The Sloan Digital Sky Survey Reverberation Mapping Project: the CIV Blueshift, Its Variability, and Its Dependence Upon Quasar Properties
Mouyuan Sun, Yongquan Xue, Gordon T. Richards, Jonathan R. Trump, Yue Shen, W. N. Brandt, D. P. Schneider
aa r X i v : . [ a s t r o - ph . GA ] J a n D RAFT VERSION J ANUARY
17, 2018Typeset using L A TEX twocolumn style in AASTeX61
THE SLOAN DIGITAL SKY SURVEY REVERBERATION MAPPING PROJECT: THE C IV BLUESHIFT, ITSVARIABILITY, AND ITS DEPENDENCE UPON QUASAR PROPERTIES M OUYUAN S UN ,
1, 2 Y ONGQUAN X UE ,
1, 2 G ORDON
T. R
ICHARDS , J ONATHAN
R. T
RUMP , Y UE S HEN ,
5, 6, ∗ W. N. B
RANDT ,
7, 8, 9
AND
D. P. S
CHNEIDER
7, 81
CAS Key Laboratory for Research in Galaxies and Cosmology, Department of Astronomy, University of Science and Technology of China, Hefei 230026, China;[email protected]; [email protected] School of Astronomy and Space Science, University of Science and Technology of China, Hefei 230026, China Department of Physics, Drexel University, 3141 Chestnut St., Philadelphia, PA 19104, USA Department of Physics, University of Connecticut, Storrs, CT 06269, USA Department of Astronomy, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA National Center for Supercomputing Applications, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA Department of Astronomy & Astrophysics, 525 Davey Lab, The Pennsylvania State University, University Park, PA 16802, USA Institute for Gravitation and the Cosmos, 525 Davey Lab, The Pennsylvania State University, University Park, PA 16802, USA Department of Physics, 104 Davey Lab, The Pennsylvania State University, University Park, PA 16802, USA (Revised
Draft: January 17, 2018)
ABSTRACTWe use the multi-epoch spectra of 362 quasars from the Sloan Digital Sky Survey Reverberation Mapping project to investigatethe dependence of the blueshift of C IV relative to Mg II on quasar properties. We confirm that high-blueshift sources tend to havelow C IV equivalent widths (EWs), and that the low-EW sources span a range of blueshift. Other high-ionization lines, such asHe II , also show similar blueshift properties. The ratio of the line width (measured as both the full-width at half maximum and thevelocity dispersion) of C IV to that of Mg II increases with blueshift. Quasar variability might enhance the connection betweenthe C IV blueshift and quasar properties (e.g., EW). The variability of the Mg II line center (i.e., the wavelength that bisects thecumulative line flux) increases with blueshift. In contrast, the C IV line center shows weaker variability at the extreme blueshifts.Quasars with the high-blueshift CIV lines tend to have less variable continuum emission, when controlling for EW, luminosity,and redshift. Our results support the scenario that high-blueshift sources tend to have large Eddington ratios. Keywords: black hole physics — galaxies: active — quasars: emission lines — quasars: general-surveys ∗ Alfred P. Sloan Research Fellow INTRODUCTIONBroad emission lines (hereafter BELs) from the “broadline region” (BLR) are unambiguous features in quasar spec-tra. These BELs can be divided into two main categoriesbased upon their ionization potential: high-ionization BELs(e.g., C IV , He II ) with ionization energy E ion &
50 eV ,and low-ionization BELs (e.g., H α , H β , Mg II ) with ioniza-tion energy E ion .
50 eV . Compared with low-ionizationBELs, high-ionization BELs are believed to be producedcloser to the central supermassive black holes (SMBHs).Interestingly, high-ionization BELs, such as C IV , are of-ten significantly shifted blueward with respect to their low-ionization counterparts (e.g., Gaskell 1982; Wilkes 1986;Corbin 1990; Sulentic et al. 2000a, 2007; Baskin & Laor2005; Richards et al. 2002, 2011; Wang et al. 2011; Denney2012; Shen et al. 2008; Shen & Liu 2012; Coatman et al.2016, 2017). The blueshift velocity can be as large as ∼ − (e.g., Luo et al. 2015). This result raises sev-eral important questions, including the origin of the blueshift,its effect on the estimation of the mass of the central SMBHs( M BH ), and its role in quasar unification.The blueshift is often attributed to accretion-disk winds(e.g., Gaskell 1982; Murray & Chiang 1997; Leighly 2004;Leighly et al. 2007; Richards et al. 2011; Denney 2012;Chajet & Hall 2013). Such winds can disturb the veloc-ity field of the BLR material and therefore can affect theline profiles. An alternative possibility is that the blueshiftis due to the scattering between the inflowing gases andthe BEL photons (Gaskell 2009). There are also specula-tions that the blueshift arises due to radiative transfer effects(Richards et al. 2002).Different physical drivers can be assessed using correla-tions of blueshift with quasar properties. For instance, itis likely that high-blueshift quasars favor distinct regionsof quasar parameter space or the quasar Eigenvector 1 se-quence (Boroson & Green 1992; Sulentic et al. 2000b, 2017;Dong et al. 2009; Runnoe et al. 2014; Shen & Ho 2014),e.g., high Eddington ratios λ Edd (Baskin & Laor 2005;Coatman et al. 2016), soft (i.e., with weak X-ray emission)spectral energy distributions (SEDs; see e.g., Leighly 2004;Richards et al. 2011; Luo et al. 2015), low inclination angles(Denney 2012) and/or low variability.In addition to the origin of the blueshift and its correla-tion with quasar properties, it is also important to considerthe effects of the blueshift on M BH estimation. As notedby Richards et al. (2011), the widely adopted single-epochblack-hole mass estimators (e.g., Vestergaard & Peterson2006; Vestergaard & Osmer 2009; Shen et al. 2011; for arecent review, see Shen 2013a) are derived using a low-blueshift reverberation-mapped quasar sample. There aretwo reasons why such estimators might not be valid for high-blueshift quasars. First, as mentioned before, the observedline profiles are likely due to a mixture of virial (i.e., dom-inated by the gravitational potential of the central SMBHs)and non-virial motions. Second, there are indications thatthe empirical BLR radius-quasar optical luminosity relation for H β depends on quasar SEDs (Kilerci Eser et al. 2015)and/or Eddington ratios (Du et al. 2016). If the high-blueshiftquasars indeed occupy a distinct region of quasar parameterspace, the current radius-luminosity relation could be invalidfor those quasars.The blueshift was not the first BEL feature showing sig-nificant changes across the quasar distribution. Ratherthat was the well-known Baldwin Effect (Baldwin 1977),which represents an anti-correlation between the C IV EW and luminosity (e.g., Dietrich et al. 2002; Wu et al.2009). Weak BEL quasars also tend to show large C IV blueshifts (e.g., Marziani et al. 1996, 2016; Richards et al.2011; Plotkin et al. 2015). Hence, there could be possi-ble connections between the C IV blueshift and EW. It isnow well-established that both the EW and blueshift areneeded to minimally characterize the range of properties ex-hibited by C IV (Sulentic et al. 2007; Richards et al. 2011;Marziani et al. 2016).In this work, we explore the high-ionization BEL blueshiftphenomenon taking advantage of the first epochs ofspectra from the Sloan Digital Sky Survey ReverberationMapping project (SDSS-RM; for a technical overview, seeShen et al. 2015). Compared with previous works, the SDSS-RM project provides a high S/N composite spectrum for eachof the quasars, allowing us to measure accurately theblueshift for both strong and weak emission lines. By ana-lyzing the spectra epoch by epoch, we can also measure thevariability properties of the blueshift. Finally, understandingthe blueshift properties of the SDSS-RM sample is also cru-cial for the project since one of its main goals is to provideunbiased M BH estimators for a wide variety of quasars.This paper is formatted as follows. In Section 2, we dis-cuss our spectral fitting procedures, and our measurements ofquasar properties. In Section 3 we show our analysis of thehigh S/N composite spectra. In Section 4, we present mea-surements of the variability of the blueshift. In Section 5,we discuss the implication of our results. A summary of ourwork appears in Section 6. We adopt a flat Λ CDM cosmol-ogy with h = 0 . and Ω M = 0 . . Throughout this work, thewavelengths of quasar features always refer to the rest-frame,unless otherwise specified. SPECTRAL MEASUREMENTThe SDSS-RM project is an ancillary program within theSDSS-III (Eisenstein et al. 2011) BOSS survey (Dawson et al.2013) using a dedicated . m telescope at Apache PointObservatory (Gunn et al. 2006). The spectrograph has awavelength range of – ˚ A with a spectral reso-lution of R ∼ (Smee et al. 2013). Each of the epochs has a typical exposure time of hours. The spec-tra were pipeline-processed (Bolton et al. 2012) and wereflux calibrated via a custom scheme (Shen et al. 2015). TheSDSS-RM sample consists of BEL quasars. We only se-lect . < z < . sources for which Mg II and C IV wereboth covered in the BOSS spectra. We focus on (three ofthe epochs are discarded due to low S/N) epochs of theSDSS-RM (Shen et al. 2015) spectra and the resulting high UN ET AL . 3S/N composite spectra. As mentioned in Shen et al. (2015)and Sun et al. (2015), there are spectra with flux anomalies. An epoch was identified as an outlier if its flux is more than magnitude away from the median of all epochs (Sun et al.2015). These outliers are rejected. Below we explain ourspectral fitting approach.2.1. Spectral fitting
Continuum fitting
Our spectral-fitting approach is similar to that of Trump et al.(2009) and Shen et al. (2011). For each spectrum, we firstfit a double power-law continuum (i.e., f λ = A λ β if λ < ˚ A ; f λ = A λ β if λ > ˚ A ) plus abroadened iron template (Vestergaard & Wilkes 2001) to thefollowing relatively emission line-free wavelength ranges, ˚ A < λ < ˚ A , ˚ A < λ < ˚ A , ˚ A < λ < ˚ A , and ˚ A < λ < ˚ A .During the continuum and the subsequent emission-line fit-ting, we rejected data points that are σ below the -pixelboxcar-smoothed spectrum. The purpose of this procedureis to reduce the effects of narrow absorption lines. We per-formed an iterative χ minimization to optimize the fits.The continuum and iron best fit is then subtracted from thespectrum. We then fit the resulting line spectrum with severalGaussian functions. In the following sections, we present ourmodeling procedures for Mg II and C IV (and He II λ Line fitting Mg II : We fit the continuum- and iron-subtracted flux inthe wavelength range of ˚ A < λ < ˚ A with threeGaussian functions, each with an unconstrained full-widthat half maximum (FWHM), i.e., we do not consider narrowMg II subtraction. We calculate the Mg II line-profile prop-erties from the overall line profile which is the summation ofthe multiple best-fit Gaussian functions. Any Gaussian func-tion with the ratio of its flux to the total line flux < . isignored.C IV : We adopted different line-modeling procedures forthe composite and single-epoch spectra.For the composite data, the pseudo-continuum subtractedspectrum from ˚ A < λ < ˚ A was modeled withsix Gaussian functions: two Gaussians for C IV , two Gaus-sians for He II λ , and the remaining two Gaussians forO III ] λ . Therefore, the ˚ A feature of the compositedata is modeled by the superposition of the red tail of C IV and a broad He II . This approach is similar to some previousstudies (e.g., Fine et al. 2010; Marziani et al. 2010). Similarto that of Mg II , we do not set limits on the FWHMs of theGaussian functions. As noted by Shen et al. (2015), such spectra might be obtained due tothe dropping of the fiber during spectroscopic exposures. We use kmpfit , a Python version of the least squares fitroutine, to perform our fitting. This routine is available asa part of the Kapteyn package, which can be downloaded from . For the single-epoch spectra, we only considered the fol-lowing wavelength range: ˚ A < λ < ˚ A . Theresulting pseudo-continuum subtracted spectrum was fittedwith two Gaussian functions. We do not model either He II or O III ] since the main purpose of the single-epoch spectralfitting is to constrain the variability of line properties. Themeasurement errors of He II or O III ] are relatively large sincethe two lines are weak. Therefore, it is nontrivial to constrainreliably the intrinsic variability of their line properties.Similar to Mg II , any Gaussian function with the ratio of itsflux to the total flux < . is removed from considerationwhen calculating the line-profile properties.In Figure 1, we present examples of our fits to the high S/Ncomposite spectra of RMID =660 and RMID=784.2.1.3. Uncertainty estimation
We adopted a Monte Carlo approach to estimate the un-certainties of the spectral-fitting parameters. A total of ( for single-epoch spectra) mock spectra were synthesized,where the flux in each wavelength pixel was generated byadding the flux density noise to the best-fit models. We thenfit the mock spectra following the same fitting recipe. Theuncertainties are estimated from the statistical dispersion ofthe best-fit models of the mock spectra. The statistical dis-persion was estimated by . x ) where IQR( x ) is theinterquartile range (IQR) of the variable x . The constant . normalizes the IQR to be equivalent to the standard deviationof a Gaussian distribution. Unlike the standard deviation, theIQR is robust against outliers or tails in the distribution.Following Shen et al. (2013b), we justified our uncertaintyestimation by exploring the distributions of quasar propertiesbetween close (i.e., rest-frame time interval < days) pairs.We then compared these distributions with the expected onesfrom the measurement errors. Our Monte Carlo approach un-derestimated the true uncertainties, so we enlarged the uncer-tainties by a constant factor until the expected distributionsfrom the measurement errors matched the observed close-pair distributions. The constant factor varies from ∼ . to ∼ . , depending on the physical quantities we are interestedin. In the following analyses, we will scale our uncertaintiesup by the constant factor.2.2. Emission-line properties
We calculated the following parameters of the emission-line properties. All line measurements are from the total lineprofile, which is the sum of the multiple Gaussians (exclud-ing Gaussian components that contribute less than of thetotal flux).1. The shift velocity, V shift ( V shift , se for the single-epochdata), is defined as c × ( λ h − λ va ) /λ va , where c , λ va ,and λ h are the speed of light, the central wavelength ofthe emission line in vacuum, and the line center. The RMID is the index of sources in the SDSS-RM catalog (see Table 1 ofShen et al. 2015). . . . . . . F l u x [ a r b i tr a r y un i t s ] RMID = 660
DataBest fit Pseudo continuumPower law Fe emisionCIV or MgII HeIIOIII] . . . . . RMID = 660 λ [ ˚ A] F l u x [ a r b i tr a r y un i t s ] RMID = 784 λ [ ˚ A] . . . . . RMID = 784
Figure 1.
Examples of multi-component fits to the composite spectra. The upper and lower panels are for RMID = and , respectively.The left and right panels are for C IV (with He II and O III ]) and Mg II , respectively. latter is defined as the wavelength that bisects the cu-mulative total line flux (Coatman et al. 2017). Figure 2presents an illustration of the definition of λ h .2. The offset of C IV for the coadded spectrum is, V off , CIV = V shift , CIV − V shift , MgII . That is, nega-tive values indicate blueshift toward the observer (i.e.,“outflows”). We define the single-epoch offset velocityas V off , se = V shift , CIV , se − V shift , MgII . The observedvariations of V off , se are due to the line-shift variabilityof C IV . The offsets of other lines are defined in a simi-lar way. Our definition of the offset of C IV might be anunderestimation of the true value because Mg II linesalso show offsets with respect to H β (Marziani et al.2013) or the host galaxy (e.g., Shen et al. 2016) by amedian blueshift velocity of
65 km s − with an intrin-sic scatter of ∼
200 km s − .3. The emission-line velocity width can be measured byFWHM or dispersion ( σ ) of the profile. Comparedwith FWHM, σ is more sensitive to the wing of theemission line. We adopted this definition because, for the redshift ranges consideredhere, Mg II is the best practical redshift estimator (Shen et al. 2016).
4. The emission-line shape is defined as D = FWHM /σ .For a perfect Gaussian profile, D = 2 . . Largervalues of D indicate that the profiles are more “boxy”.5. The equivalent width is calculated using EW = R λ va +200 ˚ A λ va − ˚ A f line ( λ )f cont ( λ ) d λ , where f line ( λ ) and f cont ( λ ) are both obtained from our spectral fitting results.We measured the ˚ A and the ˚ A continuum lu-minosities (hereafter L and L , respectively) fromthe best-fit double power-law component. We adopted the ˚ A continuum luminosity multiplied by a monochro-matic bolometric correction of (Richards et al. 2006) as anestimator of the bolometric luminosity, L Bol . We adopt the Mg II virial estimator to measure M BH (seeEq. (8) of Shen et al. 2011). The Eddington ratio is λ Edd = L Bol / (1 . × M BH /M ⊙ erg s − ) .2.3. Sample properties
We flagged our fits to the composite spectra by visual in-spection. The spectra of quasars for which reliable emission-line parameters could not be estimated are rejected. These Our results do not critically depend on L Bol . Therefore, our conclu-sions do not change if we instead adopt luminosity-dependent bolometriccorrection factors (e.g., Lusso et al. 2012; Krawczyk et al. 2013).
UN ET AL . 5 λ [ ˚ A] . . . . . . . . F l u x [ a r b i tr a r y un i t s ] λ h Figure 2.
An illustration of the definition of λ h . That is, the totalflux of the wavelengths shortward (i.e., λ < λ h ) equals to that of thewavelengths longward (i.e., λ > λ h ). .
50 1 .
75 2 .
00 2 .
25 2 . Redshift l og L B o l [ e r g s − ] Figure 3.
Distribution of our sample in the L Bol -redshift plane. Oursample consists of sources with . < z < . . The quasarluminosity range spans two orders of magnitude. spectra contain strong broad absorption lines, or have mul-tiple absorption features around the line centers of Mg II orC IV . Our final sample consists of sources. In Fig. 3, wepresent the distribution of L Bol as a function of redshift forthis sample. The quasar luminosity, M BH and λ Edd rangesspan two orders of magnitude, and therefore it is suitable toexplore the dependencies of the C IV blueshift upon quasarproperties. THE COMPOSITE SPECTRAWe are now in a position to explore V off , CIV as a functionof emission line properties.3.1.
The blueshift and EW
Figure 4 shows the distribution of our sample in the C IV EW-offset velocity parameter space. The distribution of theC IV offset velocity is not symmetric, with a long tail ofblueshifted velocities. For instance, ∼ of sources have V off , CIV >
550 km s − while ∼ of sources (highlightedas blue dots) have V off , CIV < −
550 km s − .We also binned the sources according to the C IV EW orthe offset velocity. Consistent with previous works (e.g.,Richards et al. 2011; Luo et al. 2015), the sources with ex-treme C IV blueshifts tend to have weak C IV . However, weakC IV is an insufficient condition for a quasar to have a strongC IV blueshift. We will further discuss the connection be-tween the C IV blueshift and EW in Section 5. Our sourcescan be divided into three subsamples that will be used insome of the subsequent analyses:1. Sample A: the “blueshift” sub-sample, i.e., sourceswith offset velocities < −
550 km s − and log EW < . (we selected this limit because of sources withoffset velocities < −
550 km s − satisfy this limit).There are sources in this sample.2. Sample B: sources with offset velocities > −
550 km s − (i.e.,weak or no blueshift) and log EW < . (i.e., weakC IV ). sources belong to this sample.3. Sample C: sources with offset velocities > −
550 km s − (i.e., weak or no blueshift) and log EW > . (i.e.,strong C IV ). This sample consists of sources.There are only sources in the high-blueshift and high-EWspace.We have constructed composite spectra of Mg II , He II andC IV for samples A, B, and C. The procedures to stack in-dividual spectra into a composite spectrum are as follows.First, we normalize each individual spectrum by its best-fitting ˚ A continuum flux. Second, for each wavelength,we take the median flux from the best-fitting line profiles ofthe normalized individual spectra. Third, we shift the wave-length to ensure that V shift , MgII = 0 , i.e., we adopt Mg II asthe redshift estimator. In Figure 5, we show the three com-posite spectra of Mg II , He II and C IV . Both C IV and He II show blueshift with respect to Mg II . The line shapes of C IV and He II also changes with the C IV blueshift.It is interesting that another high ionization line, He II , dis-plays similar properties. Indeed, the He II blueshift (with re-spect to Mg II ) and that of C IV are strongly correlated (Fig-ure 6). We adopted the Bayesian linear regression methodin Kelly (2007) to fit the data. The best-fit relation is y =(0 . ± . x − (583 ±
20) km s − with an intrinsic scat-ter of ±
16 km s − , where x and y correspond to theblueshifts of C IV and He II relative to Mg II , respectively. The blueshift velocity of He II is statistically larger than that Unlike Denney et al. (2016a) who tried only to include the narrow com-ponent, we considered the full He II line (i.e., the summation of broad andnarrow profiles). Therefore, it is not straightforward to compare our resultswith those of Denney et al. (2016a). − − CIV V off [km s − ] . . . . . . l og E W [ ˚ A ] A BCD
Figure 4.
Distribution of our sample in the C IV offset velocity-EWplane. Sources with offset velocity < −
550 km s − are highlightedby blue colors. The green (yellow) triangles represent the mean log EW (C IV V off ) in each C IV V off ( log EW ) bin. It is clearthat the high-blueshift quasars tend to have small EWs. However,the scatter of the correlation is not negligible. We defined threesamples A, B, and C according to the distribution (black dashedlines; see texts). Similar to the results of Richards et al. (2011),our sources tend to avoid the high-blueshift and high-EW (i.e., “D”)space. The black cross indicates the typical uncertainties of the C IV offset velocity and EW. λ [ ˚ A] . . . . F l u x [ a r b i tr a r y un i t s ] Sample A Sample B Sample C λ [ ˚ A] . . . Figure 5.
Composite spectra from C IV , He II and Mg II for thethree regions of the C IV offset velocity-EW plane (see Figure 4).The composite spectra are normalized to the best-fitting ˚ A continuum. We adopted Mg II as the redshift estimator. It is evidentthat both C IV and He II show blueshift with respect to Mg II . TheC IV shape parameter D of Sample A is larger than that of SampleB or C. Therefore, the C IV line profile of Sample A is more boxythan Sample B or C. of C IV . Meanwhile, the slope of the correlation is shallower However, this conclusion depends on the definition of the shift velocity.If we measure the shift velocity via the line peak and do not exclude thecomponents with the ratio of their flux to the total flux < . , the blueshiftvelocity of He II is statistically similar to that of C IV (Shen et al. 2016). − − − − CIV V off [km s − ] − − − − H e II V o ff [ k m s − ] Figure 6.
The He II offset as a function of the C IV offset (bothare relative to Mg II ). The red solid line and the shadowed regionrepresent the best-fit relation and its σ confidence band. The best-fit relation is y = (0 . ± . x − (583 ±
20) km s − , where x and y correspond to the blueshifts of C IV and He II relative toMg II , respectively. The intrinsic scatter is ±
16 km s − . − − − − V off [km s − ] . . . . . . . l og E W [ ˚ A ] HeIICIV
Figure 7.
Distribution of our sample in the offset velocity-EWplane. C IV and He II share similarly shaped two dimensional distri-butions. than the one-to-one relation. Hence, for sources with C IV V off < − − , the blueshift velocities of C IV arelarger than those of He II . The distribution of our sample inthe He II EW-offset parameter space is similar to that of C IV (Figure 7). Our correlation is unlikely to be due to possi-ble redshift biases because, similar to those of C IV , sourceswith extreme He II blueshifts also tend to have small EWs.Instead, these similarities suggest that the blueshifts of He II and C IV share the same physical origin, and the blueshiftmight be a common feature of all high-ionization emissionlines. UN ET AL . 7C IV EW correlates well with both Mg II EW and He II EW (left panels of Figure 8). As a result, both Mg II EWand He II EW are correlated with the C IV blueshift. Mean-while, the difference between C IV and Mg II EWs and theC IV blueshift are anti-correlated (the lower right panel ofFigure 8). Such anti-correlation is also obtained between H β and C IV (Sulentic et al. 2017). A similar anti-correlation isnot evident for C IV and He II (the upper right panel of Fig-ure 8). Compared with Mg II (whose ionization energy is ∼
10 eV ), the ionization energy of high-ionization BELs(e.g., C IV or He II ) is much higher (i.e., ∼
50 eV ). Theratio of the flux of a BEL to that of quasar continuum at theionization energy measures the covering factor of the BLRgas. If the ratio of the covering factor of Mg II to that of C IV (or He II ) is independent of the C IV blueshift, our results in-dicate that, with the presence of a large C IV blueshift, thehigh-energy ( E ∼
50 eV ) ionizing continuum is preferen-tially reduced with respect to the lower-energy ( E ∼
10 eV )one. 3.2.
The blueshift and line widths
Previous works (e.g., Sulentic et al. 2007; Shen & Liu2012; Runnoe et al. 2013; Coatman et al. 2016, 2017) oftenargue that, unlike low-ionization emission lines (e.g., H β ,Mg II ), C IV is a biased estimator of M BH . Notably, thereis an anti-correlation between FWHM
CIV / FWHM
MgII andthe C IV blueshift (see, e.g., Shen et al. 2008). We also usedour data to explore this anti-correlation since our coaddeddata do not suffer from short-timescale (i.e., rest-frame ∼ days) quasar variability. We confirm an anti-correlation be-tween FWHM
CIV / FWHM
MgII and the C IV blueshift (Fig-ure 9), indicating that MgII- and CIV-based M BH will beinconsistent. Similar relations are also observed between theratio of the line dispersion σ CIV /σ MgII and the C IV blueshift(Figure 10).These results suggest that, in the presence of the C IV blueshift, both FWHM
CIV and σ CIV are biased, and correc-tions are required. Unlike FWHM, the line dispersion σ ismore sensitive to the wings of the line profile. The correctionfor σ CIV is much smaller than that for
FWHM
CIV . There-fore, the core of the C IV profile is preferentially “broadened”as a function of the C IV blueshift. This result also indicatesthat σ is a more reliable estimator of the virial motions. Inpractice, we prefer to estimate M BH by making empiricalcorrections to FWHM
CIV (see also Coatman et al. 2017).The proposed approach has two advantages. First, FWHM isbetter constrained than σ in low S/N spectra. Second, for ex-treme blueshift sources, there is no clear correlation between σ and the blueshift.The slope of our anti-correlation between the ratio of FWHM
CIV to FWHM
MgII and the C IV blueshift is con-sistent with the correlation between C IV and H β found byCoatman et al. (2017). The intercept and the intrinsic scat-ter of our relation are larger than those of Coatman et al.(2017) by a factor of two. The differences could be causedby the imperfect one-to-one relation between FWHM
MgII and
FWHM H β (Trakhtenbrot & Netzer 2012) or the dif- ferences in how we estimate C IV blueshift compared withCoatman et al. (2017).The shape of C IV , D CIV (for its definition, see Sec-tion 2.2), is also expected to be anti-correlated with the C IV blueshift since the corrections for FWHM and σ are differ-ent. This speculation is confirmed by the Spearman rankcorrelation ( ρ = − . , and p = 0 . ; see Figure 11),albeit with a substantial scatter. In other words, the high-blueshift sources tend to avoid the small D CIV (i.e., moreboxy) space. Therefore, like the C IV blueshift, D CIV canalso be adopted as a viable and practical proxy to correct
FWHM
CIV (Denney 2012), although the correction may beinadequate at extreme C IV blueshift. There are two possibleexplanations for the anti-correlation. First, the spectra of thelow-blueshift sources have significant narrow C IV compo-nents. The contribution of the narrow C IV component to thetotal flux is smaller for the high-blueshift sources. If this sce-nario is correct, we would expect the shape of Mg II to showthe same anti-correlation. However, we found that, accordingto the Spearman rank correlation test, there is no significantcorrelation between the shape of Mg II and the C IV blueshift( ρ = − . and p = 0 . ; see Figure 11). Second, there isan intrinsic anti-correlation between the shape of C IV and itsblueshift. This anti-correlation seems to be inconsistent withthe scenario proposed by Gaskell (2009) where the line pro-files of high-blueshift sources are expected to be less boxy(i.e., have smaller values of D CIV ). VARIABILITY OF THE LINE SHIFTWe can also study the C IV blueshift for each of the epochs. The intrinsic variability of the blueshift of C IV canbe constrained by calculating the “excess of variance” (see,e.g., Sun et al. 2015), VAR . ( V shift , se ) = q (0 . V shift , se )) − f V (1)where IQR( V shift , se ) and V err are the − interquar-tile range and the uncertainty of V shift , se , respectively. f V represents the median value of the variable V .The dynamical timescale of the BLR is T dyn ∼ π Ω K = 253( R BLR R S ) / M BH × M ⊙ days (2)where Ω K , R BLR and R S are the Keplerian angular ve-locity, the radial distance of the BLR to the SMBH, andthe Schwarzschild radius, respectively. With timescaleswe consider here (i.e., . rest-frame days, which aremuch smaller than T dyn ), the variability is likely driven bythe quasar continuum variations. For instance, let us con-sider that the time lags between the ionizing continuum andthe BELs depend on the line-of-sight velocities (see, e.g.,Denney et al. 2009; Grier et al. 2013). The time lags of theblue and red wings will be different. If the time lag ofthe blue wing is shorter than the red wing, as the quasarcontinuum increases (decreases), we will observe a quickresponse of the blue wing, i.e., an apparent blue (red) shift log EW (CIV) [ ˚ A] . . . . . l og E W ( M g II ) [ ˚ A ] ρ = 0 . , p = 7 × − − − CIV V off [km s − ] − . − . . . . l og E W ( M g II / C I V ) ρ = − . , p = 7 × − . . . . . l og E W ( H e II ) [ ˚ A ] ρ = 0 . , p = 3 × − − − − . − . − . . l og E W ( H e II / C I V ) ρ = − . , p = 0 . Figure 8.
Upper-left: He II EW as a function of C IV EW. Upper-right: log EW(HeII) − log EW(CIV) as a function of the C IV blueshift.Lower panels: the same as the upper panels, but for Mg II . In each panel, the high-blueshift sources are highlighted as blue squares, and thecorrelation between the x -axis and y -axis variables is evaluated via the Spearman rank correlation (i.e., Spearman’s ρ ). It appears that, inpresence of a large C IV blueshift, C IV and He II are preferentially suppressed with respect to Mg II . − − − − CIV V off [km s − ] . . . . . . . . . F W H M C I V / F W H M M g II Figure 9.
Following Coatman et al. (2017), we fit
FWHM
CIV / FWHM
MgII as a function of the C IV blueshifts.The best-fit relation (via the Bayesian linear regression method;see Kelly 2007) is y = ( − . ± . x + (1 . ± . with anintrinsic scatter of . ± . , where x = V off / km s − . until the red wing responds to the increase (decrease) ata later epoch. Therefore, being either blueshifted or red-shifted, and the variations on short timescales are expected(hereafter the “line-of-sight velocity-dependent reverbera-tion”; see Barth et al. 2015). − − − − CIV V off [km s − ] . . . . . . . . σ C I V / σ M g II Figure 10.
The behavior of σ CIV /σ MgII as a function of the C IV blueshifts. The best-fit relation is y = ( − . ± . x + (1 . ± . with an intrinsic scatter of . ± . . For the extremeblueshift sources, there is no clear correlation between σ CIV /σ MgII and the C IV blueshift. This result, along with Figure 9, suggeststhat, for wind dominated sources, the core of the C IV profile is pref-erentially “broadened”. The line shift of Mg II To understand the line-shift variations due to reverberation,we first study Mg II . The upper panel of Figure 12 illus- UN ET AL . 9 − − − − CIV V off [km s − ] . . . . . . D = F W H M / σ CIVMgII
Figure 11.
The behavior of the line shape D as a function ofthe C IV blueshifts. For C IV , there is a weak anti-correlation be-tween D CIV and the C IV blueshifts ( ρ = − . and p = 0 . ).However, there is no anti-correlation between D MgII and the C IV blueshifts ( ρ = − . and p = 0 . ). trates the distribution of the variability of the line shift ofMg II (VAR.( V shift , MgII , se )) for our three subsamples. Notethat the uncertainty of VAR.( V shift , MgII , se ) is generally notsmall for each source. Hence, we focus only on the me-dian VAR.( V shift , MgII , se ) of each subsample. Meanwhile, wehave visually inspected those sources that show very largeVAR.( V shift , MgII , se ) (i.e., ≫
300 km s − ) and found that, inmany cases, their single-epoch spectra are noisy and the best-fitting results are not robust. Therefore, we rejected thesesources.To better control the effect of EW(C IV ), we created controlsamples matched in EW(C IV ), L , and redshift. Morespecifically, for each source in Sample A (i.e., the “high-blueshift, small-EW” sample), we randomly (with replace-ment) selected a quasar from the sources in Samples B orC with similar C IV EW (within . dex). We then adoptthe Anderson-Darling test to measure the probability thatthe randomly-selected sample is drawn from the same parentpopulation as sample A in terms of C IV EW, L , and red-shift. The two samples are consistent with being matched ifthe null-hypothesis probability p > . . We then calculatethe distributions of VAR.( V shift , MgII , se ) for sample A and thecontrol sample. We repeated this procedure times. Ourresults are presented in the lower panel of Figure 12. The me-dian VAR.( V shift , MgII , se ) of sample A is . ± . − ,whereas, the median VAR.( V shift , MgII , se ) of the control sam-ple is . ± . − . Of our realizations, the The Anderson-Darling test is found to be more sensitive at recogniz-ing the difference between two distributions than the popular Kolmogorov-Smirnov test (e.g., Hou et al. 2009). The uncertainty is the dispersion of the realizations. . . . F r a c t i o n High blueshift , small EW (sample A)Low blueshift , small EW (sample B)Low blueshift , large EW (sample C) VAR . ( V shift , MgII , se ) [km s − ] High blueshift , small EWControl sample Figure 12.
Upper panel: Distributions of the intrinsic variabil-ity of V shift , MgII , se for three different subsamples. Most sourcesshow weak variability of V shift , MgII , se . Lower panel: A comparisonbetween the distribution of the Mg II line shift variations of sam-ple A sources (i.e., the “high-blueshift, small-EW” sources) to thatof the “controlled sample” (matched in C IV EW, L , and red-shift). The median VAR.( V shift , MgII , se ) of extreme blueshift sourcesis larger than that of the control sample. possibility to have VAR.( V shift , MgII , se ) ≥ . − is lessthan . We thus conclude that VAR.( V shift , MgII , se ) of sam-ple A is statistically larger than that of the control sample.4.2. The line shift of C IV In this section, we check the variability properties of theC IV blueshift with respect to the coadded high S/N Mg II .Therefore, the variations presented in this section are onlydue to the line shift of C IV . As in Section 4.1, we focus onlyon the median VAR.( V shift , CIV , se ) , discarding sources withVAR.( V shift , CIV , se ) >
300 km s − .Figure 13 shows the results of the analysis of VAR.( V shift , CIV , se ).The median VAR.( V shift , CIV , se ) of Sample A (i.e, the“high blueshift, small EW” sample) is . ± . − .For the control sample, the median VAR.( V shift , CIV , se ) is . ± . − . Of our realizations, none hasVAR.( V shift , CIV , se ) ≤ . − . Therefore, contrary tothe results for Mg II , sample A sources have slightly weakerC IV line shift variability than that of the control sample.Meanwhile, the line shift of Mg II varies more strongly thanthat of C IV . Indeed, the Mann-Whitney U test of the line-shift variability on Mg II and C IV indicates that the variationamplitude of Mg II is statistically larger than that of C IV (the p -value is . ). These differences indicate that the struc-ture of C IV evolves as a function of the C IV blueshift in away that is different from that of Mg II . The Mann-Whitney test is a nonparametric test of the null hypothesisthat the distributions of two populations are equal. . . . . F r a c t i o n High blueshift , small EW (sample A)Low blueshift , small EW (sample B)Low blueshift , large EW (sample C) VAR . ( V shift , CIV , se ) [km s − ] High blueshift , small EWControl sample Figure 13.
Upper panel: Distributions of the intrinsic variability of V shift , CIV , se for three different subsamples. Similar to that of Mg II ,most sources show weak variability of V shift , CIV , se . Lower panel:A comparison between the distribution of the C IV line shift vari-ations of sample A sources (i.e., the “high blueshift, small EW”sources) and that of “controlled sample” (matched in C IV EW, L , and redshift). The high-blueshift sources have smaller me-dian VAR.( V shift , CIV , se ), opposite to what is observed for Mg II . In the high-redshift universe, broad Mg II or C IV linesare sometimes adopted to determine the quasar redshift. Asdiscussed by Denney et al. (2016a,b) and Shen et al. (2016),such redshift estimation is significantly biased, depending onquasar properties. The bias can be corrected to be betterthan ∼
200 km s − by some empirical guidelines (Shen et al.2016). However, our results (Figures 12 and 13) indicatethat quasar variability places lower limits on the accuracy ofmeasuring quasar redshifts with only single-epoch BELs. Forsome sources, the accuracy can be worse than
200 km s − . DISCUSSION5.1. C IV blueshift and quasar variability Our analyses in Section 4 demonstrate that the ob-served C IV blueshift can vary due to line-of-sight velocity-dependent reverberation. Therefore, quasar variability mighthave effects on the C IV EW-offset velocity connection.Let us first determine how much quasars can move in theC IV EW-offset velocity plane due to quasar variability. Todo so, we selected three sources, i.e., one from sample A,one from sample B, and one from sample C. The selectioncriteria are as follows. For each sample, we selected thesource with the highest ratio of the VAR.( V shift , CIV , se ) tothe median measurement error of C IV V shift , CIV , se such thatVAR.( V shift , CIV , se ) is the most robust one. Figure 14 presentsour results. For the three selected sources, ∼ of theirmotions along the C IV blueshift are due to measurement un-certainties. These sources do not rapidly change their posi-tions over the time period of our observations (see also Fig-ure 13). − − − − CIV V off [km/s] . . . . . . l og E W [ ˚ A ] Figure 14.
The locations of three sources in the C IV EW-offsetvelocity plane. The three sources are selected from sample A (bluesquares), sample B (orange triangles) and sample C (green stars),respectively. For each source, the symbol size increases with time.The grey color indicates the probability density distribution of ourwhole sample. Consistent with Figure 13, sources do not stronglychange their positions.
We then explored the possible correlation between V off , se and EW over the epochs for each source. Figure 15presents the distributions of the Spearman rank correlationcoefficient ρ . For most of our sources, the correlation is sta-tistically insignificant. This result is not totally unexpectedas the timescale of our multi-epoch data is short ( ∼ days)and quasar variability is weak on short timescales. However,on average, V off , se and EW are positively correlated over the epochs. We examined further the sources with statis-tically significant correlations and found that only four outof them have negative correlations ( / , / , and / forsamples A, B, and C, respectively). In Figures 16, 17, and18, we illustrate the single-epoch C IV EW as a function of V off , se for these sources. Most of their motions along theC IV EW are due to random fluctuations that are driven bythe measurement uncertainties.In Figures 19, 20, and 21, we show three examples ofthe variations of the C IV profile. As an aid to visual in-spection, for each source, we divided its observations intothree groups according to the increasing single-epoch C IV EW (i.e., the observations are sorted by the single-epochC IV EW; the st- th, th- th, and th- th re-orderedobservations belong to Groups , , and , respectively); foreach group, we created the variance-weighted high S/N meanspectrum; we fitted the mean spectra following the spectra-fitting approach mentioned in Section 2.1. When being plot-ted, all spectra and best fits of C IV profiles are normalized tothe best-fitting ˚ A continuum. Therefore, the intensitiesof C IV in the figures are proportional to their EWs.The positive correlation between C IV EW and offset ve-locity could be induced by some bias in our spectral fitting
UN ET AL . 11 − . − . − . − .
25 0 .
00 0 .
25 0 .
50 0 .
75 1 . ρ High blueshift , small EWLow blueshift , small EWLow blueshift , large EW Figure 15.
The distributions of the Spearman rank correlation co-efficient ρ between V off , se and EW over the epochs. There is,on average, a positive correlation between V off , se and EW, which isconsistent with the global C IV EW-offset connection. − − − − −
500 0 V off , se [km s − ] . . . . . . l og E W s e [ ˚ A ] Figure 16.
The single-epoch C IV EW as a function of V off , se forthe five sources in sample A (i.e., the “high-blueshift, small-EW”sample). For these sources, there are statistically significant positivecorrelations between the single-epoch C IV EW and V off , se . procedure. As mentioned in Section 2.1.2, we only fit C IV in [ ˚ A , ˚ A ] for single epoch spectra. It is possi-ble that out code interprets weak He II as the red wing ofC IV . As a result, we would expect an artificial positive cor-relation between C IV EW and offset velocity. We thereforeperformed a simple simulation to account for this bias. Weselected RMID= as an example. A total of mock We chose this source because the boundary between C IV and He II isnot visually evident in its composite spectrum. Therefore, the bias can belarge. We also tested some other sources and found similar results. − . . . l og E W s e [ ˚ A ] . . . .
250 1000 V off , se [km s − ] . . . l og E W s e [ ˚ A ] V off , se [km s − ] . . . . Figure 17.
The single-epoch C IV EW as a function of V off , se forthe sources in sample B (the “low-blueshift, small-EW” sam-ple). For these sources, there are statistically significant positive(eighteen sources) or negative (four sources) correlations betweenthe single-epoch C IV EW and V off , se . To avoid severe overlappingand confusion, the sources spread across four panels. Each ofthe upper and lower-left panels contains six sources; the lower-rightpanel contains four sources. V off , se [km s − ] . . . l og E W s e [ ˚ A ] V off , se [km s − ] . . Figure 18.
The single-epoch C IV EW as a function of V off , se for thenine sources in sample C (the “low-blueshift, large-EW” sample).For these sources, there are statistically significant positive corre-lations between the single-epoch C IV EW and V off , se . To avoidsevere overlapping and confusion, the nine sources spread acrosstwo panels. The left (right) panel contains five (four) sources. spectra were generated, where the flux in each wavelengthpixel was determined by adding the single-epoch flux den-sity noise to the best-fit model of the composite spectrum.We then fitted the mock spectra following the same fittingrecipe. We find that the variability in the mock spectra ismostly due to measurement errors and the variability ampli-tude is much less than the true single-epoch spectra. In addi-tion, for the mock spectra, the correlation between C IV EWand offset velocity is statistically insignificant. We thereforeconclude that the bias we mentioned cannot be responsiblefor the observed positive correlation.2 λ [ ˚ A] . . . . . . λ [ ˚ A] F l u x [ − e r g s − ˚ A − ] Group=1 Group=2 Group=3
Figure 19.
The time evolution of the C IV profile for a sample AQSO (RMID= ). The left and right panels represent the high S/Nmean spectrum in each group (for its definition, see texts for moredetails) and the best fits of C IV . The spectra and the best fits of C IV profiles are normalized to the best-fitting ˚ A continuum. In thisexample, groups with higher C IV EW tend to be less blueshifted. λ [ ˚ A] . . . . . λ [ ˚ A] F l u x [ − e r g s − ˚ A − ] Group=1 Group=2 Group=3
Figure 20.
The time evolution of the C IV profile for a sample BQSO (RMID= ). The left and right panels represent the high S/Nmean spectrum in each group (for its definition, see texts for moredetails) and the best fits of C IV . The spectra and the best fits of C IV profiles are normalized to the best-fitting ˚ A continuum. In thisexample, groups with higher C IV EW tend to be more blueshifted. λ [ ˚ A] λ [ ˚ A] F l u x [ − e r g s − ˚ A − ] Group=1 Group=2 Group=3
Figure 21.
The time evolution of the C IV profile for a sample CQSO (RMID= ). The left and right panels represent the high S/Nmean spectrum in each group (for its definition, see texts for moredetails) and the best fits of C IV . The spectra and the best fits of C IV profiles are normalized to the best-fitting ˚ A continuum. In thisexample, groups with higher C IV EW tend to be less blueshifted.
How do we understand these statistically significant pos-itive or negative correlations? As we mentioned in Sec-tion 4, the observed line-shift variations are likely drivenby the “line-of-sight velocity-dependent reverberation”.Meanwhile, as revealed by many multi-wavelength vari-ability studies, quasars tend to be bluer when they becomebrighter (e.g., Giveon et al. 1999; Vanden Berk et al. 2004;Guo & Gu 2016) and such a behavior is more prominent onshort timescales ( ∼ days, see Sun et al. 2014; Zhu et al.2016). Let us again image that the time lag of the blue partis shorter than that of the red one. As the quasar continuumincreases (decreases), the blue wing will respond faster thanthe red wing, which results in an apparent blueshift (red-shift); we will observe a harder (softer) quasar SED, i.e., alarger (smaller) EW. Therefore, these positive or negativecorrelations might be driven by the dependency of the BELtime lag on the line-of-sight velocity and the color variabilityof quasars.The overall positive correlation is consistent with theglobal C IV EW-offset velocity connection (i.e., sources withstrong C IV tend to be less blueshifted; see, e.g., Figure 4).As a result, we expect that quasar variability acts in such away to enhance the global C IV EW-offset velocity connec-tion. Compared with the single-epoch data, our high S/Ncomposite spectra do not suffer from short-timescale ( ∼ rest-frame days) variability. Therefore, we can assess theeffect of quasar variability by comparing the scatter of theC IV EW-offset velocity connection of the composite spectrawith that of the single-epoch data. We first added randomnoise to the C IV EW and V off (i.e., measurements fromthe composite spectra) such that their S/N are identical tothose of the C IV EW se and V off , se (i.e., measurements fromthe single-epoch spectra). The C IV velocity offset for eachepoch we adopted here is with respect to the composite Mg II profile (i.e., V off , se ). We adopted this definition to focus onthe variability of C IV alone. We then calculated Spearman’s ρ between the C IV EW-offset velocity for the S/N down-graded composite data; Figure 22 presents our results. Theexact value of ρ depends on epochs since the S/N of spec-tra changes with epochs. We also obtained the median ρ over the epochs. The differences between the correlationcoefficient of the S/N downgraded composite data and thatof the single-epoch data are due to quasar variability. Themedian ρ of the single-epoch spectra is slightly larger (by . ) than that of the S/N downgraded composite data. Ourresults indicate that quasar variability might slightly enhancethe connection between the C IV EW-offset velocity.5.2.
The physical origin of the C IV blueshift It remains unclear why sources with blueshift tend to havelow EW, but many other low EW quasars do not have ablueshift. We compare these two types of sources in termsof other quasar properties, especially the Eddington ratio,as there are suggestions that the C IV EW is tightly cor-related with the Eddington ratio (e.g., Bachev et al. 2004;Baskin & Laor 2004; Shemmer & Lieber 2015). To reduceeffects caused by other factors, for Sample A (sources with
UN ET AL . 13
Epochs . . . . . . . ρ Coadded spectraS / N downgraded coadded spectraSingle - epoch spectra Figure 22.
The Spearman rank correlation coefficient, ρ , be-tween the C IV EW and offset velocity for each epoch (small greensquares). The black dashed line corresponds to the measurementsof the high S/N composite spectra. Small red triangles are for themeasurements of the composite spectra with downgraded S/N (i.e.,matched in the S/N of each single-epoch measurements). The largered triangle and the large green square represent the mean corre-lation coefficient for the S/N downgraded composite data and thesingle-epoch data, respectively. The green squares are, on aver-age, above the red triangles, which indicates that the correlation isslightly tighter for the single-epoch data. That is, quasar variabilitycan enhance the connection between the C IV EW and offset veloc-ity. low EW and high blueshift), we again made control samplesmatched in the C IV EW, L , and redshift. Our results arepresented in Fig 23. While the median logarithmic Eddingtonratio of sample A is − . , the median logarithmic Eddingtonratio of the control sample is − . ± . . Therefore, aftercontrolling for the C IV EW, quasar luminosity, and redshift,the C IV blueshift sources tend to have significantly larger( ∼ . dex) Eddington ratios.There are two simple scenarios that can explain this re-sult. First, the observed correlation is simply an orientationeffect (Denney 2012). In this scenario, the high- and low-blueshift sources might intrinsically have similar Eddingtonratios. However, the high-blueshift sources are viewed moreface-on. When being viewed face on, the geometrically thinaccretion disk will be more luminous than the edge-on case.In addition, the face-on systems suffer less from absorption(due to, e.g., a torus) than the edge-on ones. For a polar wind,the line-of-sight blueshift velocity would be higher for theface-on case. As a result, the extreme blueshift sources ap-parently have larger Eddington ratios than the low blueshiftcounterparts. This scenario is, however, challenged by somerecent observations. For instance, Runnoe et al. (2014) mea-sured orientation for a quasar sample via the radio core dom-inance parameter; they found that there is no correlation be-tween the C IV blueshift and orientation. Second, the C IV − . − . − . . . log λ Edd
High blueshift , small EWControl sample Figure 23.
The distributions of the Eddington ratio, λ Edd . Filledblue bars represent sample A sources (the “high-blueshift, small-EW” sources). Open red bars with errors represent the controlsample (matched in C IV EW, L , and redshift). The extremeblueshift sources tend to have large Eddington ratios. blueshift sources are intrinsically more active, i.e., the high-blueshift sources have larger Eddington ratios than those ofthe low-blueshift sources. These two scenarios can be fur-ther tested by exploring the variability of the quasar contin-uum. The variability of the quasar optical/UV continuumis observed to be anti-correlated with the Eddington ratio(e.g., Ai et al. 2010; MacLeod et al. 2010; Kelly et al. 2013;Kozłowski 2016; Rumbaugh et al. 2017), after controllingfor quasar luminosity, and redshift. Therefore, according tothe first scenario, the high- and low-blueshift sources sharesimilar variations of the quasar continuum; however, thevariability amplitude of the quasar continuum for the high-blueshift sources would be smaller than the low-blueshiftones in the second scenario.We then calculated the r -band intrinsic variability forsample A and the corresponding control sample. We firstcalculated the synthetic flux in the r band by convolving the r -band bandpass with the spectra. As a second step, we cal-culated the structure function (we adopted the IQR estimator;see e.g., Sun et al. 2015) for sample A and the control sam-ple. Our results are presented in Figure 24. It is clear thatsources in sample A are intrinsically less variable. There-fore, our results disfavor the orientation scenario, but supportthe idea that high-blueshift sources often be more active (i.e.,have higher Eddington ratios). Luo et al. (2015) exploredthe X-ray properties of PHL 1811 analogs and weak-linequasars that also show evident C IV blueshifts. They foundthat these sources tend to suffer from significant (intrinsic) We chose this band because the corresponding wavelength is around ∼ ˚ A , which has the smallest spectrophotometric uncertainty (Sun et al.2015). Δt [days] r b a n dS t r u c t u r e F un c t i o n [ m a g ] High blueshiftΔ small EWControl sample
Figure 24.
The r -band variability amplitude as a function of therest-frame time interval, ∆ t . Sample A sources (the “high-blueshift,small-EW” sources) are less variable, indicating higher Eddingtonratios. As reported by Sun et al. (2015), the variability estimation ontimescales < days are biased; we therefore only consider quasarvariability on longer timescales. X-ray absorption. They proposed that PHL 1811 analogs andweak-line quasars can be well explained if these sources havevery large Eddington ratios, which appears in line with ourEddington-ratio scenario.As the accretion rate increases, the temperature of the ac-cretion disk increases, which produces more UV photons.Meanwhile, as revealed by recent radiation magnetohydro-dynamic simulations (e.g., Jiang et al. 2014a), the energydissipation efficiency of the X-ray corona decreases withthe accretion rate. The X-ray corona will be more effi-ciently cooled due to inverse Compton scattering of theseUV photons, i.e., the SED becomes softer (i.e., havinglarger α ox ) with the increasing Eddington ratio. In addi-tion, the inner accretion disk is puffed up at high Eddingtonratios (e.g., λ Edd & . ) due to radiation pressure (e.g.,Abramowicz et al. 1988; Wang & Netzer 2003; Jiang et al.2014b; Sa¸dowski et al. 2014). The puffed-up disk could actas a “shielding” gas (e.g., Leighly 2004; Wu et al. 2011;Luo et al. 2015) that blocks both the X-ray coronal emissionand the ionizing continuum, i.e., the SED is expected to besofter. Quasars with such softer SEDs (i.e., weaker X-rayemission) can launch strong winds from the accretion disk(e.g., Murray & Chiang 1997; Leighly 2004; Richards et al.2011; Chajet & Hall 2013, 2017; Luo et al. 2014).The Eddington-ratio scenario has important implicationsfor RM. According to the SED-evolution picture, the radius-optical luminosity relation of the low-blueshift sources is in-valid for the high-blueshift ones; in those cases, the radiusand M BH will be overestimated. Therefore, for the high-blueshift sources, the Eddington ratios we measured (for themethodology, see Section 2.2) might be lower limits on thetrue values. A direct test can be applied to our scenarioby performing RM campaigns (e.g., SDSS-RM; Shen et al. 2015) for the extreme blueshift sources, and exploring theradius-optical luminosity relation as a function of the C IV blueshift (also see Richards et al. 2011).5.3. The connection between the C IV blueshift and quasarproperties This Eddington-ratio scenario can also explain additionalobservational results in this work. According to our scenario,when the Eddington ratio increases, the SED becomes softerand the covering factor of the shielding gas to the BLR in-creases (Luo et al. 2015). Hence, we expect the BELs to beweaker. C IV and He II (or other high-ionization lines) wouldbe preferentially reduced with respect to Mg II (or other low-ionization lines) as the ionization energy of the former islarger. Therefore, the Eddington-ratio scenario can plausiblyexplain our results in Figure 8.How do we explain the fact that FWHM and σ are bothanti-correlated with the blueshifts (Figures 9 & 10)? Thisrelation is apparently inconsistent with previous resultsfrom Denney (2012), who found that the rms spectra ofC IV (which are assumed to represent the variable emis-sions) are broader than the mean spectra, indicating that thenon-variable blueshift component should be narrower thanthe “canonical” (or disk) C IV profile. As pointed out byBarth et al. (2015), rms spectra only evaluate the relativevariability amplitude as a function of the line-of-sight veloc-ity. It is possible to produce very broad rms spectra (broaderthan the single-epoch disk profile) if the high line-of-sightvelocity gas responds more efficiently than that of the lowline-of-sight velocity gas. However, as high velocity compo-nents are generally produced in the high-ionization region, itis unlikely that this emission is more sensitive to the contin-uum variations (Korista & Goad 2004). The Eddington-ratioscenario could provide a plausible explanation for why theratio of the line width (measured as both FWHM and σ ) ofC IV to that of Mg II increases with the C IV blueshift. Theradius of the BLR gas should scale as L . , where L ion isthe ionizing continuum luminosity. The ratio of the C IV ra-dius to that of Mg II decreases with the increasing Eddingtonratio. Therefore, FWHM
CIV / FWHM
MgII and σ CIV /σ MgII are expected to be correlated with the C IV blueshift. Theaccretion-disk winds may produce singly-peaked (which isdue to the radiative-transfer effects, e.g., the escape proba-bility is anisotropic; see Murray & Chiang 1997) and boxy(i.e., large values of D ) C IV profiles since they are generatedin the inner high-speed regions. Therefore, the line profilesof high-blueshift sources are more boxy (i.e., large valuesof D ) than those of the low-blueshift ones. It is also con-ceivable that, as the Eddington ratio increases, the radiationpressure plays a more important role in accelerating clouds(especially low column density ones; see, e.g., Marziani et al.2010). Such clouds could produce blueshifted broad C IV .This mechanism could also be (at least partially) responsiblefor the observed anti-correlation between the line-width ratioand the C IV blueshift.A remaining question is why quasars with different Ed-dington ratios/SEDs can have similar C IV EWs. Previ-
UN ET AL . 15ous works suggest that the “Baldwin effect” might be in-duced by the tight correlation between EWs and the Ed-dington ratios (e.g., Bachev et al. 2004; Baskin & Laor 2004;Shemmer & Lieber 2015). However, the C IV or He II EWmeasures only the ratio of the product of the E ion &
50 eV extreme UV (EUV) emission and the effective covering fac-tor of the BLR clouds to L . It is possible that either L or EUV emission cannot effectively track the diskemission (also see Vasudevan & Fabian 2007) and/or the ef-fective covering factor varies among quasars. Therefore, theC IV or He II EW is not an accurate indicator of the Edding-ton ratio or quasar SED.5.4.
The evolution of the line-shift variability
How do we understand the evolution of the line-shift vari-ability as a function of the C IV blueshift? Recall that theobserved line-shift variations are driven by the “line-of-sightvelocity-dependent reverberation” (see the first paragraph ofSection 4). According to this scenario, the variability am-plitude depends on the time-lag difference between the blueand red wings and on the variations of the ionizing contin-uum. The time-lag difference is expected if the BLR gas hassignificant radial motions (see also Barth et al. 2015).The line-shift variability of Mg II is more extreme than thatof C IV . At a first glance, this result is not expected since theionization energy of C IV is much larger than that of Mg II ,and the variability amplitude of quasar continuum emissiongenerally increases with energy. We argue that our resultcould be explained as follows. The distance of the locationof the Mg II gas should be larger than that of C IV , suggestingthat the Mg II gas is radially more extended. As a result, theMg II time-lag difference between the blue and red wings islarger than that of C IV , which leads to larger line-shift vari-ability of Mg II .For C IV , high-blueshift sources tend to have small line-shift variability. This connection might simply reflect thefact that, as mentioned in Section 5.2, high-blueshift sourcesare less variable in terms of quasar continua. However, theevolution of VAR.( V shift , MgII , se ) along the C IV blueshift isnot entirely expected. The evolution can only be explainedif the time-lag difference of the blue and red wings and/orthe ratio of the radial motions to the virial motions increaseswith Eddington ratio. Such a correlation may exist becausethe ratio of the radiative force to the gravitational poten-tial of the SMBH increases with the Eddington ratio (e.g.,Shakura & Sunyaev 1973). The radiative pressure can helpgenerate non-virial motions in the BLR.5.5. Is Eddington ratio the sole factor?
As we discussed above, our results can be explainedif the C IV blueshift is driven by the Eddington ratio(which is the driver of the quasar main sequence; see, e.g.,Boroson & Green 1992; Sulentic et al. 2000b; Shen & Ho2014). However, is the Eddington ratio the sole factor to de-termine the C IV blueshift? Figure 25 presents the Eddingtonratio as a function of the C IV blueshift. In general, the Ed-dington ratio and the C IV blueshift tend to be anti-correlated. − − CIV V off [km s − ] − − − l og λ E dd Figure 25.
Distribution of our sample in the C IV offset velocity- λ Edd plane. Sources with offset velocity < −
550 km s − are high-lighted by blue colors. The green (yellow) triangles represent themean log λ Edd (C IV V off ) in each C IV V off ( log λ Edd ) bin. Largeblueshift velocities tend to correspond to large λ Edd . However, thereverse is not true.
However, some extreme blueshift sources have low Edding-ton ratios, similar to those of no blueshift sources. This couldsimply be caused by the M BH estimation bias discussedabove. However, there are high Eddington-ratio sources thatshow almost no blueshift. There are several possibilities forthese sources. For instance, the high Eddington ratio mightonly be a necessary but insufficient condition for drivingaccretion-disk winds (Baskin & Laor 2005; Coatman et al.2016). Another possibility is that the M BH estimator weadopted in this work (i.e., the Mg II single-epoch virial M BH estimator; see Section 2.2) has significant intrinsic scatter( ≥ . dex; see, e.g., Vestergaard & Peterson 2006) andsuffers from considerable Eddington biases (Shen & Kelly2010). These speculations can be tested by controlling theEddington ratio and exploring the blueshift as a function ofquasar luminosity or other properties. Unfortunately, such astudy requires unbiased estimation of M BH over the entirequasar population, which is presently unavailable. The on-going RM campaigns, such as SDSS-RM (Shen et al. 2015)and other multi-object RM campaigns (King et al. 2015),have the potential to effectively test our scenario. SUMMARY AND CONCLUSIONWe have investigated the C IV blueshift as a function ofquasar properties, and constrained the intrinsic variability ofthe C IV blueshift in single-epoch spectra using the epochsof SDSS-RM spectra. Our primary results are as follows.1. We confirmed that the extreme blueshift sources gen-erally have small EWs, while the reverse is not true(Figure 4) with our high S/N composite spectra. Otherhigh-ionization emssion lines, such as He II , also showa blueshift, and the blueshift velocities are correlated6 with those of C IV (Figure 6). Furthermore, the depen-dence of the He II blueshift on EWs is similar to that ofC IV (Figure 7). These results suggest that the blueshiftbehavior is common for high-ionization emission lines(Section 3.1).2. Compared with Mg II , C IV is preferentially sup-pressed for the extreme blueshift sources (Figure 8).This result indicates a reduction of the high-energyionizing continuum over the low-energy one (Sec-tion 3.1).3. FWHM
CIV / FWHM
MgII anti-correlates with theC IV blueshift (Figure 9). Similar relations are alsofound for σ CIV /σ MgII , albeit the correlation is notapparent at the extreme blueshift (Figure 10). Theserelations can be used to make corrections for the C IV M BH estimators (Section 3.2).4. We also investigated the line-shift variability of Mg II (Figure 12) and C IV (Figure 13). The line-shift vari-ability of C IV and Mg II are different in terms ofvariability amplitude and their relation with the C IV blueshift. These differences indicate that the structuresof C IV and Mg II evolve differently as a function of theC IV blueshift (Sections 4.1 & 4.2). We also found thatquasar variability can slightly enhance the connectionbetween the C IV blueshift and EW (Figure 22; Sec-tion 5.1)5. We presented the variability of quasar continua as afunction of the C IV blueshift. The extreme blueshiftsources are less variable (Figure 24), indicating thatthe high-blueshift sources tend to have high Eddingtonratios (Figure 23, Section 5.2).6. All these results can be explained if quasar SEDs be-come softer with increasing Eddington ratios and withthe presence of X-ray shielding by the inner accretiondisk. However, a high Eddington ratio might be an in-sufficient condition for the C IV blueshift (Figure 25).Future multi-object RM experiments can probe ourscenario. M.Y.S. thanks K. D. Denney for beneficial discussion.We thank the anonymous referee for his/her helpful com-ments that improved the paper. M.Y.S. and Y.Q.X. acknowl-edge the support from NSFC-11603022, NSFC-11473026,NSFC-11421303, the 973 Program (2015CB857004), theChina Postdoctoral Science Foundation (2016M600485),the CAS Frontier Science Key Research Program (QYZDJ-SSW-SLH006), and the Fundamental Research Funds forthe Central Universities. W.N.B. acknowledges support fromNSF Grant AST-1516784 and Chandra X-ray Center grantGO6-17083X.Funding for SDSS-III has been provided by the Al-fred P. Sloan Foundation, the Participating Institutions, theNational Science Foundation, and the U.S. Departmentof Energy Office of Science. The SDSS-III web site is .SDSS-III is managed by the Astrophysical ResearchConsortium for the Participating Institutions of the SDSS-III Collaboration including the University of Arizona, theBrazilian Participation Group, Brookhaven National Labora-tory, Carnegie Mellon University, University of Florida, theFrench Participation Group, the German Participation Group,Harvard University, the Instituto de Astrofisica de Canarias,the Michigan State/Notre Dame/JINA Participation Group,Johns Hopkins University, Lawrence Berkeley National Lab-oratory, Max Planck Institute for Astrophysics, Max PlanckInstitute for Extraterrestrial Physics, New Mexico State Uni-versity, New York University, Ohio State University, Penn-sylvania State University, University of Portsmouth, Prince-ton University, the Spanish Participation Group, Universityof Tokyo, University of Utah, Vanderbilt University, Univer-sity of Virginia, University of Washington, and Yale Univer-sity. Software:
Astropy (Astropy Collaboration et al. 2013),Matplotlib (Hunter 2007), Numpy & Scipy (Van Der Walt et al.2011)REFERENCES
Abramowicz, M. A., Czerny, B., Lasota, J. P., & Szuszkiewicz, E.1988, ApJ, 332, 646Ai, Y. L., Yuan, W., Zhou, H. Y., et al. 2010, ApJL, 716, L31Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013,A&A, 558, A33Bachev, R., Marziani, P., Sulentic, J. W., et al. 2004, ApJ, 617, 171Baldwin, J. A. 1977, ApJ, 214, 679Barth, A. J., Bennert, V. N., Canalizo, G., et al. 2015, ApJS, 217,26Baskin, A., & Laor, A. 2004, MNRAS, 350, L31Baskin, A., & Laor, A. 2005, MNRAS, 356, 1029 Brotherton, M. S., Runnoe, J. C., Shang, Z., & DiPompeo, M. A.2015, MNRAS, 451, 1290Bolton, A. S., Schlegel, D. J., Aubourg, ´E., et al. 2012, AJ, 144,144Boroson, T. A., & Green, R. F. 1992, ApJS, 80, 109Chajet, L. S., & Hall, P. B. 2013, MNRAS, 429, 3214Chajet, L. S., & Hall, P. B. 2017, MNRAS, 465, 1741Coatman, L., Hewett, P. C., Banerji, M., & Richards, G. T. 2016,MNRAS, 461, 647Coatman, L., Hewett, P. C., Banerji, M., et al. 2017, MNRAS, 465,2120
UN ET AL . 17
Corbin, M. R. 1990, ApJ, 357, 346Dawson, K. S., Schlegel, D. J., Ahn, C. P., et al. 2013, AJ, 145, 10Denney, K. D. 2012, ApJ, 759, 44Denney, K. D., Horne, K., Brandt, W. N., et al. 2016, ApJ, 833, 33Denney, K. D., Horne, K., Shen, Y., et al. 2016, ApJS, 224, 14Denney, K. D., Peterson, B. M., Pogge, R. W., et al. 2009, ApJL,704, L80Dietrich, M., Hamann, F., Shields, J. C., et al. 2002, ApJ, 581, 912Dong, X.-B., Wang, T.-G., Wang, J.-G., et al. 2009, ApJL, 703, L1Du, P., Lu, K.-X., Zhang, Z.-X., et al. 2016, ApJ, 825, 126Eisenstein, D. J., Weinberg, D. H., Agol, E., et al. 2011, AJ, 142, 72Fine, S., Croom, S. M., Bland-Hawthorn, J., et al. 2010, MNRAS,409, 591Gaskell, C. M. 1982, ApJ, 263, 79Gaskell, C. M. 2009, NewAR, 53, 140Giveon, U., Maoz, D., Kaspi, S., Netzer, H., & Smith, P. S. 1999,MNRAS, 306, 637Grier, C. J., Peterson, B. M., Horne, K., et al. 2013, ApJ, 764, 47Gunn, J. E., Siegmund, W. A., Mannery, E. J., et al. 2006, AJ, 131,2332Guo, H., & Gu, M. 2016, ApJ, 822, 26Hou, A., Parker, L. C., Harris, W. E., & Wilman, D. J. 2009, ApJ,702, 1199Hunter, J. D. 2007, Computing in Science and Engineering, 9, 90Jiang, Y.-F., Stone, J. M., & Davis, S. W. 2014a, ApJ, 784, 169Jiang, Y.-F., Stone, J. M., & Davis, S. W. 2014b, ApJ, 796, 106Kelly, B. C. 2007, ApJ, 665, 1489Kelly, B. C., Treu, T., Malkan, M., Pancoast, A., & Woo, J.-H.2013, ApJ, 779, 187Kilerci Eser, E., Vestergaard, M., Peterson, B. M., Denney, K. D.,& Bentz, M. C. 2015, ApJ, 801, 8King, A. L., Martini, P., Davis, T. M., et al. 2015, MNRAS, 453,1701Krawczyk, C. M., Richards, G. T., Mehta, S. S., et al. 2013, ApJS,206, 4Korista, K. T., & Goad, M. R. 2004, ApJ, 606, 749Kozłowski, S. 2016, ApJ, 826, 118Leighly, K. M. 2004, ApJ, 611, 125Leighly, K. M., Halpern, J. P., Jenkins, E. B., & Casebeer, D. 2007,ApJS, 173, 1Luo, B., Brandt, W. N., Hall, P. B., et al. 2015, ApJ, 805, 122Luo, B., Brandt, W. N., Alexander, D. M., et al. 2014, ApJ, 794, 70Lusso, E., Comastri, A., Simmons, B. D., et al. 2012, MNRAS,425, 623MacLeod, C. L., Ivezi´c, ˇZ., Kochanek, C. S., et al. 2010, ApJ, 721,1014Marziani, P., Mart´ınez Carballo, M. A., Sulentic, J. W., et al. 2016,Ap&SS, 361, 29Marziani, P., Sulentic, J. W., Dultzin-Hacyan, D., Calvani, M., &Moles, M. 1996, ApJS, 104, 37 Marziani, P., Sulentic, J. W., Negrete, C. A., et al. 2010, MNRAS,409, 1033Marziani, P., Sulentic, J. W., Plauchu-Frayn, I., & del Olmo, A.2013, ApJ, 764, 150Murray, N., & Chiang, J. 1997, ApJ, 474, 91Plotkin, R. M., Shemmer, O., Trakhtenbrot, B., et al. 2015, ApJ,805, 123Proga, D., Stone, J. M., & Kallman, T. R. 2000, ApJ, 543, 686Richards, G. T., Lacy, M., Storrie-Lombardi, L. J., et al. 2006,ApJS, 166, 470Richards, G. T., Kruczek, N. E., Gallagher, S. C., et al. 2011, AJ,141, 167Richards, G. T., Vanden Berk, D. E., Reichard, T. A., et al. 2002,AJ, 124, 1Rumbaugh, N., Shen, Y., Morganson, E., et al. 2017,arXiv:1706.07875Runnoe, J. C., Brotherton, M. S., Shang, Z., & DiPompeo, M. A.2013, MNRAS, 434, 848Runnoe, J. C., Brotherton, M. S., DiPompeo, M. A., & Shang, Z.2014, MNRAS, 438, 3263Sa¸dowski, A., Narayan, R., McKinney, J. C., & Tchekhovskoy, A.2014, MNRAS, 439, 503Shakura, N. I., & Sunyaev, R. A. 1973, A&A, 24, 337Shemmer, O., & Lieber, S. 2015, ApJ, 805, 124Shen, Y., Brandt, W. N., Dawson, K. S., et al. 2015, ApJS, 216, 4Shen, Y., Brandt, W. N., Richards, G. T., et al. 2016, ApJ, 831, 7Shen, Y., Greene, J. E., Strauss, M. A., Richards, G. T., &Schneider, D. P. 2008, ApJ, 680, 169-190Shen, Y., & Ho, L. C. 2014, Nature, 513, 210Shen, Y., & Kelly, B. C. 2010, ApJ, 713, 41Shen, Y., & Liu, X. 2012, ApJ, 753, 125Shen, Y. 2013a, Bulletin of the Astronomical Society of India, 41,61Shen, Y., Liu, X., Loeb, A., & Tremaine, S. 2013b, ApJ, 775, 49Shen, Y., Richards, G. T., Strauss, M. A., et al. 2011, ApJS, 194, 45Smee, S. A., Gunn, J. E., Uomoto, A., et al. 2013, AJ, 146, 32Sulentic, J. W., Bachev, R., Marziani, P., Negrete, C. A., & Dultzin,D. 2007, ApJ, 666, 757Sulentic, J. W., del Olmo, A., Marziani, P., et al. 2017,arXiv:1708.03187Sulentic, J. W., Marziani, P., & Dultzin-Hacyan, D. 2000a,ARA&A, 38, 521Sulentic, J. W., Zwitter, T., Marziani, P., & Dultzin-Hacyan, D.2000b, ApJL, 536, L5Sun, M., Trump, J. R., Shen, Y., et al. 2015, ApJ, 811, 42Sun, Y.-H., Wang, J.-X., Chen, X.-Y., & Zheng, Z.-Y. 2014, ApJ,792, 54Trakhtenbrot, B., & Netzer, H. 2012, MNRAS, 427, 3081Trump, J. R., Impey, C. D., Kelly, B. C., et al. 2009, ApJ, 700, 498