A defining property of type-1 quasars is the presence of broad and narrow optical and UV lines emitted by ionic species over a wide range of ionization potentials (IPs, Vanden Berk et al., 2001) which can be conveniently grouped in high-ionization lines (HILs) involving IP ≳ 50 eV, and low-ionization lines from ionic species with IP ≲ 20 eV (Table 1). Optical and UV lines do not all show the same profiles, and quasar redshifts measured on different lines often show significant differences. Internal line shifts (i.e., differences in redshift from different emission lines measured for the same object) involve both broad and narrow emission lines and have offered a powerful diagnostic tool of the quasar innermost structure and of the emitting region dynamics since a few years after the discovery of quasars (Burbidge and Burbidge, 1967). This is true even if the assumption that unobscured type-1 quasars have very similar properties has remained widespread until very recent times. This assumption has been, in our opinion, one of the most damaging prejudices in the development of quasar research. At a meeting in 1999 in Mexico City Deborah Dultzin half-jokingly suggested that “a thousand spectra are worth more than one average spectrum” as an extension of the aphorism “a spectrum is worth a thousand pictures” (Dultzin-Hacyan et al., 2000). The developments in the last 15+ years have proved that this is indeed the case, although the value of single-epoch observations may have gone under appreciated with respect to other lines of evidence. We will therefore focus the scope of this review mainly to the organization of single-epoch spectra of large samples of quasars, following the “bottom-up” approach developed by Jack Sulentic and by his collaborators.

Over the years, the UV resonance line CIVλ1549 was considered as representative of broad HILs, and [OIII]λ4959,5007 were employed as strong and easily accessible narrow HILs. Typical broad LILs include Balmer lines, FeII emission as well as MgIIλ2800. Their analysis requires an accurate measurement of the quasar redshift. Narrow LILs (Hβ and [Oii]λ3727) have been found to be the best estimator of the quasar systemic redshift which defines the quasar “rest-frame” (e.g., Eracleous and Halpern, 2003; Hu et al., 2008; Bon et al., in preparation). The representative narrow and broad HILs [Oiii]λλ4959,5007 and Civλ1549 show systematic blueshifts with respect to LILs in a large fraction of type-1 AGN (see e.g., Gaskell 1982; Tytler and Fan 1992; Corbin and Boroson 1996; Marziani et al. 1996; Richards et al. 2002; Zamanov et al. 2002; Zhang et al. 2011; Marziani et al. 2016b, and Shen et al. 2016 for broad and narrow lines, respectively). Broad LILs could be used as a last resort especially in high-redshift quasars where the UV rest-frame is accessible from optical observations and no narrow lines are observable (Negrete et al., 2014). Even if broad LILs can show significant shifts with respect to rest frame, these are infrequent, and rarely as large as those found among HILs.

The interpretation of inter-line shifts in quasar spectra is mainly based on the Doppler effect due to gas motion with respect to the observer, along with selective obscuration. This explanation is almost universally accepted. For [OIII]λ4959,5007 blueshifts, it is consistent with a moderately dense outflow (logn~2−5 [cm⁻³]) of optically thin gas. For CIVλ1549, the explanation is not fully consistent if line emission occurs from optically thick clouds distributed, for example, symmetrically in a bicone whose axis is aligned with the spin of the black hole (e.g., Zheng et al., 1990). Such optically thick outflows might more easily give rise to a net redshift, if the receding part of the outflow remains visible. While it is no longer under discussion that large Civ blueshifts (amplitude ≳ 1,000 km/s) involve radial motion + obscuration, Gaskell and Goosmann (2013) suggested an alternative explanation involving infall and “reflection.” The observer does not see photons from the line emitting gas, but photons backscattered toward herself from a sea of hot electrons over the accretion disk. If the photons were originally emitted from infalling gas (i.e., approaching the disk), the observer should see a net blueshift. This explanation has some appeal for CIVλ1549, but sounds very unlikely for [OIII]λ4959,5007 because [Oiii]λλ4959,5007 is emitted on spatial scales ranging from a few pc to thousands of pc, where a suitable “mirror” as the one potentially offered by hot electrons surrounding the accretion disk may not exist. In view of source commonality in terms of CIVλ1549 and [OIII]λ4959,5007 blueshifts (Marziani et al., 2016a,b), we will follow the most widely accepted interpretation that blueshifts involve outflow and obscuration for both CIVλ1549 and [OIII]λ4959,5007. The interpretation of LIL blue- and redshifts will follow the same assumption (with some caveats, section 8.1).

This review will be focused on the way internal line shifts and other quasar properties can be efficiently organized. A major step was an application to quasar spectra of the Principal Component Analysis (PCA) carried out in the early 1990s (Boroson and Green, 1992; Francis et al., 1992). Boroson and Green (1992) measured the most prominent emission features in the Hβ spectral region, and found the first hint of the quasar “main sequence” (MS; their Figure 9). The PCA and other statistical techniques require measurable parameters for a set of sources. The starting point is therefore the definition of a set of parameters (section 2) that may be conductive to the identification of fundamental correlations (the Eigenvectors) as well as to physics. The PCA of type-1 quasars (we remark in section 3 that we are dealing exclusively with type-1, unobscured quasars) yields a first eigenvector from which the MS is defined (section 4). After reviewing the MS correlates (section 5), we show how the “empirical” eigenvector 1 can be connected to the main physical parameters of quasars seen as accreting systems (sections 6 and 7). The most intriguing results point toward two different accretion structures in type-1 quasars (section 8), which are however largely self-similar over a wide range in luminosity. As an example of the power of the MS to identify sources that are physically similar, we consider quasars located at the extreme tip of the MS which are potential distance indicators (section 9).

Single-epoch spectroscopy of large quasar samples yields data defined by limits in multiplexing (i.e., by the ability to obtain a record of signals in different wavebands with the same observations): synoptical observations of the UV, visual, NIR rest frame have been challenging until a few years ago (and, in part, they are still challenging to-date). Given the limit in multiplexing, observations of low- and high-z quasars provide different information since different rest-frame wavelength ranges are covered: visual spectrometers provide the rest-frame Hβ range at low z but the rest-frame UV at z ≳ 1.5. To cover the rest-frame UV at low-z, space-based observations are needed. To cover Hβ at high z, NIR spectroscopy is needed. These limitations are being overcome by new generation instruments mounted at the focus of 8m-class telescopes which provide simultaneous coverage of visual and NIR, but these facilities were not available at the time most of the work reviewed in this paper was done. Synoptic observations of visual and UV at low-z still require coordinated ground and space-based observations which are especially hard to come by.

Table 1 provides an overview of the lines covered in the different domains. When we speak of intermediate to high-z objects (a population of behemoth quasars that is now extinguished), we are speaking of sources that are not anymore observed at z ≲ 1 (mainly for the “downsizing” of nuclear activity, see e.g. Springel et al., 2005; Sijacki et al., 2015; Fraix-Burnet et al., 2017, and references therein). On the other hand, at high-z, only relatively few quasars with luminosities comparable to those of low-z quasars are known since their apparent magnitude would be too faint. They aren't yet efficiently sampled by the major optical source of quasar discovery, the Sloan Digital Sky Survey (SDSS, Blanton et al. 2017, and references therein).

Unification schemes have provided a powerful conceptual framework suitable for organizing the analysis of AGN. The precursor distinction between type 1 and 2 Seyfert (Khachikian and Weedman, 1974) gained a convincing interpretation when Antonucci and Miller (1985) reported the discovery of a broad line component visible in the polarized spectrum of Seyfert 2 nuclei but invisible in natural light: the broad line region is hidden from view and only photons scattered by hot electrons toward our line of sight are received by the observer. This explanation remains alive and widely accepted today (e.g., Eun et al., 2017), even if we now know that is only a part of the story: type-2 AGN differ for environmental properties (Dultzin-Hacyan et al., 1999; Koulouridis et al., 2006; Villarroel and Korn, 2014), may intrinsically lack a BLR (Laor, 2000) at very low accretion rates, or may even be unobscured normal type-1 under special conditions (Marinucci et al., 2012). The point here is that unification models of RQ AGN separate two types of quasars on the basis of the viewing angle between the line of sight and the symmetry axis of the system (i.e., the spin axis of the black hole or the angular momentum vector of the inner accretion disk) but make no prediction on unobscured type 1 AGN. Orientation effects are expected also for unobscured type-1s, as we should observe them in the range of viewing angles, 0 ≲ θ ≲ 45 − 60. There is little doubt that broad line width is affected by orientation, especially for RL sources (e.g., Wills and Browne, 1986; Rokaki et al., 2003; Sulentic et al., 2003; Jarvis and McLure, 2006; Runnoe et al., 2014): a comparison between RL sources that are core-dominated (believed to be oriented with the jet axis close to the line-of-sight) and lobe-dominated (misaligned) shows that the Hβ FWHM is larger in the latter class. This result strongly suggests a flattened, axisymmetric structure for the BLR. For RQ objects, the evidence is not obvious and an estimate of θ remains an unsolved problem at the time of writing. However, orientation effects are certainly not enough to explain the diversity of quasar spectral properties.

At the time the Boroson and Green (1992) paper appeared, studies based on moderately sized samples (20-30 objects) were common and often reached confusing results from correlation analysis. The best example is the Baldwin effect (an anti-correlation between equivalent width of HILs and luminosity) which was found in some and then not found in similar samples without apparent explanation.¹ In this respect the sheer size of the Boroson and Green (1992) sample was a key improvement. A novel aspect was also the application of the PCA which considers each parameter as a dimension of a parameter space, and searches for a new parameter space with fewer dimensions (defined by linear combinations of the original parameters) as needed to explain most of the data variance (Murtagh and Heck, 1987; Marziani et al., 2006). The application of the PCA was not unprecedented in extragalactic astronomy (e.g., Diaz et al., 1989) but was well suited to quasar data that appeared weakly correlated among themselves without providing a clear insight of which correlations were the most relevant ones.

The quasar Eigenvector 1 was originally defined by a PCA of ≈ 80 Palomar-Green (PG) quasars and associated with an anti-correlation between strength of FeIIλ4570, R_FeII (or [OIII] 5007 peak intensity) and FWHM of Hβ (Boroson and Green, 1992). The parameter R_FeII is defined as the ratio between the integrated flux of FeIIλ4570 blend of multiplets, and that of the Hβ broad component:² R_FeII = I(FeIIλ4570)/I(Hβ). Since 1992, various aspects of the Eigenvector 1 (E1) of quasars involving widely different datasets as well multi-frequency parameters have been discussed in more than 400 papers, as found on NASA ADS in August 2017 (Boroson and Green, 1992; Dultzin-Hacyan et al., 1997; Sulentic et al., 2000a,b, 2007; Shang et al., 2003; Grupe, 2004; Kuraszkiewicz et al., 2009; Mao et al., 2009; Tang et al., 2012). Earlier analyses have been more recently confirmed by the exploitation of SDSS-based samples (Yip et al., 2004; Wang et al., 2006; Zamfir et al., 2008; Kruczek et al., 2011; Richards et al., 2011; Marziani et al., 2013b; Shen and Ho, 2014; Brotherton et al., 2015; Sun and Shen, 2015).

The second eigenvector – Eigenvector 2 – was found to be proportional to luminosity, and eventually associated with the HIL Baldwin effect(s) that are the most-widely discussed luminosity effects in quasar samples (Baldwin et al., 1978; Dietrich et al., 2002; Bian et al., 2012). The smaller fraction of variance carried by the Eigenvector 2 indicates that luminosity is not the major driver of quasar diversity, especially if samples are restricted to low-z. We will not further consider HIL luminosity effects³ but only discuss the influence of luminosity on the LIL FWHM (section 8.2).

The distribution of data points in the optical plane of the Eigenvector 1 FWHM(Hβ) vs. R_FeII traces the quasar main sequence (MS, Figure 2), defined for quasars of luminosity logL ≲ 47 [erg s⁻¹], and z < 0.7. The MS shape allows for the definition of a sequence of spectral types (Figure 2), and motivates subdividing the 4DE1 optical plane into a grid of bins of FWHM(Hβ) and FeII emission strength. Bins A1, A2, A3, A4 are defined in terms of increasing R_FeII with bin size Δ R_FeII = 0.5, while bins B1, B1+, B1++, etc. are defined in terms of increasing FWHM(Hβ) with Δ FWHM = 4,000 km s⁻¹. Sources belonging to the same spectral type show similar spectroscopic measurements (e.g., line profiles and line flux ratios). Spectral types are assumed to isolate sources with similar broad line physics and geometry. Systematic changes are reduced within each spectral type. If so, an additional advantage is that an individual quasar can be taken as a bona fide representative of all sources within a spectral type. The binning adopted (see Figure 2) has been derived for low-z (< 0.7) quasars. Systematic changes may not be eliminated in full, if an interpretation scheme such as the one of Marziani et al. (2001) applies, who posited a continuous effect of Eddington ratio and viewing angle (at a fixed M_BH) as the origin of the MS shape (section 7.5 provides further explanations).

Developments in the analysis before late 1999 of low-z quasar spectral properties are reviewed in Sulentic et al. (2000a). Data and ideas were in place as early as in year 2000 to introduce the idea of two quasar populations, A and B: Population A with FWHM(Hβ) ≤ 4,000 km s⁻¹; Population B (broader) with FWHM(Hβ) > 4,000 km s⁻¹ (Sulentic et al., 2000a,b). Later developments have confirmed that the two populations are two distinct quasar classes. Population A may be seen as the class that includes local NLSy1s as well as high accretors (Marziani and Sulentic, 2014), and Population B as a class capable of high-power radio-loudness (Zamfir et al., 2008, see section 5). It now seems unlikely that the two populations are just the opposite extremes of a single quasar “main sequence” defined in the plane FWHM(Hβ) vs. R_FeII (Sulentic et al., 2011, section 5), although the subdivision at FWHM(Hβ) = 4,000 km s⁻¹ is not widely considered in literature. It is therefore worth analyzing the issue in some more detail after considering the main correlates along the MS.

Several correlates have been proved as especially relevant in the definition of the MS multifrequency properties.

Tables reporting main-sequence correlates are provided in several recent review and research papers (e.g., Sulentic et al., 2011; Fraix-Burnet et al., 2017), and in Chapter 3 and 6 of D'Onofrio et al. (2012). To restrict the attention of a subset of especially significant parameters, Sulentic et al. (2000a) introduced a 4DE1 parameter space. In addition to FWHM(Hβ) and R_FeII, two more observationally “orthogonal” parameters, Γ_soft and c(12) Civλ1549 are meant to help establish a connection between observations and physical properties. The 4DE1 parameters clearly support the separation of Population A (FWHM Hβ < 4,000 km s⁻¹) and Population B(broader) sources,⁴ although the non-optical parameters are not always useful since they are MS correlates and often unavailable. The immediate interpretation of the 4DE1 parameters is summarized in Table 2. In the simplest term, the FWHM Hβ is related to the velocity dispersion in the LIL emitting part of the BLR. On the converse, c(12) Civ yields a measurement affected by the high-ionization outflow detected in the HIL profile. The largest c(12) values indicate a decoupling between the strongest HIL and LIL features, with the latter remaining symmetric and unshifted (Marziani et al., 1996). The parameter R_FeII is of more complex interpretation. R_FeII is affected by the metallicity (obviously, if [Fe/H] = −10, R_FeII ≈ 0) but metallicity is most likely not all of the story (Joly et al., 2008), since Feii strength tends to saturate for high metallicity values. The main dependence is probably on ionization conditions, density and column density (section 7.5). A Γ_soft > 2 is usually ascribed to Compton thick soft X-ray emission from a hot corona above the disk, but may also be the high-energy tail of the spectral energy distribution of the disk itself, in case the inner disk is very hot (e.g., Done et al., 2012; Wang et al., 2014a).

Several physical parameters (black hole mass M_BH, Eddington ratio, spin, and the aspect angle θ) are expected to affect the parameters of the E1 MS, even if in an indirect way as in the case of the spin. Establishing a connection between a physical and an observational set of parameters is precisely the aim of the 4DE1 parameter space. Table 2 lists the main physical parameters on which the empirical 4DE1 parameters are expected to be dependent. We still have a very incomplete view of the physics along the Eigenvector 1 sequence because we are able to make only coarse estimates of physical parameters. However, Eddington ratio and viewing angle θ are likely to be the main culprits affecting the location of a quasar along the MS, as discussed in the next sections.

The accretion process is apparently largely self-similar over several order of magnitudes in black hole mass and luminosity. The general fundamental plane of accreting black holes emphasizes the invariance of the accretion disk-jet scaling phenomena (Merloni et al., 2003). The Eigenvector 1 scheme is also emphasizing a scale invariance, albeit more oriented toward the radiatively-efficient domain and, in the quasar context, limited to unobscured type-1 quasars. As far as the radiatively efficient accretion mode is concerned, the invariance may hold over a factor of almost ~10¹⁰ in solar mass (Zamanov and Marziani, 2002). Restricting the attention to RQ quasars, very large blueshifts are observed over at least four orders of magnitude in luminosity. The distribution of data points in Figure 5 is clearly affected by selection effects; however, it shows that large Civ c(12)≲-1,000 km s⁻¹ do occur also at relatively low luminosity. Outflow velocities are apparently more related to Eddington ratio than to L or M_BH: large c(12) blueshifts may occur only above a threshold at about L/L_Edd ≈ 0.2 − 0.3 which may in turn correspond to a change in accretion disk structure (Abramowicz and Straub, 2014, and references therein). Intriguingly, the threshold matches the FWHM(Hβ)≈ 4, 000 km s⁻¹ limit separating Population A from Population B (if L ≲ 10⁴⁷ erg s⁻¹). At this limiting FWHM we observe also a change in Hβ profile shape, from Lorentzian-like to double Gaussian (Sulentic et al., 2002; Zamfir et al., 2010).

Sources belonging to spectral types A3 and A4 (i.e., satisfying the criterion R_FeII> 1) are found to be radiating at the highest Eddington ratio. They show a relatively small dispersion along a well-defined, extreme value of order (1)⁷ (Marziani and Sulentic, 2014). The xA selection criterion is consistent with the parameter ∝ FWHM/σ+R_FeII used to identify super-massive extremely accreting BHs (SEAMBHs Wang et al., 2013; Du et al., 2016). Since xA sources show Lorentzian profiles, the criterion based on R_FeII should be sufficient unless relatively broad profiles with FWHM≳4,000 km s⁻¹ are considered, a case still under scrutiny.

If L/M_BH ≈ const., then the luminosity can be retrieved once the mass is known. xA sources show very similar spectra over a broad range of luminosity. The self-similarity in terms of diagnostic line ratios justifies the use of the scaling law rBLR∝L0.5 that implies spectral invariance. The luminosity can then be written as L∝(δv)⁴, where δv is a suitable VBE. It is interesting to note that this relation is in the same form of the Faber-Jackson law as originally defined (Faber and Jackson, 1976). Assuming that spheroidal galaxies are virialized systems, radiating at a constant L/M_BH ratio implies that their luminosity is L∝σ4/μB, where μ_B is the surface brightness within the effective radius. Later developments yielded a different exponent for the dependence on σ (e.g., D'Onofrio et al., 2017, and references theerin), as the assumption of a constant surface brightness proved untenable. There should be no similar difficulties for quasars, although statistical and systematic sources of error should be carefully assessed (Marziani and Sulentic, 2014). The distance modulus computed from the virial equation, μ = 2.5[logL(δv)−BC]−2.5log(f_λλ)−const.+5·log(1+z), where f_λλ is the flux at 5100 Å, and BC the bolometric correction, is in agreement with the expectation of the concordance ΛCDM cosmology, providing a proof of the conceptual validity of the virial luminosity L∝(δv)⁴ estimates (a plot of μ for several quasar samples is provided by Marziani et al., 2017b, and by Negrete et al. 2017).

The quasar MS provides a tool to systematically organize quasar multifrequency properties. It allows the identification of spectral types with fairly well-defined spectral properties. In this paper, we have described a simple parameterization that is able to describe the quasar emission line profiles, as well as a heuristic technique motivated by the internal line shifts yielding the separation of components in different physical conditions.

We then considered the MS defined in the so-called optical plane of the 4D eigenvector 1 parameter space, and analyzed several correlates involving the profile of Hβ and Mgiiλ2800 (the representative LILs), of the Civλ1549 (the representative HIL), broad-line UV diagnostic ratios that provide trends in density and ionization level, [Oiii]λλ4959,5007 shifts, the prevalence of radio-loudness, and the soft X-ray spectral slope.

The trends defined in this paper offer a consistent, empirical systematization of quasar properties for z ≲ 1, low-to moderate luminosity quasars. The difference between Pop. A and B is evident at the extremes: spectral type A4 sources typically show very strong Feii (R_FeII ≳ 1.5), blueward asymmetry in Hβ, large Civλ1549 blueshift, weak and blue shifted [Oiii]λλ4959,5007(Negrete et al., 2017). Extreme Pop. B show undetectable Feii, very broad red-ward asymmetric Hβ profiles, small-amplitude Civλ1549 shifts, prominent [Oiii]λλ4959,5007. There is evidence that the two populations represent structurally different objects. The blue shifted excess (BLUE) is interpreted as due to outflowing gas dominating HIL emission only in Pop. A (unless sources at very high L are considered). A redshifted VBC is present only in Pop. B, for values of FWHM Hβ above 4,000 km s⁻¹, and has been interpreted as due to gas close to the central continuum source. The relative balance between gravitational and radiation forces (i.e., L/L_Edd) appears as a major factor influencing both the dynamics and the physical conditions of the line emitting gas (Sect. 8.2), and an accretion mode change may be associated with a critical L/L_Edd ~0.2±0.1, leading to the two quasar populations: A (wind-dominated, following Richards et al. 2011), and B (virial or disk-dominated).

Luminosity trends are weak as they become significant only over a wide luminosity range ~10⁴³−10⁴⁸ erg/s. They involve an overall increase in virial line broadening (LILs) and an increase of blueshift frequency and amplitude consistent with the dominance of radiation forces (HILs).

The presence of a virialized subregion identified along the MS at low-L as well as at high-L has important consequences. xA quasars at the high R_FeII end (spectral type A3 and A4) may be suitable as Eddington standard candles since their Eddington ratio scatters around a well-defined value (e.g., Negrete et al., 2017).

The contextualization offered by the MS is instrumental to the development of better-focused physical models along the quasar main sequence. As examples of the advantages offered by the E1 MS, we just mention the possibility of contextualizing orientation effects in RQ quasars, and of performing a meaningful comparison between RL and RQ samples with similar optical properties. An aspect still to clarify is the connection between disk structure and the dynamics of the line emitting gas.

PM wrote the paper; DD, MD, GS, EB, and NB contributed with a critical reading and suggestions; AD, CN, MM-A, and JS contributed with a critical reading.