In Fig. we have plotted the FWHMs (corrected for the instrumental broadening) of each individual component, i.e. of each set of lines used to fit the blue and red spectra, as listed in Table (cols. 5 and 10, respectively). The good correlation found between the blue and red FWHMs gives confidence in the fitting analysis.
Figure shows the log(5007/H) vs. log(6583/H) and log(5007/H) vs. log(6300/H) diagrams traditionally used to classify nuclear emission-line regions into H II regions, Liners or Seyfert 2s. We have delimited in the two diagrams three regions, each corresponding to one of these classes. In Figs. a and b we have plotted all objects for which line ratios are available in the literature and which are unambiguously classified as H II regions (crosses), Seyfert 2s (open circles) or Liners (open squares); we have also plotted the 61 observed objects suspected of having a ``transition'' spectrum (filled circles): they fall, at least in one of the diagrams, in a ``zone of avoidance'', i.e. outside the regions arbitrary assigned to the classical emission-line regions. In figures c and d, which are the same as a and b respectively, we have plotted the individual components used to fit the spectra, as given in Table .
It is apparent that most of the ``transition objects'' belong to one of the three following categories:
Fig. is the histogram of the parameter log (6300/5007) for 159 Seyfert 2s and Liners after correction of the line fluxes for reddening, assuming that the intrinsic Balmer decrement is /= 3.1 (Osterbrock & Dahari 1983) [Binette et al. (1990) suggested an even higher value for the intrinsic Balmer decrement in AGNs: / = 3.4]. The histogram has two main peaks showing a clear separation between strong [O III]5007 objects (Seyfert 2s) and weak [O III]5007 objects (Liners). Although our sample is heterogeneous and incomplete, this suggests that there is no continuity between the two classes of objects. Heckman (1980) originally defined Liners as objects with 63005007 > 0.33; it seems that 63005007 > 0.25 would be a more realistic definition, as the observed distribution of this ratio really shows a minimum centered around this value.
According to Ho et al. (1997a), the separation between the two principal ionization sources (young stars vs. AGNs) and between the two AGN excitation classes (Seyfert 2 vs. Liners) does not have sharp, rigorously defined boundaries. Fig. shows that this is not the case. In fact, the three areas containing the H II regions, the Seyfert 2s and the Liners are clearly separated; almost every ``transition object'' turns out to be ``composite'' when observed with sufficient resolution.
Several authors had already suspected this to be the case. Kennicutt et al. (1989) and Ho et al. (1997c) have shown that the distribution of H II nuclei in the 5007/H vs. 6583/ plane parallels the disk H II region sequence, the most striking feature being a clear offset between the two classes of objects, the H II nuclei having larger 6583/H ratios for the same excitation; this effect could be due to the presence of a weak active nucleus in many of these galaxies. Binette (1985) also suggested that mixed cases of starburst and Liner spectra might be relatively common, providing a possible interpretation for objects which have an unusually strong 6300/H ratio compared to H II regions (NGC 3994, for example). Filippenko & Terlevich (1992) suggested that Liners with weak Oi emission (6300/H < 1/6) might be powered by hot main-sequence stars; however, Ho et al. (1993a) showed that these objects are most probably ``composite''.
Ho et al. (1993b) reported the discovery of a non random trend in the dispersion of emission-line intensity ratios for Seyfert 2s. 6300/H and 6583/H were found to be correlated with 5007/H, suggesting the influence of a single underlying physical parameter - the hardness of the ionizing continuum. Our data do not show these correlations, which could be artifacts due to the inclusion in the sample of ``composite'' spectra.
Examination of Fig. shows that the points representative of Seyfert 2 galaxies are not distributed at random in the region assigned to them. Figure is the histogram of the quantity log(6583/H); it shows a sharp maximum at -0.05, with broad wings. Our sample of (131) Seyfert 2 galaxies is not complete in any sense and this could therefore be due to observational biases although this seems unlikely, as the 6583/H ratio is not used for finding Seyfert 2 galaxies. We have no explanation for this fact.