5 Discussion

   
5.1 Evidence for cold cirrus-like dust

The observations show that, though the SEDs usually peak at 60-100 $\mu$m, they can stay on a plateau before the onset of the Rayleigh-Jeans tail. The analysis of Sects. 4.3.1 and 4.3.2 showed that the observed SEDs can be modelled by either one single tepid (50-75 K) blackbody with $\beta \le 2$, partly inferring a high opacity even in the FIR, or by several cool to cold transparent dust bodies with $\beta = 2$. Which of these scenarios (or which combination) might more realistically apply for the ULIRGs?

In the following discussion, we will check for observational evidence for the existence of cold cirrus-like dust in ULIRGs being in line with the multiple blackbody model. One prerequisite to allow a fit of the SEDs with several blackbodies is that $\beta \approx 2$. Even in the case of several blackbodies the opacity cannot be low at all, since the derived dust mass and the size of the emission area exceed other constraints. In order to reduce the amount of dust mass and size, the opacity of some of the blackbodies must be increased (whereby the dust temperature also rises, resulting in a lower dust mass and a smaller region). Thus, we end up with the picture of several blackbodies with $\beta = 2$, which are partly optically thick.

   
5.1.1 Evidence against the single blackbody model

The following arguments can be brought forward against the concept of one single blackbody to describe the FIR-submm SEDs of most of our ULIRGs (i.e all of our mm-subsample except Arp 220):

1)

In general, the optical morphology is disturbed and suggests a variety of patchy dust complexes with less dense regions in between. Since in such a scenario various dust temperatures are also expected, multiple blackbodies appear to provide a better physical description than one single blackbody;

2)

For a variety of objects several authors found observational evidence that $\beta \approx 2$ in the FIR-submm range ( $\lambda > 100$ $\mu$m): was found for interstellar dust in our Galaxy (e.g. Mathis et al. 1983; Lagache et al. 1998), $\beta \approx 2.0$ for normal spiral galaxies (e.g. Bianchi et al. 1999) and the archetypal starburst galaxy M82 (e.g. Thuma et al. 2000), and $\beta = 2.0 \pm 0.2$ for active (Markarian) galaxies (Chini et al. 1989b). In contrary to that, Dunne et al. (2000) found an average value $\beta \approx 1.3$ from their SCUBA Local Universe Galaxy Survey. However, one weakness of this study appears to be the restriction to three wavelength points at 60, 100, and 850 $\mu$m without probing the maximum and the start of the Rayleigh Jeans branch in sufficient detail. One common object between their sample and ours is Arp220. While their single temperature fit provides $\beta = 1.2$ our single modified blackbody fit to a much better sampled SED gives $\beta = 1.7$. Also, the spread of $\beta$ between 1.2 and 2.0 in our ULIRG sample would suggest a large variety of dust properties, like grain size distribution, among the same type of object with probably similar evolution histories. It appears much more plausible to explain the SEDs by a composition of dust components all with consistent emissivity laws of $\beta \approx 2$;

3)

We developed the following quantitative method to check the consistency of the derived $\tau_{ 100~\mu\rm m}$ values from the single blackbody fit with other extinction sensitive quantities. One is strength of the PAH 7.7 $\mu$m feature which is ubiquitous in ULIRGs (Rigopoulou et al. 1999). There must be a sufficient number of UV photons to excite the PAHs, however, an intense starburst is not a prerequisite, as can be seen from the wide distribution of PAH emission in our Galaxy (Mattila et al. 1996) and in NGC891 (Mattila et al. 1999). Mattila et al. (1999) found that the PAH distribution is similar to the one of large dust grains and neutral molecular clouds. It is reasonable to assume that the PAH carriers are mixed with other constituents of the ISM, at least on the spatial scales of kpcs we are looking at with the resolution of our observations. If the PAHs are not cospatial with the dust component emitting the bulk of the FIR emission, then there is a second noticable dust component per se.
In the case of all the FIR emission being irradiated by one single blackbody, we would expect a mixture of the PAH carriers with the FIR emitting dust and a strong decrease of the 7.7 $\mu$m PAH strength (published by Rigopoulou et al. 1999, see examples in Fig. 4 and Table 3) relative to the submm and FIR fluxes with increasing $\tau_{ 100~\mu\rm m}$. This decrease should follow the direction of the A$_{\rm V}$-vector in Fig. 5, (using the galactic extinction curve by Mathis et al. 1983). However, Fig. 5 shows that the normalised 7.7 $\mu$m PAH fluxes are largely independent of $\tau_{ 100~\mu\rm m}$, except for Arp 220 which is further discussed in Haas et al. (2001).

