1: \documentclass{article}
2:
3: \usepackage{graphicx}
4: \usepackage{amsmath}
5: \usepackage{times}
6: \usepackage{natbib}
7: \usepackage{xspace}
8: \usepackage{emulateapj}
9:
10: % \shorttitle{Quasi-Periodic Oscillation in Seyfert galaxies}
11: % \shortauthors{Benlloch et al.}
12:
13: \begin{document}
14:
15: \title{Quasi-Periodic Oscillation in Seyfert galaxies: Significance
16: levels. \\ The Case of Mrk~766}
17:
18: \author{
19: Sara Benlloch\altaffilmark{1},
20: J\"orn~Wilms\altaffilmark{1},
21: Rick Edelson\altaffilmark{2},
22: Tahir Yaqoob\altaffilmark{3,4},
23: R\"udiger Staubert\altaffilmark{1},
24: }
25:
26: \altaffiltext{1}{Institut f\"ur Astronomie und
27: Astrophysik--Astronomie, University of T\"ubingen, Waldh\"auser
28: Stra\ss{}e 64, D-72076 T\"ubingen, Germany}
29:
30: \altaffiltext{2}{University of California, Los Angeles, Department of Astronomy, Los Angeles, CA 90095-1562}
31:
32: \altaffiltext{3}{Laboratory for High Energy Astrophysics, NASA Goddard
33: Space Flight Center, Greenbelt, MD 20771}
34:
35: \altaffiltext{4}{Johns Hopkins University, Department of Physics and
36: Astronomy, Homewood Campus, 3400 North Charles Street, Baltimore, MD
37: 21218}
38:
39: %\email{benlloch@astro.uni-tuebingen.de}
40:
41: \begin{abstract}
42: We discuss methods to compute significance levels for the existence of
43: quasi-periodic oscillations (QPOs) in Active Galactic Nuclei (AGN) which
44: take the red-noise character of the X-ray lightcurves of these objects
45: into account. Applying epoch folding and periodogram analysis to the
46: \textsl{XMM-Newton} observation of the Seyfert galaxy Mrk~766, a possible
47: QPO at a timescale of 4200\,s has been reported. Our computation of the
48: significance of this QPO, however, shows that the 4200\,s peak is not
49: significant at the 95\% level. We conclude that the 4200\,s feature is
50: an artifact of the red-noise process and not the result of a physical
51: process within the Active Galactic Nuclei.
52: \end{abstract}
53:
54: \keywords{accretion disks --- galaxies: individual (Mrk~766) --- galaxies:
55: Seyferts --- X-rays: galaxies}
56:
57: %\clearpage
58:
59:
60: \section{Introduction}
61:
62: X-ray quasi-periodic oscillations (QPOs) are among the most important
63: observational properties of galactic X-ray binaries \citep[XRBs; see][for a
64: recent review]{vanderklis:00a}, yielding important constraints on the mass
65: of the central black hole (BH), $M_{\text{BH}}$, and providing theoretical
66: clues and constraints on the operative physical processes and geometry in
67: the regime of strong gravity. Although their origin is still a matter of
68: scientific debate \citep[][and references therein]{psaltis:00a}, there is
69: general agreement that QPOs originate close to the central BH.
70:
71: To date, QPOs at timescales of a few kiloseconds have also been claimed for
72: some active galactic nuclei (AGN): NGC~4151 \citep{fiore:89a}, NGC~6814
73: \citep{mittaz:89a}, NGC~5548 \citep{papadakis:93b}, NGC~4051
74: \citep{papadakis:95b}, RX~J0437.4$-$4711 \citep{halpern:96a},
75: IRAS~18325$-$5926 \citep{iwasawa:98a}, MCG$-$6-30-15 \citep{lee:00a},
76: Mrk~766 \citep[][]{boller:01a}, and IRAS~13224$-$3809
77: \citep{pfefferkorn:01a}. In addition, possible long term periodicities with
78: periods of months have been claimed for the radio loud AGN Mrk~421,
79: Mrk~501, and PKS~2155$-$304 \citep{osone:01a}. Some of the kilosecond QPO
80: findings, however, are controversial, with the strongest EXOSAT result
81: (NGC~5548) being disputed by \citet{tagliaferri:96a}, who attribute it
82: mostly to periodic swapping of detectors. NGC~6814 turned out to be
83: confused with a cataclysmic variable \citep{madejski:93a,staubert:94a}.
84:
85: Given the important implications based upon detection of QPO in AGN, it
86: seems worthwhile to study the methods employed to determine their
87: significance in detail. In this \textit{Letter}, we present a study of how
88: to compute this significance using Monte Carlo simulations of lightcurves
89: with the method of \citet{timmer:95a} and using periodogram analysis and
90: epoch folding to detect the periodicity (\S\ref{sec:qpo}). We then apply
91: our methods to a reanalysis of the \textit{XMM-Newton} lightcurve of
92: Mrk~766 and find that the QPO claimed in \citet{boller:01a} has in fact low
93: statistical significance (\S\ref{sec:mkn}). In \S\ref{sec:disc} we discuss
94: our results and comment on further work.
95:
96:
97: \begin{figure*}
98: \centerline{\includegraphics[width=0.95\textwidth]{f1.eps}}
99: \figcaption{Three examples of typical red-noise lightcurves for a process
100: with an {\scriptsize{$f^{-1.9}$}} spectrum binned to a resolution of
101: 100\,s using the method described by \citet{timmer:95a}. \textit{Upper
102: panels}: Simulated red-noise lightcurves. \textit{Lower panels}:
103: {\scriptsize{$\chi^2(P)$}} curves from epoch folding (solid line) of
104: the corresponding simulations. The dashed lines represent the 99\%,
105: 95\%, and 68\% ``local significance'' levels (see text for definition)
106: obtained for a sample of 5000 Monte Carlo simulations using the
107: \citet{timmer:95a} method. The dotted line represents the 99\%
108: significance level obtained with the ``phase randomization'' method,
109: which overestimates the significance of the peaks in the
110: {\scriptsize{$\chi^2(P)$}} curves. Note that the presence of one or
111: more peaks in the {\scriptsize{$\chi^2(P)$}} curve is far from unusual
112: for red-noise lightcurves.\label{fig:sim_plots}}
113: \end{figure*}
114:
115: \section{Significance of QPO Detections}\label{sec:qpo}
116:
117: Currently, two methods for the variability analysis of astronomical
118: sources, especially in searching for periodic signals, are common in
119: astronomy: periodogram analysis and epoch folding. We only give a brief
120: description of these methods here, see, e.g., \citet{vanderklis:89a},
121: \citet{leahy:83a} and \citet{davies:90a} for in-depth discussions. Based
122: on the Fourier decomposition of the lightcurve, periodogram analysis (often
123: called power spectrum density analysis, PSD) is especially sensitive to
124: periodic signals with a modulation that is close to sinusoidal. On the
125: other hand, epoch folding ($\chi^2(P)$) is based on comparing pulse
126: profiles obtained from binning the data into phase bins at a test period,
127: $P$, with a constant count rate using a $\chi^2$ test. In both methods,
128: the detection of a periodic signal is claimed if the PSD value or the
129: $\chi^2$ value at the period of interest is significantly above the values
130: of the testing statistics surrounding this period.
131:
132: In order to safely use either of these methods, it is critically important
133: to accurately estimate the significance of QPO features. This is not
134: trivial for real AGN lightcurves, which have much worse sampling and signal
135: to noise than X-ray binary lightcurves. Were the X-ray lightcurve purely
136: dominated by white noise and evenly sampled, the significance could be
137: easily determined from the statistical properties of the PSD or from the
138: $\chi^2(P)$ statistics. This assumption, however, does not apply to AGN,
139: where the PSD can be well approximated by a power-law $P(f)\propto
140: f^{-\beta}$ with $\beta\sim 1\ldots 2$ for the frequencies of interest
141: \citep{lawrence:93a,green:93a,koenig:97a,edelson:99a}.
142:
143: When studying a feature in $\chi^2(P)$ or a PSD, we can ask two different
144: questions: 1.~What is the significance of a QPO peak at a given
145: (predefined) frequency (the ``local significance'' of the QPO), and,
146: 2.~what is the significance of a QPO feature seen in a given frequency
147: range (the ``global significance'' of the QPO). Note that these questions
148: are really different questions, since we do have more knowledge about the
149: QPO in the case of the ``local significance'' (we do know its frequency),
150: while in the case of the ``global significance'', we are only interested in
151: having the question answered that a feature somewhere in the interesting
152: frequency range is significant or not. To our knowledge, however, no
153: formulae for determining the significance of a peak in the PSD or in
154: $\chi^2(P)$ exist for such red-noise lightcurves. A common approach,
155: therefore, is to resort to Monte Carlo simulations \citep[see,
156: e.g.,][]{horne:86a}. In these simulations, a large number ($\gtrsim 1000$)
157: of lightcurves with the same statistical properties and same temporal
158: sampling as the original lightcurve are generated and their $\chi^2(P)$ is
159: computed.
160:
161: For computing the ``local significance'', one determines the statistical
162: distribution of the resulting $\chi^2(P)$ value at the frequency of
163: interest. A deviation of the measured $\chi^2(P)$ from the red-noise
164: $\chi^2(P)$ is significant if it is above a certain threshold determined
165: from this distribution. Typically, threshold values of 99\% or even 99.9\%
166: are used \citep{bevington}. An identical procedure can be used to test the
167: significance of a peak in the PSD.
168:
169: For determining the ``global significance'', a similar approach is used.
170: Since we are looking at $\chi^2(P)$ values determined at different periods,
171: one has to take the red-noise character of the data and window effects into
172: account. Instead of directly using $\chi^2(P)$ values, more robust tests
173: can be devised by first dividing the computed individual $\chi^2(P)$ values
174: by the average $\langle\chi^2(P)\rangle$, obtained by averaging the
175: $\chi^2(P)$ curves of the simulated lightcurves. To determine the ``global
176: significance'' of the maximum $\chi^2(P)$-peak of an observed lightcurve,
177: we therefore propose to compare this value to the distribution of $\xi_{\rm
178: max}:=\max\left\{ \chi^2(P)/\langle \chi^2(P)\rangle\right\}$ from the
179: simulated lightcurves. Here, the maximum is to be taken over the period
180: range of interest. We would consider an observed peak as likely due to a
181: physical effect only if it is above the 99\% threshold determined from the
182: distribution. The maximum $\chi^2(P)$-peaks present in the red-noise
183: lightcurves of Fig.~\ref{fig:sim_plots}b and~c have a ``global
184: significance'' of 61\% and 80\% respectively. Similar global tests can
185: also be devised for the distribution of the 2nd or 3rd largest $\chi^2(P)$
186: value and for the significance of the largest peaks in a PSD, although we
187: will not use them here.
188:
189: It is crucially important how lightcurves are produced by Monte Carlo
190: simulations, if the significance level of any peak seen in a red-noise PSD
191: is to be determined. One usually starts with a model PSD and performs an
192: inverse Fourier transformation to obtain a lightcurve. An often used
193: algorithm, called the ``phase randomization'' method \citep{done:92a},
194: determines the Fourier amplitude from the square root of the power-law
195: shaped PSD and assumes the Fourier phase to be uniformly distributed in
196: $[0,2\pi[$. Although the PSDs produced by ``phase randomization'' are
197: $\propto f^{-\beta}$, it was pointed out by \citet{timmer:95a} and
198: \citet{papadakis:95b} that the resulting lightcurves do not resemble the
199: pure red-noise process in all of their statistical properties. First, this
200: procedure chooses a deterministic amplitude for each frequency and only
201: randomizes the phases. All simulated lightcurves thus exhibit a trend
202: caused by the dominating lowest frequency. Secondly, the periodogram of a
203: red-noise lightcurve must obey the usual periodogram statistics: the PSD
204: follows approximately a $\chi^2$ distribution with two degrees of freedom,
205: $\chi^2_2$, i.e., the standard deviation of each PSD point is of the same
206: magnitude as the PSD value itself such that the periodogram is fluctuating
207: wildly \citep[see, e.g.,][]{vanderklis:89a}. ``Phase randomization'' does
208: not take into account this randomness of the periodogram according to the
209: $\chi^2_2$ distribution and therefore the uncertainty of the related
210: distribution of the estimated periods is significantly underestimated. In
211: other words, red-noise PSDs have a larger scatter than those obtained from
212: ``phase randomization'' -- including the possibility of outliers that
213: strongly deviate from the general $f^{-\beta}$ behavior. This is important
214: since these outliers could be interpreted as quasi periodic oscillations,
215: while in reality they result from the statistics of the red-noise process.
216: We note that the frequent occurrence of random peaks in red-noise PSDs is
217: avoided in X-ray binary work by averaging many PSDs to obtain the ``true
218: PSD'' of a source \citep{nowak:99b,vanderklis:89a}. For AGN, averaging the
219: PSDs is unfortunately not possible due to the prohibitively long
220: observation times required.
221:
222: As we have mentioned above, the ``phase randomization'' algorithm will
223: not produce lightcurves with the appropriate red-noise statistical
224: characteristics. Indeed, were one to use the Monte Carlo approach outlined
225: above to lightcurves computed with ``phase randomization'', the QPO
226: significance would be overestimated. Instead, an algorithm that produces
227: lightcurves with the correct statistical red-noise behavior has to be used.
228: For this purpose \citet{timmer:95a} proposed a new algorithm. This
229: algorithm \citep[used also by][]{green:99a} allows randomness both in phase
230: and in amplitude producing the desired $\chi^2_2$ distribution in the
231: periodogram. In the following, we will use the \citet{timmer:95a} algorithm
232: to simulate red-noise lightcurves. As an illustration, three simulated
233: examples of red-noise lightcurves with significance levels obtained from
234: the Monte Carlo simulations are displayed in Fig.~\ref{fig:sim_plots}. We
235: also display the significance levels obtained with ``phase randomization''
236: to indicate that these levels are in fact much smaller than the correct
237: levels and therefore would imply an overestimate in the significance of an
238: apparent period.
239:
240: We note that Poisson noise introduces additional ``observational noise'' in
241: the measured lightcurves. This observational noise has to be added to the
242: simulated lightcurves after the inverse Fourier transform has been
243: performed. For each time bin, the number of observed photons is drawn from
244: a Poisson distribution with its mean given by $r(t) \Delta t$, where $r(t)$
245: is the simulated count rate and $\Delta t$ is the binning. For data with
246: new instruments such as \textit{XMM-Newton}, with a high signal to noise
247: ratio, this latter step can be ignored if the source is bright enough. For
248: example, with $\Delta t=100$\,s, ``observational noise'' contributes less
249: than 3\% to the Mrk~766 lightcurve. For earlier instruments such a
250: simplification of the Monte Carlo algorithm is not possible. We note that
251: in earlier work employing ``phase randomization'' and the addition of
252: observational noise, the resulting PSD statistics would asymptotically
253: approach the $\chi^2_2$ distribution and therefore the overestimation of
254: the period significance would be less than with newer data with a high
255: signal to noise ratio.
256:
257:
258: \centerline{\includegraphics[width=0.45\textwidth]{f2.eps}}
259: \figcaption{XMM-Newton EPIC-pn lightcurve of Mrk~766 for the 0.2--2\,keV
260: band with a resolution of 100\,s.\label{fig:lc}}
261:
262: \section{The Case of Mrk~766}\label{sec:mkn}
263:
264: We now apply the methods outlined in the previous section to the putative
265: QPO in \textit{XMM-Newton} data of Mrk~766 taken during revolution 0082. We
266: concentrate on data from the EPIC-pn instrument \citep{strueder:01a} which
267: was operated in the small window mode during the observation.
268:
269: \subsection{Data Extraction}
270:
271: We extracted source photons from a circle of 9.5 pixels radius centered on
272: the source (detector coordinates (37.5,54)). For the background, data from
273: an off-axis position (18,17.5) extracted with a circle of then same radius
274: were used. The time range of the lightcurve was chosen to be consistent
275: with the approach of \citet{boller:01a} and results in an exposure time of
276: 29\,ksec. After background subtraction, we corrected the measured
277: count-rates as is appropriate for the $\sim$71\% live-time during the
278: 5.7\,ms readout cycle of the pn-CCD \citep{kuster:99a}. Fig.~\ref{fig:lc}
279: displays the resulting lightcurve.
280:
281:
282:
283: \subsection{Lightcurve Analysis}
284:
285: We display the PSD and the $\chi^2(P)$ curves in Fig.~\ref{fig:sig} (the
286: frequency range is $0.4 \times 10^{-4}$\,Hz to $30 \times 10^{-4}$\,Hz with
287: 57 independent frequencies, the period range is 2000--5000\,s with 132 test
288: periods). A peak at a period of $\sim$4200\,s (corresponding to a
289: frequency of $\sim 2.5 \times 10^{-4}$\,Hz) that is consistent with the
290: period claimed by \citet{boller:01a} is seen. In order to determine the
291: significance of the peaks seen in Fig.~\ref{fig:sig}, we computed
292: significance levels using the methods outlined in section~\ref{sec:qpo}.
293:
294: Before we can perform these simulations, we need to determine the shape of
295: the PSD. For this purpose, we apply the ``response method'' of
296: \citet{done:92a} and \citet{green:99a}, where a model power spectrum
297: generated through the combination of a large number of red noise simulated
298: lightcurves is compared to the observed power spectrum. Varying the slope
299: and normalization of the model periodogram, the best $\chi^2$ fit gives a
300: slope of $\beta=1.9$, and a normalization determined by the variance of the
301: original XMM lightcurve ($\chi^2$/dof = 17.89/55). This low value
302: indicates that the observed PSD is fully consistent with a power law and
303: that no additional components, such as a QPO, are required.
304:
305: \vspace*{0.2cm}
306: \centerline{\includegraphics[width=0.45\textwidth]{f3.eps}}
307: \figcaption{\textit{a} PSD in \citet{leahy:83a} normalization and best
308: power-law fit using the \citet{done:92a} method (continuous lines) and,
309: \textit{b} {\scriptsize{$\chi^2(P)$}} curve (continuous line) of the XMM
310: Mrk~766 lightcurve of Fig.~\ref{fig:lc}. The set of dashed lines in both
311: panels represent in ascending order the $68\%$, $95\%$, and for $99\%$
312: ``local significance'' levels for a set of 5000 Monte Carlo red-noise
313: simulations with $\beta =1.9$.\label{fig:sig}} \vspace*{0.2cm}
314:
315:
316: We note, however, that the short duration of the observation only allows
317: for a poor determination of the PSD shape. Methods complementary to testing
318: the consistency of the PSD with a power law should be used. We therefore
319: simulated the desired 5000 lightcurves from a $f^{-1.9}$-PSD, with 29000\,s
320: of duration using a sampling interval of 100\,s, a mean count rate
321: $\mu=9.15\,\text{counts}\,\text{s}^{-1}$, a lightcurve variance
322: $\sigma^2=2.46\,\text{counts}^2\,\text{s}^{-2}$. Examples of such
323: simulated red-noise lightcurves are shown in Fig.~\ref{fig:sim_plots}.
324:
325:
326: To obtain the ``local significance'' of the peak at $\sim$4200\,s , we
327: calculated $\chi^2(P)$ and the PSD of each of the 5000 Monte Carlo
328: realizations. For each trial period, we then computed the distribution of
329: the $\chi^2(P)$ and PSD values at this trial period from all realizations.
330: These distributions were used to determine the 99\%, 95\%, and 68\% ``local
331: significance'' levels using the method described in section~\ref{sec:qpo}.
332: Our results are shown in Fig.~\ref{fig:sig}. The peak at $\sim$4200\,s
333: claimed as QPO by \citet{boller:01a} is below the 95\% significance curve.
334:
335:
336: \vspace*{0.2cm}
337: \centerline{\includegraphics[width=0.45\textwidth]{f4.eps}}
338: \figcaption{Probability distribution of the {\scriptsize{$\xi_{\rm max}$}}
339: values (see text for definition) from the 5000 red-noise simulated
340: lightcurves in the 2000--5000\,s period range. The ``global
341: significance'' indicates the probability of finding a lightcurve with a
342: maximum peak with a value greater than the correspondent
343: {\scriptsize{$\xi_{\rm max}$}} value. The dashed line marks the position
344: of the {\scriptsize{$\xi_{\rm max}$}} value of the 4200\,s feature in
345: Mrk~766 in the 2000--5000\,s period range. The ``global significance''
346: of this peak corresponds to a value of 45\%.
347: \label{fig:rel_sig}}
348: \vspace*{0.2cm}
349:
350: We also applied the ``global significance'' test to the Mrk~766 lightcurve
351: for the period range from 2000\,s to 5000\,s. Fig.~\ref{fig:rel_sig} shows
352: the probability distribution of $\xi_{\rm max}$, which corresponds to the
353: probability of finding a red-noise lightcurve with a normalized maximum
354: peak value less than or equal to the corresponding $\xi_{\rm max}$ value.
355: For Mrk~766, the putative QPO peak lies at the 45\% mark. In other words,
356: $\sim$50\% of the simulated red-noise lightcurves show peaks that are more
357: significant than the peak observed in Mrk~766. We conclude that the
358: observed 4200\,s ``QPO'' in Mrk~766 may be an artifact of the red-noise
359: process and not the result of a physical process.
360:
361: As we mentioned above, the value of $\beta$ determined from the
362: observations is rather uncertain. However, our result is independent of the
363: specific value of $\beta$: Simulations with $\beta$ values ranging from 1
364: to~2, i.e., over the typical $\beta$ range seen in AGN, yielded similar
365: results.
366:
367:
368: \section{Conclusions}\label{sec:disc}
369: In this \textit{Letter} we have described two methods to determine
370: the significance of possible quasi-periodic signals in the red-noise
371: lightcurves observed from Active Galactic Nuclei using Monte Carlo
372: simulations; a frequency-dependent ``local significance'' test and a
373: ``global significance'' test. Reiterating arguments by \citet{timmer:95a},
374: we showed that ``phase randomization'' techniques should not be used for
375: the generation of simulations since the resulting lightcurves do not
376: exhibit true red-noise characteristics. The periodograms of lightcurves
377: produced by this method do not obey the $\chi^2$ statistics which results
378: from a random process, overestimating therefore the significance of peaks
379: present in the periodogram of an underlying red-noise lightcurve. Instead
380: of ``phase randomization'' we recommend the algorithm of \citet{timmer:95a}
381: to simulate red-noise lightcurves with the correct statistical properties.
382:
383: The ``local significance'' test based on red-noise power simulations
384: generated with the \citet{timmer:95a} algorithm shows that the 4200\,s
385: feature in Mrk~766 is not significant at the 95\% level. This statement
386: holds for both, PSD and $\chi^2(P)$ analysis (Fig.~\ref{fig:sig}). The
387: ``global significance'' of the 4200\,s feature is 45\%, i.e., higher peaks
388: in the 2000--5000\,s period range are found in roughly half of all simulated
389: lightcurves. Thus, the presence of a peak in a red-noise $\chi^2(P)$ or PSD
390: is far from unusual. We are therefore led to the conclusion that we cannot
391: confirm the claim of a $\sim$4200\,s QPO. Rather, we attribute the peak at
392: 4200\,s to a random occurrence which is due to the red-noise character of
393: AGN data.
394:
395: We are currently in the process of checking the kilosecond QPO detections
396: claimed for AGN using archival data and the methods outlined in this
397: \textit{Letter}. We will report on our results in a forthcoming
398: publication.
399:
400: \acknowledgments
401:
402: We thank R.E.~Rothschild for useful discussions. We acknowledge support
403: from DLR grant 50~OX~0002, and NASA grants NAG~5-7317, NAG~5-9023, and
404: NCC5-447. This work is based on observations obtained with
405: \textit{XMM-Newton}, an ESA science mission with instruments and
406: contributions directly funded by ESA Member States and the USA (NASA).
407:
408: %\clearpage
409: \begin{thebibliography}{}
410:
411: \bibitem[\protect\astroncite{Bevington \& Robinson}{1992}]{bevington}
412: Bevington, P.~R., \& Robinson, D.~K., 1992,
413: \newblock Data Reduction and Error Analysis for the Physical Sciences,
414: \newblock (2d ed.; New York: McGraw-Hill)
415:
416: \bibitem[\protect\astroncite{Boller et~al.}{2001}]{boller:01a}
417: Boller, T., Keil, R., Tr{\"u}mper, J., O'Brien, P.~T., Reeves, J., \& Page, M.,
418: 2001, A\&A, 365, L146
419:
420: \bibitem[\protect\astroncite{Davies}{1990}]{davies:90a}
421: Davies, S.~R., 1990, MNRAS, 244, 93
422:
423: \bibitem[\protect\astroncite{Done et~al.}{1992}]{done:92a}
424: Done, C., Madejski, G.~M., Mushotzky, R.~F., Turner, T.~J.,
425: Koyama, K., \& Kunieda, H., 1992, ApJ, 400,
426: 138
427:
428: \bibitem[\protect\astroncite{Edelson \& Nandra}{1999}]{edelson:99a}
429: Edelson, R., \& Nandra, K., 1999, ApJ, 514, 682
430:
431: \bibitem[\protect\astroncite{Fiore, Massaro \& Perola}{1989}]{fiore:89a}
432: Fiore, F., Massaro, E., Perola, G.~C., \& Piro, L., 1989, ApJ, 347, 171
433:
434: \bibitem[\protect\astroncite{Green, McHardy \& Done}{1999}]{green:99a}
435: Green, A.~R., McHardy, I.~M., \& Done, C., 1999, MNRAS, 305, 309
436:
437: \bibitem[\protect\astroncite{Green, {McHardy} \& Lehto}{1993}]{green:93a}
438: Green, A.~R., {McHardy}, I.~M., \& Lehto, H.~J., 1993, MNRAS, 265, 664
439:
440: \bibitem[\protect\astroncite{Halpern \& Marshall}{1996}]{halpern:96a}
441: Halpern, J.~P., \& Marshall, H.~L., 1996, ApJ, 464, 760
442:
443: \bibitem[\protect\astroncite{Horne \& Baliunas}{1986}]{horne:86a}
444: Horne, J.~H., \& Baliunas, S.~L., 1986, ApJ, 302, 757
445:
446: \bibitem[\protect\astroncite{Iwasawa et~al.}{1998}]{iwasawa:98a}
447: Iwasawa, K., Fabian, A.~C., Brandt, W.~N., Kunieda, H., Misaki, K., Reynolds,
448: C.~S., \& Tershima, Y., 1998, MNRAS, 295, L20
449:
450: \bibitem[\protect\astroncite{K{\"o}nig \& Timmer}{1997}]{koenig:97a}
451: K{\"o}nig, M., \& Timmer, J., 1997, A\&AS, 124, 589
452:
453: \bibitem[\protect\astroncite{Kuster et~al.}{1999}]{kuster:99a}
454: Kuster, M., Benlloch, S., Kendziorra, E., \& Briel, U.~G., 1999,
455: \newblock in EUV, X-ray and Gamma-Ray Instrumentation for Astronomy X, ed.
456: O.~H. Siegmund, K.~A. Flanagan, (Bellingham, WA: SPIE), 673
457:
458: \bibitem[\protect\astroncite{Lawrence \& Papadakis}{1993}]{lawrence:93a}
459: Lawrence, A., \& Papadakis, J., 1993, ApJ, 414, L85
460:
461: \bibitem[\protect\astroncite{Leahy et~al.}{1983}]{leahy:83a}
462: Leahy, D.~A., Darbro, W., Elsner, R.~F., Weisskopf, M.~C., Sutherland, P.~G.,
463: Kahn, S., \& Grindlay, J.~E., 1983, ApJ, 266, 160
464:
465: \bibitem[\protect\astroncite{Lee et~al.}{2000}]{lee:00a}
466: Lee, J.~C., Fabian, A.~C., Reynolds, C.~S., Brandt, W.~N., \& Iwasawa, K.,
467: 2000, MNRAS, 318, 857
468:
469: \bibitem[\protect\astroncite{Madejski et~al.}{1993}]{madejski:93a}
470: Madejski, G.~M., Done, C., Turner, T.~J., Mushotzky, R.~F., Serlemitsos, P.,
471: Fiore, F., Sikora, M., \& Begelman, M.~C., 1993, Nature, 365, 626
472:
473: \bibitem[\protect\astroncite{Mittaz \& {Branduardi-Raymont}}{1989}]{mittaz:89a}
474: Mittaz, J. P.~D., \& {Branduardi-Raymont}, G., 1989, MNRAS, 238, 1029
475:
476: \bibitem[\protect\astroncite{Nowak et~al.}{1999}]{nowak:99b}
477: Nowak, M.~A., Vaughan, B.~A., Wilms, J., Dove, J.~B., \& Begelman, M.~C.,
478: 1999, ApJ, 510, 874
479:
480: \bibitem[\protect\astroncite{Osone, Teshima \& Mase}{2001}]{osone:01a}
481: Osone, S., Teshima, M., \& Mase, K., 2001,
482: \newblock in Proc.\ 27th ICRC
483: \newblock (astro-ph/0106223)
484:
485: \bibitem[\protect\astroncite{Papadakis \& Lawrence}{1993}]{papadakis:93b}
486: Papadakis, I.~E., \& Lawrence, A., 1993, Nature, 361, 233
487:
488: \bibitem[\protect\astroncite{Papadakis \& Lawrence}{1995}]{papadakis:95b}
489: Papadakis, I.~E., \& Lawrence, A., 1995, MNRAS, 272, 161
490:
491: \bibitem[\protect\astroncite{Pfefferkorn et~al.}{2001}]{pfefferkorn:01a}
492: Pfefferkorn, F., Boller, T., Burwitz, V., \& Predehl, P., 2001,
493: \newblock in Proc. MAXI workshop on AGN variability
494: \newblock (astro-ph/0106203)
495:
496: \bibitem[\protect\astroncite{Psaltis}{2000}]{psaltis:00a}
497: Psaltis, D., 2000, Adv. Space Res.,
498: \newblock submitted (astro-ph/0012251)
499:
500: \bibitem[\protect\astroncite{Staubert et~al.}{1994}]{staubert:94a}
501: Staubert, R., K{\"o}nig, M., Friedrich, S., Lamer, G., Sood, R.~K., James,
502: S.~D., \& Sharma, D.~P., 1994, A\&A, 288, 513
503:
504: \bibitem[\protect\astroncite{{Str{\"u}der} et~al.}{2001}]{strueder:01a}
505: {Str{\"u}der}, L., et~al., 2001, A\&A, 365, L18
506:
507: \bibitem[\protect\astroncite{Tagliaferri et~al.}{1996}]{tagliaferri:96a}
508: Tagliaferri, G., Bao, G., Israel, G.~L., Stella, L., \& Treves, A., 1996, ApJ,
509: 465, 181
510:
511: \bibitem[\protect\astroncite{Timmer \& K{\"o}nig}{1995}]{timmer:95a}
512: Timmer, J., \& K{\"o}nig, M., 1995, A\&A, 300, 707
513:
514: \bibitem[\protect\astroncite{{van der Klis}}{1989}]{vanderklis:89a}
515: {van der Klis}, M., 1989,
516: \newblock in Timing neutron stars, ed. H. \"Ogelman, E. van~den Heuvel,
517: (Dordrecht: Kluwer), ~27
518:
519: \bibitem[\protect\astroncite{{van der Klis}}{2000}]{vanderklis:00a}
520: {van der Klis}, M., 2000, ARA\&A, 38, 717
521:
522: \end{thebibliography}
523:
524:
525:
526:
527: \end{document}
528:
529:
530: