1: \documentclass[usenatbib,numberedappendix]{emulateapj}
2:
3: \newcommand{\be}{\begin{equation}}
4: \newcommand{\ee}{\end{equation}}
5: \newcommand{\bfig}{\begin{figure}}
6: \newcommand{\efig}{\end{figure}}
7: \newcommand{\bfige}{\begin{figure*}}
8: \newcommand{\efige}{\end{figure*}}
9: \newcommand{\bea}{\begin{eqnarray}}
10: \newcommand{\eea}{\end{eqnarray}}
11: \newcommand{\md}{M$_{\odot}$~}
12:
13: \newcommand{\chandra}{{\it Chandra}~}
14:
15:
16: \shorttitle{M33 X-ray variability}
17: \shortauthors{Grimm et al.}
18:
19: \begin{document}
20:
21: \title{The X-ray binary population in M33: II. X-ray spectra and
22: variability}
23: \author{H.-J. Grimm\altaffilmark{1}, J.~McDowell\altaffilmark{1},
24: A. Zezas\altaffilmark{1}, D.-W. Kim\altaffilmark{1},
25: G.~Fabbiano\altaffilmark{1}}
26: \affil{Harvard-Smithsonian Center for Astrophysics, 60 Garden Street,
27: Cambridge, MA 02138}
28:
29: \begin{abstract}
30: In this paper we investigate the X-ray spectra and X-ray spectral
31: variability of compact X-ray sources for 3 \chandra observations of
32: the Local Group galaxy M33. The observations are centered on the
33: nucleus and the star forming region NGC 604. In the observations 261
34: sources have been detected. For a total of 43 sources the number of
35: net counts is above 100, sufficient for a more detailed spectral
36: fitting. Of these sources, 25 have been observed in more than one
37: observation, allowing the study of spectral variability on
38: $\sim$months timescales. A quarter of the sources are found to be
39: variable between observations. However, except for two foreground
40: sources, no source is variable within any observation above the 99\%
41: confidence level. Only six sources show significant spectral
42: variability between observations. A comparison of $N_H$ values with HI
43: observations shows that X-ray absorption values are consistent with
44: Galactic X-ray binaries and most sources in M33 are intrinsically
45: absorbed. The pattern of variability and the spectral parameters of
46: these sources are consistent with the M33 X-ray source population being
47: dominated by X-ray binaries: Two thirds of the 43 bright sources have
48: spectral and timing properties consistent with X-ray binaries; we also
49: find two candidates for super-soft sources and two candidates for
50: quasi-soft sources.
51: \end{abstract}
52:
53: \keywords{X-rays: binaries --- galaxies: Local Group --- galaxies: individual (M33)}
54:
55: \section{Introduction}
56: \label{sec:intro}
57: M33 is a late-type spiral galaxy, Sc II-III, and the third largest
58: galaxy in the Local Group. It is a unique galaxy in the Local Group
59: since morphologically it is of intermediate type between the large
60: early-type spiral galaxies and the numerous dwarf galaxies. Other
61: galaxies of this type cannot be investigated with the same depth even
62: with \chandra. At a distance of 840 kpc \citep{freedman:91} from the
63: Milky Way (MW) M33 is the second nearest major galaxy. It spans
64: roughly 73'$\times$45' on the sky. The line-of-sight absorption
65: column density is small, $N_H \sim 6 \times 10^{20}$ cm$^{-2}$
66: \citep{stark:92}. M33 is more actively star forming than either the MW
67: or M31 \citep{hippelein:03}, particularly compared to its much smaller
68: mass.
69:
70: M33 has been studied with every X-ray mission since Einstein
71: \citep{markert:83}. But only recently have high angular resolution and
72: high sensitivity instruments like \chandra and XMM allowed us to study
73: the X-ray source population in depth. \citet{grimm:05} and Pietsch et
74: al. (2005) have provided source lists and fluxes for the X-ray source
75: population in M33, from the \chandra and XMM observations
76: respectively. In this paper we follow up on the \citet{grimm:05}
77: \chandra survey and present an analysis of the X-ray spectra and
78: variability behavior of X-ray sources in M33. Apart from fluxes,
79: variability and spectral energy distributions are an important
80: diagnostic tool for understanding the emission mechanism(s) in X-ray
81: sources, and for classifying these sources. Moreover, spectral and
82: time variability are the main characteristics of X-ray binaries, as
83: shown by the detailed work done for X-ray binaries in the Milky Way,
84: see e.g. \citet{klis:05}.
85:
86: The paper is organized as follows. In Sec. \ref{sec:data} we describe
87: the data processing, followed in Sec. \ref{sec:analysis} by the
88: analysis procedures. The results and their implications for the nature
89: of the X-ray source population in M33 are discussed in
90: Sec. \ref{sec:results}; we also discuss individual bright sources
91: in detail in Sec. \ref{sec:sources}. We conclude with a summary of the
92: results in Sec. \ref{sec:summary}. The figures showing the results for
93: the spectral analysis for sources with more than one observation are
94: shown in the Appendix in the electronic edition only.
95:
96: \section{Data}
97: \label{sec:data}
98:
99: M33 has been observed with the ACIS instrument on \chandra four
100: times, see \citet{grimm:05}. In this paper we use three observations
101: whose ObsIds and dates are given in Table \ref{tab:obs}. Due to its
102: angular extent the observed parts of M33 cover all active chips, the
103: standard ACIS-S configuration for \dataset[ADS/Sa.CXO\#obs/00786]{ObsId
104: 786}, and the standard ACIS-I configuration for
105: \dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730} and
106: \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}. There is considerable overlap
107: between the different observations. However, due to the decreasing
108: resolution/sensitivity with increasing off-axis angle only the inner
109: part of M33 ($\sim$ 8-10 arc minutes) has a significant number of
110: sources in two observations, \dataset[ADS/Sa.CXO\#obs/00786]{786} and
111: \dataset[ADS/Sa.CXO\#obs/01730]{1730}. A fourth observation, ObsId
112: 787, was disregarded because it was aimed at studying the nucleus and
113: suffers from both high background and small FOV.
114:
115: \begin{table}[h]
116: \begin{center}
117: \caption{List of ACIS observations of M33}
118: \begin{tabular}{l|c|r|c}
119: \hline
120: ObsId & Date & Aim point & Duration [ks]\\
121: \hline
122: \dataset[ADS/Sa.CXO\#obs/00786]{786} & 2000-08-30 & Nucleus & 45\\
123: \dataset[ADS/Sa.CXO\#obs/01730]{1730} & 2000-07-12 & Nucleus & 45\\
124: \dataset[ADS/Sa.CXO\#obs/02023]{2023} & 2001-07-06 & NGC 604 & 90\\
125: \hline
126: \end{tabular}
127: \label{tab:obs}
128: \end{center}
129: \end{table}
130:
131: The data from the three observations were processed according to the
132: standard data processing procedure with CIAO versions 3.1, including
133: exposure correction. Source detection was performed with {\em
134: wavdetect} with scales of 1, 2, 4, 8, 10, 12, and 16 in the energy
135: range 0.3--8 keV. The signal detection threshold was set to
136: $10^{-6}$. Source regions correspond to the 95\% encircled energy
137: area. For more details of the data analysis see \citet{grimm:05}.
138:
139: \section{Analysis}
140: \label{sec:analysis}
141:
142: We separate the analysis in three parts, short term,
143: long term, and spectral variability. The different methods are
144: discussed in the following subsections. We assume that the variability
145: of individual sources is independent, so there is no correlation
146: between short term and long term variability. Because the number of
147: counts observed in a single observation is not very large even for
148: bright sources we restrict the spectral variability analysis to a
149: comparison between different observations, when applicable.
150:
151: The nucleus was excluded from the following analysis because it
152: suffers from strong pile-up in two observations.
153:
154: \subsection{Short and long term variability}
155:
156: In order to establish short timescale variability we perform a
157: Bayesian block analysis of the lightcurves of individual sources. A
158: Bayesian block analysis computes the best approximation to the
159: lightcurve shape in terms of piecewise constant flux levels or
160: blocks. The discriminator between a single flux level for the whole
161: lightcurve or two flux levels is the ratio of likelihoods of
162: describing a data segment with one or two blocks. In this analysis the
163: algorithm used is iterative. It starts with the whole lightcurve and
164: subsequently divides it until the likelihood ratio for dividing a
165: lightcurve segment becomes smaller than a predefined prior, in
166: this case corresponding to $\sim$99\% confidence level. The Bayesian
167: block analysis is particularly suited for burst-like variability. For
168: more details about the principle of Bayesian blocks and the algorithm
169: used in this analysis see \citet{scargle:98}. The implementation used
170: here is the same as used by the CHAMP project and described in
171: \citep{kim:04}.
172:
173: Since a Bayesian block analysis uses only the photon arrival times
174: for computation of the blocks, no binning is necessary. Therefore
175: there is no intrinsic restriction to the number of photons the
176: algorithm is applicable to. Obviously establishing variability
177: with a certain confidence becomes less likely for fainter sources.
178:
179: Because a Bayesian block analysis is particularly sensitive to
180: burst-like variability we also performed a search for periodic
181: variability with the XRONOS v5.21 tool {\em efsearch}. This analysis
182: did not yield any source with significant short term variability.
183:
184: To establish long timescale variability we use simple Poisson
185: statistics. This is justified because, with the exception of two
186: foreground sources, no source exhibits strong short term
187: variability. We compare the difference in fluxes between two
188: observations to the quadratically added errors for the fluxes. If a
189: source is undetected in an observation in which it was in the field of
190: view of \chandra, we calculate an upper limit to the source flux using
191: an algorithm developed by \citet{kraft:91}. Since the upper limit
192: value already corresponds to the 99\% confidence level, for a
193: comparison with a detected flux we compare the 3 sigma error of the
194: detection directly with the upper limit.
195:
196: \subsection{Spectra and spectral variability}
197:
198: For 43 out of the total 261 sources the number of counts in at least
199: one observation was larger than 100 net counts, which we consider
200: sufficient to attempt spectral fitting. Of these 43 sources 25 have
201: been observed in at least two observations. For these sources we also
202: compare the spectral properties with time. The spectral fitting was
203: done with XSPEC v11.3.1. Because of the generally low number of counts
204: the fitting was done for all sources with Cash statistics
205: \citep{cash:79}. At 100 counts we expect only $\sim$3 background
206: sources based on the CDF-N LogN--LogS \citep{alexander:03}, so the
207: vast majority of these sources are likely to belong to M33.
208:
209: We first fit the bright sources with two simple absorbed spectral
210: models, power law and bremsstrahlung, to check the validity of our
211: assumptions about the spectral shape of the faint sources that do not
212: have sufficient counts for detailed spectral modeling (see derivation
213: of the X-ray luminosity function in \citet{grimm:05}). The absorption
214: was in one case a free parameter in the other case it was fixed to the
215: Galactic value. After validating that the assumptions about the
216: general power law shape are correct, we proceeded to fit more complex
217: model to the bright sources.
218:
219: The results of the simple spectral fitting are presented in
220: Sec. \ref{sec:results}. Spectra of sources with more than one
221: observation with best fit values and corresponding confidence contour
222: plots are shown in Appendix \ref{sec:spectra} in the electronic
223: edition, except for the spectra of M33 X-7 (\object[]{CXO
224: J013334.1+303210}) and M33 X-9 (\object[]{CXO J013358.8+305004}),
225: which are shown in Figs. \ref{fig:x7} and \ref{fig:x9}, and will be
226: discussed in Sec. \ref{sec:sources}.
227:
228: Based on the confidence regions for the fit parameters, shown in
229: Appendix \ref{sec:spectra}, only one of the X-ray sources shows
230: significant spectral variability from observation to observation, M33
231: X-4 (\object[]{CXO J013315.1+305317}). For the other sources in different
232: observations none of the parameters of a source spectrum show changes
233: corresponding to more than 99\% confidence level. This is despite the
234: fact that some sources show significant time variability between
235: observations. However, the number of counts available for most sources
236: for fitting is insufficient to determine spectral parameters to an
237: accuracy good enough to compare two observations. Moreover,
238: degeneracies between model parameters complicate the establishing of
239: variability.
240:
241: Another approach to spectral variability is the use of hardness
242: ratios. These are relatively crude estimators of spectral changes but
243: because of the smaller number of degrees of freedom compared to a
244: spectral fit can be statistically preferable. We therefore take all
245: sources that were observed in at least two observations, divide each
246: observation in three equally long parts, and construct hardness ratios
247: from these intervals. Because only sources with more than 100 counts
248: are used in this part of the analysis the separation in three parts
249: still gives sufficient number of counts ($\ge 20-30$) in each time bin
250: for a hardness ratio analysis \citep{prestwich:03}. Note that
251: \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023} is roughly twice as long
252: as the other observations. The energy bands used are the same as in
253: \citet{grimm:05}; 0.3--1.0 keV for the soft band, 1.0--2.1 keV for the
254: medium band, and 2.1--8.0 keV for the hard band. The hardness ratios
255: are defined as
256: \begin{equation}
257: HR1 = \frac{M-S}{T},\hspace{1cm}
258: HR2 = \frac{H-M}{T}
259: \end{equation}
260: where $S$, $M$, and $H$ are the background and exposure corrected
261: counts in the soft, medium, and hard band, and $T$ are the corrected
262: counts in the whole energy band. To compare two observations
263: it is important to take into account the different effective areas of
264: the source location in each observation, especially if the source is
265: located on a front-illuminated chip and in another observation on a
266: back-illuminated chip. We compute the effective area of the source in
267: each observation for each energy band. For comparison with the other
268: observations of the source we normalize the effective area in each
269: band by the effective area value of the aim point of
270: \dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730}. Note that the choice of
271: the normalization constant is arbitrary. The results are combined for
272: all available observations. The results for all sources that show
273: evidence of variability are shown in Appendix \ref{sec:spectra} in the
274: electronic edition.
275:
276: \section{Results and Discussion}
277: \label{sec:results}
278:
279: \subsection{Short timescale variability}
280:
281: Except for two sources which are foreground Galactic stars, no source
282: presents variability above the 99\% confidence level. In
283: Fig. \ref{fig:b1} we show the results of the Bayesian block analysis
284: for the only two sources that are variable on short timescales. The
285: dotted histograms represent the count rate binned in 400 second
286: intervals. Note that the binning is for plotting purposes only. The
287: green line is the power spectrum of the lightcurve. The thick solid
288: line is the result of the Bayesian block analysis. The small square to
289: the right of the lightcurve is a thumbnail image of the source from
290: the observation.
291:
292: Source {\bf \object[]{CXO J013327.7+304645}} shows a clear X-ray outburst in
293: \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}. This outburst, lasting
294: about 10,000 seconds completely accounts for the long term variability
295: of the source. The lightcurve of the outburst is well fit by an
296: exponential decay with a decay timescale of 3900 seconds. The
297: persistent luminosity in \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}
298: is consistent within the errors with the luminosities in the other
299: observations. Outburst and persistent counts are not large enough to
300: investigate changes in spectral shape. An X-ray color comparison of
301: the burst and persistent emission shows a softening of the emission in
302: HR2 and a slight hardening in HR1 that is significant only at the
303: 2-sigma level. The optical counterpart to source \object[]{CXO
304: J013327.7+304645} is a star in the USNO catalog with a V magnitude of
305: 16.8. We have analyzed archival HST WFPC2 data. The object was
306: detected in bands corresponding to U, B, V, and I. The colors are
307: consistent with an early M type star, either on the main sequence or a
308: giant. Using normal V magnitudes for M stars this puts the star at a
309: distance of 90--320 pc for a main sequence star, or 24--29 kpc for a
310: giant. Considering the relatively high galactic latitude of M33,
311: $b=-31$ deg, the lower distance value, and thus a main sequence star
312: is more likely. Assuming a main sequence star the peak luminosity is
313: between $\sim10^{29}-1.5\times 10^{30}$ ergs s$^{-1}$, and the total
314: energy between $\sim10^{32}-1.5\times 10^{33}$ ergs.
315:
316: \bfige[h]
317: \resizebox{0.5\hsize}{!}{\includegraphics[angle=-90]{f1a.eps}}
318: \resizebox{0.5\hsize}{!}{\includegraphics[angle=-90]{f1b.eps}}
319: \caption{{\em Left}: Lightcurve and Bayesian block division of the
320: outburst of CXO J013327.7+304645, an X-ray active
321: star. The panel shows the lightcurve binned in 400 second intervals
322: (dashed histogram). Note that the lightcurve binning is for plotting
323: purposes only. The Bayesian blocks are shown as the thick black
324: histogram. The right part of the panel shows a thumbnail picture of
325: the source.{\em Right}: Lightcurve and Bayesian block division of
326: the outburst of CXO J013341.8+303848, another X-ray
327: active star.}
328: \label{fig:b1}
329: \efige
330:
331: Source {\bf \object[]{CXO J013341.8+303848}}, a foreground star as
332: well, has two outbursts in \dataset[ADS/Sa.CXO\#obs/00786]{ObsId
333: 786}. The first outburst lasts about 2600 seconds, the second outburst
334: 18000 seconds later lasts about 800 seconds. The persistent level of
335: X-ray emission increases slightly after each outburst. As shown in the
336: long term lightcurve the source also has a long term trend to
337: increasing luminosity for the 3 observations. The decay of the
338: outbursts cannot be fit uniquely; an exponential decay with a decay
339: constant of $\sim$1000 seconds or a linear decay are both
340: possible. During both outbursts the source becomes softer, but only at
341: the $1-2\sigma$ level. This source has no HST coverage.
342:
343: \subsection{Long term variability}
344:
345: Of a total of 261 sources, 198 have been detected in at least
346: two observations, and 62 in all three observations. The luminosities
347: for comparison are taken from spectral fits for the brighter sources
348: (more than $\sim$100 counts), or from the assumed spectrum, absorbed
349: power law with a photon index of 2 and Galactic absorption. We find
350: that 49 of 198 sources show variability between observations above 3
351: sigma. An additional 29 sources show signs of variability between 2
352: and 3 sigma. Of the 49 variable sources only 16 are apparently
353: persistent.
354:
355: The strength of the variability can in fact be used to confirm the
356: nature of most of these sources as X-ray binaries. AGN in general show
357: variability on weeks or months timescale only up to factors of 2--3
358: \citep{mushotzky:93,paolillo:04}. Most of the sources in M33 show
359: stronger variability, particularly since the flux ratios for most
360: sources are only lower limits. Moreover, the sources with flux ratios
361: below 2 are bright sources that are unlikely to be AGN based only on
362: brightness.
363:
364: Taking into account that $\sim$80--90 of the 198 sources are likely
365: background AGN, somewhat less than half the sources in M33 are
366: variable. Of these two thirds (34/49) may be candidate
367: transients. This number is somewhat higher than the about 50\% in the
368: Milky Way and Magellanic Clouds \citep{liu:00,liu:01}. But the
369: detection limits (few $10^{34}$ erg s$^{-1}$) do not allow to
370: establish true transient behavior for undetected sources so the number
371: of two thirds is only an upper limit.
372:
373: For the brighter sources ($L_{X} > 5\times10^{36}$ erg s$^{-1}$) past
374: X-ray satellites also provide data for long term variability.
375: Fig. \ref{fig:long} shows data from Einstein HRI and IPC
376: \citep{trinchieri:88}, ROSAT HRI \citep{schulman:95} and PSPC
377: \citep{long:96}, BeppoSAX \citep{parmar:01} as well as from this work
378: for the bright sources in the field of view of \chandra. The quoted
379: luminosities are converted to the \chandra band of 0.3--8.0 keV and,
380: if they are not individual measurements of spectra, converted to the
381: spectral shape of a power law with photon index of 2 and the Galactic
382: absorption value towards M33, $6\times10^{20}$ cm$^{-2}$. The error
383: bars in the plot only contain errors due to counting statistics. Other
384: errors, e.g. due to the conversion of energy bands and different
385: assumed spectral shapes, add generally another 20--30\% uncertainty in
386: the luminosity. In addition different instrument responses,
387: cross-calibration issues and other systematic effects add another
388: source of systematic errors, which is hard to quantify. An exception
389: is the supernova remnant {\bf M33 X-14}, for which converting the
390: luminosity given by \citet{long:96} would result in variability at
391: over $6\sigma$ compared to other observations. However,
392: \citet{long:96} assume a power law spectrum with a photon index of 2
393: to compute the source luminosity. Although this is a reasonable choice
394: for X-ray binaries or AGN, the spectrum of M33 X-14 is quite soft. We
395: extracted a spectrum of X-14 from the ROSAT observation rp600023a00,
396: the longest of the PSPC observations with $\sim$29 ks. The spectrum is
397: well fit by an absorbed blackbody model with a column density of
398: $4\times 10^{21}$ cm$^{-2}$ and a temperature of 0.09 keV. Although
399: this model is rather unphysical for a SNR we are interested only in
400: the flux, which is sufficiently accurate for the purpose of
401: comparison. The ROSAT data also agree with the spectrum obtained from
402: \chandra. Extrapolating this spectrum to the \chandra energy range
403: gives a luminosity of $3.7\times 10^{36}$ ergs s$^{-1}$ which agrees
404: very well with other observations of the source, in particular the
405: \chandra observed value of $3.6\times 10^{36}$ ergs s$^{-1}$.
406:
407: It is clear from Fig. \ref{fig:long} that four of the bright X-ray
408: sources are variable. The sources that do not show evidence of
409: variability at the $3\sigma$ level are M33 X-1, X-2, and the SNR X-14.
410:
411: \bfige
412: \resizebox{\hsize}{!}{\includegraphics[angle=-90]{f2.eps}}
413: \caption{Long term lightcurves of bright X-ray sources in M33
414: observed with Einstein, ROSAT, BeppoSAX, and \chandra.}
415: \label{fig:long}
416: \efige
417:
418: The variability pattern is very similar to that observed in other
419: X-ray source populations, regardless of galaxy type, e.g. The
420: Antennae \citep{zezas:06}, M101 \citep{jenkins:05}, NGC 4697
421: \citep{sivakoff:05}. In particular that only very few, if any, sources
422: show short-term variability is very common, although this is most
423: likely due to limited photon statistics. On the other hand a
424: large fraction ($\sim$10-40\%) of sources exhibit long-term
425: variability. Moreover, the fraction of sources with strong long term
426: variability is quite confidently identified as X-ray binaries, and not
427: background AGN.
428:
429: \subsection{Spectra}
430:
431: Fitting all the 43 sources detected with more than 100 net counts with
432: a power law and a thermal bremsstrahlung spectrum, with column density
433: fixed to the Galactic value or as a free fit parameter, provides about
434: half of the sources with a good fit to the data. The other sources
435: require either different models, e.g. black body or other thermal
436: plasma models, or multiple components. The results of the simple power
437: law/bremsstrahlung fits support the validity of our assumptions of a
438: fixed spectral model for conversion from counts to fluxes for fainter
439: sources, assuming that the spectral properties of bright and faint
440: sources are not systematically different \citep{grimm:05}.
441:
442: In Fig. \ref{fig:spec_prop} we show histograms for the best fit power
443: law and bremsstrahlung values for all sources. The upper left panel
444: shows the comparison between photon indices $\Gamma$ of a power law
445: with column density being a free fit parameter versus column density
446: fixed to the Galactic value. The Galactic absorption towards M33 is
447: only $6 \times 10^{20}$ cm$^{-2}$. The upper right panel shows the
448: same for bremsstrahlung temperature $kT$. The lower left panel shows
449: the histogram of $\Gamma$ for the case of $N_H$ as a free fit
450: parameter. The power law slopes are concentrated around 2, the
451: canonical value for X-ray binaries and AGN. The peak at $\Gamma = 5$
452: comprises all sources with photon index larger than or equal to
453: 5. Individually, all these sources are fit well with a blackbody or
454: plasma model with temperatures of 0.1--0.13 keV (blackbody) or
455: 0.2--0.3 keV (plasma model). The lower right panel finally shows the
456: distribution of bremsstrahlung temperatures. Note that the influence
457: of $N_H$ is stronger in case of a power law than for a bremsstrahlung
458: spectrum as is evident in the generally smaller deviations from the
459: one-to-one correlation for the bremsstrahlung temperatures compared to
460: the photon indices in the upper panels.
461:
462: \bfige
463: \resizebox{0.5\hsize}{!}{\includegraphics{f3a.eps}}
464: \resizebox{0.5\hsize}{!}{\includegraphics{f3b.eps}}
465: \resizebox{0.5\hsize}{!}{\includegraphics{f3c.eps}}
466: \resizebox{0.5\hsize}{!}{\includegraphics{f3d.eps}}
467: \caption{Overall spectral properties of the sample of 43 bright
468: X-ray sources in M33. The panels show the correlations of photon
469: index $\Gamma$ with $N_H$ free versus fixed (upper left),
470: bremsstrahlung temperature $kT$ with $N_H$ free versus fixed
471: (upper right). Note that the influence of $N_H$ is stronger in case
472: of a power law than for a bremsstrahlung spectrum as evident in the
473: smaller deviations from the one-to-one correlation for the
474: bremsstrahlung temperature. The lower panels show the distributions
475: of $\Gamma$ (lower left) and $kT$ (lower right) for the case of
476: $N_H$ as a free fit parameter. The power law slopes are mainly in
477: the range from 1.4 to 2.5, the canonical values for X-ray binaries and
478: AGN.}
479: \label{fig:spec_prop}
480: \efige
481:
482: As expected, fits with a fixed low column density produce smaller
483: photon indices or larger bremsstrahlung temperatures, respectively.
484: However, changes of temperature and photon index between fits with
485: fixed and free column density are not significantly larger than the
486: errors in the majority of cases. Comparing the difference between
487: photon indices divided by the square root of the errors shows that
488: $\sim$80\% of the sources have values of less than 3. For thermal
489: bremsstrahlung model the corresponding value is 73\%. Thus the
490: assumption of a general power law spectrum with $\Gamma = 2$, and
491: Galactic absorption is quite good \citep{grimm:05}.
492:
493: \bfige[t]
494: \resizebox{0.32\hsize}{!}{\includegraphics{f4a.eps}}
495: \resizebox{0.32\hsize}{!}{\includegraphics{f4b.eps}}
496: \resizebox{0.32\hsize}{!}{\includegraphics{f4c.eps}}
497: \caption{The left panel shows the best fit photon indices for
498: sources that are well fit by a single power law (open circles) or
499: a multi-component spectrum containing a power law (open circles with
500: filled triangles) versus luminosity. Two thirds of the single power
501: law sources are also well fit by a disk black body model. The
502: middle panel shows the disk black body temperatures versus
503: luminosity for sources that are well fit by a single power law,
504: but also by a simple disk black body model (filled circles). As
505: expected for X-ray binaries the disk temperatures are in the range
506: from 1--3 keV. The temperatures are also in the same range as the
507: temperatures for sources that require a disk black body component in
508: the spectrum (open triangles). The right panel shows temperatures of
509: sources that are either well fit by a bremsstrahlung model (open
510: squares) or a black body model (filled triangles) versus
511: luminosity. With the exception of two spectra, the bremsstrahlung
512: sources are also well fit by a thermal plasma model, XSPEC model
513: {\it apec}, as discussed in the text. Errors are 90\% confidence
514: level.}
515: \label{fig:slopes}
516: \efige
517:
518: Using the results of the more detailed spectral fitting of the 43
519: bright sources, we can confirm the results of the relatively blind
520: spectral fitting of all sources with a power law model. The left panel
521: of Fig. \ref{fig:slopes} shows the photon indices of spectra that are
522: well fit by either a single power law (open circles) or a combination
523: of a power law and other components (filled triangles) versus
524: luminosity. Errors are 90\% errors on the slope. There is a clear
525: trend that photon indices in multi-component fits are larger than for
526: single power law fits, and there is an indication that brighter
527: sources are more likely well fit by multi-component fits than by
528: single power laws. The reason is most likely that the single power law
529: sources have lower counts and multiple components are not
530: distinguishable. We also fit the single power law sources with a disk
531: black body model, XSPEC model {\it diskbb}, which has the same number
532: of degrees of freedom than the power law model. Two thirds of the
533: spectra are well fit by a disk black body model, the best fit
534: temperatures being in the range from 1 to 3 keV, as expected for X-ray
535: binaries \citep{tanaka:01}. The inner disk temperatures versus
536: luminosity are shown in the middle panel of Fig. \ref{fig:slopes}. The
537: values are the same as for the sources that require disk black bodies
538: for a good fit. This suggests that the single power law sources are
539: likely to be X-ray binaries and only the low number of counts allows a
540: good fit with a single power law.
541:
542: There are 10 spectra (5 sources) that are best fit with a single
543: bremsstrahlung model with temperatures of $\sim$0.1--0.3 keV. The
544: bremsstrahlung temperatures versus luminosity are shown in the right
545: panel of Fig. \ref{fig:slopes} (open squares). Since these values are
546: unusually low for real bremsstrahlung sources, and the luminosities of
547: the sources relatively low, $1-5 \times 10^{36}$ erg s$^{-1}$, we
548: fit these spectra with a thermal plasma model (XSPEC model {\it apec})
549: as well. With one exception we obtain good fits with the plasma model
550: as well, the temperatures are in the range from 0.1 keV to 0.4 keV,
551: and metallicities from 0.05 to 0.3 solar with considerable
552: uncertainties. These values are at or below the expected metallicity
553: for M33 \citep{blair:85}. The good fit quality is not surprising
554: considering that the {\it apec} model has one degree of freedom more
555: than the simple bremsstrahlung model. However, given the low
556: temperatures and the consistent metallicities we consider the {\it
557: apec} model to be the more physical model for these sources. The
558: sources with good fits for the {\it apec} model are marked in Table
559: \ref{tab:spectra} (Appendix \ref{sec:spectra}) and the {\it apec}
560: model parameters are given, in Fig. B1 (Appendix
561: \ref{sec:spectra}) the contour plots for these fits are shown as
562: well. Also shown in the right panel of Fig. \ref{fig:slopes} are six
563: sources well fit by a black body model with very low temperatures at
564: or below 0.1 keV (filled triangles). For two sources the black body is
565: the only component in the spectrum. These sources are candidates for
566: super-soft sources and are discussed below in Sec. \ref{sec:sources}.
567: Three of the other four sources have an additional power law
568: component, that ranges from hard ($\Gamma\sim1.3$) to very soft
569: ($\Gamma\sim4.6$). The black body temperatures are even lower than the
570: temperatures inferred from ULX intermediate mass black hole
571: candidates, see e.g. \citet{miller:04}. A truncated disk would be a
572: possibility to explain the low temperature, similar to the scenario
573: suggested by \citet{kubota:04}. However, given the luminosities the
574: sources would be in the hard state (e.g. \citet{maccarone:03}) but,
575: except for \object[]{CXO J013444.6+305535}, the photon indices are
576: larger than 2. \object[]{CXO J013444.6+305535} could indeed be in the
577: hard state with a photon index of 1.3 but that number is not well
578: constrained. This hard source could also be a candidate for a magnetic
579: CV that are known to have hard spectra and relatively high
580: luminosities \citep{kuulkers:06}. The soft power law in the other
581: sources might on the other hand be another thermal
582: component. Alternatively, the soft emission might be generated in an
583: outflow from the system, or heating of a surrounding medium.
584:
585: Fig. \ref{fig:nh_comp} shows a comparison of measured values for
586: column density overlaid on a 1.49 GHz contour map of the central part
587: of M33 from VLA \citep{condon:87}. The resolution of the VLA map is
588: about 1 arcminute, and the confusion limit is given as 0.1 mJy. There
589: is no spatial correlation between the contour map and the magnitude of
590: the measure column densities. Thus the X-ray absorption value is in
591: parts due to location of sources in front of/behind HI gas, and
592: in parts due to intrinsic absorption around the X-ray source.
593:
594: \bfige
595: \resizebox{\hsize}{!}{\includegraphics[]{f5.eps}}
596: \caption{Measured values for column density overlaid on a 1.49 GHz
597: contour map of the central part of M33 from VLA \citep{condon:87}.
598: The resolution of the VLA map is about 1 arcminute, and the RMS
599: is about 0.1 mJy. Prominently visible is the HII region NGC 604 at
600: the upper left. Also visible towards south of the center is the
601: southern spiral arm. There is no correlation between the contour
602: map and the magnitude of the measure column densities.}
603: \label{fig:nh_comp}
604: \efige
605:
606: To compare actual values for the column density due to HI, we compute
607: the brightness temperature of the HI gas according to
608: \be
609: S_{\nu} = \frac{2k\nu^{2}\Omega}{c^2}T_b,
610: \ee
611: with $S_{\nu}$ the radio flux in units of Jy/steradian at frequency
612: $\nu$ (1.49 GHz), $\Omega$ the opening angle of the beam (1'), and
613: $T_b$ the brightness temperature; $k$ is the Boltzmann constant, $c$
614: the speed of light. Assuming optically thin emission we compute the HI
615: column density according to
616: \be
617: N_H = 1.83 \times 10^{18} T_b [\rm{\,K\, cm}^{-2}].
618: \ee
619: Within the size of a beam the radio flux has an RMS of $\sim$5 mJy
620: around the zero point. Therefore we take 5 mJy as the upper limit on
621: the sensitivity in a beam. This flux corresponds to $\sim$320 K or
622: $N_H$ of $5.7\times 10^{20}$ cm$^{-2}$, which is very close to the
623: Galactic absorption value. Fig. \ref{fig:nh_nh} shows the comparison
624: between the X-ray absorption values from spectral fitting and the
625: ``expected'' absorption from HI.
626: \bfige
627: \resizebox{\hsize}{!}{\includegraphics[]{f6.eps}}
628: \caption{Measured values for column density from X-ray spectral
629: fitting versus absorption values inferred from HI radio
630: flux. Upper limits are shown as arrows. Vertical and horizontal
631: lines denote the Galactic absorption value. Filled circles are
632: sources in M33, crosses are Galactic X-ray binaries, see text. The
633: plot can be separated in three regions: 1. At/below the horizontal
634: line, sources with apparently {\it less} absorption than the
635: Galactic value. 2. Below the diagonal line, sources are probably
636: located in front of at least parts of the HI emission. 3. Above the
637: diagonal line, sources are also intrinsically absorbed.}
638: \label{fig:nh_nh}
639: \efige
640: Filled circles are sources in M33, crosses are a selection of Galactic
641: X-ray binaries from \citet{vrtilek:91}. Upper limits are shown as
642: arrows. The vertical and horizontal lines denote the Galactic
643: absorption value. The figure can be separated in roughly three
644: regions: 1. Sources at or below the horizontal line, i.e. sources with
645: apparently {\it less} absorption than the Galactic value, which
646: probably have an undetected thermal component in the spectrum.
647: 2. Sources below the diagonal line which are probably located in front
648: of at least parts of the HI emission. These sources are most likely on
649: the Milky Way facing side of M33. 3. Sources above the diagonal line
650: are intrinsically absorbed. No conclusion as to their location can be
651: drawn.
652:
653: Values for Galactic X-ray binaries are taken from \citet{vrtilek:91}
654: and \citet{liu:01} for $A_V$ values missing in the former. We convert
655: the optical extinction to neutral hydrogen column density according to
656: \citet{predehl:95} $N_H = 1.77 \times A_V$ in units of $10^{21}$
657: cm$^{-2}$, and use $A_V = 3\times E(B-V)$ in cases where only the
658: reddening in available. The comparison with Galactic X-ray binaries
659: shows that M33 X-ray sources also often show excess absorption at
660: levels comparable with Galactic X-ray binaries. The larger values for
661: $N_H$(HI) in the Milky Way can be explained simply by our location in
662: the Milky Way, compared to the nearly face-on orientation of M33.
663:
664: Only six of the 25 sources observed in more than one observation with
665: more than 100 counts show significant variability in the hardness
666: ratio diagram, see Appendix \ref{sec:spectra}. Three of the sources
667: are the well known sources, M33 X-4, M33 X-5, and M33 X-13. The other
668: three are \object[]{CXO J013329.0+304216}, \object[]{CXO
669: J013329.3+304508}, and \object[]{CXO J013410.5+303946}. There is a
670: clear separation in that most sources have a very small significance
671: for moving in the hardness ratio diagram, whereas four of the ones
672: that show changes in hardness ratio are above 99.99\% confidence
673: level.
674:
675: Based on spectral shape (photon indices, temperatures) and absorption
676: properties we conclude that M33 X-ray sources as a population are very
677: similar to Galactic X-ray binaries.
678:
679: \subsection{Individual sources}
680: \label{sec:sources}
681:
682: Here we discuss the properties of individual bright M33 sources. These
683: include previously known sources (M33 X-...) and sources newly
684: discovered in our \chandra observations. Wide band spectral studies
685: exist in the literature for some of these sources. In particular,
686: BeppoSAX has observed the brightest sources in M33 and measured
687: spectra of the sources M33 X-4 to M33 X-10 in the energy range from
688: 2.0-10.0 keV \citep{parmar:01}. Of these sources X-6 and X-10 are not
689: in the field of view of the \chandra observations, and X-8 is the
690: nucleus which is not discussed here.
691:
692: {\bf M33 X-4 (\object[]{CXO J013315.1+305317})}: The X-ray spectrum of
693: this soft source measured by BeppoSAX is well fit by either a power
694: law model with photon index of $\sim$3 or a bremsstrahlung model with
695: $kT\approx2$ keV. This source has been observed by \chandra twice, in
696: ObsIds \dataset[ADS/Sa.CXO\#obs/00786]{786} and
697: \dataset[ADS/Sa.CXO\#obs/02023]{2023}. Between the observations, the
698: flux increased by a factor of two from $5\times 10^{37}$ erg s$^{-1}$
699: to $10^{38}$ erg s$^{-1}$. The variability rules out the
700: interpretation of the source as a SNR \citep{okada:01}. Using the
701: overlapping energy range with the BeppoSAX observations, 2.0--8.0 keV,
702: we obtain the same values for absorption, photon index, and
703: bremsstrahlung temperature well within the errors and $\chi^2$-values
704: below 1 for observation \dataset[ADS/Sa.CXO\#obs/00786]{786}. For observation
705: \dataset[ADS/Sa.CXO\#obs/02023]{2023}, although the best fit values are
706: consistent with the BeppoSAX and \dataset[ADS/Sa.CXO\#obs/00786]{ObsId
707: 786} values, the $\chi^2$-values are well above 2. Moreover, for the
708: whole \chandra range, 0.3--8.0 keV, none of these models is a good
709: fit. A statistically satisfactory fit for both ObsIds can be achieved
710: with a model consisting of a black body and a power law. The black
711: body temperatures ($\sim$0.65 keV) and photon indices (3-4.7) for the
712: two observations are consistent within the errors. However, the
713: absorption to the source changes. Taking into account the different
714: values for absorption there is a clear distinction at 99\% confidence
715: level in the contour plots, for $\Gamma$ versus $N_H$ and kT versus
716: $N_H$. The difference in $\Gamma$ versus kT is significant only at
717: 90\% confidence level. The black body temperature is relatively low,
718: and does not change significantly, but similar values for disk black
719: body temperatures have been observed in Galactic BHs. These
720: temperatures occur at luminosities of $\sim$1--5\% of the Eddington
721: luminosity \citep{gierlinski:04}. With a luminosity increasing from
722: $\sim5\times 10^{37}$ erg s$^{-1}$ to $10^{38}$ erg s$^{-1}$, this
723: interpretation would make the compact object in X-4 a rather massive
724: but still stellar mass BH given the luminosity. The photon index of
725: the power law component is much softer than expected for a low-hard
726: state black hole \citep{tanaka:01} but is in agreement with high-soft
727: state Galactic black holes \citep{mcclintock:06}. The change in the
728: photon index between the observations (from $\sim$3 to 4.7) does not
729: indicate a state change from low-hard to high-soft. Even a photon
730: index of 3 is too large for the low-hard state ($\Gamma \sim$1.7). The
731: photon index is also in the range observed for the Galactic XRB Sco
732: X-1 which has a neutron star primary \citep{bradshaw:03}, but Sco X-1
733: does not exhibit a soft component like M33 X-4 \citep{kahn:84}. Thus
734: based on the X-ray data the source is likely to be a BH binary.
735:
736: The hardness ratio analysis shows that the hardness of the source
737: varies significantly between the observations. The source becomes
738: harder in HR1 and softer in HR2, while the luminosity increases by a
739: factor of $\sim$2. This change is due to the increase in the medium
740: band (1.0--2.1 keV) with respect to the soft and hard band, which is
741: reflected in the spectral fits by a higher absorption column
742: and a lower power law photon index.
743:
744: {\bf M33 X-5 (\object[]{CXO J013324.4+304401})}: This source has been
745: observed in all three \chandra observations. For M33 X-5 (\object[]{CXO
746: J013324.4+304401}) BeppoSAX as well finds good fits with a soft power
747: law ($\Gamma\approx3$) or bremsstrahlung model ($kT\approx2.8$
748: keV). In all \chandra observations the overlapping energy range is fit
749: satisfactorily with these values. However, for the whole \chandra
750: energy range neither a single power law nor a bremsstrahlung spectrum
751: give good fits. A good fit is obtained with a disk black body
752: ($\sim$1.3 keV) and power law ($\Gamma= 4-9$) combination. The contour
753: plots in Fig. \ref{fig:spectra} (Appendix \ref{sec:spectra}) show no
754: significant variation of the spectral parameters given the large
755: uncertainties. The errors on the photon index are large enough to not
756: make the changes between observations significant. However, a disk
757: black body alone does not yield a good fit. The disk black body
758: temperature is in good agreement with observations of Galactic BH
759: binaries at moderately sub-Eddington luminosities
760: \citep{gierlinski:04}. The luminosity of the sources changes between
761: $3\times 10^{37}$ erg s$^{-1}$ and $5\times 10^{37}$ erg s$^{-1}$;
762: thus it is significantly below the Eddington limit for a black hole
763: but in the range of transition luminosities from the low-hard to the
764: high-soft state of Galactic black holes \citep{maccarone:03}. Another
765: interpretation is that the source is a neutron star and the steep
766: power law corresponds to the harder thermal component from the neutron
767: star surface \citep{tanaka:01}. Due to the complexity of the combined
768: model only the disk temperature is well determined.
769:
770: However, the hardness ratio analysis shows significant variation between
771: ObsIds \dataset[ADS/Sa.CXO\#obs/01730]{1730} and
772: \dataset[ADS/Sa.CXO\#obs/00786]{786} on the one hand and
773: \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023} on the other. X-5 becomes
774: significantly softer in HR1 but does not change significantly in
775: HR2. There seems to be no correlation with luminosity as ObsIds
776: \dataset[ADS/Sa.CXO\#obs/00786]{786} and
777: \dataset[ADS/Sa.CXO\#obs/02023]{2023} have the same luminosity but
778: significantly different HR1 values.
779:
780: {\bf M33 X-7 (\object[]{CXO J013334.1+303210})}: This source has been
781: observed only in \chandra observation
782: \dataset[ADS/Sa.CXO\#obs/01730]{1730}. For M33 X-7 (\object[]{CXO
783: 013334.1+303210}) BeppoSAX as well finds good fits with a power law
784: ($\Gamma\approx2-3$) or bremsstrahlung model ($kT\approx3.7$ keV). In
785: the 2.0--8.0 keV energy range both models provide satisfactory (power
786: law) to good (bremsstrahlung) fits to the \chandra data. The photon
787: indices for fixed and variable absorption are larger than in the
788: BeppoSAX observations but still within the rather large errors. The
789: bremsstrahlung temperatures for fixed and variable absorption are
790: smaller, and in the case of fixed absorption inconsistent with the
791: lower limit of the BeppoSAX observation. Although the errors are quite
792: large this indicates that the source was softer in the \chandra
793: observations. Moreover, M33 X-7 was about five times brighter in the
794: \chandra observation ($\sim7\times 10^{37}$ erg s$^{-1}$) compared to
795: the time BeppoSAX observed the source ($\sim1.6\times 10^{37}$ erg
796: s$^{-1}$) (note that both luminosities are based on spectral
797: fits). The only satisfactory fit to the whole energy range of \chandra
798: is an absorbed disk blackbody with an inner disk temperature of $1\pm
799: 0.02$ keV. This value is consistent with inner disk temperatures of
800: Galactic black holes \citep{mcclintock:06}. The column density is not
801: well constrained, but below $6.5\times 10^{20}$ cm$^{-2}$, putting the
802: upper limit of the absorption at the Galactic value. \citet{pietsch:04}
803: obtain the same values within the errors for an XMM observation, and
804: also new \chandra observations give the same spectral parameters
805: \citep{pietsch:06}. This might indicate the presence of an additional
806: soft component that due to insufficient counts presents itself as a
807: low absorption value. An additional soft component, however, does not
808: improve the fit, and its parameters are not well
809: determined. Fig. \ref{fig:x7} shows the spectrum and the contour plot
810: of absorption column density versus inner disk temperature. Given the
811: softer spectrum in the \chandra observation, a possible additional
812: soft component, and the five times higher luminosity, it is possible
813: that X-7 underwent a state transition. On the other hand the photon
814: index of the power law in the BeppoSAX spectrum is already softer than
815: expected for a low-hard state source \citep{tanaka:01}.
816:
817: \bfige
818: \resizebox{0.5\hsize}{!}{\includegraphics[angle=-90]{f7a.eps}}
819: \resizebox{0.5\hsize}{!}{\includegraphics[angle=-90]{f7b.eps}}
820: \caption{Energy spectrum and contour plot for M33 X-7. The model spectrum
821: is an absorbed disk blackbody with an inner disk temperature of
822: $1\pm 0.02$ keV and $N_H < 6.5\times 10^{20}$ cm$^{-2}$. The
823: reduced $\chi^2$ is 1.2.}
824: \label{fig:x7}
825: \efige
826:
827: {\bf \object[]{CXO J013343.4+304630}} and {\bf \object[]{CXO
828: J013409.9+303219}}: These sources are the best candidates for
829: super-soft sources (SSS) in M33. \object[]{CXO J013343.4+304630} is
830: well fit by a black body spectrum with a temperature of about 75 eV
831: and moderate absorption (N$_H \sim 8\times 10^{21}$ cm$^{-2}$). The
832: source shows no significant spectral variability, but is variable on
833: timescales of months. \object[]{CXO J013409.9+303219} has been
834: observed only once. Its spectrum is also well fit by a black body with
835: a temperature of 37 eV and also moderate absorption (N$_H \sim
836: 7\times10^{21}$ cm$^{-2}$). The temperatures are in the range for
837: SSSs, and the sources are not significantly detected above 1 keV, see
838: e.g. \citet{kahabka:97}. Given their luminosities of $\sim3-10\times
839: 10^{35}$ erg s$^{-1}$ (\object[]{CXO J013343.4+304630}) and
840: $\sim3\times 10^{36}$ erg s$^{-1}$ (\object[]{CXO J013409.9+30321})
841: these sources are most likely nuclear burning white dwarfs.
842:
843: {\bf M33 X-9 (\object[]{CXO J013358.8+305004})}: This source was observed in
844: \chandra observations \dataset[ADS/Sa.CXO\#obs/00786]{786} and
845: \dataset[ADS/Sa.CXO\#obs/02023]{2023}. It is well fit by a power law
846: with photon index of $\sim$1.2 in the BeppoSAX observation. For a
847: bremsstrahlung model only lower limits to the temperature are provided
848: ($>2.6$ keV). It is important to note that in the BeppoSAX
849: observations X-9 is unresolved, but ROSAT observations have shown that
850: X-9 actually consists of 3 sources, at least two of which seem to be
851: variable and of comparable flux \citep{long:96}. Since BeppoSAX does
852: not resolve the sources, it is impossible to say if any of them
853: dominate the spectrum and if yes which one. The \chandra source is
854: associated with the source X-9a based on the ROSAT position. In the
855: whole \chandra energy range a power law is a good fit. The absorption
856: column is not well constrained in this fit, but the values for the
857: photon index are $\sim1.7$, with a reduced $\chi^2$ of
858: 0.87. Fig. \ref{fig:x9} shows the spectrum and the contour plot of
859: absorption column density versus photon index. Assuming for the sake
860: of the argument that a significant fraction of the BeppoSAX flux
861: originated from X-9a, the source was thus significantly harder during
862: the BeppoSAX observation. Even if there is no spectral change in X-9
863: the source was then about a factor of ten brighter during the ROSAT
864: observations where it reached $\sim1.4\times 10^{37}$ erg
865: s$^{-1}$. Between the two \chandra observations the source luminosity
866: decreases by a factor of $\sim$2 from $1.7\times 10^{36}$ erg s$^{-1}$
867: to $7\times 10^{35}$ erg s$^{-1}$ without a significant change in
868: spectral parameters. Based on the variability, either spectral and/or
869: timing, and the luminosity the source is thus highly likely to be an
870: X-ray binary.
871:
872: \bfige
873: \resizebox{0.5\hsize}{!}{\includegraphics[angle=-90]{f8a.eps}}
874: \resizebox{0.5\hsize}{!}{\includegraphics[angle=-90]{f8b.eps}}
875: \caption{Energy spectrum and contour plot for M33 X-9. The model spectrum
876: is an absorbed power law with a photon index of $1.88 \pm 0.04$. The
877: reduced $\chi^2$ is 0.87.}
878: \label{fig:x9}
879: \efige
880:
881: {\bf M33 X-13 (\object[]{CXO J013354.8+303309})} and {\bf
882: \object[]{CXO J013329.0+304216}}: These two sources have very soft
883: bremsstrahlung temperatures ($\sim$ 0.2 keV) but are also variable on
884: timescales of months, thus excluding SNRs or HII regions as
885: counterparts.
886:
887: M33 X-13 (\object[]{CXO J013354.8+303309}) is well fit with a
888: bremsstrahlung model with temperature $0.22$ keV and absorption of
889: $\sim3\times10^{21}$ cm$^{-2}$ in both observations. The source is
890: coincident with a SNR \citep{gordon:99}, but the flux varies by a
891: factor of two between two observations. It also shows significant
892: change in the hardness ratios. The source becomes softer in HR1 and
893: harder in HR2. The flattening of the spectrum is accompanied by an
894: increase in luminosity $2\times 10^{36}$ erg s$^{-1}$ to $4\times
895: 10^{36}$ erg s$^{-1}$.
896:
897: \object[]{CXO J013329.0+304216} shows a significant softening in HR1 and
898: hardening in HR2 from \dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730} to
899: \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}, the same behavior at
900: similar hardness ratios like M33 X-13 (\object[]{CXO
901: J013354.8+303309}). The spectrum of the source is well fit by a
902: bremsstrahlung spectrum with a temperature decreasing from
903: $0.24\pm0.01$ keV (\dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730}) to
904: $0.21\pm0.006$ keV (\dataset[ADS/Sa.CXO\#obs/00786]{ObsId 786}) and to
905: $0.19\pm0.006$ keV (\dataset[ADS/Sa.CXO\#obs/02023]{ObsId
906: 2023}). However, the temperature is somewhat degenerate with the
907: column density which increases from $(0.29\pm0.1)\times10^{-22}$
908: cm$^{-2}$ (\dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730}) to
909: $(0.36\pm0.1)\times10^{-22}$ cm$^{-2}$
910: (\dataset[ADS/Sa.CXO\#obs/00786]{ObsId 786}) and to
911: $(0.42\pm0.17)\times10^{-22}$ cm$^{-2}$
912: (\dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}). Although the
913: column density changes systematically, the large errors make it
914: impossible to choose which, if any, value is correct. Thus although
915: the errors on the temperatures are small the significance of the
916: temperature change from the spectral fit is not large. Between the
917: \chandra observations the luminosity increases from $2.7\times
918: 10^{36}$ erg s$^{-1}$ to $3.6\times 10^{36}$ erg s$^{-1}$.
919:
920: Temperature and variability of these sources indicate that they could
921: be quasi-soft sources, although a simple black body does not provide a
922: good fit. However, more complicated models (e.g. black body plus power
923: law) do not improve the fits due to the small number of observed
924: counts. For quasi-soft sources the luminosities would be relatively
925: low (a few $10^{36}$ erg s$^{-1}$) but still in the range observed in
926: other galaxies \citep{distefano:04}. The sources are too bright ($\sim
927: 2-4 \times 10^{36}$ erg s$^{-1}$) for normal CVs ($\sim10^{30-32}$ erg
928: s$^{-1}$), and magnetic CVs that can reach such luminosities have very
929: hard spectra with effective temperatures of several keV
930: \citep{kuulkers:06}. On the other hand the sources have too high
931: temperatures for super-soft sources \citep{kahabka:97}.
932:
933: {\bf \object[]{CXO J013329.3+304508}} and {\bf \object[]{CXO
934: J013410.5+303946}}: The two sources show evidence for spectral
935: variability in their hardness ratio diagrams. Source \object[]{CXO
936: J013329.3+304508} shows a significant softening in HR1 but no change
937: in HR2 from \dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730} to
938: \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}. The best fit model for
939: the spectrum is a bremsstrahlung model, however the temperature is not
940: well determined in the fit. Moreover, the source is variable between
941: ObsIds \dataset[ADS/Sa.CXO\#obs/00786]{786} and
942: \dataset[ADS/Sa.CXO\#obs/02023]{2023}, between which the luminosity
943: decreases by $\sim$30\% from $3\times 10^{36}$ erg s$^{-1}$ to
944: $2\times 10^{36}$ erg s$^{-1}$. Source \object[]{CXO J013410.5+303946} shows a
945: softening in HR1 from \dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730} to
946: \dataset[ADS/Sa.CXO\#obs/00786]{ObsId 786} but no significant change
947: in HR2. From \dataset[ADS/Sa.CXO\#obs/00786]{ObsId 786} to
948: \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023} the HR1 hardens again so
949: that the hardness ratio for \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}
950: is consistent with the hardness ratio for
951: \dataset[ADS/Sa.CXO\#obs/01730]{ObsId 1730}. The spectrum of ObsIds
952: \dataset[ADS/Sa.CXO\#obs/01730]{1730} and
953: \dataset[ADS/Sa.CXO\#obs/02023]{2023} is well fit by a disk black body
954: with an inner disk temperature 1.2 keV. The spectrum of
955: \dataset[ADS/Sa.CXO\#obs/00786]{ObsId 786} on the other hand is well
956: fit by a power law with a photon index of 1.7. Both photon index and
957: disk temperatures are consistent with Galactic X-ray binary spectra
958: \citep{tanaka:01}. It is interesting to note that the luminosity
959: increases between the first two observations and stays high in the
960: third, although the spectral shape in the first and third observation
961: are basically identical. The source increased its luminosity between
962: ObsIds \dataset[ADS/Sa.CXO\#obs/01730]{1730} and
963: \dataset[ADS/Sa.CXO\#obs/00786]{786} from $\sim1.3\times 10^{36}$ erg
964: s$^{-1}$ to $\sim2.1\times 10^{36}$ erg s$^{-1}$ and had the same high
965: luminosity during \dataset[ADS/Sa.CXO\#obs/02023]{ObsId 2023}.
966: Although the power law spectrum is relatively hard, the softening of
967: the hardness ratio and the increasing luminosity do not indicate a
968: state change in the system.
969:
970: The spectral and timing behavior of the bright X-ray sources in M33 is
971: varied but can be well described within the framework of Galactic
972: X-ray sources and X-ray sources in other galaxies observed with
973: \chandra.
974:
975: \section{Summary and Conclusion}
976: \label{sec:summary}
977:
978: In the three \chandra observations of M33, performed between the years
979: 2000 and 2001, 261 sources have been detected in the range of X-ray
980: luminosities from $L_{X}\sim 10^{34-38}$ erg s$^{-1}$ (0.3--8.0
981: keV). Of this total, 198 sources have been detected in at least two
982: observations, and 62 sources have been observed in all three
983: observations. We find that 49 sources show variability between
984: observations above 3 sigma. For a total of 43 sources, 25 of which in
985: more than one observation, the number of counts is sufficient for a
986: more detailed spectral fitting. Given the angular extent of our survey
987: we expect $\sim$3--4 AGN interlopers for sources with luminosities
988: above $10^{36}$ erg s$^{-1}$, and about one for sources with
989: luminosities above $10^{37}$ erg s$^{-1}$ (see \citet{grimm:05}).
990: All sources in M33, except for the nucleus, have luminosities below
991: $\sim10^{38}$ erg s$^{-1}$.
992:
993:
994: {\bf Time variability:} Roughly a quarter (49/198) of all \chandra
995: sources exhibits long timescale variability ($\sim$weeks--years),
996: therefore we can exclude SNRs and HII regions in M33 as their
997: counterparts. Except for two Galactic star interlopers none of the M33
998: sources show any short timescale X-ray variability
999: ($\sim$seconds--hours). Detection of such variability would separate
1000: stellar mass objects from AGNs, since in AGNs the light travel time
1001: is too large to produce variability on time scales shorter than
1002: hours. However, the amplitude of the long-term variability (larger
1003: than a factor of 2--3) makes it unlikely for the variable sources to
1004: be AGN (generally less than a factor of 2--3).
1005:
1006: A significant number of sources showing long-term variability but
1007: little to no short-term variability is very similar to other
1008: extragalactic X-ray source populations. The comparison of the
1009: amplitude of long term variability with AGN and known X-ray binaries
1010: shows that a large majority of these sources are X-ray binaries.
1011:
1012: {\bf X-ray spectra:} The large majority of source spectra is well fit
1013: by models containing either a power law or a thermal plasma
1014: (bremsstrahlung or {\it apec}), but in the most luminous sources more
1015: complex models are required. Of the 43 sources with enough counts for
1016: detailed spectral modeling,
1017: \begin{itemize}
1018:
1019: \item ten are well fit with spectral models consistent with accreting
1020: X-ray binaries, namely a disk black body or a black body plus power
1021: law. Three of the sources well described with a black body plus power law
1022: combination have black body temperatures at or below 0.1 keV. These
1023: are even lower than temperatures inferred from ULX intermediate mass
1024: black hole candidates, see e.g. \citet{miller:04}. The luminosities of
1025: these sources also do not indicate the presence of an intermediate
1026: mass black hole.
1027:
1028: \item In addition, 16 sources are well fit by a single power law. With
1029: one exception all the photon indices are in the range $\sim$1--2,
1030: consistent with Galactic X-ray binaries \citep{tanaka:01}. The photon
1031: indices are not inconsistent with AGN but we expect only $\sim$3
1032: background AGN among the bright sources. Two thirds of the sources
1033: well fit by a single power law are also well fit by a disk black body
1034: model with inner disk temperatures in the range 1.0--3.0 keV, which
1035: would be consistent with Galactic BH and NS binaries
1036: \citep{tanaka:01}.
1037:
1038: \item There are also 4 sources with bremsstrahlung spectra and
1039: temperatures between $\sim$2--7 keV.
1040: \end{itemize}
1041:
1042: Moreover, 12 of 30 sources also show long term variability, confirming
1043: their nature as accreting objects. Therefore more than half (30) of
1044: the sources are consistent with being X-ray binaries based on the
1045: X-ray spectrum.
1046:
1047: The majority (12) of the remaining 13 sources are well fit by thermal
1048: plasma models (bremsstrahlung or {\it apec} models as discussed
1049: above), 7 of which with relatively low temperatures around 0.2
1050: keV. Five of these 12 sources are variable. Three sources with
1051: unusually low bremsstrahlung temperatures from 0.1--0.3 keV are also
1052: well fit by an {\it apec} model as well; one is variable. The {\it
1053: apec} temperatures of these sources are around 0.3 keV and abundances
1054: are consistent with the overall metallicity of M33 \citep{blair:85}
1055: (see Fig. \ref{fig:slopes}). A black body model does not provide a
1056: good fit to the data. Five non-variable of the 12 sources are
1057: coincident with SNRs \citep{gordon:99}. Variable sources in this
1058: category are candidates for quasi-soft sources.
1059:
1060: Two sources are well fit by a single black body with temperatures
1061: of $\sim$75 eV and $\sim$37 eV respectively. One of the sources is a
1062: long term variable with no significant spectral variability. The other
1063: source has been observed only once. Both sources are not significantly
1064: detected above 1 keV. These sources are the best candidates for
1065: super-soft sources in M33.
1066:
1067: Given the spectra, and variability, the majority of these sources are
1068: likely X-ray binaries, with a sizable fraction of SNRs/HII regions
1069: confirming a young X-ray source population.
1070:
1071: {\bf Spectral variability:} Six of the bright sources show evidence
1072: for spectral variability. The type of variability is varied, as has
1073: been seen in other galaxies as well \citep{zezas:06}. The spectral
1074: variability is sometimes correlated with luminosity (M33 X-4, X-13),
1075: and sometimes not (M33 X-5). In the source X-4 and X-5 the spectral
1076: change seems to originate in the column density--photon index
1077: degeneracy. In both cases, the thermal component (black body or disk
1078: black body) remains constant, whereas the photon index and column
1079: density change their values. With the current data it is not possible
1080: to decide, if there is a change in column density within the system or
1081: a genuine change in the hard component, assuming that the model is a
1082: correct description. M33 X-13 and \object[]{CXO J013329.0+304216} both
1083: show a continuous softening of the spectrum that is correlated in both
1084: cases with an increase in luminosity, and they could be identified
1085: with quasi-soft sources. The other two sources show significant change
1086: only in the soft band with only moderate change in luminosity.
1087:
1088: Given the brightness, spectra, and variability (temporal and spectral)
1089: these sources are highly likely X-ray binaries.
1090:
1091: A comparison of $N_H$ values measured from the X-ray spectra and HI
1092: radio observations of M33 shows that the X-ray absorption is larger
1093: for most sources than inferred from HI column density. A few sources
1094: have lower X-ray absorption than expected and may be located in front
1095: of the HI gas. However, the majority of sources is intrinsically
1096: absorbed. The absorption in X-rays can be compared with values observed
1097: in Galactic X-ray binaries, and is generally moderate at a few
1098: $10^{21}$ cm$^{-2}$.
1099:
1100: This comparison allows the localization of a few sources in M33, and
1101: indicates for most sources intrinsic absorption at the same level as
1102: observed in Galactic X-ray binaries.
1103:
1104: The luminosity function of M33 sources and optical counterparts from
1105: \citet{grimm:05} have shown that the X-ray source population in M33 is
1106: dominated by young objects. The long term variability, spectra, and
1107: spectral variability presented in this paper provide evidence that a
1108: large majority of sources are X-ray binaries. The properties of the
1109: X-ray binaries (variability, spectral parameters including absorption)
1110: are very similar to HMXBs in the Milky Way or the Magellanic
1111: Clouds. Thus X-ray binary populations do not show strong variations
1112: over the parameters, e.g. metallicity, covered by the Milky Way, M33,
1113: and the Magellanic Clouds. Thus the M33 X-ray source population
1114: confirms our knowledge about HMXB populations in the Milky Way and
1115: other galaxies. Therefore M33 can also be used as a comparison template
1116: for similar galaxies at larger distances, which is important as
1117: galaxies of this type are numerous.
1118:
1119: Comparisons with previous X-ray missions show that repeated
1120: observations of a galaxy are important to study the X-ray source
1121: population in detail. This analysis also shows that especially with
1122: repeat observations it is possible to identify the major parts of an
1123: X-ray population on X-ray data alone. The ongoing deep \chandra survey
1124: of M33 will add significantly more data to such an analysis, and
1125: provide a more detailed picture of the M33 X-ray source population.
1126:
1127: \section{Acknowledgments}
1128: This work has been supported by NASA grant GO2-3135X and by grant
1129: AR6-7007X from Chandra X-ray Center operated by the SAO for NASA. This
1130: research has made use of the NASA/IPAC Extragalactic Database (NED)
1131: which is operated by the Jet Propulsion Laboratory, California
1132: Institute of Technology, under contract with the National Aeronautics
1133: and Space Administration. We thank the referee for constructive
1134: comments on the paper.
1135:
1136: \bibliographystyle{apj}
1137: \bibliography{ms}
1138:
1139: \begin{appendix}
1140:
1141: \include{tab1}
1142: \include{tab2}
1143: \include{fig1}
1144:
1145: \end{appendix}
1146:
1147: \end{document}
1148: