3 Results

The GRB host galaxies span a broad range of redshifts (see Table .1 of Appendix A), but the current sample of these GRB-selected sources is actually too small to study the evolution of their characteristics with different lookback times. On the other hand, it can be particularly interesting to consider these objects as a whole sample of high-z sources, and compare their properties with other field galaxies selected by different observing techniques. In this section, we compare the observed and absolute magnitudes of the GRB host galaxies at different wavelengths (e.g., B-, R-, K-band) as well as their R-K colours, with those of high redshift sources detected in the optical, NIR, mid-infrared and submillimeter deep surveys.

   
3.1 "K-z'' and "R-z'' diagrams

Figure 2: Observed K magnitudes of the GRB host galaxies versus redshift ( filled diamonds) derived from Table 2. The GRB 001011 and GRB 981226 hosts are indicated assuming an arbitrary redshift z=1. Photometric uncertainties are reported with vertical solid lines. The K magnitudes of NIR-selected field sources with spectroscopic ( crosses) and photometric ( dots) redshifts are given for comparison (see Sect. 3.1 for references). We also indicate the K-band photometry of the NIR counterparts to high redshift ISO galaxies ( open squares) and SCUBA sources with determined spectroscopic redshifts ( filled squares).

Figure 3: Observed R magnitudes of the GRB host galaxies versus redshift ( filled diamonds) derived from Table .1. The uncertainties are indicated with vertical solid lines. The largest ones reflect, for a given object, the scatter of the various magnitudes given in the literature (see Appendix A). The R magnitudes of optically-selected field sources with spectroscopic ( crosses) and photometric ( dots) redshifts are given for comparison (see Sect. 3.1 for references).

The Hubble "K-z'' diagram of the GRB host galaxies is illustrated in Fig. 2. The spectroscopic redshifts of the GRB 981226 and GRB 001011 hosts have not been so far determined. Based on their K magnitude and R-K colour (see Sect. 3.2), we estimate that these objects could be located in the $0.7\la z \la1.4$ redshift range. To allow comparisons with other sources in the field, we overplotted the Kmagnitudes of galaxies reported from various surveys. Nearby sources were taken from the Hawaii K-band galaxy survey ( $\bar{z}=0.35$, Cowie et al. 1994; Songaila et al. 1994), while galaxies at intermediate redshift ( $\bar{z}=0.8$) were derived from the Hawaii Deep Surveys by Cowie et al. (1996). Those at higher z( $\bar{z}=1.5$) were taken from the catalog of photometric redshifts in the Hubble Deep Field (HDF, Fernández-Soto et al. 1999). We also indicated the K magnitudes of the ISO sources observed in the CFRS and HDF with flanking fields as given by Flores et al. (1999), Hogg et al. (2000) and Cohen et al. (2000), as well as those of the NIR counterparts to the SCUBA sources with confirmed spectroscopic redshifts, obtained by Smail et al. (2002a) and Dey et al. (1999). The K-band luminosities of these SCUBA galaxies were de-magnified from gravitational amplification for the lensed cases.

The comparison suggests that in the NIR, the GRB host galaxies do not particularly distinguish themselves from the field sources selected in optical/NIR deep surveys. No particular bias of detection toward the luminous sources is in fact apparent. There is however a significant contrast between their K magnitudes and those of the ISO and SCUBA sources which, like the GRB hosts, are birthplaces of massive star formation. These differences of magnitudes and the implications on their absolute luminosities will be more firmly established in Sect. 3.3 and discussed in Sect. 4.

To further address the nature of galaxies selected by GRBs relative to other field sources at high redshift, we also present in Fig. 3  the Hubble "R-z'' diagram for the sample of GRB hosts detected at optical wavelengths. Their R magnitudes are given in Table .1. They were obtained from various papers of the literature and homogenized following the method described in Appendix A. This sample is significantly larger than the one selected in the K-band. In addition to the hosts which have not been imaged in the NIR yet, there is indeed a number of GRB host galaxies which were both observed at optical and NIR wavelengths, but only detected in the visible. This can be explained from the fact that the GRB hosts display blue colours (see Sect. 3.2) and that, for the faintest sources at $R\sim26$-29, optical deep observations are generally more sensitive than NIR images to detect blue objects. It is also the reason why the scatter in the optical magnitudes appears larger than in the K-band.

In this "R-z'' diagram, we have also indicated the R magnitudes of optically-selected galaxies from the Caltech Faint Galaxy Redshift Survey (Hogg et al. 2000) and the Hubble Deep Field (Fernández-Soto et al. 1999). For the latter, the R-band photometry was derived from the V and I magnitudes of the catalog assuming a linear interpolation between the mean wavelengths of the V-band (F606 WFPC filter, 6031 Å) and the I-band (F801 WFPC filter, 8011 Å). The conversion from the standard AB magnitudes to the Vega system used throughout this paper was obtained using the calibrations of Fukugita et al. (1995) and Allen (2000).

Again, the GRB hosts in the visible appear just typical of the other optically-selected galaxies at high redshift (see Fig. 3). Yet, it is worth mentioning a particular feature of the GRB host sample, which is clearly apparent in this "R-z'' diagram. Whereas most of field sources at $R\rm ~mag\ga25$ have a redshift only determined with photometric techniques, the GRB hosts have an accurate spectroscopic redshift identification. These redshifts were derived using the emission and/or absorption features detected in the X-ray/optical spectra of GRBs and their afterglows. Such a method is independent of the GRB host luminosities, and only depends on the possibility to rapidly perform spectroscopy of the GRB transient before it has begun to fade. This advantage of GRBs for the selection of high-z sources lies in stark opposition with the deep survey approach. Note that it is particularly apparent in the "R-z'' diagram, but it is not that exceptional at NIR wavelengths (see Fig. 2). As mentioned previously, it is due to the blue colours of GRB hosts, which thus allow the faintest of these hosts detected in the K-band to be spectroscopically observed in the visible.

   
3.2 Colours

So far, the various works related to the understanding of the high redshift Universe have made an extensive use of the integrated R-Kcolours of galaxies as an indicator of their nature. These colours provide indeed a crucial information on the importance of the old stellar populations - as traced by the NIR emission - relative to the contribution of young stars dominating the optical light. For example, unobscured star-forming galaxies are typically blue objects ($R-K\sim2$-3) while old elliptical sources at $z\ga1$exhibit extremely red colours ($R-K\ga5$). Furthermore, large R-K colours in distant sources can also suggest dust obscuration, and may thus sign-post powerful dust-enshrouded starburst galaxies.

In Table 2, we indicate the observed R-K colours for the K-selected sub-sample of GRB host galaxies. The corresponding diagram showing these colours versus redshift is presented in Fig. 4. As in the "K-z'' relation displayed in Fig. 2, we also plotted the R-K colours of optically-selected sources taken from the HDF (Fernández-Soto et al. 1999), and those of the ISO and SCUBA sources already considered in the previous section. The R magnitudes of the ISO galaxies from the CFRS were derived using an interpolation between the V and I magnitudes of Flores et al. (1999). Those of the HDF ISO detections and SCUBA sources were taken from the papers mentioned in Sect. 3.1.

In this diagram, we also indicate the hypothetical colours of typical present-day galaxies if they were moved to higher redshift assuming no evolution of their physical properties. These galactic templates were chosen to be mostly representative of the local Hubble sequence, and include both early-type (E/Sc) and late-type (Scd/Irr) sources. To compute the evolution of their R-K colours with redshift, we used the optical/NIR templates of Mannucci et al. (2001) for the E and Sc types, and the optical Scd and Irr SEDs of Coleman et al. (1980). For the latter, the extrapolation to the near-infrared was derived using the NIR portion of the Mannucci et al. Sc template. This decision was justified from the prescriptions of Pozzetti et al. (1996), which show that the NIR continuum emission of dust-free galaxies longward of 1$~\mu$m always appears dominated by the same stellar populations, and therefore does not vary much from one type to another.

Figure 4: Observed R-K colours versus redshift for the sample of GRB host galaxies listed in Table 2 ( filled diamonds). The estimated uncertainties are indicated with vertical solid lines. The colours and redshifts for optically-selected field sources ( dots) were derived from the HDF source catalog of Fernández-Soto et al. (1999). Solid curves indicate the observed colours of local E, Sc, Scd and Irr galaxies if they were moved back to increasing redshifts (see text for explanations). We also indicated the colours of the ISO sources from the HDF ( open squares) and those of SCUBA galaxies with confirmed redshifts ( filled squares). See Sect. 3.2 for references.

As it can be seen in Fig. 4, the GRB hosts exhibit rather blue colours that are typical of the faint blue galaxy population in the field at $z \sim 1$. Besides, we note that most of them appear even bluer than the colours predicted from the SED of local irregular galaxies. This is similar to what has been already noticed for a large fraction of blue sources detected in the optical deep surveys (Volonteri et al. 2000). Such blue colours originate from the redshifted blue continuum of the OB stars found in HII regions. They are characteristic of unobscured star-forming galaxies, which is not surprising in the scenario linking GRBs to massive star formation. It is moreover in full agreement with the results of Sokolov et al. (2001) who found that the optical SEDs of the GRB host galaxies are consistent with those of the blue starbursts observed in the nearby Universe.

   
3.3 Absolute K magnitudes

Figure 5: Absolute K magnitudes of the GRB host galaxies listed in Table 2, compared to optically/NIR-selected field sources and ISO/SCUBA galaxies in a $\Lambda$CDM Universe ( $\Omega _m=0.3$ and $\Omega _\lambda =0.7$). Legend, photometry and redshift catalogs are similar to those used for Fig. 2, except for the NIR-selected objects with spectroscopic redshifts which are also indicated by dots in this plot. Horizontal lines indicate the magnitudes of 0.1 L*, L*and 3 L* galaxies assuming M*=-25.

We computed the absolute K magnitudes for the sample listed in Table 2, using the optical and NIR galaxy templates described in Sect. 3.2 to derive the k-corrections. For all but two sources, the latter were obtained assuming an SED typical of Irr-type objects as suggested by their blue R-K colours (see Sect. 3.2). In the case of the GRB 990506 and GRB 970828 hosts, we rather used an Scd-type template as indicated by their redder SEDs (see Fig. 4). Luminosity distances were computed assuming a $\Lambda$CDM Universe with $\Omega _m=0.3$and $\Omega _\lambda =0.7$. We parametrized the Hubble constant using h₆₅=H₀ (km s^-1 Mpc^-1) / 65.

The absolute K magnitudes are reported in Table 2 and illustrated in the Hubble diagram of Fig. 5. Again, we also compared the GRB host galaxies with other field sources quoted from the catalogs mentioned in Sect. 3.1. For the galaxies of the HDF, the k-corrections used to compute these magnitudes were derived assuming the best spectral type estimations of Fernández-Soto et al. (1999). For the lower-redshift samples taken from the NIR Hawaii Surveys and for the ISO sources, we arbitrarily assumed an Irr galaxy template relying on the fact that k-corrections are hardly type-dependent up to $z\sim1.5$ in K. The absolute Kmagnitudes of the SCUBA galaxies were determined assuming the SED suggested by their R-K colour (see Fig. 4).

We also indicated the absolute magnitudes of galaxies with luminosities of 0.1 L_* (M=-22.5), L_* (M=-25) and 3 L_* (M=-26.2), assuming M_*=-25. This value was roughly estimated from the Schechter parametrizations of the K-band luminosity function for high redshift galaxies, taken from Cowie et al. (1996) and Kashikawa et al. (2003). Both were determined assuming a matter-dominated Universe with $\Omega_m=1$. Nevertheless, the differences in comoving distance between a Universe with H₀ = 50 ;  $\Omega_m=1$ and one characterized by H₀ = 65 ;  $\Omega _m=0.3$; $\Omega _\lambda =0.7$ imply absolute magnitude variations of only $\Delta m=0.4$ on the $0.7\leq z\leq3$ redshift range. Therefore, it should not significantly affect our qualitative comparison.

With a median-averaged $\overline{M_K}=-22.25$ (corresponding to $\overline{L}\sim0.08~L_*$), the GRB host galaxies are significantly sub-luminous in the K-band. We also note a large difference with the luminosities of massive starbursts probed with ISO and SCUBA, as it was already noticed in Sect. 3.1. Since the NIR emission of galaxies gives a good indication on their mass, the low K-band luminosities of GRB hosts indicate that GRBs, so far, were not observed toward massive objects.