INSIGHTS ON THE DUSTY TORUS AND NEUTRAL TORUS FROM OPTICAL AND X-RAY OBSCURATION IN A COMPLETE VOLUME LIMITED HARD X-RAY AGN SAMPLE

Our first criterion for selecting a sample, and the only astrophysical one for the active sample, is to define a set of local AGNs in a well-characterized way so that the selection effects are (as far as possible) understood. The largest optical or infrared surveys that select AGNs are often incomplete for nearby galaxies, and also tend to use AGN tracers that are anisotropic or impose a bias against star formation (by using large apertures to measure features that can be produced by both AGNs and star formation). For example, [O iii]$\lambda 5007$ Å, a typical optical line used in the selection of AGNs, is not isotropic: in comparison to [O iv]$\lambda 25.9\;\mu$m, which correlates well with 10–200 keV emission, it is underluminous in some Sy 2s—especially those that are Compton thick—by up to an order of magnitude compared to Sy 1s (Meléndez et al. 2008; Diamond-Stanic et al. 2009; Weaver et al. 2010). The same effect is seen in the 2–10 keV emission when compared to the [O iv] line. However, high ionization lines may not be an ideal way to select AGNs since they are not always observed: Koss et al. (2013) detected Ne v only in 2/3 of luminous infrared galaxies for which an AGN was confirmed by the Swift-BAT survey.

In contrast to emission line and infrared tracers, the very hard 14–195 keV band of the Swift-BAT survey measures direct emission from the AGNs rather than scattered or reprocessed emission, and is much less sensitive to obscuration in the line-of-sight than softer X-ray or optical wavelengths (selecting only against highly Compton thick AGNs). Indeed it is widely accepted as the least biased survey for AGNs with respect to host galaxy properties, and as such it has been well studied. There is a vast amount of ancillary data on the larger scale host galaxy properties (Winter et al. 2009, 2010; Koss et al. 2010, 2011; Meléndez et al. 2014; Mushotzky et al. 2014), as well as analysis of the bolometric corrections which enable one to make an estimate of the AGN luminosities (Vasudevan et al. 2010; Winter et al. 2012; Meléndez et al. 2014). Because it is an all-sky survey with roughly uniform sensitivity, we can select a complete, volume-limited sample of AGNs. Furthermore, the continuous nature of the survey means that more recent versions of the catalog (58 and 70 months, Baumgartner et al. 2013) average over variability during the last 5–6 years.

In addition to the single astrophysical criterion described above, we also impose two observational criteria to ensure homogeneous high resolution and observability. Our sample therefore consists of all 20 AGNs in the 58 months Swift-BAT catalog that meet the following three criteria: (i) 14–195 keV luminosities ${\rm log} {{L}_{14-195}}\gt 42.5$ (using redshift distance), (ii) redshift $z\lt 0.01$ (corresponding to a distance of $\lesssim 40$ Mpc), and (iii) observable from the VLT ($\delta \lt 15{}^\circ$) so that they tend to be in the southern sky. Note that the first two criteria are adjusted to intersect the flux limit of the 58 months catalog for 90% of the sky, that is $1.46\times {{10}^{-11}}$ erg s⁻¹ cm⁻². The selection, and its completeness, are graphically represented in Figure 1.

Zoom In Zoom Out Reset image size

The second criterion for our selection is that there should be a properly matched sample of inactive galaxies, so that it is possible to discern which features (stellar age, inflow rates, etc.) are related specifically to the AGN activity. We describe the selection of the inactive sample below. Although it has no part in the analysis presented in this paper in Sections 4–5, it does play a key role in many of the analyses which we will present in future papers.

The characteristics on which the inactive sample matching is based are: host galaxy morphology (Hubble type), inclination (axis ratio) and H-band luminosity (as a proxy for stellar mass). For the purposes of target selection, we have derived the inclination from the axis ratio without any compensation for finite thickness. This is not expected to yield a significant bias when comparing two galaxies with similar morphological type, and the inclinations of the host galaxies will be assessed more carefully in future analyses. The relation between H-band luminosity and stellar mass has been calibrated from a sample of similar galaxies for which the stellar mass has been derived using multi-band photometry (Koss et al. 2011). Figure 2 shows that the scatter in this relation is 0.2 dex.

Zoom In Zoom Out Reset image size

We have quantified the matching criteria above using a χ² metric with tolerances of $\pm 1$ for Hubble type, $\pm 15{}^\circ$ for inclination, and $\pm 0.3$ dex for H-band luminosity. In addition we require that the presence/absence of a bar is matched if possible. Finally, also where possible, we have selected inactive galaxies with redshifts less than or equal to those of their active pairs. All inactive galaxies were selected from the RC3 catalog (de Vaucouleurs et al. 1991), rejecting any listed in the Véron-Cetty and Véron (2010) catalog of known AGNs. While some may still host very low luminosity AGNs, the decisive factor is that (based on the lack of any obvious signature of BH acretion in optical, radio, and X-ray data) it is orders of magnitude weaker than in the X-ray selected sample.

The inactive galaxy selection is based on matching the characteristics of individual galaxies in the active sample. This is used as a robust way to ensure that the samples, as a whole, match; it is not specifically our intention to compare pairs of active and inactive galaxies. Part of the reason is that some inactive galaxies are a good match for several active galaxies, which means there is no unique pair matching. The final matched inactive sample contains 19 galaxies, which are listed in Table 2 together with the host properties used for the selection process.

Table 2.  Summary of Control Sample Properties

Name	Distance	log LH	Hubble	Axis	Bar
	(Mpc)	L$_{\odot }$	Stage	Ratio
(1)	(2)	(3)	(4)	(5)	(6)
NGC 718	23	9.89	1	0.87	AB
NGC 1079	19	9.91	0	0.60	AB
NGC 1315	21	9.47	−1	0.89	B
NGC 1947	19	10.07	−3	0.87	⋯
ESO 208-G021	17	10.88	−3	0.70	AB
NGC 2775	21	10.45	2	0.77	⋯
NGC 3175	14	9.84	1	0.26	AB
NGC 3351	11	10.07	3	0.93	B
ESO 093-G003	22	9.86	0.3	0.60	AB
NGC 3717	24	10.39	3	0.18	⋯
NGC 3749	42	10.40	1	0.25	⋯
NGC 4224	41	10.48	1	0.35	⋯
NGC 4254	15	10.22	5	0.87	⋯
NGC 4260	31	10.25	1	0.31	B
NGC 5037	35	10.30	1	0.32	⋯
NGC 5845	25	10.46	−4.6	0.63	⋯
NGC 5921	21	10.08	4	0.82	B
IC 4653	26	9.48	−0.5	0.63	B
NGC 7727	26	10.41	1	0.74	AB

Columns: (1) common name; (2) distances are, where possible, redshift independent (NED); (3) integrated H-band luminosity (as a proxy for stellar mass). (4) Hubble stage (NED); (5) axis ratio (NED); (6) B and AB denote presence of a bar or weak bar respectively (NED).

Download table as:  ASCIITypeset image

A graphical comparison of the main host galaxy properties is given in Figure 3. By design the distributions of these properties should be similar, as is apparent from the figure. This is confirmed for L_H, axis ratio, and Hubble stage, by Kolmogorov–Smirnov (KS) tests which indicate that the differences are not significant. Formally, the probabilities that the differences between the active and inactive samples could arise by chance exceed 20%. Both the active and inactive samples show the same distribution in H-band luminosity: the active sample has a mean of ${\rm log} {{L}_{{\rm H}}}[{{L}_{\odot }}]=10.3$ with a $1\sigma$ spread of 0.3 while the inactive sample has a mean of 10.2 and a spread of 0.4. Both samples cover the full range of axis ratio (or equivalently inclination), with about half the sample each side of 0.7, equivalent to 45° for a disk. And both samples cover a wide range of morphological types, with a clear preference for Hubble stage at $\sim 1$ corresponding to S0 and Sa "early disk" types. This distribution is consistent with the analysis presented by Koss et al. (2011), who found for a larger sample of Swift-BAT AGNs that only $\sim 10$% were in ellipticals, while $\sim 30$% had intermediate (S0) and $\sim 40$% early type spiral (Sa–Sb) host galaxies. Koss et al. (2011) noted that these fractions differ from the distribution of normal galaxies—16% ellipticals, 26% spirals. But the distribution of host types of the Swift-BAT AGNs are consistent with this once one takes into account the detection rate of Seyferts (rather than LINERs) reported by Ho (2008), which peaks for S0–Sb types. Davies et al. (2014) discussed the role of host type and environment in the context of the origin of the gas fuelling the AGN, and argued that this has an impact on what gas inflow mechanisms one may see in the circumnuclear region. This issue of morphology, including the presence or absence of a bar, will be revisited for our complete sample in a future work. The main point here is that the difference in distributions underlines the need to include morphology as a criterion for matching the inactive sample as we have done.

Zoom In Zoom Out Reset image size

The largest difference between the samples occurs for the distance distribution, which a KS test indicates may be marginally significant at a level of 2.5σ. This is a direct result of our requirement to select, whenever possible, inactive galaxies that are closer, rather than more distant, to their active pair. We note that the active and inactive samples have mean distances of 31 and 24 Mpc, respectively, and in both cases the $1\sigma$ spread is about 9 Mpc.

Much of the discussion below is set in the context of Merloni et al. (2014). As such, we follow these authors and adopt ${{N}_{{\rm H}}}\gt {{10}^{21.5}}$ cm⁻² as the criterion for defining an AGN to be X-ray absorbed. For typical gas-to-dust ratios this is equivalent to an optical screen extinction of A$_{V}\sim 2$ mag, sufficient to mildy obscure the line of sight to the BLR at visible wavelengths.¹⁴ In our volume limited 14–195 keV sample, we find 50%–60% of the AGNs are optically obscured while $\sim 60$% are X-ray absorbed. Most of the obscured AGNs are both optically obscured and X-ray absorbed; none are just optically obscured; only three are just X-ray absorbed. These exceptions are "minor" since they are all close to the boundary between the regimes. The X-ray absorbed AGN NGC 1365 is optically classified as Sy 1.8 and, although in many studies is taken to be a Sy 2, here we adopt a strict definition and so classify it as Sy 1. Similarly MCG-05-23-016 is X-ray absorbed and has only weak broad Hα, hence its previous classification as a Sy 2 which we have revised to Sy 1.9. The Sy 1.8 NGC 2992 has a column only marginally above our X-ray absorption threshold and so is on the borderline in both classifications. It would be classified as X-ray unabsorbed according to other definitions that set the threshold at 10²² cm⁻² (Panessa & Bassani 2002). As such, there could be two more optically obscured and one less X-ray absorbed AGN than found using the strict definitions we apply in Figure 6. This is reflected by the ranges shown in Figure 7.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The joint optical/X-ray classifications are summarized in Figure 6 where we adopt the terminology of Merloni et al. (2014) in which the first/second digit of the classification denotes the optical/X-ray type. This leads to the following types:

• optically unobscured and X-ray unabsorbed;
• optically classified as Sy 2 and X-ray absorbed;
• optically classified as Sy 2 but X-ray unabsorbed;
• optically unobscured but X-ray absorbed.

Thus, with reference to the curves in Figure 7 adapted from Merloni et al. (2014) to the 14–195 keV luminosity scale, we also have:

• all X-ray absorbed AGNs (regardless of their optical classification);
• all AGNs optically classified as Sy 2 (independent of whether they are X-ray absorbed).

Comparing the AGNs in our sample to the left edge of Figure 7 immediately highlights a discrepancy. Above, we noted that 50%–60% of our complete sample are optically obscured and classed as Sy 2s, a fraction similar to that estimated by Lawrence & Elvis (2010). However, Merloni et al. (2014) class $\sim 90$% of their AGNs in an overlapping luminosity range ${\rm log} {{L}_{14-195}}\sim 43$–43.5 as Sy 2s. The difference is due to the large fraction of AGNs classified by these authors as type 21, that is X-ray unabsorbed Sy 2 galaxies (Pappa et al. 2001; Panessa & Bassani 2002; Brightman & Nandra 2008; Bianchi et al. 2012).

X-ray unabsorbed Sy 2s are believed to be galaxies with a direct view to the AGNs but in which there is no BLR (such objects are also known as pure or true Sy 2s). This view is supported by, for example, the six objects identified by Hawkins (2004) as having optical spectra typical of Sy 2s (HβFWHM $\lt$ 1000 km s⁻¹ and [O iii]/Hβ$\;\gt \;$ 3) but large amplitude variations typical of Sy 1s. Subsequent X-ray observations of three by Gliozzi et al. (2007) confirmed that these do not have significant absorption in the 0.3–8 keV band. These authors also showed that these AGNs are relatively luminous (intrinsic ${\rm log} {{L}_{0.5-8}}\gt 43.2$). Thus, despite their rather low Eddington ratios ${{L}_{{\rm bol}}}/{{L}_{{\rm Edd}}}\lt 0.01$ (Gliozzi et al. 2007), the absence of a BLR cannot be explained by low luminosities and/or accretion rates either being unable to sustain a disk wind (Nicastro 2000; Elitzur & Ho 2009) or meaning that high dispersion prevents BLR clouds from surviving (Laor 2003). At present, the reason that they may not have a BLR is still open.

These objects may be related to the Sy 2s for which spectropolarimetry shows no evidence of a hidden BLR. There have been many studies of polarized emission from hidden BLRs. The largest indicate that Sy 2s in which a hidden BLR has not been detected have less luminous AGNs than other Sy 2s but are not more optically obscured (Tran 2001, 2003; Gu & Huang 2002)—qualitatively, but not necessarily quantitatively, consistent with the idea that BLR may not form in low luminosity AGNs. It also suggests that a BLR is completely absent in at least 50% of Sy 2s. However, this fraction is difficult to put in perspective because of complications due to the impact of detection limits as well as the expected scattering efficiency. The latter issue was addressed by Heisler et al. (1997) who showed that the ability to detect polarized emission was related to the far-infrared colors in a way that suggested the scattering particles were in the throat of the torus, and so even they would be hidden at high inclinations. Following on from work of Ramos Almeida et al. (2011) which shows that the intrinsic properties (rather than just the inclination) of the tori in Sy 1s and Sy 2s are different, Ichikawa et al. (2015) have fitted torus models to Sy 2s with and without polarized emission. Their results again indicate that whether a hidden BLR is seen depends on the location of the scattering material in the throat of the torus; but that rather than inclination, it is the opening angle and covering factor of the torus that determines this difference. As such, the fraction of Sy 2s in which polarized scattered BLR emission is not observed could be very different from the fraction without a hidden BLR. Since, based on these models, the polarized BLR emission may in some cases be only relatively lightly obscured, deep and/or near-infrared spectropolarimetry may shed further light on this issue. NGC 7172 and NGC 7582 (see Table 1) are cases in point here: Bian & Gu (2007) and Marinucci et al. (2012) both report that no polarized broad Hα has been detected but Smajić et al. (2012) have seen broad Paα and Brγ directly in NGC 7172 and Reunanen et al. (2003) broad Brγ in NGC 7582.

Despite the uncertainties, the evidence that at least some relatively luminous AGNs do not have a BLR is convincing. The key question for this paper is how common such objects are, since this can have a significant impact on the total fraction of AGNs optically classified as Sy 2.

In their analysis, Risaliti et al. (1999) found that about 4% of Sy 2 galaxies had ${{N}_{{\rm H}}}\lt {{10}^{22}}$ cm⁻². However, Panessa & Bassani (2002) presented a sample of 17 objects which had been classified as Sy 2 in the literature, and which had similarly low column densities. They suggested that X-ray unabsorbed Sy 2s may be much more common, being 10%–30% of all Sy 2s. More recently, other authors have confirmed that such objects do exist, but without a consensus on how common they are, because many candidates are ruled out on closer examination. For example, Brightman & Nandra (2008) found that of eight candidates, four were underluminous in the 2–10 keV bands suggesting that they were Compton thick—with the unabsorbed softer X-rays originating in a scattered continuum or from host galaxy contamination. Similarly, in an examination of eight candidates—including some from the Panessa & Bassani (2002) sample—Bianchi et al. (2012) ruled out 4 for a variety of reasons, and confirmed only 3 as unabsorbed Sy 2s. Using a variety of metrics for a critical assessment of 24 objects that were previously claimed to be unabsorbed Sy 2s (again including many from Panessa & Bassani 2002), Shi et al. (2010) could confirm only two as genuine, with broad emission lines 2–3 orders of magnitude fainter than typical Sy 1s—although several more candidates may have anomalously weak broad lines, this could not be confirmed with existing data. These authors concluded that 1% or less of Sy 2s are X-ray unabsorbed.

In contrast, a much higher fraction of such sources—30% of AGNs at ${\rm log} {{L}_{14-195}}\sim 43$—appears in the Merloni et al. (2014) sample. Whether this indicates that there is a large population of X-ray unabsorbed Sy 2s or is due to a classification bias was discussed in depth by these authors. With reference to stacked optical spectra and spectral energy distributions, they suggested that many of the lower luminosity AGNs may have been incorrectly photometrically classified as Sy 2 due to the low contrast of the AGNs against the relatively bright host galaxy. On the other hand, spectroscopic data are sensitive to features associated with Sy 1s, such as broad lines, even when they are weak. Such signatures would not be evident in broad-band integrated photometry. This difference is reflected in their data: while 77% of the AGNs with only integrated broad-band photometric data are classified as Sy 2, only 56% of AGNs with spectroscopic data are classified as Sy 2s. The lower total fraction of Sy 2s based on spectroscopic classifications implies in turn a much lower fraction of X-ray unabsorbed Sy 2s.

Figure 6 shows that in our volume limited sample of Swift-BAT AGNs, there are no sources classified as X-ray unabsorbed Sy 2, implying these sources are rare. In this context, a simulation of the AGN population can provide valuable insight, since we know the Sy 2 fractions in both flux limited and volume limited samples. We describe this simulation in Section 4.3, aiming to answer the question of which curve from Merloni et al. (2014) in Figure 7 properly traces the fraction of Sy 2s: is it types 22 + 21 or just type 22?

Our aim here is to simulate parent populations of AGNs with different prescriptions for the intrinsic fraction of Sy 2s. After applying observational limits, we can then use constraints from the flux-limited sample of Winter et al. (2009) and our own volume-limited sample to discriminate between the different prescriptions, giving us a handle on the true fraction of Sy 2s.

We begin by constructing a population of AGNs using a 2–10 keV luminosity function from Aird et al. (2010) suitable for low redshift:

$$\begin{eqnarray*}&&\phi (L_{2-10}^{\operatorname{int}})\ \propto \ {{[{{(L_{2-10}^{\operatorname{int}}/{{L}_{0}})}^{0.70}}+{{(L_{2-10}^{\operatorname{int}}/{{L}_{0}})}^{3.14}}]}^{-1}}\end{eqnarray*}$$

where ${\rm log} {{L}_{0}}=44.96$. We have then randomly assigned each AGN to be a Sy 1 or Sy 2 according to the obscured fraction as a function of luminosity. We use two functions, which are discussed below. We allow the AGNs to be observed by calculating ${{L}_{14-195}}$ from $L_{2-10}^{\operatorname{int}}$, assigning them random locations in a local volume, deriving their fluxes, and assessing whether they would be detected in the Swift-BAT survey. In doing this, we have not included the effects of X-ray absorption in the 14–195 keV band since, as we have seen, it has little impact at such high energies for the absorbing columns expected. We use a limit appropriate for the 9 months survey in order to select a flux limited sample matching that used by Winter et al. (2009); and a limit appropriate for the 58 months survey to match our volume limited selection outlined in this paper.

The left panels in Figure 8 show the two obscuration functions used in our simulations. The dotted line in the upper left panel is taken directly from Merloni et al. (2014) and was designed to follow the curve for types 22+21; the dashed line in the lower left panel is our modification of this function to match the type 22 data, those AGNs that are both optically obscured and X-ray absorbed. The results of our simulations using these functions are also given in Figure 8. The center panels show the Sy 1 and Sy 2 populations generated—which can be compared to the real data in Figure 1. The right panels show their resulting distributions as a function of luminosity, which can be compared to Figure 9. In these panels, the sharp decrease in the number of Sy 2s around ${\rm log} {{L}_{14-195}}\sim 44$ simply reflects the rapid change in the fraction of obscured AGNs at this luminosity. The flux limit means that low luminosity AGNs are not detected at larger distance, and hence biasses the apparent fraction of obscured AGNs at ${\rm log} {{L}_{14-195}}\lt 43.5$. In contrast, because our sample is complete to ${\rm log} {{L}_{14-195}}=42.5$, it provides an unbiassed indication of the obscured fraction at these luminosities.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

It is immediately clear that the lower panels better match the observed distribution of Sy 1 and Sy 2 sources reported in Winter et al. (2009). Similarly, the lower panels correspond to finding a fraction of 54% of Sy 2s (in contrast to 87% for the upper panels) in a complete volume limited sample equivalent to that described in Section 2, again providing a good match to the fraction actually measured in the Swift-BAT sample.

Our conclusion is that, over the luminosity range $43{\mkern 1mu} \lt {\rm log} {{L}_{14-195}}\lt 44.5$ where the "type 22" and "types 22+21" curves differ in Figure 7, the true fraction of Sy 2s is given by the "type 22" curve. This confirms that X-ray unabsorbed Sy 2s are rare, at most a few percent of the Sy 2 population; and that the high number of type 21 objects in the Merloni et al. (2014) sample is due to a bias in photometric classification. It also implies that, in low luminosity systems, optical obscuration and X-ray absorption are usually found together—and hence most likely originate in the same shared obscuring structure.

At luminosities below ${\rm log} {{L}_{14-195}}\sim 44$, the optically obscured AGN fraction of $\sim 60$% is similar to the X-ray absorbed fraction, and is independent of luminosity. Since the dust sublimation radius R_dust increases with luminosity, this independence must reflect the geometrical properties of the torus which we can constrain as follows. A luminosity of ${\rm log} {{L}_{14-195}}=44$ implies ${{R}_{{\rm dust},44}}\sim 0.2$ pc for graphite grains of size 0.05 μm (Netzer 2014). An obscured fraction of 60% implies $H/R\sim 0.75$—a half opening angle of $\sim 53{}^\circ$—and hence ${{H}_{{\rm dust},44}}\sim 0.15$ pc. And that the obscured fraction is constant with luminosity implies that $H/R$ cannot increase at radii smaller than ${{R}_{{\rm dust},44}}$ (and the rapid drop in the optically obscured fraction at higher luminosities indicates that $H/R$ also decreases at larger radii out to a few times ${{R}_{{\rm dust},44}}$). This is illustrated in the left side of Figure 10. There are two implications. First, it is the same geometrically thick structure that causes the optical obscuration and the X-ray absorption, and this structure is synonymous with the dusty torus. The implied half-opening angle of $\sim 53{}^\circ$ is at the upper end of the 30°–50° range derived from 3D models of the ionization cone kinematics (Müller-Sánchez et al. 2011; Fischer et al. 2013), a marginal consistency that could reflect uncertainties in the derivations of these numbers or may have a more physical origin. Second, the inner wall of the dusty torus—which defines the optically obscured and X-ray absorbed fractions—has universal properties at ${\rm log} {{L}_{14-195}}\lt 44$: the average torus properties are an inner radius of $\sim 0.2$ pc and a scale height of $\sim 0.15$ pc, independent of luminosity. Here we make two notes. The first is that this is a constraint on the geometrically thick nature of the torus, and any dust that exists at smaller radii will not contribute to the obscured fraction because it is not driving the maximum value of $H/R$ (Netzer & Laor 1993; Netzer 2014). The second is that at luminosities below that of our sample, one would expect the torus properties to change if there is a luminosity threshold below which the torus disappears (Elitzur & Shlosman 2006).

Zoom In Zoom Out Reset image size

While none of the AGNs in our sample are this luminous, the data of Merloni et al. (2014), as reproduced in Figure 7, show a dramatic difference in the Sy 2 and X-ray absorbed fractions. At these luminosities the Sy 2 fraction is not expected to suffer a significant bias due to mis-classification simply because the AGNs are brighter with respect to the host galaxy; indeed, the fraction of AGNs classified as type 21 by Merloni et al. (2014) does decrease dramatically. As such, one can consider the Sy 2 fraction as the optically obscured fraction. This fraction drops rapidly to $\sim 20$% following the ${{L}^{0.5}}$ dependence of the dust sublimation radius, as understood in the "receding torus" concept (Lawrence 1991; Simpson 1998, 2005; Lusso et al. 2014), without requiring any change in the scale height of the torus. In contrast, the X-ray absorbed fraction remains nearly constant, falling only from 60% to 45%.

The implicit population of X-ray absorbed Sy 1s were discussed by Merloni et al. (2014), who argued that the X-ray absorption is not occuring on large galaxy-wide scales and suggested instead that it is more plausibly due to dust-free gas within the BLR. Below, we argue that this is indeed likely, since BLR clouds are expected to contain a significant column of neutral gas; and associate the absorption with a "neutral torus" that co-exists with the BLR, and which has an impact on how the "receding torus" concept should be understood.

However, first it is important to clarify whether warm absorbers, which have been observed in about 50% of Sy 1s (Komossa 1999; Blustin et al. 2005; Winter et al. 2012; Laha et al. 2014), may be responsible for the X-ray absorbed Sy 1s here. Warm absorbers imprint their signature on an X-ray spectrum through absorption lines and are most prominent in soft X-ray bands. They are classically identified by the O vii and O viii absorption edges at 0.74 and 0.87 keV, respectively, rather than through their impact on the spectral shape (Komossa 1999; Winter et al. 2012; Laha et al. 2014). Warm absorbers have a smaller impact than neutral gas on the global shape of the spectrum, as evident from the models of Page et al. (2011) whose fits to 0.2–10 keV QSO spectra show that a warm absorber requires an order of magnitude more gas than a neutral absorber to reproduce the same spectral shape. The median ionized column derived by Winter et al. (2012) for the Swift-BAT AGNs, as well as by Laha et al. (2014) on a different sample, is ∼10²¹ cm⁻², with columns an order of magnitude lower for AGNs without strong O vii and O viii detections. While significant, this is less than $\sim {{10}^{21.5}}$ cm⁻² threshold adopted here for defining X-ray absorbed systems. Coupled with the fact that Merloni et al. (2014) used either the full 0.5–10 keV band or the 0.5–2 keV versus 2–10 keV hardness ratio to derive the absorbing column from its effect across the whole band rather than in specific features, this strongly argues that the absorptions they derived are due to neutral rather than ionized gas.

The BLR provides an ample source of neutral gas that could be responsible for the derived absorption. Photoionization models by Netzer (2013) show that for a typical BLR ionization parameter of $U\sim 0.01$ (Leighly & Casebeer 2007; Negrete et al. 2013) and a density of 10¹⁰ cm⁻³ the temperature drops below 10⁴ K at a column of ∼10²¹ cm⁻² and the fractional abundance of H i increases dramatically. Thus, for clouds with a column density of 10²³ cm⁻², as typically expected, the majority of gas is neutral rather than ionized. This has been used to explain the cause of Sy 1s which change state on short timescales, in terms of a BLR cloud passing across the line of sight to the central engine (see Risaliti et al. 2010 and Torricelli-Ciamponi et al. 2014, and references therein). In this interpretation, the clouds have cores with column densities of at least a few ×10²³ cm⁻² that cause the neutral absorption, and ionized tails that are the origin of the warm absorption (Risaliti et al. 2009, 2011).

Since the BLR clouds contain abundant neutral gas, it is natural to associate them with the X-ray absorbed Sy 1s. The remarkable constancy of the fraction of X-ray absorbed AGNs across the luminosity range in Figure 7 suggests that the geometry of the cold gas absorber also changes rather little: both the inner edge and scale height of the thick gas distribution remain similar to those at low luminosities. The implication is that there is a "neutral torus" that co-exists with the BLR, and extends out to the start of the dusty molecular torus as illustrated in the right-hand side of Figure 10. In this context, we note that Minezaki & Matsushita (2015) and Gandhi et al. (2015) have independently suggested the core of the fluorescent Fe Kα line may originate at radii extending from the BLR out to the dusty torus, implying the presence of gas at intermediate radii. There is no clear boundary between the BLR and the neutral torus since it is the same clouds that contribute to both. The BLR emission will follow a radial dependence associated with the ionization parameter. Since $U\propto {{L}_{{\rm AGN}}}/({{R}^{2}}\ {{n}_{{\rm H}}})$, the rapid decrease of U with radius R may be largely mitigated if the cloud density n_H also decreases with R. Thus the BLR could extend far into the neutral torus. How far it extends is given by the scaling of the BLR size with AGN luminosity (Kaspi et al. 2000). Using the most recent measurement of that relation (Bentz et al. 2013), and adopting a ratio ${{L}_{{\rm AGN}}}/\lambda {{L}_{5100}}\sim 15$ (Grupe et al. 2004), we estimate the characteristic radius of the BLR to be a factor of a few less than the dust sublimation radius. However, it is also possible that the BLR emission could still occur as far as the dust sublimation radius.

In the context above, the idea of the "receding torus," in which the location of the inner wall of the torus is set by the dust sublimation radius, can be misleading. We argue that the inner dust boundary and the inner gas boundary of the torus should be considered separately. The discussion above suggests that the location of the inner gas boundary of the geometrically thick structure (causing the X-ray absorption), is roughly independent of AGN luminosity in the range $43\lt {\rm log} {{L}_{14-195}}\lt 45$. And it is within this geometrically thick gas structure that the inner dust boundary (i.e., the "receding torus") simply represents the location at which the predominant gas phase changes from ionized/neutral to dusty/molecular. The thick gas structure itself remains nearly unchanged. This means that at low luminosities, the thick gas structure is synonymous with the standard dusty torus; but at high luminosities, the thick gas structure has an outer part that is the dusty torus, as well as an inner part that contributes to the BLR emission but also acts as a neutral dust-free torus.

Intriguingly, Winter et al. (2012) found that the strength of the O vii and O viii edges tracing the warm absorbers in Swift-BAT AGNs are correlated with the neutral column density—suggesting that material in the neutral torus is associated with a more-or-less proportional amount of ionized material. This can be put in the context of AGNs showing rapid variations in absorbing column, in which neutral clouds have an ionized outflowing tail (Risaliti et al. 2009, 2011)—an interpretation that applies to individual clouds in the BLR. Given that warm absorbers are outflowing (Blustin et al. 2005; Laha et al. 2014), and that the distance of the warm absorbers from the AGN is typically larger than the BLR but within a factor of a few of the dust sublimation radius (Blustin et al. 2005), one might speculate that the ionized material seen by Winter et al. (2012) is associated with the neutral torus and outflowing from the clouds as they are ablated.