Perhaps the most important result of the PCA is that a great fraction of the emission-line does not vary together with the continuum. This fraction has a line profile that is usually too broad to originate only in the NLR and thus the classical division of the emission-line into a variable BLR and a constant NLR contribution has to be reviewed. The PCA decomposition gives support to the model proposed by Brotherton et al. (1994) in which the traditional BLR is divided in two components: a very broad line region (VBLR) and an intermediate line region (ILR), which have typical velocities of 7 000 and 2 000 , respectively. The line profile in the principal component would then be the signature of the VBLR, whereas the rest component's profile would be the signature of the ILR plus a constant contribution from the NLR.
Under this assumption, the delay of 25 days that we measured between the rest component and the UV continuum in NGC 5548 (Sect. 3.2) can be understood as the distance of the ILR. This result is in good agreement with Brotherton et al. (1994), who estimate that the ILR is about 10 times more distant from the ionizing continuum than the VBLR. We note however that the line profile in the rest component has wings which extend to significant fractions of those in the principal component. This implies the presence of high velocity gas in the ILR and hence suggests that the VBLR and the ILR are not completely separated.
The ratio of the line on the continuum variability naturally decreases when the light-travel time across the line-emitting region is of the order of the typical variability timescale of the ionizing continuum . If , the entire line-emitting region will respond as a whole to continuum variations, whereas if , different sub-regions will respond with different delays to the ionizing continuum averaging out the variations (Rosenblatt et al. 1994). In our interpretation, the lines emitted in the ILR have weak variations, because the extended ILR implies that , whereas the lines emitted in the VBLR vary in tune with the continuum, because this region is small enough to satisfy . A simple way to formally link the fraction of the line in the principal component (i.e. emitted in the VBLR) to the two characteristic timescales is to write
which shows that the BLR size increases with respect to as the luminosity increases. If we assume that is proportional to the observed delays between the line and the continuum and thus that it varies roughly with the luminosity as (Kaspi et al. 1996), the relation in Eq. (14) suggests that the variability timescale of the ionizing continuum is independent of the object luminosity.
The usually observed decrease of the C IV /Ly ratio with increasing continuum flux was interpreted as being due to a population of optically thin clouds (Shields et al. 1995). Such a model predicts that the broad emission-line component emitted in the inner BLR by optically thin clouds should be less variable than the narrower component emitted in the outer BLR by optically thick clouds. This prediction seems to be verified for the H line in some Seyfert galaxies (Mrk 590: Peterson et al. 1993; Mrk 335: Kassebaum et al. 1997), but was never observed to our knowledge for the Ly or the C IV line. Our result that the most varying line-part is usually also the broadest is in contradiction with such a model.
We propose that the C IV /Ly ratio problem is due to a significant decrease of the variability timescale between the ionizing thresholds of H I and C IV, respectively at 912 Å (13.6 eV) and at 192 Å (64.5 eV). If we assume that the light-crossing time of the BLR is similar for the two lines, this would lead to a better line-to-continuum response for the Ly than for the C IV line (Eq. (12) and (13)), implying a decrease of the C IV /Ly ratio with increasing continuum flux.
The lack of far UV observations does not allow us to test this interpretation. However, there are evidences that the continuum varies more rapidly at higher energies than in the UV. Nandra et al. (1991) found that the X-ray flux in NGC 5548 varies significantly on timescales of hours and that it can vary by a factor of two in less than a few days. More recently, Marshall et al. (1997) found that the extreme UV (150-200 eV) can vary by a factor of two on timescales of 0.5 day. Such a decrease of the variability timescale toward shorter wavelength from the UV to the extreme UV domain is qualitatively in agreement with our interpretation.
© European Southern Observatory (ESO) 1998
Online publication: December 16, 1997