Spitzer Reveals Hidden Quasar Nuclei in Some Powerful FR II Radio Galaxies Patrick Ogle Spitzer Science Center, California Institute of Technology, Mail Code 220-6, Pasadena, CA 91125 6 0 [email protected] 0 2 David Whysong1 & Robert Antonucci n a Physics Dept., University of California, Santa Barbara, CA 93106 J 0 2 ABSTRACT 1 v 5 We present a Spitzer mid-infrared survey of 42 Fanaroff-Riley class II ra- 8 dio galaxies and quasars from the 3CRR catalog at redshift z < 1. All of the 4 1 quasars and 45±12% of the narrow-line radio galaxies have a mid-IR luminosity 0 6 of νL (15µm) > 8 × 1043 erg s−1, indicating strong thermal emission from hot ν 0 dust in the active galactic nucleus. Our results demonstrate the power of Spitzer / h to unveil dust-obscured quasars. The ratio of mid-IR luminous narrow-line radio p - galaxies to quasars indicates a mean dust covering fraction of 0.56±0.15, assum- o r ing relatively isotropic emission. We analyze Spitzer spectra of the 14 mid-IR t s luminous narrow-line radio galaxies thought to host hidden quasar nuclei. Dust a v: temperatures of 210-660 K are estimated from single-temperature blackbody fits i X to the low and high-frequency ends of the mid-IR bump. Most of the mid-IR r luminous radio galaxies have a 9.7 µm silicate absorption trough with optical a depth < 0.2, attributed to dust in a molecular torus. Forbidden emission lines fromhigh-ionizationoxygen, neon, andsulfur indicateasourceoffar-UVphotons in the hidden nucleus. However, we find that the other 55±13% of narrow-line FR II radio galaxies are weak at 15 µm, contrary to single-population unification schemes. Mostofthesegalaxiesarealsoweakat30µm. Mid-IRweakradiogalax- ies may constitute a separate population of nonthermal, jet-dominated sources with low accretion power. Subject headings: galaxies: active, galaxies: quasars, galaxies: jets, infrared: galaxies 1 now at NRAO, Array Operations Center, P. O. Box O, 1003 Lopezville Rd., Socorro, NM 87801-0387 – 2 – 1. Unification of Quasars and Radio Galaxies The nature of the energy source in active galactic nuclei (AGNs) is a fundamental problem. The basic model attributes the large luminosity of these systems to gravitational energy release in an accretion disk around a supermassive black hole. A jet may be driven by magnetic fields threading the disk (Blandford & Payne 1982). The black hole spin energy may also be tapped and converted into electromagnetic Poynting flux and particles in a relativistic jet (Blandford & Znajek 1977; Punsly & Coroniti 1990; Meier 1999; De Villiers et al. 2004). Extragalactic radio sources are categorized by their morphology as either of two types (Fanaroff & Riley 1974). Fanaroff-Riley (FR) type I sources are edge-darkened, while FR IIs are edge-brightened. The different morphology of FR Is indicates that they are not related to FR IIs by orientation. FR Is also have lower radio luminosities than FR IIs for a given host galaxy luminosity (Owen & Ledlow 1994; Bicknell 1996) and most have low-ionization nuclear emission region (LINER) spectra (Hine & Longair 1979). However, not all FR Is can be characterized by low accretion power (Whysong & Antonucci 2004; Cao & Rawlings 2004). The present paper focuses on FR IIs, which contain powerful jets with bright terminal hot spots and lobes. Furthermore, we count broad-line radio galaxies (BLRGs) as low-luminosity quasars. Quasarsandnarrow-lineradiogalaxies(NLRGs)maybeunifiedbyorientation-dependent obscuration. Radio galaxies are thought to host quasar nuclei that are obscured by circum- nuclear dusty torialignedwiththeradio jets(Antonucci 1984). Unificationof radiogalaxies and quasars can therefore explain the lack of quasars viewed at large angles to the radio axis (Barthel 1989). The percentage of high-redshift radio galaxies (60% of the 3CRR FR II sample at z > 0.5) would then indicate a torus covering fraction of ∼ 0.6. However, there appears tobe a discrepancy between theredshift distributions of quasars and radio galaxies at z < 0.5, with a factor of ∼ 4 more narrow-line radio galaxies than quasars (Singal 1993). Furthermore, the median projected linear size of these ’excess’ radio galaxies is smaller than expected for quasars seen in the sky plane (Singal 1993; Whysong 2005). The unification hypothesis may be modified to include a second population of lower luminosity, low-excitation FR II radio galaxies (Wall & Jackson 1997; Grimes, Rawlings, & Willott 2004). Alternatively, it has been argued that the torus covering fraction may increase with decreasing radio luminosity (Lawrence 1991). The unification hypothesis has been qualitatively confirmed by spectropolarimetry of radio galaxies, many of which have been shown to have highly polarized broad emission lines and blue continuum, scattered from material which has a direct view of the active galactic – 3 – nucleus (Cimatti et al. 1997; Cohen et al. 1999). Of particular note are the original discovery of highly polarized broad Hα from the hidden quasar nucleus in 3C 234 (Antonucci 1984), and the discovery of highly polarized broad Hα in the spectrum of the powerful radio galaxy Cygnus A (Ogle et al. 1997). However, this method of detecting hidden quasars relies on an appropriately placed scattering region to view the otherwise hidden nucleus. Such a region is not guaranteed to exist for all radio galaxies, and thus spectropolarimetry can easily yield false negatives. Polarimetry is also ineffective at determining the luminosity of the hidden nucleus, since the scattering efficiency is usually unknown. Anotherwaytosearchforhiddenquasarnucleiistoobserveradiogalaxiesinthemid-IR. If the unification hypothesis is correct, the dusty torus should serve as a crude calorimeter of the central engine (Meisenheimer et al. 2001; Siebenmorgen et al. 2004; Haas et al. 2004; Whysong & Antonucci 2004). Optical, UV, and X-ray photons from the quasar nucleus are absorbed by dust in the torus and the energy is re-emitted in the thermal infrared. This explains why blue, UV color-selected quasars emit 10-50% of their luminosity in the IR (Sanders et al 1989;Haas et al. 2000). There appears to be no connection between the bulk of this IR emission and nonthermal radio emission, except in core-dominated radio sources such as blazars. Observations of matched 3CR quasars and radio galaxies by ISO indicate similar IR luminosities, consistent with the unification picture (Meisenheimer et al. 2001; Haas et al. 2004). However, differences in 24 µm/ 70 µm color may indicate that mid-IR emission from the torus is anisotropic by a factor of ≤ 3 (Shi et al. 2005). We present Spitzer observations of a sample of 42 FR II radio galaxies and quasars selected from the 3CRR survey. The goals are to search for mid-IR emission from hidden quasar nuclei and test the ubiquity of the unification hypothesis. The Spitzer Infrared Spectrograph (IRS) combines the advantages of unprecedented sensitivity from 5-36.5 µm to measure the mid-IR continuum and spectral resolution to measure high ionization emission lines powered by hidden AGNs. In the current paper, we present evidence for hidden quasar nucleibasedonmid-IRphotometryextractedfromtheIRSspectra. Weexamineindetailthe spectra of the subset of 14 mid-IR luminous radio galaxies which appear to contain hidden quasar nuclei. Spectra of the quasars and mid-IR weak radio galaxies and a statistical study of the complete sample will be presented in separate papers. 2. Sample We begin by selecting a well-defined, radio flux-limited and redshift-limited sample of 55 radio galaxies and quasars from the 3CRR catalog (Laing, Riley, & Longair 1983). We – 4 – include all 3CRR sources with FR II radio morphology, a flux of S > 16.4 Jy1 at 178 MHz, 178 and a redshift of z < 1. The original 3CRR catalog has a flux limit of 10 Jy at 178 MHz, is restricted to northern declinations (δ > 10◦), and has galactic latitude |b| > 10◦. It is the canonical low-frequency selected catalog of bright radio sources, has optical identifications and redshifts for all entries, and has been extensively observed in most wavebands. We select only sources with FR II radio morphology. We verify or update the FR classification of all sources by inspection of the latest published radio maps. Compact, steep- spectrum sources (CSSs: 3C 48, 138, 147, 286, and 309.1) with radio major axis D < 10 kpc (Fanti et al. 1985) are excluded from the sample because they may constitute a class of young or frustrated radio sources. Here and throughout this paper, we assume a cosmology with H = 70 km s−1 Mpc−1, Ω = 0.3, and Ω = 0.7. Size and morphology indicate that 0 m Λ CSSs are not related to FR IIs by orientation. It is essential for our unification studies that we select a sample based on isotropic radio lobe flux, and not on optical or IR properties, so that it is unbiased by orientation- dependent selection effects. In particular, our sample includes no blazars. No sources make the flux limit only because of beamed emission from the core of the radio jet. Our sample includes quasars as well as radio galaxies, and we use the quasar subsample as a control. We aim to determine whether and which narrow-line FR II radio galaxies have mid-IR power comparable to quasars or broad-line radio galaxies of similar radio lobe flux and redshift. The 42/55 sources in our sample which we have observed with Spitzer, or which have Spitzer data in the public archive are listed in Tables 1 and 2. The 25 mid-IR luminous sources with νL (15µm) > 8×1043 erg s−1 (14 NLRGs and 11 quasars or BLRGs) are listed ν in Table 1, and the 17 mid-IR weak galaxies with νL (15µm) < 8×1043 erg s−1 are listed ν in Table 2. The reason for this particular division is explained below. Optical source classifications are based on emission line properties. Type 1 sources have directly visible broad emission lines (quasars and BLRGs), and type 2 sources (NLRGs) do not. The NLRGs are further classified using their forbidden emission lines (Jackson & Rawlings 1997;Wilottetal. 1999)2. High-excitationgalaxies(HEGs) aredefined tohave [O iii]λ5007equivalent widthsof> 10˚A and[O iii]λ5007/[Oii]λ3727> 1. Thesources which do not meet these criteria are classified as low-excitation galaxies (LEGs). The equivalent width criterion ensures that [O iii] is measurable in moderate S/N spectra. However, it 1Theradiofluxlimitis15JyusingLaing,Riley,&Longair (1983)fluxvaluesand16.4Jyonthestandard Baars (1977) scale. 2 Updatedopticalclassificationsareavailableathttp://www-astro.physics.ox.ac.uk/∼cjw/3crr/3crr.html. – 5 – remains to be seen whether some sources with low [O iii] equivalent width might have [O iii] λ5007/[O ii] λ3727 > 1. In addition, we caution that [O ii] and [O iii] may be subject to differing amounts of extinction. 3. Observations We observed the sources in our sample with the Infrared Spectrograph (IRS) on the SpitzerSpaceTelescope (Houcket al. 2004;Werner etal. 2004). Weusedthelow-resolution (R ∼ 64−128) modules Short-Low (SL) and Long-Low (LL) for accurate spectrophotometry over the wavelength range 5-36.5 µm. Wavelengths 36.5-40 µm are unusable because of low-S/N and 2nd-order bleed-through caused by filter delamination in LL 1st order (LL1). The absolute and relative flux accuracies of IRS are generally better than 10% and 4%, respectively, as judged from observations of bright standard stars. However, additional low- level instrumental artifacts may become important for faint sources. We used IRS in standard flux-staring mode, for 2 cycles at 2 nod positions in each of the modules SL1, SL2 (SL 1st and 2nd order), and LL2 (LL 2nd order). We executed 1 cycle at 2 nod positions for LL1, which covers the 20-36.5 µm range. A typical observation includes 240 s of on-source exposure time in each of SL1, SL2, and LL1, and 480 s in LL2, for a total of 2000 s per target (including overhead). Archived IRS data are used for 14 sources which were observed partly or in full by other investigators (with similar or longer exposure times). Nodoroff-slitobservationsweresubtractedtoremove foregroundemissionfromthetele- scope, zodiacal light, and interstellar medium. Spectra were then extracted from the Basic- Calibrated Datasets (BCDs), using the Spitzer IRS Custom Extraction (SPICE3 version 1.1) software and standard tapered extraction windows. The extraction window full-widths are proportional to wavelength in each order to match the diffraction-limited telescope point- spread function (SL2: 7′.′2 at 6 µm, SL1: 14′.′4 at 12 µm, LL2: 21′.′7 at 16 µm, LL1: 36′.′6 at 27 µm). We rebinned portions of the spectra of 3C 55, 172, 220.1, 244.1, 263.1, 280, and 330 by factors of 4-8 in order to improve the S/N at short wavelengths. Spectral orders were trimmed at the edges and merged to produce final spectra. TheSLandLLslitshavewidthsof3′.′7and10′.′6,respectively. Standardpoint-sourceflux calibrations (version 12.0) were applied to correct for slit and aperture losses and convert the spectra from electron s−1 to Jy. In most cases, fluxes match to < 15% across order boundaries, consistent with a point source that is well-centered in all of the slits. However, 3http://ssc.spitzer.caltech.edu/postbcd/spice.html – 6 – in 5 cases (3C 192, 216, 220.1, 380, and 381) SL2 fluxes are larger by 17-35% relative to the other orders. Assuming that these mismatches owe to variable slit-loss caused by pointing errors, the orders with low flux are adjusted upward to match the orders with high flux. Order mismatches may alternatively be an indication of extended mid-IR emission. The results for a few sources with nearby neighbors in the slit should be viewed with caution. In the case of 3C 310, a nearby companion galaxy (to the east) may contribute a significant fraction of the flux (< 50%) in the LL1 slit. Similarly, a nearby source may potentially contribute to the LL spectra of 3C 438 (which is, however undetected at 15 µm). The SL2 spectrum of 3C 388 may be weakly affected by flux from a nearby star on the slit (< 20%). The northern component of the double nucleus in 3C 401 falls outside of the SL slits, but falls inside the LL slit used to measure the 15 µm flux. 3.1. Mid-Infrared and Radio Luminosities We measure the mean 6.5−7.5 µm and 13.0−17.0 µm flux densities F (7 and 15 µm, ν rest)ofeachtarget(Tables1&2). AllSpitzerfluxdensitiesinthispaperareinobservedunits at a constant rest-frame wavelength defined by λ = λ /(1+z), where z is the redshift rest obs measured from optical emission lines and cataloged in the NASA Extragalactic Database (NED4). This avoids any complication frompotentially largecosmological K-corrections that could otherwise beintroduced by a steep IR continuum slopeor redshifted silicate absorption features. We choose to measure the mid-IR flux at 7 and 15 µm to avoid the 9.7 µm trough and the deepest part of the 18 µm silicate absorption trough. We exclude the 14.0-14.5 µm and 15.3-15.8 µm wavelength regions from our photometry, to avoid emission from Ne v and Ne iii. The 7 and 15 µm bands are within the Spitzer IRS bandpass for redshifts z < 1.28. However, the archival LL data for two quasars (3C 254 and 275.1) are not yet public. We extrapolate their SL spectra to obtain F (15 µm, rest) using F (7 µm, rest) and the observed ν ν (relatively line-free) 5-7 µm spectral index. Radio luminosities νL (178 MHz, rest) are estimated from the observed 178 MHz fluxes ν and K-corrected using the 178-750MHz radio spectral index (Laing, Riley, & Longair 1983). The sources in our sample display a large range of nearly 3 orders of magnitude in mid-IR to radio luminosity: νL (15 µm)/νL (178 MHz) = 0.8−680 (Fig.1). This quantity is thought ν ν to reflect the relative importance of accretion luminosity and jet kinetic power dissipation. 4http://nedwww.ipac.caltech.edu – 7 – However, different size and time scales are probed by the radio (10 kpc-1 Mpc) and mid-IR (0.1-100pc), andtheradio power maybesensitive todifferences inenvironmental conditions. For the purpose of studying quasar and radio galaxy unification, it is natural to divide the sample into the mid-IR luminous NLRGs which emit as powerfully as quasars or BLRGs, and the mid-IR weak NLRGs that do not. We adopt an empirical dividing line of νL (15 ν µm) > 8 × 1043 erg s−1 to separate hidden quasars from mid-IR weak radio galaxies. The cutoff is set at 1/2 the luminosity of the mid-IR weakest BLRG (3C 219) to allow for some degree of anisotropy at 15 µm. Fourteen NLRGs satisfy our criterion and are thus likely to contain hidden quasar or BLRG nuclei (Table 1). Notably, all of these NLRGs are optically classified as HEGs. The 17 mid-IR weak NLRGs with νL (15 µm) < 8 × 1043 erg s−1 (Table 2) have ν mixed optical classifications, including both HEGs and LEGs. These sources have lower S/N mid-IR spectra, which will be considered in detail in a later paper. Six mid-IR weak NLRGs (including 2 HEGs and 4 LEGs) are undetected by Spitzer at 15 µm, and one is also undetected at 7 µm. 3.2. Hidden Quasar Spectra 3.2.1. Continuum Emission We now present Spitzer spectra of the 14 mid-IR luminous NLRGs that ostensibly con- tain hidden quasar nuclei (Figs. 2-4). We also plot the spectral energy distributions (SEDs) of the sources with published near-IR photometry (Fig. 5). The collected photometric data were measured in the J, H, K, L′, and M wavelength bands from the ground (Lilly & Longair 1984; Lilly, Longair, & Miller 1985; Simpson, Rawlings, & Lacy 1999; Simpson & Rawlings 2000). The photometric apertures range in size from 3−11′′, with preference given to the apertures that most closely match the Spitzer SL slit width. Where available, the ground- based L′ and M-band photometry agrees with Spitzer spectrophotometry remarkably well. There is no indication of variability over the time span of 20 yr. Fourofthelow-redshift NLRGs(3C33,234,381,and452)havebroadpeaksintheirνL ν spectra (and SEDs) at 1.5−2.5×1013 Hz (12-20 µm). A maximum and spectral curvature near 20 µm are also suggestive of broad peaks in the Spitzer spectra of 3C 55, 244.1, 265, and 330. The large amplitude (∼ 0.5 − 1.0 dex) of the mid-IR bump (Fig. 5) excludes a large contribution of synchrotron emission to the mid-IR continuum of most sources. This is not surprising if the equatorial plane of the dusty torus is roughly perpendicular to the radio jet, such that jet emission is beamed away. The high redshifts of the NLRGs 3C 172, – 8 – 220.1, 263.1, 268.1, and 280 preclude the identification of a mid-IR bump in the SEDs of these sources. The unusually flat, blue SED of 3C 433 may indicate a quasar viewed at low inclination (Section 3.2.2). We attribute the mid-IR continuum bump visible in most sources to thermal emission from warm or hot dust. Fitting the mid-IR peak with a single-temperature blackbody model indicates dust with a temperature of 210− 225 ± 0.5 K (Fig. 5). While this temperature characterizes the peak of the mid-IR SED, hotter dust must also be present. At frequencies greater than the peak of the SED (2.0 − 7.5 × 1013 Hz), the continuum emission of most sources can be characterized using a power law with spectral index α = 1.1−2.1 (Table 1 & Fig. 6). This emission likely comes from a continuous distribution of dust temperature. We measure the spectral index between 7 and 15 µm, avoiding the 9.7 µm and 18 µm silicate absorption troughs. The most blue and apparently hottest mid-IR luminous NLRG is 3C 265, while the most red and coolest are 3C 55 and 3C 268.1 (Fig. 6). In comparison, some mid-IR weak sources such as 3C 310 and 3C 388 are quite blue (α ∼ −0.1−+0.7), indicating a large contribution of starlight from the host galaxy to the 7 µm continuum. The near-IR continuum shifts into the Spitzer IRS passband for the highest redshift (z > 0.7) sources. The spectra of the NLRGs 3C 55 and 3C 265 steepen above 7.5 × 1013 Hz (below 4 µm). Fitting these spectral turnovers with single-temperature blackbodies, we find emission from hot dust with temperatures of 520±10 K and 660±10 K, respectively. Altogether, the mid-IR luminous radio galaxies in our sample show emission from dust with temperatures distributed in the range 210-660 K. Hotter temperature dust (up to the sublimation temperature) may be present but not visible for radio galaxy tori viewed at high inclination (Pier & Krolik 1992). Extinction by cold foreground dust in the host galaxy may also affect the spectral index. For Galactic-type dust, A(7,15,35 µm)/A(V) = (0.020,0.015,0.004) (Mathis 2000). An extinction of A(V) = 100 would steepen the 7-15 µm spectral index by δα = 0.6 (Fig. 6a). The observed range in spectral index for the mid-IR luminous NLRGs is δα = 1.0, corresponding to A(V,7,15,35 µm) = (167,3.3,2.5,0.6) mag. Thus if reddening by a cold foreground dust screen accounted entirely for the range in mid-IR slope, the mid-IR emission could be anisotropic by factors of f (7,15,35 µm) ∼ (22,10,1.3). However, these are upper A limits since variations in the spectral index are also controlled by the physical temperature distribution of the visible dust. The SEDs of several sources (3C 33, 55, 172, 265, and 452) have upturns at short wavelengths, which we attribute to stellar emission from the host galaxy (Fig. 5). The wavelength oftheupturn(1-5µm)isanindicator oftherelativestrengthofthemid-IRbump seen from our direction vs. host galaxy light, occurring at shorter wavelength for sources – 9 – with a stronger mid-IR bump. This may have important consequences for understanding the K-z Hubble diagram for 3C radio galaxies, for which it has been argued that AGNs contribute a negligible fraction of the K-band flux (e.g., Simpson & Rawlings 2000). This may be incorrect for a few of the most luminous mid-IR sources in our sample, including 3C 234 and 3C 280 where there appears to be much emission from hot dust in the K band. Detailed spectral modeling, combined with radio orientation indicators, promises to further characterize the temperature distribution, optical depth, and inclination of the dusty torus that is thought to produce most of the mid-IR emission from hidden quasar nuclei. Such an analysis is, however, outside the scope of the present paper. 3.2.2. Silicate Absorption The silicate absorption trough at 9.7 µm is detected in 12/14 of the mid-IR luminous NLRGspectra(Table3&Figs. 2-4). TheequivalentwidthEW andapparentopticaldepth 9.7 τ are measured relative to a local continuum fit to either side of the trough, indicated in 9.7 Figures 2-4. The optical depth is averaged over the trough bottom (rest 9.2-10.2 µm) to improve the S/N. It should be kept in mind that the apparent τ is just a convenient 9.7 parameterization of (and lower limit to) the total optical depth since there must also be broad-band silicate absorption of the adjacent continuum. Theapparentsilicateopticaldepthsaresmall(τ = 0.02−0.2),forallbut3C55and3C 9.7 433. If attributed to foreground dust screens, this would indicate optical extinction of only A = 0.2−5.1 mag (Fig. 6b), assuming a Galactic extinction law with A /τ = 12.3−25.6 V V 9.7 mag (Rieke & Lebofsky 1985; Draine & Lee 1984). The extinction values are clearly underestimated since they imply that the hidden nuclei in 3C 234, 265, 381, and 452, which haveτ ≤ 0.1,shouldbereddenedbutdirectlyvisibleatHα. Thesamediscrepancybetween 9.7 τ and estimates of extinction at shorter wavelengths is seen for the hidden quasar nucleus 9.7 in Cygnus A, and attributed to a radial gradient in torus dust temperature (Imanishi & Ueno 2000). The observed range of τ may correspond to a range of equatorial silicate 9.7 dust column densities in the torus, or alternatively a range of viewing angles. In this regard, more detailed modeling of the torus, including its geometric and temperature structure is clearly called for. Filling-in of the silicate troughs by silicate emission from the torus or narrow-line region (NLR)mayalsoreducetheapparent silicateopticaldepthsinsomesources. Thisispredicted for an optically thick torus viewed at an intermediate or face-on inclination (Pier & Krolik 1992). Recently, strong silicate emission features were detected by Spitzer in several radio- – 10 – loud (3C) and radio quiet (PG) quasars (Siebenmorgen et al. 2005; Hao et al. 2005). The failure of previous attempts to observe this feature inspired torus models with large dust grain size (Laor & Draine 1993) or a spatial distribution of optically thick clumps (Nenkova, Ivezic, & Elitzur 2002). However, it appears that past non-detections owe to inadequate wavelength coverage to determine the underlying continuum. The NLRGs 3C 55 and 3C 433 have significantly deeper silicate troughs than other NLRGs, with τ = 0.9 and 0.7, respectively (Fig. 6b). We suggest that their active nuclei 9.7 and tori are absorbed by an additional (kpc-scale) cold dust screen in the host galaxy. As noted above, the NLRG 3C 433 is unusual in having a flat, blue continuum (similar to some of the quasars in our sample). A blue mid-IR spectrum is not necessarily at odds with deep silicate absorption features. It can be understood if this is a quasar viewed at low inclination to the jet and torus axes, but through an (A ∼ 10) cold dust screen. This amount of V extinction would result in very little reddening at 7-15 µm (δα = 0.05−0.14), but would be sufficient to create the deep 9.7 µm trough (Fig. 6b) and would obscure any optical broad lines. The NLRG 3C 433 is also unique in having the only unambiguously detected 18 µm silicate trough, with equivalent width EW = 0.42± 0.01 µm and apparent optical depth 18 τ = 0.07±0.03 (averaged over 17-19 µm). The ratio of τ to τ apparent silicate trough 18 18 9.7 depths is 0.10±0.04, consistent with a 0.11 ratio for Galactic-type silicate dust (Draine & Lee 1984). We do not see the full 18 µm silicate trough in the spectrum of 3C 55 because of inadequate rest-wavelength coverage. 3.2.3. Forbidden Emission Lines All of the mid-IR luminous NLRGs with high S/N spectra have forbidden emission lines from highly ionized metals, including [O iv] λ25.89 µm, [Ne ii] λ12.81, [Ne iii] λ15.55, [Ne v] λ14.3, [Ne v] λ24.31, [Ne vi] λ7.65, [S iii] λ18.71, [S iii] λ33.48, and [S iv] λ10.51 (Figs. 2-4). We measure the line flux and rest equivalent width of each emission line relative to the local continuum level (Table 4). Formal uncertainties are computed from the noise in the continuum to either side of the line. Upper limits are estimated for undetected emission lines, assuming they are unresolved. The large range of ionization states (especially high-ionization Ne v, Ne vi, and S iv) indicates photoionization by a hidden source of far-UV photons (Alexander et al. 1999; Sturm et al. 2002; Armus et al. 2004), e.g. a quasar nucleus. Low critical densities in the range 103 −106 cm−3 (Alexander et al. 1999) indicate that the forbidden lines arise in the