The emission-line spectra of Seyfert galaxies have been very well studied in the optical spectral region. They are similar to the spectra of gaseous nebulae, but cover a wider range of ionization. In the simplest classification scheme, they may be divided into two types, Seyfert 1 and Seyfert 2. The Seyfert 2 nuclei have emission-line spectra with line widths typically 350 km s-1 full width at half maximum (FWHM), ranging from 200 to 700 km s-1 in different objects (Dahari and De Robertis 1988a). These lines include both permitted lines of H I, He I and He II, and forbidden lines of which the strongest are [O III] 4959, 5007, [N II] 6548, 6583; other lines include [O I] 6300, 6364, [S II] 6716, 6731, often [Fe VII] 6087, and in many cases [Fe X] 6375. The Seyfert 1 nuclei spectra include all these 'narrow' broad emission lines, plus much broader emission lines of H I, He I, He II and Fe II, typically with FWHM 3000 km s-1, ranging from 500 to 7000 km s-1 in different objects. The full widths at nearly zero intensity (FW0I) of these same lines, so well as they can be defined, range from 5000 to 30,000 km s-1 (Osterbrock and Shuder 1982).
It is evident from these line spectra that in AGNs, as in nearly all other astronomical objects we know, H is by far the most abundant element, He next, O, Ne, N, C next, etc. The simplest interpretation is that the narrow lines are emitted in a `narrow-line region' (NLR) in which the velocity field ranges up to a few hundreds of km s-1, and the broad lines in a `broad-line region' (BLR) in which the velocity field ranges up to as high as 10 or 15 km s-1. The (nearly) complete absence of forbidden line emission from the BLR can only mean that in it they are all collisionally de-excited; this implies that the electron density throughout this region is much higher than the critical densities for collisional de-excitation of all the strong forbidden lines observed from the NLR and in gaseous nebulae (see, e.g., Osterbrock 1989). From the known transition probabilities and collision strengths for collisional excitation of the forbidden lines of the expected more abundant ions, this limit is roughly Ne 108 cm-3.
An upper limit to the electron density in the BLRs may be set from the observed presence in many Seyfert 1 and QSO spectra of semi-forbidden C III] 1909 emission, with a FWHM comparable to those of the permitted lines. Its transition probability, A = 96 s-1, intermediate between those of typical permitted and forbidden lines, and its critical electron density Nec 3 x 109 cm-3. The density in the regions of the BLR in which C III] 1909 is emitted with appreciable strength cannot be much higher than this, and an intermediate density Ne 109 cm-3 has frequently been taken as roughly representative of the BLR. However, recent evidence discussed in section 2.7 argues for higher densities, and thus for appreciable collisional de-excitation.
The interpretation then of a Seyfert 1 AGN is that it contains a BLR and a NLR; of a Seyfert 2, that it contains only a NLR, or as we shall discuss in some detail in section 6, that it contains a NLR and a `hidden' BLR that we do not directly observe, because of intervening material that cuts off the direct radiation from it toward us. Intermediate cases can also be recognized. Seyfert 1.5 is the classification type generally used for objects whose spectra show H I emission lines with strong broad and strong narrow components, while Seyfert 1.8 and 1.9 are the types used for objects in which the broad component of H is very weak or undetectable, respectively (Osterbrock 1984).
Diagnostic information is available on the physical conditions within the NLRs. From the measured ratios of the strengths of emission lines with different upper levels, such as [O III] [I(4959) + I(5007)] / I(4363) and [S II] I(6716) / I(6731), a representative `mean' electron temperature and density may be derived for each observed NLR. Typical values are T 15,000 K and Ne 3 x 103 cm-3. There is no direct diagnostic that gives a representative mean temperature for the BLR, but on general grounds an estimate T 15,000 K also is plausible.
Of course these are highly simplified representations of what must be in reality an extremely complicated situation. In the somewhat analogous case of a stellar a atmosphere we know that the density and temperature increase inward, but for the roughest semi-quantitative description we may represent it by a single `effective' temperature and a single `mean' density. A better representation is to take two representative points, still better, three, and the only true description is the detailed run of mean temperature and density with depth, with the fluctuations around these mean values in space and time. In the case of a Seyfert 1 galaxy a description in terms of one representative point is too simplified to be useful; we must specify at least two points `the BLR' and `the NLR'. In fact there must be a continuous transition between these two idealized `regions', and large variations in the physical conditions within each of them. Certainly diagnostic line ratios of different ions give different mean temperatures and/or densities, indicating real variations in T and Ne within the NLR; the quoted values are the best overall fits to the various different determinations. In all gaseous nebulae we know, ionized gas tends to be clumped in condensations on all scales down to the smallest observable, so the mean density indicated by emission-line processes, dominated by the denser regions of the condensations, is higher than the mean density given by the total number of particles within the whole volume of the nebula. This can be represented by a `filling factor' f, giving the fraction of the volume occupied by dense condensations. It corresponds to describing the nebula as consisting of two phases: gas, with electron density Ne and relative volume f, and vacuum or a much hotter, lower-density gas with the same pressure and relative volume 1 - f. For typical nebulae f 10-2.
From this picture the approximate size of a typical NLR in a Seyfert galaxy nucleus can be estimated from its luminosity in a particular emission line, for instance H. It is well understood as arising largely from recombination of protons and electrons to levels n 4 followed by downward radiative transitions, and the luminosity in H can be written
Here H is the effective recombination coefficient for emission of an H photon, h is its energy, and the product of the first four factors is the emission coefficient per unit volume, while V is the total volume of the NLR. To a sufficiently good approximation Ne = Np + l.5 NHe (H completely ionized to protons, and He half to He+ and half to He2+). The recombination coefficient H varies only slowly with T, and if to be specific we assume T = 104 K and a spherical volume,
will give the radius R. The most luminous NLRs of Seyfert galaxies have L(H) 2 x 108 L, which gives M 7 x 105 (104 / Ne) M and R 20 f-1/3 (104 / Ne)2/3 pc. Such an NLR with Ne = 104 cm-3 will therefore have a mass of ionized gas M 106 M and, for an assumed filling factor f = 10-2, R 90 pc. In fact a few of the nearest Seyfert 2 NLRs are resolved on direct images, and have apparent diameters of the order 100 pc or so.
Typical BLRS are much smaller. The observed Balmer decrements, or ratios of intensities of H I lines I(H) : I(H) : I(H) etc. show that other processes in addition to recombination contribute to the H I line emission in these denser objects. However, for a rough estimate we can ignore them and use the same approximate calculation. The most luminous Seyfert 1 AGNs typically have L(H) 109 L, which gives M 36 M (109 / Ne) and R 0.015 f-1/3 (109 / Ne)-2/3 pc. Thus for a representative density Ne 109 cm-3, the mass of ionized gas in the BLR is only M 40 M, and for an assumed f 10-2, R 0.07 pc 0.2 light year. This is far too small to hope to resolve, even in the nearest Seyfert is, and to date none has been resolved. Figure 1 is a highly schematic drawing (not necessarily all at the same scale) of these various regions, with the central black hole, its BLR shown as cylindrically symmetric, and the NLR as spherically symmetric.
Figure 1. Schematic representation of structure of an active galactic nucleus, including central black hole and accretion disk (black), cylindrically symmetric broad-line region (BLR) containing ionized (BLG+) and neutral (BLG0) gas, with highest stages of ionization such as He2+ concentrated closest to ionization source. Narrow-line region (NLR) taken here to be spherically symmetric, but ionized (NLG+) only in the core in which ionizing radiation from central source can escape or penetrate through the NLR. The figure is necessarily not to scale. (Osterbrock 1978a)
Next let us discuss the energy-input mechanism for the ionized gas, first in the NLR. The relatively low electron temperature, T 15,000 K, together with the ionization extending to such high stages as O2+ and Ne2+, shows that the main energy-input mechanism must be photoionization. The only other forms of energy input we know, conversion of kinetic energy into heat either through shock waves (`cloud-cloud collisions') or beams of particles being stopped, require much higher temperatures for the observed degree of ionization.
Furthermore AGNs have a strong continuous spectrum. Presumably it arises from the accretion disk. In typical objects, in the optical and observable ultraviolet spectral regions, it follows approximately a power-law form
Figure 2. Reddening-corrected [O III] 5007 / H vs [N II] 6583 / H intensity ratios. Symbols for various objects are as shown in the key in the lower left corner of the figure. Short broken curves represent calculated ratios from AGN models with power-law input spectrum as described in the text, and solar and 0.1 solar abundances (upper and lower respectively). The long broken curve is the prediction from the composite model described in text. The chain curve is the prediction of the shock wave model. The full curve divides AGNs from H II-region like objects (Veilleux and Osterbrock 1987).
with n 1.2. This spectrum, extrapolated further into the ultraviolet and x-ray regions, apparently is the source of the ionizing photons. This is indicated by the fact that the equivalent width of the observed H emission, that is
is approximately constant for most Seyfert 2 NLRs. This is exactly what is expected under photoionization by a spectrum of fixed form, such as a power law, if the total number of ionizing photons is balanced by the total number of recombinations
where B is the effective recombination coefficient while the luminosity in H comes from a definite fraction of those recombinations
A power-law form for the photoionizing spectrum is harder than any O star spectrum, which always falls off roughly as exp (-h / kT*). This explains both the higher stages of ionization in the AGN spectra than that in H II regions, and the great strength of the lines of low ionization, such as [O I] and [S II], which arise in the long partly ionized region maintained by the high-energy, penetrating photons.
These qualitative conclusions are confirmed by quantitative models, calculated using various power-law input spectra, chosen to match approximately the observed luminosities and frequency dependence of the observed NLRs. Some of the most complete of such calculated models are those by Aldrovandi and Contini (1984, 1985), Contini and Aldrovandi (1983), Ferland and Netzer (1983), Halpern and Steiner (1983), Stasinska (1984a, 1984b), and Binette (1985). For instance, figures 2, 3 and 4 show the line ratios I([N II] 6583) / I(H), I([O I] 6300) / I(H), and I ([S II] 6716 + 6731) / I(H) plotted against I([O III] 5007) / I(H). These are the best diagnostic ratios to distinguish between AGN spectra and H II region or starburst-galaxy spectra of those originally proposed by Baldwin et al (1981). The black circles represent measured ratios for the AGNs, and the open circles are the measured ratios for H II regions in other galaxies, starburst, and H II region galaxies, all known to be photoionized by O stars. The full line on each diagram is the empirical division between AGNs and H II regions. The short broken lines represent sequences of AGN models, calculated with input power-law spectra with n = 1.5, Ne 103 cm-3, and either essentially solar abundances of the elements (upper right curves on all three diagrams) or with the abundances of all the heavy elements reduced by a factor 10 relative to H and He (lower left curves). Along each of these curves the ionization parameter
essentially the ratio of density of ionizing photons to density of free electrons, decreases from = 10-1.5 at the upper left end to = 10-4 at the lower right. The long broken lines represent three models calculated with the same input spectrum and solar abundances, but containing two types of clouds, with Ne = 102 and 106 cm-3, to mimic roughly the effects of density variations within the object and collisional de-excitation.
It can be seen that the calculated models with power-law input spectra match the observed line ratios for AGNs reasonably well, and do not agree at all with the observed ratios for H II regions and galaxies. These latter, however, are well represented by models with O star input photoionizing spectra (Veilleux and Osterbrock 1987). Note that for the composite density and hence presumably the most realistic AGN models, the [O I] and [S II] data are matched well with `normal' solar abundances, but for [N II] an increase in the abundance (in the first order, of N alone) by roughly a factor of three is indicated by the observed ratios.
The only justification for assuming a power-law form of spectrum is that it is simple. A broken power law, more nearly flat at high energies, may be the next best approximation to reality (see section 3.3). Much more complicated forms, based on observational data and reasonable extrapolations, have been used by Mathews and Ferland (1987), Ferland and Persson (1989), and others. Some other types of input spectra, all sharing the property of including photoionizing spectra over a wide range of energies, have been investigated by Binette et al (1988).
It should be noted that, in addition to recombination, collisional excitation also makes an appreciable contribution to the strength of the H I lines, particularly of L and H. The reason is that the gas clouds in AGNs, photoionized by a `hard' spectrum (containing relatively many high-energy photons in comparison with O-star spectra) have large partly ionized zones in which both H0 and free electrons exist, in contrast to gaseous nebulae, in which the ionization of H and the free electron density drop very abruptly together at the edge or boundary of the neutral region. The NLR models typically give calculated intensity ratios I(H) / I(H) = 3.1, which must be used to determine the reddening by dust observationally, rather than 2.85 as in pure recombination (Gaskell and Ferland 1984).
2.4. Broad-line region
It is not so clear that photoionization is the main energy-input mechanism to the ionized gas in the BLR, but this interpretation seems most likely. The main diagnostic evidence is that the equivalent widths of the H I emission lines, specifically either H or H, expressed in terms of the featureless continuum, are more or less the same for Seyfert 1 and 2 galaxies, QSOs, and radio galaxies, and fit the predicted relationship for a power law with n 1.2 (Yee 1980, Shuder 1981). This is the expected result of photoionization, as discussed in section 2.3, but it is not unique; if some other energy-input mechanism, say injection of fast particles or high-kinetic-energy clouds, were closely proportional to the luminosity in the featureless continuum, the observational result would be the same. No such process has been suggested. In addition, the observed tight correlation between continuum and broad emission-line variability makes it difficult to consider any other source. Furthermore, detailed photoionization models can be adjusted to fit approximately the observed spectra. However, it is quite possible that a non-negligible fraction of the heating in some or all BLRs results from dissipation of mechanical energy, as well as from photoionization.
Calculating the expected emission-line spectra of model photoionized BLRs is more complicated than for gaseous nebulae and NLRs. In the latter objects the density is low enough so that the fundamental nebular approximation, that essentially all ions and atoms are in their lowest energy levels and that processes involving ions in excited levels can be neglected, except for emission of photons, is generally valid. Furthermore the optical depths in nearly all the lines are small, and for other lines, such as the Lyman series of H, it is a good approximation to consider them infinitely thick optically. Very complete calculations are available for the H I and He II recombination spectra in this nebular approximation (Hummer and Storey 1987). They take into account all the radiative and collisional processes that are relevant up to densities Ne 107 cm-3 for the H I spectrum and 109 cm-3 for He II. The energy-level diagram of He I is more complicated, but calculations on a reasonably good approximation are also available for it (Brocklehurst 1972, Almog and Netzer 1989). They also include all relevant collisional processes (but the various cross sections are not as accurately known for this two-electron atom), and in addition approximately include radiative-transfer effects in a uniform finite-thickness slab approximation. As a result these He I calculations are stated to be correct (within the limitations of the highly simplified model geometry and the available values of the collision cross sections) up to Ne 1014 cm-3.
Most of the earlier theoretical work on BLRs was done on the H I line spectrum of a dense, optically thick region (Kwan and Krolik 1981). Generally these papers assumed uniform density and temperature, but went beyond the low-density nebular approximations in including collisional and radiative excitation and ionization from excited levels, and also line radiative-transfer effects in a simplified approximation. One of the striking early observational discoveries in the spectra of redshifted quasars is that H I L is relatively weak with respect to the Balmer lines, so that the intensity ratio I(L) / I(H) 10 typically, rather than 30 as expected in nebular-type (low-density) spectra. The reason is that in dense BLRs L photons are not simply scattered, but are `destroyed' by Balmer-line absorption processes and collision processes which remove H0 atoms from tile excited 22P level before they emit a L photon. Dust extinction also plays a role, but the discrepancy does not result from it alone.
The best recent papers take into account heating by photoionization, include other ions in addition to H and He, determine the equilibrium temperature at each point from the balance with radiative cooling, and calculate the emergent spectrum (Kwan 1984). The relevant parameters, in addition to the form of the photoionizing spectrum, are the input ionization parameter , the assumed constant density N0 and the total optical depth of the model `cloud' (semi-infinite slab) at the Lyman limit. In a general way, models with ionization parameters derived from the luminosities and reasonable interpolations and extrapolations of the observed spectra of AGNs into the photoionizing region, Ne 109 cm-3 and sizes of BLRs previously mentioned approximately fit their observed emission-line spectra.
One very interesting but complicated set of diagnostics is the strength of the Fe II features in the optical region. These are blends of large numbers of individual emission lines belonging to several multiplets especially concentrated in the regions around 4570, 5120, 5320. They make up a significant fraction of the flux in the optical part of the spectrum in many Seyfert 1 galaxies (Wills et al 1985).
All the observed lines arise from upper levels connected with the ground configuration of Fe II by permitted ultraviolet resonance transitions. They are evidently excited by collisional and radiative fluorescence processes, and the optical depths in the resonance lines are clearly large (0 ~ 104 in the stronger transitions). Since Fe+ is a relatively low stage of ionization (the ionization potential of Fe0 is 7.9 eV), the Fe II lines are emitted quite strongly in the large partly ionized zone, or `transition zone', in which the ionization of H is dropping from nearly completely H+ to nearly completely H0 (Netzer and Wills 1983). In this region the bulk of the ionization occurs by highly penetrating x-ray photons. The Mg II emission 2798, 2803 lines observed in many (redshifted) QSOs and quasars, and Ca II 8498, 8542, 8662, observed in many (low-redshift) Seyfert 1s (Persson and McGregor 1985, Persson 1988) also arise in this zone, as does a significant fraction of the H I line emission (Collin-Souffrin et al 1982, Collin-Souffrin and Dumont 1986).
Quite recently evidence from time variations of the strengths and profiles of broad emission lines in some AGNs has suggested that their BLRs may be considerably smaller than previously estimated (see section 2.7). This would imply that their mean densities are considerably higher than 109 cm-3. Calculated models are therefore required with densities up to Ne 1013 cm-3. In this regime the physical conditions are more nearly similar to those in stellar atmospheres then in typical gaseous nebulae. Three-body recombination, large optical depths in many lines, and even Stark broadening (which affects the optical depths) must be taken into account. The models calculated to date with mean densities Ne 1013 cm-3 do not match the observed line spectra and line profiles of BLRs nearly as well as the models with characteristic mean densities Ne 109 cm-3, but the models with mean densities up to Ne 1010.5 cm-3 are quite acceptable (Rees et al 1989). All the published models assume simple density laws, but undoubtedly the actual BLRs are much more complicated.
One problem in any photoionization model, whether of an NLR or a BLR, is the physical origin of the small filling factor, that is, the clumping as gas into density condensations or `clouds'. If not confined they should quickly dissipate, essentially at the speed of sound ( 10 km s-1) and after the `sound crossing time', approximately 104 y for a BLR with R = 0.07 pc, should no longer exist. Perhaps they are constantly regenerated by turbulent processes; perhaps they are confined by a hot, low-density invisible gas with which they are in pressure equilibrium. The most widely accepted picture is that this intercloud gas is at T 108 K (Krolik et al 1981). However, recent theoretical studies by Fabian et al (1986) and Mathews and Ferland (1987) show that such gas could not escape detection somewhere in the x-ray, ultraviolet or infrared continua of AGNs, as it has to date. Thus whether it exists, and its nature if it does, are uncertain. Rees (1987) has suggested that magnetic fields may provide the confinement mechanism. The models previously described simply adopt various power-law forms for the pressure of the assumed confining medium without specifying their physical cause.
So many energy levels and so many transitions are involved in the Fe II emission-line spectrum that it is difficult to model with currently available computers and calculated atomic parameters. Ca II, with only five relevant levels, is much more straightforward. The model calculations show that its emission lines must arise in slightly ionized regions at very large optical depths for ionizing radiation from the central source, that is, large column densities, NH 1024.5 cm-2. At these large densities and large optical depths the radiative heating and cooling processes familiar in most nebular contexts are ineffective, and processes such as Compton scattering of high-energy photons by bound electrons, photoionization of C0 and H- by photons with h < h 0, and photoionization of Fe0 K and L electrons by x-rays all come into play (Collin-Souffrin et al 1988, Ferland and Persson 1989). Even small amounts of dissipation of kinetic energy by heat would also be important (Collin-Souffrin et al 1986).
2.5. Accretion disk continuum
The observed continuous spectra of AGNs are very complicated mixtures of several components. One is a power law extending from the infrared to soft x-ray spectral regions, as previously described, but superimposed on it is a broad continuum which extends over the range roughly 1014.5 Hz < < 1016.5 Hz, with a peak somewhere between. This is generally called the `big blue bump' (meaning on a plot of flux Fv, or even better F against frequency), and it is further contaminated by the `little blue bump', in the near ultraviolet, consisting of Balmer continuum emission and many unresolved Fe II emission multiplets (Wills et al 1985). The big blue bump is generally attributed to dissipation of energy in an accretion disk near the central black hole (Shields 1978). Various types of accretion disks might exist, for instance thin disks, radiation tori and ion tori, differing in the physical processes that dominate in fixing their structure (Begelman 1985). Which actually occurs, or which occur in which specific objects, depends on details of the release of energy in the accretion disk that are not yet understood. The simplest and most straightforward case to calculate is the (geometrically) thin disk, which is optically thick. Most published interpretations of observational data have been made in terms of this model. For the high-luminosity QSOs, in which the underlying integrated-stellar galaxy continuum spectrum makes the smallest contribution and the fit to the blue bump is best determined, an accretion-disk continuum provides a significantly better match to the observed spectrum than a blackbody does. Both have two parameters, M and for the accretion disk, or L and T for the blackbody. The calculated spectra of accretion disks about either a Schwarzschild (non-rotating) or Kerr (rotating) black hole fit equally well; the indicated mass of the Schwarzschild black hole is about two to three times that of the corresponding Kerr black hole. The calculated spectrum of an accretion disk depends on the angle i between its normal and the line of sight, but this angle cannot be determined from the fitted data; to the accuracy of the data an increase in i can be compensated for in the emergent spectrum by an increase in M and (Malkan 1983).
Observations over a wide frequency range are necessary to eliminate or fit the power-law continuum, the Balmer continuum, the Fe II features, and the underlying integrated-stellar galaxy continuum. The energy dissipation in the accretion disk increases inward toward the black hole, as a result of the Keplerian velocity field in the disk. This leads to an increase of effective temperature inward, with the highest energy photons coming from the inner edge of the accretion disk. The emergent spectrum at each radius has been calculated for various theoretical models in various approximations (Czerny and Elvis 1987, Wandel and Petrosian 1988). The most recent published treatment takes into account detailed calculations based on the Kerr metric of a rotating black hole spun up to the maximum angular momentum-to-mass ratio (Sun and Malkan 1989). On physical grounds it should be the best approximation to the actual situation in nature (Thorne 1974). There are relativistic effects especially on the radiation from near the inner edge of the accretion disks. The emergent spectrum depends on the mass of the black hole M, the accretion rate , and the inclination of the disk to the line of sight. All the cited papers agree that their respective models and fitting procedures give AGN black holes masses ranging from 107.5 to 109.5 M, the higher range of masses M = 108 to 109.5 M corresponding to QSOs and the lower range 107.5 to 108.5 M corresponding to Seyfert 1 nuclei. Furthermore, the most luminous QSOs tend to have mass accretion rates and luminosities nearly as large as the Eddington luminosities LE corresponding to the derived masses of their black holes, while the less luminous Seyfert 1 nuclei typically have luminosities only a few percent of their Eddington luminosities. The most recent, and to date most physically complete model calculations, are those of Laor and Netzer (1989). They found that the upper limit to the luminosity for a thin disk is L < 0.3LE, and that all the AGN-model thin disks are dominated by radiation pressure.
However, it must be recognized that real problems remain in fitting the observed spectra with thin accretion disk models. Earlier calculations predicted a large absorption discontinuity at the Lyman limit (912), while the Laor and Netzer (1989) calculations predict a large emission discontinuity there. The observations show neither, and furthermore the polarization predictions are difficult to reconcile with the observational data (Antonucci et al 1989). The case is not closed.
2.6. Ultraviolet line spectra
The ultraviolet line spectrum of AGNs is simply an extension of the optical line spectrum to the region beyond the atmospheric cutoff imposed by the Earth's atmosphere. In low-redshift Seyfert nuclei the ultraviolet region is not observable from the ground; in high-redshift QSOs it is redshifted into the optical region and becomes observable, but correspondingly the optical region is redshifted toward the infrared and out of the region in which optical detectors work. For instance in a QSO with z = 1.7, L is shifted to ~ 3280, barely observable with any efficiency at a ground based observatory, but H is shifted to ~ 13125, beyond the range of sensitivity of a CCD. As a result, reasonably complete data are available on the emission lines over a wide range in wavelengths only for fairly bright Seyfert galaxies, observable in the ultraviolet with the IUE satellite. They show that the ultraviolet spectra of Seyfert 1 galaxies are quite similar to those of QSOs (Wu et al 1983, Clavel and Joly 1984).
In QSOs with emission lines, H I L, if in the observable range, is nearly always the strongest line. Very frequently C IV 1549, a close doublet which is unresolved in all broad-line objects, is also quite strong. Thus in objective prism or grism surveys aimed at finding more QSOs, if only a single emission line is seen in a spectrum, the best working hypothesis is that it is L, or if not L, 1549.
The best tests of any model for a particular AGN or a particular class of AGNs are comparison of the predicted spectrum with observational data over the entire ultraviolet, optical and infrared regions, to as long a wavelength as possible (Ferland and Osterbrock 1986, Ferland and Osterbrock 1987, Oliva and Moorwood 1990).
2.7. Temporal variations
A considerable fraction of all observed Seyfert 1 nuclei, and many QSOs as well, have been observed to vary in light. Among those which vary with reasonably large amplitude for which time-resolved spectral data are available, the strengths of the broad emission lines and their profiles have also been observed to vary. From the physical picture it is seen that the gas in the BLR is photoionized by continuum radiation from the central source, and that the photoionizing continuum varies together with the optical continuum, these variations must give information on the size and geometry of the BLR. This subject has been very well summarized recently by Peterson (1988).
The first observational recognition of these time-variable broad-line profiles occurred from comparisons of spectra of the same object taken at times separated by months or even years. Systematic study of the phenomenon requires observations at much more frequent intervals. These are difficult to arrange, except with a telescope, spectral scanner and a team of observers dedicated to the project. The most complete data now available have observations of NGC 5548 with average sampling intervals of 3 to 4 nights. In general, the response of a finite BLR to a pulse of photoionizing continuum radiation is expected to occur on a time scale LT = R/c, the light-travel time across the region. (The recombination time, on which the gas responds to the pulse on its arrival, is short compared to typical light-travel times.) Cross-correlating the continuum and individual broad emission-line light curves (for instance H I L, Mg II 2798, C III] 1909, C IV 1549, or H) for different assumed lags, and finding the lag for which the cross-correlation is a maximum is the most objective method of determining the light-travel time. This phase shift depends on the geometrical structure of the object and can be calculated for simple models; it is straightforward to show that, for instance, for a thin spherical shell of gas centered on the nucleus, the phase shift is exactly LT as defined previously (Gaskell and Sparke 1986).
In these space-based observations of NGC 5548, different time delays have been found for different lines, namely R 4 to 10 light days for the highest ionization He II 1640 and N V 1240 emission lines, ~ 12 light days for H I L, 8 to 16 for C IV 1549, and 26 to 32 for C III] 1909, while for Mg II 2798 the variations are small and the phase lag is not well determined but is even longer (Clavel et al 1991). These results agree with the photoionization model predictions that the degree of ionization decreases outward from the central source (unless the density decreases very strongly) and the usual assumption that the density probably also decreases outward, but not faster than R-2. Ground based measurements on the variations of H and H in NGC 5548 by Netzer et al (1990) over a 5-month interval in 1988 give a phase lag corresponding to R = 7 light days. These authors assumed various simplified models for the BLR structure, used the observed continuum light curve as the input photoionizing spectrum, calculated the response of each model, and varied the parameters to get the best fit. For a spherical thin-shell model they found R = 7±3 light days. (This agrees with the IUE results for L.) Thin shells with R 14 light days could be excluded at the 95% confidence level, likewise thick shells with inner radius Ri = 4 light days and outer radius Ro > 40 light days, also thin disks inclined by 60° to the line of sight with Ri = 7 light days and Ro 30 light days. On the other hand a large consortium of observers, observing the same nucleus over 10 months in 1988-89, with what appeared to be better signal-to-noise-ratio data and at a time of larger variation, found a phase lag R ~ 20 light days (Peterson et al 1991). Their analysis is still in progress, but there appears to be a clear contradiction between the two sets of data on the phase lag of the H variation. Note the difference with the phase lag ~ 12 days found for L for essentially the same time interval (Clavel et al 1991). It is not theoretically understood at this time.
For this object the standard photoionization model which best fits the relative emission-line strengths, = 10-2 (Mushotzky and Ferland 1984), has, for mean Ne = 1010 cm-3, a radius R = 150 light days, larger by a factor 20 than that implied by the time-variation measurements interpreted by the spherical thin-shell models of Netzer et al (1990).
Two Seyfert AGNs extensively studied for variation are NGC 4151 and Akn 120. For NGC 4151 the early data gave a wide range of values of LT (or R, expressed in light days) for different lines, but closer analysis shows the differences are dominated by observational uncertainties resulting from inadequate time sampling and insufficient signal-to-noise-ratio data. All the measurements agree with R ~ 7 light days, with an uncertainty by a factor two. This does not disagree significantly with the value from the static photoionization model that best fits the observed mean broad-line spectrum, R ~ 16 light days.
The recent optical variation measurements for NGC 4151 are among the best available for any AGN, because this low-luminosity Seyfert 1 is close and relatively bright. Hence it was measured with a very good signal-to-noise ratio (Maoz et al 1991). The cross-correlation technique, applied to the broad H and H emission lines, give R = 9 ± 2 days, in reasonable agreement with the previous measurements and with the photoionization models. However, with the recent best data it proved possible to investigate the `transfer function', or response of several possible simplified BLR models to the light variation of the central continuum source. The models with a large ratio of the outer to inner radii (~ 10) and with the line emission either constant or decreasing outward from the center fit the observed variations best.
On the other hand for Akn 120 the cross-correlation technique gives R 30 light days, but the ionization parameter that best fits the mean observed spectrum then demands Ne 1011 cm-3. Thus for some AGNs, including NGC 4151, the best BLR time-variation data agree with the photoionization models for mean densities Ne 1010 cm-3, but for several others including NGC 5548, Akn 120 and Mrk 279 (Maoz et al 1990) the sizes, at least for H, are smaller than expected at these densities. For this reason the models with mean densities as high as Ne = 1013 previously mentioned are being investigated. Comparisons of their predictions with the observed spectra seem to show that most of the line emissions comes from lower densities, with Ne = 1010 cm-3 probably a more representative mean density, but up to Ne = 1011 cm-3 is possible (Rees et al 1989, Ferland and Persson 1989). Some of these BLR models assume a wide range of densities, with a power-law decrease outward. A crucial test is the C III 977 / C III] 1909 line ratio, which is predicted to be about 4 at Ne = 1011 cm-3, but about 0.3 if Ne = 1010 cm-3 both at T = 15 000 K, a reasonable temperature estimate (Nussbaumer and Schild 1979). There are few published observations of C III 977, partly because of its short wavelength and the related problem that it lies in the L `forest' of absorption lines in the spectra of distant QSOs. However, in the IUE observations of six AGNs with the IUE by Green et al (1980), four show C III] 1909 emission, and two of these are listed as having measured C III 977, with intensity ratios I(C III 977) / I(C III] 1909) = 2 for PKS 1302-102 and 0.9 for PG 1247 + 268. These would indicate mean densities (as measured by the C III lines) around 3 x 1010 cm-3 for these two AGNs. However, later, better data suggest that the purported measurement of 977 for PKS 1302-102 was not correct. For several other QSOS observed with the IUE, C III 977 has not been detected (Gondhalekar 1990a, Kinney et al 1991). For these objects a very rough upper limit appears to be I(977) / I(1909) < 0.5, corresponding to mean densities on any models published to date Ne < 3 x 1010 cm-3. All these QSOs, however, are considerably more luminous than the Seyfert 1 galaxies observed for variability, and the question of the viability of standard photoionization models is definitely an open one at present.
An alternative interpretation to the smaller size and consequent higher densities derived from time variations is that the geometry is anisotropic. If the ray from varying nucleus to the gas which responds to the variation makes an angle to the ray from the nucleus to the observer, then the time lag of the variation by this element of gas is r(1 - cos) / c. If the gas which is observed to vary lies within a cone of small half-angle , the phase lag
may be much less than R/c. This type of model, with a cylindrical geometry, a small opening angle, and the rear side occulted was proposed for 3C 446 by Bregman et al (1986). It is the preferred interpretation of Perez et al (1989a, 1989b) for their time variation data for many AGNs. It would seem to require that only a small fraction of AGNs be observed to vary. However, if the radiation is seen down the axis of a torus, which is opaque in directions near its equatorial plane, so the BLR itself is not visible from those directions, all observed BLRs could be variable. These cylindrically symmetric type models are discussed in section 4.
Some data are also available on the variability of the broad emission lines L and C IV 1549 in high-luminosity QSOs, from multiple IUE observations (Gondhalekar 1990b). Since the observed emission-line spectra are essentially independent of luminosity, the ionization parameter and electron density Ne are supposed to be independent of L and hence the time scale for variation
However, the available observational data do not show this dependence, but in fact suggest that the time scale for variation is more nearly independent of luminosity. Thus again, more sophisticated models, probably with variable density and non-spherical geometry, will be necessary to fit the observed AGNs (Gondhalekar 1990b).
Variations are not expected in the `ordinary' narrow emission lines in times less than hundreds of years, and none have been observed. However, variations have been observed in [Fe X] 6375 in a few high-ionization AGNs, most(certainly in NGC 5548, on time scales of a few years. This is expected from photoionization models, as such high stages of ionization as Fe+9 are predicted to be at distances of 0.1 to 1 pc from the ionization source (Veilleux 1988).
2.8. Completeness of Seyfert galaxy samples
As mentioned in section 1, the first known Seyfert galaxies were recognized and classified on the basis of their emission lines on slit spectrograms. As only a few percent of luminous galaxies are Seyferts, this is a relatively inefficient method of finding new ones. Since many Seyfert 1 and some Seyfert 2 nuclei have strong blue and ultraviolet continua, it is possible to identify candidates on this basis from low-dispersion objective-prism photographic spectra surveys. This program was very successfully carried out at Byurakan Observatory by the late B. E. Markarian and his collaborators, and is now generally referred to as the First Byurakan Survey or FBS (Lipovetsky et al 1987). About 10% of the galaxies isolated by their ultraviolet continuum in this survey turned out to be Seyfert galaxies, on the basis of slit spectra; the other 90% are mostly star-burst galaxies in which the blue continuum comes from the hot stars rather than the accretion disk. More recently Markarian began a second survey (SBS), which has been carried on since his death by his collaborators, using fine-grain plates, objective prisms of various dispersions, and multiple limiting exposures with the prisms in different orientations (Markarian et al 1987). With these techniques they have been able to find much fainter Seyfert galaxies than had been previously recognized. There are many examples as faint as apparent magnitude mB = 17 or 18. In many cases slit spectra are obtained with the 6-m telescope before the identification is published. Nearly every galaxy listed in this SBS has been confirmed as a Seyfert galaxy. However, the level of completeness, that is what fraction of the Seyferts in the field have been found, is not at all well known.
Many of the known Seyfert galaxies were first identified as candidates in the FBS. But many Seyfert 2 galaxies were not found, because their blue continua are too faint to be picked up. Programs for obtaining slit spectra of all emission-line galaxies, or all galaxies down to a given magnitude, such as the Center for Astrophysics (CfA) redshift survey, have shown this by turning up many `new' Seyfert galaxies (Huchra et al 1982, Phillips et al 1983). By comparison with the CfA survey, it is known that among the brighter galaxies the FBS is 67% complete for Seyfert 1s, but only 44% for Seyfert 2s (Lipovetsky et al 1987).
Other, more recent, objective-prism surveys have aimed at finding Seyfert-galaxy candidates primarily by their emission lines, especially [O III] 4959, 5007, rather than by the blue continuum. The largest area is covered by the University of Michigan survey (MacAlpine and Williams 1981). This method was applied, using a relatively high-dispersion objective prism to detect relatively weak emission lines, to a large field surrounding the North Galactic Pole (Wasilewski 1983). The limit of completeness of such a survey clearly cannot really be given simply in terms of a limiting apparent magnitude; instead, it depends on a combination of the magnitude and the strength of the emission line or lines detected. Faint galaxies with strong emission lines can be detected as well as brighter galaxies with weaker emission lines (Salzer 1989).
Thus a true complete magnitude-limited sample is difficult to obtain, or even to define. Earlier attempts at such samples were seriously deficient in lower-luminosity Seyfert 2 galaxies. The best approximation to a complete, magnitude-limited sample is one drawn from a slit-spectrum survey like the CfA, and taking the sample only down to a magnitude limit well above that of the catalogue. This leaves only a few objects, but is relatively well defined physically (Edelson 1987). Even so, it implies some ill-defined limit to the strength of the emission lines. It is clearly quite possible that many galaxies that are not called Seyfert galaxies have the same or similar phenomenon going on in their nuclei, but the characteristic emission lines are too faint to be detected. Some of the LINERS discussed in section 6.4 fit into this category, but `less active' less luminous, nuclei may well remain undetected.