Spectroscopic Observations of Lyman–Break Galaxies at Redshift \sim 4, 5 and 6 in the GOODS–South FieldBased on observations made at the European Southern Observatory Very Large Telescope, Paranal, Chile (ESO programme 170.A-0788 The Great Observatories Origins Deep Survey: ESO Public Observations of the SST Legacy / HST Treasury / Chandra Deep Field South). Also based on observations obtained with the NASA/ESA Hubble Space Telescope obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc. (AURA) under NASA contract NAS 5-26555.

Spectroscopic Observations of Lyman–Break Galaxies at Redshift ∼ 4, 5 and 6 in the GOODS–South Field1

Abstract

We report on observations of Lyman–break galaxies (LBGs) selected from the Great Observatories Origins Deep Survey (GOODS) at mean redshifts , 5 and 6 (–, – and –band dropouts, respectively), obtained with the red–sensitive FORS2 spectrograph at the ESO VLT. This program has yielded spectroscopic identifications for 114 galaxies ( 60% of the targeted sample), of which 51 are at , 31 at , and 32 at . We demonstrate that the adopted selection criteria are effective, identifying galaxies at the expected redshift with minimal foreground contamination. Of the 10% interlopers, 83% turn out to be Galactic stars. Once selection effects are properly accounted for, the rest–frame UV spectra of the higher–redshift LBGs appear to be similar to their counterparts at . As at , LBGs at and are observed with Ly both in emission and in absorption; when in absorption, strong interstellar lines are also observed in the spectra. The stacked spectra of Ly absorbers and emitters also show that the former have redder UV spectra and stronger but narrower interstellar lines, a fact also observed at and . At , sensitivity issues bias our sample towards galaxies with Ly in emission; nevertheless, these spectra appear to be similar to their lower–redshift counterparts. As in other studies at similar redshifts, we find clear evidence that brighter LBGs tend to have weaker Ly emission lines. At fixed rest–frame UV luminosity, the equivalent width of the Ly emission line is larger at higher redshifts. At all redshifts where the measurements can be reliably made, the redshift of the Ly emission line turns out to be larger than that of the interstellar absorption lines, with a median velocity difference  km s at 4 and 5, consistent with results at lower redshifts. This shows that powerful, large–scale winds are common at high redshift. In general, there is no strong correlation between the morphology of the UV light and the spectroscopic properties. However, galaxies with deep interstellar absorption lines and strong Ly absorption appear to be more diffuse than galaxies with Ly in emission.

cosmology: observations — galaxies: formation — galaxies: evolution — galaxies: distances and redshifts

1 Introduction

The study of galaxies at high redshift is crucial for understanding the formation of the Hubble sequence, the growth of visible structures in the Universe and the processes leading to the reionization of the intergalactic hydrogen at the end of the Dark Ages. In the past decade, the empirical investigation of galaxies at high redshifts (i.e., ) has made rapid progress thanks to advances in telescopes and instrumentation and to the development of optimized selection techniques based on the observed colors of galaxies through either broad or narrow passbands. Color–selection criteria, which are designed to target galaxies with a range of spectral energy distributions (SEDs) within a targeted redshift window, are generally very efficient and allow one to build large samples with reasonably well controlled systematics (e.g., Steidel et al., 1999; Daddi et al., 2004; van Dokkum et al., 2003; Taniguchi et al., 2005), suitable for a broad range of studies, both statistical in character or based on the properties of the individual sources.

Among the various types of galaxies at high redshifts identified by color selection, the Lyman–break galaxies (LBGs; e.g., Guhathakurta et al., 1990; Steidel et al., 2003; Giavalisco et al., 2004b, for a review, see Giavalisco 2002) are the best studied and their samples are the largest both from a statistical point of view, and in terms of the the cosmic time covered (reaching back to less than one billion years after the Big Bang). The reason is mostly practical: since these galaxies are selected on the basis of luminous rest–frame ultraviolet (UV) emission, which at redshifts is redshifted into the optical and near–infrared windows, the observations are among the easiest to carry out, taking advantage of very sensitive instrumentation with large areal coverage.

LBGs at redshift have been intensively studied with both very large (Steidel et al., 2003) and deep samples (Papovich et al., 2001), including high–quality spectra for more than a thousand galaxies. Current surveys at provide the largest data set to study the properties of galaxies during a relatively early phase of galaxy evolution ( corresponds to when the Universe was % of its current age), at least for one spectral type, namely star–forming galaxies with moderate dust obscuration. A number of follow–up studies have been carried out following the discovery of these galaxies (Steidel et al., 1996a, b), including studies of their morphology and size (Giavalisco et al., 1996; Papovich et al., 2003; Ravindranath et al., 2006; Law et al., 2007; Lotz et al., 2006; Ferguson et al., 2004), their ages and stellar masses (Papovich et al., 2000; Shapley et al., 2000; Dickinson et al., 2003a), their chemical evolution (Pettini et al., 2000; Shapley et al., 2003), and of their clustering properties (Giavalisco et al., 1998; Adelberger et al., 1998; Giavalisco & Dickinson, 2001; Adelberger et al., 2003).

At higher redshifts, LBGs appear fainter, the observations require higher sensitivity and the samples are still relatively small. As a consequence, the properties of the most–distant LBGs are less well characterized. Initial studies of the spectral properties and luminosity function have been carried out at (Steidel et al., 1999) and (Madau et al., 1998) based on fairly small samples. More recently, larger samples at have been gathered, from both ground– and space–based observatories (Shimasaku et al., 2005; Iwata et al., 2003; Giavalisco et al., 2004b; Dickinson et al., 2004; Bunker et al., 2004; Bouwens et al., 2006, 2007); these samples, however, are exclusively photometric ones, with small numbers of spectroscopic identifications. Such studies have investigated a wide spectrum of statistical properties of LBGs at and up to , such as spatial clustering, morphology, UV luminosity function, stellar mass and properties of the Ly line (see, for example, Hamana et al., 2006; Lee et al., 2006; Ravindranath et al., 2006; Lotz et al., 2006; Giavalisco et al., 2004b; Ferguson et al., 2004; Ouchi et al., 2005; Bouwens et al., 2007; Ando et al., 2006, 2007; Yan et al., 2006). However, these results are based on the assumption that the spectral properties of LBGs at are the same as at . It is reasonable to expect that the higher–redshift samples should bear a similarity to those at , since LBG color selection at all redshifts are tuned to select galaxies with a similar rest-frame UV SED. However, without spectroscopic information, it is impossible to know if evolutionary effects are introducing systematic biases in the observed statistics. For example, if the distribution of surface brightness, UV SED or Ly emission line properties evolve with redshift, this will affect the redshift distribution and the completeness of the samples, which in turn will bias derived properties such as the LBG spatial clustering, luminosity function, and the evolution of these quantities.16

In this paper we present results from a program of spectroscopic follow–up of LBGs at redshift selected from optical images obtained with the Advanced Camera for Surveys (ACS; Ford et al. 1993) on board the Hubble Space Telescope (HST) in the four passbands, , ,  and , as part of the Great Observatories Origins Deep Survey, or GOODS (for an overview of the GOODS project, see Renzini et al., 2002; Dickinson et al., 2003b; Giavalisco et al., 2004a). The spectra have been obtained at the ESO VLT with the FORS2 spectrograph. The data from this program, specifically all the spectra of galaxies in the redshift range 0.5-6.3, have already been released and described in previous papers (Vanzella et al. 2005, 2006, 2008). Here we focus on a sample of LBGs, which has yielded 114 spectroscopic identifications in the redshift interval . Re-analyzing the whole LBG sample, we introduce very few differences (mainly in the quality redshift) to respect the previous global release (Vanzella et al.  2008), improvements that have been marked in the reported list of the present work. Currently, it represents one of the largest and most homogeneously–selected spectroscopic samples in this redshift range.

Throughout this paper magnitudes are in the AB scale (Oke 1974), and the world model, when needed, is a flat universe with density parameters , and Hubble constant km s Mpc.

2 Data and Sample Selection

2.1 ACS Images and Source Catalogs

We have selected samples of LBGs at mean redshift , 5 and 6 (in the following referred to as –, –, and –band dropouts, respectively) from the latest version (v2.0) of the GOODS images, obtained with the ACS on HST. The v2.0 mosaics are nearly identical in shape and size to the v1.0 ones. They cover the two GOODS fields, the northern one encompassing the Hubble Deep Field North (HDF–N) and the southern one located at the center of the Chandra Deep Field South (CDF–S). Each subtend an area of approximately arcmin on the sky, for a total areal coverage of about 0.1 square degrees. As with v1.0, the v2.0 images consist of two sets of mosaics observed in the , ,  and  filters. The depth of the ,  and  mosaics, however, has been increased over version v1.0 by including additional observations taken during the continuation of the original GOODS survey for high-redshift Type Ia supernovae (Riess et al., 2004, 2005). Since these additional data were obtained using the same observational strategy as the original ACS program (e.g., the same Phase-II files were used to carry out the observations), integrating them into the existing mosaics has been straightforward and has resulted in doubling the original exposure time in the  band as well as a more modest depth increase in the other bands.17

2.2 Photometric Samples of Lyman–Break Galaxies

We have selected samples of LBGs using color criteria very similar to those presented by Giavalisco et al. (2004b; G04b hereafter), with some minor modifications applied to the definition of –band dropouts to explore the redshift distribution of galaxies near the border of that color–color selection window. The exact locations of such windows balance the competing desires of completeness and reliability. Windows are designed to include as complete a sample of target galaxies as possible given the dispersion of observed colors — due both to both observational scatter and the intrinsic dispersion in galaxy UV SEDs (e.g., related to varying dust content, ages, metallicities, Ly equivalent widths, etc.). On the other hand, windows are designed to avoid significant numbers of galaxies at redshifts outside (usually lower than) the targeted one.

In the present work, –band dropouts are defined as objects that satisfy the color equations:

 (B450−V606)≥1.1+(V606−z850)   ∧   (B450−V606)≥1.1   ∧   (V606−z850)≤1.6, 1.1

where and are the logical AND and OR operators. These criteria extend the selection of candidates to slightly bluer (-) and redder (-) colors than those in G04b. As can be seen in Figure 1, which shows the selection windows corresponding to both sets of color equations, the sample selected with the new criteria (solid line) fully includes the one selected with the G04b criteria (dashed line). We have decided to use these more general criteria to define the sample of –band dropouts, which is the largest among the three LBG samples targeted for the spectroscopic observations, to explore both changes in the low–end of the targeted redshift range and contamination rates from low–redshift interlopers.

The definitions of the color equations of –band and –band dropouts are unchanged from those used in G04b and Dickinson et al. (2004), and are given by the color equations

 [(V606−i775)>1.5+0.9×(i775−z850)]   ∨
 ∨   [(V606−i775)>2.0]   ∧   (V606−i775)≥1.2   ∧   (i775−z850)≤1.3
 ∧   [(S/N)B<2] 1.2

and

 (i775−z850)>1.3   ∧   [(S/N)B<2]   ∨   [(S/N)V<2], 1.3

respectively (see Figure 2 for a collapsed representation of the selection windows for – and -band dropouts). For all three selection criteria above, when the isophotal in a given band is less than one, limits on the colors have been calculated using the 1 error on the isophotal magnitude.

We have restricted the photometric samples to galaxies with isophotal in the  band, and we have visually inspected each candidate, removing sources that were deemed artifacts. In addition, we have estimated the number of spurious detections using counts of negative sources detected in the same data set. Together, these amount to a negligible number of spurious sources for the – and –dropout samples, and % for the –dropouts. We have also eliminated all sources with stellar morphology down to apparent magnitude , i.e., where such a morphological classification is reliable. This accounts for an additional %, % and % of the –, – and –dropout samples, respectively. While this procedure biases our samples against LBGs (and high–redshift quasars) that are unresolved by ACS, it minimizes contamination from Galactic stars. Note that we have spectroscopically observed a few point-like sources that obey the dropout selections in order to verify that such sources are indeed Galactic. In practice, these cullings of the dropout samples result in negligible changes to the spectroscopic samples and to key measured quantities, such as the specific luminosity density.

Down to , roughly the % completeness limit for unresolved sources, the culled samples include 1544, 490 and 213 –, –and –band dropouts, respectively. With a survey area of 316 arcmin, this corresponds to surface density , and galaxies per arcmin for the three types of dropouts, respectively. Error bars simply reflect Poisson fluctuations.

We note that while the – and -band dropout samples are mutually exclusive (i.e., vs. , respectively), the intersection of the – and -band dropout samples may be non-zero. However, in this latter case, no sources in common have been found down to the , and only one galaxy satisfies both criteria when the magnitude limit is extended down to (i.e., GDS J033245.88-274326.3).

We have used Monte Carlo simulations to estimate the redshift distribution function of our LBG samples and compared the results to observations. The technique is the same as that used in G04b and consists of generating artificial LBGs distributed over a large redshift range (we used ) with assumed distribution functions for UV luminosity (we used a flat distribution, discussed below), SED, morphology and size. We adjusted the input SED and size distribution functions by requiring that the distribution functions recovered from the simulations match the observed sample, the largest of the three GOODS samples. In this way, both simulations and observations are subject to similar incompleteness, photometric errors (in flux and color), blending, and other measurement errors. The model SED used for the simulations is based on a synthetic spectrum of a continuously star–forming galaxy with age yr, Salpeter IMF and solar metallicity (Bruzual & Charlot 2000). We reddened it with the starburst extinction law (Calzetti 2000) and randomly extracted from a Gaussian distribution with and . In other words, the dispersion of the LBG UV SEDs is modeled as only due to the dispersion in the amount of obscuration for the same unobscured SED, neglecting the effects of age and metallicity of the stellar populations. This is obviously a crude approximation, but, thanks to the strong degeneracy between age, obscuration and metallicity on the broad–band UV colors of star–forming galaxies, it is adequate here since we are only interested in measuring the selection effects due to the specifics of the observations. For the cosmic opacity, we have adopted the Madau (1995) prescription, extrapolated to higher redshifts when necessary. To model the dispersion of the morphologies of the galaxies we have used an equal number of and exponential profiles with random orientation, and size extracted from a log–normal distribution function (see Ferguson et al., 2004). We found the average redshift and standard deviation of the redshift distribution to be and for the –dropout sample, and for the –dropout sample, and and for the –dropout sample.

2.3 The Spectroscopic Sample

We have selected a sample of 202 LBGs from the three samples defined above as primary targets of the FORS2 spectroscopic observations. While the criteria to include a galaxy in the target list were mostly based on its apparent magnitude, as we detail below, we did not set a strict flux limit for the spectroscopic sample in these initial high-redshift LBG spectroscopic studies. This allowed us to empirically assess how often the presence of Ly emission allows the measurement of the redshift of galaxies which are too faint for absorption spectroscopy.

Targets were assigned slits in the FORS2 multi–object spectroscopic masks according to an algorithm in which two competing factors combine to maximize (i) the number of targets and (ii) the likelihood of success, under the assumption that brighter targets are more likely to result in successful identifications. In practice, while brighter galaxies were more likely to be assigned a slit, (slightly) fainter targets could still win the competition if their coordinates allowed a larger total number of targets on a given mask. Relatively faint targets in close proximity to brighter one were also assigned a slit if their inclusion could be made without penalty. Where possible, we assigned faint targets to multiple masks.

When the number of available slits in a mask exceeded that of available targets, we populated the remaining slits with “filler” targets selected to test target selection criteria and to identify lower–redshift galaxies in the range (for a summary of the global target selection of the FORS2 campaign, see Vanzella et al., 2006). In particular, we selected some filler targets using LBG criteria that extended the primary – and –band dropout selection criteria to galaxies with less pronounced “Lyman drops” and bluer UV continuum, thus getting closer to the locus of general field galaxies. As discussed earlier, such observations are useful for exploring the dependence of the redshift distribution function of the confirmed LBGs on the details of the color selection, as well as for measuring contamination by low–redshift interlopers. In what follows we refer to these more loosely defined – and –band dropouts simply as “fillers”. Only three such spectroscopically-identified fillers are considered below, two –dropout fillers (GDS J033234.40-274124.3 at , QF=B and GDS J033251.81-275236.5 at , QF=A) and one –dropout filler (GDS J033239.82-275258.1 at , QF=C). We will report more extensively on these tests in following papers, which will also include spectroscopic observations of GOODS galaxies obtained with different instrumental configurations.

3 FORS2 Spectroscopic Observations

The details of the observations, including journals of the observing runs, data reduction, the extractions of the spectra have been reported in Vanzella et al. (2005, 2006, 2008), and we refer the reader to those papers. We recall that the wavelength coverage was typically 5700Å-10000Å with a spectral resolution of , corresponding to 13Å at 8600Å. No order separation filter was used.

In the vast majority of cases, the redshift has been calculated through the identification of prominent features of LBG spectra, e.g., Ly either in emission or absorption, and Si ii 1260Å, O i+Si ii 1302Å (a blend at the spectral resolution of our instrumental setup), C ii 1335Å, Si iv 1394,1403Å, Si ii 1527Å, C iv 1548, 1551Å in absorption.

Redshift determinations have been made based on visual identification of spectral features as well as by cross–correlating the observed spectra against high–fidelity LBG templates of differing spectral types using the rvsao package in the IRAF environment. In particular, we used emission and absorption line LBG templates from Shapley et al. (2003) as well as the lensed absorption line LBG cB58 (Pettini et al.  2000). Each two-dimensional spectrum has been visually inspected, including consideration of its slit orientation on the sky. In many cases where no continuum has been detected, we derive a redshift measurement from Ly emission.

We have co–added all repeated spectra to improve the final S/N. The typical exposure time for each mask was about 14,400 seconds and for co-added sources, total exposure times range from 20,000 to 80,000 sec (e.g., see Vanzella et al., 2008).

We have assigned each measured redshift a quality flag (QF), with values of either A (unambiguous identification), B (likely identification; e.g., based on only one line or a continuum break), or C (uncertain identification). The presence of Ly emission in the second order spectrum (at 10,000Å), has also been used on occasion, especially for faint sources with low QFs based on absorption features. For example, one –band dropout source (GDS J033221.05-274820.5) shows an apparently featureless continuum with a second-order emission line at 10400Å, implying 3.3. Indeed, recently the VIMOS spectroscopic observations have confirmed this galaxy to be at (Popesso et al. 2008).

We have assigned a redshift to 118 galaxies of the initial list of 202 targets, or 58.4 of the input list; this relatively low success rate is, in large part, due to two factors: (i) the target list includes a relatively large fraction of faint sources — 65 or 32.2% of the sample have ; and (ii) the difficulty in deriving redshifts for galaxies at with our instrumental configuration. In the latter case, depending on the slit position, Ly and the UV absorption features are often blueward of the spectral range available.

Of the 118 spectroscopically identified sources, 106 have redshifts in the expected range for their adopted color selection. Note that some of these redshifts have already been published in Vanzella et al. (2005, 2006). Of the sources outside the expected redshift range, one source is a low–redshift galaxy from the –band dropout sample, one is a low–redshift galaxy from the –band dropout sample, and 10 are Galactic stars (1, 3 and 6 from the –, –, and –dropout samples, respectively). We note that one faint star, GDS J033238.80-274953.7 ( = 25.16), that we spectroscopically classify with QF=C, has been confirmed Galactic in nature due to the detection of its proper motion (M. Stiavelli, private communication). Excluding the Galactic stars and the two low–redshift interlopers, the final list of spectroscopically identified LBGs includes 46 –band dropouts, 32 –band dropouts, and 28 –band dropouts (reported in Tables 1, 2 and 3).

As mentioned above, we have assigned redshifts to three high–redshift filler targets found to be in the same redshift range as the primary LBG sample. This brings the total number of high-redshift () spectroscopic identifications to 109, of which 32 have QF=C. Of these 109 galaxies, 70 have redshift (24 with QF=C); 37 have redshift (13 have QF=C); and 32 have redshift (11 with QF=C; see Table 4 for a summary).

Finally, we also found five serendipitously–identified, high–redshift galaxies. These fell, as second or third sources, on slitlets assigned to other primary targets. For four of them the redshift identification relies upon a Ly emission line; only in one case does the redshift rely upon absorption features. These five serendipitous sources are, in right ascension order:

• GDS J033218.27-274712.0, at (QF=C), marked in Figure 2 with a red pentagon, is close to the -band dropout selection window (-1.901, -=0.501).

• GDS J033219.41-274728.4, at (QF=C). This galaxy shows a flat continuum and an absorption doublet interpreted as C iv 1548, 1551Å.

• GDS 03322.89-274521.0, at (QF=C). This source, clearly visible in the  band (from which the coordinates were measured) is not detected in the , , or  bands. We assume that the emission line is Ly, though lacking firm constraints on the continuum SED, we can not rule out another interpretation such as [O ii]3727 at redshift . This source is not used in the following analysis.

• GDS J033228.94-274128.2, at (QF=B), is discussed in Vanzella et al. (2005, see their Fig. 13). The source is not present in the ACS catalogs because of blending with a bright galaxy.

• GDS J033243.16-275034.6, at (QF=C), is discussed in Vanzella et al. (2006, see their Fig. 2, top panel). This source is not present in the ACS catalogs because of blending with a bright star.

Figures 3 and 4 show the one-dimensional spectra for all confirmed LBGs, separated depending on whether Ly is in emission or absorption. Figure 5 shows the two-dimensional spectra of confirmed LBGs at . Table 4 summarizes the characteristics of each dropout sample compared with those expected from the Monte Carlo simulations of the redshift selection described in §2.2, while Figure 6 shows the observed redshift distribution of each dropout category. Note the (small) overlap between the redshift distribution functions for – and –band dropouts at and between – and –band dropouts at .

4 Efficiency of the Photometric Selections

The effectiveness of the LBG color selection has been verified at by means of an extensive program of spectroscopic confirmations of over a thousand –band dropouts (Steidel et al., 2003). At higher redshift, the spectroscopic samples of LBGs collected by various groups (e.g., Steidel et al., 1999; Vanzella et al., 2006, 2008; Popesso et al., 2008; Yoshida et al., 2006; Ando et al., 2007) are rather small and estimates of successful identification rates for a given set of color criteria remain correspondingly uncertain. Details of the filters and color criteria used by various surveys can result in different relative proportions of successfully-identified LBGs (i.e., in the targeted redshift range), interlopers (i.e., outside of the targeted redshift range), as well as other types of unwanted sources (e.g., AGN in the targeted redshift range). The GOODS data set is being widely used for a variety of studies of the properties of galaxies at high redshifts (e.g., Bouwens et al. 2007), and many of these studies use samples of photometrically–selected, high–redshift galaxies from the GOODS ACS data without spectroscopic verification of the effective composition of the samples. Even under the most optimistic assumption that the samples include negligible fractions of interlopers and unwanted sources, fundamental quantities such as the shape of the redshift distribution, which is important for the measures of the spatial clustering and luminosity function, remain largely unknown.

The spectroscopic sample obtained with FORS2 discussed here is our initial effort to characterize the effectiveness of the GOODS LBG color selection criteria in selecting star–forming galaxies at high redshifts; the numbers cited below are summarized in Table 4.

Of the 85 –band dropout candidates selected for spectroscopic observations, we have secure redshifts for 48 sources down to   (56%). Of the 48 identifications, 46 have redshifts in the expected range for –band dropouts, and only two are foreground objects. One is a Galactic star (QF=B, =23.43, SExtractor stellarity index S/G=0.99) and the other is a galaxy at identified from [O ii] 3727 emission (QF=B, =25.49)18. We have classified 8 of the 48 identifications as having QF=C. Assuming that all the identifications with QF=C are correct, the efficiency of the –band dropout selection is 46/48=96. Omitting the two foreground sources, the mean and rms of the –band dropout redshift distribution are and =0.328, respectively, fully consistent with the prediction from Monte Carlo simulations (see §2.2 and Table 4).

Figure 1 shows the (-) vs. (-) color–color diagram for the entire FORS2 spectroscopic sample. The region of –band dropout sources is marked with a solid line and the size of the symbols scale linearly with redshift for sources at ; at redshift lower than 3.1, the size of the symbol is fixed. The majority of the galaxies with lie in the –band dropout region. It is evident from Figure 1 that the lower tail of the redshift distribution is located in the lower part of the selection region: the eight sources with have a mean (-)=1.690.23 and (-)=0.440.21. In this redshift range, the selection criteria are more uncertain and depend on the intrinsic properties of the sources. Photometric errors may also scatter sources across the boundary of the selection region.

For the sample of –band dropouts, we have assigned spectroscopic slits to 52 candidates and derived a redshift identification for 36 of them (69%) down to , of which 11 have been given QF=C. Among the confirmed redshifts, 32 are in the range expected for –band dropouts (11 with QF=C), three are Galactic stars (all of them with and S/G=0.99), and one, GDS J033220.31-274043.4, is a low–redshift interloper at (QF=B)19. Assuming that all of the QF=C identifications are correct, the efficiency of the LBG selection is 32/36=89; omitting the foreground sources, the mean and rms of the redshift distribution are and =0.386, respectively. Again, the observed redshift distribution agrees well with the Monte Carlo predictions (Table 4).

Figure 2 shows the (-) vs. (-) color–color diagram for the entire FORS2 spectroscopic sample. The selection window for the –band dropouts (solid lines) and –band dropouts (dotted line; 1.3) are plotted. Galaxies confirmed in the redshift interval are marked with open circles. The majority of galaxies at are located within the selection regions.

Finally, of the 65 –band dropouts selected for spectroscopic observations, we have secured redshifts for 34 down to (52%). Of these, 28 have redshifts in the range , of which 23 are based on the identification of an observed emission line as redshifted Ly (seven have QF=C) and five are based on the identification of an observed continuum break as the onset of the high–redshift Ly forest (four have QF=C and one has QF=B). The remaining six –band dropouts are Galactic stars (four with QF=C and two with QF=B; all with stellarity index S/G0.91 and 25.4). Assuming that all of the QF=C identifications are correct, the efficiency of the LBG selection is thus 28/34=. The average redshift and standard deviation of the successfully identified –band dropouts are =5.898 and =0.184. While the average of the distribution is consistent with the predicted one, we note that the standard deviation is almost a factor of two narrower. This may be an indication of large–scale structure at this redshift; from ACS grism spectroscopy, Malhotra et al. (2005) note structure at this same mean redshift in the HUDF.

We note that several sources from our –dropout spectroscopic sample were previously published, including spectroscopic observations. One has a well-observed spectrum showing Ly emission at (Stanway et al., 2004a; Dickinson et al., 2004; Bunker et al., 2004). An additional two sources were observed with the low–dispersion ACS grism and show strong spectral breaks interpreted as due to the Ly forest at z5.9 (Malhotra et al., 2005). These galaxies are present in our list with redshifts and 5.95 (see Table 3). The source GDS J033234.55-274756.0, for which the FORS2 spectrum yielded an inconclusive redshift determination, is presented in Malhotra et al. (2005) at redshift .

Figure 2 shows the selection diagram for the -band dropouts (dotted line). Galaxies confirmed in the redshift interval are marked with open squares. The majority of galaxies at have (-) redder than 1.3, as per the adopted selection criteria. The confirmation of galaxies at redshift beyond five are almost exclusively due to the presence of a single emission line, identified as Ly (see §5.4 for a dedicated discussion).

In the redshift interval , the – and –band dropout selection criteria overlap. In this redshift range, five spectroscopically–confirmed galaxies meet our –band dropout selection criteria, while two meet the –band dropout criteria. As discussed below, the presence of the Ly emission line may play an important role in this respect.

Finally, as can be seen in Figures 1 and 2, of 114 high–redshift galaxies (109 targeted and five serendipitous), 12 are outside of the primary color selection windows. Three of them are the above mentioned “fillers”, five are serendipitous sources discussed in §3, and nine are galaxies with colors close to the –, – or –band dropout selection windows (the mean “distance” in terms of color from the selection windows is 0.04). These galaxies were selected as –, – or –band dropouts from the previous (v1.0) ACS catalog, though in the current v2.0 catalog, they no longer meet the dropout selection criteria (see Table 5). We further note that seven out of these nine galaxies have been classified with QF=C. In particular, GDS J033233.52-275532.2, an -band dropout at redshift (QF=C), satisfied the –band dropout selection criteria using the v1.0 catalog (in the v1.0 catalog, -=1.791), but not using the v2.0 catalog; it is no longer identified at S/N5 in the  band. A visual inspection of the  image suggests a faint source, as evident in Figure 7. However, further investigations will be needed to clarify this target; we do not include this source in the following analysis.

5 Composite Spectra

We now describe, for each category of dropout, the general spectral properties observed. As in the case of LBGs at , and depending on the S/N ratio, the most prominent rest–frame UV features observed in our samples are the HI Ly line (seen in emission, absorption, or a combination of both), low-ionization, resonant interstellar metal lines such as Si ii 1260Å, O i + Si ii 1302Å, C ii 1335Å, Si ii 1527Å. Fe ii 1608Å and Al ii 1670Å, and high-ionization metal lines such as Si iv 1394,1403Å and C iv 1548, 1550Å associated with P-Cygni stellar wind features and ionized interstellar gas. In one case, N iv] 1485Å emission has been detected together with Ly in emission (GDS J033218.92-275302.7; Vanzella et al.  2006). As to be expected, the number of robustly identified lines decreases drastically with apparent magnitude over the range from to . At the faintest magnitudes, given our typical exposure times, continuum flux is at the limit of measurability (S/N1 per resolution element) and therefore no absorption lines are reliably observed; the only observable feature for the faintest sources is Ly emission.

5.1 B435-Band Dropout Composite Spectra

Among 46 –band dropouts with spectroscopic redshift at , 15 of them show Ly in emission line (the “em.” class), 21 have redshift identified by means of absorption lines only (the “abs.” class) — typically SiII 1260.4Å, CII 1335.1Å, SiIV 1393.8,1402.8Å, CIV 1548.2,1550.8Å — and 10 sources show both emission and absorption features (the “comp.” class). As mentioned above, two -band dropouts have Ly blueward of the observed spectral range, but this line was visible in second order at 10,000Å; only absorption features were used to derive the redshifts for these sources. These two galaxies have not been used to make the composite spectra.

The composite spectra, normalized at 1450Å, for emission line sources (“em.” class), emission and absorption line sources (“comp.” class) and absorption line sources (“abs.” class) at 3.8 are shown in the left panel of Figure 8. The composite spectra include sources with QF = A, B and C. A continuum break blueward of Ly, due to the intergalactic medium, is clearly evident. Stellar and interstellar lines are also easily recognized.

The absorption lines clearly differ between the “em.”, “comp.” and “abs.” classes. Figure 9 superposes the composite spectra of the -band dropouts with and without the Ly emission line (“em.” and “abs.”). Low-ionization interstellar absorption lines are more pronounced in the “abs.” class composite spectrum; e.g., compare the OI, CII and FeII lines. Figure 9 also shows that the non-emitter population has a redder spectral slope, consistent with the previous work based solely on photometric data; e.g., Pentericci et al. (2007) find and , where . A similar trend has also been noted by Shapley et al. (2003) from their sample of LBGs. In particular, Shapley et al. (2003) find that the average extinction, , decreases as a function of increasing Ly emission strength. The similar trends seen here at suggests that the emission line –dropout LBGs are, on average, less extincted than the absorption line –dropout LBGs (c.f., Pentericci et al., 2007).

5.2 V606-Band Dropout Composite Spectra

As reported above, 32 –band dropouts are at . As expected for the larger distance modulus and correspondingly fainter sample, the fraction of spectroscopically–confirmed –band dropouts with Ly in emission is higher compared to the –band dropout sample; 19 sources are in the “em.” class and 13 show only absorption lines or a continuum break (“abs.” class). The rest–frame composite spectra of emission and absorption galaxies with QF = A, B and C are shown in the top left panel of Figure 8. Stellar and interstellar absorption lines, as well as the strong continuum discontinuity at Ly, are clearly observed in the composite spectra. The composite “em”-class spectrum looks quite similar to the –band dropout composite “em” spectrum. For the “abs.” class, the composite is dominated by low quality (QF=C) spectra and only the strong Lyman- forest break is apparent.

5.3 i775-Band Dropout Composite Spectrum

The rest–frame composite spectrum of the emission line (“em.” class) –band dropouts is shown in Figure 10. Among the 28 –band dropouts with spectroscopic redshift at , 22 sources show Ly in emission (7 with QF=C).

Given the exposure times, these galaxies are generally too faint to measure a continuum (e.g., see Figure 3 or Figure 5) and only Ly emission has been detected. As shown in Figure 11, this is particularly true for the fainter –band dropouts. Nevertheless, the composite –band dropout spectrum shows signal redward of the Ly line, with tentative detection of the Si ii 1260Å and O i + Si ii 1302Å absorption lines despite the sky lines at these wavelengths being stronger and denser. At these high redshifts (), we find a very opaque IGM blueward of the redshifted Ly line. Consistent with quasar results (e.g., Songaila, 2004), the IGM transparency is estimated to be of the order of 1.

5.4 Single–Line Redshift Identifications

For most of the LBGs in our sample, the redshift identifications are based on a single emission line — assumed to be redshifted Ly — in an otherwise featureless and/or low–S/N ratio spectrum. A question naturally arises: how robust are these identifications (e.g., Stern et al. 2000)? To be selected as dropouts, the broad–band SED of these galaxies must satisfy the color–selection criteria, which require the signature of the Lyman limit and/or Ly forest blanketing. The most plausible candidate for an alternate identification is [O ii]3727 at , though H at , [O iii]5007 at and H at are also possible. Such possibilities, however, will generally be inconsistent with the broad–band colors of the galaxies, since low–redshift solutions would be star–forming galaxies with relatively blue continua. This is illustrated in Figure 12, which plots the observed equivalent width versus the (-) color for [O ii]–emitting galaxies at redshift and Ly–emitting LBGs at redshift (only galaxies with QF=A are plotted). Color and equivalent width do an effective job at separatating the low–redshift, star–forming galaxies from their high–redshift counterparts, particularly for the –band dropouts. A few –band dropouts do overlap with the low–redshift galaxies, but are easily separated using (-) color, which is better suited for galaxies at .

The composite spectra of emission line –band dropouts (Figure 8, right) and –band dropouts (Figure 10) also provides evidence that most of the single–line Ly identifications are correct. The composite spectrum, which is a lower–S/N version but otherwise virtually identical to the composite spectrum (Figure 8, left), shows a number of absorption features that would not be observed due to dilution if most of the identifications were wrong. The composite spectrum is similar, except that the lower S/N ratio results in a lower–S/N detection, or no detection at all, of absorption features. The continuum discontinuity across the Ly, however, is clearly detected with a jump larger than one order of magnitude in the continuum flux density (in fact, the continuum blueward of the Ly line is consistent with being zero). This is larger than other continuum discontinuties observed in distant galaxies (c.f., Spinrad et al. 1998; Stern et al. 2000).

Finally, another discriminant between high–redshift and low–redshift single emission-line sources is provided by the line profile: high–redshift Ly lines are asymmetric due to intervening Hi absorption, while other lines will generally be symmetric. However, the low S/N, low spectral resolution (660) reported here makes the detection of a clear asymmetry challenging in most individual spectra. In a few cases, however, an asymmetric profile has been detected in the brigher Ly–emitting LBGs reported here.

6 Outflows at z∼ 4 and 5

Evidence of powerful winds in LBGs at (Shapley et al. 2003) and in galaxies at selected from UV colors (Shapley et al. 2005) has been inferred from the systematic redshift of the Ly emission line and the blueshift of interstellar absorption lines with respect to the systemic redshift of the galaxies, as traced by rest-frame optical nebular lines. In this scenario the redshifted Ly emission line forms in the receding part of a generally bipolar flow of gas, while the blueshifted interstellar lines originate in the part along the line of sight moving toward the observer.

6.1 Outflows in B435–Band Dropouts (z∼4)

It is of interest to see if LBGs at also show the same phenomenon, and compare its magnitude to that of the lower-redshift galaxies, looking for evolutionary effects. Obtaining spectroscopic observations of the rest–frame optical nebular emission lines is not a trivial task. The [O ii]3727, [O iii] 4959 and 5007Å lines have been identified for only nine galaxies in our sample, as a part of the AMAZE project, aimed at estimating the mass-metallicity relation at high redshift (Maiolino et al. 2008). In fact, these features become unreachable from the ground for redshift when the lines go beyond the -band. For such sources, information about the possible presence of winds is derived from the velocity differences between Ly emission and interstellar absorption lines.

The AMAZE project (Maiolino et al. 2008) has determined the redshift of nebular lines using the integral field spectrometer SINFONI at the VLT, adopting a spectral resolution =1500 in the spectral range m. For each source, the redshift derived from [O iii] 4959, 5007Å and [O ii]3727 agree within . We have calculated “nebular redshifts” for each galaxy by averaging these three lines. The redshift of the interstellar medium has been derived from the low-ionization interstellar absorption lines (ISL, e.g. Si ii 1260Å, O i+Si ii 1302Å, C ii 1335Å, and Si ii 1527Å), and the redshift of the hydrogen gas is estimated from the Ly line.

We then compare the various redshift estimates arising from the different physical regions within the LBGs, i.e. the velocities , and . We find that:

1. the relative median velocity observed between the Ly emission lines and the interstellar absorption lines is + (derived from 16 galaxies at an average redshift of 3.700.2). The Ly emission is always redshifted relative to the interstellar lines. Adopting the model of Verhamme et al. (2006), the velocity of the expanding neutral hydrogen shell is of order of ;

2. the relative median velocity between the Ly emission line and the nebular lines is + (derived from four galaxies at z 3.65);

3. the relative median velocity between the interstellar absorption lines and the redshift of the nebular lines is (derived from nine galaxies at z 3.7);

The galaxy GDS J033217.22-274754.4, with its peculiar, double-peaked Ly profile is already been discussed in detail in Vanzella et al. (2008).

Figure 13 shows the histogram of for the 16 galaxies from the -band dropout sample for which this measurement has been possible. The histogram does not include quality QF=C spectra. In all cases, the redshift of the Ly is measured by fitting a Gaussian profile to the line,20 while the redshift of the interstellar absorption lines is derived cross-correlating the individual spectra with templates (viz., the lensed galaxy cB58 and the composite spectrum without Ly emission from Shapley et al., 2003), after excluding the Ly line from the analysis. The typical redshift error is (Vanzella et al., 2008, derived from multiple, independent observations) and translates into a final error on the velocity difference at . Figure 13 shows that the Ly line is systematically redshifted relative to the interstellar absorption lines and a few galaxies have velocity differences in excess of .

Though derived from relatively small samples, these numbers are similar to LBGs at (Shapley et al., 2003; Adelberger et al., 2003). In particular, Figure 11 of Shapley et al. (2003) shows that with increasing Ly emission strength, the kinematic offset implied by the relative redshifts of Ly emission and low-ionization interstellar absorption lines decreases monotonically from to . If we assume this trend remains true at and consider the mean rest–frame Ly equivalent width of our sample (20 Å), the comparison is even more consistent with the results at . We also note that the derived by Adelberger et al. (2003) is similar to the value derived here at slightly higher redshift, .

6.2 Outflows in V606–Band Dropouts and at Redshifts Beyond 5

In the case of –dropouts (), the mean velocity difference is more difficult to estimate for individual galaxies because the S/N is generally lower, due to both the faintness of the targets and to the UV absorption features entering a spectral region affected by strong sky emission lines at . For this reason, we have resorted to estimating from the composite spectrum. All –band dropouts with Ly in emission have been co-added, registering their redshift with respect to theeir Ly lines. This procedure will lead to a slight smoothing of the interstellar lines and thus a larger uncertainty. Nevertheless, an absorption signal remains clearly detected in the composite spectrum.

The UV absorption features SiII 1260.4Å, CII 1335.1Å and SiIV 1393.8,1402.8Å (see Figure 8) show an average blueshift of with respect to the Ly line, similar to the –dropout results. With the aim of extending this measurement to yet higher redshift, we have selected a subsample of eight LBGs with detected continuum at from the  and –band dropout samples, at an average redshift of 5.6 and  magnitude 25.6 (three –band dropouts and five –band dropouts; QF=C LBGs have not been considered). Similar to the sample of pure –band dropouts, the composite spectrum shows a velocity offset of .

In order to check if the above estimations give realistic measurements of the offset, we have re-calculated from the –band dropout composite spectrum. We find . Though a bit higher, this value is consistent with the number derived from individual measurements.

This analysis performed therefore supports the interpretation that outflows at and are present and similar to those seen at lower redshifts ().

7 Lyα Equivalent Width and the UV Luminosity

For all galaxies with Ly in emission, we have estimated the rest frame equivalent width of the line. In the critical cases where this line is the only feature detected in the spectrum, the continuum has been estimated from the available photometry assuming a flat spectrum with spectral index (). Depending on the redshift, the  (),  () or () magnitudes have been used to determine the continuum level. In the highest–redshift case, we use magnitudes (the filter has a central wavelength of 1.24m and width of 0.16m, it allows an accurate photometry) from the GOODS-MUSIC catalog (Grazian et al., 2006), or the NIC3 F110W band magnitude (Thompson et al. 2006) for sources in the HUDF. If the magnitude is a lower limit, the resulting equivalent width is a lower limit (indicated by an arrow in the figures). The absolute magnitude has been derived from the -band, assuming a template (drawn from SB99) of a star–forming galaxy with spectral index .

Figure 14 shows the distribution of the rest–frame equivalent widths versus the absolute magnitude calculated at 1450Å for all sources in the sample. A cosmic time between 0.9 to 1.6 Gyr after the Big-Bang is covered (,  and -band dropouts are marked with different symbols). At fainter luminosities (), the estimated equivalent widths span a wide range of values, from a few Angstroms up to 300Å. There is a natural observational bias that the redshift of the faintest, high–redshift galaxies can only be measured if they contain a strong, high equivalent width Ly line. However, there is no such bias against high equivalent width for brighter galaxies, but these are not observed. This absence of large equivalent widths of Ly lines at bright luminosities has already been noted by several groups studying samples of Ly emitters (LAEs) and LBGs at redshift between 3 and 6 (e.g., Shapley et al., 2003; Ando et al., 2006, 2007; Tapken et al., 2007; Verhamme et al., 2008).

The equivalent width of the Ly line (or the escape fraction of the Ly photons) is related to the velocity expansion of the medium, the column density of the neutral gas , the dust extinction and the geometry of the media (clumpy or continuum geometry). A possible scenario is that the brighter galaxies are experiencing (or have already experienced) a higher burst of star formation and supernovae explosions with an associated production of dust. Thus, the more luminous galaxies would be dustier and more metal rich, have correspondingly more efficient Ly absorption, and thus exhibit lower observed Ly equivalent widths. Larger equivalent widths are expected for objects dominated by younger ( Myr) stellar populations; lower equivalent widths are expected in dusty and/or post-starburst galaxies (e.g., Schaerer & Verhamme, 2008). This hypothesis implies that brighter LBGs would be dustier, more chemically enriched, and show lower equivalent widths (Ly). One would expect that ultimately the main underlying parameter governing the trends with UV magnitude might be the galaxy mass.

Finally we note that fixing the redshift (i.e., the dropout flavor) in Figure 14, the deficiency of strong lines at bright UV magnitudes remains, though better statistics are clearly needed, particularly at the faint end of the redshift distributions.

On the other side of the distribution, the presence of large Ly equivalent widths for faint sources may be a combination of selection effects and intrinsic properties of these galaxies:

1. Observational bias:

• Spectroscopy. Obviously, from the spectroscopic point of view, faint galaxies (mainly –band dropouts) are confirmed thanks to the presence of a Ly emission line that can be observed also in the middle of the sky emission (see Figure 15). In the current spectroscopic sample, fainter galaxies tend to be at higher redshifts. Figure 16 shows the behavior of Ly luminosity versus redshift. There is an indication that the fraction of stronger lines increases with redshift.

• Photometry. Strong Ly emission also affects photometric color selection — in particular, for –band dropouts which rely on a single color (i.e.,   1.3). At , the contribution of Ly emission to the (-) color range up to 0.5 (0.8) magnitudes for Ly rest–frame equivalent widths of 100 Å  (150 Å), consistent with the measurements in our spectroscopic sample (see Figure 17). The (-) color is increased or decreased depending on the strength of the line and the redshift of the source. Two clear examples (marked with star symbols in Figure 17) are:

1. GDS J033218.92-275302.7 () shows a Ly emission line with rest–frame equivalent width of Å falling within the  band. This has the effect of reducing the apparent  color by 0.59 magnitudes. This galaxy has been selected as a –band dropout and is also discussed as a candidate “Balmer Break galaxy” based on its bright IRAC flux and apparent “break” in the m colors (Wiklind et al., 2008).

2. GDS J033223.84-275511.6 () shows a Ly emission line with no continuum detected in 80 ks of spectroscopy. The rest–frame equivalent width is Å, and the measured (-) color is a lower limit ((-) ). In this case the  apparent magnitude (and the (-) color) is increased by the line.

In order to explore such effects as a function of redshift, Ly equivalent width and  magnitude, we have calculated various color tracks as shown in Figure 18. We find that, when the Ly line enters the  band () and leaves the  band (), depending on the equivalent width, it favors the -band dropout selection criteria. For fainter sources (26.5), only the emitters tend to survive.

2. Intrinsic effects: the large spread in Ly equivalent widths at faint magnitudes () observed by numerous authors may also be due to a relatively small amount of dust, which would not filter out the stronger Ly lines, and to a larger variety of star formation histories and timescales — i.e., an enhanced role of “stochastic star formation events.” Such a scenario is most likely to have a strong effect for galaxies of smaller absolute scale (either mass or total star formation rate, Verhamme et al., 2008).

We further note that at , the age of the Universe is of the order of the duration of the LBG phase ( Gyr — Shapley et al., 2001; Papovich et al., 2001; Lee et al., 2008).

Assuming an initial interval of time () in which the LBG is active as a LAE (i.e., shows conspicuous Ly emission, with rest–frame equivalent width greater than 100Å), whose duration should be of the order of Myr (e.g., Mori & Umemura, 2006; Verhamme et al., 2008), the probability to observe a LBG in the LAE phase should increase with redshift when observing galaxies in a Universe younger than Gyr (roughly, the fraction of emitters versus non-emitters is proportional to ). Future surveys of LBGs at redshift beyond seven should show this trend even more clearly (albeit subject to the observational selection effects discussed above).

8 Correlation with Morphological Properties

We have derived basic morphological parameters for the galaxies in our spectroscopic samples from the ACS –band image. With an effective wavelength Å (for a typical LBG UV spectrum), the  filter probes the rest–frame far–UV emission of –dropout galaxies at Å. In general, it is difficult to interpret the results of analyses of the UV morphologies of high–redshift galaxies in terms of the evolution of traditional Hubble types, in part because these are mostly known at optical rest–frame wavelengths (c.f., Giavalisco et al. 1996a,b), and also because it is not obvious what is the typical morphology of the present–day spectral types that are most similar to the LBGs.

We have measured parametric and non–parametric morphological indicators separately for the two sub–samples of “emitters” and “absorbers” in the primary sample. The basic morphological parameters have been drawn from the v2.0 ACS catalogs, direct outputs of the SExtractor algorithm during the segmentation process in the  band, and are summarized in Table 6 with their average values and 1- standard deviations. Tabulated quantities are the major semi-axis (), half–light radius (), isophotal (AREAF) and FWHM. As shown in Table 6, LBGs with Ly in emission have more compact morphologies relative to those with rest–frame UV features observed in absorption. In detail, the physical sizes at the half light radius for emitters and absorbers are on average 1.1 and 1.6 kpc, respectively.

To further investigate the correlation between morphology and Ly properties, we have computed the Gini coefficient for our sample. We have utilized the formulation described in Abraham et al. (2003, , their Eq. 3). The Gini coefficient provides a measure of the degree of central concentration of the source. Values range between 0 (uniform surface brightness) and 1 (highly nucleated). Lotz et al. (2006), Ravindranath et al. (2006) and, more recently, Lisker (2008), have analyzed the stability of the Gini coefficient, based on a comparison of HST/ACS imaging data from the GOODS and UDF surveys. They find the Gini coefficient depends strongly on the S/N ratio and at all S/N levels, the Gini coefficient shows a strong dependence on the choice of aperture within which it is measured. This complicates comparisons of Gini parameters derived in different studies. However, relative values from measurements done the same way within a given data set should be meaningful. In the present case we restrict the analysis for the brighter –dropout sample and assume that systematics are similar for both emitter– and absorber–class LBGs. The pixels of each source used in the calculation are those with flux above F 1 percentile of the median background. Adopting F=2 and the  band (F=2,  band), we find that the “em.” and “abs.” classes have = and =, respectively. With (F=3,  band) the values are = and =. The same calculation performed in the  band, produces the following median values: =, = (F=3,  band) and =, = (F=2,  band).

These calculations show that the two LBG spectroscopic classes have different average morphologies, with emitters intrinsically more nucleated than the absorbers. This distinction seems to increase with greater Ly equivalent width. The behavior is shown in Figure 19 (middle panel, F=2,  band), where the Gini coefficient is plotted versus the Ly equivalent width. Though this result is, on average, in qualitative agreement with the observations at 2 and 3 by Law et al. (2007), we note that cases of nucleated absorbers and “fuzzier” emitters are also present. Larger galaxy samples at these redshifts are needed in order to put this result on a firmer statistical footing.

As shown in Figure 19 (top panel), there also seems to be an inverse correlation between Ly emission equivalent width and galaxy size, namely galaxies with larger equivalent widths are smaller. To some extent, this correlation can be explained by the fact that galaxies with larger equivalent widths are more likely to be fainter; this is the case for the –dropout sample (Figure 19, bottom panel). The correlation, however, seems to persist even when subsamples cut by absolute luminosities are considered, as illustrated in the top panel of Figure 19 where the size of the symbols scale with apparent  magnitude. In this latter case, only sources with spectroscopically-detected detected and have been considered (at , Ly enters the  band). One potential physical explanation of this size behavior could lie in the masses of the objects. Ly–emitting LBGs at are found to have smaller stellar masses than objects lacking this emission (e.g. Gawiser et al., 2006). Pentericci et al. (2007) also found for the –band dropout sample in the present work an average stellar mass of M and M for the emitters and absorbers, respectively. This further suggests that emitters may be associated with less massive dark matter halos and hence have experienced a different star formation history compared to the absorption line galaxies. At higher redshift, this analysis is more critical because high quality near–infrared images (e.g., NICMOS) and deep spectroscopy are needed to identify the absorbers. Dow-Hygelund et al. (2007) found that Ly–emitting –band dropouts seem to be morphologically distinct from the general –band dropout LBGs. We only report here that, on average, the for our sample of –dropouts emitters is consistent with other observations (e.g. Stanway et al., 2004a, b; Dow-Hygelund et al., 2007), with (Figure 19, lower panel). The only –band dropout with QF=B without Ly in emission (GDS J033233.19-273949.1) has . High quality and deeper near-IR images and spectroscopy are necessary to investigate this issue.

9 Conclusions

In the present work, we have addressed the spectroscopic properties of LBGs at high redshift, selected from the GOODS survey. We have discussed the efficiency of the photometric selection criteria adopted. We have extracted preliminary information from the spectral features and UV luminosity and compared it with analogous studies at lower redshift. Summarizing,

1. 109 out of 202 targeted LBGs have been spectroscopically confirmed in the redshift range , according to –,– and –band dropout selections. This relatively low confirmation rate is largely due to the following two reasons: i) the target list includes a relatively large fraction of faint sources, with 65 out of 202 or 32.2% of the sample having ; and ii) the difficulty in determining redshifts for galaxies at given our instrumental set-up. Considering sources with determined redshifts, 96, 89 and 82 of the observed –,– and –band dropout samples have been confirmed in the expected redshift range, respectively. Twelve low–redshift interlopers have also been confirmed, 10 stars and two galaxies at . Five high–redshift galaxies have been serendipitously discovered, yielding a total of 114 redshifts measured beyond redshift 3.1 (38 of these with QF=C).

2. From the composite spectra of the three flavors of dropout (,  and –band dropouts), we detect the typical spectral features of star–forming galaxies, namely a flat spectrum redward of Ly, IGM attenuation and the Lyman limit blueward of Ly, UV absorption lines (both high and low ionization), and Ly seen in both emission or absorption. In particular, at , a comparison between the composite spectra of emitters and absorbers shows steeper spectral slopes and weaker UV absorption features for the emitters.

3. Galactic outflows have been identified at by measuring the velocity offset between interstellar, Ly and nebular lines. The measured  km s is consistent with results at by Shapley et al. (2003), considering the portion of their sample with similar Ly equivalent widths to our sample. We derive of  km s, similar to the  km s derived by Adelberger et al. (2003) at lower () redshift. A similar offset (but less accurate because it is derived from the composite spectrum) has been detected in the –dropout sample (redshift 5), i.e.,  km s. This supports the interpretation that outflows similar to those taking place at 2 and 3 are also observed in our samples of LBGs at 4 and 5.

4. The presence of a weaker Ly equivalent widths for dropouts with brighter UV luminosities () is clear in the current spectroscopic sample (considering all categories). This trend has been recently noted by several authors, and may be naturally explained by a different evolution of bright UV LBGs with respect to the fainter ones. The brighter galaxies should be dustier and more evolved (and probably more massive) than the fainter ones, which show a larger spread of Ly equivalent widths possibly due to assorted SF histories.

5. The sample at exhibits correlations between certain basic UV rest–frame morphological properties and spectroscopic properties such as the presence and strength of Ly emission. In particular, emitters appear more compact and nucleated than absorbers. Law et al. (2007) find a similar “nucleation effect” at 2 and 3 in their BM/BX and LBG samples, and interpret this as a consequence of more dust in the absorbers leading to redder colors and more diffuse morphologies. Pentericci et al. (2008) analyze the photometric properties of the same sample as discussed here, and find that emitters are less massive and less dusty than absorbers. Focusing on the emitters, increasing Ly equivalent widths correspond to decreasing stellar masses and extinction. The emitters, especially those with a large Ly equivalent widths, could be systems forming a relatively large fraction of their stellar mass during an intense burst of star formation. These putative proto–spheroids observed at could include in significant numbers the progenitors of the compact massive early–type galaxies identified at (e.g., Cimatti et al. 2008; van Dokkum et al. 2008; Buitrago et al. 2008). Such an evolutionary link is generally consistent with the observed spatial clustering properties and the stellar populations of LBGs at and (Giavalisco et al., 1998; Giavalisco & Dickinson, 2001; Lee et al., 2006, 2008; Ouchi et al., 2005) and those of the BzK and DRG galaxies at (Kong et al., 2006; Quadri et al., 2008). The strength of spatial clustering increases with the mass of the galaxies and with redshift, as a consequence of gravitational evolution of structure. The observed larger spatial correlation length and larger stellar mass of the UV/optical–selected galaxies at (Daddi et al., 2007; van Dokkum et al., 2006) are in overall quantitative agreement with the expected dependence of clustering with both mass and time, when compared to the less strongly clustered and less massive (in stellar content) UV–only selected galaxies at and (Adelberger et al., 2005). This suggests that the same populations of dark matter halos is being observed at different evolutionary stages of the growth of their galaxy hosts and spatial clustering.

We are grateful to the ESO staff in Paranal and Garching who greatly helped in the development of this programme. We acknowledge financial contribution from contract ASI/COFIN I/016/07/0 and PRIN INAF 2007 “A Deep VLT and LBT view of the Early Universe”. EV thanks STScI and NOAO for hospitality during a visit in which this paper was conceived and partially written. EV thanks F. Calura for useful discussions about the dust properties of high redshift galaxies. The work of DS was carried out at Jet Propulsion Laboratory, California Institute of Technology, under a contract with NASA.

Footnotes

1. affiliation: Based on observations made at the European Southern Observatory Very Large Telescope, Paranal, Chile (ESO programme 170.A-0788 The Great Observatories Origins Deep Survey: ESO Public Observations of the SST Legacy / HST Treasury / Chandra Deep Field South). Also based on observations obtained with the NASA/ESA Hubble Space Telescope obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc. (AURA) under NASA contract NAS 5-26555.
2. affiliationmark:
3. affiliationmark:
4. affiliationmark:
5. affiliationmark:
6. affiliationmark:
7. affiliationmark:
8. affiliationmark:
9. affiliationmark:
10. affiliationmark:
11. affiliationmark:
12. affiliationmark:
13. affiliationmark:
14. affiliationmark:
15. affiliationmark:
16. Note that a simple analysis of the observed colors or sizes is not sufficient to establish the presence of evolutionary effects since it is not possible to separately measure the distribution functions of color, size and luminosity. See discussion in Reddy et al. (2008).
17. The v2.0 exposure times in the  ,  ,  and  bands are 7200, 5450, 7028 and 18232 seconds, respectively. Details of the ACS observations, as well as major features of the GOODS project, can be found in Giavalisco et al. (2004a); additional information about the latest v2.0 release of the GOODS ACS images and source catalogs can be found at www.stsci.edu/science/goods/v2.0 and will be described in detail in an upcoming paper (Giavalisco et al., in prep.).
18. This galaxy is well detected in the ,  and  bands (with signal to noise ratios of 13.9, 14.1 and 26.1, respectively). This fact excludes the Ly possibility at redshift 6.8. Other possible emission lines are [O iii]5007 and/or H at z 0.9, however in this case the [O ii] 3727 should be detected at 7049Å, region free from skylines. If we assume H line, we should expect to observe at least H at 7014Å. Therefore the most probable interpretation is [O ii] 3727 at z=1.541.
19. This galaxy is interesting in its own right. It has colors (