BAL and non-BAL quasars: continuum, emission, and absorption properties establish a common parent sample

ABSTRACT

1 INTRODUCTION

Quasar-driven outflows are now widely invoked in galaxy formation models in order to reproduce the observed properties of massive galaxies. Broad absorption lines (BALs) observed in the ultraviolet (UV) spectra of quasars are long established as providing evidence for the presence of high-velocity outflows among a substantial fraction of the quasar population (Weymann et al. 1991; Hewett & Foltz 2003; Allen et al. 2011). Of the intrinsic, extinction-corrected, ≃40 per cent of luminous, optically selected quasars classified as BAL quasars (Dai, Shankar & Sivakoff 2008; Allen et al. 2011), the majority, so-called HiBALs, show absorption of highly ionized species such as C iv λ1549, Si iv λ1397, N v λ1240, and O iv λ1034. LoBAL quasars also show absorption of low-ionization species (e.g. Al iii λ1857 and Mg ii λ2800), while FeLoBALs are quasars that additionally display Fe ii and Fe iii absorption over extended wavelength intervals. The broad, blueshifted absorption is believed to result from the presence of outflowing gas along the direct line of sight. Evidence for the presence of infalling or rapidly rotating material from BALs is rare (Hall et al. 2013).

Following an original observation by Gaskell (1982), increasing attention over the last decade has been devoted to quantifying the outflow signatures evident from the high-ionization emission lines in the UV (e.g. Sulentic et al. 2000; Leighly 2004; Richards et al. 2011) – particularly the blueshifting of the prominent C iv λ1549 emission.

Disc winds (Murray et al. 1995; Elvis 2000; Proga, Stone & Kallman 2000; Proga 2003) are generally considered to explain the origin of both the observed BALs and the blueshifted emission. From a thin accretion disc, a wind emerges as a result of radiation pressure from the disc’s UV emission (see fig. 1 of Murray et al. 1995). Some models invoke shielding of the wind (Murray et al. 1995; Proga et al. 2000) or perhaps disc geometry (Leighly 2004; Luo et al. 2015) to explain the high velocities that the gas can reach without being overionized by the central source. Radiation line driving by UV photons has been considered as an acceleration mechanism of the wind (e.g. Giustini & Proga 2019) and signatures of this process, such as line locking, have been observed in both narrow absorption lines (NALs) and BALs (Bowler et al. 2014; Mas-Ribas & Mauland 2019). However, acceleration by magnetohydrodynamics has also been considered (Blandford & Payne 1982) and it is probable that both acceleration mechanisms are responsible for the observed winds.

Figure 1.

Example spectra (black; smoothed by 5 pixels) and reconstructions (red) showing the quality of the reconstructions across the full range of spectrum S/N. Also plotted are the residuals (spectrum − reconstruction) divided by the spectrum noise at each pixel. Non-BAL quasars are plotted in the top row and classically defined BAL quasars are plotted in the bottom row. The middle row contains spectra with absorption too narrow to have positive BI but with positive AI. The spectrum S/N increases from left to right.

Open in new tabDownload slide

In a number of models, the properties of outflowing material are orientation dependent, e.g. relative to the black hole spin axis, and the probability a quasar is observed to possess BAL features depends on viewing angle. There is, however, still no agreement on whether BALs do arise as the result of viewing angle or represent a particular phase in the fuelling-outflow cycle, with the wind potentially clearing the absorbing gas from the galaxy as the quasar transitions to a non-BAL quasar. With multiple observations of the same object now available, it is possible to observe the variability of the BALs on rest-frame time-scales of ≤5 yr with some objects transitioning from BAL quasars to non-BAL quasars (Filiz Ak et al. 2012, 2013; Sameer et al. 2019) and vice versa (Rogerson et al. 2018). Such transformations may arise due to changes in the ionizing state of the outflowing gas or result from wind motion transverse to the line of sight (e.g. Gibson et al. 2010; Sameer et al. 2019).

The focus of this paper is the far more common HiBAL objects among the BAL quasar population and we exclude consideration of the much smaller fractions of LoBAL and FeLoBAL quasars. Careful comparisons of the emission-line properties of HiBAL and non-BAL quasars have been undertaken (e.g. Weymann et al. 1991; Reichard et al. 2003; Baskin, Laor & Hamann 2013, 2015). Results of such investigations have shown that, in general, the UV and optical spectral energy distributions (SEDs), unaffected by the presence of BAL troughs, of BAL and non-BAL quasar populations are very similar. Relatively weak emission features, however, which are diagnostics of the far-UV ionizing spectrum, such as He ii λ1640, have been found to show systematic differences between the populations (Baskin et al. 2013). To date, the strong absorption present in BAL quasars has precluded a direct comparison of the kinematic properties of high-ionization UV emission lines, such as C iv λ1549, in the BAL and non-BAL quasar populations. As a consequence, investigations of the outflow properties of luminous quasars have been confined to evidence from either absorber or emission properties.

The key aim of this paper is to provide the first direct comparison of the underlying UV SEDs of BAL and non-BAL quasars and thus to investigate outflow properties using both absorber and emission diagnostics. The ultimate goal is to use the relationships between the absorption and emission outflow properties to constrain the geometry and physical parameters of a disc wind or other models. In this first paper, however, we concentrate on presenting the observational results for a large sample of BAL and non-BAL quasars.

The structure of the paper is as follows. In Section 2, we review the task of recovering the unabsorbed UV SEDs of BAL quasars and the approach taken to perform the investigation here. The selection of the quasar sample is described in Section 3 before the specific recipe for the spectral reconstructions is presented in Section 4. The question of the definition of the sample of quasars that possess ‘broad absorption’ is the topic of Section 5, where both the classic definitions involving the balnicity index (BI) from Weymann et al. (1991) and the more extensive absorption index (AI) from Hall et al. (2002) are employed. The results of the investigation are presented in Section 6 before a short discussion of the implications in Section 7. An overview of the main conclusions is included in Section 8.

Vacuum wavelengths are employed throughout the paper and we adopt a Lambda cold dark matter cosmology with h₀ = 0.71, Ω_M = 0.27, and Ω_Λ = 0.73 when calculating quantities such as quasar luminosities.

2 THE PROBLEM AT HAND: RECONSTRUCTING QUASAR SPECTRA

Performing a statistical analysis of quasar UV spectra to produce a small number of ‘components’, linear combinations of which are capable of reproducing individual quasar SEDs, is long established (Francis et al. 1992; Yip et al. 2004). Application of the component reconstruction techniques has proved successful in the context of BAL quasars, where it is necessary to reconstruct the unabsorbed SED at wavelengths affected by the presence of BAL troughs. The Sloan Digital Sky Survey Data Release 12 (SDSS DR12) BAL catalogue (Pâris et al. 2017) is the most extensive sample. The non-negative matrix factorization (NMF)-based analysis of the SDSS DR6 spectra (Allen et al. 2011) resulted in very effective reconstructions, albeit over a restricted range in wavelength, including the C iv λ1549 emission line. Implementations involving principal component analysis (PCA) now form key elements of the redshift determination for quasars and the identification of BALs in recent large-scale surveys, particularly the SDSS (Pâris et al. 2017; Abolfathi et al. 2018; Pâris et al. 2018).

From the previous work, it is clear that the accuracy of any component-based reconstructions of quasar spectra needs to be high in order to parametrize differences in the emission-line SEDs such as the He ii λ1640 properties, particularly for spectra with extensive wavelength ranges affected by BAL troughs. In order to undertake such a quantitative investigation of the underlying UV SEDs of high-ionization BAL quasars and non-BAL quasars using the very large quasar spectra samples available from the SDSS (Pâris et al. 2018), we have revisited the question of how to obtain high-accuracy reconstructions of individual BAL and non-BAL spectra that possess extensive ranges in SED properties.

The PCA, NMF, and mean-field independent component analysis (MFICA) (see later) approaches all involve reconstructions based on linear combinations of derived components. There is, however, a significant variation in the form of quasar UV SEDs due to multiplicative changes as a function of wavelength; the effect of extinction due to dust along the line of sight is a familiar example. Such intrapopulation non-linear variations are not optimally characterized via PCA/NMF/MFICA analyses and we have therefore applied an empirical large-wavelength scale correction to the overall shape of the quasar spectra prior to deriving components and generating reconstructions of individual spectra. Essentially, the individual spectra are given the same overall shape and we term the process ‘morphing’ throughout the paper. The application of the large-scale multiplicative shape morphing has the advantage of reducing the number of components required to produce a reconstruction of specified accuracy. A full description of the procedure is given in Section 4.

To generate the components that allow spectrum reconstructions, we have used MFICA decompositions (Højen-Sørensen, Winther & Hansen 2002; Opper & Winther 2005; Allen et al. 2013). For a specified number of components, the MFICA works as well or better than any decomposition scheme we have investigated (Allen & Hewett, in preparation). As a result, relatively few components are required to produce reconstructions, making reconstructions of only partially complete spectra (as required for the BAL quasars) more stable.

The question of reconstruction reliability and stability for spectra with extensive missing regions, such as the underlying emission spectra for BAL quasars with large troughs, is key and discussed in some detail by Allen et al. (2011). In our analysis, we are able to place priors on the component weights when reconstructing wavelength regions affected by BALs (particularly the C iv λ1549 emission and wavelength regions to the blue) using the properties of the individual quasar spectra unaffected by BAL troughs. In more detail, our scheme uses priors on the MFICA component weights based on the properties of the C iii]λ1908–Si iii]λ1892–Al iii λ1857 complex, which are closely related to the morphology of the emission in the 1200–1600 Å range as evident in fig. 11 of Richards et al. (2011).

Using the MFICA component scheme, we have the ability to reconstruct 95 per cent of (non-)BAL quasar spectra to an accuracy¹ of 93 (94) per cent over wavelength range of 1265–3000 Å. The procedure is described in detail in Sections 4 and 5, but readers interested primarily in the results may wish to inspect Fig. 1, which shows the quality of the reconstructions achieved, and move on to Section 6.

3 QUASAR SAMPLE

The quasar sample is based on the catalogue compiled from the SDSS DR14Q (Pâris et al. 2018), which contains ≃526 000 quasars. The Baryon Oscillation Spectroscopic Survey (BOSS) involved observations of galaxies as well as quasars to allow an investigation of the intervening Lyman-α forest absorption. Unlike the initial SDSS survey, which resulted in the Schneider et al. (2010) quasar catalogue drawn from the SDSS DR7, determination of a quantitative selection function for quasars as a function of SED, redshift, and magnitude was not attempted for the BOSS survey. As a consequence, completeness information for quasars (Richards et al. 2002) and BAL quasars (Allen et al. 2011) is not available. The situation is compounded by the fainter magnitude limit to which SDSS DR14 extends and the significant fraction of quasar spectra with very low signal-to-noise ratio (S/N).

Fortunately, for the comparative investigation presented here, it is not necessary to determine the detection probability P_qso(SED, m, |$z$|⁠) but only to ensure that the probabilities for BAL and non-BAL quasars are similar. The detectability of BAL troughs decreases significantly as the spectrum continuum S/N decreases (Allen et al. 2011). Only the most extreme BAL quasars are therefore identifiable in the DR14 quasars that possess low continuum S/N. As the intention is to investigate the relationship between BAL quasar absorption properties and their underlying emission-line properties, a minimum spectrum S/N threshold for inclusion of quasars is adopted.

The majority of spectra included in the sample possess an average S/N ≥ 5.0 threshold, per SDSS spectrum pixel, over the rest-frame wavelength interval of 1600–2900 Å (when Mg ii λ2800 is present in the spectrum) or 1600–2000 Å (at higher redshifts).² Fig. 2 plots the fraction of BAL and non-BAL quasars as a function of average S/N. As expected, the fraction of BAL quasars decreases at low S/N.

Figure 2.

Top: Distribution of average S/N for our quasar sample. Bottom: Fraction of BAL and non-BAL quasars as a function of S/N with Poisson errors on the BAL fraction. At each S/N bin, the fraction of BALs and non-BALs adds to unity. Note the drop in BAL quasars at low S/N is due to the difficulty in detecting troughs in noisy spectra.

Open in new tabDownload slide

The classification of quasars as HiBALs or non-BALs is based on the C iv λλ1548,1551 emission line and the 25 000 km s⁻¹ region blueward of the line (Section 5). Minimum redshifts of |$z$| = 1.56 for quasars with SDSS-III spectra (minimum λ_obs ≃ 3600 Å) and |$z$| = 1.67 for quasars that possess only spectra from DR7 (minimum λ_obs ≃ 3800 Å) are therefore adopted. The number of quasars with redshifts |$z$| > 3.5 whose spectra satisfy the S/N threshold is small and a maximum redshift of |$z$| = 3.5 is also imposed. The sample of quasars satisfying the redshift interval constraints and the spectrum S/N threshold numbers ≃144 000. Our sample contains only the default spectra listed in DR14Q of quasars for which duplicate observations exist.

The estimation of systemic redshifts for quasars in the SDSS data releases has improved significantly since DR7 (Schneider et al. 2010) and the re-analysis of Hewett & Wild (2010). Extensive discussion of the effectiveness of the improved schemes is contained in Pâris et al. (2017, 2018) and the implications for clustering investigations are reviewed by Zarrouk et al. (2018). Notwithstanding the improvements, our own investigations demonstrate that substantial advances in the accuracy of systemic redshift estimates can be made relative to those included in Pâris et al. (2018).

Here, quasar redshifts are calculated using spectrum reconstructions based on an MFICA scheme (Allen et al. 2013) with the redshift as a free parameter. The reconstructions are deliberately confined to the 1600–3000 Å region, thereby excluding the C iv emission line. The stability of the rest-frame wavelengths of the low-ionization emission lines O i λ1304 + Si ii λ1307, C ii λ1335, Al iii λ1857, Si iii]λ1892, C iii]λ1908, and Mg ii λ2800, independent of the large range of UV SEDs, including high-ionization emission-line blueshifts, is used to verify the effectiveness of the redshift determinations. In terms of the quasar rest-frame velocity differences, in excess of 35 000 quasars show a shift of >500 km s⁻¹ between the DR14 and our new redshifts, with ≃9800 quasars possessing shifts of >1000 km s⁻¹. While the use of the new redshifts has an impact on the effectiveness of the spectral reconstructions and which objects are classified as BAL quasars (Section 5), none of the statistical results presented in Section 6 change if the DR14 redshifts are employed.

The redshift distribution of the base 143 664-quasar sample to be analysed is presented in Fig. 3 along with a comparison of the SDSS DR14 redshifts with our new values. The first stage in the analysis is to generate reconstructions for each of the quasar spectra as described in the next section.

Figure 3.

Comparison of DR14Q redshifts, |$z$|_SDSS, with our new redshifts, |$z$|_new. Top: Distribution of corrected redshifts of all quasars in our base sample (black) and the subsample of classically defined BAL quasars (red; see Section 5). The BAL quasar counts have been multiplied by 4 to better show the n(⁠|$z$|⁠) distribution. Bottom: Difference between |$z$|_new and |$z$|_SDSS, as a velocity in the rest frame, as a function of |$z$|_new.

Open in new tabDownload slide

4 QUASAR SPECTRA RECONSTRUCTIONS

The rest-frame wavelength range selected for the investigation is 1260–3000 Å. The long-wavelength limit ensures the inclusion of the Mg ii λ2800 emission line for quasars with redshifts up to |$z$| ≃ 2.4 in the DR14 spectra. The short-wavelength limit allows BAL troughs associated with the O iv + Si iv emission at 1400 Å to be included as well as emission due to the low-ionization species O i λ1304 + Si ii λ1307 and C ii λ1335. The strong BAL absorption associated with the N v λ1242 emission and Lyman-α emission is in principle also of interest. The presence of absorption from the Lyman-α forest, however, complicates the problem of reconstructing the intrinsic quasar SEDs, particularly where the aim is to utilize only a small number of components to achieve the reconstructions.

4.1 Quasar spectrum shape standardization: ‘morphing’

Unlike the original SDSS DR7 quasar spectra, the observational constraints relating to the placement of the fibres for the BOSS DR14 observations mean that differential atmospheric refraction has a significant effect on the spectrophotometry, particularly at blue wavelengths. An extensive discussion of the effect and a scheme for performing post-reduction corrections to the spectrophotometry are presented in Margala et al. (2016).

The differential atmospheric refraction effects give rise to ±10 per cent multiplicative variations in the spectrophotometry. Variations in the amount of dust extinction affecting the UV continuum and emission-line fluxes from quasar to quasar also produce ‘intrinsic’ multiplicative differences in the large-scale shape of the observed spectra. As heralded in Section 2, the presence of such wavelength-dependent multiplicative changes complicates any scheme whereby spectra are reconstructed using linear combinations of fixed ‘components’. The first stage in the spectrum reconstruction procedure is therefore to ensure that all the individual quasar spectra possess the same overall shape. The reference SED used in the spectrum shape morphing is a model quasar spectrum as described in Maddox et al. (2012). By construction, the model reproduces the UV SED of luminous, unreddened quasars and a mean spectrum from the SDSS DR7 quasar catalogue would work equally well.

The shape morphing proceeds as follows. Eight MFICA components (Section 4.2), to allow spectrum reconstructions, are generated using ≃4000 spectra³ of non-BAL quasars with a range of spectrum shapes that have not been shape-morphed. Individual spectrum reconstructions are then calculated using the masking procedure (to exclude wavelengths affected by both broad and narrow absorption features) described in Section 4.3 for all spectra. The scheme thus allows spectra with high-EW (equivalent width) BAL troughs to be shape-morphed while ensuring that all spectra are processed in the same way. The resulting reconstructions capture the overall large-scale shape of the spectra. The continuum in the reference model and each individual reconstruction is then interpolated across the locations of the strongest emission lines. The ratio of the reconstruction continuum and reference model is calculated and median filtered using a 601-pixel (≃250 Å) window. The original spectrum is then multiplied by the resulting smooth function (see top two rows of Fig. 4 for examples of the morphing procedure). ∼76 per cent of spectra are modified by factors of less than ±10 per cent over the full wavelength range.

Figure 4.

A non-BAL quasar spectrum (left) and BAL quasar spectrum (right) illustrating, from top to bottom, the main stages in the MFICA reconstruction procedure. The two quasar spectra both show significant differences in shape compared to the majority of the quasar population. The original spectra in the top panels (blue) are reconstructed using the BAL masking routine and the MFICA components that account for large-scale spectrum shape information. The initial BAL mask (dotted blue) and the iteratively produced final BAL mask (solid orange) are also shown, along with the NAL mask in purple. From the reconstructions in the top panels (red), the morph arrays are produced (green, middle panel; see text for details). The original spectra are multiplied by the morph arrays to produce the black spectra in the middle panels that now have the same overall spectral shape as the master quasar template spectrum. The morphed spectra are then fitted with MFICA-derived components to reconstruct the 1260–3000 Å wavelength interval. The morphed spectra (black), component reconstructions (red), and final BAL and narrow absorption mask (orange and purple, respectively) are shown in the bottom panels. It is from such reconstructions that all emission-line and continuum-derived parameters are calculated.

Open in new tabDownload slide

4.2 MFICA component generation

Samples of morphed spectra are used to calculate the components to be employed in the definitive spectrum reconstructions. First, bad pixels identified in the SDSS reduction pipeline are masked along with strong sky lines (e.g. the 5578 Å [O i] emission) in the blue. Narrow absorption features are also masked by applying a 61-pixel median filter to define a pseudo-continuum and masking the pixels where the spectrum lies below this pseudo-continuum by at least 3σ, where σ is the spectrum noise. Three pixels blueward and redward of each such pixel are also masked.

The MFICA components are calculated (see Allen et al. 2013) using a sample of ≃4000 quasar spectra with complete coverage of the 1260–3000 Å wavelength range⁴ and without any BAL troughs. Linear combinations of the resulting 10 components reproduce ≃99 per cent of the non-BAL quasars, with average fractional errors, (spectrum − reconstruction)/reconstruction, less than 2 per cent. Systematic deviations are greatest for quasars with the most extreme C iv emission EWs. For the purpose here, the form of the MFICA components is not important; the goal is simply to reproduce the quasar SEDs accurately. Two further sets of components were therefore generated using ≃2000 spectra with C iv(EW) > 40 Å (7 components⁵) and ≃2000 more with C iv(EW) < 20 Å (10 components) specifically to allow more accurate reconstructions of spectra with extreme emission-line properties. Each quasar with weak or strong C iv(EW) – measured from an initial reconstruction using the standard set of components – is fitted with the corresponding extreme C iv components, and from the standard and the extreme reconstructions, the best is chosen based on a χ² measurement.

4.3 Fitting the whole quasar sample

The routine for fitting the MFICA components utilizes the Python package lmfit (Newville et al. 2014) for non-linear least-squares minimization using the Levenberg–Marquardt algorithm with priors. The priors adopted for the component weights are based on the strong correlation between the morphology of the C iii] complex and that of the C iv line (see Appendix  A for details). Reconstruction of the intrinsic C iv emission when much of it has been absorbed or masked is greatly improved by the use of the priors on the component weights (see discussion later). Fig. 4 illustrates the full fitting procedure.

In addition to the shape morphing and masking of the NALs, in order to successfully fit BAL quasars with the MFICA components, it is crucial that the wavelength regions of the spectra that are affected by BAL troughs are masked. For HiBALs, these are regions blueward of 1400 Å for Si iv + O iv and 1550 Å for C iv. Troughs differ greatly in velocity width and position (up to thousands of  km s⁻¹) across BAL quasar spectra and to avoid masking more or less of each spectrum than necessary an optimal mask for each spectrum was defined in a similar manner to Allen et al. (2011). The routine also takes into account absorption redward of the BAL regions presented above resulting from errors in redshift. The steps applied to the morphed spectra, with flux F_λ and noise σ, are as follows:

• The initial mask covers regions of the spectra just blueward of the peak of the emission lines, where troughs are likely to appear: 1295–1400 Å for Si iv + O iv, 1430–1546 Å for C iv, and 1780–1880 Å for Al iii.

• The components are then fitted to the unmasked parts of the spectrum.

• A new mask is created by considering every pixel in turn. For pixel cpix, if the majority of the pixels in [cpix − 30, cpix + 30] have (reconstruction − spectrum F_λ) > Nσ, then pixel cpix is masked. Initially, N = 2.

• Once the pixels have been identified, 10 extra pixels (∼4 Å at 1550 Å) are masked blueward and redward of each of the masked regions.

• The BAL troughs associated with the Si iv + O iv λ1400 emission are weaker than those associated with C iv λ1550. For much of the quasar redshift range, the troughs lie at wavelengths where the spectrum S/N is lower and the masking algorithm is not always fully effective. To ensure absorption associated with Si iv + O iv is masked effectively, the masked region from C iv in velocity space is applied to the Si iv region, in addition to pixels masked in the iterative process.

Steps (ii) through (iv) are repeated until the mask converges, normally after two or three iterations. When there is no mask convergence after 10 iterations, the process is started from scratch at step (i) with N = N + 0.25 up to N = 4, at which point the spectrum is flagged as not having converged.

When fitted without masking, 97 per cent of the 7000 non-BAL quasar spectra had f(EW) < 10 per cent and a median f(EW) of only 2 per cent. The investigation confirmed the importance of employing priors on the component weights (see Appendix  A). Unsurprisingly, when the reconstructions are relatively poorly constrained, due to extended masked regions or low spectrum S/N, the use of priors for the component weights is essential. f(EW) is anticorrelated with S/N but only rises to a median of 6 per cent for S/N ≃ 5.

When the spectra were masked to simulate BAL quasars but the priors were not implemented, only 78 per cent of the sample had f(EW) < 10 per cent even though the median f(EW) increased to only 4 per cent. Upon employing the priors to fit the masked non-BAL quasar spectra, the percentage of spectra with f(EW) < 10 per cent increased to 85 per cent and maintained a low median f(EW) of 4 per cent. Statistically, the average gain of using the priors appears small for the sample as a whole. However, for the relatively small fraction of spectra with extended masks (corresponding to quasars with large BI values), the reconstructions improve significantly. For the actual BAL spectra, in the iterative process of fitting the components and defining the mask, the component weight priors prevent a bias towards narrower BAL masks and the presence of ‘dips’ in the reconstructions where the troughs are present. The use of priors derived from the properties of the C iii] complex emission is predicated on the assumption that the relationship between the C iv emission properties and the C iii] complex emission is the same for both BAL and non-BAL quasar populations. While, by definition, it is not possible to test the assumption directly for individual BAL quasars, it is possible to perform a powerful test for the BAL population as a whole. The results, presented in Appendix  B, demonstrate that the assumption regarding the similarities of emission-line properties for the BAL and non-BAL quasar populations is true to very high accuracy. Visual inspection of the sparsely occupied regions at the extremes of the C iv emission EW versus blueshift space (Section 6), specifically at (i) large negative C iv emission blueshift (<−1200 km s⁻¹), (ii) small C iv(EW) (EW < 10 Å), and (iii) the region at the bottom left of the C iv emission space, with relatively high negative C iv emission blueshift and low C iv(EW) (⁠|$\log _{10}\rm{C \,{\small {iv}}(EW)} \lt -2.3077\times 10^{-4}\times \rm{C\,{\small {iv}}(blueshift)}+1.3231$|⁠), reveals a number of intrinsically pathological quasar spectra and some reconstructions that are clearly suboptimal. The number of spectra in all three regions is small: 3027 spectra or 2 per cent of the quasar sample. 2283 of the spectra removed in this way are classed as non-BAL quasars by the trough parametrization code (see Section 5) but upon visual inspection include a significant fraction of LoBAL and FeLoBAL objects and their classification (and large, negative blueshift) is often due to insufficient masking. Spectra in these regions are therefore excluded from subsequent analysis. A further 189 spectra were excluded because the reconstruction scheme failed to converge due to poor morphing or the reconstructions had reduced χ² > 2. The excluded spectra were distributed evenly across C iv emission space.

For the remaining 140 448 quasars, each spectrum has the same overall spectral shape and a continuum plus emission-line reconstruction is available. An associated mask array contains the wavelengths of pixels where absorption has been identified. Quantification of the absorber properties and the definition of subsamples of BAL quasar spectra is now possible.

5 DEFINING THE BAL QUASAR POPULATION

The DR14Q catalogue contains measurements of BI, but the updated redshifts used in this investigation as well as our own reconstructions mean some spectra change classification based on the BI and AI values. We remeasured the BI and AI using python code written by Rogerson (2019) with minor modifications, allowing us to use our reconstructions to produce the normalized spectra, f(V), in equations (2) and (3). Each pixel in the normalized spectra has a value calculated using the inverse variance weighted flux within a 5-pixel window, thereby reducing the effect of single-pixel noise spikes. The code also calculates a variety of individual absorption trough measurements, including minimum and maximum ejection velocities and trough depth. Discussion of the additional trough measurements is included in Section 6.4.

We compare our BI values with those in the DR14Q catalogue in Fig. 5. The redshift changes have primarily affected narrow troughs near to the minimum (3000 km s⁻¹) and maximum (25 000 km s⁻¹) velocity boundaries used for the BI measurements; more or less of the absorption is shifted into the BI velocity window. In particular, if the C iv emission is considered when calculating the redshifts of BAL quasars with abrupt onset of absorption in the blue half of the line, these quasars can have systematically higher redshifts in SDSS. Our statistically lower redshifts lead to less of the absorption appearing at velocities >3000 km s⁻¹ and we can calculate lower or even zero BI.

Figure 5.

Comparison of the BI_DR14Q values with our BI values. Top: Counts of BI_DR14Q and BI and AI calculated for this paper. Bottom: ΔBI = BI − BI_DR14Q. For visualization purposes, we have removed BI = 0 or AI = 0 quasars from each of the histograms. In this work, the calculated BI values are, on average, 700 km s⁻¹ larger than those in DR14Q. Below 1000 km s⁻¹, there is a decrease in the number of BI-defined BAL quasars according to our measurements and compared to the DR14Q values, which we believe is a result of changes in redshift. Above 1000 km s⁻¹, there are more BI-defined quasars due to higher continua in our spectrum reconstructions.

Open in new tabDownload slide

Differences in the continuum level of the component reconstructions can also result in classification switches. Here, changes result as the contribution of absorption to the BI calculation depends sensitively on the absorption depth threshold, f(V) < 0.9, continuously for 2000 km s⁻¹. In Fig. 6, we illustrate these effects by considering the 2359 spectra for which DR14Q listed zero BI, whereas we measure BI > 0. As a result, using the BI definition we gain 1503 BAL quasars, while the 2359 objects in Fig. 6 become non-BAL quasars.⁶

Figure 6.

Our BI values against our calculations of the minimum ejection velocity with points coloured by trough depth for the 2359 spectra with BI > 0 but zero BI_DR14Q. Most of these spectra have minimum velocities close to the 3000 km s⁻¹ cut-off where redshift differences are responsible. The spectra with large V_min have shallow troughs and changes in classification due to continuum reconstructions dominate.

Open in new tabDownload slide

We exclude 1668 LoBAL quasars from our sample (in addition to those in outlying regions of C iv emission space), defined by a positive BI₀ of Al iii troughs where BI₀ is a modified BI, where the integral in equation (2) starts at 0 km s⁻¹. Additional 18 FeLoBAL quasars are also removed from the sample by means of the BAL mask (Section 4.3): spectra with more than one-third of pixels redward of C iv masked are flagged as potential FeLoBAL quasars. The final sample of quasars thus numbers 138 762, with 73 806 of those defined as non-BAL quasars, 14 887 classically defined BAL quasars (positive BI), and further 50 069 with positive AI but a BI of zero.

5.1 Two AI populations

Upon plotting the distribution of AI in log space (Fig. 7), we observe two populations not unlike Knigge et al. (2008, see their fig. 1), although note that those authors adopt the AI prescription of Trump et al. (2006) that includes only absorption wider than 1000 km s⁻¹ and extends the 25 000 km s⁻¹ maximum velocity to 29 000 km s⁻¹. The large-AI population of quasars is dominated by spectra with positive BI (AI(BI > 0)), whereas the classically defined non-BAL quasars dominate the low-AI population (AI(BI = 0)). The distributions of BI and AI(BI > 0) overlap since many of the spectra, thus same trough measurements, appear in both samples. The shift of the AI(BI > 0) towards larger values as compared to the BI values is a result of the minimum velocity cut-off on the BI. The number of quasars in the AI(BI > 0) population is also lower than the number in the BI population due to the |$\chi ^2_{\text{trough}}\geq $| 10 requirement of the AI troughs. The properties of the AI(BI = 0) and AI(BI > 0) samples are discussed later in Section   6.4.

Figure 7.

AI distribution for quasars with AI > 0 (black). The AI population is divided into quasars with zero BI (blue; left column of Fig. 16, where we discuss the trough parameters) and non-zero BI (orange; middle column of Fig. 16). The form of the AI distribution suggests strongly that two populations of absorbers contribute. The BI distribution for quasars with BI > 0 is also plotted (red, hatched; right column of Fig. 16), which overlaps with the AI(BI > 0) population since many of the same spectra appear in both populations.

Open in new tabDownload slide

6 RESULTS

6.1 C iv emission line profile

Figure    8 shows the C iv EW versus the C iv blueshift (hereafter the C iv emission space) for the samples of non-BAL and BAL quasars. Let us first consider the non-BAL quasar population: consistent with Richards et al. (2011), weak lines (low EW) are often highly blueshifted, providing strong evidence of outflows, while strong lines are close to symmetric (zero blueshifts). As Richards et al. (2011) point out, however, the relationship is not one to one as there are quasars with low blueshifts and low EWs. Note also that non-BAL refers here to quasars with both AI = 0 and BI = 0, i.e. no absorption present. If, however, the AI > 0 quasars (still BI = 0) are included, the non-BAL C iv distribution does not in fact change significantly. Composite non-BAL quasars constructed from spectra in three regions of C iv emission space are presented in the top panels of Fig. 9. In detail, a high-EW composite is generated from spectra with C iv EWs and blueshifts of 103–116 Å and 262–787 km s⁻¹, respectively; a low-EW, low-blueshift composite from 23–29 Å and −262 to 262 km s⁻¹; and a low-EW, high-blueshift composite from 23–29 Å and 2360–2890 km s⁻¹. The spectra are smoothed using inverse variance weighting over a 7-pixel window. The close to constant pixel-by-pixel median absolute deviation (MAD), shown as the shaded regions, redward of C iv is almost entirely due to the finite S/N of the spectra.⁸ Such composite spectra, covering the full occupied region of the C iv emission space, are available to view in an interactive plot online.⁹

Figure 8.

C iv emission space for the sample of non-BAL quasars (blue contours/dots) and classically defined BAL quasars (red contours/dots). Marginalized distributions of blueshift and EW are also shown. The faint points in the sparsely populated lower left of the figure have been excluded from the analysis (Section 4.3). The two populations are distributed very similarly within the C iv emission space except at high EW and low blueshift, where there is a notable lack of BALs at the lowest blueshifts.

Open in new tabDownload slide

Figure 9.

Composite spectra and reconstructions of non-BAL (top) and BAL quasars (bottom) for select regions in C iv emission space. Left: High-EW and low-blueshift C iv; middle: low-EW and low-blueshift C iv; right: low-EW and high-blueshift C iv. The number of spectra contributing to each composite is indicated in the top left. In the top panel of each pair, the black line is the median composite SDSS spectrum and the red line is the composite reconstruction. The grey and red shaded regions mark the pixel-by-pixel MAD of the spectra and reconstructions, respectively. The lower panel of each pair presents the composite spectrum normalized by the composite reconstruction. The shaded region is the MAD of the normalized spectra. The vertical, dashed, grey line corresponds to the rest wavelength of C iv and the horizontal, solid grey line in the normalized BAL quasar panels marks f(V) = 0.9, the flux below which BAL troughs contribute to BI and AI calculations. Note that all panels containing the normalized spectra have the same y-scale for comparison. While the C iv emission profiles vary in C iv emission space by design, the BAL troughs also vary systemically. Composite non-BAL and BAL spectra are available to view in an interactive plot of the C iv emission space.⁹

Open in new tabDownload slide

The full spectrum reconstructions for the BAL quasar population allow a direct comparison of the non-BAL and BAL quasar populations within the C iv emission space (Fig. 8) and the bottom panels of Fig. 9 contain composite BAL quasars taken from the same areas in the C iv emission space as the non-BAL quasar composites above. The same C iv emission line shapes are apparent as for the non-BAL quasars. The BAL quasar composites show systematically different BAL trough properties moving around the C iv emission space (see Section 6.4). Again, an interactive plot is available for the BAL quasars.⁹ The increase in the pixel-by-pixel MAD (shaded regions) within the troughs highlights the diversity in the absorber properties at a given velocity. The diversity of troughs is particularly evident in spectra with weak and blueshifted C iv emission (lower right region of Fig. 8), an observation made in an important early paper discussing BAL quasar properties by Turnshek (1988). We find that averaging over spectra possessing relatively deep individual troughs at different velocities results in the composites showing weak but broad BAL troughs. The kinematics and properties of BAL troughs are explored more fully in Section 6.4.

As discussed in Section 3, the probability that a BAL quasar is included in the DR14 catalogue as a function of SED, magnitude, and redshift is not available. The level of uncertainty in the spectrophotometry of the DR14 quasar spectra also makes consideration of the amount of reddening experienced by the BAL quasars far from straightforward. As a consequence, the quantitative assessment of the intrinsic frequency of BAL quasars as a function of their BAL trough properties and reddening is not possible (cf. Allen et al. 2011, particularly fig. 22). Nevertheless, it is instructive to review whether the observed fraction of BAL quasars in the sample changes across the C iv emission parameter space.

Figure 10.

The observed fraction of quasars that are classically defined BAL quasars as a function of the C iv emission properties. Only bins with 10 or more quasars are plotted. The grey hexagons indicate the sparsely populated region of C iv emission space excluded from analysis. As discussed in the text, the intrinsic fraction of BALs is not known; however, at fixed EW, the probability a quasar possesses BAL troughs increases with increasing blueshift suggesting that the fraction of quasars with signatures of outflows in absorption increases with increasing evidence of outflows in emission. It is true that the fraction of quasars that are BALs, at a particular location in C iv emission space, is highest at high EW and positive blueshift; however, these quasars have low BIs and the relative number of BAL quasars here is low compared to those at higher C iv blueshifts and lower EWs, as highlighted by the marginal distributions (same as for Fig. 8).

Open in new tabDownload slide

6.2 C iv and other UV emission lines

The composite spectra allow us to compare the BAL quasars to the non-BAL quasars in different regions of C iv emission space. Figure    11 shows the composites in Fig. 9 over the wavelength range of 1260–3000 Å. The BAL and non-BAL composite reconstructions in the top panel of Fig. 11 are extremely similar across the whole wavelength range. The composite spectra themselves are also very similar, apart from the presence of troughs in the BAL spectra, in agreement with Weymann et al. (1991) and Matthews, Knigge & Long (2017). There are, however, subtle differences in some of the emission lines at certain areas in C iv emission space that can only be seen when dividing the BAL spectra (reconstructions) by the non-BAL spectra (reconstructions) as shown in the bottom panel. Note that the y-scale in the bottom panel has been chosen to illustrate low-amplitude spectral differences.

Figure 11.

Composite spectra and reconstructions of the quasars in and surrounding the three EW and blueshift regions used to produce Fig. 9 (top panel). The size of the areas in EW and blueshift has been extended to increase the number of quasars contributing to each composite. The BAL quasars follow the colours in the legend, while the non-BAL quasars are plotted in black. The composite BAL (dot–dashed, coloured lines) and non-BAL (dot–dashed, black lines) quasar spectra are also plotted. Almost identical composites and reconstructions, such that the dot–dashed spectra are hardly visible underneath the reconstructions, mean the reconstructions are capturing the pertinent features in the spectra. The high-EW, low-blueshift composites have been shifted up by 1.5 and the low-EW, low-blueshift composite shifted up by 3 for presentation purposes. The bottom panel contains the BAL quasar composite spectra (reconstructions) divided by the non-BAL quasar composite spectra (reconstructions) in black (legend colours). Note that the y-scale has been chosen to emphasize even small differences between spectra (reconstructions). The high-EW composite and low-EW, low-blueshift composite have been shifted up by 0.6 and 0.3, respectively. Identical BAL and non-BAL quasar composite spectra (reconstructions) would produce divided spectra (reconstructions) of unity.

Open in new tabDownload slide

Consider first the direct empirical differences evident in the BAL quasar/non-BAL spectrum ratios. The BAL absorption blueward of the C iv and Si iv emission is most evident as expected. All three composites also show a systematic difference in Mg ii λ2800 emission with the BAL possessing weaker emission, although the reduction is small. The difference has been seen in other investigations (see fig. 8 in Baskin et al. 2015, for example) but is particularly evident here. Otherwise, the two composites drawn from the lower C iv(EW) regions show no spectral differences greater than ≃2 per cent shortward of the C iv emission line.

The composite from the high C iv(EW) and low C iv blueshift region does, however, possess features coincident with wavelengths of prominent emission lines: peaks are present in the BAL quasar/non-BAL spectrum ratio at O i + Si ii, Si iv, C iv, He ii, O iii], and C iii]. The location of quasars within the C iv emission space means that the C iv(EW)s are the same and care has been taken to ensure that both the distributions of C iv(EW) and blueshift contributing to the non-BAL and BAL composites are carefully matched. The peak differences in emission-line strength are not large, a maximum of 10 per cent, but are highly significant. Inspection of the ratio spectrum reveals systematic reductions in the BAL spectrum to the blue and red of several of the emission-line core excesses. The BAL quasar emission lines are narrower, in full width at half-maximum (FWHM), than those of the non-BAL quasars by ≃14 per cent. The origin of the emission-line differences is therefore due to the BAL spectra possessing slightly narrower emission-line profiles than the non-BAL spectra. As far as we are aware, the observation represents a new result. In the low-EW, high-blueshift composites, the converse is true but with less significance: the BAL quasar C iv emission lines are wider in FWHM by ≃7 per cent.

Moving to the BAL/non-BAL reconstruction ratio spectra, the effectiveness of the reconstruction scheme can be assessed via comparison with the features evident in the BAL/non-BAL spectrum ratios, in particular the differences in emission lines between the two populations. If the reconstructions were perfect, the ratio spectra would essentially mimic the features seen in the spectrum ratios, other than at wavelengths affected by BAL troughs. Such is very definitely not the case across the Mg ii λ2800 line. The reason, however, is the Baldwin effect, i.e. the anticorrelation between emission-line EW and luminosity (Baldwin 1977), which we have chosen not to remove. In more detail, the MFICA components were generated using quasars with redshifts |$z$| = 2.0–2.3 (Section 4.3), thereby defining the relative emission-line strengths for the UV and Mg ii lines. The quasars with the wavelength coverage of Mg ii emission that contribute to the composites in Fig. 11, however, have a significantly lower mean redshift, hence lower luminosity and thus higher emission-line EWs. The reconstruction emission-line strength for each quasar is dominated by the C iv emission and other lines in the far-UV. The MFICA components possess the flexibility to reproduce the extended range of emission-line EW at any redshift, but, statistically, the higher emission-line dominated MFICA component contributions to the lower redshift objects result in an overprediction of the Mg ii emission. For the two low-EW composites in Fig. 11, the BAL and non-BAL UV emissions have the same strength and the reconstructions across the Mg ii line are identical. For the high-EW composite, the BAL UV emission is stronger by 10 per cent and the BAL quasar Mg ii emission is thus also stronger, producing the visible ‘peak’ at the location of the core of the Mg ii line in the reconstruction ratio.

The Baldwin effect induced mismatch to the Mg ii emission aside the reconstruction ratio spectra closely follow the spectrum ratios. Away from the Si iv and C iv BAL troughs, the reconstructions are within 2 per cent of the actual composite spectra. The reproduction of the individual emission excesses for the BAL quasars in the top composite is striking. The similarities between the composite ratio spectra and the composite ratio reconstructions for all three regions in the C iv emission space highlight the effectiveness of our reconstruction scheme.

High He ii EW is indicative of a stronger soft X-ray spectrum (Leighly 2004). Baskin et al. (2013, 2015) noted the decrease in He ii λ1640 strength with increasing C iv blueshift by means of BAL and non-BAL quasar composites. With our reconstructions, we have been able to consider the relationship between the C iv and He ii on an object-by-object basis. We calculate the He ii λ1640 EW between 1620 and 1650 Å (identical to Baskin et al. 2013) following the same procedure as for the C iv doublet but using the 1610–1620 and 1700–1705 Å windows for the power-law estimation. Our observations are in agreement with Baskin et al. (2013, 2015) and the hypothesis that disc winds form only when the ionizing quasar SED is soft enough that electrons remain bound to nuclei and radiation line driving contributes to the acceleration of material. The dependence of He ii EW as a function of location in C iv emission space confirms the anticorrelation between the hardness of the ionizing SED and the strength of outflows seen in both absorption and emission. A new result is the demonstration that the systematic He ii EW trends for the non-BAL, AI-, and BI-defined BAL quasar populations (Fig. 12) are close to identical.

Figure 12.

C iv emission space with hexagons coloured by median He ii EW for the non-BAL quasars (left), AI-defined quasars (middle; black histogram in Fig. 7), and BI-defined BAL quasars (right; red, hatched histogram in Fig. 7). Only hexagons with ≥5 quasars are plotted. The spectra in the lower left of C iv emission space that were excluded and previously plotted as grey hexagons (Fig. 10) are not plotted here. The less extensive C ivcoverage in the BI sample is a result of fewer quasars in this population. The He iiEW distributions are extremely similar in all three populations.

Open in new tabDownload slide

6.3 Black hole mass and luminosities

The bolometric luminosity, L_bol, has been calculated from the monochromatic luminosity at 3000 Å or at 1350 Ź¹ for spectra where 3000 Å is not present, using the bolometric corrections BC₃₀₀₀ = 5.15 and BC₁₃₅₀ = 3.81 from Shen et al. (2011). L_bol is displayed in Fig. 13 as a function of location in C iv emission space for the non-BAL, AI-, and BI-defined quasar populations. In agreement with the Baldwin effect (Baldwin 1977), albeit over a smaller dynamic range of luminosity, there is an anticorrelation between L_bol and C iv EW at fixed blueshift. L_bol also increases towards the high-blueshift end of the C iv emission space in all three populations. The three populations have very similar luminosities (median log₁₀(L_bol) = 46.39, 46.57, and 46.54 for the non-BAL, AI and BI quasars, respectively). Consideration of the intrinsic luminosities of the populations would require accurate extinction estimates for the quasars that are not available. It is, however, likely that the quasars with absorbers possess modest E(B − V) values (see Allen et al. 2011, section 8.5). Adopting an average E(B − V) = 0.05 for BAL quasars (from Allen et al. 2011), BAL quasars are predicted to appear 25 per cent fainter at 3000 Å and 46 per cent fainter at 1350 Å. The reference wavelength transition occurs at |$z$| ≃ 2.3 when 3000 Å rest frame moves beyond the red limit of the SDSS spectra. The intrinsic luminosities of the BAL quasar samples should thus be larger by ≃0.12 (⁠|$z$| ≤ 2.3) and 0.27 (⁠|$z$| ≥ 2.3) in log₁₀(L_bol). The factors are modest relative to the dynamic range in luminosity present in the quasar samples and should not affect the trends within the C iv emission space. Again, the systematic trends within the C iv emission space for all three populations closely mimic each other.

Figure 13.

As for Fig. 12, with hexagons occupied by five or more quasars coloured by median L_bol. Similar systemic trends within the C iv emission space are present in all three populations.

Open in new tabDownload slide

Only C iv λ1550 emission is present in all the quasar spectra; thus, we derive back hole mass estimates from the C iv line and employ the correction of Coatman et al. (2017) to account for the excess, non-virial, blue emission for quasars with C ivblueshift >500 km s⁻¹. The Coatman et al. (2017) correction is not well defined for modest positive blueshifts or negative blueshifts of any size. As a consequence, the mass correction has not been applied where the C iv blueshift <500 km s⁻¹.

From the black hole mass estimates, we compute the Eddington luminosity, L_Edd, and thus the Eddington ratio using L_bol from above, which we present as a function of C iv emission space in Fig. 14. Since the Eddington ratio depends on the Eddington luminosity that, in turn, depends on the black hole mass, we have only plotted the Eddington ratio for spectra with C iv blueshift >500 km s⁻¹. The Eddington ratios of the non-BAL, AI-, and BI-defined quasars are similarly distributed in C iv emission space, indicating that quasars with evidence for the strongest outflowing winds based on their emission-line properties have the highest L/L_Edd.

Figure 14.

As for Fig. 12, with hexagons occupied by five or more quasars coloured by median Eddington ratio. We only plot the Eddington ratio for spectra with C iv blueshift >500 km s⁻¹ for which we correct the black hole masses using Coatman et al. (2017). The Eddington ratio increases with C iv blueshift for all three populations but is slightly lower at fixed C iv parameters for the non-BAL population compared to the AI- and BI-defined samples (see also Fig. 13).

Open in new tabDownload slide

6.4 BAL trough parameters

In Fig. 9 and the interactive plot,⁹, we see the systematic trend in BAL trough properties with location in C iv emission space. Strong and symmetric C iv is often accompanied by relatively narrow and deep troughs, while spectra with weak and blueshifted emission have broader and more diverse troughs. To investigate any systematic relationships between absorber trough and emission-line properties, various trough parameters are measured. Each trough is defined as a region where the BI (or AI) conditions are met. Table 1 lists the parameters and they are illustrated using an example spectrum in Fig. 15. Where spectra have multiple troughs, the parameters pertaining to individual trough measurements are reported as follows: we report the minimum V_min (i.e. the lowest velocity for which there is broad absorption), the maximum V_max (i.e. the highest velocity for which there is broad absorption), the deepest trough F_depth measurement and the accompanying V_Fdepth, and the widest trough V_width.

Figure 15.

Example BAL quasar spectrum and reconstruction (top panel) alongside the normalized spectrum (bottom panel) in velocity space, inverse variance weighted with a window of 5 pixels. The grey, dotted vertical lines mark the minimum and maximum velocities considered in the calculation of BI and thus the trough parameters. The dashed, grey line at zero velocity marks the laboratory wavelength of C iv. The measured trough parameters for this spectrum are marked in the lower panel (see Table 1 for descriptions of each parameter). There are two troughs in this BAL quasar spectrum (shaded regions) with the lower velocity, blue trough up against the minimum allowed velocity, although significant absorption at velocities below 3000 km s⁻¹ exists.

Open in new tabDownload slide

In Fig. 16, we present the AI and BI trough parameters as a function of location in C iv emission space. Considering the two populations of AI > 0 quasars (Fig. 7), we split the AI trough parameters into the two populations: AI troughs in spectra with BI = 0 (AI(BI = 0), left-hand panels; blue histogram of Fig. 7) and AI parameters for spectra with BI > 0 (AI(BI > 0), middle panels; orange histogram of Fig.   7). The right-hand panels contain the BI trough parameters. By definition, almost all of the spectra with BI > 0 have AI > 0 and so appear in the middle and right-hand panels.¹² The extent in velocity of the absorption troughs therefore increases systematically moving from left to right in Fig. 16.

Figure 16.

The AI- and/or BI-defined quasar populations binned in C iv emission profile space where each hexagonal bin is coloured by trough parameters outlined in Table 1. Left: AI trough parameters for spectra with BI = 0; middle: AI trough parameters for spectra with BI > 0; right: BI trough parameters for classically defined BAL quasars. The quasars have been binned in C iv emission space and the median trough parameter in each bin plotted on the colour axis, for bins with at least five quasars. The number of spectra contributing to each panel is noted in the top right of every panel; each row has the same number of spectra. From top to bottom, AI or BI, F_depth and V_Fdepth. The figure continues overleaf with V_min, V_max, ΔV, and V_width. The dot–dashed lines in the lower right panel locate the boundaries in C iv emission space used to produce the absorber rest-frame composites in Fig. 17. The structures of both the AI- and BI-defined troughs vary systematically, but differently, in C iv emission space.

Open in new tabDownload slide

Figure 16

– continued

Open in new tabDownload slide

Figure 17.

Median absorber rest-frame composites of the 14 887 BI-defined quasar spectra. Where spectra have more than one trough, the deepest trough is selected. The numbering of the composites matches that of the C iv regions in the lower right panel of Fig. 16; however, only the odd-numbered composites are plotted for clarity. The BI-defined troughs extend to larger velocities at higher C iv blueshifts with the maximum velocity increasing more rapidly than the minimum velocity or the velocity of the deepest part of the trough. The depth of the troughs does not vary systematically in C iv emission space.

Open in new tabDownload slide

From first glance at Fig. 16, the AI(BI = 0) and AI(BI > 0) trough systematics differ, while the latter are almost identical to the BI trough systematics, reinforcing the evidence for the existence of two AI populations described by Knigge et al. (2008) and presented in Fig. 7. For spectra in the middle panels with only one AI trough, the AI trough parameters are identical to the BI trough parameters in the right-hand panels except where V_min,450 < 3000 km s⁻¹. It is also possible for there to be more than one AI trough per spectrum in the middle panels; however, only one of these AI troughs has to satisfy the BI conditions for the object to be included in the right-hand panels; thus, the AI and BI trough parameters will differ. Approximately 54 per cent of the spectra in the middle panel have more than one trough.

In the top row of panels, we see the dependence of absorption outflow strength within the C iv emission space, via the AI and BI measures. Clear differences exist between the AI(BI = 0) and the BI troughs. While the BI troughs become stronger with increasing C iv emission blueshift, the AI(BI = 0) troughs show this same trend at fixed high EW, but at low EW the AI measures decrease for large C iv blueshifts.

The left-hand panel of the second row, showing the dependence of trough depth, explains the AI(BI = 0) behaviour in the top row – i.e. at log₁₀ C iv(EW) ≲ 1.6, the troughs become shallower as C iv blueshift increases, and thus AI also decreases (equation 3). The opposite is true at high C iv EW. If there is any such systematic present for the BI-defined troughs, it is far less significant, in contrast to what is seen in the composite troughs in Fig. 9. The weakening of the troughs in the composite spectra is a result of averaging out the individual troughs that are more diverse in velocity structure at higher C iv blueshift and lower EW. The comparable depths of the troughs are clear in Fig. 17, which shows absorber rest-frame composites of the BI-defined BAL spectra in different regions of C iv emission space. Baskin et al. (2015) presented similar absorber frame composites and found trough depth not to be correlated with He ii EW (see their fig. 7), in agreement with what we present here in C ivemission space.

Figure 18.

Mean normalized composite of 18 740 single, AI(BI = 0) troughs shifted to the rest frame of the deepest part of the trough. The composite is dominated by the well-defined deep, narrow absorption in which both features in the C iv λλ1548,1551 doublet are seen. Additional weaker absorption is present in some troughs, explaining the broader depression, particularly to shorter wavelengths.

Open in new tabDownload slide

The trends evident in the third, fourth, and fifth rows, all showing measures of the absorber trough velocities, are, however, far more similar. For both AI- and BI-defined troughs, the minimum and maximum velocities where absorption is present and the velocity of maximum absorption depth all show strong systematics as a function of C iv blueshift. As the strength of the emission outflow signature increases, the velocities of the absorbers increase. Very few objects have V_Fdepth at very large velocities (even at high C iv blueshift) – i.e. the deepest part of each trough often occurs in the lower velocity half of the trough – and this can also be seen in Fig. 17. Hamann et al. (2019, see their fig. 4) observed these skewed absorption profiles in composites binned by BI. For both AI- and BI-defined samples, the highest velocity troughs are in quasars with the highest Eddington ratio as observed in Fig. 14.

The sixth and seventh rows, examining the behaviour of the velocity extent (ΔV) and width (V_width) of all and individual absorbers, respectively, within the spectra, provide insight into the systematic differences between the AI(BI = 0) and BI-defined absorbers. For the BI-defined absorbers, as the C iv emission blueshift increases, the trough velocities increase and the trough velocity extent increases, as would be expected for an accelerating wind. The BAL troughs present in spectra in the lower right of C iv emission space are much broader than those in spectra at the upper left. However, for the AI(BI = 0) absorbers, the velocities increase as the C iv blueshift increases, but the velocity extent of the absorbers is close to constant. 16 390 AI-defined quasars have multiple troughs and in many they are separated extensively in velocity, up to 24 950 km s⁻¹. In the AI(BI = 0), AI(BI > 0), and BI quasar populations, 67, 46, and 83 per cent of the quasars, respectively, have only one trough. Spectra with more than one trough typically have large trough velocities to allow for a second trough at lower velocities and thus have large C iv blueshifts. Many of the AI(BI > 0) quasars have a BI-defined trough (therefore also AI-defined) and a second AI-defined trough, accounting for the high fraction of spectra with N₄₅₀ > 1.

The AI(BI = 0) spectra are dominated by single, relatively narrow, troughs such that BI = 0. These narrow troughs appear in spectra with the full range in C iv blueshift as evidenced by the swathe of constant and low AI, ΔV₄₅₀, and V_width,450 in the AI(BI = 0) plots. We can test the hypothesis that the AI(BI = 0) absorbers are different from the AI(BI > 0) absorbers rather than just much narrower by investigating the structure of the zero-BI troughs. We select 18 740 AI(BI = 0) quasars with only one AI trough and generate a mean composite of the spectra shifted to the rest frame of the maximum absorption depth. A well-defined C iv doublet is observed (Fig. 18). In many cases, the 450 km s⁻¹ minimum velocity required by the AI measurement is sufficiently narrow to include NALs with very different internal kinematic properties compared to BI-defined absorbers. The AI-defined absorption is participating in the outflows, but we conclude that AI(BI = 0) absorbers are dominated by material in structures with velocity spreads of a few hundred km s⁻¹, very different from much of the material giving rise to the classical BI-defined troughs.

We have yet to consider the high-EW, low-blueshift (≲1000 km s⁻¹) region of C iv emission space. Inspection of the ΔV distribution for the AI(BI = 0) troughs shows increased ΔV at high C ivEW. Some of these troughs have V_width,450 > 2000 km s⁻¹ but have BI = 0 as most of the absorption is below 3000 km s⁻¹. In contrast with the NALs discussed earlier, as the C iv blueshift increases, over an admittedly small range of velocities present at high EW, AI and V_width,450 increase while the troughs deepen (F_depth,450 increases). In other words, while the minimum velocity remains constant, the maximum velocity increases. The original BI calculation was designed with the 3000 km s⁻¹ starting velocity to exclude strong associated absorbers that are believed to result from a different phenomenon compared to the BAL troughs (Weymann et al. 1991). In the circumstances, the use of a BI₀ definition for absorbers (where the BI metric is extended down to zero velocity) will include a larger fraction of absorbers resulting from outflows but at the expense of including true ‘associated absorbers’.

7 DISCUSSION

The main aim of this paper is to present the observational differences and similarities between the UV spectra of non-BAL and BAL quasars, including the outflow signatures evident in the C iv λ1550 emission. Our approach has been designed to allow a direct investigation of the systematics observed in the absorption and emission outflow properties for the non-BAL and high-ionization BAL populations as a whole. In short, we can definitively locate BAL quasars in C iv emission-line parameter space, even in the face of significantly absorbed C iv profiles. The results should provide constraints on models, particularly in the case of disc winds, but we do not attempt to propose a quantitative model, deferring such considerations to a future paper.

7.1 Quasar physical properties

The observational manifestations of outflows observed in absorption and emission are quite different, particularly in terms of the solid angle over which material contributes to the spectrum. BAL and narrow absorption occurs due to gas present along a particular line of sight. For absorbing structures of limited physical extent, a small change in viewing angle may result in very different observed absorber properties. Similar changes may result temporally due to rapid motion of absorbing material across the line of sight or changes in the ionization parameter. Emission-line properties instead derive from gas covering a much larger solid angle from the perspective of the observer. Photons produced by emission, in gas participating in an outflow, over a significant angle contribute to the observed spectrum and small changes in viewing angle are not, in general, expected to result in significant changes to the emission-line spectrum. Observed emission-line changes due to larger angular variations in viewing angle are of course potentially powerful probes of non-spherical geometries, including disc wind models.

A main result is the similarity in the observed UV emission and physical properties between the BAL and non-BAL quasar populations. For every BAL quasar with a certain black hole mass, bolometric luminosity, Eddington ratio, C iv emission profile, and He ii λ1640 EW, there exists a non-BAL quasar with essentially identical properties, consistent with the two objects possessing the same orientation and/or geometry. Matthews et al. (2017) suggest that BAL quasars can be observed at similar inclinations to non-BAL quasars. The converse is not always true; while for the majority of the C iv emission space one can find a BAL quasar for each non-BAL quasar of specified properties, there is a lack of BAL quasars possessing large C iv emission EW, no significant C iv emission blueshifts, and hard SEDs.

Richards et al. (2011) used the observed relation between the morphology of the C iii]λ1908–Si iii]λ1892–Al iii λ1857 complex and the blueshift of the C ivemission¹³ to deduce the locations of BAL quasars within the C iv emission space. The conclusions of Richards et al. (2011) are consistent with the results presented here, but now far more information is available about the location of BAL quasars in the C iv emission space as a function of the BAL trough properties (see Fig. 16 in particular).

The distribution of quasars in C iv emission space (Fig. 8) can be explained by a disc wind model (e.g. Richards et al. 2011): the Doppler blueshifted emission is thought to occur in an outflowing wind component; thus, spectra with symmetric and high-EW C iv show little to no evidence of a disc wind and instead are dominated by disc emission. Quasars with extremely blueshifted emission are all weak lined. Evidence from the strength of C iv and He iiemission (see later) indicates that quasars at the top left of the C iv emission space have the hardest SEDs, with a systematic softening of the ionizing continuum moving towards the bottom right. In a scenario where line driving is important for producing the wind, only when the number of high-energy ionizing photons is small enough that electrons remain bound to nuclei can the wind be accelerated to high velocities. As a consequence, quasars with strong C iv emission and high blueshift are not found.

As a recombination line, He ii λ1640 can be used as an indicator of the number of ionizing photons with energies above 54 eV, i.e. as an SED hardness diagnostic. In the line-driven wind scenario, only the spectra with the lowest He ii EW, thus fewest high-energy ionizing photons, are able to produce the strongest disc winds. Baskin et al. (2013, 2015) have quantified the strength of the dependence of BAL trough properties as a function of the He ii emission EW (broader and faster moving troughs are observed when He ii is weak). The systematic behaviour of He ii emission as a function of C iv blueshift for non-BAL quasars has also been identified by Baskin et al. (2015). Again, our results (Fig. 12) for the strength of He ii emission in the BAL and non-BAL populations across the C iv emission space show striking systematic behaviour and demonstrate the very close similarity in properties for BAL and non-BAL populations.

Simple comparisons of emission-line properties for the BAL and non-BAL populations as a whole will show differences as a result of the lack of BAL quasars with hard SEDs and strong, symmetric C iv λ1550 emission. The dearth of BAL quasars at the top left of the C iv emission space (Fig. 8) has naturally led to the conclusion that BAL quasars are more common among the quasar population with softer SEDs. Our analysis indicates the situation is somewhat more complex as BAL quasars exist across the entire remainder of the C iv emission space occupied by the non-BAL quasars. While SED hardness is important, a stronger emission outflow signature at fixed C iv EW is also relevant for increasing the probability that a quasar is a BAL quasar (Fig. 10). Overall, a key conclusion from the results presented in Section 6 is that, when an outflow is present, i.e. C iv blueshift >0, the physical properties (including L, L/L_Edd, and SED hardness) and UV emission-line properties of the non-BAL and BAL populations are essentially indistinguishable.

7.2 Absorber properties

Knigge et al. (2008) were the first to propose the existence of two populations of absorbers among quasars with positive AI values. In our absorber classification, a physically significant difference is evident between the AI(BI = 0)-defined absorbers and those defined by AI(BI > 0). Row 5 of Fig. 16 shows how, independent of the AI or BI value of troughs, the maximum absorber outflow velocity increases with increasing C iv emission blueshift. Row 6, however, reveals a clear difference in the velocity extent of troughs, with AI(BI = 0)-defined absorbers possessing a constant velocity width (ΔV), while the AI(BI > 0) absorbers show a strong systematic trend of increasing ΔV with increasing maximum outflow velocity. The latter behaviour is consistent with outflowing absorbing material in an accelerating wind. The composite absorption spectrum of the AI(BI = 0) absorbers (Fig. 18) by contrast shows both components of the C iv λλ1548,1551 doublet are visible. The typical velocity spread within the absorber structures, independent of outflow velocity, must therefore be no more than ≃250 km s⁻¹.

It is in the high-EW region of C iv emission space where quasars have the hardest ionizing SEDs that we observe slightly narrower emission lines in the BAL quasars than the non-BALs (Fig. 11). If absorption is present in these spectra, it is typically deep and also narrower than the majority of the BI-defined troughs but is among the broader AI-defined troughs. The narrower emission lines and hard SEDs, combined with the differing trough systematics, could suggest a different orientation of the quasars or different outflow-driving mechanism (Richards 2012) from that of the outflows traced by the absorption seen in all other areas of C iv emission space. Alternatively, certain sightlines may probe different parts of the outflows, which may be dominated by different driving mechanisms.

The investigation presented here has focused on quasars with BAL troughs, while a study of the outflow properties of narrow C ivabsorption lines in non-BAL and BAL quasars was undertaken by Bowler et al. (2014). Results included the detection of line-locked absorbers but, most relevant to findings presented here, also showed that the kinematics and population statistics of narrow absorbers in the non-BAL and BAL quasar subpopulations were similar. Combining the results with those presented here leads to a picture where, in respect of outflow signatures, the non-BAL and BAL quasar populations possess the same emission and NAL outflow properties strongly suggesting a direct link between apparent ‘non-BAL’ and ‘BAL’ quasars.

While the results in Section 7.1 support the idea that BAL quasars have the same parent population as non-BALs (Richards 2006), it is clear that the properties of the absorption troughs change significantly across the C iv emission space, which is an observation that is expected from theory (see Giustini & Proga 2019). The observation may be important in the context of understanding the origin of ‘BAL quasars’ as population statistics, such as the fraction of quasars classified as BALs, are potentially a strong function of the BAL definition. Put another way, BAL quasars can be found across nearly the full C iv emission space but BAL trough properties change in such a way that grouping all BALs together into the same subclass of objects may restrict the physical insight that can be gained into the origin and properties of outflows.

8 CONCLUSIONS

Using MFICA, we have successfully reconstructed the UV spectra of both high-ionization BAL and non-BAL quasars from the SDSS DR14 quasar catalogue with redshifts 1.56 ≤ |$z$| ≤ 3.5 and S/N ≥ 5, ≃ 139 000 quasar spectra in total (Fig.   1). Using our reconstructions and employing improved redshifts (Fig. 3), we have redefined the BAL quasar population (Fig. 5) and found evidence for two AI-defined populations, in agreement with Knigge et al. (2008): BI = 0 and BI > 0 quasars (Fig. 7).

By reconstructing the quasar spectra, we have recovered the intrinsic C iv emission of BAL quasars even where the said emission is heavily absorbed. This has allowed us to, for the first time, place the BAL quasars alongside the non-BAL quasars in C iv emission profile space (Fig. 8). We find that for every BAL quasar with certain C iv emission properties, there exists a non-BAL quasar with the same properties. The converse is not necessarily true; there are almost no BAL quasars at the highest EWs in C iv emission space with zero blueshift, compared to the non-BAL quasars.

In addition to C iv emission, BAL and non-BAL quasars are extremely similar in respect of their He ii λ1640 EW, bolometric luminosity, and Eddington ratio (Figs 12–14). The similarities in properties, some measured from the reconstructed emission spectra, suggest that the broad absorption observed in BAL quasars is the result of clumpy, outflowing gas along the line of sight (broadly consistent with Yong et al. 2018) and that all quasars – barring perhaps those with the hardest ionizing SEDs at the highest C iv emission EWs and lowest blueshifts – have the potential to be seen as BAL quasars but with varying probability (Fig. 10). The trough parameters also vary systematically in C iv emission space (Fig. 16), and we have gained insight into the trough structure of the two AI-defined populations. At the highest C iv EWs, the BAL quasars have slightly narrower emission lines compared to the non-BAL quasars (Fig. 11) and it is also these BAL quasars that have quite different troughs from the rest of the BAL quasars – deep, narrow, and close to systemic velocity – which perhaps could suggest a different driving mechanism of the outflows and/or a different quasar orientation.

Virtually every paper on BAL quasars starts by noting the two main hypotheses, which are that BALs are ubiquitous in the quasar population but have only a ∼20 per cent covering fraction or that they represent a distinct 20 per cent of the quasar population (with nearly 100 per cent covering fraction). With this paper, we argue that the question is largely resolved (at least for high-ionization BALs): the similarity of BAL and non-BAL quasars demands that BALs are not a distinct class of quasars. Rather, BAL quasars are normal quasars observed along a particular line of sight or at a particular time. Conceivably, both may be true given that BAL troughs can be transient (e.g. Capellupo et al. 2012; Rogerson et al. 2018; Yi et al. 2019). Models invoking non-spherical geometries and particular viewing angles, e.g. BALs seen when the line of sight probes just above/below some form of obscuring torus, are not straightforward to reconcile with the strong systematic relationship between absorber and emission-line kinematics across the full range of C iv emission properties.

SUPPORTING INFORMATION

CIV_BAL_nonBAL.html.zip

Please note: Oxford University Press is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

ACKNOWLEDGEMENTS

We thank Bob Carswell, Norm Murray, and Vivienne Wild for helpful discussions. An anonymous referee provided a comprehensive review for which we are grateful. ALR acknowledges funding via the award of a Science and Technology Facilities Council (STFC) studentship. MB acknowledges funding from the Royal Society via a University Research Fellowship. PCH acknowledges funding from STFC via the Institute of Astronomy, Cambridge, Consolidated Grant.

Funding for the Sloan Digital Sky Survey IV has been provided by the Alfred P. Sloan Fundation, the U.S. Department of Energy Office of Science, and the Participating Institutions. SDSS acknowledges support and resources from the Center for High-Performance Computing at the University of Utah. The SDSS web site is www.sdss.org.

SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions of the SDSS Collaboration including the Brazilian Participating Group, the Carnegie Institution for Science, Carnegie Mellon University, the Chilean Participation Group, the French Participation Group, Harvard-Smithsonian Center for Astrophysics, Instituto de Astrofísica de Canaries, The Johns Hopkins University, Kavli Institute for the Physics and Mathematics of the Universe (IPMU) / University of Tokyo, the Korean Participation Group, Lawrence Berkeley National Laboratory, Leibniz Institut für Astrophysik Potsdam (AIP), Max-Planck-Institut für Astronomie (MPIA Heidelberg), Max-Planck-Institut für Astrophysik (MPA Garching), Max-Planck-Institut für Extraterrestrische Physik (MPE), National Astronomical Observatories of China, New Mexico State University, New York University, University of Notre Dame, Observatório Nacional / MCTI, the Ohio State University, Pennsylvania State University, Shanghai Astronomical Observatory, United Kingdom Participation Group, Universidad Nacional Autónoma de México, University of Arizona, University of Colorado Boulder, University of Oxford, University of Portsmouth, University of Utah, University of Virginia, University of Washington, University of Wisconsin, Vanderbilt University, and Yale University.

Footnotes

REFERENCES

APPENDIX A: CREATION OF PRIORS

The correlation between the morphology of the C iii]λ1908 + Si iii]λ1892 + Al iii λ1857 emission complex and the extent of the blueshift of C iv emission in luminous quasar spectra has been known for some time (see fig. 16 of Richards et al. 2011, and the top panels of our Fig. A1). While the C iv blueshift is a result of shifting the centroid of the line, the observed C iii] ‘blueshift’ is caused by an increase in the ratio of Si iii] to C iii] emission. Richards et al. (2011) used the C iii] ‘blueshift’ to parametrize the blueshift of the absorbed C iv emission in a sample of BAL quasars. Here, we use the C iii] ‘blueshift’ to place priors on the 10 component weights for the BAL quasars since, for the high-ionization BAL quasars under investigation, the 1600–2900 Å wavelength interval is free of high-ionization transitions where broad absorption is observed.

Figure A1.

Priors for first component (left-hand panels) and seventh (right-hand panels) as examples, created from ∼4000 non-BAL quasars. Top: C iv blueshift against that of C iii] for the non-BAL quasars, with points coloured by component weight. Vertical grey lines mark the C iii] ‘blueshift’ bin edges. Middle: Component weight against C iii] ‘blueshift’. Grey lines are again the bin edges. Bottom: Median component weight in each C iii] ‘blueshift’ bin against C iii] ‘blueshift’ (blue points). The blue line and shaded regions are created via a quadratic spline interpolation of the median and the MAD component weights. The green points are the same as those in the middle panels. The orange points mark the priors for an example spectrum with a C iii] blueshift of 2000 km s⁻¹, where the points and error bars mark the centre and width of the normal priors for each component, respectively.

Open in new tabDownload slide

The component weight priors were constructed using a sample of ∼4000 non-BAL quasars with S/N ≥ 9 per pixel. Similarities between the BAL quasar and non-BAL quasar spectra in the wavelength range of 1600–3000 Å allow the non-BAL quasars to be used to infer the BAL quasar component weights. The components covering 1260–3000 Å were fitted to the spectra and the C iii] ‘blueshift’ measured between 1820 and 1970 Å using the recipe detailed in Section 6.1. The continuum was fitted using the wavelength intervals of 1790–1810 and 2015–2035 Å. The middle panels of Fig. A1 show two examples of the component weights against the C iii] ‘blueshift’ for the non-BAL quasar sample. The spectra (and component weights) were binned according to the C iii] ‘blueshift’. The edges of the bins are drawn on the top four panels in Fig. A1. The correlation between the C iii] and C iv is evident in the composite reconstruction for each bin in Fig. A2: with increasing C iii] ‘blueshift’ (increasing bin number), the C iv blueshift also increases. Note also that the C iv EW decreases and that He ii λ1640 decreases in strength with increasing blueshift, the latter of which is in agreement with Richards et al. (2011) and Baskin et al. (2013, 2015).

Figure A2.

Non-BAL quasar composite reconstructions for the odd-numbered bins in C iii] ‘blueshift’. From the first bin to the ninth bin, as the C iii] ‘blueshift’ increases the C iv blueshift also increases.

Open in new tabDownload slide

The median and MAD of each component weight in each bin were calculated (see examples in the bottom panels of Fig. A1). From these measurements, for each C iii] ‘blueshift’ measured in the whole quasar sample, normal priors can be created by implementing quadratic spline interpolation via the Python library scipy (shaded regions in the same panels). As an example, a spectrum with a C iii] blueshift of 2000 km s⁻¹ will have a normal prior for component 1 (7) with centre 0.018 (0.009) and width 0.005 (0.003).

APPENDIX B: THE UNABSORBED SEDS OF BAL AND NON-BAL QUASAR POPULATIONS

All quasar spectra in our sample, whether BAL or non-BAL, are reconstructed using MFICA components generated from a sample of non-BAL quasars. Much of the BAL spectra blueward of 1600 Å is affected by absorption such that to reconstruct the BAL quasar spectra we rely on the assumption that the correlation between the C iv emission and the spectra redward of 1600 Å of the non-BAL quasars can also be applied to the BAL quasar spectra. To test the validity of the assumption that, excluding absorption, the BAL and non-BAL spectra are similar, we have generated composite BAL and non-BAL spectra for different regions in C iv emission space. The regions are those marked on the C iv emission space in the top-right panel of Fig. B1. Pixels masked due to narrow or broad absorption in individual spectra do not contribute to the composites. In regions 1–4, where the distributions of BAL and non-BAL quasars in the C iv emission space differ significantly, the objects contributing to each composite have been matched carefully. Specifically, non-BAL composites have been constructed using the same number of spectra contributing to the BAL composite, where each non-BAL spectrum has been selected to lie close to a BAL spectrum in the C iv emission space.

Figure B1.

Composite spectra from the numbered C iv regions in the top-right panel. BAL composite spectra are in black and non-BAL quasars in colours. Each composite is numbered according to its location in C iv emission space. For three of the regions, we plot the C iv emission in the lower right panels. Composite BAL reconstructions are also plotted here (coloured dotted lines). The C iv parameters measured from the composite spectra (crosses) and from individual quasars (points) are plotted in the C iv emission space with non-BALs in blue and BALs in red.

Open in new tabDownload slide

Fig. B1 presents the composite spectra of the BAL and non-BAL quasars, illustrating that for all regions in C iv emission space the median non-BAL and BAL spectra are extremely similar. Only by adopting a very contrived model could the results of the comparison (for the two populations) be explained in circumstances where the assumption was not also valid for individual BAL quasars. There are only slight differences in a few regions, namely 3, 6, and 9, where the continuum blueward of C iv is lower in the BAL composites than the non-BAL composites (1, 2, and 1 per cent difference on average for the three regions) on account of imperfect masking of the absorption troughs. The composite BAL reconstructions (lower right panels of Fig. B1) are almost identical to the non-BAL spectra. The improved match comes about because of the continuity over hundreds of  km s⁻¹ in the reconstruction process, further reducing the effect of the modest depression in the BAL spectra.

The C iv emission properties computed from the BAL and non-BAL composite spectra and the BAL reconstructions show only very small differences. Even using the composite spectra themselves, the differences in the C iv EW and blueshift are small, typically only ≃3 per cent in EW and ≃70 km s⁻¹ in blueshift. No blueshift difference exceeds 150 km s⁻¹ and the most extreme difference in EW occurs for the composites in region 9, where the 2 Å change corresponds to a 9 per cent increase for the BAL composite. The measurements from the BAL and non-BAL reconstructions, used throughout the paper, are even more similar.

The differences in the C iv emission properties between the actual BAL and non-BAL composite spectra are negligible in the context of all measurements presented in the paper. The result provides clear evidence that the assumption regarding the similarity between the BAL and non-BAL SEDs is valid to a high degree of accuracy for the two populations.