arX

iv1

905

0432

0v1

[as

tro-

phG

A]

10

May

201

9Draft version May 14 2019Preprint typeset using LATEX style emulateapj v 121611

HOT DUST-OBSCURED GALAXIES WITH EXCESS BLUE LIGHT

RJ Assef1 M Brightman2 DJ Walton3 D Stern4 FE Bauer567 AW Blain8 T Dıaz-Santos1PRM Eisenhardt4 RC Hickox9 HD Jun10 A Psychogyios11 C-W Tsai12 JW Wu12

Draft version May 14 2019

ABSTRACT

Hot Dust-Obscured Galaxies (Hot DOGs) are among the most luminous galaxies in the UniversePowered by highly obscured possibly Compton-thick active galactic nuclei (AGNs) Hot DOGs arecharacterized by SEDs that are very red in the mid-IR yet dominated by the host galaxy stellar emis-sion in the UV and optical An earlier study identified a sub-sample of Hot DOGs with significantlyenhanced UV emission One target W0204ndash0506 was studied in detail and based on Chandra obser-vations it was concluded that the enhanced emission was most likely due to either extreme unobscuredstar-formation (SFR gt 1000 M⊙ yrminus1) or to light from the highly obscured AGN scattered by gas ordust into our line of sight Here we present a follow-up study of W0204ndash0506 as well as two more HotDOGs with excess UV emission For the two new objects we obtained ChandraACIS-S observationsand for all three targets we obtained HSTWFC3 F555W and F160W imaging We conclude that theexcess UV emission is primarily dominated by light from the central highly obscured hyper-luminousAGN that has been scattered into our line of sight We cannot rule out however that star-formationmay significantly contribute to the UV excess of W0204ndash0506Keywords galaxies active mdash galaxies evolution mdash galaxies high-redshift mdash quasars general mdash

infrared galaxies

1 INTRODUCTION

Hot Dust-Obscured Galaxies (Hot DOGsEisenhardt et al 2012 Wu et al 2012) are some ofthe most luminous galaxies in the Universe withbolometric luminosities Lbol gt 1013 L⊙ and a signif-icant fraction with Lbol gt 1014 L⊙ (Wu et al 2012Tsai et al 2015) Discovered by the Wide-field InfraredSurvey Explorer (WISE Wright et al 2010) HotDOGs are characterized by very red mid-IR colorsand spectral energy distributions (SEDs) that peakat rest-frame sim 20microm This implies that Hot DOGsare powered by highly obscured hyper-luminous AGNthat dominate the SED from the mid- to the far-IR(Eisenhardt et al 2012 Wu et al 2012 2014 Fan et al

1 Nucleo de Astronomıa de la Facultad de Ingenierıa y Cien-cias Universidad Diego Portales Av Ejercito Libertador 441Santiago Chile

2 Cahill Center for Astrophysics California Institute of Tech-nology 1216 East California Boulevard Pasadena CA 91125USA

3 Institute of Astronomy University of Cambridge MadingleyRoad Cambridge CB3 0HA UK

4 Jet Propulsion Laboratory California Institute of Technol-ogy 4800 Oak Grove Drive Pasadena CA 91109 USA

5 Instituto de Astrofısica Facultad de Fısica Pontificia Uni-versidad Catolica de Chile 306 Santiago 22 Chile

6 Millennium Institute of Astrophysics (MAS) NuncioMonsenor Sotero Sanz 100 Providencia Santiago Chile

7 Space Science Institute 4750 Walnut Street Suite 205 Boul-der Colorado 80301

8 Physics amp Astronomy University of Leicester 1 UniversityRoad Leicester LE1 7RH UK

9 Department of Physics and Astronomy Dartmouth College6127 Wilder Laboratory Hanover NH 03755 USA

10 School of Physics Korea Institute for Advanced Study 85Hoegiro Dongdaemun-gu Seoul 02455 Republic of Korea

11 Department of Physics University of Crete 71003 Herak-lion Greece

12 National Astronomical Observatories Chinese Academy ofSciences 20A Datun Road Chaoyang District Beijing 100012Peoplersquos Republic of China

2016a Dıaz-Santos et al 2016 Tsai et al 2018) Asexpected from their luminosities Hot DOGs are rarewith one object every 31plusmn4 deg2 Yet their numberdensity is comparable to that of similarly luminousunobscured quasars (Assef et al 2015) and of heavilyreddened type 1 quasars (Banerji et al 2015)X-ray studies have shown that the obscuration of the

central engine in Hot DOGs is very high with col-umn densities ranging from somewhat below to abovethe Compton-thick limit (ie NH gt 15 times 1024 cmminus2

Stern et al 2014 Piconcelli et al 2015 Assef et al 2016Ricci et al 2017 Vito et al 2018) As the AGN emissionis highly obscured the host galaxy is observable at rest-frame UV optical and near-IR wavelengths A study oftheir SEDs by Assef et al (2015) showed that their stel-lar masses as derived from their rest-frame near-IR lumi-nosities imply that either the super-massive black holes(SMBHs) are accreting well above the Eddington limitor that their SMBH masses (MBH) are well above the lo-cal relations betweenMBH and the mass of the spheroidalcomponent of the host galaxy (see eg Magorrian et al1998 Bennert et al 2011) Indeed recent results byWu et al (2018) and Tsai et al (2018) suggest that HotDOGs are radiating at or above the Eddington limitwhich in turn suggests that Hot DOGs are likely experi-encing strong AGN feedback that could easily affect thewhole host galaxy and its immediate environment In-deed Dıaz-Santos et al (2016) presented a study of the[C ii] 1577microm emission line in the highest luminosity HotDOG and possibly the most luminous galaxy knownWISEA J22460756ndash0526349 (W2246ndash0526 Tsai et al2015) and determined based on the emission-line kine-matics that the central gas of the host galaxy is likelyundergoing an isotropic outflow event Further ionizedgas outflow signatures have been observed in the opticalnarrow emission lines of some other Hot DOGs (Wu et al2018 Jun et al in prep) supporting the presence of

2

strong AGN feedback in the ISM of these targetsAssef et al (2015 also see Eisenhardt et al 2012

Tsai et al 2015 2018) showed that the UV through mid-IR SED of the majority of Hot DOGs (specifically ldquoW12ndashdropsrdquo with z gt 1) can be well modeled as a combina-tion of a star-forming galaxy that dominates the opti-calUV emission and a luminous obscured AGN thatdominates the mid-IR SED and the bolometric lumi-nosity of the system However this is not the case forall Hot DOGs In a later work Assef et al (2016 A16hereafter) presented a small sample of eight Hot DOGswhose opticalUV emission is not well modeled by a star-forming galaxy but instead needs a second unobscuredAGN component that is only sim1 as luminous as theobscured component A16 posited that the SED couldbe explained by three different scenarios i) that theUVoptical emission is dominated by leaked or scatteredlight from the hyper-luminous highly obscured AGNii) that the system is a dual quasar with a more lumi-nous highly obscured quasar and a less luminous un-obscured one and iii) that the system is undergoing anextreme star-formation event with little dust obscurationsuch that the broad-band UVoptical SED is similar tothat of an AGNOne of these objects WISEA J02044613ndash0506408

(W0204ndash0506 hereafter) was serendipitously observedby the Chandra X-ray Observatory as part of theLarge-Area Lyman Alpha survey (LALA Rhodes et al2000) A16 studied this object in detail usingthese observations along with broad-band SED andoptical spectroscopic observations A16 determinedthat the X-ray spectrum of W0204ndash0506 is consis-tent with a single hyper-luminous highly absorbedAGN (logL2minus10 keVerg sminus1 = 449+086

minus014 NH =

63+81minus21 times 1023 cmminus2) and highly inconsistent with a

secondary unobscured AGN with the luminosity nec-essary to explain the opticalUV emission InsteadA16 found that the UVoptical continuum was bet-ter explained by a starburst with a star-formation rateamp 1000 M⊙ yrminus1 or by scattered light from the hyper-luminous highly obscured central engine While star-formation rates (SFRs) amp 1000 M⊙ yrminus1 are routinelyfound through far-IRsub-mm observations of highly ob-scured systems such as SMGs and ULIRGs rates abovesim 300 M⊙ yrminus1 have never been observed throughUVoptical wavelengths in Lyman break galaxies whichhave the strongest UVoptical star-formations measured(Barger et al 2014) Due to the large SFR needed to ex-plain the opticalUV SED of this object as a starburstA16 favored the scattered AGN-light scenarioIn this paper we present Hubble Space Telescope

(HST) observations of W0204ndash0506 to further exploreits nature and we explore in detail two more Blue-Excess Hot DOGs (BHDs) WISE J02205212+0137116(W0220+0137 hereafter) and WISE J01160141-0505040(W0116ndash0505 hereafter) using HST and Chandra obser-vations as well as optical spectroscopy and broad-bandUV through mid-IR SEDs In sect2 we present the samplestudied here as well as the different observations avail-able for each target while in sect3 we discuss the modelingof the Chandra X-ray observations In sect4 we presenta detailed discussion of the source of the excess blueemission analyzing each possible scenario in light of

the available observations Our conclusions are sum-marized in sect5 Throughout the article all magnitudesare presented in their natural system unless otherwisestated namely AB for ugriz and Vega for all the restWe assume a concordance flat ΛCDM cosmology withH0 = 70 km sminus1 Mpcminus1 ΩΛ = 07 and ΩM = 03 Forall quantities derived from X-ray spectra we quote 90confidence interval while for all other quantities we quote683 confidence intervals instead

2 SAMPLE AND OBSERVATIONS

21 Blue-Excess Hot DOGs

A16 identified 8 BHDs from a sample of 36 HotDOGs with W4lt72 mag spectroscopic redshifts z gt 1and ugriz modelMag13 photometry in the SDSS DR12database with SN gt 3 in at least one of the SDSSbands This spectroscopic sample is biased towards op-tical emission and after considering the selection effectsA16 estimated BHDs could comprise as much as 8 ofthe Hot DOG population with W4lt72 mag althoughmost likely a smaller fraction when considering fainterW4 fluxesTo select this sample A16 started by modeling the

SEDs of the aforementioned 36 Hot DOGs using thegalaxy and AGN SED templates and modeling algo-rithm of Assef et al (2010) following the prescriptionpresented by Assef et al (2015) In short the broad-bandSED of any given object is modeled as a linear non-negative combination of four empirically derived SEDtemplates an ldquoErdquo template which resembles the SEDof an old stellar population an ldquoSbcrdquo template whichresembles the SED of an intermediately star-forminggalaxy an ldquoImrdquo template which resembles a local star-burst galaxy and a type 1 AGN template We also fitfor the reddening of the AGN template parametrizedby E(B minus V ) assuming RV = 31 and a reddening lawthat follows that of the SMC at short wavelengths butthat of the Milky Way at longer wavelengths A singleIGM absorption strength is also fit for all templates whenneeded (see Assef et al 2010 2015 A16 for details) Us-ing this approach A16 modeled the SED of each objectin the following broad bands the ugriz SDSS DR12modelMag photometry SpitzerIRAC [36] and [45] pho-tometry from Griffith et al (2012) and the WISE W3and W4 photometry from the WISE All-Sky Data Re-lease (Cutri et al 2012) Additionally whenever possi-ble A16 used the J Ks and deeper r-band imaging pre-sented by Assef et al (2015) For the three sources con-sidered in this article the deeper r-band imaging was ob-tained using the 41m Southern Astrophysical ResearchTelescope (SOAR) with the SOAR Optical Imager (SOI)For W0116ndash0505 images were obtained with an exposuretime of 3times600 s on the night of UT 2013 August 28 Forthe other two sources the images were obtained on UT2011 November 20 with exposure times of 3times500 s forW0204ndash0506 and of 2times500 s for W0220+0137 In allcases the images were reduced following standard pro-cedures and the photometric calibration was performedby comparing bright stars in each field with their respec-tive SDSS magnitudes The details of the NIR imagingcan be found in Assef et al (2015) All magnitudes areshown in Table 1

13 httpwwwsdssorgdr12algorithmsmagnitudesmag_model

3

Table 1Photometric Data

WISE ID J01160141ndash0505040 J02044613ndash0506408 J02205212+0137116

SDSS u 23571plusmn0685 23004plusmn0600 23470plusmn0587SDSS g 21464plusmn0054 22660plusmn0166 21779plusmn0059F555W 21679plusmn0019 22441plusmn0047 21772plusmn0018SDSS r 21383plusmn0054 22488plusmn0234 21841plusmn0086SOI r 21515plusmn0078 22357plusmn0166 21803plusmn0047SDSS i 21740plusmn0094 21797plusmn0175 22060plusmn0132SDSS z 21368plusmn0257 22026plusmn0667 21607plusmn0273J middot middot middot 20768plusmn0216 20790plusmn0149F160W 20648plusmn0007 20390plusmn0007 21077plusmn0010Ks middot middot middot middot middot middot 18604plusmn0117W1 17130plusmn0184 17343plusmn0115 17875plusmn0225[36] 16800plusmn0040 17182plusmn0056 17722plusmn0091[45] 15725plusmn0021 16340plusmn0033 16806plusmn0051W2 15564plusmn0156 16103plusmn0158 16575plusmn0253W3 10213plusmn0059 10245plusmn0056 10512plusmn0075W4 7014plusmn0084 7062plusmn0090 7076plusmn0092

A16 found that the approach described above was notable to accurately model the UVoptical emission for afraction of their sample which were significantly bluerthan allowed by the SED templates They identifiedeight objects for which an additional secondary AGNcomponent with independent normalization and redden-ing provided a significant improvement in χ2 to the best-fit SED model A16 presented a detailed study of theproperties of one of these targets W0204ndash0506 Herewe extend this analysis to an additional one of theseeight targets W0220+0137 as well as to another verysimilar target W0116ndash0505 The W1=1713plusmn018 magof W0116ndash0505 is slightly brighter than the formal HotDOG selection limit (W1gt174 Eisenhardt et al 2012)and hence it was excluded from the final list presented byA16 despite meeting all other selection criteria We findthere is only a 27 probability that the improvementin χ2 by the secondary AGN component is spurious forthis source A16 argued that these probabilities are likelyoverestimated and hence conservative as the F-test usedto estimate them does not take into account the con-straints provided by the non-negative requirement of thelinear combination of templates for the best-fit modelThe broad band SEDs as well as best-fit SED models

of the three targets are shown in Figure 1 We note thatthe SED of W0204ndash0506 differs slightly from that pre-sented by A16 as for consistency with the other twosources the SED presented here only uses the SDSSDR12 bands in the UVoptical instead of the deeperimaging of Finkelstein et al (2007) Table 2 shows foreach target the best-fit E(B minus V ) to both the primaryand secondary AGN components The table also showsthe reddening-corrected monochromatic luminosities at6microm L6microm calculated from the template fit to each AGNcomponent In all three targets the secondary AGN hasa much lower ( 1) monochromatic luminosity at 6micromThe uncertainties for the parameters shown in Table 2

have been estimated using a Monte-Carlo method follow-ing a similar prescription to that used by A16 For eachobject we first apply a scaling factor to the photomet-ric uncertainties such that the best-fit SED model has areduced χ2 (χ2

ν) of 1 We then create 1000 realizationsof the observed SED of each object by re-sampling itsphotometry according to the aforementioned scaled un-certainties and assuming a Gaussian distribution We fit

Figure 1 UV through mid-IR SEDs of the three BHDs discussedin this study The green solid points show the observed flux den-sities in the photometric bands discussed in sect21 The solid blackline shows the best-fit SED model to the photometric data pointsthat consists of a non-negative linear combination of a primaryluminous obscured AGN (dashed magenta line) a secondary lessluminous unobscured or mildly obscured AGN (solid blue line)an old stellar population (dotted red line) an intermediate stel-lar population (dashed green line) and a young stellar population(cyan dotted-dashed line) The open triangles show the predictedflux density for each photometric band based on the best-fit SEDmodel For each object we indicate also the redshift and the prob-ability PRan that the improvement in χ2 gained from adding thesecondary AGN component is spurious

each of the 1000 simulated SEDs and compile the dis-tribution of each parameter We assign the uncertaintiesto the 683 intervals of these distributions around thevalues of the best-fit model

22 Optical Spectra

We obtained optical spectra of W0116ndash0505 andW0220+0137 using the Multiple Mirror Telescope

4

Table 2Best-fit SED Parameters

Primary AGN Secondary AGNObject Redshift logL6microm E(B minus V ) logL6microm E(B minus V ) Pran

(erg sminus1) (mag) (erg sminus1) (mag) (10minus2)

W0116minus0505 3173 4724+017minus011

424+271minus123

4518+004minus003

000+002minus000

27

W0204minus0506 2100 4687+003minus008

1000+174minus206

4498+004minus022

010+000minus005

45

W0220+0137 3122 4733+016minus016

733+267minus232

4496+008minus011

000+002minus000

02

(MMT) spectrograph on the night of UT 2010 December4 We used the blue channel with the 1200 linesmmgrating and obtained 3times600 s exposures on each targetthrough a longslit with a width of 15primeprime The optical spec-trum of W0204ndash0506 was obtained using the GMOS-Sspectrograph on the Gemini South telescope on UT 2011November 27 using a longslit with a width of 15primeprime aswell These observations have been previously presentedby A16 and we refer the reader to that study for furtherdetails on these observations All three spectra were re-duced using standard toolsThe optical spectra of W0116ndash0505 W0204ndash0506 and

W0220+0137 are presented in Figures 2 3 and 4 respec-tively In addition to the MMT spectra we also showlower SN spectra obtained by SDSS for W0116ndash0505and W0220+0137 on UT 2013 October 3 and UT 2015September 13 respectively The difference between theequivalent widths of the emission lines suggests eitherthat there is a small amount of variability in the contin-uum andor the emission lines or that the emission linescome from a region that has a different spatial exten-sion than the continuum such that differences in extrac-tion apertures can account for the discrepancy Unlikethe MMT spectra the SDSS observations were obtainedthrough a much larger 3primeprime fiber The spectra of W0116ndash0505 and W0220+0137 show clear broad high ioniza-tion emission lines that are typically observed in type 1quasars Single Gaussian fits to the C iv emission linefollowing the prescription of Assef et al (2011 and ref-erences therein) to fit the continuum and define the spec-tral region on which to fit the emission line have FWHMof approximately 2800 km sminus1 and 3500 km sminus1 respec-tively for W0116ndash0505 andW0220+0137 Based on theseemission lines we measure a redshift of z = 3173plusmn 0002for W0116ndash0505 and z = 3122plusmn0002 for W0220+0137In particular both spectra show blended Lyβ and Oviemission features W0204ndash0506 is at a significantly lowerredshift of z = 2100plusmn0002 and hence we cannot deter-mine if these emission lines are present in the spectrumas they fall shortwards of the atmospheric UV cut-off

23 HST Observations

A joint program between Chandra and HST was ap-proved during Chandra Cycle 17 (PID 17700696) to ob-tain HST imaging in two bands of all three targetsand obtain ChandraACIS-S observations of W0116ndash0505 and W0220+0137 These targets were selectedfor having some of the clearest blue excess emission interms of the χ2 improvement and for having some ofthe highest expected count rates in ACIS-S The archivalChandraACIS-I observations for W0204ndash0506 presentedby A16 are sufficient to accomplish our science goals sono further observations were requested In this section

Figure 2 Optical spectrum of W0116ndash0505 obtained with theMMT spectrograph (black) and by SDSS (gray)

Figure 3 Optical spectrum of W0204ndash0506 obtained with theGMOS-S instrument at the Gemini South Observatory

Figure 4 Optical spectrum of W0220+0137 obtained with theMMT spectrograph (black) and by SDSS (gray)

5

we focus on the HST observations while the Chandraobservations are described in the next sectionImaging observations were obtained using the WFC3

camera onboardHST of all three BHD targets in both theF555W and the F160W bands Each target was observedduring one orbit with two exposures in the F555W bandfollowed by three exposures in the F160W band Theexposure times in the F555W band were 738 s and 626 sfor each image for W0116ndash0505 and W0204ndash0506 and735 s and 625 s each for W0220+0137 All exposure timesin the F160W band were 353 s For the F160W band weuse the reduced images provided by the HST archiveCutouts of 5primeprimetimes5primeprime centered on the F160W coordinatesof the target are shown in the middle panels of Figure5For the F555W band we do not use the archive pro-

vided reductions as the pipeline cosmic ray rejection issignificantly compromised by the acquisition of only twoimages Instead we took the fully-reduced single framesprovided by the archive including the charge transfer ef-ficiency correction and used the LACOSMIC algorithm(van Dokkum 2001) to remove cosmic rays We thenused those cosmic-ray corrected images to continue withthe pipeline processing and combine the frames Wealigned the F555W image to the F160W image usingstars detected in both bands The final images are shownin the left panels of Figure 5 Table 1 presents the 4primeprime di-ameter aperture magnitudes measured in each band foreach objectThe right panel of Figure 5 show an RGB composite

of the images created using the Lupton et al (2004) al-gorithm as implemented through the astropy v20114

function make lupton rgb We assigned the F555W im-age to the blue channel and the F160W image to the redchannel while leaving the green channel empty Beforeproducing the RGB composite we convolve the F555Wimage with a Gaussian kernel to match its PSF to thatof the F160W image We assume that the PSFs of bothimages are well modeled by Gaussian PSFs with the re-spective FWHM as provided by the WFC3 documenta-tion15 namely 0067primeprime for the F555W channel and 0148primeprime

for the F160W channel Hence the Gaussian kernel usedon the F555W image corresponds to a Gaussian functionwith FWHM2

kernel = FWHM2F160W minus FWHM2

F555WFigure 6 shows the radial profiles of each source com-

pared to that of a fiducial point source with a GaussianPSF The emission of the three objects is clearly resolvedin both bands For W0116ndash0505 and W0220+0137 themorphologies seem to be broadly undisturbed in bothbands with the F160W emission having a larger extentand a higher luminosity The emission peaks in bothbands are spatially co-located W0204ndash0506 is on theother hand quite clearly disturbed with the F160Wmorphology (rest-frame 5200A) suggestive of a recent in-teraction The F555W emission (rest-frame 1800A) ispatchy reminiscent of a starburst We discuss the impli-cations of this UV morphology further in sect4To more quantitatively assess the morphology of these

sources we have measured different coefficients com-monly used in the literature Specifically we follow

14httpwwwastropyorg

15httpwwwstscieduhstwfc3ins_performancegroundcomponentsfilters

httpwwwstscieduhstwfc3documentshandbookscurrentIHBc06_uvis07html391868

Lotz et al (2004) to measure the Gini M20 and A co-efficients (Lotz et al 2004 and references therein) TheGini coefficient (Abraham et al 2003) measures how uni-formly distributed is the light among the pixels of agalaxy in an image such that Gini is 0 if all pixels havea uniform brightness and is 1 if all brightness is concen-trated in a single pixel The M20 coefficient measures thesecond order moment of the brightest 20 of the flux ofthe galaxy as compared to the total second order mo-ment Mtot The moments are computed around a cen-ter chosen to minimize Mtot The A coefficient measuresthe rotational asymmetry of a galaxy by subtracting animage of the galaxy rotated by 180 degrees The rota-tional center is chosen to minimize A For further detailson these coefficients we refer the reader to Lotz et al(2004) and Conselice (2014)We start by subtracting the background using

SExtractor (v2195 Bertin amp Arnouts 1996) as wellas obtaining the centroid of each object in each bandWe then compute the Petrosian radius (Petrosian 1976)and generate the segmentation map following Lotz et al(2004) and finally proceed to measure the coefficientsdiscussed above The values and uncertainties of theGini M20 and A coefficients for each object in each bandare shown in Table 3 We estimate the uncertainties ineach parameter through a Monte Carlo approach For agiven object in a given band we use the uncertainty ineach pixel to generate 1000 resampled images assumingGaussian statistics We then repeat the measurementin each resampled image following the procedure out-lined above We assign the measurement error to be thedispersion of the coefficient measurements in the 1000resampled imagesRecently Farrah et al (2017) measured these coeffi-

cients for 12 Hot DOGs using HSTWFC3 images in theF160W Using the boundaries proposed by Lotz et al(2004) in the GinindashA plane and by Lotz et al (2008)in the GinindashM20 plane Farrah et al (2017) determinedthat while Hot DOGs have a high merger fraction (sim80) this fraction is consistent with that found for mas-sive galaxies at z sim 2 leading them to conclude thatHot DOGs are not preferentially associated with merg-ers These results are generally consistent with those ofFan et al (2016b) who also found a high merger fraction(62 plusmn 14) among Hot DOGs as well as with those re-cently presented by Dıaz-Santos et al (2018) who foundevidence with sub-mm ALMA sim 200 microm imaging of atriple major merger in the the most luminous Hot DOGW2246ndash0526 If we adopt the same boundaries used byFarrah et al (2017) to classify our sources according totheir F160W morphologies and noting that all caveatsidentified by Farrah et al (2017) also apply here we findthat the host galaxies of W0116ndash0505 and W0220+0137are not consistent with mergers but instead are classifiedas undisturbed early-type galaxies For W0204ndash0506 onthe other hand we find that its host galaxy morphologyis best classified as an on-going merger These results areconsistent with our visual characterization of the hostgalaxies

24 Chandra Observations

We have obtained ChandraACIS-S observations oftwo of our targets W0116ndash0505 and W0220+0137 (pro-posal ID 17700696) Each object was observed with a

4

Table 2Best-fit SED Parameters

Primary AGN Secondary AGNObject Redshift logL6microm E(B minus V ) logL6microm E(B minus V ) Pran

(erg sminus1) (mag) (erg sminus1) (mag) (10minus2)

W0116minus0505 3173 4724+017minus011

424+271minus123

4518+004minus003

000+002minus000

27

W0204minus0506 2100 4687+003minus008

1000+174minus206

4498+004minus022

010+000minus005

45

W0220+0137 3122 4733+016minus016

733+267minus232

4496+008minus011

000+002minus000

02

(MMT) spectrograph on the night of UT 2010 December4 We used the blue channel with the 1200 linesmmgrating and obtained 3times600 s exposures on each targetthrough a longslit with a width of 15primeprime The optical spec-trum of W0204ndash0506 was obtained using the GMOS-Sspectrograph on the Gemini South telescope on UT 2011November 27 using a longslit with a width of 15primeprime aswell These observations have been previously presentedby A16 and we refer the reader to that study for furtherdetails on these observations All three spectra were re-duced using standard toolsThe optical spectra of W0116ndash0505 W0204ndash0506 and

W0220+0137 are presented in Figures 2 3 and 4 respec-tively In addition to the MMT spectra we also showlower SN spectra obtained by SDSS for W0116ndash0505and W0220+0137 on UT 2013 October 3 and UT 2015September 13 respectively The difference between theequivalent widths of the emission lines suggests eitherthat there is a small amount of variability in the contin-uum andor the emission lines or that the emission linescome from a region that has a different spatial exten-sion than the continuum such that differences in extrac-tion apertures can account for the discrepancy Unlikethe MMT spectra the SDSS observations were obtainedthrough a much larger 3primeprime fiber The spectra of W0116ndash0505 and W0220+0137 show clear broad high ioniza-tion emission lines that are typically observed in type 1quasars Single Gaussian fits to the C iv emission linefollowing the prescription of Assef et al (2011 and ref-erences therein) to fit the continuum and define the spec-tral region on which to fit the emission line have FWHMof approximately 2800 km sminus1 and 3500 km sminus1 respec-tively for W0116ndash0505 andW0220+0137 Based on theseemission lines we measure a redshift of z = 3173plusmn 0002for W0116ndash0505 and z = 3122plusmn0002 for W0220+0137In particular both spectra show blended Lyβ and Oviemission features W0204ndash0506 is at a significantly lowerredshift of z = 2100plusmn0002 and hence we cannot deter-mine if these emission lines are present in the spectrumas they fall shortwards of the atmospheric UV cut-off

23 HST Observations

A joint program between Chandra and HST was ap-proved during Chandra Cycle 17 (PID 17700696) to ob-tain HST imaging in two bands of all three targetsand obtain ChandraACIS-S observations of W0116ndash0505 and W0220+0137 These targets were selectedfor having some of the clearest blue excess emission interms of the χ2 improvement and for having some ofthe highest expected count rates in ACIS-S The archivalChandraACIS-I observations for W0204ndash0506 presentedby A16 are sufficient to accomplish our science goals sono further observations were requested In this section

Figure 2 Optical spectrum of W0116ndash0505 obtained with theMMT spectrograph (black) and by SDSS (gray)

Figure 3 Optical spectrum of W0204ndash0506 obtained with theGMOS-S instrument at the Gemini South Observatory

Figure 4 Optical spectrum of W0220+0137 obtained with theMMT spectrograph (black) and by SDSS (gray)

5

we focus on the HST observations while the Chandraobservations are described in the next sectionImaging observations were obtained using the WFC3

camera onboardHST of all three BHD targets in both theF555W and the F160W bands Each target was observedduring one orbit with two exposures in the F555W bandfollowed by three exposures in the F160W band Theexposure times in the F555W band were 738 s and 626 sfor each image for W0116ndash0505 and W0204ndash0506 and735 s and 625 s each for W0220+0137 All exposure timesin the F160W band were 353 s For the F160W band weuse the reduced images provided by the HST archiveCutouts of 5primeprimetimes5primeprime centered on the F160W coordinatesof the target are shown in the middle panels of Figure5For the F555W band we do not use the archive pro-

vided reductions as the pipeline cosmic ray rejection issignificantly compromised by the acquisition of only twoimages Instead we took the fully-reduced single framesprovided by the archive including the charge transfer ef-ficiency correction and used the LACOSMIC algorithm(van Dokkum 2001) to remove cosmic rays We thenused those cosmic-ray corrected images to continue withthe pipeline processing and combine the frames Wealigned the F555W image to the F160W image usingstars detected in both bands The final images are shownin the left panels of Figure 5 Table 1 presents the 4primeprime di-ameter aperture magnitudes measured in each band foreach objectThe right panel of Figure 5 show an RGB composite

of the images created using the Lupton et al (2004) al-gorithm as implemented through the astropy v20114

function make lupton rgb We assigned the F555W im-age to the blue channel and the F160W image to the redchannel while leaving the green channel empty Beforeproducing the RGB composite we convolve the F555Wimage with a Gaussian kernel to match its PSF to thatof the F160W image We assume that the PSFs of bothimages are well modeled by Gaussian PSFs with the re-spective FWHM as provided by the WFC3 documenta-tion15 namely 0067primeprime for the F555W channel and 0148primeprime

for the F160W channel Hence the Gaussian kernel usedon the F555W image corresponds to a Gaussian functionwith FWHM2

kernel = FWHM2F160W minus FWHM2

F555WFigure 6 shows the radial profiles of each source com-

pared to that of a fiducial point source with a GaussianPSF The emission of the three objects is clearly resolvedin both bands For W0116ndash0505 and W0220+0137 themorphologies seem to be broadly undisturbed in bothbands with the F160W emission having a larger extentand a higher luminosity The emission peaks in bothbands are spatially co-located W0204ndash0506 is on theother hand quite clearly disturbed with the F160Wmorphology (rest-frame 5200A) suggestive of a recent in-teraction The F555W emission (rest-frame 1800A) ispatchy reminiscent of a starburst We discuss the impli-cations of this UV morphology further in sect4To more quantitatively assess the morphology of these

sources we have measured different coefficients com-monly used in the literature Specifically we follow

14httpwwwastropyorg

15httpwwwstscieduhstwfc3ins_performancegroundcomponentsfilters

httpwwwstscieduhstwfc3documentshandbookscurrentIHBc06_uvis07html391868

Lotz et al (2004) to measure the Gini M20 and A co-efficients (Lotz et al 2004 and references therein) TheGini coefficient (Abraham et al 2003) measures how uni-formly distributed is the light among the pixels of agalaxy in an image such that Gini is 0 if all pixels havea uniform brightness and is 1 if all brightness is concen-trated in a single pixel The M20 coefficient measures thesecond order moment of the brightest 20 of the flux ofthe galaxy as compared to the total second order mo-ment Mtot The moments are computed around a cen-ter chosen to minimize Mtot The A coefficient measuresthe rotational asymmetry of a galaxy by subtracting animage of the galaxy rotated by 180 degrees The rota-tional center is chosen to minimize A For further detailson these coefficients we refer the reader to Lotz et al(2004) and Conselice (2014)We start by subtracting the background using

SExtractor (v2195 Bertin amp Arnouts 1996) as wellas obtaining the centroid of each object in each bandWe then compute the Petrosian radius (Petrosian 1976)and generate the segmentation map following Lotz et al(2004) and finally proceed to measure the coefficientsdiscussed above The values and uncertainties of theGini M20 and A coefficients for each object in each bandare shown in Table 3 We estimate the uncertainties ineach parameter through a Monte Carlo approach For agiven object in a given band we use the uncertainty ineach pixel to generate 1000 resampled images assumingGaussian statistics We then repeat the measurementin each resampled image following the procedure out-lined above We assign the measurement error to be thedispersion of the coefficient measurements in the 1000resampled imagesRecently Farrah et al (2017) measured these coeffi-

cients for 12 Hot DOGs using HSTWFC3 images in theF160W Using the boundaries proposed by Lotz et al(2004) in the GinindashA plane and by Lotz et al (2008)in the GinindashM20 plane Farrah et al (2017) determinedthat while Hot DOGs have a high merger fraction (sim80) this fraction is consistent with that found for mas-sive galaxies at z sim 2 leading them to conclude thatHot DOGs are not preferentially associated with merg-ers These results are generally consistent with those ofFan et al (2016b) who also found a high merger fraction(62 plusmn 14) among Hot DOGs as well as with those re-cently presented by Dıaz-Santos et al (2018) who foundevidence with sub-mm ALMA sim 200 microm imaging of atriple major merger in the the most luminous Hot DOGW2246ndash0526 If we adopt the same boundaries used byFarrah et al (2017) to classify our sources according totheir F160W morphologies and noting that all caveatsidentified by Farrah et al (2017) also apply here we findthat the host galaxies of W0116ndash0505 and W0220+0137are not consistent with mergers but instead are classifiedas undisturbed early-type galaxies For W0204ndash0506 onthe other hand we find that its host galaxy morphologyis best classified as an on-going merger These results areconsistent with our visual characterization of the hostgalaxies

24 Chandra Observations

We have obtained ChandraACIS-S observations oftwo of our targets W0116ndash0505 and W0220+0137 (pro-posal ID 17700696) Each object was observed with a

5

we focus on the HST observations while the Chandraobservations are described in the next sectionImaging observations were obtained using the WFC3

camera onboardHST of all three BHD targets in both theF555W and the F160W bands Each target was observedduring one orbit with two exposures in the F555W bandfollowed by three exposures in the F160W band Theexposure times in the F555W band were 738 s and 626 sfor each image for W0116ndash0505 and W0204ndash0506 and735 s and 625 s each for W0220+0137 All exposure timesin the F160W band were 353 s For the F160W band weuse the reduced images provided by the HST archiveCutouts of 5primeprimetimes5primeprime centered on the F160W coordinatesof the target are shown in the middle panels of Figure5For the F555W band we do not use the archive pro-

vided reductions as the pipeline cosmic ray rejection issignificantly compromised by the acquisition of only twoimages Instead we took the fully-reduced single framesprovided by the archive including the charge transfer ef-ficiency correction and used the LACOSMIC algorithm(van Dokkum 2001) to remove cosmic rays We thenused those cosmic-ray corrected images to continue withthe pipeline processing and combine the frames Wealigned the F555W image to the F160W image usingstars detected in both bands The final images are shownin the left panels of Figure 5 Table 1 presents the 4primeprime di-ameter aperture magnitudes measured in each band foreach objectThe right panel of Figure 5 show an RGB composite

of the images created using the Lupton et al (2004) al-gorithm as implemented through the astropy v20114

function make lupton rgb We assigned the F555W im-age to the blue channel and the F160W image to the redchannel while leaving the green channel empty Beforeproducing the RGB composite we convolve the F555Wimage with a Gaussian kernel to match its PSF to thatof the F160W image We assume that the PSFs of bothimages are well modeled by Gaussian PSFs with the re-spective FWHM as provided by the WFC3 documenta-tion15 namely 0067primeprime for the F555W channel and 0148primeprime

for the F160W channel Hence the Gaussian kernel usedon the F555W image corresponds to a Gaussian functionwith FWHM2

kernel = FWHM2F160W minus FWHM2

F555WFigure 6 shows the radial profiles of each source com-

pared to that of a fiducial point source with a GaussianPSF The emission of the three objects is clearly resolvedin both bands For W0116ndash0505 and W0220+0137 themorphologies seem to be broadly undisturbed in bothbands with the F160W emission having a larger extentand a higher luminosity The emission peaks in bothbands are spatially co-located W0204ndash0506 is on theother hand quite clearly disturbed with the F160Wmorphology (rest-frame 5200A) suggestive of a recent in-teraction The F555W emission (rest-frame 1800A) ispatchy reminiscent of a starburst We discuss the impli-cations of this UV morphology further in sect4To more quantitatively assess the morphology of these

sources we have measured different coefficients com-monly used in the literature Specifically we follow

14httpwwwastropyorg

15httpwwwstscieduhstwfc3ins_performancegroundcomponentsfilters

httpwwwstscieduhstwfc3documentshandbookscurrentIHBc06_uvis07html391868

Lotz et al (2004) to measure the Gini M20 and A co-efficients (Lotz et al 2004 and references therein) TheGini coefficient (Abraham et al 2003) measures how uni-formly distributed is the light among the pixels of agalaxy in an image such that Gini is 0 if all pixels havea uniform brightness and is 1 if all brightness is concen-trated in a single pixel The M20 coefficient measures thesecond order moment of the brightest 20 of the flux ofthe galaxy as compared to the total second order mo-ment Mtot The moments are computed around a cen-ter chosen to minimize Mtot The A coefficient measuresthe rotational asymmetry of a galaxy by subtracting animage of the galaxy rotated by 180 degrees The rota-tional center is chosen to minimize A For further detailson these coefficients we refer the reader to Lotz et al(2004) and Conselice (2014)We start by subtracting the background using

SExtractor (v2195 Bertin amp Arnouts 1996) as wellas obtaining the centroid of each object in each bandWe then compute the Petrosian radius (Petrosian 1976)and generate the segmentation map following Lotz et al(2004) and finally proceed to measure the coefficientsdiscussed above The values and uncertainties of theGini M20 and A coefficients for each object in each bandare shown in Table 3 We estimate the uncertainties ineach parameter through a Monte Carlo approach For agiven object in a given band we use the uncertainty ineach pixel to generate 1000 resampled images assumingGaussian statistics We then repeat the measurementin each resampled image following the procedure out-lined above We assign the measurement error to be thedispersion of the coefficient measurements in the 1000resampled imagesRecently Farrah et al (2017) measured these coeffi-

cients for 12 Hot DOGs using HSTWFC3 images in theF160W Using the boundaries proposed by Lotz et al(2004) in the GinindashA plane and by Lotz et al (2008)in the GinindashM20 plane Farrah et al (2017) determinedthat while Hot DOGs have a high merger fraction (sim80) this fraction is consistent with that found for mas-sive galaxies at z sim 2 leading them to conclude thatHot DOGs are not preferentially associated with merg-ers These results are generally consistent with those ofFan et al (2016b) who also found a high merger fraction(62 plusmn 14) among Hot DOGs as well as with those re-cently presented by Dıaz-Santos et al (2018) who foundevidence with sub-mm ALMA sim 200 microm imaging of atriple major merger in the the most luminous Hot DOGW2246ndash0526 If we adopt the same boundaries used byFarrah et al (2017) to classify our sources according totheir F160W morphologies and noting that all caveatsidentified by Farrah et al (2017) also apply here we findthat the host galaxies of W0116ndash0505 and W0220+0137are not consistent with mergers but instead are classifiedas undisturbed early-type galaxies For W0204ndash0506 onthe other hand we find that its host galaxy morphologyis best classified as an on-going merger These results areconsistent with our visual characterization of the hostgalaxies

24 Chandra Observations

We have obtained ChandraACIS-S observations oftwo of our targets W0116ndash0505 and W0220+0137 (pro-posal ID 17700696) Each object was observed with a

6

W0116-0505 F555W

N

E 1

PSF FWHM

F160W

PSF FWHM

W0204-0506 F555W

N

E 1

PSF FWHM

F160W

PSF FWHM

W0220+0137 F555W

N

E 1

PSF FWHM

F160W

PSF FWHM

Figure 5 HSTWFC3 images of W0116ndash0505 (top row) W0204ndash0506 (middle row) and W0220+0137 (bottom row) in the F555W (leftpanels) and F160W (middle panels) bands The right panels show a color-composite where the F160W band has been mapped to red andthe F555W band has been mapped to blue and we have matched the PSF of the F555W band to that of F160W Each panel shows a5primeprimetimes5primeprime region centered on the F160W centroid of each target The magenta circle in the F555W image of W0204ndash0506 shows the brightestUV clump

Table 3Morphology of Blue Excess Hot DOGs

Source Band Gini M20 A

W0116ndash0505 F555W 0499plusmn0003 ndash200plusmn016 0133plusmn0010F160W 0527plusmn0008 ndash208plusmn002 0116plusmn0049

W0204ndash0506 F555W 0529plusmn0016 ndash119plusmn014 0599plusmn0027F160W 0633plusmn0011 ndash081plusmn002 0278plusmn0010

W0220+0137 F555W 0491plusmn0004 ndash177plusmn009 0112plusmn0009F160W 0559plusmn0016 ndash216plusmn004 0172plusmn0016

7

Figure 6 Radial profile of the flux surface density Σ in theF555W (left panel) and F160W (right panel) bands of all threesources namely W0116ndash0505 (solid red lines) W0204ndash0506 (solidblue lines) and W0220+0137 (solid green lines) The dashed graylines show the radial profile of a Gaussian PSF with the appropriateFWHM for each band All profiles are shown from a minimumradius of 1 pixel and all profiles have been normalized to Σ = 1 atthat pixel

total exposure time of 70 ks W0116ndash0505 was observedcontinuously while the observations of W0220+0137were split into one 30 ks and two 20 ks visits spreadthroughout seven days It is worth noting that these ob-servations have previously been presented by Vito et al(2018) in the context of a larger sample of Hot DOGs ob-served in X-rays They find both sources are heavily ab-sorbed at those wavelengths Goulding et al (2018) an-alyzed the observations for W0220+0137 as well but inthe context of a large sample of Extremely Red Quasars(ERQs) and also found the source to be heavily absorbedat X-ray energies qualitatively consistent with the restof the ERQ population analyzed Here we analyze thedata following the approach of A16 who analyzed thearchival ChandraACIS-I observations of W0204ndash0506We use ciao v47 to analyze these data The spec-

tral data products including the source and backgroundspectra and the response files were created using thespecextract tool Source events were extracted fromcircular regions with 2primeprime radii centered on the sourcewhile background events were extracted from annuliwith inner and outer radii of 3 and 6primeprime respectivelyFor W0220+0137 the spectral products from the threeobservations were combined into one using the toolcombine spectra We use the heasoft tool grppha togroup the spectra with a minimum of one count per binAfter subtracting the background 74 counts are de-

tected for W0116ndash0505 and only 18 for W0220+0137Figures 7 and 8 show their respective unfolded spec-tra For reference we also show the ACIS-I spectrumof W0204ndash0506 in Figure 9 which had a significantlylonger exposure time of 160 ks The shape of all threespectra differ significantly from that of an unabsorbedpower-law suggesting the emission is dominated by ahighly obscured AGN as expected from the SED model-ing presented in sect21 In the next section we model thesespectra and discuss their implications for the nature ofthe BHDs

3 X-RAY DATA MODELING

The X-ray spectra of W0116ndash0505 and W0220+0137are clearly hard implying the emission is most likely

Figure 7 (Top panel) X-ray spectrum of W0116ndash0505 obtainedusing ChandraACIS-S (see sect24 for details) The solid black lineshows the best-fit absorbed AGN model to the spectrum as de-scribed in sect3 The dashed-gray line shows the emission expectedfor a second unobscured AGN in the system powering the ob-served UVoptical emission (Bottom panel) The data points showthe ratio between the observed spectrum and the best-fit model

Figure 8 Same as Fig 7 but for the ChandraACIS-S spectrumof W0220+0137

8

Figure 9 Same as Fig 7 but for the ChandraACIS-I spectrumof W0204ndash0506 Adapted from Fig 4 of A16

dominated by a highly obscured AGN To better con-strain the properties of the obscured AGN we fitthe emission of both objects using the models ofBrightman amp Nandra (2011) following the same ap-proach as in A16 These models predict the X-ray spec-trum as observed through an optically thick mediumwith a toroidal geometry as posited by the AGN uni-fied scheme The models employ Monte-Carlo techniquesto simulate the transfer of X-ray photons through theoptically-thick neutral medium self-consistently includ-ing the effects of photoelectric absorption Compton scat-tering and fluorescence from Fe K amongst other ele-ments Treating these effects self consistently rather thanseparately has the advantage of reducing the number offree parameters and of gaining constraints on the spec-tral parameters It is therefore particularly useful forlow count spectra such as those we are fitting here Wetherefore carry out the parameter estimation by mini-mizing the Cash statistic (Cash 1979) modified throughthe W-statistic provided by XSPEC16 to account for thesubtracted background Also following the approach ofA16 we require the photon index Γ to be ge 16 as it ispoorly constrained by our data and lower values are onlyappropriate for low Eddington ratios In practice thefitting procedure we used allows values of Γ in the range16ndash30 and values of NH in the range 1020minus 1026 cmminus2Figures 7 and 8 show the best-fit models to the

spectra of W0116ndash0505 and W0220+0137 respectivelyThe best-fit absorbed AGN model to W0116ndash0505has an absorption column density of neutral hydro-gen of NH = 12+10

minus07 times 1024 cmminus2 a photon-index of

Γ = 19+07minus03 and an absorption-corrected luminosity of

logL2minus10 keVerg sminus1 = 4563+058minus024 The best-fit model

16httpsheasarcgsfcnasagovxanaduxspecmanualXSappendixStatisticshtml

has a Cash statistic of C = 563 for ν = 65 degrees offreedom For W0220+0137 the best-fit model has NH =32 times 1024 cmminus2 Γ = 25 and logL2minus10 keVerg sminus1 =4554+030

minus216 with C = 80 and ν = 15 Note that be-cause of the low number of counts for W0220+0137 the90 confidence uncertainties in the best-fit Γ and NH

are determined by the boundaries of the model Thesame is true for the lower-bound of the best-fit Γ inW0116ndash0505 The best-fit values of NH and L2minus10 keV

are consistent within the uncertainties with those foundby Vito et al (2018) for both sources The best-fit valuesfor W0220+0137 are also consistent within the (large)error bars with those found by Goulding et al (2018)For W0204ndash0506 A16 found that the best-fit absorbedAGN has NH = 063+081

minus021 times 1024 cmminus2 Γ = 16+08minus00 and

logL2minus10 keVerg sminus1 = 449+086minus014 with C = 6608 and

ν = 77The spectra of all three objects are likely dominated

by a luminous AGN with very high absorption Inthe case of W0116ndash0505 and W0220+0137 the ab-sorption is consistent with the objects being Compton-thick (ie NrmH gt 15 times 1024 cmminus2) This isin qualitative agreement with the SED modeling pre-sented in sect21 From the SED model of each ob-ject we can estimate the rest-frame intrinsic (ie ob-scuration corrected) specific luminosity at 6microm L6micromwhich has been shown to be well correlated with theL2minus10 keV X-ray luminosity by a number of authors(Fiore et al 2009 Gandhi et al 2009 Bauer et al 2010Mateos et al 2015 Stern 2015 Chen et al 2017) Weuse the best-fit relation of Stern (2015) between L6microm

and L2minus10 keV which accurately traces this relation upto very high L6microm and is hence most appropriate forour targets From the L6microm of the most luminous andobscured AGN component of W0116ndash0505 this relationpredicts logLPredicted

2minus10 keVerg sminus1 = 4553 plusmn 062 which isin excellent agreement with the luminosity of the best-fit model to the X-ray data of logL2minus10 keVerg sminus1 =4563+058

minus024 For W0220+0137 we also find excellent

agreement with logLPredicted2minus10 keVerg sminus1 = 4558 plusmn 062

and logL2minus10 keVerg sminus1 = 4554+030minus216 A16 nominally

found a good agreement as well for W0204ndash0506 as theyestimated logLPredicted

2minus10 keVerg sminus1 = 4536plusmn037 and found

logL2minus10 keVerg sminus1 = 449+086minus014 from the best-fit X-ray

model However when jointly considering this with thebest-fit and expected absorption their Figure 5 suggestsW0204ndash0506 may be somewhat X-ray weakFrom the SED modeling we also have an estimate

of the amount of dust that is obscuring the luminousAGN that dominates in both the mid-IR and the X-rays Comparing to the column densities of neutral hy-drogen constrained by the modeling of X-ray spectrawe find dust-to-gas ratios of E(B minus V )NH = 35 plusmn

20 times 10minus24 cm2 mag for W0116ndash0505 where the un-certainty corresponds to the 683 confidence intervaland has been derived for simplicity assuming Gaussianstatistics For W0220+0137 we find E(B minus V )NH =23 times 10minus24 cm2 mag As NH is not constrained atthe 90 level within the model boundaries we cannotderive a meaningful confidence interval For W0204ndash0506 A16 found a larger ratio of E(B minus V )NH =154 plusmn 126 times 10minus23 cm2 mag For comparison the me-

7

Figure 6 Radial profile of the flux surface density Σ in theF555W (left panel) and F160W (right panel) bands of all threesources namely W0116ndash0505 (solid red lines) W0204ndash0506 (solidblue lines) and W0220+0137 (solid green lines) The dashed graylines show the radial profile of a Gaussian PSF with the appropriateFWHM for each band All profiles are shown from a minimumradius of 1 pixel and all profiles have been normalized to Σ = 1 atthat pixel

total exposure time of 70 ks W0116ndash0505 was observedcontinuously while the observations of W0220+0137were split into one 30 ks and two 20 ks visits spreadthroughout seven days It is worth noting that these ob-servations have previously been presented by Vito et al(2018) in the context of a larger sample of Hot DOGs ob-served in X-rays They find both sources are heavily ab-sorbed at those wavelengths Goulding et al (2018) an-alyzed the observations for W0220+0137 as well but inthe context of a large sample of Extremely Red Quasars(ERQs) and also found the source to be heavily absorbedat X-ray energies qualitatively consistent with the restof the ERQ population analyzed Here we analyze thedata following the approach of A16 who analyzed thearchival ChandraACIS-I observations of W0204ndash0506We use ciao v47 to analyze these data The spec-

tral data products including the source and backgroundspectra and the response files were created using thespecextract tool Source events were extracted fromcircular regions with 2primeprime radii centered on the sourcewhile background events were extracted from annuliwith inner and outer radii of 3 and 6primeprime respectivelyFor W0220+0137 the spectral products from the threeobservations were combined into one using the toolcombine spectra We use the heasoft tool grppha togroup the spectra with a minimum of one count per binAfter subtracting the background 74 counts are de-

tected for W0116ndash0505 and only 18 for W0220+0137Figures 7 and 8 show their respective unfolded spec-tra For reference we also show the ACIS-I spectrumof W0204ndash0506 in Figure 9 which had a significantlylonger exposure time of 160 ks The shape of all threespectra differ significantly from that of an unabsorbedpower-law suggesting the emission is dominated by ahighly obscured AGN as expected from the SED model-ing presented in sect21 In the next section we model thesespectra and discuss their implications for the nature ofthe BHDs

3 X-RAY DATA MODELING

The X-ray spectra of W0116ndash0505 and W0220+0137are clearly hard implying the emission is most likely

Figure 7 (Top panel) X-ray spectrum of W0116ndash0505 obtainedusing ChandraACIS-S (see sect24 for details) The solid black lineshows the best-fit absorbed AGN model to the spectrum as de-scribed in sect3 The dashed-gray line shows the emission expectedfor a second unobscured AGN in the system powering the ob-served UVoptical emission (Bottom panel) The data points showthe ratio between the observed spectrum and the best-fit model

Figure 8 Same as Fig 7 but for the ChandraACIS-S spectrumof W0220+0137

8

Figure 9 Same as Fig 7 but for the ChandraACIS-I spectrumof W0204ndash0506 Adapted from Fig 4 of A16

dominated by a highly obscured AGN To better con-strain the properties of the obscured AGN we fitthe emission of both objects using the models ofBrightman amp Nandra (2011) following the same ap-proach as in A16 These models predict the X-ray spec-trum as observed through an optically thick mediumwith a toroidal geometry as posited by the AGN uni-fied scheme The models employ Monte-Carlo techniquesto simulate the transfer of X-ray photons through theoptically-thick neutral medium self-consistently includ-ing the effects of photoelectric absorption Compton scat-tering and fluorescence from Fe K amongst other ele-ments Treating these effects self consistently rather thanseparately has the advantage of reducing the number offree parameters and of gaining constraints on the spec-tral parameters It is therefore particularly useful forlow count spectra such as those we are fitting here Wetherefore carry out the parameter estimation by mini-mizing the Cash statistic (Cash 1979) modified throughthe W-statistic provided by XSPEC16 to account for thesubtracted background Also following the approach ofA16 we require the photon index Γ to be ge 16 as it ispoorly constrained by our data and lower values are onlyappropriate for low Eddington ratios In practice thefitting procedure we used allows values of Γ in the range16ndash30 and values of NH in the range 1020minus 1026 cmminus2Figures 7 and 8 show the best-fit models to the

spectra of W0116ndash0505 and W0220+0137 respectivelyThe best-fit absorbed AGN model to W0116ndash0505has an absorption column density of neutral hydro-gen of NH = 12+10

minus07 times 1024 cmminus2 a photon-index of

Γ = 19+07minus03 and an absorption-corrected luminosity of

logL2minus10 keVerg sminus1 = 4563+058minus024 The best-fit model

16httpsheasarcgsfcnasagovxanaduxspecmanualXSappendixStatisticshtml

has a Cash statistic of C = 563 for ν = 65 degrees offreedom For W0220+0137 the best-fit model has NH =32 times 1024 cmminus2 Γ = 25 and logL2minus10 keVerg sminus1 =4554+030

minus216 with C = 80 and ν = 15 Note that be-cause of the low number of counts for W0220+0137 the90 confidence uncertainties in the best-fit Γ and NH

are determined by the boundaries of the model Thesame is true for the lower-bound of the best-fit Γ inW0116ndash0505 The best-fit values of NH and L2minus10 keV

are consistent within the uncertainties with those foundby Vito et al (2018) for both sources The best-fit valuesfor W0220+0137 are also consistent within the (large)error bars with those found by Goulding et al (2018)For W0204ndash0506 A16 found that the best-fit absorbedAGN has NH = 063+081

minus021 times 1024 cmminus2 Γ = 16+08minus00 and

logL2minus10 keVerg sminus1 = 449+086minus014 with C = 6608 and

ν = 77The spectra of all three objects are likely dominated

by a luminous AGN with very high absorption Inthe case of W0116ndash0505 and W0220+0137 the ab-sorption is consistent with the objects being Compton-thick (ie NrmH gt 15 times 1024 cmminus2) This isin qualitative agreement with the SED modeling pre-sented in sect21 From the SED model of each ob-ject we can estimate the rest-frame intrinsic (ie ob-scuration corrected) specific luminosity at 6microm L6micromwhich has been shown to be well correlated with theL2minus10 keV X-ray luminosity by a number of authors(Fiore et al 2009 Gandhi et al 2009 Bauer et al 2010Mateos et al 2015 Stern 2015 Chen et al 2017) Weuse the best-fit relation of Stern (2015) between L6microm

and L2minus10 keV which accurately traces this relation upto very high L6microm and is hence most appropriate forour targets From the L6microm of the most luminous andobscured AGN component of W0116ndash0505 this relationpredicts logLPredicted

2minus10 keVerg sminus1 = 4553 plusmn 062 which isin excellent agreement with the luminosity of the best-fit model to the X-ray data of logL2minus10 keVerg sminus1 =4563+058

minus024 For W0220+0137 we also find excellent

agreement with logLPredicted2minus10 keVerg sminus1 = 4558 plusmn 062

and logL2minus10 keVerg sminus1 = 4554+030minus216 A16 nominally

found a good agreement as well for W0204ndash0506 as theyestimated logLPredicted

2minus10 keVerg sminus1 = 4536plusmn037 and found

logL2minus10 keVerg sminus1 = 449+086minus014 from the best-fit X-ray

model However when jointly considering this with thebest-fit and expected absorption their Figure 5 suggestsW0204ndash0506 may be somewhat X-ray weakFrom the SED modeling we also have an estimate

of the amount of dust that is obscuring the luminousAGN that dominates in both the mid-IR and the X-rays Comparing to the column densities of neutral hy-drogen constrained by the modeling of X-ray spectrawe find dust-to-gas ratios of E(B minus V )NH = 35 plusmn

20 times 10minus24 cm2 mag for W0116ndash0505 where the un-certainty corresponds to the 683 confidence intervaland has been derived for simplicity assuming Gaussianstatistics For W0220+0137 we find E(B minus V )NH =23 times 10minus24 cm2 mag As NH is not constrained atthe 90 level within the model boundaries we cannotderive a meaningful confidence interval For W0204ndash0506 A16 found a larger ratio of E(B minus V )NH =154 plusmn 126 times 10minus23 cm2 mag For comparison the me-

8

Figure 9 Same as Fig 7 but for the ChandraACIS-I spectrumof W0204ndash0506 Adapted from Fig 4 of A16

dominated by a highly obscured AGN To better con-strain the properties of the obscured AGN we fitthe emission of both objects using the models ofBrightman amp Nandra (2011) following the same ap-proach as in A16 These models predict the X-ray spec-trum as observed through an optically thick mediumwith a toroidal geometry as posited by the AGN uni-fied scheme The models employ Monte-Carlo techniquesto simulate the transfer of X-ray photons through theoptically-thick neutral medium self-consistently includ-ing the effects of photoelectric absorption Compton scat-tering and fluorescence from Fe K amongst other ele-ments Treating these effects self consistently rather thanseparately has the advantage of reducing the number offree parameters and of gaining constraints on the spec-tral parameters It is therefore particularly useful forlow count spectra such as those we are fitting here Wetherefore carry out the parameter estimation by mini-mizing the Cash statistic (Cash 1979) modified throughthe W-statistic provided by XSPEC16 to account for thesubtracted background Also following the approach ofA16 we require the photon index Γ to be ge 16 as it ispoorly constrained by our data and lower values are onlyappropriate for low Eddington ratios In practice thefitting procedure we used allows values of Γ in the range16ndash30 and values of NH in the range 1020minus 1026 cmminus2Figures 7 and 8 show the best-fit models to the

spectra of W0116ndash0505 and W0220+0137 respectivelyThe best-fit absorbed AGN model to W0116ndash0505has an absorption column density of neutral hydro-gen of NH = 12+10

minus07 times 1024 cmminus2 a photon-index of

Γ = 19+07minus03 and an absorption-corrected luminosity of

logL2minus10 keVerg sminus1 = 4563+058minus024 The best-fit model

16httpsheasarcgsfcnasagovxanaduxspecmanualXSappendixStatisticshtml

has a Cash statistic of C = 563 for ν = 65 degrees offreedom For W0220+0137 the best-fit model has NH =32 times 1024 cmminus2 Γ = 25 and logL2minus10 keVerg sminus1 =4554+030

minus216 with C = 80 and ν = 15 Note that be-cause of the low number of counts for W0220+0137 the90 confidence uncertainties in the best-fit Γ and NH

are determined by the boundaries of the model Thesame is true for the lower-bound of the best-fit Γ inW0116ndash0505 The best-fit values of NH and L2minus10 keV

are consistent within the uncertainties with those foundby Vito et al (2018) for both sources The best-fit valuesfor W0220+0137 are also consistent within the (large)error bars with those found by Goulding et al (2018)For W0204ndash0506 A16 found that the best-fit absorbedAGN has NH = 063+081

minus021 times 1024 cmminus2 Γ = 16+08minus00 and

logL2minus10 keVerg sminus1 = 449+086minus014 with C = 6608 and

ν = 77The spectra of all three objects are likely dominated

by a luminous AGN with very high absorption Inthe case of W0116ndash0505 and W0220+0137 the ab-sorption is consistent with the objects being Compton-thick (ie NrmH gt 15 times 1024 cmminus2) This isin qualitative agreement with the SED modeling pre-sented in sect21 From the SED model of each ob-ject we can estimate the rest-frame intrinsic (ie ob-scuration corrected) specific luminosity at 6microm L6micromwhich has been shown to be well correlated with theL2minus10 keV X-ray luminosity by a number of authors(Fiore et al 2009 Gandhi et al 2009 Bauer et al 2010Mateos et al 2015 Stern 2015 Chen et al 2017) Weuse the best-fit relation of Stern (2015) between L6microm

and L2minus10 keV which accurately traces this relation upto very high L6microm and is hence most appropriate forour targets From the L6microm of the most luminous andobscured AGN component of W0116ndash0505 this relationpredicts logLPredicted

2minus10 keVerg sminus1 = 4553 plusmn 062 which isin excellent agreement with the luminosity of the best-fit model to the X-ray data of logL2minus10 keVerg sminus1 =4563+058

minus024 For W0220+0137 we also find excellent

agreement with logLPredicted2minus10 keVerg sminus1 = 4558 plusmn 062

and logL2minus10 keVerg sminus1 = 4554+030minus216 A16 nominally

found a good agreement as well for W0204ndash0506 as theyestimated logLPredicted

2minus10 keVerg sminus1 = 4536plusmn037 and found

logL2minus10 keVerg sminus1 = 449+086minus014 from the best-fit X-ray

model However when jointly considering this with thebest-fit and expected absorption their Figure 5 suggestsW0204ndash0506 may be somewhat X-ray weakFrom the SED modeling we also have an estimate

of the amount of dust that is obscuring the luminousAGN that dominates in both the mid-IR and the X-rays Comparing to the column densities of neutral hy-drogen constrained by the modeling of X-ray spectrawe find dust-to-gas ratios of E(B minus V )NH = 35 plusmn

20 times 10minus24 cm2 mag for W0116ndash0505 where the un-certainty corresponds to the 683 confidence intervaland has been derived for simplicity assuming Gaussianstatistics For W0220+0137 we find E(B minus V )NH =23 times 10minus24 cm2 mag As NH is not constrained atthe 90 level within the model boundaries we cannotderive a meaningful confidence interval For W0204ndash0506 A16 found a larger ratio of E(B minus V )NH =154 plusmn 126 times 10minus23 cm2 mag For comparison the me-

9

dian dust-to-gas ratio in AGN found by Maiolino et al(2001) is 15 times 10minus23 cm2 mag This value is compa-rable to that found in W0204ndash0506 while those foundin W0116ndash0505 and W0220+0137 are lower Unfortu-nately the large uncertainties in this quantity make thisresult difficult to interpret but it is worth noting that re-cently Yan et al (2019) identified a very low dust-to-gasratio of asymp 4 times 10minus25cm2 for a heavily obscured nearbyquasar at z = 0218 with NH asymp 3 times 1025 cmminus2 witharound NH asymp 1023 cmminus2 coming from the ISM whichcould be a better analog to our objects If the dust-to-gas ratio is indeed significantly lower in W0116ndash0505and W0220+0137 than in W0204ndash0506 it could eitherimply a low metallicity for the former systems such thatthere is a deficit of dust overall in the host galaxy orthat a higher than typical fraction of absorbing gas ex-ists within the dust sublimation radius of the accretiondisk We speculate the latter could be consistent with therecent results of Wu et al (2018) that show Hot DOGsare accreting close to the Eddington limit perhaps as aresult of higher gas densities in the vicinity of the SMBHTaken together these results could imply that W0116ndash

0505 and W0220+0137 represent a different class of ob-ject than W0204ndash0506 as the former are either dust-pooror gas-rich in the nuclear regions but have normal X-rayluminosities while the latter has a normal amount ofdust but might be somewhat X-ray weak The morphol-ogy of the HST imaging strongly differs between theseobjects as discussed in sect23 further supporting thisGoulding et al (2018) points out that W0220+0137 isalso classified as an ERQ by Hamann et al (2017) andfinds that W0116ndash0505 fulfills most of the criteria andhence classifies it as ERQ-like This supports a view inwhich ERQs and Hot DOGs are not independent popu-lations but possibly related to each other with BHDs be-ing the link between them We speculate that Hot DOGsmight correspond to the highly obscured AGN phase ofgalaxy evolution proposed by eg Hopkins et al (2008)or Alexander amp Hickox (2012) and as the obscurationstarts clearing out (see Hickox amp Alexander 2018 fora description of the different physical scales of the ob-scuring materials) the object transforms into a BHDand then an ERQ before transitioning into an unob-scured quasar The significant levels of outflowing ion-ized gas identified by Zakamska et al (2016) for fourERQs by Dıaz-Santos et al (2016) for the most lumi-nous Hot DOG W2246ndash0526 and by Wu et al (2018)for two more Hot DOGs support the view that bothtypes of objects are experiencing strong AGN feedback

4 SOURCE OF THE EXCESS BLUE EMISSION

41 Dual AGN

One of the possible scenarios proposed by A16 is thatBHDs could be powered by two AGNs instead of onewhere a primary luminous highly obscured AGN dom-inates the mid-IR emission and a secondary fainterunobscured or lightly obscured AGN dominates theUVoptical emission As discussed above the formerwould be expected to dominate the hard X-ray emissionof these sources and that is exactly what is observedHowever the less luminous component would contributesignificant soft X-ray emission that can be constrainedby the Chandra observations In Table 2 we list the

expected intrinsic 6microm luminosity of the primary andsecondary best-fit AGN components for both W0116ndash0505 and W0220+0137 It is important to note that forthe secondary AGN components we have no useful con-straints in the IR as the rest-frame near-IR is dominatedby the host galaxy and the mid-IR is dominated by theprimary AGN component Its 6microm luminosity comes in-stead indirectly from the template fit to the rest-frameUVoptical SED As we did in sect3 we can estimate theexpected 2ndash10 keV luminosity using the relation of Stern(2015) Hence if the secondary component is a real sec-ond AGN in the system for W0116ndash0505 we expect itto have an X-ray luminosity of logLPredicted

2minus10 keVerg sminus1 =4443plusmn 037 and for W0220+0137 we expect it to havelogLPredicted

2minus10 keVerg sminus1 = 4429 plusmn 062 The gray-dashedcurves in Figures 7 and 8 show the expected X-rayspectrum of these secondary components for W0116ndash0505 and W0220+0137 respectively We assume power-law spectra with Γ = 19 and no absorption asboth secondary components show no reddening in theUVoptical Figure 9 also shows the expected X-rayspectrum of the secondary AGN as determined from theanalysis of A16We can determine with 90 confidence that a sec-

ondary power-law component would have a luminosityof logL2minus10 keVerg sminus1 lt 4395 in W0116ndash0505 and oflogL2minus10 keVerg sminus1 lt 4393 in W0220+0137 Theselimits are marginally consistent with the 2ndash10 keV lumi-nosities expected given the opticalUV luminosities ob-served For W0204ndash0506 on the other hand A16 wasable to rule out this scenario with high confidence Un-like the analysis presented here A16 reached this con-clusion by comparing the change in Cash statistic of theX-ray spectra modeling obtained by requiring or not thepresence of the secondary AGN emission with the ex-pected luminosity Specifically A16 found that includ-ing the secondary component resulted in an increase inthe Cash statistic ∆C = 12838 which allowed to ruleout the dual AGN scenario with gt 999 confidenceWe do not replicate this analysis for W0116ndash0505 andW0220+0137 as the interpretation of the change in theC statistic (∆C = 173 and ∆C = 187 respectively)is complicated by the lower number of counts detectedparticularly in the case of W0220+0137Hence the X-ray spectra of all three objects are better

described by the single highly absorbed AGN modelsuggesting that BHDs are not dual AGN The case isstrongest for W0204ndash0506 while for W0116ndash0505 andW0220+0137 we cannot completely reject the dual AGNscenario with high confidence using the current data sets

42 Extreme Star-formation

Another possibility discussed by A16 is that theUVoptical SED of BHDs is powered by unobscured ex-treme star-formation rather than by unobscured AGNemission This would account for a very blue UVopticalSED without the X-ray contribution expected for a sec-ondary AGNThis scenario was studied in detail by A16 for W0204ndash

0506 Modeling the UVoptical SED of this object usingthe Starburst99 v700 code (Leitherer et al 1999 20102014 Vazquez amp Leitherer 2005) in combination withthe EzGal package of Mancone amp Gonzalez (2012) they

10

determined that the SED could be consistent with beingpowered by a young starburst but only if the SFR wasvery high Specifically they assumed the latest Genevamodels available for the used version of Starburst 99(see Leitherer et al 2014 for details) a constant SFRand a solar metallicity and determined that the SEDcould only be powered by a starburst of age 5 Myrwith SFR amp 1000 M⊙ yrminus1 with 90 confidence Alower metallicity somewhat eases these constraints withthe lowest metallicity available for the Geneva models inStarburst99 of Z = 0001 implying age 100 Myr andSFR amp 250M⊙ yrminus1 However A16 considered that sucha low metallicity was unlikely given the large amount ofdust available in the inner regions of the system thatgive rise to the high specific luminosities in the mid-IRFurthermore due to the large unobscured SFR impliedby the solar metallicity models A16 considered that theUVoptical SED was unlikely powered by a starburstHowever the morphology of the UV emission in the

HST imaging we have obtained (see sect23 for details)seems to imply that starburst activity is present inW0204ndash0506 As shown in Figure 5 the flux tracedby the F555W band (rest-frame sim1750A) is distributedalong the NE section of the galaxy and concentrated ina few distinct regions The bulk of the F555W emis-sion is considerably offset from the emission of the olderstars traced by the F160W band (rest-frame sim5100A)Furthermore the morphology of the system is consistentwith a merger (see sect23) which can trigger significantstar-formation activityThe analysis of A16 in conjunction with the HST

imaging available for W0204ndash0506 then imply that if itsUVoptical SED is solely powered by a starburst thenthe system must be in a very uncommon state On onehand it could be that the system has a very large metal-licity gradient such that in the outskirts where star-formation dominates the metallicity is close to primor-dial and SFR is only amp 250M⊙ yrminus1 yet near the SMBHthe metallicity is high enough to allow for the substantialamount of dust needed to obscure the hyper-luminousAGN The other possibility would be that W0204ndash0506does not have a substantial metallicity gradient but isinstead powered by the strongest unobscured starburstknown with SFR amp 1000 M⊙ yrminus1A third and more likely option is that while a moder-

ate starburst is ongoing in the system the UVopticalemission is still dominated by light leaking from the cen-tral highly obscured AGN As shown in Table 3 (alsosee discussion in sect23) the light distribution of W0204ndash0506 in the F555W band has a somewhat larger Gini anda significantly larger M20 coefficient than the other twoBHDs studied While the large M20 is consistent withthe observed patchiness of the system the high Gini co-efficient implies that the light is strongly concentrated inthe brightest regions In the left panel of Figure 5 it canbe appreciated that the NW UV clump (marked by themagenta circle 02primeprime diameter) is brighter than the restcontaining approximately 10 of the total F555W fluxmeasured in the 4primeprime radius aperture This region is closeto the geometrical center of the F160W light distributionand could correspond to the position of the buried AGNThat the optical spectrum of this source (Fig 3) showsa mixture of narrow and broad emission lines is also con-

sistent with this picture as A16 reported a FWHM of1630plusmn 220 km sminus1 for C iv but of only 550plusmn 100 km sminus1

for C iii]For W0116ndash0505 and W0220+0137 the situation is

somewhat different The optical spectra shown in Fig-ures 2 and 4 show clear broad high-ionization featurescharacteristic of type 1 AGNs The UV emission whilespatially extended is strongly concentrated in both ob-jects (see discussion in sect23) which is more consistentwith the expectations for the dual AGN or the leakedAGN light scenarios instead of the star-formation sce-nario If we model their broad-band SEDs as starburstsas in A16 we find the best-fit models shown in Figures 10and 11 for W0116ndash0505 and W0220+0137 respectivelyEach figure shows the best fit obtained assuming a so-lar metallicity and a metallicity of Z = 0001 discussedabove As we do not include nebular emission we only usethe bands that are redward of the Lyα emission line andexclude the F160W band which can be strongly contami-nated by [O ii] emission Indeed if we include the F160Wband we find best-fit χ2 values a factorsim3 larger indicat-ing either that nebular emission is prevalent in this bandor that it is dominated by an older stellar population Weset a minimum photometric uncertainty of 005 mag assystematic differences between the measurements are un-likely to be below that level In practice this only affectsthe uncertainty used for the F555W band The best-fitSFR age and obscuration of the stellar population areshown in the figures as well however the values are quitedegenerate as shown in Figure 12 particularly as thereare no constrains longwards of sim4000A The only otherlonger wavelength broadbands that we have are in re-gions of the SED dominated by either an older stellarpopulation or by the highly obscured luminous AGN asshown in Figure 1 and hence are not useful for constrain-ing these fits We note however that the χ2 values ofthe best fits are quite large when it is considered that weare fitting three different parameters This coupled withUVoptical spectral features (ie the presence of broademission lines) and the morphology in the HST imagingsuggest that the UVoptical emission in these objects isunlikely dominated by unobscured starbursts

43 Leaked AGN Light

The third possibility to explain the nature of BHDsis that the blue excess emission found in these objectscorresponds to light coming from the highly obscuredprimary AGN that is leaking into our line of sight Asdiscussed by A16 this could happen either due to dust orgas scattering of the AGN emission into our line of sightor due to a small gap in the dust that allows for a par-tial view towards the accretion disk and the broad-lineregion However the latter is unlikely as discussed byA16 as the UVoptical SED is consistent with the emis-sion of an unobscured accretion disk As radiation atprogressively shorter wavelengths is emitted in progres-sively inner regions of the accretion disk a gap that onlyallows sim1 of the emitted light through but does notdistort the accretion disk spectrum would need to cover99 of the effective disk size at each wavelength Whilenot impossible the shape of such a gap would be exceed-ingly contrived making this unlikely Furthermore wewould not expect the UVoptical emission to be spatially

11

Figure 10 (Upper panel) The solid black line shows the best-fitStarburst99 SED model to the UVoptical broad-band photome-try of W0116ndash0505 assuming solar metallicity (see text for details)The gray shaded area shows all SED shapes within the 90 confi-dence interval (Bottom panel) Same as in the top panel but for ametallicity of Z = 0001

Figure 11 Same as Fig 10 but for W0220+0137

extended as found in sect23This suggests that the most likely source of the blue

excess emission in the three BHDs studied is scatteredlight In this scenario 99 of the emission from theaccretion disk and the broad-line region would be ab-

Figure 12 The contours show a χ2 map of the best-fit Star-burts99 models as discussed in the text The contours for W0116ndash0505 and W0220+0137 are shown in the top and bottom panelsrespectively Dark contours assume solar metallicity while graycontours assume Z = 0001 The solid (dotted) contour shows the683 (90) confidence region while the solid dots show the valuesof the best-fits models shown in Figures 10 and 11

sorbed by dust while 1 will be scattered into our lineof sight by either dust or gas or both Reflection nebu-lae are known to make the reflected SED bluer than theemitted SED in the UV (λ 2500A eg see Draine2003a) suggesting that the scattering medium in BHDsis more likely the gas surrounding the AGN Howeveras there is likely dust on scales larger than the torus(Dıaz-Santos et al 2016 Tsai et al 2018) we note thatour data are not sufficient to rule out an SED that hasbeen made bluer by dust reflection and redder by dustabsorption With respect to the X-rays we note thatwhile they should also be scattered into our line of sightalong with the UV emission the scattering cross sectionby either dust or free electrons is significantly smaller atthe high energy ranges probed by the Chandra observa-tions (Draine 2003ab) Hence the non-detection of aclear unabsorbed component in the X-ray spectra in sect3is consistent with this scattering scenario

5 SUMMARY

We have investigated the source of the blue excessemission in three BHDs two of which were identi-fied as such by A16 and a third one which has anSED consistent with that of a Hot DOG although itdoes not meet the formal selection criteria due to be-ing slightly too bright in the W1 band While allHot DOGs are characterized by mid-IR emission thatis most naturally explained by a highly obscured hyper-luminous AGN (Eisenhardt et al 2012 Assef et al 2015Tsai et al 2015) BHDs have a UVoptical SED thatis significantly bluer than expected based on templatefitting results Using a similar approach to that of

12

Assef et al (2015) we find that the SEDs of BHDs arebest modeled using two AGN components a primaryhyper-luminous highly obscured AGN that dominatesthe mid-IR emission and a secondary lower luminos-ity but unobscured AGN that dominates the UVopticalemission The bolometric luminosity of the secondaryAGN SED is sim1 of that of the primary componentA16 identified three possible scenarios to produce the ex-cess blue emission namely (i) a secondary less luminousbut unobscured AGN in the system (ii) an extreme star-burst or (iii) leaked UVoptical light from the primaryhighly luminous highly obscured AGN that dominatesthe mid-IRFor one of the sources (W0204ndash0506) A16 ruled out

a secondary AGN as the source of the blue excess emis-sion and instead concluded that the excess was causedby either unobscured star formation with an SFR amp1000 M⊙ yrminus1 or by UVoptical light from the cen-tral engine leaking into our line of sight due to scatter-ing or through a partially obscured sight-line with thescattered AGN light hypothesis deemed more likely Inthis paper we have presented HSTWFC3 imaging ofW0204ndash0506 showing a morphology consistent with anon-going merger and evidence of an on-going widespreadstarburst Considering however the very high SFRneeded to explain the UV emission by star-formationalone we conclude it is more likely that the UV emissionof W0204ndash0505 arises from a combination of scatteredAGN light and star-formationWe also studied in detail two other BHDs W0116ndash

0505 and W0220+0137 We present observations ob-tained with ChandraACIS-S and interpret them us-ing the Brightman amp Nandra (2011) models We findthat the X-ray spectra are consistent with single lumi-nous highly absorbed AGNs dominating the X-ray emis-sion We find that the L2minus10 keV luminosities of theseAGNs are consistent with those expected for the pri-mary AGNs based on their estimated L6microm according tothe L6microm minus L2minus10 keV relation of Stern (2015) We alsofind that the gas-to-dust ratios of the AGNs in thesesystems are somewhat below the median value foundin AGNs by Maiolino et al (2001) and lower than thatfound in W0204ndash0506 suggestive of a lower metallicityor of a higher fraction of absorbing gas within the dust-sublimation radius of the AGN Based on the UV throughmid-IR SED models of these sources we estimate the ex-pected X-ray luminosity of the putative secondary AGNcomponents assuming it is a second independent AGNin the system We found that the X-ray observations areonly marginally consistent with the presence of a secondAGN component in bothW0116ndash0505 andW0220+0137suggesting the dual AGN scenario is unlikelyWe followed A16 and modeled the UV emission of

W0116ndash0505 and W0220+0137 assuming a pure star-burst scenario and found that while the best-fit SFRsare generally high comparable to those found by A16for W0204ndash0506 they are not well constrained due tothe large degeneracies between SFR age and metallicityWe found however that the χ2 values of the best-fit star-burst models are large (sim 12 for W0116ndash0506 and sim 8for W0220+0137) despite the small number of degrees offreedom (1 and 2 respectively) implying a pure starburstis not a good description of the observed UV SED Addi-tionally the rest-frame UV spectra shows broad emission

lines characteristic of AGN activity further suggestingthat star-formation does not dominate the observed UVemissionFinally we also studied the morphologies observed in

the HSTWFC3 F555W and F160W images of W0116ndash0505 and W0220+0137 and found them to be undis-turbed with the UV emission being centrally concen-trated An analysis based on the Gini M20 and A coef-ficients showed that these systems are best characterizedas undisturbed early type galaxies consistent with theleaked AGN scenario Considering all of this we con-clude that the source of the UV emission in W0116ndash0505and W0220+0137 is scattered light from the hyperlu-minous highly obscured AGN that powers the mid-IRSED Given the detail of our data and SED modelingwe cannot determine whether the scattering material isprimarily gas dust or a mixture of bothThat all three BHDs we have investigated are due to

scattered light from the highly obscured hyperluminousAGN highlights how powerful the central engine is in HotDOGs with only 1 of the emission of the accretiondisk scattered into our line of sight it is still more lumi-nous than the entire stellar emission of the host galaxy inthe UV This is in general agreement with recent resultswhich show that the SMBHs in Hot DOGs are accretingabove the Eddington limit (Wu et al 2018 Tsai et al2018) and are injecting large amounts of energy into theISM of their host galaxies (Dıaz-Santos et al 2016) andhence are experiencing strong events of AGN feedback

We thank J Comerford and B Weiner for carry-ing out observations presented in this article RJAwas supported by FONDECYT grants number 1151408and 1191124 DJW acknowledges financial supportfrom STFC Ernest Rutherford fellowships HDJwas supported by Basic Science Research Programthrough the National Research Foundation of Korea(NRF) funded by the Ministry of Education (NRF-2017R1A6A3A04005158) FEB acknowledges supportfrom CONICYT-Chile (Basal AFB-170002) and the Min-istry of Economy Development and Tourismrsquos MilleniumScience Initiative through grant IC120009 awarded toThe Millenium Institute of Astrophysics MAS JW issupported by the NSFC Grant 11690024 and SPRP CASgrant XDB23000000 Based on observations made withthe NASAESA Hubble Space Telescope obtained at theSpace Telescope Science Institute which is operated bythe Association of Universities for Research in Astron-omy Inc under NASA contract NAS 5-26555 Theseobservations are associated with program 14358 Sup-port for program14358 was provided by NASA througha grant from the Space Telescope Science Institute whichis operated by the Association of Universities for Re-search in Astronomy Inc under NASA contract NAS5-26555 The scientific results reported in this articleare based to a significant degree on data obtained fromthe Chandra X-ray Observatory and observations madeby the Chandra X-ray Observatory and published previ-ously in cited articles Support for this work was pro-vided by the National Aeronautics and Space Adminis-tration through Chandra Award Number 17700696 is-sued by the Chandra X-ray Center which is operatedby the Smithsonian Astrophysical Observatory for and

13

on behalf of the National Aeronautics Space Adminis-tration under contract NAS8-03060 This publicationmakes use of data products from the Wide-field InfraredSurvey Explorer which is a joint project of the Universityof California Los Angeles and the Jet Propulsion Labo-ratoryCalifornia Institute of Technology funded by theNational Aeronautics and Space Administration Thiswork is based in part on observations made with theSpitzer Space Telescope which is operated by the JetPropulsion Laboratory California Institute of Technol-ogy under a contract with NASA The Pan-STARRS1Surveys (PS1) have been made possible through con-tributions of the Institute for Astronomy the Univer-sity of Hawaii the Pan-STARRS Project Office theMax-Planck Society and its participating institutes theMax Planck Institute for Astronomy Heidelberg andthe Max Planck Institute for Extraterrestrial PhysicsGarching The Johns Hopkins University Durham Uni-versity the University of Edinburgh Queenrsquos Univer-sity Belfast the Harvard-Smithsonian Center for Astro-physics the Las Cumbres Observatory Global TelescopeNetwork Incorporated the National Central Universityof Taiwan the Space Telescope Science Institute theNational Aeronautics and Space Administration underGrant No NNX08AR22G issued through the PlanetaryScience Division of the NASA Science Mission Direc-torate the National Science Foundation under Grant NoAST-1238877 the University of Maryland and EotvosLorand University (ELTE) and the Los Alamos NationalLaboratory Funding for SDSS-III has been provided bythe Alfred P Sloan Foundation the Participating Insti-tutions the National Science Foundation and the USDepartment of Energy Office of Science The SDSS-III web site is httpwwwsdss3org SDSS-III ismanaged by the Astrophysical Research Consortium forthe Participating Institutions of the SDSS-III Collabo-ration including the University of Arizona the BrazilianParticipation Group Brookhaven National LaboratoryCarnegie Mellon University University of Florida theFrench Participation Group the German ParticipationGroup Harvard University the Instituto de Astrofisicade Canarias the Michigan StateNotre DameJINA Par-ticipation Group Johns Hopkins University LawrenceBerkeley National Laboratory Max Planck Institute forAstrophysics Max Planck Institute for ExtraterrestrialPhysics New Mexico State University New York Uni-versity Ohio State University Pennsylvania State Uni-versity University of Portsmouth Princeton Universitythe Spanish Participation Group University of TokyoUniversity of Utah Vanderbilt University University ofVirginia University of Washington and Yale UniversitySome of the observations reported here were obtained atthe MMT Observatory a joint facility of the SmithsonianInstitution and the University of Arizona

REFERENCES

Abraham R G van den Bergh S amp Nair P 2003 ApJ 588218

Alexander D M amp Hickox R C 2012 NewAR 56 93Assef R J Kochanek C S Brodwin M et al 2010 ApJ 713

970Assef R J Denney K D Kochanek C S et al 2011 ApJ

742 93Assef R J Eisenhardt P R M Stern D et al 2015 ApJ

804 27

Assef R J Walton D J Brightman M et al 2016 ApJ 819111

Banerji M Alaghband-Zadeh S Hewett P C amp McMahonR G 2015 MNRAS 447 3368

Barger A J Cowie L L Chen C-C et al 2014 ApJ 784 9Bauer F E Yan L Sajina A amp Alexander D M 2010 ApJ

710 212Bennert V N Auger M W Treu T Woo J-H amp Malkan

M A 2011 ApJ 726 59Bertin E amp Arnouts S 1996 AampAS 117 393Brightman M amp Nandra K 2011 MNRAS 413 1206Cash W 1979 ApJ 228 939Chen C-T J Hickox R C Goulding A D et al 2017 ApJ

837 145Conselice C J 2014 ARAampA 52 291Cutri R M Wright E L Conrow T et al 2012 Explanatory

Supplement to the WISE All-Sky Data Release Products Techrep

Dıaz-Santos T Assef R J Blain A W et al 2016 ApJ 816L6

mdash 2018 Science 362 1034Draine B T 2003a ApJ 598 1017mdash 2003b ApJ 598 1026Eisenhardt P R M Wu J Tsai C-W et al 2012 ApJ 755

173Fan L Han Y Nikutta R Drouart G amp Knudsen K K

2016a ApJ 823 107Fan L Han Y Fang G et al 2016b ApJ 822 L32Farrah D Petty S Connolly B et al 2017 ApJ 844 106Finkelstein S L Rhoads J E Malhotra S Pirzkal N amp

Wang J 2007 ApJ 660 1023Fiore F Puccetti S Brusa M et al 2009 ApJ 693 447Gandhi P Horst H Smette A et al 2009 AampA 502 457Goulding A D Zakamska N L Alexandroff R M et al

2018 ApJ 856 4Griffith R L Kirkpatrick J D Eisenhardt P R M et al

2012 AJ 144 148Hamann F Zakamska N L Ross N et al 2017 MNRAS

464 3431Hickox R C amp Alexander D M 2018 ARAampA 56 625Hopkins P F Hernquist L Cox T J amp Keres D 2008

ApJS 175 356Jun H et al in prepLeitherer C Ekstrom S Meynet G et al 2014 ApJS 212 14Leitherer C Ortiz Otalvaro P A Bresolin F et al 2010

ApJS 189 309Leitherer C Schaerer D Goldader J D et al 1999 ApJS

123 3Lotz J M Primack J amp Madau P 2004 AJ 128 163Lotz J M Davis M Faber S M et al 2008 ApJ 672 177Lupton R Blanton M R Fekete G et al 2004 PASP 116

133Magorrian J Tremaine S Richstone D et al 1998 AJ 115

2285Maiolino R Marconi A Salvati M et al 2001 AampA 365 28Mancone C L amp Gonzalez A H 2012 PASP 124 606Mateos S Carrera F J Alonso-Herrero A et al 2015

MNRAS 449 1422Petrosian V 1976 ApJ 209 L1Piconcelli E Vignali C Bianchi S et al 2015 AampA 574 L9Rhodes J Refregier A amp Groth E J 2000 ApJ 536 79Ricci C Assef R J Stern D et al 2017 ApJ 835 105Stern D 2015 ApJ 807 129Stern D Lansbury G B Assef R J et al 2014 ApJ 794 102Tsai C-W Eisenhardt P R M Wu J et al 2015 ApJ 805

90Tsai C-W Eisenhardt P R M Jun H D et al 2018 ApJ

868 15van Dokkum P G 2001 PASP 113 1420Vazquez G A amp Leitherer C 2005 ApJ 621 695Vito F Brandt W N Stern D et al 2018 MNRAS 474 4528Wright E L Eisenhardt P R M Mainzer A K et al 2010

AJ 140 1868Wu J Tsai C-W Sayers J et al 2012 ApJ 756 96Wu J Bussmann R S Tsai C-W et al 2014 ApJ 793 8Wu J Jun H D Assef R J et al 2018 ApJ 852 96Yan W Hickox R C Hainline K N et al 2019 ApJ 870 33

10

determined that the SED could be consistent with beingpowered by a young starburst but only if the SFR wasvery high Specifically they assumed the latest Genevamodels available for the used version of Starburst 99(see Leitherer et al 2014 for details) a constant SFRand a solar metallicity and determined that the SEDcould only be powered by a starburst of age 5 Myrwith SFR amp 1000 M⊙ yrminus1 with 90 confidence Alower metallicity somewhat eases these constraints withthe lowest metallicity available for the Geneva models inStarburst99 of Z = 0001 implying age 100 Myr andSFR amp 250M⊙ yrminus1 However A16 considered that sucha low metallicity was unlikely given the large amount ofdust available in the inner regions of the system thatgive rise to the high specific luminosities in the mid-IRFurthermore due to the large unobscured SFR impliedby the solar metallicity models A16 considered that theUVoptical SED was unlikely powered by a starburstHowever the morphology of the UV emission in the

HST imaging we have obtained (see sect23 for details)seems to imply that starburst activity is present inW0204ndash0506 As shown in Figure 5 the flux tracedby the F555W band (rest-frame sim1750A) is distributedalong the NE section of the galaxy and concentrated ina few distinct regions The bulk of the F555W emis-sion is considerably offset from the emission of the olderstars traced by the F160W band (rest-frame sim5100A)Furthermore the morphology of the system is consistentwith a merger (see sect23) which can trigger significantstar-formation activityThe analysis of A16 in conjunction with the HST

imaging available for W0204ndash0506 then imply that if itsUVoptical SED is solely powered by a starburst thenthe system must be in a very uncommon state On onehand it could be that the system has a very large metal-licity gradient such that in the outskirts where star-formation dominates the metallicity is close to primor-dial and SFR is only amp 250M⊙ yrminus1 yet near the SMBHthe metallicity is high enough to allow for the substantialamount of dust needed to obscure the hyper-luminousAGN The other possibility would be that W0204ndash0506does not have a substantial metallicity gradient but isinstead powered by the strongest unobscured starburstknown with SFR amp 1000 M⊙ yrminus1A third and more likely option is that while a moder-

ate starburst is ongoing in the system the UVopticalemission is still dominated by light leaking from the cen-tral highly obscured AGN As shown in Table 3 (alsosee discussion in sect23) the light distribution of W0204ndash0506 in the F555W band has a somewhat larger Gini anda significantly larger M20 coefficient than the other twoBHDs studied While the large M20 is consistent withthe observed patchiness of the system the high Gini co-efficient implies that the light is strongly concentrated inthe brightest regions In the left panel of Figure 5 it canbe appreciated that the NW UV clump (marked by themagenta circle 02primeprime diameter) is brighter than the restcontaining approximately 10 of the total F555W fluxmeasured in the 4primeprime radius aperture This region is closeto the geometrical center of the F160W light distributionand could correspond to the position of the buried AGNThat the optical spectrum of this source (Fig 3) showsa mixture of narrow and broad emission lines is also con-

sistent with this picture as A16 reported a FWHM of1630plusmn 220 km sminus1 for C iv but of only 550plusmn 100 km sminus1

for C iii]For W0116ndash0505 and W0220+0137 the situation is

somewhat different The optical spectra shown in Fig-ures 2 and 4 show clear broad high-ionization featurescharacteristic of type 1 AGNs The UV emission whilespatially extended is strongly concentrated in both ob-jects (see discussion in sect23) which is more consistentwith the expectations for the dual AGN or the leakedAGN light scenarios instead of the star-formation sce-nario If we model their broad-band SEDs as starburstsas in A16 we find the best-fit models shown in Figures 10and 11 for W0116ndash0505 and W0220+0137 respectivelyEach figure shows the best fit obtained assuming a so-lar metallicity and a metallicity of Z = 0001 discussedabove As we do not include nebular emission we only usethe bands that are redward of the Lyα emission line andexclude the F160W band which can be strongly contami-nated by [O ii] emission Indeed if we include the F160Wband we find best-fit χ2 values a factorsim3 larger indicat-ing either that nebular emission is prevalent in this bandor that it is dominated by an older stellar population Weset a minimum photometric uncertainty of 005 mag assystematic differences between the measurements are un-likely to be below that level In practice this only affectsthe uncertainty used for the F555W band The best-fitSFR age and obscuration of the stellar population areshown in the figures as well however the values are quitedegenerate as shown in Figure 12 particularly as thereare no constrains longwards of sim4000A The only otherlonger wavelength broadbands that we have are in re-gions of the SED dominated by either an older stellarpopulation or by the highly obscured luminous AGN asshown in Figure 1 and hence are not useful for constrain-ing these fits We note however that the χ2 values ofthe best fits are quite large when it is considered that weare fitting three different parameters This coupled withUVoptical spectral features (ie the presence of broademission lines) and the morphology in the HST imagingsuggest that the UVoptical emission in these objects isunlikely dominated by unobscured starbursts

43 Leaked AGN Light

The third possibility to explain the nature of BHDsis that the blue excess emission found in these objectscorresponds to light coming from the highly obscuredprimary AGN that is leaking into our line of sight Asdiscussed by A16 this could happen either due to dust orgas scattering of the AGN emission into our line of sightor due to a small gap in the dust that allows for a par-tial view towards the accretion disk and the broad-lineregion However the latter is unlikely as discussed byA16 as the UVoptical SED is consistent with the emis-sion of an unobscured accretion disk As radiation atprogressively shorter wavelengths is emitted in progres-sively inner regions of the accretion disk a gap that onlyallows sim1 of the emitted light through but does notdistort the accretion disk spectrum would need to cover99 of the effective disk size at each wavelength Whilenot impossible the shape of such a gap would be exceed-ingly contrived making this unlikely Furthermore wewould not expect the UVoptical emission to be spatially

11

Figure 10 (Upper panel) The solid black line shows the best-fitStarburst99 SED model to the UVoptical broad-band photome-try of W0116ndash0505 assuming solar metallicity (see text for details)The gray shaded area shows all SED shapes within the 90 confi-dence interval (Bottom panel) Same as in the top panel but for ametallicity of Z = 0001

Figure 11 Same as Fig 10 but for W0220+0137

extended as found in sect23This suggests that the most likely source of the blue

excess emission in the three BHDs studied is scatteredlight In this scenario 99 of the emission from theaccretion disk and the broad-line region would be ab-

Figure 12 The contours show a χ2 map of the best-fit Star-burts99 models as discussed in the text The contours for W0116ndash0505 and W0220+0137 are shown in the top and bottom panelsrespectively Dark contours assume solar metallicity while graycontours assume Z = 0001 The solid (dotted) contour shows the683 (90) confidence region while the solid dots show the valuesof the best-fits models shown in Figures 10 and 11

sorbed by dust while 1 will be scattered into our lineof sight by either dust or gas or both Reflection nebu-lae are known to make the reflected SED bluer than theemitted SED in the UV (λ 2500A eg see Draine2003a) suggesting that the scattering medium in BHDsis more likely the gas surrounding the AGN Howeveras there is likely dust on scales larger than the torus(Dıaz-Santos et al 2016 Tsai et al 2018) we note thatour data are not sufficient to rule out an SED that hasbeen made bluer by dust reflection and redder by dustabsorption With respect to the X-rays we note thatwhile they should also be scattered into our line of sightalong with the UV emission the scattering cross sectionby either dust or free electrons is significantly smaller atthe high energy ranges probed by the Chandra observa-tions (Draine 2003ab) Hence the non-detection of aclear unabsorbed component in the X-ray spectra in sect3is consistent with this scattering scenario

5 SUMMARY

We have investigated the source of the blue excessemission in three BHDs two of which were identi-fied as such by A16 and a third one which has anSED consistent with that of a Hot DOG although itdoes not meet the formal selection criteria due to be-ing slightly too bright in the W1 band While allHot DOGs are characterized by mid-IR emission thatis most naturally explained by a highly obscured hyper-luminous AGN (Eisenhardt et al 2012 Assef et al 2015Tsai et al 2015) BHDs have a UVoptical SED thatis significantly bluer than expected based on templatefitting results Using a similar approach to that of

12

Assef et al (2015) we find that the SEDs of BHDs arebest modeled using two AGN components a primaryhyper-luminous highly obscured AGN that dominatesthe mid-IR emission and a secondary lower luminos-ity but unobscured AGN that dominates the UVopticalemission The bolometric luminosity of the secondaryAGN SED is sim1 of that of the primary componentA16 identified three possible scenarios to produce the ex-cess blue emission namely (i) a secondary less luminousbut unobscured AGN in the system (ii) an extreme star-burst or (iii) leaked UVoptical light from the primaryhighly luminous highly obscured AGN that dominatesthe mid-IRFor one of the sources (W0204ndash0506) A16 ruled out

a secondary AGN as the source of the blue excess emis-sion and instead concluded that the excess was causedby either unobscured star formation with an SFR amp1000 M⊙ yrminus1 or by UVoptical light from the cen-tral engine leaking into our line of sight due to scatter-ing or through a partially obscured sight-line with thescattered AGN light hypothesis deemed more likely Inthis paper we have presented HSTWFC3 imaging ofW0204ndash0506 showing a morphology consistent with anon-going merger and evidence of an on-going widespreadstarburst Considering however the very high SFRneeded to explain the UV emission by star-formationalone we conclude it is more likely that the UV emissionof W0204ndash0505 arises from a combination of scatteredAGN light and star-formationWe also studied in detail two other BHDs W0116ndash

0505 and W0220+0137 We present observations ob-tained with ChandraACIS-S and interpret them us-ing the Brightman amp Nandra (2011) models We findthat the X-ray spectra are consistent with single lumi-nous highly absorbed AGNs dominating the X-ray emis-sion We find that the L2minus10 keV luminosities of theseAGNs are consistent with those expected for the pri-mary AGNs based on their estimated L6microm according tothe L6microm minus L2minus10 keV relation of Stern (2015) We alsofind that the gas-to-dust ratios of the AGNs in thesesystems are somewhat below the median value foundin AGNs by Maiolino et al (2001) and lower than thatfound in W0204ndash0506 suggestive of a lower metallicityor of a higher fraction of absorbing gas within the dust-sublimation radius of the AGN Based on the UV throughmid-IR SED models of these sources we estimate the ex-pected X-ray luminosity of the putative secondary AGNcomponents assuming it is a second independent AGNin the system We found that the X-ray observations areonly marginally consistent with the presence of a secondAGN component in bothW0116ndash0505 andW0220+0137suggesting the dual AGN scenario is unlikelyWe followed A16 and modeled the UV emission of

W0116ndash0505 and W0220+0137 assuming a pure star-burst scenario and found that while the best-fit SFRsare generally high comparable to those found by A16for W0204ndash0506 they are not well constrained due tothe large degeneracies between SFR age and metallicityWe found however that the χ2 values of the best-fit star-burst models are large (sim 12 for W0116ndash0506 and sim 8for W0220+0137) despite the small number of degrees offreedom (1 and 2 respectively) implying a pure starburstis not a good description of the observed UV SED Addi-tionally the rest-frame UV spectra shows broad emission

lines characteristic of AGN activity further suggestingthat star-formation does not dominate the observed UVemissionFinally we also studied the morphologies observed in

the HSTWFC3 F555W and F160W images of W0116ndash0505 and W0220+0137 and found them to be undis-turbed with the UV emission being centrally concen-trated An analysis based on the Gini M20 and A coef-ficients showed that these systems are best characterizedas undisturbed early type galaxies consistent with theleaked AGN scenario Considering all of this we con-clude that the source of the UV emission in W0116ndash0505and W0220+0137 is scattered light from the hyperlu-minous highly obscured AGN that powers the mid-IRSED Given the detail of our data and SED modelingwe cannot determine whether the scattering material isprimarily gas dust or a mixture of bothThat all three BHDs we have investigated are due to

scattered light from the highly obscured hyperluminousAGN highlights how powerful the central engine is in HotDOGs with only 1 of the emission of the accretiondisk scattered into our line of sight it is still more lumi-nous than the entire stellar emission of the host galaxy inthe UV This is in general agreement with recent resultswhich show that the SMBHs in Hot DOGs are accretingabove the Eddington limit (Wu et al 2018 Tsai et al2018) and are injecting large amounts of energy into theISM of their host galaxies (Dıaz-Santos et al 2016) andhence are experiencing strong events of AGN feedback

We thank J Comerford and B Weiner for carry-ing out observations presented in this article RJAwas supported by FONDECYT grants number 1151408and 1191124 DJW acknowledges financial supportfrom STFC Ernest Rutherford fellowships HDJwas supported by Basic Science Research Programthrough the National Research Foundation of Korea(NRF) funded by the Ministry of Education (NRF-2017R1A6A3A04005158) FEB acknowledges supportfrom CONICYT-Chile (Basal AFB-170002) and the Min-istry of Economy Development and Tourismrsquos MilleniumScience Initiative through grant IC120009 awarded toThe Millenium Institute of Astrophysics MAS JW issupported by the NSFC Grant 11690024 and SPRP CASgrant XDB23000000 Based on observations made withthe NASAESA Hubble Space Telescope obtained at theSpace Telescope Science Institute which is operated bythe Association of Universities for Research in Astron-omy Inc under NASA contract NAS 5-26555 Theseobservations are associated with program 14358 Sup-port for program14358 was provided by NASA througha grant from the Space Telescope Science Institute whichis operated by the Association of Universities for Re-search in Astronomy Inc under NASA contract NAS5-26555 The scientific results reported in this articleare based to a significant degree on data obtained fromthe Chandra X-ray Observatory and observations madeby the Chandra X-ray Observatory and published previ-ously in cited articles Support for this work was pro-vided by the National Aeronautics and Space Adminis-tration through Chandra Award Number 17700696 is-sued by the Chandra X-ray Center which is operatedby the Smithsonian Astrophysical Observatory for and

13

on behalf of the National Aeronautics Space Adminis-tration under contract NAS8-03060 This publicationmakes use of data products from the Wide-field InfraredSurvey Explorer which is a joint project of the Universityof California Los Angeles and the Jet Propulsion Labo-ratoryCalifornia Institute of Technology funded by theNational Aeronautics and Space Administration Thiswork is based in part on observations made with theSpitzer Space Telescope which is operated by the JetPropulsion Laboratory California Institute of Technol-ogy under a contract with NASA The Pan-STARRS1Surveys (PS1) have been made possible through con-tributions of the Institute for Astronomy the Univer-sity of Hawaii the Pan-STARRS Project Office theMax-Planck Society and its participating institutes theMax Planck Institute for Astronomy Heidelberg andthe Max Planck Institute for Extraterrestrial PhysicsGarching The Johns Hopkins University Durham Uni-versity the University of Edinburgh Queenrsquos Univer-sity Belfast the Harvard-Smithsonian Center for Astro-physics the Las Cumbres Observatory Global TelescopeNetwork Incorporated the National Central Universityof Taiwan the Space Telescope Science Institute theNational Aeronautics and Space Administration underGrant No NNX08AR22G issued through the PlanetaryScience Division of the NASA Science Mission Direc-torate the National Science Foundation under Grant NoAST-1238877 the University of Maryland and EotvosLorand University (ELTE) and the Los Alamos NationalLaboratory Funding for SDSS-III has been provided bythe Alfred P Sloan Foundation the Participating Insti-tutions the National Science Foundation and the USDepartment of Energy Office of Science The SDSS-III web site is httpwwwsdss3org SDSS-III ismanaged by the Astrophysical Research Consortium forthe Participating Institutions of the SDSS-III Collabo-ration including the University of Arizona the BrazilianParticipation Group Brookhaven National LaboratoryCarnegie Mellon University University of Florida theFrench Participation Group the German ParticipationGroup Harvard University the Instituto de Astrofisicade Canarias the Michigan StateNotre DameJINA Par-ticipation Group Johns Hopkins University LawrenceBerkeley National Laboratory Max Planck Institute forAstrophysics Max Planck Institute for ExtraterrestrialPhysics New Mexico State University New York Uni-versity Ohio State University Pennsylvania State Uni-versity University of Portsmouth Princeton Universitythe Spanish Participation Group University of TokyoUniversity of Utah Vanderbilt University University ofVirginia University of Washington and Yale UniversitySome of the observations reported here were obtained atthe MMT Observatory a joint facility of the SmithsonianInstitution and the University of Arizona

REFERENCES

Abraham R G van den Bergh S amp Nair P 2003 ApJ 588218

Alexander D M amp Hickox R C 2012 NewAR 56 93Assef R J Kochanek C S Brodwin M et al 2010 ApJ 713

970Assef R J Denney K D Kochanek C S et al 2011 ApJ

742 93Assef R J Eisenhardt P R M Stern D et al 2015 ApJ

804 27

Assef R J Walton D J Brightman M et al 2016 ApJ 819111

Banerji M Alaghband-Zadeh S Hewett P C amp McMahonR G 2015 MNRAS 447 3368

Barger A J Cowie L L Chen C-C et al 2014 ApJ 784 9Bauer F E Yan L Sajina A amp Alexander D M 2010 ApJ

710 212Bennert V N Auger M W Treu T Woo J-H amp Malkan

M A 2011 ApJ 726 59Bertin E amp Arnouts S 1996 AampAS 117 393Brightman M amp Nandra K 2011 MNRAS 413 1206Cash W 1979 ApJ 228 939Chen C-T J Hickox R C Goulding A D et al 2017 ApJ

837 145Conselice C J 2014 ARAampA 52 291Cutri R M Wright E L Conrow T et al 2012 Explanatory

Supplement to the WISE All-Sky Data Release Products Techrep

Dıaz-Santos T Assef R J Blain A W et al 2016 ApJ 816L6

mdash 2018 Science 362 1034Draine B T 2003a ApJ 598 1017mdash 2003b ApJ 598 1026Eisenhardt P R M Wu J Tsai C-W et al 2012 ApJ 755

173Fan L Han Y Nikutta R Drouart G amp Knudsen K K

2016a ApJ 823 107Fan L Han Y Fang G et al 2016b ApJ 822 L32Farrah D Petty S Connolly B et al 2017 ApJ 844 106Finkelstein S L Rhoads J E Malhotra S Pirzkal N amp

Wang J 2007 ApJ 660 1023Fiore F Puccetti S Brusa M et al 2009 ApJ 693 447Gandhi P Horst H Smette A et al 2009 AampA 502 457Goulding A D Zakamska N L Alexandroff R M et al

2018 ApJ 856 4Griffith R L Kirkpatrick J D Eisenhardt P R M et al

2012 AJ 144 148Hamann F Zakamska N L Ross N et al 2017 MNRAS

464 3431Hickox R C amp Alexander D M 2018 ARAampA 56 625Hopkins P F Hernquist L Cox T J amp Keres D 2008

ApJS 175 356Jun H et al in prepLeitherer C Ekstrom S Meynet G et al 2014 ApJS 212 14Leitherer C Ortiz Otalvaro P A Bresolin F et al 2010

ApJS 189 309Leitherer C Schaerer D Goldader J D et al 1999 ApJS

123 3Lotz J M Primack J amp Madau P 2004 AJ 128 163Lotz J M Davis M Faber S M et al 2008 ApJ 672 177Lupton R Blanton M R Fekete G et al 2004 PASP 116

133Magorrian J Tremaine S Richstone D et al 1998 AJ 115

2285Maiolino R Marconi A Salvati M et al 2001 AampA 365 28Mancone C L amp Gonzalez A H 2012 PASP 124 606Mateos S Carrera F J Alonso-Herrero A et al 2015

MNRAS 449 1422Petrosian V 1976 ApJ 209 L1Piconcelli E Vignali C Bianchi S et al 2015 AampA 574 L9Rhodes J Refregier A amp Groth E J 2000 ApJ 536 79Ricci C Assef R J Stern D et al 2017 ApJ 835 105Stern D 2015 ApJ 807 129Stern D Lansbury G B Assef R J et al 2014 ApJ 794 102Tsai C-W Eisenhardt P R M Wu J et al 2015 ApJ 805

90Tsai C-W Eisenhardt P R M Jun H D et al 2018 ApJ

868 15van Dokkum P G 2001 PASP 113 1420Vazquez G A amp Leitherer C 2005 ApJ 621 695Vito F Brandt W N Stern D et al 2018 MNRAS 474 4528Wright E L Eisenhardt P R M Mainzer A K et al 2010

AJ 140 1868Wu J Tsai C-W Sayers J et al 2012 ApJ 756 96Wu J Bussmann R S Tsai C-W et al 2014 ApJ 793 8Wu J Jun H D Assef R J et al 2018 ApJ 852 96Yan W Hickox R C Hainline K N et al 2019 ApJ 870 33

11

Figure 10 (Upper panel) The solid black line shows the best-fitStarburst99 SED model to the UVoptical broad-band photome-try of W0116ndash0505 assuming solar metallicity (see text for details)The gray shaded area shows all SED shapes within the 90 confi-dence interval (Bottom panel) Same as in the top panel but for ametallicity of Z = 0001

Figure 11 Same as Fig 10 but for W0220+0137

extended as found in sect23This suggests that the most likely source of the blue

excess emission in the three BHDs studied is scatteredlight In this scenario 99 of the emission from theaccretion disk and the broad-line region would be ab-

Figure 12 The contours show a χ2 map of the best-fit Star-burts99 models as discussed in the text The contours for W0116ndash0505 and W0220+0137 are shown in the top and bottom panelsrespectively Dark contours assume solar metallicity while graycontours assume Z = 0001 The solid (dotted) contour shows the683 (90) confidence region while the solid dots show the valuesof the best-fits models shown in Figures 10 and 11

sorbed by dust while 1 will be scattered into our lineof sight by either dust or gas or both Reflection nebu-lae are known to make the reflected SED bluer than theemitted SED in the UV (λ 2500A eg see Draine2003a) suggesting that the scattering medium in BHDsis more likely the gas surrounding the AGN Howeveras there is likely dust on scales larger than the torus(Dıaz-Santos et al 2016 Tsai et al 2018) we note thatour data are not sufficient to rule out an SED that hasbeen made bluer by dust reflection and redder by dustabsorption With respect to the X-rays we note thatwhile they should also be scattered into our line of sightalong with the UV emission the scattering cross sectionby either dust or free electrons is significantly smaller atthe high energy ranges probed by the Chandra observa-tions (Draine 2003ab) Hence the non-detection of aclear unabsorbed component in the X-ray spectra in sect3is consistent with this scattering scenario

5 SUMMARY

We have investigated the source of the blue excessemission in three BHDs two of which were identi-fied as such by A16 and a third one which has anSED consistent with that of a Hot DOG although itdoes not meet the formal selection criteria due to be-ing slightly too bright in the W1 band While allHot DOGs are characterized by mid-IR emission thatis most naturally explained by a highly obscured hyper-luminous AGN (Eisenhardt et al 2012 Assef et al 2015Tsai et al 2015) BHDs have a UVoptical SED thatis significantly bluer than expected based on templatefitting results Using a similar approach to that of

12

Assef et al (2015) we find that the SEDs of BHDs arebest modeled using two AGN components a primaryhyper-luminous highly obscured AGN that dominatesthe mid-IR emission and a secondary lower luminos-ity but unobscured AGN that dominates the UVopticalemission The bolometric luminosity of the secondaryAGN SED is sim1 of that of the primary componentA16 identified three possible scenarios to produce the ex-cess blue emission namely (i) a secondary less luminousbut unobscured AGN in the system (ii) an extreme star-burst or (iii) leaked UVoptical light from the primaryhighly luminous highly obscured AGN that dominatesthe mid-IRFor one of the sources (W0204ndash0506) A16 ruled out

a secondary AGN as the source of the blue excess emis-sion and instead concluded that the excess was causedby either unobscured star formation with an SFR amp1000 M⊙ yrminus1 or by UVoptical light from the cen-tral engine leaking into our line of sight due to scatter-ing or through a partially obscured sight-line with thescattered AGN light hypothesis deemed more likely Inthis paper we have presented HSTWFC3 imaging ofW0204ndash0506 showing a morphology consistent with anon-going merger and evidence of an on-going widespreadstarburst Considering however the very high SFRneeded to explain the UV emission by star-formationalone we conclude it is more likely that the UV emissionof W0204ndash0505 arises from a combination of scatteredAGN light and star-formationWe also studied in detail two other BHDs W0116ndash

0505 and W0220+0137 We present observations ob-tained with ChandraACIS-S and interpret them us-ing the Brightman amp Nandra (2011) models We findthat the X-ray spectra are consistent with single lumi-nous highly absorbed AGNs dominating the X-ray emis-sion We find that the L2minus10 keV luminosities of theseAGNs are consistent with those expected for the pri-mary AGNs based on their estimated L6microm according tothe L6microm minus L2minus10 keV relation of Stern (2015) We alsofind that the gas-to-dust ratios of the AGNs in thesesystems are somewhat below the median value foundin AGNs by Maiolino et al (2001) and lower than thatfound in W0204ndash0506 suggestive of a lower metallicityor of a higher fraction of absorbing gas within the dust-sublimation radius of the AGN Based on the UV throughmid-IR SED models of these sources we estimate the ex-pected X-ray luminosity of the putative secondary AGNcomponents assuming it is a second independent AGNin the system We found that the X-ray observations areonly marginally consistent with the presence of a secondAGN component in bothW0116ndash0505 andW0220+0137suggesting the dual AGN scenario is unlikelyWe followed A16 and modeled the UV emission of

W0116ndash0505 and W0220+0137 assuming a pure star-burst scenario and found that while the best-fit SFRsare generally high comparable to those found by A16for W0204ndash0506 they are not well constrained due tothe large degeneracies between SFR age and metallicityWe found however that the χ2 values of the best-fit star-burst models are large (sim 12 for W0116ndash0506 and sim 8for W0220+0137) despite the small number of degrees offreedom (1 and 2 respectively) implying a pure starburstis not a good description of the observed UV SED Addi-tionally the rest-frame UV spectra shows broad emission

lines characteristic of AGN activity further suggestingthat star-formation does not dominate the observed UVemissionFinally we also studied the morphologies observed in

the HSTWFC3 F555W and F160W images of W0116ndash0505 and W0220+0137 and found them to be undis-turbed with the UV emission being centrally concen-trated An analysis based on the Gini M20 and A coef-ficients showed that these systems are best characterizedas undisturbed early type galaxies consistent with theleaked AGN scenario Considering all of this we con-clude that the source of the UV emission in W0116ndash0505and W0220+0137 is scattered light from the hyperlu-minous highly obscured AGN that powers the mid-IRSED Given the detail of our data and SED modelingwe cannot determine whether the scattering material isprimarily gas dust or a mixture of bothThat all three BHDs we have investigated are due to

scattered light from the highly obscured hyperluminousAGN highlights how powerful the central engine is in HotDOGs with only 1 of the emission of the accretiondisk scattered into our line of sight it is still more lumi-nous than the entire stellar emission of the host galaxy inthe UV This is in general agreement with recent resultswhich show that the SMBHs in Hot DOGs are accretingabove the Eddington limit (Wu et al 2018 Tsai et al2018) and are injecting large amounts of energy into theISM of their host galaxies (Dıaz-Santos et al 2016) andhence are experiencing strong events of AGN feedback

We thank J Comerford and B Weiner for carry-ing out observations presented in this article RJAwas supported by FONDECYT grants number 1151408and 1191124 DJW acknowledges financial supportfrom STFC Ernest Rutherford fellowships HDJwas supported by Basic Science Research Programthrough the National Research Foundation of Korea(NRF) funded by the Ministry of Education (NRF-2017R1A6A3A04005158) FEB acknowledges supportfrom CONICYT-Chile (Basal AFB-170002) and the Min-istry of Economy Development and Tourismrsquos MilleniumScience Initiative through grant IC120009 awarded toThe Millenium Institute of Astrophysics MAS JW issupported by the NSFC Grant 11690024 and SPRP CASgrant XDB23000000 Based on observations made withthe NASAESA Hubble Space Telescope obtained at theSpace Telescope Science Institute which is operated bythe Association of Universities for Research in Astron-omy Inc under NASA contract NAS 5-26555 Theseobservations are associated with program 14358 Sup-port for program14358 was provided by NASA througha grant from the Space Telescope Science Institute whichis operated by the Association of Universities for Re-search in Astronomy Inc under NASA contract NAS5-26555 The scientific results reported in this articleare based to a significant degree on data obtained fromthe Chandra X-ray Observatory and observations madeby the Chandra X-ray Observatory and published previ-ously in cited articles Support for this work was pro-vided by the National Aeronautics and Space Adminis-tration through Chandra Award Number 17700696 is-sued by the Chandra X-ray Center which is operatedby the Smithsonian Astrophysical Observatory for and

13

on behalf of the National Aeronautics Space Adminis-tration under contract NAS8-03060 This publicationmakes use of data products from the Wide-field InfraredSurvey Explorer which is a joint project of the Universityof California Los Angeles and the Jet Propulsion Labo-ratoryCalifornia Institute of Technology funded by theNational Aeronautics and Space Administration Thiswork is based in part on observations made with theSpitzer Space Telescope which is operated by the JetPropulsion Laboratory California Institute of Technol-ogy under a contract with NASA The Pan-STARRS1Surveys (PS1) have been made possible through con-tributions of the Institute for Astronomy the Univer-sity of Hawaii the Pan-STARRS Project Office theMax-Planck Society and its participating institutes theMax Planck Institute for Astronomy Heidelberg andthe Max Planck Institute for Extraterrestrial PhysicsGarching The Johns Hopkins University Durham Uni-versity the University of Edinburgh Queenrsquos Univer-sity Belfast the Harvard-Smithsonian Center for Astro-physics the Las Cumbres Observatory Global TelescopeNetwork Incorporated the National Central Universityof Taiwan the Space Telescope Science Institute theNational Aeronautics and Space Administration underGrant No NNX08AR22G issued through the PlanetaryScience Division of the NASA Science Mission Direc-torate the National Science Foundation under Grant NoAST-1238877 the University of Maryland and EotvosLorand University (ELTE) and the Los Alamos NationalLaboratory Funding for SDSS-III has been provided bythe Alfred P Sloan Foundation the Participating Insti-tutions the National Science Foundation and the USDepartment of Energy Office of Science The SDSS-III web site is httpwwwsdss3org SDSS-III ismanaged by the Astrophysical Research Consortium forthe Participating Institutions of the SDSS-III Collabo-ration including the University of Arizona the BrazilianParticipation Group Brookhaven National LaboratoryCarnegie Mellon University University of Florida theFrench Participation Group the German ParticipationGroup Harvard University the Instituto de Astrofisicade Canarias the Michigan StateNotre DameJINA Par-ticipation Group Johns Hopkins University LawrenceBerkeley National Laboratory Max Planck Institute forAstrophysics Max Planck Institute for ExtraterrestrialPhysics New Mexico State University New York Uni-versity Ohio State University Pennsylvania State Uni-versity University of Portsmouth Princeton Universitythe Spanish Participation Group University of TokyoUniversity of Utah Vanderbilt University University ofVirginia University of Washington and Yale UniversitySome of the observations reported here were obtained atthe MMT Observatory a joint facility of the SmithsonianInstitution and the University of Arizona

REFERENCES

Abraham R G van den Bergh S amp Nair P 2003 ApJ 588218

Alexander D M amp Hickox R C 2012 NewAR 56 93Assef R J Kochanek C S Brodwin M et al 2010 ApJ 713

970Assef R J Denney K D Kochanek C S et al 2011 ApJ

742 93Assef R J Eisenhardt P R M Stern D et al 2015 ApJ

804 27

Assef R J Walton D J Brightman M et al 2016 ApJ 819111

Banerji M Alaghband-Zadeh S Hewett P C amp McMahonR G 2015 MNRAS 447 3368

Barger A J Cowie L L Chen C-C et al 2014 ApJ 784 9Bauer F E Yan L Sajina A amp Alexander D M 2010 ApJ

710 212Bennert V N Auger M W Treu T Woo J-H amp Malkan

M A 2011 ApJ 726 59Bertin E amp Arnouts S 1996 AampAS 117 393Brightman M amp Nandra K 2011 MNRAS 413 1206Cash W 1979 ApJ 228 939Chen C-T J Hickox R C Goulding A D et al 2017 ApJ

837 145Conselice C J 2014 ARAampA 52 291Cutri R M Wright E L Conrow T et al 2012 Explanatory

Supplement to the WISE All-Sky Data Release Products Techrep

Dıaz-Santos T Assef R J Blain A W et al 2016 ApJ 816L6

mdash 2018 Science 362 1034Draine B T 2003a ApJ 598 1017mdash 2003b ApJ 598 1026Eisenhardt P R M Wu J Tsai C-W et al 2012 ApJ 755

173Fan L Han Y Nikutta R Drouart G amp Knudsen K K

2016a ApJ 823 107Fan L Han Y Fang G et al 2016b ApJ 822 L32Farrah D Petty S Connolly B et al 2017 ApJ 844 106Finkelstein S L Rhoads J E Malhotra S Pirzkal N amp

Wang J 2007 ApJ 660 1023Fiore F Puccetti S Brusa M et al 2009 ApJ 693 447Gandhi P Horst H Smette A et al 2009 AampA 502 457Goulding A D Zakamska N L Alexandroff R M et al

2018 ApJ 856 4Griffith R L Kirkpatrick J D Eisenhardt P R M et al

2012 AJ 144 148Hamann F Zakamska N L Ross N et al 2017 MNRAS

464 3431Hickox R C amp Alexander D M 2018 ARAampA 56 625Hopkins P F Hernquist L Cox T J amp Keres D 2008

ApJS 175 356Jun H et al in prepLeitherer C Ekstrom S Meynet G et al 2014 ApJS 212 14Leitherer C Ortiz Otalvaro P A Bresolin F et al 2010

ApJS 189 309Leitherer C Schaerer D Goldader J D et al 1999 ApJS

123 3Lotz J M Primack J amp Madau P 2004 AJ 128 163Lotz J M Davis M Faber S M et al 2008 ApJ 672 177Lupton R Blanton M R Fekete G et al 2004 PASP 116

133Magorrian J Tremaine S Richstone D et al 1998 AJ 115

2285Maiolino R Marconi A Salvati M et al 2001 AampA 365 28Mancone C L amp Gonzalez A H 2012 PASP 124 606Mateos S Carrera F J Alonso-Herrero A et al 2015

MNRAS 449 1422Petrosian V 1976 ApJ 209 L1Piconcelli E Vignali C Bianchi S et al 2015 AampA 574 L9Rhodes J Refregier A amp Groth E J 2000 ApJ 536 79Ricci C Assef R J Stern D et al 2017 ApJ 835 105Stern D 2015 ApJ 807 129Stern D Lansbury G B Assef R J et al 2014 ApJ 794 102Tsai C-W Eisenhardt P R M Wu J et al 2015 ApJ 805

90Tsai C-W Eisenhardt P R M Jun H D et al 2018 ApJ

868 15van Dokkum P G 2001 PASP 113 1420Vazquez G A amp Leitherer C 2005 ApJ 621 695Vito F Brandt W N Stern D et al 2018 MNRAS 474 4528Wright E L Eisenhardt P R M Mainzer A K et al 2010

AJ 140 1868Wu J Tsai C-W Sayers J et al 2012 ApJ 756 96Wu J Bussmann R S Tsai C-W et al 2014 ApJ 793 8Wu J Jun H D Assef R J et al 2018 ApJ 852 96Yan W Hickox R C Hainline K N et al 2019 ApJ 870 33

12

Assef et al (2015) we find that the SEDs of BHDs arebest modeled using two AGN components a primaryhyper-luminous highly obscured AGN that dominatesthe mid-IR emission and a secondary lower luminos-ity but unobscured AGN that dominates the UVopticalemission The bolometric luminosity of the secondaryAGN SED is sim1 of that of the primary componentA16 identified three possible scenarios to produce the ex-cess blue emission namely (i) a secondary less luminousbut unobscured AGN in the system (ii) an extreme star-burst or (iii) leaked UVoptical light from the primaryhighly luminous highly obscured AGN that dominatesthe mid-IRFor one of the sources (W0204ndash0506) A16 ruled out

a secondary AGN as the source of the blue excess emis-sion and instead concluded that the excess was causedby either unobscured star formation with an SFR amp1000 M⊙ yrminus1 or by UVoptical light from the cen-tral engine leaking into our line of sight due to scatter-ing or through a partially obscured sight-line with thescattered AGN light hypothesis deemed more likely Inthis paper we have presented HSTWFC3 imaging ofW0204ndash0506 showing a morphology consistent with anon-going merger and evidence of an on-going widespreadstarburst Considering however the very high SFRneeded to explain the UV emission by star-formationalone we conclude it is more likely that the UV emissionof W0204ndash0505 arises from a combination of scatteredAGN light and star-formationWe also studied in detail two other BHDs W0116ndash

0505 and W0220+0137 We present observations ob-tained with ChandraACIS-S and interpret them us-ing the Brightman amp Nandra (2011) models We findthat the X-ray spectra are consistent with single lumi-nous highly absorbed AGNs dominating the X-ray emis-sion We find that the L2minus10 keV luminosities of theseAGNs are consistent with those expected for the pri-mary AGNs based on their estimated L6microm according tothe L6microm minus L2minus10 keV relation of Stern (2015) We alsofind that the gas-to-dust ratios of the AGNs in thesesystems are somewhat below the median value foundin AGNs by Maiolino et al (2001) and lower than thatfound in W0204ndash0506 suggestive of a lower metallicityor of a higher fraction of absorbing gas within the dust-sublimation radius of the AGN Based on the UV throughmid-IR SED models of these sources we estimate the ex-pected X-ray luminosity of the putative secondary AGNcomponents assuming it is a second independent AGNin the system We found that the X-ray observations areonly marginally consistent with the presence of a secondAGN component in bothW0116ndash0505 andW0220+0137suggesting the dual AGN scenario is unlikelyWe followed A16 and modeled the UV emission of

W0116ndash0505 and W0220+0137 assuming a pure star-burst scenario and found that while the best-fit SFRsare generally high comparable to those found by A16for W0204ndash0506 they are not well constrained due tothe large degeneracies between SFR age and metallicityWe found however that the χ2 values of the best-fit star-burst models are large (sim 12 for W0116ndash0506 and sim 8for W0220+0137) despite the small number of degrees offreedom (1 and 2 respectively) implying a pure starburstis not a good description of the observed UV SED Addi-tionally the rest-frame UV spectra shows broad emission

lines characteristic of AGN activity further suggestingthat star-formation does not dominate the observed UVemissionFinally we also studied the morphologies observed in

the HSTWFC3 F555W and F160W images of W0116ndash0505 and W0220+0137 and found them to be undis-turbed with the UV emission being centrally concen-trated An analysis based on the Gini M20 and A coef-ficients showed that these systems are best characterizedas undisturbed early type galaxies consistent with theleaked AGN scenario Considering all of this we con-clude that the source of the UV emission in W0116ndash0505and W0220+0137 is scattered light from the hyperlu-minous highly obscured AGN that powers the mid-IRSED Given the detail of our data and SED modelingwe cannot determine whether the scattering material isprimarily gas dust or a mixture of bothThat all three BHDs we have investigated are due to

scattered light from the highly obscured hyperluminousAGN highlights how powerful the central engine is in HotDOGs with only 1 of the emission of the accretiondisk scattered into our line of sight it is still more lumi-nous than the entire stellar emission of the host galaxy inthe UV This is in general agreement with recent resultswhich show that the SMBHs in Hot DOGs are accretingabove the Eddington limit (Wu et al 2018 Tsai et al2018) and are injecting large amounts of energy into theISM of their host galaxies (Dıaz-Santos et al 2016) andhence are experiencing strong events of AGN feedback

We thank J Comerford and B Weiner for carry-ing out observations presented in this article RJAwas supported by FONDECYT grants number 1151408and 1191124 DJW acknowledges financial supportfrom STFC Ernest Rutherford fellowships HDJwas supported by Basic Science Research Programthrough the National Research Foundation of Korea(NRF) funded by the Ministry of Education (NRF-2017R1A6A3A04005158) FEB acknowledges supportfrom CONICYT-Chile (Basal AFB-170002) and the Min-istry of Economy Development and Tourismrsquos MilleniumScience Initiative through grant IC120009 awarded toThe Millenium Institute of Astrophysics MAS JW issupported by the NSFC Grant 11690024 and SPRP CASgrant XDB23000000 Based on observations made withthe NASAESA Hubble Space Telescope obtained at theSpace Telescope Science Institute which is operated bythe Association of Universities for Research in Astron-omy Inc under NASA contract NAS 5-26555 Theseobservations are associated with program 14358 Sup-port for program14358 was provided by NASA througha grant from the Space Telescope Science Institute whichis operated by the Association of Universities for Re-search in Astronomy Inc under NASA contract NAS5-26555 The scientific results reported in this articleare based to a significant degree on data obtained fromthe Chandra X-ray Observatory and observations madeby the Chandra X-ray Observatory and published previ-ously in cited articles Support for this work was pro-vided by the National Aeronautics and Space Adminis-tration through Chandra Award Number 17700696 is-sued by the Chandra X-ray Center which is operatedby the Smithsonian Astrophysical Observatory for and

13

on behalf of the National Aeronautics Space Adminis-tration under contract NAS8-03060 This publicationmakes use of data products from the Wide-field InfraredSurvey Explorer which is a joint project of the Universityof California Los Angeles and the Jet Propulsion Labo-ratoryCalifornia Institute of Technology funded by theNational Aeronautics and Space Administration Thiswork is based in part on observations made with theSpitzer Space Telescope which is operated by the JetPropulsion Laboratory California Institute of Technol-ogy under a contract with NASA The Pan-STARRS1Surveys (PS1) have been made possible through con-tributions of the Institute for Astronomy the Univer-sity of Hawaii the Pan-STARRS Project Office theMax-Planck Society and its participating institutes theMax Planck Institute for Astronomy Heidelberg andthe Max Planck Institute for Extraterrestrial PhysicsGarching The Johns Hopkins University Durham Uni-versity the University of Edinburgh Queenrsquos Univer-sity Belfast the Harvard-Smithsonian Center for Astro-physics the Las Cumbres Observatory Global TelescopeNetwork Incorporated the National Central Universityof Taiwan the Space Telescope Science Institute theNational Aeronautics and Space Administration underGrant No NNX08AR22G issued through the PlanetaryScience Division of the NASA Science Mission Direc-torate the National Science Foundation under Grant NoAST-1238877 the University of Maryland and EotvosLorand University (ELTE) and the Los Alamos NationalLaboratory Funding for SDSS-III has been provided bythe Alfred P Sloan Foundation the Participating Insti-tutions the National Science Foundation and the USDepartment of Energy Office of Science The SDSS-III web site is httpwwwsdss3org SDSS-III ismanaged by the Astrophysical Research Consortium forthe Participating Institutions of the SDSS-III Collabo-ration including the University of Arizona the BrazilianParticipation Group Brookhaven National LaboratoryCarnegie Mellon University University of Florida theFrench Participation Group the German ParticipationGroup Harvard University the Instituto de Astrofisicade Canarias the Michigan StateNotre DameJINA Par-ticipation Group Johns Hopkins University LawrenceBerkeley National Laboratory Max Planck Institute forAstrophysics Max Planck Institute for ExtraterrestrialPhysics New Mexico State University New York Uni-versity Ohio State University Pennsylvania State Uni-versity University of Portsmouth Princeton Universitythe Spanish Participation Group University of TokyoUniversity of Utah Vanderbilt University University ofVirginia University of Washington and Yale UniversitySome of the observations reported here were obtained atthe MMT Observatory a joint facility of the SmithsonianInstitution and the University of Arizona

REFERENCES

Abraham R G van den Bergh S amp Nair P 2003 ApJ 588218

Alexander D M amp Hickox R C 2012 NewAR 56 93Assef R J Kochanek C S Brodwin M et al 2010 ApJ 713

970Assef R J Denney K D Kochanek C S et al 2011 ApJ

742 93Assef R J Eisenhardt P R M Stern D et al 2015 ApJ

804 27

Assef R J Walton D J Brightman M et al 2016 ApJ 819111

Banerji M Alaghband-Zadeh S Hewett P C amp McMahonR G 2015 MNRAS 447 3368

Barger A J Cowie L L Chen C-C et al 2014 ApJ 784 9Bauer F E Yan L Sajina A amp Alexander D M 2010 ApJ

710 212Bennert V N Auger M W Treu T Woo J-H amp Malkan

M A 2011 ApJ 726 59Bertin E amp Arnouts S 1996 AampAS 117 393Brightman M amp Nandra K 2011 MNRAS 413 1206Cash W 1979 ApJ 228 939Chen C-T J Hickox R C Goulding A D et al 2017 ApJ

837 145Conselice C J 2014 ARAampA 52 291Cutri R M Wright E L Conrow T et al 2012 Explanatory

Supplement to the WISE All-Sky Data Release Products Techrep

Dıaz-Santos T Assef R J Blain A W et al 2016 ApJ 816L6

mdash 2018 Science 362 1034Draine B T 2003a ApJ 598 1017mdash 2003b ApJ 598 1026Eisenhardt P R M Wu J Tsai C-W et al 2012 ApJ 755

173Fan L Han Y Nikutta R Drouart G amp Knudsen K K

2016a ApJ 823 107Fan L Han Y Fang G et al 2016b ApJ 822 L32Farrah D Petty S Connolly B et al 2017 ApJ 844 106Finkelstein S L Rhoads J E Malhotra S Pirzkal N amp

Wang J 2007 ApJ 660 1023Fiore F Puccetti S Brusa M et al 2009 ApJ 693 447Gandhi P Horst H Smette A et al 2009 AampA 502 457Goulding A D Zakamska N L Alexandroff R M et al

2018 ApJ 856 4Griffith R L Kirkpatrick J D Eisenhardt P R M et al

2012 AJ 144 148Hamann F Zakamska N L Ross N et al 2017 MNRAS

464 3431Hickox R C amp Alexander D M 2018 ARAampA 56 625Hopkins P F Hernquist L Cox T J amp Keres D 2008

ApJS 175 356Jun H et al in prepLeitherer C Ekstrom S Meynet G et al 2014 ApJS 212 14Leitherer C Ortiz Otalvaro P A Bresolin F et al 2010

ApJS 189 309Leitherer C Schaerer D Goldader J D et al 1999 ApJS

123 3Lotz J M Primack J amp Madau P 2004 AJ 128 163Lotz J M Davis M Faber S M et al 2008 ApJ 672 177Lupton R Blanton M R Fekete G et al 2004 PASP 116

133Magorrian J Tremaine S Richstone D et al 1998 AJ 115

2285Maiolino R Marconi A Salvati M et al 2001 AampA 365 28Mancone C L amp Gonzalez A H 2012 PASP 124 606Mateos S Carrera F J Alonso-Herrero A et al 2015

MNRAS 449 1422Petrosian V 1976 ApJ 209 L1Piconcelli E Vignali C Bianchi S et al 2015 AampA 574 L9Rhodes J Refregier A amp Groth E J 2000 ApJ 536 79Ricci C Assef R J Stern D et al 2017 ApJ 835 105Stern D 2015 ApJ 807 129Stern D Lansbury G B Assef R J et al 2014 ApJ 794 102Tsai C-W Eisenhardt P R M Wu J et al 2015 ApJ 805

90Tsai C-W Eisenhardt P R M Jun H D et al 2018 ApJ

868 15van Dokkum P G 2001 PASP 113 1420Vazquez G A amp Leitherer C 2005 ApJ 621 695Vito F Brandt W N Stern D et al 2018 MNRAS 474 4528Wright E L Eisenhardt P R M Mainzer A K et al 2010

AJ 140 1868Wu J Tsai C-W Sayers J et al 2012 ApJ 756 96Wu J Bussmann R S Tsai C-W et al 2014 ApJ 793 8Wu J Jun H D Assef R J et al 2018 ApJ 852 96Yan W Hickox R C Hainline K N et al 2019 ApJ 870 33

13

on behalf of the National Aeronautics Space Adminis-tration under contract NAS8-03060 This publicationmakes use of data products from the Wide-field InfraredSurvey Explorer which is a joint project of the Universityof California Los Angeles and the Jet Propulsion Labo-ratoryCalifornia Institute of Technology funded by theNational Aeronautics and Space Administration Thiswork is based in part on observations made with theSpitzer Space Telescope which is operated by the JetPropulsion Laboratory California Institute of Technol-ogy under a contract with NASA The Pan-STARRS1Surveys (PS1) have been made possible through con-tributions of the Institute for Astronomy the Univer-sity of Hawaii the Pan-STARRS Project Office theMax-Planck Society and its participating institutes theMax Planck Institute for Astronomy Heidelberg andthe Max Planck Institute for Extraterrestrial PhysicsGarching The Johns Hopkins University Durham Uni-versity the University of Edinburgh Queenrsquos Univer-sity Belfast the Harvard-Smithsonian Center for Astro-physics the Las Cumbres Observatory Global TelescopeNetwork Incorporated the National Central Universityof Taiwan the Space Telescope Science Institute theNational Aeronautics and Space Administration underGrant No NNX08AR22G issued through the PlanetaryScience Division of the NASA Science Mission Direc-torate the National Science Foundation under Grant NoAST-1238877 the University of Maryland and EotvosLorand University (ELTE) and the Los Alamos NationalLaboratory Funding for SDSS-III has been provided bythe Alfred P Sloan Foundation the Participating Insti-tutions the National Science Foundation and the USDepartment of Energy Office of Science The SDSS-III web site is httpwwwsdss3org SDSS-III ismanaged by the Astrophysical Research Consortium forthe Participating Institutions of the SDSS-III Collabo-ration including the University of Arizona the BrazilianParticipation Group Brookhaven National LaboratoryCarnegie Mellon University University of Florida theFrench Participation Group the German ParticipationGroup Harvard University the Instituto de Astrofisicade Canarias the Michigan StateNotre DameJINA Par-ticipation Group Johns Hopkins University LawrenceBerkeley National Laboratory Max Planck Institute forAstrophysics Max Planck Institute for ExtraterrestrialPhysics New Mexico State University New York Uni-versity Ohio State University Pennsylvania State Uni-versity University of Portsmouth Princeton Universitythe Spanish Participation Group University of TokyoUniversity of Utah Vanderbilt University University ofVirginia University of Washington and Yale UniversitySome of the observations reported here were obtained atthe MMT Observatory a joint facility of the SmithsonianInstitution and the University of Arizona

REFERENCES

Abraham R G van den Bergh S amp Nair P 2003 ApJ 588218

Alexander D M amp Hickox R C 2012 NewAR 56 93Assef R J Kochanek C S Brodwin M et al 2010 ApJ 713

970Assef R J Denney K D Kochanek C S et al 2011 ApJ

742 93Assef R J Eisenhardt P R M Stern D et al 2015 ApJ

804 27

Assef R J Walton D J Brightman M et al 2016 ApJ 819111

Banerji M Alaghband-Zadeh S Hewett P C amp McMahonR G 2015 MNRAS 447 3368

Barger A J Cowie L L Chen C-C et al 2014 ApJ 784 9Bauer F E Yan L Sajina A amp Alexander D M 2010 ApJ

710 212Bennert V N Auger M W Treu T Woo J-H amp Malkan

M A 2011 ApJ 726 59Bertin E amp Arnouts S 1996 AampAS 117 393Brightman M amp Nandra K 2011 MNRAS 413 1206Cash W 1979 ApJ 228 939Chen C-T J Hickox R C Goulding A D et al 2017 ApJ

837 145Conselice C J 2014 ARAampA 52 291Cutri R M Wright E L Conrow T et al 2012 Explanatory

Supplement to the WISE All-Sky Data Release Products Techrep

Dıaz-Santos T Assef R J Blain A W et al 2016 ApJ 816L6

mdash 2018 Science 362 1034Draine B T 2003a ApJ 598 1017mdash 2003b ApJ 598 1026Eisenhardt P R M Wu J Tsai C-W et al 2012 ApJ 755

173Fan L Han Y Nikutta R Drouart G amp Knudsen K K

2016a ApJ 823 107Fan L Han Y Fang G et al 2016b ApJ 822 L32Farrah D Petty S Connolly B et al 2017 ApJ 844 106Finkelstein S L Rhoads J E Malhotra S Pirzkal N amp

Wang J 2007 ApJ 660 1023Fiore F Puccetti S Brusa M et al 2009 ApJ 693 447Gandhi P Horst H Smette A et al 2009 AampA 502 457Goulding A D Zakamska N L Alexandroff R M et al

2018 ApJ 856 4Griffith R L Kirkpatrick J D Eisenhardt P R M et al

2012 AJ 144 148Hamann F Zakamska N L Ross N et al 2017 MNRAS

464 3431Hickox R C amp Alexander D M 2018 ARAampA 56 625Hopkins P F Hernquist L Cox T J amp Keres D 2008

ApJS 175 356Jun H et al in prepLeitherer C Ekstrom S Meynet G et al 2014 ApJS 212 14Leitherer C Ortiz Otalvaro P A Bresolin F et al 2010

ApJS 189 309Leitherer C Schaerer D Goldader J D et al 1999 ApJS

123 3Lotz J M Primack J amp Madau P 2004 AJ 128 163Lotz J M Davis M Faber S M et al 2008 ApJ 672 177Lupton R Blanton M R Fekete G et al 2004 PASP 116

133Magorrian J Tremaine S Richstone D et al 1998 AJ 115

2285Maiolino R Marconi A Salvati M et al 2001 AampA 365 28Mancone C L amp Gonzalez A H 2012 PASP 124 606Mateos S Carrera F J Alonso-Herrero A et al 2015

MNRAS 449 1422Petrosian V 1976 ApJ 209 L1Piconcelli E Vignali C Bianchi S et al 2015 AampA 574 L9Rhodes J Refregier A amp Groth E J 2000 ApJ 536 79Ricci C Assef R J Stern D et al 2017 ApJ 835 105Stern D 2015 ApJ 807 129Stern D Lansbury G B Assef R J et al 2014 ApJ 794 102Tsai C-W Eisenhardt P R M Wu J et al 2015 ApJ 805

90Tsai C-W Eisenhardt P R M Jun H D et al 2018 ApJ

868 15van Dokkum P G 2001 PASP 113 1420Vazquez G A amp Leitherer C 2005 ApJ 621 695Vito F Brandt W N Stern D et al 2018 MNRAS 474 4528Wright E L Eisenhardt P R M Mainzer A K et al 2010

AJ 140 1868Wu J Tsai C-W Sayers J et al 2012 ApJ 756 96Wu J Bussmann R S Tsai C-W et al 2014 ApJ 793 8Wu J Jun H D Assef R J et al 2018 ApJ 852 96Yan W Hickox R C Hainline K N et al 2019 ApJ 870 33

14

Zakamska N L Hamann F Paris I et al 2016 MNRAS 4593144