Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~pgreen/Papers/aox_composites98.ps Äàòà èçìåíåíèÿ: Mon Sep 19 19:23:12 2005 Äàòà èíäåêñèðîâàíèÿ: Tue Oct 2 03:24:50 2012 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: dust disk

THE ASTROPHYSICAL JOURNAL, 498 : 170õ180, 1998 May 1
1998. The American Astronomical Society. All rights reserved. Printed in U.S.A.
(
DIFFERENCES BETWEEN THE OPTICAL/ULTRAVIOLET SPECTRA OF X­RAY BRIGHT
AND X­RAY FAINT QSOs
PAUL J. GREEN
Harvard­Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138 ; pgreen=cfa.harvard.edu
Received 1997 July 29 ; accepted 1997 December 4
ABSTRACT
We contrast measurements of composite optical and ultraviolet (UV) spectra constructed from samples
of QSOs deïned by their soft X­ray brightness. X­ray bright (XB) composites show stronger emission
lines in general, but particularly from the narrow­line region. The di+erence in the [O III]/Hb ratio is
particularly striking, and is even more so when blended Fe II emission is properly subtracted. The corre­
lation of this ratio with X­ray brightness was a principal component of QSO spectral diversity found by
Boroson & Green. We ïnd here that other, much weaker narrow optical forbidden lines ([O II] and
[Ne V]) are enhanced by factors of 2 to 3 in our XB composites, and that narrow line emission is also
strongly enhanced in the XB UV composite. Broad permitted­line ÿuxes are slightly larger for all XB
spectra, but the narrow­/broad­line ratio stays similar or increases strongly with X­ray brightness for all
strong permitted lines except Hb.
Spectral di+erences between samples divided by their relative X­ray brightness (as measured by a ox )
exceed those seen between complementary samples divided by luminosity or radio loudness. We propose
that the Baldwin e+ect may be a secondary correlation to the primary relationship between and emis­
a ox
sion line equivalent width. We conclude that either (1) depends primarily on the shape of the ionizing
W j
continuum, as crudely characterized here by or (2) both and are related to some third param­
a ox , W j a ox
eter characterizing the QSO physics. One such possibility is intrinsic warm absorption ; a soft X­ray
absorber situated between the broad and narrow line regions can successfully account for many of the
properties observed.
Subject headings : galaxies : active õ quasars : emission lines õ quasars : general õ
ultraviolet : galaxies õ X­rays : galaxies
1. INTRODUCTION
Most QSOs have been discovered by virtue of their
strong optical/UV emission lines or nonstellar colors in this
bandpass. Our understanding to date of the violent inner
regions of active galactic nuclei (AGNe) also derive in large
part from their optical/ultraviolet (OUV) spectra. The pro­
duction of QSO emission lines is widely attributed to
photoionization and heating of the emitting gas by the UV
to X­ray continuum (see, e.g., & Shields
Ferland 1985 ;
& Kallman Studies investigating the relation­
Krolik 1988).
ship of emission lines to continuum radiation have a long
history in the ïeld, but several strong observational
relationships remain unexplained.
If the proportionality between line and continuum
strength were linear, then diagnostics such as line ratios and
would be independent of continuum luminosity.
W j ïrst noticed that, in high­redshift quasars,
Baldwin (1977)
the of the C IV 1550 emission line in quasars decreases
W j ñ
with increasing UV (1450 luminosity. The Baldwin e+ect
ñ)
(BE+, hereafter) was also found to be strong for ions such as
O VI, N V, He II, C III], Mg II, and Lya (see, e.g., &
Tytler
Fan et al. The initial excitement
1992 ; Zamorani 1992).
about the potential for the BE+ as a standard candle and
cosmological probe has faded ; the dispersion in the
relationship is too large. Neither the source of that disper­
sion nor the cause of the BE+ itself have been deïnitively
identiïed. However, some possible explanations for the
BE+ have been o+ered, one being a dependence of the con­
tinuum spectral energy distribution (SED) on luminosity
& Malkan
(Zheng 1993 ; Green 1996).
Many important lines respond primarily to the extreme
ultraviolet (EUV) or soft X­ray continuum. Unfortunately,
the EUV band is severely obscured by Galactic absorption.
However, constraints on the ionizing continuum are avail­
able through analysis of the adjacent UV and soft X­ray
windows. In a small, uniform sample of optically selected
QSOs et al. the strongest correlation found
(Laor 1997),
between X­ray continuum and optical emission line param­
eters was of the soft X­ray spectral slope (where
a X f l P
with the FWHM of the Hb emission line. Strong
l~aX)
correlations between Fe II/Hb, and the
a X , L *O III+ ,
[O III]/Hb ratio were seen both there and in previous QSO
studies (e.g., & Green The latter authors
Boroson 1992).
found that most of the variation in the observed properties
of low­redshift QSOs can be represented in a principal com­
ponent analysis by eigenvectors linking Fe II, [O III], Hb,
and He II emission line properties and continuum proper­
ties, such as radio loudness, and the relative strength of
X­ray emission, as characterized by (deïned below). A
a ox
recent, and possibly related result is that Seyfert 1 galaxies
with broad Hb emission lines tend to have hard (ÿat) X­ray
spectral slopes (see, e.g., Mathur, & Elvis In a
Brandt, 1997).
hard X­ray selected sample of (mostly Seyfert) AGNe,
narrow [O III] ÿux correlates well with X­ray ÿux, while
broad Balmer lines do not The physical
(Grossan 1992).
origin of these diverse and interrelated correlations has yet
to be determined.
We are launching a large­scale e+ort to probe these
e+ects in large samples, using both data and analysis as
homogeneous as possible. Many physically informative
trends intrinsic to QSOs may be masked by dispersion in
the data caused by either low signal­to­noise ratios or varia­
bility. An important tool for studying global properties of
QSOs is the coaddition of data for samples of QSOs. In this
170

SPECTRA OF X­RAY BRIGHT AND X­RAY FAINT QSOs 171
paper, we concentrate on an analysis of composite optical/
UV spectra of subsamples of QSOs grouped by the relative
strength of their soft X­ray emission.
2. ANALYSIS
2.1. Constructing Comparison Samples
Although the signal­to­noise ratio (S/N) for the individ­
ual spectra in both samples we study here tends to be about
10 or less per resolution element, the coaddition (averaging)
of spectra with similar continuum properties allow us to
increase the S/N and constrain the properties of the average
QSO. Through averaging, a greater number of emission
lines, with a wider range of ionization energies, and ïner
details in emission line proïles become measurable. This
technique has been applied in several recent studies (see,
e.g., & Vio et al.
Cristiani 1990 ; Francis 1991 ; Osmer,
Porter, & Green et al. What is lost is a
1994 ; Zheng 1996).
reliable measure of the intrinsic dispersion in the observed
correlations. However, it is important to ïrst discover the
correlations intrinsic to the average QSO. The sources of
dispersion in the relationship can later be studied if data of
adequate S/N are available for a large enough sample.
summarizes mean continuum properties for the
Table 1
X­ray bright and X­ray faint subsamples we culled from the
Large Bright Quasar Survey and IUE QSO samples
described below. Optical and X­ray luminosities are taken
from et al. and and assume
Green (1995) Green (1996)
km s~1 Mpc~1 and The slope of a hypo­
H 0 \ 50 q 0 \ 0.5.
thetical power law connecting 2500 and 2 keV is deïned
ñ
as log so that is larger for objects
a ox \ 0.384 (l opt /l X ), a ox
with stronger optical relative to X­ray emission.
2.1.1. L BQS Sample
The Large Bright Quasar Survey (LBQS ; Foltz,
Hewett,
& Cha+ee is a sample of more than 1000 QSOs, uni­
1995)
formly selected over a wide range of redshifts. LBQS QSO
candidates were selected using the Automatic Plate Mea­
suring Machine & Trimble to scan UK
(Irwin 1984)
Schmidt direct photographic and objective prism plates. A
combination of quantiïable selection techniques were used,
including color selection, selection of objects with strong
emission lines, and selection of objects having redshifted
absorption features or continuum breaks. The technique
appears to be highly efficient at ïnding QSOs with
0.2 \ z \ 3.3, a signiïcantly broader range than in past
work. Follow­up (6õ10 resolution) optical spectra with
ñ
S/NB 10 (in the continuum at D4500 were obtained at
ñ)
the Multiple Mirror Telescope (MMT) and the 2.5 m
duPont telescope. The digital spectra used here were gra­
ciously provided to the author by Craig Foltz.
Soft X­ray data for LBQS QSOs was selected from the
ROSAT All­Sky Survey (RASS) as detailed in et al.
Green
Of the 908 QSOs in the LBQS/RASS sample, 92 are
(1995).
detected in X­rays. For the nondetections, we assign an
upper limit of 4 p to the raw counts.
Before generating composite spectra, we remove QSOs
with known broad absorption lines (BAL QSOs) from the
LBQS/RASS sample. Above z \ 1, an insufficient fraction
of QSOs are detected to construct X­ray bright and faint
subsamples of comparable size. We therefore exclude QSOs
with redshifts [1, for which the RASS data are insuffi­
ciently sensitive to be of use. To create composite spectra of
similar S/N, we choose a dividing point of that results in
a ox
similarly sized subsamples. We construct the X­ray bright
sample (XB ; 60 QSOs) of only detected objects, with a ox \
1.475. The X­ray faint sample (XF ; 54 QSOs) includes both
detections and lower limits with QSOs with
a ox º 1.475.
lower limits below that value could rightly belong either to
the XB or XF sample and so are excluded from consider­
ation. Note also that the value of 1.475 does not imply that
is measured to such accuracy. Rather, it provides a con­
a ox
venient dividing line near the median of the small range of
values (D1.2õ1.6) typically measured in QSOs.
a ox Continuum properties of the ïnal XF and XB (a ox ­selec­
ted) samples are listed in In survival analysis, if the
Table 1.
TABLE 1
CONTINUUM PARAMETERS FOR QSO SUBSAMPLES
SAMPLE SIZE
PROBABILITYc
PARAMETER SUBSAMPLE N total N limits a MEDIAN MEAN RMS TRUNCATEDb (%)
Redshift . . . . . . LBQS/XF 54 0 0.46 0.49 0.03 No 43
LBQS/XB 60 0 0.45 0.51 0.03 No
IUE/XF 23 0 0.35 0.57 0.09 No 2.5
IUE/XB 25 0 0.21 0.34 0.03 No
log l opt . . . . . . . LBQS/XF 54 0 30.41 30.44 0.06 No 17
LBQS/XB 60 0 30.30 30.30 0.06 No
IUE/XF 23 0 31.03 31.05 0.15 No 25
IUE/XB 25 0 30.50 30.57 0.08 No
a ox . . . . . . . . . . . . LBQS/XF 54 42 1.61 1.61 0.01 1.67 \10~4
LBQS/XB 60 0 1.34 1.35 0.01 No
IUE/XF 23 4 1.57 1.60 0.03 No \10~4
IUE/XB 25 0 1.31 1.28 0.02 No
log l X . . . . . . . . . LBQS/XF 54 42 22.55 25.96 0.08 25.64 \10~4
LBQS/XB 60 0 26.78 26.78 0.06 No
IUE/XF 23 0 26.91 26.78 0.20 No 1
IUE/XB 25 0 27.17 27.25 0.09 No
NOTE.õFor LBQS/XF, z\1 and for LBQS/XB, z \ 1 and detections only ; for IUE/XF,
a ox º 1.475 ; a ox \ 1.475,
and for IUE/XB, detections only.
a ox º 1.4 ; a ox \ 1.4,
a All limits are upper limits except for a ox .
b This column lists the value of an outlying nondetection, if it was necessarily redeïned, truncating the distribution to
allow normalization. The resulting mean value is biased.
c Probability that the two subsample distributions for this parameter are randomly drawn from the same parent
distribution, from ASURV logrank test.

2000 2200 2400 2600 2800
20
40
60
3200 3400 3600 3800 4000
20
40
60
3800 4000 4200 4400 4600
20
40
60
4200 4400 4600 4800 5000
20
40
60
172 GREEN Vol. 498
TABLE 2
COMPOSITE NORMALIZATION DEFINITIONS
Wavelength Region j n Continuum Band Continuum Band Line Name
LBQS Spectra
4150õ5100 . . . . . . . . . . 4863 4710õ4780 5040õ5090 Hb
3800õ4800 . . . . . . . . . . 4102 4020õ4050 4150õ4270 Hd
3200õ4000 . . . . . . . . . . 3426 3360õ3400 3450õ3500 [Ne V]
2000õ3000 . . . . . . . . . . 2798 2645õ2700 3020õ3100 Mg II
IUE Spectra
1000õ2000 . . . . . . . . . . 1549 1450õ1480 1680õ1700 C IV
NOTE.õEach individual rest­frame QSO spectrum is divided by the ÿux of the ït line at the
normalization wavelength before addition to a composite spectrum. The continuum level
j n
at is found by ïtting a linear continuum between two points, the mean ÿux values in the
j n
two continuum bands shown, at their central wavelength points.
lowest (highest) point in the data set is an upper (lower)
limit, the mean is not well deïned since the distribution is
not normalizable, and so the outlying censored point is
redeïned as a detection. For the XF sample, we list here the
value of that redeïned limit at which the distribution is
truncated. The resulting mean value is biased, so that the
true mean values of and log for the LBQS XF sample
a ox l X
are probably even more X­ray faint than those listed.
We ïnd no signiïcant di+erences in the distributions of
redshift, or between our XB and XF subsamples of
N H Gal, l opt
the LBQS. In any case, the emission line properties of the
LBQS as a whole show no strong dependence on either
luminosity or redshift Hooper, & Impey
(Francis, 1993).
2.1.2. IUE Sample
To explore changes in QSO UV spectra as a function of
we include a previously compiled sample of QSOs
a ox ,
FIG. 1.õHistograms of the number of QSOs whose spectra contribute
to each of eight LBQS composite spectra. Four separate continuum points
were used to generate LBQS composite spectra for XB and XF subsamples
in four separate wavelength regions (see Composites normalized
Table 2).
at the four di+erent continuum points have di+erent histograms, since they
require the rest frame continuum bands to be present in all contributing
observed frame spectra. Di+erent selection criteria for the XB (solid lines)
and the XF subsamples (dotted lines) also yield slightly di+erent histo­
grams.
observed by both the International Ultraviolet Explorer
(IUE) and Einstein. This sample was selected as described in
by requiring that the QSOs in the IUE
Green (1996),
sample have Einstein soft X­ray data available. In et
Wilkes
al. we deïne the IUE/Einstein sample of 49 objects.
(1994),
Again, we remove all known BAL QSOs from the IUE
sample. With this sample, because of selection e+ects, it is
not possible to achieve similar­sized subsamples with
similar redshift distributions by selecting an appropriate a ox
value at which to divide the sample. Similar redshift dis­
tributions are obtained for but then the XB sub­
a ox \ 1.3,
sample contains only about a dozen QSOs, compared to 37
in the XF subsample. Similar subsample sizes are obtained
for for which, however, the requirement of a
a ox \ 1.4,
detection in the XB sample results in a lower mean redshift,
due mostly to extra QSOs between 0.1 \ z \ 0.2. Since
QSO spectral evolution between their mean redshifts (0.34
and 0.57, respectively) is negligible, we emphasize the sub­
sampleîs split at which is in any case closer to the
a ox \ 1.4,
dividing value for the LBQS subsamples. However, we also
check the subsampleîs split at to ensure that evolu­
a ox \ 1.3
FIG. 2.õHistograms of the number of QSOs whose spectra contribute
to the IUE composite spectra for the XB subsample (solid line) and the XF
subsample (dotted line). The IUE composites were all normalized at the
rest wavelength of C IV.

No. 1, 1998 SPECTRA OF X­RAY BRIGHT AND X­RAY FAINT QSOs 173
tion or luminosity e+ects do not signiïcantly bias our
results.
2.2. Constructing Composite Spectra
We ïrst deredshift each individual spectrum by dividing
the linear dispersion coefficients (initial wavelength and
wavelength per pixel) by (1 ] z). An estimate of the rest
frame continuum ÿux at a chosen normalization point is
f c,l
then derived by ïtting a linear continuum between two
continuum bands. The widths and centers of these bands,
typically chosen to straddle an important emission line, are
listed for each wavelength region in We then nor­
Table 2.
malize the entire spectrum via division by f c,l .
Each normalized spectrum is ïrst rebinned to the disper­
sion of the composite spectrum, conserving ÿux via inter­
polation. We choose 2.5 bins, similar to the majority of
ñ
the individual LBQS spectra. The IUE spectra are binned
to 1.18 as in the original short­wavelength prime (SWP)
ñ,
camera spectra. Each normalized, rebinned spectrum is
then stored as a vector in a two­dimensional array. Finally,
for each pixel in the completed array, the number of spectra
N with ÿuxes in that rest frame wavelength bin is tallied. If
Nº 3, the median and the mean of all the ÿux values in the
bin are computed and stored in the ïnal one­dimensional
composite spectra, as is N in the ïnal histogram array. A
similar procedure is performed for each bin, until the
spectra are completed. The strong sky line at j5577, some­
times poorly subtracted, was omitted from the LBQS com­
posite. Geocoronal Lya lines were similarly excluded when
building the IUE composites.
We chose four separate continuum points to generate
LBQS composite spectra in four wavelength regions (Table
Composites normalized at di+erent continuum points
2).
may have di+erent histograms, since they require the rest
frame continuum bands to be present in all contributing
observed frame spectra. Histograms of the number of QSOs
contributing to each LBQS composite are shown in Figure
The IUE composites were all normalized at the wave­
1.
length of C IV, and the histograms for those spectra are
shown in Figure 2.
The median composites appear to be smoother than the
means, since they are less a+ected by spikes and low S/N
features in the individual spectra. As a result, the s2 values
of model ïts are also lower for median composites.
However, since the median omits QSOs with more extreme
spectral properties, we prefer to analyze the average com­
posites. We check the signiïcance of every strong di+erence
between XB and XF average composites by measurement of
the median composites to determine whether outliers domi­
nate the feature in question.
3. EMISSION LINE MEASUREMENTS : LBQS
No analysis of continuum slope is presented for the
LBQS spectra, since these observations are not fully
spectrophotometric. Using the IRAF task SPECFIT (Kriss
we ït simple empirical models to the LBQS compos­
1994),
ite spectra in four wavelength regions, deïned in In
Table 2.
each region, we begin by ïtting a power­law (PL) contin­
uum using only comparatively line­free regions (listed for
each region below). The resulting continuum ït parameters
are slope a and intercept such that ÿux
f 1000 , f j \
Since the LBQS spectra are not spectro­
f 1000 (j/1000)a.
photometric, the continuum ït results should be used only
to derive exact composite equivalent widths from
(normalized) line ÿuxes if desired. The intercept value reÿec­
ts the arbitrary normalization of individual spectra to unity
at the chosen wavelength shown in
j n Table 2.
The emission line components are assumed to be Gauss­
ian and symmetric (skew ïxed at unity), so output from the
ïts includes ÿux, centroid, and FWHM for each line.
Results from these ïts are shown in and
Table 3 Figure 3.
The errors listed in both Tables and are directly from the
3 4
SPECFIT task, which assumes that the errors on the input
spectrum follow a Gaussian distribution. The tabulated
errors represent 1 p for a single interesting parameter.
3.1. T he Hb and [O III] Region
We ït a PL continuum using these relatively line­free
regions : jj4020õ4050, 4150õ4270, 4420õ4450, 4710õ4780,
and 5070õ5130. The resulting ït parameters are ïxed, while
the following single Gaussian components are ït to the
strong emission lines between 4150 to 5100 (1) broad
ñ:
Hb ; (2) narrow Hb, with FWHM ïxed to that of [O III]
j5007 ; (3) [O III] j5007 ; and (4) [O III] j4959, ïxed relative
to the central wavelength and (1/3) ÿux of component (3).
Finally, the blended iron emission above the PL continuum
is summed over the wavelength range jj4434õ4684.
As can be seen from and by far the
Figure 3 Table 3,
strongest e+ect is the great relative strength of [O III] emis­
sion in the XB composite ; [O III] emission is D2.5 times
stronger in X­ray bright sample. Broad Hb emission is
about 40% stronger in the XB composite, although the
FWHM are similar at D5700 km s~1. The strengths of the
narrow components of Hb are comparable. In both XB and
XF composites, broad Hb is redshifted relative to rest (i.e.,
relative to [O III] j5007 or narrow Hb) by about 300 km
s~1.
& Green hereafter in a principal
Boroson (1992 ; BG92),
component analysis of low­redshift QSO spectra, found
that the principal eigenvector of their sample (the linear
combination of parameters that represents the largest
variance in the sample) is primarily an anticorrelation
between the strength of [O III] emission and optical Fe II
strength. This trend is in the same direction as we ïnd here,
with the large ([O III]) XB sample having the lower Fe II
W j
equivalent width, as judged by the iron emission near j4600
(Table 3).
We can make more precise measurements in this region if
we subtract Fe II emission. Marianne Vestergaard has
accomplished this task, using a technique similar to that of
A narrow optical Fe II template spectrum (of IZw1 ;
BG92.
is convolved with a Gaussian function and sub­
BG92)
tracted from the LBQS composites, ïtting iteratively by eye.
The ÿux in the iron template between j4434 and j4684 is
within D10% of that found via our simple summation tech­
nique. Because of a strong iron multiplet with lines at
jj4924, 5015, and 5169, however, subtraction of Fe II
decreases the apparent ([O III]) in both the XF and XB
W j
composites. In the XF composite, where iron emission is
strong, the measured ([O III]) at j5007 decreases from
W j
13.5 to 7.5, and its measured FWHM decreases from 1330
to 650 km s~1. The XF Hb line changes by comparison only
slightly after iron subtraction (from a total equivalent width
of 53 to 47). In the XB composite, where iron emission is
weak, the measured change in equivalent width of both
[O III] and Hb after Fe II subtraction are within the errors
as are the changes in FWHM. Since the
(*W j \ 3 ñ),
strength of [O III] in the XF composite is initially low, the

174 GREEN Vol. 498
FIG. 3.õFinal normalized composite spectra of the LBQS sample. The left­hand column contains plots of the XB composites, while the right­hand
column shows XF composites. Each row shows a portion of a composite constructed from spectra normalized at a di+erent rest wavelength continuum point
(see j2798 (top row), j3426 (second row), j4102 (third row), and j4863 (bottom row). Individual lines and their parameters measured using SPECFIT
Table 2) :
are listed in Fitted components are shown in each plot. These include the best power­law continuum ït to relatively line­free continuum
(Kriss 1994) Table 3.
regions (dashed lines) and Gaussian emission lines (dotted lines). Lines are identiïed by name in the X­ray faint plots.
di+erence in the measured [O III]/Hb ratio between XF and
XB composites is substantially increased by Fe II subtrac­
tion. Much smaller e+ects are expected on other emission
lines treated here. An important extension of this technique
to the UV will be described in an upcoming paper
& Green
(Vestergaard 1998).
3.2. T he Hd Region
First, we ït a PL continuum through the windows :
jj3790õ3845, 4020õ4050, 4150õ4270, 4420õ4450, and
4710õ4780. Gaussian components are ït between 3800 and
4800 to the following lines : [Ne III] j3869, He I j3889,
ñ
[S II]/Hd j4068/jj(4076 ] 4102), and Hc/[O III]
jj(4340 ] 4363). The only signiïcant di+erence seen
between emission lines in this region is that the Hc/[O III]
blend is stronger (at the 2.7 p level) in the XF composite.
3.3. T he [Ne V] and [O II] Region
Line­free regions at jj3200õ3400, 3500õ3700, and 3800õ
4000 are used for the PL ït. Then single Gaussian com­
ponents are ït for [Ne V] j3426 and [O II] j3727. As can be
seen from and both the [Ne V] and [O II]
Figure 3 Table 3,
line ÿuxes are much stronger in the XB composite, consis­
tent with the relative strengths of other unblended narrow
lines measured. For the XF sample, the very weak [O II]
line could not be successfully ït when leaving both and
j c
FWHM free to vary, so we ïxed the central wavelength at
3727 Similarly, the weakness of [Ne V] demanded that its
ñ.

No. 1, 1998 SPECTRA OF X­RAY BRIGHT AND X­RAY FAINT QSOs 175
TABLE 3
SPECFIT RESULTS FOR LBQS COMPOSITES
Component Parameter XB Composite XF Composite
2000õ3000 ñ continuum . . . . . . f 1000 2.44 ^ 0.07 1.79 ^ 0.04
a 0.89 ^ 0.03 0.59 ^ 0.02
Fe II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . f a 41.1 ^ 0.2 36.8 ^ 0.1
Mg II broadb . . . . . . . . . . . . . . . . . . . f 42.0 ^ 2.5 36.4 ^ 2.1
j c 2796.9 ^ 3.5 2801.6c
Mg II narrow . . . . . . . . . . . . . . . . . . f 15.94 ^ 1.38 11.26 ^ 1.11
j c 2798.8 ^ 0.8 2801.6 ^ 0.8
FWHM 2650 ^ 230 2530 ^ 240
3200õ4000 ñ continuum . . . . . . f 1000 31.3 ^ 1.2 29.2 ^ 0.8
a 2.81 ^ 0.03 2.76 ^ 0.02
Ne Vd . . . . . . . . . . . . . . . . . . . . . . . . . . . f 1.89 ^ 0.22 1.06 ^ 0.15
j c 3425.6 ^ 1.2 3425.5 ^ 1.8
[O II]e . . . . . . . . . . . . . . . . . . . . . . . . . . f 2.10 ^ 0.24 0.65 ^ 0.17
FWHM 900 ^ 120 1090 ^ 290
3800õ4800 ñ continuum . . . . . . f 1000 13.5 ^ 1.3 35.6 ^ 3.1
a 1.82 ^ 0.07 2.50 ^ 0.06
[Ne III] . . . . . . . . . . . . . . . . . . . . . . . . . f 2.9 ^ 1.2 2.3 ^ 0.6
j c 3870.8 ^ 3.6 3867.3 ^ 4.1
FWHM 1330 ^ 240 1200f
He I . . . . . . . . . . . . . . . . . . . . . . . . f 1.45 ^ 1.3 1.43 ^ 0.6
j c 3894 ^ 15 3887 ^ 6
FWHM 2100 ^ 1500 1200f
[S II] ]Hd . . . . . . . . . . . . . . . . . . . . . f 4.06 ^ 0.6 4.10 ^ 1.0
j c 4103.0 ^ 1.5 4106.7 ^ 6.1
FWHM 1330 ^ 240 3440 ^ 780
Hc][O III] . . . . . . . . . . . . . . . . . . . . f 15.3 ^ 1.0 20.7 ^ 1.7
j c 4347.5 ^ 1.5 4348.4 ^ 2.8
FWHM 3240 ^ 250 5200 ^ 620
4150õ5100 ñ continuum . f 1000 30.0 ^ 3.3 47.0 ^ 3.4
a 2.17 ^ 0.07 2.43 ^ 0.05
Fe II . . . . . . . . . . . . . f g 26.5 ^ 0.2 31.1 ^ 0.1
Hb broad . . . . . . . . . . . . . . . . . . . . . . . f 64.0 ^ 2.6 42.8 ^ 2.5
j c 4869.3 ^ 1.7 4872.5 ^ 2.5
FWHM 5660 ^ 290 5730 ^ 360
Hb narrow . . . . . . . . . . . . . . . . . . . . . f 9.1 ^ 1.4 10.7 ^ 1.7
j c 4865.6 ^ 1.1 4864.2 ^ 1.3
FWHMh 976 1330
[O III] j5007 . . . . . . . . . . . . . . . . . . . f 32.32 ^ 1.11 13.46 ^ 0.95i
j c 5010.2 ^ 0.3 5009.9 ^ 0.8
FWHM 976 ^ 43 1330 ^ 108
NOTE.õAll ÿuxes are summed over a locally normalized continuum (see text) and so
are approximately equivalent widths. The value of the (normalized) continuum ÿux at
any wavelength j can be obtained via (j/1000)~a. Emission line centroids
f c \ f 1000 j c
are in while FWHM are shown in km s~1.
ñ,
a Fe II summed over PL ït from jj2210õ2730.
b FWHM ïxed at 15,000 km s~1.
ïxed at narrow Mg II values.
c j c
d FWHM ïxed at 1000 km s~1.
ïxed at 3727
e j c ñ.
f FWHM ïxed at 1200 km s~1.
g Fe II summed over PL ït from jj4434õ4684.
h FWHM ïxed at [O III] j5007 FWHM.
([O III]) of XF composite decreases to 7.5 ^ 1 after Fe II subtraction.
W j
FWHM be ïxed because otherwise SPECFIT seeks to ït
neighboring features with an unrealistically broad line. The
use of composite spectra enables us here for the ïrst time to
show that these narrow lines are about 3 times as strong in
the X­ray bright QSOs as in their X­ray faint analogs.
3.4. T he Mg II Region
Here a PL continuum is ït to jj2195õ2275, 2645õ2700,
and 3020õ3100 windows. The spectrum from 2000 to 3200
is then ït with the ïxed PL plus these Gaussian com­
ñ
ponents : narrow Mg II and broad Mg II, ïxed in wavelength
to the narrow component. Without ïxing the broad line
width, the ït would wander o+ trying to include nearby
blended iron lines. Finally, the blended iron emission above
the PL continuum is summed over the wavelength range
2210õ2730 ñ.
We ïnd here that the XB sample has stronger UV Fe II
emission opposite to the trend for optical Fe II.
(Table 4),
et al. also found that QSOs in the LBQS with
Green (1995)
strong UV Fe II emission are particularly bright in the soft
X­ray bandpass. UV Fe II lines have their principal ionizing
and heating continuum above 500 eV ; Fe II lines in the
optical are principally caused by continuum above 800 eV.
4. EMISSION LINE MEASUREMENTS : IUE
For the IUE spectra, convincing model ïts to the com­
posite spectra could be obtained when ïtting a simple PL
continuum simultaneously with lines. The following single

176 GREEN Vol. 498
TABLE 4
SPECFIT RESULTS FOR IUE COMPOSITES
Component Parameter XB Composite XF Composite
1000õ2000 ñ continuum a 1.51 ^ 0.02 1.57 ^ 0.02
f 1000 0.75 ^ 0.02 0.75 ^ 0.03
Lyb/O VI . . . . . . . . . . . . . . . . . . . . . . . f 19.6 ^ 1.1 10.1 ^ 1.0
j c 1033.5 ^ 0.4 1032.5 ^ 0.6
FWHM 5230 ^ 400 4560 ^ 450
Lya broad . . . . . . . . . . . . . . . . . . . . . . f 87.3 ^ 3.9 78.7 ^ 2.9
j c 1218.8 ^ 0.8 1218.7 ^ 0.5
FWHM 12840 ^ 410 12700 ^ 400
Lya narrow . . . . . . . . . . . . . . . . . . . . f 47.3 ^ 1.7 36.8 ^ 2.2
j c 1214.6 ^ 0.1 1214.9 ^ 0.2
FWHM 3030 ^ 80 3470 ^ 130
N V . . . . . . . . . . f 3.9 ^ 2.5 1.2 ^ 0.3
j c 1240 ^ 3 1244 ^ 1
FWHM 6000 ^ 1200 6700 ^ 3200
O I . . . . . . . . . . . . f 2.6 ^ 0.9 6.8 ^ 1.9
j c 1310 ^ 4 1312 ^ 4
FWHM 6700 ^ 3400 6400 ^ 2100
S IV . . . . . . . . . f 8.4 ^ 1.0 11.2 ^ 1.6
j c 1398 ^ 1.7 1395 ^ 2.3
FWHM 6250 ^ 870 6270 ^ 950
C IV broad . . . . . . . . . . . . . . . . . . . . . f 52.9 ^ 1.3 39.2 ^ 9.0
j c 1546.2 ^ 1.6 1548.3 ^ 1.6
FWHM 12500 ^ 2800 12000 ^ 2500
C IV narrow . . . . . . . . . . . . . . . . . . . . f 30.6 ^ 6.6 14.8 ^ 7.9
j c 1549.09 ^ 0.2 1548.34 ^ 0.6
FWHM 3200 ^ 350 3400 ^ 940
Absorption . . . . . . . . . . . . . . . . . . . . . f 5.8 ^ 3.9 8.8 ^ 1.4
j c 1487 ^ 19 1480 ^ 12
FWHM 12900 ^ 1800 12500 ^ 2800
NOTE.õParameters have the same meaning as in Table 3.
Gaussian components are ït to the spectrum between 1000
and 2000 (1) Lyb/O VI, (2) broad Lya, (3) narrow Lya, (4)
ñ:
N V, (5) O I, (6) S IV, (7) broad C IV, (8) narrow C IV, and
(9) C IV absorption. Again, Gaussian components are
assumed to be symmetric. The data do not warrant multiple
Gaussian ïts to lines as weak or weaker than N V. However,
some such lines are included in the ït partly to insure that a
reasonable continuum estimate is derived.
Results from these ïts, shown in and are
Table 4 Figure 4,
described below.
1. The PL continuum ïts are similar in slope. At the 2 p
level, the XF sample may be marginally steeper. If veriïed
in other samples, this could be caused by reddening or
absorption by dust. However, the similarity of the slopes
suggests that the relative attenuation between the XB and
XF spectra is gray.
2. Just as for the LBQS composites, we ïnd for the IUE
sample that line emission is signiïcantly stronger in the XB
composite. Lyb/O VI, Lya, and C IV, when ït with a single
Gaussian component, are each signiïcantly ([3 p) stronger
in the XB composite. For Lya, the narrow/broad line ÿux
ratios are higher in the XB composite as well. Some margin­
al evidence (D1.5 p) for the same e+ect is seen for C IV.
From it is clear that the Lyb/O VI emission line
Table 4,
shows a large contrast. The inverse correlation between
Lyb/O VI and was also remarked recently in a sample of
a ox
individual Hubble Space Telescope (HST ) Faint Object
Spectrograph (FOS) spectra Kriss, & Davidsen
(Zheng,
Among weaker emission lines, O I j1302 and Si IV/
1995).
O IV] show a marginal (\2 p) trend in the opposite direc­
tion, with smaller in the XB composite. There is also a
W j
FIG. 4.õFinal normalized composite spectra of the IUE/Einstein
sample. The top plot is the XB composite, and the bottom shows the XF
composite. Both these IUE composites were constructed from spectra nor­
malized at j1549, the rest wavelength of C IV. Individual lines and their
measured parameters using SPECFIT are listed in
(Kriss 1994) Table 4.
The best­ït components are shown in each plot. These include a power­law
continuum (dashed lines) ït simultaneously with Gaussian emission lines
(dotted lines). Lines are identiïed by name in the X­ray faint plot. Note the
apparent continuum deïcit centered near j1480 in the XF composite.

No. 1, 1998 SPECTRA OF X­RAY BRIGHT AND X­RAY FAINT QSOs 177
suggestion that in the XF composite, N V is weak or non­
existent. These should be investigated at higher S/N.
3. There is marginal evidence for a continuum ÿux deïcit
blueward of C IV of the XF continuum.
4.1. Evidence for Absorption
It proved difficult to ït the IUE XF spectrum with a
convincing PL continuum because of a dip blueward of the
C IV emission line, which we tentatively interpret as absorp­
tion. We ït a simple Gaussian component to the absorp­
tion, which is sufficient to characterize the absorbed ÿux
and velocity width. The result is that the C IV emission line
ïts remain unchanged while the overall ït improves, but
only for the XF spectrum. (The formal improvement in s2
would appear marginal because the absorbed region covers
only a few percent of the spectrum.) The equivalent width
of the absorption is 8.8 ^ 1.4 centered at 1480 ^ 10
(W j ) ñ,
with an FWHM of 12,500 ^ 2800 km~1 (see
ñ, Table 4).
Some evidence for a similar dip can be seen even in the XB
spectrum, but it is smaller, and its is not signiïcant in the
W j
ït in comparison to the formal errors (\1.5 p).
If individual absorbers contribute to the dip in the IUE
XF composite, they are likely to be narrow, weak, and
undetected in the noise of the individual spectra. When the
spectra of XF QSOs are combined, however, the enlarged
width and higher S/N of the composite absorber could
enable detection of a broad feature. High­velocity intrinsic
narrow­line absorption in QSOs is just now being recog­
nized in individual QSOs et al. the chal­
(Hamann 1997) ;
lenge is to properly distinguish it from intervening
absorption.
In summary, there is evidence for a dip blueward of the
C IV emission line in the XF composite, independent of any
reasonable continuum choice. If the dip in the XF compos­
ite is caused by C IV absorption, the absorber is highly
ionized and is probably situated near the ionizing source.
The large velocity width also suggests proximity to the
broad­line region (BLR). The weak X­ray emission com­
bined with evidence for high­velocity ionized absorbers is
reminiscent of recent results associating soft X­ray and UV
absorption Elvis, & Wilkes
(Mathur 1994 ; Mathur, 1995 ;
& Mathur et al.
Green 1996 ; Green 1997).
5. DISCUSSION
We now consider a variety of possibilities to account for
the optical and UV spectral di+erence between X­ray bright
and X­ray faint QSOs : (1) luminosity e+ects, (2) radio loud­
ness, (3) absorption, and (4) changes in the intrinsic spectral
energy distribution (SED). The strength of some of these
e+ects is directly testable using the samples at hand.
5.1. L uminosity Subsamples
Both selection e+ects and secondary correlations must be
considered when evaluating the signiïcance of observed
correlations such as these. Two well­known e+ects could
conspire to produce an overall weakening of emission lines
with increasing First, is known to increase with
a ox . a ox
optical luminosity et al. et al. at
(Wilkes 1994 ; Green 1995),
least for optically selected samples et al.
(LaFranca 1995).
Secondly, as luminosity increases, line equivalent width
decreases (i.e., the Baldwin e+ect ; Could
Baldwin 1977).
these e+ects combine to produce the anticorrelation of a ox
and line strength observed here?
As can be seen from the LBQS subsamples are
Table 1,
well matched in optical luminosity, so that no Baldwin
e+ect is expected. Furthermore, the strength of the Baldwin
e+ect in the optical is known to be weak.
Our results from the IUE subsamples are less immune to
a Baldwin e+ect/SED conspiracy because our IUE sub­
samples are not as well matched in luminosity and because
the Baldwin e+ect is much stronger in the UV. We therefore
perform a stringent check, applying the same spectral
averaging techniques now to new subsamples deïned by
UV luminosity.
We divided the IUE sample at the mean UV luminosity1
value of 30.6. The resulting low UV luminosity (UVLO ; 27
QSOs) and high UV luminosity (UVHI ; 22 QSOs) sub­
samples both had mean SPECFIT pro­
a ox \ 1.4 ^ 0.04.
cedures identical to those of the XB and XF samples were
applied. Virtually all spectral di+erences were less signiï­
cant in the UV luminosity subsamples than in the sub­
a ox
samples. Only narrow Lya emission changes more strongly
between UV subsamples than between the subsamples.
a ox
Indeed, the Baldwin e+ect appears to be strongest in the
narrow­line components of both Lya and C IV. The bulk of
the e+ect could be caused by di+erences in narrow­line
region (NLR) emission, as was also suggested by et
Osmer
al. (1994).
Since emission line correlations are stronger with than
a ox
with luminosity, we conclude that either (1) depends
W j
primarily on the shape of the ionizing continuum, crudely
characterized here by or (2) both and are related
a ox , W j a ox
to some third parameter characterizing the QSO physics.
One such possibility is absorption.
Could the correlation of to luminosity cause the
a ox
Baldwin e+ect? Although by design we have selected sub­
samples of similar luminosity for our XF and XB compos­
ites, we may suppose that the primary relationship between
and is propagated into the relationship between
W j a ox a ox
and luminosity, and we test the strength of the predicted
secondary relationship between and luminosity that
W j
results, i.e., the Baldwin e+ect.
We begin by simply contrasting the observed response
here of (C IV) and (Mg II) to with that predicted
W j W j a ox
in the most recent comprehensive study of the Baldwin
e+ect in optically selected QSOs et al. The
(Zamorani 1992).
change between XF and XB subsamples is
*a ox D 0.3
similar in our IUE and LBQS samples. The change in log
(Mg II) predicted by the BE+ in this line is *log
W j W j
(Mg The change we actually measure
II) D[0.68a ox \ 0.18.
between XB and XF composites is (from the values in Table
* log (Mg II) \ 1.0. Similarly, the change predicted by
3) W j
the BE+ in log (C IV) seen by Zamorani et al. is * log
W j W j
(C while we actually measure *log
IV) D[1.16a ox \ 0.30,
(C IV) \ 1.57. Thus, the e+ect of on emission line
W j a ox
strengths is some 5õ6 times larger than that expected if it
were secondary to a luminosity e+ect. Indeed, it seems likely
that the BE+ is secondary to the relationship between and
a ox
line equivalent width.
If the BE+ is caused by a change in concomitant with
a ox
luminosity, then the strongest relationship we see,
W j (a ox )
that of ([O III]), predicts a BE+ in [O III]. This is indeed
W j
seen in the data, where the probability of no corre­
BG92
lation (the null hypothesis) between and ([O III])
M V W j
j5007 is \1%.
1 The log rest frame luminosity, log in ergs cm~2 s~1 Hz~1 at 1450
l UV ,
is deïned in
ñ, Green (1996).

178 GREEN Vol. 498
5.2. Radio Properties
Radio and X­ray loudness are correlated ; et al.
Green
conïrmed that radio loud QSOs (RLQs) are more
(1995)
soft X­ray luminous than radio quiet QSOs (RQQs) in the
LBQS. Unfortunately, the di+erence in radio loudness
between the LBQS XF and XB subsamples cannot be well
characterized : although 4 (of 17 with radio data) are radio
detected in the LBQS XB subsample, only one QSO (of 38)
in the XF subsample is a radio detection.
However, OUV spectral di+erences as a function of radio
loudness have been extensively studied. Di+erences between
emission lines in RLQ and RQQ spectra longward of 1600
are quite subtle, and there is a remarkable similarity in
ñ
Mg II and C III] emission lines between RLQs and RQQs
et al. Distinctions found by include a
(Francis 1993). BG92
redward asymmetry of Hb in RLQs, whereas RQQs show
about equal numbers of red and blue asymmetries (BG92).
Radio loud QSOs tend to have strong [O III] and weak
optical Fe II. In Brinkmann, & Bergeron
Wang, (1996),
neither Hb nor optical Fe II measurements correlate with
radio loudness.
To check the e+ects of radio loudness on UV emission
lines, we created radio loud (RL) and radio quiet (RQ) com­
posite UV spectra from the IUE/Einstein sample. We deter­
mined radio loudness from Sherwood, & Patnaik
Falcke,
for all the UV­excess selected (PG) QSOs, and from
(1996)
& Veron or the NED database for
Veron­Cetty (1993)
others. Two QSOs are omitted because of intermediate
radio loudness, and two more have no published radio data.
Radio subsamples are identical in luminosity and well
matched in redshift if we impose z [ 0.15. This yields
N\ 26 and for the RL subsample, N\ 13 and
z
6 \ 0.48
for RQ QSOs, and log for both. There is
z
6 \ 0.52 l opt \ 30.9
some remaining di+erence between distributions for the
a ox
IUE/Einstein radio subsamples, as expected, such that the
RL sample is somewhat more X­ray bright (a ox \ 1.36
^ 0.04) than the RQ sample (a ox \ 1.47 ^ 0.05).
We ïnd Lya emission to be somewhat stronger in the RQ
composite. C IV emission is of similar strength and FWHM
in both RQ and RL composites but is displaced slightly to
the red in the RL composite. Similar results were found
independently by & Brotherton Thus, overall,
Wills (1995).
the emission line trends due to radio loudness in our sample
are weaker than, and tend to diminish, those seen in The
a ox .
important result is that spectral di+erences between compos­
ites binned by are signiïcantly larger than between com­
a ox
posites binned by radio loudness.
5.3. Absorption and the Relation
a ox (l opt )
Absorption by ionized gas near the nucleus can extin­
guish soft X­ray emission without having signiïcant e+ects
on the observed optical emission. The BAL QSOs represent
an extreme example, wherein observed soft X­ray ÿuxes are
at least an order of magnitude below those of non­BAL
QSOs of similar optical brightness & Mathur
(Green 1996).
Even AGNe with much less spectacular UV absorption
(e.g., narrow absorption lines) show signiïcant soft X­ray
absorption et al. Since most
(Mathur 1994 ; Mathur 1995).
QSOs have either low S/N, low resolution, or no available
UV spectra, many such absorbers await recognition in
current QSO samples.
Is it possible that the correlation is itself caused
a ox (l opt )
by absorption? The hypothesis might be tested if all types of
absorbed QSOs were removed, including QSOs with
known narrow­line intrinsic, damped Lya, and Lya forest
absorbers. A higher S/N sample that has been systemati­
cally searched for absorption in both UV and soft X­ray
spectra is needed. We are building such a sample from the
HST FOS and ROSAT public archives and will study these
issues in an upcoming paper.
We note that, if at least some of the correlation is caused
by increased warm absorption at higher luminosities, then
the correlation would be expected to be further weakened in
soft X­ray selected samples. More common absorption in
the UV bandpass is suspected anecdotally at higher lumi­
nosities but remains to be demonstrated statistically. In the
X­ray bandpass, the same can be said for the popular
assumption that absorption is less common at high lumi­
nosity. This latter notion should be particularly suspect,
since higher energy rest frame X­ray emission is observed in
most high­luminosity objects and requires a higher intrinsic
absorbing column for detection. Furthermore, the most
luminous objects in samples to date have also been radio
loud (Reynolds 1997).
Variability may play a signiïcant role in the a ox (l opt )
correlation. Typically, an optically selected AGN is found
near the survey ÿux limit and therefore preferentially in a
bright phase and then is later followed up in an X­rayõ
pointed observation. Overall, more distant (and thus more
luminous) optically selected QSOs would tend to have
larger measured Even if variability­related biases are
a ox .
not responsible for the observed correlation, such
a ox (l opt )
variability may be expected to introduce scatter into the
true relation. As such, (the power­law spectral index
a ix
between 1 km and 2 keV) might prove a more variability­
resistant measure. Indeed, et al. have found
Lawrence (1997)
that some of the primary correlations discussed here
become more signiïcant when replaces Not only
a ix a ox .
variability but also absorption more strongly a+ects the
optical than the infrared.
5.4. Intrinsic Spectral Energy Distributions
We have o+ered one interpretation of our measurements
that assumes that the intrinsic broadband continuum emis­
sion from the QSO central engine is constant in shape,
independent of luminosity, but that the spectral energy dis­
tribution (SED) seen by the NLR and/or by us may be
strongly a+ected by intervening, possibly ionized absorbing
clouds. However, the change in and the accompanying
a ox ,
changes in emission lines may at least in part be caused by
changes in the intrinsic SED.
Many QSOs have soft keV) X­ray emission that
([1
exceeds the extrapolation from the power­law continuum
observed at higher energies (see, e.g., & Pounds
Turner
et al. This X­ray ```` soft excess îî has
1989 ; Masnou 1992).
often been interpreted as the high­energy continuation of
the UV/EUV/soft X­ray ```` big blue bump îî (BBB), possibly
thermal emission from the surface of an accretion disk (cf.
Several workers (beginning with &
Barvainis 1993). Malkan
Sargent have proposed as an explanation of the
1982)
Baldwin e+ect that as QSO luminosity increases, the BBB
shifts toward lower energies at higher (OUV) luminosities.
As QSOs become more luminous in the optical/UV band,
they thereby undergo a weaker increase in and, therefore,
l X
an increase in The response of line ÿux and depends
a ox . W j
in a fairly complicated manner on the peak energy of the
BBB, on the BBB normalization relative to the power­law
continuum, and on the ionization and heating continuum of

No. 1, 1998 SPECTRA OF X­RAY BRIGHT AND X­RAY FAINT QSOs 179
the line species in question. Detailed photoionization mod­
eling using a variety of input continua, impinging on an
ensemble of clouds from the broad­ to the narrow­line
region, including full self­shielding and optical depth e+ects,
is called for (see, e.g., et al. et al.
Korista 1997a ; Baldwin
Ferland, & Peterson On average, the
1995 ; Shields, 1995).
strongest e+ect may be that higher luminosity QSOs may
undergo spectral evolution such that fewer photons from a
soft X­ray excess/BBB component are available for ioniza­
tion.
A shift of the BBB to lower energies at higher luminosity
implies that the soft X­ray excess should decrease to higher
luminosities. An apparent hardening of the soft X­ray PL
spectral index has been seen in composite ROSAT spectra
of LBQS QSOs toward higher redshifts (see, e.g., et
Schartel
al. that gibes with this picture, but again, higher
1996)
energy rest frame X­ray emission is observed in most high­
luminosity objects, for which a higher intrinsic column is
required before absorption can be detected. In this picture,
a stronger BE+ might be expected for species of higher ion­
ization energy. There is some evidence for such a trend
Fang, & Binette The intrinsic SED model
(Zheng, 1992).
does not predict absorption features but does imply that the
relation should persist even in soft X­ray selected
a ox (l opt )
samples.
6. SUMMARY
By contrasting the composite optical/UV spectra of large
samples of X­ray bright and X­ray faint QSOs, we have
unveiled signiïcant new correlations with X­ray brightness
as characterized by We ïnd that [O III] emission is at
a ox .
least 2.5 times stronger in our XB sample. Proper subtrac­
tion of Fe II suggests a true ratio closer to 5. We ïnd that
other, much weaker narrow optical forbidden lines ([O II]
and [Ne V]) are enhanced by factors of 2 to 3 in our com­
posites. Narrow­line emission is also strongly enhanced in
the XB UV composite. Broad permitted line ÿuxes are
slightly larger for both XB composites, but velocity widths
in the broad emission­line region are not signiïcantly
a+ected. The narrow/broad line ratio stays similar or
increases with X­ray brightness for all strong lines except
Hb.
We also ïnd that UV Fe II and optical Fe II emission are
correlated in opposite senses with Optical Fe II equiva­
a ox .
lent widths decrease with X­ray brightness, while UV iron
equivalent widths increase. This conïrms similar sugges­
tions elsewhere (see, e.g., et al.
Green 1995 ; Lipari 1994 ;
& Meyers Proper modeling of nonradiative
Boroson 1992).
heating, optical depth e+ects, and iron recombination rates,
along with improved methods of Fe II emission measure­
ments, may all be needed to contribute to the solution of
this intriguing observation.
Broad emission lines di+er less between XB and XF
QSOs than do narrow lines. Our interpretation of the data
is that absorbers tend to obscure the line of sight to the
central ionizing source both from us and from the NLR. A
similar conclusion indicating that less altered continuum
radiation reaches the BLR was recently suggested from dif­
ferent lines of evidence by Ferland, & Baldwin
Korista,
(1997b).
Our tests on complementary subsamples indicate that
spectral di+erences between subsamples divided by a ox
exceed those seen between samples divided by luminosity or
radio loudness. In particular, we propose that the Baldwin
e+ect may be a secondary correlation to the primary
relationship between and emission line equivalent width.
a ox
One test of this is that the Baldwin e+ect should be domi­
nated by narrow­line components.
We note that for 23 UV­excess selected (PG) QSOs
observed in the ROSAT bandpass et al. corre­
(Laor 1997),
lations of emission line parameters (FWHM [Hb] and
optical Fe II and [O III] strengths) are clearly stronger with
than with However, a correlation exists between
a X a ox . a ox
and X­ray bright QSOs (with small tend to have ÿat
a X ; a ox )
(hard) At these and slightly lower luminosities (Seyfert 1
a X .
galaxies ; see, e.g., Brandt, & Fink only objects
Boller, 1996),
with ÿat (hard) have large FWHM (Hb). If Hb is repre­
a X
sentative, the combination of these two trends would
predict that XB QSOs should on average have broader
emission lines, counter to the overall trends seen in our
samples. Of the emission lines to which we can ït two com­
ponents (Lya, C IV, Mg II, and Hb), it is the only line for
which the percentage of total line ÿux in the narrow­line
component may be larger in the XF composite. Since Hb
appears to be unique among the larger variety of emission
lines studied here, it is clear that Hb may not be the best or
only representative of BLR line widths. Fitting the Hb line
in our optical composites with a single Gaussian com­
ponent leads to conclusions similar to those of these pre­
vious studies (i.e., that FWHM [Hb] is larger in the XF
composite). However, ïtting narrow­ and broad­line com­
ponents separately reveals that Hb has a stronger broad
component in the XB composite and a larger broad/narrow
line ratio, while the actual FWHM is similar to that in the
XF composite. This reveals that single Gaussian ïts to com­
pound lines must be interpreted with caution. Furthermore,
we also caution that the luminosity ranges for these (PG
and LBQS) samples are disjoint. However, since and not
a X
seems to have the more fundamental correlation with
a ox
both Hb and [O III] in et al. we highlight the
Laor (1997),
need for correlation of with a wider range of emission
a X
line measurements.
The parameters emphasized here, and narrow­line
a ox
emission, also appear to be linked to the following quan­
tities : X­ray spectral slope Fe II strength, luminosity,
a X ,
radio loudness (see, e.g., et al. et al.
Laor 1997 ; Lawrence
et al. We suggest here that the as
1997 ; Green 1995 ; BG92).
yet mysterious physical link between these diverse proper­
ties is intimately related to high­velocity outÿowing winds
near the nucleus that, by absorbing the intrinsic nuclear
continuum, strongly a+ect the radiation observed at larger
distances. The continuum impinging on the NLR is closest
to that received by distant observers like ourselves but is
quite di+erent from that arriving at the BLR. Material in
the BLR itself may reprocess the intrinsic continuum
issuing from the unshrouded QSO nucleus.
Only some of the correlations between measured line and
continuum parameters are intrinsic, and others simply add
dispersion to more primary correlations. The principal
physical processes must be extracted from the principle
observational eigenvectors in a multivariate, multi­
wavelength approach, with careful attention to continuum
slopes and detailed emission line ïts. We are accumulating
a high­quality homogeneous database including all this
information for a large sample of QSOs, primarily from the
ROSAT and HST archives. We believe that multi­
wavelength studies such as this show promise for signiïcant
advances in our understanding.

180 GREEN
Thanks to Marianne Vestergaard for performing the iron
subtraction, to Craig Foltz for the LBQS spectra, to Ken
Lanzetta for the IUE QSO atlas, and to Todd Boroson for
the optical Fe II template. The author gratefully acknow­
ledges support provided by NASA through grant NAG5­
1253 and contract NAS 8­39073 (ASC), as well as
HF­1032.01­92A awarded by the Space Telescope Science
Institute, which is operated by the Association of Uni­
versities for Research in Astronomy, Inc., under NASA con­
tract NAS 5­26555. This research has made use of the
NASA/IPAC Extragalactic Database (NED) which is oper­
ated by the Jet Propulsion Laboratory, California Institute
of Technology, under contract with the National Aeronau­
tics and Space Administration.
REFERENCES
J. A. 1977, ApJ, 214,
Baldwin, 679
J. A., Ferland, G., Korista, K., & Verner, D. 1995, ApJ, 455,
Baldwin, 119
R. 1993, ApJ, 412,
Barvainis, 513
T., Brandt, W. N., & Fink, H. 1996, A&A, 305,
Boller, 53
T. A., & Green, R. F. 1992, ApJS, 80,
Boroson, 109
T. A., & Meyers, K. A. 1992, ApJ, 397,
Boroson, 442
W. N., Mathur, S., & Elvis, M. 1997, MNRAS, 285,
Brandt, 25P
S., & Vio, R. 1990, A&A, 227,
Cristiani, 385
H., Sherwood, W., & Patnaik, A. R. 1996, ApJ, 471,
Falcke, 106
G. F., & Shields, G. A. 1985, in Astrophysics of Active Galaxies
Ferland,
and Quasi­stellar Objects, ed. J. Miller (Mill Valley : University Science),
157
P. J., Hewett, P., Foltz, C., Cha+ee, F., Weymann, R. J., & Morris,
Francis,
S. 1991, ApJ, 373, 465
P. J., Hooper, E. J., & Impey, C. D. 1993, AJ, 106,
Francis, 417
P. J. 1996, ApJ, 467,
Green, 61
P. J., Aldcroft, T., Mathur, S., & Schartel, N. 1997, ApJ, 484,
Green, 135
P. J., & Mathur, S. 1996, ApJ, 462,
Green, 637
P. J., et al. 1995, ApJ, 450,
Green, 51
B. A. 1992, Ph.D. thesis,
Grossan, MIT
F., Barlow, T. A., Cohen, R. D., Junkkarinen, V. T., & Burbidge,
Hamann,
E. M. 1997, in ASP Conf. Proc. 128, Mass Ejection from Active Galactic
Nuclei, ed. N. Arav, I. Shlosman, & R. Weymann (San Francisco : ASP),
19
P. C., Foltz, C. B., & Cha+ee, F. H. 1995, AJ, 109,
Hewett, 1498
M. J., & Trimble, V. 1984, AJ, 89,
Irwin, 93
K., Baldwin, J. A., Ferland, G., & Verner, D. 1997a, ApJS, 108,
Korista, 401
K., Ferland, G., & Baldwin, J. A. 1997b, ApJ, 487,
Korista, 555
G. 1994, in ASP Conf. Proc. 61, Astronomical Data Analysis Soft­
Kriss,
ware and Systems III, ed. D. R. Crabtree, R. J. Hanisch, & J. Barnes (San
Francisco ASP), 437
J. H., & Kallman, T. R. 1988, ApJ, 324,
Krolik, 714
Franca, F., Franceschini, A., Cristiani, S., & Vio, R. 1995, A&A, 299,
La 19
A., Fiore, F., Elvis, M., Wilkes, B. J., & McDowell, J. C. 1997, ApJ,
Laor,
477, 93
A., Elvis, M., Wilkes, B. J., McHardy, I., & Brandt, N. 1997,
Lawrence,
MNRAS 285, 879
S. 1994, ApJ, 436,
Lipari, L102
M. A., & Sargent, W. L. W. 1982, ApJ, 254,
Malkan, 22
J.­L., Wilkes, B. J., Elvis, M., Arnaud, K. A., & McDowell, J. C.
Masnou,
1992, A&A, 253, 35
S. 1994, ApJ, 431,
Mathur, L75
S., Elvis, M., & Wilkes, B. J. 1995, ApJ, 452,
Mathur, 230
P. S., Porter, A. C., & Green, R. F. 1994, ApJ, 436,
Osmer, 678
C. S. 1997, MNRAS, 286,
Reynolds, 513
N., et al. 1996, MNRAS,
Schartel, 283,1015
J. C., Ferland, G. J., & Peterson, B. M. 1995, ApJ, 441,
Shields, 507
T. J., & Pounds, K. A. 1989, MNRAS, 240,
Turner, 833
D., & Fan, X.­M. 1992, ApJS, 79,
Tytler, 1
M.­P., & Veron, P. 1996, ESO Sci. Rep., 17,
Veron­Cetty, 1
M., & Green, P. J. 1998, in
Vestergaard, preparation
T., Brinkmann, W., & Bergeron, J. 1996, A&A, 309,
Wang, 81
B. J., Tananbaum, H., Worrall, D. M., Avni, Y., Oey, M. S., &
Wilkes,
Flanagan, J. 1994, ApJS, 92, 53
B. J., & Brotherton, M. S. 1995, ApJ, 448,
Wills, 81
G., Marano, B., Mignoli, M., Zitelli, V., & Boyle, B. J. 1992,
Zamorani,
MNRAS, 256, 238
W., Fang, L. Z., & Binette, L. 1992, ApJ, 392,
Zheng, 74
W., Kriss, G. A., & Davidsen, A. F. 1995, ApJ, 440,
Zheng, 606
W., Kriss, G. A., Telfer, R. C., Grimes, J. P., & Davidsen, A. F. 1996,
Zheng,
ApJ, 475, 469
W., & Malkan, M. A. 1993, ApJ, 415,
Zheng, 517