1: %\documentclass[preprint, 12pt]{aastex}
2: \documentclass[apj]{emulateapj}
3: \usepackage{apjfonts}
4: \usepackage{graphicx}
5:
6: \renewcommand{\vec}[1]{\mbox{\boldmath $\displaystyle #1$}}
7: \newcommand{\grad}{\vec{\nabla}}
8: \newcommand{\lap}{\nabla^2}
9: \newcommand{\vdot}{\vec{\cdot}}
10: \newcommand{\vcross}{\vec{\times}}
11: \newcommand{\divr}{\grad\vdot\,}
12: \newcommand{\curl}{\grad\vcross\,}
13: \newcommand{\Ye}{Y_{\mathrm{e}}}
14: \newcommand{\Ledd}{L_{\mathrm{Edd}}}
15:
16: \usepackage{color}
17: \newcommand{\xxx}[1]{\textcolor{red}{\textbf{[#1]}}}
18:
19: \begin{document}
20:
21: \title{Millihertz Quasi-Periodic Oscillations from Marginally Stable
22: Nuclear Burning on an Accreting Neutron Star}
23:
24: \author{
25: Alexander Heger\altaffilmark{1,2},
26: Andrew Cumming\altaffilmark{3}, \&
27: Stanford E.~Woosley\altaffilmark{2}}
28: \altaffiltext{1}{Theoretical Astrophysics Group, T-6, MS B227,
29: Los Alamos National Laboratory, Los Alamos, NM 87545; aheger@lanl.gov}
30: \altaffiltext{2}{Department of Astronomy and Astrophysics, University
31: of California, Santa Cruz, CA 95064; alex@ucolick.org, woosley@ucolick.org}
32: \altaffiltext{3}{Physics Department, McGill University, 3600 rue
33: University, Montreal, QC, H3A 2T8, Canada; cumming@physics.mcgill.ca}
34:
35: \begin{abstract}
36: We investigate marginally stable nuclear burning on the surface of
37: accreting neutron stars as an explanation for the mHz quasi-periodic
38: oscillations (QPOs) observed from three low mass X-ray binaries. At
39: the boundary between unstable and stable burning, the temperature
40: dependence of the nuclear heating rate and cooling rate almost
41: cancel. The result is an oscillatory mode of burning, with an
42: oscillation period close to the geometric mean of the thermal and
43: accretion timescales for the burning layer. We describe a simple
44: one-zone model which illustrates this basic physics, and then present
45: detailed multizone hydrodynamical calculations of nuclear burning
46: close to the stability boundary using the \textsc{Kepler} code. Our
47: models naturally explain the characteristic 2 minute period of the mHz
48: QPOs, and why they are seen only in a very narrow range of X-ray
49: luminosities. The oscillation period is sensitive to the accreted
50: hydrogen fraction and the surface gravity, suggesting a new way to
51: probe these parameters. A major puzzle is that the accretion rate at
52: which the oscillations appear in the theoretical models is an order of
53: magnitude larger than the rate implied by the X-ray luminosity when
54: the mHz QPOs are seen. We discuss the implications for our general
55: understanding of nuclear burning on accreting neutron stars. One
56: possibility is that the accreted material covers only part of the
57: neutron star surface at luminosities $L_X\gtrsim 10^{37}\ {\rm erg\
58: s^{-1}}$.
59: \end{abstract}
60:
61: \keywords{accretion, accretion disks---X-rays:bursts---stars:neutron}
62:
63:
64: \section{Introduction}
65:
66: Low mass X-ray binaries, in which a neutron star or black hole
67: accretes from a low mass companion, exhibit a range of periodic and
68: quasi-periodic phenomena, ranging in frequency from very low frequency
69: (mHz) noise to kHz quasi-periodic oscillations (QPOs) (see van der
70: Klis 2004 for a recent review). This variability has mostly been
71: associated with orbiting material in the accretion flow close to the
72: compact object. In the case of a neutron star accretor, an important
73: question is whether any of these phenomena originate from or are
74: associated with the neutron star surface. This is important for
75: identifying the compact object as a neutron star or a black hole and
76: offers a probe of the neutron star surface layers.
77:
78: Unstable nuclear burning on neutron star surfaces has been studied for
79: many years, and is observed as Type I X-ray bursts. The accreted
80: hydrogen (H) and helium (He) fuel accumulates on the surface of the
81: star and undergoes a thin shell instability, giving rise to a
82: $10$--$100$ second burst of X-rays with typical energy $10^{39}$ ergs
83: (for reviews, see Lewin, van Paradijs, \& Taam 1993, 1995; Strohmayer
84: \& Bildsten 2003). Not all sources show Type I X-ray bursts, however,
85: and in many sources the bursts are not frequent enough to burn all of
86: the accreted fuel (van Paradijs, Penninx, \& Lewin 1988; in 't Zand et
87: al.~2003). Bildsten (1993, 1995) suggested that a different mode of
88: nuclear burning, involving slowly propagating fires over the neutron
89: star surface, operates at high accretion rates, and manifests itself
90: in the power spectrum of the source as very low frequency noise
91: (VLFN). He found an anti-correlation between bursting and VLFN,
92: supporting this picture.
93:
94: Revnivtsev et al.~(2001) discovered a new class of mHz QPOs in three
95: Atoll sources, 4U~1608-52, 4U~1636-53, and Aql X-1, which they
96: proposed were from a special mode of nuclear burning on the neutron
97: star surface rather than from the accretion flow. These mHz QPOs have
98: frequencies in the range $7$--$9\ {\rm mHz}$ (timescales of
99: $1.9$--$2.4$ minutes). The associated flux variations are at the few
100: per cent level, and are strongest at low photon energies ($\lesssim 5$
101: keV). This is in contrast to all other observed QPOs, whose amplitude
102: generally increases with photon energy. Revnivtsev et al.~(2001)
103: showed that the centroid frequency of the mHz QPO was stable on year
104: timescales in a given source, and the same to within tens of per cent
105: in all three sources in which it was detected. In addition, the
106: presence of the mHz QPOs is affected by Type I X-ray bursts: in
107: 4U~1608-52, the mHz QPOs disappeared immediately following a Type I
108: X-ray burst. In 4U~1608-52, a transient source whose luminosity is
109: observed to change by orders of magnitude, the mHz QPO was only
110: present within a narrow range of luminosity, $L_X\approx
111: 0.5$--$1.5\times 10^{37}\ {\rm erg\ s^{-1}}$.
112:
113: The association of the mHz QPOs with a surface phenomenon was
114: strengthened by the results of Yu \& van der Klis (2002), who showed
115: that the kHz QPO frequency is anti-correlated with the luminosity
116: variations during the mHz oscillation. This is opposite to the long
117: term trend, which is that the kHz QPO frequency varies proportional to
118: the X-ray luminosity, consistent with the inner edge of the accretion
119: disk being pushed inwards at higher accretion rates. The
120: anti-correlation observed by Yu \& van der Klis (2002) during the mHz
121: QPO cycle suggests that the inner edge of the disk moves outwards
122: slightly as the luminosity increases during the cycle, perhaps
123: consistent with enhanced radiation drag as the gas orbiting close to
124: the neutron star is radiated by emission from the neutron star
125: surface.
126:
127:
128:
129: The fact that the properties of the mHz QPOs suggest that they are
130: associated with the neutron star surface led Revnivtsev et al.~(2001)
131: to propose that a special mode of nuclear burning operates at
132: luminosities $L_X\approx 0.5$--$1.5\times 10^{37}\ {\rm erg\ s^{-1}}$.
133: The nature of the burning and the physics underlying the
134: characteristic timescale of $\approx 2$ minutes, however, were
135: unexplained. Bildsten (1993) gave the characteristic timescales of the
136: burning layer that might be associated with sub-hertz phenomena. The
137: thermal timescale of the layer is $t_{\rm therm}\approx 10$ s, the
138: time to accrete the fuel is $t_{\rm accr}\approx 1000$ s at the
139: Eddington accretion rate, and the time for a nuclear burning ``fire''
140: to propagate around the surface is estimated to be $\sim 10^4$
141: s. Bildsten (1993) proposed that if several fires are propagating around
142: the star at a given time, the time-signature would be broad band noise
143: with frequencies in the mHz range. None of the timescales he
144: identified match the $\approx 2$ minute mHz QPO period, however.
145:
146:
147: The luminosity at which mHz QPOs are observed ($L_X\approx 10^{37}\
148: {\rm erg\ s^{-1}}$) is significant because it is similar to the
149: luminosity at which a transition in burning behavior occurs, from
150: frequent Type I X-ray bursting at low accretion rates to the
151: disappearance of Type I X-ray bursts at high accretion rates. This
152: transition is common to many X-ray bursters (e.g., Cornelisse et
153: al.~2003), and is expected theoretically because at high accretion
154: rates the fuel burns at a higher temperature, reducing the
155: temperature-sensitivity of helium burning and quenching the thin shell
156: instability. An outstanding puzzle is that theory predicts
157: a transition accretion rate close to the Eddington rate (Bildsten
158: 1998), which corresponds to luminosities $L_X\sim 10^{38}\ {\rm erg\
159: s^{-1}}$, much larger than observed. Paczynski (1983) pointed out that
160: near the transition from instability to stability, oscillations are
161: expected because the eigenvalues of the system are complex. Narayan \&
162: Heyl (2003) extended Paczynski's one-zone analysis, calculating linear
163: eigenmodes of truncated steady-state burning models. They too found
164: complex eigenvalues near the stability boundary, and suggested that
165: this might explain the mHz QPOs observed by Revnivtsev et al.~(2001).
166: The oscillation frequencies, however, were an order of magnitude too
167: small.
168:
169: In this paper, we show that the mHz QPO frequencies are, in fact,
170: naturally explained as being due to marginally stable nuclear burning
171: on the neutron star surface. At the boundary between unstable and
172: stable burning, the temperature dependence of the nuclear heating rate
173: and cooling rate almost cancel. The result is an oscillatory mode of
174: burning, with an oscillation period close to the geometric mean of the
175: thermal and accretion timescales for the burning layer, $(t_{\rm
176: therm}t_{\rm accr})^{1/2}\approx 100\ {\rm s}$. In \S 2, we describe a
177: simple one-zone model which illustrates this basic physics, and then
178: present detailed hydrodynamical calculations of nuclear burning close
179: to the stability boundary in \S 3. We discuss the implications of our
180: results in \S 4. In particular, if the mHz QPOs are due to marginally
181: stable nuclear burning, the local accretion rate onto the star must be
182: close to the Eddington rate, even though the global accretion rate
183: inferred from the X-ray luminosity is ten times lower.
184:
185:
186: \section{A one-zone model}
187:
188: In this section, we discuss a simplified model of the burning layers
189: which illustrates the basic physics underlying the oscillations
190: observed in our multizone numerical simulations. Following Paczynski
191: (1983), we consider a one-zone model of the burning layer. The
192: temperature, $T$, and thickness of the fuel layer, $y$ (which we
193: measure as a column depth, in units of mass per unit area), obey
194: \begin{eqnarray}\label{eq:heateqn}
195: c_P{dT\over dt}&=&\epsilon -{F\over y}
196: \\\label{eq:conteqn}
197: {dy\over dt}&=&\dot m - {\epsilon\over E_\star}y.
198: \end{eqnarray}
199: (equivalent to eqns.~[8] of Paczynski 1983). Equation
200: (\ref{eq:heateqn}) describes the heat balance, including heating of
201: the layer by nuclear reactions $\epsilon$, and radiative cooling
202: $-\divr\vec{F}/\rho=dF/dy\approx F/y$. The heat capacity at constant
203: pressure is $c_P$, and $F$ is the outwards heat flux. Equation
204: (\ref{eq:conteqn}) tracks the burning depth, allowing for accretion of
205: new fuel at a rate given by the local accretion rate $\dot m$, as well
206: as burning of fuel on a timescale $E_\star/\epsilon$, where $E_\star$
207: is the energy per gram released in the burning. Note that the pressure
208: at the base of the layer is $P=gy$ from hydrostatic balance, where $g$
209: is the local gravity. We first study equations (\ref{eq:heateqn}) and
210: (\ref{eq:conteqn}) analytically (\S 2.1), and then show some numerical
211: integrations (\S 2.2).
212:
213: \subsection{Analytic estimates}
214:
215:
216:
217: Equations (\ref{eq:heateqn}) and (\ref{eq:conteqn}) constitute a
218: nonlinear oscillator. To understand its behavior, we consider linear
219: perturbations to the steady state solution, which has
220: \begin{equation}\label{eq:steadystate}
221: \epsilon y = F = \dot mE_\star.
222: \end{equation}
223: The nuclear energy generation is generally a strong function of
224: temperature, and we will assume $\epsilon\propto T^\alpha$, where
225: $\alpha\equiv d\ln\epsilon/d\ln T$. The heat flux is approximately
226: $F\approx acT^4/3\kappa y$ (e.g., Bildsten 1998). To simplify the
227: algebra in this section, we assume that $\epsilon$ depends only on
228: temperature, and that the opacity $\kappa$ is a constant. We write the
229: deviations from steady state as $\delta y$ and $\delta T$, giving
230: \begin{eqnarray}
231: c_P{\partial\delta T\over \partial t}&=&\alpha\epsilon{\delta T\over
232: T}-{4F\over y}{\delta T\over T}+{2F\over y}{\delta y\over y}\\
233: {\partial \delta y\over\partial t}&=&-{\epsilon\over
234: E_\star}\left(\delta y+{\alpha y\over T}\delta T\right).
235: \end{eqnarray}
236: Defining a thermal timescale for the layer $t_{\rm
237: therm}=c_pT/\epsilon=c_pTy/F$ and an accretion timescale $t_{\rm
238: accr}=y/\dot m$, and using the steady-state relations of equation
239: (\ref{eq:steadystate}) gives
240: \begin{eqnarray}\label{eq:pert1}
241: {\partial\over\partial t}\left({\delta T\over
242: T}\right)&=&\left({\alpha-4\over t_{\rm therm}}\right){\delta T\over
243: T}+\left({2\over t_{\rm therm}}\right){\delta y\over y}\\
244: \label{eq:pert2}
245: {\partial\over\partial t}\left({\delta y\over
246: y}\right)&=&-\left({1\over t_{\rm accr}}\right){\delta y\over
247: y}-\left({\alpha\over t_{\rm accr}}\right){\delta T\over T}.
248: \end{eqnarray}
249: It is useful to write $\delta T=f(t)\exp(-t/t_{\rm accr})$ and $\delta
250: y=g(t)\exp(-t/t_{\rm accr})$. This simplifies equations
251: (\ref{eq:pert1}) and (\ref{eq:pert2}), allowing us to combine them
252: into a single differential equation for $f$,
253: \begin{equation}\label{eq:osc}
254: {\partial^2f\over\partial t^2}+\left({4-\alpha\over t_{\rm
255: therm}}-{1\over t_{\rm accr}}\right){\partial f\over \partial
256: t}+{2\alpha\over t_{\rm accr}t_{\rm therm}}f=0.
257: \end{equation}
258: Equation (\ref{eq:osc}) is the equation for a damped simple harmonic
259: oscillator. The solution is $\delta T\propto\exp(\lambda t)$, with
260: \begin{equation}
261: \lambda={1\over 2}\left({\alpha-4\over t_{\rm therm}}-{1\over t_{\rm
262: accr}}\right) \pm \left[{1\over 4}\left({\alpha-4\over t_{\rm
263: therm}}+{1\over t_{\rm accr}}\right)^2-{2\alpha\over t_{\rm
264: therm}t_{\rm accr}}\right]^{1/2}.
265: \end{equation}
266: Note that the ``damping'' term in equation (\ref{eq:osc}) can be
267: positive or negative depending on the relative temperature
268: sensitivities of the heating and cooling.
