THE NATURE OF OPTICALLY DULL ACTIVE GALACTIC NUCLEI IN COSMOS^*

We draw the sample of optically dull AGNs from the COSMOS (Scoville et al. 2007), a survey over 2 deg² of sky with deep multiwavelength observations. The XMM-Newton observations of COSMOS reach limiting fluxes of 1.7 × 10⁻¹⁵ erg cm⁻² s⁻¹ and 9.3 × 10⁻¹⁵ erg cm⁻² s⁻¹ in the 0.5–2 keV and 2–10 keV energy bands, respectively (Cappelluti et al. 2009). The optical and infrared counterparts to the X-ray sources are presented in M. Brusa et al. (2009, in preparation) and all counterpart matches were visually inspected. Spectroscopic follow-up of X-ray targets with i_AB ⩽ 23 is described in Trump et al. (2009a). In particular, the 48 optically dull AGNs are the objects of Trump et al. (2009a) classified as "a" types (absorption line spectra) with 90% redshift confidence. All objects lack strong emission lines (see Section 2.2) in the optical spectra and satisfy one of the two X-ray AGN criteria:

$$\begin{equation} L_{0.5\hbox{--}10\,{\rm keV}} > 3 \times 10^{42}\;{\rm erg\;s}^{-1}, \end{equation} \tag{ 1 }$$

$$\begin{equation} -1 \le X/O \le 1. \end{equation} \tag{ 2 }$$

In Equation (2), X/O = log f_X/f_O = log(f_{0.5–2 keV}) + i_AB/2.5 + 5.352. These constraints are set by the limit on X-ray luminosity in local star-forming galaxies of L_X ≲ 10⁴² erg s⁻¹ (e.g., Fabbiano 1989; Colbert et al. 2004) and the traditional "X-ray AGN locus" of Maccacaro et al. (1988). These equations have been shown to be quite reliable in selecting AGNs, although they are probably overly conservative (e.g., Hornschemeier et al. 2001; Alexander et al. 2001; Bauer et al. 2004; Bundy et al. 2007). Of the 48 optically dull AGNs, 44 meet both criteria, with only four meeting one criterion but not the other. The optically dull AGNs are additionally restricted to z ⩽ 1, since beyond these redshifts the 4000 Å break shifts beyond the observed spectral range of (Trump et al. 2009a) and it becomes extremely difficult to measure redshifts from absorption lines. The 48 optically dull AGNs are all of the z < 1 AGNs within the 2 deg² of COSMOS that meet either of the X-ray criteria and have Magellan/Inamori Magellan Areal Camera and Spectrograph (IMACS) or Sloan Digital Sky Survey (SDSS) spectroscopy.

Of the optically dull AGNs, 45/48 have optical spectra from observations with the IMACS (Bigelow et al. 1998) on the 6.5 m Magellan/Baade telescope. These spectra have wavelength ranges of 5600–9200 Å, with a resolution element of 10 Å (5 pixels). All targets were selected as AGN candidates by their X-ray emission. Details of the observations and reductions are presented by Trump et al. (2009a), and all of the spectra are publicly available on the COSMOS IRSA server (http://irsa.ipac.caltech.edu/data/COSMOS). The optically dull AGNs in this work all have high-confidence redshifts (z_conf ⩾ 3), which empirically corresponds to a 90% likelihood of the correct redshift measurement (Trump et al. 2009a).

Three of the optically dull AGN spectra come from archivalSDSS (York et al. 2000) observations. These sources were selected by their X-ray emission, but were excluded from the main Magellan/IMACS survey because their redshifts were already known. Their wavelength coverage is 3800–9200 Å and their resolution element is 3 Å (3 pixels).

