EDP Sciences
Highlight
Free access
Issue
A&A
Volume 534, October 2011
Article Number A15
Number of page(s) 13
Section Extragalactic astronomy
DOI http://dx.doi.org/10.1051/0004-6361/201117649
Published online 22 September 2011

© ESO, 2011

1. Introduction

Accurate measurement of star formation rates (SFR) is a key ingredient for studying galaxy evolution and deriving the census of the star formation activity, both in the distant and in the local universe. To this end, it has been shown that the contribution of luminous infrared galaxies (with infrared luminosities LIR  >  1011 L) to the star formation density is progressively rising as we look back in cosmic time, at least up to z ~ 2. Indeed, although they were found to be rare in the local Universe and to account for only  ~5% of the total infrared energy emitted by galaxies at z ~ 0 (Soifer et al. 1991; Kim & Sanders 1998), LIRGs and ULIRGs (LIR  >  1012 L), dominate the SFR density at z ~ 1−2, accounting for 70% of the star formation activity at these epochs (Papovich et al. 2004; Le Floc’h et al. 2005; Caputi et al. 2007).

The study of infrared sources was greatly facilitated by the advent of the Spitzer Space Telescope (Spitzer Werner et al. 2004). Extragalactic surveys carried out using the MIPS 24   μm band on-board Spitzer confirmed the strong evolution of these sources first indicated by infrared and submillimeter observations using ISO and SCUBA, respectively (Blain et al. 1999a; Elbaz et al. 1999; Serjeant et al. 2001; Dole et al. 2001). Such surveys are believed to detect the bulk of the dusty star forming galaxies up to z ~ 2. However there are two important caveats. The first is that the conversion of 24   μm flux densities to total LIR (and therefore SFR) is subject to large uncertainties, as it relies on extrapolations that strongly depend on the assumed spectral energy distribution (SED) libraries (Chary & Elbaz 2001; Lagache et al. 2003; Dale & Helou 2002). The second comes from the prominent emission and absorption features between 3- and 19   μm in the spectra of star forming galaxies that parade through the 24   μm band at various redshifts.

A large number of studies using the InfraRed Spectrograph (IRS, Houck et al. 2004) have revealed that the vast majority of local LIRGs and ULIRGs exhibit a broad silicate absorption feature centred at 9.7   μm with silicate optical depths ranging from τ9.7 ~ 0.4 to τ9.7 ≥ 4.2 (Brandl et al. 2006; Armus et al. 2007; Pereira-Santaella et al. 2010). Furthermore, using a sample of local ULIRGs, Desai et al. (2007) found strong PAHs and prominent silicate absorption in the H ii and LINER sources and weak PAHs and silicate absorption in Seyferts, suggesting that ULIRGs with strong PAHs but weak silicate absorption are rare. Similar features have been observed in the mid-IR spectra of high-z galaxies (Higdon et al. 2004; Houck et al. 2005). For example, Menéndez-Delmestre et al. (2009) and Farrah et al. (2008) report a median τ9.7 ~ 0.31 for a sample of submillimetre galaxies (SMGs) and IRAC selected ULIRGs respectively, while Sajina et al. (2007) found deeper silicate absorption features (τ9.7 > 1.1) in a sample of z ~ 2 radio-loud galaxies. Although for LIRGs and ULIRGs the identification and measurement of the silicate optical depths is straightforward, this is not the case for normal galaxies (LIR  <  1010 L), as for the latter, it is difficult to discriminate between moderate PAH emission superimposed on a silicate-absorbed continuum and strong PAH features with a relatively weak underlying continuum (Smith et al. 2007).

Whatever its origin, the existence of this broad dip in the mid-IR spectra of star forming galaxies 10   μm could be of particular importance for galaxies in the redshift range of 1 < z < 1.8. At these redshifts the 24   μm filter samples this part of the spectrum and sources with such features would appear faint at 24   μm or even be undetected in this band (depending on the depth of the 24   μm data). A second broad dip that is common in the spectra of star forming galaxies is caused by another silicate absorption feature at 18   μm. This feature would have a similar effect for sources at 0.2 < z < 0.6.

The impact of these features on the mid-IR colours as a function of redshift were presented in detail by Takagi & Pearson (2005), who predicted a population of infrared luminous galaxies at z ~ 1.5 which, due to strong absorption at 9.7   μm, are not detected in the 24   μm band. Subsequent studies that focussed on the search for such silicate absorbed systems employed the 16   μm IRS peak-up image. Kasliwal et al. (2005) suggested that such objects account for more than half of all the sources at z ~ 1−2 predicted by various models. It has also been proposed that the mid-IR colour anomalies introduced by the silicate absorption feature can serve as a redshift indicator for dusty infrared luminous galaxies at z ~ 1.5 (Charmandaris et al. 2004; Teplitz et al. 2011; Armus et al. 2007). Similar claims have also been presented by Pearson et al. (2010), using the AKARI IRC L18W to MIPS24 band colour. These studies raised concerns about a possible bias introduced by the 24   μm selection, in the sense that a significant fraction of z ≤ 2 LIRGs and ULIRGs could remain undetected in 24   μm surveys. However, with little or no information about the far-IR part of the spectrum, these studies were subject to large extrapolation and hence suffered from large uncertainties.

With the successful launch of the Herschel Space Observatory (Herschel) (Pilbratt et al. 2010), we now have access to wavelengths that directly probe the peak of the far-IR emission of high-z galaxies and are in a position to measure with unprecedented accuracy their bolometric output. Deep Herschel extragalactic surveys can be used to determine the accuracy of our extrapolations of the far-IR properties of high-z galaxies as well as test previous claims that 24   μm surveys miss a population of z < 2 LIRGs and ULIRGs (24   μm dropouts). In this paper, we use the deepest Herschel observations to date, as part of the GOODS-Herschel (GOODS-H) program (PI D. Elbaz), covering both the north and the south part of the GOODS fields (GOODS-N and GOODS-S respectively), (Dickinson et al. 2003; Giavalisco et al. 2004), to search for such sources. In Sect. 2 we present the Herschel data, introduce the GOODS-H sample of galaxies, and identify 24   μm dropout sources, i.e., sources detected in the PACS bands but not at 24   μm. In Sect. 3 we investigate the properties of this population, while in Sect. 4 we extend our study to the whole GOODS-H sample. Finally in Sect. 5 we provide estimates of the fraction of z < 2 MIPS dropout sources as a function of the 24   μm, 100   μm and 160   μm sensitivity limits and summarize our results.

2. Herschel data and sample selection

Herschel observations were obtained as part of the open time key program GOODS-H (PI D. Elbaz). The full 10′ × 16′ GOODS-N field was imaged with the PACS (Poglitsch et al. 2010) and SPIRE (Griffin et al. 2010) instruments at 100, 160   μm (PACS) and 250, 350, 500   μm (SPIRE). The total observing time was 124.6 h (~2.5 h/sky position) and 31.1 h for PACS and SPIRE respectively. Similarly a 7′ × 7′ part of the GOODS-S field was observed by PACS over a total of 264 h (~15 h/sky position). Observations of both fields were carried out by adopting the intermediate speed (20′′   s-1) scan-map mode. Both PACS and SPIRE data were processed through the standard Herschel reduction pipeline, version 6.0.3, within the HCSS environment. Additionally, we employed custom procedures aimed at removing of interference patterns, tracking anomalies, re-centering positional offsets, and mapping. A full description of the data reduction procedures will be given in a companion paper (Leiton et al. 2011, in prep.).

2.1. Prior based source extraction; the GOODS-H sample

Given the large beam size of the Herschel bands, (FWHM ~ 6.7′′, 11.2′′, 18.0′′, 25.0′′, 36.0′′ for PACS 100- and 160   μm and SPIRE 250, 350 and 500   μm), a common approach to performing source extraction has been a guided extraction using priors. Here we will give a brief summary of the procedure as an extensive description of the method is given in Elbaz et al. (2011). Source extraction and photometry were obtained from point source fitting at prior positions defined by 24   μm sources with fluxes brighter than S24  ~  20 μJy for the 100   μm maps and down to S24  ~  30 μJy for the 160   μm and 250   μm maps. For the other two SPIRE bands, a secondary criterion was needed, as the 24   μm sources were far too numerous and would lead to an over-deblending of the actual sources. Hence, only sources with S/N > 2 at 250   μm were considered as priors for the longer wavelength SPIRE bands. This choice was optimized by Monte Carlo (MC) simulations to avoid artificial over-deblending of a source, but also to give clean residual maps. Flux uncertainties were based on local estimates of the background noise at the position of the sources from residual images produced after subtracting detected sources, while global noise estimates for the maps were derived from Monte Carlo simulations. This used artificial sources injected into the Herschel maps and source extraction performed in the same manner as for the real sources. The dispersion between input and recovered fluxes provides a secure estimate of the completeness and the noise properties of the map. The two noise estimates were found to be in good agreement.

To construct the GOODS-H sample, we considered sources with flux densities down to 3σ in the PACS bands, i.e. 1.0 and 2.6 mJy (0.7 and 2.6 mJy) at 100 and 160   μm in GOODS-N (GOODS-S). For the GOODS-N sample, where SPIRE data are available we also considered sources down to the 5σ detection limit, i.e. 6.3, 7.1, and 15.0 mJy at 250, 350 and 500   μm, respectively. The choice of a higher S/N cut for the SPIRE catalogues was dictated by the larger beam size and the confusion noise that significantly affects the SPIRE observations (for more details see Elbaz et al. 2011). Herschel catalogs were then matched with the existing multi-wavelength data of the GOODS team to create a multi-band merged catalogue of GOODS galaxies including HST ACS BViz, J, K, IRAC, Spitzer MIPS 24 and 70   μm, Herschel PACS 100 and 160   μm, and Herschel SPIRE 250, 350 and 500   μm. Among our sources, 65% have secure spectroscopic redshifts, while for the rest, we use the reliable compilation of photometric redshifts by Le Borgne et al. (2009). Hereafter we will refer to this sample as the GOODS-H sample.

2.2. Blind source extraction; the 24   μm dropout sample

Since the main aim of this work is to investigate whether Herschel observations reveal a population of galaxies that were previously missed by 24   μm surveys, we also performed blind source extraction in the two PACS bands using Starfinder, a point spread function (PSF) fitting code (Diolaiti et al. 2000). We first extracted PSF profiles from the final science maps that were used to perform source extraction. Aperture corrections were derived based on calibration observation of the asteroid Vesta, while the flux uncertainties were derived based both on the error maps and Monte Carlo simulations, as described above. Monte Carlo simulations were also employed to obtain the level of completeness and the fraction of spurious sources. Both the derived fluxes and the noise properties of the maps are in good agreement with those obtained by the prior based source extraction. Finally, a critical parameter in Starfinder is the correlation threshold (ct), a measure of the similarity between the PSF used for source extraction and the profile of the extracted source, with ct = 1 corresponding to identical profiles. MC simulations indicate that high ct values result in catalogs immune to spurious detections but with lower completeness levels. Similarly lower ct values correspond to higher completeness but also to higher fractions of spurious sources. For our blind catalogs we consider sources with flux densities above the 4σ detection limit at each band and ct > 0.67, for which the fraction of spurious sources is  <3%.

thumbnail Fig. 1

ACS V-band (5′′ × 5′′), IRAC 3.6   μm (20′′ × 20′′), MIPS24   μm (30′′ × 30′′), and PACS 100-, 160   μm (30′′ × 30′′) cut-out images of the 24   μm dropout sources from our sample. The red circles are centred at the IRAC 3.6   μm positions of the sources and their diameter corresponds to the FWHM at each band. The size of each image is denoted on the top of each column.

Open with DEXTER

In principle, our aim was to find sources detected in either of the PACS bands but undetected at 24   μm. Therefore, we first matched the PACS 100- and 160   μm blind catalogs with the MIPS 24   μm sample down to S24  ~  20 μJy (3σ), i.e. the one that served as a pool for the prior based source extraction, starting from the longest wavelength available and using search radii of 7′′ and 11′′ respectively. Sources with 24   μm counterparts were omitted while the rest were matched to the IRAC 3.6 μm catalogue and subsequently to the master GOODS multi-bands catalogue described in the previous section. We also performed photometry in the 24   μm maps at the position of the PACS sources to ensure that there is no 24   μm source at this position, possibly missing from the 24   μm catalogue. We calculated the corrected Poissonian probability that an association of a PACS source within the search radius is a chance coincidence (see Downes et al. 1986) and all sources were found to have a robust (p < 0.05) 3.6 μm counterpart. We also note that flux boosting due to insufficient de-blending, which is the main caveat of blind source extraction, should not, by definition, be an issue for the 24   μm dropout sample. All sources were also inspected by eye and a quality flag was attributed to them. In particular, sources with multiple IRAC counterparts within the PACS beam and sources close to bright objects in the PACS bands were flagged as low quality sources. The final sample consists of 21 MIPS dropout sources, all detected at 100   μm and two at 160   μm, accounting for  ~2% of the total sources detected in the PACS bands. Hence, we find that even at the confusion limit of the 100- and 160   μm passbands (0.7- and 2.6 mJy at a 3σ level), 98% of the Herschel sources possess a robust 24   μm counterpart brighter than 20 μJy. The 21 MIPS dropout sources are shown in Fig. 1 where we present cut-out images at several bands.

3. 24   μm dropout sources

The small number of 24   μm dropouts indicates that the vast majority of PACS sources do have a 24   μm counterpart. In other words, the “normal” SED behaviour of galaxies in the GOODS sample is the one where the relative sensitivity at 24   μm overpowers that of the PACS bands. In Fig. 2 (left) we plot the flux density at 24   μm over that at 100   μm for the whole GOODS-H sample, as well as for the dropout sources and see that the latter depart from the general trend and are relatively faint at 100   μm. We also note that similarly to the MIPS dropouts, some sources with 24   μm detection tend to exhibit redder S100/S24 colours than the bulk of the population while they span a wide range of S24. Here we will study the origin of the departure of the MIPS dropouts from the bulk of the GOODS-H sample and our investigation will be driven by their property that intrigued our interest in the first place i.e. their unusual S100/S24 colour.

3.1. Far-IR properties

The total infrared luminosity (LIR = L8 − 1000   μm) of galaxies in the sample was determined from the 100   μm flux density using the templates of (Chary & Elbaz 2001, CE01) and Dale & Helou (2002, DH02). Despite the lack of data points at longer wavelengths, we note that the monochromatic derivations of total IR luminosities from the 100   μm flux density tend to be robust up to z ~ 1.5 (Elbaz et al. 2010). For the two sources with 160   μm detection LIR was determined from the best fit of the two PACS points, using the whole library of SED templates from CE01 independently of their luminosity (i.e. allowing normalization of all SEDs to the observations) as well as the DH02 templates. The derived luminosities range from 8 × 109 to 2 × 1012 L, with 11 of the MIPS dropout sources having LIR  >  1011 L, belonging to the class of luminous infrared galaxies. To illustrate the completeness of our sample in terms of LIR, we plot the LIR for the GOODS-H sample, as well as for the drop-out sources as a function of redshift, along with the corresponding detection limits at 100   μm and 24   μm (Fig. 2 middle). Furthermore, In Fig. 2 (right), we also plot the S100/S24 colour as a function of LIR as derived from the Herschel data, both for the whole GOODS sample as well as for the MIPS dropouts (crosses with arrows). For the latter we compute lower limits to S100/S24 assuming the 3σ detection limit of the 24   μm maps. The points are also colour-coded based on their redshift. We see that 24   μm dropouts have LIR values similar to that of the whole GOODS-H sample (for a given redshift), while they exhibit significantly higher S100/S24 colours for the whole range of luminosities. As we discussed above, LIR scales with S100. Hence, given the richness of features in the rest frame MIR emission (i.e. Armus et al. 2007), it is more likely that the dropouts have a suppressed S24 emission rather than an excess at S100 when compared to the rest of the GOODS-H sample. In what follows we investigate the origin of this S24 deficit.

thumbnail Fig. 2

Left: S24 vs. S100 flux densities for the whole GOODS-H sample (black circles) as well as for the 24   μm dropout sources (red arrows). For the dropouts we consider a 3σ upper limit for the S24. Dropouts as well as some 24   μm detected sources tend to depart from the bulk of the GOODS-H population, exhibiting redder S100/S24 colours. The cyan line corresponds to S100/S24 = 43. As discussed latter in the paper, sources with S100/S24  >  43 are classified as silicate-break galaxies. Middle: detection limits as a function of redshift for the GOODS-N and GOODS-S PACS 100   μm and MIPS 24   μm observations. Red squares correspond to the drop-out sources. Right: S100/S24 as a function of LIR as derived by Herschel for the whole GOODS-H sample (circles) and lower limits for the 24   μm dropouts (arrows). Both samples are colour coded based on their redshift. Sources with a black cross are AGNs based on their X-ray emission. Filled symbols denote sources in the GOODS-H sample with spectroscopic redshifts while open symbols sources with photometric redshift. Similarly, yellow circles on top of the arrows indicate that a spectroscopic redshift is available for that 24   μm dropout source. The horizontal black dashed line corresponds to S100/S24 = 43.

Open with DEXTER

thumbnail Fig. 3

Redshift distribution of sources in the MIPS dropout sample. A KMM test suggests a bimodal distribution centred at z ~ 0.4 and z ~ 1.3. Blue shadowed area corresponds to the distribution of sources with spectroscopic redshift.

Open with DEXTER

thumbnail Fig. 4

S100/S24 as a function of redshift for the whole GOODS-H sample (circles) and lower limits for the 24   μm dropouts (arrows). Both samples are colour coded based on their LIR. Sources with a black cross are AGNs based on their X-ray emission. Filled symbols denote sources in the GOODS-H sample with spectroscopic redshift, while open symbols are sources with photometric redshift. Similarly, black circles on top of the arrows indicate that a spectroscopic redshift is available for that MIPS dropout source. Solid lines correspond to different observed SEDs of local LIRGs/ULIRGs (see Fig. 5) and horizontal black dashed line to S100/S24 = 43. The pop panel shows the SED of Arp220 at various redshifts along with the MIPS 24- and PACS 100   μm bands.

Open with DEXTER

3.2. S100/S24 colour and redshift distribution

We wish to examine whether the MIPS dropout sources tend to be found at specific redshifts. We note that nine sources have spectroscopic redshift, while for the rest we adopt the photometric redshifts derived by the GOODS team. It appears that 24   μm dropouts are distributed in two redshift bins, one centred at z ~ 0.4 and one at z ~ 1.3 (Fig. 3). This bimodality is verified by a KMM test (Ashman et al. 1994) at a 4.3σ confidence level. Therefore, it seems that our sample is populated by low-z (0.2 < z < 0.6) and high-z (0.9 < z < 1.7) sources. Similarly to Fig. 2, we now plot the S100/S24 colour as a function of redshift (Fig. 4). In this figure we also overplot the S100/S24 colour as a function of redshift for a number of local LIRGs/ULIRGs based on their observed SED as constructed by IRS observations of their mid-IR spectrum (Armus et al. 2007) and IRAS observations of their far-IR emission (Rieke et al. 2009). The observed templates were chosen to span a wide range of LIR and silicate optical depths (τ9.7) and their full SEDs are presented in Fig. 5, in order of increasing τ9.7. The τ9.7 measurements are adopted from Armus et al. (2007) (Arp200, NGC 22491), Pereira-Santaella et al. (2010) (ESO 320-G030, NGC 2369, Zw 049.057, NGC 3256) and da Cunha et al. (2010) (IRAS 17128). On the top of Fig. 4 we also show the SED of Arp220, along with the 24- and 100   μm bands at several redshifts. It is evident that the S100/S24 colour of the templates varies significantly as a function of redshift, mainly due to the presence of the silicate absorption features at 9.7- and 18   μm that enter the MIPS 24   μm filter at z ~ 1.4 and  ~ 0.3 respectively. We also note a wide range of S100/S24 colours for a given redshift, indicative of different amounts of extinction as well as different dust temperatures. Indeed, Fig. 5 suggests a weak trend for SEDs of sources with deeper silicate absorption to peak at shorter wavelengths. Therefore, the S100/S24 values increase because S24 is suppressed by the silicate absorption features, but are also further elevated due to higher dust (big grain) temperatures and hence S100 values.

thumbnail Fig. 5

A large range of rest frame SEDs of local ULIRGs and LIRGs, presented in increasing order of silicate optical depth. The mid-IR part is the IRS spectrum (Armus et al. 2007), while the far-IR comes from Rieke et al. (2009) templates. Blue (orange) stripes correspond to MIPS 24- and PACS 100   μm bands at z = 1.3 (0.3).

Open with DEXTER

Looking back at Fig. 4, we see that 24   μm dropouts, although they have higher S100/S24 values when compared to the whole sample, exhibit colours that are consistent with those of local star formation dominated ULIRGs/LIRGs. The fact that the redshift distribution of 24   μm dropouts peaks at redshifts where the 9.7- and 18   μm silicate absorption features enter the 24   μm band indicates that these sources could have moderate/strong silicate absorption features. On the other hand, the high S100/S24 ratio could also result from lower levels of observed dust continuum emission at 24   μm, but as we will see later that our data disfavour this scenario. In any case, all 24   μm dropouts fall within the envelope defined by the templates, so they form an extreme rather than an extraordinary population of star forming galaxies. We also note that none of the drop-out sources have an X-ray detection or meet the criteria for a power-law AGN.

Previous studies have attempted to identify infrared luminous sources that are undetected at 24   μm by combining 16   μm IRS peak-up imaging with 24   μm data (Kasliwal et al. 2005; Teplitz et al. 2011) and following an approach similar to the one presented in this work. They searched for sources at z ~ 1.3, that are faint at 24   μm, due to the shift of the silicate absorption features into the MIPS band, but bright at 16   μm due to 7.7   μm PAH emission, concluding that sources with S16/S24  >  1.2 tend to be found at z > 1.1. They also investigated the S16/S24 colours of several local LIRGs/ULIRGs and report an average ratio of  ~1–2 for galaxies with strong silicate absorption features. From our 24   μm dropout sample, none of the sources are detected at 16   μm down to the 3σ detection limit (S16  ~  40 μJy for GOODS-N and  ~65 μJy for GOODS-S, Teplitz et al. 2011). Apart from the fact that some of our sources are outside the area covered by the IRS peak-up image in the GOODS fields, the non-detection of the rest of the sources at 16   μm is somewhat expected from the discussion above. Even if we adopt a S24 = 20 μJy for all sources in the sample and assume a S16/S24 = 2.0 (Arp220 case, Armus et al. 2007) then our sources should only be marginally detected at the depth of the 16   μm GOODS maps.

thumbnail Fig. 6

The SED of a dropout at z = 1.68, detected in both the 100- and 160   μm bands, although missed at 24   μm. The optical part is overlaid with the best fit BC03 model (green line) and the infrared part with a range of observed and model SEDs. The red square denotes the upper limit at 24   μm. Cut-out images of the source are shown in Fig. 1, 6th row.

Open with DEXTER

3.3. A z  ~ 1.68 ULIRG, missed by MIPS

Although the study of the properties of individual 24   μm dropout sources is beyond the scope of this study, here we wish to have a closer look at a specific source for which, Herschel data indicate an infrared luminosity LIR  >  1012 L. This is the only source in our high-z sample with detection at 100   μm (S100 = 5.2 ± 0.42 mJy), 160   μm (S160 = 6.9 ± 1.1 mJy) and at 250   μm. The source also has also a  ~6σ detection at the GOODS-N VLA 1.4 GHz map (S1.4 = 32.1 ± 5.1   μJy). Unfortunately, the source is out of the area covered by IRS 16   μm peak-up imaging. Cut out images of this galaxy are shown in the sixth row of Fig. 1. To derive the far-IR properties of the source we fit the observed Herschel points with CE01 models, but also with a range of observed SEDs of local ULIRGs, described in the previous section. The observed multi-band photometry, together with the best fit SEDs are shown in Fig. 6.

The derived infrared luminosity of the source is LIR = 1.3 − 1.6 × 1012 L (depending on the assumed SED), but it is evident that only templates with a strong silicate absorption feature at 9.7   μm can reproduce the non-detection down to 20 μJy at 24   μm. Fitting the optical part of the SED with BC03 models, yields a stellar mass of 8.9 ( ± 0.5) × 1010 M, while based on the β slope of the UV spectrum we derive a reddening E(B − V) = 2.1, consistent with a heavily obscured source. Assuming a Salpeter IMF and the Kennicutt (1998) relation we convert the Herschel-based LIR to SFR and derive SFR ~ 280   M yr-1. Similarly, we convert the radio flux to LIR (Condon 1992) and subsequently to SFR, finding that the two estimates are in perfect agreement (SFRradio = 300 ± 32 M yr-1). The source is not detected in X-rays and exhibits the 1.6   μm stellar bump, indicating that there is no near-to-mid-IR evidence for the presence of an AGN. The above analysis suggests that this is a heavily obscured starburst galaxy with no signs of AGN activity. Although unique in our sample, this galaxy raises the question of how many such objects we might have missed in the pre-Herschel era and what would be their contribution to the cosmic SFR density. A detailed discussion on this point will be presented in Sect. 5.

4. Silicate-break galaxies

So far, we have demonstrated that Herschel data have revealed a small but interesting class of infrared luminous galaxies, which are undetected at 24   μm. The main characteristic of these sources is their atypically red S100/S24 colour. Furthermore, Fig. 2 (left), reveals that sources with S100/S24 colours similar to those of the dropouts can be found among sources with 24   μm detection. In what follows we extend our study to the whole GOODS-H sample, searching for such objects.

To select them, we adopt a cut off, S100/S24  >  43, that corresponds to the bluest lower limit among the MIPS dropout sample and the colour of the local LIRG ESO320 shifted at z = 1.4. In Fig. 7 we plot the redshift distribution of sources with S100/S24  >  43 (excluding the dropouts), finding that it bears a remarkable resemblance to that of the 24   μm dropouts in Fig. 3. Indeed, a K-S test reveals that there is no significant difference between the two samples with a p value of 0.61. In contrast, we find that the sample has a redshift distribution quite different than that of the whole GOODS-N sample (Fig. 7, top panel) at a confidence level of  >97.5%. Similar to the dropout sample, sources with S100/S24  >  43 exhibit a bimodal redshift distribution at the 3.5σ level, centred at z ~ 0.48 and  ~1.3. Therefore, in what follows, we suggest that the S100/S24 colour could serve as redshift indicator.

thumbnail Fig. 7

Top: redshift distribution of sources in GOODS-N with PACS 100   μm and MIPS 24   μm detections. Bottom: redshift distribution of sources in GOODS-N with S100/S24  >  43. Sources are clustered around z = 0.4 and z = 1 .3. The blue histogram corresponds to the redshift distribution of MIPS dropouts.

Open with DEXTER

thumbnail Fig. 8

S100/S24 vs. S16/S8 colour − colour diagram for the whole GOODS-H sample in GOODS-N with 16   μm detection (black circles). Circles filled with green and orange colour are sources with S100/S24  >  43 at 0.2 < z < 0.6 and 1.0 < z < 1.7, respectively. Vertical and horizontal red lines indicate a cut off ratio of S16/S8  >  4 and S100/S24  >  43. We see that this diagram can serve as a redshift diagnostic, for sources at 1.0 < z < 1.7.

Open with DEXTER

4.1. S100/S24 colour as a redshift indicator

In the pre-Herschel era, there were several attempts to use anomalous MIR colours as a crude redshift indicator. These studies mainly employed the 16   μm IRS peak-up imaging and proposed that a blue S16/S24 colour would peak in a narrow redshift bin, 1.0 < z < 1.8 (Armus et al. 2007; Teplitz et al. 2011), as well as being useful for selecting infrared luminous galaxies in this redshift bin that were undetected by the 24   μm band. The main idea behind this criterion is that at these redshifts the 9.7   μm silicate absorption feature enters in the 24   μm passband, while the strong PAH emission features at 6.2- and 7.7 μm are shifted into the 16   μm band, producing a distinctive blue S16/S24 colour. For example, Kasliwal et al. (2005) used a S16/S24  >  1.2 ratio to select objects at 1.1 < z < 1.6, and called them “silicate-break galaxies”, attributing their blue S16/S24 colour to the existence of a strong silicate absorption feature at 9.7   μm. This approach had two main caveats. First, the blue S16/S24 colour is not necessarily produced by a silicate absorption feature, as similar colours can appear for sources with strong PAHs and low dust continuum. Second, many objects at lower redshifts fall within the same colour cut (Teplitz et al. 2005). A more recent attempt by Teplitz et al. (2011), reports that a higher ratio (S16/S24  >  1.4), would eliminate many but not all of the low-z interlopers.

Here, we face the same situation. The cut in the S100/S24 ratio that we have adopted selects sources in two redshifts bins. In order to reject the low-z sources in our sample, simply increasing the S100/S24 cut is not useful as apart from the low-z, we also miss many high-z sources and the selection is still not pure enough. Alternatively, we can employ a second colour criterion, based on the 16   μm and 8   μm flux densities. In Fig. 8 we plot the S100/S24 vs. S16/S8 colour − colour diagram for sources that have a 16- and 8   μm detection. We see that if a ratio cut of S16/S8  >  4 is combined with a S100/S24 cut  > 43, then we successfully reject all low-z interlopers, while selecting sources in the 1.0 < z < 1.7 redshift bin. The selection of S16/S8 as a second colour criterion was driven by the fact that at 1.0 < z < 1.7 the 16   μm band probes the PAH complex (6.2 − 7.7   μm), boosting the value of S16 while for the low redshift sources only traces emission from a warm dust continuum at  > 10   μm. We note that none of our sources is classified as an AGN, based on their X-ray emission, their optical spectra or their mid-IR colours (i.e. power law AGNs). It therefore seems that we have found a way to select star-forming high-z galaxies in a narrow redshift bin. In what follows we argue that these sources are compact starbursts with moderate/strong silicate features in their MIR spectrum.

4.2. Evidence for silicate absorption

We have already discussed in the Introduction, that for normal galaxies, i.e. those with LIR  <  1010 L, discriminating between moderate PAH emission superposed on a silicate-absorbed continuum and strong PAH features with a relatively weak underlying continuum is a difficult task (e.g. Smith et al. 2007), even when high quality mid-IR spectra are available. A typical example is M 82, for which Sturm et al. (2000) suggests that there is no silicate absorption since the 10   μm dip can be reproduced by a superposition of strong PAHs and VSG continuum. On the other hand, things are more straightforward for more luminous infrared sources, where there is clear evidence for the existence of a wide range of silicate optical depths, both for local and high-z LIRGs and ULIRGs (Armus et al. 2007; Pereira-Santaella et al. 2010). Unlike the sample based on the blue S16/S24 colour, for which the LIR estimates were based on large/uncertain extrapolations, our study benefits from more robust Herschel-based LIR estimates. The fact that all sources with S100/S24  >  43 at 1.0 < z < 1.7 in our sample (including the high-z 24   μm dropouts) have LIR  >  1011 L, coupled with the lack of observational evidence of sources with similar luminosities and low dust continuum emission, is a first hint that these sources have a silicate absorption feature at 9.7   μm.

We have already demonstrated in Fig. 4, that these sources have S100/S24 colours consistent with those of local templates of LIRGs and ULIRGs with moderate/high silicate strength. To further investigate this, we plot the distribution of the S100/S24 colour of the GOODS-H (GOODS-N and GOODS-S) sources with spectroscopic redshift at 1.0 < z < 1.7, along with the range of the S100/S24 colour at this redshift range for 5 LIRG/ULIRG observed templates with a wide range of τ9.7 values (Fig. 9). In practice, this plot examines the detailed distribution of galaxies in the 1.0 < z < 1.7 region of Fig. 4. Fitting the colour distribution with a Gaussian indicates a clear excess in the red tail due to sources with S100/S24  >  43. It is crucial to stress that this excess is absent in other redshifts bins. Furthermore, we see that based on the template SEDs, sources with higher optical depth exhibit redder S100/S24 colours in this redshift bin. In other words, it seems that the S100/S24 colour could serve as a rough indicator of the silicate strength. Our sources, i.e. those with S100/S24  >  43, have colours consistent with 0.7 < τ9.7 < 1.9, while it appears that none of them have a silicate feature as strong as that of Arp220. We note however, that we cannot rule out a source in our sample with stronger silicate but colder Tdust.

thumbnail Fig. 9

Distribution of the S100/S24 colour among sources in the GOODS-H sample with spectroscopic redshift at 1.0 < z < 1.7. Green line and green shadowed area indicates the Gaussian fit to the distribution and the area beneath it. Coloured bars indicate the range of the S100/S24 colour of several observed SEDs of local LIRGs and ULIRGs at this redshift bin. The silicate optical depths of the local sources, as derived by IRS spectroscopy are also overlaid. The black arrow indicates the position of the high-z 24   μm dropout sample.

Open with DEXTER

thumbnail Fig. 10

S160/S100 vs. S100/S24 colour − colour diagram for sources in the GOODS-H sample at 1.0 < z < 1.7. Sources with S100/S24  >  43 and S16/S8  >  4 are filled with orange colour. This plot demonstrates that our selection is not biased towards warmer sources.

Open with DEXTER

An alternative explanation of the high S100/S24 ratios of these sources would be enhanced 100   μm emission, simply because the sources are warmer, and hence their SEDs peak at shorter wavelengths. To test this, we investigate the S100/S24 colour as a function of dust temperature, as indicated by the S160/S100 ratio. In Fig. 10, we present this colour − colour diagram for sources at 1.0 < z < 1.7, which are detected at 160-and 100   μm. It appears that the adopted S100/S24 cut does not introduce a bias towards warmer sources (i.e. sources with low S160/S100 colours), as it selects sources with a wide range of dust temperatures similar to that found for the whole GOODS-H sample in this redshift bin, suggesting that an enhanced S100 is not the main reason for their high S100/S24 ratios. We note that a large dispersion of dust temperature of high-z galaxies has recently been demonstrated by Hwang et al. (2010) and Magdis et al. (2010d).

One of the sources that meets the selection criteria is the well studied sub-millimetre source at z = 1.21, GN26 for which Pope et al. (2008) have published an IRS spectrum. The detection of this source in the PACS bands has also been discussed by Dannerbauer et al. (2010) and Magnelli et al. (2010) while Frayer et al. (2008) has reported the detection of CO(2 → 1) emission. In Fig. 11, we show the full SED of the source, overlaid with the best CE01 template for the far-IR part and the observed IRS spectrum for the mid-IR part of the SED. The IRS spectrum of GN26 is also presented separately in an inset panel. This source has S100/S24  ~  70, and according to Fig. 9, it should have a silicate optical depth of  ~1. Although the S/N of the IRS spectrum does not allow for a robust measurement of τ9.7, its IRS spectrum is very similar to that of the composite spectrum of SMGs presented in Pope et al. (2008), and for which they report τ9.7 ~ 1 in excellent agreement with what we expected based on this source’s S100/S24 colour. We should also note that GN26 was one of the 13 sources used for the construction of the composite spectrum. Taken together, the evidence suggests that the red S100/S24 colour of our sample is caused by the existence of a moderate/strong silicate absorption feature at 9.7   μm, that enters the 24   μm band at these redshifts. We therefore choose to characterize these sources as silicate-absorbed galaxies.

thumbnail Fig. 11

Full SED of GN26, a source at z = 1.21 with S100/S24  ~  70. The far-IR part of the SED is derived from Herschel data and is overlaid with the best fit CE01 model. The mid-IR part of the SED is the observed IRS spectrum from Pope et al. (2008), that is also presented separately in the inset figure. We also show the optical/near-IR part of the SED along with the best fit BC03 model.

Open with DEXTER

thumbnail Fig. 12

S100/S24 as a function of the Herschel derived LIR over the LIR derived based only on the 24   μm flux densities. Green circles are sources at 0.2 < z < 0.6 and orange circles are sources at 1.0 < z < 1.7. The top right box (defined by S100/S24  >  43 and LIR   Herschel/LIR   24 > 3) consists mainly of sources at 1.0 < z < 1.7. Vertical dashed lines indicate the area where the two LIR estimates agree within a factor of 1.5.

Open with DEXTER

thumbnail Fig. 13

Left: the SFR − M relation for sources with S100/S24  >  43 at 1.1 < z < 1.7 (black circles). Orange dots indicate sources with spectroscopic redshift. Red cross corresponds to the ULIRG dropout that we presented in Sect. 3.3. The solid green line depicts the SFR − M correlation at z ~ 1.4 (Elbaz et al. 2011), while the dashed lines indicate its dispersion. Right: specific star formation rate (sSFR), as a function of redshift, for the same sample as in the left panel, as well as for the whole GOODS-H sample (grey dots). The gray thick line denotes the evolution of the sSFR with redshift, as derived by individual detections and stacking analysis of the GOODS-H sample by Elbaz et al. (2011). The blue shaded area along with the dashed gray lines which correspond to the dispersion of the evolution (~0.3 dex), indicate the region of main sequence galaxies. Sources with S100/S24  >  43 tend to be found above the blue shaded area, populating the starburst region. The green text indicates the position of Arp 220 and M 82 (shifted to z = 0.1 for clarity).

Open with DEXTER

4.3. The far-IR properties of the high-z sample

In the pre-Herschel era the far-IR properties of large samples of galaxies were derived based on large extrapolations of their 24   μm flux density using SED templates based on local IR-luminous galaxies. Recent studies using Herschel data have confirmed the validity of these extrapolations, demonstrating that the 24   μm flux density is a good proxy of the total LIR, at least up to z ~ 1.5 (Elbaz et al. 2010). On the other hand, here we have found a population of high-z sources (1.0 < z < 1.7), that exhibit atypically red S100/S24 colours. For these sources, we expect that LIR estimates based on their S24, and using average template SEDs, such as CE01 and DH02, would be severely underestimated.

In Fig. 12 we plot the S100/S24 colour of all GOODS-H galaxies as a function of the ratio between the Herschel based LIR and the one derived using only the 24   μm flux density. Although for the majority of the sources the two LIR estimates are in good agreement (within a factor of  ~1.5), for the high-z sources with S100/S24  >  43 the 24   μm flux density would underestimate the true LIR on average by a factor of  > 3. This is not the case for the low-z sample where the two estimates are in better agreement. For the high-z sample there is a clear trend between the S100/S24 colour and the ratio of the two LIR estimates, in the sense that the true LIR is progressively underestimated for sources with redder S100/S24 colours, depicting the limited variety of SEDs used for the derivation of the LIR. This indicates that although the average template SEDs are representative for the bulk of the galaxy population at z < 1.5, this is not the case for a population at 1.0 < z < 1.7 with very red S100/S24 colours (>43 for this study).

4.4. The starburst nature of the high-z sample

As the silicate absorption feature merely requires a mass of warm dust obscured by a significant column of cooler dust, it does not provide any insight into the mechanism that is heating the warm dust. Hence, it is difficult to establish a correlation between the strength of this feature and the source that powers the mid-IR emission of infrared luminous galaxies, as it could equally be produced by a deeply buried AGN or a compact starburst (Farrah et al. 2008; Imanishi et al. 2009; Armus et al. 2007).

We have already reported than none of our sources show direct signs of AGN activity, since none of them are either (a) detected in X-rays in the 2 Msec Chandra observations (Alexander et al. 2003), (b) satisfies the criteria of power-law AGN or (c) has high excitation lines in their optical spectra (where available). A deeply obscured AGN though, cannot be ruled out. To investigate this, we stack the Chandra X-ray data on these sources. We find a strong detection (7σ) in the 0.5 − 2 keV band corresponding to L2 − 10 = 3 × 1041 erg s-1 and no detection in the 2 − 7 keV band (3σ upper limit 1.28 × 10-17 erg cm-2 s-1). The derived upper limit in the hard band along the stacking results in the soft band suggest that the sources are dominated by star formation (Nandra et al. 2002; Lehmer et al. 2008). Furthermore, in most cases the radio based LIR is in good agreement with that derived by Herschel, again indicating that those sources are dominated by a nuclear starburst, rather than an AGN.

It has recently been shown that normal star-forming galaxies exhibit a correlation between their SFR and stellar mass at any given redshift. This correlation was first found among z ~ 0 galaxies (Brinchmann et al. 2004) and was subsequently extended to higher redshifts, z ~ 1 Elbaz et al. (2007), Noeske et al. (2007), z ~ 2 Daddi et al. (2007), Pannella et al. (2009), z ~ 3 Magdis et al. (2010a,b,c) and z ~ 4 Daddi et al. (2009). It has also been shown that star forming galaxies that do not follow this correlation tend to undergo a rapid starburst phase and have more compact geometries (Elbaz et al. 2011). For instance, at z ~ 0 and z ~ 2, respectively, local ULIRGs and SMGs have SFRs that greatly exceed the SFR − M correlation. Both classes of objects are thought to host compact starbursts (Daddi et al. 2010; Tacconi et al. 2010). It would therefore be interesting to investigate the position of the sources with red S100/S24 colour in the SFR–M diagram. Using the Kennicutt (1998) relation and a Salpeter IMF, we convert the Herschel based LIR to SFR for the sources with S100/S24  >  43 and 1.1 < z < 1.7 and we plot the derived SFR versus the stellar mass of the galaxies (Fig. 13, left). We also overplot the SFR − M   correlation at the median redshift of the sample i.e., z = 1.4 (Elbaz et al. 2011). Clearly the sources are off the correlation, exhibiting enhanced star formation activity for their stellar mass. Similarly the ULIRG dropout source that we presented in Sect. 3.3 is also off the correlation. To eliminate the effect of the evolution of the SFR − M with redshift, we also show the specific SFR (sSFR), defined as SFR/M, as function of redshift for sources with S100/S24  >  43, as well as for the whole GOODS-H sample (Fig. 13, right). We also overplot the evolution of sSFR with time as derived by Elbaz et al. (2011), based both on detected sources and stacking analysis of the GOODS-H sample. According to Elbaz et al. (2011), sources within the blue shaded area are main sequence, normal star forming galaxies galaxies while sources above it, are considered to undergo a starburst phase. We find that sources with S100/S24  >  43 tend to have higher sSFRs when compared to main sequence star forming galaxies at this redshift range, and populate the starbursts region. This result is coherent with the picture where these sources are compact starbursts with high sSFR and where (like local ULIRGs) a strong silicate absorption feature at 9.7   μm is present in their mid-IR spectra.

thumbnail Fig. 14

The fraction of sources expected to be missed in the 24   μm band at 0.2 < z < 0.6 (left) and 1.0 < z < 1.7 (right) as a function of the 24   μm vs. 100   μm (top) and the 24   μm vs. 160   μm (bottom) depths. Black circles indicate the depths of PACS and MIPS24   μm surveys for some of the most important cosmological fields, i.e. EGS, Lockman and COSMOS. We note that the region where the constant fraction isocontours run horizontally, cannot be constraint by the data. The adopted detection limits (5σ) at 24   μm are 80   μJy for COSMOS and 50   μJy for EGS and Lockman. Similarly the 3σ detection limits at 100   μm are 5.0-, 3.8- and 3.6 mJy (COSMOS, EGS and Lockman), while at 160   μm are 10.2-, 8.6- and 7.5 mJy.

Open with DEXTER

5. Discussion

Previous studies predicted that the number of such silicate break sources could reach  ~900 to 1500 sources per square degree, depending on the assumed model (e.g. Tagagi & Pearson 2005). Furthermore, Kasliwal et al. (2005), based on 16   μm IRS data, reported that such sources account for more than half of all galaxies at z ~ 1−2 predicted by various models. The samples of silicate-break galaxies and MIPS dropout sources that we have found here can place strong limits on the number of 1.0 < z < 2.0 infrared luminous galaxies with similar properties.

In the 250 arcmin2 covered by PACS in the two GOODS fields, we have detected  ~30 (7) silicate absorbed candidates at 1.0 < z < 2.0 with S100/S24  >  43 and LIR  >  1011 L (LIR  >  1012 L). Assuming that all sources with S100/S24  >  43 at this redshift are silicate absorbed starbursts, and neglecting the effects of cosmic variance, this implies a surface density of  ~540 (220) sources deg-2, significantly lower than previous predictions. Using the co-moving volume within 1 < z < 2, this implies a space density of Si absorbed ULIRGs in this redshift bin of  ~2.0 ( ± 0.3) × 10-5 Mpc-3. Comparing the numbers with the rest high-z sample of GOODS-H, we find that these sources account for the  ~8% and 16% of the LIR  >  1011 L and LIR  >  1012 L sources at 1.0 < z < 2.0. We note though, that our sample is not complete. As illustrated in Fig. 9, sources with silicate absorption feature but cold Tdust have S100/S24  <  43 and are missed from the selection. We therefore conclude that this estimate is a rather conservative lower limit. Moving to the dropout sample, we have identified 11 sources with LIR  >  1011 L, accounting for approximately 1 − 2% of the population of 1 < z < 2 infrared luminous galaxies in GOODS-H as a whole. We note that by comparing to the whole GOODS-H sample in the same redshift range, all effects of incompleteness are taken into account. Finally, these number should be treated as upper limits, given that some of these could be spurious detections, as discussed in Sect. 2.

thumbnail Fig. 15

The fraction of sources expected to be missed by the 24   μm band at all redshifts as a function of the 24 μm vs. 250   μm and the 24   μm vs. to 350   μm depth. Black circles indicate the depths of PACS and MIPS24   μm surveys for some of the most important cosmological fields, i.e. EGS, SWIRE and COSMOS.

Open with DEXTER

These estimates confirm that for the GOODS surveys, 24   μm observations recover the vast majority of z < 2 sources and do not introduce a strong selection bias. Furthermore, we confirm that using the 24   μm catalogues to define priors for the extraction of PACS sources should only miss a small fraction of high-z sources. Of course the ratio between the number of sources missed by the 24   μm band and the total number of sources in the PACS bands strongly depends on the depth of the MIPS 24   μm and PACS observations. This is illustrated in Fig. 14, where we show the fraction of the expected dropouts at 0.2 < z < 0.6 and 1.0 < z < 1.7 as a function of the 24- to 100   μm and the 24- to 160   μm depth. These estimates are solely based on observations (i.e. on the GOODS-H sample), without any further assumptions. We simply calculate the ratio of fdrop:

where α and β are free parameters. These diagrams can be used as diagnostic of the fraction of sources that will be missed by a source extraction method based on 24   μm priors for several extragalactic Herschel surveys, e.g. PEP (Lutz et al. 2011), HerMES (Oliver et al., in prep.). We conclude that for the major extragalactic surveys, there is not a large population of silicate break galaxies that would have been undetected in 24   μm Spitzer data as some authors have previously suggested. This discrepancy could be indicative of an evolution of the strength of the silicate absorption feature with time. If the silicate features were weaker in early galaxies due to more extended star forming regions, then theoretical models based on local templates would grossly overestimate the number density of such sources. This scenario is in line with recent findings, where the star formation activity of the majority of high-z galaxies (including those with high LIR), is similar to that of local spiral galaxies (e.g. Daddi et al. 2010; Genzel et al. 2010).

Finally, although in this work we have mainly focused on PACS data, it is worth attempting to quantify the fraction of sources that would be missed by a 24   μm prior based source extraction in the SPIRE bands. The philosophy behind Fig. 15 is identical to that of Fig. 14, but this time considering the whole redshift range of the sources and the S250 (left) and S350 (right) flux density limits. We note that these detection limits, indicate the expected instrumental noise and do not take into account the confusion noise, which is dominant in the SPIRE bands.

6. Conclusions

We have presented a study dedicated to sources that exhibit atypical S100/S24 colours using the deepest PACS data to date, obtained as part of the GOODS-H program. By performing blind source extraction we searched for sources that are bright in the far-IR but undetected at 24   μm, i.e. for 24   μm dropout galaxies. Then we investigated the properties of sources in the GOODS-H sample with red S100/S24 colours searching for the population of silicate-break galaxies. The main results are summarized below:

  • We have identified 21 PACS sources that are undetected at24   μm (down to a 3σ detection limit of  ~20 μJy). These 24   μm dropout sources are found to have a bimodal redshift distribution, with peaks centred at z ~ 0.4 and  ~1.3, and are expected to exhibit strong silicate absorption features, responsible for their depressed 24   μm emission. Among the sources in higher redshift we identify 10 LIRGs and one ULIRG at z = 1.68. This enables us to place upper limits in the fraction of LIRGs/ULIRGs that are missed by 24   μm surveys.

  • The vast majority of Herschel PACS sources are detected at 24   μm, indicating that a prior-based source extraction based on the 24   μm emission of the galaxies suffers only very modest incompleteness, with MIPS dropout sources accounting only for  ~2% of the infrared luminous population in the GOODS fields. Although this fraction is negligible for the GOODS surveys, 24   μm dropouts may be a concern for other Herschel extragalactic surveys with shallower 24   μm data.

  • Based on the mid- and far-IR colours of sources in the GOODS-H sample, we demonstrated that sources with S100/S24  >  43 and S16/S8  >  4 are located in a narrow redshift bin, 1.0 < z < 1.7. Furthermore, we provided evidence that sources selected in this manner are starburst-dominated and with compact geometries. Similarly to the dropouts, the red S100/S24 colours of these sources are attributed to the 9.7   μm silicate absorption feature in their mid-IR spectra that enters into the 24   μm band. We characterize them as silicate-absorbed galaxies.

  • The infrared luminosity of these silicate-absorbed galaxies, when derived based on their monochromatic 24   μm flux density, is on average underestimated by a factor  ~3. They account for about 16% of the ULIRGs in the GOODS fields with a space density of 2.0 × 10-5 Mpc-3.

  • We provide diagnostic diagrams to estimate the fraction of sources expected to be missed in the 24   μm band for several Herschel extragalactic surveys, and predict that for most of them that fraction is less that 10%.

Acknowledgments

G.E.M. acknowledge the support of the Centre National de la Research Scientifique (CNRS) and the University of Oxford. H.S.H. and D.E. acknowledge the support of the Centre National d’Études Spatiales (CNES) PACS has been developed by a consortium of institutes led by MPE (Germany) and including UVIE (Austria); KU Leuven, CSL, IMEC (Belgium); CEA, LAM (France); MPIA (Germany); INAFIFSI/OAA/OAP/OAT, LENS, SISSA (Italy) and IAC (Spain). This development has been supported by the funding agencies BMVIT (Austria), ESA-PRODEX (Belgium), CEA/CNES (France), DLR (Germany), ASI/INAF (Italy), and CICYT/MCYT (Spain). SPIRE has been developed by a consortium of institutes led by Cardiff University (UK) and including Univ. Lethbridge (Canada); NAOC (China); CEA, LAM (France); IFSI, Univ. Padova (Italy); IAC (Spain); SNSB (Sweden); Imperial College London, RAL, UCL-MSSL, UKATC, Univ. Sussex (UK); and Caltech, JPL, NHSC, Univ. Colorado (USA). This development has been supported by national funding agencies: CSA (Canada); NAOC (China); CEA, CNES, CNRS (France); ASI (Italy); MCINN (Spain); Stockholm Observatory (Sweden); STFC (UK); and NASA (USA). This work is based [in part] on observations made with Herschel, a European Space Agency Cornerstone Mission with significant participation by NASA. Support for this work was provided by NASA through an award issued by JPL/Caltech.

References

All Figures

thumbnail Fig. 1

ACS V-band (5′′ × 5′′), IRAC 3.6   μm (20′′ × 20′′), MIPS24   μm (30′′ × 30′′), and PACS 100-, 160   μm (30′′ × 30′′) cut-out images of the 24   μm dropout sources from our sample. The red circles are centred at the IRAC 3.6   μm positions of the sources and their diameter corresponds to the FWHM at each band. The size of each image is denoted on the top of each column.

Open with DEXTER
In the text
thumbnail Fig. 2

Left: S24 vs. S100 flux densities for the whole GOODS-H sample (black circles) as well as for the 24   μm dropout sources (red arrows). For the dropouts we consider a 3σ upper limit for the S24. Dropouts as well as some 24   μm detected sources tend to depart from the bulk of the GOODS-H population, exhibiting redder S100/S24 colours. The cyan line corresponds to S100/S24 = 43. As discussed latter in the paper, sources with S100/S24  >  43 are classified as silicate-break galaxies. Middle: detection limits as a function of redshift for the GOODS-N and GOODS-S PACS 100   μm and MIPS 24   μm observations. Red squares correspond to the drop-out sources. Right: S100/S24 as a function of LIR as derived by Herschel for the whole GOODS-H sample (circles) and lower limits for the 24   μm dropouts (arrows). Both samples are colour coded based on their redshift. Sources with a black cross are AGNs based on their X-ray emission. Filled symbols denote sources in the GOODS-H sample with spectroscopic redshifts while open symbols sources with photometric redshift. Similarly, yellow circles on top of the arrows indicate that a spectroscopic redshift is available for that 24   μm dropout source. The horizontal black dashed line corresponds to S100/S24 = 43.

Open with DEXTER
In the text
thumbnail Fig. 3

Redshift distribution of sources in the MIPS dropout sample. A KMM test suggests a bimodal distribution centred at z ~ 0.4 and z ~ 1.3. Blue shadowed area corresponds to the distribution of sources with spectroscopic redshift.

Open with DEXTER
In the text
thumbnail Fig. 4

S100/S24 as a function of redshift for the whole GOODS-H sample (circles) and lower limits for the 24   μm dropouts (arrows). Both samples are colour coded based on their LIR. Sources with a black cross are AGNs based on their X-ray emission. Filled symbols denote sources in the GOODS-H sample with spectroscopic redshift, while open symbols are sources with photometric redshift. Similarly, black circles on top of the arrows indicate that a spectroscopic redshift is available for that MIPS dropout source. Solid lines correspond to different observed SEDs of local LIRGs/ULIRGs (see Fig. 5) and horizontal black dashed line to S100/S24 = 43. The pop panel shows the SED of Arp220 at various redshifts along with the MIPS 24- and PACS 100   μm bands.

Open with DEXTER
In the text
thumbnail Fig. 5

A large range of rest frame SEDs of local ULIRGs and LIRGs, presented in increasing order of silicate optical depth. The mid-IR part is the IRS spectrum (Armus et al. 2007), while the far-IR comes from Rieke et al. (2009) templates. Blue (orange) stripes correspond to MIPS 24- and PACS 100   μm bands at z = 1.3 (0.3).

Open with DEXTER
In the text
thumbnail Fig. 6

The SED of a dropout at z = 1.68, detected in both the 100- and 160   μm bands, although missed at 24   μm. The optical part is overlaid with the best fit BC03 model (green line) and the infrared part with a range of observed and model SEDs. The red square denotes the upper limit at 24   μm. Cut-out images of the source are shown in Fig. 1, 6th row.

Open with DEXTER
In the text
thumbnail Fig. 7

Top: redshift distribution of sources in GOODS-N with PACS 100   μm and MIPS 24   μm detections. Bottom: redshift distribution of sources in GOODS-N with S100/S24  >  43. Sources are clustered around z = 0.4 and z = 1 .3. The blue histogram corresponds to the redshift distribution of MIPS dropouts.

Open with DEXTER
In the text
thumbnail Fig. 8

S100/S24 vs. S16/S8 colour − colour diagram for the whole GOODS-H sample in GOODS-N with 16   μm detection (black circles). Circles filled with green and orange colour are sources with S100/S24  >  43 at 0.2 < z < 0.6 and 1.0 < z < 1.7, respectively. Vertical and horizontal red lines indicate a cut off ratio of S16/S8  >  4 and S100/S24  >  43. We see that this diagram can serve as a redshift diagnostic, for sources at 1.0 < z < 1.7.

Open with DEXTER
In the text
thumbnail Fig. 9

Distribution of the S100/S24 colour among sources in the GOODS-H sample with spectroscopic redshift at 1.0 < z < 1.7. Green line and green shadowed area indicates the Gaussian fit to the distribution and the area beneath it. Coloured bars indicate the range of the S100/S24 colour of several observed SEDs of local LIRGs and ULIRGs at this redshift bin. The silicate optical depths of the local sources, as derived by IRS spectroscopy are also overlaid. The black arrow indicates the position of the high-z 24   μm dropout sample.

Open with DEXTER
In the text
thumbnail Fig. 10

S160/S100 vs. S100/S24 colour − colour diagram for sources in the GOODS-H sample at 1.0 < z < 1.7. Sources with S100/S24  >  43 and S16/S8  >  4 are filled with orange colour. This plot demonstrates that our selection is not biased towards warmer sources.

Open with DEXTER
In the text
thumbnail Fig. 11

Full SED of GN26, a source at z = 1.21 with S100/S24  ~  70. The far-IR part of the SED is derived from Herschel data and is overlaid with the best fit CE01 model. The mid-IR part of the SED is the observed IRS spectrum from Pope et al. (2008), that is also presented separately in the inset figure. We also show the optical/near-IR part of the SED along with the best fit BC03 model.

Open with DEXTER
In the text
thumbnail Fig. 12

S100/S24 as a function of the Herschel derived LIR over the LIR derived based only on the 24   μm flux densities. Green circles are sources at 0.2 < z < 0.6 and orange circles are sources at 1.0 < z < 1.7. The top right box (defined by S100/S24  >  43 and LIR   Herschel/LIR   24 > 3) consists mainly of sources at 1.0 < z < 1.7. Vertical dashed lines indicate the area where the two LIR estimates agree within a factor of 1.5.

Open with DEXTER
In the text
thumbnail Fig. 13

Left: the SFR − M relation for sources with S100/S24  >  43 at 1.1 < z < 1.7 (black circles). Orange dots indicate sources with spectroscopic redshift. Red cross corresponds to the ULIRG dropout that we presented in Sect. 3.3. The solid green line depicts the SFR − M correlation at z ~ 1.4 (Elbaz et al. 2011), while the dashed lines indicate its dispersion. Right: specific star formation rate (sSFR), as a function of redshift, for the same sample as in the left panel, as well as for the whole GOODS-H sample (grey dots). The gray thick line denotes the evolution of the sSFR with redshift, as derived by individual detections and stacking analysis of the GOODS-H sample by Elbaz et al. (2011). The blue shaded area along with the dashed gray lines which correspond to the dispersion of the evolution (~0.3 dex), indicate the region of main sequence galaxies. Sources with S100/S24  >  43 tend to be found above the blue shaded area, populating the starburst region. The green text indicates the position of Arp 220 and M 82 (shifted to z = 0.1 for clarity).

Open with DEXTER
In the text
thumbnail Fig. 14

The fraction of sources expected to be missed in the 24   μm band at 0.2 < z < 0.6 (left) and 1.0 < z < 1.7 (right) as a function of the 24   μm vs. 100   μm (top) and the 24   μm vs. 160   μm (bottom) depths. Black circles indicate the depths of PACS and MIPS24   μm surveys for some of the most important cosmological fields, i.e. EGS, Lockman and COSMOS. We note that the region where the constant fraction isocontours run horizontally, cannot be constraint by the data. The adopted detection limits (5σ) at 24   μm are 80   μJy for COSMOS and 50   μJy for EGS and Lockman. Similarly the 3σ detection limits at 100   μm are 5.0-, 3.8- and 3.6 mJy (COSMOS, EGS and Lockman), while at 160   μm are 10.2-, 8.6- and 7.5 mJy.

Open with DEXTER
In the text
thumbnail Fig. 15

The fraction of sources expected to be missed by the 24   μm band at all redshifts as a function of the 24 μm vs. 250   μm and the 24   μm vs. to 350   μm depth. Black circles indicate the depths of PACS and MIPS24   μm surveys for some of the most important cosmological fields, i.e. EGS, SWIRE and COSMOS.

Open with DEXTER
In the text