All objects with measurable [Ne V] in our sample have [Ne V] ratios indicating a low density ( $n_{\rm e} = a$ few 100 to a few 1000 cm^-3), well below the critical densities ( $\approx$ $5 \times 10^4$ and $5 \times 10^5$ cm^-3), see Fig. 1. The average ratio is $1.1\pm0.4$ for the (4) Seyfert 1s, and $1.3\pm0.3$ for the (6) Seyfert 2s, i.e. NLR densities for the two types agree for this fairly small sample, consistent with unification.

$\begin{figure} \par\resizebox{8.8cm}{!}{\includegraphics{H3806F2.eps}}\par\end{figure}$

Figure 2: Temperature-sensitive ratio of the 3426 Å and 24.3 $\mu$ m lines of [Ne V] as a function of electron density and temperature. The dashed line indicates the ratio for NGC 1068 and the dotted lines its uncertainty range considering measurement errors and uncertainty of extinction.

5.2 Temperature

$\begin{figure} \par\resizebox{8.8cm}{!}{\includegraphics{H3806F3.eps}}\end{figure}$

Figure 3: Temperature-sensitive ratio of the 6087 Å and the 9.53 $\mu$ m (top) and 7.81 $\mu$ m (bottom) lines of [Fe VII], as a function of electron density and temperature. The dashed line indicates the ratio for NGC 1068 and the dotted lines its uncertainty range considering measurement errors and uncertainty of extinction.

One interesting species is [Ne V] (lower ionisation potential 97 eV) which has very bright optical and mid-infrared transitions. Figure 2 shows the ratio predicted for the 3426 Å and 24.3 $\mu$ m lines, using the atomic data of Lennon & Burke (1994) and Nussbaumer & Rusca (1979). The range of allowed values for NGC 1068 (combining our data with optical data of Marconi et al. 1996 and adopting A_V=0.71) is also indicated, adding a 20% measurement error for each line, and a 0.25 mag uncertainty in A_V to produce the maximum deviation. These uncertainties (particularly the one on extinction) lead to a wide range of electron temperatures ( $\sim$ 17 000-45 000 K) being consistent with the NGC 1068 data for its density of 2000 cm^-1.

Another high excitation species offers the potential of reducing the extinction effects. The brightest optical line of [Fe VII] is found at the longer wavelength of 6087 Å. With a lower ionisation potential of 99 eV, [Fe VII] samples a region similar to [Ne V]. The ratio of the optical [Fe VII] 6087 Å line to the mid-infrared [Fe VII] lines is little sensitive to electron density in the regime determined above, and forms a diagnostic of electron temperature in the NLR. Figure 3 shows the ratios from solving the rate equation using the atomic data of Berrington et al. (2000, and priv. communication). We have observed the mid-IR [Fe VII] lines in NGC 1068 and Circinus. The extinction corrected line ratios for NGC 1068 lead to very different electron temperatures: $\approx$ 43 000 K from 6087 Å/9.53 $\mu$ m but $\approx$ 16 000 K from 6087 Å/7.81 $\mu$ m. This discrepancy is also reflected in the extinction corrected ratio of the 9.53 $\mu$ m and 7.81 $\mu$ m lines which is 1.4, significantly different from the $\approx$ 3-3.5 expected for a wide range of conditions from the atomic data. For Circinus, the extinction corrected ratio of the 9.53 $\mu$ m and 7.81 $\mu$ m lines is similarly low, about 1.1. There is no immediate explanation for this inconsistency between mid-IR FeVII fluxes that is observed independently in two sources. An explanation by extinction uncertainties is unlikely, since $\approx$ 10 mag of additional visual extinction would be needed to selectively weaken the 9.53 $\mu$ m line which is inside the silicate feature. Uncertainties in the atomic data might be another possibility but are only partially supported by observations of the same lines in the planetary nebula NGC 7027: Salas et al. (2001) observe a ratio of 1.84 for the two mid-IR lines, closer to the value of 2.73 expected for the conditions in this nebula. We summarize that, while there are indications for high (20 000-30 000 K) electron temperatures in the part of the narrow line region sampled by $E_{\rm ion}\sim100$ eV species, there is considerable uncertainty on extinction corrections and atomic data which prevents firm conclusions.

$\begin{figure} \par\resizebox{8.8cm}{!}{\includegraphics{H3806F4.eps}} % \end{figure}$

Figure 4: Line ratio diagram of typical NLR tracers. Diamonds: Seyfert 1s. Stars: Seyfert 2s, Plus signs: NLXGs. Numbering of galaxies as in Fig. 1. The supernova remnant RCW103 (Oliva et al. 1999) is shown as a triangle. The [Ne VI]/[O IV] values for three additional sources are indicated as horizontal arrows on the right hand y-axis (from top to bottom: 3C120, Mkn 509, Mkn 573). To constrain ionization parameter and density we have overplotted a model grid by Spinoglio et al. (2000) for a power law ionizing continuum of index $\alpha = -1.0$ : solid lines are log U = -2.5, -2, and -1.5, dashed lines are for n = 10², 10³, and 10⁴.

Table 5: Classes of line profile asymmetries.
Class I Class II Class IIIa Class IIIb

optical [O III] profile symmetric asymmetric asymmetric asymmetric

MIR [O IV] profile ${\rm symmetric, =optical}$ symmetric ${\rm asymmetric, =optical}$ asym./non-gauss., $\neq{\rm optical}$

examples 3C 120¹, PKS 2048² Mkn 509², NGC 5643² NGC 3783²,NGC 4151,³ NGC 1068³,Mkn 1¹, Mkn 3³,

NGC 7469¹, To l0109² Mkn 463², NGC 5506³

¹ Optical line profile from Vrtilek & Carleton (1985).
² Optical line profile from Whittle (1985).
³ Optical line profile from Veilleux (1991).

**Table 5:** Classes of line profile asymmetries.
	Class I	Class II	Class IIIa	Class IIIb
optical [O III] profile	symmetric	asymmetric	asymmetric	asymmetric
MIR [O IV] profile	${\rm symmetric, =optical}$	symmetric	${\rm asymmetric, =optical}$	asym./non-gauss., $\neq{\rm optical}$
examples	3C 120¹, PKS 2048²	Mkn 509², NGC 5643²	NGC 3783²,NGC 4151,³	NGC 1068³,Mkn 1¹, Mkn 3³,
			NGC 7469¹, To l0109²	Mkn 463², NGC 5506³

5.3 Excitation

Varying relative line strengths of low and high ionizaton lines, e.g. of the Neon sequence, seem to indicate a large variation in excitation among the galaxies in our sample. Line ratios from ions in different stages of excitation along with a photoionization model can be used to reconstruct the NLR radiation field produced by the central ionizing source. Mid-Infrared lines are particularly suited because they are little sensitive to extinction and electron temperature, and because they span a wide range of ionization potentials. In detailed studies we have analyzed three nearby Seyfert nuclei of our sample with elaborate photoionization models: Circinus (Seyfert 2 plus starburst), NGC 4151 (Seyfert 1.5) and NGC 1068 (Seyfert 2), see Moorwood et al. (1996), and T. Alexander et al. (1999, 2000). These models were able to reconstruct the intrinsic spectral energy distribution (SED) of the ionizing source in the extreme UV, where a "Big UV Bump'' around 100 eV is expected as a signature from a hot thin accretion disk. The results are consistent with such a "Big UV Bump'', but also suggest for some sources of both types (Seyfert 1 and 2) the presence of neutral absorbers between the AGN's extreme ultraviolet emitting source and the NLR. Such (UV) absorbers have been suggested independently by studies of UV absorption lines (Kriss et al. 1992; Kraemer et al. 1999, 2001). This detailed photoionization modeling requires a large number of mid-IR lines (of good S/N) supplemented with UV/optical/near-IR lines. For most of the galaxies in our sample the compilation of lines from ISO and the literature is not as complete as for the three examples given above. Hence, we do not attempt a similar modeling for them. Instead, we try to constrain the NLR excitation by directly comparing our observations to standard photoionization models from the literature.

The main input parameters for photoionization models are the electron density $n_{\rm e}$ and the ionization parameter U, i.e. the number of ionizing photons per hydrogen atom at the inner face of the ionized cloud. The ratio of lines from ions of similar ionization potential but with different critical densities (e.g. the [Ne V] 14/24 ratio) are good tracers of the electron densities (see Sect. 5.1). Vice versa, the ratio of lines with similar critical density but from ions of different ionization potential (like [Ne VI]/[O IV]) is sensitive to the ionization parameter. In Fig. 4 we have used this to construct a diagram to constrain $n_{\rm e}$ and U. We have chosen the lines of [Ne V], [Ne VI] and [O IV] because they are not affected by photoionization by stars and because they are generally among the brightest high ionization lines. For comparison we show in Fig. 4 the location of the supernova remnant RCW103 (Oliva et al. 1999), as an example for a strong shock source, and a photoionization model grid taken from Spinoglio et al. (2000). This model grid was computed for a power law with an ionizing continuum of index $\alpha = -1.0$ and for various ionization parameters and electron densities. We draw two conclusions from this comparison: firstly, all galaxies are consistent with these standard photoionization models, with average ionization parameters log U between -1.5 and -2 and (as seen already in Sect. 5.1) average densities between a few 100 and a few 1000 cm^-3. Such a simple comparison can not, however, distinguish between simple power laws and more complex "Big UV Bump'' models. As noted earlier, the detailed modeling of single sources required for such an analysis is outside the scope of this paper. Secondly, within our (small) sample, we do not see significant differences between the AGN sub-types.

5.4 NLR line profiles

Structures and velocity fields in the NLR can be studied by an analysis of line profiles. Numerous studies of the emission line profiles in Seyfert galaxies have shown that in many cases the lines exhibit blueward asymmetries and are blueshifted with respect to the galaxies systemic velocity. The common interpretation for these profile asymmetries is that they are caused by differential extinction in an outflow or inflow of clouds with a modest amount of mixed-in dust. Observations in the infrared can obviously test these scenarios, because infrared lines suffer more than an order of magnitude less extinction than in the optical. Hence, they should not or only marginally show asymmetries. Sturm et al. (1999) and Lutz et al. (2000b) have presented the first such studies of optical-to-mid-IR line profile comparisons for NGC 4151 and NGC 1068. In these two sources the outflow-plus-dust scenario seems to be (at least partially) wrong, since the MIR lines show asymmetries similar to their optical counterparts.

$\begin{figure} \par\resizebox{8.8cm}{!}{\includegraphics{H3806F5.eps}}\end{figure}$

Figure 5: Comparisons between optical and MIR line profiles (continuum subtracted and normalized to peak flux density), for four different cases as discussed in the text. Solid lines show the MIR [O IV] 26 $\mu$ m line, taken from the ISO-SWS data set. Dashed lines are optical [O III]5007 Å lines, smoothed to the SWS resolution. References for the optical line profiles are given in Table 5. The velocity scale on the x-axis (in [km s^-1]) is relative to the systemic velocities given in Table 1.

Optical [O III] line profile information exists for many objects in our data set. These can be used for a comparison with the MIR [O IV] line, as described in Sturm et al. (1999). We have convolved the optical line profiles with the SWS instrumental profile (Gauss profiles with FWHM as given in Table 2) in order to smooth these profiles to the resolution of SWS. Four such examples of an optical-MIR comparison are shown in Fig. 5. The objects in our sample can be grouped roughly into three different classes in terms of line profile asymmetries and agreement of MIR lines with their optical counterparts, as summarized in Table 5. In some cases (Class I) the optical line profile of [O III] is quite symmetric and of Gaussian shape. In these cases, not surprisingly, the MIR and optical line profiles match each other very well. Another group of objects (Class II) shows strong blueward asymmetries in the optical, while the MIR lines are rather symmetric. This is exactly what scenarios with infall our outflow of NLR clouds with mixed-in dust predict. We only found two such cases in our sample. The third class of objects has asymmetric optical lines, but MIR lines which are inconsistent with these scenarios. This class can be further divided into two sub-classes: in Class IIIa the MIR lines are asymmetric, too, and agree well with the optical lines. This case has been studied in more detail in an analysis of NGC 4151 by Sturm et al. (1999), and can be explained, for instance, by a true asymmetry in the distribution of the NLR clouds, or, in the case of NGC 4151, by a central, optically very thick, but geometrically thin absorber on parsec scales. In Class IIIb the NIR lines are asymmetric (or symmetric but with non-Gaussian profiles), but different from their optical counterparts. One member of this class is NGC 1068, which has been analyzed by Lutz et al. (2000b). This suggests that parts of the NLR are significantly obscured in the optical, but not enough to also block the MIR lines. Similar to Class IIIa, the remaining MIR profile asymmetries may be either due to an intrinsic asymmetry of the NLR, or due to a very high density obscuring component which is hiding part of the NLR even from infrared view. We note that for some of the objects with MIR lines of good S/N (Cen A, M 51, MKN 573, NGC 1365, and NGC 7582) there is, to our knowledge, no (suitable) optical line profile information available in the literature.

Many objects in our sample exhibit differences in the profiles of lines with different ionization potential. For instance, the Circinus galaxy shows symmetric low ionization lines, but asymmetric high ionization lines with the typical blueward asymmetries. Vice versa, in NGC 7582 high ionization lines are very symmetric, while the low ionization lines are strongly asymmetric. Correlations, as well as anti-correlations, of line asymmetries with critical density and/or ionization potential have been claimed in many publications. For NGC 7582 a contribution to the low ionization lines from a starburst component with asymmetric spatial distribution could be an additional/alternative solution. For the Circinus galaxy the situation is even more complex, since Oliva et al. (1994) reported asymmetric low ionization lines. Finally, for some objects in our sample all MIR line profiles appear to be quite similar. For instance, the profiles of Cen A, M 51, MKN 573, NGC 1365 (for which no optical counterparts exist) are quite symmetric and of Gaussian shape, regardless of ionization potential or critical density.

It appears that there is no unique answer to what causes the line profile asymmetries in Seyfert galaxies. Extinction by dust on different spatial scales and with varying column densities, sub-structure and true asymmetries in the spatial distribution of NLR clouds may all play a role with varying degrees of importance. For some objects the assignment to a certain group is not unique and depends also on S/N. We use this classification purely to obtain an overview of this complex issue. We refrain from a more detailed analysis of line profiles in this paper because it requires a careful study of the effects of aperture differences (spatial resolutions) between the optical and MIR observations, and, in many cases, a larger data set of high ionization lines with good S/N.

$\begin{figure} \par\resizebox{8.8cm}{!}{\includegraphics{H3806F6.eps}} % \end{figure}$

Figure 6: The ratio [Ne VI]/[O IV] vs. [Ne VI]/[Ne II]. Same symbols and numbering as in Fig. 1. A typical error bar (according to a 20% individual line flux error) is shown in the upper left corner. An empirical dividing line is drawn as dashed line. Composite sources according to their PAH spectra are encircled. For galaxies #26, 28 and 29 no ISO PAH spectra exist.

5 Physical conditions in the NLR

5.1 Density

5.2 Temperature

5.3 Excitation

5.4 NLR line profiles