In the appendix, we show the PCA decomposition of the C IV and Ly lines for all objects. If the component's line profiles are similar for the C IV and the Ly line of a same object, they can be very different from one AGN to the other. This great diversity forced us to consider only very simple parameters to quantify the line properties of the components in order to find significant trends. The general results and possible trends with luminosity are presented in Sect. 5.1, whereas comments on individual objects are given in Sect. 5.2.
Tables 1 and 2 display the most meaningful quantities derived from our analysis together with the number n of considered spectra and the redshift z of the object as given by the set of identifications, measurements and bibliography for astronomical data (SIMBAD). log(L) is the logarithm of the luminosity density at 1350 Å expressed in . It was derived from the fitted continuum assuming a Hubble constant of .
Table 1. List of the objects in our sample with some characteristic parameters derived from the PCA. log(L) is the logarithm of the luminosity expressed in , the FWHM is expressed in . Other quantities have no units. The parameters are defined in Sect. 5.1
To quantify the variability in the principal component with respect to other variations, we choose two parameters: and , which are respectively the relative importance of the principal component and the ratio of the flux dispersions in the rest and in the principal component, as defined in Sect. 3.1. Both parameters show that in general the dominant variations are well described by the principal component alone ( ≳ 50% and ≲ 50%).
According to the discussion in Sect. 3.2, the fraction of the line that varies together with the continuum is given by the ratio of the mean line fluxes in the principal and in the total component . Its dependence with luminosity shown in Fig. 3 has a slope of /decade. Here, as in the following, the slope was determined by using the "OLS bisector" unweighted linear regression of Isobe et al. (1990), as advised by these authors and the trend is highlighted in the figure by the 3- uncertainty on the regression. We applied a Spearman's test (e.g. Bevington 1969) to determine whether the correlation suggested by Fig. 3 is significant or not. The probability that such a correlation could occur by chance is 1.6%.
We quantified the width of the Ly and the C IV lines by evaluating their FWHM both in the principal and in the rest component. The FWHM values displayed in Table 1 are expressed in with an uncertainty on the measure of . Some noisy components had to be smoothed to determine properly their width and in low redshift objects, for which the blue side of the Ly line is not in the component, the FWHM was extrapolated from the half width at half maximum. Even so, we could not determine a meaningful width for the Ly line in the rest component of NGC 4151, because of geocoronal Ly contamination and for the C IV line in the principal component of 3C 273, because there is nearly no line feature in this component.
The distribution of the FWHM in Fig. 4 shows that there is no obvious difference between the C IV and the Ly line. The average width of the two lines is twice as large in the principal component (6 850 ) as in the rest component (3 400 ). Individually, the line is always broader in the principal component than in the rest component, except for the Ly line in 3C 120 and the C IV line in Mrk 279. This clear result shows that the line-part that varies with the continuum is broader in general than the line-part that varies less. In many objects however, the line in the rest component is broader than the typical width (300-1 000 ) of a line emitted in the NLR (Netzer 1990).
The C IV line profile in the principal component is clearly double-peaked in 3C 390.3, NGC 3516 and NGC 4151. A double-peaked profile is known to be the signature of a thin rotating disk viewed close to edge-on (Welsh & Horne 1991). Recently, Goad & Wanders (1996) showed that double-peaked profiles can also originate due to a non-uniform lighting of a spherical BLR by a predominantly biconical continuum emission. Their model is able to reproduce a wide range of observed profiles and predicts that the FWHM should be generally larger in double- than in single-peaked profiles. The FWHM that we measured for our three C IV double-peaked profiles ( 8 000 ) are among the highest in the sample, in good agreement with their prediction. The line profile in Mrk 509 can be seen as the transition between double- and single-peaked profiles.
We quantified the variability of the continuum and of the Ly and C IV lines by the ratio of the flux dispersion on the mean flux . This ratio corresponds to the parameter first defined by Clavel et al. (1991) to describe the flux variability. We calculated the values shown in Table 2 as explained in Sect. 4. The luminosity dependence of the line variability shown in Fig. 5a has a slope of /decade. The Spearman's test probability that such an anticorrelation could occur by chance is 1.8%, but rises to 10%, when we consider only the Ly or the C IV line and reaches 70% for the continuum variability. This absence of correlation between the continuum variability and the luminosity is in contradiction with Paltani & Courvoisier (1994), who found a strongly significant anticorrelation in a sample of 72 Seyfert-like objects with a slope of -0.046/decade. The fact that this trend remains hidden to us with a sub-sample of 18 objects is most probably due to its weak slope and to the high scatter of the points. This increases our confidence in the significant trends that we found for the two emission-lines.
The ratio of the line variability on the continuum variability is perhaps even more meaningful, since it describes the line-to-continuum response. Its luminosity dependence shown in Fig. 5b has a steeper slope ( /decade) than that of the line variability, but the Spearman's test probability is similar (1.7%) and the scatter of the points is higher. The great similarity with the relation in Fig. 3 is due to the fact that and are closely related, as shown by the following theoretical argument.
This straightforward calculation shows that both and describe in a way the line-to-continuum response.
In all but two objects (3C 390.3 and Mrk 926), the line response is stronger within the uncertainties for the Ly than for the C IV line. This naturally leads to a decrease of the C IV /Ly ratio with increasing continuum flux, as it was observed in some Seyfert galaxies (Peterson 1993). 3C 390.3 was until now the only object in which the C IV /Ly ratio was observed to increase with increasing continuum flux (Wamsteker et al. 1997). Our results confirm this peculiar behavior in 3C 390.3 and predict a similar one in Mrk 926.
We did not quantify line profile asymmetries, but a look at the components displayed in the appendix shows that most of their line profiles are roughly symmetric. The observed absence of strong line asymmetries in the components of most objects excludes that their line-emitting region is dominated by infall or outflow (Rosenblatt et al. 1994). However, minor line asymmetries might reflect some radial motion within a predominantly chaotic or rotational velocity field.
Apart from 3C 390.3, which has a different C IV /Ly ratio behavior than most AGN (Sect. 5.1), other objects have some peculiarities. The most luminous radio-loud quasar in our sample is 3C 273, its principal component has only a very broad and nearly undetectable line feature. This leads to a very weak line variability, as was already noticed by Ulrich et al. (1993) for the Ly line. Since 3C 273 is the only well enough observed luminous object, it is difficult to know whether or not its blazar characteristics are responsible for this peculiarity. However, the fact that 3C 273 is well inside the obtained luminosity trend suggests that other luminous quasars would also have nearly constant emission-lines.
Fairall 9, on the other hand, is an object which does not follow the general luminosity trend. Its line variability behavior is typical for a Seyfert galaxy, whereas its luminosity is that of a quasar. This was already pointed out by Rodríguez-Pascual et al. (1997), who found that the emission-line lags in Fairall 9 (Ly : 14-20 days; He II 1640: ≲ 4 days) are comparable to those in NGC 5548, despite the difference in luminosity of a factor of ten. It suggests that the luminosity of Fairall 9 during the first years of IUE observations was extraordinary high and hence that its average luminosity is overestimated.
3C 120 and NGC 7469 are two other objects that are always outside of the 3- uncertainty curves. Their line variability is very small, because they are the two only objects with a clear asymmetric line profile in their first component. The blue wing of the line is not correlated with the continuum in 3C 120, whereas it seems even anticorrelated in NGC 7469. Such asymmetries are qualitatively consistent with an infalling BLR, since the blue wing is then emitted behind the continuum source having thus a higher emission-line lag than the red wing emitted in front of the source. However, both objects are not well enough observed to draw strong conclusions only based on these PCA results.
© European Southern Observatory (ESO) 1998
Online publication: December 16, 1997