1:
2: %% The command below calls the preprint style
3: %% which will produce a one-column, single-spaced document.
4: \documentclass[12pt,preprint]{aastex}
5: %\usepackage{txfonts}
6:
7: %% manuscript produces a one-column, double-spaced document:
8: %\documentclass[manuscript]{aastex}
9:
10: %% preprint2 produces a double-column, single-spaced document:
11: %% \documentclass[preprint2]{aastex}
12:
13: %% Sometimes a paper's abstract is too long to fit on the
14: %% title page in preprint2 mode. When that is the case,
15: %% use the longabstract style option.
16: %% \documentclass[preprint2,longabstract]{aastex}
17:
18: %% If you want to create your own macros, you can do so
19: %% using \newcommand. Your macros should appear before
20: %% the \begin{document} command.
21:
22: %\newcommand{\vdag}{(v)^\dagger}
23: %\newcommand{\myemail}{skywalker@galaxy.far.far.away}
24:
25: %% You can insert a short comment on the title page using the command below.
26:
27: \slugcomment{To be published in ApJ }
28:
29: \shorttitle{Time Domain Coherence}
30: \shortauthors{Che, Liu \& Li}
31:
32: \begin{document}
33:
34: \title{X-ray Variability Coherence in the Time Domain}
35:
36: \author{Xiao Che\altaffilmark{1}, Cong-Zhan Liu\altaffilmark{1} and Ti-Pei Li\altaffilmark{1,2,3}}
37:
38: \altaffiltext{1}{Department of Engineering Physics \& Center for
39: Astrophysics, Tsinghua University, Beijing, China}
40: \altaffiltext{2}{Department of Physics \& Center for Astrophysics,
41: Tsinghua University, Beijing, China} \altaffiltext{3}{Particle Astrophysics Lab.,
42: Institute of High Energy Physics, Chinese Academy of Sciences, Beijing}
43:
44: \begin{abstract}
45: A technique for calculation of variability coherence at different
46: timescales performed directly in the time domain is introduced.
47: Simulations are made to compare the coherence spectrum derived by the
48: time domain technique with the
49: coherence function by Fourier analysis. The results indicate
50: that in comparison with the Fourier analysis the time domain
51: technique is more powerful for revealing signal coherence in noisy data.
52: We apply the time domain technique to the real data of the black
53: hole binaries Cygnus X-1 and GX 339-4 and compare the results with
54: their Fourier coherence spectra.
55: \end{abstract}
56:
57: \keywords{Methods: data analysis --- X-ray: stars --- Stars:binaries}
58:
59: \section{Introduction}
60: The complex aperiodic variability of high-energy emission
61: shown in different time scales is a common character for X-ray binaries
62: and super massive black
63: holes. Temporal analysis is an important approach to study undergoing
64: physics in objects from observed light curves. The spectral lag
65: between variation signals in different energy bands provides useful
66: information on their producing and propagating processes in
67: celestial objects. Besides the time lag, the spectral coherence, a
68: measure of the degree of linear correlation between two light curves
69: in different energy bands, is another useful quantity to provide
70: constraints on models of physical processes in observed objects. For
71: studying variability coherence in the frequency domain, the
72: coherency function -- the normalized average amplitude of the cross
73: power density as a function of Fourier frequency -- can be used.
74: Vaughan and Nowak (1997) have presented a procedure for estimating
75: the coherence function in the presence of counting noise, which has
76: been applied to studying black hole candidates (e.g. Nowak \&
77: Vaughan 1996; Vaughan \& Nowak 1997; Cui, Zhang \& Chen 2000; Nowak
78: et al. 2001; Nowak, Wilms \& Dove 2002; Ji et al. 2003).
79:
80: The time domain quantity corresponding to the coherence in the frequency domain
81: is the coefficient of linear correlation between two light curves
82: in different energy bands.
83: The correlation coefficient has not been extensively applied in
84: temporal analysis as the coherence function in the frequency domain.
85: One of the reasons is that the variability is caused by various physical processes
86: at different timescales,
87: a single value of correlation coefficient
88: from two observed light curves is not enough to provide measures of
89: association between complex variables. What needed is a distribution
90: of correlation coefficient over timescales. Here we present a
91: procedure for deriving the coherence spectrum in the time domain, or
92: a timescale spectrum of the correlation coefficient for a complex process
93: in the presence
94: of background noise and compare the coherence spectrum in the time
95: domain with the Fourier frequency-dependent function of coherence by
96: simulation in \S 2. In our simulation, the result from the time
97: domain technique is consistent with the coherence function obtained
98: from the Fourier analysis when the signal to noise ratio is high. In
99: general, the time domain technique can provide a more accurate and
100: stable result than the Fourier analysis. We apply the time domain
101: technique to the black hole X-ray binaries Cygnus X-1 and GX 339-1
102: and present the results in \S 3 and give discussion on timelag effect
103: in \S 4.
104:
105: \section{Timescale Spectrum of Correlation Coefficient}
106: A proper statistic for measuring the degree of linear correlation
107: between variations in two $N$-point signal counting series $x_s(i)$
108: and $y_s(i)$ is their correlation coefficient
109: \begin{equation}
110: r=\frac{\sum_{i=1}^N(x_s(i)-\overline{x}_s)(y_s(i)-\overline{y}_s)/N}{\sigma(x_s)\sigma(y_s)}.
111: \end{equation}
112: For two observed lightcurves $x(i)=x_s(i)+x_n(i)$ and $y(i)=y_s(i)+y_n(i)$
113: with independent noise $x_n$ and $y_n$ respectively, we have
114: \[ \sum_i(x(i)-\overline{x})(y(i)-\overline{y})=
115: \sum_i(x_s(i)-\overline{x}_s)(y_s(i)-\overline{y}_s), \]
116: and \[
117: \sigma^2(x)=\sigma^2(x_s)+\sigma^2(x_n), \hspace{4mm} \sigma^2(y)=\sigma^2(y_s)+\sigma^2(y_n),
118: \]
119: then the signal's correlation coefficient
120: can be calculated as
121: \begin{equation}
122: r=\frac{\sum_{i=1}^N(x_s(i)-\overline{x}_s)(y_s(i)-\overline{y}_s)/N}{\sigma(x_s)\sigma(y_s)}
123: =\frac{\sum_{i=1}^Ni(x(i)-\overline{x})(y(i)-\overline{y})/N}{\sqrt{\sigma^2(x)-\sigma^2(x_n)}
124: {\sqrt{\sigma^2(y)-\sigma^2(y_n)}}}~.
125: \end{equation}
126: It should be pointed out that one can not take two so called
127: ``background-subtracted lightcurves'' as signal's lightcurves $x_s$ and $y_s$
128: to calculate the correlation coefficient simply by Eq.~(1), because background
129: lightcurves given by a database are usually smoothed curves not including
130: the fluctuation of real noise at shorter timescales.
131:
132: We apply a timescale analysis technique (Li, Feng \& Chen 1999; Li 2001; Li et al. 2004)
133: to calculate correlation coefficients for different timescales from two observed
134: lightcurves.
135: From an original lightcurve $x(j; \delta t) ~(j=1,2,\cdots)$ with a time resolution $\delta t$,
136: we can construct $M$ different lightcurves with the same larger time step $\Delta t = M \delta t$
137: by combining $M$ successive bins
138: \begin{equation}
139: x_{m}(i; \Delta t)=\sum_{j=(i-1)M+m}^{iM+m-1}x(j; \delta t)~, \hspace{5mm}
140: (m=1,\cdots,M)~,
141: \end{equation}
142: where the combination starts from the $m$th bin of the original lightcurve.
143: From two simultaneously observed counting series
144: $x(j;\delta t)$ and $y(j;\delta t)$, the correlation coefficient
145: $r(\Delta t)$ for a certain timescale $\Delta t=M_{\Delta t}\delta t$
146: can be calculated by
147: \begin{equation}
148: r(\Delta t)=\frac{1}{M_{\Delta t}}\sum_{m=1}^{M_{\Delta t}}
149: \frac{\sum_{i=1}^N(x_m(i)-\overline{x}_m)(y_m(i)-\overline{y}_m)/N}{\sqrt{\sigma^2(x_m)-\sigma^2(x_n)}
150: {\sqrt{\sigma^2(y_m)-\sigma^2(y_n)}}}~.
151: \end{equation}
152: The total observation period is divided into $L$ segments. The
153: correlation coefficient $r_k(\Delta t)$ is calculated by Eq.~(4) for
154: a pair of lightcurves on each time segment $k$ separately.
155: The average $\overline{r}(\Delta t)=\sum_{k=1}^Lr_k(\Delta t)/L$
156: and its standard deviation $\sigma(\overline{r})$ are defined as the coherence
157: coefficient for the timescale $\Delta t$ and its error.
158:
159: To check the procedure described above and compare it with the
160: Fourier coherence function we produce two 1000 s signal time series,
161: $x_s$ and $y_s$, with time bin $\delta t=1$~ms sampled from a
162: two-dimensional normal distribution
163: \begin{eqnarray}
164: p(x_s,y_s)&=&\frac{1}{2\pi\sigma(x_s)\sigma(y_s)\sqrt{1-\rho^2}}
165: \exp\left\{-\frac{1}{2(1-\rho^2)}
166: \left[\left(\frac{x_s-\langle x_s\rangle }{\sigma(x_s)}\right)^2\right.\right. \nonumber \\
167: & &\left.\left.-2\rho\frac{x_s-\langle x_s\rangle }{\sigma(x_s)}\frac{y_s-\langle y_s\rangle }{\sigma(y_s)}
168: +\left(\frac{y_s-\langle y_s\rangle }{\sigma(y_s)}\right)^2\right]\right\}~,
169: \end{eqnarray}
170: where $\langle x_s \rangle , \sigma(x_s), \langle y_s \rangle ,
171: \sigma(y_s)$ are means and standard deviations of $x_s$ and $y_s$
172: respectively, and $\rho$ is the correlation coefficient between
173: $x_s$ and $y_s$. By giving these parameters in Eq.~(5), we are able
174: to produce two signal time series with a certain correlation
175: coefficient. To simplify, we specify $\langle x_s \rangle=\langle
176: y_s \rangle=49$, $\sigma(x_s)=\sigma(y_s)=7$. The correlation
177: coefficient $\rho$ in Eq.~(5) is the theoretical value of coherence,
178: and it is used to compare with results of time domain coherence and
179: Fourier coherence function. We add independent Poisson noise $x_n$
180: and $y_n$ on the signal series $x_s$ and $y_s$ respectively to get
181: simulated lightcurves $x=x_s+x_n$ and $y=y_s+y_n$. Different signal
182: to noise ratio (SNR) could be set by adjusting the mean of
183: independent Poisson noise.
184: The simulated lightcurve is divided into ten segments, each has a
185: length of 100 s. For each segment, coherences at different timescales
186: are calculated by Eq.~(4) and Fourier coherence coefficients at different
187: frequencies are also computed with the procedure
188: described by Vaughan and Nowak (1997). Finally, the average
189: and its standard deviation are computed from the ten time domain coherences
190: and Fourier coherence coefficients separately.
191:
192: Figure~1 shows coherence
193: spectra from simulated light curves with $\rho=0.8$ and SNR = 5, 2
194: and 1 separately, where Fourier coherence function in the frequency
195: domain are transformed into the time domain by $time
196: scale=1/frequency$. From Fig.~1 we can see that, in comparing with
197: Fourier coherence functions, time domain coherence distributions are
198: more consistent and accurate. For the case of low SNR, Fourier
199: coherence functions fluctuate seriously and have a large error bar,
200: while the time domain procedure has a better capability to reveal a
201: given coherence, especially at short timescales. Figure 2 show the
202: results for a certain SNR = 1 and different $\rho=0.9$, 0.8 and 0.6
203: respectively, indicating that the time domain procedure is more
204: powerful to reveal lower measured coherence.
205:
206: \begin{figure}[b]
207: \plotone{f1.eps} \caption{Coherence between two simulated
208: lightcurves with a given signal coherence $r^2=0.64$ and different SNR.
209: From left to right, SNR = 5, 2 and 1 respectively.
210: {\sl Plus sign} and {\sl solid line} represent the coherence and
211: its deviation obtained by the time domain technique.
212: {\sl Circle} and {\sl dotted line} show Fourier coherence function. \label{fig1}}
213: \end{figure}
214:
215: \begin{figure}[b]
216: \plotone{f2.eps} \caption{Coherence between two simulated
217: lightcurves with a given SNR = 1 and different signal correlation
218: coefficients. From left to right, the signal correlation coefficient
219: $r^2=0.81$, 0.64, and 0.36 respectively. {\sl Plus sign} and
220: {\sl solid line} represent the coherence and its deviation obtained
221: by the time domain technique. {\sl Circle} and {\sl dotted line}
222: show Fourier coherence function. \label{fig2}}
223: \end{figure}
224:
225: The test and comparison are extended to different signal distributions
226: other than the binormal one.
227: The joint probability density function of binormal distribution can be factorized into two factors
228: \begin{equation}
229: p(x,y)=p(x)p(y|x)~,
230: \end{equation}
231: where
232: \begin{equation} p(x)=\frac{1}{\sqrt{2\pi}\sigma_{x}}\exp\left[-\frac{1}{2}\left(\frac{x-\langle
233: x\rangle}{\sigma_{x}}\right)^2\right]~,
234: \end{equation}
235: and
236: \begin{equation}
237: p(y|x)=\frac{1}{\sqrt{2\pi}\sigma_{y}\sqrt{1-\rho^2}}\exp\left[-\frac{1}{2}\left(\frac{y-\langle
238: y\rangle-\rho\frac{x-\langle x\rangle }{\sigma_{x}}\sigma_{y}
239: }{\sigma_{y}\sqrt{1-\rho^2}}\right)^2\right]~.
240: \end{equation}
241: The first factor $p(x)$ of Eq.~(6) is a one-dimensional Gaussian density function of the variable
242: $x$ which is independent of $y$. The second factor $p(y|x)$ is also
243: a Gaussian density but with average $\overline
244: y+\rho\frac{x-\overline x }{\sigma_{x}}\sigma_{y}$, standard error
245: $\sigma_{y}\sqrt{1-\rho^2}$ and the correlation coefficient $\rho$
246: between x and y, which can be seen as a conditional density of $y$
247: given $x$. With Eq.~(6) we can produce a pair of time series with a given
248: correlation coefficient $\rho$ with a density function
249: $p(x)$ different to Gaussian by sampling $x$ from the density
250: $p(x)$ and $y$ from the density of Eq.~(8). The exponential
251: distribution, uniform distribution and Poisson distribution are
252: adopted as $p(x)$ to produce correlated pairs of time series
253: separately, and then compute their coherence distributions in the
254: time domain and their Fourier correlation functions. All results are
255: similar with Fig.~1 and Fig.~2, showing that the time domain
256: procedure is an effective approach to coherence analysis.
257:
258: \section{X-ray Coherence of Black Hole Binaries}
259: Here we apply the time domain technique of coherence analysis to two
260: black hole candidates, Cyg X-1 in the low state and GX 339-4 in the
261: low and high states. We use observational data from PCA/$RXTE$.
262: PCA/$RXTE$ observation has dead time $10\mu s$ much smaller than
263: time resolution 1ms of original lightcurves we use in this paper.
264: And for Cyg X-1, dead-time affection is about $1\%$ (Maccarone and
265: Coppi 2000), so it could be ignored.
266:
267: Variabilities in lightcurves observed for an astronomical object
268: are usually caused by various processes at different timescales with
269: different coherences. The coherence coefficient $r(\Delta t)$
270: calculated by Eq.~(4) reflects not only the variation property
271: on the timescale $\Delta t$, but is also affected by that
272: on larger ones up to the total time period used in the calculation.
273: To limit the effect of larger timescales we divide the total observation
274: period into segments, each has a duration just 10$\Delta t$.
275: The resultant correlation coefficient should be then related to the
276: limited time scale region of $\Delta t - 10\Delta t$.
277:
278: In our calculation for Cyg X-1 and GX 339-4 data,
279: for a given time bin $\Delta t$ the studied observation period is divided into
280: segments with a duration of 10$\Delta t$ each and the coherence coefficient
281: is calculated for each segment by using Eq.~(4).
282: The all segments are divided into ten groups with the segment numbers
283: in each group being nearly equal to each other.
284: We calculate 10 avarage coherence coefficients $r_k(\Delta t)$ for each group $k~(k=1,\dots,10)$
285: respectively and take their average $r(\Delta t)=\sum_{k=1}^{10}r_k(\Delta t)/10$
286: and the standard deviation $\sigma [r(\Delta t)]$ as the final result
287: of the observation. The Fourier coherence coefficients and their
288: average and the standard deviation of the average are also calculated from
289: the 10 time groups.
290: The results are shown in Figure 3.
291: For both Cyg X-1 and GX 339-4 in
292: the low state shown in the left and middle panels of Fig.~3, both
293: Fourier coherence and time domain coherence show similar trend of
294: change from near-perfect coherence at large time scale to low
295: coherence at small time scale. Two other black hole candidates, XTE
296: J1550-564 (Cui, Zhang and Chen 2000) and GRS 1915+105 (Ji et al.
297: 2003), have Fourier coherence function with near-unity coherence at
298: large time scale and low coherence at small time scale. We also
299: apply the time domain technique to these sources and get similar
300: results. As the coherence is much more easily to be lost than
301: maintained, only when one linear physical process is involved could
302: coherence be preserved (Vaughan \& Nowak 1997). Any nonlinear
303: process or more than one independent linear processes involved could
304: degrade coherence. The near-perfect coherence at large time scale
305: may provide a strong evidence that there is a powerful physical
306: process in these sources with a large characteristic timescale
307: dominates the emission of photons in the studied energy bands.
308:
309: It is a remarkable feature that all the values of time domain
310: coherence are very low in the whole studied timescale range for GX
311: 339-4 in the high state, shown in the right panel of Fig.~3, indicating
312: that there is only little relationship between variabilities of
313: 2-5 keV and 5-7 keV emissions of GX339-4 in its high state.
314:
315: \begin{figure}
316: \plotone{f3.eps} \caption{Variability coherence for black hole X-ray
317: binaries. {\sl Plus}: coherence obtained by the time domain
318: technique. {\sl Circle}: Fourier coherence function. {\sl Left
319: panel}: Cyg X-1 in the low state, PCA/$RXTE$ observation in
320: Dec-17-1996, 2-5 keV vs. 5-13 keV. {\sl Middle panel}: GX 339-4 in
321: the low state, PCA/$RXTE$ observation in Sep-19-1997, 2-6 keV vs.
322: 6-8 keV. {\sl Right panel}: GX 339-4 in the high state, PCA/$RXTE$
323: observation in Jan-17-1998, 2-5 keV vs. 5-7 keV.
324: \label{fig3}}
325: \end{figure}
326:
327: To see whether the relationship between variabilities of two light curves
328: coincides with the time domain coherence computed from them, we
329: simultaneously plot two observed light curves of Cyg X-1 in two energy bands
330: for different time scales at the same time period, shown in Figure~4.
331: From Fig.~4 one can see that the time domain coherence calculated by
332: Eq.~(4) can reflect the real strength of the linear relationship between
333: two light curves.
334:
335: \begin{figure}
336: \plotone{f4.eps} \caption{ Lightcurve segments of Cyg X-1 in the
337: low state, PCA/$RXTE$ observation in Dec-17-1996. $Solid~line$:
338: 2-5 keV, $Dashed~line$: 5-13 keV. {\sl Top panel}: 10 ms time bin, time domain coherence
339: in Fig.~3 is $\sim 0.4$. {\sl Middle panel}: 100 ms time bin,
340: coherence $\sim 0.9$. {\sl Bottom panel}: 1000 ms time bin,
341: coherence is very near to unit.
342: \label{fig4}}
343: \end{figure}
344:
345: \section{Discussion}
346: The Fourier coherence function $r(f)$ of two lightcurves $x$ and $y$ is defined in the space of Fourier frequency
347: \begin{equation} r(f)=\frac{|\left<C(f)\right>|}{\sqrt{\left<|X(f)|^2\right>\left<|Y(f)|^2\right>}}~. \end{equation}
348: Where, $X(f)$ and $Y(f)$ are the Fourier transforms of $x$ and $y$ respectively,
349: the cross spectrum $C(f)=X^*(f)Y(f)$, and
350: angle brackets denote an average over the used segments of light curve.
351: If the ratio of two Fourier transforms at a frequency $f$,
352: $H(f)=X(f)/Y(f)$, is the same for all segments of the two processes,
353: or, equivalently
354: \begin{equation} y(t)=\int h(t-\tau)x(\tau)d\tau~, \end{equation}
355: then $r(f)=1$, the processes are said to be coherent at frequency $f$.
356: But the unit coherence in the time domain,
357: $r(\Delta t)=1$, indicates a linear correlation between the variabilities of intensity
358: in the two energy bands on the timescale $\Delta t$
359: \begin{equation}
360: y(i;\Delta t)=hx(i;\Delta t)~, \hspace{5mm} (i=1,\dots,N_T)~, \end{equation}
361: where $h$ is a constant during the observation period $T=N_T\Delta t$, which is a stronger constraint than
362: Eq.~(10) from an unit coherence, $r(f)=1$, in the frequency domain.
363: It is an advantage for Fourier coherence function that it can obtain
364: coherence at all frequencies simultaneously and exactly, because
365: amplitude at all frequencies could be obtained independently from
366: Fourier transformation of original lightcurve. Improved Fourier
367: coherence function (Vaughan \& Nowak 1997) can calculate coherence
368: of noisy lightcurves and uncertainty as well.
369: While the coherence coefficient calculated in the time domain by Eq.~(4) can be confined
370: in only a timescale region, $\Delta t - N\Delta t$ with $N$ being the number of time bins
371: in a calculated time segment, and the error of the average coherence coefficient is
372: estimated by the statistical deviation in coherence coefficients of
373: different time segments.
374:
375: Spectral time lags are usually existed in observed processes.
376: To calculate the time domain coherence $r(\Delta t)$, one should firstly derive
377: their timescale spectrum of time lag, then make time lag correction for light curves
378: at the timescale $\Delta t$.
379: But, in practice, the coherence derived by the time domain
380: technique is not sensitive to the spectral time lag in analyzing
381: real observational data.
382: From lightcurves of Cyg X-1 in its low
383: state in energy band 2-5 keV and 5-13 keV observed by PCA/$RXTE$, we
384: use the modified cross-correlate function (MCCF) (Li, Feng \& Chen 1999; Li
385: et al. 2004) to calculate the spectral
386: time lag, the result is similar to
387: what obtained from Fourier analysis (Nowak et al 1999).
388: Then we calculate time domain coherences between the two lightcurves
389: with and without time lag correlation separately. The result shown in Fig.~5
390: does not indicate any essential effects of the spectral lag on the time
391: domain coherence. To see how seriously time lag could
392: affect results of time domain coherence, we make a modification
393: to the original observed lightcurves to produce an artificial deviation: the band
394: 5-13 keV lags behind the band 2-5 keV by 10 times of time lag
395: measured from the observed lightcurves by MCCF, then calculate time domain
396: coherences between the modified lightcurves. As we can see in Fig.~5,
397: only a litter deviation can be found between the result from the original
398: lightcurves and the modified ones. We also test other data
399: used in this paper to see whether Fig.~5 is only an exception, the
400: results turns out to be that although the spectral time lag exists
401: in most data, it has just little influence to the coherence obtained
402: by the time domain technique.
403:
404: \begin{figure}
405: \plotone{f5.eps} \caption{X-ray coherence obtained by the time
406: domain technique. Cyg X-1 in the low state, PCA/$RXTE$ observation
407: in Dec-17-1996, 2-5 keV vs. 5-13 keV. {\sl Circle}: from original
408: observed lightcurves. {\sl Cross}: from lightcurves after spectral
409: timelag correction. {\sl Plus}: from modified lightcurves whose timelag is
410: artificially amplified by 10 times. \label{fig5}}
411: \end{figure}
412:
413: To understand reasons of insensitivity of time domain coherence to
414: spectral timelag in observed X-ray data, we make simulation studies.
415: Two 1000 s lightcurves with 1 ms time bin and a given correlation
416: coefficient $\rho=0.8$ (or expected coherence $r^2=0.64$) are
417: produced by sampling from the two-dimensional normal distribution
418: Eq.~(5) described in \S 2. One of the lightcurves is then
419: artificially shifted by 10 ms. We use MCCF to derive timescale
420: distribution of time lags and calculate time domain coherences
421: between the two lightcurves with and without timelag correction. The
422: results are shown in the left panel of Figure~6. We can see that,
423: though the effect of spectral timelag can not be neglected in
424: general, the difference of coherences obtained with and without
425: timelag correction is small at large timescale. For deriving time
426: domain coherences at large timescales, we need to reconstruct
427: original lightcurves with a larger time bin (see the details in \S
428: 2), which impairs the effect of a timelag which is much smaller than
429: the time bin. Spectral timelags measured at a timescale are usually
430: much shorter than the timescale itself. For example, the X-ray time
431: lag of Cyg X-1 between 2-5 keV and 5-13 keV at the timescale of 1 s
432: (or frequency of 1 Hz) is only about 10 ms. From the left panel of
433: Fig.~6 we can see that the effect of timelag on time domain
434: coherence is small when timelag is much smaller than time bin.
435:
436: \begin{figure}
437: \plotone{f6.eps} \caption{Spectral timelag effect on coherence
438: obtained by the time domain technique. {\sl Circle} : no timelag
439: correction. {\sl Plus}: after timelag correction. {\sl Left panel}:
440: coherence between two simulated lightcurves based on two-dimensional
441: normal distribution model with timelag 10ms at all time scale, the
442: given coherence is $r^2=0.64$. {\sl Right panel}: coherence between
443: two simulated lightcurves based on shot model with timelag 10ms at
444: all time scale.
445: \label{fig6}}
446: \end{figure}
447:
448: The correlation existed in
449: successive time bins in observed lightcurves may be another reason of the insensitivity
450: of time domain coherence to spectral timelag in observed X-ray data.
451: The right panel of Fig.~6 shows results from two simulated lightcurves
452: based on shot model with 10 ms timelag.
453: We produce a 1000 s lightcurve with 1 ms time bin consisted of
454: random shots similar to what observed from Cyg X-1 (Feng, Li \& Chen 1999).
455: The front of the shot is represented by a single exponential and
456: the time constant of both ascending and descending
457: front is 70 ms. The interval between two successive shots follows an
458: exponential distribution with an average of 450 ms.
459: The above produced lightcurve is shifted by 10 ms to create the second one.
460: The right panel of Fig.~6 shows coherence spectra derived by
461: the two lightcurves with and without timelag correction.
462: From Fig.~6 we see that the timelag effect on time domain coherence
463: for lightcurves produced by shots (right panel)
464: is much smaller than that for lightcurves sampled from two-dimensional
465: normal distribution (left panel).
466: This is because successive bins in a lightcurve consisted of shots have a strong
467: connection.
468:
469: In summary, we propose a time domain technique in this paper to calculate
470: variability coherence for different timescales. In general, the time domain
471: coherence distribution is
472: consistent with the Fourier coherence function, but more reliable
473: for the case of low signal to noise ratio and more powerful to
474: reveal lower coherence.
475: In practical data analysis, the
476: effect of spectral time lags on the time domain coherence can be
477: neglected. This technique provides an useful tool for temporal
478: analysis.
479:
480: \acknowledgments
481: The referee is thanked for helpful comments and suggestions.
482: This research was supported by the National Natural
483: Science Foundation of China and made use
484: of data obtained through the High Energy Astrophysics Science Archive Research
485: Center Online Service,
486: provided by the NASA/Goddard Space Flight Center.
487:
488: \begin{thebibliography}{}
489: \bibitem[Bendat, J., \& Piersol, A. 1986]{} Bendat, J., \& Piersol, A.
490: 1986, Random Data: Analysis and Measurement Procedures
491: \bibitem[Cui, Zhang \& Chen 2000]{cui00}Cui W., Zhang S.N. \& Chen W. 2000, \apjl,
492: 531, L45
493: \bibitem[Feng, Li \& Chen 1999]{Feng99}Feng Y.X., Li, T.P. \& Chen L. 1999, ApJ, 514,
494: 373
495: \bibitem[Ji et al. 2003]{ji00}
496: Ji J.F., Zhang S.N., Qu J.L. \& Li T.P. 2003, \apjl, 584, L23
497: \bibitem[Li T.P. 2001]{li01}Li T.P. 2001, \cjaa, 1, 313; astro-ph/0109468
498: \bibitem[Li, Feng \& Chen 1999]{li99}Li, T.P., Feng Y.X. \& Chen L. 1999, \apj, 521, 789
499: \bibitem[Li T.P. et al. 2004]{li04}
500: Li T.P., Qu J.L., Feng H., Song L.M., Ding G.Q. \& Chen L. 2004,
501: \cjaa, 4, 583; astro-ph/0407458
502: \bibitem[Maccarone \& Coppi]{Mac2000} Maccarone,T.J \& Coppi,P.S. 2000, \apjl, 537,
503: L107
504: \bibitem[Nowark \& Vaughan 1996]{now96} Nowark,M.A. \& Vaughan,B.A. 1996, \mnras, 280, 227
505: \bibitem[Nowark et al. 1999]{now99}Nowark, M.A., Vaughan, B.A., Wilms, J., Dove, J.B.
506: \& Begelman, M.C. 1999, \apj, 510, 874
507: \bibitem[Nowark et al. 2001]{now01}Nowark,M.A., Wilms, J., Heindl W.A., Pottschmidt K.,
508: Dove,J.B. \& Begelman,M.C. 2001, \mnras, 320, 316
509: \bibitem[Nowark,M.A., Wilms,J. and Dove,J.B.(2002)]{Now02} Nowark,M.A., Wilms,J. \& Dove,J.B. 2002, \mnras,
510: 332, 856
511: \bibitem[Vaughan,B.A. and Nowark,M.A.(1997)]{vau96} Vaughan,B.A. \& Nowark,M.A. 1997, \apjl,
512: 474, L43
513:
514:
515: \end{thebibliography}
516:
517:
518: \end{document}
519: