5. NUCLEAR PROPERTIES

5.1. Ionizing Continuum Radiation

AGNs, at least when unobscured, reveal themselves as pointlike nuclear sources with power-law spectra at optical and UV wavelengths, typically described by a continuum flux density f , with -0.5 (e.g., Vanden Berk et al. 2001). In unbeamed sources, this featureless continuum traces the low-frequency tail of the "big blue bump" (Shields 1978; Malkan & Sargent 1982), which supplies the bulk of the ionizing photons. This feature is extremely difficult to detect in LLAGNs, both because the big blue bump is weak or absent (Section 5.8) and because the sources are exceedingly faint. The optical nuclei of LINERs can have M_B -10 mag (Ho 2004b), at least 10⁴ times fainter than their (usually bulge-dominated) hosts (M_B M^* -20 mag). To overcome this contrast problem, searches for nuclear point sources in the optical and near-IR have relied on HST images (e.g., Chiaberge, Capetti & Celotti 1999; Quillen et al. 2001; Verdoes Kleijn et al. 2002; Chiaberge, Capetti & Macchetto 2005; Balmaverde & Capetti 2006; González-Martín et al. 2006). But resolution alone is not enough. Given the extreme faintness of the nucleus, the intrinsic cuspiness of the underlying bulge profile, complexities of the point-spread function, and the often irregular background marred by circumnuclear dust features, one must pay very close attention to how the measurements are made. Simple aperture photometry or searching for central excess emission can yield very misleading results. The most robust technique to extract faint nuclei in the presence of these complications employs two-dimensional, multi-component fitting (Ho & Peng 2001; Ravindranath et al. 2001; Peng et al. 2002). Using this method, nuclear sources with optical magnitudes as faint as ~ 20 have been measured, with limits down to ~ 22-23 mag possible for nearby galaxies. Due to the computational requirements of two-dimensional fitting, however, not many LLAGNs have yet been analyzed in this manner, and fewer still have enough photometric points to define even a crude spectral slope.

In a few cases, the optical featureless continuum has been detected spectroscopically. From the ground, this was only possible for a couple of the brightest sources. The stellar features of NGC 7213 (Halpern & Filippenko 1984) and Pictor A (Carswell et al. 1984; Filippenko 1985) show dilution by a featureless continuum, which can be described approximately by a power law with a spectral index of -1.5. The nuclear continuum is much more readily seen in small-aperture spectra that help to reject the bulge starlight. HST spectra have isolated the optical continuum in several LINERs (Ho, Filippenko & Sargent 1996; Nicholson et al. 1998; Ho et al. 2000; Shields et al. 2000; Barth et al. 2001a; Sabra et al. 2003), although in most objects it remains too faint to be detected spectroscopically (Sarzi et al. 2005). In all well-studied cases, the optical continuum is quite steep, with -1 to -2. This range in spectral slopes is consistent with the broad-band optical (Verdoes Kleijn et al. 2002) and optical-UV colors (Chiaberge et al. 2002) of the cores frequently detected in the LINER nuclei of FR I radio galaxies.

The predominantly old population of present-day bulges ensures that the stellar contamination largely disappears in the UV, especially at high resolution. A number of attempts have been made to detect UV emission in LINERs using IUE, but most of these efforts yielded ambiguous results (see review in Filippenko 1996), and real progress had to await the HST. Two dedicated HST UV (~ 2300 Å) imaging studies have been completed. Using the pre-COSTAR FOC, Maoz et al. (1996) surveyed a complete sample of 110 large, nearby galaxies, and among the subset with spectral classifications from Palomar, Maoz et al. (1995) discovered that ~25% of the LINERs show an unresolved UV core. Barth et al. (1998) found similar statistics in a more targeted WFPC2 study. They also made the suggestion, later confirmed by Pogge et al. (2000), that dust obscuration is probably the main culprit for the nondetection of UV emission in the majority of LINERs. The implication is that UV emission is significantly more common in LINERs than indicated by the detection rates. In some type 2 objects (e.g., NGC 4569 and NGC 6500), the UV emission is spatially extended and presumably not related to the nuclear source. Second-epoch UV observations with the ACS/HRC revealed that nearly all of the UV-bright sources exhibit long-term variability (Maoz et al. 2005), an important result that helps assuage fears that the UV emission might arise mainly from young stars (Maoz et al. 1998). Importantly, both type 1 and type 2 LINERs vary. UV variability has also been discovered serendipitously in a few other sources (Renzini et al. 1995; O'Connell et al. 2005).

5.2. Radio Cores

AGNs, no matter how weak, are almost never silent in the radio. Barring chance superposition with a supernova remnant, the presence of a compact radio core is therefore a good AGN indicator. Because of the expected faintness of the nuclei, however, any search for core emission must be conducted at high sensitivity, and arcsecond-scale angular resolution or better is generally needed to isolate the nucleus from the surrounding host, which emits copious diffuse synchrotron radiation. In practice, this requires an interferometer such as the VLA.

The prevalence of weak AGNs in nearby early-type galaxies has been established from the VLA radio continuum studies of Sadler, Jenkins & Kotanyi (1989) and Wrobel & Heeschen (1991), whose 5 GHz surveys with 5" report a high incidence (~ 30 - 40%) of radio cores in complete, optical flux-limited samples of elliptical and S0 galaxies. Interestingly, the radio detection rate is similar to the detection rate of optical emission lines (Figure 4), and the optical counterparts of the radio cores are mostly classified as LINERs (Phillips et al. 1986; Ho 1999a). Conversely, Heckman (1980b) showed that LINERs tend to be associated with compact radio sources. The radio powers are quite modest, generally in the range of 10¹⁹ - 10²¹ W Hz^-1 at 5 GHz. When available, the spectral indices tend to be relatively flat (e.g., Wrobel 1991; Slee et al. 1994). With the exception of a handful of well-known radio galaxies with extended jets (Wrobel 1991), most of the radio emission is centrally concentrated.

No comparable radio survey has been done for spiral galaxies. Over the last few years, however, a number of studies, mostly using the VLA, have systematically targeted sizable subsets of the Palomar galaxies, to the point that by now effectively the entire Palomar AGN sample has been surveyed at arcsecond ( 0.15" - 2.5") resolution (Filho, Barthel & Ho 2000, 2002a, 2006; Nagar et al. 2000, 2002; Ho & Ulvestad 2001; Filho et al. 2004; Nagar, Falcke & Wilson 2005; Krips et al. 2007). Because the sensitivity, resolution, and observing frequency varied from study to study, each concentrating on different subclasses of objects, it is nontrivial to combine the literature results. The only survey that samples a significant fraction of the three LLAGN classes at a uniform sensitivity and resolution is that by Nagar et al. (2000, 2002; Nagar, Falcke & Wilson 2005), which was conducted at 15 GHz and = 0.15". The main drawback is that the sensitivity of this survey (rms 0.2 mJy) is rather modest, and mJy-level sources can be missed if they possess relatively steep spectra. Despite these limitations, Nagar et al. detected a compact core, to a high level of completeness, in 44% of the LINERs. Importantly, to the same level of completeness, the Seyferts exhibit a very similar detection rate (47%). LINER 2s have a lower detection rate than LINER 1s (38% versus 63%; see Table 1), but the same pattern is reflected almost exactly within the Seyfert population (detection rate 30% for type 2s versus 72% for type 1s). Transition objects, on the other hand, clearly differ, showing a markedly lower detection rate of only 16%, consistent with the 8.4 GHz survey of Filho, Barthel & Ho (2000, 2002a, 2006), where the detection rate is ~ 25%. The statistical differences in the Hubble type distributions of the three AGN classes (Ho, Filippenko & Sargent 2003) slightly complicate the interpretation of these results. To the extent that radio power shows a mild dependence on bulge strength or BH mass (Nagar, Falcke & Wilson 2005; see Ho 2002a), the detection rates, strictly speaking, should be renormalized to account for the differences in morphological types among the three classes. This effect, however, will not qualitatively change the central conclusion: if a compact radio core guarantees AGN pedigree, then LINERs, of either type 1 or type 2, are just as AGN-like as Seyferts, whereas a significant fraction of transition objects (roughly half) may be unrelated to AGNs.

