1: %\documentstyle[12pt,aasms4]{article}

2: \documentstyle[12pt,aaspp4,flushrt]{article}

3: %\documentstyle[emulateapj]{article}

4: %\documentstyle[]{article}

5: 

6: \slugcomment{Submitted to The Astrophysical Journal}

7: 

8: \catcode`\@=11 % This allows us to modify PLAIN macros.

9: \def\@versim#1#2{\vcenter{\offinterlineskip

10:         \ialign{$\m@th#1\hfil##\hfil$\crcr#2\crcr\sim\crcr } }}

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

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

13: \def\lsim{\mathrel{\mathpalette\@versim<}}

14: \def\gsim{\mathrel{\mathpalette\@versim>}}

15: \def\vp{v_\phi}

16: \def\la{\langle}

17: \def\ra{\rangle}

18: \def\pp{p_\perp}

19: \def\ppar{p_\parallel}

20: \def\kpar{k_\parallel}

21: \def\qp{q_\perp}

22: \def\qpar{q_\parallel}

23: \def\dpp{\delta p_\perp}

24: \def\dppar{\delta p_\parallel}

25: \def\drho{\delta \rho}

26: \def\db{\delta B}

27: \def\om{\tilde \omega}

28: %\def\b{|{\bf B_)}|}

29: 

30: %\input psfig

31: 

32: \begin{document}

33: \title{The Magnetohydrodynamics of Convection-Dominated Accretion

34: Flows} 

35: \author{Ramesh Narayan\altaffilmark{1}} 

36: 

37: \affil{Department of Astrophysical Sciences, Princeton University,

38: Princeton, NJ 08540; narayan@cfa.harvard.edu}

39: 

40: \altaffiltext{1}{Permanent Address: Department of Astronomy,

41: Harvard-Smithsonian Center for Astrophysics, 60 Garden Street,

42: Cambridge, MA 02138}

43: 

44: 

45: \author{Eliot Quataert}

46: 

47: \affil{Department of Astronomy, University of California, 501 Campbell

48: Hall, Berkeley, CA 94720; eliot@astron.berkeley.edu}

49: 

50: \author{Igor V. Igumenshchev}

51: 

52: \affil{Laboratory for Laser Energetics, University of Rochester, 250

53: East River Road, Rochester, NY 14623; iigu@lle.rochester.edu}

54: 

55: \author{Marek A. Abramowicz}

56: 

57: \affil{Department of Astronomy and Astrophysics, G\"oteborg

58: University/Chalmers University of Technology, S-41296, G\"oteborg,

59: Sweden; marek@fy.chalmers.se}

60: 

61: 

62: \medskip

63: 

64: 

65: \begin{abstract}

66: 

67: Radiatively inefficient accretion flows onto black holes are

68: unstable due to both an outwardly decreasing entropy (``convection'')

69: and an outwardly decreasing rotation rate (the ``magnetorotational

70: instability''; MRI).  Using a linear magnetohydrodynamic

71: stability analysis, we show that long-wavelength modes with $\lambda/H

72: \gg \beta^{-1/2}$ are primarily destabilized by the entropy gradient

73: and that such ``convective'' modes transport angular momentum {\it

74: inwards} ($\lambda$ is the wavelength of the mode, $H$ is the scale

75: height of the disk, and $\beta$ is the ratio of the gas pressure to

76: the magnetic pressure).  Moreover, the stability criteria for the

77: convective modes are the standard H{\o}iland criteria of hydrodynamics.

78: By contrast, shorter wavelength modes with $\lambda/H \sim

79: \beta^{-1/2}$ are primarily destabilized by magnetic tension and

80: differential rotation.  These ``MRI'' modes transport angular momentum

81: {\it outwards.}  The convection-dominated accretion flow 

82: (CDAF) model, which has been proposed for radiatively inefficient

83: accretion onto a black hole, posits that inward angular momentum

84: transport and outward energy transport by long-wavelength convective

85: fluctuations are crucial for determining the structure of the

86: accretion flow. Our analysis suggests that 

87: % , as originally proposed,

88: the CDAF model is applicable to a magnetohydrodynamic accretion flow

89: provided the magnetic field saturates at a sufficiently

90: sub-equipartition value ($\beta \gg 1$), so that long-wavelength

91: convective fluctuations with $\lambda/H \gg \beta^{-1/2}$ can fit

92: inside the accretion disk.  Numerical magnetohydrodynamic simulations

93: are required to determine whether such a sub-equipartition field is in

94: fact obtained.

95: 

96: \

97: 

98: \noindent {\it Subject Headings:} accretion --- accretion disks ---

99: black hole physics --- convection --- instabilities ---MHD ---

100: turbulence

101: 

102: \end{abstract}

103: 

104: 

105: \section{Introduction}

106: 

107: Models of radiatively inefficient accretion flows provide a useful

108: framework for interpreting observations of low-luminosity black hole

109: X-ray binaries and active galactic nuclei (see, e.g., Narayan,

110: Mahadevan \& Quataert 1998; Quataert 2001; Narayan 2002 for reviews).

111: In the past few years, there has been rapid theoretical progress in

112: understanding the dynamics of such flows.  Much of this advance has

113: been driven by numerical simulations.  In particular, a number of

114: hydrodynamic simulations have been reported in which angular momentum

115: transport is put in ``by hand'' using an $\alpha$ prescription (e.g.,

116: two-dimensional simulations by Igumenshchev, Chen, \& Abramowicz 1996;

117: Igumenshchev \& Abramowicz 1999, 2000; Stone, Pringle, \& Begelman

118: 1999; and three-dimensional simulations by Igumenshchev, Abramowicz \&

119: Narayan 2000).  Such simulations reveal a flow structure very

120: different from advection-dominated accretion flow models (ADAFs) that

121: were proposed to describe the structure of radiatively inefficient

122: accretion flows (Ichimaru 1977; Narayan \& Yi 1994; Abramowicz et

123: al. 1995).  In an ADAF, the gas accretes rapidly, and the

124: gravitational potential energy of the accreting gas is stored as

125: thermal energy and advected into the central black hole.  By contrast,

126: in the hydrodynamic simulations, the rate of mass accretion is much

127: smaller and strong radial convection efficiently transports energy

128: outwards.  The simulations have been interpreted in terms of a

129: ``convection-dominated accretion flow'' model (CDAF; Narayan,

130: Igumenshchev \& Abramowicz 2000; Quataert \& Gruzinov 2000; Abramowicz

131: et al. 2002).  In a CDAF, convection simultaneously transports energy

132: outwards and angular momentum inwards, strongly suppressing the

133: accretion rate onto the central black hole.

134: 

135: The relevance of the hydrodynamic simulations, and the CDAF model

136: derived from them, is unclear.  This is primarily because of the {\it

137: ad hoc} treatment of angular momentum transport.  It is believed that

138: angular momentum transport in accretion flows is primarily due to MHD

139: turbulence initiated by the magnetorotational instability (MRI; Balbus

140: \& Hawley 1991; BH91).  Balbus \& Hawley have argued that, because of

141: the fundamental role played by magnetic fields, hydrodynamic models

142: typically cannot describe the structure of the accretion flow (or

143: differentially rotating systems more generally; e.g., Balbus \& Hawley

144: 1998; Balbus 2000; 2001).  They have applied this criticism in detail

145: to the CDAF model (Hawley, Balbus, \& Stone 2001; Balbus \& Hawley

146: 2002; hereafter BH02).  

147: In an independent argument, BH02 also suggest

148: that CDAF models violate the second law of thermodynamics.  We do not

149: consider this issue here, but intend to deal with it in a separate

150: paper.

151: 

152: In this paper we discuss the physics of the CDAF model within the

153: framework of MHD.  Apart from clarifying the theoretical underpinnings

154: of the CDAF model, we believe that our analysis also provides a useful

155: illustration of the conditions under which a hydrodynamic model is

156: applicable to an intrinsically magnetohydrodynamic situation; this is

157: important in other astrophysical contexts, e.g., the excitation of

158: density waves at Lindblad resonances in a magnetized accretion flow,

159: or diskoseismology models of QPOs.

160: 

161: 

162: The paper is organized as follows.  In \S2 we discuss the linear

163: stability of a differentially rotating and thermally stratified plasma

164: in the presence of a weak vertical magnetic field.  We identify and

165: explain the difference between ``convective'' modes and ``MRI'' modes

166: and show that the former transport angular momentum inwards while the

167: latter usually transport angular momentum outwards.  Then, in \S2.1 we

168: generalize this result to an arbitrary field orientation and arbitrary

169: stratification.  In \S3 we discuss the implications of our results.

170: In particular, we use the linear analysis to speculate about the

171: conditions under which the saturated nonlinear turbulence in an

172: accretion flow might give rise to a CDAF-like structure. In \S4 we

173: summarize our results.

174: 

175: \section{Convective Modes in a Magnetized Differentially Rotating Plasma}

176: 

177: We consider the linear stability of a differentially rotating and

178: thermally stratified plasma in the presence of a weak magnetic field.

179: We specialize to the case of a vertical magnetic field, $B_z \equiv

180: B$, and modes with purely vertical wavevectors, $k_z \equiv k$.  We

181: also take the rotation rate, pressure, and density to be constant on

182: cylinders, i.e., $\Omega(R)$, $P(R)$, and $\rho(R)$.  We thus employ

183: the same equations as those in BH02 (which are a simplification of the

184: more general analysis in BH91).  Although the problem we analyze is

185: identical to that considered by BH02, we differ % significantly

186: in the

187: interpretation of the results.  In particular, we identify a set of

188: modes that have all the features of growing convective modes in

189: hydrodynamics, including inward transport of angular momentum. 

190: 

191: % DROP  THese modes were not considered to be important by BH02

192: 

193: The thermal stratification of the flow can be described by the

194: Brunt-V{\"a}is{\"a}l{\"a} frequency, \beq

195: N^2=-N_R^2={3\over5\rho}{\partial P\over\partial R} {\partial\ln

196: P\rho^{-5/3}\over\partial R}. \label{brunt} \eeq In the absence of

197: rotation, the system is convectively unstable if $N^2 > 0$.  In a

198: rotating medium, but ignoring magnetic fields, convection is present

199: only if $N^2 > \kappa^2$, where $\kappa^2 = 4 \Omega^2 + d \Omega^2 /d

200: \ln R$ is the epicyclic frequency.\footnote{The requirements for

201: convection in an unmagnetized rotating medium are the H{\o}iland

202: criteria (e.g., Tassoul 1978).  These reduce to the condition $N^2 >

203: \kappa^2$ for the case considered here.  For simplicity, we will refer

204: to this condition as the H{\o}iland criterion in this section.  \S2.1

205: treats the general case.}  Note that $N^2 > \kappa^2$ requires a

206: sound speed comparable to the rotation speed.

207: 

208: For a weakly magnetized medium, BH91 derived the behavior of linear

209: perturbations including the effects of thermal stratification.  For

210: fluid displacements of the form ${\bf \xi} \propto \exp[ikz + \gamma

211: t]$, and for the particular geometry considered here, the growth rate

212: $\gamma$ is given by \beq \gamma^2 = -(k v_A)^2 + {1\over2}\left[N^2 -

213: \kappa^2 \pm \sqrt{(N^2-\kappa^2)^2 + 16\Omega^2(kv_A)^2}\right],

214: \label{growth} \eeq where $v_A = B/\sqrt{4 \pi \rho}$ is the Alfv\'en

215: speed.  For the remainder of this section we focus on the growing mode

216: in equation (\ref{growth}); this has $\gamma^2 > 0$ and corresponds to

217: the positive sign of the square root term.  In the Appendix we present

218: a more general analysis of both growing and oscillatory modes that

219: helps clarify some of the issues discussed in this section.

220: 

221: To understand the physics behind the dispersion relation in equation

222: (\ref{growth}) consider the equations for the radial and azimuthal

223: displacements of fluid elements (BH91): \beq {\partial^2 \xi_R \over

224: \partial t^2} - 2 \Omega {\partial \xi_\phi \over \partial t } =

225: -\left({d \Omega^2 \over d \ln R} + (kv_A)^2\right)\xi_R + N^2 \xi_R ,

226: \label{radial} \eeq \beq {\partial^2 \xi_\phi \over \partial t^2} + 2 

227: \Omega {\partial \xi_R \over \partial t } = -(kv_A)^2

228: \xi_\phi. \label{azimuthal} \eeq Solving equation (\ref{azimuthal})

229: for $\xi_\phi$ we can substitute into equation (\ref{radial}) to find

230: \beq {\partial^2 \xi_R \over \partial t^2} = a_H + a_{M,s} + a_{M,d},

231: \label{radialtot} \eeq 

232: where the three acceleration terms are of the form \beq a_H=(N^2 -

233: \kappa^2)\xi_R, \qquad a_{M,s} = - (kv_A)^2 \xi_R, \qquad a_{M,d} =

234: \left[{4 \Omega^2 k^2 v_A^2 \over \gamma^2 +

235: k^2v_A^2}\right]\xi_R. \label{terms} \eeq Since $\partial^2

236: \xi_R/\partial^2 t = \gamma^2 \xi_R$ it is straightforward to solve

237: equation (\ref{radialtot}) and obtain the dispersion relation in

238: equation (\ref{growth}).

239: 

240: The first acceleration term in equation (\ref{radialtot}), $a_H$, is a

241: pure ``hydrodynamic'' or ``H{\o}iland'' term (hence the subscript $H$)

242: which is present even in the absence of magnetic fields.  This term

243: can be either stabilizing or destabilizing; it is stabilizing when

244: $N^2<\kappa^2$ (the medium is convectively stable by the H{\o}iland

245: criteria) and destabilizing when $N^2>\kappa^2$ (convectively

246: unstable).  The second and third terms in equation (\ref{terms}) owe

247: their existence to the magnetic field (hence subscript $M$).  One of

248: them is always stabilizing (subscript $s$) and the other is always

249: destabilizing (subscript $d$).

250: 

251: Equations (\ref{radial})--(\ref{terms}) have a very clean physical

252: interpretation, as we explain in the rest of this section.  For the

253: simple geometry considered here (vertical field and vertical

254: wavevector), gas and magnetic pressure forces are unimportant.  If

255: $N^2 = 0$, the only forces in the problem are magnetic tension and

256: rotating frame dynamics (e.g. the Coriolis force).  Equations

257: (\ref{radial}) and (\ref{azimuthal}) then describe the behavior of

258: fluid elements that are coupled by the tension of magnetic field

259: lines.  As Balbus \& Hawley (1992, 1998) have discussed, these

260: equations are identical to those describing fluid elements coupled by

261: a spring of spring constant $(kv_A)^2$.  For a stratified medium with

262: $N^2 \ne 0$, the radial acceleration of a fluid element has an

263: additional component due to the radial buoyancy force.  This is the

264: $N^2 \xi_R$ term in equations (\ref{radial}) and (\ref{radialtot}).

265: When $N^2<\kappa^2$, this term does not have a large effect and

266: introduces modest changes to the values of numerical factors.

267: However, when $N^2>\kappa^2$, which corresponds to a convectively

268: unstable system by the H{\o}iland criterion, the acceleration term

269: $a_H$ becomes positive and destabilizing.  This introduces

270: qualitatively new effects.

271: 

272: The angular momentum flux of a linear mode is given by $R \Omega

273: T_{R\phi}$, where $T_{R \phi}$ is the $R-\phi$ component of the fluid

274: stress tensor, \beq T_{R \phi} = \rho(\delta v_R \delta v_\phi -

275: \delta v_{AR} \delta v_{A\phi}), \eeq $\delta {\bf v}$ is the

276: perturbed fluid velocity, and $\delta {\bf v_A} \equiv \delta {\bf

277: B}/\sqrt{4 \pi \rho}$.  Using the eigenfunctions in BH91 we confirm

278: BH02's expression for the stress tensor, namely \beq {T_{R\phi} \over

279: \rho |\delta v_R^2|} = {\Omega \over D \gamma} \left[{k^2v_A^2 \over

280: \gamma^2}\left(\left|{d\ln \Omega \over d \ln R}\right| + 2 \right)-

281: {\kappa^2 \over 2 \Omega^2}\right] \equiv \left({\Omega \over \gamma}

282: \right) t_{R \phi}, \label{stress} \eeq where $D = 1 +

283: k^2v_A^2/\gamma^2$ and $t_{R\phi}$ is a useful dimensionless stress

284: whose sign gives the sign of $T_{R \phi}$.  

285: 

286: %DROP BH02 suggest that since equation (\ref{stress}) does not explicitly

287: %contain $N^2$, the angular momentum transport does not ``know anything

288: %about convection'' and so there cannot be ``convective'' modes as

289: %distinct from ``MRI'' modes.  However, since the growth rate $\gamma$

290: %is a function of $N^2$ (eq. [\ref{growth}]), and since $\gamma$

291: %certainly appears in equation (\ref{stress}), this conclusion is not

292: %at all obvious.  Indeed, we shall see that it is incorrect.

293: 

294: \placefigure{fig1}

295: 

296: We are now in a position to elucidate the physics of the linear

297: instability calculation.  Figure 1a shows a plot of the growth rate

298: $\gamma/\Omega$ as a function of the dimensionless wavevector

299: $kv_A/\Omega$ for 5 values of the Brunt-V{\"a}is{\"a}l{\"a} frequency:

300: $N^2/\Omega^2 = 0, 0.5, 1, 1.5,$ and $2$.  We have assumed a Keplerian

301: rotation profile, for which $\kappa^2 = \Omega^2$.  Thus, the first

302: two choices of $N^2$ correspond to a stable entropy gradient by the

303: H{\o}iland criterion, the third is neutrally stable, and the last two

304: choices correspond to an unstable entropy gradient.  Figure 1b shows

305: the dimensionless stress $t_{R \phi}$ as a function of $kv_A/\Omega$

306: for the same five values of $N^2$.  It is useful to note that the

307: self-similar ADAF model gives $N^2/\Omega^2 = 15(\gamma_{ad} -

308: 1)/4\gamma_{ad}$, where $\gamma_{ad}$ is the adiabatic index of the

309: gas (Narayan \& Yi 1994).\footnote{This expression is not valid if

310: $\gamma_{ad} = 5/3$ since then $\Omega^2 = 0$ in the analytical

311: models.}  For example, if $\gamma_{ad} = 1.5$, $N^2/\Omega^2 = 1.25$.

312: 

313: For $N^2 = 0$, the results reproduce the usual MRI.  The growth rate

314: is very small for $k v_A \ll \Omega$ and peaks when $kv_A$ is $\approx

315: \Omega$, with a peak value of $\gamma = 3 \Omega/4$; the angular

316: momentum flux is always outwards ($T_{R\phi} > 0$).  The instability

317: is clearly triggered by the magnetic field via the destabilizing term

318: $a_{M,d}$, since the other two terms in equation (\ref{radialtot}) are

319: both stabilizing (negative).  The MRI survives with qualitatively the

320: same behavior even for non-zero $N^2$, so long as the H{\o}iland

321: criterion for convective stability ($N^2<\kappa^2$) is satisfied.

322: 

323: Consider, however, a flow with $N^2 > \kappa^2$.  Such a flow is

324: unstable to convection according to hydrodynamics, and is therefore

325: the case of interest for CDAFs.  Figure 1 shows that, for $N^2 >

326: \kappa^2$, modes with $k v_A \sim \Omega$ are not very different from

327: their $N^2 = 0$ counterparts.  The angular momentum flux is still

328: outwards and the growth rate of the mode is comparable (though

329: somewhat larger).  By contrast, long-wavelength modes are very

330: different.  As Figure 1 shows, long-wavelength modes are strongly

331: unstable and they transport angular momentum {\it inwards}.  In fact,

332: as can be seen by inspection, the $k v_A \ll \Omega$ limit of the

333: dispersion relation in equation (\ref{growth}) (with the + root) gives

334: $\gamma^2 = N^2 - \kappa^2$, and the corresponding stress tensor is

335: $T_{R \phi}/(\rho |\delta v_R|^2) = - \kappa^2/[2 \Omega (N^2 -

336: \kappa^2)^{1/2}] < 0$.  The long-wavelength modes are clearly

337: independent of the magnetic field and behave very differently from MRI

338: modes.

339: 

340: The physics of the $k v_A \ll \Omega$ modes is as follows.  In the

341: pure MRI problem, namely $N^2=0$ (or more generally $N^2<\kappa^2$),

342: long-wavelength modes are only weakly unstable because the magnetic

343: tension forces that are central to the MRI are weak; the field lines

344: are hardly bent by a long-wavelength perturbation.  By contrast, when

345: $N^2>\kappa^2$, the fluid has two destabilizing forces, the buoyancy

346: term ($a_H$) due to an unstable entropy gradient and the standard MRI

347: term ($a_{M,d}$).  The buoyancy force that drives convection is

348: independent of the wavelength of the perturbation, whereas the MRI

349: depends on $k$.  For long wavelengths, buoyancy is much more important

350: and controls all the physics of the mode, both the growth rate and the

351: angular momentum transport.  Moreover, we show in the Appendix that

352: the ``convective'' mode driven by buoyancy is the {\it only} unstable

353: long-wavelength mode in MHD for a H{\o}iland-unstable medium ($N^2 >

354: \kappa^2$).  In particular, long-wavelength perturbations for which

355: magnetic tension dominates (referred to as $-$ modes in the Appendix)

356: are stable oscillatory waves rather than unstable modes (see

357: eq. \ref{minus} and Fig. 3). 

358: 

359: \placefigure{fig2}

360: 

361: Figure 2 explicitly compares the relative importance of the two

362: destabilizing terms, the buoyancy term $a_H$ and the magnetic tension

363: term $a_{M,d}$.  Let us identify a mode as an ``MRI'' mode if

364: $a_H\lsim a_{M,d}$ and as a ``convective'' mode if $a_H>a_{M,d}$.

365: With this natural identification, we see that, for $N^2<\kappa^2$, all

366: unstable modes are MRI modes, regardless of the wavelength.  Even when

367: $N^2>\kappa^2$, short-wavelength modes with $kv_A\gsim\Omega$ are

368: still MRI modes.  However, unstable long-wavelength modes with

369: $kv_A\ll\Omega$ are clearly convective modes.  The growth of these

370: modes is due entirely to the unstable buoyancy force; correspondingly,

371: the modes transport angular momentum inwards (Fig. 1b), just as

372: indicated by hydrodynamic studies (e.g., Ryu \& Goodman 1992).  Indeed

373: these modes are virtually indistinguishable from their hydrodynamic

374: counterparts.  {\it All of the properties of {hydrodynamic} convective

375: instabilities therefore carry over to a {magnetohydrodynamic}

376: analysis.}

377: 	

378: % DROP BH02 defined ``convective'' modes as those that are unstable only 

379: %when $N^2 \ne 0$ (i.e., $\gamma > 0$ because $N^2 > 0$).  They showed that

380: %such modes have $ |d\Omega^2/d \ln R| < k^2 v_A^2 < N^2 + |d\Omega^2/d

381: %\ln R|$ and transport angular momentum outwards.  It should now be

382: %clear that this definition misses the relevant ``convective'' modes in

383: %a magnetized medium.

384: 

385: \subsection{The Hydrodynamic Limit}

386: 

387: Magnetic tension should be weak for fluctuations with $k v_A \ll

388: \Omega$ regardless of the details of the magnetic field geometry and

389: thermal stratification.  In this subsection, we show this explicitly

390: by considering the long-wavelength limit of the most general

391: axisymmetric MHD dispersion relation.

392: 

393: Balbus (1995) considered the linear, adiabatic, and axisymmetric

394: stability of a differentially rotating weakly magnetized plasma in the

395: presence of both vertical and radial stratification.  He showed that

396: the dispersion relation is given by \beq \om^4{k^2 \over k_z^2} -

397: \om^2\left[{3 \over 5 \rho} (D P)(D \ln [P \rho^{-5/3}]) + {1 \over

398: R^3} D(R^4\Omega^2)\right] - 4 \Omega^2 ({\bf k \cdot v_A})^2 = 0,

399: \label{disp} \eeq where \beq D = \left({k_R \over k_z}{\partial \over

400: \partial z} - {\partial \over \partial R}\right), \eeq and $\om^2 =

401: -\gamma^2 - ({\bf k} \cdot {\bf v_A})^2$ (for fluctuations $\propto

402: \exp[\gamma t + i {\bf k \cdot r}]$). Unlike in the previous section,

403: $\Omega$, $P$, and $\rho$ are now allowed to be functions of both $z$

404: and $R$, the magnetic field has an arbitrary direction (i.e., $B_R$,

405: $B_\phi$, and $B_z$ components), and the wavevector is given by ${\bf

406: k} = k_R {\bf \hat R} + k_z {\bf \hat z}$.

407: 

408: For ${\bf k \cdot v_A} = 0$ (but $k/k_z$, $k_R/k_z$, etc., finite),

409: equation (\ref{disp}) reduces to \beq \gamma^2 {k^2 \over k_z^2} +

410: \left[{3 \over 5 \rho} (D P)(D \ln [P \rho^{-5/3}]) + {1 \over

411: R^3}D(R^4\Omega^2)\right] = 0. \label{dispk0} \eeq Equation

412: (\ref{dispk0}) is independent of the magnetic field.  Indeed, as noted

413: by Balbus (1995), it describes the linear, adiabatic, axisymmetric,

414: hydrodynamic stability of a differentially rotating and thermally

415: stratified medium (e.g., Goldreich \& Shubert 1967).  As is readily

416: confirmed, equation (\ref{dispk0}) has unstable solutions if and only

417: if the H{\o}iland criteria are satisfied (e.g., Tassoul 1978).

418: 

419: The hydrodynamic limit in equation (\ref{dispk0}) formally requires

420: setting ${\bf k \cdot v_A} = 0$ in the MHD dispersion relation (eq.

421: [\ref{disp}]).  If ${\bf k \cdot v_A} \ne 0$, the full dispersion

422: relation (eq. [\ref{disp}]), which has twice the number of modes

423: (because it is fourth order in $\gamma$ rather than second order),

424: must be used.  However, if the medium is unstable by the H{\o}iland

425: criteria, and if we focus on long-wavelength MHD instabilities with

426: ${\bf k \cdot v_A} \ll \Omega $, then the MHD system will behave like

427: its hydrodynamic counterpart; in particular, the only unstable mode is

428: effectively hydrodynamical and its growth rate is given by equation

429: (\ref{dispk0}) with small corrections $\sim ({\bf k \cdot

430: v_A}/\Omega)^2 \ll 1$.  Note that equation (\ref{dispk0}) describes

431: the unstable growing mode only when the medium is H{\o}iland unstable.

432: If the medium is H{\o}iland stable, the only possible instabilities

433: are intrinsically MHD in nature and equation (\ref{disp}) must be used

434: even for ${\bf k \cdot v_A} \ll \Omega$.  This point is explained in

435: more detail in the Appendix.

436: 

437: The analysis in this section shows that all of the properties of

438: axisymmetric hydrodynamic instabilities carry over to MHD in the

439: long-wavelength limit.  In particular, the results of the previous

440: section are general, and are not an artifact of the simplifying

441: assumptions made there.  Although our quantitative analysis is

442: restricted to axisymmetry, we suspect that long-wavelength

443: non-axisymmetric modes will behave hydrodynamically as well (since

444: they also do not significantly bend the magnetic field lines).

445: 

446: \section{Implications for CDAFs}

447: 

448: To assess the implications of our linear analysis it is useful to

449: consider wavelengths relative to the scale height of the disk, $H

450: \approx c_s/\Omega$, where $c_s$ is the sound speed.  The maximally

451: growing MRI mode has $k H \sim \beta^{1/2}$, where $\beta \approx

452: c^2_s/v^2_A$ is the ratio of the thermal energy to the magnetic energy

453: in the disk.  By contrast, the longer wavelength convective modes have

454: $k H \ll \beta^{1/2}$; more quantitatively, the linear analysis

455: suggests that buoyancy forces become important for $k H \lsim

456: 0.3\beta^{1/2}$ (Fig. 2).  For these modes to be of interest they must

457: fit in the disk and so must have $k H \gsim 1$.  This in turn requires

458: $\beta \gg 1$, i.e., that the magnetic field must be sub-equipartition

459: ($\beta \gsim 10$ may suffice).

460: 

461: Given the above analysis, we propose the following identification

462: between the hydrodynamic CDAF model and the present MHD results.  If

463: the magnetic field saturates at a sub-equipartition value, the most

464: unstable MRI mode operates on a scale $\sim \beta ^{-1/2} H \ll H$ and

465: transports angular momentum outwards. Moreover, if the medium is

466: H{\o}iland unstable, {\it only} such short-wavelength fluctuations

467: transport angular momentum outwards.  Longer wavelength convective

468: fluctuations are dominated by scales $\sim H$ and transport angular

469: momentum inwards (Fig. 1).

470: 

471: Which fluctuations are more important?  Or, more precisely, which part

472: of the power spectrum of fluctuations is more important, those with $k

473: H \sim 1$ or those with $k H \sim \beta^{-1/2}$?  This cannot really

474: be addressed analytically because we do not understand the saturation

475: of the linear instabilities (and it is clear that they are coupled;

476: e.g., convection, by stirring the fluid, can help to build up the

477: magnetic field).  Nonetheless, it is typically the case that the

478: energy in a turbulent plasma is dominated by the largest unstable

479: scales in the medium.  This suggests that, if $\beta \gg 1$ (perhaps

480: $\beta \gsim 10$ is sufficient), the larger-scale convective

481: fluctuations will dominate the dynamics, as proposed in the

482: hydrodynamic models.  Moreover, the inward angular momentum transport

483: by convective modes depends on the magnitude of $N^2 - \kappa^2$

484: (Fig. 1b).  Thus there is a natural way for the inward convective

485: transport to self-adjust to counteract outward transport by small

486: scale MRI fluctuations.  In the limit of efficient convection and

487: $\beta \gg 1$ we would expect marginal stability to convection

488: according to the H{\o}iland criteria, just as in the hydrodynamic models

489: (Quataert \& Gruzinov 2000).

490: 

491: It is important to stress several points: 

492: 

493: (1) Although we believe that our linear stability analysis sheds

494: important light on the physics of radiatively inefficient accretion

495: flows, it does not directly address the fully developed nonlinear

496: turbulence that is of primary importance to the accretion flow

497: structure.  This requires numerical simulations.

498: 

499: (2) The picture proposed here is very similar to the hydrodynamic

500: models of CDAFs; the primary change is that we have theoretical

501: support from a magnetohydrodynamic analysis.

502: 

503: (3) Balbus (2001) has shown that magnetized dilute plasmas can be

504: convectively unstable in the presence of an outwardly decreasing

505: temperature, rather than just an outwardly decreasing entropy.  It

506: would be interesting to incorporate this into future work.

507: 

508: (4) The extent to which one can usefully distinguish scales in the

509: power spectrum that are convective from those that are

510: magnetorotational depends on $\beta$, and becomes more and more useful

511: if $\beta \gg 1$.  It would be straightforward if the convective and

512: MRI modes occupied distinct and well-separated regions of wavevector

513: space.  Then even the nonlinear physics of the instabilities could be

514: distinct.  To cite an example from plasma physics, heat transport in

515: fusion devices is due to instabilities on both proton and electron

516: Larmor radii scales and one can often distinguish the physics of each

517: mechanism separately even in the nonlinear regime (e.g., Dorland et

518: al. 2000; Rogers, Dorland \& Kotschenreuther 2000).  As Figure 1

519: shows, such a separation in wavevector space is not present for our

520: problem.  On the other hand, Figs. 1 and 2 do show quite clearly that

521: the physics of the convective modes is very different from that of the

522: MRI modes.

523: 

524: (5) For a H{\o}iland-unstable medium ($N^2>\kappa^2$), there is a nice

525: continuity between the hydrodynamic limit and the strong field MHD

526: limit.  In wavevector space, the field-affected modes (the MRI modes)

527: are restricted to larger values of $k$ where magnetic tension is

528: important.  For very weak fields, these modes are at very large $k$,

529: and most of the long-wavelength modes that probably dominate the

530: dynamics are convective.  As the field strength increases, however,

531: the MRI modes expand downwards in $k$ space and become more and more

532: important.  Another way to look at this is as follows.  When the fluid

533: is unstable according to the H{\o}iland criterion, the switch that

534: determines whether or not convection is important is not the presence

535: or absence of a magnetic field, but rather the magnitude of the

536: dimensionless parameter $kv_A/\Omega\approx k H/\beta^{1/2}$.  When

537: $kv_A/\Omega$ is small, convection dominates the physics, whereas when

538: it is $\gsim1$, the magnetic field dominates.

539: 

540: \section{Conclusion}

541: 

542: In this paper we have used linear stability theory to assess the

543: conditions under which a hydrodynamic analysis is appropriate for an

544: intrinsically magnetohydrodynamic accretion flow.  We have shown that,

545: when the medium is unstable by the H{\o}iland criteria,

546: long-wavelength instabilities with $k v_A \ll \Omega$ are effectively

547: hydrodynamical.  Physically, this is because magnetic tension is

548: unimportant for sufficiently long-wavelength perturbations.  Formally,

549: the dispersion relation for a magnetohydrodynamic plasma reduces in

550: this limit to that of hydrodynamics (\S2). This shows that all of the

551: properties of hydrodynamic instabilities carry over to an MHD analysis

552: in the long-wavelength limit.

553: 

554: For the particular case of radiatively inefficient accretion flows

555: that are unstable to both convection and the MRI, the relevant

556: parameter that distinguishes convective behavior from

557: magnetorotational behavior is the quantity $kv_A/\Omega\approx

558: H/(\lambda\beta^{1/2})$.  Long-wavelength modes for which $\lambda/H

559: \gg \beta^{-1/2}$ are convective, while modes for which $\lambda/H

560: \sim \beta^{-1/2}$ are magnetorotational.  The convective modes are

561: primarily driven by buoyancy, not magnetic tension (Fig. 2).

562: Moreover, they transport angular momentum {\it inwards} rather than

563: outwards (Fig. 1b), just as indicated by hydrodynamic studies.  For

564: the long-wavelength modes to be of interest the accretion flow must

565: have $\beta \gg 1$ so that the convective fluctuations fit inside the

566: disk.  Under these conditions we believe that the hydrodynamic CDAF

567: models can accurately describe the structure of the accretion flow.

568: The precise $\beta$ that allows a hydrodynamical analysis is uncertain

569: and can be determined only by magnetohydrodynamic simulations (the

570: linear analysis suggests that $\beta \gsim 10$ may suffice).

571: 

572: It is interesting to note that the saturation $\beta$ of the MRI does

573: not appear to be universal.  Given the plausible dependence on $\beta$

574: highlighted above, this may have important implications for the

575: structure of radiatively inefficient accretion flows.

576: Three-dimensional MHD simulations find that that if the field is

577: initially toroidal a radiatively inefficient accretion flow develops

578: into a CDAF-like state (Igumenshchev, Abramowicz \& Narayan 2002),

579: while if the field has a vertical component the structure is very

580: different (Hawley, Balbus \& Stone 2001; Hawley \& Balbus 2002; see

581: also Stone \& Pringle's 2001 2D simulations).

582: 

583: To conclude, it is important to emphasize that the primary conclusion

584: of the analytic arguments in this paper is not that CDAFs {\it

585: actually do} describe the structure of radiatively inefficient

586: accretion flows, but rather that they {\it could}.  In particular, our

587: linear MHD analysis supports all of the key CDAF assumptions so long

588: as $\beta \gg 1$.

589: 

590: 

591: \acknowledgements

592: 

593: RN, IVI, and EQ were supported in part by NSF grant AST-9820686, RFBR

594: grant 00-02-16135, and NASA grant NAG 5-12043, respectively.

595: 

596: 

597: %\newpage

598: 

599: \begin{thebibliography}{999}

600: 

601: \bibitem[]{} Abramowicz, M. A., Chen, X.-M., Kato, S., Lasota, J.-P., 

602: \& Regev, O. 1995, \apj, 438, L37

603: \bibitem{} Abramowicz, M. A., Igumenshchev, I. V., Quataert, E., \& Narayan, 

604: R. 2002, ApJ, 565, 1101

605: \bibitem[Balbus(1995)]{1995ApJ...453..380B} Balbus, S.~A.\ 1995, \apj, 453, 380

606: \bibitem[Balbus(2000)]{2000ApJ...534..420B} Balbus, S.~A.\ 2000, \apj, 534, 420

607: \bibitem[Balbus(2001)]{2001ApJ...562..909B} Balbus, S.~A.\ 2001, \apj, 562, 909

608: \bibitem[Balbus \& Hawley(1991)]{1991ApJ...376..214B} Balbus, S.~A.,~\& 

609: Hawley, J.~F.\ 1991, \apj, 376, 214 (BH91)

610: \bibitem[Balbus \& Hawley(1992)]{1992ApJ...392..662B} Balbus, S.~A.,~\& 

611: Hawley, J.~F.\ 1992, \apj, 392, 662

612: \bibitem[Balbus \& Hawley (1998)]{bh97} Balbus, S.A., \& Hawley, J.F. 1998, 

613: Rev. Mod. Phys., 70, 1

614: \bibitem{} Balbus, S. A., \& Hawley, J. F., 2002, \apj, submitted, 

615: astro-ph/0201428 (BH02)

616: %\bibitem[]{} Begelman, M. C., \& Meier, D. L. 1982, \apj, 253, 873

617: %\bibitem[]{} Chen, X., Abramowicz, M., \& Lasota, J.-P., 1997, ApJ, 476, 61

618: \bibitem[]{} Dorland, W., Jenko, F., Kotschenreuther, M., \& Rogers, B. N.

619: 2000, Phys. Rev. Lett., 85, 5579

620: \bibitem{} Goldreich, P., \& Shubert, G., 1967, ApJ, 150, 571

621: \bibitem{} Hawley, J. F. \& Balbus, S. A., 2002, ApJ in press (astro-ph/0203309)

622: \bibitem[Hawley, Balbus, \& Stone(2001)]{2001ApJ...554L..49H} Hawley, J.~F., 

623: Balbus, S.~A., \& Stone, J.~M.\ 2001, \apjl, 554, L49

624: \bibitem[]{} Ichimaru, S. 1977, \apj, 214, 840

625: %\bibitem[]{} Igumenshchev, I. V. 2000, \mnras, 314, 54

626: \bibitem[]{} Igumenshchev, I. V., \& Abramowicz, M. A. 1999, \mnras, 303, 309

627: \bibitem[]{} Igumenshchev, I. V., \& Abramowicz, M. A. 2000, \apjs, 130, 463

628: \bibitem[]{} Igumenshchev, I. V., Chen, X., \& Abramowicz, M. A. 1996, \mnras, 

629: 278, 236

630: \bibitem[]{} Igumenshchev, I. V., Abramowicz, M. A., \& Narayan, R. 2000,

631: \apj, 537, L27

632: \bibitem[]{} Igumenshchev, I. V., Abramowicz, M. A., \& Narayan, R. 2002,

633: in preparation

634: \bibitem[]{} Narayan, R. 2002, in {\it Lighthouses of the Universe},

635: ed. M. Gilfanov, \& R. Sunyaev (Heidelberg: Springer-Verlag), in press,

636: astro-ph/0201260

637: \bibitem[]{} Narayan, R., Igumenshchev, I. V., \& Abramowicz, M. A.

638:      2000, \apj, 539, 798

639: \bibitem[]{} Narayan, R., Mahadevan, R., \& Quataert, E. 1998, in {\it

640: Theory of Black Hole Accretion Discs}, ed. M. A. Abramowicz,

641: G. Bjornsson, \& J. E. Pringle (Cambridge: Cambridge Univ. Press), p148

642: \bibitem[]{} Narayan, R., \& Yi, I. 1994, \apj, 428, L13

643: \bibitem[]{} Quataert, E. 2001, in {\it Probing the Physics of AGN

644: with Multiwavelength Monitoring}, ed. B. M. Peterson, R. S. Polidan, \&

645: R. W. Pogge (San Francisco: Astr. Soc. Pacific), p71q

646: \bibitem[]{} Quataert, E., \& Gruzinov, A. 2000, \apj, 539, 809

647: \bibitem[]{} Rogers, B. N., Dorland, W., \& Kotschenreuther, M. 2000,

648: Phys. Rev. Lett., 85, 5536

649: \bibitem[Ryu \& Goodman(1992)]{1992ApJ...388..438R} Ryu, D.,~\& Goodman, 

650: J.\ 1992, \apj, 388, 438

651: \bibitem[]{} Stone, J. M., Pringle, J. E., \& Begelman, M. C. 1999, \mnras, 

652: 310, 1002

653: \bibitem{} Stone, J. \& Pringle, J. E., 2001, MNRAS, 322, 461

654: \bibitem{} Tassoul, J.-L., 1978, {\it Theory of Rotating Stars} 

655: (Princeton: Princeton University Press)

656: 

657: \end{thebibliography}

658: 

659: \clearpage

660: \begin{appendix}

661: \section{\bf Appendix}

662: \bigskip

663: In the main text we focused exclusively on unstable modes and

664: therefore considered only the positive sign of the root in equation

665: (\ref{growth}).  Here we generalize the discussion and consider both

666: unstable and stable modes.  To this end, we rewrite equation

667: (\ref{growth}) as follows \beq \gamma^2 = -(k v_A)^2 +

668: {(N^2-\kappa^2)\over2}\left[1 \pm \sqrt{1 +

669: 16\Omega^2(kv_A)^2/(N^2-\kappa^2)^2}\right].\label{disppm} \eeq The +

670: and $-$ signs in equation (\ref{disppm}) map directly onto the + and

671: $-$ signs in equation (\ref{growth}) when $N^2-\kappa^2>0$, but the

672: mapping is reversed when $N^2-\kappa^2<0$.  Equation (\ref{disppm})

673: is, of course, completely equivalent to equation (\ref{growth}), but

674: for this Appendix we find it more convenient to work with the $\pm$

675: convention in equation (\ref{disppm}) because of its behavior when

676: $N^2 - \kappa^2$ changes sign.  In what follows, we refer to the +

677: sign in equation (\ref{disppm}) as the ``+ mode'' and the $-$ sign as

678: the ``$-$ mode.''

679: 

680: Let us focus on the long-wavelength limit $kv_A/\Omega \ll 1$.  The

681: growth rates of the + and $-$ modes are then given by \beq \gamma_+^2

682: \approx (N^2-\kappa^2) + \left( {4\Omega^2\over N^2-\kappa^2} -1

683: \right )(kv_A)^2, \label{plus} \eeq \beq \gamma_-^2 \approx - \left(

684: {4\Omega^2\over N^2-\kappa^2} + 1 \right )(kv_A)^2.

685: \label{minus}

686: \eeq 

687: 

688: In the limit $kv_A/\Omega \ll 1$, the growth rate of the + mode is the

689: same as in hydrodynamics: $\gamma_+^2=N^2-\kappa^2$.  If

690: $N^2>\kappa^2$ (H{\o}iland unstable), $\gamma_+^2$ is positive and the

691: mode is unstable, while if $N^2<\kappa^2$, $\gamma_+^2$ is negative

692: and the mode is oscillatory in nature.  In either case, the mode is

693: completely insensitive to the magnetic field.

694: 

695: The $-$ mode, on the other hand, has a growth rate that vanishes as

696: $kv_A\to0$, which is a clear indication that this mode is strongly

697: influenced by the magnetic field.  Indeed, the mode owes its very

698: existence to the presence of the magnetic field.  The $-$ mode is

699: associated with the classic MRI of BH91.

700: 

701: An inspection of equations (\ref{plus}) and (\ref{minus}) reveals the

702: following interesting result: assuming that $|N^2-\kappa^2| <

703: 4\Omega^2$ (which is almost always the case), only one of the two

704: modes can be unstable.  When the medium is convectively stable ($N^2 <

705: \kappa^2$), the + mode (which we argued above is hydrodynamic in

706: nature) is obviously stable.  In this case, the $-$ mode is unstable

707: and gives the long-wavelength limit of the MRI.  However, when the

708: medium is convectively unstable ($N^2 > \kappa^2$), then it is the +

709: mode that grows, while the $-$ mode is stable.  Thus, in the long

710: wavelength limit, the MRI actually disappears if the medium is

711: unstable by the H{\o}iland criterion.  Fortunately, this is precisely

712: when the medium is convectively unstable so there are always unstable

713: modes available, regardless of the sign of $N^2-\kappa^2$.

714: 

715: \placefigure{fig3}

716: 

717: Figure (\ref{fig3}) shows $\gamma^2$ as a function of $(kv_A)^2$ for

718: the + and $-$ modes for representative values of $N^2/\Omega^2$.  We

719: see that modes that are stable ($\gamma^2<0$) in the limit $kv_A\to0$

720: remain stable for all values of $k$, while modes that are unstable at

721: small values of $kv_A$ reach a maximum growth rate for $kv_A \sim

722: \Omega$ and then become stable with increasing $kv_A$.

723: 

724: \end{appendix}

725: 

726: \clearpage

727: 

728: \begin{figure}

729: \plotone{f1.eps}

730: \caption{(a) Dimensionless growth rate $\gamma/\Omega$

731: as a function of dimensionless wavevector $kv_A/\Omega$ in a Keplerian

732: system ($\kappa^2=\Omega^2$), for five choices of the Brunt-V{\"a}is{\"a}l{\"a}

733: frequency: $N^2/\Omega^2 =$ 0 (solid line), 0.5 (dotted line), 1

734: (short-dashed line), 1.5 (long-dashed line), 2 (dot-dashed line).  The

735: system is unstable by the H{\o}iland criterion for $N^2/\Omega^2=$ 1.5

736: and 2.  For these two cases, the growth rate remains finite in the

737: long-wavelength limit $kv_A/\Omega\ll1$ and the modes are

738: indistinguishable from the corresponding unstable hydrodynamical

739: modes.  (b) Dimensionless shear stress corresponding to the same modes

740: as in (a).  For $N^2/\Omega^2=$ 1.5, 2, and $kv_A/\Omega\ll1$, the

741: stress is negative, implying that these convective modes transfer

742: angular momentum inwards, just as in hydrodynamics.

743: \label{fig1}}

744: \end{figure}

745: 

746: \begin{figure}

747: \plotone{f2.eps}

748: \caption{Variation of $a_H/a_{M,d}$ versus

749: dimensionless wavevector $kv_A/\Omega$ for $N^2/\Omega^2=$ 0 (solid

750: line), 0.5 (dotted line), 1.5 (long-dashed line), 2 (dot-dashed line).

751: The ratio $a_H/a_{M,d}$ measures the importance of the hydrodynamic

752: buoyancy force relative to the leading destabilizing force due to

753: magnetic fields (see eqs. [\ref{radialtot}], [\ref{terms}]).  For the two

754: cases in which the system is convectively unstable according to the

755: H{\o}iland criterion, $viz., N^2/\Omega^2=$ 1.5, 2, the buoyancy force

756: clearly dominates at long wavelengths, confirming that these modes are

757: primarily convective, not magnetorotational.

758: \label{fig2}}

759: \end{figure}

760: 

761: \begin{figure}

762: \plotone{f3.eps}

763: \caption{The dimensionless growth rate $\gamma^2/\Omega^2$ versus

764: dimensionless wavevector $kv_A/\Omega$ for $N^2/\Omega^2=0, ~0.5,

765: ~1.5, ~2$.  A mode with a positive value of $\gamma^2$ is unstable,

766: while one with a negative value of $\gamma^2$ is stable (oscillatory).

767: The solid and dashed lines correspond to the + and $-$ modes,

768: respectively, in equation (\ref{disppm}).  Each curve is labeled by

769: the value of $N^2/\Omega^2$.

770: \label{fig3}}

771: \end{figure}

772: 

773: \end{document}

774: 

775: