1: %% Stability by KAM confinement of certain wild, nongeneric relative
2: %% equilibria of underwater vehicles with coincident centers of mass and
3: %% bouyancy
4: %%
5: %% George W. Patrick
6: %% Department of Mathematics and Statistics
7: %% University of Saskatchewan
8: %%
9: %% April 02, 2002
10: %%
11:
12: \documentclass{article}
13: \usepackage{amssymb,amsmath,theorem,chicago,epsf}
14: \usepackage[]{titlesec}
15:
16: \titleformat{\section}[hang]
17: {\scshape\normalsize\filcenter}{\S\thesection}{1em}{}
18: \titleformat{\subsection}[hang]
19: {\scshape\normalsize}{\S\thesubsection}{1em}{}
20: \titleformat{\subsubsection}[runin]
21: {\itshape\normalsize}{\S\thesubsubsection}{.5em}{}[.]
22: \titlespacing{\subsubsection}{0pc}{*1}{1em}
23:
24: {\theorembodyfont{\slshape}
25: \newtheorem{theorem}{\slshape Theorem}
26: \newtheorem{proposition}{\slshape Proposition}
27:
28: \renewcommand\figurename{\footnotesize Figure}
29:
30: \newcommand{\qedsymbol}{\rule{.3\baselineskip}{.35\baselineskip}}
31: \newenvironment{proof}{\begin{trivlist}\item[]{\it Proof.}}{\hspace*{\fill}$\qedsymbol$\end{trivlist}}
32: \newcommand\noqedsymbol{\def\qedsymbol{\relax}}
33:
34: \newcommand{\erf}[1]{\expandafter(\ref{#1})} % for equation numbers
35: \newcommand{\thrf}[1]{\expandafter\ref{#1}} % for theorems and such
36: \newcommand{\ct}[1]{\expandafter\citeANP{#1}~[\expandafter\citeyearNP{#1}]}
37: \newcommand{\lb}[1]{\expandafter\label{#1}}
38: \newcommand{\elb}[1]{\expandafter\label{#1}}
39:
40: \newcommand\mfk\mathfrak
41: \newcommand\mcl\mathcal
42: \newcommand\mbb\mathbb
43: \newcommand\mtl\mathit
44: \newcommand\mbf\mathbf
45:
46: \newcommand\onm\operatorname
47: \newcommand{\set}[2]{\left\{\vphantom{\bigl(}\mskip1mu #1:#2\mskip1mu\right\}}
48: \newcommand{\sset}[1]{\left\{\vphantom{\bigl(}\mskip1mu#1\mskip1mu\right\}}
49:
50: \title{Stability by KAM confinement of certain wild,~nongeneric
51: relative equilibria
52: of underwater vehicles with coincident centers of~mass~and~bouyancy}
53: \author{George W. Patrick}
54: \date{April 2002}
55:
56: \begin{document}\maketitle
57: %%
58: \begin{abstract}
59: Purely rotational relative equilibria of an ellipsoidal underwater
60: vehicle occur at nongeneric momentum where the symplectic reduced
61: spaces change dimension. The stability these relative equilibria
62: under momentum changing perturbations is not accessible by Lyapunov
63: functions obtained from energy and momentum. A blow-up construction
64: transforms the stability problem to the analysis symmetry-breaking
65: perturbations of Hamiltonian relative equilibria. As such, the
66: stability follows by KAM theory rather than energy-momentum
67: confinement.
68: \end{abstract}
69: %%
70: \section*{Introduction}
71: %%
72: The phase space $T\mtl{SE}(3)$ with Lagrangian
73: \begin{equation}\elb{00}
74: L(A,a,\Omega,v)\equiv\frac12\Omega^t\mbf I\Omega+\frac12v^t\mbf Mv
75: \end{equation}
76: approximately models the motion of a neutrally buoyant vehicle
77: submerged in an inviscid irrotational fluid (see~\ct{LeonardNE-1997.1}
78: and the references therein), in the case of coincident centers of mass
79: and buoyancy. Here tangent vectors of $\mtl{SE}(3)$ are represented
80: by left translation and elements of $\mtl{SE}(3)$ parameterize the
81: configurations of the vehicle by embedding a reference vehicle into
82: the fluid. $\mbf I$ and $\mbf M$ are constant, positive definite,
83: $3\times 3$ matrices that can be calculated from the shape and mass
84: distribution of the vehicle. This system admits the $\mtl{SE}(3)$
85: symmetry of the left action of $\mtl{SE}(3)$ on itself.
86:
87: For an ellipsoidal vehicle with principle axes of inertia along the
88: axes of symmetry of the ellipsoid, $\mbf I=\onm{diag}(I_1,I_2,I_3)$
89: and $\mbf M=\onm{diag}(M_1,M_2,M_3)$ (i.e. $\mbf I$ and $\mbf M$ are
90: diagonal). If $M_1=M_2$ and $I_1=I_2$ (or similarly if $M_1=M_3$ and
91: $I_1=I_3$ etc.) then there is a further material symmetry of the
92: system: $\mtl{SO}(2)=\sset{\onm{exp}(\mbf k^\wedge\theta)}$ acts as a
93: subgroup of $\mtl{SE}(3)$ by inverse multiplication on the right. In
94: the case that $\mbf I$ and $\mbf M$ are both constants of the identity
95: then the material symmetry is $\mtl{SO}(3)$.
96:
97: Lie-Poisson reduction yields the Poisson
98: phase space $\mfk{se}(3)^*=\bigl\{(\pi,p)\bigr\}$ where $\pi=\mbf
99: I\Omega$ and $p=\mbf Mv$. The equations of motion are
100: \begin{equation}\elb{60}
101: \frac{d\pi}{dt}=\pi\times\Omega+p\times v,\qquad\frac{dp}{dt}=p\times\Omega,
102: \end{equation}
103: and by direct substitution, for each $\alpha_e\in\mbb R$,
104: \begin{equation*}
105: p_e^{\alpha_e}:\quad\pi=\alpha_e\mbf{k},\qquad p=0,
106: \end{equation*}
107: is an equilibrium of the Poisson reduced systems and hence a relative
108: equilibrium of the original system. The generator is
109: \begin{equation*}
110: \Omega_e^{\alpha_e}\equiv\frac{\alpha_e}{I_3}\mbf k,
111: \quad v_e^{\alpha_e}\equiv0,
112: \end{equation*}
113: so the relative equilibrium corresponds to a stationary vehicle
114: rotating about an axis of symmetry which is aligned with the vertical.
115: This article is concerned with the stability of these relative
116: equilibria in the case where $I_3$ is not an intermediate axis of
117: symmetry, i.e. assuming $I_1<I_2<I_3$ or $I_3<I_2<I_1$.
118:
119: The symplectic leaves of $\mtl{se}(3)^*$ are as follows. On the
120: complement of $p=0$ lie the generic symplectic leaves, all
121: diffeomorphic to $TS^2$, and which are the level sets of the two
122: Casimirs $|p|$ and $\pi\cdot p$. Nongeneric leaves occur within the
123: set $p=0$ and are the level sets of the subcasimir $|\pi|$. Thus the
124: relative equilibria $p_e^{\alpha_e}$ correspond to Lyapunov stable
125: equilibria on the (nongeneric) symplectic leaves of $\mtl{se}(3)^*$
126: since the energy has a definite critical point when restricted to
127: those leaves. Were the symmetry group to be compact this
128: \emph{leafwise stability} would imply stability of the equilibrium
129: modulo the isotropy group of the momentum. $\mtl{SE}(3)$, of course,
130: is not compact. The question is whether or not $p_e^{\alpha_e}$ are
131: stable under perturbations from nongeneric leaves into nearby generic
132: leaves.
133:
134: \ct{LeonardNEMarsdenJE-1997.1} have identified this question as
135: particularly delicate, and the theory
136: of~\ct{PatrickGWRobertsRMWulffC-2001.1}, the sharpest possible for the
137: problem of establishing the stability of relative equilibria by
138: energy-momentum confinement in the case of noncompact symmetry,
139: corroborates that opinion. Patrick \emph{et al.} separate generators
140: of relative equilibria into two complementary classes, \emph{tame} and
141: \emph{wild}. The generator of $p_e^{\alpha_e}$ is tame if and only if
142: $\alpha_e=0$, corresponding to a stationary, nonrotating vehicle, in
143: which case $\mtl{SE}(3)$-stability follows directly, since the energy
144: has zero derivative at $p_e^{\alpha_e}$ and has positive definite
145: Hessian there. However, if $\alpha_e\ne0$, the generator is wild and
146: the theory does not imply stability.
147:
148: So it is an open question whether the relative equilibria
149: $p_e^{\alpha_e}$, $\alpha_e\ne0$, are $\mtl{SE}(3)$-stable or not, and
150: the problem appears inaccessible by energy-momentum confinement. This is
151: due to the presence of a noncompact symmetry group and
152: wild generators.
153:
154: \section{The blow-up construction}
155: %%
156: In its essence the stability issue is one of perturbation from a
157: nongeneric symplectic leaf to nearby, higher dimensional, generic
158: leaves. In order to bridge to Hamiltonian perturbation theory, which
159: is usually cast in a setting of a fixed canonical phase space, it is
160: natural to begin by normalizing the generic leaves. The leaf
161: corresponding to
162: \begin{equation*}
163: |p|=a,\quad\pi\cdot p=b,
164: \end{equation*}
165: for $a>0$ is diffeomorphic to
166: \begin{equation*}
167: TS^2=\set{(w,\dot w)\in\mbb R^3\times\mbb R^3}{|w|=1,w\cdot\dot w=0}
168: \end{equation*}
169: by the map
170: \begin{equation*}
171: w=\frac p{|p|},\quad\dot w=\pi-\frac{\pi\cdot p}{|p|^2}p,
172: \end{equation*}
173: the inverse map being given by, for fixed $a>0$ and $b\in\mbb R$,
174: \begin{equation}\elb{4}
175: p=aw,\quad\pi=\dot w+\frac ba w.
176: \end{equation}
177: Having normalized the symplectic leaves to the constant manifold
178: $TS^2$, one seeks to extend this to the nongeneric leaves within
179: $p=0$, which means extending it to $a=0$, since $p=aw$. As it
180: stands~\erf{4} is poorly defined for $a=0$, but for fixed ratios of
181: $b/a$ it is well defined even for arbitrarily small $a$, suggesting
182: that the proper way to approach the nongeneric leaves from generic
183: ones is through constant $\pi\cdot p/|p|$. \emph{Setting
184: $\gamma\equiv b/a$ and using the parameters $a$ and $\gamma$ instead
185: of $a$ and $b$ codes the generic leaves so they fit smoothly into the
186: nongeneric ones, thus allowing the possibility of an effective
187: perturbation approach.} The map $p=aw$, $\pi=\dot w+\gamma w$ for
188: $a=0$ is many-to-one and so the three dimensional set of nongeneric
189: leaves $p=0$ is ``blown-up'' by this map to the five dimensional set of
190: $TS^2\times\mbb R=\sset{\bigl((w,\dot w),\gamma\bigr)}$. Thus one is
191: led to define the \emph{blown-up space} of $\mtl{se}(3)^*$ as
192: \begin{equation*}
193: \hat P\equiv TS^2\times\mbb R_{\ge 0}\times\mbb R
194: \equiv\set{(w,\dot w,a,\gamma)}{|w|=1,w\cdot w=0, a\ge0}
195: \end{equation*}
196: with \emph{blow-down map}
197: \begin{equation*}
198: p=aw,\quad\pi=\dot w+\gamma w
199: \end{equation*}
200: and corresponding \emph{blow-up map}, defined on the generic ($p\ne0$)
201: leaves only,
202: \begin{equation*}
203: w=\frac p{|p|},\quad\dot w=\pi-\frac{\pi\cdot p}{|p|^2}p,\quad a=|p|,
204: \quad \gamma=\frac ba.
205: \end{equation*}
206: The blow-up map is a diffeomorphism from the (open) set of generic
207: leaves to the (open) set $a>0$ in the blown-up space (the
208: \emph{generic sector}), such that each generic leaf is sent to the
209: constant manifold $TS^2$. The evolution of the generic leaves is
210: transformed to a evolution on $TS^2$ parameterized by the Casimir
211: values $a$ and $\gamma$. The blow-down map takes the set $a=0$ in the
212: blown-up space (the \emph{nongeneric sector}) to the set of nongeneric
213: leaves, and is many-to-one on that sector. Increasing the parameter
214: $a$ from zero corresponds to leaving the nongeneric leaves and moving
215: to the generic ones, while $\gamma$ parameterizes the possible avenues
216: of departure.
217:
218: The utility of the blow-up to support perturbation arguments depends
219: on whether or not the dynamics of the generic sector can be continued
220: smoothly to the nongeneric sector. On the generic sector the vector
221: field that generates the dynamics is
222: \begin{equation}\elb{8}\begin{split}
223: \frac{dw}{dt}&=\frac1a\frac{dp}{dt}=\frac1ap\times\mbf I^{-1}\pi=
224: w\times\mbf I^{-1}(\dot w+\gamma w),\\
225: \frac{d\dot w}{dt}&=\frac{d\pi}{dt}-\gamma\frac{dw}{dt}\\
226: &=\pi\times\mbf I^{-1}\pi+p\times\mbf M^{-1}p-\gamma\frac{dw}{dt}\\
227: &=(\dot w+\gamma w)\times\mbf I^{-1}(\dot w+\gamma w)+
228: aw\times\mbf M^{-1}aw-\gamma w\times\mbf I^{-1}(\dot w+\gamma w)\\
229: &=\dot w\times\mbf I^{-1}(\dot w+\gamma w)+a^2w\times\mbf M^{-1}w.
230: \end{split}\end{equation}
231: Obviously this is smooth in $a$ for all $a\ge 0$, as required.
232: Dynamics on $a=0$ that is robust enough to continue through
233: perturbation to small positive $a$ will have implications for the
234: original system. By continuity in $a$, the blown-up vector field is a
235: lift by the smooth blow-down map of the vector field for the original
236: system, even through the nongeneric sector. Thus the flow on the
237: nongeneric sector corresponds through the blow-down map to the flow of
238: the original system on the union of the nongeneric leaves.
239:
240: Actually, the blow-up has a very transparent reformulation, since
241: $\hat P$ is diffeomorphic to $S^2\times\mbb R^3\times\mbb R_{\ge
242: 0}=\sset{(w,\pi,a)}$ by the map $\pi=\gamma w+\dot w$. Through this
243: diffeomorphism the blow-down map is simply $p=aw$, which is to say
244: that $w$ by itself is enough to desingularize the foliation by
245: symplectic leaves, but not enough to normalize the leaves. The
246: blow up map is a proper map since the map $(a,w)\mapsto aw$ is proper.
247:
248: Some exploration of the nongeneric sector of the blown-up space may be
249: helpful for visualization purposes. For fixed $\pi_0$ the equation
250: $\gamma w+\dot w=\pi_0$ has solution $w\in S^2$ and $\gamma=\pi_0\cdot
251: w$. Consequently the blow-up of the point $p=0$, $\pi=\pi_0$ is in all
252: cases a two sphere. This sphere intersects fixed $\gamma$ such that
253: $|\gamma|<|\pi_0|$ in a circle, $\gamma=\pm|\pi_0|$ in a point, and
254: $|\gamma|>|\pi_0|$ not at all. Thus departure from the point $p=0$,
255: $\pi=\pi_0$ along $\gamma>|\pi_0|$ is impossible. The nongeneric
256: symplectic symplectic leaf $|\pi|=r>0$ blows up to $\gamma^2+|\dot
257: w|^2=r^2$ which is diffeomorphic to $S^2\times S^2$, and which for
258: fixed $|\gamma|<r$ is a circle bundle and for $|\gamma|=r$ is a
259: sphere.
260:
261: The Hamiltonian pulls back through the smooth blow-down map to
262: \begin{equation*}\begin{split}
263: &\hat H\equiv\frac12\pi^t\mbf I^{-1}\pi+\frac12p^t\mbf M^{-1}p
264: =\hat H^0+a^2\hat H^1,\\
265: &\hat H^0\equiv=\frac12(\dot w+\gamma w)^t\mbf I^{-1}(\dot w +\gamma w),\quad
266: \hat H^1\equiv\frac{1}2w^t\mbf M^{-1}w,
267: \end{split}\end{equation*}
268: $\hat H$ is written this way in anticipation of perturbation arguments
269: from $a=0$ to small nonzero $a$. The symplectic form $\hat\omega$ on
270: the nongeneric sector can be calculated from the formula for the
271: coadjoint orbit symplectic forms of $\mtl{SE}(3)$
272: in~\ct{MarsdenJERatiuTS-1994.1}, with the result that
273: \begin{equation*}\begin{split}
274: &\hat\omega(w,\dot w)\bigl((\delta w_1,\delta\dot w_1),
275: (\delta w_2,\delta\dot w_2)\bigr)\\
276: &\qquad\mbox{}=-w\cdot(\delta w_1\times\delta\dot w_2
277: -\delta w_2\times\delta\dot w_1)
278: -\gamma w\cdot(\delta w_1\times\delta w_2).
279: \end{split}\end{equation*}
280: By continuity, the relation $i_{X_{\hat H}}\hat\omega=d\hat H$
281: persists from $a=0$ to $a>0$, so the vector field~\erf{8} is
282: Hamiltonian at $a=0$ with symplectic form $\hat\omega$ and Hamiltonian
283: $\hat H^0$. Thus the evolution on the nongeneric sector is
284: Hamiltonian, in a way that smoothly continues the Hamiltonian
285: structure of the generic sector.
286:
287: The dynamics on the invariant submanifold $p=0$ in the original space
288: $P$ admits the subcasimir $|\pi|$. This conserved quantity
289: (\emph{conserved on $p=0$ only}) pulls back to a conserved quantity
290: $|w+\gamma\dot w|$ for the \emph{nongeneric sector} of the blow-up
291: space $\hat P$. Since $w\cdot\dot w=0$ and $|w|=1$, this gives the
292: conserved quantity $|\dot w|^2$ and hence the conserved quantity
293: $f(|\dot w|)$, where $f$ is any function. The Hamiltonian
294: vector field of $f(|\dot w|^2)$ is
295: \begin{equation*}
296: \frac{dw}{dt}=-\frac{f^\prime(|\dot w|)}{|\dot w|}\dot w\times w,\quad
297: \frac{d\dot w}{dt}=-\gamma\frac{f^\prime(|\dot w|)}{|\dot w|}w\times\dot w.
298: \end{equation*}
299: Note that $\tilde m\equiv\dot w+\gamma w$ is conserved by these
300: equations, so that
301: \begin{equation*}
302: \frac{dw}{dt}=-\frac{f^\prime(|\dot w|)}{|\dot w|}\tilde m\times w,\quad
303: \frac{d\dot w}{dt}=-\gamma\frac{f^\prime(|\dot w|)}{|\dot w|}
304: \tilde m\times\dot w,
305: \end{equation*}
306: the solution of which is rotations about $\tilde m$. To normalize the
307: period at $2\pi$ and the righthand sense about $\tilde m$, choose
308: \begin{equation*}
309: \frac{f^\prime(|\dot w|)}{|\dot w|}|\tilde m|=
310: \frac{f^\prime(|\dot w|)}{|\dot w|}\sqrt{\gamma^2+|\dot w|^2}=-1,
311: \end{equation*}
312: which gives $f(|\dot w|)=-\sqrt{\gamma^2+|\dot w|^2}$. \emph{Thus the
313: nongeneric sector has an additional $\mtl{SO}(2)$ symmetry}, which acts
314: by
315: \begin{equation*}
316: \theta\cdot(w,\dot w)\equiv\bigl(\onm{exp}(m^\wedge\theta)w,
317: \onm{exp}(m^\wedge\theta)\dot w\bigr),
318: \quad m\equiv\frac{\dot w+\gamma w}{\sqrt{\gamma^2+|\dot w|^2}},
319: \end{equation*}
320: and has momentum
321: \begin{equation*}
322: \hat J\equiv-\sqrt{\gamma^2+|\dot w|^2}.
323: \end{equation*}
324: This extra $\mtl{SO}(2)$ symmetry arises from a subcasimir of the
325: original system. The action and the corresponding momentum are defined
326: on the nongeneric sector where $\gamma$ and $\dot w$ are not both
327: zero. The set where $a=\gamma=0$ and $\dot w=0$ exactly corresponds
328: through the blow-down/up to the set where $p=0$ and $\pi=0$, so that
329: the action and its momentum are defined only on an open subset does
330: not affect the analysis of the relative equilibria $p_e^{\alpha_e}$.
331:
332: Here are some aspects of the $\mtl{SO}(2)$ action and its relation
333: to the blow-down/up map.
334: \begin{enumerate}
335: \item
336: The action is free except on the set $\dot w=0$, which is a 2-sphere
337: of fixed points. This two sphere is also the level $-|\gamma|$ of the
338: momentum $\hat J$, is a symplectic submanifold of $\hat P$, and as
339: such is equal to its own singular reduction.
340: \item
341: The orbit relation of the action together with the parameter $\gamma$
342: exactly absorb the additional phase space from blowing up the
343: nongeneric leaves. Indeed, for fixed $\gamma$ the blow-down map is a
344: quotient map for the action, and the orbit space is therefore smooth,
345: irrespective of the fact that the action is not free.
346: \item
347: The blow-down map restricts to a quotient map for the (singular or
348: nonsingular) symplectic reduced space associated to the $\hat
349: J=\hat\mu$ level set. As such this reduced space is symplectomorphic
350: to the nongeneric leaf $p=0$, $|\pi|=-\hat\mu$.
351: \end{enumerate}
352: Only the verification of the third item in the nonsingular
353: ($\hat\mu<-|\gamma|$) case is troublesome. For that, it is easily
354: verified that the map $\pi=\gamma w+\dot w$ is a quotient map for the
355: $\mtl{SO}(2)$ action on $\hat J^{-1}(\hat\mu)$ which has image the
356: sphere $TS_{-\hat\mu}^2=\set{\pi}{|\pi|=-\hat\mu}$. To pull down the
357: symplectic form~$\hat\omega$ by that map, first let $(\pi,\delta\pi_i)\in
358: TS_{-\hat\mu}^2$, $i=1,2$, and seek $(w,\dot w,\delta w_i,\delta\dot
359: w_i)$ such that
360: \begin{gather*}
361: |w|=1,\quad w\cdot\dot w=0,\quad w\cdot\delta w_i=0,\quad
362: \delta w_i\cdot\dot w+w\cdot\delta\dot w_i=0,\\
363: -\sqrt{\gamma^2+|\dot w|^2}=\hat\mu,\quad \delta\dot w_i\cdot\dot w=0,\quad
364: \pi=\gamma w+\dot w,\quad \delta\pi_i=\gamma\delta w_i+\delta\dot w_i.
365: \end{gather*}
366: To solve these equations choose and $w$ such that
367: $w\cdot\dot\pi=\gamma$ and set $\dot w=\pi-\gamma w$. Expanding $\delta w_i$
368: and $\delta\dot w_i$ in the basis $w$, $\dot w$, $w\times\dot w$ gives
369: \begin{equation*}
370: \delta w_i=-\frac{w\cdot\delta\pi_i}{\hat\mu^2-\gamma^2}(\pi-\gamma w),\quad
371: \delta\dot w_i=(w\cdot\delta\pi_i)w
372: +\frac{(w\times\pi)\cdot\delta\pi_i}{\hat\mu^2-\gamma^2}w\times\pi.
373: \end{equation*}
374: Substitution into~$\hat\omega$ then gives
375: \begin{equation*}\begin{split}
376: &\hat\omega(w,\dot w)\bigl((\delta w_1,\delta\dot w_1),
377: (\delta w_2,\delta\dot w_2)\bigr)\\
378: &\qquad\mbox{}=\frac1{\hat\mu^2-\gamma^2}
379: \Bigl(\bigl(\delta\pi_1\cdot w\bigr)\bigl((\pi\times\delta\pi_2)\cdot w\bigr)
380: -\bigl(\delta\pi_2\cdot w\bigr)\bigl((\pi\times\delta\pi_1)\cdot w\bigr)\Bigr)
381: \\
382: &\qquad\mbox{}=\frac1{\hat\mu^2-\gamma^2}\bigl(w\times(w\times\pi)\bigr)\cdot
383: (\delta\pi_1\times\delta\pi_2)\\
384: &\qquad\mbox{}=\frac1{\hat\mu^2-\gamma^2}\bigl((w\cdot\pi)w-\pi\bigr)\cdot
385: (\delta\pi_1\times\delta\pi_2)\\
386: &\qquad\mbox{}=\frac1{\hat\mu^2-\gamma^2}\left((w\cdot\pi)
387: \frac{w\cdot\pi}{|\pi|^2}\pi
388: -\pi\right)\cdot
389: (\delta\pi_1\times\delta\pi_2)\\
390: &\qquad\mbox{}=\frac1{\hat\mu^2-\gamma^2}\left(\frac{\gamma^2}{\hat\mu^2}-1
391: \right)\pi\cdot(\delta\pi_1\times\delta\pi_2)\\
392: &\qquad\mbox{}=-\frac{1}{|\pi|^2}\,\pi\cdot(\delta\pi_1\times\delta\pi_2).
393: \end{split}\end{equation*}
394: This last expression is the symplectic form on the nongeneric leaf
395: $|\pi|=-\hat\mu$, as required.
396:
397: In short, the $\mtl{SO}(2)$ symmetry arises and exactly absorbs the
398: additional dimensions resulting from the blow-up construction. The
399: symplectic reductions of the nongeneric sector by this symmetry
400: exactly coincide with the original system restricted to the nongeneric
401: symplectic leaves of the phase space $P$.
402:
403: The pull-back of the relative equilibria~$p_e^{\alpha_e}$ by the
404: blow-down map is the set of $(w,\dot w,a,\gamma)$ such that
405: \begin{equation*}
406: p=aw=0,\quad \pi=\alpha_e\mbf k=\dot w+\gamma w.
407: \end{equation*}
408: Since $|w|=1$ the first equation is equivalent to $a=0$ (the relative
409: equilibria are of course in the nongeneric sector), and dotting the
410: second with $w$ shows it is equivalent to $\gamma=\alpha_e\mbf k\cdot
411: w$ and $\dot w=\alpha_e\mbf k-\gamma w$. Since $\gamma^2+|\dot
412: w|^2={\alpha_e}^2$ and $\alpha_e\ne0$, all these solutions are in are
413: within the open set where the $\mtl{SO}(2)$ action is defined (i.e.
414: where $\gamma$ and $\dot w$ are not both zero). Thus the relative
415: equilibria~$p_e^{\alpha_e}$ blow up to
416: \begin{equation}\elb{13}
417: \hat p_e^{\alpha_e}:\quad|w|=1,\quad \dot w=\alpha_e\mbf k
418: -\alpha_e(\mathbf k\cdot w)w,\quad a=0,\quad
419: \gamma=\alpha_e\mbf k\cdot w.
420: \end{equation}
421: Since $w$ is unconstrained in~\erf{13}, except for the first equation,
422: $\hat p_e^{\alpha_e}$ is diffeomorphic to $S^2$. Since $m=\mbf k$ on
423: $\hat p_e^{\alpha_e}$, the $\mtl{SO}(2)$ action on $\hat
424: p_e^{\alpha_e}$ is by rotation of the pair $(w,\dot w)$ about $\mbf
425: k$. By substitution of $\hat p_e^{\alpha_e}$ into~\erf{8}, each point
426: of $\hat p_e^{\alpha_e}$ is a relative equilibrium for the
427: $\mtl{SO}(2)$ symmetry, except for the the two points $w=\pm\mbf k,
428: \dot w=0,a=0,\gamma=\pm\alpha_e$, which are equilibria that reside at
429: singular points of the action. Each relative equilibrium in $\hat
430: p_e^{\alpha_e}$ has the same generator, namely
431: \begin{equation}\elb{19}
432: \hat\xi_e^{\alpha_e}\equiv-\frac{\alpha_e}{I_3}
433: \end{equation}
434: and by substitution into $\hat J$, the same
435: momentum, namely
436: \begin{equation*}
437: \hat\mu_e^{\alpha_e}\equiv-\alpha_e.
438: \end{equation*}
439:
440: Fixing $\gamma$, which means fixing a parameter,
441: $\alpha_e=\gamma/\cos\phi$, where $\phi$ is the angle between $\mbf k$
442: and $w$. Thus for fixed $\gamma$ there are 2-submanifolds of relative
443: equilibria, as $\alpha_e$ is varied, as expected for an $\mtl{SO}(2)$
444: symmetric Hamiltonian system. Along those submanifolds there is the
445: \emph{momentum-generator} relation
446: \begin{equation}\elb{22}
447: \hat\xi_e^{\alpha_e}=\frac1{I_3}\hat\mu_e^{\alpha_e},
448: \end{equation}
449: which will give a crucial component in the KAM twist condition to
450: follow.
451:
452: \section{Normal forms in the blown-up system}
453: %%
454: Proposition~\thrf{StabProp} below relates the stability of the
455: relative equilibrium~$p_e^{\alpha_e}$ to the stability of its blow-up
456: $\hat p_e^{\alpha_e}$.
457:
458: \begin{proposition}\lb{StabProp}
459: Suppose that, for some fixed $\alpha_e$, $\hat p_e^{\alpha_e}$ is
460: stable for the flow $\hat F_t$ on $\hat P$, in the sense that for all
461: neighborhoods $\hat U$ of $\hat p_e^{\alpha_e}$ there is a
462: neighborhood $\hat V$ of $\hat p_e^{\alpha_e}$ such that $\hat
463: F_t(\hat p)\in\hat U$ for all $\hat p\in\hat V$. Then $p_e^{\alpha_e}$
464: is a stable relative equilibrium.
465: \end{proposition}
466:
467: \begin{proof}
468: Suppose $U$ is a neighborhood of $p_e^{\alpha_e}$. $U$ pulls back by
469: the blow-down map to an open neighborhood $\hat U$ of $\hat
470: p_e^{\alpha_e}$. Let $\hat V$ be a neighborhood as in the statement of
471: the proposition. Then it suffices to show that $\hat V$ pushes forward
472: by the blow-down map to a neighborhood of $p_e^{\alpha_e}$. But this
473: follows since the blow-down map is proper.
474: \end{proof}
475:
476: In particular, if all of the relative equilibria and both equilibria
477: in $\hat p_e^{\alpha_e}$ are stable under perturbation
478: both within the phase space $TS^2$ and in the parameters $a$ and
479: $\gamma$, then the original relative equilibrium~$p_e^{\alpha_e}$ is
480: stable. When $a$ is perturbed away from $0$ this is a $\mtl{SO}(2)$
481: symmetry breaking perturbation. As $\onm{dim}TS^2=4$, the blown-up
482: system is integrable when $a=0$ and hence the stability issue is one
483: of the stability of periodic orbits of a near integrable Hamiltonian
484: system.
485:
486: Assume, without loss of generality, that $\alpha_e>0$. Since the
487: $\mtl{SO}(2)$ symmetry on $\hat p_e^{\alpha_e}$ is by rotation about
488: $\mbf k$, it suffices to consider the stability of orbits in $\hat
489: p_e^{\alpha_e}$ emanating from points $\hat p_e^{\alpha_e,\theta}$
490: obtained by substituting $w=\sin\theta\mbf i+\cos\theta\mbf k$
491: into~\erf{13}, for $\theta\in[0,\pi]$.
492:
493: \subsection{Normal form for the relative equilibria}\lb{200}
494: %%
495: Consider first the \emph{relative equilibria} in $\hat
496: p_e^{\alpha_e,\theta}$; i.e., exclude the \emph{equilibria}
497: corresponding to $\theta=0$ and $\theta=\pi$. The argument proceeds
498: by adapting and incrementally refining, to the order required for the
499: stability analysis, the normal form near relative equilibria developed
500: in~\ct{PatrickGW-1995.1}.
501:
502: Below $O(x;y)^k$, $x\in\mbb R^n$, $y\in\mbb R^m$ will denote the set
503: of smooth $y$ dependent functions such that $O(x;y)/|x|^k$ is bounded
504: near $0$. The product $O(x;y)^kO(x^\prime;y^\prime)^{k^\prime}$ denotes
505: the set of finite sums of products of elements of $O(x;y)^k$ and
506: $O(x^\prime;y^\prime)^{k^\prime}$.
507:
508: \subsubsection{Initial normal form}\lb{201}
509: %%
510: This is constructed from the linearization of the relative
511: equilibrium, which means the linearization at $\hat p_e^{\alpha_e,\theta}$
512: of Hamiltonian vector field $X_{\hat H_{\xi^{\alpha_e}_e}}$ where
513: \begin{equation*}
514: \hat H_{\xi^{\alpha_e}_e}^0\equiv\hat H^0-\hat\xi_e\hat J.
515: \end{equation*}
516: The characteristic polynomial of the linearization is $x\mapsto
517: x^2(x^2+{\omega_e}^2)$, where
518: \begin{equation}\elb{23}
519: \omega_e\equiv\pm{\alpha_e}\sqrt{\left(\frac1{I_3}-\frac1{I_1}\right)
520: \left(\frac1{I_3}-\frac1{I_2}\right)},
521: \end{equation}
522: For later convenience define $\omega_e$ positive if $I_3>I_1$ and
523: $I_3>I_2$ and negative if $I_3<I_1$ and $I_3<I_2$. The linearization
524: has a $0$ and $\pm i\omega_e$~generalized eigenspaces, both of
525: dimension~$2$. Introducing the parameter
526: \begin{equation*}
527: D\equiv\frac{I_2(I_3-I_1)}{I_1(I_3-I_2)},
528: \end{equation*}
529: the vectors
530: \begin{equation*}\begin{split}
531: &v_1\equiv\frac {D^{\frac14}}{\sqrt{\alpha_e}}\left[\begin{array}{cccccc}0&\cos\theta&0&0&\alpha_e\sin^2\theta&0
532: \end{array}\right]\\
533: &v_2\equiv\frac {D^{-\frac14}}{\sqrt{\alpha_e}}\left[
534: \begin{array}{cccccc}\cos\theta&0&-\sin\theta&\alpha_e\sin^2\theta&0&
535: \alpha_e\sin\theta\cos\theta\end{array}\right]\\
536: &v_3\equiv\sin\theta\left[
537: \begin{array}{cccccc}0&1&0&0&-\alpha_e\cos\theta&0
538: \end{array}\right]\\
539: &v_4\equiv\frac1{\alpha_e\sin\theta}\left[
540: \begin{array}{cccccc}-\cos^2\theta&0&\cos\theta\sin\theta
541: &\alpha_e\cos^3\theta&0&-\alpha_e\sin\theta(1+\cos^2\theta)\end{array}\right]
542: \end{split}\end{equation*}
543: form a basis of $T_{\hat p_e^{\alpha_e,\theta}}S^2$ which satisfies
544: the following:
545: \begin{enumerate}
546: \item the basis is symplectically canonical, so that the symplectic form
547: with respect to it is
548: \begin{equation*}
549: \omega(\hat p_e^{\alpha_e,\theta})=\left[
550: \begin{array}{cccc}0&1&0&0\\-1&0&0&0\\0&0&0&1\\0&0&-1&0\end{array}\right];
551: \end{equation*}
552: \item with respect to the basis the derivative of the momentum is
553: \begin{equation*}
554: d\hat J\bigl(\hat p_e^{\alpha_e,\theta}\bigr)=
555: \left[\begin{array}{cccc}0&0&0&1\end{array}\right];
556: \end{equation*}
557: \item the third basis vector~$v_3$ is the infinitesimal generator
558: action corresponding to~$1\in\mtl{so}(2)$;
559: \item the first two vectors $v_1$, $v_2$
560: span the~$\omega_e$ generalized eigenspace and the last two $v_3$, $v_4$ span
561: the $0$~generalized eigenspace of the linearization;
562: \item the linearization of the relative equilibrium is
563: \begin{equation}\elb{27}
564: dX_{\hat H_{\xi^{\alpha_e}_e}^0}(\hat p_e^{\alpha_e,\theta})
565: =\left[\begin{array}{cccc}
566: 0&\omega_e&0&0\\-\omega_e&0&0&0\\0&0&0&\kappa_e\\0&0&0&0\end{array}\right],
567: \end{equation}
568: where
569: \begin{equation}\elb{25}
570: \kappa_e\equiv 1/I_3.
571: \end{equation}
572: \end{enumerate}
573: Consequently, the basis effects a Witt-Moncrief decomposition
574: \begin{equation*}
575: T_{\hat p_e^{\alpha_e,\theta}}\hat P=N_1\oplus\mtl{so}(2)
576: \oplus\mtl{so}(2)^*
577: \equiv\onm{span}(v_1,v_2)\oplus\mbb Rv_3\oplus\mbb Rv_4.
578: \end{equation*}
579: Here $N_1$, the \emph{symplectic normal}, may be
580: identified with the tangent space to the symplectic reduced space at
581: $\hat p_e^{\alpha_e,\theta}$ for the $\mtl{SO}(2)$ action. The
582: appearance of the nilpotent part of the linearization is the
583: foundational element of~\ct{PatrickGW-1995.1}. The value of
584: $\kappa_e$ coincides with the derivative
585: $d\hat\xi_e^{\alpha_e}/d\hat\mu_e^{\alpha_e}$
586: from~\erf{22}, as predicted by the general theory.
587:
588: The initial normal form can now transcribed from the data above, and is
589: \begin{equation}\begin{split}\elb{21}
590: &\hat H=\frac{\omega_e}2(q^2+p^2)+\hat\xi_e^{\alpha_e}\nu
591: +\frac12\kappa_e\nu^2+R(q,p,\nu)
592: +\frac{a^2}{2}\hat H^1(q,p,\varphi,\nu),\\
593: &R=O(q,p,\nu)^3
594: \end{split}
595: \end{equation}
596: on the product of $\mbb R^2\times
597: T^*\mtl{SO}(2)=\sset{(q,p),(\varphi,\nu)}$ with the product symplectic
598: form $dq\wedge dp+d\varphi\wedge d\nu$, with $\mtl{SO}(2)$ acting by
599: lifts of its left action on itself, and with the momentum map
600: $\nu-\alpha_e$. The transcription is that there is an
601: $\mtl{SO}(2)$~intertwining symplectic diffeomorphism from a
602: neighborhood of the $\mtl{SO}(2)$ orbit of $\hat p_e^{\alpha_e}$
603: to a neighborhood of $0$ times the zero section of $T^*\mtl{SO}(2)$
604: which
605: \begin{enumerate}
606: \item sends the relative equilibrium $\hat p_e^{\alpha_e,\theta}$ to
607: $p=q=\nu=\varphi=0$.
608: \item has derivative at $\hat p_e^{\alpha_e}$ the identity map with
609: respect to the basis~$v_i$ and the standard basis of $\mbb R^2\times
610: T^*\mtl{SO}(2)$;
611: \item intertwines the momentum maps $\hat J$ and $\nu-\alpha_e$;
612: \end{enumerate}
613: Thus the transription is \emph{strucure preserving} in that it is
614: symplectic and it preserves the $\mtl{SO}(2)$ symmetry and momentum,
615: so the blown-up system near the group orbit of the relative
616: equilibrium $\hat p_e^{\alpha_e,\theta}$ can be replaced by the
617: entirely equivalent system~\erf{21} near $q=p=\nu=0$.
618:
619: \subsubsection{Elimination of $qO(\nu)^2$, $pO(\nu)^2$, and $(q^2-p^2)\nu$}\lb{202}
620: %%
621: The remainder term of~\erf{21} can be expanded as
622: \begin{equation*}\begin{split}
623: R=&c_1(\nu)q+c_2(\nu)p+c_3\nu(q^2-p^2)+c_4\nu(q^2+p^2)\\
624: &\qquad\qquad+O(q,p)^3+O(\nu)^3+O(q,p)^2O(q,p,\nu)^2,
625: \end{split}\end{equation*}
626: where $c_1(\nu)=O(\nu)^2$,$c_2(\nu)=O(\nu)^2$, and $c_3$, $c_4$ are
627: constants. The transformation
628: \begin{equation*}
629: \tilde q=q+\frac{c_1}{\omega_e},\quad
630: \tilde\varphi=\varphi+\frac{p\nu}{\omega_e}\,\frac{dc_1}{d\nu},
631: \end{equation*}
632: suggested by completing the square in $\frac12\omega_eq^2+c_1q\nu^2$,
633: is structure preserving and changes the Hamiltonian to the same form
634: but without terms of the form $qO(\nu)^2$. Similarly one eliminates
635: $pO(\nu)^2$. The transformation
636: \begin{equation}\elb{64}
637: \tilde q=\frac q{f(\nu)},\quad\tilde p=f(\nu)p,
638: \quad f(\nu)=\left(
639: \frac{1-\frac{2c_3}{\omega_e}\nu}{1+\frac{2c_3}{\omega_e}\nu}
640: \right)^{\frac14},
641: \end{equation}
642: takes the fragment $\frac{\omega_e}2(q^2+p^2)+c_3(q^2-p^2)\nu$ to
643: \begin{equation*}
644: \left(\frac{\omega_e}2+c_3\right)q^2+\left(\frac{\omega_e}2-c_3\right)p^2
645: =\frac{\omega_e}2(\tilde q^2+\tilde p^2)+\tilde q^2O(\nu)^2+\tilde p^2O(\nu)^2,
646: \end{equation*}
647: while the symplectic form becomes
648: \begin{equation*}\begin{split}
649: dq\wedge dp+d\varphi\wedge d\nu=&d\tilde q\wedge d p+d\varphi\wedge d\nu+
650: \frac{f^\prime}{f}(qdp+pdq)\wedge d\nu\\
651: =&d\tilde q\wedge d p+d\left(\varphi+\frac{f^\prime}{f}qp\right)\wedge d\nu.
652: \end{split}\end{equation*}
653: Adjoining $\tilde\varphi=\varphi+(f^\prime/f)qp$ to~\erf{64} gives a
654: structure preserving symplectic transformation that
655: eliminates the term $c_3(q^2-p^2)$. Thus, without loss of generality,
656: \begin{equation}\elb{67}
657: R=c_4(q^2+p^2)+O(q,p)^3+O(\nu)^3+O(q,p)^2O(q,p,\nu)^2.
658: \end{equation}
659:
660: \subsubsection{Normal form for the rigid body}\lb{203}
661: %%
662: We will require the first two terms of the normal form corresponding
663: to the equilibrium $\pi=\alpha_e\mbf k$ of the blown-up system reduced
664: by its $\mtl{SO}(2)$ symmetry, i.e. the symplectic reduced spaces of
665: the rigid body $\frac12\pi\mbf I^{-1}\pi$. The map
666: \begin{equation*}
667: \pi=\Bigl(\bigl(\alpha_e-{\textstyle\frac14}(Q^2+P^2)\bigr)^{\frac12}P,
668: \bigl(\alpha_e-{\textstyle\frac14}(Q^2+P^2)\bigr)^{\frac12}Q,\alpha_e
669: {\textstyle-\frac12}(Q^2+P^2)\Bigr)
670: \end{equation*}
671: is a symplectic chart on the reduced space $|\pi|=\alpha_e$ and in
672: these coordinates, the Hamiltonian becomes, up to a constant,
673: \begin{equation*}
674: \frac12\pi^t\mbf I^{-1}\pi=\frac12\bigl(\alpha_e
675: -{\textstyle\frac14}(Q^2+P^2)\bigr)
676: \left(\left(\frac1{I_1}-\frac{1}{I_3}\right)P^2+
677: \left(\frac1{I_2}-\frac{1}{I_3}\right)Q^2\right).
678: \end{equation*}
679: Action-angle variables for the linearized flow are
680: \begin{equation*}
681: Q=\sqrt{2I}D^{\frac14}\sin\psi,\qquad P=\sqrt{2I}D^{-\frac14}\cos\psi,
682: \end{equation*}
683: and the Hamiltonian is then
684: \begin{equation*}
685: \frac12\pi^t\mbf I^{-1}\pi=\omega_eI-\frac{\omega_e}{2\alpha_e}(D^{\frac12}\sin^2\psi
686: +D^{-\frac12}\cos^2\psi)I^2.
687: \end{equation*}
688: By averaging over $\psi$,
689: \begin{equation}\elb{14}
690: \frac12\pi^t\mbf{I}\pi=\omega_eI+\frac12\upsilon_eI^2+O(Q,P)^5
691: \end{equation}
692: where
693: \begin{equation}\elb{24}
694: \upsilon_e=-\frac{\omega_e}{2\alpha_e}(D^{\frac12}+D^{-\frac12})
695: =\frac12\left(\frac2{I_3}-\frac1{I_1}-\frac1{I_2}\right).
696: \end{equation}
697:
698: \subsubsection{Matching and normalizing the reduced spaces at
699: $\hat p_e^{\alpha_e,\theta}$}\lb{204}
700: %%
701: For $a=0$ the symplectic reduced space through $\hat
702: p_e^{\alpha_e,\theta}$ of the blown-up system is the $|\pi|=\alpha_e$
703: symplectic reduced space of the rigid body $\frac12\pi^t\mbf I\pi$.
704: For $a=0$ the symplectic reduced space of the normal form~\erf{21}
705: through $q=p=\varphi=\nu=0$ is $\mbb R^2$ with symplectic form
706: $dq\wedge dp$ and Hamiltonian $\hat H|_{\nu=0}$. Since the
707: intertwining map between the blown-up system and the normal form is
708: structure preserving, it descends to symplectomorphisms of reductions
709: of the these two systems. Consequently, by symplectomorphism on
710: $(q,p)$ only, the normal form Hamiltonian~\erf{21} at $q=p=\nu=0$ can
711: be equated to the rigid body normal form~\erf{14}, after which the
712: normal is correct to fourth order in pure $q$ and $p$ and
713: \begin{equation*}
714: R=O(q,p)^5+\nu O(q,p)^2O(q,p,\nu).
715: \end{equation*}
716:
717: \subsubsection{Refinement by matching the normal forms and
718: generators along the relative equilibria near $\hat
719: p_e^{\alpha_e,\theta}$}\lb{205}
720: %%
721: Advantage may be obtained by comparing reduced normal along the
722: relative equilibria $\hat p_e^{\alpha_e+z,\theta}$ as $z$
723: varies. These relative equilibria occur (for both systems) at momentum
724: $-(\alpha_e+z)$. For the rigid body the only $z$ dependent adjustment
725: is in the $\alpha_e$ dependence of the linearized frequency, which
726: becomes $\omega_e(\alpha_e+z)/\alpha_e$, so the normal form is
727: \begin{equation}\elb{72}
728: \left(\omega_e+\frac{\omega_e}{\alpha_e}z\right)I+\frac12\upsilon_eI^2
729: +O(q,p;z)^5.
730: \end{equation}
731: For~\erf{21} it is the normal form of the reduction at $\nu=-z$, so it
732: is the normal form of the Hamiltonian
733: \begin{equation*}
734: \omega_eI+\frac12\upsilon_eI^2-2c_4Iz+O(q,p)^2O(q,p,z)^2
735: \end{equation*}
736: which is
737: \begin{equation}\elb{73}
738: \bigl(\omega_e-2c_4z+O(z)^2\bigr)I+O(q,p;z)^3.
739: \end{equation}
740: Comparison of~\erf{72} and~\erf{73} at first order in $I$ gives a
741: crucial fact:
742: \begin{equation*}
743: c_4=-\frac{\omega_e}{2\alpha_e}.
744: \end{equation*}
745:
746: Also, the $SO(2)$ generator of the blown-up system at $\hat
747: p_e^{\alpha_e+z,\theta}$, which is $-(\alpha_e+z)/I_3$, and the
748: $\mtl{SO}(2)$ generator of system~\erf{21} at the relative equilibrium
749: $q=p=0$ are the same. Equating these gives
750: \begin{equation*}
751: \hat\xi_e^{\alpha_e}+\kappa_e\nu+\left.\frac{\partial R}{\partial\nu}\right|_
752: {\!\!\!\renewcommand{\arraystretch}{.2}\begin{array}[c]{l}\scriptstyle q=p=0\\\scriptstyle\nu=-z\end{array}}\!\!\!=-\frac1{I_3}(\alpha_e+z)=\hat\xi_e^{\alpha_e}-\kappa_ez,
753: \end{equation*}
754: which means that $R$ has no pure $\nu$ terms. Particularly, the
755: $O(\nu)^3$ term in~\erf{67} is zero.
756:
757: \subsubsection{Symmetry breaking term}\lb{206}
758: %%
759: The transcription to the initial normal form is known to first order since it
760: has derivative the identity map along the $\mtl{SO}(2)$ orbit of $\hat
761: p_e^{\alpha_e,\theta}$. Consequently, $\hat H^1$ can be calculated to
762: first order by substitution of
763: \begin{equation*}
764: w=\exp(\varphi\mbf k)P_w\bigl(\hat p_e^{\alpha_e,\theta}+(qv_1+pv_2
765: +\nu v_4)\bigr)
766: \end{equation*}
767: into $w^t\mbf M^{-1}w$, where $P_w(w,\dot w)=w$.
768:
769: \subsubsection{Altogether}\lb{207}
770: %%
771: Putting all the foregoing together, the normal form is
772: \begin{equation}\begin{split}\elb{69}
773: \hat H=&\omega_eI+\frac12\upsilon_eI^2+\hat\xi_e^{\alpha_e}\nu
774: +\frac12\kappa_e\nu^2-\frac{\omega_e}{\alpha_e}I\nu\\
775: &\quad\mbox{}+O(q,p)^5+\nu O(q,p)^2O(q,p,\nu)\\
776: &\quad\mbox{}+a^2\hat H^{1,0}(q,p,\varphi,\nu)+a^2\hat H^{1,1}(q,p,\varphi,\nu)
777: +a^2O(q,p,\nu;\varphi)^2,
778: \end{split}\end{equation}
779: where
780: \begin{equation}
781: \hat H^{1,0}\equiv\frac{M_2-M_1}{2M_1M_2}\sin^2\theta\cos^2\varphi
782: \end{equation}
783: and
784: \begin{equation*}\begin{split}
785: \hat H^{1,1}&\equiv\frac{(M_2-M_1)\sin2\theta}{4M_1M_2\sqrt{\alpha_e}}
786: \left(-D^{\frac14}q\sin2\varphi+D^{-\frac14}p\cos2\varphi\right)\\
787: &\qquad\mbox{}-\frac{\cos^2\theta}{\alpha_e}\left(\frac1{M_1}\cos^2\varphi
788: +\frac1{M_2}\sin^2\varphi-\frac1{M_3}\right)\nu\\
789: &\qquad\mbox{}-\frac{\sin2\theta}{4D^{\frac14}\sqrt\alpha_e}
790: \left(\frac2{M_3}-\frac1{M_1}-\frac1{M_2}\right)p.
791: \end{split}\end{equation*}
792: The details of the symmetry breaking term $\hat H^1$ are not required
793: for the stability analysis and are displayed here for the sake of
794: completeness. The functional form of $\hat H^{1,1}$ depends on the
795: choice of the basis $v_i$ and further normalization or analysis would
796: be required to extract information from it.
797:
798: \subsection{Normal form for the equilibria}\lb{300}
799: %%
800: There remains to consider the two equilibria $\hat p_e^{\alpha_e,0}$ and
801: $\hat p_e^{\alpha_e,\pi}$ corresponding to $w=\mbf k$ and $w=-\mbf k$,
802: respectively. These equilibria are fixed points of the action of
803: $\mtl{SO}(2)$ and the analysis requires a transparent extension of the
804: normal form in~\ct{PatrickGW-1995.1} to \emph{equilibria} which have
805: $\mtl{SO}(2)$ isotropy.
806:
807: It suffices to consider $\hat p_e^{\alpha_e,0}$; the case of $\hat
808: p_e^{\alpha_e,\pi}$ is similar. There is a one parameter family of
809: possible linearizations of the equilibrium, namely the linearizations
810: at $\hat p_e^{\alpha_e,0}$ of the Hamiltonian vector fields $X_{\hat
811: H_\lambda^0}$ where $\hat H_\lambda^0-\lambda\hat J$. These
812: linearizations have characteristic polynomials
813: \begin{equation*}
814: x\mapsto\bigl(x^2+(\alpha_e+\lambda I_3)^2\bigr)\bigl(x^2+{\omega_e}^2\bigr).
815: \end{equation*}
816: Choosing $\lambda=-\alpha_e/I_3$ gives the largest possible null space
817: and therefore the largest number of intrinsically defined higher order
818: terms. The vectors
819: \begin{equation*}\begin{split}
820: &v_{1,0}\equiv\frac{D^{\frac14}}{\sqrt{\alpha_e}}
821: \left[\begin{array}{cccccc}0&1&0&0&&0\end{array}\right],\quad
822: v_{2,0}\equiv\frac {D^{-\frac14}}{\sqrt{\alpha_e}}\left[
823: \begin{array}{cccccc}1&0&0&0&0&0\end{array}\right]\\
824: &v_{3,0}\equiv\frac 1{\sqrt{\alpha_e}}\left[
825: \begin{array}{cccccc}0&1&0&0&-\alpha_e&0\end{array}\right]\quad
826: v_{4,0}\equiv\frac 1{\sqrt{\alpha_e}}\left[
827: \begin{array}{cccccc}-1&0&0&\alpha_e&0&0
828: \end{array}\right],
829: \end{split}\end{equation*}
830: form a basis of $T_{\hat p_e^{\alpha_e,0}}S^2$ which is symplectically
831: canonical and such that the linearization
832: $dX_{H_\lambda^0}(\hat p_e^{\alpha_e,0})$ is the
833: linearization~\erf{27} with $\kappa_e$ replaced by zero. The first
834: two vectors span the tangent space to the (singular) reduced space
835: through $p_e^{\alpha_e,0}$, which is the $|\pi|=\alpha_e$ symplectic
836: reduced space of the rigid body $\frac12\pi^t\mbf I^{-1}\pi$.
837:
838: Higher order terms of the Taylor expansion of $X_{H_\lambda^0}$ are
839: intrinsically polynomials on $\onm{ker}dX_{H_\lambda^0}(\hat
840: p_e^{\alpha_e,0})$. Alternately one can compute the higher order
841: terms of the Taylor expansion of the Hamiltonian on the null space.
842: Letting $(x,y)$ be the coordinates on $\onm{ker}dX_{H_\lambda^0}(\hat
843: p_e^{\alpha_e,0})$ indicated by the last of the two basis vectors
844: above, the Hamiltonian on the null space is easily computed to be
845: \begin{equation*}
846: \frac{1}{8I_3}(x^2+y^2)^2+O(x,y)^5=\frac{\kappa_e}8(x^2+y^2)^2+O(x,y)^5.
847: \end{equation*}
848: The initial normal form, obtained by the Equivariant Darboux Theorem,
849: is the Hamiltonian
850: \begin{equation*}\begin{split}
851: \hat H=&\frac{\omega_e}2(q^2+p^2)
852: -\frac{\alpha_e}{2I_3}(x^2+y^2)+\frac{\kappa_e}{8}(x^2+y^2)^2\\
853: &\qquad\mbox{}+O(q,p)O(q,p,x,y)^2+O(x,y)^5+a^2\hat H^1(q,p,x,y)
854: \end{split}\end{equation*}
855: on the phase space $\mbb R^2\times\mbb R^2=\sset{(q,p),(x,y)}$, with
856: symplectic form $dq\wedge dp+dx\wedge dy$, with $\mtl{SO}(2)$ acting
857: by counterclockwise rotation on $(x,y)$, and with the momentum mapping
858: $-\frac12(x^2+y^2)$. The transcription is by local symplectic
859: diffeomorphism with analogous properties to those stated in
860: Section~\thrf{201}.
861:
862: Manipulations similar to those in Section~\thrf{200} are required, as follows:
863: \begin{enumerate}
864: \item Linear terms in $q$ of the form $qO(x,y)^2$ can be removed as in
865: Section~\thrf{202}, and similarly linear terms in $p$, $x$, and $y$.
866: Quartic terms of the form $O(q,p)^2O(x,y)^2$ must by $\mtl{SO}(2)$ invariance
867: be in $(x^2+y^2)O(q,p)$, and so can be written as sums of
868: $(q^2+p^2)(x^2+y^2)$ and $(q^2-p^2)(x^2+y^2)$, and the latter kind removed,
869: as in Section~\thrf{202}.
870: \item Pure $q$ and $p$ terms up to order~4 can be found by matching the
871: reduced system of the initial normal form to rigid body reduced spaces.
872: \item By Items~1 and~2 all terms up to and including order 4 are
873: removed or calculated, except for the coefficient of the term
874: $(q^2+p^2)(x^2+y^2)$. This can be found by matching normal forms along
875: the equilibria $q=p=0$ (which are fixed points of the action of
876: $\mtl{SO}(2)$), and the resulting term is
877: $-\frac{\omega_e}{4\alpha_e}(q^2+p^2)(x^2+y^2)$.
878: \item The remainder after all of that, having no terms linear in any
879: variable, and being at least degree $5$, is of the form
880: $O(q,p,x,y)^2O(q,p,x,y)^3$, and is $\mtl{SO}(2)$ invariant.
881: \item The symmetry breaking term $\hat H^1$ can be calculated as in
882: Section~\thrf{206} by substituting
883: \begin{equation*}
884: w=P_{w}(\hat p_e^{\alpha_e,0}+qv_{1,0}+pv_{2,0}+xv_{3,0}+yv_{4,0})
885: \end{equation*}
886: into $w^t\mbf M^{-1}w$, and keeping the leading terms, which are order~2.
887: \end{enumerate}
888: Altogether, the normal form is
889: \begin{equation}\begin{split}\elb{325}
890: \hat H=&\omega_eI+\frac12\upsilon_eI^2
891: +\xi_e^{\alpha_e}\nu+\frac12\kappa_e\nu^2-\frac{\omega_e}{\alpha_e}I\nu
892: +O(q,p,x,y)^2O(q,p,x,y)^3\\
893: &\qquad\qquad\mbox{}+\frac{a^2}2\hat H^{1,1}(q,p,x,y)+a^2O(q,p,x,y)^3
894: \end{split}\end{equation}
895: where $I=\frac12(q^2+p^2)$ and $\nu=\frac12(x^2+y^2)$, and
896: \begin{equation*}
897: \hat H^{1,1}\equiv\frac{(M_3-M_1)(D^{-\frac14}p-y)^2}{\alpha_eM_1M_3}+
898: \frac{(M_3-M_2)(D^{\frac14}q+x)^2}{\alpha_eM_2M_3}.
899: \end{equation*}
900:
901: \section{Stability}
902: %%
903: The rescaling
904: \begin{equation*}
905: I=a^{2c}\tilde I,\quad\nu=a^{2c}\tilde\nu
906: \end{equation*}
907: is symplectic with multiplier $a^{2c}$. Substituting into~\erf{69} and
908: dropping the tildes gives
909: \begin{equation*}\begin{split}
910: \frac1{a^{2c}}\hat H&=\omega_eI+\frac12\upsilon_eI^2a^{2c}
911: +\hat\xi_e^{\alpha_e}\nu+\frac12\kappa_e\nu^2a^{2c}-\omega_eI\nu a^{2c}
912: +O(a;q,p,\varphi,\nu)^{3c}\\
913: &\qquad\mbox{}+a^{2-2c}\hat H^{1,0}+O(a;q,p,\varphi,\nu)^{2-c}
914: \end{split}\end{equation*}
915: Matching the exponents of $a$ in first nontrivial terms of the
916: integrable part, i.e. $\frac12\kappa_e\nu^2a^{2c}$ and
917: $\frac12\upsilon_eI^2a^{2c}$, with the first term of the nonintegrable
918: part, gives $2c=2-2c$, or $c=\frac12$. After putting $\epsilon=\sqrt
919: a$, and disposing the factor $1/a^{2c}$ of $\hat H$, which merely
920: reparameterizes time, one has
921: \begin{equation}\elb{15}
922: \hat H=\omega_eI+\hat\xi_e^{\alpha_e}\nu+\left(\frac12\kappa_e\nu^2
923: -\frac{\omega_e}{\alpha_e}I\nu+\frac12\upsilon_eI^2+\hat H^{1,0}\right)\epsilon^2+O(\epsilon)^3,
924: \end{equation}
925: where the dependence of $O(\epsilon)^3$ on all of $q$, $p$, $\varphi$,
926: and $\nu$ has been notationally suppressed. For $\epsilon=0$ the
927: Hamiltonian~\erf{15} has a periodic orbit cylinder by varying
928: $\varphi$ and $\nu$ with $I=0$. The orbit $I=\nu=0$ corresponds to the
929: relative equilibrium $\hat p_e^{\alpha_e,\theta}$.
930:
931: For determining stability it suffices to approximate the Poincar\'e
932: map for the orbit corresponding to $\nu=0$ in the zero energy
933: level. Solving~\erf{15} for $\nu$ when $H=0$ gives
934: \begin{equation*}
935: \nu=-\frac{\omega_eI}{\hat\xi_e^{\alpha_e}}+O(\epsilon)^2
936: \end{equation*}
937: and the equations of motion for~\erf{15} are
938: \begin{equation}\elb{16}\begin{split}
939: &\frac{d\psi}{dt}=\omega_e+\left(\upsilon_eI-\frac{\omega_e}{\alpha_e}\nu\right)\epsilon^2
940: +O(\epsilon)^3,
941: \quad\frac{dI}{dt}=O(\epsilon)^3,\\
942: &\frac{d\varphi}{dt}=\hat\xi_e^{\alpha_e}+\left(\kappa_e\nu-\frac{\omega_e}{\alpha_e}I\right)\epsilon^2
943: +O(\epsilon)^3,
944: \quad\frac{d\nu}{dt}=-\frac{\partial\hat H^{1,0}}{\partial\varphi}\epsilon^2
945: +O(\epsilon)^3.
946: \end{split}\end{equation}
947: On the zero energy level the equation for the evolution of $\varphi$ is
948: \begin{equation}\elb{17}
949: \frac{d\varphi}{dt}=\hat\xi_e^{\alpha_e}
950: -\frac{\omega_e(\kappa_e\alpha_e+\hat\xi_e^{\alpha_e})}{\alpha_e\hat\xi_e^{\alpha_e}}
951: I\epsilon^2+O(\epsilon)^3.
952: \end{equation}
953: With an initial condition $I=I_0$, the second equation of~\erf{16}
954: gives $I=I_0+O(\epsilon)^3$, so the return time of the Poincar\'e map
955: is, from~\erf{17},
956: \begin{equation*}
957: T\equiv-\frac{2\pi}{\hat\xi_e^{\alpha_e}}\left(1+\frac{(\alpha_e\kappa_e
958: +\hat\xi_e^{\alpha_e})\omega_eI_0}{\alpha_e(\hat\xi_e^{\alpha_e})^2}
959: \epsilon^2\right)+O(\epsilon)^3
960: \end{equation*}
961: Solving the first two equations of~\erf{16} over this period,
962: dropping the subscript~0 for the initial conditions, and using
963: the relation $\hat\xi_e^{\alpha_e}=-\kappa_e\alpha_e$ to eliminate $\alpha_e$
964: gives
965: \begin{equation}\begin{split}\elb{18}
966: &I^\prime=I+O(\epsilon)^3\\
967: &\psi^\prime=
968: \psi-\frac{2\pi\omega_e}{\hat\xi_e^{\alpha_e}}
969: -\epsilon^2\frac{2\pi}{(\hat\xi_e^{\alpha_e})^3}
970: \bigl(\upsilon_e(\hat\xi_e^{\alpha_e})^2-\kappa_e{\omega_e}^2\bigr)I
971: +O(\epsilon)^3
972: \end{split}\end{equation}
973: This is of the form (\ct{MeyerKRHallGR-1991.1}, Theorem~2, page~231),
974: namely $(I,\psi)\mapsto(I^\prime,\psi^\prime)$ by
975: \begin{equation*}\begin{split}
976: &I^\prime=I+\epsilon^{r+s}c(I,\psi,\epsilon),\\
977: &\psi^\prime=\psi+\omega+\epsilon^sh(I)+\epsilon^{s+r}d(I,\psi,\epsilon)
978: \end{split}\end{equation*}
979: with $r=1$, $s=2$, and
980: \begin{equation*}
981: h(I)\equiv-\frac{2\pi}{(\hat\xi_e^{\alpha_e})^3}
982: \bigl(\upsilon_e(\hat\xi_e^{\alpha_e})^2-\kappa_e{\omega_e}^2\bigr)I.
983: \end{equation*}
984: The twist condition $dh/dI\ne0$ is
985: \begin{equation}\elb{30}
986: \upsilon_e(\hat\xi_e^{\alpha_e})^2-\kappa_e{\omega_e}^2\ne0,
987: \end{equation}
988: which, after substituting~\erf{19}, \erf{23}, \erf{25}, and~\erf{24},
989: is
990: \begin{equation*}\begin{split}
991: &-\frac12\left(\frac2{I_3}-\frac1{I_1}-\frac1{I_2}\right)+I_3\left(\frac1{I_3}-\frac1{I_1}\right)\left(\frac1{I_3}-\frac1{I_2}\right)\\
992: &\qquad\qquad\qquad\mbox{}=\frac{1}{I_1I_2}\left(I_3-\frac12(I_1+I_2)\right)\ne0.
993: \end{split}\end{equation*}
994: This is certainly true if $I_3$ is not between $I_1$ and $I_2$.
995:
996: As for the equilibria $\hat p_e^{\alpha_e,0}$ of Section~\thrf{300},
997: Arnold's Stability Theorem (\ct{MeyerKRHallGR-1991.1}, Theorem~1,
998: page~235) together with the normal form~\erf{325} imply that the
999: equilibrium is stable when $a=0$ if
1000: \begin{equation*}
1001: \left.\omega_eI+\frac12\upsilon_eI^2
1002: +\hat\xi_e^{\alpha_e}\nu+\frac12\kappa_e\nu^2-\frac{\omega_e}{\alpha_e}I\nu
1003: \right|_
1004: {\!\!\!\renewcommand{\arraystretch}{.2}\begin{array}[c]{l}\scriptstyle
1005: I=\hat\xi_e^{\alpha_e}\\\scriptstyle\nu=-\omega_e\end{array}}
1006: \!\!\!\!\!\!=\frac12(\upsilon_e(\hat\xi_e^{\alpha_e})^2-\kappa_e{\omega_e}^2)\ne0,
1007: \end{equation*}
1008: which is the same as the twist condition~\erf{30}. Stability follows
1009: for sufficiently small nonzero~$a$ since $a$ contributes continuously.
1010:
1011: Thus $p_e^{\alpha_e}$ are stable as equilibria on the Poisson reduced
1012: space $\sset{(\pi,p)}$. From~\ct{PatrickGWRobertsRMWulffC-2001.1}, and
1013: since the momentum at $p_e^{\alpha_e}$ has zero translational part and
1014: rotational part parallel to $\mbf k$, this implies $\mtl{SO}(2)\times
1015: C$ stability of the original equilibrium, where $C$ is any cone about
1016: $\mbf k$.
1017:
1018: \begin{figure}[t]
1019: \setlength{\unitlength}{1in}
1020: \centerline{{\begin{picture}(4.8,1.225)
1021: \put(.1,.04){{\epsfysize=1.2in\epsfbox{f1.eps}}}
1022: \put(0,.9){\footnotesize$\sqrt[4]{D}\pi_x$}
1023: \put(.8,0){\footnotesize$\pi_y/\sqrt[4]{D}$}
1024: \put(1.65,.04){{\epsfysize=1.2in\epsfbox{f2.eps}}}
1025: \put(1.64,.9){\footnotesize$\psi^\prime-\psi$}
1026: \put(2.5,0){\footnotesize$I$}
1027: \put(3.425,.04){{\epsfysize=1.2in\epsfbox{f3.eps}}}
1028: \put(3.43,.9){\footnotesize$\frac{I}{h(I)}$}
1029: \put(4.2,0){\footnotesize$I$}
1030: \end{picture}}}
1031: \caption{\label{101} Numerical verification of the twist predicted
1032: by~\erf{18}. Leftmost: the curved leading edge of the numerically
1033: computed Poincar\'e map indicates a twist map by visibly showing
1034: faster rotation as $I$ increases. The twist overlays a constant
1035: rotation (in $I$) which is caused by high order terms in $a$ and
1036: decreases as $a$ decreases. Center: the rotation angle per iteration
1037: of the Poincar\'e map on the right as a function of $I$. The slope
1038: corresponds to the twist predicted by~\erf{18}. Right: reciprocal of
1039: the twist for $I_3=1$, $I_2=I_3-I_1$ as $I_1$ ranges from $0$ to $1$
1040: compared to the parabola predicted by~\erf{18}.}
1041: \end{figure}
1042:
1043: The foregoing sort of analysis will always lead to some stability
1044: condition, irrespective of possible errors in the derivation, so it
1045: necessary to check~\erf{18} by comparing it with numerically generated
1046: Poincar\'e maps. Substitution of $\epsilon=1$ after truncation of
1047: $O(\epsilon)^3$ into~\erf{18} gives the leading behavior of the
1048: Poincar\'e map when $a$, $\nu$, and $I$, are of comparable order (they
1049: are all order $\epsilon^2$). The Poincar\'e map is determined to leading
1050: order in $I$ by the first order twist term $h(I)$ in when the
1051: zero~order term $2\pi\omega_e/\hat\xi_e^{\alpha_e}$ has a vanishing
1052: effect (i.e. is a multiple of $2\pi$). This happens to occur when
1053: $I_1+I_2=I_3$, as is easily verified. After substitution of
1054: $I_2=I_3-I_1$, the first order twist is
1055: \begin{equation*}
1056: h(I)=-\frac{\pi {I_3}^2}{I_1(I_3-I_1)}I,
1057: \end{equation*}
1058: whereupon
1059: \begin{equation*}
1060: \frac{I}{h(I)}=-\frac1{\pi {I_3}^2}I_1(I_3-I_1),
1061: \end{equation*}
1062: which is a parabola in $I_1/I_3$. As can be seen in Figure~\thrf{101},
1063: this compares well with numerical integrations of the
1064: original (as opposed to the blown-up) system.
1065:
1066: \section*{Summary}
1067: %%
1068: The following theorem has been proved.
1069: \begin{theorem}
1070: Within the context of the Lagrangian system~\erf{00}, the motion of an
1071: underwater ellipsoid rotating about a long or short principle axis of
1072: inertia is stable modulo $\mtl{SO}(2)\times C$ where $\mtl{SO}(2)$
1073: acts around the rotation axis and $C$ is any cone
1074: containing that axis.
1075: \end{theorem}
1076: This theorem follows from KAM confinement after a blow-up construction
1077: and normal form analysis rather than confinement by Lyapunov functions
1078: derived from energy and momentum.
1079:
1080: \section*{Acknowledgments}
1081: %%
1082: This work was supported by an NSERC individual research grant and an
1083: EPSRC visiting fellowship. I thank the University of Warwick
1084: Mathematics Institute for its hospitality during a sabbatical visit
1085: while this paper was written.
1086:
1087: \frenchspacing\footnotesize
1088: %\bibliographystyle{chicagoa}\bibliography{abbrev,ref}
1089: \begin{thebibliography}{}
1090:
1091: \bibitem[\protect\citeauthoryear{Leonard}{Leonard}{1997}]{LeonardNE-1997.1}
1092: Leonard, N.~E. [1997].
1093: \newblock Stability of a bottom-heavy underwater vehicle.
1094: \newblock {\em Automatica---J. IFAC\/}~{\em 33}, 331--346.
1095:
1096:
1097: \bibitem[\protect\citeauthoryear{Leonard and Marsden}{Leonard and
1098: Marsden}{1997}]{LeonardNEMarsdenJE-1997.1}
1099: Leonard, N.~E. and J.~E. Marsden [1997].
1100: \newblock Stability and drift of underwater vehicle dynamics: mechanical
1101: systems with rigid motion symmetry.
1102: \newblock {\em Physica D\/}~{\em 105}, 130--162.
1103:
1104:
1105: \bibitem[\protect\citeauthoryear{Marsden and Ratiu}{Marsden and
1106: Ratiu}{1994}]{MarsdenJERatiuTS-1994.1}
1107: Marsden, J.~E. and T.~S. Ratiu [1994].
1108: \newblock {\em Introduction to Mechanics and Symmetry}, Volume~17 of {\em Texts
1109: in Applied Mathematics}.
1110: \newblock Springer-Verlag.
1111:
1112:
1113: \bibitem[\protect\citeauthoryear{Meyer and Hall}{Meyer and
1114: Hall}{1991}]{MeyerKRHallGR-1991.1}
1115: Meyer, K.~R. and G.~R. Hall [1991].
1116: \newblock {\em Introduction to {Hamiltonian} dynamical systems and the $N$-body
1117: problem}.
1118: \newblock Springer-Verlag.
1119:
1120:
1121: \bibitem[\protect\citeauthoryear{Patrick, Roberts, and Wulff}{Patrick
1122: et~al.}{2002}]{PatrickGWRobertsRMWulffC-2001.1}
1123: Patrick, G., R.~M. Roberts, and C.~Wulff [2002].
1124: \newblock Stability of {P}oisson equilibria and {H}amiltonian relative
1125: equilibria by energy methods.
1126: \newblock Preprint.
1127:
1128: \bibitem[\protect\citeauthoryear{Patrick}{Patrick}{1995}]{PatrickGW-1995.1}
1129: Patrick, G.~W. [1995].
1130: \newblock Relative equilibria of {Hamiltonian} systems with symmetry:
1131: linearization, smoothness, and drift.
1132: \newblock {\em J. Nonlinear Sci.\/}~{\em 5}, 373--418.
1133:
1134:
1135: \end{thebibliography}
1136:
1137: \end{document}
1138: