SPECTROSCOPY OF BROAD-LINE BLAZARS FROM 1LAC

This paper reports on a multi-year observing campaign to follow-up the Fermi blazars. A principal aim is to achieve high-redshift completeness for the 1LAC sample (Abdo et al. 2010a).

In this paper, we discuss the spectra of FSRQs and other LAT blazar associations with strong emission lines. A major contribution is new spectroscopy of 165 of these blazars. To extend the analysis, we also measured archival spectra of 64 Sloan Digital Sky Survey (SDSS) blazars in the sample, for a total of 229 spectra.

This work takes the 1LAC high-latitude sample to 96% type completeness, with 316 FSRQs, 322 BL Lac objects, 33 other AGNs, 4 LINERs, and 4 galaxies. There are 30 remaining associated flat-spectrum radio sources of unknown type—generally these represent objects that are optically extremely faint (R > 23) or show faint continuum-dominated spectra, where current spectroscopy does not have sufficient signal-to-noise ratio (S/N) to unambiguously confirm a BL Lac type ID.

The most important sub-set of this emission-line sample are the objects with traditional FSRQ properties—in addition to the flat-spectrum radio core emission which allows the LAT counterpart association, we require emission lines with kinematic FWHM >1000 km s⁻¹ and bolometric luminosity >10⁴² erg s⁻¹. We find that 188 FSRQs meet these criteria, including 10 low-latitude sources with 1LAC FSRQ associations. In addition, some 11 BL Lac objects show well-detected broad lines. For this paper, we adopt the traditional heuristic BL Lac definition: continuum-dominated objects with observed frame line equivalent width (EW) of <5 Å and, where measured, Balmer break strength of <0.5 (Healey et al. 2008). We classify an object as "BL Lac" if it meets these spectroscopic criteria at any epoch. For 6 of the 11 BL Lac objects our spectra include epochs in a "low" state where decreased continuum reveals broad emission lines with >5 Å EW. The other five objects satisfy the BL Lac criteria in all of our spectra, but nevertheless show highly significant, albeit low EW, broad lines.

The emission-line sample contains 29 other objects—spectroscopically these are 9 galaxies, 5 LINERs, and 15 other AGNs. These show only strong narrow lines. While the line strengths rule out BL Lac IDs for these objects, they are manifestly different from our typical FSRQs. They may be misaligned radio galaxies, intrinsically weak AGNs, narrow-line Seyfert galaxies, and other less common types of γ-ray emitters (Abdo et al. 2010c).

Note that this sample of digital spectra does not include measurements for a number of bright, famous blazars, with historical spectroscopic classifications in the literature. Indeed, 137 of the 316 Fermi FSRQs fall into this category. Measured in R, the sources with archival spectra are brighter by 1.4 mag than our sample, so their omission may introduce systematic effects, discussed briefly in Section 5.3.

We have used medium and large telescopes in both hemispheres in a many-faceted assault on this sample. Observations were obtained from the Marcario Low Resolution Spectrograph (LRS) on the Hobby–Eberley Telescope (HET), the Large Cass Spectrometer (LCS) on the 2.7 m at McDonald Observatory, on the ESO Faint Object Spectrograph and Camera (EFOSC2; Buzzoni et al. 1984) and ESO Multi-Mode Instrument (EMMI; Dekker et al. 1986) at the New Technology Telescope at La Silla Observatory (NTT), on the Double Spectrograph (DBSP) on the 200'' Hale Telescope at Mt. Palomar, on the FOcal Reducer and low dispersion Spectrograph (FORS2; Appenzeller et al. 1998) on the Very Large Telescope (VLT) at Paranal Observatory, and on the Low Resolution Imaging Spectrograph (LRIS) at the W. M. Keck Observatory (WMKO). Observational configurations and objects observed are listed in Table 1.

The observing runs jointly targeted emission-line objects discussed in this paper and BL Lac objects to be discussed in a companion paper (M. S. Shaw et al., in preparation). All objects discussed in this work have highly significant emission lines and confirmed spectroscopic redshifts.

All spectra are taken at the parallactic angle, except for LRIS spectra using the atmospheric dispersion corrector, where we observed in a north–south configuration. In a few cases, we rotated the slit angle to minimize contamination from a nearby star. At least two exposures are taken of every target for cosmic-ray cleaning. Typical exposure times are 2 × 600 s.

With the variety of telescope configurations and varying observing conditions, the quality of the spectra are not uniform: resolutions vary from 4 to 15 Å, exposure times from 360 s to 2400 s, and telescope diameters from 2.7 m to 10 m. During this campaign, we generally used the minimum exposure required for a reliable redshift, rather than exposure to a uniform S/N. This may introduce selection effects into the sample, discussed briefly in Section 4.

Data reduction was performed with the IRAF package (Tody 1986; Valdes 1986) using standard techniques. Data were overscan subtracted and bias subtracted. Dome flats were taken at the beginning of each night, the spectral response was removed, and all data frames were flat-fielded. Wavelength calibration employed arc lamp spectra and was confirmed with checks of night sky lines. For these relatively faint objects, we employed an optimal extraction algorithm (Valdes 1992) to maximize the final S/N. For HET spectra, care was taken to use sky windows very near the long-slit target position so as to minimize spectroscopic residuals caused by fringing in the red, whose removal is precluded by the rapidly varying HET pupil. Spectra were visually cleaned of residual cosmic-ray contamination affecting only individual exposures.

We performed spectrophotometric calibration using standard stars from Oke (1990) and Bohlin (2007). In most cases standard exposures were available from the data night. Since the HET is queue scheduled, standards from subsequent nights were sometimes used. At all other telescopes, multiple standard stars were observed per night under varying atmospheric conditions and different air masses. The sensitivity function was interpolated between standard star observations when the solution was found to vary significantly with time.

For blue objects, broad-coverage spectrographs can suffer significant second-order contamination. In particular, the standard HET configuration using a Schott GG385 long-pass filter permitted second-order effects redward of 7700 Å. The effects on object spectra were small, but for blue WD spectrophotometric standards, second-order corrections were needed for accurate determination of the sensitivity function. This correction term was constructed following Szokoly et al. (2004). In addition, since BL Lac spectra are generally simple power laws, we used BL Lac objects observed during these runs (M. S. Shaw et al., in preparation) to monitor second-order contamination and residual errors in the sensitivity function. This resulted in excellent, stable response functions for the major data sets.

Spectra were corrected for atmospheric extinction using standard values. We followed Krisciunas et al. (1987) for WMKO LRIS spectra, and used the mean KPNO extinction table from IRAF for P200 DBSP spectra. Our NTT, VLT, and HET spectra do not extend into the UV and so suffer only minor atmospheric extinction. These spectra were also corrected using the KPNO extinction tables. We removed Galactic extinction using IRAF's de-reddening function and the Schlegel maps (Schlegel et al. 1998). We made no attempt to remove intrinsic reddening (i.e., from the host galaxy).

Telluric templates were generated from the standard star observations in each night, with separate templates for the oxygen and water line complexes. We corrected separately for the telluric absorptions of these two species. We found that most telluric features divided out well, with significant residuals only apparent in spectra with very high S/N. On the HET spectra, residual second-order contamination prevented complete removal of the strong water band redward of 9000 Å.

When we had multiple epochs of these final cleaned, flux-calibrated spectra with the same instrumental configuration, we checked for strong continuum variation. Spectra with comparable fluxes were then combined into a single best spectrum, with individual epochs weighted by S/N.

Due to variable slit losses and changing conditions between object and standard star exposures, we estimated that the accuracy of our absolute spectrophotometry is ∼30% (Healey et al. 2008), although the relative spectrophotometry is considerably better.

Redshifts were generally confirmed by multiple emission lines and derived by cross-correlation analysis using the rvsao package (Kurtz & Mink 1998). For a few objects only single emission lines were measured with high S/N. In general we could use the lack of otherwise expected features to identify the species and the redshift with high confidence. Nevertheless, single-line redshifts are marked (Tables 3 and 4) with a colon (:) and discussed in Section 2.4. Velocities are not corrected to heliocentric or LSR frames.

Table 3. Line Properties for z < 1 Broad-line 1LAC AGNs

Name				Hβ				Mg ii				Telescope
	z	$F_{\nu,\,10^{14.7}}$	α	L5100	LHβ	FWHM	Mass	L3000	$L_{\hbox{Mg\,{\sc {ii}}}}$	FWHM	Mass
	(10−28 erg cm−2 s−1 Hz−1)			(erg s−1)	(erg s−1)	(km s−1)	log (M/M☉)	(erg s−1)	(erg s−1)	(km s−1)	log (M/M☉)
J0004−4736	0.880	11.4 ± 0.0	−0.99 ± 0.08	...	...	...	...	45.339 ± 0.011	42.896 ± 0.151	2700 ± 500	7.85 ± 0.36	NTT
J0008+1450	0.045	73.8 ± 0.1	−1.79 ± 0.23	43.216 ± 0.002	40.921 ± 0.120	10800 ± 2600	8.19 ± 0.48	...	...	...	...	SDSS
J0017−0512	0.226	30.4 ± 0.1	−0.46 ± 0.36	44.353 ± 0.004	42.388 ± 0.107	2300 ± 500	7.55 ± 0.45	...	...	...	...	HET
J0024+0349	0.545	27.1 ± 0.0	−1.88 ± 0.08	...	...	...	...	45.110 ± 0.001	42.585 ± 0.057	3000 ± 200	7.76 ± 0.14	WMKO
J0043+3426	0.966	6.6 ± 0.0	−1.24 ± 0.03	...	...	...	...	45.201 ± 0.006	42.807 ± 0.072	3400 ± 300	8.01 ± 0.16	WMKO
J0050−0452	0.922	7.7 ± 0.1	−0.92 ± 0.19	...	...	...	...	45.198 ± 0.013	43.140 ± 0.141	3300 ± 600	8.20 ± 0.34	HET
J0102+4214	0.874	30.1 ± 0.1	−1.63 ± 0.86	46.001 ± 0.004	43.460 ± 0.038	1900 ± 200	7.92 ± 0.16	45.409 ± 0.007	43.593 ± 0.072	3300 ± 300	8.49 ± 0.17	WMKO
J0102+5824	0.644	167.0 ± 0.3	−1.25 ± 0.36	...	...	...	...	46.093 ± 0.076	43.449 ± 0.256	4100 ± 1200	8.57 ± 0.61	HET
J0112+2244	0.265	18.6 ± 0.2	−2.71 ± 0.79	...	...	...	...	...	...	...	...	HET
J0113+1324	0.685	15.2 ± 0.0	−0.01 ± 0.26	45.063 ± 0.006	43.094 ± 0.062	3800 ± 500	8.35 ± 0.25	45.263 ± 0.006	43.389 ± 0.082	4300 ± 400	8.57 ± 0.19	SDSS

Note. The J4+4 names represent radio positions of the objects, as found in Abdo et al. (2010a).

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

Table 4. Line Properties for z > 1 Broad -line 1LAC AGNs

Name				Mg ii				C iv				Telescope
	z	$F_{\nu,\,10^{14.7}}$	α	L3000	$L_{\hbox{Mg\,{\sc {ii}}}}$	FWHM	Mass	L1350	$L_{\hbox{C\,{\sc {iv}}}}$	FWHM	Mass
	(10−28 erg cm−2 s−1 Hz−1)			(erg s−1)	(erg s−1)	(km s−1)	log (M/M☉)	(erg s−1)	(erg s−1)	(km s−1)	log (M/M☉)
J0011+0057	1.493	5.2 ± 0.1	−0.51 ± 0.05	45.486 ± 0.021	43.541 ± 0.081	4900 ± 500	8.80 ± 0.19	45.800 ± 0.069	43.493 ± 0.435	2500 ± 1300	8.09 ± 1.02	SDSS
J0015+1700	1.709	8.0 ± 0.0	−2.75 ± 0.38	46.070 ± 0.004	44.178 ± 0.043	5900 ± 300	9.36 ± 0.10	45.818 ± 0.016	44.180 ± 0.068	5900 ± 500	9.15 ± 0.17	WMKO
J0023+4456	2.023 :	1.3 ± 0.1	−0.20 ± 0.50	...	...	...	...	45.213 ± 1.000	43.336 ± 0.875	1900 ± 2200	7.78 ± 2.29	HET
J0042+2320	1.426	5.1 ± 0.1	−1.40 ± 0.08	45.533 ± 0.017	43.412 ± 0.137	6900 ± 1100	9.01 ± 0.32	...	...	...	...	HET
J0044−8422	1.032	9.7 ± 0.0	−0.14 ± 0.03	45.401 ± 0.007	43.667 ± 0.113	3900 ± 500	8.68 ± 0.26	...	...	...	...	VLT
J0048+2235	1.161	11.5 ± 0.0	−1.34 ± 0.18	45.639 ± 0.002	43.143 ± 0.049	4300 ± 200	8.43 ± 0.11	45.916 ± 0.027	43.272 ± 0.115	3400 ± 500	8.25 ± 0.28	WMKO
J0058+3311	1.369	3.1 ± 0.1	−2.49 ± 0.37	45.331 ± 0.003	42.805 ± 0.054	3400 ± 200	8.01 ± 0.13	45.177 ± 0.035	43.451 ± 0.076	2200 ± 200	7.97 ± 0.19	WMKO
J0104−2416	1.747	8.8 ± 0.0	−0.81 ± 0.22	45.817 ± 0.010	43.613 ± 0.111	5000 ± 600	8.85 ± 0.26	45.989 ± 0.025	44.121 ± 0.292	4900 ± 1700	8.97 ± 0.69	NTT
J0157−4614	2.287	1.9 ± 0.0	−1.07 ± 0.33	45.632 ± 0.030	43.216 ± 0.171	2500 ± 500	7.98 ± 0.40	45.522 ± 1.000	44.083 ± 7.015	3000 ± 24900	8.52 ± 20.15	VLT
J0226+0937	2.605	16.9 ± 0.0	−0.58 ± 0.26	...	...	...	...	46.811 ± 0.006	44.582 ± 0.720	8500 ± 7500	9.65 ± 1.76	P200

Note. The J4+4 names represent radio positions of the objects, as found in Abdo et al. (2010a).

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

Reduced spectra of the newly measured objects are presented in Figure 1 (full figure available in the online journal).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View figure set (17 panels)

Tarball of figure set images

A few observations require individual comment.

J0023+4456. This spectrum had low S/N (∼2) and, while C iv, C iii, and Mg ii lines were tentatively identified, we mark the final solution as uncertain.

J0654+5042. This blazar, 05 from a bright star, was observed using DBSP with the slit along the blazar-star axis. We were able to isolate the stellar light by extraction from the wings of the convolved object. This spectrum was scaled and subtracted from the blazar spectrum to remove the stellar rest wavelength absorption features, giving a relatively clean blazar spectrum. While the emission line and redshift measurements are unambiguous, the continuum spectrophotometry should be treated with caution.

J0949+1752, J1001+2911, J1043+2408, and J1058+0133. These blazars show only a single broad emission line; good spectral coverage gives high confidence of identification as Mg ii.

J1330+5202. This blazar shows only strong [O ii] in emission at 3727 Å, with associated host Ca H/K absorptions. No broad emission lines are observed.

J1357+7643. Only one broad line is detected with high significance. A weaker corroborating feature implies C iv, Mg ii at z = 1.585. However, a z = 0.431 solution cannot be ruled out so we mark this solution as tentative.

J2250−2806. With a single strong narrow line (assigned to [O ii] 3727 Å), this blazar has an uncertain redshift. No broad emission lines are observed.

For some objects, multiple exposures at different epochs show significant spectral variability. In general, follow-up epochs have higher resolution or S/N or targeted the blazar in a lower flux state. In a few cases SDSS spectra were published after we had obtained independent data. We adopt the spectrum showing the highest S/N for broad-line detection as the primary spectrum used in this analysis. In general, this is the spectrum used to solve for the source redshift. As noted above, for particularly faint objects we can make S/N-weighted combinations of spectra (for which no strong flux variation is seen) to obtain the best primary spectrum. This is particularly valuable for the HET data, as limited track time requires short observations, but queue scheduling with very stable configurations allows easy combination of multiple epochs.

As noted above, in six cases, sources in this sample transitioned from a bright (continuum-dominated) state to a lower state. For these the "Primary" spectrum is from the low state, allowing better line measurements, during which the source is a nominal FSRQ (broad-line EW >5 Å). We nevertheless retain the original BL Lac typing (see Section 3.7). Clearly a more physical separation between the classes, and attention to duty cycles in the various "states" is needed.

We have fit power-law continua to all our spectra, using the scipy.optimize.leastsq routine⁴ based on Levenberg–Marquardt fitting to estimate parameter values and statistical errors. For these fits we first excise regions around the strongest emission lines in the blazar rest frame: λλ1200–1270, 1380–1420, 1520–1580, 1830–1970, 2700–2900, 3700–3750, 4070–4130, 4300–4380, 4800–5050, 6500–6800. We then fit simple power-law flux spectra to the entire spectral range, excluding regions at the blue or red end with uncharacteristically large noise. We do not here attempt to separate the thermal disk contribution. The results are presented as power-law indices, α, and continuum fluxes at 10^14.7 Hz (∼5980 Å), the center of our spectral range, as measured in the observer frame. These data may be combined with multi wavelength data to study the spectral energy distribution (SED) of the blazers in our sample.

The statistical errors on F_ν (Tables 3 and 4) are generally small, but should be convolved with the overall fluxing errors, estimated at 30%. We estimate errors on the spectral index by independently fitting the red and blue halves of the measured spectrum. We then sum the differences in quadrature for an estimated error:

$$\begin{equation} \alpha _{\rm err} = ((\alpha _{\rm red} - \alpha)^2 + (\alpha _{\rm blue} - \alpha)^2)^{1/2}. \end{equation} \tag{ 1 }$$

Note that large error bars generally indicate deviation from power-law continua rather than poor statistics. We find that extreme values of α seem to correlate with large Galactic A_V, suggesting errors in the assumed extinction and residual curvature in the corrected spectrum.

All line measurements are conducted in the object's rest frame. The local continuum and line parameters are measured with the scipy.optimize.leastsq routines. We first isolate the local continua surrounding the line, as in Section 3.2 (including an additional pseudocontinuum for Mg ii from the Fe ii/Fe iii line complex). Then, the line is fit to the continuum-subtracted data. Reported errors are purely statistical.

We follow Shen et al. (2011) in their line measurement techniques to facilitate comparison of our results with measurements in the extensive SDSS quasar catalog. We focus on lines used for BH mass estimates: Hβ at 4861 Å, Mg ii at 2800 Å, and C iv at 1550 Å.

Before reporting kinematic widths, we subtract in quadrature the observational resolution, to eliminate bias between objects measured by different telescopes, and to improve estimates of narrower lines. In higher redshift sources, emission lines are sometimes contaminated by intervening or associated absorption line systems. We visually check all emission lines.

We found strong absorption lines (associated and self-absorption systems) in a number of objects and edited these out before fitting the emission lines. Associated Mg ii absorbers were found in J0043+3426 (Δv ∼ −2400 km s⁻¹), J0252−2219 (Δv ∼ −3500 km s⁻¹ and Δv ∼ −2800 km s⁻¹), J1120+0704 (Δv ∼ −1600 km s⁻¹), J1639+4705 (Δv ∼ −600 km s⁻¹), and J2212+2355 (Δv ∼ −2400 km s⁻¹). In J0325+2224 we find self-absorbed C iv absorption. In J2321+3204, we find two systems (Δv ∼ −400 km s⁻¹ and Δv ∼ 100 km s⁻¹) and remove the absorptions from both Mg ii and C iv before fitting. Finally, in J2139−6732, we edit out an intervening (z = 0.923) Fe ii absorption system coincident with the C iv emission line. We next summarize particular issues for the three major species fit.

For Hβ, it is essential to include narrow components in the line fit. After power-law continuum removal, we simultaneously fit broad and narrow Hβ, narrow [O iii] 4959 Å, and narrow [O iii] 5007 Å (McLure & Dunlop 2004). We fix the rest wavelengths of narrow Hβ and the [O iii] lines at the laboratory values and fix widths at the spectral resolution, as measured from sky lines; the broad Hβ center and width are free to vary. All lines are modeled with single Gaussian profiles. The continuum is measured at 5100 Å.

For Mg ii, we fit the pseudocontinuum from broad Fe ii/Fe iii along with a power law. For the former, we use a template (Vestergaard & Wilkes 2001) convolved with the observational resolution. The power law is fit using the λλ2200–2700 and 2900–3100, portions of the spectrum; both continuum component fluxes and the power-law index vary freely.

After continuum subtraction, we fit the Mg ii line with broad and narrow Gaussian components. For this line, the continuum measurement is made at 3000 Å to minimize residual Fe contamination (McLure & Dunlop 2004).

Here, we fit the power-law continuum over 1445–1464 Å  and 1700–1705 Å. After subtraction, we fit the C iv line with three Gaussians, following Shen et al. (2011), and report the full FWHM of the line. Any narrow self-absorption components are visually identified and removed prior to the fit, as with the intervening absorbers described above.

Several blazars were classified as BL Lac objects in initial epoch observations. At the "primary" spectrum epoch, with low continuum, each was a nominal FSRQ. The objects which changed (and continuum decrease) were: J0058+3311 (8 ×), J0923+4125 (4 ×), J1001+2911 (6 ×), J1607+1551 (5 ×), J2031+1219 (4 ×), and J2244+4057 (10 ×).

With very high S/N observations, we were able to detect broad lines at high significance at EW levels <5 Å in several objects. These were thus "BL Lac objects" at all of our epochs, but can be analyzed along with the FSRQ. The BL Lac objects (and strongest broad-line EWs) were: J0430−2507 (Mg ii at EW = 0.9 Å), J0516−6207 (C iv at EW = 1.6 Å; C iii, Mg ii also present), J1058+0133 (Mg ii at EW = 2.2 Å), J2236+2828 (Mg ii at EW = 4.9 Å), and J2315−5018 (Mg ii at EW = 3.8 Å). These EW measurements are in observed frame. Clearly these sources are transitional between our standard FSRQs and BL Lac types.

For 53 of the 64 SDSS spectra in our sample Shen et al. (2011) tabulate standard measurements of the continuum and emission-line properties. This allows us to check the consistency of our measurements techniques.

We find that our measurements of continuum and line luminosities are consistent with those of Shen et al. (2011), although some scattered differences in line luminosity are seen, as the luminosity is sensitive to the fitting details. In particular, for Mg ii and Hβ, the variations in the narrow-line component flux can change the measured broad-line luminosity. Nevertheless, we see no systematic offset between our results and those of Shen et al. (2011).

However, we do note that our fitted kinematic FWHM values are systematically ∼10% larger (for the same sources) than those of Shen et al. (2011). For Mg ii and Hβ, we attribute this to small differences in the fit to the narrow component. For C iv, three Gaussians are used and we suspect that our central (narrowest) Gaussian is systematically weaker than that found in the SDSS fits. As we shall see, interesting differences from the SDSS quasars include narrower lines, on average, so we conservatively chose not to bias our fitter to force agreement. Nevertheless FWHM effects may be suspect at the 10% scale.

To place our Fermi sample in the context of the broader QSO population, we compare our measurements to those of the largely (>90%) radio-quiet SDSS Data Release 7 (DR7) quasar sample (Shen et al. 2011). We find significant offsets in the mean values of the continuum luminosities, line luminosities, and line FWHMs of the two classes as plotted in Figure 2. Since our sample extends fainter than the SDSS QSOs, we also plot the distributions for a magnitude cut i < 19.1 sample (Schneider et al. 2010).

Zoom In Zoom Out Reset image size

To quantify these differences, we compute the medians as well as the semi interquartile range (SIQR) for each distribution. We can use these values to estimate statistical errors on the median as $\sigma = {\rm SIQR} / \sqrt{n}$, where n is the number of points in the sample. In Figure 2, we report both the median offset, Δ, and its estimated significance, Δ/σ, for each parameter. The corresponding medians and errors for the magnitude-cut sample are given in parentheses.

Several distributions show large offsets from the SDSS population. Our C iv and Mg ii lines are significantly less luminous than those of the SDSS QSOs. Comparing the values for the flux-cut sample, we see that a significant difference persists. We conclude that we are sampling a population with intrinsically weaker lines or that fainter, weaker-line objects are lifted into our samples by the addition of extra non-thermal continuum flux. It seems that the local continua around Mg ii and C iv are similar for our objects; we do notice that our Hβ continua seem substantially brighter than those of the QSO population. We suspect that a lower redshift distribution and larger host-light contribution (see below) may explain this effect.

For all species (and for the sub-sample) we find that our measured lines are significantly narrower than those the SDSS average. We attribute this to preferential alignment, further discussed in Section 6.

In order to characterize the non-thermal emission in our spectra, we estimate a predicted continuum luminosity from the emission-line fluxes, scaling to a non-blazar sample. Shen et al. (2011) report line and continuum luminosities for the DR7 quasars. These are highly correlated (rxy = 0.85, 0.91, and 0.68 for Hβ, Mg ii, and C iv respectively), and they fit continuum luminosities to line luminosities for Mg ii and C iv, finding

$$\begin{equation} \log L_{3000} = 1.016 \pm 0.003 \cdot \log L_{\hbox{Mg\,{\sc {ii}}}} + 1.22 \pm 0.11 \end{equation} \tag{ 2 }$$

$$\begin{equation} \log L_{1350} = 0.863 \pm 0.009 \cdot \log L_{\hbox{C\,{\sc {iv}}}} + 7.66 \pm 0.41. \end{equation} \tag{ 3 }$$

We extend this treatment by fitting the SDSS QSO results for the Hβ line:

$$\begin{equation} \log L_{5100} = 0.802 \pm 0.049 \cdot \log L_{\hbox{\sc h}\beta } + 1.574 \pm 0.060. \end{equation} \tag{ 4 }$$

Since the line flux continuum predictions are calibrated to a non-blazar sample, they should be independent of the non-thermal jet activity. Of course for both samples, at low redshift, there may still be galaxy light contaminating the Hβ region, increasing the scatter. We compare the predicted to observed continuum levels for our Fermi FSRQs, finding that the lines predict (on average) 44%, 40%, and 32% of observed continuum luminosities for Hβ, Mg ii, and C iv respectively. We attribute the excess to non-thermal "jet" emission.

It is of interest to inspect the non-thermal dominance (NTD ≡ L_obs/L_pred) for individual objects in our sample. While most objects have NTD ∼ 1 (i.e., mostly thermal continua), for 57 blazars we have NTD > 2 and 18 spectra show NTD > 10 based on measurements of at least one line. We find significant correlation in same-object NTD measured from the three species (rxy = 0.51 for Mg ii and C iv; rxy = 0.73 for Hβ and Mg ii) Hβ generally measures larger NTD, which we attribute to contaminating galaxy light. Thus for 10% of our sample, the spectra are BL Lac-like and analysis of the optical properties must account for this dominant non-thermal emission.

In Figure 3, we report on the NTD of our sample. An optically selected population, such as that of Shen et al. (2011), peaks at log NTD = 0, the line-predicted thermal flux is comparable to the measured flux. Unsurprisingly the Fermi blazars extend to much larger NTD levels. This plays an important role in continuum-calibrated estimates of hole mass in Section 5.

Zoom In Zoom Out Reset image size

We further find a weak, but interesting correlation between NTD and Fermi γ-ray spectral index. The Fermi FSRQ population has 〈γ〉 = 2.44, whereas the BL Lac population has 〈γ〉 = 2.04. Thus the γ-ray hardness is a good predictor of the continuum strength in the broadband SED. Indeed, we find that the objects in our sample with the largest NTD show relatively hard Fermi spectra. It will be of interest to trace this trend deeper into the BL Lac population, although NTD estimates will be more difficult for this sample. Note that while actual BL Lac objects (circles), unsurprisingly, are mostly in the high NTD regime, a number of nominal FSRQs are found in this region, as well.

As in Shen et al. (2011), we estimate traditional virial BH masses from a relation of the form

$$\begin{equation} \log \left(\frac{M_{{\rm BH}}}{M_\odot }\right) = a + b \log \left(\lambda L_\lambda \right) + 2 \log \left(\hbox{FWHM}\right)\!, \end{equation} \tag{ 5 }$$

where a and b are calibrated from reverberation mapping for each line species (McLure & Dunlop 2004). The broad-line FWHM, in km s⁻¹, and the continuum luminosity λL_λ, in units of 10⁴⁴ erg s⁻¹, are measured as described in Sections 3.4–3.6. Values of a and b are tabulated in Table 2 from Vestergaard & Peterson (2006; hereafter VP06) for Hβ and C iv and Vestergaard & Osmer (2009; hereafter, VO09) for Mg ii. A scatter of ∼0.4 dex has been inferred for virial masses in optically selected samples (Shen et al. 2011).

For our Fermi sample we estimate mass from Hβ for 50 objects, from Mg ii for 176 objects, and from C iv for 68 objects. For 39 and 50 objects respectively, we measure mass from both Hβ and Mg ii, and from Mg ii and C iv. Masses are reported in Tables 3 and 4.

We urge caution in a naive interpretation of these masses, as applied to blazers. Due to the significant NTD of our sample, we expect continuum luminosity to not scale as in the reverberation mapping sample, as discussed below in Section 5.2. Further, preferential alignment and a non-spherical BLR yields systematically narrower lines, as discussed in Section 6.

Bearing in mind the substantial NTD of much of our Fermi sample, we wish to make mass estimates from the line luminosity and line kinematic width alone. Greene & Ho (2005) and Kong et al. (2006) calibrate such estimators from reverberation mapping samples. Shen et al. (2011) calibrate estimators for Mg ii and C iv from the SDSS sample. We have augmented this, using the SDSS mass estimators and developed our own coefficients for Hβ with values consistent with the predicted continuum luminosity and mass estimates of Section 5.1. The coefficients for these estimators are listed in Table 2 for a formula of the form of Equation (5), replacing λL_λ with line luminosity in units of 10⁴⁴ erg s⁻¹.

In Figure 4, we compare the line-estimated masses for our radio-γ-selected sample with estimates from the traditional continuum virial mass equations. We find that the line masses are smaller than the traditional virial masses by 0.14 dex on average. This is a small, but significant effect. More interesting is the comparison for different species and different epochs. For example the nearby, low-mass, low-luminosity Hβ mass sample shows a larger average line-mass decrease than the powerful C iv FSRQ. We attribute this primarily to host-light pollution in the former and thermal disk domination in the latter.

Zoom In Zoom Out Reset image size

For those objects where we have multiple epochs of observation (see Section 3.1 for details), we calculate the mass in both spectra. Due to the statistical uncertainty in measuring the line kinematic width, and the large role that plays in mass calculations, in all but three cases, the measurement in the multiple spectra are not significantly different from that in the primary spectrum. In the right panel of Figure 4, we also plot estimated mass changes from continuum luminosity fluctuations in spectra too poorly measured to estimate line properties. The general trend is for variations to extend above the line-determined values, which show smaller fluctuation (i.e., continuum variations dominate and bias mass estimates, as expected).

Using the less biased masses derived using line strengths, we compare in Figure 5 our radio-γ sample to the optically selected SDSS QSOs (Shen et al. 2011). Interestingly, the mean mass varies with redshift in parallel to the QSO sample, but with a lower normalization. For the full Fermi sample the offset is 0.44 dex. For the magnitude-cut sub-sample (directly comparable to the SDSS QSO), the offset is 0.34 dex.

Zoom In Zoom Out Reset image size

The origin of this offset is unclear. Certainly the NTD of our sample causes some intrinsically fainter, less massive blazars to be boosted into our flux cut sample. Further, our Fermi objects sample a preferred orientation (see Section 6). Another possibility is that these sources are on average in a more active accretion state than the typical QSO (see Section 5.4). We also allow that the brighter blazars with historical spectra in the literature may have systematically larger hole masses, decreasing the magnitude of this offset.

Following the unified model (Urry & Padovani 1995), we expect the quasi-isotropic thermal emissions of blazars to be similar to those in the underlying QSO population.

To probe this, we measure the disk luminosity in Eddington units, L_Edd, using line measurements alone. We first use the line fluxes to estimate local continuum luminosities, as above, then convert these to bolometric fluxes using factors from Richards et al. (2006) (L_bol/L₅₁₀₀ = 9.26, L_bol/L₃₀₀₀ = 5.15, L_bol/L₁₃₅₀ = 3.81). This provides a bolometric thermal (= accretion power) luminosity, independent of any non-thermal contribution. This luminosity may be compared with the line-estimated hole masses to derive the Eddington ratio L_bol/L_Edd for the population. Figure 6 displays this ratio against line-determined hole mass. Our sample has a systematically higher Eddington ratio and lower mass than the optically selected SDSS quasars. The radio-γ selection probes an intrinsically more active population than an optically selected sample, both in flux limited samples.

Zoom In Zoom Out Reset image size

Ghisellini et al. (2011) have proposed that FSRQs are more active accretors than BL Lac objects. While our high Eddington ratios for this FSRQ support this picture, we do not find any inverse correlation between Eddington ratio and NTD as might be expected if high values indicate a transition to a BL Lac type state.