3 Discussion

Figure 3: Restframe SEDs compared with the MED of low-redshift quasars compiled by Elvis et al. (1994). Dashed lines show the $1\sigma$ deviation (actually 68 Kaplan-Meier percentile) envelopes. Asterisks denote data taken by Richards et al. (1997) and Schneider et al. (1994). The dot-dashed and dot lines for B 1422+231 shows the MED of radio-loud quasars from Elvis et al. (1994) and the SED of 3C 273 from Lichti et al. (1995), respectively. Note that $3\sigma$ upper limits correspond three times uncertainties listed in Table 3

All the quasars have been detected in the ISOCAM LW2 and LW3 or LW10 band. Combining optical and near-infrared observations, SEDs from the UV to mid-infrared have been obtained for eight quasars at z = 1.4-3.7. Figure 3 shows the restframe SEDs ($L_{\nu}$). The down arrows indicate the $3\sigma$ upper limits obtained with the ISOPHOT observations. Note that $3\sigma$ upper limits in Fig. 3 correspond to three times uncertainties listed in Table 3. The mean spectral distributions (MEDs) of low redshift quasars by Elvis et al. (1994) are also shown by fitting to the ISO $6.7~\mu$m flux density. Note that $6.7~\mu$m in the observing frame is approximately $1.25~\mu$m in the restframe. The dash lines indicate the $1\sigma$ deviation (strictly speaking 68 Kaplan-Meier percentile) envelopes. The rise from the 1 $\mu$m minimum toward 3 $\mu$m is evident at least in z = 3.5 quasar PC 1548+4637. This rise is naturally explained by the sublimation of dust grains at $\sim$$1\,500$ K (e.g., Kobayashi et al. 1993) as expected from the current unified models of active galactic nuclei. Hence, our data suggest that the obscuring torus model could be applied to high-redshift quasars.

The original objective of this work was to study whether there was any evidence for redshift-dependent changes in SEDs from the UV to far-infrared. As shown in Fig. 3, the data from the UV to mid-infrared do not deviate significantly from the MED of the low-redshift sample. However, this cannot be taken as evidence for no SED evolution in the UV to mid-infrared, because our sample is too small in number and too narrow in spectral coverage. Our far-infrared observations only provide $3\sigma$ upper limits, and are not very helpful to check SED evolution in the far-infrared.

The far-infrared detection with ISO have been severely limited by the confusion due to galaxies and local peaks of the IR cirrus (Kawara et al. 1998; Herbstmeier et al. 1998; Matsuhara et al. 2000). Simulating the 90 and 170 $\mu$m source counts in the Lockman Hole, Kawara et al. (2000a, 2000b) concluded that the effect of confusion due to crowded sources (presumably extragalactic) was significant in the flux range below 200 mJy, and 50% of 140 mJy sources at 170 $\mu$m were left undetected due to confusion noise. The expected fluxes of our sample quasars ranges from 7-170 mJy at 170 $\mu$m. Such confusion is clearly seen in Fig. 2. There are a few sources scattering around the quasar, which hampers detection in the far-infrared. It should be noted that the brightest source in the PG 1630+377 field has 106 mJy and the one in PG 1715+535 has 190 mJy (Oyabu & Kawara 2000).

It is clear that larger aperture telescopes like FIRST, which provide spatial resolution finer than ISO, are required for far-infrared studies of high-redshift quasars. In cases where similar aperture telescopes such as IRIS(ASTRO-F) and SIRTF are used, data sampling, namely a step size of raster mapping, should be carefully designed so that far-infrared images can be improved by deconvolution without violating the sampling theorem (Magain et al. 2000).

Acknowledgements

We wish to thank the staff of Vilspa, CTIO, OAO, and the Kiso observatory for their assistance and hospitality. We are grateful to T. Tsuji for his continuous support to this work. We thank the referee, P. Andreani, for valuable suggestions on how to improve this work.