What makes red quasars red? Observational evidence for dust extinction from line ratio analysis

Red quasars are very red in the optical through near-infrared (NIR) wavelengths, which is possibly due to dust extinction in their host galaxies as expected in a scenario in which red quasars are an intermediate population between merger-driven star-forming galaxies and unobscured type 1 quasars. However, alternative mechanisms also exist to explain their red colors: (i) an intrinsically red continuum; (ii) an unusual high covering factor of the hot dust component, that is, $\rm CF_{HD} = {\it L}_{HD} / {\it L}_{bol}$, where the ${L}_{\rm HD}$ is the luminosity from the hot dust component and the ${L}_{\rm bol}$ is the bolometric luminosity; and (iii) a moderate viewing angle. In order to investigate why red quasars are red, we studied optical and NIR spectra of 20 red quasars at $z\sim$0.3 and 0.7, where the usage of the NIR spectra allowed us to look into red quasar properties in ways that are little affected by dust extinction. The Paschen to Balmer line ratios were derived for 13 red quasars and the values were found to be $\sim$10 times higher than unobscured type 1 quasars, suggesting a heavy dust extinction with $A_V>2.5$ mag. Furthermore, the Paschen to Balmer line ratios of red quasars are difficult to explain with plausible physical conditions without adopting the concept of the dust extinction. The $\rm CF_{HD}$ of red quasars are similar to, or marginally higher than, those of unobscured type 1 quasars. The Eddington ratios, computed for 19 out of 20 red quasars, are higher than those of unobscured type 1 quasars (by factors of $3 \sim 5$), and hence the moderate viewing angle scenario is disfavored. Consequently, these results strongly suggest the dust extinction that is connected to an enhanced nuclear activity as the origin of the red color of red quasars, which is consistent with the merger-driven quasar evolution scenario.


Introduction
Large area surveys in X-ray, ultraviolet (UV), optical, and radio wavelengths have uncovered nearly a half million quasars to date (Grazian et al. 2000;Becker et al. 2001;Anderson et al. 2003;Croom et al. 2004;Risaliti & Elvis 2005;Schneider et al. 2005;Véron-Cetty & Véron 2006;Im et al. 2007;Lee et al. 2008;Young et al. 2009;Pâris et al. 2014;Kim et al. 2015c). In the UV to optical wavelength range, the spectra of quasars show a blue power-law continuum with broad and narrow emission lines or a host galaxy continuum with only narrow emission lines, which are classified as type 1 and 2 quasars, respectively. The unification model (Urry & Padovani 1995) has been proposed to explain different types of quasars. In the model, a quasar is composed of a black hole (BH), accretion disk, dust torus, broad line regions (BLRs), and narrow line regions (NLRs). Under the unification model, the type 1 and 2 quasars are physically the same, but an obscuring dust torus prevents us from seeing the accretion disk and the BLR in a certain line-of-sight direction for type 2 quasars.
Interestingly, red quasars are expected in simulations where merger-driven galaxy evolution is stressed (Menci et al. 2004;Hopkins et al. 2005Hopkins et al. , 2006Hopkins et al. , 2008. In such simulations, major mergers of galaxies trigger both star formation and quasar activity, often appearing as ultra-luminous infrared galaxies (ULIRGs; Sanders et al. 1988;Sanders & Mirabel 1996). After that, the central BH grows rapidly but it is still obscured by the remaining dust in their host galaxies. Finally, the obscured quasar evolves to an unobscured quasar after the feedback from the central BH sweeps away cold gas and dust. In this picture, red quasars should appear during the intermediate stage between ULIRGs and unobscured quasars. These quasars are red owing to dust extinction by the remaining dust and gas in their host galaxies. So far, observational signatures of red quasars tend to support this picture. For example, red quasars have (i) higher BH accretion rates than unobscured type 1 quasars by factors of 4 to 5 Kim et al. 2015b), (ii) enhanced star formation activities (Georgakakis et al. 2009), (iii) a high fraction of merging features (Urrutia et al. 2008;Glikman et al. 2015), and (iv) young radio jets (Georgakakis et al. 2012).
However, several studies have proposed varying explanations for the red continuum of red quasars. Wilkes et al. (2002) and Rose et al. (2013) suggested that the red J − K colors (J − K > 2) of red quasars come from a moderate viewing an-A&A proofs: manuscript no. redcolor gle in the unification model when the photons from the accretion disk and the BLR are blocked by the dust torus and not by the dust in host galaxies. Other studies suggest that some (∼40%; Young et al. 2008) red quasars have intrinsically red continuum (Puchnarewicz & Mason 1998;Whiting et al. 2001;Young et al. 2008;Rose et al. 2013;Ruiz et al. 2014) or an unusual covering factor of hot dust (Rose 2014). Also a synchrotron emission peak at NIR wavelength from radio jet (Whiting et al. 2001) has been proposed as a possible reason for the intrinsically red continuum of red quasars.
The hydrogen line ratios can be used to infer the amount of dust extinction over the BLRs. Rose et al. (2013) showed that 2MASS selected (J − K > 2) red quasars have higher L Hα /L Hβ ratios than unobscured type 1 quasars. However, the difference is modest (4.9 ± 0.5 versus 3.2 ± 0.4 for red and unobscured type 1 quasars, respectively) and the modest difference could originate from not only the dust extinction but also different physical condition in the BLRs.
In this study, we use Pβ and Pα lines with Balmer lines to investigate the red colors of red quasars that are shown to have broad emission lines (i.e., type 1). The use of the line ratios over a wide range of wavelength makes it easier to understand if the red colors are due to dust extinction or other mechanisms. Additionally, the use of AGN NIR diagnostics (e.g., Kim et al. 2010) allows us to measure black hole masses and bolometric luminosities in a way that is almost dust free. This fact is very advantageous when trying to distinguish several plausible mechanisms for red colors. For example, in a simple viewing angle scenario, we expect to find red quasars to have Eddington ratios similar to ordinary type 1 quasars, but not so in the intermediate population scenario. Throughout this paper, we use a standard ΛCDM cosmological model of H 0 =70 km s −1 Mpc −1 , Ω m =0.3, and Ω Λ =0.7, supported by previous observations (e.g., Im et al. 1997).

Sample and data
In this work, we used 20 red quasars at 0.186 < z < 0.842 that are composed of 16 red quasars (z > 0.5 and L bol > 10 46 erg s −1 ) studied in Kim et al. (2015b) and four additional red quasars (z < 0.5 and L bol ∼ 10 46 erg s −1 ). These 20 red quasars are a subsample of ∼ 80 spectroscopically confirmed red quasars in Glikman et al. (2007) and Urrutia et al. (2009), which were selected to be red quasars based on their broadband colors (R − K > 4 and J − K > 1.7 mag in Glikman et al. 2007; r ′ − K > 5 and J − K > 1.3 mag in Urrutia et al. 2009) from radio-detected 2MASS point sources. In this work, "radio-detected" means the detection of the object in the FIRST radio catalog (Becker et al. 1995). We note that radio-detected AGNs are not necessarily radio-loud. We define radio-loudness as R i = log( f (1.4 GHz)/ f (7480 Å)) (Ivezić et al. 2002). In our red quasars sample, we find that only one-third are radio-loud (e.g., R i > 2; Karouzos et al. 2014) and the remaining two-thirds are in the radio-intermediate regime (1 < R i < 2). Additionally, these quasars are chosen to be at z ∼ 0.3 or z ∼ 0.7 so that we can sample both the Balmer and Paschen lines.
In this study, we compare several properties of red quasars to those of unobscured type 1 quasars. We used 623 unobscured type 1 quasars at z ∼ 0.3 and 0.7, both selected from Sloan Digital Sky Survey (SDSS) in an approach similar to the red quasars (also see Section 5.3), and 37 unobscured type 1 quasars at z < 0.5 for which Paschen line information is available from Kim et al. (2010). The 37 unobscured type 1 quasars are relatively bright (K < 14.5 mag). The unobscured type 1 quasars cover a wide range in the K band luminosities (−30.0 < K mag <  Glikman et al. (2007) and Urrutia et al. (2009) from which our sample is drawn. Four red quasars (0036−0113, 0817+4354, 1209−0107, and 1307+2338), which were not used in Kim et al. (2015b), are denoted by filled red circles. The comparison sample is made of 37 unobscured type 1 quasars from Kim et al. (2010;blue open triangles), and 623 SDSS unobscured type 1 quasars at 0.275 < z < 0.363 and 0.562 < z < 0.842 (gray dots). −21.3), which overlaps well with the K-band luminosities of our red quasars (−30.3 < K mag < −25.7). Although the unobscured type 1 quasars from Kim et al. (2010) include six faint (K mag > −24) sources, the host galaxy contamination is negligible (< 8 %) due to a narrow slit width (Glikman et al. 2006;Kim et al. 2010). Figure 1 indicates the redshifts and absolute K-band magnitudes of the 20 red quasars and the comparison quasar sample. The absolute K-band magnitudes are K−corrected using a composite spectrum of unobscured type 1 quasars (NIR: Glikman et al. 2006;MIR: Kim et al. 2015a). Here, we do not apply the dust extinction correction, and we note that the amount of extinction is rather small for K band. The mean E(B − V) of red quasars is 0.88 (A K = 0.45 mag).
For 13 of the 20 red quasars, both the optical and NIR spectra are available and the remaining objects have only NIR spectra. The optical spectra of the red quasars come from Glikman et al. (2007Glikman et al. ( , 2012 while the NIR spectra are from Glikman et al. (2007Glikman et al. ( , 2012 and Kim et al. (2015b). Additionally, new NIR spectra were obtained for two red quasars (0036−0113 and 1307+2338) using the SpeX instrument ) on the NASA Infrared Telescope Facility (IRTF). The observation was performed with a set of the short cross-dispersion mode (SXD; 0.8-2.5 µm) and a 0 ′′ . 8 slit width to achieve a resolution of R ∼ 750 (400 km s −1 in FWHM) under clear weather and good seeing conditions of ∼ 0 ′′ . 7. In order to get fully reduced spectra, we used the Spextool software (Vacca et al. 2003;Cushing et al. 2004). The spectrophotometric calibration was carried out using the standard star spectra taken before or after the spectrum of each quasar was obtained. These flux-calibrated spectra were matched with a K−band photometry from 2MASS and additional flux scaling was performed when necessary. The fully reduced and calibrated NIR spectra of the two red quasars are provided in Table 3 by a machine readable table form. Among our sample, we detected Pβ in 19 and Pα in two red quasars (only one red quasar is detected in both Pβ and Pα lines) at signal-to-noise ratios (S/N) of > 5. Among 13 red quasars with Article number, page 2 of 14  (Kim et al. 2015b). Then, the 5-sigma limit of the integrated luminosity density within the wavelength width of the FWHM value is adopted.
As summarized in Table 1, 12 red quasars have both Hβ and Pβ detections or upper limits, while only Pβ line is detected for 7 red quasars. A few red quasars have Hα or Pα line detected.

Analysis
We corrected the spectra for the Galactic extinction by adopting the reddening map of Schlafly & Finkbeiner (2011), and the extinction curve with R V = 3.1 of Fitzpatrick (1999). Then, we shifted the corrected spectra to the rest frame. In order to estimate the Hβ luminosity, first we determined the continuum spectrum around the Hβ line. Spectrum of type 1 quasar around the Hβ line is generally complex, which contains a power-law continuum component, a host galaxy component, blended Fe ii multiplets, and several narrow emission lines (e.g., [O ii] λ3726, Hγ λ4340, and [O iii] λλ4959, 5007 doublet).
We fit the continuum of the red quasars with the power-law component and Fe template (Boroson & Green 1992). To avoid contamination from the several narrow emission lines around the Hβ line, the continuum-fitting regions were chosen as 4150-4280 Å, 4400-4750 Å, 4910-4945 Å, and 5070-5400 Å. For the fit, a MPFIT (Markwardt 2009) code based on an interactive data language (IDL) procedure was used to determine the power-law continuum and Fe blends. The Fe blends were determined by broadening and scaling the Fe template from the spectrum of IZw1 (Boroson & Green 1992). Figure 2 shows the optical spectrum of 2339−0912 with the fitted continuum model (the powerlaw continuum and Fe blends) and the continuum-subtracted spectrum.
We omitted the host galaxy component from the fit for two reasons. First, among the 11 Balmer line measurable red quasars, the measured rest-frame 5100 Å continuum luminosities (L 5100 = λL λ at 5100 Å) of 6 red quasars (0825+4716, 1113+1244, 1227+5053, 1434+0935, 1720+6156, and 2339−0912) are very luminous (extinction-corrected L 5100 = L bol /9.2 > 10 45 erg, s −1 ; Table 2). For type 1 AGNs as luminous as the red quasars in L 5100 (after the extinction correction), the host galaxy contamination is known to be well below 1% (Shen et al. 2011), and we assume the same for the red quasars. Then, even if we assume that the dust extinction occurs only in the nuclear part 1 with E(B − V) = 1, the host contamination would be < 30 % at 5100 Å. Second, we measured the equivalent width (EW) of Ca ii K λ3934 absorption line for the remaining 5 red quasars, since an EW of 1.5 Å for Ca ii K corresponds to the host galaxy contribution of ∼10 % (Greene & Ho 2005;Kim et al. 2006). We could not detect the Ca ii K absorption line for 3 red quasars, and for the remaining 2, we measured EWs of 3.8 Å, which corresponds to the host galaxy contribution of ∼25 % (0036−0113 and 1307+2338), by scaling the results of Greene & Ho (2005) and Kim et al. (2006). Even if the host galaxy contamination of red quasars is bigger than 25 %, we stress that it will not affect the line luminosity and width measurements much since the continuum is well subtracted for each object.
After the continuum subtraction, we fit the narrow component of Hβ line using [O iii] line at 5007 Å as a template when possible. The [O iii] line was fitted with a double Gaussian function to reproduce both the symmetric component and asymmetric outflow component (Greene & Ho 2005), and the fitted result was used as the template. This was possible for nine red quasars.
Among the nine [O iii]-fitted red quasars, the Hβ line of four red quasars (0036−0113, 0915+2418, 1307+2338, and 1532+2415) is well fitted by the narrow component only, although 0036−0113 and 1307+2338 were reported to have a broad Hβ component (Glikman et al. 2007). For these four red quasars, we estimated the upper limit of the Hβ line luminosities. In order to estimate the upper limit of the Hβ line luminosities, we performed the same procedures (see Section 2) in the narrow component subtracted spectra.
A&A proofs: manuscript no. redcolor For the [O iii]-unfitted three red quasars (1309+6042, 1720+6156, and 2339−0912), we fit broad Hβ line with a single (1309+6042) or double (1702+6156 and 2339−0912) Gaussian function, and the fit is performed by the MPFIT. Similar to the procedure above, the central wavelength, line width, and line flux are set to be free parameters. The FWHMs of the fitted components of the three red quasars are bigger than 800 km s −1 and this result means that the fitted component represent the broad components. The flux of the Hβ line is taken to be the sum of the two broad components for 1704+6156 and 2339−0912, while only the flux of the single broad component is taken to be the broad line flux for 1309+6042. Figure 3 shows the Hβ line-fitting results.
To fit the Hα line, we used a similar procedure as for Hβ line fitting. However, for the fitting the continuum, we use a single power-law component only because of the negligible contribution of the Fe blends around the Hα line. After the continuum subtraction, fitting the Hα is somewhat more complicated than fitting the Hβ line because the Hα line is blended with a [N ii] λλ 6548, 6583 doublet. For Hα line measurable red quasars 0036−0113, 0817+4354, and 1307+2338, we con- The Pβ and Pα lines are detected for 19 and 2 red quasars, respectively. Among these, the Pβ luminosities of 16 red quasars are adopted from Kim et al. (2015b). The Paschen line luminosi- ties of the other red quasars are newly measured, following the procedure in Kim et al. (2010). Figure 5 shows the fitting of the Paschen lines. This procedure fits the line with a single or double Gaussian functions, neglecting the NLR component, since the S/N and the resolution of the spectra do not allow us to measure the NLR component. The measured line luminosities and line widths are taken as the values for the broad line. Hence, the derived broad line luminosities and widths can be somewhat biased. To correct for the bias we applied the mean correction factors that are derived from well-resolved Balmer lines of 26 local unobscured type 1 quasars (Kim et al. 2010). They measured the fluxes and FWHMs of the well-resolved Balmer lines with a single, double, and multiple Gaussian functions, and derived the correction factors by comparing the measured properties. The mean correction factors are flux multi /flux double = 1.05 and flux multi /flux single = 1.06 for correcting the luminosities and FWHM multi /FWHM double = 0.85 and FWHM multi /FWHM single = 0.91 for correcting FWHM values. For more details, see Section 2.3 of Kim et al. (2010).
When no Hβ line is detected (0817+4354 and 1656+3821), we estimated the upper limit of the Hβ line luminosity. In order to estimate the upper limit of the Hβ line luminosity, we used the same procedure mentioned above. The measured upper limits of Hβ line luminosities of the two red quasars are provided in Table  1.
In total, we obtained the broad line luminosities of eight Hβ, three Hα, 19 Pβ, and two Pα lines. The line luminosities and FWHMs are summarized in Table 1. We note that all the values in Table 1 are only corrected for Galactic extinction.

Results
In order to investigate the factor leading to the red color of red quasars, we compared the luminosity ratios of the Pβ/Hβ, Pβ/Hα, Pα/Hβ, and Pα/Hα of red quasars to those of unobscured type 1 quasars. For a typical red quasar with the E(B − V) = 2 mag, its Hα and Hβ luminosities would be suppressed by a factor of 100 and 1000, respectively, but the Pα and Pβ fluxes are suppressed by a factor of only 2.3 and 4.7, respectively. Therefore, the analysis using optical to NIR emission lines is a useful way to investigate the dust obscuration, where the Paschen lines can serve as a good measure of the unobscured light.   For this analysis, the line luminosity ratios of unobscured type 1 quasars are adopted from Kim et al. (2010). Although several faint (L Paschen < 10 42 erg s −1 ) unobscured type 1 quasars are included in this comparison, the line luminosity ratios are insensitive to the luminosity (see Figure 6 in Kim et al. 2010). Figure  6 compares line luminosities of the red quasars and unobscured type 1 quasars. We find that the observed Balmer line luminosity for a given Paschen line luminosity is 1.5-290 (∼12 on average) times weaker for the red quasars than for unobscured type 1 quasars. Figure 7 shows the distributions of the Pβ/Hβ luminosity ratios of the red quasars and unobscured type 1 quasars. We find that red quasars have the log(L Pβ /L Hβ ) values much higher than those of the unobscured type 1 quasars; median log(L Pβ /L Hβ ) of the red quasars are 0.27±0.53, in contrast to that of unobscured type 1 quasars (-0.49±0.17). The Kolmogorov-Smirnov (K-S) test confirms this significant difference in the line luminosity ratios between the two quasar populations. We performed the K-S test using the KSTWO code based on the IDL. The maximum deviation between the cumulative distributions of these two Pβ/Hβ luminosity ratios, D, is 1.0, and the probability of the result given the null hypothesis, p, is only 9.22×10 −7 .
If red quasars are dust-reddened, we expect the dustcorrected line luminosity ratios of the red quasars to be consistent with those of unobscured type 1 quasars. For this  Fig. 7. Distributions of the Pβ/Hβ luminosity ratios of the red quasars (color-hatched histogram) and unobscured type 1 quasars (gray histogram). The blue and red histograms indicate the distributions of the extinction corrected and uncorrected Pβ/Hβ luminosity ratios of the red quasars, respectively. After applying the extinction correction, the Pβ/Hβ distribution of red quasars agree broadly with that of unobscured type 1 quasars, but with a much larger scatter suggesting that the extinction-correction prescription is not perfect. test, we adopted the E(B − V) values from previous studies (Glikman et al. 2007;Urrutia et al. 2009) and applied the ex-  The extinction-corrected median log(L Pβ /L Hβ ) of the red quasars is -0.63±0.81, that is almost the same as that of unobscured type 1 quasars. Furthermore, our K-S test for the histogram of the extinction-corrected Pβ/Hβ luminosity ratios show that the measured D and p values of the K-S statistic are 0.44 and 0.12, respectively, against unobscured type 1 quasars. For this K-S test, we broadened the Pβ/Hβ luminosity ratios distribution of unobscured type 1 quasars by adding scatters to Pβ/Hβ luminosity ratios through a Monte Carlo simulation by the amount that could be produced during dereddening of Pβ/Hβ luminosity ratios of red quasars assuming a typical scatter in E(B − V) of 0.5 (Glikman et al. 2007). Therefore, we conclude that, on average, both the continuum colors and the line ratios of red quasars can be explained by dust extinction.
One interesting question is whether we can accurately determine E(B − V) values of red quasars. In an attempt to make this determination The large scatter between E(B − V) cont and E(B − V) line is likely due to the wide range of continuum slopes that quasars can have and the difficulty in estimating the intrinsic continuum shape in advance. Another possible reason is that the dust obscuration region varies between the continuum emitting and line emitting regions. We suggest that E(B−V) estimated through the continuum shape contains a large scatter (∼ 0.72 in E(B − V)).
This agrees with what we saw in Figure 7, where the histogram of L Pβ /L Hβ after the extinction correction with E(B − V) cont is much broader than the same histogram for unobscured type 1 A&A proofs: manuscript no. redcolor quasars and the two histograms are virtually indistinguishable after taking into account this broadening effect.

Physical condition as a cause for high line luminosity ratio
Here, we test a hypothesis that the observed L Paschen /L Balmer of red quasars are due to a different physical condition of broad emission line regions (BELRs) without dust extinction. To do so, we explore what physical conditions of BELRs can reproduce the observed line luminosity ratios by computing theoretically expected line luminosity ratios under different physical conditions using the CLOUDY code (version 13.03; Ferland et al. 1998).
We set the plausible ranges of input parameters as the following. Quasars do not show any broad forbidden lines (e.g., see Glikman et al. 2007;Urrutia et al. 2009). The absence of the broad forbidden lines implies that the hydrogen density (n H ) in the BELR is higher than the critical density of the forbidden lines, which gives us the lower limit of n H = 10 7 cm −3 . The upper limit of n H is set to 10 14 cm −3 considering the existence of strong Fe blends (∼ 10 12 cm −3 ; Collin-Souffrin et al. 1982;Rees et al. 1989). For the other parameters, we vary the values of the shape of the ionizing continuum (α = −2 ∼ 2) and the ionization parameter (U = 10 −5 ∼ 10 5 ) to cover various physical conditions. Line luminosity ratios of L Pα /L Hβ , L Pβ /L Hβ , and L Hα /L Hβ are sensitive to n H and U. The line luminosity ratios decrease when the n H and U values are increased, which is indicated in the B and C panels of Figure 9. Figure 9 shows the line luminosity ratios as a function of wavelength for red quasars and those from the CLOUDY calculation. The line luminosity ratios of unobscured type 1 quasars can be successfully reproduced by the CLOUDY calculation with a set of parameters, α = −1.0, U = 10 −1.5 , and n H = 10 9 cm −3 (Kim et al. 2010), which is represented by the dotted line in Figure 9. For the Balmer lines, although the Hα/Hβ luminosity ratios of red quasars are not measured in this study, these luminosity ratios of local red AGNs are only moderately different from those of unobscured type 1 quasars (Rose et al. 2013).
However, the line luminosity ratios start to demand unusual physical conditions when Paschen lines are included. The median log(L Pβ /L Hβ ) is 0.43 ± 0.53 for red quasars, while it is only −0.48 ± 0.17 for unobscured type 1 quasars. Among the eight L Pβ /L Hβ ratios measured red quasars, five (63 %) red quasars have higher line ratios than the maximum line luminosity ratios in the CLOUDY calculation (log(L Pβ /L Hβ ) ≤ 0.18). In other words, the L Pβ /L Hβ ratios of the five red quasars are much higher than the line luminosity ratios from all the plausible physical parameters for BLR. For the remaining three (38 %) red quasars, the L Pβ /L Hβ ratios can be explained with a condition of low n H = 10 7 cm −3 , low U = 10 −3 , and α = 2. The parameters are found by minimizing χ 2 , which is a function of the line luminosity ratios where N is the number of line luminosity ratios, and two types of R i are the line luminosity ratios either from observation or the CLOUDY model, and σ i is the uncertainty in the measured line luminosity ratio. However, this physical condition, that is, low n H and low U, is somewhat similar to that of NLRs (e.g., n e = 10 3.5 ∼ 10 7.5 cm −3 and U ∼ 10 −2 ; Osterbrock 1991; Netzer 2013), rather than the broad lines that we are studying here (FWHM > 800 km s −1 ), and this disfavors the physical condition as a reason for the observed line luminosity ratios.

5.2.
High hot dust covering factor as a cause for redness Rose et al. (2013) suggested that some of AGNs with red J − K colors are red owing to a relatively large hot dust covering factor (CF HD = L HD /L bol ; Maiolino et al. 2007;Kim et al. 2015a). We address this point here with our sample. For the comparison between the CF HD of red quasars and unobscured type 1 quasars, we used 37 unobscured type 1 quasars from Kim et al. (2010). The unobscured type 1 quasars are bright (K < 14.5 mag and M i < −23 mag) and located at z < 0.5.
Because the L bol and L HD are easily under-or overestimated from the spectral energy distribution (SED) model fitting depending on a set of the dust extinction value and extinction law, we used the Paschen line luminosities and NIR continuum luminosities (L 2 and L 3.5 ; λL λ at 2 µm and 3.5 µm in the 10     rest-frame) as proxies for the L bol and L HD , respectively. Considering previous results that found, first, the temperature of hot dust torus is ∼1000-1500 K (Barvainis 1987;Glikman et al. 2006;Kim et al. 2015a;Hernán-Caballero et al. 2016) and, second, 2 µm and 3.5 µm are closed to the peak wavelengths of blackbody radiation from the hot dust component, L 2 and L 3.5 can represent L HD (Glikman et al. 2006;Kim et al. 2015a). Although the stellar emission can peak at 1.6 µm, the hot dust component can be a dominant component with its peak at 2-3.5 µm. This is supported by J − K colors of red quasars that are significantly different from those of normal galaxies and starforming galaxies (e.g., see Figure 1 in Glikman et al. 2012). Although the NIR continuum luminosities of red quasars have a possibility to be overestimated by the stellar emission contam-ination, the NIR contribution of the stellar emission is known to be ≤10 % for quasars, when NIR continuum luminosity is over than 10 43.5 erg s −1 (Hernán-Caballero et al. 2016). There is no plausible reason to believe that the hot and warm dust emission should be much weaker for red quasars than type 1 quasars. If red quasars are red due to unusually high CF HD , this would make the host galaxy contribution to the NIR continuum luminosities even smaller. The Paschen line luminosities are used as a tracer for the L bol (Kim et al. 2015b) by employing the excellent correlation between L P , L Hα , L 5100 , and L bol (Kim et al. 2010;Shen et al. 2011;Jun et al. 2015). The wavelengths of 2 µm and 3.5 µm are not too far from the Paschen lines (Pβ: 1.2818 µm and Pα: 1.8751 µm) and so the L 2 /L P and L 3.5 /L P are rather insensitive to the exact values of dust extinction. A&A proofs: manuscript no. redcolor The NIR continuum luminosities are measured by interpolating the 2MASS (Skrutskie et al. 2006) and WISE (Wright et al. 2010) photometric data. The Pβ and Pα luminosities are adopted from Kim et al. (2010) for 37 and 27 unobscured type 1 quasars, respectively.
In this study, we do not consider the Baldwin effect because the Baldwin effect in the Balmer lines is weak (Dietrich et al. 2002), and the Paschen lines have a strong correlation with the Balmer lines (Kim et al. 2010). Figures 10 and 11 show the comparisons between the NIR continuum luminosities versus Paschen line luminosities of the red quasars and unobscured type 1 quasars. The unobscured type 1 quasars have the mean log(L 2 /L Pβ ) and log(L 3.5 /L Pβ ) of 2.58±0.01 and 2.64±0.01 with dispersions of 0.55 and 0.51, respectively. The measured luminosity ratios are only slightly smaller than those of red quasars (3.06±0.01 and 3.21±0.01 with dispersions of 0.27 and 0.28). Moreover, the mean log(L 2 /L Pα ) and log(L 3.5 /L Pα ) of the unobscured type 1 quasars are 2.51±0.01 and 2.56±0.01 with dispersions of 0.64 and 0.60, respectively, and those of red quasars (2.72±0.02 and 2.97±0.02 with dispersions of 0.32 and 0.35) are almost same.
The result indicates that the covering factor of red quasars are not much different from type 1 quasars, and an unusually large covering factor is not the reason for their red colors. This result is consistent with the viewing angle or torus obscuration scenario, but we address this scenario below. Wilkes et al. (2002) suggested that the redness of red quasars arises from a moderate viewing angle in the quasar unification model when the accretion disk and the BLR are viewed through a dust torus. If so, red quasars should show properties very similar to unobscured type 1 quasars.

Viewing angle as a cause for redness
However, previous studies Kim et al. 2015b) showed that red quasars have significantly higher accretion rates than unobscured type 1 quasars, which cannot be explained by the viewing angle scenario. Among the 19 Pβ luminosity measured red quasars in our sample, the Eddington ratios of 16 red quasars at z ∼ 0.7 were already studied in Kim et al. (2015b). We have three additional Pβ measured red quasars at z ∼ 0.3 (0036−0113, 1209−0107, and 1307+2338). Below, we first examine the Eddington ratios of the three additional quasars and then compare the Eddington ratios of 16 red quasars at z ∼ 0.7 to the L bol matched unobscured type 1 quasars.
As a comparison sample, we select unobscured type 1 quasars from the quasar catalog (Schneider et al. 2010) of the SDSS Seventh Data Release (DR7; Abazajian et al. 2009). In order to avoid the sample selection bias, the unobscured type 1 quasars are selected by the same selection criteria as red quasars as follows: (i) the same redshift range of the three red quasars (0.275< z <0.363), (ii) radio detection in Faint Images of the Radio Sky at Twenty-Centimeters (FIRST) survey (Becker et al. 1995), and (iii) photometric detection in 2MASS. This selection yields 213 unobscured type 1 quasars as the comparison sample.
In order to estimate the BH masses of red quasars, we corrected the Pβ luminosities by adopting E(B − V) values from previous studies (Glikman et al. 2007;Urrutia et al. 2009). The E(B − V) values were determined using their continuum shape. After then, we used the Pβ based M BH estimator from Equation (1) of Kim et al. (2015b), which is modified from the Pβ scaling relation (Kim et al. 2010) by adopting a recent virial coefficient of log f = 0.05   from Shen et al. (2011) and the M BH estimator of Equation (2) in Kim et al. (2015b).
The Hβ-based M BH values are lower than the Pβ-based M BH values by 0.45 dex before the extinction correction. No offset can be seen after the Hβ-based M BH values are derived by taking the extinction effect into account, but this introduces a large scatter (rms∼ 0.56 dex) in M BH values from the two methods, justifying the use of Pβ-based M BH here.
To obtain L bol values of the red quasars, we translated the L Pβ using the relationship between the L bol and L Pβ of Equation (4) in Kim et al. (2015b). The L bol values of the unobscured type 1 quasars are converted from the L 5100 values with a bolometric correction factor of 9.26 (Shen et al. 2011). We note that the measured M BH and L bol values of the red quasars are summarized in Table 2.
The L bol values of the red quasars (10 45.50 erg s −1 < L bol < 10 46.51 erg s −1 ) and the comparison sample of unobscured type 1 quasars (10 45.04 erg s −1 < L bol < 10 46.55 erg s −1 ) are similar but not identical. For that reason, we also performed the comparison of the two populations using L bol -matched samples too.
To estimate the Eddington ratio, the Pβ is used as both an indicator of L bol and M BH . Although these two different quantities are derived from the Pβ line and include the same line luminosity term, the M BH has a larger uncertainty than L bol because the (B) 46.6 <log (L bol / erg s -1 ) < 47.2 Fig. 13. (A) Eddington ratio distributions of red quasars and unobscured type 1 quasars at z ∼ 0.7. The red solid and blue dashed histograms represent red quasars and unobscured type 1 quasars, respectively. The Eddington ratios of red quasars at z ∼ 0.7 are adopted from Kim et al. (2015b). (B) Eddington ratio distributions for high-luminous quasars (46.6 < log(L bol /erg s −1 ) < 47.2). (C) Eddington ratio distributions for low-luminous quasars (46.0 < log(L bol /erg s −1 ) < 46.6).
M BH values include FWHM square term and the square-rooted line luminosity term. The median uncertainties of the Pβ based L bol and M BH are 0.03 and 0.08 dex, respectively. The Eddington ratio is proportional to L 0.5 /FWHM 2 and this gives the combined uncertainty of 0.08 dex 2 We compared the Eddington ratios (L bol /L Edd , where L Edd is the Eddington luminosity) of red quasars and unobscured type 1 quasars in Figure 12. The median Eddington ratios of unobscured type 1 quasars is 0.12 with an rms scatter of 0.34. Meanwhile, the measured Eddington ratios of three red quasars are 0.83, 0.30, and 0.52 for 0036−0113, 1209−0107, and 1307+2338, respectively. Among 213 unobscured type 1 quasars, only 55 unobscured type 1 quasars have higher Eddington ratios than 0.30. The probability is only 1.7 % that three randomly chosen unobscured type 1 quasars from the comparison sample have the Eddington ratios higher than 0.30. No systematic bias is expected between L bol /L Edd from Pβ (red quasars) and L 5100 with FWHM Hβ (unobscured type 1 quasars), except for an added scatter of 0.36 dex that comes from the scatter in the correlation of different quantities (Kim et al. 2015b).
Since several works suggest that the Eddington ratios are dependent on L bol (e.g., Lusso et al. 2012;Suh et al. 2015), we also show the Eddington ratio distribution of the sample in a narrow L bol range that matches the L bol of red quasars. The objects 0036−0113 and 1307+2338 have similar L bol of 10 45.79 erg s −1 and 10 45.50 erg s −1 , respectively, but 1209−0107 has a much higher L bol of 10 46.51 erg s −1 . For comparison, we selected 48 unobscured type 1 quasars with the same L bol range (10 45.50 < L bol < 10 45.79 erg s −1 ) of the 2 red quasars 0036−0113 and 1307+2338. The median Eddington ratio of the 48 unobscured type 1 quasars changes to 0.15±0.33, but this is still much lower than those of the red quasars. Among the 48 unobscured type 1 quasars, only 9 unobscured type 1 quasars have higher Eddington ratios than the minimum Eddington ratio, 0.52, of 0036−0113 and 1307+2338. The probability is only 3.2 % that 2 randomly chosen unobscured type 1 quasars from the comparison sample have the Eddington ratios higher than 0.52.
Previous studies Kim et al. 2015b) showed red quasars at different redshift have higher Eddington ratios than unobscured type 1 quasars. Kim et al. (2015b) measured Eddington ratios for 16 red quasars at z ∼ 0.7 and the 16 red quasars are included in our 20 red quasars. Kim et al. (2015b) compared the Eddington ratios of the 16 red quasars to those of unobscured type 1 quasars that are matched in M BH . We show how the Eddington ratios of 16 red quasars compare with those of L bol -matched unobscured type 1 quasars. The unobscured type 1 quasars are selected from the SDSS DR7 quasar catalog (Schneider et al. 2010) with the same redshift range of the 16 red quasars (0.56 < z < 0.84). We note that the selection method of the comparison sample is identical to Kim et al. (2015b) and the details of the comparison sample selection are described in Section 2.1 of Kim et al. (2015b).
We divided the red quasars and the unobscured type 1 quasars into two luminosity bins. Figure 13 shows the comparison of the Eddington ratio distributions and our results show that the Eddington ratios of 16 red quasars at z ∼ 0.7 are significantly higher (factors of ∼3 to 4) than those of unobscured type 1 quasars. Even if we assume a maximal amount of 35 % contamination on the Paschen line flux by a narrow line (see Section 3), the Eddington ratio decreases by only ∼15 % or 0.07 dex, and we find that the analysis result does not change.
For the low luminosity sample, 9 red quasars and 246 unobscured type 1 quasars are selected, and their log(M BH /M ⊙ ) values have ranges of 7.98-9.07 and 7.71-10.27 for the red quasars and the unobscured type 1 quasars, respectively. The median Eddington ratios of the red quasars and the unobscured type 1 quasars are 0.62 and 0.15, respectively. The D and p values from a K-S test between these two distributions are 0.64 and 7.6×10 −4 . The high luminosity sample includes 7 red quasars and 36 unobscured type 1 quasars. The log(M BH /M ⊙ ) values have ranges of 8.26-9.09 and 8.58-10.18 for the red quasars and unobscured type 1 quasars, respectively. The median Eddington ratios of the red quasars and unobscured type 1 quasars are 0.97 and 0.35, respectively. The measured D and p values between these two distributions are 0.66 and 5.6 × 10 −3 .
Overall, the additional analysis of the three red quasars and the re-analysis of the luminosity matched 16 red quasars in Kim et al. (2015b) strengthens the previous results of Urrutia et al. (2012) and Kim et al. (2015b) that red quasars tend to have higher Eddington ratios than unobscured type 1 quasars. Therefore, we suggest that many of the red quasars, if not all, are not seen as red simply because of the viewing angle (also see Onori et al. 2017).