1: \documentclass[12pt,preprint]{aastex}
2: %\documentclass[manuscript]{aastex}
3: %\documentclass[preprint2]{aastex}
4: %\documentclass{aastex}
5: %\usepackage{emulateapj5}
6:
7: \def\xr {X--ray}
8: \newcommand{\gcc}{g~cm$^{-3}\ $}
9: \newcommand{\sfun}[2]{$#1(#2)\ $}
10: \newcommand{\rhonot}{$\rho_{\circ}\ $}
11: \newcommand{\msun}{$M_{\odot}\ $}
12: \newcommand{\greq}{$\stackrel{>}{ _{\sim}}$}
13: \newcommand{\lteq}{$\stackrel{<}{ _{\sim}}$}
14: \newcommand{\etal}{{\it et al.}}
15: \newcommand{\lsim}{\raisebox{-0.3ex}{\mbox{$\stackrel{<}{_\sim} \,$}}}
16: \newcommand{\gsim}{\raisebox{-0.3ex}{\mbox{$\stackrel{>}{_\sim} \,$}}}
17: \def\gta{\ifmmode {\mathbin{\lower 3pt\hbox %> or of order
18: {$\,\rlap{\raise 5pt\hbox{$\char'076$}}\mathchar"7218\,$}}}
19: \else {${\mathbin{\lower 3pt\hbox
20: {$\rlap{\raise 5pt\hbox{$\char'076$}}\mathchar"7218\,$}}}
21: $}\fi}
22: \def\lta{\ifmmode {\,\mathbin{\lower 3pt\hbox %< or of order
23: {$\,\rlap{\raise 5pt\hbox{$\char'074$}}\mathchar"7218\,$}}}
24: \else {${\mathbin{\lower 3pt\hbox
25: {$\rlap{\raise 5pt\hbox{$\char'074$}}\mathchar"7218\,$}}}
26: $}\fi}
27:
28:
29: \shorttitle {Thermonuclear flame spreading on neutron stars}
30: \shortauthors {Bhattacharyya \& Strohmayer}
31:
32: \begin{document}
33:
34: \title{Spreading of thermonuclear flames on the neutron star in SAX
35: J1808.4--3658: an observational tool}
36:
37: \author {Sudip Bhattacharyya\altaffilmark{1,2}, and Tod
38: E. Strohmayer\altaffilmark{2}}
39:
40: \altaffiltext{1}{Department of Astronomy, University of Maryland at
41: College Park, College Park, MD 20742-2421}
42:
43: \altaffiltext{2}{X-ray Astrophysics Lab,
44: Exploration of the Universe Division,
45: NASA's Goddard Space Flight Center,
46: Greenbelt, MD 20771; sudip@milkyway.gsfc.nasa.gov,
47: stroh@clarence.gsfc.nasa.gov}
48:
49: \begin{abstract}
50:
51: We analyse archival Rossi X-Ray Timing Explorer (RXTE) proportional
52: counter array (PCA) data of thermonuclear X-ray bursts from the 2002
53: outburst of the accreting millisecond pulsar SAX J1808.4--3658. We
54: present evidence of nonmonotonic variations of oscillation
55: frequency during burst rise, and correlations among the time
56: evolution of the oscillation frequency, amplitude, and the inferred
57: burning region area. We also discuss that the amplitude and burning
58: region area evolutions are consistent with thermonuclear flame
59: spreading on the neutron star surface. Based on this discussion, we
60: infer that for the 2002 Oct. 15 thermonuclear burst, the ignition
61: likely occured in the mid-latitudes, the burning region took $\sim
62: 0.2$ s to nearly encircle the equatorial region of the neutron star,
63: and after that the lower amplitude oscillation originated from the
64: remaining asymmetry of the burning front in the same hemisphere where
65: the burst ignited. Our observational findings and theoretical
66: discussion indicate that studies of the evolution of burst oscillation
67: properties during burst rise can provide a powerful tool to understand
68: thermonuclear flame spreading on neutron star surfaces under extreme
69: physical conditions.
70: \end{abstract}
71:
72: \keywords{equation of state --- methods: data analysis --- stars:
73: neutron --- X-rays: binaries --- X-rays: bursts --- X-rays: individual
74: (SAX J1808.4--3658)}
75:
76: \section {Introduction} \label{sec: 1}
77:
78: X-ray bursts are produced by thermonuclear burning of matter
79: accumulated on the surfaces of accreting neutron stars (Woosley, \&
80: Taam 1976; Lamb, \& Lamb 1978). During many bursts, millisecond period
81: brightness oscillations are generated by the combination of rapid
82: stellar rotation and an asymmetric brightness pattern on the neutron
83: star surface (Strohmayer, \& Bildsten 2003). The period of these
84: oscillations is very close to the stellar spin period (Chakrabarty et
85: al. 2003; Strohmayer et al. 2003). Moreover, as this timing feature
86: originates from the surface of the neutron star, its detailed modeling
87: may be useful to constrain stellar structure parameters, and hence the
88: equation of state models of the dense matter in the neutron star core
89: (Bhattacharyya et al. 2005; Miller, \& Lamb 1998; Nath, Strohmayer, \&
90: Swank 2002; Muno, \"Ozel, \& Chakrabarty 2002).
91:
92: Modeling of burst
93: oscillations can also be useful to understand neutron star
94: atmospheres, surface fluid motions, and for mapping the magnetic field
95: structure on the stellar surface. For example, the evolution of
96: frequency of these oscillations during the burst rise phase, may
97: provide information on the spreading of thermonuclear flames under the
98: extreme physical conditions that exist on neutron stars (e.g.,
99: Bhattacharyya \& Strohmayer 2005). This is because bursts almost
100: certainly ignite at a particular point on the stellar surface (as
101: simultaneous ignition over the whole surface would require very fine
102: tuning), and then spread to burn all the surface fuel (Fryxell \&
103: Woosley 1982; Cumming \& Bildsten 2000; Spitkovsky, Levin, \&
104: Ushomirsky 2002; Bhattacharyya \& Strohmayer 2006a). This slow
105: (compared to the rotational speed) movement and spreading of the
106: burning region (along with other physical effects such as the
107: increased scale height due to burning) may give rise to complex
108: frequency evolution of the observed burst oscillations. This
109: spreading would also cause the observed burst intensity to increase,
110: and the oscillation amplitude to decrease. Moreover, the increase in
111: emission area can be estimated by spectral analysis (Strohmayer, Zhang
112: \& Swank 1997; Bhattacharyya \& Strohmayer 2006a; 2006b). Therefore,
113: simultaneous modeling of the evolution of burst intensity, oscillation
114: frequency and amplitude, and spectral properties can, in principle, be
115: a powerful tool to understand the propagation of burning fronts on
116: neutron star surfaces under conditions of extreme radiative pressure,
117: magnetic field and gravity.
118:
119: Frequency evolution during burst rise oscillations has so far been
120: observed from two low mass X-ray binary (LMXB) sytems: SAX
121: J1808.4--3658 and 4U 1636--536 (Chakrabarty et al. 2003; Bhattacharyya
122: \& Strohmayer 2005). The Rossi X-ray Timing Explorer (RXTE) observed
123: the 401 Hz X-ray pulsar SAX J1808.4--3658 in October and November of
124: 2002, when it was in outburst. Four type I X-ray bursts were detected
125: during these observations (Chakrabarty et al. 2003), three of which
126: showed strong millisecond period brightness oscillations during burst
127: rise. A previous study found that as the burst intensity rises, the
128: oscillation frequency also increases by $\sim 5$~Hz and may overshoot
129: the stellar spin frequency (Chakrabarty et al. 2003). Here, we
130: analyse these archival data in order to model the time evolution of
131: different burst properties and search for correlations among them.
132:
133: In our study we find evidence of nonmonotonic variations in the oscillation
134: frequency during the rising phase of bursts. This is the first report
135: of such variations from any source. The frequency modulation is
136: correlated with the evolution of oscillation amplitude and burning
137: region area. In \S~3, we discuss that
138: the correlated amplitude and area evolution is consistent with
139: thermonuclear flame propagation on the neutron star surface.
140:
141: \section {Data Analysis and Results} \label{sec: 2}
142:
143: We analyse the archival RXTE proportional counter array (PCA) data of
144: the 2002 outburst from SAX J1808.4--3658. Three thermonuclear bursts
145: with significant millisecond oscillations during the rising phase are
146: found in the ObsIds: 70080-01-01-000 (Oct 15), 70080-01-02-000 (Oct
147: 18), and 70080-01-02-04 (Oct 19). First we explore the frequency
148: evolution of burst rise oscillations during these bursts using three
149: procedures. We calculate dynamic power spectra (Strohmayer \&
150: Markwardt 1999) with time sampling short enough to resolve the burst
151: rise interval, but large enough to accumulate sufficient signal power
152: above the noise level. The dynamic spectra (panel {\it a} of Fig. 1,
153: and Fig. 2) provide indications of the frequency evolution
154: behavior. To confirm the indications in the dynamic spectra, we carry
155: out a phase timing analysis (Muno et al. 2000). We divide the burst
156: rise time interval into several bins of a fixed chosen length, and
157: then assuming a frequency evolution model, we calculate the average
158: phase $(\psi_k)$ in each bin $(k)$. The corresponding $\chi^2$ is
159: calculated using the formula $\chi^2 = \sum^M_{k=1}
160: (\psi_k-\bar\psi_k)^2/\sigma^2_{\psi_k}$ (Strohmayer \& Markwardt
161: 2002), where $M$ is the number of bins. For this study we used
162: extensive burst rise simulations to evaluate the uncertainty,
163: $\sigma_{\psi_k}$, as a function of the $Z^2$ power in each time bin.
164: We find the best fit parameter values for various frequency evolution
165: models by minimizing $\chi^2$, and we calculate the uncertainty in
166: each parameter by finding the change which produces the appropriate
167: increase in $\chi^2$ (Strohmayer \& Markwardt 2002; Press et
168: al. 1992). Finally, we calculate the total $Z^2$ power (Strohmayer \&
169: Markwardt 2002) for the entire rise interval (ie. without binning)
170: using the best fit frequency evolution model parameters, and ensure
171: that this power is close to the maximum power obtained from any
172: parameter values.
173:
174: We first fit the oscillations during the rising phase of the Oct. 15
175: burst with a constant frequency model. This gives a best fit frequency
176: of $400.91$~Hz and $\chi^2$/dof $= 30.04/8$. We evaluate the
177: significance of this $\chi^2$ value using simulations, and find a
178: probability of 0.005 to obtain such a value by chance. Since the
179: constant frequency model does not describe the data well, we next
180: consider more complex models. These are; (1) linear frequency
181: increase, (2) second order polynomial, and (3) linear increase and
182: subsequent linear decrease models (see Table 1 for a description of
183: the models). These models give $\chi^2$/dof values of $17.73/7$,
184: $17.35/6$, and $11.68/5$ respectively. Although these models give
185: better fits, they still have reduced $\chi^2$ values $> 2$, and do not
186: describe the data very well. A slightly more complex model which fits
187: the data well has a linear frequency increase followed by a second
188: order polynomial, and gives a $\chi^2$/dof value of 3.36/4. Of the
189: models tested we consider this the best description of the Oct. 15
190: burst, and the best fit parameter values are given in Table 1. Panel
191: {\it a} of Fig. 1 shows that this model is consistent with the dynamic
192: power contours, and it indicates a frequency increase (by a few Hertz)
193: for the first $\sim 0.2$ s from burst onset, then a frequency decrease
194: (by $\sim 1$~Hz), and a subsequent increase. From Table 1, we note
195: that the model parameters $\nu_{\rm 0}$ \& $\dot\nu_{\rm 1}$ are
196: practically unconstrained from the lower and upper sides respectively.
197: This implies that the initial frequency increase can be well fit by a
198: very steep model. However, the other sides of these parameters are
199: reasonably well constrained, which shows that a constant frequency
200: model is insufficient to model this portion. Next we fit the rise
201: oscillations from the Oct. 18 \& Oct. 19 bursts with the same
202: frequency evolution models. The constant frequency model for the
203: Oct. 18 burst gives a $\chi^2$/dof $= 154.51/8$, and hence can be
204: strongly rejected. This confirms the conclusions of Chakrabarty et
205: al. (2003) who first noted the large frequency increase present during
206: this burst. A linear frequency increase model gives a $\chi^2$/dof $=
207: 19.35/7$ that, though better, is still uncomfortably large to be
208: acceptable. The next more complex model (a constant frequency, \&
209: subsequent linear increase) gives a $\chi^2$/dof $= 8.44/6$ (see Table
210: 1, and the upper panel of Fig. 2), and is statistically
211: acceptable. However, we note that the dynamic power contours (upper
212: panel, Fig. 2) are suggestive of an initial frequency increase,
213: decrease, and increase behavior (shown by the dotted curve)
214: qualitatively similar to that for the Oct. 15 burst. Although higher
215: signal to noise ratios per time bin would likely be required to
216: confirm this behavior, this (dotted curve) model does give a higher
217: total $Z^2$ power than that given by the best fit $\chi^2$ model
218: (solid curve) in Table 1.
219: Finally, the constant frequency model for the Oct. 19 burst gives a
220: $\chi^2$/dof $= 22.03/8$, which is also unacceptably high. A linear
221: frequency increase model (see Table 1, and the lower panel of Fig. 2)
222: gives a $\chi^2$/dof $= 6.46/7$, and is acceptable for this burst.
223: We note that although the oscillation frequency behavior of the
224: Oct. 19 burst is different from that of the Oct. 15 \& 18 bursts, the
225: former burst seems otherwise similar to the latter ones. For example,
226: all three bursts show photospheric radius expansion, and their rise
227: times and durations are $\le 1$ s and a few tens of seconds
228: respectively. Thus, the variations in inferred frequency evolution
229: must be associated with some variable which does not drastically alter
230: the gross properties of the burst. One possible variable might be the
231: initial lattitude of ignition, although large variations in this
232: quantity might be expected to affect other burst properties, such as
233: the rise time, as well. Panel {\it b} of Fig. 1 shows the rms
234: amplitude variation with time during the rise of the Oct. 15 burst.
235: From the beginning of the burst, the amplitude decreases for $\sim
236: 0.2$ s, and then assumes a near constant value, with some
237: fluctuations. This behavior is qualitatively similar to that seen in
238: bursts from the LMXB systems 4U 1728-34 and 4U 1636--536 (Strohmayer,
239: Zhang \& Swank 1997; Bhattacharyya \& Strohmayer 2005). We also
240: perform time resolved spectral fitting (using a blackbody model)
241: during the rise of the Oct. 15 burst. The inferred source radius
242: (which provides some relative indication of the size of the burning
243: region on the stellar surface) shows evidence for a modest increase
244: for the first $\sim 0.2$ s, and then remains almost constant (panel
245: {\it c}, Fig. 1). Panel {\it d} gives the corresponding source
246: temperature variation.
247:
248: \section {Discussion} \label{sec: 3}
249:
250: Taken together our results for the three bursts from SAX J1808.4--3658
251: suggest that the oscillation frequency can evolve in a complex manner
252: during burst rise. In all cases a constant frequency model is a poor
253: description of the oscillations. In two bursts (Oct. 15 and 18) the
254: data can best be described by a complex modulation in frequency
255: whereby it initially increases, then decreases before increasing
256: again. Moreover, the oscillation amplitude and the inferred burning
257: region area are found to be correlated with the frequency. We now
258: discuss from a theoretical perspective how spreading of thermonuclear
259: flames can plausibly account for these observations. In our
260: discussions we use the observational results mostly from the Oct. 15
261: burst as characteristic. The salient features of this burst are as
262: follows: (1) from the start of the burst, the oscillation frequency
263: and burning region area increase and the oscillation amplitude
264: decreases for $\sim 0.2$ s; (2) after $\sim 0.2$ s, the frequency
265: first decreases and then increases, and both amplitude and burning
266: region area reach a nearly constant value (with some fluctuations).
267:
268: The burst begins when the fuel (i.e., accumulated matter) ignites at a
269: particular point, and then the flame propagates over the surface
270: (Fryxell \& Woosley 1982; Spitkovsky et al. 2002; Bhattacharyya \&
271: Strohmayer 2006a; 2006b). Before spreading has engulfed the entire
272: star, temperature variations due to surface waves may not be able to
273: explain the brightness oscillation or its frequency evolution (as has
274: been proposed for the burst tail oscillations; Heyl 2005; Lee \&
275: Strohmayer 2005), as the rapid spreading and temperature increase may
276: wash out this effect. This explanation of oscillations during burst
277: rise was also shown to be unfavored by Bhattacharyya \& Strohmayer
278: (2005). However, thermonuclear burning in a limited portion (hot
279: spot) of the neutron star surface can give rise to these oscillations,
280: and the propagation of the burning front may explain the observed time
281: evolution of oscillation properties.
282:
283: In order to understand how thermonuclear flame spreading can give rise
284: to the observed evolution of oscillation amplitude and burning region
285: area, we first review some relevant results from previous work
286: (Spitkovsky et al. 2002): (1) the greater scale height of the burning
287: region than the cold fuel gives rise to a shearing speed (as the
288: horizontal pressure gradient in the burning front increases with
289: height). As a result, the cold fuel is drawn into the burning front
290: and ignited. This enables the burning front to propagate. (2) The
291: shearing speed is greater nearer the equator than the pole (due to the
292: latitude dependence of the Coriolis parameter; Spitkovsky et
293: al. 2002). Thus, the burning front propagates with the shearing speed
294: $(\vartheta)$ (assuming the mixing time scale is very small;
295: Spitkovsky et al. 2002; Fujimoto 1988; Cumming \& Bildsten 2000).
296: Now, if the fuel ignites at a mid-latitude (say, in the northern
297: hemisphere), it will propagate faster towards the equator (than the
298: pole), and the east-west width of the burning region will increase
299: much faster near the equator. Therefore, after a certain time, the
300: burning region will encircle the equator and propagate more or less
301: symmetrically towards the south pole. The northern burning front
302: propagates towards the north pole, keeping the asymmetry (due to the
303: variation of east-west width with latitude; see Figure 8 of Spitkovsky
304: et al. 2002), which vanishes near the pole. At the beginning, the
305: burning region is relatively small, and hence the oscillation
306: amplitude can be large. As the burning region grows, the oscillation
307: amplitude naturally diminishes (Fig. 1). This effect was also reported
308: for bursts from other sources (Strohmayer, Zhang, \& Swank
309: 1997). After the burning region encircles the equator with a
310: considerable north-south width, the observed burning area does not
311: increase much (hence the near constant radius after the initial
312: increase; see Fig. 1). From this time, the oscillation is due to the
313: residual asymmetry of the northern burning front (with the persistent
314: background due to the azimuthally symmetric portion of the burning
315: region), and hence the amplitude attains a near constant value till
316: the asymmetry vanishes. Therefore, according to our explanation, for
317: the Oct. 15 burst, the burning region takes $\sim 0.2$ s to nearly
318: encircle the stellar equator. However, we note that this explanation
319: does not include the effects of magnetic field, which may affect the
320: burning front propagation considerably (see below).
321:
322: Thermonuclear flame propagation may give rise to the observed
323: frequency evolution, if the eastbound and the westbound burning fronts
324: have different accelerations, causing acceleration (either eastwards
325: or westwards) of the center of the burning region relative to the
326: star.
327: %But a possible problem with this picture is, a burst
328: %oscillation frequency, which is considerably different from the
329: %stellar spin frequency (by a few Hertz), would mean a rapid flame
330: %spreading in the azimuthal direction. Such a fast spreading will make
331: %the burning region to encircle the star very quickly (in a fraction of
332: %a second) removing any azimuthal asymmetry, and hence causing the
333: %oscillations to disappear. This may make the observed oscillations
334: %during the later part of burst rise difficult to explain. However, a
335: %residual asymmetry (as discussed in the previous paragraph; Spitkovsky
336: %et al. 2002) can solve this problem by giving rise to low amplitude
337: %oscillations in the later phase of burst rise, and gives confidence to
338: %the theoretical model of Spitkovsky et al. (2002).
339: Now considering this picture of the azimuthal shift of the hot spot
340: center, an observed oscillation frequency lower than the stellar spin
341: frequency may be caused by the westward motion of the burning region
342: center (for an eastward rotating star). An increased scale height in
343: this region may cause such motion, because, as the hot portion of the
344: burning region puffs up, its top portion slips westwards to conserve
345: angular momentum. If the shearing speed due to this is $v$, the center
346: of the burning region moves westward with a speed $v$ relative to the
347: stellar surface. However, this effect alone can not explain an
348: oscillation frequency that is more than $\sim 2$ Hz lower than the
349: stellar spin frequency (Cumming \& Bildsten 2000). Therefore, as the
350: initial oscillation frequency of the Oct. 15 burst is $\sim 3-4$ Hz
351: less than the neutron star spin frequency, this physical effect can
352: only partially explain the initial frequency of this burst. Moreover,
353: the increased scale height effect can not explain the following
354: observed features: (1) initial quick increase of frequency (as the
355: scale height does not decrease during the beginning of burst rise),
356: (2) oscillation frequency overshooting the stellar spin frequency, and
357: (3) a decrease and subsequent increase of oscillation frequency during
358: the later stage of the burst rise. A physical effect of variable
359: magnitude, that can move the center of the expanding burning region
360: (hot spot) both eastwards and westwards, is required to explain these
361: aspects. This effect can add to the increased scale height effect at
362: the burst onset (if moving the hot spot center westwards), and can
363: compete with the scale height effect to produce the other observed
364: features of frequency evolution (if moving the hot spot center
365: eastwards). At present, such an effect is not known, but the surface
366: magnetic field may be a promising candidate. This is because the
367: surface magnetic field may have its strength amplified and its
368: geometry modified by differential rotation during the flame spreading.
369: This may enhance the magnetic force, which can subsequently modify the
370: shearing flows, and hence can influence the propagation of the burning
371: front (e.g., Cumming et al. 2002). The exact nature and magnitude of
372: this influence will depend on the geometry and the strength of the
373: field. This effect may be particularly important for SAX
374: J1808.4--3658, as this source is a pulsar (and hence probably has a
375: higher magnetic field than non-pulsars), and because the oscillation
376: frequency overshooting the neutron star spin frequency has been
377: observed only from this source. This source also seems different from
378: other sources in other properties such as the quick and large increase
379: of oscillation frequency during burst rise. However, we note that the
380: uniqueness of SAX J1808.4--3658 in terms of burst rise oscillation
381: frequency evolution is not yet well proven, as so far only seven
382: bursts with significant frequency evolution during the rising phase
383: have been observed (three from this source, and four from 4U
384: 1636--536; Bhattacharyya \& Strohmayer 2005). Nevertheless, detailed
385: numerical simulations of thermonuclear flame spreading (including
386: magnetic field effects) may be able to explain the observed correlated
387: evolution of oscillation properties during burst rise, and may provide
388: important information about how the flame speed and other flame
389: spreading properties can be influenced by stellar spin, compactness,
390: magnetic field and burst strength. Therefore, we emphasize that burst
391: rise oscillations provide a potentially powerful tool to understand
392: thermonuclear flame spreading on neutron star surfaces under extreme
393: physical conditions.
394:
395: \acknowledgments
396:
397: \clearpage
398:
399: \begin{thebibliography}{}
400:
401: %\bibitem[]{413} Belian, R. D., Conner, J. P., \& Evans, W. D. 1976, \apj,
402: %206, L135.
403:
404: \bibitem[]{419} Bhattacharyya, S., \& Strohmayer, T. E. 2005, \apj,
405: 634, L157.
406:
407: \bibitem[]{416} Bhattacharyya, S., \& Strohmayer, T. E. 2006a, \apj,
408: 636, L121.
409:
410: \bibitem[]{422} Bhattacharyya, S., \& Strohmayer, T. E. 2006b, \apj, in press.
411:
412: \bibitem[]{424} Bhattacharyya, S., Strohmayer, T. E., Miller, M. C. \&
413: Markwardt, C. B. 2005, \apj, 619, 483.
414:
415: \bibitem[]{427} Chakrabarty, D. et al. 2003, Nature, 424, 42.
416:
417: \bibitem[]{429} Cumming, A., \& Bildsten, L. 2000, \apj, 544, 453.
418:
419: \bibitem[]{} Cumming, A., Morsink, S. M., Bildsten, L., Friedman,
420: J. L., \& Holz, D. E. 2002, 564, 343.
421:
422: \bibitem[]{431} Fujimoto, M. Y. 1988, A\&A, 198, 163.
423:
424: \bibitem[]{} Fryxell, B. A., \& Woosley, S. E. 1982, \apj, 261, 332.
425:
426: %\bibitem[]{433} Grindlay, J. E. et al. 1976, \apj, 205, L127.
427:
428: \bibitem[]{435} Heyl, J. S. 2005, \mnras, 361, 504.
429:
430: %\bibitem[]{437} Joss, P. C. 1977, Nature, 270, 310.
431:
432: \bibitem[]{439} Lamb, D. Q., \& Lamb, F. K. 1978, \apj, 220, 291.
433:
434: \bibitem[]{441} Lee, U., \& Strohmayer, T. E. 2005, \mnras, 361, 659.
435:
436: \bibitem[]{443} Miller, M. C. \& Lamb, F. K. 1998, \apj, 499, L37.
437:
438: \bibitem[]{445} Muno, M. P., Fox, D. W., Morgan, E. H. \& Bildsten,
439: L. 2000, \apj, 542, 1016.
440:
441: \bibitem[]{448} Muno, M.~P., {\"O}zel, F., \& Chakrabarty, D.\ 2003, \apj,
442: 595, 1066.
443:
444: \bibitem[]{451} Nath, N. R., Strohmayer, T. E. \& Swank, J. H. 2002,
445: \apj, 564, 353.
446:
447: \bibitem[]{454} Press, W. H., Teukolsky, S. A., Vetterling, W. T., \&
448: Flannery, B. P. 1992, Numerical Recepies in FORTRAN (New York: Cambridge
449: University Press), 687-693.
450:
451: %\bibitem[]{434} Psaltis, D., \& Chakrabarty, D. 1999, \apj, 521, 332.
452:
453: \bibitem[]{458} Spitkovsky, A., Levin, Y. \& Ushomirsky, G. 2002,
454: \apj, 566, 1018.
455:
456: \bibitem[]{460} Strohmayer, T. E., \& Bildsten, L. 2003, in {\it Compact
457: Stellar X-ray Sources}, Eds. W.H.G. Lewin and M. van der Klis,
458: (Cambridge University Press: Cambridge), (astro-ph/0301544).
459:
460: \bibitem[]{464} Strohmayer, T. E. \& Markwardt, C. B. 1999, \apj, 516, L81.
461:
462: \bibitem[]{466} Strohmayer, T. E. \& Markwardt, C. B. 2002, \apj, 577, 337.
463:
464: \bibitem[]{389} Strohmayer, T. E., Markwardt, C.B., Swank, J. H., \&
465: in 't Zand, J. 2003, \apj, 596, L67.
466:
467: \bibitem[]{471} Strohmayer, T. E., Zhang, W. \& Swank, J. H. 1997,
468: \apj, 487, L77.
469:
470: \bibitem[]{473} Woosley, S. E., \& Taam, R. E. 1976, Nature, 263, 101.
471:
472: \end{thebibliography}{}
473:
474: \clearpage
475:
476: \begin{deluxetable}{cccccccc}
477: \tablecolumns{8} \tablewidth{0pc} \tablecaption{Frequency evolution
478: model parameters\tablenotemark{a} (with 1$\sigma$ error) for burst
479: rise oscillations of three bursts from SAX J1808.4--3658.}
480: \tablehead{Burst date & $\nu_{\rm 0}$ & $\dot\nu_{\rm 1}$ & $t_{\rm
481: b_{\rm 1}}$ & $\dot\nu_{\rm 2}$ & $\ddot\nu_{\rm 2}$ & $\chi^{\rm
482: 2}$/dof & $Z^2_1$\tablenotemark{b}} \startdata 2002 Oct 15 &
483: $396.72^{+2.00}_{-21.73}$ & $29.64^{+236.21}_{-13.95}$ &
484: $0.20^{+0.05}_{-0.10}$ & $-15.54^{+4.73}_{-4.94}$ &
485: $23.67^{+11.52}_{-9.81}$ & $3.36/4$ & $152.16$ \\ \\ 2002 Oct 18 &
486: $398.22^{+0.20}_{-0.21}$ & -- & $0.31^{+0.04}_{-0.04}$ &
487: $19.21^{+5.19}_{-3.39}$ & -- & $8.44/6$ & $62.83$ \\ \\ 2002 Oct 19 &
488: $399.64^{+0.23}_{-0.14}$ & $1.69^{+0.07}_{-0.42}$ & -- & -- & -- &
489: $6.46/7$ & $86.64$ \\ \enddata \tablenotetext{a}{Frequency evolution
490: model: $\nu(t) = \nu_{\rm 0} + \dot\nu_{\rm 1} t$ (for $t \le t_{\rm
491: b_{\rm 1}}$, and $\nu_{\rm 0} = \nu(0))$; $\nu(t) = \nu(t_{\rm b_{\rm
492: 1}}) + \dot\nu_{\rm 2} (t-t_{\rm b_{\rm 1}}) + \ddot\nu_{\rm 2}
493: (t-t_{\rm b_{\rm 1}})^2$ (for $t \ge t_{\rm b_{\rm 1}})$.}
494: \tablenotetext{b}{Fundamental power during burst rise.}
495: \end{deluxetable}
496:
497: \clearpage
498: \begin{figure}
499: \hspace{-1.1 cm}
500: \epsscale{0.9}
501: \plotone{f1.ps}
502: \vspace{-3.5 cm}
503: \caption {Time evolution of different observed burst properties during
504: the rise of the Oct 15 burst from SAX J1808.4--3658. Panel {\it a}
505: gives the detected intensity (histogram), power contours (minimum and
506: maximum power values are $16$ and $51$) from the dynamic power spectra
507: (for 0.15 s duration at 0.01 s intervals), the best fit model from
508: Table 1, and the neutron star spin frequency (broken horizontal line).
509: Panel {\it b} shows the rms amplitude of the oscillations. Here the
510: horizontal lines give the binsize. Panel {\it c} gives the inferred
511: radius (assuming 10 kpc source distance) of the source, while panel
512: {\it d} gives the corresponding temperature. For panels {\it b}, {\it
513: c}, and {\it d}, persistent emission is subtracted and a deadtime
514: correction is applied, and the error bars are $1\sigma$ values.}
515: \end{figure}
516:
517: \clearpage
518: \begin{figure}
519: %\hspace{1.3 cm}
520: \epsscale{0.8}
521: \plotone{f2.ps}
522: \vspace{-3.0 cm}
523: \caption {Similar to panel {\it a} of Fig. 1 but for the Oct 18
524: (upper) and Oct 19 (lower) bursts. For the upper panel dynamic power
525: spectra are calculated for 0.15 s duration at 0.01 s intervals, while
526: for the lower panel these numbers are 0.2 s and 0.01 s. For the
527: calculation of power contours, minimum and maximum power values are
528: 15 and 50 for the upper panel, and 17 and 111 for the lower panel. The
529: dotted line in the upper panel gives the frequency evolution for which
530: the total power $(74)$ is higher than that $(63)$ corresponding to the
531: best fit model (solid line) from Table 1.}
532: \end{figure}
533:
534: \end{document}
535:
536: