1: \documentclass[12pt,preprint]{aastex}
2: %\documentclass[useAMS,usenatbib]{emulateapj}
3: \usepackage{latexsym,graphicx,natbib}
4:
5: %\usepackage{rotating}
6: %\usepackage{lscape}
7: % Style
8: \def\simless{{\th \rlap{\raise 0.5ex\hbox{$\scriptstyle {<}$}}
9: {\lower 0.3ex\hbox{$\scriptstyle {\sim}$}} \th }} %< or of order
10: \def\simgreat{{\th \rlap{\raise 0.5ex\hbox{$\scriptstyle {>}$}}
11: {\lower 0.3ex\hbox{$\scriptstyle {\sim}$}} \th }} %> or of order
12: \def\greateq{{\th \rlap{\raise 0.5ex\hbox{$\scriptstyle {>}$}}
13: {\lower 0.3ex\hbox{$\scriptstyle {-}$}} \th }} %< or of order
14: \def\lesseq{{\th \rlap{\raise 0.5ex\hbox{$\scriptstyle {<}$}}
15: {\lower 0.3ex\hbox{$\scriptstyle {-}$}} \th }} %> or of order
16: \def\th{\thinspace}
17: \def\ts{{\raise 0.3ex\hbox{$\scriptstyle {\th \sim \th }$}}}
18:
19:
20: \usepackage{epsfig}
21: \newcommand{\etal}{\mbox{et al.}}
22: \newcommand{\saxjfull}{\mbox{SAX J1748.9$-$2021}}
23: \newcommand{\saxj}{\mbox{SAX J1748}}
24: \newcommand{\RXTE}{\textit{RXTE}}
25: \newcommand{\us}{$\mu$s}
26: \newcommand{\uHz}{$\mu$Hz}
27: \newcommand{\fluxunits}{~erg~cm$^{-2}$~s$^{-1}$}
28: \newcommand{\ra}[4]{$#1^{\rm h} #2^{\rm m} #3\fs#4$}
29: \newcommand{\dec}[4]{$#1^\circ #2^\prime #3\farcs#4$}
30: \newcommand{\tempo}{\textsc{tempo}}
31: \newcommand{\Tempo}{\textsc{Tempo}}
32:
33:
34: \begin{document}
35:
36: \shortauthors{Patruno \etal}
37: \shorttitle{Timing of \saxjfull\ }
38:
39: %\submitted{Submitted to ApJ Letters}
40: \title{Phase-coherent timing of the accreting millisecond pulsar SAX J1748.9-2021}
41: \author{Alessandro Patruno \altaffilmark{1}, Diego Altamirano \altaffilmark{1}, Jason W.T. Hessels \altaffilmark{1},
42: Piergiorgio Casella \altaffilmark{1}, Rudy Wijnands \altaffilmark{1}, Michiel van der Klis \altaffilmark{1} }
43: \altaffiltext{1}{Astronomical Institute ``Anton Pannekoek,'' University of
44: Amsterdam, Kruislaan 403, 1098 SJ Amsterdam, The Netherlands, E-mail: apatruno@science.uva.nl}
45:
46: \email{apatruno@science.uva.nl}
47:
48:
49: \begin{abstract}
50:
51: We present a phase-coherent timing analysis of the intermittent
52: accreting millisecond pulsar \saxjfull. A new timing solution for the
53: pulsar spin period and the Keplerian binary orbital parameters was
54: achieved by phase connecting all episodes of intermittent pulsations
55: visible during the 2001 outburst. We investigate the pulse profile
56: shapes, their energy dependence and the possible influence of Type I
57: X-ray bursts on the time of arrival and fractional amplitude of the
58: pulsations. We find that the timing solution of \saxjfull\ shows an
59: erratic behavior when selecting different subsets of data, that is
60: related to substantial timing noise in the timing post-fit
61: residuals. The pulse profiles are very sinusoidal and their fractional
62: amplitude increases linearly with energy and no second harmonic is
63: detected. The reason why this pulsar is intermittent is still unknown
64: but we can rule out a one-to-one correspondence between Type I X-ray
65: bursts and the appearance of the pulsations.
66:
67: \end{abstract}
68: \keywords{stars: individual (\saxjfull\ ) --- stars: neutron --- X-rays: stars}
69:
70: \section{Introduction}
71:
72: Recently it has been shown (\citealt{Kaaret}, \citealt{Galloway07},
73: \citealt{Casella}, \citealt{Gavriil}, \citealt{Altamirano}) that
74: intermittent pulsations can be observed during the transient outbursts
75: of some accreting millisecond pulsars (AMPs). It was emphasized
76: (\citealt{Casella}, \citealt{Gavriil}, \citealt{Altamirano}) that a
77: new class of pulsators is emerging with a variety of phenomenology
78: i.e., pulsations lasting only for the first two months of the outburst
79: (HETE J1900.1-2455, \citealt{Kaaret}, \citealt{Galloway07}), switching
80: on and off throughout the observations (SAX J1748.9-2021,
81: \citealt{Altamirano}, and HETE J1900.1-2455 \citealt{Galloway07}), or
82: even appearing for only ~150 s over 1.3 Ms of observations (Aql X-1,
83: \citealt{Casella}). This is different from the behavior of other
84: known AMPs, where the pulsations persist throughout the outburst until
85: they drop below the sensitivity level of the instrumentation.
86:
87: %with the approaching of quiescence.
88:
89: \citet{Galloway07} and \citet{Altamirano} pointed out the clear,
90: albeit non-trivial, connection between the appearance of pulsations in
91: HETE J1900.1-2455 and SAX J1748.9-2021 and the occurrence of
92: thermonuclear bursts (Type I X-ray bursts). \citet{Galloway07}
93: demonstrated that increases of pulse amplitude in HETE J1900.0$-$2455
94: sometimes but not always follow the occurrence of Type I
95: bursts nor do all bursts trigger an increase in pulse
96: amplitude. \citet{Altamirano} showed that in \saxjfull\ the pulsations
97: can re-appear after an absence of pulsations of a few orbital periods
98: or even a few hundred seconds. All the pulsing episodes occur close in
99: time to the detection of a Type I X-ray burst, but again not all the
100: Type I bursts are followed by pulsations so it is unclear if the
101: pulsations are indeed triggered by Type I X-ray bursts.\\ The pulse
102: energy dependence and the spectral state in which pulsations are
103: detected also differ between these objects. As shown by \citet{Cui},
104: \citet{Galloway05}, \citet{Gierlinski} and \citet{Galloway07}, the
105: fractional amplitude of the pulsations in AMPs monotonically decreases
106: with increasing energies between 2 and $\approx 20$keV. However in IGR
107: J00291+5934, \citet{Falanga} found a slight decrease in the pulse
108: amplitude between 5 and 8 keV followed by an increase up to energies
109: of 100 keV. For the single pulse episode of Aql X-1 the pulsation
110: amplitude increased with energy between 2 and $\approx 20$keV
111: (\citealt{Casella}). Interestingly in Aql X-1 the pulsations appeared
112: during the soft state, contrary to all the other known AMPs which
113: pulsate solely in their hard state (although many of these objects
114: have not been observed in a soft state). Therefore it is interesting
115: to investigate these issues for \saxjfull\ as well.\\
116: In order to study these various aspects of \saxjfull\ an
117: improved timing solution is necessary. The spin frequency of the
118: neutron star and the orbital parameters of \saxjfull\ were measured by
119: \citet{Altamirano} using a frequency-domain technique that measures
120: the Doppler shift of the pulsar spin period due to the orbital motion
121: of the neutron star around the center of mass of the binary. The
122: source was found to be spinning at a frequency of 442.361 Hz
123: in a binary with an orbit of ~8.7 hrs and a donor companion mass of
124: $\sim 0.1-1M_{\odot}$. However, the timing solution obtained with this
125: technique was not sufficiently precise to phase connect between epochs
126: of visible pulsations. The intermittency of the pulsations creates
127: many gaps between measurable pulse arrival times, complicating the use
128: of standard techniques to phase connect the times of arrival
129: (TOAs). The current work presents a follow-up study to the discovery
130: paper of intermittent pulsations in \saxjfull\ (\citealt{Altamirano})
131: and provides for the first time a timing solution obtained by phase
132: connecting the pulsations observed during the 2001 outburst. In \S 2
133: we present the observations and in \S 3 we explain the technique
134: employed to phase connect the pulsations. In \S 4 we show our results
135: and in \S 5 we discuss them. In \S 6 we outline our conclusions.
136:
137:
138: \section{X-ray observations and data reduction}
139:
140: We reduced all the pointed observations from the {\it{RXTE}}
141: Proportional Counter Array (PCA, \citealt{Jahoda}) that cover the
142: 1998, 2001 and 2005 outbursts of \saxjfull. The PCA instrument
143: provides an array of five proportional counter units (PCUs) with a
144: collecting area of 1200 $\rm\,cm^{2}$ per unit operating in the 2-60
145: keV range and a field of view of $\approx 1^{\circ}$. The number of
146: active PCUs during the observations varied between two and
147: five. %note: this value in 2001 and 2005 outbursts is 2-4 PCUs.
148: We used all the data available in the Event mode with a time resolution
149: of $122\rm\,\mu s$ and 64 binned energy channels and in the Good Xenon
150: mode with a time resolution of $1\rm\,\mu s$ and 256 unbinned
151: channels; the latter were re-binned in time to a resolution of $
152: 122\rm\,\mu s$. All the observations are listed in Table 1.
153:
154: %########### TABLE 1###########################
155: \begin{table}
156: \caption{Observations analyzed for each outburst\label{tbl:obslist}}
157:
158: %\centering
159: \scriptsize
160: \begin{tabular}{lrrrl}
161: \hline
162: \hline
163: Outburst & Start & End & Time & Observation IDs \\
164: (year) & (MJD) & (MJD) & (ks) &\\
165: 1998 & 51051.3 & 51051.6 & 14.8 & {\tt 30425-01-01-00}\\
166: 2001 & 52138.8 & 52198.4 & 138.0 & {\tt 60035-*-*-*}~$-$~{\tt 60084-*-*-*}\\
167: 2005 & 53523.0 & 53581.4 & 82.2 & {\tt 91050-*-*-*}\\
168: \hline
169: \end{tabular}
170:
171: \label{table1:data}
172: \end{table}
173: %########### END TABLE #######################
174:
175:
176:
177:
178: For obtaining the pulse timing solution we selected only photons with
179: energies between $\approx5$ and 24 keV to avoid the strong background at
180: higher energies and to avoid the region below $\approx 5$keV where the
181: pulsed fraction is below $\approx 1$\% rms. The use of a wider energy band
182: decreases the signal to noise ratio of the pulsations.
183: We used only the good time
184: intervals excluding Earth occultations, passages of the satellite
185: through the South Atlantic Anomaly and intervals of unstable
186: pointing. We barycentered our data with the tool {\it{faxbary}} using
187: the JPL DE-405 ephemeris along with a spacecraft ephemeris including
188: fine clock corrections that together provide an absolute timing
189: accuracy of $\approx 3.4$ $\rm\,\mu s$ (\citealt{Rots}). The best available
190: source position comes from {\it{Chandra}} observations (\citealt{Zand}
191: and \citealt{Pooley}, see Table 2). The background was subtracted
192: using the FTOOL {\it{pcabackest}}.
193: %\footnote{http://heasarc.gsfc.nasa.gov/docs/xte/recipes/pcabackest.html}
194:
195: \section{The timing technique}
196:
197: The intermittent nature of \saxjfull\ requires special care when
198: trying to obtain a full phase connected solution (i.e., a solution
199: that accounts for all the spin cycles of the pulsar between periods of
200: visible pulsations). With pulsations detected in only $\approx 12\%$
201: of the on-source data of the 2001 outburst, and only a few hundred
202: seconds of observed pulsations during the 2005 outburst, the solution
203: of \citet{Altamirano} does not have the precision required to directly
204: phase connect the pulses. Their spin frequency uncertainty of
205: $\sigma_{s}=10^{-3}$Hz implies a phase uncertainty of half a spin
206: cycle after only $500\rm\,s$, while the gaps between pulse episodes in
207: the 2001 outburst were as long as $2$ $\rm\,days$. Moreover, the low
208: pulsed fraction ($\simless 2.5\%$ rms in the 5-24 keV band) limits the
209: number of TOAs with high signal to noise ratio.\\ Since the solution
210: of Altamirano et al. (2008) lacks the precision required to phase
211: connect the TOAs, we had to find an improved initial set of ephemeris
212: to be use as a starting point for the phase-connection. To obtain a
213: better initial timing model we selected all the chunks of data where
214: the pulsations could clearly be detected in the average power density
215: spectrum (see \citealt{Altamirano} for a description of the
216: technique). The total amount of time intervals with visible pulsations
217: were M=11 with a time length $500\rm\,s\simless\rm
218: t_{obs,i}\rm\simless 3000\rm\,s$ in the i-th chunk, with i=1,...,M.
219: We call the Modified Julian Date (MJD) where the pulsations appear in
220: the i-th chunk as $\rm\,MJD_{start,i}$ while the final MJD where the
221: pulsation disappears is
222: $\rm\,MJD_{end,i}=\rm\,MJD_{start,i}+t_{obs,i}$. We then measured the
223: spin period $P_{s,i}$, the spin period derivative $\dot{P}_{s,i}$ and,
224: when required, the spin period second derivative $\ddot{P}_{s,i}$ to
225: align the phases of the pulsations (folded in a profile of 20 bins)
226: for each chunk of data, i.e., to keep any residual phase drift to less
227: than one phase bin (0.05 cycles) over the full length $t_{obs,i}$.
228: These three spin parameters are a mere \emph{local} measure of the
229: spin variation as a consequence of, primarily, the orbital Doppler
230: shifts and cannot be used to predict pulse phases outside the i-th
231: chunk of data; they provide a 'local phase-connection' of the pulses
232: within the i-th chunk. Using this set of $3\times\rm\,M$ parameters we
233: generated a series of spin periods $P_{s}\left(t\right)$ in each i-th
234: chunk of data with $t\in\left[\rm\,MJD_{start,i},
235: \rm\,MJD_{end,i}\right]$. The predicted spin period of the pulsar
236: $P_{s}$ at time $t$ is given by the equation:
237: \begin{equation}
238: P_{s}\left(t\right)=P_{s,i}+\dot{P}_{s,i}t+\frac{1}{2}\ddot{P}_{s,i}t^{2}
239: \end{equation}
240: with $i=1...M$. We then fitted a sinusoid representing a circular Keplerian
241: orbit to all the predicted spin periods $P_{s}\left(t\right)$
242: and obtained an improved set of orbital and spin parameters.
243:
244: With this new solution we folded 300$\rm\,s$ intervals of data to
245: create a new series of pulse profiles with higher signal to noise
246: ratio. We then fitted a sinusoid plus a constant to these folded
247: profiles and selected only those profiles with a S/N (defined as the
248: ratio between the pulse amplitude and its 1$\sigma$ statistical error)
249: larger than 3.4. With a value of $\rm\,S/N\greateq 3.4$ we expect less
250: than one false detection when considering the total number of folded
251: profiles. The TOAs were then obtained by cross-correlating these
252: significant folded profiles with a pure sinusoid. The determination
253: of pulse TOAs and their statistical uncertainties closely resembles
254: the standard radio pulsar technique (see for example
255: \citealt{Taylor}). \\ If the pulsar is isolated and the measured TOAs
256: are error-free, the pulse phase $\phi$ at time \emph{t} can be
257: expressed as a polynomial expansion:
258: \begin{equation}\label{eq:phi}
259: \phi_{P}\left(t\right)=\phi_{0} + \nu_{0}(t-t_{0}) + \frac{1}{2}\dot{\nu}_{0}(t-t_{0})^{2} + ...
260: \end{equation}
261:
262:
263: where the subscript ``0'' refers to the epoch time $t_{0}$. In an
264: AMP, the quantities $\nu_{0}$ and $\dot{\nu}_{0}$ are the spin
265: frequency of the pulsar and the spin frequency derivative (related to
266: the spin torque), respectively. We note that the detection of a spin
267: up/down in AMPs is still an open issue, since many physical processes
268: can mimic a spin up/down (see for example \citealt{Hartman}, but see
269: also \citealt{Burderi}, \citealt{Papitto}, \citealt{Riggio} for a
270: different point of view). Indeed in the real situation the pulse
271: phase of an AMP can be expressed as:
272: \begin{equation}\label{eq:phitot}
273: \phi\left(t\right) = \phi_{P}\left(t\right) + \phi_{O}\left(t\right) +
274: \phi_{M}\left(t\right) + \phi_{N}\left(t\right)
275: \end{equation}
276: with the subscripts P, O, M, and N referring respectively to the
277: polynomial expansion in eq.~\ref{eq:phi} (truncated at the second
278: order), the orbital and the measurement-error components and any
279: intrinsic timing noise of unknown origin that might be included in the
280: data. For example the timing noise is observed in young isolated radio
281: pulsars (e.g., \citealt{Groth}, \citealt{Cordes}) or has been detected
282: as red noise in SAX J1808.4-3658 (\citealt{Hartman}) or in high mass
283: X-ray binaries (\citealt{Bildsten97}). The term
284: $\phi_{O}\left(t\right)$ takes into account the phase variation
285: introduced by the orbital motion of the pulsar around the companion.
286: We expect that the error-measurement component
287: $\phi_{M}\left(t\right)$ is given by a set of independent values and
288: is normally distributed with an amplitude that can be predicted from
289: counting statistics (e.g., \citealt{Taylor}). Fitting the TOAs with a
290: constant frequency model and a zero-eccentricity orbit we phase
291: connected the 2001 outburst pulsations in a few iterations, re-folding
292: the profiles with the improved solution and fitting the TOAs using the
293: software package TEMPO2 \citep{Hobbs}.
294: %\footnote{http://www.atnf.csiro.au/research/pulsar/tempo2/}.
295: No significant improvement is obtained fitting for a spin derivative
296: and an eccentricity (using the ELL1 model in TEMPO2). We did not fit
297: for the orbital period in TEMPO2, because our initial fit, described
298: earlier in this section, included data from the 2005 outburst and thus
299: provided more precision on this parameter. A consistency check was
300: made folding the pulsations of the 2005 outburst with the two
301: different solutions. With the phase-connected solution of the 2001
302: outburst we were not able to properly fold the pulsations in the 2005
303: outburst, due to the relatively large error in the fitted orbital
304: period, whereas with the hybrid solution (pulse spin frequency $\nu$,
305: epoch of ascending node passage $\rm\,T_{asc}$, and projected
306: semi-major axis $a_{x}\rm\,sin$$\,i$ from the TEMPO2 fit to 2001 only;
307: $\rm\, P_{orb}$ kept from the sinusoidal fit to
308: $P_{s}\left(t\right)$ for all the data) we were able to fold the pulsations
309: and obtained a $\approx 8\rm\,\sigma$ pulse detection with approximately
310: 500$\rm\, s$ of the 2005 outburst data.
311:
312:
313:
314: \section{Results}
315:
316: \subsection{Timing solution}
317:
318: With the new timing solution we were able to search with higher
319: sensitivity for additional pulsation episodes throughout the 1998,
320: 2001 and 2005 outbursts. We folded chunks of data with a length
321: between approximately $300$s (to minimize the pulse smearing due to
322: short timescale timing noise) and $1$ hr (to increase the S/N when the
323: pulsations are weak). Two new pulsation episodes of length $\approx
324: 300$s each, were detected on MJD~52180.7 and MJD~52190.5, about eleven
325: days and one day before the first earlier detection respectively,
326: with a significance of $\approx 3.5\sigma$. We refitted a constant
327: frequency model and a zero eccentricity orbit to the new ensemble of
328: TOAs and obtained new timing residuals and a new timing solution. The
329: post-fit residuals are all less than $\approx 0.2$ spin cycles, but their
330: $\rm\, rms$ amounts to $164\rm\,\mu$s,
331: well in excess of the value expected from counting statistics of
332: $\simeq 70\mu$s; the reduced $\chi^{2}$
333: is $9$ (for 86 degrees of freedom, see the timing residuals in Figure~\ref{res}). Trying to fit higher frequency derivatives or an eccentricity does not
334: significantly improve the fit.
335: \begin{figure}
336: \begin{center}
337: \rotatebox{-90}{\includegraphics[width=0.7\columnwidth]{f1.eps}}
338: \end{center}
339: \caption{Post-fit timing residuals of the 5-24 keV TOAs of the 2001
340: outburst. A negative/positive value of the residuals means that a
341: pulsation is leading/ lagging with respect to the timing model. All
342: the pulse profiles used have $\rm\,S/N > 3.4$. Strong timing noise
343: appears as a large scatter within each group of points and
344: misalignment between groups.
345: \label{res}}
346: \end{figure}
347: At first glance, the residual variance,
348: described by the timing noise term $\phi_{N}$ in eq~\ref{eq:phitot},
349: has two components: a long-term and a short-term component to the
350: timing noise $\phi_{N}$. The short term timing noise acts on a
351: timescale of a few hundred seconds and creates the large scatter
352: observed within each group of points in Fig.~\ref{res}. The long term
353: timing noise acts on a timescale of a few days and can be recognized
354: by a misalignment of the TOA residuals between groups, where with a
355: simple white noise component we would expect within each group a
356: distribution of TOAs spread around zero. The short term and the long term
357: timing noise, which could of course well be part of the same process,
358: are together responsible for the bad $\chi^{2}$ of the fit and the
359: large $\rm\,rms$ of the residuals. \\ A histogram of the timing
360: residuals shows a non-Gaussian distribution with an extended lower
361: tail. A superposition of two Gaussians
362: distribution can fit this. The first Gaussian is composed of TOAs that
363: are on average earlier than predicted by the timing solution. The
364: second Gaussian comprises TOAs that are on average later than
365: predicted. The separation between the two mean values of the Gaussians
366: is a few hundred microseconds. We found that the
367: lagging TOAs correspond to systematically higher S/N pulsations than
368: the leading ones. Since pulsations with higher S/N usually have high
369: fractional amplitude, we analyzed the fractional amplitude dependence
370: of the residuals. As can be seen in Fig.~\ref{banana} above a
371: fractional amplitude of $\approx 1.8\%$ rms (the ``upper group''), the TOAs
372: are on average 103$\rm\,\mu$s later than predicted by the timing
373: solution; the remaining TOAs (the ``lower group'') are
374: 42$\rm\,\mu$s early. The lower group TOAs are not
375: related in a one-to-one relation with the fractional amplitude
376: (Fig.~\ref{banana}), meaning that probably some kind of noise process
377: is affecting the lower group TOAs in addition to the effect the
378: varying amplitude has on the TOAs.
379:
380: \begin{figure}[t]
381: \begin{center}
382: \rotatebox{-90}{\includegraphics[width=0.7\columnwidth]{f2.eps}}
383: \end{center}
384: \caption{Dependence of the residuals on fractional amplitude of the
385: pulsations. The vertical dashed line divides the plot in the upper (right side of the
386: panel) and
387: lower group (left side) of TOAs at a fractional amplitude of 1.8$\%$ rms.
388: \label{banana}}
389: \end{figure}
390:
391:
392: \begin{figure*}[t]
393: \begin{center}
394: \rotatebox{-90}{\includegraphics[width=0.3\textwidth]{f3.eps}}
395: \end{center}
396: \caption{Timing residuals of the upper (right panel) and lower (left
397: panel) group. All the TOAs have at least a S/N larger than 3.4 and
398: refer to the 5-24 keV energy band. The short timescale timing noise
399: affects more strongly the lower group than the upper one, which is
400: reflected in the bad $\chi^{2}$ of the fit (4.4 and 9.6 for the upper
401: and lower group respectively). The new TOA around MJD 52181 belong to
402: the lower group. The two plots have the same scale on the
403: residuals-axis to put in evidence the large scatter of the post-fit
404: residuals in the lower group with respect to the upper one.
405: \label{res_two}}
406: \end{figure*}
407: We fitted separate constant frequency and zero eccentricity
408: models to just the upper and lower groups of TOAs and again checked the
409: post-fit residuals (see Fig.~\ref{res_two}).
410: The reduced $\chi^{2}$ values of the two solutions are still large
411: ($\approx4.4$ and $\approx9.6$ for the upper and lower solution,
412: respectively). The maximum residual amplitude is about 200$\rm\,\mu$s
413: and 450$\rm\,\mu$s in the upper and lower solution, respectively.
414: Long term timing noise is still apparent, contributing more to the
415: lower TOAs group than to the upper one. While the fitted orbital
416: parameters ($\rm\,T_{asc}$ and $a_{x}\rm\,sin$$\,i$) are in agreement
417: to within 1$\sigma$, the two solutions deviate in pulse frequency by
418: $1.2\times10^{-7}\rm\,Hz$, about $4\sigma$.
419:
420: We ascribe this
421: difference to the unmodeled timing noise component $\phi_{N}$, which is
422: ignored in the estimate of the parameter errors. Selecting groups of
423: TOAs with fractional amplitude thresholds different from $1.8\%$ rms
424: always gives different timing solutions, with deviations larger than
425: $3\sigma$ sometimes also in $\rm\,T_{asc}$ and
426: $a_{x}\rm\,sin$$\,i$.
427:
428: To investigate if this indicates a systematic
429: connection between timing noise excursions and fractional amplitude or
430: is just a consequence of selecting subsets of data out of a record
431: affected by systematic noise, we selected subsets of data using
432: different criteria: we split the data in chunks with TOAs earlier and
433: later than a fixed MJD or we selected only TOAs with S/N larger than
434: 3.4. In each case the fitted solutions deviated by several standard
435: deviations, indicating that the differences are due to random
436: selection of data segments out of a record affected by correlated
437: noise (the timing noise). To take this effect into account in
438: reporting the timing solution we increased our statistical errors by a
439: constant factor such that the reduced $\chi^{2}$ of the timing
440: solution fit was close to unity and refitted the parameters.
441: The parameters and the errors for the global solution, calculated in
442: this way, are reported in Table 2. They provide an improvement of
443: three and five orders of magnitude with respect to the Keplerian and
444: spin frequency parameter uncertainties, respectively, when compared
445: with the solution reported in \citet{Altamirano}.
446:
447: %########### TABLE 2 ###########################
448: \begin{table}
449: \caption{Timing parameters for \saxjfull\ }
450:
451: %\centering
452: \scriptsize
453: \begin{tabular}{lc}
454: \hline
455: \hline
456: Parameter & Value \\
457: \hline
458: Right Ascension ($\alpha$) (J2000)$^{a}$ & $17^{h}48^{m}52^{s}.163$ \\
459: Declination ($\delta$) (J2000)$^{a}$ & $-20^{\circ}21^{'}32^{''}.40 $\\
460: Orbital period, P$_{orb}$(hr) \dotfill & 8.76525(3) hr \\
461: %
462: Projected semi major axis, $a_\mathrm{x} \sin i$ (light-ms)\dotfill & 387.60(4) \\
463: %
464: Epoch of ascending node passage $T_\mathrm{asc}$ (MJD, TDB) \dotfill & 52191.507190(4) \\
465: %
466: Eccentricity, $e$ (95\% confidence upper limit) \dotfill & $<2.3\times 10^{-4}$ \\
467: %
468: Spin frequency $\nu_0$ (Hz) \dotfill & 442.36108118(5) \\
469: %
470: Reference Epoch (MJD) & 52190.0\\
471: %
472: \hline
473: %
474: All the uncertainties quoted correspond to 1$\sigma$ confidence level (i.e., $\Delta\chi^{2}=1$).\\
475: $^{a}$X-ray position from Chandra \citep{Zand,Pooley}. The
476: pointing uncertainty is 0''.6.
477: \end{tabular}
478:
479: \label{table2:data}
480: \end{table}
481: %########### END TABLE #######################
482:
483:
484: To check whether the timing noise is energy dependent, we selected three
485: energy bands: a soft band (5-13 keV), an intermediate band (11-17 keV)
486: and a hard one (13-24 keV). Then we repeated the procedure, fitting
487: all the significant TOAs with a constant spin frequency and circular
488: Keplerian orbit, without increasing the statistical errors on the
489: TOAs.
490:
491: The two solutions of the soft and intermediate bands were in
492: agreement to within 1$\sigma$ for $\nu$, $\rm\,T_{asc}$ and
493: $a_{x}\rm\,sin$$\,i$, while the hard band solution deviated by more
494: than 3$\sigma$ for the spin frequency. The reduced $\chi^{2}$ of the
495: sinusoidal fit to the pulse profiles in the highest energy band was
496: systematically larger than in the other two bands. Fitting a second
497: and a third harmonic did not significantly improve the fit according
498: to an F-test.
499:
500: The bad $\chi^{2}$ can be ascribed to a non-Poissonian
501: noise process that dominates at high energies or to an intrinsic
502: non-sinusoidal shape of the pulse profiles. Since the hard energy band
503: has a count rate approximately 10 times smaller than the soft band,
504: the influence of this on the global timing solution calculated for the
505: energy band 5-24 keV is negligible. We refolded all the profiles and
506: calculated another solution for the 5-17 keV energy band to avoid the
507: region of high energies and found an agreement to within 1$\sigma$
508: with the solution reported in Table 2 and with the solution obtained
509: for the 5-13 keV and for the 11-17 keV energy bands. This also shows
510: that the timing noise is substantially independent from the energy
511: band selected for energies below 17 keV.
512:
513: We checked for the possible existence of characteristic frequencies in
514: a power density spectrum (PDS) of the timing residuals using a
515: Lomb-Scargle technique. No peaks above three sigma were found in
516: either the PDS of the upper and lower group TOAs or the PDS of the
517: global solution combining these two groups. The post-fit timing
518: residuals are also not correlated with the reduced $\chi^{2}$ of each
519: fitted pulse profile. We also checked the influence of very large
520: events (VLE) due to energetic particles in the detector on the TOA
521: residuals but no correlation was found. There is also no link between
522: the source flux and the magnitude of the timing residuals and between
523: the timing residuals and the orbital phase.
524:
525: \subsection{Pulse shape variations}\label{sec:shape}
526:
527: To look for possible pulse profile changes, we fitted all significant
528: folded pulse profiles with a fundamental and a second harmonic
529: sinusoid (plus a constant level). No significant second harmonic
530: detections ($>3\rm\sigma$) were made in the 5-24 keV energy band with
531: upper limits of 0.9\%, 0.5\% and 0.4\% rms amplitude at 98$\%$
532: confidence for the 1998, 2001 and 2005 pulsations episodes
533: respectively. We repeated the procedure for different energy bands,
534: obtaining similar results.\\ The pulse profiles of \saxjfull\ are
535: therefore extremely sinusoidal with the fundamental amplitude at least
536: $\approx 11$ times that of the second harmonic %(see Fig.~\ref{profiles})
537: both in the soft and hard energy bands. Thus there are no detectable pulse
538: profile shape variations nor the possibility to detect sudden changes
539: of the relative phase between the fundamental and the second harmonic
540: such as observed in SAX J1808.4-3658 (\citealt{Burderi},
541: \citealt{Hartman}) and XTE J1814-338 \citep{Papitto}.
542:
543:
544: \subsection{Pulse energy dependence and time lags}
545:
546: We analyzed the energy dependence of the pulse profiles selecting ten
547: energy bands and measuring the strength of the pulsations. We folded
548: all the observations where we detect significant pulsations. The pulse
549: fraction is as small as $\approx 0.5\%$ rms between 2.5 and 4 keV and
550: increases up to $\approx4.0\%$ rms above 17 keV following a linear trend.
551: Counting statistics prevents the measurement of pulsations above 24
552: keV. Fitting a linear relation to the points in
553: Fig.~\ref{energy} gives a slope of $(0.17\pm 0.01)\rm\,
554: \%\,rms\,keV^{-1}$.
555:
556: The pulses of \saxjfull\ exhibit a similar energy dependence to
557: Aql X-1 \citep{Casella}, which is opposite to that measured in three
558: other AMPs (SAX J1808.4-3658 \citealt{Cui}, XTE J1751-305
559: \citealt{Gierlinski}, XTE J0929-314 \citealt{Galloway07}), where the
560: amplitude is stronger at low energies and decreases at high
561: energies. In another source however (IGR J00291+5934) \citet{Falanga}
562: pointed out an increase of the fractional amplitude with energy which
563: seems to resemble what has been observed for \saxjfull\ and Aql
564: X-1. In IGR J00291+5934 a slight decrease in the fractional amplitude
565: between 5 and 8 keV is followed by a slight increase up to 100
566: keV. However between 6 and 24 keV the slight increase is also
567: consistent with a constant when considering the 1$\sigma$ error bars.
568: In \saxjfull\ and Aql X-1 the increase in fractional amplitude with
569: energy is much steeper.
570:
571: We did not detect any significant time lag between soft and hard
572: photons, with an upper limit of $250\mu$s at the $3\sigma$ level,
573: using a coherent analysis between the 2.5 and 24 keV bands
574: (see fig.~\ref{energy}).
575: If a time lag with a magnitude smaller than our limit exists, this
576: might explain the dependence of TOA on pulse amplitude if for example
577: low amplitude pulses are dominated by soft photons whereas the high
578: amplitude pulses are dominated by hard photons or vice versa. To check
579: this possibility we re-folded all the low and high amplitude pulses
580: selecting a low energy (5-13 keV) and an high energy band (13-24
581: keV). The folded profiles do not show any strong energy dependence,
582: with the low and high energy bands equally contributing to the low and
583: high amplitude profiles. Therefore energy dependent lags are unlikely
584: to be the origin of the TOA dependence on amplitude.
585:
586: \begin{figure}[t]
587: \begin{center}
588: \rotatebox{-90}{\includegraphics[width=0.3\columnwidth]{f4.eps}}
589: \end{center}
590: \caption{Time lags (left panel) and energy dependence of the
591: fractional amplitudes of the pulsations (right panel). The plots stop
592: at 24 keV after which the counts strongly drop below the detection
593: level. The plots refer to all the observations where we detected
594: pulsations. \textbf{Left}: no significant time lags are measured,
595: with all the points being consistent with a zero lag at the three
596: sigma level with respect to the first energy band, chosen here to be
597: our reference time. \textbf{Right}: the fractional amplitude starts
598: with a very low value of $\approx 0.5\%$ rms around 2 keV and linearly
599: increases up to 4$\%$ rms at around 15 keV.
600: \label{energy}}
601: \end{figure}
602:
603: \subsection{Pulse dependence on flux}
604: In both the 2001 and 2005 outbursts the pulsations disappear at both
605: low and high X-ray flux. In Fig.~\ref{lc_pulses} we show the light
606: curve of the 2001 outburst, with pulsating episodes indicated with
607: filled circles. The pulses appear in a broad flux band but not below
608: $\approx 150$ and not above $\approx 240\rm\,Ct/s/PCU$. It is also
609: intriguing that pulsations sporadically disappear on a timescale of a
610: few hundred seconds even when the flux is in the band where we see
611: pulsations. The few non-pulsating episodes at high and low flux could
612: be random, however some non-pulsing episodes are observed at the same
613: fluxes where we observe pulsations at different times. This indicates
614: the absence of a flux threshold above or below which the pulse
615: formation mechanism is at work.
616: \begin{figure}[t!]
617: \begin{center}
618: \rotatebox{-90}{\includegraphics[width=0.7\columnwidth]{f5.eps}}
619: \end{center}
620: \caption{Light curve of the 2001 outburst from the first pulsating
621: episode until the end of the outburst. The time resolution of the
622: light curve is 300 s, and each point corresponds to a folded light
623: curve. When a pulsation is detected a filled circle is plotted. The
624: dotted lines are limits of highest and lowest flux where the pulses
625: appear. Note that even between the dashed lines, there are
626: non-pulsating episodes. These are often followed by pulsations on a
627: timescale of a few hundred seconds.
628: \label{lc_pulses}}
629: \end{figure}
630:
631:
632: \subsection{Type I X-ray bursts}
633:
634: The possibility that the pulsations are triggered by Type I X-ray
635: bursts can be tested by measuring the evolution of the fractional
636: amplitude of the pulsations in time. Our time resolution corresponds
637: to the time interval selected to fold the data, i.e., $\approx300$s. As
638: can be seen in Fig.~\ref{resamp}, there are cases in which the
639: fractional amplitude increases after a Type I burst (e.g., MJD
640: 52191.74, 52192.25, 52192.31, 52196.24, 52198.2, 52198.36, 53534.46)
641: on timescales of a few hundred seconds, while after other Type I
642: bursts the fractional amplitude does not change significantly (MJD
643: 52190.38, 52190.47, 52193.32, 52193.40, 52193.45, 52195.3, 52195.5) or
644: increased prior to the Type I burst (MJD 52195.62). In four cases
645: pulsations are present without any occurrence of a Type I burst (MJD
646: 52181,52190.30, 52191.65, 52198.27) and in one case the fractional
647: amplitude decreases after the burst (MJD 52192.40). Clearly there is
648: no single response of the pulsation characteristics to the occurrence
649: of a burst. There is also no clear link between the strength of a Type
650: I burst and the increase of the pulsation amplitudes as is evident
651: from Fig.~\ref{resamp}. The maximum fractional amplitude ($3.5\%$ rms)
652: is reached during the 2005 outburst, after a Type I X-ray
653: burst. However, after $\approx 500$s the pulsations were not seen
654: anymore.
655:
656:
657:
658: \subsection{Color--color diagram}\label{sec:ccd}
659:
660: We use the 16 s time resolution Standard 2 mode data to calculate
661: X-ray colors. Hard and soft colors are defined respectively as the
662: 9.7--16.0 keV / 6.0--9.7 keV and the 3.5--6.0 keV / 2.0--3.5 keV count
663: rate ratio. The energy-channel conversion is done using the
664: pca\_e2c\_e05v02 table provided by the RXTE Team.
665: Type I X-ray bursts were removed, background
666: subtracted, and dead-time corrections made. In order to correct for
667: the gain changes as well as the differences in effective area between
668: the PCUs themselves, we normalized our colors by the corresponding
669: Crab Nebula color values \cite[see][ see table 2 in
670: \citealt{Altamirano} for average colors of the Crab Nebula per
671: PCU]{Kuulkers,Straaten} that are closest in time but in the same
672: RXTE gain epoch, i.e., with the same high voltage setting of the PCUs
673: \citet{Jahoda}.
674:
675: The PCA observations sample the source behavior during three different
676: outbursts \citep[see Fig 1 in][]{Altamirano}. In Figure~\ref{ccd} we
677: show the color--color diagram for all the observations of the three
678: outbursts.
679: Grey circles mark the 16 second averaged colors while black crosses
680: and triangles mark the average color of each of the 8 observations
681: from which pulsations were detected. As it can be seen, pulsations
682: only appear in soft state of the source (banana state). During
683: the observations of the 1998 outburst, the source was always in the so called
684: Island/Extreme island state while during the 2001 and 2005 outburst the
685: source was observed in both island and banana state.
686:
687:
688: \begin{figure*}
689: \begin{center}
690: \rotatebox{0}{\includegraphics[width=1.0\textwidth]{f6.eps}}
691: \end{center}
692: \caption{Time dependence of the fractional amplitude and the TOA
693: residuals in the 5-24 keV band during the 2001 outburst. Each diagram
694: displays the residuals (upper panel) and their fractional amplitude (lower panel)
695: versus time. All the points correspond to pulses with a
696: $\rm\,S/N\greateq 3.4$. Type I bursts are also plotted in both
697: panels in arbitrary units but scaled by a common constant factor.
698: \label{resamp}}
699: \end{figure*}
700: \begin{figure}
701: \begin{center}
702: \rotatebox{-90}{\includegraphics[width=0.7\columnwidth]{f7.eps}}
703: \end{center}
704: \caption{Color-color diagram for the 1998, 2001 and 2005 outbursts
705: (values are normalized by the Crab Nebula, see
706: Section~\ref{sec:ccd}). Gray circles mark the 16 seconds average
707: color of all available data. Black crosses (2001 outburst) and
708: triangles (2005 outburst) mark the average color per observation in
709: which we find pulsations.
710: \label{ccd}}
711: \end{figure}
712:
713:
714: \subsection{Radio pulse search}
715: We also used the new phase-coherent timing solution presented here to search
716: for possible radio pulsations from SAX J1748.9$-$2021. For this, we used
717: 2-GHz archival radio data from Green Bank Telescope (GBT) observations of
718: the 6 known radio millisecond pulsars (MSPs) in the globular cluster (GC)
719: NGC~6440 (see \citealt{Freire} for a description of these data). The
720: known radio pulsars in NGC~6440 have spin periods from 3.8$-$288.6\,ms and
721: dispersion measures (DMs) between 219$-$227\,pc cm$^{-3}$. With a spin
722: period of 2.3\,ms, SAX~J1748.9$-$2021 is thus the fastest-spinning pulsar
723: known in NGC~6440, where the average spin period of the radio pulsars
724: (11.3\,ms when the 288-ms pulsar B1745$-$20 is excluded) is relatively long
725: compared to other GCs.
726:
727: We searched 11 data sets taken between 2006 December and 2007 March. These
728: data sets have total integration times between $0.5-5$\,hr and a combined
729: total time of 20.3\,hr (73\,ks). The data were dedispersed into 10 time
730: series with trial DMs between 219$-$228\,pc cm$^{-3}$. To check if
731: pulsations were present, these time series were then folded with the timing
732: solution presented in this paper. Because potential radio pulsations may
733: also be transient, we folded not only the full data sets, but also
734: overlapping chunks of 1/4, 1/10, and 1/50 of the individual data set (thus
735: probing timescales between 100\,s and hours). Furthermore, because these data
736: were taken 6 years after the data that was used to construct the timing
737: solution, we folded the data using both the exact period prediction from the
738: timing solution as well as allowing for a small search around this value.
739:
740: No obvious pulsations were detected in this analysis, where the
741: reduced $\chi^2$ of the integrated pulse profile was used to judge if
742: a given fold was worthy of further investigation. This is perhaps not
743: surprising as radio pulsations have, as of yet, never been detected
744: from an LMXB or AMP. Nonetheless, we plan to continue these searches
745: on the large amount of available radio data which we have not yet
746: searched with this technique.
747:
748:
749:
750: \section{Discussion}
751:
752: We have presented a new timing solution for \saxjfull\, obtained by
753: phase connecting the 2001 outburst pulsations. We discovered the
754: presence of timing noise on short (hundred seconds) and long (few
755: days) timescales. We cannot exclude the presence of timing noise on
756: different timescales, since the short and the long timescales found
757: are also the two timescales the TOAs probe. The pulse profiles of
758: \saxjfull\ keep their sinusoidal shape below 17 keV throughout the
759: outburst, but do experience considerable shifts relative to the
760: co-rotating reference frame both apparently randomly as a function of
761: time and systematically with amplitude. Above 17 keV the pulse
762: profiles show deviations from a sinusoidal shape that cannot be
763: modeled adding a 2nd harmonic. The fit needs a very high number of
764: harmonics to satisfactory account the shape of the pulsation.
765: This is probably due to the effect of some underlying unknown
766: non-Poissonian noise process that produces several sharp spikes in the
767: pulse profiles. The lack of a detectable second harmonic
768: prevents us from studying shape variations of the pulse profiles such
769: as was done for SAX J1808.4--3658 \citep{Burderi,Hartman} where sudden
770: changes between the phase of the fundamental and the second harmonic
771: were clearly linked with the outburst phase.\\ Another interesting
772: aspect is the energy dependence of the pulse profiles, which can be a
773: test of current pulse formation theories. In one model of
774: AMPs, \citet{Poutanen} explain the pulsations by a modulation of
775: Comptonized radiation whose seeds photons come from blackbody
776: radiation. The thermal emission is given by the hot spot and/or
777: emitting column produced by the in-falling material that follows the
778: magnetic dipole field lines of the neutron star plasma rotating with
779: the surface of the neutron star. Part of the blackbody photons can be
780: scattered to higher energies by a slab of shocked plasma that forms a
781: comptonizing region above the hot spot \citep{Basko}.
782:
783: In three AMPs
784: \citep{Cui,Gierlinski,Poutanen,Watts06,Galloway07} the fractional
785: pulse amplitudes decreases toward high energies with the soft photons
786: always lagging the hard ones. However in \saxjfull\ the energy
787: dependence is opposite, with the fractional pulse amplitudes
788: increasing toward higher energies and no detectable lags (although the
789: large upper limit of $250\rm\,\mu$s does not rule out the presence of
790: time lags with magnitude similar to those detected in other
791: AMPs). Moreover, all the pulsating episodes happen when the source is
792: in the soft state (although not all soft states show pulsations). This
793: is similar to Aql X-1 (see \citealt{Casella}) where the only pulsating
794: episode occurred during a soft state, and the pulsed fraction also
795: increased with energy. Remarkably all the other sources which show
796: persistent pulsations have hard colors typical of the extremely island
797: state.
798:
799: Both the energy dependence and the presence of the pulsations
800: during the soft state strongly suggest a pulse formation pattern for
801: these two intermittent sources which is different from that of the
802: other known AMPs and the intermittent pulsar HETE J1900.1-2455. As
803: discussed in \citealt{MunoA}, \citealt{MunoB}, a hot spot region
804: emitting as a blackbody with a temperature constrast with respect to
805: the neutron star surface produces pulsations with an increasing
806: fractional amplitude with energy in the observer rest frame. The exact
807: variation of the fractional amplitude with energy however has a
808: complex dependence on several free parameters as the mass and radius
809: of the neutron star and the number,size, position and temperature of
810: the hot spot and viewing angle of the observer. The observed slope of
811: 0.2$\rm\,\%\,rms\,keV^{-1}$ is consistent with this scenario, i.e., a
812: pure blackbody emission from a hot spot with a temperature constrast
813: and a weak comptonization (see \citealt{FalangaTit}, \citealt{MunoA},
814: \citealt{MunoB}). However it is not possible to exclude a
815: strong comptonization given the unknown initial slope of the
816: fractional amplitude.
817:
818: The pulse shapes above 17 keV are
819: non-sinusoidal and are apparently affected by some non-Poissonian
820: noise process or can be partially produced by an emission mechanism
821: different from the one responsible of the formation of the soft pulses
822: (see for example \citet{Poutanen} for an explaination of how the soft
823: pulses can form). We found that the TOAs of the pulsations are
824: independent of the energy band below 17 keV but selection on the
825: fractional amplitude of the pulsations strongly affects the TOAs: high
826: amplitude pulses arrive later. However, this does not affect the
827: timing solution beyond what is expected from fitting other data
828: selections: apparently the TOAs are affected by correlated timing
829: noise. If we take into account the timing noise ($\phi_{N}$) in
830: estimating the parameter errors, then the timing solutions are
831: consistent to within 2$\sigma$. In the following we examine several
832: possibilities for the physical process producing the $\phi_{N}$ term.
833:
834: %\subsection{Random walk}
835: %Groth (1975) studied how a random walk in the phase, frequency or
836: %spin torque of a pulsar (PN,FN and SN processes respectively)
837: %will produce timing noise in the residuals of radio pulsars. The
838: %origin of the stochastic process behind the random walk is poorly
839: %understood. In X-ray binaries it might be related with random
840: %torque components during the accretion process (\citealt{LambS,LambP,BD85}
841: %but see also \citealt{Baykal,Cordes} for a
842: %discussion of the timing noise in radio pulsars). As demonstrated by
843: %\citet{Groth}, a random walk in the n--th time derivative of the phase
844: %is mathematically equivalent to a white-noise process in the (n + 1)--th
845: %time derivative of the phase. This means that if a PN, FN or SN
846: %processes are at work in \saxjfull\, we should see a Gaussian
847: %distribution of the residuals with zero-mean after fitting a
848: %polynomial frequency, a frequency derivative and a frequency second
849: %derivative to the TOAs. However this is not seen in the residuals of
850: %\saxjfull. We found that this ``whitening'' of the timing residuals
851: %cannot be achieved even after including up to the 9th time derivative
852: %of the spin frequency. We attribute this to the presence of
853: %timing noise that cannot be easily modeled with a polynomial model
854: %such as in eq.~\ref{eq:phi}.
855:
856: \subsection{Influence of Type I X-ray bursts on the TOAs}
857:
858: The occurrence of Type I X-ray bursts in coincidence with the
859: appearance of the pulsations suggests a possible intriguing relation
860: between the two phenomena. However as we have seen in \S 4.4, there is
861: not a strict link between the appearance of pulsations and Type I
862: burst episodes. Only after $\approx 30\%$ of the Type I
863: bursts the pulsations appeared or increased their fractional
864: amplitude. The appearance of pulsations seems more related with a
865: period of global surface activity during which both the pulsations and
866: the Type I bursts occur. This is different from what has been observed
867: in HETE J1900.1-2455 \citep{Galloway07} where the Type I X-ray bursts
868: were followed by an increase of the pulse amplitudes that were
869: exponentially decaying with time. We note that during the pulse
870: episodes at MJD 52191.7, 52193.40, 52193.45 and 52195.5 the TOA
871: residuals just before and after a Type I burst are shifted by
872: 300-500$\rm\,\mu$s suggesting a possible good candidate for the large
873: timing noise observed. However in two other episodes no substantial shift is
874: observed in the timing residuals (see Fig.~\ref{resamp}). We also note that
875: there is no relation between the Type I burst peak flux
876: and the magnitude of the shifts in the residuals.
877:
878:
879:
880: \subsection{Other possibilities}
881:
882:
883: The timing noise and its larger amplitude in the weak pulses might be
884: related with some noise process that becomes effective only when the
885: pulsations have a low fractional amplitude. \citet{Romanova} have
886: shown that above certain critical mass transfer rates an unstable
887: regime of accretion can set in, giving rise to low-$m$ modes in the
888: accreting plasma which produce an irregular light curve. The
889: appearance of these modes inhibits magnetic channeling and hence
890: dilutes the coherent variability of the pulsations. Such a model can
891: explain pulse intermittency by positing that we see the pulses only
892: when the accretion flow is stable. The predominance of timing noise in
893: the low fractional amplitude group could be explained if with the
894: onset of unstable accretion the $m$-modes gradually set in and affect
895: the pulsations, lowering their fractional amplitude until they are
896: undetectable. However the lack of a correlation between pulse
897: fractional amplitude and X-ray flux is not predicted by the model.
898:
899: Another interesting aspect are the apparent large jumps that we observe
900: in the phase residuals on a timescale of a few hundred seconds.
901: Similar jumps have been observed for example in SAX J1808.4-3658
902: during the 2002 and 2005 outbursts (\citealt{Burderi,Hartman}. In
903: that case however the phases jump by approximately 0.2 spin cycles on
904: a timescale of a few days, while here we observe phase jumps of about
905: 0.4 cycles on a timescale of hundred seconds. Other AMPs where phase
906: jumps were observed (XTE J1814-338 \citealt{Papitto}, XTE J1807-294
907: \citealt{Riggio}) have shown timescales of a few days similar to SAX
908: J1808.4-3658. This is a further clue that the kind of noise we are
909: observing in \saxjfull\ is somewhat different from what has been
910: previously observed.
911:
912:
913: \section{Conclusions}
914:
915: We have shown that it is possible to phase connect the intermittent
916: pulsations seen in \saxjfull\ and we have found a coherent timing
917: solution for the spin period of the neutron star and for the Keplerian
918: orbital parameters of the binary.\\ We found strong correlated timing
919: noise in the post-fit residuals and we discovered that this noise is
920: strongest in low fractional amplitude pulses and is not related with
921: the orbital phase. Higher-amplitude ($\simgreat 1.8\%$ rms) pulsations
922: arrive systematically later than lower-amplitude ($\simless 1.8\%$ rms)
923: ones, by on average 145$\mu$s. The pulsations of \saxjfull\ are
924: sinusoidal in the 5-17 keV band, with a fractional amplitude
925: linearly increasing in the energy range considered. The pulsations
926: appear when the source is in the soft state, similarly to what has
927: been previously found in the intermittent pulsar Aql X-1. The origin
928: of the intermittency is still unknown, but we can rule out a
929: one-to-one correspondence between Type I X-ray bursts and the
930: appearance of pulsations.
931:
932: \acknowledgements{Acknowledgements: We would like to thank P. Soleri,
933: B. Stappers and J. Verbiest for useful discussions. J.W.T.H. thanks
934: NSERC and the Canadian Space Agency for a postdoctoral fellowship and
935: supplement respectively.The National Radio Astronomy Observatory is a
936: facility of the National Science Foundation operated under
937: cooperative agreement by Associated Universities, Inc.}
938:
939:
940: \begin{thebibliography}{32}
941: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
942: \expandafter\ifx\csname url\endcsname\relax
943: \def\url#1{{\tt #1}}\fi
944: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
945:
946: \bibitem[{{Altamirano} et~al.(2007){Altamirano}, {Casella}, {Patruno},
947: {Wijnands}, \& {van der Klis}}]{Altamirano}
948: {Altamirano} D., {Casella} P., et~al., Aug. 2007, ArXiv e-prints, 708
949:
950: \bibitem[{{Basko} \& {Sunyaev}(1976)}]{Basko}
951: {Basko} M.M., {Sunyaev} R.A., May 1976, \mnras, 175, 395
952:
953: \bibitem[{{Bildsten} et~al.(1997){Bildsten}, {Chakrabarty}, {Chiu}
954: et~al.}]{Bildsten97}
955: {Bildsten} L., {Chakrabarty} D., et~al., Dec. 1997, \apjs, 113, 367
956:
957: \bibitem[{{Burderi} et~al.(2006){Burderi}, {Di Salvo}, {Menna}, {Riggio}, \&
958: {Papitto}}]{Burderi}
959: {Burderi} L., {Di Salvo} T., et~al., Dec. 2006, \apjl, 653, L133
960:
961: \bibitem[{{Casella} et~al.(2007){Casella}, {Altamirano}, {Wijnands}, \& {van
962: der Klis}}]{Casella}
963: {Casella} P., {Altamirano} D., et~al., Aug. 2007, ArXiv e-prints, 708
964:
965: \bibitem[{{Cordes} \& {Helfand}(1980)}]{Cordes}
966: {Cordes} J.M., {Helfand} D.J., Jul. 1980, \apj, 239, 640
967:
968: \bibitem[{{Cui} et~al.(1998){Cui}, {Morgan}, \& {Titarchuk}}]{Cui}
969: {Cui} W., {Morgan} E.H., {Titarchuk} L.G., Sep. 1998, \apjl, 504, L27+
970:
971: \bibitem[{{Falanga} \& {Titarchuk}(2007)}]{FalangaTit}
972: {Falanga} M., {Titarchuk} L., Jun. 2007, \apj, 661, 1084
973:
974: \bibitem[{{Falanga} et~al.(2005){Falanga}, {Kuiper}, {Poutanen}
975: et~al.}]{Falanga}
976: {Falanga} M., {Kuiper} L., et~al., Dec. 2005, \aap, 444, 15
977:
978: \bibitem[{{Freire} et~al.(2007){Freire}, {Ransom}, {Begin} et~al.}]{Freire}
979: {Freire} P.C.C., {Ransom} S.M., et~al., Nov. 2007, ArXiv e-prints, 711
980:
981: \bibitem[{{Galloway} et~al.(2005){Galloway}, {Markwardt}, {Morgan},
982: {Chakrabarty}, \& {Strohmayer}}]{Galloway05}
983: {Galloway} D.K., {Markwardt} C.B., et~al., Mar. 2005, \apjl, 622, L45
984:
985: \bibitem[{{Galloway} et~al.(2007){Galloway}, {Morgan}, {Krauss}, {Kaaret}, \&
986: {Chakrabarty}}]{Galloway07}
987: {Galloway} D.K., {Morgan} E.H., et~al., Jan. 2007, \apjl, 654, L73
988:
989: \bibitem[{{Gavriil} et~al.(2007){Gavriil}, {Strohmayer}, {Swank}, \&
990: {Markwardt}}]{Gavriil}
991: {Gavriil} F.P., {Strohmayer} T.E., et~al., Nov. 2007, \apjl, 669, L29
992:
993: \bibitem[{{Gierli{\'n}ski} \& {Poutanen}(2005)}]{Gierlinski}
994: {Gierli{\'n}ski} M., {Poutanen} J., Jun. 2005, \mnras, 359, 1261
995:
996: \bibitem[{{Groth}(1975)}]{Groth}
997: {Groth} E.J., Nov. 1975, \apjs, 29, 453
998:
999: \bibitem[{{Hartman} et~al.(2007){Hartman}, {Patruno}, {Chakrabarty}
1000: et~al.}]{Hartman}
1001: {Hartman} J.M., {Patruno} A., et~al., Aug. 2007, ArXiv e-prints, 708
1002:
1003: \bibitem[{{Hobbs} et~al.(2006){Hobbs}, {Edwards}, \& {Manchester}}]{Hobbs}
1004: {Hobbs} G.B., {Edwards} R.T., {Manchester} R.N., Jun. 2006, \mnras, 369, 655
1005:
1006: \bibitem[{{in't Zand} et~al.(2001){in't Zand}, {van Kerkwijk}, {Pooley}
1007: et~al.}]{Zand}
1008: {in't Zand} J.J.M., {van Kerkwijk} M.H., et~al., Dec. 2001, \apjl, 563, L41
1009:
1010: \bibitem[{{Jahoda} et~al.(2006){Jahoda}, {Markwardt}, {Radeva} et~al.}]{Jahoda}
1011: {Jahoda} K., {Markwardt} C.B., et~al., Apr. 2006, \apjs, 163, 401
1012:
1013: \bibitem[{{Kaaret} et~al.(2006){Kaaret}, {Morgan}, {Vanderspek}, \&
1014: {Tomsick}}]{Kaaret}
1015: {Kaaret} P., {Morgan} E.H., et~al., Feb. 2006, \apj, 638, 963
1016:
1017: \bibitem[{{Kuulkers} et~al.(1994){Kuulkers}, {van der Klis}, {Oosterbroek}
1018: et~al.}]{Kuulkers}
1019: {Kuulkers} E., {van der Klis} M., et~al., Sep. 1994, \aap, 289, 795
1020:
1021: \bibitem[{{Muno} et~al.(2002){Muno}, {{\"O}zel}, \& {Chakrabarty}}]{MunoA}
1022: {Muno} M.P., {{\"O}zel} F., {Chakrabarty} D., Dec. 2002, \apj, 581, 550
1023:
1024: \bibitem[{{Muno} et~al.(2003){Muno}, {{\"O}zel}, \& {Chakrabarty}}]{MunoB}
1025: {Muno} M.P., {{\"O}zel} F., {Chakrabarty} D., Oct. 2003, \apj, 595, 1066
1026:
1027: \bibitem[{{Papitto} et~al.(2007){Papitto}, {di Salvo}, {Burderi}
1028: et~al.}]{Papitto}
1029: {Papitto} A., {di Salvo} T., et~al., Mar. 2007, \mnras, 375, 971
1030:
1031: \bibitem[{{Pooley} et~al.(2002){Pooley}, {Lewin}, {Verbunt} et~al.}]{Pooley}
1032: {Pooley} D., {Lewin} W.H.G., et~al., Jul. 2002, \apj, 573, 184
1033:
1034: \bibitem[{{Poutanen} \& {Gierli{\'n}ski}(2003)}]{Poutanen}
1035: {Poutanen} J., {Gierli{\'n}ski} M., Aug. 2003, \mnras, 343, 1301
1036:
1037: \bibitem[{{Riggio} et~al.(2007){Riggio}, {Di Salvo}, {Burderi} et~al.}]{Riggio}
1038: {Riggio} A., {Di Salvo} T., et~al., Oct. 2007, ArXiv e-prints, 710
1039:
1040: \bibitem[{{Romanova} et~al.(2007){Romanova}, {Kulkarni}, \&
1041: {Lovelace}}]{Romanova}
1042: {Romanova} M.M., {Kulkarni} A.K., {Lovelace} R.V.E., Nov. 2007, ArXiv e-prints,
1043: 711
1044:
1045: \bibitem[{{Rots} et~al.(2004){Rots}, {Jahoda}, \& {Lyne}}]{Rots}
1046: {Rots} A.H., {Jahoda} K., {Lyne} A.G., Apr. 2004, \apjl, 605, L129
1047:
1048: \bibitem[{{Taylor}(1992)}]{Taylor}
1049: {Taylor} J.H., 1992, Philosophical Transactions of the Royal Society of London,
1050: 341, 117-134 (1992), 341, 117
1051:
1052: \bibitem[{{van Straaten} et~al.(2003){van Straaten}, {van der Klis}, \&
1053: {M{\'e}ndez}}]{Straaten}
1054: {van Straaten} S., {van der Klis} M., {M{\'e}ndez} M., Oct. 2003, \apj, 596,
1055: 1155
1056:
1057: \bibitem[{{Watts} \& {Strohmayer}(2006)}]{Watts06}
1058: {Watts} A.L., {Strohmayer} T.E., Dec. 2006, \mnras, 373, 769
1059:
1060: \end{thebibliography}
1061:
1062:
1063:
1064: \end{document}
1065: