Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~bradw/cv/papers/maximBG.ps.gz Äàòà èçìåíåíèÿ: Mon Jan 13 23:51:09 2003 Äàòà èíäåêñèðîâàíèÿ: Sat Dec 22 03:16:26 2007 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï

CHANDRA SPECTRA OF THE SOFT X­RAY DIFFUSE BACKGROUND
M. Markevitch, M. W. Bautz, 1 B. Biller, Y. Butt, R. Edgar, T. Gaetz, G. Garmire, 2 C. E. Grant, 1
P. Green, M. Juda, P. P. Plucinsky, D. Schwartz, R. Smith, A. Vikhlinin, S. Virani,
B. J. Wargelin, and S. Wolk
Harvard­Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138; maxim@head­cfa.harvard.edu
Received 2002 June 25; accepted 2002 October 2
ABSTRACT
We present an exploratory Chandra ACIS­S3 study of the di#use component of the cosmic X­ray
background (CXB) in the 0.3--7 keV band for four directions at high Galactic latitudes, with emphasis on
details of the ACIS instrumental background modeling. Observations of the dark Moon are used to model
the detector background. A comparison of the Moon data and the data obtained with ACIS stowed outside
the focal area showed that the dark Moon does not emit significantly in our band. Point sources down to
3 # 10 #16 ergs s #1 cm #2 in the 0.5--2 keV band are excluded in our two deepest observations. We estimate the
contribution of fainter, undetected sources to be less than 20% of the remaining CXB flux in this band in all
four pointings. In the 0.3--1 keV band, the di#use signal varies strongly from field to field and contributes
between 55% and 90% of the total CXB signal. It is dominated by emission lines that can be modeled by a
kT ¼ 0:1 0:4 keV plasma. In particular, the two fields located away from bright Galactic features show a
prominent line blend at E # 580 eV (O vii+O viii) and a possible line feature at E # 300 eV. The two
pointings toward the North Polar Spur exhibit a brighter O blend and additional bright lines at 730--830 eV
(Fe xvii). We measure the total 1--2 keV flux of 1:0 1:2 # 0:2
Ï ÷ # 10 #15 ergs s #1 cm #2 arcmin #2 (mostly
resolved) and the 2--7 keV flux of 4:0 4:5 # 1:5
Ï ÷ # 10 #15 ergs s #1 cm #2 arcmin #2 . At E > 2 keV, the di#use
emission is consistent with zero, to an accuracy limited by the short Moon exposure and systematic
uncertainties of the S3 background. Assuming Galactic or local origin of the line emission, we put an upper
limit of #3 # 10 #15 ergs s #1 cm #2 arcmin #2 on the 0.3--1 keV extragalactic di#use flux.
Subject headings: intergalactic medium --- ISM: general --- methods: data analysis ---
X­rays: di#use background --- X­rays: ISM
1. INTRODUCTION
The existence of a cosmic X­ray background (CXB) was
one of the first discoveries of extrasolar X­ray astronomy
(Giacconi et al. 1962). In the intervening four decades,
observations with improving angular and spectral resolu­
tion have enhanced our understanding of the components
that make up this background. Several broadband, all­sky
surveys have been performed using proportional counter
detectors (Marshall et al. 1980; McCammon et al. 1983;
Marshall & Clark 1984; Garmire et al. 1992; Snowden et
al. 1995, 1997; for a review of pre­ROSAT results, see
McCammon & Sanders 1990). These surveys form a consis­
tent picture of the angular distribution of X­ray emission in
the various bands. Above 2 keV, the emission is highly iso­
tropic on large angular scales and has an extragalactic ori­
gin. Below 2 keV, the X­ray background is a mixture of
Galactic di#use emission (see, e.g., Kuntz & Snowden 2000
and references therein), heliospheric and geocoronal di#use
components (see, e.g., Cravens 2000), and extragalactic flux
from point sources and, possibly, from intergalactic warm
gas that may contain the bulk of the present­day baryons
(see, e.g., Cen &Ostriker 1999).
The earliest observations could provide only limited spec­
tral information on the background. Marshall et al. (1980)
found that the spectrum in the 3--50 keV range was well fit­
ted by a thermal bremsstrahlung model with kT # 40 keV.
In the 3--10 keV band, this can be approximated by a power
law with a photon index of #1.4. At energies below 1 keV,
the background surface brightness exceeds the extrapola­
tion of this power law (Bunner et al. 1969). Later observa­
tions with gas scintillation proportional counters and solid­
state detectors (Inoue et al. 1979; Schnopper et al. 1982;
Rocchia et al. 1984) suggested emission lines in the 0.5--
1.0 keV band, most likely from oxygen. The evidence for
emission lines in this band has become more convincing in
recent observations using CCDs (Gendreau et al. 1995;
Mendenhall & Burrows 2001) and in a high­resolution spec­
trum obtained by McCammon et al. (2002) in a microcalori­
metric experiment. A definitive demonstration of spectral
lines in the 0.15--0.3 keV band (at low Galactic latitude) was
obtained by Sanders et al. (2001), using a Bragg crystal spec­
trometer. Observations that combine high spectral and
angular resolution are essential for disentangling the many
CXB soft emission components.
Chandra and XMM should soon provide a wealth of
new information on the CXB. Several works have already
taken advantage of the Chandra's arcsecond resolution to
study the point­source component of the CXB (see, e.g.,
Mushotzky et al. 2000; Brandt et al. 2001; Rosati et al.
2002). The first XMM results are starting to appear as well
(De Luca & Molendi 2002; Warwick 2002; Lumb et al.
2002), utilizing the large e#ective area of that observatory.
In this paper, we present a Chandra ACIS study of the dif­
fuse CXB at high Galactic latitudes. The main advantage of
Chandra over all other instruments is its ability to resolve
point sources down to very low fluxes and probe the true dif­
1 Center for Space Research, Massachusetts Institute of Technology,
Cambridge, MA 02139.
2 Department of Astronomy and Astrophysics, Pennsylvania State
University, 525 Davey Laboratory, University Park, PA 16802.
The Astrophysical Journal, 583:70--84, 2003 January 20
# 2003. The American Astronomical Society. All rights reserved. Printed in U.S.A.
70

fuse background. In addition to that, compared to ROSAT
PSPC (which had lower detector background), ACIS has
energy resolution su#cient to identify spectral lines. Com­
pared to XMM EPIC, ACIS appears to be less a#ected by
instrumental background flares (although during quiescent
periods, the ACIS detector background per unit sky signal
is higher). To the extent that results can be compared,
we confirm many of the recent findings made with other
instruments.
Technical aspects of our study, especially the ACIS
instrumental background modeling, are quite complex, and
we discuss them here in detail. Much of our analysis proce­
dure may be useful for studies of extended sources such as
clusters of galaxies. Uncertainties are 1 # unless specified
otherwise.
2. DATA SET
For this exploratory study of the di#use CXB, we selected
four Chandra ACIS­S observations at high Galactic lati­
tudes, listed in Table 1. Our main focus is two relatively deep
(90--100 ks) observations, observation IDs (ObsIDs) 3013
and 3419, obtained at positions away from any bright
Galactic features seen in the RASS (ROSAT All­Sky
Survey) R4--R5 ( 3
4 keV) band (Snowden et al. 1997). For
comparison, we also analyze two shorter archival observa­
tions toward an edge (ObsID 869) and the middle (ObsID
930) of the North Polar Spur, 3 which exhibits bright emis­
sion in the ROSAT R4--R5 band. In Figure 1, positions of
the four observations are overlaid on the RASS R4--R5
map. None of these observations' original goals were related
to CXB, and there are no nearby cataloged extended sour­
ces, except an irregular galaxy that was a target of ObsID
869 and is spatially excluded from our analysis.
3. ACIS INSTRUMENTAL BACKGROUND
A critical part of this study is modeling of the ACIS
instrumental background. This background is caused by
cosmic charged particles and consists of a slowly changing
quiescent component and at least two species of highly vari­
able background flares, whose spectra are very di#erent
from that of the quiescent component. Below we describe in
detail how these components were dealt with from a practi­
cal perspective; their exact physical nature is beyond the
scope of this paper.
3.1. Background Flare Filtering
We use data from the ACIS back­side--illuminated (BI)
chip S3. Compared to the ACIS front­side--illuminated (FI)
chips, S3 has a higher sensitivity at low energies. However,
its sensitivity to low­energy X­rays also renders it more sen­
sitive to particle events, which results in more frequent
background flares than in FI chips (Plucinsky & Virani
2000; Markevitch 2001 4 ). The quiescent background is
stable and predictable; therefore, when the accuracy of the
background subtraction is critical, it is best to exclude flare
periods from the analysis. The spectra of the flaring and qui­
escent background components (discussed below) are such
that the best energy band in which to look for flares in chip
S3 is approximately 2.5--7 keV. Figure 2 shows light curves
in this energy band for our four observations (from the
whole S3 chip, excluding celestial sources). The time bin size
(#1 ks) is chosen to limit the statistical scatter while provid­
ing a reasonably detailed light curve.
In observations 3419, 869, and 930, the quiescent rate is
easily identifiable and very close to that in most other obser­
vations performed during 2000--2001. To limit the back­
ground modeling uncertainty, we exclude from further
analysis all time bins above and below a factor of 1.2 of this
rate (the rate can be lower, for example, because of occa­
sional short intervals of missing telemetry and bad aspect,
etc.). The resulting clean exposures are given in Table 1.
3.1.1. Anomalous Background in Field 3013
Observation 3013 is unusual in that the apparent `` quies­
cent '' rate between the numerous flare intervals (Fig. 2) is
about 30% higher than in other observations. It is also more
variable than the quiescent rate usually is. It appears that, in
fact, all of this exposure is a#ected by a long flare, which
requires special treatment (and, unfortunately, will add to
the systematic uncertainty of the results).
Again, we exclude the time periods above a factor of 1.2
of the apparent quiescent rate. Assuming that the back­
ground excess in the rest of the exposure is indeed a flare, we
can try to model it by taking advantage of the empirical
finding, based on a number of observations, that the spec­
tral shape of the most frequent, `` soft '' species of the BI
flares stays the same even while their flux varies strongly in
time. To illustrate this, we derive spectra of the background
flare components from the rejected, high count rate periods
of observations 3013 and 930. We first check the light curves
of FI chips also used in these observations, in order to
exclude flares of a di#erent, `` hard '' species that is less fre­
quent, a#ects both the BI and FI chips, and has a di#erent
spectrum. There are 1--2 ks of such flares within the already
excluded time periods in each observation. After excluding
those, we extract the spectra from the remaining high­rate
periods and subtract from them the spectra of the quiescent
periods of the same observations, normalizing them by
the respective exposure ratios (thereby subtracting all
3 The North Polar Spur, part of the Loop I supershell of emission in the
RASS (Egger & Aschenbach 1995), is thought to be the collision of explo­
sive remnants with the Local Bubble.
4 For Markevitch (2001), see http://asc.harvard.edu/cal, sections
`` ACIS,'' `` Background,'' `` General discussion.''
Fig. 1.---ROSAT PSPC all­sky map of CXB in the R4--R5, or 3
4 keV,
band (Snowden et al. 1997), in Galactic coordinates. Positions of our obser­
vations are marked; labels give ObsIDs.
DIFFUSE SOFT X­RAY BACKGROUND 71

time­independent sky and instrumental background com­
ponents). The resulting flare spectra are shown in Figure 3.
In a similar manner, we also extracted flare spectra from
several other archival observations (ObsIDs 766, 326, 2206,
2076, 2213, and 1934) spanning a period 2000 February--
2001 September and a range of flare intensities. Flares in all
those observations can be described by a model consisting
of a power law with a photon index of #0.15 and an expo­
nential cuto# at 5.6 keV, without the application of the tele­
scope e#ective area and the CCD quantum e#ciency
(commands arf none or model/b in XSPEC). In Figure 3,
this model is overplotted on the flare spectra from our CXB
fields 930 and 3013, fixing the spectral shape and fitting only
the normalization. As the figure shows, flares in both our
observations have nearly the same spectral shape, despite
their time di#erence of 1.5 yr and very di#erent intensities.
3013 3419
869 930
Fig. 2.---Light curves of the four CXB observations in the 2.5--7 keV band, where the contribution of flares is most easily detected. ObsIDs are marked in
each panel. For ObsID 869, only 72% of the chip area is used. Very Faint mode filtering was applied (x 3.3). Shaded bins are above or below a factor of 1.2 of
the quiescent level and are excluded.
TABLE 1
Data Summary
Parameter ObsID 3013 ObsID 3419 ObsID 869 ObsID 930 Moon
l; b
Ï ÷ (deg).......................................................................... (259.6, +56.9) (187.1, #31.0) (36.6, +53.0) (358.7, +64.8) . . .
Galactic N H (#10 20 cm #2 ) .................................................. 4.1 11.4 4.3 1.8 . . .
ROSATR4--R5 flux (#10 #6 counts s #1 arcmin #2 ) .............. 160 90 200--250 a 400 . . .
Observation date................................................................ 2001 Dec 13 2002 Jan 8 2000 Jun 24 2000 Apr 19 2001 Jul 26
Total (uncleaned) exposure (ks) ......................................... 112 98 57 40 16
Exposure for source detection (ks) ..................................... 101 92 52 28 . . .
Exposure for CXB spectra (ks) ........................................... 69 86 52 20 11
Field solid angle (arcmin 2 ).................................................. 69 69 51 69 70
a A#ected by an artifact in the ROSAT all­sky map.
72 MARKEVITCH ET AL. Vol. 583

Freeing the spectral shape parameters, we obtained a
photon index of #0.10 # 0.07 and a cuto# at 5:2 # 0:7 keV
for observation 930 and #0.1 # 0.3 and 7:2 ×1
#3:4 keV for
3013, consistent with the above fit for the composite flare
spectrum.
Therefore, if the background excess a#ecting the useful
period of observation 3013 is indeed a residual flare, we can
expect that it has the same spectral shape, only a still lower
normalization. As is seen in x 6, its spectrum is indeed con­
sistent with this assumption at the energies for which the
comparison is possible. To try modeling the residual back­
ground excess in 3013, we chose to fix the shape parameters
to the best­fit values from this particular observation (given
above), even though they are less strongly constrained than
those from the composite spectrum, to account for any pos­
sible slow evolution of the spectrum, which cannot be ruled
out with the data at hand. This choice has no significant
e#ect on our results. The normalization of this flare model is
determined in x 6, after subtraction of the quiescent back­
ground component and removal of point sources.
For comparison, Figure 3 also shows a quiescent instru­
mental background spectrum from the dark­Moon observa­
tions (discussed below). We note that the BI soft flare
spectrum at high energies is much softer than the quiescent
background; unless the flare is very strong, its contribution
above 10 keV is unnoticeable (this is not true for the other,
hard flare species mentioned above). This fact is used below.
3.2. Quiescent Background
3.2.1. Dark­Moon Observations
To separate the CXB and instrumental components of
the ACIS background, Chandra observed the dark Moon in
2001 July in a series of six short pointings (ObsIDs 2469,
2487, 2488, 2489, 2490, and 2493), tracking the Moon for a
total of about 15 ks. ACIS chips S2, S3, I2, and I3 were on
and telemetered data in Very Faint (VF) mode. A second
installment of Moon observations in 2001 September
exposed chips I2 and I3. Technical di#culties encountered
in these two runs, related mostly to the fact that the Chandra
aspect camera cannot be used near the Moon, prevented
further dark­Moon observations. Here we use only the S3
data from 2001 July. During that run, optical flux from the 1
3
of the lunar disk that was illuminated was imaged onto to
the ACIS focal plane. The ACIS optical blocking filters
were not designed to reject visible light from the sunlit
Moon, and a detectable o#set signal (bias error) was pro­
duced. The e#ect was most severe in chip I2, but very small
in chip S3. A correction to each event's pulse­height ampli­
tude (PHA) was calculated individually by averaging the
lowest 16 pixels of the 5 # 5 pixel VF mode event island.
This o#set in S3 was well below the threshold of a#ecting
the event grades, so this problem did not result in any loss of
events due to the onboard grade rejection (as was the prob­
lem for chip I2 in this data set). The average correction to
the energy for events in chip S3 was within a few eV, negli­
gible for our purposes.
A 2.5--7 keV S3 light curve for all 2001 July dark­Moon
observations is shown in Figure 4. The end of the exposure
was a#ected by an apparent faint flare, which was filtered
using the same factor of 1.2 threshold as in x 3.1 (we had to
use smaller time bins, which resulted in some statistical devi­
ations that were also excluded for consistency). The result­
ing clean Moon exposure for S3 is 11,400 s. At high
energies, where the CXB contribution is negligible, the
Moon quiescent background rate was within a few percent
of that in other recent observations. Scientific results from
the Moon observations will be discussed by C. E. Grant et
al. (2003, in preparation).
3.2.2. Event Histogram Mode Data
An independent approach to calibrating the ACIS instru­
mental background utilizes the event histogram mode
(EHM) data (Biller, Plucinsky, & Edgar 2002). 5 These data
are collected during science observations by the HRC­I
detector while ACIS is stowed inside the detector support
structure. This structure blocks celestial X­rays but does not
a#ect the particle rate significantly (as is seen below). In this
mode, the telemetry capacity available for ACIS is small, so
the only information transmitted is a PHA histogram for
the events from a predefined region of the chip. For this
reason, the usual exclusion of bad pixels and the position­
dependent gain correction cannot be applied. At this
location, ACIS is also faintly illuminated by the internal
Fig. 3.---Spectra of the excluded background flares (see text) in ObsIDs
930 (red ) and 3013 (black), compared to the quiescent spectrum from the
dark Moon (blue), for the whole S3 chip. Red histograms show a model that
was fitted to a combination of other observations with flares and renormal­
ized, without a change of shape, to match the ObsID 930 and 3013 spectra.
Fig. 4.---Same as Fig. 2, but for the 2.5--7 keV light curve for the dark­
Moon observations. The rate is shown before VFmode filtering (x 3.3).
5 See http://asc.harvard.edu/cal, sections `` ACIS,'' `` Background,''
`` Event Histogram mode.''
No. 1, 2003 DIFFUSE SOFT X­RAY BACKGROUND 73

calibration line source. The flare component discussed in
x 3.1 is never observed in the stowed position.
An EHM spectrum from the whole S3 chip accumulated
over the 2001 July--October period (straddling the date of
the Moon observation) is shown in Figure 5 (see also Biller
et al. 2002). For comparison, we overlaid a dark­Moon
spectrum. In order for it to be directly comparable to the
EHM spectrum, in deriving it we did not exclude bad CCD
pixels, apply gain corrections (that is, we used PHA rather
than pulse height invariant, or PI, values), or apply the addi­
tional VFmode filtering (x 3.3).
Figure 5 shows that away from the calibration source
lines, the agreement between the spectra is quite remarkable
over the entire energy range, within the statistical accuracy
of the Moon data set. One might expect both the dark­
Moon and the EHM spectra to exhibit emission above the
non--cosmic background level seen in ordinary observa­
tions; e.g., ROSAT and ASCA Moon data suggested emis­
sion at low energies (Schmitt et al. 1991; Kamata et al.
1999), and the detector support structure may be radioac­
tive. In principle, one could also imagine a component of
the quiescent particle background that can be blocked by
that structure. While the coincidence of the two spectra in
Figure 5 does not rule out a conspiracy of these possibilities,
it makes each of them very unlikely. It also supports the
conclusion by Freyberg (1998) that the ROSAT emission in
the direction of the dark Moon was actually fluorescent
emission from a region around the Earth (i.e., below the
Chandra orbit). Therefore, we assume that both the dark­
Moon and the EHM data give the true quiescent back­
ground, and we use the Moon data set as an instrumental
background model for the sky data below.
The EHM data were also used directly as the background
model for a CXB study by R. J. Edgar et al. (2003, in prepa­
ration). Another measure of the ACIS background was
obtained in 1999 August, just prior to opening the Chandra
telescope door. That data set is analyzed in Bagano# (1999),
which may be consulted for background line identifications
and other qualitative information. 6 Unfortunately, those
data cannot be used directly for recent observations,
because both the background and the ACIS detector have
evolved significantly since then. Forthcoming calibration
observations with ACIS stowed and working in the full
imaging mode (as opposed to EHM), could be used for the
most recent observations.
3.2.3. Background Time Dependence
From the analysis of a large number of ACIS observa­
tions and monitoring of the rate of ACIS events rejected on
board (Grant, Bautz, & Virani 2002), we know that the
high­energy quiescent background slowly declined from
launch until around the end of 2000 and has been relatively
constant during 2001 (Fig. 6 shows this behavior in the 5--10
keV band, where the instrumental background dominates),
in anticorrelation with the solar cycle. At low energies, the
qualitative behavior is similar, but it is di#cult to tell exactly
because of the di#ering sky signal. In addition to the slow
evolution, there are small variations of the quiescent rate on
short timescales. Since the background can change between
the observations of our fields and the Moon, the Moon
background may need a correction, and the background
uncertainty must be included in the final results.
The EHM data are free from the sky signal and flares and
can be used for checking the quiescent background time
dependence, especially in the soft band most important for
the present study. From the comparison of EHM (as well as
blank­field) data in di#erent energy bands, it appears that
during the short­term quiescent background fluctuations,
such as those in Figure 6, its spectral shape does not change
significantly and only the normalization varies. This behav­
Fig. 5.---Spectra of dark­Moon (black) and EHM data (red ). This Moon spectrum is extracted in PHA channels in a special manner to be directly
comparable to the EHM data (see text). The energy scale is approximate (PHA values multiplied by 4.7 eV per channel). Apart from the faint internal
calibration source lines present in the EHM data (most notably at 5.9 and 1.5 keV), there is very good agreement between the spectra.
6 See http://asc.harvard.edu/cal, sections `` ACIS,'' `` Background,''
`` Measurement with closed mirror cover.''
74 MARKEVITCH ET AL. Vol. 583

ior enables us to account for these variations by normalizing
the model instrumental background by the ratio of rates at
high energies, e.g., 10--12 keV, where the contribution from
celestial sources and possible faint undetected flares (x 3.1.1)
is negligible.
Figure 7 shows EHM rates from the full S3 chip in the
0.5--2, 2--7, and 5--10 keV bands divided by the rates in the
PHA interval of 2500--3000 ADU (approximately 10--12
keV; PHA is preferred over energy because many science
observations use an onboard cuto# at 3000 ADU). The
period from 2001 July is shown because earlier EHM data
were collected from parts of the chip. The scatter of these
ratios is quite small; in all energy bands, the intrinsic scatter
around the mean required in addition to the Poissonian
scatter is 1.5%--2% (1 #) or less, a reduction from 3%--4%
for fluxes not normalized by the high­energy rate. One also
notices that neither the rates nor the spectral shape changed
systematically between the dates of the Moon observation
and our deep observations 3013 and 3419.
We use such high­energy rate matching in our CXB anal­
ysis by normalizing the Moon spectrum by the ratio of the
respective 2500--3000 ADU rates. For ObsIDs 3013, 3419,
930, and 869, such normalizations are 1.06, 0.97, 1.01, and
0.93 times the ratio of the Moon exposure to the respective
exposures, so this correction is, as expected, small. We
adopt a systematic uncertainty of the resulting quiescent
background normalization of 2% (1 #), as derived above.
We note that over longer periods, the spectral shape of
the detector background can change---the spectra of the
empty­field observations from 2000 and 2001 di#er by #5%
at energies where the noncosmic component dominates.
This means that the Moon spectrum with a simple normal­
ization adjustment might not be a good model for our
earlier observations 930 and 869. However, as seen below,
the soft di#use signal in those observations is so strong that
the background uncertainty does not matter.
3.3. VF Mode Background Filtering
In VF ACIS telemetry mode, the detector background
can be reduced significantly by rejecting events with signal
above the split threshold in any of the outer pixels of the
5 # 5 pixel event island, after an approximate correction for
the charge transfer ine#ciency. Details of the method can
be found in Vikhlinin (2001). 7 Figure 8 shows the e#ect of
such filtering on the spectra of the background (the dark
Moon) and real X­ray events from an extended celestial
source una#ected by photon pileup. It results in a significant
reduction of the background rate, especially at the lowest
and highest energies, while rejecting only about 2% of the
real events. The background reduction is stronger for FI
chips (not used in this work). All four of our CXB observa­
tions, as well as all Moon observations, were telemetered in
VFmode and filtered in this manner.
4. POINT SOURCES
We now detect and exclude point sources in our CXB
fields. All of the S3 chip is within 7 0 of the optical axis, and
the point­spread function (PSF) is narrow over the whole
field, so point­source detection is photon­limited and back­
ground is relatively unimportant for it. It is therefore advan­
tageous to include periods of moderately high background
if it significantly increases the exposure. For observations
3013, 3419, and 930, we applied a less restrictive light­curve
filtering (using the 0.3--10 keV band, and for 930 a higher
threshold factor of 2, instead of 1.2) to increase the expo­
sures for source detection (see Table 1).
Fig. 6.---Time dependence of the S3 chip 5--10 keV quiescent background
rate in various blank­field observations since launch (see Markevitch 2001).
Statistical errors are comparable to the symbol size. Red symbols denote
observations used in this work (ObsID 3013 is shown before the residual
flare correction).
Fig. 7.---Ratios of the S3 background rates in the 0.5--2, 2--7, and 5--
10 keV bands to the 2500--3000 ADU (#10--12 keV) rate, for EHM obser­
vations that used the full chip. Horizontal lines show average values.
7 See http://asc.harvard.edu/cal, sections `` ACIS,'' `` Background,''
`` VFmode.''
No. 1, 2003 DIFFUSE SOFT X­RAY BACKGROUND 75

In the three fields with pointlike original targets (ObsIDs
3013, 3419, and 930), we excluded r ¼ 30 00 circles around the
target from the analysis. In ObsID 869, whose original tar­
get is a galaxy, we masked that source liberally and used the
remaining 72% of the chip area.
Source detection was performed in two bands, 0.3--2 and
2--7 keV. The source candidates were identified in a standard
manner, by applying wavelet filtering to the image to reduce
statistical noise and then searching for local brightness max­
ima using the code of Vikhlinin et al. (1998). The source
positions were then refined by photon centroiding and the
source fluxes calculated within the 90% PSF--encircled
energy radii r 90 (and then divided by 0.9), using as the local
background a wavelet decomposition component contain­
ing details on the largest angular scale. For simplicity, we
set a relatively high, spatially uniform lower limit on the
source flux that corresponds approximately to 8--10 photons
anywhere in the field in each observation. This ensures that
all detected sources are real and that the background contri­
bution to the source flux is always small, even though we
may miss some obvious fainter sources near the optical axis.
As is seen below, these omissions do not a#ect our results
significantly.
The resulting cumulative source counts from the 0.3--2
keV images as a function of the unabsorbed 0.5--2 keV
source flux (assuming Galactic absorption and a power law
with a #1.4 photon index for the source spectrum, relevant
for the faintest sources) are presented in Figure 9. The figure
also shows fits to the low­flux end of the 0.5--2 keV source
counts from much deeper observations of Chandra Deep
Field North (CDF­N; Brandt et al. 2001) and South
(CDF­S; Rosati et al. 2002), which used the same assump­
tion about the average source spectrum. Our curves are in
good agreement with those results. The field­to­field di#er­
ence (a factor of #2) is also similar to that between CDF­N
and CDF­S (see also Barcons, Mateos, & Ceballos 2000),
although with such low absolute source numbers that this
result is not particularly significant.
For the interpretation of our di#use measurements, it is
useful to estimate the expected contribution of point sources
below our detection limits. Figure 10 shows the cumulative
0.3--2 keV flux of all sources above a certain flux, divided by
the total CXB signal in each observation (the total flux
minus the Moon background; this quantity is di#erent in all
observations, being higher for ObsIDs 869 and 930). We
can extrapolate these curves below our limits assuming, for
Fig. 8.---(a) Spectra of the S3 detector background (dark Moon) before (red ) and after (black) VF mode cleaning. (b) Ratio of the spectra of a bright
celestial source (without pileup) after and before VF mode cleaning. The noncosmic background is significantly reduced at low and high energies, while the
e#ect on the real X­rays is energy­independent and very small.
Fig. 9.---Cumulative numbers of excluded point sources as a function of
their 0.5--2 keV flux. Only the sources detected in the 0.3--2 keV band are
shown. Labels give ObsIDs. The curves end at our adopted lower flux cuts,
which approximately correspond to 8--10 photons from the source. For
comparison, low­flux fits to the CDF­N (Brandt et al. 2001) and CDF­S
(Rosati et al. 2002) source counts are shown.
76 MARKEVITCH ET AL. Vol. 583

example, the faint source distribution from CDF­N, as
shown in the figure. The arrows show the asymptotic limits
of those extrapolations at the zero flux. One can see that at
most 60% of the 0.3--2 keV total CXB flux in our two main
observations can be due to point sources (barring the emer­
gence of an unknown source population at fluxes below the
CDF limits or a significant change in the average source
spectrum at the lowest fluxes; such possibilities are beyond
the scope of this paper). The contribution of sources below
our detection limits is less than 10% of the total CXB flux
(and less than 20% of the unresolved flux) in all four obser­
vations; thus, the uncertainty of our extrapolation toward
lower fluxes should not a#ect any of our conclusions.
At energies above 2 keV, the accuracy of our present
measurements does not warrant a detailed analysis of the
point­source contribution. Lists of sources detected in the
0.3--2 and 2--7 keV bands were merged, and circles of radius
2r 90 around the sources (3r 90 for sources with more than 100
counts) were excluded from the spectral analysis below.
Table 1 gives the resulting solid angles after the target and
source exclusion.
5. SPECTRA AND INSTRUMENT RESPONSES
Di#use CXB spectra were extracted in PI channels from
the whole S3 chip, excluding the source regions. To extract
the instrumental background spectra from exactly the same
chip areas, we converted the dark­Moon event list into the
corresponding sky coordinate frame using the aspect track
of each CXB observation, assigning each Moon event a time
tag selected randomly from the time interval spanned by
that observation. The dark­Moon spectrum was then
extracted from the same region specified in sky coordinates
and normalized as described in x 3.2.3. All observations,
including those of the Moon, are performed at the same
focal plane temperature (#120 # C), so the Moon spectra
have the same resolution and can be directly subtracted
from the CXB spectra.
A spectral redistribution matrix (RMF) for each observa­
tion was calculated by averaging the position­dependent
matrices over the extraction region. Auxiliary response files
(ARFs) included the telescope e#ective area and CCD
quantum e#ciency averaged over the extraction region.
Regions of the masked sources were also excluded from the
response averaging (this has any e#ect only for ObsID 869,
for which a relatively large area is excluded). ARFs and
RMFs were calculated using A. Vikhlinin's tools calcarf
and calcrmf.
A recently discovered slow systematic decline of the ACIS
quantum e#ciency at low energies (Plucinsky et al. 2002)
was taken into account in the ARFs. The present calibration
accuracy of the quantum e#ciency at E ' 0:5 0:7 keV is
#10%.
6. RESIDUAL BACKGROUND EXCESS
Our first­iteration spectra of the unresolved CXB from
each observation are shown in Figure 11. For illustration,
a 90% systematic uncertainty of the quiescent back­
ground normalization of #3% is also shown. There is
one more background correction that we need to per­
form. In our deepest observation, 3419, the di#use spec­
trum above 2 keV is fully consistent with zero. In 3013,
however, there is an obvious hard excess that cannot
originate from the sky because of its unphysical spec­
trum. This is the excess discussed in x 3.1.1, where we
proposed that it is a long flare of the same kind that
a#ects the discarded time intervals in this observation,
only fainter. Indeed, as the figure shows, the spectral
shape of the hard excess is consistent with that of the
flare model derived in x 3.1.1 (while being clearly incon­
sistent with an elevated quiescent background, for exam­
ple). The model normalization was fitted to this spectrum
in the 2.5--10 keV interval.
Thus, we proceed with the above assumption. The flare
model also gives a significant contribution in the softer
band. To correct it, we add this best­fit model to the detector
background spectrum (this correction was already included
in Fig. 10). Unfortunately, by doing so we are setting the dif­
fuse flux above 2 keV in field 3013 to zero by definition, but
such a correction is consistent with expectations from the
light curve and the flare spectrum. A #14% normalization
uncertainty (1#) of this flare model is included in the error
budget for this observation.
Figure 11 also shows marginally significant excesses
above 2 keV in ObsIDs 869 and 930, which can also be
described by the flare model (for those fits, we adopted the
parameters derived for the 930 flare in x 3.1.1). However,
while the excess spectrum in 3013 does match the flare
model, the same cannot be said with certainty for these two
observations. The excesses are also comparable to the quies­
cent background uncertainty. As the figure shows, the possi­
ble flare contamination below 1 keV in these observations is
negligible and can be safely ignored in the analysis below.
However, the above discussion illustrates the di#culty of
the di#use CXB measurements at Ee2 keV using the ACIS
BI chip S3; FI chips may be better suited for that band
(x 8.4).
Fig. 10.---Contribution of the detected point sources above a certain flux
to the total 0.3--2 keV count rate (minus detector background), as a func­
tion of the source 0.5--2 keV flux. Labels give ObsIDs. Dotted lines show
extrapolations toward lower fluxes, assuming source counts as in CDF­N
(Fig. 9); arrows indicate their asymptotic limits at zero flux.
No. 1, 2003 DIFFUSE SOFT X­RAY BACKGROUND 77

In the Appendix, we summarize the above background
subtraction procedure and describe how it can be applied
for other ACIS obsevations of extended sources.
7. RESULTS
Table 2 gives total and di#use fluxes in our four CXB
fields in three energy bands. It includes both the directly
measured unresolved fluxes (that is, excluding the detected
sources) and the estimated true di#use fluxes after the small
correction for undetected sources (see x 4 and Fig. 10). The
2--7 keV part of the table omits ObsID 3013 and the di#use
fluxes for ObsIDs 869 and 930 because of the residual back­
ground flare uncertainty (see x 6) that a#ects this band most
significantly. The errors on these fluxes include statistical
errors of the data and the Moon background and systematic
uncertainties of the background. For ease of comparison
with earlier results, Table 2 also gives the normalization of a
power­law fit to the total spectra in the 1--7 keV band, with
photon index fixed at #1.4 (a slope consistent with all four
spectra, although only marginally so for field 930 because of
a strong soft excess), assuming Galactic absorption. The
average of the four fields is 10:7 # 0:9 photons s #1 cm #2
keV #1 sr #1 . No attempt was made to correct our `` total ''
fluxes for the missing very bright sources that we were
unlikely to encounter because of the small field of view (cov­
ering a total of 0.07 deg 2 ), so these fluxes may not be repre­
sentative of the sky average. The brightest sources found in
our fields have unabsorbed 0.5--2 keV fluxes in the range
1 5
Ï ÷ # 10 #14 ergs s #1 cm #2 .
Fig. 11.---First­iteration spectra of the di#use component for the four fields. ObsIDs are marked in each panel. For ObsIDs 3013, 869, and 930, models of
the residual flares, whose normalizations were fitted in the 2.5--10 keV band, are shown as red histograms (see text). The soft flux in fields 869 and 930 is above
the limit of the plots. Green bars illustrate the 90% quiescent background normalization uncertainty of #3%. For 3013, the flare component is obvious and well
described by the model; it is subtracted as an additional background component. For 869 and 930, it is marginally consistent with 0, given the background
uncertainty. In all observations, the (possible) flare components are small compared to the di#use signal in the soft band.
78 MARKEVITCH ET AL. Vol. 583

Figure 12 shows relative contributions of di#erent back­
ground components---instrumental, cosmic di#use, and
point sources---in our cleanest observation, 3419, which
also has the lowest soft di#use signal. Figure 13 shows the
same spectra of the di#use and total (including point
sources) CXB after the instrumental background subtrac­
tion. Below 1 keV, a bright di#use component dominates
the CXB spectrum. In the 1--2 keV band, point sources
dominate, continuing into the 2--5 keV band, where the dif­
fuse component disappears.
Figure 14 shows the final di#use spectra in the 0.25--1.2
keV band, with field 3013 corrected for the residual flare
(see x 6). All exhibit a ubiquitous O vii feature around 570
eV (the He­like K# blend), with possible contributions from
K# at 665 eV and the O viii Ly# line at 654 eV (these com­
ponents cannot be resolved with the present statistics). The
TABLE 2
Wideband Fluxes
Flux ObsID 3013 ObsID 3419 ObsID 869 ObsID 930
0.3--1 keV a
Total ....................... 4.3 # 0.2 2.2 # 0.2 7.0 # 0.2 7.8 # 0.2
Unresolved ............. 2.7 # 0.2 1.6 # 0.2 6.7 # 0.2 7.4 # 0.2
Di#use b ................... 2.4 # 0.2 1.4 # 0.2 6.4 # 0.2 7.2 # 0.2
1--2 keV a
Total ....................... 1.1 # 0.2 1.3 # 0.2 1.0 # 0.2 1.5 # 0.2
Unresolved ............. 0.3 # 0.2 0.6 # 0.15 0.7 # 0.2 1.2 # 0.2
Di#use b ................... 0.1 # 0.2 0.3 # 0.15 0.4 # 0.2 1.0 # 0.2
2--7 keV a
Total ....................... . . . 4.0 # 1.5 4.7 # 1.8 4.5 # 1.5
Unresolved ............. . . . 1.6 # 1.6 . . . . . .
Di#use b ................... . . . 0.8 # 1.6 . . . . . .
Power Law c
Total ....................... 8.9 # 1.6 12.1 # 1.9 8.8 # 1.6 12.8 # 1.7
a In units of 10 #15 ergs s #1 cm #2 arcmin #2 , uncorrected for absorption.
b Unresolved flux minus estimated contribution of undetected sources.
c Normalization of a power­law fit to the 1--7 keV total spectrum, corrected for
absorption. Photon index is fixed at 1.4; E ¼ 1 keV. Units are photons s #1 cm #2 keV #1
sr #1 .
Fig. 12.---Spectra of ObsID 3419 (our field with the lowest di#use signal) and the Moon background. Black crosses show the Moon renormalized to match
the 10--12 keV rate of 3419 (see x 3.2.3). Red crosses show the total spectrum (excluding the target), and blue squares show the di#use component after the
exclusion of all point sources (blue error bars are similar to the red ones and are omitted for clarity).
No. 1, 2003 DIFFUSE SOFT X­RAY BACKGROUND 79

best­fit mean energies and fluxes of this line blend are given
in Table 3. The line brightness strongly increases in the
direction of the North Polar Spur (fields 869 and 930), and
an additional pair of bright features (around 730 and 820
eV, primarily from Fe xvii) emerges from the brighter part
of the Spur (field 930). The lines are fitted well with a simple
thermal plasma model (MEKAL; Kaastra 1992) with solar
abundances, consisting of either one component (for the
two fields o# the Spur) or a two­temperature mixture (the
Spur fields). The model parameters are given in Figure 14;
the temperature range is 0.1--0.4 keV, and the best­fit
absorption column is lower than the full Galactic value for
all four observations. For ease of comparison, the best­fit
models from Figure 14 are plotted together in Figure 15,
taking away the time dependence of the ACIS e#ciency.
There is also an apparent linelike feature at E # 300 eV,
seen in all four spectra. Its brightness appears to change
together with that of the O lines, and it is not present in the
Moon spectrum (Fig. 12), which suggests that it is real.
However, because of the present calibration uncertainties at
the lowest energies, it is di#cult to quantify its flux and even
assess its reality. It should also be kept in mind that the same
Moon spectrum with large statistical uncertainties was used
as the background for all four observations.
8. DISCUSSION
8.1. Comparison with Earlier Results
We can compare our total (di#use+sources) CXB
results with the recent XMM work (Lumb et al. 2002).
Since the brightest sources in our four fields have fluxes
similar to the lowest fluxes of the excluded sources in
Lumb et al. [ 1 2
Ï ÷ # 10 #14 ergs s #1 cm #2 in the 0.5--2
keV band], our total fluxes can be directly compared with
their values after the bright source exclusion. Starting
from high energies, normalizations of our power­law fits
in the 1--7 keV band given in Table 2 are in agreement
with the XMM average of 8.4 photons s #1 cm #2 keV #1
sr #1 at 1 keV, if we exclude our bright field 930. In the
narrower 1--2 keV band, our fluxes (again, except 930)
are in good agreement with that of Lumb et al.,
1:0 # 10 #15 ergs s #1 cm #2 arcmin #2 , calculated from their
best­fit model. Our 2--7 keV fluxes also agree with
3:2 # 10 #15 ergs s #1 cm #2 arcmin #2 , converted from their
2--10 keV flux (we note here that these three values in
Table 2 are not entirely statistically independent, because
the error is dominated by the same Moon data set). Our
uncertainties in the 2--7 keV band are large; future ways
to reduce them are described in x 8.4.
Our 1--7 keV power­law normalizations also agree with
9--11 photons s #1 cm #2 keV #1 sr #1 at 1 keV, derived by
Miyaji et al. (1998) from ASCA GIS data for two high
Galactic latitude fields, and with 11.7 photons s #1 cm #2
keV #1 sr #1 , derived by Vecchi et al. (1999) using BeppoSAX.
Note that those studies cover much greater solid angles and
therefore are more likely to include rare, bright sources, so
this comparison is only approximate.
At E < 1 keV, the scatter between our fields (a factor of
5) is higher than that in Lumb et al. (2002; a factor of 2--3),
but this, of course, is because of our specific selection of
fields spanning a range of ROSAT fluxes.
At the primary O­line energy, the CXB is dominated by
the di#use component (Fig. 13). Thus, we can compare our
line fluxes with those recently derived, in a microcalorimet­
ric experiment by McCammon et al. (2002), from a 1 sr area
mostly away from bright Galactic features. They report an
average flux in the O vii+O viii lines of 5:4 # 0:8
Ï ÷ # 10 #7
photons s #1 cm #2 arcmin #2 . This is in the range of our val­
ues for the o#­Spur observations 3013 and 3419 given in
Table 3. McCammon et al. (2002) also observed lines at
lower energies, some of which may explain our 300 eV fea­
ture (if it is real). Oxygen line fluxes derived from earlier
Fig. 13.---Spectra of ObsID 3419 after the detector background subtrac­
tion. Red shows the total spectrum and black shows the di#use component
from Fig. 12. Errors are dominated by the low statistics of the Moon. Green
bars illustrate a 90% systematic uncertainty of the quiescent background
normalization of #3%. Above 2 keV, the di#use component is consistent
with zero.
TABLE 3
Oxygen Line and Continuum Fluxes
ObsID
O­Line Energy
(eV)
O­Line Flux a
(#10 #7 photons s #1 cm #2 arcmin #2 )
0.3--1 keV Continuum b
(#10 #15 ergs s #1 cm #2 arcmin #2 )
3013 ................. 570 # 10 9.8 # 1.5 3.2 # 0.5
3419 ................. 580 # 14 5.6 # 1.5 3.5 # 1.0
869 ................... 590 # 6 15 # 2 . . .
930 ................... 585 # 8 12 # 2 . . .
a Uncorrected for absorption.
b Flux excluding O­line, corrected for absorption assuming full Galactic column.
80 MARKEVITCH ET AL. Vol. 583

experiments (see, e.g., Inoue et al. 1979; Gendreau et al.
1995) are also within our range. Our results show, however,
that even for these high Galactic latitude areas away from
bright Galactic features, the line brightness varies signifi­
cantly from field to field. The general shape of the North
Polar Spur spectrum in our field 930 is consistent with that
reported in earlier works (e.g., Schnopper et al. 1982;
Rocchia et al. 1984; Warwick 2002).
8.2. Extragalactic Di#use Component
The di#use flux in fields 869 and 930 is obviously domi­
nated by the North Polar Spur. The O blend in 3013 and
3419 probably has a Galactic or local origin as well (extra­
galactic sources with redshifts greater than 0.05 are excluded
by our measured line energy; furthermore, the high­
resolution spectrum of McCammon et al. 2002 excludes any
redshift), although we cannot exclude, for example, its
Local Group origin. Thus, the continuum component in the
two low­brightness fields can give an approximate upper
limit on the flux from the vast quantities of the putative
warm intergalactic gas (see, e.g., Cen & Ostriker 1999) that
should emit a mixture of lines and continua from di#erent
redshifts. For a conservatively high estimate of this contin­
uum component, we fit the 0.3--1 keV di#use spectra by a
power­law model plus the line, applying the full Galactic
absorbing column.
The resulting unabsorbed continuum fluxes are given in
Table 3; they correspond to the spectral density of
2:4 2:5
Ï ÷ # 10 #15 ergs s #1 cm #2 arcmin #2 keV #1 at E ¼ 0:7
keV. This is well above the typical theoretical predictions,
which range between 0:3 # 10 #15 and 10 #15 ergs s #1 cm #2
arcmin #2 keV #1 (see, e.g., Cen & Ostriker 1999; Phillips,
Ostriker, & Cen 2001, Voit & Bryan 2001; but see Bryan &
Voit 2001 for a higher predicted flux from simulations with­
out the inclusion of cooling and preheating). Note that the
Fig. 14.---Final soft di#use spectra. ObsIDs are marked in each panel. For 3013, the residual flare is subtracted (see Fig. 11). Histograms show simple
one­ or two­temperature thermal plasma fits, with parameters shown in the panels.
No. 1, 2003 DIFFUSE SOFT X­RAY BACKGROUND 81

predicted average brightness values from the published sim­
ulations are not directly comparable to our result, because
they are dominated by nearby galaxy groups and clusters
that are easily detected and excluded from our and other
CXBmeasurements.
Thus, our crude estimate does not constrain the warm
intergalactic gas models. The constraint may be improved
in the future by better modeling and subtraction of the
Galactic emission and spatial fluctuation analysis (as in,
e.g., Kuntz, Snowden, & Mushotzky 2001); however, since
the Galaxy dominates at these energies, such constraints will
necessarily be model­dependent.
8.3. Origin of Line Emission
The observed energies of the spectral lines suggest local
(in the Local Group, Galaxy, or our immediate vicinity) ori­
gin of the dominant fraction of the soft di#use CXB. Exten­
sive literature exists that models its various components
under the well­justified assumption of their thermal plasma
origin (see, e.g., Kuntz & Snowden 2000 and references
therein). Leaving such modeling for future work, here we
mention an interesting alternative possibility.
It is likely that a significant fraction of the line flux comes
from charge exchange (CX) between highly charged ions in
the solar wind (primarily bare and hydrogenic O and C) and
neutral gas occurring throughout the heliosphere and in the
geocorona. Most of the flux that would be observed from
heliospheric CX (with H and He) originates within a few
tens of AU of the Sun. Geocoronal emission arises where
residual atmospheric H is exposed to the solar wind, at dis­
tances of the order of 10 Earth radii.
In the CX process, a collision between a solar wind ion
and a neutral atom leads to the transfer of an electron from
the neutral species to a high­n energy level in the ion, which
then decays and emits an X ray. Dennerl, Englhauser, &
Tru ˜ mper (1997) and Cox (1998) were the first to suggest that
these photons might contribute to the CXB, and Cravens
(2000) estimated that heliospheric emission might account
for roughly half of the observed soft CXB. Cravens,
Robertson, & Snowden (2001) also argued that the excess
time­variable di#use flux often observed by ROSAT was
due to fluctuations in heliospheric and especially geocoronal
CX emission, caused by `` gusts '' in the solar wind.
In the Chandra Moon observations, the heliospheric com­
ponent will be blocked, but geocoronal emission should be
present. We estimate, however, that the typical intensity of
that signal, about a few times 10 #8 photons s #1 cm #2
arcmin #2 in O K# (the strongest line), is smaller than the
statistical uncertainties in our measurement. Note also that
the geocoronal signature is not likely to be present in our
net CXB spectra, since it is subtracted as part of the Moon
spectrum. Heliospheric CX emission should be stronger and
less time­variable than geocoronal emission and should be
present in our spectra. We have constructed a numerical
model, based on Cravens' (2000) work and similar to that
described in Wargelin & Drake (2001, 2002), that predicts
that roughly half the flux in the O line(s) in fields 3013 and
3419 may come from heliospheric CX. Those observations
were at low ecliptic latitude, within the `` slow '' and more
highly ionized solar wind. CX flux in fields 869 and 930 is
expected to be much lower, because those observations
looked through the `` fast '' solar wind, which has a much
smaller fraction of bare and H­like O ions (von Steiger et al.
2000). Observations of a sample of fields selected specifically
to test this possibility are required for a more quantitative
analysis, which is forthcoming (B. J. Wargelin et al. 2003, in
preparation).
8.4. Future Work
For further CXB studies with Chandra, it is useful to look
into the error budget of the present results. The errors in the
1--7 keV band are dominated by the detector background
uncertainty: the statistical error of the short Moon data set,
the uncertainty on its normalization, and the possible resid­
ual flare component. Forthcoming calibration observations
with ACIS stowed but working in the full imaging mode
should take care of the first component. The other two point
toward the use of FI chips for studies in this band. The scat­
ter of the quiescent background normalization probably
cannot be reduced below our 2%--3% estimate; however, the
FI detector background itself is lower by a factor of 2--3,
depending on the energy. The FI chips also are much less
a#ected by the background flares. An ACIS­I study of the
CXB will be presented in a forthcoming paper (S. Virani et
al. 2003, in preparation).
9. SUMMARY
We have analyzed four high Galactic latitude, empty
fields observed with Chandra ACIS­S3 and, for the first
time, derived spectra of the di#use X­ray background,
directly excluding the point­source contribution. The total
(di#use and point­source) CXB brightness in all bands is in
agreement with most previous experiments. In the 0.3--1
keV band, the di#use signal varies strongly between all four
fields. In the two fields far from known bright Galactic fea­
tures, it contributes about half of the total CXB seen by
instruments with poorer angular resolution. It is dominated
Fig. 15.---Model fits for di#erent fields from Fig. 14. To be directly com­
parable, models are convolved with the same telescope response without
taking into account its time change. ObsIDs and major line energies are
marked.
82 MARKEVITCH ET AL. Vol. 583

by emission lines (most prominently, the O vii+O viii blend
at E # 580 eV, indications of which were also seen in pre­
vious experiments) and can be described by a thermal
plasma model with kT ¼ 0:1 0:2 keV. The line brightness
increases strongly, and additional lines appear, in the direc­
tions of the North Polar Spur.
At higher energies, the background is more uniform. Dif­
fuse emission is detected with high significance in the 1--2
keV band in the brighter of the two North Polar Spur fields.
In other fields at E > 1 keV, the di#use component is weak
or consistent with zero at our present accuracy (to be
improved in the forthcoming analysis of ACIS­I deep obser­
vations). Our current uncertainties for the di#use CXB---
one of the lowest surface brightness celestial objects---are
large because of the di#culty of modeling the instrumental
background in the time­variable particle environment
encountered in high­altitude orbits like Chandra's.
The results presented here are made possible by the suc­
cessful e#ort of the entire Chandra team to launch and oper­
ate the observatory. We thank the referee for helpful
comments and suggestions. Partial support was provided by
NASA contract NAS 8­39073, grant NAG 5­9217, and the
Smithsonian Institution.
APPENDIX
ACIS BACKGROUND SUBTRACTION
For the ACIS analysis of extended objects, the background can be modeled and subtracted as we did in x 3. In addition to
the 2001 July--September dark­Moon data and the forthcoming observations with ACIS stowed that contain only the detector
background, large data sets combining a number of high Galactic latitude, relatively source­free fields are available (see
Markevitch 2001 8 for details). The latter data are cleaned of flares and point sources, but include the cosmic di#use emission,
so they are relevant when this emission is part of the background, as opposed to signal. Separate blank­sky data sets are
assembled for three ACIS temperatures (which determine the spectral resolution) and for di#erent time periods for the present
#120 # C temperature, to track the background evolution.
The following steps summarize our background subtraction procedure as it could be applied in a typical extended source
analysis, using the combined blank­field data sets. In anomalous cases, this procedure may need more than one iteration.
1. To identify flares, we select a region of the chip(s) free of bright source emission and use it to create a light curve with time
bins large enough to be able to detect rate changes by a factor of 1.2 (the fiducial factor used in creation of the blank­sky data
sets). Using the relevant background data set (determined by the observation date), we evaluate the nominal count rate in the
selected region and reject all intervals with the rate above a factor of 1.2 of this expected rate (usually, the flares are well defined
in time and the exact threshold is not important). BI and FI chips should be cleaned separately, because BI chips are much
more prone to flares. Given the spectra of the flares, for the BI chips, a 2.5--7 keV (or 2.5--6 keV for S1) energy band should be
used, while for the FI chips, the full 0.3--12 keV band can be used.
If the target is in S3 and covers the whole chip, chip S1 can be used for flare detection. Despite the di#erent quiescent
backgrounds, S1 and S3 show a very similar response to flares, including the flare timing, spectra, and intensities. If S1 is not
available either, then faint, nonobvious flares may escape detection and significantly a#ect the results, because flare and
quiescent background components have very di#erent spectral shapes (see, e.g., a discussion in Markevitch 2002 and x 3.1.1).
In such cases, one should at least check the validity of the results by trying to fit a flare model to the spectrum above 2 keV
extracted from low­brightness target areas (as in x 6). That spectral model applies only to the soft flare species that does not
a#ect the FI chips.
If there is a faint, long flare in the observation whose proper exclusion results in too little useful exposure left, and if there
are source­free areas in the chip of the same type, one can try to model the flare contribution using those areas (as was
attempted in, e.g., Markevitch et al. 2002 and Markevitch 2002). However, the flare flux is not spatially uniform; its spatial
distribution is under investigation.
2. After the flare exclusion, the background data set can be normalized by the ratio of high­energy rates, e.g., in the PHA
interval of 2500--3000 ADU. This ratio should be within #10% of the exposure ratio. Step 1 may need to be repeated with the
corrected nominal rate if this correction is large. A 90% systematic uncertainty on the background normalization derived in
this manner is about 3% (x 3.2.3); however, if any residual faint flare is suspected, it should be taken into account in the final
uncertainty.
3. If the observation was performed in VF mode, and if a VF mode background data set is available for that time period,
additional background filtering (Vikhlinin 2001) 9 can be applied.
4. As seen from our results (e.g., Table 2), the soft (Ed1 keV) CXB component varies strongly from field to field even at
high Galactic latitudes, so the blank­sky background may not have the correct soft spectrum. One can, for example, compare
the RASS R4--R5 flux (Snowden et al. 1997) for a given observation to the average R4--R5 flux for the background data set to
determine if a correction is needed (for low­b fields, it may be needed even if those wideband fluxes are similar, since the soft
spectra may still be very di#erent). If so, one can try to use regions of the field of view free of sources (not necessarily in the
same chip) to model the sky soft excess or deficiency with respect to the blank­sky background (e.g., by the simple models used
in this paper) and subtract this model from the regions of interest after the proper vignetting correction (as in, e.g., Markevitch
& Vikhlinin 2001 and Markevitch 2002). An alternative is to use the dark­Moon data or the ACIS stowed data (if the
observation date is su#ciently close to those data sets) and model away the whole di#use CXB component.
8 See http://asc.harvard.edu/cal, sections `` ACIS,'' `` Background,'' `` General discussion.''
9 See http://asc.harvard.edu/cal, sections `` ACIS,'' `` Background,'' `` VFmode.''
No. 1, 2003 DIFFUSE SOFT X­RAY BACKGROUND 83

REFERENCES
Bagano#, F. 1999, CXCACIS background memo (Cambridge: CXC)
Barcons, X., Mateos, S., &Ceballos, M. T. 2000, MNRAS, 316, L13
Biller, B., Plucinsky, P., & Edgar, D. 2002, CXC ACIS background memo
(Cambridge: CXC)
Brandt, W. N., et al. 2001, AJ, 122, 2810
Bryan, G. L., &Voit, G. M. 2001, ApJ, 556, 590
Bunner, A. N., Coleman, P. L., Kraushaar, W. L., McCammon, D.,
Palmieri, T. M., Shilepsky, A., &Ulmer, M. 1969, Nature, 223, 1222
Cen, R., &Ostriker, J. P. 1999, ApJ, 514, 1
Cox, D. P. 1998, in IAU Colloq. 166, The Local Bubble and Beyond,
ed. D. Breitschwerdt, M. J. Freyberg, & J. Tru ˜ mper (LNP 506; Berlin:
Springer), 121
Cravens, T. E. 2000, ApJ, 532, L153
Cravens, T. E., Robertson, I. P., & Snowden, S. L. 2001, J. Geophys. Res.,
106, 24883
De Luca, A., &Molendi, S. 2002, in New Visions of the X­Ray Universe in
the XMM­Newton and Chandra Era (ESA SP­488; Noordwijk: ESA), in
press (astro­ph/0202480)
Dennerl, K., Englhauser, J., & Tru ˜ mper, J. 1997, Science, 277, 1625
Egger, R. J., &Aschenbach, B. 1995, A&A, 294, L25
Freyberg, M. J. 1998, in IAU Colloq. 166, The Local Bubble and Beyond,
ed. D. Breitschwerdt, M. J. Freyberg, & J. Tru ˜ mper (LNP 506; Berlin:
Springer), 113
Garmire, G. P., Nousek, J. A., Apparao, K. M. V., Burrows, D. N., Fink,
R. L., &Kraft, R. P. 1992, ApJ, 399, 694
Gendreau, K. C., et al. 1995, PASJ, 47, L5
Giacconi, R., Gursky, H., Paolini, F. R., & Rossi, B. B. 1962, Phys. Rev.
Lett., 9, 439
Grant, C., Bautz, M., & Virani, S. 2002, in ASP Conf. Ser. 262, High­
Energy Universe at Sharp Focus: Chandra Science, ed. E. M. Schlegel &
S. D. Vrtilek (San Francisco: ASP), 401
Inoue, H., Koyama, K., Matsuoka, M., Ohashi, T., Tanaka, Y., &
Tsunemi, H. 1979, ApJ, 227, L85
Kaastra, J. S. 1992, An X­Ray Spectral Code for Optically Thin Plasmas
(Internal SRON­Leiden Rep., Version 2; Leiden: SRON)
Kamata, Y., Takeshima, T., Okada, T., & Terada, K. 1999, Adv. Space
Res., 23, 1829
Kuntz, K. D., & Snowden, S. L. 2000, ApJ, 543, 195
Kuntz, K. D., Snowden, S. L., &Mushotzky, R. F. 2001, ApJ, 548, L119
Lumb, D. H., Warwick, R. S., Page, M., &DeLuca, A. 2002, A&A, 389, 93
Markevitch, M. 2001, CXCACIS background memo (Cambridge: CXC)
Markevitch, M. 2002, preprint (astro­ph/0205333)
Markevitch, M., Gonzalez, A. H., David, L., Vikhlinin, A., Murray, S.,
Forman, W., Jones, C., &Tucker, W. 2002, ApJ, 567, L27
Markevitch, M., &Vikhlinin, A. 2001, ApJ, 563, 95
Marshall, F. E., Boldt, E. A., Holt, S. S., Miller, R. B., Mushotzky, R. F.,
Rose, L. A., Rothschild, R. E., & Serlemitsos, P. J. 1980, ApJ, 235, 4
Marshall, F. J., &Clark, G. W. 1984, ApJ, 287, 633
McCammon, D., Burrows, D. N., Sanders, W. T., & Kraushaar, W. L.
1983, ApJ, 269, 107
McCammon, D., & Sanders, W. T. 1990, ARA&A, 28, 657
McCammon, D., et al. 2002, ApJ, 576, 188
Mendenhall, J. A., &Burrows, D. N. 2001, ApJ, 563, 716
Miyaji, T., Ishisaki, Y., Ogasaka, Y., Ueda, Y., Freyberg, M. J., Hasinger,
G., &Tanaka, Y. 1998, A&A, 334, L13
Mushotzky, R. F., Cowie, L. L., Barger, A. J., & Arnaud, K. A. 2000,
Nature, 404, 459
Phillips, L. A., Ostriker, J. P., &Cen, R. 2001, ApJ, 554, L9
Plucinsky, P. P., &Virani, S. N. 2000, Proc. SPIE, 4012, 681
Plucinsky, P. P., et al. 2002, Proc. SPIE, in press
Rocchia, R., Arnaud, M., Blondel, C., Cheron, C., Christy, J. C.,
Rothenflug, R., Schnopper, H. W., &Delvaille, J. P. 1984, A&A, 130, 53
Rosati, P., et al. 2002, ApJ, 566, 667
Sanders, W. T., Edgar, R. J., Kraushaar, W. L., McCammon, D., &
Morgenthaler, J. P. 2001, ApJ, 554, 694
Schmitt, J. H. M. M., Snowden, S. L., Aschenbach, B., Hasinger, G.,
Pfe#ermann, E., Predehl, P., & Tru ˜ mper, J. 1991, Nature, 349, 583
Schnopper, H. W., et al. 1982, ApJ, 253, 131
Snowden, S. L., et al. 1995, ApJ, 454, 643
---------. 1997, ApJ, 485, 125
Vecchi, A., Molendi, S., Guainazzi, M., Fiore, F., & Parmar, A. N. 1999,
A&A, 349, L73
Vikhlinin, A. 2001, CXCACIS background memo (Cambridge: CXC)
Vikhlinin, A., McNamara, B. R., Forman, W., Jones, C., Quintana, H., &
Hornstrup, A. 1998, ApJ, 502, 558
Voit, G. M., &Bryan, G. L. 2001, ApJ, 551, L139
von Steiger, R., et al. 2000, J. Geophys. Res., 105, 27217
Wargelin, B. J., &Drake, J. J. 2001, ApJ, 546, L57
---------. 2002, ApJ, 578, 503
Warwick, R. S. 2002, in New Visions of the X­Ray Universe in the XMM­
Newton and Chandra Era (ESA SP­488; Noordwijk: ESA), in press
(astro­ph/0203333)
84 MARKEVITCH ET AL.