1: 

2: %\documentclass[12pt,preprint]{aastex}

3: \documentclass{emulateapj}

4: 

5: % math commands

6: \newcommand{\be}{\begin{eqnarray}}

7: \newcommand{\ee}{\end{eqnarray}}

8: \newcommand{\beq}{\begin{equation}}

9: \newcommand{\eeq}{\end{equation}}

10: \def\simless{\mathbin{\lower 3pt\hbox

11:       {$\rlap{\raise 5pt\hbox{$\char'074$}}\mathchar"7218$}}}

12: \def\simgreat{\mathbin{\lower 3pt\hbox

13:       {$\rlap{\raise 5pt\hbox{$\char'076$}}\mathchar"7218$}}} %> or of order

14: 

15: % variables

16: \renewcommand{\vec}[1]{\mbox{\boldmath $\displaystyle #1$}}

17: \newcommand{\grad}{{\mbox{\boldmath $\nabla$}}}

18: \newcommand{\calO}{ {\cal O} }

19: 

20: % -----------------------------------------------------------

21: % -----------------------------------------------------------

22: 

23: \begin{document}

24: 

25: \title{ Quasi-Periodic Oscillations from Magnetorotational Turbulence}

26: 

27: \author{Phil Arras\altaffilmark{1}, Omer Blaes\altaffilmark{2},

28: Neal J. Turner\altaffilmark{3} }

29: 

30: \altaffiltext{1}{Kavli Institute for Theoretical Physics, Kohn Hall,

31: University of California, Santa Barbara, CA 93106;

32: arras@kitp.ucsb.edu}

33: \altaffiltext{2}

34: {Department of Physics, Broida Hall, University of California, Santa

35: Barbara, CA 93106; blaes@physics.ucsb.edu}

36: \altaffiltext{3}{Jet Propulsion Laboratory, MS 169-506, California

37: Institute of  Technology, Pasadena, CA 91109; neal.turner@nasa.jpl.gov}

38: 

39: \begin{abstract}

40: 

41: Quasi-periodic oscillations (QPOs) in the X-ray lightcurves of accreting

42: neutron star and black hole binaries have been widely interpreted as

43: being due to standing wave modes in accretion disks. These disks are

44: thought to be highly turbulent due to the magnetorotational instability

45: (MRI). We study wave excitation by MRI turbulence in the shearing

46: box geometry. We demonstrate that axisymmetric sound waves and radial

47: epicyclic motions driven by MRI turbulence give rise to narrow, distinct

48: peaks in the temporal power spectrum. Inertial waves, on the other hand,

49: do not give rise to distinct peaks which rise significantly above the

50: continuum noise spectrum set by MRI turbulence, even when the fluid

51: motions are projected onto the eigenfunctions of the modes. This is 

52: a serious problem for QPO models based on inertial waves.

53: 

54: \end{abstract}

55: 

56: \keywords{accretion, accretion disks --- MHD --- turbulence --- waves ---

57: X-rays: binaries} 

58: 

59: \section{ Introduction }

60: 

61: Millisecond variability has been observed for some time with the Rossi

62: X-Ray Timing Explorer in both neutron star and black hole binaries

63: (e.g. McClintock \& Remillard 2004, van der Klis 2004).  In addition

64: to a broad continuum in the power spectrum, QPOs are observed. These

65: QPOs have been modeled as hydrodynamic waves in geometrically thin

66: accretion disks (for reviews see e.g. Wagoner 1999 and Kato 2001),

67: or ``torus-like'' flows with significant radial pressure gradients

68: \citep{2004A&A...427..251G,2003MNRAS.344L..37R, 2004ApJ...603L..89K,

69: 2004ApJ...603L..93L,BAF}.

70: 

71: Turbulence seeded by the magnetorotational instability (MRI;

72: \citealt{bal91}) is widely believed to provide the stresses responsible

73: for angular momentum transport and accretion.  This raises a series

74: of questions. Do waves idealized as hydrodynamic perturbations on

75: a laminar background still exist in turbulent, magnetized accretion

76: flows? Does the time-averaged flow act as a resonant cavity supporting

77: standing waves? What is the steady-state wave amplitude due to turbulent

78: excitation? The simplest and most controlled geometry to explore

79: these questions is that of the shearing box (e.g. \citealt{haw95}).

80: One published attempt has been made to look for discrete mode frequencies

81: in a shearing box MRI simulation \citep{2005AN....326..787B}, with

82: negative results (see figure 7 of that paper). Analytic estimates for 

83: turbulent excitation of waves in accretion disks have been

84: reported in \citet{1993ApJ...418..187N, 1995MNRAS.274...37N}.

85: 

86: We revisit this problem, finding that MRI turbulence does indeed

87: excite distinct peaks in the power spectrum, and that these peaks can

88: be identified with certain classes of hydrodynamic modes, specifically

89: axisymmetric acoustic waves and a global axisymmetric epicyclic oscillation.

90: Hydrodynamic inertial wave modes, however, are {\it not} clearly detected,

91: a serious problem for QPO models that are based on these modes

92: (``$g$-modes'' in diskoseismology parlance).

93: 

94: The plan of the paper is as follows. In section \ref{sec:sbox} we review the

95: shearing box and compute the linear wave frequencies. We present power

96: spectra from numerical simulations in section \ref{sec:power}, and discuss

97: the results.  We present our conclusions and discuss their relevance

98: to models for QPOs in section \ref{sec:conclusions}.

99: 

100: \section{ The shearing box }

101: \label{sec:sbox}

102: 

103: We solve the magnetohydrodynamic (MHD) equations as applied

104: to an isothermal, differentially rotating, magnetized accretion

105: disk, in the unstratified shearing box approximation:

106: \be

107:  \frac{\partial \vec{v}}{\partial t}  +  \vec{v} \cdot \grad \vec{v} &+&

108: 2 \vec{\Omega} \times \vec{v}  = 

109:  - \frac{1}{\rho} \grad \left( P

110:  +   \frac{B^2}{8\pi} \right)  \\

111:  &+&  \frac{\vec{B} \cdot \grad \vec{B} }{4\pi \rho}

112:  +  2q\Omega^2 x \vec{e}_x

113: \\

114: \frac{\partial \rho}{\partial t} & + &  \grad \cdot \left( \rho \vec{v} \right)

115:  =  0

116:  \\

117: \frac{\partial \vec{B} }{\partial t} & = &  \grad \times \left( \vec{v}

118: \times \vec{B} \right)

119: \\

120: P & = & c_s^2 \rho.

121: \ee

122: Here $\rho$, $P$, $\vec{v}$, and $\vec{B}$ are the mass density,

123: gas pressure, velocity and magnetic field. The sound speed $c_s$

124: is constant. The simulation domain represents a small region of the

125: disk near the midplane orbiting with frequency $\Omega$. The

126: non-inertial reference frame of the orbit is taken into account by

127: including a Coriolis force, as well as ``tidal" forces due to the

128: difference of gravitational and centrifugal forces, represented by

129: the term $2q\Omega^2 x \vec{e}_x$, where $q=-d\ln \Omega/d\ln R$

130: is the shear parameter.  Cartesian coordinates are used, with the

131: radial, azimuthal, and vertical coordinates labeled $x$, $y$, $z$,

132: respectively. The azimuthal and vertical boundaries are periodic,

133: and the radial boundaries are shearing-periodic (fluid passing

134: through one radial boundary appears on the other at an azimuth that

135: varies in time according to the difference in orbital speed across

136: the box).

137: 

138: An initially weak magnetic field will drive turbulence in the shearing box

139: due to the MRI, which is a nearly

140: incompressible instability.  A second branch of nearly incompressible

141: perturbations which we call ``inertial waves'' also exists, at least on a

142: laminar background flow.

143: In the limit of zero magnetic field, the specific angular momentum

144: gradient is the restoring force for these inertial waves. To understand the

145: two solutions with a simple example,  

146: linearize about a time-independent background with constant density $\rho$

147: and pressure $P$, constant 

148: magnetic field $\vec{B}=B_y \vec{e}_y + B_z \vec{e}_z$, and velocity 

149: $\vec{v}=-q\Omega x \vec{e}_y$. For incompressible, axisymmetric waves

150: with space-time dependence

151: $\exp(ik_x x + ik_z z - i\omega t)$, the dispersion relation is

152: \citep{bal91}

153: \beq

154: \omega^2  = k_z^2 v_{{\rm A}z}^2 + \frac{1}{2} \kappa^2 \frac{k_z^2}{k^2}

155: \pm \frac{1}{2} \left[ \left(\kappa \frac{k_z}{k} \right)^4

156: + 16 k_z^2 v_{{\rm A}z}^2 \Omega^2 \frac{k_z^2}{k^2} \right]^{1/2},

157: \label{eq:incomp}

158: \eeq

159: where the positive sign gives inertial waves and the negative sign the

160: MRI. Here $v_{{\rm A}z}=B_z/(4\pi \rho)^{1/2}$ is the vertical Alfven speed and

161: $\kappa=[2(2-q)]^{1/2}\Omega$ is the fluid epicyclic frequency.\footnote{

162:   Not to be confused with the epicyclic frequency of a particle in a

163:   gravitational field. }

164: %For this field geometry and $k_y=0$, the MRI instability relies upon having

165: %$k_z \neq 0$. 

166: %For $k_x=0$, maximum occurs for growth rate

167: %$\gamma_{\rm max}=q\Omega/2$ at a wavenumber $k_{\rm z,max}=(q-q^2/4)^{1/2}

168: %\Omega/v_A$.

169: In the zero field limit, incompressible inertial waves

170: have the dispersion relation $\omega^2=\kappa^2 k_z^2/k^2$.

171: 

172: Although inertial modes modified by magnetic tension exist for the

173: constant background chosen above, once the medium becomes turbulent,

174: their possible existence as normal modes of oscillation becomes highly

175: questionable. Low order inertial waves have frequencies of the same

176: order as the energy-bearing MRI ``eddies'', and hence are susceptible to

177: nonlinear interactions with the MRI turbulence.

178: 

179: The simulations described here have zero net magnetic flux in each

180: direction. Furthermore, the time and space-averaged magnetic pressure

181: is small. We attempt to understand waves in such a turbulent background

182: by again linearizing about a time-independent background with constant

183: density and pressure, velocity $\vec{v}=-q\Omega x \vec{e}_y$, but now

184: including compressibility and neglecting magnetic fields.  In this case,

185: axisymmetric inertial-acoustic waves satisfy the dispersion relation

186: \beq

187: \omega^2  =  \frac{1}{2} \left( \kappa^2 + c_s^2 k^2 \right) \pm

188: \frac{1}{2} \left[ \left(\kappa^2+c_s^2 k^2 \right)^2 

189: - 4\kappa^2 c_s^2 k_z^2 \right]^{1/2}.

190: \label{eq:disprel}

191: \eeq

192: For zero wavenumber one recovers the fluid epicyclic mode

193: $\omega^2=\kappa^2$. In the limit $\kappa^2 \ll c_s^2 k^2$, which is

194: a good approximation even for the

195: lowest wavenumbers in our simulations, inertial

196: waves again have the dispersion relation $\omega^2=\kappa^2

197: k_z^2/k^2$, while acoustic waves satisfy $\omega^2=c_s^2 k^2$.

198: In a shearing box of dimensions $(L_x,L_y,L_z)$, these waves will occur

199: at discrete frequencies because the wavenumbers will be quantized according

200: to

201: \beq

202: k_x={2\pi n_x\over L_x}\,\,\,\,\,{\rm and}\,\,\,\,\,k_z={2\pi n_z\over L_z},

203: \eeq

204: where $n_x$ and $n_z$ are integers.

205: 

206: Because of the shearing boundary conditions, non-axisymmetric waves in

207: the shearing box have an amplitude and frequency which change with time

208: \citep{gol65}, and therefore do not exist as normal modes. The radial

209: wavenumber  changes with time as $k_x(t)=k_x'+q\Omega (t-t_0) k_y'$,

210: where $k_x'$ and $k_y'$ are the wavenumbers with respect to comoving

211: coordinates $[x,y'=y+q\Omega x (t-t_0)]$, and $t_0$ is the time of minimum

212: $k_x(t)$. When $k_x(t) \sim k_x'$, non-axisymmetric waves have similar

213: frequencies to axisymmetric waves. For $k_x(t) \gg k_x'$, an approximate

214: WKB dispersion relation for the time-dependent frequency is (see eq. 72 in

215: \citealt{gol65}) $\omega(t)\simeq c_s k_y' q\Omega t$, showing  that the

216: frequency changes on the shear timescale of the disk. As $k_x(t)$ becomes

217: large, eventually the wave will damp away. If the wave has lifetime $T$,

218: then we expect it to give rise to a broad peak in the power spectrum of

219: width $\sim ck_y'q\Omega T$ around the corresponding axisymmetric mode.

220: 

221: \section{ Power spectra } 

222: \label{sec:power}

223: 

224: \begin{figure*}

225: \epsscale{0.9}

226: %\plotone{power.ps}

227: \plotone{f1.eps}

228: \caption{``Position space" power spectrum for density

229:   (upper), radial velocity (middle), and vertical velocity

230:   fluctuations (lower). In each plot, the upper curve is for the

231:   non-axisymmetric data and the lower curve is for the axisymmetric

232:   data. The sharp peak at the orbital frequency in the middle plot is

233:   the box averaged data. The frequency is in units of cycles per

234:   orbital period. Power has been rebinned into logarithmically spaced

235:   frequency bins, which effectively multiplies the power spectrum by

236:   one power of frequency. Axisymmetric hydrodynamic normal mode frequencies

237:   from eq.\ (\ref{eq:disprel}) are shown as arrows.  The epicyclic mode

238:   is at the orbital frequency (unity). Inertial modes and sound waves

239:   should lie below and above the epicyclic frequency, respectively. From

240:   right to left below the epicyclic frequency, arrows show the expected

241:   axisymmetric inertial mode frequencies for $(n_x,n_z)=(1,2), (1,1),

242:   (2,1), (3,1), (4,1)$ and $(5,1)$. From left to right above the epicyclic

243:   frequency, the axisymmetric acoustic waves have $(n_x,n_z)=(1,0),(0,1),

244:   (1,1),(2,0),(2,1),(0,2),(1,2),(3,0),(0,3),(4,0)$, and $(5,0)$,

245:   respectively. } 

246: \label{fig:power}

247: \end{figure*}

248: 

249: Our simulations use box size $(L_x,L_y,L_z)=(1.25,4,1)$ with

250: $(N_x,N_y,N_z)=(40,64,32)$ grid points in each direction, respectively.

251: The shear parameter $q=3/2$, corresponding to Keplerian rotation.

252: The net magnetic flux in each direction is zero and the initial weak

253: field has a spatial dependence given by ${\bf B}=B_0\cos(2\pi x/L_x){\bf

254: e}_y+B_0\sin(2\pi x/L_x){\bf e}_z$. The initial magnetic pressure is 400

255: times less than the gas pressure. The sound speed, orbital frequency,

256: and average density have the constant values $c_s=\Omega=10^{-3}$

257: and $\rho=1$. The fiducial simulation is run for 1000 orbits. Averaged

258: over time and the box, the total magnetic pressure in the turbulence

259: that results is $0.65\%$ of the average gas pressure.  However, there

260: is significant variance, with small regions of local magnetic pressure

261: reaching $10\%$ of the gas pressure.

262: 

263: Figure \ref{fig:power} shows ``position space" power spectra for density,

264: radial velocity, and vertical velocity fluctuations.

265: Three spectra are plotted for each. Data

266: recorded at a specific point $(x,y,z)$ in the box contain fully

267: non-axisymmetric motions. Fluid motions with nodes along the $y$-direction

268: contain most of the power in MRI turbulence, hence we also record time

269: series for azimuthally averaged data at the same $(x,z)$ and box-averaged

270: data in order to increase signal to noise for oscillatory modes.

271: We interpolated the data to uniform spacing in time,  padded the array

272: to a multiple of two, multiplied by the Bartlett window function

273: to reduce spectral leakage \citep{1992nrfa.book.....P}, and used

274: a Discrete Fast Fourier Transform to produce the power spectrum.

275: The power spectra are normalized to the root-mean-square of the time

276: series. To produce figure \ref{fig:power} we binned the power into

277: 512 logarithmically spaced frequency bins in the frequency range

278: shown. This has the effect of multiplying the power spectrum by

279: a factor of $\omega$. The frequency $\omega/2\pi$ is expressed in

280: units of inverse orbital time.  Arrows show the position of low order

281: acoustic and inertial waves using eq.\ (\ref{eq:disprel}). While all

282: three panels show curves for non-axisymmetric and axisymmetric data,

283: the box-averaged power is only visible in the plot for $v_x$, appearing

284: as a sharp peak at $\omega=\kappa=\Omega$, the epicyclic frequency.

285: This epicyclic mode has been reported in previous shearing box simulations

286: \citep{haw95}. Numerical damping causes the sharp decline in power at

287: high frequencies. In runs at higher resolution, this dropoff was pushed

288: to higher frequencies, as expected.

289: 

290: The most striking feature in figure \ref{fig:power} is the axisymmetric

291: sound waves, which appear as sharply defined peaks. The power in

292: these peaks approaches the broad continuum due to non-axisymmetric

293: MRI turbulence, and clearly rises above the continuum for at least two

294: of the peaks. Axisymmetric sound waves with $n_z=0$  have much larger

295: density and radial velocity fluctuations than those with $n_x=0$. This

296: is consistent with the larger continuum seen in $v_x$ relative to

297: $v_z$. A number of previous investigations (e.g. \citealt{sto96,gar05})

298: have noted non-axisymmetric sound waves in snapshots of the density

299: profile. These non-axisymmetric sound waves also have $n_z=0$ and $n_x

300: \neq 0$, consistent with our findings for axisymmetric waves.

301: 

302: There are no significant peaks observed for inertial modes in figure

303: \ref{fig:power}. However, this is not proof of their non-existence.

304: The epicyclic mode was buried in the noise for non-axisymmetric

305: and axisymmetric data, and only became visible by box-averaging. Moreover,

306: the {\it observable} luminosity variation decreases strongly as wavelength

307: decreases, due to averaging of hot and cold spots on the disk. Hence

308: large lengthscale, small power inertial modes may {\it in principle} be

309: observable as compared to short lengthscale, large power MRI eddies. We

310: now show that this is in fact not the case.

311: 

312: \begin{figure}

313: \epsscale{1.2}

314: %\plotone{power_sft_axi.ps}

315: \plotone{f2.eps}

316: \caption{ ``Momentum space" power spectrum for axisymmetric

317:   radial velocity (upper), and vertical velocity (lower) fluctuations.

318:   Each curve is labelled by $(n_x,n_z)$ used for the spatial Fourier transform.

319:   Power has been rebinned into logarithmically spaced

320:   frequency bins, which effectively multiplies the power spectrum by

321:   one power of frequency. Axisymmetric hydrodynamic inertial mode frequencies

322:   from eq.\ (\ref{eq:disprel}) are shown as arrows with $(n_x,n_z)=(0,0),

323:   (1,2), (1,1), (2,1)$ from right to left.}

324: \label{fig:ftpower}

325: \end{figure}

326: 

327: Figure \ref{fig:ftpower} shows the ``momentum space" power spectrum.

328: Time series for azimuthally averaged $v_x$ and $v_z$ were computed by

329: taking spatial transforms in the $x$ and $z$ directions, for $n_x$ and

330: $n_z$ nodes, respectively. Hence, turbulent eddies and inertial modes

331: are on the same footing in this plot, as {\it we compare their power at

332: the same lengthscale}. The temporal power spectrum for the resulting time

333: series was computed as in figure \ref{fig:power} for 128 frequency bins in

334: the range shown. The normalization of the curves is such that the sum over

335: all $n_x$, $n_z$, and frequency bins gives the root-mean-square of the

336: time series (compare to axisymmetric curves in figure \ref{fig:power}).

337: 

338: The epicyclic mode is again clearly visible in figure \ref{fig:ftpower}.

339: The inertial modes are qualitatively different. Two of the lowest

340: order modes show broak peaks of width $\delta \omega \sim \omega$ at

341: the expected frequencies. Without the analytic mode frequencies as

342: a guide, it would be hard to recognize these peaks as significant. 

343: Higher order modes (not plotted here) show even

344: less evidence of peaks at the expected frequencies. The lack of clear

345: peaks for inertial modes suggests that they are strongly affected by MRI

346: turbulent stresses. We note that although the epicyclic mode has frequency

347: similar to inertial modes, it is immune to magnetic tension forces as

348: it is nodeless in the shearing box, and hence is less susceptible to

349: turbulent buffeting.

350: 

351: Overall, we find that axisymmetrizing or box-averaging the data

352: from simulations greatly enhances the signal from oscillation modes,

353: a technique easily adapted to global simulations. In addition, long

354: integration times will increase the signal to noise of the nearly

355: sinusoidal waves relative to the stochastic background from turbulence. It

356: is not immediately obvious, however, why we clearly see peaks in our

357: non-axisymmetric data whereas \cite{2005AN....326..787B} did not. Perhaps

358: this is because his simulation was stratified in the vertical direction.

359: 

360: \section{Conclusions} 

361: \label{sec:conclusions}

362: 

363: We find that turbulence driven by the MRI naturally excites radial

364: epicyclic motion and axisymmetric sound waves, but there is scant

365: evidence of inertial waves. This may imply that inertial modes

366: are strongly affected by MRI turbulent stresses, and cease to exist

367: as high quality oscillators. We note that the zero net magnetic flux

368: simulations presented here give rise to the {\it weakest} MRI turbulence

369: possible. Inertial waves could be even more strongly affected in

370: simulations with net magnetic flux.

371: 

372: Inertial waves, called ``$g$-modes'' in the diskoseismology literature,

373: have been proposed as the origin of some observed QPOs in X-ray binaries

374: \citep{1999PhR...311..259W, 2001PASJ...53....1K}. However, our results

375: imply that all but the longest wavelength of these hydrodynamic modes

376: are destroyed by the turbulence. Even the longest wavelength modes are

377: substantially altered, perhaps to the point of becoming unobservable. This

378: presents a serious problem for models invoking $g$-modes to explain

379: observed QPOs.

380: 

381: We are not clearly able to pick out non-axisymmetric, shearing

382: waves in the analysis done so far. There are broadened peaks in

383: the non-axisymmetric data around the frequencies of axisymmetric

384: waves, but this is not clear confirmation. Furthermore, ``real''

385: accretion disks may in fact admit standing nonaxisymmetric waves

386: with constant pattern speeds. There is substantial global simulation

387: data which demonstrates this fact, at least for hydrodynamic disks

388: (e.g. \citealt{1988ApJ...326..277B}).  The shearing box cannot capture

389: these modes, because the shearing box boundary conditions preclude

390: their existence.

391: 

392: In summary, this paper demonstrates that QPOs can be found in simulations

393: of MRI turbulence, at least in a shearing box geometry.  The well-defined

394: boundary conditions of this geometry help ensure the existence of modes,

395: but it is not clear that such well-defined boundaries will exist in real

396: accretion disks.  Global MRI simulations could in principle determine

397: self-consistently whether wave cavities exist and contain trapped modes.

398: Traveling acoustic waves have certainly been observed in such simulations

399: (e.g.  \citealt{dev03}), but no evidence for modes at discrete frequencies

400: has yet been reported.  We encourage extensions of our analysis techniques

401: to these simulations.

402: 

403: \acknowledgements

404: 

405: We thank Steve Balbus, Shane Davis, Mike Nowak and Aristotle Socrates for useful

406: discussions. This research was supported by the National Science Foundation

407: under grant nos. PHY99-07949 and AST 03-07657.

408: 

409: \begin{thebibliography}{}

410: \bibitem[Balbus \& Hawley(1991)]{bal91} Balbus, S.~A., \& 

411: Hawley, J.~F.\ 1991, \apj, 376, 214 

412: \bibitem[Blaes \& Hawley(1988)]{1988ApJ...326..277B} Blaes, O.~M., \& 

413: Hawley, J.~F.\ 1988, \apj, 326, 277 

414: \bibitem[Blaes et.al. (2006)]{BAF} Blaes, O.~M., Arras, P., \&

415:   Fragile, C., MNRAS, submitted

416: \bibitem[Brandenburg(2005)]{2005AN....326..787B} Brandenburg, A.\ 2005, 

417: Astronomische Nachrichten, 326, 787 

418: \bibitem[De Villiers \& Hawley(2003)]{dev03} De Villiers, J.-P., \& Hawley,

419: J. F. 2003, ApJ, 592, 1060

420: \bibitem[Gardiner \& Stone(2005)]{gar05} Gardiner, T. A., \& Stone, J.

421: M.\ 2005, in Magnetic Fields in the Universe, AIP Conf. Proc., 784, 475

422: \bibitem[Giannios \& Spruit(2004)]{2004A&A...427..251G} Giannios, D., \& 

423: Spruit, H.~C.\ 2004, \aap, 427, 251 

424: \bibitem[Goldreich \& Lynden-Bell(1965)]{gol65} Goldreich, 

425: P., \& Lynden-Bell, D.\ 1965, \mnras, 130, 125  

426: %\bibitem[Goldreich \& Nicholson(1977)]{1977Icar...30..301G} Goldreich, P., 

427: %\& Nicholson, P.~D.\ 1977, Icarus, 30, 301 

428: %\bibitem[Goodman \& Oh(1997)]{1997ApJ...486..403G} Goodman, J., \& Oh, 

429: %S.~P.\ 1997, \apj, 486, 403 

430: \bibitem[Hawley et al.(1995)]{haw95} Hawley, J.~F., Gammie, 

431: C.~F., \& Balbus, S.~A.\ 1995, \apj, 440, 742 

432: \bibitem[Kato(2001)]{2001PASJ...53....1K} Kato, S.\ 2001, \pasj, 53, 1 

433: \bibitem[Klu{\'z}niak et al.(2004)]{2004ApJ...603L..89K} Klu{\'z}niak, W., 

434: Abramowicz, M.~A., Kato, S., Lee, W.~H., \& Stergioulas, N.\ 2004, \apjl, 

435: 603, L89 

436: \bibitem[Lee et al.(2004)]{2004ApJ...603L..93L} Lee, W.~H., Abramowicz, 

437: M.~A., \& Klu{\'z}niak, W.\ 2004, \apjl, 603, L93 

438: \bibitem[McClintock \& Remillard(2004)]{2004astro.ph..6213M} McClintock, 

439: J.~E., \& Remillard, R.~A.\ 2004, in Black Hole Binaries, eds. W. H. G.

440: Lewin \& M. van der Klis (Cambridge: Cambridge Univ. Press), in press

441: \bibitem[Nowak \& Wagoner(1993)]{1993ApJ...418..187N} Nowak, M.~A., \& 

442: Wagoner, R.~V.\ 1993, \apj, 418, 187 

443: \bibitem[Nowak \& Wagoner(1995)]{1995MNRAS.274...37N} Nowak, M.~A., \& 

444: Wagoner, R.~V.\ 1995, \mnras, 274, 37 

445: \bibitem[Press et al.(1992)]{1992nrfa.book.....P} Press, W.~H., Teukolsky, 

446: S.~A., Vetterling, W.~T., \& Flannery, B.~P.\ 1992, (Cambridge: Cambridge

447: Univ.  Press)

448: \bibitem[Rezzolla et al.(2003)]{2003MNRAS.344L..37R} Rezzolla, L., Yoshida, 

449: S., Maccarone, T.~J., \& Zanotti, O.\ 2003, \mnras, 344, L37 

450: \bibitem[Stone et al.(1996)]{sto96} Stone, J. M., Hawley, J. F., Gammie,

451: C. F., \& Balbus, S. A.\ 1996, ApJ, 463, 656

452: \bibitem[van der Klis(2004)]{2004astro.ph.10551V} van der Klis, M.\ 2004, 

453: in Black Hole Binaries, eds. W. H. G. Lewin \& M. van der Klis (Cambridge:

454: Cambridge Univ. Press), in press

455: \bibitem[Wagoner(1999)]{1999PhR...311..259W} Wagoner, R.~V.\ 1999, 

456: \physrep, 311, 259 

457: \end{thebibliography}

458: 

459: \end{document}

460: