J. Pradas - J. Kerp
Radioastronomisches Institut der Universität Bonn, Auf dem Hügel 71, 53121 Bonn, Germany
Received 3 March 2005 / Accepted 30 July 2005
Abstract
We present a study of the in-flight performance of the XMM-Newton EPIC MOS and pn detectors, with focus on the influence of proton flares and vignetting on the data. The very wide range in the conditions of our sample of observations, in terms of exposure length and background intensities, allows the detection of a wide range in the spectra of the proton flares, in contrast to the hard-spectrum flares proposed by Lumb et al. (2002, A&A, 389, 93)
or Read & Ponman (2003, A&A, 409, 395). We also find an up to now unreported contamination in the low energy regime (
keV) of the MOS1 observations, consisting of a significant increase in the measured intensities in two CCDs at the edges of the detector. This contamination yields in bright CCDs in the observations. Its effect must be taken into account for the study of sources detected in the affected CCDs. With respect to vignetting, we present in-flight exposure maps and we propose a method to repeat this calculation for user-definable energy bands. All the results presented here, have the goal to enable the study of very faint extended sources with XMM-Newton, like nearby galactic X-ray halos or the soft X-ray background.
Key words: X-rays: general - X-rays: diffuse background - methods: data analysis
Here we present our newly developed data reduction method for the analysis of diffuse X-rays observations with the XMM-Newton EPIC MOS and pn detectors. In order to develop this method, we made extensive use of the XMM-Newton Science Archive and carried out the in-flight background analysis for XMM-Newton with largest accumulated exposure time at present (about 3.8 Ms). In this work, we focus on the FullFrame mode, although the developed method can be straightforwardly applied to the remaining modes.
We also report on an overestimation in the measured X-ray intensity in the softest energy regime (
keV) for CCDs 2 and 5 of the EPIC-MOS 1 camera. These two CCDs occasionally show up with up to a factor of two higher X-ray background intensity than the other CCDs. In Sect. 5 we will see that about 15% of the observations in our database are affected by this effect in CCD 5. In case of CCD 2, more than 50% of the observations present contamination detectable with a first visual inspection of the data.
The structure of this paper is as follows: in Sect. 2 we present the X-ray data and its calibration. In Sect. 3, we present our new data reduction method with emphasis on the detection of proton flares (see Sect. 3.2). In Sect. 4, we explain the construction of the new background maps based on the in-flight performance of XMM-Newton. In Sect. 5, we report on the detection of bright CCDs in several observations of the EPIC-MOS 1 camera. In Sect. 6, we present our conclusions.
In order to maximize the background signal of the sample, observations with very bright point sources were excluded from our data selection. We also excluded observations with less than 8 ks exposure time. This criterion assures an expected background rate of at least
for the soft energy regime, defining a set with acceptable statistical significance. The derived value of 8 ks is based on the ROSAT all-sky survey (RASS) data, taken as a first estimation of the level of the cosmic X-ray background (XRB) towards the fields of interest. In fact, the minimal X-ray intensity for the XRB among the observations in
Tables 3 to 6 of
observed by ROSAT in the R2 energy band corresponds to
in one EPIC MOS detector (
)
with the medium filter (
expected in about 8 ks). Note, that the differences in effective area between both mirrors must be considered for this calculation.
With the two subsets presented in this section, we accumulated the largest database to the present (about 3.8 Ms) for the analysis of the XMM-Newton data calibration based on the in-flight performance of the satellite (see Sects. 3.2 and 4). The observations were calibrated with respect to the calibration database synchronized on October 24, 2002. For that purpose, we use the standard SAS 5.3.3 tasks grouped in the pipeline procedure called emchain. Further processing of the so produced calibrated event lists, was performed using a combination of SAS 5.3.3 and self-developed software (see Sect. 3).
In contrast to the RASS, there is no standard tool to perform such an absolute calibration for faint extended emission for XMM-Newton. There are some problems in the data reduction procedure that require the development of special tools to be solved. For example, Read & Ponman (2003) and Lumb et al. (2002) agree in the necessity for a reliable proton flare filtering, alternative to the method presented in the ABC Guide to XMM-Newton Data Analysis (Snowden et al. 2004) available at the web pages of the High Energy Astrophysics Science Archive Research Center, and propose slightly different methods in order to achieve this goal. Read & Ponman (2003) and Kappes et al., in prep, also deepen in the analysis of the XMM-Newton vignetting and propose methods to overcome the insufficiency of the standard data reduction tools to deal with this effect. In practice, the standard tools produce an artificial enhancement of the intensity at the edges of the detector that can be described as a tunnel effect (see Sect. 4 and Fig. 3).
Now, we present the method that we developed for the data reduction based on the in-flight performance of XMM-Newton, and which has the goal of allowing a reliable study of very faint diffuse emission. We emphasize in the effect of proton flares (Sect. 3.2) and vignetting (Sect. 4) and make use of the largest database for this kind of investigation today.
Table 1: Selected XMM-Newton energy bands.
Energy bands C1, C2 and C3 are only related to source detection and proton flare filtering. The choice of these bands is justified in the following Sect. 3.2.
![]() |
Figure 1:
Light curves of the EPIC MOS1 camera in the three C1, C2 and C3 energy bands (indicated on top of the panels) for two observations representative of the sample and indicated on the right side of the panels. The dashed lines in each panel marks the two sigma level used for the rejection of contaminated time intervals at the final iteration step (see Sect. 3.2). Mean count rates as determined by our programs are shown in the upper parts of the panels as ![]() |
Open with DEXTER |
When the imaging instruments of the satellite cross interplanetary clouds of electrically charged particles, the count rate increases by up to several orders of magnitude. The low energy protons of these charged clouds are ejected from the Sun (Marty et al. 2003) and show a broad variety of X-ray spectra (see Fig. 1), with particle energies covering the entire energy coverage of the EPIC cameras. It has been proposed (Lumb et al. 2002) that proton flares are composed of several components with different spectra and turn on times. However, Lumb et al. (2002); Read & Ponman (2003) suggest that the most important fraction of the proton flares show a hard X-ray spectrum with their dominant contribution at energies higher than
and, therefore, they focus their methods for proton flare filtering only on the hardest energy bands, corresponding approximately to the C3 band in Table 1. We now revise this suggestion by systematically extending the search for proton flares to all energy bands noted in Table 1. The very broad bands C1, C2 and C3 are used because their high signal-to-noise ratio gives very stable count rates in the phases when the detectors are not being affected by proton flares. Finally, all three bands together cover the entire EPIC energy range without any overlap.
Today, there is no method available to predict the occurrence of proton flares. Therefore, their effect can only be corrected in a post-observation data analysis. An accurate method of detecting the presence, beginning and end of a proton flare is required to keep the longest usable observation time. We developed such a temporal-filter method based on an iterative algorithm with an user definable -level. This
-level gives the minimal relative contribution of a proton flare to the total count rate necessary to flag out the corresponding time interval.
For each energy band, we compute the mean
and standard deviation
of the count rate and search for observing intervals with a rate exceeding a threshold defined by the user, typically
or
.
These bad time intervals are flagged and
and
are calculated for the remaining observing time. This iteration continues until the difference of the mean values of two consecutive iteration steps stays below the statistical uncertainty of the data (
). This "stop condition'' is generally fulfilled after less than five iterations. Then, we compute the intersection of the good time intervals obtained for the different energy bands and obtain the maximal observing time with all bands free of proton flare contamination. Then, good time intervals with a shorter duration than four minutes are rejected. Bad time intervals shorter than two minutes are not rejected. These two constrains are obtained by trial and error from a large number of observations, although they can be changed in the software in order to "fine tune'' it to specific situations. Based on our experience, we find that, in most cases, time intervals identified by the programs as good but with a shorter duration than four minutes, correspond to a decrease in the intensity of a flare between two epochs of higher intensity. However, the intensity decreased epoch is still contaminated by a flare. Time intervals identified as bad and shorter than two minutes, usually correspond to anomalous data in only a single energy band, and can be considered as good time intervals in all other energy bands. However, the here proposed interval duration constrains might be inadequate in some special cases and the corresponding parameters in the software should be changed accordingly. For an investigation of a large number of observations, like presented in this work, we can safely neglect the imprint on the final result of time intervals in the minute range.
After the application of the proton flare detection on the complete event lists, a first source detection (see Sect. 3.4 for more details) is carried out. The obtained source lists are then used to filter the complete event lists and produce event lists containing only background events. With these new source filtered lists used as input, we invoke the proton flare filter a second time. By eliminating the contribution of point sources from the input for this second application of the filter, we mitigate the confusion created by sources at the step of calculating the background count rates for each observation. With that, we achieve an important improvement in the sensitivity of the proton flare detection, with a difference of about 10% in the final total effective time of the sample. This difference is mainly because of flares, whose intensities are dominated in the total light curves of their observations by an important contribution of point sources. Then, after selecting the events corresponding to point sources, light curves of "pure'' background events can be calculated. By using these light curves, the flares become statistically significant and can be detected by the programs.
In the final step, the filtered event lists corresponding to the good time intervals are calculated. From the initial 3.8 Ms exposure time included in the total sample, about 3.0 Ms effective time is left after the application of our filtering of proton flares for all EPIC cameras (see Cols. 5 to 9 in Tables 3 to 6), although the obtained effective exposure time for MOS2 is in general slightly larger than for MOS1.
Equipped with our tools to automatically search for proton flares, we investigated a large number of XMM-Newton observations (see Tables 3 to 6). We found that proton flares show up in all XMM-Newton energy regimes with a broad variety of X-ray spectra. The results presented in Fig. 2 indicate that it is not safe to restrict the search for proton flares to the high energy regime. This is in opposition to the proposals by Lumb et al. (2002) or Read & Ponman (2003), which suggest a search for proton flares restricted to the
.
Our results show that the search for proton flares should be extended to all energy regimes as part of the standard reduction of XMM-Newton data. This is in agreement with the results recently published by Nevalainen et al. (2005), who also find soft energy flares in their investigation of the X-ray emission of galaxy clusters.
![]() |
Figure 2: Hardness ratios of the proton flares detected with our procedure for observations performed with the medium filter (see Sect. 3.2.1). With the energy bands C1, C2 and C3 we make use of the whole EPIC energy coverage. For clarity, we omit the error bars which vary in a broad range. This is because there are significant differences in the respective intensities of the proton flares represented in the figure, from very faint to very bright which correspond to relatively large (about 35%) and small (about 10%) error bars respectively. We can see that the hardness ratios of the proton flares are wide spread in all ranges. In contrast to the suggestions by Lumb et al. (2002); Read & Ponman (2003), there is an important fraction of proton flares with soft and medium spectra in the lower half of the diagram. |
Open with DEXTER |
![]() |
Figure 3: X-ray background map towards the field indicated in the upper part of the panel, as obtained by the standard SAS data reduction tasks: the SAS-calibrated photon image is "source filtered'' (see Sect. 3.4), smoothed and finally corrected by the exposure map calculated by the corresponding SAS task ( eexpmap). The increase in intensity towards the rims of the detector (see inverted tunnel effect in Sect. 4) is created by the use of the task eexpmap. This resulting background map is unacceptable since, at the B4 band, the X-ray background is dominated by the extragalactic background (Read & Ponman 2003) which must yield in very homogeneous intensity distributions. |
Open with DEXTER |
The contribution of the un-rejected cosmic-ray induced particle background (CRB) is negligible (Lumb et al. 2002). The internal background is dominated by the contribution of the XRB in the energy range
(Fig. 8 from Lumb et al. 2002).
In the case of fluorescent lines in the energy regime from B1 to B4, like the Al-K line at
(Lumb et al. 2002), we must consider that the contribution of narrow lines vanishes when we focus on the study of broad energy bands. Consequently, we exclude a treatment of this effect in our data reduction process. However, this assumption will be revised in the future, when the available database would permit a precise spectral investigation of the vignetting by making use of narrower energy channels as those used here (see Table 1) Then, a higher statistical certainty in the data, could permit testing the spatial distribution of the fluorescent lines and a more specific analysis of the most intense lines.
On the contrary, the contribution of the XRB is maximized for
and, in principle, it cannot be neglected in our study. In order to avoid a specific analysis of the XRB for each observation, which would dramatically increase the computation time for an automatic procedure applied to a large database like in the work at hand, we have chosen our sample to include a very wide range in Galactic coordinates (Cols. 2 and 3 in Tables 3 to 6). We also have a wide range in the column density of the photoelectric absorbing material
(
). These selections ensure a broad range in the XRB intensities of our sample, as already confirmed by the RASS. Additionally, the large number of selected observations yields a rich variety in structures of absorber column density present in the individual fields. Then, by averaging all fields of the sample, the differences in the structure of the XRB for each field are compensated. This will be of special importance in our study of the in-flight vignetting of XMM-Newton in Sect. 4 and the following. There, we will be interested in the radial slope of the vignetting function and the differences in intensities among the fields will be canceled by normalizing the images to the value at the center of the field. Consequently, a specific study of the XRB in each field will be neglected in our investigation of the vignetting in XMM-Newton.
After completing the source detection, we make use of a self-developed tool to automatically eliminate the contribution of all point sources. This was done by the creation of so-called cheese images and posterior refilling of the gaps with the background level of a nearby source-free region. The refilling is necessary in order to avoid an artificial decrease in the intensity of our accumulated database towards the center of the detectors due to the higher source detection rate in the central area of the cameras. We explain this in some more detail in the following.
![]() |
Figure 4: Normalized mean exposure time vs. radial distance to the optical center of the EPIC MOS2 camera for the four energy bands shown at the upper right part of each diagram. The thick solid line shows our new developed exposure maps. To illustrate the quality of these exposure maps, we also show the background intensity distribution, without correction for vignetting, of four different observations (thin solid lines). The error bars are calculated based on the photon statistics of the inner 5 arcmin of the observations. The solid lines can be compared to the SAS calculated exposure maps (dashed lines) for the corresponding observations and bands. The use of version 5.4.1 of the SAS software and a more modern calibration database (synchronization of March, 2004), as mentioned in Sect. 5, yields only slight variations in the shape of the obtained exposure maps. |
Open with DEXTER |
![]() |
Figure 5:
Normalized exposure maps for the four energy bands indicated on the top of the panels obtained with our method applied to the EPIC MOS1 detector with the medium filter. The grids of each map are presented in detector coordinates. The gaps in the maps correspond to the detector mask. Apart from a small residual contamination from CCD2 (see Sect. 5) in the lower right part of the B1 band map, a good radial symmetry is present in all panels. The total effective exposure times used to calculate the maps of this figure are
![]() ![]() ![]() |
Open with DEXTER |
![]() |
Figure 6:
Normalized exposure maps for the four energy bands indicated on the top of the panels obtained with our method applied to the EPIC MOS2 detector with the medium filter. The grids of each map are presented in detector coordinates. The gaps in the maps correspond to the detector mask. A good radial symmetry is present in all panels. The total effective exposure times used to calculate the maps of this figure are
![]() |
Open with DEXTER |
![]() |
Figure 7:
Normalized exposure maps for the four energy bands indicated on the top of the panels obtained with our method applied to the EPIC pn detector with the medium filter. The grids of each map are presented in detector coordinates. The gaps in the maps correspond to the detector mask. Like in the case of MOS1 (see Fig. 5), there are clear unexpected asymmetries in the radial distribution of the B1 band vignetting. The total effective exposure times used to calculate the maps of this figure are
![]() |
Open with DEXTER |
![]() |
Figure 8: Radial profiles of the exposure maps for the energy bands indicated in the upper right part of the diagrams in the cases of the thin (dotted line), medium (solid line) and thick (dashed line) filters. For clarity, only the error bars for the medium filter are shown. These are calculated based on the variation of the exposure across the rings used for the calculation of the radial profile. There are clear differences in the profiles for the different filters, like for the thick filter in the B2 panel. |
Open with DEXTER |
In Sect. 4 we will average the background maps obtained here and, for this aim, maps with a high number of sources close to the center of the field are unacceptable since they yield in a significant loss of XRB intensity in the inner 5 arcmin. A refilling step for every field is necessary. In this step, we fill each hole in the cheese images with an approximate background intensity value according to a source free region in its vicinity. For this purpose, we implemented an iterative procedure based on a Gaussian smoothing task (asmooth of SAS with smoothtype = "simple'') with
.
In each iteration, background areas (without sources) are kept constant. Then, the value obtained by the Gaussian filter for the areas with sources is used as input for the same areas in the next iteration step. In general at most five iteration steps are necessary until the difference between the areas with sources of two successive steps stays below the accuracy level of the data (see also Sect. 3.2).
Our choice of a smoothing size of
is justified with similar arguments as for the exposure times in Sect. 2. At least one X-ray photon is expected in each point of the new resolution grid with 1.5 arcmin angular spacing.
We tested the SAS 5.3.3 tool eexpmap for the calculation of the exposure maps. We detected an overestimation of the correction towards the rims of the FOV (see Fig. 3), which yields final intensity maps with very bright rims (inverted tunnel effect). Read & Ponman (2003) faced this problem and proposed a different method to correct for vignetting based on the in-flight performance of XMM-Newton. We extended their investigation to a longer accumulated exposure time(from
to
)
and take advantage of the improved data reduction tools presented in the preceeding sections.
We use the background maps explained in Sect. 3.4 as a weighted sample to calculate the mean background intensity distribution affected by vignetting as observed by XMM-Newton in-flight. The weighting used for observations in the sample is proportional to the effective observation time (after proton filtering) as shown in Cols. 7 to 9 in Tables 3 to 6. In all cases, the background maps are converted to detector coordinates (XMM-Newton Users' Handbook, Ehle et al. 2004, available at the VILSPA web pages). Thus, and since the observations are summed up always in the same orientation (see Sect. 5), we can study not only the radial distribution of the mean exposure maps, which is valid only if there is radial symmetry in the vignetting, but the complete 2-D distribution.
In Fig. 4, we present our results for the medium and low energy regimes (B1 to B4, i.e.
). We focus on these regimes because:
![]() |
Figure 9: EPIC MOS1 exposure maps calculated for the three separated data sets indicated in the top of each figure for the medium filter. The white grid and numbers shown in the upper panel indicate the positions of the seven CCDs in the MOS1 camera. The maps are normalized to a value of 1 in the center of the map. The count rate increase in the contaminated observations accounts up to a value of 1.5 in the affected region of the normalized exposure map. This corresponds up to a factor of 2 above the uncontaminated case (shown in the upper left panel of Fig. 5). |
Open with DEXTER |
Concerning the use of our exposure maps for specific XMM-Newton pointings, the maps must be re-scaled to the effective exposure time of the observation. In some cases, an additional re-scaling is necessary to adjust the overall shape of the exposure maps and the observed X-ray distribution before performing the correction for vignetting. This is because individual observations of the background - mainly in the low energy regime B1 and B2 - show real structure partly due to variations in the absorbing column density in the FOV and partly to the limitations in the source detection procedures which may fail to detect point sources with very faint emission. These effects occasionally result in an enhancement or, alternatively, in a decrease of the background intensity in the center region of the detector for the individual observations. Since this inner region (
)
is used as default reference to normalize the re-scaling of the exposure maps, a new scaling (based in another detector region and generally within 10% of the default solution) might be needed in some cases to avoid an over- or underestimation of the total background count rate. At this point, the exposure maps are ready to be used.
The final results for our calculations for the medium filter case are shown in Figs. 5 to 7. There, the effect of bright CCD contamination (see next section) has been almost completely eliminated and, therefore, the statistical significance of the final MOS1 exposure maps for the B1 and B2 energy bands is lower than for the remaining cases.
We also detected differences in the vignetting for alternative filters, as can be seen in Fig. 8. There, the radial profile of the exposure map for the MOS2 detector is compared for observations with the thin, medium and thick filters. Although in most of the cases the profiles are in agreement according to the error bars, some clearly deviate, like the profile for the thick filter in the B2 band or for medium in B4. Because of this differences we recommend the use of separate sets of exposure maps for each filter.
In the near future, electronic versions of the exposure maps presented in Figs. 5 to 7 (also for the cases of the thin and thick filters) will be available, together with dedicated software for the regridding of the maps from detector into the sky coordinates grid of a given XMM-Newton observation.
We performed an individual study of the observations in our database (all filters) and
detected a number of observations which showed a clear enhancement, up to a
factor of two, in the observed count rate in certain regions of the
MOS1 detector, independently of the used filter. The count rate increase is confined to CCD2 in most
cases. However, among these affected observations, some present an additional contamination of CCD5. In Tables 3 to 6, the superscripts 2 and 5 in the first column
mark observations which present contamination, in CCD2 and 5 respectively. We
selected as contaminated the observations with mean background intensities in
CCD2 or CCD5 exceeding those of the
uncontaminated CCDs by more than the level of accuracy of the data. More
formally:
It is important to note, that we also made use of more actual versions of the data analysis software (SAS 5.4.1) and calibration database (synchronization of March, 2004) to re-analyze the contaminated fields. The results obtained with the newest tools are only marginally different from those obtained with the older. The bright CCD contamination appears in the same manner in both cases. Furthermore, the use of the SAS task eexpmap to calculate the exposure maps in the contaminated observations yields results that do not reflect the intensity enhancement registered during the observations, although the shape of the exposure maps calculated with the newest tools differs slightly from those obtained with the older versions. To summarize, the standard tools do not perform any correction for the contamination reported here.
We also compared our set of affected observations with those analyzed by Read & Ponman (2003) to compute their EPIC background maps for the medium filter. From a total of 21 pointings in their sample, at least 3 contain CCD5 contamination (observation IDs 0021740101, 0022740301 and 008564021) and several more shown up with CCD2 contamination. The use of this contaminated observations produces an up to now undetected asymmetry in their MOS1 event files and exposure maps for the soft energy regime (B1) equivalent to the one presented in Fig. 9.
After separating the observations in three sets, we repeated the calculation of the exposure maps for the cases: without bright CCDs, with bright CCD2 and with bright CCD5. The results for the contaminated sets are shown in the middle and lower panels of Fig. 9. With this first rough classification of the observations, the map obtained for the cases without bright CCDs (upper left panel in Fig. 5) is not completely satisfactory because the effect of CCD2 contamination is still present in the resulting exposure map. Furthermore, CCD2 contamination is also clearly present in the map calculated with the CCD5 contaminated observations (lower panel in Fig. 9). All this indicates that probably all observations are affected by the CCD2 contamination, as we might derive from the upper panels of Fig. 10. From those panels, we deduce that about 90% of the observations included in our sample show a higher background rate in CCD2 than in the remaining CCDs, although the excess is above the precision of the data only in about 50% of the cases (Eq. (1)). In contrast, CCD5 contamination is confined to only a reduced fraction of the observations (15-20%) and its effect does not appear when calculating exposure maps with the remaining observations in the database: CCD2 contaminated and uncontaminated pointings in the middle and lower panels of Fig. 9 respectively. Another difference between both contaminations is given by the shape of the contaminated region in each case (see different shape for the contaminated regions in the middle and lower panels of Fig. 9). Additionally, the contamination of CCD5 extends into the B2 band, while the case of CCD2 is confined only to B1. All these considerations point to different origins for the contaminations of CCD2 and CCD5. In the following Sect., we present our first results in constraining these different origins. We will focus on energy band B1 because the bright CCD contamination in B2 is only present in 10% of the observations and never exceeds a factor 1.3 above the uncontaminated background intensity. However, the observations contaminated in the B2 band are excluded from the calculation of the final exposure map shown in the upper right panel of Fig. 5.
![]() |
Figure 10: Upper panels: scatter plots of the CCD2 contamination as defined in Eq. (2) vs. the quantities indicated in the horizontal axes in logarithmic scale. The points with error bars represent contaminated observations (Eq. (1)), where solid squares correspond to observations which also present CCD5 contamination. Empty squares represent uncontaminated observations. The curves correspond to the regression lines of the data. The statistical significance of the regression lines is shown in Table 2. The mean background intensities of the left panel are calculated using the data gained with CCDs 3, 4, 5 and 7. Lower panels: equivalent to the upper panels for the CCD5 contamination. Points with error bars are CCD5 contaminated, solid squares show CCD2 contaminated observations and empty squares represent uncontaminated observations. No regression lines are shown because the statistical tests do not reject the zero correlation hypothesis in any of the three cases shown here (see Table 2 and Sect. 5.1). Visually, the CCD5 contaminated set is well separated from the remaining observations which are located close to the C5 = 1 line as expected for pointings free of contamination. |
Open with DEXTER |
Table 2: Correlation coefficients of the CCD contamination rates as defined in Sect. 5 vs. the observation parameters listed in the left column of the table. The last row shows the critical value of a the statistical test for the significance of the correlation coefficient of the sample at a 95% confidence level adopted from Crow et al. (1960). We shown the 99% confidence level value in parenthesis. Note that the zero correlation is only rejected with the values of the CCD2 column.
CCD2 contaminated observations (points with error bars in the upper panels of Fig. 10) are concentrated towards the high background count rate part of the diagram. There is also a clear tendency to present higher contamination rates for higher count rates. This is confirmed by the statistical test for the significance of the correlation coefficient of the sample, which rejects the zero correlation hypothesis between contamination rate (Eq. (2)) and the background count rates (see Table 2). With respect to the effective exposure time, after proton flare filtering, a negative correlation can be rejected at a 95% significance level but not at a 99% level. For this case, a larger sample of contaminated observations is necessary to derive definitive statistical conclusions. The correlation of CCD2 contamination rate with total counts is also accepted by the test, but the reliability of the correlation is significantly lower with respect to the case of the correlation with the background count rate. Furthermore, there are also uncontaminated observations in the high effective exposure time and high total counts parts of the diagrams (different to the count rate case). This indicates that probably the correlations found between the contamination rate and these two parameters (effective exposure time and total counts) are not real. As already mentioned, more data are necessary to derive definitive conclusions in this sense.
For the CCD5 contaminated data set, we detect a systematic overestimation in the CCD5 intensities of about a factor of two (points with error bars in the lower panels of Fig. 10). In contrast to the case of CCD2, there is no evidence for grouping of the contaminated observations towards any count rate, exposure time or total counts regime for CCD5 contamination. The correlation coefficients shown in the right column of Table 2 do not reject the un-correlation in any of the three cases with contamination rate.
We also tested whether the bright CCD contamination is triggered by extreme values in the X-ray background hardness ratios of the observations or is correlated with the exposure date (see Fig. 11 and Tables 3 to 6) and no relation could be found.
From our analysis of the bright CCD contaminations, we conclude that they probably
have their origin in software problems or in the electronics of the CCD. The contamination is always
present and grows with mean background count rate in the case of
CCD2. Therefore, the use of diffuse X-ray data from this CCD2 should be avoided until a
consistent method to mitigate the effect of the contamination is developed. For
CCD5 case, since the frequency in the presence of this contamination is below 20% and the
contamination rate is always high (
C5 > 1.5 using an equivalent definition to that of Eq. (2)), we recommend to check whether the intensity
increment is visible in the raw data before performing further analysis of the diffuse X-ray
data gained with this CCD.
![]() |
Figure 11: Hardness ratios of the X-ray backgrounds as detected with the EPIC MOS1 camera for the observations in Tables 3 to 6 (all filters). Empty squares correspond to observations free of CCD2 or CCD5 contamination, solid squares to CCD2 contaminated observations and the points with the error bars to CCD5 contaminations. However, in order to avoid the influence of this contamination in the hardness ratios presented here, we calculate the values based only on the data provided by uncontaminated CCDs. The results shown in this figure rule out a correlation between the hardness ratio of the X-ray background towards the observed field and the presence of bright CCD contamination, either for the CCD5 or for the CCD2 case. This can be deduced from the fact that the contaminated observations are spread in the panel with a similar distribution to that of the uncontaminated observations. |
Open with DEXTER |
Since the standard tools for the vignetting correction in the XMM-Newton EPIC detectors are not sufficient to allow a sensitive analysis of faint signals in the soft energy regime (
keV), as we can see in Fig. 3, we have developed a new task to calculate the real exposure map of XMM-Newton. It is based on the in-flight performance of the observatory and uses longest accumulated exposure time today. Our results point to differences between the vignetting of observations performed with different filters. The process to calculate exposure maps can be repeated automatically for user defined energy bands and observations. An extension to modes alternative to the FullFrame mode presented here can be performed by exploiting the XMM-Newton Science Archive.
We detect an artificial increase in the measured count rates (softest energy,
)
in the EPIC MOS1 CCDs 2 and
5. This contamination appears to be present in practically all observations and it seems to
be associated to high background rates in the case of CCD2. The set of
affected observations accumulated for CCD5 contamination until today is not
sufficient to draw definitive conclusions on the origin of the
contamination. The count rate in the affected CCDs is increased up to a factor
of two. Such a significant difference must be taken into account when
investigating diffuse emission or the hardness ratios of faint sources
detected by one of these two CCDs. Also, the EPIC pn exposure map presents asymmetries in the lower energy regime, which are and obstacle for use of the outer parts of the detector to investigate faint diffuse emission. However, the asymmetries in the pn case are lower in magnitude that those for the MOS1 detector.
To summarize, the actual status of the data reduction for XMM-Newton still presents fundamental problems which prevent a reliable study of faint soft X-ray sources (e.g., Bright CCD contamination). These difficulties have to be further investigated and solved before being able of using the full capabilities of the satellite. Interesting topics that could be in principle investigated with XMM-Newton, like for example:
Acknowledgements
J.P. likes to thank the Deutsches Zentrum für Luft- und Raumfahrt and the Graduiertenkolleg 787 of the universities of Bonn and Bochum for the financial support.
Table 3:
Summary of the observations extracted from the XMM-Newton Science Archive for our investigation of the XMM-data reduction. This table shows observations performed with the thin filter. The fourth column shows the total column densities derived from the newest all-sky stray-radiation corrected 21-cm line surveys (Kalberla et al., submitted). The
columns show the total integration time of the observation for the MOS and pn instruments. The
columns show the effective time after proton flare filtering (see Sect. 3.2) for each observation in MOS1, MOS2 and pn as indicated. The accumulated integration time of the thin filter sample is
for MOS and
for pn. The accumulated effective time amounts to
for MOS and
for pn.
Table 4:
Like Table 3 for the case of the observations with the medium filter. The accumulated integration time of this sample is
for MOS and
for pn. The accumulated effective time amounts to
for MOS and
for pn.
Table 5:
Like Table 3 for the case of the observations with the thick filter. The accumulated integration time of this sample is
for MOS and
for pn. The accumulated effective time amounts to
for MOS and
for pn.
Table 6: Summary of of our sample of pointings performed with XMM-Newton (see also Table 3). The directions of the pointings have been specifically selected to study the contribution of the SXRB as seen by XMM-Newton.