8. IMPLICATIONS FOR ACCRETION PHYSICS

8.1. Why Are LLAGNs So Dim?

This review highlights a class of galactic nuclei that are extraordinary for being so ordinary. At their most extreme manifestation, LLAGNs emit a billion times less light than the most powerful known quasars. When quasars were first discovered, the challenge then was to explain their tremendous luminosities. Ironically, more than four decades later, the problem has been reversed: the challenge now is to explain how dead quasars can remain so dormant. The luminosity deficit problem was noted by Fabian & Canizares (1988), who drew attention to the fact that elliptical galaxies, despite being suffused with a ready supply of hot gas capable of undergoing spherical accretion, have very dim nuclei. We can no longer speculate that these systems lack supermassive BHs, for we now know that they are there, at least in galaxies with bulges. And as I have shown in this review, the problem is by no means confined to ellipticals either.

Explanations of the luminosity paradox fall in several categories.

• Obscuration     This trivial possibility can be summarily dismissed as a general solution in light of the evidence presented in Sections 5.3, 5.6.

• Low accretion rate     A more plausible strategy is to starve the BH. Present-day massive galaxies, after all, should be relatively gas-poor, especially in their central regions, which are largely devoid of significant ongoing star formation. This argument quickly falls apart when one realizes just how little material is needed to light up the nuclei. The bolometric luminosities of nearby nuclei span ~ 10³⁸ - 10⁴⁴ ergs s^-1, with a median value of L_bol = 3 × 10⁴⁰ ergs s^-1 and half of the sample lying between 3 × 10³⁹ and 3 × 10⁴¹ ergs s^-1. For a canonical radiative efficiency of = 0.1, the required accretion rate is merely = L_bol / c² = 5 × 10^{-6 ± 1} M yr^-1. This is a pitifully miniscule amount, in comparison with the amount of fuel actually available to be accreted. Galactic nuclei unavoidably receive fuel from two sources: ordinary mass loss from evolved stars and gravitational capture of gas from the hot interstellar medium.

  During the normal course of stellar evolution, evolved stars return a significant fraction of their mass to the ISM through mass loss. For a Salpeter stellar initial mass function with a lower-mass cutoff of 0.1 M, an upper-mass cutoff of 100 M, solar metallicities, and an age of 15 Gyr, Padovani & Matteucci (1993) estimate that

  This result is consistent, within a factor of ~2, with the work of Ciotti et al. (1991) and Jungwiert, Combes & Palous (2001). HST images reveal that galaxies contain central density concentrations, either in the form of nuclear cusps or photometrically distinct, compact stellar nuclei. The cusp profiles continue to rise to the resolution limit of HST (0.1"), which is r 10 pc at a distance of 20 Mpc, where 10-10³ L_{, V} pc^-3 for the "core" ellipticals and 10²-10⁴ L_{, V} pc^-3 for the "power-law" ellipticals and bulges of early-type spirals and S0s (e.g., Faber et al. 1997). Within a spherical region of r = 10 pc, the diffuse cores have L 4 × 10⁴ - 4 × 10⁶ L_{, V}, which yields * 1 × 10^-6 - 1 × 10^-4 M yr^-1; for the denser power-law cusps, L 4 × 10⁵ - 4 × 10⁷ L_{, V}, or _* 1 × 10^-5 - 1 × 10^-3 M yr^-1. Centrally dominant nuclear star clusters, present in a large fraction of disk galaxies, typically have luminosities L 10⁷ L (Carollo et al. 1997; Böker et al. 2002), and hence _* 10^-3 M yr^-1.

  Diffuse, hot gas in the central regions of galaxies holds another potential fuel reservoir. Low-angular momentum gas sufficiently close to a BH can accrete spherically in the manner described by Bondi (1952). To estimate the Bondi accretion rate, one needs to know the gas density and temperature at the accretion radius, R_a GM_BH / c_s², where c_s 0.1 T^1/2 km s^-1 is the sound speed of the gas at temperature T. The mass accretion rate follows from the continuity equation, _B = 4 R_a² _a c_s, where _a is the gas density at R_a. Expressed in terms of typical observed parameters (see below),

  Chandra observations with sufficient resolution to resolve R_a find that the diffuse gas in the central regions of elliptical galaxies typically has temperatures of kT 0.3-1 keV and densities of n 0.1-0.5 cm^-3 (e.g., Di Matteo et al. 2001, 2003; Loewenstein et al. 2001; Pellegrini 2005). Our knowledge of the hot gas content in the central regions of the bulges of spiral and S0 galaxies is more fragmentary. Chandra has so far resolved the hot gas in the centers of a handful of bulges (Milky Way: Baganoff et al. 2003; M31: Garcia et al. 2005; M81: Swartz et al. 2002; NGC 1291: Irwin, Sarazin & Bregman 2002; NGC 1553: Blanton, Sarazin & Irwin 2001; Sombrero: Pellegrini et al. 2003a). These studies suggest that bulges typically have gas temperatures of kT 0.3-0.6 keV. Information on gas densities is sketchier, but judging from the work on M81 and the Sombrero, a fiducial value might be n 0.1 cm^-3.

  If, for simplicity, we assume that the hot gas in most bulges is characterized by n = 0.1 cm^-3 and kT = 0.3 keV, then _B 10^-5-10^-3 M yr^-1 for M_BH = 10⁷-10⁸ M. In elliptical galaxies M_BH 10⁸ - 10⁹ M, and for n = 0.2 cm^-3 and kT = 0.5 keV, _B 10^-4 - 10^-2 M yr^-1. We note that these estimates of the Bondi accretion rates, which fall within the range given in recent compilations (e.g., Donato, Sambruna & Gliozzi 2004; Pellegrini 2005; Soria et al. 2006), are probably lower limits because the actual densities near R_a are likely to be higher than we assumed. For the above fiducial temperatures and BH masses, R_a 1-10 pc for bulges and ~ 10-100 pc for ellipticals, roughly an order of magnitude smaller than the typical linear resolution achieved by Chandra for nearby galaxies. In well-resolved cases, the gas temperature profile generally remains constant to within ~ 50% whereas the density typically increases by a factor of a few toward the center (e.g., Milky Way: Baganoff et al. 2003; M31: Garcia et al. 2005; M87: Di Matteo et al. 2003; NGC 1316: Kim & Fabbiano 2003).

  Although the above estimates are very rough, and they are valid only in a statistical sense, one cannot escape the conclusion that in general in the central few parsecs of nearby galaxies _B + _* >> . Although meager, the joint contributions from stellar mass loss and Bondi accretion, if converted to radiation with = 0.1, would generate nuclei far more luminous than actually observed. The net accretion from the Bondi flow would be considerably reduced if the gas possesses some angular momentum at large radii (Proga & Begelman 2003) - as inevitably it must - but even so it seems likely that the BH still has plenty of food at its disposal. LLAGNs are by no means fuel-starved. Moreover, the above estimates have erred on the conservative side. For example, I have assumed that all of the stars are evolved, although in reality most nuclei have composite populations and hence larger mass loss rates. Furthermore, I have neglected additional dissipation from larger scales (e.g., due to nuclear bars or spirals), as well as discrete, episodic events such as stellar tidal disruptions, which can provide a significant source of fuel, especially for BHs with masses 10⁷ M (Milosavljevic, Merritt & Ho 2006). All of these additional sources will only exacerbate the fuel surplus crisis.

• Low radiative efficiency     If accretion does proceed at a reasonable fraction of the supply rate, then one has no option but to conclude that the radiative efficiency is much less than = 0.1, the standard value for an optically thick, geometrically thin disk. This type of argument has been invoked to explain the apparent conflict between the nuclear luminosities and Bondi accretion rates in many early-type galaxies (e.g., Fabian & Rees 1995; Reynolds et al. 1996; Di Matteo & Fabian 1997; Mahadevan 1997; Di Matteo et al. 2001, 2003; Loewenstein et al. 2001; Ho, Terashima & Ulvestad 2003; Pellegrini et al. 2003a; Donato, Sambruna & Gliozzi 2004; Evans et al. 2006). Accretion flows with low radiative efficiency, of which the most popular version is the ADAF (see reviews in Narayan 2002; Yuan 2007), arise when the accreting medium is sufficiently tenuous that its cooling time exceeds the accretion timescale. RIAFs are predicted to exist for accretion rates below a critical threshold of _crit 0.3² _Edd 0.01 _Edd, where the Shakura & Sunyaev (1973) viscosity parameter is taken to be 0.1-0.3 and _Edd L_Edd / c² = 0.22 ( / 0.1) (M_BH / 10⁸ M) M yr^-1. LLAGNs lie comfortably below this threshold.

• Inefficient accretion/jet feedback     Precisely how low can become depends on how much of the native fuel supply actually gets accreted. In the presence of some rotation in the ambient medium, numerical simulations find that the amount of material accreted is much less than is available at large radii (e.g., Stone & Pringle 2001; Igumenshchev, Narayan & Abramowicz 2003). Since the gravitational binding energy in a RIAF cannot be radiated efficiently, it must be lost by nonradiative means (Blandford & Begelman 1999), either through convective transport of energy and angular momentum to large radii or through a global outflowing wind (see review by Quataert 2003). The net effect of either process is to flatten the density profile near the center and to dramatically reduce the accretion rate. At very low accretion rates (10^-5 - 10^-6) _Edd, such as in Sgr A^* and some giant elliptical galaxies, electron heat conduction will further suppress the accretion rate (Johnson & Quataert 2007). While these effects will ease the burden of invoking extremely small and perhaps physically unrealistic radiative efficiencies, it is important to note that these more recent models are still radiatively inefficient.

  Whether the outflows generated in RIAFs can lead directly to relativistic jets is unclear, but what observations have established is the tendency for lowly accreting systems to become increasingly jet-dominated. We see this not only in FR I radio galaxies (Chiaberge, Capetti & Celotti 1999; Donato, Sambruna & Gliozzi 2004; Kharb & Shastri 2004; Chiaberge, Capetti & Macchetto 2005; Balmaverde & Capetti 2006; Wu, Yuan & Cao 2007), but it is also reflected in the nuclear properties of more run-of-the-mill LLAGNs (Sections 5.3, 5.8), as well as in nearly quiescent nuclei (Pellegrini et al. 2007; Wrobel, Terashima & Ho 2008). Detailed analysis of some sources, in fact, indicates that most of the accretion power is not radiated but instead channeled into the kinetic energy of relativistic jets (M87: Di Matteo et al. 2003; IC 1459: Fabbiano et al. 2003; IC 4296: Pellegrini et al. 2003b). By analogy with the situation in cooling flows in galaxy clusters, the propensity for LLAGNs to become radio-loud opens up the possibility that the kinetic energy from small-scale jets or collimated outflows provides a major source of "feedback" into the circumnuclear environment, perhaps to the extent that accretion can be significantly interrupted or curtailed (Binney & Tabor 1995; Di Matteo et al. 2003; Pellegrini et al. 2003a; Omma et al. 2004). Indeed, calculations by Körding, Jester & Fender (2008; see also Heinz, Merloni & Schwab 2007) show that the total kinetic energy injected by LLAGN jets is very substantial. Given the prominent hard X-ray component in LLAGN spectra, inverse-Compton scattering of the hard photons might also provide another avenue to heat the ambient medium (Ciotti & Ostriker 2001). Either form of energy injection - mechanical or radiative - can lead to unsteady, intermittent accretion with a short duty cycle.

• Subluminous disk     A thin disk can be tolerated if it can be made extremely subluminous during periods of intermittent activity (Shields & Wheeler 1978). This situation would arise if accretion disks in AGNs undergo the thermal-viscous ionization instability (Lin & Shields 1986; Siemiginowska, Czerny & Kostyunin 1996), in which case they spend most of their time in quiescence, punctuated by brief episodes of intense outbursts. Menou & Quataert (2001) questioned the applicability of the ionization instability in AGNs, but they suggested that low-luminosity systems (with 10^-3 M yr^-1) may contain disks in which mass accumulates in a stable, nonaccreting "dead zone." Others have managed to stall accretion by condensing the hot flow into an inner cold, inert disk (Nayakshin 2003; Tan & Blackman 2005; Jolley & Kuncic 2007), which may form naturally from Compton cooling of the corona (Liu et al. 2007). Finally, Merloni & Fabian (2002) proposed that LLAGNs do contain a cold thin disk, but because of their low mass accretion rates, they liberate a large fraction of their gravitational energy in a strongly magnetized, unbound corona. Since a cold disk component is present in all these models, they face a serious, and in my opinion insurmountable, challenge because LLAGNs generally do not show fluorescent Fe K emission or reflection features in their X-ray spectra. The Merloni & Fabian model may be spared of this criticism, as the disk may be highly ionized, but it does predict strong and rapid X-ray variability, which is generally not observed in LLAGNs (Section 5.3; Ptak et al. 1998).

8.2. The Disk-Jet Connection

As the mass accretion rate drops and the radiative efficiency declines, an increasing fraction of the accretion power gets channeled into a relativistic jet whose energy release is mainly kinetic rather than radiative. The principal evidence for the growing importance of jets in LLAGNs comes from the broad-band SEDs, which invariably are prominent in the radio, with the degree of radio-loudness rising systematically (albeit with significant scatter) with decreasing Eddington ratio (Section 5.8; Figure 10b). Where available, VLBI imaging on milliarcsecond scales reveals unresolved cores with nonthermal brightness temperatures and a flat or slightly inverted spectrum - classical signposts of a relativistic jet (Blandford & Königl 1979). Detailed modeling of the SEDs of individual sources often shows that the accretion flow itself does not produce enough radio emission to match the data: that extra "something else" is most plausibly attributed to the jet component (Quataert et al. 1999; Ulvestad & Ho 2001b; Fabbiano et al. 2003; Pellegrini et al. 2003b; Anderson, Ulvestad & Ho 2004; Ptak et al. 2004; Wu & Cao 2005; Nemmen et al. 2006; Wu, Yuan & Cao 2007). Moreover, RIAF models predict radio spectral indices of +0.4 (Mahadevan 1997), whereas the observed values more typically fall in the range -0.2 to +0.2.

The jet may contribute substantially outside of the radio band, especially in the optical and X-rays. Some advocate that the jet, in fact, accounts for most or even all of the emission across the broad-band SED. For example, Yuan et al. (2002) successfully fitted the multiwavelength data of NGC 4258 with effectively a jet-only model. In their picture, a radiative shock at the base of the jet gives rise to synchrotron emission in the near-IR and optical regions, whose self-Compton component then explains the X-rays; the flat-spectrum radio emission comes from further out in the jet. Similar models have been devised for the Galactic Center source Sgr A^* (Falcke & Markoff 2000; Yuan, Markoff & Falcke 2002). The gross similarity between the SEDs of some FR I nuclei and BL Lac objects, which are jet-dominated sources but otherwise also low-accretion rate systems (Wang, Staubert & Ho 2002), has also been noted (e.g., Bower et al. 2000; Capetti et al. 2000; Chiaberge et al. 2003; Meisenheimer et al. 2007).

Statistical samples that are larger but more limited in spectral coverage have come from combining radio data with high-resolution optical or X-ray observations. Studies that specifically target radio galaxies, particularly FR I sources and weak-line FR IIs, report that the core radio power scales tightly with the optical and/or X-ray continuum luminosity, a finding often taken to support a common nonthermal, jet origin for the broad-band emission (Worrall & Birkinshaw 1994; Canosa et al. 1999; Chiaberge, Capetti & Celotti 1999, 2000; Capetti et al. 2002; Verdoes Kleijn et al. 2002; Donato, Sambruna & Gliozzi 2004; Balmaverde & Capetti 2006; Balmaverde, Capetti & Grandi 2006; Evans et al. 2006; for a counterargument, see Rinn, Sambruna & Gliozzi 2005 and Gliozzi et al. 2008). A similar radio-optical correlation, after correcting for Doppler boosting, is also seen among BL Lac objects (Giroletti et al. 2006), strengthening the case that FR I radio galaxies and BL Lac objects are intrinsically the same but misoriented siblings. Many FR II systems, on the other hand, especially those with broad lines, deviate systematically from the baseline FR I correlations, by exhibiting stronger optical and X-ray emission for a given level of radio emission (Chiaberge, Capetti & Celotti 2000, 2002; Varano et al. 2004). In concordance with the frequent detection of X-ray absorption and Fe K emission (Evans et al. 2006), this suggests that FR IIs have higher accretion rates and a much more dominant accretion flow component, relative to the jet, than FR Is.

Any attempt to explain the broad-band spectrum of LLAGNs with either just a RIAF or just a jet runs the risk of oversimplification. Clearly both are required. The trick is to figure out a reliable way to divvy up the two contributions to the SED. We cannot deny that there is a jet, because we see it directly in the radio at a strength far greater than can be attributed to the RIAF. The jet emission must contribute at some level outside of the radio band. At the same time, the jet cannot exist in isolation; it is anchored to and fed by some kind of accretion flow, of which a promising configuration is a vertically thick RIAF (Livio, Ogilvie & Pringle 1999; Meier 1999). An outstanding problem is that the interpretation of the data is not unique. Because many of the model parameters are poorly constrained and the broad-band data remain largely fragmentary and incomplete, the SEDs often can be fit with pure jet models, pure accretion flow models, or some combination of the two. The recent detection of high levels of polarization in the optical nuclei of FR Is (Capetti et al. 2007) strongly points toward a synchrotron origin in the jet for the optical continuum, but even this observation cannot be considered definitive, because a RIAF can also produce nonthermal flares (e.g., in Sgr A^*; Quataert 2003).

The so-called BH fundamental plane - a nonlinear correlation among radio luminosity, X-ray luminosity, and BH mass - offers a promising framework to unify accreting BHs over a wide range in mass and accretion rates. Merloni, Heinz & Di Matteo (2003) first demonstrated that the correlation between L_rad and L_X tightens considerably after including M_BH as a third variable. Combining observational material for several Galactic stellar BHs and a large sample of nearby LLAGNs, they find that

This empirical correlation agrees well with the theoretical relations between radio flux, BH mass, and accretion rate derived from the scale-invariant disk-jet model of Heinz & Sunyaev (2003). The BH fundamental plane, however, appears to be a very blunt tool. In an independent analysis, Falcke, Körding & Markoff (2004) obtained a similar empirical relation, but unlike Merloni, Heinz & Di Matteo these authors explained the scaling coefficients entirely in terms of a jet-dominated model. Moreover, as emphasized by Körding, Falcke & Corbel (2006), objects with very different emission processes, including luminous quasars and BL Lac objects, sit on the same correlation, albeit with larger scatter.

I illustrate this point in Figure 12a, which includes all Palomar LLAGNs with suitable data, along with the collection of high-luminosity sources from L.C. Ho (in preparation). With the exception of a handful of radio-loud quasars, the vast majority of the objects fall on a well-defined swath spanning ~ 10 orders of magnitude in luminosity. There are no obvious differences among the various subclasses of LLAGNs, except that the type 1 sources appear more tightly correlated. Plotting the residuals of the fundamental plane relation versus the Eddington ratio reveals two interesting points (Figure 12b). First, although the intrinsic scatter of the relation is quite large, it markedly increases for objects with high Eddington ratios, at L_bol / L_Edd 10^{-1 ± 1}, as already noted by Maccarone, Gallo & Fender (2003) and Merloni, Heinz & Di Matteo (2003). The scatter flares up because the radio-loud quasars lie offset above the relation and the radio-quiet quasars on average lie offset below the relation. At the opposite extreme, sources with L_bol / L_Edd 10^-6.5 may also show a systematic downturn, in possible agreement with the proposal by Yuan & Cui (2005) that below a critical threshold, L_X 10^-5.5 L_Edd, both the radio and the X-rays should be dominated by emission from the jet. M31 (Garcia et al. 2005), NGC 821 (Pellegrini et al. 2007), and NGC 4621 and NGC 4697 (Wrobel, Terashima & Ho 2008) seem to conform to Yuan & Cui's prediction, but M32 and especially Sgr A^* clearly do not. Additional deep radio and X-ray observations of ultra-low-luminosity nuclei would be very valuable to clarify the situation in this regime.

Figure 12. (a) Fundamental plane correlation among core radio luminosity, X-ray luminosity, and BH mass. (b) Deviations from the fundamental plane as a function of Eddington ratio.

If, as surmised, the relative proportions between jet and disk output depend on accretion rate, with the bulk of the radiated power, even in the X-rays, originating from the former in the lowest accretion rate systems, two important consequences ensue. With respect to the microphysics of RIAFs, it implies that the radiative efficiencies are even lower than previously inferred under the assumption that the X-rays emanate solely from the accretion flow. On a more global, environmental scale, shifting the emphasis from the disk to the jet changes the balance between kinetic versus radiative output, with important implications for prescriptions of AGN feedback in models of galaxy formation because BHs spend most of their lives in a low-state. From empirical and theoretical considerations (Heinz, Merloni & Schwab 2007; Körding, Jester & Fender 2008), the jet carries a substantial fraction of the accreted rest mass energy: P_jet 0.2 c² 7.2 × 10³⁶ (L_rad / 10³⁰ ergs s^-1)^12/17 ergs s^-1. In fact, the total kinetic energy injected by LLAGN jets is comparable to or perhaps even greater than the contribution from supernovae. At low redshifts, radiative feedback from quasars, which is commonly assumed to operate with an efficiency of ~ 5%, may be less important then jet-driven feedback from LLAGNs (Körding, Jester & Fender 2008).

Figure 13. A cartoon of the central engine of LLAGNs, consisting of three components: an inner, radiatively inefficient accretion flow (RIAF), an outer, truncated thin disk, and a jet or outflow. (Courtesy of S. Ho.)

8.3. The Central Engine of LLAGNs

The preceding sections argue that the weak nuclear activity seen in the majority of nearby galaxies traces low-level BH accretion akin to the more familiar form observed in powerful AGNs. However, multiple lines of evidence indicate that LLAGNs are not simply scaled-down versions of their more luminous cousins. They are qualitatively different. From the somewhat fragmentary clues presented in this review, we can piece together a schematic view of the structure of the central engine in LLAGNs (Ho 2002b, 2003, 2005). As sketched in Figure 13, it has three components.

1. Radiatively inefficient accretion flow     In the present-day Universe, and especially in the centers of big bulges, the amount of material available for accretion is small, resulting in mass accretion rates that fall far below 10^-2 _Edd. In such a regime, the low-density, tenuous material is optically thin and cannot cool efficiently. Rather than settling into a classical optically thick, geometrically thin, radiatively efficient disk - the normal configuration for luminous AGNs - the accretion flow puffs up into a hot, quasi-spherical, radiatively inefficient distribution, whose dynamics may be dominated by advection, convection, or outflows. This is an area of active ongoing theoretical research. In the interest of brevity, I will gloss over the technical details and simply follow Quataert (2003) by calling these RIAFs. The existence of RIAFs, or conversely the absence of a standard thin disk extending all the way to small radii (a few Schwarzschild radii R_s), is suggested by the feeble luminosities of LLAGNs, by their low Eddington ratios, and especially by their low inferred radiative efficiencies. The great disparity between the available fuel supply and the actually observed accretion luminosity demands that the radiative efficiency of the accretion flow be much less than = 0.1 (Section 8.1). Additional support for RIAFs comes from considerations of the SED, particularly the absence of the big blue bump, a classical signature of the thin disk, and the preponderance of intrinsically hard X-ray spectra.

2. Truncated thin disk     Beyond a transition radius R_tr 100-1000 R_s, the RIAF switches to a truncated optically thick, geometrically thin disk. The observational evidence for this component comes in three forms. First, the SEDs of some well-studied LLAGNs require a truncated thin disk to explain the big red bump - the prominent mid-IR peak and the steep fall-off of the spectrum in the optical-UV region (Section 5.8). The thermal disk emission is cool (red) not only because of a low accretion rate (Lawrence 2005) but also because the inner radius of the disk does not extend all the way in to a few R_s as in luminous AGNs. Second, the very same truncated disk structure employed to model the SED simultaneously accounts for the relativistically broadened, double-peaked emission-line profiles observed in some sources (Section 5.5). Indeed, in the case of NGC 1097 (Nemmen et al. 2006), the transition radius derived from modeling the SED (R_tr = 225 R_s) agrees remarkably well with the inner radius of the disk obtained from fitting the double-peaked broad H profile. Ho et al. (2000) suggested that double-peaked broad emission lines are commonplace in LLAGNs. By implication, the truncated disk configuration inferred from this special class of line profiles must be commonplace too. Lastly, the striking absence of broad Fe K emission in the X-ray spectra of LLAGNs (Section 5.3), a feature commonly attributed to X-ray fluorescence off of a cold accretion disk extending inward to a few R_s in bright Seyfert 1 nuclei (e.g., Nandra et al. 1997b, 2007), strongly suggests that in low-luminosity sources such a structure is either absent or truncated interior to some radius, such that it subtends a significantly smaller solid angle. Similar lines of reasoning have been advanced for broad-line radio galaxies that show weak Fe K emission and weak Compton reflection (Wozniak et al. 1998; Eracleous, Sambruna & Mushotzky 2000; Lewis et al. 2005), although these characteristics can be mimicked by an ionized but otherwise untruncated disk (Ballantyne, Ross & Fabian 2002).

3. Jet/outflow     The empirical connection between LLAGNs and jets has been established unequivocally from radio observations. Not only are the SEDs of LLAGNs generically radio-loud, but the strength of the radio emission generally cannot be fit without recourse to a jet component, which in many cases can be seen directly from VLBI-scale radio images. From a theoretical point of view, jets may share a close physical connection with RIAFs. As emphasized by Narayan & Yi (1995) and Blandford & Begelman (1999), RIAFs have a strong tendency to drive bipolar outflows due to the high thermal energy content of the hot gas. Whether such outflows can develop into highly collimated, relativistic ejections remains to be seen, but they at least provide a promising starting point. RIAFs may be additionally conducive to jet formation because its vertically thick structure enhances the large-scale poloidal component of the magnetic field, which plays a critical role in launching jets (Livio, Ogilvie & Pringle 1999; Meier 1999; Ballantyne & Fabian 2005; Ballantyne 2007). It is interesting to recall that the original motivation for ion-supported tori (Rees et al. 1982), an early incarnation of RIAFs, was to explain the low luminosity of radio galaxies. Rees et al. postulated that the puffed-up structure of the ion torus may help facilitate the collimation of the jet.

The above-described three-component structure has been applied to model the broad-band spectrum of a number of LLAGNs, including NGC 4258 (Lasota et al. 1996; Gammie, Narayan & Blandford 1999), M81 and NGC 4579 (Quataert et al. 1999), NGC 3998 (Ptak et al. 2004), and NGC 1097 (Nemmen et al. 2006). For the handful of LLAGNs with available estimates of the transition radii, R_tr seems to scale roughly inversely with L_bol / L_Edd (Yuan & Narayan 2004). This trend may be in agreement with models for disk evaporation (Liu & Meyer-Hofmeister 2001). As the latter authors note, however, disks attain their maximum evaporation efficiency at R_tr 300 R_s, making sources such as M81 and NGC 4579, both with R_tr 100 R_s (Quataert et al. 1999), difficult to explain. At a qualitative level, at least, the general idea that the thin disk recedes to larger and larger radii as the accretion rate drops is probably correct. In an analysis of 33 PG quasars with Fe K emission detected in XMM-Newton spectra, Inoue, Terashima & Ho (2007) find that the iron line profile varies systematically with Eddington ratio. Specifically, the Fe K profile becomes narrower with decreasing L_bol / L_Edd, a result that can be interpreted as a systematic increase in the inner radius of the accretion disk at low accretion rates.

The basic schematic proposed in Figure 13 is hardly new. To my knowledge, a hybrid model consisting of a RIAF - then called an ion-supported torus - plus a truncated thin disk was most clearly articulated in a prescient paper by Chen & Halpern (1989) in their description of Arp 102B, later elaborated by Eracleous & Halpern (1994) in the general context of double-peaked broad-line radio galaxies. Chen & Halpern identified the 25 µm peak in the SED with the turnover frequency of the synchrotron peak from the RIAF, whose elevated structure illuminates an outer thin disk that emits the double-peaked broad optical lines. The overall weakness of the UV continuum in Arp 102B (Halpern et al. 1996) further corroborates a truncated thin disk structure and potentially provides an explanation for the low-ionization state of the emission-line spectrum. As for the jet component, it was assumed to be present, at least implicitly, insofar as the double-peaked broad-line AGNs were thought to reside preferentially in radio-loud AGNs, and the very concept of ion-supported tori was invented with reference to radio galaxies (Rees et al. 1982).

Recent developments add important refinements and modifications to Chen & Halpern's original picture. First, the mid-IR peak in most objects is dominated by thermal emission from the truncated thin disk rather than by the synchrotron peak of the RIAF. Second, the jet component, which was once regarded as somewhat incidental, has emerged as a natural and perhaps inevitable outgrowth of the inner accretion flow itself. Third, although the original model was invented to explain a small minority of the AGN population (double-peaked radio-loud sources), now we have good reason to believe that similar physical conditions prevail in LINERs as a class (Ho et al. 2000), and, by extension, in the majority of nearby accreting BHs.

The physical picture outlined above for LLAGNs shares strong similarities with that developed for X-ray binaries in their hard state (see Maccarone, Fender & Ho 2005 and references therein), suggesting that the basic architecture of the central engine around accreting BHs - across 10 orders of magnitude in mass - is essentially scale-invariant (Meier 2001; Maccarone, Gallo & Fender 2003; Merloni, Heinz & Di Matteo 2003; Falcke, Körding & Markoff 2004; Ho 2005; Körding, Jester & Fender 2006).