Next Contents Previous


3.1. Observational Methods and Results

Our understanding of the BE has grown in recent years through application of several distinct approaches to the study of AGN emission lines. These techniques include:

1) Measurement of lines in high signal-to-noise ratio spectra, with bivariate analysis. This approach is the simplest to grasp, and the same as that employed by Baldwin (1977a), but improvements in technology have made it possible to study ensembles with very high signal-to-noise ratio. This method arguably reached its pinnacle in the studies by Laor et al. (1994, 1995) of quasar spectra acquired with HST. The data were of sufficient signal-to-noise ratio to permit deconvolution of line blends and accurate measurement of line wings, which are often lost in spectra with only modest signal-to-noise ratio. With a sample of only 18 QSOs, Laor et al. were able to discern a BE in nearly all of the lines of at least moderate strength between 1000-2000 Å (9 features in total), with the notable exception of N V lambda1240.

2) Construction of composite spectra. Subdividing a sample into luminosity intervals, and subsequently generating an average or median spectrum for each bin, offers several advantages for identifying luminosity-dependent behavior. Combining spectra into a composite decreases the shot noise present in any individual spectrum, and also diminishes the ``noise'' contributed by the intrinsic peculiarities of any single source. Composites permit identification of luminosity-dependent line effects by simple inspection, or more quantitative methods. Working with actual spectra, rather than line measurements, also makes it possible to distinguish luminosity-dependent behavior across the profiles of individual lines. Examples of this approach include studies by Véron-Cetty et al. (1983), Osmer et al. (1994, OPG), Laor et al. (1995), and Francis & Koratkar (1995). An example of such a comparison is shown in Figure 3, which compares composite spectra assembled by OPG for high- and low-luminosity quasars at z > 3. The BE is apparent from inspection of most of the stronger lines, and the direct comparison also reveals qualitative differences in the strength of the effect (e.g. large variation in C IV but only a small difference in the 1400 Å feature).

Figure 3

Figure 3. The OPG composite spectra for high- and low-luminosity quasars with z > 3.

3) Principle component analysis (PCA). PCA examines correlations between the variances of parameters measured for a sample, and builds a minimum set of basis vectors across this parameter space that describes the total variance in the system (see Wills & Francis, this proceedings). PCA has been employed in two modes in the study of quasar spectra. The first directly examines variations in spectra, with the measured parameters comprised of the flux density as a function of wavelength after putting the spectra on a common normalization. Francis et al. (1992, FHFC) used this method to study the spectra of 232 objects with 1.8 < z < 2.7 from the Large Bright Quasar Survey to classify the ultraviolet features. They found that the first three principal components accounted for about 75% of the observed variance in the spectra (Fig. 4). These components can be interpreted approximately as largely independent variables describing the strength of the emission line cores, the slope of the continuum, and the prominence of broad absorption line features.

Figure 4

Figure 4. The Francis et al. (1992) principal components found for the LBQS.

The alternative application of PCA operates on a parameter set of quantities measured from the spectra, such as line equivalent widths, profile indices, etc. Boroson & Green (1992) published an influential study of this type, using optical rest-frame spectra of 87 QSOs in the Bright Quasar Survey with z < 0.5. In this case, 61% of the observed variance in the spectra was accounted for with three principal components, and the first eigenvector is dominated by an anticorrelation between measures of Fe II and [O III] line strength.

Conventional studies of line strengths as well as the innovations discussed above have led to reports of a BE in lines of Lybeta lambda1026, O VI lambda1034, Lyalpha lambda1216, O I lambda1304, C II lambda1335, C IV lambda1549, He II lambda1640, Al III lambda1857, C III] lambda1909, and Mg II lambda2798. Given the length of this list, it is of interest to examine those lines that apparently do not participate in the BE. Most studies have shown little or no evidence of a BE for the Si IV + O IV] lambda1400 blend (e.g., Cristiani & Vio 1990; FHFC; OPG; Francis & Koratkar 1995). The interpretation of this result is unclear, and the situation remains ambiguous in light of a strong BE signal for this feature reported by Laor et al. (1995).

Stronger concurrence is available for the behavior of N V, which essentially all studies indicate does not show a BE; the equivalent width of the line is apparently independent of source luminosity (e.g., OPG; Francis & Koratkar 1995; Laor et al. 1995). Hamann & Ferland (1992, 1993) have argued that this behavior can be understood as an abundance effect, with N V emission enhanced relative to other lines in luminous sources resulting from selective (secondary) enrichment of nitrogen in vigorously star-forming environments. Expressed in terms of line intensity ratios, the N V / C IV and N V / He II ratios are enhanced in more luminous systems, and the study by OPG suggests that this is truly a luminosity, rather than redshift, effect. Under this interpretation of the N V line strength, the decoupling of Wlambda(N V) and L would result from the luminosity dependence of N enrichment largely canceling out the normal BE signature.

In the optical bandpass, several studies have addressed the luminosity dependence of broad Hbeta, which also presents little evidence of a BE (e.g., Yee 1980; Binette et al. 1993; Boroson & Green 1992). Only very limited information has been published on the luminosity dependence of the narrow emission lines. Baldwin (1987) has argued that the ultraviolet lines in luminous QSOs are unexpectedly weak, in comparison with Seyfert 2 galaxies (see also Wills et al. 1993b; Laor et al. 1994, 1995). Boroson & Green (1992) found a positive correlation, significant at > 99% confidence, between absolute magnitude MV and Wlambda([O III] lambda5007) for PG QSOs, consistent with a BE. Brotherton (1996) performed a PCA analysis on radio-loud quasars, and while no BE was apparent in a simple bivariate analysis for this sample, his first eigenvector displayed strong BE-like behavior in Hbeta and [O III]. This trend is evidently masked by scatter from other parameters in the full spectra, leading to the lack of a simple correlation.

Even if the narrow lines do exhibit a BE, the extent to which this behavior shares an origin with the broad-line BE may be open to question. Wills et al. (1993b) have argued that the weakness of the UV narrow lines in luminous sources may be due to reddening by dust, part of which is associated with the narrow-line region (NLR) itself; but the broad-line region (BLR) is unlikely to harbor significant dust (e.g., Laor & Draine 1993). More generally, the NLR in QSOs may extend to kpc scales, at which point the properties of the emitting gas may be dictated primarily by the distribution of matter in the host galaxy rather than more exotic processes operative on small scales. If the NLR plasma resides in a disk with a scale height h independent of the luminosity L, a BE would in fact be expected. The length scale r characteristic of the NLR will move out (increase) with L. The solid angle describing NLR coverage scales with h / r, so that the NLR covering factor, and hence equivalent width Wlambda of line emission, will decrease in more luminous systems. This scenario is a variant of the ``receding torus'' model described by Lawrence (1991; see also Simpson 1998), whereby a dusty torus obscures less of the central nucleus in more luminous systems; the dust sublimation radius grows with increasing L. In this picture, NLR emission arises in the unobscured cones along the torus axis. As the torus recedes, the cones expand in opening angle, and the covering factor of NLR gas may then increase with larger L. The luminosity dependence of narrow-line emission in AGNs may thus yield interesting insights into AGN structure, but may have only limited connection to the classical broad-line BE (3).

3 The ambiguity inherent in ascribing common physical processes to sources with similar phenomenology is underscored by the claim that Wolf-Rayet stars exhibit a BE (Morris et al. 1993).

Next Contents Previous