1: \documentclass[12pt,preprint]{aastex}
2: \begin{document}
3: \title{Discovery of the Neutron Star Spin Frequency in EXO 0748--676}
4: \author{Adam R. Villarreal\altaffilmark{1}} \affil{Department of
5: Physics, University of Arizona, Tucson, AZ 85721;
6: adamv@physics.arizona.edu} \and \altaffiltext{1}{NASA Harriett G.
7: Jenkins Fellow} \author{Tod E. Strohmayer} \affil{Laboratory for
8: High Energy Astrophysics, NASA Goddard Space Flight Center,
9: Greenbelt, MD 20771; stroh@milkyway.gsfc.nasa.gov}
10: \begin{abstract}
11:
12: We report the results of a search for burst oscillations during
13: thermonuclear X-ray bursts from the low mass X-ray binary (LMXB)
14: EXO 0748--676. With the proportional counter array (PCA) onboard
15: the {\it Rossi X-ray Timing Explorer} (RXTE) we detected a 45 Hz
16: oscillation in the average power spectrum of 38 thermonuclear
17: X-ray bursts from this source. We computed power spectra with 1 Hz
18: frequency resolution for both the rising and decaying portions of
19: 38 X-ray bursts from the public RXTE archive. We averaged the 1 Hz
20: power spectra and detected a significant signal at 45 Hz in the
21: decaying phases of the bursts. The signal is detected at a
22: significance level of $4 \times 10^{-8}$. No similar signal was
23: detected in the rising intervals. A fit to the oscillation peak
24: at 0.25 Hz resolution gives a frequency of $\nu_0 = 44.7 \pm 0.06$
25: Hz and an oscillation quality factor of $Q = \nu_0 /
26: \Delta\nu_{fwhm} = 80 \pm 18$. The average signal amplitude is
27: $\approx 3\%$ (rms). The detection of 45 Hz burst oscillations
28: from EXO 0748--676 provides compelling evidence that this is the
29: neutron star spin frequency in this system. We use the inferred
30: spin frequency to model the widths of absorption lines from the
31: neutron star surface and show that the widths of the absorption
32: lines from EXO 0748--676 recently reported by Cottam et al. are
33: consistent with a 45 Hz spin frequency as long as the neutron star
34: radius is in the range from about 9.5 -- 15 km. With a known spin
35: frequency, precise modelling of the line profiles from EXO
36: 0748--676 holds great promise for constraining the dense matter
37: equation of state.
38:
39: \end{abstract}
40: \keywords{binaries: general---stars: individual (EXO
41: 0748--676)---stars: neutron---stars: rotation---X-rays:
42: bursts---X-rays: stars}
43:
44: \section{Introduction}
45:
46: High time resolution observations of thermonuclear X-ray bursts
47: with the \textit{Rossi X-ray Timing Explorer} (\textit{RXTE}) have
48: now found X-ray brightness oscillations, or burst oscillations,
49: from more than a dozen LMXBs. A large body of evidence supports
50: the conclusion that such oscillations are produced by spin
51: modulation of the X-ray burst flux (for a recent review see
52: Strohmayer \& Bildsten 2003 and references therein). In the last
53: two years the detection of burst oscillations at the known spin
54: frequencies of two accreting millisecond pulsars; SAX
55: J1808.4--3658 (Chakrabarty et al. 2003), and XTE J1814--338
56: (Strohmayer et al. 2003) has provided particularly strong
57: confirmation of the spin modulation hypothesis.
58:
59: EXO 0748--676 (hereafter EXO 0748) is a well studied low-mass
60: X-ray binary (LMXB) discovered by \textit{EXOSAT} in 1985 (Parmar
61: et al. 1985). This transient X-ray burst source exhibits
62: irregular X-ray dips and periodic eclipses (Gottwald et al. 1986).
63: Timing of the eclipses revealed a 3.82 hr orbital period (Parmar
64: et al. 1986; Hertz et al. 1996; Wolff et al. 2002). Based on the
65: eclipse duration and orbital period the system inclination angle
66: is constrained to be in the range 75--82$\degr$ (Parmar et al.
67: 1986), which depends on assumptions about the neutron star's
68: companion. X-ray timing studies of the source have revealed the
69: presence of $\approx$1 Hz quasi-periodic oscillations (QPOs), as
70: well as a single kilohertz QPO, which is presumably the lower kHz
71: peak of a pair (Homan et al. 1999; Homan \& van der Klis 2000).
72: Homan \& van der Klis (2000) searched a total of 10 X-ray bursts
73: from EXO 0748 for burst oscillations. They searched the 100--1000
74: Hz frequency band (with 2 Hz resolution), but found no significant
75: signals. They placed upper limits on the amplitude of oscillations
76: to be between 4 \% and 11 \% (rms) during the rise of the bursts
77: in the 2--60 keV band.
78:
79: EXO 0748 is one of few neutron star LMXBs for which
80: high-resolution spectroscopic observations have been obtained for
81: a large number of X-ray bursts. Cottam et al. (2002) reported the
82: presence of narrow absorption lines in a study of EXO 0748 burst
83: spectra observed with the Reflection Grating Spectrometer (RGS) on
84: \textit{XMM--Newton}. Using co-added data from 28 bursts, they
85: found evidence for absorption features between 13--14 $\AA$ which
86: they interpreted as redshifted absorption lines of the n = 2 to 3
87: transitions from hydrogen- and helium-like Fe. Their line
88: identifications imply a neutron star surface redshift of $z =
89: 0.35$. Here, $1 + z = (1 - 2GM/c^2 R)^{-1/2}$, where $M$ and $R$
90: are the neutron star mass and radius, respectively. A measurement
91: of $z$ fixes the mass to radius ratio, or the compactness, but
92: additional information is required to determine both the mass and
93: radius separately.
94:
95: A measurement of the neutron star spin frequency in EXO 0748 could
96: provide the additional information required to measure both its
97: mass and radius. This is possible because the observed width of
98: surface spectral lines provides information about the surface
99: rotational velocity, $v_{rot}$, and thus the stellar radius if the
100: spin frequency, $\nu_{spin}$, is known, viz. $v_{rot} \propto
101: \nu_{spin} R\,$. Even at modest rotation rates it is expected that
102: rotational Doppler broadening will dominate over thermal Doppler
103: and Stark broadening (see Bildsten, Chang \& Paerels 2003).
104: Rotational broadening depends on the surface rotational velocity,
105: $v_{rot} = 2\pi \nu_{spin} R \sin i$, where $i$ is the system
106: inclination (assuming the rotation axis is perpendicular to the
107: orbital plane). Since the inclination of EXO 0748 is tightly
108: constrained by the presence of eclipses, a measurement of the
109: stellar radius would, in principle, be possible if the spin
110: frequency were known. A radius determination, combined with $z$,
111: would then allow for a constraint to be placed on the mass. Until
112: now, this has not been possible because burst oscillations and
113: surface absorption lines have not been observed in the same
114: source. Precise neutron star mass and radius measurements are
115: crucial for understanding the equation of state (EOS) of
116: supranuclear density matter (see, for example, Lattimer \& Prakash
117: 2001; Strohmayer 2004).
118:
119: In this Letter we report the discovery of 45 Hz burst oscillations
120: from EXO 0748 with the proportional counter array (PCA) onboard
121: \textit{RXTE}. We investigate the implications of this finding for
122: the widths of absorption lines from the neutron star surface. In
123: \S 2 we summarize our search and detection of a 45 Hz burst
124: oscillation signal in the average power spectrum of the decay
125: phases of 38 bursts from EXO 0748. In \S 3 we show that the width
126: of the lines observed from EXO 0748 with the RGS are consistent
127: with a 45 Hz spin frequency if the neutron star radius is 9.5--15
128: km. We conclude in \S 4 with a brief summary of our results.
129:
130: \section{Observations and Data Analysis}
131:
132: Since oscillation searches in individual bursts from EXO 0748 have
133: been unsuccessful, we elected to search by ``stacking''
134: (averaging) the power spectra from all available bursts. We began
135: by searching the public \textit{RXTE} data archive and found a
136: total of 38 type I X-ray bursts from EXO 0748. All the bursts were
137: observed (between 15 August, 1996 and 19 February, 2003) with the
138: PCA and high time resolution event mode data were available. For
139: the purposes of computing power spectra we used light curves
140: sampled at 4096 Hz, yielding a Nyquist frequency of 2048 Hz. Since
141: burst oscillation amplitudes generally increase with photon energy
142: (see Strohmayer et al. 1997; Muno, \"{O}zel \& Chakrabarty 2003),
143: we computed power spectra in the energy band 6--60 keV. All power
144: spectra were normalized such that a pure Poisson noise process
145: would be flat with a mean of 2 (see Leahy et al. 1983).
146:
147: From the sample of 38 bursts we computed two average power
148: spectra; one each for the rises and decays of the burst profiles.
149: We selected rising intervals from just prior to burst onset to
150: near the burst peak. We started the decay intervals at the end of
151: the rise intervals and stopped when the countrate had fallen to
152: about 5\% of the peak rate (above the pre-burst level). For both
153: the rises and decays we ``rounded-off'' the intervals so that the
154: length of each was an even multiple of the shortest length.
155: Because we have bursts of differing length this allowed us to
156: linearly rebin the power spectra of each burst to the same
157: frequency resolution before averaging. This procedure resulted in
158: rise intervals lasting 5 or 10 s and decay intervals from 16 to
159: 256 s, with most being 64 or 128 s.
160:
161: Since burst oscillations can drift in frequency by order 1--2 Hz
162: we rebinned each individual burst power spectrum to 1 Hz
163: resolution and then averaged the 1 Hz power spectra of all the
164: bursts (rises and decays separately). We estimated the errors on
165: the individual burst power spectra using the statistical error
166: associated with rebinning (averaging) $N_i$ independent powers for
167: the $i^{\rm th}$ burst, viz. $\sigma_i = 2/\sqrt(N_i)$. We then
168: propagated these uncertainties to estimate the errors for the
169: averaged spectra. Figure 1 shows the average power spectrum
170: computed from the decaying portion of the bursts in the 1--2048 Hz
171: band. This spectrum contains a prominent peak at $\approx$45 Hz.
172: The increase in mean power below about 20 Hz is not unexpected,
173: since, by definition, the decay light curves are all trending down
174: in countrate.
175:
176: In order to quantify the significance of this peak we need to
177: understand the noise power distribution of our power spectrum.
178: First, we fitted the continuum power level from 20--2048 Hz with a
179: constant + power-law model, excluding the 45 Hz peak to avoid
180: biasing the fit to higher values. We then rescaled our power
181: spectrum by dividing by the best fitting continuum model. Figure 2
182: shows the power spectrum and best fitting continuum model (top)
183: along with the rescaled spectrum (bottom). To estimate the noise
184: power distribution we computed a histogram of the number of noise
185: powers with power between $p_i$ and $p_i+\Delta p$, using $\Delta
186: p = 0.01$. Figure 3 shows the resulting noise power histogram and
187: best fitting $\chi^2$ distribution (solid). A $\chi^2$
188: distribution with a mean of 2 and 3,446 degrees of freedom fits
189: extremely well. This distribution is, to high accuracy, a Gaussian
190: with a mean of $2.000 \pm 0.0011$ and a standard deviation of
191: $\sigma=0.0481 \pm 0.0008$. It is not unexpected that the
192: distribution is effectively Gaussian, especially given the large
193: number of independent powers averaged (van der Klis 1989).
194:
195: We now estimate the significance of the 45 Hz peak using the
196: fitted $\chi^2$ distribution. The 45 Hz peak has a power value of
197: 2.335 (vertical dashed line in Figure 3), and the single trial
198: probability of obtaining this value from the fitted distribution
199: is $2.1 \times 10^{-11}$. Using a conservative number of trials of
200: 2048 (the number of bins in our power spectrum), we arrive at a
201: significance of $4.3 \times 10^{-8}$, indicating a strong
202: detection. At 0.25 Hz resolution we could just resolve the peak. A
203: fit with a Lorentzian function gives a frequency $\nu_0 = 44.7 \pm
204: 0.06$ Hz, yielding a quality factor of $Q = \nu _0 /
205: \Delta\nu_{fwhm} = 80 \pm 18$. This peak corresponds to an
206: average signal amplitude of $\approx 3\%$ (rms).
207:
208: To further investigate the association of the 45 Hz signal with
209: the X-ray bursts from EXO 0748 we computed two additional power
210: spectra in the same manner as before; one for the first half (ie.
211: the brighter half) of the burst decay intervals and the other for
212: the second half. These two power spectra are compared in Figure 4.
213: The power spectrum from the first half (top) shows a strong peak
214: at 45 Hz while no signal is detected in the other (fainter) half.
215: This demonstrates that the 45 Hz signal is associated with the
216: brighter portions of the burst decay profiles, as is typical for
217: burst oscillations in other sources. As an additional test of
218: robustness, we divided the burst sample into two sets (each with
219: 19 bursts) and computed average power spectra for each set. We
220: detected the 45 Hz peak in both sets.
221:
222: \section{The Spin Frequency and Line Profiles}
223:
224: We have detected an oscillation signal at 45 Hz during X-ray
225: bursts from EXO 0748. The frequency width of the signal and its
226: clear association with the X-ray bursts (see Figure 4) indicates
227: that it is most likely a burst oscillation signal similar to those
228: seen in more than a dozen other LMXBs, therefore establishing a
229: spin frequency of 45 Hz for the neutron star in EXO 0748.
230: Interestingly, this is the slowest spin period yet measured for a
231: burst oscillation source. This begs the question, why is the spin
232: frequency so much lower in EXO 0748 compared to other LMXBs? One
233: possibility is that the source is in spin equilibrium with a
234: neutron star magnetic field that is stronger than in other LMXBs.
235: Using typical estimates of the accretion torque (see, for example,
236: Psaltis \& Chakrabarty 1999) one would require a magnetic field
237: ranging from $B \approx 1.5-5 \times 10^9$ G, assuming a
238: time-averaged mass accretion rate in the range 0.01--0.1 $\dot
239: m_{Edd}$, where $\dot m_{Edd}$ is the Eddington accretion rate.
240: Knowledge of the spin frequency also has implications for the kHz
241: QPO frequencies expected from the system. Based on behavior from
242: other sources (ie. those with $\nu_{spin} < 400$ Hz), one would
243: expect a frequency separation of 45 Hz for the kHz QPO pair. This
244: would make a nice test case; however, the comparatively small
245: frequency separation (comparable to the width of the lower kHz QPO
246: in some systems) could make resolving both peaks difficult.
247:
248: The detection of the spin frequency in EXO 0748 is very important
249: given the evidence for narrow, gravitationally redshifted
250: absorption lines from this object (Cottam et al. 2002). To
251: investigate the consistency of the absorption line widths observed
252: by Cottam et al. (2002) with a neutron star spin frequency of 45
253: Hz, we computed model line profiles from a rotating neutron star
254: (see also, Datta \& Kapoor 1988; {\" O}zel \& Psaltis 2003). Our
255: modelling builds on that previously described by Nath, Strohmayer
256: \& Swank (2002) and is discussed in detail by Strohmayer (2004).
257: Briefly, the model includes photon deflection in the Schwarzschild
258: metric, relativistic beaming and aberration, gravitational
259: redshifts, and allows for arbitrary viewing geometries. We assume
260: the whole surface of the neutron star is involved in line
261: formation, and that the rotational axis is perpendicular to the
262: orbit plane. We use an inclination of $78.5 \degr$. We model the
263: intrinsic line shape as a Gaussian with a width (FWHM) fixed at
264: the value due to Stark and thermal Doppler broadening estimated by
265: Bildsten, Chang \& Paerels (2003) for the H$\alpha$ transition.
266: Finally, we convolve the model line profile observed at infinity
267: with an RGS response model appropriate for the 13--14 $\AA$ band.
268:
269: A detailed model fit to the RGS absorption line data is beyond the
270: scope of this paper; however, we can compare the measured line
271: width with our line broadening model and place some preliminary
272: limits on the stellar radius. To do this we used data from Cottam
273: et al. (2002, Figure 1) to model the widths of the absorption
274: lines at 13 and 13.75 $\AA$. We find, from a joint fit using
275: Gaussian profiles and only one width parameter for both lines, a
276: value of $\Delta\lambda / \lambda_{fwhm} = 0.018 \pm 0.004\,$.
277: Figure 5 shows absorption line profiles for a neutron star
278: spinning at 45 Hz with different radii. As expected, larger
279: neutron stars produce broader absorption lines, and even at 45 Hz
280: the spin Doppler broadening dominates. At the half-intensity value
281: of each line profile in Figure 5 is a horizontal line. The thin
282: part of each line represents the maximum absorption line width,
283: while the thick part denotes the minimum line width (both at
284: $1\sigma$). The vertical dotted lines denote the best-fit value of
285: $\Delta\lambda / \lambda_{fwhm} = 0.018\,$. One can see from
286: Figure 5 that radii in the range $9.5 < R < 15$ km (or, assuming
287: $z$ is known, masses in the range $1.5\, M_{\sun} < M < 2.3\,
288: M_{\sun}$) are consistent with the derived line widths, and that
289: the observed lines are best matched with $R \approx 11.5$ km ($M
290: \approx 1.8\, M_{\sun}$).
291:
292: This comparison demonstrates that the observed line widths are
293: consistent with a 45 Hz neutron star spin frequency and a
294: reasonable range of neutron star radii. It also shows that precise
295: measurements of the absorption line widths can, in principle, lead
296: to an accurate measurement of the stellar radius. We will attempt
297: to derive more precise radius constraints from detailed model
298: fitting in future work.
299:
300: \section{Summary}
301:
302: Our discovery of burst oscillations from EXO 0748--676 has
303: established a spin frequency of 45 Hz for the accreting neutron
304: star in this system. We have also shown that the observed widths
305: of the absorption lines claimed by Cottam et al. (2002) to be
306: gravitationally redshifted lines from the neutron star surface are
307: consistent with such a spin frequency and with reasonable neutron
308: star radii. This provides at least indirect support for the idea
309: that the absorption lines {\it could} indeed come from the neutron
310: star surface (ie. they are {\it not} too narrow). Finally, we
311: have shown that detailed modelling of absorption line profiles
312: combined with spin measurements will likely provide a means to
313: accurately measure both the masses and radii of neutron stars, and
314: thus tightly constrain the dense matter EOS.
315:
316: A. R. V. is grateful for the NASA Harriett G. Jenkins Pre-doctoral
317: Fellowship Program Mini-Grant Award in support of this work.
318:
319: \begin{thebibliography}{}
320: \bibitem[Barret et al.(1997)]{bar97} Barret, D. et al. 1997, IAU Circ., 6793.
321: \bibitem[]{BCP03}Bildsten, L., Chang, P., \& Paerels, F. 2003, ApJ,
322: 591, L29.
323: \bibitem[]{chaky03}Chakrabarty, D., et al. 2003, Nature, 424, 42.
324: \bibitem[Cottam et al.(2002)]{cot02} Cottam, J., Paerels, F., \&
325: Mendez, M. 2002, Nature, 420, 51.
326: \bibitem[]{DK88}Datta, B. \& Kapoor, R. C. 1988, ApJ, 331, 784.
327: \bibitem[Gottwald et al.(1986)]{got86} Gottwald, M., Haberl, F., Parmar,
328: A. N., \& White, N. E. 1986, ApJ, 308, 213.
329: \bibitem[Homan et al.(1999)]{hom99} Homan, J., Jonker, P. G.,
330: Wijnands, R., van der Klis, M., \& van Paradijs, J. 1999, ApJ,
331: 516, L91.
332: \bibitem[Homan \& van der Klis(2000)]{hom00} Homan, J. \& van der
333: Klis, M. 2000, ApJ, 539, 847.
334: \bibitem[]{LP01}Lattimer, J. M. \& Prakash, M. 2001, ApJ, 550, 426.
335: \bibitem[]{L83}Leahy, D. A., et al. 1983, ApJ, 266, 160.
336: \bibitem[Marshall \& Markwardt(1999)]{mar99} Marshall, F. E. \&
337: Markwardt, C. B. 1999, IAU Circ., 7103.
338: \bibitem[]{MOC03}Muno, M. P., \"{O}zel, F., \& Chakrabarty, D. 2003, ApJ,
339: 595, 1066.
340: \bibitem[]{NSS02}Nath, N. R., Strohmayer, T. E., \& Swank, J. H. 2002, ApJ,
341: 564, 353.
342: \bibitem[]{OP03}{\" O}zel, F. \& Psaltis, D. 2003, ApJ, 582, L31.
343: \bibitem[Parmar et al.(1986)]{par86} Parmar, A. N., White, N. E.,
344: Giommi, P., \& Gottwald, M. 1986, ApJ, 308, 199.
345: \bibitem[Parmar et al.(1985)]{par85} Parmar, A. N., White, N. E.,
346: Giommi, P., \& Haberl, F. 1985, IAU Circ., 4039.
347: \bibitem[]{psa99} Psaltis, D. \& Chakrabarty, D. 1999, ApJ, 521,
348: 332.
349: \bibitem[]{Stroh04}Strohmayer, T. E. 2004, in X-ray Timing 2003: Rossi and
350: Beyond, ed. P. Kaaret, F. K. Lamb, \& J. H. Swank, (AIP Conference
351: Proceedings, Vol. 714), 245-252.
352: \bibitem[]{SB03} Strohmayer, T. E. \& Bildsten, L. 2003, in Compact
353: Stellar X-ray Sources, ed. W. H. G. Lewin \& M. van der Klis
354: (Cambridge: Cambridge Univ. Press), in press (astro-ph/0301544).
355: \bibitem[Strohmayer et al.(2003)]{str03} Strohmayer, T. E.,
356: Markwardt, C. B., Swank, J. H., \& in't Zand, J. J. M. 2003, ApJ,
357: 597, L67.
358: \bibitem[Strohmayer \& Markwardt(2002)]{str02} Strohmayer, T. E.
359: \& Markwardt, C. B. 2002 ApJ, 577, 337.
360: \bibitem[Strohmayer et al.(1997)]{str97} Strohmayer, T. E.,
361: Jahoda, K., \& Giles, A. B. 1997, ApJ, 486, 355.
362: \bibitem[]{vdk89}van der Klis, M. 1989, in Timing Neutron Stars, ed.
363: H. Ogelman \& E. P. J. van den Heuvel, (Kluwer Academic
364: Publishers), 27-69.
365: \end{thebibliography}
366:
367: \pagebreak
368:
369: \begin{figure}
370: \begin{center}
371: \includegraphics[width=6in, height=6in]{f1.eps}
372: \end{center}
373: Figure 1: Average Leahy-normalized power spectrum of the decay
374: intervals of 38 X-ray bursts from EXO 0748 in the 1--2048 Hz band.
375: The frequency bins are 1 Hz and the Nyquist frequency is 2048 Hz.
376: Note the prominent peak at $\approx$45 Hz. The increase in power
377: towards low frequencies is due to the decrease in countrate with
378: time during the burst decays. A characteristic error bar is also
379: shown.
380:
381: \end{figure}
382: \clearpage
383:
384: \begin{figure}
385: \begin{center}
386: \includegraphics[width=6in,
387: height=6in]{f2.eps}
388: \end{center}
389: Figure 2: The same power spectrum from Figure 1 is shown in the
390: 20--2048 Hz band along with the best fitting constant + power law
391: model of the continuum (top). The renormalized power spectrum
392: obtained by dividing by the continuum model (and multiplying by 2)
393: is also shown (bottom). The renormalized spectrum has been
394: displaced by $-0.6$ for clarity.
395: \end{figure}
396: \clearpage
397:
398: \begin{figure}
399: \begin{center}
400: \includegraphics[width=6in, height=6in]{f3.eps}
401: \end{center}
402: Figure 3: Distribution of noise powers for the average
403: renormalized power spectrum shown in Figure 2 (histogram). The
404: best fitting $\chi^2$ model is also shown (solid). The
405: distribution has been scaled to zero mean and the ordinate is in
406: units of standard deviations ($\sigma$'s). The power level of the
407: 45 Hz peak is marked by the vertical dashed line and is
408: $\approx$7$\sigma$ from the mean.
409: \end{figure}
410: \clearpage
411:
412: \begin{figure}
413: \begin{center}
414: \includegraphics[width=6in, height=6in]{f4.eps}
415: \end{center}
416: Figure 4: Comparison of average power spectra computed from the
417: 1st half (by time) of the burst decay intervals (top) and the 2nd
418: half (bottom). The 45 Hz signal is clearly associated with the 1st
419: half of the decay intervals.
420: \end{figure}
421: \clearpage
422:
423: \begin{figure}
424: \begin{center}
425: \includegraphics[width=6in, height=6in]{f5.eps}
426: \end{center}
427: Figure 5: Model absorption line profiles from a neutron star
428: rotating at 45 Hz for radii of 9.5 (dashed), 11.5 (solid) and 15
429: (dot-dashed) km. A horizontal line is located at the
430: half-intensity point for each profile. The thin portion of each
431: line represents the maximum width consistent with the RGS data
432: from Cottam et al. ($1\sigma$), while the thick portion represents
433: the minimum width (again, at $1\sigma$). The vertical dotted lines
434: denote the best-fit line width. Comparison of the model profiles
435: and the measured line widths suggests $R = 11.5^{+3.5}_{-2.5}$.
436: The model lines have been convolved with a line response function
437: appropriate for the RGS at 13.5 $\AA$. This accounts for the
438: shallow but broad wings of the profiles.
439: \end{figure}
440: \clearpage
441:
442: \end{document}
443: