1: %===================================================================
2: \documentclass[12pt,oneside,a4paper,leqno]{article}
3: %===================================================================
4:
5: \usepackage{amsfonts,amssymb}
6: \usepackage{geometry}\geometry{a4paper,scale={0.8,0.9}}
7: \usepackage[centertags]{amsmath}
8: \usepackage[all]{xy}\xyoption{all}
9: \usepackage{graphicx}
10: \usepackage{amsthm}
11: % \usepackage{comment}
12: %===================================================================
13: % HYPHENATION
14: \hyphenation{Leo-nar-do}
15: %===================================================================
16: \newcounter{lemma}
17: \swapnumbers
18: % THEOREM ENVIRONMENTS
19: \theoremstyle{plain}
20: \newtheorem{lemma}[equation]{Lemma}\numberwithin{lemma}{section}
21: \newtheorem{theo}[equation]{Theorem}
22: \newtheorem{propo}[equation]{Proposition}
23: \newtheorem{coro}[equation]{Corollary}
24: %
25: \theoremstyle{definition}
26: \newtheorem{defi}[equation]{Definition}
27: \newtheorem{example}[equation]{Example}
28: \newtheorem{remark}[equation]{Remark}
29: %
30: \theoremstyle{remark}
31:
32: \numberwithin{equation}{section}
33: \renewcommand{\headheight}{14pt}
34: \renewcommand{\headsep}{14pt}
35:
36: \renewcommand{\qedsymbol}{\emph{q.e.d.\/}}
37: %===================================================================
38: % NEW COMMANDS
39: \makeatletter
40: % \def\@eqnnum{{\normalfont \normalcolor --\theequation}}
41: % \def\theequation{\@arabic\c@equation}
42: % \def\theequation{\@arabic\c@equation}
43: % it does not work with amsmath! \def\@eqnnum{XXXXX}
44: \def\tagform@#1{\maketag@@@{\ignorespaces#1\unskip\@@italiccorr}}
45: \renewcommand{\theequation}{(\arabic{section}.\arabic{equation}\/)}
46: % \let\c@enumi\c@equation
47: % it does not work, because it secounter=1 at the beginning of enumerate.
48: \renewcommand{\labelenumi}{$(\roman{enumi})$}
49: \makeatother
50: %===================================================================
51: \newcommand{\ldiag}[2]{%
52: \begin{equation}\label{#1}\begin{aligned}\xymatrix{#2}\end{aligned}\end{equation}%
53: }
54: \newcommand{\diag}[1]{%
55: \begin{equation}\begin{aligned}\xymatrix{#1}\end{aligned}\end{equation}%
56: }
57: \newcommand{\ndiag}[1]{%
58: \begin{equation*}\begin{aligned}\xymatrix{#1}\end{aligned}\end{equation*}%
59: }
60: %===================================================================
61:
62:
63:
64: %===================================================================
65: % XY-PIC SPECIFIC NEW COMMANDS
66: \newdir{ >}{!/8pt/@{ }*@{>}}
67: \newdir{< }{!/-8pt/@{ }*@{<}}
68: %===================================================================
69:
70: %============================================================================
71: \theoremstyle{plain}
72: \newtheorem{assumption}[equation]{Collision Assumption}
73: % \usepackage{epigraph}
74: \usepackage{paralist}
75: % \setlength{\epigraphrule}{0pt}
76: %============================================================================
77: \newcommand{\potenziale}{\mathcal{U}}
78: \newcommand{\cinetica}{\mathcal{K}}
79: \newcommand{\C}{\mathbb{C}}
80: \newcommand{\R}{\mathbb{R}}
81: \newcommand{\CP}{\mathbb{CP}}
82: \newcommand{\RR}{\mathbb{R}}
83: \newcommand{\RP}{\mathbb{RP}}
84: \newcommand{\FN}{\mathbb{F}_n\mathbb{C}}
85: \newcommand{\CC}{\mathrm{CC}}
86: \newcommand{\fix}{\mathrm{Fix}}
87: \newcommand{\fbar}{\overline{f}}
88: \newcommand{\from}{\colon}
89: \newcommand{\Chi}{\mathcal{X}}
90: \newcommand{\minus}{\smallsetminus}
91: \newcommand{\crit}{\mathrm{Crit}}
92: \newcommand{\barCC}{\overline{\mathrm{CC}}}
93: \newcommand{\action}{\mathcal{A}}
94: \newcommand{\nn}{\mathbf{n}}
95: \newcommand{\todo}{....\marginpar{\fbox{\large{....}}}}
96: \newcommand{\st}{\ \mathrm{:} \ }
97: \newcommand{\ze}{\mathbb{Z}}
98: \newcommand{\Usf}{U_{\mathrm{sf}}}
99: %============================================================================
100:
101: %===================================================================
102: \begin{document}
103: \pagenumbering{arabic}
104:
105: \title{%
106: Symmetric periodic orbits for the $n$-body problem:
107: some preliminary results
108: }
109:
110: \author{Davide L.~Ferrario}
111:
112: \date{}
113: \maketitle
114:
115: \begin{abstract}
116: We show the existence of some infinite families of
117: periodic solutions of the planar Newtonian $n$-body problem --with
118: positive masses-- which are symmetric with
119: respect to suitable actions of finite groups
120: (under a strong--force assumption or only numerically).
121: The method is by minimizing a discretization of the action functional
122: under symmetry constraints.
123: \end{abstract}
124:
125: % \subjclass{
126: % Primary
127: %todo
128: % Secondary
129: %todo
130: % }
131:
132: % \keywords{%
133: %todo
134: % }
135:
136:
137:
138: %======================================================================
139:
140: \section{Introduction}
141: \label{sec:intro}
142: Exact periodic solutions of the Newtonian $n$-body problem
143: have long been a topic of interest, and not only among the specialists
144: of celestial mechanics.
145: The recently found ``eight'' choreography of Montgomery and Chenciner
146: \cite{monchen}
147: has been the starting point of the numerical discovery
148: by Sim\'o and others of several
149: interesting periodic orbits of the same kind,
150: which are symmetric under the action
151: of a finite group \cite{chenven,chenciner}.
152: In this note we simply define suitable actions
153: of some finite groups on the configuration space
154: and infer the existence of periodic
155: orbits
156: by the well-known variational approach. For details and
157: further reading the reader is referred to
158: \cite{ambbad,argate,mate,setate,mate95,
159: monchen,chenven,andrea,ambrosetticoti,ambrocoti}.
160: The most recent advances in this theory will appear
161: in Chenciner's papers for the ICM 2002 (see also \cite{chenciner}).
162: We need to consider the assumption that the local minima
163: of the Lagrangian action for the corresponding Bolza problem
164: is collision-free.
165: This has been proved to be true by Marshall (in a preprint)
166: in case the symmetry group is cyclic and the potential is Newtonian;
167: it deserves some effort to be proved in general. But in any case
168: we can circumvent this problem by a strong--force perturbation of
169: the potential.
170: The paper is roughly structured as follows: we will prove rigorously
171: (even if omitting full details, that can be find in the cited papers)
172: some needed
173: results, and then use numerical simulations to verify the
174: collision assumption when needed and to compute the orbits.
175: We obtain some non--trivial periodic orbits in the following cases.
176:
177: \begin{inparaenum}[\itshape (a)]
178: \item
179: $3$-body with all equal masses: the Montgomery--Chenciner choreography
180: (section \ref{subsec:threeeq}).
181: \item
182: $3$-body with two equal masses:
183: infinitely many periodic orbits (section \ref{subsec:3_2eq} and
184: figure \ref{fig:candidate}).
185: Among these, there are
186: the orbits with all the three masses equal (figure \ref{fig:3b}).
187: \item
188: $4$-body with all equal masses:
189: the method only yields some numerical examples (section \ref{subsec:4eq}),
190: like the one in figure \ref{fig:nonhomographic}.
191: \item
192: $4$-body with two pairs of equal masses:
193: an infinite family of solutions
194: (section \ref{subsec:twopairs} and
195: figure \ref{fig:rough}).
196: \item
197: $4$-body with all equal masses and an additional symmetry:
198: an infinite family of periodic orbits (section \ref{subsec:4eqp},
199: figures \ref{fig:mac}, \ref{fig:caseq6}).
200: \item
201: $4$-body with two pairs of equal masses and an additional symmetry:
202: an infinite family (section \ref{subsec:22eqp},
203: figure \ref{fig:arch}).
204: \item
205: Plane choreographies (``eight'' shape) with $n$ odd equal masses
206: (section \ref{sec:choreo}, figures \ref{fig:chor5}, \ref{fig:chor5bis}).
207: \end{inparaenum}
208:
209: Many of these orbits are well-known (but in general only numerically),
210: other might not. This is an attempt ---most
211: far from being ultimate--- to give a general method
212: for proving the existence of such periodic orbits with variational
213: methods combined with a computer--assisted proof. More general
214: results (not confined to dihedral groups and without the collision
215: assumption) will come in subsequent papers.
216: Most of the ideas are borrowed from Susanna Terracini, who helped
217: me to understand the problem and willingly shared her knowledge
218: during several discussions on the topic. Sincere thanks
219: are also due to Andrea Venturelli, who generously gave his
220: much appreciated comments.
221:
222:
223: \section{Preliminaries}
224: Let $n\geq 2$ be the number of bodies in the
225: Euclidean space $V=E^k \approx \R^k$ of dimension $k\geq 2$.
226: Here we consider the $n$-body problem with masses $m_i$, $i=1,\dots,n$.
227: Following the notation of \cite{chenciner},
228: let $\Chi$ denote the subspace of $V^n$
229: of all points $x=(x_1,x_2,\dots,x_n)$ such that
230: $\sum_{i=1}^n m_ix_i = 0$, i.e., with center of mass in the origin $O$.
231: Let $\Delta_{i,j}$ denote the collision set in $\Chi$ of the $i$-th
232: and the $j$-th particles,
233: and $\Delta = \cup_{i<j} \Delta_{i,j}$ the collision set
234: in $\Chi$.
235: Let $U\from \Chi \to \R$ be the potential function
236: \[
237: U(x) = \sum_{i<j} \dfrac{1}{|x_i - x_j|}.
238: \]
239: We can consider also a deformation $\Usf$ of $U$ in a small
240: neighborhood of the collision set, so that $U$ satisfies
241: the \emph{strong force} condition.
242: The kinetic energy, defined on the tangent bundle of $\Chi$,
243: is
244: $K=\sum_{i=1}^n \dfrac{1}{2} m_i |\dot x_i|^2$, and the Lagrangian
245: is $L=K+U$.
246: Moreover $I=\sum_{i} m_i |x_i|^2$ is the moment of inertia with
247: respect to the center of mass.
248: Let $T^1\subset \R^2$ denote the unit circle and
249: $\Lambda = H^1(T^1,\Chi)$ the Sobolev space
250: of the $L^2$ loops $T^1 \to \Chi$ with $L^2$ derivative.
251: Then,
252: the critical points of the positive-defined action functional
253: $\action \from \Lambda \to \R\cup \{\infty\} $
254: are
255: periodic orbits of the Newtonian $n$-body problem;
256: the action is defined by
257: \begin{equation}\label{eq:action}
258: \action(x) = \int_{T^1} L(x(t),\dot x(t) ) dt
259: \end{equation}
260: for every loop $x=x(t) \in \Lambda$.
261: The action functional is called
262: \emph{coercive} if it goes to infinity as
263: the moment of inertia $I$ goes to infinity.
264: (i.e., if the $H^1$-norm of $x$ goes to infinity).
265:
266: Let $G$ be a finite group, acting on a space $X$. The space $X$ is
267: then called \emph{$G$-equivariant} space.
268: We recall some
269: standard notation.
270: If $H\subset G$ is a subgroup of $G$, then
271: $G_x = \{g\in G \st gx=x \}$ is termed the \emph{isotropy} of $x$,
272: or the \emph{fixer} of $x$ in $G$.
273: The space $X^H\subset X$ consists of all the points
274: $x\in X$ which are fixed by $H$, that is
275: $X^H=\{x\in X \st G_x \supset H \}$.
276: Given two $G$-equivariant spaces $X$ and $Y$,
277: an \emph{equivariant map} $f\from X \to Y$ is a map
278: with the property that $f(g\cdot x) = g\cdot f(x)$
279: for every $g\in G$ and every $x\in X$.
280: An equivariant map induces, by restriction to the spaces
281: $X^H$ fixed by subgroups $H\subset G$, maps
282: $f^H\from X^H \to Y^H$.
283:
284: \section{Symmetry constraints}
285: We give here an introduction to symmetry
286: constraints which is slightly different than the well-known
287: in the literature.
288: Let $G$ be a finite group, $\tau$ an orthogonal representation of
289: $G$ on
290: $T^1$ and
291: $\rho$ an orthogonal representation on the Euclidean space
292: $V$. Furthermore, let $\sigma$
293: be a group homomorphism $\sigma\from G \to \Sigma_n$
294: from $G$ to the symmetric group on $n$ elements. Therefore
295: we let $G$ act on the space $V$, on the time $T^1$ and on the
296: set of indexes of the masses $\mathbf{n}=\{1,2,\dots, n\}$.
297: Let $\R[\mathbf{n}]$ denote the
298: $n$-dimensional real vector space generated by the $n$
299: elements of $\mathbf{n}$,
300: and
301: $\R_0[\nn]$ the linear subspace
302: of elements with coordinates $(\lambda_1,\lambda_2,\dots,\lambda_n)$
303: such that $\sum_{i=1}^n m_i\lambda_i = 0 $.
304: Given $\sigma\from G \to \Sigma_n$, $G$ acts on
305: $\R_0[\nn]$ by the action on $\nn$,
306: provided that $\sigma(g)(i) = j$ implies $m_i = m_j$
307: for every $g\in G$ and every $i=1,\dots, n$.
308: Two homomorphisms $\sigma$ and $\sigma'$ yield
309: the same real representation on $\R_0[\nn]$
310: if they are conjugated by an element of $\Sigma_n$,
311: that is if one is obtained by permuting coordinates on the other.
312: With an abuse of notation we call $\sigma$ the representation
313: induced by the homomorphism $\sigma$. It is equivalent to
314: the representation
315: given by $[N] - [1]$ (in the representation ring of $\Sigma_n$),
316: where $N$ is the natural representation and $1$ the trivial
317: representation.
318: Now, since
319: \[
320: \Chi \cong V\otimes_\R \R_0[\nn],
321: \]
322: given $\rho$ and $\sigma$ we have
323: an orthogonal action of $G$ on $\Chi$.
324:
325: Furthermore, since $G$ acts on $T^1$
326: and on $\Chi$, then there is the standard diagonal action
327: on the loop space $\Lambda$, defined by
328: $x(t) \mapsto g x( g^{-1} t)$.
329: Let us note that
330: the loops in $\Lambda$ fixed by $G$ are
331: the equivariant maps $x\from T^1 \to \Chi$.
332: In this terminology, a \emph{symmetry constraint} is
333: a such action of $G$ on $\Lambda$. Since the action
334: functional is invariant with respect to
335: the $G$-action,
336: we have a restricted action
337: \begin{equation}
338: \label{eq:lambdag}
339: \action^G\from \Lambda^G \subset \Lambda \to \R,
340: \end{equation}
341: and
342: the following proposition (Palais principle of symmetric
343: criticality -- see \cite{chenciner}).
344:
345: \begin{lemma}\label{palais}
346: A critical point of $\action^G$ in $\Lambda^G$ is a critical point of
347: $\action$ on $\Lambda$.
348: \end{lemma}
349:
350:
351: Now, the problem arises about which representations yield
352: symmetry constraints that are sufficient to imply
353: the existence of nontrivial (i.e., non-homographic) periodic solutions
354: in the equivariant loops class. As shown
355: by Chenciner in \cite{chenciner}, we have to consider the problem
356: of collisions, coercivity and non-triviality.
357: We start by trying to see which conditions on $\tau$, $\rho$
358: and $\sigma$ might give the desired properties.
359:
360:
361: Let $\ker\tau$, $\ker \rho$ and
362: $\ker \sigma$ be the kernels of the corresponding
363: homomorphisms
364: $\rho\from G\to O(k)$, $\tau\from G \to O(2)$
365: and $\sigma\from G \to \Sigma_n$.
366: Without loss of generality we can assume that
367: $\ker\tau \cap \ker\rho \cap \ker\sigma = 1$.
368: Moreover, assume that $g\in \ker\tau \cap \ker\rho$:
369: this implies that $g\not\in \ker\sigma$.
370: Then, if $x(t)$ is an equivariant loop, then the restriction
371: map $x(t)^g\from T^1={T^1}^g \to \Chi^g$
372: sends every point of $T^1$ to $\Chi^g$;
373: since $g\in \ker \rho$ but $g\not\in \ker \sigma$, the space
374: $\Chi^g$ consists entirely of collisions.
375: Therefore we must have $\ker\tau \cap \ker\rho = 1$
376: in order to avoid necessary collisions.
377:
378:
379: Furthermore, assume that $g\in \ker\tau \cap \ker\sigma$,
380: and thus $g\not\in \ker\rho$.
381: Again, every configuration in the orbit
382: $x(t)^g = x(t)$ needs to belong to $\Chi^g$,
383: which is nothing but the subspace of configurations
384: of points in $V^g$, which is a linear proper
385: subspace of $V$. Thus we can consider it as a sub-problem,
386: and assume that $\ker\tau \cap \ker\sigma = 1$ as well.
387:
388:
389: Finally, consider $g\in \ker\rho\cap \ker\sigma$.
390: Its action on $T^1$ can be a rotation or a reflection.
391: In case it is a rotation, we are considering $n$-bodies that
392: actually tread the same loop more than once, and clearly
393: the problem can be solved by solving the problem concerning
394: loops with just one iteration. So,
395: we can assume that
396: every element of
397: $\ker\rho\cap \ker\sigma$
398: acts as a reflection on $T^1$.
399: But if there are two distinct such elements
400: $g_1\neq g_2 \in \ker\rho\cap \ker\sigma$,
401: then their product $g_1g_2$ would act as a rotation on $T^1$,
402: hence $g_1g_2$ must be trivial, i.e., $g_1=g_2^{-1} = g_2$.
403: This implies that
404: $\ker\rho\cap \ker\sigma$ has at most one non-trivial
405: element, that is it is a subgroup of order at most $2$.
406: If $\ker \rho\cap \ker \sigma \neq 1$, then
407: every loop in $\Lambda^G$ can be decomposed as $\gamma \gamma^{-1}$,
408: i.e., it is a loop that runs along a path $\gamma$ from $g(0)$ to
409: $\gamma(1)$ in $1/2$ of the time,
410: and then from $\gamma(1)$ to $\gamma(0)$ in the second half
411: of the time interval.
412: % here these are the hip-hop-like solutions...
413:
414: \begin{remark}
415: The equivariant loops $x\in \Lambda^G$ can be seen as
416: $G/\ker \tau$-equivariant
417: loops $T^1 \to \Chi^{\ker \tau}$. Thus we can consider the same
418: problem related to
419: a finite subgroup of
420: $O(2)$
421: (thus a subgroup of a dihedral group)
422: and a linear subspace
423: $\Chi^{\ker \tau} \subset \Chi$.
424: Moreover, if $\Chi^{\ker \tau} \subset \Delta$ then all the loops
425: are just made of collision points.
426: Therefore we assume that $\Chi^{\ker \tau} \not\subset \Delta$.
427: \end{remark}
428:
429: If for a choice of $\sigma$, $\rho$ and $\tau$
430: one of the following cases occurs, we say that the action
431: of $G$ is \emph{degenerate}.
432: \begin{inparaenum}[\itshape (a)]
433: \item
434: $\ker \tau \cap \ker\sigma \cap \ker \rho \neq 1$.
435: \item
436: Every loop in $\Lambda^G$ has collisions.
437: \item
438: There is a proper linear subspace $S$ of $E^k$ such that
439: for every $t\in T^1$ and
440: for every $x \in
441: \Lambda^G$, the body $x_i(t)$ belongs
442: to $S$.
443: \item
444: For every loop $x(t)$ in $\Lambda^G$
445: there is a loop $y(t)$ and $k\in \ze$, $k\neq 0,\pm 1$, such that
446: for every $t\in T^1$ we have $x(t)=y(kt)$.
447: \end{inparaenum}
448:
449: \begin{lemma}
450: If the action of $G$ is non-degenerate, then
451: $G$ is a finite subgroup of
452: $O(2)\times O(k)$ and a finite subgroup
453: of $O(2)\times \Sigma_n$.
454: \end{lemma}
455: \begin{proof}
456: Since $\ker \tau \cap \ker \rho = 1$,
457: the homomorphism $\tau \times \rho \from G \to
458: O(2)\times O(k)$ has trivial kernel.
459: The same happens to the homomorphism $\tau \times \sigma$.
460: \end{proof}
461:
462: For example, in case $k=2$, this implies that
463: $G$ is a subgroup of the direct product of two dihedral groups,
464: and hence metabelian.
465: For $k=3$, $G$ is a subgroup of the direct product
466: of a dihedral group and a finite subgroup of $O(3)$,
467: hence either $G$ is metabelian or
468: it is an extension of a finite metabelian group
469: with with a finite subgroup of $O(3)$.
470: Hence the only nonsolvable group occurs if $G$ projects onto the icosahedral
471: group $A_5$ in $O(3)$.
472:
473: \section{Coercivity}
474:
475: \begin{lemma}
476: \label{lemma:coercive}
477: The symmetric action functional $\Lambda^G$ is coercive
478: if and only if $\Chi^G = 0$.
479: \end{lemma}
480: \begin{proof}
481: Consider the group $G/\ker \tau$. It is a finite subgroup
482: of $O(2)$, hence it is either a cyclic group or a dihedral group.
483: Let us consider first the cyclic case. Let $c$ be a generator of
484: $G/\tau$. Let $X=\Chi^{\ker\tau}$. Since $\Chi^G=X^c$, we have
485: $X^c=0$. Therefore $X$ can be decomposed into irreducible
486: components $X=\R+\R+\dots+\R+\C+\dots+\C$,
487: where on the one-dimensional components $\R$ the action
488: of $c$ is given by $c(s)=-s$, while on the two-dimensional
489: components we have
490: $c(z) = e^{2\pi i / l}z$ for a suitable $l\in \ze$.
491: Thus, using the same argument as in Bessi and Coti-Zelati \cite{bessi},
492: it is possible to show that $\Lambda^G$ is coercive.
493: Now consider the case $G/\tau$ is dihedral. Let $h_1$ and
494: $h_2$ be two generators of order $2$ of $G/\tau$.
495: Again, $X$ can be decomposed as
496: $X=\R+\R+\dots +\R + \C \dots + \C$, where
497: on the one-dimensional irreducible components
498: the action is either $r_1(s)=-s=r_2(s)$ or $r_1=-r_2(s)=\pm s$,
499: while on the two-dimensional irreducible components
500: $\C$ is a dihedral representation.
501: Thus, again exactly with same argument as \cite{bessi}
502: it can be shown that there is $\alpha>0$
503: such that $|x|_{L^2} \leq \alpha |\dot x|_{L^2}$,
504: i.e., that the action functional is coercive.
505:
506: For the converse, if $\Chi^G \neq 0$ let $x_0$ denote
507: a loop in $\Lambda^G$ (possibly with collision) with finite action
508: $\action^G$. Then $x_0 + v$, with $v\in \Chi^G$, is again a
509: loop in $\Lambda^G$, with action $\action^G(x_0+v) < \action^G(x_0)$.
510: But as $|v|\to \infty$ also $x_0 + v$ goes to infinity,
511: hence $\action^G$ is not coercive.
512: \end{proof}
513:
514:
515:
516:
517:
518: \section{Dihedral orbits}
519: Consider the time circle $T^1\subset \R^2$ of radius $\frac{T}{2\pi}$,
520: where $T$ is the period of a periodic orbit. Let $h_1$ and
521: $h_2$ be two reflections in $\R^2$ that fix two lines
522: forming an angle of $\frac{\pi}{l}$, with $l>1$.
523: Then the group $G$ generated by $h_1$ and $h_2$ is the
524: \emph{dihedral group} $D_{l}$ of order $2l$.
525: Consider as above a $k$-dimensional orthogonal representation
526: of $G$ and an homomorphism $\sigma\from G \to \Sigma_n$
527: to the symmetric group of order $n!$.
528: This means that $h_1$ and $h_2$ act on $V=E^k$ (with a symmetry of order $2$
529: along a plane, a line or the origin) and on the set
530: $\{1,2,\dots, n\}$ of indexes via the homomorphism $\sigma$.
531: The elements $\sigma(h_i)$ need to be of order $2$ in $\Sigma_n$,
532: whenever they are not trivial.
533: Given these data, $G$ acts on $\Chi$ by
534: \[
535: g (x_1,\dots, x_n) = (gx_{\sigma(1)},\dots, gx_{\sigma(n)} ),
536: \]
537: where we mean $\sigma(i) = \sigma(g)(i)$ and on $T^1$.
538: Then $G$ acts on the loop space $\Lambda$ by
539: \[
540: g\cdot \gamma\from t \to g \gamma(g^{-1}t)
541: \]
542: for every $t\in T^1$ and every $\gamma\in \Lambda$.
543: The space $\Lambda^G$ consists of the equivariant loops.
544: It is easy to see that $\Lambda^G$ is homeomorphic to the space $P$
545: of all the paths $\lambda\from [0,1] \to \Chi$
546: with the property that $\lambda(0) \in \Chi^{h_1}$ and
547: $\lambda(1)\in \Chi^{h_2}$.
548: The homeomorphism is given by the restriction function
549: $r\from \Lambda^G \to P$.
550: The action functional can be defined in exactly
551: the same way on $P$, by integrating $L$
552: along $\lambda$ with a rescaled time.
553: Let is denote it by $\action_P$.
554: If $L$ is invariant with respect to the action of $G$,
555: then $2l \action_P r (\gamma) = \action^G (\gamma)$,
556: for every $\gamma\in \Lambda^G$.
557: Hence $\gamma$ is a stationary point for $\action^G$ if and only if
558: its restriction $r(\gamma)$ is stationary for $\action_P$.
559: We can hence consider critical points and local minima
560: of $\action_P$ in $P$.
561: This is a sort of
562: \emph{generalized Bolza problem}.
563: \begin{lemma}
564: Any critical point of $\action_P$ in $P$ yields
565: a critical point of $\action^G$ in $\Lambda^G\subset \Lambda$, which
566: is a critical point of $\Lambda$.
567: \end{lemma}
568:
569: \section{Minimizing on constrained paths}
570: More generally, assume that $h_1$ and $h_2$ are two elements of order $2$
571: acting isometrically on $E^k$ and
572: $\{1,2,\dots,n\}$.
573: Let $X_1$ and $X_2$ be two the fixed
574: subspaces $\Chi^{h_1}$ and $\Chi^{h_2}$ of $\Chi$
575: and let $P$ denote the Sobolev space of all the paths
576: $\gamma\from [0,1] \to \Chi$ such that
577: $\gamma(0) \in X_1$ and $\gamma(1) \in X_2$.
578: That is, $P = H^1( ([0,1],0,1), (\Chi,X_1,X_2) )$ is the Sobolev space
579: of the $L^2$ paths $[0,1] \to \Chi$ with $L^2$ derivative
580: with the constraints at the endpoints of the interval $[0,1]$.
581: Let $\action_P$ be defined on $P$ as above.
582: \begin{lemma}
583: Any local minimum of $\action_P$ can be extended
584: to a solution (in the weak sense)
585: $x\from \R \to \Chi$ which is periodic
586: in a rotating frame.
587: \end{lemma}
588:
589: Now we have to face the problem of the possible existence of collisions
590: in (local) minima of the action functional.
591:
592: \begin{assumption}
593: \label{assumption}
594: If $G$ is a finite group and $\Lambda^G\subset \Lambda$ is
595: defined as in \ref{eq:lambdag}, then
596: all local minima of the action $\action^G$ in $\Lambda^G$ are
597: collision-free.
598: \end{assumption}
599:
600: We cannot term \ref{assumption} a lemma, since it has not yet been
601: proved in full generality. However, there is a certain evidence
602: that it holds for general actions.
603: In fact, if the group $G$ is cyclic and acts
604: in the standard way on $T^1$ (that is, yielding choreographies),
605: then it was proved recently by Marchall (in some unpublished notes).
606: The proof can be extended without significant change to some
607: other group actions, but will not work in full generality.
608: On the other hand Majer--Terracini methods on collisions singularities
609: \cite{serter1,mate,setate,mate95,rate,mate93,serter2,mate93a}
610: can be extended to the equivariant case, if $n$ is $3$ or $4$
611: and under some further assumptions.
612: But there are still some gaps in the proofs, so that we hope
613: to provide a complete proof in the future.
614: For the purpose of this note, it suffices either to
615: consider a strong--force perturbation $\Usf$
616: or to consider the numerical hint that the algorithm stops
617: at a non-collision loop and determines it as a global minimum.
618:
619:
620:
621:
622: \section{Three bodies in the plane}
623: Now we can start to investigate which symmetry constraints
624: yield non-trivial periodic solutions.
625: We start with the case of $3$-bodies.
626: Recall that $G$ now is a dihedral group with standard
627: generators $h_1$ and $h_2$.
628:
629: \subsection{Three equal masses}
630: \label{subsec:threeeq}
631: We can assume that $m_i=1$ for $i=1,2,3$.
632: If $G$ acts on $\{1,2,3\}$ without fixed points,
633: then it must be $\sigma(h_1) = (1 2)$
634: and $\sigma(h_2)=(2 3)$, up to an inner permutation of the indexes.
635: So that the symmetric group $\Sigma_3$ is a homomorphic image of $G$.
636: To determine $G$ and its action on $E^2$
637: we consider now the cases for $h_1$ and $h_2$.
638:
639: First case: both are rotations (of angle $\pi$) on $E^2$. Then the minimal
640: $G$ with this property is the dihedral group $D_3\cong \Sigma_3$ of order $6$.
641: The space $\Chi^{h_1}$ is the space of all the configurations
642: with $x_3=0$ and $x_1 = - x_2$, while $\Chi^{h_2}$ is given by
643: all the configurations with $x_1=0$ and $x_2=-x_3$.
644: It is clear that $\Chi^{h_1} \cap \Chi_{h_2} = 0$,
645: hence by \ref{lemma:coercive} the minimum $x=x(t)$ in $\Lambda^G$ exists
646: and is collision-free by \ref{assumption}.
647: Since the product $h_1h_2$ is a rotation of $T/3$ in the time circle
648: and acts trivially on $E^2$, we have that $x$ is a choreography.
649: It cannot be an Euler or Lagrange solution, hence it is a non-trivial
650: choreography.
651: It is possible to show that it has an ``eight'' shape.
652: There is the natural question, whether
653: it is the same as the Montgomery-Chenciner orbit or not.
654:
655: Second case: $h_1$ acts on $E^2$ by reflection along a line and
656: $h_2$ by rotation of angle $\pi$.
657: Since the product $h_1h_2$ acts as a reflection in $E^2$
658: and as the cyclic permutation $(1 2 3)$ in $\Sigma_n$,
659: $G$ needs to be the dihedral group $D_6$ of order $12$.
660: The configurations in
661: $\Chi^{h_1}$ are those such that $(x_1,x_2,x_3)$ is a triangle
662: symmetric with respect to the line fixed by $h_1$ and the configurations in
663: $\Chi^{h_2}$ are those such that again $x_1=0$ and $x_2=-x_3$.
664: This is the action described in \cite{monchen}, and the corresponding
665: solution
666: is the figure eight choreography.
667:
668: Third case: both $h_1$ and $h_2$ act on $E^2$ by a reflection along
669: a line ($l_1$ and $l_2$ respectively).
670: If $l_1=l_2$, then the product $h_1h_2$ acts trivially
671: on $E^2$, hence the minimal $G$ is the dihedral group $D_6$.
672: The minimum exists and numerical experiments
673: let one guess that it is the Lagrange orbit.
674: Otherwise, $l_1$ and $l_2$ intersect with an angle $\pi/q$,
675: with $q>1$ integer.
676: The minimal $G$ is therefore $D_q$ if $q$ is $0 \mod 3$
677: and $D_{3q}$ otherwise. Since the Lagrange orbit
678: belongs to $\Lambda^G$ it can be the minimum.
679: We did not check whether it is a minimum for every $q$ or not.
680:
681: \subsection{Two equal masses}
682: \label{subsec:3_2eq}
683: Now assume that the first two masses are equal ($m_1=m_2$).
684: We again list the possible cases.
685: Without loss of generality we can assume
686: $\sigma(h_2) = (1 2)$, since at least one of $h_1$ and $h_2$
687: needs to act non-trivially on the indexes.
688:
689: First case: $\sigma(h_1) = (1 2)$.
690: If both $h_1$ and $h_2$ act rotating on $E^2$,
691: then $\Chi^{h_1} = \Chi^{h_2} = \Chi^{G}$, and the functional
692: is not coercive.
693: If $h_1$ acts by reflection and $h_2$ by rotation, then
694: again the functional is not coercive,
695: and the same is true if they act by reflecting along the
696: same line.
697: So it is left to check the case in which they act
698: on $E^2$ by reflection along two distinct lines.
699: In this case the functional is coercive, and $\Lambda^G$ contains
700: the Lagrange solution so that it is of no interest.
701:
702: Second case: $\sigma(h_1) = ()$.
703: Since $h_1$ does not move the indexes,
704: to avoid collisions it is is necessary that
705: $h_1$ does not act on $E^2$ with a rotation.
706: It cannot have a trivial action, since otherwise $\Chi^{h_1} = \Chi$
707: and the orbit would not be dihedral, so that it will be
708: a reflection. Now, it is left to determine the action of
709: $h_2$. If $h_2$ is a rotation of angle $\pi$, then
710: the action functional is not coercive.
711: \begin{remark}
712: We can restrict the space of paths considering only paths
713: $x$
714: with a prescribed order of the configuration $x(0)$.
715: If we look at the configurations
716: such that $x_3(0)$ does not lie between $x_1$ and $x_2$,
717: then $\action$ is coercive.
718: In a strong-force settings a collision-free minimum need to exist.
719: Numerical experiments show that
720: such minima might exist even for a potential of type $1/r^{a}$, with
721: $a\geq 1.3$ (see figure \ref{fig:1}).
722: \ref{fig:1}.
723: \begin{figure}
724: \begin{center}
725: \includegraphics[width=4truecm]{figs/fig1.eps}
726: \caption{Two equal masses }
727: \label{fig:1}
728: \end{center}
729: \end{figure}
730: This is the solution of braid $b_1^2b_2^{-2}$ of Moore \cite{moore},
731: pag. 3677.
732: \end{remark}
733:
734: Otherwise, $h_2$ is a reflection along a line.
735: If this line coincides with the line fixed by
736: $h_1$, then again the functional is not coercive,
737: since there are orbits in which $x_1$ and $x_2$ rotate in a circle
738: very far from $x_3$.
739: So we can assume that the two lines intersect with an angle
740: $0<\alpha \leq \pi/2$.
741: If the angle $\alpha$ is equal to $\pi/2$ then again $\action$ is not
742: coercive, so we assume $0<\alpha<\pi/2$.
743: Now the functional is coercive, and there is a minimum.
744: If $q$ is an integer, then
745: the minimal group $G$ is $D_q$ if $q$ is even
746: and $D_{2q}$ if $q$ is odd.
747: At $t=2T/q$ the configuration is the same as the
748: configuration
749: at $t=0$ with the two bodies interchanged and rotated by
750: an angle of $2\pi/q$; at $t=4T/q$ it
751: is exactly the configuration at $t=0$ rotated of an angle
752: $4\pi/q$.
753: If $q>2$ is not an integer, then one obtains an orbit
754: periodic with respect to a rotating frame.
755: Unfortunately the Euler orbits belong to this class,
756: so that the minimum can be achieved on a Euler solution.
757: Some numerical simulations give a hint that this is not the case:
758: orbits like the one in figure \ref{fig:3b} can be found
759: with constrained optimization techniques, with an action
760: less than the action of the corresponding Euler orbit.
761:
762: \begin{figure}
763: \begin{center}
764: \includegraphics[width=4truecm]{figs/fig_3b.eps}
765: \caption{$m_1=m_2=m_3=1$}
766: \label{fig:3b}
767: \end{center}
768: \end{figure}
769:
770: To prove the existence of such orbits, provided that because
771: of \ref{assumption} there are no collisions in a minimizing
772: orbit, it suffices to use the following level estimates.
773: % \subsection{Level estimates}
774: We are going to compare the levels of the action
775: of suitable symmetric orbits with the action of the Euler orbit.
776: Let $m_1=m_2 = 1$, $m_3=m>0$ be the masses.
777: Let $c$, $r_0$ and $l$ be three positive constants (to be determined),
778: and consider the path in $P$ determined by the equations
779: \begin{equation}
780: \label{orbit1}
781: \begin{split}
782: x_1(t) & =
783: le^{i\theta t} + (r_0+ct)e^{i(\theta - \frac{\pi}{2})t}
784: \\
785: x_2(t) & =
786: le^{i\theta t} - (r_0+ct)e^{i(\theta - \frac{\pi}{2})t}
787: \\
788: x_3(t) & =
789: -\frac{2l}{m} e^{i\theta t}.
790: \\
791: \end{split}
792: \end{equation}
793:
794:
795: The kinetic contribution of $(1)$ and $(2)$ to the action $\action(x)$ is
796: \[
797: \begin{split}
798: K_1 + K_2 =
799: 1/12\,{c}^{2}{\pi }^{2}+1/4\,{{r_0}}^{2}{\pi }^{2}+ {r_0}\,{
800: \theta}^{2}c
801: - c{r_0}\,\theta\,\pi +{c}^{2}+1/3\,{c}^{2}{\theta}^{2
802: }-{{r_0}}^{2}\theta\,\pi +\\
803: + {l}^{2}{\theta}^{2}-1/3\,{c}^{2}\theta
804: \,\pi +{{r_0}}^{2}{\theta}^{2}+1/4\,{r_0}\,{\pi }^{2}c. \\
805: \end{split}
806: \]
807: The kinetic term coming from $(3)$ is simply
808: \[
809: K_3 = 2\frac{l^2 \theta}{m}.
810: \]
811: Now consider the terms corresponding to the potential.
812: The term corresponding to the interaction between $(2)$ and $(3)$
813: is equal to
814: \[
815: U_3 = \frac{1}{2c} \log(1+\frac{c}{r_0}).
816: \]
817: Now, the term of the interaction between $(1)$ and $(3)$ is bounded
818: by
819: \[
820: U_2 \leq
821: m \left( (r_0 + c)^2 + l^2(1+2/m)^2 \right)^{-1/2},
822: \]
823: and a similar inequality holds for the term $(2)-(3)$:
824: \[
825: U_1 \leq
826: \dfrac{m^2}{m(l-r_0)+2l}.
827: \]
828: Let $\action_D = K_1+K_2+K_3 + U_1 + U_2 + U_3$ denote the Lagrangian action of the path
829: \ref{orbit1}.
830: The action of the Euler solution with the body $x_3$ in the center of mass
831: is
832: \[
833: \action_E =
834: \frac{3}{2} \left[
835: (1/2 + 2m)^2(\pi/2 -\theta)
836: \right]^{1/3}.
837: \]
838: With some computations it is possible to simplify the difference as
839: \begin{equation}
840: \label{eq:difference}
841: \begin{split}
842: \action_D-\action_E =
843: l^2 \theta^2 (1+2/m) + c^2
844: + \frac{1}{12} (3r_0^2 + c^2 + 3cr_0)(\pi-2\theta)^2
845: + \dfrac{m^2}{m(l-r_0)+2l} +
846: \\
847: + \dfrac{m}{ \sqrt{ (r_0 + c)^2 + l^2(1+2/m)^2 }} +
848: \frac{1}{2c} \log(1+\frac{c}{r_0})
849: -
850: \frac{3}{2} \left[
851: (1/2 + 2m)^2(\pi/2 -\theta)
852: \right]^{1/3}. \\
853: \end{split}
854: \end{equation}
855:
856: Now, let $D\subset \R^2$ the domain of all the pairs $(m,\theta)$
857: such that
858: \begin{equation}
859: \label{eq:defD}
860: \begin{split}
861: \mathrm{inf} \{ \action_D - \action_E \st l>r_0>0, l>c>0 \} < 0.
862: \end{split}
863: \end{equation}
864:
865: The following proposition is a trivial consequence of the definition
866: of $D$.
867: \begin{propo}\label{propo:first}
868: If $(m,\theta) \in D$ then
869: there are $l$,$r_0$ and $d$ such that
870: the action of the orbit \ref{orbit1}
871: is less than the action of the Euler orbit. Therefore
872: non-homographic dihedral orbits exists for every $(m,\theta)\in D$.
873: \end{propo}
874:
875: \begin{propo}\label{propo:last}
876: The set $D$ is a non-empty open subset of $\R^2$.
877: \end{propo}
878: \begin{proof}
879: Since $\inf$ is upper semi continuous, $D$ is open. We only need to
880: show that it is not empty:
881: let $m=2$, $\theta=\pi/8$,
882: $l=1$, $r_0=0.4$, $c=0.3$.
883: Evaluating for such values an approximation of \ref{eq:difference} yields
884: $-.124390105 < 0$.
885: A candidate for the corresponding minimum can be seen
886: in figure \ref{fig:candidate}.
887: \begin{figure}
888: \begin{center}
889: \includegraphics[width=4truecm]{figs/candidate.eps}
890: \caption{$m_1=m_2=1$, $m=m_3=2$ and $\theta=\pi/8$.}
891: \label{fig:candidate}
892: \end{center}
893: \end{figure}
894: \end{proof}
895:
896:
897:
898: \begin{remark}
899: Of course propositions \ref{propo:first} and \ref{propo:last}
900: do not imply that the orbit in figure \ref{fig:3b} exists.
901: The minimization in that case has been done with a piecewise linear
902: path, which at the moment cannot be reproduced symbolically,
903: whose action is less than $\action_E$. The orbit in figure
904: \ref{fig:3b} is the same as the orbit
905: with braid $b_1^2b_2b_1^{-2}b_2$, found numerically
906: by Moore in \cite{moore}.
907: \end{remark}
908:
909:
910:
911:
912:
913:
914:
915:
916: \section{Four bodies in the plane}
917: \label{sec:4}
918: Now we analyze in the same way the situation of $4$ bodies in the plane.
919: Let again $h_1$ and $h_2$ be the reflections in $T^1$ generating $G$.
920:
921: \subsection{Four equal masses}
922: \label{subsec:4eq}
923: Assume that $m_i=1$, $i=1,\dots, 4$.
924: We want that $\sigma(h_1)$ and $\sigma(h_2)$ generate
925: a subgroup of $\Sigma_4$ that acts transitively
926: on the indexes $\{1,2,3,4\}$.
927:
928: First case: $\sigma(h_1) = (1 2)$. Then the only transitive
929: subgroup up to inner automorphism of $\Sigma_4$ is
930: given by the choice $\sigma(h_2) = (1 3)(2 4)$.
931: The action of $h_1$ on the plane $E^2$
932: cannot be a rotation (since fixes the indexes $3$ and $4$).
933: Since it cannot be trivial (otherwise
934: at $t=0$ a collision is not avoidable)
935: it needs to be a reflection along a line $l_1$.
936: Now, if $h_2$ acts by reflection along a line $l_2$, there
937: are the following sub-cases.
938: If $l_2=l_1$, then the homographic solution of a rotating
939: square can be a minimum (actually, apparently it is the minimum),
940: so we do not consider this case.
941: If $l_2$ and $l_1$ meet at an angle $\pi/4$, then
942: the action is not coercive (there is a big square with stationary
943: masses). On the other hand even
944: if the angle is different from $\pi/4$ then
945: the homographic solution of the rotating square can be a minimum,
946: hence we do not consider this case too (even if it might be possible
947: that the minimum is achieved by a non-homographic orbit).
948: % todo next
949:
950:
951: So, it is only left the case in which
952: $h_2$ acts by rotation of angle $\pi$ on $E^2$.
953: % see themis
954: % and tt4 max=8
955: Since $\Chi^{h_1} \cap \Chi^{h_2} \neq 0$,
956: by lemma \ref{lemma:coercive} the action is not coercive.
957: %\begin{comment}
958: %Homographic solutions cannot be equivariant under this action:
959: %the only central configurations that can belong to
960: %$\Chi^{h_1}$ are the square and Albouy's one-axis configuration,
961: %and the ones in $\Chi^{h_2}$ are the collinear Moulton
962: %configuration and the square. But it is easy to see that
963: %a rotation of a square in
964: %$\Chi^{h_1}$ cannot belong to $\Chi^{h_2}$,
965: %unless the ordering of the masses is changed.
966: %So the solution is not homographic. About collisions,
967: %we can apply \ref{assumption},
968: %so that we have a collision-free solution.
969: %A possible picture of the trajectories is in figure
970: %\ref{fig:4b}.
971: %The group acting is the dihedral group
972: %$D_4$ of size $8$.
973: %\begin{figure}
974: %\begin{center}
975: %\includegraphics[width=6truecm]{figs/fig_4b.eps}
976: %\caption{Four equal masses with a $D_4$ action.}
977: %\label{fig:4b}
978: %\end{center}
979: %\end{figure}
980: %\end{comment}
981:
982:
983: Now consider the second case $\sigma(h_1)=(1 2)(3 4)$.
984: Then the only choice of $\sigma(h_2)$ that has has not yet been
985: considered for a transitive action is $\sigma(h_2) = (1 3)(2 4)$.
986: This time $h_1$ can act on $E^2$ either as a reflection or as a rotation,
987: and the same holds for $h_2$.
988: If one is a reflection and one is a rotation,
989: then
990: the action is not coercive, since a sequence of increasing stationary
991: squares can have as small as possible action.
992: If both act as rotations, then the group $G$ is equal
993: to the group generated by $\sigma(h_1)$ and $\sigma(h_2)$,
994: i.e., the elementary abelian group $\ze_2^2$ of order $4$.
995: The
996: resulting symmetric orbit can be two coupled Kepler orbits,
997: hence there is no coercivity.
998: So it is left the case in which both are reflections along lines
999: $l_1$ and $l_2$ in $E^2$.
1000: If the lines $l_1=l_2$ coincide,
1001: then again it is possible to see that the functional not is coercive,
1002: by taking two symmetric Keplerian orbits that have increasing
1003: distance (i.e., by applying lemma \ref{lemma:coercive}).
1004:
1005:
1006: Otherwise, if they are orthogonal, then again it is easy to see that
1007: the functional is not
1008: coercive. If they are not orthogonal then
1009: the functional is coercive, but in the class of symmetric paths there are
1010: the homographic orbits of rotating squares.
1011: Of course the question arises whether the homographic
1012: orbits achieve the minimum or not.
1013: Numerical simulations lead to think that
1014: the minimum might be achieved by non-homographic orbits,
1015: like the one depicted in figure \ref{fig:nonhomographic}.
1016: \begin{figure}
1017: \begin{center}
1018: \includegraphics[width=4truecm]{figs/fig_nonhomographic.eps}
1019: \caption{Four equal masses with $l_1/l_2$-angle $\pi/8$.}
1020: \label{fig:nonhomographic}
1021: \end{center}
1022: \end{figure}
1023: It could be of some interest to prove some estimates
1024: like those in the previous section, to actually prove
1025: or disprove their existence. This is true also for other
1026: examples listed below, and we will not rise the question again.
1027:
1028:
1029: \subsection{Three equal masses}
1030: Assume now that there are three equal masses $m_1=m_2=m_3=1$
1031: and a fourth mass $m_4=m$. Then the subgroup
1032: of $\Sigma_4$ generated by $\sigma(h_1)$ and $\sigma(h_2)$
1033: needs to act transitively only on the set $\{1,2,3\}$.
1034: The only possibility, up to rearranging the indexes,
1035: is $\sigma(h_1)=(1 2)$ and $\sigma(h_2) = (1 3)$,
1036: like in the case of $3$ bodies.
1037: Since both fix two indexes, the actions of $h_1$ and
1038: $h_2$ on $E^2$ need to be reflections along the lines
1039: $l_1$ and $l_2$ respectively.
1040: If the lines coincide, then a rotating triangle with the mass $(4)$
1041: in the center can be the minimum.
1042: If the angle is $\pi/3$, then the problem is no longer coercive,
1043: since any constant equilateral triangle with bodies $(1)$, $(2)$ and $(3)$
1044: with $(4)$ in the center is symmetric with respect to this action.
1045: On the other hand, such a triangle when rotating at a suitable speed
1046: always belongs to the set of
1047: symmetric loops $\Lambda^G$, however the two lines intersect.
1048: We do not know whether it is a minimum in $\Lambda^G$.
1049: % todo: try with some masses...
1050: % seems to be fine is the mass is $2$ and order=5.
1051:
1052:
1053: \subsection{Two pairs of equal masses}
1054: \label{subsec:twopairs}
1055: Assume that $m_1=m_2=1$ and $m_3=m_4=m$.
1056: Since at least one from $\sigma(h_1)$ and $\sigma(h_2)$
1057: is non-trivial, we can assume that
1058: $\sigma(h_1) = (1 2)$ or $\sigma(h_1) = (1 2)(3 4)$.
1059:
1060: In the first case, $\sigma(h_1)=(1 2)$, necessarily it must be
1061: $\sigma(h_2) =(3 4)$.
1062: The only possible action of $h_1$ and $h_2$ on $E^2$
1063: is given by reflections along lines $l_1$ and $l_2$.
1064: If the lines coincide, then the functional is not coercive
1065: by \ref{lemma:coercive}.
1066: It is not
1067: coercive also if they are orthogonal: a square in increasing
1068: size can give a sequence going to infinity with bounded
1069: action.
1070: If the lines meet with an angle $\pi/q$, then it is coercive,
1071: but a rotating central configuration with the masses
1072: at the vertexes of a parallelogram belongs to $\Lambda^G$,
1073: and hence it can be the homographic minimum.
1074: Again, as above, the question arises whether the minimum
1075: is homographic or not.
1076: % todo: themis (n>5). why are they choreographies for m=1?
1077:
1078: % \subsection{General case}
1079: Consider now the second case, $\sigma(h_1) = (1 2)(3 4)$.
1080: There are three possibilities for $\sigma(h_2)$:
1081: up to rearranging indexes,
1082: the trivial $()$, or $(1 2)$ or $(1 2)(3 4)$.
1083: Consider $\sigma(h_2) = ()$. Then $h_2$ must act on
1084: $E^2$ by reflection along a line $l_2$.
1085: However $h_1$ acts on $E^2$, as a rotation or as a reflection,
1086: it is possible to find
1087: a rotating collinear central configuration belonging to
1088: $\Lambda^G$.
1089: % is it the minimum? two cases left...
1090: So we consider the next case, $\sigma(h_2) = (1 2)$.
1091: Again $h_2$ needs to act as a reflection along a line $l_2$,
1092: and rotating collinear configurations now cannot belong to $\Lambda^G$.
1093: If $h_1$ acts by rotation, then a rotating parallelogram belongs
1094: to $\Lambda^G$, hence we consider only the case of $h_1$ acting
1095: by reflection along a line $l_1$.
1096: If the two lines coincide or are orthogonal, then the action functional is not
1097: coercive. Otherwise it is coercive, and hence
1098: there is a minimum, which is collisionless due to theorem
1099: \ref{assumption}.
1100: Can it be homographic? No: at the time $t=0$ (i.e., the time in $T^1$
1101: fixed by $h_1$) the lines through $(1)\mbox{--}(2)$ and
1102: $(3)\mbox{--}(4)$ are parallel (both orthogonal to $l_1$),
1103: while at time $1$ (i.e., the time in $T^1$ fixed by $h_2$)
1104: they are orthogonal.
1105: Such orbits can be described as follows: two masses in a
1106: roughly Keplerian orbit outside, and two masses in a
1107: retrograde approximate Keplerian orbit inside,
1108: like in figure
1109: \ref{fig:rough}.
1110: % todo: themis. try different things: they are very nice!
1111: \begin{figure}
1112: \begin{center}
1113: \includegraphics[width=4truecm]{figs/fig_rough.eps}
1114: \caption{$m_1=m_2=1$ and $m_3=m_4=4$, with $D_3$-symmetry.}
1115: \label{fig:rough}
1116: \end{center}
1117: \end{figure}
1118:
1119: Now it is left the case $\sigma(h_1) = \sigma(h_2) = (1 2)(3 4)$.
1120: Since we are assuming the action of $G$ on $T^1$ to be faithfully
1121: dihedral, $h_1$ and $h_2$ must act on $E^2$ in different ways,
1122: so that at least one of them acts as a reflection.
1123: Let us assume that $h_1$ acts by reflecting along a line $l_1$.
1124: If $h_2$ acts by rotation, then the functional is not coercive;
1125: if it acts as a reflection along a line $l_2\neq l_1$,
1126: then homographic orbits belong to $\Lambda^G$,
1127: so that this case is of minor interest.
1128:
1129:
1130: \section{Orbits with an additional central symmetry}
1131: Consider the symmetries in the previous sections.
1132: If we can find an element $\sigma_3$ in $\Sigma_n$ of order $2$
1133: that fixes at most one index and commutes
1134: with $\sigma(h_1)$ and $\sigma(h_2)$,
1135: we can consider the following additional central symmetry:
1136: the symmetry group is $G\times \ze_2$ (where $G$ is the group
1137: in the example under consideration), where the direct
1138: factor $\ze_2$ is generated by $h_3$; this element
1139: acts trivially on $T^1$, acts as a rotation of angle $\pi$
1140: in the plane $E^2$, and is sent to $\sigma_3$ by
1141: the homomorphism $\sigma$.
1142: This means that for every $t\in T^1$ the configuration
1143: at time $t$ is in $\Chi^{h_3}$, that is, bodies
1144: (with the same mass) in the same cycle in $\sigma_3$
1145: are symmetric with respect $0\in E^2$, while the possible
1146: body with index fixed by $\sigma_3$ lies in $0\in E^2$.
1147: If $n=3$, then such orbits are trivial, since
1148: they need to be always collinear.
1149: So consider the case $n=4$. We can analyze the cases
1150: exploited in section \ref{sec:4}
1151: to see when these conditions are fulfilled,
1152: and if the additional central symmetry $h_3$ yields
1153: non-homographic orbits.
1154: We omit the details of this case-by-case analysis, and
1155: exhibit only a particular family of dihedral orbits.
1156:
1157: \subsection{Four equal masses}
1158: \label{subsec:4eqp}
1159: In section \ref{subsec:4eq}
1160: consider the case of
1161: $\sigma(h_1) = (1 2)(3 4)$
1162: and $\sigma(h_2) = (1 3)(2 4)$,
1163: where $h_1$ and $h_2$ act on $E^2$ by reflection
1164: along different lines $l_1$ and $l_2$.
1165: If $l_1$ and $l_2$ are orthogonal, then
1166: the functional is not coercive, even
1167: when we add the additional symmetry $\sigma_3 = (1 2)(3 4)$.
1168: %\begin{comment}
1169: %as explained above.
1170: %Clearly there are no homographic solutions in this class,
1171: %so that we have found a dihedral periodic orbit (see figure
1172: %\ref{fig:another}).
1173: %\begin{figure}
1174: %\begin{center}
1175: %\includegraphics[width=6truecm]{figs/fig_another.eps}
1176: %\caption{$m_i=1$, with $(D_2\times \ze_2)$-symmetry}
1177: %\label{fig:another}
1178: %\end{center}
1179: %\end{figure}
1180: %\end{comment}
1181:
1182: In case $l_1$ and $l_2$ meet at an angle $\pi/q$, with
1183: $q>2$, we can avoid the homographic solution
1184: again by the same additional symmetry $\sigma_3$.
1185: The group $G$ acts faithfully on $T^1$ and
1186: is equal to the dihedral group of order $2k$,
1187: where $k$ is the least common multiple of $2$ and $q$.
1188: Thus we obtain an infinite family of
1189: periodic orbits in the $4$-body problem with equal masses.
1190: We can see the case
1191: $q=4$ in figure
1192: \ref{fig:mac}
1193: \begin{figure}
1194: \begin{center}
1195: \includegraphics[width=4truecm]{figs/fig_mac.eps}
1196: \caption{$m_i=1$, with $(D_4\times \ze_2)$-symmetry}
1197: \label{fig:mac}
1198: \end{center}
1199: \end{figure}
1200: and
1201: $q=3$ in figure
1202: \ref{fig:caseq6}.
1203: \begin{figure}
1204: \begin{center}
1205: \includegraphics[width=4truecm]{figs/fig_caseq6.eps}
1206: \caption{$m_i=1$ with $(D_6\times \ze_2)$-symmetry.}
1207: \label{fig:caseq6}
1208: \end{center}
1209: \end{figure}
1210: The orbit of figure \ref{fig:mac} is very likely the orbit found
1211: by Chen \cite{chen}.
1212:
1213:
1214:
1215:
1216:
1217:
1218:
1219: \subsection{Two pairs of equal masses}
1220: \label{subsec:22eqp}
1221: The periodic solutions of section
1222: \ref{subsec:twopairs}
1223: can be endowed with the additional symmetry given by
1224: $\sigma_3=(1 2)(3 4)$. So we consider
1225: $\sigma(h_1) = (1 2)(3 4)$,
1226: $\sigma(h_2) = (1 2)$,
1227: the action of $h_1$ and $h_2$ on $E^2$ is by reflection
1228: along two lines $l_1$ and $l_2$ that intersect
1229: at an angle $\pi/q$ with $q>2$.
1230: We can see the case $m_1=m_2=1$, $m_3=m_4=2$
1231: and $q=4$ in figure \ref{fig:arch}.
1232: \begin{figure}
1233: \begin{center}
1234: \includegraphics[width=4truecm]{figs/fig_arch.eps}
1235: \caption{$m_1=m_2=1$, $m_3=m_4=2$, with $(D_4\times \ze_2)$-symmetry.}
1236: \label{fig:arch}
1237: \end{center}
1238:
1239: \end{figure}
1240:
1241:
1242:
1243:
1244:
1245:
1246:
1247: \section{Some plane choreographies for $n>3$ bodies}
1248: \label{sec:choreo}
1249: As shown in \cite{chenciner},
1250: it is not difficult to generalize the eight-shaped choreography
1251: of Montgomery-Chenciner to the case of $n>3$ odd bodies with
1252: equal masses.
1253: Consider
1254: the following permutations on $\{1,2,\dots,n\}$:
1255: \begin{equation}
1256: \begin{split}
1257: \sigma_1\from i \to n-i \mod n\\
1258: \sigma_2\from i \to n-i+1 \mod n,\\
1259: \end{split}
1260: \end{equation}
1261: where we understand that $0 \equiv n \mod n$.
1262: That is, for $n=3$ we have
1263: $\sigma_1 = (1 2)$, $\sigma_2=(1 3)$;
1264: for $n=5$ we have
1265: $\sigma_1=(1 4)(2 3)$ and
1266: $\sigma_2=(1 5)(2 4)$;
1267: for $n=7$ it is
1268: $\sigma_1=(1 6)(2 5)(3 4)$
1269: and $\sigma_2=(1 7)(2 6)(3 5)$.
1270: The product $\sigma_1 \sigma_2$ ($\sigma_2 \sigma_1$ in functional
1271: notation) sends $i$ to $i+1 \mod n$, i.e.,
1272: $\sigma_1 \sigma_2$ is the cyclic permutation
1273: $(1 2 \dots n)$.
1274: Thus the subgroup generated by $\sigma_1$ and $\sigma_2$
1275: is a dihedral group of order $2n$. We can define
1276: $\sigma_1$ and $\sigma_2$ in a geometrical way:
1277: consider a regular $n$-gon with consecutive vertices
1278: $(1),(2), \dots, (n)$. Then $\sigma_1$ is the reflection
1279: with axis through the vertex $(n)$ and $\sigma_2$ the
1280: reflection fixing the vertex $(i)$ with $i=(n+1)/2$.
1281: We will of course choose $\sigma(h_i) = \sigma_i$,
1282: with $i=1,2$, where $h_i$ are the generators
1283: of the symmetry group $G$ as above (it is not assumed
1284: that the homomorphism $\sigma$ is a monomorphism;
1285: it will depend on the choice of the action of $h_1$ add $h_2$
1286: on $E^2$).
1287:
1288: Hence, it is only left
1289: to choose the action of $h_1$ and $h_2$
1290: on the plane $E^2$.
1291: Again, $h_1$ and $h_2$ need to be of order two,
1292: hence they can be either rotations of angle $\pi$
1293: or reflections along lines $l_1$ or $l_2$.
1294: Let us consider first the case in which $h_1$ and $h_2$
1295: are both the rotation of angle $\pi$.
1296: The group $G$ is therefore the dihedral group
1297: of order $2n$.
1298: By
1299: lemma \ref{lemma:coercive}
1300: the functional is coercive, hence
1301: it attains the minimum;
1302: by \ref{assumption},
1303: the minimum is collision-free.
1304: It is only left to show that this minimum
1305: is not homographic.
1306: This is easy, since at time $t=0$
1307: the body $(n)$ is in the origin $0\in E^2$,
1308: while at time $T/(2n)$
1309: the body in the origin is the body $(i)$ with $i=(n+1)/2$
1310: (figure \ref{fig:chor5}).
1311: % todo: figures for n=5,7,9
1312: \begin{figure}
1313: \begin{center}
1314: \includegraphics[width=4truecm]{figs/fig_chor5.eps}
1315: \caption{$5$-body choreography}
1316: \label{fig:chor5}
1317: \end{center}
1318: \end{figure}
1319:
1320: If $h_1$ acts by rotating and $h_2$ by a reflection,
1321: then the same results hold,
1322: only this time the group $G$ is the dihedral group
1323: of order $4n$.
1324: Again, we have a choreography with $n$ equal masses.
1325: It might be interesting to see whether
1326: it is the same as the choreography with $h_i$ rotations or not
1327: (we have tried some simulations, obtaining something like
1328: figure \ref{fig:chor5bis}, which is very similar to
1329: figure \ref{fig:chor5}).
1330: % todo: figures for n=5,7,9
1331: \begin{figure}
1332: \begin{center}
1333: \includegraphics[width=4truecm]{figs/fig_chor5bis.eps}
1334: \caption{$5$-body choreography}
1335: \label{fig:chor5bis}
1336: \end{center}
1337: \end{figure}
1338:
1339:
1340: In case $h_1$ and $h_2$ both act
1341: by reflections, then clearly a rotating regular $n$-gon
1342: belongs to $\Lambda^G$ (if the functional is coercive),
1343: so that the minimum can be homographic and we cannot
1344: apply the previous results.
1345:
1346:
1347:
1348: % \section{Periodic orbits in the space}
1349:
1350: \section{Remarks}
1351: % generalizations with $n>3,4$
1352: \begin{remark}
1353: The methods used in the paper for proving the existence
1354: of non-homographic periodic orbits are quite simple,
1355: once the collision assumption is proved,
1356: and can be extended in a straightforward way to
1357: the case of $n>4$ bodies and non--dihedral groups.
1358: A full classification of non-degenerate actions of dihedral groups
1359: or abelian extensions of dihedral groups is not difficult,
1360: and will be the content of a forthcoming paper.
1361: \end{remark}
1362:
1363: % 3dimension
1364: \begin{remark}
1365: So far, in this note we have considered explicitly
1366: only planar periodic orbits. This is not a serious restriction,
1367: since the only change needed to deal with the case $E^3$
1368: of non-planar periodic orbit is to add a one-dimensional irreducible
1369: representation of $G$ to the representation $\rho$ (there are
1370: no $3$-dimensional irreducible representation of the dihedral group).
1371: The known periodic orbits in the space (such as
1372: the Chenciner--Venturelli ``hip--hop'' orbit \cite{chenven})
1373: obtained by symmetry constraints use a $3$-dimensional representation
1374: which can be decomposed into an irreducible $2$-dimensional
1375: representation and a $1$-dimensional representation.
1376:
1377: In fact, the method is quite simple. Consider a periodic orbit
1378: arising from the method given above, with generators $h_1$
1379: and $h_2$ of $G$. Then $h_1$ and $h_2$ can act on $E^2\subset E^3$
1380: by this action, and on the third
1381: orthonormal coordinate of $E^3$ as $\pm 1$.
1382: If they act both trivially, then the problem will not be coercive.
1383: Otherwise, we can have three choices ($(+,-)$,$(-,+)$ and $(-,-)$)
1384: that will yield non-degenerate coercive actions on $\Chi$.
1385: In some cases $(+,-)$ and $(-,+)$ will yield the same
1386: periodic orbit, up to a time shift, but in general we will
1387: get three periodic orbits in the space.
1388: Numerical simulations can be done exactly as in the planar case:
1389: we obtained some interesting analogues of the ``hip--hop''
1390: orbit.
1391: \end{remark}
1392: % software : which program:
1393: \begin{remark}
1394: The program used for the simulations is rather simple.
1395: We consider a PL discretization of the loops, and
1396: so we obtain a finite dimensional space $\Lambda$.
1397: Then by relaxation dynamics on $\Lambda^G$ (that is,
1398: a very simple gradient method) and a
1399: random method for avoiding poor progress
1400: (that is, we restart the relaxation process after a
1401: random small variation within $\Lambda^G$ if the progress is not good,
1402: until it is apparent that the program is in a local minimum)
1403: we obtain an approximation of the minimum.
1404: Now we can compare the action functional on such a path (that can be computed
1405: with a reasonable precision), and compare with other
1406: known values (like homographic solutions).
1407: The language used was FORTRAN 95 with double precision arithmetic,
1408: and the NETLIB SLATEC
1409: library for the ODE solver and error-handling routines.
1410: The figures are produced with GNUPLOT run on the raw data files.
1411: \end{remark}
1412: % todo.
1413: % stability
1414: % optimization theory applications (quote the guy in Princeton).
1415: % find in the net other people and quote everybody
1416: \begin{remark}
1417: This is a very preliminary report. Not only the collision
1418: assumption \ref{assumption} is still present, but also the program used
1419: is quite bad designed and
1420: has a very poor performance.
1421: While writing the program I was more
1422: concerned about flexibility, robustness and simplicity, than performance
1423: or very good approximation of the solutions.
1424: Thus the algorithm is very slow
1425: and does not give a very good approximation of the orbits.
1426: Moreover, I did not compute the linear stability of the orbits,
1427: nor I used the more efficient approaches of variational
1428: optimization techniques in symmetric periodic problems,
1429: available in the literature.
1430: %The Glass Bead Game
1431: %Das Glasperlenspiel
1432: \end{remark}
1433:
1434:
1435:
1436: %===================================================================
1437: % \bibliographystyle{abbrvnat}
1438: % \bibliography{bibliog}
1439:
1440: % \begin{comment}
1441: \def\cfudot#1{\ifmmode\setbox7\hbox{$\accent"5E#1$}\else
1442: \setbox7\hbox{\accent"5E#1}\penalty 10000\relax\fi\raise 1\ht7
1443: \hbox{\raise.1ex\hbox to 1\wd7{\hss.\hss}}\penalty 10000 \hskip-1\wd7\penalty
1444: 10000\box7} \def\cprime{$'$} \def\cprime{$'$} \def\cprime{$'$}
1445: \def\cprime{$'$}
1446: \begin{thebibliography}{10}
1447:
1448: \bibitem{ambbad}
1449: {\sc Ambrosetti, A., and Badiale, M.}
1450: \newblock Homoclinics: {P}oincar\'e-{M}elnikov type results via a variational
1451: approach.
1452: \newblock {\em Ann. Inst. H. Poincar\'e Anal. Non Lin\'eaire 15}, 2 (1998),
1453: 233--252.
1454:
1455: \bibitem{ambrocoti}
1456: {\sc Ambrosetti, A., and Coti~Zelati, V.}
1457: \newblock {\em Periodic solutions of singular {L}agrangian systems}.
1458: \newblock Birkh\"auser Boston Inc., Boston, MA, 1993.
1459:
1460: \bibitem{ambrosetticoti}
1461: {\sc Ambrosetti, A., and Coti~Zelati, V.}
1462: \newblock Non-collision periodic solutions for a class of symmetric $3$-body
1463: type problems.
1464: \newblock {\em Topol. Methods Nonlinear Anal. 3}, 2 (1994), 197--207.
1465:
1466: \bibitem{argate}
1467: {\sc Arioli, G., Gazzola, F., and Terracini, S.}
1468: \newblock Minimization properties of {H}ill's orbits and applications to some
1469: ${N}$-body problems.
1470: \newblock {\em Ann. Inst. H. Poincar\'e Anal. Non Lin\'eaire 17}, 5 (2000),
1471: 617--650.
1472:
1473: \bibitem{bessi}
1474: {\sc Bessi, U., and Coti~Zelati, V.}
1475: \newblock Symmetries and noncollision closed orbits for planar ${N}$-body-type
1476: problems.
1477: \newblock {\em Nonlinear Anal. 16}, 6 (1991), 587--598.
1478:
1479: \bibitem{chen}
1480: {\sc Chen, K.-C.}
1481: \newblock Action-minimizing orbits in the parallelogram four-body problem with
1482: equal masses.
1483: \newblock {\em Arch. Ration. Mech. Anal. 158}, 4 (2001), 293--318.
1484:
1485: \bibitem{chenciner}
1486: {\sc Chenciner, A.}
1487: \newblock Action minimizing periodic orbits in the {N}ewtonian $n$-body
1488: problem.
1489: \newblock In {\em Celestial mechanics (Evanston, IL, 1999)}. Amer. Math. Soc.,
1490: Providence, RI, 2002, pp.~71--90.
1491:
1492: \bibitem{monchen}
1493: {\sc Chenciner, A., and Montgomery, R.}
1494: \newblock A remarkable periodic solution of the three-body problem in the case
1495: of equal masses.
1496: \newblock {\em Ann. of Math. (2) 152}, 3 (2000), 881--901.
1497:
1498: \bibitem{chenven}
1499: {\sc Chenciner, A., and Venturelli, A.}
1500: \newblock Minima de l'int\'egrale d'action du probl\`eme newtonien de 4 corps
1501: de masses \'egales dans ${\bf {r}}\sp 3$: orbites ``hip-hop''.
1502: \newblock {\em Celestial Mech. Dynam. Astronom. 77}, 2 (2000), 139--152 (2001).
1503:
1504: \bibitem{mate93a}
1505: {\sc Majer, P., and Terracini, S.}
1506: \newblock Periodic solutions to some $n$-body type problems: the fixed energy
1507: case.
1508: \newblock {\em Duke Math. J. 69}, 3 (1993), 683--697.
1509:
1510: \bibitem{mate93}
1511: {\sc Majer, P., and Terracini, S.}
1512: \newblock Periodic solutions to some problems of $n$-body type.
1513: \newblock {\em Arch. Rational Mech. Anal. 124}, 4 (1993), 381--404.
1514:
1515: \bibitem{mate}
1516: {\sc Majer, P., and Terracini, S.}
1517: \newblock Multiple periodic solutions to some $n$-body type problems via a
1518: collision index.
1519: \newblock In {\em Variational methods in nonlinear analysis (Erice, 1992)}.
1520: Gordon and Breach, Basel, 1995, pp.~245--262.
1521:
1522: \bibitem{mate95}
1523: {\sc Majer, P., and Terracini, S.}
1524: \newblock On the existence of infinitely many periodic solutions to some
1525: problems of $n$-body type.
1526: \newblock {\em Comm. Pure Appl. Math. 48}, 4 (1995), 449--470.
1527:
1528: \bibitem{moore}
1529: {\sc Moore, C.}
1530: \newblock Braids in classical dynamics.
1531: \newblock {\em Phys. Rev. Lett. 70}, 24 (1993), 3675--3679.
1532:
1533: \bibitem{rate}
1534: {\sc Ramos, M., and Terracini, S.}
1535: \newblock Noncollision periodic solutions to some singular dynamical systems
1536: with very weak forces.
1537: \newblock {\em J. Differential Equations 118}, 1 (1995), 121--152.
1538:
1539: \bibitem{setate}
1540: {\sc Serra, E., Tarallo, M., and Terracini, S.}
1541: \newblock On the existence of homoclinic solutions for almost periodic second
1542: order systems.
1543: \newblock {\em Ann. Inst. H. Poincar\'e Anal. Non Lin\'eaire 13}, 6 (1996),
1544: 783--812.
1545:
1546: \bibitem{serter1}
1547: {\sc Serra, E., and Terracini, S.}
1548: \newblock Collisionless periodic solutions to some three-body problems.
1549: \newblock {\em Arch. Rational Mech. Anal. 120}, 4 (1992), 305--325.
1550:
1551: \bibitem{serter2}
1552: {\sc Serra, E., and Terracini, S.}
1553: \newblock Noncollision solutions to some singular minimization problems with
1554: {K}eplerian-like potentials.
1555: \newblock {\em Nonlinear Anal. 22}, 1 (1994), 45--62.
1556:
1557: \bibitem{andrea}
1558: {\sc Venturelli, A.}
1559: \newblock Une caract\'erisation variationnelle des solutions de {L}agrange du
1560: probl\`eme plan des trois corps.
1561: \newblock {\em C. R. Acad. Sci. Paris S\'er. I Math. 332}, 7 (2001), 641--644.
1562:
1563: \end{thebibliography}
1564: %\end{comment}
1565: %===================================================================
1566: %\epigraph{{\itshape If you understand, things are just as they are;
1567: %if you do not understand, things are just as they are.}}{(Zen proverb)}
1568:
1569:
1570:
1571:
1572: \end{document}
1573: %===================================================================
1574:
1575: