3 Observations and data reduction

   
3.1 ISOPHOT 10-200 $\mu$m data

The observations were carried out with ISOPHOT (Lemke et al. 1996), the photometer on board ISO (Kessler et al. 1996). A description of the ISOPHOT observation templates (AOTs) is given in Laureijs et al. (2000). The MIR 10 to 25 $\mu$m observations were done in triangular chopped mode (AOT P03) with a 52 $\hbox {$^{\prime \prime }$ }$ aperture and a chopper throw of 60 $\hbox {$^{\prime \prime }$ }$ which matched the compact size of the targets well. The FIR 60 to 200 $\mu$m observations were performed with the array cameras C100 and C200 in the "sparse map'' off-on mode (AOT P37-39) with background positions 3-4 $\hbox{$^\prime$ }$ north of the source. Beam sizes used for photometry are given in Table 1. The four faint sources at higher redshift were observed in mini-map mode (AOT P22) in a $4 \times 2$ raster and restricted to the filters at 120, 150, 180, and 200 $\mu$m.

The data were reduced using the PHT Interactive Analysis tool (PIA) (V7.3.3e) in standard processing mode, together with the calibration data set V4.0 (ISOPHOT Data User Manual V4.0, Laureijs et al. 1998). This includes correction for non-linearity of the electronics, deglitching (removal of data disturbed by cosmic particle events), and correction for signal dependence on the reset interval time. To handle the signal transients in the time series of bright FCS illuminations, only the last half was taken, when the values approached the final signal level. The calibration of detector responsivity and its changes was performed using associated measurements of the thermal fine calibration source (FCS) on board.

For the chopped measurements with the P1 and P2 detectors at wavelengths 10-25 $\mu$m we inspected the sequence of chopper plateaux and removed outliers, with the criterion that the uncertainty was larger than the average three sigma of the measurements. By this procedure typically the first chopper plateau of a series and those with residual glitches were removed.

Finally the fluxes were corrected for aperture/beam size effects. For the $3 \times 3$ pixel C100 array (60 and 90 $\mu$m) the fluxes were derived with two methods using (1) only the central pixel (46 $\hbox {$^{\prime \prime }$ }\times 46\hbox {$^{\prime \prime }$ }$) and (2) the whole array (138 $\hbox{$^{\prime\prime}$ }\times 138\hbox{$^{\prime\prime}$ }$). Both methods yield essentially the same flux (with some higher noise for the whole array), providing evidence that the objects are point-like and not resolved in the FIR.

The error propagation in ISOPHOT data reduction is described in Laureijs & Klaas (1999). The statistical errors derived from signal processing are about 5-20%, depending on the wavelength range and object brightness (see caption to Table 1), but systematic errors due to absolute calibration accuracy are estimated to be 30% (Klaas et al. 1998b). To account for the overall uncertainty in the signal derivation as well as relative and absolute photometric calibration we have adopted a general photometric uncertainty of 30%.

   
3.2 SCUBA 450 and 850 $\mu$m data

The 450 and 850 $\mu$m observations were obtained on July 1st and 3rd, 1999, using the Submillimetre Common User Bolometer Array (SCUBA, Holland et al. 1999) at the James Clerk Maxwell Telescope (JCMT) on Mauna Kea, Hawaii. In general, photometry with 1 $\hbox{$^\prime$ }$ chopper throw, mini-jiggle and beam switching was performed. We used the SCUBA narrow band filters in order to minimize possible CO line contributions (see also Sect. 4.1). The atmospheric transmission was determined every hour from measurement series at various zenith distances (sky dips). The conditions were excellent and stable ( $\tau_{850~\mu{\rm m}} = 0.134 \pm 0.002$ on July 1st and $0.209 \pm 0.003$ on July 3rd, respectively). $\tau_{450~\mu{\rm m}}$ was extrapolated from the 850 $\mu$m skydips ( $\tau_{450~\mu{\rm m}} = 0.65 \pm 0.01$ on July 1st and $1.09 \pm 0.02$ on July 3rd, respectively). Mars and Uranus served as standard calibrators, and repeated photometry measurements gave a reproducibility of better than 5% and 14% at 850 $\mu$m and 450 $\mu$m, respectively. The observing time per source was 10-20 min (actually the dome carousel driver was out of order, so that the targets could only be observed during their oblique rising passage across the dome slit pointing towards south-east). The data were reduced using the SCUBA User Reduction Facility (SURF) with special emphasis on identifying noisy bolometer pixels and removing the sky noise determined from off-source pixels.

In order to check for possible extended submm flux, jiggle maps covering 2 $\hbox{$^\prime$ }$ were obtained for three sources. These sources were NGC 6240 and 17208-0014, both revealing a submm flux excess (as discussed below), and 20046-0623 providing the same good upper limits for different beams. None of the sources seems to be extended at the spatial resolution of about 8 $\hbox {$^{\prime \prime }$ }$ and 15 $\hbox {$^{\prime \prime }$ }$ FWHM, respectively. These test cases suggest that for the other sources the bulk of submm flux is also contained within our photometry beams and that some possible extended submm flux is small and lies within the total adopted uncertainties of about 20-30%, in particular since NGC6240 and 17208-0014 lie in the low redshift range of our sample.

   
3.3 SEST 1300 $\mu$m data

The $1300~\mu$m observations were obtained between July 4th and 9th, 1999, using the MPIfR single channel facility bolometer (Kreysa 1990) at the Swedish ESO Submillimetre Telescope (SEST) on La Silla, Chile. The system provides a beamsize of 24 $\hbox {$^{\prime \prime }$ }$ and was used in the ON-OFF mode with dual beam switching and a beam separation of 68 $\hbox {$^{\prime \prime }$ }$ in azimuth. Pointing and focus were checked at least every two hours and the pointing accuracy was always better than 4 $\hbox {$^{\prime \prime }$ }$. The atmospheric transmission as determined by sky dips was stable at about $\tau_{1300} \approx 0.1$. Uranus served as a standard calibrator. The observing time per source was about 1 hour (on-source). From repeated calibrator measurements we estimate the absolute accuracy to be about 20-30%.

Figure 1: Spectral energy distributions of ultra-luminous IR galaxies, ordered along the columns by RA. Upper limits are indicated by a downward arrow and in the case of ISO, SCUBA and SEST measurements in addition by open symbols. The redshifts and, if known, the optical spectral types are listed. The wavelength and frequency ranges are as observed and not corrected with regard to the rest frame of the objects. The dotted lines represent modified blackbody fits. Emissivity exponent $\beta$ and opacity $\tau_{\rm 100~ \mu m}$ are listed in the plots. The resulting temperatures T are corrected for redshift. For those sources with no submm/mm data available we used an average $\beta = \langle$1.6$\rangle$. The solid lines indicate the shapes of the SEDs, dash-dotted lines are used where they are uncertain. At NIR-MIR wavelengths the SED shape is obtained by connecting the data points, and at long FIR and submm wavelengths it is determined by the modified blackbody fit. Luminosities were derived by integrating the SEDs between 10 and 1000 $\mu$m. Dashed lines in the mm range indicate synchrotron extrapolations from cm observations. The literature data are from Sanders et al. (1988a,b), Crawford et al. (1996), Marx et al. (1994), Murphy et al. (1996), Rigopoulou et al. (1996), Duc et al. (1997), Rigopoulou et al. (1999) and Lisenfeld et al. (2000). For some sources the NIR fluxes of the central regions are plotted with a "$\times$'' (Diameter = 1 $\hbox {$^{\prime \prime }$ }$, from Scoville et al. 2000, except 5 $\hbox{$.\!\!^{\prime\prime}$ }$5 for 18090+0130 from Hill et al. 1988).

Figure 1: continued.

Figure 1: continued.

Figure 1: continued.

   
3.4 MAGIC 1.2 and 2.2 $\mu$m data

J and K-band images were obtained at the Calar Alto 2.2 m telescope on August 21 and 22, 2000, using the $256 \times 256$ pixel NIR camera MAGIC (Herbst et al. 1993) with a pixel scale of 0 $\hbox{$.\!\!^{\prime\prime}$ }$6416. The observations were performed in standard dithering mode with a total exposure time per source of about 10 and 30 min in J and K, respectively. The seeing was about 2 $\hbox {$^{\prime \prime }$ }$ preventing the identification of spatial details in the sources (see Appendix A). But the conditions were photometric, in particular during the second night. The UKIRT standard stars FS2, FS27 and FS35 were observed for flux calibration. The data reduction followed the standard procedures. Based on the cross calibration of the standard stars, we estimate the photometric accuracy to be about 15%. In the case of IRAS18090+0130 the uncertainty is somewhat higher due to superposition of two stars.