1: 

2: \documentstyle[preprint,psfig]{aastex}

3: %\documentstyle[manuscript]{aastex}

4: 

5: \def\lesssim{\mathrel{\hbox{\rlap{\hbox{\lower4pt\hbox{$\sim$}}}\hbox{$<$}}}}

6: \def\gtrsim{\mathrel{\hbox{\rlap{\hbox{\lower4pt\hbox{$\sim$}}}\hbox{$>$}}}}

7: 

8: 

9: \begin{document}

10: 

11: \title{Global MHD Simulation of the Inner Accretion Disk in a 

12: Pseudo-Newtonian Potential}

13: 

14: \author{John F. Hawley}

15: \affil{Department of Astronomy, University of Virginia,

16: Charlottesville VA 22903}

17: 

18: \author{Julian H. Krolik}

19: \affil{Physics and Astronomy Department, Johns Hopkins University,

20:     Baltimore, MD 21218}

21: 

22: \shorttitle{Global MHD Simulation}

23: 

24: \begin{abstract}

25: 

26: We present a detailed three dimensional magnetohydrodynamic (MHD)

27: simulation describing the inner region of a disk accreting onto a black

28: hole.   To avoid the technical complications of general relativity, the

29: dynamics are treated in Newtonian fashion using the pseudo-Newtonian

30: Pacz\'ynski-Wiita potential.  The disk evolves due to angular momentum

31: transport that is produced naturally from MHD turbulence generated by

32: the magnetorotational instability.  We find that the resulting stress

33: is continuous across the marginally stable orbit, in contradiction

34: with the widely-held assumption that the stress should go to

35: zero there.  As a consequence,

36: the specific angular momentum of the matter accreted into the hole is

37: smaller than the specific angular momentum at the marginally stable

38: orbit.  The disk exhibits large fluctuations in almost every quantity,

39: both spatially and temporally.  In particular, the ratio of stress to

40: pressure (the local analog of the Shakura-Sunyaev $\alpha$ parameter)

41: exhibits both systematic gradients and large fluctuations; from $\sim

42: 10^{-2}$ in the disk midplane at large radius, it rises to $\sim 10$

43: both at a few gas density scale heights above the plane at large radius,

44: and near the midplane well inside the plunging region.   Driven in part

45: by large-amplitude waves excited near the marginally stable orbit, both

46: the mass accretion rate and the integrated stress exhibit large

47: fluctuations whose Fourier power spectra are smooth ``red" power-laws

48: stretching over several orders of magnitude in timescale.

49: 

50: \end{abstract}

51: 

52: \keywords{accretion, accretion disks, MHD, black holes}

53: 

54: \section{Introduction}

55: 

56: For many years, the physical nature of angular momentum transport

57: in accretion disks has been one of the great outstanding

58: questions of high-energy astrophysics.  As a result of nearly a decade

59: of effort, it is now becoming increasingly clear that the mechanism is

60: turbulent magnetic stress in which the energy for the magnetic field is

61: largely drawn from the free energy of orbital shear (as reviewed by

62: Balbus \& Hawley 1998).  Theoretical grounds for this conclusion were

63: first found in the rediscovery of the magneto-rotational instability

64: (MRI) by Balbus \& Hawley (1991).  Increasingly detailed simulations of

65: local ``shearing boxes'' have bolstered our confidence that this is

66: indeed the process found in nature (e.g., Hawley, Gammie \& Balbus

67: 1995, 1996; Brandenburg et al. 1995; Stone et al. 1996; 

68: Miller \& Stone 2000).

69: 

70: What has been missing until recently is studies of the radial structure

71: of disks accreting under the influence of magnetohydrodynamic (MHD)

72: turbulent stresses.  Hitherto, only a few such three-dimensional global

73: simulations have been presented (Armitage 1998; Matsumoto 1999; Hawley

74: 2000; Machida, Hayashi, \& Matsumoto 2000).  

75: In global simulations it is difficult to obtain adequate

76: resolution over the wide range of length scales involved, namely the

77: MHD turbulent length scales within the disk, the disk scale height $H$,

78: and the radial distance from the central object $R$.  These length

79: scales are generally quite disparate, but near the inner boundary of an

80: accretion disk they should become comparable.  In this paper we present

81: a global simulation of an accretion disk in which we focus on this

82: inner region where adequate numerical resolution can be obtained.

83: 

84: With the results of this new simulation, we can begin to answer a

85: number of questions of central interest to the mechanics of accretion.

86: For example, the most common view of the inner portions of black hole

87: accretion disks is based on the work of Novikov \& Thorne (1973) (e.g.

88: Abramowicz \& Zurek 1981; Muchotrzeb \& Paczy\'nski 1982; Matsumoto et

89: al. 1984).  In this picture, the disk is (nearly) time-steady and

90: axisymmetric.  Page \& Thorne (1974) argued on heuristic grounds that

91: the $R$-$\phi$ component of the stress (the one responsible for angular

92: momentum transport) should go to zero at the marginally stable orbit;

93: later work by Abramowicz and Kato (1989) showed that if the stress is a

94: (constant) small fraction of the local pressure, it automatically

95: approaches zero near that location.

96: 

97: However, because the MHD turbulence is the result of a general, local,

98: rapidly-growing instability, assumptions such as stationarity and

99: axisymmetry are likely inappropriate.  One might also question whether

100: there is any reason for the stress to diminish at the disk's inner

101: edge, given that the growth rate of the MRI does not appreciably

102: diminish near the marginally stable orbit (cf. Krolik 1999; Gammie 1999;

103: see also the footnote in Page \& Thorne 1974 in which they speculate

104: that their heuristic arguments might fail if strong magnetic fields were

105: present).  Further, because magnetic

106: stresses do not automatically maintain a fixed ratio to the pressure,

107: conclusions based upon $\alpha$ parameterizations (e.g., Abramowicz \&

108: Kato 1989) may not directly apply.  This issue is important because

109: continued stress at the marginally stable orbit could alter both the

110: energy and angular momentum with which matter arrives at the black

111: hole; that is, this boundary condition determines the efficiency of

112: accretion and the rate at which the black hole is spun up.

113: 

114: Because the greatest luminosity should be released near the inner edge

115: of the accretion disk, the detailed behavior and structure of the

116: region near the marginally stable orbit is crucial to

117: understanding the observations of black hole systems.  For example, the

118: luminosity of essentially all accreting black holes exhibits sizable

119: fluctuations with a very broad-band distribution of fluctuation power

120: with timescale (e.g., as discussed by Sunyaev \& Revnivtsev 2000); can we

121: identify the specific dynamics that drive these variations?

122: Occasionally, the lightcurve exhibits

123: quasi-periodic oscillations (QPOs); can we either identify the

124: mechanism responsible for these, or test some of those mechanisms that

125: have been proposed?

126: 

127: In the following section we will present a technical description of the

128: three-dimensional global simulation we report.  We will then set out

129: our results in \S 3, with their qualitative implications outlined in \S

130: 4.  In \S 5 we will discuss the degree to which limitations of our

131: simulation hamper direct application of our results, and suggest in

132: which directions the greatest improvement might soon be made.

133: We summarize our conclusions in \S 6.

134: 

135: 

136: \section{Numerical Method}

137: 

138: \subsection{Equations}

139: 

140: An investigation of the inner edge of a disk orbiting a black hole

141: would be best carried out with a fully general relativistic

142: simulation.  However, we have not yet written a three dimensional

143: general relativistic MHD code.  In the present study we instead make

144: use of the existing three dimensional Newtonian MHD code (Hawley 2000).

145: The code evolves the equations of ideal MHD, i.e.,

146: \begin{equation}\label{mass}

147: {\partial\rho\over \partial t} + \nabla\cdot (\rho {\bf v}) =  0

148: \end{equation}

149: \begin{equation}\label{mom}

150: \rho {\partial{\bf v} \over \partial t}

151: + (\rho {\bf v}\cdot\nabla){\bf v} = -\nabla\left(

152: P + {\mathcal Q} +{B^2\over 8 \pi} \right)-\rho \nabla \Phi +

153: \left( {{\bf B}\over 4\pi}\cdot \nabla\right){\bf B}

154: \end{equation}

155: \begin{equation}\label{ene}

156: {\partial\rho\epsilon\over \partial t} + \nabla\cdot (\rho\epsilon

157: {\bf v}) = -(P+{\mathcal Q}) \nabla \cdot {\bf v}

158: \end{equation}

159: \begin{equation} \label{ind}

160: {\partial{\bf B}\over \partial t} =

161: \nabla\times\left( {\bf v} \times {\bf B} \right)

162: \end{equation}

163: where $\rho$ is the mass density, $\epsilon$ is the specific internal

164: energy, ${\bf v}$ is the fluid velocity, $P$ is the pressure, $\Phi$ is

165: the gravitational potential, ${\bf B}$ is the magnetic field vector,

166: %%%NEW

167: and ${\mathcal Q}$ is an explicit artificial viscosity of

168: the form described by Stone \& Norman (1992a).

169: %%%

170: The global disk code is written in cylindrical coordinates, $(R,\phi,z)$.  

171: To model important effects associated with the relativistic Schwarzschild

172: metric, we employ the pseudo-Newtonian potential of Paczy\'nski \& Wiita

173: (1980). The pseudo-Newtonian potential has the form

174: \begin{equation}\label{pwp}

175: \Phi = - {G M \over r-r_g}

176: \end{equation}

177: where $r$ is spherical radius,

178: and $r_g \equiv 2GM/c^2$ is the ``gravitational radius,''

179: akin to the black hole horizon.   For this potential,

180: the Keplerian specific angular momentum (i.e., that

181: corresponding to a circular orbit) is

182: \begin{equation}\label{pwl}

183: \ell_{kep} = (GMr)^{1/2} {r \over r-r_g} ,

184: \end{equation}

185: and the angular frequency $\Omega = \ell/R^2$.  The innermost marginally

186: stable circular orbit is located at $r_{ms}=3r_g$. 

187: 

188: We use an adiabatic equation of state, $P=\rho\epsilon(\Gamma

189: -1) = K\rho^\Gamma$, with $\Gamma = 5/3$, and ignore radiation

190: transport and losses.  Since there is no explicit resistivity or

191: physical viscosity, the gas can heat only through adiabatic

192: compression or by the action of the artificial viscosity

193: which acts in shocks.

194: Equations (\ref{mass})-(\ref{ind}) are solved using time-explicit

195: Eulerian finite differencing.  The numerical algorithm is that employed

196: by the ZEUS code for hydrodynamics (Stone \& Norman 1992a) and MHD

197: (Stone \& Norman 1992b; Hawley \& Stone 1995).  We adopt units where

198: $GM=1$ and $r_g = 1$ (so that $c = \sqrt{2}$).

199: 

200: 

201: In this paper we repeat---with significantly better

202: resolution---simulation GT4 from Hawley (2000), whose initial state was

203: a thick torus with an angular velocity distribution $\Omega \propto

204: R^{-1.68}$, slightly steeper than Keplerian.  The angular momentum

205: within the torus is equal to the Keplerian value at $R=10$, which

206: determines the location of the pressure maximum in the torus.  In

207: somewhat arbitrary units, the pressure at this point is equal to 0.036

208: and the density $\rho_{max}=34$.  The initial magnetic field is

209: constructed by setting the toroidal component of the vector potential

210: %%%NEW

211: $A_\phi (R,z) = \rho (R,z) -

212: \rho_{min}$, for all $\rho$ greater than a minimum value, $\rho_{min} =

213: 0.1$.  Poloidal field components are then constructed from ${\bf B} =

214: \nabla \times {\bf A}$.  This procedure produces large-scale poloidal

215: field lines that follow the equal-density contours inside the torus.

216: The magnetic field is initially zero everywhere $\rho \leq \rho_{min}$.

217: As discussed in

218: Hawley (2000), this configuration is motivated by the desire to have

219: well-resolved unstable modes of the poloidal field instability, and to

220: have the initial field fully contained within the torus.  The magnetic

221: field level is renormalized so that the total magnetic energy in the

222: torus is 0.01 the total integrated gas pressure.  This corresponds to an

223: overall value of $\beta=P_{\rm gas}/P_{\rm mag} = 100$.

224: %%%%

225: 

226: %NEW

227:   The aim of this initial condition is not so much to

228: follow the evolution of this particular torus, as it is to construct a

229: disk that is nearly Keplerian with an inner boundary that is initially

230: close to, but still outside, the marginally stable orbit.  To that end,

231: the torus is perturbed with low-amplitude random adiabatic pressure

232: fluctuations.  These seed the magneto-rotational instability whose

233: rapid evolution leads to the desired state.  Further disk evolution

234: results in inflow through $r_{ms}$;

235: %

236: it is this resulting accretion flow that will be studied in detail.

237: 

238: The computational grid has $128\times 128 \times 128$ zones.  This is the

239: same number of zones as used in GT4 of Hawley (2000), but here,

240: to increase the resolution within the disk itself and in the inflow

241: region,  we locate more of the zones near the marginally stable orbit

242: and around the equator.  The center of the coordinate system is

243: excised, i.e., the radial coordinate begins at a nonzero

244: $R_{\rm min} =1.5$; this choice avoids both the coordinate singularity

245: associated with the axis and the singularity at $r_g$.  The outer

246: radius is set at $R=31.5$.  The $z$ coordinate is centered on the

247: equatorial plane, and runs from -10 to $+10$.  The angle $\phi$ is

248: periodic and covers the full $2\pi$.  In the radial direction, there

249: are 30 equally spaced zones in $R$ from 1.5 to 4; the radial zones are

250: then graded logarithmically from 4 to the outer boundary at 31.5.  In

251: the vertical direction, half of the $z$ zones are equally spaced around

252: the equator from $-2$ to 2, then graded from those locations out to the

253: vertical boundaries at $z = \pm 10$.  The grid cells are equally spaced

254: in $\phi$.  The simulation GT4 from Hawley (2000) used equally spaced

255: zones in $R$ and $z$, and the outer radial boundary was set at 21.5.

256: We have also run this initial torus at lower resolution for comparison,

257: using $64^3$ grid zones.

258: 

259: The boundary conditions on the grid are simple zero-gradient outflow

260: conditions; no flow into the computational domain is permitted.  The

261: magnetic field boundary condition is set by requiring the transverse

262: components of the field to be zero outside the computational domain,

263: while the perpendicular component satisfies the divergence-free

264: constraint.  Although this produces a nonzero stress at the boundary,

265: it works well in preventing an artificial buildup of field at the

266: boundary.  Most of the time, for $|z| \leq 1$ along the inner radial

267: boundary and $R \leq 2$ in the equatorial plane, the poloidal velocity

268: is greater than the magnetosonic speed, so that the influence of the

269: boundary condition on the flow upstream is limited.

270: 

271: Timescales of interest in the simulation are set by the circular

272: orbital period $P_{orb} = 2\pi \Omega^{-1} = 2\pi r^{3/2} (r-r_g)/r$ at

273: significant points within the disk.  In units where $GM=r_g=1$,

274: $P_{orb}$ at the pressure maximum ($R=10$) is $179$, and at the

275: marginally stable orbit, where the orbital frequency $\Omega =

276: 1/(2\sqrt 3)$, the orbital period is 21.8.  The local accretion

277: timescale is set by the effective stress, namely $t_{acc} = R/\langle

278: v_R \rangle \sim \ell/W_{R\phi} ( \gg P_{orb})$, where $W_{R\phi}$ is

279: the $R$--$\phi$ component of the stress, $W_{R\phi} = \langle \delta

280: v_R \, \delta v_\phi - {v_{AR} v_{A\phi} \rangle}$, and $v_A$ is an

281: Alfv\'en velocity.  The goal is to evolve long enough to observe a

282: number of accretion times near the marginally stable orbit.  Since the

283: code is time-explicit and therefore Courant-limited, this is difficult

284: to achieve over large radial extents.  Here we ran the simulation out

285: to time $t=1500$, which is 69 orbits at $r_{ms}$.

286: 

287: \subsection{Diagnostics}

288: 

289: The quantity of data associated with a three dimensional simulation is

290: daunting, and this makes analyzing the outcome a challenge.  Diagnostics

291: can be computed during the simulation and in post-processing.  The best 

292: procedures for this are still being developed.  Here we briefly describe

293: some of the diagnostic procedures used in the present simulation.

294: 

295: The time-evolution of the disk can be visualized through animation

296: sequences.  Animation frames of the $\phi = 0$ slice and the equatorial

297: ($z=0$) slice are dumped every 4 units in time.  Full 3D density

298: information is saved every 10 units of time, and these are used to make

299: an animation of a three dimensional volumetric rendering.

300: 

301: The time-evolution can also be studied in a manageable way through

302: space-time $(R,t)$ studies of azimuthally- and vertically-integrated quantities.

303: Examples include the averaged mass density 

304: \begin{equation}\label{sigma}

305: \langle \rho\rangle = {\int \rho R d\phi dz \over \int R d\phi dz},

306: \end{equation}

307: the average angular momentum,

308: \begin{equation}\label{angm}

309: \langle \rho \ell \rangle  = {\int \rho \ell R d\phi dz \over \int R

310: d\phi dz},

311: \end{equation}

312: and average specific angular momentum,

313: \begin{equation}\label{el}

314: \langle \ell \rangle= {\langle \rho \ell \rangle \over \langle\rho\rangle }.

315: \end{equation} 

316: The net radial mass flux is

317: \begin{equation}\label{mdot}

318: \langle \dot M\rangle = \langle -\rho v_R \rangle = \int -\rho v_R R d\phi dz,

319: \end{equation}

320: here defined to be positive for net accretion (inflow).

321: Similarly, the angular momentum flux,

322: \begin{equation}\label{lflux}

323: \langle \rho v_R \ell \rangle = \int \rho v_R \ell R d\phi dz,

324: \end{equation}

325: can be computed.

326: 

327: Radial angular momentum transport in the disk is due to

328: the $R$--$\phi$ component of the stress,

329: \begin{equation}\label{stress}

330: T^{R\phi} = \rho \delta v_R \, \delta v_\phi - {B_R B_\phi \over 4\pi},

331: \end{equation}

332: where $\delta v_R$ and $\delta v_\phi$ are the {\it turbulent} portions

333: of the velocity, that is, the departures from the mean flow.  The

334: kinematic portion is the Reynolds stress and the magnetic portion is

335: the Maxwell stress.  In the simulation we have no way of

336: knowing what the mean flow is.   For example, the typical radial

337: velocity is much larger than the time-averaged net radial drift

338: velocity, and there are significant differences between the

339: instantaneous angular velocity and the value associated with a circular

340: orbit (which one might assume should be the mean orbital flow).  As a

341: result, there is no  unique prescription to compute the Reynolds

342: component of the total stress.  Hawley (2000) used a definition for the

343: perturbed orbital velocity $\delta v_\phi$ in terms of the difference

344: between the total instantaneous angular momentum flux, and the mass

345: flux times the average specific angular momentum.  The difference then

346: represents the excess or deficit angular momentum transport due to

347: orbital velocity fluctuations compared to the mean.  Specifically,

348: \begin{equation}\label{rstress}

349: \langle \rho \delta v_R \delta v_\phi \rangle =

350: \langle \rho v_R v_\phi \rangle

351: - \langle \rho v_R \rangle \langle \ell \rangle / R .

352: \end{equation}

353: 

354: In contrast, we can easily compute the stress due solely to magnetic

355: forces, the second term on the right-hand-side of equation

356: (\ref{stress}); we denote this quantity by $M^{R\phi}$.  The

357: vertically- and azimuthally-averaged Maxwell stress is

358: \begin{equation}\label{maxwell}

359: \langle M^{R\phi}\rangle = {\int (-B_R B_\phi / 4\pi) R d\phi dz

360: \over \int Rd\phi dz} .

361: \end{equation}

362: %%%NEW

363: We do not distinguish the total magnetic field from the fluctuating field

364: because, as we shall show later, where the field is dynamically important,

365: the magnitude of the fluctuating portion is large compared to any time-averaged

366: mean field.

367: %%%%

368: 

369: In addition to these vertically- and azimuthally-averaged quantities,

370: we examine full data dumps of the simulation at specific moments in

371: time.  From these we can examine not only the full three dimensional

372: structure, but also azimuthally-averaged and vertically-averaged

373: slices in $(R,z)$ and $(R,\phi)$.

374: 

375: \section{Results}

376: 

377: \subsection{Quasi-stationary state}

378: 

379: The initial torus was in hydrodynamical equilibrium, but it evolves

380: rapidly as the simulation proceeds.  The overall evolution is very

381: similar to that of GT4 in Hawley (2000; see Fig. 13).

382: The radial field is sheared by the differential rotation, creating

383: toroidal field.  Turbulence develops within the disk with the onset of

384: the magnetorotational instability.  The resulting Maxwell stresses

385: drive angular momentum transfer; the disk begins to accrete into the

386: central hole and expands radially outward.  Relatively rapidly the

387: inner portion of the disk achieves an approximate steady-state.  By $t

388: \simeq 700$, the mass accretion rate reaches roughly the long-time

389: mean; after $t \simeq 1000$, the shape of the disk's average radial

390: density distribution $\langle \rho\rangle (R)$ no longer changes

391: appreciably.  Thus, we observe the disk in a quasi-stationary state

392: for several dozen orbital periods at the marginally stable orbit.

393: In much of what follows, we will present a detailed analysis of the

394: state of the disk during this time frame, focusing particularly on

395: the end of the simulation, using that ``snapshot" as a fairly typical

396: sample of conditions in the quasi-stationary disk.

397: However, it must be borne in mind that there are always

398: significant fluctuations around all these mean states.

399: 

400: \subsection{Gas density}

401: 

402: A contour plot of the azimuthally-averaged gas density at the initial

403: and final times is presented in Figure~1.  As might be expected, the

404: density diminishes inside the marginally stable orbit and outside the

405: initial outer radial boundary of the disk, while also decreasing with

406: increasing altitude away from the equatorial plane.  Although the

407: vertical heights of the highest contour levels at the final time are

408: roughly independent of radius within the disk proper, the

409: azimuthally-averaged exponential scale-height of the pressure, $H$,

410: increases very nearly linearly with radius from $R \simeq 2$ out to $R

411: \simeq 20$; at the marginally stable orbit ($R = 3$), $H \simeq 0.4$,

412: while at $R = 10$, $H \simeq 1.5$.  This particular disk shape is

413: almost certainly a consequence of the equation of state, in which all

414: fluid elements retain the same specific entropy except for shock

415: heating.

416: %NEW

417: As a result of this special choice of equation of state,

418: the temperature rises inward as the gas is compressed; by coincidence,

419: this occurs at such a rate as to keep the ratio of sound speed to

420: orbital speed, and therefore $H/R$, almost constant at $\simeq 0.15$.

421: %

422: In a real disk, one in which turbulent dissipation introduces

423: heat and radiation vents it, the shape might be rather different.

424: 

425: Density gradients in radius and altitude are determined primarily

426: by the overall dynamical effects of approximate hydrostatic equilibrium.

427: No such requirements exist for azimuthal fluctuations, whose

428: amplitudes characterize the turbulence.  We

429: quantify the azimuthal fluctuation level by defining

430: \begin{equation}\label{fluctuation}

431: {\delta \rho \over \rho}\left(R,z\right) = 

432: {1 \over \langle \rho\rangle_\phi}  \left\{{1 \over 2\pi} \int \, d\phi \,

433:              \left[\rho - \langle \rho \rangle_\phi \right]^2 \right\}^{1/2} ,

434: \end{equation}

435: where ${\langle X \rangle}_\phi$ refers to the azimuthal average at

436: fixed $R$ and $z$ of the quantity $X$.  Measured in this way,

437: $\delta\rho/\rho \simeq 0.2$ -- 0.6 over most of the problem volume,

438: with smaller values typically nearer the equatorial plane.  In a few

439: places, $\delta \rho/\rho$ rises to be greater than unity.  Thus, quite

440: sizable density fluctuations are generic.

441: 

442:     These fluctuations are primarily sheared strips whose radial extent is

443: $\lesssim 0.1 R$, but extend $\sim 1$~radian in azimuthal angle (Figure~2). 

444: This pattern (tightly wrapped spiral features) is characteristic of

445: fluctuations in almost all quantities.  Part of their nature is

446: revealed by constructing spacetime diagrams in which the surface

447: density at a specific radius is plotted as a function of $t$ and $\phi$.

448: From this visualization we are able to measure the angular speed of

449: these features; it is generally quite close to the local orbital frequency

450: $\Omega$.  We can also measure their persistence at any particular radius;

451: it is usually less than an orbital period.  In this sense, they can be

452: regarded as fluctuations that merely ``ride" with the local orbital motion,

453: but see further discussion in \S 3.5.

454: 

455: \subsection{Magnetic field}

456: 

457: The magnetic field distribution is rather more complicated than the

458: density distribution.  A poloidal projection (i.e., averaged

459: azimuthally) shows the magnetic pressure to have its largest value

460: %%%NEW

461: immediately

462: %%%%

463: outside the dense regions of the torus, with large variations inside

464: the torus (Figure~3).  In the shearing-box simulations of Miller \&

465: Stone (2000), the field strength likewise peaked away from the midplane,

466: but we find somewhat greater contrast.  Field-strength

467: fluctuations are also larger than

468: those for density: the {\it rms} fractional fluctuations in

469: %%%NEW

470: $|\vec B|$ in

471: %%%

472: the azimuthal direction are $\simeq 1$ -- 1.5 in the disk body, falling to

473: $\simeq 0.25$ several gas scale-heights immediately above and below the

474: disk.

475: %%%NEW

476: Another measure of the magnitude of the fluctuations is

477: $|\langle B_i \rangle_\phi |/\langle |B_i | \rangle_\phi$, the ratio of the

478: mean value of an individual field component to the mean of its absolute

479: value.  This quantity is typically $\sim 0.2$ for all components

480: wherever the gas density

481: is significant, indicating that the field is extremely variable as a

482: function of azimuth (and therefore as a function of time at fixed azimuth,

483: as orbital rotation carries the field around).

484: Only where the gas density is very low does the field

485: become relatively smooth and steady

486: %NEW

487: (a result also qualitatively similar to the findings of Miller

488: \& Stone 2000).

489: %

490: %%%

491: Like the density, the characteristic shape of fluctuations is

492: sheared arcs that run for $\sim 1$~radian before losing coherence.

493: However, the peaks in magnetic field strength tend to lie adjacent to,

494: rather than coincident with, peaks in gas density.

495: 

496: %%%NEW

497:      To illustrate some of the field structure, we show in Figure~4

498: poloidal slices taken at $\phi = \pi/2$.  Panel a) is a close-up of the inner,

499: high-resolution region; panel b) shows (almost) the entire disk.

500: This view, of course, suppresses the toroidal component (which is

501: typically larger than the poloidal component by a factor $\sim 1$ -- 10);

502: in addition, in order to make the figures visually comprehensible, we have

503: omitted small-scale field variations (smaller than 0.125 length

504: units in panel a), smaller than 0.5 length units in panel b).  Even

505: with this smoothing, it is apparent that the field is very tangled.  The

506: only field property exhibiting any stability is the roughly dipolar

507: pattern seen in the low density regions.  This is a remnant of the

508: initial condition.  Because it only persists in these very low density

509: regions, and they do not contribute much to the stress (see \S 3.6),

510: we do not regard this remnant artifact as dynamically significant.

511: %%%

512: 

513: As one expects from magnetic fields in shearing plasma,

514: $B_R$ and $B_\phi$ are correlated in the sense that the mean ratio of

515: the Maxwell stress to the magnetic pressure, $\langle -2 B_R B_\phi

516: /B^2 \rangle > 0$ and has (approximately) constant magnitude $\simeq

517: 0.2$ -- 0.3 in the body of the disk.  Inside the

518: marginally stable orbit, this ratio increases slowly in magnitude as

519: radial accretion ``combs'' the field out into radial loops that are

520: strongly sheared, increasing the correlation between $B_R$ and

521: $B_\phi$.  In fact, the Maxwell stress is so correlated with the

522: average magnetic pressure $\langle B^2/8\pi\rangle$, that a figure

523: showing the stress is almost indistinguishable from one showing the

524: magnetic pressure.

525: 

526: \subsection{Accretion rate}

527: 

528: Perhaps the single most basic quantity of interest is the accretion

529: rate, $\langle \dot M\rangle$, defined in equation (\ref{mdot}).

530: The disk is never perfectly time-steady; $\rho v_R$ varies

531: substantially from place to place and from time to time.  In fact, disk

532: quantities tend to be quite nonaxisymmetric, and much of the final

533: accretion inside $r_{ms}$ takes place in a spiral flow.  However,

534: time-averaging can remove much of this variability, and

535: Figure 5 shows the time-averaged accretion rate as a function of $R$

536: along with the instantaneous values of $\dot M (R)$ at $t= 1000$

537: and 1500.  The time-averaged $\dot M(R)$ is nearly constant as a

538: function of radius inside $R=10$, with $\dot M = 5$.  We shall adopt

539: this as the fiducial steady state accretion rate for the inner disk.

540: Between $R=9$ and 14 there is inflow at a decreasing rate; outside

541: $R=14$ the net flux is outward.  In this simulation, the initial disk

542: is entirely contained within the grid, and the outer part of the disk

543: must move outward in order to ``soak up" the angular momentum received

544: from the inner disk material.  Over the course of the simulation the

545: total disk mass decreases by 14\%; about 90\% of this has gone into the

546: black hole.

547: 

548: Figure~6 shows a detailed time-history of the accretion rate through

549: the inner radial boundary.  As can be seen from this figure, the

550: accretion rate shows significant time variations after its initial

551: growth phase.  The nature of the time-variability can be studied using

552: the Fourier power spectrum of the accretion rate

553: during the latter portion of the simulation (beyond $t=600$, i.e., after the

554: initial growth phase is over).  This is shown in Figure~7.  It is

555: most simply described as a ``red" power-law; roughly speaking,

556: $P(f) \propto f^{-1.7}$.  However, there is also some curvature, in the

557: sense that it steepens with increasing frequency.  Because the accretion

558: rate does appear to have a well-defined mean value, $d\ln P/d\ln f$ must

559: be $> -1$ at the lowest frequencies.  At the high frequency end,

560: the spectrum is largely featureless, extending almost to the inverse

561: free-fall time at the innermost boundary.  Other than a smooth

562: steepening, there is no special feature at the orbital frequency

563: of the marginally stable orbit (marked on the figure).  The flow is

564: unsteady everywhere, even well inside the plunging region.

565: 

566: Figure~6 provides some idea of how nonsteady the accretion is; the mass

567: flux is highly spatially inhomogeneous as well.  To gain some sense of

568: the magnitude of the spatial variations, we show the accretion

569: rate as a function of position on a cylinder at $R= 5$ (Figure~8).  As

570: this figure shows, the fluctuations are very large.  If we suppose that

571: the mean accretion rate of 5 is evenly distributed over the range of

572: altitudes $-1 \leq z \leq 1$, the mean mass flux would be $\sim 0.1$.

573: As the figure makes plain, there are several places where the mass flux

574: is five times as great as this; there are also several places where the

575: mass flux is $\simeq 0.3$ -- 0.4 {\it outward}.  Relatively large

576: fluctuations compared to the mean are to be expected when the transport of

577: angular momentum in the disk is by turbulence. 

578: 

579: \subsection{Propagation of fluctuations}

580: 

581:   As we have stressed all along, one of the most striking features

582: of this simulation is the presence of large amplitude nonaxisymmetric

583: modulations in the density, mass flux, and magnetic field strength.

584: We have already discussed how $(\phi,t)$ spacetime plots enable us to

585: measure their angular velocity.  It is possible to learn more

586: by constructing analogous plots in $(R,t)$.  For example, in Figure~9

587: we show the history of the azimuthally- and vertically-averaged

588: mass flux (eq. \ref{mdot}) in this simulation.  The diagonal stripes

589: show the radial motion through the disk of fluctuations in the mass

590: flux, their slopes indicating their radial group speeds.  Typically

591: these disturbances begin near $R \simeq 5$ and travel both outward and

592: inward.  In the outward direction, they move at a very nearly constant

593: speed $\simeq 0.07$, while in the inward direction, their speed is at

594: least three times faster.  Because the outward speed very nearly

595: matches the magnetosonic speed, we infer that these disturbances

596: propagate as magnetosonic waves in the outward direction.  On the other

597: hand, inward-going mass flux perturbations are simply advected along

598: with the dynamical inflow.

599: 

600:    We conjecture that their origin lies in the way in which material

601: enters the region of unstable orbits.  When a fluid element begins to

602: accelerate inward, it creates a rarefaction wave that moves outward at

603: the magnetosonic speed, while traveling azimuthally at the (faster)

604: orbital speed.  On the other hand, the radial inflow speed

605: near and inside the marginally stable orbit is comparable to the

606: magnetosonic speed, so fluctuations travel inward rather more quickly.

607: These fluctuations are never smoothed out into a completely steady flow

608: because individual fluid elements move into the plunging region as a

609: result of the continually changing torques exerted by the MHD

610: turbulence (see the next subsection).  Thus, at this level of detail,

611: there is no way to achieve a stationary state.

612: 

613:     Although strong nonaxisymmetric pressure waves were generated by MHD

614: turbulence in the shearing box simulations, they had no

615: significant impact on the evolution of those simulations.  In this

616: global simulation, however, as Figure~9 shows, they play

617: a significant role in the creation of large fluctuations in the

618: accretion rate.

619: 

620: \subsection{$R$--$\phi$ Stress}

621: 

622: The total stress must satisfy the equation of angular momentum

623: conservation.  In a steady-state disk, this means that locally the

624: torque must be equal to the mass accretion rate (which is constant)

625: times the difference between the local specific angular momentum $\ell$

626: and the specific angular momentum carried into the black hole

627: $\ell(r_{g})$.  Thus, the stress is equal to

628: \begin{equation}\label{steadystress} 

629: \bar

630: S = {\dot M \Omega(R) \over 2\pi} \left[1 - {\ell(r_g) 

631: \over \ell(R)}\right], 

632: \end{equation} 

633: where $\dot M$ is the total mass accretion rate, $\Omega(R)$ the

634: rotational frequency (for the pseudo-Newtonian potential, $\Omega(R) =

635: 1/[R^{1/2}(r-1)]$), and $\ell$ is the specific angular momentum ($\ell

636: \equiv R^2 \Omega$).

637: In the time-steady, axisymmetric ``Novikov-Thorne" disk, it is assumed

638: that $\ell(r_g) = \ell(r_{ms})$ because the stress is assumed to be zero

639: for all $R \leq r_{ms}$.

640: 

641: 

642: For the purpose of studying angular momentum transport, it is useful to

643: look at the azimuthally- and vertically-integrated magnetic stress,

644: $\langle \int dz  M^{R\phi}\rangle_\phi$, as this is the quantity most directly

645: related to transfer of the $z$-component of angular momentum.  In particular,

646: we compare this quantity to the total stress per unit area

647: that would be expected in the steady state ``Novikov-Thorne" disk.  As

648: discussed above, in the inner part of the disk the mean accretion rate

649: is about 5 in code units.  In Figure~10 $\bar S$ (for $\dot M=5$ and

650: $\ell(r_g) = \ell(r_{ms})$) is

651: contrasted with $\langle \int \, dz \, M^{R\phi}\rangle_\phi$ averaged

652: from $t=750$ to $t=1500$.  As that figure

653: shows, in the main body of the disk (approximately $4 \leq R \leq 15$),

654: the average magnetic stress is almost exactly equal to the expected stress.

655: At large radius the two curves diverge because

656: equation (\ref{steadystress}) assumes a time-steady disk, whereas ours

657: has only a finite mass, so the behavior at its radial exterior is quite

658: different.  At the inner edge of the disk, in contrast to the

659: prediction of the conventional model, the azimuthally- and vertically-averaged

660: magnetic stress does not

661: diminish, but, if anything, increases slightly inward (the final upward

662: spike is an artifact of the boundary condition).  

663: Note, however, that the local value of the ratio

664: $|M^{R\phi}/(\rho v_R v_\phi)|$ in the plunging region varies widely:

665: from as much as $\sim 10$ several scale-heights out of the equatorial

666: plane to as little as $\sim 10^{-4}$ at some locations in the plane.

667: 

668:    In the context of time-steady and axisymmetric disks, the stress is often

669: discussed in terms of the dimensionless parameter introduced by Shakura

670: \& Sunyaev (1973),

671: \begin{equation}

672: \alpha_{SS} \equiv \langle {\int \, dz \, T^{R\phi} \over \int \, dz \, p}

673:                    \rangle_\phi .

674: \end{equation}

675: %%%NEW

676: In many papers following this seminal work, the ``$\alpha$ model" is

677: generalized so that the {\it local} ratio of stress to pressure

678: is measured by the quantity $\alpha \equiv |T^{R\phi}|/p$.  Often,

679: it is also assumed that the stress is due to some variety of

680: turbulent viscosity.  However,

681: here we follow the spirit of the original Shakura \& Sunyaev work

682: by supposing that $\alpha$ is merely a convenient dimensionless

683: way of measuring the stress in units of the pressure.  We do not

684: particularly care whether the stress is due to chaotic or regular

685: mechanisms.  In this spirit,

686: we define $\alpha \equiv |M^{R\phi}|/p$.

687: %%%

688: Unlike $\alpha_{SS}$, which, in

689: most studies, is taken to be a constant independent of time and location,

690: $\alpha$ in this simulation is a dynamical and highly variable quantity

691: that is determined by local and transient magnetic dynamics.  For this

692: reason, the absolute value is significant; although $M^{R\phi}$ is

693: positive most places, there are counter-examples.

694: 

695: To give a sense of its scale and range of variation, we first examine

696: $\langle \alpha \rangle_\phi$, the azimuthally-averaged version of this

697: ratio, as found in the late-time snapshot (Figure~11).  In the main

698: portion of the disk, $\langle \alpha \rangle_\phi$ ranges between $\sim

699: 10^{-2}$ and $\sim 10^{-1}$.  However, it rises dramatically (to

700: $\sim10$) at several scale-heights above and below the mid-plane.  The

701: same tendency has been seen (with much better vertical resolution)

702: in stratified shearing box simulations (Miller \& Stone 2000).

703: $\langle \alpha \rangle_\phi$ also increases sharply at small radius.

704: In the mid-plane, $\langle \alpha \rangle_\phi \simeq 0.2$ at $R=3$,

705: but rises to $\sim 3$ at the inner radial edge of the simulation.

706: Moreover, the vertical

707: rise in $\langle\alpha\rangle_\phi$ also becomes sharper at small

708: radius, so that at $R=3$, $\langle \alpha \rangle_\phi \sim 10$ at

709: $|z|$ as small as 0.5.

710: 

711: In addition to these systematic trends, $\alpha$ is also subject to

712: sizable fluctuations.  Measured in the same way as for the magnetic

713: energy density and the gas density (i.e., in terms of fluctuations in

714: azimuth at fixed $R$ and $z$), the {\it rms} fractional fluctuation

715: amplitude in the bulk of the disk ranges from $\simeq 1.5$ to $\simeq

716: 3$, but at large altitude above the disk midplane, the fluctuations can

717: be much greater.  The character of these fluctuations is illustrated by

718: a different projection of $\alpha$, which we call $\bar{\alpha}$:

719: \begin{equation}

720: \bar{\alpha} \equiv {\int \, dz \, \rho \alpha \over \int \, dz \, \rho},

721: \end{equation}

722: the density-weighted vertical average of $\alpha$.  In this projection

723: (Figure~12), we clearly see the nature of the azimuthal fluctuations:

724: just as for the density fluctuations (shown on smaller scale in Figure~2),

725: the magnetic fluctuations are sheared filaments of limited angular coherence

726: length.

727: 

728: These results highlight the limitations of the primary assumption made

729: in many previous studies of disk structure, that $\alpha_{SS}$ is a

730: universal constant, independent of time and location.  By contrast, we

731: find that the ratio of stress to pressure varies strongly from place to

732: place, both in the sense of having large fluctuations and in the sense

733: of possessing strong systematic gradients.  In many places the gas

734: pressure is relatively smooth while the magnetic stress varies

735: considerably.  In other places the variation in $\alpha$ is simply due

736: to changes in the background pressure.  For example, the magnetic

737: stress does not drop with height as rapidly as the pressure does.

738: 

739: For purposes of comparison with standard disk models, it is possible to

740: compute (albeit with some trepidation) the value of $\alpha_{SS}$ found

741: in this simulation.  This quantity, defined in terms of the Maxwell

742: stress, is shown as a function of radius in Figure~13 for the disk at

743: the end of the simulation.  It is $\sim 0.1$ through most of the disk,

744: but, starting just outside the marginally stable orbit, it rises inward.

745: It reaches a peak $\sim 1$ near the inner radial boundary of the

746: simulation.   The rise is primarily due to the rapid drop in gas

747: pressure as the accretion flow accelerates inward.  Also shown is the

748: ratio of the Maxwell stress to the magnetic pressure (dashed line).

749: This $\alpha_{mag}$ is 0.2--0.3 throughout most of the disk, and

750: increases inside $r_{ms}$.

751: 

752: Whether $\alpha_{SS}$ is meaningful depends on how it is used.  As the

753: previous discussion emphasized, the local ratio of magnetic stress to

754: pressure has strong vertical gradients that are completely obscured

755: when the vertically-integrated version is used.  Similarly, there are

756: large fluctuations in azimuth and as a function of time.   For these reasons,

757: $\alpha_{SS}$ is useless as a local dynamical variable.  However,

758: the approximate constancy of $\alpha_{SS}$ as a function of radius {\it

759: in the disk proper} means that it can be used as a rough estimator of

760: the surface density in the body of the disk, provided its use is not

761: extended to local properties nor is it thought to be accurate at better

762: than ``factor of a few" accuracy.  This conclusion is entirely

763: consistent with many standard uses of $\alpha$, and within the spirit

764: with which it was originally introduced by Shakura and Sunyaev (1973)

765: as a scaling parameter, not a fundamental variable.

766: 

767: \subsection{Accreted angular momentum and energy}

768: 

769: Another significant diagnostic of the flow is the mean specific angular

770: momentum distribution.  There are a variety of ways to characterize the

771: specific angular momentum; one is by vertical and

772: azimuthal average, $\langle \ell \rangle$, as defined in equation

773: (\ref{el}).  This function was discussed in Hawley (2000) for the model

774: GT4 (see Fig.~14 in Hawley 2000);  the present model is very similar.

775: At late time the average specific angular momentum is slightly

776: sub-Keplerian throughout most of the disk.  The most striking feature

777: is that $\langle \ell \rangle$ continues to decline inside $r_{ms}$

778: rather than leveling off.  If all stress ceased at the marginally

779: stable orbit, material would travel inward from there retaining

780: precisely its energy and angular momentum.  Conversely, the degree to

781: which fluid changes its specific energy and angular momentum is a

782: measure of the stresses that exist in the plunging region.  The mean

783: specific angular momentum diminishes from 2.60 at $r_{ms}$ to 2.48 at

784: the innermost radius.  Thus, on average in this snapshot, the accreted

785: matter has lost $\sim$5\% of the angular momentum it had when it

786: was at the marginally stable orbit.  It is likely (for reasons

787: explained in \S 5) that this number is an underestimate, and in reality

788: the specific angular momentum at the innermost radius is rather

789: smaller.

790: 

791: The total rate at which angular momentum is carried across the inner

792: radial boundary is slightly smaller than these numbers would suggest

793: because there is also a small outward electromagnetic angular momentum

794: flux.  This electromagnetic flux is due to a generic property of

795: magnetic fields in shearing plasma: when the orbital frequency

796: decreases outward, the field is swept back so as to transport angular

797: momentum outward, i.e., $\langle B_R B_\phi \rangle < 0$.  However,

798: near the inner radial boundary this is a small effect, the

799: electromagnetic angular momentum flux there is only about 2\% of the

800: angular momentum flux carried by the matter.

801: 

802: Further complexity is revealed by a closer examination of the $(R,z)$

803: distribution of the late time azimuthally-averaged angular momentum

804: inside the plunging region.  In fact, the specific angular momentum

805: in the cells lying along the equatorial plane

806: is nearly constant inside $r_{ms}$.  However, the value of

807: $\ell$ drops with $z$ above and below the equator.  This is consistent

808: with the distribution of magnetic field which, of course, accounts for

809: the nonzero stress.  The low value of field near the equator is in part

810: a consequence of the assumed symmetry of the initial distribution of

811: field.  The initial field loops are symmetric with respect to the

812: equator ($B_R$ and $B_\phi$ are zero there), and throughout the run the

813: strongest fields tend to be located above and below the equatorial

814: plane.

815: 

816: Just as for every other quantity, there is also considerable spiral structure.

817: Figure~14 shows the density-weighted vertically-averaged specific

818: angular momentum.  It makes plain how non\-axisymmetric the distribution of

819: torque is in the plunging region.

820: 

821:     For a number of reasons, the accreted energy is harder to determine

822: than the accreted angular momentum (see further discussion in \S 5.2).

823: To introduce this topic, we begin by displaying (Figure~15) the

824: specific energy $e$ in rest-mass units,

825: \begin{equation}\label{specenergy}

826: ec^2 = {1\over 2} v^2 + {(\Gamma -1)P\over \rho } 

827: + {B^2\over8\pi \rho } - {1\over (r-r_g)},

828: \end{equation}

829: in the equatorial plane at late time.  The specific energy for a

830: circular orbit precisely at $r_{ms}$ is -0.0625; here, $\langle e

831: \rangle_\phi$ is a bit greater than that ($\simeq -0.055$) because of

832: the thermal, magnetic, and turbulent kinetic energy contributions to

833: the total energy.  In the equatorial plane, $\langle e \rangle_\phi$ is

834: almost constant from $R=3$ to just outside $R = 2$, but actually {\it

835: rises} as $R$ decreases from there to $R=1.6$, reaching a peak that is

836: very nearly unbound: -0.01.  The azimuthally-averaged energy in the

837: equatorial plane then falls sharply to -0.07.  Even this wild swing

838: underestimates the fluctuations: the smallest specific energy found in

839: the equatorial plane is -0.19.  Thus, we are confident in asserting

840: that there is a large amount of energy transfer within the plunging

841: region.  However, given these large fluctuations, we do not feel that

842: we can compute a well-defined mean energy for the accreted matter.  It

843: is likely, moreover, that at least some of the very large fluctuations

844: in the innermost radial zones are not physical, but due to the inner

845: boundary condition.

846: 

847: In fact, the situation is somewhat worse than this.  The reason is that

848: dissipation is crucial for computing the accreted energy, and our

849: simulation contains no real dissipation physics.  Absent numerical

850: losses (see \S 5.1, 5.2), energy taken from plunging matter may go

851: several places.  One possibility is to build magnetic field energy in

852: the plunging region itself.  If the field energy is advected into the

853: black hole without dissipation and radiative losses, the transfer of

854: energy from fluid to magnetic field makes no difference to the accreted

855: energy.  It is also possible for magnetic stresses to transfer energy

856: all the way back to the disk proper (and indeed we see significant

857: magnetic stress throughout the plunging region).   Again, without

858: dissipation, the energy transferred in that way stays in the fluid,

859: whether in the form of kinetic or magnetic energy.  Only in the event

860: that the energy is transferred all the way out to non-accreting matter

861: would it be possible for that energy to avoid being carried back in.

862: Thus, to the degree that this simulation avoids numerical losses, it is

863: very difficult for it to show any significant transfer of energy out of

864: the plunging region.  By contrast, in a real disk, dissipation changes

865: energy into heat, which may then be radiated if the thermal timescale

866: is shorter than the inflow timescale; in our simulation, neither

867: physical dissipation nor radiation is present.  

868: 

869: \section{Discussion}

870: 

871: \subsection{Black hole spin-up}

872: 

873: We are now ready to begin answering some of the questions posed in the

874: Introduction.  Most significantly, this simulation has shown that the

875: stress due to MHD turbulence does not automatically cease at the

876: marginally stable orbit.  In fact, it continues to be present

877: throughout the plunging region.  As an immediate consequence, matter

878: continues to lose angular momentum as it travels from the marginally

879: stable orbit toward the event horizon.  Because it then enters the

880: black hole with smaller angular momentum than if no stresses had been

881: present, the rate at which the black hole is spun up is slower than it

882: would have been otherwise (Agol \& Krolik 2000; Gammie 1999).  It

883: further follows that estimates of the black hole spin distribution are

884: shifted toward slower rotation (e.g., Moderski \& Sikora 1996).

885: Although in this simulation the magnitude of this effect is small

886: ($\simeq 5\%$), it is possible that more realistic conditions will

887: increase its size (see \S 5.1).

888: 

889: \subsection{Additional dissipation in the inner disk}

890: 

891:     We have already remarked on the fact that our simulation does not

892: directly compute the dissipation rate.  In this section we attempt to

893: estimate what dissipation might be associated with dynamics of the sort

894: seen here.  In conventional disk models, the local dissipation rate is

895: directly tied to the local stress:

896: \begin{equation}\label{dissipation}

897: Q = \left|{d\Omega \over d\ln r}\right| \int \, dz T^{R\phi},

898: \end{equation}

899: where $Q$ is the dissipation rate per unit area.  This relationship is

900: predicated upon the notion that the same stress producing local

901: transport also produces local dissipation.  This remains true for the

902: mean flow dynamics of MHD turbulence (Balbus \& Papaloizou 1999); in

903: fact, numerical dissipation mimicking the physical dissipation at the

904: short wavelength end of the turbulent cascade is the way in which

905: matter in this simulation loses energy as it moves slowly inward in the

906: body of the disk.  However, this identification is not necessarily

907: correct as a prediction for the instantaneous rate of dissipation in

908: specific fluid elements.  It is especially questionable in the plunging

909: region.  The stress is not necessarily turbulent, and there may not be

910: enough time for the energy deposited in the fluid by the stress to be

911: transferred to small enough lengthscales for it to be dissipated.

912: Nonetheless, in this subsection we adopt the speculative assumption that

913: the vertically-integrated dissipation rate is proportional to the

914: vertically-integrated stress in order to speculate about the

915: consequences for disk radiation that may result.

916: 

917: One possible consequence is an altered radial distribution of the disk

918: surface brightness.  As shown in Figure~10, the azimuthally-averaged

919: stress in the simulation is nearly constant inside $R\simeq 5$.  If the

920: relation of equation (\ref{dissipation}) holds, the total dissipation

921: rate outside $R=3$ in this simulation would be $\simeq 50\%$ greater

922: than in a disk with the same accretion rate, but zero stress at the

923: marginally stable orbit.

924: 

925: If this were to happen, there would be a number of effects of

926: considerable phenomenological interest (Agol \& Krolik 2000).  Because

927: this additional heat is released in a relatively small area of the

928: disk, it leads to relatively high temperatures, and therefore

929: contributes to the highest-frequency portion of the spectrum.  Another

930: consequence stems from the fact that the dissipation is located in the

931: innermost part of the disk, where relativistic effects are strongest.

932: The light that is radiated due to this extra heating should be

933: comparatively strongly beamed into the equatorial plane by a

934: combination of special relativistic Doppler beaming driven by the

935: orbital motion and general relativistic trajectory-bending.

936: Trajectory-bending also forces a large fraction of the

937: photons to return to the disk at larger radius, to be reprocessed there

938: into lower-frequency light.

939: 

940: Whether or not the enhanced stresses in the plunging region 

941: lead to additional radiative losses

942: depends on the dissipation and photon diffusion rates.  In the disk

943: proper, dissipation takes place in the turbulent cascade, in small

944: lengthscale fluctuations that are short-lived, exhibit little spatial

945: coherence, but are smoothly distributed in a statistical sense.  In the

946: time-varying accretion of the plunging region this turbulent cascade

947: may not have time to go to completion.  On the other hand, there might

948: be dissipative events, mediated by reconnection for example, that

949: involve structures with relatively large spatial coherence lengths and

950: which are not necessarily smoothly distributed throughout the region.

951: As we will discuss in \S 5.1, our limited spatial resolution makes it

952: difficult for us to estimate fairly the rate of such events in the

953: plunging region.  Moreover, assessing the radiative properties of such

954: dissipative events is beyond the the capabilities of this simulation.

955: However, if events of this sort do occur, significant luminosity might

956: result because the reduced optical depth of the plunging matter (as

957: compared to the disk proper) also greatly reduces the thermal time.

958: Note that this mechanism results in an enhanced radiative efficiency

959: without the transfer of any energy from the plunging region to the disk

960: outside the marginally stable orbit.

961: 

962: Finally, although the azimuthally-averaged stress matches conventional

963: expectations in much of the disk, there are large azimuthal variations

964: in the form of sheared filaments (Figure~12).  Translated into a

965: dissipation rate, these spiral waves of enhanced (or diminished) stress

966: become spirals of enhanced (or diminished) heating.  Whether this

967: concentration of heating has an effect on the emitted radiation depends

968: on the characteristic timescale of these fluctuations {\it as seen in a

969: specific fluid element}.  If a particular fluid element has an elevated

970: heating rate for as long as its thermal timescale, the effective

971: radiating area of the disk is reduced to the area of the regions where

972: the heating is greatest, and the typical effective temperature is

973: raised accordingly.  On the other hand, if over a thermal timescale

974: individual fluid elements see both positive and negative fluctuations,

975: the effects will average out.   We have already estimated that the

976: characteristic speed of fluctuations through the disk is of order the

977: magnetosonic speed; on the other hand, the thermal timescale is $\sim

978: \alpha_{SS}^{-1}$ times the orbital period.  Consequently, the ratio of

979: the coherence time of fluctuations to the thermal time should be, very

980: roughly, $\sim \alpha_{SS} v_{orb}/(v_{A}+c_s)$; in this simulation this

981: ratio is $\sim 1$.

982: 

983: \subsection{Fluctuations in light output}

984: 

985:     We have emphasized throughout this paper the highly non-stationary

986: behavior that is intrinsic to disk behavior as found in this simulation.

987: One form these fluctuations can take is the coherent waves that

988: are continually radiated outward through the disk from the region of

989: the marginally stable orbit.  Even if, on average, dynamics in the

990: plunging region change the mean accretion efficiency by only a small

991: amount, the large fluctuations we observe in the simulation should cause

992: substantial time-dependent variations in the light output.

993: 

994: Once again, because there is no direct connection between the dynamics

995: traced by this simulation and dissipation, we cannot directly predict the time

996: variation of the disk luminosity.  In fact, to do so would also require

997: calculating the photon diffusion rate from inside the (optically thick)

998: disk.  However, some indication of what may occur may be gleaned from

999: Figures~6 and 7.  Some of the fluctuations in the accretion rate

1000: (particularly those at the highest frequencies) are due to dynamics in

1001: the plunging region; these may or may not lead to dissipation (see \S

1002: 4.2), and therefore may or may not contribute to fluctuations in the

1003: light output.  For this reason, and following the {\it ansatz} of

1004: equation (\ref{dissipation}), in Figure~7 we also plot the Fourier

1005: power spectrum of the volume-integrated magnetic stress.  Whether the

1006: accretion rate or the stress is more closely related to the dissipation

1007: rate, the actual output is the convolution of fluctuations in the

1008: heating rate with the probability distribution for the photon escape

1009: time.  In frequency space, this amounts to a (position-dependent)

1010: low-pass filter.  When the heating occurs deep inside the disk, the

1011: cut-off frequency for this filter is $\sim \alpha_{SS}\Omega$;

1012: fluctuations in the light due to heating closer to the surface are

1013: subject to a less stringent frequency cut-off (in the optically thin

1014: limit, there is no cut-off at all).  Consequently, we expect the

1015: Fourier power spectrum of the disk luminosity to be rather strongly cut

1016: off at frequencies above $\sim 10^{-2}$, but the power spectrum of

1017: fluctuations associated with coronal dynamics may be different.

1018: 

1019: The observation of QPOs in black hole candidates has

1020: motivated a number of studies of possible sources of systematic time

1021: variability in accretion disks.  In a series of papers (Nowak \&

1022: Wagoner 1991, 1992; Perez et al. 1997) Wagoner and collaborators have

1023: explored the linear theory of fluid oscillations in the relativistic

1024: portions of stationary accretion disks.  In these disks only

1025: hydrodynamic effects are considered, and angular momentum transport is

1026: modeled by a pseudo-viscosity parameterized by a constant

1027: $\alpha$.  These disks possess a variety of normal modes, including

1028: $p$-modes that are trapped between $r_{ms}$ and the radius where the

1029: epicyclic frequency is maximum, and $g$-modes, whose greatest amplitude

1030: is found under the peak of the epicyclic frequency curve.  In the

1031: Paczy\'nski-Wiita potential we use, the position of the maximum

1032: epicyclic frequency is $R \simeq 3.6$; with the additional correction

1033: terms introduced by Nowak and Wagoner (1991), it moves outward to

1034: $R=4$.  We would not expect to find exactly the same mode frequencies

1035: as they because of the difference in the potentials used; however, the

1036: mode frequencies should in general be $\sim \Omega(r_{ms})$.  A perusal

1037: of Figure~7 fails to find any evidence for special frequencies in this

1038: range.

1039: 

1040: There are several reasons why trapped disk oscillations do not appear

1041: here.  First, the normal modes discussed in the previous paragraphs are

1042: predicted on the basis of a purely {\it hydrodynamic} analysis; no

1043: account is made of magnetic forces.  Here, however, magnetic forces

1044: play a major role in determining the dynamics of fluid elements.  For

1045: example, if one repeated the derivation of the ``diskoseismic"

1046: dispersion relation, but allowed for a weak magnetic field, one would

1047: rediscover the magneto-rotational instability, thus vitiating the

1048: hydrodynamic ``normal modes".  More fundamentally, our simulation

1049: indicates that there is no underlying quiet flow to support coherent

1050: oscillations.  The disk is so turbulent that there is no steady

1051: equilibrium against which linear perturbations can develop.  Put

1052: another way, we find that the typical radial motion timescale in the

1053: vicinity of $r_{ms}$ (where these normal modes are supposed to be

1054: concentrated) is comparable to the predicted mode frequencies, so the

1055: fundamental assumption that these modes are perturbations to circular

1056: orbits is inappropriate.

1057: 

1058: \section{Limitations}

1059: 

1060: The results presented here and in Hawley (2000) obviously represent

1061: only the first steps in detailed simulations of global accretion

1062: disks.  In this section we outline how the approximations we have made

1063: may impact our conclusions.  Of course the only definitive way to

1064: investigate the effects of these approximations would be with

1065: additional simulations, and this discussion should be viewed as a guide

1066: for future work.

1067: 

1068: \subsection{Effects of resolution}

1069: 

1070: Foremost among the limitations is the numerical resolution of the

1071: simulation.  Although we placed 30 radial zones between $R=1.5$ and

1072: $R=4$, that is probably insufficient.  There are two reasons for

1073: believing this to be the case.  First, we may apply the standard test

1074: for convergence of comparing two simulations with the same physics but

1075: different resolution, and ask whether they show significant

1076: differences.  Simulation GT4, reported in Hawley (2000), and a new

1077: $64^3$ run (hereafter referred to as run LR for Lower Resolution) serve

1078: as good comparisons.  Both GT4 and LR share identical initial and

1079: boundary conditions.  GT4 had equally spaced grid zones over a distance

1080: of 20 in both $R$ and $z$ for $\Delta R = \Delta z = 0.156$.

1081: Simulation LR used equally spaced radial zones over a range of 30 for

1082: $\Delta R = 0.47$, and used half of the $z$ grid zones centered around

1083: the equator within $z = \pm 2$, for $\Delta z= 0.125$.  By contrast,

1084: here $\Delta R = 0.0833$ and $\Delta z = 0.0625$ in the inner region.

1085: 

1086: All three simulations are qualitatively similar in global properties,

1087: as measured by volume integration, but differ in certain other respects

1088: that are important to physical interpretation.  The time-averaged

1089: accretion rate in GT4 is about 40\% smaller than in the work reported

1090: here (although the time-averaged accretion rate in LR is only about 2\%

1091: smaller).  However, the most striking contrast is in magnetic field

1092: strength.  Higher resolution results in larger magnetic fields,

1093: particularly in the poloidal components.  Contrasting maps of the

1094: magnetic field intensity at similar late time for GT4 and our new

1095: simulation (Figure~16), we find that the field is consistently stronger

1096: in the new simulation everywhere inside $R=5$, rising to typically a

1097: factor of 3 greater in magnitude inside $R=4$.  Significantly, outside

1098: $R=5$, the field is generally stronger in the old simulation (sometimes

1099: by more than a factor of 10), but that, too, is in keeping with the

1100: sense of the resolution contrast: GT4 has better resolution in the

1101: outer regions.

1102: 

1103: The field intensity is controlled by a trade-off between the turbulent

1104: MHD dynamo driven by the magneto-rotational instability and numerical

1105: dissipation (recall that there is no explicit resistivity in this

1106: simulation).  Whatever the real physics of resistivity in accretion

1107: disks may be, it is unlikely that it is as large as the effective

1108: numerical resistivity induced by our modest spatial resolution.  The

1109: contrast between the field intensity in the two simulations is

1110: therefore very likely explained by the smaller numerical resistivity of

1111: the better-resolved simulation.

1112: 

1113: Another piece of evidence supporting the belief that the magnetic field

1114: is weaker than it would be with better resolution is presented in

1115: Figure~17, where we display the magnetic field structure in the

1116: equatorial plane within $R=4$, plotted for every fourth grid zone.  

1117: The planar component of the field is

1118: organized into long sheared loops.  Because the loops are wrapped

1119: around one another, there are many places where the field direction

1120: reverses within 4 -- 8 grid cells.  This means that with a modest

1121: amount of flow convergence numerical effects could cause the field to

1122: decay on a relatively short timescale.

1123: 

1124: Finally there is the issue of the computational domain employed, in

1125: particular the location of the boundaries.  Very little outflow

1126: leaves through the $z$ boundaries (typically only a few

1127: percent of the accretion rate).  The outer radial boundary was

1128: extended in the present simulation compared to GT4, which had the

1129: effect of keeping more of the disk on the grid, but otherwise had no

1130: obvious effect on the evolution.  We expect that there is also little

1131: dependence upon the location of the inner boundary because the

1132: flow across that boundary is generally supermagnetosonic.

1133: Although we made no tests here of the effect of varying $r_{in}$, tests

1134: of this sort were made on the similar simulations in Hawley (2000),

1135: and it appeared to make little difference.

1136: 

1137: \subsection{Energy conservation and physical dissipation rates}

1138: 

1139: The artificial dissipation caused by insufficient resolution also has

1140: consequences for energy conservation.  To very good accuracy, angular

1141: momentum is conserved in this simulation.  The primary source of

1142: numerical dissipation, the truncation error associated with transport

1143: terms in the equation, does not alter the total angular momentum,

1144: although it may artificially change its distribution.

1145: Total energy, however, is different because the energy equation that is

1146: solved is the internal energy equation.  Numerical dissipation can

1147: destroy kinetic energy in regions of strong velocity gradients and

1148: magnetic energy in regions of strong magnetic field gradients, but

1149: there is no provision in this simulation for putting that energy back

1150: into heat (except for dissipation in shocks through the use of an

1151: artificial viscosity).  Both sorts of dissipation can also occur in

1152: real systems, but at rates that might be very different from what

1153: happens in the simulation.

1154: 

1155: In the body of the disk, the problems that numerical dissipation can

1156: cause are limited.  The single largest items in the energy budget are

1157: potential energy and rotational kinetic energy; neither one is

1158: significantly affected by numerical effects.  The thermal energy is

1159: only a small fraction of those energies; this is a prime motivation

1160: for solving an internal rather than total energy equation.

1161: Artificial energy losses occur in the dissipation of relatively small

1162: amounts of energy bound up in small-scale random fluid motions and in

1163: magnetic field.  In fact, in this regard, numerical dissipation acts in

1164: a way that is consistent with the physical dissipation of energy at the short

1165: lengthscale end of the turbulent cascade.  In both cases, the losses

1166: occur on the shortest lengthscales, and in both cases (assuming the

1167: disk radiates efficiently) energy is lost

1168: from the disk on timescales short compared to the inflow time.

1169: 

1170: However, near and inside the marginally stable orbit, detailed

1171: questions of energy transfer become significant.  The contributions to

1172: the energy budget due to both turbulent motions and magnetic field

1173: grow.  At the same time, their importance also grows.   For example, it

1174: would be of great interest to determine whether plunging fluid elements

1175: can, by exerting magnetic stresses, transfer energy back to the disk

1176: proper (cf. \S 3.7).  As we have seen, significant magnetic stresses

1177: exist throughout this region, so there is some possibility that this happens. 

1178: However, the rate at which this may occur is regulated by the magnetic field

1179: strength and topology, and these are subject to significant artificial

1180: numerical dissipation.  In addition, any energy that is delivered is

1181: likely to take the form of MHD turbulence; whether this turbulence is

1182: dissipated (and the energy radiated) quickly or slowly compared to the

1183: inflow is critical for evaluating the radiative efficiency that

1184: results.  Because the inflow, thermal, and dissipation timescales all

1185: change significantly in the vicinity of $r_{ms}$, it is unlikely that

1186: in this region numerical dissipation serves as a successful stand-in

1187: for physical dissipation and radiative losses in the way it does at

1188: larger radii. 

1189: 

1190: \subsection{Equation of state}

1191: 

1192: Dissipation and radiation are, of course, key elements in the gas

1193: thermodynamics.  In the present simulation, we describe the gas's

1194: internal energy in only an approximate way: we assume it is

1195: adiabatic except for heating due to artificial viscosity.

1196: Explicit treatment of dissipation and radiation would fill in the

1197: missing terms in the gas energy equation.

1198: 

1199: In the absence of a full treatment of the energy equation, we can

1200: at best make educated speculations about the impact of our assumption

1201: that the gas behaves adiabatically. Some consequences were already

1202: discussed in the previous subsection.  There are several more, not

1203: as closely related to resolution issues.

1204: 

1205: As we have remarked earlier (\S 3.7), one of the most dramatic features

1206: of this simulation is the coherent disturbances that propagate as

1207: magnetosonic waves from the vicinity of the marginally stable orbit

1208: outward through the disk.  Their behavior depends strongly, of course,

1209: on the sound speed in the disk.  However, in this simulation, in which

1210: all gas is given the same specific entropy initially and then evolved

1211: adiabatically, the sound speed could be quite different from what it

1212: would be in a more realistic simulation that includes heating from

1213: dissipation of MHD turbulence and cooling via radiation.  We

1214: expect, therefore, that properties that depend on pressure forces (such

1215: as the propagation of these waves, or the vertical thickness of the

1216: disk) are not well-described in this simulation.

1217: 

1218: Another consequence of our assumed equation of state, combined with

1219: the particular value of specific entropy we assigned the gas, is that

1220: the disk is modestly thick.  As a result, it begins from a non-Keplerian,

1221: partially pressure-supported state and radial pressure gradients

1222: drive some of its initial evolution.   Magnetic stresses

1223: rapidly change the angular momentum distribution within the torus to

1224: nearly Keplerian, but without cooling, the internal pressure continues to

1225: push matter away from the pressure maximum.  Some accretion,

1226: therefore, is driven not by magnetic stresses but simply by the pressure

1227: gradient.  At late time, however, the inner disk establishes a

1228: quasi-steady state and nearly all the accretion can be accounted for by

1229: the Maxwell stress.  It is likely that the observed dynamics are

1230: reasonably representative of generic inner disk accretion.  

1231: 

1232: Finally, although the present disk is moderately thin, with $H/R

1233: \simeq 0.15$ in the inner region, it is conceivable that new

1234: phenomena might appear in a still thinner disk.  Although our results

1235: provide no particular reason to predict this, it is possible that in a

1236: disk with still smaller $H/R$ there might be modes with wavelength

1237: large compared to $H$ yet still small compared to $R$.  Questions such

1238: as these could be investigated by contrasting the present work with

1239: simulations of the inner regions of more extended, thinner, initially

1240: Keplerian disks.

1241: 

1242: A possible example of this sort of comparison is provided by the work

1243: of Armitage et al. (2000).  Like us, they also simulated the inner

1244: region of an accretion disk in a pseudo-Newtonian potential using a 

1245: three dimensional

1246: MHD code.  They find a smaller reduction of $\ell$ inside $r_{ms}$ than

1247: we find in the simulation reported here, an effect they conjecture

1248: could be due to the smaller temperature and scale height in their

1249: disk.  However, their scale height is $\simeq 0.08$, not much smaller

1250: than our typical value of $\simeq 0.15$.  Moreover, there are other

1251: differences between the two simulations that may also account for

1252: quantitative contrasts: in their stratified simulation, only a limited

1253: vertical extent is simulated; their equation of state is isothermal;

1254: and they simulate a $\pi/3$ wedge, rather than a full $2\pi$ in

1255: azimuth.  Because in our simulation fluctuations with azimuthal

1256: wavelength $\sim 1$~radian or more have considerable amplitude, the

1257: limitation to a wedge could have significant consequences.  In any

1258: event, as they emphasize, the differences between their simulations and

1259: ours are simply quantitative, not qualitative.  They, too, find a

1260: nonzero Maxwell stress inside $r_{ms}$, but its amplitude is lower.

1261: Further work is required to determine what sets the stress levels in a

1262: variety of different situations.

1263: 

1264: Thus, for all these reasons, it will be important for future simulations

1265: to incorporate improved treatment of gas heating and cooling.  It

1266: should not be difficult to integrate a total energy equation along with

1267: the gas momentum (i.e., force) equation.  Heating could be treated in a

1268: controlled way by inserting an explicit resistivity into the magnetic

1269: field evolution equation and an explicit viscosity into the fluid force

1270: equation, and placing corresponding terms in the internal energy equation.

1271: Radiative losses could be described with either an escape probability

1272: formalism, or, in the longer term, solution of the radiation transfer

1273: problem.

1274: 

1275: A simulation with both  heating and radiation would simulate

1276: the behavior of thin, radiative disks.  There has been no previous

1277: work of this sort; the significance of such an effort is obvious.

1278: 

1279: Heating alone would, in effect, lead to a simulation of the popular

1280: advection dominated accretion flow (ADAF) model.  Some previous

1281: simulations of this model have used a phenomenological prescription for

1282: viscosity (e.g., Igumenshchev \& Abramowicz 1999; Igumenshchev,

1283: Abramowicz, \& Narayan

1284: 2000; Stone, Pringle \& Begelman 1999).  Others have computed the

1285: stress from a global magnetic field (in axisymmetry: Uchida \& Shibata

1286: 1985 and Stone \& Norman 1994; in three dimensions:  Matsumoto 1999).

1287: Still another approach has been to study the stress created by MHD

1288: turbulence including resistive dissipation, but in axisymmetry (Stone

1289: \& Pringle 2000).  Addition of explicit dissipation to the simulation

1290: presented here would, in effect, extend the Stone \& Pringle work to

1291: three dimensions.

1292: 

1293: \subsection{Duration} 

1294: 

1295: The timestep of the simulation is limited by the Courant condition.

1296: The most stringent limitation comes from the high orbital velocities

1297: and the radial plunging speeds in the inner part of the grid where the

1298: resolution is reasonably fine.   The total duration of the simulation

1299: was 1500 time units, requiring almost 400,000 timesteps.  As discussed

1300: in \S3.1, this is long enough for a quasi-steady state to be

1301: established in the inner part of the disk.  Since $\alpha \sim 0.1$,

1302: the simulation covers several accretion times in the inner part of the

1303: disk.  Whether or not the results obtained are truly representative,

1304: must, however, be tested with simulations of longer duration.  An

1305: evolution beginning with a nearly Keplerian disk that extends out to

1306: large radius could allow the inner disk region to establish itself

1307: self-consistently from a long-term accretion flow.

1308: 

1309: \subsection{Field topology}

1310: 

1311: What are the limitations due to the choice of initial disk and field

1312: topology?  The initial poloidal field loops used here are amenable to

1313: rapid development of the MRI on large scales.  Further, the presence of

1314: significant initial radial field quickly generates through simple shear

1315: amplification a toroidal field of near-equipartition strength.  Thus,

1316: the initial field topology guarantees magnetically-driven evolution

1317: within a few orbits.  In earlier work (Hawley 2000) it was shown that

1318: in this regard (generating magnetically-driven inflow after a few orbits),

1319: initially poloidal and initially toroidal field configurations 

1320: are essentially

1321: equivalent.  However, the symmetry inherent in this

1322: choice of initial field topology is partly responsible

1323: for the low field intensity near the equatorial plane and the weak

1324: dipolar structure characteristic of the low-density regions (\S 3.3).

1325: 

1326:      Whether the field is contained entirely within the problem area,

1327: or, equivalently, whether there is zero net flux through the box,

1328: has a greater qualitative impact.  Shearing box simulations (Hawley,

1329: Gammie \& Balbus 1995, 1996) also found strong similarities between

1330: simulations with initial toroidal field and simulations with random

1331: initial fields.  However, they found dramatically different behavior

1332: when there was net vertical field.  Such field topologies tend to

1333: produce larger net angular momentum transport and more violent

1334: evolution.  These effects have also been observed in 

1335: global simulations (Matsumoto 1999).

1336: 

1337: \subsection{Displacement current}

1338: 

1339: The equations of MHD as we have written them (see especially

1340: equation~2) omit the displacement current term.  This approximation is

1341: valid so long as the Alfv\'en speed $v_A < c$.  This condition is

1342: satisfied almost everywhere in this simulation, with the only exception

1343: a small region near the inner radial boundary ($1.5 \leq R \leq 2$ and

1344: $1 \leq |z| \leq 2$).  In the equatorial plane, where most of the mass

1345: flows, the largest value of $v_A \simeq 0.3$, and that occurs only at

1346: the inner radial boundary; everywhere outside $R = 2$ in the midplane,

1347: $v_A \leq 0.15$.  Future simulations in which the displacement current

1348: term is included should be somewhat more accurate, but we doubt that

1349: this correction will cause any major changes.

1350: 

1351: \subsection{Relativity}

1352: 

1353: Obviously, it would be preferable to run the simulation using genuine

1354: general relativistic dynamics, rather than the approximation of

1355: Newtonian physics in a Paczy\'nski-Wiita potential.  We might hope that

1356: this approximation is qualitatively reasonable when attempting to mimic

1357: dynamics around a non-rotating black hole; after all, the most

1358: important aspect of this case is the existence of dynamically unstable

1359: orbits inside a critical radius and enforced inward radial motion

1360: within a still smaller distance from the center.  Further, Gammie \&

1361: Popham (1998) compute the growth rate of the MRI in full relativity and

1362: have shown that it remains undiminished $r_{ms}$.  The

1363: Paczy\'nski-Wiita potential, however, has no representation of the

1364: frame-dragging created by a rotating black hole, so it cannot be a good

1365: approximation to accretion dynamics around a Kerr black hole.  Gammie

1366: (1999) and Gammie \& Popham (1998) examine some of the interesting

1367: possibilities created by the Kerr geometry, but examination of those

1368: effects must await a code with true relativistic dynamics.

1369: 

1370: \section{Conclusions}

1371: 

1372: The simulation reported here represents another step towards more

1373: realistic simulations of accretion disk dynamics.  By improving the

1374: grid resolution in the inner region, we have shown that magnetic

1375: effects become increasingly important near and inside the radius of the

1376: marginally stable orbit.   Contrary to the most widely-held assumption,

1377: the $R$--$\phi$ component of the stress tensor does not go to zero at

1378: $r_{ms}$; averaged over azimuth, height, and time, it is almost

1379: constant as a function of $R$ for $R \leq 2 r_{ms}$.  As a result,

1380: matter in the region of unstable orbits does not follow simple energy-

1381: and angular momentum-conserving free-fall trajectories, although we are

1382: not yet ready to make quantitative predictions about how much the

1383: energy and angular momentum change.

1384: 

1385: The ratio $T^{R\phi}/P \equiv \alpha$ is commonly used as a

1386: measure of the magnitude of the stress, and almost as commonly is

1387: thought to be a constant, independent of location and time.  Our

1388: simulation showed clearly that while it can provide a good

1389: qualitative measure of the stress when averaged over several orbits

1390: in time,  at any particular instant $\alpha$ has large systematic

1391: gradients (as a function of both $R$ and $z$) and substantial

1392: temporal fluctuations.  On average the value of $\alpha$ in the

1393: disk is $\sim 0.1$.

1394: 

1395: This simulation has shown that substantial time- and space

1396: variations are the rule.  The inner regions of accretion disks around

1397: black holes are highly turbulent, non-steady systems.  Large-amplitude

1398: fluctuations sheared into spiral fragments come and go, many of them

1399: initiated in the region immediately outside the marginally stable

1400: orbit.  As a result, there are continuing fluctuations in the accretion

1401: rate; we expect that some of these fluctuations will be mirrored in the

1402: light curves of accreting black holes.

1403: 

1404: Future simulations, guided by these results, will probe the beyond the

1405: limitations of the present simulation.  Within the context of the same

1406: pseudo-Newtonian potential and the present adiabatic equation of state,

1407: we can explore the effects of different initial disk configurations and

1408: magnetic field strengths and topologies, at higher resolution and

1409: for longer time.  Beyond this, it will be important to incorporate

1410: improved physics into the models.   These include general relativistic

1411: dynamics and improved treatments of the equation of state and

1412: dissipative processes.   This future work should bring us still closer

1413: to a quantitative understanding of the dynamics of accretion onto black

1414: holes.

1415: 

1416: \begin{acknowledgments}

1417: 

1418: JHK would like to thank Eric Agol for numerous helpful conversations,

1419: and Nancy Levenson for invaluable instruction in the wiles of IDL.

1420: He would also like to thank Giora Shaviv for comments on the difference

1421: between heating rate fluctuations and luminosity fluctuations.

1422: JFH thanks Charles Gammie for useful comments on the manuscript.

1423: This work was supported by NSF grant AST-0070979, and NASA grants

1424: NAG5-9266 and NAG5-7500 to JFH, and NSF Grant AST-9616922 and NASA

1425: grant NRA-99-01-ATP-031 to JHK.  Simulations were carried out on the

1426: Cray T3E system of the San Diego Supercomputer Center of the National

1427: Partnership for Advanced Computational Infrastructure, funded by the

1428: NSF.

1429: 

1430: \end{acknowledgments}

1431: 

1432: \begin{thebibliography}{}

1433: 

1434: \bibitem[Abramowicz \& Kato]{ak89} Abramowicz, M.A., \& Kato, S. 1989, ApJ,

1435: 336, 304

1436: \bibitem[Abramowicz \& Zurek 1981]{az81} Abramowicz, M.A., \& Zurek, W.H. 1981,

1437: ApJ, 246, 314

1438: \bibitem[Agol \& Krolik 2000]{ak00} Agol, E., \& Krolik, J.H. 2000, 

1439: ApJ, 528, 161

1440: \bibitem[Armitage 1998]{a98} Armitage, P.J. 1998, ApJ, 501, L189

1441: \bibitem[Armitage 2000]{a00} Armitage, P.J., Reynolds, C.S., \& Chiang,

1442: J. 2000, ApJ, submitted

1443: \bibitem[Balbus \& Hawley 1991]{bh91} Balbus, S.A., \& Hawley, J.F. 1991, ApJ,

1444: 376, 214

1445: \bibitem[Balbus \& Hawley 1998]{bh98} Balbus, S.A., \& Hawley, J.F. 1998, Rev

1446: Mod Phys, 70, 1

1447: \bibitem[Balbus \& Papaloizou 1999]{bp99} Balbus, S.A., \& Papaloizou,

1448: J. C. B. 1999, ApJ, 521, 650

1449: \bibitem[Brandenburg et al. 1996]{b96} Brandenburg, A.,  Nordlund, \AA,

1450: Stein, R.~F., \& Torkelsson, U. 1995, ApJ, 446, 741

1451: \bibitem[Gammie 1999]{g99}Gammie, C.F. 1999, ApJ, 522, L57

1452: \bibitem[Gammie 1998]{gp98}Gammie, C. F., \& Popham, R. 1998, ApJ, 498, 313

1453: \bibitem[Hawley 2000]{h00} Hawley, J.F. 2000, ApJ, 528, 462

1454: \bibitem[Hawley, Gammie \& Balbus 1995]{hgb95} Hawley, J.F., Gammie, C.

1455: F.,  \& Balbus, S.A. 1995, ApJ, 440, 742

1456: \bibitem[Hawley, Gammie \& Balbus 1995]{hgb96} Hawley, J.F., Gammie, C.

1457: F.,  \& Balbus, S.A. 1996, ApJ, 464, 690

1458: \bibitem[Hawley \& Stone 1995]{hs95}Hawley, J. F., \& Stone, J. M. 1995,

1459: Comp Phys Comm, 89, 127

1460: \bibitem[Igumenshchev \& Abramowicz 1999]{ia99}

1461: Igumenshchev, I. V., \& Abramowicz, M. A. 1999, MNRAS, 303, 309

1462: \bibitem[Igumenshchev et al. 2000]{i00}

1463: Igumenshchev, I. V., Abramowicz, M. A., \& Narayan, R. 2000, ApJ, 537, 27

1464: \bibitem[Krolik 1999]{h99}Krolik, J.H. 1999, ApJ, 515, L73

1465: \bibitem[Machida et al. 2000]{mhm00} Machida, M., Hayashi, M.~R., \&

1466: Matsumoto, R. 2000, ApJ, 532, L67

1467: \bibitem[Matsumoto 1999]{m99} Matsumoto, R. 1999, in Numerical Astrophysics, 

1468: eds. Miyama, S., Tomisaka, K. \& Hanawa, T (Kluwer:  Dordrecht), 195

1469: \bibitem[Matsumoto et al. 1984]{m84} Matsumoto, R., Kato, S., Fukue, J.,

1470: \& Okazaki, A.T. 1984, PASJ, 36, 71

1471: \bibitem[Miller \& Stone 2000]{ms00} Miller, K.A., \& Stone, J.M. 2000, ApJ,

1472: 534, 398

1473: \bibitem[Moderski \& Sikora 1996]{ms96} Moderski, R., \& Sikora, M. 1996, 

1474: MNRAS, 283, 854

1475: \bibitem[Muchotrzeb\& Pacy\'nski 1982]{mp82} Muchotrzeb, B., \& Paczy\'nski, B.

1476: 1982, Acta. Astr., 32, 1

1477: \bibitem[Novikov \& Thorne 1973]{nt73} Novikov, I.D., \& Thorne, K.S. 1973, in

1478: Black Holes, eds. C. de Witt and B. de Witt (New York: Gordon \& Breach), 343

1479: \bibitem[Nowak \& Wagoner 1991]{nw91} Nowak, M., \& Wagoner, R.V. 1991,

1480: ApJ, 378, 656

1481: \bibitem[Nowak \& Wagoner 1992]{nw92} Nowak, M., \& Wagoner, R.V. 1992,

1482: ApJ, 393, 697

1483: \bibitem[Paczy\'nski \& Wiita 1980]{pw80}Paczy\'nski, B., \& Wiita, P. J.

1484: 1980, A\&A, 88, 23

1485: \bibitem[Page \& Thorne 1974]{pt74} Page, D.N., \& Thorne, K.S. 1974, 

1486: ApJ 191, 499

1487: \bibitem[Perez et al. 1997]{p97} Perez, C.A., Silbergleit, A.S., Wagoner,

1488: R.V., \& Lehr, D.E. 1997, ApJ, 476, 589

1489: \bibitem[Shakura \& Sunyaev 1973]{ss73} Shakura, N.~I., \& Sunyaev, R.~A.

1490: 1973, A\&A, 24, 337

1491: \bibitem[Stone et al. 1996]{s96} Stone, J. M., Hawley, J. F., Gammie, C.

1492: F., \& Balbus, S. A.  1996, ApJ, 463, 656

1493: \bibitem[Stone \& Norman 1992a]{sn92a} Stone, J. M., \& Norman, M. L.

1494: 1992a, ApJS, 80, 753

1495: \bibitem[Stone \& Norman 1992b]{sn92b} Stone, J. M., \& Norman, M. L.

1496: 1992b, ApJS, 80, 791

1497: \bibitem[Stone \& Norman 1994]{sn94} Stone, J. M., \& Norman, M. L. 1994,

1498: ApJ, 433, 746

1499: \bibitem[Stone \& Pringle 2000]{sp00} Stone, J. M, \& Pringle, J. E.

1500: 2000, MNRAS, in press

1501: \bibitem[Stone et al. 1999]{spb99} Stone, J. M, Pringle, J. E., \&

1502: Begelman, M. C. 1999, MNRAS, 310, 1002

1503: \bibitem[Sunyaev \& Revnivtsev 2000]{sr00} Sunyaev, R., \& Revnivtsev,

1504: M. 2000, A\&A, 358, 617

1505: \bibitem[Uchida \& Shibata 1985]{us85} Uchida, Y., \& Shibata, K. 1985,

1506: PASJ, 37, 515

1507: \end{thebibliography}

1508: 

1509: \clearpage

1510: \centerline{Figure Captions}

1511: 

1512: \figcaption{Azimuthally-averaged gas density distribution at the beginning

1513: ($t=0$: panel a) and the end ($t=1500$: panel b) of the simulation.

1514: There are 14 evenly spaced contours.  The maximum density at $t=0$ is 34.2;

1515: at $t=1500$, $\rho_{max}=14.3$.}

1516: 

1517: \figcaption{Density in the equatorial plane for $R \leq 10$ at $t=1500$.}

1518: 

1519: \figcaption{Azimuthally-averaged magnetic pressure, $\langle

1520: B^2/8\pi\rangle_\phi$, plotted on a logarithmic scale from $10^{-5}$

1521: to $10^{-2}$.  The mean value of magnetic pressure

1522: within the region where $\rho > 0.01$ is $7.1\times 10^{-4}$.

1523: Using the mean gas pressure this corresponds to an averge 

1524: $\beta = 13$ in the disk.

1525: }

1526: 

1527: 

1528: \figcaption{Poloidal magnetic field structure in the slice at $\phi = \pi/2$,

1529: overlaid on density contours.  Panel a) shows the inner, high-resolution

1530: region; panel b) shows the disk body.  The length of each arrow

1531: is proportional to the magnitude of the poloidal field, but the scales

1532: are different in the two panels.}

1533: 

1534: \figcaption{Accretion rate (in code units) as a function of

1535: radius at times $t=1000$ 

1536: and $t=1500$ (solid lines). The dashed curve is the time-average of the

1537: accretion rate between those two times. Note that

1538: negative accretion rate means outward flow.}

1539: 

1540: \figcaption{Mass accretion rate $\langle \dot M \rangle$ through the

1541: inner radial boundary as a function of time.  To capture the

1542: time-variability accurately, the accretion rate was measured every 10

1543: timesteps in the simulation, corresponding to an average interval of

1544: 0.0037 in time.}

1545: 

1546: \figcaption{

1547: Fourier power density of the accretion rate through the inner boundary

1548: (solid curve) and volume-integrated Maxwell stress (dashed curve).

1549: Frequency is in code units; the orbital frequency (i.e., the inverse of

1550: the orbital period $\Omega(r_{ms})/2\pi$) at the marginally stable

1551: orbit is marked with an arrow.}

1552: 

1553: \figcaption{ Accretion rate at cylindrical radius $R=5$; inward mass flow

1554: is given a positive sign.  Note that the 

1555: {\it local} mass flux is almost as likely to be outward as inward.}

1556: 

1557: \figcaption{Vertically- and azimuthally-averaged mass flux as a function of

1558: radius $R$ and time.  The diagonal stripes denote radially-propagating

1559: waves.  Reds and yellows correspond to accretion, blues to outflow.

1560: }

1561: 

1562: \figcaption{Azimuthally-averaged and vertically-integrated magnetic

1563: stress averaged from $t=750$ until the end of the simulation 

1564: (solid curve) and nominal mean stress per unit area (dashed curve).}

1565: 

1566: \figcaption{Absolute value of $\langle \alpha \rangle_\phi$, in a

1567: logarithmic scale in the snapshot at $t=1500$.

1568: This quantity increases with decreasing radius, and

1569: with increasing distance above or below the disk midplane.}

1570: 

1571: \figcaption{Absolute value of the vertically-averaged $\bar{\alpha}$, plotted

1572: on a logarithmic scale for the snapshot at $t=1500$.  Note the

1573: large fluctuations, even when values are compared at fixed radius.

1574: $\bar{\alpha}$ is large at both large and small $R$, but for different

1575: reasons (see text).}

1576: 

1577: \figcaption{Shakura-Sunyaev $\alpha_{SS}$ as a function of radius 

1578: (solid line) constructed from vertical and azimuthally  averaging the

1579: Maxwell stress and the gas pressure.  Also shown is the averaged

1580: $\alpha_{mag}$ as a function of radius (dashed line), constructed using

1581: the magnetic pressure in place of the gas pressure.  $\alpha_{mag}$ is

1582: relatively constant in the disk, whereas $\alpha_{SS}$ varies with the

1583: gas pressure.

1584: }

1585: 

1586: \figcaption{Density-weighted vertically-averaged specific angular momentum

1587: in the inner portion of the accretion flow.}

1588: 

1589: \figcaption{Specific energy of matter (in rest-mass units) in the

1590: equatorial plane within $R=4$ (the high-resolution segment of the

1591: simulation).  Note the large amplitude spiral fluctuations.  The sharp

1592: features immediately surrounding the inner radial boundary are 

1593: artifacts.} 

1594: 

1595: \figcaption{Logarithm of the ratio of the density-weighted mean magnetic

1596: field intensity in the new simulation to the same quantity in GT4 at

1597: analogous late times.  Only the region inside $R=10$ is shown.

1598: Because these are different simulations, one cannot expect features 

1599: to match up precisely.  However, there is a clear general

1600: tendency for the field in the new simulation to be stronger than in the

1601: old simulation inside $R=5$ and weaker outside.}

1602: 

1603: \figcaption{Planar component of the magnetic field in the equatorial

1604: slice of the high-resolution region of the simulation.  The length of

1605: each arrow is proportional to the intensity of the magnetic field.  For

1606: clarity, the field is shown in every fourth cell.}

1607: 

1608: \end{document}

1609: 

1610: 

1611: