THE OPTICALLY UNBIASED GAMMA-RAY BURST HOST (TOUGH) SURVEY. I. SURVEY DESIGN AND CATALOGS^*

Following the discovery of long-duration gamma-ray burst (GRB) optical afterglows (e.g., van Paradijs et al. 2000), evidence rapidly accumulated pointing to their origin in the deaths of massive stars (Galama et al. 1998; Bloom et al. 2002; Stanek et al. 2003; Hjorth et al. 2003). This soon led to their being touted as a potentially powerful means of discovering and reckoning star formation activity, which could bypass many of the biases that hamper other star formation measures (e.g., Wijers et al. 1998). The detection of GRBs is at once both independent of the luminosity of their host galaxy and, at high energies, unaffected by dust obscuration. They are also extremely luminous at all wavelengths, so that they can be used to perform transmission spectroscopy of their host galaxies, even where such hosts are too faint to be detected photometrically (e.g., Thöne et al. 2010). This also permits them to be found across the universe from very low to very high redshift: Indeed, GRB 090423 holds the record at z = 8.2 for the most distant spectroscopically confirmed object known (Tanvir et al. 2009; Salvaterra et al. 2009).

The history of baryonic structure in the universe to a large extent revolves around the co-evolution of stars, galaxies, metals, and dust. In recent years, a range of methods has been applied to the identification and study of different populations of high-redshift galaxies, but all are variously affected, to a greater or lesser extent, by the problems of dust extinction, flux-limited samples, source confusion, and incomplete redshift determinations. Notable examples are Lyman break galaxies (LBGs; Steidel et al. 2003), sub-mm galaxies (SMGs; e.g., Barger et al. 2000; Blain et al. 2002), Lyα emitters (LAEs; e.g., Fynbo et al. 2003b, 2005; Sawicki & Thompson 2006), distant red galaxies (DRGs; Daddi et al. 2004), damped Lyα absorbers (DLAs; Møller et al. 2002; Wolfe et al. 2005), and high-redshift dropout galaxies (e.g., Bouwens et al. 2010). Studies of these populations have led to major advances in our understanding of galaxy evolution, but our view remains far from complete, particularly at higher redshifts.

The major stumbling blocks to the routine use of GRBs as tools for the investigation of cosmic history have been the relatively slow buildup of statistically useful samples and the disconnect between the high-energy properties of the burst and parameters useful for measuring the host and its star formation rate (SFR). These latter parameters, such as the redshift, the host-galaxy bolometric luminosity, metallicity, and dust content can only be determined from longer wavelength observations in the UV, optical, infrared, millimeter, and radio regimes.

Accurate X-ray localizations, as an intermediate step between the gamma-rays and the optical/infrared, have turned out to be critical to finding the afterglows and the host galaxies of GRBs. The first detections provided by BeppoSAX (Costa et al. 1997) heralded the afterglow era. Since 2005, Swift (Gehrels et al. 2004) with its autonomous slewing capability and onboard X-Ray Telescope (XRT; Burrows et al. 2005) has been extremely successful in this regard, delivering high-accuracy (few arcseconds) positions for approximately a hundred bursts every year. It has become the GRB facility par excellence thanks to its innovative, intelligent design, open access policy, energetic team, and active community support.

In spite of Swift's success, some of the problems that beset most other star formation indicators also affect GRBs. Despite many attempts to infer distances from the prompt high-energy properties (e.g., Guidorzi 2005), UV/optical/NIR observations (and preferably spectroscopy) remain the only route to reliably find the redshift of a GRB or its host, and we must observe the hosts of GRBs directly across a range of wavelengths to connect SFR to GRB rate. Thus, to avoid optical biases in GRB selection requires a sample as complete as possible in redshift and host-galaxy properties. Swift's excellent X-ray localization completeness and large sample size provide a first opportunity for this to be attempted.

Several previous studies of GRB hosts have found that the majority of GRB hosts are faint, blue galaxies (Fruchter et al. 1999; Le Floc'h et al. 2003; Christensen et al. 2004) with low stellar masses (Castro Cerón et al. 2006, 2010; Savaglio et al. 2009). The GRBs generally occur within the UV-bright parts of their hosts (Bloom et al. 2002; Fruchter et al. 2006), consistent with their association with star formation. They are also typically fainter and have lower metallicities than the galaxies selected in Lyman-break surveys in the same redshift range (Jakobsson et al. 2005; Savaglio 2006; Fynbo et al. 2008b). Accounting for systematic incompleteness biases due to dust obscuration, i.e., the so-called dark burst problem (Fynbo et al. 2001, 2009; De Pasquale et al. 2003; Jakobsson et al. 2004; Perley et al. 2009; Krühler et al. 2011; Svensson et al. 2012; Rossi et al. 2012; Melandri et al. 2012), has also been attempted with some success (e.g., Jakobsson et al. 2006b; Fynbo et al. 2009).

Submillimeter (submm) and radio surveys of GRB hosts have shown that only a few could be ultraluminous infrared galaxies (Berger et al. 2003), indicating that obscured star formation is not dominant in GRB-selected galaxies (Tanvir et al. 2004). Moreover, the candidate submm bright GRB hosts are bluer, hotter, and at lower redshift than typical of SMGs (e.g., Berger et al. 2003; Michałowski et al. 2008).

Most of the studies mentioned above have been limited to fairly low-redshift samples (z ≲ 1.5), which calls for an extension of such statistical investigations to higher redshifts. Recently, more complete surveys (Greiner et al. 2011) have indicated the existence of a fraction of GRB host galaxies being more chemically evolved, more dusty, and of higher mass (Krühler et al. 2011). This highlights the importance of minimizing optical selection effects in using GRBs as tracers of high-redshift star formation and the cosmic star formation history (e.g., Tanvir et al. 2012; Salvaterra et al. 2012; Robertson & Ellis 2012; Elliott et al. 2012; Basa et al. 2012; Trenti et al. 2012).

We have conducted a survey of galaxies hosting GRBs, suitable for statistical studies and for observations with other existing or future facilities. In particular, we have imaged 69 GRB host galaxies to deep limits in R and K_s and obtained follow-up spectroscopy to determine redshifts and study Lyα emission in the higher-redshift systems. The improved redshift completeness is also important for statistical studies of GRB properties themselves. The details of the survey design and target selection are described in Section 2. The observations are described in Section 3, along with the catalog of GRB and host galaxy properties, including magnitudes and redshifts. Section 4 provides a summary of some of the immediate scientific results and Section 5 concludes. More detailed discussion and results are presented in four companion papers by D. Malesani et al. (2012, in preparation), Jakobsson et al. (2012), Milvang-Jensen et al. (2012), and Krühler et al. (2012). Supplementary radio, Hubble Space Telescope (HST), and X-ray observations are reported in Michałowski et al. (2012), S. Schulze et al. (2012, in preparation), and D. Watson et al. (2012, in preparation).

All magnitudes given in this paper are in the Vega system. Throughout our work, we adopt a cosmology with Ω_m = 0.27, Ω_Λ = 0.73, H₀ = 71 km s⁻¹ Mpc⁻¹.

A statistically meaningful and optically unbiased sample is essential in addressing issues such as the GRB redshift distribution, the importance of dust obscuration, and the determination of the burst and host luminosity functions (see also Salvaterra et al. 2012). We have applied the following selection criteria to provide a large and homogeneous sample of targets and so as to ensure that selection effects are minimized and well understood. The criteria are akin to those introduced by Jakobsson et al. (2006b) and refined by Fynbo et al. (2009).

• 1.  
  Our starting point is the sample of Swift-detected GRBs. Swift is the only current facility which can provide a large, homogeneous, well-localized sample of GRBs. We pick only events triggered onboard by the Burst Alert Telescope (BAT) and disregard those (very few) discovered during ground analysis.
• 2.  
  We limit ourselves to long-duration GRBs, and discard all events with T₉₀ < 2 s (Kouveliotou et al. 1993), based on the catalog of Sakamoto et al. (2011).
• 3.  
  The bursts must have a detected X-ray afterglow, the location of which is made available less than 12 hr after the trigger. The requirement of the existence of an X-ray afterglow effectively makes our sample X-ray selected.

The above criteria are essentially constraints on the nature of the GRBs. The requirement on the timing in Criterion 3 additionally ensures that efficient afterglow searches could be performed. It also works as an effective Sun constraint, since those bursts with lines of sight close to the Sun upon discovery cannot be observed quickly. It should be noted that the fraction of long GRBs without a detected X-ray afterglow is very low, providing the XRT is repointed rapidly (see below).

Looking for the hosts of all these GRBs would be very demanding. We therefore apply further cuts, with the purpose of reducing the sample size in an unbiased way. Imposing restrictions 4–6 does not bias the sample toward optically bright afterglows; instead each GRB in the sample has favorable observing conditions, i.e., useful follow-up observations are likely to be secured.

• 4.  
  The Galactic foreground optical extinction as determined by Schlegel et al. (1998) must be A_V ⩽ 0.5 mag. This makes the detection level almost uniform (i.e., no bias against systems that happen to be faint just because of Galactic extinction). As a bonus, all fields have high Galactic latitude (|b| ≳ 15°) ensuring that crowding by foreground stars is not an issue (Figure 1).
• 5.  
  As another visibility constraint, we require that the distance on the sky between the GRB and the Sun at the time of explosion is more than 55° (similar to Jakobsson et al. 2006b). This is relevant since bursts closer to the Sun than this will offer very limited opportunity for ground follow-up. In practice, it is not a very demanding constraint because Criterion 3 effectively enforces a solar angle constraint of this order to satisfy the Swift Sun-avoidance condition for rapid slews for the narrow-field instruments (XRT and the Ultraviolet-Optical Telescope, UVOT) of 47°. In fact, during the survey period, only GRB 070810A had an X-ray afterglow announced within 12 hr but had a discovery line of sight closer to the Sun.
• 6.  
  We require that there is no nearby star or bright contaminating object that would render host-galaxy identification and spectroscopy difficult. To make this quantitative, we note that, under the typical observing conditions of our R-band imaging survey, bright stars affect a region of radius r ≈ 1.8 + 0.4 × exp [(20 − R)/2.05] arcsec, where R is the USNO-B magnitude (Monet et al. 2003). This relation is valid in the range 11 ≲ R ≲ 19. We thus require that no star with R < 19 is located on the CCD detected within this distance from the X-ray error circle edge. In two cases, GRBs 060604 and 070721B, very bright stars (R < 10) at distances ≳ 1' would nominally violate the above limit, but their effect was minimized by offsetting the pointing such that they were located outside of the detector.

Zoom In Zoom Out Reset image size

The aforementioned six criteria are applicable in principle to any survey of Swift GRB positions. Our survey was carried out with the VLT, and was restricted to bursts detected during a specific 2.5 yr period, so that for this sample we applied two additional criteria.

• 7.  
  Bursts triggered between 2005 March 1 and 2007 August 10. During this period Swift was fully operational and automatic slews were routinely enabled. There were 236 GRBs (229 triggered) in this time.⁹
• 8.  
  In order to be well placed for observation with the VLT we limit the declination range to be −70° < δ < +27° (J2000.0).

These criteria can easily be extended by using other observatories and different points in time.

After enforcing these criteria, we end up with a sample of 69 GRBs. The advantage of our selection is illustrated by the large success of afterglow follow-up for bursts in this subset. Based on Jochen Greiner's Web site,¹⁰ optical/NIR afterglows have been discovered for 75% (52/69) of the bursts in our sample, compared to 66% for the overall Swift sample. More importantly, spectroscopic redshifts are available for 55% of our targets (38 GRBs) compared to only 37% in the overall Swift sample.

Our final criterion concerns the localization accuracy of the GRBs.

• 9.  
  The X-ray localization uncertainty must be better than or equal to 2 arcsec (90% error radius). This constraint is dictated by the necessity to perform meaningful host searches. Larger error boxes would lead to too many ambiguous detections. For bursts without an optical/NIR localization, we adopt the best constrained position among the catalogs provided by Butler (2007b) and Evans et al. (2009). In principle, a more accurate localization could be available from optical or radio data, but we only base our selection on the X-ray position to retain a dependence solely on high-energy properties.

We summarize the selection criteria in Table 1. Our final sample consists of 69 GRBs and is presented in Table 2. Figure 1 shows the distribution in the sky of all GRBs detected by Swift during the chosen time period with the TOUGH sample explicitly marked.

Below we discuss the effects of the various selections. Figure 2 shows the 1 s BAT 15–150 keV peak photon flux (from Sakamoto et al. 2011) versus the date. As can be seen, our targets uniformly sample the full BAT distribution and there is no obvious long-term change in the detection threshold as a function of time, which seems to be consistently at a level of about 0.4γ cm⁻² s⁻¹ (the faintest target GRB 060218 is detected at 0.25γ cm⁻² s⁻¹). This shows that we are starting from a stable, well-controlled set of events (Criterion 1), although we note that the effective trigger threshold does vary somewhat with many factors, such as the current background level, the position of the burst within the BAT field, and also the form of the particular GRB light curve (for example, the detection of some bursts is only made after integrating up more than the 1 s binning we consider here). Out of the 69 TOUGH bursts, 56 (13) were rate (image) triggers.

Zoom In Zoom Out Reset image size

Our Criterion 3 is the detection of an X-ray afterglow. During the considered period, only 25 triggered GRBs were not detected (or observed) by XRT (11% of the overall Swift sample). Restricting to long-duration bursts (Criterion 2) with prompt slew (less than 3600 s), however, we find that only 5 events out of 180 (2.8%) were not detected by XRT. This shows that, in practice, XRT selection has a negligible impact on the size of our sample, although it does mean that rare GRBs with very faint X-ray afterglows will be systematically underrepresented.

Figures 3 and 4 illustrate the distributions of the times to first XRT observation (taken from the Swift GRB table¹¹; akin to but not identical to Criterion 3) and the Galactic V-band extinction (Schlegel et al. 1998; Criterion 4) for the TOUGH sample as well as the full distribution of GRBs.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Criterion 9 introduces a potential bias in our requirement of an accurate XRT position, which might lead us to favor optically bright afterglows. The accuracy of the X-ray position is influenced by many factors, several of which are not directly related to the GRB properties. For example, the astrometrically corrected positions (Butler 2007b) require the detection of serendipitous X-ray sources in the field other than the GRB. This depends more on the total exposure time than on the afterglow brightness. Figure 5 shows that the accuracy of X-ray positions has been stable during the considered time span. In fact, many of the UVOT-enhanced positions are dominated by the systematic error term which is independent of the afterglow brightness. There is a possible dearth of very accurate positions after 2007 March, which may be due to shallower exposures following a change of the Swift pointing strategy. In Figure 6, we investigate the impact of rejecting bursts without an accurate localization. We pick all GRBs which would pass selection criteria 1–8, regardless of the X-ray accuracy. We then plot their 90% error radii (Butler 2007a; Evans et al. 2009) as a function of the count rate at 12 hr, obtained by interpolating or extrapolating the light curves posted online¹² (Evans et al. 2009). Figure 6 shows that, indeed, faint afterglows have on the average larger error radii, but the effect is very small (a factor of ≈3 across more than two orders of magnitude in the count rate). Thus, among the 71 GRBs considered here, only 2 are discarded (less than 3%), showing that the impact on our sample is limited.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Our observational program consists of deep R and K_s imaging and spectroscopy obtained with the FORS1, FORS2, ISAAC, and X-shooter instruments at the ESO VLT. All VLT observations were conducted at least 50 days after the GRB explosion in order to avoid possible contamination from the afterglow or an associated supernova. The possible afterglow contamination in our images is quantified in D. Malesani et al. (2012, in preparation). Here, we briefly describe the overall results, as summarized in Table 3. Detailed survey results are reported in four accompanying papers. Results on individual targets or preliminary results are reported in Ruiz-Velasco et al. (2007), Jakobsson et al. (2007, 2008, 2009, 2011a, 2011b, 2011c), Tanvir et al. (2008), Thöne et al. (2008, 2010), Jaunsen et al. (2008), Xu et al. (2009), Elíasdóttir et al. (2009), Ferrero et al. (2009), Fynbo et al. (2009, 2008a), Malesani et al. (2009, 2010), Svensson et al. (2012), and Schulze et al. (2012). In Tables 2 and 3, we also list 10 extra hosts that were observed for our program, but that do not obey all of the TOUGH sample criteria.

3.1. Optical Imaging

For the purposes of identification of hosts and brightness estimation we have obtained deep images of the targets in the R band (typically rest-frame UV) to magnitude limits in the range R = 26–28 mag. The seeing FWHM of the R-band images ranges from 050 to 126 with a median of 071. D. Malesani et al. (2012, in preparation) discuss our procedure for identifying a host galaxy, including a quantification of the chance probability of wrongly identifying an unrelated galaxy in the error circle as the host. We note that there is always a small risk of wrong host identification in both imaging and spectroscopy. While this could lead to erroneous conclusions when discussing individual systems, it will have little effect on statistical results inferred from a sufficiently large sample, like the one presented here, as long as the probability for false identification is low.

We have identified host galaxies for 55 systems, corresponding to 80% of the sample. The coordinates of the identified host galaxies are reported in Table 3. In Figure 7, we show a mosaic of GRB host galaxies that are only localized by the XRT (a mosaic of all identified host galaxies is presented in D. Malesani et al. 2012, in preparation). These 17 galaxies (25% of our sample) are typically missed from other surveys. The observed magnitudes are shown in Figure 8. It is evident that host galaxies are generally sub-luminous, in the sense of being fainter than L* at z = 0. We note that the simple detection of a host galaxy in the R band can be used to set an upper limit to the redshift of the GRB of z ≲ 5.6 (Ruiz-Velasco et al. 2007; Jakobsson et al. 2012).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.2. Near-infrared Imaging

3.3. Spectroscopy

We have verified that there are no obvious trends in the redshifts or magnitudes of the TOUGH host galaxies as a function of GRB epoch.

Below we summarize some of the results of the TOUGH survey, with special emphasis on the implications of having compiled an X-ray-selected sample of GRB host galaxies, and obtaining photometry and spectroscopy of a sample of GRB host galaxies extending to faint and high-redshift systems.

4.1. Redshift Distribution

In Figure 9, we show the redshift distribution prior to our survey and compare to the redshift distribution resulting from our survey. All redshifts are given in Table 3. The new redshift distribution is more smooth with a conspicuous peak at around redshift 2 with a median redshift of 2.14 ± 0.18. The median redshift of the entire TOUGH sample (i.e., including systems with no redshift constraint) is robustly determined to be between 1.5 and 2.5 (Table 4). As discussed by Jakobsson et al. (2012) our redshifts and redshift constraints allow us to distinguish models for the GRB rate and luminosity function and formation biases.

Zoom In Zoom Out Reset image size

Table 4. Median Properties of TOUGH and TOUGH Subsamples

Sample	R	Ks	〈z〉low	Median z	〈z〉high
(1)	(2)	(3)	(4)	(5)	(6)
TOUGH	25.52 ± 0.23	>22.53	1.50	...	2.53
FORS	23.59 ± 0.51	20.78 ± 1.30	...	1.25 ± 0.29	...
X-shooter	24.48 ± 0.15	21.13 ± 0.27	...	2.24 ± 0.15	...
OA	25.80 ± 0.31	>22.53	1.95	...	2.32
XRT	24.36 ± 0.40	21.13 ± 0.29	...	...	...
z	25.11 ± 0.31	>22.53	...	2.14 ± 0.18	...
No z	25.88 ± 0.24	>21.77	...	...	...
R	24.73 ± 0.21	22.15 ± 0.69	1.29	...	2.39
No R	...	...	3.07	...	3.97
Ks	24.27 ± 0.28	20.74 ± 0.16	1.10	...	1.94
No Ks	26.61 ± 0.43	...	2.14	...	3.35

Notes. Column 1: TOUGH sample or subsample; Column 2: median R magnitude of subsample; Column 3: median K_s magnitude of subsample; Column 4: minimum median redshift of subsample, assuming all unknown redshifts are less than the minimum redshift; Column 5: median redshift; Column 6: maximum median redshift of subsample, assuming all unknown redshifts are larger than the maximum redshift. Errors in the median are estimated as 1.4826 × MAD/$\sqrt{N-1}$, where MAD is the median absolute deviation, N is the number of systems used to determine the median, and 1.4826 is a scale factor that applies to Gaussian distributions. The subsamples are the following: FORS: the six well-defined FORS emission-line redshifts, i.e., excluding the dropout redshifts (GRBs 050822, 051006, 051117B, 060908, 061021, and 070306); X-shooter: the nine new X-shooter redshifts; OA: systems with an optical/NIR afterglow; XRT: systems with no optical/NIR afterglow; z: systems with a known redshift; No z: systems without a known redshift; R: systems with a measured R magnitude; No R: systems without a measured R magnitude (including the high-redshift systems GRB 050904 and GRB 060927 which were not observed); K_s: systems with a measured K_s magnitude; No K_s: systems without a measured K_s magnitude (including the high-redshift GRB 050904 but excluding GRB 050819 and GRB 051006 which were not observed).

Download table as:  ASCIITypeset image

To illustrate the importance of sample selection criteria for the determination of the median redshift for GRBs we plot the redshift completeness as a function of BAT (15–150 keV) peak photon flux in Figure 10. The median 1 s peak flux for the TOUGH sample is 1.4γ cm⁻² s⁻¹ (Figure 10). Subsamples with or without redshift give a consistent value for the median 1 s peak flux, as do subsamples with or without an optical/NIR afterglow. About 75% of the GRBs in our sample have lower peak photon flux than the cutoff adopted by Salvaterra et al. (2012) (⩾2.6γ cm⁻² s⁻¹). Our redshift completeness is in the range 70%–80% below that value and reaches almost 90% slightly above. The median redshift increases almost monotonically with lower peak flux, from 1.38 at a cutoff at 2.6γ cm⁻² s⁻¹ to 2.12 for a cutoff at 0.4γ cm⁻² s⁻¹. Tying to the BATSE luminosity function, Salvaterra et al. (2012) modeled their results and predicted a median redshift of z = 2.05 ± 0.15 to that limit, consistent with our value.

Zoom In Zoom Out Reset image size

However, obtaining redshifts of fainter galaxies may shift the median redshift somewhat higher. As indicated by Krühler et al. (2012) the faint host galaxies targeted with X-shooter have been found at significantly higher redshifts compared to brighter targets. We therefore suspect that the remaining very faint host galaxies may lead to even higher average redshifts. Figure 11 shows the median redshift as a function of the median R-band magnitude of the hosts (the corresponding values are given in Table 4). The median redshift of the TOUGH subsamples is lowest for the bright hosts for which the redshift is determined from emission lines using FORS, higher for fainter hosts where the redshift is determined using X-shooter, and higher still for bursts where the host remains undetected. K_s-detected host galaxies are at significantly lower redshifts than galaxies which are not detected in K_s. The overall trend seems to be consistent with the expectation that fainter GRB host galaxies are on average at higher redshifts.

Zoom In Zoom Out Reset image size

TOUGH for the first time picks up high-redshift GRB hosts in significant numbers. The compilations of Chen et al. (2009) and Savaglio et al. (2009) count eight optically detected GRB host galaxies at z > 2 of which two are Swift GRBs. Table 3 lists a total of 18 TOUGH-detected GRB host galaxies at z > 2 of which 17 are new, as well as one non-TOUGH z > 2 GRB host galaxy. Such a large sample opens the field for more detailed studies of individual systems and of high-redshift GRB hosts as a whole (see also Section 4.3).

4.2. Galaxy Colors and Magnitudes

Of special interest is the study of dark bursts which are believed to be caused mostly by dust obscuration (e.g., Djorgovski et al. 2001; Rol et al. 2007; Jaunsen et al. 2008; Tanvir et al. 2008; Perley et al. 2009; Krühler et al. 2011; Svensson et al. 2012; Rossi et al. 2012). Hence, it is expected that the properties of the host galaxies of XRT-only GRBs may be different from those hosting GRBs with detected optical/NIR afterglows.

Figure 12 highlights the difference between systems with and without an optical/NIR afterglow. There is a significant dearth of XRT-only galaxies at very faint magnitudes in the R band. This is also reflected in the median R-band magnitude of 25.8 ± 0.3 for galaxies hosting an optically/NIR-detected afterglow, as opposed to 24.4 ± 0.4 for those with undetected ones (Table 4). XRT-only localized host galaxies are also relatively brighter in K_s (median magnitude of K_s = 21.1 ± 0.3 opposed to K_s > 22.5 for galaxies hosting GRBs with optical/NIR afterglows). A clue to the origin of these remarkable differences comes from the lower panel of Figure 12 which demonstrates that XRT-only hosts have equivalent X-ray absorption column densities at z = 0 extending to much higher values (see also Krühler et al. 2012). These differences clearly demonstrate that statistical studies based on optically selected GRBs will be strongly biased.

Zoom In Zoom Out Reset image size

We note that our procedure of identifying the host as the brightest galaxy inside the XRT error circle naturally will lead to a bias toward brighter galaxies if any of the identifications are incorrect. To quantify this bias we tested our host identification procedure in the XRT error circles of GRBs with an optical/NIR afterglow. We found that in 6 out of 50 cases (12%) we would misidentify the host (this would indicate that about 1–2 of the galaxies in Figure 7 may have been misidentified). This procedure led to a biased median R-band magnitude of 25.5 ± 0.2 for the XRT error circles of GRB with optical/NIR afterglows but XRT-only identified host galaxies. Hence, the difference between host galaxies with and without optical/NIR afterglows remains significant.

Regarding systems with both R and K_s detections, we find that the median magnitude of hosts of GRBs with or without optical/NIR afterglows is consistently R ≈ 24.3 ± 0.3. However, the median colors of XRT-only GRB host galaxies detected in the K_s band are significantly redder, R − K_s = 3.7 ± 0.3, than those for which an optical/NIR afterglow was detected, R − K_s = 2.8 ± 0.1. While this difference in color could partly be a redshift effect, their larger X-ray absorption column densities suggest that GRBs without an optical/NIR afterglow suffer stronger attenuation due to dust and that dust extinction causes optical/NIR GRB afterglows to drop out of detection. This is in line with other studies which find that host galaxies of dark GRBs are redder and brighter (more massive) than those of GRBs with optical afterglows (Krühler et al. 2011; Svensson et al. 2012). A more detailed discussion of the R − K_s distribution as a function of redshift and a comparison to other high-redshift galaxies are presented in D. Malesani et al. (2012, in preparation).

4.3. Star Formation in GRB Host Galaxies

Long-duration GRBs provide a complementary view of the nature of "typical" star-forming galaxies over most of cosmic history. However, to date, most samples of GRB hosts have been heterogeneous and had a strong bias toward GRBs with detected optical afterglows. In this paper we have described a large, homogeneous sample of hosts targeted with the VLT, using X-ray positions where no optical/NIR afterglow was found. We report the main catalog in Table 3. We detect the majority of the hosts, namely 80%, in the R band down to a detection limit of ∼27.5, including 17 systems at z > 2. Within the TOUGH sample, we detect 54 hosts of which 32 were not previously detected. GRB host galaxies cover a wide range of luminosities from <0.1L* to L*. From the K_s-band imaging we confirm the result of more biased studies that GRB host galaxies tend to be blue. We also find that galaxies with no optical/NIR afterglow are significantly brighter and redder than those with an optical/NIR afterglow, highlighting the importance of TOUGH in being X-ray selected. With our newly reported redshifts, the redshift distribution consisting of 77% of the GRBs now appears smooth with a median redshift above 2.

Our sample will form the basis for statistical studies of GRB host galaxies, such as their luminosity function, colors and SFRs, redshift distribution, Lyα properties, radio emission, and X-ray absorption properties.

Future observational work will focus on completing the sample, in particular in redshift, and to complement it at other wavelengths, in particular, radio, submm (ALMA, Herschel), and NIR/MIR (Spitzer, HST, JWST). It is also useful for selecting subsamples for further study, such as SED studies or emission-line diagnostics, e.g., metallicity determination. Finally, it is straightforward to enlarge in a well-defined way, in time or declination, as we are doing with VLT/X-shooter and Gemini.

We thank the members of the former GRACE collaboration as well as participants of the former EU FP5 Research Training Network "Gamma-Ray Bursts: An Enigma and a Tool," for their initial support of and continued interest in this project. We are particularly grateful to Paul M. Vreeswijk for his enthusiastic contributions to the project. We are grateful to Scott Barthelmy, Nat Butler, Phil Evans, Samantha Oates, and Patricia Schady for their assistance in defining our sample selection criteria. We acknowledge the use of Jochen Greiner's GRB Web site. This work made use of data supplied by the UK Swift Science Data Centre at the University of Leicester. D.M. acknowledges support from the Instrument Center for Danish Astrophysics. P.J. and S.S. acknowledge support by a Project Grant from the Icelandic Research Fund. J.P.U.F. and B.M.-J. acknowledge support from the ERC-StG grant EGGS-278202. The research activities of J.G. are supported by the Spanish research programs AYA2008-03467/ESP and AYA2009-14000-C03-01. M.J.M. acknowledges the support of the Science and Technology Facilities Council. The Dark Cosmology Centre is funded by the Danish National Research Foundation.

Facilities: VLT:Antu (ISAAC,FORS2) - Very Large Telescope (Antu), VLT:Kueyen (FORS1,XSHOOTER) - Very Large Telescope (Kueyen), Swift (BAT,XRT) - Swift Gamma-Ray Burst Mission