1: %\documentclass[12pt,preprint]{aastex}
2:
3: \documentclass{emulateapj}
4:
5: \usepackage{graphicx}
6:
7: \slugcomment{The Astrophysical Journal, In Press}
8:
9: \shorttitle{UNSTABLE DISK GALAXIES. II.}
10: \shortauthors{M. A. Jalali}
11:
12: \begin{document}
13:
14: \title{UNSTABLE DISK GALAXIES. II. THE ORIGIN OF GROWING AND STATIONARY MODES}
15:
16: \author{Mir Abbas Jalali
17: %\altaffilmark{1,2}
18: }
19: \affil{Sharif University of Technology, Azadi Avenue, Tehran, Iran;
20: mjalali@sharif.edu \\
21: Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540}
22:
23: %\altaffiltext{1}{E-mail: mjalali@sharif.edu}
24:
25: \begin{abstract}
26:
27: I decompose the unstable growing modes of stellar disks to
28: their Fourier components and present the physical mechanism
29: of instabilities in the context of resonances. When the equilibrium
30: distribution function is a non-uniform function of the orbital
31: angular momentum, the capture of stars into the corotation
32: resonance imbalances the disk angular momentum and triggers
33: growing bar and spiral modes. The stellar disk can then recover
34: its angular momentum balance through the response of non-resonant
35: stars. I carry out a complete analysis of orbital structure
36: corresponding to each Fourier component in the radial angle,
37: and present a mathematical condition for the occurrence of
38: van Kampen modes, which constitute a continuous family.
39: I discuss on the discreteness and allowable pattern speeds
40: of unstable modes and argue that the mode growth is saturated
41: due to the resonance overlapping mechanism. An individually
42: growing mode can also be suppressed if the corotation and
43: inner Lindblad resonances coexist and compete to capture a
44: group of stars. Based on this mechanism,
45: I show that self-consistent scale-free disks with a sufficient
46: distribution of non-circular orbits should be stable under
47: perturbations of angular wavenumber $m>1$. I also derive a
48: criterion for the stability of stellar disks against
49: non-axisymmetric excitations.
50: \end{abstract}
51:
52: \keywords{stellar dynamics,
53: instabilities,
54: methods: analytical,
55: galaxies: kinematics and dynamics,
56: galaxies: spiral,
57: galaxies: structure}
58:
59: \section{INTRODUCTION}
60: \label{sec:intro}
61:
62: Both $N$-body simulations \citep{H71} and analytical methods
63: (Kalnajs 1978; Jalali \& Hunter 2005, hereafter JH) show that global
64: instabilities can generate barred structures in galactic disks.
65: Apart from the bar mode, which is an isolated event in frequency
66: space, global spiral modes seen in the eigenspectra of cored stellar
67: disks (Jalali 2007, hereafter Paper I) constitute a discrete family
68: that bifurcates form stationary \citet{vK55} modes. But not all
69: spiral structures are global modes as disturbances induced by close
70: neighbors \citep{BH92} and density inhomogeneities \citep{T90} may
71: also create the spiral patterns of the observed galaxies.
72:
73: A mode of a stellar disk is a mathematical entity that comes out of
74: an eigenvalue problem. However, its physical origin in isolated
75: systems has not yet been understood clearly. Lynden-Bell \& Kalnajs
76: (1972, hereafter LBK) attempted to explain a mode through the
77: transport of angular momentum between different parts of the disk.
78: They suggested that the inner Lindblad resonance (ILR) releases the
79: angular momentum of central regions and the spiral structure
80: transports it to the outer parts through the corotation (CR) and
81: outer Lindblad resonances (OLR). This mechanism is favored by some
82: galactic dynamicists \citep{ATHA03}, but it is seriously challenged
83: by Toomre's (1981) theory that says that feedback through the
84: galactic center is a critical ingredient for growing modes. In
85: Toomre's theory, on the other hand, the modeling of feedback as the
86: reflection of a leading spiral wave at the galactic center and its
87: emergence as a trailing one, is a simple description of a very
88: complicated dynamics that governs the motions of stars. Unresolved
89: issues concerning the evolution of unstable modes include the
90: following: (i) The swing amplification theory is not capable of
91: predicting the fate of a growing mode against other stationary
92: and unstable modes that coexist in the eigenspectrum of a given
93: model. (ii) How does the bar mode saturate? \citep{KJKJ07}.
94: (iii) Why does the nonlinear bar terminate almost at the corotation
95: radius? \citep{S81} (iv) We should also understand the origin of
96: different species in an eigenspectrum and interpret their continuous
97: or discrete nature, and the distribution of their pattern speeds and
98: growth rates.
99:
100: Stellar orbits in a galactic disk begin to evolve once the surface
101: density deviates from its equilibrium state,
102: $\Sigma_0(\textbf{\textit{x}})$, and develops a time-varying
103: mean-field potential $V_1(\textbf{\textit{x}},t)$. Here
104: $\textbf{\textit{x}}$ denotes the position vector of stars at the
105: time $t$. In the linear regime, we are usually interested in density
106: waves that grow/decay according to the exponential law $e^{st}$
107: and rotate with the fixed pattern speed $\Omega_p$. For a wave
108: of $m$-fold symmetry, the perturbed potential becomes
109: %
110: \begin{equation}
111: V_1=\epsilon e^{st} \tilde V
112: \left (\textbf{\textit{x}},m\Omega_p t \right ).
113: \label{eq:perturbed-potnetial-introduction}
114: \end{equation}
115: %
116: Since we are dealing with infinitesimal perturbations,
117: I have introduced the small parameter $\epsilon$ so that
118: $\epsilon e^{st}\ll 1$.
119: From (\ref{eq:perturbed-potnetial-introduction}) one
120: arrives at the equations of motion
121: %
122: \begin{equation}
123: \dot \textbf{\textit{x}} = \textbf{\textit{v}},~~
124: \dot \textbf{\textit{v}} =
125: -\frac{\partial V_0}{\partial \textbf{\textit{x}} } -
126: \epsilon e^{st}
127: \frac{\partial \tilde V}{\partial \textbf{\textit{x}} },
128: \label{eq:equation-motion-for-x-and-v}
129: \end{equation}
130: %
131: where $V_0(\textbf{\textit{x}})$ is the equilibrium potential
132: field generated by galactic stars and a possible dark matter
133: halo. In writing equation (\ref{eq:equation-motion-for-x-and-v}),
134: I have assumed that the motion of stars is restricted to the disk
135: plane. When the equilibrium state is axisymmetric and the dark
136: component is spherical, $V_0$ becomes a function of radial distance
137: to the galactic center and the unperturbed equations (with $\epsilon=0$)
138: are integrable. In such a circumstance, the phase space is filled
139: by rosette orbits denoted by $[\textbf{\textit{x}}_0(t),\textbf{\textit{v}}_0(t)]$.
140: The growth of perturbations, whatever the magnitude of $\epsilon e^{st}\ll 1$
141: may be, deforms stellar orbits. Orbital deformations are measured by
142: $\tilde \textbf{\textit{x}}=\textbf{\textit{x}}-\textbf{\textit{x}}_0$
143: and $\tilde \textbf{\textit{v}}=\textbf{\textit{v}}-\textbf{\textit{v}}_0$,
144: which can be used in (\ref{eq:equation-motion-for-x-and-v}) to obtain
145: %
146: \begin{equation}
147: \frac{d\tilde \textbf{\textit{x}} }{dt}=\tilde \textbf{\textit{v}},~~
148: \frac{d\tilde \textbf{\textit{v}} }{dt}=- \left [
149: \frac{\partial^2 V_0}{\partial \textbf{\textit{x}}^2 }
150: \right ]_{ \textbf{\textit{x}}_0 } \!\!\!\! \cdot \tilde \textbf{\textit{x}}
151: - \epsilon e^{st} \left [
152: \frac{\partial \tilde V}{\partial \textbf{\textit{x}} }
153: \right ]_{ \textbf{\textit{x}}_0 }.
154: \label{eq:equation-motion-for-tilde-x-and-v}
155: \end{equation}
156: %
157:
158: Although a proper equilibrium distribution function $f_0(\textbf{\textit{x}},\textbf{\textit{v}})$ can self-consistently
159: reproduce $\Sigma_0(\textbf{\textit{x}})$ using rosette orbits,
160: the perturbed density $\Sigma_1(\textbf{\textit{x}},t)$
161: (corresponding to $V_1$) cannot be supported by rosette
162: orbits alone and orbital deformations are necessary for
163: the self-consistency of density waves.
164: According to equations (\ref{eq:equation-motion-for-x-and-v})
165: and (\ref{eq:equation-motion-for-tilde-x-and-v}), orbital
166: deformations of ${\cal O}\left ( \epsilon e^{st} \right )$
167: are sufficient to support the growth of density/potential
168: perturbations up to the same order of magnitude of such
169: deformations over a time scale of $1/{\cal O}\left ( \epsilon e^{st} \right )$.
170: As the time is elapsed, the amplitude of perturbations increases
171: exponentially and the solution of the linearized collisionless
172: Boltzmann equation (CBE) fails when $\epsilon e^{st}\sim 1$.
173: During my mode calculations, I realized that the orbital axes
174: of certain stars librate in a coordinate frame that rotates with
175: the density pattern. This {\it resonant capture} initially seemed
176: to be a higher-order nonlinear effect but further experiments
177: showed that the resonant gap is constrained by the magnitude
178: of density perturbations. The complex behavior of stars for
179: infinitesimally small yet non-zero $\epsilon e^{st}\ll 1$,
180: and the role of resonant stars in the generation of discrete
181: galactic modes, are investigated in this paper.
182:
183: I use the results of Paper I and introduce a new dynamical
184: mechanism that sparks unstable modes and governs the singular
185: oscillations of \citet{vK55} modes. Resolving the origin of
186: instabilities and amplitude saturation precede my nonlinear
187: calculations, which were made feasible in Paper I by the
188: Petrov-Galerkin method and reducing the CBE to a system of
189: nonlinear ordinary differential equations. Those reduced equations,
190: however, are valid only when orbits are regular and averaging
191: over angle variables is allowed. As unstable modes grow, chaotic
192: orbits come into existence and the weighted residual form of the
193: CBE must be modified to handle them. I quote some of the results
194: of such modifications in this paper when I discuss the issue of
195: mode saturation. In Paper III, I will give a full account of the
196: mathematical and numerical modeling of stochastic layers, and will
197: analyze modal interactions after their saturation phase.
198:
199: For the cored exponential disk embedded in the field of the cored
200: logarithmic potential, I describe the decomposed Fourier components
201: of unstable bar and spiral modes in \S\ref{sec:mode-decomposition}
202: and highlight the existence of a phase shift between different
203: components. In \S\ref{sec:capture-into-resonances}, I derive a
204: condition for the corotation of the orbital axes of an ensemble of
205: stars and explain the role of such a synchronous motion in pattern
206: formation. I dedicate \S\ref{sec:perturbed-stellar-dynamics} to
207: exploring the orbital structure of a perturbed stellar disk and
208: identify a resonance mechanism that can generate both stationary and
209: growing modes. I reveal the mechanism of angular momentum transfer
210: between Fourier components and derive analytical expressions for
211: the growth of resonance zone. I address the origin of instabilities
212: in \S\ref{sec:origin-of-instabilities}, present a saturation
213: mechanism for unstable modes in \S\ref{sec:mode-saturation},
214: and discuss about the global stability of soft-centered and
215: scale-free disks. I explain the restrictions of LBK's mode mechanism
216: in \S\ref{sec:discussion-and-conclusions} and end up the paper
217: with concluding remarks.
218:
219:
220: \section{THE MODEL}
221: \label{sec:used-model}
222:
223: The calculations of the present study are carried out for the cored
224: exponential disks of JH whose eigenfrequency spectra and mode shapes
225: have been completely explored in Paper I. The model has a dark
226: matter halo and the motion of stars in the equilibrium state is
227: governed by the cored logarithmic potential
228: %
229: \begin{equation}
230: V_0(R)=v_0^2 \ln \sqrt{1+R^2/R_C^2 },
231: \end{equation}
232: %
233: which is the resultant gravitational potential of luminous and
234: dark components. Throughout the paper, all length and velocity
235: variables are normalized, respectively, to the core radius of the
236: potential ($R_C$) and the asymptotic velocity of stars on circular
237: orbits ($v_0$) by setting $R_C=v_0=1$. To describe the physical
238: quantities in the configuration space, I will use the usual polar
239: coordinates $(R,\phi)$ and their Cartesian counterparts $(x,y)$
240: where $R$ is the radial distance from the galactic center and $\phi$
241: is the azimuthal angle. The specific model that I
242: adopt here is a relatively cold, near-maximal disk with no dark
243: matter concentration in the region with rising rotation curve. The
244: model parameters are set to $\left (N,\lambda,\alpha \right )=
245: (6,1,0.42)$ where $N$ is an integer exponent that controls
246: the proportion of circular orbits and the disk temperature. Larger
247: values of $N$ give rise to colder disks. For $N=6$, the parameter
248: $Q$ of Toomre (1964) is marginally larger than 1. The parameter
249: $\lambda$ is defined as the ratio $R_C/R_D$ with $R_D$ being the
250: core radius of the equilibrium density. $\alpha$ is a factor
251: that controls the total mass of the stellar component. The chosen
252: model with $\alpha=0.42$ is near-maximal within a radius
253: of $\approx 2.5 R_C$ and dominated by dark matter beyond it.
254: The calculations of Paper I revealed the eigenspectrum of this
255: model that includes a compact bar mode B1 and a sequence of
256: spiral modes S1, $\cdots$, S6. I select modes B1 and S2 as my
257: case studies of sections \ref{sec:mode-decomposition} through \ref{sec:perturbed-stellar-dynamics}. The reason for
258: choosing mode S2 is its extensive and prominent spiral structure
259: and modest growth rate. In section \ref{sec:mode-saturation},
260: I will also display some results for modes S1, S3 and S6.
261:
262:
263: \section{DECOMPOSITION OF UNSTABLE MODES}
264: \label{sec:mode-decomposition}
265:
266: One of the advantages of the method developed in Paper I is that the
267: density function of a mode can be readily decomposed to its constituent
268: Fourier components in angle-action space. The phase shifts between
269: density components determine the magnitude and direction of the
270: torque that is exerted on each component. So we can probe the
271: transfer of angular momentum and identify the direction of its flow
272: once a global mode develops.
273:
274: The perturbed distribution and Hamiltonian functions of an unstable
275: mode can be expanded as (Paper I)
276: %
277: \begin{eqnarray}
278: f_1(\Theta,\textbf{\textit{J}},t) \! &=& \! {\rm Re} \!\!\!\!\!\!
279: \sum_{m,l=-\infty}^{\infty}\sum_{j=0}^{\infty} \! \epsilon
280: d^{ml}_j(t) \Phi^{ml}_j(\textbf{\textit{J}})e^{\imath
281: \left(m\theta_\phi+l\theta_R \right)},
282: \label{eq:expansion-f1} \\
283: {\cal H}_1(\Theta,\textbf{\textit{J}},t) \! &=& \! {\rm Re} \!\!\!\!\!\!
284: \sum_{m,l=-\infty}^{\infty}\sum_{j=0}^{\infty} \! \epsilon
285: b^{ml}_j(t) \Psi^{ml}_j(\textbf{\textit{J}})e^{\imath
286: \left(m\theta_\phi+l\theta_R \right)}, \label{eq:expansion-V1}
287: \end{eqnarray}
288: %
289: where $\Theta=\left (\theta_R,\theta_{\phi} \right )$ and
290: $\textbf{\textit{J}}=\left (J_R,J_{\phi} \right )$ are the angle
291: and action variables, respectively. The angles are defined based
292: on the radial and azimuthal frequencies
293: %
294: \begin{equation}
295: {\bf \Omega}=\left (\Omega_R,\Omega_{\phi} \right )=
296: \left ( \frac{\partial {\cal H}_0}{\partial J_R},
297: \frac{\partial {\cal H}_0}{\partial J_{\phi}}
298: \right ),
299: \end{equation}
300: %
301: of stars on rosette orbits so that
302: %
303: \begin{equation}
304: \dot \theta_R = \Omega_R(\textbf{\textit{J}}),~~
305: \dot \theta_{\phi} = \Omega_{\phi}(\textbf{\textit{J}}).
306: \end{equation}
307: %
308: A dot stands for the time derivative and
309: ${\cal H}_0(\textbf{\textit{J}})$ is the integrable Hamiltonian
310: of the axisymmetric equilibrium state. For a normal mode the
311: amplitude functions $d^{ml}_j(t)$ and $b^{ml}_j(t)$ depend on
312: the time variable $t$ through the simple exponential law
313: $\exp(-{\imath}\omega t)$ with $\omega=m\Omega_p+{\imath}s$ and
314: $\imath=\sqrt{-1}$. Here $\Omega_p$ and $s$ are the pattern speed
315: and growth rate of an unstable mode of angular wavenumber $m$.
316: On the other hand, the perturbed potential function $V_1$ and
317: its associated surface density $\Sigma_1$ can be expanded in
318: the configuration space as
319: %
320: \begin{eqnarray}
321: V_1(R,\phi,t) &=& {\rm Re} \!\!\!\!\!
322: \sum_{m=-\infty}^{\infty}\sum_{j=0}^{\infty} \epsilon
323: a^{m}_j(t) \psi^{\vert m\vert}_j(R) e^{\imath m\phi},
324: \label{eq:expansion-V1-configuration} \\
325: \Sigma_1(R,\phi,t) &=& {\rm Re} \!\!\!\!\!
326: \sum_{m=-\infty}^{\infty}\sum_{j=0}^{\infty} \epsilon
327: a^{m}_j(t) \sigma^{\vert m\vert}_j(R) e^{\imath m\phi},
328: \label{eq:expansion-Sigma1-configuration}
329: \end{eqnarray}
330: %
331: where $\sigma^{\vert m\vert}_j(R)$ and $\psi^{\vert m\vert}_j(R)$
332: are surface density and potential basis functions, respectively.
333: They satisfy Poisson's integral and the bi-orthogonality
334: condition
335: %
336: \begin{equation}
337: D_j(m) \delta_{j,j'}\delta_{m,m'}=
338: 2\pi \int\limits_{0}^{\infty} \psi^{\vert m\vert}_j(R)
339: \sigma^{\vert m'\vert}_{j'}(R) RdR.
340: \end{equation}
341: %
342: Here $\delta_{m,m'}$ is the Kronecker delta and $D_j(m)$
343: are some constants that depend on our special choice of
344: basis functions. Noting $V_1\equiv {\cal H}_1$, one can
345: equate (\ref{eq:expansion-V1}) and
346: (\ref{eq:expansion-V1-configuration}), multiply the identity
347: by $\exp[-\imath(l\theta_R+m\theta_{\phi})]$ and integrate
348: over the angle variables to obtain
349: %
350: \begin{equation}
351: \sum_{j=0}^{\infty} b^{ml}_j(t)
352: \Psi^{ml}_j(\textbf{\textit{J}})=
353: \sum_{j=0}^{\infty} a^{m}_j(t)
354: \tilde \Psi^{ml}_j(\textbf{\textit{J}}),
355: \end{equation}
356: %
357: \begin{equation}
358: \tilde \Psi^{ml}_j(\textbf{\textit{J}})\!=\!
359: \frac{1}{\pi}\int\limits_{0}^{\pi} \! \psi^{\vert m\vert}_j(R)\cos
360: \left [ l\theta_R+m\left (\theta_{\phi}-\phi \right ) \right ]
361: d\theta_R.
362: \end{equation}
363: %
364: Setting $\Psi^{ml}_j(\textbf{\textit{J}})=
365: \tilde \Psi^{ml}_j(\textbf{\textit{J}})$
366: gives $b^{ml}_j(t)=a^m_j(t)$, which is a remarkable
367: simplification. A relation between $a^m_j(t)$ and
368: $d^{ml}_j(t)$ then follows from the weighted residual
369: form of the fundamental equation
370: %
371: \begin{equation}
372: f_1(\Theta,\textbf{\textit{J}},t)d\textbf{\textit{J}}
373: d\Theta=\Sigma_1(R,\phi,t)RdRd\phi.
374: \end{equation}
375: %
376: I obtain
377: %
378: \begin{eqnarray}
379: a^m_j(t)&=& \frac
380: {4\pi^2}{D_j(m)}\sum_{l=-\infty}^{\infty}\sum_{k=0}^{\infty}
381: \Lambda^{ml}_{jk}d^{ml}_k(t),\label{eq:a-versus-d} \\
382: %
383: \Lambda^{ml}_{jk} &=& \int\limits \Psi^{ml}_j(\textbf{\textit{J}})
384: \Phi^{ml}_k(\textbf{\textit{J}}) d \textbf{\textit{J}},
385: \label{eq:define-Lambda}
386: \end{eqnarray}
387: %
388: where
389: %
390: \begin{equation}
391: \Phi^{ml}_k(\textbf{\textit{J}})=
392: \frac{l\left (\partial f_0/\partial J_R\right )
393: +m\left (\partial f_0/\partial J_{\phi} \right )}
394: {l\Omega_R+m\Omega_{\phi}}
395: \Psi^{ml}_k(\textbf{\textit{J}}).
396: \label{eq:define-Phi-vs-Psi}
397: \end{equation}
398: %
399:
400: Let ${\cal L}$ be the total angular momentum of the disk, which must
401: be conserved in an isolated galaxy. This means that the torque
402: $d{\cal L}/dt$ must vanish. Defining
403: %
404: \begin{eqnarray}
405: a^m_j(t) &=& e^{-{\imath}\omega t} \tilde a^m_j=e^{-{\imath}\omega t}
406: \left ( u^m_j+{\imath}v^m_j \right ), \\
407: d^{ml}_j(t) &=& e^{-{\imath}\omega t} \tilde d^{ml}_j =
408: e^{-\imath \omega t} \left ( U^{ml}_j+{\imath}V^{ml}_j \right ),
409: \end{eqnarray}
410: %
411: one obtains
412: %
413: \begin{eqnarray}
414: {d{\cal L}\over dt} &=& 2m \pi^2 \epsilon ^2
415: e^{2st}\sum_{l=-\infty}^{\infty}
416: L_m(l)=0, \label{eq:total-torque}\\
417: L_m(l) &=& \sum_{j,k=0}^{\infty} \Lambda^{ml}_{jk}\left (v^m_j
418: U^{ml}_k -u^m_j V^{ml}_k \right ). \label{eq:torque-component-1}
419: \end{eqnarray}
420: %
421: The constant vectors $\left (U^{ml}_k,V^{ml}_k \right )$ and $\left
422: (u^m_j,v^m_j \right )$ are obtained from the linear eigenvalue
423: equations of Paper I and a subsequent use of equation
424: (\ref{eq:a-versus-d}). Equation (\ref{eq:total-torque}) is analogous
425: to equation (B6) in JH and $L_m(l)$ shows the rate of angular
426: momentum flow into/from the $l$th Fourier component (in the
427: radial angle $\theta_R$) whose corresponding density in the
428: configuration space is
429: %
430: \begin{equation}
431: \Sigma^{l}_1=\! \epsilon e^{st} {\rm Re} \!\!
432: \sum_{j,k=0}^{\infty} \! \frac{4\pi^2}{D_j(m)}
433: \Lambda^{ml}_{jk}\sigma^{|m|}_j(R) \tilde d^{ml}_k
434: e^{{\imath}m\left ( \phi-\Omega_p t \right )}.
435: \end{equation}
436: %
437: The angular momentum conservation of the disk implies
438: that some $L_m(l)$ take negative and some other positive values with
439: a total vanishing sum. This means that $\Sigma^{l}_1$, the pattern
440: corresponding to $L_m(l)$, is subject to a positive torque from
441: other components if $L_m(l)>0$, and a negative torque otherwise.
442:
443: %
444: \begin{figure*}
445: \centerline{\hbox{\includegraphics[width=0.9\textwidth]
446: {f1.eps} }}
447: \caption{The angular momentum content of different
448: Fourier components of modes B1 and S2. The sum
449: $\sum_{l=-\infty}^{+\infty}L_m(l)$ vanishes because
450: the total angular momentum of the disk is conserved.
451: The function $L_m(l)$ is normalized so that the sum of
452: positive components is unity.
453: \label{pic:ang-momentum-modes-B1-S2} }
454: \end{figure*}
455: %
456: %
457: \begin{figure*}
458: \centerline{\hbox{\includegraphics[width=0.9\textwidth]
459: {f2.eps} }} \caption{Modes B1 and S2 and their
460: components that release and absorb angular momentum. Left panels
461: show the mode shapes, which rotate counter-clockwise. The pattern
462: speeds and growth rates of modes B1 and S2 are
463: $(\Omega_p,s)=(0.918,1.160)$ and $(\Omega_p,s)=(0.454,0.216)$,
464: respectively. Middle and right panels demonstrate the
465: $\Sigma^{l-}_1$ and $\Sigma^{l+}_1$ componenets, respectively. Solid
466: lines in top panels highlight the orientations of the bar mode and
467: its components. The prominent phase shift between $\Sigma^{l-}_1$
468: and $\Sigma^{l+}_1$ is responsible for the gravitational torque
469: between them. The isocontours show the positive part of the density
470: from 10\% to 90\% of the maximum with increments of 10\%.
471: \label{pic:components-modes-B1-S2}}
472: \end{figure*}
473: %
474:
475: Figure \ref{pic:ang-momentum-modes-B1-S2} shows the variation
476: of $L_2(l)$ versus $l$ for modes B1 and S2. The results
477: are in accordance with the bar charts of JH: components with $l<0$
478: lose angular momentum and those with $l\ge 0$ gain it.
479: As JH had already pointed out, a few components ensure the
480: convergence of Fourier series in the $\theta_R$-direction.
481: The superposition of the $l\ge 0$ components of the perturbed
482: density, defined as
483: %
484: \begin{equation}
485: \Sigma^{l+}_1(R,\phi,t)=\sum_{l=0}^{\infty}\Sigma^{l}_1(R,\phi,t),
486: \end{equation}
487: %
488: will thus be a pattern that experiences a positive torque exerted by
489: %
490: \begin{equation}
491: \Sigma^{l-}_1(R,\phi,t)=\sum_{l=-\infty}^{-1}\Sigma^{l}_1(R,\phi,t).
492: \end{equation}
493: %
494: In response, the reaction torque of $\Sigma^{l+}_1$ will drain the
495: angular momentum of $\Sigma^{l-}_1$ so that ${\cal L}$ is conserved.
496: Figure \ref{pic:components-modes-B1-S2} shows modes B1 and S2 and
497: their components that emit and absorb angular momentum. It is seen
498: that in both modes, the phase of $\Sigma^{l+}_1$ lags that of
499: $\Sigma^{l-}_1$ by a magnitude of $\approx 90^{\circ}$ and the
500: positive parts of $\Sigma^{l-}_1$ fill the regions of negative
501: $\Sigma^{l+}_1$. Due to this phase shift, the angular momentum is
502: transferred between $l<0$ and $l\ge 0$ components and a
503: counterclockwise torque is exerted on $\Sigma^{l+}_1$. This
504: phenomenon is more obvious in mode B1 that has a definite spatial
505: orientation. The most important question relevant to the origin of
506: instabilities arises now: Are the density components
507: $\Sigma^{l}_1(R,\phi,t)$ generated by stellar orbits at
508: resonances? Analytical calculations of sections
509: \ref{sec:capture-into-resonances} and
510: \ref{sec:perturbed-stellar-dynamics} provide the answer.
511:
512:
513: \section{SYNCHRONOUS PRECESSION OF THE ORBITAL AXES}
514: \label{sec:capture-into-resonances}
515:
516: The perturbed density $\Sigma_1$ of a mode with the pattern
517: speed $\Omega_p$ must be supported by the slow motion of stars
518: in a rotating coordinate frame of angular velocity $\Omega_p$.
519: In other words, by sitting on a moving coordinate system one
520: must be able to identify a group of stars that are losing
521: their angular velocity, and another group of the same kind
522: but with different initial conditions, gaining it. A density
523: peak is expected in the region that the angular velocity of
524: the slow ensemble is minimum. Since the angular velocity of
525: a star is minimum at its orbital apocenter, the precession
526: rate of the orbital axes of participating stars in the
527: pattern formation should be in resonance with the pattern
528: speed. The above scenario is legitimate as long as pattern
529: stars stay far from Lindblad resonances that can generate
530: higher density regions due to the radial slowing of stars.
531: My computations show that all unstable modes fulfill this
532: requirement and azimuthal slowing is the main origin of
533: density perturbations (see \S\ref{sec:perturbed-stellar-dynamics}).
534:
535: The evolution of $\phi(t)$ is crucial for understanding the slow
536: dynamics in the configuration space, but we usually have the perturbed
537: distribution and potential functions in the angle-action space. It is
538: thus useful to find how $\phi$ depends on $(\Theta,\textbf{\textit{J}})$.
539: The simplest relation can be obtained by
540: expanding $\exp[\imath m (\phi-\theta_{\phi})]$ in Fourier series
541: of $\theta_R$ as
542: %
543: \begin{equation}
544: e^{ {\imath} m \left (\phi-\theta_{\phi}\right )} =
545: \sum_{l=-\infty}^{+\infty}
546: \xi_{ml}\left ( \textbf{\textit{J}} \right )e^{{\imath}l
547: \theta_R}, \label{eq:relation-between-phi-angles}
548: \end{equation}
549: %
550: where
551: %
552: \begin{equation}
553: \xi_{ml}(\textbf{\textit{J}})=
554: \frac 1{2\pi}\oint \cos \left [ l\theta_R+
555: m\left (\theta_{\phi}-\phi \right )\right ] d\theta_R.
556: \label{eq:fourier-coeff-phi-vs-angles}
557: \end{equation}
558: %
559: The functions $\xi_{ml}(\textbf{\textit{J}})$ are computed by
560: integrating along rosette orbits (over a period of radial
561: oscillation), and they vanish for circular orbits when $l\not=0$
562: and for radial orbits when $m\not =-2l$.
563: Multiplying equation (\ref{eq:relation-between-phi-angles})
564: by $\exp[-\imath m(\phi-\theta_{\phi})]$ and integrating the
565: identity over a period of $\theta_R$ yields the
566: following useful relation
567: %
568: \begin{equation}
569: \sum_{l=-\infty}^{+\infty}
570: \left [\xi_{ml}(\textbf{\textit{J}})\right ]^2=1.
571: \label{eq:square-fourier-identity}
572: \end{equation}
573: %
574: Moreover, taking the partial derivative of
575: (\ref{eq:relation-between-phi-angles}) with respect
576: to $\theta_R$, multiplying both sides of the resulting
577: equation by $\exp[-\imath m(\phi-\theta_{\phi})]$ followed
578: by an integration over $\theta_R$, leads to (Scott Tremaine,
579: private communication)
580: %
581: \begin{equation}
582: \sum_{l=-\infty}^{+\infty}
583: l \left [\xi_{ml}(\textbf{\textit{J}})\right ]^2=0.
584: \label{eq:l-multiply-square-fourier-identity}
585: \end{equation}
586: %
587:
588: I now multiply (\ref{eq:relation-between-phi-angles}) by
589: $\exp(\imath m \theta_{\phi})$, differentiate both sides
590: of the resulting equation with respect to $t$ and obtain
591: %
592: \begin{eqnarray}
593: m\dot \phi &=& -{\rm Re} \!\! \sum_{l=-\infty}^{+\infty} \!\!
594: \imath \left ( \frac {\partial \xi_{ml} }{\partial J_R}
595: \dot J_R+
596: \frac {\partial \xi_{ml} }{\partial J_{\phi} }
597: \dot J_{\phi}
598: \right ) e^{\imath \left (l\theta_R+m\theta_{\phi}-m\phi \right )}
599: \nonumber \\
600: &+&
601: {\rm Re} \sum_{l=-\infty}^{+\infty} \!\! \xi_{ml}
602: \left ( l\dot \theta_R +m\dot \theta_{\phi} \right )
603: e^{\imath \left (l\theta_R+m\theta_{\phi}-m\phi \right )}.
604: \label{eq:diff-relation-phi-angles}
605: \end{eqnarray}
606: %
607: The perturbed motions of stars are governed by Hamilton's
608: equations
609: %
610: \begin{equation}
611: \dot \Theta = {\bf \Omega}(\textbf{\textit{J}}) +
612: \frac{\partial {\cal H}_1}{\partial \textbf{\textit{J}}},~~
613: \dot \textbf{\textit{J}} =
614: - \frac{\partial {\cal H}_1}{\partial \Theta},
615: \end{equation}
616: %
617: that can be used to rewrite (\ref{eq:diff-relation-phi-angles})
618: in the form
619: %
620: \begin{eqnarray}
621: m\dot \phi &=& {\rm Re}
622: \sum_{l=-\infty}^{+\infty} \!\!
623: \xi_{ml} \left [
624: l\Omega_R+m\Omega_{\phi} \right ]
625: e^{\imath \left (
626: l\theta_R+m\theta_{\phi}-m\phi
627: \right ) } \nonumber \\
628: &-& {\rm Re} \sum_{l=-\infty}^{+\infty}
629: \imath e^{-\imath m\phi}
630: \left [ e^{{\imath}(l\theta_R+m\theta_{\phi} )}
631: \xi_{ml},{\cal H}_1 \right ],
632: \label{eq:relation-diff-phi-angles}
633: \end{eqnarray}
634: %
635: where $[\cdots,\cdots]$ denotes a Poisson bracket taken over the
636: angle-action space. On substituting for the complex conjugate of
637: $\exp(\imath m\phi)$ from (\ref{eq:relation-between-phi-angles}) in
638: (\ref{eq:relation-diff-phi-angles}) and noting that the second term
639: on the right hand side of (\ref{eq:relation-diff-phi-angles}) is of
640: ${\cal O}\left (\epsilon e^{st} \right )$, one finds
641: %
642: \begin{equation}
643: m\dot \phi = \! {\rm Re} \!\!\!
644: \sum_{l,k=-\infty}^{+\infty} \!\!
645: \!\! \xi_{ml} \xi_{mk} \left [
646: l\Omega_R+m\Omega_{\phi} \right ]
647: e^{\imath (l-k)\theta_R }+{\cal O}
648: \left (\epsilon e^{st}\right ),
649: \label{eq:relation-diff-phi-theta-r}
650: \end{equation}
651: %
652: which shows the functional dependence of $\dot \phi$ on the
653: actions and the radial angle.
654:
655: Equation (\ref{eq:relation-diff-phi-theta-r}) can be
656: written as
657: %
658: \begin{equation}
659: m\dot \phi = m\Omega_{\phi}(\textbf{\textit{J}}) +
660: \eta(\textbf{\textit{J}},\theta_R)+
661: {\cal O}\left ( \epsilon e^{st} \right ),
662: \label{eq:simple-relation-for-dot-phi}
663: \end{equation}
664: %
665: where the constant part $m\Omega_{\phi}$ is obtained by
666: setting $l=k$ in the double summation of
667: (\ref{eq:relation-diff-phi-theta-r}) and subsequent
668: application of relations (\ref{eq:square-fourier-identity})
669: and (\ref{eq:l-multiply-square-fourier-identity}).
670: $\eta(\textbf{\textit{J}},\theta_R)$ is a periodic
671: function of $\theta_R$ that stands for all other terms
672: of (\ref{eq:relation-diff-phi-theta-r}) with $l\not =k$.
673: $\Omega_{\phi}(\textbf{\textit{J}})$ is thus the precession
674: rate of the orbital axis of stars whose energy and angular
675: momentum correspond to the action vector $\textbf{\textit{J}}$.
676: Those stars can contribute to a developing density peak when
677: they linger at the apocenter of their orbits.
678: Such a mass deposition will continue in a rotating frame
679: of angular velocity $\Omega_p$ and over the time scale
680: $1/{\cal O}\left (\epsilon e^{st} \right )$ if the condition
681: for synchronous
682: precession
683: %
684: \begin{equation}
685: \left \langle \dot\phi \right \rangle _{\theta_R}
686: \!\!\! -\Omega_p \approx {\cal O} \left (\epsilon e^{st} \right ),
687: \label{eq:precession-frame-resoannce}
688: \end{equation}
689: %
690: holds. By defining
691: %
692: \begin{equation}
693: \mu^l_m(\textbf{\textit{J}})=
694: l\Omega_R(\textbf{\textit{J}})+
695: m\Omega_{\phi}(\textbf{\textit{J}})-m\Omega_p,
696: \label{eq:define-mu}
697: \end{equation}
698: %
699: and using (\ref{eq:simple-relation-for-dot-phi}), the
700: condition (\ref{eq:precession-frame-resoannce})
701: takes the convenient form
702: %
703: \begin{equation}
704: \mu^0_m(\textbf{\textit{J}})
705: \approx {\cal O} \left ( \epsilon e^{st} \right ),
706: \label{eq:condition-secular-resonance}
707: \end{equation}
708: %
709: which is the definition of the CR. The ILR and OLR occur
710: if $\vert \mu^l_m\vert$ diminishes for $l=-1$ and $l=1$,
711: respectively.
712:
713: A graphical representation of (\ref{eq:condition-secular-resonance})
714: will help to sharpen our understanding of the ensemble of stars that
715: can support a rotating pattern. Since there is a one-to-one, onto
716: and invertible map $\textbf{\textit{J}} \rightarrow {\bf \Omega}$
717: for initially axisymmetric disks (excluding Keplerian and harmonic
718: oscillator potentials), any functional form of the actions can be
719: described in terms of ${\bf \Omega}$ as well. For instance, one may
720: write $\mu^0_{m}(\textbf{\textit{J}})\equiv \mu^0_{m}({\bf \Omega})$.
721: I therefore identify resonant regions in the frequency space.
722: For demonstrating the distribution of dependent quantities,
723: I will use the pair $\left (\Omega_R,\Omega_i \right )$ as the
724: coordinates where $\Omega_i=\Omega_{\phi}-\frac 12 \Omega_R$.
725: The reason for this choice is that the orbital frequencies of
726: soft-centered stellar disks fill a very narrow region in the
727: ${\bf \Omega}$-space (Hunter 2002, hereafter H02; Figure 1 in JH),
728: which does not provide enough resolution for the visual identification
729: of some fine structures.
730: Denoting $\Omega_0=\left [\Omega_{\phi}\right ]_{\rm max}$
731: in a cored stellar disk, the center of the disk and infinity
732: correspond to $(\Omega_{R},\Omega_i)=(2\Omega_0,0)$ and
733: $(\Omega_{R},\Omega_i)=(0,0)$, respectively. Figure \ref{pic:CR-zones}
734: shows the frequency space of the cored logarithmic potential.
735: The lower boundary ($\Omega_R$-axis) corresponds to radial orbits
736: with $J_{\phi}=0$ and the upper curved boundary
737: $\Omega_i=\Gamma_c(\Omega_R)$ is determined by the orbital
738: frequencies of circular orbits with $J_R=0$.
739:
740: I have plotted in Figure \ref{pic:CR-zones} the countours of $\vert
741: \mu^0_{m}({\bf \Omega})\vert$ for $m=2$ and $\Omega_p=0.454\Omega_0$,
742: which is the pattern speed of mode S2. Lighter regions mark
743: the stars that are closer to the exact CR. The intersection of
744: the straight line $\mu^0_2({\bf \Omega})=0$ with the frequency
745: space corresponds to the stars at the CR. The centerline of the
746: CR zone begins at the location of a star moving on the corotation
747: circle, and extends into inner regions.
748:
749: %
750: \begin{figure}
751: \centerline{\hbox{\includegraphics[width=0.45\textwidth]
752: {f3.eps} }} \caption{The frequency space of the cored
753: logarithmic potential with $R_C=v_0=\Omega_0=1$. The upper boundary
754: $\Omega_i=\Gamma_c(\Omega_R)$ corresponds to circular orbits with
755: $J_R=0$. Contours show the variation of the CR indicator $\vert
756: \mu^0_2({\bf \Omega})\vert$ over the frequency space for
757: $\Omega_p=0.454$. Lighter regions correspond to smaller values of
758: $\vert \mu^0_2({\bf \Omega})\vert$. The function $\mu^0_2({\bf
759: \Omega})=2(\Omega_{\phi}-\Omega_p)$ vanishes at the intersection of
760: the drawn straight line and the frequency space. The star at the
761: intersection of the line $\mu^0_2({\bf \Omega})=0$ and the upper
762: boundary moves on the corotation circle. \label{pic:CR-zones}}
763: \end{figure}
764: %
765:
766: By decreasing $\Omega_p$, the CR zone is pushed to the outskirts
767: of the disk where orbital frequencies are small and the surface density
768: has dropped substantially. Furthermore, the line $\mu^{0}_m({\bf \Omega})=0$
769: will not intersect the frequency space, and there will not be a CR if
770: $\Omega_p>\Omega_0$. For a similar reason, the ILR will be absent
771: if
772: %
773: \begin{eqnarray}
774: &{}& \Omega_p > \Omega_{\rm ILR}(m), \label{eq:condition-for-NO-ILR} \\
775: &{}& \Omega_{\rm ILR}(m) = \left [ \Gamma_c(\Omega_R) +\left (\frac m2 -1 \right )
776: \Omega_R \right ]_{\rm max}.
777: \label{eq:define-Omega-ILR}
778: \end{eqnarray}
779: %
780: Inequality (\ref{eq:condition-for-NO-ILR}) is satisfied
781: when the line $\mu^{-1}_m({\bf \Omega})=0$ does not cross
782: the boundary curve $\Omega_i=\Gamma_c(\Omega_R)$.
783: These properties appear to correlate with the computed
784: eigenfrequencies of unstable modes: according to the
785: eigenspectra of Paper I, the pattern speeds of growing
786: modes lie in the interval
787: %
788: \begin{equation}
789: \Omega_{\rm ILR}(m) < \Omega_p < \Omega_0.
790: \label{eq:constraint-pattern-speed-unstable}
791: \end{equation}
792: %
793: The lower limit may be violated by cold disks and disks with
794: inner cutouts (see \S\ref{sec:role-of-ILR}). The upper limit is
795: violated by some rapidly rotating modes in galaxies with dark
796: matter, and by mode B1 of cutout disks. After exploring the orbital
797: structure of perturbed disks in \S\ref{sec:perturbed-stellar-dynamics},
798: I will explain the constraint (\ref{eq:constraint-pattern-speed-unstable})
799: and the exceptional cases that may not satisfy it.
800:
801: %
802: \begin{figure*}
803: \centerline{ \hbox{\includegraphics[width=0.33\textwidth]{f4a.eps}}
804: \hbox{\includegraphics[width=0.33\textwidth]{f4b.eps}}
805: \hbox{\includegraphics[width=0.33\textwidth]{f4c.eps}} }
806: \caption{The instantaneous isocontours of the averaged Hamiltonian
807: $\overline {\cal K}(2,0)$ for mode S2. The factor $\epsilon e^{st}$
808: is dealt with as a constant and has been set to 0.01 ({\em left panel}),
809: 0.05 ({\em middle panel}) and $0.1$ ({\em right panel}).
810: The maximum of $\overline {\cal K}_1$ has been normalized
811: to unity. The crescent-like resonant zone, driven by the
812: instantaneous elliptic point $c_0$, has been shaded. Darker
813: regions correspond to smaller values of $\overline {\cal K}(2,0)$.
814: The separatrices intersect at the instantaneous hyperbolic
815: point $h_0$.
816: \label{pic:resonance-cavity}}
817: \end{figure*}
818: %
819:
820:
821: \section{STELLAR DYNAMICS IN THE PERTURBED DISK}
822: \label{sec:perturbed-stellar-dynamics}
823:
824: The orbital dynamics of stars can be understood
825: through investigating the induced dynamics by each Fourier
826: component. We usually split the phase space to
827: several subspaces by using a Fourier expansion in $\theta_R$,
828: and each subspace is associated with a pattern component
829: (e.g., Figure \ref{pic:components-modes-B1-S2}). In the
830: linear regime, the perturbed motion of stars due to the
831: $l$th component can be traced by carrying out a canonical
832: transformation
833: %
834: \begin{equation}
835: \left (\theta_R,\theta_{\phi},J_R,J_{\phi} \right )
836: \rightarrow \left ( w_1,w_2,I_1,I_2 \right ),
837: \end{equation}
838: defined by the generating function \citep{L-B93}
839: %
840: \begin{equation}
841: {\cal S}=\left (l \theta_R+
842: m\theta_{\phi}- m\Omega_p t \right )I_1+\theta_R I_2.
843: \end{equation}
844: %
845: This yields the transformation rules
846: %
847: \begin{equation}
848: \textbf{\textit{w}} = \frac{\partial {\cal S}}
849: {\partial \textbf{\textit{I}}},~~
850: \textbf{\textit{J}} =
851: \frac{\partial {\cal S}}{\partial \Theta},
852: \end{equation}
853: %
854: and the new Hamiltonian
855: %
856: \begin{equation}
857: {\cal K} = {\cal H}_0(\textbf{\textit{I}})+
858: {\cal H}_1(\textbf{\textit{w}},\textbf{\textit{I}},t)
859: +\partial {\cal S}/\partial t,
860: \end{equation}
861: %
862: so that
863: %
864: \begin{eqnarray}
865: {\cal K} &=& {\cal H}_0(\textbf{\textit{I}})-m \Omega_p I_1
866: \nonumber \\
867: &+& \epsilon e^{st} ~{\rm Re} \! \sum_{k=-\infty}^{\infty} \sum_{j=0}^{\infty}
868: \tilde a^{m}_j \Psi^{mk}_j(\textbf{\textit{I}}) e^{{\imath} \left [
869: w_1 +\left (l-k \right )w_2 \right ]}.
870: \label{eq:new-K-Hamiltonian}
871: \end{eqnarray}
872: %
873: From the equations of motion
874: %
875: \begin{equation}
876: \dot \textbf{\textit{I}}=-\frac {\partial {\cal K} }
877: {\partial \textbf{\textit{w}} },~~
878: \dot \textbf{\textit{w}}=\frac {\partial {\cal K} }
879: {\partial \textbf{\textit{I}} },
880: \end{equation}
881: %
882: one can verify that $w_2=\Omega_R t+{\cal O}(\epsilon e^{st})$
883: is an increasing function of time, which can be made a cyclic
884: coordinate by averaging ${\cal K}$ over $w_2$. Consequently,
885: the action $I_2=J_R-l I_1=J_R-l J_{\phi}/m$ becomes an
886: adiabatic invariant so that $\dot I_2={\cal O}(\epsilon^2 e^{2st})$,
887: and the dynamics is reduced to the flows governed by the averaged
888: Hamiltonian
889: %
890: \begin{eqnarray}
891: \overline {\cal K}(m,l) &\equiv& \langle {\cal K} \rangle _{w_2} =
892: \overline {\cal K}_0+\epsilon e^{st} \overline {\cal K}_1,
893: \label{eq:averaged-bar-K} \\
894: \overline {\cal K}_0(\textbf{\textit{I}}) &=&
895: {\cal H}_0(\textbf{\textit{I}})-m\Omega_p I_1 \\
896: \overline {\cal K}_1 &=& A^{l}_{m}(\textbf{\textit{I}})
897: \cos \left [ w_1+\vartheta^{l}_{m}(\textbf{\textit{I}}) \right ],
898: \label{eq:averaged-bar-K-w}
899: \end{eqnarray}
900: %
901: where I have kept the real (physical) part of
902: $\langle {\cal K} \rangle _{w_2}$ and
903: %
904: \begin{eqnarray}
905: A^{l}_{m}(\textbf{\textit{I}}) &=&
906: \sqrt{X^2+Y^2},~~
907: \vartheta^{l}_{m}(\textbf{\textit{I}}) =
908: \arctan \left ({Y \over X} \right ), \\
909: X &=& \sum_{j=0}^{\infty} u^{m}_j
910: \Psi^{ml}_j(\textbf{\textit{I}}),~~
911: Y = \sum_{j=0}^{\infty} v^{m}_j
912: \Psi^{ml}_j(\textbf{\textit{I}}).
913: \end{eqnarray}
914: %
915: The action $I_2$ remains a constant parameter over a
916: time scale of $1/{\cal O}(\epsilon e^{st})$.
917:
918: The angle $w_1$ can become a rotating or librating angle
919: depending on the initial value of $\mu^{l}_m(\textbf{\textit{I}})$.
920: Stars will be near a mean-motion resonance if $w_1$ librates.
921: For simplicity, I introduce the slow angle
922: $\theta=w_1+\vartheta^{l}_{m}(\textbf{\textit{I}})-\pi$, which
923: casts $\overline {\cal K}(m,l)$ into the form
924: %
925: \begin{equation}
926: \overline {\cal K}(m,l) =
927: {\cal H}_0(\textbf{\textit{I}})-m\Omega_p I_1 -
928: \epsilon e^{st} A^{l}_{m}(\textbf{\textit{I}})
929: \cos \theta. \label{eq:averaged-bar-K-theta}
930: \end{equation}
931: %
932: The distribution function (DF) $f=f_0+f_1$ is thus conserved
933: along the trajectories determined by
934: %
935: \begin{eqnarray}
936: \dot \theta \! &=& \frac{\partial \overline {\cal K}(m,l)}{\partial I_1} =
937: \! \mu^{l}_m(\textbf{\textit{I}}) \! -\! \epsilon
938: e^{st} \frac {\partial A^{l}_{m}(\textbf{\textit{I}})}{\partial I_1}
939: \cos \theta,
940: \label{eq:averaged-equations-vs-theta-1} \\
941: \dot I_1 \! &=& -\frac{\partial \overline {\cal K}(m,l)}{\partial \theta} =
942: \! -\epsilon e^{st} A^{l}_{m}(\textbf{\textit{I}})
943: \sin \theta. \label{eq:averaged-equations-vs-theta-2}
944: \end{eqnarray}
945: %
946: These are equations of perturbed orbits bound to the $l$th
947: Fourier component in the linear regime. According to the
948: dynamical mechanism presented in \S\ref{sec:capture-into-resonances},
949: the $l=0$ Fourier component is capable of keeping
950: $\mu^{l}_m(\textbf{\textit{I}})\approx {\cal O}\left (
951: \epsilon e^{st} \right )$, which in turn, can lead to the
952: resonant capture of stars in the linear regime once $\theta$
953: begins to evolve slowly.
954:
955: I have frozen the (small) exponential factor $\epsilon e^{st}$
956: and plotted the instantaneous isocontours (IICs) of
957: $\overline {\cal K}(2,0)$ in Figure \ref{pic:resonance-cavity}
958: for the initially circular orbits ($I_2=0$) of mode S2. The
959: isocontours have been displayed in the coordinate plane of
960: $u_1=\sqrt{2I_1}\sin \theta$ and $u_2=\sqrt{2I_1}\cos \theta$
961: where the transformation $(\theta,I_1) \rightarrow (u_1,u_2)$
962: is canonical. Note that the angle $\theta$ is measured clockwise
963: with respect to the positive $u_2$-direction.
964:
965: The IICs of Figure \ref{pic:resonance-cavity} resemble the
966: topology of the celestial three-body problem (e.g., Wisdom 1980;
967: Winter \& Murray 1997). Two homoclinic loops that intersect at
968: the instantaneous hyperbolic point $h_0$ surround a crescent-like,
969: resonant region where the angle $\theta$ librates. The instantaneous
970: elliptic point $c_0$ corresponds to stable orbits at exact resonance.
971: For a specified $I_2$, the coordinates of $c_0$ and $h_0$ are
972: $(\theta,I_1)=[\pi,r_{\pi}(t)]$ and $(\theta,I_1)=[0,r_0(t)]$,
973: respectively, where $r_{\pi}(t)$ and $r_0(t)$ are the real roots of
974: %
975: \begin{equation}
976: \mu^{0}_m(r_z,I_2) -
977: \epsilon e^{st} \frac {\partial A^{0}_{m}(r_z,I_2)}{\partial r_z}
978: \cos z =0,~~z =0,\pi, \label{eq:stationary-points}
979: \end{equation}
980: %
981: at a given time $t$.
982:
983: To this end, I show that the elliptic point $c_0$ always lies on
984: the negative $u_2$-axis with $\theta=\pi$. I assume the small
985: variations $\tilde \theta=\theta-\pi$ and $\tilde I_1=I_1-r_{\pi}(t)$
986: for $l=0$, and linearize equations (\ref{eq:averaged-equations-vs-theta-1})
987: and (\ref{eq:averaged-equations-vs-theta-2}) to obtain
988: %
989: \begin{eqnarray}
990: \frac {d\tilde \theta}{dt} &=&
991: a_{12}(t) \tilde I_1,
992: \label{eq:linearized-equations-vs-theta-1} \\
993: \frac {d\tilde I_1}{dt} &=&
994: a_{21}(t) \tilde \theta -\dot r_{\pi}(t),
995: \label{eq:linearized-equations-vs-theta-2}
996: \end{eqnarray}
997: %
998: where the time-dependent coefficients are defined as
999: %
1000: \begin{eqnarray}
1001: a_{12}(t) &=& \left [
1002: \frac {\partial \mu^{0}_m(\textbf{\textit{I}})}
1003: {\partial I_1}
1004: +\epsilon e^{st}
1005: \frac {\partial^2 A^{0}_m(\textbf{\textit{I}})}
1006: {\partial I_1^2}
1007: \right ]_{I_1=r_{\pi}(t)}, \\
1008: %
1009: a_{21}(t) &=& \epsilon e^{st} \left [
1010: A^{0}_{m}(\textbf{\textit{I}}) \right ]_{I_1=r_{\pi}(t)}.
1011: \end{eqnarray}
1012: %
1013: The quantity $\dot r_{\pi}(t)$ is determined by
1014: differentiating (\ref{eq:stationary-points}) with respect
1015: to $t$. I obtain
1016: %
1017: \begin{eqnarray}
1018: \dot r_{\pi}(t) &=& -\frac {a_{13}(t)}{a_{12}(t)}, \\
1019: a_{13}(t) &=& \epsilon s e^{st}
1020: \left [ \frac {\partial A^{0}_m(\textbf{\textit{I}})}
1021: {\partial I_1} \right ]_{I_1=r_{\pi}(t)}.
1022: \end{eqnarray}
1023: %
1024: Eliminating $\tilde I_1$ from the linearized equations
1025: leads to
1026: %
1027: \begin{eqnarray}
1028: &{}& \frac {d^2\tilde \theta}{dt^2} +
1029: a_{0}(t) \tilde \theta =a_{13}(t)+{\cal O}
1030: \left (\epsilon^2 e^{2st} \right ),
1031: \label{eq:2nd-order-tilde-theta} \\
1032: &{}& a_{0}(t)= - \left [
1033: \epsilon e^{st} A^{0}_{m}(\textbf{\textit{I}})
1034: \frac {\partial \mu^{0}_m(\textbf{\textit{I}})}
1035: {\partial I_1}
1036: \right ]_{I_1=r_{\pi}(t)}.
1037: \label{eq:spring-coefficient}
1038: \end{eqnarray}
1039: %
1040: The sign of the time-dependent {\it spring coefficient} $a_0(t)$
1041: is determined by the sign of $\partial \mu^0_m/\partial I_1$. In
1042: cored stellar disks whose equilibrium potential fields are monotonic
1043: functions of $R$, the following conditions always hold
1044: %
1045: \begin{equation}
1046: \frac {\partial \Omega_{R}(\textbf{\textit{J}})}{\partial J_{\phi}}\le 0,~~
1047: \frac {\partial \Omega_{\phi}(\textbf{\textit{J}})}{\partial J_{\phi}}\le 0.
1048: \label{eq:negative-diff-omgs-d-I1}
1049: \end{equation}
1050: %
1051: The equality sign corresponds to the galactic center. The isochrone,
1052: Kuzmin-Toomre and cored exponential disks (investigated in JH and
1053: Paper I) fulfill (\ref{eq:negative-diff-omgs-d-I1}), which implies
1054: $\partial \mu^0_m/\partial I_1<0$. Therefore, the spring coefficient
1055: $a_0(t)$ is positive and the homogeneous solution of
1056: (\ref{eq:2nd-order-tilde-theta}) is bounded. This proves that the
1057: instantaneous critical point $(\theta,I_1)=[\pi,r_{\pi}(t)]$ is of
1058: elliptic type.
1059:
1060: The resonant zone disappears for $l\not =0$ when the pattern speed
1061: lies in the interval (\ref{eq:constraint-pattern-speed-unstable}).
1062: This is because all participating stars in the pattern formation
1063: satisfy (\ref{eq:condition-secular-resonance}) that yields
1064: %
1065: \begin{equation}
1066: \mu^{l}_m(\textbf{\textit{I}}) \approx l\Omega_R+
1067: {\cal O}\left (\epsilon e^{st} \right ).
1068: \label{eq:mu-for-other-l-near-SR}
1069: \end{equation}
1070: %
1071: Accordingly, $\mu^l_m(\textbf{\textit{I}})$ cannot flip
1072: sign for $l\not =0$ and equation (\ref{eq:stationary-points})
1073: will not have any real root for small perturbations
1074: of ${\cal O}(\epsilon e^{st})$. In such a circumstance,
1075: IICs will constitute a bundle of unidirectional closed
1076: curves that encircle the origin.
1077: In \S\ref{sec:modes-in-frequency-space}, I will verify for
1078: the spiral mode S2 that pattern stars satisfy
1079: (\ref{eq:mu-for-other-l-near-SR}).
1080:
1081:
1082: \subsection{The Expansion of Resonant Zones}
1083: \label{sec:resonant-zones-expand}
1084:
1085: It is evident that the structure of IICs in
1086: Figure \ref{pic:resonance-cavity} evolves with time as
1087: the perturbations grow proportional to $\epsilon e^{st}$.
1088: For small perturbations the critical points $c_0$ and $h_0$
1089: are preserved although they are displaced. As a consequence,
1090: there always exists a resonant zone for $\epsilon e^{st}\ll 1$
1091: and it would be interesting to learn how that zone {\it expands}
1092: in a growing mode. To answer this question analytically,
1093: I measure the variation of the quantity
1094: %
1095: \begin{equation}
1096: \Delta = d_{o}-d_{i},
1097: \end{equation}
1098: %
1099: where $d_{o}$ and $d_{i}$ are, respectively,
1100: the values of $I_1$ on the outer and inner homoclinic loops at
1101: $\theta=\pi$. The implicit functional form of homoclinic
1102: loops is given by
1103: %
1104: \begin{eqnarray}
1105: &{}& \overline {\cal K}_0(I_1,I_2) -
1106: \epsilon e^{st} A^{0}_{m}(I_1,I_2) \cos \theta =\nonumber \\
1107: &{}& \qquad \overline {\cal K}_0[r_0(t),I_2] -
1108: \epsilon e^{st} A^{0}_{m}[r_0(t),I_2],
1109: \end{eqnarray}
1110: %
1111: whose temporal derivative at $\theta=\pi$ results in
1112: %
1113: \begin{eqnarray}
1114: &{}& \!\! \dot d_{\nu} \left [ \mu^0_m +
1115: \epsilon e^{st}
1116: \frac {\partial A^{0}_{m}}{\partial I_1}
1117: \right ]_{I_1=d_{\nu}} \!\!\!\!\! +
1118: \epsilon s e^{st} A^{0}_{m}(d_{\nu},I_2)
1119: = \nonumber \\
1120: &{}& \!\! \dot r_0 \left [ \mu^0_m \!-\!
1121: \epsilon e^{st}
1122: \frac {\partial A^{0}_{m}}{\partial I_1}
1123: \right ]_{I_1=r_0} \!\!\!\!\! -
1124: \epsilon s e^{st} A^{0}_{m}(r_0,I_2),
1125: \label{eq:growth-of-Io-and-Ii}
1126: \end{eqnarray}
1127: %
1128: for $\nu \equiv o,i$. The bracket on the right hand side of
1129: equation (\ref{eq:growth-of-Io-and-Ii}) vanishes because of
1130: (\ref{eq:stationary-points}) and the bracket on the left
1131: hand side is simply $d\theta_{\nu}/dt$ ($\nu$=$o,i$).
1132: From (\ref{eq:averaged-equations-vs-theta-1}) and
1133: (\ref{eq:averaged-equations-vs-theta-2}) one can
1134: verify that $d\theta_o/dt<0$ and $d\theta_i/dt>0$.
1135: Equation (\ref{eq:growth-of-Io-and-Ii}) thus becomes
1136: %
1137: \begin{equation}
1138: \dot d_{\nu} = - \frac {\epsilon s e^{st} }
1139: {\dot \theta_{\nu} }
1140: \left [ A^{0}_{m}\left (d_{\nu},I_2 \right )
1141: + A^{0}_{m} \left (r_0,I_2 \right ) \right ],~~
1142: \nu \equiv o,i. \label{eq:growth-of-Io-and-Ii-closed-form}
1143: \end{equation}
1144: %
1145: Subtracting $\dot d_i$ from $\dot d_o$ leads to
1146: %
1147: \begin{eqnarray}
1148: \dot \Delta &=& s \left ( \epsilon e^{st} \right )
1149: \Biggl [
1150: \frac {A^{0}_{m}\left (d_o,I_2 \right )}{|\dot \theta_o|} +
1151: \frac {A^{0}_{m}\left (d_i,I_2 \right )}{|\dot \theta_i|}
1152: \nonumber \\
1153: &{}& \qquad \qquad + \frac {A^{0}_{m} \left (r_0,I_2 \right )}
1154: {|\dot \theta_o|} +
1155: \frac {A^{0}_{m} \left (r_0,I_2 \right )}
1156: {|\dot \theta_i|} \Biggr ].
1157: \label{eq:growth-of-Delta}
1158: \end{eqnarray}
1159: %
1160: The angular rates $\dot \theta_o$ and $\dot \theta_i$
1161: can be estimated using the linearized equation
1162: (\ref{eq:linearized-equations-vs-theta-1}) as
1163: %
1164: \begin{equation}
1165: \dot \theta_o=a_{12}(t) \left [ d_o-r_{\pi}(t) \right ],~~
1166: \dot \theta_i=a_{12}(t) \left [ d_i-r_{\pi}(t) \right ].
1167: \label{eq:relation-for-dot-theta_o-and-theta_i}
1168: \end{equation}
1169: %
1170: Since the resonant zone is thin, one may assume
1171: $r_{\pi}(t)\approx \left ( d_o +d_i \right )/2$,
1172: which can be combined with
1173: (\ref{eq:relation-for-dot-theta_o-and-theta_i})
1174: to obtain
1175: %
1176: \begin{equation}
1177: |\dot \theta_o | \approx
1178: |\dot \theta_i | \approx
1179: |a_{12}(t) | \frac{\Delta}{2}.
1180: \label{eq:approx-relation-for-dot-theta_o}
1181: \end{equation}
1182: %
1183: Substituting from (\ref{eq:approx-relation-for-dot-theta_o})
1184: in (\ref{eq:growth-of-Delta}) yields
1185: %
1186: \begin{eqnarray}
1187: \Delta \dot \Delta & \approx &
1188: \frac{2 s \left ( \epsilon e^{st} \right )}{|a_{12}(t) |}
1189: \Bigl [
1190: 2 A^{0}_{m}\left (r_0,I_2 \right ) \nonumber \\
1191: &{}& \qquad +
1192: A^{0}_{m}\left (d_i,I_2 \right ) +
1193: A^{0}_{m}\left (d_o,I_2 \right ) \Bigr ],
1194: \label{eq:growth-of-Delta-2nd-relation}
1195: \end{eqnarray}
1196: %
1197: where the quotient of the barcketed terms and $|a_{12}(t)|$ is
1198: of ${\cal O}(1)$. Integrating (\ref{eq:growth-of-Delta-2nd-relation})
1199: results in
1200: %
1201: \begin{equation}
1202: \Delta \approx 2 \sqrt{ \epsilon e^{st} } {\cal O}(1),~~
1203: \dot \Delta \approx s \sqrt{\epsilon e^{st} } {\cal O}(1),
1204: \label{eq:growth-of-Delta-and-dotDelta}
1205: \end{equation}
1206: %
1207: from which the expansion rate of the resonance width per unit
1208: length is estimated:
1209: %
1210: \begin{equation}
1211: \dot \Delta/\Delta \sim s/2.
1212: \label{eq:growth-of-Delta-versus-s}
1213: \end{equation}
1214: %
1215: This shows that stars are steadily captured into resonance
1216: by the $l=0$ Fourier component as the mode grows in the linear
1217: regime.
1218:
1219: My calculations show that the maximum of $A^0_m(\textbf{\textit{I}})$
1220: in unstable modes does not coincide neither with $h_0$ nor with $c_0$,
1221: but it falls inside the resonant zone within a time scale
1222: $\ll 1/{\cal O}\left (\epsilon e^{st} \right )$ as $\Delta$
1223: grows from zero width. The observed shift from the location
1224: of $c_0$ depends on the gradient $\partial f_0/\partial J_{\phi}$.
1225: All growing modes obey the following rules near $c_0$:
1226: %
1227: \begin{eqnarray}
1228: &{}&
1229: \frac{\partial A^0_m(\textbf{\textit{I}})}{\partial I_1}
1230: <0<\mu^0_m(\textbf{\textit{I}})~~{\rm if}~~
1231: \frac{\partial f_0(\textbf{\textit{J}})}{\partial J_{\phi}}<0,
1232: \label{eq:negative-gradient-of-A} \\
1233: &{}&
1234: \frac{\partial A^0_m(\textbf{\textit{I}})}{\partial I_1}
1235: >0>\mu^0_m(\textbf{\textit{I}})~~{\rm if}~~
1236: \frac{\partial f_0(\textbf{\textit{J}})}{\partial J_{\phi}}>0.
1237: \label{eq:positive-gradient-of-A}
1238: \end{eqnarray}
1239: %
1240: In both cases the magnitude of $\mu^0_m(\textbf{\textit{I}})$
1241: remains small near the maximum of $A^0_m(\textbf{\textit{I}})$,
1242: confirming the theory of \S\ref{sec:capture-into-resonances}
1243: that the perturbed density profile is supported by the
1244: synchronous precession of orbital axes.
1245:
1246:
1247: %
1248: \begin{figure*}
1249: \centerline{\hbox{\includegraphics[width=0.9\textwidth]
1250: {f5.eps} }} \caption{The density components
1251: $\Sigma^{l}_1$ (top panels) and the contours of $A^l_2({\bf
1252: \Omega})$ (bottom panels) for mode S2. In all figures dashed lines
1253: mark the stars at exact CR with $\mu^0_2({\bf \Omega})=0$. The line
1254: $\mu^{-1}_2({\bf \Omega})=0$ lies outside the plot range because it
1255: is parallel to the $\Omega_R$-axis with $\Omega_i=0.454$. Straight
1256: solid line in the bottom-right panel is defined by $\mu^1_2({\bf
1257: \Omega})=0$. Its relative location with respect to the highly
1258: populated regions of the frequency space shows that stars bound to
1259: the $l=1$ component evolve far from the OLR. The contour levels of
1260: $A^l_2({\bf \Omega})$ range from 10\% to 90\% of the maximum with
1261: increments of 10\%. \label{pic:components-and-zones-B1-S2}}
1262: \end{figure*}
1263: %
1264:
1265: \subsection{Angular Momentum Transfer}
1266: \label{sec:angular-momentum-at-resonance}
1267:
1268: The location of the resonant zone in the $(u_1,u_2)$-plane
1269: provides valuable information of the behavior of resonant
1270: stars. The Hamiltonian varies according to the equation
1271: %
1272: \begin{equation}
1273: \frac {d \overline {\cal K}(m,l)}{dt}=
1274: \frac {\partial \overline {\cal K}(m,l)}{\partial t}=
1275: -\epsilon s e^{st} A^l_m(\textbf{\textit{I}}) \cos \theta.
1276: \label{eq:formula-for-bar-K-variation}
1277: \end{equation}
1278: %
1279: One can thus determine how the orbital energy of trapped stars
1280: changes over time. Recalling that the instantaneous elliptic point
1281: $c_0$ occurs at $\theta=\pi$, for most resonant orbits the
1282: minimum of $\theta$ is larger than $\pi/2$
1283: (see Figure \ref{pic:resonance-cavity}) and equation
1284: (\ref{eq:formula-for-bar-K-variation}) yields
1285: %
1286: \begin{equation}
1287: \frac {d \overline {\cal K}(m,0)}{dt}>0.
1288: \end{equation}
1289: %
1290: In other words, stars gain energy during a librational
1291: motion around $\theta=\pi$. Since $I_2=J_R$ (for $l=0$)
1292: is an adiabatic invariant in the resonance zone, an
1293: increase in the orbital energy of stars boosts their
1294: angular momentum.
1295:
1296: There is a different mechanism for the angular momentum transfer
1297: to/from stars whose $\theta$ is rotational. Such stars will definitely
1298: experience both $\dot I_1>0$ and $\dot I_1<0$ states over a complete
1299: period of $\theta$, but which one overwhelms the other and how does it
1300: affect the orbital angular momentum? Outside the resonant zone we have
1301: %
1302: \begin{equation}
1303: \left \vert \mu^l_m(\textbf{\textit{I}}) \right \vert
1304: \gg \left \vert \epsilon e^{st}
1305: \frac {\partial A^{l}_{m}(\textbf{\textit{I}})}{\partial I_1}
1306: \right \vert,
1307: \end{equation}
1308: %
1309: which guarantees the rotation of $\theta$ according
1310: to (\ref{eq:averaged-equations-vs-theta-1}). I now divide
1311: equation (\ref{eq:averaged-equations-vs-theta-2}) by
1312: (\ref{eq:averaged-equations-vs-theta-1}) and ignore
1313: all terms of ${\cal O}(\epsilon^2 e^{2st})$ to obtain
1314: %
1315: \begin{equation}
1316: \frac {dI_1}{d\theta}=-\frac {\epsilon}{\mu^l_m(\textbf{\textit{I}}_0)}
1317: \exp \left [\frac{s\theta}{\mu^l_m(\textbf{\textit{I}}_0)} \right ]
1318: A^l_m(\textbf{\textit{I}}_0)\sin \theta,
1319: \label{eq:I1-vs-theta}
1320: \end{equation}
1321: %
1322: where $\textbf{\textit{I}}_0$ is the initial value of the
1323: action vector $\textbf{\textit{I}}=(I_1,I_2)$ at $\theta=0$.
1324: For a clockwise rotation of phase space flows, $\theta$ ranges
1325: from $0$ to $2\pi$ and reversely for their counter-clockwise
1326: rotation. Moreover, the direction of rotation is determined by
1327: the sign of $\mu^l_m(\textbf{\textit{I}}_0)$.
1328: Given these points, integrating (\ref{eq:I1-vs-theta}) over
1329: a complete cycle of $\theta$ results in the incremental
1330: change of $I_1$ as
1331: %
1332: \begin{equation}
1333: \Delta I_1(l)=\epsilon A^l_m
1334: \frac {\mu^l_m
1335: {\rm sign}\left (\mu^l_m \right )}
1336: {\left ( \mu^l_m \right )^2+s^2}
1337: \left [ \exp \left ( \frac {2\pi s}{\mu^l_m} \right ) -1
1338: \right ].
1339: \label{eq:delta-I1-vs-l}
1340: \end{equation}
1341: %
1342: It is seen that $\Delta I_1$ is positive for
1343: $\mu^l_m(\textbf{\textit{I}}_0)>0$ and stars gain
1344: angular momentum as the perturbations grow. The opposite
1345: happens for $\mu^l_m(\textbf{\textit{I}}_0)<0$. Combining
1346: (\ref{eq:mu-for-other-l-near-SR}) and (\ref{eq:delta-I1-vs-l})
1347: shows that the angular momentum gain or loss is decided
1348: by the sign and magnitude of the Fourier number $l$. The Fourier
1349: components with $l<0$ and $l>0$ respectively drain and boost the
1350: angular momentum of stars so that
1351: %
1352: \begin{eqnarray}
1353: &{}& \!\! \Delta I_1(-1)<\Delta I_1(-2)<
1354: \Delta I_1(-3)<\cdots <0, \label{eq:sort-negative-l} \\
1355: &{}& \!\! \Delta I_1(+1)>\Delta I_1(+2)>
1356: \Delta I_1(+3)>\cdots >0. \label{eq:sort-positive-l}
1357: \end{eqnarray}
1358: %
1359:
1360: Not all stars bound to the $l=0$ component are in the
1361: resonant zone, specially in the limit of
1362: $\epsilon e^{st}\rightarrow 0$. Hence, the sign of
1363: $\Delta I_1(0)$ is determined both by resonant and
1364: non-resonant stars. Based on my arguments presented
1365: after equation (\ref{eq:formula-for-bar-K-variation}),
1366: resonant stars always gain angular momentum and have a
1367: positive contribution to $\Delta I_1(0)$. However, It is
1368: the sign of $\partial f_0/\partial J_{\phi}$
1369: that determines whether non-resonant stars of the $l=0$
1370: component are emitting or absorbing angular momentum.
1371: Equations (\ref{eq:negative-gradient-of-A}) and
1372: (\ref{eq:positive-gradient-of-A}) together with
1373: (\ref{eq:delta-I1-vs-l}) show that non-resonant stars
1374: of the $l=0$ component gain and lose angular momentum for
1375: $\partial f_0/\partial J_{\phi}<0$ and
1376: $\partial f_0/\partial J_{\phi}>0$, respectively.
1377: Therefore, $\Delta I_1(0)$ is certainly positive if the
1378: mode develops in a region of the phase space with
1379: $\partial f_0/\partial J_{\phi}<0$. Since most stars of
1380: the $l=0$ component are non-resonant as
1381: $\epsilon e^{st}\rightarrow 0$, the occurrence of a
1382: growing mode as a result of $\partial f_0/\partial J_{\phi}>0$
1383: will lead to $\Delta I_1(0)<0$. The results of the above
1384: analysis are consistent with the results of JH and the
1385: bar charts of Figure \ref{pic:ang-momentum-modes-B1-S2}.
1386:
1387:
1388: \subsection{Mode Components in the Frequency Space}
1389: \label{sec:modes-in-frequency-space}
1390:
1391: To understand how the mechanism of resonant trapping operates
1392: on non-circular orbits in the frequency space, I have plotted
1393: the components of mode S2 in Figure \ref{pic:components-and-zones-B1-S2}.
1394: Top panels display the pattern components in the configuration space,
1395: and in bottom panels I have shown the isocontours of $A^l_2({\bf \Omega})\equiv
1396: A^l_2(\textbf{\textit{I}})$ together with the lines $\mu^1_2({\bf
1397: \Omega})=0$ (straight solid line) and $\mu^0_2({\bf \Omega})=0$
1398: (dashed lines). Darker regions in the bottom panels correspond to
1399: larger values of $A^l_2({\bf \Omega})$. The contour plots of
1400: $A^l_2({\bf \Omega})$ show which stars in the frequency/action space
1401: are engaged with the $l$th component. As one could anticipate, the
1402: highly populated regions of all components are close to the CR zone
1403: of Figure \ref{pic:CR-zones}. The maxima of $A^l_2({\bf \Omega})$
1404: have not been located on the line $\mu^0_2({\bf \Omega})=0$
1405: because the initial density gradient of the equilibrium state
1406: displaces the center of mass of pattern stars. In modes with
1407: $\partial f_0/\partial J_{\phi}<0$ near the CR zone, the maxima
1408: are shifted to regions with $\mu^0_2({\bf \Omega})>0$
1409: (see \S\ref{sec:resonant-zones-expand}).
1410:
1411: Whilst the $l=0$ component has trapped near-circular orbits into the
1412: CR (Figure \ref{pic:components-and-zones-B1-S2}), the stars bound to
1413: other components evolve far from mean-motion resonances and the
1414: angle $\theta$ becomes rotational for them. In fact, stars bound
1415: to the $l=-1$ component avoid the line $\mu^{-1}_m({\bf \Omega})=0$
1416: and the quantity $\mu^{-1}_m({\bf \Omega})$ is always negative as long
1417: as inequality (\ref{eq:condition-for-NO-ILR}) holds. Through a similar
1418: mechanism, stars avoid the line $\mu^{1}_m({\bf \Omega})=0$ and the
1419: quantity $\mu^{1}_m({\bf \Omega})$ remains positive regardless of the
1420: magnitude of $\Omega_p$. These results show that the dynamical mechanism
1421: for growing modes of finite $s$ differers with LBK's theory established
1422: in the limit of $s\rightarrow 0$. In other words, in a growing spiral
1423: mode similar to mode S2 of this paper, the $l=-1$ and $l=1$ components
1424: are not associated with the ILR and OLR. A new dynamical origin of
1425: instabilities, which interprets the results of this paper and has not
1426: the restrictions of LBK's theory, is presented in the next section.
1427:
1428:
1429: \section{THE ORIGIN OF INSTABILITIES}
1430: \label{sec:origin-of-instabilities}
1431:
1432: Consider a randomly generated, small-amplitude, $m$-fold density wave
1433: of pattern speed $\Omega_p$ whose potential field can be generally
1434: expanded in the Fourier series of the angle variables. Independent
1435: of the form of initial phase space distribution, determined by
1436: $f_0(\textbf{\textit{J}})$, the points
1437: $\left (\textbf{\textit{I}},\theta \right )=
1438: \left (\textbf{\textit{I}}_0, 0\right )$ and
1439: $\left (\textbf{\textit{I}},\theta \right )=
1440: \left (\textbf{\textit{I}}_0, \pi \right )$ will emerge as
1441: time-invariant stationary points of the flows generated by
1442: $\overline {\cal K}(m,0)$ if the following condition holds
1443: %
1444: \begin{equation}
1445: \frac {\partial A^0_m(\textbf{\textit{I}})}{\partial I_1}
1446: =\mu^0_m(\textbf{\textit{I}})=0,
1447: \label{eq:condition-for-stationary-points}
1448: \end{equation}
1449: %
1450: at $\textbf{\textit{I}}=\textbf{\textit{I}}_0$.
1451: In such a circumstance, orbits associated with either of these
1452: stationary points will never change their action vector
1453: $\textbf{\textit{I}}=(I_1,I_2)$ and their orbital energy must be
1454: conserved. This condition will be satisfied only if $s=0$ in
1455: (\ref{eq:formula-for-bar-K-variation}). This is how a \citet{vK55}
1456: mode is born. The state of (\ref{eq:condition-for-stationary-points})
1457: can occur at any point in the infinite dimensional action space
1458: and for arbitrary values of $\Omega_p$. Consequently, stationary
1459: modes constitute a continuous family and Mathur's (1990) isolated,
1460: pure oscillatory modes are not feasible in stellar disks.
1461:
1462: Let me now suppose that for some pattern speed $\Omega_p$ in
1463: the interval (\ref{eq:constraint-pattern-speed-unstable}), the
1464: quantity $\mu^0_m(\textbf{\textit{I}})$ accidentally remains
1465: small (for a group of stars) over a finite duration of time,
1466: but the condition (\ref{eq:condition-for-stationary-points})
1467: is violated. This (likely) symmetry-breaking phenomenon creates
1468: a slim resonant zone of the width $\Delta(t)$ (see
1469: \S\ref{sec:resonant-zones-expand}). There is not any
1470: physical constraint on the evolution of $\Delta(t)$ when
1471: the perturbations are in their early stages. For a steady
1472: rate given by (\ref{eq:growth-of-Delta-versus-s}) more stars
1473: are trapped by the resonant zone if $s>0$ and the density
1474: wave corresponding to the Fourier number $l=0$ is magnified.
1475: As I discussed in \S\ref{sec:angular-momentum-at-resonance},
1476: the angular momentum of the captured stars increases and some
1477: other {\it reacting stars} should therefore respond in order
1478: to recover the angular momentum balance of the disk.
1479: The reacting stars live in the same CR zone defined by
1480: $\mu^0_m({\bf \Omega})\approx {\cal O}
1481: \left (\epsilon e^{st} \right )$.
1482:
1483: As more stars are trapped into resonance by the $l=0$
1484: component, non-resonant components with $\vert l\vert >0$
1485: should also involve more stars to compensate the angular
1486: momentum deficit. So the amplitudes of other wave components
1487: increase as well. This is what I am considering as the
1488: origin of instabilities: triggering unsteady density
1489: perturbations by the capture of stars into the CR.
1490: Angular momentum transfer between non-resonant
1491: stars cannot be facilitated if a resonant gap does not open
1492: in the phase space. In fact, the initial destabilizing
1493: imbalance of ${\cal L}$ is generated by an irreversible
1494: engagement of resonant stars even if the amount of angular
1495: momentum that they absorb is small (e.g., Figure 8 in JH).
1496: A perturbation with the property of
1497: (\ref{eq:condition-for-stationary-points}) fixes
1498: $s=0$, does not awake stars of $\vert l\vert >0$ components,
1499: and it creates a stationary mode. I conclude that most
1500: (if not all) unstable modes disappear for $\Omega_p>\Omega_0$
1501: just because the CR is destroyed in such a circumstance.
1502:
1503: The $l=0$ component is not always an angular momentum absorber.
1504: For example, in models with an inner cutout of the DF, one has
1505: $\partial f_0/\partial J_{\phi}>0$ which enforces $L_m(0)<0$
1506: (discussions in \S\ref{sec:angular-momentum-at-resonance} and JH).
1507: Figures 2 and 11 in JH demonstrate two modes that show angular
1508: momentum emission by the $l=0$ component. The pattern speeds
1509: of some C-modes explored in Paper I exceed $\Omega_0$ marginally.
1510: Those modes also have $L_m(0)<0$. Unstable modes with
1511: $\Omega_p>\Omega_0$ are barely observed in modal calculations
1512: because their large pattern speeds shrink them to the galactic
1513: center. Coexistence of mode B1 and a compact C-mode can explain
1514: the origin of double-barred galaxies. For a softened-gravity
1515: model of the exponential disk with an inner cutout, Alar Toomre
1516: (private communication) also finds an inner edge mode that has
1517: a corotation circle inside its pattern. Toomre's inner edge mode
1518: is another example of a mode with $\partial f_0/\partial J_{\phi}>0$
1519: and $L_m(0)<0$. Axisymmetric features/grooves introduced by
1520: \citet{SK91} to the surface density and DF also develop a spiral
1521: mode whose pattern is discontinuous at the location of the groove.
1522: That discontinuity is due to the rapid change in the sign of
1523: $\partial f_0/\partial J_{\phi}$: while the $l=0$ component
1524: boosts the angular momentum of stars outside the groove,
1525: it has an opposite effect on inner stars.
1526:
1527: Each unstable mode occupies a finite region in the frequency space
1528: due to the finiteness of its CR zone. This is how divisions (gaps)
1529: are created between the pattern speeds of unstable modes, the ${\bf
1530: \Omega}$-space is quantized by resonances, and finally, unstable
1531: modes constitute a discrete family. Growth rates are sorted according
1532: to the likelihood of resonant capture, which is proportional to the
1533: initial population of stars determined by the equilibrium DF. In a
1534: stellar disk with a falling density profile, the growth rates of
1535: linear modes are sorted in a decreasing order from compact modes
1536: to more extensive ones. Most barred and spiral modes obey this
1537: rule (see Paper I). The perturbation theory of
1538: \S\ref{sec:perturbed-stellar-dynamics} says that stars live
1539: on the integral manifolds of constant $I_2$ before and after
1540: capture into resonance. In order to guarantee a sustained growth
1541: of the CR zone, the initial DF $f_0(\textbf{\textit{J}})$ must
1542: be smooth and non-zero along the curves of constant $I_2$,
1543: and $\mu^{0}_m(\textbf{\textit{I}})$ has to flip its sign
1544: there. A resonant zone cannot grow in a region of the phase space
1545: which is devoid of stars from the beginning, or its stars have been
1546: forced to migrate by unsteady processes. The first-order perturbation
1547: theory is valid as long as the CR zone of each mode is inaccessible
1548: by the stars of other modes and $I_2$ is an adiabatic invariant.
1549:
1550: %
1551: \begin{figure}
1552: \centerline{\hbox{\includegraphics[width=0.45\textwidth]
1553: {f6.eps} } }
1554: \caption{The resonant zones of modes B1, S1, S3 and S6 for
1555: the cored exponential disk with $(N,\lambda,\alpha)=(6,1,0.42)$.
1556: Contour plots show the IICs of $\overline {\cal K}(2,0)$ for
1557: $I_2=0$ and $\epsilon e^{st}=0.01$. Darker regions
1558: correspond to larger values of $\overline {\cal K}(2,0)$.
1559: Contour levels of each zone have been normalized to
1560: the maximum value of $\overline {\cal K}(2,0)$ in the
1561: same zone.
1562: \label{pic:zones-B1-S1-S3-S6}}
1563: \end{figure}
1564: %
1565:
1566: %
1567: \begin{figure*}
1568: \centerline{\hbox{\includegraphics[width=0.45\textwidth]
1569: {f7a.eps}}
1570: \hbox{\includegraphics[width=0.45\textwidth]
1571: {f7b.eps}} }
1572: \caption{{\it Left panel}: Contour plots of $A^0_2({\bf \Omega})$
1573: for modes B1, S1, S3 and S6 of the cored exponential disk with
1574: $(N,\lambda,\alpha)=(6,1,0.42)$. The levels of isocontours have
1575: been normalized to the maximum amplitude. {\it Right panel}:
1576: The emergence of ILR through the intersection of the lines
1577: $\mu^{-1}_m({\bf \Omega})=0$ with the frequency space.
1578: The line $\mu^0_m({\bf \Omega})=0$ (it is independent
1579: of $m$) has also been plotted to show the location of the CR.
1580: In all cases the pattern speed has been set to $\Omega_p=0.1$.
1581: There is no ILR associated with $m=1$ perturbations for
1582: any $\Omega_p>0$. \label{pic:zones-overlap}}
1583: \end{figure*}
1584: %
1585:
1586: \section{MODE SATURATION AND STABLE DISKS}
1587: \label{sec:mode-saturation}
1588:
1589: Perhaps the most important issue related to the growth of
1590: instabilities is that the amplitudes of unstable modes are
1591: saturated after several pattern rotations. Saturation
1592: is indeed a nonlinear phenomenon and can have complex routes.
1593: \citet{SB02} suggest that the growth of a mode is stopped by
1594: the emergence of horseshoe orbits. What they call
1595: ``horseshoe orbits" are simply the librational orbits captured
1596: by the CR. I found those orbits the main cause of instabilities.
1597: I agree with \citet{SB02} that the linear perturbation theory
1598: must fail at some stage, but it does not necessarily imply
1599: that resonant zones can self-control their expansion beyond
1600: the linear regime. Using the phase space geometry and the
1601: nonexistence of the first integral $I_2$ at the boundaries
1602: of neighboring resonances, I argue that resonance overlapping
1603: is an alternative and efficient mechanism to stop the growth of
1604: unstable modes. The stabilization of stellar disks in the
1605: presence of the ILRs is also explained by the same mechanism.
1606: I then present a simple criterion for the stability of
1607: non-axisymmetric perturbations, and discuss about the
1608: stability of self-consistent scale-free disks.
1609:
1610:
1611: \subsection{Resonance Overlapping}
1612: \label{sec:resonance-overlap}
1613:
1614: As a mode grows, the width of its CR zone increases proportional
1615: to $\sqrt{\epsilon e^{st}}$ until two adjacent zones, corresponding
1616: to modes of different pattern speeds, overlap and compete for trapping
1617: the stars in the overlapping zone. This is the moment that a chaotic
1618: layer occurs in the phase space according to Chirikov's (1979) theory,
1619: and stars living in that layer repeatedly migrate between different
1620: resonant zones. Thus, the share contributed to each competing resonant
1621: zone from stars in the overlapping region drops substantially and the
1622: growth is suppressed. A natural consequence of the emergence of a
1623: chaotic layer is the radial migration of stars that heats up the
1624: disk. Some modes may be gradually dissolved as their
1625: Kolmogorov-Arnold-Moser (KAM) tori around stable periodic orbits are
1626: destroyed by modes with larger resonant zones and growth rates. I
1627: remark that the weighted residual form of the CBE that I obtained in
1628: Paper I is valid as long as we are allowed to average out angle
1629: variables from dynamical equations. After the occurrence of chaotic
1630: orbits it is impossible to average out resonant angles and extra
1631: coupling terms between the amplitudes of unstable modes are
1632: contributed to the reduced CBE. Details of my numerical
1633: simulations in the nonlinear regime and in the presence of
1634: chaotic orbits will be presented in Paper III.
1635:
1636: Figure \ref{pic:zones-B1-S1-S3-S6} shows the $l=0$ resonant
1637: zones of modes B1, S1, S3, and S6 for the initially circular
1638: orbits ($I_2=0$) of the model introduced in \S\ref{sec:used-model}.
1639: This is the first demostartion of its kind that shows how the
1640: resonant zones of different unstable modes partition the phase
1641: space of a stellar disk. Since the perturbations grow according
1642: to an exponential law, resonant zones had been slim rings infinitely
1643: long ago. The structure shown in Figure \ref{pic:zones-B1-S1-S3-S6}
1644: has been obtained by setting $\epsilon e^{st}=0.01$ for all modes.
1645: Correspondingly, resonant gaps are of comparable size at the
1646: displayed moment but they will not remain so because the growth
1647: rates are different. Left panel in Figure \ref{pic:zones-overlap}
1648: shows the amplitude function $A^0_2({\bf \Omega})$ of the same
1649: modes of Figure \ref{pic:zones-B1-S1-S3-S6} but this time in the
1650: frequency space that covers all possible values of $I_2$.
1651: Arrows indicate the expansion direction of the $l=0$ Fourier
1652: component as the modes grow. The growth rates of modes are
1653: different and mode S6, which is at the bifurcation point
1654: of the S-family (see Figure 2{\em a} in Paper I), has the
1655: smallest $s$. The configuration of the CR zones in the
1656: phase/frequency space and their thickening according to
1657: equation (\ref{eq:growth-of-Delta-versus-s}) suggest that
1658: resonance overlapping is likely in stellar disks with rich
1659: branches of unstable modes in their eigenspectra.
1660:
1661: To find out how and when an overlap takes place, I assume
1662: two growing waves with the eigenvalues $m\Omega_{p,1}+\imath s_1$
1663: and $m\Omega_{p,2}+\imath s_2$ that successively appear in the
1664: spectrum with $\Omega_{p,1}>\Omega_{p,2}$. These waves develop
1665: CR zones of widths $\Delta_1(I_2)$ and $\Delta_2(I_2)$ on each
1666: manifold of constant $I_2$, and I define $d_{12}(I_2)$ as the
1667: distance between the instantaneous elliptic points of those zones.
1668: Let me suppose that the global minimum of $d_{12}(I_2)$ corresponds
1669: to the critical action $I_2=I^{\rm cr}_2$. An overlap thus occurs if
1670: %
1671: \begin{equation}
1672: d_{12}(I^{\rm cr}_2) \lessapprox
1673: \frac 12 \left [ \Delta_1(I^{\rm cr}_2)+
1674: \Delta_2(I^{\rm cr}_2)
1675: \right ].
1676: \label{eq:condition-for-1st-overlap}
1677: \end{equation}
1678: %
1679: On the other hand, the magnitude of $d_{12}(I^{\rm cr}_2)$
1680: can be estimated as
1681: %
1682: \begin{equation}
1683: d_{12}(I^{\rm cr}_2)\approx
1684: I_{1,2}-I_{1,1}+{\cal O}\left (\epsilon e^{st} \right ),
1685: \label{eq:estimate-of-d12}
1686: \end{equation}
1687: %
1688: where $\Omega_{\phi}\left (I_{1,j},I^{\rm cr}_2 \right)=\Omega_{p,j}$
1689: ($j$=1,2). By combining (\ref{eq:condition-for-1st-overlap}) and
1690: (\ref{eq:estimate-of-d12}), and using
1691: (\ref{eq:growth-of-Delta-and-dotDelta}), I obtain
1692: %
1693: \begin{equation}
1694: I_{1,2} - I_{1,1}
1695: \lessapprox \left [ \sqrt{\epsilon e^{s_1t}}{\cal O}_1(1)
1696: \!+\!
1697: \sqrt{\epsilon e^{s_2t}}{\cal O}_2(1) \right ]_{I_2=I^{\rm cr}_2},
1698: \label{eq:final-condition-for-overlap}
1699: \end{equation}
1700: %
1701: which will be satisfied in the linear regime at some $t=t_{\rm cr}$
1702: for a sufficiently small value of $I_{1,2}-I_{1,1}$. If the post-linear
1703: changes in the DF allow for a sustained expansion of resonant zones (when
1704: the amplitude growth is no longer exponential), one can still anticipate
1705: an overlap for larger values of $I_{1,2}-I_{1,1}$. Cooling a stellar disk
1706: increases the number of unstable modes (Figure 6 in Paper I) and decreases
1707: the gaps between the CR zones. Consequently, transition to chaos is faster
1708: in cold disks than hot ones. This result is in harmony with Sellwood's
1709: (2007, private communication) $N$-body experiments. The very first overlapping
1710: does not necessarily occur between the modes of the same azimuthal wavenumber
1711: $m$. The pattern speeds of two rapidly growing modes of different $m$ may
1712: result in a very small $I_{1,2}-I_{1,1}$ and thus cause an early-stage
1713: overlap.
1714:
1715: The fundamental bar mode may still be saturated in the
1716: absence of spiral modes and in a shorter time scale than
1717: chaotic diffusion of stars. When the resonant zone of mode
1718: B1 grows, the inner homoclinic loop rapidly shrinks to
1719: central regions, the supply of stars to the CR zone is cut
1720: from the galactic center, and the bar mode saturates there.
1721: On the other hand, the bar continues to elongate because
1722: of the expansion of the outer homoclinic loop. The elongation
1723: rate, however, will slow down if $\partial f_0/\partial J_{\phi}<0$.
1724: A declining rate of amplitude growth gives time to the trapped
1725: stars to fill their invariant tori and maintain the bar in
1726: outer regions.
1727:
1728:
1729: \subsection{The Role of Inner Lindblad Resonances}
1730: \label{sec:role-of-ILR}
1731:
1732: Disappearance of unstable modes when the pattern speed violates
1733: (\ref{eq:condition-for-NO-ILR}) can also be explained by the same
1734: mechanism that saturates instabilities. The ILR emerges when the
1735: line $\mu^{-1}_m({\bf \Omega})=0$ crosses the circular orbit
1736: boundary $\Omega_i=\Gamma_c(\Omega_R)$. This happens while the line
1737: $\mu^0_m({\bf \Omega})=0$ has a crossing too and the CR zone has
1738: already been created. The competition between the ILR and CR near
1739: their neighboring boundaries can therefore prohibit both resonances
1740: from broadening their zones and the disk develops a stationary wave.
1741: Right panel in Figure \ref{pic:zones-overlap} shows
1742: the frequency space of the cored exponential disk together with the
1743: lines $\mu^{0}_m({\bf \Omega})=0$ and $\mu^{-1}_m({\bf \Omega})=0$
1744: that have been drawn for $\Omega_p=0.1$ and $m=$1,2,3.
1745: The line $\mu^{-1}_1({\bf \Omega})=0$ never intersects the
1746: frequency space independent of the magnitude of $\Omega_p>0$.
1747: A crossing is possible for $m\ge 2$ and the slope of
1748: $\mu^{-1}_m({\bf \Omega})=0$ increases proportional to $m$.
1749: Subsequently, the distance between the CR and ILR zones decreases
1750: by increasing $m$ and the disk shows more reluctance against
1751: developing an unstable mode (see Figure 1 in Paper I).
1752: \citet{Bertin77} had also reached a similar result for their
1753: three- and four-armed modes, and interpreted it based on the
1754: closeness of ILR and CR but with no mention of the increased
1755: likelihood of resonance overlapping.
1756:
1757: Having resonant zones of finite width for both the CR and ILR
1758: is an essential prerequisite for their subsequent competition.
1759: The $l$th Fourier component in the radial angle can trap stars
1760: into a growing resonant zone if $f_0(\textbf{\textit{J}})$ is
1761: smooth and non-zero on a range of manifolds of constant
1762: $I_2=J_R-lJ_{\phi}/m$, and $\mu^{l}_m({\bf \Omega})$ can switch
1763: its sign in that neighborhood. Having some population of stars
1764: on circular orbits always guarantees the creation and growth
1765: of the CR zone because $I_2$ is identical to $J_R$ and
1766: $I_1=J_{\phi}/m$ can vary independently. However, if the ILR
1767: exists, its adiabatic invariant is defined by $I_2=J_R+J_{\phi}/m$
1768: and the variation of $I_1$ in the vicinity of a resonant value
1769: implies the variation of $J_R$. Therefore, the ILR can develop
1770: a growing resonant zone if $f_0(\textbf{\textit{J}})$ has a
1771: functional dependence on both actions. Stellar disks with such
1772: a property will have non-zero radial velocity dispersions
1773: (warm disks).
1774:
1775: Let me now consider a group of stars moving on near-circular orbits
1776: that satisfy $\mu^{-1}_m({\bf \Omega})>0$ and have been suddenly
1777: trapped into resonance by the ILR. On the resonant tori, the orbits
1778: of those stars are deformed so that $I_2=J_R+J_{\phi}/m$ remains
1779: constant. This means that orbits are elongated as $J_R$ continuously
1780: increases from its tiny initial value and $J_{\phi}$ drops until
1781: $\mu^{-1}_m({\bf \Omega})$ becomes negative and phase space flows
1782: reach their turning points. At this moment, some stars are likely
1783: to find themselves on the integral manifolds of constant $J_R \gg 0$
1784: and join the CR zone while they carry energy and angular momentum
1785: away from the ILR. A reverse migration is also possible. The
1786: switchings can only take place in the overlapping region of the ILR
1787: and CR where highly eccentric orbits live and
1788: $\mu^{-1}_m({\bf \Omega})<0<\mu^{0}_m({\bf \Omega})$. An effective
1789: early-stage overlap requires a non-zero $f_0(\textbf{\textit{J}})$
1790: in the cross section ${\cal S}^{0}_{m} \cap {\cal S}^{-1}_{m}$ of
1791: the sets
1792: %
1793: \begin{eqnarray}
1794: {\cal S}^{0}_m &=& \left \{ {\bf\Omega} :
1795: \mu^{0}_m({\bf\Omega}) \approx{\cal O}
1796: \left( \epsilon \right ) \right \}, \\
1797: {\cal S}^{-1}_m &=& \left \{ {\bf\Omega} :
1798: \mu^{-1}_m({\bf\Omega})\approx{\cal O}
1799: \left( \epsilon \right )
1800: \right \}.
1801: \end{eqnarray}
1802: %
1803: As a result, infinitesimal wave amplitudes (in the linear regime)
1804: can suppress their own growth due to the competition between
1805: the CR and ILR. In the overlapping region
1806: ${\cal S}^{0}_{m} \cap {\cal S}^{-1}_{m}$ neither $I_2=J_R$
1807: (for $l=0$) nor $I_2=J_R+J_{\phi}/m$ (for $l=-1$) are conserved
1808: and that region is filled by chaotic orbits. The chaotic layer
1809: should be very narrow in the linear regime, but it will become
1810: thick in steady large-amplitude bars \citep{C83}.
1811:
1812: The stability of warm stellar disks for $\Omega_p<\Omega_{\rm ILR}(m)$
1813: can be deduced from an independent mathematical reasoning as follows.
1814: Equation (\ref{eq:growth-of-Delta-versus-s}) applies to any growing
1815: resonant gap in the phase space. Let me assume that the ILR and CR
1816: of a wave with the pattern speed $\Omega_p$ are widening their
1817: resonant zones with the rates
1818: %
1819: \begin{equation}
1820: \dot \Delta_{\rm CR}/\Delta_{\rm CR} \sim s_1/2,~~
1821: \dot \Delta_{\rm ILR}/\Delta_{\rm ILR} \sim s_2/2.
1822: \label{eq:Delta-for-CR-and-ILR}
1823: \end{equation}
1824: %
1825: In an equilibrium disk with $d\Sigma_0/dR<0$, the probability
1826: of capture into the ILR is more than the CR, and according to
1827: (\ref{eq:Delta-for-CR-and-ILR}), we need $s_2>s_1$ to assure
1828: a faster expansion rate of the ILR zone. This inequality
1829: contradicts the basic assumption of linear analysis that
1830: separates the time-dependent part of perturbations as
1831: $\exp(-\imath \omega t)$ and assigns the same growing envelope
1832: and pattern speed to all parts of a normal mode. Therefore,
1833: the only possibility for a disk with a monotonic density
1834: gradient will be $s_1=s_2=0$. In cold disks with
1835: $f_0(\textbf{\textit{J}})=\delta(J_R)g_0(J_{\phi})$,
1836: where $\delta(J_R)$ is Dirac's delta function, the ILR
1837: cannot open a resonant gap because it suffers from the
1838: lack of fuel (stars) needed for a possible expansion
1839: along the curves of constant $I_2=J_R+J_{\phi}/m$.
1840: This automatically implies $s_2=0$ and a unique value
1841: is assigned to $s_1$. The CR zone can thus grow freely
1842: and destabilize the disk.
1843:
1844: It is worth noting that with an inner cutout, the surface
1845: density rises in central regions, reaches to a maximum
1846: and then decays outwards. Denote $R_{\rm CR}$ and $R_{\rm ILR}$
1847: as the corotation and inner Lindblad radii, respectively.
1848: If the cutout is not sharp and the density peak lies fairly
1849: between $R_{\rm ILR}$ and $R_{\rm CR}$
1850: (see the leftmost panel in Figure 8 of Evans \& Read 1998b),
1851: it will be possible to find two points of the same density
1852: on the curve of $\Sigma_0(R)$: one near the ILR on the
1853: rising part of $\Sigma_0(R)$ and the other near the CR
1854: on the falling side, i.e.,
1855: $\Sigma_0(R_{\rm ILR})\approx \Sigma_0(R_{\rm CR})$.
1856: Thus, the ILR and CR can manage to grow with the same
1857: rate $s_1=s_2$ and an unstable mode emerges. Such modes
1858: have been demonstarted in Figure 7 of \citet{ER98b} whose
1859: cutout functions immobilize the circular orbits of the
1860: galactic center and all near-radial orbits. Accordingly,
1861: the population of active stars in the overlapping region
1862: of the ILR and CR declines substantially and the simultaneous
1863: growth of both resonances becomes likely. By decreasing the
1864: pattern speed, the ILR moves outside the cutout radius and
1865: its resonant zone becomes visible to the CR. The stellar
1866: disk is therefore stabilized. The condition $s_1=s_2\not =0$
1867: cannot be reached if $\Sigma_0(R_{\rm CR})$ differs
1868: remarkably from $\Sigma_0(R_{\rm ILR})$. This happens
1869: when both the ILR and CR lie outside the cutout radius or
1870: when the cutout is so sharp.
1871:
1872: The above analysis shows how diverse the response of
1873: disk galaxies may be to non-axisymmetric excitations.
1874: \citet{T81} suggested that the two parameters $Q$ and $X$
1875: decide the fate of non-axisymmetric density waves.
1876: Wave mechanics based on the behavior of resonances,
1877: shows more complex decisive factors that govern the
1878: evolution of density waves, specially near the critical
1879: values of $\Omega_p\approx \Omega_{\rm ILR}(m)$ and
1880: for non-monotonic gradients of equilibrium density.
1881:
1882:
1883: \subsection{A Criterion for Global Stability}
1884: \label{sec:criterion-for-stability}
1885:
1886: I am now in a position to provide a simple criterion for the
1887: stability of stellar disks in the absence of the ILR. It is the CR
1888: that captures stars, contributes an angular momentum imbalance to
1889: the stellar disk, and triggers unstable modes. Circumferential waves
1890: will be {\it globally stable} if resonant gaps do not open near
1891: the CR for all possible pattern speeds. Consequently, $L_m(0)$
1892: should remain zero, which according to
1893: equation (\ref{eq:torque-component-1}) implies
1894: %
1895: \begin{equation}
1896: \Lambda^{m0}_{jk}=0,~~\forall j,k\ge 0.
1897: \end{equation}
1898: %
1899: This condition is satisfied only if [see equations
1900: (\ref{eq:define-Lambda}) and (\ref{eq:define-Phi-vs-Psi})]
1901: %
1902: \begin{equation}
1903: \frac{\partial f_0}{\partial J_{\phi}}=0,~~ {\rm
1904: for~all}~\textbf{\textit{J}} \in \Re^{+}\times \Re^{+}.
1905: \label{eq:condition-stability-1}
1906: \end{equation}
1907: %
1908: The stability criterion given in (\ref{eq:condition-stability-1})
1909: can also be deduced from the earlier works of LBK and JH as I
1910: explain below. An alternative form of (\ref{eq:total-torque}) is
1911: given in equation (B6) of JH as
1912: %
1913: \begin{eqnarray}
1914: {d{\cal L} \over dt} &=&
1915: -2ms\pi^2e^{2st} \sum_{l=-\infty}^{\infty}
1916: \int d\textbf{\textit{J}}
1917: \left(l{\partial f_0\over \partial J_R}
1918: +m{\partial f_0\over \partial J_{\phi}} \right)
1919: \nonumber \\
1920: &{}& \times
1921: {|\tilde V_{l}|^2 \over |l\Omega _R+m\Omega _{\phi} -\omega|^2}.
1922: \label{eq:dL-dt-from-JH}
1923: \end{eqnarray}
1924: %
1925: where
1926: %
1927: \begin{equation}
1928: \tilde V_{l}=\sum_{j=0}^{\infty} \tilde a^m_j
1929: \tilde \Psi^{ml}_j(\textbf{\textit{J}}).
1930: \end{equation}
1931: %
1932: Dividing equation (\ref{eq:dL-dt-from-JH}) by $16\pi^4$
1933: and changing the sign of $\omega$ lead to equation (28) in LBK.
1934: The amount of angular momentum emitted/absrobed by the $l=0$
1935: component is therefore determined by
1936: %
1937: \begin{equation}
1938: \left [ {d{\cal L} \over dt} \right ]_{l=0}=-2(m\pi)^2
1939: se^{2st} \int d\textbf{\textit{J}}
1940: {\partial f_0\over \partial J_{\phi}}
1941: {|\tilde V_{0}|^2 \over |m\Omega _{\phi} -\omega|^2},
1942: \label{eq:dL-dt-from-JH-CR}
1943: \end{equation}
1944: %
1945: which reduces to equation (96) of \citet{GT79} as $s\rightarrow 0$.
1946: It is seen that the sign of $\partial f_0/\partial J_{\phi}$
1947: determines whether the angular momentum is added to, or drained from
1948: the $l=0$ wave component. Furthermore, the CR will not imbalance
1949: ${\cal L}$ if the condition (\ref{eq:condition-stability-1}) holds.
1950: This means that the condition (\ref{eq:condition-for-stationary-points})
1951: cannot be violated according to my discussion of
1952: \S\ref{sec:origin-of-instabilities}, and $s$ should vanish for
1953: arbitrary perturbations. A stellar disk that satisfies
1954: (\ref{eq:condition-stability-1}) will be hot because it is
1955: impossible to reproduce the surface density gradient in the limit of
1956: $J_R\rightarrow 0$ with a DF that does not depend on $J_{\phi}$. It
1957: has now become more transparent why there is a correlation between
1958: Toomre's (1964) $Q$ and the disk stability for non-axisymmetric
1959: excitations. An illustrative example has been given in Figure 6 of
1960: Paper I. Most of S-modes disappear when the parameter $N$ of the DF
1961: is decreased and more stars are distributed on near-radial orbits.
1962:
1963:
1964: \subsection{The Stability of Scale-Free Disks}
1965: \label{sec:scale-free-disks}
1966:
1967: Stability analysis of cuspy mass models has been a big puzzle in
1968: stellar dynamics since \citet{Z76} computed the global modes of
1969: Mestel's (1963) disk and showed the importance of $m=1$ excitations.
1970: To tackle the singular center of Mestel's disk, he introduced
1971: an inner cutout that freezes central stars with diverging orbital
1972: frequencies. \citet{ER98a,ER98b} followed the same procedure but
1973: for different cusp slopes. In spite of all these remarkable
1974: efforts, what H02 illustrated for the frequency space of
1975: scale-free potentials was indeed the key to resolve the
1976: long-standing stability problem of scale-free disks without
1977: any simplifying cutouts. The map of the angular sector that
1978: H02 plotted in his Figure 1 for the scale-free potentials
1979: \citep{TT97}
1980: %
1981: \begin{eqnarray}
1982: V_0(R)= \biggl \{
1983: \begin{array}{ll}
1984: {\rm sign}(\beta)R^{\beta}, & ~ \beta \not =0,~
1985: -1< \beta < 2, \\
1986: \ln R, & ~ \beta=0,
1987: \end{array}
1988: \end{eqnarray}
1989: %
1990: will be an angular sector in the $(\Omega_R,\Omega_i)$-space
1991: with the $\Omega_R$-axis and the straight line
1992: %
1993: \begin{equation}
1994: \Omega_i=\left ( \frac{1}{\sqrt{2+\beta}}-\frac 12 \right )\Omega_R,
1995: \end{equation}
1996: %
1997: being its lower and upper boundaries, respectively. The slope of the
1998: upper boundary is always less than or equal to $1/2$. It is therefore
1999: obvious that except for $m=1$, all $\mu^{-1}_m({\bf \Omega})=0$ and
2000: $\mu^{0}_m({\bf \Omega})=0$ lines will intersect the frequency space
2001: (whatever $\Omega_p$ may be) and resonance overlapping between the CR
2002: and ILR can occur if there are enough number of non-circular orbits.
2003: In other words, the radial velocity dispersion $\tilde \sigma_u$
2004: (defined in Evans \& Read 1998a) should exceed a critical value
2005: $\tilde \sigma_{u,\rm min}(m)$ to give rise to nonempty cross sections
2006: ${\cal S}^{0}_{m}\cap {\cal S}^{-1}_{m}$. This is possible only
2007: for $m>1$ excitations and it suggests that scale-free disks are
2008: subject to $m=1$ instabilities. Finding $\tilde \sigma_{u,\rm min}(m)$
2009: is an open, yet conceivable, problem that can generalize Toomre's
2010: (1964) criterion to non-axisymmetric perturbations of scale-free
2011: disks. I explained in \S\ref{sec:role-of-ILR} that the ILR cannot
2012: help to stabilize cold disks. That argument applies to self-consistent
2013: scale-free disks as well, and strongly justifies the existence
2014: of $\tilde \sigma_{u,\rm min}(m)$ for $m>1$. Moreover, a stability
2015: condition like $\tilde \sigma_u> \tilde \sigma_{u,\rm min}(m)$ may
2016: correlate with Lynden-Bell's (1993, equation 6-23) criterion, which
2017: has been formulated based on the angular dispersion of lobe tumble
2018: rates.
2019:
2020:
2021:
2022: \section{DISCUSSION AND CONCLUSIONS}
2023: \label{sec:discussion-and-conclusions}
2024:
2025: I had already given an evidence in \S\ref{sec:modes-in-frequency-space}
2026: that LBK's theory cannot be extended to large growth rates. Here I provide
2027: a more detailed analysis of LBK's mode mechanism. Taking the $s\rightarrow 0$
2028: limit of (\ref{eq:dL-dt-from-JH}) leads to equation (30) of LBK.
2029: The resulting integrand of the $l$th component involves the term
2030: $\delta(l\Omega_R+m\Omega_{\phi}-\omega)$ but its
2031: argument never becomes zero for $l\le -1$ as long as $\Omega_p$
2032: exceeds $\Omega_{\rm ILR}(m)$. Once the delta function vanishes for
2033: $\Omega_p>\Omega_{\rm ILR}(m)$, the angular momentum content of $l\le -1$
2034: components remains zero. Therefore, angular momentum transfer between
2035: inner and outer resonances as suggested by LBK, is feasible only for
2036: $\Omega_p < \Omega_{\rm ILR}(m)$. This is a constraint on the
2037: applicability of LBK's theory.
2038:
2039: According to the arguments of \S\ref{sec:role-of-ILR} and the WKB
2040: theory, the ILR is not transparent to short wavelength disturbances
2041: with $X\gg 1$ \citep{BT08,Mark74} and any developing spiral structure
2042: must be damped unless the ILR remains invisible to the CR. Therefore,
2043: an inconsistent point in LBK's paper is the imagination of an
2044: angular-momentum-transferring spiral structure while the condition
2045: $\Omega_p <\Omega_{\rm ILR}(m)$ has already abandoned the existence
2046: of such structures. An issue yet needs to be explained: If there is
2047: no spiral structure in stable disks, which mechanism does transfer
2048: the angular momentum between inner and outer resonances?
2049: In fact, in the limit of $s\rightarrow 0$
2050: the density components $\Sigma^{l-}_1$ and $\Sigma^{l+}_1$ become
2051: stationary waves that are azimuthally separated by a phase shift of
2052: $90^{\circ}$ (like the components of mode B1 in
2053: Figure \ref{pic:components-modes-B1-S2}). Thus, they exert
2054: opposite gravitational torques on each other without any need
2055: for a communicating spiral structure.
2056:
2057: In this paper I decomposed unstable modes of a model disk galaxy to
2058: its constituent Fourier components and showed how different
2059: components experience a gravitational torque. My results clearly
2060: showed that only the Fourier component associated with the CR
2061: generates a resonant zone in the phase space, and other components
2062: exchange angular momentum far from resonances. This result led me
2063: to introduce a new dynamical mechanism that triggers unstable
2064: modes. According to my calculations, an irreversible resonant
2065: capture of stars into the CR causes a synchronous precession of
2066: their orbital axes, which in turn, support a rotating density
2067: pattern. The emerged pattern grows because the resonant zone
2068: expands in the frequency space.
2069:
2070: The irreversibility of resonant trapping is the most destructive
2071: event that happens in a stellar disk when a group of resonant stars with
2072: $d_i<I_1<d_o$ gain angular momentum forever. An immediate consequence
2073: of this phenomenon is the angular momentum imbalance in the disk,
2074: which requires proper reaction of other stars in order to conserve
2075: ${\cal L}$. This is how the Fourier components of $l\not =0$ are
2076: generated and the angular momentum transfer between them is initiated.
2077: Reacting stars do not have a librating $\theta$ and they cooperate
2078: in a way that the quantity $\vert \mu^0_m({\bf \Omega})\vert$ remains
2079: small for all components.
2080:
2081: Using the maps of resonance zones in the phase and frequency spaces,
2082: I argued that resonance overlapping should stop the growth of unstable
2083: modes and stabilize stellar disks in the presence of an ILR. I showed
2084: that for $m>1$ the emergence of an ILR is inevitable in scale-free
2085: models. The competition between the CR and ILR can therefore stabilize
2086: sufficiently warm scale-free disks against $m>1$ excitations.
2087: \citet{ER98b} suggested in \S6.1 of their paper that self-consistent
2088: scale-free disks (without inner cutouts) do not admit growing
2089: non-axisymmetric modes at all, and they ruled out the possibility of
2090: a critical temperature. Their prediction is in agreement with my results
2091: except in two aspects: (i) Since the line $\mu^{-1}_{1}({\bf \Omega})=0$
2092: does not intersect the frequency space of scale-free models, there is
2093: no ILR to compete with a growing CR and density waves will be amplified
2094: for $m=1$. (ii) The ILR in cold disks is very special and it cannot
2095: develop a resonant zone of finite size. It is therefore hard to believe
2096: a serious influence by the ILR on the phase space flows up to a critical
2097: temperature.
2098:
2099: Note that each unstable mode dominates an isolated region of the
2100: frequency (action) space until a chaotic layer emerges due to resonance
2101: overlapping. The rate by which the chaotic layer diffuses itself
2102: in the phase space, is determined by the resistance of KAM tori
2103: against competing resonant zones. The limited space that the fastest
2104: growing bar mode occupies after its saturation (Khoperskov et al. 2007),
2105: is an $N$-body evidence for such a resistance of spiral modes that
2106: fill outer regions of the cored exponential disk.
2107:
2108: The other implication of resonance overlapping, which has
2109: observational support too, is that the central bar in grand-design
2110: barred spiral galaxies must join the spiral arms at the tips of the
2111: bar, which are the overlapping regions of two neighboring B and S
2112: modes in the frequency space (see Figure \ref{pic:zones-overlap}). I
2113: note that the pattern speeds of the spiral and bar components are
2114: not the same because these structures are associated with different
2115: modes. The overlapping region is filled by chaotic orbits that yield
2116: the turbulent mixing of density waves. This process can enhance star
2117: formation near the tips of the bar. If we accept this scenario,
2118: barred spiral galaxies should have been formed from hot stellar
2119: disks whose eigenfrequency spectra include few spiral modes. On the
2120: other hand, flocculent spirals with many wave packets will be the
2121: natural destiny of initially cold disks that give birth to a rich
2122: family of spiral modes.
2123:
2124: The mathematical background for the nonlinear evolution of modes was
2125: developed in Paper I but that formulation is valid as long as
2126: averaging is allowed over angle variables. In the presence of
2127: chaotic orbits it is impossible to average out resonant angles and
2128: some modifications are required to deal with the CBE in its full
2129: nonlinear form. I will present such modifications in Paper III with
2130: the aim of discovering the dynamical processes that generate
2131: different classes of barred and spiral structures.
2132:
2133:
2134: \acknowledgments
2135:
2136: I thank Scott Tremaine and Chris Hunter for their encouragement
2137: and stimulating discussions in the course of my calculations and
2138: writing the manuscript. I also express my sincere thanks to the
2139: referee whose illuminating comments helped me to further my
2140: calculations and thoroughly revise the presentation and
2141: arguments of sections 5 through 8. The major part of this
2142: research was supported by the Institute for Advanced Study
2143: at Princeton, where I was an astrophysics visitor.
2144:
2145:
2146: \begin{thebibliography}{}
2147:
2148: \bibitem[Athanassoula (2003)]{ATHA03} Athanassoula, E. 2003,
2149: \mnras, 341, 1179
2150:
2151: \bibitem[Bertin et al. (1977)]{Bertin77}
2152: Bertin, G., Lau, Y. Y., Lin, C. C., Mark J. W.-K.,
2153: \& Sugiyama, L. 1977, Proc. Natl. Acad. Sci. USA, 74, 4726
2154:
2155: \bibitem[Binney \& Tremaine (2008)]{BT08} Binney, J., \& Tremaine, S.
2156: 2008, Galactic Dynamics (Princeton: Princeton Univ. Press)
2157:
2158: \bibitem[Byrd \& Howard (1992)]{BH92}
2159: Byrd, G. G., \& Howard, S. 1992, \aj, 103, 1089
2160:
2161: \bibitem[Chirikov (1979)]{Ch79} Chirikov, B. V. 1979, Phys. Rep., 52, 263
2162:
2163: \bibitem[Contopoulos (1983)]{C83} Contopoulos, G. 1983, A\&A, 117, 89
2164:
2165: \bibitem[Evans \& Read (1998a)]{ER98a} Evans, N. W., \& Read, J. C. A.
2166: 1998a, \mnras, 300, 83
2167:
2168: \bibitem[Evans \& Read (1998b)]{ER98b} Evans, N. W., \& Read, J. C. A.
2169: 1998b, \mnras, 300, 106
2170:
2171: \bibitem[Goldreich \& Tremaine (1979)]{GT79} Goldreich P.,
2172: \& Tremaine, S. 1979, \apj, 233, 857
2173:
2174: \bibitem[Hohl (1971)]{H71} Hohl, F. 1971, \apj, 168, 343
2175:
2176: \bibitem[Hunter (2002)]{H02} Hunter, C. 2002, in
2177: Disks of Galaxies: Kinematics, Dynamics and Perturbations,
2178: ASP Conference Proceedings, Vol. 275, ed E. Athanassoula,
2179: A. Bosma, \& R. Mujica (San Francisco: Astronomical Society
2180: of the Pacific), 293 (H02)
2181:
2182: \bibitem[Jalali (2007)]{J07} Jalali M. A. 2007, \apj, 669, 218 (Paper I)
2183:
2184: \bibitem[Jalali \& Hunter (2005)]{JHa} Jalali M. A.,
2185: \& Hunter, C. 2005, \apj, 630, 804 (JH)
2186:
2187: \bibitem[Kalnajs (1978)]{K78} Kalnajs, A. J. 1978, in IAU Symp. 77,
2188: Structure and Properties of Nearby Galaxies, ed. E. M. Berhuijsen \&
2189: R. Wielebinski (Dordrecht: Reidel) 113
2190:
2191: \bibitem[e.g., Khoperskov et al. (2007)]{KJKJ07} Khoperskov, A. V.,
2192: Just, A., Korchagin, V. I., \& Jalali, M. A. 2007, A\&A, 473, 31
2193:
2194: %\bibitem[Lau \& Bertin (1978)]{LB78}
2195: % Lau, Y. Y., \& Bertin, G. 1978, \apj, 226, 508
2196:
2197: \bibitem[Lynden-Bell (1993)]{L-B93} Lynden-Bell, D. 1993, in
2198: Galactic Dynamics and N-Body Simulations, editors Contopoulos, G.,
2199: Spyrou, N. K., \& Vlahos, L., Springer-Verlag, Berlin, 3
2200:
2201: \bibitem[Lynden-Bell \& Kalnajs (1972)]{LBK72} Lynden-Bell, D.,
2202: \& Kalnajs, A. J. 1972, \mnras, 157, 1 (LBK)
2203:
2204: \bibitem[Mark (1974)]{Mark74} Mark, J. W.-K. 1974, \apj, 193, 539
2205:
2206: \bibitem[Mathur (1990)]{M90} Mathur, S. D. 1990, \mnras, 243, 529
2207:
2208: \bibitem[Mestel (1963)]{M63} Mestel, L. 1963, \mnras, 126, 553
2209:
2210: \bibitem[Sellwood (1981)]{S81} Sellwood, J. A. 1981,
2211: A\&A, 99, 362
2212:
2213: \bibitem[Sellwood \& Kahn (1991)]{SK91} Sellwood, J. A.,
2214: \& Kahn, F. D. 1991, \mnras, 250, 278
2215:
2216: \bibitem[Sellwood \& Binney (2002)]{SB02} Sellwood, J. A.,
2217: \& Binney, J. J. 2002, \mnras, 336, 785
2218:
2219: \bibitem[Toomre (1964)]{T64} Toomre, A. 1964, \apj, 139, 1217
2220:
2221: \bibitem[Toomre (1981)]{T81} Toomre, A. 1981, in Structure
2222: and Evolution of Normal Galaxies, ed S. M. Fall \& D. Lynden-Bell
2223: (Cambridge: Cambridge Univ. Press), 111
2224:
2225: \bibitem[Toomre (1990)]{T90} Toomre, A. 1990, in Dynamics and
2226: Interactions of Galaxies, ed. Wielen, R., Springer-Verlag,
2227: Berlin, 292
2228:
2229: \bibitem[Touma \& Tremaine (1997)]{TT97} Touma, J.,
2230: \& Tremaine, S. 1997, \mnras, 292, 905
2231:
2232: \bibitem[van Kampen (1955)]{vK55} van Kampen, N. G. 1955,
2233: Physica, 31, 949
2234:
2235: \bibitem[Winter \& Murray (1997)]{WM97} Winter, O.C., \&
2236: Murray, C.D. 1997, A\&A, 319, 290
2237:
2238: \bibitem[Wisdom (1980)]{W80} Wisdom, J. 1980,
2239: \aj, 85, 1122
2240:
2241: \bibitem[Zang (1976)]{Z76} Zang, T. A. 1976, PhD Thesis,
2242: Massachusetts Institute of Technology, Cambridge, MA
2243:
2244: \end{thebibliography}
2245:
2246: \end{document}
2247:
2248: