Physical Properties of Luminous Dust Poor Quasars
aa r X i v : . [ a s t r o - ph . C O ] O c t Draft version August 3, 2018
Preprint typeset using L A TEX style emulateapj v. 5/2/11
PHYSICAL PROPERTIES OF LUMINOUS DUST POOR QUASARS
Hyunsung David Jun and Myungshin Im Draft version August 3, 2018
ABSTRACTWe identify and characterize a population of luminous dust poor quasars at 0 < z <
5, similar inphotometric properties to the objects found at z > L bol > . erg s − . After fitting the broad-band spectral energy distributions(SEDs) with accretion disk and dust components, we find 0.6% of our sample to be hot dust poorwith a rest-frame 2.3 µ m to 0.51 µ m flux density ratio of − . Subject headings: galaxies: active — galaxies: evolution — infrared: galaxies — quasars: general INTRODUCTION
Active galactic nuclei (hereafter AGNs) are knownto be radiating through a wide range of wavelengths,where the unified model (e.g., Krolik & Begelman 1988;Antonucci 1993; Urry & Padovai 1995) postulates dustystructures surrounding the central black hole to be re-sponsible for the infrared emission. Thermally repro-cessed emission coming from heated dust, is prominent inthe infrared spectral energy distribution (hereafter SED)of AGNs. The hottest dust reaches temperatures up to1500 K and gives off near-infrared (hereafter NIR) excessradiation above the light from the accretion disk andthe host galaxy, but the dust does not get much hotterdue to sublimation of dust grains (e.g., Barvainis 1987).Hot dust in AGNs is common (e.g., Glikman et al. 2006),and is often modeled to lie in the inner dust torus (e.g.,Barvainis 1987).Recently, this common feature of NIR excess due to hotdust emission was found to be missing for a few quasarsat z ∼ , log ( L . /L . ),they found two quasars (out of 21 at z ∼
6) with excep-tionally small flux ratios such that the hot dust emissioncomponent is almost absent, while no such AGNs werefound at lower redshifts. These particular examples wereinterpreted as the existence of quasars before the forma-tion of dusty structure, observed in their early lifetimes.It was then by Hao et al. (2010, hereafter H10) whenthey selected hot dust poor quasars to show weak NIRcontinuum shapes from the diagram of NIR against op- Center for the Exploration of the Origin of the Universe(CEOU), Astronomy Program, Department of Physics and As-tronomy, Seoul National University, Seoul 151-742, Republic ofKorea; [email protected], [email protected]. We define L λ = L λ ( µ m) =4 πd L λF λ as the rest-framemonochromatic luminosity, where d L is the luminosity distanceand F λ is the rest-frame specific flux. tical slopes, and reported &
10% of quasars to be hotdust poor even at low redshifts, although their definitionof “hot dust poor” is different from that of J10. H10provided wider possibilities in explaining the deficiencyof hot dust radiation, raising dust destruction models asthe origin. Meanwhile, Mor & Trakhtenbrot (2011, here-after M11) found 1.7% of their quasars at redshifts of0.75 < z < , having zero hot dustcovering factors (hereafter CF hd = L hd /L bol ), and & ∼ σ distribu-tion of CF hd . Still contrary to J10, they could not finda redshift dependence on the distribution of CF hd , or onthe fraction of their hot dust poor quasars.Not only the number evolution but also the origin ofquasars to be hot dust poor, is yet controversial. Al-though J10 found hot dust poor quasars to have lowerblack hole masses ( M BH ) and higher Eddington ratios( f Edd ), H10 could not locate their hot dust poor quasarsto be clustered in a specific range of M BH , f Edd , orbolometric luminosity ( L bol ). Since both studies sufferfrom relatively small sample sizes of a few hundreds, alarger parent sample might help clarify the properties ofdust poor quasars sizable in number. Based on 15,000quasars with SDSS/WISE coverage, M11 still failed tofind a meaningful link between the hot dust coveringfactor with M BH or f Edd , and from the scatter in thecorrelations suggested that CF hd is independent on theevolutionary stage of the black hole. The latest study ofMa & Wang (2013) about the properties of warm dustcovering factors (CF wd ) of 12,000 quasars, agrees withM11 for the anti-correlation between CF wd and L bol , butdisagrees for the anti-correlation in CF wd and M BH , fur-ther complicating the case.Previous studies on dust poor quasars follow mutually Throughout this paper we use the term “dust poor” to indicateweak IR dust emission below a certain threshold, and “dust free”to be completely absent of dust emission.
Jun and Imdifferent definitions of being hot dust poor/free, thusadding an uncertainty when trying to compare the re-sults on each of selected dust poor populations. For lu-minous quasars, the selection criterion of J10 based onthe NIR-to-optical flux ratio, verifies that it picks outSEDs with an apparently weak NIR bump originatingfrom the hot dust radiation. On the other hand, for lessluminous ( L bol . erg s − ) quasars where host galaxycontamination to the SED becomes meaningful, it wouldbe better to use dust emission strength indicators takinginto account the fraction of host galaxy light (H10), orto apply a luminosity dependent average galaxy contami-nation correction to the observed fluxes (e.g., Shen et al.2011, hereafter S11). Moreover, the selection of hot dustpoor quasars up to date relies on the weak NIR emission,and it is yet unclear whether hot dust poor quasars arealso dust poor in warmer phases, where the mid-infrared(hereafter MIR) is thought to be the peak wavelength ofAGN dust emission (e.g., Richards et al. 2006). Lastly,all previous studies lack either the sample size (J10, H10)or redshift coverage (M11, Ma & Wang 2013) to mean-ingfully disentangle number evolution and physical pa-rameter characteristics of dust poor quasars.In this paper, we aim to identify the lower red-shift counterpart to the high redshift hot dust poorquasars of J10, to compare and understand the obser-vational features of local to high redshift populations.Where our sample quasars are defined to be luminousenough to have almost negligible host contamination( § §
2) of 41,000 quasars at0 < z < § § § H = 70 km s − Mpc − , Ω m = 0 .
3, and Ω Λ = 0 . SAMPLE DEFINITION AND AGN DATA SET
For the selection of hot dust poor quasars we followJ10 to use NIR-to-optical flux ratios, in order to indicatethe relative strength of the dust emission. Defining f λ asthe λ µ m-to-optical flux ratio , f λ = log ( λF λ / . F . ) = log ( L λ /L . ) , (1)we assign “hot dust poor” quasars to satisfy f . < − . f . is chosen for the 2.3 µ m data to be effective in con-straining the hot dust emission, as it is the wavelengthwhere the 1250 K black body component for the SED fit-ting ( § F λ space. In addition, the hotdust poor criterion is set to select objects with a weakNIR bump at 2.3 µ m, equivalent to J10 objects that arebelow the lower 3 σ distribution in f . ( . − . f . can be derived from We use rest-frame flux for F λ , F ν throughout this paper, inupper cases to prevent confusion with the rest-frame flux ratio f λ . Fig. 1.—
Graphical examples of two hot dust poor quasar SEDs(black lines) with f . = − . f . are cal-culated from the logarithmic ratio of 2.3 µ m to 0.51 µ m fluxes in λf λ space. These cases falling on the marginal limit of hot dustpoor quasar selection, correspond to SEDs that are either purelypower-law (black solid line) or power-law plus weak hot dust emis-sion (black dashed line), with both f . values lying below 3 σ ofthe distribution in J10. Overplotted are the two extreme SEDscomparable in f . to dust free quasars in J10 (gray lines). a range of α , in F ν ∝ ν α . In the figure, we show twoexample SEDs of different continuum slopes that satisfy f . = − .
5, with α lying within 1 σ to the average ofall quasars in this work. If α = − .
24, the SED (blacksolid line) is purely power-law in the optical–NIR, andwhen α = 0.10 the SED (black dashed line) is the sum ofa power-law, plus a weak hot dust emission componentsimilar in strength to that of J1411+1217 from J10. The f . for both examples are − .
64 and − .
57, which arehigher than f . . − σ range of f . in Figure2 of J10, separating weak hot dust radiating AGNs fromthe rest of the distribution.The search for hot dust poor quasars is based onthe Sloan Digital Sky Survey quasar catalog (DR9 ver-sion, Schneider et al. 2010; Pˆaris et al. 2012), which pro-vides the largest number (185,801) of optically selectedtype-1 AGNs together with ancillary AGN properties(S11), from a survey area of 14,555 deg . In additionto the SDSS optical imaging and spectroscopic data, wegathered UV through MIR imaging data from GALEX,2MASS, UKIDSS and WISE surveys (Martin et al. 2005;Skrutskie et al. 2006; Lawrence et al. 2007; Wright et al.2010) covering ∼ , all-sky, ∼ , andall-sky, respectively. The multi-wavelength data set en-ables the measurement of hot and warm dust emissionstrengths, f . and f , where f is to be used to probethe warm dust emission of hot dust poor quasars in com-parison with f . . Also, additional NIR data were ob-tained from our own UKIRT imaging observations, toimprove the photometric accuracy of 113 type-1 AGNsthat have low S/N in the 2MASS data, including 32 hotdust poor objects at z &
2. Out of the hot dust poorquasars observed by UKIRT, 19 satisfy the criterion of f . < − . ′′ radii, with multi-wavelength data from UVthrough MIR: GALEX GR7 for UV, 2MASS andUKIDSS DR10 for NIR, and WISE all-sky for MIR.ust poor quasars 3 Fig. 2.—
Left: An illustration of Sloan i -band limit for a marginal hot dust poor quasar with f . = − . z = 1, set for the rest-frame2.3 µ m to fall on the WISE (in this case W
2) S/N=2 detection limit. Under the i -band flux limit not only hot dust rich ( f . > − . µ m, hot dust poor ( f . < − .
5) objects are either detected or given with upper limits deeper than the 2.3 µ m fluxthat meets f . = − .
5, such that the classification of hot dust poor / rich quasars becomes more complete under the given WISE sensitivity.Right: The SDSS i -magnitude cut (solid line) as a function of redshift. f . = − . µ m detection limit for t exp = 97 s. Overplotted dots are objects satisfying the i -band sensitivity limit for their respective WISE limits,mostly preserving the main magnitude limit ( i < Since the GALEX and UKIDSS data consist of severalseparate surveys (AIS, MIS, DIS, NGS, GII, CAI forGALEX, and LAS, GCS, DXS for UKIDSS), we mergedeach catalog consisting the UV and NIR survey data re-spectively, with overlapping sources to have a single pho-tometry of better quality. The matching was performedfor all the UV–MIR data set, but for the 2MASS data,we instead took the matched catalog in Schneider et al.(2010), which adopts the closest neighbor for multiplematches of 1.5% occurrence. On the other hand, mul-tiple SDSS to UKIDSS NIR matches of 0.72% occur-rence were removed instead of keeping the closest pair,to prevent source confusion in UV/MIR data under poorseeing, where UKIDSS data have the best seeing to tellwhether or not there is a confusion. In the absence ofUKIDSS coverage the number of multiple SDSS-2MASSmatches could not be fully trusted, due to the higher in-cidence of multiple matching than that of SDSS-UKIDSSat an identical 2 ′′ matching radius. For this reason, wedid not reject the multiple SDSS-2MASS matched ob-jects without UKIDSS coverage in principle, but visu-ally re-examined the SDSS images of the vicinity of hotdust poor quasars ( § f . > − .
5) objectswithout UKIDSS data are exposed under a mild level ofconfusion, expected to be 0.72% from the SDSS-UKIDSSmatching, though we regard this fraction to be negligi-ble in counting the number of hot dust rich quasars lateron. Finally, objects undetected or blended in WISE weregiven with upper flux limits, while those undetected inthe NIR were removed, in order to provide careful con-straints on the dust emission strength.The photometric data collected from multi-wavelengthcatalogs can potentially suffer from the following prob-lems. First, the photometry of AGNs include contribu-tions from the host galaxy that needs to be minimizedor subtracted somehow. Second, the mixture of differ-ent definitions of Kron, PSF, aperture, and profile fitmagnitudes for UV, optical, NIR, and MIR data, brings inaccurate photometry of extended objects from PSF oraperture magnitudes. Both of these potential problemscan be avoided by minimizing the host galaxy contam-ination. Thus, we limited our sample with a bolomet-ric luminosity cut of L bol > . erg s − , derived from L . > . erg s − ( §
3) using the optical to bolomet-ric correction of 9.26 (S11). By doing so, host contam-ination is limited to less than 10% in 5100 ˚A(S11). Af-ter applying the luminosity cut, we double checked iffixed-aperture (PSF or aperture) magnitudes bring con-sistent photometry to total magnitudes by comparingPSF versus Petrosian magnitudes for SDSS ugriz filters,and aperture versus Petrosian magnitudes for 2MASS orUKIDSS
JHK filters. We found the median differencebetween the magnitude systems to be at most 0.03 magat all filters, with the rms scatter to fall within 0.1 magfor SDSS or UKIDSS data, and within 0.2 mag for that of2MASS. Therefore we consider our luminosity cut sam-ple to have a compatible set of magnitudes dominated bythe central AGN contribution.As a next step, we imposed an i -band flux limit thatvaries with redshift and the WISE depth for each ob-ject. This is necessary since the WISE flux limits are notdeep enough for some of the SDSS quasars to determineif the object is f . < − .
5. In other words, we selectedobjects that are bright enough in both observed-frame i - and WISE bands, to allow us to determine if the ob-ject is dust poor or not. Since each observed band tracesdifferent rest-frame wavelength at different redshift, the i -band cut changes as a function of redshift. Further-more, the WISE depths are not uniform over the entiresky, so that the i -band depth needs to be different ateach location on the sky. The i -band flux limits were de-termined by imposing a quasar with f . = − . i -magnitude, to have 2 σ detection in the WISE band thatcovers the rest-frame 2.3 µ m. Figure 2a shows a fiducial i -band magnitude limit which assumes the typical WISElimit with 97 s exposure (11 frames, Wright et al. 2010).When the exposure time, t exp , was different at the po-sition of another quasar, we adjusted the fiducial i -band Jun and Imlimit by adding 1.25 log( t exp /97 s). The objects passingthe i -band limit are depicted in Figure 2b.Lastly, we required at least two data points to lie inthe rest-frame 0.3–1 µ m with one of the points at shorterthan 0.6 µ m, so that the optical continuum slope could bereliably measured from the SED fitting. In short, book-keeping the number of matches through the sample selec-tion process, 100%, 54% of SDSS quasars have NIR cov-erage and detection, where almost all the undetectionscome from the shallow 2MASS data. From then, 100%,91% are covered and detected in the WISE MIR, yield-ing 49% of the initial catalog to be matched with con-tiguous rest-frame UV–NIR information, of which 82%are matched in the GALEX UV. The bolometric lumi-nosity cut passes through 81% of the multi-wavelengthdata, while the i -band limit leaves a further 56% of theremaining sample. Finally, the constraints on the shal-low WISE upper limits and the number of optical datapoints sum up to a 2% rejection, leaving 22% of the initialSDSS sample or 40,825 objects, for the SED fitting anal-ysis. We summarize the final data set and observationsin Table 1. FITTING OF BROAD-BAND SED AND SPECTRA
We modeled individual quasar SEDs within rest-frame0.3–20 µ m boundaries, as a combination of power-lawcontinuum feature from the accretion disk, and blackbody emission from the heated dust in hot and warmphases. For all quasars a default 1250 K hot dust emis-sion was considered because it well describes the NIRpart of the SED (Glikman et al. 2006). In cases wherethe rest-frame 3.5 µ m and 9 µ m were available, 500 K and200 K black bodies were added respectively to model theMIR continuum emission of AGNs through warm dustphases, where the combination of warm dust tempera-tures are found to well fit the observed SEDs of AGNs(e.g., Barvainis 1987; Hao et al. 2005). Quantitativelyexpressing the model SED as F λ = F λ,disk + F λ,dust , (2)it consists of an accretion disk component, F λ,disk = c disk λ − (2+ α ) , and a combination of dustemission components, F λ,dust = c hd B λ (1250 K) + c id B λ (500 K) + c wd B λ (200 K), where α is the power-law continuum slope in F ν ∝ ν α , and c hd , c id , c wd are contributions from the hot (1250 K), intermediate(500 K), and warm (200 K) dust black bodies. c id , c wd were used when the rest-frame IR data included3.5 µ m/9 µ m, the geometric mean of the peak wave-lengths out of hot–intermediate/intermediate–warmcomponents, while the coefficients were fixed to zerootherwise. The peak of the black body radiation fromhot and warm dust models in F λ space are at 2.3 and14 µ m each, which means that WISE covers hot dustemission mostly within W W z = 0–4,while warm dust emission are probed usually under theinclusion of W z . §
2) due to shallow WISE upper limitsor insufficient rest-frame optical coverage, respectively.The median reduced chi-square χ ν , are not the best at5.4 and 3.0, for all and hot dust poor objects separately. TABLE 1Summary of imaging data
Name Filters
N t exp (s)Cataloged dataGALEX FUV/NUV 17,797/33,741 368/1,504SDSS ugriz
JHK
Y JHK W W Y JHK
113 100
Note . — N is the number of objects detected in at least onefilter within the corresponding data set, out of the N=40,825 finalmulti-wavelength matched sample limited in bolometric luminos-ity, i -band flux, and the number of optical data points, from theinitial N=185,801 SDSS quasars. t exp is the median exposuretime of a single-band image in the corresponding data set. Still, we find the large χ ν sources ( § χ ν values do not indicate the poor determination of thecontinuum slope, since they are caused by wiggly fea-tures in the continuum such as broad emission lines orFeII complex, and photometric variabilities between dif-ferent wavelength data sets. A further test on the accu-racy of the SED fitting under special conditions, whenthe gap between the WISE W W f . , is described in § f . , f , to quantify hot/warm dustemission strengths over that of the optical continuum.We measured f . , f from the fluxes on the fitted SEDcurve at the corresponding rest-frame wavelengths, while f was computed only when the warm (200 K) dustcomponent was used for the SED fit. Upper limitsto f . , f were provided from the upper flux limitsfor the case of WISE undetection, assuming the upperflux limits as detections. The use of f . , f insteadof α, c disk , c hd , c id , c wd , is more straightforward to selectquasars with weak dust emission, as c hd , c id , c wd are eas-ily coupled with α which varies by object. An exam-ple is a red quasar small in α due to optical extinction,and large in f . , f from strong hot–warm dust emis-sion. Because the extrapolated red power-law componentcould take care of the NIR–MIR emission, the resultantsmall c hd , c id , c wd for instance, are not necessarily goodindicators of little dust emission.We wanted to pay careful attention for the modeling tobe robust under systematic effects. First, broad emissionlines could disturb the optical broad-band fluxes froma simple power-law continuum model, where H α is thestrongest and the only meaningful ( > α , if the χ ν containing that datapoint became larger than that without. We addition-ally note that, although the rest-frame 10 µ m is oftensurrounded by polycyclic aromatic hydrocarbon emissionand silicate absorption features, we expect the line con-taminations to be negligible considering the very wide fil-ter transmission of W W
4. Second, variability canbe a problem with multi-wavelength data taken at differ-ust poor quasars 5
Fig. 3.—
Top: An example of CIV λ λ − )from S11 and from our fits, are printed. Bottom: An exampleof CIV and CIII] both showing BAL features, where the M BH information is dropped. ent epochs, for it would possibly bring flux offsets. As asanity check we computed the scatter in the magnitudedifference between the 2MASS versus UKIDSS J and K AB magnitudes, to find the median rms of the mag-nitude difference to be ∼ M BH esti-mators in the form oflog (cid:16) M BH M ⊙ (cid:17) = a + b log (cid:16) L λ erg s − (cid:17) + c log (cid:16) F W HM km s − (cid:17) , (3)with L λ = ( L . , L . , L . ) and ( a, b, c )= { (6.69, 0.50,2.1), (6.91, 0.50, 2), (6.66, 0.53, 2) } , for spectral re-gions around (H α , H β , CIV) respectively. Forthe MgII estimator L λ = L . and ( b, c ) = (0.62, 2),while a =(6.75, 6.81, 6.79) depending on the nar-row line to be subtracted/included (DR7), or un-used for the fitting (DR9), individually. Each(H α , H β , MgII, CIV) recipe was originally fromGreene & Ho (2005), Vestergaard & Peterson (2006),McLure & Dunlop (2004), and Vestergaard & Peterson(2006), but we primarily followed the cross calibratedform of ( a, b ) from S11 to provide consistency between M BH from different lines. Within the following red-shift intervals, we used the black hole mass estima-tor of H α ( z < . β ( z < . . < z < z > M BH in S11 from different method- Fig. 4.—
Composite SED of hot dust poor (purple, down-ward triangles) and all (black circles) quasars, gridded on therest-frame 400, 700, 1000 ˚A–GALEX–SDSS–UKIDSS–WISE wave-lengths. Solid lines are polynomial fits for < µ m, and power-lawcontinuum+black body dust fits at > µ m. Individual hot dustpoor SEDs based on the detected data points are overplotted withthin gray lines, while we note that the MIR part are generallypoorer in sensitivity due to shallow W W ologies. The continuum and line fits by S11 were re-producible overall, but with cases of problematic fittedresults. First of all, we checked the H β fits by followingthe method of S11 except for using the FeII template ofTsuzuki et al. (2006), to fit the spectra around the H β region. For a total of 9 quasars at z < .
84, we found agood agreement in our H β M BH to S11 values, to havea − . ± β M BH of S11. Likewise, we per-formed CIV fits of 92 hot dust poor quasars at z > λ M BH of remaining 11 ob-jects. The CIV M BH without absorption spectra, areoffset by − . ± M BH for the better treatmentof absorption features.For the MgII fitting, we note that the DR7 and DR9spectra are treated differently in S11, such that the DR7broad MgII FWHMs are measured with and without sub-tracting the narrow ( < − ) component, whilethe narrow line component itself is not used for the DR9fitting. In order to shift each MgII mass estimator tobe mutually consistent, we followed the argument of S11to normalize each MgII M BH recipe to the H β M BH ofVestergaard & Peterson (2006), obtaining the coefficient a in Equation 3 while keeping b and c fixed. This pro-cess was performed for all 2489 objects covering bothH β and MgII emission lines at 0.7 < z < >
5. We find a = 6.75 when using the MgII linewidth measured subtracting the narrow component, con-sistent with 6.74 in S11, and a = 6.81 when the narrowcomponent is included. Next, we normalized the DR9MgII M BH with that of 1125 overlapping DR7 objects at Jun and Im Fig. 5.—
Selected list of hot dust poor quasar SEDs, sorted in redshift. Along with the observed data points and 2 σ upper limits, modelfits with accretion disk (blue) and dust components of T=1250, 500 and 200 K phases (red) are overplotted. We include the 500 K and200 K black bodies only when the longest wavelength of each SED exceeds 3.5 µ m and 9 µ m respectively, except for calculating upper limitsin f . / f . In addition, the composite SED of luminous ( L bol & erg s − ) SDSS quasars (gray, Richards et al. 2006) normalized tothe data at 5100 ˚A, are overplotted. SDSSJ144706.80+212839.3 at z = 3 .
23 (bottom, 2nd panel from left), is the quasar with the smallestNIR-to-optical flux ratio of f . = − .
88, reaching down to similar values to J10 dust free quasars.
TABLE 2Hot dust poor quasar properties
Name z f . f log L bol log M BH f Edd
J001754.17+011123.4 1.465 -0.77 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Note . — Catalog of hot dust poor quasars sorted in RA, where only the first five rows are displayed. The units for L bol and M BH are erg s − and M ⊙ . Empty parameters are fixed to 99. The entire table is published in the electronic edition ofthe paper. A portion is shown here for guidance regarding its form and content. < z < a = 6.79, irrespective of whether thenarrow MgII lines of DR7 quasars are subtracted or not,and even when these two DR7 masses are averaged. Sinceit is still debatable whether to subtract the narrow com-ponent for the MgII line width measurement (e.g., S11),we averaged the narrow component subtracted/included M BH for the DR7 sample giving the least scatter withthe overlapping DR9 M BH , but at the same time cautionon the accuracy of S11 MgII M BH until the fitting depen-dent systematic uncertainties are clarified in the future.Lastly, we substituted 7 MgII M BH s at z < L , which is derived fromour own broad-band SED fit. This approach may reducethe systematic uncertainties from extinction or BAL fea- tures, compared to when converting the rest-frame UVinto bolometric luminosity. We used a constant bolomet-ric correction of 9.26 (S11), derived from the compositequasar SED in Richards et al. (2006). To check whetherour photometrically determined 5100 ˚A luminosities arereliable, we compared our values with that from the spec-tral fitting in S11, to find a good agreement with a dif-ference of L − L ,S = 0.06 ± RESULTS
Number counts and SEDs of dust poor quasars
There are 253 objects that meet the hot dust poorquasar criterion f . < − .
5. Visually inspecting indi-vidual images and spectra we find 17 blended images inSDSS, and 3 stellar spectra that are rejected. After all,we keep 233 hot dust poor quasars out of a total of 40,805,which is 0.6% in fraction. Figure 4 shows the compos-ust poor quasars 7
Fig. 6.—
Correlations between optical and NIR monochromatic luminosities (left), and between MIR and NIR luminosities (right). Thedistributions of hot dust poor quasars (purple, filled circles represent detections and open arrows are based on WISE 2 σ upper limits atcorresponding wavelengths) and of the rest of the sample (black dots, detections only) are plotted, together with an overall median errorbar on the upper left of each panel. Linear fits to luminosity correlations are obtained by applying the ordinary least squares bisectormethod (Isobe et al. 1990) to the entire sample with detections, where the fitting coefficients and rms scatter are displayed. ite SED of hot dust poor quasars together with that ofthe whole sample. Both composites were constructed bythe geometric mean as to preserve the global continuumshape (Vanden Berk et al. 2001), after normalizing indi-vidual SEDs at 0.51 µ m. We fitted the composite SEDsby 3rd order polynomials in the UV ( < µ m) and withoptical power-law plus IR black body combinations ( § § µ m than the composite of all SEDs.Individual hot dust poor SEDs overplotted to theircomposite in Figure 4, are also depicted with model fitsin Figure 5, while their properties are summarized in Ta-ble 2. We find that the majority of hot dust poor quasarsexhibit warm dust emission under the detection of NIR–MIR, even if the hot dust component is negligible. Ex-ceptions are objects for which upper limits in the rest-frame MIR do not allow us, to assess whether they arelacking the warm dust component or not. Specifically,the optical–NIR spectral shapes of the upper W z > L . is in general proportional to L . , meaning luminous AGNs show stronger dust emis-sion than faint AGNs (e.g., Haas et al. 2003). With re-spect to this relation, we find that hot dust poor quasarsare objects defining the lower ∼ σ envelope. Next, Fig. 7.—
Distribution of f . along the continuum slope α (left),following the color and symbol layouts of Figure 6. The averageand standard deviation of α are displayed for both the entire anddust poor samples. The data points forming the diagonal line justabove the gray highlighted line, are SEDs fitted as purely power-law in the optical–NIR, with no hot dust component contribution( c hd = 0). The gray shaded region are prohibited since they require c hd < warm and hot dust emission strengths are comparedthrough the 2.3 µ m versus 10 µ m luminosities in Figure6b, which shows a linear relation between the MIR–NIRluminosities. Interestingly, the detected hot dust poordata points lie moderately below the relation by 1.8 σ on average, suggesting perhaps a shifted MIR–NIR lu-minosity relation for hot dust poor SEDs. Nonetheless,we may find hot dust poor quasars to stay on the warmdust poor side, as their 10 µ m luminosities are lowerby 1.9 σ on average to the linear relation, log( L ) =(1 . ± . L . ) − (6 . ± . , rms = 0 . µ m detections.Therefore we find it reasonable from our NIR/MIR lu-minosity correlation of hot dust poor quasars, althoughwith a possible offset, to deduce that hot dust emission isa marginally good tracer of warm dust emission in AGNs.The IR portion of the quasar SED to peak in the MIR(e.g., Richards et al. 2006), and the acceptable correla-tion between MIR/far-IR (hereafter FIR) luminosities of Jun and Im Fig. 8.—
Left: Composite spectra of dust poor quasars (purple) and of general SDSS quasars (black, from Vanden Berk et al. 2001),smoothed by 3 pixels to display the spectral features better. The number of spectra used to construct the composite from the top to bottompanels are 163, 125, 22 for dust poor quasars, and around 500, 1800, 700 from Vanden Berk et al. (2001). Right: Spectral fits around theH β region for dust poor (top) and ordinary (bottom) composites. The measured HeII λ quasars (e.g., Haas et al. 2003), further suggest hot dustpoor sources to stay relatively dust poor through the en-tire IR. For these reasons, we choose to refer to “hot dustpoor” quasars as “dust poor” quasars from now on.To investigate how dust poor quasars differ from ordi-nary quasars in wavelengths other than the NIR/MIR,we plot f . as a function of optical continuum slope α , in Figure 7. The average slope and its scat-ter of <α> = − . ± − .
44 of Vanden Berk et al.(2001). This is consistent with the results of Davis et al.(2007) where more luminous quasars have bluer contin-uum slopes, as our sample are high luminosity selectedSDSS quasars. We also note that differences in the fittingrange or the method to measure α , could further shift thevalues of α (e.g., Table 5 in Vanden Berk et al. 2001). Inany case <α> = 0.10 ± α = of optically thick and locallyheated accretion disk model predictions, and of polar-ized observations of quasar SEDs seen through the dust(Kishimoto et al. 2008). This indicates that the opticalcontinuum from the accretion disk of dust poor quasarsare mostly unobscured to our line of sight, consistentwith the weak IR re-emission observed. Meanwhile, thereare points forming a diagonal line across the lower leftpart of α – f . space in Figure 7, which are optical toNIR SEDs best fit by the continuum component only.Although these objects do not seem to involve the hotdust component for the SED fitting, satisfying c hd = 0 inEquation 2, the NIR emission could be taken over by theextrapolated optical continuum component, especially atthe red end of small α . Therefore we regard at least partof the diagonal sequence in Figure 7, to effectively meanlower limits in the measured α .In addition to the optical continuum slope measure-ments from photometric data points, we constructed acomposite UV–optical spectrum of dust poor quasarsplotted in Figure 8a, along with the composite of gen-eral SDSS quasars in Vanden Berk et al. (2001). Com-pared to typical quasars, dust poor quasars have abluer NUV–optical continuum slope. Moreover, fit- Fig. 9.—
Distribution of HeII λ β broad line fluxes, fol-lowing the color and symbol layouts of Figure 6. The H β andHeII λ z < >
10 spectra 1899 have non-detections in H β or HeII,or fail to meet H β line S/N >
2, where they are excluded from theplot. The average and standard deviation of HeII/H β on the plot,are printed for both the entire and dust poor samples. ting all prominent UV to optical emission lines we findthe HeII λ β and [OIII] lines to better decompose the HeII line emis-sion.Because the HeII λ β line ratios for all z < >
10 and H β line S/N >
2. Theust poor quasars 9
Fig. 10.—
Redshift–bolometric luminosity (left) and black hole mass–Eddington ratio distribution (right) of our sample. The rangeof data points from M11 are highlighted (black dashed lines) for comparison. Furthermore, the 361
XMM -COSMOS spectroscopicallyconfirmed type-1 AGNs in Lusso et al. (2010) that are similar to the sample of H10, are overplotted (black circles). Since the entire sampleis large in number, a quarter of the sample are randomly selected and plotted (black dots) in both panels, to better resolve and comparethe overall distribution to that of dust poor objects (color circles). We subdivide our sample into four bolometric luminosity bins (colordot dashed lines), for the forthcoming analysis to be independent of luminosity selection effects. results are plotted in Figure 9, where HeII and H β linefluxes of dust poor quasars are compared with those ofordinary sources. Dust poor quasars occupy regions withabout 1.8 times higher HeII/H β line ratios compared totypical quasars, being on average 13% and 7% respec-tively, consistent with the results from Figure 8b. Thiscan be understood as a result of the strong UV contin-uum source, as the HeII/H β line ratio and the sourcespectral shape in the extreme-UV (EUV) form a relation(Penston & Fosbury 1978) as L HeII λ L H β = 1 . × α EUV , (4)where α EUV indicates the EUV slope of the source thatbrings HI and HeII emission. Equation 4 yields a bluerslope of α EUV ∼ − . Parameter space study of dust poor quasars
Figure 10 shows the distribution of AGNs in our sam-ple in the L bol versus redshift, and f Edd versus M BH .Also plotted are the range of the parameter space cov-ered by other studies (H10 and M11). Here the dashedboundary for the M11 sample is determined followingtheir selection criteria, and we mimic the distribution ofthe H10 sample by plotting with thick black points theobjects in Lusso et al. (2010) sample which contain 88%of the 408 AGNs in H10. Overall, our sample are brighterthan of H10 in L bol , and less extensive in L bol but moreextensive in redshift space coverage than of M11. FromFigure 10, we find that the L bol and i -band limits ( § L bol and the hot dust covering factor CF hd (e.g.,M11), may act as a luminosity selection effect ( § f . or CF hd when plottedagainst a parameter closely associated with L bol , such as M BH or f Edd . To minimize the luminosity selection ef-fect, we split the whole sample into four volume-limitedsubsamples where 204 dust poor quasars remain insidethe dot dashed limits in Figure 10a.Now, we use the fraction of dust poor quasars (here-after, p hdp ), to visualize the global trends of dust poorquasars in observed parameters. When the p hdp are plot-ted against L bol in Figure 11a, we find a positive cor-relation between the two products, in accord with thedistribution of f . decreasing with L bol in Figure 11c.The first parameter to look into through p hdp under thevolume-limits is the redshift, as we would like to know theevolution of the fraction of dust poor quasars indepen-dent of luminosity selection. In Figures 11b and 11d wefind not only a trend of lower f . at higher redshift over-all (bottom panel), but also a clear tendency of increased p hdp at higher redshifts at given luminosity (top panel),mixed with the p hdp increasing with luminosity at a givenredshift. Compiling our p hdp with that inferred from J10,we fit the z > L bol > erg s − data points in Figure11b to model the redshift evolution of the dust poor frac-tion as p hdp = (1 . ± . × − (1 + z ) . ± . .Since we are dealing with a small number of specialobjects, some of dust poor quasars could be misclassi-fied normal quasars due to large photometric errors, orsparse wavelength sampling in the WISE data points. Toestimate what fraction of dust poor quasars could havearisen from these artifacts, we selected 7834 SDSS DR7quasars in our sample at z < K and W i =18–19.5 mag. Start-ing from a set of observed fluxes and errors linearly inter-polated to simulated redshifts ( F obs , ∆ F obs ), we assignedthe flux errors corresponding to the simulated fluxesas ∆ F sim = ∆ F obs . m sim − m obs ) where m obs , m sim areobserved and simulated magnitudes. Then, we addeda Gaussian random noise ( σ gauss = ∆ F sim ) to the sim-0 Jun and Im Fig. 11.—
Top left: L bol –hot dust poor fraction ( p hdp ) plot. Color filled symbols represent the fraction of quasars in each luminosity bin,to be hot dust poor. The error bars are calculated based on the Poisson number counting statistics. Bottom left: L bol – f . plot. Individualdistribution of hot dust poor quasars (purple) with randomly selected one quarter (following Figure 10) of the entire sample (black), areplotted. In addition, the average f . along L bol are overplotted in thick solid lines. Top right: z – p hdp plot. Color filled symbols slightlyshifted to each other to avoid overlaps, represent the fraction of quasars in each luminosity and redshift bin to be hot dust poor, witheach number of objects per data point printed following the arrangement of the data. Meanwhile, the open circle is the z = 6 result of J10.Upper limits in p hdp are calculated as 1 /N bin , when there are no dust poor quasars in the bin with size N bin . The gray gradational circlesare test results of our z < K and W i -band magnitude. Bottom right: z – f . plot. The layout follows that of the bottom left panel. ulated fluxes, to obtain a set of fluxes and errors( F sim , ∆ F sim ). Every object was repeatedly simulated50 times with randomly added errors, for a given sim-ulated redshift and magnitude. Finally the redshifted,error scaled, and noise added set of SEDs were fitted tocalculate the dust poor fraction. The dust poor fractionsfrom this simulation are overplotted in Figure 11b withgray circles.The figure indicates that the objects selected for thistest are representative of general quasars at z <
1, sincethe simulated p hdp regardless of the test magnitude, arewithin errors to the observed p hdp at z <
1. But then, thetest gives the possibility of dust rich ( f . > − .
5) quasarsto be artificially classified as dust poor, as the simulated p hdp starts to increase at z >
1. Not only does p hdp in-crease for fainter simulated magnitudes, it also has a peakat 2 < z < µ m lies in betweenthe widely separated W /W p hdp of L bol = 10 . − erg s − sources in Fig-ure 11b for example (yellow and gray points), we estimate ∼
60% of the faintest objects ( i = 18.5–19, see Figure 2b)in the dust poor sample to be possibly misclassified nor-mal or border-line quasars at 2 < z <
3. However, at z > p hdp aregenuinely higher at z >
3. Furthermore, taking into ac-count the higher possibility of identifying false dust poorobjects at 2 < z <
3, the intrinsic distribution of true dustpoor quasars will show a stronger positive z – p hdp rela-tion at z >
2, for any given luminosity bin. Therefore weregard the overall trend of increased p hdp with redshiftto remain unchanged under possible artifacts.Next, p hdp versus black hole mass and Eddington ra-tio are plotted in Figures 12a and 12b for four different L bol bins. We find higher p hdp for more luminous L bol binned subsamples, which is consistent with Figures 11aand 11b, and with predictions of smaller dust coveringfactors at brighter L bol , such as from the receding dusttorus model (Lawrence 1991). Therefore, we note the im-portance of controlling the range of L bol in order to ac-curately trace the properties of dust poor quasars. In ad-dition, because our dust poor quasars are systematicallysmaller in CF hd , or equivalently larger in NIR bolomet-ust poor quasars 11 Fig. 12.—
Left: M BH – p hdp plot (top), and the distribution of M BH – f . (bottom). Right: f Edd – p hdp plot (top), and the distribution of f Edd – f . (bottom). The meaning of the data point symbols follows that of Figure 11. ric corrections (hereafter BC, BC λ =BC λ ( µm ) ) than aver-age, we would like to double check whether the L bol – p hdp trend is affected by a systematically different BC for dustpoor SEDs. Utilizing Figure 4 we integrated the model fiton the composite photometric SED of all quasars in the0.04–20 µ m, to find L . − /L . = 6 .
46, or 70% of thebolometric luminosity to be bounded within the selectedUV through MIR wavelengths, adopting BC . = 9.26from S11. Now, assuming that the SED of the aver-age and dust poor composites do not systematically varywith each other in the γ -ray–X-ray or FIR–radio (see § L < . + L > ) /L . ] hdp =2 .
80 = ( L < . + L > ) /L . . From our measurement of( L . − /L . ) hdp = 5.04 out of the composite dust poorSED, we get BC . ,hdp = 7.84, meaning that bolometricluminosities of dust poor quasars are about 18% over-estimated with respect to ordinary quasars, under theadoption of a single BC . = 9.26.Drawing Figures 11–12 again with a 18% smallerBC . for dust poor quasars, however, does not changethe strength of p hdp trends found in ( z , L bol , M BH , f Edd )spaces. Therefore we conclude that possible system-atic biases in the bolometric correction for dust poorquasars does not artificially produce the L bol – p hdp re-lation nor change the observed p hdp trends within themain observed parameters, although our test brings cau-tion when applying the BC to dust poor quasars fromthe monochromatic luminosity alone. After all, reading Figures 12a and 12b within each luminosity bin decou-pled from the L bol dependence on p hdp , we find moredust poor quasars at lower black hole masses and highEddington ratios.Finally, we plot p hdp against other AGN observablesin Figure 13, starting from the broad line width and ve-locity offset. We used the FWHM from S11 as broadline widths, updated and replaced for some dust poorquasars in §
3. Meanwhile, we took the broad line veloc-ity offset measurements from S11, defined as the relativeshift of the broad line center with respect to the sys-temic redshift. The choice of broad line for the line widthand offset measurements at each redshift, was identicalas to measure M BH ( § p hdp and broad line FWHM in Figure 13a, consistent withthe trend found for M BH , and a meaningful tendencyfor velocity offset towards blueshifted broad emission,in Figure 13b. The implication of these results willbe discussed in the next section. Next, reading Figure13c, dust poor quasars with respect to the FeII(2200–3090 ˚A)/MgII λ p hdp , with R radio = F ν, /F ν, from S11. The ra-dio data are from the FIRST survey (White et al. 1997)roughly coinciding with the SDSS coverage and reach-ing a typical 5 σ sensitivity of 1 mJy. This is similar to2 Jun and Im Fig. 13.— p hdp in parameters related to broad emission lines and radio properties. In panels a–d we plot p hdp versus broad line FWHM,broad line blueshift from the systemic redshift, FeII(2200–3090 ˚A)/MgII λ R radio = F ν, cm /F ν, .The meaning of the data point symbols follows that of Figure 11 except for the left pointing arrows in panel d, where they are p hdp estimated without radio detections, but with a rough R radio .
10 limit. the WISE W F ν from the MIR to radio, implying that mostradio loud/intermediate sources are able to be radio de-tected. However, the FIRST detection threshold (5 σ )is higher than that of WISE (2 σ ), and misses the ma-jority of radio quiet sources, which results in only 10%of the final AGN sample used in this work to be de-tected. But still, the radio loudness is roughly completedown to R radio ∼
10 within the detection limit, while thedistribution of R radio shows a trend expected to fall be-low R radio ∼
10 under nondetections, for both dust poorand entire AGNs. To further investigate the complete-ness of the radio data, we calculated the p hdp in Fig-ure 13d while adding the FIRST undetections into the R radio <
10 bin, to find the p hdp to be consistent withinerrors, to that with the detections only. After all, wefind no significant preference of dust poor quasars in the R radio – p hdp distribution. Likewise, we tried to searchfor the X-ray loudness features from the RASS observedcounts in Schneider et al. (2010), but failed to compileenough number of matches (N=7) for proper analysis. DISCUSSION
Observational characteristics of dust poor quasars
To quantify how distinct dust poor quasars are fromtypical quasars in various physical properties, we per-formed a set of Kolmogorov–Smirnov (K–S) tests by com- paring the distribution of dust poor against average sam-ples. The result is summarized in Table 3. Although itbecomes uncertain at L bol < . erg s − , most of theK–S probabilities at L bol > . erg s − are smaller than1%, indicating that dust poor quasars are likely to bedrawn from a different population to average luminousquasars. Moreover the directions of the inequality be-tween dust poor and average quasar properties in Table3, are consistent with Figures 11–13. To begin summa-rizing the main results from the table and figures, dustpoor quasars are at higher redshifts for a given luminos-ity, with this trend found to be secure at z >
2. This isconsistent with J10 and H10, but not with M11 wherethey do not find an evolution in the dust poor fraction.Next, we find dust poor quasars of relatively lower blackhole masses in the 10 M ⊙ order and high Eddington ra-tios of super-Eddington accretion, at a given luminosity,in agreement with J10 at z > M BH and f Edd . In addition to these fundamentalproperties, dust poor quasars show narrower broad lineFWHM of 2000–3000 km s − , blueshifted line centers inthe 10 km s − order, and higher FeII/MgII line ratiosby a few times of factors than average.Although we find some of the parameter trends sum-marized above to agree with previous studies, there arealso debatable results. To investigate further in betweenust poor quasars 13 TABLE 3K–S test on dust poor quasar fraction in observed parameters log L bol (erg s − ) 45.7–46.5 46.5–47 47–47.5 > > p KS p KS p KS p KS N hdp N hdp N hdp N hdp z -0.09 3.1e-06 1.2e-03 0.13 18 60 93 33 M BH -0.64 -1.1e-04 -7.7e-12 -3.4e-03 18 60 90 25 f Edd v blueshift R radio Note . — p KS is the K–S probability for the distribution of dust poor quasars in each parameter and luminosity bin, to beindistinguishable from that of the rest of our whole sample. The configuration of volume-limited luminosity bins follow Figure 10a.Positive p KS implies dust poor quasars to have a larger parameter value than average quasars, and vice versa for negative signs,while statistically meaningful probabilities set as | p KS | < N hdp is the number of dust poorquasars used in each bin to calculate p KS , ordered by the same luminosity range as for p KS . Blank fields correspond to bins withfew quasars to perform a test. conflicting arguments, we would first like to note thelimitation of small number statistics. The number ofour volume-limited dust poor quasars is N=204, which isenough to precisely look for parameter trends as we findmany of the K–S probabilities in Figure 3 to be statisti-cally significant. Also, our study fills the lower redshiftcounterpart to J10 as they miss the z < z – p hdp trend in this work thatpredicts lesser dust poor quasars at lower– z . Meanwhile,although H10 selected dust poor quasars with a relativelyhigh fraction ( & f . < − . p hdp in H10 as from the dif-ferent definition of being dust poor rather than from theX-ray selection of AGNs to elevate the p hdp .Next, we note the difference of studying CF hd in over-all distributions from previous works, and under volume-limited subsampling here. For instance, the overall dis-tribution of CF hd is roughly independent of M BH in J10( z < p hdp for less massive M BH ata given luminosity of L bol > . erg s − . If the samplehad a wide dynamical range of L bol , there is a higherchance of the anti-correlation between L bol and CF hd tomix up the distribution of CF hd when plotted against aparameter dependent on L bol . In other words, quasarswith higher M BH are more likely to be selected as dustpoor objects since M BH ∝ √ L bol (e.g., Equation 3 withconstant bolometric correction), and because there aremore dust poor quasars with high L bol (Figure 11a). Re-visiting Figures 12a and 12c, this luminosity selectioneffect is indeed visible where the bottom panel undera wide range of L bol displays flatter f . with M BH , ormore chance of dust poor quasars at higher M BH thanfrom the upper panel under a fixed range of L bol . There- fore we expect the disconnection between f . or CF hd versus M BH in the z < L bol that would smooth out the possible underlyingcorrelation between M BH and f . /CF hd .Likewise, the luminosity selection would act on theEddington ratio as f Edd ∝ L bol /M BH ∝ √ L bol , select-ing more dust poor quasars at higher f Edd than whenvolume limited. This time the positive f Edd – p hdp re-lation of Figure 12b is consistent with the f Edd –CF hd anti-correlation in Figure 12d, which can be explainedby the stronger f Edd –CF hd anti-correlation than thatof L bol –CF hd (e.g,. Figures 12d versus 11c). Further-more, we expect the luminosity selection effect to besomewhat mixed by the redshift selection effect, dueto the z – p hdp trend and the luminosity distribution ofour sample such that higher luminosity sources are athigher redshift on average. We tested the redshift se-lection effect by drawing Figures similar to 12a and 12bat a fixed L bol = 10 − . erg s − interval, while compar-ing the p hdp between 1 < z < < z <
3. We finda higher normalization of p hdp at higher redshift, butthe difference in the p hdp is smaller than when the red-shift interval is fixed as one the two above, and insteadadjacent subsamples in L bol to 10 − . erg s − are com-pared. Therefore, the luminosity selection effect on p hdp in this work, is only mildly mixed with that from redshiftselection.As a way to remove the luminosity selection effect, M11defined their hot dust poor quasars as the lower outlyingobjects in CF hd , dependent on L bol . Therefore, the p hdp based parameter trends of dust poor quasars in M11 canbe directly compared to the volume limited p hdp basedtrends in this work, as both studies are corrected for theluminosity selection, while the sample size of M11 is aslarge as ours such that they do not suffer from smallnumber statistics. The L bol – p hdp trend would be differ-ent for the two studies though, as the definition of thedust-poorness in M11 inherently leads to a flat p hdp – L bol relation, while our result of the increasing p hdp with L bol is free from such a selection. Apart from this exception,we expect the p hdp related parameter trends from M11and this work to be consistent with each other, but westill witness somewhat conflicting results in z , M BH , and f Edd spaces. However, the redshift range of 0.75 < z < p hdp above z = 2. Next, M11 did not find the p hdp to be depend-ing on M BH or f Edd while we find negative and positivedependences on M BH and f Edd , respectively. A caveatis that the dust poor selection of M11 passes through &
16% of the sample, which is much higher than 0.6%from this work, or ∼
1% from Figure 11b within the red-shift interval 0.75 < z < hd sources, may be choosing too many objects to be dustpoor, making it hard to find distinctive properties of dustpoor quasars from the rest of the sample. To summarize,we conclude that the differences in p hdp trends betweenM11 and this work, originate from the different redshiftcoverage and the selection criteria of dust poor quasars.We note that the observational characteristics of dustpoor quasars could be different in lower AGN luminosi-ties, as our K–S probabilities in Table 3 are weaker at L bol < . erg s − . Combining our results with thestudy of Mor & Netzer (2012, hereafter M12) helpsto constrain the properties of fainter dust poor AGNs,since their sample are Seyfert 1 galaxies mostly covering L bol = 10 − erg s − . Although their dust poor/totalsample sizes are only ∼
10 and 115, they do find lowCF hd sources for narrow-line Seyfert 1s in low M BH andhigh f Edd , consistent with our results at > . erg s − .Since the optical luminosities of AGNs in M12 are likelyto be affected by host galaxy contamination (e.g., S11), itwould be interesting for future studies to check whetherthe parameter trends of intermediate luminosity ( L bol . erg s − ) dust poor AGNs as in M12, are valid af-ter correcting for the host contamination in the opticalluminosity.Physically connecting our observational results of lu-minous dust poor quasars, the L bol – p hdp relation is wellexplained by the receding torus model (Lawrence 1991),such that more luminous quasars have smaller dust cov-ering factors due to the torus being located far fromthe central light source. In addition, the link betweenFWHM– p hdp in Figure 13a is closely connected with the M BH – p hdp and f Edd – p hdp relations in Figures 12a and12b, since quasars with narrow FWHM would be lowerin M BH and higher in f Edd at a given L bol . This impliesthat quasars not only luminous but violent in growth andsmall in cumulative mass, or actively growing quasars,would have a higher chance of being dust poor. Sinceour dust poor quasars, although skewed to higher– z , arefairly spread within the z – p hdp relation, we refer to thedust poor epoch of quasars as positioned in a buildupstate within individual black hole growth histories, ratherthan a specific global cosmic epoch (J10).In addition to the key trends, the v offset – p hdp relation-ship for dust poor quasars preferring blueshifted broademission, could be related to radiative outflows affectingthe broad line emission (e.g., Richards et al. 2011), whichare necessary for radiative quasar feedback or QSO dustproduction through the expansion of broad line clouds( § / MgII ofdust poor quasars could be signs of metal seeds where theQSO dust can grow ( § The origin of dust poor quasars
We now consider possible scenarios for the origin ofdust poor quasars. First, we would like to list the expla-nations that the occurrence of dust poor objects are de-pendent on the geometry of the surrounding dusty struc-ture. One of the descriptions is the receding torus model(Lawrence 1991) which predicts smaller dust coveringfactors at higher luminosities, for the obscuring structureto be located more outward. Our study provides furtherconstraints to the model, as our dust poor quasars arenot only more luminous, but also less massive and highin Eddington ratio. To fit our observational results theviewing angle of dust poor quasars could be consideredas being observed from a face-on direction, because theunified model predicts obscuration from the dusty toruswhen observed through an inclined angle. Assuming thatthe broad line region follows the inclination of the torus(Gaskell et al. 2007), and a simplified planar geometryfor these AGN substructures, the observed line widthswould meet FWHM obs = FWHM sin i , where i is the in-clination angle. Therefore, we may expect the broadFWHMs of dust poor quasars to be systematically nar-rower merely due to the projected sin i factor for a planargeometry.While this orientation based approach is capable ofexplaining many parameter space features (FWHM, M BH , f Edd ) of dust poor quasars, the problem is thatit cannot describe the redshift dependence of p hdp , un-less higher redshift quasars are systematically biasedto lower inclination angle objects at given luminosity.Moreover, when the dusty torus is observed face-on,the unified model suggests more chances of being di-rectly showered by radio jets (e.g., Figure 3 in Antonucci1993), whereas in Table 3 we do not find dust poorquasars to be radio louder, although number statisticsmay not be secure. Several other geometric explana-tions to the origin of dust poor quasars include a lowlevel misalignment of the torus with respect to the accre-tion disk (Kawaguchi & Mori 2011), or mildly misaligneddisks even without the torus structure (Lawrence & Elvis2010). At this time where our observations are not ableto strictly validate each geometric scenario, we leave itas an open question to explain dust poor quasars in ageometric sense, though we do stress the need for futuremodels to be able to further explain our observationalfeatures ( § L bol > . erg s − ) more than 50% of quasarsare triggered by major mergers. Merger triggered quasaractivity are often observationally interpreted to initiatefrom the stage of luminous infrared galaxies, showingstrong signatures of dense gas and dust up to kpc scalesust poor quasars 15(e.g., Sanders et al. 1988; Surace et al. 1998). Therefore,for the majority of luminous dust poor quasars to beoriginated from galaxy merging, not only the host galax-ies are expected to have been dusty during the mergingstage, but also the obscured host galaxies require feed-back mechanisms that blow out the surrounding mate-rial to become more transparent to the quasar radiationalike normal quasar systems. This step is predicted inthe merger driven AGN model of Hopkins et al. (2008)as the blowout stage, where the gas and dust surroundingthe quasar host galaxy are expelled.Depending on the extent of merger driven quasar feed-back, red quasars are thought to be in the blowout stagewhere the surrounding dust have not yet been removed(e.g., Figure 5e in Hopkins et al. 2008; Urrutia et al.2008), while we may now place dust poor quasars atthe end of the blowout stage for the dust to have suffi-ciently been dispersed. Our study further helps to obser-vationally constrain the evolutionary boundary of dustpoor quasars as we find it to agree with the blowoutphase predictions of rapidly growing BHs (low M BH , high f Edd ) and intense feedback (high L bol and blue UV con-tinuum) from Hopkins et al. (2008), and with observa-tions of high f Edd objects within the red quasar sampleof Urrutia et al. (2012). At the same time however, dustpoor quasars are closer to normal quasar phases as theiraccretion disk dominated SEDs indicate the nucleus tobe less obscured from its surrounding, compared to redquasars. Therefore, we consider it the most plausibleto assign dust poor quasars in between the blowout andtraditional quasar phases, where dust poor quasars canbe explained to have just become unobscured as theywent through intense feedback during the rapid blackhole growth, but are still relatively low in M BH andhigh in f Edd on their way of becoming normal quasars.When our dust poor fraction of 0.6 % is translated intothe visible dust poor timescale, it becomes ∼ p hdp fromthis work is further explained. At high redshift theamount of dust observed in quasars or GRBs can be ex-plained by dust producing sources mainly from super-novae and the most massive AGB stars, whereas QSOdust or contributions from less massive AGB stars arerelatively limited (e.g., Pipino et al. 2011, Jang et al.2011). The restricted dust production routes within theshort age of high redshift quasar systems therefore, couldbe the cause of the z – p hdp relation of dust poor quasars astheir progenitors may not have enriched enough dust tosurvive under the presence of AGN feedback, observed tobe in action up to the early universe (e.g., Maiolino et al.2012). Hence, the increased fraction of dust poor quasars We use the term QSO dust for the dust production to originatefrom the quasar activity, and to distinguish from stellar relatedsources. at higher redshift is consistent with the current dust for-mation model predictions, and with observations of bluerUV continuum slopes of inactive galaxies at high redshift(e.g., Bouwens et al. 2009).Stemming from the evolutionary model for dust poorquasars, the fact that we find BAL quasars in our dustpoor sample has interesting implications about the dustorigin of BAL quasars. Utilizing the BAL flag in S11based on the criterion from Gibson et al. (2009), we findthe BAL fraction of 4.3 ± ± The future of dust poor quasars
Having considered dust poor quasars to possibly be ex-plained by geometric or evolutionary models, to be ob-served by an unobscured orientation or to have under-gone a duration of strong feedback, we would finally liketo comment on their close future by comparing with av-erage optically selected quasars. If dust poor quasars areoriginated from a geometric reason, we may expect thecovering factor to become larger at later quasar phaseswhen they become quieter in luminosity or Eddington ra-tio, as the covering factors are anti-correlated with L bol or f Edd (Lawrence 1991; Kawaguchi & Mori 2011).On the other hand, assuming an evolutionary originof dust poor quasars, we need to consider dust forma-tion mechanisms during the AGN evolution. Althoughour observational explanation of dust poor quasars to berapidly growing ( § L bol < × erg s − quasars though, since the radia-tive flux density shining the QSO dust forming region isweaker than that around giant stars (Elvis et al. 2002).Considering that 75% of our dust poor quasars are lessluminous than this limit, the extreme case of continueddust destruction can mostly be rejected as the fate ofdust poor quasars. Therefore, provided that the dustproduction is effectively working, we suggest the nearfuture of dust poor quasars to be richer in dust includ-ing the contributions from QSO dust, which is againstobservation based scenarios where ordinary quasars be-come dust poor objects (Haas et al. 2003; Hao et al. 2012though for a fainter luminosity range), but supports dustenrichment within the torus (Elvis et al. 2002; J10) orcircumnuclear regions (e.g., Sim˜oes Lopes et al. 2007) ofactive galaxies during its activity.But what if dust poor quasars are not causally relatedto ordinary quasars? It could be the case where dustpoor quasars are indeed rare objects out of AGN evo-lution, to be placed in intrinsically dust poor environ-ment within their host galaxies and/or the QSO dustformation is weak, such that they do not become asdust rich as ordinary quasars during the quasar phase.This idea may be supported by observations of early-type AGN host galaxies, where the ratio of dust massover 4.5 µ m luminosity representing the stellar mass (e.g.,Jun & Im 2008), is widely ranged at given 4.5 µ m lumi-nosity (Martini et al. 2013). Within this framework, therelative lifetimes of each AGN evolutionary stage basedon obscuration, could be scattered due to the variety ofdust content within each host environment. For exam-ple, although the AGN evolutionary models predict ob-scured AGNs to be younger than the unobscured, underan especially low dust-to-stellar mass ratio environmentof the host, AGNs would quickly shine optically luminouswithout a typically long duration being obscured (e.g.,Hopkins et al. 2005). Thus, even if dust poor quasarsmay be explained to lie in a rare dust poor environment,it shifts the visible dust poor lifetime to the relativelyearlier growing state of the black hole, still consistentwith our rapid growth explanation ( § CONCLUSION
Out of a large area, multi-wavelength sample of lu-minous quasars, we identified 233 (0.6%) objects thatsatisfy the hot dust poor criterion analogous to that ofJiang et al. (2010). The selected dust poor quasars areweak in both NIR/MIR-to-optical flux ratios, and show ablue continuum with the mean slope of α ∼ M BH , f Edd , whilethe rapid growth of black holes and the dust poornessshould be linked closely with the evolution of quasars as a function of redshift.To explain the observational characteristics of dustpoor quasars, we suggest a scenario in which dust poorquasars are transient phenomena during an evolution-ary process, when they became unobscured by mergerdriven AGN feedback. Luminous quasars are often trig-gered by violent major merging (Treister et al. 2012), in-volving large extinctions from their host galaxies (e.g.,Sanders et al. 1988). From the onset of strong AGNactivity, radiative feedback blows out (Hopkins et al.2008) the surrounding galactic dust to become dust poorquasars, for a very short period if they produce QSOdust (Elvis et al. 2002; Pipino et al. 2011), or for longerif they are in an especially dust poor environment. Theredshift evolution of the dust poor fraction is an indi-cation of the dust content at different redshift, wherethe dust poor phase is more easily identified and lastslonger in higher redshift, since the elements that makeup dust grains are scarce early in the universe. An al-ternative scenario would be to explain dust poor quasarsto be a distinct population with small covering factors,due to the geometry/orientation of the obscuring ma-terial. Such models (Lawrence 1991; Lawrence & Elvis2010; Kawaguchi & Mori 2011) are consistent with mostof the observed properties, but not with the redshift evo-lution of the dust poor fraction.We find the short timescale from the rare populationof dust poor quasars, to provide meaningful evolutionarypredictions. Deep and resolved imaging of the AGN hostwill tell us if these dust poor quasars are closely linked tothe evolutionary phase, as the merging features would beclearer at earlier times than the rapid fading of tidal fea-tures (Hopkins et al. 2008). Besides, extrapolating therough relation p hdp ∝ (1 + z ) . ( § p hdp = 18% and 37% for luminousquasars at z = 7 and 10. Therefore, discovery of the high-est redshift quasars will help us know whether the higherincidence of dust poor quasars, blue in UV continuum,boost the quasar contribution in reionizing the universe(e.g., Fan et al. 2006), or if they are born in extremelydust poor galaxies as hinted by the UV slopes of lowluminosity galaxies at z ∼7 (e.g., Bouwens et al. 2010).Future deep multi-wavelength studies of dust poor AGNsshould place better constraints on the multi-temperaturedust emission of the AGN-host system and the redshiftevolution of these objects.This work was supported by the National ResearchFoundation of Korea (NRF) grant, No. 2008-0060544,funded by the Korea government (MSIP). We thank thereferee’s helpful comments to improve the paper, andthe CEOU team members for their contribution in tak-ing the UKIRT data. Based on observations made withthe NASA Galaxy Evolution Explorer. GALEX is op-erated for NASA by the California Institute of Tech-nology under NASA contract NAS5-98034. This pub-lication makes use of data products from the Two Mi-cron All Sky Survey, which is a joint project of theUniversity of Massachusetts and the Infrared Process-ing and Analysis Center/California Institute of Technol-ogy, funded by the National Aeronautics and Space Ad-ministration and the National Science Foundation. Thispublication makes use of data products from the Unitedust poor quasars 17Kingdom Infrared Deep Sky Survey, and the data takenwith the UKIRT. The United Kingdom Infrared Tele-scope is operated by the Joint Astronomy Centre on be-half of the Science and Technology Facilities Council ofthe U.K. The UKIDSS photometric (Hewett et al. 2006)data are from WFCAM imaging (Casali et al. 2007),where the data are processed (Irwin et al. 2009, in prep), calibrated (Hodgkin et al. 2009), and science archived(Hambly et al. 2008). This publication makes use ofdata products from the Wide-field Infrared Survey Ex-plorer, which is a joint project of the University of Cal-ifornia, Los Angeles, and the Jet Propulsion Labora-tory/California Institute of Technology, funded by theNational Aeronautics and Space Administration.