1: \documentclass[apj]{emulateapj}

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

3: \usepackage{amsmath}

4: \bibliographystyle{apj}

5: 

6: \newcommand \p{\partial}

7: \newcommand \lya{Ly$\alpha$ }

8: \def\dim#1{\mbox{\,#1}}

9: \newcommand \chimps{\ h^{-1} \dim{Mpc}}

10: 

11: \begin{document}

12: 

13:   \title{Resolving Gas Dynamics in the Circumnuclear Region of a Disk Galaxy

14:   in a Cosmological Simulation}

15:   

16:   \author{Robyn Levine\altaffilmark{1,2,3}}

17:   \author{Nickolay Y. Gnedin\altaffilmark{3,4,5}}

18:   \author{Andrew J. S. Hamilton\altaffilmark{1,2}}

19:   \author{Andrey V. Kravtsov\altaffilmark{4,5,6}}

20:   \altaffiltext{1}{JILA, University of Colorado, Boulder, CO 80309;

21:   robyn.levine@colorado.edu}

22:   \altaffiltext{2}{Department of Astrophysical \& Planetary Sciences,

23:   University of Colorado, Boulder, CO 80309, USA}

24:   \altaffiltext{3}{Particle Astrophysics Center, Fermi National

25:   Accelerator Laboratory, Batavia, IL 60510, USA}

26:   \altaffiltext{4}{Kavli Institute for Cosmological Physics, The

27:   University of Chicago, Chicago, IL 60637, USA}

28:   \altaffiltext{5}{Department of Astronomy \& Astrophysics, The

29:   University of Chicago, Chicago, IL 60637, USA}

30:   \altaffiltext{6}{Enrico Fermi Institute, The University of Chicago,

31:   Chicago, IL 60637, USA}

32: 

33:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

34:   

35:   \begin{abstract}

36:     Using a hydrodynamic adaptive mesh refinement code, we simulate

37:     the growth and evolution of a galaxy, which could potentially host

38:     a supermassive black hole, within a cosmological volume. Reaching

39:     a dynamical range in excess of 10 million, the simulation follows

40:     the evolution of the gas structure from super-galactic scales all

41:     the way down to the outer edge of the accretion disk. Here, we

42:     focus on global instabilities in the self-gravitating, cold,

43:     turbulence-supported, molecular gas disk at the center of the

44:     model galaxy, which provide a natural mechanism for angular

45:     momentum transport down to sub-pc scales. The gas density profile

46:     follows a power-law $\propto r^{-8/3}$, consistent with an

47:     analytic description of turbulence in a quasi-stationary

48:     circumnuclear disk. We analyze the properties of the disk which

49:     contribute to the instabilities, and investigate the significance

50:     of instability for the galaxy's evolution and the growth of a

51:     supermassive black hole at the center.

52:   \end{abstract}

53:   

54:   \keywords{galaxies: evolution---galaxies: high-redshift---galaxies:

55:   ISM---galaxies: nuclei---galaxies: structure }

56: 

57:   

58:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

59: 

60:   \section{INTRODUCTION AND MOTIVATION}

61: 

62:   It is increasingly evident that the growth histories of different

63:   components of galaxies: stars, gas, and supermassive black holes,

64:   are intricately connected. Observations indicate a relationship

65:   between the masses of supermassive black holes (SMBHs) and various

66:   properties of their host galaxies, such as the spheroid mass

67:   \citep{Magorrian98} and the velocity dispersion of stars in the

68:   bulge \citep{FM00, Geb00, Trem02}. Recent simulations have shown

69:   that feedback from accreting SMBHs can regulate the growth of the

70:   black holes as well as the evolution of their host galaxies, making

71:   feedback a potentially important piece of the theory of SMBH-host

72:   galaxy co-evolution \citep[e.g.][]{DiMatteo05, Springel05,

73:   DiMatteo07, Sijacki07}.

74: 

75:   Building up the mass of a galaxy and driving AGN feedback requires a

76:   continuous replenishment of fuel in the center of the galaxy. A

77:   comprehensive understanding of such fueling is not possible without

78:   detailed knowledge about matter transport from large scales to the

79:   vicinity of the black hole. That transport helps determine, in

80:   particular, both the amount of material available for accretion,

81:   feedback, and star formation, as well as the strength and duration

82:   of the fueling. Large scale gravitational tidal fields created

83:   during major mergers of galaxies are thought to be effective at

84:   funneling matter toward the centers of galaxies. Indeed, simulations

85:   and semi-analytic modeling of hierarchical growth scenarios have

86:   shown that a combination of accretion and black hole mergers can

87:   effectively build-up the local black hole population

88:   \citep{KaufHae00,YooME04,VolRees05,Malbonetal06,VolRees06,Lietal07},

89:   as well as assemble the $\sim 10^9$ M$_{\sun}$ black holes already

90:   observed to be present in quasars at $z\approx6$

91:   \citep{Fan03,Lietal07}.

92: 

93:   Secular evolution of galaxies, particularly in the absence of major

94:   merger events, can drive fuel down to the center as well. Global bar

95:   instabilities are thought to be efficient at transporting angular

96:   momentum, allowing material in the disk to move toward the center of

97:   the galaxy

98:   \citep{Robertsetal79,Simkinetal80,Noguchi88,barswinbars,KormKenn04,RegTeu04}.

99:   In the ``bars within bars'' scenario of \citet{barswinbars}, a large

100:   scale galactic bar drives gas inward where it forms a

101:   self-gravitating disk. As this disk becomes unstable, a smaller

102:   secondary bar forms, driving material down to scales of order $10$

103:   pc, at which point other physics can take over and transport the

104:   material the rest of the way toward the black hole

105:   \citep[e.g.][]{ShlosRev90}.

106:   

107:   Given the complexity of transporting matter from cosmological scales

108:   all the way down to the circumnuclear region of a galaxy, and

109:   ultimately to the vicinity of a SMBH, it is clear that a thorough

110:   understanding of the relationship between SMBHs and galaxy formation

111:   requires modeling of a wide range of scales. Recently, cosmological

112:   SPH simulations have been combined with smaller scale simulations of

113:   a SMBH host galaxy environment to study the effects of merger driven

114:   fueling and AGN feedback on supermassive black hole growth and

115:   demography \citep{DiMatteo07, Sijacki07}. Reaching a large dynamic

116:   range in N-body+SPH simulations, \citet{Mayeretal07} have studied

117:   galaxy dynamics during mergers resulting in a supermassive black

118:   hole binary. Small scale simulations have addressed gas dynamics in

119:   sub-galactic scale disks with high resolution \citep{Fukuda00,

120:   Wada01, WadaNorman01,Escala07,WadaNorman07}, finding the

121:   development of a turbulent, multi-phase interstellar medium (ISM).

122: 

123:   The study we present here follows the evolution of the circumnuclear

124:   region in a typical galaxy environment rather than a dramatic, rare

125:   event such as a major merger, in a self-consistent cosmological

126:   simulation. We have specifically chosen a simulated galaxy that will

127:   evolve into a typical $L_*$ galaxy at $z=0$. The simulation follows

128:   the galaxy during a phase of its evolution in which it has not

129:   undergone any major mergers within several dynamical time scales, in

130:   order to follow the development of instabilities in the

131:   circumnuclear disk, which may drive the transport of matter and

132:   angular momentum from large to small scales. The cosmological

133:   simulations use the adaptive mesh refinement (AMR) technique to

134:   self-consistently model the gas dynamics in a single galaxy at high

135:   resolution (sub-pc resolution in the center of the galaxy). A large

136:   dynamic range ($> 10$ million), achievable with AMR technique,

137:   allows us to bridge cosmological scales to scales relevant for

138:   molecular cloud formation (the birthplace of stars) and AGN

139:   fueling. After studying the physics in this basic model galaxy, we

140:   can begin to include physical processes that are directly relevant

141:   to the problem of SMBH growth in the context of galaxy evolution,

142:   such as AGN feedback.

143: 

144:   It is a complex task to implement mergers, feedback and secular

145:   evolution in large, cosmological simulations all at once. Our

146:   approach is to split the problem into pieces to be addressed one at

147:   a time, ultimately building a more realistic simulation. While the

148:   present paper focuses on the structure of the galaxy, and the

149:   development of instabilities which may be responsible for driving

150:   matter and angular momentum transport within the galaxy, we plan to

151:   explore the time evolution of such transport, specifically the

152:   evolution of the accretion rates of mass and angular momentum, in a

153:   subsequent paper. The nature of the circumnuclear region and a

154:   quantitative analysis of accretion rates each have important

155:   consequences for galaxy evolution.

156: 

157:   The organization of the paper is as follows. In Section

158:   \ref{sec:sim}, we describe the details of the simulation and the

159:   adopted ``zoom-in'' method. In Section \ref{sec:res}, we describe

160:   the features of the highly-resolved galaxy in a single zoom-in

161:   episode at $z=4$, including a stability analysis of the disk and the

162:   potential role of physics not included in the current

163:   simulation. Section \ref{sec:miss} examines the potential role of

164:   physics missing from the simulations. Finally, Section

165:   \ref{sec:disc} contains a summary of our conclusions and their

166:   interpretation.

167: 

168:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

169: 

170:   \section{SIMULATION}

171:   \label{sec:sim}

172: 

173:   Sections \ref{subsection:art} and \ref{subsection:ref} briefly

174:   explain details of the simulation and of our methods. Additional

175:   details regarding tests of angular momentum conservation, and the

176:   treatment of dark matter particle discreteness effects will be

177:   available in the PhD thesis of the author Levine (also see the

178:   discussion of angular momentum conservation in Appendix

179:   \ref{sec:con}).

180:   

181:   \subsection{Simulation Code}

182:   \label{subsection:art}

183: 

184:   It is unfeasible to follow the evolution of a galaxy all the way

185:   down to the scale of a black hole hole accretion disk over

186:   cosmological times. Such a simulation spans a large range of scales

187:   (with a dynamic range $> 10^7$), and each spatial scale has a

188:   different temporal scale. As a result, the Courant condition for

189:   numerical stability requires very small time steps in the most

190:   highly refined regions, making it computationally expensive to

191:   follow the evolution of the galaxy over long periods of time. It is

192:   much more feasible to start with a lower resolution cosmological

193:   simulation, and zoom in to the small-scale region with increasingly

194:   smaller cell sizes and time steps. As the simulation evolves on

195:   small scales, the large scale portion of the simulation does not

196:   undergo much evolution, and does not need as high resolution.

197: 

198:   The cosmological simulations presented in this paper are conducted

199:   with the Adaptive Refinement Tree (ART) code \citep{Kravtsovetal97,

200:   KravtsovPhD, Kravtsovetal02}. The ART code includes the technique of

201:   adaptive mesh refinement (AMR), allowing high resolution of a galaxy

202:   residing in a small region of the cosmological simulation, while

203:   following the rest of the simulated region with lower

204:   resolution. The technique is appropriate for studying the structure

205:   and evolution of a galaxy over a large dynamical range and in a

206:   cosmological context.

207:  

208:   The ART code includes a range of physics for modeling dark matter,

209:   stars, and gas dynamics. Gas is converted into stars in cells with

210:   densities greater than $\rho_{\textrm{SF}}$ and temperatures less

211:   than $T_{\textrm{SF}}$, where $\rho_{\textrm{SF}} = 1.64

212:   \dim{M}_{\sun} \dim{pc}^{-3}$ and $T_{\textrm{SF}} = 9000$ K

213:   \citep[see][for more details]{Kravtsov03}, resulting in a star

214:   formation efficiency consistent with a Kennicutt law on kiloparsec

215:   scales \citep{Kennicutt98} and with observations on $100$ pc scales

216:   as well \citep[e.g.][]{Youngetal96, WongBlitz02}. The code follows

217:   ISM physics, such as molecular hydrogen formation, and gas cooling

218:   by heavy elements and dust under the assumption of collisional

219:   ionization equilibrium. The cooling and heating rates are tabulated

220:   as functions of gas density, temperature, metallicity, and redshift

221:   over the temperature range $10^2<T<10^9$ K using CLOUDY

222:   \citep{Cloudy98}, which accounts for the metallicity of the gas, and

223:   formation of molecular hydrogen and cosmic dust. In future studies,

224:   we plan to include the ART code's radiative transfer capabilities,

225:   which will be necessary for implementing AGN feedback.

226: 

227: 

228:   %%%%%%%%%%%%%%%%%%%%%%%%%%%

229: 

230:   \subsection{``Zooming-In'' to the Center of a Galaxy}

231:   \label{subsection:ref}

232: 

233:   We begin with a cosmological simulation, evolved from a realization

234:   of a random Gaussian density field at $z=50$, with periodic boundary

235:   conditions, measuring $6 h^{-1}$ comoving Mpc across. The simulation

236:   was first run with low-resolution in order to select a galactic-mass

237:   halo for subsequent study. A Lagrangian region the size of five

238:   virial radii of the halo at $z=0$ was then identified at $z=50$, and

239:   re-sampled and run with higher resolution to $z=2.8$ \citep[see][for

240:   a description of the technique]{Klypinetal01}. The Lagrangian region

241:   of the simulation was automatically refined three levels, as a

242:   minimum. There are $2.64 \times 10^6$ dark matter particles in the

243:   Lagrangian region, each with mass $9.18 \times 10^5 h^{-1}

244:   \dim{M}_{\sun}$. Subsequent refinement and de-refinement in this

245:   region followed a dark matter mass criterion, in which a cell was

246:   refined if its total dark matter mass is greater than $1.8 \times

247:   10^6 h^{-1} \dim{M}_{\sun}$. At $z=4$, the highest matter density

248:   peaks in the simulation have a maximum resolution of $\approx 50$ pc

249:   (in physical units; corresponding to $9$ levels of refinement on top

250:   of a $64^3$ root grid). The largest halo has a total mass of $2

251:   \times 10^{11} \dim{M}_{\sun}$ at $z=4$, and it contains the

252:   progenitor of an $L^*$ spiral galaxy. This initial cosmological

253:   simulation was run including metal enrichment and energy feedback

254:   from supernova, and radiative transfer in addition to the physics

255:   described in Section \ref{subsection:art}.

256: 

257: %\clearpage

258:   \begin{figure*}

259:     \center{\includegraphics[width=1.9\columnwidth]{f1.eps}}

260:     \caption{\label{fig:pan} Volume rendering of the $3$-dimensional

261:     gas density on several different scales at $z=4$, from $200$ kpc

262:     in the upper left panel, to $2$ pc in the lower right panel. The

263:     color bar shows the density scale in units of

264:     $\dim{cm}^{-3}$. Note the presence of the spiral arms and

265:     instabilities on a wide range of scales. (This figure is best

266:     viewed in color.)} 

267:   \end{figure*}

268: %\clearpage

269: 

270:   This particular $z=4$ cosmological simulation was chosen for this

271:   study because it contains a galaxy that has not been disturbed by a

272:   major merger since it merged with a galaxy with $25\%$ its mass at

273:   $z\approx6$. The time since this dynamically active period ($\approx

274:   600$ million years) is significantly longer than the dynamical time of

275:   the galactic disk at $z\approx4$ ($\approx 15$ million years). The

276:   galaxy is, however, far from quiescent, having grown in mass by

277:   $\approx 25\%$ from $z=5$ to $z=4$ via minor mergers and accretion

278:   from its environment.

279: 

280: %\clearpage

281:   \begin{figure*}

282:     \center{\includegraphics[width=1.9\columnwidth]{f2.eps}}

283:     \caption{\label{fig:star} The stellar component of the galaxy at

284:     $z=4$ on a scale of $3$ kpc. The stars in the image are divided

285:     into two populations, with yellow stars corresponding to the oldest

286:     stars in the galaxy, and blue stars corresponding to the youngest

287:     stars. The youngest stars are found primarily in the disk of the

288:     galaxy. (This figure is best viewed in color.)}

289:   \end{figure*}

290: %\clearpage

291: 

292:   In the cosmological simulation, we allow further refinement (at $z

293:   \approx 4$), one level at a time, in a $1.5$ kpc region centered on

294:   the galaxy, effectively zooming in to the center of the galaxy with

295:   increasing resolution. The ``zoom-in'' technique is similar to the

296:   one used by \citet{Abeletal02} in simulations of the formation of

297:   the first star, which required an even larger dynamical

298:   range. Because we are starting simply in this first pilot study, and

299:   building a more realistic simulation in pieces, radiative transfer

300:   and stellar feedback have been switched off in the zoom-in

301:   portion of the cosmological simulation (they will be examined in

302:   future studies). The high-resolution portion of the simulation

303:   employs additional refinement criteria, refining according to a

304:   level-dependent mass criterion on levels $11$ and below\footnote{It

305:   should be noted that we adopt the convention of referring to level

306:   $0$ as the ``top'' level, and the maximum (most-refined) level as

307:   the ``bottom'' level, so that the terms ``above'' and ``below''

308:   refer to lower and higher resolution, respectively.} in the zoom-in

309:   region. The refinement criterion is defined so that the finer the

310:   resolution, the more aggressively the mesh refines, ensuring that

311:   there are enough highly refined cells to resolve structures on small

312:   scales. Specifically, the mesh refinement is super-Lagrangian in the

313:   circumnuclear region of the galaxy, and cells are marked for

314:   refinement if the gas mass in a cell is

315:   $m_{\textrm{r}}^{\textrm{level}-10}$ times the Lagrangian mass

316:   criterion ($9 \times 10^7 h^{-1} \dim{M}_{\sun}$), where

317:   $m_{\textrm{r}}=0.7$ is our fiducial value. The factor $0.7$ was

318:   chosen through experimentation, so as to make the refinement in the

319:   central region more aggressive, but not so aggressive that the

320:   simulation becomes too computationally expensive.

321: 

322:   During the zoom-in period, we increase the maximum level of

323:   refinement gradually, one level at a time, allowing the simulation

324:   to reach a quasi-stationary state on each level before moving to the

325:   next. The slow initial refinement allows the ART code to resolve

326:   spatial scales within the simulated galaxy while avoiding transient

327:   effects. Since the time steps depend on the sound speed of the gas,

328:   they are therefore related to the dynamical time, and a fixed

329:   minimum number of time steps on each level forces the mesh to evolve

330:   for several dynamical times. We have run parallel simulations with

331:   $100$, $300$, and $1000$ minimum steps on each level, and have

332:   determined that a requirement of a minimum of $300$ on each level

333:   sufficiently reduces transient numerical effects that result from

334:   refining too quickly.

335: 

336:   Throughout the initial zoom-in, we take precautions to avoid

337:   numerical artifacts.  As the resolution of the simulation increases

338:   and throughout its subsequent evolution, the density of dark matter

339:   is smoothed on scales corresponding to level $9$ resolution and

340:   above. This ensures that each cell contains $\approx 15-20$ dark

341:   matter particles, avoiding effects resulting from their

342:   discreteness.  It is well known that poor resolution in

343:   hydrodynamics codes can lead to artificial fragmentation in the

344:   gas. \citet{Truelove97} investigated these numerical instabilities

345:   and found that a simulation must resolve the Jeans length to avoid

346:   artificial fragmentation. In addition to the mass criteria described

347:   above, the ART code's refinement scheme meets the Jeans condition of

348:   \citet{Truelove97} on the maximum level of refinement (which

349:   increases gradually during the initial zoom-in episode), requiring

350:   that $\Delta x/\lambda_{\textrm{J}} < 0.25$, where $\Delta x$ is the

351:   resolution, or cell size, and $\lambda_{\textrm{J}}$ is the Jeans

352:   length. Additionally, as the mesh refines, rapidly collapsing

353:   structures can cause numerical instabilities at steep density

354:   gradients. The ART code implements artificial pressure support (as

355:   described in \citet{Machacek01}) on the maximum level of refinement,

356:   in order to avoid the over-collapsing of structures.

357: 

358:   After the initial zoom-in, the highest level cells are $0.03$ pc

359:   across (corresponding to $20$ levels of refinement), allowing us to

360:   determine the location for a supermassive black hole test particle

361:   with high precision, but without resolving the black hole accretion

362:   disk, which would require additional physics (such as

363:   magneto-hydrodynamics, or MHD) not currently included in the ART

364:   code. Our results extend reliably down to a scale of at least $0.12$

365:   pc, or the size of four level $20$ cells, which we consider to be

366:   our resolution limit. After reaching this maximum resolution (level

367:   $20$), we replace $3\times10^7 \dim{M}_{\sun}$ of the gas from the

368:   center of the simulated galaxy with a black hole point mass of equal

369:   mass and momentum. At present, we focus on the dynamical properties

370:   of the circumnuclear disk on scales where the gravity of the gas

371:   dominates that of the black hole. The addition of the black hole

372:   point mass here simply allows us to follow the black hole's location

373:   and velocity as a reference point. However the successful

374:   introduction of the black hole particle will be essential in

375:   subsequent simulations involving physics associated with the black

376:   hole (such as AGN feedback).

377: 

378:   After the initial zoom-in and the successful introduction of the

379:   black hole test particle, the simulation then evolves at high

380:   resolution for approximately one dynamical time at the $100

381:   \dim{pc}$ scale, showing a highly resolved galactic disk. For a

382:   quasi-Keplerian disk, the number of orbital periods,

383:   $N_{\textrm{orb}}$, undergone by the simulation at radii less than

384:   $100 \dim{pc}$, is $(R/100\dim{pc})^{-3/2}$. Figure \ref{fig:pan}

385:   shows the $3$-dimensional gas density in an approximately face-on

386:   view of the galactic disk, on several different spatial scales, for

387:   a single time step, demonstrating the large dynamic range of the

388:   simulation. The distribution of stars is shown in Figure

389:   \ref{fig:star}, in both a face-on and an edge-on view of the galaxy

390:   on a scale of $3$ kpc. The two different colors correspond to old

391:   and young stars (yellow and blue, respectively), with the youngest

392:   stars in the galaxy located primarily in the disk, and the oldest

393:   stars distributed more isotropically, populating the galaxy's bulge.

394: 

395:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

396: 

397:   \section{RESULTS}

398:   \label{sec:res}

399:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

400:   \subsection{Structure of the Circumnuclear Disk}

401:   \label{subsection:disk}

402: 

403:   Volume rendered images of the gas density in the $z=4$ simulation,

404:   such as those in Figure \ref{fig:pan}, clearly show the spiral disk

405:   structure of the galaxy. The gas disk extends inward all the way to

406:   sub-pc scales, which are the smallest scales resolved in our

407:   simulation\footnote{In particular, the simulated disk does not

408:   contain the commonly observed toroidal structure in the inner few pc

409:   of the galaxy. The failure of our simulation to reproduce the AGN

410:   torus may be the result of additional physics still missing in the

411:   simulation, as we will discuss in \S \ref{subsection:torus}.}. The

412:   geometric structure of the circumnuclear region is best illustrated

413:   by the eigenvalues of the inertia tensor, calculated in spherical

414:   shells around the black hole. The inertia tensor is given by

415: 

416:   \begin{equation}\label{eq:iner}

417:     I_{jk} = \sum_{\alpha=1}^N m_{\alpha}x_j^{\alpha}x_k^{\alpha},

418:   \end{equation}

419: 

420:   \noindent where $m_{\alpha}$ is the gas mass of cell $\alpha$ at radius

421:   $r_{\alpha} = |\bf{x}_{\alpha}|$ inside a spherical shell containing

422:   $N$ cells, and centered on the black hole. The eigenvalues of the

423:   inertia tensor define the principal axes of each shell, and their

424:   ratios describe the shape. Figure \ref{fig:shape} shows the average

425:   radial profile of the ratios of the inertia tensor's

426:   eigenvalues. Above $0.1$ pc, $c$ is significantly smaller than $b$

427:   over several orders of magnitude in radius, indicating a disk

428:   structure. However, the fact that $b/a$ is significantly less than

429:   one in parts of the disk indicates that the disk is not axially

430:   symmetric, and that it has large-scale structures, such as spiral

431:   waves and bars.

432: 

433: %\clearpage

434:   \begin{figure}[t]

435: %    \centering \epsscale{1.}  

436:     \centering \epsscale{1.2}  

437:     \plotone{f3.eps}

438:     \caption{\label{fig:shape} Ratios of eigenvalues of the inertia

439:     tensor in spherical shells centered on the black hole. The ratio

440:     $c/a \ll b/a$, indicating a disk structure.}

441:   \end{figure}

442: %\clearpage

443: 

444:   The gas disk in the simulation is fully rotationally supported, with

445:   the tangential component of velocity dominating over the radial

446:   component by at least an order of magnitude, as is illustrated in

447:   Figure \ref{fig:vel}. The top panel of Figure \ref{fig:vel} shows

448:   the ratio of the radial to the tangential component of velocity,

449:   which is small throughout much of the disk, as the motion of the gas

450:   is almost entirely rotational. The average radial velocity is

451:   negative, indicating the inflow of gas. The bottom panel shows the

452:   tangential component of velocity in units of a quasi-Keplerian

453:   velocity, $\sqrt{GM(r)/r}$, determined by the interior total mass,

454:   $M(r)$, at each radius. The velocity is close to being Keplerian,

455:   but since the mass distribution in the disk resembles a flattened

456:   ellipsoid, rather than a spherical distribution, the rotation is

457:   slightly super-Keplerian.

458: 

459: %\clearpage

460:   \begin{figure}[t]

461: %    \centering \epsscale{1.} 

462:     \centering \epsscale{1} 

463:     \plotone{f4.eps}

464:     \caption{\label{fig:vel} {\it Top:} Ratio of radial and tangential

465:     velocity components of the gas, each averaged over $550{,}000$

466:     years. {\it Bottom:} Average tangential velocity component in units

467:     of the average quasi-Keplerian velocity.}

468:   \end{figure}  

469: %\clearpage

470: 

471:   A remarkable feature of the circumnuclear disk is that the average

472:   gas density profile, measured within spherical shells centered on

473:   the black hole, follows an almost perfect power-law with little

474:   evolution in time. Figure \ref{fig:dens} shows the gas density

475:   profile from two different snapshots of the simulation, as well as

476:   the average of the profile over a $\approx 550{,}000$ year

477:   period. Both the snapshots and the average profiles of the gas

478:   density increase by $\approx 8$ orders of magnitude in the inner

479:   $100$ pc of the simulation, obeying a steep power-law with slope

480:   $-8/3$. The stability of the gas density profile indicates that the

481:   disk is in a quasi-stationary state on timescales of several hundred

482:   thousand years. Figure \ref{fig:dens} also shows the dark matter and

483:   stellar mass density profiles. In the central kiloparsec of the

484:   galaxy, gas comprises $\sim 62 \%$ of the mass ($\sim 2.3 \times

485:   10^{10} h^{-1} \dim{M}_{\sun}$ total in the central kpc), dominating

486:   the dynamics of the disk. In contrast, the stellar and dark matter

487:   populations comprise $\sim 13$ and $\sim 25 \%$ of the galaxy's mass

488:   inside the central kiloparsec. The disk is extremely gas rich at

489:   $z=4$, because the galaxy is still actively growing and has not yet

490:   formed all of its stars. Interestingly, we find that both the

491:   stellar and dark matter profiles measured in the central $\sim 200$

492:   pc of the galaxy follow simple $\propto r^{-2}$ power-laws. The

493:   profiles match those predicted for the adiabatic contraction of dark

494:   matter from an NFW profile \citep{NFW97}, using the model of

495:   \citet{Gnedinetal04}, all the way down to the $1\dim{pc}$ scale.

496: 

497: %\clearpage

498:   \begin{figure}[t]

499: %    \centering \epsscale{1.} 

500:     \centering \epsscale{1.2} 

501:     \plotone{f5.eps}

502:     \caption{\label{fig:dens} Radial profiles of the gas, dark matter,

503:       and stellar mass density. Snapshots of the gas density are shown

504:       at $100{,}000$ and $500{,}000$ years after the initial

505:       refinement ({\it thin solid lines; green and red respectively,

506:       in the color version}), in addition to a $550{,}000$ year

507:       average ({\it thick solid line}) . The snapshots do not show

508:       much variation from the averaged profile, demonstrating that on

509:       timescales of several hundred thousand years, the disk is in a

510:       quasi-stationary state. The dotted line shows a power-law with

511:       slope $-8/3$ for comparison, which matches the gas density slope

512:       well. The dark matter and stellar mass density profiles are

513:       shown by the long-dashed ({\it blue in the color version}) and

514:       short-dashed ({\it magenta in the color version}) lines,

515:       respectively. In the inner $\sim 200$ pc, the slope of the dark

516:       matter and stellar density profiles is $\sim -2$ consistent with

517:       the adiabatic contraction model.}

518: 

519:   \end{figure}  

520: %\clearpage

521: 

522:   Figure \ref{fig:vcmp} shows that the RMS velocity dispersion of the

523:   gas (measured between neighboring cells, and therefore depending on

524:   the resolution) greatly exceeds the sound speed in the inner $100$

525:   pc, indicating supersonic turbulence in the disk. Supersonic

526:   turbulence decays on a dynamical time scale, so the persistence of

527:   the turbulence in the circumnuclear disk of the galaxy indicates a

528:   driving mechanism, which will be addressed in the following

529:   sections. The mean sound speed of the gas indicates a cold,

530:   molecular gas disk within the central $100 \dim{pc}$ of the

531:   simulated galaxy. The floor in the sound speed shown in Figure

532:   \ref{fig:vcmp} is determined by the minimum temperature to which our

533:   adopted cooling rates (computed using the CLOUDY package) apply. The

534:   mean sound speed is computed for cells with $T<20{,}000$ K, because

535:   the spherical averages shown in Figure \ref{fig:vcmp} include the

536:   low density gas, infalling perpendicular to the plane of the

537:   galactic disk, and shock-heated to high temperatures ($\gtrsim 10^6$

538:   K).  Additionally, turbulence inside the disk produces localized

539:   shocks that briefly heat individual cells to similarly high

540:   temperatures. The inclusion of shock-heated gas produces a broad

541:   temperature distribution and raises the mean temperature of the gas

542:   so that it does not effectively describe the typical sound speed of

543:   the gas in the disk. Therefore, the mean is computed over a

544:   temperature range that selects cells in the disk, while excluding

545:   atypical cells, resulting in a more representative quantity for

546:   describing the thermal properties of the disk.

547: 

548: %\clearpage  

549:   \begin{figure}[t]

550:     \centering

551: %    \epsscale{1.}

552:     \epsscale{1.2}

553:     \plotone{f6.eps}

554:     \caption{\label{fig:vcmp} A comparison of the radial profiles of

555:       the tangential velocity component of the gas ({\it solid}), its

556:       RMS velocity dispersion, or turbulent velocity ({\it dashed}),

557:       and the mean sound speed, computed in cells with $T<20{,}000$ K

558:       ({\it dotted}), time averaged over $550{,}000$ years. The sound

559:       speed is low, indicating cold, molecular gas within the central

560:       $20 \dim{pc}$.}

561:   \end{figure}    

562: %\clearpage

563: 

564:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

565:   \subsection{Angular Momentum Transport}

566:   \label{subsection:ang}

567: 

568:   The gas density profile of Figure \ref{fig:dens} motivates a simple

569:   analytic description of angular momentum transport in the

570:   disk. Averaged over a sufficiently long time, the disk can be

571:   considered to be approximately azimuthally symmetric. Additionally,

572:   the rotational velocity of the disk dominates over other velocity

573:   components: over the local velocity dispersion, and over the sound

574:   speed, as demonstrated in Figure \ref{fig:vcmp}. Therefore, the

575:   circumnuclear disk in the simulation is well approximated by a thin,

576:   rotationally supported, viscous, molecular gas disk. Conservation of

577:   angular momentum in such a disk is described by the following

578:   equation in cylindrical coordinates \citep{Pringle81}:

579: 

580:   \begin{equation}\label{eq:ang}

581:     \frac{\p}{\p t}(J_{\textrm{z}}) + \frac{1}{R}\frac{\p}{\p R}(R

582:     v_{\textrm{R}} J_z) = \frac{1}{2\pi R}\frac{\p G}{\p R},

583:   \end{equation}

584: 

585:   \noindent where the angular momentum density normal to the disk is

586:   given by $J_{\textrm{z}} = \Sigma R^2 \Omega$, where

587:   $\Omega=v_{\textrm{t}}/R$. The surface density of the gas is

588:   $\Sigma$, and $v_{\textrm{R}}$ is the radial component of velocity,

589:   which is small compared to the rotational velocity of the gas. The

590:   right hand side of equation (\ref{eq:ang}) describes the effects of

591:   a viscous torque, $G$, generated by turbulent motions of the

592:   gas. \citet{Pringle81} parameterizes the viscous torque $G$, which

593:   necessarily vanishes in solid body rotation, $\p \Omega / \p R = 0$,

594:   as

595: 

596:   \begin{equation}\label{eq:torqa}

597:     G(R,t) = 2 \pi \nu \Sigma R^3 \frac{\p \Omega}{\p R} 

598:   \end{equation}

599: 

600:   \noindent where $\nu$ is the coefficient of turbulent

601:   viscosity. During the initial zoom-in, the gas disk quickly reaches

602:   the power-law density profile shown in Figure \ref{fig:dens}. On

603:   short time scales of a few hundred thousand years, traced by a

604:   single high-resolution zoom-in episode of our simulation,

605:   time-averaged radial motions through the disk are small compared to

606:   the rotational motion, as shown in Figure \ref{fig:vel}. Over longer

607:   time scales, angular momentum is transported outward by a viscous

608:   torque driven by the turbulent motions of the gas, allowing

609:   accretion to slowly feed the black hole. The left hand side of

610:   equation (\ref{eq:ang}) can therefore be considered to be small on

611:   time scales of a few hundred thousand years. It then follows that

612:   the right hand side of equation (\ref{eq:ang}) should also be small,

613:   and that the torque $G$ is approximately constant with radius.

614: 

615:   Given that the gas density $\rho$ follows a power law (Figure

616:   \ref{fig:dens}), it is reasonable to approximate the surface density

617:   of the gas $\Sigma$ by a power law with the radius $R$:

618: 

619:   \begin{equation}\label{eq:surf}

620:     \Sigma \propto R^{-\beta}.

621:   \end{equation}

622: 

623:   \noindent Since the gravitational potential $\Phi$ satisfies

624:   $\nabla^2 \Phi = 4 \pi G \rho$ and $\Sigma = 2\rho R$, it follows

625:   that

626: 

627:   \begin{equation}\label{eq:phi}

628:     \Phi \propto R^{1-\beta}.

629:   \end{equation}

630: 

631:   \noindent Since the disk is almost entirely rotationally supported,

632:   the tangential velocity $v_{\textrm{t}}$ is effectively 

633:   the circular velocity $v_{\textrm{c}}$, defined by $v_{\textrm{c}}^2

634:   = R\ d\Phi/dR$. Thus, the angular speed is

635: 

636:   \begin{equation}

637:     \Omega = \frac{1}{R} \left(R \frac{d\Phi}{dR}\right)^{1/2} \propto

638:     R^{-(1+\beta)/2}, 

639:   \end{equation}

640: 

641:   \noindent and the viscous torque

642: 

643:   \begin{equation}\label{eq:torqb}

644:     G \propto R^{3(1-\beta)/2}\nu.

645:   \end{equation}

646: 

647:   \noindent If $\nu \propto R$ (which, as we will show in Section

648:   \ref{subsection:stab}, is the case in our simulations), then

649:   equation (\ref{eq:torqb}) implies a power-law slope of $\beta

650:   \approx 5/3$. It then follows that the volume density of the gas

651:   should scale as $\rho \propto R^{-8/3}$, consistent with the slope

652:   measured in Figure \ref{fig:dens}. Therefore, the simple description

653:   of angular momentum transport, maintained by turbulence in the disk,

654:   potentially provides a consistent interpretation of the power-law

655:   slope of the gas density profile of Figure \ref{fig:dens}.

656: 

657: 

658:   %%%%%%%%%%%%%%%%%%%%%%%%%%%

659: 

660:   \subsection{Disk Stability and the Source of Turbulence}

661:   \label{subsection:stab}

662: 

663:   In Section \ref{subsection:disk}, we showed that the simulated

664:   galaxy contains a self-gravitating, turbulent, cold, molecular gas

665:   disk within the central $\approx 100 \dim{pc}$. Such disks are

666:   susceptible to instabilities and to fragmentation. In the snapshots

667:   shown in Figure \ref{fig:evol}, the presence of instabilities is

668:   illustrated by waves moving through the simulated disk. The

669:   snapshots trace the evolution of disk structure on scales of

670:   $\approx 125$ pc, and $\approx 12.5$ pc (inset). The rotational

671:   period of the disk at the outer radius is such that the disk has

672:   undergone at least one full rotation (several more at smaller radii)

673:   over the time scales followed by the present simulation. The panels

674:   show the somewhat chaotic formation and re-formation of spiral

675:   structures on these scales, in contrast with the more ordered spiral

676:   structure seen on kiloparsec scales (Figure \ref{fig:pan}). This

677:   section includes an analysis of the behavior of structures in the

678:   disk on the scales shown in Figure \ref{fig:evol}.

679: 

680: %\clearpage

681:   \begin{figure*}

682:     \center{\includegraphics[width=1.9\columnwidth]{f7.eps}}

683:     \caption{\label{fig:evol} The evolution of global instabilities in

684:       the disk over time, on a scale of $\approx 125$ pc and on a scale

685:       of $\approx 12.5$ pc (inset). Each panel shows a volume rendering

686:       of the gas density (in $\dim{cm}^{-3})$ at a different

687:       epoch. The similarities between the appearance of the disk on

688:       the two different scales indicate the hierarchical structure of

689:       the disk. (This figure is best viewed in color.)} 

690:   \end{figure*}

691: %\clearpage

692: 

693:   The Toomre $Q$ parameter describes the stability of the disk against

694:   linear perturbations, and is given by $Q = \sigma_{\textrm{turb}}

695:   \kappa / \pi G\Sigma$, where $\kappa$ is the epicyclic frequency,

696:   and $\Sigma$ is the surface density of the gas, both determined

697:   locally \citep{Toomre64,GoldLB65}. Although the disk is cold, and

698:   the sound speed is low, the RMS velocity dispersion is substantially

699:   higher than the sound speed inside $100$ pc, indicating that the gas

700:   is turbulent. Therefore, in place of the sound speed typically used

701:   in the definition of the Toomre $Q$ parameter, we have substituted a

702:   turbulent velocity, given by the quadrature sum of the sound speed

703:   and the RMS velocity dispersion of the gas. The disk is, for the

704:   most part, quasi-Keplerian (as described in Section

705:   \ref{subsection:disk}), so that the epicyclic frequency $\kappa$ is

706:   proportional to the angular speed of the disk,

707:   $\Omega=v_{\textrm{t}}/r$.

708: 

709:   The top panel of Figure \ref{fig:Q} shows the Toomre $Q$ parameter

710:   for the simulated disk, corresponding to the region shown in Figure

711:   \ref{fig:evol}. Where $Q<1$, the disk is susceptible to local

712:   gravitational instabilities resulting from axisymmetric

713:   perturbations. For $Q>1$ the disk is likely to be stable against

714:   axisymmetric perturbations, but is still susceptible to

715:   instabilities arising from non-axisymmetric perturbations, which are

716:   less stable than radial perturbations \citep{Polyachenko}. For the

717:   density profile shown in Figure \ref{fig:dens}, the disk becomes

718:   stable for $Q\gtrsim2$. The shaded region in Figure \ref{fig:Q}

719:   ($1<Q<2$) therefore represents marginally stable values of the $Q$

720:   parameter, where the disk still might become unstable. Figure

721:   \ref{fig:Q} suggests that the disk in the simulated galaxy lies

722:   mostly in the region of marginal stability inside $\approx

723:   1\dim{kpc}$. In the case of axisymmetric perturbations, the fastest

724:   growing unstable mode corresponds to the scale

725:   $\lambda_{\textrm{fast,r}}$, at which $d\omega^2 / dk = 0$ (where

726:   $\omega$ is the angular frequency of waves in the disk), given by

727: 

728:   \begin{equation}\label{eq:fastr}

729:     \lambda_{\textrm{fast,r}} = \frac{2 \sigma_{\textrm{turb}}^2}{G

730:     \Sigma}.

731:   \end{equation}

732: 

733:   \noindent Using the condition for marginal stability from

734:   \citet{Polyachenko}, the fastest growing mode for all modes

735:   (axisymmetric and non-axisymmetric), $\lambda_{\textrm{fast,all}}$,

736:   is given by 

737: 

738:   \begin{equation}\label{eq:fasta}

739:     \lambda_{\textrm{fast,all}} = \frac{\lambda_{\textrm{fast,r}}}{2}

740:       = \frac{ \sigma_{\textrm{turb}}^2}{G \Sigma}.

741:   \end{equation}

742: 

743:   \noindent The bottom panel of Figure \ref{fig:Q} shows the ratio

744:   $\lambda / r$ for each of the fastest growing modes. In the inner

745:   $100$ pc, the scales of the fastest growing modes are smaller than

746:   the radius by only a factor of a few, implying that the disk is

747:   stable on scales $\lambda \ll R$ (at least in the linear

748:   regime). This is an indication that perturbations in the disk

749:   operate on a range of scales, but always on scales that are an

750:   appreciable fraction of the size of the system, driving global

751:   instabilities, which generate turbulence on smaller scales. The disk

752:   remains locally stable all the way into a region less than $1$ pc

753:   from the black hole. There is no catastrophic fragmentation into

754:   clumps small enough to form stars, so that angular momentum

755:   transport may continue uninterrupted by bursts of star formation.

756: 

757: %\clearpage

758:   \begin{figure}[t]

759: %    \epsscale{1.}

760:     \epsscale{1.2}

761:     \centering

762:     \plotone{f8.eps}

763:     \caption{\label{fig:Q} {\it Top:} Toomre $Q$ parameter. For $Q<1$

764:       the disk is unstable, and for $Q\gtrsim2$ the disk is stable. In

765:       the regime $1<Q<2$, the disk is marginally stable (stable

766:       against axisymmetric modes, but not non-axisymmetric

767:       modes). {\it Bottom:} Average values of the the fastest growing

768:       unstable mode for axisymmetric perturbations,

769:       $\lambda_{\textrm{fast,r}}$ ({\it solid}), and the fastest

770:       growing unstable mode for all perturbations,

771:       $\lambda_{\textrm{fast,all}}$ ({\it dashed}), each divided by

772:       radius. The ratio $\lambda/r \gtrsim 0.1$ throughout the

773:       circumnuclear disk, so there is no catastrophic fragmentation

774:       down to small scales.}

775:   \end{figure}

776: %\clearpage

777: 

778:   By dimensional analysis, the turbulent kinematic viscosity $\nu$

779:   discussed in Section \ref{subsection:ang} can be described by

780:   $\lambda \sigma_{\textrm{turb}}$, where $\lambda$ is a

781:   characteristic scale for turbulence and $\sigma_{\textrm{turb}}$,

782:   the turbulent velocity shown in Figure \ref{fig:vcmp}, is a

783:   characteristic velocity. It is natural to identify the

784:   characteristic length scale with the length scale corresponding to

785:   the fastest growing unstable mode, $\lambda_{\textrm{fast,r}}$, so

786:   that $\nu \sim \lambda_{\textrm{fast,r}}

787:   \sigma_{\textrm{turb}}$. The top panel of Figure \ref{fig:visc}

788:   shows that the turbulent viscosity $\nu$, given approximately by

789:   $\lambda_{\textrm{fast,r}} \sigma_{\textrm{turb}}$, scales linearly

790:   with the radius over much of the disk, so that $\nu \propto

791:   R$. Although the behavior of the turbulent viscosity $\nu$ can

792:   potentially change with time, the viscous torque $G$ should remain

793:   approximately constant in radius for the above interpretation to be

794:   valid. The bottom panel of Figure \ref{fig:visc} shows the average

795:   viscous torque $G$, normalized to its value at $10$ pc. The

796:   normalized value is close to unity over most of the circumnuclear

797:   region, which is indeed the condition needed to explain the $-8/3$

798:   slope of the spherically-averaged gas density profile of Figure

799:   \ref{fig:dens}.

800: 

801: %\clearpage

802:   \begin{figure}[t] 

803:     \centering

804: %    \epsscale{1.}

805:     \epsscale{1.2}

806:     \plotone{f9.eps}

807:     \caption{\label{fig:visc} {\it Top:} Ratio of kinematic viscosity

808:     $\nu$, given approximately by $\lambda_{\textrm{fast,r}}

809:     \sigma_{\textrm{turb}}$, to radius $R$. The ratio $\nu / R$ is

810:     roughly constant in radius. {\it Bottom:} The viscous torque $G$

811:     normalized to its $10$ pc value. The viscous torque is constant in

812:     radius, consistent with the power-law gas density slope $\rho

813:     \propto R^{-8/3}$.}

814:   \end{figure} 

815: %\clearpage

816: 

817:   \noindent Thus, a simple description of the structure of the

818:   circumnuclear disk based on equations (\ref{eq:ang}-\ref{eq:torqb})

819:   and Fig.\ \ref{fig:visc} provides a good match to the simulation

820:   results, supporting the idea that global instabilities in the disk

821:   are responsible for generating turbulence in the gas, resulting in

822:   the power-law slope described in Sections \ref{subsection:disk} and

823:   \ref{subsection:ang}.

824: 

825:   The conditions in the circumnuclear disk are potentially conducive

826:   to gaseous bar formation, the conditions for which have been studied

827:   extensively for different geometries and physical conditions,

828:   resulting in a variety of criteria

829:   \citep[e.g.][]{OP73,Efstat82,Christo95_1,Christo95_2,Bottema03,Wyse04}.

830:   A comparison with the criterion of \citet{OP73}, which characterizes

831:   stability in terms of the ratio $t=T/|W|$ of kinetic to

832:   gravitational potential energy, indicates that the disk is both

833:   secularly and dynamically unstable to bar formation on all scales $r

834:   \gtrsim 0.1 \dim{pc}$ adequately resolved by the simulation. Because

835:   of the chaotic and transient behavior of the instabilities, the

836:   structures shown in Figure \ref{fig:evol} do not resemble a single,

837:   well defined bar, but rather a highly perturbed ``bar.''

838: 

839:   Since the Toomre criterion applies only to small linear

840:   perturbations and the Ostriker-Peebles criterion describes global

841:   modes, the question remains whether non-linear effects lead to

842:   fragmentation on smaller scales, below the resolution of the

843:   simulation. The top panel of Figure \ref{fig:jeans} shows the ratio

844:   of the local Jeans length of the gas, $\lambda_{\textrm{J}}$, to

845:   cell size, $\Delta x$, for all simulation cells in the circumnuclear

846:   disk (with temperatures less than $10^3$ K). In most cells, the

847:   Jeans length is larger than the cell size, preventing the gas from

848:   fragmenting into sub-cell sized clouds, ultimately leading to star

849:   formation. The Truelove criterion for preventing numerical

850:   fragmentation is enforced on the maximum level of refinement (level

851:   $20$) only, so there are no numerical restrictions to prevent gas

852:   from collapsing all the way to level $20$. However, only the central

853:   sub-pc part of the disk reaches levels $19$ and $20$, as shown by

854:   the histogram in the lower panel of Figure \ref{fig:jeans},

855:   indicating that the disk is stable to non-linear effects. While some

856:   cells in the disk have $\lambda_{\textrm{J}} < \Delta x$ and may

857:   form stars, Figure \ref{fig:jeans} demonstrates that most of the

858:   disk is stable against collapse, and that there is no widespread

859:   burst of star formation. The few cells with

860:   $\lambda_{\textrm{J}}/\Delta x < 1$ may correspond to resonances

861:   where the pattern speed of waves in the disk is the same as the

862:   rotational velocity, leading to higher gas densities, and possible

863:   star formation. However, a detailed analysis of the behavior of

864:   these resonances with time is beyond the scope of this paper.

865: 

866: %\clearpage  

867:   \begin{figure}[t] 

868:     \centering

869: %    \epsscale{1.}

870:     \epsscale{1.2}

871:     \plotone{f10.eps}

872:     \caption{\label{fig:jeans} {\it Top:} Scatter plot showing the

873:       ratio of the local Jeans length, $\lambda_{\textrm{J}}$, to the

874:       cell size, $\Delta x$ at $300{,}000$ years after the initial

875:       refinement. Only cool cells with temperatures less than $10^3$ K

876:       are shown. {\it Bottom:} Histogram showing the volume of level

877:       $20$ and of level $19$ cells as a function of radius. The

878:       simulation only refines to the maximum levels in the center of

879:       the circumnuclear disk.}

880:   \end{figure}  

881: %\clearpage

882: 

883:   %%%%%%%%%%%%%%%%%%%%%%%%%%%

884: 

885:   \subsection{The Nature of the Turbulence}

886:   \label{subsection:turb}

887: 

888:   Turbulence has been widely studied in modeling of the ISMs of

889:   galaxies and star forming regions \citep[for recent reviews,

890:   see][]{MacLowKlessen04,McKeeOst07}. It is important to understand

891:   the nature of the turbulence in the present simulated galaxy as it

892:   plays a key role in the disk properties. The global instabilities

893:   arising in the simulated circumnuclear disk generate turbulence on a

894:   range of scales, maintaining the quasi-steady state of the disk.

895: 

896:   Turbulence dissipates energy in such a way that the turbulent

897:   velocity scales as $\sigma_{\textrm{turb}} \propto \lambda^q$, where

898:   $q=1/3$ for incompressible, sub-sonic (Kolmogorov) turbulence and

899:   $q=1/2$ for compressible turbulence in the zero-pressure limit

900:   (Burgers turbulence). The supersonic turbulence in the simulated

901:   disk falls between these two limits. In AMR simulations, it is

902:   straightforward to measure the turbulent velocity on different

903:   scales because the information on different levels of refinement is

904:   readily available. However, the measurements are only accurate to

905:   within a factor of two in scale because the cell size decreases by a

906:   factor of two with each refinement. The approximate scaling of the

907:   turbulence is shown in the top panel of Figure \ref{fig:spec}. The

908:   figure shows the turbulent velocity (given by the RMS velocity

909:   dispersion between neighboring cells) at two different radii for two

910:   different simulation runs. The first is the fiducial run described

911:   in the previous sections, and the second is a short portion of the

912:   fiducial run, with a more aggressive refinement criterion on levels

913:   $13$ and below, given by $m_{\textrm{r}}^{\textrm{level}-10}

914:   \textrm{max}[0.5^{\textrm{level}-12},\ 0.125]$ (where

915:   $m_{\textrm{r}}$ is an empirically determined parameter defined in

916:   Section \ref{subsection:ref}). The more aggressive run probes

917:   smaller scales at a given radius, in order to capture the scale of

918:   the turbulence. The turbulent velocities shown in Figure

919:   \ref{fig:spec} demonstrate the scaling of the turbulence down to

920:   small scales. It is difficult to determine the precise slope of the

921:   scaling using measurements from the present simulation because of

922:   the limitation imposed by the refinement scheme. While a more

923:   precise characterization of the turbulence spectrum is beyond the

924:   scope of the present simulation, Figure \ref{fig:spec} shows that

925:   the slope approximately falls between the Kolmogorov and Burgers

926:   turbulence limits, as expected.

927: 

928: %\clearpage

929:   \begin{figure}[t] 

930:     \centering

931:     \epsscale{1.}

932:     \plotone{f11.eps}

933:     \caption{\label{fig:spec} {\it Top:} Turbulent velocity as a

934:     function of scale (or cell size) at $3$ and $8$ pc, for the

935:     fiducial and aggressive refinement runs. The more aggressive

936:     refinement run resolves the scaling of the turbulent down to

937:     smaller scales. {\it Bottom:} A comparison of the smallest

938:     resolved scale of turbulence and the fastest growing unstable

939:     mode, $\lambda_{\textrm{fast,all}}$ (where

940:     $\lambda_{\textrm{fast,all}}$ is the same as in Figure

941:     \ref{fig:Q}).}

942:   \end{figure}  

943: %\clearpage

944: 

945:   At scales below the resolution at a given location in the disk, the

946:   turbulent velocities in Figure \ref{fig:spec} level off at constant

947:   values because the turbulence scaling can only be measured down to

948:   the resolution limit. The more aggressive refinement run ({\it

949:   triangles} in Figure \ref{fig:spec}) probes smaller scales for a

950:   given radius, and therefore levels off at smaller scales. The bottom

951:   panel of Figure \ref{fig:spec} shows the approximate scale at which

952:   the flattening of the spectrum occurs for several different radii in

953:   the circumnuclear disk of the fiducial run. For comparison, the

954:   fastest growing unstable mode, $\lambda_{\textrm{fast,all}}$ is

955:   shown as well. The turbulent velocity measured at the resolution of

956:   the simulation is a numerical quantity, but it roughly corresponds

957:   to the turbulence on the scale of $\lambda_{\textrm{fast,r}}$, which

958:   is a physical quantity. Figure \ref{fig:spec} demonstrates that the

959:   simulation resolves all scales that are Toomre $Q$ unstable,

960:   supporting the resulting interpretation that instability driven

961:   turbulence maintains the quasi-stationary state of the circumnuclear

962:   disk.

963: 

964:   The density structure of the disk is also consistent with the

965:   description of turbulence in the disk. Supersonic turbulence

966:   typically imposes a log-normal density distribution on the gas, as

967:   demonstrated by models of turbulence in molecular clouds

968:   \citep[e.g.][]{Vazquez94, PassotVaz98, Wada01, Kritsuketal07,

969:   McKeeOst07}, such as

970: 

971:   \begin{equation}\label{eq:pdf}

972:     \frac{dV}{d\rho} = \frac{1}{\rho \sqrt{2\pi\sigma^2}} \

973:     \textrm{exp}\left[\frac{-(\textrm{ln} \rho - \mu)^2}{2

974:     \sigma^2}\right],

975:   \end{equation}

976: 

977:   \noindent where the mean, $\mu$, is defined as

978:   $\overline{\textrm{ln}\ \rho}$, and $\sigma$ is the

979:   dispersion. Figure \ref{fig:pdf} shows the volume-weighted

980:   probability distribution function (PDF), time-averaged over $\sim

981:   250,000$ years, for a shell of gas at a radius of $30$ pc. The PDF is

982:   well fit by a log-normal distribution over at least $4$ orders of

983:   magnitude in density, consistent with models of supersonic turbulence.

984: 

985: %\clearpage  

986:   \begin{figure}[t]

987:     \centering

988: %    \epsscale{1.}

989:     \epsscale{1.2}

990:     \plotone{f12.eps}

991:     \caption{\label{fig:pdf} A volume-weighted PDF of the gas density

992:       in cells at a scale of $\sim 30$ pc, with temperatures $< 10^3$

993:       K (in order to exclude the hot corona outside the disk),

994:       averaged over $\sim 550,000$ years. The PDF is well fit by a

995:       log-normal distribution over the range $10^2 \lesssim \rho

996:       \lesssim 10^6 \dim{M}_{\sun} \dim{pc}^{-3}$ with a mean density

997:       of $10^{3.9} \dim{M}_{\sun} \dim{pc}^{-3}$ ({\it line}).}

998:   \end{figure}    

999: %\clearpage

1000: 

1001: 

1002:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1003: 

1004:   \section{Possible Effects of Missing Physics}

1005:   \label{sec:miss}

1006: 

1007:   The simulations presented here do not include all of the potentially

1008:   important physics for galaxy evolution. Therefore, in this section

1009:   we discuss the potential effect that the physics missing from the

1010:   simulations might have on our results.

1011: 

1012:   \subsection{Optically Thick Cooling}

1013:   %  \label{subsubsection:torus}

1014:   \label{subsection:torus}

1015: 

1016:   Cooling in the high density, central region of the galaxy is not

1017:   treated entirely correctly by the ART code. The column density of

1018:   the molecular gas in the center is so high that the gas is expected

1019:   to become optically thick to its own cooling radiation

1020:   \citep[e.g.][]{RipAbel04}. Additionally, the presence of dust grains

1021:   in this region traps radiation and halts cooling further

1022:   \citep[see][]{DraineLee84, Ossenkopf94}. Figure \ref{fig:col} shows

1023:   the column density vertically through the circumnuclear gas

1024:   disk. Horizontal lines are shown for comparison, corresponding to

1025:   the column densities at which dust and H$_2$ each become optically

1026:   thick to their own cooling radiation ($\sim 10^{26} \dim{cm}^{-2}$

1027:   and $\sim 10^{27} \dim{cm}^{-2}$, respectively, at $0.1$ solar

1028:   metallicity, which is the metallicity of the circumnuclear disk in

1029:   the simulation at $z=4$). In the inner $\approx 10$ pc, the column

1030:   density of the gas is large enough that the opacity of dust and

1031:   molecular gas must be accounted for to accurately describe cooling

1032:   in the simulation.

1033: 

1034: %\clearpage

1035:   \begin{figure}[t] 

1036:     \centering \epsscale{1.2} \plotone{f13.eps}

1037:     \caption{\label{fig:col} Column density of the gas. The horizontal

1038:     lines show the column densities at which dust and H$_2$ become

1039:     optically thick to their own cooling radiation.}

1040:   \end{figure}  

1041: %\clearpage

1042: 

1043:   The simulation runs presented here do not include optically thick

1044:   radiative transfer. This may explain why the gas remains in a thin

1045:   disk all the way in to sub-pc scales, and does not resemble the

1046:   obscuring tori observed in the inner few parsecs of AGN. Should the

1047:   opacity of the gas to its own cooling radiation be included, the gas

1048:   around the mid-plane of the disk would not be able to cool and would

1049:   heat to temperatures corresponding to the energy dumped into the gas

1050:   by the turbulence. These higher temperatures may be able to provide

1051:   a substantial vertical pressure support in the central few parsecs

1052:   of the disk, resulting in a thicker, more toroidal-like

1053:   structure\footnote{Notice, that the outer, optically thin layers of

1054:   such a disk would continue cooling efficiently, covering the hot

1055:   interior with a cold molecular ``skin''. Such a skin may become

1056:   Rayleigh-Taylor unstable, leading to fragmentation of the torus into

1057:   individual clouds.}. However, it has been suggested that

1058:   hydromagnetic disk winds, and not hydrostatic pressure support, may

1059:   be responsible for sustaining the optically and geometrically thick

1060:   obscuration region, or ``obscuring torus,'' in the nuclei of

1061:   galaxies \citep[e.g.][]{KonigKartje94,ElitzShlos06}. In which case,

1062:   the inclusion of optically thick radiative transfer may not be

1063:   sufficient to create a ``torus'' in our simulations.

1064: 

1065:   In future simulations we plan to remedy this limitation of the

1066:   present simulation by accounting for the opacity of the high 

1067:   density region in the center, which will allow us to consistently

1068:   incorporate radiative feedback from a central source and to test the

1069:   above hypothesis.

1070: 

1071:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1072:   \subsection{Magnetic Fields}

1073:   \label{subsection:mag}

1074: 

1075:   The ART code does not include magnetic fields, whereas MHD is

1076:   certainly important for accretion disk physics \citep[see][and

1077:   references therein]{BalbusHawley98}. It is for this reason that the

1078:   present study has been restricted to scales larger than $\approx 0.1

1079:   \dim{pc}$, corresponding to about $10^4$ Schwarzschild radii for a

1080:   black hole of mass $3 \times 10^7$ M$_{\sun}$.

1081: 

1082:   At scales larger than the accretion disk, the absence of magnetic

1083:   fields in the ART code is probably not important. An estimate of the

1084:   strength that an equipartition magnetic field would need in order to

1085:   affect the disk on small scales is given by $B_{\textrm{eq}}^2

1086:   \approx 4\pi \rho \sigma_{\textrm{turb}}^2$. In order to affect the

1087:   large-scale dynamics of the gas, the magnetic field would have to

1088:   have an even larger strength of order $B_{\textrm{dyn}}^2 \approx

1089:   4\pi \rho v_{\textrm{t}}^2$. Figure \ref{fig:B} shows estimates of

1090:   $B_{\textrm{eq}}$ and $B_{\textrm{dyn}}$, in order to demonstrate

1091:   how high the magnetic field would need to be to significantly

1092:   influence gas dynamics in the simulated galaxy. For most of the

1093:   galactic disk, the above estimates for the magnetic field are far

1094:   higher than the few $\mu$G fields observed in real galaxies

1095:   \citep[e.g.][]{ZweibHeil97,Beck01}. Even in the sub-pc region, where

1096:   water maser observations indicate stronger fields of a few tens of

1097:   mG \citep[e.g.][]{Modjazetal05,Vlemmetal07}, the magnetic field is

1098:   still too low to affect the gas dynamics in the model galaxy. We

1099:   therefore conclude that magnetic fields, which are not included in

1100:   our simulations, will have a negligible effect on the dynamical

1101:   state of this disk unless their strength greatly exceeds

1102:   observational measurements.

1103: 

1104: %\clearpage

1105:   \begin{figure}[t] 

1106:     \centering

1107:     \epsscale{1.2}

1108:     \plotone{f14.eps}

1109:     \caption{\label{fig:B} Equipartition ({\it solid}) and dynamical

1110:     ({\it dashed}) magnetic fields needed to influence the gas

1111:     dynamics in the disk.}

1112:   \end{figure}  

1113: %\clearpage

1114: 

1115:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1116:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1117: 

1118:   \section{DISCUSSION AND CONCLUSIONS}

1119:   \label{sec:disc}

1120: 

1121:   We have used a large-dynamic range cosmological simulation to study

1122:   gas dynamics in the circumnuclear disk of typical mass galaxy at

1123:   $z\approx4$ (evolving into an $L_*$ galaxy at $z=0$), resolving the

1124:   distribution of matter from megaparsec scales all the way down to

1125:   sub-pc scales (with $20$ levels of refinement). The simulation

1126:   reveals a cold, fully molecular, self-gravitating, and turbulent

1127:   rotationally-supported gas disk, which is globally unstable but

1128:   locally stable.

1129: 

1130:   The global instability, operating on a range of scales comparable to

1131:   the size of the system, generates turbulence down to the smallest

1132:   scale resolved in the simulation. On small scales, the turbulence

1133:   supports most of the disk against gravitational fragmentation and

1134:   collapse. The disk, therefore, remains locally stable and reaches a

1135:   quasi-stationary state. In that state, global instability drives

1136:   bar-like and spiral-wave-like structures on time scales of the order

1137:   of $100{,}000$ years at $100\dim{pc}$ scales (and on proportionally

1138:   shorter time scales at smaller radii), but on a $500{,}000$ year

1139:   time-scale the structure of the disk remains quasi-steady.

1140: 

1141:   The disk develops a power-law gas density profile with a well

1142:   defined slope of $-5/3$ in surface density and $-8/3$ in

1143:   spherically-averaged volume density. This slope is a natural

1144:   consequence of the dynamical state of the disk: the turbulence,

1145:   generated by the global instability, redistributes the angular

1146:   momentum in the disk on a dynamical time scale, reducing the

1147:   gradient of the viscous torque. The resulting (quasi-)steady state

1148:   uniquely determines the gas density profile we find in the

1149:   simulation and the distribution of the angular momentum with gas

1150:   mass.

1151: 

1152:   The turbulence in the disk drives the local outward transport of

1153:   angular momentum and the inward flow of gas toward the supermassive

1154:   black hole on time scales of at least 10 million years, which are

1155:   too long to follow in a single zoom-in episode of the

1156:   simulation. Thus, we only capture a single snapshot in the

1157:   cosmological life of the supermassive black hole. In follow-up

1158:   work, we plan to consider several snapshots taken at different

1159:   cosmological times in order to describe the system on longer time

1160:   scales.

1161: 

1162:   The dynamical state of the disk that we find in our simulation

1163:   appears to be consistent with the results of previous simulations of

1164:   isolated circumnuclear disks in galaxies \citep{Fukuda00,

1165:   Wada01, WadaNorman01, Escala07}. A distinctive feature of our

1166:   approach is that we follow the dynamics of the circumnuclear disk

1167:   within cosmological simulations. While most of the volume of the

1168:   simulation evolves little on time scales relevant for the dynamics

1169:   of the circumnuclear disk, cosmological scales provide realistic

1170:   boundary conditions for the dynamics on sub-kpc scales. Since we

1171:   find that the circumnuclear disk rapidly reaches a quasi-stationary

1172:   state, its evolution is entirely governed by the boundary

1173:   conditions. For the same reason, the fact that the cosmological

1174:   simulation that was used for the initial conditions did not resolve

1175:   the scale of the circumnuclear disk, does not compromise our

1176:   results: the quasi-stationary state of the disk does not depend on

1177:   the initial conditions, and so the lack of power on scales below

1178:   about $100\dim{pc}$ in the initial cosmological simulation is not

1179:   important.

1180: 

1181:   The adopted approach of this work is comparable to the recent study

1182:   by \citet{Mayeretal07}, who used a cosmological simulation to model the

1183:   coalescence of two supermassive black holes. While the detailed

1184:   treatment of gas physics and the scientific questions answered by

1185:   the two studies are different, many of our results are consistent

1186:   with those of \citet{Mayeretal07}. For example, they find a similar

1187:   slope for the spherically averaged density profile, although they do

1188:   not elaborate on the physical mechanisms for the formation of this

1189:   profile. 

1190: 

1191:   The local stability of the disk, supported by highly supersonic

1192:   turbulence, implies that the disk is capable of continuously feeding

1193:   a central black hole, uninterrupted by catastrophic bursts of star

1194:   formation, which could have consumed the available fuel were the

1195:   disk locally unstable. This interesting dynamical state of the disk

1196:   provides a potential solution to the problem of how AGN fueling is

1197:   maintained by self-gravitating gas disks \citep[for a related

1198:   discussion, see e.g.][]{ShlosBeg89,Riceetal05,NayakshinKing07}.

1199: 

1200:   The circumnuclear disk in the simulation extends all the way to the

1201:   maximally resolved scale of $\approx0.1\dim{pc}$, which corresponds

1202:   to the outer part of the black hole accretion disk. We find no

1203:   toroidal-like structures on several parsec scales, which are

1204:   commonly inferred to exist around AGN. A possible reason that the

1205:   disk does not form an AGN torus on the appropriate scales (if it

1206:   indeed, should) is the limitation imposed by our implementation of

1207:   gas cooling in the code. The ART code, like all existing

1208:   cosmological codes, assumes that the cooling radiation from cosmic

1209:   gas escapes freely into the IGM. This assumption breaks down at the

1210:   densities and temperatures reached by the simulation in the inner

1211:   $10\dim{pc}$. At this scale the disk becomes optically thick to its

1212:   own cooling radiation from dust and molecular hydrogen. In that

1213:   regime the disk may to heat up and acquire a substantial

1214:   amount of pressure support, which may result in a puffier, more

1215:   toroidal-like configuration for the inner several parsecs of the

1216:   disk. A consistent treatment of a putative AGN torus will require

1217:   simulations that incorporate optically-thick cooling.

1218: 

1219:   \acknowledgements

1220: 

1221:   The authors thank Oleg Gnedin for his analysis of the mass profiles

1222:   from the simulations. The authors also thank Mitch Begelman, Moshe

1223:   Elitzur, Lucio Mayer, Brant Robertson, Isaac Shlosman, Volker

1224:   Springel, Marta Volonteri, \& Keiichi Wada for comments on the

1225:   paper. This work was supported in part by the DOE and the NASA grant

1226:   NAG 5-10842 at Fermilab and by the NSF grants AST-0134373 and

1227:   AST-0507596. Supercomputer simulations were run on the IBM P690

1228:   array at the National Center for Supercomputing Applications and San

1229:   Diego Supercomputing Center (under grant AST-020018N).

1230: 

1231: 

1232:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1233:   

1234:   \bibliography{ms}

1235: 

1236:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1237: 

1238:   \appendix

1239: 

1240:   \section{Measuring the Viscosity}

1241:   \label{sec:mvisc}

1242: 

1243:   Here we address the validity of our estimate of the turbulent

1244:   kinematic velocity parameter $\nu$, which motivates the

1245:   interpretation that turbulent transport of angular momentum

1246:   maintains local stability and the quasi-steady state of the

1247:   circumnuclear disk. Instead of a direct measurement of the

1248:   viscosity, we have assumed that the viscosity depends on

1249:   characteristic velocity and length scales, given by

1250:   $\sigma_{\textrm{turb}}$ and $\lambda_{\textrm{fast,r}}$,

1251:   respectively. The accuracy of this estimate can be tested by

1252:   measuring the transport of gas due to turbulent motions.

1253: 

1254:   The ART code does not follow individual parcels of gas across cell

1255:   boundaries, but rather advected quantities describing the

1256:   gas. Therefore, in order to follow the turbulent motions of the gas,

1257:   we introduce a passive scalar into the simulation which has a value

1258:   of unity inside a spherical region centered on the black hole

1259:   particle, and a value of zero outside this region, effectively

1260:   ``painting'' the gas. As the simulation evolves, turbulent motions

1261:   of the gas cause the profile of the scalar quantity to deviate from

1262:   its original spherical form. The evolution of the profile depends on

1263:   the turbulent kinematic viscosity, $\nu$, and in the early stages of

1264:   diffusion, while the deviations from the initial profile are small,

1265:   the profile approximately satisfies the linear diffusion equation

1266:   for initial conditions,

1267: 

1268:   \begin{equation}

1269:     P(r,t=0) = 

1270:     \begin{cases} 

1271:       1,  & \mbox{if }r<r_0, \\

1272:       0,  & \mbox{if }r>r_0, 

1273:     \end{cases}

1274:   \end{equation}

1275: 

1276:   \noindent (where $r_0$ is the radius of the initial sphere) allowing

1277:   a more direct calculation of the viscosity. The solution to the

1278:   linear diffusion equation for the above initial conditions, is

1279: 

1280:   \begin{equation}\label{eq:prof}

1281:     P(x,q) = \frac{1}{2}\left[ \textrm{erf}\left(\frac{x+1}{\sqrt{q}}\right) -

1282:     \textrm{erf}\left(\frac{x-1}{\sqrt{q}}\right)\right] - \frac{q}{x \sqrt{\pi

1283:     q}}\ e^{-(1+x^2)/q}\ \sinh\left(\frac{2x}{q}\right),

1284:   \end{equation}

1285: 

1286:   \noindent where

1287: 

1288:   \begin{equation}

1289:     x = \frac{r}{r_0}\ \ \&\ \ q = \frac{4 \nu t}{r_0^2}.

1290:   \end{equation}

1291: 

1292:   Figure \ref{fig:paint} shows samples of the profile evolving from a

1293:   $30$ pc sphere for different times ($q$). The measured profiles

1294:   agree well with the solution given in Equation

1295:   \ref{eq:prof}. Starting from several painted regions at $1, 3, 10,

1296:   30, \textrm{and}\ 100$ pc radii, we followed the evolution of the

1297:   paint-weighted density profiles and measured the turbulent

1298:   viscosity, $\nu$, as a function of radius. Figure \ref{fig:vpaint}

1299:   shows the best-fit measurements of $\nu/r$ in relation to the

1300:   estimate given by $\sigma_{\textrm{turb}} \lambda_{\textrm{fast,r}} /

1301:   r$. The lines in Figure \ref{fig:vpaint} show individual snapshots

1302:   of $\sigma_{\textrm{turb}} \lambda_{\textrm{fast,r}} / r$

1303:   corresponding to the timescale over which the diffusion was followed

1304:   (whereas Figure \ref{fig:visc} showed a time-averaged estimate).

1305:   The best-fit measurements match the estimates for $\nu/r$ well, thus

1306:   lending justification to the estimate for $\nu$ used in the analytic

1307:   arguments of Section \ref{subsection:stab}.

1308: 

1309: %\clearpage

1310:   \begin{figure}[t] 

1311:     \centering

1312:     \epsscale{1.}

1313:     \plotone{fa1.eps}

1314:     \caption{\label{fig:paint} Mass-weighted profiles of the passive

1315:     scalar as a function of radius, and evolving over time from a $30$

1316:     pc sphere. The symbols are measurements from the simulation, and

1317:     the lines are the best fits to Equation \ref{eq:prof}.}

1318:   \end{figure}  

1319: %\clearpage

1320: 

1321: %\clearpage

1322:   \begin{figure}[t] 

1323:     \centering

1324:     \epsscale{1.}

1325:     \plotone{fa2.eps}

1326:     \caption{\label{fig:vpaint} Ratio of kinematic viscosity $\nu$, to

1327:     radius $R$. The lines show estimates for $\nu/R$ given

1328:     approximately by $\lambda_{\textrm{fast,r}}

1329:     \sigma_{\textrm{turb}}$, for three different snapshots. The points

1330:     are estimates from a second method measuring the advection of gas

1331:     at five different radii in the simulation. The snapshots

1332:     correspond to the profile fits at different times, so the $1$ pc

1333:     point should be compared to the solid line, the next point to the

1334:     dashed line, and the last three to the dotted line.}

1335:   \end{figure}  

1336: %\clearpage

1337: 

1338:   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1339: 

1340:   \section{Angular Momentum Conservation}

1341:   \label{sec:con}

1342: 

1343:   Anomalous numerical transport of angular momentum is sometimes

1344:   presented as a concern for adaptive mesh refinement simulations

1345:   using interpolation schemes for calculating velocities on the

1346:   mesh. We have tested conservation of angular momentum in the code by

1347:   conducting a separate set of simulations of the collapse of an

1348:   isothermal sphere (the details are included in the PhD thesis of the

1349:   author Levine). The test was conducted for a mesh which refined

1350:   along with the collapsing sphere, and for a pre-refined mesh, in

1351:   order to test the conservation of angular momentum both as the mesh

1352:   refines and as gas moves through the mesh. The sphere collapses into

1353:   a thin disk ($h/r << 1$), which conserves angular momentum over

1354:   several rotation periods.

1355: 

1356:   In Figure \ref{fig:jvm} we demonstrate the conservation of angular

1357:   momentum in the present simulation. As the maximum resolution of the

1358:   simulation increases, the angular momentum profile initially

1359:   evolves, as the viscous torque, $G$, described in Section

1360:   \ref{subsection:ang} redistributes angular momentum within the

1361:   disk. But after reaching the maximum, level $20$ resolution (the

1362:   point at which we introduce the black hole particle into the

1363:   simulation and let it evolve), the profile remains rather steady

1364:   with time, because the disk has reached a quasi-stationary

1365:   state. Figure \ref{fig:jvm} shows the angular momentum as a function

1366:   of enclosed gas mass at several different times during a single

1367:   zoom-in episode of the simulation. The figure shows the evolution of

1368:   the angular momentum distribution from $200,000$ years before the

1369:   introduction of the black hole particle, when the simulation has

1370:   refined to level $11$, to $500,000$ years after the introduction of

1371:   the black hole particle, when the simulation is fully refined, and

1372:   has evolved for several hundred thousand years in a quasi-steady

1373:   state.

1374: 

1375:   On the scale of the circumnuclear disk, the simulation has made

1376:   several thousand time steps between $t=100$ kyrs and $t=500$ kyrs,

1377:   meaning that the simulation has undergone significant evolution on

1378:   this spatial scale. Numerical effects, accumulating with each time

1379:   step, would cause deviations in the angular momentum profile. Over

1380:   longer time scales, the slow inward transport of matter will alter

1381:   the angular momentum profile. However, on the hundred thousand year

1382:   time scale we expect the profile to remain rather steady, reflecting

1383:   the quasi-stationary state of the disk, unless there is anomalous

1384:   numerical transport of angular momentum. Figure \ref{fig:jvm} shows

1385:   no systematic deviations in the angular momentum profile after the

1386:   initial refinement and the corresponding re-distribution of the

1387:   angular momentum, demonstrating that there is no sign of unphysical

1388:   numerical angular momentum transport on the time scales of the

1389:   simulation. On time scales much longer than those simulated in the

1390:   present paper, however, these effects may become important.

1391: 

1392: %\clearpage

1393:   \begin{figure}[t]

1394:     \centering

1395:     \epsscale{1.}

1396:     \plotone{fb1.eps}

1397:     \caption{\label{fig:jvm} Angular momentum as a function of

1398:     enclosed gas mass for several different times. The times are measured

1399:     from the introduction of the black hole. The vertical dashed lines

1400:     correspond to enclosed masses on scales of $\approx 10$ and $\approx

1401:     1000$ pc. At each of these scales, the simulation has made

1402:     approximately $230{,}000$ and $3{,}600$ steps, respectively

1403:     (corresponding to the resolution at each scale), between $t=100$

1404:     and $t=500$ kyrs.}

1405:   \end{figure}

1406: %\clearpage

1407: 

1408: 

1409: 

1410: 

1411: \end{document}

1412: