BAT AGN Spectroscopic Survey-XXIII. A New Mid-Infrared Diagnostic for Absorption in Active Galactic Nuclei
Ryan W. Pfeifle, Claudio Ricci, Peter G. Boorman, Marko Stalevski, Daniel Asmus, Benny Trakhtenbrot, Michael J. Koss, Daniel Stern, Federica Ricci, Shobita Satyapal, Kohei Ichikawa, David J. Rosario, Turgay Caglar, Ezequiel Treister, Meredith Powell, Kyuseok Oh, C. Megan Urry, Fiona Harrison
DDraft version February 9, 2021
Typeset using L A TEX twocolumn style in AASTeX62
BAT AGN Spectroscopic Survey-XXIII. A New Mid-Infrared Diagnostic for Absorption in Active Galactic Nuclei
Ryan W. Pfeifle, Claudio Ricci,
2, 3, 1
Peter G. Boorman,
4, 5
Marko Stalevski,
6, 7
Daniel Asmus, Benny Trakhtenbrot, Michael J. Koss, Daniel Stern, Federica Ricci, Shobita Satyapal, Kohei Ichikawa,
13, 14
David J. Rosario, Turgay Caglar, Ezequiel Treister, Meredith Powell, Kyuseok Oh, C. Megan Urry, and Fiona Harrison Department of Physics & Astronomy, George Mason University, 4400 University Drive, MSN 3F3, Fairfax, VA 22030, USA Núcleo de Astronomía de la Facultad de Ingeniería, Universidad Diego Portales, Av. Ejército Libertador 441, Santiago, Chile Kavli Institute for Astronomy and Astrophysics, Peking University, Beijing 100871, China Astronomical Institute, Academy of Sciences, Bo˘cní II 1401, CZ-14131 Prague, Czech Republic Department of Physics & Astronomy, Faculty of Physical Sciences and Engineering, University of Southampton, Southampton SO171BJ, UK Astronomical Observatory, Volgina 7, 11060 Belgrade, Serbia Sterrenkundig Observatorium, Universiteit Gent, Krijgslaan 281-S9, Gent, 9000, Belgium Department of Physics & Astronomy, University of Southampton, SO17 1BJ, Southampton, Hampshire, United Kingdom School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel Eureka Scientific, 2452 Delmer Street Suite 100, Oakland, CA 94602-3017, USA Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Mail Stop 169-221, Pasadena, CA 91109, USA Instituto de Astrofísica, Facultad de Física, Pontificia Universidad Catolica de Chile, Casilla 306, Santiago 22, Chile Frontier Research Institute for Interdisciplinary Sciences, Tohoku University, Sendai 980-8578, Japan Astronomical Institute, Tohoku University, Aramaki, Aoba-ku, Sendai, Miyagi 980-8578, Japan Centre for Extragalactic Astronomy, Department of Physics, Durham University, South Road, DH1 3LE Durham, UK Leiden Observatory, PO Box 9513, 2300 RA Leiden, The Netherlands Instituto de Astrofísica, Facultad de Física, Pontificia Universidad Católica de Chile, Casilla 306, Santiago 22, Chile Department of Physics, Yale University, 217 Prospect St, New Haven, CT 06511, USA Korea Astronomy and Space Science Institute, Daedeokdae-ro 776, Yuseong-gu, Daejeon 34055, Republic of Korea Department of Physics and Yale Center for Astronomy and Astrophysics, Yale University, 217 Prospect St, New Haven, CT 06511,USA Cahill Center for Astronomy and Astrophysics, California Institute of Technology, Pasadena, CA 91125, USA
ABSTRACTIn this study, we use the
Swift /BAT AGN sample, which has received extensive multiwavelengthfollow-up analysis as a result of the BAT AGN Spectroscopic Survey (BASS), to develop a diagnosticfor nuclear obscuration by examining the relationship between the line-of-sight column densities( N H ), the 2–10 keV-to- µ m luminosity ratio, and WISE mid-infrared colors. We demonstratethat heavily obscured AGNs tend to exhibit both preferentially “redder” mid-infrared colors andlower values of L X , Obs . / L µ m than less obscured AGNs, and we derive expressions relating N H tothe L X , Obs . / L µ m and L µ m / L . µ m luminosity ratios as well as develop diagnostic criteria usingthese ratios. Our diagnostic regions yield samples that are (cid:38) % complete and (cid:38) % pure forAGNs with log( N H / cm − ) ≥ , as well as (cid:38) % pure for AGNs with log( N H / cm − ) (cid:38) . . Wefind that these diagnostics cannot be used to differentiate between optically star forming galaxiesand active galaxies. Further, mid-IR contributions from host galaxies that dominate the observed µ m emission can lead to larger apparent X-ray deficits and redder mid-IR colors than the AGNswould intrinsically exhibit, though this effect helps to better separate less obscured and moreobscured AGNs. Finally, we test our diagnostics on two catalogs of AGNs and infrared galaxies,including the XMM-Newton
XXL-N field, and we identify several known Compton-thick AGNs aswell as a handful of candidate heavily obscured AGNs based upon our proposed obscuration diagnostics. INTRODUCTIONKnown to reside at the centers of most galaxies (e.g.Ferrarese & Merritt 2000; Kormendy & Ho 2013), super- massive black holes (SMBHs) grow and evolve throughperiods of activity characterized by the accretion of largequantities of gas. Classically, these active galactic nuclei a r X i v : . [ a s t r o - ph . GA ] F e b R. W. Pfeifle et al. (AGNs) are categorized based upon the characteristicsof their optical spectroscopic emission lines, where theapparent differences between AGNs may be reconciledthrough a unification scheme involving a dusty obscur-ing torus (e.g. Antonucci 1993; Urry & Padovani 1995;Netzer 2015; Ramos Almeida & Ricci 2017), for whichdifferent inclination angles of the torus correspond tothe observation of different AGN classes.One ubiquitous observational signature of accretiononto SMBHs is X-ray emission, produced very close tothe accretion disk (Fabian et al. 2009) due to inverseCompton scattering of optical and ultraviolet (UV) pho-tons from the accretion disk by hot electrons in thecorona (Haardt & Maraschi 1991, 1993). In the X-rayband, the line-of-sight gas column density, N H , is largelytransparent to the 2–10 keV X-ray flux even up to col-umn densities of a few times cm − , however, signif-icant attenuation and reprocessing of the 2–10 keV X-ray emission does occur for Compton-thick (CT) AGNs,which possess gas column densities of (cid:38) cm − (e.g.Lansbury et al. 2014; Ricci et al. 2015; Bauer et al. 2015;Puccetti et al. 2016; LaMassa et al. 2019; Toba et al.2020). CT AGNs even at low redshift have thus provento be very difficult to find and characterize (Alexander& Hickox 2012) using lower energy X-ray observatoriessuch as Chandra and
XMM-Newton , since the X-ray fluxbelow 10 keV suffers significant photoelectric absorptionand Compton scattering. This prevents the detection ofsome sources, and even those detected have fewer ob-served photons, which reduces the accuracy of spectralmodeling. Important spectral signatures used to charac-terize the nature of AGNs can be missed without higherenergy X-ray observations (Lansbury et al. 2015; Kosset al. 2016; Ricci et al. 2017a).CT AGNs are particularly important among the gen-eral AGN population, as large fractions of CT AGNs arerequired to reproduce the observed Cosmic X-ray Back-ground (CXB, Gilli et al. 2007; Buchner et al. 2015;Ananna et al. 2019). CT AGNs likely represent a sig-nificant fraction of the total intrinsic AGN populationin the local Universe ( ∼ ∼ ± % and ± % of thetotal intrinsic AGN population up to both z ∼ . and z ∼ . , respectively, though many previous synthesismodels predicted lower intrinsic fractions of CT AGNsthan this, including but not limited to Gilli et al. (2007);Treister et al. (2009); Akylas et al. (2012); Ueda et al.(2014). Furthermore, questions remain regarding the ex-act nature of the obscuring structure and how it relatesto, for example, the host environment: some CT AGNs could represent the evolutionary phase predicted to oc-cur as a result of galaxy mergers (e.g. Hopkins et al.2008; Kocevski et al. 2015; Ricci et al. 2017b; Blechaet al. 2018). Identifying further cases of CT AGNs iscrucial for providing a full census of accreting SMBHs,placing constraints on the CXB, as well as placing con-straints on evolutionary and unification models.UV radiation from the accretion disk is also repro-cessed by the obscuring dusty torus, wherein the radia-tion is scattered and absorbed by the dust grains and re-emitted thermally with a peak usually at mid-infrared(mid-IR) wavelengths. While the classically acceptedorigin of the mid-IR emission is the dusty torus itself,recent high angular resolution infrared observations ofAGNs suggest that the mid-IR emission is in fact dom-inated by a dusty polar outflow rather than the torusitself (e.g. Hönig et al. 2012, 2013; Tristram et al. 2014;Asmus et al. 2016; Hönig & Kishimoto 2017; Stalevskiet al. 2018; Hönig 2019). Other recent studies (Baron &Netzer 2019a,b) have actually attributed mid-IR emis-sion to dusty outflows located on the order of tens tohundreds of parsecs from the centers of the galaxies.A correlation between the intrinsic (unabsorbed) hardX-ray 2–10 keV luminosity ( L X ) and mid-IR luminosity( L MIR ) of AGNs was reported as early as Elvis et al.(1978). Universally, studies find no difference ( < . dex) in the ratios of L X to L MIR between Type 1 andType 2 AGNs (Lutz et al. 2004; Levenson et al. 2009;Gandhi et al. 2009; Ichikawa et al. 2012; Mateos et al.2015; Asmus et al. 2015) and this ratio is also insensi-tive to the neutral gas column density along the line-of-sight (Gandhi et al. 2009; Mateos et al. 2015; Asmuset al. 2015), even in the case of CT AGNs after cor-recting the X-rays for absorption (Gandhi et al. 2009).Many previous studies have also pointed out that thisrelation can serve as a useful tool to select obscured,particularly CT, AGNs, because CT AGNs tend to ex-hibit severe deficits in their absorbed X-ray emissionwhen compared to their mid-IR emission, and thus CTAGNs fall significantly off of the L X -to- L MIR relation(Alexander et al. 2008; Goulding et al. 2011; Georgan-topoulos et al. 2011; Rovilos et al. 2014; Asmus et al.2015). Moreover, Asmus et al. (2015) also demonstratedthat the ratio of L X / L MIR can be used to predict col-umn densities for significantly obscured objects, derivingan equation that relates L X / L MIR to log( N H / cm − ) forlog( N H / cm − ) > . .It is important to gather large samples of powerfulAGNs across a range of column densities to test theutility of the X-ray to mid-IR relation as a tracer of nu-clear obscuration. Hard X-ray selection provides one ofthe least biased methods of identifying powerful AGNs, ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN < cm − (see Fig. 1 from Ricci et al. 2015). The Swift /BAT ultra-hard X-ray (14–195 keV) all-sky survey has dramaticallyincreased the number of known hard X-ray extragalacticsources (Baumgartner et al. 2013; Oh et al. 2018), andhas therefore been the focus of a large multiwavelengthfollow-up campaign (the BAT AGN Spectroscopic Sur-vey, or BASS ) designed to characterize the most pow-erful AGNs in the local Universe (Koss et al. 2017; Ricciet al. 2017a; Lamperti et al. 2017). A second release ofoptical spectroscopy (BASS DR2) will also soon be pub-licly available (Koss et al., in prep; Oh et al., in prep).Ricci et al. (2017a) presented a detailed X-ray analy-sis of 838 ultra hard X-ray detected Swift /BAT AGNs,providing constraints on column densities ( N H ) as wellas the absorbed (observed) and unabsorbed 2–10 keVluminosities, while Ichikawa et al. (2017) provided mid-IR to far-infrared photometric data and correspondingIR luminosities for the 604 mid-IR-detected Swift /BATAGNs. These two catalogs provide precisely the infor-mation needed to test the relationship between the ab-sorbed hard X-ray and mid-IR emission with respect tothe line-of-sight obscuration in AGNs.In this paper, we present a new diagnostic for absorp-tion in AGNs which combines the power of the known L X / L µ m correlation with WISE mid-IR colors. Usingmultiwavelength catalogs available for the
Swift /BAThard X-ray-selected sample of AGNs, we show this di-agnostic reliably identifies the most obscured AGNs, atleast for nearby X-ray bright AGN. Our proposed diag-nostics could prove valuable in the search for obscuredAGNs in the ongoing eROSITA survey (Predehl et al.2010; Merloni et al. 2012). In Section 2 we describe oursample. In Section 3 we describe the analysis of the sam-ple and propose our new absorption diagnostics as wellas develop expressions that constrain column densities.In Section 4 we explore the emission ratios of optically-selected star forming galaxies, we compare our diagnos-tics for obscuration to other recent studies, and we applyour diagnostics to the
XMM-Newton
XXL North field(Pierre et al. 2016, 2017). In Section 5 we detail ourconclusions. Throughout this manuscript we assume thefollowing cosmological values: H = 70 km s − Mpc − , Ω M = 0 . , Ω Λ = 0 . . All luminosities quoted in thiswork are given in units of erg s − . SAMPLE CONSTRUCTIONWe selected our sample from the 70-month
Swift /BATX-ray properties catalog (Ricci et al. 2017a) which de- tails the broadband 0.3–150 keV X-ray spectral proper-ties of the 838 AGNs detected in the ultra hard X-ray14–195 keV band by Swift /BAT and reported in the 70-month source catalog (Baumgartner et al. 2013). Wematched this catalog to the 70-month
Swift /BAT in-frared catalog of Ichikawa et al. (2017), which providesthe complete near-infrared to far-infrared photometryfor 604
Swift /BAT non-beamed AGNs at high Galacticlatitudes (|b|>10 ◦ ). We refer the reader to Ricci et al.(2017a) and Ichikawa et al. (2017) for further detailson the construction of these catalogs. These catalogsyielded a parent sample of 604 non-beamed AGNs; anysystems flagged as beamed in the Ricci et al. (2017a)catalog were removed during the matching process.In order to conduct our analysis, we required X-rayand mid-IR detections in all four WISE bands for theAGNs in our sample, which excluded another 78 AGNsfrom the final sample . We adopted the hard X-ray 2–10 keV luminosities (observed, uncorrected for intrinsicabsorption) from Ricci et al. (2017a) and the infraredluminosities in all four WISE bands ( . µ m , . µ m , µ m , and µ m ) from Ichikawa et al. (2017), whichare not corrected for any host galaxy contributions. Wefurther limited the sample to z < . AGNs to avoidredshift effects; this redshift cut removed another 70AGNs from the sample. The Ricci et al. (2017a) cat-alog includes independent estimates of N H from a torusmodel in the event that the column density found withthe phenomenological model was ≥ cm − . In thesecases, we instead use the column density inferred fromthe torus model, because torus models more accuratelyaccount for the 2–10 keV emission of CT AGNs.Following the matching of the catalogs and the appli-cation of the above requirements, the final sample wascomposed of 456 nearby, non-beamed AGNs with a me-dian redshift of z (cid:39) . and mid-IR luminosities inthe range . × ≤ L µ m / erg s − ≤ . × . DATA ANALYSISComparing the observed X-ray 2–10 keV luminosities( L X , Obs . ) to the µ m luminosities ( L µ m ), we see inFigure 1 that unobscured Swift /BAT AGNs tend to fol-low the relation between the intrinsic (unabsorbed) 2–10 X-ray and nuclear µ m luminosities (dashed blackline) established by Asmus et al. (2015), whereas ob-scured AGNs appear X-ray suppressed when compared While detected in the
WISE W [ . µ m ] and W [ . µ m ]bands, 67 AGNs did not satisfy the WISE data quality cuts es-tablished in Section 2.2.1 of Ichikawa et al. (2017). A further 11AGNs did not possess detections at either µ m or µ m . AllAGNs within the parent sample of 604 AGNs possessed 2-10 keVX-ray detections. R. W. Pfeifle et al. L
12 m (erg s )10 L k e V , O b s . ( e r g s ) l o g ( N H / c m ) Figure 1.
The observed 2–10 keV X-ray vs. µ m luminosi-ties. We color code the data points using the derived columndensity from the Ricci et al. (2017a) catalog, where the colormap is denoted on the auxiliary axis. AGNs with N H up-per limits of log( N H / cm − ) ≤ . are denoted with graytriangles. When comparing the 2–10 keV X-ray luminosityto the µ m luminosity for our sample of 456 Swift /BATAGNs, we see a general decrease in the X-ray-to-mid-IR ra-tio with increasing column density, as expected. The relationbetween the AGN intrinsic 2–10 keV luminosity (correctedfor absorption) and the nuclear 12 µ m luminosity derived byAsmus et al. (2015) is represented by a dashed black line,whereas we plot the logarithmic ratio of log( L X , Obs . / L µ m ) = − . with a black dotted line. Most heavily obscured (CT)AGNs exhibit luminosity ratios log( L X , Obs . / L µ m ) < − . . to L µ m . This decrease in L X , Obs . compared to L µ m with increasing column density is expected, of course,since the X-ray emission will suffer greater attenuationthan the mid-IR emission, and we color code the data inFigure 1 according to the column densities adopted inSection 2. As a population, CT AGNs are generally thefurthest offset from the Asmus et al. (2015) relation,exhibiting luminosity ratios of log( L X , Obs . / L µ m ) < − . (this was previously discussed in, e.g. Alexanderet al. 2008); we plot this ratio between log( L X , Obs . ) andlog( L µ m ) as a dotted black line in Figure 1. As hasbeen done in previous works (e.g. Alexander et al. 2008;Goulding et al. 2011; Asmus et al. 2015), we can use thisratio between L X , Obs . and L µ m as a diagnostic tool todifferentiate between less obscured and heavily obscuredor CT AGNs.Ratios of two mid-IR luminosities, sufficiently sepa-rated in wavelength, could exhibit the same trend asis observed for the L X , Obs . / L µ m ratio, in that theshorter wavelength mid-IR emission could appear sup-pressed compared to the longer wavelength emissiondue to the obscuring material surrounding the AGN.We test a new diagnostic tool based on the ratio of log( N H /cm )0.500.250.000.250.500.751.00 l o g ( L m / L . m ) log( N H /cm ) < 2222 log( N H /cm ) < 24 log( N H /cm ) 24log( N H /cm ) 20 Figure 2.
The ratio of the µ m and . µ m WISE lumi-nosities as a function of column density (reported in Ricciet al. 2017a), and we differentiate between unobscured (bluecircles), Compton-thin (green squares), and Compton-thick(red diamonds) AGNs, as denoted in the legend. AGNs with N H upper limits of log( N H / cm − ) ≤ . are denoted withgray triangles. While there is a large amount of scatter, thereis a general upturn in the luminosity ratio at the highest col-umn densities, beginning at ∼ – × cm − . the WISE µ m to . µ m luminosity ( L µ m / L . µ m )in Figure 2, and we find that the logarithmic ratio increases with column density, with the most signifi-cant increase in WISE ratio corresponding to the high-est column densities (although with significant scatter).For example, we find a mean luminosity ratio and 1- σ uncertainty of log( L µ m / L . µ m ) = 0 . ± . forAGNs with log( N H / cm − ) ≥ , whereas we find amean ratio and 1- σ uncertainty of log( L µ m / L . µ m ) =0 . ± . for unobscured AGNs with column densitieslog( N H / cm − ) < . This suggests that WISE colorscan also be used as a diagnostic tool for identifying heav-ily absorbed AGNs.We combine the L X , Obs . / L µ m and L µ m / L . µ m ratio diagnostics in Figure 3, plotting the two diagnosticratios against one another in Panel A and color codingthe data points by column density on the auxiliary axis.As the column density increases, the AGNs tend to ex-hibit lower values of L X , Obs . / L µ m and higher values of L µ m / L . µ m , and thus we find that the most heavilyobscured AGNs predominantly occupy the lower rightportion of the parameter space (i.e. the largest X-raydeficits and highest L µ m / L . µ m ratios). In Panel B ofFigure 3 we show this same result after binning the databy L µ m / L . µ m . In Panel C we bin by L µ m / L . µ m ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN L
22 m / L )210 l o g ( L k e V , O b s . / L m ) (A) 202122232425 l o g ( N H / c m ) L
22 m / L )2.01.51.00.50.0 l o g ( L k e V , O b s . / L m ) (B)0.5 0.0 0.5 1.0log( L
22 m / L )202122232425 l o g ( N H / c m ) (C) Figure 3.
The logarithmic L X , Obs . / L µ m and L µ m / L . µ m luminosity ratios (Panel A) with each pointcolor coded to indicate the column density (as indicatedby the auxiliary axis). Sources with N H upper limits of log( N H / cm − ) < . are denoted with gray triangles. Bin-ning by the L µ m / L . µ m luminosity ratio (Panel B), wesee a general trend of decreasing X-ray-to-mid-IR ratio withincreasing values of the L µ m / L . µ m ratio. Binning bythe L µ m / L . µ m luminosity ratio and comparing it tothe column density (Panel C), we see a general trend ofincreasing WISE luminosity ratios with increasing columndensity. Solid error bars in Panels B and C represent thestandard error of the mean while dashed error bars rep-resent the standard deviation computed for the respectivebin.
12 m / L )210 l o g ( L k e V , O b s . / L m ) (A) 202122232425 l o g ( N H / c m ) L
12 m / L )2.52.01.51.00.50.0 l o g ( L k e V , O b s . / L m ) (B)0.50 0.25 0.00 0.25 0.50 0.75log( L
12 m / L )21222324 l o g ( N H / c m ) (C) Figure 4.
In an identical fashion to that shown in Figure 3,we show the logarithmic L X , Obs . / L µ m and L µ m / L . µ m luminosity ratios (Panel A) with the column density givenon the auxiliary axis. We find very similar results tothose shown in Figure 3 when binning and comparing the L X , Obs . / L µ m and L µ m / L . µ m luminosity ratio (PanelB) and when binning and comparing the L µ m / L . µ m ratio and N H (Panel C). R. W. Pfeifle et al. and examine the scatter in the column density by bin; wenotice a positive correlation between the column densityand the mid-IR luminosity ratio. All solid error bars inPanels B and C represent the standard error of the meanwhile dashed error bars represent the standard deviationcomputed for the respective bin. Considering these threepanels together, we may define a parameter space usingthe L X , Obs . / L µ m and the L µ m / L . µ m luminosityratio, which could be used to identify the most heavilyabsorbed AGNs. We repeated this analysis for an al-ternative WISE luminosity ratio of L µ m / L . µ m andfound very similar results (see Figure 4). We exploredseveral AGN selection methods to potentially mitigateor at least account for the scatter observed, which wediscuss in the Appendix. Ultimately, we did not applyany selection criteria to our sample of Swift /BAT AGNsduring the analysis described below. Of important note,however, is the interesting result that the correlationbetween the mid-IR color and N H holds true for both WISE (selected via W − W > . ; Stern et al. 2012) as well as non- WISE
AGNs.3.1.
A Relation for N H as a Function of L X , Obs . / L µ m Since the L X , Obs . / L µ m ratio is known to correlatewith obscuring column (e.g. Ichikawa et al. 2012; Asmuset al. 2015; Yan et al. 2019), we derived an expressionto describe this relationship using the Swift /BAT sam-ple studied here. Note that, in contrast to Asmus et al.(2015) (see Section 4.2 for further details), we do notexclude sources with log( N H / cm − ) < . from the fit-ting process.For our fitting process, we incorporate the luminos-ity ratios and the associated uncertainties. We pull theuncertainties in the mid-IR fluxes from Ichikawa et al.(2017) and Ichikawa et al. (2019); while uncertaintiesin the observed X-ray luminosities were not availablein the Ricci et al. (2017a) catalog, we conservativelyadopt uncertainties of log( L X , Obs . ) = ± . for AGNswith log( N H / cm − ) < and log( L X , Obs . ) = ± . forAGNs with log( N H / cm − ) ≥ . In order to take intoaccount the asymmetric uncertainties associated withlog( N H / cm − ) and log( L X , Obs . / L µ m ), we employedthe following Monte Carlo fitting routine:1. Bootstrap : To fully account for the effect of out-liers, we generated a new data set as a sample ofthe original, allowing repeats.2.
Monte Carlo : For each point in the boot-strapped data set, we generate a new pointgiven the uncertainties in log( N H / cm − ) andlog( L X , Obs . / L µ m ). We incorporate asymmetric error bars by randomly drawing from a Gaussiandistribution separately for the negative and pos-itive error bars. For upper limits in log( N H ) ,we generate a new point using a uniform dis-tribution from the limit to an arbitrarily smalllog( N H / cm − ) value below it. After experi-menting a number of different values, each giv-ing similar results, we settled for a lower boundof log( N H / cm − ) = 19, which is only marginallylower than the smallest log( N H / cm − ) valuethrough the Milky Way according to the mapsby (Kalberla et al. 2005).3. Orthogonal Distance Regression : We fit eachMonte Carlo-bootstrapped data set with a func-tion of the form y = a · b · ( x − + c using orthog-onal distance regression (from the Python scipy package odr (Brown et al. 1990, pp. 186; Vir-tanen et al. 2020). This ensures we account forminimisation in both variables during the fittingprocedure.4. Parameter Estimation : Steps 1 – 3 were performedmany times, giving a distribution for the parame-ters a, b and c. For each parameter we report the th percentile and uncertainties derived from thethe th and th percentiles.The relationship between L X , Obs . / L µ m and N H canbe expressed as: log(L X , Obs . / L µ m ) = ( − . +0 . − . ) +( − . +0 . − . ) × ( N H / cm − ) (0 . +0 . − . ) (1)We plot the Swift /BAT sample and Equation 1 (orangeline) in Figure 5. To visualize the uncertainty in thisrelation, we show the 1000 realizations of the best fit asgrey lines.We attempted to fit directly for log( N H ) as a functionof log( L X , Obs . / L µ m ) but the fitting routine could notsuccessfully converge on a reasonable fit to the data.Instead, we chose to invert Equation 1 to recover theobscuring column density as a function of the X-ray tomid-IR luminosity ratio: log(N H / cm − ) = 20 + (1 . +0 . − . ) × log (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) log (cid:16) L X , Obs . L µ m (cid:17) + (0 . +0 . − . )( − . +0 . − . ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (2)To better quantify the uncertainty in log( N H / cm − )associated with this relation, we tabulated values of Note there are no limits for log( L X , Obs . / L µ m ). ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN Figure 5.
Relation for the column density vs.log( L X , Obs . / L µ m ) for the Swift /BAT sample (white cir-cles). New data samples simulated during the MC fittingroutine are displayed as blue dots. The best-fit trendline(orange line, given by Equation 1), represents the median of1000 realizations (gray lines) obtained through orthogonaldistance regression during the MC fitting process (see Sec-tion 3.1). The best-fit relation from Asmus et al. (2015) isshown as a red dashed line. log( N H ) and the uncertainty for specific values oflog( L X , Obs . / L µ m ), as shown in Table 1. Using the pa-rameter distributions calculated during the MC fittingroutine, we derived a distribution of log( N H / cm − ) val-ues using an expression of the form shown in Equation 2(inverted from Equation 1) and use the th , th , and th percentiles of the distribution to list the median col-umn density and associated uncertainty for each chosenvalue of log( L X , Obs . / L µ m ). While strictly empirical,Equations 1 and 2 reproduce the observed trend of de-creasing values of log( L X , Obs . / L µ m ) with increasingcolumn density. We do note that this relation is largelyinsensitive to ratios of log( L X , Obs . / L µ m ) > − . , withthe relation appearing nearly flat for column densitiesof log( N H / cm − ) < . . Equations 1 and 2 are mosteffective for obscured AGNs with column densities oflog( N H / cm − ) ≥ . , and readers should keep this inmind when using these expressions to derive estimatesfor column densities.We compare these results to those found in Asmuset al. (2015) in Section 4.2, and we plot the best fitrelation for log( N H ) versus log( L X , Obs . L µ m ) found byAsmus et al. (2015) in Figure 5. Table 1. log( N H / cm − ) Derived from Equations 2 and 9log( L X , Obs . / L µ m ) log( N H / cm − ) log( N H / cm − )(this work) (Asmus et al. 2015) − . . +0 . − . . ± . − . . +0 . − . . ± . − .
75 23 . +0 . − . . ± . − . . +0 . − . . ± . − . . +0 . − . . ± . − . . +0 . − . . ± . − . . +0 . − . . ± . − . . +0 . − . . ± . − . . +0 . − . . ± . Note —Column densities as a function oflog( L X , Obs . / L µ m ). Column 1: logarithmic ratio ofthe 2–10 keV to 12 µ m luminosities; Equation 2 is in-sensitive to ratios of log( L X , Obs . / L µ m ) > − . , whichare largely exhibited by AGNs with log( N H / cm − ) < .Column 2: column density and associated error, derivedby Equation 2; the parameter distributions found duringthe MC fitting process described in Section 3.1 were readinto an inverted expression of the form in Equation 2, fromwhich we retrieved the median column density and upperand lower bounds using the th , th , and th percentilesof the resulting log( N H / cm − ) distribution. Column 3:column density derived using Equation 6 from Asmus et al.(2015). Relation between N H and the Mid-Infrared Colors Given the correlation between L µ m / L . µ m and N H ,as shown in Figure 3, we followed the same fitting proce-dure as discussed above to develop an expression relat-ing the mid-IR luminosity ratio and N H . The equationrelating these properties can be expressed as: log(L µ m / L . µ m ) = (0 . +0 . − . ) +(0 . +0 . − . ) × ( N H / cm − ) (0 . +0 . − . ) (3)In Figure 6 we plot the Swift /BAT sample and the best-fitting trendline (orange line) along with 1000 realiza-tions of the best fit found during the fitting procedure(shown as gray lines).As mentioned in Section 3.1, we attempted to fit di-rectly for N H as a function of L X , Obs . / L µ m , but thefitting routine could not successfully converge on a rea-sonable line of best fit. We therefore simply invert Equa-tion 3 like before to recover the column density as a R. W. Pfeifle et al.
Figure 6.
Relation for column density vs.log( L µ m /L . µ m ) for the Swift /BAT sample (whitecircles). The fitting process for this relation is identical tothat in Figure 5. The best-fit trendline (orange line, givenby Equation 1) represents the median of 1000 realizations(gray lines) obtained during the MC fitting process. Theblack dashed line represents a theoretical prediction for asimple model consisting of monochromatic radiation passingthrough a homogeneous screen of dust (see Section 4.1.) function of the mid-IR luminosity ratio: log(N H / cm − ) = 20 + (3 . +1 . − . ) × log (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) log (cid:16) L µ m L . µ m (cid:17) − (0 . +0 . − . )(0 . +0 . − . ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (4)In an identical fashion to the calculations in Section 3.1,we tabulated values of log( N H / cm − ) and the associateduncertainties for specific values of log( L µ m / L . µ m )in Table 2. Equation 4 is not sensitive to values oflog( L µ m / L . µ m ) < and, as in the case of Equa-tion 2, readers should use this relation cautiously andbear in mind that it is really only effective for obscuredAGNs with log( N H / cm − ) ≥ . .3.3. Diagnostic Regions for Heavily Absorbed AGNs
The correlations between L X , Obs . / L µ m , L µ m / L . µ m ,and N H discussed above suggest that these relation-ships may be combined to help differentiate between Swift /BAT AGNs according to the levels of obscuration.We probed this potential diagnostic for heavily absorbedAGNs by plotting L µ m / L . µ m vs. L X , Obs . / L µ m and binning the full sample by absorbing column as Table 2. log( N H / cm − ) Derived from Equation 4log( L µ m / L . µ m ) log( N H / cm − )0.0 . +1 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . Note —Column densities as a function oflog( L µ m / L . µ m ). Column 1: logarithmic ratio ofthe 22 µ m to 4.6 µ m luminosities; Equation 2 is insensitiveto ratios of log( L µ m / L . µ m ) > . Column 2: columndensity and associated error, derived by Equation 4; themedian value and associated uncertainties were derived inan identical fashion to that described in Table 1. shown in Figure 7. The sample was divided into fourbins in obscuration corresponding to:• unobscured[ log( N H / cm − ) < , Panel A]• Compton-thin “lightly obscured”[ ≤ log( N H / cm − ) < , Panel B]• Compton-thin “moderately obscured”[ ≤ log( N H / cm − ) < , Panel C]• “heavily obscured” to Compton-thick[ log( N H / cm − ) ≥ , Panel D]We also split the two Compton-thin bins into two sub-bins each in increments of log( N H / cm − ) = 0 . , as sum-marized in Table 3.Contours were computed for each bin and/or sub-binand in each case are designed to encompass ∼ of thepopulation of each respective bin. The binned subplotsshown in Figure 7 demonstrate much more clearly that, in general , the heavily obscured sources (Panel D) tendto occupy a separate region of space than the unobscured(Panel A) or Compton-thin ‘lightly obscured’ (Panel B)sources. While there is some overlap between the CTand Compton-thin ‘moderately obscured’ populations,this is predominantly due to Compton-thin ‘moderatelyobscured’ sources with . ≤ log( N H / cm − ) < . Inlight of this, appropriate selection criteria can be usedto construct diagnostic regions in this parameter space ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN log( L X , Obs . / L µ m ) < − . (5)which is shown as a black dashed line in Figure 7, pro-vides a simple yet robust method of differentiating be-tween the most heavily obscured AGNs and less ob-scured AGNs in the Swift /BAT sample. We reportthe sample statistics for this cut in Table 3. To derivethe population statistics for Table 3 (and all percent-ages quoted hereafter), we calculated the median ( th percentile) value for each population in question, whilethe uncertainties on the fractions are the th and th quantiles of a binomial distribution, all computed fol-lowing Cameron (2011). The criterion in Equation 5yields . +4 . − . % completeness for the heavily obscuredAGNs, a . +6 . − . % pure sample, and a mean columndensity of log( N H / cm − ) = 24 . ± . for the diagnos-tic region. It is important to note that the majorityof impurities selected with Equation 5 arise from AGNswith column densities . ≤ log( N H / cm − ) < . ; thediagnostic region is in fact ∼ % pure for AGNs with log( N H / cm − ) ≥ . and suffers minimal impuritiesfrom AGNs with lower column densities.In a similar fashion, we could define a vertical cutbased on the WISE colors in this space: log(L µ m / L . µ m ) > . (6)and while this criterion also yields a highly completesample ( +4 . − . %) of heavily obscured AGN, the se-lected sample is only . +2 . − . % pure for heavily ob-scured AGNs and is greatly contaminated by moder-ately obscured, lightly obscured, and even unobscuredAGNs. This is not at all surprising, given the scatterin the mid-IR ratio as demonstrated in Figures 3 and 5.Mid-IR selection alone is therefore not sufficient whenattempting to select both a highly complete and fairlypure sample of heavily obscured sources.Next we defined a slightly more stringent box region(gray dash-dotted line and black dotted line in PanelsA–D of Figure 7), which encompasses the majority ofthe most heavily absorbed sources with minimal overlapwith the unobscured and Compton-thin bins, using thefollowing relations: . < log( L µ m / L . µ m ) < . − . < log( L X , Obs . / L µ m ) < − . (7) Table 3. log( L X , Obs . / L µ m ) ≤ . Diagnostic Cut log( N H / cm − ) Completeness Purity ≥ . . +4 . − . . +6 . − . < . . +1 . − . . +6 . − . [23 . , .
0) 16 . +3 . − . . +6 . − . [23 . , .
0) 30 . +6 . − . . +6 . − . [23 . , .
5) 4 . +3 . − . . +3 . − . [22 . , .
0) 5 . +2 . − . . +4 . − . [22 . , .
0) 5 . +3 . − . . +3 . − . [22 . , .
5) 6 . +4 . − . . +3 . − . < . . +0 . − . . +2 . − . Note —Statistics derived from the log(L
Obs . X / L µ m ) < − . threshold, defined in Section 3 (Equation 5), for various N H bins and sub-bins. Column 1: N H bin. Column 2: Complete-ness, or the fraction of AGNs selected (per column densitybin). Column 3: Purity of the sample, or the percentagecontribution to the diagnostic box. Table 4. log( L µ m / L . µ m ) ≥ . Diagnostic Cut log( N H / cm − ) Completeness Purity ≥ . . +4 . − . . +2 . − . < . . +2 . − . . +2 . − . [23 . , .
0) 75 . +3 . − . . +2 . − . [23 . , .
0) 81 . +5 . − . . +2 . − . [23 . , .
5) 69 . +5 . − . . +2 . − . [22 . , .
0) 56 . +5 . − . . +2 . − . [22 . , .
0) 60 . +6 . − . . +2 . − . [22 . , .
5) 51 . +7 . − . . +1 . − . < . . +3 . − . . +3 . − . Note —Statistics derived from the mid-IRlog( L µ m / L . µ m ) ≥ . criteria (without invokingany cut in log[L Obs . X / L µ m ] ) defined in Section 3 (Equa-tion 6), for various N H bins and sub-bins. Columns 1-3:The same as Table 3. We report the population statistics for this diagnos-tic box in Table 5. This box offers a completeness of . +5 . − . % for the heavily obscured AGN population, a . +6 . − . % pure sample, and a mean column density oflog( N H / cm − ) = 24 . ± . . As with Equation 5, thelargest source of impurities within this region are AGNswith . ≤ log( N H / cm − ) < ; the region is ∼ %0 R. W. Pfeifle et al.
Table 5. L µ m / L . µ m Diagnostic Box Statistics log( N H / cm − ) Completeness Purity ≥ . . +5 . − . . +6 . − . < . . +1 . − . . +6 . − . [23 . , .
0) 15 . +3 . − . . +6 . − . [23 . , .
0) 28 . +6 . − . . +6 . − . [23 . , .
5) 4 . +3 . − . . +3 . − . [22 . , .
0) 3 . +2 . − . . +4 . − . [22 . , .
0) 5 . +3 . − . . +3 . − . [22 . , .
5) 3 . +3 . − . . +3 . − . < . . +0 . − . . +2 . − . Note —Statistics derived from the diagnostic box developedfor the L µ m / L . µ m luminosity ratio (defined by Equa-tion 7 in Section 3) for various N H bins and sub-bins.Columns 1-3: The same as Table 3. pure for AGNs with log( N H / cm − ) ≥ . and suffersfew impurities from AGNs of lower column densities.We repeated this analysis for the alternative L µ m / L . µ m luminosity ratio, and we show the contoured populationsbinned by column density along with an alternative di-agnostic box for heavily absorbed sources in Panels E–Hof Figure 7. We construct this box with the followingrelations: . < log( L µ m / L . µ m ) < . − . < log( L X , Obs . / L µ m ) < − . (8)and we find that this box yields a completeness of . +5 . − . % for heavily absorbed AGNs, a purity of . +6 . − . %, and a median column density of log( N H / cm − ) =24 . ± . . Again, AGNs with column densities . ≤ log( N H / cm − ) < contribute the most tothe impurity of the sample, while AGNs with lowercolumn densities do not contribute as significantly.The diagnostic metrics defined above — developed us-ing the well-constrained X-ray and mid-IR propertiespreviously found for Swift /BAT AGNs (e.g. Ricci et al.2017a; Ichikawa et al. 2017) — carve out parameterspaces that yield fairly complete and fairly pure sam-ples of heavily obscured AGNs, offering an efficient andeffective method for identifying heavily obscured or CTAGN candidates, particularly in large samples of AGNs. DISCUSSION4.1.
The physical origin of the trend in WISE ratios asa function of N H The observed trend of the µ m / . µ m and µ m / . µ m WISE ratios increasing with column density can be readily understood from dust absorption and emissionproperties and basics of the radiation transfer. For me-dia optically thin to the mid-IR radiation, the shapeof the resulting SED will be determined predominantlyby the dust temperature and its gradient throughoutthe dusty structure. If the dusty medium is opticallythick to its own radiation, the outgoing emission will bereshaped due to a number of reasons: (i) Warm dustemission at shorter wavelengths will be absorbed andre-emitted at longer wavelengths. (ii) Dust emissionat shorter wavelengths will suffer more extinction thanthe long-wavelength emission, owing to the wavelength-dependent extinction for typical AGN dust (Laor &Draine 1993). (iii) Warm dust emission originates closerto the inner rim of the torus, while colder emission orig-inates farther out. As a consequence, longer wavelengthmid-IR radiation has to travel a shorter path throughthe dust before reaching us and thus, suffers even lessextinction than the emission of shorter wavelengths.These three effects result in an increased ratio of longer-to-shorter wavelength mid-IR emission and scale withthe column density of the medium through which the X-ray radiation is traversing. (iv) Additionally, a disk-likemolecular structure in hydrostatic equilibrium is ex-pected to have a vertical gradient (Hönig 2019). In thiscase, the observed trend of increasing
WISE ratios withincreasing N H can be explained simply as an inclinationeffect: the closer our viewing angle is to the equator,the higher is the column density along the line-of-sightand, at the same time, the dust emission becomes “red-der”. All these effects contribute to the observed trendof increasing WISE ratios with N H and also explainwhy the effect is more pronounced in the µ m / . µ m than at µ m / . µ m luminosity ratio (for illustration,see torus model SEDs in, e.g., Hönig & Kishimoto 2010;Stalevski et al. 2012).There are a few caveats: The dust emission is oftendegenerate, as SEDs of similar shape can be producedby different combinations of the geometrical and phys-ical parameters of the torus, some of which can con-spire to work against or hide the trend in luminosityratios. However, the above reasoning should hold in gen-eral since it relies on universal radiative transfer effects.Another deviation can be introduced by the presence ofsilicate dust grains which exhibit a strong increase of ab-sorption efficiency around and µ m . The apparentstrength of these features appear in an SED depends onseveral factors, including the amount of silicates, grainsize distribution, and radiative transfer effects.We illustrate these effects in Figure 6 with a blackdashed line, which represents a theoretical expectationfor a very simple model: monochromatic radiation pass- ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN l o g ( L k e V , O b s . / L m ) (A)
22 log( N H /cm ) (B)
22 log( N H /cm ) < 22.522.5 log( N H /cm ) < 23
22 m / L )3.02.52.01.51.00.50.00.5 l o g ( L k e V , O b s . / L m ) (C)
23 log( N H /cm ) < 23.523.5 log( N H /cm ) < 24
22 m / L ) (D)
24 log( N H /cm ) (E)
22 log( N H /cm ) (F)
22 log( N H /cm ) < 22.522.5 log( N H /cm ) < 23
12 m / L ) (G)
23 log( N H /cm ) < 23.523.5 log( N H /cm ) < 24
12 m / L ) (H)
24 log( N H /cm ) Figure 7. L µ m / L . µ m (Panels A–D) and L µ m / L . µ m (Panels E–H) diagnostics for AGNs in the Swift /BAT samplebinned by N H , where we have unobscured (Panel A, E; log( N H / cm − ) < ), Compton-thin ‘lightly obscured’ (Panel B,F; ≤ log( N H / cm − ) < ), Compton-thin ‘moderately obscured’ (Panel C, G; ≤ log( N H / cm − ) < ), and ‘heavilyobscured’ to Compton-thick (Panel D, H; log( N H / cm − ) ≥ ). The Compton-thin bins are broken into two sub-bins each: ≤ log( N H / cm − ) < . (black points), . ≤ log( N H / cm − ) < (blue points), ≤ log( N H / cm − ) < . (blackpoints), and . ≤ log( N H / cm − ) < (orange points). Contours were computed to encompass ∼ of the populationfor each respective bin or sub-bin. Note in general that the most obscured sources tend to populate a different region of theparameter space than do the unobscured sources and Compton-thin ‘lightly obscured’ sources, while there is some overlap withCompton-thin ‘moderately obscured’ sources. ing through a homogeneous screen of dust. For this ex-ample, we assumed a typical Galactic interstellar dustmixture of silicates and graphite (e.g., Stalevski et al.2016). The grain size distribution are from Mathis et al.(1977) and optical properties from Laor & Draine (1993)and Li & Draine (2001). The conversion between theoptical depth and N H assumes Galactic relation be-tween extinction and column density found by Predehl& Schmitt (1995). We see that the theoretical curve isfollowing the trend of the data at lower column densi-ties, but is reaching the breaking point sooner. This isbecause the simple dust screen model does not accountfor a number of radiative transfer effects (self-consistentabsorption and re-emission of the thermal IR radiation),which together with geometry of the dusty medium andorientation shape the resulting SED, and thus, the ob-served trend of luminosity ratios with column density.4.2. Comparison to Asmus et al. (2015)
Using sub-arcsecond resolution mid-IR observations– which enabled the isolation of the nuclear mid-IRemission ( F nuc12 µ m ) – Asmus et al. (2015) found a sig- nificant correlation between log(F nuc12 µ m / F obs2 −
10 keV ) and log( N H / cm − ) for 53 AGNs with reliable X-ray observa-tions and column densities log( N H / cm − ) > . , whichwas expressed as (see Equation 6 in Asmus et al. 2015): log (cid:18) N H . − (cid:19) = (0 . ± . . ± .
12) log (cid:32) F nuc12 µ m F obs2 −
10 keV (cid:33) (9)This expression is plotted in Figure 5 (dashed red line)along with our Equation 2 (solid orange line) and the
Swift /BAT sample. Along with values of log( N H / cm − )derived using Equation 2 in Section 3.1, we use therelation from Asmus et al. (2015) to derive values oflog( N H / cm − ) and the uncertainties for specific val-ues of log( L X , Obs . / L µ m ) in order to compare to ourown results. Despite the fact that Asmus et al. (2015)removed AGNs with log( N H / cm − ) < . and uti-lized subarcsecond-resolution mid-IR emission (whereasin this work we utilized lower angular resolution mid-IRphotometry for the Swift /BAT sample), it does appear2
R. W. Pfeifle et al. that the two relations generally agree (within the uncer-tainties) for ratios of − . (cid:46) log ( L X , Obs . / L µ m ) (cid:46) − . . The two relations differ more severely for higherratios of log( L X , Obs . / L µ m ), though this is expecteddue to (1) the wide range of ratios that unobscuredAGNs exhibit and (2) the fact that our relation turnsover to account for less obscured sources while the As-mus et al. (2015) relation does not take into account lessobscured sources.4.3. Diagnosing Column Densities with UncertainDust Heating Sources
While the diagnostic boxes defined in Section 3 pro-vide a reliable way to identify the most heavily ob-scured AGNs, star formation activity can contributenon-negligibly to the mid-IR colors of an AGN host.The mid-IR colors assumed to originate from the AGNitself could therefore be overestimated without perform-ing detailed spectral energy decomposition (SED) fittingto differentiate between the AGN and star formationcontributions to the mid-IR continuum. Furthermore,Satyapal et al. (2018) demonstrated, using
Cloudy (Fer-land et al. 2013, 2017) radiative transfer models, thatheavily obscured star formation activity can actuallymimic the mid-IR colors of AGNs. These two pointssuggest that our diagnostic boxes defined in Section 3may (1) misdiagnose the column density of an AGN ifsignificant star formation is present, as the contaminat-ing stellar emission could lead to much redder colorsthan the AGN intrinsically exhibits, or (2) mislead usto think an AGN is present in cases where the dom-inant dust heating sources are actually stellar-relatedrather than AGN-related (Satyapal et al. 2018). To in-vestigate this potential contamination of the diagnos-tic boxes, we constructed a catalog of optically-selectedgalaxies whose optical spectroscopic line ratios suggeststar formation dominates the observed emission, and weexamined methods – for example mid-IR or X-ray selec-tion criteria – through which this contamination couldbe mitigated.Beginning with the MPA-JHU catalog of galaxy prop-erties (from the SDSS data release 8, Aihara et al. 2011),we first selected systems with redshifts z < . and in-cluded only systems which are classified as star form-ing systems (“BPTClass” = 1) based upon their Bald-win, Phillips, Telervich (BPT; Baldwin et al. 1981) op-tical spectroscopic emission line ratios. We also re-moved any systems with QSO and AGN flags withinthe “TARGETTYPE,” “SPECTROTYPE,” and “SUB-CLASS” columns, and then narrowed the sample to onlysystems with WISE counterparts and X-ray counter-parts from the 4XMM point source catalog (Webb et al., L
22 m / L )5432101 l o g ( L k e V , O b s . / L m ) Optically Normal Galaxies
WISE -Selected (Optically Normal) Galaxies l o g ( L k e V , O b s . / e r g s ) L
12 m / L )5432101 l o g ( L k e V , O b s . / L m ) Optically Normal Galaxies
WISE -Selected (Optically Normal) Galaxies l o g ( L k e V , O b s . / e r g s ) Figure 8.
Top (bottom): Diagnostic box defined in Fig-ure 7 and Equation 7 (Equation 8) for the L µ m / L . µ m ( L µ m / L . µ m ) ratio with a population of star forminggalaxies (see Section 4.3) overlaid as triangles. The horizon-tal dashed black line is given by Equation 5. The observedX-ray luminosity is denoted on the auxiliary axis. The di-agnostics presented here cannot unambiguously differentiatebetween AGNs and star forming systems since a significantfraction of the star forming galaxy population falls withinthe absorption diagnostic box. This contamination can bemitigated with a mid-IR WISE cut of W − W > . (Stern et al. 2012); only six optically normal galaxies satisfythis mid-IR criteria (red open circles), and these systems falloutside of the diagnostic box defined by Equation 7 (Equa-tion 8). submitted). These criteria yielded a full parent sampleof 448 galaxies which we assume are ‘purely’ star form-ing systems based upon optical spectroscopic measure-ments. We make no distinction between morphologicalclasses of galaxies.We plot our population of optically-selected star form-ing galaxies (color coded according to the observed ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN L µ m / L . µ m and L µ m / L . µ m diagnostic boxes in Figure 8. While starformation dominated galaxies tend to exhibit lower ra-tios of log( L X , Obs . / L µ m ) than the majority of the Swift /BAT sample, they do tend to exhibit similar X-ray deficits as well as L µ m / L . µ m and L µ m / L . µ m mid-IR colors as those exhibited by the heavily obscured Swift /BAT AGNs; in fact, . ± . of this star form-ing population overlaps the L µ m / L . µ m diagnosticregion, while . +2 . − . % of the population overlaps the L µ m / L . µ m region. We therefore caution that thisdiagnostic is emphatically not designed to differentiatebetween star forming and AGN-dominated systems andshould not be used as a diagnosis of the dominant dustheating source. Nevertheless, the contamination fromoptically-selected star forming systems can be mitigatedthrough the use of reliable mid-IR and X-ray selectioncriteria traditionally used for identifying AGNs.We applied the two band WISE
AGN selection cut( W . µ m] – W . µ m] > . ) from Stern et al. (2012)to the sample of star forming galaxies, which removed allbut six systems (red empty circles, shown in Figure 8).As expected, requiring traditional mid-IR AGN selec-tion criteria eliminates virtually all contamination byoptically-selected star forming systems within the diag-nostic boxes. While six star forming galaxies ( . +0 . − . %of the total population) succeed in meeting the Sternet al. (2012) cut, these do still fall outside of ourdiagnostic boxes. Thus, use of mid-IR AGN selec-tion tools could be used to avoid misdiagnosing thedominant photoionization process of the sources withinthe diagnostic regions. However, it is important tobear in mind that the relation between L X , Obs . / L µ m , L µ m / L . µ m (and L µ m / L . µ m ), and N H holds truefor both WISE and non- WISE selected AGNs (see Ap-pendix), and therefore requiring a
WISE cut could ingeneral remove true AGNs as well as star forming sys-tems. For example, imposing the W – W > . cuton the Swift /BAT sample examined in this work wouldremove 240 systems (or . +2 . − . % of the parent sampleof 456 AGNs); in the parent sample of 456, there are71 AGNs which possess column densities in excess of × cm − , and 40 of these would be removed withthis mid-IR cut.We also found that requiring an observed X-ray lu-minosity of L X , Obs . > erg s − removes nearly all L X , Obs . / L µ m : (1) acompact star forming region in a galaxy ∼ Mpc away, and (2)a galaxy at z = 0 . , (3) a pair of merging galaxies at z = 0 . ,(4) a pair of merging galaxies which actually host a candidate dualAGN at z = 0 . (Pfeifle et al. 2019). of the optically star forming population from the di-agnostic region (5 galaxies, or . +0 . − . %, remain withinthe L µ m / L . µ m diagnostic box), though this methodmust also be used judiciously to avoid removing heavilyobscured AGNs, which could exhibit lower X-ray lumi-nosities.Ideally, the usage of this diagnostic should be limitedto systems whose dominant photoionization processesare unambiguous or for which detailed SED fitting canbe performed to differentiate between AGN and hostemission. Otherwise, we recommend proceeding cau-tiously, taking into account the various caveats outlinedabove to avoid inaccurate estimations of the obscurationalong the line-of-sight.4.4. Mid-IR Emission Contributions from GalaxiesHosting Obscured AGNs
The realization in Section 4.3 that optically star form-ing galaxies, which presumably do not host AGNs,can exhibit luminosity ratios similar to those exhib-ited by more heavily obscured or CT
Swift /BAT AGNsraises the intriguing point of how much host galaxiesmay contribute to the observed luminosity ratios de-rived for the
Swift /BAT AGNs. Figure 9 shows thelog( L X , Obs . / L µ m ) and log( L µ m / L . µ m ) ratios ofthe Swift /BAT AGNs and is color-coded according tothe fractional contribution by the AGN to the observed12 µ m emission ( f µ mAGN ) derived through detailed SEDfitting in Ichikawa et al. (2019). While host-dominatedsystems at 12 µ m ( f µ mAGN < . ) can be found across thisparameter space, a significant fraction ( . +6 . − . %, 22out of 51) of AGNs within the diagnostic region (Equa-tion 7) reside in host-dominated systems. This suggeststhat the host galaxies could contribute significantly tothe observed mid-IR colors of the heavily obscured AGNpopulation in particular, presumably via dust emissionheated through star formation.Ichikawa et al. (2019) provided decomposed loga-rithmic AGN 12 µ m luminosities for the Swift /BATAGNs, which we can use here to examine how thelog( L X , Obs . / L µ m ) ratios may change if we use theAGN 12 µ m luminosity ( L µ m , AGN ) instead of the to-tal observed 12 µ m luminosity ( L µ m ). Figure 10(top panel) shows that host-dominated systems are pre-dominantly occupied by AGNs with log( N H / cm − ) (cid:38) . ; half of the CT Swift /BAT AGNs reside in hostdominated systems. After recalculating the luminosityratio using L µ m , AGN instead (bottom panel), host-dominated systems exhibit a shift toward higher lu-minosity ratios, with an average difference in ratio of ∆ log( L X , Obs . / L µ m ) ≈ . for AGNs with f µ mAGN < . , although we note that these shifts are not limited4 R. W. Pfeifle et al. L
22 m / L )210 l o g ( L k e V , O b s . / L m ) f ( m ) A G N Figure 9.
The logarithmic L X , Obs . / L µ m vs. L µ m / L . µ m ratios of the Swift /BAT AGN sample,where the data are color-coded according to the fractionalcontribution of the AGN µm emission to the totalobserved µ m emission ( f µ mAGN ). While there is no clearoffset in this parameter space between AGN-dominatedand host-dominated systems, several of the most heavilyobscured AGNs (i.e. sources with significant X-ray deficitsand log( L µ m / L . µ m ) > . ) do reside in systems wherethe host dominates the µ m emission. only to heavily obscured AGNs. Here we have assumedthat the X-ray emission is AGN-dominated, rather thanhost-dominated; in reality, if some portion of the X-rayemission is due to the host as well, the observed ratioshifts will not be as large.Figure 9 demonstrates that host galaxies do indeedcontribute significantly to the diagnostic ratios probedin this work, at least for systems in which the host dom-inates the mid-IR emission at 12 µ m . In these cases,the host contribution to the mid-IR leads to a perceivedlarger X-ray deficit for the AGN at a given columndensity. It does appear, though, that generally thiseffect actually works in our favor when attempting toidentify CT AGNs, as these more severe X-ray deficitsand presumably ‘redder’ mid-IR colors aid in separat-ing this population from less obscured populations incolor space. Decomposed AGN 22 µ m and 4.6 µ m lumi-nosities were not included in Table 1 of Ichikawa et al.(2019) and therefore could not be examined in a similarfashion here. While it is beyond the scope of this paper,an analysis of the interplay between the mid-IR colors,host galaxy emission, and AGN emission with regard tothe selection diagnostics presented in this work shouldbe performed more rigorously in a future study.4.5. Comparison to Kilerci-Eser et al. (2020)
In a very recent study, Kilerci-Eser et al. (2020) se-lected a subsample of the 105 month
Swift /BAT catalog(Oh et al. 2018) and proposed a new selection method for log( N H /cm )32101 l o g ( L k e V , O b s . / L m ) log( N H /cm )32101 l o g ( L k e V , O b s . / L m , A G N ) f ( m ) A G N f ( m ) A G N Figure 10. log N H / cm − ) vs. log( L X , Obs . / L µ m ) for (top)the total 12 µ m emission and (bottom) the decomposed AGN12 µ m emission from Ichikawa et al. (2019). The auxiliaryaxes represent the fractional contribution by the AGN tothe total observed 12 µ m emission. (Top) As is alreadyknown, log( L X , Obs . / L µ m ) decreases with increasing col-umn density, however a significant number of the obscuredand CT AGNs contribute less than of the total ob-served 12 µ m ( f µ mAGN ). (Bottom) The relationship betweenlog( L X , Obs . / L µ m ) and log( N H / cm − ) is still present whenrecalculating the ratio using the decomposed AGN 12 µ m luminosity ( L µ m , AGN ), although obscured and CT AGNsexhibit smaller deficits than when using the total 12 µ m lu-minosity. Host galaxies can therefore contribute significantlyto the observed X-ray to mid-IR ratios of AGNs, especiallyCT AGNs. CT AGNs using mid-IR and far-IR photometry. Theyreport, as we do here in Section 3 of this study, a shiftin infrared colors (specifically mid-IR and far-IR) to-ward ‘redder’ colors with increasing column density, andthey define a physically motivated color-color diagram(see Figure 11, henceforth F , in Kilerci-Eser et al.2020) and selection method using the [ µ m ]–[ µ m ] ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN µ m ]–[ µ m ] colors. Of the 32 CT Swift /BATAGN for which there exists the relevant photometry,this selection criteria identifies four CT AGNs (a suc-cess rate of . +6 . − . % ). However, it is evident from F that these color cuts cannot reliably distinguishbetween unobscured, obscured, and CT AGNs, as theAGNs from these three different obscuration bins largelyoccupy the same [ µ m ]–[ µ m ] and [ µ m ]–[ µ m ] pa-rameter space. In the case of Swift /BAT, the color-colorcriteria proposed by Kilerci-Eser et al. (2020) yields a farlower success rate of identifying heavily obscured andCT AGNs than the criteria set forth in this work (e.g.Equations 7 and 8; see Tables 3, 5, and 6).To further test their diagnostic, they applied this colorselection criteria to the
AKARI infrared galaxy catalogdeveloped in Kilerci Eser & Goto (2018), which con-tains over 17,000 galaxies, and recover one known CTAGN (NGC 4418, e.g. Sakamoto et al. 2013). The re-mainder of the infrared galaxy sample of Kilerci Eser& Goto (2018) is represented with blue contours in F that partially overlap a significant number of Swift /BATCT, obscured, and unobscured AGNs, suggesting thatsome portion of these infrared galaxies may in fact hostheavily obscured AGNs. Despite finding a few cases ofCT AGNs between the
Swift /BAT and Kilerci Eser &Goto (2018) samples, the diagnostic criteria set forth inKilerci-Eser et al. (2020) does not appear to provide acomplete or reliable (see Section 5.4 of Kilerci-Eser et al.2020) method of selecting CT AGNs.As an additional comparison between our selectionmethod and that proposed by Kilerci-Eser et al. (2020),we turned our attention to the infrared galaxy catalogfrom Kilerci Eser & Goto (2018). We matched this sam-ple to the AllWISE catalog and the 4XMM DR9
XMM-Newton
Serendipitous Source Catalog (Webb et al. sub-mitted), using a match radius of 10 (cid:48)(cid:48) for each, whichyielded a sample of 401 local ( z < . ) infrared galax-ies with XMM-Newton and
WISE detections. In Fig-ure 11, we show the resulting sample of infrared galax-ies (red stars), along with our diagnostic criteria fromEquations 5 and 7. As in Section 4.3, it is impossible todiscern whether or not any of these galaxies host AGNswithout a reliable method for removing star formationdominated systems. We tried four different mid-IR AGNselection criteria, defined in Jarrett et al. (2011), Sternet al. (2012), Assef et al. (2018), and Satyapal et al. The selection criteria also recovers two other sources:NGC 7714, an unobscured AGN (Gonzalez-Delgado et al. 1995;Smith et al. 2005), and NGC 1614, which has no clear evidenceof an AGN (e.g. Xu et al. 2015; Pereira-Santaella et al. 2011;Herrero-Illana et al. 2014). L
22 m / L )432101 l o g ( L k e V , O b s . / L m ) Kilerci-Eser + 2018, IR GalaxiesStern+2012 Selected Mid-IR AGNs
Figure 11.
The L X , Obs . / L µ m vs. L µ m / L . µ m ratiosfor the infrared galaxies (red stars) cataloged by Kilerci Eser& Goto (2018) along with the obscuration diagnostics estab-lished in Equations 5 and 7. After applying the Stern et al.(2012) mid-IR cut to search for AGNs within the sample, wefind a significant population of candidate heavily obscured orCT AGNs contained within the Kilerci Eser & Goto (2018)catalog, a result not found using the mid-IR to far-IR color-color criteria proposed by Kilerci-Eser et al. (2020). See Ta-ble 7 for more details on these AGNs. (2018) (with the understanding that some heavily ob-scured AGNs will be missed with this simple approach),and in all four cases we recover a significant numberof candidate heavily obscured or CT AGNs. We showmid-IR AGNs selected as a result of the Stern et al.(2012) cut in Figure 11 (blue squares); these candidateCT AGNs likely inhabited the blue contoured promi-nence that overlapped the Swift /BAT AGNs in F ,but were missed due to fact that they did not satisfythe criteria proposed by Kilerci-Eser et al. (2020).Table 7 in the Appendix lists the 36 candidate CTAGNs selected from Kilerci Eser & Goto (2018) usingthe L µ m / L . µ m diagnostic region and at least oneof the mid-IR selection cuts listed above. We includein the table the source coordinates, redshifts, luminos-ity ratios, and alternative identifiers, and we also cat-egorize the column densities of the sources (using thecolumn density bins defined in Section 3.3) based uponany available measurements in the literature. Of thecandidates that have inferred or directly measured col-umn densities in the literature (24/36), we find eightCT AGNs, six moderately obscured AGNs, two lightly6 R. W. Pfeifle et al. obscured AGNs, and three unobscured AGN, while a re-maining five AGNs have conflicting measurements of N H in the literature (all five of which have been reported asCT at least once in the past). Therefore, we concludethat our diagnostic criteria proposed in Equations 5, 7,and 8 offer a more reliable method for identifying candi-date CT AGNs than the mid-IR to far-IR color criteriaproposed by Kilerci-Eser et al. (2020).In light of Section 4.4, we caution that some portionof these AGNs may not dominate the observed µ m emission and, therefore, may exhibit larger X-ray deficitsand redder mid-IR colors than might be expected for theAGN alone due to additional mid-IR contributions fromthe host galaxy.4.6. Diagnosis of N H in the XXM-XXL Field To test the power of our absorption diagnostic, we turnour attention to its application in the
XMM
XXL Northfield (Pierre et al. 2016, 2017). Menzel et al. (2016) pre-sented a rigorous multiwavelength analysis of 8445 X-raysources detected by
XMM-Newton in an 18 deg area ofthe XMM
XXL North field (hereafter XXL-N), with alimiting flux of F . −
10 keV > − erg cm − s − , pro-viding optical Sloan Digital Sky Survey (SDSS) and mid-IR WISE counterparts to the
XMM-Newton sources.In a complementary investigation, Liu et al. (2016)presented a thorough X-ray spectral analysis of the2512 XXL-N AGNs, deriving the spectral properties(e.g. photon index Γ , N H , L X , Obs . ) for those AGNsusing a Bayesian statistical approach contained withinthe Bayesian X-ray Astronomy (BXA) software package(Buchner et al. 2014).We combined the catalogs from Menzel et al. (2016)and Liu et al. (2016) to obtain the observed 2–10 keVluminosities and mid-IR WISE magnitudes, from whichwe derived the relevant luminosity ratios examined inSection 3. Initially we limited the XXL-N sample toonly local ( z < . ) AGNs, and as with the Swift /BATAGNs, we did not employ any mid-IR or X-ray selectioncriteria. We plot the resulting luminosity ratios of thelow redshift AGNs from the XXL-N field in the left panelof Figure 12 along with our L µ m / L . µ m diagnosticbox (Equation 7) and horizontal cut in L X , Obs . / L µ m (Equation 5). The data and auxiliary axis in Figure 12are color coded to represent the derived th -percentile N H values from the Liu et al. (2016) catalog and themarkers denote the obscuration bin for each AGN. Inthe right panel of Figure 12 we plot the L X , Obs . / L µ m There are other moderately and heavily obscured
WISE
AGNsin this sample which did exhibit log( L X / L µ m ) < − . but falloutside of our more stringent diagnostic region. ratio against the derived log( N H / cm − ) values from Liuet al. (2016), where the error bars represent the th and th percentiles and the data points use the samemarker and color scheme as the left panel. A dearth oflow redshift AGNs in the XXL-N field is immediatelyapparent, and while it appears that most of the moreobscured AGNs do exhibit “redder” colors, there is someoverlap in L X , Obs . / L µ m ratios exhibited by AGNs ofstarkly different obscuration bins, e.g. heavily obscuredand unobscured AGNs. Unfortunately, we cannot drawdefinitive conclusions about the reliability of our diag-nostic for the XXL-N field with such poor AGN statis-tics.We repeated this analysis for the entire sample ofXXL-N AGNs, breaking the sample into redshift bins of ∆ z = 0 . each. While a small number of obscured AGNsoverlap with the diagnostic region defined by Equa-tion 7, the majority of the heavily absorbed AGNs stilloccupy the same parameter space as the Compton-thinand unobscured AGNs, even in the local ( z < . ) red-shift bin, in stark contrast to the results found with the Swift /BAT AGNs. We find a very similar result whenexamining the L µ m / L . µ m diagnostic ratio. Dueto the fact that at higher redshift the L µ m / L . µ m and L µ m / L . µ m diagnostics do not correspond to thesame wavelength ranges as they do at local redshifts, wethen turned to the luminosity ratio of L µ m /L . µ m .Yet again the heavily absorbed AGNs generally occupythe same parameter space as unobscured AGNs. Thespectral curvature method (Koss et al. 2016) may pro-vide a more effective means of selecting heavily obscuredAGNs at higher redshift in the XXL-N field, and indeedBaronchelli et al. (2017) demonstrated its effectivenessin selecting high redshift ( z > ) CT AGNs in both the Chandra
Deep Field South and the
Chandra
COSMOSlegacy survey.There are a few explanations for why the luminosityratios of the XXL-N AGNs do not as clearly differen-tiate between obscuration levels like that seen for the
Swift /BAT AGNs. For one, there may simply not beenough local redshift XXL-N AGNs for a proper statis-tical comparison to the results found for the
Swift /BATAGNs. Secondly, it is possible that the chosen redshiftbins are not fine enough and we are including AGNsacross too large of redshift ranges. After splitting the z < . bin into five sub-bins, however, we find the sameresult as before: the unobscured and heavily obscuredAGNs coexist within the parameter space.Another explanation lies in the reliability of the re-sults of the X-ray spectral fitting performed in Liu et al.(2016): while the Bayesian statistical framework em-ployed in that work is a powerful method for constrain- ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN
22 m / L )3.02.52.01.51.00.50.0 l o g ( L k e V , O b s . / L m ) UnobscuredLightly obscuredModerately obscuredCompton-thick l o g ( N H / c m ) , t h % - t il e
20 21 22 23 24 25 26log( N H /cm )3.02.52.01.51.00.50.0 l o g ( L k e V , O b s . / L m ) UnobscuredLightly obscuredModerately obscuredCompton-thick
Figure 12. (Left) L X , Obs . / L µ m vs. L µ m / L . µ m logarithmic ratios with each point color-coded to indicate the th percentile log( N H / cm − ) value. (Right) Logarithmic L X , Obs . / L µ m ratio vs. log( N H / cm − ), where the th percentilelog( N H / cm − ) values and error values ( th and th percentiles) are drawn from Liu et al. (2016), and the data points arecolor-coded according to the auxiliary axis to the left. To aid the reader, we use different markers to denote the obscuration binfor each source. Unlike the Swift /BAT AGNs, there is not as clear of relationship between the luminosity ratios and columndensity for the XXL-N AGNs, though this comparison should be viewed with caution because of the large uncertainties onlog( N H / cm − ) and because these AGNs were selected with a softer energy band (0.3–10 keV). ing the spectral properties for AGNs with low counts,our results suggest that the column densities derivedfor the AGNs may still be inaccurate due to the lowcounts acquired. For example, for all AGNs detectedin the z < . bin, the median number of counts de-tected in the epic pn and in the two epic mos detec-tors is only . and . counts, respectively. Further-more, identification of Compton-thick AGNs via XMM-Newton spectroscopy is quite difficult due to the softerX-ray passband (0.3-10 keV) probed with the
XMM-Newton imaging. It can be very difficult to distinguishbetween the scenario in which the source is heavily ob-scured, and the X-ray emission is dominated by repro-cessed radiation from the circumnuclear material, andthat in which the source is unobscured but the emissionis dominated by relativistic reflection from the accre-tion disk when using low signal-to-noise X-ray spectraalone due to the strong model-dependent degeneraciesinvolved (see e.g., Gandhi et al. 2009, Treister et al.2009). Our mid-IR ratio predictions for the highest X-ray column densities can hence be very complementaryto help classify sources in which unobscured and ob-scured reflection models can fit an observed X-ray spec-trum equally well.Future, deeper
XMM-Newton or NuSTAR follow-upobservations could improve upon the current photonstatistics and could provide robust constraints on thecolumn densities for at least the local XXL-N AGNswhich lie within the diagnostic boxes. For now, we iden-tify these systems as candidate heavily obscured or CTAGNs, and we list their spectral properties in Table 8. CONCLUSIONSUsing the well-studied
Swift /BAT sample of ultrahard X-ray selected AGNs, we have presented an analy-sis of the AGN L X , Obs . / L µ m luminosity ratio and twodifferent mid-IR luminosity ratios: L µ m / L . µ m and L µ m / L . µ m . Using the well constrained X-ray (Ricciet al. 2017a) and mid-IR (Ichikawa et al. 2017) prop-erties of the Swift /BAT AGNs, we probed the utilityof these luminosity ratios as tools for inferring line-of-sight column densities and identifying the most heavilyobscured AGNs in the local Universe ( z < . . Wesummarize the results of our analysis as follows:• We have derived expressions relating the col-umn density N H to both the L X , Obs . / L µ m and L µ m / L . µ m luminosity ratios, which are de-fined by Equations 2 and 4. These expressionscan be inverted to give N H as a function of theluminosity ratios. We provide these expressionsagain here: log(N H / cm − ) = 20 + (1 . +0 . − . ) × log (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) log (cid:16) L X , Obs . L µ m (cid:17) + (0 . +0 . − . )( − . +0 . − . ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) and log(N H / cm − ) = 20 + (3 . +1 . − . ) × log (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) log (cid:16) L µ m L . µ m (cid:17) − (0 . +0 . − . )(0 . +0 . − . ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) R. W. Pfeifle et al. • We have demonstrated that unobscured andheavily obscured AGNs tend to exhibit different L X , Obs . / L µ m , L µ m / L . µ m , and L µ m / L . µ m luminosity ratios. All three ofour diagnostic regions (Equations 5, 7, and8) identify (in general) the most heavily ab-sorbed AGNs, with average column densities of log( N H / cm − ) ≥ . for each defined parameterspace. These regions are all (cid:38) % complete and (cid:38) % pure for AGNs with log( N H / cm − ) ≥ .The greatest impurities arise due to AGNs with . (cid:46) log( N H / cm − ) < ; these regions are (cid:38) % pure for AGNs with log( N H / cm − ) (cid:38) . .• While optically star forming systems can fallwithin our diagnostic regions, this contaminationcan be virtually eliminated via mid-IR or X-rayselection criteria. Such selection criteria shouldbe used judiciously to avoid removing non-mid-IRAGNs. These diagnostic regions should not beused to differentiate between AGNs and galaxiesdominated by star formation.• Swift /BAT AGNs which do not dominate the to-tal observed µ m emission tend to exhibit red-der colors and larger X-ray deficits with increasingcolumn density, suggesting that host galaxy con-tributions to at least the mid-IR emission can bea significant factor in the luminosity ratios exam-ined here, particularly in the case of mildly ob-scured and CT AGNs selected with our diagnos-tics. However, it appears that this effect actuallyaids in the identification of CT AGNs, as the hostcontributions result in a larger separation in colorspace between less obscured and more obscuredAGNs.• We find that the selection criteria proposed hereare more reliable at identifying obscured and CTAGNs than the mid-IR and far-IR selection cri-teria proposed by Kilerci-Eser et al. (2020). Weidentify several known obscured and CT AGNs,as well as several candidate CT AGNs, within theIR galaxy catalog of Kilerci Eser & Goto (2018)(see Table 7).• We applied our diagnostics to the XMM-Newton
XXL-N field and found, in contrast to
Swift /BATAGNs, that obscured and unobscured XXL-NAGNs do not appear to exhibit distinctly differentluminosity ratios. This disparity could be due topoor photon statistics or the softer X-ray energiesprobed for the XXL-N AGNs, which could leadto inaccurate column density values. Although, given the small number of z < . XXL-N AGNsand the large errors associated with several N H values, this comparison should be viewed withcaution.Identifying heavily obscured AGNs remains an impor-tant yet difficult task, though the study of such AGNsis an important step in the development of our under-standing of the evolution of AGNs. In a future study, wecould expand our analysis to include diagnostic regionsappropriately modified to differentiate between unob-scured and heavily obscured sources at higher redshift,although it is difficult to speculate at the moment howthe emission ratios may change with redshift, as bothstar formation activity and AGN activity are expectedto increase with redshift.The selection criteria presented here offers a comple-mentary approach to the spectral curvature method de-veloped in Koss et al. (2016), which is very effectiveat selecting heavily obscured AGNs at local- z , with thecaveats that one must already have hard X-ray ( > keV) measurements with NuSTAR or Swift /BAT andthat it is most effective for brighter AGNs. Softer X-raymissions can only be utilized for higher redshift sources( z ∼ ) for which the hard X-ray emission has shiftedinto the rest frame 10–30 keV passband. The diagnos-tics presented here, on the other hand, do not requirehigher energy passbands in order to select local- z sourcesand can take advantage of softer X-ray missions suchas Chandra and
XMM-Newton . The synergy betweenthese two approaches is best summed up by the factthat they select many of the same sources using differ-ent passbands, and that they select heavily obscuredAGNs missed by one another (see Figure 15 in the Ap-pendix), yielding a more complete census of heavily ob-scured
Swift /BAT AGNs overall.The diagnostic regions proposed in this study, aswell as the expressions derived relating N H to the L X , Obs . / L µ m and L µ m / L . µ m luminosity ratios,could be used to differentiate between unobscuredand heavily obscured AGNs in future, large samplesof AGNs, such as those now being detected by theeROSITA all-sky survey (Predehl et al. 2010; Merloniet al. 2012). In particular, the eROSITA survey will pro-vide the first all-sky X-ray imaging survey at energiesup to 10 keV, yielding a highly complementary catalogto those of other all-sky missions, such as WISE . Futureworks could cross-match the
WISE and eROSITA cat-alogs and use the diagnostics presented here to identifymany more cases of CT AGN candidates, select tar-gets for deeper follow-up multiwavelength observations,and to compute the CT fraction for the future sample,all of which will be crucial in the quest to construct a
ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN
Facilities:
Chandra, GALEX, NuSTAR, SDSS,Suzaku, Swift, WISE, XMM-Newton.
Software: odr (Brown et al. 1990; Virtanen et al.2020), pandas (McKinney 2010), scipy (Virtanen et al.2020), numpy (Oliphant 2006; van der Walt et al. 2011;Oliphant 2015), matplotlib Hunter (2007), BXA (Buch-ner et al. 2014) .This publication makes use of data products from the
Wide-field Infrared Survey Explorer
R. W. Pfeifle et al. APPENDIX6.1.
Exploring the Origin of the Scatter in the L µ m / L . µ m vs. N H Correlation
In the process of our analysis, we explored whetherany quality cuts or selection cuts could be applied tothe data to reduce the scatter observed in Panel A ofboth Figures 3 and 4. Our parent sample was dividedinto four sub-samples as shown in Figure 13:• AGN-dominated systems with W – W > . (Stern et al. 2012, blue squares).• AGNs not selected using the aforementioned WISE cut, i.e. W – W < . (red triangles).• AGNs with > spectral counts in the X-rayspectra, which provides a statistically significantnumber of counts to constrain N H in the X-rayspectral fitting analysis (Ricci et al. 2017a, in-verted cyan triangles).• AGNs with observed 2–10 keV luminosities in ex-cess of erg s − (green diamonds).In Figures 13 and 14 we compare these sub-samples forthe L µ m / L . µ m and L µ m / L . µ m luminosity ratios,respectively, and how they correlate with N H . The re-sulting mean values per bin for each different sub-sampleare consistent with the values (the error bars representthe standard deviation of each subsample) originallyfound for the parent sample; interestingly, the observedcorrelation between WISE color and column densityholds for AGNs that satisfy the Stern et al. (2012) crite-rion as well as
AGNs which do not satisfy that criterion.We observe a larger difference in the L µ m / L . µ m ra-tios when moving to higher column densities than for the L µ m / L . µ m ratios. Additionally, as shown previouslyin Figure 4, while there is a large amount of scatter inthe lowest- N H bin of the bottom panel of Figure 14, thisscatter is likely due to the low number of sources withinthat bin, and this still does not overlap the bin probingthe highest obscuring columns. Due to the consistencybetween the results for the sub-samples and that foundfor the parent sample, we do not implement any of thesecuts during our analysis.From Figure 13 and 14, it becomes clear that therea number of WISE
AGNs (W1-W2 > 0.8) within the
Swift /BAT sample which exhibit extremely red mid-IRcolors, with log( L µ m / L . µ m ) > . , suggesting sig-nificant obscuration. We tabulate these WISE
AGNs inTable 9. Indeed, the majority of these AGNs (18/23)are moderately to heavily obscured with log( N H / cm − ) > , though there are a few exceptions, notably: Table 6. L µ m / L . µ m Diagnostic Box Statistics log( N H / cm − ) Completeness Purity ≥ . . +5 . − . . +6 . − . < . . +1 . − . . +6 . − . [23 . , .
0) 16 . +3 . − . . +6 . − . [23 . , .
0) 30 . +6 . − . . +6 . − . [23 . , .
5) 4 . +3 . − . . +3 . − . [22 . , .
0) 5 . +2 . − . . +4 . − . [22 . , .
0) 5 . +3 . − . . +3 . − . [22 . , .
5) 6 . +4 . − . . +3 . − . < . . +0 . − . . +2 . − . Note —A breakdown of the statistics derived from the diag-nostic box developed for the L µ m / L . µ m luminosity ratio(defined by Equation 8 in Section 3) for various N H bins andsub-bins. Columns 1-4: The same as Table 3. • HS0328+0528, an unobscured Seyfert 1.• IRAS05189-2524, a lightly obscured Seyfert 2.• 2MASXJ09172716-6456271, an unobscured Seyfert2.• MCG-1-24-12, a lightly obscured Seyfert 2.• NGC4253, an unobscured Seyfert 1. WISE selection based on the cut defined in Sternet al. (2012) is not, however, a necessarily good methodfor selecting obscured over unobscured AGNs. Aswe discussed in Section 4.3, 40/71 of AGNs withlog( N H / cm − ) > . × cm − in the Swift /BATsample studied here do not meet a color cut of W1-W2> 0.8, reinforcing our choice to not invoke such a cut onthe parent sample.
ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN
20 21 22 23 24 25log( N H /cm )0.00.10.20.30.40.50.6 l o g ( L m / L . m ) Full sampleWISE AGNNon-WISE AGNCounts cut ( > 300 cts)Luminosity cut ( > 10 erg s )
22 m / L )21.021.522.022.523.023.524.0 l o g ( N H / c m ) Full sampleWISE AGNNon-WISE AGNCounts cut ( > 300 cts)Luminosity cut ( > 10 erg s ) Figure 13.
We applied four different cuts to our full
Swift /BAT sample to explore the origin of the scatter ob-served in Figure 3. Here we present two different compar-isons of the derived values for N H from Ricci et al. (2017a)and the L µ m / L . µ m mid-IR ratios, where we have (toppanel) binned by log( N H ) and (bottom panel) binned bythe L µ m / L . µ m ratio. We observe more scatter whenbinning by the mid-IR ratio than we do when binning in-stead by N H , but nonetheless the general trend is the same:we observe increasing mid-IR ratios of L µ m / L . µ m withincreasing column density regardless of the sub-sample.
20 21 22 23 24 25log( N H /cm )0.00.10.20.3 l o g ( L m / L . m ) Full sampleWISE AGNNon-WISE AGNCounts cut ( > 300 cts)Luminosity cut ( > 10 erg s )
12 m / L )20.521.021.522.022.523.023.524.0 l o g ( N H / c m ) Full sampleWISE AGNNon-WISE AGNCounts cut ( > 300 cts)Luminosity cut ( > 10 erg s ) Figure 14.
Analogous to Figure 13 except here we ex-amine the alternative mid-IR diagnostic ratio which de-pends upon L µ m and L . µ m . We observe an increase inthe mid-IR ratio of L µ m / L . µ m with column density aswas observed with L µ m / L . µ m , although with a largeamount of scatter in the N H values for the lowest mid-IRratio bin. However, given that our focus is on the mostheavily obscured sources ( > a few times cm − ) thisscatter is not a concern. R. W. Pfeifle et al.
Table 7.
Mid-IR AGNs from Kilerci Eser & Goto (2018) Selected via Equation 7
AKARI
I.D. RA Dec z Selection log (cid:16) L µ m L . µ m (cid:17) log (cid:16) L X , Obs . L µ m (cid:17) Alternate Obscuration N H Method I.D. Class Ref.0041533+402120 10.473 40.355 0.071 3, 4 0.62 -1.52 Mrk 957 . . . . . .0138053-125210 24.522 -12.87 0.04 1, 2, 3, 4 0.47 -2.47 IRAS 01356-1307 Heavily Obscured 10143576+022059 25.991 2.35 0.017 1, 2, 3, 4 0.35 -2.29 Mrk 573 / UGC 1214 Heavily Obscured 20150029-072549 27.511 -7.43 0.018 1, 2, 3, 4 0.64 -1.44 IRAS 01475-0740 Unobscured or Heavily Obscured 3, 40222435-084305 35.682 -8.719 0.045 1, 2, 3, 4 0.37 -1.72 NGC 905 . . . . . .0325256-060832 51.356 -6.144 0.034 4 0.36 -1.45 Mrk 609 Unobscured a b
8, 70525179-460023 81.325 -46.006 0.042 1, 2, 3, 4 0.42 -1.56 ESO 253-3 Moderately obscured 90742406+651031 115.674 65.177 0.037 1, 2, 3, 4 0.49 -1.58 Mrk 78 Heavily or moderately obscured 10, 7, 110759401+152314 119.917 15.387 0.016 4 0.41 -2.74 UGC 4145 . . . . . .0807411+390015 121.921 39.004 0.023 4 0.83 -2.04 Mrk 622 Heavily Obscured 70810401+481233 122.668 48.209 0.077 3, 4 0.74 -2.11 2MASX J08104028+4812335 Moderately Obscured 10904011+012733 136.004 1.458 0.054 4 0.6 -2.54 IRAS 09014+0139 . . . . . .0935514+612112 143.965 61.353 0.039 1, 2, 3, 4 0.12 -2.28 UGC 5101 Heavily Obscured 7, 121010432+061157 152.681 6.2 0.098 1, 2, 3, 4 0.57 -2.39 2MASS J10104334+0612013 . . . . . .1021428+130655 155.428 13.115 0.076 4 0.92 -2.71 3XMM J102142.6+130654 Unobscured 13, 11034080+600152 158.536 60.031 0.051 1, 2, 3, 4 0.47 -1.97 Mrk 34 Heavily Obscured 141034381+393820 158.661 39.641 0.043 1, 2, 3, 4 0.15 -1.32 7C 103144.10+395402.00 Unobscured 151100183+100255 165.075 10.049 0.036 3, 4 0.68 -2.28 LEDA 200263 Lightly Obscured 161219585-355743 184.996 -35.960 0.058 1, 2, 3, 4 0.22 -2.74 6dFGS gJ121959.0-355735 . . . . . .1307059-234033 196.775 -23.677 0.01 4 0.55 -2.46 NGC 4968 Heavily Obscured 171344421+555316 206.175 55.887 0.037 1, 2, 3, 4 0.89 -1.97 Mrk 273 Moderately obscured 18, 81347044+110626 206.768 11.106 0.023 1, 2, 3, 4 0.5 -1.81 Mrk 1361 . . . . . .1356027+182222 209.012 18.372 0.051 1, 2, 3, 4 0.17 -2.29 Mrk 463 Moderately Obscured 191550415-035314 237.673 -3.888 0.03 1, 2, 3, 4 0.53 -1.97 IRAS 15480-0344 Heavily Obscured 18, 101651053-012747 252.774 -1.463 0.041 4 0.38 -1.58 LEDA 1118057 . . . . . .1847441-630920 281.934 -63.157 0.015 3, 4 0.52 -2.43 IC 4769 Heavily Obscured 101931212-723919 292.839 -72.656 0.062 1, 2, 3, 4 0.52 -2.23 ‘Superantennae’ Unobscured or Heavily Obscured 20, 82019593-523716 304.996 -52.622 0.017 4 0.34 -1.8 IC 4995 Moderately or Heavily Obscured 2, 21, 102059127-520024 314.804 -52.006 0.05 1, 2, 3, 4 0.33 -2.48 ESO 235-26 . . . . . .2316006+253326 349.003 25.557 0.027 1, 2, 3, 4 0.73 -2.23 IC 5298 Moderately Obscured 222351135+201349 357.808 20.230 0.044 4 0.63 -1.37 MCG+03-60-031 . . . . . .
Note —Column 1:
AKARI
ID from Kilerci Eser & Goto (2018). Columns 2-4: right ascension, declination, and redshift of theAGN. Column 5: selection method satisfied by this AGN, where 1, 2, 3, and 4 correspond to the selection criteria in Stern et al.(2012), Jarrett et al. (2011), Assef et al. (2018), and Satyapal et al. (2018). 6-7: log( L µ m / L . µ m ) and log( L X , Obs . / L µ m )luminosity ratios. Column 8: alternative identifier for candidate CT AGN from other literature studies. Column 9: Obscurationclass, assigned based on column density found in the literature. Class labels are adopted from Section 3. Column 9: Reference(s)for column densities (used to classify the objects here): (1) Terashima et al. (2015), (2) Guainazzi et al. (2005), (3) Huang et al.(2011), (4) Brightman & Nandra (2008), (5) LaMassa et al. (2014), (6) Marchesi et al. (2018), (7) Ricci et al. (2017a), (8)Teng et al. (2015), (9) Asmus et al. (2015), (10) Severgnini et al. (2012), (11) Gilli et al. (2010), (12) Oda et al. (2017), (13)Teng & Veilleux (2010), (14) Gandhi et al. (2014), (15) González-Martín (2018), (16) Dutta et al. (2018), (17) LaMassa et al.(2019), (18) Brightman & Nandra (2011), (19) Bianchi et al. (2008), (20) Braito et al. (2009), (21) Noguchi et al. (2009), (22)Torres-Albà et al. (2018). a Despite a lack of broad optical lines, Mrk 609 shows no sign of obscuration at X-ray wavelengths (LaMassa et al. 2014) b Based upon current measurements, it is believed that IRAS 05189-2524 is currently lightly obscured, although it is possible that it mayhave been heavily obscured in the past (Teng et al. 2015).
ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN Table 8.
Low-Redshift Candidate Heavily Obscured AGNs from
XMM -XXL North Selected via Equation 7UXID α δ z F −
10 keV log( N H /cm − ) log( L X , Obs . / L µ m ) log( L µ m / L . µ m )(erg cm − s − )N_96_28 33.3335 -3.48755 0.0754 . × − . +0 . − . -1.6 0.27N_42_10 34.7902 -5.42083 0.0987 . × − . +2 . − . -2.11 0.71N_97_13 35.7049 -5.56736 0.0687 . × − . +2 . − . -2.09 0.33N_35_14 36.0105 -5.22831 0.0843 . × − . +0 . − . -1.48 0.37N_45_48 36.4573 -4.00696 0.0433 . × − . +0 . − . -1.68 0.75N_30_7 36.5185 -4.99197 0.0539 . × − . +0 . − . -2.11 0.11N_113_19 37.3037 -5.18955 0.0736 . × − . +1 . − . -2.09 0.32N_105_14 37.5322 -4.53268 0.0444 . × − . +0 . − . -1.47 0.11 Note —All eight low-redshift ( z < . ) candidate heavily obscured or CT AGNs within the XMM -XXL-N field pulled from theMenzel et al. (2016) and Liu et al. (2016) catalogs selected using the diagnostic box defined by Equation 7. Column 1: uniqueX-ray identification string used by both Menzel et al. (2016) and Liu et al. (2016). Column 2-3: X-ray source right ascension(RA) and declination (Dec) values (uncorrected for systematic offsets) given in degrees. Column 3: redshift. Column 4: HardX-ray 2–10 keV fluxes. Column 5: th -percentile logarithmic line-of-sight column density derived by Liu et al. (2016). Errorbounds are calculated using the th and th -percentile column density values. Column 6-7: values for the log( L X , Obs . / L µ m )and log( L µ m / L . µ m ) luminosity ratios. R. W. Pfeifle et al.
Table 9.
WISE -Selected
Swift /BAT AGNs with log( L µ m / L . µ m ) > 0.5 SWIFT
I.D. RA Dec z log (cid:16) L µ m L . µ m (cid:17) log (cid:16) L X , Obs . L µ m (cid:17) Alternate I.D. log( N H /cm − )SWIFTJ0107.7-1137B 16.9152 -11.65320 0.0475 0.51 -1.10 2MASXJ01073963-1139117 . +0 . − . SWIFTJ0122.8+5003 20.6435 50.05500 0.0204 0.76 -1.44 MCG+8-3-18 . +0 . − . SWIFTJ0308.2-2258 47.0449 -22.96080 0.0360 0.89 -1.78 NGC1229 . +1 . − . SWIFTJ0331.3+0538 52.7174 5.64040 0.0460 0.62 -0.35 HS0328+0528 . +0 . − . SWIFTJ0521.0-2522 80.2561 -25.36260 0.0426 0.51 -1.74 IRAS05189-2524 . +0 . − . SWIFTJ0615.8+7101 93.9015 71.03750 0.0135 0.82 -1.20 Mrk3 . +0 . − . SWIFTJ0656.4-4921 104.0498 -49.33060 0.0410 0.58 -1.79 LEDA478026 . +0 . − . SWIFTJ0743.0+6513 115.6739 65.17710 0.0371 0.66 -1.82 Mrk78 . +0 . − . SWIFTJ0804.2+0507 121.0244 5.11380 0.0135 0.77 -1.09 Mrk1210 . +0 . − . SWIFTJ0843.5+3551 130.9375 35.82830 0.0540 0.58 -1.09 CASG218 . +0 . − . SWIFTJ0917.2-6457 139.3634 -64.94090 0.0860 0.55 -0.07 2MASXJ09172716-6456271 . +0 . − . SWIFTJ0920.8-0805 140.1927 -8.05610 0.0196 0.54 -0.28 MCG-1-24-12 . +0 . − . SWIFTJ1214.3+2933 183.5741 29.52860 0.0632 0.53 -0.94 Was49b . +0 . − . SWIFTJ1218.5+2952 184.6105 29.81290 0.0129 0.56 -0.81 NGC4253 . +0 . − . SWIFTJ1225.8+1240 186.4448 12.66210 0.0084 0.61 -0.77 NGC4388 . +0 . − . SWIFTJ1238.6+0928 189.6810 9.46017 0.0829 0.64 -0.93 SDSSJ123843.43+092736.6 . +0 . − . SWIFTJ1322.2-1641 200.6019 -16.72860 0.0165 0.63 -1.86 MCG-3-34-64 . +0 . − . SWIFTJ1717.1-6249 259.2478 -62.82060 0.0037 0.51 -1.16 NGC6300 . +0 . − . SWIFTJ1800.3+6637 270.0304 66.61510 0.0265 0.94 -1.62 NGC6552 . +0 . − . SWIFTJ2052.0-5704 313.0098 -57.06880 0.0114 0.73 -1.45 IC5063 . +0 . − . SWIFTJ2207.3+1013 331.7582 10.23340 0.0267 0.71 -1.70 UGC11910 . +0 . − . SWIFTJ2304.9+1220 346.2361 12.32290 0.0079 0.82 -2.65 NGC7479 . +0 . − . SWIFTJ2343.9+0537 355.9982 5.64000 0.0560 0.52 -0.62 LEDA3092070 . +0 . − . Note — WISE -selected
Swift /BAT AGN ( W − W > . , Stern et al. 2012) which exhibit mid-IR ratios of log( L µ m / L . µ m ) > . . Column 1: SWIFT
I.D. number. Column 2-4: Right ascension, declination, and redshift. Column 5-6: log( L X , Obs . / L µ m )and log( L µ m / L . µ m ) luminosity ratios. Column 7: Alternative I.D. drawn from Ricci et al. (2017a). Column 8: Columndensity derived in Ricci et al. (2017a). ASS-XXIII: Mid-Infrared Diagnostic for Absorption in AGN log( N H /cm )0.00.20.40.60.8 Sp e c t r a l C u r v a t u r e CT (Koss et al. 2016)AGNs selected via Eq. 4
Figure 15.
A comparison between the X-ray and mid-IRselection criteria introduced in this work and the spectralcurvature method from Koss et al. (2016). The spectral cur-vature of each point was calculated using observations from
NuSTAR , whereas the column densities come from Ricciet al. (2017a). Only AGNs with spectral curvature errorsof < . are included in this plot. The dashed grey linerepresents a spectral curvature value of 0.4, above which anAGN is considered to be CT. Red points are AGNs selectedusing the diagnostic box defined in Equation 7, whereas bluepoints are not selected via Equation 7. The spectral cur-vature method and the diagnostic defined in this work findseveral of the same sources, and in fact these two approachesalso find CT AGNs missed by one another: a few CT AGNsfall below the 0.4 spectral curvature cutoff, yet the diag-nostic presented here selects them, meanwhile the spectralcurvature method recovered a CT AGN missed by our newdiagnostic. There is one outlier which is selected by Equa-tion 7 and exhibits a relatively high spectral curvature, yet itpossesses a column density of only ∼ cm − . This source,NGC 1365, is a well known variable absorber and has gonethrough massive absorption transitions. Recent high signal-to-noise observations have shown that the column densityremains substantial, above cm − (e.g. Risaliti et al.2009a,b,c; Maiolino et al. 2010; Walton et al. 2010; Brenne-man et al. 2013) and occasionally increases to the extent ofbecoming CT (Risaliti et al. 2005). It is possible that themid-IR emission is tracing a higher absorption period seenin past observations. R. W. Pfeifle et al.
REFERENCES
Aihara, H., Allende Prieto, C., An, D., et al. 2011, ApJS,193, 29, doi: 10.1088/0067-0049/193/2/29Akylas, A., Georgakakis, A., Georgantopoulos, I.,Brightman, M., & Nandra, K. 2012, A&A, 546, A98,doi: 10.1051/0004-6361/201219387Alexander, D. M., & Hickox, R. C. 2012, NewAR, 56, 93,doi: 10.1016/j.newar.2011.11.003Alexander, D. M., Chary, R. R., Pope, A., et al. 2008, ApJ,687, 835, doi: 10.1086/591928Ananna, T. T., Treister, E., Urry, C. M., et al. 2019, ApJ,871, 240, doi: 10.3847/1538-4357/aafb77Antonucci, R. 1993, ARA&A, 31, 473,doi: 10.1146/annurev.aa.31.090193.002353Asmus, D., Gandhi, P., Hönig, S. F., Smette, A., & Duschl,W. J. 2015, MNRAS, 454, 766,doi: 10.1093/mnras/stv1950Asmus, D., Hönig, S. F., & Gandhi, P. 2016, ApJ, 822, 109,doi: 10.3847/0004-637X/822/2/109Assef, R. J., Stern, D., Noirot, G., et al. 2018, ApJS, 234,23, doi: 10.3847/1538-4365/aaa00aBaldwin, J. A., Phillips, M. M., & Terlevich, R. 1981,PASP, 93, 5, doi: 10.1086/130766Baron, D., & Netzer, H. 2019a, MNRAS, 482, 3915,doi: 10.1093/mnras/sty2935—. 2019b, MNRAS, 486, 4290, doi: 10.1093/mnras/stz1070Baronchelli, L., Koss, M., Schawinski, K., et al. 2017,MNRAS, 471, 364, doi: 10.1093/mnras/stx1561Bauer, F. E., Arévalo, P., Walton, D. J., et al. 2015, ApJ,812, 116, doi: 10.1088/0004-637X/812/2/116Baumgartner, W. H., Tueller, J., Markwardt, C. B., et al.2013, ApJS, 207, 19, doi: 10.1088/0067-0049/207/2/19Bianchi, S., Chiaberge, M., Piconcelli, E., Guainazzi, M., &Matt, G. 2008, MNRAS, 386, 105,doi: 10.1111/j.1365-2966.2008.13078.xBlecha, L., Snyder, G. F., Satyapal, S., & Ellison, S. L.2018, ArXiv e-prints. https://arxiv.org/abs/1711.02094Braito, V., Reeves, J. N., Della Ceca, R., et al. 2009, A&A,504, 53, doi: 10.1051/0004-6361/200811516Brenneman, L. W., Risaliti, G., Elvis, M., & Nardini, E.2013, MNRAS, 429, 2662, doi: 10.1093/mnras/sts555Brightman, M., & Nandra, K. 2008, MNRAS, 390, 1241,doi: 10.1111/j.1365-2966.2008.13841.x—. 2011, MNRAS, 413, 1206,doi: 10.1111/j.1365-2966.2011.18207.x Brown, P. J., Fuller, W. A., et al. 1990, Statistical analysisof measurement error models and applications:Proceedings of the AMS-IMS-SIAM joint summerresearch conference held June 10-16, 1989, with supportfrom the National Science Foundation and the US ArmyResearch Office, Vol. 112 (American Mathematical Soc.)Buchner, J., Georgakakis, A., Nandra, K., et al. 2014,A&A, 564, A125, doi: 10.1051/0004-6361/201322971—. 2015, ApJ, 802, 89, doi: 10.1088/0004-637X/802/2/89Burlon, D., Ajello, M., Greiner, J., et al. 2011, ApJ, 728,58, doi: 10.1088/0004-637X/728/1/58Cameron, E. 2011, PASA, 28, 128, doi: 10.1071/AS10046Dutta, R., Srianand, R., & Gupta, N. 2018, MNRAS, 480,947, doi: 10.1093/mnras/sty1872Elvis, M., Maccacaro, T., Wilson, A. S., et al. 1978,MNRAS, 183, 129, doi: 10.1093/mnras/183.2.129Fabian, A. C., Zoghbi, A., Ross, R. R., et al. 2009, Nature,459, 540, doi: 10.1038/nature08007Ferland, G. J., Porter, R. L., van Hoof, P. A. M., et al.2013, RMxAA, 49, 137. https://arxiv.org/abs/1302.4485Ferland, G. J., Chatzikos, M., Guzmán, F., et al. 2017,RMxAA, 53, 385. https://arxiv.org/abs/1705.10877Ferrarese, L., & Merritt, D. 2000, ApJL, 539, L9,doi: 10.1086/312838Gandhi, P., Horst, H., Smette, A., et al. 2009, A&A, 502,457, doi: 10.1051/0004-6361/200811368Gandhi, P., Lansbury, G. B., Alexander, D. M., et al. 2014,ApJ, 792, 117, doi: 10.1088/0004-637X/792/2/117Georgantopoulos, I., Rovilos, E., Akylas, A., et al. 2011,A&A, 534, A23, doi: 10.1051/0004-6361/201117400Gilli, R., Comastri, A., & Hasinger, G. 2007, A&A, 463, 79,doi: 10.1051/0004-6361:20066334Gilli, R., Vignali, C., Mignoli, M., et al. 2010, A&A, 519,A92, doi: 10.1051/0004-6361/201014039Gonzalez-Delgado, R. M., Perez, E., Diaz, A. I., et al. 1995,ApJ, 439, 604, doi: 10.1086/175201González-Martín, O. 2018, ApJ, 858, 2,doi: 10.3847/1538-4357/aab7ecGoulding, A. D., Alexander, D. M., Mullaney, J. R., et al.2011, MNRAS, 411, 1231,doi: 10.1111/j.1365-2966.2010.17755.xGuainazzi, M., Matt, G., & Perola, G. C. 2005, A&A, 444,119, doi: 10.1051/0004-6361:20053643Haardt, F., & Maraschi, L. 1991, ApJL, 380, L51,doi: 10.1086/186171—. 1993, ApJ, 413, 507, doi: 10.1086/173020Herrero-Illana, R., Pérez-Torres, M. Á., Alonso-Herrero, A.,et al. 2014, ApJ, 786, 156,doi: 10.1088/0004-637X/786/2/156