Annu. Rev. Astron. Astrophys. 1999. 37: 487-531 Copyright © 1999 by Annual Reviews. All rights reserved

2.6. Abundance Diagnostics and Results

2.6.1. Intercombination Lines

Shields (1976) proposed using various collisionally excited intercombination (semiforbidden) lines to derive metal-to-metal abundance ratios in QSOs. He emphasized the strengths of NIII] 1750 and NIV] 1486 compared with OIII] 1664, CIII] 1909 and CIV 1549 as potential diagnostics of the overall metallicity. As noted above, the metallicity dependence stems from the expected Z² scaling of N via secondary nucleosynthesis (also Section 6 below). Shields selected lines with similar ionization and excitation energies, so that their ratios are insensitive to the uncertain temperature, ionization, and geometry. Comparisons with the measured line ratios in QSOs (see also Davidson 1977, Baldwin & Netzer 1978, Davidson & Netzer 1979, Osmer 1980, Uomoto 1984) suggested that N/C and N/O are often solar or higher, consistent with solar or higher metallicities. Gaskell et al. (1981) extended this analysis to SiIII] 1892 and other lines to show that the refractory elements cannot be substantially depleted by dust in BELRs.

One drawback of the intercombination lines is that most of them are weak and therefore difficult to measure. Nonetheless, the best recent measurements (Wills et al. 1995, Laor et al. 1995, Boyle 1990, Baldwin et al. 1996) support the earlier results. It is now possible to gather even more data for these lines over a range of redshifts. A note of caution is that the strong feature generally attributed to CIII] 1909 can have large contributions from other lines (Laor et al. 1995, 1997; Baldwin 1996), so that ratios like NIII]/CIII] might systematically underestimate N/C if line blending is not accounted for.

A more serious concern is that the early theoretical work did not consider the range of high densities now believed to be present in the BELR (Section 2.2). The intercombination lines probably form at or near their critical densities (typically 3 × 10⁹ to 3 × 10¹¹ cm^-3 for = 1 in Equation 5). Lines with different n_crit could have different degrees of collisional suppression. For example, the calculated results with n_e n_H = 10¹⁰ cm^-3 in Figures 5 and 6 favor large NIII] / CIII] at a given N/C abundance because CIII] is collisionally suppressed above its n_crit 3 × 10⁹ cm^-3. If there is a range of densities, lines with different n_crit might form in different regions (even if they have similar ionizations), leading to a geometry dependence. Nonetheless, line ratios involving similar n_crit and similar sensitivities to other parameters, such as NIII]/OIII], could still be robust abundance indicators when they are measurable. More theoretical work is needed to explore the parameter sensitivities and selection effects that can influence these lines in complex BELRs.

2.6.2. Permitted Lines

There are several possibilities for abundance diagnostics among the permitted UV lines. Figure 6 shows that NIII 991/CIII 977 and NV 1240/CIV 1549 should be good tracers of N/C. Another possibility is NV/HeII 1640, or perhaps NV / (CIV + OVI 1034). The NV, OVI and HeII lines form in overlapping regions (Figure 3), as do NIII and CIII, so their flux ratios should be insensitive to the global BELR structure. Also, as noted above, the N lines are not important coolants and are thus responsive to abundance changes. There are practical problems with most of these lines, however; NV is blended with Ly; CIII, NIII, and HeII are weak; and CIII, NIII, and OVI lie in the "forest" of intervening Ly absorption lines. Nonetheless, improvements in the quality of data (for example, high resolution and high signal-to-noise spectra in the Ly forest) are permitting increasingly accurate measurements of these lines in large QSO samples.

2.6.3. NV/HeII and NV/CIV

Some of the first studies of QSO samples noted that NV 1240 is often stronger than predicted by photoionization models using solar abundances (Osmer & Smith 1976, 1977). The NV/HeII and NV/CIV ratios have since received particular attention as abundance diagnostics (Hamann & Ferland 1992, Hamann & Ferland 1993a, Hamann & Ferland 1993b, Ferland et al. 1996). Figure 7 shows the measured ratios in these lines for QSOs at different redshifts (from Hamann & Ferland 1993b, Hamann et al. 1997a, with some new data and modifications based mainly on Wills et al. 1995, Baldwin et al. 1996). NV/HeII is the ratio of a collisional to recombination line, with the expected strong temperature dependence (Section 2.4). Calculations similar to (but more exhaustive than) those shown in Figures 4, 5 and 6, indicate that NV/HeII reaches a maximum value linked to the maximum temperature in photoionized clouds (Ferland et al. 1996). The maximum NV/HeII ratio is ~ 2-3 for solar N/He abundances, depending on how "hard" a continuum shape one considers realistic for QSOs. (Beware that the highest ratios in Figure 4 occur for parameters where both lines are growing weak, cf the EW(CIV) plot or Korista et al. 1997b.) Nominal BELR parameters predict NV/HeII near unity for solar N/He (Figures 4, 5 and 6). These predictions fall well below most of the measured ratios (Figure 7), implying that QSOs typically have super-solar N/He. The ad hoc (high) temperatures that would be needed to explain the observed NV/HeII ratios with solar N/He are inconsistent with photoionization equilibrium and would lead to strong far-UV emission lines. The fact that these far-UV line strengths are not seen sets an upper limit on the temperature and supports the result for super-solar N/He (Ferland et al. 1996).

Figure 7. Measured NV/HeII and NV/CIV flux ratios versus redshift (left panels) and continuum luminosity (right). The upper and lower ranges might be undersampled (especially for NV/HeII at redshifts > 1) because limits on weak lines (e.g. HeII) were often not available from the literature. The two asterisks in each panel represent mean values measured by Osmer et al. (1994) for high- and low-luminosity QSOs at redshift > 3. The solid curves are predictions based on chemical evolution models (discussed in Section 6).

The NV/CIV lines are collisionally excited with similar energies, so the temperature dependence in this ratio is smaller than NV/HeII. Nominal BELR parameters predict NV/CIV of ~ 0.1 for solar N/C (Figures 4, 5 and 6; see also Hamann & Ferland 1993a, b). Comparisons with the data in Figure 7 thus indicate super-solar N/C for most QSOs.

The two NV ratios together therefore imply that (a) quasar metallicities are often solar or higher, especially in high-redshift, high-luminosity objects, and (b) nitrogen (e.g. N/C) is typically enhanced compared with solar ratios (Ferland 1996, Hamann & Ferland 1993a, b). The conclusion for enhanced N/C is based largely on NV/CIV, but we note that the scaling of N Z² leads to self-consistent estimates of Z based on NV/HeII and NV/CIV (Figure 7). The actual Z values are uncertain, but the main point is that many observed ratios require Z Z. Figures 6 and 7 combined suggest that the nominal metallicity range is 1 Z 10 Z for standard photoionization parameters and N Z².

Hamann & Ferland 1993b noted that the observed NV ratios tend to be higher in more luminous sources (Figure 7). Most BELs exhibit the well-known Baldwin effect, that is, lower equivalent widths at higher luminosities (Baldwin 1977b). This effect is well-established in CIV and appears to be even stronger in OVI (Zheng et al. 1995, Kinney et al. 1990, Osmer & Shields 1999). Surprisingly, NV does not show this effect (Osmer et al. 1994, Laor et al. 1995, Francis & Koratkar 1995) even though its ionization is intermediate between CIV and OVI, and its electron structure is identical. We proposed that the peculiar NV behavior is caused by generally higher metallicities and more enhanced N abundances in more luminous QSOs. The recent theoretical study of the Baldwin effect by Korista et al. (1998) gives quantitative support to that proposal. In Section 7, we argue that this proposed metallicity-luminosity trend in QSOs could naturally result from a mass-metallicity correlation among their host galaxies.

The abundance results based on NV have been questioned by Turnshek et al. (1988, 1996), Krolik & Voit (1998), who argue that the NV BEL forms largely by resonance scattering in an outflowing BAL region. NV emission might be selectively enhanced by this mechanism because it can scatter both the continuum and the underlying Ly BEL. However, explicit calculations of the line scattering (Hamann et al. 1993;, Hamann & Korista 1996; F Hamann, KT Korista, & GJ Ferland, manuscript in preparation) do not support this scenario. For example, (a) the amount of NV scattering estimated by Krolik & Voit (1998) is too large by a factor of ~ 3 on average, because BALRs do not generally have the right velocity/optical-depth structure to scatter all of the incident Ly photons. In particular, NV BALs are not usually black across the Ly emission-line wavelengths (see Figure 1 in Hamann et al. 1999a). (b) It is difficult for NV ions in high-velocity BAL winds to scatter Ly photons into simple BEL profiles with the observed half-widths of typically 2000-2500 km s^-1. For example, isotropic scattering of the Ly flux would produce BEL half-widths of ~ 6000 km s^-1 (the velocity separation between the NV and Ly lines). Anisotropic scattering (e.g. in BALRs with equatorial or bipolar geometries) would lead to strong orientation effects and systematically broader BEL profiles in BAL versus non-BAL QSOs. These differences are not observed (Weymann et al. 1991). (c) It is not clear why, in individual spectra, the NV emission profiles should closely resemble those of other BELs if the former are produced by scattering in a high-velocity BAL wind whereas the latter are collisionally excited in a separate region [i.e. the usual BELR - whose velocity field is not mostly radial based on the reverberation studies (Türler & Courvoisier 1997, Korista et al. 1995)]. Finally, (d) large scattering contributions to NV would minimally require much larger global BALR covering factors (the fraction of the sky covered by the BALR as seen from the central QSO) than expected from their observed detection frequency in (randomly oriented) QSO samples. Goodrich (1997), Krolik & Voit (1998) argue that larger global covering factors could occur, but that issue is not settled.

Another concern is that complex BELR geometries might cause the NV/HeII and NV/CIV abundance indicators to fail - but they would fail in opposite directions. Specifically, clouds that are truncated at different physical depths (see Figure 3) could produce strong HeII with little or no NV and CIV emission, or strong HeII and NV with little or no CIV. For a given abundance set, this type of truncation could therefore either lower the observed NV/HeII ratio or increase NV/CIV. Comparing the data with simulations that do not take truncation into account (Figures 4, 5 and 6) might then lead to underestimated N/He abundances or overestimated N/C. However, we have already shown that these two line ratios yield similar metallicities when compared with the nontruncated simulations, so we are not likely being misled by complex BELR geometries. Moreover, the NV/HeII ratio provides in any case a secure lower limit on N/He.

2.6.4. FeII/MgII

The broad FeII emission lines pose unique problems because the atomic physics is complex and many blended lines contribute to the spectrum, particularly at the wavelengths ~ 2000-3000 Å and ~ 4500-5500 Å. Nonetheless, FeII is worth the effort because a delay of ~1 Gyr in the Fe enrichment, relative to elements such as O, Mg, or Si, might provide a clock for constraining the ages of QSOs and the epoch of their first star formation (see Sections 6-7 below; Hamann & Ferland 1993b).

A series of important papers on FeII emission (Osterbrock 1977;, Phillips 1977, 1978;, Grandi 1981) culminated with Wills & Netzer (1983), Wills et al. (1985). They performed sophisticated calculations showing that the large observed FeII fluxes, in particular FeII(UV)/MgII 2799, require that either Fe is several-fold overabundant (compared with solar ratios) or some unknown process dominates the FeII excitation. One process that might selectively enhance FeII emission is photoexcitation by Ly photons (Johansson & Jordan 1984). The absorption of Ly radiation can pump electrons from the lower (metastable) energy levels of Fe⁺ into specific high-energy states, leading to fluorescent cascades. Wills et al. (1985) discounted this mechanism because it appeared insignificant in their simulations, but Penston (1987) noted that Ly pumping is known to be important in some emission line stars, such as the symbiotic star RR Tel, and therefore might be important in QSOs (also Graham et al. 1996, Laor et al. 1997). More recent FeII simulations with better atomic data and exploring a wider range of physical conditions (Sigut & Pradhan 1998, Verner et al. 1999) suggest that Ly can be important in some circumstances, but it is not yet clear whether those circumstances occur significantly in QSOs.

Recent observations have renewed interest in this question by showing that the FeII(UV)/MgII emission fluxes can be larger than the Wills et al. (1985) predictions even at z > 4, with the tentative conclusion that Fe/Mg is at least solar (and thus the objects are at least ~ 1 Gyr old) (Taniguchi et al. 1997, Yoshii et al. 1998, Thompson et al. 1999, and references therein). New theoretical efforts, such as Sigut & Pradhan (1998), Verner et al. (1999), are needed to test these conclusions and quantify the uncertainties. However, a better way to measure Fe / might be with the intrinsic NALs (see below).