MAMBO 1.2 mm OBSERVATIONS OF LUMINOUS STARBURSTS AT z ∼ 2 IN THE SWIRE FIELDS

After the discovery by IRAS of local Ultraluminous InfraRed Galaxies (ULIRGs, with L_FIR ≳ 10¹² L_☉) and follow-up studies (see reviews by Sanders & Mirabel 1996; Lonsdale et al. 2006), the advent of sensitive submillimeter bolometer arrays heralded a revolution in our understanding of star formation at high redshift. Extragalactic surveys with SCUBA/JCMT (Hughes et al. 1998; Scott et al. 2002; Mortier et al. 2005; Coppin et al. 2006) and the Max Planck Millimeter Bolometer (MAMBO) array (Kreysa et al. 1998) at the Institut de Radioastronomie Millimétrique (IRAM) 30m telescope (Greve et al. 2004; Voss et al. 2006) show that the number of dusty, infrared-luminous galaxies (high-z ULIRGs, commonly "Submillimeter Galaxies" or SMGs) at 2 ≲ z ≲ 3 is several orders of magnitude higher than in the local universe and that these distant SMGs contribute significantly to the universal star formation history (see e.g., Smail et al. 2004). SMGs appear to be "maximal starbursts" that form stars at the maximum global rate until a significant fraction of their original large gas mass is converted into a massive stellar system. Numerous follow-up studies indicate that SMGs are the most likely objects to trace the major episodes of star formation leading to the production of massive (>2L*) ellipticals (Greve et al. 2005; Tacconi et al. 2006). This picture is consistent with the claim that evolved massive ellipticals have formed the bulk of their stars in short (< 2 Gyr), intense bursts (Dunlop et al. 1996; Blakeslee et al. 1999). The evolution of SMGs into massive elliptical galaxies is often attributed to a negative feedback mechanism induced by Active Galactic Nuclei (AGN) activity (Sanders et al. 1988). It is speculated that as the super-massive black hole (SMBH) in these systems reaches a certain mass and luminosity, the AGN will heat the surrounding gas, and thus halt any further star-forming activity and SMBH growth. This scenario is supported by deep X-ray observations of SMGs that show that some SMGs contain fast growing SMBHs (Alexander et al. 2005a; Borys et al. 2005).

In this context, identifying a large sample of ULIRGs at the epoch when these processes were taking place, i.e., z ≳ 2, is particularly important to understand the formation of the most massive galaxies, investigate the contribution of SMGs to the star formation history of the universe, and the presence and role of AGN activity. These goals have up to now been stymied by the limitations in mapping speed and sensitivity of the current submillimeter bolometers. The total number of known SMGs from all blank-field surveys with any instrument is less than five hundred. Their number and luminosity function are known from MAMBO/SCUBA counts, but remain uncertain at both low and high luminosities. Moreover, in order to quantify the relative importance of BH growth, stellar mass assembly, and the negative feedback induced into the host galaxies by these processes, it is necessary to separate any starburst and AGN power in these systems. Such an analysis requires multiwavelength coverage and good characterization of the spectral energy distributions (SEDs) and spectra. For these reasons, numerous efforts are being carried out to collect large samples of ULIRGs at z ≃ 2 with well sampled SEDs and spectra.

With the advent of wide area Spitzer surveys, new breakthroughs in the study of SMGs are now possible as the statistics of the high redshift ULIRG population increase and the power sources in SMGs are quantified. In particular, the Spitzer Wide Area Infrared Extragalactic Survey (SWIRE¹⁷; Lonsdale et al. 2003), the largest (∼50 deg²) of the Spitzer suite of extragalactic surveys, with its sensitivity is able to detect many sources found in blank-field submillimeter/millimeter surveys. The SWIRE 5σ limits are 4.2, 7.5, 46, 47 μJy in the 4 IRAC (Fazio et al. 2004) bands at 3.6, 4.5, 5.8, and 8.0 μm, respectively, and 0.21, 30, and 180 mJy in the 3 MIPS (Rieke et al. 2004) bands at 24, 70, and 160 μm (Surace et al. 2005)¹⁸. Many previously known SMGs lie in the SWIRE fields and a large fraction of them are detected in 2 or more IR bands (see statistics below and Clements et al. 2008).

In order to estimate what fraction of SMGs would be detected by Spitzer to the SWIRE sensitivities, we gathered from the literature a sample of 90 SMGs with published Spitzer data, or that lie in the SWIRE fields (Greve et al. 2004; Ivison et al. 2004; Egami et al. 2004; Frayer et al. 2004a, 2004b; Chapman et al. 2004; Borys et al. 2005; Pope et al. 2006). This sample is not meant to be complete or exhaustive, but it includes a large number of objects that are representative of submillimeter selected SMGs (`classical SMGs' hereafter).

From an analysis of the SMGs with f_1.2 mm > 2.5 mJy or S_850 μm > 7 mJy (22 `bright SMGs'), we expect the large majority to be detected in at least 2 SWIRE bands. Ten (45%) of these 22 `bright SMGs' are detected above the SWIRE 24 μm ∼5σ–6σ flux density limit (∼250 μJy), 7 of them (32%) are brighter than 400 μJy, 20 (91%) are detected at both 3.6 and 4.5 μm above the SWIRE 5σ limits, 18% are detected at 5.8 μm and 27% at 8.0 μm¹⁹.

The density of bright (f_1.2 mm > 2.5 mJy or S_850 μm > 7 mJy) SMGs is ∼230 per sq. deg. (Voss et al. 2006; Coppin et al. 2006), which means that in ∼50 sq. deg. SWIRE would detect with IRAC about ∼10,000 sources that are typical of those found in submillimeter surveys, and ∼5,000 at 24 μm (>3,000 brighter than 400 μJy), which is 10 times larger than all existing MAMBO-SCUBA surveys put together. Appreciable SWIRE detectability rates are also found for the submillimeter fainter SMGs (3.2 < S_850 μm < 7 mJy): 36% detectable at 24 μm (14% brighter than 400 μJy); 86% at 3.6 and 4.5 μm, 5% at 5.8 μm and 9% at 8 μm, so many thousand more submillimeter fainter SMGs are also to be found among SWIRE sources, including many 24 μm-detected sources. Moreover, Spitzer offers the opportunity to investigate possible selection biases inherent to submillimeter surveys in favor of objects dominated by cool dust.

The identification of high-z ULIRGs among the large SWIRE population of galaxies is not trivial, however, because they are faint in the optical and their optical–mid-IR SEDs can be similar to those of more quiescent star forming galaxies. Moreover, only long wavelength data (far-IR and mm/submillimeter) and reliable redshifts can confirm the extreme luminosities of the ULIRG candidates, but only 0.4% of the SWIRE population is detected at λ⩾ 70 μm, and most of these ULIRG candidates are at z < 1 or too faint in the optical to be followed-up spectroscopically.

X-ray, radio, IR and CO studies (Alexander et al. 2005a; Chapman et al. 2005; Greve et al. 2005; Valiante et al. 2007; Menéndez-Delmestre et al. 2007) indicate that the far-IR (FIR) emission of the bulk of SMGs is dominated by starbursts, but a large fraction of them also host weak AGNs. Spitzer holds a key to this question because the diagnostic provided by the MIR broadband SEDs and IRS (Houck et al. 2004) spectra can discriminate warm AGN-dominated emission. Indeed, using IRS spectra, Pope et al. (2008) constrain the AGN contribution to the MIR emission of SMGs to <30%. However, Spitzer alone cannot determine the FIR luminosity or dust temperature of these systems because most of them are not detected at 70 and 160 μm.

Therefore, millimeter or submillimeter observations are needed to identify these objects and directly measure the long wavelength emission and estimate their FIR luminosities. The combination of SWIRE and MAMBO offers the possibility to explore the role of ULIRGs at z ∼ 2 and determine their power sources.

We have thus undertaken a systematic observing program with MAMBO at 1.2 mm to search for luminous ULIRGs among SWIRE sources. The results from the first set of observations from this program are presented here.

This work is organized as follows. Section 3 describes the selection of the best candidates from the four northern SWIRE fields. Their observations with MAMBO are presented in Section 4. The spectral energy distributions (SEDs) of all sources and photometric redshift estimates are presented in Section 5. Millimeter and 24 μm fluxes, broadband SEDs, and colors of the selected sources are compared with those of submillimeter selected galaxies in Section 6. FIR luminosities and star formation rates (SFRs) are estimated by fitting greybody models with dust temperatures typically measured in SMGs, and presented in Section 7. Since the majority of the sources are not detected with MIPS at 70 and 160 μm, we measure average values from stacked MIPS images. The stacked FIR fluxes at 70 and 160 μm, combined with the average 24 μm and 1.2 mm fluxes, are used to characterize the FIR emission of these sources, and to compare it with the FIR emission of SMGs in Section 7. Stellar masses, near-IR (NIR) luminosities and star formation timescales are discussed in Section 8. The presence of AGN activity is investigated using different tracers, MIR spectra, X-ray, and radio observations in Section 9. The possible nature of the selected sources and their relation with classical SMGs are discussed in Section 10. Finally, our results are summarized in Section 11. Most of the analysis described here is applied to the subsample of sources from the Lockman Hole field. We only report the observations and resulting data on the sub-samples from the other fields. This choice is due to the fact that the Lockman Hole sub-sample is larger and better observed than those in the other fields. The sub-samples from the other fields do not allow us to carry out an accurate statistical analysis. We adopt a flat cosmology with H₀ = 71 km s⁻¹ Mpc⁻¹, Ω_M = 0.27 and Ω_Λ = 0.73 (Spergel et al. 2003).

The most sensitive of the Spitzer MIPS bands for identification of distant ULIRGs is the 24 μm band. Therefore, we have searched for extreme ULIRGs among SWIRE sources with f₂₄ > 250 μJy, corresponding to ≳5σ. Most systems above this f₂₄ limit and z > 1 are expected to be ULIRGs. In addition distant ULIRGs will be faint at optical wavelengths, therefore we limited our overall z > 1 ULIRG samples to sources with r' > 23. This is a brighter cut than adopted in previous works targeting extreme 24 μm sources (e.g., Houck et al. 2005; Yan et al. 2005). For our purpose, we prefer to adopt a brighter optical limit because a fainter cut can reject ULIRGs with colors like some local well known examples, e.g., an Arp 220-like system (z = 0.018 and L(IR) = 1.45 × 10¹² L_☉) at z = 1 with an observed f₂₄ = 250 μJy would have a luminosity of L_IR = 10^12.6 L_☉ and an observed magnitude r' = 23.4, while an Arp 220-like system with L_IR = 10¹³ L_☉ at z = 1 would have an observed r' = 22.4.

The selection based on bright 24 μm and faint r'-band fluxes yields consequently sources with large f_24 μm/f_r ratios, typically ≳100. Bright 24 μm Spitzer samples with large IR/optical flux ratios (i.e., f_24 μm/f_R > 500) tend to be dominated by AGN light in the MIR (Houck et al. 2005; Yan et al. 2005; Sajina et al. 2007; Brand et al. 2006; Dey et al. 2008). Moreover MAMBO observations of these have shown that they are weak at 1.2 mm compared to submillimeter selected SMGs (Lutz et al. 2005). An additional selection criterion was thus applied to disfavor AGN-dominated and favor star-forming-dominated systems. The additional selection criterion is based on the evidence in the IRAC bands for the ∼ 1.6 μm rest-frame Planck spectrum peak from evolved stars. Note that IRAC was in part designed for photometric selection of galaxies at z = 1.5–3 displaying this feature (Simpson & Eisenhardt 1999; Sawicki 2002). Such a peak is not observed in templates of local AGN-dominated sources such as Mrk 231, IRAS 19254−7245 (Berta et al. 2003), or in type 1 QSOs (Elvis et al. 1994). On the other hand, the SED of sources displaying such a peak can be well fitted by local star formation-dominated templates, such as M 82, and Arp 220 (Silva et al. 1998).

Furthermore, we have selected sources where this peak lies in the 5.8 μm band in order to favor sources at z = 1.5–3. We refer to such sources as `5.8 μm-peaker' (or `b3' for bump in the third IRAC channel). Formally a source is defined `5.8 μm-peaker' if satisfies the following condition: f_3.6 < f_4.5 < f_5.8 > f_8.0 where f is the flux density in μJy and f_8.0 can also be an upper limit (typically at 5σ).

We have previously demonstrated that this selection method is effective in finding starburst-dominated objects at z = 1.5–3 (Weedman et al. 2006; Berta et al. 2007b; Farrah et al. 2008). IRS MIR spectroscopic observations of a similar sample of 32 SWIRE sources displaying a rest-frame NIR bump peaking at 4.5 μm demonstrate that they are star-forming dominated (Farrah et al. 2008). Most of their spectra exhibit prominent PAH emission features consistent with a starburst origin of their MIR emission. Similarly, Weedman et al. (2006) find that all SWIRE sources with a NIR bump at 5.8 μm, display PAH emission features in their IRS spectra. Moreover, eight of the sources selected here were observed in the MIR by the IRS instrument on Spitzer (six in the LH Weedman et al. 2006, and two in the EN2 field, C. J. Lonsdale 2009, in preparation), and their spectra display strong PAH features as expected from starburst-dominated emission. The IRS observations yield spectroscopic redshifts that, in all but one case, confirm the `5.8 μm-peakers' photometric redshift selection method. The agreement between photometric and IRS redshifts is discussed further in Section 5 (see also Farrah et al. 2008). Based on these considerations and studies, we infer that the combination of a strong 24 μm flux and a peak in the IRAC bands is a good indication of the presence of PAH emission at z ∼ 2, and thus of a significant, and likely dominant, starburst contribution to the NIR and MIR emission. Although in some cases an AGN component in the MIR can be present, its contribution is never dominant compared to the starburst component (Berta et al. 2007a).

In order to further increase the chance of selecting z ∼ 2 star-forming galaxies, we derived photometric redshifts for all source candidates (for a detailed description of the procedure see Section 5).

We have selected in this manner 1023 `5.8 μm-peaker' ULIRG candidates with f<sub>24</sub> > 250 μJy, r' > 23, and photometric redshift in the range 1.5–3, from the SWIRE's largest field (11 deg<sup>2</sup>), the Lockman Hole (LH). We restricted the search to the region in the LH with available optical imaging (∼9.5 deg<sup>2</sup>). At similar redshifts and for fixed SED shape the most luminous objects will be the brightest in flux. Therefore, to cull from this large sample the most luminous candidates we selected the brightest 24 μm sources, i.e., f<sub>24</sub> > 400 μJy, among all `5.8 μm-peakers', and with ⩾5σ detections in the first 3 IRAC bands. There are 520 sources with f₂₄ > 400 μJy, corresponding to 55 sources per square degree. We have also required unambiguous (unique or not detected counterpart) source identification at all optical-MIPS wavelengths in the images, so that source confusion would not affect the SED interpretation. Our selection is likely to strongly favor sources with strong 7.7 μm PAH features in 24 μm band in the z = 2 range.

Finally we made estimates of the expected 1.2 mm flux densities by fitting templates of various local starbursts and ULIRGs to the optical (Ugriz) and infrared (3.6–24 μm) bands. Since most of the candidates are not detected at 70 or 160 μm, their FIR through mm spectral shape is unconstrained. The MAMBO flux predictions therefore rest on the adoption of a long wavelength SED template. To select appropriate templates we build the rest-frame SEDs of ∼30 SMGs with known redshifts and available IRAC and 24 μm fluxes, and normalized them at rest-frame 1.6 μm. We then selected templates of local starburst-dominated ULIRGs which pass approximately through the midst of the distribution of 850 μm and 1.2 mm points for these objects, as shown in Figure 1. Most of the SMGs have FIR SEDs similar to Arp 220 or IRAS 22491−1808 (I22491 hereafter). The main difference between the Arp 220 and I22491 templates, a starburst dominated and a composite starburst+AGN system, respectively, resides at optical and MIR wavelengths. However, since with some extinction the optical emission of I22491 resembles that of Arp 220, and in the MIR we do not have enough spectral coverage to favor one template with respect to the other, we decided to fit each candidate source with an Arp 220 template. In addition we decided to also fit a template with a much lower FIR/NIR flux ratio than Arp 220, to obtain minimum estimates of the predicted 1.2 mm flux for each source. For this we chose NGC 6090, a local starburst with a similar SED shape to the well known starburst galaxy M 82. Examples of these two fits are shown for 3 sources in Figure 2, where it can be seen that the two templates produce similar photometric redshifts from the IRAC data, but the predictions for the bolometric luminosity, and for the 1.2 mm flux density differ by a factor of about 10.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To select the best candidates for observations with MAMBO, we averaged the predicted 1.2 mm flux density for each source from both templates and ordered the sources according to decreasing predicted averaged 1.2 mm flux. While this procedure may not predict the 1.2 mm flux of an individual source accurately, because the actual long wavelength SED is unconstrained, it does provide a robust way to prioritize them by decreasing the expected 1.2 mm flux under the assumption that they have similar SED shapes, and taking into account the sources redshift.

The same selection procedure was applied to other three SWIRE fields, XMM-LSS (XMM), ELAIS-N1 (EN1) & ELAIS-N2 (EN2). A description of the fields sizes and multiwavelength coverage can be found in Surace et al. (2005). The top 61 sources from the 4 SWIRE fields, (LH, XMM, EN1 & EN2), were observed with MAMBO at 1.2 mm. The largest number of sources was selected in the LH field, 29 sources, compared to 17 in EN1, 9 in EN2, and 6 in XMM. The 29 LH sources correspond to 5.6% of the entire sample of 520 f₂₄ > 400 μJy `5.8 μm peakers'. The predicted 1.2 mm fluxes (averaged on all templates) range from 1.4 to 25 mJy. The list of selected sources, coordinates and optical–infrared–mm fluxes are reported in Tables 1 and 2.

Note that because of the uncertainties on the IRAC fluxes, especially at 5.8 and 8.0 μm, which are less sensitive bands than those at shorter wavelengths, a `5.8 μm-peaker' (z ∼ 2.2–3.2) can become `4.5 μm' or `8.0 μm-peakers' (`b2', or `b4') in different versions of the data processing. Indeed, some of the sources selected here, that were classified `5.8 μm-peakers' based on a previous version of the SWIRE catalog used during the selection process, show a peak at shorter (`b2'; z∼1.5–2.2) or longer wavelengths (`b4'; z⩾3.2) or a flat IRAC SED (`fl') in the latter SWIRE catalogs that is used in this work. This change is observed in 14 out of 61 selected objects (see last column of Table 1).

Observations were spread over the pool observing sessions at the IRAM 30m telescope in fall/winter 2005/2006, using the 117 element version of the MAMBO array (Kreysa et al. 1998) operating at a wavelength of 1.2 mm (250 GHz). On the first run in late October - early November 2005 we observed most of the Lockman sample (29 sources), and the 6 XMM sources. During the winter run, mainly early March 2006, the observations of the Lockman sample were completed and exploratory observations of the 17 EN1 and 9 EN2 sources were carried out.

We used the standard on-off photometry observing mode, chopping between the target and sky at 2 Hz, and nodding the telescope every 10 or 20 s. On-off observations were typically obtained in blocks of six scans of 16 or 20 10 s subscans each, and repeated in later observing nights. The atmospheric transmission was intermediate with τ(1.2 mm) at zenith between 0.1 and 0.4. The absolute flux calibration was established by observations of Mars and Uranus, resulting in a flux calibration uncertainty of about 20%.

On average, the noise of the channel used for point-source observations was about 35–40 mJy/$\sqrt{t}$/beam, where t is the exposure time in seconds, consistent with the MAMBO time estimator for winter conditions. In order to identify a maximum number of strong sources with flux density ∼3–5 mJy within the allocated time (57 hr, i.e., ∼38 hr of integration time), we adopted a two-step procedure: a first exploratory phase generally aiming at an rms ∼1.0–1.2 mJy; then longer integrations on sources showing hints of detection in order to reach rms ≈ 0.6–0.7 mJy and achieve 3–5σ detections of sources stronger than ∼2–3 mJy. However, we were able to apply such a complete procedure only to the LH sample (29 sources), which had the best selection and was the only one observable in good conditions (during night or early morning) both in fall and winter runs. The observations of the sources of the three other fields (EN1 and EN2, XMM; 32 sources) were more or less limited to the exploratory phase.

A major change occurred in the 30 m telescope control system in 2005. MAMBO pool observations in our October to early November session were carried out with a well known system which had been in use for years. A new system, just installed before the winter pool session, brought some problems such as stronger accelerations of the telescope during nodding with possible effects on tracking accuracy and temperature stability of the helium cryostat in some circumstances; anomalous noise in some MAMBO channels for a few cases; etc. These problems induced some delays and loss of time, but they were practically all well under control for most of our observations in 2006 February–March. The data were reduced with standard procedures in the MOPSIC package developed by R. Zylka. Great care was taken to check the quality of the data obtained with the new control system, and we discarded the data from about ten scans displaying obvious problems such as anomalous noise in some channels, as well as ∼3% of the data taken in periods when we are not sure that the problems with the new system were well under control.

Table 1 lists the resulting 1.2 mm flux densities and their statistical uncertainties. In the more completely studied Lockman sample (29 sources, Table 1), more than 40% of the sources (12) show a signal at >2.5σ, and fluxes ≳2 mJy (corresponding to ∼5 mJy at 850 μm for z ∼ 2, see Greve et al. 2004). Eight of them have 1.2 mm fluxes in the 3–5 mJy range, with signal-to-noise ratio (S/N) ≳ 4–5, which guarantees the robustness of the detections. However, it is possible that the flux densities of the detected sources are slightly overestimated by the Malmquist bias, by an amount that probably does not exceed 10% on average. The average flux density of the 29 LH sources is 1.88 ± 0.28 mJy (see Table 3), corresponding to ∼5 mJy at 850 μm at z ∼ 2).

As quoted, the remaining 32 sources, from EN1 & EN2 and XMM, were not observed as completely. However, the average strength of the sources in the other fields is significantly weaker (1.14 ± 0.20 mJy) than in the Lockman sample. The fraction of sources showing at least a 2.5σ signal is only 25% (8 sources), there is practically no source stronger than 3 mJy, and the average of the 23 other undetected sources is only 0.60 ± 0.17 mJy. Possible reasons for such a behavior are either a pointing problem caused by the system changes implemented in 2005, or a bias introduced by the slight different selection criteria (see Section 6.1).

Table 3 summarizes the information about the number of sources showing a signal at various levels and the average values for two sub-samples and the totality of the 61 sources.

In order to classify the spectral energy distributions (SEDs) and estimate photometric redshifts of the sources in the sample, optical and IR data (⩽24 μm) are combined and fitted with various galaxy templates. The SEDs are fitted using the ${\sf Hyper{-}z}$ code (Bolzonella et al. 2000). ${\sf Hyper{-}z}$ finds the best-fit by minimizing the χ² derived from the comparison of the observed SED and expected SEDs at various redshifts, and using a template library and the same photometric system. The effects of dust extinction are taken into account by reddening the reference templates according to a selected reddening law. We use the prescription for extinction measured in high-redshift starbursts by Calzetti et al. (2000). We limit the additional extinction A_V to be less than 2.0 mag and include templates of highly extinguished objects. We used two template libraries, one with 105 templates of starburst galaxies (Chary & Elbaz 2001), and one with 8 templates including 2 late-spirals, 5 starbursts, and 1 composite (starburst+AGN) template covering the wavelength range between 1000 Å and 1000 μm. The starburst templates correspond to the SEDs of NGC 6090, NGC 6240, Arp 220, IRAS 20551−4250, and I22491 (Silva et al. 1998; Berta 2005). The composite (AGN+SB) template corresponds to the SED of the Seyfert 2 galaxy IRAS 19254−7245 South (I19254; Berta et al. 2003). A description of the library and of the method can be found in Polletta et al. (2007).

We fit the SEDs using the data up to 24 μm, including upper limits. Although the optical upper limits place some constraints on the acceptable fits, for these optically faint objects the photometric redshifts are based primarily on the IRAC bands. The photometric redshifts are therefore limited in accuracy by (1) the small number of available photometric bands (3–4 IRAC bands, MIPS[24], and up to a maximum of 5 optical bands among Ugriz), and (2) the fact that the main spectral feature available for fitting (the peak in the stellar spectrum) is broad and detected through broad filters. Therefore our photometric redshifts are expected to have uncertainties of ±0.5 in redshift or higher, even for the highest SNR sources. In order to illustrate the range of acceptable solutions and photometric redshifts, we report multiple solutions if they give similar χ². The fits were also done without including the 24 μm measurements. Since the comparison with spectroscopic redshifts was slightly better when those data were included, we decided to adopt the results obtained using also the 24 μm data. The inclusion of 24 μm data in the fitting procedure can yield better results, if templates including dust emission, like the libraries used here, are employed (see e.g., Polletta et al. 2007).

The reliability and accuracy of the photometric redshifts can be estimated using the available spectroscopic data. Spectroscopic redshifts from the IRS are available for 8 sources and are listed in Table 1 (Weedman et al. 2006, C. J. Lonsdale et al. 2009, in preparation). We define the fractional error Δz, the systematic mean error $\overline{\Delta z}$, the 1σ dispersion σ_z, and the rate of catastrophic outliers, defined as the fraction of sources with |Δz|> 0.2. Δz is defined as:

$$\begin{equation} \Delta z = \left(\frac{z_{\rm phot}-z_{\rm spec}}{1+z_{\rm spec}}\right) \end{equation} \tag{ 1 }$$

and

$$\begin{equation} \sigma _z^2 = \sum \left(\frac{z_{\rm phot}-z_{\rm spec}}{1+z_{\rm spec}}\right)^2\bigg/N \end{equation} \tag{ 2 }$$

with N being the number of sources with spectroscopic redshifts. The systematic mean error, $\overline{\Delta z}$, is 0.03. The rms of $\overline{\Delta z}$, σ_z is 0.08, and there are no outliers with |Δz|> 0.2. The comparison between the spectroscopic and final photometric redshifts is shown in Figure 3 for the eight sources. There is a good agreement for five sources, and for three sources there is a difference ≳0.05 in log(1 + z). In order to better asses the reliability of our estimates, we applied the same technique to a sample of 24 μm detected SMGs from the literature with known spectroscopic redshift and with at least three detections in the IRAC bands consistent with the presence of a peak at either 4.5 μm or 5.8 μm. We found 15 sources that satisfy these criteria out of the 90 SMGs for which we collected data from the literature (see Section 1). Note that most of the sources from the literature do not benefit from observations from as many bands as the SWIRE-MAMBO sample. After combining the SWIRE-MAMBO and the literature sample, $\overline{\Delta z}=0.03$, σ_z = 0.19, and we find two outliers (|Δz|> 0.2), corresponding to ∼9% of the sample. The results obtained for the literature sample confirm the accuracy of our estimates on the SWIRE-MAMBO sample.

Zoom In Zoom Out Reset image size

SED fits for all the SWIRE sources in the sample are provided in the Appendix and the best-fit templates and photometric redshifts are listed in Table 4.

The redshift distribution of the sample, based on the photometric redshifts (solid line) and on the spectroscopic redshifts (dotted line) when available, is shown in Figure 4. The distribution peaks at z ≃ 2.0, which corresponds to about the minimum redshift for a 5.8 μm peaker. The z peak of our sources is slightly lower than the redshift distribution of submillimeter-selected sources (Frayer et al. 2004b; Egami et al. 2004; Borys et al. 2005; Chapman et al. 2005; Greve et al. 2004). Note that the difference could be even larger because of the possible overestimate of z_phot in case, for example, of a shift of the peak of the stellar bump due to an AGN contribution (e.g., a `b2' can appear as a `b3' and its redshift be overestimated by Δz = 1; see e.g., Berta et al. 2007b; Daddi et al. 2007).

Zoom In Zoom Out Reset image size

6.1. Millimeter and 24 μm Fluxes

The distribution of observed 1.2 mm flux densities (separated in sources with f_1.2 mm ⩾ 2σ and < 2σ) is illustrated in Figure 5. The fraction of > 2σ sources, 41% (24% at 4σ), and the large average 1.2 mm flux observed in our samples (1.5 ± 0.2 mJy), and especially in the more thoroughly observed LH sample (1.9 ± 0.3 mJy), show that our criteria are quite successful in selecting submillimeter-bright galaxies from MIR samples.

Zoom In Zoom Out Reset image size

The XMM, EN1 & EN2 sample show systematically lower fluxes compared to the LH sample. Although the LH sample was more thoroughly observed than the other samples, with typical exposure times >1.5 hr, compared to ∼40 minutes for the other sources, the origin of the different detection rate and average fluxes is more likely due to intrinsic properties of the sources or of the fields, e.g., cosmic variance, than to the observational parameters. The most plausible explanation is that our selection introduced different biases in the selected samples possibly due to the different optical coverage or to the analysis carried out on the SEDs and on the images during the source selection process. The latter was indeed more extensive in the LH field.

In Figure 5, we also show the distribution of 1.2 mm fluxes of the 90 SMGs selected from the literature (see Section 2). In case of lack of flux measurement at 1.2 mm for the SMGs from the literature, we estimate it from the flux at 850 μm by applying the following equation:

$$\begin{equation} {\rm Log}(f_{1.2\,{\rm mm}}) = -0.36+0.98\times {\rm Log}(f_{850\,\mu {\rm m}}) \end{equation} \tag{ 3 }$$

where f_1.2 mm and f_850 μm are the flux densities at 1200 and 850 μm, respectively. This relationship is derived from a sample of 21 SMGs for which measurements at both 850 and 1200 μm are available (Greve et al. 2004).

Based on Figure 7 in Voss et al. (2006) and the 850 μm source counts from SHADES (Coppin et al. 2006), the expected number of SMGs with f_1.2 mm > 5 mJy, 4 mJy, and 3 mJy in the 9 deg² LH field is ∼250, 800, and 2,000, respectively. On the other hand, we have identified only two, three, and four sources brighter than 5 mJy, 4 mJy, and 3 mJy, respectively out of 29 24 μm bright sources in the 9 deg² LH field. Accounting for the fact that we have sampled with MAMBO ∼5.6% of the LH 5.8 μm peakers which are brighter than 400 μJy at 24 μm, we find that we are detecting about 15% of the predicted number of SMGs brighter than 5 mJy. The small number of mm-bright sources in our sample and in the sample in Lutz et al. (2005) suggests that very luminous millimeter sources are not common among sources with bright 24 μm fluxes (f_24 μm > 400 μJy).

In Figure 6, we show the distribution of 24 μm fluxes of the SWIRE-MAMBO sample and of the SMGs from the literature. All SWIRE-MAMBO sources have 24 μm greater than 400 μJy as imposed by the selection criteria. The 24 μm fluxes of our sample range from 0.4 to 1.5 mJy, with a median value of 0.76 mJy and a mean value of 1.09 ± 1.06 mJy. The SWIRE-MAMBO sources with f_1.2 mm < 2σ show a similar range of 24 μm fluxes, but fewer of the faintest 24 μm sources are detected by MAMBO. Figure 6 clearly illustrates that the majority of SMGs are characterized by 24 μm fluxes smaller than 400 μJy. More specifically, out of the 90 SMGS from the literature, 20 (22%) have 24 μm fluxes greater than 400 μJy. Therefore, about 78% of classical SMGs would be missed by our selection criteria. Indeed, more than half of SMGs at z ∼ 2 are not detected at the SWIRE sensitivity at 24 μm, ∼250 mJy (and even at 5.8 or 8.0 μm, see Section 1). The lowest 24 μm flux in the SMGs reaches 21 μJy, but there are 16 sources that are not detected, four in GOODS (f_24 μm< 26–50 μJy) and the others in the SWIRE fields (f_24 μm< 250 μJy). Since our sources are at similar redshifts and show similar mm fluxes to SMGs, they either represent a subset of 24 μm bright SMGs (∼20%) or a different class of SMGs with enhanced MIR emission. This question will be further investigated in the following sections.

Zoom In Zoom Out Reset image size

6.2. Comparison with Other Spitzer Samples

6.3. Spectral Energy Distributions

The spectral energy distributions (SEDs) of our sources are compared with those of submillimeter selected SMGs in Figure 7. Only sources with spectroscopic redshifts are shown in the right panel. All SEDs are shown in the rest-frame and normalized at 1.6 μm in the rest frame. The SWIRE sources are predominantly high in the 24 μm band at the location of the rest-frame 7.7 μm PAH emission feature, and low in the rest-frame FIR compared to the SMGs from the literature. The FIR emission of the SWIRE sources is in between that expected for the typical starburst galaxies as M 82, and NGC 6090, and a ULIRG like Arp 220, after normalizing them at the 1.6 μm in the rest-frame.

Zoom In Zoom Out Reset image size

Figure 7 shows clearly the selection bias that is present in our sample: by selecting the 24 μm-brightest sources with a peak at 5.8 μm, we have strongly favored sources at z ∼ 2 with large NIR and MIR luminosities. The classical SMG sample is naturally more biased towards the most mm-bright systems. The location of our sources in Figure 7 with respect to various template tracks confirms that their SEDs are intermediate between M 82 and Arp 220.

6.4. Colors and Nature of the Sources

We compare the 1.2 mm/24 μm flux ratio of the SWIRE and literature SMG samples in Figure 8. The SWIRE sources show on average lower flux ratios than the literature sources at the same redshift. The difference in the 1.2 mm/24 μm can not be simply explained by a redshift difference but is likely due to an excess in the MIR (i.e., 24 μm) with respect to the FIR in the SWIRE sample compared to the literature sample. Such an excess, also visible in the SEDs (see Figure 7), might be due to an AGN contribution in the MIR, or to enhanced PAH emission, or to a deficiency at mm wavelengths due to lower SFRs or higher temperatures. Indeed objects characterized by warmer dust temperatures are expected to show lower flux ratios (e.g., Chapman et al. 2003; Lewis et al. 2005).

Zoom In Zoom Out Reset image size

In Figure 8, we also show for comparison the 1.2 mm/24 μm flux ratios measured in the sample of composite sources from Lutz et al. (2005), and in high-z radio galaxies (HzRG) detected at both 850 μm and 24 μm (Seymour et al. 2007). The 1.2 mm flux is derived from the 850 μm by applying equation 3. These two samples, from Lutz et al. (2005), and Seymour et al. (2007), show even smaller 1.2 mm over 24 μm flux ratios. In this case, the most plausible explanation for such small ratios is the contribution from AGN-heated hot dust emission to the MIR since both samples contain mainly AGNs. Since our sources show intermediate flux ratios in between those seen in the literature SMGs and HzRGs, it is quite possible that an AGN contributes to their MIR, more than typically observed in SMGs and less than observed in HzRGs.

To better investigate the origin of such an excess we consider in more detail the SMGs that fall in the same region covered by the SWIRE sources, and analyze the MIR spectra when available. From the literature subsample with available spectroscopic redshifts and available 24 μm flux (65 sources), we find three sources that fall in the same region in the f_1.2 mm/f_24 μm vs z diagram as our sources, J123635.59+621424.1 (f_1.2 mm/f_24 μm = 1.35) at z = 2.005 from the GOODS North field (Borys et al. 2005), CSMM J163650+4057 (f_1.2 mm/f_24 μm = 3.07) at z = 2.384 from the ELAIS-N2 field (Greve et al. 2004), and SMM J105238.19+571651.1 (f_1.2 mm/f_24 μm = 3.34) at z = 1.852 (Chapman et al. 2005). The first two sources are AGNs, their MIR SEDs are consistent with a power-law model, their optical spectra show broad emission lines, and one source is also a luminous X-ray source. The latter source is instead classified as starburst galaxy based on its optical spectrum, and it also shows the highest 1.2 mm/24 μm flux ratio of the three selected sources. This analysis thus reinforces the hypothesis that an AGN component might be responsible for the lower flux ratios observed in the SWIRE sample, but this is based on only 2 sources and their SEDs are very different from those of our sources as a dominating AGN component is clearly seen in the MIR.

The best and probably unique way to constrain the presence and the contribution of an AGN component in the MIR is to measure the equivalent width of the PAH features. A warm dust continuum due to AGN-heated hot dust will decrease their equivalent width (see e.g., Clavel et al. 2000; Valiante et al. 2007; Menéndez-Delmestre et al. 2007; Sajina et al. 2007; Desai et al. 2007). This measurement can be performed using MIR spectra.

Based on the IRS spectra of similar samples and the few available here, a significant AGN contribution is disfavored for the majority of our sample. MIR spectra from IRS are available only for eight sources in our SWIRE-MAMBO sample (Weedman et al. 2006, C. J. Lonsdale et al. 2009, in preparation). Altogether, six out of these eight sources show PAH dominated spectra and weak or absent continuum, implying large equivalent widths and negligible or absent AGN contribution. They are, in approximate order of PAH strength, LH-07, LH-06, LH-26, EN2-04, LH-12, and LH-05. The only exceptions are two sources (LH-23/B6, and EN2-03) that show, in addition to PAH emission, a warm dust continuum that might be associated with an AGN component. We do not find any trend between the presence of continuum, the strength of the PAHs, the 1.2 mm flux, and the f_1.2 mm/f_24 μm flux ratio. Therefore, for this analysis we consider the average spectrum of all eight sources, independently of their properties. The average IRS spectrum of all eight SWIRE-MAMBO sources is shown in Figure 9. We also show for comparison the composite spectrum of 12 SMGs from the Hubble Deep Field North (HDFN) (Pope et al. 2008), the IRS spectrum of a starburst, NGC 3079 (Weedman et al. 2005), a power-law model representing emission from warm dust, and the sum of the starburst spectrum and of the warm dust continuum. The average spectrum of the SWIRE-MAMBO sources clearly shows PAH features, implying a strong starburst component, but also a warm dust continuum. The average spectrum is well reproduced by the sum of the starburst spectrum and the warm dust continuum. The continuum is modeled with a power-law model $(f_{\nu }\propto \lambda ^{\alpha _{IR}})$, with spectral index α_IR = 2. We have applied the same analysis to the average spectrum after removing the 2 sources with larger continuum contamination and for the sources with >2σ and <2σ 1.2 mm fluxes and the result is still valid with only small differences that would require a larger sample to better quantify. The contribution of the power-law component to the average spectrum is ∼34% in the 6–12 μm rest-frame wavelength range. Assuming that the 24 μm flux includes a 34% contribution from an AGN on average, the AGN-subtracted f_1.2 μm/f_24 μm flux ratio would be 1.5 times higher. Even after such a correction, our sources remain systematically on the lower bound region of the flux ratio distribution. Note that for consistency we would have to apply a similar correction to the literature SMGs as well. Indeed, the majority of classical SMGs also show both PAH features and a weak warm dust continuum in their MIR spectra (Valiante et al. 2007; Menéndez-Delmestre et al. 2007; Pope et al. 2008). The PAH equivalenth withs of our sources and of classical SMGs are in fact very similar, as shown in the left panel of Figure 9. Thus, the hypothesis that our sources might have an enhanced 24 μm flux due to the contribution from AGN emission, is disfavored. However, our sources are more luminous at 24 μm than SMGs. Their PAHs and MIR continuum are thus, on average, more luminous than in SMGs.

Zoom In Zoom Out Reset image size

In order to estimate FIR luminosities, and star formation rates (SFRs) of the SWIRE-MAMBO sources, their FIR SEDs need to be characterized. Since most of the sources are too faint to be detected at 70 and 160 μm with MIPS at the SWIRE sensitivity (typically 18, and 108 mJy, respectively), we can only use the observed 1.2 mm flux and make reasonable assumptions about their FIR SEDs. We assume that the FIR–mm emission of our sources is similar to that seen in SMGs, and model their FIR–mm emission with a greybody model normalized at the observed 1.2 mm flux, or 3σ upper limit if the flux is <2σ. To constrain the temperature of the FIR–mm emitting dust, it would be necessary to have at least two flux measurements at FIR–mm wavelengths. Since the dust temperature can not be directly estimated for our sources, we fix the temperature to the values typically observed in SMGs after taking into account the dependency of the dust temperature on the redshift. From the sample of SMGs with multiple submillimeter and mm data studied by Kovács et al. (2006), we derive:

$$\begin{equation} T_{\rm dust} = 3.9+9.5\times (1+z) \,\,\,\,\,(K). \end{equation} \tag{ 4 }$$

We use a dust emissivity index β = 1.5 as in Kovács et al. (2006). Higher values of β give slightly larger temperatures (see Figure 3 in Kovács et al. 2006), and larger FIR luminosities. At parity of flux and temperature, the FIR luminosity derived with β = 2 is on average 2.3 times larger than the FIR luminosity derived with β = 1.5. Two submillimeter/mm flux measurements are available only for 4 sources in our sample (A. Kovacs et al. 2009, in preparation). In these four cases, the estimated temperatures are consistent with the measured ones, giving support to our assumption. The dust temperatures range from 23 K to 41 K, with a mean value of 33 ± 4 K. In order to take into account the redshift uncertainty, the greybody fits are performed three times: (1) at the best redshift of the source, z_best; (2) at a lower redshift, z₋ = z_best − 0.12 × (1 + z_best); and (3) at a higher redshift, z₊ = z_best + 0.12 × (1 + z_best). The coefficient 0.12 corresponds to 1.5 times the dispersion found in the comparison between photometric and spectroscopic redshifts (see Section 5). The FIR luminosities are reported in Table 5. The FIR luminosities for the sources with a > 2σ signal at 1.2 mm range from 1 to 6.3 × 10¹² L_☉, confirming that our SWIRE-MAMBO sources are ULIRGs. This is true for the majority of the sources also at z = z₋. A similar luminosity range is obtained for the rest of the sample, but they are likely upper limits to the true FIR luminosity.

The greybody models are always below the 70 μm, and 160 μm upper limits of each source, implying that the assumed temperatures are never overestimated. If the temperature of the fits were increased to come closer to the FIR upper limits, the estimated FIR luminosities would increase.

We carry out an additional analysis to characterize the average FIR SED of our sources. We estimate average fluxes at 70 μm and 160 μm by co-adding all the 70 μm and all the 160 μm images to create a stacked 70 μm image and a stacked 160 μm image. The co-adding technique used in this work follows the method described in Dole et al. (2006). This technique was applied to all MIPS images (24, 70, and 160 μm) and only to the sources in the LH field since the number of observed and detected sources is significantly larger than in the other fields. The stacking of the 24 μm images was carried out to validate the method by comparing the stacking result with the mean flux of the single measurements. In the rest of the analysis, we will use the mean of the single measurements at 24 μm instead of the stacking value. Two sources, LH-18, and LH-28, were excluded from the stacking because the former is located at the edge of the 160 μm image, and the latter is detected at 70 μm. We divided the remaining 27 sources in the Lockman Hole sample in two sub-groups: sources with 1.2 mm fluxes above 2σ (12), and below 2σ (15).

At each of the positions of the sources in each subgroup, we extract a 72''× 72'' subimage from the full-field SWIRE mosaics. The subimages are placed into a stack twice, once with their original orientation, and again with a 90° rotation, to remove instrumental effects such as gradients affecting the stacked flux. However, since the MIPS maps are already filtered following the techniques described in Frayer et al. (2006), the rotation should not give a significant difference. Indeed, a test on the the 70 μm stacks with and without image rotation yields a difference of 1.4% which is negligible compared to the uncertainty. The mean flux image is computed from the stack, and the flux measured in an aperture (with sky subtraction in an annulus included). An uncertainty estimate is derived using a bootstrap technique (Efron & Tibshirani 1993), in which samples of the subimages are drawn with replacement from the image stack, and the mean image and aperture flux measurement are recomputed. The standard deviation of the aperture flux measurements of 200 repetitions of this process yields the uncertainty estimate. We have validated these uncertainty estimates by stacking positions randomly offset by 60 to 80 arcseconds from our source positions; the standard deviation of 500 such stacks is slightly lower than the uncertainties from the bootstrap method. All of the measured stacked fluxes are reported in Table 6. The mean of the single measurements at 24 μm agrees in all three cases with the stacked mean 24 μm flux, within the uncertainties.

The stack of all 27 LH sources results in a detection at both 70 and 160 μm, at ∼3.8σ. The stack of sources > 2σ 1.2 mm fluxes is well (∼4σ) detected at both wavelengths. The stack of the sample with f_1.2 mm < 2σ does not result in a detection at either wavelength. Note that the significance level of these measurements reflects mainly differences in noise level due to small number statistics. However, these measurements provide some constraints on the average FIR flux and spectrum of our sources.

The average FIR SEDs (from 24 to 1200 μm, including the stacked 70 and 160 μm fluxes) of the LH subsample with f_1.2 mm > 2σ is shown in Figure 10. The figure also shows the predicted average flux at 70 and 160 μm assuming the greybody models used to compute the FIR luminosities. The stacked mean fluxes are well above the greybody models predictions (solid black curve in Figure 10). This discrepancy cannot be explained by any of our uncertainty (in redshifts for example). At 160 μm the mean observed flux predicted by the greybody models is 4 times lower than measured in the stacked images. This discrepancy suggests that our sources are more FIR luminous and warmer than estimated by the greybody models that are based on the SMGs temperatures. In order to fit the SWIRE-MAMBO sources average FIR SED, the dust component should be on average 10 K higher than predicted. Indeed, a greybody model that fits the average FIR SED assuming the mean redshift of the LH sub-sample, z = 2.03, has a temperature of 42.5 K (dashed black curve in Figure), instead of 32.7 K as predicted by Equation (4). Note that while the excess at 160 μm can be explained by a temperature effect, at 70 μm (≳20 μm in the rest-frame) a different contribution from small grains is more likely. In this work, we do not investigate the difference at 70 μm, as we focus our analysis on the FIR part of the spectrum.

Zoom In Zoom Out Reset image size

For comparison with the classical SMGs, we consider the FIR properties of >30 SMGs from the HDFN sample (Pope et al. 2006, 2008). The mean FIR SED of this sub-sample is also shown in Figure 10. The 850 μm fluxes of these SMGs range from 1.7 to 20.3 mJy (corresponding to 0.7–8.3 mJy at 1.2 mm), and show a mean value of 5.7 mJy (corresponding to 2.4 mJy at 1.2 mm), consistent with our SWIRE-MAMBO sources with f_1.2 mm > 2σ. Their average 70 and 160 μm fluxes, derived from the stacking of a sub-sample of 26 sources by Huynh et al. (2007), are significantly lower than those measured in the SWIRE sample. Such a difference suggests that the HDFN SMGs are, on average, characterized by cooler dust temperatures and lower FIR luminosities. Note that the sub-sample used by Huynh et al. (2007) does not include 2 sources that are detected in the MIPS images. However, even including those sources the average FIR SED of classical SMGs would remain well below that of our sources.

In summary, the analysis of the average FIR SED of the SWIRE-MAMBO sources with f_1.2 mm > 2σ indicates that their FIR luminosities are higher than estimated using the greybody models and that the FIR emitting dust is at higher temperatures than assumed. On average, our sources are thus more luminous and warmer than classical SMGs, given that the two samples have similar redshift distributions.

thus more luminous and warmer than classical SMGs.

Our IRAC-based selection corresponds to a rest-frame NIR selection, therefore, in absence of a significant AGN contribution at these wavelengths, we are directly sampling the stellar component. Moreover a NIR selection is minimally affected by dust extinction. Therefore it is possible to derive accurate stellar masses, by means of spectrophotometric synthesis. We focus this analysis on the better observed and larger LH sample.

We use the code Sim-Phot-Spec (SPS), developed in Padova (Berta et al. 2004; Poggianti et al. 2001) to estimate the stellar mass in our sources. The code performs mixed stellar population (MSP) spectrophotometric synthesis. The adopted synthesis stellar population (SSP) library is based on the Padova evolutionary sequences of stellar models (Fagotto et al. 1994a, 1994b) and isochrones (Bertelli et al. 1994), and was computed by assuming a solar metalicity and a Salpeter (1955) initial mass function (IMF) between 0.15 and 120 M_☉. The SSP spectra are built using the Pickles (1998) stellar spectral library, and extended to the NIR with the Kurucz (1993) stellar atmosphere models. Nebular features are added based on the ionization code CLOUDY (Ferland 1996). The spectra thus obtained provide a reliable description of simple stellar generations up to ∼3 μm in the rest frame. Note that our models do not take fully into account the behavior of AGB stars, in other words they underestimate the contribution from intermediate-age SSPs. We estimate that this results in overestimating stellar masses of at most a factor of 2 (see a discussion in Berta et al. 2007a). Another overestimating factor comes from the assumption of a Salpeter IMF, indeed a Kroupa et al. (1993) or a Chabrier (2003) IMF would yield about 50% lower masses (see Berta et al. 2007a).

The mixed stellar population synthesis combines several phases, adopting a different SFR for each age. This code is well suited for modeling stellar populations of star-forming galaxies, because it allows several, independent episodes of star formation during the life of a galaxy, for example resembling multiple merger-driven starburst events. Age-selective extinction is applied, assuming that the oldest stars have abandoned the dusty medium long ago. Since disk populations are on average affected by a moderate extinction (A_V ≲1 mag, e.g., Kennicutt 1998), we adopt a maximum allowed absorption for stars older than 1 Gyr of A_V = 0.3–1.0 magnitudes. For younger populations the color excess gradually increases, but is limited to A_V ⩽5. The best fit is sought by means of χ² minimization. In the case of nondetections 3σ upper limits are adopted. For each source in the sample, the code explores the SFR extinction parameter space using an Adaptive Simulated Annealing algorithm and records all the attempted models. The uncertainty in the stellar mass estimate is then given by all those solutions whose χ² values lie within 3σ from the best fit. Details on the fitting method are described in Berta et al. (2007a).

Assuming that the total energy absorbed in the ultraviolet (UV)–optical domain is processed by dust in the thick molecular clouds embedding young stars and re-emitted in the MIR and FIR (5–1000 μm), we exploit the 24 μm observed flux to constrain the amount of dust and the luminosity of young stellar populations (see Berta et al. 2004, for more details).

The estimated stellar masses and associated uncertainties are listed in Table 7. The stellar masses range from 2.3 to 56 ×10¹⁰ M_☉, with a median value of 1.8 × 10¹¹ M_☉. Although the near-IR SED is dominated by stellar emission, it is possible that an AGN component is also present. Thus these estimates should be considered as upper limits to the true stellar masses. It is also possible that the derived stellar masses overestimate the true masses because our models do not take into account the TP-AGB contribution (Maraston 2005). Indeed such a contribution might yield lower masses, especially at the redshifts of our sources. The stellar masses as a function of redshift are shown in Figure 11. We find that at the same redshift, the stellar masses vary by a factor of 4. No difference in stellar mass is observed between 1.2 mm detected and nondetected objects. Similar masses are observed in objects with a wide range of luminosities and viceversa. For example, all high-z (z > 1.5) sources with available spectroscopic redshifts have similar FIR luminosities, ∼10^12.3 L_☉, but their stellar masses can vary by a factor of 5, i.e., M_* = 8–20 × 10¹⁰ M_☉.

Zoom In Zoom Out Reset image size

8.1. NIR Luminosities

In order to carry out a consistent comparison between the stellar masses of the SWIRE-MAMBO sources with the SMGs found in the literature, we compare their rest-frame luminosity at 1.6 μm, where the stellar emission peaks. The rest-frame luminosity at 1.6 μm is estimated by interpolating the observed IRAC fluxes. The estimated values as a function of redshift are reported in Figure 12. They range from 5.8 × 10¹⁰ L_☉ to 3.8 × 10¹² L_☉, and have a median value of 9.5 × 10¹¹ L_☉. We compare the luminosities of sources with a spectroscopic redshift and those with only a photometric redshift, and sources with 1.2 mm fluxes greater or lower than 2σ. No apparent difference in NIR luminosity is observed among these subsamples. We also divide the SMGs from the literature in two subgroups (shown as triangles and diamonds in Figure 12), those with a clear stellar bump in the IRAC bands and those for which a bump is not visible because of a different spectral shape or because there are no detections at λ = 5.8 and 8.0 μm. There is no significant difference in NIR luminosities between the SMGs with and without a stellar bump. The SWIRE-MAMBO sources are systematically more luminous in the NIR, suggesting that they might be more massive than literature SMGs. On average our sources are 3 times more luminous at 1.6 μm than submillimeter selected SMGs. This result is not surprising since our selection is based on a detection at 5.8 μm where the SWIRE 5σ limit is ≃50 μJy, while the literature sample is, on average, characterized by lower 5.8 μm fluxes, and both samples are at similar redshifts.

Zoom In Zoom Out Reset image size

It is well established that starburst sources dominate the bulk population of SMGs with only a minor contribution from AGNs. Indeed, although a weak AGN emission is observed in a significant fraction of SMGs, at X-ray energies (Alexander et al. 2005a), in optical spectra lines (Chapman et al. 2005), and in MIR spectra (Valiante et al. 2007; Menéndez-Delmestre et al. 2007; Pope et al. 2008), its contribution to their bolometric luminosity is usually modest (< 10%–20%).

In Section 6.4, the contribution from an obscured AGN to the observed MIR emission of our sources was invoked to explain their unusual colors and SEDs. Based on the available IRS spectra, we estimate, on average, a 34% contribution from a warm dust continuum in the MIR (6–12 μm in the rest frame). Such a component might be generated by an AGN, or by H ii regions (Laurent et al. 2000; Dopita et al. 2006). However, this estimate is based on only eight sources. Moreover, the IRS targets include the brightest 24 μm sources of the sample and might thus be biased towards sources with enhanced PAH emission or larger AGN contribution (Brand et al. 2006). To further investigate the presence of an AGN in our sample, we searched for possible signatures using X-ray and radio data.

X-ray observations from Chandra of moderate depth (70 ks) are available for 3 LH sources, LH-10, LH-19, and LH-28, from the Chandra/SWIRE survey (Polletta et al. 2006), and the CLASX survey (Yang et al. 2004). Although the XMM field has a wide shallow (10 ks) X-ray coverage from XMM-Newton  only one source, XMM-03, has been observed by XMM-Newton (Pierre et al. 2007), and it is at the edge of the field. Out of four sources with X-ray coverage, only LH-10 is detected in the X-ray. The X-ray 0.4–8 keV flux of LH-10 is 2.1 × 10⁻¹⁵ ergs cm⁻² s⁻¹, and the hardness ratio is HR = −0.7 (Yang et al. 2004), as expected for a steep unabsorbed power-law spectrum. The estimated 0.4–8 keV luminosity is 5.7 × 10⁴³ ergs s⁻¹, consistent with being AGN powered and with those measured in X-ray detected SMGs in HDFN (Alexander et al. 2005b). Although, based on the hardness ratio, the X-ray spectrum appears unabsorbed, the derived L_MIR/L_X-ray ratio, ≃100, is much higher than commonly observed in AGNs, 0.3–10, (see e.g., Franceschini et al. 2005; Polletta et al. 2007). The large L_MIR/L_X-ray ratio suggests that the X-ray luminosity is underestimated and that the observed X-ray emission corresponds to a small fraction of the intrinsic X-ray emission, perhaps due to a reflected component (warm scattering), while the intrinsic X-ray emission is completely absorbed. The lack of X-ray detections of the other 3 sources does not rule out the presence of an absorbed X-ray luminous AGN (L_X-ray > 10⁴⁴ ergs s⁻¹) or an unabsorbed AGN of moderate luminosity (L_X-ray < 10⁴⁴ ergs s⁻¹). Indeed the detection limit of the Chandra observations corresponds to X-ray luminosities ⩾10⁴⁴ ergs s⁻¹ for an unabsorbed AGN at the redshift of our sources. An AGN of lower luminosity or absorbed, as typically found in SMGs, would not be detected at our depths.

Deep VLA radio observations at 20 cm are available in a 0.4 deg² area centered on the Chandra/SWIRE field (Polletta et al. 2006, Owen & Morrison 2008) in the LH, and shallow radio observations are available from the VLA FIRST survey in all fields²⁰ (Condon et al. 1998). LH-11, and LH-22 are the only sources that fall in the deep VLA field and are both detected, with 20 cm fluxes of 82 and 51 μJy, respectively. A third source, LH-28, is detected by the FIRST survey. The radio fluxes are reported in Table 1. The radio source associated with LH-28 is extended, 666 × 316, corresponding to 57 kpc × 27 kpc. The bright radio flux and radio extension, combined with the lack of mm detection, indicate that the radio emission of LH-28 is powered by a jet and it is, thus, an AGN. LH-11, and LH-22 are not resolved in the radio images, < 1 24 and 0 87, respectively. Although their 1.4 GHz rest-frame radio luminosities are large, respectively, 10^28.2 and 10^27.8 W Hz⁻¹, they are not in excess of what is expected based on the FIR-radio relationship found in both starburst and radio-quiet AGNs (Condon et al. 1998). Therefore, the available radio data do not imply or rule out the presence of an AGN in these two luminous radio sources.

In summary, the available X-ray and radio data indicate that at least one out of three sources contains an AGN. This fraction is consistent with that derived from IRS spectroscopy (⩾ 30%), but all these samples are limited by small number statistics and there is no overlap among them, leaving the question on the AGN contribution and role in our SWIRE-MAMBO sample unanswered. Moreover, the subset of sources observed with the IRS might be biased towards a certain type of sources, for example with enhanced PAHs, and might not be representative of the whole SWIRE-MAMBO sample. Yet, since there is no apparent difference in the 1.2 mm/24 μm flux ratios of the X-ray and radio AGNs and the IRS sample, the estimated AGN contribution might be valid for the whole SWIRE-MAMBO sample. However, altough it is quite likely that an AGN is present in a large fraction of our sources, it is not expected to significantly contribute to their MIR emission, as is also the case for the classical SMGs (see e.g., Pope et al. 2008). Indeed, even by taking this contribution into account, we still find smaller f_1.2 μm/f_24 μm flux ratios, larger FIR luminosities, and a FIR peak at shorter wavelengths than in classical SMGs. Thus, as already stated in Section 6.4, it is unlikely that the difference between the SWIRE-MAMBO sources and classical SMGs is due to a more important AGN contribution in our sample.

In the previous sections we have investigated the multiwavelength properties of the SWIRE-MAMBO sources selected here and compared them with those of classical SMGs. Our sources are characterized by redshifts, 1<z<3, and millimeter fluxes in the same range as classical SMGs. Compared to classical SMGs, they are in general brighter at mid and near infrared wavelengths. Their FIR luminosities, constrained for the sources detected at 1.2 mm, are in the ULIRG range, ⩾10¹² L_☉. Their FIR emission is likely dominated by warmer dust than estimated in classical SMGs, but this result needs to be confirmed by additional submillimeter data. Their stellar masses are large, 10^10.3 to 10^11.8 M_☉, and their NIR luminosities and, likely stellar masses, are also higher by about a factor of 3 than in classical SMGs. Although some of these estimates are characterized by large uncertainties, it is a clear that our sources do show some differences compared to classical SMGs, or that they belong to a subclass of SMGs with larger MIR/FIR flux ratios than the bulk of SMGs (Figure 8). Indeed, SMGs with a relatively small 1.2 mm/24 μm flux ratio represent only a minority of all SMGs reported in the literature.

The possible origin of the observed differences between our sources and classical SMGs are discussed here. Warm dust emission is an indicator of strong overall heating of the dust. A large infrared luminosity and faint optical/UV emission implies elevated dust opacity, which in turn might imply a larger density of interstellar dust closer to the heating sources, or a more efficient heating mechanism relative to the dust (Dale et al. 2007; Kovács et al. 2006). Dust heating is predominantly fueled by the emission from massive (> 8 M_☉) stars in star-forming regions (Kovács et al. 2006), or by an AGN radiation field. The peak of IR emission of cold dust can thus shift towards shorter wavelengths if (1) the radiation field is higher because of younger stellar populations (Piovan et al. 2006), or (2) a significant contribution from AGN-heated dust is present. Thus, the difference in the FIR emission peak between our sample and classical SMGs (see Section 7) might be an indication of more massive stars relative to dust mass, or of a more significant AGN contribution in our SWIRE-MAMBO sources compared to classical SMGs.

The latter scenario has been investigated by measuring the intensity and equivalent width of the PAH features; weak PAH emission is indicative of an important AGN contribution. Since the average MIR spectrum, and thus the PAH equivalent width, of our sources and of classical SMGs are consistent (see Figure 9), a significant AGN contribution is disfavored. This result supports the alternative scenario of a higher contribution of massive stars relative to the dust mass. However, the available data do not allow us to test this hypothesis and rule out other explanations, perhaps related to the star forming region geometry and distribution (see e.g., Farrah et al. 2008).

Our results indicate that our selection has revealed a class of ULIRGs that is underrepresented in current SMG samples. The SWIRE-MAMBO sources indeed cover a different region in the luminosity-temperature (L − T_d) diagram than classical SMGs and bridge the gap between the latter and local ULIRGs (see Chapman et al. 2003; Lewis et al. 2005). Similar results have been obtained by a study of a similar sample of Spitzer-selected sources with known spectroscopic redshifts from strong PAH features, and multiple FIR detections (Younger et al. 2008).

Using the SWIRE survey and the MAMBO camera on the IRAM 30m telescope, we have identified 21 ULIRGs at z ≃ 1–3 with average 1.2 mm fluxes of 2.9 ± 0.2 mJy. They are characterized by starburst-dominated SEDs and by unusually large MIR/FIR flux ratios. These sources were discovered among a sample of 61 ULIRG z > 1 candidates selected in the SWIRE survey that were followed-up with MAMBO observations at 1.2 mm.

The primary selection criteria of the SWIRE sample of ULIRG candidates are a peak in the IRAC 5.8 μm band due to the red-shifted near-infrared spectrum of evolved stars, a bright (>400 μJy) detection at 24 μm, and a faint optical counterpart (r' > 23). The average 1.2 mm flux of the whole sample is 1.5 ± 0.2 mJy.

The optical–MIR SEDs of the selected sources are consistent with those of starburst galaxies. Photometric redshifts are estimated by fitting the SEDs with galaxy templates. For eight sources, spectroscopic redshifts from MIR spectra obtained with IRS on board of Spitzer are available and are in reasonable agreement with the estimated photometric redshifts.

Our analysis focuses on 29 sources in the Lockman Hole field where the average 1.2 mm flux (1.9 ± 0.3 mJy) is higher than in the XMM-LSS, ELAIS-N1 & ELAIS-N2 fields (1.1 ± 0.2 mJy). However, we also present basic results for 32 sources from the other fields, which were less thoroughly observed. The fraction of LH sources with 1.2 mm fluxes ⩾ 2σ is ∼40% (24% at 4σ; see Table 3). This fraction is much higher than previously found for Spitzer MIR selected samples (23%, 18%, and 10% at ⩾ 2, 3, and 4σ, respectively; Lutz et al. 2005). We attribute this difference to the fact that earlier samples favored AGN-dominated, rather than starburst-dominated systems. Our sample, on the other hand, shows systematically lower 1.2 mm/24 μm flux ratios than the majority of Spitzer-detected submillimeter-selected SMGs. Such a difference is attributed to a bias in our selection in favor of 24 μm-bright systems compared to the selection of SMGs.

Far-infrared luminosities are estimated assuming a z-dependent dust temperature as observed in classical SMGs, and normalizing a greybody model to the observed mm flux. Average 70 and 160 μm fluxes from stacked Spitzer/MIPS images are measured. The mean FIR stacked fluxes are ∼4 times larger than the fluxes predicted by the greybody models, suggesting higher FIR luminosities and a FIR peak at shorter wavelengths than in classical SMGs. The SWIRE-MAMBO sources are ULIRGs in the z = 1.5–3 range with FIR luminosities from 10¹² to 10^13.3 L_☉. Stellar masses, estimated by modeling the optical–infrared data with stellar synthesis populations, are also large, ∼ (0.2–6) × 10¹¹ M_☉, suggesting that these systems might be the precursors of the most massive ellipticals.

Compared to submillimeter selected galaxies (SMGs), the SWIRE-MAMBO sources are among those with the largest 24 μm mm⁻¹ flux ratios. The origin of such large ratios is investigated by comparing the average mid-infrared spectra and the stacked far-infrared spectral energy distributions of our sources and classical SMGs.

The mid-infrared spectra, available for a handful of sources, show strong PAH features and a warm dust continuum. The warm dust continuum contributes ∼34% of the mid-infrared emission, and is likely associated with an AGN component. This constribution is consistent with what is found in SMGs, and indeed the PAH equivalent widths are, on average, similar in our sources and in SMGs. Thus, even taking into account the AGN component, the 24 μm/1.2 mm flux ratios remain larger than typically observed in SMGs. The hypothesis of contribution from AGN emission to the MIR is thus disfavored, and it is more likely that the large ratios are due to more luminous PAH emission than in classical SMGs.

The analysis of the stacked far-infrared SEDs yields warmer dust temperatures than typically observed in classical SMGs. Thus, our selection has favored a warm class of ultraluminous infrared sources at high-z that is rarely found in current SMG samples. Indeed classical SMGs are not common among bright 24 μm sources such as the SWIRE-MAMBO sample (e.g., only about 20% of SMGs have f_24 μm > 0.4 mJy). Our sample occupies a region in the L − T_d diagram that is underpopulated by classical SMGs, implying that ULIRGs at high-z might exhibit a similar range in dust temperatures as local ULIRGs (see Chapman et al. 2003; Lewis et al. 2005). The analysis of our sample suggests that warmer ULIRGs are brighter MIR sources due to exhibiting more luminous PAH emission than cooler ULIRGs. However, to confirm such a correlation, an analysis of the MIR spectra for a larger sample of 24 μm-selected objects with known mm fluxes would be necessary. This analysis will be soon possible with other Spitzer samples that have been or will be observed with the IRS and with MAMBO (Younger et al. 2008, G. Lagache et al. 2009, in preparation).

Our sample is the largest Spitzer-selected sample detected at millimeter wavelengths currently available.

We thank the referee for useful comments that improved the paper. C.J.L. and M.P. thank Bruce Elmegreen and Jakob Walcher for helpful discussion. We are grateful to Elisabetta Valiante for providing the average IRS spectrum of a sample of SMGs. M.P. acknowledges financial support from the Marie-Curie Fellowship grant MEIF-CT-2007-042111. This work is based on observations made with the IRAM 30m Telescope and the Spitzer Space Telescope. The IRAM 30m Telescope is funded by the Centre National de la Recherche Scientifique (France), the Max-Planck Gesellschaft (Germany), and the Instituto Geografico Nacional (Spain). The Spitzer Space Telescope is operated by the Jet Propulsion Laboratory, California Institute of Technology under NASA contract 1407. Support for this work, part of the Spitzer Space Telescope Legacy Science Program, was provided by NASA through an award issued by the Jet Propulsion Laboratory, California Institute of Technology under NASA contract 1407. We thank the IRAM staff for their support during the observations, and E. Kreysa and his group for providing the MAMBO bolometer array. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.

Facilities: Spitzer (IRAC,MIPS), IRAM(MAMBO).

The best and acceptable fits to the optical (Ugriz) and MIR (3.6–24 μm) SEDs of each source, obtained as described in Section 5 are shown in Figure 13. We report a maximum of three solutions, those with the lowest 3 χ². In case a spectroscopic redshift is available, the best-fit template at the spectroscopic redshift is reported as dotted curve. Although the templates extend to FIR wavelengths, the FIR data are not used in the fitting procedure and in the χ² computation. The parameters (template, and redshift) of each solution are reported in Table 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Eight sources have spectroscopic redshifts and are shown in Figures 13(a), (c), and (f). It can be appreciated from this figure that with only the broad NIR bump to constrain the photometric redshift, using the broad IRAC filters, a redshift accuracy of 10% in (1+z) is a reasonable expectation. In some cases, a NIR bump can be present in sources at lower redshift. Indeed the presence of warm dust continuum can make the NIR bump or the turnover at 8 μm less pronounced and even shift the stellar peak to longer wavelengths (Berta et al. 2007b; Daddi et al. 2007). Also strong emission from the 3.3 μm PAH feature falling into the 5.8 μm filter can produce a peak in the IRAC SED that would be wrongly interpreted if no optical or NIR data were available. These two effects increase the uncertainty associated with the photometric redshift estimate. Some examples can be seen in the fit to source LH-12 (z_spec = 1.89 and z_phot = 2.65) and LH-23 (z_spec = 0.98 and z_phot = 1.21) in Figures 13(a) and (c).