Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~bradw/cv/papers/darkmoon.pdf Äàòà èçìåíåíèÿ: Mon May 17 21:51:52 2004 Äàòà èíäåêñèðîâàíèÿ: Tue Oct 2 07:05:04 2012 Êîäèðîâêà: ISO8859-5 Ïîèñêîâûå ñëîâà: moon

The Astrophysical Journal, 607:596 - 610, 2004 May 20
# 2004. The American Astronomical Society. All rights reserved. Printed in U.S.A.

A

CHANDRA OBSERVATIONS OF THE ``DARK'' MOON AND GEOCORONAL SOLAR WIND CHARGE TRANSFER
B. J. Wargelin, M. Markevitch,1 M. Juda, V. Kharchenko, R. Edgar, and A. Dalgarno
Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138; bwargelin@cfa.harvard.edu Received 2003 December 19; accepted 2004 February 3

ABSTRACT We have analyzed data from two sets of calibration observations of the Moon made by the Chandra X-Ray Observatory. In addition to obtaining a spectrum of the bright side that shows several distinct fluorescence lines, we also clearly detect time-variable soft X-ray emission, primarily O vii K and O viii Ly , when viewing the optically dark side. The apparent dark-side brightness varied in time by at least an order of magnitude, up to $2 Ò 10Ð6 photons sÐ1 arcminÐ2 cmÐ2 between 500 and 900 eV, which is comparable to the typical 3 keV- band 4 background emission measured in the ROSAT All-Sky Survey. The spectrum is also very similar to background spectra recorded by Chandra in low- or moderate-brightness regions of the sky. Over a decade ago, ROSAT also detected soft X-rays from the dark side of the Moon, which were tentatively ascribed to continuum emission from energetic solar wind electrons impacting the lunar surface. The Chandra observations, however, with their better spectral resolution, combined with contemporaneous measurements of solar wind parameters, strongly favor charge transfer between highly charged solar wind ions and neutral hydrogen in the Earth's geocorona as the mechanism for this emission. We present a theoretical model of geocoronal emission and show that predicted spectra and intensities match the Chandra observations very well. We also model the closely related process of heliospheric charge transfer and estimate that the total charge transfer flux observed from Earth amounts to a significant fraction of the soft X-ray background, particularly in the ROSAT 3 keV band. 4 Subject headings: atomic processes -- Moon -- solar wind -- X-rays: diffuse background -- X-rays: general On-line material: color figures

1. INTRODUCTION As reported by Schmitt et al. (1991), an image of the Moon in soft X-rays (0.1 - 2 keV) was obtained by the Rontgen? Satellit (ROSAT ), using its Position Sensitive Proportional Counter (PSPC) on 1990 June 29. This striking image showed an X-ray - bright, sunlit half-circle on one side and a much dimmer but not completely dark side outlined by a brighter surrounding diffuse X-ray background. Several origins for the dark-side emission were considered, but the authors' favored explanation was continuum emission arising from solar wind electrons sweeping around to the unlit side and impacting on the lunar surface, producing thick-target bremsstrahlung. Given the very limited energy resolution of the PSPC, however, emission from multiple lines could not be ruled out. A significant problem with the bremsstrahlung model was explaining how electrons from the general direction of the Sun could produce events on the opposite side of the Moon, with a spatial distribution that was ``consistent with the telescopevignetted signal of a constant extended source'' (Schmitt et al. 1991). An elegant alternative explanation would be a source of X-ray emission between the Earth and the Moon, but at the time, no such source could be envisioned. If this source were also time-variable, it would account for the long-term enhancements (LTEs) seen by ROSAT. These occasional increases in the counting rate of the PSPC are vignetted in the same way as sky-background X-rays, indicating an external origin (Snowden et al. 1995). LTEs are distinct from the particle-induced background and are uncorrelated with the
Also Space Research Institute (IKI), Russian Academy of Sciences, Profsoyuznaya 84/32, Moscow 117810, Russia.
1

spacecraft's orientation or position (geomagnetic latitude, etc.), although Freyberg (1994) noted that the LTEs appeared to be related, by a then unknown mechanism, to geomagnetic storms and solar wind variations. The final ROSAT All-Sky Survey (RASS) diffuse background maps (Snowden et al. 1995, 1997) removed the LTEs, so far as possible, by comparing multiple observations of the same part of the sky, but any constant or slowly varying ( k 1 week) emission arising from whatever was causing the LTEs would remain. A conceptual breakthrough came with the ROSAT observation of comet Hyakutake (Lisse et al. 1996) and the suggestion by Cravens (1997) that charge transfer (CT) between the solar wind and neutral gas from the comet gave rise to the observed X-ray emission. In solar wind CT, a highly charged ion in the wind (usually oxygen or carbon) collides with neutral gas (mostly water vapor, in the case of comets) and an electron is transferred from the neutral species into an excited energy level of the wind ion, which then decays and emits an X-ray. This hypothesis has been proved by subsequent observations of comets such as C/1999 S4 (LINEAR), by Chandra (Lisse et al. 2001), and Hyakutake, by the Extreme Ultraviolet Explorer (EUVE; Krasnopolsky & Mumma 2001; see also the review by Cravens 2002), and is supported by increasingly detailed spectral models (Kharchenko et al. 2003; Kharchenko & Dalgarno 2000). A more extensive history of the evolution of the solar wind CT concept can be found in Cravens, Robertson, & Snowden (2001) and Robertson & Cravens (2003a). Citing the cometary emission model, Cox (1998) pointed out that CT must occur throughout the heliosphere as the solar wind interacts with atomic H and He within the solar system. Freyberg (1998) likewise presented ROSAT High Resolution 596


X-RAY MOON AND CHARGE TRANSFER
TABLE 1 Obse rvation Information Exposure (s) 2930 2982 2747 2998 2830 2993 3157 2223 4772 3998

597

Date 2001 Jul 26 ...................

ObsID 2469 2487 2488 2489 2490 2493 2468 3368 3370 3371

CCDs I2, I3, S2, S3 I2, I3, S2, S3 I2, I3, S2, S3 I2, I3, S2, S3 I2, I3, S2, S3 I2, I3, S2, S3 I2, I3, S1, S2, S3 I2, I3, S1, S2, S3 I2, I3, S1, S2, S3 I2, I3, S1, S2, S3

Chandra Time 112,500,070 112,503,320 112,507,858 112,510,900 112,515,450 112,518,500 117,529,483 117,532,850 117,536,678 117,541,880 - - - - - - - - - - 112,503,000 112,506,302 112,510,605 112,513,898 112,518,280 112,521,493 117,532,640 117,535,073 117,541,450 117,545,878

2001 Sep 22 ..................

Notes.--Chandra time 0 corresponds to the beginning of 1998. Observations were July 26 02:01:10 - 07:58:13 UT and September 22 07:04:43 - 11:37:58 UT.

Imager data that provided some evidence for a correlation between increases in the apparent intensity of comet Hyakutake and in the detector background; he further suggested that this could be caused by CT of the solar wind with the Earth's atmosphere. A rough, broadband, quantitative analysis by Cravens (2000) predicted that heliospheric emission, along with CT between the solar wind and neutral H in the Earth's tenuous outer atmosphere (geocorona), accounts for up to half of the observed soft X-ray background (SXRB). Intriguingly, results from the Wisconsin Soft X-Ray Background sky survey (McCammon & Sanders 1990) and RASS observations (Snowden et al. 1995) indicate that roughly half of the 1 keV 4 background comes from a ``local hot plasma.'' Cravens et al. (2001) also modeled how variations in solar wind density and speed should affect heliospheric and geocoronal CT emission observed at Earth and found strong correlations between the measured solar wind proton flux and temporal variations in the ROSAT counting rate. In this paper we present definitive spectral evidence for geocoronal CT X-ray emission, obtained in Chandra observations of the Moon. Data analysis is discussed in x 2, and results are presented in x 3. As we show in x 4, model predictions of geocoronal CT agree very well with the observed Chandra spectra. In x 5, we estimate the level of heliospheric CT emission, discuss the overall contribution of CT emission to the SXRB, and assess the observational prospects for improving our understanding of this subtle but ubiquitous source of X-rays. 2. THE DATA The Moon was observed with the Chandra Advanced CCD for Imaging Spectroscopy (ACIS) in two series of calibration observations, on 2001 July 26 and September 22, totaling 17.5 and 14 ks, respectively (see Table 1). The intention was to determine the intrinsic ACIS detector background by using the Moon to block all cosmic X-ray emission. Four of the ACIS CCDs were used in July (I2 and I3 from the ACIS Imaging array and S2 and S3 from the ACIS Spectroscopy array), and the S1 chip was added in September. Two of the chips, S1 and S3, are back-illuminated (BI) and have better quantum efficiency (QE) at low energies than the front-illuminated (FI) chips, I2, I3, and S2. As can be seen in Figure 1, however, the BI chips have higher intrinsic background than the FI chips and also poorer energy resolution. Telemetry limits prevented the operation of more CCDs when using ACIS Very Faint

(VF) mode, which was desired because of its particle background rejection utility.2 The ACIS detector background was also calibrated in an alternative manner using event histogram mode (EHM).3 The July Moon and EHM spectra from the S3 ship were compared by Markevitch et al. (2003) and showed good agreement, although there was a noticeable but statistically marginal excess near 600 eV in the Moon data. The dark-Moon versus EHM comparison strongly supports the assumption that the high-energy particle background inside the detector housing where EHM data are collected is the same as in the focal position. With that in mind, new calibration measurements were made on 2002 September 3, with ACIS operating with its standard imaging setup in a ``stowed'' position, where it was both shielded from the sky and removed from the radioactive calibration source in its normal off-duty position.4 These data (observation ID [ObsID] 62850, 53 ks) provide the best available calibration of the intrinsic detector background and are used in the analysis that follows. 2.1. Data Preparation All data were processed to level 1 using Chandra Interactive Analysis of Observations (CIAO) software, Pipeline release 6.3.1, with bad-pixel filtering. Start and stop times for each observation were chosen to exclude spacecraft maneuvers. Apart from the inclusion of charge transfer inefficiency (CTI) corrections (see below) and a more aggressive exclusion of any possibly questionable data, our data processing is essentially the same as that described by Markevitch et al. (2003), who limited their analysis to the July S3 data. Here we use data from all chips during both the July and September observations and include data from periods of partial darkMoon coverage by using spatial filtering (see x 2.2). Although ACIS has thinly aluminized filters to limit optical contamination, the sunlit side of the Moon was so bright that an excess bias signal was sometimes produced in the CCDs, particularly in the I2 and I3 chips that imaged that region during July. As described in Markevitch et al. (2003),
2 See A. Vikhlinin (2001), at http://asc.harvard.edu /cal; follow links ``ACIS,'' ``Background,'' and ``Reducing ACIS Quiescent Background.'' 3 See B. Biller, P. Plucinsky, & R. Edgar (2002) at http://asc.harvard.edu / cal; follow links ``ACIS,'' ``Background,'' and ``Using Chandra Level 0 Event Histogram Files.'' 4 See M. Markevitch (2002) at http://asc.harvard.edu /cal; follow links ``ACIS,'' ``Background,'' and ``Particle Background Observations.''


598

WARGELIN ET AL.

Vol. 607

a bias correction to each event's pulse-height amplitude was calculated by averaging the 16 lowest signal pixels of the 5 Ò 5 pixel VF mode event island. ObsID 2469 suffered by far the most optical contamination, so that all data from the I2 chip during that observation had to be discarded. The I3 chip also had significant contamination, but it was small enough to be largely corrected. As explained in x 2.3, however, I3 data from that observation were also excluded as a precaution. Two other July observations required exclusion of some I2 data because of optical leaks (930 s in ObsID 2490 and 1140 s in ObsID 2493), but in all the remaining data the typical energy correction was no more than a few eV, which is insignificant for our purposes. To improve the effective energy resolution, we applied standard CTI corrections, as implemented in the CIAO tool acis_ process_events, to data from the FI chips. CTI is much less of a problem in the BI chips, S1 and S3, and no corrections were made to those chips' data. Finally, VF-mode filtering was applied to all the data5 to reduce the particleinduced detector background. The ACIS stowed background data were treated in the same way, except that no opticalcontamination corrections were required. 2.2. Spatial Extractions As seen in Figure 1, Chandra pointed at a fixed location on the sky during each observation as the Moon drifted across the field of view. Using ephemerides for the Moon and Chandra's orbit, we calculated the apparent position and size of the Moon in 1 minute intervals and extracted data from its dark side, as well as from the bright side and from the unobscured cosmic X-ray background (CXB) within the field of view for comparison. Because the Moon moved by up to 1600 per minute, and to avoid X-ray ``contamination'' of data within each extraction region ( particularly spillover of bright-side or CXB photons into the dark side), we used generous buffers of 9000 from the terminator and limb for the dark-side extraction, 3000 from the terminator and 6000 from the limb for bright-side data, and 6000 from the limb for the CXB data. Chandra has a very tight point-spread function, with an on-axis encircled energy fraction of nearly 99% at 500 eV within 1000 ; although off-axis imaging is involved here, estimated X-ray contamination is less than $2% within our chosen extraction regions. Data from all ObsIDs were analyzed in several energy bands to look for discrete sources, but none were found, and light curves for each observation behave as would be expected for uniform emission within each extraction region. Effective exposure times (as if one CCD were fully exposed) were computed for each ObsID/chip/extraction combination by computing extraction areas for each 1 minute interval (accounting for spacecraft dither, which affects area calculations near the chip edges) and summing the area Ò time products. Results are listed in Tables 2 and 3. Because the detector background is not perfectly uniform across each chip, background data were projected onto the sky and extracted using the same regions as for the dark-Moon data. Exposure-weighted and epoch-appropriate detector response functions (RMFs and ARFs) were then created using standard CIAO threads, including the corrarf routine, which applies the ACISABS model to account for contaminant buildup on ACIS. The detector background rate varies slightly on timescales of months, so we renormalized the background
See A. Vikhlinin (2001), at http://asc.harvard.edu /cal; follow links ``ACIS,'' ``Background,'' and ``Reducing ACIS Quiescent Background.''
5

Fig. 1.--Moon motion across the ACIS chips during July and September observations. Green denotes the Moon position at the start of the observation and red that at the end. The illuminated portion of the Moon is shown by the crescent on the right. Dark-side gibbous extraction regions provide a 9000 buffer along the Moon limb and terminator. The S1 and S3 chips are darker because of their higher background.


No. 1, 2004

X-RAY MOON AND CHARGE TRANSFER
TABLE 2 Effe ctive Exposur e T imes: July Dark-Side Region ObsID 2469....................... 2487....................... 2488....................... 2489....................... 2490....................... 2493....................... Total .................. I2 0 1702 1204 1491 1071c 1456d 6924
b

599

Bright-Side Region S3 2924 2982 2747 2977 2714 2572 16916 I2 0 722 979 984 984 238 3503
b

I3 0 1759 1233 1375 1306 1637 7310
c

S2 2930 2812 2714 2978 2811 2862 17107

I3 0 704 965 1062 1062 847 4560
c

S2 0 0 0 0 0 0 0

S3 0 0 0 0 0 0 0

Note.--All exposure times are in chip-seconds. a Severe optical leak. All data excluded. b Unreliable energies and possible event loss from optical leak. All data excluded. c Optical leak. Time range 112,516,560 - 112,517,490 (930 s) excluded. d Optical leak. Time range 112,520,070 - 112,521,210 (1140 s) excluded.

data to match the corresponding observational data in the energy range 9.2 - 12.2 keV, where the detected signal is entirely from intrinsic background. The required adjustments were only a few percent. 2.3. Spectra As described by Markevitch et al. (2003), the BI chips, and very rarely the FI chips, often experience ``soft'' background flares because of their higher sensitivity to low-energy particles. A relatively bright flare was found in ObsID 3370, and 400 s of data were removed. Weaker flares are more common, and we judged it better to model and subtract their small effects rather than exclude large intervals of data. Soft flares have a consistent spectral shape (a power law with a highenergy cutoff ), and their intensity at all energies can be determined by integrating the excess signal (above the stowed background) in the energy range 2.5 - 7 keV, where the relative excess is most significant. We find that spectra from ObsIDs 2468 and 3368 have minor soft-flare components, and we have accounted for them in the results presented later. An essentially negligible soft-flare excess is also seen and accounted for when all the July S3 data are combined. One last complication is the effect of optical contamination on the energy calibration. Although raw energy offsets were removed from the data during the bias correction process, more subtle effects remained, mostly related to CTI. Opticalleak events partially fill the charge traps in the CCDs, thus reducing the net CTI. When standard CTI corrections were applied to the data to improve the energy resolution of the FI chips, this overcorrected and pushed energies too high for data

affected by the optical leak. The effect varies between and within chips based on optical exposure, but we can place an upper limit on it by examining the bright-Moon data, which are most affected. Figure 2 shows the spectrum of the combined July bright-Moon data, which come from the I2 and I3 chips. The K-shell fluorescence lines of O, Mg, Al, and Si are easily identified, and we find a positive offset of 50 Ö 5 eV from their true values of 525, 1254, 1487, and 1740 eV, respectively. Bright-side data from ObsID 2469 I3 were excluded because they showed an offset of roughly 80 eV, with a distorted shape for the O peak; to be conservative, we excluded the corresponding dark-side data from further consideration as well. As noted above, I2 data from ObsID 2469 were already excluded because of their much larger optical contamination. Energy offsets in the dark-side and CXB data should be much smaller, particularly for the S chips, which did not image the bright side of the Moon. Judging from the positions of weak fluorescence lines present in the detector background and the agreement of astrophysical line positions in the FI and S3 spectra with each other and theoretical models (see x 4.1), the dark-side energy errors indeed appear to be negligible. 3. RESULTS 3.1. Dark-Side Spectra X-ray spectra were created from the event files using CIAO dmextract. Data from the three FI chips (I2, I3, and S2), which have lower QE than the BI chips below 1 keV, were always

TABLE 3 Effe ctive Exposure T imes : Septe mber Dark-Side Region ObsID 2468.............. 3368.............. 3370.............. 3371.............. Total ......... I2 3157 2223 4772 3995 14147 I3 3150 2223 4770 3995 14138 S1 1026 162 1281a 265 2734 S2 2534 1364 3479 1631 9008 S3 2958 2031 3611a 1803 10403 I2 0 0 0 0 0 I3 0 0 0 0 0 CXB Region S1 1462 1680 2157a 3024 8323 S2 177 275 439 1114 2005 S3 6 0 20a 950 976

Note.--All exposure times are in chip-seconds. a Background flare in BI chips. Time range 117,538,500 - 117,538,900 (400 s) excluded.


600

WARGELIN ET AL.

Vol. 607

Fig. 2.--Spectrum of the bright side of the Moon, combining I2 and I3 data from all July observations except ObsID 2469, binned by two PI channels (29.2 eV). The dotted curve represents detector background. K-shell fluorescence lines from O, Mg, Al, and Si are shifted up by 50 eV from their true values because of residual errors when correcting for detector sensitivity to optical photons (see text). Optical contamination effects likewise cause slight mismatches in energies of intrinsic detector features such as the Au M complex (2.2 keV), Ni K (7.5 keV), and Au L (9.7 keV). There are $1300 counts in the O K line. [See the electronic edition of the Journal for a color version of this figure.]

combined in order to improve statistics. Scaled background spectra, with soft-flare corrections as needed, were created for each observed dark-side spectrum and subtracted to reveal any excess X-ray emission. Summing all the July data, apart from ObsID 2469, reveals no significant excesses in either the S3 or combined FI spectra (see top half of Fig. 3). The S3 spectrum from ObsID 2469, however, has a noticeable emission feature (more than 3 ) at $600 eV. The corresponding FI spectrum for ObsID 2469 has too few counts to be used for corroboration, as it includes only data from the S2 chip (because of the optical leaks in I2 and I3). Much stronger evidence for excess emission near 600 eV appears in the September dark-Moon spectra, in all chips (bottom half of Fig. 3). It is immediately obvious that this cannot be particle- or photon-induced O fluorescence, which would occur in a single line at 525 eV, nor is it electron-impact continuum, as posited by Schmitt et al. (1991). To assess the significance of any excesses, we selected three energy ranges for statistical study, with the a priori assumption that solar wind CT is the source of the emission (x 4). Each range (311-511, 511-711, and 716-886 eV) was chosen to extend $50 eV below and above the strongest CT lines expected within each range (x 4.1). For comparison, the 886 - 986 eV band was also studied (in which we might hope to see Ne ix K at 905 - 922 eV), along with four 200 eV wide bands from 1000 to 1800 eV. The most important results are shown in Table 4, with excesses of more than 2.5 marked by asterisks. Emission in the 511 - 711 eV range has an excess of more than 4 in three of the four S3 spectra from September and 2.3 in the other (ObsID 3370). The same pattern (the most significant excess in ObsID 2468, the least in ObsID 3370) holds for the combined FI spectra. When spectra from the three ObsIDs showing the largest excesses (2468, 3368, and 3371, henceforth referred to as the ``bright September ''

ObsIDs) were summed, the feature significance was more than 6 in both the S3 and FI spectra (see Table 4). The ratio of net counting rates for S3 and the FI chips (4:3 Ö 0:9) also matches well with the ratio of those chips' effective areas in that energy range, consistent with this being an X-ray signal from the sky. Results from the S1 chip were consistent with those from S3, but with lower significance because of the S1's much shorter dark-Moon exposure times and somewhat higher background; we do not discuss S1 dark-side results further. The 716 - 886 eV range, which we expect to contain O viii Ly and O viii Ly emission, also showed significant excesses in the S3 and combined FI spectra for the bright September ObsIDs ($2.5 and $4 , respectively), with excesses in individual ObsIDs roughly following the time pattern seen in the 511 - 711 eV band. The S3 spectrum for ObsID 3371 stands out, with a 4 excess. The same ObsID also has 2 excesses in the 886 - 986 eV range in both the S3 and FI spectra. This energy range, like the 716 - 886 eV range, contains CT emission lines ( Ne ix K at 905 - 922 eV) from an ion that is only abundant when the solar wind is especially highly ionized. As we discuss in x 4.3, there is evidence for such a situation during ObsID 3371. Above 1000 eV, no significant excesses were seen for any of the ObsID combinations listed in Table 4, with the possible exception of 1200 - 1400 eV, which had a 2.9 excess in the bright September FI spectrum (and a 1.0 excess in the S3 spectrum). Again, 3371 was the individual ObsID recording the largest excesses in the FI (1.8 ) and S3 (1.3 ) spectra. Although the evidence is not compelling, we believe that the observed excess probably represents a detection of He-like Mg xi K ($1340 eV). In the 311 - 511 eV range, where C vi Lyman emission might be detectable, only a 2.2 excess appears in the S3 data. Over the full range of O emission (511 - 886 eV), which is relevant to the discussion in x 4, the bright September S3 and FI spectra both have an excess of 7.4 , with net rates of ? 287 Ö 39îÒ 10Ð6 and ?78 Ö 11î Ò 10Ð6 counts sÐ1 arcminÐ2, respectively. The summed July S3 data have a rate of ?62 Ö 19î Ò 10Ð6 counts sÐ1 arcminÐ2 between 511 and 886 eV. If ObsID 2469 is excluded because of its obviously stronger O emission, the rest of the July S3 data have a statistically insignificant 1.7 excess, with a rate of ?34 Ö 20îÒ 10Ð6 counts sÐ1 arcminÐ2. (As an aside, we note that the S3 Moon spectrum from July, combining data from all six ObsIDs, was used as a measure of the detector background by Markevitch et al. 2003. Even with the ObsID 2469 excess, the oxygen emission in that background is much less than the sky-background emission discussed in that paper, and so the authors' results are not significantly affected.) 3.2. Comparison with CXB If the dark-side emission arises between the Earth and Moon, then it must also be present at the same level on the bright side and in the CXB beyond the Moon's limb. Unfortunately, the bright side was observed using the I2 and I3 chips and only in July, when the dark-side emission was barely detectable even with the more sensitive S3 chip. Such a weak signal would in any case be swamped by the bright-side fluorescence X-rays. Spectra of the CXB were obtained only in September, primarily by the S1 chip (see Table 3), which is back-illuminated like S3 and has similar QE. In Figure 4, which shows CXB S1


No. 1, 2004

X-RAY MOON AND CHARGE TRANSFER

601

Fig. 3.--Observed and background-subtracted spectra from the dark side of the Moon, with two-channel (29.2 eV) binning. In the July S3 data (top four panels), no excess emission is obvious, except in ObsID 2469. Features around 1750 eV (Si K) and 2150 eV (Au M) are from particle-induced fluorescence of the detector assembly. The bottom four panels show S3 and combined FI spectra from the three September ObsIDs with the strongest emission excesses. Oxygen emission from CT is clearly seen in both spectra, and energy resolution in the FI chips is sufficient that O viii Ly is largely resolved from O vii K . High-n O viii Lyman lines are also apparent in the FI spectrum, along with what is likely Mg xi K around 1340 eV. [See the electronic edition of the Journal for a color version of this figure.]

spectra from all four September observations, it is apparent that the CXB is much brighter than the dark-Moon signal. In fact, this is one of the brightest regions of the sky, with a complex spatial structure; see Table 5, which lists the centroid of each CXB extraction region and the corresponding R45 ( 3 keV- band) RASS rate.6 The typical SXRB recorded in 4 the RASS is roughly one-quarter as bright, comparable to the bright September dark-Moon brightness.
6 Background maps are available online at http://www.xray.mpe.mpg.de/ rosat/survey/sxrb/12/ass.html. An X-ray Background Tool is also available at http://heasarc.gsfc.nasa.gov/cgi-bin /Tools/xraybg/xraybg.pl. Given the limited resolution of the PSPC, data are usually divided into three energy bands: R12 (aka the 1 keV band, effectively defined on the high-energy end by the C 4 absorption edge at 0.284 keV), R45 ( 3 keV band; roughly 0.4 - 1.0 keV), and 4 R67 (1.5 keV band; roughly 1.0 - 2.0 keV).

One can also see that emission around 600 eV is strongest for ObsID 2468, which is also the ObsID that shows the most significant dark-side X-ray emission. While this correlation is suggestive, direct comparisons among the CXB spectra are not possible, because they were taken from slightly different regions of the sky (because of the Moon's motion), nor can the CXB data be adequately normalized using RASS background rates, because statistical uncertainties are too large. 4. INTERPRETATION: GEOCORONAL CHARGE TRANSFER The primary result of our analysis is that highly significant time-variable emission is seen looking toward the dark side of the Moon at energies between 500 and 900 eV. As we discuss


602

WARGELIN ET AL.
TABLE 4 Ne t E mis s io n wi th in S el e ct ed En e rg y B an d s 311 - 511 eV Counting Ratea 2 Ð1 2 Ð11 Ð5 1 Ð2 Ð36 30 22 51 54 58 26 63 Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö 19 6 12 13 9 11 7 34 18 16 31 54 37 49 28 Excess () 0.1 Ð0.2 0.1 Ð0.9 Ð0.5 0.1 Ð0.3 Ð1.1 1.7 1.4 1.7 1.0 1.5 0.5 2.2 511 - 716 eV Counting Ratea 25 3 65 76 12 26 52 126 22 43 218 256 77 247 228 Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö Ö 20 6 14* 18* 9 12 8* 39* 15 14* 33* 61* 34 60* 31* Excess () 1.2 0.5 4.5* 4.2* 1.3 2.2 6.3* 3.2* 1.4 3.0* 6.6* 4.2* 2.