Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~kim/pap/2007_M33_II.pdf Äàòà èçìåíåíèÿ: Mon Dec 6 20:20:05 2010 Äàòà èíäåêñèðîâàíèÿ: Tue Oct 2 00:19:51 2012 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 ï ï ï ï ï ï

The Astrophysical Journal Supplement Series, 173:70 Y 84, 2007 November
# 2007. The American Astronomical Society. All rights reserved. Printed in U.S.A.

A

THE X-RAY BINARY POPULATION IN M33. II. X-RAY SPECTRA AND VARIABILITY
H.-J. Grimm, J. McDowell, A. Zezas, D.-W. Kim, and G. Fabbiano
Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138 Received 2006 August 8; accepted 2007 April 24

ABSTRACT In this paper we investigate the X-ray spectra and X-ray spectral variability of compact X-ray sources for 3 Chandra observations of the Local Group galaxy M33. The observations are centered on the nucleus and the star-forming region NGC 604. In the observations, 261 sources have been detected. For a total of 43 sources the number of net counts is above 100, sufficient for a more detailed spectral fitting. Of these sources, 25 have been observed in more than one observation, allowing the study of spectral variability on timescales of $months. A quarter of the sources are found to be variable between observations. However, except for two foreground sources, no source is variable within any observation above the 99% confidence level. Only six sources show significant spectral variability between observations. A comparison of NH values with H i observations shows that X-ray absorption values are consistent with Galactic X-ray binaries and most sources in M33 are intrinsically absorbed. The pattern of variability and the spectral parameters of these sources are consistent with the M33 X-ray source population being dominated by X-ray binaries: Two-thirds of the 43 bright sources have spectral and timing properties consistent with X-ray binaries; we also find two candidates for supersoft sources and two candidates for quasi-soft sources. Subject headingg galaxies: individual ( M33) -- Local Group -- X-rays: binaries s: Online material: color figures, extended figure set 1. INTRODUCTION M33 is a late-type spiral galaxy, Sc II Y III, and the third largest galaxy in the Local Group. It is a unique galaxy in the Local Group since morphologically it is of intermediate type between the large early-type spiral galaxies and the numerous dwarf galaxies. Other galaxies of this type cannot be investigated with the same depth even with Chandra. At a distance of 840 kpc ( Freedman et al. 1991) from the Milky Way ( MW ) M33 is the second nearest major galaxy. It spans roughly 73 0 ; 45 0 on the sky. The lineof-sight absorption column density is small, NH $ 6 ; 10 20 cmþ2 (Stark et al. 1992). M33 is more actively star forming than either the MW or M31 ( Hippelein et al. 2003), particularly compared to its much smaller mass. M33 has been studied with every X-ray mission since Einstein ( Markert & Rallis 1983), but only recently have high angular resolution and high-sensitivity instruments like Chandra and XMMNewton allowed us to study the X-ray source population in depth. Grimm et al. (2005) and Pietsch et al. (2004) have provided source lists and fluxes for the X-ray source population in M33, from the Chandra and XMM-Newton observations, respectively. In this paper we follow up on the Grimm et al. (2005) Chandra survey and present an analysis of the X-ray spectra and variability behavior of X-ray sources in M33. Apart from fluxes, variability and spectral energy distributions are an important diagnostic tool for understanding the emission mechanism(s) in X-ray sources and for classifying these sources. Moreover, spectral and time variability are the main characteristics of X-ray binaries, as shown by the detailed work done for X-ray binaries in the Milky Way (see, e.g., van der Klis 2005). The paper is organized as follows. In x 2 we describe the data processing, followed in x 3 by the analysis procedures. The results and their implications for the nature of the X-ray source population in M33 are discussed in x 4; we also discuss individual bright sources in detail in x 4.4. We conclude with a summary of the results in x 5. The figures showing the results for the spectral analysis for sources with more than one observation are shown in Appendix B. 70 2. DATA M33 has been observed with the ACIS instrument on Chandra four times (see Grimm et al. 2005). In this paper we use three observations whose ObsIDs and dates are given in Table 1. Due to its angular extent the observed parts of M33 cover all active chips, the standard ACIS-S configuration for ObsID 786, and the standard ACIS-I configuration for ObsID 1730 and ObsID 2023. There is considerable overlap between the different observations. However, due to the decreasing resolution /sensitivity with increasing off-axis angle only the inner part of M33 ($80 Y 100 ) has a significant number of sources in two observations, 786 and 1730. A fourth observation, ObsID 787, was disregarded because it was aimed at studying the nucleus and suffers from both high background and small FOV. The data from the three observations were processed according to the standard data processing procedure with CIAO versions 3.1, including exposure correction. Source detection was performed with wavdetect with scales of 1, 2, 4, 8, 10, 12, and 16 in the energy range 0.3 Y 8 keV. The signal detection threshold was set to 10þ6. Source regions correspond to the 95% encircled energy area. For more details of the data analysis (see Grimm et al. 2005). 3. ANALYSIS We separate the analysis in three parts, short-term, long-term, and spectral variability. The different methods are discussed in the following subsections. We assume that the variability of individual sources is independent, so there is no correlation between short-term and long-term variability. Because the number of counts observed in a single observation is not very large even for bright sources, we restrict the spectral variability analysis to a comparison between different observations, when applicable. The nucleus was excluded from the following analysis because it suffers from strong pile-up in two observations. 3.1. Short- and Long-Term Variability In order to establish short-timescale variability we perform a Bayesian block analysis of the light curves of individual sources.


M33 X-RAY VARIABILITY
TABLE 1 List of ACIS Observations of M33 Duration ( ks) 45 45 90

71

ObsID 786........................ 1730...................... 2023......................

Date 2000 Aug 30 2000 Jul 12 2001 Jul 06

Aim Point Nucleus Nucleus NGC 604

A Bayesian block analysis computes the best approximation to the light curve shape in terms of piecewise constant flux levels or blocks. The discriminator between a single flux level for the whole light curve or two flux levels is the ratio of likelihoods of describing a data segment with one or two blocks. In this analysis the algorithm used is iterative. It starts with the whole light curve and subsequently divides it until the likelihood ratio for dividing a light curve segment becomes smaller than a predefined prior, in this case corresponding to $99% confidence level. The Bayesian block analysis is particularly suited for burstlike variability. For more details about the principle of Bayesian blocks and the algorithm used in this analysis (see Scargle 1998). The implementation used here is the same as used by the CHAMP project and described in Kim et al. (2004). Since a Bayesian block analysis uses only the photon arrival times for computation of the blocks, no binning is necessary. Therefore, there is no intrinsic restriction to the number of photons the algorithm is applicable to. Obviously, establishing variability with a certain confidence becomes less likely for fainter sources. Because a Bayesian block analysis is particularly sensitive to burstlike variability, we also performed a search for periodic variability with the XRONOS v5.21 tool efsearch. This analysis did not yield any source with significant short-term variability. To establish long-timescale variability we use simple Poisson statistics. This is justified because, with the exception of two foreground sources, no source exhibits strong short-term variability. We compare the difference in fluxes between two observations to the quadratically added errors for the fluxes. If a source is undetected in an observation in which it was in the field of view of Chandra, we calculate an upper limit to the source flux using an algorithm developed by Kraft et al. (1991). Since the upper limit value already corresponds to the 99% confidence level, for a comparison with a detected flux we compare the 3 error of the detection directly with the upper limit. 3.2. Spectra and Spectral Variability For 43 out of the total 261 sources the number of counts in at least one observation was larger than 100 net counts, which we consider sufficient to attempt spectral fitting. Of these 43 sources, 25 have been observed in at least two observations. For these sources we also compare the spectral properties with time. The spectral fitting was done with XSPEC v11.3.1. Because of the generally low number of counts the fitting was done for all sources with Cash statistics (Cash 1979). At 100 counts we expect only $3 background sources based on the CDF-N log N Y log S (Alexander et al. 2003), so the vast majority of these sources are likely to belong to M33. We first fitted the bright sources with two simple absorbed spectral models, power law and bremsstrahlung, to check the validity of our assumptions about the spectral shape of the faint sources that do not have sufficient counts for detailed spectral modeling (see derivation of the X-ray luminosity function in Grimm et al. 2005). The absorption was in one case a free parameter, in the other case it was fixed to the Galactic value. After validating

that the assumptions about the general power-law shape are correct, we proceeded to fit more complex models to the bright sources. The results of the simple spectral fitting are presented in x 4. Spectra of sources with more than one observation with best-fit values and corresponding confidence contour plots are shown in Appendix B, except for the spectra of M33 X-7 (CXO J013334.1+ 303210) and M33 X-9 (CXO J013358.8+305004), which are shown in Figures 7 and 8 and will be discussed in x 4.4. Based on the confidence regions for the fit parameters, shown in Appendix B, only one of the X-ray sources shows significant spectral variability from observation to observation, M33 X-4 (CXO J013315.1+305317). For the other sources in different observations none of the parameters of a source spectrum show changes corresponding to more than 99% confidence level. This is despite the fact that some sources show significant time variability between observations. However, the number of counts available for most sources for fitting is insufficient to determine spectral parameters to an accuracy good enough to compare two observations. Moreover, degeneracies between model parameters complicate the establishing of variability. Another approach to spectral variability is the use of hardness ratios. These are relatively crude estimators of spectral changes, but because of the smaller number of degrees of freedom compared to a spectral fit can be statistically preferable. We therefore take all sources that were observed in at least two observations, divide each observation in three equally long parts, and construct hardness ratios from these intervals. Because only sources with more than 100 counts are used in this part of the analysis, the separation in three parts still gives sufficient number of counts (!20 Y 30) in each time bin for a hardness ratio analysis ( Prestwich et al. 2003). Note that ObsID 2023 is roughly twice as long as the other observations. The energy bands used are the same as in Grimm et al. (2005); 0.3 Y 1.0 keV for the soft band, 1.0 Y 2.1 keV for the medium band, and 2.1 Y 8.0 keV for the hard band. The hardness ratios are defined as HR1 ¼ M þS ; T HR2 ¼ H þM ; T Ï 1÷

where S, M, and H are the background and exposure corrected counts in the soft, medium, and hard band, and T represents the corrected counts in the whole energy band. To compare two observations it is important to take into account the different effective areas of the source location in each observation, especially if the source is located on a front-illuminated chip and in another observation on a back-illuminated chip. We compute the effective area of the source in each observation for each energy band. For comparison with the other observations of the source we normalize the effective area in each band by the effective area value of the aim point of ObsID 1730. Note that the choice of the normalization constant is arbitrary. The results are combined for all available observations. The results for all sources that show evidence of variability are shown in Appendix B. 4. RESULTS AND DISCUSSION 4.1. Short-Timescale Variability Except for two sources that are foreground Galactic stars, no source presents variability above the 99% confidence level. In Figure 1 we show the results of the Bayesian block analysis for the only two sources that are variable on short timescales. The dotted histograms represent the count rate binned in 400 s intervals. Note that the binning is for plotting purposes only. The green line is the power spectrum of the light curve. The thick solid line is


72

GRIMM ET AL.

Vol. 173

Fig. 1.-- Left: Light curve and Bayesian block division of the outburst of CXO J013327.7+304645, an X-ray active star. The panel shows the light curve binned in 400 s intervals (dashed histogram). Note that the light curve binning is for plotting purposes only. The Bayesian blocks are shown as the thick black histogram. The right part of the panel shows a thumbnail picture of the source. Right : Light curve and Bayesian block division of the outburst of CXO J013341.8+303848, another X-ray active star. [See the electronic edition of the Supplement for a color version of this figure.]

the result of the Bayesian block analysis. The small square to the right of the light curve is a thumbnail image of the source from the observation. Source CXO J013327.7+304645 shows a clear X-ray outburst in ObsID 2023. This outburst, lasting about 10,000 s completely accounts for the long-term variability of the source. The light curve of the outburst is well fitted by an exponential decay with a decay timescale of 3900 s. The persistent luminosity in ObsID 2023 is consistent within the errors with the luminosities in the other observations. Outburst and persistent counts are not large enough to investigate changes in spectral shape. An X-ray color comparison of the burst and persistent emission shows a softening of the emission in HR2 and a slight hardening in HR1 that is significant only at the 2 level. The optical counterpart to source CXO J013327.7+304645 is a star in the USNO catalog with a V magnitude of 16.8. We have analyzed archival Hubble Space Telescope Wide Field Planetary Camera 2 (HST WFPC2) data. The object was detected in bands corresponding to U, B, V, and I. The colors are consistent with an early M type star, either on the main sequence or a giant. Using normal V magnitudes for M stars, this puts the star at a distance of 90 Y 320 pc for a main-sequence star, or 24 Y 29 kpc for a giant. Considering the relatively high Galactic latitude of M33, b ¼ þ31 , the lower distance value, and thus a main-sequence star is more likely. Assuming a main-sequence star the peak luminosity is between $10 29 and 1:5 ; 10 30 ergs sþ1, and the total energy between $1032 and 1:5 ; 10 33 ergs. Source CXO J013341.8+303848, a foreground star as well, has two outbursts in ObsID 786. The first outburst lasts about 2600 s, the second outburst 18,000 s later lasts about 800 s. The persistent level of X-ray emission increases slightly after each outburst. As shown in the long-term light curve the source also has a long-term trend to increasing luminosity for the three observations. The decay of the outbursts cannot be fitted uniquely; an exponential decay with a decay constant of $1000 s or a linear decay are both possible. During both outbursts the source becomes softer, but only at the 1 Y 2 level. This source has no HST coverage. 4.2. Long-Term Variability Of a total of 261 sources, 198 have been detected in at least two observations, and 62 in all three observations. The luminosi-

ties for comparison are taken from spectral fits for the brighter sources (more than $100 counts), or from the assumed spectrum, absorbed power law with a photon index of 2 and Galactic absorption. We find that 49 of 198 sources show variability between observations above 3 . An additional 29 sources show signs of variability between 2 and 3 . Of the 49 variable sources, only 16 are apparently persistent. The strength of the variability can in fact be used to confirm the nature of most of these sources as X-ray binaries. Active galactic nuclei (AGNs) in general show variability on weeks or months timescale only up to factors of 2 Y 3 ( Mushotzky et al. 1993; Paolillo et al. 2004). Most of the sources in M33 show stronger variability, particularly since the flux ratios for most sources are only lower limits. Moreover, the sources with flux ratios below 2 are bright sources that are unlikely to be AGNs based only on brightness. Taking into account that $80 Y 90 of the 198 sources are likely background AGNs, somewhat less than half the sources in M33 are variable. Of these, two-thirds (34/49) may be candidate transients. This number is somewhat higher than the about 50% in the Milky Way and Magellanic Clouds ( Liu et al. 2000; Liu et al. 2001). But the detection limits ( few 1034 ergs sþ1) do not allow us to establish true transient behavior for undetected sources, so the number of two-thirds is only an upper limit. For the brighter sources (LX > 5 ; 10 36 ergs sþ1) past X-ray satellites also provide data for long-term variability. Figure 2 shows data from Einstein HRI and IPC ( Trinchieri et al. 1988), ROSAT HRI (Schulman & Bregman 1995) and PSPC ( Long et al. 1996 ), and BeppoSAX ( Parmar et al. 2001), as well as from this work for the bright sources in the field of view of Chandra. The quoted luminosities are converted to the Chandra band of 0.3 Y 8.0 keV and, if they are not individual measurements of spectra, converted to the spectral shape of a power law with photon index of 2 and the Galactic absorption value toward M33, 6 ; 10 20 cmþ2. The error bars in the plot only contain errors due to counting statistics. Other errors, e.g., due to the conversion of energy bands and different assumed spectral shapes, add generally another 20% Y 30% uncertainty in the luminosity. In addition different instrument responses, cross-calibration issues and other systematic effects add another source of systematic errors, which


No. 1, 2007

M33 X-RAY VARIABILITY

73

Fig. 2.-- Long-term light curves of bright X-ray sources in M33 observed with Einstein, ROSAT, BeppoSAX, and Chandra.

is hard to quantify. An exception is the supernova remnant M33 X-14, for which converting the luminosity given by Long et al. (1996 ) would result in variability at over 6 compared to other observations. However, Long et al. (1996 ) assume a power-law spectrum with a photon index of 2 to compute the source luminosity. Although this is a reasonable choice for X-ray binaries or AGNs, the spectrum of M33 X-14 is quite soft. We extracted a spectrum of X-14 from the ROSAT observation rp600023a00, the longest of the PSPC observations with $29 ks. The spectrum is well fitted by an absorbed blackbody model with a column density of 4 ; 10 21 cmþ2 and a temperature of 0.09 keV. Although this model is rather unphysical for a SNR we are interested only in the flux, which is sufficiently accurate for the purpose of comparison. The ROSAT data also agree with the spectrum obtained from Chandra. Extrapolating this spectrum to the Chandra energy range gives a luminosity of 3:7 ; 10 36 ergs sþ1, which agrees very well with other observations of the source, in particular the Chandra observed value of 3:6 ; 10 36 ergs sþ1. It is clear from Figure 2 that four of the bright X-ray sources are variable. The sources that do not show evidence of variability at the 3 level are M33 X-1, X-2, and the SNR X-14. The variability pattern is very similar to that observed in other X-ray source populations, regardless of galaxy type, e.g., The Antennae ( Zezas et al. 2006 ), M101 (Jenkins et al. 2005), and NGC 4697 (Sivakoff et al. 2005). In particular that only very few, if any, sources show short-term variability is very common, although this is most likely due to limited photon statistics. On the other hand, a large fraction ($10% Y 40%) of sources exhibit long-term variability. Moreover, the fraction of sources with strong long-term variability is quite confidently identified as X-ray binaries, and not background AGNs. 4.3. Spectra Fitting all the 43 sources detected with more than 100 net counts with a power-law and a thermal bremsstrahlung spectrum, with column density fixed to the Galactic value or as a free fit parameter, provides about half of the sources with a good fit to the data. The other sources require either different models, e.g., blackbody or other thermal plasma models, or multiple components. The results of the simple power law/ bremsstrahlung fits support the validity of our assumptions of a fixed spectral model for conversion from counts to fluxes for fainter sources, assuming that the spectral properties of bright and faint sources are not systematically different (Grimm et al. 2005).

In Figure 3 we show histograms for the best-fit power-law and bremsstrahlung values for all sources. The upper left panel shows the comparison between photon indices þ of a power law with column density being a free fit parameter versus column density fixed to the Galactic value. The Galactic absorption toward M33 is only 6 ; 10 20 cmþ2. The upper right panel shows the same for bremsstrahlung temperature kT. The lower left panel shows the histogram of þ for the case of NH as a free fit parameter. The power-law slopes are concentrated around 2, the canonical value for X-ray binaries and AGNs. The peak at þ ¼ 5 comprises all sources with photon index larger than or equal to 5. Individually, all these sources are fitted well with a blackbody or plasma model with temperatures of 0.1 Y 0.13 keV ( blackbody) or 0.2 Y 0.3 keV ( plasma model). The lower right panel finally shows the distribution of bremsstrahlung temperatures. Note that the influence of NH is stronger in case of a power law than for a bremsstrahlung spectrum, as is evident in the generally smaller deviations from the one-to-one correlation for the bremsstrahlung temperatures compared to the photon indices in the upper panels. As expected, fits with a fixed low column density produce smaller photon indices or larger bremsstrahlung temperatures, respectively. However, changes of temperature and photon index between fits with fixed and free column density are not significantly larger than the errors in the majority of cases. Comparing the difference between photon indices divided by the square root of the errors shows that $80% of the sources have values of less than 3. For the thermal bremsstrahlung model the corresponding value is 73%. Thus, the assumption of a general power-law spectrum with þ ¼ 2, and Galactic absorption is quite good (Grimm et al. 2005). Using the results of the more detailed spectral fitting of the 43 bright sources, we can confirm the results of the relatively blind spectral fitting of all sources with a power-law model. The left panel of Figure 4 shows the photon indices of spectra that are well fitted by either a single power law (open circles) or a combination of a power law and other components ( filled triangles) versus luminosity. Errors are 90% errors on the slope. There is a clear trend that photon indices in multicomponent fits are larger than for single power-law fits, and there is an indication that brighter sources are more likely well fitted by multicomponent fits than by single power laws. The reason is most likely that the single power-law sources have lower counts and multiple components are not distinguishable. We also fit the single power-law sources with a disk blackbody model, XSPEC model diskbb,which has the same number of degrees of freedom than the power-law model. Two-thirds of the spectra are well fitted by a disk blackbody model, the best-fit temperatures being in the range from 1 to 3 keV, as expected for X-ray binaries ( Tanaka 2001). The inner disk temperatures versus luminosity are shown in the middle panel of Figure 4. The values are the same as for the sources that require disk blackbodies for a good fit. This suggests that the single power-law sources are likely to be X-ray binaries and only the low number of counts allows a good fit with a single power law. There are 10 spectra (5 sources) that are best fitted with a single bremsstrahlung model with temperatures of $0.1 Y 0.3 keV. The bremsstrahlung temperatures versus luminosity are shown in the right panel of Figure 4 (open squares). Since these values are unusually low for real bremsstrahlung sources, and the luminosities of the sources relatively low, (1 Y 5) ; 10 36 ergs sþ1, we fit these spectra with a thermal plasma model ( XSPEC model apec) as well. With one exception, we obtain good fits with the plasma model as well, the temperatures are in the range from 0.1 to 0.4 keV, and metallicities from 0.05 to 0.3 solar with considerable uncertainties. These values are at or below the expected metallicity for


74

GRIMM ET AL.

Vol. 173

Fig. 3.-- Overall spectral properties of the sample of 43 bright X-ray sources in M33. The upper panels show the correlations of photon index þ with NH free vs. fixed (upper left) and bremsstrahlung temperature kT with NH free vs. fixed (upper right). Note that the influence of NH is stronger in case of a power law than for a bremsstrahlung spectrum, as evident in the smaller deviations from the one-to-one correlation for the bremsstrahlung temperature. The lower panels show the distributions of þ (lower left)and kT (lower right) for the case of NH as a free fit parameter. The power-law slopes are mainly in the range from 1.4 to 2.5, the canonical values for X-ray binaries and AGNs.

M33 ( Blair & Kirshner 1985). The good fit quality is not surprising considering that the apec model has one degree of freedom more than the simple bremsstrahlung model. However, given the low temperatures and the consistent metallicities we consider the apec model to be the more physical model for these sources. The sources with good fits for the apec model are marked in Table 3 (Appendix B), and the apec model parameters are given in Figure Set 9 (Appendix B), and the contour plots for these fits are shown as well. Also shown in the right panel of Figure 4 are six sources well fitted by a blackbody model with very low temperatures at or below 0.1 keV ( filled triangles). For two sources the

blackbody is the only component in the spectrum. These sources are candidates for supersoft sources and are discussed below in x 4.4. Three of the other four sources have an additional powerlaw component that ranges from hard ( þ $ 1:3) to very soft ( þ $ 4:6). The blackbody temperatures are even lower than the temperatures inferred from ULX intermediate mass black hole candidates (see, e.g., Miller et al. 2004). A truncated disk would be a possibility to explain the low temperature, similar to the scenario suggested by Kubota & Done (2004). However, given the luminosities the sources would be in the hard state (e.g., Maccarone 2003) but, except for CXO J013444.6+305535, the photon indices


No. 1, 2007

M33 X-RAY VARIABILITY

75

Fig. 4.-- Left panel shows the best-fit photon indices for sources that are well fitted by a single power-law (open circles) or a multicomponent spectrum containing a power law (open circles with filled triangles) vs. luminosity. Two-thirds of the single power-law sources are also well fitted by a disk blackbody model. The middle panel shows the disk blackbody temperatures vs. luminosity for sources that are well fitted by a single power law, but also by a simple disk blackbody model ( filled circles). As expected for X-ray binaries the disk temperatures are in the range from 1 to 3 keV. The temperatures are also in the same range as the temperatures for sources that require a disk blackbody component in the spectrum (open triangles). Right panel shows temperatures of sources that are either well fitted by a bremsstrahlung model (open squares) or a blackbody model ( filled triangles) vs. luminosity. With the exception of two spectra, the bremsstrahlung sources are also well fitted by a thermal plasma model, XSPEC model apec, as discussed in the text. Errors are at 90% confidence level. [See the electronic edition of the Supplement for a color version of this figure.]

are larger than 2. CXO J013444.6+305535 could indeed be in the hard state with a photon index of 1.3, but that number is not well constrained. This hard source could also be a candidate for a magnetic CV that are known to have hard spectra and relatively high luminosities ( Kuulkers et al. 2006 ). The soft power law in the other sources might on the other hand be another thermal component. Alternatively, the soft emission might be generated in an outflow from the system, or heating of a surrounding medium. Figure 5 shows a comparison of measured values for column density overlaid on a 1.49 GHz contour map of the central part of M33 from VLA (Condon 1987 ). The resolution of the VLA map is about 10 , and the confusion limit is given as 0.1 mJy. There is no spatial correlation between the contour map and the magnitude of the measure column densities. Thus, the X-ray absorption value is in part due to location of sources in front of / behind H i gas, and in part due to intrinsic absorption around the X-ray source. To compare actual values for the column density due to H i,we compute the brightness temperature of the H i gas according to S ¼ 2k 2 Tb ;