Figure 4: Examples of ISOPHOT-S 5.8-11.6 $\mu$m spectra for the ULIRG Arp 220 and the comparison galaxy NGC23. The dotted line indicates the continuum subtracted for the estimate of the PAH 7.7 $\mu$m peak flux strength. Arp 220 has a strong silicate 9.7 $\mu$m absorption feature (e.g. Dudley 1999), which is accounted for in the continuum estimate.

 

 

Table 3: PAH 7.7 $\mu$m peak flux strength, 100 $\mu$m and 850 $\mu$m continuum fluxes used for the assessment of flux ratios in Fig. 5. Measured fluxes are in Jy. The MIR continuum around the PAH line was subtracted, also the extrapolated synchrotron continuum from the 850 $\mu$m flux. Uncertainties are less than 30%. For the ULIRG sample PAH fluxes are from Rigopoulou et al. (1999), 850 $\mu$m fluxes are from Rigopoulou et al. (1996), Lisenfeld et al. (2000) and from this work. For the sample of normal comparison galaxies PAH fluxes are from this work and 850 $\mu$m fluxes are from Dunne et al. (2000). The 100 $\mu$m fluxes are all from IRAS.

ULIRGs	PAH	100	850
	7.7 $\mu$m	$\mu$m	$\mu$m
05189-2524	0.220	11.73	0.048
UGC5101	0.186	21.24	0.143
12112+0305	0.080	9.73	0.030
Mrk 231	0.289	30.33	0.045
Mrk 273	0.193	24.58	0.077
14348-1447	0.068	7.60	0.016
15250+3609	0.128	5.80	0.027
Arp 220	0.414	126.7	0.744
NGC6240	0.420	28.13	0.137
17208-0014	0.274	35.66	0.119
19254-7245	0.084	5.38	0.029
20100-4156	0.041	5.20	<0.019
22491-1808	0.043	5.06	0.012
23365+3604	0.086	8.00	0.014
23389-6139	0.028	4.33	<0.009
Comparison Galaxies
Arp 148	0.250	10.99	0.092
MGC+02-04-025	0.380	9.60	0.039
Mrk 331	0.650	20.86	0.132
NGC23	0.560	14.96	0.144
NGC695	0.700	13.80	0.136
NGC1122	0.600	15.15	0.084
NGC1667	0.400	16.54	0.163
NGC5256	0.350	10.35	0.082
NGC5653	0.700	21.86	0.205
NGC5713	1.000	36.27	0.359
NGC5962	0.300	20.79	0.317
NGC6052	0.270	10.18	0.095
NGC7591	0.400	13.52	0.135
NGC7592	0.450	10.50	0.108
NGC7674	0.250	7.91	0.108
NGC7679	0.350	10.65	0.093
UGC2238	0.600	15.22	0.104
UGC2369	0.300	11.10	0.072
UGC2982	0.900	17.32	0.176
UGC8387	0.500	24.90	0.113

Figure 5: Normalised 7.7 $\mu$m PAH flux versus opacity $\tau_{ 100~\mu\rm m}$ as formally derived from the single blackbody fits. Different symbols for each spectral type as in Fig. 2. The arrows show the effect of extinction, if the PAH carriers are mixed with the FIR opaque dust. The filled circle and the thick vertical bar on the left-hand side show the range for normal galaxies "NGs'' (see text).

Further clues on the relation between PAH and FIR-submm emitting dust comes from the comparison of the ULIRGs with normal galaxies: we examined the PAH/100 $\mu$m and PAH/850 $\mu$m flux ratios of the ULIRGs with that for normal galaxies, for which both 850 $\mu$m and PAH data are available. Among the SCUBA sample of 104 galaxies obtained by Dunne et al. (2000), 20 sources were also measured with ISOPHOT-S (by several observers) covering a similar area (24 $\hbox {$^{\prime \prime }$ }$) as the SCUBA observations. We derived the PAH7.7 $\mu$m fluxes from the ISO data archive products processed with OLP Version 9.0 (Laureijs et al. 2000). The values are listed in Table 3 and an example for the MIR spectra is shown in Fig. 4. Note, that now we consider only the vertical distribution in Fig. 5 and ignore the horizontal one: the range for the normal galaxies is indicated by the thick vertical bars on the left-hand side. Strikingly the normal galaxies populate a similar PAH/850 $\mu$m range ( $3.9 \pm 2.3$) as the "typical'' ULIRGs ( $3.7 \pm 1.5$, excluding Arp 220), but the PAH/100 $\mu$m range ( $0.03 \pm 0.014$) lies higher than for the ULIRGs ( $0.01 \pm 0.004$). This difference cannot be explained by PAH destruction in the ULIRGs, since it should be reflected in the PAH/850 $\mu$m ratio as well. Neither can this difference be due to extinction, because the shift of the normal galaxies with respect to the ULIRGs must then be larger at 850 $\mu$m than at 100 $\mu$m (see the length and orientation of the extinction arrows in Fig. 5). The difference of the PAH/100 $\mu$m ratio between normal galaxies and ULIRGs can be best explained by the higher dust temperatures of the ULIRGs of about T = 30-60 K compared with T = 15-30 K in normal galaxies, which increases the 100 $\mu$m flux with respect to that at 850 $\mu$m. This behaviour also suggests that the PAH carriers not only trace the dust in active regions, but are even more related to the entire dust content;

4)

We compared the dust mass ( $M_{\rm d}^{\rm\tau free}$ in Table 2) with the molecular gas mass derived from integrated CO luminosities (M(H$_{\rm 2}$) in Table 2). The resulting gas-to-dust mass ratio lies in the range between 500 and 2500, with an average of about $1500 \,\pm \,500$. This is much higher than the canonical value of about 150 for normal galaxies (e.g. Stickel et al. 2000). This discrepancy becomes even higher, when considering the additional contribution of the atomic gas. Since it is unlikely that the CO-to H$_{\rm 2}$ conversion factor for ULIRGs is so much different from that for normal galaxies, a larger dust mass than $M_{\rm d}^{\rm\tau free}$ is required for ULIRGs in order to match the canonical value. Such a suitably high dust mass can be achieved with the combination of several dust components.

In summary, we listed several items where the single blackbody model appears to be insufficient or inconsistent. This can be solved by the application of multiple blackbodies. In order to place several blackbodies underneath the Rayleigh-Jeans tail, $\beta = 2$ seems to provide a reasonable choice.

   
5.1.2 Constraining the FIR opacity for the multiple blackbody model

For a low FIR opacity and several cold to cool blackbodies with $\beta = 2$, the estimated total dust mass $M_{\rm d(total)}^{\rm\beta = 2}$ is considerably higher (on average by a factor of ten) than for the single optically thick blackbody (cf. Table 2). This is because of the contribution from the additional cold component ( $M \propto T^{-6}$ for $\beta = 2$). For the mm-subsample $M_{\rm d(total)}^{\rm\beta = 2}$ lies in the range $\approx$10$^{\rm 8}$-10$^{\rm 9}$ $M_{\odot}$. As mentioned in Sect. 4.5, the uncertainty for $M_{\rm d(total)}^{\rm\beta = 2}$ is quite large, because the decomposition into several blackbodies is not unique. Nevertheless, when comparing the $M_{\rm d(total)}^{\rm\beta = 2}$ for the most reliable fits (marked by "+'' in Table 2) with the molecular gas mass derived from integrated CO luminosities (Table 2), the gas-to-dust mass ratio lies in the range between 30 and 300 with an average of $165 \pm 120$, close to the canonical value of about 150. The multiple blackbody model seems to yield a consistent total dust mass.
Stricter constraints on the dust mass can be obtained, when comparing it with the extension of CO gas inferred from interferometric data (Downes & Solomon 1998). In order to keep the opacity, i.e. column density, low, the extent of the dust region must exceed a minimum size. As a check, we distribute the dust mass associated with the bulk FIR emission, $M_{\rm d(FIR)}^{\rm\beta = 2}$ (Sect. 4.5) evenly within a disk of minimal radius $r_{\tau = 1}$, so that $\tau_{ 100~ \mu\rm m} = 1$ ( $r_{\rm\tau }$ in Table 2). In each object the radius $r(\tau_{\rm 100~\mu m} = 1)$ is smaller than the radius of the optical-NIR image sizes. With the possible exception of Arp 220, the extent of low opacity dust is consistent with the size of the CO disk, where available (from Tables 3 and 4 in Downes & Solomon 1998): Mrk 231:0 $\hbox{$.\!\!^{\prime\prime}$ }$9, Mrk 273:0 $\hbox{$.\!\!^{\prime\prime}$ }$9-3 $\hbox{$.\!\!^{\prime\prime}$ }$1, 17208-0014:1 $\hbox{$.\!\!^{\prime\prime}$ }$8, 23365+3604:1 $\hbox{$.\!\!^{\prime\prime}$ }$0. For $\tau_{ 100~ \mu\rm m} = 0.3$, corresponding to $A_{\rm V} \approx 50$, the minimal radius $r_{\rm\tau }$ is about a factor two larger than listed in Table 2, and exceeds the CO disk sizes. If the dust and the molecular gas are well mixed, then the opacity of the bulk FIR dust component cannot be low ( $\tau_{ 100~ \mu\rm m} \le 0.3$), rather it must be "moderately'' high ( $\tau^{\rm bulk}_{ 100~ \mu\rm m} \approx 1$, corresponding to $A_{\rm V} \approx 150$) in many ULIRGs. Furthermore, with regard to our previous findings on the PAH/850 $\mu$m and PAH/100 $\mu$m ratios, which required relatively low extinction in the MIR (except for Arp 220), we conclude that the bulk FIR dust component cannot contain the majority of the PAH carriers. The relative constancy of the PAH/850 $\mu$m flux ratio with regard to normal galaxies is preserved, since the cold component dominates at 850 $\mu$m, as can be seen from the examples in Fig. 3.

In summary, an emissivity exponent $\beta = 2$ works well for the ULIRGs, as in normal spirals and Markarian galaxies. In some ULIRGs the single component fits already give $\beta = 2$, and for those with $\beta_{\rm fitted} < 2$ several blackbodies with $\beta = 2$ can be used. However, in order to match the dust mass and the size of the emitting regions with the constraints from the CO observations, the opacity of the bulk FIR dust component might be moderately high ( ).

   
5.1.3 The proposed scenario: Cold cirrus dust and cool, partly opaque starburst dust

The previous discussion suggests that the FIR-submm SEDs are composed of several (at least two) $\lambda^{\rm -2}$ modified blackbodies with cool to cold temperatures. Actually, the dust might exhibit a continuous temperature range, but, when applying a formal model, within the uncertainties, a restriction to a few components works best. These dust components represent two stages: (1) the "cirrus'' and (2) "starburst'' stages.

The nuclear regions are probably surrounded by clumpy "Super-Orion'' complexes providing the heating power for the starburst dust. The gaps in between the cloud complexes and the outer regions with lower density and weaker interstellar radiation field host the cirrus-like cold dust. In this picture the PAH emission comes preferentially from the submm emitting dust. Some fraction of the FIR peak dust emission in the nucleus or in dense clouds can be opaque in the FIR.

5.2 Starburst and AGN heated dust

In this section we assess for each optical spectral class (see Table 4) the typical shape of the IR-to-mm SEDs. While the SEDs of AGNs and SBs appear to be indistinguishable at FIR and submm wavelengths, they differ in the NIR-MIR. This suggests that the cool FIR emitting dust is not connected to the AGN, and that the AGN only powers the warm and hot dust. The difference at short IR wavelengths provides a NIR diagnostic tool based on the J-K colours, in order to reveal the presence of an AGN (with red colours, while SBs have blue colours). This tool is successful in the sense that if an AGN is seen by other diagnostics (optical or MIR spectroscopy or X-rays), then in most cases it is also seen via the NIR colours.

 

 

Table 4: Morphologies and spectroscopic classifications. The morphologies are from optical and NIR images as referenced. The extinction E_B-V has been compiled from H $_{\rm\alpha }$/H $_{\rm\beta }$ ratios. The optical spectral types distinguish between Seyfert 1, Seyfert 2, LINER (galaxies with Low Ionisation Nuclear Emission Regions), and HII/starburst types, as classified in the respective references. The MIR spectral types distinguish between AGN=Seyfert1+Seyfert2 and SB=LINER+HII types (Lutz et al. 1999). The NIR classifications are derived from the J-K colours (AGN, SB, SB/n=SB for total galaxy and red 1 $\hbox {$^{\prime \prime }$ }$ nucleus, SB/f=SB for total galaxy and 1 $\hbox {$^{\prime \prime }$ }$ nucleus). The X-ray evidence for AGN is taken from Risaliti et al. (2000).

Name	Morph	EB-V	Optical	MIR	NIR	X-ray
	.	[mag]	Sp-type	type	class	AGN
00199-7426	double m1		LINER? s1		SB
00262+4251	merger m2,*				SB*
00406-3127r1	merger m6		Seyf 2 s6
03068-5346	double?DSS
03158+4227	compactm2					No
03538-6432r1	compactDSS
04232+1436	merger *				AGN*
05189-2524	merger m3	2.03	Seyf 2 s3	AGNa	AGN	Yes
06035-7102	double m1	1.34	HII/SB s1	SB	SB
06206-6315	double m1	2.06	Seyf 2 s1	SB	AGN
12112+0305	double m3	0.92	LINER s3	SB	SB/n
Mrk 231	merger m3	0.70	Seyf 1 s3	AGN	AGN	Yes
Mrk 273	merger m3	1.22	Sy2 s3/LINs8	AGN	SB/n	No
Mrk 463	double m4	0.60	Seyf 2 s4	AGN	AGN	Yes
14348-1447	merger m3	1.15	LINER s3	SB	SB/n
14378-3651	merger m1	1.26	LINER s1		SB
15245+1019	merger m2
15250+3609	ring gal. m3	0.76	LINER s3	SB	SB
Arp 220	merger m3	1.05	LINER s3	SB	SB/n	No
15462-0450	merger m2	0.60	Seyf 1 s5		AGN
16090-0139	merger m1,2	1.55	LINER s1,5		SB
NGC 6240	double m5	1.65	LINER s5	SB	SB/n	Yes
17208-0014	merger m1,2	1.75	HII/SB s1	SB	SB/n	No
17463+5806r1	compactDSS
18090+0130	merger *				SB*
18470+3234	double m2,*	1.13	HII/SB s5		SB*
19254-7245	double m1,2	2.92	Seyf 2 s1	AGN	AGN	Yes
19458+0944	double m2
20046-0623	double m1,2,*		LINER *		SB
20087-0308	merger m1,2	1.43	LINER s1		SB
20100-4156	double m1	1.12	HII/SB s1	SB	SB
20414-1651	merger? m1,2	1.31	LINER s1
ESO 286-19	merger m1	1.07	LINER s1	SB	SB	No
21130-4446	merger m1	0.50	HII/SB s1		SB
21504-0628	merger m1,*		LINER *		SB*
22491-1808	merger m1	0.81	HII/SB s1	SB	SB/f
ESO 148-2	merger m1	1.01	HII/SB s1	SB	SB	No
23230-6926	merger m1	1.49	LINER s1		SB
23365+3604	merger m7,*	0.88	LINER s7	SB	SB*
23389-6139	double m1	3.06	Seyf 2 s1		SB
23515-3127r1	compactDSS		Seyf 2 s6

^r1 z > 0.3; ^* this work, see Appendix; ^e0 Boksenberg et al. (1977), Krabbe et al. (1997); ^m1,s1 Duc et al. (1997); ^m2 Murphy et al. (1996); ^m3,s3 Sanders et al. (1988a); ^m4 Sanders et al. (1988b); ^s4 Véron-Cetty & Véron (1985); ^m5 Fried & Schulz (1983); ^s5 Kim et al. (1995) and Veilleux et al. (1995); ^m6 Clements et al. (1996); ^s6 Allen et al. (1991); ^{m7, s7} Klaas & Elsässer (1991); ^s8 Colina et al. (1999); ^a Clavel et al. (2000).

5.2.1 Seyferts

Our sample contains eight galaxies (with z < 0.3) which have optical Seyfert spectra, i.e. are sources known to house an AGN. Two are Seyfert1s and six are Seyfert2s (see Table 4).

The SEDs of the Seyferts show a variety of shapes in the NIR and MIR range:

$\bullet$

On the one hand, a power-law-like flux increase is seen in the Seyfert1s (Mrk231 and 15462-0450) and the Seyfert2s 05189-2524, 19254-7245 and Mrk463 (and probably also 06206-6315);

$\bullet$

On the other hand, Mrk273 (a LINER nucleus with an off-nucleus Seyfert2 nebula, cf. Colina et al. 1999) has a flat NIR flux plateau with a steep rise at about 10 $\mu$m;

$\bullet$

23389-6139, showing a damped power-law flux increase, appears to be in between these extremes.

The power-law shape might be attributed to the central, relatively unobscured AGN. The red J-K colours might not be caused by extinction only, rather the hottest dust clouds at a temperature just below the evaporation temperature of the dust of about 1500 K, corresponding to a peak wavelength of about 2 $\mu$m, might be seen.

In a torus geometry the flat sources like 23389-6139 could be seen more edge-on, so that the hot AGN heated dust is more hidden. Then the flat NIR flux plateau - with blue colours - may be understood in terms of light scattered at bipolar cones. Alternatively, young hot stars in the host galaxies might contribute significantly to the NIR flux, shifting the colours towards the blue. Also, the AGN could be weak compared with the strength of the starbursts; perhaps Mrk273 belongs to this group.

A somewhat indefinite case is the double source 06206-6315, which is classified by Duc et al. (1997) as a Seyfert2, but as starburst via PAH diagnostics (Lutz et al. 1999; Rigopoulou et al. 1999), although both spectra look quite noisy. The steep optical to NIR 2.2 $\mu$m flux increase argues in favor of a relatively unobscured AGN.

In the FIR and submm range the Seyfert SEDs also exhibit some diversity, though only a relatively moderate one. Examples are Mrk463 and 19254-7245, both with double nuclei (and similar power-law NIR flux increase): while Mrk463 only shows little FIR flux compared to the MIR flux, 19254-7245 has a strong FIR bump and even a considerable amount of mm flux due to cold dust. A similar difference is found for Mrk231 and 05189-2524, both with a "single'' nucleus, i.e. probably mergers in an advanced state: both show a strong FIR bump, but Mrk231 has a steeper Rayleigh-Jeans tail and less cold dust than 05189-2524. Thus, the host galaxies of AGNs may or may not have cold dust.

5.2.2 LINERs and HII/SBs

Our sample contains 11 LINERs (Low Ionisation Nuclear Emission Regions) and 9 HII/SBs. In general, the SEDs look very similar for both optical spectral types. Therefore, we discuss them together here.

The SEDs appear quite homogeneous from the NIR to the FIR, with some diversity in the submm range: all SEDs show a relatively flat NIR flux plateau with a faint 2 $\mu$m bump, PAH emission and/or Si 9.7 $\mu$m absorption features, then a steep rise in flux at about 10 $\mu$m, i.e. a 10 $\mu$m knee. The flux peaks in the FIR at about 60-100 $\mu$m. The only variety appears in the submm range: some sources show a steep Rayleigh-Jeans tail, others exhibit a flatter tail providing evidence for additional cold dust. Typical sources with and without cold dust are 23230-6926 and 23365+3604, respectively, among the LINERs, and 17208-0014 and ESO286-19, respectively, among the HII/SBs.

The homogeneity and extreme similarity of the NIR to FIR SEDs suggests that the dust in LINERs and HII/SBs has similar properties, in particular concerning the spatial distribution and heating mechanism. In fact, this similarity places the LINERs closer to HII/SB than AGN dominated ULIRGs. A similar conclusion (that LINERs are not AGN dominated) was drawn from the PAH diagnostics (Lutz et al. 1999).

5.2.3 The SED shapes and $\mathsfsl{J}$- $\mathsfsl{K}$ colours as diagnostic tool

Figure 6: Two-colour diagrams 2.2 $\mu$m/1.25 $\mu$m versus 25 $\mu$m/60 $\mu$m (top) and versus 120 $\mu$m/200 $\mu$m (bottom). Different symbols for each spectral type as in Fig. 2.

Figure 7: Histogram showing the correspondence of the NIR SED shape classification (via the J-K colour) with the optical emission line diagnostic and MIR (PAH class) spectroscopic classifications.

The NIR-MIR power-law flux increase for the AGN-type ULIRGs and the flat NIR flux plateau with a 10 $\mu$m knee for the SB-type ULIRGs suggests the utilization of these two different SED shapes for a classification scheme. Earlier investigations (e.g. Hill et al. 1988) showed that red NIR colours provide a high probability of finding an AGN among moderately luminous IRAS galaxies. As a quantitative characterisation measure for the SED shapes we use the J-K colours: J-$K \ge 2$ mag identifies power-law shape SEDs; this value corresponds to a 2.2 $\mu$m/1.25 $\mu$m flux ratio of about 2.34.

Although longer MIR wavelengths are less sensitive to extinction, this range suffers from possible confusion by strong PAH emission and 9.7 $\mu$m silicate absorption which complicates the analysis; detailed work has been done by several authors, e.g. Laurent et al. (2000), Imanishi & Dudley (2000) and Tran et al. (2001).

As shown in Fig. 6, the J-K colours (2.2 $\mu$m/1.25 $\mu$m flux ratios) provide a much clearer separation of AGN- and SB-ULIRGs than the MIR colours (25 $\mu$m/60 $\mu$m) used to identify warm AGN-type objects (de Grijp et al. 1985), to say nothing of the FIR colours (120 $\mu$m/200 $\mu$m).

In Fig. 7 we show in histograms the correspondence of this NIR photometry classification scheme with the spectroscopic optical emission line and the MIR PAH diagnostic classification as compiled in Table 4. With respect to the optical classification, we find that the NIR SED shape confirms for 22 out of 22 (100%) the optical classification as a HII/LINER object and for 6 out of 8 (75%) the classification as Seyfert1/2. With respect to the MIR PAH diagnostics, we find that the NIR SED shape confirms for 12 out of 14 (86%) the PAH classification as SB and for 4 out of 5 (80%) the PAH classification as AGN.

These classifications are further confirmed by comparing them with the hard X-ray classifications (Table 4). Thus, the J-K colours represent an attractive tool to explore the nature of ULIRGs and other dust rich IR galaxies. This is of particular advantage, if the galaxies are too distant or too faint for spectroscopy. Then the intrinsic J-K colours can provide constraints on the nature of cosmologically interesting ULIRGs.

All these methods have some limitations, of course, and in some cases provide indefinite classifications. For example, the J-K colour does not reveal any AGN in Mrk273 which is optically classified as Seyfert2 and has a PAH/continuum flux ratio of 1.9 favouring the starburst dominance (Rigopoulou et al. 1999). Also X-ray observations with Beppo-Sax unveiled an AGN in NGC 6240 (Vignati et al. 1999), which is optically classified as a LINER, SB dominated according to PAH diagnostics, and shows no AGN-typical power-law NIR-MIR flux increase (except for the 1 $\hbox {$^{\prime \prime }$ }$ nuclear region).

5.2.4 The nature of the 10 $\mu$m knee

Pure starburst objects have a flat NIR flux plateau followed by a steep rise longwards of 10 $\mu$m. Part of this pronounced 10 $\mu$m knee could be due to the 9.7 $\mu$m silicate absorption. In contrast, galactic compact HII regions housing O5-O9 stars show SEDs with a steep power-law-like rise. As for AGNs this rise already starts at about 1 $\mu$m (Chini et al. 1987). Also the prototype starburst galaxy M 82 has such a SED shape (e.g. Thuma et al. 2000). Thus, a power-law SED could, in principle, also occur in ULIRGs with starburst classification. But this is not observed for our sample.

HST NICMOS images (Scoville et al. 2000) reveal that the central 1 $\hbox {$^{\prime \prime }$ }$ region is redder than the outer regions. Figure 1 shows that for those 9 sources in common with our sample (05189-2524, 12112+0305, Mrk273, 14348-1447, 15250+3609, Arp220, NGC6240, 17208-0014 and 22491-1808) most of the nuclear SEDs (marked with "$\times$'' symbols) exhibit - even for LINERs and HIIs - a power-law-like flux increase in the NIR. J-K colours of these nuclear regions are similar to the colours in larger apertures found for the AGN-type ULIRGs. For 05189-2524 the NICMOS fluxes (hardly to recognize in Fig. 1) are practically identical with the ground based fluxes showing that this source is completely dominated by the central point-like source. For the others the much higher and bluer ground-based fluxes suggest that the sources are extended on the 5-10 $\hbox {$^{\prime \prime }$ }$ scale, in particular for 17208-0014 and 22491-1808 for which already the NICMOS images indicate that the SEDs are composed of many different components.

With regard to the Balmer decrement LINERs and HII-starburst galaxies have low E_B-V (Table 4), as found for Seyfert1s. The highest E_B-V are those for Seyfert 2s, indicating that their emission regions are the most deeply embedded ones. A deeply embedded HII region should suffer from extinction in the same way. However, deeply embedded young stars will probably be outshone by those located closer to the outer surface of the galaxy and will not dominate the NIR colours. In conclusion, the relatively flat blue NIR flux plateau comes from the practically unobscured outer regions. The strong flux rise longwards of 10 $\mu$m is caused by more deeply embedded luminous star forming regions. This naturally explains the appearance of the 10 $\mu$m knee.

5.2.5 Does the AGN power the FIR emission of ULIRGs?

It is now widely accepted that a powerful (and not obscured) AGN creates a strong NIR-MIR emission (e.g. Sanders et al. 1988a, 1988b; Pier & Krolik 1992, 1993; Rowan-Robinson 1995; Haas et al. 2000a). However, it is still a matter of debate whether the FIR/submm emission in ULIRGs (as well as quasars) is also mainly powered by the AGN (e.g. Sanders 1999) or by circumnuclear SBs (e.g. Rowan-Robinson 1995; Genzel et al. 1998).

The power-law flux increase of Mrk231 appears similar to that of the quasars PG0050+124 and PG1613+658 (Haas et al. 2000a). For these quasars it is not clear whether and in which wavelength range a starburst contributes to their power-law SEDs: energetically, the AGN radiation would be sufficient to heat the dust, and since pure starbursts have different SED shapes (with knees, as shown above), the superposition of a significant starburst component would probably destroy the smooth power-law shape. For Mrk231, however, the FIR luminosity exceeds the MIR- and the (dereddened) UV-optical-luminosity (Downes & Solomon 1998). Downes & Solomon (1998) found two rotating gas disks with radii of 460 and 1150 pc, respectively, and concluded that the outer one is starburst dominated, since it is too extended and FIR-luminous to intercept sufficient power from the central region. Although the AGN is clearly visible, it dominates neither the FIR luminosity nor the total energy output. However, the PAH/continuum diagnostics indicates a dominating AGN in Mrk231 (Genzel et al. 1998; Lutz et al. 1999). The reason for this might be that the unobscured AGN continuum diminishes the relative PAH strength, mimicking a weak starburst and overemphasising the role of the AGN. Thus, it is necessary to consider not only the ratio PAH/continuum, but also the absolute PAH and continuum levels, both of which may be affected by different amounts of extinction due to different sites of origin in the galaxies. For the diagnostics of moderately luminous Seyfert 1s and Seyfert 2s this has already been stressed by Clavel et al. (2000).

As shown in Fig. 6, the ULIRGs with and without AGN signatures cannot be distinguished via their FIR colours ( $F_{\rm 120~\mu m}$/ $F_{\rm 200~\mu m}$). This suggests that either the FIR/submm emission is largely independent of the presence of an AGN, or that every ULIRG not classified as AGN contains a hidden AGN. The relative high number of ULIRGs with signs of strong starbursts versus ULIRGs with AGN signatures in our sample (22:10 from optical emission line diagnostics, 13:5 from PAH diagnostics) would argue in favour of the first alternative. Since in Sect. 5.1.2 we concluded that the bulk FIR component might be moderately opaque ( $\tau^{\rm bulk}_{\rm 100~\mu m} \approx 1$) even IR diagnostics, in particular the PAH/850 $\mu$m flux ratio, have limited meaningfullness in this context. An exception seems to be Arp220 where the dust geometry leads to a strong deficiency of PAH strength versus 850 $\mu$m flux and dereddening would yield a quasar-like IR continuum (Haas et al. 2001). Again, this is only 1 out of 22 sources of the submm ULIRG subsample showing this behaviour. Therefore, we see some evidence from our sample that the role of the AGN for powering the FIR/submm emission in nearby ULIRGs is negligible.

5.3 Cold dust and search for evolutionary trends

ULIRGs may well have considerable amounts of cold dust at 10-30 K, a temperature typical for dust in less active spiral galaxies. If the cold dust (from the parent galaxies) is continuously heated during the merging process, then we would expect more advanced mergers to show a relatively larger amount of warm dust than mergers in a beginning phase. To check this hypothesis, we use the morphological appearance and the separation of the galaxy nuclei as a measure for the merger state. This is a simplification, since the encounter might not lead to a monotonous approach of the two galaxies. According to simulations by Dubinski et al. (1999) the merging process contains repeated approaches with semi-elastical collisions followed by a drift apart. Nevertheless, we divided the sources into (1) single and (2) double (see Table 4). We use the MIR, FIR and submm luminosities as a measure for the amount of warm, cool and cold dust, respectively. Figure 8 does not show any trend of MIR-, FIR- and submm-luminosity ratios with the simple morphological classifications. Within our sample, which covers only a small range in bolometric luminosity, the relative amounts of warm, cool and cold dust are not correlated with the merger state inferred from the morphology.

Figure 8: Distribution of the MIR/(FIR+submm) luminosity ratio versus total IR luminosity (top) and FIR/submm luminosity ratio versus FIR+submm luminosity (bottom). Different symbols for each spectral type as in Fig. 2. The symbol size indicates sources with either a single nucleus (small and filled or thick) or a double nucleus (large and open or thin).

A similar independence was found between the strength of the PAH-feature/MIR-continuum and the (projected) separation of the galaxy nuclei (Lutz et al. 1998; Rigopoulou et al. 1999). Either the (projected) separation of the nuclei is not well suited to measure the progress of the merging process, or the dust heating does not steadily increase during the merging. It seems likely that while the merging nuclei approach each other relatively slowly and/or repeatedly, the dust clouds undergo several phases of compression, turning into cold proto-starburst clouds which subsequently undergo heating by starbursts (and by an AGN, if present). These distinct phases could occur simultaneously in a merger, but at different locations. Photometry of the entire galaxy then shows a mixture of dust complexes at different phases. This is actually revealed by spatially resolved submm observations of the colliding IR luminous (not yet ultra-luminous) galaxy pair NGC4038/39 (Haas et al. 2000b).

Figure 9: Two colour diagram (25 $\mu$m/60 $\mu$m versus 150 $\mu$m/850 $\mu$m) of ULIRGs. Different symbols for each spectral type as in Fig. 2. Note the relation between this diagram and that in Fig. 2 ($\tau$ versus $\beta$): 25 $\mu$m/60 $\mu$m is related to 1/$\tau$ and 150 $\mu$m/850 $\mu$m to $\beta$.

Irrespective of the morphology, the $F_{\rm 25}$/ $F_{\rm 60}$ versus $F_{\rm 150}$/ $F_{\rm 850}$ two-colour diagram (Fig. 9) illustrates the distribution of our ULIRG mm-subsample. Schematically, the cold cirrus-like dust is represented by the 850 $\mu$m emission, the cool SB dust by the 150 and 60 $\mu$m emission, and the warm AGN dust by the 25 $\mu$m emission. In this diagram trends are recognizable: strong AGNs are located in the upper half ( $F_{\rm 25}$/ ), SBs in the lower half. The galaxies with and without cold cirrus-like dust lie towards the left and right side, respectively. This diagram illustrates that despite similar luminosities the nearby ULIRG sample exhibits quite some diversity. Whether or not this is due to evolution is still a puzzle.

Finally, it is interesting to note that the four ULIRGs at medium redshift ( $z \approx 0.3$) have similar dust temperatures as the low redshift ULIRGs, but exhibit the highest FIR luminosities among the sample. While their MIR SEDs and their spectral types can be determined soon, the question of whether they also contain cold dust will have to be left to future observations with the Herschel Space Observatory or the Atacama Large Millimetre Array.