Table 1. Median statistical properties of LLAGNs

Spectral	LH	LX	Prad	Lbol/LEdd	LX/LH	log Ro	log RX	fb	fr	fx
Class	(ergs s-1)	(ergs s-1)	(W Hz-1)					(%)	(%)	(%)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
S1	1.9×1040	1.4×1041	8.5×1020	1.1×10-3	7.3	0.7	-3.8	52	72	100
S2	1.7×1039	1.3×1039	8.5×1019	5.9×10-6	0.75	0.4	-3.3	…	30	86
L1	3.7×1039	8.8×1039	2.6×1020	1.0×10-5	4.6	1.3	-2.9	23	63	95
L2	0.5×1039	1.2×1039	4.7×1019	4.8×10-6	1.6	0.8	-2.9	…	38	74
T	0.7×1039	0.5×1039	2.4×1019	1.2×10-6	0.41	0.4	-2.8	3	16	74
NOTE.- Column (1) Spectral class. (2) Extinction-corrected luminosity of the H line; for type 1 sources, it represents both the narrow and broad components. (3) X-ray luminosity in the 2-10 keV band. (4) Radio power at 5 GHz. (5) Bolometric Eddington ratio, with Lbol estimated from LX and LEdd derived from BH masses estimated using the MBH - relation of Tremaine et al. 2002. (6) Ratio of X-ray to H luminosity. (7) Optical radio-loudness parameter. (8) X-ray radio-loudness parameter. (9) Detection fraction of broad H emission. (10) Detection fraction of radio cores at 15 GHz. (11) Detection fraction of X-ray cores.

The detection rates from the Nagar et al. survey can be viewed as firm lower limits. At = 1" and rms = 0.04 mJy at 1.4 and 5 GHz, for example, the detection rate for the Palomar Seyferts rises to 75% (Ho & Ulvestad 2001). Although no lower frequency survey of LINERs has been completed so far (apart from the lower resolution studies of Sadler, Jenkins, & Kotanyi 1989 and Wrobel & Heeschen 1991 confined to early-type galaxies), the preliminary study by Van Dyk & Ho (1998) of 29 LINERs at 5 and 3.6 GHz ( = 0.5"; rms = 0.05-0.1 mJy) yielded a detection rate of over 80%, again suggesting that LINERs and Seyferts have a comparably high incidence of radio cores.

Importantly, a sizable, flux-limit subset of the 15 GHz detections has been reobserved with the Very Long Baseline Array at 5 GHz, and essentially all of them have been detected at milliarcsecond resolution (Nagar, Falcke & Wilson 2005). The high brightness temperatures ( 10⁶ - 10¹¹ K) leaves no doubt that the radio cores are nonthermal and genuinely associated with AGN activity.

Where multifrequency data exist, their spectra tend to be flat or even mildly inverted ( -0.2 to +0.2; Ho et al. 1999b; Falcke et al. 2000; Nagar et al. 2000; Nagar, Wilson & Falcke 2001; Ulvestad & Ho 2001b; Anderson, Ulvestad & Ho 2004; Doi et al. 2005; Krips et al. 2007), seemingly more optically thick than Seyferts (median = -0.4; Ulvestad & Ho 2001a), and variability on timescales of months is common (Nagar et al. 2002; Anderson & Ulvestad 2005). Both of these characteristics suggest that the radio emission in LINERs is mainly confined to a compact core or base of a jet. Seyfert galaxies contain radio cores as well, but they are often accompanied by linear, jetlike features resolved on arcsecond scales (e.g., Ulvestad & Wilson 1989; Kukula et al. 1995; Ho & Ulvestad 2001; Gallimore et al. 2006). This extended component appears to be less prevalent in LINERs, although a definitive comparison must await a survey matched in resolution, sensitivity, and wavelength with that performed for the Seyferts (Ho & Ulvestad 2001). Higher resolution images on milliarcsecond scales do resolve elongated structures akin to subparsec-scale jets, but most of the power is concentrated in a compact, high-brightness temperature core (Bietenholz, Bartel & Rupen 2000; Falcke et al. 2000; Ulvestad & Ho 2001b; Filho, Barthel & Ho 2002b; Anderson, Ulvestad & Ho 2004; Filho et al. 2004; Krips et al. 2007). The comprehensive summary presented in Nagar, Falcke & Wilson (2005) indicates that the incidence of milliarcsecond-scale radio cores is similar for LINERs and Seyferts, but that subparsec-scale jets occur more frequently in LINERs.

5.3. X-ray Cores

X-ray observations provide another very effective tool to isolate LLAGNs and to diagnose their physical properties. Ultra-faint LLAGNs can be identified where none was previously known in the optical (e.g., Loewenstein et al. 2001; Ho, Terashima & Ulvestad 2003; Fabbiano et al. 2004; Pellegrini et al. 2007; Wrobel, Terashima & Ho 2008). Here, too, sensitivity and resolution are critical, as the central regions of galaxies contain a plethora of discrete nonnuclear sources, often suffused with a diffuse thermal plasma. Chandra, whose ACIS camera delivers ~ 0.5" images, is the instrument of choice, although in some instances even data at ~ 5" resolution (e.g., ROSAT HRI) can still provide meaningful constraints, especially if accompanied by spectral information (e.g., XMM-Newton).

As in the radio, no truly unbiased high-resolution X-ray survey has yet been performed of an optical flux-limited sample of nearby galaxies. The closest attempt was made by Roberts & Warwick (2000), who searched for X-ray nuclear sources in 83 Palomar galaxies (~ 20% of the total sample) having archival ROSAT HRI data. This subset is probably not unbiased, but it does encapsulate all the nuclear spectral classes in the Palomar survey. In total, X-ray cores were detected in 54% of the sample, with Seyferts and LINERs (including transition objects) both showing a higher detection rate (~ 70%) than absorption (30%) or H II nuclei (40%). The high detection rate among the optically classified LLAGNs agrees well with other ROSAT studies of Palomar sources (Koratkar et al. 1995; Komossa, Böhringer & Huchra 1999; Halderson et al. 2001; Roberts, Schurch & Warwick 2001), but the nonnegligible detection rate among the inactive members suggests that a significant fraction of the "core" flux may be nonnuclear emission [X-ray binaries (XRBs) and diffuse gas] insufficiently resolved by ROSAT.

Observations with Chandra (e.g., Ho et al. 2001; Eracleous et al. 2002) confirm the suspicion that earlier X-ray studies may have suffered from confusion with extranuclear sources (Figure 5). Importantly, the sharp resolution and low background noise of ACIS enable faint point sources to be detected with brief (few ks) exposures. This makes feasible, for the first time, X-ray surveys of large samples of galaxies selected at non-X-ray wavelengths. In a snapshot survey of a distance-limited sample of Palomar LLAGNs, Ho et al. (2001) find that ~ 75% of LINERs, both type 1 and type 2, contain X-ray cores, some as faint as ~ 10³⁸ ergs s^-1 in the 2-10 keV band. Terashima & Wilson (2003b) report an even higher detection rate (100%) for a sample of LINERs chosen for having a flat-spectrum radio core. To date, roughly 50% of the entire Palomar sample, among them 40% of the AGNs, have been observed by Chandra. This rich archival resource has been the basis of a number of recent investigations focused on quantifying the AGN content of LINERs, chief among them Satyapal, Sambruna & Dudik (2004), Dudik et al. (2005), Pellegrini (2005), Satyapal et al. (2005), Flohic et al. (2006), and González-Martín et al. (2006). A common conclusion that can be distilled from these studies is that the incidence of X-ray cores among LINERs is quite high, ranging from ~ 50% to 70%, down to luminosity limits of ~ 10³⁸ ergs s^-1. The incidence of X-ray cores in LINERs is somewhat lower than, but still compares favorably to, that found in Palomar Seyferts (~ 90%), the vast majority of which now have suitable X-ray observations, as summarized in Cappi et al. (2006) and Panessa et al. (2006). While the impact of selection biases cannot be assessed easily, they are probably not very severe because most of the observations were not originally intended to study LINERs, nor were they targeting famous X-ray sources.

Figure 5. Chandra/ACIS images of three LLAGNs, illustrating the diversity and complexity of the X-ray morphologies of their circumnuclear regions. The cross marks the near-IR position of the nucleus. (Courtesy of H.M.L.G. Flohic and M. Eracleous.)

It is of interest to ask whether the incidence of X-ray cores in LINERs depends on the presence of broad H emission. The moderate-resolution ROSAT/HRI studies of Roberts & Warwick (2000) and Halderson et al. (2001) showed roughly comparable detection rates for type 1 and type 2 LINERs, suggesting that the two classes are intrinsically similar and that obscuration plays a minor role in differentiating them. On the other hand, detailed X-ray spectral analysis has raised the suspicion that LINER 2s may be a highly heterogeneous class, with the bulk of the X-ray emission possibly arising from stellar processes. An important caveat is that these studies were based on large-beam observations, mostly using ASCA (Terashima, Ho & Ptak 2000; Terashima et al. 2000a, 2002; Roberts, Schurch & Warwick 2001) and the rest using BeppoSAX (Georgantopoulos et al. 2002; Pellegrini et al. 2002). A clearer, more consistent picture emerges from the recent Chandra work cited above. Although the individual samples remain small, most Chandra surveys detect LINER 2s with roughly similar frequency as LINER 1s, ~ 50%-60%. To gain a more comprehensive census, I have assembled Chandra measurements for all Palomar LINERs from the literature, along with unpublished material for a significant number of additional objects in public archives, which were analyzed following Ho et al. (2001). Although clearly heterogeneous and incomplete, the final collection of 64 LINERs (20 type 1, 44 type 2) does constitute 70% of the entire Palomar sample. The detection rate among all LINERs is 86%, broken down into 95% for LINER 1s and 74% for LINER 2s. For completeness, note that a similar exercise for 36 transition objects (55% of the parent sample) yields a detection rate of 74%, identical to that of LINER 2s and only marginally lower than that of Seyfert 2s (86%; Table 1).

The X-ray spectral properties of LLAGNs, particularly LINERs, have most thoroughly been investigated using ASCA (Yaqoob et al. 1995; Ishisaki et al. 1996; Iyomoto et al. 1996, 1997, 1998a, 1998b; Ptak et al. 1996, 1999; Terashima et al. 1998a, 1998b, 2000a, 2000b, 2002; Ho et al. 1999a; Terashima, Ho & Ptak 2000; Roberts, Schurch & Warwick 2001), with important contributions from BeppoSAX (Pellegrini et al. 2000a, 2000b, 2002; Iyomoto et al. 2001; Georgantopoulos et al. 2002; Ptak et al. 2004). A seminal study on M81 was done using BBXRT (Petre et al. 1993). Although the nuclear component was not spatially isolated because of the poor angular resolution of these telescopes, they had sufficient effective area to yield good photon statistics over the energy range ~ 0.5-10 keV to spectrally isolate the hard, power-law AGN signal. The most salient properties are the following. (1) Over the region ~ 0.5-10 keV, the continuum can be fit with a power law with an energy index of -0.4 to -1.2. Although this range overlaps with that seen in more luminous sources, the typical value of ~ -0.8 in LLAGNs may be marginally flatter than in Seyfert 1s (<> = -0.87 ± 0.22; Nandra et al. 1997b) or radio-quiet quasars (<> = -0.93 ± 0.22; Reeves & Turner 2000), perhaps being more in line with radio-loud quasars (< > = -0.6 ± 0.16; Reeves & Turner 2000). (2) With a few notable exceptions (e.g., M51: Fukazawa et al. 2001, Terashima & Wilson 2003a; NGC 1052: Weaver et al. 1999, Guainazzi et al. 2000; NGC 4258: Makishima et al. 1994, Fiore et al. 2001; NGC 4261: Matsumoto et al. 2001), the power-law component shows very little intrinsic absorption. This trend conflicts with the tendency for the degree of obscuration to increase with decreasing luminosity (e.g., Lawrence & Elvis 1982). (3) Signatures of X-ray reprocessing by material from an optically thick accretion disk, in the form of Fe K emission or Compton reflection (Lightman & White 1988; George & Fabian 1991), are weak or absent; the weakness of the Fe K line in LLAGNs runs counter to the inverse correlation between iron line strength and luminosity observed in higher luminosity AGNs (Nandra et al. 1997b). (4) In the few cases where Fe K emission has been detected, it is always narrow. (5) Apart from the hard power law, most objects require an extra soft component at energies 2 keV that can be fit by a thermal plasma model with a temperature of kT 0.4-0.8 keV and near-solar abundances. (6) Contrary to the trend established for luminous sources (Nandra et al. 1997a), short-term, large-amplitude X-ray variability is rare in LLAGNs (Ptak et al. 1998).

More recent observations with Chandra and XMM-Newton have refined, but not qualitatively altered, the above results. Where detailed spectral analysis is possible (e.g., Böhringer et al. 2001; Kim & Fabbiano 2003; Pellegrini et al. 2003a; Terashima & Wilson 2003b; Filho et al. 2004; Page et al. 2004; Starling et al. 2005; Flohic et al. 2006; González-Martín et al. 2006; Soria et al. 2006), the hard power-law component (except in objects previously known to be heavily absorbed) continues to be relatively unabsorbed, even among many type 2 sources, and to show little signs of reflection. No convincing case of a relativistic Fe K line has yet surfaced in an LLAGN. The marginally broad iron lines discovered with ASCA in M81 (Ishisaki et al. 1996) and NGC 4579 (Terashima et al. 1998a) has now been resolved into multiple components (Dewangan et al. 2004; Page et al. 2004; Young et al. 2007), none of which can be associated with a canonical disk. At the same time, the equivalent width limits for even the narrow component have become impressively low (e.g., Ptak et al. 2004). Interestingly, a soft thermal component is still required in many objects (Section 5.4), but there is no evidence for blackbody-like soft excess emission commonly seen in Seyferts and quasars (e.g., Turner & Pounds 1989; Inoue, Terashima & Ho 2007).

5.4. Circumnuclear Thermal Plasma

Early X-ray observations of LLAGNs using ASCA have consistently revealed the presence of a diffuse, thermal component, typically with a temperature of kT 0.5 keV (Ptak et al. 1999; Terashima et al. 2002). The uniform analysis of ROSAT data by Halderson et al. (2001) concluded that ~ 80% of the Palomar sources contain an extended component. However, without better resolution, it was impossible to know the extent of confusion with point sources, how much of the gas is truly associated with the nuclear region of the galaxy, or the density and temperature profile of the gas.

Our view of the diffuse component in the nuclear region has been dramatically sharpened with Chandra and XMM-Newton. Not only has the near ubiquity of diffuse gas been confirmed in many nearby galaxies (Ho et al. 2001; Eracleous et al. 2002; Terashima & Wilson 2003b; Pellegrini 2005; Rinn, Sambruna & Gliozzi 2005; Cappi et al. 2006; González-Martín et al. 2006; Soria et al. 2006), including our own (Muno et al. 2004) and our close neighbor M31 (Garcia et al. 2005), but quantitative, statistical properties of the gas are now becoming available. In the comprehensive investigation of 19 LINERs by Flohic et al. (2006), the diffuse emission, detected in 70% of the sample, is concentrated within the central few hundred pc. With an average 0.5-2 keV luminosity of ~ 10³⁸ ergs s^-1, it accounts for more than half of the total central luminosity in most cases. The average spectrum is similar to that seen in normal galaxies: it can be described by a thermal plasma with kT = 0.5 keV plus a power-law component with = -0.3 to -0.5. I will return to the nature of the hard component in Section 6.5. What is the origin of the thermal plasma? Given what we know about the stellar populations (Section 4.2), a starburst origin, as suggested by González-Martín et al. (2006), seems improbable. In normal elliptical galaxies, the X-ray-emitting gas represents the repository of thermalized stellar ejecta generated from mass loss from evolved stars and Type Ia supernovae (e.g., Awaki et al. 1994). There is no reason not to adopt the same picture to explain the hot plasma in LINERs and other LLAGNs. High-resolution X-ray spectroscopy of the highly ionized gas around the nucleus of M81 (Page et al. 2003; Young et al. 2007) and NGC 7213 (Starling et al. 2005) reveals that the plasma is collisionally ionized. Starling et al. note that this may be a property unique to LINERs, as thermal gas in luminous Seyferts is usually photoionized rather than collisionally ionized (e.g., Kinkhabwala et al. 2002).

5.5. Broad-line Region

Luminous, unobscured AGNs distinguish themselves unambiguously by their characteristic broad permitted lines. The detection of broad H emission in ~ 25% of LINERs (Ho et al. 1997e) thus constitutes strong evidence in favor of the AGN interpretation of these sources. LINERs, like Seyferts, come in two flavors - some have a visible BLR (type 1), and others do not (type 2). The broad component becomes progressively more difficult to detect in ground-based spectra for permitted lines weaker than H. However, HST spectra of LINERs, when available, show broad higher-order Balmer lines as well as UV lines such as Ly, C IV 1549, Mg II 2800, and Fe II multiplets (Barth et al. 1996; Ho, Filippenko & Sargent 1996). A subset of LINERs contain broad lines with double-peaked profiles (Figure 6), analogous to those seen in a minority of radio galaxies (Eracleous & Halpern 1994), where they are often interpreted as a kinematic signature of a relativistic accretion disk (Chen & Halpern 1989). Most of the nearby cases have been discovered serendipitously, either as a result of the broad component being variable (NGC 1097: Storchi-Bergmann, Baldwin & Wilson 1993; M81: Bower et al. 1996; NGC 3065: Eracleous & Halpern 2001) or because of the increased sensitivity to weak, broad features afforded by small-aperture measurements made with HST (NGC 4450: Ho et al. 2000; NGC 4203: Shields et al. 2000; NGC 4579: Barth et al. 2001a). Double-peaked broad-line AGNs may be more common than previously thought, especially among LLAGNs, perhaps as a consequence of their accretion disk structure (Section 8).

Figure 6. LINERs with broad, double-peaked H emission discovered with HST. A model fit for the disk profile in NGC 4450 is shown for illustration (green curve). (Adapted from Ho et al. 2000, Shields et al. 2000, and Barth et al. 2001a.)

A pressing question, however, is: What fraction of the more numerous LINER 2s are AGNs? By analogy with the Seyfert 2 class, do LINER 2s contain a hidden LINER 1 nucleus? At first sight, it might seem that there is no a priori reason why the orientation-dependent unification model, which has enjoyed much success in the context of Seyfert galaxies, should not apply equally to LINERs. If we suppose that the ratio of LINER 2s to LINER 1s is similar to the ratio of Seyfert 2s to Seyfert 1s - 1.6:1 in the Palomar survey - we can reasonably surmise that the AGN fraction in LINERs may be as high as ~ 60%. That at least some LINERs do indeed contain a hidden BLR was demonstrated by the deep Keck spectropolarimetric observations of Barth, Filippenko & Moran (1999a, 1999b). In a survey of 14 LLAGNs, mostly LINERs, these authors detected broad H emission in three objects (~ 20%) polarized at a level of 1%-3%. Interestingly, all three objects are elliptical galaxies with double-sided radio jets. NGC 315 and NGC 1052 technically qualify as type 1.9 LINERs (Ho et al. 1997e), whereas NGC 4261 is a LINER 2. Although the sample is small, these observations prove two important points: (1) the weak broad H features detected in direct light is not always scattered emission (Antonucci 1993), since polarized emission was not detected in several other LINER 1.9s included in Barth, Filippenko & Moran's survey; (2) an obscured nucleus does lurk in some LINER 2s.

At the same time, other bright LINER 2s have resisted detection by spectropolarimetry. As in the case of Seyferts (Tran 2001), however, the nondetection of polarized broad lines does not necessarily imply that there is no hidden BLR. Nevertheless, the BLR in some type 2 AGNs, especially LINERs but also Seyferts, may be intrinsically absent, not obscured. In the case of some Seyferts, mostly weak sources, the evidence comes from low absorbing X-ray column densities (Bassani et al. 1999; Pappa et al. 2001; Panessa & Bassani 2002; Gliozzi et al. 2004; Cappi et al. 2006; Gliozzi, Sambruna & Foschini 2007; Bianchi et al. 2008; but see Ghosh et al. 2007) as well as optical variability (Hawkins 2004). LINERs, as a class, very much conform to this picture. As discussed further below, LINERs of either type generally show very little sign of absorbing or reprocessing material, and UV variability is common. A few exceptions exist (e.g., NGC 1052: Guainazzi et al. 2000; NGC 4261: Sambruna et al. 2003, Zezas et al. 2005), but, interestingly, these are precisely the very ones for which Barth, Filippenko, & Moran discovered hidden BLRs. NGC 4258, also highly absorbed in the X-rays (Fiore et al. 2001), shows polarized narrow lines rather than broad lines (Barth et al. 1999).

An excellent of a LINER with a naked type 2 nucleus is the Sombrero galaxy. Although clearly an AGN, it shows no trace of a broad-line component, neither in direct light (Ho et al. 1997e), not even when very well isolated with a small HST aperture (Nicholson et al. 1998), nor in polarized light (Barth, Filippenko & Moran 1999b). Its Balmer decrement indicates little reddening to the NLR. For all practical purposes, the continuum emission from the nucleus looks unobscured. It is detected as a variable UV source (Maoz et al. 1995, 2005) and in the soft and hard X-rays (Nicholson et al. 1998; Ho et al. 2001). The X-ray spectrum is only very mildly absorbed (Nicholson et al. 1998; Pellegrini et al. 2002, 2003a; Terashima et al. 2002), with no signs of Fe K emission expected from reprocessed material, consistent with the modest mid-IR emission reported by Bendo et al. (2006). In short, there is no sign of anything being hidden or much doing the hiding. So where is the BLR? It is just not there.

The lack of a BLR in very low-luminosity sources may be related to a physical upper limit in the broad-line width (Laor 2003). If LLAGNs obey the same BLR-luminosity relation as in higher luminosity systems, their BLR velocity depends on the BH mass and luminosity. At a limiting bolometric luminosity of L_bol 10^41.8 (M_BH / 10⁸ M)² ergs s^-1, v 25,000 km s^-1, above which clouds may not survive due to excessive shear or tidal forces. Alternatively, if BLR clouds arise from condensations in a radiation-driven, outflowing wind (Murray & Chiang 1997), a viewpoint now much espoused, then it is reasonable to expect that very low-luminosity sources would be incapable of generating a wind, and hence of sustaining a BLR. For example, the clumpy torus model of Elitzur & Shlosman (2006) predicts that the BLR can no longer be sustained for L_bol 10⁴² ergs s^-1. In the scenario of Nicastro (2000), the BLR originates from a disk outflow formed at the transition radius between regions dominated by gas and radiation pressure. As this radius shrinks with decreasing L_bol / L_Edd, where L_Edd = 1.3 × 10³⁸ (M_BH / M) ergs s^-1, the BLR is expected to disappear for L_bol / L_Edd 10^-3. The apparent correlation between BLR line width and L_bol / L_Edd qualitatively supports this picture (Xu & Cao 2007). Although the existing data are sparse, they indicate that LINERs generally lack UV resonance absorption features indicative of nuclear outflows (Shields et al. 2002). The models by Elitzur & Shlosman and Nicastro are probably correct in spirit but not in detail, because many of the Palomar LLAGNs plainly violate their proposed thresholds (Section 5.10).

Nonetheless, the statistics within the Palomar survey already provide tentative support to the thesis that the BLR vanishes at the lowest luminosities or Eddington ratios. Which of the two is the controlling variable is still difficult to say. For both Seyferts and LINERs, type 1 sources are almost a factor of 10 more luminous than type 2 sources in terms of their median total H luminosity (Table 1). (The statistical differences between type 1 and type 2 sources cannot be ascribed to sensitivity differences in the detectability of broad H emission. Type 1 objects do have stronger line emission compared to the type 2s, but on average their narrow H flux and equivalent width are only ~ 50% higher, and the two types overlap significantly. Moreover, as noted in Section 3.4, the broad H detection rates turn out to be quite robust even in light of the much higher sensitivity afforded by HST.) The differences persist after normalizing by the Eddington luminosities: adopting a bolometric correction of L_bol 16 L_X, L_bol / L_Edd = 1.1 × 10^-3 and 5.9 × 10^-6 for Seyfert 1s and Seyfert 2s, respectively, whereas the corresponding values for LINER 1s and LINER 2s are 1.0 × 10^-5 and 4.8 × 10^-6. Two caveats are in order. First, while most of the type 1 sources have X-ray data, only 60% of the LINER 2s and 70% of the Seyfert 2s do. Second, the X-ray luminosities, which pertain to the 2-10 keV band, have been corrected for intrinsic absorption whenever possible, but many sources are too faint for spectral analysis. The lower X-ray luminosities for the type 2 sources must be partly due to absorption, but considering the generally low absorbing columns, particularly among the LINERs (Georgantopoulous et al. 2002; Terashima et al. 2002), it is unclear if absorption alone can erase the statistical difference between the two types. The tendency for Seyfert 2s to have lower Eddington ratios than Seyfert 1s has previously been noted, for the Palomar sample (Panessa et al. 2006) and others (Middleton, Done & Schurch 2008).

Several authors have raised the suspicion that LINER 2s may not be accretion-powered. Large-aperture X-ray spectra of LINER 2s, like those of LINER 1s, can be fit with a soft thermal component plus a power law with -0.7 to -1.5 (Georgantopoulos et al. 2002; Terashima et al. 2002). But this alone does not provide enough leverage to distinguish AGNs from starburst galaxies, many of which look qualitatively similar over the limited energy range covered by these observations. We cannot turn to the iron K line or variability for guidance, because LLAGNs generally exhibit neither (Section 5.3). The hard X-ray emission in LINER 2s is partly extended (Terashima et al. 2000a>; Georgantopoulos et al. 2002), but the implications of this finding are unclear. Just because the X-ray emission surrounding the LLAGN is morphologically complex and there is evidence for circumnuclear star formation (e.g., NGC 4736; Pellegrini et al. 2002) does not necessarily imply that there is a causal connection between the starburst and the LLAGN. Roberts, Schurch & Warwick (2001) advocate a starburst connection from the observation that LINER 2s have a mean flux ratio in the soft and hard X-ray band (~ 0.7) similar to that found in NGC 253. This interpretation, however, conflicts with the stellar population constraints discussed in Section 4.2. It is also not unique. Luminous, AGN-dominated type 1 sources themselves exhibit a tight correlation between soft and hard X-ray luminosity, with a ratio not dissimilar from the quoted value (Miniutti et al. 2008).

An important clue comes from the fact that many LINER 2s have a lower L_X / L_H ratio than LINER 1s (Ho et al. 2001). In particular, the observed X-ray luminosity from the nucleus, when extrapolated to the UV, does not have enough ionizing photons to power the H emission (Terashima et al. 2000a). This implies that (1) the X-rays are heavily absorbed, (2) nonnuclear processes power much of the optical line emission, or (3) the ionizing SED is different than assumed. As discussed in Section 6.4, this energy budget discrepancy appears to be symptomatic of all LLAGNs in general, not just LINER 2s, and most likely results from a combination of the second and third effect. There are some indications that the SEDs of LINER 2s indeed differ systematically from those of LINER 1s (e.g., Maoz et al. 2005; Sturm et al. 2006). In light of the evidence given in Sections 5.3, < ahref="#5.6">5.6, I consider the first solution to be no longer tenable. One can point to objects such as NGC 4261 (Zezas et al. 2005) as examples of LINER 2s with strong obscuration, but such cases are rare.

From the point of view of BH demographics, the most pressing issue is what fraction of the LINER 2s should be included in the AGN tally. Some cases are beyond dispute (M84, M87, Sombrero). What about the rest? The strongest argument that the majority of LINER 2s are AGN-related comes from the detection frequency of radio (Section 5.2) and X-ray (Section 5.3) cores, which is roughly 60% of that of LINER 1s. On the other hand, the detection rate of Seyfert 2s are similarly lower compared to Seyfert 1s, most likely reflecting the overall reduction of nuclear emission across all bands in type 2 LLAGNs as a consequence of their lower accretion rates. In summary, the AGN fraction among LINER 2s is at least 60%, and possibly as high as 100%.

5.6. Torus

In line with the absence of a BLR discussed above and using very much the same set of evidence, a convincing case can be made that the torus also disappears at very low luminosities. In a large fraction of nearby LINERs, the low absorbing column densities and weak or undetected Fe K emission (Section 5.3) strongly indicate that we have a direct, unobstructed view of the nucleus. Ghosh et al. (2007) warn that absorbing columns can be underestimated in the presence of extended soft emission, especially when working with spectra of low signal-to-noise ratio. While this bias no doubt enters at some level, cases like the Sombrero (Section 5.3) cannot be so readily dismissed. By analogy with situation in luminous AGNs (e.g., Inoue, Terashima & Ho 2007; Nandra et al. 2007), type 1 LLAGNs, if they possess tori, should also show strong, narrow fluorescent Fe K emission. This expectation is not borne out by observations. NGC 3998, which has excellent X-ray data, offers perhaps the most dramatic example. Apart from showing no signs whatsoever for intrinsic photoelectric absorption, it also possesses one of the tightest upper limits to date on Fe K emission: EW < 25 eV (Ptak et al. 2004). Our sight line to the nucleus is as clean as a whistle. Satyapal, Sambruna & Dudik (2004) claim that many LINERs have obscured nuclei, but this conclusion is based on IR-bright, dusty objects chosen from Carrillo et al. (1999); as I have discussed in Section 3.2, I regard these objects not only as biased, but also confusing with respect to their nuclear properties.

Palomar Seyferts, whose luminosities and Eddington ratios are about an order of magnitude higher than those of LINERs (Section 5.10), show markedly larger absorbing column densities and stronger Fe K lines. In an XMM-Newton study of a distance-limited sample of 27 Palomar Seyferts, Cappi et al. (2006) detect strong Fe K emission in over half of objects. The distribution of absorbing columns is nearly continuous, from N_H 10²⁰ to 10²⁵ cm^-2, with 30%-50% of the type 2 sources being Compton-thick (Panessa et al. 2006). This seems consistent with the tendency for Seyferts to be more gas-rich than LINERs, to the extent that this is reflected in their higher NLR densities (Ho, Filippenko & Sargent 2003).

The trend of increasing absorption with increasing luminosity or Eddington ratio observed in Palomar LLAGNs has an interesting parallel among radio galaxies. A substantial body of recent work indicates that the nuclei of FR I sources, most of which are, in fact, LINERs, are largely unobscured (e.g., Chiaberge, Capetti & Celotti 1999; Donato, Sambruna & Gliozzi 2004; Balmaverde & Capetti 2006). In contrast, FR II systems, especially those with broad or high-excitation lines (analogs of Seyferts), show clear signs of absorption and Fe K emission (Evans et al. 2006).

Even if we are fooled by the X-ray observations, substantial absorption must result in strong thermal reemission of "waste heat" in the IR. While sources such as Cen A provide a clear reminder that every rule has its exception (Whysong & Antonucci 2004), the existing data do suggest that, as a class, FR I radio galaxies tend to be weak mid-IR or far-IR sources (Haas et al. 2004; Müller et al. 2004). The same holds for more nearby LINERs. Their SEDs do show a pronounced mid-IR peak (Section 5.8), but as I will argue later, it is due to emission from the accretion flow rather than from dust reemission.

5.7. Narrow-line Region Kinematics

The kinematics of the NLR are complex. At the smallest scales probed by HST, Verdoes Kleijn, van der Marel & Noel-Storr (2006) find that the velocity widths of the ionized gas in the LINER nuclei of early-type galaxies can be modeled as unresolved rotation of a thin disk in the gravitational potential of the central BH. The subset of objects with FR I radio morphologies, on the other hand, exhibit line broadening in excess of that expected from purely gravitational motions; these authors surmise that the super-virial motions may be related to an extra source of energy injection by the radio jet. Walsh et al. (2008) use multiple-slit STIS observations to map the kinematics of the inner ~ 100 pc of the NLR in a sample of 14 LLAGNs, mostly LINERs. Consistent with earlier findings (Ho et al. 2002; Atkinson et al. 2005), the velocity fields are generally quite disorganized, rarely showing clean signatures of dynamically cold disks undergoing circular rotation. Nevertheless, two interesting trends can be discerned. The emission line widths tend to be largest within the sphere of influence of the BH, progressively decreasing toward large radii to values that roughly match the stellar velocity dispersion of the bulge. The luminous members of the sample, on the other hand, show more chaotic kinematics, as evidenced by large velocity splittings and asymmetric line profiles, reminiscent of the pattern observed by Rice et al. (2006) in their sample of Seyfert galaxies. Walsh et al. suggest that above a certain luminosity threshold - one that perhaps coincides with the LINER/Seyfert division - AGN outflows and radio jets strongly perturb the kinematics of the NLR.

A large fraction (~ 90%) of the Palomar LLAGNs have robust measurements of integrated [N II] 6583 line widths, which enable a crude assessment of the dynamical state of the NLR and its relation to the bulge. Consistent with what has been established for more powerful systems (Nelson & Whittle 1996; Greene & Ho 2005a), the kinematics of the ionized gas are dominated by random motions that, to first order, trace the gravitational potential of the stars in the bulge. Among the objects with available central stellar velocity dispersions, _NLR / _* 0.7-0.8 for the weakest sources (L_H 10³⁸ ergs s^-1), systematically rising to _NLR / _* 1.2 in the more luminous members (L_H 10^41.5 ergs s^-1). L.C. Ho (in preparation) speculates that the central AGN injects a source of dynamical heating of nongravitational origin to the NLR, either in the form of radiation pressure from the central continuum or mechanical interaction from radio jets. Given the empirical correlation between optical line luminosity and radio power (e.g., Ho & Peng 2001; Ulvestad & Ho 2001a; Nagar, Falcke & Wilson 2005), and the near ubiquity of compact radio sources, it is a priori difficult to determine which of these two sources acts as the primary driver. The tendency for extended radio emission to be more prevalent in Seyferts (Section 5.2) suggests that jets may be more important.

5.8. Spectral Energy Distribution

The broad-band SED provides one of the most fundamental probes of the physical processes in AGNs. Both thermal and nonthermal emission contribute to the broad-band spectrum of luminous AGNs such as quasars and classical Seyfert galaxies. In objects whose intrinsic spectrum has not been modified severely by relativistic beaming or absorption, the SED can be separated into several distinctive components (e.g., Elvis et al. 1994): radio synchrotron emission from a jet, which may be strong ("radio-loud") or weak ("radio-quiet"); an IR excess, now generally considered to be predominantly thermal reradiation by dust grains; a prominent optical to UV "big blue bump," usually interpreted to be pseudo-blackbody emission from an optically thick, geometrically thin accretion disk (Shields 1978; Malkan & Sargent 1982); a soft X-ray excess, whose origin is still highly controversial (Done et al. 2007; Miniutti et al. 2008); and an underlying power law, which is most conspicuous at hard X-ray energies but is thought to extend down to IR wavelengths, that can be attributed to Comptonization of softer seed photons.

Within this backdrop, there were already early indications that the SEDs of LINERs may deviate from the canonical form. Halpern & Filippenko (1984) succeeded in detecting the featureless optical continuum in NGC 7213, and while these authors suggested that a big blue bump may be present in this object, they also noted that it possesses an exceptionally high X-ray-to-optical flux ratio, although perhaps one not inconsistent with the extrapolation of the trend of increasing X-ray-to-optical flux ratio with decreasing luminosity seen in luminous sources (Zamorani et al. 1981; Avni & Tananbaum 1982). A more explicit suggestion that LINERs may possess a weak UV continuum was made in the context of double-peaked broad-line AGNs such as Arp 102B and Pictor A, whose narrow-line spectra share many characteristics with LINERs (Chen & Halpern 1989; Halpern & Eracleous 1994). The HST spectrum of Arp 102B, in fact, shows an exceptionally steep optical-UV nonstellar continuum ( -2.1 to -2.4; Halpern et al. 1996). Halpern & Eracleous (1994) further suggested that the SEDs are flat in the far-IR. In an important study of M81, Petre et al. (1993) proposed that the relative weakness of the UV continuum compared to the X-rays is a consequence of a change in the structure of the central accretion flow, from a standard thin disk to an ion-supported torus (see Section 8.3). Parameterizing the two-point spectral index between 2500 Å and 2 keV by _ox [logL(2500 Å) - logL(2 keV)] / [log(2500 Å) - log(2 keV)], M81 and possibly other LINERs (Mushotzky 1993) have _ox -1, to be compared with _ox -1.4 for quasars and _ox -1.2 for Seyferts (Mushotzky & Wandel 1989).

The full scope of the spectral uniqueness of LLAGNs only became evident once the modern, albeit still fragmentary, multiwavelength data could be assembled. The initial studies concentrated on individual objects, emphasizing the weakness of the UV bump (M81: Ho, Filippenko & Sargent 1996; Sombrero: Nicholson et al. 1998) and the overall consistency of the SED with spectral models generated from advection-dominated accretion flows (ADAFs; see Narayan 2002 and Yuan 2007 for reviews) as unique attributes of systems with low Eddington ratios (NGC 4258: Lasota et al. 1996, Chary et al. 2000; M87: Reynolds et al. 1996; M60: Di Matteo & Fabian 1997). Ho (1999b) systematically investigated the SEDs of a small sample of seven LLAGNs with available BH mass estimates and reliable small-aperture fluxes from radio to X-ray wavelengths. This was followed by a study of another five similar objects, which have the additional distinction of having double-peaked broad emission lines (Ho et al. 2000; Ho 2002b). Figure 7 gives the latest update from a comprehensive analysis of the SEDs of 150 nearby type 1 AGNs spanning 4 dex in BH mass (M_BH 10⁵-10⁹ M) and 6.5 dex in Eddington ratio (L_bol / L_Edd 10^-6 - 10^0.5). Let us focus on two regimes: L_bol / L_Edd = 0.1 to 1, typical of classical, luminous AGNs, and L_bol / L_Edd < 10^-3.0, which characterizes most nearby LLAGNs (Section 5.10). I defer the discussion of the physical implications until Section 8, but for now list the most notable features concerning the LLAGN SED, some of which are also apparent in the composite LINER SED assembled by Eracleous, Hwang & Flohic (2008a). (1) The big blue bump is conspicuously absent. (2) Instead, a broad excess is shifted to the mid-IR, forming a "big red bump"; this component is probably related to the mid-IR excess previously noted by Lawrence et al. (1985), Willner et al. (1985), and Chen & Halpern (1989), and more recently from Spitzer observations (e.g., Willner et al. 2004; Bendo et al. 2006; Gu et al. 2007). (3) As a consequence of this shift, the optical-UV slope is exceptionally steep, generally in the range _ou -1 to -2.5, to be compared with _ou -0.5 to -0.7 for luminous AGNs (Vanden Berk et al. 2001; Shang et al. 2005); the X-ray-to-optical ratio is large, resulting in _ox -1. (4) There is no evidence for a soft X-ray excess. (5) Lastly, the overall SED can be considered radio-loud, defined here by the convention that the radio-to-optical luminosity ratio exceeds a value of R_o L(5 GHz) / L(B) = 10. Radio-loudness, in fact, seems to be a property common to essentially all nearby weakly active nuclei (Ho 1999b, 2002a; Ho et al. 2000) and a substantial fraction of Seyfert nuclei (Ho & Peng 2001). Defining radio-loudness based on the relative strength of the radio and X-ray emission, R_X L(5 GHz)/ L_X, Terashima & Wilson (2003b) also find that LINERs tend to be radio-loud, here taken to be R_X > 10^-4.5. Moreover, the degree of radio-loudness scales inversely with L_bol / L_Edd (Ho 2002a; Terashima & Wilson 2003b; Wang, Luo & Ho 2004; Greene, Ho & Ulvestad 2006; Panessa et al. 2007; Sikora, Stawarz & Lasota 2007; L.C. Ho, in preparation; see Figure 10b).

Figure 7. Composite SEDs for radio-quiet AGNs binned by Eddington ratio. The SEDs are normalized at 1 µm. (Adapted from L.C. Ho, in preparation.)

In a parallel development, studies of the low-luminosity, often LINER-like nuclei of FR I radio galaxies also support the notion that they lack a UV bump. M84 (Bower et al. 2000) and M87 (Sabra et al. 2003) are two familiar examples, but it has been well documented that FR I nuclei tend to exhibit flat _ox (Donato, Sambruna & Gliozzi 2004; Balmaverde, Capetti & Grandi 2006; Gliozzi et al. 2008) and steep slopes in the optical (Chiaberge, Capetti & Celotti 1999; Verdoes Kleijn et al. 2002) and optical-UV (Chiaberge et al. 2002).

Finally, I note that the UV spectral slope can be indirectly constrained from considering the strength of the He II 4686 line. While this line is clearly detected in Pictor A (Carswell et al. 1984; Filippenko 1985), its weakness in NGC 1052 prompted Péquignot (1984) to deduce that the ionizing spectrum must show a sharp cutoff above the He⁺ ionization limit (54.4 eV). In this respect, NGC 1052 is quite representative of LINERs in general. He II 4686 was not detected convincingly in a single case among a sample of 159 LINERs in the entire Palomar survey (Ho, Filippenko & Sargent 1997a). Starlight contamination surely contributes partly to this, but the line has also eluded detection in HST spectra (e.g., Ho, Filippenko & Sargent 1996; Nicholson et al. 1998; Barth et al. 2001b; Sabra et al. 2003; Sarzi et al. 2005; Shields et al. 2007), which indicates that it is truly intrinsically very weak. To a first approximation, the ratio of He II 4686 to H reflects the relative intensity of the ionizing continuum between 1 and 4 Ryd. For an ionizing spectrum f , case B recombination predicts He II 4686 / H = 1.99 × 4 (Penston & Fosbury 1978). The current observational limits of He II 4686 / H 0.1 thus imply -2, qualitatively consistent with the evidence from the SED studies.

Maoz (2007) has offered an alternative viewpoint to the one presented above. Using a sample of 13 LINERs with variable UV nuclei, he argues that their SEDs do not differ appreciably from those of more luminous AGNs, and hence that LINERs inherently have very similar accretion disks compared to powerful AGNs. Maoz does not disagree that LINERs have large X-ray-to-UV flux ratios or that they tend to be radio-loud; his data show both trends. Rather, he contends that because LINERs lie on the low-luminosity extrapolation of the well-known relation between _ox and luminosity (Zamorani et al. 1981; Avni & Tananbaum 1982; Strateva et al. 2005) they do not form a distinct population. And while LINERs do have large values of R_o, they nonetheless occupy the "radio-quiet" branch of the R_o versus L_bol / L_Edd plane (Sikora, Stawarz & Lasota 2007). In my estimation, the key point is not, and has never been, whether LINERs constitute a disjoint class of AGNs, but whether they fit into a physically plausible framework in which their distinctive SEDs, among other properties, find a natural, coherent explanation. Section 8 attempts to offer such a framework.

It should be noted that Maoz's results strongly depend on his decision to exclude all optical and near-IR data from the SEDs, on the grounds that they may be confused by starlight. I think this step is too draconian, as it throws away valuable information. While stellar contamination is certainly a concern, one can take necessary precautions to try to isolate the nuclear emission as much as possible, either through high-resolution imaging (e.g., Ho & Peng 2001; Ravindranath et al. 2001; Peng et al. 2002) or spectral decomposition. In well-studied sources, there is little doubt that the optical continuum is truly both featureless and nonstellar (e.g., Halpern & Filippenko 1984; Ho, Filippenko & Sargent 1996; Ho et al. 2000; Bower et al. 2000; Sabra et al. 2003). Given what we know about the nuclear stellar population, we cannot assign the featureless continuum to young stars. In a few cases, the nonstellar nature of the nucleus can even be established through variability in the optical (Bower et al. 2000; Sabra et al. 2003; O'Connell et al. 2005) and mid-IR (Rieke, Lebofsky & Kemp 1982; Grossan et al. 2001; Willner et al. 2004).

While the SEDs of LINERs differ from those of traditional AGNs, it is important to recognize that they are decidedly nonstellar and approximate the form predicted for radiatively inefficient accretion flows (RIAFs) onto BHs, often coupled to a jet (Quataert et al. 1999; Yuan, Markoff & Falcke 2002; Yuan et al. 2002; Fabbiano et al. 2003; Pellegrini et al. 2003b; Ptak et al. 2004; Nemmen et al. 2006; Wu, Yuan & Cao 2007). They bear little resemblance to SEDs characteristic of normal stellar systems. Inactive galaxies or starburst systems not strongly affected by dust extinction emit the bulk of their radiation in the optical-UV and in the thermal IR regions, with only an energetically miniscule contribution from X-rays.

5.9. Luminosity Function

Many astrophysical applications of AGN demographics benefit from knowing the AGN luminosity function, (L, z). Whereas (L, z) has been reasonably well charted at high L and high z using quasars, it is very poorly known at low L and low z. Indeed, until very recently there has been no reliable determination of (L, 0). The difficulty in determining (L, 0) can be ascribed to a number of factors, as discussed in Huchra & Burg (1992). First and foremost is the challenge of securing a reliable, spectroscopically selected sample. Since nearby AGNs are expected to be faint relative to their host galaxies, most of the traditional techniques used to identify quasars cannot be applied without introducing large biases. The faintness of nearby AGNs presents another obstacle, namely how to disentangle the nuclear emission - the only component relevant to the AGN - from the usually much brighter contribution from the host galaxy. Finally, most optical luminosity functions of bright, more distant AGNs are specified in terms of the nonstellar optical continuum (usually the B band), whereas spectroscopic surveys of nearby galaxies generally only reliably measure optical line emission (e.g., H) because the featureless nuclear continuum is often impossible to detect in ground-based, seeing-limited apertures.

A different strategy can be explored by taking advantage of the fact that H luminosities are now available for nearly all of the AGNs in the Palomar survey. Figure 8 shows the H luminosity function for the Palomar sources, computed using the V / V_max method (L.C. Ho, A.V. Filippenko & W.L.W. Sargent, in preparation). Two versions are shown, each representing an extreme view of what kind of sources should be regarded as bona fide AGNs. The open symbols include only type 1 nuclei, whose AGN status is incontrovertible. This may be regarded as the most conservative assumption and a lower bound, since we know that genuine narrow-line AGNs do exist. The filled symbols lump together all sources classified as LINERs, transition objects, and Seyferts, both type 1 and type 2. This represents the most optimistic view and an upper bound, if some type 2 sources are in fact AGN impostors, although, as I argue in Section 6.5, this is likely to be a small effect. The true space density of local AGNs lies between these two possibilities. In either case, the differential luminosity function can be approximated by a single power law from L_H 10³⁸ to 3 × 10⁴¹ ergs s^-1, roughly of the form L^{-1.2 ± 0.2}. The slope seems to flatten below L_H 10³⁸ ergs s^-1, but the luminosity function is highly uncertain at the faint end because of density fluctuations in our local volume. Nevertheless, it is remarkable that the Palomar luminosity function formally begins at L_H 6 × 10³⁶ ergs s^-1, roughly the luminosity of the Orion nebula (Kennicutt 1984). In units more familiar to the AGN community, this corresponds to an absolute B-band magnitude of roughly -8 (using the H-optical continuum conversion of Greene & Ho 2005b), no brighter than a single supergiant star.

Figure 8. The H nuclear luminosity function of nearby AGNs derived from the Palomar survey. The top axis gives an approximate conversion to absolute magnitudes in the B band, using the H-continuum correlation of Greene & Ho (2005b). The unfilled circles include only type 1 sources, while the filled circles represent both type 2 and type 1 sources. The luminosities have been corrected for extinction, and in the case of type 1 nuclei, they include both the narrow and broad components of the line. For comparison, I show the z < 0.35 luminosity function for SDSS Seyfert galaxies (types 1 and 2; dashed line; Hao et al. 2005b). (Adapted from L.C. Ho, A.V. Filippenko & W.L.W. Sargent, in preparation.)

For comparison, I have overlaid the H luminosity function of z 0.35 Seyfert galaxies derived from the SDSS by Hao et al. (2005b). The Palomar survey reaches ~ 2 orders of magnitude fainter in H luminosity than SDSS, but the latter extends a factor of 10 higher at the bright end. Over the region of overlap, the two surveys show reasonably good agreement, especially considering the small number statistics of the Palomar survey and the fact that Hao et al.'s sample only includes Seyferts.

5.10. Bolometric Luminosities and Eddington Ratios

To gain further insight into the physical nature of LLAGNs, it is more instructive to examine their bolometric luminosities rather than their luminosities in a specific band or emission line. Because AGNs emit a very broad spectrum, their bolometric luminosities ideally should be measured directly from their full SEDs. In practice, however, complete SEDs are not readily available for most AGNs, and one commonly estimates L_bol by applying bolometric corrections derived from a set of well-observed calibrators. As discussed in Section 5.8, the SEDs of LLAGNs differ quite markedly from those of conventionally studied AGNs. Nonetheless, they do exhibit a characteristic shape, which enables bolometric corrections to be calculated. The usual practice of choosing the optical B band as the reference point should be abandoned for LLAGNs, not only because reliable optical continuum measurements are scarce but also because the optical/UV region of the SED shows the maximal variance with respect to accretion rate (Section 5.8) and depends sensitively on extinction. What is available, by selection, is nuclear emission-line fluxes, and upper limits thereon. Although the H luminosity comprises only a small percentage of the total power, its fractional contribution to L_bol turns out to be fairly well defined: from the SED study of L.C. Ho (in preparation), L_bol 220 L_H, with an rms scatter of ~ 0.4 dex, consistent with the calibration given in Greene & Ho (2005b, 2007a). Because of the wide range of ionization levels among LLAGNs, a bolometric correction based on H should be more stable than one tied to [O III] 5007 (e.g., Heckman et al. 2005). Nevertheless, in light of the nonnuclear component of the nebular flux in LLAGNs (Section 6.4), the luminosity of the narrow H line will tend to overestimate L_bol. I recommend that, whenever possible, L_bol should be based on the hard X-ray (2-10 keV) luminosity, bearing in mind the added complication that the bolometric correction in this band is luminosity-dependent. Making use again of the database from L.C. Ho (in preparation), I estimate L_bol / L_X 83, 28, and 16 for quasars, luminous Seyferts, and LLAGNs, respectively.

Figure 9 shows the distributions of L_bol and their values normalized with respect to the Eddington luminosity for Palomar galaxies with measurements of L_X and central stellar velocity dispersion. The M_BH - relation of Tremaine et al. (2002) was used to obtain L_Edd. Although there is substantial overlap, the four spectral classes clearly delineate a luminosity sequence, with L_bol decreasing systematically as S L T A. The differences become even more pronounced in terms of L_bol / L_Edd, with Seyferts having a median value (1.3 × 10^-4) 20 times higher than in LINERs (5.9 × 10^-6), which in turn are higher than transition objects by a factor of ~ 5. Among Seyferts and LINERs, type 1 sources are systematically more luminous than type 2s. Notably, the vast majority of nearby nuclei have highly sub-Eddington luminosities. The total distribution of Eddington ratios is characterized by a prominent peak at L_bol / L_Edd 10^-5 dominated by Seyfert 2s, LINERs, and transition objects, and a precipitous drop toward larger Eddington ratios. Contrary to previous claims (Wu & Cao 2005; Hopkins & Hernquist 2006) based on the smaller sample of Ho (2002a), the distribution of Eddington ratios shows no bimodality. The systematic difference in Eddington ratios between LINERs and Seyferts has been noticed before in the Palomar survey (Ho 2002b, 2003, 2005) and in SDSS (Kewley et al. 2006), but this is the first time that the more subtle differences among the different subclasses can be discerned.

Figure 9. Distribution of (a) bolometric luminosity, Lbol, and (b) ratio of bolometric luminosity to the Eddington luminosity, Lbol / LEdd, for all objects, Seyferts (S), LINERs (L), transition objects (T), and absorption-line nuclei (A). Lbol is based on the X-ray (2-10 keV) luminosity. The hatched and open histograms denote detections and upper limits, respectively; type 1 objects are plotted in blue, type 2 objects in red. (Adapted from L.C. Ho, in preparation.)