Searching for the signature of radiative line driving: on the absence of Lyα–N v line-locking features in a large sample of BALQSOs

Abstract

We have searched the hybrid broad absorption line quasar (BALQSO) catalogue of Scaringi et al. derived from Data Release 5 of the Sloan Digital Sky Survey in order to compile the largest sample of objects displaying spectral signatures which may be indicative of radiative line driving. The feature in question is the ‘ghost of Lyα’, a line-locking feature previously identified in the broad C iv and Si iv absorption lines of a small fraction of BALQSOs, and formed via the interaction of Lyα photons with N v ions.

We test, where possible, the criteria required to produce an observable ghost feature. These criteria include significant broad absorption, strong intrinsic Lyα emission, narrow Lyα, strong N v absorption and a weak far-ultraviolet continuum. No single ghost candidate meets all of these criteria. Furthermore, we find that these criteria are not met significantly more often in ghost candidates than in a comparison sample chosen to exhibit relatively featureless broad absorption troughs. Indeed, the only significant difference we find between our ghost candidate and comparison samples is that on average our ghost-candidate sample displays (i) significantly stronger N v absorption, and (ii) the onset of absorption occurs at lower velocities in our ghost-candidate objects.

Significantly, we find no evidence for an excess of objects whose absorption troughs bracket the location of the Lyα–N v line-locking region, rather the location of ghost-like features appears to be independent of any systematic velocity, with comparable numbers appearing both redward and blueward of the ghost zone. Thus, the majority of objects identified here as strong ghost candidates are likely multitrough interlopers whose absorption feature simply brackets the region of interest.

galaxies: active, quasars: absorption lines, quasars: general

1 INTRODUCTION

Broad absorption line quasars (BALQSOs), as their name suggests, show strong broad blueshifted absorption lines in their spectra believed to be indicative of high velocity (∼0.1c) out-flowing winds. BALQSOs represent approximately 15 per cent of quasars in general (Reichard et al. 2003a; Trump et al. 2006; Knigge et al. 2008; Scaringi et al. 2009). The majority (∼85 per cent; Sprayberry & Foltz 1992; Reichard et al. 2003b) of BALQSOs is known as HiBALs, displaying absorption in lines of high ionization only (e.g. N v, Si iv and C iv). The remainder are classified as LoBALs and show in addition broad absorption in lines of low-ionization species, most notably Al iii and Mg ii. LoBALs are further subclassified according to the presence of Fe absorption, the Fe LoBals. Since the spectra of LoBals are in general redder than HiBals, Becker et al. (2000) suggest that BALQSOs and in particular Fe LoBALs may be an early phase in the development of emerging or refuelled quasars.

There has been much debate about the relationship between BALQSOs and the general quasar population as a whole. Simple unification schemes suggest that BALQSOs and non-BALQSOs are similar objects and that any observed differences in their spectra arise due to orientation effects (Weymann et al. 1991; Ogle et al. 1999; Schmidt & Hines 1999; Elvis 2000). In this picture, the relative fraction of BALQSOs to non-BALQSOs has a simple geometric interpretation, representing the fraction of sky, as seen from the source, obscured by gas (i.e. the source covering fraction), which in the simplest case, can be related to the flow geometry (e.g. opening angle).

In recent years interest in BALQSOs outflows has risen sharply, principally because of the realization that such high-velocity outflows carry a substantial amount of energy and momentum into the interstellar medium (ISM), and may therefore be important in driving active galactic nuclei (AGN) feedback as well as providing a mechanism for quenching star formation (Scannapieco, Silk & Bouwens 2005). Indeed, the discovery of highly ionized, very high velocity X-ray outflows (e.g. Pounds 2003a,b; Pounds & Reeves 2009), for which the energy transport (in terms of mechanical energy) is large enough to interrupt the growth of the host galaxy, may provide the causal link behind the well-known correlation between the mass of the central black hole and the mass of the bulge (e.g. Ferrarese & Merritt 2000; Gebhardt et al. 2000; Tremaine et al. 2002).

However, despite the increasing importance of these outflows the precise mechanism responsible for accelerating them to high velocity remains uncertain, with radiative acceleration (Shlosman, Vitello & Shaviv 1985; Arav & Li 1994; Murray et al. 1995; Proga, Stone & Kallman 2000; Chelouche & Netzer 2001), magnetohydrodynamic (MHD) driven winds (Blandford & Payne 1982; Emmering, Blandford & Shlosman 1992; Konigl & Kartje 1994; Bottorff et al. 1997) and thermally driven winds seen as the main contenders (see e.g. Proga 2007 for a review). A major hindrance to progress in this area is the absence of a clean discriminatory observational signature of what are essentially orthogonal wind geometries. For radio-quiet objects, AGN unification schemes tend to favour equatorial wind geometries (e.g. Elvis 2000). By contrast, observations of radio-loud BALQSOs (Becker et al. 1997, 2000; Brotherton et al. 1998) and in particular polarization observations of PKS 0040−005 indicate a non-equatorial broad absorption line (BAL) outflow (Brotherton, de Breuck & Schaefer 2006). One possible explanation for the apparent difference in wind geometry between radio-loud and radio-quiet objects is that different acceleration mechanisms operate in these two classes of objects.

Perhaps the strongest indicator that radiative driving is responsible for accelerating at least some of the high-velocity flows is the appearance of line-locking features in the spectra of a small fraction of BAL quasars (e.g. Turnshek 1984; Weymann et al. 1991; Korista et al. 1993; Arav 1996; Vilkoviskij & Irwin 2001). The most well studied of these is the so-called ghost of Lyα, a small hump seen in the absorption troughs of a small fraction (less than a few per cent) of BALQSOs formed via the interaction between Lyα photons and N v ions (see e.g. Korista et al. 1993; Arav et al. 1995; Arav 1996, and references therein). However, the reality of ghosts remains very much an open question. Previous studies have been limited to small samples of relatively low-quality spectra, with strong ghosts often only becoming apparent in composite spectra (Arav et al. 1995; North, Knigge & Goad 2006). Moreover, distinguishing between line-locking features and similar features caused by the chance alignment of multiple absorption systems is difficult with relatively small samples. Indeed, Korista et al. (1993) showed that in a sample 72 objects, evidence for line-locking features was merely suggestive rather than convincing.

The aim of this paper is twofold: (i) to compile the largest sample of uniformly selected ghost-candidate spectra, and (ii) attempt to determine the origin of their ghost features, by testing on a source by source basis, whether they meet the original criteria as set out by Arav (1995) necessary for ghost features to be observed. In Section 2 we describe the mechanism proposed for ghost formation and outline Arav's criteria for the production of strong ghost signatures. Selection of those objects comprising our ghost-candidate spectra and our non-ghost control sample is described in Section 3. In Section 4 we present the results of testing each of the spectra in our ghost-candidate and comparison samples against each of the criteria in turn necessary for the formation of an observable ghost feature. We discuss the implications of our findings in Section 5. Our conclusions are summarized in Section 6.

2 THE GHOST OF LYα

The feature that is poetically known as the ghost of Lyα is a hump seen in the absorption trough of a small fraction (less than 5 per cent; Arav et al. 1995, North et al. 2006) of BALQSOs located 5900 km s⁻¹ blueward of the centre of the line to which the absorption is attributed. The feature was first investigated by Korista et al. (1993) after being seen in a difference spectrum of BALQSO and non-BALQSO spectral composites produced by Weymann et al. (1991) who pointed out that the difference between the two absorption troughs was approximately equal to the velocity separation of N v and Lyα. In a series of papers in the mid 1990s (Arav & Begelman 1994; Arav & Li 1994; Arav, Li & Begelman 1994; Arav et al. 1995; Arav 1996) Arav suggested that the feature seen at this velocity was a result of the increased radiative pressure on N v ions at this velocity within the outflow. In this picture N v ions moving at 5900 km s⁻¹, see the Lyα emission from the AGN at the energy of their own emission resulting in a large increase in the scattering cross-section of the Lyα photons. Consequently, out-flowing N v ions receive an increase in radiation pressure leading to a large injection of momentum into the outflow. The out-flowing ions of other species are dragged to higher velocities by electromagnetic interactions with the N v ions, leading to a deficit of material at this velocity (because each out-flowing ion spends very little time at this velocity before being accelerated to greater velocities). In turn this causes a decrease in the opacity at this velocity which manifests as a peak within the absorption troughs of the out-flowing species.

This simple mechanism explains both the presence of the feature and why its position is linked to the rest frame of the source.1Arav (1996) suggested that the reason for the feature being most prominent in C iv absorption rather than in other species is that the doublet separation of C iv is far smaller (498 km s⁻¹). By comparison the larger doublet separations of Si iv (1933 km s⁻¹) and O vi (1647 km s⁻¹) result in the feature being blurred over several thousand km s⁻¹, producing a low-contrast feature which is difficult to detect in low signal-to-noise ratio (S/N) data. For this reason, in this study we confine our search for ghost features to the C iv BAL only.

Since Lyα and N v are amongst the strongest UV lines in AGN spectra, an obvious question to ask is why are not ghost-features ubiquitous amongst the general BALQSO population? To answer this, Arav proposed a set of physically motivated criteria which had to be met before ghost features would be observable. These are the following:

• significant broad absorption in the region between 3000 and 9000 km s⁻¹ blueward of line centre;

• strong intrinsic Lyα emission;

• narrow Lyα emission;

• strong broad N v absorption;

• little far-ultraviolet (UV) flux between λλ200 and 1000 Å.

The first criterion is simply a statement that ghost features can only be observed if significant photon scattering is occurring. The second criterion implies that there must be a significant flux of Lyα photons, while the third requires that the line-emitting gas has a relatively narrow spread in velocities so that the resulting feature is not spread over a large range of velocity reducing the contrast. Since the dominant scattering ion is N v, strong N v absorption is also needed. The final requirement, and the most difficult to measure, recognizes that in order for Lyα–N v line locking to dominate the dynamics of the flow, there cannot be significant far-UV line driving (Korista et al. 1993).

3 SELECTION OF GHOST-CANDIDATE SPECTRA

BALQSOs represent ∼15 per cent of the quasar population (see e.g. Knigge et al. 2008 and references therein), of these, between ∼20 and 25 per cent show evidence for multiple troughs (Korista et al. 1993; North et al. 2006). Amongst the multitrough objects, ghost candidates are likely rare representing less than a few per cent of BALQSOs (Arav et al. 1995) because of the stringent requirements for ghost observability. Thus in order to find significant numbers of ghost candidates, large quasar samples are required.

All previous ghost-candidate samples have been limited to just a handful of objects (Arav 1996; North et al. 2006). By comparison, Data Release 5 (DR5) of the Sloan Digital Sky Survey (SDSS) contains 77 429 quasars of which 28 421 have redshifts between 1.7 and 4.2 allowing C iv and any associated absorption to be seen in their optical spectra. We require a visible C iv BAL because although radiative line driving will produce a feature in all absorption troughs, the resultant hump will be most easily observed in the C iv absorption trough because of the small doublet separation of this line. Several BALQSO samples have been compiled from the various data releases of the SDSS (Reichard et al. 2003a; Trump et al. 2006; Knigge et al. 2008) and from DR5 (Gibson et al. 2008; Scaringi et al. 2009). Here we adopt the BALQSO catalogue of Scaringi et al. (2009), containing 3552 BALQSOs. The Scaringi et al. sample differs from other samples in that is does not rely on simple metrics such as the balnicity index (BI; Weymann et al. 1991) or absorption index (AI; Hall et al. 2002; Trump et al. 2006) to identify and classify BALQSOs. Rather it uses a robust hybrid method, involving gross classification with a supervised neural network (learning vector quantisation) and the visual inspection of outliers.

To identify strong ghost candidates we have performed five cuts on this BALQSO sample. The first cut eliminates noisy spectra (S/N < 4 pixel⁻¹), estimated in two line-free continuum bands from λλ1650–1700 and λλ1700–1750 Å. To reduce contamination from emission, absorption and cosmic rays that may be present in either bin we take the greater of the two values as our S/N estimate. The second cut removes all objects which show an apparently single smooth absorption trough. Following this cut we are left with only those objects which display multiple absorption trough features (1019 in total).2

A third rejection cut removes those objects that contain two or more peaks within the absorption trough (numbering 761). These objects are eliminated to reduce contamination by objects which display absorption from multiple out-flowing regions which by chance happen to lie at velocities corresponding to the ghost zone. This leaves just 258 objects which contain a single peak (or alternatively double trough) within the C iv absorption region.

Fig. 1 (upper panel) shows our multitrough composite spectrum (solid line) created from taking the geometric mean of all objects rejected at this stage after normalizing the flux between λλ1700–1750 Å to unity. In order to highlight the absorption regions, we also plot a non-BALQSO composite spectrum (dashed line). The non-BALQSO composite is the geometric mean of all of the quasar spectra in the SDSS DR5 quasar catalogue in the redshift range 1.7 < z < 4.2 excluding those objects which are also in the BALQSO catalogue of Scaringi et al., and reddened to fit to the continuum windows in the multitrough composite spectrum using the Small Magellanic Cloud (SMC) extinction law of Pei (1992). For completeness, the middle and lower panels show two example spectra rejected at his stage. We note that our multitrough composite spectrum 1 shows no evidence for coherent structure within the C iv trough. This suggests that the positions of the various absorption systems in the individual spectra are in general uncorrelated, that is they do not contain a large number of ghost candidates that happen to show additional peaks within the absorption. This is consistent with the hypothesis that these multiple-trough spectra are the result of multiple unconnected absorbing systems.

Figure 1

Top panel: geometric mean composite spectrum of objects (1019) rejected due to having multiple trough features (solid line) fit with a reddened DR5 QSO composite (dashed line). Middle and lower panels: example spectra of individual objects rejected at this stage. The dashed vertical lines indicate the location of the ghost zone.

For the 258 objects which show a single peak in their C iv absorption we make a further cut to remove those objects (189) which show peaks outside of the ghost zone as defined by North et al. (2006). The ghost-zone edges are 5900 km s⁻¹ blueward of the peaks of the doublet of the emission line in question (in this work C iv), this zone is then expanded to take into account redshift errors by multiplying these boundaries by 1 ±Δz/(1 +z_med), where z_med is the median redshift of our sample (z_med= 2.14), and Δz is the average redshift error (Δz= 0.01). This results in a ghost zone extending between λλ1513.2 and 1525.4 Å for C iv. Our single-peak composite comprising 189 objects whose peak lies outside of the ghost zone is shown in Fig. 2 (upper panel, solid line). The middle and lower panels show example spectra of objects rejected at this stage. Each object shows a clear peak located just outside of the limits of the ghost zone. Interestingly, this composite spectrum shows some evidence of a feature at the low-velocity edge of the ghost zone. One possible explanation for this feature is that these objects are genuine ghost candidates with poorly assigned redshifts. However, on closer inspection of the objects rejected at this stage we find no evidence for features outside of the ghost zone due to random redshift errors. Alternative explanations for the origin of this feature include (i) intrinsically weaker absorption at lower velocities, or (ii) an abrupt change in the intensity of the (overlying) emission at these velocities.

Figure 2

Top panel: geometric mean composite spectrum of objects (189) rejected due to having double trough, ghost-like features that appear outside of the ghost zone (solid line) fit with a reddened DR5 composite (dashed line). Middle and lower panels: example spectra of individual objects rejected at this stage. The dashed vertical lines indicate the location of the ghost zone.

We make one further cut to remove those objects which do not show significant broad absorption either side of the ghost feature extending from 3000 to 9000 km s⁻¹ blueward of the emission line. This cut is chosen to eliminate objects with deep narrow absorption features whose alignment mimics the presence of a ghost feature. Fig. 3 (upper panel, solid line) displays the composite spectrum of all objects rejected at this stage, while in the middle and lower panels, we give examples of individual objects rejected by this cut. Though the composite spectrum shows a clear feature within the ghost zone, it is not smooth, instead displaying sharp narrow features, whose separation is larger than the doublet separation of C iv. This reflects both the narrowness of the absorption features (since intrinsically broad humps would be smoothed out in the averaging process yielding low-contrast features) which make up this composite, and in addition small differences in their velocities relative to the rest frame of the source. These differences may result from these features being unrelated to the ghost mechanism and appearing in the ghost zone by chance though we cannot preclude the fact that some of the objects rejected at this stage are indeed true ghost candidates.

Figure 3

Top panel: geometric mean composite spectrum of objects (26) rejected due to having insufficient absorption either side of the ghost feature. Middle and lower panels: example spectra of individual objects rejected at this stage. The dashed vertical lines indicate the location of the ghost zone.

Our final sample of 43 ghost-candidate objects comprises the largest sample of objects exhibiting the ghost of Lyα and the first to allow a statistical analysis of the properties of these objects. Fig. 4 shows the geometric mean composite ghost-candidate spectrum (solid line), together with a non-BALQSO composite spectrum (dashed line), and two example ghost-candidate spectra (middle and lower panels). Our final ghost-candidate composite spectrum displays a clear smooth double-trough structure located firmly within the ghost zone of a relatively broad absorption trough. Visual comparison of this composite with that formed from objects rejected at the previous stage (i.e. those with insufficient absorption either side of the ghost zone) indicates that our final sample displays on average narrower Lyα and C iv emission lines, stronger Lyα emission and deeper N v absorption (fulfilling three of Arav's criteria for forming observable ghost features), as well as broader Si iv absorption, and suggests that we are indeed isolating those objects most likely to form observable ghost features. Fig. 5 summarizes the various cuts made in the creation of our ghost-candidate sample.

Figure 4

Top panel: geometric mean composite spectrum of our ghost-candidate sample of 43 objects. Middle and lower panels: example spectra of individual objects making up this sample. The dashed vertical lines indicate the location of the ghost zone.

Figure 5

Flow diagram showing the number of objects rejected at each step in our ghost selection algorithm.

3.1 Selecting a comparison sample

Of the 43 strong ghost candidates, 21 are at large enough redshift z > 2.15 to place Lyα above the atmospheric cut-off and thus allow us to test for each object all of the criteria deemed necessary for ghost formation. These are listed in the upper half of Table 1. For the remainder, their redshifts are too low to allow a direct measurement of the strength and width of the Lyα emission line. For these objects we estimate the strength and width of the Lyα emission line using the C iv emission line as a surrogate, and substantiated by known correlations between the two lines (see e.g. Wilkes 1984; Ulrich 1989). These objects are listed in the lower half of Table 1. For completeness, we have also compiled a sample of BALQSOs showing no evidence for a ghost feature or multiple troughs. This allows us to compare the relative frequency with which the ghost-formation criteria are met amongst BALQSOs both with and without ghost features. In order to create such a sample we have performed a number of cuts on the Scaringi et al. (2009) BALQSO sample. As for our ghost-candidate sample we first select for S/N > 4, and then select those objects with z > 2.15 to allow inspection of the N v and Lyα emission lines necessary to test for the presence of strong narrow Lyα emission and strong N v absorption. The remaining spectra are then visually inspected and any object showing multiple absorption troughs or any other hint of a ghost feature is removed from the sample. We also remove those objects which do not show significant absorption between 3000 and 9000 km s⁻¹. This leaves just 26 objects showing significant C iv absorption between 3000 and 9000 km s⁻¹ with no evidence for a ghost feature or multiple trough. One of these objects SDSS J101056.68+355833.3 shows evidence for a ghost feature in the Si iv absorption line (see A5) and is therefore also removed from our comparison sample leaving just 25 objects. A summary of the rejection cuts made in the creation of our comparison sample is given in Fig. 6. The geometric mean composite spectrum of our comparison sample is indicated by the dotted line in Fig. 7. Aside from substantially weaker N v absorption, the strengths and widths of the emission lines in the comparison sample composite spectrum match those of the ghost candidate composite spectrum remarkably well. The main difference between these two composite spectra appears to be the absence of the low-velocity absorption trough in the comparison sample spectrum. We do not believe this difference arises from the way in which the two samples were selected.

Table 1

Our ghost sample and the results of testing Arav's criteria along with the FWHM of the ghost feature.

Object name	Redshift	Strong N v BAL	Strong Lyα emission	Narrow emission lines FWHM (km s−1)	He ii EW scalefactor	FWHM ghost (km s−1)
024514.93−004101.9	2.802	√	√	X(3624)	X 1.06 (0.89–1.31)	2646
080621.45+191441.0	2.272	√	X	X(4307)	√ 1.58 (1.23–2.17)	1785
083749.59+364145.4	3.416	√	X	√(2807)	√ 2.67 (2.05–3.73)	2005
084316.93+050640.1	2.409	√	√	√(2690)	X 0.96 (0.85–1.11)	1744
085104.05+051539.8	3.213	√	√	X(4324)	√ 1.56 (1.31–1.93)	1013
090430.49+343108.9	3.408	√	√	X(4067)	X 0.89 (0.78–1.00)	1696
092557.52+044035.9	2.266	√	√	X(5219)	√ 1.41 (1.09–2.07)	1248
093742.27+403351.2	2.258	√	√	√(2725)	X 1.00 (0.81–1.23)	1125
094427.27+614424.6	2.337	√	√	√(2426)	X 0.76 (0.88–0.87)	782
094454.85+481403.5	2.291	√	√	√(2504)	X 0.65 (0.55–0.72)	1833
101844.45+544015.6	3.253	√	√	√(2729)	X 0.43 (0.37–0.53)	2168
103450.21+313326.1	2.496	√	X	X(3943)	X 1.41 (0.99–2.48)
105058.28+614406.0	2.797	√	√	√(2375)	X 0.60 (0.53–0.69)	2185
110623.51−004326.0a	2.443	√	√	X(4040)	√ 3.35 (>1.82)	2996
114704.47+153243.3	3.081	√	√	√(3258)	X 0.89 (0.79–1.00)	771
131505.19+603716.4	2.330	√	√	X(3976)	X 0.97 (0.83–1.18)	1313
134818.03+423205.1	3.066	√	√	?	X 0.89 (0.73–1.14)	944
142244.45+382330.6	3.728	√	√	√(2146)	X 1.16 (0.93–1.51)	2505
144032.04+060554.1	2.297	√	√	√(2723)	X 0.89 (0.79–1.00)	1337
162657.47+405848.0	3.051	√	√	√(2520)	X 0.49 (0.44–0.55)	1333
170322.41+231243.3	2.634	√	√	√(2138)	X 0.93 (0.79–1.05)	1032
005109.45+001636.3	2.036	?	?	X(4850)	X 1.30 (0.94–2.00)	2800
023252.80−001351.1a	2.033	?	?	X(4016)	X 0.94 (0.82–1.09)	709
033048.50−002819.6a	1.779	?	?	?	√ > 2.00	1977
081158.10+343624.1	2.106	?	√	√(3147)	X 1.13 (0.86–1.63)	4395
082006.60+522158.7	2.015	?	?	√(3217)	??b	2660
091109.46+410155.2	2.128	?	√	X(4062)	X 0.56 (0.49–0.64)	1006
091313.09+411014.2	1.737	?	?	?	√ 3.61 (2.62–5.72)	2607
092221.25+084312.8	1.969	?	?	?	√ > 1.43c	1398
102908.06+365155.2	1.750	?	?	X(3870)	X 0.31 (0.28–0.34)	2432
110736.67+000329.4	1.741	?	?	√(2443)	X 1.18 (1.06–1.31)	2041
111255.70+040600.5	1.989	?	?	X(3706)	X 0.96 (0.85–1.08)	1055
113831.42+351725.3	2.118	?	√	√(2422)	X 0.78 (0.69–0.89)	1336
122107.07+074437.0	1.900	?	?	√(3222)	X 0.91 (0.77–1.08)	503
132304.58-003856.5a	1.827	?	?	X(4133)	X 1.09 (0.95–1.28)	2077
133428.06−012349.0	1.876	?	?	?	√(??)d	1616
134458.82+483457.5	2.052	?	?	?	X 0.89 (0.73–1.12)	1046
141843.95+373750.8	1.782	?	?	X(3596)	X 0.38 (0.33–0.44)
155505.10+442151.3	1.798	?	?	X(4425)	√ 1.90 (1.51–2.59)	1386
160335.43+225612.9	2.079	?	√	√(2479)	X 0.68 (0.60–0.78)	2614
163231.60+294929.7	1.905	?	?	√(2587)	X 0.61 (0.54–0.66)	2140
170056.85+602639.7a	2.123	?	√	X(4708)	X 1.15 (0.99–1.35)	1913
172001.31+621245.7a	1.760	?	?	√(2528)	X 1.73 (1.24–2.95)	2034

^a Objects identified by North et al. (2006) as their best ghost candidates, their ghost candidate final cut.

^b Poor continuum fit.

^c He ii absorbed.

^d He ii too weak to be measured.

Figure 6

Flow diagram showing the number of objects rejected at each step in our comparison sample selection algorithm.

Figure 7

Comparison between the geometric mean composites formed from (i) 21 ghost candidates with z > 2.15 (solid line), (ii) 25 BALs with featureless broad absorption troughs (dotted line) and (iii) non-BALQSOs from the DR5 QSO catalogue (dashed line).

In Section 4 we use our ghost-candidate sample to test, where possible, the criteria set out by Arav (1996) required by radiative acceleration models to produce an observable ghost feature. We compare the relative frequency by which our ghost candidate spectra meet the criteria necessary for the formation of observable ghost features with our non-ghost comparison sample.

4 TESTING THE CRITERIA FOR THE FORMATION OF OBSERVABLE GHOST FEATURES

In this section we present the results of testing the criteria for the formation of observable ghost features. We first concentrate on the sample of 21 ghost-candidate objects for which all of the criteria may be tested and compare these results with those obtained using our non-ghost comparison sample. We then move on to our remaining objects and test, where possible, whether these objects satisfy the necessary criteria.

4.1 Significant C iv broad absorption line

All of the objects in our ghost-candidate sample will clearly show significant absorption in the region between 3000 and 9000 km s⁻¹ blueward of the C iv emission line as this is required in order to see any potential ghost feature. This criterion is required purely because in order to effect the out-flowing C iv ions in an observable way this outflow must reach significant optical depth and appear as a BAL trough within the spectrum. Our comparison sample has also been chosen to exhibit strong broad absorption, so naturally also fulfils this criterion.

4.2 Strong intrinsic Lyα emission

In order for the effects of Lyα radiation on N v to cause a significant increase in the radiation pressure on the outflow as a whole the Lyα emission line must be intrinsically strong. However, since the ghost of Lyα is produced by scattering Lyα photons it can be extremely difficult to measure the strength of the intrinsic Lyα emission line. Hence Arav et al. (1995) modified this criterion to allow the C iv and N v emission lines to be used in situations where the Lyα emission line was either significantly absorbed, or had insufficient spectral coverage. This modified criterion requires Lyα emission to be at least 300 per cent above the continuum and/or both C iv and N v to be at least 100 per cent above the continuum.3 They note that ‘the latter criterion is hardly ideal but observations of non-BAL quasars show that the line strengths roughly scale together’ (Arav et al. 1995). Here we isolate the continuum emission using the specfit package in iraf. We model the continuum as a reddened power law and fit to the continuum in regions free of contaminating emission- and absorption lines, where possible. This can be particularly difficult for objects with the highest velocity out-flows as there are few emission/absorption line-free regions. Furthermore, we have not attempted to fit for the continuum shortward of λ1280 Å as this region is typically so absorbed and the line emission so highly blended that no clean continuum regions can be identified. Instead, we fix the continuum strength at short wavelengths to that measured between λλ1315 and 1350 Å. This approach is chosen to avoid overestimating the strength of the broad absorption, at the expense of overestimating the strength of the line emission in a few cases.

In objects whose spectra cover the Lyα region, 86 per cent (18/21) of the ghost-candidate sample meet the criterion for strong emission lines compared to 88 per cent (22/25) of the comparison sample. Inspection of the composite spectra for the two samples (Fig. 7) indicates that the strength of the Lyα emission line (relative to the continuum) is broadly similar in the two samples, and both are weaker relative to the non-BALQSO composite, due to the presence of strong N v and Lyα absorption. For the other lines, the N v emission line appears stronger on average in our ghost-candidate and comparison samples relative to non-BALQSOs, while there appears little difference in the strength of the C iv emission line, though we note that in so far as we can measure them, their emission-line widths appear to be narrower on average.

4.3 Narrow emission lines

Models of outflows accelerated by radiative line driving produced by Arav & Begelman (1994) and Arav et al. (1995) suggest that once the linewidths exceed 3500 km s⁻¹[full width at half-maximum (FWHM)] potential ghost features become undetectable. This is because broader emission lines will result in a broader, lower contrast feature (since the core of the line, where most of the scattering occurs, is weaker relative to the wings in a broad emission line c.f. a narrow emission line) which is difficult to detect particularly in low S/N data. Unfortunately, determining the width of the underlying emission lines in BALQSOs is extremely difficult as the broad absorption in these objects prevents accurate measurements of the shape and strength of the underlying emission lines and continuum. Given how our ghost-candidate sample was selected we expect all of the ghost candidates to show significant absorption in the blue wing of Lyα, N v and C iv, while our comparison sample was chosen to exhibit strong relatively featureless absorption in the blue wing of C iv.

Since we cannot make an accurate measurement of the width of the Lyα emission line we instead attempt to estimate its likely width using the C iv emission-line width as a surrogate. For the majority of sources, C iv is severely absorbed, so we estimate its underlying unabsorbed width using model fits to the red wing of the emission line. We use the same underlying fit for the continuum as was used to measure the emission-line strength (Section 4.2). We fit the red wing with a two-component model, a Gaussian core and Lorentzian wings, with the central wavelengths fixed together but allowed to vary slightly from the rest wavelength of C iv. We allow the strength and width of the two components to vary independently and fit to the red wing taking care to avoid those regions of the spectrum contaminated by broad He iiλ1640 and [O iii]λ1663 emission. The measured width is taken to be the FWHM of the composite fit. If the measured FWHM < 3500 km s⁻¹, then the object meets the criterion for the formation of an observable ghost feature.

For our 21 ghost candidates with z > 2.15, only 12 (57 per cent) have FWHM (C iv) < 3500 km s⁻¹. In the remainder, those for which the C iv width can actually be measured (17 objects), 8/17 (47 per cent) also meet this criterion. This fraction is similar to that found for our comparison sample where 12 out of the 20 objects (60 per cent) for which C iv width measurements can be made, also meet this criterion (see Table 2 for details). Taking the average of all of the FWHM measured for the ghost candidates gives an average FWHM of 3345 ± 842 km s⁻¹ which is similar to the upper limit proposed by Arav (1996) and to the average FWHM of C iv found for the comparison sample (3534 ± 1160 km s⁻¹). A Kolmogorov–Smirnov (K–S) test on these populations shows no significant difference between the ghost and comparison samples. The difficulty in fitting the C iv line in many of these objects results in large uncertainties in the estimated widths, however, it is clear that several of our ghost candidates have emission that is wider than the 3500 km s⁻¹ upper limit proposed by Arav (1996). There are a number of possible explanations for this. First, C iv may be a relatively poor surrogate for the width of Lyα. Alternatively, our sample may suffer from contamination by objects in which the potential ghost feature is produced by a chance alignment of multiple absorption systems or through other mechanisms unrelated to radiative acceleration. If the potential ghost features are indeed due to line locking, then it may in fact be possible to produce an observable ghost feature from emission lines with widths in excess of 3500 km s⁻¹.

Table 2

As for Table 1, indicating our comparison sample of 25 objects selected to have broad absorption and no ghost features and the results of testing Arav's criteria for ghost formation.

Object name	Redshift	Strong N v BAL	Strong Lyα emission	Narrow emission lines FWHM (km s−1)	He ii EW scalefactor
013724.43−082419.9	2.5663	X	√	√(3174)	X 0.39 (0.35–0.44)
014648.52−001051.8	2.3853	√	√	√(2186)	X 0.83 (0.77–1.38)
021219.54+141739.1	2.1902	√	X	X(6457)	X 0.94 (0.82–1.18)
024413.76−000447.5	2.7925	√	√	X(3804)	√ 2.72 (>2.46)
031331.22−070422.8	2.7548	√	X	?	√ 3.35 (2.43–5.56)
081906.14+394813.8	3.2071	X	√	X(4142)	X 0.43 (0.39–0.47)
084348.68+445226.9	2.5821	X	√	X(4024)	X 1.06 (0.89–1.30)
084554.24+423003.5	2.5568	X	√	√(2378)	√ 3.92 (>1.69)
090115.18+371822.8	2.6173	X	√	√(2322)	X 0.30 (0.27–0.33)
093804.52+120011.4	2.2273	X	X	?	√ 2.03 (1.59–2.79)
095220.88+371622.9	3.0928	√	√	√(3331)	X 0.80 (0.66–1.00)
101324.20+064900.3	2.7675	√	X	?	√ > 4.07
101420.52+325931.9	2.3935	X	√	√(3256)	X 0.41 (0.37–0.45)
104245.48+365642.2	2.8545	√	√	√(2235)	X 0.44 (0.40–0.50)
110928.51+092403.8	2.1539	X	√	√(3452)	X 0.49 (0.43–0.55)
111437.25+503445.9	2.2075	X	√	X	√ >3.63
115901.75+065619.0	2.1906	√	√	X	X 1.02 (0.79–1.43)
135559.03−002413.7	2.3438	X	√	√(3180)	√ 1.30 (1.11–1.53)
135912.20+450338.1	2.2794	X	√	X(3802)	X 0.95 (0.68–1.58)
141225.35+041951.9	2.3882	√	√	X(4748)	√ 0.72 (0.60–0.89)
143559.60+034153.7	2.4646	√	X	X(4003)	X 1.05 (0.85–1.32)
164148.19+223225.2	2.5061	X	√	√(2347)	X 1.12 (1.00–1.46)
165816.78+231653.7	2.5753	X	√	X(6305)	√ 1.88 (1.26–3.66)
214113.05−003545.8	2.2329	X	√	√(3099)	X 0.49 (0.43–0.54)
223841.88+142154.9	2.2898	√	√	√(2488)	X 0.47 (0.43–0.52)

4.4 Strong N v broad absorption line

As described in Section 2, the ghost feature is produced by the increased radiation pressure on out-flowing N v ions by Lyα photons. Since in this model the N v ions scatter a significant fraction of the incident flux, strong N v absorption is a prerequisite.

In order to measure the strength of absorption due to N v we use a modified BI (Weymann et al. 1991). We take the continuum to be equal to the mean flux between λλ1315 and 1330 Å, and require that the flux drops to below 90 per cent of this value continuously for >1000 km s⁻¹ between 0 and 7000 km s⁻¹ blueward of the N v emission line. We use the 7000 km s⁻¹ limit to avoid contamination by absorption due to Lyα. Any object with a non-zero value for this measure is considered to show strong N v broad absorption.

Significant N v absorption is measured in all 21 objects from our ghost-candidate sample whose redshift allows detection of N v absorption. Only 44 per cent (11/25) of our comparison sample satisfy this criterion. The geometric composites of Fig. 7 confirm that our ghost-candidate sample displays significantly stronger N v absorption on average than our comparison sample. Moreover, the broadly similar N v emission-line strengths of the ghost-candidate and comparison sample emphasizes the increased importance of N v absorption in the ghost-candidate spectra.

4.5 Little far-UV flux

Photoionization models suggest that the far-UV spectra of quasars contain a significant number of emission lines between λλ200 and 1000 Å that will contribute to the radiation pressure on any BAL outflow (Korista et al. 1993). In order for a ghost feature to be observed the radiation pressure of Lyα on the N v ions in the outflow must be at least comparable in strength to the radiation pressure on all other ions in the outflow. This led Arav to suggest that the far-UV continuum must therefore be necessarily weak in objects showing ghost features. Absorption of far-UV photons both within the host galaxy as well as within our own Galaxy preclude direct measurement of the far-UV continuum. However, Arav et al. (1995) note that the strength of the He ii λ1640 Å emission line can be used as a surrogate for the far-UV ionizing continuum. A strong He ii line suggests a similarly strong far-UV flux and vice versa as this line is produced by photons with energies corresponding to λ228 Å (≈54 eV).

However, the He ii emission-line strength is notoriously difficult to measure even in high-quality spectra. This is because of its close proximity to the [O iii]λ1663 emission line, as well as its location near the red wing of C iv, which may extend to relatively high velocities. Thus isolating the broad He ii component generally requires multicomponent fitting (see e.g. Goad and Koratkar 1998) which without prior information (for example, variability data) is rather subjective.

Since the quality (in terms of S/N) of the individual spectra in both our ghost-candidate and comparison samples is relatively low, we do not attempt to fit to the individual line components. Instead, we use our non-BALQSO composite spectrum as a template for the region of interest, and scale each spectrum to this template, by minimizing the residual flux in the region of interest. Our optimization routine fits for the underlying continuum in both spectra, and adjusts for both redshift errors and the scalefactor over the region of interest (nominally λλ1600–1680 Å). Since both spectra are first normalized to the flux in the continuum band between λλ1700 and 1750 Å, if we assume that the He ii and [O iii] emission-line strengths scale together, then the derived scaling factor between the ghost-candidate and non-BALQSO spectra is a measure of the ratio of their He ii equivalent widths (EWs). Fig. 8 shows an example of a fit. In the top panel we show the non-BALQSO composite (solid line) and a ghost-candidate spectrum (dotted line). In the middle panel we show the spectra together with their continuum fits spanning the C iv–He ii region, while in the lower panel we show a close-up of the non-BALQSO composite and the scaled ghost-candidate spectrum after removal of the underlying continuum. The residual flux under the He ii–[O iii] lines is indicated by the dot–dashed line. We caution that this fitting process is sensitive to (i) small redshift errors between the spectra, (ii) an appropriate choice of continuum bands (in general the choice of the continuum bands depends on the width of the C iv absorption trough) and most importantly (iii) the S/N of the spectra. Consequently the error on the scalefactor is generally quite large. We have visually inspected the results of the fitting process for all of our ghost-candidate and comparison spectra to verify that the fitting procedure has converged correctly, adjusting the continuum bands and region over which the spectra are scaled when necessary. In Table 1 we list the He ii scalefactors, that is the multiplication factor necessary to produce the same EW in the ghost-candidate spectrum. A number less than 1 indicates that the ghost-candidate spectrum has larger EW in the line in the region of He ii than the non-BALQSO spectrum. Conversely, numbers greater than 1 indicate that the ghost feature has a proportionately weaker He ii line. In order to determine error estimates on the derived scalefactors, after determining a best-fitting solution, we fix all parameters apart from the scalefactor, and then minimize on this parameter only. We then calculate the 90 per cent confidence interval on this one interesting parameter from the best-fitting model by varying the scalefactor until the χ² value has increased by 2.71. Both the best-fitting scalefactor and its range (based on the 90 per cent confidence interval) are given in Tables 1 and 2.

Figure 8

An illustration of our method for estimating the He ii strength. Upper panel: non-BALQSO composite (solid line) and an example ghost-candidate spectrum (dashed line). Middle panel: model fits to the underlying continua. Lower panel: non-BALQSO composite (solid line) and scaled ghost-candidate spectrum, after subtraction of the underlying continuum. The residual flux is indicated by the dot–dashed line.

For our sample of 21 ghost candidates for which all criteria can be tested, five indicate a significantly weaker He ii EW, suggesting a weaker than average UV continuum. Only five objects show evidence for a stronger than average He ii EW. For the rest, there is no significant difference (within the errors) in their He ii EWs when compared to the non-BALQSO composite. For the 22 ghost candidates for which only some of the criteria can be tested, five show evidence for a weaker than average He ii EWs, six indicate stronger than average He ii EWs and the rest indicate no significant difference in the He ii EW when compared to the non-BALQSO composite. Thus only 11/43 (25 per cent) of our ghost-candidate sample appear to satisfy the criteria for a weak UV continuum. Repeating this test on our non-ghost BALQSO comparison sample, eight objects indicate weaker than average He ii EW, nine objects have stronger than average He ii EW and six objects show no discernible difference in their He ii EW relative to the non-BALQSO composite. For the remaining two objects, the fits did not converge, though visual inspection of one of these shows no evidence for significant He ii emission.

4.6 Supplementary requirements – the width of the ghost feature

If we ignore the likely complex interaction between the Lyα photons and N v ions (i.e. we assume the optical depth in the flow is effectively constant with velocity), then naively one might expect that the ghost feature should be at least as broad as the Lyα emission line responsible for its formation. We note that in fact scattering is likely dominated by the line core where the optical depth is larger. By requiring that any observed ghost feature is at least as broad as Lyα, we can further refine our ghost-candidate sample to leave only the best ghost candidates. While the width of any ghost feature is relatively easy to measure (unless the S/N is low), we are once again limited by the accuracy to which we can measure the width of Lyα, as this emission is absorbed in the process by which the ghost is formed. Here we again make use of the width of the C iv emission line (see Section 4.3) as a surrogate for the width of Lyα. We isolate the ghost feature by removing the underlying absorption trough by fitting a simple function (cubic spline). We then fit a single Gaussian to the ghost feature to provide an estimate of its width. The measured widths are listed in Table 1. We find no correlation between the width of the ghost feature and the width of the C iv emission line (Fig. 9). Instead, we find that in the vast majority of cases the ghost feature is narrower than the C iv emission line [the solid line in Fig. 9 is FWHM(C iv) = FWHM(ghost)].

Figure 9

FWHM (km s⁻¹) of the C iv emission line and the ghost feature. The solid line shows FWHM(C iv) = FWHM(ghost).

If we require that the FWHM of the ghost feature matches that of the C iv emission line to within 1500 km s⁻¹ then 11/20 (for one object we are unable to make emission-line width measurements) of our ghost candidates for which all ghost-formation criteria can be tested also match this criterion. For our remaining ghost candidates 8/17 (five objects have no emission-line width information) also match this criterion. Of the 11 candidates with similar ghost and C iv widths for which all the criteria are testable all 11 show N v absorption, 10 show strong emission lines, nine have narrow C iv emission but only two show evidence for weaker than average He ii emission. Of these two, one has C iv FWHM of 4040 km s⁻¹ while the other fails the criteria for strong emission lines.

5 DISCUSSION

For the 43 objects in our ghost candidate sample (a composite is shown in Fig. 10), 21 are at large enough redshift to allow us to test all of the criteria necessary for the formation of an observable ghost feature (these are shown in Figs 11 and 12). Of these 21/21 satisfy the condition for strong N v absorption. 18/21 also satisfy the condition for strong intrinsic Lyα emission. However, less than half of the objects (12/21) satisfy the condition for narrow emission lines, and even fewer (5/21) show weaker than average He ii EW. While 11/21 objects satisfy the first three criteria, no single object meets all of the requirements for the formation of an observable ghost feature. For the remaining 22 objects (shown in Figs 13 and 14), only three objects pass two of the first three criteria, and of these all show stronger than average He ii. Of the five objects which indicate weaker than average He ii strengths, only one can be tested for any of the other criteria, in this case, the width of the emission line, which it fails.

Figure 10

Composite spectrum made up of all our ghost candidates fit with a reddened DR5 composite (dotted) and a BALQSO composite (dashed).

Figure 11

Spectra of ghost candidates from the best-candidate sample for which all criteria can be tested.

Figure 12

Spectra of ghost candidates from the best-candidate sample for which all criteria can be tested.

Figure 13

Ghost candidates for which only some of the criteria can be tested.

Figure 14

Ghost candidates for which only some of the criteria can be tested.

None of the objects in our ghost-candidate sample is present in the samples produced by Arav (1996) or Korista et al. (1993) due to differences in the redshift ranges over which they were constructed. North et al. (2006) identified seven strong ghost-candidate spectra in Data Release 3 (DR3) of the SDSS. six of those objects are also in our sample of 43 ghost candidates. The other, SDSS J142050.34−002553.1. has an assigned redshift of 2.085 in DR5 compared to the 2.103 used by North et al. (2006) and is therefore rejected by our ghost zone cut. Only one of North et al.'s ghost zone final cut is of sufficient redshift to test all of the criteria for the formation of an observable ghost feature, SDSS J110623.52−004326.0. This object fails the narrow emission-line requirement. For the other five objects, three fail the test for narrow emission lines, and four fail the test for weaker than average He ii EW.

For our non-ghost comparison sample, 12/25 meets the criterion for strong N v absorption. 20/25 meet the requirement of strong Lyα, while 12/22 meet the narrow emission-line width constraint. 9/25 objects show evidence for weaker than average He ii strengths. However, only three objects meet three out of four criteria, and none of these meets the weaker than average He ii strength. Thus, none of the objects in our comparison sample meets all of the criteria necessary for the formation of an observable ghost feature.

A comparison between our ghost-candidate and non-ghost samples suggests that the main difference between them lies in the strength of the absorption, with our ghost-candidate sample displaying more objects with strong N v absorption. Comparison of the geometric mean composite spectra of these two samples, and their ratio [Fig. 15, dashed (ghost candidate) and dotted (non-ghost comparison spectra) lines] indicates that the ghost-candidate composite shows significantly stronger absorption at lower velocities in all of the strong lines Lyα, N v, Si iv and C iv.

Figure 15

Upper panel: composite ghost-candidate (solid line) and comparison (dashed line) spectra. Lower panel: ratio of ghost-candidate to comparison spectra.

In order to test whether peaks are more common within the ghost zone than elsewhere, we repeat the ghost selection method using two ‘fake’ ghost zones (redward and blueward of the original ghost zone) in a similar fashion to North et al. (2006). These zones are created in precisely the same way as the ghost zone except that the red zone is centred at 4000 km s⁻¹, while the blue zone is centred at 8000 km s⁻¹ blueward of the C iv emission line. Of the 258 single-peaked objects, 82 have peaks within the ‘fake’ red zone and 50 have peaks within the fake ‘blue zone’. Since the original ghost zone contained 69 objects with single peaks, the evidence for an excess of objects with peaks at a preferred velocity is weak. That is, single peaks within the ghost zone are no more likely than single peaks at other velocities.

We have also examined the link, if any, between the peaks within the C iv absorption trough and N v absorption. In order to select N v BALs we use a modified BI (Weymann et al. 1991). We take the continuum to be equal to the mean flux between λλ1315 and 1330 Å, and require that the flux drops to below 90 per cent of this value continuously for >1000 km s⁻¹ between 0 and 7000 km s⁻¹ blueward of the N v emission line. In order to perform this test we require objects with redshifts in excess of 2.15. From a sample of 1747 objects with z > 2.15, we find 1258 N v BALs and 489 N v non-BALs. 27 per cent (340) of the N v BALs show multiple troughs in their C iv absorption compared to only 15.5 per cent (76) of the N v non-BALs. Similarly, 7.6 per cent (95) of the N v BALs and only 3.1 per cent (15) of the N v non-BALs show a single peak in the C iv absorption trough. Among the N v non-BALs we find (i) two objects with single peaks within the ghost zone, (ii) two objects with a single peak in the fake blue zone and (iii) no objects with a single peak in the fake red zone. For the N v BALs, we find (i) 26 objects with single peaks within the ghost zone, (ii) 14 with single peaks within the fake blue zone and (iii) 33 with single peaks within the fake red zone. While these results indicate a strong link between the presence of N v absorption and the mechanism responsible for producing features within the C iv absorption trough, line locking between Lyα and N v does not appear to be the dominant mechanism.

In summary, N v BALs are more likely to have multiple troughs within the C iv absorption than N v non-BALs (factor of 2). Further, N v BALs are also more likely to display single peaks within their C iv absorption than N v non-BALs. Approximately 25 per cent of the single-peaked objects are located within the ghost zone. However, similar numbers are found in both the blue and red fake ghost zones. Thus while strong N v absorption appears to be a strong requirement for the appearance of features within the C iv absorption trough of BALQSOs, there is no preferred velocity for the location of these features.

6 SUMMARY OF RESULTS

In an analysis of the largest sample of BALQSOs derived from DR5 of SDSS, we have identified 43 objects which show evidence for radiative driving, as indicated by the appearance of feature in the C iv absorption, and thought to be associated with the interaction between Lyα photons and N v ions. We have attempted to explore the reality of these ghost features by testing the physically motivated criteria, as set out by Arav, for the formation of observable features via this mechanism. Of the 21 objects for which we can test all of the criteria, none satisfies all of the conditions necessary for the formation of an observable feature. Two of the criteria, the width of the driving line and the strength of the UV continuum are particularly difficult to measure. Of the five objects that clearly pass the latter criteria having weak He ii, four have widths that are marginally larger than the upper limit imposed by Arav necessary to produce strong ghost signatures, while the other appears to have only weak emission lines. We recognize that because of the large uncertainty on the measurements of the He ii strength, eight further objects from the sample of 21 could potentially satisfy all of the criteria necessary for ghost formation. Addressing this issue will require higher S/N data than currently available for these objects.

However, taken at face value either the conditions necessary for line locking to occur is less stringent than previously proposed, or a large fraction of these objects are multitrough interlopers masquerading as ghost candidates. This possibility is supported by our finding that the ‘fake’ ghost zones bracketing the ‘real’ ghost zone contain comparable numbers of objects as the ‘real’ ghost zone. This suggests that the peaks observed within BAL troughs at the position of the ghost zone are not a physical effect of radiative line driving, but are instead independent of any systematic velocity.

Since our analysis has shed real doubt as to the existence of Lyα–N v line-locking features, in Appendix A we repeat for the interested reader, the sequence of tests by Korista et al. (1993) and re-investigate the incidence of double-trough phenomena in a large sample of BALQSOs. In summary, the results of these tests show that there is no strong evidence for an excess of double troughs bracketing the region of ghost formation, and that the evidence for line locking between Lyα photons and N v ions in BALQSO spectra is substantially lacking.

1

This is not the case for shadowing of N v by Lyα (Korista et al. 1993) where the separation of the troughs is constant but their velocities are determined by the velocity of the first trough.

2

Approximately 28 per cent (1019 out of 3552) of our BALQSOs show evidence for multiple troughs. This compares favourably with previous studies by North et al. (2006) who found that 58 out of 224 (26 per cent) of BALQSOs from SDSS EDR showed evidence of multiple troughs, and Korista et al. (1993) who found 16 out of 72 objects (22 per cent) showed multiple troughs.

3

We note that measuring the strength of a line according to its peak flux can be misleading, since it implies that the underlying line profiles are similar for all objects (and thus the total flux in the line simply scales with the peak flux). For example, a comparison between the C iv emission line of the composite spectra in Fig. 7 suggests that both our ghost-candidate and comparison samples display a narrower underlying C iv emission line. If the widths of the other emission lines scale with the C iv emission-line width, then we would expect both the Lyα and N v emission lines to be similarly narrow.

This work is supported at the University of Leicester and the University of Southampton by the Science and Technology Facilities Council (STFC). The SDSS website is http://www.sdss.org/.

Appendix

APPENDIX A: ON THE REALITY OF THE DOUBLE-TROUGH PHENOMENA

Since the criteria by which observable ghosts are formed appears less stringent than has previously been claimed, we have performed a series of sanity checks to test whether ghost features and double troughs in general are not simply a result of the random superposition of multiple absorption systems. The tests performed are similar to those described in Korista et al. (1993), though we note that our parent BALQSO sample is considerably larger and thus the quality of the statistics is a significant improvement over previous work.

A1 Random sampling

In the first test we compute the geometric mean residual intensity spectrum (the ratio of the geometric mean BALQSO composite divided by a geometric mean non-BALQSO composite spectrum) from two random samples (without replacement) taken from the BALQSO catalogue of Scaringi et al. (2009)[Fig. A1, sample 1 (solid line), sample 2 (dashed line)]. These samples (consisting of 1776 objects each) were generated by sorting the BALQSO catalogue into RA order, and then selecting even-numbered objects for the first sample, and odd-numbered objects for the second sample. The mean residual intensity spectra for both samples are virtually indistinguishable, aside from a small difference (∼15 per cent) in the residual intensity in the velocity range 3900–7800 km s⁻¹ blueward of line centre, which intriguingly, is precisely where the purported ghost feature should be found. We have checked to see whether there exists an excess of ghost-candidate spectra in either sample, and find that the ghost candidates are spread evenly amongst both samples, with 21 in sample A and 22 in sample B. While there are small apparent differences in the mean residual intensity spectrum in the region of the ghost feature, neither sample shows convincing evidence of a double-trough feature.

Figure A1

Geometric mean composite residual intensity spectrum of two randomly selected BALQSO samples (without replacement).

To explore these findings further, we have attempted to address whether there exists any substantive evidence for absorption components at preferred velocities. If ghost features are relatively common, we would expect an excess abundance of peaks within the absorption trough at velocities 5900 km s⁻¹ blueward of line centre. To address this question we have performed two additional tests. In these tests we work with fluxes instead of residual intensities to avoid complications introduced by uncertainties in fitting the C iv emission line blueward of line centre. In the first, we take the LVQ BALQSO sample of Scaringi et al. (2009), and compare the fluxes in three adjacent velocity bins (A, B and C) each with widths of 1000 km s⁻¹, where B is the reference bin and A and C represent the adjacent bins. Sliding these bins from a velocity of 0–30 000 km s⁻¹ blueward of C iv rest, we record the number of times the mean flux in the central bin (bin B) exceeds the mean flux in adjacent bins A and C by more than 3σ. To show the effects of selecting different binwidths over which the fluxes are measured, we repeat this exercise for binwidths of 500 and 1500 km s⁻¹. The results of this procedure are shown in Fig. A2 (upper left-hand panel). The distribution of peaks shows a prominent peak at zero velocity, which we associate with the location of the C iv emission-line peak. Ignoring this feature, the peak distribution is a generally smooth function of velocity with no indication of an excess number of peaks at 5900 km s⁻¹ blueward of line centre. The same analysis performed on our ghost-candidate sample, displays a similar peak at zero velocity, and in addition further peaks at around 6000 and 11 000 km s⁻¹. While the numbers of peaks at higher velocities is only marginally less than that at 5900 km s⁻¹, visual inspection of our ghost-candidate sample suggest that the significance of these secondary peaks is rather low.

Figure A2

Distribution functions showing the incidence of peaks as a function of velocity blueward of Civ rest wavelength (see text for details). Upper left-hand panel: LVQ BALQSOs, peak distribution functions (solid line – 500 km s⁻¹ bins, dotted line – 1000 km s⁻¹ bins, dashed line – 1500 km s⁻¹ bins). Upper right-hand panel: same as for upper left-hand panel but for the ghost-candidate sample. Lower panels: as above except now using slope comparisons.

We have also performed a variation on this method in which instead of measuring the intensities within each bin, we measure the mean gradient of the intensities within each of the bins. A peak is recorded if and only if the gradient in bin A is positive, the gradient in bin A is greater than the gradient in bin B, the gradient in bin B is greater than gradient of C and the gradient in bin C is negative. Fig. A2 lower panels show the result of this test. Again, we associate the peak at zero velocity with the peak of the C iv emission line. Thereafter the distribution of the locations of the peaks follows a smooth broad function with a maximum at approximately 15 000 km s⁻¹. As with the previous test, varying the binwidths between 500 and 1500 km s⁻¹ does not significantly alter the results.

These additional tests suggest that for the general BALQSO population, there is no substantive evidence that there exist preferred velocities for the location of peaks within the broad absorption troughs of BALQSOs. Thus if ghosts are real they are certainly rare, as was first suggested by Arav. We note that the incidence of single peaks (or double troughs) in our sample 258/3552 (7 per cent) is significantly smaller than that found by Korista et al. (1993, 22 per cent), likely a result of the small number statistics of this earlier study.

A2 Appearance of double troughs at other velocities

In the second test we verify whether there exist associated troughs (via line locking) at velocities other than systemic. We do this by aligning each spectrum according to the location of the red edge of the first BAL trough and thereby search for evidence of a bi-modal distribution of subtroughs. Here, the red edge is defined as in Korista et al. (1993) to be where the residual intensity falls below 0.75 and remains below this level for more than 1000 km s⁻¹. The resultant geometric mean residual intensity spectrum is shown in Fig. A3. As one might expect, the residual intensity is somewhat steeper at lower velocities, a result of aligning the first troughs in velocity space. However, the remainder of the spectrum is a smooth function of velocity, with no indication of additional subtroughs. Thus, as with Korista et al. (1993) we find no evidence for line-locking features at velocities other than systemic.

Figure A3

Residual intensity spectra after aligning all spectra to the position of the first (lowest velocity) absorption trough. We see no indication of a second trough at the known separation of Lyα–N v.

Unlike Korista et al. (1993) the distribution of velocities of the first trough (see Fig. A4) peaks at velocities between 1500 and 2000 km s⁻¹ blueward of line centre, thereafter the number of objects contributing to each velocity bin shows a smooth linear decline. Thus while there appears a preference for the onset of BALs toward lower velocities, there is no indication of other preferred blueward velocities. This again suggests that the incidence of the double-trough signature among the BALQSO population as a whole is generally low. We note that the distribution of redshifted troughs is similarly peaked toward lower velocities and may indicate a preferred launch radius for the out-flowing gas.

Figure A4

The frequency distribution of velocities of the first absorption trough.

A3 K–S test

In our third test, we compute the mean residual intensities in three separate velocity bins, one centred on the expected location of any ghost feature, the other taken from bins either side of this feature, i.e. velocity bins at 4000–5000, 5500–6500 (the reference bin) and 10 000–11 000 km s⁻¹, respectively. We then use the K–S test to check whether the distribution of residual intensities within each of these velocity bins belongs to the same population as the reference bin. The residual intensity spectra we use for this test are (i) the total sample of BALQSOs from Scaringi et al. (2009), the randomly selected samples 1 and 2 described above, our multitrough sample from rejection cut 2, our single-peak sample from rejection cut 3 and our candidate ghost sample. The results of this test are presented in Table A1. The residual intensity spectra of the samples are shown in Fig. A5.

Table A1

K–S tests on the multitrough phenomenon.

BALQSO sample	Sample size	K–S probabilities × 100 per cent
		Subtrough 1	Subtrough 2
LVQ BALs	3552	10−2	5.9
MTS	1019	5 × 10−3	23.8
MTS A	510	1.2	34.9
MTS B	509	2.1	4.1
Single-peak sample	258	3.7	41.4
Ghost-zone sample	69	0.2	0.1

Figure A5

Residual intensity spectra of (i) the multitrough sample, (ii) the single-peak sample and (iii) ghost-candidate sample. See text for details.

Table A1 gives the probability that the K–S statistic of the subtrough with the reference trough (i.e. the ghost zone) would be exceeded by chance if they were drawn from the same population. That is, lower K–S probabilities indicate more significant differences between the chosen velocity bin and the reference bin. Our analysis shows that only for our ghost sample do both subtroughs 1 and 2 indicate significant differences from the reference bin (K–S probabilities less than 0.1 per cent). This is as expected, since members of this sample were selected precisely because they showed significant ghost-like features. The single-peak sample shows no evidence that the residual intensity of the ghost zone is drawn from a different population to that of subtrough 2. This suggests that the single-peak sample does not have a systematic peak and that the position of peaks within the absorption is random and unrelated to Lyα–N v line locking.

A4 A Monte Carlo test for ghosts

Korista et al. (1993) described one further test for an excess of double-trough features bracketing the Lyα–N v line-locking region, which we repeat here for completeness. In this Monte Carlo simulation, we test how frequently the depth of the double-trough structure is exceeded by a random arrangement of residual intensity differences among each of the three velocity bins (see Korista et al. 1993 for details). For each spectrum, we can form six residual intensity differences from three velocity bins. For each spectrum in our BALQSO catalogue we randomly pick two from the six possible residual intensity differences and calculate their mean value, i.e. an average residual intensity difference. We then define 〈D〉_random to represent the mean of these average intensity differences across our BALQSO sample. 〈D〉_random is then compared to 〈D〉, formed by averaging the mean residual intensity differences between subtroughs 1 and 2 with the reference bin, again averaged across our whole sample. 〈D〉 can take on any value between +1 and −1. Values of 〈D〉 near +1 indicate a strong double-trough structure. Conversely, a value of 〈D〉 near −1 indicates a strong trough at the position of the reference bin. Values near zero are indicative of a smooth trough passing through the velocity bins.

To test how frequently a double-trough structure having a contrast at least as large as that observed in the sample mean could arise by chance, we repeat this process 10 000 times for each sample, and determine the number of times 〈D〉_random exceeds 〈D〉, normalized to 10 000. The results of these simulations are shown in Table A2. Since we only find strong evidence for the double-trough structure within our ghost zone sample (chosen for precisely that reason), we conclude that there is indeed no strong evidence for an excess of objects with subtroughs bracketing the ghost zone among the BALQSO population at large. Indeed, the high probabilities found for both the LVQ sample and for the multitrough sample, strongly suggest that multitrough features are randomly distributed in velocity. We are therefore conclude that many of the objects previously identified as strong ghost candidates in previous studies may simply be multitrough interlopers masquerading as ghosts due to the chance alignment of multiple unassociated absorption systems.

Table A2

Monte Carlo simulations of the double-trough phenomenon.

BALQSO sample	〈D〉	Monte Carlo probability × 100 per cent
LVQ BALs	−0.01197	98.88
MTS	−0.03918	99.8
Single-peak sample	0.007355	35.8
Ghost-zone sample	0.16932	0.0

A5 SDSS J101056.68+355833.3: Si iv ghost?

This is an object selected for our comparison sample as it shows no hint of a ghost feature within the deep broad C iv absorption. The spectra are shown in Fig. A6 along with the flux in velocity space relative to C iv and Si iv showing clearly the potential ghost feature in the Si iv absorption. SDSS J101056.68+355833.3 meets the criteria for strong emission due to C iv and N v emission exceeding 100 per cent of the continuum flux at their peak, shows clear N v absorption, has C iv FWHM measured as 3483 km s⁻¹ and an EW of He ii measured to be 3.58 ± 0.82 Å. These measurements place this object very close to the boundary on several criteria and it would not be surprising to see a ghost feature in this object. While no ghost feature can be seen in the C iv trough, there is a clear feature within the Si iv trough (∼6200 km s⁻¹) within the ghost zone. Because of the greater separation of the Si iv doublet (1933 km s⁻¹) in comparison to the C iv doublet (498 km s⁻¹) a ghost feature in the Si iv would be expected to be wider and weaker. This larger doublet separation is the reason for the wider ghost zone in the S iv BAL trough. The absence of a C iv feature could be due to an extremely high optical depth in the flow such that the flow remains optically thick even with the decrease in optical depth caused by the acceleration due to N v ions. However, the absorption troughs of C iv and Si iv look very similar with the exception of the potential ghost feature suggesting the optical depths are similar. There is a slight hint of an additional feature within the Si iv at ∼8200 km s⁻¹ which is in the appropriate region to be due to the lower wavelength doublet feature from Si iv.

Figure A6

SDSS J101056.68+355833.3: a possible Si iv ghost showing no evidence of a C iv feature along with close-ups in velocity space of the C iv (middle) and Si iv(top) ghost regions. Vertical lines represent the ghost zone.