269:
270:
271:
272: \begin{figure}
273: \plotone{timedep3a.ps}
274: \caption{Lightcurves for the one-zone model at three different
275: accretion rates. The system evolves from the arbitrary initial values
276: $y=2\times 10^8\ {\rm g\ cm^{-2}}$ and $T=2\times 10^8\ {\rm K}$. At
277: $\dot m=0.95\ \dot m_{\rm Edd}$, bursts occur with a recurrence time
278: of 34 minutes. At $\dot m=0.998\ \dot m_{\rm Edd}$, oscillations are
279: seen with a period of 4.7 minutes. At $\dot m=1.05\ \dot m_{\rm Edd}$,
280: after a few transient oscillations, the burning evolves to a steady
281: state. The steady state flux is $\dot mE_{\rm nuc}$, where $E_{\rm
282: nuc}\approx 5$ MeV per nucleon.\label{Fig:lc1}}
283: \end{figure}
284:
285:
286:
287: \begin{figure}
288: \plotone{timedep2a.ps}
289: \caption{The trajectories in the temperature-column depth plane for
290: the models shown in Figure 1. In each case, the model starts at the
291: arbitrary point $y=2\times 10^8\ {\rm g\ cm^{-2}}$ and $T=2\times
292: 10^8\ {\rm K}$.\label{Fig:tra1}}
293: \end{figure}
294:
295:
296:
297:
298:
299: The key to understanding the behavior of the burning is to note that
300: the timescales $t_{\rm therm}$ and $t_{\rm accr}$ are very
301: different. Steady burning at accretion rates near Eddington leads to
302: ignition at a column depth $y\approx 10^8\ {\rm g\ cm^{-2}}$ and
303: temperatures $T\approx 5\times 10^8\ {\rm K}$ (Schatz et
304: al.~1999). The accretion timescale is therefore $y/\dot m\approx 1000\
305: {\rm s}$ for accretion at $\dot m\approx 10^5\ {\rm g\ cm^{-2}\
306: s^{-1}}$ (roughly the Eddington rate). For stable burning $F=\dot m
307: E_\star$, giving $t_{\rm therm}=c_PTy/F=(c_PT/E_\star)(y/\dot
308: m)\approx 0.01\ t_{\rm accr}\approx 10\ {\rm s}$, since for an ideal
309: gas, $c_PT\approx (5/2)(k_BT/m_p)=7.2\ {\rm keV}\ T_8$ per nucleon,
310: whereas $E_\star\approx 5\ {\rm MeV}$ per nucleon (Schatz et
311: al.~1999)\footnote{In fact, the gas is partially degenerate at the
312: burning location, with $k_BT\approx E_F$, and so there will be a small
313: correction factor to the ideal gas expression for $c_P$.}.
314:
315: Because $t_{\rm therm}\ll t_{\rm accr}$ the damping term is usually
316: much larger than the oscillatory term, and the system is either
317: positively or negatively overdamped depending on the sign of
318: $(\alpha-4)$. When $\alpha>4$, the damping coefficient is negative,
319: leading to strong exponential growth: the steady state solution is
320: linearly unstable. When $\alpha <4$, the damping coefficient is
321: positive, leading to strong damping: the steady-state solution is
322: linearly stable. Close to marginal stability when
323: $\alpha\approx 4$, however, the effective thermal timescale $t_{\rm
324: therm}/\left|\alpha-4\right|$ becomes much longer than the accretion
325: time, and the inequality is reversed, giving weakly damped or excited
326: oscillations with oscillation frequency $\omega\approx (2\alpha/t_{\rm
327: accr}t_{\rm therm})^{1/2}$. The condition to observe oscillations is
328: that the oscillation frequency should be larger than the damping rate,
329: \begin{equation}
330: \left({2\alpha\over t_{\rm accr}t_{\rm therm}}\right)^{1/2}>{1\over 2}\left({\alpha-4\over t_{\rm therm}}+{1\over t_{\rm accr}}\right).
331: \end{equation}
332: Note that this condition can be satisfied when $\lambda$ has either
333: positive or negative real part, so that the oscillations may be
334: excited or damped.
335:
336:
337: Neglecting the slight
338: modification from the damping term, the oscillation period is $P_{\rm
339: osc}=2\pi/\omega$, or
340: \begin{eqnarray}\label{eq:posc}
341: P_{\rm osc}&\approx &{2\pi\over (2\alpha)^{1/2}}\left({c_PT\over
342: E_\star}\right)^{1/2}\left({y\over \dot m}\right)\nonumber\\ &=&3.1\
343: {\rm mins}\ \left({T_8\over 5}\right)^{1/2}y_8\left({\dot m\over \dot
344: m_{\rm Edd}}\right)^{-1},
345: \end{eqnarray}
346: where we set $\alpha=4$ at marginal stability, and take $E_\star=5$
347: MeV per nucleon. This estimate is remarkably close to the observed mHz
348: QPO period of $2$ minutes.
349:
350:
351:
352:
353:
354: The physics of the oscillations can be understood by considering
355: equations (\ref{eq:pert1}) and (\ref{eq:pert2}). For most accretion
356: rates, the effective thermal time is much smaller than the accretion
357: timescale, or $t_{\rm therm}/\left|4-\alpha\right|\ll t_{\rm
358: accr}$. The perturbations are then effectively at constant pressure,
359: or $\delta y=0$, as commonly assumed (see, e.g., Fujimoto, Hanawa, \&
360: Miyaji 1981; Bildsten 1998), and equation (\ref{eq:pert1}) leads
361: directly to exponential growth or decay depending on the relative
362: temperature sensitivities of the heating and cooling rates. Close to
363: marginal stability, $(\alpha-4)\approx 0$, however, the effective
364: thermal timescale becomes very long compared to the accretion time.
365: Equations (\ref{eq:pert1}) and (\ref{eq:pert2}) in this limit are
366: \begin{equation}\label{eq:pert3}
367: {\partial\over\partial t}\left({\delta T\over T}\right)\approx {2\over t_{\rm
368: therm}}{\delta y\over y}
369: \end{equation}
370: \begin{equation}\label{eq:pert4}
371: {\partial\over\partial t}\left({\delta y\over y}\right)\approx -{\alpha\over t_{\rm
372: accr}}{\delta T\over T}.
373: \end{equation}
374: These equations nicely summarize the physics of the
375: oscillations. Consider an upwards fluctuation in temperature $\delta
376: T>0$. At the point of marginal stability, the increase in heating rate
377: almost exactly cancels the $T^4$ dependence of the cooling rate. The
378: main effect is that the hotter temperature leads to faster burning of
379: the accreting fuel, and a decrease in thickness $\delta y<0$ on a
380: timescale $\sim t_{\rm accr}$ (eq.~[\ref{eq:pert4}]). But a thinner
381: layer cools faster since $F/y\propto 1/y^2$. Therefore, the
382: temperature fluctuation now begins to decrease, this time on the
383: faster timescale $\sim t_{\rm therm}$ (eq.~[\ref{eq:pert3}]). These
384: changes are out phase, driving an oscillation on the intermediate
385: timescale $\approx (t_{\rm therm}t_{\rm accr})^{1/2}$.
386:
387:
388: The behavior we describe here is shown by the canonical example of a
389: nonlinear oscillator, the van der Pol oscillator (e.g., Abarbanel,
390: Rabinovich, \& Sushchik 1993). This oscillator consists of an LC
391: circuit with an active element that can behave as a ``negative
392: resistor'', originally a vacuum tube. The governing equation is of the
393: form $\ddot x+k(x^2-1)\dot x+\omega^2 x=0$. Depending on the choice of
394: control parameter $k$, the behavior of this circuit is a limit cycle
395: (relaxation oscillations) with fast and slow timescales dominating at
396: different parts of the cycle ($k>1$), a strongly damped system which
397: evolves to a steady-state ($k<-1$), or oscillatory with growing or
398: damped oscillations ($|k|<1$). These three states are analogous to
399: bursting, stable burning, and oscillations near the stability boundary
400: in the one-zone model.
401:
402: \subsection{Numerical integrations}
403:
404: We have integrated equations (\ref{eq:heateqn}) and (\ref{eq:conteqn})
405: in time to determine the non-linear evolution of the one-zone
406: model. For the nuclear burning, we use the triple alpha reaction
407: ($3\alpha\rightarrow ^{12}$C) rate as given by Fushiki \& Lamb (1987).
408: We allow for the presence of hydrogen in the accreted fuel, however,
409: by enhancing the energy release from the triple alpha reaction by a
410: factor $E_{\rm nuc}/E_{3\alpha}$, where $E_{3\alpha}=0.606$ MeV per
411: nucleon is the energy release from the triple alpha reaction, and
412: $E_{\rm nuc}$ is the energy release from burning the accreted mixture
413: of hydrogen and helium to iron group. We assume that $E_{\rm
414: nuc}=1.6+4.9X_0$ MeV per nucleon, where $X_0$ is the mass fraction of
415: hydrogen in the accreted layer. This expression for $E_{\rm nuc}$
416: includes an energy loss of 25\% from neutrino emission during
417: \textsl{$\alpha$p}- and \textsl{rp}-process burning (e.g., Fujimoto et
418: al.~1987), and gives $E_{\rm nuc}=5$ MeV per nucleon for $X_0=0.7$, in
419: good agreement with the energy release in the steady state burning
420: models of Schatz et al.~(1999). We write the flux as $F=acT^4/3\kappa
421: y$ (Bildsten 1998), where the opacity $\kappa$ is calculated as
422: described by Schatz et al.~(1999). In addition, we include a flux
423: heating the layer from below of 0.15 MeV per nucleon, and a
424: contribution to the heating rate from hot CNO hydrogen burning in the
425: accumulating fuel layer. Neither of these extra contributions to the
426: heat balance make a significant difference to our results. Note that
427: the amount of hydrogen burned by hot CNO cycle prior to helium
428: ignition is very small at the rapid accretion rates considered here.
429:
430:
431: Figure~\ref{Fig:lc1} shows lightcurves from the one-zone integrations
432: at different accretion rates. To enable a direct comparison with the
433: multizone simulations discussed in \S 3, we set the local gravity to
434: be the Newtonian gravity for a 1.4 $M_\odot$, 10 km neutron star,
435: $g=1.9\times 10^{14}\ {\rm cm^2\ s^{-1}}$. Throughout this paper, we
436: define the local Eddington accretion rate to be $\dot m_{\rm
437: Edd}\equiv 8.8\times 10^4\ {\rm g\ cm^{-2}\ s^{-1}}$. By
438: coincidence\footnote{The physics which stabilizes the burning, the
439: decreasing temperature sensitivity of the triple alpha reaction with
440: increasing temperature, has nothing to do with the physics setting the
441: Eddington luminosity. By coincidence, the transition to stability
442: occurs close to the Eddington accretion rate (e.g., Bildsten 1998).},
443: the stability boundary for this one zone model is very close to the
444: Eddington accretion rate. In Figure~\ref{Fig:lc1}, we show lightcurves
445: at $\dot m=0.95$, $0.998$, and $1.05\ \dot m_{\rm Edd}$.
446: Figure~\ref{Fig:tra1} shows the corresponding tracks in the
447: temperature-column depth plane. We start the simulations with the
448: arbitrary conditions $T=2\times 10^8\ {\rm K}$, and $y=2\times 10^8\
449: {\rm g\ cm^{-2}}$. At accretion rates below the boundary, the system
450: evolves quickly into a limit cycle corresponding to Type I X-ray
451: bursts: slow accumulation of fuel followed by rapid burning. Close to
452: the stability boundary, the recurrence time is $\approx 30$
453: minutes. At accretion rates above the boundary the evolution is to a
454: steady state in which the fuel burns at $T\approx 5\times 10^8\ {\rm
455: K}$ and $y\approx 7\times 10^7\ {\rm g\ cm^{-2}}$. For accretion rates
456: very close to the transition, $\dot m\approx 1\ \dot m_{\rm Edd}$, we
457: find oscillations around the steady state conditions, with oscillation
458: period $\approx 4$ minutes and amplitude a few percent of the
459: Eddington flux. For the oscillatory case, Figure~\ref{Fig:tra1mag}
460: shows a more detailed view of the trajectory of the solution in the
461: temperature-column depth phase space.
462:
463:
464: \begin{figure}
465: \plotone{timedep2.ps}
466: \caption{The trajectory in the temperature-column depth plane for
467: $\dot m=0.998\dot m_{\rm Edd}$, as shown in Figure 2, but zooming in
468: on the oscillations around the steady burning
469: location.\label{Fig:tra1mag}}
470: \end{figure}
471:
472:
473: \begin{figure*}
474: \epsscale{1.0}
475: \plotone{timedepom.ps}
476: \epsscale{1.0}
477: \caption{Oscillation period as a function of accretion rate in the
478: one-zone model, for different choices of surface gravity and accreted
479: hydrogen fraction. At each accretion rate, we integrate the model for
480: $10^6\ {\rm s}$, and plot the mean oscillation period, only including
481: data for $t>100$ minutes. The solid symbols indicate models for which
482: the oscillations grow and reach a steady amplitude; the open symbols
483: indicate models for which the oscillations are damped.\label{Fig:P1}}
484: \end{figure*}
485:
486:
487: Although the one-zone model is approximate, it is useful because it
488: allows us to investigate how the properties of the oscillations change
489: with parameters such as surface gravity and accreted composition. The
490: accreted composition could differ from system to system either due to
491: metallicity variations, or variations in the accreted hydrogen
492: fraction. Intermediate-mass binary evolution models
493: (e.g., Podsiadlowski et al.~2002) predict that the companion star in
494: many systems is hydrogen deficient, so that the hydrogen mass fraction
495: is reduced below the solar composition value of $X_0\approx 0.7$. The
496: importance of the hydrogen fraction is that the nuclear energy release
497: $E_{\rm nuc}$ changes significantly with only small changes in
498: $X_0$. In contrast, we expect that metallicity will not have a large
499: effect on the transition accretion rate or the oscillation period. This is because the
500: metallicity of the accreted material mainly enters into the one-zone
501: model as hot CNO hydrogen burning in the accumulating layer, but
502: the flux from hot CNO burning, $\epsilon_{\rm CNO}y\approx
503: 5\times 10^{21}\ {\rm erg\ cm^{-2}\ s^{-1}}\ y_8(Z_{\rm CNO}/0.01)$,
504: is much smaller than the steady burning flux, $\dot mE_{\rm
505: nuc}\approx 5\times 10^{23}\ {\rm erg\ cm^{-2}\ s^{-1}}\ (\dot m/\dot
506: m_{\rm Edd})$, where $Z_{\rm CNO}$ is the mass fraction of CNO
507: elements.
508:
509: Figure~\ref{Fig:P1} shows the dependence of the oscillation period on
510: accretion rate for different choices of gravity and accreted hydrogen
511: fraction. We show results for $g_{14}=1.9$, corresponding to the
512: Newtonian gravity of a 1.4 $M_\odot$, 10 km neutron star (or the
513: general relativistic gravity for a 1.4 $M_\odot$, 12.3 km neutron
514: star), $g_{14}=2.45$, the gravity of a 1.4 $M_\odot$, 10 km neutron
515: star taking general relativistic corrections into account, and a
516: stronger gravity $g_{14}=3.1$ which corresponds to the general
517: relativistic surface gravity for a $2\ M_\odot$, 11 km neutron
518: star. We also show a model with a hydrogen fraction below solar,
519: $X_0=0.5$. Only the accretion rate range at which oscillations are
520: observed is shown. At each accretion rate, we integrate the one-zone
521: model for $10^6$ seconds, and plot the mean oscillation period after
522: discarding the first 100 minutes of data.
523:
524: For each choice of $g_{14}$ and $X_0$, the pattern is similar. The
525: overall range of accretion rates for which oscillations are seen is
526: very narrow, $\Delta \dot m/\dot m\approx 1$\%. For most of this range
527: the oscillations are decaying. As the stability boundary is approached
528: from below, we first see oscillations which reach a steady amplitude
529: (indicated by solid squares in Fig.~\ref{Fig:P1}) whose frequency
530: drops rapidly with increasing $\dot m$. At larger accretion rates
531: (open squares in Fig.~\ref{Fig:P1}), the oscillations decay with time,
532: on timescales $<1$ day, and have a frequency that is less sensitive to
533: $\dot m$. The transition from growing to decaying oscillations is very
534: rapid. In each case, the model indicated by the last closed square
535: shows stable oscillations for $10^6$ seconds, whereas with only a
536: small increment in accretion rate the next model (first open square)
537: has oscillations which decay on a timescale of $\sim 10^5$
538: seconds. Most interesting is that the oscillation period is sensitive
539: to $g_{14}$ and $X_0$. Increasing gravity or decreasing $X_0$ moves
540: the transition from unstable to stable burning to higher accretion
541: rates, where the oscillation period is shorter. As $X_0$ decreases,
542: the accretion rate range over which the oscillations are growing
543: rather than decaying is larger.
544:
545:
546: \section{Multizone calculations of burning near the stability boundary}
547:
548: We now present detailed multi-zone models of nuclear burning at
549: accretion rates close to the transition from unstable to stable
550: burning. These models are extensions of the calculations presented by
551: Woosley et al.~(2004) using the implicit 1D hydrodynamic code
552: \textsc{Kepler} (Weaver et al.~1978). Woosley et al.~(2004) calculated
553: sequences of X-ray bursts at accretion rates $\dot M\approx 0.03$ and
554: $0.1\ \dot M_{\rm Edd}$. In this paper we show the first results of
555: an extension of these calculations to higher accretion rates.
556:
557: Following Woosley et al.~(2004), we take the gravitational mass of the
558: neutron star to be $1.4\ M_\odot$ and $R=10\ {\rm km}$, giving a
559: Newtonian gravity $g_{14}=1.9$. The effects of general relativity are
560: not included in the simulations itself, but because the burning layer
561: is very thin the effects of general relativity are small over the
562: extent of the simulated burning layer and a local Newtonian frame is a
563: good approximation. General relativity may be accounted for using
564: appropriate redshift factors, as discussed in \S 4.4 of Woosley et
565: al.~(2004). The results we provide here do not include these redshift corrections;
566: the simulated conditions apply for different combinations of neutron
567: star radius and mass that give the same surface acceleration in the
568: local frame, using appropriate scaling of surface area, accretion
569: rate, and luminosity.
570:
571: Our code includes an adaptive nuclear reaction network that
572: automatically adjusts to include or remove isotopes as needed to
573: follow the details of the nucleosynthesis, out of a reaction rate
574: library of about 5,000 nuclei (Rauscher et al.~2002). The
575: calculations presented here use up to 1300 different isotopes. The
576: reaction rate library includes recent measurements and estimates of
577: critical nuclear reaction rates. The effect of uncertainties in these
578: rates are discussed in detail in Woosley et al.~(2004). The energy
579: generation from the reaction network is directly and consistently
580: coupled into the implicit hydrodynamic solver. The numerical grid
581: adaptively refines and derefines the Lagrangian grid to resolve
582: gradients, but in the hydrogen-rich layer in effect is essentially at
583: constant mass resolution, i.e., linear in column depth. Accretion is
584: modeled by periodically adding an extra zone at the surface of the
585: star (of column depth $\approx 1.6\times 10^6\ {\rm g\ cm^{-2}}$)
586: (Woosley et al.~2004). We follow both the compositional and thermal
587: profiles of the layer, including radiative and convective transport,
588: with a time-dependent mixing length treatment for convection,
589: semiconvection, and thermohaline convection.
590:
591: \begin{figure}
592: \includegraphics[width=\columnwidth]{trans-lc2.ps}
593: \caption{Light curves for different accretion rates (the \textsl{upper
594: right corner} gives accretion rate in units of Eddington accretion
595: rate). Panel~A shows regular bursting with stable recurrence times.
596: Panel~B shows weaker bursts with a partial oscillatory behavior in the
597: tail of the burst light curves. Panel~C shows oscillatory rather
598: behavior and no bursts. Panel~D shows very small oscillations,
599: essentially stable behavior. The small dips seen in all lightcurves
600: on about 15 s time-scale are an artifact of our treatment of
601: accretion, in which an entire small outer zone is added
602: periodically. The beginning of this figure is 10,000 s after the
603: beginning of the simulation to allow the model to reach
604: quasi-stationary conditions.\label{Fig:lc}}
605: \end{figure}
606:
607: Here we present results for an accreted material metallicity of 1/20
608: solar. At an accretion rate $\dot M=0.1\ \dot M_{\rm Edd}$, Woosley et
609: al.~(2004) (their model zM) found a sequence of regular bursts with
610: recurrence times close to 3 hours. Increasing the accretion rate in
611: our new sequence of models, we find a transition from unstable to
612: stable burning at $\dot M=0.924\ \dot M_{\rm Edd}$.
613: Figure~\ref{Fig:lc} shows the behavior close to the transition
614: accretion rate. The lightcurves show a progression from regular
615: periodic bursting at $\dot M=0.7\ \dot M_{\rm Edd}$ (with recurrence
616: times close to $20$ minutes), to a combination of irregular bursts
617: intermixed with oscillations at $\dot M=0.923\ \dot M_{\rm Edd}$, to a
618: regular sequence of oscillations at $\dot M=0.925\ \dot M_{\rm Edd}$,
619: and finally to stable burning at $\dot M=0.95\ \dot M_{\rm Edd}$, with
620: the oscillation amplitude rapidly decreasing as the accretion rate is
621: increased.
622:
623: The oscillation period at $\dot M=0.923\ \dot M_{\rm Edd}$ is $185\pm
624: 5$ seconds. Figure~\ref{Fig:kd} shows a portion of the lightcurve at
625: this accretion rate. The oscillations have an asymmetric profile, with
626: the decay lasting twice as long as the rise. Revnitsev et al.~(2001)
627: noted marginal evidence that the peaks of the mHz QPOs were
628: asymmetric, with a steep rise and shallower decline. A more detailed
629: comparison of our models with observed lightcurves would be
630: interesting, but clearly there should be significant harmonic
631: components. The peak-to-peak amplitude of the oscillation in
632: Figure~\ref{Fig:kd} is $\approx 5\times 10^{36}\ {\rm erg\ s^{-1}}$,
633: with minimum luminosity $\approx 5\times 10^{36}\ {\rm erg\ s^{-1}}$
634: and maximum luminosity $\approx 10^{37}\ {\rm erg\ s^{-1}}$. This is
635: in good agreement with the one-zone model (compare the middle panel of
636: Fig.~\ref{Fig:lc1}; for a 10 km star, a flux of $\approx 10^{24}\ {\rm
637: erg\ cm^{-2}\ s^{-1}}$ corresponds to a luminosity of $\approx
638: 10^{37}\ {\rm erg\ s^{-1}}$).
639:
640:
641:
642:
643: \begin{figure}
644: \noindent
645: \includegraphics[bb=133 253 459 558,clip=true,width=\columnwidth]{qpo-colcoord.ps}
646: \includegraphics[bb=135 325 459 486,clip=true,width=\columnwidth]{qpo-rad.ps}
647: \includegraphics[width=\columnwidth]{qpo-rad-cm.ps}
648: \caption{Detailed light curve (\textsl{upper panel}) and specific
649: nuclear energy generation as a function of time and column depth
650: (\textsl{lower panel}). In the lower thee panels, each darker shading
651: of blue corresponds to a value of energy generation one order of
652: magnitude higher; see scale on right hand side of the figures. In the
653: second figure from the top we label each depth with a Lagrangian
654: column depth. Following a given column depth to the right shows the
655: evolution of that fluid element in time. The sloping black line
656: indicates the surface of the star (the slope gives the accretion
657: rate). The lower two panels show the evolution s a function of radius
658: coordinate. Zero is chosen to correspond to the location where
659: hydrogen is depleted. The upper panel gives specific nuclear energy
660: generation rate (same as the panel above), the bottom panel gives
661: energy generation rate per unit depth.\label{Fig:kd}}
662: \end{figure}
663:
664:
665:
666: \begin{figure}
667: \includegraphics[width=\columnwidth]{trans-phase5.ps}
668: \caption{Snapshots of structure (temperature: \textsl{thick gray
669: line}; density: \textsl{thick gray dashed line}; specific nuclear
670: energy generation: \textsl{black line}) and composition (select
671: isotopes, \textsl{colored lines}) during one oscillation cycle; each
672: panel is advanced in time by $P/4$ where $P$ is the oscillation
673: period; the bottom panel is advanced by one full cycle. The bottom
674: axis for each figure gives column depth, the top axis the
675: corresponding time since accretion began. This is the same model as
676: shown in Fig.~\ref{Fig:lc}, Panel~C. The white and gray stripes
677: correspond to one cycle of oscillation each, with the interfaces
678: corresponding to the time of a maximum in the light curve at the time
679: of the accretion of that layer. The small inserts at upper right
680: corners indicate the position in the light curve (\textsl{red}) cycle
681: of the snapshot (\textsl{black dot}; intentionally aligned with a
682: layer interface). Note that the decreases of some of the radioactive
683: isotopes in the right-hand side of the figure is due to their
684: radioactive decay. An animation of this figure can be found at
685: \texttt{http://xrayburst.org/qpo}.
686: \label{Fig:phase}}
687: \end{figure}
688:
689:
690: \begin{figure}
691: \includegraphics[angle=90,width=\columnwidth]{xrba60x.32000.586.591.ps}
692: \caption{Variation of the composition (summed up for each mass number
693: to be invariant against $\beta^+$ decays) in the ashes layer at a
694: column depth close to $2\times10^8\,$g$\,$cm$^{-2}$ (at the base of
695: the composition profile shown in
696: Fig.~\ref{Fig:phase}).\label{Fig:comp}}
697: \end{figure}
698:
699:
700: The nuclear burning in the oscillation mode is powered by
701: \textsl{$\alpha$p}- and \textsl{rp}-process burning (Wallace \&
702: Woosley 1981; Schatz et al.~1999), beginning with seed nuclei produced
703: by breakout reactions from the CNO cycle, and terminating at mass
704: number $\approx 80$ (the most abundant nucleus is
705: $^{80}$Sr). Figure~\ref{Fig:kd} shows the energy generation as a
706: function of column depth, radius, and time through several oscillation
707: cycles. Figure~\ref{Fig:phase} shows the composition profile of the
708: layer at different phases of the oscillation cycle. Beneath the
709: hydrogen burning layer, the composition of the ashes shows periodic
710: variations with a spacing in column depth of $\dot m P_{\rm osc}$,
711: where $P_{\rm osc}=185$ seconds is the oscillation period. Note,
712: however, that the hydrogen burning depth is relatively constant during
713: the oscillation cycle. The hydrogen burns at a depth $\approx 10^8\
714: {\rm g\ cm^{-2}}$ which is $\approx 7$ times larger than the amount of
715: mass accumulated in one oscillation cycle. The underlying ashes record
716: the variations in burning temperature and the resulting variation of
717: the rp-process ashes during the oscillation cycle. This is
718: reminiscent of the growth of annual rings in a tree trunk.
719: Figure~\ref{Fig:comp} shows the distribution of nuclei in the ashes by
720: mass number, and the slight variation in the composition through the
721: cycle.
722:
723: At the peak in the oscillation light curve, the increased temperature
724: in the burning region drives heat transport inwards, heating the
725: underlying material. This is similar to the substrate heating during
726: Type I X-ray bursts discussed by Woosley et al.\ (2004), and leads to
727: increase of burning of the $^4$He remaining in the ashes layer by the
728: triple-$\alpha$ process and by $\alpha$ captures. In
729: Figure~\ref{Fig:kd} this can be seen as additional spikes in nuclear
730: energy generation below the main burning band of the hydrogen-rich
731: zone. Of particular interest is the amount of carbon remaining in the
732: ashes. Stable burning of accreted H/He has been suggested as the
733: source of carbon fuel for superbursts (in 't Zand et al.~2003; Schatz
734: et al.~2003). Figure~\ref{Fig:comp} shows that the amount of carbon at
735: $y\approx 2\times 10^8\ {\rm g\ cm^{-2}}$ is $\approx 2$\,\% by mass,
736: which is very close to the asymptotic value of slightly below $2$\,\% in
737: the deeper layers where all the helium has been burnt.
738:
739: The layering of different compositions in the ashes does not persist
740: to great depths. The different layers have different values of the
741: number of electrons per baryon $\Ye$, which determines the specific
742: weight of fluid elements under the degenerate conditions in the ashes
743: layer. The variation in composition is stable to the Rayleigh-Taylor
744: instability because of the thermal buoyancy. Secular doubly-diffusive
745: instabilities, however, cannot be suppressed. We find that the
746: thermohaline or salt-finger instability grows in the layers which have
747: an outwardly-decreasing $Y_e$ profile. The subsequent mixing is
748: apparent in Figure~\ref{Fig:phase} as the ``erosion'' of the peaks in
749: the $^{68}$Ge profile and the valleys of the $^{61}$Cu profiles at
750: column depths above 1.5$\times10^{8}\,$g cm$^2$. The mixing eventually
751: leads to homogenization of the ashes layer at depths $>3\times 10^8\
752: {\rm g\ cm^{-2}}$. Our calculations followed that process to a point
753: where the typical column-depth scale of the homogenized regions was
754: several times that of the composition oscillations made by the
755: oscillations in the burning region.
756:
757: Schatz et al.~(1999) calculated the nucleosynthesis in steady-state
758: burning models at $\dot m=1\ \dot m_{\rm Edd}$. They found that the
759: rp-process endpoint terminated at $A\approx 70$, and the carbon mass
760: fraction was $\approx 5$\%. We find slightly heavier rp-process ashes
761: ($A\approx 80$), and a smaller carbon mass fraction ($\approx
762: 2$\%). This is consistent with the general anti-correlation found by Schatz et al.~(2003) between
763: the mass of nuclei produced in the rp-process and the carbon yield. In our case, a longer rp-process gives
764: more time for helium to burn before the hydrogen runs out, leading to
765: less carbon production following hydrogen exhaustion. Our burning
766: temperature is approximately 10--20\% hotter than the models of Schatz
767: et al.~(1999) which might explain the more extensive rp-process. This
768: is perhaps because of a difference in radiative opacities: Schatz et
769: al.~(1999) use a fit to the results of Itoh et al.~(1991) for
770: free-free opacity, whereas the \textsc{Kepler} code uses the fit of
771: Iben (1975) to the radiative opacities of Cox \& Stewart (1970a,b). We
772: will investigate this difference further in future calculations.
773:
774:
775: \section{Discussion}
776:
777: The fact that oscillatory burning is naturally expected at the
778: transition from unstable to stable nuclear burning was pointed out by
779: Paczynski (1981). We have investigated the properties of marginally
780: stable burning in this paper with a simplified one-zone model (\S2)
781: and with detailed multizone simulations (\S3) using the
782: \textsc{Kepler} code, extending the Type I X-ray burst calculations of
783: Woosley et al.~(2004) to higher accretion rates. The period,
784: amplitude, and shape of the oscillations agree well in both
785: models. Remarkably, the basic physics of the oscillations is the same
786: physics underlying a nonlinear relaxation oscillator such as the van
787: der Pol oscillator (e.g., Abarbanel et al.~1993). Usually, the
788: positive or negative damping term dominates, giving rise to the
789: familiar X-ray bursts at low accretion rates or stable burning at a
790: fixed temperature and density at high accretion rates. Close to the
791: marginally stable point, however, the effective thermal timescale is
792: very long, allowing the underlying oscillation period of the system to
793: be seen. This period is close to the geometric mean of the thermal
794: time and accumulation time of the burning layer (eq.~\ref{eq:posc}).
795:
796: \begin{figure}
797: \includegraphics[bb=73 210 556 672,width=\columnwidth]{timedepx.ps}
798: \caption{Oscillation frequency as a function of hydrogen mass
799: fraction, $X_0$, for different surface gravities, corresponding to
800: different radii and masses of the underlying neutron star. The data
801: is taken from one-zone models and rescaled to the oscillation
802: frequency found in the multi-zone model with $X_0=0.7$ and
803: $g_{14}=1.9$. Proper general relativistic redshifts corrections and
804: radii have been applied. The curve for $g_{14}=0.9$ corresponds to
805: $M=1.4\,M_\odot$ (gravitational mass) and $R=15.2\,$km, the curve for
806: $g_{14}=1.48$ corresponds to $M=1.4\,M_\odot$ and $R=11.9\,$km,
807: $g_{14}=1.9$ (used in the multi-zone calculation) corresponds to
808: $M=1.4\,M_\odot$ and $R=12.3\,$km, the curve for $g_{14}=2.45$ is for
809: $M=1.4\,M_\odot$ and $R=10\,$km, and the curve for $g_{14}=3.1$
810: corresponds to $M=2\,M_\odot$ and $R=11\,$km. The \textsl{dotted
811: lines} indicate the boundaries of the observed range of QPO
812: frequencies. \label{Fig:F(X)}}
813: \end{figure}
814:
815: This behavior naturally reproduces two properties of the mHz QPOs
816: observed by Revnivtsev et al.~(2001): the observed periods of $\approx
817: 2$ minutes, and the fact that mHz QPOs were observed in only a narrow
818: range of luminosities in 4U~1608-52, $(0.5$--$1.5)\times 10^{37}\ {\rm
819: erg\ s^{-1}}$. Identification of the mHz QPOs with marginally stable
820: nuclear burning would for the first time relate a feature of the
821: persistent X-ray emission to the neutron star surface in sources which
822: are not X-ray pulsars. Further, our one-zone models indicate that the
823: oscillation period is very sensitive to the surface gravity and
824: accreted hydrogen fraction (Fig.~\ref{Fig:P1}). One
825: of the difficulties in comparing X-ray burst properties with
826: theoretical models is the uncertain relation
827: between X-ray luminosity and accretion rate (e.g., Cumming 2003). This uncertainty is
828: removed for marginally stable burning because it occurs at a specific
829: accretion rate. The dependencies on surface gravity and hydrogen
830: fraction need to be confirmed with multizone models, although unfortunately scanning the parameter space for the location of the transition is computationally very expensive. Our one-zone model
831: results (Fig.~\ref{Fig:P1}) suggest that the observed periods of $\approx 2$ minutes require either $X_0<0.7$, as predicted by intermediate mass evolution
832: models (e.g., Podsiadlowski et al.~2002), or surface gravity
833: $g_{14}\approx 3$, for example corresponding to a $2\ M_\odot$
834: (gravitational mass) star with $R=11\ {\rm km}$. This can be seen in Figure~\ref{Fig:F(X)}, which shows the oscillation period as a function of $X_0$ and $g_{14}$. In this figure, we have rescaled the one-zone model results to match the multizone model for $X_0=0.7$ and $g_{14}=1.9$; in addition, we include the gravitational redshift factor\footnote{Gravitational redshift increases the oscillation periods shown in Figure~\ref{Fig:P1} by $\approx 30$\%. The numerical results in \S3, however, suggest that
835: the one-zone model overpredicts the oscillation period by a similar
836: factor. Therefore, the oscillation periods from the one-zone model without redshifting are in fact approximately those we expect from multizone
837: models corrected for gravitational redshift.}. Future comparisons of theoretical models with
838: mHz QPO periods and lightcurves are potentially sensitive probes of
839: the surface gravity and accreted composition.
840:
841: Two observed features of mHz QPOs remain to be explained. The first
842: is the $Q$ value of the oscillation, which Revnivtsev et al.~2001
843: found to be $Q\equiv\nu/\Delta\nu\approx 3$--$4$. This may be related
844: to the range of accretion rates for which oscillatory nuclear burning
845: can be observed. Revnivtsev et al.~(2001) constrained the range of
846: luminosities at which mHz QPOs are present to be $(0.5$--$1.5)\times
847: 10^{37}\ {\rm erg\ s^{-1}}$ in 4U~1608-52. In contrast, the
848: theoretical range where oscillations are seen is much smaller, within
849: a range $\Delta \dot m/\dot m\approx 0.01$ around the transition
850: accretion rate. Moreover, for much of this range, the oscillations
851: decay in time. Further observations which constrain the range of
852: luminosities for which mHz QPOs can be observed would be valuable.
853:
854: The second puzzle is that our theoretical models, in agreement with
855: previous estimates (Fujimoto, Hanawa, \& Miyaji 1981; Ayasli \& Joss
856: 1982; Bildsten 1998), find that the transition to stable burning
857: occurs at an accretion rate close to the Eddington rate ($\dot
858: M=0.924\ \dot M_{\rm Edd}$ for the model presented in \S 3). In
859: contrast, the X-ray luminosity at which the mHz QPOs are observed in
860: 4U~1608-52 is $\approx 0.5$--$1.5\times 10^{37}\ {\rm erg\ s^{-1}}$,
861: implying an accretion rate ten times lower, $\dot M\approx 0.1\ \dot
862: M_{\rm Edd}$. The same factor of ten appears when comparing the
863: theoretical and observed oscillation amplitudes. The observed
864: amplitudes of mHz QPOs correspond to flux variations of $\approx
865: 1$--$2$\% (Revnivtsev et al.~2001). As noted by Revnivtsev et
866: al.~(2001), this is consistent with the ratio of nuclear energy
867: release from burning the accreted material to gravitational energy
868: release from accretion. We find fractional amplitudes of the expected
869: few percent level in the theoretical models. The absolute
870: peak-to-peak flux variation is a factor of ten larger than observed,
871: however, roughly $5\times 10^{23}\ {\rm erg\ cm^{-2}\ s^{-1}}$ or
872: $5\times 10^{36}\ {\rm erg\ s^{-1}}$. This is more than $30$\% of the
873: observed persistent luminosity. Note that the oscillation amplitude
874: decreases rapidly with increasing accretion rate above the transition
875: accretion rate, so there does exist a small range of accretion rates
876: where the amplitude does match the observed value.
877:
878: A simple explanation for both of these discrepancies is that the {\em
879: local} accretion rate onto the star is close to the Eddington rate,
880: even though the {\em global} accretion rate is much lower. Because the
881: burning layer is very thin, the properties of the burning depend only
882: on the local accretion rate, which may vary across the surface of the
883: star. In particular, if the accreted material spread over only
884: $\approx 10$\% of the surface, the local accretion rate would be ten
885: times larger, comparable to the Eddington rate, and the emitted
886: luminosity would be ten times lower than if the burning covered the
887: entire stellar surface. This would allow marginally stable burning to
888: occur at a global rate of $\approx 0.1$ Eddington, while giving flux
889: variations of a few percent as observed.
890:
891: The physics that might cause confinement of the fuel onto a small
892: fraction of the surface is not obvious. The pressure at the base of
893: the burning layer is $P=gy=10^{22}\ {\rm erg\ cm^{-3}}\ g_{14}y_8$,
894: implying that magnetic fields of strength approaching $\approx
895: \sqrt{8\pi P}\approx 3\times 10^{11}\ {\rm G}$ would be required to
896: confine the fuel. This is much larger than the $\approx 10^8$--$10^9\
897: {\rm G}$ fields assumed for the neutron stars in LMXBs (believed to be
898: the progenitors of the millisecond radio pulsars; Bhattacharya et
899: al.~1995), although small scale fields of these strengths might exist
900: on the surface. The need to transport angular momentum could also
901: potentially delay spreading of material accreted from a disk onto the
902: equator of the star. Inogamov \& Sunyaev (1999) studied this problem
903: with a one-zone model of the spreading layer and a basic prescription
904: for angular momentum transport. They found column depths $<10^4\ {\rm
905: g\ cm^{-2}}$ in the spreading layer, much smaller than the burning
906: depth.
907:
908: The possibility that the covering fraction changes with accretion rate
909: was suggested previously by Bildsten (2000), but in the opposite
910: sense, with covering fraction increasing with $\dot m$. The motivation
911: was to explain a puzzling change in burst behavior that is observed to
912: occur at a luminosity $\sim 10^{37}\ {\rm erg\ s^{-1}}$. EXOSAT
913: observations of several Atoll sources showed that as X-ray luminosity
914: increased, burst properties changed from regular, frequent bursts
915: ($t_{\rm recur}\approx$ hours) with energetics consistent with burning
916: all of the accreted fuel in bursts, to irregular, infrequent ($t_{\rm
917: recur}> 1$ day) bursts whose energetics indicate that only a small
918: fraction of the fuel burns in bursts (van Paradijs et al.~1988). RXTE
919: and BeppoSAX observations confirmed this result, with particularly
920: good coverage for the transient source KS~1731-260 (Muno et al.~2000;
921: Cornelisse et al.~2003). Cornelisse et al.~(2003) found that
922: observations of nine bursters with BeppoSAX were consistent with this
923: pattern of bursting, with a universal transition luminosity
924: $L_X\approx 2\times 10^{37}\ {\rm erg\ s^{-1}}$.
925:
926: This change in bursting behavior is not predicted by the standard
927: theory, in which regular bursting should continue up to the stability
928: boundary at $\dot m\approx\dot m_{\rm Edd}$. Several theoretical
929: explanations for the discrepancy were put forward, including mixing by
930: Rayleigh-Taylor (Wallace \& Woosley 1984) or shear instabilities
931: (Fujimoto et al.~1987) which might allow more rapid burning of
932: hydrogen, a new mode of burning involving slowly propagating fires at
933: $\dot m\gtrsim 0.1\ \dot m_{\rm Edd}$ (Bildsten 1993), or that the
934: covering fraction of accreted material increases at higher accretion
935: rates, lowering the accretion rate per unit area and lengthening the
936: time between bursts, giving hydrogen time to burn stably (Bildsten
937: 2000). We also mention here that Narayan \& Heyl (2003) calculated
938: linear eigenmodes of truncated steady-state burning models, and found
939: stability for accretion rates $\dot M>0.25\ \dot M_{\rm Edd}$, more
940: consistent with observations. The lower accretion rate is likely the
941: cause of the small oscillation frequencies that they found for
942: marginally stable burning (period $\sim 20$ minutes). We find,
943: however, that bursting continues unabated up to $\dot M\approx \dot
944: M_{\rm Edd}$. Further work comparing linear stability analysis with
945: numerical calculations seems to be required.
946:
947: The observations of mHz QPOs at a luminosity close to $\approx
948: 10^{37}\ {\rm erg\ s^{-1}}$ and their interpretation as marginally
949: stable oscillatory burning at a local accretion rate $\dot m\approx
950: \dot m_{\rm Edd}$ provide new input for these ideas. As we have
951: discussed, if changes in the covering fraction are responsible, this
952: suggests that the covering fraction {\em decreases} rather than
953: increases at this luminosity. The observed Type I bursts at
954: $L_X>10^{37}\ {\rm erg\ s^{-1}}$ could be accommodated if there was a
955: slow ``leak'' of fuel away from the stably burning region. This fuel
956: would deplete hydrogen as it accumulated, giving occasional short
957: helium-rich bursts. Other mechanisms, such as stable burning driven by
958: mixing of fuel by shear instabilities might also lead to oscillatory
959: burning. More theoretical work is needed. One clue is that the
960: ultracompact source 4U~1820-30 which most likely accretes pure helium
961: (Bildsten 1995; Cumming 2003) shows a similar transition at a similar
962: luminosity, implying that the nature of the transition does not depend
963: on accreted composition.
964:
965: This question is also likely to be relevant for superbursts and Type I
966: burst oscillations. Superbursts are long duration, rare, and extremely
967: energetic Type I X-ray bursts (up to 1000 times the duration and
968: energy and less frequent than normal bursts) believed to be due to
969: unstable carbon ignition (Cumming \& Bildsten 2001; Strohmayer \&
970: Brown 2002). Superbursts are only observed at luminosities above
971: $\approx 10^{37}\ {\rm erg\ s^{-1}}$, and from sources for which burst
972: energetics indicate that bursts burn only a small fraction of the
973: accreted fuel (in 't Zand et al.~2003). This fits nicely with the
974: theoretical result that stable burning is much more efficient than
975: unstable burning at producing the carbon fuel (Schatz et al. 2003).
976: How to achieve stable burning theoretically at $\dot m<\dot m_{\rm
977: Edd}$, however, has been an open question. Type I burst oscillations
978: are high frequency oscillations during Type I X-ray bursts believed to
979: be due to burning asymmetries on the surface. Burst oscillations are
980: preferentially seen at higher accretion rates, in the banana branch of
981: the color-color diagram (e.g., Muno et al.~2000). Incomplete covering
982: of the surface might help to explain the origin of burst oscillations,
983: facilitating inhomogeneous burning when Type I X-ray bursts are able
984: to occur.
985:
986: Finally, we note that Atoll sources undergo a transition from the
987: island state to banana branch in the color-color diagram at a
988: luminosity of $L_X\approx 10^{37}\ {\rm erg\ s^{-1}}$. It is
989: well-known for these sources that X-ray luminosity does not track
990: accretion rate on short timescales (van der Klis 2001). One
991: explanation for the transition from island state to banana branch in
992: Atolls is that a hot quasi-spherical accretion flow at low rates is
993: replaced by a thin disk at high rates (e.g., Gierlinski \& Done 2002).
994: If this picture is correct, it could be that the change in accretion
995: geometry affects the distribution of fuel on the neutron star
996: surface. An interesting observational question is whether the
997: appearance of the mHz QPOs is linked to a particular luminosity range,
998: or a particular part of the color-color diagram, for example the
999: island to banana transition.
1000:
1001:
1002:
1003: \acknowledgements We thank Michiel van der Klis and Dong Lai for useful
1004: discussions. This research is supported by the DOE Program for
1005: Scientific Discovery through Advanced Computing (SciDAC;
1006: DE-FC02-01ER41176). AH is supported at LANL by DOE contract
1007: W-7405-ENG-36 to the Los Alamos National Laboratory. AC acknowledges
1008: support from McGill University startup funds, an NSERC Discovery
1009: Grant, Le Fonds Qu\'eb\'ecois de la Recherche sur la Nature et les
1010: Technologies, and the Canadian Institute for Advanced Research. SEW
1011: has been supported by the NSF (AST 02-06111) and NASA (NAG5-12036).
1012:
1013:
1014:
1015: \begin{references}
1016:
1017: \noindent
1018: Abarbanel, H.~D.~I., Rabinovich, M.~I., \& Sushchik, M.~M.\ 1993,
1019: ``Introduction to Nonlinear Dynamics for Physicists'' (Singapore:
1020: World Scientific)
1021:
1022: \noindent
1023: Ayasli, S., \& Joss, P.~C.\ 1982, \apj, 256, 637
1024:
1025: \noindent
1026: Bildsten, L.\ 1993, \apjl, 418, L21
1027:
1028: \noindent
1029: Bildsten, L.\ 1995, \apj, 438, 852
1030:
1031: \noindent
1032: Bildsten, L.\ 1998, NATO ASIC Proc.~515: The Many Faces of Neutron
1033: Stars., 419
1034:
1035: \noindent
1036: Bildsten, L.\ 2000, American Institute of Physics Conference Series,
1037: 522, 359
1038:
1039: \noindent
1040: Cornelisse, R., et al.\ 2003, \aap, 405, 1033
1041:
1042: \noindent
1043: Cox, A.~N., \& Stewart, J.~N.\ 1970, \apjs, 19, 243
1044:
1045: \noindent
1046: Cox, A.~N., \& Stewart, J.~N.\ 1970, \apjs, 19, 261
1047:
1048: \noindent
1049: Cumming, A.\ 2003, \apj, 595, 1077
1050:
1051: \noindent
1052: Cumming, A.\ 2004, Nuclear Physics B Proceedings Supplements, 132, 435
1053:
1054: \noindent
1055: Cumming, A., \& Bildsten, L.\ 2001, \apjl, 559, L127
1056:
1057: \noindent
1058: Fujimoto, M.~Y., Hanawa, T., \& Miyaji, S.\ 1981, \apj, 247, 267
1059:
1060: \noindent
1061: Fujimoto, M.~Y., Sztajno, M., Lewin, W.~H.~G., \& van Paradijs, J.\
1062: 1987, \apj, 319, 902
1063:
1064: \noindent
1065: Fushiki, I., \& Lamb, D.~Q.\ 1987, \apj, 317, 368
1066:
1067: \noindent
1068: Gierli{\'n}ski, M., \& Done, C.\ 2002, \mnras, 337, 1373
1069:
1070: \noindent
1071: Iben, I.\ 1975, \apj, 196, 525
1072:
1073: \noindent
1074: in~'t Zand, J.~J.~M., Kuulkers, E., Verbunt, F., Heise, J., \&
1075: Cornelisse, R.\ 2003, \aap, 411, L487
1076:
1077: \noindent
1078: Inogamov, N.~A., \& Sunyaev, R.~A.\ 1999, Astronomy Letters, 25, 269
1079:
1080: \noindent
1081: Lewin, W.~H.~G., van Paradijs, J., \& Taam, R.~E.\ 1993, Space Science Reviews, 62, 223
1082:
1083: \noindent
1084: Lewin, W.~H.~G., van Paradijs, J., \& Taam, R.~E.\ 1995, in X-Ray Binaries,
1085: ed.\ W.~H.~G.~Lewin, J.~van Paradijs, \& E.~P.~J.~van den Heuvel (Cambridge:
1086: CUP), 175
1087:
1088: \noindent
1089: Muno, M.~P., Fox, D.~W., Morgan, E.~H., \& Bildsten, L.\ 2000, \apj,
1090: 542, 1016
1091:
1092: \noindent
1093: Narayan, R., \& Heyl, J.~S.\ 2003, \apj, 599, 419
1094:
1095: \noindent
1096: Paczynski, B.\ 1983, \apj, 264, 282
1097:
1098: \noindent
1099: Podsiadlowski, Ph., Rappaport, S., \& Pfahl, E.~D.\ 2002, \apj, 565, 1107
1100:
1101: \noindent
1102: Rauscher, T., Heger, A., Hoffman, R.~D., Woosley, S.~E.\ 2002, \apj,
1103: 576, 323
1104:
1105: \noindent
1106: Revnivtsev, M., Churazov, E., Gilfanov, M., \& Sunyaev, R.\ 2001, \aap,
1107: 372, 138
1108:
1109: \noindent
1110: Schatz, H., Bildsten, L., Cumming, A., \& Wiescher, M.\ 1999, \apj,
1111: 524, 1014
1112:
1113: \noindent
1114: Schatz, H., Bildsten, L., Cumming, A., \& Ouellette, M.\ 2003,
1115: Nucl. Phys. A, 718, 247
1116:
1117: \noindent
1118: Strohmayer, T.~E., \& Bildsten, L.\ 2003, in ``Compact Stellar X-Ray
1119: Sources'', eds. W.H.G. Lewin and M. van der Klis (Cambridge: Cambridge
1120: University Press) (astro-ph/0301544)
1121:
1122: \noindent
1123: Strohmayer, T.~E., \& Brown, E.~F.\ 2002, \apj, 566, 1045
1124:
1125: \noindent
1126: van der Klis, M.\ 2001, \apj, 561, 943
1127:
1128: \noindent
1129: van der Klis, M.\ 2004, in ``Compact Stellar X-Ray
1130: Sources'', eds. W.H.G. Lewin and M. van der Klis (Cambridge: Cambridge
1131: University Press) (astro-ph/0410551)
1132:
1133: \noindent
1134: van Paradijs, J., Penninx, W., \& Lewin, W.~H.~G.\ 1988, \mnras, 233,
1135: 437
1136:
1137: \noindent
1138: Wallace, R.~K., \& Woosley, S.~E.\ 1981, \apjs, 45, 389
1139:
1140: \noindent
1141: Wallace, R.~K., \& Woosley, S.~E.\ 1984, ``High Energy Transients in
1142: Astrophysics'', 319
1143:
1144: \noindent
1145: Weaver, T.~A., Zimmerman, G.~B., \& Woosley, S.~E.\ 1978, \apj, 225,
1146: 1021
1147:
1148: \noindent
1149: Woosley, S.~E., et al.\ 2004, \apjs, 151, 75
1150:
1151: \noindent
1152: Yu, W., \& van der Klis, M.\ 2002, \apj, 567, L67
1153:
1154: \end{references}
1155:
1156: \end{document}
1157:
1158:
1159:
1160: