Near-IR Spectroscopy of Luminous LoBAL Quasars at 1 < z < 2.5

Our sample is drawn from the BAL QSO catalog from Allen et al. (2011). They measured BAL properties for the high-ionization lines Si ivλ1400 and C ivλ1550 and the low-ionization lines Al iiiλ1860 and Mg iiλ2800 from quasar spectra in the Sloan Digital Sky Survey (SDSS) DR6 spectroscopic survey (Schneider et al. 2007, 2010). A BAL QSO in their sample is defined as having a nonzero balnicity index (BI), where the BI is defined following Weymann et al. (1991) and measures the presence of a continuous broad absorption feature below a threshold of 0.9 with respect to the normalized continuum.

Using this definition, their sample contains 368 LoBAL QSOs, identified by a $\mathrm{BI}\gt 0$ in either Mg ii or Al iii. We are selecting our targets in two redshift windows from this LoBAL QSO sample. These are chosen such that Hα falls well within the atmospheric window not strongly affected by telluric absorption bands in either H band or K band. In addition, for most targets Hβ falls into J band or H band, respectively. Specifically, we target the following:

(1) The redshift range $1.32\lt z\lt 1.60$, BI(Mg ii) > 0, and a 2MASS H-band magnitude $H\lt 16.7$ mag, giving a sample of 23 targets. We removed SDSS J014349.15+002128.3 from the sample, since this object is not a true LoBAL QSO but a superposition of a normal QSO with intervening absorption by a foreground star (W. Liu 2017, private communication). These 22 LoBAL QSOs form our initial z ∼ 1.5 sample. We obtained NIR spectroscopy for 12 of these LoBAL targets. For 10 of these we have spectroscopy for both Hα and Hβ.

(2) The redshift range $2.2\lt z\lt 2.5$, BI(Al iii) > 0, and a 2MASS K-band magnitude $K\lt 15.3$ mag. The relatively bright luminosity cut is motivated by the smaller-size telescopes used for NIR spectroscopy of the sample in this redshift bin to ensure acceptable signal-to-noise ratio (S/N) in the spectra. Focusing on these bright LoBALs gives a sample of 11 targets, forming our initial z ∼ 2.3 sample. We obtained K-band spectra for 10 of them, covering Hα, and additionally H-band spectra covering Hβ for seven of these, where six have acceptable S/N and are used here. For nine of the z ∼ 2.3 LoBALs BOSS spectra are available that cover also the Mg ii line at these redshifts, contrary to SDSS-I/II spectra. The presence of a clear absorption trough also in Mg ii can be confirmed from the BOSS spectra.

The basic sample properties are provided in Table 1. We indicate clear cases of FeLoBALs in our sample. In total, we identified three FeLoBALs in the z ∼ 1.5 sample and another three in the z ∼ 2.3 sample, based on visual inspection. SDSS J0841+2005 is known as an FeLoBAL that shows very strong changes in its absorption systems (Rafiee et al. 2016; Stern et al. 2017).

Table 1.  Sample Summary

Name	R.A. (J2000)	Decl. (J2000)	${z}_{\mathrm{NIR}}$	${z}_{\mathrm{HW}}$	BI(Mg ii)	BI(Al iii)	${H}_{2\mathrm{MASS}}$	${K}_{2\mathrm{MASS}}$	Inst.(Hβ)	Inst.(Hα)	Type
SDSS J0033+0632	00:33:35.638	+06:32:07.58	1.502	1.505	188.7	457.3	15.785	15.748	TSPEC	TSPEC	Lo
SDSS J0132−0046	01:32:45.302	−00:46:10.01	1.469	1.475	147.6	342.4	16.539		TSPEC	TSPEC	Lo
SDSS J0859+4239	08:59:10.400	+42:39:11.38	1.497	1.499	5212.0	496.2	15.381	15.853	TSPEC	TSPEC	Lo
SDSS J0952+0257	09:52:32.212	+02:57:28.39	1.358	1.358	1034.5	648.0	15.493	15.424	TSPEC	TSPEC	Lo
SDSS J0957+4406	09:57:21.361	+44:06:42.91	1.459	1.468	90.0	381.5	15.665	15.169	TSPEC	TSPEC	Lo
SDSS J1019+0225	10:19:27.371	+02:25:21.44	1.364	1.364	4648.4	0.0	15.215	15.114	TSPEC	TSPEC	Lo
SDSS J1128+0623	11:28:51.837	+06:23:15.38	1.513	1.497	67.3	0.0	14.754	14.438	TSPEC	TSPEC	Lo
SDSS J1440+3710	14:40:02.245	+37:10:58.52	1.401	1.414	1540.7	0.0	15.316	14.761	⋯	ISLE	FeLo
SDSS J1448+0424	14:48:42.451	+04:24:03.12	1.539	1.546	64.3	126.1	14.482	14.402	TSPEC	TSPEC	Lo
SDSS J1508+6055	15:08:48.805	+60:55:51.93	1.529	1.532	3068.7	0.0	15.241	14.756	NOTCam	ISLE	FeLo
SDSS J1511+4905	15:11:13.846	+49:05:57.37	1.368	1.361	369.8	1264.5	14.606	14.284	⋯	IRCS	Lo
SDSS J1556+3517	15:56:33.783	+35:17:57.39	1.501	1.495	9926.2	0.0	14.905	14.787	TSPEC	TSPEC	FeLo
SDSS J0841+2005	08:41:33.153	+20:05:25.81	2.345	2.276	⋯	11540.6	14.411	13.620	NOTCam	ISLE	FeLo
SDSS J0943−0100	09:43:38.218	−01:00:19.33	2.368	2.376	⋯	395.0	⋯	14.947	NOTCam	NOTCam	Lo
SDSS J1011+5155	10:11:08.895	+51:55:53.82	2.472	2.465	⋯	3182.4	15.979	15.245	⋯	NOTCam	Lo
SDSS J1019+4108	10:19:12.850	+41:08:07.41	2.471	2.460	⋯	6287.6	15.828	14.685	⋯	NOTCam	Lo
SDSS J1028+5110	10:28:50.317	+51:10:53.11	2.418	2.426	⋯	128.7	15.591	14.660	NOTCam	NOTCam	FeLo
SDSS J1132+0104	11:32:12.920	+01:04:41.35	2.377	2.328	⋯	884.0	16.254	15.177	⋯	NOTCam	Lo
SDSS J1134+3238	11:34:24.642	+32:38:02.45	2.454	2.461	⋯	3178.3	14.844	13.987	NOTCam	ISLE	FeLo
SDSS J1516+0029	15:16:36.786	+00:29:40.51	2.252	2.251	⋯	597.7	15.701	15.007	⋯	NOTCam	Lo
SDSS J1554+2218	15:54:33.131	+22:18:42.08	2.410	2.418	⋯	41.3	15.810	14.731	NOTCam	NOTCam	Lo
SDSS J1709+6303	17:09:30.996	+63:03:57.13	2.380	2.407	⋯	482.7	15.551	14.624	NOTCam	ISLE	Lo

Note. ${z}_{\mathrm{NIR}}$ is the redshift measured from the peak of either Hα or [O iii]; ${z}_{\mathrm{HW}}$ is the improved SDSS redshift from Hewett & Wild (2010); BI(Mg ii) and BI(Al iii) are the balnicity indices for these broad lines as measured by Allen et al. (2011); ${H}_{2\mathrm{MASS}}$ and ${K}_{2\mathrm{MASS}}$ are the 2MASS magnitudes taken from Schneider et al. (2010); Inst. gives the instrument used for spectroscopy of either Hα or Hβ (note that only TSPEC spectroscopy covers Hα and Hβ simultaneously); Type indicates whether the object is a regular LoBAL (Lo) or an obvious FeLoBAL (FeLo).

Download table as:  ASCIITypeset image

In the z ∼ 1.5 LoBAL sample we have discovered two cases with strong intrinsic Balmer absorption lines in Hα and Hβ through our observations. Intrinsic Balmer absorption in quasar spectra is a rare phenomenon, with only a handful of cases known so far (e.g., Aoki et al. 2006; Hall 2007; Zhang et al. 2015). We include these objects here in our statistical LoBAL study and reserve a more detailed discussion for a separate paper (A. Schulze et al. 2017, in preparation).

For our z ∼ 1.5 sample we used the NIR spectrograph TripleSpec (Wilson et al. 2004) at the Palomar Hale 200-inch telescope to observe nine of our z ∼ 1.5 targets in 2014 January under good conditions. TripleSpec provides simultaneous coverage from 1.0 to 2.4 μm at a spectral resolution of R ∼ 2700. A slit width of 1'' was used. Total exposure times varied between 40 and 60 minutes. Observations for three other quasars were obtained using ISLE on the 1.88 m telescope at Okayama Astrophysical Observatory (OAO), NOTCam on the 2.56 m Nordic Optical Telescope (NOT), and IRCS on the 8.2 m Subaru telescope (Kobayashi et al. 2000).

For the z ∼ 2.3 sample we mainly used the 2.56 m NOT, supplemented by the OAO 1.88 m telescope for three targets. Using NOTCam on the NOT, we obtained low-resolution $R=2500$ spectroscopy in either J, H, or K band with a slit width of 06. Observations were carried out during two runs in 2016 March and 2017 March under very good conditions with an average seeing of 07. Typical exposure times range between 30 and 60 minutes.

Observations at OAO were obtained during several runs from 2015 to 2017 under mostly poor conditions with varying seeing between 1'' and 25. Due to the seeing limitations, we increased the slit width to 2'' to avoid significant slit losses, leading to a reduced resolution of R ∼ 1000. Exposure times per target are between 1 and 2 hr. For each quasar a full calibration set (including dome flats and Xe and Ar arc lamps) was observed, together with a telluric standard star at similar airmass and position either before or after the quasar. We performed an ABBA dither pattern along the slit to improve the sky subtraction for all targets at each of these telescopes.

The data reduction for the spectroscopic data from TripleSpec is carried out using the modified IDL-based Spextool3 package (Cushing et al. 2004), as described in Zuo et al. (2015). This involved flat-field correction, sky background subtraction, wavelength calibration, and telluric correction. The telluric correction is based on several A0V stars observed each night. The data reduction of the spectra from the other facilities was performed using the IRAF software following the standard reduction steps for sky subtraction, flat-fielding, and telluric correction. We extracted the 1D spectrum and performed a wavelength calibration using either an Ar or Xe arc lamp.

We did not perform spectrophotometric flux calibration but tied the absolute flux calibration to their 2MASS magnitudes. Simultaneous NIR K-band observations of three of our targets using the Wide-Field Imager mounted on the 91 cm telescope at OAO showed that the current NIR photometry is fully consistent with their 2MASS photometry. All spectra are corrected for galactic extinction (Cardelli et al. 1989; Schlegel et al. 1998).

Spectral measurements are obtained from fitting the spectral regions around the broad Hα and Hβ line with a continuum+line profile model. Our procedure for continuum fitting and emission line modeling is similar to, e.g., Shen et al. (2011). Furthermore, we have masked out regions in the NIR that are strongly affected by telluric absorption.

For Hα we first fit a local power-law continuum to wavelength regions free from emission lines. To the continuum-subtracted spectrum we fit a line model over the range 6200–7000 Å rest frame consisting of up to three Gaussians for the broad Hα. We add a set of narrow lines if justified by the data, including a single Gaussian each for narrow Hα, [N ii] λλ6548, 6584, and [S ii] λλ6717, 6731, whose line widths and velocity offsets are tied together, while the flux ratio of the [N ii] lines is fixed to 2.96.

For Hβ we fit a local pseudo-continuum, consisting of a power-law continuum and an optical iron template (Boroson & Green 1992). In a few cases where the spectral quality in particular at the edges of the NIR spectra does not allow a reliable fit of the pseudo-continuum, we instead fit a power-law continuum to areas free of emission lines and strong iron contribution and allow for iron contribution by modeling the strongest iron features around Hβ and [O iii] by a double Gaussian centered at 4924 and 5018 Å (e.g., Schulze et al. 2009). We fit a line model over the range 4700–5100 Å to the pseudo-continuum-subtracted spectra. We fit again up to three Gaussians for the broad Hβ line. We use up to two Gaussians each to model the narrow [O iii] $\lambda \lambda 4959,5007$ lines, which are coupled together in their shape and their line ratio of 3.0. This allows us to capture the often asymmetric shape of the [O iii] line profile with a core and a blue wing component. The narrow Hβ line is fitted with a single Gaussian, with its velocity offset and line width fixed to the core component of [O iii].

We perform all our measurements (line and continuum luminosities, FWHM) from our best-fit spectral model. Uncertainties on these parameters are derived via a Monte Carlo approach. For each spectrum we generate 100 simulated spectra by adding Gaussian random noise to the spectra, with the standard deviation at each pixel taken from the flux error. Each simulated spectrum is automatically fitted with the same model, and the uncertainties for each measured parameter are obtained as the 1σ dispersion from the fits to the set of simulated spectra.

We here use the FWHM of the broad component of the Balmer lines as the preferred measure of line width. Compared to the line dispersion, FWHM is less dependent on the wings of the profile and thus tends to be more robust at low S/N. Furthermore, FWHM is the reported width measure in the literature we use for our comparison with the non-BAL QSOs.

We have remeasured the systemic redshifts for the LoBAL QSO sample from our spectral models. In cases where [O iii] is significantly detected we base our redshift estimate on the model peak of the [O iii] $\lambda 5007$ emission line, while we use the model peak of the total Hα profile otherwise. Our NIR redshift measurements are given in Table 1, together with the optical SDSS redshift determined by Hewett & Wild (2010). Given the potential difficulty of obtaining a reliable redshift based on the BAL-affected UV lines, the Hewett & Wild (2010) redshifts are in good agreement with the NIR redshifts, with only three objects having $| {z}_{\mathrm{HW}}-{z}_{\mathrm{NIR}}| \gt 0.02$ (and all within 0.07). Excluding these outliers, when defining the difference between the two redshift estimates by $c({z}_{\mathrm{HW}}-{z}_{\mathrm{NIR}})/(1+{z}_{\mathrm{NIR}})$ we find a mean offset of ∼150 km s⁻¹ and a dispersion of ∼820 km s⁻¹.

The continuum and line measurements are given in Table 2. Example line fits are shown in Figure 1. Line fits for the full sample are given in the Appendix.

Zoom In Zoom Out Reset image size

Table 2.  Spectral Measurements and Derived Black Hole Properties

Name	${\mathrm{FWHM}}_{{\rm{H}}\alpha }$	$\mathrm{log}{L}_{{\rm{H}}\alpha }$	${\mathrm{FWHM}}_{{\rm{H}}\beta }$	$\mathrm{log}{L}_{{\rm{H}}\beta }$	logL5100	$\mathrm{log}{L}_{\mathrm{bol}}$	$\mathrm{log}{M}_{\mathrm{BH}}$	$\mathrm{log}\lambda$
	(km s−1)	(erg s−1)	(km s−1)	(erg s−1)	(erg s−1)	(erg s−1)	(${M}_{\odot }$)
SDSS J0033+0632	2475 ± 29	44.95 ± 0.01	3165 ± 107	44.08 ± 0.02	45.60 ± 0.03	47.06 ± 0.01	8.96 ± 0.01	−0.01 ± 0.01
SDSS J0132−0046	4973 ± 130	44.42 ± 0.01	7611 ± 1745	43.37 ± 0.06	45.20 ± 0.01	46.53 ± 0.01	9.35 ± 0.02	−0.93 ± 0.02
SDSS J0859+4239	8317 ± 1145	45.02 ± 0.01	7895 ± 500	44.11 ± 0.03	45.97 ± 0.01	47.14 ± 0.01	10.11 ± 0.18	−1.09 ± 0.18
SDSS J0952+0257	3048 ± 18	44.85 ± 0.01	3594 ± 184	43.95 ± 0.02	46.03 ± 0.01	46.97 ± 0.01	9.11 ± 0.01	−0.25 ± 0.01
SDSS J0957+4406	7880 ± 126	44.64 ± 0.01	5853 ± 473	43.78 ± 0.04	46.02 ± 0.01	46.75 ± 0.01	9.88 ± 0.02	−1.24 ± 0.02
SDSS J1019+0225	6440 ± 3573	45.01 ± 0.01	8545 ± 919	44.07 ± 0.06	45.94 ± 0.01	47.12 ± 0.01	9.87 ± 0.35	−0.86 ± 0.35
SDSS J1128+0623	2574 ± 61	45.22 ± 0.01	2580 ± 208	44.54 ± 0.02	46.39 ± 0.01	47.33 ± 0.01	9.13 ± 0.02	0.10 ± 0.02
SDSS J1440+3710	4584 ± 104	44.57 ± 0.01	⋯	⋯	⋯	46.68 ± 0.01	9.35 ± 0.02	−0.77 ± 0.02
SDSS J1448+0424	5531 ± 158	45.36 ± 0.01	8664 ± 537	44.60 ± 0.03	46.53 ± 0.01	47.47 ± 0.01	9.90 ± 0.03	−0.54 ± 0.03
SDSS J1508+6055	3601 ± 1957	44.98 ± 0.01	4694 ± 373	44.18 ± 0.03	46.22 ± 0.01	47.09 ± 0.01	9.32 ± 0.33	−0.34 ± 0.33
SDSS J1511+4905	2792 ± 1086	45.10 ± 0.14	⋯	⋯	⋯	47.21 ± 0.14	9.14 ± 0.24	−0.04 ± 0.24
SDSS J1556+3517	4688 ± 227	45.17 ± 0.01	4153 ± 176	44.05 ± 0.03	46.16 ± 0.01	47.29 ± 0.01	9.66 ± 0.04	−0.48 ± 0.04
SDSS J0841+2005	6107 ± 181	45.70 ± 0.01	6225 ± 1019	44.93 ± 0.07	47.04 ± 0.01	47.82 ± 0.01	10.15 ± 0.02	−0.45 ± 0.02
SDSS J0943−0100	3804 ± 165	45.35 ± 0.01	3849 ± 3783	44.58 ± 0.12	46.27 ± 0.02	47.46 ± 0.01	9.55 ± 0.04	−0.20 ± 0.04
SDSS J1011+5155	5186 ± 296	45.02 ± 0.03	⋯	⋯	⋯	47.14 ± 0.03	9.68 ± 0.05	−0.65 ± 0.05
SDSS J1019+4108	3157 ± 308	45.08 ± 0.05	⋯	⋯	⋯	47.19 ± 0.05	9.25 ± 0.09	−0.16 ± 0.09
SDSS J1028+5110	3109 ± 334	45.38 ± 0.02	3562 ± 386	44.91 ± 0.03	46.61 ± 0.02	47.49 ± 0.02	9.38 ± 0.09	0.01 ± 0.09
SDSS J1132+0104	4656 ± 399	45.16 ± 0.03	⋯	⋯	⋯	47.28 ± 0.03	9.65 ± 0.08	−0.48 ± 0.08
SDSS J1134+3238	6242 ± 178	45.57 ± 0.01	4682 ± 516	44.63 ± 0.06	46.94 ± 0.01	47.68 ± 0.01	10.11 ± 0.03	−0.54 ± 0.03
SDSS J1516+0029	3826 ± 1571	45.25 ± 0.06	⋯	⋯	⋯	47.37 ± 0.06	9.51 ± 0.28	−0.25 ± 0.28
SDSS J1554+2218	3762 ± 353	45.34 ± 0.02	6300 ± 1080	44.49 ± 0.05	46.55 ± 0.01	47.46 ± 0.02	9.54 ± 0.07	−0.19 ± 0.07
SDSS J1709+6303	4551 ± 720	45.10 ± 0.06	8329 ± 1034	44.81 ± 0.06	46.60 ± 0.02	47.21 ± 0.06	9.59 ± 0.14	−0.49 ± 0.14

Note. ${L}_{\mathrm{bol}}$, ${M}_{\mathrm{BH}}$, and $\lambda$ have been derived from the broad Hα line as discussed in the text.

Download table as:  ASCIITypeset image

We note that for SDSS J0841+2005 Stern et al. (2017) recently published a K-band MOSFIRE/Keck spectrum with a higher S/N than the one we present here. Our measurements of redshift, Hα FWHM, and luminosity are fully consistent with the values obtained from the MOSFIRE spectrum.

We derive estimates of the black hole mass for our sample based on the virial method for single-epoch broad-line AGN spectra (e.g., McLure & Dunlop 2004; Greene & Ho 2005; Vestergaard & Peterson 2006). We have Hα measurements for all objects in our sample and also Hβ measurements for a major fraction (in particular for the z ∼ 1.5 sample). We will use black hole mass estimates based on the broad Hα line as our primary black hole mass estimator and use Hβ for comparison (see Section 3.3). While Hβ is most directly calibrated to reverberation mapping results (Vestergaard & Peterson 2006), Hα serves equally as a reliable black hole mass estimator (Greene & Ho 2005; Shen & Liu 2012; Mejía-Restrepo et al. 2016) and for our sample has the advantage of higher S/N and being less affected by reddening. We therefore argue and show below that Hα is indeed the preferred black hole mass estimator for our sample. At the luminosities of our samples the host galaxy contribution to L₅₁₀₀ is negligible (Shen et al. 2011).

As our reference relation we use the formula for Hβ by Vestergaard & Peterson (2006):

$$\begin{eqnarray}&&{M}_{\mathrm{BH}}({\rm{H}}\beta )={10}^{6.91}{\left(\displaystyle \frac{{L}_{5100}}{{10}^{44}\mathrm{erg}{{\rm{s}}}^{-1}}\right)}^{0.50}{\left(\displaystyle \frac{\mathrm{FWHM}}{3000\mathrm{km}{{\rm{s}}}^{-1}}\right)}^{2}{M}_{\odot }.\end{eqnarray} \tag{ 1 }$$

This relationship is directly derived from reverberation mapping studies. We use it to compute black hole masses from Hβ and calibrate our Hα black hole masses to this relation. For Hα we use the FWHM and luminosity of the broad Hα line to estimate the black hole mass as done in previous work (Greene & Ho 2005; Shen & Liu 2012; Jun et al. 2015). We use the relations between Hα and Hβ FWHM from Jun et al. (2015, their Equation (4)), derived from a compilation of studies covering a large luminosity range. For the scaling relation between L₅₁₀₀ and ${L}_{{\rm{H}}\alpha }$ we adopt the scaling presented in the same paper (their Equation (2)). This relation is consistent with Greene & Ho (2005) over the lower luminosity range covered in that study, but at the same time it provides a better fit to the high-luminosity regime, as studied in Shen & Liu (2012), Jun et al. (2015), and also in this work.

Combining these relations gives the following virial black hole mass estimator for Hα:

$$\begin{eqnarray}&&{M}_{\mathrm{BH}}({\rm{H}}\alpha )={10}^{6.711}{\left(\displaystyle \frac{{L}_{{\rm{H}}\alpha }}{{10}^{42}\mathrm{erg}{{\rm{s}}}^{-1}}\right)}^{0.48}{\left(\displaystyle \frac{\mathrm{FWHM}}{3000\mathrm{km}{{\rm{s}}}^{-1}}\right)}^{2.12}{M}_{\odot }.\end{eqnarray} \tag{ 2 }$$

To compute bolometric luminosities, we use the broad Hα luminosity ${L}_{{\rm{H}}\alpha }$ adopting a bolometric correction factor of 130 (Stern & Laor 2012), i.e., ${L}_{\mathrm{bol}}=130{L}_{{\rm{H}}\alpha }$. The Eddington ratio is then given by $\lambda ={L}_{\mathrm{bol}}/{L}_{\mathrm{Edd}}$, where ${L}_{\mathrm{Edd}}$ is the Eddington luminosity for the object, given its black hole mass.

Since LoBAL QSOs typically show high levels of dust extinction (including our sample as shown in Sections 4.2 and 4.3), this might also affect our measurement of ${L}_{{\rm{H}}\alpha }$ and thus ${M}_{\mathrm{BH}}$ and ${L}_{\mathrm{bol}}$. Assuming a typical value of $E(B-V)=0.14$ would indicate an underestimate of the flux at 6565 Å of 0.12 dex, assuming an SMC-like extinction curve. To test for a systematic bias in ${L}_{{\rm{H}}\alpha }$ in our sample, we use an alternative estimate of the intrinsic luminosity, based on the mid-IR (MIR) luminosity at $4.6\,\mu {\rm{m}}$, which for typical LoBAL $E(B-V)$ values are largely unaffected by dust extinction. This luminosity is obtained from the Wide-Field Infrared Survey Explorer (WISE) W2 magnitude (see Section 4.2 for details on the MIR data for our sample). We use the quasar SED template from Richards et al. (2006) to obtain the typical ratio to rest-frame 5100 Å and convert the W2 magnitude to L₅₁₀₀ (the typical ratio for our sample is ∼1). We then use the relation by Jun et al. (2015) to convert to ${L}_{{\rm{H}}\alpha }$. Comparing this intrinsic ${L}_{{\rm{H}}\alpha }$ estimate with our measurements for the z ∼ 1.5 sample, we find zero offset with a dispersion of 0.28 dex. The z ∼ 2.3 sample on average even shows larger measured ${L}_{{\rm{H}}\alpha }$ than the intrinsic estimate. We thus conclude that we do not see evidence for a systematic bias in ${L}_{{\rm{H}}\alpha }$ compared to the general quasar population due to dust extinction.

In Figure 2 we show the location of our two LoBAL samples (z ∼ 1.5 and z ∼ 2.3) in the black hole mass–luminosity plane and the black hole mass—Eddington ratio plane. We can already see that given the high luminosity limit of the respective samples they cover a broad range of ${M}_{\mathrm{BH}}$ and $\lambda$ with $8.7\lt \mathrm{log}{M}_{\mathrm{BH}}\lt 10.2$ and $-1.1\lt \mathrm{log}\lambda \lt 0.3$. As expected by our selection, the z ∼ 2.3 sample shows on average higher bolometric luminosities and also more massive black holes. We will discuss the black hole masses and Eddington ratios of these LoBAL QSOs in comparison to normal QSOs in more detail in Section 4.1.

Zoom In Zoom Out Reset image size

The primary goal of this paper is not to address the cross-calibration of different black hole mass estimators based on different broad lines. While this is a highly relevant topic that has drawn significant attention (McGill et al. 2008; Assef et al. 2011; Ho et al. 2012; Shen & Liu 2012; Trakhtenbrot & Netzer 2012; Matsuoka et al. 2013; Park et al. 2013; Jun et al. 2015; Mejía-Restrepo et al. 2016; Bisogni et al. 2017), our sample is by design not particularly suited for such studies, since the broad Mg ii and C iv lines are severely affected by the broad absorption troughs and therefore do not serve as a black hole mass estimator. This is, in general, not the case for the Balmer lines, apart from the case of SDSS J1019+0225, which shows unusual blueshifted absorption features in both Hα and Hβ (see Figure 9; for the emission-line fit of this object we have masked out this region).

We here investigate the consistency of the broad-line widths, luminosities, and black hole mass estimates between Hα and Hβ for our LoBAL quasar sample in comparison to the general quasar population. In Figure 3 we compare the FWHMs, ${L}_{{\rm{H}}\alpha }$ versus L₅₁₀₀, and the resulting black hole mass estimates. The correlation between Hα and Hβ FWHM from Jun et al. (2015), the relation between ${L}_{{\rm{H}}\alpha }$ and L₅₁₀₀ from Jun et al. (2015), and the one-to-one relation for ${M}_{\mathrm{BH}}$ are indicated by the solid black lines in each of the panels. In addition, we show the values for a representative non-BAL QSO sample of 60 objects at similar redshift and luminosity from Shen & Liu (2012) underlying as gray circles. They targeted normal luminous (L_bol> a few ×10⁴⁶ erg s⁻¹) quasars at redshift $1.5\lt z\lt 2.2$, with the majority of them at $1.5\lt z\lt 1.7$.

Zoom In Zoom Out Reset image size

We find a reasonably good correlation for the FWHM, with the z ∼ 1.5 sample showing a mean offset of −0.04 and a standard deviation of 0.10 dex. The z ∼ 2.3 sample shows a slightly larger scatter ($\sigma =0.14$ dex), due to the typical lower S/N in the Hβ line, but shows basically no mean offset (mean ${\rm{\Delta }}\mathrm{log}\,\mathrm{FWHM}=0.001$).

For the luminosity comparison we find the z ∼ 2.3 sample to be in good agreement with the relation for normal quasars. For the z ∼ 1.5 sample about half of the sample follows the relation for normal quasars by Jun et al. (2015), while the other half shows lower L₅₁₀₀ than expected. This lower L₅₁₀₀ is caused by increased intrinsic reddening in these objects. Indeed, we find these objects to show overall the strongest reddening based on the SED and the UV–optical spectral shape, as discussed further below and shown in Figures 9–12. We conclude that for our LoBAL sample ${L}_{{\rm{H}}\alpha }$ is the spectroscopic intrinsic luminosity indicator that is least affected by dust reddening and therefore our preferred estimator of ${L}_{\mathrm{bol}}$ and ${M}_{\mathrm{BH}}$.

The right panel in Figure 3 compares the derived black hole masses using Equations (1) and (2). We find a clear correlation between both mass estimates. For the z ∼ 1.5 sample both ${M}_{\mathrm{BH}}$ estimates are in good agreement. We find a mean difference of 0.18 and a standard deviation $\sigma =0.19$ dex. The offset toward less massive Hβ ${M}_{\mathrm{BH}}$ is at least partly due to the objects with lower L₅₁₀₀ discussed above. The z ∼ 2.3 sample again shows larger scatter due to the typical low S/N for Hβ in this sample, but the black hole masses are overall consistent (mean ${\rm{\Delta }}\mathrm{log}{M}_{\mathrm{BH}}=-0.01$ and $\sigma =0.30$ dex). This verifies the reliability of using Hα as a black hole mass estimator (e.g., Greene & Ho 2005; Ho et al. 2012; Mejía-Restrepo et al. 2016). We conclude that Hα remains a reliable black hole mass estimator also for the quasar subpopulation of LoBAL QSOs. We therefore use ${M}_{\mathrm{BH}}$ and $\lambda$ based on broad Hα for the rest of the paper as listed in Table 2.

A potential implication of the scenario that LoBAL QSOs represent a special evolutionary phase, corresponding to a young AGN, is that they might on average accrete at a higher rate than non-BAL QSOs, since they should still have ample fuel supply while just being in the process of blowing off their dusty envelope. Thus, we would expect to find on average higher Eddington ratios for our LoBAL quasar sample compared to a luminosity-matched non-BAL quasar sample. This is the hypothesis we test in this section.

To construct our non-BAL comparison sample, we follow two different approaches. The first uses the SDSS DR7 quasar catalog (Shen et al. 2011) to find a large sample matched directly to our LoBAL sample in redshift and either H-band magnitude (for the z ∼ 1.5 sample) or K-band magnitude (for the z ∼ 2.3 sample). We then use black hole masses derived from Mg ii (for the z ∼ 1.5 sample) or C iv (for the z ∼ 2.3 sample) to compare the observed black hole mass and Eddington ratio distributions to our LoBAL sample. This has the advantage of having a large comparison sample with the closest match to the LoBAL quasar sample, but at the risk of introducing potential systematics due to the use of different black hole mass estimators.

We therefore also use representative NIR spectroscopic samples of non-BAL quasars with measured properties of either broad Hα or Hβ that cover the same redshift and NIR magnitude range as our samples. For the z ∼ 1.5 sample we use the study by Shen & Liu (2012), restricted to $1.5\lt z\lt 1.8$, containing 55 quasars with Hα measurements. They obtained NIR observations of normal SDSS quasars selected for high-S/N optical spectra, which led to bolometric luminosities of their sample of $\mathrm{log}{L}_{\mathrm{bol}}\gt 46.4$, well matched to our LoBAL sample.

For the z ∼ 2.3 sample we construct a comparison sample based on the study by Coatman et al. (2017). They presented a large sample of 230 luminous quasars at redshift $1.5\lt z\lt 4.0$ with broad Hα and/or Hβ measurements from NIR spectroscopy. We restrict their sample to objects within $2.0\lt z\lt 2.6$ with broad Hα measurements and ${L}_{\mathrm{bol}}\gt {10}^{47}$ erg s⁻¹ to approximately match their sample to our LoBAL sample. The luminosity cut corresponds to the flux limit within our LoBAL sample. This selection results in a z ∼ 2.3 comparison sample of 70 non-BAL QSOs. For both the z ∼ 1.5 and z ∼ 2.3 comparison samples we estimated black hole masses and bolometric luminosities in the same way as for our LoBAL sample from the reported FWHMs and luminosities.

In addition, we augment our two redshift bins with a lower-z bin, again based on the BAL catalog from Allen et al. (2011). We select all objects with BI(Mg ii) $\gt \,0$ within the redshift range $0.4\lt z\lt 0.9$, allowing simultaneous coverage of Mg ii and Hβ in the optical SDSS spectra. We further require detection by 2MASS in J band and a black hole mass estimate based on Hβ in the SDSS DR7 quasar catalog from Shen et al. (2011). This gives a z ∼ 0.6 LoBAL quasar sample of 34 objects. We match it with a non-BAL quasar sample (BI(Mg ii) = 0), where we match them as close as possible in redshift and J-band magnitude with 20 non-BALs for every LoBAL.

We show the comparison between these samples in the ${M}_{\mathrm{BH}}$–$\lambda$ plane in Figure 4. We see no apparent difference in the ${M}_{\mathrm{BH}}$ or $\lambda$ distribution of LoBAL QSOs and non-BAL QSOs. To quantify this visual impression, we performed a two-sample Kolmogorov–Smirnov (K-S) test and an Anderson–Darling (A–D) test on the distributions of $\mathrm{log}{L}_{\mathrm{bol}}$, ${M}_{\mathrm{BH}}$, and $\lambda$ for each of the above five combinations of LoBAL sample and matched sample to test the null hypothesis that the LoBAL QSO and the matched non-BAL QSO sample are drawn from the same distribution. The results for the K-S test and the mean values of the distributions are given in Table 3. The A–D test provided consistent results. In no case do we find a statistically significant difference between any of these distributions, i.e., the distributions are statistically indistinguishable between the LoBAL and non-BAL QSO samples. In particular, we do not see clear evidence for a higher Eddington ratio in the LoBAL samples. The z ∼ 0.6 sample shows slightly higher mean $\lambda$ and lower ${M}_{\mathrm{BH}}$ by ∼0.1 dex for the LoBAL QSO sample, qualitatively consistent with Zhang et al. (2010), but not at a high significance for the sample we use here. More important, at $z\gt 1$ we cannot confirm such a trend. Our results are robust against the details of the matching and of the black hole mass estimator in use. Potential systematics between different lines for estimating ${M}_{\mathrm{BH}}$ would tend to increase the difference in the apparent distributions of ${M}_{\mathrm{BH}}$ and $\lambda$.

Zoom In Zoom Out Reset image size

Table 3.  Comparison of ${M}_{\mathrm{BH}}$ and $\lambda$ Distributions of LoBALs and Non-BALs

	$\mathrm{log}{M}_{\mathrm{BH}}$			$\mathrm{log}\lambda$			$\mathrm{log}{L}_{\mathrm{bol}}$
Sample	LoBAL	non-BAL	${p}_{\mathrm{KS}}$	LoBAL	non-BAL	${p}_{\mathrm{KS}}$	LoBAL	non-BAL	${p}_{\mathrm{KS}}$
z ∼ 0.6—SDSS	8.70 ± 0.08	8.83 ± 0.02	0.0233	−0.65 ± 0.06	−0.76 ± 0.02	0.224	46.15 ± 0.05	46.17 ± 0.01	0.535
z ∼ 1.5—SDSS	9.48 ± 0.11	9.50 ± 0.03	0.488	−0.54 ± 0.12	−0.51 ± 0.03	0.534	47.05 ± 0.08	47.09 ± 0.02	0.951
z ∼ 2.3—SDSS	9.64 ± 0.09	9.55 ± 0.06	0.761	−0.34 ± 0.06	−0.25 ± 0.05	0.558	47.41 ± 0.07	47.40 ± 0.04	0.925
z ∼ 1.5—NIR	9.48 ± 0.11	9.44 ± 0.04	0.838	−0.54 ± 0.12	−0.58 ± 0.04	0.604	47.05 ± 0.08	46.97 ± 0.04	0.125
z ∼ 2.3—NIR	9.64 ± 0.09	9.53 ± 0.05	0.466	−0.34 ± 0.06	−0.12 ± 0.03	0.0392	47.41 ± 0.07	47.52 ± 0.04	0.353

Note. We list the mean ${M}_{\mathrm{BH}}$, $\lambda$, and L_bol for the LoBAL sample and respective comparison sample, together with its uncertainty and the K-S test probability.

Download table as:  ASCIITypeset image

We conclude that we do not find evidence for LoBAL QSOs constituting a separate population in terms of their ${M}_{\mathrm{BH}}$ and $\lambda$. They are rather consistent with having the same black hole mass and Eddington ratio distributions as non-BAL QSOs.

We have also tested for any correlation of ${M}_{\mathrm{BH}}$, $\lambda$, and $\mathrm{log}{L}_{\mathrm{bol}}$ with the BAL properties, as measured by Allen et al. (2011). We do not find any significant correlation of these with their balnicity BI, mean BAL depth, or the minimum and maximum velocities, based on their Spearman rank-order correlation coefficients.

We next test whether our LoBAL sample shows any significant difference compared to the matched non-BAL sample in their SED. All of our targets posses multiwavelength photometry from the far-UV to the MIR. In particular, we collect data from four surveys that provide 14 bands in total. Optical data are taken from the SDSS DR7 quasar catalog (Schneider et al. 2010) in the $u,g,r,i$, and z bands. The NIR data in the J, H, and K bands come from the 2MASS catalog (Skrutskie et al. 2006), with the matching provided by Schneider et al. (2010). We augment this photometry with UV data obtained by the all-sky Galaxy Evolution Explorer (GALEX ) space mission (Martin et al. 2005). GALEX provides measurements in the far-UV (1350–1750 Å) and the near-UV (1750–2750 Å). Finally, we add MIR data from the all-sky WISE (Wright et al. 2010) mission, covering four bands at 3.4, 4.6, 12, and 22 μm, respectively. We take the WISE data for our SDSS QSOs from Lang et al. (2016), obtained from forced photometry of the WISE all-sky release imaging at SDSS positions. All of the photometric data have been corrected for Galactic extinction (Schlegel et al. 1998). We correct for missing data in a similar way to Richards et al. (2006). We use the AGN SED template by Richards et al. (2006) normalized at the neighboring band to derive the typical flux density in the missing band. For the LoBAL QSOs we add a typical LoBAL reddening to the SED template.

In Figure 5 we present the SED for each individual source from our LoBAL QSO sample, as well as their geometric mean SED in the three redshift bins studied above. We note that a detailed SED modeling for our sources is beyond the scope of this work, and we here highlight the average SED shape in comparison to the general quasar population. For this, we compare the LoBAL photometry with the SED obtained in the same way from the matched SDSS non-BAL QSO sample presented in Section 4.1 (black solid line) and with the AGN SED template from Richards et al. (2006) (magenta dashed line). We do not correct the SED for contributions other than the AGN continuum, like host galaxy contamination or the contribution by emission lines. With the matching of the LoBAL and the non-BAL samples these will contribute in a similar way to the SED, and we are here interested not in determining the intrinsic SED of LoBAL QSOs but only in the comparison to the general quasar population. The enhancement in the band around 6400 Å we see in the z ∼ 1.5 and z ∼ 2.3 samples in comparison to the Richards et al. (2006) SED template is likely due to the contribution from the Hα line (given its large equivalent width of ∼400 Å), which is by design of these samples centered on the respective band.

Zoom In Zoom Out Reset image size

Comparing the thus-constructed SEDs, the most obvious difference between our LoBAL QSOs and non-BAL QSOs is the significantly reduced flux at ${\lambda }_{\mathrm{rest}}\lesssim 3000$ Å leading to red colors, in agreement with previous work (Weymann et al. 1991; Reichard et al. 2003; Gibson et al. 2009). The redder color is usually interpreted as excess dust reddening in LoBAL quasars (e.g., Sprayberry & Foltz 1992).

Apart from the stronger reddening in the rest-frame UV, we do not see a clear difference between our LoBAL samples and the non-BAL QSOs. In particular, there is no apparent difference on the red side of the accretion disk emission at ${\lambda }_{\mathrm{rest}}\gtrsim 5000$ Å and in the dust torus emission in the NIR to MIR.

The MIR emission in QSOs originates from reprocessed UV–optical emission from the so-called torus, a cold, dusty obscuring medium distributed on spatial scales of $\gt 1$ pc. The evolutionary scenario for LoBAL QSOs implies a large dust covering fraction of the BAL wind, i.e., the BAL is visible along most orientation angles but only present in a small fraction of the quasar population. As pointed out by Gallagher et al. (2007), in this case it might be expected that BAL QSOs will have enhanced MIR emission due to the larger emitting volume of dust. We do not find evidence for such an enhancement and thus no support for the evolutionary scenario, consistent with previous work on HiBAL and LoBAL QSOs (Gallagher et al. 2007; Lazarova et al. 2012; but for a different result for radio-loud BAL QSOs, see DiPompeo et al. 2013).

In Figure 6 we show the WISE $W1-W2$ colors for our LoBAL QSO samples and the matched non-BAL QSO samples. Both populations are consistent with being drawn from the same population, based on a two-sample K-S test. All LoBALs in the two samples at $z\lt 2$ satisfy the WISE AGN selection criterion from Stern et al. (2012), $W1-W2\geqslant 0.8$. At $z\gt 2$ the criterion by Stern et al. (2012) is less complete, as also shown for our z ∼ 2.3 LoBAL sample. Wu et al. (2012) proposed a less strict criterion $W1-W2\geqslant 0.57$ to select $z\lt 3.2$ quasars. Most of our LoBALs, as well as those at $z\gt 2$, satisfy this criterion. While optical color selection is biased against LoBAL QSOs due to their red colors, WISE MIR selection is a promising technique for obtaining an unbiased census of the luminous LoBAL population at $z\lesssim 2$ and possibly beyond.

Zoom In Zoom Out Reset image size

We conclude that apart from the well-known higher reddening of LoBAL QSOs in the UV, their optical (${\lambda }_{\mathrm{rest}}\gtrsim 4000$ Å) to MIR SEDs are consistent with the general quasar population.

To investigate potential differences in the spectral continuum and emission-line properties between our LoBAL QSO sample and the general QSO population, we generate composite spectra for both populations. We generate individual composite spectra for the three LoBAL samples at z ∼ 0.6 (based on SDSS DR7 spectra), z ∼ 1.5, and z ∼ 2.3. As a non-BAL comparison sample for the z ∼ 0.6 sample we use the matched SDSS QSO sample discussed in Section 4.1, including 680 objects. For the z ∼ 1.5 and z ∼ 2.3 sample we use the sample from Shen & Liu (2012) with spectra made publicly available by Shen (2016), restricted to the same broad redshift bin, including 55 and 5 objects, respectively. The non-BAL samples cover the same luminosity range as their respective LoBAL QSO samples, so we are not affected by luminosity-dependent effects like the Baldwin effect in lines like [O iii] (Netzer et al. 2004; Stern & Laor 2013; Zhang et al. 2013) or the amount of host galaxy contribution for each z-bin comparison.

Each spectrum is shifted into rest frame, rebinned to a common wavelength scale, and normalized at 5100 Å for the z ∼ 0.6 sample and at 6350 Å otherwise, and a stacked spectrum is generated using the geometric mean. Uncertainties are derived from bootstrapping the sample where we applied the Monte Carlo approach discussed in Section 3.1 to every bootstrapped object. The derived composite spectra for the three LoBAL samples and non-BAL QSO comparison samples are shown in Figures 7 and 8. For the z ∼ 2.3 LoBAL sample we use BOSS spectra for the rest-frame UV coverage when available (in 9/10 cases), based on SDSS DR12 (Alam et al. 2015), applying the improved spectrophotometric calibration by Margala et al. (2016).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The most prominent continuum difference is again the significant reddening for the LoBALs in all three z bins at ${\lambda }_{\mathrm{rest}}\lesssim 4000$ Å, consistent with previous LoBAL composite spectra studies (Weymann et al. 1991; Brotherton et al. 2001; Reichard et al. 2003; Zhang et al. 2010). In addition, the strong broad absorption troughs are clearly visible. The z ∼ 2.3 LoBAL sample shows the most extreme BAL properties, as also seen in their individual spectra in Figures 11 and 12. Assuming SMC-like dust extinction, our data are consistent with a reddening of $E(B-V)\sim 0.14,0.17$, and 0.10 for the z ∼ 0.6, z ∼ 1.5, and z ∼ 2.3 samples, respectively.

The rest-frame optical properties between the LoBAL QSOs and the non-BAL QSOs are at first sight remarkably similar. We do not see any major differences in the broad Balmer line profiles. Boroson & Meyers (1992) reported an excess blue wing component in Hα for their small LoBAL QSO sample compared to their control sample. We cannot confirm this trend for our sample. The z ∼ 0.6 sample shows on average weak [O iii] and strong iron emission, consistent with previous work on low-z LoBAL QSOs (Weymann et al. 1991; Boroson & Meyers 1992; Zhang et al. 2010) and the small sample of LoBALs (mainly) located at $0.6\lt z\lt 1.2$ from Runnoe et al. (2013). For our two intermediate-z samples these trends are less clear. Neither the z ∼ 1.5 sample nor the z ∼ 2.3 sample shows significantly enhanced iron emission. The z ∼ 1.5 LoBAL composite may even have weaker optical Fe ii emission than the non-BAL composite. The z ∼ 2.3 sample shows on average weak or absent [O iii] emission; however, the [O iii] observations for this sample suffer from small number statistics and low S/N. Only one of six objects shows a clear strong [O iii] line. The z ∼ 1.5 sample shows [O iii] emission at least as strong as for the non-BAL comparison sample. While the composite spectrum derived from all 12 z ∼ 1.5 LoBAL QSOs even shows enhanced [O iii], this is mainly driven by the two objects with broad Balmer absorption lines and very strong [O iii] emission (Schulze et al. 2017, in preparation). When excluding these two from the stack, the [O iii] profile of the LoBAL QSO composite is fully consistent with the non-BAL QSO composite (see magenta line in Figure 8).

The [O iii] emission arises at larger distances from the nucleus in the narrow-line region (NLR) and is therefore largely an isotropic quantity. Significantly different [O iii] equivalent widths for LoBAL QSOs would be difficult to explain in a pure orientation scenario and rather support a large covering fraction of the BAL wind, which might shield the NLR from part of the ionizing radiation. Unfortunately, our results are inconclusive on this. While we see reduced [O iii] in the z ∼ 0.6 and z ∼ 2.3 samples, the [O iii] strength in the z ∼ 1.5 sample is consistent with the general quasar population.

Furthermore, the [O iii] profile in AGNs often shows a broad blue wing component indicative of outflows on NLR scales. Connecting the BAL wind originating on small scales and the large-scale ionized outflows traced via [O iii] can help us understand the outflow phenomena and the role AGN winds play for AGN feedback (e.g., Fiore et al. 2017). If LoBAL QSOs are young AGNs in the process of blowing off their dusty environment, an ubiquitous existence of powerful outflows might be expected. Indeed, our LoBAL sample shows several cases of broad [O iii] lines (FWHM $\gt \,1000$ km s⁻¹) and extended wings indicating the presence of powerful outflows in these objects. But while their demographics are still not well understood, signatures of outflows seem to be common in the general luminous quasar population, in particular at highz (Netzer et al. 2004; Harrison et al. 2014; Brusa et al. 2015; Carniani et al. 2015; Shen 2016; Bischetti et al. 2017). The composite spectra in Figures 7 and 8 again shed light on the prevalence of powerful outflows in LoBAL QSOs or otherwise different [O iii] outflow properties. In addition, we list the FWHM of the [O iii] line derived from a spectral fit to the composite spectra in Table 4. We find that the [O iii] profile in the LoBAL QSO composites is largely consistent with that of the non-BAL QSO composites of comparable luminosity. Thus, we do not see clear evidence of more powerful outflows traced via [O iii] in the LoBAL population. These results are fully consistent with an orientation interpretation of the LoBAL QSO phenomenon.