Testing AGN Outflow and Accretion Models With C IV and He II Emission Line Demographics in 2 Quasars

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

MNRAS 000, 1–20 (2023) Preprint 10 January 2023 Compiled using MNRAS LATEX style file v3.

Testing AGN outflow and accretion models with C iv and He ii emission


line demographics in 𝑧 ≈ 2 quasars

Matthew J. Temple ,1★ James H. Matthews ,2,3 Paul C. Hewett ,3 Amy L. Rankine ,4
Gordon T. Richards ,5 Manda Banerji ,6 Gary J. Ferland ,7 Christian Knigge 6 and Matthew Stepney 6
1 Instituto de Estudios Astrofísicos, Universidad Diego Portales, Av. Ejército Libertador 441, Santiago 8370191, Chile
2 Department of Physics, Astrophysics, University of Oxford, Denys Wilkinson Building, Keble Road, Oxford OX1 3RH, UK
3 Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK
arXiv:2301.02675v1 [astro-ph.GA] 6 Jan 2023

4 Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK
5 Department of Physics, Drexel University, 32 S. 32nd Street, Philadelphia, PA 19104, USA
6 School of Physics & Astronomy, University of Southampton, Southampton SO17 1BJ, UK
7 Department of Physics and Astronomy, The University of Kentucky, Lexington, KY 40506, USA

Submitted to MNRAS 2022 December 22

ABSTRACT
Using ≈190,000 spectra from the seventeenth data release of the Sloan Digital Sky Survey, we investigate the ultraviolet
emission line properties in 𝑧 ≈ 2 quasars. Specifically, we quantify how the shape of C iv 𝜆1549 and the equivalent width (EW)
of He ii 𝜆1640 depend on the black hole mass and Eddington ratio inferred from Mg ii 𝜆2800. Above 𝐿/𝐿 Edd & 0.2, there is a
strong mass dependence in both C iv blueshift and He ii EW. Large C iv blueshifts are observed only in regions with both high
mass and high accretion rate. Including X-ray measurements for a subsample of 5,300 objects, we interpret our observations in
the context of AGN accretion and outflow mechanisms. The observed trends in He ii and 2 keV strength are broadly consistent
with theoretical qsosed models of AGN spectral energy distributions (SEDs) for low spin black holes, where the ionizing
SED depends on the accretion disc temperature and the strength of the soft excess. High spin models are not consistent with
observations, suggesting SDSS quasars at 𝑧 ≈ 2 may in general have low spins. We find a dramatic switch in behaviour at
𝐿/𝐿 Edd . 0.2: the ultraviolet emission properties show much weaker trends, and no longer agree with qsosed predictions,
hinting at changes in the structure of the broad line region. Overall the observed emission line trends are generally consistent
with predictions for radiation line driving where quasar outflows are governed by the SED, which itself results from the accretion
flow and hence depends on both the SMBH mass and accretion rate.
Key words: quasars: emission lines

1 INTRODUCTION ploring the properties which are observed to vary the most. Such
diversity in the observed quantities must ultimately be driven by
1.1 Observational context: spectroscopic properties of quasars some of the physics which we would like to use to better constrain
The spectroscopic properties of type-1 quasars have long been appre- both the growth of SMBHs and their effect on the galactic ecosystems
ciated for their potential to provide insight into the physical processes in which they reside. The most famous result of these investigations
responsible for luminous active galactic nuclei (AGN; Baldwin & is arguably the identification of the so-called ‘eigenvector 1’ (EV1),
Netzer 1978; Davidson & Netzer 1979; Kwan & Krolik 1981; Krolik which accounts for the largest amount of correlated variance in the
& Kallman 1988; Elvis 2000). These processes include the excitation optical spectra of low-redshift (𝑧 < 1) type-1 AGN spectra. Most
of various line- and continuum-emitting regions, and mechanisms authors now agree that the EV1 is driven by the mass-normalised
for launching outflows which might ‘feed back’ energy to their host accretion rate (the Eddington ratio), possibly convolved with some
galaxies. Such processes are ultimately powered by accretion onto orientation effect (Boroson & Green 1992; Wills et al. 1999; Sulentic
supermassive black holes (SMBHs; Lynden-Bell 1969), and thus de- et al. 2000; Shen & Ho 2014; Sun & Shen 2015; Sulentic & Marziani
pend primarily on the mass of the SMBH and the accretion rate, with 2015; Wolf et al. 2020).
potential second-order drivers including the spin of the SMBH and
Similarly, the ultraviolet emission features in quasar spectra also
the metal content of the accreting material.
show a rich phenomenology (Croom et al. 2002; Jensen et al. 2016;
The search for insight has gained much from identifying and ex-
Brodzeller & Dawson 2022). Early work by Baldwin (1977) showed
that the equivalent widths (EWs) of various ultraviolet lines, most
notably C iv 𝜆1549, were anti-correlated with the ultraviolet contin-
★ E-mail: [email protected] uum luminosity. Shang et al. (2003) showed that this ‘Baldwin effect’

© 2023 The Authors


2 M. J. Temple et al.
was independent of EV1 in 22 quasars with 𝑧 < 0.4, implying dif- For example, we cannot rule out the possibility that two objects with
ferent physical drivers for these correlations. Early observations also different 𝑀BH and accretion rate have similar (or indeed identical)
demonstrated that the centroid of the C iv emission line is commonly C iv emission.
shifted to the blue (Gaskell 1982; Wilkes 1984; Richards et al. 2002).
Within a sample of 87 Palomar-Green quasars, Baskin & Laor (2004,
2005) found that large C iv blueshifts were only seen in objects with
1.2 Theoretical context: AGN outflows and SEDs
high Eddington ratios, although not all quasars with high Eddington
ratios had large C iv blueshifts. The EV1 formalism was extended by Mass outflows from AGN can be launched by thermal pressure, mag-
Bachev et al. (2004) and Sulentic et al. (2007) to include the velocity netic forces, or radiation (Laha et al. 2021). Thermal winds can only
shift of C iv, again finding that large C iv blueshifts are seen only in be launched at large radii [𝑅 & 105 𝑅g ≈ 5 × (𝑀BH /109 𝑀 ) parsec]
so-called ‘Population A’ quasars with high Eddington ratios. with terminal velocities of order 100–1000 km s−1 (Begelman et al.
With the start of the Sloan Digital Sky Survey (SDSS; York et al. 1983; Woods et al. 1996; Mizumoto et al. 2019). Faster outflows with
2000), large samples of rest-frame ultraviolet quasar spectra became speeds > 2000 km s−1 , as commonly seen in broad high-ionization
available. Equally as important were methods to accurately charac- ultraviolet absorption features, are most likely launched on smaller
terize the systemic redshift of each quasar (Hewett & Wild 2010), (sub-parsec) scales. Magnetically driven winds (Blandford & Payne
which is necessary to infer the velocity shift of any emission features. 1982; Emmering et al. 1992; Konigl & Kartje 1994; Fukumura et al.
A notable work by Richards et al. (2011) summarized the state of 2010; Yang et al. 2021a) may be important in this context, but we cur-
the field a decade ago at the time of the seventh data release (DR7; rently lack predictive models for how such winds would translate into
Schneider et al. 2010) from SDSS. Using ≈35 000 quasar spectra, observable quantities (although see e.g. Bottorff et al. 2000; Chajet
Richards et al. (2011) confirmed the Baldwin effect and showed & Hall 2013). On the other hand, radiation line driving (Castor et al.
that the EW of C iv line also anti-correlates with the magnitude of 1975; Murray et al. 1995; Murray & Chiang 1995; Elvis 2000; Proga
the C iv blueshift: quasars with higher luminosities show, on aver- et al. 2000; Proga & Kallman 2004; Proga 2007; Risaliti & Elvis
age, weaker C iv emission which is more strongly blueshifted. C iv 2010; Elvis 2017; Nomura & Ohsuga 2017; Nomura et al. 2020; Zhu
blueshifts could be a signature of emission from ionized gas being et al. 2022) is intrinsically linked to the spectral energy distribution
driven away from the accretion disc along the line-of-sight to the ob- (SED) of the continuum which is responsible for both ionizing the
server (Leighly & Moore 2004), in which case the results of Richards transitions and then accelerating the flow by providing the source
et al. (2011) can be interpreted as brighter objects showing stronger of radiation pressure. By considering how the SED changes with
emission from outflowing gas and weaker emission from the virial- SMBH mass and accretion rate, authors such as Giustini & Proga
ized broad line region (BLR). We discuss this interpretation further (2019) have developed unifying frameworks which make testable
in Section 5.2.2, but do not assume anything about the origin of C iv predictions for luminous AGN based on the physics of radiation line
line shifts when presenting our observational results in Section 4.1. driven winds.
Richards et al. (2011) also demonstrated that the C iv properties are The ionizing continuum SED depends on the structure of the
strongly correlated with the EW of the nearby He ii 𝜆1640 emission accretion flow, which in turn is set by the SMBH mass 𝑀BH , the
line. More recent work by Rankine et al. (2020) has shown that mass-normalised accretion rate 𝑚¤ = 𝑀¤ BH / 𝑀¤ Edd and the SMBH
the correlations between the EW of He ii and both the EW and spin 𝑎 ∗ . Empirically, the optical–to–X-ray SEDs of AGN are seen to
blueshift of C iv are also present in quasars with broad absorption contain at least three distinct components (Elvis et al. 1994; Casebeer
features. We now know that the C iv and He ii properties are strongly et al. 2006; Leighly et al. 2007; Done et al. 2012; Jin et al. 2012).
correlated with the properties of other ultraviolet emission features First, any optically thick accretion disc will emit thermally, with
such as Ly 𝛼, N v, Si iv and O iv] (Temple et al. 2021b), Fe iii, Al iii, larger radii being cooler, giving rise to a multi-temperature blackbody
Si iii] and C iii] (Temple et al. 2020), as well as the optical [O iii] which is expected to peak in the near-ultraviolet (𝑀BH > 108 𝑀 ),
emission (Vietri et al. 2018; Coatman et al. 2019; Vietri et al. 2020), far-ultraviolet or soft X-rays (𝑀BH < 108 𝑀 ). This part of the SED
the strength of near infrared emission from dust at the sublimation is commonly observed to peak around 1100 Å (Shang et al. 2005;
temperature (Temple et al. 2021a), the strength of the far infrared Laor & Davis 2014; Stevans et al. 2014; Vanden Berk et al. 2020),
emission (Maddox et al. 2017), the radio detection fraction (Rankine although this peak might be expected to depend on 𝑀BH if larger
et al. 2021) and the strength of the 2 keV X-ray continuum (Kruczek SMBHs have accretion flows which truncate at lower temperatures
et al. 2011; Zappacosta et al. 2020; Timlin et al. 2020, 2021; Lusso (eq. 5.3.1 of Novikov & Thorne 1973). Second, a hot Comptonised
et al. 2021; Marlar et al. 2022; Rivera et al. 2022). Tentative links ‘corona’ emits a non-thermal power law which dominates the X-
have also been found between the C iv blueshift and the amount of ray continuum above 1 keV (Haardt & Maraschi 1991; Titarchuk
continuum reddening ascribed to nuclear dust (Calistro Rivera et al. 1994). Finally, a ‘soft excess’ is seen in the X-rays below ≈1 keV,
2021; Fawcett et al. 2022). which is usually attributed to an intermediate warm Comptonising
The existence of such correlations - between parameters which component (Petrucci et al. 2018). This soft excess may be a significant
trace emission at different wavelengths and from different physical contributor to the ionizing SED in the ≈100–1000 Å (≈10–100 eV)
regions - suggests that they are driven (either directly or indirectly) range, where many of the ultraviolet transitions are excited, but where
by changes in some of the fundamental physical parameters which direct observations of the continuum are not possible due to the high
govern the properties of a SMBH and its surrounding regions, such as opacity of neutral hydrogen along the line-of-sight.
the SMBH mass, spin, and accretion rate. The space spanned by C iv For a line-driven disc-wind to emerge, the system needs strong
blueshift and C iv EW therefore appears to be just as important as ultraviolet emission to produce sufficient radiation pressure, but also
EV1 in understanding the physics of luminous AGN. However, while a soft enough SED to avoid over-ionizing the gas (Murray et al. 1995;
the location of a given quasar spectrum on either EV1 or the C iv Higginbottom et al. 2014). For each relevant line, the flux at the line
blueshift–EW space must ultimately be a function of the fundamental energy combined with the line opacity determines the line-driving
SMBH parameters, there is no guarantee that such a function is boost beyond radiation pressure from Thomson scattering. The line
linear, or even injective (i.e. one-to-one with a well-defined inverse). opacity depends on the ionization state, which is primarily sensitive

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 3
to the flux beyond some ionization edge (64 eV for C iv). Line driving strength of He ii as a proxy for the strength of the ‘unseen’ EUV
results when this effect is summed across many lines, each with their continuum which is ionizing the BLR.
own energies, leading to a complex interplay between the flux of
the SED underneath all the relevant lines in the ultraviolet and the
flux of the SED beyond all the relevant ionization edges. Giustini & 1.4 This work
Proga (2019) suggest that both 𝐿/𝐿 Edd > 0.25 and 𝑀BH > 108 𝑀 The first goal of this paper is to provide an up-to-date summary of
are required to satisfy these criteria and hence to power a strong our knowledge of the ultraviolet spectral properties of type-1 quasars,
outflow through radiation line driving. Giustini & Proga (2019) also using the final data release (DR17) from the fourth iteration of SDSS.
expect the 𝑀BH dependence of the observed outflow properties to be This sample contains an order of magnitude more quasars than the
different above and below an 𝑚¤ of around 0.25, where they expect SDSS DR7 sample used by Richards et al. (2011). The large sample
the cold, optically thick accretion disc to extend down towards the size allows us to consider the emission properties as a function of
innermost stable circular orbit (ISCO) and replace the hot, optically both mass and Eddington ratio simultaneously, and thus provide a
thin, inner accretion flow which is present at lower accretion rates. test of current SED models and disc-wind theories, which is our
In other words, they require 𝐿/𝐿 Edd > 0.25 to ensure emission from second goal.
thermal disc emission dominates over that from the hot corona, to To best compare with theory and simulations, we present observed
accelerate a strong line-driven wind without over-ionizing the gas. quantities such as the C iv blueshift, He ii EW, and 𝛼ox as a function
With the quantity and quality of spectroscopic data which are now of three physical parameters: the ultraviolet continuum luminosity,
available from large surveys, these predictions from the Giustini & the SMBH mass estimated from the Mg ii 𝜆2800 emission line, and
Proga (2019) framework can be tested empirically. the inferred Eddington ratio. This relatively simple exercise has long
been used to gain insight into the physics of AGN (Dibai 1980), but
1.3 Observational probes of quasar SEDs is subtly different from purely empirical approaches which observe
trends in emission line properties (e.g. EV1 or the C iv blueshift-EW
From an observational viewpoint, it is relatively easy to constrain the plane) and then try to infer which underlying physical parameters are
SED of an unobscured type-1 AGN in the rest-frame infrared, optical driving those trends. By contrast, theoretical models make predic-
and X-ray wavebands, as photometric measurements can place direct tions for the SED and outflow properties as a function of the SMBH
constraints on the emission. For example, the strength of the rest- mass and accretion rate. In this work we confront such predictions
frame 2 keV X-ray continuum relative to the near-ultraviolet contin- directly with observations, showing that the ultraviolet emission lines
uum has been shown to anti-correlate with the ultraviolet continuum display different behaviour above a threshold of 𝐿 bol /𝐿 Edd ≈ 0.2,
luminosity in the so-called 𝛼ox –𝐿 2500 Å relation (Avni & Tananbaum consistent with predictions for radiation line-driven winds, and find-
1982, 1986; Steffen et al. 2006; Just et al. 2007; Lusso & Risaliti ing good qualitative agreement between state-of-the-art SED models
2016; Timlin et al. 2021), and the fractional contribution of the 2- and observed continuum tracers in regions of parameter space where
10 keV emission to the total bolometric power of the AGN is known the models were not calibrated.
to vary as a function of the accretion rate 𝑚¤ (Vasudevan & Fabian The structure of this paper is as follows. In Section 2 we present the
2007, 2009). However, while this X-ray waveband can make a not observational data, while in Section 3 we describe the SED models
insignificant contribution to the total emitted energy, it contributes a to which we compare. We present our key results in Section 4 and
negligible number of ionizing photons to the photoionization budget discuss their implications and limitations in Section 5. Throughout
of the BLR gas (see Appendix A). The number of ionizing photons this work, wavelengths are given in vacuum in units of Ångströms,
is instead dominated by photons at the ionization edges themselves, and we assume a flat ΛCDM cosmology with Ω𝑚 = 0.27, ΩΛ = 0.73
which is of the order of 10-100 eV for the ultraviolet BLR (e.g. pro- and 𝐻0 = 71 km s−1 Mpc−1 . Energies, frequencies and wavelengths
duction edges of 15 eV for Mg ii, 54 eV for He ii and 64 eV for C iv). are given in the rest-frame unless stated otherwise.
This extreme ultraviolet (EUV) part of the SED is not directly ob-
servable due to intervening absorption along the line-of-sight, but
plays a crucial role in the physics of the BLR. To add to the complex- 2 OBSERVATIONAL DATA
ity, the relative contribution of the warm Comptonising soft excess
to the total EUV emission is likely to be varying as a function of 2.1 Rest-frame ultraviolet spectra
𝑀BH and 𝑚, ¤ meaning that the observable 2 keV continuum may not
The first aim of this paper is to quantify the behaviour of He ii 𝜆1640
be a reliable proxy for the strength of the EUV SED at the ionization
and C iv 𝜆1549 as a function of SMBH mass 𝑀BH and Eddington
edges.
ratio 𝐿/𝐿 Edd . The 𝑀BH inferred from single-epoch measurements
C iv is a resonant doublet transition with a complicated ionic struc-
of C iv is known to be biased as a function of the emission line
ture, so the strength of C iv emission is not necessarily a good tracer
properties (Baskin & Laor 2005; Shen et al. 2008; Coatman et al.
of the ionizing SED. However, it is instead possible to probe the
2016, 2017; Mejía-Restrepo et al. 2018), so we will instead use the
EUV continuum using the He ii 𝜆1640 recombination line, which
velocity width of the Mg ii 𝜆2800 line to infer 𝑀BH . We construct
arises from a simple hydrogenic (i.e. single electron) system. Under
a sample of quasars from the SDSS with coverage of rest-frame
the assumptions that the He ii emitting region is in equilibrium and
wavelengths 1450-3000 Å to include C iv, He ii and Mg ii (Fig. 1).
that the He ii continuum is optically thick, the total rate of He ii-
The original selection of the SDSS DR17 quasar sample was de-
ionizing photons must balance the total number of recombinations
scribed by Lyke et al. (2020) and Abdurro’uf et al. (2022). We post-
such that each He ii 𝜆1640 line photon can be associated with an ion-
process each spectrum using a sky subtraction routine conceptually
izing continuum photon at or above 54 eV. This method was first used
similar to that described by Wild & Hewett (2005)1 . Systemic red-
by Zanstra (1929) to infer stellar temperatures using the strength of
Hydrogen recombination lines (section 5.10 of Osterbrock & Ferland
2006). Following previous works (Mathews & Ferland 1987; Baskin 1 Measurements of spectrum properties derived from observed-frame wave-
et al. 2013; Ferland et al. 2020; Timlin et al. 2021), we will use the lengths >6700 Å improve somewhat but none of the results, or conclusions,

MNRAS 000, 1–20 (2023)


4 M. J. Temple et al.

1044
100

Number of quasars per bin


2000
10−0.2 < L/LEdd < 10−0.1
1000
109.4 < MBH /M < 109.5
400
L/LEdd

200
10−1
100 10−1.1 < L/LEdd < 10−1.0
1043 109.7 < MBH /M < 109.8

Lλ [erg s−1 Å−1 ]


40
20
10
10−2
108 109 1010 10−0.5 < L/LEdd < 10−0.4
MBH [M ] 108.4 < MBH /M < 108.5

1042
3 CIV HeII
1700Å
fλ /f

2
HeII
10−1.6 < L/LEdd < 10−1.5
CIV 108.9 < MBH /M < 109.0 MgII
1
41
1500 1550 1600 1650 10
1400 1600 1800 2000 2200 2400 2600 2800 3000
Rest Wavelength [Å] Rest Wavelength [Å]

Figure 1. Top left panel: the distribution of our sample of 191 391 quasars with redshifts 1.5 < 𝑧 < 2.65 in the 𝑀BH –𝐿/𝐿Edd plane. Throughout this work, we
only consider hexagonal bins where there are five or more quasars per bin. By construction, the FWHM of Mg ii 𝜆2800 increases from top-left to bottom-right
of this parameter space, while the 3000 Å continuum luminosity increases from bottom-left to top-right. Right panel: composite spectra taken from different
regions of the 𝑀BH –𝐿/𝐿Edd plane, as indicated by coloured boxes in the top left panel. A diversity of emission line properties is seen with changing SMBH
mass and Eddington ratio. Bottom left panel: comparing the composite spectra in the region around C iv 𝜆1549 and He ii 𝜆1640. Here the spectra have been
normalised at 1700 Å and plotted on a linear y-axis. The EW of He ii can be seen to correlate with the profile of C iv: the high-mass, high-Eddington composite
in blue displays weak lines and blueshifted C iv while the low-mass, high-Eddington composite in green shows much stronger line emission with no blue excess
in C iv, consistent with fig. 11 of Richards et al. (2011) and fig. A2 of Rankine et al. (2020). The difference here is that, instead of being constructed from C iv
or C iii] emission properties, objects were included based on the FWHM of Mg ii and 𝐿3000 to represent regions of the 𝑀BH –𝐿/𝐿Edd plane, and also that the
larger sample from SDSS DR17 includes fainter objects such as those contributing to the composite in red. Composite spectra spanning the full range of the
𝑀BH –𝐿/𝐿Edd space will be made available as supplemental online-only material with the journal.

shifts are calculated as described in section 3 of Rankine et al. (2020). ≥3.0 over the rest-frame interval 1700-2200 Å. These criteria leave
Our redshift estimation routine uses the rest-frame 1600-3000 Å re- a sample of 191 391 quasars.
gion, deliberately excluding the C iv emission line, which is a key To compute the EW of C iv emission, a power law continuum is de-
difference compared to the approach employed in the SDSS quasar fined using the median flux in the 1445-1465 and 1700-1705 Å wave-
catalogues. The improved redshifts and sky-subtracted spectra will be length windows. This continuum is then subtracted from the spec-
described in a forthcoming publication by P. C. Hewett. To measure trum to isolate the line flux in the 1500-1600 Å wavelength region.
the emission line properties, we employ the spectral reconstructions The He ii EW is measured in the same way across the 1620-1650 Å
from the Mean-Field Independent Component Analysis (ICA) car- wavelength region, using windows at 1610-1620 and 1700-1705 Å to
ried out by Rankine et al. (2020), which we have successfully used in define the continuum model. The C iv emission line ‘blueshift’ is de-
our previous investigations into quasar emission line physics (Tem- fined as the Doppler shift of the wavelength bisecting the continuum-
ple et al. 2020, 2021a,b). The ICA-reconstructions provide a signif- subtracted line flux:
icant improvement in the measurement of emission line properties,  
𝜆 rest − 𝜆 median
reducing the impact of the modest signal-to-noise ratio in the origi- C iv blueshift ≡ 𝑐 × (1)
𝜆rest
nal spectra and the effect of weak absorption lines (e.g. intervening
or outflowing C iv 𝜆𝜆1548,1550 doublets). We exclude objects with where 𝑐 is the speed of light, 𝜆median is the rest-frame wavelength
broad low-ionization absorption features (LoBALs) and poor ICA of the observed line centroid, and 𝜆rest = 1549.48 Å is the mean
reconstructions. To include both C iv and Mg ii in the observed spec- rest-frame wavelength of the C iv 𝜆𝜆1548.19,1550.77 doublet.
trum, we limit our sample to redshifts 1.5 < 𝑧 < 2.65. Spectra from
before the start of the BOSS survey (MJD 55000) were observed 2.2 X-ray data
using the original SDSS spectrograph which had a more limited
wavelength coverage; for these objects we require 1.6 < 𝑧 < 2.2 to In addition to the rest-frame ultraviolet emission features, we can
ensure coverage of C iv and Mg ii. The spectra of quasars are required use the rest-frame 2 keV X-ray continuum emission to gain further
to possess a mean signal-to-noise ratio (per 69 km s−1 SDSS pixel) insight into the spectral energy distributions of the quasars in our
sample. We therefore cross-match our sample of 191 391 objects
to various X-ray catalogues from the literature, in order to build a
of this paper change if the original DR17 reductions of the spectra are used large sample of rest-frame 2 keV measurements. 4189 objects from
instead. our sample of 1.5 < 𝑧 < 2.65 objects with ultraviolet spectra are

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 5
included in the recent study of quasar X-ray properties by Rivera average. The error budget on our resulting 𝑀BH is dominated by the
et al. (2022), including 2820 with XMM-Newton observations from 0.55 dex uncertainty on the single-epoch estimator as described by
Lusso et al. (2020), 1337 with Chandra observations from Timlin Vestergaard & Osmer (2009).
et al. (2020), and 32 with XMM-Newton observations from Liu et al. We infer 𝐿 3000 by fitting a quasar SED model (Temple et al.
(2020). We make use of data from the second ROSAT All-Sky Survey 2021c) to griz photometry. For sources in SDSS DR16 we use the
(2RXS; Boller et al. 2016) for objects included in the SDSS DR16 SDSS photometry reported by Lyke et al. (2020), and for eFEDS-
SPIDERS programme (Dwelly et al. 2017; Comparat et al. 2020). The selected sources in SDSS DR17 we use the Hyper-Suprime Cam
flux limit for this survey is relatively bright so we use the Bayesian (HSC) photometry reported by Salvato et al. (2022). Eddington lu-
measurements described by Coffey et al. (2019) which account for minosities are calculated in the usual way, balancing the gravitational
the Eddington bias. A total of 45 objects from 2RXS are included in and radiation forces in a Hydrogen-only plasma, and assuming the
our sample. Finally, we include 13, 11, and 8 objects with Chandra dominant source of opacity is Thomson electron scattering:
observations from Timlin et al. (2021), Ni et al. (2018, 2022) and Fu 4𝜋𝐺 𝑀BH 𝑚 p 𝑐
 
𝑀BH
et al. (2022) respectively. These last three sub-samples were selected 𝐿 Edd = = 1.26 × 1038 erg s−1 . (4)
𝜎T 𝑀
to have high 𝐿 UV , weak C iv and strong C iv respectively, but the
number of quasars is small and our results would be unchanged if The Eddington ratio 𝐿 bol /𝐿 Edd (hereafter 𝐿/𝐿 Edd ) is then estimated
we were to exclude them. To improve our statistics for the number assuming a constant bolometric correction of 𝐿 bol = 5.15 × 𝐿 3000 .
of objects with X-ray detections, we augment our sample with 1059 We discuss this assumption further in Section 5.1.1, and show how
objects from the eROSITA Final Equatorial Depth Survey (eFEDS; our key observables depend directly on FWHM(Mg ii) and 𝐿 3000 in
Liu et al. 2022) with redshifts 1.5 < 𝑧 < 2.65 and with SDSS griz Appendix B.
photometry (Lyke et al. 2020) or HSC griz photometry from the Our sample of 191 391 quasars spans 2.5 dex in luminosity, with
eFEDS catalogue (Salvato et al. 2022). 𝐿 3000 ≈ 1044.5−47 erg s−1 and 𝐿 bol ≈ 1045−47.5 erg s−1 . We infer
The compilation results in a sample of 5325 quasars with measure- SMBH masses in the 108−10 𝑀 range and Eddington ratios from
ments of both their ultraviolet (2500 Å) and X-ray (2 keV) continuum 0.01 to unity, with the distribution of our sample shown in Fig. 1.
fluxes. We use directly observed X-ray fluxes with no spectral fitting,
though we have verified that the conclusions of this work would not
change if we excluded objects which may be affected by absorp- 3 MODELING THE QUASAR SED
tion. From these fluxes we compute luminosities assuming a consis-
tent cosmology (Section 1.4) across all sub-samples. We then infer Our second goal is to confront observational data with models for
𝛼ox , the logarithm of the ratio of the rest-frame 2 keV and 2500 Å accretion and outflow in quasars; more specifically, we aim to test
monochromatic luminosities: if the changes in observed emission line and continuum properties
    with 𝑀BH and Eddington ratio are consistent with theoretical models
𝛼ox = log10 𝜈𝐿 𝜈 − log10 𝜈𝐿 𝜈 , (2) for the SED of the ionizing continuum. To this end, we make use
2 keV 2500 Å
of qsosed2 (Kubota & Done 2018). We used the implementation of
as a measure of the relative strength of the X-ray emission in each qsosed in xspec (Arnaud 1996) to calculate SEDs, via the PyXSPEC
source. Objects with smaller (i.e. more negative) 𝛼ox have weaker python wrapper (Gordon & Arnaud 2021).
2 keV X-ray emission relative to the ultraviolet continuum. In qsosed, the radiation originates from three characteristic re-
gions: an outer thermal disc, an inner hot Comptonising ‘corona’
and an intermediate warm Comptonising component. These three
2.3 Black hole masses and Eddington ratios
regions are assumed to be radially stratified as defined by four criti-
We estimate SMBH masses using the single-epoch virial estimator cal radii: 𝑅ISCO < 𝑅hot < 𝑅warm < 𝑅out . The inner and outer radii
described by Vestergaard & Osmer (2009), using the full width at are defined by the radius of the innermost stable circular orbit 𝑅ISCO
half maximum (FWHM) of the Mg ii line: and the self-gravitation radius 𝑅out . The hot X-ray component origi-
 2   0.5 nates from 𝑅ISCO < 𝑅 < 𝑅hot , and has a luminosity set by the sum
FWHM(Mg ii) 𝐿 3000 of the directly dissipated power, 𝐿 diss,hot , and the seed photon lumi-
𝑀BH = 106.86 𝑀 . (3)
1000 km s−1 1044 erg s−1 nosity, 𝐿 seed . One of the key aspects of the model is the empirically
motivated assumption that the dissipated power is always 2 per cent
where 𝐿 3000 is the rest-frame monochromatic continuum luminosity
of the Eddington luminosity; this constraint defines the value of 𝑅hot .
𝜈𝐿 𝜈 at 3000 Å. This 𝑀BH estimator assumes a relationship between
The outer radius of the warm Comptonising component 𝑅warm is set
the radius of the Mg ii-emitting region and the observed 𝐿 3000 which
to be twice 𝑅hot . For 𝑅warm < 𝑅 < 𝑅out , the thermal disc component
is independent of the shape of the ionizing SED, or more generally,
is assumed to emit as described by Novikov & Thorne (1973).
independent of any changes in the accretion disc structure which
qsosed has four physical input parameters: the cosine of the incli-
may arise with changing 𝑀BH or accretion rate. We discuss this
nation, cos 𝑖, the SMBH mass, 𝑀BH , the dimensionless spin param-
assumption further in Section 5.1.2. We infer the FWHM of Mg ii
eter, 𝑎 ∗ , and the Eddington-scaled accretion rate, 𝑚¤ ≡ 𝑀¤ acc / 𝑀¤ Edd .
from our ICA reconstructions, which provide a more robust model
We fix cos 𝑖 = 0.5 and calculate grids of models in (𝑀BH , 𝑚) ¤ parame-
of the intrinsic Mg ii profile than a conventional Gaussian fit in low
ter space, for non- and maximally-spinning SMBHs 𝑎 ∗ ∈ (0, 0.998).
signal-to-noise (𝑆/𝑁) spectra. Using a sub-sample with 𝑆/𝑁 > 10,
We calculate models with 21 logarithmically-spaced grid points in
we have verified that our Mg ii FWHM measurements are consistent
each direction, spanning the ranges 8 ≤ log(𝑀BH /𝑀 ) ≤ 10 and
with those obtained from fitting a single Gaussian to Mg ii together
−1.65 ≤ log 𝑚¤ ≤ 0, corresponding to intervals of 0.1 and 0.0825 dex.
with an iron template (Vestergaard & Wilkes 2001) using the routine
To compare with observations, we take the input SMBH mass and
described by Shen et al. (2011). The key results of this paper would
not change if we were to instead use such a Gaussian model for Mg ii,
but there would be significantly more scatter in lower luminosity 2 https://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/
regions of parameter space where the spectral 𝑆/𝑁 is poorer on node132.html

MNRAS 000, 1–20 (2023)


6 M. J. Temple et al.
calculate the Eddington ratio from 𝐿 3000 using the same bolometric 1047 100
correction of 5.15 that we apply to the observational data (but see
MBH = 109 M
Section 5.1.1 and Appendix C). Here, and in Section 4.2, we present
models for only the non-spinning case, as these are in much better 1046
agreement with the data. We discuss the impact of SMBH spin and

νLν (erg s−1 )


system inclination in Section 5.2.3 and models with 𝑎 ∗ = 0.998 are
presented in Appendix D.


1045
Although the emission line properties must depend on the ioniz-
ing SED, the exact relationship between, for example, C iv EW and 10−1
the SED is complex due to a number of confounding factors such as
BLR geometry, density and radiative transfer. The relationship to any 1044
kinematic signatures such as C iv blueshift is even more complicated
and would require a physical model for the line formation region and
associated flow dynamics. A somewhat simpler case is the EW of
He ii 𝜆1640, which is a recombination line and therefore a reasonable 1047 1010

‘photon counter’. He ii has history as a tracer of the EUV continuum: ṁ = 0.15


for example, in cataclysmic variables Hoare & Drew (1991) applied
a modified Zanstra (1929) method to infer boundary layer temper- 1046
atures, and in quasars, Leighly (2004) note that a high He ii EW is

νLν (erg s−1 )

MBH [M ]
indicative of a strong X-ray continuum. Assuming Case B recombi-
nation, Mathews & Ferland (1987) give the He ii 𝜆1640 EW in terms 109
1045
of the 228 Å continuum flux. Their equation can be inverted to give
the proportionality
𝐹𝜈 (𝜆228) Ω 1044
He II
∝ EW(He ii 𝜆1640) , (5)
𝐹𝜈 (𝜆1640) 4𝜋
αox
where Ω/4𝜋 is the covering fraction and the proportionality constant
is dependent on the shape of the SED (Mathews & Ferland 1987 108
1014 1015 1016 1017 1018 1019
considered a power law in 𝐹𝜈 at 228 Å). In this work we assume, ν (Hz)
based on the above equation, that the observed He ii 𝜆1640 EW is a
reasonable proxy for the ratio of continuum luminosities 𝐿 228 /𝐿 1640 . Figure 2. Output SEDs in 𝜈𝐿𝜈 units from qsosed for 𝑎∗ = 0 and cos 𝑖 = 0.5.
In Fig. 2 we present output SEDs from qsosed, in which the three The vertical lines show, from left to right, the frequencies at 2500 Å, 1640 Å,
radially stratified components can be seen as separate ‘bumps’ in 228 Å (= 54 eV) and 2 keV which together determine 𝛼ox and the EW of
the spectrum. In these plots, we show how the model SEDs change He ii 𝜆1640. The He ii ionization edge at 54 eV (1.3×1016 Hz) lies in the EUV
as a function of Eddington-scaled accretion rate, 𝑚¤ (for fixed mass, regime where the intermediate warm Comptonising component in qsosed is
top panel) and SMBH mass, 𝑀BH (for fixed 𝑚, ¤ bottom panel). The most important, but the EW of He ii can also be seen to depend on the location
important frequencies for determining He ii EW (corresponding to of the peak of the ionizing SED. Top panel: SEDs with fixed SMBH mass of
228 Å and 1640 Å) and 𝛼ox (corresponding to 2500 Å and 2 keV) are 109 𝑀 and varying 𝑚 ¤ in logarithmic intervals. As 𝑚 ¤ increases the peak of
the SED moves to the blue, the luminosity increases, and the hard X-ray power
marked. Increasing 𝑚¤ increases the overall luminosity of the system
law spectral index becomes softer. Bottom panel: SEDs with fixed 𝑚 ¤ = 0.15
and pushes the peak of the outer thermal disc component to higher
and varying 𝑀BH in logarithmic intervals. As SMBH mass increases the peak
frequencies. Simultaneously, the hard X-ray slope becomes signifi- of the SED moves to the red, and the luminosity increases. A maximal spin
cantly softer and 𝐿 2keV only increases slowly. As a result, the higher analogue to this plot is shown in Fig. D1.
Eddington fraction objects are more X-ray weak relative to their ul-
traviolet flux. Increasing 𝑀BH also increases the total luminosity, but
now the peak of the thermal component moves to lower frequencies stronger emission line blueshifts. However, when considering the
and the hard X-ray slope stays fairly constant. In both panels of the observed C iv properties as a function of both 𝑀BH and 𝐿/𝐿 Edd
plot the peak of the SED can be found on either side of the low (bottom panel), we see a more complicated behaviour. To observe
frequency pivot points for both He ii EW and 𝛼ox , resulting in an the strongest C iv blueshifts (which are associated with the smallest
interesting interplay between these quantities and the fundamental EWs), we need to look at objects with both 𝑀BH & 109 𝑀 and
AGN parameters. 𝐿/𝐿 Edd & 0.2. Moreover, the contours of constant C iv blueshift
follow acute-angled ‘wedge’ shapes, which are somewhat orthogonal
to lines of constant luminosity (running diagonally top-left to bottom-
4 RESULTS right in the 𝑀BH –𝐿/𝐿 Edd space). At the same time, objects with the
strongest C iv EWs & 100 Å, which have strong symmetric emission
4.1 Observed properties in 𝑀BH –𝐿/𝐿 Edd space
with little or no blueshift, are found at high 𝐿/𝐿 Edd & 0.2 and
The first observational result from this work is the behaviour of relatively low 𝑀BH . 109 𝑀 .
the C iv 𝜆1549 emission line morphology as a function of SMBH To help us to understand the physical drivers behind the trends
mass (𝑀BH ) and Eddington ratio (𝐿/𝐿 Edd ), shown in Fig. 3. In the seen in C iv, in Figs. 4 and 5 we also show 𝛼ox and He ii EW across
left panel, we show the C iv emission line blueshift (as defined in the same 𝑀BH –𝐿/𝐿 Edd parameter space. The 𝛼ox behaviour is as
Eq. 1) and in the right panel the EW of C iv. In the top panels, expected from previous works (e.g. Mitchell et al. 2022), largely with
consistent with previous works, we find that more luminous quasars more luminous objects displaying relatively weaker X-ray emission
show weaker emission line strengths relative to the continuum and which is quantified by a more negative 𝛼ox . A more interesting result

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 7

0
1047 1047

200
120
[erg s−1 ]

[erg s−1 ]
0
175
1046 1046
3000Å

3000Å
100

0
150
νLν |

νLν |
CIV λ1550 blueshift [km s ]
−1
1045 1045

EW(CIV λ1550) [Å]


0
125
80

0
100
100 100

60

750
L/LEdd

L/LEdd
500

10−1 10−1
40
250

10−2 10−2 20
0

108 109 1010 108 109 1010


MBH [M ] MBH [M ]

Figure 3. The median observed C iv blueshift (left) and EW (right) in bins of SMBH mass, 3000 Å ultraviolet continuum luminosity (top) and Eddington ratio
(bottom). Data are shown only for bins which contain five or more objects. The C iv blueshift and EW are seen to anti-correlate: areas of parameter space with
strong blueshifts have weak EW and vice versa. 𝐿/𝐿Edd & 0.2 is a necessary but not sufficient condition for observing the largest C iv blueshifts. The strongest
C iv blueshifts are observed only at large SMBH mass and large Eddington ratio, while high EW C iv emission is observed at large Eddington ratio and smaller
mass. The Baldwin effect can be observed in the sense that objects with brighter 3000 Å luminosities tend to have weaker C iv EWs on average. However, the
C iv EW behaviour as a function of 𝑀BH and 𝐿/𝐿Edd shows that the underlying drivers of the Baldwin effect are more complicated than a simple dependence
on the ultraviolet luminosity.

is seen in the EW of He ii, which is even more striking than the 4.2 Comparison with model SEDs
behaviour seen in C iv. With the He ii EW, there is a clear transition
In the right-hand panels of Figs. 4 and 5 we show how 𝛼ox and
around 𝐿/𝐿 Edd ≈ 0.2, with both the strongest and weakest line
𝐿 228 /𝐿 1640 , respectively, vary with mass and Eddington fraction, as
emission only seen above this threshold. Below this Eddington limit,
modeled by qsosed. These plots can be compared to the respective
there is little change in the average line properties as a function
plots from the observational sample (left-hand panels), albeit with
of mass, but at 𝐿/𝐿 Edd & 0.2 there is a strong mass dependence
some caveats regarding bolometric corrections (Section 5.1.1) and
with diagonal wedge-shaped contours similar to those observed in
𝑀BH estimates (Section 5.1.2). In a qualitative sense, the models do
C iv. By contrast, the contours of constant 𝛼ox are much less closely
a reasonably good job of reproducing the trends observed in the data.
aligned with contours of constant C iv blueshift.
Focusing first on 𝛼ox , we can see that the general trend of decreasing
To test the robustness of these trends, we divide the 𝑀BH –𝐿/𝐿 Edd 𝛼ox with Eddington fraction is reproduced, and, in addition, the
into square bins of 0.1 by 0.1 dex and compute the median absolute gradient is stronger at high 𝑀BH , as observed in the data. To put
deviation (MAD) in each bin. The typical MAD is 290 km s−1 in this another way, in both the data and model results, the contour of
C iv blueshift, 13 Å in C iv EW and 0.5 Å in He ii EW. The typical fixed 𝛼ox curves around, from being nearly horizontal at high 𝑀BH
scatter within each bin is therefore significantly less than the dynamic to being closer to vertical at low 𝑀BH . The dynamic range of model
range in the average emission line properties shown in Figs. 3 and 𝛼ox values is comparable to that observed, but the models do not
5, meaning that one is unlikely to find individual objects which go produce soft enough spectra to match the data; 𝛼ox ≈ −1.9 can be
against the overall trend of the population. Dividing through by the found in some bins in the quasar sample but the minimum value of
median in each bin, the typical MAD/median in each bin is 0.29 𝛼ox in the models is −1.79.
and 0.24 for the He ii and C iv EWs respectively, meaning that the The comparison of the model 𝐿 228 /𝐿 1640 ratio and the observed
typical range of emission line EW within each 𝑀BH –𝐿/𝐿 Edd bin is He ii 𝜆1640 EW is also broadly encouraging, at least at relatively
a factor of 3.5 and 4 for He ii and C iv respectively, compared with high Eddington fractions. This finding is perhaps more interesting as
the dynamic range of more than a factor of six seen in the median the He ii EW is probing a portion of the SED that is not accessible
per-bin line properties. directly. The basic behaviour, of decreasing He ii EW with 𝑀BH at

MNRAS 000, 1–20 (2023)


8 M. J. Temple et al.

−1.2 −1.2
1047 1047

[erg s−1 ]
[erg s−1 ]

−1.3 −1.3

1046 1046

3000Å
3000Å

−1.4 −1.4

ν Lν |
νLν |

1045 1045

−1.5 −1.5

αox

αox
100 100
−1.6 −1.6
L/LEdd

L/LEdd
−1.7 −1.7
10−1 10−1

−1.8 −1.8

10−2 10−2
108 109 1010 −1.9 108 109 1010 −1.9
MBH [M ] MBH [M ]

Figure 4. Left panel: The median observed 𝛼ox in bins of SMBH mass, 3000 Å ultraviolet continuum luminosity (top) and Eddington ratio (bottom) for the
5325 objects from our sample with 2 keV X-ray measurements. Data are shown only for bins which contain five or more objects. Right panel: the predicted 𝛼ox
from low spin qsosed models in the same parameter space. The observations show good agreement with the models, with 𝛼ox more negative (i.e. more X-ray
weak) in objects with brighter ultraviolet luminosities. In Fig. D1 we show equivalent models but with high spin, which do not show such agreement with the
observations, suggesting that the 𝑧 ≈ 2 SDSS quasar population may be more consistent with low SMBH spins on average.

high Eddington fractions, is well matched by the models. The models 5.1 Key assumptions and limitations
also capture the diagonal contours of constant He ii EW, in which
the transition to low He ii EWs occurs at higher masses for higher 5.1.1 Bolometric corrections
Eddington fractions. As discussed above, at low Eddington fractions A large part of this work has attempted to quantify the ‘unseen’ ex-
(𝐿/𝐿 Edd . 0.2), something fundamentally switches in the data, with treme ultraviolet (EUV) portion of the SED which is not directly
gradients generally being shallower and along a different direction in observable, but which can instead be probed via the He ii emission
the parameter space. This relatively sharp change is not reproduced line. This portion of the SED contributes a significant amount to the
by the models, and may be telling us something fundamental about bolometric luminosity of a quasar. To estimate bolometric luminosi-
the quasar accretion process (see Section 5.2.1 for a discussion). ties (and Eddington ratios 𝐿 bol /𝐿 Edd ), we have assumed a constant
bolometric correction 𝑓bol ≡ 𝐿 bol /𝐿 3000 of 5.15, consistent with
previous works in the literature (e.g. Richards et al. 2006; Krawczyk
et al. 2013). However, we have also shown that the He ii strength
is changing as a function of 𝑀BH and 𝐿/𝐿 Edd , so we expect the
strength of the EUV continuum and hence the bolometric correction
to be varying with 𝑀BH and 𝐿/𝐿 Edd . Using our qsosed models,
5 DISCUSSION
we attempt to quantify this effect in Fig. 6. While our chosen value
We have quantified the average behaviour of C iv 𝜆1549, He ii 𝜆1640 of 𝑓bol = 5.15 lies within the range of values spanned by our grid
and 𝛼ox as a function of both 𝑀BH and 𝐿/𝐿 Edd , and compared our of model SEDs, there is variation of around a factor of two in 𝑓bol
observations with predictions from qsosed models. In this section depending on the values of 𝑀BH and 𝐿/𝐿 Edd we consider. While
we now discuss these results. We first outline the key caveats in this could in principle lead to systematic biases in our estimation of
our findings (Sections 5.1.1 and 5.1.2), before discussing possible 𝐿/𝐿 Edd , we show in Appendix C that these biases are likely to be
interpretations of our results within the context of AGN accretion and small compared to the magnitude of the trends we observe.
outflow theories (Sections 5.2.1 and 5.2.2). Finally, we discuss some We can however, briefly describe what might happen if we were to
wider implications and possible future applications (Section 5.3), adopt a non-constant bolometric correction when inferring 𝐿/𝐿 Edd
before summarizing our key conclusions in Section 6. from our observations. For two objects, both at 𝑚¤ = 0.2, the 𝑓bol

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 9

1047 1047

[erg s−1 ]
[erg s−1 ]

0.1
4
1046 1046

3000Å
3000Å

ν Lν |
νLν |

1045 1045

EW(HeII λ1640) [Å]

L228 /L1640
2
0.06

100 100
L/LEdd

L/LEdd
1 0.04
10−1 10−1

0.03

10−2 10−2
108 109 1010 0.5 108 109 1010
MBH [M ] MBH [M ]

Figure 5. Left panel: The median observed He ii EW in bins of SMBH mass, 3000 Å ultraviolet continuum luminosity (top) and Eddington ratio (bottom).
Data are shown only for bins which contain five or more objects. Right panel: the predicted strength of He ii ionizing photons at 228 Å relative to the 1640 Å
continuum from qsosed models. Above an Eddington ratio of ≈0.2, there is a strong trend as a function of SMBH mass, with high mass objects showing the
weakest He ii emission and low mass objects showing the strongest He ii emission. The model predictions show qualitatively similar behaviour in this region of
parameter space, explaining the diagonal contours in constant He ii. Below 𝐿/𝐿Edd . 0.2, the observed He ii displays much weaker trends, and does not agree
with the model predictions, suggesting that in this regime either the SED models are less accurate or the structure of the BLR is changing.

inferred from the qsosed models would be ≈6 and ≈3 for 𝑀BH = us that any random scatter or noise in our 𝑀BH estimates is small
108 𝑀 and 1010 𝑀 respectively. This would skew the observations enough not to ‘wash out’ the observed trends.
in Fig. 5, moving the location of the strongest He ii EW (at low
We used the FWHM of the Mg ii line to estimate 𝑀BH . Shen
𝑀BH ) to larger 𝐿/𝐿 Edd , more in line with the 𝐿/𝐿 Edd threshold at
et al. (2008) showed that such Mg ii-derived 𝑀BH estimates corre-
high 𝑀BH above which we see the weakest He ii and largest C iv
late tightly with those derived from H𝛽 across the full 108−10 𝑀
blueshifts. 
H𝛽  Mg ii
mass range, with the distribution of log 𝑀BH 𝑀BH following
a Gaussian with mean 0.034 and dispersion 0.22 dex. Shen & Liu
(2012) extended this analysis to higher redshifts and higher lumi-
5.1.2 Black hole mass estimates nosities, more appropriate for the objects in this work, and again
found that the Mg ii properties remained well correlated with those
As well as the assumption of a constant bolometric correction, we
of H𝛽. The Mg ii-derived 𝑀BH estimates we use in this work are
have used a single-epoch virial estimator to estimate SMBH masses
therefore unlikely to be biased compared to those which we would
throughout this work. The caveats associated with such estimates
have derived from a single-epoch H𝛽 measurement. The possibility
are numerous and have been reviewed by Shen (2013). Here we
remains, however, that such estimates are biased as a function of the
discuss some of the issues which are most relevant to our method
SED, or equivalently, as a function of 𝑀BH and 𝐿/𝐿 Edd .
and results. Most notably, the BLR radius–luminosity relation (as
encoded through the virial 𝑓 factor) may depend on the shape of the Early concerns about the universality of the BLR radius–
SED. Other uncertainties arising from (for example) orientation are luminosity relation were discussed by Kaspi et al. (2005) and Collin
likely to be random, in the sense that they will add scatter to our et al. (2006). More recently, various authors have tried to account
𝑀BH estimates but should not bias our results. While it is possible for possible SED-dependent biases in single-epoch 𝑀BH estimates
that our observed distribution of quasars in the 𝑀BH –𝐿/𝐿 Edd plane (Du & Wang 2019; Dalla Bontà et al. 2020; Fonseca Alvarez et al.
is not the same as the intrinsic distribution, the fact that we do still 2020; Martínez-Aldama et al. 2020), either using the accretion rate
observe such striking behaviour in the He ii and C iv emission line directly or by using the strength of optical iron emission 𝑅Fe ii as a
properties as a function of our inferred 𝑀BH and 𝐿/𝐿 Edd is telling proxy. However, Khadka et al. (2022a,b) and Yu et al. (2022a) find

MNRAS 000, 1–20 (2023)


10 M. J. Temple et al.

1010 𝐿/𝐿 Edd & 0.2) the strength of He ii emission is set directly by the
ionizing photon luminosity at 54 eV, and thus that He ii is providing
12 a probe of the EUV which is not directly observable. Moreover, the
observed He ii EW behaviour provides further evidence for the soft
excess to be an intermediate, warm Comptonising component which
10
behaves in the way in which the qsosed models predict. The strongest
and weakest 228 Å emission (relative to the 1640 Å continuum) are

MBH [M ]
both produced at high Eddington ratios, at low (≈ 108 𝑀 ) and high
fbol

8
109
(≈ 1010 𝑀 ) SMBH mass respectively.
However, the match between the observed He ii and the predicted
6 strength of the 54 eV ionizing luminosity is not perfect, especially
fbol = 5.15 in the 𝐿/𝐿 Edd . 0.2 regime. This mismatch might suggest a decou-
pling between the He ii EW and the 228 Å continuum flux at these
4 Eddington ratios, perhaps if changes in the BLR covering factor lead
to differences in the fraction of the continuum source which is re-
108 processed into emission lines. Another possibility is that the He ii
1044 1045 1046 1047
−1 continuum becomes optically thin, for instance if the density of the
νLν | [erg s ]
3000Å BLR were to decrease (which could indicate the absence of a dense
outflow). Alternatively, the mismatch could suggest that the SED
Figure 6. The predicted bolometric correction, 𝑓bol ≡ 𝐿bol /𝐿3000 , as a func- models are inaccurate in this Eddington ratio regime. Intriguingly,
tion of 𝐿3000 , from qsosed models. The points are colour-coded by 𝑀BH with this regime is similar to the region of the 𝑀BH –𝐿 2500 Å space where
a logarithmic normalisation, and points of constant mass are joined with solid Mitchell et al. (2022) find a mismatch between the observed and pre-
lines so that the trends with Eddington ratio can be understood by following dicted 𝛼ox . Either way, the observed switch in He ii behaviour above
individual lines from left to right. The adopted bolometric correction in this and below 𝐿/𝐿 Edd ≈ 0.2 (which is not reflected in the qsosed mod-
work, 𝑓bol = 5.15, is shown as a horizontal dashed line. 𝑓bol ranges from els) suggests that something fundamental is changing in the structure
≈ 3 − 10, and our adopted 𝑓bol is bounded by this range; however 𝑓bol does
of either the BLR or the accretion flow.
have a clear dependence on mass and luminosity in the model SEDs. Our
assumption of a fixed 𝑓bol could lead to an artificially reduced dynamic range
C iv is a resonant doublet with a more complicated ionic structure
in the inferred 𝐿bol at 𝑀BH = 108 𝑀 and an artificially increased range of than He ii. However, the close correspondence of the C iv blueshift
𝐿bol at 𝑀BH = 1010 𝑀 . and EW with the He ii EW, allied with the fact that the He ii behaviour
can be consistently explained with trends in the SED, suggests that the
C iv morphology is governed by accretion physics - specifically the
the opposite result, with the inclusion of 𝑅Fe ii having no effect on shape of the SED in the near and extreme ultraviolet regions. Given
the scatter in either the Mg ii or H𝛽 radius–luminosity relations. the proximity of the C iv and He ii ionization edges, at 64 eV and
While the literature remains divided, we argue it is still true that 54 eV respectively, this result is perhaps unsurprising. More notable
any SED-dependent bias in our single-epoch 𝑀BH estimates must be is the fact that the observed 𝛼ox behaviour does not correspond so
contained within the scatter on the BLR radius–luminosity relation, well to the C iv morphology, as demonstrated by the differing trends
i.e. within 0.3-0.5 dex. This scatter is smaller than the range spanned in Figs. 3 and 4. This shows how the 2 keV X-ray emission is a weaker
by our sample by a factor of ≈3, meaning that SED-dependent bi- proxy than He ii for the physical mechanisms which control the C iv
ases in our 𝑀BH cannot explain the observed trends presented in emission.
Section 4.1. Finally, the fact that our 𝐿/𝐿 Edd & 0.2 observations of He ii and
𝛼ox can be well explained by changes in the predicted model SEDs is
significant. The qsosed models were calibrated using observations
5.2 Quasar physics of just three objects with 𝑀BH = 108 𝑀 (Kubota & Done 2018), and
had not been critically assessed at higher SMBH masses prior to the
5.2.1 AGN accretion models recent work of Mitchell et al. (2022). The fact that the models agree
In Section 4.2, we compared our observations with predictions from with our observations across a relatively wide range of parameter
the qsosed models of Kubota & Done (2018). The predictions for 𝛼ox space was not necessarily to be expected. Moreover, the theoretical
made by these models have recently been tested over a much broader models show that the different phenomenological behaviours ob-
parameter space (𝑀BH ≈ 107−10 𝑀 and 𝐿 3000 ≈ 1043.5−47 erg s−1 ) served in 𝛼ox and He ii (Figs. 4 and 5) have natural origins in the
by Mitchell et al. (2022), who find that the qsosed model predicts SED, and that the observations presented in this work are broadly in
the optical and X-ray SED fairly well for 𝑀BH . 109 𝑀 , but that agreement with our current understanding of the accretion physics in
at higher masses the outer accretion disc spectra are predicted to be AGN.
too cool to match the observed data, especially at lower Eddington
ratios. This finding is consistent with our result (in Fig. 4) that the
5.2.2 AGN outflow models
2 keV emission is predicted to be slightly stronger (relative to the
2500 Å emission) than observed at 𝑀BH ≈ 109.5 𝑀 . In this subsection we test the predictions made by Giustini & Proga
In this work we have also quantified the He ii emission, which (2019), who summarize current understanding of AGN accretion and
provides a new, complementary probe of the ionizing SED across outflow mechanisms with a particular focus on the physical condi-
the 𝑀BH –𝐿/𝐿 Edd space. In Section 4.2 we found that, for 𝐿/𝐿 Edd & tions required to drive powerful winds from the accretion disc through
0.2, the observed He ii EW is qualitatively similar to the behaviour radiation line driving. We note again that the picture described by
of the 54 eV ionizing SED predicted by the qsosed models. The Giustini & Proga (2019) might not be the only plausible model for
observations are consistent with a scenario in which (at least for AGN outflows, but we choose to compare with their picture as it pro-

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 11
vides clear testable predictions within a well-defined framework. In observational results: the different behaviour of He ii EW and 𝛼ox
particular, Giustini & Proga (2019) suggest that both 𝐿/𝐿 Edd > 0.25 as a function of 𝑀BH and 𝐿/𝐿 Edd points to the presence of a third
and 𝑀BH > 108 𝑀 are required to power strong outflows from AGN spectral component in the EUV which can vary separately from the
through radiation line driving: below these thresholds the X-ray flux disc and corona.
is strong enough to over-ionize material and the ultraviolet flux will Other physical effects could also be at play. In particular, as the
be too low to accelerate a line-driven wind. accretion rate increases above 𝑚¤ & 0.3, we expect the disc structure
For the purposes of this comparison, we assume that any blue- to transition between geometries akin to slim discs and thin discs
wing excess in the C iv emission line profile is tracing an outflow (Abramowicz et al. 1988; Abramowicz & Fragile 2013). At low ac-
along the line-of-sight from the accretion disc, and hence that the cretion rates, slim discs are well approximated by the Novikov &
blueshift presented in Fig. 3 is a measure of the strength of emission Thorne (1973) thin disc solution, as used in qsosed, but we expect
from the outflowing wind (Leighly & Moore 2004; Richards et al. this to be less accurate as 𝑚¤ increases. In other words, the regime
2011). The origin of the C iv emission line blueshift is still debated in which qsosed appears to best match our data is also the regime
(see Gaskell & Goosmann 2013, 2016, for an alternative view), but a in which we might expect it to be least accurate. The origin of the
growing body of work is connecting the C iv emission morphology apparent transition around 𝐿/𝐿 Edd ≈ 0.2 in Figs. 3 and 5 is there-
with more unambiguous tracers of line driven winds. For example, fore still uncertain and further work is required to fully understand
the strengths and velocities of broad C iv absorption troughs have the interplay between AGN accretion flows, the ionizing SEDs they
been shown to correlate with the C iv emission blueshift (Rankine produce, and the outflows they drive.
et al. 2020; Rodríguez Hidalgo & Rankine 2022), and the velocities of Line-driven winds from high Eddington ratio AGN are often cited
narrow C iv line-locked ‘triplet’ absorption features are also strongly as a potentially important component of radiative-mode (quasar-
correlated with the emission blueshift (Rankine et al. in preparation). mode) feedback (Zubovas & King 2012). While difficult to observe
For the discussion in this subsection, we therefore assume that directly, such feedback modes are required to regulate galaxy growth
objects with larger C iv blueshifts have stronger disc winds. To be and explain the tight SMBH-galaxy correlations observed in the local
more precise, the C iv blueshift is taken as a measure of the strength universe (see Fabian 2012, for a review). However, most SMBHs in
of emission from outflowing gas relative to the strength of emission the local universe do not have masses above 109 𝑀 , so our results
from virialized gas at the systemic redshift. In this paradigm, the might suggest that the line-driven winds traced by C iv cannot have
trends seen in Fig. 3 are in good agreement with the picture proposed a significant effect on their host galaxies’ growth and co-evolution
by Giustini & Proga (2019). We see large (& 1000 km s−1 ) median as they never reach the SMBH masses required to launch strong
C iv blueshifts only in bins with high SMBH masses and high Ed- winds. There are at least two solutions to this apparent problem.
dington ratios. Furthermore, we do indeed see a more complicated First is that radiative-mode feedback could still be operating through
mass dependence above 𝐿/𝐿 Edd ≈ 0.2. High 𝐿/𝐿 Edd is therefore ionized gas outflows, but that the gas is in a different ionization phase
a necessary, but not sufficient, condition for observing large C iv and is not seen in C iv, but instead in other bands such as the X-ray
blueshifts, consistent with the results of Baskin & Laor (2005). ‘ultra-fast outflows’ (Laha et al. 2021). Second could be that quasar-
In detail, we only observe strong outflow signatures in objects mode feedback is only effective when coupled to dusty gas (Fabian
with 𝑀BH & 109 𝑀 , which is somewhat higher than the criterion et al. 2008; Ishibashi et al. 2018; Ricci et al. 2022), thus having
of 𝑀BH & 108 𝑀 proposed by Giustini & Proga (2019). Requiring most impact when the AGN is obscured by dust (Temple et al. 2019;
𝑀BH & 109 𝑀 and 𝐿/𝐿 Edd & 0.2 together ensures that the crite- Lansbury et al. 2020; Jun et al. 2021; Assef et al. 2022).
rion 𝐿 bol & 1045.5 erg s−1 is satisfied. Above this 𝐿 bol threshold,
Zakamska & Greene (2014) suggest that quasar winds are capable of
5.2.3 SMBH spin and system inclination
driving ionized gas (as traced by [O iii] 𝜆5008 emission) beyond the
escape velocity of the host galaxy. The kinematics of C iv and [O iii] In our qsosed modeling, we kept inclination fixed at cos 𝑖 = 0.5
are known to correlate (Coatman et al. 2019), and our observed C iv and only presented the non-spinning SMBH case, 𝑎 ∗ = 0. However,
blueshift behaviour is therefore consistent with the conclusion of Za- both of these parameters have an impact on the predicted SEDs. The
kamska & Greene (2014) that 𝐿 bol & 1045.5 erg s−1 is required for impact of SMBH spin is particularly pronounced; plots matching
quasar feedback to operate. those in the right-hand panels of Figs. 4 and 5 are presented in
For 108 𝑀 . 𝑀BH . 109 𝑀 and 𝐿/𝐿 Edd & 0.2, we see the Appendix D. The basic finding from the maximal spin models is
strongest He ii and strongest non-outflowing C iv line emission. One that the observed trends of 𝛼ox with mass and Eddington fraction
possible explanation for this behaviour would be that this emission are not reproduced, for reasons that are explained in Appendix D. In
represents ionized material which has been launched from the ac- fact, all of the maximal spin models have 𝛼ox & −1.5, meaning that
cretion disc, but lacks the ultraviolet luminosity to accelerate the the X-ray luminosity is always quite high compared to the optical
outflow, meaning that such material falls back and virializes instead and ultraviolet, and the observed soft spectra at high mass and high
of escaping. In such a scenario the strong symmetric emission from Eddington fraction are not reproduced for 𝑎 ∗ = 1. If there are a
high-ionization ultraviolet lines would represent a failed line-driven significant proportion of maximally spinning SMBHs in our quasar
wind, analogous to models of the low-ionization BLR which repre- sample, this would imply that the model predictions are not valid
sent a failed dust-driven wind (Czerny & Hryniewicz 2011; Baskin for high spin objects, potentially undermining many of the results
& Laor 2018, see also Elvis 2017). discussed in Section 5.2.1. Alternatively, if the qsosed models are
While we observe a reasonably good qualitative agreement be- correct, the good agreement at low spin and poor agreement at high
tween the C iv blueshift behaviour and the Giustini & Proga (2019) spin would imply that most SDSS quasars at 𝑧 ≈ 2 typically have low
predictions for line-driven winds, the reality is likely more compli- or moderate SMBH spins.
cated. In particular, Giustini & Proga (2019) do not consider any SMBH spin is most commonly estimated from broad iron line
emission from a ‘soft excess’. Instead they assume that the ionizing emission in the X-ray band (Reynolds 2019). Spin measurements
SED consists of just two components, emitted from a thermal disc tend to be rather high, with the majority of X-ray measurements in
and a hot corona. Such a simple model is unlikely to explain our AGN consistent with maximally spinning SMBHs. This apparent

MNRAS 000, 1–20 (2023)


12 M. J. Temple et al.
preference might initially appear to be inconsistent with our results. Schindler et al. 2020; Yang et al. 2021b; Lai et al. 2022; Wang et al.
However, there are a number of factors at work. First, discs around 2022). Such quasars are (by selection) very luminous, and generally
maximally spinning SMBHs have higher radiative efficiencies and display large C iv blueshifts. From our results, we would argue that
are thus more luminous. As shown in figure 3 of Reynolds (2019), comparative studies should match AGN samples not just in luminos-
this might lead to high spins being over-represented in a sample. ity, but in two independent parameters which trace 𝐿/𝐿 Edd and 𝑀BH .
One could also imagine further selection effects if spins are eas- Stepney et al. (in preparation) will discuss this further in a study of
ier to measure when they are close to maximal and the iron line C iv and He ii emission in SDSS quasars with redshifts 𝑧 > 3.5.
is broader. Second, the majority of spin measurements are at lower Current samples of 𝑧 & 6 quasars include a significant number of
masses (𝑀BH . 108 𝑀 ) than in our sample, with only a handful of objects with inferred 𝐿/𝐿 Edd > 1, which lie outside the parameter
spin measurements in our considered mass regime. In fact, there is space explored in this work. We have verified that the sample of 37
some tentative evidence for a decrease of SMBH spin with increasing quasar spectra with redshifts 𝑧 > 6.3 presented by Yang et al. (2021b)
mass (Sisk-Reynés et al. 2022), behaviour that is expected if accre- typically show narrower Mg ii profiles than 𝑧 ≈ 2 SDSS objects with
tion is coherent at low masses and more incoherent at higher masses, the same 3000 Å luminosities, suggesting smaller 𝑀BH and larger
as predicted by both semi-analytic models and hydrodynamic sim- 𝐿/𝐿 Edd (for a given 𝐿 3000 ) than the quasars characterized in this
ulations (King et al. 2008; Sesana et al. 2014; Zhang & Lu 2019; work. For such objects it is therefore not surprising that their typical
Bustamante & Springel 2019). If our results do indeed favour low or ultraviolet emission properties are different from the majority of the
moderate SMBH spins in luminous quasars, then they are consistent SDSS population at 𝑧 ≈ 2.
with this proposed trend, especially since it is the high-mass quasars
which cannot be matched by the maximally spinning qsosed models.
5.3.2 BLR metallicities
Any conclusions drawn here should be treated with caution, given (i)
the difficulties in obtaining reliable and unbiased spin measurements, The relative strengths of many ultraviolet emission lines are known
and (ii) the lack of knowledge about the impact of spin on the EUV to correlate with the C iv and He ii emission properties (Richards
and X-ray regions of the quasar SED. et al. 2011; Temple et al. 2020). In particular, the flux ratios of high
We fixed inclination in the qsosed models, adopting cos 𝑖 = 0.5. ionization ultraviolet lines such as C iv, N v 𝜆1240 and Si iv 𝜆1400
Following Copernican reasoning, we expect AGN to have a random, are tightly correlated with the C iv blueshift and He ii EW (Temple
isotropic distribution of viewing angles, in which case the mean et al. 2021b). Assuming no changes in density or ionization structure
viewing angle of all AGN is cos 𝑖 = 0.5. Factoring in obscuration or geometry of the BLR, changes in these line ratios are sometimes
by a putative ‘torus’ beyond some maximum inclination, and fore- taken to reflect changes in the metal content of the BLR (Nagao et al.
shortening/limb darkening of the disc continuum might be expected 2006). Such an interpretation, combined with the results in this work,
to bias this mean viewing angle to lower (more face-on) inclinations gives rise to a paradigm where the metal content of quasar BLRs is
(e.g. Krolik & Voit 1998; Matthews et al. 2017). If we were to adopt largest in objects with the largest 𝑀BH and 𝐿/𝐿 Edd , as noted by Xu
a different inclination in qsosed, we can think about how the model et al. (2018) and Śniegowska et al. (2021) respectively.
predictions would change. However, as shown in Temple et al. (2021b), the variation in these
The hard X-ray source in qsosed is isotropic, whereas the warm line ratios can be explained with changes in the density of the emitting
and thermal components have a disc-like geometry and thus pro- gas, and need not involve changes in metallicity (see also appendix
duced an observed luminosity ∝ cos 𝑖, such that lower inclinations A4 of Casebeer et al. 2006). In particular, the ultraviolet emission
have higher luminosities. The impact of inclination on the outputs line ratios seen in objects with large C iv blueshifts can be explained
from qsosed can thus be straightforwardly understood. Changing by emission from relatively dense gas which is located closer to
inclination from 𝑖1 → 𝑖 2 results in a fractional change in 𝐿 3000 of the ionizing source, while the line ratios in objects with high EW,
(cos 𝑖 2 /cos 𝑖 1 ). Since, for a given input 𝑚,
¤ we calculate 𝐿/𝐿 Edd from symmetric C iv emission are consistent with emission from less dense
𝐿 3000 , we obtain a linear scaling of the 𝑦-axis of the right-hand gas at larger radii. Given the trends seen in Figs. 3 and 5, this is a
panels of Figs 5 and 4 by the same factor. The change in 𝛼ox is much more natural explanation: objects with different SMBH masses
Δ𝛼ox = −0.3838 log10 (cos 𝑖2 /cos 𝑖1 ); adopting a lower inclination and accretion rates have different accretion flows, which give rise to
with cos 𝑖 = 0.75 would result in a more negative 𝛼ox in all simula- different EUV SEDs (as traced by He ii) and different kinematic and
tion bins by ≈ 0.07. Finally, the ratio 𝐿 228 /𝐿 1640 undergoes small density structures in the BLR (traced by the C iv blueshift and high
changes with cos 𝑖, but these are fairly uniform across the simulation ionization line ratios respectively). Under this alternative paradigm
grid and thus unimportant, given that the proportionality constant the BLR metallicity would be free to vary independently of 𝑀BH and
between 𝐿 228 /𝐿 1640 and He ii EW is not known. Furthermore, the 𝐿/𝐿 Edd , and need not be super-solar in the early universe (cf. Lai
He ii EW depends on the 𝐿 228 seen by the He ii gas, rather than et al. 2022; Wang et al. 2022).
the 𝐿 228 seen by a distant observer, meaning that the true inclination
dependence of He ii EW would depend on the BLR geometry. We
have explicitly checked that the anticipated changes in 𝐿 3000 and 𝛼ox 5.3.3 Quasar cosmology
are indeed reproduced in qsosed, except for small departures in the Quasars are visible out to large cosmological distances, and display
𝛼ox due to contamination of the 2 keV flux by the warm component. remarkably homogeneous behaviour across cosmic time. A grow-
ing body of work has proposed the non-linear scaling between 𝐿 UV
and 𝐿 2 keV (i.e. the 𝛼ox –𝐿 2500 Å relation) as a way to use quasars
5.3 Wider implications and future work as standardizable candles for cosmological measurements (Risaliti
& Lusso 2015, 2017, 2019; Lusso & Risaliti 2017; Salvestrini et al.
5.3.1 Comparison with other populations
2019; Lusso et al. 2020; Sacchi et al. 2022; Khadka & Ratra 2022).
Recent work has attempted to compare the ultraviolet emission prop- More recently, however, Petrosian et al. (2022) have shown that an
erties in high redshift (𝑧 & 6) quasars with their lower redshift ana- independent determination of the cosmological distance–redshift re-
logues (Mazzucchelli et al. 2017; Meyer et al. 2019; Shen et al. 2019; lation cannot be constructed simply by using the empirical correlation

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 13
between two continuum luminosities, as such luminosities must as- (i) As shown in previous works (Richards et al. 2011; Rankine
sume a cosmological model to be inferred from fluxes and redshifts, et al. 2020), the blueshift and EW of C iv correlate with the EW of
leading to circular reasoning. He ii. Objects with strong He ii have high EW C iv with little or no
In this work we have shown that it should be possible to break this blue excess, while objects with weaker He ii show smaller EW C iv
circularity by including information from the emission line prop- with larger C iv blueshifts.
erties. With knowledge of the Mg ii velocity width, and either the
(ii) We recover a Baldwin effect, but instead of simply correlating
He ii strength or the C iv properties, one should be able to locate
with the ultraviolet luminosity, we find that the C iv and He ii prop-
an object in the 𝑀BH –𝐿/𝐿 Edd plane, and hence infer the intrinsic
erties display more complicated trends in the 𝑀BH –𝐿/𝐿 Edd plane.
luminosity in a cosmology-independent way. By comparing to the
The dynamic range in He ii EW is greatest at Eddington ratios &0.2
observed fluxes one could then (in principle) infer a constraint on
(Fig. 5). The largest C iv blueshifts are only observed at high 𝐿/𝐿 Edd
the Hubble parameter 𝐻 (𝑧). However, further work is still required.
and high 𝑀BH , while the highest EWs are seen only at high 𝐿/𝐿 Edd
In particular, we need to build a sample of quasars with ultraviolet
and relatively low 𝑀BH (Fig. 3). Composite spectra from these two
emission line measurements which have independent measurements
extrema are shown in blue and green in Fig. 1.
of the luminosity distance, in order to calibrate our 𝑀BH –𝐿/𝐿 Edd
space in a cosmology-independent manner, in an analogous way to (iii) In contrast to the ultraviolet emission line properties, but
the use of the ‘inverse distance ladder’ to calibrate type Ia supernovae consistent with previous work in the literature, 𝛼ox displays a much
as standard candles (e.g. Freedman et al. 2001; Riess et al. 2021). simpler behaviour across the 𝑀BH –𝐿/𝐿 Edd plane (Fig. 4). 𝛼ox cor-
relates with the ultraviolet continuum luminosity in a more direct
manner than the emission lines, albeit in a much smaller sample. Fu-
5.3.4 Time variability and upcoming surveys ture data releases from eROSITA, SDSS-V and 4MOST will increase
Changes in 𝐿/𝐿 Edd for a quasar with fixed 𝑀BH will lead to changes the number of known quasars with X-ray data.
in the emitted spectrum, but such changes in SMBH fueling are (iv) Under the assumption that blueshifted C iv emission is trac-
expected to generally occur on the viscous time-scale, which is on the ing a disc wind accelerated by radiation line driving, we find our
of order of tens to thousands of years. However, SMBH accretion is results are consistent with the global scheme for accretion and out-
inherently stochastic and the emitted flux varies by a factor of a few on flow mechanisms proposed by Giustini & Proga (2019). In particular,
shorter time-scales of just years. The time-scale and amplitude of this an Eddington-scaled mass accretion rate 𝑚¤ & 0.25 is required for the
intrinsic ‘flickering’ are now known to depend on the SMBH mass formation of the strongest line-driven winds. Giustini & Proga (2019)
and accretion rate (Yu et al. 2022b), and this stochastic flickering will suggest that 𝑀BH > 108 𝑀 is also required to launch strong line-
contribute to the scatter within each binned region of our parameter driven winds, however we only observe the largest C iv blueshifts in
space (Section 4.1). objects with Mg ii-inferred 𝑀BH & 109 𝑀 . Strong line emission at
In terms of spectroscopic variability, Rivera et al. (2020) showed 𝑀BH . 109 𝑀 could perhaps indicate a ‘failed’ line-driven wind.
that individual SDSS-RM quasars with multiple epochs of spec-
troscopy (i.e. with fixed 𝑀BH ) can vary in essentially every direction (v) Absent large changes in the density or geometry of the broad
in the C iv blueshift–EW space, although objects with large blueshifts line region, the strength of He ii is probing the strength of 54 eV ion-
tend to show a change in blueshift and objects with strong EW show izing radiation in the ‘unseen’ portion of the ultraviolet SED. Above
a change in EW. In the near future, SDSS-V (Kollmeier et al. 2017) 𝐿/𝐿 Edd ≈ 0.2, we find that the EW of He ii is broadly consistent
will provide multi-epoch spectroscopic data for tens of thousands of with the qsosed model (Kubota & Done 2018). In other words, the
luminous quasars, providing new insights into AGN variability. relative strength of the 54 eV flux (which is photoionizing the broad
At the same time, surveys such as DESI (Alexander et al. 2022; line region) compared to the 1640 Å continuum is consistent with a
Chaussidon et al. 2022) and 4MOST (Merloni et al. 2019; Eltvedt relatively simple model where the peak temperature of the accretion
et al. 2022) will probe fainter, yielding spectra of lower luminosity disc blackbody changes as a function of SMBH mass and accre-
quasars than the sample investigated in this work, and future data tion rate and the strength of the ‘soft excess’ is adjusted to give the
releases from the eROSITA all-sky survey will include X-ray flux correct bolometric luminosity while keeping the strength of the hot
measurements for millions of AGN. Together these surveys will pro- coronal emission fixed at two per cent of the Eddington luminosity
vide new constraints on the spectroscopic properties and ionizing (as proposed by Kubota & Done 2018).
SEDs of luminous AGN across the 𝑀BH –𝐿/𝐿 Edd parameter space. (vi) Below 𝐿/𝐿 Edd ≈ 0.2, something appears to change in the
physics of the broad line region, with no strong C iv blueshifts ob-
served and much weaker trends in He ii. The simple SED models do
6 CONCLUSIONS not provide as good a match to the observed He ii trends, consistent
with the results of Mitchell et al. (2022) who find a discrepancy
We have investigated the rest-frame ultraviolet emission line proper-
between the observed and predicted 𝛼ox in the same region of the
ties in 191 391 SDSS quasars with redshifts 1.5 < 𝑧 < 2.65. We can
𝑀BH –𝐿/𝐿 Edd parameter space.
infer 𝛼ox , the logarithmic ratio of the rest-frame 2 keV and 2500 Å
luminosities, for 5325 quasars in our sample. Using the FWHM of (vii) Similar to Mitchell et al. (2022), we also find no strong
Mg ii 𝜆2800 as a proxy for the virial velocity, we quantify the av- evidence for high SMBH spins in our quasar sample: the zero-spin
erage properties of the C iv 𝜆1549 and He ii 𝜆1640 emission lines qsosed models provide an acceptable match to the SED probes across
across the two-dimensional space spanned by 𝑀BH and 𝐿/𝐿 Edd , and a significant portion of our observed parameter space while the max-
use these observations to confront qualitative predictions of when imally spinning models do not. If a significant fraction of our quasar
radiation-driven outflows should dominate kinetic feedback mecha- sample have maximally spinning SMBHs, this would suggest that
nisms (Giustini & Proga 2019) and theoretical SEDs arising from the qsosed model assumptions are not valid for high spin objects.
models of AGN accretion flows (Kubota & Done 2018). Our main Alternatively, taking the model results at face value would suggest
conclusions are: low or moderate spins in typical SDSS quasars at 𝑧 ≈ 2.

MNRAS 000, 1–20 (2023)


14 M. J. Temple et al.
ACKNOWLEDGEMENTS Bachev R., Marziani P., Sulentic J. W., Zamanov R., Calvani M., Dultzin-
Hacyan D., 2004, ApJ, 617, 171
We gratefully acknowledge useful discussions with Chris Done and Baldwin J. A., 1977, ApJ, 214, 679
Jake Mitchell. MJT thanks Chiara Mazzucchelli, Claudio Ricci and Baldwin J. A., Netzer H., 1978, ApJ, 226, 1
Roberto Assef for insightful comments, and Jinyi Yang for sharing Baskin A., Laor A., 2004, MNRAS, 350, L31
the sample of 𝑧 > 6 quasar spectra from Yang et al. (2021b). Baskin A., Laor A., 2005, MNRAS, 356, 1029
MJT acknowledges support from a FONDECYT postdoctoral fel- Baskin A., Laor A., 2018, MNRAS, 474, 1970
lowship (3220516). JHM acknowledges funding from the Royal So- Baskin A., Laor A., Hamann F., 2013, MNRAS, 432, 1525
ciety. ALR acknowledges support from UKRI (MR/T020989/1). Begelman M. C., McKee C. F., Shields G. A., 1983, ApJ, 271, 70
Funding for the Sloan Digital Sky Survey IV has been provided by Blandford R. D., Payne D. G., 1982, MNRAS, 199, 883
Boller T., Freyberg M. J., Trümper J., Haberl F., Voges W., Nandra K., 2016,
the Alfred P. Sloan Foundation, the U.S. Department of Energy Office
A&A, 588, A103
of Science, and the Participating Institutions. SDSS-IV acknowledges Boroson T. A., Green R. F., 1992, ApJS, 80, 109
support and resources from the Center for High-Performance Com- Bottorff M. C., Korista K. T., Shlosman I., 2000, ApJ, 537, 134
puting at the University of Utah. The SDSS web site is www.sdss.org. Brodzeller A., Dawson K., 2022, AJ, 163, 110
SDSS-IV is managed by the Astrophysical Research Consortium Bustamante S., Springel V., 2019, MNRAS, 490, 4133
for the Participating Institutions of the SDSS Collaboration including Calistro Rivera G., et al., 2021, A&A, 649, A102
the Brazilian Participation Group, the Carnegie Institution for Sci- Casebeer D. A., Leighly K. M., Baron E., 2006, ApJ, 637, 157
ence, Carnegie Mellon University, the Chilean Participation Group, Castor J. I., Abbott D. C., Klein R. I., 1975, ApJ, 195, 157
the French Participation Group, Harvard-Smithsonian Center for As- Chajet L. S., Hall P. B., 2013, MNRAS, 429, 3214
trophysics, Instituto de Astrofísica de Canarias, The Johns Hopkins Chaussidon E., et al., 2022, arXiv e-prints, p. arXiv:2208.08511
Coatman L., Hewett P. C., Banerji M., Richards G. T., 2016, MNRAS, 461,
University, Kavli Institute for the Physics and Mathematics of the Uni-
647
verse (IPMU) / University of Tokyo, the Korean Participation Group,
Coatman L., Hewett P. C., Banerji M., Richards G. T., Hennawi J. F.,
Lawrence Berkeley National Laboratory, Leibniz Institut für Astro- Prochaska J. X., 2017, MNRAS, 465, 2120
physik Potsdam (AIP), Max-Planck-Institut für Astronomie (MPIA Coatman L., Hewett P. C., Banerji M., Richards G. T., Hennawi J. F.,
Heidelberg), Max-Planck-Institut für Astrophysik (MPA Garching), Prochaska J. X., 2019, MNRAS, 486, 5335
Max-Planck-Institut für Extraterrestrische Physik (MPE), National Coffey D., et al., 2019, A&A, 625, A123
Astronomical Observatories of China, New Mexico State Univer- Collin S., Kawaguchi T., Peterson B. M., Vestergaard M., 2006, A&A, 456,
sity, New York University, University of Notre Dame, Observatário 75
Nacional / MCTI, The Ohio State University, Pennsylvania State Comparat J., et al., 2020, A&A, 636, A97
University, Shanghai Astronomical Observatory, United Kingdom Croom S. M., et al., 2002, MNRAS, 337, 275
Participation Group, Universidad Nacional Autónoma de México, Czerny B., Hryniewicz K., 2011, A&A, 525, L8
Dalla Bontà E., et al., 2020, ApJ, 903, 112
University of Arizona, University of Colorado Boulder, University
Davidson K., Netzer H., 1979, Reviews of Modern Physics, 51, 715
of Oxford, University of Portsmouth, University of Utah, Univer-
Dibai E. A., 1980, Soviet Ast., 24, 389
sity of Virginia, University of Washington, University of Wisconsin, Done C., Davis S. W., Jin C., Blaes O., Ward M., 2012, MNRAS, 420, 1848
Vanderbilt University, and Yale University. Du P., Wang J.-M., 2019, ApJ, 886, 42
For the purpose of open access, the authors will apply a Cre- Dwelly T., et al., 2017, MNRAS, 469, 1065
ative Commons Attribution (CC BY) licence to any Author Accepted Eltvedt A., et al., 2022, arXiv e-prints, p. arXiv:2211.07324
Manuscript version arising from this submission. Elvis M., 2000, ApJ, 545, 63
Elvis M., 2017, ApJ, 847, 56
Elvis M., et al., 1994, ApJS, 95, 1
Emmering R. T., Blandford R. D., Shlosman I., 1992, ApJ, 385, 460
DATA AVAILABILITY Fabian A. C., 2012, ARA&A, 50, 455
Fabian A. C., Vasudevan R. V., Gandhi P., 2008, MNRAS, 385, L43
The spectroscopic data underlying this article are available from
Fawcett V. A., Alexander D. M., Rosario D. J., Klindt L., Lusso E., Morabito
SDSS.3 The X-ray data sets underlying this article are available from
L. K., Calistro Rivera G., 2022, MNRAS, 513, 1254
the references given in Section 2.2. The composite spectra in Fig. 1 Ferland G. J., et al., 2013, Rev. Mex. Astron. Astrofis., 49, 137
are available as online supplementary material. Ferland G. J., et al., 2017, Rev. Mex. Astron. Astrofis., 53, 385
Ferland G. J., Done C., Jin C., Landt H., Ward M. J., 2020, MNRAS, 494,
5917
Fonseca Alvarez G., et al., 2020, ApJ, 899, 73
REFERENCES
Freedman W. L., et al., 2001, ApJ, 553, 47
Abdurro’uf et al., 2022, ApJS, 259, 35 Fu S., Brandt W. N., Zou F., Laor A., Garmire G. P., Ni Q., Timlin John D.
Abramowicz M. A., Fragile P. C., 2013, Living Reviews in Relativity, 16, 1 I., Xue Y., 2022, ApJ, 934, 97
Abramowicz M. A., Czerny B., Lasota J. P., Szuszkiewicz E., 1988, ApJ, 332, Fukumura K., Kazanas D., Contopoulos I., Behar E., 2010, ApJ, 715, 636
646 Gaskell C. M., 1982, ApJ, 263, 79
Alexander D. M., et al., 2022, arXiv e-prints, p. arXiv:2208.08517 Gaskell C. M., Goosmann R. W., 2013, ApJ, 769, 30
Arnaud K. A., 1996, in Jacoby G. H., Barnes J., eds, Astronomical Society Gaskell C. M., Goosmann R. W., 2016, Ap&SS, 361, 67
of the Pacific Conference Series Vol. 101, Astronomical Data Analysis Giustini M., Proga D., 2019, A&A, 630, A94
Software and Systems V. p. 17 Gordon C., Arnaud K., 2021, PyXspec: Python interface to XSPEC spectral-
Assef R. J., et al., 2022, ApJ, 934, 101 fitting program, Astrophysics Source Code Library, record ascl:2101.014
Avni Y., Tananbaum H., 1982, ApJ, 262, L17 (ascl:2101.014)
Avni Y., Tananbaum H., 1986, ApJ, 305, 83 Haardt F., Maraschi L., 1991, ApJ, 380, L51
Hewett P. C., Wild V., 2010, MNRAS, 405, 2302
Higginbottom N., Proga D., Knigge C., Long K. S., Matthews J. H., Sim
3 https://www.sdss4.org/dr17/ S. A., 2014, ApJ, 789, 19

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 15
Hoare M. G., Drew J. E., 1991, MNRAS, 249, 452 Petrosian V., Singal J., Mutchnick S., 2022, ApJ, 935, L19
Ishibashi W., Fabian A. C., Ricci C., Celotti A., 2018, MNRAS, 479, 3335 Petrucci P. O., Ursini F., De Rosa A., Bianchi S., Cappi M., Matt G., Dadina
Jensen T. W., et al., 2016, ApJ, 833, 199 M., Malzac J., 2018, A&A, 611, A59
Jin C., Ward M., Done C., Gelbord J., 2012, MNRAS, 420, 1825 Proga D., 2007, ApJ, 661, 693
Jun H. D., Assef R. J., Carroll C. M., Hickox R. C., Kim Y., Lee J., Ricci C., Proga D., Kallman T. R., 2004, ApJ, 616, 688
Stern D., 2021, ApJ, 906, 21 Proga D., Stone J. M., Kallman T. R., 2000, ApJ, 543, 686
Just D. W., Brandt W. N., Shemmer O., Steffen A. T., Schneider D. P., Chartas Rankine A. L., Hewett P. C., Banerji M., Richards G. T., 2020, MNRAS, 492,
G., Garmire G. P., 2007, ApJ, 665, 1004 4553
Kaspi S., Maoz D., Netzer H., Peterson B. M., Vestergaard M., Jannuzi B. T., Rankine A. L., Matthews J. H., Hewett P. C., Banerji M., Morabito L. K.,
2005, ApJ, 629, 61 Richards G. T., 2021, MNRAS, 502, 4154
Khadka N., Ratra B., 2022, MNRAS, 510, 2753 Reynolds C. S., 2019, Nature Astronomy, 3, 41
Khadka N., Martínez-Aldama M. L., Zajaček M., Czerny B., Ratra B., 2022a, Ricci C., et al., 2022, ApJ, 938, 67
MNRAS, 513, 1985 Richards G. T., Vanden Berk D. E., Reichard T. A., Hall P. B., Schneider
Khadka N., Zajaček M., Panda S., Martínez-Aldama M. L., Ratra B., 2022b, D. P., SubbaRao M., Thakar A. R., York D. G., 2002, AJ, 124, 1
MNRAS, 515, 3729 Richards G. T., et al., 2006, ApJS, 166, 470
King A. R., Pringle J. E., Hofmann J. A., 2008, MNRAS, 385, 1621 Richards G. T., et al., 2011, AJ, 141, 167
Kollmeier J. A., et al., 2017, arXiv e-prints, p. arXiv:1711.03234 Riess A. G., Casertano S., Yuan W., Bowers J. B., Macri L., Zinn J. C., Scolnic
Konigl A., Kartje J. F., 1994, ApJ, 434, 446 D., 2021, ApJ, 908, L6
Krawczyk C. M., Richards G. T., Mehta S. S., Vogeley M. S., Gallagher S. C., Risaliti G., Elvis M., 2010, A&A, 516, A89
Leighly K. M., Ross N. P., Schneider D. P., 2013, ApJS, 206, 4 Risaliti G., Lusso E., 2015, ApJ, 815, 33
Krolik J. H., Kallman T. R., 1988, ApJ, 324, 714 Risaliti G., Lusso E., 2017, Astronomische Nachrichten, 338, 329
Krolik J. H., Voit G. M., 1998, ApJ, 497, L5 Risaliti G., Lusso E., 2019, Nature Astronomy, 3, 272
Kruczek N. E., et al., 2011, AJ, 142, 130 Rivera A. B., Richards G. T., Hewett P. C., Rankine A. L., 2020, ApJ, 899, 96
Kubota A., Done C., 2018, MNRAS, 480, 1247 Rivera A. B., Richards G. T., Gallagher S. C., McCaffrey T. V., Rankine A. L.,
Kwan J., Krolik J. H., 1981, ApJ, 250, 478 Hewett P. C., Shemmer O., 2022, ApJ, 931, 154
Laha S., Reynolds C. S., Reeves J., Kriss G., Guainazzi M., Smith R., Veilleux Rodríguez Hidalgo P., Rankine A. L., 2022, ApJ, 939, L24
S., Proga D., 2021, Nature Astronomy, 5, 13 Sacchi A., et al., 2022, A&A, 663, L7
Lai S., et al., 2022, MNRAS, 513, 1801 Salvato M., et al., 2022, A&A, 661, A3
Lansbury G. B., Banerji M., Fabian A. C., Temple M. J., 2020, MNRAS, 495, Salvestrini F., Risaliti G., Bisogni S., Lusso E., Vignali C., 2019, A&A, 631,
2652 A120
Laor A., Davis S. W., 2014, MNRAS, 438, 3024 Schindler J.-T., et al., 2020, ApJ, 905, 51
Leighly K. M., 2004, ApJ, 611, 125 Schneider D. P., et al., 2010, AJ, 139, 2360
Leighly K. M., Moore J. R., 2004, ApJ, 611, 107 Sesana A., Barausse E., Dotti M., Rossi E. M., 2014, ApJ, 794, 104
Leighly K. M., Halpern J. P., Jenkins E. B., Casebeer D., 2007, ApJS, 173, 1 Shang Z., Wills B. J., Robinson E. L., Wills D., Laor A., Xie B., Yuan J.,
Liu T., et al., 2020, ApJS, 250, 32 2003, ApJ, 586, 52
Liu T., et al., 2022, A&A, 661, A5 Shang Z., et al., 2005, ApJ, 619, 41
Lusso E., Risaliti G., 2016, ApJ, 819, 154 Shen Y., 2013, Bulletin of the Astronomical Society of India, 41, 61
Lusso E., Risaliti G., 2017, A&A, 602, A79 Shen Y., Ho L. C., 2014, Nature, 513, 210
Lusso E., et al., 2020, A&A, 642, A150 Shen Y., Liu X., 2012, ApJ, 753, 125
Lusso E., et al., 2021, A&A, 653, A158 Shen Y., Greene J. E., Strauss M. A., Richards G. T., Schneider D. P., 2008,
Lyke B. W., et al., 2020, ApJS, 250, 8 ApJ, 680, 169
Lynden-Bell D., 1969, Nature, 223, 690 Shen Y., et al., 2011, ApJS, 194, 45
Maddox N., et al., 2017, MNRAS, 470, 2314 Shen Y., et al., 2019, ApJ, 873, 35
Marlar A., Shemmer O., Brotherton M. S., Richards G. T., Dix C., 2022, ApJ, Sisk-Reynés J., Reynolds C. S., Matthews J. H., Smith R. N., 2022, MNRAS,
931, 41 514, 2568
Martínez-Aldama M. L., Zajaček M., Czerny B., Panda S., 2020, ApJ, 903, Śniegowska M., Marziani P., Czerny B., Panda S., Martínez-Aldama M. L.,
86 del Olmo A., D’Onofrio M., 2021, ApJ, 910, 115
Mathews W. G., Ferland G. J., 1987, ApJ, 323, 456 Steffen A. T., Strateva I., Brandt W. N., Alexander D. M., Koekemoer A. M.,
Matthews J. H., Knigge C., Long K. S., 2017, MNRAS, 467, 2571 Lehmer B. D., Schneider D. P., Vignali C., 2006, AJ, 131, 2826
Mazzucchelli C., et al., 2017, ApJ, 849, 91 Stevans M. L., Shull J. M., Danforth C. W., Tilton E. M., 2014, ApJ, 794, 75
Mejía-Restrepo J. E., Trakhtenbrot B., Lira P., Netzer H., 2018, MNRAS, Sulentic J., Marziani P., 2015, Frontiers in Astronomy and Space Sciences,
478, 1929 2, 6
Merloni A., et al., 2019, The Messenger, 175, 42 Sulentic J. W., Zwitter T., Marziani P., Dultzin-Hacyan D., 2000, ApJ, 536,
Meyer R. A., Bosman S. E. I., Ellis R. S., 2019, MNRAS, 487, 3305 L5
Mitchell J. A. J., Done C., Ward M. J., Kynoch D., Hagen S., Lusso E., Landt Sulentic J. W., Bachev R., Marziani P., Negrete C. A., Dultzin D., 2007, ApJ,
H., 2022, arXiv e-prints, p. arXiv:2210.11977 666, 757
Mizumoto M., Done C., Tomaru R., Edwards I., 2019, MNRAS, 489, 1152 Sun J., Shen Y., 2015, ApJ, 804, L15
Murray N., Chiang J., 1995, ApJ, 454, L105 Temple M. J., Banerji M., Hewett P. C., Coatman L., Maddox N., Peroux C.,
Murray N., Chiang J., Grossman S. A., Voit G. M., 1995, ApJ, 451, 498 2019, MNRAS, 487, 2594
Nagao T., Marconi A., Maiolino R., 2006, A&A, 447, 157 Temple M. J., Ferland G. J., Rankine A. L., Hewett P. C., Badnell N. R.,
Ni Q., et al., 2018, MNRAS, 480, 5184 Ballance C. P., Del Zanna G., Dufresne R. P., 2020, MNRAS, 496, 2565
Ni Q., et al., 2022, MNRAS, 511, 5251 Temple M. J., Banerji M., Hewett P. C., Rankine A. L., Richards G. T., 2021a,
Nomura M., Ohsuga K., 2017, MNRAS, 465, 2873 MNRAS, 501, 3061
Nomura M., Ohsuga K., Done C., 2020, MNRAS, 494, 3616 Temple M. J., Ferland G. J., Rankine A. L., Chatzikos M., Hewett P. C.,
Novikov I. D., Thorne K. S., 1973, in Black Holes (Les Astres Occlus). pp 2021b, MNRAS, 505, 3247
343–450 Temple M. J., Hewett P. C., Banerji M., 2021c, MNRAS, 508, 737
Osterbrock D. E., Ferland G. J., 2006, Astrophysics of gaseous nebulae and Timlin J. D., Brandt W. N., Ni Q., Luo B., Pu X., Schneider D. P., Vivek M.,
active galactic nuclei Yi W., 2020, MNRAS, 492, 719

MNRAS 000, 1–20 (2023)


16 M. J. Temple et al.

109
Timlin John D. I., Brandt W. N., Laor A., 2021, MNRAS, 504, 5556
Titarchuk L., 1994, ApJ, 434, 570

4πν Jν
Vanden Berk D. E., Wesolowski S. C., Yeckley M. J., Marcinik J. M., Quash-
nock J. M., Machia L. M., Wu J., 2020, MNRAS, 493, 2745
Vasudevan R. V., Fabian A. C., 2007, MNRAS, 381, 1235
Vasudevan R. V., Fabian A. C., 2009, MNRAS, 392, 1124 108

4π ν Jν [erg cm-2 s-1]


Verner D. A., Ferland G. J., Korista K. T., Yakovlev D. G., 1996, ApJ, 465,
487
Vestergaard M., Osmer P. S., 2009, ApJ, 699, 800
Vestergaard M., Wilkes B. J., 2001, ApJS, 134, 1
Vietri G., et al., 2018, A&A, 617, A81 107
Vietri G., et al., 2020, A&A, 644, A175
Wang S., et al., 2022, ApJ, 925, 121
Wild V., Hewett P. C., 2005, MNRAS, 358, 1083
Wilkes B. J., 1984, MNRAS, 207, 73
Wills B. J., Laor A., Brotherton M. S., Wills D., Wilkes B. J., Ferland G. J., 106 ν Fν Softest SED (αox=-2.05)
Shang Z., 1999, ApJ, 515, L53 ν Fν Hardest SED (αox=-0.05)
Wolf J., et al., 2020, MNRAS, 492, 3580
He, C L
Wolfire M. G., Vallini L., Chevance M., 2022, ARA&A, 60, 247 H 2 keV
Woods D. T., Klein R. I., Castor J. I., McKee C. F., Bell J. B., 1996, ApJ, C K
461, 767 105 14
10 1015 1016 1017 1018 1019
Xu F., Bian F., Shen Y., Zuo W., Fan X., Zhu Z., 2018, MNRAS, 480, 345
106
Yang X.-H., Ablimit K., Li Q.-X., 2021a, ApJ, 914, 31
Yang J., et al., 2021b, ApJ, 923, 262

φν
York D. G., et al., 2000, AJ, 120, 1579
Yu Z., et al., 2022a, arXiv e-prints, p. arXiv:2208.05491
Yu W., Richards G. T., Vogeley M. S., Moreno J., Graham M. J., 2022b, ApJ,
936, 132 1000
Zakamska N. L., Greene J. E., 2014, MNRAS, 442, 784
φν [cm-2 Hz-1 s-1]

Zanstra H., 1929, Publs Dom. Ap. Obs. Victoria, 4, 209


Zappacosta L., et al., 2020, A&A, 635, L5
Zhang X., Lu Y., 2019, ApJ, 873, 101
Zhu Y., Bu D.-F., Yang X.-H., Yuan F., Lin W.-B., 2022, MNRAS, 513, 1141 1
Zubovas K., King A., 2012, ApJ, 745, L34

10−3
APPENDIX A: THE ROLE OF X-RAYS IN φν Softest SED (αox=-2.05)
PHOTOIONIZATION φν Hardest SED (αox=-0.05)

This appendix shows that the X-ray portion of the SED is an insignif- He, C L
H 2 keV
icant source of ionization for typical AGN emission line regions. 10−6
C K
This is surprising since the SEDs shown in Fig. 2 have a signifi-
1014 1015 1016 1017 1018 1019
cant fraction of their power at high energies. The ultraviolet–X-ray
hardness ratio (𝛼ox ) does correlate with emission line properties, as Hz
summarized above, but this is not likely a direct relation with the
observed X-rays.
To illustrate this point, we use the hardest and softest SEDs from Figure A1. The upper panel shows the hardest and softest SEDs presented in
this paper. The lower panel shows the flux of photons 𝜙𝜈 for the same models.
our model grid (Section 3), corresponding to 𝑚¤ = 0.027 and 𝑚¤ =
Some important energies are indicated by the hashed lines near the bottom
1.000 at 𝑀BH = 1010 𝑀 . These SEDs are shown in the top panel of
of each panel. These show the ionization potentials of ground-state H0 and
Figure A1, and have 𝛼ox = −0.05 in the hardest and 𝛼ox = −2.05 in He2+ , the L and K shells of C2+ , and 2 keV. The flux of ionizing photons is
the softest cases. orders of magnitude smaller at 2 keV compared to the flux at the He ii and
The photoionization rate for a given shell 𝑛 is C iv photoionization edges even for the hardest SED.
∫ ∞
Γ𝑛 = 𝜎𝜈 𝜙 𝜈 𝑑𝜈[s−1 ] (A1)
𝜈0 ∼6 dex fainter than the value near the peak. Photoionization is photon-
where 𝜈0 , 𝜎𝜈 , and 𝜙 𝜈 are the photoionization threshold of shell 𝑛, its counting and there are relatively few high-energy photons despite
energy-dependent photoionization cross section [cm2 ] , and the flux their significant energy.
of ionizing photons [cm−2 s−1 Hz−1 ] (Osterbrock & Ferland 2006). The photon flux is multiplied by the photoionization cross section
The total photoionization rate is the sum over all shells, to derive the photoionization rate (Eq. A1). We concentrate on C2+
∑︁ since photoionization of that ion produces C3+ and C iv 𝜆1549 emis-
Γtotal = Γ𝑛 [s−1 ] (A2) sion. The shell-dependent cross sections for photoionization of C2+ ,
𝑛 taken from Verner et al. (1996), are shown in Fig. A2. Both the 1s2 K
The flux of ionizing photons 𝜙 𝜈 enters in the photoionization shell in the X-ray and the lower energy 2s2 L shell are shown. Both
rate (Eq. A1). This is the ratio 𝜙 𝜈 = 4𝜋𝜈𝐽𝜈 /(ℎ𝜈 2 ) and is shown in shells have two electrons and, as expected, the peak photoionization
the lower panel of Fig. A1. The photon flux near 2 keV is typically cross sections are similar.

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 17

C2+ cross section


Light-matter interaction rate
109
10−18
K
L
106
σphoto [cm2]

φν σν [s-1]
10−19

1
10−20
φν σν Softest SED (αox=-2.05)
φν σν Hardest SED (αox=-0.05)
10−3
He, C L
H 2 keV
−21
10 C K
1014 1015 1016 1017 1018 1019 1014 1015 1016 1017 1018 1019
Hz Hz
Figure A3. This shows the rate that photons interact with matter, the product
Figure A2. The K and L shell photoionization cross sections for C2+ are
of the flux of photons and the total gas opacity. X-rays interact with matter
shown as a function of energy. The cross section are from the calculations by
with a rate about ∼ 7 dex slower than the rate near the peak rate around
Verner et al. (1996). These, together with the flux of photons (lower panel of
50 − 912Å.
Figure A1), determine the photoionization rate (Equation A1) and the effects
of light upon matter.
since emission from the gas which absorbed the softer radiation
would be far stronger. This is discussed in Section 4.1 of Ferland
Table A1. Photoionization rates for H0 and the K and L shells of C2+ . et al. (2013). The ultraviolet luminosity of a realistic SED has more
power than the relatively hard X-ray portion that drives an XDR.
The full SED striking a cloud produces successive H+ /H0 /H2 layers,
Shell ΓSoftest [s−1 ] ΓHardest [s−1 ] which are brighter than the deep X-ray heated regions. Emission
from regions powered by lower-energy light would dominate over
H0 K 8.24e+00 5.52e+00 the XDR.
C2+ L 2.66e-01 4.66e-01
C2+ K 4.49e-03 2.71e-02
Correlations between the X-ray portion of the SED and emission-
line properties are observed. These are likely due to other correlations
with the softer portions of the SED, as evident in Figure 2, and not
the X-rays themselves.
The photoionization rates for H0 and the two shells of C2+ are
listed in Table A1. The C2+ K-shell rate is 17 to 60 times smaller
then the L-shell rate. Both are ∼ 10 − 30 times smaller than the H0
photoionization rate. From this comparison we expect that the effects APPENDIX B: ADDITIONAL OBSERVATIONS
of the EUV and XUV will be more important than the X-ray, mainly
In this Appendix we present additional observational results. First,
due to the larger number of softer photons.
in Fig. B1 we present the He ii EW, the C iv blueshift and 𝛼ox as a
Figure A3 shows the rate at which photons interact with matter for
function of the FWHM of Mg ii 𝜆2800 and the 3000 Å continuum
our two reference SEDs and a solar composition. Calculations are
luminosity. These two parameters are measured directly from the
done with Cloudy version 22.01, as last described by Ferland et al.
SDSS spectroscopy and photometry respectively. By contrast, the
(2017). Cloud parameters are typical of the C iv emitting region of
plots in the main text show observed properties as a function of
an AGN. The vertical axis is the total light-matter interaction rate at
0.5
a particular frequency and is the product of the photon flux and the 𝑀BH ∝ 𝐿 3000 FWHM2Mg ii (B1)
total gas opacity, evaluated for the appropriate chemical composition
and degree of ionization. The 50-912 Å ultraviolet region is ∼7 dex and
more interactive than 2 keV X-rays. 0.5
𝐿/𝐿 Edd ∝ 𝐿 3000 /𝑀BH ∝ 𝐿 3000 FWHM−2
Mg ii . (B2)
The result that the X-rays hardly matter at all is surprising given the
shape of the SEDs in the upper panel of Figure A1. This is because Given that both 𝑀BH and 𝐿/𝐿 Edd depend on the observed parameters
photoionization is photon counting, and the paucity of X-ray photons FWHMMg ii and 𝐿 3000 , this might lead to induced correlations in the
cannot make up for their great energy. High-energy photons would 𝑀BH –𝐿/𝐿 Edd space. However, in practice our inferred 𝑀BH –𝐿/𝐿 Edd
dominate the physics if softer parts of the SED were extinguished space is simply a rotation and reflection of the FWHMMg ii –𝐿 3000
so only X-rays strike the gas. Indeed, this is the ‘XDR’ model of space, where we (reassuringly) still see clear trends. Furthermore, we
atomic and molecular regions of clouds exposed to ionizing radiation see the same He ii behaviour as a function of FWHMMg ii and 𝐿 3000
(Wolfire et al. 2022). It would be difficult to detect this XDR emission in the X-ray detected sub-sample as in our full sample, meaning that

MNRAS 000, 1–20 (2023)


18 M. J. Temple et al.

1047 All sources 1047 X-ray detected sources

4 4

EW(HeII λ1640) [Å]

EW(HeII λ1640) [Å]


[erg s−1 ]

[erg s−1 ]
1046 2 1046 2
3000Å

3000Å
νLν |

νLν |
1 1
1045 1045

0.5 0.5
−1.2

0
125 150 175 200
47 47
10 10

CIV λ1550 blueshift [km s−1 ]


0
−1.3

0
[erg s−1 ]

[erg s−1 ]
−1.4

1046 0 1046 −1.5

αox
0
100
3000Å

3000Å
−1.6
750
νLν |

νLν | −1.7
500

1045 1045
250

−1.8
0

−1.9
2500 5000 7500 10000 2500 5000 7500 10000
−1 −1
FWHM(MgII) [km s ] FWHM(MgII) [km s ]

Figure B1. Observed quasar properties as a function of the FWHM of Mg ii 𝜆2800 and the 3000 Å continuum luminosity. Left panel: The median He ii EW
(top) and C iv blueshift (bottom) in our full sample of 191 391 objects. Right panel: The median He ii EW (top) and 𝛼ox (bottom) in our sub-sample of 5325
X-ray detected sources. The He ii behaviour is identical in both panels (modulo the sample size), suggesting that our X-ray detected sub-sample is not biased
in terms of its ultraviolet emission properties. Moreover, clear differences are seen in the behaviour of He ii and 𝛼ox within the X-ray subsample: the strongest
He ii emission is seen only at low Mg ii FWHM while the strongest 2 keV X-ray emission is seen only at the lowest 3000 Å luminosities.

our X-ray detected objects are not obviously biased compared to our paring our qsosed models with data we tried to match scalings and
full sample. biases in the data introduced by the fixed bolometric correction by
Second, we show the 2 keV X-ray continuum and He ii 𝜆1640 line applying appropriate transformations to the qsosed outputs. For the
luminosities in Fig. B2. Assuming no changes in the BLR covering SMBH mass, the estimate in the data from the Mg ii line width but
factor, and that the He ii continuum is optically thick, 𝐿 He ii can also requires an estimate of the line formation radius, for which we
be taken as a proxy for the continuum luminosity at 54 eV. These follow the usual method and assume that the BLR radius scales as
two observables show qualitatively different behaviour: contours of 𝑅BLR ∝ 𝐿 1/2 . The 𝐿 in this expression should really be some appro-
constant 𝐿 2 keV are largely aligned with lines of constant 𝑀BH , which priate ionizing luminosity, but 𝐿 bol is normally used and we follow
is consistent with the assumption in qsosed that the hard X-ray power this convention. As a result, the bolometric correction enters into the
law component emits a constant fraction of the Eddington luminosity. SMBH mass estimate and implies a bias in the SMBH mass estimates
He ii behaves in a much more complex manner, with the gradient with respect to the true SMBH mass by factor of ( 𝑓bol /5.15) −1/2 .
vector of increasing 𝐿 He ii changing depending on the location in the As a result, when plotting 𝑀BH along the 𝑥-axis of Figs. 5 and 4, we
𝑀BH –𝐿/𝐿 Edd space. apply the scaling

𝑀BH = ( 𝑓bol /5.15) −1/2 𝑀BH, q , (C1)

APPENDIX C: BOLOMETRIC CORRECTIONS where 𝑀BH, q denotes the input qsosed grid value (the ‘true’ SMBH
mass). For 𝐿/𝐿 Edd , correction factors appear in both the numer-
In the data, we apply a fixed bolometric correction of 5.15 to estimate ator and denominator. 𝐿 Edd ∝ 𝑀BH , introducing a bias factor
𝐿 bol from 𝜈𝐿 𝜈 at 𝜈 = 3000Å. In reality, the bolometric correction ( 𝑓bol /5.15) −1/2 into the Eddington ratio estimate, while the numera-
will vary as a function of 𝑀BH and 𝐿/𝐿 Edd . We discussed the vari- tor is 𝐿 bol and so contains a straightforward bias factor of 𝑓bol /5.15.
ation of the bolometric correction from the qsosed model grid in As a result, the relationship between the 𝐿/𝐿 Edd plotted in Figs. 5
Section 5.1.1, showing a range in 𝑓bol by a factor of ≈ 2 − 3, where and 4, and the dimensionless, Eddington-scaled accretion rate used
𝑓bol ≡ 𝜈𝐿 𝜈 | 3000Å /𝐿 bol is calculated from each individual qsosed as input to qsosed is given by
model. Although a true ‘Apples versus Apples’ comparison is only
really possible with full knowledge of the intrinsic SED, in com- 𝐿/𝐿 Edd = 𝑚¤ × ( 𝑓bol /5.15) × ( 𝑓bol /5.15) −1/2 . (C2)

MNRAS 000, 1–20 (2023)


C iv and He ii emission in quasars 19
1044

1047 1047

1045
[erg s−1 ]

[erg s−1 ]
1046 1046
3000Å

3000Å
νLν |

νLν |
1045 1045

LHeII λ1640 [erg s−1 ]

νLν |2 keV [erg s−1 ]


1043
100 100
L/LEdd

L/LEdd
10−1 10−1
1044

10−2 10−2
1042
108 109 1010 108 109 1010
MBH [M ] MBH [M ]

Figure B2. He ii line luminosity (left) and 2 keV X-ray continuum luminosity (right) as a function of SMBH mass, luminosity and Eddington ratio. Contours of
constant 𝐿2 keV are largely aligned with lines of constant 𝑀BH , consistent with the assumption in qsosed that the hard X-ray emission is equal to 2 per cent of
the Eddington limit.

100 100 100


0.1
/LEdd ) × 5.15
ṁ ≡ Ṁ /ṀEdd

L/LEdd

L228 /L1640
10−1 10−1 10−1 0.06
3000Å,q
(ν Lν |

0.04

10−2 10−2 10−2


0.03
108 109 1010 108 109 1010 108 109 1010
MBH,q [M ] MBH,q [M ] MBH ≡ MBH,q × (fbol /5.15)−1/2 [M ]

Figure C1. An illustration of how the qsosed predictions change with differing treatments of the bolometric correction, focusing on the bottom-right panel of
Fig. 4. Left panel: the model outputs plotted as a function of the intrinsic, input values of 𝑚¤ and 𝑀BH, q . Centre panel: as in the left panel, but with the 𝑦-axis
is replaced with 𝐿bol /𝐿Edd where 𝐿bol is calculated from 𝐿3000 using a constant bolometric correction of 5.15. Right panel: as in the centre panel, but with the
𝑥-axis scaled by ( 𝑓bol /5.15) −1/2 and the 𝑦-axis scaled by ( 𝑓bol /5.15) 1/2 to capture the impact of the bolometric correction on SMBH mass estimates in the
observed data. See the main Appendix text for details.

The effect of introducing these scaling factors as transformations APPENDIX D: BLACK HOLE SPIN
from the initial qsosed grid is shown in Fig. C1, to show how the
right-hand panels of Figs. 4 and 5 would change if we had made a In Fig. D1 we show how the predictions of 𝛼ox and our He ii 𝜆1640
different presentation choice. The scale factors twist and distort the EW proxy (𝐿 228 /𝐿 1640 ) from qsosed change if we instead con-
simulation grid slightly from the original uniform parameter space, sider a maximally spinning SMBH. While the qualitative trends
but, overall, the effects are fairly modest because only square-root in the 𝐿 228 /𝐿 1640 are broadly in line with the low spin case, the
terms distinguish the rightmost panel from the original input grid. 𝑎 ∗ = 1 models fail to reproduce the observed low values of 𝛼ox at
high Eddington fractions and SMBH masses (see discussion in Sec-

MNRAS 000, 1–20 (2023)


20 M. J. Temple et al.
1047 100
−1.2

MBH = 109 M 1047 1047

1046

[erg s−1 ]

[erg s−1 ]
−1.3
νLν (erg s−1 )

1046 1046


1045

3000Å

3000Å
−1.4
10−1

ν Lν |

ν Lν |
1044 1045 1045

−1.5

L228 /L1640
αox
1047 1010
100 100
ṁ = 0.15 −1.6

1046
νLν (erg s−1 )

MBH [M ]

L/LEdd

L/LEdd
−1.7
0.1
1045 109 10−1 10−1

−1.8
1044
He II

0.06
αox
10−2 10−2
108 108 109 1010 −1.9 108 109 1010
1014 1015 1016 1017 1018 1019
ν (Hz) MBH [M ] MBH [M ]

Figure D1. Left panel: As Fig. 2, but for the maximally spinning case, 𝑎∗ = 0.998. The grey shaded area shows the range of the SEDs shown in Fig. 2 for the
non-spinning case. Centre and right panels: qsosed predictions for 𝛼ox and 𝐿228 /𝐿1640 for a maximally spinning SMBH with 𝑎∗ = 0.998 (cf. Figs. 4 and 5).

tion 5.2.3). The reason for this can be understood from the left-hand
panel of Fig. D1, where we show the qsosed broadband spectrum
(the analogue to Fig. 2) for the maximally spinning case. Inspection
of the high 𝑚¤ models in the top-panel reveals that the 𝛼ox behaviour
is driven by a combination of stronger X-rays and the movement of
the peak of the thermal component. At high spin, the thermal peak
moves blueward to higher energies, such that the lower frequency
pivot point falls further from the peak and has lower flux compared
to the low spin model. In qsosed, this behaviour comes about in
a slightly convoluted way, but is driven by the decrease of the ra-
dius 𝑅warm (and corresponding temperature increase). This decrease
happens because 𝑅warm = 2𝑅hot , and 𝑅hot must move inwards as
spin increases, because 𝑅ISCO moves closer to the SMBH so 𝑅hot
must also decrease from eq. 2 of Kubota & Done (2018) to maintain
the model assumption that the dissipated power is 2 per cent of the
Eddington luminosity. One could clearly construct other models in
which the critical radii change in different ways when the spin is
changed, which is partly why we caution against over-interpreting
the fact that maximally spins appear difficult to reconcile with the
data.

This paper has been typeset from a TEX/LATEX file prepared by the author.

MNRAS 000, 1–20 (2023)

You might also like