Figure 1 shows the measured [O ii] (λ3727 Å) and Hβ (λ4861 Å) narrow emission line luminosities for the optically dull AGNs (black squares), along with a comparison sample of Type 2 AGNs (blue diamonds) from Trump et al. (2009a). To compute the line luminosity, we first define a straight-line continuum by averaging the spectral regions 30–40 Å redward and blueward of the line region. The line luminosity is then measured across the continuum-subtracted region 1000 km s⁻¹ about the line center. The 1000 km s⁻¹ width is a conservative limit since <1% of Type 2 AGNs have narrow emission lines broader than 1000 km s⁻¹ (Hao et al. 2005). The measured error for each line luminosity is computed using both the spectral error and the error of the continuum fit. When the line luminosity was less than its 5σ error, we used the 5σ error as an upper limit on line luminosity.

Zoom In Zoom Out Reset image size

All 38 optically dull AGNs with Hβ in the observed wavelength range have L_Hβ < 5σ_L(Hβ) and thus are assigned only upper limits in the bottom panel of Figure 1. However, 9/40 optically dull AGNs with [O ii] in the observed wavelength range have a line luminosity greater than the 5σ threshold, despite the fact that the classification of Trump et al. (2009a) identified no emission lines. Still, even when detected, the emission line luminosities of the optically dull AGNs are much lower than those of typical Type 2 AGNs. If the optically dull AGNs were simply Type 2 AGNs observed at low signal-to-noise ratio (S/N), we might expect poorly constrained upper limits on line luminosity. This is not the case, as the upper limits are 10–100 times lower than the line luminosities of typical Type 2 AGNs. Optically dull AGNs are not Type 2 AGNs with low S/N, but have much less luminous emission lines than Type 2 AGNs of similar X-ray luminosities.

The optical and infrared photometry of the optically dull AGN is drawn from the catalog of P. Capak et al. (2009, in preparation). Table 1 shows the depths, wavebands, and year of observation for the COSMOS photometry used here.

Table 1. COSMOS Optical and Infrared Photometry

Filter	Telescope	Center λ (Å)	FWHM (Å)	Depth (30) (magAB)	Epoch UTC
BJ	Subaru	4460	897	27.7	2004
g+	Subaru	4750	1265	27.1	2005
VJ	Subaru	5484	946	27.0	2004
r+	Subaru	6295	1382	27.1	2004
i+	Subaru	7640	1497	26.7	2004
z+	Subaru	9037	856	25.7	2004
IA427	Subaru	4271	210	26.5	2006
IA464	Subaru	4636	227	26.0	2006
IA484	Subaru	4842	227	26.5	2007
IA505	Subaru	5063	232	26.2	2006
IA527	Subaru	5272	242	26.5	2007
IA574	Subaru	5743	271	26.2	2007
IA624	Subaru	6226	299	26.3	2006
IA679	Subaru	6788	336	26.1	2006
IA709	Subaru	7082	318	26.3	2007
IA738	Subaru	7373	322	26.1	2007
IA767	Subaru	7690	364	25.9	2007
IA827	Subaru	8275	364	25.8	2006
NB711	Subaru	7126	73	25.4	2006
NB816	Subaru	8150	119	26.1	2005
IRAC1	Spitzer	35263	7412	23.9	2006
IRAC2	Spitzer	44607	10113	23.3	2006
IRAC3	Spitzer	56764	13499	21.3	2006
IRAC4	Spitzer	77030	28397	21.0	2006

Download table as:  ASCIITypeset image

The optical photometry was taken on the Subaru telescope, with observations of the 6 broad, 12 intermediate, and 2 narrow bands described in Taniguchi et al. (2007) and Y. Taniguichi et al. (2009, in preparation). Fluxes were measured in 30 diameter apertures, on point-spread function (PSF)-matched images with FWHM of 15, and simulations (Capak et al. 2007) show that the 30 diameter aperture contains 76% of the total flux for a point source. We additionally correct each optical magnitude by the zero-point correction from Ilbert et al. (2009).

The infrared photometry is derived from Spitzer/IRAC observations. The closest IRAC source within 10 of the optical counterpart to the XMM-Newton source was chosen as the infrared counterpart. IRAC fluxes are given in the COSMOS–IRAC catalog for 38 diameter apertures, so we translate these into 30 diameter aperture fluxes as described in Salvato et al. (2009). All of our optically dull AGNs were unambiguously detected in all four IRAC bands.

We discuss morphological data of the optically dull AGN host galaxies from observations with the Advanced Camera for Surveys (ACS) on the Hubble Space Telescope (HST), fully described in Koekemoer et al. (2007). The COSMOS field was imaged in the F814W filter for 583 orbits, reaching a limiting magnitude of AB(F814W) ⩽27.2 (5σ). Because the HST/ACS imaging only covers 1.64 deg² of the 2 deg² COSMOS field, 3/48 optically dull AGNs lack HST/ACS coverage. Gabor et al. (2009) provides morphological data for 37 of the remaining optically dull AGNs, from the point-source subtracted host galaxies. (The other eight optically dull AGNs have HST/ACS imaging, but do not have morphological data because the resultant fit was wildly unphysical or did not converge.)

All of the optically dull AGNs are detected in the Subaru and IRAC photometry, and all within the HST areal coverage were detected in ACS, so these do not affect the completeness limits. The soft 0.5–2 keV X-ray limit of 1 × 10⁻¹⁵ erg cm⁻² s⁻¹ means the X-ray AGN sample is complete to all AGNs meeting the luminosity criterion (Equation (1)) of L_{0.5–10 keV} > 3 × 10⁴² erg s⁻¹ at z ≲ 1 (see Figure 9 of Trump et al. 2009a). Correct identification of optically dull AGNs is also limited to z ≲ 1, since at higher redshifts the 4000 Å break in these objects is redshifted beyond the observed wavelength range, and high-confidence identification becomes difficult. The Magellan/IMACS spectroscopy is uniformly 90% complete to i_AB ⩽ 22 absorption line objects. The optically dull AGN sample is then limited by z ≲ 1 and i_AB ⩽ 23, but is 90% complete to only i_AB ⩽ 22.

Each optically dull AGN spectrum lacks strong emission lines and has the red shape and absorption signature (H+K lines, 4000 Å break, etc.) of an old, red elliptical galaxy. However, the spectra often have low S/N, and most (45/48) are limited by the 5600–9200 Å wavelength range of Magellan/IMACS. The 20 bands of high-S/N optical photometry allow us to take a broader look at the optical SED.

We fit the optical photometry of each optically dull AGN with an "r+q" template that is a mix of a red galaxy component (the SDSS red galaxy composite from Eisenstein et al. 2001) and a blue AGN component (the SDSS quasar composite from Vanden Berk et al. 2001). The scale of each component is an independent free parameter. The two components of the hybrid "r+q" template are well motivated for two reasons: (1) from the X-ray properties the objects must have an AGN and (2) the optical spectrum most closely resembles a red galaxy. While both the host and any underlying AGN will not be perfectly described by the "r" and "q" components of the template, we explore the minor systematic deviations below.

We find the red galaxy and AGN components in the best-fitting template by maximizing the Bayesian probability function, $P = \prod \frac{1}{\sqrt{2\pi \sigma _m^2}} \exp {(\frac{-0.5 (m-m_t)^2}{\sigma _m^2})}$. Here, m is the observed magnitude, σ_m is its error, and m_t is the template magnitude computed by measuring the template flux through the same wavelength response function as the observed magnitude. (The χ²₀ parameter is the logarithm of this probability function, but we choose the Bayesian approach because it maps out the probability distribution, not just the best-fit values.) In the fits for all objects, the best-fit fractions of AGN and red galaxy are tightly constrained: the 99% confidence intervals for the fit contain deviations of <3%.

Systematic errors will dominate over the fitting errors, however, because the "r+q" template is not likely to be a perfect fit to the observed data. First, the optically dull AGNs may have active or recent star formation contributing to the blue emission, causing us to overestimate the AGN emission. The contribution from a young stellar population (O/B star) is likely to be minor, since the emission line luminosities for the optically dull AGNs are very low (see Section 2.2). A moderate age (A star) stellar population would not have strong emission lines, but must also be a minor contributor at best because none of the optically dull AGN spectra show a Balmer break. We estimate the effect of any blue star-forming component as <20%, since any higher contribution would lead to emission lines or a recognizable Balmer break for even the lowest S/N optically dull AGNs. In addition, the AGN template of Vanden Berk et al. (2001) may not well describe the optical shape of the true underlying AGN, since AGNs can be heavily reddened (Hopkins et al. 2004) or obscured (Elitzur 2008), and even unobscured quasars are known to exhibit a wide variety of optical spectral shapes (Richards et al. 2006). Still, Vanden Berk et al. (2001) notes that the Type 1 AGNs have a variation of only σ < 20% from the mean SED in their blue (λ < 4000 Å) continua. So we can assume that our AGN fractions are valid, with the caveat that the true optical AGN emission may differ by up to 20%. While the "r+q" template may not recover the true optical AGN fraction of the optically dull AGNs, our estimated AGN fraction is useful as rough estimate for studying the host-subtracted X/O fraction.

We show examples of our template fits in Figure 2. In each panel, the black points show the photometry, with the x-error bar showing the bandwidth and the y-error showing the photometric error. The black histogram is the observed spectrum from Magellan/IMACS or the SDSS, and the blue histogram is the best-fit "r+q" template. Figure 2 additionally includes a best-fit red-galaxy-only ("r") template, shown in red, to illustrate the improvement of including a blue AGN component in the template fit. Reduced χ² values, and the blue AGN contribution of the "r+q" template fit, are shown in the upper left of each panel. Note that the χ²₀ values are quite large because the optical photometry has very small errors and the 14 narrow and intermediate bands are sensitive to details which are not well described by our templates. But while the fits do not perfectly describe the details of the optically dull AGN SEDs, the templates are useful for studying the shape of the SEDs and providing a rough estimate of the relative blue AGN and red host components.

Zoom In Zoom Out Reset image size

The top panel represents optically dull AGNs with virtually no quasar contribution in the optical photometry, where the highest probability "r+q" template had zero quasar fraction. Five of the 48 optically dull AGNs had similar fits, with virtually no blue AGN emission. The upper limit on any blue quasar excess in these objects is typically only 2% blue AGN component.

The second panel of Figure 2 shows an example of an optically dull AGN with a significant quasar component in the fit. In this example, the best-fit "r+q" template is significantly better than the best-fit "r" template, with a much lower reduced χ². The majority of the optically dull AGNs, 28/48 objects, exhibited similar fits, with χ²₀(r) ⩾ 2χ²₀(r + q). The blue AGN contribution in these objects is typically 15%–35%.

The third panel represents optically dull AGNs where the best-fit "r+q" template is only a slight improvement over the plain "r" red galaxy template. These AGNs have only a very weak excess of blue emission, completely invisible in the observed spectrum and only barely detected in the optical photometry. Of the 48 optically dull AGNs, 15 exhibit similar fits, with blue AGN contribution of about 3%–7%.

We can additionally compare the predicted line fluxes of the quasar component in the best-fit template to the line flux limits in the optical spectrum. The Hβ and [O ii] narrow emission lines (shown in Figure 1) do not work well for this comparison because we use a quasar template in our fit, and these narrow lines are often weak or nonexistent in Type 1 AGN. However the [O iii] (λ5007) narrow emission line is typically strong in both Type 1 and Type 2 AGN and so is useful for the comparison. Only 30 optically dull AGNs have [O iii] in their observed wavelength range, and all of these are upper limits only. Most (19/30) of these AGNs have predicted [O iii] fluxes from the best-fit model which lie below the upper limit on [O iii] flux from the spectrum. Since these AGNs have low predicted line fluxes and the measured [O iii] fluxes are only limits, it is not a strong constraint, but it does suggest that these 19 optically dull AGNs could be diluted Type 1 or Type 2 AGNs.

The optical photometry also reveals significant variability in four optically dull AGNs. When comparing the observations from 2004, 2006, and 2007 (see Table 1), these four AGNs exhibited changes in flux 5σ beyond the photometry errors. We show an example of a variable optically dull AGN in Figure 3. Magnitudes from each of 2004, 2006, and 2007 are shown in each panel in blue, with the corresponding template fit shown in red in each panel. The template fit to all 20 bands of optical photometry (from all years) is shown in gray, along with the Magellan/IMACS spectrum in black, for comparison in each panel. The optically dull AGN decreases in flux from 2004 to 2007, but almost all of this change is in the blue emission. In the template fit, the red galaxy component remains nearly the same in each year while the AGN component decreases from 43% to 15% contribution.

Zoom In Zoom Out Reset image size

Old red galaxies do not change in flux over different years of observations. Type 1 AGNs, however, can vary by as much as factors of a few on year timescales (e.g., Morokuma et al. 2008; Kelly et al. 2009). The four variable optically dull AGNs must then have a weak AGN causing the variability. The source of the variability must be ⩽1 lt-yr in size, making obscuration or reddening extremely unlikely. The variable optically dull AGNs are instead likely to be diluted "normal" AGNs. Indeed, close inspection of Figure 3 shows that the optical spectrum may have a weak Hβ broad emission line, although it is difficult to positively identify the line because of low S/N in that part of the spectrum. (Section 2.2, however, showed that for this and other optically dull AGNs, the narrow [O ii] and Hβ lines are not hidden by low S/N, but are instead very weak compared to those of Type 2 AGNs.) So while optically dull AGNs do not have strong emission lines, the four variable objects in particular show evidence for a diluted (not obscured) AGN. These objects are likely to be normal, unobscured Type 1 AGNs diluted by extranuclear light (as we explore in Section 3.5).

The defining characteristic of optically dull AGNs is that they are bright in X-rays while their optical spectra have no sign of emission lines. But while optically dull AGNs lack the emission line signature of an AGN, Section 3.1 showed that they do have excess blue emission which might be attributed to a diluted AGN. But are optically dull AGNs simply diluted by a bright host, or is their optical emission actually depressed when compared to their bright X-rays?

We present the ratio between the X-ray and optical flux in Figure 4, where log f_X/f_O = log(f_X) + i_AB/2.5 + 5.352. The optically dull AGNs are shown as squares, and the four variable objects are indicated by filled squares. For comparison the Type 1 (broad line) and Type 2 (narrow line) X-ray AGNs of Trump et al. (2009a) are shown in gray. At left, we use the total f_O for the optically dull AGNs, and all but two of the optically dull AGNs have f_X/f_O values consistent with typical AGNs. At right, the i_AB magnitude includes only the AGN fraction as determined in Section 3.1. It is important to note that X-ray K-corrections will cause Compton-thick AGNs at higher redshifts to have higher f_X/f_O ratios, (e.g., Comastri et al. 2003), although this effect should be minimal in our sample because very few of the optically dull AGNs are Compton thick (see Section 3.4) and all have z < 1.

Zoom In Zoom Out Reset image size

Even after subtracting out the host component, 33/48 optically dull AGNs have f_X/f_O values consistent with typical AGNs. These optically dull AGNs might be normal AGNs diluted by their hosts. However, we note that host dilution should push objects to the left in Figure 4, so host dilution may be unlikely for AGNs with f_X/f_O ∼ 1 and 33/48 may be an upper limit on the true fraction of optically dull AGNs diluted by their hosts. The 15 AGNs with f_X/f_O > 1 present the most interesting case, since host dilution is impossible and some physical effect must depress their optical emission while they remain X-ray bright.

Bright AGNs are well known to have redder Spitzer/IRAC colors than normal galaxies (Lacy et al. 2004; Stern et al. 2005) as a result of strong mid-IR power-law continua (Sajina et al. 2005; Donley et al. 2007). The IRAC emission is generally associated with the hot, dusty "torus," or outer accretion disk of AGNs. Since optically dull AGNs are optically fainter than normal AGNs, they might also have different mid-IR properties, with the power-law continuum either diluted or absent.

The IRAC colors are shown in Figure 5. The optically dull AGNs are marked with red squares, while Type 1 and Type 2 AGNs from Trump et al. (2009a) are shown as blue crosses and green diamonds, respectively. Type 1 AGNs generally have the reddest IRAC colors, followed by Type 2 AGNs, while most optically dull AGNs have IRAC colors consistent with normal galaxies. The [3.6 μm]_AB − [4.5 μm]_AB color does a particularly good job of separating the various AGN types. We additionally show the [3.6 μm]_AB − [4.5 μm]_AB color with soft X-ray luminosity in the right panel of Figure 5. Donley et al. (2007) suggested that the mid-IR power-law continuum disappears at low X-ray luminosities, but this does not appear to be the case for our optically dull AGNs. While many Type 1 AGNs are more X-ray luminous than the optically dull AGNs, many have similar luminosities, and there is no apparent correlation between IRAC color and X-ray luminosity in Figure 5. The optically dull AGNs have IRAC colors consistent with normal galaxies even though they are as X-ray luminous as some Type 1 AGNs.

Zoom In Zoom Out Reset image size

Several authors have suggested that the optical emission of optically dull AGNs is obscured, either by material near the central engine (Comastri et al. 2002; Civano et al. 2007) or by gas and dust in the host galaxy (Rigby et al. 2006). But if the optical emission is obscured, then the X-ray emission would probably be obscured as well (so long as the obscuring material for X-ray and optical emission is cospatial). For the 28 optically dull AGNs with >50 full band counts in their XMM-Newton or Chandra (Elvis et al. 2009; G. Lanzuisi et al. 2009, in preparation) observations, we estimate N_H from X-ray spectral analysis. We fit each X-ray spectrum as an intrinsically absorbed power law with Galactic absorption (N_H,gal = 2.6 × 10²⁰ cm² in the direction of the COSMOS field), with the power-law slope and N_H as free parameters. The best-fit N_H value and its 2σ error are found using the Cash (1979) statistic. For the remaining 20 optically dull AGNs, we estimate a less accurate N_H from their hardness ratio, HR = (H − S)/(H + S), following the relation between N_H and HR from Mainieri et al. (2007). Here, H is the counts in the hard 2–4.5 keV XMM-Newton band and S is the counts in the soft 0.5–2 keV XMM-Newton band.

Figure 6 shows the column density N_H with the [3.6 μm]_AB − [4.5 μm]_AB color. Black squares and upper limits show those optically dull AGNs with over 50 counts in their XMM-Newton or Chandra observation, while gray diamonds show those objects with N_H estimates from the hardness ratio only. Most (31) optically dull are relatively unobscured in their X-rays, with N_H < 10²² cm⁻², and at most only 2–3 are Compton-thick (N_H > 10²⁴ cm⁻²). The X-ray column densities are similar to those of Type 1 and Type 2 AGNs in COSMOS (Mainieri et al. 2007), so there is no X-ray evidence for additional obscuration in optically dull AGNs.

Zoom In Zoom Out Reset image size

The HST/ACS imaging in COSMOS allows for detailed studies of the host galaxies of the optically dull AGNs. We show postage stamps of the 46 objects with HST/ACS coverage in Figure 7. Immediately, it is evident that the optically dull AGNs reside in a wide variety of hosts (in contrast with Rigby et al. 2006), despite the fact that they have spectra consistent with old, red elliptical galaxies. A few hosts appear as isolated spheroids or ellipticals, while others have clumpy and dusty disks, and some are disturbed. Type 1 and Type 2 AGNs have similarly been shown to exist in a wide range of host galaxy morphologies (Jahnke et al. 2004; Sánchez et al. 2004; Gabor et al. 2009). Marking the spectroscopic aperture (10 × 54 IMACS slit or 30 diameter SDSS fiber) over each of the images, however, reveals a common thread: several of the optically dull AGNs appear to be have significant extranuclear light within the aperture. Both of the optically dull AGNs with SDSS spectroscopy in Figure 7 have bright elliptical hosts filling the fiber aperture. At least eight objects with IMACS spectroscopy have a nearby companion falling in the slit, while the hosts of at least eight others appear to have a bar or disk oriented along the slit. In all of these cases, the AGN optical emission is likely to be diluted by the continua of one or more normal galaxies. This scenario can explain the optically dull AGNs with normal f_X/f_O ratios, since the extranuclear host galaxy light would increase the total optical brightness.

Zoom In Zoom Out Reset image size

Rigby et al. (2006) used HST/ACS images in the Chandra Deep Field South (CDF-S) to show that optically dull AGNs have preferentially edge-on hosts compared to other AGNs, further suggesting that optically dull AGNs are optically obscured by extranuclear dust in their host galaxies. To test this hypothesis, we present axis ratios with redshift for 37 optically dull AGNs in Figure 8, along with a sample of 93 Type 2 AGN from Gabor et al. (2009) in gray. No Type 1 AGN are shown because the point sources are too bright for accurate host galaxy decompositions: in Gabor et al. (2009), 2/3 of Type 1 AGN hosts had unphysical best-fit parameters.

Zoom In Zoom Out Reset image size

We measure b/a, the ratio of minor to major axis, using the publicly available galaxy fitting software GALFIT (Peng et al. 2002) and following the procedures of Gabor et al. (2009). Sources in the AGN vicinity (<35 h⁻¹ kpc projected on the sky) are identified using the Source Extractor (Bertin & Arnouts 1996) and are either masked out of the image or simultaneously fit by GALFIT (if they are too close to the AGN for masking). We use the Source Extractor outputs based on isophotal profiles to generate initial guesses for the magnitude and shape parameters of the galaxy images in GALFIT. We fit each AGN image two separate times to explore different light distribution models. In one fit, we model the galaxy as a single Sérsic profile, and another uses a Sérsic profile plus a point source representing a nuclear point source (although over half of the optically dull AGNs have no strong nuclear point source; see below). Gabor et al. (2009) explored fits with additional components (e.g., a disk + Sérsic profile), but found that such fits typically give unphysical results because they are unstable for the S/N of the HST/ACS images. By constraining the fits in radius, magnitude, and shape, we prevent GALFIT from exploring wildly unphysical parameter space, but we flag as unacceptable any fits which run into the boundaries or yield strikingly unphysical results (for more details, see Gabor et al. 2009).

The GALFIT analysis yields good fits for 37 of the optically dull AGNs. Most (21) of the optically dull AGNs are best fit with a single Sérsic component and no nuclear point source. Of the remaining 16 with two-component fits, 13 have only marginal contributions from a point source, and fitting these 13 AGNs with Sérsic-only components does not affect their b/a values. Onlythree optically dull AGNs have significant nuclear point sources which would bias their b/a measurements to high (less elongated) values if not included in the fit. These nuclear point source contributions are consistent with the template fitting in Section 3.1, which showed that the optically dull AGNs have blue AGN contributions of 35% or less. The GALFIT axis ratios correlate strongly with those measured using Source Extractor, with a mean absolute difference of 0.11 in b/a. This suggests that our axis ratios are robust.

The optically dull AGNs and Type 2 AGNs in COSMOS have nearly identical ranges of axis ratio, with the optically dull AGN mean b/a = 0.56 ± 0.20 and the Type 2 AGN mean b/a = 0.56 ± 0.18. While our optically dull AGNs have consistent axis ratios to those of Rigby et al. (2006), our Type 2 AGNs do not show the face-on preference that Rigby et al. (2006) claim for their "optically active" AGN sample. Part of this difference comes from the differences in sample definitions: the six "optically active" AGNs of Rigby et al. (2006) include four broad-line Type 1 AGNs, while we compare to only Type 2 AGNs. In addition, the 45 "optically dull" objects of Rigby et al. (2006) included 36 (∼80%) AGNs with weakly detected narrow emission lines, while only 9/48 (∼20%) of our optically dull AGN sample includes AGNs with weakly detected narrow emission. Rigby et al. (2006) showed that weak-lined AGNs are quite different from Type 1 AGNs, while our Figure 8 shows that line-less optically dull AGNs have similar hosts to Type 2 AGNs. In Gabor et al. (2009), it was shown that morphological fits to Type 1 AGN hosts suffer from many systematic errors. In particular, a Type 1 host could have an incorrectly high b/a value, since even a slightly incorrect point source removal would leave a symmetric halo and a corresponding round residual. In any case, the fact that Type 2 and optically dull AGNs have similar axis ratios indicates that edge-on hosts are not causing the lack of narrow emission lines in optically dull AGNs.

At the redshifts of the optically dull AGNs in the sample, our spectroscopic slit generally includes nearly all of the host galaxy, and occasionally even includes a nearby companion. This is especially evident in Figure 7, where at least 10 host galaxies contaminate the spectroscopic aperture and at least eight others have a companion galaxy in the slit. One can imagine that many "optically normal" local Seyfert AGNs would appear "optically dull" if observed with spectroscopic apertures including extranuclear galaxy emission. Indeed, Moran et al. (2002) obtained integrated spectra for 18 local Seyfert 2 galaxies, and found that 11 (∼60%) of them would appear optically dull when observed in a 5'' × 1'' spectroscopic slit at z ≳ 0.5. Many of our optically dull AGNs may then be analogs to local Seyfert 2 AGNs. Dilution provides the simplest explanation for the four variable optically dull AGNs, all of which have a clear blue component in their optical photometry and (f_X/f_O) < 1 for the AGN fraction of the template fit. Dilution by a host galaxy might explain all 33/48 (70%) of the optically dull AGN with f_X/f_O ratios consistent with Type 1 and Type 2 AGN (that is, log(f_X/f_O) < 1). While only 18 show obvious evidence for extranuclear galaxy light in the slit, the other log(f_X/f_O) < 1 objects might be weak AGN with the emission lines diluted by a bright host. AGN activity is typically correlated with host luminosity (e.g., Hickox et al. 2009; Silverman et al. 2009), but there is a large scatter in the relation. Under the dilution hypothesis, some optically dull AGNs may represent the weak AGN / bright host tail of the relation.

However, dilution cannot explain all optically dull AGNs. Locally, 10%–20% of local AGNs are undiluted and remain optically dull (La Franca et al. 2002; Hornschemeier et al. 2005). And in COSMOS, 15 optically dull AGNs are optically under-luminous compared to their X-ray emission, with log(f_X/f_O)>1. Dilution by a host galaxy, on the other hand, would cause AGNs to become more optically luminous compared to their X-ray emission. Indeed, optically dull AGNs with log(f_X/f_O) ∼ 1 may also not fit the dilution paradigm, since presumably the additional host light would drive the optical flux of "normal" AGNs well below this cutoff. This suggests that 15/48 (∼30%) is a lower limit for the optically dull AGNs not explained by dilution.

The optically dull AGNs in COSMOS do not show signs of strong obscuration, with X-ray column densities similar to Type 2 AGNs and blue IRAC colors. Their host galaxies are not preferentially edge-on compared to the hosts of Type 2 AGNs, suggesting that obscuration by the host is not the cause of their missing narrow emission lines. With no evidence for obscuration, the undiluted optically dull AGNs must be intrinsically weak in their optical emission. AGNs with low accretion rates are expected to be optically underluminous, with very weak or missing emission lines, in just this fashion. In the next section we investigate the properties of the 15/48 (30%) optically dull AGNs which are not explained by obscuration or dilution to see if they fit the properties expected for low accretion rate AGNs.