. (2008). ^hSilverman et al. (2010).

Download table as:  ASCIITypeset image

We decided that using the one-dimensional spectra formed from our version of the pipeline software rather than the one-dimensional spectra obtained from the data cube would enable us to make the most direct comparison with brighter emitters in the field, since their spectra had been formed from the pipeline software. As a bonus, it would also provide an independent confirmation of the emission-line selection. In Figure 8, we compare for two of the emitters the one-dimensional spectra formed both ways. Despite the very different procedures used, the two methods produce spectra that agree well in shape and normalization.

Zoom In Zoom Out Reset image size

For each of the 28 objects, we checked that the discovered line was not clearly another line rather than Lyα by using the combined FUV and NUV spectra. For the spectrum shown in Figure 5, the discovered line was not Lyα, but as we noted above, this object was from the pipeline sample rather than from our data cube sample. All of the lines in our data cube sample appear to be Lyα.

We measured the redshifts, the Lyα fluxes, and the line widths by fitting a Gaussian to the primary line in each spectrum. Since, in general, the continua are faint in the spectra, we obtained the EWs of the lines by dividing the measured Lyα flux by the continuum flux measured from the broadband NUV continuum image. We also made a classification of whether the emitter was definitely an AGN based on the presence of high-excitation emission lines in its UV spectrum, or whether it could be a star-forming galaxy (see, e.g., Cowie et al. 2010). Our candidate star-forming galaxy sample will still contain some remaining AGNs, so optical spectra and/or X-ray imaging data are needed to make a final determination of whether a galaxy is, in fact, star-forming.

3.4. Completeness of Recovery versus Flux

Because of the low spatial and spectral resolution of the GALEX grism data, the emitters almost all appear as unresolved objects in the narrowband slices or in the two-dimensional spectral images (Figure 6). The one exception to this is GALEX033246-274154, which is clearly extended. In Figure 9(a), we show a contoured image (black) of a typical emitter, and in Figure 9(b), we show a similar image for GALEX033246-274154. In both panels, we also show the point-spread function contours formed from the average of the three brightest sources in the sample (red). There is a clear distinction between the two panels. In order to be spatially resolved, the object must be very extended. Thus, we shall interpret GALEX033246-274154 as a Lyα blob in the discussion section.

Zoom In Zoom Out Reset image size

Since nearly all the emitters in our sample are unresolved, it is relatively straightforward to estimate the completeness with which we can detect emitters of a given flux, since we do not need to consider morphological or size differences. In addition, since the observed-frame EWs of the objects are large compared to the spectral resolution, the emission-line peaks have a high contrast with the continuum level (Figure 7). Thus, the selection depends only on the emission-line flux and not on the continuum level, or, as a corollary, on the EW.

In order to estimate the completeness, we added in 3000 objects (100 emitters in each of the 30 narrowband slices) of a specified flux with shapes corresponding to the spatial point-spread function measured using bright objects in the data cube. The emitters were added into the data cube at random spatial positions and uniformly distributed through the narrowband slices. We then ran our standard selection procedure and found the number of recovered objects. At fluxes above 10⁻¹⁵ erg cm⁻² s⁻¹, our selection is more than 80% complete, but it drops extremely rapidly below this value (see Figure 10). We shall use this in calculating the Lyα luminosity function.

Zoom In Zoom Out Reset image size

In Table 1, we summarize the properties of the 28 new z ∼ 1 LAEs in the CDFS-00 found in our data cube search. These are all objects that were not present in the pipeline sample. In Column 1, we give the GALEX name; in Columns 2 and 3, the J2000 right ascension and declination; in Columns 4 and 5, the NUV and FUV AB magnitudes; in Column 6, the redshift from the GALEX UV spectrum; in Column 7, the Lyα fluxes with 1σ errors; in Column 8, the logarithm of the Lyα luminosity; in Column 9, the rest-frame EW(Lyα) with 1σ errors; in Column 10, an entry of "AGN," if the source is classified as an AGN based on its UV spectrum; in Column 11, the logarithm of the 2–8 keV flux, if the source is detected in the ECDFS, or an entry of "ECDFS," if the source is in the ECDFS region but is not detected; in Column 12, the optical ground-based redshift; and in Column 13, the optical spectral classification.

We measured the NUV and FUV AB magnitudes in 8'' diameter apertures centered on the emitter positions. We corrected these to total magnitudes by measuring for the brighter objects (20–23 mag range) the offset between the aperture magnitudes and the magnitudes of the pipeline sample. The 1σ errors obtained by measuring magnitudes at random positions in the images are 26.0 for the NUV and 26.7 for the FUV. All of the objects are bright in the NUV, but some are not detected in the FUV. For objects where we measure a negative flux in the FUV, the quantity in Column 5 is the magnitude corresponding to the absolute value of the flux with a minus sign in front to indicate that the flux was negative.

When allowance is made for the masking done on bright sources (see Sections 3.2 and 3.3), the area covered is 2286 arcmin², and the data cube emission-line search procedure covers a redshift range for Lyα detections of z = 0.67–1.16. The observed comoving volume is 2.3 × 10⁶ Mpc³.

Thirteen of the LAEs lie in the area covered by the deep X-ray exposures of the ECDFS (Lehmer et al. 2005; Virani et al. 2006); six of these thirteen LAEs are detected in X-rays (see Column 11). At z ∼ 1, the ECDFS sensitivity limit (f_2–8 keV ∼ 6.7 × 10⁻¹⁶ erg cm⁻² s⁻¹) is close to the X-ray luminosity threshold of 10⁴² erg s⁻¹ that is usually used to define AGN activity (e.g., Hornschemeier et al. 2001; Barger et al. 2002; Szokoly et al. 2004; Silverman et al. 2005), so the sources that are not detected have X-ray luminosities consistent with being star-forming galaxies. We hereafter refer to the sources that are X-ray detected (and hence are clearly AGNs based on their X-ray luminosities) as X-ray AGNs.

Nine of the LAEs have optical spectroscopic redshifts (see Column 12). For all of these sources, the optical redshift confirms the UV redshift derived from the data cube search. Four of these nine are X-ray AGNs, which is not too surprising, given that the spectroscopy in the ECDFS region was concentrated on X-ray-selected objects. The mean offset and dispersion between the ground-based optical and the GALEX UV redshifts is z_ground − z_galex = −0.004 ± 0.003.

All three of the sources in the ECDFS region that were classified as AGNs based on the presence of high-excitation lines in their UV spectra are also X-ray AGNs. In the other direction, three of the six sources classified as X-ray AGNs were already classified as AGNs based on their UV spectra. Of the remaining three sources, GALEX033131-273429 is classified as a Seyfert 2 (Sy2) based on the presence of [Ne v]λ3426 in its optical spectrum, and, as can be seen in Figure 7(a), has a single narrow but very strong Lyα emission line in the UV but no obvious C iv λ1549 emission; GALEX033200-274319 is classified as a broadline AGN from its optical spectrum; and GALEX033335-273934 has not been optically observed. In the case of the latter two objects, their redshifts are poorly positioned to detect either C iv λ1549 (O vi λ1036 for GALEX033335-273934) in their UV spectra, and only their Lyα lines are seen (though GALEX033200-274319 may have weak O vi).

Classifying sources as AGNs based on their UV spectra can be challenging, since quite often C iv is not accessible, and, as can be seen from Figure 7, O vi is typically quite weak. The fraction of LAEs in the ECDFS region that were not classified as AGNs based on their UV spectra (10 sources) but are classified as AGNs based on their optical spectra (two sources) is 20%. This is very similar to the ∼20% found for the LAEs in the GALEX pipeline samples (Finkelstein et al. 2009b; Scarlata et al. 2009; Cowie et al. 2010; see Cowie et al. 2011 for a discussion of the variations in the measurements of this quantity).

We conclude that the bulk of the LAEs found in our data cube search are real and that the UV spectroscopic redshifts based on the Lyα identifications are reliable.

We now consider how the properties of the CDFS-00 LAEs in our data cube sample compare with those of the CDFS-00 LAEs in the pipeline sample. In Figure 11(a), we illustrate how, by construction, the pipeline misses sources fainter than the pipeline magnitude limit of NUV ∼ 22 (dashed line). We show the GALEX redshifts versus the NUV magnitudes for the CDFS-00 LAEs found by Cowie et al. (2010, 2011; black squares) and the new CDFS-00 LAEs found by our NUV data cube search (blue diamonds). In contrast, in Figure 11(b) (symbols and colors as in Figure 11(a)), we show how there is substantial overlap in the Lyα emission-line fluxes of the two samples, since this quantity depends on both the NUV magnitude and the EW(Lyα). Here we can see that the pipeline extraction begins to miss sources below a Lyα flux of about 8 × 10⁻¹⁵ erg cm⁻² s⁻¹ and becomes progressively more incomplete at fainter fluxes, until below 2 × 10⁻¹⁵ erg cm⁻² s⁻¹, very few sources are detected in the pipeline extraction.

Zoom In Zoom Out Reset image size

In Figure 11(a), we use larger red symbols to denote sources with high-excitation UV lines (in addition to Lyα) in their GALEX spectra. These sources are classified as AGNs based on their UV spectra. We can see that at NUV < 22, most of the sources are AGNs. Thus, the only way to probe the bulk of the LAE galaxy population at these redshifts is to do a data cube search of the present type.

The sources found in the data cube search and missed in the pipeline are those with high EWs. The progressive increase in incompleteness with decreasing Lyα flux can be illustrated by the number of sources lying above the EW threshold at a given Lyα flux. We show this in Figure 12 (symbols and colors as in Figure 11), where we plot rest-frame EW versus Lyα flux. For a given Lyα flux and redshift, the NUV = 22 continuum cut corresponds to a given EW. We show these cuts in the figure at the upper (dashed diagonal) and lower (solid diagonal) end of the z = 0.67–1.16 redshift range. They become progressively lower as we move to fainter Lyα fluxes, such that fewer of the emitters are included in the pipeline sample.

Zoom In Zoom Out Reset image size

We next inspected the UV/optical images of the full CDFS-00 LAE sample (i.e., this includes both the pipeline and the data cube LAE samples) in the redshift range z = 0.67–1.12. All but one of these LAEs are covered by deep U-band imaging data obtained with the wide-field (1 deg² field-of-view) imaging camera MegaPrime/MegaCam on the 3.6 m Canada-France-Hawaii Telescope (CFHT). The Canadian Astronomy Data Centre reduced the data using the MegaCam Image Stacking Pipeline (MegaPipe; Gwyn 2008) and give the 5σ AB magnitude limits as ranging from 26.5 to 27.0. Only one of the observed objects (GALEX033246-274154) does not have a clear counterpart in the CFHT U band.

In addition, 20 of the 70 LAEs in the full CDFS-00 sample are in the GEMS region (Rix et al. 2004), where high spatial resolution HST images are available. We show the GEMS images in Figure 13. We have divided these into two groups: AGNs (based on either their UV spectral signatures or their X-ray luminosities, which turns out to include all of the objects that were classified as AGNs based on their optical spectra) and candidate star-forming galaxies (no high-excitation UV lines and not detected in X-rays). It appears that the morphologies of the objects follow their classifications: the AGNs are dominated by compact nuclei, while the star-forming galaxies are primarily resolved small blue galaxies, often with multiple components.

Zoom In Zoom Out Reset image size

In addition to not being detected in the CFHT U-band, GALEX033246-274154 (last thumbnail of the bottom row in Figure 13) is not clearly detected in the HST images. GALEX033246-274154 appears to be a rather unusual object. It is clearly visible both in the NUV continuum image and in the UV spectral image (Figure 6) and hence is real. Furthermore, it is spatially resolved (Figure 9(b)). Its major axis is 18'' and its minor axis is about 9'', giving a linear major-axis diameter of 120 kpc and placing it in the category of the giant Lyα blobs, though its Lyα luminosity is at the low end of these objects (Matsuda et al. 2011).

Fortunately, a much deeper U-band image than the CFHT MegaPrime image was taken (Nonino et al. 2009) with the VIMOS instrument on the ESO Very Large Telescope and covers the central region of the field, including this object. In Figure 14, we compare the FUV image (first thumbnail) of the source (located at the center of each image) and the NUV image (second thumbnail) with the VIMOS U-band image (third thumbnail) and a K_s-band image (final thumbnail). In Figure 15, we overlay the contours of the emission-line region on the VIMOS U-band image. From these images, we see that the emission-line region corresponds to a faint diffuse U-band continuum structure. The object appears very similar to higher redshift Lyα blobs (e.g., Prescott et al. 2012).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We have marked the nearest X-ray-selected AGN (circle) on the final thumbnail in Figure 14. It lies at about 14'' from the blob. However, this AGN has a redshift of z = 0.740 and hence cannot be the excitation source for the blob. A more probable candidate (square) is an AGN that lies 25'' (170 kpc) from the blob. Its redshift of z = 0.982 roughly matches that of the blob (z = 0.977).

Matsuda et al. (2011) found 10 giant Lyα blobs at z = 3.1 in a comoving volume that is almost identical to the present one. This suggests that the comoving volume density of giant blobs has dropped by almost an order of magnitude between z = 3 and z = 1 (see also Keel et al. 2009). However, a larger sample and a fuller understanding of the selection effects in the two samples would be needed to refine this result. We postpone a detailed analysis to a future paper (I. Wold et al. 2012, in preparation), where we plan to conduct data cube searches of all of the deep GALEX grism fields.

Finally, we computed the Lyα luminosity function of the CDFS-00 LAE galaxy sample in the redshift range z = 0.67–1.16 using the 1/V technique (Felten 1976), calculating the accessible volumes from the area of the sample. We used a flux limit of 7 × 10⁻¹⁶ erg cm⁻² s⁻¹ and only included sources that were not classified as AGNs in any way and that have rest-frame EW(Lyα) ⩾ 20 Å. (Note that the EW(Lyα) ⩾ 20 Å criterion is what is normally used to define the high-redshift LAE population; e.g., Hu et al. 1998.) With these criteria, the sample includes two CDFS-00 LAE galaxies from the pipeline sample of Cowie et al. (2011, their Table 2), and the rest come from our data cube sample. We show the Lyα luminosity function (black squares) in Figure 16, along with the Lyα luminosity functions at z = 0.3 (blue dotted curve) and z = 3.1 (red solid curve) using the Schechter (1976) function fits given in Cowie et al. (2010) and Gronwall et al. (2007), respectively. The open squares show the computation when we do not correct for the effects of incompleteness, and the solid squares when we do using the results from the simulations (Figure 10). The dynamic range in luminosity is not large enough to justify fitting a Schechter function, but as we illustrate with the red dashed curve, the data can be well described by the Gronwall et al. (2007) luminosity function with the normalization (ϕ_⋆) reduced by a factor of 30.

Zoom In Zoom Out Reset image size

This confirms what was previously reported by Cowie et al. (2011). On the figure we also show the luminosity function they calculated for their five NUV < 22 pipeline extracted, optically confirmed, star-forming LAE galaxies from multiple fields with rest-frame EW(Lyα) ⩾ 20 Å (green open triangles). In addition, we show their luminosity function after they corrected it (green solid triangles) to allow for sources with fainter continuum magnitudes (see Deharveng et al. 2008 and Cowie et al. 2010 for details). Since many of the LAEs with Lyα luminosities ∼10⁴³ erg s⁻¹ have NUV > 22, this correction is substantial. It is these corrected points that we should compare with the present luminosity function.

The z = 0.67–1.16 luminosity function presented here needs to be corrected for contamination from any remaining unidentified AGNs, which will slightly lower the overall normalization. We postpone these corrections to a subsequent paper, where we will present LAE catalogs determined from data cube extractions of all the deepest GALEX grism fields, together with the optical spectroscopic follow-up necessary to remove the residual AGNs in the sample (I. Wold et al. 2012, in preparation). However, even allowing for this, we can see that there is a dramatic evolution in the luminosity function between z ∼ 0.3 and z ∼ 0.9 and that high-luminosity Lyα sources seen at still higher redshifts are already present at z ∼ 1.

In this paper, we described a new method for obtaining a flux-limited sample of LAEs from GALEX grism data. We showed how for intensively observed fields with large numbers of independent grism images we could construct three-dimensional (two spatial axes and one wavelength axis) data cubes from the grism images. We then constructed the data cube for the central region of the CDFS-00, the field with the deepest GALEX grism observations, and used it to carry out a search for emission-line objects.

We detected 28 new z  ∼  1 LAEs in our data cube search of the CDFS-00. By comparing with the ECDFS X-ray data, existing optical spectroscopy, and deep U-band imaging, we determined that the bulk of our new LAEs are real and that the UV spectroscopic redshifts based on the Lyα identifications are reliable. In looking at the high-resolution HST images that are available for some of the LAEs in our new sample, as well as for some of the LAEs from the GALEX pipeline sample, we found that the sources classified as AGNs are dominated by compact nuclei, while the sources that appear to be star-forming galaxies are primarily resolved small blue galaxies, often with multiple components. One of our new LAEs is spatially resolved with a linear major-axis diameter of 120 kpc. This places it in the category of the giant Lyα blobs, though at the low end in terms of its Lyα luminosity.

The LAEs that had previously been found by the GALEX pipeline were limited by the pipeline magnitude limit of NUV = 22, but the new LAEs extend to much fainter magnitudes. Moreover, they also have higher EWs, since for a given Lyα flux and redshift, the NUV = 22 continuum cut corresponds to an EW, which becomes progressively lower as one moves to lower Lyα fluxes. The consequences of this are severe: the pipeline progressively misses more sources from a Lyα flux of ∼8 × 10⁻¹⁵ erg cm⁻² s⁻¹ down to a Lyα flux of ∼2 × 10⁻¹⁵ erg cm⁻² s⁻¹, at which point it misses nearly all of them.

Finally, we computed the CDFS-00 LAE galaxy luminosity function for the redshift interval z = 0.67–1.16 using both the very small sample of LAE galaxies from the pipeline sample of Cowie et al. (2011; two galaxies) and the larger sample of LAE galaxies from our data cube search. We confirmed a dramatic evolution in the luminosity function between z ∼ 0.3 and z ∼ 1.

In the future, we plan to undertake data cube searches on the other deep GALEX fields to obtain a much larger sample of z ∼ 1 LAEs (I. Wold et al. 2012, in preparation). This will enable us to make a precise determination of the luminosity function at this redshift. We also plan to do extensive optical spectroscopic follow-up to better assess the level of AGN contamination in the z ∼ 1 LAEs.

We thank the referee for providing an interesting and thoughtful report that helped us to improve the paper. We thank Michael Cooper for supplying spectra from the Arizona CDFS Environment Survey (ACES). We gratefully acknowledge support from the University of Wisconsin Research Committee with funds granted by the Wisconsin Alumni Research Foundation and the David and Lucile Packard Foundation (A.J.B.), as well as from NSF grant AST-0709356 (L.L.C.).