COMPLETE INFRARED SPECTRAL ENERGY DISTRIBUTIONS OF MILLIMETER DETECTED QUASARS AT z > 5

All objects were observed with the PACS (Poglitsch et al. 2010) at 100 and 160 μm using the mini-scan map observing template. For each source we obtained two maps with different scan angles using observing parameters as recommended in the mini-scan map Astronomical Observation Template release note.⁶ For each scan direction, five repetitions were executed. This resulted in a total on-source integration time of ∼900 s for each object.

Data reduction was performed within the Herschel Interactive Processing Environment (HIPE; Ott 2010), version 8.0.1. We followed standard procedures for deep field data reduction, including source masking and high-pass filtering. The maps of the two scan directions were processed individually and mosaicked at the end of the work flow. A first version of the combined map was used to create a source mask for high-pass filtering. Masking was performed by hand through visual inspection of the mosaicked maps. This proved to be more reliable than a strict sigma cut, as it also allowed the masking of faint structures which can potentially influence the fluxes of faint targets if located close to the object of interest. A second processing was then performed, including the mask. The resulting maps (individual and mosaic) were inspected visually and if necessary the mask or width of the high-pass filter was adjusted. Images around the target positions at 100 and 160 μm are presented in Figure 1.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Aperture photometry of the final mosaics was performed in IDL. We used relatively small apertures of 6''–10'' radius to maximize the signal-to-noise ratio. Appropriate aperture corrections were determined from tabulated values of the encircled energy fraction of unresolved sources. Details on the properties of the PACS point spread function (PSF) can be found on the PACS calibration Web sites.⁷

The uncertainty in the photometry of our maps cannot be determined directly from the pixel-to-pixel variations because the final PACS maps suffer from correlated noise (where the level of pixel noise correlation depends on the details of the data reduction and final map projection). In order to estimate robust photometric uncertainties we therefore implemented the following procedure: for any given map a set of 500 apertures of the same size as used for the QSO photometry was placed on random positions on the sky (see also Lutz et al. 2011; Popesso et al. 2012). The only constraint on the placement of these background apertures was that the central pixel of the aperture has to have at least 75% of the integration time as the position of the QSO (to exclude noisy areas at the edge of the map). The distribution of the measured fluxes in these 500 apertures was then fitted by a Gaussian. The sigma value of this Gaussian was taken as the 1σ photometric uncertainty of this map. The resulting flux values for the quasars are given in Table 2.

All quasars in the sample were also observed with the SPIRE (Griffin et al. 2010) at 250, 350, and 500 μm in small scan map mode for five repetitions and a total on-source integration time of ∼190 s per source. Data reduction followed standard procedures in HIPE as recommended by the SPIRE instrument team. Source extraction was performed with the HIPE built-in task "sourceExtractorSussextractor" (Savage & Oliver 2007) using information on the PSF (e.g., FWHM) given in the SPIRE Observer's Manual.⁸

Our observations are dominated by confusion noise which is on the order of 6–7 mJy beam⁻¹ in the SPIRE photometric bands (Nguyen et al. 2010) as determined from deep extragalactic fields. In order to estimate the uncertainties due to confusion noise specifically in our target fields, we implemented the following procedure (see also Elbaz et al. 2011; Pascale et al. 2011). First, the source extractor was run over the full calibrated maps. Detections within less than half the FWHM from the nominal target position were tentatively considered to belong to the quasar, pending further confirmation from our check for confusion with nearby FIR bright sources. We then created an artificial image which included all the sources found by the source extractor and subtracted this "source image" from the observed map. On this "residual map" we determined the pixel-to-pixel rms in a box with a size of 8 times the FWHM (FWHM size: 182, 249, and 363 for default map pixel sizes of 6'', 10'', and 14'' at 250, 350, and 500 μm, respectively), centered on the nominal position of the QSO. The size of this box was chosen large enough to allow an appropriate sampling of the surroundings of the source, but small enough to avoid including the lower coverage areas at the edges of the map even for the longest wavelengths. In addition, the number of pixels per FWHM is approximately constant for the three wavelengths in the final maps (2.5–3.0 pixel FWHM⁻¹) which translates into a similar number of pixels used for determining the rms in the background box. The resulting estimates for the noise (limited by confusion) are comparable to the average values given in Nguyen et al. (2010), but have a tendency to be slightly lower. The fluxes and uncertainties we determine are given in Table 2 and the 250 μm maps are presented in Figure 1.

We note that Table 2 lists a number of SPIRE flux measurements which are nominally below the estimated 3σ value of the noise. In these cases, the inspection of the images revealed a clear excess of flux at the position of the quasar. It has been shown that the use of positional priors can reduce the effect of confusion noise by 20%–30% (Roseboom et al. 2010). While our strategy is somewhat different from that work, we benefit not only from accurate (relative and absolute) positional information, but also from information on the SEDs of the quasar and potential confusing sources in the field via our multi-wavelength data. This leads us to include these flux measurements in this study, although they have to be treated with caution. Similarly, fluxes at 500 μm should be considered tentative because at this wavelength the beam is large (∼36'' FWHM), the confusion noise is high, and the significance of the detections is often low.

For all Herschel targets (except J2054−0005) we also have available mid-infrared (MIR) imaging from Spitzer at 3.6, 4.5, 5.8, and 8.0 μm with the InfraRed Array Camera (IRAC; Fazio et al. 2004) as well as at 24 μm with the Multiband Imaging Photometer for Spitzer (MIPS; Rieke et al. 2004). For the redshifts of our sources these passbands cover the rest frame optical and near-infrared (NIR) wavelengths (∼0.5–4 μm). The Spitzer data have been processed using standard procedures within the Mopex software package provided by the Spitzer Science Center (SSC). Aperture photometry was performed in IDL using standard sets of aperture radii and appropriate aperture corrections as outlined in the respective instrument handbooks (also available from the SSC Web site). Errors were estimated in a similar fashion as for PACS by measuring the flux in randomly placed apertures on empty parts of the background and determining the variations between these background flux measurements. All objects in this paper with Spitzer data are detected at high significance in the five bands. The resulting photometry is summarized in Table 2 and is usually consistent with measurements published previously, where available (e.g., Jiang et al. 2006, 2010; Hines et al. 2006). The 24 μm images are presented alongside the PACS and SPIRE 250 μm images in Figure 1.

Our multi-wavelength data set, and in particular the Spitzer 24 μm images, provide a tool for determining the exact position of the quasar in the Herschel bands. Since we can often identify several sources per field that are visible both at Spitzer and at Herschel wavelengths, the exact location of the quasar in the FIR maps can be determined from the relative positional information. With this procedure we can robustly identify faint Herschel detections with the quasars as well as avoid mis-identifications due to nearby objects. During this exercise we observe absolute spatial offsets between Spitzer and Herschel of typically ≲2'', in line with expectations from the absolute pointing accuracies.

The all-sky data release of the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010) was queried for photometry or upper limits in the 12 μm band which can fill the gap in the Spitzer photometry between 8 and 24 μm. Only three quasars in this paper are detected at 12 μm and their significance is low (≲3.5σ, Table 2). Some objects (e.g., J2054−0005) are not detected in any WISE band, and consequently no upper limits are available in the point-source catalog. For such sources we performed aperture photometry on the WISE atlas images to determine upper limits, following the guidelines in the explanatory supplements to the WISE All-Sky Data Release Products (Cutri et al. 2012).

Additional data from the literature were compiled, mainly in the observed NIR or millimeter regimes and often taken from the discovery papers. The latest data release of the UKIDSS survey (Lawrence et al. 2007) was also checked which yielded additional photometry in the NIR for six objects.

In combination with other supplemental data from the literature (Section 2), we compile SEDs covering a rest frame wavelength range of typically 0.1–400 μm (see Figures 2 and 3). These SEDs are then fitted with a combination of models to represent the different components contributing to the observed SED. For this purpose we have divided our sample into two groups, depending on the amount of data available to constrain the fitted components. Objects detected in at least two Herschel bands were subject to full SED fits (five sources), except J2054−0005, for which the lack of strong photometric constraints at rest frame wavelengths <10 μm (no Spitzer observations and only WISE upper limits) prevented a full SED fit. In our fits we consider four distinct components.

• 1.  
  A power law in the UV/optical regime which represents the emission from the accretion disk. We extend this component into the NIR and introduce a break to the Rayleigh–Jeans slope of F_ν∝ν² at 3 μm in the rest frame (Hönig & Kishimoto 2010). In the fitting, the power-law slope in the UV/optical and the overall normalization are free parameters.
• 2.  
  A blackbody of typically 1300 K temperature, thus peaking in the rest frame NIR. Empirically, such a component is often required to fit the optical through MIR SEDs of luminous quasars (e.g., Barvainis 1987; Gallagher et al. 2007; Mor et al. 2009; Leipski et al. 2010a) and is generally interpreted as a signature of hot (graphite) dust close to the sublimation temperature.
• 3.  
  A clumpy torus model from the library of Hönig & Kishimoto (2010) to account for the active galactic nucleus (AGN) heated dust from the "dusty torus" in the central parts of the AGN. This component dominates the MIR and is important to disentangle, to first order, the contributions from the nuclear dust to the rest frame FIR emission. Besides the choice of a particular model (see below), the absolute scaling of the model is the only free parameter for this component.
• 4.  
  A modified blackbody to account for possible FIR excess emission (over the AGN heated torus) which we here interpret as powered by star formation. For this component we fix the emissivity index β to a value of 1.6 (e.g., Beelen et al. 2006; Wang et al. 2008a; but see Section 4.3). The temperature of the modified blackbody and the normalization are free parameters.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The torus models we consider here are available for seven different inclinations (starting at 0° and increasing in steps of 15°). For each inclination, Hönig & Kishimoto (2010) provide models of various combinations of parameters (opening angle, radial dust distribution, etc.). Our fitting procedure takes one of the torus models and fits a linear combination of the four components to the observed SED via chi-square minimization using Mpfit (Markwardt 2009) in IDL. We then cycle through all torus models in the library and repeat the fitting for each of them. However, for efficiency we limit the torus component to models with inclinations ⩽45°, which seems reasonable given that we observe luminous and largely unreddened type-1 quasars. This leaves a total of 959 different torus models. We visually inspect the best-fitting 10% of the model combinations to confirm the fits.

With this approach we do not intend to develop a highly accurate model for the full SED emission in these objects. For the current work we aim to construct a physically motivated approximation that yields a reasonable description of the observed SEDs which allows us to isolate excess FIR emission and to account for (to first order) contributions of the AGN heated nuclear dust to the FIR photometry.

As outlined above, our fitting includes an additional, empirically motivated 1300 K blackbody in the NIR. We have also performed the fits excluding this component, only fitting a power-law, a torus model and a FIR modified blackbody. The comparison between the two cases shows that the fits including the NIR blackbody generally represent the observed photometry better. This is particularly apparent in wavelength regions dominated by very hot dust (λ_rest  ∼ 1–3 μm), in the overlap region between the torus and the FIR blackbody (λ_rest  ∼ 20–30 μm) and in the fit at λ_rest ≳ 100 μm. The temperatures of the FIR dust component also come out consistently lower (by about 5–10 K) in fits including the additional NIR blackbody.

Recall that the additional NIR component contributes significantly (or dominantly) to the short infrared wavelengths. In cases where this component is absent, torus models with a strong emphasis on emission at λ ≲ 10 μm are favored in the fits to accommodate (in particular) the MIPS photometry. By design such torus models contribute less flux at longer wavelengths (λ ≳ 20 μm), thus requiring a hotter FIR component to match the Herschel photometry. This in turn negatively affects the fit in the observed submillimeter/millimeter regime. Including the NIR component, more power in the torus component can shift to slightly longer wavelengths, allowing a cooler FIR component and providing a better overall fit to the data. The need to add an additional hot component to torus models when fitting type-1 AGN SEDs has also been noted by e.g., Mor & Netzer (2012). NIR reverberation mapping observations (e.g., Suganuma et al. 2006) show that the size of the emitting region of this very hot dust is a factor of ∼20 smaller than the torus as measured in the MIR via interferometry (L. Burtscher et al., in preparation), supporting this additional complexity in the distribution of the AGN-heated dust. The following results and discussion will therefore be based on the fits including the additional NIR blackbody.

In our new Herschel observations we detect six out of eleven sources (Table 2). Typically, the quasars are either detected in all five bands or not at all with Herschel. The exceptions are J0927+2001 which is detected with SPIRE but not with PACS and J1044−1025 for which the opposite is the case.

From the 10 objects observed with Spitzer, all are detected in all bands. With WISE at 12 μm only three objects are detected and at low significance. More detailed information on the individual objects can be found in the Appendix.

Previous studies of star formation in high-redshift (z > 5) quasars often had to rely on single band millimeter emission as a tracer for the starburst heated dust (e.g., Bertoldi et al. 2003; Wang et al. 2008a). FIR luminosities were determined by fitting a single modified blackbody to the millimeter photometry and integrating under this component. Since no knowledge about the temperature of the dust was at hand for most cases, typical values found for FIR bright quasars at lower redshift (z ∼ 2–4) were assumed (e.g., T = 47 K, β = 1.6; Beelen et al. 2006). This approach has been tentatively supported for some high-redshift quasars by ground-based observations at 350 μm (Wang et al. 2008b, 2010).

The new multi-wavelength FIR photometry now allows us to estimate the temperature of the FIR emitting dust directly while simultaneously accounting for the contributions from AGN heated nuclear dust to the infrared. In Figure 2 we present the SEDs of the five objects which are detected in at least two Herschel bands and their accompanying best fits.

As explained above, details of the SED of the AGN heated dust torus affects the shape (and temperature) of the FIR component. We have taken this into account when calculating the uncertainties of the dust temperature (see Section 4.5). The values of T_FIR we obtain here for these five objects are reported in Table 3. We clearly see a range of temperatures among these objects, spanning almost 20 K.

One important result from the SED fits in Figure 2 is that the flux at λ_rest ≳ 50 μm is usually dominated by the FIR excess emission with only minor contributions from the AGN heated torus. This is further illustrated in Figure 4 where we show the fractional contribution of the hot dust plus dusty torus component (both presumably powered by the AGN) compared with the contributions from the FIR excess component, which may be powered by star formation. As indicated in this figure, the SPIRE 350 μm band is typically dominated by emission from the FIR blackbody for the redshifts considered here. In fact, fitting only a single modified blackbody (β = 1.6) to the photometry at λ ⩾ 350 μm gives very similar dust temperatures compared to the full SED fits.

This result can immediately be utilized for an estimate on the dust temperature for objects which do not qualify for full SED fits (see Figure 3). In these cases we use the 250 GHz detection as an anchor for the (modified) blackbody while the FIR upper limits at λ ⩾ 350 μm (mostly from SPIRE) allow us to constrain the maximum permitted temperature of this component. Because of our findings from the full SED fits (Figure 4) we can assume that our upper limits to the dust temperature are reasonably robust when limiting the fits to λ ⩾ 350 μm, even without suitable constraints on the AGN dust emission in these objects. We determine dust temperatures of ⩽57 K, with J0818+1722 being the only exception (T_FIR ⩽ 71 K). However, the latter value has to be taken with caution because the 250 GHz photometry of this source could be contaminated by the emission from the nearby galaxy that is detected in most infrared bands (see notes on individual objects in the Appendix and Figure 1).

In the previous fits, the emissivity index β was fixed to a value of 1.6 to enable the comparison with earlier literature studies. The good photometric coverage in the FIR and submillimeter for some of our objects now allows us to explore how far β can be constrained using the quasar SEDs. For such a study, additional photometry at lower frequencies (typically around 90 GHz in the observed frame) is very important as it helps to further constrain the Rayleigh–Jeans tail of the fitted dust component. In fact, considering only photometry at ν_obs > 250 GHz does not provide good constraints on β (or the FIR dust temperature) if both parameters are kept free during fitting. Photometry at ∼90 GHz in combination with Herschel FIR detections is available for four objects in our sample. Re-fitting the SEDs as outlined previously but now keeping β as a free parameter, we find relatively high β values (2.0–2.7) combined with relatively low temperatures (∼33 K, but still 54 K for J1148+5251). The integrated luminosity of the FIR dust when determined from these new fits remains virtually unchanged as compared to a fixed β approach. We caution, however, that a reliable measure of β is hard to obtain in these objects because the peak of the dust emission is not well defined (or isolated) in the SED due to the strong nuclear dust emission from the torus. Therefore, the peak wavelength (and temperature) of the FIR blackbody depends on the choice of the torus model, which adds additional uncertainty in the determination of β.

Integrating the emission of the FIR excess component only (i.e., the modified blackbody) between 8 and 1000 μm in the rest frame allows us to determine its luminosity L_FIR. The FIR luminosities we derive are of the order of ∼10¹³ L_☉ (see Table 3), which corresponds to ∼25%–60% of the bolometric FIR luminosity in these objects. For sources with full SED fits, the values for L_FIR do not change significantly (≲ 10%) if we fit only a single modified blackbody to the photometry at λ_obs ⩾ 350 μm.

Assuming that the FIR emission is powered by star formation, we converted the FIR luminosities into star-formation rates using the formula in Kennicutt (1998). The results are reported in Table 3. Using our fits of the FIR excess component we can also derive an estimate for the dust masses in the star forming regions:

$$\begin{equation} M_{{\rm dust}} = \frac{S_{250\,{\mu }{\rm m}} D_{L}^2}{\kappa _{250\,{\mu }{\rm m}} B_{\nu }(250\,{\mu }{\rm m},T_{{\rm FIR}})}, \end{equation} \tag{ 1 }$$

where S_250 μm is flux level at a rest frame wavelength of 250 μm as determined from the fit, D_L is the luminosity distance, κ_250μm is the dust absorption coefficient at 250 μm as determined from the models of Draine (2003), and B_ν(250 μm, T_FIR) is the value of the Planck function with temperature T_FIR at a wavelength of 250 μm. The results are also reported in Table 3.

In order to estimate uncertainties in the derived parameters we studied the distribution of their values resulting from all the fitted models (N ∼ 1000; as outlined in Section 3.2 we only include torus models with inclinations of ⩽45°). This also allows us to account for the influence the choice of a particular torus model has on these parameters.

As a first step, we calculated for all the fitted models the residuals between the global fit and the observed data in the infrared (λ_rest > 1 μm) and determined the error-weighted rms for these points. A typical distribution of the resulting rms values is presented in panel "(a)" of Figure 5 (we here use the quasar J0756+4104 as an example to demonstrate our approach). In this figure we see a clear peak at low rms values representing a family of good fits, with an extended tail to large rms values corresponding to increasingly worse model representations of the observed SED. We then fitted the right side of the rms peak with a Gaussian (dashed line). All fits with an rms value within 3σ of the centroid value of the Gaussian are identified as acceptable model fits. The values corresponding to these fits are marked as blue and hashed regions in all panels of Figure 5 and are used for estimating uncertainties on the derived values.

Zoom In Zoom Out Reset image size

Each of the N ∼ 1000 model fits provides a value for the temperature of the modified blackbody in the FIR (T_FIR, panel "(b)" in Figure 5). From the temperature and the normalization of this component we can then calculate (see Section 4.4) L_FIR (panel "(c)"), a star-formation rate (panel "(d)"), and a dust mass M_dust (panel "(e)"). In each histogram we fit the parameter values obtained from the acceptable fits (as determined from the residual rms distribution; blue and hashed regions) with a Gaussian (dashed line). This Gaussian fit provides us with a mean parameter value (centroid) and an uncertainty (σ) for each fitted object. These results are reported in Table 3.

The average temperature of the modified blackbody used to model the FIR emission is comparable to the ∼47 K measured for lower redshift FIR bright quasars (Beelen et al. 2006). This finding is confirmed by the average SEDs presented in Section 5.2 below.

While this is true on average, we see a significant spread in dust temperature (∼20 K) between individual objects, even for comparable FIR luminosities. Despite the low number of objects for which such fits can be performed, this finding highlights that the choice of the dust temperature can add uncertainty to the estimate of L_FIR and M_dust, in particular for objects with only single photometric measurements and thus no individual constraints on T_FIR.

A related issue is that of possible AGN contributions to the heating of the FIR dust which will be discussed briefly in the following section.

We also find that in our modeling strategy the FIR component can be isolated from the torus component if only data at λ_rest ≳ 50 μm are considered (see Section 4.2 and Figure 4). Consequently, single-component fits to data at these wavelengths yield estimates of the FIR luminosity, temperature and dust mass that match the values based on the full SED fits. This result validates the approach in previous studies of high-z quasars (Bertoldi et al. 2003; Beelen et al. 2006; Wang et al. 2008a, 2008b) in which single-component fits to (ground-based) photometry at λ_obs ⩾ 350 μm were used to derive physical parameters (with the caveat of unknown dust temperature in some of these studies). It also adds further significance to the upper limits on T_FIR, L_FIR, and SFR we determine for the remainder of our sample where we are limited to single-component fits at λ_obs ⩾ 350 μm.

The strong overlap between the torus and the FIR components in our fits does not provide good constraints on the emissivity index β of the latter component. The difficulty of determining reliable β estimates in objects with a strong AGN is also apparent from the literature: Priddey & McMahon (2001) find a high β value of ∼2 (with a FIR dust temperature of 41 K) by combining the available photometry for a number of z ∼ 4 quasars into a single SED and using this global SED to constrain a modified blackbody fit. On the other hand, using a similar approach (and much of the same data), Beelen et al. (2006) find β ∼ 1.6 and T_FIR ∼ 47 K for a sample of quasars with z = 1.8–6.4.

In order to better constrain the FIR emission of the SPIRE non-detected objects, we stacked the individual SPIRE observations at the nominal position of the quasar (excluding J0818+1722 due to possible confusion issues).⁹ Even in the stacked images no significant detection was achieved (see Figure 6). Stacking the corresponding PACS data we recover a faint (∼3σ) average signal at 100 and 160 μm. We iterated during the stacking, leaving a different source out of the stack for every iteration to verify that the result is not biased by any individual object. Differences between these stacks were usually smaller than the uncertainty on the photometry of the total stack. The photometry was performed in an identical manner to the individual frames and as outlined in Section 2.

Zoom In Zoom Out Reset image size

Using the stacked Herschel fluxes and combining them with stacked WISE data as well as averaged Spitzer and millimeter photometry, we can produce an average SED for the FIR non-detected objects which is presented in Figure 7. Performing the same stacking/averaging procedure for the objects detected in the FIR individually (excluding J1148+5251 due to possible confusion issues) also provides us with an average SED for these objects.

Zoom In Zoom Out Reset image size

Fitting the average FIR SEDs with a modified blackbody (at λ_obs ⩾ 350 μm, β = 1.6) reveals dust temperatures of 47 K for the stack of the individually detected objects, as expected considering the individual FIR dust temperatures of the objects in this stack (Table 3). For the average SED of the objects individually undetected in SPIRE we determine an upper limit of 49 K which is lower than the upper limits determined individually (Table 3). These temperatures are very similar to the values commonly adopted for millimeter detected quasars with only few or single photometric data points (e.g., Beelen et al. 2006; Wang et al. 2008b). They also agree well with the dust temperatures observed for submillimeter galaxies of comparable FIR luminosity (Magnelli et al. 2012).

In Figure 7 we also compare our average SEDs with the Sloan Digital Sky Survey (SDSS) quasar template of Richards et al. (2006). When normalized at 1 μm (using a power-law fit to the rest frame UV/optical emission), we see that below ∼30 μm the average SED of the SPIRE non-detections (blue points in Figure 7) looks similar to the typical SDSS quasar template. At longer wavelengths we see an additional cool dust component for our sources, as indicated by the available millimeter observations. Not considering these longer wavelengths, we see AGN powered emission with possibly increased emphasis on the hottest dust (NIR) when compared to the template.

In contrast, for a similar scaled UV/optical emission, the average infrared SED of the SPIRE detected quasars (red points in Figure 7) exceeds the scaled template significantly at any wavelength λ_rest ≳ 1 μm. This includes the presumably AGN powered dust emission at short infrared wavelengths (λ_rest ≲ 20 μm) as well as the FIR emission possibly powered by additional star formation.

The observed discrepancy between the average SEDs in Figure 7 is somewhat puzzling. Both groups of objects have similar UV/optical properties (on average), indicating that their black holes grow at comparable rates. Now, assuming that the NIR and MIR emission is powered by the AGN, the objects with individual Herschel detections (red symbols) convert a higher fraction of their accretion luminosity into reprocessed dust emission. This could, in principle, be caused by a different dust geometry (or dust content) in the inner parts of these objects. This in turn could also lead to increased contributions of AGN powered dust emission at FIR wavelengths. If AGN-powered dust emission extends further into the FIR than anticipated by the torus models we utilize here, the values for the inferred star-formation luminosity (and star formation rates) in these objects (Table 3) would be overestimated.

While a detailed discussion of this issue is beyond the scope of this paper, it is worth keeping in mind that the AGN may be contributing to the heating of the FIR dust depending on the dust distribution and geometry.

Object names are given in the format Jhhmm+ddmm. For full, NED compatible object designations see Table 1.

J0203+0012 (z = 5.72). The combined optical and NIR spectroscopy by Mortlock et al. (2009) reveals broad absorption line features in this quasar which could be one explanation for the abrupt change in SED slope observed at λ_rest ≲ 0.3 μm (Figure 3). It is fairly radio bright at 1.4 GHz compared to most z ∼ 6 quasars (Wang et al. 2008a) and is detected at 250 GHz with a flux of 1.85 ± 0.46 mJy (Wang et al. 2011). These authors did not detect the quasar in CO(6–5) nor in the corresponding continuum. While clearly seen in all our Spitzer observations, no detection was achieved with Herschel. Several nearby galaxies can be identified in SDSS and IRAC, some of which are prominent in most infrared channels. The closest bright object is ∼30'' northeast of the quasar. From SDSS spectroscopy of some galaxies in this area it appears that they might belong to a foreground cluster at redshift of z ∼ 0.077.

J0338+0021 (z = 5.03). This source was detected at 250 GHz (3.7 ± 0.3 mJy), but remained undetected at 1.4 GHz (Carilli et al. 2001). The 850 μm flux is measured to be 11.9 ± 2.0 mJy (Priddey et al. 2003). Maiolino et al. (2007) report CO(5–4) emission from the quasar, but no continuum at 95.6 GHz is detected. Inspection of the field at optical through FIR wavelengths reveals a nearby source ∼15'' to the west of the quasar. This object is clearly visible in the SDSS r and i bands, as well as with IRAC and MIPS. In all these filters, the quasar is typically the brighter source, a situation which reverses at 100 and 160 μm where the nearby object is ∼1.4 and ∼1.3 times brighter, respectively, than the quasar. In all SPIRE bands, however, we only detect a single source. Interestingly, the quasar and the nearby object can both be seen in ground-based 350 μm observations (Wang et al. 2008b) and the flux of the quasar (17.7 ± 4.4 mJy beam⁻¹) is very comparable to the SPIRE 350 μm flux determined for the single detection (18.5 ± 6.0 mJy). This suggests that the SED of the source close to the QSO peaks somewhere around the PACS bands but does not contribute significantly at wavelengths ≳ 350 μm.

J0756+4104 (z = 5.11). The 250 GHz flux (5.5 ± 0.5 mJy) has been presented by Petric et al. (2003) who also detect the source at 1.4 GHz and constrain the size of the radio emission to <23. Priddey et al. (2008) report detections at 850 μm (11.2 ± 1.0 mJy) and 450 μm (16 ± 5 mJy) where "the source clearly appears elongated" in the 850 μm map at a position angle of ∼70°. We detect the quasar in all our five Herschel bands, but the significance is often marginal. Our flux at 350 μm is consistent with the ground-based measurements of Wang et al. (2008b).

J0818+1722 (z = 6.00). Radio continuum emission at 1.4 GHz is clearly detected (Wang et al. 2007) and the 250 GHz continuum is observed at the 3σ level (1.19 ± 0.38 mJy; Wang et al. 2008a). Close inspection of our multi-wavelength images reveals a resolved foreground galaxy ∼6'' north-east of the quasar. Both objects, the galaxy and the quasar, are individually detected by MIPS at 24 μm and in shorter wavelength bands. However, only a single source is detected in PACS and SPIRE (note that the spatial resolution of Herschel/PACS at 100 μm is comparable to Spitzer/MIPS at 24 μm). From the relative positions of other sources in the field we can identify the detection at 100, 160, and possibly at 250 μm with the foreground galaxy. It is conceivable that the faint (3.5σ) detection at 350 μm is also due to this source. No detection is achieved at 500 μm. In the light of these results, higher resolution millimeter observation are clearly needed to determine the source of the 250 GHz continuum emission.

J0840+5624 (z = 5.84). We do not detect this source in our PACS nor in the SPIRE data. The quasar has been detected at 250 GHz (3.20 ± 0.64 mJy), but not at 1.4 GHz (Wang et al. 2007). CO emission is seen in this source (but no continuum at either 85 GHz or 101 GHz) and the 350 μm emission is "marginally detected" from the ground (Wang et al. 2010), which is consistent with our SPIRE 350 μm upper limit. Wang et al. (2010) also report the presence of another source visible at 350 μm as well as at 1.4 GHz located ∼30'' north-west of the quasar. This source is also detected in all our infrared bands. Inspection of the IRAC maps reveals this detection to coincide with two close objects which could be two slightly resolved galaxies separated by ∼18 as seen on an archival HST/WFC3 image in the F105W filter.

J0927+2001 (z = 5.77). Previously detected at 250 GHz (4.98 ± 0.75 mJy; Wang et al. 2007) as well as in CO and in the 90 GHz continuum (Carilli et al. 2007). The 350 μm observations by Wang et al. (2008b) show the quasar (17.7 ± 5.7 mJy beam⁻¹) and a secondary source of equal brightness 15'' to the southeast. While the quasar detection was confirmed in Wang et al. (2010) with better sensitivity (11.7 ± 2.4 mJy beam⁻¹), the secondary source was not. We detect the quasar with SPIRE, but not with PACS.

J1044−0125 (z = 5.78). This well studied quasar shows a broad C iv absorption feature in its spectrum (e.g., Maiolino et al. 2001; Goodrich et al. 2001). The continuum emission is detected at 850 μm (5.6 ± 1.0 mJy; Priddey et al. 2008) and at 250 GHz (1.82 ± 0.43 mJy; Wang et al. 2008a). Wang et al. (2010) report the detection of CO(6–5) but can only give an upper limit on the continuum at 102 GHz. The quasar is not seen in ground-based observations at 350 μm (Wang et al. 2010) and 1.4 GHz (Petric et al. 2003). We detect the quasar with PACS, but not with SPIRE.

J1048+4637 (z = 6.23). While clearly detected in the available Spitzer bands, this quasar remains undetected in our Herschel photometry. Wang et al. (2008b) only give an upper limit on the 350 μm flux, just like Robson et al. (2004) at 450 and 850 μm. These authors, however, note that based on the published 1.2 mm detection (3.0 ± 0.4 mJy; Bertoldi et al. 2003), the SCUBA 850 μm observations are deep enough to enable the detection of the source with 4σ significance given a dust temperature of 40 K. At even longer wavelengths, Wang et al. (2010) detect CO(6–5) emission as well as the continuum at 96 GHz. Observations at 1.4 GHz only provide an upper limit on the source flux (Wang et al. 2007).

J1148+5251 (z = 6.42). This famous object was the highest redshift quasar known for half a decade (Fan et al. 2003; Willott et al. 2007) and as such has been studied at many wavelengths, including the millimeter and submillimeter regime. We also have observed this quasar previously with Herschel/PACS and reported detections at 100 and 160 μm (Leipski et al. 2010b). Surprisingly, we discovered a secondary object ∼10'' north-west of the quasar which is brighter at 160 μm but can still be identified at 100 μm. Ground-based data at 350 μm (Beelen et al. 2006) and 250 GHz (Bertoldi et al. 2003) revealed an intriguing elongation of the quasar detection in the direction of the second source. We argued in our previous paper that this could be an indication for the presence of the secondary source. A possible counterpart is also seen in the 24 μm images. In the IRAC band, three sources are detected around the position of this secondary source, two of which can also be identified on deep Hubble Space Telescope (HST) images with the Advanced Camera for Surveys camera in the F850LP filter. Recently, new deep HST images from WFC3 in the NIR revealed also the third source seen in IRAC. This object is clearly detected, but faint in F125W and gets significantly brighter in F160W. This is our best candidate for a counterpart of the secondary source seen at 24, 100, and 160 μm. Surprisingly, however, follow-up observations with the Plateau de Bure Interferometer at 1.2 mm at ∼1'' resolution did not yield a detection and the 3σ upper limit we derive is 0.9 mJy.

Since our initial photometry (Leipski et al. 2010b) we have re-observed the quasar with Herschel and obtained new images at 70, 160, 250, 350, and 500 μm. While the quasar itself is faintly detected at 70 μm, there is no sign for a secondary source. The new 160 μm observations confirm our earlier findings that the flux appears to come from two sources. The source complex is also detected in all SPIRE bands, but the spatial resolution is too low to identify a possible double source. Combining the multi-wavelength photometry of the secondary source, we find that the SED is consistent with a star-forming galaxy at z ∼ 2 with ultraluminous infrared galaxy-like luminosity (L_{8–1000 μm}  ∼ few times 10¹² L_☉).

J1335+3533 (z = 5.90). The optical spectrum of this source is quite unusual as it shows a typical quasar continuum but virtually no Lyα emission (Fan et al. 2006). At longer wavelengths, the quasar is seen in the Spitzer bands, but not in our Herschel data. Wang et al. (2010) report the detection of the CO(6–5) transition and give upper limits on the continuum at 350 μm and 100 GHz. The 250 GHz (2.34 ± 0.50 mJy) and 1.4 GHz continuum was detected by Wang et al. (2007).

J2054−0005 (z = 6.04). This is the only source in our sample for which we do not have Spitzer observations. The SDSS imaging featured a sufficient number of sources that could also be identified on the PACS maps to determine the position of the quasar accurately. The QSO is detected at 250 GHz (2.38 ± 0.53 mJy), but not at 1.4 GHz (Wang et al. 2008a). CO observations revealed the (6–5) transition but no continuum at 98 GHz (Wang et al. 2010). We see the source at 160 μm with PACS, but can only give an upper limit on the 100 μm flux. A faint 3σ–4σ source is visible at 250 and 350 μm. At 500 μm we run into confusion issues with a source located ∼30'' north of the quasar's nominal position, which can also be identified (separate from the quasars) in the other SPIRE bands and at 160 μm. Since our photometry indicates that the FIR peak of the quasar in F_ν falls close to the 250 μm band, we do not expect significant flux in the 500 μm channel. In a NIR spectrum, Ryan-Weber et al. (2009) see a very strong Mg ii absorber at z_abs = 2.598.