Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~pgreen/Papers/chandrabals01.ps Äàòà èçìåíåíèÿ: Wed Sep 14 22:17:24 2005 Äàòà èíäåêñèðîâàíèÿ: Tue Oct 2 01:50:24 2012 Êîäèðîâêà: koi8-r Ïîèñêîâûå ñëîâà: extreme ultraviolet

THE ASTROPHYSICAL JOURNAL, 558 109õ118, 2001 September 1
( 2001. The American Astronomical Society. All rights reserved. Printed in U.S.A.
A CHANDRA SURVEY OF BROAD ABSORPTION LINE QUASARS
PAUL J. GREEN,1 THOMAS L. ALDCROFT,1 SMITA MATHUR,2 BELINDA J. WILKES,1 AND MARTIN ELVIS1
Received 2001 March 4 ; accepted 2001 May 15
ABSTRACT
We have carried out a survey with the Chandra X­Ray Observatory of a sample of 10 bright broad
absorption line (BAL) quasars (QSOs). Eight of 10 sources are detected. The six brightest sources have
only high­ionization BALs (hiBALs), while the four faintest all show low­ionization BALs (loBALs). We
perform a combined spectral ït for hiBAL QSOs (384 counts total ; 0.5õ6 keV) to determine the mean
spectral parameters of this sample. We derive an underlying best­ït power­law slope !\ 1.8 ^ 0.35,
which is consistent with the mean slope for radio­quiet QSOs from ASCA, but BAL QSOs require a
(rest­frame) absorbing column of cm~2, with a partial covering fraction of The
6.5 ~3.8 ‘4.5 ] 1022 D80 ~17 ‘9 %.
opticalõtoõX­ray spectral slope from 2500 to 2 keV) varies from 1.7 to 2.4 across the full sample,
(a ox ñ
consistent with previous results that BAL QSOs appear to be weak soft X­ray emitters. Removing the
absorption component from our best­ït spectral model yields a range of from 1.55 to 2.28. All six
a ox
hiBAL QSOs have deabsorbed X­ray emission consistent with non­BAL QSOs of similar luminosity.
The spectral energy distributions of the hiBAL QSOsõboth the underlying power­law slope and
the ïrst conclusive evidence that BAL QSOs have appeared to be X­ray weak because of
a ox õprovide
intrinsic absorption and that their underlying emission is consistent with non­BAL QSOs. By contrast,
the removal of the best­ït absorption column detected in the hiBAL QSOs still leaves the four loBAL
QSOs with values of that are unusually X­ray faint for their optical luminosities, which is consis­
a ox [ 2
tent with other evidence that loBALs have higher column density, dustier absorbers. Important questions
of whether BAL QSOs represent a special line of sight toward a QSO nucleus or rather an early evolu­
tionary or high­accretion phase in a QSO lifetime remain to be resolved, and the unique properties of
loBAL QSOs will be an integral part of that investigation.
Subject headings : galaxies : active õ quasars : emission lines õ quasars : general õ
ultraviolet : galaxies õ X­rays : galaxies
1. INTRODUCTION
While large surveys are rapidly increasing the number of
known quasars (QSOs), our understanding of the QSO phe­
nomenon grows more slowly. However, absorption lines
caused by material intrinsic to the QSO hold great promise
for revealing the conditions near the supermassive black
holes that power them. The richest and most extreme
absorption lines are found in quasars with broad absorp­
tion lines (BALs). About 10%õ15% of optically selected
QSOs have rest­frame ultraviolet spectra showing these
BALsõdeep absorption troughs displaced blueward from
the corresponding emission lines in the high­ionization
transitions of C IV, Si IV, N V, and O VI (high­ionization
BALs [hiBALs]). About 10% of BAL QSOs also show
broad absorption in lower ionization lines of Mg II or Al III
(low­ionization BALs [loBALs]). BAL QSOs in general
have higher optical/UV polarization than non­BAL QSOs,
but the loBAL subsample tends to have particularly high
polarization (Schmidt & Hines 1999) along with signs of
reddening by dust (Sprayberry & Foltz 1992 ; Egami et al.
1996). All the BALs are commonly attributed to material
along our line of sight ÿowing outward from the nucleus
with velocities of 5000 up to D50,000 km s~1. The observed
ratios of broad emission and absorption line equivalent
widths and the detailed proïles of C IV BALs
W j em/W j abs
1 Harvard­Smithsonian Center for Astrophysics, 60 Garden Street,
Cambridge, MA 02138 ; pgreen=cfa.harvard.edu, aldcroft=
cfa.harvard.edu, bwilkes=cfa.harvard.edu, melvis=cfa.harvard.edu.
2 Department of Astronomy, Ohio State University, 140 West 18th
Avenue, Columbus, OH 43210­1173 ; smita=astronomy.ohio­state.edu.
both imply that the covering factor of the BAL region must
be less than 20% (Hamann, Korista, & Morris 1993). This
observation, together with the similar fraction of QSOs
showing BALs, suggests that most or possibly all QSOs
contain BAL­type outÿows. The optical/UV emission lines
and continuum slopes of hiBAL QSOs are remarkably
similar to those of non­BAL QSOs (Weymann et al. 1991).
BAL QSOs may thus provide a unique probe of conditions
near the nucleus of most QSOs. Ironically, although viewed
from an obscured direction, BAL QSOs may nevertheless
be particularly revealing.
In the last decade, a signiïcant observational e+ort has
been dedicated to BAL QSOs in the UV and X­ray band­
passes. The absorbing columns typically inferred from the
UV spectra for the BAL clouds themselves (e.g., N H D
atoms cm~2 ; Korista et al. 1992) appear low
1020õ1021
enough that we would a priori expect very little X­ray
absorption (q > 1). It was initially a surprise then to dis­
cover that BAL QSOs are markedly underluminous in soft
X­rays compared to their non­BAL QSO counterparts.
Contrasting a complete sample of 36 BAL QSOs in the
Large Bright Quasar Survey (LBQS) and the ROSAT
All­Sky Survey (RASS) with carefully chosen comparison
samples, Green et al. (1995) revealed deïnitively that BAL
QSOs are soft X­ray quiet as a class. Deeper archival
ROSAT Position Sensitive Proportional Counter (PSPC)
pointings of 11 bona ïde BAL QSOs conïrmed this (Green
& Mathur 1996, hereafter GM96), yielding unusually steep
opticalõtoõX­ray slopes for BAL QSOs relative
(a ox º 1.93)
3 The variable is the slope of a hypothetical power­law from 2500
a ox ñ
to 2 keV ; a ox \ 0.384 log (L 2500 /L 2 keV ).
109

110 GREEN ET AL. Vol. 558
to non­BAL QSOs in the ROSAT bandpass. By
(a ox D 1.6)
assuming that the intrinsic (unabsorbed) spectral energy dis­
tributions (SEDs) of BAL QSOs are similar to those of
non­BAL QSOs, GM96 found that absorbing columns of
cm~2 are necessary to quench the X­ray ÿux to
N H intr D 1023
the observed (or upper limit) levels. Gallagher et al. (1999)
studied a sample of eight BAL QSOs with ASCA, of which
only two were detected. They estimated column densities of
º5] 1023 cm ~2 to explain the nondetections, which are
even higher than the ROSAT estimates. In some cases, the
absorber is probably Compton­thick (i.e., N H intr Z 1024
cm~2), as in ASCA observations of PG 0946]301 (Mathur
et al. 2000).
If the UV and X­ray absorption in quasars arises in the
same region (see, e.g., Mathur et al. 1994), the large derived
X­ray columns increase the best UV­derived estimates of
both the ionization and mass outÿow rate of BALs by 2õ3
orders of magnitude. These highly ionized BAL outÿows
then represent a signiïcant component of the QSO energy
budget, but a single­zone photoionization model may not
be appropriate, and other intriguing possibilities remain.
BAL QSOs have been interpreted as normal QSOs seen
along a line of sight either ablating o+ the edge of an
obscuring torus or accelerated from the surface of the accre­
tion disk in a wind (see, e.g., Murray & Chiang 1995 ;
deKool & Begelman 1995 ; Elvis 2000). In this case, the
inner wind­driven X­ray absorber shields the UV BAL
clouds so that the UV BAL zone has a lower ionization
than the X­ray absorber.
Even if the X­ray and UV absorbers are identical, the
geometry, covering factor, temperature, density, metallicity,
and ionization parameter of the absorbing clouds are
poorly constrained from UV absorption line studies alone.
The few absorption lines observed provide little if any con­
straint on the ionization of the absorbing material, leading
to the simplifying assumption that the observed ions are the
dominant species. Furthermore, BALs in the UV are often
saturated (Wang et al. 1999), and column densities derived
from UV measurements may also be signiïcantly underesti­
mated because of partial covering of the continuum source
(Hamann 1998 ; Arav et al. 1999). Higher ionization
absorbers are indicated not just by the X­ray absorption
but by the detection of UV absorption in Ne VIII, O VI, and
Si XII (Telfer et al. 1998). UV spectropolarimetry implies
columns consistent with X­ray results (Goodrich 1997)õthe
most common UV BALs are saturated but partially ïlled in
with scattered light.
Many results support the picture that BAL QSOs are
intrinsically normal QSOs, with the BAL region an impor­
tant part of every QSOîs structure. Suggestive links between
low­ionization BAL QSOs and IR­luminous mergers
(Fabian 1999) and similarities between BAL QSOs and
narrow­line Seyfert 1 galaxies (Mathur 2000 ; Brandt & Gal­
lagher 2000) may also support a scenario where BAL QSOs
are adolescent quasars in a transition phase, evolving from
active high (high Eddington fraction) to normal
L /L
Edd
QSOs. If the BAL phase represents a high accretion rate
period in a quasarîs lifetime, then an intrinsic power law
steeper than that for non­BAL QSOs might be expected, by
analogy to narrow­line Seyfert galaxies and Galactic black
hole candidate binary systems in outburst (Leighly 1999 ;
Pounds, Done, & Osborne 1995).
Are the intrinsic SEDs of BAL QSOs really the same as
non­BAL QSOs? X­ray spectroscopy can conïrm the
absorption interpretation and verify whether the underlying
(unabsorbed) emission supports the hypothesis that BAL
QSOs are typical QSOs seen from a privileged line of sight
or rather a di+erent phase or species of QSO. Unfor­
tunately, because of low observed ÿuxes, there is only the
following handful of BAL QSOs with X­ray spectroscopy :
1. In a 100 ks ASCA spectrum, Mathur et al. (2000)
found evidence that PG 0946]341 is Compton­thick, but
this again was based on assumptions that the underlying
spectrum and normalization was that of a normal QSO
since the counts were too few for detailed spectral ïtting.
2. Mathur et al. (2001) analyzed an ASCA spectrum of
the prototype BAL QSO PHL 5200 (with z \ 1.98), wherein
intrinsic absorption of was required,
N H intr D 5 ] 1023
covering 80% of the source. Intriguingly, the best­ït power­
law photon index4 in the 2õ10 keV range (! D 2.4õ2.8) for
PHL 5200 is steeper than typical for non­BAL QSOs.
3. The simultaneous ASCA/ROSAT ïtting of PG
1411]442 (Wang et al. 1999) shows a hard X­ray slope
typical for non­BAL QSOs (! D 2 ; George et al. 2000 ;
Reeves & Turner 2000), but there is also evidence for a
strong, steep (! \ 3) component of soft X­ray emission, in
which non­BAL QSOs typically show !D 2.5. At z \ 0.09,
however, PG 1411]442 is the least luminous BAL QSO
and su+ers signiïcant contamination from star­forming
regions in its host galaxy.
4. Gallagher et al. (2001) found one BAL QSO, PG
2112]059 (B \ 15.5, z \ 0.457), which has perhaps the
brightest ÿux of any BAL QSO. A best­ït power law of
slope partially covered by
!\ 1.98 ~0.27 ‘0.40, (97 ~26 ‘3 %) 1.0 ~0.49 ‘1.4
cm~2 of intrinsic absorption, suggests that this
] 1022
object could be a shrouded example of a typical QSO.
However, while the objectîs ```` balnicity îî index5 of 2980 km
s~1 seems to classify it ïrmly as a BAL QSO, the BALs are
atypically shallow and the derived column rather low.
Further X­ray spectroscopy is critical to our basic under­
standing of BAL QSOs, but it is needed for some more
typical objects and for as large a sample as is feasible. To
begin to address this problem systematically, we performed
a snapshot X­ray survey of BAL QSOs during Cycle 1 of
the Chandra X­Ray Observatory. We describe below the
chosen sample (° 2) and their Chandra observations, ensem­
ble spectral ïtting (° 3), X­ray brightness (° 4), and the sig­
niïcance of our ïndings (° 5). We summarize our ïndings in
° 6 and present a brief discussion of individual objects in the
sample in the Appendix.
2. SAMPLE AND OBSERVATIONS
We compiled a list of bona ïde BAL QSOs with magni­
tudes (usually B or brighter than 17. We derived
m pg )
expected count rates using the Chandra Portable Interactive
Multimission Simulator, assuming that the intrinsic SED
(before absorption) of BAL QSOs is similar to that of
typical radio­quiet QSOs at similar luminosities. For the
X­ray spectral photon index !, we used 2.5 in the soft X­ray
4 The photon index ! is related to the energy index a by a \![ 1.
5 Weymann et al. (1991) deïne the balnicity index by summing the
equivalent width (in units of km s~1) of any contiguous absorption that
falls in the 3000õ25,000 km s~1 range from the systemic redshift, if the
absorption feature exceeds 2000 km s~1 in width and is at least 10% below
the continuum level.

No. 1, 2001 CHANDRA SURVEY OF BROAD ABSORPTION LINE QUASARS 111
band (Schartel et al. 1996) and 1.8 above 2 keV (Lawson &
Turner 1997). The power­law normalizations were derived
from the observed optical magnitudes using values of a ox
typical for normal QSOs Green et al. 1995). We
(a ox \ 1.6 ;
then calculated the absorbed Chandra broadband ÿux
assuming an (z \ 0) absorbing column of cm~2,
N H \ 1022
which corresponds to an intrinsic column of N H intr D 1023
cm~2 at typical sample redshifts. We thus calculated our
proposed Chandra exposure times to result in a strong
detection for each source.
The resulting sample spans a wide range of BAL QSO
phenomena, including redshifts from 0.1 to 2.4, four dusty
loBAL QSOs, two loBAL QSOs with metastable excited
states of Fe II and Fe III (Hazard et al. 1987), a radio­
moderate BAL QSO (Becker et al. 1997), and a gravita­
tionally lensed BAL QSO. Table 1 lists the sample in order
of increasing right ascension and includes mostly nonõX­
ray information.
All sources were observed between 1999 December 30
and 2000 May 15 using the back­illuminated S3 chip of the
Advanced CCD Imaging Spectrometer (ACIS) on board
Chandra. For the (optically) brightest object IRAS
07598]6508 (B \ 14.3 mag), we used a subarray for more
rapid readout to avoid the possible pileup of counts in
ACIS. Table 2 lists the Chandra Observation ID (ObsID)
and exposure times, observation dates, observed count
rates, or 3 p upper limits. The total exposure time for the
sample of 10 objects is 36.2 ks. For each detected target,
X­ray celestial coordinates matched optical counterpart
coordinates to within D1A so that there is no ambiguity
about identiïcation.
3. DATA ANALYSIS AND SIMULTANEOUS
SPECTRAL FITTING
We used reprocessed6 data and extracted ACIS gain­
corrected pulse­height invariant (PI) spectra from a 2A. 5
region around each QSO using the PSEXTRACT script
6 CXCDS versions R4CU5UPD11.1 and higher, along with ACIS cali­
bration data from the Chandra CALDB 2.0.
TABLE 1
SAMPLE PROPERTIES
Ba N H Gal Polarization
Target z (mag) (]1020cm~2) BAL Ionization (%) Referencesb Commentsc
Q0059[2735 . . . . . . . . . . . . . 1.595 18.0 1.97 Low 1.43 ^ 0.16 1 Metastable Fe II and Fe III
Q0135[4001 . . . . . . . . . . . . . 1.850 17.3 1.97 High . . .
Q0254[334 . . . . . . . . . . . . . . . 1.863 17.8 2.26 High 0.0 ^ 0.04 2 N V and O VI BALs
IRAS 07598]6508 0.148 14.3 4.34 Low 1.45 ^ 0.14 3 IRAS and ASCA detection
FIRST J0840]3633 1.220 17.1 3.44 Low 4 4 Metastable Fe II and Fe III, Radio­moderate
Q0842]3431 . . . . . . . . . . . . . 2.120 17.5 3.41 High 0.55 ^ 0.02 5
UM 425 . . . . . . . . . . . . . . . . . . . 1.465 16.5 4.09 High 1.93 ^ 0.17 2 Gravitational lens ?, O VI BALs
LBQS 1235]1807B . . . . 0.449 16.9 1.96 Low 0.00 ^ 0.07 1 IRAS
Q1246[0542 . . . . . . . . . . . . . 2.236 16.4 2.17 High 0.87 ^ 0.07 2 ROSAT detection
SBSG 1542]541 . . . . . . . 2.371 16.8 1.27 High . . . Very high ionization
a The magnitudes from USNOA­2.0 (Monet 1998) for all but UM 425 (Michalitsianos et al. 1997). Magnitudes are uncorrected for the BALs.
B J
b The references are for polarization only.
c The references for comments can be found in the Appendix, where individual objects are discussed.
REFERENCES..õ(1) Lamy & Hutsemekers 2000 ; (2) Hutsemekers et al. 1998 (3) Schmidt & Hines 1999 (4) Brotherton et al. 1997 ; (5) Ogle et al. 1999.
TABLE 2
SAMPLE OBSERVATIONS AND DERIVED PROPERTIES
log F X (0.5õ8 keV)
Chandra TIME DATE OF COUNT RATE
TARGET OBSERVATION ID (ks) Observation COUNTS (counts ks~1) Absorbed Deabsorbed log L 2 keV a ox
Q0059[2735 . . . . . . . . . . . . . 813 4.39 2000 May 15 \5 \1.1 [14.19 [13.96 \26.13 [2.00
Q0135[4001 . . . . . . . . . . . . . 814 4.90 2000 Jan 2 23 4.7 [13.59 [13.29 26.94 1.84
Q0254[334a . . . . . . . . . . . . . . 815 2.43 2000 Jan 2 33 15.2 [12.96 [12.75 27.44 1.57
135 1.04 2000 Feb 15 27 27.9
IRAS 07598]6508 . . . 816 1.34 2000 Mar 21 10 6.7 [13.38 [13.19 24.73 2.34
FIRST J0840]3633 817 4.17 1999 Dec 30 8 1.9 [13.97 [13.85 25.98 2.11
Q0842]3431 . . . . . . . . . . . . . 818 4.09 2000 Jan 22 51 11.7 [13.17 [12.91 27.48 1.65
UM 425 . . . . . . . . . . . . . . . . . . . 819 2.61 2000 Apr 7 113 43.7 [12.53 [12.28 27.74 1.60
LBQS 1235]1807B 820 1.30 2000 Jan 21 \5 \3.8 [13.66 [13.43 \25.45 [2.01
Q1246[0542 . . . . . . . . . . . . . 821 5.41 2000 Feb 8 43 8.1 [13.34 [13.12 27.30 1.90
SBSG 1542]541 822 4.55 2000 Mar 22 78 19.7 [13.05 [12.79 27.64 1.73
NOTE.õUnits of and are ergs cm~2 s~1 and ergs s~1 Hz~1, respectively. The deabsorbed ÿux values and are all calculated using our
F X L 2 keV L 2 keV a ox
best­ït partial covering spectral model from Table 3, with the intrinsic (redshifted) absorption component removed from the best­ït model. We note that the
use of the absorbed ÿuxes would decrease by about 0.23 and thereby increase by about 0.1.
log L X a ox
a Fluxes and luminosities calculated from average count rate of the two Chandra observations.

112 GREEN ET AL. Vol. 558
TABLE 3
SPECTRAL FIT PARAMETERS
N H intr Covering
Model ! (]1022 cm~2) Fraction s2 (DOF)a
A . . . . . . 1.08 ~0.13 ‘0.13 . . . . . . 75.8 (62)
B . . . . . . 1.44 ~0.22 ‘0.23 1.6 ~0.8 ‘0.9 . . . 64.6 (61)
C . . . . . . 1.80 ~0.35 ‘0.35 6.5 ~3.8 ‘4.5 0.80 ~0.17 ‘0.09 56.9 (60)
NOTE.õThe ïtted parameters are based on simultaneous ïtting of
unbinned spectra using Cash statistics. Uncertainties are 90% con­
ïdence limits. Models A ïts a global power­law continuum of photon
index ! with individual neutral Galactic absorption of column (see
N H Gal
Table 3) ; Model B includes global neutral absorption of column at
N H intr
each quasarîs redshift Model C allows for a global partial covering
fraction of the continuum by N H intr .
a The values of s2 are based on spectra binned to 5 counts bin~1
using given ïtted parameters.
described in the standard thread for the Chandra Interactive
Analysis of Observations (CIAO2.0). This script creates an
aspect histogram ïle, and the response matrix and ancil­
lary response7 calibration ïles (RMFs and ARFs, respect­
ively) appropriate to the source position on chip (which is
time­dependent because of dither) and CCD temperature
([120 C). We extract background in PI space using an
annulus extending typically from 5A to 50A around the
source. In every case, the total background normalized to
the source extraction area was less than 1 count, so we
henceforth ignore background. In all analyses, we ignored
channels below 0.5 keV since the ACIS response at lower
energies is not well calibrated. Above 0.5 keV, the cali­
bration is accurate to better than 10%. Channels above 6
keV were also ignored because of insufficient counts. Two
ïnal PI spectrum ïles were created for each source, one
with no binning and one binned to a minimum of 5 counts
bin~1.
We perform spectral modeling for the six sources from
Table 2 with more than 20 counts. We used SHERPA, a
generalized modeling and ïtting environment within
CIAO2.0. Since each source spectrum taken individually
has insufficient counts to usefully constrain the intrinsic
absorption or power­law spectral index, instead we simulta­
neously ïtted all six spectra. We ïtted only the six BAL
QSOs from Table 2 with more than 20 counts each. Note
that these sources are all hiBAL QSOs, so the spectral pa­
rameters we derive may not apply to loBAL QSOs. We
tested several source models, for which the best­ït values
are recorded in Table 3.
Model A is simply a global power law with an individual
ÿux normalization for each QSO and (z \ 0) absorption
ïxed to the Galactic value for each QSO :
N(E) \A i E~!e~NH G a i l p(E) photons cm~2 s~1 keV~1 .
In this formula, is the normalization for the ith spectrum,
A i
but ! is a global power­law emission component. is the
N H, i
equivalent Galactic neutral hydrogen column density that
characterizes the e+ective absorption (by cold gas at solar
abundance) for the ith source, with p(E) being the corre­
sponding absorption cross section (Morrison & McCam­
7 RMFs are used to convert the ACIS pulse height (deposited charge) to
energy. ARFs calibrate the e+ective collecting area of a speciïed source
region on ACIS as a function of incident photon energy.
mon 1983). This simple ït yields an unusually ÿat
continuum slope (! \ 1.08 ^ 0.13), which is a signal that
intrinsic absorption may be present. For determining the
best­ït parameter values, we use Powell optimization with
Cash statistics. This allows the use of unbinned spectral
data, and we quote 90% conïdence limits on ïtted parame­
ters in Table 3 and hereafter.
In Model B, we add a neutral absorber at the systemic
redshift of each spectrum by multiplying Model A by a
further term Here the key feature is that all
e~NH intr p*E(1‘zi)+.
the intrinsic column density parameters are linked
N H intr
together, giving just a single free ```` intrinsic absorption îî
component. Similarly, the overall model amplitudes are free
to vary individually, but again only one global power­law
spectral index is ïtted. The best­ït slope of Model B is
!\ 1.44 ^ 0.23, with intrinsic (rest­frame) absorption
cm~2.
N H intr \ 6.5 ~3.8 ‘4.5 ] 1022
We examined the relative quality of di+erent model ïts
using s2 statistics, which must be performed on binned data.
We binned the photon events to 5 counts bin~1 and esti­
mate the variance using the background and source model
amplitudes rather than the observed counts data
(STATISTIC CHI MVAR in CIAO2.0). Table 3 presents
the best­ït (Cash) model parameters together with their
reduced s2 statistics. The results of s2 ïtting conïrm that
the inclusion of a redshifted absorber (Model B) improves
the ït at 99.7% (3 p) conïdence (using the F­test).
Inclusion of a global partial covering parameter for
C f
the redshifted absorbers (Model C) substitutes the intrinsic
absorption term in Model B with the expression
C f e~NH intr p*E(1‘zi)+ ] (1 [C f ) .
Here the last term in parentheses represents the fraction of
light that escapes the source without absorption. Model C
improves the ït, again at 99.5% conïdence (F­test), relative
to a redshifted absorber with no partial covering. The
```` composite îî BAL QSO has intrinsic (rest­frame) absorp­
tion cm~2 covering of the
N H intr \ 6.5 ~3.8 ‘4.5 ] 1022 80 ~17 ‘9 %
source, whose intrinsic power­law energy index
!\ 1.80 ^ 0.35.
In Figure 1, we present the summed Chandra X­ray spec­
trum for the six BAL QSOs with more than 20 counts. The
sum of all the individual source models from the global
best­ït Model C is overplotted, both with and without the
absorber. The dashed line shows the ```` deabsorbed îî model
spectrum, where the intrinsic absorption component is
removed from the best­ït model. Residuals for (similarly
summed) models A, B, and C are also shown. We caution
that this is essentially a composite of residuals from individ­
ual sources with di+erent values of redshift and galactic
absorption, and so features do not correspond directly to
those expected in a single spectrum. However, the result is
useful for visualization purposes since the redshifts for the
spectral subsampleõfrom 1.465 to 2.371, with mean z \
not to range so widely as in the full
1.98 ^ 0.33õhappen
sample. Neither is the counts­weighted redshift of 1.93 sig­
niïcantly di+erent from this mean.
Figure 2 shows the conïdence contours for Model C, in
which it can be seen that the absorption is required at more
than 2 p conïdence. The best­ït power­law index ! for our
BAL QSO sample is entirely consistent with the mean of
D1.89 ^ 0.05 with a dispersion of p \ 0.27 ^ 0.04 seen
with ASCA for radio­quiet (RQ) QSOs at redshifts z [ 0.05
(Reeves & Turner 2000). Measurements in a similar redshift

No. 1, 2001 CHANDRA SURVEY OF BROAD ABSORPTION LINE QUASARS 113
FIG. 1.õL eft : Summed Chandra X­ray spectrum for the BAL QSOs with more than 20 counts. The sum of the all the individual source models is plotted
over the merged event lists of all six objects. The solid line shows the global best­ït model (Model C in Table 3). The dashed line shows the ```` deabsorbed îî
model spectrum, in which the intrinsic absorption component is removed from Model C after ïtting. Right : Residuals for models A, B, and C (Table 3 and
° 3). These represent the overall sum of the residuals in the observed frame, so that remaining rest­frame features would appear blurred in this representation.
range are perhaps more relevant, so we compiled ASCA
measurements from Reeves & Turner (2000) and Vignali et
al. (1999) for all 10 of the z [ 1.3 RQ QSOs with measured
power­law energy indices. Redshifts for this comparison
sample range from 1.3 to 3.0, with a mean of 2.1. The
average index for the comparison sample is !\ 1.8 with
dispersion 0.15, which is indistinguishable from our results
for the Chandra BAL QSO sample.
FIG. 2.õJoint (1, 2, and 3 p) conïdence intervals for spectral ït parameters for our simultaneous ït to the six Chandra BAL QSOs with more than 20
counts using Model C (Table 3 and ° 3). L eft : Redshifted intrinsic absorption and power­law spectral index !. Right : Conïdence intervals for redshifted
absorption and covering fraction.

114 GREEN ET AL. Vol. 558
The quality of the spectra are not sufficient to also con­
strain ionization or metallicity of the absorber, justifying
the assumption of neutral absorbers with solar metallicity
in our modeling. Modeling with either higher metallicities
or with ionized absorbers would only increase the required
intrinsic column in the best­ït models but is very unlikely to
substantially change the power­law slope.
4. X­RAY BRIGHTNESS
Now that we have a measured mean spectral shape for
hiBAL QSOs, for the ïrst time we can calculate ÿuxes con­
sistently using the best­ït model with the redshifted absorp­
tion component removed. This tells us what values BAL
a ox
QSOs would have without their intrinsic absorption since
their (deabsorbed) intrinsic SEDs are well characterized by
the above slope. We use the best­ït composite X­ray spec­
tral model to calculate the observed ÿuxes in the 0.5õ8 keV
band in Table 2. We derive the deabsorbed ÿux in the same
band and use these to calculate the monochromatic rest­
frame luminosities at 2 keV, also shown in Table 2. Optical
magnitudes from Table 1 are used to derive the 2500 ñ
luminosities, and from these we calculated the opticalõtoõ
X­ray index All luminosities are calculated assuming
a ox .
km s~1 Mpc~1 and with speciïc optical
H 0 \ 50 q 0 \ 0.5,
normalization from Marshall et al. (1984).
Using the deabsorbed ÿuxes from our full best­ït model
in the observed Chandra band (0.5õ8 keV) and also a consis­
tent power­law slope !\ 1.8 for the K­correction, the
resulting values (or limits) range from 1.56 to 2.36, with a
a ox
mean of 1.87. We note that use of the absorbed ÿuxes would
decrease by about 0.23 and thereby increase by
log L X a ox
about 0.1.
We must be careful when we compare for our BAL
a ox
QSO sample to previous results derived from observed
ÿuxes in di+erent (e.g., ROSAT ) bandpasses or assuming
di+erent X­ray slopes. As a consistency check with previous
results (e.g., GM96), we ïrst simulate what would have been
seen by ROSAT . To do this, we calculate with our full
best­ït model the ÿux that would be observed in the
ROSAT (0.5õ2 keV) band. The resulting values range
a ox
from 1.7 to 2.5, with a mean of 2.0, which is consistent with
the ROSAT BAL QSO results for GM96 (most of which
were nondetections).
Figure 3 shows the deabsorbed luminosities and for
a ox
our sample relative to the composite points for large
samples of radio­quiet QSOs observed by ROSAT (Green
et al. 1995 ; Yuan et al. 1998). We caution that those
ROSAT points are calculated in the ROSAT bandpass
assuming a steeper slope !\ 2.5, applicable to ROSAT ­
observed radio­quiet quasars. With the modeled intrinsic
absorption removed, the hiBAL QSOs in our sample ït
reasonably well along the empirical trend of increasing a ox
(weakening X­ray emission) with increasing On the
L opt .
other hand, the four low­ionization BAL QSOs in our
sample are extremely X­ray weak. Two are not detected at
all (for which we assign 5 counts as an upper limit). Of the
two loBAL QSOs that are detected, one is the most nearby
object (at z \ 0.148), and the other is a radio­intermediate
BAL QSO.
5. DISCUSSION
Previous estimates of column densities in BAL QSO
samples came by assuming that each BAL QSO had an
intrinsic X­ray continuum of shape and normalization
FIG. 3.õL eft : The log of the monochromatic (2 keV) X­ray luminosity plotted against the log of monochromatic 2500 optical luminosity for quasars
ñ
(both in units of ergs s~1 Hz~1). Right : OpticalõtoõX­ray spectral slope (from 2500 to 2 keV), also plotted against In both panels, the circles
a ox ñ log L 2500 .
depict the 10 BAL QSOs in our Chandra sample. The ïlled circles are those objects known to have loBALs. The X­ray luminosity and are deabsorbed, i.e.,
a ox
calculated without using our best­ït Model C. The arrows mark limits to X­ray luminosity in our Chandra exposures. The len