THE NATURE OF TRANSITION BLAZARS

Our data are taken from the SDSS-III (Gunn et al. 2006; Eisenstein et al. 2011) Data Release 10 (DR10; Ahn et al. 2014), which is publicly available and also includes all data from the SDSS-I/II portions of the survey. DR10 includes ∼4 × 10⁶ optical spectra over approximately 14,000 deg² of sky, including ∼1.6 × 10⁵ quasars (Pâris et al. 2014), selected primarily from the imaging portion of the survey in a variety of science programs (Ross et al. 2012). In our investigation, we include all optical spectra taken from the primary survey portion of SDSS-I/II on the SDSS spectrograph, as well as spectra from the BOSS spectrograph used in SDSS-III (spectral resolution R ∼ 2000). Due to differences between the spectrographs, fiber system, and spectral reduction pipeline between these phases of the SDSS survey (see Bolton et al. 2012; Smee et al. 2013), some small differences in absolute flux calibration may be present in spectra taken at different epochs (Dawson et al. 2013). These differences do not strongly affect our results, which are primarily based on spectral properties which are normalized to the continuum (e.g., EWs).

Occasionally, repeat spectra of the same objects in SDSS were taken, either due to low signal-to-noise ratios in the first epoch or as part of the survey plan. To create a sample of known blazars with repeat spectra in SDSS-III, we match all objects with repeat spectra available in DR10 to the 3149 blazars in the ROMA-BZCAT⁷ catalog of Massaro et al. (2009). This comparison results in a sample of 354 blazars with repeat observations, including a few with more than two epochs. Since the classical EW-based classification is based on whether any BELs in the observed optical spectrum are above or below 5 Å, we will compare the BEL EWs in pairs of repeat spectra of each blazar to identify cases where the BEL crossed this EW threshold. For blazars with more than two epochs of spectra, we will compare all unique pairs, leading to a total of 602 unique pairs of repeat spectra of the 354 blazars. Blazars are known to be highly variable even in intra-night observations, and the timescales on which blazars can transition between the FSRQ and BLL subclasses is not known; we thus place no cuts on the time lag between repeat observations for our sample, which range from intra-night to approximately 12.5 yr in the observed frame. The distribution of time lags for our sample is shown in Figure 1.

Zoom In Zoom Out Reset image size

We correct all repeat blazar spectra in our sample for Galactic extinction, using the maps of Schlegel et al. (1998) and the Milky Way reddening law of Cardelli et al. (1989). We also shift the repeat spectra of each blazar into the rest frame using redshifts provided by the ROMA-BZCAT catalog, aware of the fact that the redshifts of many blazars in this catalog are missing, uncertain, or occasionally incorrect. To ensure that these redshifts we use are not obviously incorrect, we visually inspect all 602 unique pairs of repeat blazar spectra in our sample. In the visual inspection, we mask pixels in each spectrum with SDSS flags set for NOPLUG, BADTRACE, BADFLAG, BADARC, MANYBADCOLUMNS, MANYREJECTED, NEARBADPIXEL, LOWFLAT, FULLREJECT, SCATTEREDLIGHT, NOSKY, BRIGHTSKY, COMBINEREJ, and REDMONSTER (for details on these flags, see Stoughton et al. 2002). Due to this spectrum quality mask, 16 of the 602 unique pairs of repeat blazar spectra in our sample were removed.

As expected, our visual inspection reveals that the overwhelming majority of blazars appear to have spectra that are either featureless (canonical BLLs) or have moderately strong BELs (canonical FSRQs). Some outliers include BLLs that possessed strong host-galaxy contributions (e.g., misaligned BLLs), blazars with intrinsic and line-of-sight absorption features, as well as a small handful in which the emission lines strongly varied in EW between the different spectral epochs. To quantitatively identify blazars that have transitioned from FSRQs to BLLs (or vice versa) in our sample, we measure the properties of the C iv λ1546, C iii] λ1906, Mg ii λ2800, and Hα λ6565 BELs (in vacuums wavelengths) where possible.

The details of our BEL fitting closely follow Shen et al. (2011, hereafter SH11) to allow for comparison of the BEL properties of our transition blazars to the general SDSS quasar population. The primary difference in our fitting in comparison to SH11 is that we opt not to fit templates for Fe emission since the continua in transition blazars are strongly dominated by jet emission, and thus the Fe line complexes are not well detected in their spectra. For C iv, Mg ii, and Hα, we fit a power law to the continuum wavelength windows used by SH11; the flux densities in the BEL wavelength windows (also the same as in SH11) are then normalized by the fitted continuum and integrated to find the EW. For C iii], which was not analyzed in SH11, we use continuum windows of 1810–1830 Å and 1976–1996 Å, and the BEL window is 1830–1976 Å. The choice of these wavelength windows for C iii] is informed by the SDSS composite quasar spectrum of Vanden Berk et al. (2001).

For each BEL, we fit both a broad component and a narrow component, each using a Gaussian with the same line center fixed to their lab vacuum wavelengths. However, the higher-ionization C iv line is well known to exhibit intrinsic blueshifts (Gaskell 1982; Richards et al. 2002, 2011), and so we allow the line center (assumed to be the same for both broad and narrow components) to be a single free parameter. The FWHM of the narrow component is constrained to be <1200 km s⁻¹, and the amplitudes of all components are constrained to be positive. For Hα, we fit additional Gaussians for the narrow [N ii] λλ6548, 6584 lines, as well as [S ii] λλ6717, 6731. We visually inspect the fit of each emission line and continuum region to ensure that it was not strongly affected by problematic pixels or absorption features. The statistical uncertainties on all quantities derived from the fitting (notably the EWs, luminosities, and FWHMs) are estimated by simulating the noise in each spectrum, by resampling each pixel from a Gaussian distribution with σ set to the uncertainty in the pixel flux density. All pixels in each spectrum are resampled in this fashion and refitted with the above procedure 10³ times; the uncertainties on the fitted quantities are then based on their 1σ range in the simulated spectra. Several studies have indicated a likely dependence of quasar BEL FWHMs on orientation (Wills & Browne 1986; Jarvis & McLure 2006; Fine et al. 2011; Runnoe et al. 2013), which may bias our FWHM measurements and black hole mass estimates due to the strong jet-alignment of blazars. However, this bias cannot be significantly larger than the ≲0.2 dex dispersion in the log-normal distribution of BEL FWHMs of quasars (Shen et al. 2008), and thus cannot account for the 1–2 dex difference in the Eddington ratio between our transition blazars (see Section 3.1) and that typical of radiatively inefficient accretion.

We identify six blazars for which any BEL in the spectrum crossed the 5 Å EW threshold in the rest-frame, transitioning between the canonical BLL and FSRQ classifications (or vice versa). The properties of these transition blazars are listed in Table 1 (and discussed in Section 3), and Figures 2–8 show their multi-epoch spectra, the fitting of the continuum-normalized BEL region in each of the epochs, as well as their long-term photometric light curves (discussed in Section 3.4). The observed changes in the EW in the six blazars across the 5 Å threshold vary wildly, up to a factor of >60 for SDSS J101603.13+051302.3. Due to the small sample size, we will not attempt to investigate the incidence of these blazar subclass transitions in our current study, and instead focus on understanding the nature of this phenomenon.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Since transition blazars intermittently show small BEL EWs by definition, redshifts can be difficult to obtain from spectra. Indeed, for two of these transition blazars as well as one additional blazar (SDSS J083223.22+491321.0) which did not transition, no previous redshifts have been reported. We report new secure redshifts from the SDSS pipeline using the spectral epochs where the BELs have the largest EWs (are thus most well detected). Although SDSS J083223.22+491321.0 did not transition, we nevertheless show its spectra and BEL fits in Figure 8 due to its newly determined redshift and its special implications for transition blazars (discussed in Section 5). We discuss the properties of each of these individual blazars in more detail in Section 4.

The homogeneous nature of SDSS spectra allows us not only to identify a sample of transition blazars with well-determined redshifts, but also to characterize robustly their emission line properties. Using the BEL fits described in Section 2.1, we calculate their intrinsic line luminosities over their corresponding BEL wavelength windows; these line luminosities are also listed in Table 1. Figure 9 (left panel) shows the BEL EW versus the fitted continuum luminosities in the BEL wavelength windows, for all epochs of the BELs of each transition blazar in temporal order. A clear anti-correlation is present across different BELs and in each of the different blazars for nearly all epochs. This anti-correlation strongly suggests that the EW variability causing the transition blazar phenomenon in our sample is primarily due to swamping of the emission lines by the jet continuum emission, which is highly variable due to Doppler boosting. However, the emission lines themselves can also have intrinsic variability in their luminosities, which could contribute to their observed EW changes. Figure 9 (right panel) shows the line luminosity versus the continuum luminosity of the same lines as the left panel. The lines are variable up to ≳0.5 dex in luminosity; in some instances (e.g., SDSS J125032.57+021632.1), the BEL luminosities appear to show an additional anti-correlation with the continuum luminosity (i.e., the lines become more luminous as the continuum luminosity decreases), further increasing the EW variability. Variations in luminosity of BELs of 0.6 dex have also been previously observed in the namesake BLL object (Corbett et al. 1996), which is a known transition blazar.

Zoom In Zoom Out Reset image size

We use the line luminosities to estimate the bolometric Eddington ratio (L_bol/L_Edd) as well as the Eddington ratio of the broad line emission (L_BLR/L_Edd). These two quantities which probe the accretion state are often used to investigate blazar subclasses in large samples (e.g., Ghisellini et al. 2011; Shaw et al. 2012), allowing us to place the accretion state of our transition blazars in a broader context. We calculate L_bol/L_Edd of the transition blazars in our sample following the method of Shaw et al. (2012), largely based on the results of SH11. We first estimate the unbeamed thermal continuum luminosity at 3000 Å using its correlation with the Mg ii line luminosity determined by SH11 from a large sample of quasars,

$$\begin{equation} \rm {log}\left(\frac{{L}_{\rm {cont}}}{\rm {erg\,s^{-1}}}\right) = {a}\cdot \rm {log}\left(\frac{{L}_{\rm {line}}}{erg\,s^{-1}}\right) + {b}, \end{equation} \tag{ 1 }$$

where a = 1.016 ± 0.003 and b = 1.22 ± 0.11 for the Mg ii BEL. We use this correlation to estimate the continuum luminosity rather than a direct measurement because the continuum emission in blazars is often dominated by non-thermal jet emission. The bolometric luminosity is estimated by multiplying the continuum luminosity at 3000 Å from Equation (1) by the bolometric correction of L_bol/L₃₀₀₀ = 5.15 from Richards et al. (2006). Using the luminosity and the fitted FWHMs of the Mg ii lines, we estimate black hole masses M_BH using the relation

$$\begin{equation} \rm {log}\left(\frac{{M}_{\rm {BH}}}{{M}_{\odot }}\right) = {c} + {d}\cdot \rm {log}\left(\frac{{L}_{\rm {line}}}{10^{44}erg\,s^{-1}}\right) + \rm {2\, log}\left(\frac{\rm {FWHM}}{\rm {km\,s^{-1}} }\right), \end{equation} \tag{ 2 }$$

where c = 1.70 ± 0.07 and d = 0.63 ± 0.00, also determined by SH11.

For the low-z blazar SDSS J083353.88+422401.8, Mg ii is not in the observed spectral wavelength range, and so the Eddington ratio must be calculated using other BELs. Hα is the only measurable BEL in this blazar, as Hβ is too weak to be discernible from the continuum, even in the low continuum luminosity epochs. We thus estimate M_BH using its relation with the Hα line luminosity and FWHM, determined by SH11 (their Equation (10)) and based on Greene & Ho (2005). Because SH11 did not measure correlations between continuum and line luminosities for Hα, we estimate the continuum luminosity L₅₁₀₀ by first estimating the FWHM of Hβ (not observable) using its correlation with the Hα FWHM determined in SH11. We can then estimate L₅₁₀₀ from its relation with the M_BH and the Hβ FWHM, using Equation (5) from Shaw et al. (2012) with their constants a = 0.672 and b = 0.61, determined by McLure & Dunlop (2004). Finally, the bolometric luminosity is calculated using L_bol/L₅₁₀₀ = 9.26 (Richards et al. 2006).

We estimate the M_BH and L_bol/L_Edd of each blazar using the spectral epochs for which the BELs have the largest EWs; these values are also listed in Table 1. All uncertainties on these derived quantities include the propagated statistical uncertainties from the spectral line fitting, as well as the reported statistical and systematic uncertainties of the constants in each of the scaling relations. We are also able to estimate L_bol and M_BH for SDSS J101603.13+051302.3 using C iv independently in addition to Mg ii; this calculation is performed using a = 0.863  ±  0.009 and b = 7.66  ±  0.41 in Equation (1) (SH11) where the line luminosity is for C iv and the continuum luminosity is for 1350 Å, and using a bolometric correction of L_bol/L₁₃₅₀ = 3.81 (Richards et al. 2006). The M_BH based on C iv is then calculated by setting c = 1.52 ± 0.22 and d = 0.46  ±  0.01 in Equation (2) (SH11). The M_BH and L_bol/L_Edd calculated independently from C iv and Mg ii in this object are consistent to within the uncertainty.

We also estimate the total luminosity of the broad line emission to estimate the Eddington ratio of the BELs (L_BLR/L_Edd). This calculation is performed in a manner similar to the method of Celotti et al. (1997), who found that the ratio of the total luminosity of all BELs to the line luminosities (L_BLR/L_line) for Hα, Mg ii, and C iv are 555.76/77, 555.76/34, and 555.76/63, respectively, based primarily on BEL flux ratios in the quasar composite spectrum of Francis et al. (1991). Using these factors, we calculate L_BLR for each of our transition blazars at the spectral epoch with largest EW, similar to that done for L_bol/L_Edd above. For SDSS J101603.13+051302.3, the L_BLR independently based on C iv is also highly consistent with Mg ii. Although a single measurement of L_bol/L_Edd is most helpful for classification in cases where multiple BELs are available (e.g., by averaging estimates derived from multiple lines), we opt to use the estimate from Mg ii as the fiducial L_bol/L_Edd for J101603.13+051302.3 due to its smaller scatter in the M_BH scaling relations we use in comparison to that of C iv, which instead serves as a helpful consistency check. Figure 10 presents the bolometric Eddington ratio of the transition blazars against the Eddington ratio of the BELs for the transition blazars in our sample, as well as for the blazar which did not transition. The Eddington ratios of these blazars are similar to the Eddington ratios of typical FSRQs (and much larger than those from radiatively inefficient accretion flows in BLLs). We discuss our interpretation of these results in Section 5.

Zoom In Zoom Out Reset image size

Blazars are well known to emit strongly at a wide range of wavelengths, and their SEDs are dominated by synchrotron emission at radio to soft X-ray wavelengths and inverse-Compton emission at hard X-ray to γ-ray wavelengths. The peaks of the synchrotron and inverse-Compton emission tend to occur at lower frequencies for FSRQs than for BLLs in general, although BLLs have often been further subdivided into low-, intermediate-, and high-synchrotron peaked BLLs, leading to the idea of a possible continuous "blazar sequence" of SEDs (Fossati et al. 1998; Ghisellini & Tavecchio 2008). We compile SEDs for the transition blazars in our sample using the ASI Science Data Center Virtual Observatory SED builder⁸ (Stratta et al. 2011), conveniently linked to all blazars in the ROMA-BZCAT catalog. The archival multi-wavelength data available are most complete at radio to optical wavelengths, and visual inspection of the SEDs of our sample of transition blazars shows that the synchrotron peak is well determined for all but one (SDSS J143758.67+300207.1), which has few data points in its SED.

We estimate the frequency and luminosity of the synchrotron peak of the transition blazars (except for SDSS J143758.67+300207.1) by fitting a third-order polynomial to the synchrotron emission in the SEDs. In the fitting, we have purposely excluded data points at frequencies >10¹⁷ Hz to avoid inverse-Compton emission. The estimated synchrotron peak frequencies and luminosities are listed in Table 2. These order-of-magnitude estimates are approximate since we are not fitting a physically motivated model to the SED and the archival SED data points are not contemporaneous. However, the SED peaks are reasonably well determined, and sufficient for classification of SED type. All the measured peak frequencies of these transition blazars are similar to the 10^{13 − 14} Hz range of low-synchrotron peaked blazars, and their luminosities are also consistent with the >10⁴³ erg s⁻¹ luminosities of blazars in this SED class. This result is in strong contrast to the ∼10^{14 − 16} Hz and ∼10^16–18 Hz synchrotron peak frequencies of intermediate- and high-synchrotron peaked blazars, respectively (Abdo et al. 2010a). Thus, the SED properties of these transition blazars (as well as those of SDSS J083223.22+491321.0) are consistent with FSRQs and low-synchrotron peaked BLLs.

Table 2. γ-Ray and Long-term Photometric Variability Properties of the Transition Blazars in Our Sample

SDSS	log10Lγa	αγb	log$_{10}\nu _{\rm peak}^{\rm c}$	log$_{10}L_{\rm peak}^{\rm d}$	log10SF∞e	log10τf
Object	(erg s−1)		(Hz)	(erg s−1)	(mag)	(days)
J083223.22+491321.0			13.12	45.51	−0.14	1.33
J083353.88+422401.8	45.22	2.33	14.11	44.80	−0.21	1.38
J101603.13+051302.3	47.99	2.08	13.27	45.94	0.31	0.79
J125032.57+021632.1			13.74	45.49	0.05	1.76
J130823.70+354637.0	46.89	2.28	12.88	45.89	−0.09	0.81
J143758.67+300207.1					0.13	0.93
J220643.28−003102.5	46.61	2.16	13.04	46.02	−0.06	0.78

Notes. ^a0.1–100 GeV luminosity from Nolan et al. (2012) ^b0.1–100 GeV photon number power-law index from Nolan et al. (2012) ^cEstimated frequency of the synchrotron peak in the SED. ^dEstimated luminosity of the synchrotron peak in the SED. ^eOptical photometric variability amplitude on long timescales. ^fCharacteristic timescale of optical photometric variability in the rest-frame.

Download table as:  ASCIITypeset image

The Large Area Telescope (Atwood et al. 2009) on board Fermi has surveyed the γ-ray sky to unprecedented depth, thus far detecting 1873 γ-ray sources in the two year 2FGL catalog (Nolan et al. 2012), including 1121 AGNs in the second Fermi AGN catalog (Ackermann et al. 2011). The overwhelming majority of Fermi-detected AGNs are blazars, approximately evenly divided between FSRQs and BLLs. Analysis of these Fermi blazars has revealed that BLLs tend to have γ-ray luminosities L_γ ≲ 10⁴⁶ erg s⁻¹ and γ-ray spectral indices α_γ ≲ 2.2 in the 0.1–100 GeV band, while FSRQs exceed these values. However, there is significant overlap in these γ-ray properties between the two subclasses (Ackermann et al. 2011), likely due to some combination of large measurement uncertainties and misclassification, in addition to astrophysical origins.

We match our transition blazars to the Fermi 2FGL catalog to investigate their γ-ray properties. Four of the blazars in our sample are matched to 2FGL sources that have association probabilities >95% (i.e., the blazar is highly likely to be the optical counterpart of the γ-ray source); their γ-ray luminosities and spectral indices are also listed in Table 2. None of the four transition blazars in our sample detected by Fermi can be clearly classified based on their γ-ray properties, and all lie in the area occupied by FSRQs and low-to-intermediate synchrotron peaked BLLs. Ghisellini et al. (2011) argued that for such objects, a better classifier to use is the Eddington ratio based on the BEL (L_BLR/L_Edd), with a dividing value of ∼5 × 10⁻⁴. We note that although the synchrotron peak frequencies and luminosities of these transition blazars (where determined) are similar, not all are detected in γ-rays.

Blazars are well known to be among the most variable extragalactic objects detected in large-scale optical time-domain surveys (Bauer et al. 2009). An investigation of long-term optical light curves recalibrated by Sesar et al. (2011) from the Lincoln Near-Earth Asteroid Research (LINEAR; Stokes et al. 2000) survey by Ruan et al. (2012) revealed that the jet synchrotron-dominated continuum emission in blazars is systematically more variable than the thermal disk-dominated continuum in typical Type 1 quasars. In particular, Ruan et al. (2012) found that blazars have typical optical variability amplitudes of ∼0.5 mag in the r-band on rest-frame timescales greater than ∼100 days; below this characteristic timescale, the optical variability is self-correlated and smaller in amplitude. The strong variability amplitudes and short variability timescales observed in comparison to typical quasars was interpreted by Ruan et al. (2012) as due to the effects of Doppler boosting in the relativistic jet of blazars.

We compile publicly available optical light curves of the transition blazars in our sample from the LINEAR and Catalina Sky Survey/Catalina Real-Time Transient Survey (Drake et al. 2009) databases to investigate their optical variability properties. The optical light curves constructed using these two surveys are also shown in Figures 2–8; they are well sampled, with 98–978 epochs (median of 444 epochs) spread over ∼11 yr in the observed frame. All transition blazars in our sample display strong optical variability with maximum amplitude in the range of ∼1.5–6 mag in the data.

To compare the optical variability properties of these transition blazars to the large sample of FSRQs and BLLs from Ruan et al. (2012), we fit each of these light curves to a first-order continuous autoregressive model (i.e., a damped random walk) using the method of Kozłowski et al. (2010), and discussed in detail in MacLeod et al. (2010). Figure 11 compares the rest-frame characteristic timescale of variability τ, and the variability amplitude on long timescales (when the variability is uncorrelated and strongest) SF_∞, of these seven blazars to the large sample of 60 bright blazars from Ruan et al. (2012). The seven blazars in our sample have larger variability amplitudes and shorter timescales of variability than typical blazars. A two-sample Kolmogorov–Smirnov test between the distributions of τ and SF_∞ in Figure 11 for our transition blazars and the blazar sample of Ruan et al. (2012) results in a p value of 3.1 × 10⁻³ for τ and 5.6 × 10⁻⁴ for SF_∞.

Zoom In Zoom Out Reset image size

Although the light curves we construct use data from both LINEAR and the Catalina surveys, the filters and calibrations are not exact between these two surveys; this issue will lead to a systematic offset between their measured magnitudes. However, both surveys used white-light filters and the systematic offset in magnitude is small (<0.1 mag), especially in comparison to the several-magnitude variability observed in our transition blazar light curves. Thus, any differences in calibration between the two surveys are dwarfed by the intrinsic variability of transition blazars and do not significantly affect our analysis.