A Brief History of Active Galactic Nuclei

The basic parameters of the region of gas emitting the narrow emission lines were fairly quickly established. In one of the first physical analyses of "emission nuclei" in galaxies, Woltjer (1959) derived a density N_e≈10⁴ cm⁻³ and temperature T≈20,000 K from the [S ii] and [O iii] line ratios of Seyfert galaxies. The region emitting the narrow lines was just resolved for the nearest Seyfert galaxies, giving a diameter of order 100 pc (e.g., Walker 1968; Oke & Sargent 1968). Oke & Sargent derived a mass of ∼10⁵ M_⊙ and a small volume filling factor for the narrow line gas in NGC 4151. Burbidge, Burbidge, & Prendergast (1958) found that the nuclear emission lines of NGC 1068 were much broader than could be accounted for by the rotation curve of the galaxy and concluded that the material was in a state of expansion.

A key question was why, in objects showing broad wings, these were seen on the permitted lines but not the forbidden lines. (Seyfert galaxies with broad wings came to be called "Seyfert 1" or "Sy 1" and those without them "Sy 2" [Khachikian & Weedman 1974].) Were these wings emitted by the same gas that emits the narrow lines? Woltjer (1959) postulated a separate region of fast‐moving, possibly gravitationally bound gas to produce the broad Balmer line wings of Seyfert galaxies. Souffrin (1969a) adopted such a model in her analysis of NGC 3516 and NGC 4151. Alternatively, broad Balmer line wings might be produced by electron scattering (Burbidge et al. 1966). Oke & Sargent (1968) supported this possibility for NGC 4151. Their analysis of the emission‐line region gave an electron scattering optical depth τ_e∼0.1. Multiple scattering of Balmer line photons by the line opacity might increase the effective electron scattering probability, explaining the presence of wings only on the permitted lines. However, analysis of electron scattering profiles by other authors (e.g., Weymann 1970) indicated the need for a dense region only a tiny fraction of a light year across. Favoring mass motions were the irregular broad line profiles in some objects (Anderson 1971), which demonstrated the presence of bulk velocities of the needed magnitude. In addition, Shklovskii (1964) had argued for an electron scattering optical depth τ_es<1 in 3C 273 to avoid excessive smoothing of the continuum light variations. The picture of broad lines from a small region of dense, fast‐moving clouds ("broad line region" or BLR) and narrow lines from a larger region of slower moving, less dense clouds ("narrow line region" or NLR) found support from photoionization modes (Shields 1974).

Early workers (e.g., Seyfert 1943) had noted that the narrow line intensities resembled those of planetary nebulae, and photoionization was an obvious candidate for the energy input to the emitting gas for both the broad and narrow lines. For 3C 273, Shklovskii (1964) noted that the kinetic energy of the emission‐line gas could power the line emission only for a very short time, whereas the extrapolated power in ionizing ultraviolet radiation was in rough agreement with the emission‐line luminosities. Osterbrock & Parker (1965) argued against photoionization because of the observed weakness of the Bowen O iii fluorescence lines. Also eliminating thermal collisional ionization because of the observed wide range of ionization stages, they proposed ionization and heating by fast protons resulting from high‐velocity cloud collisions. Souffrin (1969b) rejected this on the basis of thermal equilibrium considerations and argued along with Williams & Weymann (1968) that thermal collisional ionization was inconsistent with observed temperatures. Noting that an optical–ultraviolet continuum of roughly the needed power is observed, and that the thermal equilibrium gives roughly the observed temperature, Souffrin concluded that a nonthermal ultraviolet continuum was "the only important source of ionization." Searle & Sargent (1968) likewise noted that the equivalent widths of the broad Hβ emission lines were similar among AGN over a wide range of luminosity and were consistent with an extrapolation of the observed "nonthermal" continuum as a power law to ionizing frequencies. Detailed models of gas clouds photoionized by a power‐law continuum were calculated with the aid of electronic computers, with application to the Crab Nebula, binary X‐ray sources, and AGN (Williams 1967; Tarter & Salpeter 1969; Davidson 1972; MacAlpine 1972). Such models showed that photoionization can account for the intensities of the strongest optical and ultraviolet emission lines. In particular, the penetrating high‐frequency photons can explain the simultaneous presence of very high ionization stages and strong emission from low‐ionization stages, in the context of a "nebula" that is optically thick to the ionizing continuum. Photoionization quickly became accepted as the main source of heating and ionization in the emission‐line gas.

Attention then focused on improving photoionization models and understanding the geometry and dynamics of the gas emitting the broad lines. It was clear that the emitting gas had only a tiny volume filling factor, and one possible possible geometry was the traditional nebular picture of clouds or "filaments" scattered through the BLR volume. Photoionization models typically assumed a slab geometry representing the ionized face of a cloud that was optically thick to the Lyman continuum. Model parameters included the density and chemical composition of the gas and the intensity and energy distribution of the incident ionizing continuum. Various line ratios, such as C iii]/C iv, were used to constrain the "ionization parameter," i.e., the ratio of ionizing photon density to gas density. Chemical abundances were assumed to be approximately solar but were hard to determine because the high densities prevented a direct measurement of the electron temperature from available line ratios.

A challenge for photoionization models was the discovery that the Lyα/Hα ratio was an order of magnitude smaller than the value of ∼50 predicted by photoionization models at the time (Baldwin 1977a; Davidsen, Hartig, & Fastie 1977). This stimulated models with an improved treatment of radiative transfer in optically thick hydrogen lines (e.g., Kwan & Krolik 1979). These models found strong Balmer line emission from a "partially ionized zone" deep in the cloud, heated by penetrating X‐rays, from which Lyman line emission was unable to escape. The models still did not do a perfect job of explaining the observed ratios (e.g., Lacy et al. 1982) of the Paschen, Balmer, and Lyman lines. Models by Collin‐Souffrin, Dumont, & Tully (1982) and Wills, Netzer, & Wills (1985) suggested the need for densities as high as N_e≈10¹¹ cm⁻³ to explain the Hα/Hβ ratio.

The X‐ray–heated region also was important for the formation of the strong Fe ii multiplet blends observed in the optical and ultraviolet. Theoretical efforts by several authors culminated in models involving thousands of Fe lines, with allowance for the fluorescent interlocking of different lines (Wills et al. 1985). These models enjoyed some success in explaining the relative line intensities, but the total energy in the Fe ii emission was less than observed. Although some of this discrepancy might involve the iron abundance, Collin‐Souffrin et al. (1980) proposed a separate Fe ii–emitting region with a high density ( N_e≈10¹¹ cm⁻³) heated by some means other than photoionization. This region might be associated with an accretion disk. The Fe ii emission and the Balmer continuum emission that combined to form the 3000 Å "little bump" still are not fully explained, nor is the tendency for radio‐loud AGN to have weaker Fe ii and steeper Balmer decrements than radio‐quiet objects (Osterbrock 1977).

A tendency for the equivalent width of the C iv emission line to decrease with increasing luminosity was found by Baldwin (1977b). Explanations of this involved a possible decrease, with increasing luminosity, in the ionization parameter and in the "covering factor," i.e., the fraction (Ω/4π) of the ionizing continuum intercepted by the BLR gas (Mushotzky & Ferland 1984). The ionization parameter was also the leading candidate to explain the difference in ionization level between classical Seyfert galaxies and the "low‐ionization nuclear emission regions" or "LINERs" (Heckman 1980; Ferland & Netzer 1983; Halpern & Steiner 1983).

The geometry and state of motion of the BLR gas has been a surprisingly stubborn problem. If the BLR was a swarm of clouds, they might be falling in (possibly related to the accretion supply), orbiting, or flying out. Alternatively, the gas might be associated with an accretion disk irradiated by the ionizing continuum (e.g., Shields 1977; Collin‐Souffrin 1987). Except for the BAL QSOs, there was little evidence for blueshifted absorption analogous to the P Cygni–type line profiles of stars undergoing vigorous mass loss. The approximate symmetry of optically thick lines such as Lyα and Hα suggested that the motion was circular or random rather than predominantly radial (e.g., Ferland, Netzer, & Shields 1979). However, for orbiting (or infalling) gas, the line widths implied rather large masses for the central object, given prevailing estimates of the BLR radius. In addition, gas in Keplerian orbit seemed likely to give a double‐peaked line profile or to have other problems (Shields 1978a). In the face of these conflicting indications, the most common assumption was that the gas took the form of clouds flying outward from the central object. The individual clouds would disperse quickly unless confined by some intercloud medium, and a possible physical model was provided by the two‐phase medium discussed by Krolik, McKee, & Tarter (1981). Radiation pressure of the ionizing continuum, acting on the bound‐free opacity of the gas, seemed capable of producing the observed velocities and giving a natural explanation of the "logarithmic" shape of the observed line profiles (Mathews 1974; Blumenthal & Mathews 1975). Interpretation of the line profiles was complicated by the recognition of systematic offsets in velocity between the high‐ and low‐ionization lines (Gaskell 1982; Wilkes & Carswell 1982; Wilkes 1984).

A powerful new tool was provided by the use of "echo mapping" or "reverberation mapping" of the BLR. Echo mapping relies on the time delays between the continuum and line variations caused by the light travel time across the BLR (Blandford & McKee 1982). Early results showed that the BLR is smaller and denser than most photoionization models had indicated (Ulrich et al. 1984; Peterson et al. 1985). Masses of the central object, by this time assumed to be a black hole, could be derived with increased confidence. The smaller radii implied smaller masses that seemed reasonable in the light of other considerations, and the idea of gravitational motions for the BLR gained in popularity. This was supported by the rough tendency of the line profiles to vary symmetrically, consistent with "chaotic" or circular motions (e.g., Ulrich et al. 1984).

The question of the ultimate energy source for AGN stimulated creativity even before the discovery of QSO redshifts. The early concept of radio galaxies as galaxies in collision gave way to the recognition of galactic nuclei as the sites of concentrated, violent activity. Burbidge (1961) suggested that a chain reaction of supernovae (SN) could occur in a dense star cluster in a galactic nucleus. Shock waves from one SN would compress neighboring stars, triggering them to explode in turn. Cameron (1962) considered a coeval star cluster leading to a rapid succession of SN as the massive stars finished their short lives. Spitzer & Saslaw (1966), building on earlier suggestions, developed another model involving a dense star cluster. The cluster core would evolve to higher star densities through gravitational "evaporation," and this would lead to frequent stellar collisions and tidal encounters, liberating large amounts of gas. Additional ideas involving dense star clusters included pulsar swarms (Arons, Kulsrud, & Ostriker 1975) and starburst models (Terlevich & Melnick 1985).

Hoyle & Fowler (1963a, 1963b) discussed the idea of a supermassive star (up to ∼10⁸ M_⊙) as a source of gravitational and thermonuclear energy. In additional to producing large amounts of energy per unit mass, all these models seemed capable of accelerating particles to relativistic energies and producing gas clouds ejected at speeds of ∼5000 km s⁻¹, suggestive of the broad emission line wings of Seyfert galaxies. In this regard, Hoyle & Fowler (1963a) suggested that "a magnetic field could be wound toroidally between the central star and a surrounding disk." The field could store a large amount of energy, leading to powerful "explosions" and jets like that of M87. Hoyle & Fowler (1963b) suggested that "only through the contraction of a mass of 10⁷–10⁸ M_⊙ to the relativistic limit can the energies of the strongest sources be obtained."

Soon after, Salpeter (1964) and Zeldovich (1964) proposed the idea of QSO energy production from accretion onto a supermassive black hole. For material gradually spiraling to the innermost stable orbit of a nonrotating black hole at r = 6GM/c², the energy released per unit mass would be 0.057c², enough to provide the energy of a luminous QSO from a reasonable mass. Salpeter imagined some kind of turbulent transport of angular momentum, allowing the matter to move closer to the hole, which would grow in mass during the accretion process.

The black hole model received limited attention until Lynden‐Bell (1969) argued that dead quasars in the form of "collapsed bodies" (black holes) should be common in galactic nuclei, given the lifetime energy output of quasars and their prevalence at earlier times in the history of the universe. Quiescent ones might be detectable through their effect on the mass‐to‐light ratio of nearby galactic nuclei. Lynden‐Bell explored the thermal radiation and fast particle emission to be expected in a disk of gas orbiting the hole, with energy dissipation related to magnetic and turbulent processes. For QSO luminosities, the disk would have a maximum effective temperature of ∼10⁵ K, possibly leading to photoionization and broad line emission. He remarked that "with different values of the [black hole mass and accretion rate] these disks are capable of providing an explanation for a large fraction of the incredible phenomena of high‐energy astrophysics, including galactic nuclei, Seyfert galaxies, quasars and cosmic rays."

Further evidence for relativistic conditions in AGN came from other theoretical arguments. Hoyle, Burbidge, & Sargent (1966) noted that relativistic electrons emitting optical and infrared synchrotron radiation would also Compton scatter ambient photons, boosting their energy by large factors. This would lead to "repeated stepping up of the energies of quanta," yielding a divergence that came to be known as the "inverse Compton catastrophe." This would be attended by rapid quenching of the energy of the electrons. They argued that this supported the idea of noncosmological redshifts. In response, Woltjer (1966) invoked a model with electrons streaming radially on field lines, which could greatly reduce Compton losses. He further noted that because "the relativistic electrons and the photons they emit both move nearly parallel to the line of sight, the time scale of variations in emission can be much shorter than the size of the region divided by the speed of light." The emission would also likely be anisotropic, reducing the energy requirements for individual objects.

Dramatic confirmation of the suspected relativistic motions came from the advancing technology of radio astronomy. Radio astronomers using conventional interferometers had shown that many sources had structure on a subarcsecond scale. Scintillation of the radio signal from some AGN, caused by the interplanetary medium of our solar system, also implied subarcsecond dimensions (Hewish, Scott, & Wills 1964). The compact radio sources in some AGN showed flat spectrum components and variability on timescales of months (Dent 1965; Sholomitsky 1965). The variability suggested milliarcsecond dimensions on the basis of light travel time arguments. The spectral shape and evolution found explanation in terms of multiple, expanding components that were optically thick to synchrotron self‐sbsorption, which causes a low‐frequency cutoff in the emitted continuum (Pauliny‐Toth & Kellermann 1966, and references therein). Such models had interesting theoretical consequences, including angular sizes (for cosmological redshifts) as small as 10⁻³ arcsec, and large amounts of energy in relativistic electrons, far exceeding the energy in the magnetic field.

These inferences made clear the need for angular resolution finer than was practical with conventional radio interferometers connected by wires or microwave links. This was achieved by recording the signal from the two antennas separately on magnetic tape, and correlating the recorded signals later by analog or digital means. This technique came to be known as "very long baseline interferometry" (VLB, later VLBI). After initial difficulties finding "fringes" in the correlated signal, competing groups in Canada and the United States succeeded in observing several AGN in the spring of 1967, over baselines of roughly 200 km (see Cohen et al. 1968). The U.S. experiments typically used the 140 foot antenna at the National Radio Astronomy Observatory in Green Bank, West Virginia, in combination with increasingly remote antennas in Maryland, Puerto Rico, Massachusetts, California, and Sweden. The latter gave an angular resolution of 00006. Within another year, observations were made between Owens Valley, California, and Parkes, Australia, a baseline exceeding 10,000 km or 80% of the Earth's diameter. A number of AGN showed components unresolved on a scale of 10⁻³ arcsec.

On 1970 October 14 and 15, Knight et al. (1971) observed quasars at 7840 MHz with the Goldstone, California–Haystack, Massachusetts, "Goldstack" baseline. 3C 279 showed fringes consistent with a symmetrical double source separated by (1.55 ± 0.03) × 10^{- 3} arcsec. Later observations on 1971 February 14 and 26 by Whitney et al. (1971) showed a double source structure at the same position angle, but separated by a distinctly larger angle of (1.69 ± 0.02) × 10^{- 3} arcsec. Given the distance implied by the redshift of 0.538, this rate of angular separation corresponded to a linear separation rate of 10 times the speed of light! Cohen et al. (1971), also using Goldstack data, observed "superlight expansion" in 3C 273 and 3C 279. Whitney et al. and Cohen et al. considered a number of interpretations of their observations, including multiple components that blink on and off (the "Christmas tree model") and noncosmological redshifts. However, most astronomers quickly leaned toward an explanation involving motion of emitting clouds ejected from the central object at speeds close to, but not exceeding, the speed of light. Rees (1966) had calculated the appearance of relativistically expanding sources, and apparent expansion speeds faster than that of light were predicted. A picture emerged in which a stationary component was associated with the central object, and clouds were ejected at intervals of several years along a fairly stable axis. (Repeat ejections were observed in the course of time by VLBI experiments.) If this ejection occurred in both directions, it could supply energy to the extended double sources. The receding components would be greatly dimmed by special relativistic effects, while the approaching components were brightened. The two observed components are then associated with the central object and the approaching cloud, respectively. The fact that the two observed components had roughly equal luminosities found an explanation in the relativistic jet model of Blandford & Königl (1979).

Apparent superluminal motion has now been seen in a number of quasars and radio galaxies, and a possibly analogous phenomenon has been observed in connection with black hole systems of stellar mass in our Galaxy (Mirabel & Rodriguez 1994).

On 1962 June 18, an Aerobee sounding rockets blasted skyward from White Sands proving ground in New Mexico. It carried a Geiger counter designed to detect astronomical sources of X‐rays. The experiment, carried out by Giacconi et al. (1962), discovered an X‐ray background and a "large peak" in a 10° error box near the Galactic center and the constellation Scorpius. A rocket experiment by Bowyer et al. (1964) also found an isotropic background, confirmed the Scorpius source, and detected X‐rays from the Crab Nebula. Friedman & Byram (1967) identified X‐rays from the active galaxy M87. A rocket carrying collimated proportional counters sensitive in the 1–10 keV energy range, found sources coincident with 3C 273, NGC 5128 (Cen A), and M87 (Bowyer, Lampton, & Mack 1970). The positional error box for 3C 273 was small enough to give a probability of less that 10⁻³ of a chance coincidence. The X‐ray luminosity, quoted as ∼10⁴⁶ ergs s⁻¹, was comparable with quasar's optical luminosity.

The first dedicated X‐ray astronomy satellite, Uhuru, was launched in 1970. Operating until 1973, it made X‐ray work a major branch of astronomy. X‐rays were reported from the Seyfert galaxies NGC 1275 and NGC 4151 (Gursky et al. 1971). The spectrum of NGC 5128 was consistent with a power law of energy index α = - 0.7, where L_ν∝ν^α; and there was low energy absorption corresponding to a column density of 9 × 10²² atoms cm⁻², possibly caused by gas in the nucleus (Tucker et al. 1973). Early variability studies were hampered by the need to compare results from different experiments, but Winkler & White (1975) found a large change in the flux from Cen A in only 6 days from OSO 7 data. Using Ariel V observations of NGC 4151, Ives, Sanford, & Penston (1976) found a significant increase in flux from earlier Uhuru measurements. Marshall, Warwick, & Pounds (1981), using Ariel V data on AGN gathered over a 5 year period, found that roughly half of the sources varied by up to a factor of 2 on times less than or equal to a year. A number of sources varied in times of 0.5–5 days. Marshall et al. articulated the importance of X‐ray variability observations, which show that the X‐rays "arise deep in the nucleus" and "relate therefore to the most fundamental aspect of active galaxies, the nature of the central 'power house'."

Strong X‐ray emission as a characteristic of Sy 1 galaxies was established by Martin Elvis and his coworkers from Ariel V data (Elvis et al. 1978). This work increased to 15 the number of known Seyfert X‐ray sources, of which at least three were variable. Typical luminosities were ∼10^42.5–10^44.5 ergs s⁻¹. The X‐ray power correlated with the infrared and optical continuum and Hα line. Seyfert galaxies evidently made a significant contribution to the X‐ray background, and limits could be set on the evolution of Seyfert galaxy number densities and X‐ray luminosities in order that they not exceed the observed background. Elvis et al. considered thermal bremsstrahlung (10⁷ K), synchrotron, and synchrotron self‐Compton models of the X‐ray emission.

HEAO 1, the first of the High Energy Astronomy Observatories, was an X‐ray facility that operated from 1977 to 1979. It gathered data on a sufficient sample of objects to allow comparisons of different classes of AGN and to construct a log N–log S diagram and improved luminosity function. HEAO 1 provided broad‐band X‐ray spectral information for a substantial set of AGN, showing spectral indices α≈ - 0.7, with rather little scatter, and absorbing columns less than 5 × 10²² cm⁻² (Mushotzky et al. 1980).

The Einstein Observatory (HEAO 2) featured grazing incidence focusing optics allowing detection of sources as faint as ∼10⁻⁷ the intensity of the Crab Nebula. Tananbaum et al. (1979) used Einstein data to study QSOs as a class of X‐ray emitters. Luminosities of 10⁴³–10⁴⁷ ergs s⁻¹ (0.5–4.5 keV) were found. OX 169 varied substantially in under 10,000 s, indicating a small source size. This suggested a black hole mass not greater than 2 × 10⁸ M_⊙, if the X‐rays came from the inner portion of an accretion flow. By this time, strong X‐ray emission was established as a characteristic of all types of AGN and a valuable diagnostic of their innermost workings.

Today, the word "continuum" in the context of AGN might bring to mind anything from radio to gamma‐ray frequencies. However, in the early days of QSO studies, the term generally meant the optical continuum, extending to the ultraviolet and infrared as observations in these bands became available. Techniques of photoelectric photometry and spectrum scanning were becoming established as QSO studies began. The variability of QSOs, including 3C 48 and 3C 273 (e.g., Sandage 1963), was known and no doubt contributed to astronomers' initial hesitation to interpret QSO spectra in terms of large redshifts. In his contribution to the four discovery papers on 3C 273, Oke (1963) presented spectrophotometry showing a continuum slope L_ν∝ν^{+ 0.3} in the optical, becoming redder toward the near‐infrared. He noted that the energy distribution did not resemble a blackbody, and inferred that there must be a substantial contribution of synchrotron radiation.

A key issue for continuum studies has been the relative importance of thermal and nonthermal emission processes in various wavebands. Early work tended to assume synchrotron radiation, or "nonthermal emission," in the absence of strong evidence to the contrary. The free‐free and bound‐free emission from the gas producing the observed emission lines was generally a small contribution. The possibility of thermal emission from very hot gas was considered for some objects such as the flat blue continuum of 3C 273 (e.g., Oke 1966). The energy distributions tend to slope up into the infrared; and for thermal emission from optically thin gas, this would would have required a rather low temperature and an excessive Balmer continuum jump. This left the possibilities of nonthermal emission or thermal emission from warm dust, presumably heated by the ultraviolet continuum.

Observational indicators of thermal or nonthermal emission include broad features in the energy distribution, variability, and polarization. For the infrared, one also has correlations with reddening, the silicate absorption and emission features, and possible angular resolution of the source (Edelson et al. 1988). For some objects, rapid optical variability implied brightness temperatures that clearly required a nonthermal emission mechanism. For example, Oke (1967) observed day‐to‐day changes of 0.25 and 0.1 mag for 3C 279 and 3C 446, respectively. For many objects, the energy distributions were roughly consistent with a power law of slope near ν^−1.2. Power laws of similar slopes were familiar from radio galaxies and the Crab Nebula, where the emission extended through the optical band. These spectra were interpreted in terms of synchrotron radiation with power‐law energy distributions for the radiating, relativistic electrons. Such a power‐law energy distribution was also familiar from studies of cosmic rays, and thus power laws seemed natural in the context of high‐energy phenomena like AGN. In addition to simple synchrotron radiation, there might be a hybrid process involving synchrotron emission in the submillimeter and far‐infrared, with some of these photons boosted to the optical by "inverse" Compton scattering (Shklovskii 1965). The idea of a nonthermal continuum in the optical, whose high‐frequency extrapolation provided the ionizing radiation for the emission‐line regions, was widely held for many years. This was invoked not only for QSOs but also for Seyfert galaxies, where techniques such as polarization were used to separate the "nonthermal" and galaxy components (e.g., Visvanathan & Oke 1968).

Infrared observations were at first plagued by low sensitivity and inadequate telescope apertures. Measurements of 3C 273 in the K filter (2.2 μm), published by Johnson (1964) and Low & Johnson (1965), showed a continuum steeply rising into the infrared. Infrared radiation from NGC 1068 was observed by Pacholczyk & Wisniewski (1967), also with a flux density (F_ν) strongly rising to the longest wavelength observed ("N" band, or 10 μm). The infrared radiation dominated the power output of this object. Becklin et al. (1973) found that much of the 10 μm emission from NGC 1068 came from a resolved source 1 ^'' (90 pc) across and concluded that most of the emission was not synchrotron emission. In contrast, variability of the 10 μm emission from 3C 273 (e.g., Rieke & Low 1972) pointed to a strong nonthermal component. Radiation from hot dust has a minimum source size implied by the blackbody limit on the surface brightness, and this is more stringent for longer wavelengths radiated by cooler dust. This in turn implies a minimum variability timescale as a function of wavelength. The near‐infrared emission of NGC 1068 was found to be strongly polarized (Knacke & Capps 1974).

Improving infrared technology, and optical instruments such as the multichannel spectrometer on the 200 inch telescope (Oke 1969), led to larger and better surveys of the AGN continuum. Oke, Neugebauer, & Becklin (1970) reported observations of 28 QSOs from 0.3 to 2.2 μm. The energy distributions were similar in radio‐loud and radio‐quiet QSOs. They found that the energy distributions could generally be described as a power law (index −0.2 to −1.6 for F_ν∝ν^α) and that they remained "sensibly unchanged" during the variations of highly variable objects. Penston et al. (1974) studied the continuum from 0.3 to 3.4 μm in 11 bright Seyfert galaxies. All turned up toward the infrared, and consideration of the month‐to‐month variability pointed to different sources for the infrared and optical continua. From an extensive survey of Seyfert galaxies, Rieke (1978) concluded that strong infrared emission was a "virtually universal" feature and that the energy distributions in general did not fit a simple power law. The amounts of dust required were roughly consistent with the expected dust in the emission‐line gas of the active nucleus and the surrounding interstellar medium. A consensus emerged that the infrared emission of Seyfert 2 galaxies was thermal dust emission, but the situation for Seyfert 1 galaxies was less clear (e.g., Neugebauer et al. 1976; Stein & Weedman 1976). From a survey of the optical and infrared energy distribution of QSOs, Neugebauer et al. (1979) concluded that the slope was steeper in the 1–3 μm band than in the 0.3–1 μm band and that an apparent broad bump around 3 μm might be dust emission. Neugebauer et al. (1987) obtained energy distributions from 0.3 to 2.2 μm for the complete set of quasars in the Palomar‐Green (PG) survey (Green, Schmidt, & Liebert 1986) as well as some longer wavelength observations. A majority of objects could be fitted with two power laws ( α≈ - 1.4 at lower frequencies, α≈ - 0.2 at higher frequencies) plus a "3000 Å bump."

Measurements at shorter and longer wavelengths were facilitated by the International Ultraviolet Explorer (IUE) and the Infrared Astronomical Satellite (IRAS), launched in 1978 and 1983, respectively. Combining such measurements with ground‐based data, Edelson & Malkan (1986) studied the spectral energy distribution of AGN over the wavelength range 0.1–100 μm. The 3–5 μm "bump" was present in most Seyfert galaxies and QSOs, involving up to 40% of the luminosity between 2.5 and 10 μm. All Sy 1 galaxies without large reddening appeared to require a hot thermal component, identified with the increasingly popular concept of emission from an accretion disk. Edelson & Malkan (1987) used IRAS observations to study the variability of AGN in the far‐infrared. The high‐polarization objects varied up to a factor 2 in a few months, but no variations greater than 15% were observed for "normal" quasars or Seyfert galaxies. The former group was consistent with a class of objects known as "blazars" that are dominated at all wavelengths by a variable, polarized nonthermal continuum. Blazars were found to be highly variable at all wavelengths, but most AGN appeared to be systematically less variable in the far‐infrared than at higher frequencies. This supported the idea of thermal emission from dust in the infrared. This was further supported by observations at submillimeter wavelengths that showed a very steep decline in flux longward of the infrared peak at around 100 μm. For example, an upper limit on the flux from NGC 4151 at 438 μm (Edelson et al. 1988) was so far below the measured flux at 155 μm as to require a slope steeper than ν^+2.5, the steepest that can be obtained from a self‐absorbed synchrotron source without special geometries. Dust emission could explain a steeper slope because of the decreasing efficiency of emission toward longer wavelengths.

Sanders et al. (1989) presented measurements of 109 QSOs from 0.3 nm to 6 cm (10¹⁰–10¹⁸ Hz). The gross shape of the energy distributions was quite similar for most objects, excepting the flat spectrum radio‐loud objects such as 3C 273. This typical energy distribution could be fitted by a hot accretion disk at shorter wavelengths and heated dust at longer wavelengths. Warping of the disk at larger radii was invoked to give the needed amount of reprocessed radiation as a function of radius. As noted by Rees et al. (1969) and others, the rather steep slope in the infrared, giving rise to an apparent minimum in the flux around 1 μm, could be explained naturally by the fact that grains evaporate if heated to temperatures above about 1500 K. Sanders et al. saw "no convincing evidence for energetically significant nonthermal radiation" in the wavelength range 3 nm to 300 μm in the continua of radio‐quiet and steep‐spectrum radio‐loud quasars. This paper marked the culmination of a gradual shift of sentiment from nonthermal to thermal explanations for the continuum of nonblazar AGN.

The blazar family comprised "BL Lac objects" and "optically violent variable" (OVV) QSOs. BL Lac objects, named after the prototype object earlier listed in catalogs of variable stars, had a nonthermal continuum but little or no line emission. OVVs have the emission lines of QSOs. These objects all show a continuum that is fairly well described as a power law extending from X‐ray to infrared frequencies. They typically show rapid (sometimes day to day) variability and strong, variable polarization. The continuum in blazars is largely attributed to nonthermal processes (synchrotron emission and inverse Compton scattering). 3C 273 seems to be a borderline OVV (Impey, Malkan, & Tapia 1989). The need for relativistic motions, described above, arises in connection with this class of objects. A comprehensive study of the energy distributions of blazars from 10⁸ to 10¹⁸ Hz was given by Impey & Neugebauer (1988). Bolometric luminosities ranged from 10⁹ to 10¹⁴ L_⊙, dominated by the 1–100 μm band. There was evidence for a thermal infrared component in many of the less luminous objects and an ultraviolet continuum bump associated with the presence of emission lines. When gamma rays are observed from AGN (e.g., Swanenburg et al. 1978), they appear to be associated with the beamed nonthermal continuum. The relationship of blazars to "normal" AGN is a key question in the effort to unify the diverse appearance of AGN.

IRAS revealed a large population of galaxies whose luminosity was strongly dominated by the far‐infrared (Soifer, Houck, & Neugebauer 1987). (Rieke 1972 had found early indications of a class of ultraluminous infrared galaxies.) The infrared emission is thermal emission from dust, energized in many cases by star formation but in some cases by an AGN. One suggested scenario was that some event, possibly a galactic merger, injected large quantities of gas and dust into the nucleus. This fueled a luminous episode of accretion onto a black hole, at first enshrouded by the dusty gas, whose dissipation revealed the AGN at optical and and ultraviolet wavelengths (Sanders et al. 1988).

The intriguing paper by Lynden‐Bell (1969) still did not launch a widespread effort to understand AGN in terms of accretion disks around black holes. Further impetus came from the discovery of black holes of stellar mass in our Galaxy. Among the objects discovered by Uhuru and other early X‐ray experiments were sources involving binary star systems with a neutron star or black hole. "X‐ray pulsars" emitted regular pulses of X‐rays every few seconds as the neutron star turned on its axis. The X‐ray power was essentially thermal emission from gas transferred from the companion star, impacting on the neutron star with sufficient velocity to produce high temperatures. Another class of source, exemplified by Cyg X‐1, showed no periodic variations but a rapid flickering (Oda et al. 1971) indicating a very small size. Analysis of the orbit gave a mass too large to be a neutron star or white dwarf, and the implication was that the system contained a black hole (Webster & Murdin 1972; Tananbaum et al. 1972). The X‐ray emission was attributed to gas from the companion O star heated to very high temperatures as it spiraled into the black hole by way of a disk (Thorne & Price 1975).

Galactic X‐ray sources, along with cataclysmic variable stars, protostars, and AGN, stimulated efforts to develop the theory of accretion disks. In many cases, the disk was expected to be geometrically thin, and the structure in the vertical and radial directions could be analyzed separately. A key uncertainty was the mechanism by which angular momentum is transported outward as matter spirals inward. In a highly influential paper, Shakura & Sunyaev (1973) analyzed disks in terms of a dimensionless parameter α that characterized the stresses that led to angular momentum transport and local energy release. General relativistic corrections were added by Novikov & Thorne (1973). This "α‐model" remains the standard approach to disk theory, and only recently have detailed mechanisms for dissipation begun to gain favor (Balbus & Hawley 1991). The α‐model gave three radial zones characterized by the relative importance of radiation pressure, gas pressure, electron scattering, and absorption opacity. The power producing regions of AGN disks would fall in the "inner" zone dominated by radiation pressure and electron scattering. Electron scattering would dominate in the atmosphere as well as the interior and modify the local surface emission from an approximate blackbody spectrum. The "inner" disk zone suffers both thermal and viscous instabilities (Pringle 1976; Lightman & Eardley 1974), but the ultimate consequence of these was unclear. A model in which the ions and electrons had different, very high temperatures was proposed for Cyg X‐1 by Eardley, Lightman, & Shapiro (1975). This led to models of "ion supported tori" for AGN (Rees et al. 1982). The related idea of "advection‐dominated accretion disks" or "ADAFs" (Narayan & Yi 1994) recently has attracted attention.

A key question was, do expected physical processes in disks explain the phenomena observed in AGN? In broad terms, this involved producing the observed continuum and, at least in some objects, generating relativistic jets, presumably along the rotation axis. Shields (1978b) proposed that the flat blue continuum of 3C 273 was thermal emission from the surface of an accretion disk around a black hole. For a mass ∼10⁹ M_⊙ and accretion rate 3 M_⊙ yr ⁻¹, the size and temperature of the inner disk was consistent with the observed blue continuum. This component dominated an assumed nonthermal power law, which would explain the infrared upturn and the X‐rays. Combining optical, infrared, and ultraviolet observations, Malkan (1983) successfully fitted the continua of a number of QSOs with accretion disk models. Czerny & Elvis (1987) suggested that the soft X‐ray excess of some AGN could be the high‐frequency tail of the thermal disk component or "big blue bump," which appeared to dominate the luminosity of some objects.

Problems confronted the simple picture of thermal emission from a disk radiating its locally produced energy. Correlated continuum variations at different wavelengths in the optical and ultraviolet were observed in the optical and ultraviolet on timescales shorter than the expected timescale for viscous or thermal processes to modify the surface temperature distribution in an AGN disk (e.g., Clavel, Wamsteker, & Glass 1989; Courvoisier & Clavel 1991). This suggested that reprocessing of X‐rays incident on the disk made a substantial contribution to the optical and ultraviolet continuum (Collin‐Souffrin 1991). Also troublesome was the low optical polarization observed in normal QSOs, typically 1% or less. The polarization generally is oriented parallel to the disk axis, when this can be inferred from jet structures (Stockman, Angel, & Miley 1979). Except for face on disks, electron scattering in disk atmospheres should produce strong polarization oriented perpendicular to the axis. Yet another problem was the prediction of strong Lyman edge absorption features, given effective temperatures similar to those of O stars (Kolykhalov & Sunyaev 1984). These issues remain under investigation today.

The question of fueling a black hole in a galactic nucleus has been difficult. Accretion rates of only a few solar masses a year suffice to power a luminous quasar, and even a billion solar masses is a small fraction of the mass of a QSO host galaxy. However, the specific angular momentum of gas orbiting a black hole at tens or hundreds of gravitational radii is tiny compared to that of gas moving with normal speeds even in the central regions of a galaxy. The angular momentum must be removed if the gas is to feed the black hole. Moreover, some galaxies with massive central black holes are not currently shining. Indeed, the rapid increase in the number of quasars with increasing look‐back time (Schmidt 1972), implies that there are many dormant black holes in galactic nuclei. What caused some to blaze forth as QSOs while others are inert? A fascinating possibility was the tidal disruption of stars orbiting close to the black hole (Hills 1975). However, the rate at which new stars would have their orbits evolve into disruptive ones appeared to be too slow to maintain a QSO luminosity (Frank & Rees 1976). The probability of an AGN in a galaxy appeared to be enhanced if it was interacting with a nearby galaxy (Adams 1977; Dahari 1984), which suggested that tidal forces could induce gas to sink into the galactic nucleus. There, unknown processes might relieve it of its angular momentum and allow it to sink closer and closer to the black hole.

The growing acceptance of the black hole model resulted, not from any one compelling piece of evidence, but rather from the accumulation of observational and theoretical arguments suggestive of black holes and from the lack of viable alternatives (Rees 1984).

After the discovery of QSOs, the widely different appearances of different AGN became appreciated. The question arose, what aspects of this diversity might result from the observer's location relative to the AGN? A basic division was between radio‐loud and radio‐quiet objects. Since the extended radio sources radiate fairly isotropically, their presence or absence could not be attributed to orientation. Furthermore, radio‐loud objects seemed to be associated with elliptical galaxies, and radio‐quiet AGN with spiral galaxies. The huge range of luminosities from Seyfert galaxies to QSOs clearly was largely intrinsic. However, some aspects could be a function of orientation. Blandford & Rees (1978) proposed that BL Lac objects were radio galaxies viewed down the axis of a relativistic jet. Relativistic beaming caused the nonthermal continuum to be very bright when so viewed, and the emission lines (emitted isotropically) would be weak in comparison. The same object, viewed from the side, would have normal emission‐line equivalent widths, and the radio structure would be dominated by the extended lobes rather than the core.

A key breakthrough occurred as a result of advances in the techniques of spectropolarimetry. Rowan‐Robinson (1977) had raised the possibility that the BLR of Seyfert 2 galaxies was obscured by dust, rather than being truly absent. Using a sensitive spectropolarimeter on the 120 inch Shane telescope at Lick Observatory, Antonucci & Miller (1985) found that the polarized flux of NGC 1068, the prototype Seyfert 2, had the appearance of a normal Seyfert 1 spectrum. This was interpreted in terms of a BLR and central continuum source obscured from direct view by an opaque, dusty torus. Electron scattering material above the nucleus near the axis of the torus scattered the nuclear light to the observer, polarizing it in the process. This allowed Seyfert 2 galaxies to have a detectable but unreddened continuum. However, the broad lines had escaped notice because the scattered light was feeble compared with the narrow lines from the NLR, which was outside the presumed obscuring torus. The same object, viewed face on, would be a Seyfert 1. Such a picture had also been proposed by Antonucci (1984) for the broad line radio galaxy 3C 234. Various forms of toroidal geometry had been anticipated by Osterbrock (1978) and others, and the idea received support from the discovery of "ionization cones" in the nuclei of some AGN (Pogge 1988). Orientation indicators were developed involving the ratio of the core and extended radio luminosities (Orr & Browne 1982; Wills & Browne 1986). The concepts of a beamed nonthermal continuum and an obscuring, equatorial torus remain fundamental to current efforts to unify AGN. Consideration of the obscuring torus supports the idea that the X‐ray background is produced mostly by AGN (Setti & Woltjer 1989).

The author is indebted to many colleagues for valuable communications and comments on the manuscript, including Stu Bowyer, Geoff and Margaret Burbidge, Marshall Cohen, Suzy Collin, Martin Elvis, Jesse Greenstein, Ken Kellermann, Matt Malkan, Bill Mathews, Richard Mushotzky, Gerry Neugebauer, Bev Oke, Martin Rees, George Rieke, Maarten Schmidt, Woody Sullivan, Marie‐Helene Ulrich, and Bev and Derek Wills. Don Osterbrock was especially supportive and helpful. This article was written in part during visits to the Department of Space Physics and Astronomy, Rice University, Lick Observatory, and the Institute for Theoretical Physics, University of California, Santa Barbara. The hospitality of these institutions is gratefully acknowledged. This work was supported in part by the Texas Advanced Research Program under grant 003658‐015 and by the National Science Foundation under grant PHY94‐07194.