1: \documentstyle[epsfig,mncite]{mn}
2: %\renewcommand{\baselinestretch}{2}
3:
4:
5: \begin{document}
6: \LARGE \normalsize \title{KiloHertz quasi--periodic oscillations
7: difference frequency exceeds inferred spin frequency in 4U~1636--53}
8:
9: \author[P. G. Jonker et al.]
10: {Peter G. Jonker$^1$ \thanks{email : peterj@ast.cam.ac.uk},
11: M. M\'endez$^2$,
12: M. van der Klis$^3$\\
13: $^1$Institute of Astronomy, Cambridge University,
14: Madingley Road, CB3 0HA, Cambridge; \\
15: $^2$SRON, National Institute for Space Research;
16: m.mendez@sron.nl\\
17: $^3$Astronomical Institute ``Anton Pannekoek'',
18: University of Amsterdam, and Center for High-Energy Astrophysics,
19: Kruislaan 403, 1098 SJ \\
20: Amsterdam; michiel@astro.uva.nl\\
21: }
22: \maketitle
23:
24: \begin{abstract}
25: \noindent
26: Recent observations of the low--mass X--ray binary 4U~1636--53 with
27: the {\it Rossi X--ray Timing Explorer} show, for the first time, a
28: kiloHertz quasi--periodic oscillation (kHz QPO) peak separation that
29: exceeds the neutron star spin frequency as inferred from burst
30: oscillations. This strongly challenges the sonic--point beat frequency
31: model for the kHz QPOs found in low--mass X--ray binaries. We detect
32: two simultaneous kHz QPOs with a frequency separation of
33: 323.3$\pm$4.3~Hz in an average Fourier power spectrum of observations
34: obtained in September 2001 and January 2002. The lower kHz QPO
35: frequency varied between 644~Hz and 769~Hz. In previous observations
36: of this source the peak separation frequency was $\sim$250~Hz when the
37: lower kHz QPO frequency was $\sim$900~Hz. Burst oscillations occur in
38: 4U~1636--53 at $\sim$581~Hz and possibly at half that frequency
39: (290.5~Hz). This is the first source where the peak separation
40: frequency is observed to change from less than (half) the burst
41: oscillation frequency to more than that. This observation contradicts
42: all previously formulated implementations of the sonic--point beat
43: frequency model except those where the disk in 4U~1636--53 switches
44: from prograde to retrograde.
45:
46: \end{abstract}
47:
48: \begin{keywords}accretion, accretion disks --- stars: individual
49: (4U~1636--53) --- stars: neutron --- X-rays: stars
50:
51: \end{keywords}
52:
53: \section{Introduction}
54: \label{intro}
55: \noindent
56: In recent years kiloHertz quasi--periodic oscillations (kHz QPOs) have
57: been found in power spectra of more than 20 low--mass X--ray binaries
58: (LMXBs) using observations made with the {\it Rossi X--ray Timing
59: Explorer (RXTE)} satellite (\pcite{va2000}). In most sources the kHz
60: QPOs are found in pairs. The high frequencies of these QPOs suggest
61: that at least one of the two peaks reflects the orbital motion of
62: matter close to the neutron star. Another high frequency phenomenon
63: was detected with {\it RXTE}: during some type I X--ray bursts nearly
64: coherent oscillations which slightly drift in frequency, the
65: so--called burst oscillations, have been observed in the power spectra
66: of ten sources (for a review see \pcite{2001AdSpR..28..511S}). The
67: X--ray flux--modulation is consistent with being due to the changing
68: aspect of a drifting hot spot on the surface of the neutron
69: star. Therefore, these burst oscillations are thought to reflect the
70: spin frequency of the neutron star. In high luminosity `Z'--source
71: LMXBs QPOs at low frequencies were found with the {\it EXOSAT} and
72: {\it GINGA} satellites (for a review see
73: \pcite{1989ARA&A..27..517V}). Observations with the {\it RXTE}
74: satellite have also revealed low--frequency QPOs in a number of atoll
75: sources (e.g. \pcite{1998ApJ...499L..41H}; \pcite{fova1998})
76:
77: The atoll source 4U~1636--53 is one of the few sources that displays
78: all of the high--frequency QPO phenomena that have been observed in
79: LMXBs; two kHz QPOs have been observed (\pcite{1996ApJ...469L..17Z};
80: \pcite{wivava1997}) as well as a sideband to the lower of the two kHz
81: QPOs (\pcite{2000ApJ...540L..29J}) and burst oscillations at $\sim$581
82: Hz (\pcite{1997IAUC.6541....1Z}; \pcite{stzhsw1998}). \scite{mi1999}
83: presented evidence that these oscillations are in fact the second
84: harmonic of the neutron star spin of $\sim$290.5 Hz. However, using
85: another dataset these findings were not confirmed
86: (\pcite{2001AdSpR..28..511S}).
87:
88: Since the frequency difference between the two kHz QPO peaks (the peak
89: separation, $\Delta\nu$) is nearly equal to (half) the burst
90: oscillation frequency a beat frequency mechanism was proposed for the
91: kHz QPOs (\pcite{stzhsw1996}). In such a model the higher frequency
92: QPO is attributed to the orbital frequency of clumps of plasma at a
93: special radius near the neutron star while the lower frequency QPO is
94: due to the beat between the orbital frequency and the stellar spin
95: frequency. A specific model incorporating the beat frequency
96: mechanism, the sonic--point beat frequency model, is now one of the
97: leading models explaining kHz QPOs in LMXBs
98: (\pcite{milaps1998}). However, subsequent findings complicated the
99: picture. As was first shown in Sco~X--1, the peak separation decreases
100: as the kHz QPO frequencies increase (\pcite{vawiho1997}). Furthermore,
101: in 4U~1636--53 the peak separation was found to be less than half the
102: burst oscillation frequency (\pcite{mevava1998b}). These findings
103: could be explained within the sonic point beat frequency model
104: (\pcite{2001ApJ...554.1210L}) by taking into account the effect of the
105: inward velocity of the plasma on the emerging kHz QPO frequencies. A
106: firm prediction of this revised sonic--point beat frequency model is
107: that for prograde spinning accretion disks the kHz QPO peak separation
108: is smaller than the neutron star spin frequency. This follows from the
109: detailed description of the plasma flow patterns in this model which
110: produce a number of corrections to the observed frequencies relative
111: to the true orbital and beat frequency, all of which are expected to
112: make $\Delta\nu$ smaller (\pcite{2001ApJ...554.1210L}). Indeed, until
113: now, in various sources the peak separation was always found to be
114: consistent with, or less than (half) the burst oscillation frequency.
115:
116: In this {\it Letter} we show that in recent observations of
117: 4U~1636--53 the kHz QPO peak separation is significantly larger than
118: half the burst oscillation frequency. Using other observations
119: \scite{mevava1998b} had shown earlier that the peak separation was
120: significantly smaller than half the burst oscillation frequency. This
121: means that in order for the sonic--point beat frequency to explain
122: both observations further changes will have to be made in the
123: description of the flow pattern. The only change in the flow pattern
124: previously described that could produce the observed change in the
125: separation frequency is that in which the accretion disk changes from
126: prograde to retrograde when the lower kHz QPO moves from $\sim$750 Hz
127: to $\sim$800 Hz.
128:
129: \section{Observations and analysis}
130: \label{analysis}
131: \noindent
132: 4U~1636--53 was observed in September 2001 and January 2002 with the
133: proportional counter array (PCA; \pcite{jaswgi1996}) onboard the {\it
134: RXTE} satellite (\pcite{brrosw1993}). Due to the reduced duty cycle of
135: the PCA detectors only a subset of the 5 detectors is operational for
136: most of the time. In this analysis we only used data in which 4 of the
137: 5 detectors were operational. A log of the observations is given in
138: Table~\ref{log}. In total $\sim$28.4 ksec of data were used. Data were
139: obtained using an Event--mode in which information on detected X--ray
140: photons is stored onboard and telemetered to the ground on an
141: event--to--event basis. The time resolution of the data is
142: $\sim$125$\mu$s and {\it RXTE}'s effective energy range (2--60 keV) is
143: covered by 64 channels.
144:
145: \begin{table*}
146: \caption{Log of the observations used in this analysis.}
147: \label{log}
148: \begin{center}
149: \begin{tabular}{cccc}
150: \hline
151:
152: Observation ID & Used segments & Amount of good data (ksec) & Count
153: rate$^a$ cnt/s/PCU\\
154: \hline
155: \hline
156: 60032-01-09-00 & 1 & 3.3 & 230 \\
157: 60032-01-09-03 & all & 2.5 & 220 \\
158: 60032-01-19-00 & 1, 2, 3 & 8.6 & 205 \\
159: 60032-01-21-00 & 7, 8 & 6.7 & 180 \\
160: 60032-01-22-00 & 1, 2 & 7.3 & 180 \\
161:
162: \end{tabular}
163: \end{center}
164: {\footnotesize$^a$ Average background subtracted count rate per PCU
165: (2--60 keV)}
166: \end{table*}
167:
168: Using this data we calculated power density spectra of segments of
169: 64~s up to a Nyquist frequency of 4096 Hz in one energy band covering
170: the total PCA energy range. We used a dynamical power spectrum
171: displaying consecutive power spectra rebinned to a frequency
172: resolution of 2 Hz and a time resolution of 128~s to visualize the
173: time evolution of the lower kHz QPO for all observations except for
174: the power spectra of observation 60032-01-21-00 which were rebinned to
175: a time resolution of 162s (segment 7) and 256s (segment 8), and the
176: power spectra of observation 60032-01-22-00, which were treated
177: separately. We traced the frequency of the lower kHz QPO and fitted
178: the power spectrum in a window of 200 Hz centered on the traced
179: frequency with a Lorentzian and a constant. All peaks in each power
180: spectrum of 128~s, 162~s, and 256~s had a significance of
181: $\sim3\sigma$ or more. The frequency of the bin associated with the
182: peak of the fitted Lorentzian was taken to be the frequency of the QPO
183: in that power density spectrum. Finally, we used the shift--and--add
184: method described by \scite{mevava1998b} to shift the lower kHz QPO to
185: the same (arbitrary) frequency in each power spectrum before
186: averaging. This assures that at least for timescales longer than 256~s
187: the determined peak separation is not influenced by the fact that the
188: two kHz QPOs have a different frequency--amplitude relation
189: (\pcite{2001ApJ...561.1016M}).
190:
191: We averaged the aligned power spectra and fitted the average power
192: spectrum from 256--1500 Hz with a constant, to represent the power
193: introduced by the Poisson counting noise, and two Lorentzians to
194: represent the two kHz QPOs. Errors on the fit parameters were
195: determined using $\Delta\chi^2 = 1.0$ ($1\sigma$ single
196: parameter). The observation 60032-01-22-00 was treated separately
197: since we were not able to find a more than 3$\sigma$ lower kHz QPO
198: using data stretches shorter than or equal to 256~s. Instead we
199: averaged the power spectra in this observation without shifting the
200: data. We note that by doing so we may have artificially introduced
201: deviations in the kHz QPO peak separation if the two kHz QPOs moved in
202: frequency during the observation for reasons outlined above.
203:
204: The fits of the average power spectra were good, with a reduced
205: $\chi^2$ slightly larger than 1 for 146 degrees of freedom. We found
206: that, combining all observations, except 60032-01-22-00, the frequency
207: separation between the two kHz QPO peaks was 323.3$\pm$4.3~Hz,
208: significantly larger than 290.5~Hz the putative spin frequency
209: according to an analysis of \scite{mi1999}, but still significantly
210: smaller than 581 Hz, the frequency of the strong burst oscillations
211: found by \scite{1997IAUC.6541....1Z} and \scite{stzhsw1998} (see
212: Figure~\ref{total}). We then divided the data set into three parts
213: according to the frequency of the lower kHz QPO and measured the peak
214: separation in each of the parts. For each part we also averaged the
215: power spectra {\it without} shifting in order to measure the average
216: frequency of the lower kHz QPO and that of a low--frequency QPO
217: changing in frequency from $\sim$24--42 Hz. In Table~\ref{meas} we
218: present our results on the kHz QPOs. The fractional rms amplitude of
219: the low--frequency QPO changed from 7.8$\pm$0.6 per cent to
220: 4.3$\pm$0.7 per cent (2--60 keV) while its frequency increased from
221: 23.6$\pm$2.4 Hz to 42$\pm$4 Hz and its FWHM was consistent with being
222: constant at 26$\pm$9 Hz. In Figure~\ref{delta} we plot the measured
223: peak separation ($\Delta\nu$) as a function of the frequency of the
224: lower and upper kHz QPO ({\it left} and {\it right} panel,
225: respectively). The squares are our new measurements using the
226: shift--and--add method, the diamond is our measurement using
227: observation 60032-01-22-00, and the dots are the measurements of
228: \scite{disalvoinprep} based on {\it RXTE} data obtained from
229: 1996--2000 (see also \pcite{wivava1997};
230: \pcite{mevava1998b}). Clearly, $\Delta\nu$ changes from more than half
231: the burst oscillation frequency to less than that. The average
232: $\Delta\nu$ of the three leftmost points is 327.2$\pm$4.2 Hz and that
233: of the dots is 247.1$\pm$1.9 Hz making the jump in
234: $\Delta\nu\,70.1\pm4.6$ Hz.
235:
236: \begin{table*}
237: \caption{The kHz QPO frequency separation ($\Delta\nu$), the
238: frequency, the fractional rms amplitude (2--60 keV), and the
239: full--width--at--half--maximum (FWHM) of the lower kHz QPO
240: ($\nu_{lower~kHz}$), the fractional rms amplitude (2--60 keV) and the
241: FWHM of the upper kHz QPO ($\nu_{upper~kHz}$).}
242: \label{meas}
243: \begin{center}
244: \begin{tabular}{cccccccc}
245: \hline
246: & Lower~kHz QPO & & & & Upper~kHz QPO & & Peak separation \\
247: \hline
248: \hline
249: rms (\%) & FWHM (Hz) & $\nu_{lower~kHz}$ (Hz)$^a$ & & rms (\%) &
250: FWHM (Hz) & & $\Delta\nu$ (Hz) \\
251: \hline
252:
253: 7.7$\pm$0.4 & 52$\pm$8 & 644.2$\pm$3.2 & & 6.9$\pm$0.5 & 52$\pm$11 & &
254: 326.9$\pm$5.3 \\
255: 7.6$\pm$0.3 & 15.7$\pm$2.3 & 687.8$\pm$2.1 & & 6.8$\pm$0.9 &
256: 74$\pm$35 & & 325.1$\pm$11.5 \\
257: 8.5$\pm$0.1 & 8.1$\pm$0.3 & 723.0$\pm$0.9 & & 5.6$\pm$0.7 & 83$\pm$31
258: & & 329.5$\pm$8.9 \\
259: 8.7$\pm$0.1 & 7.2$\pm$0.7 & 769.2$\pm$1.2 & & 5.1$\pm$0.9 &
260: 139$\pm$62 & & 288.6$\pm$31 \\
261:
262: \end{tabular}
263: \end{center}
264: {\footnotesize $^a$ The frequency of the lower kHz QPO
265: ($\nu_{lower~kHz}$) and all the parameters on the first row were
266: determined without shifting the data.}
267:
268: \end{table*}
269:
270:
271: \begin{figure*}
272: \leavevmode{\psfig{file=total_pds.ps,width=7cm}}\caption{ Average
273: power density spectrum combining all shifted data used in
274: this paper. The peak of the lower kHz QPO is off the vertical
275: scale. The kHz QPO frequency separation is 323.3$\pm$4.3~Hz. The power
276: spectra have been shifted before averaging, therefore, only the
277: frequency difference between the peaks is meaningful.}
278: \label{total}
279: \end{figure*}
280:
281: \begin{figure*}
282: \leavevmode{\psfig{file=lower_upper_delta.ps,angle=-90,width=13cm}}\caption{
283: {\it Left:} The kHz QPO frequency separation as a function of the
284: frequency of the lower kHz QPO. {\it Right:} The kHz QPO frequency
285: separation as a function of the frequency of the upper kHz QPO. The
286: frequency of half the burst oscillation frequency (290.5 Hz) is
287: indicated by the dashed line. The peak separation varies from
288: significantly more than half the burst oscillation frequency [diamond
289: (unshifted data) and squares (shifted data); this work] to
290: significantly less than that (dots; Di Salvo et al. 2002; see also
291: Wijnands et al. 1997 and M\'endez et al. 1998). Error bars on the
292: frequency of the lower kHz QPO are generally smaller than the size of
293: the symbols.}
294: \label{delta}
295: \end{figure*}
296:
297:
298: \section{Discussion}
299: \noindent
300: We found that in 4U~1636--53 the kHz QPO frequency separation
301: ($\Delta\nu$) changes as a function of the frequency of the lower kHz
302: QPO from 323.3$\pm$4.3 Hz to 242.4$\pm$3.6 Hz (Fig.~\ref{delta} {\em
303: left}); the maximum peak separation is significantly larger than 290.5
304: Hz (half the burst oscillation frequency of $\sim$581 Hz; see
305: Section~\ref{intro}), while the minimum peak separation is
306: significantly smaller than that. The change in $\Delta\nu$ as a
307: function of upper kHz QPO frequency is abrupt; the data are
308: consistent with a sudden jump at $\nu=1050$~Hz (Fig.~\ref{delta} {\em
309: right}). We note that similar jumps in $\Delta\nu$ as a function of
310: the upper kHz QPO have also been seen in 4U~1728--34 and 4U~1608--52,
311: although in those cases $\Delta\nu$ always remained smaller or equal
312: to (half) the burst oscillation frequency
313: (\pcite{1999ApJ...517L..51M}; \pcite{mevawi1998};
314: \pcite{2001ApJ...554.1210L}; \pcite{2001ApJ...553L.157M}). This is the
315: first time that the kHz peak separation has been shown to be
316: significantly larger than the inferred neutron star spin frequency and
317: also the first time that $\Delta\nu$ has been seen to vary between
318: less and more than (half) the burst oscillation frequency.
319:
320: In previous work burst oscillations detected in the dipping burster
321: 4U~1916--05 at $\sim$272 Hz also seemed to occur at too low a
322: frequency for $\Delta\nu$ (\pcite{2001ApJ...549L..85G}). However, in
323: that source the kHz QPOs are weak and sometimes very broad, therefore
324: whole segments of several ksec of data had to be averaged to obtain a
325: detection (\pcite{bobaol2000}). This means that there is no certainty
326: that the kHz QPOs were detected simultaneously or that their apparent
327: separation was not affected by frequency shifts combined with changes
328: in their power ratio. Therefore, the measured $\Delta\nu$ may not
329: reflect the true peak separation. In the current work, using the
330: shift--and--add method we excluded the possibility that effects
331: changing $\Delta\nu$ on timescales longer than 256~s have influenced
332: the observed $\Delta\nu$. On timescales shorter than 256~s, changes in
333: the rms amplitudes could bias the peak separation for reasons outlined
334: above. However, since our measurements show that the lower kHz QPO rms
335: amplitude increased while the rms amplitude of the upper kHz QPO
336: decreased this effect will have made the observed $\Delta\nu$ smaller
337: than the true peak separation assuming the rms amplitude--frequency
338: relation is the same for timescales shorter than 256~s. Fast changes
339: in the upper kHz QPO frequency, associated with the 5--6 Hz QPO in
340: Sco~X--1 ($\sim$20 Hz; \pcite{2001ApJ...559L..29Y}) could also bias
341: $\Delta\nu$ measurements. However, those $\sim$5--6 Hz QPOs are only
342: found when atoll and Z sources are in a high--luminosity state
343: (e.g. \pcite{1986ApJ...306..230M};
344: \pcite{1999ApJ...512L..39W}). Changes in the kHz QPO frequencies
345: associated with milliHz QPOs present in 4U~1636--53 and 4U~1608--52 in
346: a small luminosity interval (\pcite{2001A&A...372..138R}) are so small
347: (less than a Hz; \pcite{2002ApJ...567L..67Y}) that it is unlikely that
348: they alter our findings significantly, although we note that such a
349: milliHz QPO was present during some of the observations.
350:
351: Systematic variations in the profile of the kHz QPOs could in
352: principle cause the kHz QPO separation frequency to become large at
353: low kHz QPO frequencies and small at high frequencies since the
354: centroid frequencies have been measured assuming a Lorentzian peak
355: profile. Since our fits were good (reduced $\chi^2\sim$1) and no
356: systematic residuals were apparent, we estimate that shifts in the
357: centroid frequency due to asymmetries in the peak profiles could at
358: most be as large as the FWMH of the peak divided by its significance,
359: i.e. negligible in case of the lower kHz QPO but up to $\sim$10~Hz in
360: case of the upper kHz QPO. For power spectra with a lower kHz QPO in
361: the frequency range 650--750 Hz, the average data$-$model residuals
362: within one FWHM are less then 0.15 per cent of the power in the
363: peak. Furthermore, there is no evidence for a sudden change of the QPO
364: profile which could produce the observed abrupt decrease in
365: $\Delta\nu$. When using Gaussians to model the kHz QPO peaks the fits
366: were also good (reduced $\chi^2\sim$1) and we obtained results
367: consistent with those using Lorentzians. We conclude that it is
368: unlikely that our results can be explained by changing asymmetries in
369: the peak profile of the kHz QPOs.
370:
371: The principal motivation for a beat frequency model is the relative
372: closeness of the inferred spin frequency to $\Delta\nu$. We found that
373: $\Delta\nu$ is consistent with being distributed symmetrically around
374: half the burst oscillation frequency, emphasizing the apparent
375: connection between the inferred spin and $\Delta\nu$. The present
376: formulation of the sonic--point beat--frequency model can explain the
377: observed decrease in $\Delta\nu$ as a function of kHz QPO frequency
378: (\pcite{2001ApJ...554.1210L}). However, all versions of the
379: sonic--point beat--frequency model described so far involving a
380: prograde spinning accretion disk (\pcite{milaps1998};
381: \pcite{2001ApJ...554.1210L}) predict that $\Delta\nu$ is always less
382: than or equal to the spin frequency of the neutron star for Keplerian
383: orbital frequencies larger than the neutron star spin frequencies;
384: since our results for 4U~1636--53 show that $\Delta\nu$ can be larger
385: than the neutron star spin frequency this prediction is now falsified.
386:
387: \scite{2001AdSpR..28..511S} showed that it is still unclear whether
388: the burst oscillations at $\sim$581~Hz reflect the spin frequency of
389: the neutron star or its second harmonic as suggested by
390: \scite{mi1999}. Therefore, one might argue that the spin frequency is
391: really at 581~Hz (not 290.5~Hz) and hence that $\Delta\nu$ is less
392: than the spin frequency. This would, however, take away the principal
393: motivation of beat--frequency models, since the neutron star spin
394: frequency and the kHz QPO peak separation are in that case completely
395: different.
396:
397: That in the sonic--point beat--frequency model the neutron star spin
398: and $\Delta\nu$ are not exactly equal is explained by the way in which
399: the beat--frequency interaction produces the observed frequencies. The
400: key element is the radial motion of the plasma clumps near the sonic
401: radius (\pcite{2001ApJ...554.1210L}). This makes the observed lower
402: kHz QPO frequency somewhat larger than the beat--frequency since it
403: squeezes the spatial separation between peaks of enhanced mass
404: flow. It also makes the upper kHz QPO frequency smaller than the
405: orbital frequency since it makes the clump's footpoint move backward
406: in a corotating frame. It might be possible to find other effects in
407: the plasma flow patterns besides those discussed by
408: \scite{2001ApJ...554.1210L} (all of which make $\Delta\nu$ smaller)
409: which may lead to larger values of $\Delta\nu$ at lower kHz QPO
410: frequencies. Fully relativistic simulations of the gas flow and
411: radiation transport reported by \scite{2001ApJ...554.1210L} have not,
412: so far, shown evidence for this. If, however, the sense of rotation of
413: (part of) the accretion disk would change from prograde to retrograde
414: when $\Delta\nu$ changes from less to more than the spin frequency,
415: then our results could still be in accordance with current
416: formulations of the sonic--point beat--frequency model. This also
417: requires that in our observations for $\Delta\nu$ larger than 290.5~Hz
418: the lower kHz QPO reflects the (near) Keplerian orbital frequency,
419: whereas for those points in Figure~\ref{delta} where $\Delta\nu$ is
420: smaller than half the burst oscillation frequency, the Keplerian
421: orbital motion is represented by the upper kHz QPO.
422:
423: We detected a low--frequency QPO between $\sim$25--40 Hz in
424: 4U~1636--53 simultaneously with the kHz QPO pair. The relation between
425: this low--frequency QPO and the lower kHz QPO is similar to the one
426: found by \scite{fova1998} for 4U~1728--34. It has been suggested that
427: such low--frequency QPOs in atoll sources are related to the Z source
428: Horizontal Branch Oscillations (HBO; e.g. \pcite{1998ApJ...499L..41H};
429: \pcite{psbeva1999}). According to the magnetospheric beat--frequency
430: model these HBO originate near the magnetospheric radius
431: (\pcite{alsh1985}; \pcite{lashal1985}). The frequency of the
432: low--frequency QPO in 4U~1636--53 increased as the frequency of the
433: lower kHz QPO increased. Therefore, if the disk near the sonic radius
434: is indeed retrograde within our observations the retrograde annulus
435: must extend at least up to the magnetospheric radius if the
436: sonic--point and magnetospheric beat--frequency model both
437: apply. Partially retrograde spinning disks have been
438: proposed as an explanation for observed torque reversals
439: (\pcite{1997ApJS..113..367B}) in slowly rotating neutron stars with a
440: dipole magnetic field strength of 10$^{11-12}$ Gauss
441: (\pcite{1998ApJ...499L..27V}). Modelling by
442: \scite{1998ApJ...499L..27V} shows that the accretion disk can get
443: inclinations of more than 90 degrees. It is not clear if their work is
444: also applicable to the case of low--magnetic field LMXBs. Clearly,
445: allowing disks with alternating rotation senses complicates the
446: sonic--point beat frequency interpretation of the kHz QPOs
447: considerably.
448:
449: \section*{Acknowledgments}
450: \noindent
451: PGJ is supported by EC Marie Curie Fellowship
452: HPMF--CT--2001--01308. MK is supported in part by a Netherlands
453: Organization for Scientific Research (NWO) grant. We would like to
454: thank Cole Miller for useful discussions and the referee for comments
455: which helped improve the manuscript.
456: \vskip 0.1cm
457: \begin{thebibliography}{{{M{\'e}ndez}, {van der Klis} \& {van Paradijs}}{1998}}
458:
459: \bibitem[\protect\citefmt{{Alpar} \& {Shaham}}{1985}]{alsh1985}
460: {Alpar}~M.~A., {Shaham}~J., 1985, \nat, 316, 239
461:
462: \bibitem[\protect\citefmt{{Bildsten} {\rm et~al.}}{1997}]{1997ApJS..113..367B}
463: {Bildsten}~L. {\rm et~al.}, 1997, \apjs, 113, 367+
464:
465: \bibitem[\protect\citefmt{{Boirin} {\rm et~al.}}{2000}]{bobaol2000}
466: {Boirin}~L., {Barret}~D., {Olive}~J.~F., {Bloser}~P.~F., {Grindlay}~J.~E.,
467: 2000, \aap, 361, 121
468:
469: \bibitem[\protect\citefmt{{Bradt}, {Rothschild} \& {Swank}}{1993}]{brrosw1993}
470: {Bradt}~H.~V., {Rothschild}~R.~E., {Swank}~J.~H., 1993, \aaps, 97, 355
471:
472: \bibitem[\protect\citefmt{{Di Salvo}, {M{\'e}ndez} \& {van der
473: Klis}}{2002}]{disalvoinprep}
474: {Di Salvo}~T., {M{\'e}ndez}~M., {van der Klis}, 2002, \apj, , in prep
475:
476: \bibitem[\protect\citefmt{{Ford} \& {van der Klis}}{1998}]{fova1998}
477: {Ford}~E.~C., {van der Klis}~M., 1998, \apjl, 506, L39
478:
479: \bibitem[\protect\citefmt{{Galloway} {\rm et~al.}}{2001}]{2001ApJ...549L..85G}
480: {Galloway}~D.~K., {Chakrabarty}~D., {Muno}~M.~P., {Savov}~P., 2001, \apjl, 549,
481: L85
482:
483: \bibitem[\protect\citefmt{{Homan} {\rm et~al.}}{1998}]{1998ApJ...499L..41H}
484: {Homan}~J., {van der Klis}~M., {Wijnands}~R., {Vaughan}~B., {Kuulkers}~E.,
485: 1998, \apjl, 499, L41
486:
487: \bibitem[\protect\citefmt{{Jahoda} {\rm et~al.}}{1996}]{jaswgi1996}
488: {Jahoda}~K., {Swank}~J.~H., {Giles}~A.~B., {Stark}~M.~J., {Strohmayer}~T.,
489: {Zhang}~W., {Morgan}~E.~H., 1996, \procspie, 2808, 59
490:
491: \bibitem[\protect\citefmt{{Jonker}, {M{\' e}ndez} \& {van der
492: Klis}}{2000}]{2000ApJ...540L..29J}
493: {Jonker}~P.~G., {M{\' e}ndez}~M., {van der Klis}~M., 2000, \apjl, 540, L29
494:
495: \bibitem[\protect\citefmt{{Lamb} \& {Miller}}{2001}]{2001ApJ...554.1210L}
496: {Lamb}~F.~K., {Miller}~M.~C., 2001, \apj, 554, 1210
497:
498: \bibitem[\protect\citefmt{{Lamb} {\rm et~al.}}{1985}]{lashal1985}
499: {Lamb}~F.~K., {Shibazaki}~N., {Alpar}~M.~A., {Shaham}~J., 1985, \nat, 317, 681
500:
501: \bibitem[\protect\citefmt{{M{\' e}ndez}, {van der Klis} \&
502: {Ford}}{2001}]{2001ApJ...561.1016M}
503: {M{\' e}ndez}~M., {van der Klis}~M., {Ford}~E.~C., 2001, \apj, 561, 1016
504:
505: \bibitem[\protect\citefmt{{M{\'e}ndez} \& {van der
506: Klis}}{1999}]{1999ApJ...517L..51M}
507: {M{\'e}ndez}~M., {van der Klis}~M., 1999, \apjl, 517, L51
508:
509: \bibitem[\protect\citefmt{{M\'endez} {\rm et~al.}}{1998}]{mevawi1998}
510: {M\'endez}~M., {van der Klis}~M., {Wijnands}~R., {Ford}~E.~C., {van
511: Paradijis}~J., {Vaughan}~B.~A., 1998, \apjl, 505, L23
512:
513: \bibitem[\protect\citefmt{{M{\'e}ndez}, {van der Klis} \& {van
514: Paradijs}}{1998}]{mevava1998b}
515: {M{\'e}ndez}~M., {van der Klis}~M., {van Paradijs}~J., 1998, \apjl, 506, L117
516:
517: \bibitem[\protect\citefmt{{Middleditch} \&
518: {Priedhorsky}}{1986}]{1986ApJ...306..230M}
519: {Middleditch}~J., {Priedhorsky}~W.~C., 1986, \apj, 306, 230
520:
521: \bibitem[\protect\citefmt{{Miller}, {Lamb} \& {Psaltis}}{1998}]{milaps1998}
522: {Miller}~M.~C., {Lamb}~F.~K., {Psaltis}~D., 1998, \apj, 508, 791
523:
524: \bibitem[\protect\citefmt{{Miller}}{1999}]{mi1999}
525: {Miller}~M.~C., 1999, \apjl, 515, L77
526:
527: \bibitem[\protect\citefmt{{Muno} {\rm et~al.}}{2001}]{2001ApJ...553L.157M}
528: {Muno}~M.~P., {Chakrabarty}~D., {Galloway}~D.~K., {Savov}~P., 2001, \apjl, 553,
529: L157
530:
531: \bibitem[\protect\citefmt{{Psaltis}, {Belloni} \& {van der
532: Klis}}{1999}]{psbeva1999}
533: {Psaltis}~D., {Belloni}~T., {van der Klis}~M., 1999, \apj, 520, 262
534:
535: \bibitem[\protect\citefmt{{Revnivtsev} {\rm
536: et~al.}}{2001}]{2001A&A...372..138R}
537: {Revnivtsev}~M., {Churazov}~E., {Gilfanov}~M., {Sunyaev}~R., 2001, \aap, 372,
538: 138
539:
540: \bibitem[\protect\citefmt{{Strohmayer} {\rm et~al.}}{1996}]{stzhsw1996}
541: {Strohmayer}~T.~E., {Zhang}~W., {Swank}~J.~H., {Smale}~A., {Titarchuk}~L.,
542: {Day}~C., {Lee}~U., 1996, \apjl, 469, L9
543:
544: \bibitem[\protect\citefmt{{Strohmayer} {\rm et~al.}}{1998}]{stzhsw1998}
545: {Strohmayer}~T.~E., {Zhang}~W., {Swank}~J.~H., {White}~N.~E., {Lapidus}~I.,
546: 1998, \apjl, 498, L135
547:
548: \bibitem[\protect\citefmt{{Strohmayer}}{2001}]{2001AdSpR..28..511S}
549: {Strohmayer}~T.~E., 2001, Advances in Space Research, 28, 511
550:
551: \bibitem[\protect\citefmt{{van der Klis} {\rm et~al.}}{1997}]{vawiho1997}
552: {van der Klis}~M., {Wijnands}~R. A.~D., {Horne}~K., {Chen}~W., 1997, \apjl,
553: 481, L97
554:
555: \bibitem[\protect\citefmt{{van der Klis}}{1989}]{1989ARA&A..27..517V}
556: {van der Klis}~M., 1989, \araa, 27, 517
557:
558: \bibitem[\protect\citefmt{{van der Klis}}{2000}]{va2000}
559: {van der Klis}~M., 2000, \araa, 38, 717
560:
561: \bibitem[\protect\citefmt{{van Kerkwijk} {\rm
562: et~al.}}{1998}]{1998ApJ...499L..27V}
563: {van Kerkwijk}~M.~H., {Chakrabarty}~D., {Pringle}~J.~E., {Wijers}~R.~A.~M.~J.,
564: 1998, \apjl, 499, L27
565:
566: \bibitem[\protect\citefmt{{Wijnands} {\rm et~al.}}{1997}]{wivava1997}
567: {Wijnands}~R. A.~D., {van der Klis}~M., {van Paradijs}~J., {Lewin}~W. H.~G.,
568: {Lamb}~F.~K., {Vaughan}~B., {Kuulkers}~E., 1997, \apjl, 479, L141
569:
570: \bibitem[\protect\citefmt{{Wijnands}, {van der Klis} \&
571: {Rijkhorst}}{1999}]{1999ApJ...512L..39W}
572: {Wijnands}~R., {van der Klis}~M., {Rijkhorst}~E., 1999, \apjl, 512, L39
573:
574: \bibitem[\protect\citefmt{{Yu} \& {van der Klis}}{2002}]{2002ApJ...567L..67Y}
575: {Yu}~W., {van der Klis}~M., 2002, \apjl, 567, L67
576:
577: \bibitem[\protect\citefmt{{Yu}, {van der Klis} \&
578: {Jonker}}{2001}]{2001ApJ...559L..29Y}
579: {Yu}~W., {van der Klis}~M., {Jonker}~P.~G., 2001, \apjl, 559, L29
580:
581: \bibitem[\protect\citefmt{{Zhang} {\rm et~al.}}{1996}]{1996ApJ...469L..17Z}
582: {Zhang}~W., {Lapidus}~I., {White}~N.~E., {Titarchuk}~L., 1996, \apjl, 469, L17
583:
584: \bibitem[\protect\citefmt{{Zhang} {\rm et~al.}}{1997}]{1997IAUC.6541....1Z}
585: {Zhang}~W., {Lapidus}~I., {Swank}~J.~H., {White}~N.~E., {Titarchuk}~L., 1997,
586: \iaucirc, 6541, 1+
587:
588: \end{thebibliography}
589:
590: \end{document}
591: