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

2: \documentclass{emulateapj}

3: 

4: \newcommand{\be}{\begin{equation}}

5: \newcommand{\ee}{\end{equation}}

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

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

8: \newcommand{\bld}[1]{\mbox{\boldmath$#1$\unboldmath}}

9: \newcommand{\mycite}[1]{\citeauthor{#1}\ \citeyear{#1}}

10: 

11: \begin{document}

12: 

13: 

14: \title{Formation, Survival, and Destruction of Vortices in Accretion Disks}

15: 

16: 

17: \author{Yoram Lithwick\altaffilmark{1}}

18: \altaffiltext{1}{CITA. Toronto, Ontario, Canada; yoram@cita.utoronto.ca}

19: 

20: 

21: \begin{abstract}

22: 

23: 

24: Two dimensional hydrodynamical disks

25:  are nonlinearly unstable to the formation of vortices.  

26:  Once formed, these vortices essentially survive forever.  

27:  What happens in three dimensions?

28:  We show with incompressible shearing box simulations that  in 3D a vortex in

29:  a short box forms and survives just as in 2D.  But a vortex

30:  in a tall box is unstable and is destroyed. 

31: In our simulation, the unstable vortex decays into a transient turbulent-like state that 

32: transports angular momentum outward at a nearly constant rate

33: for hundreds of orbital times.

34: The 3D  instability that destroys vortices is a generalization of  the 2D instability

35: that forms them.

36: We derive  the conditions for these nonlinear instabilities to act

37: by calculating the coupling between linear modes, and

38: thereby derive the criterion

39: for a vortex to survive in 3D as it does in 2D: {\it the azimuthal extent of the

40: vortex must be larger than the scale height  of the accretion disk}. 

41:  When this criterion is violated, the

42: vortex is unstable and decays.

43: Because vortices are longer in azimuthal than

44: in radial extent by a factor that is inversely proportional to their

45: excess vorticity, a vortex with given radial extent will only survive in a 3D disk

46: if it is sufficiently weak.  This counterintuitive result explains why previous

47: 3D simulations always yielded decaying vortices: their 

48: vortices were too strong.

49:  Weak vortices behave two-dimensionally even if their width is much less than their height

50:  because they are stabilized by rotation, and behave as Taylor-Proudman columns.

51: We conclude that in protoplanetary disks  weak vortices can trap dust and serve

52: as the nurseries of planet formation.  Decaying strong vortices might be responsible

53: for the outwards transport of angular momentum that is required to make accretion

54: disks accrete. 

55: 

56:  

57:  \end{abstract}

58: \keywords{accretion, accretion disks --- instabilities --- solar system: formation ---turbulence }

59: 

60: \section{Introduction}

61: 

62: Matter accretes onto a wide variety of objects, such as young stars, black holes, and white

63: dwarfs, through accretion disks.  In highly ionized disks  magnetic fields are important, 

64: and they trigger turbulence via 

65: the magnetorotational instability \citep{BH98}.  However, many disks,

66:  such as those around

67: young stars or dwarf novae, are nearly neutral  \citep[e.g.,][]{SMUN00,GM98}. 

68: In these disks, 

69: the fluid motions

70: are well described by hydrodynamics. 

71: 

72: Numerical simulations of hydrodynamical 

73: disks in two-dimensions---in the plane of the disk---often

74: produce long-lived vortices \citep{GL99,UR04,JG05}.

75: If vortices really exist in accretion disks, 

76: they can have important consequences.  First and foremost,

77: they might generate turbulence.

78: Since turbulence naturally transports angular momentum outwards\footnote{

79: Energy conservation implies that turbulence transports angular momentum outwards;

80: see \S \ref{sec:pseudo}.  

81:  Nonetheless, if an external energy source  (e.g., the radiative

82: energy from the central star) drives the turbulence, then angular momentum could in principle be transported 

83: inwards.

84: }, 

85: as is required for mass to fall inwards, it might be vortices 

86: that cause accretion disks to accrete.

87: Second, in disks around young stars, long-lived vortices can trap solid particles

88: and initiate the formation of planets \citep{BS95}.

89: 

90: 

91: Why do vortices naturally form in 2D simulations?

92: Hydrodynamical disks are stable to linear perturbations.  

93: However, they are nonlinearly unstable, despite some claims to the 

94: contrary in the

95: astrophysical literature.

96: In two dimensions, the incompressible hydrodynamical

97: equations of a disk are equivalent to those of a non-rotating linear

98: shear flow \citep[e.g.,][hereafter L07]{L07}.

99: And it has long been known that such flows are nonlinearly unstable 

100: (\mycite{Gill65}; \mycite{LK88}; L07).  This nonlinear instability is just 

101: a special case of the Kelvin-Helmholtz instability. 

102: Consider a linear shear flow extending throughout the $x$-$y$ plane

103: with velocity profile $\bld{v}=-qx\bld{\hat{y}}$, where $q>0$ 

104: is the constant shear rate, so that $-q$ is the flow's vorticity.

105: (In the equivalent accretion disk, the local angular speed is 

106: $\Omega=2q/3$.)  This shear flow is linearly stable to 

107: infinitesimal perturbations.  But if the shear profile is altered

108: by a small amount, the alteration can itself be unstable

109: to infinitesimal perturbations.  To be specific, let the alteration

110: be confined

111: within a band of width $\Delta x$, and let it 

112: have vorticity $\omega=\omega(x)$ (with $|\omega|\lesssim q$), so that it induces a velocity field

113: in excess of the linear shear

114: with components $u_y\sim \omega\Delta x$ and $u_x=0$.  

115: Then this band is unstable to infinitesimal nonaxisymmetric

116: (i.e. non-stream-aligned) perturbations provided roughly that

117: \be

118: \left|k_y

119: \right|

120: \lesssim

121: {1\over q}{|\omega|\over \Delta x}

122: \ \ \Rightarrow 

123: {\rm 2D\ instability}

124: \label{eq:2dinst}

125: \ee

126: where $k_y$ is the wavenumber of the nonaxisymmetric perturbation.\footnote{

127: More precisely, the necessary and sufficient condition for instability

128: in the limit $|\omega|\ll q$ is that $|k_y|<{1\over 2 q}\int_{-\infty}^\infty {d\omega/dx\over x-x_0}dx$, where

129: $x_0$ is any value of $x$ at which $d\omega/dx=0$ \citep[][L07]{Gill65,LK88}.

130: For arbitrarily large  $\omega$, 

131: Rayleigh's inflection point theorem and Fj\o rtoft's theorem give necessary (though

132: insufficient) criteria for instability \citep{DR04}.  The former states that for 

133: instability, it is required that $d\omega/dx=0$ somewhere in the flow, i.e. that the velocity

134: field must have an inflection point.  \cite{Love99} generalize Rayleigh's inflection

135: point theorem to compressible and nonhomentropic disks.

136: \label{foot:inst}

137: }

138: For any value of $|\omega|$ and $\Delta x$, the band is always unstable 

139: to perturbations with long enough wavelength.

140: Remarkably, instability even occurs when $|\omega|$ is infinitesimal.

141: Hence we may regard this  as a true nonlinear instability.  

142: \cite{BH06} assert that detailed numerical simulations have not shown evidence for nonlinear

143: instability. The reason many simulations fail to see it

144: is that their boxes are not long enough in the $y$-direction to encompass a small

145: enough non-zero $|k_y|$.

146: 

147: 

148: In two dimensions, the outcome of this instability is a long-lived vortex (e.g., L07).

149: A vortex that has been studied in detail is the Moore-Saffman vortex, 

150: which is a localized patch of spatially

151: constant vorticity  superimposed on a linear shear flow  \citep{Saffman95}.

152: When $|\omega|\lesssim q$, 

153: where $\omega$ here refers to the spatially constant

154: excess vorticity within the patch, and when the vorticity within

155: the patch ($\omega-q$) is stronger than that of the background shear,

156: then the patch forms a stable vortex that is elongated in $y$ relative

157: to $x$ by the factor 

158: \be

159: {\Delta y\over \Delta x} \sim {q\over |\omega|} \ . \label{eq:ms}

160: \ee

161: This relation applies not only to Moore-Saffman vortices,

162: but also to vortices whose $\omega$ is not spatially constant.

163: It may be understood as follows.

164: A patch with characteristic excess vorticity $\sim\omega$ and with $\Delta y\gg \Delta x$

165: induces a velocity field in the $x$-direction with amplitude $u_x\sim|\omega|\Delta x$, 

166: independent of the value of $\Delta y$ (e.g., \S 6 in L07). As long as $|\omega|\lesssim q$,

167: the $y$-velocity within the vortex is predominantly due to the background shear, 

168: and is $\sim q\Delta x$. Therefore the time to cross the width of the vortex is

169: $t_x\sim \Delta x/u_x \sim 1/|\omega|$,

170: and the time to cross its length is 

171: $t_y\sim \Delta y/(q\Delta x)$.  Since these times must be comparable in a vortex,

172: equation (\ref{eq:ms}) follows.

173: Equation (\ref{eq:ms}) is very similar to equation (\ref{eq:2dinst}). 

174: The 2D instability

175: naturally forms into a 2D vortex.  

176: Futhermore, the exponential growth rate of the instability is $\sim |\omega|$, 

177: which is comparable to the rate at which fluid circulates around the vortex.

178: 

179: More generally,

180: an arbitrary axisymmetric profile of $\omega(x)$ tends to evolve into a 

181: distinctive

182: banded structure.  Roughly speaking, bands where $\omega<0$ contain vortices, and these

183: are interspersed 

184: with bands where $\omega>0$, which  contain no vortices.   (Recall that we take

185: the background vorticity to be negative; otherwise, the converse would hold.)

186: The reason for this  is that only regions that have $\omega<0$ can be unstable,

187:  as may be inferred either from the integral

188: criterion for instability given in footnote \ref{foot:inst}, or from Fj\o rtoft's theorem.

189: For more detail on vortex dynamics in shear flows, see

190: the review by \cite{Marcus93}.

191: 

192: 

193: What happens in three dimensions? To date, numerical simulations of 

194: vortices in 3D disks have been reported in two papers.

195: \cite{BM05b} 

196: initialized their simulation with a Moore-Saffman vortex, and

197: solved the anelastic equations in a stratified disk.

198: They found that this vortex decayed.  As it decayed,

199: new vortices were formed

200: in the disk's atmosphere, two scale heights above the midplane. The new vortices

201:  survived for the duration of the simulation.

202: \cite{SSG06} performed both 2D and 3D simulations of the compressible 

203: hydrodynamical equations in an unstratified disk, initialized with 

204: large random fluctuations.  They found that whereas the 2D simulations

205: produced long-lived vortices, 

206: in three dimensions  vortices rapidly decayed.

207: 

208:  Intuitively, it seems

209: clear that a vortex in a very thin disk will behave as it does in  2D.

210: And from the  3D simulations described above it may be inferred that placing this vortex in a

211: very thick disk will induce its decay. 

212: Our main goal in this paper is to understand  these two behaviors, and the transition

213: between them.

214: A crude explanation of our final result is that vortices decay when the 2D

215: vortex motion

216: couples resonantly to 3D modes, i.e., to modes that have vertical wavenumber

217: $k_z\ne 0$.  As described

218: above, a vortex with excess vorticity $|\omega|$ has circulation frequency $\sim|\omega|$, 

219: and $k_y/k_x\sim |\omega|/q$, where $k_x$ and $k_y$ are its ``typical'' wavenumbers.

220: Furthermore, it is well-known that the frequency of axisymmetric ($k_y=0$) inertial

221: waves is $\Omega k_z/\sqrt{k_x^2+k_z^2}$  (see eq.  [\ref{eq:axi3d}]).

222: Equating the two frequencies, and taking the $k_x$ of the 3D mode to be comparable

223: to the $k_x$ of the vortex, as well as setting $q=3\Omega/2$ for a Keplerian disk, we find

224: \be

225: k_z\sim k_y \label{eq:res}

226: \ee

227: as the condition for resonance.

228: Therefore a vortex with length $\Delta y$  will survive in a box

229: with height $\Delta z\lesssim \Delta y$, because in such a box

230: all 3D modes have too high a frequency to couple with the vortex, i.e., all

231:  nonzero $k_z$

232: exceed the characteristic $k_y\sim 1/\Delta y$.

233: But when $\Delta z\gtrsim \Delta y$, there exist $k_z$ in the box that satisfy

234: the resonance condition (\ref{eq:res}), leading to the vortex's destruction.

235: This conclusion suggests that vortices live indefinitely in disks with scale height

236: less than their length ($h\lesssim \Delta y$)

237: because in such disks all 3D modes have too high a frequency for resonant coupling.

238: This conclusion is also consistent  with the simulations of

239: \cite{BM05b} and \cite{SSG06}. Both of these works

240:  initialized their simulations with strong excess vorticity

241: $|\omega|\sim q$, corresponding to nearly circular vortices.  

242: Both had vertical domains that were comparable to the vortices' width.

243: Therefore both saw that their vortices decayed.  Had they initialized their simulations

244: with smaller $|\omega|$, and increased the box length $L_y$ to encompass

245: the resulting elongated vortices, both would have found long-lived 3D vortices.

246:  \citeauthor{BM05b}'s discovery of long-lived vortices in the disk's

247: atmosphere is simple to understand because  the local scale height is reduced

248: in inverse proportion to the height above the midplane.   Therefore 

249: higher up in the atmosphere the dynamics becomes more two-dimensional, and

250: a given vortex is better able to survive the higher it is.\footnote{

251: However, \cite{BM05b} also include buoyancy forces in their simulations,

252: which we ignore here.  How buoyancy affects the stability of vortices is a topic

253: for future work.

254: }

255: 

256: 

257: \subsection{Organization of the Paper}

258: 

259: In  \S \ref{sec:eom} we introduce the equations of motion, and

260: in \S \ref{sec:pseudo} we present

261:  two pseudospectral simulations.

262:  One illustrates the

263: formation and survival of a vortex in a short box, and the other illustrates

264: the destruction of a vortex in a tall box. 

265: 

266: In \S\S \ref{sec:lin}-\ref{sec:nonlin} we develop a theory explaining

267: this behavior.

268: The reader who is satisfied by the qualitative description leading

269: to equation (\ref{eq:res}) may

270: skip those two sections.

271: The theory that we develop is indirectly related to the transient

272: amplification scenario for the generation of turbulence.

273:  Even though 

274: hydrodynamical disks are linearly stable, linear perturbations 

275: can be transiently amplified before they decay, often by a 

276: large factor.  It has been proposed that sufficiently amplified modes might

277: couple nonlinearly, leading to turbulence \citep[e.g.,][]{CZTL03,Yecko04,AMN05}.

278: However, to make this proposal more concrete, one must work out

279: how modes couple nonlinearly.  In  L07, we did that in two dimensions.  

280: We showed that the 2D nonlinear instability of equation (\ref{eq:2dinst}) is a consequence

281: of the coupling of an axisymmetric mode with a transiently amplified mode, 

282: which may be called a ``swinging mode'' because its phasefronts are swung around

283: by the background shear. 

284: In \S \ref{sec:nonlin} we show that the 3D instability responsible

285: for the destruction of vortices is a generalization of this 2D instability. 

286: It may be understood by examining the coupling of a 

287: 3D swinging mode 

288: with an axisymmetric one.    3D modes become increasingly unstable as $|k_z|$

289: decreases, and in the limit that $k_z\rightarrow 0$, the 3D instability matches

290: smoothly onto the 2D one.

291: Thicker disks are more prone to 3D instability because they encompass

292: smaller $|k_z|$.

293: 

294: 

295: 

296: 

297: \section{Equations of Motion}

298: \label{sec:eom}

299: 

300: 

301: 

302: 

303: We solve the  ``shearing box'' equations,

304: which approximate the dynamics in an accretion disk on lengthscales

305: much smaller than the distance to the disk's center. 

306: We assume incompressibility, which is a good approximation

307: when relative motions are subsonic.

308: We also neglect vertical gravity, and hence stratification and buoyancy, which is 

309: an oversimplification. 

310: To fully understand vortices in astrophysical disks, one must consider the effects

311: of  vertical gravity as well as of shear and rotation. 

312: In this paper, we consider only two pieces of this puzzle---shear and

313: rotation.  Adding the third piece---vertical gravity---is a topic that 

314: we leave for future investigations.  See also the Conclusions for some 

315: speculations.

316: 

317: 

318: 

319: An unperturbed Keplerian disk has angular velocity profile $\Omega(r)\propto r^{-3/2}$.

320: In a reference

321: frame rotating at constant angular speed $\Omega_0\equiv \Omega(r_0)$, where $r_0$ is

322: a fiducial radius, 

323: the incompressible shearing box equations of motion  read

324: \begin{eqnarray}

325: {\partial_t {\bld v}}+{\bld{ v\cdot\nabla v}}&=&-

326: 2\Omega_0{\bld{\hat z}}{\bld{\times v}}+2q\Omega_0x{\bld{ \hat{x}}} -{\bld\nabla} P/\rho\ \ ,

327: \label{eq:bigeom}

328: \\

329: \bld{\nabla\cdot v}&=&0  

330: \end{eqnarray}

331: adopting

332:  Cartesian coordinates $x,y,z$, which are related to the disk's cylindrical $r,\theta$ 

333: via $x\equiv r-r_0$ and $y\equiv r_0(\theta-\Omega_0t)$;

334: $\bld{\hat{x}}$ and $\bld{\hat{z}}$ are unit vectors, 

335: and

336: \be

337: q\equiv -{d \Omega\over d\ln r}\Big\vert_{r_0}={3\over 2}\Omega_0 \ ,

338: \ee

339: We retain $q$ and $\Omega_0$ as independent parameters because they 

340: parameterize different effects: shear and rotation, respectively. 

341: The first term on the right-hand side of equation (\ref{eq:bigeom}) is the Coriolis force, 

342: and the second is what remains after adding centrifugal and gravitational forces.

343: Decomposing  the velocity  into

344: \be

345: \bld{v}=-qx\bld{\hat{y}}+\bld{u} \ , \label{eq:dec}

346: \ee

347: where the first term is the shear flow of the unperturbed disk,  yields

348: \begin{eqnarray}

349: \left( {\partial_t}-qx{\partial_y}\right) {\bld u}

350: +{\bld{u\cdot\nabla u}}

351: &=&

352:  q u_x\bld{\hat{y}}

353:  -2\Omega {\bld{\hat z}}{\bld{\times u}}

354:  -\bld{\nabla} P/\rho \label{eq:eom2}

355: \\

356: {\bld{\nabla\cdot u}} &=&

357: 0   \  ,

358: \label{eq:eom1}

359: \end{eqnarray}

360: dropping the subscript from $\Omega_0$, as we shall do 

361: in the remainder of this paper.  An unperturbed disk has 

362: $\bld{u}=0$.

363: 

364: In addition to the above ``velocity-pressure'' formulation, an

365: alternative ``velocity-vorticity'' formulation will prove convenient.  It

366: is given by 

367: the curl of equation (\ref{eq:eom2}), 

368: \be

369: (\partial_t-qx\partial_y)\bld{\omega}=-q\bld{\hat{y}}\omega_x+(2\Omega-q)\partial_z\bld{u}+

370: \bld{\nabla\times}(\bld{u\times \omega})

371: \label{eq:eomo}

372: \ee

373: where

374: \be

375: \bld{\omega}\equiv \bld{\nabla\times u}

376: \label{eq:omdef}

377: \ee

378: is the vorticity of $\bld{u}$.

379:  Equation (\ref{eq:eomo}),  together with the inverse of equation (\ref{eq:omdef})

380: \be

381: \bld{u}=-\nabla^{-2}\bld{\nabla\times\omega} \label{eq:ominv} \ ,

382: \ee

383: form a complete set. 

384: 

385: Equation (\ref{eq:eomo}) implies that the total vorticity field

386: is frozen into the fluid, because it is equivalent to 

387: \be

388: \label{eq:omtot}

389: \partial_t\bld{\omega}_{\rm tot}=\bld{\nabla\times}(\bld{v\times\omega}_{\rm tot}) \ ,

390: \ee

391: where

392: \be

393: {\bld \omega}_{\rm tot}\equiv (2\Omega-q)\bld{\hat{z}}+\bld{\omega} \ 

394: \ee

395: is the total vorticity; note that

396:  $-q\bld{\hat{z}}$ is the vorticity of the unperturbed shear flow in the rotating frame, 

397: and hence $(2\Omega-q)\bld{\hat{z}}$ is the unperturbed vorticity in the non-rotating frame.

398: The vorticity-velocity picture is similar to MHD, where it is the

399: magnetic field that is frozen-in because it satisfies equation (\ref{eq:omtot})

400: in place of $\bld{\omega}_{\rm tot}$.  However, in MHD the velocity field has its own dynamical

401: equation, whereas in incompressible hydrodynamics it is determined directly from the vorticity field

402: via equation (\ref{eq:ominv}).

403: 

404: 

405: %\clearpage

406: %%%%%%%%%%%%%begin figure

407: \begin{figure*}

408:   \centering

409:   \includegraphics[width=.32\textwidth]{f1a.eps}

410:   \includegraphics[width=.33\textwidth]{f1b.eps}

411:   \includegraphics[width=.33\textwidth]{f1c.eps}  

412:  \includegraphics[width=.32\textwidth]{f1d.eps}

413:   \includegraphics[width=.32\textwidth]{f1e.eps}  

414:   \includegraphics[width=.33\textwidth]{f1f.eps}  

415:   \caption{{\bf Vortex Formation and Survival in a Short Box:} 

416:    Color depicts  $\omega_z$.

417:   The initial state is unstable to vertically symmetric ($k_z=0$) perturbations,

418:   and forms into a vortex.  But it is stable to 3D ($k_z\ne 0$) perturbations, and the evolution remains

419:   two-dimensional.  The bottom panels show  horizontal slices through the boxes in the upper

420:   panels, midway through the boxes.

421:    At time=150, the vortex has already

422:   formed. Only fluid with $\omega_z<-0.08$ is shown  in the middle panels to

423:   highlight the vortex, and to illustrate that surfaces of constant

424:   $\omega_z$ remain purely vertical.  At time=500,

425:   the vortex still survives.  Its amplitude is slowly decaying by viscosity, which

426:   acts on timescale=1130.  We set  $\Omega=1$ and $q=3/2$.

427:    The number of modes in the simulation is

428:   $n_x\times n_y\times n_z=64\times 64\times 32$, and the size of the simulation box is $(L_x,L_y,L_z)=({1\over 15},1,{1\over 2})$. In this figure, $L_z$ is to scale relative to

429:   $L_y$, but $L_x$ has been expanded by a factor of 5 for clarity.

430: }

431:   \label{fig:shortsim}

432: \end{figure*}

433: %%%%%%%%%%%%%end figure

434: 

435: 

436: 

437: 

438: 

439: %%%%%%%%%%%%%begin figure

440: \begin{figure*}

441:   \centering

442:   \includegraphics[width=.32\textwidth]{f2a.eps}

443:   \includegraphics[width=.33\textwidth]{f2b.eps}

444:   \includegraphics[width=.33\textwidth]{f2c.eps}  

445:   \includegraphics[width=.32\textwidth]{f2d.eps}

446:   \includegraphics[width=.32\textwidth]{f2e.eps}  

447:   \includegraphics[width=.33\textwidth]{f2f.eps}  

448:   \caption{{\bf Vortex Destruction in a Tall Box:} The setup 

449:   is identical to the short-box simulation of Figure \ref{fig:shortsim}, except that the height

450:   $L_z$ has been increased by a factor of 4, so that it now exceeds $L_y$. 

451:    The resulting evolution is dramatically different.

452:   The initial state is now unstable not only to 2D perturbations, but to 3D ones as well.  In the

453:   middle panels, surfaces of constant $\omega_z$ are warped, and the evolution is

454:   no longer vertically symmetric. 

455:     In the right panels, the flow looks turbulent.}

456:   \label{fig:tallsim}

457: \end{figure*}

458: %%%%%%%%%%%%%end figure

459: 

460: 

461: 

462: 

463: 

464: %%%%%%%%%%%%%begin figure

465: \begin{figure}

466:   %\centering

467:   \hspace{-1cm}\includegraphics[width=.55\textwidth]{f3.eps}

468:   \caption{{\bf Energy in the Short Box:}

469:   The three contributions to the energy budget, $E_{u^2}$, $\Delta E_{\rm shear}$,

470:   and $\Delta E_{\rm visc}$, are defined in equations (\ref{eq:e1})-(\ref{eq:e3}).

471:   $E_{u^2}$ initially decays, and then rises to a peak

472:   near $t\sim 200$ as nonaxisymmetric perturbations turn the axisymmetric

473:   mode into a vortex. 

474:   Subsequently, the vortex decays due to viscosity.  The spikiness

475:   of the evolution is due to the boundary conditions, as explained in the text.

476:   Also shown in the bottom panel is the error due to numerical effects,

477:   $\Delta E_{\rm error}$ (eq. [\ref{eq:ee}]).

478:   It is unlabelled because it is mostly obscured by $\Delta E_{\rm shear}$.

479:   But it is nearly equal to zero everywhere, showing that the code accurately

480:   tracks the components of the energy budget.

481:   }

482:   \label{fig:shorten}

483: \end{figure}

484: %%%%%%%%%%%%%end figure

485: 

486: %%%%%%%%%%%%%begin figure

487: \begin{figure}

488:   %\centering

489:   \hspace{-1cm}\includegraphics[width=.55\textwidth]{f4.eps}

490:   \caption{{\bf Energy in the Tall Box:} The initial evolution is almost the same

491:   as that seen in the short box (Fig. \ref{fig:shorten}).  But 3D perturbations are unstable and force

492:   the destruction of the vortex.   In the time interval $300\lesssim t\lesssim 600$,

493:   while the initial axisymmetric disturbance decays in a turbulent-like state,  the value of $E_{u^2}$ is significantly larger than its

494:   initial value, and $\Delta E_{\rm shear}$ rises nearly linearly in time, corresponding

495:   to nearly constant outwards transport of angular momentum in a disk.  

496:   The contribution of numerical errors to the time-integrated energy budget, $\Delta E_{\rm error}$ (eq. [\ref{eq:ee}]),

497:   remains small throughout.

498:   }

499:   \label{fig:tallen}

500: \end{figure}

501: %%%%%%%%%%%%%end figure

502: %\clearpage

503: 

504: 

505: \section{Two Pseudospectral Simulations}

506: \label{sec:pseudo}

507: 

508: 

509: 

510: The pseudospectral code is described in detail in the Appendix of L07.  

511: It solves the velocity-pressure equations of motion with an explicit viscous term

512: \be

513: \nu\nabla^2{\bld{u}}

514: \label{eq:visc}

515: \ee

516: added to equation (\ref{eq:eom2}).

517: (In L07, we did not include this term because we

518: only considered inviscid flows.)

519: The equations are solved 

520: in Fourier-space by decomposing fields into

521:  spatial Fourier modes whose wavevectors

522: are advected by the background flow $-qx\bld{\hat{y}}$.

523: As a result, the boundary conditions are periodic in the $y$ and $z$ dimension, and

524: ``shearing periodic'' in $x$.

525: Most of our techniques are standard \citep[e.g.,][]{MG01,Rogallo81,BM06}.  One exception is 

526: our method for

527: remapping 

528: highly trailing wavevectors into highly leading ones, which is both simpler

529: and more accurate than the usual method.  In addition, our remapping 

530: does not introduce power into leading modes, because a mode's amplitude

531: has always been set to zero before the remap.

532: The code was extensively tested on 2D flows in L07.

533: A number of rather stringent 3D tests are performed in this  paper.  We shall show 

534: that the code correctly reproduces the linear evolution of 3D modes (\S \ref{sec:lin}), as

535: well as the nonlinear coupling between them (\S \ref{sec:nonlin}). We also show in 

536: the present section

537: that it tracks

538: the various contributions to the energy budget, and that the sum

539: of the contributions vanishes to high accuracy.

540: 

541: Figures \ref{fig:shortsim}-\ref{fig:tallen} show results 

542:   from two pseudospectral 

543: simulations.  One simulation illustrates the formation

544: and survival of a vortex, and the other illustrates

545:  vortex destruction.

546: In the first (the ``short box''),

547:  the number of Fourier modes  used 

548: is $n_x\times n_y\times n_z=64\times 64\times 32$, 

549: and the simulation box has dimensions

550: $L_x={1\over 15}$, $L_y=1$, and $L_z={1\over 2}$.  

551: In the second (the ``tall box''), the setup is identical,

552: except that it has $L_z=2$ instead of $1/2$.

553: Both simulations are initialized by setting

554: \be

555: \omega_z\Big\vert_{t=0}=-0.1\cos\left({2\pi\over L_x}x\right) \ .

556: \label{eq:initz}

557: \ee

558: In addition,  small perturbations are added to long-wavelength modes.

559: Specifically, labelling the wavevectors as

560: \be

561: (k_x,k_y,k_z)=2\pi\left({j_x\over L_x},{j_y\over L_y},{j_z\over L_z}  \right) \ ,

562: \label{eq:kj}

563: \ee 

564: with integers $(j_x,j_y,j_z)$,

565: we select all modes that satisfy

566:  $|j_x|\leq 3$, $|j_y|\leq 3$,

567: and $|j_z|\leq 3$, and set the Fourier amplitude of their $\omega_z$  to 

568:  $10^{-4}e^{i\phi}$, where $\phi$ is a random phase. 

569: But

570: we exclude the  $(j_x,j_y,j_z)=(0,0,0)$ mode, as well as 

571: $(j_x,j_y,j_z)=(\pm 1,0,0)$, which is  given by equation (\ref{eq:initz}).

572: Finally, we set $\Omega=1$, $q=3/2$, $\nu=10^{-7}$, and integration 

573: timestep $dt=1/30$.

574: 

575: With our chosen initial conditions, the mode given by equation (\ref{eq:initz})

576: is nonlinearly unstable to vertically symmetric ($k_z=0$) perturbations, and hence

577: it tends to wrap up into a vortex. From the approximate criterion for instability 

578:  (eq. [\ref{eq:2dinst}]),

579: we see that to illustrate the wrapping up into a vortex

580: of a mode with a small amplitude, one must make the simulation box

581: elongated in the $y$-direction relative to the $x$-scale of the mode in equation

582: (\ref{eq:initz}).

583: 

584: 

585: In the short box (Fig. \ref{fig:shortsim}), the evolution proceeds just as it would in two dimensions. 

586: The initial mode indeed wraps up into a vortex, and 

587:  the evolution remains vertically

588: symmetric throughout.  Once formed, the vortex can live for ever in the absence

589: of viscosity.

590: But in our simulation, there is a slow viscous decay.  The timescale for viscous decay

591: across

592: the width of the vortex is

593: $\sim 1/\nu k_x^2=1130$, taking $k_x=2\pi/L_x$.

594: 

595: In the tall box (Fig. \ref{fig:tallsim}), the evolution is dramatically different.  

596: In this case, the initial state is unstable not only to 2D  perturbations,

597:  but to 3D $(k_z\ne 0)$

598: ones as well. In the middle panel of that figure, we see 

599: that instead of forming a vertically symmetric vortex as in the short box,

600:  surfaces of constant $\omega_z$ are 

601: warped. 

602: By the third panel, the flow looks turbulent.

603: 

604: 

605: Figures \ref{fig:shorten}-\ref{fig:tallen} shows the evolution of the

606:  energy in these simulations.

607: Projecting ${\bld u}$

608: onto

609: the Navier-Stokes equation (eq. [\ref{eq:eom2}]  with viscosity included),

610: and spatially averaging, we arrive at the energy equation

611: \be

612: {d\over dt}

613: {\left\langle u^2

614: \right\rangle\over 2}=q

615: \langle u_xu_y\rangle 

616: +\nu\langle

617: {\bld u\cdot}\nabla^2\bld{u}

618: \rangle \ ,

619: \ee

620: after applying the shearing-box boundary conditions, 

621: where angled brackets denote a spatial average.

622: The time integral of this equation is

623: \be

624: E_{u^2}-E_{u^2}\vert_{t=0}=\Delta E_{\rm shear}+\Delta E_{\rm visc} \ , 

625: \label{eq:inte}

626: \ee

627: where

628: \beqn

629: E_{u^2}&\equiv& {\left\langle u^2

630: \right\rangle\over 2}

631: \label{eq:e1}

632:  \\

633: \Delta E_{\rm shear}&\equiv&

634: q\int_0^t \langle u_xu_y\rangle dt' \\

635: \Delta E_{\rm visc}&\equiv&

636: \nu\int_0^t

637: \langle

638: {\bld u\cdot}\nabla^2\bld{u}

639: \rangle dt' \ . \label{eq:e3}

640: \eeqn

641: The pseudospectral code records  each of these terms, and

642: Figure \ref{fig:shorten} shows the result in the short box 

643: simulation.  At very 

644: early times, $E_{u^2}$ decays from its initial value due to viscosity. 

645: At the same time, the small vertically symmetric perturbations are growing exponentially, 

646: and they start to give order unity perturbations by $t\sim 150$, by which time

647:  a vortex has been formed (Fig. \ref{fig:shortsim}).  

648:  As time evolves, $E_{u^2}$ gradually decays due to viscosity on the viscous

649:  timescale $=1130$.   The evolution  is very spiky.  

650:  We defer a discussion of this spikiness to the end of this section.

651:  

652:  Figure \ref{fig:tallen} shows the result in the tall box. The early evolution of 

653:  $E_{u^2}$ is similar to that seen in the short box.  Both start with the same $E_{u^2}$,

654:  and an initial period of viscous decay is interrupted by exponentially growing 

655:  perturbations. But in the tall box, not only are vertically symmetric modes

656:  growing, but modes with $k_z\ne 0$  are growing as well. By $t\sim 150$, there is a distorted

657:  vortex that subsequently decays into a turbulent-like state.  The energy $E_{u^2}$

658:  rises to a value significantly larger than its initial one, and it continues to rise

659:  until $t\sim 600$, when it starts to decay.  Throughout the time interval

660:  $300\lesssim t\lesssim 600$, $\Delta E_{\rm shear}$ rises nearly linearly in 

661:  time, showing that $\langle u_xu_y\rangle$ is positive and  nearly constant.

662:  

663: It is intriguing that  $\langle u_x u_y\rangle$ is positive for hundreds of orbits,

664: because

665:  it suggests that decaying vortices might transport

666: angular momentum outwards in disks and hence drive accretion.

667: Understanding the level of the turbulence,

668:  its lifetime, 

669:  and its nature are  topics for future work.

670:  Here we merely address the sign of $\langle u_xu_y\rangle$.

671: The quantity  $\langle u_xu_y\rangle$ is the 

672: flux of $y$-momentum in the $+x$-direction (per unit mass and spatially averaged).

673: It corresponds to the flux of angular momentum

674: in a disk.  A positive

675: $\langle u_xu_y\rangle$ implies an outwards flux of angular momentum, 

676: as is required to drive matter inwards in an accretion disk.

677:  (Even though the shearing box cannot distinguish

678:  inwards from outwards,

679: the sign of the angular momentum within a box depends on which side of the shearing

680: box one calls inwards.  Therefore, outwards transport of (positive) angular momentum

681: is well-defined in a shearing box.)

682: In the shearing box, 

683: any force that tends to diminish the background shear

684: flow $-qx\bld{\hat{y}}$ necessarily transports $y$-momentum in the $+x$-direction.

685: Hence the fact that $\langle u_xu_y \rangle >0$ in Figure \ref{fig:tallen} shows

686: that the turbulence exerts forces that resist the background shear, as one 

687: might

688: expect on physical grounds.

689: One can also understand why   $\langle u_xu_y \rangle >0$ from

690: energy considerations. 

691: Since  $\Delta E_{\rm visc}<0$, as may be seen explicitly by

692: an integration by parts, i.e. $\langle \bld{u\cdot}\nabla^2\bld{u} \rangle

693: =-\sum_{i,j}\langle(\partial_j u_i)^2\rangle$, equation (\ref{eq:inte})

694: may be rearranged to read

695: \be

696: \Delta E_{\rm shear} = \left\vert \Delta E_{\rm visc}\right\vert

697: +E_{u^2}-E_{u^2}\vert_{t=0} \ .

698: \label{eq:ee1}

699: \ee

700: If  the turbulence reaches a steady state---as it approximately does in 

701: Figure \ref{fig:tallen} during the time interval

702: $300\lesssim t\lesssim 600$---then the last two terms in the 

703: above equation are nearly constant, whereas $\left\vert \Delta E_{\rm visc}\right\vert$

704: increases linearly with time.   Hence $\Delta E_{\rm shear}$ must also increase.

705: The fact that energy dissipation implies outwards transport of angular momentum

706: is a general property of accretion disks \citep[e.g., ][]{LP}.   Since turbulence always

707: dissipates energy, it must also transport angular momentum outwards.  

708: However, this argument can be violated if  an external energy source drives

709: the turbulence,

710: in which case one would have to add this energy to the left-hand side of equation (\ref{eq:ee1}).

711: For example, the simulations of \cite{SB96} show that convective disks can transport angular momentum

712: inwards when an externally imposed heat source drives the convection. 

713: 

714:  

715:  

716:  

717: Also shown in the bottom panels of Figures \ref{fig:shorten}-\ref{fig:tallen}

718: is the integrated energy error

719: \be

720: \Delta E_{\rm error}\equiv \Delta E_{\rm shear}+\Delta E_{\rm visc}+E_{u^2}\vert_{t=0}-E_{u^2}

721: \label{eq:ee}

722: \ee

723:  due to 

724: numerical effects, which is seen to be small. 

725:  (In Figure \ref{fig:shorten}, $\Delta E_{\rm error}$ is not labelled because

726: the curve is mostly obscured by $\Delta E_{\rm shear}$; it can be seen near $t\sim 200$,

727: and is everywhere very nearly equal to zero.)

728: The fact that $\Delta E_{\rm error}$ nearly vanishes throughout the simulations is 

729: not guaranteed by the pseudospectral algorithm.  Rather, we have chosen

730: $\nu$ to be large enough that the algorithm introduces negligible  error into the

731: energy budget.  To be more precise, at each timestep in the pseudospectral code,

732: modes that have $|j_x|>n_x/3$ or $|j_y|>n_y/3$ or $|j_z|>n_z/3$, where $j_{x,y,z}$

733: are defined via equation (\ref{eq:kj}), have their amplitudes set to zero (``dealiased'').

734: This introduces an error that is analogous to

735: grid error in grid-based codes.  By choosing  $\nu$ to be sufficiently large, it

736: is the explicit viscosity that forces modes

737: with large $k$ to have small amplitudes, in which case

738: the dealiasing procedure has little effect on the dynamics.

739: Increasing the resolution $n_x\times n_y\times n_z$

740:  would allow a smaller $\nu$ to be chosen---implying

741: a larger effective Reynolds number---while keeping the energy error small.

742: 

743: 

744: 

745: The curves of $E_{u^2}$ show

746:  sharp narrow spikes every time interval $\Delta t=10$, with width $\sim 1$.

747:  Similar spikes have been seen in other simulations \citep{UR04,SSG06}, but

748:  they are stronger and narrower in our simulations because our simulation box

749:  is elongated.  These spikes are due to the

750:  shearing-periodic boundary conditions.  It is perhaps simplest to understand

751:  them by following the evolution in $k$-space, as we shall do in 

752:   \S \ref{sec:nonlin} (see also L07).  But for now, we explain their origin

753:    in real-space.

754: By the  nature of shearing-periodic boundary conditions,   associated with the simulation 

755:  box centered  at $x=0$ are  ``imaginary boxes'' centered

756:     at $x=jL_x$ with

757:  integer $j=\pm 1,\pm 2,\cdots$.  These imaginary boxes completely tile the $x-y$ plane,

758:  and each

759:   contains a virtual copy of the conditions inside the simulation box.

760:  The boxes move relative to the simulation box in 

761:  the $y$-direction, with the speed of the mean shear at the center of each box,

762:  $-qjL_x$.

763:  Therefore, in  the  time interval 

764:  $\Delta t = L_y/(qL_x)=10$, all the boxes line up.  When this happens, 

765:  the velocity field $\bld{u}$ that is induced by the vorticity within all the boxes 

766:  (via eq. [\ref{eq:ominv}]) becomes large, because all the boxes reinforce each other,

767:  and therefore $E_{u^2}$ exhibits a spike.

768: Even though the shearing-periodic boundary conditions that we use are 

769: somewhat artificial, 

770: we are confident that using more realistic open boundary conditions would not

771: affect the main results of this paper---and particularly not the stability of

772:  axisymmetric modes to 3D perturbations.

773: In L07, where we considered 2D dynamics, 

774: we investigated both open and shearing-periodic boundary conditions, and

775: showed explicitly that both give similar results.

776: We also feel that the boundary conditions likely do not affect the level

777: and persistence of the ``turbulence'' seen in Figure \ref{fig:tallen}.  However, 

778: this is less certain.  Future investigations should more carefully address the role

779: of boundary conditions.  

780: 

781: 

782: \section{Linear Evolution}

783: \label{sec:lin}

784: 

785: 

786: %\clearpage

787: %%%%%%%%%%%%%begin figure

788: \begin{figure*}

789:   \centering

790:   \includegraphics[width=1\textwidth]{f5.eps}

791:   \caption{{\bf Evolution of Wavevectors:} Modes have constant $k_y$ and $k_z$, 

792:   and $k_x=qtk_y+$const.  The three spheres depict modes that

793:    play important roles in nonlinear

794:   instability.  The mode at $(1,0,0)$ does not move in $k$-space.  The other two modes

795:   are swinging modes that are depicted in the leading phase of their swing. They will

796:   become trailing after crossing through the radially symmetric plane. The mode

797:   crossing through $(1,-1,0)$ is responsible for 2D instability that forms vortices.

798:   The one crossing through $(1,-1,1)$ is responsible for 3D instability that destroys

799:   vortices.}

800:   \label{fig:kspace}

801: \end{figure*}

802: %%%%%%%%%%%%%end figure

803: %\clearpage

804: 

805: In the remainder of this paper, we develop a theory explaining the

806: stability of vortices seen in the above numerical simulations.  

807: We first consider the linear evolution of individual modes, 

808: and then proceed to show how nonlinear coupling between

809: linear modes can explain vortex stability.

810: 

811: 

812: The linear evolution has been considered previously \citep{AMN05,JG05,BH06}.

813: Only two aspects of our treatment are new.

814: First, we give the solution in terms

815: of variables that allow the simple reconstruction of

816: the full vectors $\bld{\omega}$ and $\bld{u}$.

817: And second, we give the analytic expression 

818:   for matching a leading mode 

819: onto 

820: a trailing mode that is

821: valid for all $k_y$ and $k_z$,

822: 

823: The linearized equation of motion is (eq. [\ref{eq:eomo}])

824: \be

825: (\partial_t-qx\partial_y)\bld{\omega}=-q\bld{\hat{y}}\omega_x+(2\Omega-q)\partial_z\bld{u} \ .

826: \label{eq:elin}

827: \ee

828:  A single mode may be written as

829:  \be

830: \bld{\omega}(\bld{x},t)=\bld{\hat{\omega}}(\bld{k_0},t)e^{i 

831: \left[

832: {\bld{k}}

833: \left(

834: {\bld{k_0}},t

835: \right)

836: \right]

837: {\bld\cdot}

838: \bld{x}

839: } \ , \label{eq:omk}

840: \ee

841: where $\bld{k_0}$ is a constant vector that denotes the wavevector at time $t=0$, and

842: the  wavevector $\bld{k}=\bld{k}(\bld{k_0},t)$ has components

843: \beqn

844: k_y &=& k_{0y}={\rm const} \\

845: k_z &=& k_{0z} ={\rm const}\\

846: k_x &=& {k_{0x}}+qtk_y\ne {\rm const} \ ,

847: \eeqn

848: so that upon insertion into equation (\ref{eq:elin}), the time-derivative of the exponential

849: cancels the term $-qx\partial_y{\bld{\omega}}$.

850: The velocity field induced by such a mode 

851:  is (eq. [\ref{eq:ominv}])

852: \beqn

853: \bld{u}(\bld{x},t)=

854: \bld{\hat{u}}(\bld{k_0},t)

855: e^{i \bld{k}\left[

856: \left(

857: {\bld{k_0}},t

858: \right)

859: \right]

860: {\bld\cdot}

861: \bld{x}} 

862: \label{eq:uk}

863: \eeqn

864: where

865: \be

866: \bld{\hat{u}}=

867: i{\bld{k\times\hat{\omega}}\over k^2}

868: \label{eq:udef}

869: \ee

870: 

871: Figure \ref{fig:kspace} sketches the evolution of wavevectors. 

872: Axisymmetric modes ($k_y=0$) do not move in $k$-space, as depicted by

873: the sphere at $(1,0,0)$ in Figure \ref{fig:kspace}.

874:  ``Swinging modes'' have $k_y\ne 0$, and their $k_x$ is time-dependent.  Their

875: fronts of constant phase are advected by the background shear.  Swinging modes

876: with $k_x/k_y<0$, as depicted by the two spheres 

877: near $(1,-1,0)$ and $(1,-1,1)$ in Figure \ref{fig:kspace},

878:  have phasefronts tilted into the background shear, i.e., they are leading modes.

879:    As time evolves, 

880: the shear first swings their $k_x$ through $k_x=0$, at which point their phasefronts

881: are radially symmetric.

882: Subsequently, they become

883: trailing modes ($k_x/k_y>0$), and approach alignment with the azimuthal direction

884:  ($k_x/k_y\rightarrow \infty$).  

885: 

886: 

887: We turn now to the evolution of the Fourier amplitudes.

888: In the remainder of this paper we drop the hats

889: \beqn

890: \bld{\hat\omega}\rightarrow \bld{\omega} \ ,

891: \ \ 

892: \bld{\hat u}\rightarrow \bld{u} \ .

893: \eeqn

894: To distinguish real-space fields, we shall explicitly write their

895: spatial dependence, e.g. $\bld{\omega}(\bld{x})$.

896: 

897: Because $\bld{\omega}(\bld{x})$ is divergenceless, $\bld{{\omega}}$

898: only has two degrees of freedom, which we select to be

899:  ${\omega}_x$ and

900: \beqn

901: {\omega}_{yz}\equiv 

902: {\bld{\hat{x}\cdot}

903: \left({{\bld{k\times{\omega}}}}\right)\over k_{yz}} = 

904: \left\{

905:  \begin{array}

906:  {r@{\quad,\quad}l}

907:  -{\omega}_y & {\rm if\ } k_y=0 \\

908: {\omega}_z & {\rm if\ } k_z=0

909:  \end{array}

910:  \right.

911: \eeqn

912: where

913: \be

914: k_{yz}\equiv \sqrt{k_y^2+k_z^2} 

915: \ee

916: Our variable $\omega_{yz}$ is proportional to the variable $U$ of

917: \cite{BH06}. Adopting $\omega_x$ as the second degree of freedom

918: enables

919:   the full vectors to 

920: be reconstructed  as

921: \beqn

922: \bld{{\omega}} &=& 

923: -{\omega}_x

924: {\bld{k\times}(\bld{k\times \hat{x}})\over k_{yz}^2}

925: -{{\omega}_{yz}}{\bld{k\times\hat{x}}\over k_{yz}}

926: \label{eq:omrec}

927: \\

928: \bld{{u}} &=& -i\omega_{yz}{\bld{k\times}(\bld{k\times \hat{x}})\over k^2 k_{yz}}+

929: i\omega_x {\bld{k\times\hat{x}}\over k_{yz}^2}

930: \label{eq:urec}

931: \eeqn

932: 

933: The linearized equation  (\ref{eq:elin}) is expressed in terms of these degrees of freedom as

934: \be

935: {k_{yz}\over qk_y}{d\over dt}

936: \left(\begin{array}{c}{\omega}_x \\ {\omega}_{yz}\end{array}\right)

937: =\beta{\Omega\over \kappa}

938: \left(\begin{array}{cc}0 & -{1\over 2}{\kappa^2\over\Omega^2}{1\over 1+\tau^2} \\ 2 & 0\end{array}\right)

939: \left(\begin{array}{c}{\omega}_x \\ {\omega}_{yz}\end{array}\right) \ ,

940: \label{eq:linlim}

941: \ee

942: after introducing the  epicyclic frequency,

943: \beqn

944: \kappa \equiv \sqrt{2\Omega(2\Omega-q)}  \ ,

945: \eeqn

946: with $\kappa=\Omega$ in a Keplerian disks, and

947: \beqn

948: \tau&\equiv& {k_x\over k_{yz}} \\

949: \beta&\equiv&{\kappa\over q}{k_z\over k_y} \ .

950: \eeqn

951: As long as $k_y\ne 0$,

952: $\tau$ varies in time through its dependence on

953: $k_x=k_{0x}+qtk_y$.

954: 

955: For axisymmetric modes ($k_y=0$), $\tau$ is constant and

956: \be

957: {d^2\over dt^2}\omega_{yz}+\kappa^2{k_z^2\over k_x^2+k_z^2}\omega_{yz}=0 \ ,

958: \label{eq:axi3d}

959: \ee

960: the solution of which is sinusoidal with frequency $\kappa k_z/\sqrt{k_x^2+k_z^2}$.

961: Axisymmetric modes with phasefronts aligned with the plane of the disk ($k_x=k_y=0$) have

962: in-plane fluid velocities, and they oscillate at

963: the epicyclic frequency of a free test particle, $\kappa$.

964:   But axisymmetric modes with tilted

965: phasefronts have slower frequencies, because fluid pressure causes deviations from

966: free epicycles.  In the limit of vertical axisymmetric phasefronts ($k_z=k_y=0$), the effects of

967: rotation disappear entirely, and this zero-frequency mode merely alters the mean shear

968: flow's velocity profile.

969: 

970: 

971: For swinging modes

972: ($k_y\ne 0$),  it is convenient to employ $\tau$ as  the time variable.

973: Since 

974: \be

975: {k_{yz}\over qk_y}{d\over dt}={d\over d\tau} \ ,

976: \ee

977: we have

978: \be

979: {d^2\over d\tau^2}\omega_{yz}+{\beta^2\over 1+\tau^2}\omega_{yz}=0 \ .

980: \label{eq:bh}

981: \ee

982: \citep{BH06}.

983: Figure \ref{fig:lin} plots numerical solutions of this equation, and shows that it matches 

984: the output from the pseudospectral code, as well as the analytic theory described below.

985: Given $\omega_{yz}$, it is trivial to construct $\bld{\omega}$ and $\bld{{u}}$

986: from

987: \be

988: \omega_x={\kappa\over 2\beta\Omega}{d\omega_{yz}\over d\tau}

989: \label{eq:oxeq}

990: \ee

991: and equations (\ref{eq:omrec}) and (\ref{eq:urec}).

992: 

993: %\clearpage

994: %%%%%%%%%%%%%begin figure

995: \begin{figure} %old \begin{figure*} .... \end{figure*}

996:   %\centering

997:  \hspace{-1.cm} \includegraphics[width=.55\textwidth]{f6.eps}

998:   \caption{{\bf Linear Evolution of Mode Amplitudes for Three Values 

999:   of $\beta$:} Time runs from right to left.

1000:   Solid curves show the exact, numerically integrated solution of equation (\ref{eq:bh}).

1001:   The initial value of $d\omega_{yz}/d\tau$ was chosen so that $\omega_B=0$ initially

1002:   (eq. [\ref{eq:powerlaw}]). 

1003:     Dashed lines show the analytic solution (eq. [\ref{eq:powerlaw}])

1004:   with constant $\omega_A$ and $\omega_B=0$ for $\tau>0$; while for $\tau<0$, the

1005:   normal mode amplitudes are set to

1006:   different constants that are given by equation (\ref{eq:trans}). 

1007:   We exclude the domain $|\tau|<1$ from the dashed curve, because the analytic

1008:   approximation does not apply there.

1009:    Circles show output

1010:   from the pseudospectral code, integrated with a timestep $dt=1/15$ and with the viscosity

1011:   set to zero. 

1012:     }

1013:   \label{fig:lin}

1014: \end{figure}

1015: %%%%%%%%%%%%%end figure

1016: %\clearpage

1017: 

1018: For highly leading or trailing modes ($|\tau|\gg 1$), equation (\ref{eq:bh}) has simple power-law solutions,

1019: \beqn

1020: \omega_{yz}&=&\omega_A|\tau|^{1-\delta\over 2}+\omega_B|\tau|^{1+\delta\over 2} \ , 

1021: \ \ |\tau|\gg 1 \label{eq:powerlaw}

1022: \eeqn

1023: \citep{BH06}, where  $\omega_A$ and $\omega_B$ are constants that

1024: we shall call the ``normal-mode'' amplitudes, and

1025: \be

1026: \delta \equiv \sqrt{1-4\beta^2} \ ,

1027: \ee

1028: which is imaginary for $|\beta|>1/2$.

1029: As a mode's wavevector evolves along a line in $k$-space, its

1030: amplitude is oscillatory if this line is much closer to the $k_z$ axis

1031: than to the $k_y$ one, and  non-oscillatory if the converse is true.

1032: The transition occurs at $|\beta|=1/2$.     This

1033: behavior may be understood as a competition between shear and epicyclic oscillations.  

1034: The timescale for $k_x$ to change by an order-unity factor due to the

1035: shear is $t_{\rm shear}\sim |k_x/\dot{k}_x|=|k_x/qk_y|$, and

1036: the timescale for epicyclic oscillations of axisymmetric modes

1037:  is $t_{\rm epi}\sim \kappa^{-1}|k_x/k_z|$ for 

1038: $|k_x|\gg |k_z|$.  

1039: Therefore $|\beta|\sim t_{\rm shear}/t_{\rm epi}$, and

1040: when $|\beta|\gg 1$ the epicyclic time is shorter and

1041: so the mode's amplitude oscillates as its wavevector is slowly advected by the shear.  But

1042:  when $|\beta|\ll 1$ the shear changes the wavevector faster

1043: than the amplitude can oscillate.

1044: 

1045: 

1046: The solution (\ref{eq:powerlaw}) breaks down in mid-swing.  

1047: As a swinging wave changes from leading to trailing, its ``normal-mode

1048: amplitudes''  change abruptly  on the timescale that $\tau$ changes from

1049: $\pm 1$ to $\mp 1$ via

1050: \be

1051: \left(\begin{array}{c} \omega_A \\ \omega_B\end{array}\right)_{\rm trail}

1052: =

1053: \left(\begin{array}{cc}T_{AA} & T_{AB} \\T_{BA} & T_{BB}\end{array}\right)

1054: \left(\begin{array}{c} \omega_A \\ \omega_B\end{array}\right)_{\rm lead}

1055: \label{eq:trans}

1056: \ee

1057: where the transition matrix has components

1058: \beqn

1059: T_{AA}&=&-T_{BB}=\csc\left({\delta\pi/ 2}\right) \\

1060: T_{BA}&=&

1061: -{\cot(\delta\pi/2)^2\over T_{AB}}

1062: =-2^{\delta+1}{1\over\delta}{1-\delta\over 1+\delta}

1063: {\Gamma(1+\delta/2)^2\over \Gamma(1/2+\delta/2)^2}

1064: \eeqn

1065: and determinant $=-1$, and hence is its own inverse.

1066:  The components are complex when $|\beta|>1/2$.

1067: To derive these components, we took advantage of the fact that equation (\ref{eq:bh}) has

1068: hypergeometric solutions   \citep{JG05,BH06}, and matched

1069: these onto the normal-mode solution given above.  We omit

1070: the unenlightening details.  

1071: 

1072: 

1073: \section{Nonlinear Evolution: Formation and Destruction of Vortices}

1074: \label{sec:nonlin}

1075: 

1076: 

1077: \subsection{Qualitative Description}

1078: 

1079: The instability that destroys  vortices is a generalization of 

1080: the one that forms them. 

1081:  We review here how vortices form,

1082: before describing the instability that destroys them.  In \S \ref{subsec:stab},

1083: we make this description quantitative. 

1084: 

1085: 

1086: Vortices form out of a nonlinear instability that involves vertically symmetric ($k_z=0$)

1087: modes. (See L07 for more details of the 2D dynamics than are presented here.) 

1088:   Consider the two vertically symmetric modes shown in Figure \ref{fig:kspace}: the

1089: ``mother'' mode at $(1,0,0)$ and the ``father'' mode that is depicted crossing through $(1,-1,0)$.

1090: Triplets of integers 

1091: $(j_x,j_y,j_z)$

1092: label  values of wavevectors $(k_x,k_y,k_z)$

1093: (for example, via [\ref{eq:kj}]).

1094: The mother is both axi- and vertically-symmetric, and the father is a leading swinging mode.

1095: 

1096: 

1097: As the father swings through radial symmetry, i.e. as it crosses through

1098: the point $(0,-1,0)$, its

1099:  velocity

1100: field  is

1101: strongly amplified by the background shear.  This can be seen from  

1102:  \S \ref{sec:lin}, which shows that swinging modes with $k_z=0$ have

1103:  $\omega_{yz}$=const., and hence $u_x=i(\omega_{yz}/k_{yz})/(1+\tau^2)$, which

1104:  becomes largest when $\tau$ crosses through 0.

1105: When the father is near the peak of its transient amplification ($|\tau|\lesssim 1$), 

1106: it couples most strongly with the mother, and they produce a ``son''

1107: near $(1,-1,0)=(1,0,0)+(0,-1,0)$.  The son will then swing through radial symmetry

1108: where it will  couple (oedipally) with the mother to produce a grandson near $(1,-1,0)$,

1109: which can repeat the cycle.

1110: We summarize this 2D instability feedback loop  as

1111: \beqn

1112: {\rm linear\ amplification:}& (1,-1,0)\rightarrow (0,-1,0)  \nonumber \\

1113: {\rm nonlinear\ coupling:}& (0,-1,0)+(1,0,0)\rightarrow(1,-1,0) \nonumber

1114: \eeqn

1115: The criterion for instability is simply that the amplitude of the son's $\omega_{yz}$ be

1116: larger than that of the father.  

1117: As shown in L07, if instability is triggered, its nonlinear outcome 

1118: in two dimensions is a long-lived vortex.

1119: 

1120: The three-dimensional instability that is responsible for destroying vortices

1121: is a straightforward generalization.   

1122: The mother mode is still at $(1,0,0)$, but now the father mode starts near 

1123: $(1,-1,1)$.

1124: Symbolically, the feedback loop is

1125: \beqn

1126: {\rm linear\ amplification:}& (1,-1,1)\rightarrow (0,-1,1)  \nonumber \\

1127: {\rm nonlinear\ coupling:}& (0,-1,1)+(1,0,0)\rightarrow(1,-1,1) \nonumber

1128: \eeqn

1129: The 2D instability described above is just a special case of this 3D one in the limit that $k_z=0$.

1130: In general, the stability of a mother  mode at $(1,0,0)$ with given $k_x=\bar{k}_x$ and 

1131: ${\bld\omega}=\bld{\bar{\omega}}$ 

1132: depends on the $k_y$ and $k_z$ of the 

1133:  father-mode perturbations (as well as on the parameters

1134: $q$ and $\Omega$). Which $k_y$ and $k_z$ are accessible in turn depends

1135:  on the dimensions $L_y\times L_z$ of the simulation box---or equivalently,

1136: on the circumferential distance around a disk and the scale-height.

1137: In \S \ref{subsec:stab}, we map out quantitatively  the  region in the

1138: $k_y-k_z$ plane that leads to instability.

1139: For now, it suffices to note that the unstable region has $|k_y|\lesssim |\bar{k}_x \bar{\omega}|/q$ and

1140: $|k_z|\lesssim |k_y|$.  

1141: We conclude that a given mother mode suffers one of three

1142: possible fates, depending on $L_y$ and $L_z$.

1143: \begin{enumerate}

1144: \item

1145: If $L_y$ is less than a critical value 

1146: ($\sim q/|\bar{\omega}\bar{k}_x|$), then the mother mode is

1147: stable to all perturbations. 

1148: \item

1149:  If $L_y$ is larger than this critical value, then the 

1150: mother mode is unstable to vertically symmetric ($k_z=0$) perturbations;  if in addition $L_z$ is sufficiently

1151: small that all modes with $k_z\ne 0$ are stable, then the  mother mode turns

1152: into a long-lived vortex (Figure \ref{fig:shortsim}).

1153:  \item If both $L_y$ and $L_z$ are sufficiently large, the mother mode

1154: is unstable both to vertically symmetric and to 3D perturbations.  When this happens, the mother

1155: starts to form a vortex, but this vortex is 3D-unstable.  The result is turbulence (Figure \ref{fig:tallsim}).

1156: \end{enumerate}

1157: There is also a possibility that is intermediate between numbers 2 and 3: if the conditions

1158: described in number 2 hold, the essentially 2D dynamics that results

1159: can nonlinearly produce new mother modes that are unstable to 3D perturbations.

1160: In this paper, we shall not consider this possibility further, since it did not occur in

1161: the pseudospectral simulations of \S \ref{sec:pseudo}.  

1162: We merely note that in our simulations

1163: of this possibility (not presented in this paper), we found that when the new mother modes

1164: decayed, they also destroyed the original mother mode.

1165: 

1166: 

1167: \subsection{The Stability Criterion}

1168: \label{subsec:stab}

1169: 

1170: %\clearpage

1171: %%%%%%%%%%%%%begin figure

1172: \begin{figure}

1173:   %\centering

1174:   \hspace{-1cm}\includegraphics[width=.55\textwidth]{f7.eps}

1175:   \caption{{\bf Nonlinear Evolution of 3D Instability:} Time runs from right to left.  In the two left panels, lines show numerical 

1176:   solutions of equations (\ref{eq:lin2}) and (\ref{eq:soneq}), as well as the grandson's

1177:   equation. Also shown as circles

1178:   are  the output from a pseudospectral simulation, showing excellent agreement

1179:   with the ``exact'' solutions.    The

1180:   following parameters have been chosen: $\Omega=1$, $q=3/2$, $\bar{\omega}=0.005$, $\bar{k}_x=2\pi\cdot 15$,

1181:   $k_y=-2\pi/30$, $k_z=0.45|k_y|\Rightarrow \beta=-0.3$. The small disagreement

1182:   between pseudospectral and exact solutions  for $\omega_x'$ at

1183:   $\tau<-2\bar{\tau}$ is due to the conjugate modes that, for simplicity,  we have not included in equations 

1184:   (\ref{eq:lin2}) and (\ref{eq:soneq}); see footnote \ref{foot:cc}.   The two right panels show the mode amplitudes, defined via equation (\ref{eq:sonnm})

1185:   for the son, and similarly for the father and grandson. Although these two right

1186:   panels contain the same information as the left ones, they are helpful

1187:   in constructing the analytic form of the growth factor $\chi$ (see Appendix).  With the parameters

1188:   chosen for this figure, equation (\ref{eq:amp}) predicts $\chi=-2.2$ for the amplification

1189: factor between successive generations, in agreement with that seen in the figure.

1190:   }

1191:   \label{fig:nonlin}

1192: \end{figure}

1193: %%%%%%%%%%%%%end figure

1194: %\clearpage

1195: 

1196: 

1197: To quantify the previous discussion, we choose an initial state as

1198: in Figure \ref{fig:kspace}, with the mother

1199: mode at $(1,0,0)$ and the father a leading mode crossing through $(1,-1,1)$. 

1200:  The son mode, not depicted in the figure,

1201:  is initially

1202: crossing through the point $(2,-1,1)$.  We set its initial vorticity---as well as the

1203: initial vorticity of all modes other than the mother and father---to 

1204: zero.\footnote{\label{foot:cc}We

1205: ignore the complex conjugate modes for simplicity.

1206: Since 

1207: $\bld{\omega(x)}$ is real-valued, each mode with wavevector 

1208: and amplitude $(\bld{k},\bld{\omega)}$ is accompanied by a

1209: conjugate mode that has $(-\bld{k},\bld{\omega}^*)$.

1210: In our initial state, there are really four modes with non-zero amplitudes: the mother

1211: at $(1,0,0)$ and its conjugate at $(-1,0,0)$, and the father and its conjugate.  We may ignore the conjugate

1212: modes because they do not affect the instability described here.  

1213: As shown in L07 for the 2D case, 

1214: their main effect  is that when the son swings through $(0,-1,1)$, not only 

1215: does it couple with the mother at $(1,0,0)$ to produce a grandson at $(1,-1,1)$, but

1216: it also couples with the conjugate mother at $(-1,0,0)$ to partially kill its father,

1217:  which is then at $(-1,-1,1)$ (bringing to mind the story of Oedipus).

1218: But since the father is a trailing mode at this time, it no longer participates

1219: in the instability.  Nonetheless, the conjugate modes do play a role in the nonlinear

1220: outcome of the instability.

1221: }

1222: The father's wavevector and Fourier amplitude are labelled as in \S\ref{sec:lin}, and

1223: the mother's and son's  are labelled with bars and primes:

1224: \beqn

1225: {\rm father:\ }&{\bld{{k}}}& \ , \

1226: {\bld{\omega}}

1227:  \nonumber \\

1228: {\rm mother:\ }&{\bld{\bar{k}}}&\equiv \bar{k}{\bld{\hat{x}}}  \ \  , \ 

1229: {\bld{\bar{\omega}}}\equiv{\bar{\omega}}{\bld{\hat{z}}} 

1230: \nonumber \\

1231: {\rm son:\ }&{\bld{k'}}& \equiv \bar{k}{\bld{\hat{x}}} +{\bld{k}} \  , \ 

1232: {\bld{\omega'}}

1233: \nonumber

1234: \eeqn

1235: Note that $\bar{k}$=const., and $\bar{\omega}_x=0$ because the vorticity

1236: must be transverse to the wavevector.

1237: We also set $\bar{\omega}_y=0$; otherwise $\bar{u}_z\ne 0$, which corresponds to

1238:  a mean flow out the top of the box  and  in through the bottom.

1239: 

1240: At early times, the father mode swings through the point $(0,-1,1)$. 

1241: Since the only other nonvanishing mode at this time is the mother, there 

1242: are no mode couplings  that can nonlinearly change the father's

1243: amplitude.

1244: Therefore its amplitude

1245: is governed by the linear equation (\ref{eq:linlim}), which we reproduce here as

1246: \be

1247: {d\over d\tau}

1248: \left(\begin{array}{c}{\omega}_x \\ {\omega}_{yz}\end{array}\right)

1249: =\beta{\Omega\over \kappa}

1250: \left(\begin{array}{cc}0 & -{1\over 2}{\kappa^2\over\Omega^2}{1\over 1+\tau^2} \\ 2 & 0\end{array}\right)

1251: \left(\begin{array}{c}{\omega}_x \\ {\omega}_{yz}\end{array}\right) \ .

1252: \label{eq:lin2}

1253: \ee

1254: During its swing, it couples with the mother to change the amplitude of the son. 

1255:  The linear part  of the son's evolution is given by the

1256: above equation with primed vorticity and wavevector in place of unprimed.

1257: The nonlinear part is given by \be

1258: {d\over dt}{\bld{\omega'}}\Big\vert_{\rm nonlin}=i\bld{

1259: k'\times(\bar{u}\times \omega+u\times\bar{\omega})

1260: }

1261: \ee

1262: (eq. [\ref{eq:eomo}])

1263: where  $\bld{\bar{u}}=-i(\bar{\omega}/\bar{k}){\bld{\hat{y}}}$ and

1264:  $\bld{u}=i\bld{k\times\omega}/k^2$ (eq. [\ref{eq:udef}]).

1265: Adding the linear and nonlinear parts, and re-expressing in terms of our chosen

1266: degrees of freedom, we find

1267: \beqn

1268: {d\over d\tau}

1269: \left(\begin{array}{c}{\omega}_x' \\ {\omega}_{yz}'\end{array}\right)

1270: =\beta{\Omega\over \kappa}

1271: \left(\begin{array}{cc}0

1272: \ \ \ %spaces

1273:  & -{1\over 2}{\kappa^2\over\Omega^2}{1\over 1+(\tau+\bar{\tau})^2} \\ 

1274: 2\ \ \ %spaces

1275:   & 0\end{array}\right)

1276: \left(\begin{array}{c}{\omega}_x' \\ {\omega}_{yz}'\end{array}\right) 

1277: \nonumber

1278: \\

1279: -{\bar{\omega}\over q}

1280: \left(\begin{array}{cc} {1\over \bar{\tau}}

1281: \ \ \ %spaces

1282: &{\beta q\over\kappa}{1\over 1+\tau^2} \\

1283:  0\ \ \  %spaces

1284:  &{1\over\bar{\tau}}- {\bar{\tau}\over 1+\tau^2}\end{array}\right)

1285:  \left(\begin{array}{c}{\omega}_x \\ {\omega}_{yz}\end{array}\right)  \ ,

1286:  \label{eq:soneq}

1287: \eeqn

1288: where the dimensionless constant

1289: \be

1290: \bar{\tau}\equiv {\bar{k}_x\over k_{yz}} \ 

1291: \ee

1292: depends on both the mother's and father's wavevectors.

1293: It is the father's $\tau\equiv k_x/k_{yz}$ that is being used as the time-coordinate

1294: for evolving the son's amplitude.

1295: The grandson's equation is the obvious extension: 

1296: denoting

1297: the grandson's amplitudes with double primes, one need only

1298: make the following replacements

1299: in equation (\ref{eq:soneq}):

1300:  $\bld{\omega'}\rightarrow\bld{\omega''}$,

1301: $\bld{\omega}\rightarrow\bld{\omega'}$, and

1302: $\tau\rightarrow \tau+\bar{\tau}$.

1303: Subsequent generations

1304: evolve  analogously.

1305: 

1306: The father's equation  (\ref{eq:lin2}) is easily solved, as shown in \S \ref{sec:lin}. 

1307: Inserting this solution into equation (\ref{eq:soneq}) produces a linear inhomogeneous

1308: equation for the son's amplitude, and similarly for the grandson's.

1309: Figure \ref{fig:nonlin} plots numerical solutions of these equations.  Also shown

1310:  as circles are output from a pseudospectral simulation,

1311: showing excellent agreement.

1312: 

1313: 

1314: %\clearpage

1315: %%%%%%%%%%%%%begin figure

1316: \begin{figure*}

1317:   \centering

1318:   \hspace{-3cm}\includegraphics[width=.4\textwidth]{f8a.eps}

1319:  \hspace{1cm} \includegraphics[width=.4\textwidth]{f8b.eps}  

1320:   \caption{{\bf Curves of Marginal Stability for a Mother Mode

1321:   With  $\bld{\bar{\omega}=0.005}$ (left panel) and

1322:   $\bld{\bar{\omega}=0.05}$ (right panel):}  Left panel corresponds to Figure \ref{fig:nonlin} 

1323:   and right panel corresponds to the pseudospectral simulations of Figures \ref{fig:shortsim}-\ref{fig:tallsim}.

1324:   We set $\Omega=1$ and $q=3/2$.

1325: To make the solid lines in these plots (the ``exact solutions''), 

1326: we repeated the integrations that produced the lines in Figure \ref{fig:nonlin}, but varying

1327: $k_z$ for each $k_y$ until perturbations neither grew nor decayed.  The dashed line

1328: in the left panel

1329: shows that the analytic approximation of equation (\ref{eq:amp}) agrees reasonably well

1330: with the exact solution.  We do not show equation (\ref{eq:amp}) in the right panel because

1331: the agreement is poorer there. Right panel shows two X's for the values of the

1332: smallest non-zero $|k_y|$ and $|k_z|$ in the simulations of Figures \ref{fig:shortsim}-\ref{fig:tallsim},

1333: i.e., $|k_y/\bar{k}|=L_x/L_y=0.067$ for both simulations, and $|k_z/\bar{k}|=L_x/L_z=0.13$

1334: for the short box and $=0.033$ for the tall box.  The tall box contains a 3D-unstable mode

1335: that leads to the destruction of the vortex into a turbulent-like state.  The short box contains

1336: no such mode, and is stable to 3D perturbations.

1337:  }

1338:   \label{fig:marg}

1339: \end{figure*}

1340: %%%%%%%%%%%%%end figure

1341: %\clearpage

1342: 

1343: 

1344: In the Appendix, we solve  equation (\ref{eq:soneq}) analytically to

1345:  derive the amplification factor $\chi$, which is the ratio of the son's amplitude

1346: at any point in its evolution (e.g., when it is radially symmetric), to the father's amplitude at

1347: the same point in its evolution. 

1348: We find 

1349: \be

1350: \chi=-{\bar{\omega}\over q}\bar{\tau}^\delta\sqrt{\pi}{1+\delta\over\delta^2}

1351: \left(

1352: 1+{q\Omega\over\kappa^2}(1-\delta)

1353: \right)

1354: {\Gamma(1+\delta/2)\over\Gamma(1/2+\delta/2)}

1355: \label{eq:amp}

1356: \ee

1357: where $\delta=\sqrt{1-4\beta^2}$.

1358: Equation (\ref{eq:amp}) is applicable in the limit $|\bar{\omega}|/q\ll 1$.

1359: For 2D modes ($\beta=0\Rightarrow \delta=1$), it recovers  equation

1360: 42  of L07 (see also eq. [\ref{eq:2dinst}] of this paper): 

1361: \be

1362: \chi_{2D}=-\pi{\bar{\omega}\over q}{\bar{k}\over k_y} \ .

1363: \ee

1364: 

1365: Marginally stable modes have $|\chi|=1$.  

1366: Figure \ref{fig:marg} plots curves of marginal stability. 

1367: The left panel is for the case $\bar{\omega}=0.005$, as in Figure \ref{fig:nonlin}, and the right

1368: panel is for $\bar{\omega}=0.05$, as in the pseudospectral simulations presented 

1369: at the outset of this paper (eq. [\ref{eq:initz}]; Figs. \ref{fig:shortsim}-\ref{fig:tallsim}).

1370: The left panel shows that equation (\ref{eq:amp}) gives a fair reproduction of the exact curve.

1371: We do not show equation (\ref{eq:amp}) in the right panel, because it gives poorer

1372: agreement there (since $|\bar{\omega}|/q$ is too large). In the right panel we also plot X's for the values

1373: of the smallest nonvanishing 

1374: 3D wavenumbers in the simulations of Figures \ref{fig:shortsim}-\ref{fig:tallsim}.

1375: In the short-box simulation, all 3D modes lie in the stable zone.  Therefore the dynamics remains

1376: two-dimensional.  But in the 3D box, there is a 3D mode in the unstable zone that destroys the vortex

1377: and gives rise to turbulent-like behavior.

1378: 

1379: It is interesting to consider briefly how the instability described here connects with the

1380: Rayleigh-unstable case, which occurs when $\kappa^2<0$ . 

1381:   At small $|k_y|$, the  marginally stable curves in Figure \ref{fig:marg}

1382:  are

1383: given by $|\beta|=1/2$, where $\beta=(\kappa/q)(k_z/k_y)$.  Hence if one decreases

1384: $\kappa$ from its Keplerian value $\Omega$, the marginally stable curve becomes

1385: steeper in the $k_z-k_y$ plane, and an increasing number of 3D modes become

1386: unstable.  As $\kappa\rightarrow 0$, if a 2D mode with some $k_y$ is unstable, then

1387: so are all 3D modes with the same $k_y$.  Therefore any 2D-unstable state is also

1388: 3D-unstable, and any forming vortex would decay into turbulence.

1389: 

1390: 

1391: 

1392: \section{Conclusions}

1393: \label{sec:conc}

1394: 

1395: 

1396: Our main result follows from Figure \ref{fig:marg}, which maps

1397: out the stability of a 

1398: ``mother mode'' (i.e., a mode with wavevector $\bar{k}\bld{\hat{x}}$

1399:  and amplitude $\bar{\omega}$)

1400: to nonaxisymmetric 3D perturbations. 

1401: A mother mode is unstable 

1402: provided that the $k_y$ and $k_z$ of the

1403:  nonaxisymmetric perturbations  satisfy both

1404: $|k_y|\lesssim \bar{k}\bar{\omega}/q$ and $|k_z|\lesssim |k_y|$, dropping

1405: order-unity constants.

1406: Based on this result, we may understand the formation, survival, 

1407: and destruction of vortices.

1408: Vortices form out of mother modes that are unstable to 2D ($k_z=0$)

1409: perturbations. Mother modes that are unstable to 2D modes but stable

1410: to 3D ($k_z\ne 0$) ones, form into long-lived vortices.  Mother modes

1411: that are unstable to both 2D and 3D modes are destroyed.

1412: Therefore a mother mode with given $\bar{k}$ and $\bar{\omega}$

1413: will form into a vortex if the disk has a sufficiently large circumferential

1414: extent and a sufficiently small scale height, i.e., if $r\gtrsim \bar{k}^{-1}q/\bar{\omega}$

1415: and $h\lesssim \bar{k}^{-1}q/\bar{\omega}$, where $r$ is the distance

1416: to the center of the disk, and $h$ is the scale height.

1417: Alternatively, 

1418: the mother mode will be destroyed in a turbulent-like state

1419: if  both $r$ and $h$ are sufficiently large

1420:  ($r\gtrsim \bar{k}^{-1}q/\bar{\omega}$

1421:   and

1422: $h\gtrsim \bar{k}^{-1}q/\bar{\omega}$).

1423: 

1424: 

1425: Our result

1426: has a number of astrophysical consequences.

1427: In protoplanetary disks that do not contain any vortices,

1428: solid particles drift inward. 

1429: Gas disks orbit at sub-Keplerian speeds,  

1430: $v_{\rm gas}\sim \Omega r (1-\eta)$, where 

1431: $\Omega r$ is the Keplerian speed and $\eta\sim  (c_s/\Omega r)^2$, with

1432: $c_s$  the sound speed.  Since solid particles  would orbit

1433: at the Keplerian speed in the absence of gas, the mismatch of speeds

1434: between solids and gas produces a drag on the solid particles, removing

1435: their angular momentum and causing them to fall into the star.

1436: For 

1437: example, in the minimum mass solar nebula, meter-sized

1438: particles fall in from 1 AU in around a hundred years.  

1439: This rapid infall presents

1440: a serious problem for theories of planet formation, since it is

1441: difficult to produce planets out of dust in under a hundred years.

1442: Vortices can solve this problem \citep{BS95}.

1443: A vortex that has 

1444: excess vorticity $-\bar{\omega}$ 

1445: and radial width $1/\bar{k}$ can halt the infall of particles

1446: provided that $\bar{\omega}/\bar{k}\gtrsim (\Omega r)\eta$, 

1447: because the gas speed induced by such a vortex more than 

1448: compensates for

1449: the sub-Keplerian speed induced by gas pressure.\footnote{

1450: We implicitly assume here that the stopping time of the particle

1451: due to gas drag

1452: is comparable to the orbital time, which is true for meter-sized particles

1453: at 1 AU in the minimum mass solar nebula.  A more careful treatment

1454: shows that

1455: a vortex can stop a particle with stopping time $t_s$ provided

1456: that  $\bar{\omega}/\bar{k}\gtrsim (\Omega r)(\Omega t_s)\eta$ 

1457: \citep{Youdin08}.

1458: }

1459: Previous simulations implied that 3D vortices rapidly 

1460: decay, and so cannot  prevent the rapid infall of solid

1461: particles \citep{BM05b,SSG06}.  Our result

1462: shows that vortices can survive within disks, and so restores

1463: the viability of vortices as a solution to the infall problem. 

1464: 

1465: 

1466: A more important---and more speculative---application of our result is

1467: to the transport of angular momentum within neutral accretion disks. 

1468: In our simulation of a vortex in a tall box, we found that as the vortex 

1469: decayed it transported angular momentum outward at a nearly

1470: constant rate for hundreds of orbital times.

1471:  If decaying vortices transport a significant amount of angular momentum

1472: in disks, they would resolve one of the most important outstanding questions

1473: in astrophysics today: what causes hydrodynamical accretion disks to accrete?

1474: To make this speculation more concrete, one must understand the amplitude

1475: and duration of the ``turbulence'' that results from decaying vortices.  

1476: This is a topic for future research.

1477: 

1478: In this paper, we considered only the effects of rotation and shear on

1479: the stability of vortices, while we neglected the

1480: effect of vertical gravity.  

1481: There has been a lot of research in the geophysical community 

1482: on the dynamics of fluids in the presence of vertical gravity, since 

1483: stably stratified fluids are very common on Earth---in the atmosphere, oceans, 

1484: and lakes.  In numerical and laboratory experiments of strongly stratified

1485: flows, thin horizontal ``pancake vortices'' often form, and fully developed 

1486: turbulence is characterized by thin horizontal layers. 

1487: \citep[e.g., ][]{BBLC07}.  Pancake vortices are stabilized by vertical gravity,

1488: in contrast to the vortices studied in this paper which are stabilized by rotation.

1489: Gravity inhibits vertical motions because of buoyancy: it costs gravitational energy

1490: for fluid to move vertically.

1491: The resulting quasi-two-dimensional

1492: flow can form into a vortex.\footnote{\cite{BC00} 

1493: show that

1494: a vertically uniform

1495: vortex column in a stratified (and non-rotating and non-shearing) 

1496:  fluid suffers an instability (the ``zigzag instability'') that is 

1497: characterized by a typical vertical lengthscale $\lambda_z\sim U/N$, where $U$ is 

1498: the horizontal speed induced by the vortex, $N$ is the Brunt-V\"ais\"al\"a frequency,

1499: and the horizontal lengthscale of the vortex $L_h$ is assumed to be much

1500: greater that  $\lambda_z$ (hence the pancake structure).

1501:  We may understand

1502: Billant \& Chomaz's result in a crude fashion with an argument similar to that employed in the introduction

1503: to explain the destruction of rotation-stabilized vortices: since the frequency

1504: of buoyancy waves is $Nk_x/k_z$ (when $|k_x|\ll |k_z|$), and since the frequency

1505: at which fluid circulates around a vortex is $U/L_h\sim k_xU$, there is a resonance between

1506: these two frequencies for vertical lengthscale $1/k_z\sim U/N$.} 

1507: We may speculate that in an astrophysical disk vertical gravity provides an 

1508: additional means to stabilize vortices, in addition to rotation.  

1509: But to make this speculation concrete, the theory presented in this paper

1510: should be extended to include vertical gravity.

1511: 

1512: 

1513: 

1514: We have not addressed in this paper the origin of the axisymmetric structure

1515: (the mother modes) that give rise to surviving or decaying vortices.

1516: One possibility is that decaying vortices can produce more axisymmetric

1517: structure, and therefore they can lead to self-sustaining turbulence.

1518: This seems to us unlikely.

1519: We have not seen evidence for it in our simulations, but this could

1520: be  because

1521: of the modest resolution of our simulations.  

1522: Other possibilities for the generation of axisymmetric structure include thermal instabilities, such

1523: as the baroclinic instability, or convection, or

1524: stirring by planets.  This, too, is a topic for future

1525: research.

1526: 

1527: 

1528: 

1529: 

1530: 

1531: \appendix

1532: \section{Analytic Expression for Growth Factor $\chi$}

1533: 

1534: In this Appendix, we derive equation (\ref{eq:amp}) by analytically

1535: integrating equation (\ref{eq:soneq}) for the son's vorticity, given the father's vorticity

1536: as a function of time (\S \ref{sec:lin}),

1537: and taking the mother's vorticity $\bar{\omega}$

1538:  to be constant, which is valid when the father's amplitude is 

1539: small relative to the mother's.

1540: The numerical integral of equation (\ref{eq:soneq})  is shown in Figure \ref{fig:nonlin}.

1541: Recall that initially  $\tau=\bar{\tau}>0$ and $\tau$ decreases in time, and typically

1542: $\bar{\tau}\gg 1$.

1543: It simplifies the analysis to work with the son's ``normal-mode'' amplitudes $\omega_A'$ and

1544: $\omega_B'$, defined from $\omega_x'$ and $\omega_{yz}'$ via 

1545: (eqs. [\ref{eq:oxeq}] and [\ref{eq:powerlaw}])

1546: \be

1547: \label{eq:sonnm}

1548: \left(\begin{array}{c}{\omega}_x' \\ {\omega}_{yz}'\end{array}\right)

1549: =

1550: \left(\begin{array}{cc}

1551: {\kappa\over 2\beta\Omega}{1-\delta\over 2\tau'}|\tau'|^{1-\delta\over 2} & 

1552: {\kappa\over 2\beta\Omega}{1+\delta\over 2\tau'}|\tau'|^{1+\delta\over 2} 

1553:  \\

1554: |\tau'|^{1-\delta\over 2}   

1555:   & 

1556: |\tau'|^{1+\delta\over 2}  

1557:   \end{array}\right)

1558: \left(\begin{array}{c}{\omega}_A' \\ {\omega}_B'\end{array}\right) \ .

1559: \ee

1560: where

1561: \be

1562: \tau'\equiv\tau+\bar{\tau}

1563: \ee

1564: Substituting this into equation (\ref{eq:soneq}), the time derivative of the above

1565: matrix cancels the homogeneous term in that equation if we approximate

1566: $1+\tau^{'2}\simeq  \tau^{'2}$, which holds until just before the time

1567: that $\tau=-\bar{\tau}$.  The inhomogeneous term produces

1568: \beqn

1569: {d\over d\tau}\omega_A'={\bar{\omega}\over q}{1\over\delta}|\tau'|^{1+\delta\over 2}

1570: \left(

1571: \omega_{yz}

1572: \left[

1573: {1\over 1+\tau^2}

1574: \left(

1575: {2\beta^2q\Omega\over\kappa^2}+{1+\delta\over 2}{\bar{\tau}\over\tau'}

1576: \right)

1577: -{1+\delta\over 2}{1\over \bar{\tau}\tau'}

1578: \right]

1579: +\omega_x{2\beta\Omega\over\kappa}{1\over{\bar{\tau}}}

1580: \right) \ \ , \ \ \tau\gtrsim -\bar{\tau}

1581: \label{eq:aprime}

1582: \eeqn

1583: Since $\omega_x$ and $\omega_{yz}$ are known (\S \ref{sec:lin}), 

1584: a straightforward integration yields $\omega_A'$ just before $\tau=-\bar{\tau}$.

1585: To perform this integral, we resort to some approximations, guided by the solution

1586: shown in Figure \ref{fig:nonlin}.

1587: For the first term, we need

1588: \beqn

1589: \int_{\bar{\tau}}^{-\bar{\tau}}\omega_{yz}{(\tau+\bar\tau)^{1+\delta\over 2}\over 1+\tau^2}d\tau

1590: &\approx&

1591: %-{2\Omega\over\beta\kappa}\int_\epsilon^{-\epsilon}{d\omega_x\over d\tau}

1592: %(\tau+\bar\tau)^{1+\delta\over 2}d\tau+

1593: -{1\over \beta^2}\int_{\bar{\tau}}^{-\epsilon\bar\tau}

1594: {d^2\omega_{yz}\over d\tau^2}(\tau+\bar\tau)^{1+\delta\over 2}

1595: d\tau

1596: +

1597: \omega_B\int_{-\epsilon\bar\tau}^{-\bar\tau}

1598: |\tau|^{{1+\delta\over 2}-2}

1599: (\tau+\bar\tau)^{1+\delta\over 2}

1600: d\tau 

1601: \\

1602: &\approx&

1603: \omega_B\bar{\tau}^\delta{1+\delta\over 2\beta^2}\epsilon^{-1+\delta\over 2}

1604: -\omega_B\bar{\tau}^\delta\int_{\epsilon}^1s^{-3+\delta\over 2}(1-s)^{1+\delta\over 2}ds\\

1605: &\approx&

1606: \omega_B{\bar\tau}^\delta2^{-\delta}\sqrt{\pi}{1+\delta\over1-\delta}

1607: {\Gamma(1/2+\delta/2)\over \Gamma(1+\delta/2)}

1608: \eeqn

1609: where $\epsilon$ is a parameter that satisfies $1\gg \epsilon\gg 1/\bar\tau$.  In the first line,

1610: we used equation (\ref{eq:bh}),

1611: and

1612: we discarded the $\omega_A$ mode from the second integral 

1613: because the $\omega_B$ mode increases faster

1614: with increasing $|\tau|$. 

1615: From Figure \ref{fig:nonlin}, $\omega_A'$ nearly vanishes until $\tau\simeq 0$.  Therefore,

1616: in the second line we approximated the first integral as 

1617: $-(\bar{\tau}^{1+\delta\over 2}/\beta^2){d\omega_{yz}/d\tau}\vert_{-\epsilon}$.

1618: The third line holds in the limit of small $\epsilon$.

1619: The other three terms in equation (\ref{eq:aprime}) are integrated similarly, yielding

1620: \be

1621: \omega_A'=\omega_B{\bar{\omega}\over q}{1\over\delta}{\bar\tau}^\delta 2^{-\delta-1}\sqrt{\pi}

1622: {\Gamma(1/2+\delta/2)\over \Gamma(1+\delta/2)}

1623: \left(

1624: 1+{q\Omega\over\kappa^2}(1-\delta)

1625: \right){(1+\delta)^2\over 1-\delta}

1626: \label{eq:omap}

1627: \ee

1628: just before the time when $\tau=-\bar{\tau}$, i.e., just before the son is radially symmetric. 

1629: At this time, Figure \ref{fig:nonlin} shows that

1630: $\omega_B'$ very nearly vanishes.  Therefore, just after the son is radially symmetric, 

1631: it will have $\omega_B'=T_{BA}{\omega_A'}$ (eq. [\ref{eq:trans}]), with $\omega_A'$

1632: given by equation (\ref{eq:omap}). This gives the amplification factor $\chi\equiv\omega_B'/\omega_B$

1633: that is displayed in equation (\ref{eq:amp}).

1634: 

1635: 

1636:  

1637:  

1638: 

1639: 

1640: \bibliographystyle{apj}

1641: %\bibliography{shearbib}

1642: 

1643: \begin{thebibliography}{20}

1644: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi

1645: 

1646: \bibitem[{{Afshordi} {et~al.}(2005){Afshordi}, {Mukhopadhyay}, \&

1647:   {Narayan}}]{AMN05}

1648: {Afshordi}, N., {Mukhopadhyay}, B., \& {Narayan}, R. 2005, \apj, 629, 373

1649: 

1650: \bibitem[{{Balbus} \& {Hawley}(1998)}]{BH98}

1651: {Balbus}, S.~A. \& {Hawley}, J.~F. 1998, Reviews of Modern Physics, 70, 1

1652: 

1653: \bibitem[{{Balbus} \& {Hawley}(2006)}]{BH06}

1654: ---. 2006, \apj, 652, 1020

1655: 

1656: \bibitem[{{Barge} \& {Sommeria}(1995)}]{BS95}

1657: {Barge}, P. \& {Sommeria}, J. 1995, \aap, 295, L1

1658: 

1659: \bibitem[{{Barranco} \& {Marcus}(2005)}]{BM05b}

1660: {Barranco}, J.~A. \& {Marcus}, P.~S. 2005, \apj, 623, 1157

1661: 

1662: \bibitem[{{Barranco} \& {Marcus}(2006)}]{BM06}

1663: ---. 2006, Journal of Computational Physics, 219, 21

1664: 

1665: \bibitem[{{Billant} \& {Chomaz}(2000)}]{BC00}

1666: {Billant}, P. \& {Chomaz}, J.-M. 2000, Journal of Fluid Mechanics, 419, 29

1667: 

1668: \bibitem[{{Brethouwer} {et~al.}(2007){Brethouwer}, {Billant}, {Lindborg}, \&

1669:   {Chomaz}}]{BBLC07}

1670: {Brethouwer}, G., {Billant}, P., {Lindborg}, E., \& {Chomaz}, J.-M. 2007,

1671:   Journal of Fluid Mechanics, 585, 343

1672: 

1673: \bibitem[{{Chagelishvili} {et~al.}(2003){Chagelishvili}, {Zahn}, {Tevzadze}, \&

1674:   {Lominadze}}]{CZTL03}

1675: {Chagelishvili}, G.~D., {Zahn}, J.-P., {Tevzadze}, A.~G., \& {Lominadze}, J.~G.

1676:   2003, \aap, 402, 401

1677: 

1678: \bibitem[{{Drazin} \& {Reid}(2004)}]{DR04}

1679: {Drazin}, P.~G. \& {Reid}, W.~H. 2004, {Hydrodynamic Stability} (Hydrodynamic

1680:   Stability, by P.~G.~Drazin and W.~H.~Reid, pp.~626.~ISBN

1681:   0521525411.~Cambridge, UK: Cambridge University Press, September 2004.)

1682: 

1683: \bibitem[{{Gammie} \& {Menou}(1998)}]{GM98}

1684: {Gammie}, C.~F. \& {Menou}, K. 1998, \apjl, 492, L75+

1685: 

1686: \bibitem[{{Gill}(1965)}]{Gill65}

1687: {Gill}, A.~E. 1965, Journal of Fluid Mechanics, 21, 503

1688: 

1689: \bibitem[{{Godon} \& {Livio}(1999)}]{GL99}

1690: {Godon}, P. \& {Livio}, M. 1999, \apj, 523, 350

1691: 

1692: \bibitem[{{Johnson} \& {Gammie}(2005)}]{JG05}

1693: {Johnson}, B.~M. \& {Gammie}, C.~F. 2005, \apj, 635, 149

1694: 

1695: \bibitem[{{Lerner} \& {Knobloch}(1988)}]{LK88}

1696: {Lerner}, J. \& {Knobloch}, E. 1988, Journal of Fluid Mechanics, 189, 117

1697: 

1698: \bibitem[{{Lithwick}(2007)}]{L07}

1699: {Lithwick}, Y. 2007, \apj, 670, 789

1700: 

1701: %\bibitem[{{Lithwick} \& {Chiang}(2007)}]{LC}

1702: %{Lithwick}, Y. \& {Chiang}, E. 2007, \apj, 656, 524

1703: 

1704: \bibitem[{{Lovelace} {et~al.}(1999){Lovelace}, {Li}, {Colgate}, \&

1705:   {Nelson}}]{Love99}

1706: {Lovelace}, R.~V.~E., {Li}, H., {Colgate}, S.~A., \& {Nelson}, A.~F. 1999,

1707:   \apj, 513, 805

1708: 

1709: \bibitem[{{Lynden-Bell} \& {Pringle}(1974)}]{LP}

1710: {Lynden-Bell}, D. \& {Pringle}, J.~E. 1974, \mnras, 168, 603

1711: 

1712: \bibitem[{{Marcus}(1993)}]{Marcus93}

1713: {Marcus}, P.~S. 1993, \araa, 31, 523

1714: 

1715: \bibitem[{{Maron} \& {Goldreich}(2001)}]{MG01}

1716: {Maron}, J. \& {Goldreich}, P. 2001, \apj, 554, 1175

1717: 

1718: \bibitem[{{Rogallo}(1981)}]{Rogallo81}

1719: {Rogallo}, R.~S. 1981, NASA STI/Recon Technical Report N, 81, 31508

1720: 

1721: \bibitem[{{Saffman}(1995)}]{Saffman95}

1722: {Saffman}, P.~G. 1995, {Vortex Dynamics} (Vortex Dynamics, by P.~G.~Saffman,

1723:   pp.~325.~ISBN 0521477395.~Cambridge, UK: Cambridge University Press, February

1724:   1995.)

1725: 

1726: \bibitem[{{Sano} {et~al.}(2000){Sano}, {Miyama}, {Umebayashi}, \&

1727:   {Nakano}}]{SMUN00}

1728: {Sano}, T., {Miyama}, S.~M., {Umebayashi}, T., \& {Nakano}, T. 2000, \apj, 543,

1729:   486

1730: 

1731: \bibitem[{{Shen} {et~al.}(2006){Shen}, {Stone}, \& {Gardiner}}]{SSG06}

1732: {Shen}, Y., {Stone}, J.~M., \& {Gardiner}, T.~A. 2006, \apj, 653, 513

1733: 

1734: \bibitem[{{Stone} \& {Balbus}(1996)}]{SB96}

1735: {Stone}, J.~M. \& {Balbus}, S.~A. 1996, \apj, 464, 364

1736: 

1737: 

1738: \bibitem[{{Umurhan} \& {Regev}(2004)}]{UR04}

1739: {Umurhan}, O.~M. \& {Regev}, O. 2004, \aap, 427, 855

1740: 

1741: \bibitem[{{Yecko}(2004)}]{Yecko04}

1742: {Yecko}, P.~A. 2004, \aap, 425, 385

1743: 

1744: \bibitem[{{Youdin}(2008)}]{Youdin08}

1745: {Youdin}, A. 2008, ArXiv e-prints, 807

1746: 

1747: \end{thebibliography}

1748: 

1749: 

1750: \end{document}

1751: 

1752: 

1753: