1: %%%

2: %%%%Revisited version for nonlinearity

3: %%%%%%%%%%%%

4: % version: 1.0

5: %===================================================================

6: \documentclass[12pt,oneside,a4paper,leqno,final]{amsart}

7: %===================================================================

8: %% required packages

9: % \usepackage[cmtip,arrow,line,frame,all]{xy}

10: \usepackage[all,cmtip]{xy}

11: %% \xyoption{all}

12: \CompileMatrices

13: 

14: \newdir{ >}{!/10pt/@{ }*@{>}}

15: \newdir{< }{!/10pt/@{ }*@{<}}

16: 

17: 

18: % \usepackage{geometry}\geometry{a4paper,scale={0.72,0.80}}

19: \usepackage{graphicx}

20: \usepackage{amsfonts,amssymb}

21: \usepackage[centertags]{amsmath}

22: %%_ \usepackage{pifont}

23: %% \usepackage{newcent}

24: %% \usepackage{cmbright}

25: 

26: %%

27: \usepackage{amsthm} %[2000/10/26]

28: \newcounter{theorems}

29: \renewcommand{\thetheorems}{\Alph{theorems}}

30: \theoremstyle{plain}

31: \newtheorem{theoAA}[theorems]{Theorem}

32: 

33: %% theorem environments

34: \swapnumbers

35: \newcounter{lemma}

36: \numberwithin{equation}{section}

37: 

38: \newtheoremstyle{par}% name

39:      {\topsep}%      Space above (it was 0pt!!!)

40:      {\topsep}%      Space below (it was 0pt!!!)

41:      {\itshape}%         Body font

42:      {}%         Indent amount (empty = no indent, \parindent = para indent)

43:      {\bfseries}% Thm head font

44:      {}%        Punctuation after thm head

45:      {.5em}%     Space after thm head: " " = normal interword space;

46:                % \newline = linebreak

47:      {}%         Thm head spec (can be left empty, meaning `normal')

48: 

49: \newtheoremstyle{parrm}% name

50:      {\topsep}%      Space above (it was 0pt!!!)

51:      {\topsep}%      Space below (it was 0pt!!!)

52:      {\normalfont}%         Body font

53:      {}%         Indent amount (empty = no indent, \parindent = para indent)

54:      {\itshape}% Thm head font

55:      {}%        Punctuation after thm head

56:      {.5em}%     Space after thm head: " " = normal interword space;

57:                % \newline = linebreak

58:      {}%         Thm head spec (can be left empty, meaning `normal')

59: 

60: 

61: 

62: %\theoremstyle{plain}

63: %\numberwithin{equation}{section}

64: %%\newtheorem{lemma}[equation]{Lemma}

65: %\newtheorem{theo}[equation]{Theorem}

66: %% \newtheorem{propo}[equation]{Proposition}

67: %\newtheorem{coro}[equation]{Corollary}

68: %

69: %

70: %\theoremstyle{definition}

71: %\newtheorem{defi}[equation]{Definition}

72: %\newtheorem{example}[equation]{Example}

73: %\newtheorem{hypo}[equation]{Hypothesis}

74: %\newtheorem{property}[equation]{Property}

75: %

76: %\theoremstyle{remark}

77: %\newtheorem{remark}[equation]{Remark}

78: %

79: %\theoremstyle{par}

80: %\newtheorem{nr}[equation]{}

81: %\newtheorem{lemma}[equation]{}

82: %\newtheorem{propo}[equation]{}

83: %

84: %\theoremstyle{parrm}

85: %\newtheorem{rem}[equation]{}

86: 

87: 

88: %\newtheorem{mainthm}{\sc Theorem}           % numbered absolutely

89: \newtheorem{theo}{\sc Theorem}[section]      % numbered within each section

90: \newtheorem{coro}[theo]{\sc Corollary}        % numbered along with Theorem

91: \newtheorem{nr}[theo]{\sc Lemma}            % numbered along with Theorem

92: \newtheorem{propo}[theo]{\sc Proposition}     % numbered along with Theorem

93: %\newtheorem{add}[thm]{\sc Addendum}         % numbered along with Theorem

94: \newtheorem{defi}[theo]{\sc Definition}      % numbered along with Theorem

95: \newtheorem{remark}[theo]{\sc Remark}           % numbered along with Theorem

96: \newtheorem{example}[theo]{\sc Example}           % numbered along with Theorem

97: \newtheorem{hypo}[theo]{\sc Hypothesis}    % numbered along with Theorem

98: %\newtheorem{ass}[thm]{\sc Assumption}       % numbered along with Theorem

99: \newtheorem{property}[theo]{\sc Property}        % numbered along with Theorem

100: %\newtheorem{conj}[thm]{\sc Conjecture}      % numbered along with Theorem

101: %\newtheorem{convention}[thm]{\sc Convention}      % numbered along with Theorem

102: 

103: 

104: 

105: 

106: 

107: 

108: 

109: 

110: 

111: 

112: 

113: 

114: 

115: 

116: 

117: %% bibliography

118: % \bibliographystyle{acm}

119: % \bibliographystyle{abbrvnat}

120: %% \bibliographystyle{alpha}

121: 

122: 

123: %% page style

124: \renewcommand{\qedsymbol}{\emph{q.e.d.\/}}

125: 

126: \makeatletter

127: \def\tagform@#1{\maketag@@@{\ignorespaces#1\unskip\@@italiccorr}}

128: \renewcommand{\theequation}{(\arabic{section}.\arabic{equation}\/)}

129: %% \renewcommand{\labelenumi}{$(\roman{enumi})$}

130: \renewcommand{\labelenumi}{\theenumi}

131: \renewcommand{\theenumi}{{\it(\roman{enumi})}}

132: \makeatother

133: 

134: 

135: % newtheorem*{}{} %unnumbered

136: %===================================================================

137: 

138: 

139: 

140: 

141: 

142: \bibliographystyle{acm}

143: % \usepackage{showkeys}

144: 

145: % \usepackage{showlabels}

146: \usepackage{paralist}

147: \usepackage{comment}

148: \usepackage{subfigure}

149: %===================================================================

150: \newcommand{\RR}{\mathbb{R}}

151: \newcommand{\CC}{\mathbb{C}}

152: \newcommand{\ZZ}{\mathbb{Z}}

153: 

154: \newcommand{\from}{\colon}

155: 

156: \newcommand{\C}{\mathbb C}

157: \newcommand{\T}{\mathbb T}

158: \newcommand{\N}{I\!\!N}

159: \newcommand{\R}{\mathbb R}

160: \newcommand{\Z}{\mathbb Z}

161: 

162: \newcommand{\aaccent}{\`a}

163: \newcommand{\Ker}{\mathrm{Ker}}

164: \newcommand{\sgn}{\mathrm{sign}}

165: \newcommand{\Bsym}{\mathrm B_{\mathrm{sym}}}

166: \newcommand{\Spl}{\mathrm{Sp}}

167: \newcommand{\spl}{\mathfrak{sp}}

168: \newcommand{\Gl}{\mathrm{Gl}}

169: \newcommand{\X}{\mathcal{X}}

170: %% \newcommand{\norm}[1]{\parallel #1 \parallel}

171: \newcommand{\Hess}{\mathrm{Hess}}

172: \newcommand{\grad}{\mathrm{grad}}

173: \newcommand{\diag}{\mathrm{diag}}

174: \newcommand{\n}[1]{{\bf #1}}

175: \newcommand{\planare}{\mathrm{Pl}}

176: \newcommand{\prisma}{\mathrm{Pr}}

177: \newcommand{\antiprisma}{\mathrm{APr}}

178: \newcommand{\simmetrico}{\mathrm{sym}}

179: 

180: \newcommand{\todo}[1]{\fbox{\large ** TODO #1 **}}

181: 

182: % \newcommand{\fromto}{\rightsquigarrow}

183: \newcommand{\fromto}{\leadsto}

184: %% looparrowright

185: 

186: %%%==========================================================================

187: \usepackage{bm}

188: \newcommand{\simbolovettore}[1]{{\boldsymbol{#1}}}

189: \newcommand{\va}{\simbolovettore{a}}

190: \newcommand{\vb}{\simbolovettore{b}}

191: \newcommand{\vc}{\simbolovettore{c}}

192: \newcommand{\vd}{\simbolovettore{d}}

193: \newcommand{\ve}{\simbolovettore{e}}

194: \newcommand{\vf}{\simbolovettore{f}}

195: \newcommand{\vg}{\simbolovettore{g}}

196: \newcommand{\vh}{\simbolovettore{h}}

197: \newcommand{\vi}{\simbolovettore{i}}

198: \newcommand{\vj}{\simbolovettore{j}}

199: \newcommand{\vk}{\simbolovettore{k}}

200: \newcommand{\vl}{\simbolovettore{l}}

201: \newcommand{\vm}{\simbolovettore{m}}

202: \newcommand{\vn}{\simbolovettore{n}}

203: \newcommand{\vo}{\simbolovettore{o}}

204: \newcommand{\vp}{\simbolovettore{p}}

205: \newcommand{\vq}{\simbolovettore{q}}

206: \newcommand{\vr}{\simbolovettore{r}}

207: \newcommand{\vs}{\simbolovettore{s}}

208: \newcommand{\vt}{\simbolovettore{t}}

209: \newcommand{\vu}{\simbolovettore{u}}

210: \newcommand{\vv}{\simbolovettore{v}}

211: \newcommand{\vw}{\simbolovettore{w}}

212: \newcommand{\vx}{\simbolovettore{x}}

213: \newcommand{\vy}{\simbolovettore{y}}

214: \newcommand{\vz}{\simbolovettore{z}}

215: %% \newcommand{\zero}{\simbolovettore{o0}}

216: \newcommand{\zero}{\boldsymbol{0}}

217: 

218: 

219: 

220: %---------------------------------

221: \newcommand{\norm}[1]{\protect\left\protect\Vert\protect#1\protect\right\protect\Vert}

222: \newcommand{\sqnorm}[1]{\langle M #1, #1 \rangle}

223: 

224: \newcommand{\collisionmanifold}{\includegraphics{figs/collisionmanifold}\;}

225: \newcommand{\parabolicmanifold}{\ensuremath{P}}

226: 

227: \newcommand{\perron}{\widehat P}

228: 

229: 

230: 

231: 

232: 

233: 

234: \usepackage{multirow,rotating}

235: 

236: 

237: %%%%%%%%%%%%%%===================================================

238: % \usepackage{pstricks}

239: % \usepackage{fancybox}

240: % \usepackage{pst-grad}

241: % %

242: % \newlength{\provalen}

243: % \setlength{\provalen}{\paperwidth}

244: % \addtolength{\provalen}{-4cm}

245: % %

246: % \newcommand{\erinni}{%

247: % \fancyput*(\provalen,1in){%

248: % \psframe[linestyle=none,fillstyle=solid,fillcolor=yellow](4cm,-\paperheight)

249: % }

250: % }

251: % \DeclareGraphicsExtensions{.eps}

252: %

253: 

254: 

255: %===================================================================

256: %===================================================================

257: %===================================================================

258: \begin{document}

259: \pagenumbering{arabic}

260: 

261: 

262: \title{%

263: On the dihedral $n$-body problem }

264: 

265: \author{Davide L.~Ferrario and Alessandro Portaluri}

266: 

267: 

268: \date{%

269: \today}

270: 

271: 

272: \begin{abstract}

273: Consider $n=2l\geq 4$ point particles with equal masses in space,

274: subject to the following symmetry constraint: at each instant they

275: form an orbit of the dihedral group $D_l$, where $D_l$ is the group

276: of order $2l$ generated by two rotations of angle $\pi$ around two

277: secant lines in space meeting at an angle of $\pi/l$. By adding a

278: homogeneous potential of degree $-\alpha$ for $\alpha \in (0,2)$

279: (which recovers the gravitational Newtonian potential), one finds a

280: special $n$-body problem with three degrees of freedom, which is a

281: kind of generalisation of Devaney isosceles problem, in which all

282: orbits have zero angular momentum. In the paper we find all the

283: central configurations and we compute the dimension of the

284: stable/unstable manifolds.

285: 

286: \noindent {\em MSC Subject Class\/}: Primary 70F10; Secondary 37C80.

287: \vspace{0.5truecm}

288: 

289: \noindent {\em Keywords\/}: Dihedral $n$-body problem, McGehee

290: coordinates, central configurations.

291: \end{abstract}

292: 

293: \maketitle

294: 

295: 

296: %%%=========================================================================

297: \section{Introduction}

298: \label{sec:intro}

299: 

300: The goal of this paper is to compute all the central configurations

301: and the dimension of the stable/unstable manifolds for the dihedral

302: symmetric $n$-body problem in space under the action of a

303: homogeneous potential of degree $-\alpha$. For the Newtonian

304: potential this problem is a kind of generalisation of Devaney planar

305: isosceles three body problem \cite{Devaney80,Devaney81}.%%, following

306: %%Moeckel's approach to the study of the three body problem in space

307: %%\cite{Moeckel81,Moeckel83}.

308: The dihedral problem is a special case

309: of the full $n$-body problem which reduces to a Hamiltonian system

310: with three degrees of freedom. Briefly, one takes $n=2l\geq 4$ equal

311: masses whose initial position and velocity are symmetric with

312: respect to the dihedral group of rotations $D_l\subset SO(3)$. So

313: the masses form a (possibly degenerate and non-regular) antiprism in

314: space (and they are vertices of  two symmetric parallel $l$-gons).

315: Because of the symmetry of the problem, the masses will remain in

316: such a configuration for all time. Hence we have a system with only

317: three degrees of freedom. For $l=2$, the four bodies are at vertices

318: of a tetrahedron, and the problem has been studied in a series of

319: papers by Delgado and Vidal \cite{Vidal99,DelgadoVidal99}. The main

320: tool is the use of McGehee coordinates introduced in

321: \cite{McGehee74} but

322: for a general homogeneous potential of degree $-\alpha$. %%and with a

323: %%slight change: we consider McGehee coordinates not only for studying

324: %%the behaviour of solutions passing close to a total collision, but

325: %%also for parabolic orbits connecting central configurations,

326: %%projecting the full phase space to a codimension $1$ subspace.

327: We replace the singularity due to total collapse with an invariant

328: immersed manifold in the full phase space usually called \emph{total

329: collision manifold} which is the immersion of the parabolic manifold

330: of the projected phase space. We explicitly compute all central

331: configurations for this problem and show that just three types can

332: arise: a planar regular $2l$-gon, a regular $l$-gonal prism and a

333: $l$-gonal anti-prism. %%In particular, we prove the existence of some

334: %%heteroclinic connections between these central configurations and

335: %%some asymptotic sets on the projection of the regularised flow on

336: %%the parabolic manifold. Moreover by using this flow we are able to

337: %%establish some global results on the behaviour of solutions. We

338: %%discuss the qualitative behaviour of orbits which reach or come

339: %%close to the total collision and of those orbits which start from

340: %%total collapse.

341: 

342: The motivation in order to study this kind of problem is twofold.

343: From one side this problem is difficult enough to put on evidence

344: some chaotic behaviour of the full $n$-body problem and at the same

345: time it is simple enough to carry out some explicit computations.

346: From the other side the interest in this kind of problem is due to

347: the fact that it includes a lot of other problems with two or three

348: degrees of freedom studied in the past decades. The literature is

349: quite broad and we limit ourself to quote only some closest results;

350: among the others is the tetrahedral four body problem without and

351: with rotation, studied respectively in \cite{DelgadoVidal99} and

352: \cite{Vidal99}, the rectangular four body problem studied by Sim\'o

353: and Lacomba in \cite{SimoLacomba82}. %%What makes this problem harder

354: %%than  the tetrahedral four body problem is, also, the need of

355: %%computing all  central configurations.

356: 

357: \subsection*{Acknowledgements} We are very grateful to the anonymous

358: referees for their suggestions, comments and criticism which greatly

359: improved the manuscript.

360: 

361: 

362: %%%=========================================================================

363: \section{McGehee coordinates, projections and regularisations}

364: \label{sec:mcgehhecoord}

365: 

366: Let $V = \R^d$ denote the Euclidean space of dimension $d$ and $n

367: \geq 2$ an integer. Let $0$ denote the origin $0\in\R^d$. Let $m_1,

368: \dots, m_n$ be $n$ positive numbers (which can be thought as

369: masses). The configuration space of $n$ point particles with masses

370: $m_i$ respectively and center of mass in $0$ can be identified with

371: the subspace of $V^n$ consisting of all points $\vq=(\vq_1, \dots

372: \vq_n)\in V^n$ such that $\sum_{i=1}^n m_i\vq_i=0$. Let $\n{n}$

373: denote the set $\{1, \dots, n\}$ of the first $n$ positive integers.

374: For each pair of indexes $i,j \in \n{n}$ let $\Delta_{i,j}$ denote

375: the collision set of the $i$-th and $j$-th particles $\Delta_{i,j} =

376: \{\vq \in X|\vq_i=\vq_j\}$. Let $\Delta= \cup_{i,j} \Delta_{i,j}$ be

377: the {\em collision set\/}. %%The set of collision-free configurations

378: %%is denoted by $\hat X = X \backslash \Delta$.

379: 

380: Let $X \subset V^n$ be an open cone ($\R X =X$) and let $\alpha>0$

381: be a given positive real number. We consider the potential function

382: (the opposite of the potential energy) defined by

383: \[

384: U(\vq):= \sum_{i< j} \dfrac{m_im_j}{|\vq_i - \vq_j|^{\alpha}}.

385: \]

386: %%

387: %%

388: %%Consider $n$ point particles with masses $m_i$  and positions $\vq_i

389: %%\in X$ for $i \in\n{n}$. (we assume $X$ to be open in $\RR^k$).

390: %%%Consider a point $\vq$ in the configuration space $X \subset \RR^k$

391: %%If $M$ is the diagonal matrix (with positive entries, which can be

392: %%thought as masses) and $U(\vq)$ is a positive potential,

393: If $M$ is the diagonal matrix, then Newton equations

394: \[

395: M \ddot \vq =  \dfrac{\partial U}{\partial\vq}

396: \]

397: can be written in Hamiltonian form as

398: \begin{equation}\label{eq:hamilton}

399: \left\{

400: \begin{aligned}

401: M \dot \vq &=  \vp  \\

402: \dot \vp &= \dfrac{\partial U}{\partial \vq},

403: \end{aligned}\right.

404: \end{equation}

405: where the Hamiltonian is $H =H(\vq,\vp) = \langle\dfrac{1}{2} M^{-1}

406: \vp,\vp \rangle - U(\vq)$. Then equations \ref{eq:hamilton} can be

407: written in polar coordinates by setting the mass norm in $V^n$

408: defined for every $\vq \in X$ as

409: \[

410: \norm{\vq}^2 =  \langle M \vq, \vq \rangle

411: \]

412: and suitably rescaling the momentum as follows

413: \[\begin{aligned}

414: \rho & =  \norm{\vq} \\

415: \vs &= \dfrac{\vq}{\rho}\\

416: \vz & = \rho^{\beta} \vp \qquad\qquad \text{with $\alpha = 2

417: \beta$.}

418: \end{aligned}\]

419: In these coordinates equations~\ref{eq:hamilton} can be read as

420: \begin{equation}\label{eq:mcghee1}

421: \left\{

422: \begin{aligned}

423: \rho' &  =  \langle \vz , \vs\rangle \rho \\

424: \vs' &=   M^{-1}\vz - \langle \vz,\vs\rangle  \vs  \\

425: \vz' &= \beta \langle\vz, \vs\rangle \vz + \dfrac{\partial

426: U}{\partial \vq}(\vs),

427: \end{aligned}

428: \right.

429: \end{equation}

430: where the time has been rescaled by $dt = \rho^{1+\beta} d\tau$

431: (that is, $\dfrac{d}{d\tau} = \rho^{1+\beta}\dfrac{d}{dt}$); now the

432: energy can be written as

433: \begin{equation}\label{eq:energy}

434: H = \dfrac{1}{2} \rho^{-\alpha} \langle M^{-1}\vz,\vz\rangle  -

435: \rho^{-\alpha} U(\vs) = \rho^{-\alpha} \left( \dfrac{1}{2}  \langle

436: M^{-1}\vz,\vz\rangle  - U(\vs) \right).

437: \end{equation}

438: Let $k:=dn$ and let us consider the projection $(\vq,\vp) \mapsto

439: (\vs,\vz)$ from the full phase space $X\times  \RR^k$ to the reduced

440: space $S^{k-1}\times \RR^k$ (which is the trivial $\RR^k$-bundle  on

441: the ellipsoid $S^{k-1}$)

442: \[

443: X  \times \RR^k \to S^{k-1} \times \RR^k.

444: \]

445: In McGehee coordinates it is easy to see that the flow on $X \times

446: \RR^k$ can be projected to $S^{k-1} \times \RR^k$, that is

447: \begin{equation}\label{eq:mcghee2}

448: \left\{

449: \begin{aligned}

450: \vs' &=   M^{-1}\vz - \langle \vz,\vs\rangle  \vs  \\

451: \vz' &= \beta  \langle\vz, \vs\rangle \vz + \dfrac{\partial

452: U}{\partial \vq}(\vs).

453: \end{aligned}

454: \right.

455: \end{equation}

456: Also, being $X$ a cone, it is a cone on its $(k-1)$-dimensional

457: intersection with the ellipsoid $S^{k-1}$, which we will denote

458: simply by $S= S^{k-1} \cap X$. We define the \emph{parabolic

459: manifold} as the projection of all zero-energy orbits (or,

460: equivalently, of the zero-energy submanifold of $X\times \RR^k$) in

461: $S\times \RR^k$, that is

462: \[

463: \parabolicmanifold :=

464: \{ (\vs,\vz) \in S\times \RR^k : \frac{1}{2}  \langle

465: M^{-1}\vz,\vz\rangle  = U(s) \} \subset S^{k-1} \times \RR^k.

466: \]

467: Its dimension is $\dim S + k -1 = 2k - 2$. This is also the

468: projection of McGehee total collision manifold (see

469: \cite{McGehee74,Devaney80,Moeckel81,Moeckel83}); the manifold of

470: $(\vs,\vz)$ here is not considered as embedded in the space of

471: $(\rho,\vs,\vz)$ with $\rho=0$. By the form of equation

472: \ref{eq:mcghee1}, it is easy to prove the following proposition.

473: \begin{nr}

474: Solutions of~\ref{eq:mcghee2}  in $S \times \RR^k$ are projections

475: of solutions of~\ref{eq:mcghee1}. The parabolic manifold

476: $\parabolicmanifold$ is invariant for the flow of \ref{eq:mcghee2},

477: and solutions in $\parabolicmanifold$ can be lifted to $X\times

478: \RR^k$ by integrating the equation $\rho'/\rho   =  \langle \vz ,

479: \vs\rangle $.

480: \end{nr}

481: The parabolic manifold $\parabolicmanifold$ is the boundary of the

482: $(2k-1)$-dimensional \emph{elliptic} and \emph{hyperbolic}

483: manifolds, defined as

484: \[

485: \text{Elliptic} = \{ (\vs,\vz) \in S\times \RR^k : \dfrac{1}{2}

486: \langle M^{-1}\vz,\vz\rangle  < U(s) \} \subset S^{k-1} \times

487: \RR^k.

488: \]

489: \[

490: \text{Hyperbolic} = \{ (\vs,\vz) \in S\times \RR^k : \dfrac{1}{2}

491: \langle M^{-1}\vz,\vz\rangle  > U(s) \} \subset S^{k-1} \times

492: \RR^k.

493: \]

494: They are again invariant (even if the function $\dfrac{1}{2}

495: \langle M^{-1}\vz,\vz\rangle$ is not an invariant of the flow in

496: $S^{k-1}\times \RR^k$), and correspond to projection of

497: elliptic/hyperbolic orbits (that is, orbits with negative/positive

498: energy).  In fact, any fixed-energy (negative/positive) surface is

499: homeomorphic to the elliptic/hyperbolic manifold. Given a solution

500: of \ref{eq:mcghee2} in the elliptic or hyperbolic manifolds, for

501: each energy value $h$ the lifted solutions in $X\times \RR^k$ can be

502: found simply by applying \ref{eq:energy} as

503: \begin{equation}\label{eq:energyrelation}

504: \rho^\alpha= \dfrac{\langle M^{-1}\vz,\vz\rangle  - 2U(s)}{2h}.

505: \end{equation}

506: The  parabolic manifold $\parabolicmanifold$ is fiberwise

507: homeomorphic to a trivial $(k-1)$-sphere bundle on $S\subset

508: S^{k-1}$.

509: 

510: The next change of coordinates, due to McGehee \cite{McGehee74}

511: (with a reference to Sundman \cite{sundman}), is needed for defining

512: the Sundman--Lyapunov coordinate $v$ and for the regularisation of

513: the parabolic manifold $\parabolicmanifold$. Let $v,\vw\in \RR\times

514: \RR^k$ be defined by

515: \[

516: \begin{cases}

517: v& =\langle \vz, \vs \rangle \\

518: \vw & =M^{-1} \vz - \langle \vz, \vs \rangle \vs.

519: \end{cases}

520: \]

521: Then $\vz = vM\vs +M\vw$ and  $\langle \vw, M\vs \rangle = 0$, and

522: equations \ref{eq:mcghee2} can be replaced by

523: \begin{equation}\label{eq:mcghee3}

524: \left\{

525: \begin{aligned}

526: v' &= \norm{\vw}^2 + \beta v^2 - \alpha U(\vs)\\

527: \vs' &=  \vw \\

528: \vw' &= -\norm{\vw}^2 \vs +(\beta-1) v\vw +  M^{-1}\nabla_{\vs}

529: U(\vs),

530: \end{aligned}

531: \right.

532: \end{equation}

533: where $\nabla_{\vs}$ denotes covariant derivative, i.e.~the

534: component of the gradient tangent to the inertia ellipsoid

535: $\norm{\vq}=1$:

536: \[

537: \nabla_\vs U =

538: % \dfrac{\partial U}{\partial \vq}(\vs)   -

539: %  \langle \dfrac{\partial U}{\partial\vq}(\vs),  \vs \rangle \vs

540: % =

541: \dfrac{\partial U}{\partial \vq}(\vs)    + \alpha U(\vs) M \vs.

542: \]

543: The parabolic manifold $\parabolicmanifold$ is then defined by the

544: equation

545: \[

546: v^2 + \norm{\vw}^2   =2 U(\vs).

547: \]

548: The trivial bundle $S\times \RR^k$ is simply decomposed as the sum

549: of the normal bundle $(\vs,v)$ of $S$ in $\RR^k$ and the tangent

550: bundle $TS$ (with coordinates $(\vs,\vw)$). By the first equation in

551: \ref{eq:mcghee3}

552: \[

553: v'  = \norm{\vw}^2 + \beta v^2 - \alpha U(\vs)

554:  = (1-\beta) \norm{\vw}^2 +

555:  \alpha \left( \dfrac{1}{2} ( \norm{\vw}^2 + v^2 ) - U(\vs) \right),

556: \]

557: can be deduced the well-known fact that for $0<\alpha <2$,  $v$ is a

558: Lyapunov function on the flow in the parabolic and hyperbolic

559: manifolds, and therefore the flow is dissipative (gradient-like).

560: Moreover, the \emph{equilibrium points} in \ref{eq:mcghee3} are the

561: projections of the equilibrium points of \ref{eq:mcghee1} (and the

562: projection is one-to-one in the parabolic manifold), which can be

563: found as solutions of

564: \begin{equation}\label{eq:centralconf}

565: \left\{

566: \begin{aligned}

567: v^2 & =  2 U(\vs)\\

568: \nabla_{\vs} U(\vs) & =   \zero \\

569: \vw  &=   \zero.

570: \end{aligned}

571: \right.

572: \end{equation}

573: Hence all equilibrium points belong to the parabolic manifold

574: $\parabolicmanifold$. The constant solution in a central

575: configuration $\bar \vs$ with $v^2 = 2U(\bar\vs)$ can be lifted to

576: the full space as a homotetic parabolic orbit by  integrating (back

577: to the real time coordinate)

578: \[

579: \dot \rho = \pm \rho^{-\beta}  \sqrt{ 2U(\bar \vs)} \implies

580: %

581: \rho (t) = \left( \pm(1+\beta) \sqrt{2U(\bar \vs)} t  \right) ^{

582: 1/\left( 1+\beta \right) },

583: \]

584: assuming the total collision occurs at $t=0$ (the $+$ sign yields an

585: ejection solution, the $-$ sign yields a collision solution). More

586: generally, \emph{homotetic solutions} (i.e. $\vs'=0$, with $\vs(t)

587: \equiv \bar\vs$) can be found in the hyperbolic and elliptic

588: manifolds by setting in equations \ref{eq:mcghee3} $\vs' = \vw = 0$,

589: and therefore by integrating the single equation

590: \[

591: v' = \beta v^2 - \alpha U(\bar\vs)

592: \]

593: and then lifting the solution found to the full space using the

594: energy relation \ref{eq:energyrelation}. The graphs of homotetic

595: solutions are straight lines contained in the normal bundle of $S$

596: in $S\times \RR^k$.

597: 

598: %%\subsection{McGehee coordinates for the Kepler problem}

599: %%Consider the classical Kepler problem, in which $S\subset \RR^2$ is

600: %%a $1$-dimensional manifold. Let $(0,2\pi) \to S$ be a (maybe

601: %%partially defined) local parametrisation, which we will denote by

602: %%$\vartheta \mapsto \vs(\vartheta)$, for $\vs(\vartheta) = (

603: %%m_1^{-1/2}\cos \vartheta, m_2^{-1/2} \sin \vartheta )$, if $m_1$ and

604: %%$m_2$ are the diagonal entries of $M$. If we introduce the vector

605: %%$\hat\vs = (- m_1^{-1/2} \sin \vartheta, m_2^{-1/2} \cos \vartheta)$

606: %%and the scalar $w$ such that $\vw = w \hat\vs$,  then equations

607: %%\ref{eq:mcghee3} turn out to become

608: %%\begin{equation}\label{eq:mcgheeonedim}

609: %%\left\{

610: %%\begin{aligned}

611: %%v' &= w^2 + \beta \left(v^2 - 2 U(\vartheta)\right)\\

612: %%\vartheta' &=  w \\

613: %%w' &= (\beta-1) vw +  U'(\vartheta),

614: %%\end{aligned}

615: %%\right.

616: %%\end{equation}

617: %%with the parabolic manifold $\parabolicmanifold$ defined by the

618: %%equation

619: %%\[

620: %%v^2 + w^2 = 2U(\vartheta).

621: %%\]

622: %%Now consider the flow in part of the parabolic manifold

623: %%$\parabolicmanifold$ contained in the half-space $w>0$: by

624: %%eliminating the term $w^2$ in the equations of $v'$ and

625: %%$\vartheta'$, the projection of the flow on the

626: %%$(\vartheta,v)$-plane is contained in the region $\{ (\vartheta,v) :

627: %%v^2  < 2U(\vartheta) \}$ and is given by the system

628: %%\begin{equation}\label{eq:batta}

629: %%\left\{

630: %%\begin{aligned}

631: %%\vartheta' &= \sqrt{2U(\vartheta) - v^2 } \\

632: %%v' & = (1-\beta) (2U(\vartheta) -v^2),

633: %%\end{aligned}

634: %%\right.

635: %%\end{equation}

636: %%which can be written also as (see figures~\ref{fig:fund4}

637: %%and~\ref{fig:fund} at page~\pageref{fig:fund4})

638: %%\begin{equation}\label{eq:prima}

639: %%\dfrac{dv}{d\vartheta} = (1-\beta) \sqrt{2U(\vartheta) -v^2}.

640: %%\end{equation}

641: %%For the projection of the part in $w<0$, the first equation of

642: %%\ref{eq:batta} has to be changed in \( \vartheta' = -

643: %%\sqrt{2U(\vartheta) - v^2} \). If $U(\vartheta)\equiv U$ is

644: %%constant, then the parabolic manifold $\parabolicmanifold$ is the

645: %%cylinder of equation $v^2+w^2 = 2U$ and the flow in

646: %%$\parabolicmanifold$ (which is invariant up to translation in

647: %%$\vartheta$) is given by curves leaving the line $v=-\sqrt{2U}$,

648: %%$w=0$ of equilibrium points at $\vartheta_0$ and reaching the

649: %%equilibrium line  $v=\sqrt{2U}$, $w=0$ at $\vartheta_1 = \vartheta_0

650: %%+ \dfrac{\pi}{1-\beta}$, since

651: %%\[

652: %%(1-\beta)(\vartheta_1 - \vartheta_0) = \int_{-\sqrt{2U}}^{\sqrt{2U}}

653: %%(2U-v^2)^{-1/2} \,dv = \pi.

654: %%\]

655: %%Thus a ``bouncing'' trajectory on a collision can be seen as  a

656: %%solution of the regularised problem only for $1-\beta =

657: %%\dfrac{1}{2}$ implies that $\alpha = 1$ (this is the reason the

658: %%Levi--Civita regularisation might work only in the case $\alpha=1$).

659: %%% Nonetheless, the Sundman--McGehee regularisation can be

660: %%% applied for every $\alpha$ (yielding a less natural regularisation).

661: %%

662: %%

663: %%Assume that $S\subset S^1 \subset \RR^2$ is equal to the ellipse

664: %%minus a finite number of points, where $U(\vartheta)$ goes to

665: %%infinity. We call these points \emph{collisions}. This is the case

666: %%for example of the Kepler problem (where $U$ is constant and without

667: %%collisions), the anisotropic Kepler problem \cite{Devaney81} (still

668: %%without collisions), the planar isosceles $3$-body problem

669: %%\cite{Devaney80}, the collinear $3$-body problem \cite{McGehee74},

670: %%the rectangular $4$-body problem \cite{SimoLacomba82}, the

671: %%tetrahedral $4$-body problem without rotation \cite{DelgadoVidal99}.

672: %%% Examples: Kepler, anisotropic Kepler, isosceles, collinear,

673: %%% Simo-Lacomba 4-body, Vidal, Fusco alpha-rosette.

674: %%If $\phi(\vartheta)$ is a function which is strictly positive in

675: %%$S$, then by setting $w = \phi  (\vartheta) u$ and rescaling the

676: %%vector field (i.e. rescaling time by the factor $\phi(\vartheta)$)

677: %%equations~\ref{eq:mcgheeonedim} are transformed into

678: %%\begin{equation}\label{eq:mcgheeonedim_2}

679: %%\left\{

680: %%\begin{aligned}

681: %%v' &= u^2\phi + \beta \dfrac{v^2 - 2 U}{\phi}\\

682: %%\vartheta' &=  u \\

683: %%u' &= (\beta-1) \dfrac{u v}{\phi} + \dfrac{U'}{\phi^2} -

684: %%\dfrac{\phi'}{\phi} u^2.

685: %%\end{aligned}

686: %%\right.

687: %%\end{equation}

688: %%The parabolic manifold $\parabolicmanifold$ has equation $v^2 +

689: %%\phi^2u^2 - 2U = 0$, and the projection on the $(\vartheta,v)$ plane

690: %%of the flow on $\parabolicmanifold$ has the same equations

691: %%\ref{eq:batta} as before (for $u>0$, for $u<0$ the analogous ones),

692: %%after the same rescaling in time. If the set of collisions is not

693: %%empty, then over each connected component $C$ of $S$, by a suitable

694: %%choice of $\phi$ (for example $\phi=U$), the vector field can be

695: %%regularised on $\parabolicmanifold$ in such a way that it can be

696: %%extended to a smooth non-degenerate vector field on a smooth surface

697: %%contained in the closure $\overline{C} \times \RR^2$. In usual

698: %%McGehee coordinates, this is the Sundman regularisation, which in

699: %%general gives a regularisation on the energy manifolds for the

700: %%Newton potential with $\alpha=1$.

701: 

702: %%%=========================================================================

703: \section{The dihedral $2n$-body problem}

704: \label{sec:equivsetup} Let $\RR^3 \cong \CC \times \RR$ be endowed

705: with coordinates $(z,y)$, $z\in \CC$, $y\in \RR$. For $l\geq 1$, let

706: $\zeta_l$ denote the primitive root of unity $\zeta_l = e^{2\pi

707: i/l}$; the \emph{dihedral} group $D_{l}\subset SO(3)$ is the group

708: of order $2l$ generated by the rotations

709: \[

710: \zeta_l\from (z,y) \mapsto (\zeta_l z, y ) \text{\ and \ }

711: \kappa\from (z,y) \mapsto (\overline{z},-y),

712: \]

713: where $\overline{z}$ is the complex conjugate of $z$. The

714: non-trivial elements of $D_l = \langle \zeta_l,\kappa \rangle$ are

715: the  $l-1$ rotations around the $l$-gonal axis $\zeta_l^j$,

716: $j=1,\ldots,l-1$ and  the $l$ rotations of angle $\pi$ around the

717: $l$ digonal  axes orthogonal to the $l$-gonal axis (see figure

718: \ref{fig:fun}) $\zeta_l^j\kappa$, $j=1,\ldots ,l$. In

719: figures~\ref{fig:fund4} and~\ref{fig:fund} one can find the

720: upper-halves of the fundamental domains for the action of $D_l$

721: restricted on the unit sphere. In fact, in figure~\ref{fig:fund4}

722: corresponding to the dihedral four body problem, the fundamental

723: domain is represented by an octant of the shape sphere while

724: figure~\ref{fig:fund} represent the fundamental domain on the shape

725: sphere for the dihedral six body problem.

726: \begin{figure}

727: \centering

728: \subfigure[$l=2$]{%

729: \includegraphics[width=0.480\textwidth]{figs/fun4}

730: % \caption{Fundamental domain for the dihedral action for $l=2$}

731: \label{fig:fund4} }

732: \subfigure[$l=3$]{%

733: % \caption{Fundamental domain for the dihedral action for $l=3$}

734: \includegraphics[width=0.480\textwidth]{figs/fun}

735: \label{fig:fund} } \caption{Dihedral groups $D_l$, with the upper

736: half of the fundamental domains in white.} \label{fig:fun}

737: \end{figure}

738: %%

739: %%\begin{figure}\centering

740: %%\includegraphics[width=0.618\textwidth]{figs/fun2}

741: %%\caption{$G$-orbit of a generic point for the dihedral action with $l=3$

742: %%(6 bodies).}

743: %%\label{figs/fun2}

744: %%\end{figure}

745: 

746: Consider the permutation representation of $D_l$ given by left

747: multiplication (that is, the Cayley immersion $\sigma\from D_l \to

748: \Sigma_{2l}$ of $D_l$ into the symmetric group on the  $2l$ elements

749: of $D_l$, defined by $\sigma(g)(x) = gx$ for each $g,x\in D_l$, see

750: \cite{gg} for more details). The action of $D_l$ on $\RR^3$ induces

751: an orthogonal action on the configuration space $\RR^{6l}$ of $n=2l$

752: point particles $\vq_i\in \RR^3$ in the three-dimensional space. The

753: Newtonian potential for the $n$-body problem, homogeneous with

754: degree $-\alpha$ induces by restriction on the fixed subspace

755: $\left(\RR^{6l}\right)^{D_{l}} \cong \RR^3$ a homogeneous potential

756: defined for each $\vq\in \RR^3$ by

757: \begin{equation}\label{eq:potenziale}

758: U(\vq) = \sum_{g\in D_l \smallsetminus \{1\}} \left| \vq - g \vq

759: \right|^{-\alpha},

760: \end{equation}

761: provided we assume (without loss of generality) all masses

762: $m_i^{2}=1/l$. Now, the potential $U$ in \ref{eq:potenziale} can be

763: re-written in terms of coordinates $\vq = (z,y) \in \CC\times \RR$

764: as

765: \[\begin{aligned}

766: U(\vq) &= \sum_{j=1}^{l-1} | \vq - \zeta_l^j \vq |^{-\alpha} +

767: \sum_{j=1}^{l} | \vq - \zeta_l^j \kappa \vq |^{-\alpha} \\

768: &= \sum_{j=1}^{l-1} | z - \zeta_l^j z |^{-\alpha} + \sum_{j=1}^{l}

769: \left(  | z - \zeta_l^j \overline{z}|^2 + 4y^2 \right) ^{-\alpha/2}.

770: \end{aligned}

771: \]

772: By definition, for each $g\in D_{l}$, $U(g\vq)=U(\vq)$. Further

773: symmetries of $U$ are:

774: \begin{enumerate}

775: \item the reflection on the

776: plane $y=0$ (given by $h\from (z,y) \mapsto (z,-y)$),

777: \item

778: the $l$ reflections on the  planes containing the $l$-gonal axis and

779: one of the digonal axes,

780: \item and the $l$ reflections on the planes

781: containing the $l$-gonal axis and the points $( \zeta_l^j e^{\pi i/

782: l}  , 0 )$, $j = 1, \ldots, l$.

783: \end{enumerate}

784: It is not difficult to prove that these are (up to conjugacy and

785: multiplication with elements in $D_l$) all the elements of  the

786: normaliser of $D_l$ in $O(3)$. Thus we can study $U$ only in the

787: left-upper area of the $D_l$-fundamental domain on $S^2\subset

788: \RR^3$, as we have seen in figures \ref{fig:fund4} and

789: \ref{fig:fund}. Now, in order to simplify the expression of the

790: potential we introduce the variables $r$ and $\xi$ as follows. If

791: $y\geq 0$ and $z\neq 0$, let $r=r(z,y)$ be defined as $r = 1 +

792: 2y^2/|z|^2 - 2y/|z|\sqrt {y^2/|z|^2 + 1}$ and $\xi =

793: \dfrac{\overline{z}}{z}$. Hence $r\in (0,1]$, with $r=1$ if and only

794: if  $y=0$, $1+r^2 = r ( 2 + 4 y^2/|z^2|)$ and therefore

795: \[

796: \begin{aligned}

797: | z - \zeta_l^j \overline{z}|^2 + 4y^2  &= |z|^2 \left( | 1 -

798: \zeta_l^j \dfrac{\overline{z}}{z} |^2 + 4 \dfrac{y^2}{|z|^2} \right)

799:  = \dfrac{|z|^2}{r} \left| 1 -  \zeta_l^j r \xi \right|^2.

800: \end{aligned}

801: \]

802: In these coordinates the potential function $U(\vq)$ can be written

803: as

804: \begin{equation}\label{eq:potentialplain}

805: U = |z|^{-\alpha} \left[ \sum_{j=1}^{l-1} | 1 - \zeta_l^j

806: |^{-\alpha} + r^{-\alpha/2} \sum_{j=1}^{l} \left| 1 - \zeta_l^j r

807: \xi \right| ^{-\alpha} \right].

808: \end{equation}

809: We can now state the integral representation of the potential

810: \ref{eq:potentialplain} proven in the Appendix \ref{sec:appendice}

811: (see also \cite{BanElm} and remark \ref{remark:qqq} below).

812: \begin{propo}

813: \label{propo:potentialintegral} For $\beta\in(0,1)$, $r\in (0,1]$

814: and $\xi\in S^1\subset \CC$ the potential $U$ can be written as

815: \[

816: U = |z|^{-\alpha} \left[ c_l  + l r^{-\beta} \dfrac{\sin( \beta

817: \pi)}{ \pi } \int_0^1

818: \dfrac{(1-t)^{-\beta} t^{\beta-1} }{ %

819: {(1-tr^2)}^{\beta} } \dfrac{1 - (tr)^{2l}}{|1-(tr)^l\xi^l|^2} \,dt

820: \right],

821: \]

822: where $c_l$ is the constant $c_l  = \sum_{j=1}^{l-1} | 1 - \zeta_l^j

823: |^{-\alpha}$.

824: \end{propo}

825: \proof For the proof of this result, see Appendix

826: \ref{sec:appendice}.\qed

827: \begin{remark}

828: The above integral representation plays a fundamental role in order

829: to find all the central configuration. In fact, otherwise the

830: expression of the potential given in formula \ref{eq:potentialplain}

831: is quite difficult to deal with.

832: \end{remark}

833: \subsection{Planar type central configurations}

834: 

835: On the unit sphere $S\subset \RR^3$ (of equation $|z|^2 + y^2 = 1$),

836: parametrised by $(\varphi,\theta)\in (-\pi/2,\pi/2) \times [0,2\pi)$

837: with $y=\sin  \varphi$ and $z = \cos\varphi e^{i\theta}$, the

838: (reduced to the $2$-sphere) potential reads

839: \begin{equation}\label{potential:sphere}

840: {U}(\theta,\varphi) \!\!=\!\! (2\cos\varphi)^{-\alpha} \left[

841: \sum_{j=1}^{l-1} \left( \sin \dfrac{j\pi}{l} \right)^{-\alpha}

842: \!\!\!+\!\! \sum_{j=1}^{l} \left( \sin^2 (\dfrac{j\pi}{l}

843: \!-\!\theta ) \!+\! \tan^2 \varphi \right) ^{-\frac{\alpha}{2}}

844: \right],

845: \end{equation}

846: and by Proposition \ref{propo:potentialintegral} also as

847: \[

848: {U}(\theta,r) = \left( \dfrac{1+r^2}{4r} \right)^{\beta} \left[ c_l

849: + l r^{-\beta} \dfrac{\sin( \beta \pi)}{ \pi } I(r, \theta)\right]\]

850: \[

851: \textrm{where}\ \ I(r,\theta)=\int_0^1

852: \dfrac{(1-t)^{-\beta} t^{\beta-1} }{ %

853: {(1-tr^2)}^{\beta} } \dfrac{1 - (tr)^{2l}}{1+(tr)^{2l} -

854: 2(tr)^l\cos(2 l\theta) } \,dt ,

855: \]

856: with (just for $\varphi\in [0,\pi/2)$)

857: \[

858: r = 1 + 2\tan^2\varphi  -2 \dfrac{\tan\varphi}{\cos\varphi} =

859: \dfrac{1-\sin\varphi}{1+\sin\varphi}

860: \]

861: and hence

862: \[

863: \sin\varphi = \dfrac{1-r}{1+r} \text{\ and \ } \cos^2\varphi =

864: \dfrac{4r}{1+r^2}.

865: \]

866: In spherical coordinates, the symmetry reflections  of $U$ are (up

867: to conjugacy)

868: \begin{enumerate}

869: \item the reflection on the horizontal

870: plane: $h_\varphi\from (\theta,\varphi)  \mapsto

871: (\theta,-\varphi)$,

872: \item

873: the reflection on the plane containing the $l$-gonal axis and the

874: digonal axis $h_\theta\from  (\theta,\varphi) \mapsto

875: (-\theta,\varphi)$

876: \item and the reflection on the plane

877: containing the $l$-gonal axis and the point $( e^{\pi i/ l}  , 0 )$,

878: defined as $h'_\theta\from (\theta,\varphi) \mapsto (\pi/l - \theta

879: ,\varphi)$.

880: \end{enumerate}

881: As direct consequence of the Palais' symmetric criticality

882: principle, it follows that critical points of the restrictions of

883: the reduced potential $U$ to the $1$-spheres of such fixed planes

884: are critical points for the restriction of $U$ to the sphere, and

885: hence are central configurations for $U$. In fact, as already

886: observed this $1$-spheres are nothing but the spaces fixed by each

887: of the reflections given in $(i)$, $(ii)$ and $(iii)$. In principle

888: it can be exist other critical points for the restriction of the

889: potential $U$ to the sphere which do not lie in these fixed spaces.

890: However if we are able to show that out of this $1$-spheres the

891: derivative of the potential is bounded away from zero, we have done.

892: 

893: Now consider the derivative with respect to $\theta$ of $U$, which

894: by Proposition \ref{propo:potentialintegral} can be written as

895: follows

896: \begin{equation}\label{eq:partialtheta}

897: \dfrac{\partial U}{\partial \theta}  = -4 l^2 \sin(2l\theta) \dfrac{

898: (1+r^2)^\beta \sin (\beta \pi ) }{ \pi ( 2r )^\alpha

899: % r^\beta(\cos\varphi)^{\alpha}

900: } I(r,\theta)

901: \end{equation}

902: where $I(r,\theta)$ is  strictly positive and defined for

903: $(\theta,r) \neq (2k\pi/l,1)$, $k$ integer. Hence for each $r\in

904: (0,1]$ the derivative $\dfrac{\partial U}{\partial \theta}$ is

905: strictly negative for $\theta \in (0,\dfrac{\pi}{2l})$ and strictly

906: positive for $\theta\in (\dfrac{\pi}{2l},\dfrac{\pi}{l})$. It is

907: zero for $\theta= \dfrac{k\pi}{2l}$ and $r\in (0,1)$ and

908: $\theta=\dfrac{(2k+1)\pi}{2l}$ and $r=1$. Thus, for $\varphi=0$, we

909: have proved the following proposition:

910: \begin{nr}[Planar $2l$-gon]

911: \label{nr:nagon} For any $\alpha \in (0,2)$ central configurations

912: which are $h_\varphi$-symmetric are on the vertices $(e^{(2k+1)\pi

913: i/(2l) },0)$ of the regular $2l$-gon.

914: \end{nr}

915: \subsection{Prism type central configurations}

916: Now we have to explore the cases $\theta = k\pi /l $ and $\theta  =

917: (2k+1)\pi/(2l)$, which correspond respectively to prisms and

918: antiprisms. The derivative of \ref{potential:sphere} with respect to

919: $\varphi$ is

920: \begin{equation}\label{eq:partialvphi}\begin{aligned}

921: \dfrac{\partial U}{\partial \varphi} & = 2 \beta \dfrac{

922: \tan\varphi}{(2\cos\varphi)^\alpha} \left[ c_l - \sum_{j=1}^l

923: \dfrac{ \cos^2(j\pi/l - \theta) }{ \left( \sin^2(j\pi/l - \theta) +

924: \tan^2\varphi \right)^{\beta+1} }

925: \right] \\

926: & = 2 \beta \dfrac{ \tan\varphi}{(2\cos\varphi)^\alpha} \left[

927: f_\theta(\varphi) \right].

928: \end{aligned}

929: \end{equation}

930: The term in square brackets $f_\theta(\varphi)$ has the same sign of

931: $\dfrac{\partial U}{\partial \varphi}$, and since $c_l$ is a

932: constant and each term of the sum is strictly monotone in $\varphi$,

933: for each $\theta$ the function $f_\theta(\varphi)$ can vanish at

934: most once in the interval $(0,\pi/2)$. Since the  limit of the sum

935: as $\varphi\to \pi/2$ is zero and $c_l$ is positive, there will be a

936: unique zero in $(0,\pi/2)$ (for a fixed $\theta$) for all the values

937: $\theta$ such that $\lim_{\varphi \to 0} f_\theta(\varphi) < 0 $,

938: i.e.

939: \[

940: \lim_{\varphi \to 0} \sum_{j=1}^l \dfrac{ \cos^2(j\pi/l - \theta) }{

941: \left( \sin^2(j\pi/l - \theta) + \tan^2\varphi \right)^{\beta+1} }

942: >  c_l = \sum_{k=1}^{l-1} \left( \sin^2 (j\pi/l) \right)^{-\beta}.

943: \]

944: Now, since $ \lim_{\varphi \to 0} f_0(\varphi) = -\infty,$ there

945: exists a unique minimum $\hat\varphi$ for $\theta= k\pi/l$,

946: $k=0\ldots 2l-1$, corresponding to a prism.

947: 

948: \begin{nr}[Prisms]

949: \label{nr:prism} There are exactly $4l$ central configurations which

950: are $h_\theta$-symmetric (up to conjugacy), and they are precisely

951: on the vertices  of a prism: $(\cos\hat\varphi' e^{k\pi i/l },\pm

952: \sin \hat\varphi')$.

953: \end{nr}

954: We observe that in the dihedral four body problem these kind of

955: central configurations collapse to square type central

956: configurations.

957: \subsection{Antiprism type central configurations}

958: It is left to compute critical points for $\theta = (2k+1)\pi/(2l)$,

959: that is, to find zeroes of $f_\theta(\varphi)$ for $\theta =

960: \dfrac{\pi}{2l}$, or, equivalently, $h'_\theta$-symmetric central

961: configurations.

962: 

963: \begin{nr}[Antiprisms]

964: \label{nr:antiprism} There are exactly $2l$ central configurations

965: which are $h'_\theta$-symmetric (up to conjugacy) and $\varphi\neq

966: 0$. They are on the vertices of a prism: $(\cos\hat\varphi

967: e^{(2k+1)\pi i/(2l) },\pm \sin \hat\varphi)$.

968: \end{nr}

969: We remark that in the four body problem the antiprism type central

970: configurations reduce to tetrahedral type configurations.

971: \begin{proof}

972: It suffices to show that

973: \[

974: \sum_{j=1}^l \dfrac{ \cos^2(j\pi/l - \pi/(2l)) }{ \left(

975: \sin^2(j\pi/l - \pi/(2l)) \right)^{\beta+1} }

976: >  \sum_{k=1}^{l-1} \left( \sin^2 (j\pi/l) \right)^{-\beta}.

977: \]

978: If $\lfloor l/2 \rfloor$ denotes the greatest integer $n\leq l/2$,

979: that is

980: \[

981: \lfloor l/2 \rfloor =

982: \begin{cases}

983: (l-1)/2 & \text{ $l$ odd} \\

984: l/2 & \text{$l$ even,}

985: \end{cases}

986: \]

987: then

988: \[

989: \sum_{j=1}^l \dfrac{ \cos^2(j\pi/l - \pi/(2l)) }{ \left(

990: \sin^2(j\pi/l - \pi/(2l)) \right)^{\beta+1} } = 2

991: \sum_{j=1}^{\lfloor l/2 \rfloor} \dfrac{ \cos^2(j\pi/l - \pi/(2l) )

992: }{ \left( \sin^2(j\pi/l - \pi/(2l)) \right)^{\beta+1}}.

993: \]

994: On the other hand

995: \[

996: \sum_{j=1}^{l-1} \left( \sin^2 (j\pi/l) \right)^{-\beta} = 2

997: \sum_{j=1}^{\lfloor l/2 \rfloor} \left( \sin^2 (j\pi/l)

998: \right)^{-\beta} + d_l,

999: \]

1000: where

1001: \[

1002: d_l = \begin{cases}

1003: 1 & \text{$l$ even} \\

1004: 0 & \text{$l$ odd}

1005: \end{cases}

1006: \]

1007: Now then, since

1008: \[

1009: \dfrac{ \cos^2 x  }{ \left( \sin^2 x \right)^{\beta+1}} = \dfrac{ 1

1010: }{ \left( \sin^2 x \right)^{\beta+1}} - \dfrac{ 1  }{ \left( \sin^2

1011: x \right)^{\beta}},

1012: \]

1013: the conclusion would follow once we could prove that

1014: \[

1015:  2 \sum_{j=1}^{\lfloor l/2 \rfloor} C_j  > d_l,

1016: \]

1017: where

1018: \[

1019: C_j\! = \!\dfrac{ 1  }{ \left( \sin^2 (j\pi/l - \pi/(2l))

1020: \right)^{\beta+1}} - \dfrac{ 1  }{ \left( \sin^2 (j\pi/l - \pi/(2l))

1021: \right)^{\beta}} -

1022: \dfrac{1}{%

1023: \left( \sin^2 (j\pi/l) \right)^{\beta}}.

1024: \]

1025: If $l=2$, it turns out that $C_1 = 2^\beta - 1$ and hence $2

1026: \sum_{j=1}^{\lfloor l/2 \rfloor} C_j = 2C_1 > 0 = d_2$. If $l=3$,

1027: then $C_1 = 2^\alpha ( 3 - 3^{-\beta})$, which is greater than $2$

1028: for all $\alpha = 2\beta$, so that $2C_1 > 1 = d_1$. In general,

1029: since $j \leq l/2$, $\sin(j\pi/l) > \sin(j\pi/l - \pi/(2l))$, and

1030: therefore

1031: \[\begin{aligned}

1032: C_j & > \dfrac{ 1  }{ \left( \sin^2 (j\pi/l - \pi/(2l))

1033: \right)^{\beta+1}} - \dfrac{ 2  }{ \left( \sin^2 (j\pi/l - \pi/(2l))

1034: \right)^{\beta}} \\

1035: & = \dfrac{ 1 - 2 \sin^2( j\pi/l - \pi/(2l))  }{ \left( \sin^2

1036: (j\pi/l - \pi/(2l)) \right)^{\beta+1}}.

1037: \end{aligned}\]

1038: The first term is estimated by

1039: \[\begin{split}

1040: C_1 \geq \dfrac{ 1 - 2 \sin^2( \pi/(2l))  }{ \left( \sin^2

1041: (\pi/(2l)) \right)^{\beta+1}}

1042: > \dfrac{1 -  \dfrac{\pi^2}{2l^2} }{%

1043: \left( \dfrac{\pi}{2l} \right)^{\alpha+2} } = (1 - \dfrac{\pi^2}{2

1044: l^2}) \dfrac{2^{\alpha+2} l^{\alpha+2} }{\pi^{\alpha+2}}

1045: \\

1046: \geq

1047: % (1 - \dfrac{\pi^2}{2 l^2}) \dfrac{4 l^{2} }{\pi^{2}}

1048: \dfrac{4l^2}{\pi^2} - 2 > \dfrac{l^2}{4} - 2,

1049: \end{split}\]

1050: and all other terms $C_j$ with $j\geq 2$ are in any case greater

1051: than $-1$; thus for $l\geq 4$

1052: \[

1053:  \sum_{j=1}^{\lfloor l/2 \rfloor} C_j

1054: >

1055:  C_1 - ( \lfloor \dfrac{l}{2} \rfloor - 1 )

1056:  \geq C_1 - \dfrac{l}{2} + 1

1057:  \geq

1058:  \dfrac{l^2}{4} -  \dfrac{l}{2} -1 \geq  1,

1059: \]

1060: and thus for all $l\geq 4$ we have $2  \sum_{j=1}^{\lfloor l/2

1061: \rfloor} C_j \geq 2  > d_l,$ which concludes the proof.

1062: \end{proof}

1063: 

1064: Since there are no other central configurations, by

1065: \ref{eq:partialtheta}, we can summarise the results in the following

1066: proposition.

1067: \begin{propo}

1068: \label{propo:main} All central configurations in the dihedral

1069: $2n$-body problem are symmetric for one of the three types of

1070: reflections $h_\varphi$ Lemma \ref{nr:nagon}, $h_\theta$ Lemma

1071: \ref{nr:prism} or $h'_\theta$ Lemma \ref{nr:antiprism}. They are

1072: represented in the (upper-half) fundamental domain on the sphere in

1073: figure \ref{fig:centralconfigurations}.

1074: \end{propo}

1075: \begin{figure}\centering

1076: \includegraphics[width=0.718\textwidth]{figs/centralconfigurations}

1077: \caption{Central configurations in the upper-half fundamental

1078: domain.} \label{fig:centralconfigurations}

1079: \end{figure}

1080: In fact, in figure \ref{fig:centralconfigurations} it is drawn a

1081: geodesic triangle which represents the fundamental domain on the

1082: shape sphere for the dihedral $n$-body problem. In this figure are

1083: shown the three types of central configurations arising in the

1084: problem we are dealing with in the exact location together with.

1085: Moreover we observe that due to the symmetry constraint only two

1086: types of collisions can occur. We denoted by the name $l$-adic

1087: collision and binary collision, meaning that in the first case two

1088: clusters of $l$-bodies simultaneously collide, while in the second

1089: case $l$ clusters of 2 bodies simultaneously collide. This two types

1090: of collisions are all located on the same plane containing the

1091: planar central configurations while the $l$-adic central

1092: configurations can be represented in the north and south pole of the

1093: shape sphere.

1094: 

1095: 

1096: Now consider equations \ref{eq:mcghee3} in coordinates

1097: $(\theta,\varphi)$ on the sphere: we set $w_1$ and $w_2$ such that

1098: $\vw = w_1 \dfrac{\partial \vs}{\partial \theta} + w_2

1099: \dfrac{\partial \vs}{\partial \varphi}$, i.e.  (since $\vs =

1100: (\cos\varphi e^{i\theta}, \sin \varphi )$),

1101: \[

1102: \vw = w_1 ( i \cos\varphi e^{i\theta}, 0) + w_2 (-\sin\varphi

1103: e^{i\theta}, \cos\varphi).

1104: \]

1105: Then

1106: \[

1107: \norm{\vw} = w_1^2\cos^2\varphi + w_2^2

1108: \]

1109: and

1110: \[

1111: \nabla_{\vs} U(\vs)  = \dfrac{1}{\cos^2\varphi} \dfrac{\partial

1112: U}{\partial \theta} \dfrac{\partial \vs}{\partial \theta} +

1113: \dfrac{\partial U}{\partial \varphi} \dfrac{\partial \vs}{\partial

1114: \varphi}.

1115: \]

1116: Also, equations \ref{eq:mcghee3} become

1117: \begin{equation}\label{eq:mcgheesphere}

1118: \left\{

1119: \begin{aligned}

1120: v' &= w_1^2 \cos^2\varphi + w_2^2 +\beta v^2 - \alpha U(\theta,\varphi) \\

1121: \theta' &= w_1 \\

1122: \varphi' &= w_2 \\

1123: w_1' &= (\beta-1)vw_1 + 2 \tan \varphi~  w_1 w_2 +

1124: \dfrac{1}{\cos^2\varphi} \dfrac{\partial U}{\partial \theta}

1125: \\

1126: w_2' &= (\beta-1)v w_2 - \dfrac{1}{2} w_1^2 \sin 2\varphi  +

1127: \dfrac{\partial U}{\partial \varphi}.

1128: \end{aligned}

1129: \right.

1130: \end{equation}

1131: The linearization at equilibrium points (central configurations)

1132: \ref{eq:centralconf} is represented by the $5\times 5$ matrix  $L$

1133: \[

1134: L =

1135: \begin{bmatrix}

1136: 2\beta v & 0 & 0 & 0 & 0 \\

1137: 0 & 0 & 0 & 1 & 0 \\

1138: 0 & 0 & 0 & 0 & 1 \\

1139: 0 & \dfrac{1}{\cos^2\varphi} \dfrac{\partial^2 U}{\partial \theta^2}

1140: &

1141: \dfrac{\partial^2 U}{\partial \theta \partial \varphi } & (\beta-1)v & 0 \\

1142: 0 & \dfrac{\partial^2 U}{\partial \varphi\partial\theta} &

1143: \dfrac{\partial^2 U}{\partial\varphi^2} & 0 & (\beta-1)v \\

1144: \end{bmatrix}.

1145: \]

1146: Thus the eigenvalues of the linearization can be computed in terms

1147: of the eigenvalues of the Hessian $D^2U(\overline\vs)$  of

1148: $U(\theta,\varphi)$:

1149: \begin{propo}\label{propo:borat}

1150: The eigenvalues of $L$, at a central configuration $\overline\vs$

1151: (i.e. at the point $(\overline v, \overline \vs, \zero)$, where

1152: $\overline v = \pm \sqrt{2U(\overline\vs)}$), are equal to the roots

1153: $\lambda$ of the equation

1154: \[

1155: \lambda^2 + (1-\beta)\overline v \lambda = \gamma

1156: \]

1157: for each $\gamma$ eigenvalue of the Hessian $D^2U(\overline\vs)$.

1158: \end{propo}

1159: 

1160: By elementary calculations it follows from Proposition

1161: \ref{propo:borat} that the equilibrium points

1162: $(\pm\sqrt{2U(\overline\vs)},\overline\vs,\zero)$ are hyperbolic

1163: when the Hessian $D^2U$ is non-singular at $\overline{\vs}$, and

1164: that for each positive eigenvalue $\gamma>0$ of $D^2U$ there is a

1165: pair of real eigenvalues of $L$,  $\lambda_1>0$, $\lambda_2<0$; for

1166: each negative eigenvalue $\gamma <0$ of $D^2U$, there are two

1167: eigenvalues $\lambda_1$, $\lambda_2$ of $L$ with negative real part

1168: ($\lambda_1$, $\lambda_2$ are real if $d = (1-\beta)^2{\overline

1169: v}^2 + 4\gamma > 0$ and $\lambda_1=\lambda_2$ if $d=0$).

1170: 

1171: \begin{propo}

1172: All equilibrium points of \ref{eq:mcgheesphere} are hyperbolic.

1173: % (with non-zero

1174: % real part).

1175: \end{propo}

1176: \begin{proof}

1177: We just need to proof that the Hessian $D^2U$ is non-singular at

1178: $\overline\vs$, if $\vs$ is a central configuration. Since each

1179: central configuration $\vs$ lies in the line fixed by a reflection

1180: (which is a symmetry of $U$), the matrix $D^2U$ is diagonal at

1181: $\overline\vs$. So the result follows once we prove that

1182: $\dfrac{\partial^2 U}{\partial \theta^2}(\overline\vs) \neq 0 \neq

1183: \dfrac{\partial^2U}{\partial \varphi^2}(\overline\vs)$. But by

1184: \ref{eq:partialtheta}, since $I(r,\theta)$ is strictly positive and

1185: regular in a neighbourhood of $\overline{s}$, $\dfrac{\partial^2

1186: U}{\partial \theta^2}(\overline\vs) \neq 0$. By

1187: \ref{eq:partialvphi}, the same holds for

1188: $\dfrac{\partial^2U}{\partial \varphi^2}(\overline\vs)$.

1189: \end{proof}

1190: 

1191: 

1192: \begin{propo}\label{coro:dimensions}

1193: The dimension of the stable (unstable) manifold of $(\overline

1194: v,\overline\vs,\zero)$ with $\overline v = \sqrt{2 U(\overline\vs)}

1195: > 0$ is 3 (2) if $\overline\vs$ is a $2l$-gon or a prism; it is 2

1196: (3) if $\overline\vs$ is an antiprism. The dimension of the stable

1197: (unstable) manifold of the point $(- \overline

1198: v,\overline\vs,\zero)$ with $\overline v = \sqrt{2 U(\overline\vs)}

1199: > 0$ is equal to the dimension of the unstable (stable) manifold of

1200: $(\overline v,\overline\vs,\zero)$. The intersection of the stable

1201: (unstable) manifold of $(\overline v,\overline\vs,\zero)$ with the

1202: parabolic manifold $\parabolicmanifold$ has codimension 0 (1) in

1203: $\parabolicmanifold$ if $\overline v >0$. It has codimension 1 (0)

1204: in $\parabolicmanifold$ if $\overline v<0$.

1205: \end{propo}

1206: \begin{proof}

1207: These facts follow directly from the stable/unstable manifold

1208: theorem and the above arguments on eigenvalues of $L$. The results

1209: are summarised in table \ref{tb:tabella1}.

1210: \end{proof}

1211: 

1212: % \begin{table}\centering

1213: % \begin{tabular}{c|ccccc}

1214: % & $\overline v $&  {$\dim W^u \cap \parabolicmanifold$}

1215: % &{$\dim W^s \cap \parabolicmanifold$}& {$\dim  W^u$}&{$\dim  W^s$}\\

1216: % \hline \\

1217: % $2l$-agon and prism & $>0$  &  1 & 3& 2& 3 \\

1218: %  & $<0$  & 3&1&3&2 \\

1219: % \hline\\

1220: % Anti-prism  & $>0$& 2 &2& 3& 2   \\

1221: %  &$<0$ & 2 &2& 2 & 3   \\

1222: % \hline

1223: % \end{tabular}

1224: % \caption{Dimensions of stable and unstable manifolds}

1225: % \label{tb:tabella1}

1226: % \end{table}

1227: 

1228: \begin{table}\centering

1229: \begin{tabular}{|l|c|cc|cc|}

1230: \hline & $\overline v $&

1231: \begin{sideways}$\dim  W^s$\end{sideways} &

1232: \begin{sideways}$\dim  W^u$\end{sideways} &

1233: \begin{sideways}$\dim W^s \cap \parabolicmanifold$\end{sideways}&

1234: \begin{sideways}$\dim W^u \cap \parabolicmanifold$\end{sideways}

1235: \\ \hline

1236: \multirow{2}{*}{$2l$-gon and prism} & $>0$  &

1237: 3 & 2 & 3 & 1 \\

1238:  & $<0$

1239:  & 2 & 3 & 1 & 3 \\

1240: \hline \multirow{2}{*}{anti-prism}  & $>0$&

1241: 2 &3& 2& 2   \\

1242:  &$<0$ & 3 &2& 2 & 2   \\\hline

1243: \end{tabular}

1244: \caption{Dimensions of stable and unstable manifolds.}

1245: \label{tb:tabella1}

1246: \end{table}

1247: 

1248: 

1249: 

1250: 

1251: 

1252: %%%%%=========================================================================

1253: %%\section{Colliding parabolic connections}

1254: %%\label{sec:parabolicconnections}

1255: %%

1256: %%

1257: %%The main result of this section is to establish the existence of

1258: %%connecting orbits on the invariant subset of the

1259: %%$\parabolicmanifold$ fixed by the the reflections $h_\theta,

1260: %%h_\varphi, h'_\theta$. The idea to perform our analysis is based

1261: %%upon a study of the intersection between the stable and unstable

1262: %%manifold of the equilibria on the parabolic manifold

1263: %%$\parabolicmanifold$.

1264: %%

1265: %%\begin{figure}\centering

1266: %%\subfigure[%

1267: %%% The flow on the covering of a regularised component of the parabolic

1268: %%% manifold for the

1269: %%Prism section.

1270: %%]{%

1271: %%\includegraphics[width=0.45\textwidth]{figs/flowoncollision}

1272: %%\label{fig:prismsection} }

1273: %%\subfigure[%

1274: %%% The flow on the covering of a regularised component of the parabolic

1275: %%% manifold for the

1276: %%$2n$-gon/antiprism section. ]{

1277: %%\includegraphics[width=0.45\textwidth]{figs/flowoncollision_2}

1278: %%\label{fig:antiprismsection} } \caption{Flows on the covering of a

1279: %%regularized component of the parabolic manifold}

1280: %%\label{fig:prismsections}

1281: %%\end{figure}

1282: %%

1283: %%% \begin{figure}\centering

1284: %%% \caption{The flow on the covering of a regularised component of the parabolic

1285: %%% manifold for the prism section.}

1286: %%% \includegraphics[width=0.5\textwidth]{figs/flowoncollision}

1287: %%% \label{fig:prismsection}

1288: %%% \end{figure}

1289: %%%

1290: %%% \begin{figure}\centering

1291: %%% \caption{The flow on the covering of a regularised component of the parabolic

1292: %%% manifold for the $2n$-gon/antiprism section.}

1293: %%% \includegraphics[width=0.5\textwidth]{figs/flowoncollision_2}

1294: %%% \label{fig:antiprismsection}

1295: %%% \end{figure}

1296: %%

1297: %%%

1298: %%% \begin{remark}

1299: %%% We observe that  the result of Theorem

1300: %%% \ref{thm:calcolovarietastabinstab} is based on the knowledge on the

1301: %%% dimension of the spectral eigenspaces of the potential function at

1302: %%% central configuration. However, in the dihedral four body problem

1303: %%% all of these eigenvalues are explicitly computable. (See

1304: %%% \cite{Vid99}). In this case these eigenvalues are complex and their

1305: %%% imaginary parts will be responsible of a spiralling character of

1306: %%% some invariant manifolds.

1307: %%% \end{remark}

1308: %%

1309: %%Due to the gradient-like character of the (regularised) flow on

1310: %%$\parabolicmanifold$, each orbit  in $\parabolicmanifold$ either

1311: %%tends to a rest-point or $v \to + \infty$ as $t \to + \infty$.

1312: %%Moreover taking into account the inequality $v^2 \leq 2U(\vs),$ it

1313: %%is clear that if $v \to +\infty$ then $U(\vs) \to + \infty$ and so

1314: %%two or more particles have to collide. For $k = 0, 1, \dots, 2l-1$

1315: %%we set $c_k \equiv (e^{\frac{k\pi}{l}i},0)$ and

1316: %%we define the \emph{%

1317: %%binary collision sets}

1318: %%\[

1319: %%B^{s}_{k}:= \left\{(v,\vs,\vw) \in \parabolicmanifold \colon \ v \to

1320: %%+\infty \ \  \textrm{and}  \ \ \vs \to c_k\ \  \textrm{for}  \ \ t

1321: %%\to +\infty \right\};

1322: %%\]

1323: %%\[ B^{u}_{k}:= \left\{(v,\vs,\vw) \in \parabolicmanifold \colon \ v \to -\infty

1324: %%\ \  \textrm{and}  \ \ \vs \to c_k\ \  \textrm{for}  \ \  t \to

1325: %%-\infty\right\};

1326: %%\]

1327: %%and the \emph{$l$-gons collision  sets\/}

1328: %%\[

1329: %%P^{s}_{\pm}:= \left\{(v,\vs,\vw) \in \parabolicmanifold \colon \ v

1330: %%\to +\infty \ \  \textrm{and}  \ \ \vs \to (0,0,\pm 1)\ \

1331: %%\textrm{for}  \ \  t \to +\infty \right\};

1332: %%\]

1333: %%\[

1334: %%P^{u}_{\pm}:= \left\{(v,\vs,\vw) \in \parabolicmanifold \colon \ v

1335: %%\to -\infty \ \  \textrm{and}  \ \ \vs \to (0,0,\pm 1)\ \

1336: %%\textrm{for}  \ \  t \to -\infty\right\}.

1337: %%\]

1338: %%

1339: %%If $X$ and $Y$ are two equilibrium points or even two escape sets on

1340: %%$\parabolicmanifold$ we will write $X \fromto Y$ in order to

1341: %%indicate the existence of an orbit in $\parabolicmanifold$ with

1342: %%asymptotic behaviour $X$ for $t \to -\infty$ and $Y$ as $t \to +

1343: %%\infty$ i.e. if there exists an orbit on the total collision

1344: %%manifold which lies in $W^u(X) \cap W^s(Y)$.

1345: %%

1346: %%In order to simplify the study of the connecting orbits between the

1347: %%rest-points of the Lyapunov function $v$  some remarks are in order.

1348: %%First of all since in each rest point cycle there are two

1349: %%rest-points we introduce the following convention. We denote with

1350: %%superscript $+$ (resp. $-$), central configurations corresponding to

1351: %%the positive (resp. negative) value of the coordinate $v$. Now, we

1352: %%observe that the symmetry

1353: %%\[(v,\vs,\vw) \mapsto (-v, \vs, -\vw)\]

1354: %%in the vector field defined by equations \eqref{eq:mcghee3} implies

1355: %%that for each connection $X^+ \fromto Y^+$ there is a symmetric one

1356: %%$Y^- \fromto X^-$ which will be term \emph{dual connection}.

1357: %%Moreover, this transformation is an involution on

1358: %%$\parabolicmanifold\times \R$ which sends the stable and unstable

1359: %%manifold of the point $Q^+$ respectively on the unstable and stable

1360: %%manifold of the point $Q^-$.

1361: %%

1362: %%This remark gives us some restrictions about the existence of

1363: %%connecting orbits between rest-points, as a direct consequence of

1364: %%the increasing character of the function $v$. In fact, since $v$ is

1365: %%non-decreasing and strictly increasing away from the rest-points,

1366: %%connecting orbits between $X$ and $Y$ cannot occur if $v(Y) \leq

1367: %%v(X)$. Denoting by $\planare, \prisma, \antiprisma$ respectively the

1368: %%planar, prism and antiprism type central configurations, it is easy

1369: %%to check that for positive values of $v$ we have

1370: %%\[

1371: %%v(\planare) \geq v(\prisma) \geq v(\antiprisma).

1372: %%\]

1373: %%

1374: %%In order to prove the main results of this section we observe the

1375: %%following technical fact.

1376: %%\begin{nr}\label{thm:lemmatecnico} Given $(a,b) \subset [0,

1377: %%\pi/2]$, let $\xi \colon (a,b) \to \R^+$ be a regular and positive

1378: %%function such that $\lim_{x \to a^+} \xi(x)=\lim_{x \to b^-}

1379: %%\xi(x)=+\infty$. We assume that $c\in (a,b)$ is the only critical

1380: %%point and let us consider the following one parameter family of

1381: %%initial value problems:  %parameterised by $\delta$

1382: %%\[

1383: %%\left\{

1384: %%\begin{aligned}

1385: %%%\begin{array}{ll}

1386: %%%

1387: %%\dfrac{d x}{d z} & =

1388: %%\pm\dfrac{1}{1-\beta} \left(2\xi(x)-z^2\right)^{-1/2}\\

1389: %%x(\delta) & = 0.

1390: %%\end{aligned}\right.

1391: %%\]

1392: %%Then we have

1393: %%\begin{enumerate}

1394: %%\item[(i)] If $\delta= a$ then the backward solution will reach the curve $z=-\sqrt{2\xi(x)}$ at the right of

1395: %%the point $P\equiv(c, -\sqrt{2\xi(c)})$.

1396: %%\item[(ii)] If

1397: %%$\delta=\frac{3a+b}{2}$ then the backward solution starting at

1398: %%$\delta$ meet again the line $x=\delta$ at the point

1399: %%$R\equiv(\delta, \gamma)$ for $\gamma \in (-\sqrt{2\xi(c)},0)$.

1400: %%\end{enumerate}

1401: %%\end{nr}

1402: %%\begin{proof} It is enough to show that there exist two vector fields

1403: %%$f,g$ such that

1404: %%\[

1405: %%\Big\vert\dfrac{dx}{dz}\Big\vert \leq f, \qquad \textrm{and} \qquad

1406: %%\Big\vert\dfrac{dx}{dz}\Big\vert \geq g.

1407: %%\]

1408: %%Let $y=-z$. Thus we have $\dfrac{dx}{dy}\!=\!

1409: %%\mp\dfrac{1}{1-\beta}\left(2\xi(x)- y^2\right)^{-\frac12}$. We prove

1410: %%$(i)$ by showing that for each solution $x=x(y)$:

1411: %%\begin{itemize}

1412: %%\item[(a)] {\em (First claim). The following inequality holds:

1413: %%\[

1414: %%y(b) \geq \sqrt{2 \xi(c)}\sin[(1-\beta)(b-a)].

1415: %%\]

1416: %%\/}

1417: %%\item[(b)] {\em (Second claim). There exists $\bar x

1418: %%\in (c, b)$ with $\bar y:= y(\bar x)=\sqrt{2\xi(\bar x)}$ such that

1419: %%\[

1420: %%\dfrac{dx}{dy}(\eta)\leq 0 \qquad \forall \,\eta \in (y(b), \bar y

1421: %%).

1422: %%\] \/}

1423: %%\end{itemize}

1424: %%We prove (a). Let $F:\!=\! (1-\beta)^{-1}\left(2\xi(c)-

1425: %%y^2\right)^{-\frac12} $. Then $\dfrac{dx}{dy}\leq F$ in the interval

1426: %%$[c, b]$. Thus by direct calculation it follows that

1427: %%\[

1428: %%\int_a^{x}ds\leq

1429: %%\int_0^{y(x)}\dfrac{1}{(1-\beta)\sqrt{2\xi(c)-y^2}}\,dy,\qquad

1430: %%\forall\, x \in [a,b]

1431: %%\]

1432: %%which implies, in particular, that

1433: %%\[

1434: %%y\left(b\right)\geq \xi(c)\sin [(1-\beta)(b-a)].

1435: %%\]

1436: %%

1437: %%Claim (b). It is sufficient to prove that there exists $\bar x \in

1438: %%(c, b)$ with $\bar y:= y(\bar x)=\sqrt{2\xi(\bar x)}$ such that

1439: %%\[

1440: %%\dfrac{dx}{dy}(\eta)\leq 0, \qquad \forall \,\eta \in (y(b), \bar

1441: %%y),

1442: %%\]

1443: %%where $y$ is a solution of $ \dfrac{dx}{dy}= -

1444: %%(1-\beta)^{-1}\left(2\xi(c)- y^2\right)^{-1/2}$. Thus by direct

1445: %%integration it is enough to show that

1446: %%\begin{eqnarray*}

1447: %%\int_b^{\bar x}dx &\geq& -\int_{y(b)}^{\bar y}\dfrac{1}

1448: %%{(1-\beta)\sqrt{2\xi(c)-y^2}}dy = -

1449: %%\dfrac{1}{1-\beta}\arcsin\left(\dfrac{y(\bar x)}{\xi(c)}\right)\\

1450: %%&+&\dfrac{1}{1-\beta}  \arcsin \left(\dfrac{ y(b)}{\xi(c)}\right)

1451: %%\geq - \dfrac{1}{1-\beta}\arcsin\left(\dfrac{y(\bar

1452: %%x)}{\xi(c)}\right) + b-a,

1453: %%\end{eqnarray*}

1454: %%namely $\arcsin \left(\dfrac{y(\bar x)}{ \xi(c)}\right) \geq

1455: %%(1-\beta)(2b-a-\bar x)$. Thus we have to prove that $y(\bar x) \geq

1456: %%\xi(c) \sin [(1-\beta)(2b-a-\bar x)] \geq \xi(c) \sin

1457: %%[(1-\beta)(b-a)]$. By claim (a) and by using the inequality $y(\bar

1458: %%x) \geq y(b)$ it holds $y(\bar x) \geq \xi(c) \sin [(1-\beta)(b-a)]$

1459: %%holds. Therefore the proof of Claim (b) is complete and this

1460: %%concludes the proof of (i).

1461: %%

1462: %%To prove (ii) we argue as before. Let $\tilde x \in (0, b-c)$ and

1463: %%let us consider the solution $x(\cdot)$ starting at $\delta$. In the

1464: %%interval $[c, b]$ it satisfies the following inequality

1465: %%\[

1466: %%\dfrac{dx}{dy} \geq (1-\beta)^{-1}\left(2L_\varepsilon -

1467: %%y^2\right)^{-1/2}, \qquad \textrm{for}\ \ L_\varepsilon:=

1468: %%\xi(b-\varepsilon).

1469: %%\]

1470: %%By direct integration we have

1471: %%\[

1472: %%\int_{c}^{b-\varepsilon}dx \geq

1473: %%(1-\beta)^{-1}\int_{0}^{y(b-\varepsilon)}\left(2L_\varepsilon -

1474: %%y^2\right)^{-1/2}dy

1475: %%\]

1476: %%and therefore

1477: %%\[

1478: %%y(b-\varepsilon)\leq \sqrt{2\xi(b-\varepsilon)}\sin

1479: %%\left((b-c-\varepsilon)(1-\beta)\right).

1480: %%\]

1481: %%This implies in particular that $y(b-\varepsilon) < \sqrt{2\xi(c)}$.

1482: %%Moreover the bouncing solution in the interval $[c, b-\varepsilon]$

1483: %%satisfies the following

1484: %%\[

1485: %%\dfrac{dx}{dy}\leq - (1-\beta)^{-1}\left(2L_\varepsilon -

1486: %%y^2\right)^{-1/2},

1487: %%\]

1488: %% and integrating it follows

1489: %%\[

1490: %%\int_{b-\varepsilon}^{c}dx \leq -(1-\beta)^{-1}

1491: %%\int_{y(b-\varepsilon)}^{y(c)}\dfrac{1}{\sqrt{2L_\varepsilon -y^2}}

1492: %%\]

1493: %%which implies that $y(c) \leq y(b-\varepsilon)$. By using the above

1494: %%inequality $y(b-\varepsilon) <\sqrt{2\xi(c)}$, it holds $y(c) <

1495: %%\sqrt{2\xi(c)}$.

1496: %%\end{proof}

1497: %%

1498: %%As direct consequence of the above result, we are in position to

1499: %%prove the existence of some colliding solutions in the parabolic

1500: %%manifold $\parabolicmanifold$. Our first result is Theorem

1501: %%\ref{thm:diedraleplanare} which describes all the connecting orbits

1502: %%on the planar section, i.e. on the subset

1503: %%$\parabolicmanifold^{\planare}$ of the parabolic manifold fixed by

1504: %%the symmetry $h_\varphi$; the second result Theorem

1505: %%\ref{thm:diedraleprisma} describes all connecting orbits on the

1506: %%subset $\parabolicmanifold^{\prisma}$ of the parabolic manifold

1507: %%fixed by the symmetry $h_\theta$ and our last result, Theorem

1508: %%\ref{thm:diedralesimmetrico} describes all connecting orbits on the

1509: %%subset $\parabolicmanifold^{\antiprisma}$ of the parabolic manifold

1510: %%fixed by the symmetry $h'_\theta$. Before proceeding further with

1511: %%the description of the flow on the parabolic manifold some comments

1512: %%are in order. First of

1513: %%all in figure % \ref{fig:prismsections}

1514: %%\ref{fig:prismsection} it is represented the flow obtained by

1515: %%integrating the corresponding $1$-dimensional Cauchy problem for

1516: %%different values of the initial conditions. This flow is represented

1517: %%on the covering over each piece of $1$-sphere between the binary

1518: %%collision and the $l$-adic collision.  We remark that in the planar

1519: %%case the situation is a little bit more symmetric but the

1520: %%picture is very similar. In figure % \ref{fig:prismsections}

1521: %%\ref{fig:antiprismsection} it is represented the flow on the

1522: %%covering of a regularised component of the parabolic manifold for

1523: %%the $2n$-gon/antiprism section. In this case the situation is more

1524: %%complicated due to the presence of two

1525: %%different types of central configurations. %%By changing coordinates,

1526: %%%%locally in the neighborhood of binary collisions or $l$-adic

1527: %%%%collisions, the regularized component of the parabolic manifold in

1528: %%%%the planar and prism section resemble that studied by McGehee in

1529: %%%%\cite{McGehee74} for the collinear three body problem, while the

1530: %%%%regularized component of the parabolic manifold in the

1531: %%%%$2n$-agon/antiprism resemble that studied by Devaney in

1532: %%%%\cite{Devaney80} in the isosceles three body problem.

1533: %%\begin{theo}${\bf (h_\varphi)}$\label{thm:diedraleplanare}

1534: %%Let $\planare$ be the planar central configuration in the

1535: %%fundamental domain and for $j=0,1$ let $B^{s}_j$ be the stable

1536: %%manifold of the binary collision sets. Then the following

1537: %%connections and the dual ones occur in $\parabolicmanifold$.

1538: %%\begin{itemize}

1539: %%\item[(P1)] $\planare^+ \fromto B^{s}_j;$

1540: %%\item[(P2)] $ B^{u}_j\fromto \planare^+;$

1541: %%\item[(P3)] There is no saddle connections between two planar central configurations

1542: %%$\planare^{\pm}.$

1543: %%\end{itemize}

1544: %%\end{theo}

1545: %%\begin{proof}

1546: %%In order to describe the flow on $\parabolicmanifold^\planare$ we

1547: %%observe that in this case the potential $U$ depends only on $\theta$

1548: %%(since $\varphi=0$). Therefore the projection of the equations

1549: %%\eqref{eq:mcgheeonedim} on the parabolic manifold are represented by

1550: %%the following first order differential system

1551: %%\begin{comment}

1552: %%\todo{beta}

1553: %%\end{comment}

1554: %%\begin{equation}\label{eq:projectioneqmcgeheesfericheplanare}

1555: %%\left\{\begin{array}{ll}

1556: %%v' =  w^2 + \beta \big(v^2-2U(\theta)\big)\\

1557: %%\theta' = w\\

1558: %%w' =  (\beta-1) v w +U'(\theta)\\

1559: %%\end{array}\right.

1560: %%\end{equation}

1561: %%Now to prove this result it is enough to keep track of the unstable

1562: %%manifolds of the planar central configuration for $v>0$ on the

1563: %%$(\theta, v)$ plane. We will project the flow on

1564: %%$\parabolicmanifold^\planare$ on the $(\theta, v)$ plane and by the

1565: %%first two equations of motion

1566: %%\eqref{eq:projectioneqmcgeheesfericheplanare} we can eliminate the

1567: %%time dependence as in equation \eqref{eq:prima}, by setting

1568: %%\[

1569: %%\dfrac{dv}{d\theta} = \pm(1-\beta)\sqrt{2U(\theta)-v^2}.

1570: %%\]

1571: %%The last equation have to be integrated starting at the point

1572: %%$\planare^-$ forward up to $v=0$, namely at the point $Q$. Taking

1573: %%into account the regularisation given in formula

1574: %%\eqref{eq:mcgheeonedim_2} and by using the symmetry with respect to

1575: %%$\varphi = \pi/2l$ and $v=0$ it is clear that to have a connection

1576: %%between the lower point $\planare^-$, and upper $\planare^+$

1577: %%equilibrium points requires $\theta(Q)$ to be a multiple of

1578: %%$\pi/2l$.

1579: %%

1580: %%In order to show that $W^s(\planare^+)\not= W^u(\planare^-)$, it is

1581: %%enough to see that starting at the origin of the $(\theta, v)$ plane

1582: %%and going backwards we reach the curve $v = \sqrt{2U(\theta)}$ to

1583: %%the right of the point $\planare^-$ and starting at the point

1584: %%$P_2\equiv(\pi/2l,0)$ we reach $\theta=\pi/2l$ above point

1585: %%$\planare^-$. It is enough to prove the existence of two vector

1586: %%fields $f$ and $g$ such that

1587: %%\[

1588: %%\Big\vert\dfrac{d\varphi}{dv}\Big\vert \leq f, \qquad

1589: %%\Big\vert\dfrac{d\varphi}{dv}\Big\vert \geq g

1590: %%\]

1591: %%the assertions above are true respectively for $f$ and $g$. But this

1592: %%is a consequence of Lemma \ref{thm:lemmatecnico} simply by setting

1593: %%$x = \theta$ and $\xi(x)= U(\theta)$.

1594: %%\end{proof}

1595: %%\begin{theo}${\bf (h_\theta)}$\label{thm:diedraleprisma}

1596: %%Let $\prisma^+$ be the prism type central configurations in the

1597: %%fundamental domain for $\theta=0$  and let  $B^{u}_0$ be the binary

1598: %%collision escape set. Then the following connections and the dual

1599: %%ones occur in $\parabolicmanifold^{\prisma}$.

1600: %%\item[(R1)] $\prisma^+ \fromto P^{s}_+;$

1601: %%\item[(R2)] $ B^{u}_0\fromto \prisma^+$ and $ P^{u}_-\fromto \prisma^+;$

1602: %%\item[(R3)] $\prisma^+ \fromto B^{s}_0;$

1603: %%\item[(R4)] There is no saddle connections between two prism type central configurations

1604: %%$\prisma^{\pm}.$

1605: %%\end{theo}

1606: %%\begin{proof}

1607: %%This result can be proved by arguing as before and by applying this

1608: %%time Lemma \ref{thm:lemmatecnico} for $x= \varphi$ and $\xi(x)=

1609: %%U(\varphi)$.

1610: %%\end{proof}

1611: %%

1612: %%

1613: %%%%%=========================================================================

1614: %%\section{Non-colliding parabolic connections }

1615: %%\label{sec:noncollidingparabolic}

1616: %%

1617: %%

1618: %%\begin{theo}${\bf (h'_\theta)}$\label{thm:diedralesimmetrico}

1619: %%For $i=0,1$ let $\antiprisma_i^+, \planare\in (-\pi/2, \pi/2) $ be

1620: %%the anti-prism and planar central configurations and let $P^{s}_+$

1621: %%be the stable collision escape sets. Then the following connections

1622: %%and the dual ones occur in $\parabolicmanifold^{\antiprisma}$.

1623: %%\begin{itemize}

1624: %%\item[(Q1)] $P^{u}_\pm\fromto \planare^+$;

1625: %%\item[(Q2)] $\antiprisma_i^+\fromto \planare^+$;

1626: %%\item[(Q3)] $\antiprisma_i^+\fromto P^{s}_{\pm} $.

1627: %%\end{itemize}

1628: %%\end{theo}

1629: %%\begin{proof} To prove the result it is

1630: %%enough to show the existence of the following dual connections:

1631: %%$\planare_- \fromto P^s_\pm$, $\planare_- \fromto \antiprisma_i^-$

1632: %%and $P^u_+ \fromto \antiprisma_i^-$. First of all we observe that,

1633: %%by existence and uniqueness for the Cauchy problem the symmetries

1634: %%and the gradient-like character of the flow with respect to $v$, we

1635: %%need only to prove that

1636: %%\begin{enumerate}

1637: %%\item[(I)] Starting at the origin of the $(\varphi, v)$ plane and going

1638: %%backwards we reach the curve $v = \sqrt{2U(\varphi)}$ to the right

1639: %%of the point $A_1^- \equiv (\varphi^*, -\sqrt{2U(\varphi^*)})$

1640: %%corresponding to the projection on this plane of the central

1641: %%configuration $ \antiprisma_1^-$;

1642: %%\item[(II)] Starting at the point $A_1^- \equiv (\varphi^*, -\sqrt{2U(\varphi^*)})$

1643: %%we reach $\varphi-\pi/2$ with a negative value of $v$.

1644: %%\end{enumerate}

1645: %%In what follows we will refer to the {\em right (left) unstable

1646: %%branch of $W^u(\antiprisma_1^- )$\/} as that having $w >0$, ($w<0$)

1647: %%in a sufficiently small neighbourhood of $\antiprisma_1^- $ in

1648: %%$\parabolicmanifold^{\prisma}$.

1649: %%

1650: %%With this in mind we remark that the first item corresponds to the

1651: %%following situation. The right branch of $W^u(\antiprisma_1^-)$

1652: %%intersects $\varphi= \pi/2$ with $v<0$ and then intersects the

1653: %%section $v=0$ with $\varphi >0$. This proves the existence of the

1654: %%connecting orbit between $\antiprisma_1^-$ and $P^s_-$. The second

1655: %%item, however, corresponds to the fact that the left branch of

1656: %%$W^u(\antiprisma_1^-)$ intersects $\varphi= \pi/2$ with $v<0$ and

1657: %%then intersects the section $v=0$ with $\varphi <0$.

1658: %%

1659: %%The proof of (I) easily follows by applying Lemma

1660: %%\ref{thm:lemmatecnico} to the projection onto the $(\varphi, v)$

1661: %%plane of the right unstable branch of $\antiprisma_1^-$ by setting

1662: %%$x=\varphi$ and $\xi(x)= U(\varphi)$.

1663: %%

1664: %%

1665: %%To see this it remains only to show that the left unstable branch of

1666: %%this anti-prism central configuration meets $\varphi=-\pi/2$ at a

1667: %%negative value of the Lyapunov function $v$.

1668: %%

1669: %%We prove (II). Let us consider the projection of this orbit on the

1670: %%$(\varphi,v)$ plane. The following differential inequality holds

1671: %%true:

1672: %%\[

1673: %%\dfrac{d\varphi}{dv} \leq -\dfrac{1}{(1-\beta)\sqrt{v^2(\varphi_*)-

1674: %%v^2}}.

1675: %%\]

1676: %%By direct integration it follows that

1677: %%\[

1678: %%\int_{\varphi_*}^{-\pi/2}d\varphi \leq - \int_{-v^*}^{v(-\pi/2)}

1679: %%\dfrac{1}{(1-\beta)\sqrt{v^2(\varphi_*)-v^2}}\, dv

1680: %%\]

1681: %%and therefore $v(-\pi/2) \leq v(\varphi_*)\sin \varphi_*$, which is

1682: %%negative.

1683: %%\end{proof}

1684: %%

1685: %%

1686: %%\begin{theo}\label{thm:diedralecompleto}

1687: %%For $i=1,2$ let $\antiprisma^+, \prisma_i^+$  be the anti-prism and

1688: %%prism central configurations in the upper-half of the fundamental

1689: %%domain and for $j=0,1$ let $B^{s}_j$ and $P^{s}_+$ be the stable

1690: %%collision escape sets. Then the following connections and the dual

1691: %%ones occur in $\parabolicmanifold$.

1692: %%\begin{itemize}

1693: %%\item[(AP1)] $\antiprisma_1^+\fromto \prisma_i^+ $;

1694: %%%%\item[(AP2)] $\antiprisma^+\fromto \planare^+ $;

1695: %%\item[(AP2)] $\antiprisma_1^+\fromto B^{s}_j$;

1696: %%%%\item[(AP4)] $\antiprisma^+\fromto P^{s}_+$;

1697: %%\end{itemize}

1698: %%\end{theo}

1699: %%The proof of the last result will be the content of the last part of

1700: %%this section and it is based upon a topological idea already

1701: %%introduced in \cite{Moeckel83}.

1702: %%

1703: %%We will denote by the symbol $\parabolicmanifold_\lambda$  the level

1704: %%set of the Lyapunov function $v$ i.e.

1705: %%\[\parabolicmanifold_\lambda\, :=\, \{(v,\vs,\vw) \in \parabolicmanifold; \ \

1706: %%v \geq \lambda\}.\] Clearly since $v$ is an increasing function it

1707: %%follows that for all $\lambda \in \R$ all of these sets are

1708: %%invariant for the flow induced by the function $v$.

1709: %%

1710: %%Let $\Pi \colon \parabolicmanifold \longrightarrow S$ (defined by

1711: %%$(v,\vs,\vw) \mapsto \vs$) be the canonical projection onto the

1712: %%inertia ellipsoid. We will set $S_\lambda:=

1713: %%\Pi(\parabolicmanifold_\lambda)$. Taking into account the inequality

1714: %%$v^2 \leq 2 U(\vs)$ it follows that

1715: %%\[

1716: %%S_\lambda = \, \{\vs \in  S\colon U(\vs) \geq \lambda^2/2\}.

1717: %%\]

1718: %%

1719: %%We observe that since each central configuration is a hyperbolic

1720: %%rest-point, it follows that $W^u(\antiprisma^+)\cap

1721: %%\parabolicmanifold$ is an immersed disk. Therefore denoting by

1722: %%$c:=v(\antiprisma^+)$ the level set of the function $v$ at the

1723: %%anti-prism type central configuration and by bar its closure then

1724: %%for all positive $\varepsilon$ the set defined by

1725: %%\[

1726: %%\overline{S}_{c+\varepsilon}  \, := \, \big\{s \in S \colon \ \

1727: %%U(\vs) \geq (c+ \varepsilon)^2/2\big\}

1728: %%\] is just the ellipsoid with small disks

1729: %%about all anti-prism-type central configurations removed.

1730: %%

1731: %%\begin{nr}\label{thm:lemmettinofurbo}

1732: %%Let $\alpha \colon S^1 \to \parabolicmanifold_{c+\varepsilon}$ be an

1733: %%analytic closed curve such that $\Pi \circ \alpha \colon S^1 \to

1734: %%S_{c+\varepsilon}$ is not contractible in

1735: %%$\overline{S}_{c+\varepsilon}$. Then $\alpha$ intersects $B^{s}_j$

1736: %%and contains one-cells complementary to

1737: %%$W^s(\prisma_i^+)\cap\parabolicmanifold$.

1738: %%\end{nr}

1739: %%\begin{proof}

1740: %%(Sketch of) Given the solution curve defined by

1741: %%\[

1742: %%\gamma_\alpha\big(t,\alpha(x)\big)=\big(\vs(t,\alpha(x)),\vw(t,

1743: %%\alpha(x))\big)

1744: %%\]

1745: %% we define the closed curve $\Pi(\gamma_\alpha):=

1746: %%\vs\big(t,\alpha(x)\big)$. Since $\alpha$ is contained in a positive

1747: %%invariant set for the flow, this implies that for all $t \geq 0$ its

1748: %%projection is clearly homotopic to $\Pi(\gamma_\alpha)$ in

1749: %%$\overline {S}_{c+ \varepsilon}$. Since $\alpha$ and

1750: %%$W^s(\prisma_i^+)\cap\parabolicmanifold$ are analytic we have the

1751: %%following two situations. Either $\alpha \subset

1752: %%W^s(\prisma_i^+)\cap\parabolicmanifold$, or the intersection $\alpha

1753: %%\cap W^s(\prisma_i^+)\cap\parabolicmanifold$ consists of finitely

1754: %%many points. If $\alpha$ fails to have topologically transverse

1755: %%intersection with all $W^s(\prisma_i^+)\cap\parabolicmanifold$, then

1756: %%either $\alpha \cap W^s(\prisma_i^+)\cap \parabolicmanifold =

1757: %%\emptyset$ or $\alpha \cap W^s(\prisma_i^+)\cap \parabolicmanifold$

1758: %%consists of finitely many points near each of which $\alpha$ fails

1759: %%to cross from one complement of

1760: %%$W^s(\prisma_i^+)\cap\parabolicmanifold$ to the other. In any case

1761: %%there exists a simple closed curve $\beta \subset

1762: %%\parabolicmanifold_{c + \varepsilon}$ arbitrarily close to $\alpha$

1763: %%such that $\beta \cap

1764: %%W^s(\prisma_i^+)\cap\parabolicmanifold=\emptyset$. Moreover if

1765: %%$\beta $ is close enough to $\alpha$ then the curve will be not

1766: %%contractible in $\overline{S}_{c+\varepsilon}$. Choose a positive

1767: %%constant $C$ larger than the largest critical value of $v$. Since

1768: %%orbits in $\parabolicmanifold_C$ cannot tend to rest-points then

1769: %%$\parabolicmanifold_C \subset B^s_0 \cup B^s_1 \cup P^s_+$.

1770: %%

1771: %%Consider the one dimensional manifold $W^u(\prisma_i^+)\cap

1772: %%\parabolicmanifold$. In this case by Theorem

1773: %%\ref{thm:diedraleprisma}, one side enters in $B^s_j$ and the other

1774: %%in $P^s_+$. Therefore the projection of $W^u(\prisma_i^+)$  is

1775: %%contained respectively in one of the two arcs representing the edges

1776: %%of the upper half of the fundamental domain.

1777: %%

1778: %%Now let us consider a small disk of the $\prisma_i^+$ in the

1779: %%parabolic manifold and let $I(\prisma_i^+)$ be the cylinder obtained

1780: %%by keeping the trace of the disk flowing up to its entrance in

1781: %%$\parabolicmanifold_C$ and let $U(\prisma_i^+)$ be the projection of

1782: %%this cylinder on the ellipsoid.

1783: %%

1784: %%Suppose that $\beta \cap W^s(\prisma_i^+)=\emptyset$ and let

1785: %%$\bigcup_{\prisma_i^+ \in \mathcal D(D_l)} I(\prisma_i^+)\cup

1786: %%\parabolicmanifold_C.$ This is a positively invariant set into

1787: %%which every point of $\beta$ flows. Therefore denoting by

1788: %%$\gamma_\beta\big(t,\beta(x)\big)$ the curve given by

1789: %%$\big(\vs(t,\beta(x)), \vw(t, \beta(x))\big)$ it follows that if

1790: %%$T>0$ sufficiently large than for $t \geq T$ the projection of the

1791: %%curve $\gamma_\beta$ is contained in the region of the ellipsoid

1792: %%given by

1793: %%\[

1794: %%\bigcup_{\prisma_i^+ \in \mathcal D(D_l)} U(\prisma_i^+)\cup

1795: %%\overline{S}_C.

1796: %%\]

1797: %%Hence the curve $\Pi(\gamma_\beta)$ is contractible. Moreover this

1798: %%curve is homotopic to the curve $\beta$ (by the flow homotopy) and

1799: %%therefore also $\beta$ is contractible and this is a contradiction.

1800: %%

1801: %%We have shown that the original curve $\alpha$ contains

1802: %%complementary one cells to each stable manifold of the saddle

1803: %%central configurations and by the above construction it follows that

1804: %%any complementary one cell to $W^s(\prisma_i^+)$ intersects the

1805: %%escape set.

1806: %%\end{proof}

1807: %%

1808: %%

1809: %%

1810: %%%\begin{figure}\centering

1811: %%%\includegraphics[width=0.5\textwidth]{figs/connections}

1812: %%%\caption{Connections}

1813: %%%\label{fig:mf}

1814: %%%\end{figure}

1815: %%%

1816: %%\newcommand{\cerchio}[1]{\save[]*\frm{#1}\restore}

1817: %%

1818: %%\begin{figure}

1819: %%% \entrymodifiers={++[o][F-]}

1820: %%\begin{center}\SelectTips{xy}{}

1821: %%\[\xymatrix@C-16pt@R-18pt{%

1822: %%*+[F-:<3pt>]{\text{binary}} & & &  *+[F-:<3pt>]{\text{$l$-adic}} \\

1823: %%& *+[F-:<3pt>]{\text{$2l$-gon}} & & &  \\

1824: %%& & *+[F-:<3pt>]{\text{prism}} \ar@/_2ex/[ruu] \ar@/_4ex/[lluu] & &  \\

1825: %%& &  *+[F-:<3pt>]{\text{antiprism}} \ar[u] \ar@/_3ex/[ruuu] \ar@/^2ex/[luu] \ar@/^3ex/[lluuu]&  \\

1826: %%& &  *+[F-:<3pt>]{\text{antiprism}} \ar@/_4ex/[ruuuu]  &  \\

1827: %%& & *+[F-:<3pt>]{\text{prism}} \ar[u] \ar@/^4ex/[lluuuuu] \ar@/_6ex/[ruuuuu] & &  \\

1828: %%& *+[F-:<3pt>]{\text{$2l$-gon}} \ar@/^2ex/[ruu] \ar@/^4ex/[luuuuuu] & & & \\

1829: %%*+[F-:<3pt>]{\text{binary}}

1830: %%\ar@/^8ex/[uuuuuuurrr] \ar@/^8ex/[uuuuuur] \ar@/^5ex/[uuuuurr]

1831: %%\ar@/^6ex/[rruu] \ar@/^2ex/[ur] \ar@/^6ex/[uuurr] & & &

1832: %%*+[F-:<3pt>]{\text{$l$-adic} } \ar@/_6ex/[luu] \ar@/_6ex/[luuu]

1833: %%\ar@/_14ex/[llluuuuuuu] \ar@/_8ex/[luuuuu] \ar@/_8ex/[luuuu] }

1834: %%\]

1835: %%\end{center}

1836: %%\caption{ All connecting orbits (together with the dual ones)

1837: %%between central configurations on $\parabolicmanifold$.}

1838: %%\label{fig:mf}

1839: %%\end{figure}

1840: %%

1841: %%

1842: %%

1843: %%%%%=========================================================================

1844: %%\section{Topological transversality and Palis inclination lemma}

1845: %%\label{sec:topologicaltransversality} The main goal of this section

1846: %%is to show how to use the connecting orbits on the

1847: %%$\parabolicmanifold$ in order to establish the motion of the

1848: %%dihedral $2n$-body problem. We recall a result analogous to the {\em

1849: %%Palis inclination lemma\/} proved by Moeckel in \cite{Moeckel83}

1850: %%which will provide the key ingredient of this section.

1851: %%\begin{nr}

1852: %%\label{thm:lambdalemma} Let $p$ be a hyperbolic rest-point of a flow

1853: %%on a manifold $ N$. If a positively invariant manifold $ P$ has a

1854: %%non-empty topologically transverse intersection with $W^s(p)$, then

1855: %%$W^u(p) \subset \overline{ P}$.

1856: %%\end{nr}

1857: %%For the proof see for instance Lemma 4.2 in \cite{Moeckel83}.

1858: %%\begin{propo}\label{thm:primaconclusione} The following inclusion between

1859: %%invariant manifold hold.

1860: %%\begin{enumerate}

1861: %%\item[(i)] $W^s(\planare^-) \subset \overline{W^s(\antiprisma^-)}$ and

1862: %%$W^u(\planare^+) \subset  \overline{W^u(\antiprisma^+)}$;

1863: %%\item[(ii)]$W^s(\prisma_i^-) \subset \overline{W^s(\antiprisma^-)}$ and

1864: %%$W^u(\prisma_i^+) \subset  \overline{W^u(\antiprisma^+)}$ for

1865: %%$i=1,2$.

1866: %%\end{enumerate}

1867: %%\end{propo}

1868: %%\begin{proof}

1869: %%First of all we observe that by symmetry in order to prove

1870: %%$(i)-(ii)$ it is enough to prove that $W^u(\planare^+) \subset

1871: %%\overline{W^u(\antiprisma^+)}$ and $W^u(\prisma_i^+) \subset

1872: %%\overline{W^u(\antiprisma^+)}$. Since the proof in this two cases is

1873: %%the same, we will prove that

1874: %%\[

1875: %%W^u(Q^+) \subset \overline{W^u(\antiprisma^+)}, \qquad \textrm{for}

1876: %%\ \ Q = \planare, \prisma_i.\]

1877: %%

1878: %%This statement will be consequence of the fact that $W^u(Q^+)

1879: %%\subset  \overline{W^u(\antiprisma^+)}$ on the parabolic manifold

1880: %%and of the fact that the complementary $\Sigma$ of the parabolic

1881: %%manifold in the full phase space is open.

1882: %%

1883: %%

1884: %%Setting $W_\Sigma^{u,s}=W^{u,s}\cap \Sigma$, it will be sufficient

1885: %%to prove that $W^u_\Sigma(\antiprisma^+)$ has a topologically

1886: %%transverse intersection with $W^s_\Sigma(Q^+)$ in $\Sigma$.

1887: %%

1888: %%>From Theorem \ref{thm:diedralecompleto}, $W^u(\antiprisma^+)$ has a

1889: %%topologically transverse intersection with $W^s(Q^+)$ in

1890: %%$\parabolicmanifold$. Thus there exists a one cell

1891: %%\[

1892: %%\alpha \colon (D^1, D^1\backslash(0))\longrightarrow

1893: %%\big(W^u(\antiprisma^+),W^u(\antiprisma^+)\backslash

1894: %%W^u(\antiprisma^+)\cap W^s(Q^+)\big)

1895: %%\]

1896: %%complementary to the stable manifold of $Q^+$. By Table

1897: %%\eqref{tb:tabella1} it is worth noticing that the dimension of the

1898: %%stable manifold of $Q^+$ in the parabolic manifold is the same that

1899: %%in $\Sigma$, while the dimension of the unstable manifold in

1900: %%$\Sigma$ increases by one.

1901: %%

1902: %%We observe that  $D^1$ can be seen as a diameter of the two

1903: %%dimensional disk $D^2= D^2_+\cup D^2_-$ and since

1904: %%$(W^u(\antiprisma^+)\cap \parabolicmanifold$ divides

1905: %%$W^u_\Sigma(\antiprisma^+)$ into two components one for $r>0$ and

1906: %%one for $r<0$, we can extend $\alpha$ to a map $\bar \alpha \colon

1907: %%D^2 \to W^u_\Sigma(\antiprisma^+)$ taking $D^2_+$ to $r>0$ and

1908: %%$D^2_-$ to $r<0$.

1909: %%

1910: %%Such an extension gives a complementary two disc to

1911: %%$W^s_\Sigma(Q^+)$: to do so, it is enough to introduce local chart

1912: %%and to show that $\bar \alpha $ is a generator of the second

1913: %%homology group of  $(\R^5, \R^5\backslash \R^3 \times 0^2)$ or which

1914: %%is the same that its restriction to the boundary of $D^2$ winds one

1915: %%time around the origin in the $0^3 \times \R^2$ plane and this is

1916: %%immediate to check. \end{proof}

1917: %%

1918: %%

1919: %%

1920: %%% %%%=========================================================================

1921: %%% \section{Invariant manifolds and connections on the total collision manifold}

1922: %%% \label{sec:invariantmanifolds}

1923: %%% \[

1924: %%% A \fromto B

1925: %%% \]

1926: %%

1927: %%%%%=========================================================================

1928: %%\section{Conclusions and remarks}

1929: %%\label{sec:Closingremarks}

1930: %%

1931: %%We close this section with some dynamical consequences of

1932: %%Proposition \ref{thm:primaconclusione}.

1933: %%

1934: %%\begin{propo}

1935: %%Arbitrarily close to every planar or prism collision orbit there are

1936: %%antiprism type collision orbits. Moreover, arbitrarily close to

1937: %%every antiprism type ejection orbit there are planar and prism

1938: %%ejection orbits.

1939: %%\end{propo}

1940: %%

1941: %%Based upon the existence of connecting orbits between some central

1942: %%configurations corresponding to a negative value of the function $v$

1943: %%and some others corresponding to the positive ones, it is possible

1944: %%to derive some dynamical consequences. For instance, let us denote

1945: %%the vertices of the regular $2l$-gon by $P_k $ for $k=0, 1, \dots,

1946: %%2l-1$.

1947: %%\begin{propo}

1948: %%Suppose that  there exists the connection $B^{u}_j\fromto B^{s}_i$

1949: %%on $\parabolicmanifold$. Then there is an open set of orbits in the

1950: %%interior of $\Sigma$ with the following behavior $r(t) \to \infty \

1951: %%\ \textrm{as}\ \ t \to \pm \infty,$ but as $t \to -\infty$ the

1952: %%configuration of the system is an irregular $n$-gon in which the

1953: %%edge with vertices $P_{j-1}, P_j$ let us say $\overline{P_{j-1}P_j}$

1954: %%remains bounded and as $t \to + \infty$ the edge with vertices

1955: %%$P_{i-1}, P_i$ remains bounded.

1956: %%

1957: %%Suppose that the following connection $B^{u}_j\fromto P^{s}_\pm$

1958: %%occur on $\parabolicmanifold$. Then there is an open set of orbits

1959: %%in the interior of $\Sigma$ with the following behaviour $r(t) \to

1960: %%\infty \ \ \textrm{as}\ \ t \to \pm \infty, $ but as $t \to -\infty$

1961: %%the configuration of the system is an irregular $n$-gon in which the

1962: %%edge with vertices $P_{j-1}, P_j$ remains bounded and as $t \to +

1963: %%\infty$ the configuration of the system is a prism in which the top

1964: %%and bottom faces are regular $p$-gons which remains bounded.

1965: %%\end{propo}

1966: %%The proof of this result follows by integrating the first equation

1967: %%of the McGehee equations.

1968: %%

1969: %%The last dynamical feature of the problem is a direct consequence of

1970: %%the connecting orbits on the parabolic manifold between a central

1971: %%configuration for a negative value of $v$ and the asymptotic escape

1972: %%sets.

1973: %%

1974: %%\begin{propo}

1975: %%Suppose $R^-\fromto S $ where $R$ is prism, anti-prism or planar

1976: %%central configuration and $S$ is one escape set. Then arbitrarily

1977: %%close to every $R^-$ collision orbit, there are orbits with the

1978: %%following behavior: after reaching a minimum, $r(t)$ tends

1979: %%monotonically to infinity in such a way that the configuration is

1980: %%for all time an irregular $n$-gon (or an irregular $n$-prism) where

1981: %%$l$ edges (resp. top and bottom faces) remains bounded for all time

1982: %%$t$.

1983: %%\end{propo}

1984: %%\begin{proof}

1985: %%Let $\mathcal U$ be any neighborhood of a point $p$ on a $R^-$

1986: %%collision orbit and let us consider the forward orbits of the points

1987: %%in $\mathcal U$ namely $\psi\big(\mathcal U)$. Since

1988: %%$\psi\big(\mathcal U_p)$ is topologically transverse to $W^s(R^-)$

1989: %%and positively invariant, by using Lemma \ref{thm:lambdalemma} we

1990: %%conclude that $W^u(R^-)\subset \psi\big(\mathcal U)$ and therefore

1991: %%there are points in $\mathcal U$ which pass arbitrarily close to a

1992: %%$R^-\fromto S$ connecting orbit. Repeating to any point passing

1993: %%close enough to such orbit this argument, we conclude the proof.

1994: %%\end{proof}

1995: %%\begin{remark}

1996: %%All these results reveal in a certain sense a chaotic behaviour of

1997: %%the total collision orbits in the sense that all these orbits are

1998: %%extremely sensitive to small changes in initial conditions. For

1999: %%example in the first result of this section we have seen that after

2000: %%passing close to total planar or prism type collision, the system

2001: %%emerges with arbitrarily large kinetic energy. In particular in

2002: %%every neighborhood of every planar or prism type collision orbit

2003: %%there are anti-prism collision orbits as well as orbits which avoid

2004: %%collision and emerge from a neighborhood of the singularity in any

2005: %%of the irregular $n$-gonal or $n$-prism type central configuration

2006: %%and the same holds for orbits in a neighborhood of a anti-prism type

2007: %%central configuration.

2008: %%

2009: %%In the the particular case of the $D_2$ symmetric four body problem

2010: %%as already proved in \cite{DelgadoVidal99} when a total collapse is

2011: %%reached (asymptotical to a planar central configuration), the

2012: %%solution oscillates very rapidly passing through an arbitrarily

2013: %%number of instantaneous planar configurations (this is due to the

2014: %%spiralling character of the sink) until a double collision occurs,

2015: %%followed by a passage through a planar central configuration and

2016: %%further escape from a neighborhood of total collapse with multiple

2017: %%binary collisions. An approach to total collision asymptotically to

2018: %%a tetrahedral type central configuration is very unstable in the

2019: %%sense that after a close encounter is a double collision followed by

2020: %%a planar configuration and then an escape from total collision with

2021: %%multiple binary collisions.

2022: %%\end{remark}

2023: %%\begin{figure}\centering

2024: %%\hfill\subfigure[Four bodies with dihedral symmetry and anti-symmetric constraint.]{%

2025: %%\includegraphics[width=0.2\textwidth]{figs/klein4}\label{fig:figs/klein4}

2026: %%}\hfill

2027: %%\subfigure[Four bodies with dihedral symmetry and semi-coreography constraint.]{%

2028: %%\includegraphics[width=0.2\textwidth]{figs/dihedral4}\label{figs/dihedral4}

2029: %%} \hfill

2030: %%\subfigure[Twelve bodies with dihedral symmetry and 3-coreography constraint.]{%

2031: %%\includegraphics[width=0.2\textwidth]{figs/dihedral4_12b_choreography}\label{figs/dihedral4_12b_choreography}

2032: %%} \hfill \caption{Some symmetric periodi orbits minimizing the

2033: %%action}

2034: %%\end{figure}

2035: %%\begin{remark}

2036: %%Variational methods under symmetry constraints have been popular in

2037: %%the recent years, starting from Chenciner and Montgomery ``figure

2038: %%eight'' orbit \cite{chenICM,monchen}. Symmetric periodic minimisers

2039: %%have been found by many authors (to name a few,

2040: %%\cite{ambrosetticoti,chenven,SalomoneXia,Chen03,Chen03b}). Periodic

2041: %%orbits dihedrally symmetric can be shown to exist as well: from

2042: %%\cite{gg} (see also \cite{zz,dd,oo}) it follows that if $x(t)$ is a

2043: %%minimiser of the Lagrangean action functional, constrained to the

2044: %%Sobolev space of $G$-equivariant periodic trajectories (where $G$ is

2045: %%a cyclic extension of $D_l$), then $x(t)$ is collision-less. In the

2046: %%Figures \ref{fig:figs/klein4}, \ref{figs/dihedral4} and

2047: %%\ref{figs/dihedral4_12b_choreography}, some of the resulting orbits

2048: %%are shown, which are symmetric with respect to some cyclic

2049: %%extensions of the group $D_l$.

2050: %%\end{remark}

2051: %%

2052: 

2053: 

2054: \appendix

2055: \section[Tua madre]{An integral representation for $U$}\label{sec:appendice}

2056: 

2057: The aim of this section is to give a direct proof of the integral

2058: representation for the potential $U$ used before in order to compute

2059: all the central configurations.

2060: 

2061: For $l\geq 2$, let $\perron_l$ denote the $l$-adic Perron-Frobenius

2062: operator, defined on complex functions $f\from S^1\subset \CC \to

2063: \CC$ by

2064: \[

2065: \begin{aligned}

2066: \forall \xi=e^{i\theta} \in S^1, \perron_l(f)(\xi) & = \dfrac{1}{l}

2067: \sum_{y\,:\, y^l = \xi} f(y)

2068: %%&=

2069: %%\dfrac{1}{l}

2070: %%\sum_{\substack{j=0\ldots (l-1)\\ y_1: y_1^l = \zeta_l}} {f(\zeta_l^j y_1)} \\

2071: %%\\

2072: =  \dfrac{1}{l} \sum_{j=0}^{l-1} f( e^{\frac{i(\theta + 2j\pi)}{l} }

2073: ).

2074: \end{aligned}

2075: \]

2076: For each $k\in \ZZ$,

2077: \begin{equation}\label{eq:proprperron}

2078: \perron_l(\xi^k)(\xi)  = \dfrac{1}{l} \sum_{y\,:\, y^l = \xi} y^k =

2079: \begin{cases}

2080: \xi^{k/l}   &

2081: \text{if $k\equiv 0 \mod l$} \\

2082: 0 & \text{if $k \not\equiv 0 \mod l$.}

2083: \end{cases}

2084: \end{equation}

2085: In terms of the $l$-adic Perron-Frobenius operator,  the potential

2086: \ref{eq:potentialplain} can be written as

2087: 

2088: \begin{equation}\label{eq:potentialperron}

2089: U = |z|^{-\alpha} \left[ c_l  + r^{-\alpha/2} l \perron_l\left(

2090: \left| 1 - r \xi \right| ^{-\alpha} \right)(\xi^l) \right],

2091: \end{equation}

2092: where $c_l$ is the constant $c_l  = \sum_{j=1}^{l-1} | 1 - \zeta_l^j

2093: |^{-\alpha}$ and $\perron_l\left( \left| 1 - r \xi \right|

2094: ^{-\alpha} \right)(\xi^l)$ denotes the function $\perron_l\left(

2095: \left| 1 - r \xi \right| ^{-\alpha} \right)$ of argument $\xi$

2096: evaluated at $\xi^l$. In order to compute  $\perron_l\left( \left| 1

2097: - r \xi \right| ^{-\alpha} \right)$, we expand $|1-r\xi|^{-\alpha}$

2098: in a double power series as follows.

2099: 

2100: \begin{nr}\label{nr:stimaperron1}

2101: For each $r\in (0,1]$ and $\alpha=2\beta > 0 $

2102: \[

2103: |1-r\xi|^{-\alpha} = \sum_{n=-\infty}^{+\infty} b_n \xi^n,

2104: \]

2105: with, for each $n\geq 0$,

2106: \[

2107: b_n  = b_{-n} = \dfrac{\sin( \beta \pi)}{ \pi }   r^n \int_0^1

2108: (1-t)^{-\beta} t^{\beta-1}  t^n {(1-tr^2)}^{-\beta} \,dt.

2109: \]

2110: \end{nr}

2111: \begin{proof}

2112: \[

2113: \begin{aligned}

2114: |1-r\xi|^{-\alpha} &

2115: = (1- r\xi)^{-\beta}(1-r\xi^{-1})^{-\beta} \\

2116: &= \left( \sum_{k=0}^\infty \binom{-\beta}{k} (-r\xi)^k \right)

2117: \cdot

2118: \left( \sum_{h=0}^\infty \binom{-\beta}{h} (-r\xi^{-1})^h \right)\\

2119: &= \sum_{h,k=0}^\infty \binom{-\beta}{k}  \binom{-\beta}{h}

2120: (-r)^{k+h} \xi^{k-h}  \\ &= \sum_{n=-\infty}^\infty

2121: \underbrace{\left( (-1)^n \sum_{\substack{k-h=n\\k,h\geq 0}}

2122: \binom{-\beta}{k} \binom{-\beta}{h}  r^{k+h} \right)}_{ b_n } \xi^n

2123: \end{aligned}.

2124: \]

2125: Now, recall that  for each $\beta>0$ and $N$ integer

2126: \[\begin{aligned}

2127: \binom{-\beta}{N} =  (-1)^N \binom{N+\beta - 1}{N} & = (-1)^N

2128: \dfrac{\Gamma(N+\beta) \Gamma(1-\beta)}%

2129: {\Gamma(N+1)\Gamma(\beta)\Gamma(1-\beta)}  \\

2130: &= \dfrac{(-1)^N }{ \Gamma(\beta)\Gamma(1-\beta) } \cdot

2131: \dfrac{\Gamma(N+\beta) \Gamma(1-\beta)}{\Gamma(N+1)} \\

2132: &= \dfrac{(-1)^N \sin( \beta \pi)}{ \pi } \cdot

2133: B ( 1-\beta, N+\beta ) \\

2134: \end{aligned}\]

2135: where $B(x,y)$ denotes the beta function, defined as

2136: \[

2137: B(x,y) = \int_0^1 t^{x-1} (1-t)^{y-1}\,dt =

2138: \dfrac{\Gamma(x)\Gamma(y)}{\Gamma(x+y)}

2139: \]

2140: and we have used the equalities

2141: \[

2142: \Gamma(\beta)\Gamma(1-\beta) = \dfrac{\pi}{\sin (\beta \pi ) },

2143: \]

2144: \[

2145: \binom{-\beta}{N} = \dfrac{(-\beta)(-\beta-1) \ldots (-\beta-N+1)}{

2146: N!} = (-1)^N \dfrac{\Gamma(N+\beta)}{\Gamma(N+1)\Gamma(\beta)}

2147: \]

2148: and

2149: \[

2150: \binom{\beta}{N} =

2151: \dfrac{\Gamma(\beta+1)}{\Gamma(N+1)\Gamma(\beta-N+1)}.

2152: \]

2153: We can now use the integral representation of the binomial function

2154: \begin{equation}\label{eq:cinque}

2155: \binom{-\beta}{N} = (-1)^N\dfrac{\sin( \beta \pi)}{ \pi } \int_0^1

2156: (1-t)^{-\beta} t^{\beta-1} t^N \,dt,

2157: \end{equation}

2158: which implies that, by setting $N=h+n$,

2159: \[\begin{aligned}

2160: b_n  &= (-1)^n \sum_{h=0}^\infty \left( (-1)^{n+h}\dfrac{\sin( \beta

2161: \pi)}{ \pi } \int_0^1 (1-t)^{-\beta} t^{\beta-1} t^{n+h} \,dt

2162:  \binom{-\beta}{h}  r^{n+2h}

2163: \right) \\

2164: &= \dfrac{\sin( \beta \pi)}{ \pi }   r^n \sum_{h=0}^\infty \left(

2165: (-1)^{h} \int_0^1 (1-t)^{-\beta} t^{\beta-1} t^{n+h} \,dt

2166:  \binom{-\beta}{h}  r^{2h}

2167: \right) \\

2168: &= \dfrac{\sin( \beta \pi)}{ \pi }   r^n \int_0^1 (1-t)^{-\beta}

2169: t^{\beta-1}  t^n \sum_{h=0}^\infty

2170: \underbrace{%

2171: \left( (-1)^{h} t^{h}

2172:  \binom{-\beta}{h}  r^{2h} \right)}_{\binom{-\beta}{h} (-tr^2)^h}

2173: \,dt

2174:  \\

2175: &= \dfrac{\sin( \beta \pi)}{ \pi }   r^n \int_0^1 (1-t)^{-\beta}

2176: t^{\beta-1}  t^n {(1-tr^2)}^{-\beta} \,dt.

2177:  \\

2178: \end{aligned}

2179: \]

2180: \end{proof}

2181: 

2182: \begin{nr}

2183: \label{nr:stimaperron2} For each $\beta\in (0,1)$, $r\in (0,1]$ and

2184: integer $l\geq 2$

2185: \[

2186: \perron_l \left ( |1-r\xi|^{-\alpha} \right) =

2187: \sum_{n=-\infty}^{+\infty} b_{ln} \xi^n.

2188: \]

2189: \end{nr}

2190: \begin{proof}

2191: It follows directly from equation \ref{eq:proprperron}. The

2192: convergence is easy to check.

2193: \end{proof}

2194: 

2195: \begin{nr}\label{nr:stimaperron3}

2196: For each $\beta\in(0,1)$ and $r\in (0,1]$ and integer $l\geq 2$

2197: \[

2198: \perron_l \left ( |1-r\xi|^{-\alpha} \right) = \dfrac{\sin( \beta

2199: \pi)}{ \pi } \int_0^1

2200: \dfrac{(1-t)^{-\beta} t^{\beta-1} }{ %

2201: {(1-tr^2)}^{\beta} } \dfrac{1 - (tr)^{2l}}{|1-(tr)^l\xi|^2} \,dt.

2202: \]

2203: \end{nr}

2204: \begin{proof}

2205: \[\begin{aligned}

2206: \perron_l\left(|1-r\zeta|^{-\alpha} \right) \!\!\!&=

2207: \sum_{n=-\infty}^\infty b_{ln} \zeta^n = \sum_{n=0}^\infty b_{ln}

2208: \zeta^n +

2209: \sum_{n=1}^\infty b_{ln} \zeta^{-n} \\

2210: \!\!\!\!&=\!\!\! \sum_{n=0}^\infty \left( \dfrac{\sin( \beta \pi)}{

2211: \pi } r^{ln} \int_0^1 (1-t)^{-\beta} t^{\beta-1}  t^{ln}

2212: {(1-tr^2)}^{-\beta} \,dt \right) \zeta^n\!+

2213: \\

2214: \!\!\!& +\!\!\! \sum_{n=1}^\infty \left( \dfrac{\sin( \beta \pi)}{

2215: \pi } r^{ln} \int_0^1 (1-t)^{-\beta} t^{\beta-1}  t^{ln}

2216: {(1-tr^2)}^{-\beta} \,dt

2217: \right) \zeta^{-n} \\

2218: \!\!\!&=\!\!\! \dfrac{\sin( \beta \pi)}{ \pi }\!\! \int_0^1\!\!

2219: \dfrac{(1-t)^{-\beta} t^{\beta-1} }{ %

2220: {(1-tr^2)}^{\beta} } \left[ \sum_{n=0}^\infty (tr)^{ln} \zeta^n \!\!

2221: +\!\! \sum_{n=1}^\infty (tr)^{ln} \zeta^{-n} \right]\!\!dt.

2222: \end{aligned}\]

2223: % But

2224: % \[\begin{aligned}

2225: % \sum_{n=0}^\infty

2226: % (tr)^{ln} \zeta^n   &= \dfrac{1}{1-(tr)^l\zeta } \\

2227: % \sum_{n=1}^\infty

2228: % (tr)^{ln} \zeta^{-n} &=

2229: % \dfrac{(tr)^l\zeta^{-1}}{1-(tr)^l\zeta^{-1}}, \\

2230: % \end{aligned}

2231: % \]

2232: The conclusion follows since

2233: \[

2234: \sum_{n=0}^\infty (tr)^{ln} \zeta^n   + \sum_{n=1}^\infty (tr)^{ln}

2235: \zeta^{-n}   = \dfrac{1 - (tr)^{2l}}{|1-(tr)^l\zeta|^2}.

2236: \]

2237: \end{proof}

2238: Thus we proved the following result.

2239: \begin{propo}

2240: For $\beta\in(0,1)$, $r\in (0,1]$ and $\xi\in S^1\subset \CC$ the

2241: potential $U$ can be written as

2242: \[

2243: U = |z|^{-\alpha} \left[ c_l  + l r^{-\beta} \dfrac{\sin( \beta

2244: \pi)}{ \pi } \int_0^1

2245: \dfrac{(1-t)^{-\beta} t^{\beta-1} }{ %

2246: {(1-tr^2)}^{\beta} } \dfrac{1 - (tr)^{2l}}{|1-(tr)^l\xi^l|^2} \,dt

2247: \right],

2248: \]

2249: where $c_l$ is the constant $c_l  = \sum_{j=1}^{l-1} | 1 - \zeta_l^j

2250: |^{-\alpha}$.

2251: \end{propo}

2252: 

2253: \begin{remark}\label{remark:qqq}

2254: An analogue of the integral representation of the potential is well

2255: known, and can be traced back to F.-F. Tisserand's book

2256: \cite{Tisserand}  (chapter XVII)  for the exponent $\alpha=1$; it

2257: had been used by M. Lindow \cite{Lindow24} (section 3) in computing

2258: central configurations for the planar gravitational field  generated

2259: by a regular $n$-gon. More recently D. Bang and B. Elmabsout

2260: extended and generalised Lindow's theorem, proving an equivalent of

2261: \ref{propo:potentialintegral} (Proposition 7 and 8 of

2262: \cite{BanElm}). The proof given here is direct, and allows explicit

2263: estimates that can be used to compute the Hessian for the potential

2264: restricted to the shape sphere. Furthermore, it involves an

2265: interesting connection with the $l$-adic Ruelle--Perron--Frobenius

2266: operator (see P. Gaspard's paper \cite{Gaspard92}), which is worth a

2267: mention.

2268: \end{remark}

2269: 

2270: 

2271: 

2272: 

2273: 

2274: 

2275: 

2276: 

2277: 

2278: 

2279: 

2280: 

2281: 

2282: 

2283: 

2284: 

2285: %%%=========================================================================

2286: 

2287: 

2288: 

2289: 

2290: \begin{thebibliography}{10}

2291: 

2292: \bibitem{ambrosetticoti}

2293: {\sc Ambrosetti, A., and Coti~Zelati, V.}

2294: \newblock Non-collision periodic solutions for a class of symmetric $3$-body

2295:   type problems.

2296: \newblock {\em Topol. Methods Nonlinear Anal. 3}, 2 (1994), 197--207.

2297: 

2298: \bibitem{BanElm}

2299: {\sc Bang, D., and Elmabsout, B.}

2300: \newblock Representations of complex functions, means on the regular {$n$}-gon

2301:   and applications to gravitational potential.

2302: \newblock {\em J. Phys. A 36}, 45 (2003), 11435--11450.

2303: 

2304: \bibitem{zz}

2305: {\sc Barutello, V., Ferrario, D.~L., and Terracini, S.}

2306: \newblock Symmetry groups of the planar $3$-body problem and action--minimizing

2307:   trajectories.

2308: \newblock {\em Arch. Rational Mech. Anal.\/} (2007).

2309: \newblock to appear.

2310: 

2311: \bibitem{Chen03b}

2312: {\sc Chen, K.-C.}

2313: \newblock Binary decompositions for planar {$N$}-body problems and symmetric

2314:   periodic solutions.

2315: \newblock {\em Arch. Ration. Mech. Anal. 170}, 3 (2003), 247--276.

2316: 

2317: \bibitem{Chen03}

2318: {\sc Chen, K.-C.}

2319: \newblock Variational methods on periodic and quasi-periodic solutions for the

2320:   {$N$}-body problem.

2321: \newblock {\em Ergodic Theory Dynam. Systems 23}, 6 (2003), 1691--1715.

2322: 

2323: \bibitem{chenICM}

2324: {\sc Chenciner, A.}

2325: \newblock Action minimizing solutions of the {N}ewtonian {$n$}-body problem:

2326:   from homology to symmetry.

2327: \newblock In {\em Proceedings of the International Congress of Mathematicians,

2328:   Vol. III (Beijing, 2002)\/} (Beijing, 2002), Higher Ed. Press, pp.~279--294.

2329: 

2330: \bibitem{monchen}

2331: {\sc Chenciner, A., and Montgomery, R.}

2332: \newblock A remarkable periodic solution of the three-body problem in the case

2333:   of equal masses.

2334: \newblock {\em Ann. of Math. (2) 152}, 3 (2000), 881--901.

2335: 

2336: \bibitem{chenven}

2337: {\sc Chenciner, A., and Venturelli, A.}

2338: \newblock Minima de l'int\'egrale d'action du probl\`eme newtonien de 4 corps

2339:   de masses \'egales dans ${\bf {r}}\sp 3$: orbites ``hip-hop''.

2340: \newblock {\em Celestial Mech. Dynam. Astronom. 77}, 2 (2000), 139--152 (2001).

2341: 

2342: \bibitem{DelgadoVidal99}

2343: {\sc Delgado, J., and Vidal, C.}

2344: \newblock The tetrahedral {$4$}-body problem.

2345: \newblock {\em J. Dynam. Differential Equations 11}, 4 (1999), 735--780.

2346: 

2347: \bibitem{Devaney80}

2348: {\sc Devaney, R.~L.}

2349: \newblock Triple collision in the planar isosceles three-body problem.

2350: \newblock {\em Invent. Math. 60}, 3 (1980), 249--267.

2351: 

2352: \bibitem{Devaney81}

2353: {\sc Devaney, R.~L.}

2354: \newblock Singularities in classical mechanical systems.

2355: \newblock In {\em Ergodic theory and dynamical systems, I (College Park, Md.,

2356:   1979--80)}, vol.~10 of {\em Progr. Math.} Birkh\"auser Boston, Mass., 1981,

2357:   pp.~211--333.

2358: 

2359: \bibitem{dd}

2360: {\sc Ferrario, D.~L.}

2361: \newblock Symmetry groups and non-planar collisionless action-minimizing

2362:   solutions of the three-body problem in three-dimensional space.

2363: \newblock {\em Arch. Rational Mech. Anal. 179\/} (2006), 389--412.

2364: 

2365: \bibitem{gg}

2366: {\sc Ferrario, D.~L.}

2367: \newblock Transitive decomposition of symmetry groups for the $n$-body problem.

2368: \newblock {\em Adv. in Math. 2\/} (2007), 763--784.

2369: 

2370: 

2371: \bibitem{oo}

2372: {\sc Ferrario, D.~L., and Terracini, S.}

2373: \newblock On the existence of collisionless equivariant minimizers for the

2374:   classical n-body problem.

2375: \newblock {\em Invent. Math. 155}, 2 (2004), 305--362.

2376: 

2377: \bibitem{Gaspard92}

2378: {\sc Gaspard, P.}

2379: \newblock {$r$}-adic one-dimensional maps and the {E}uler summation formula.

2380: \newblock {\em J. Phys. A 25}, 8 (1992), L483--L485.

2381: 

2382: \bibitem{Lindow24}

2383: {\sc Lindow, M.}

2384: \newblock Der kreisfall im problem der $n+1$ k\"orper.

2385: \newblock {\em Astron. Nach. 228}, 5461 (1924), 234--248.

2386: 

2387: \bibitem{McGehee74}

2388: {\sc McGehee, R.}

2389: \newblock Triple collision in the collinear three-body problem.

2390: \newblock {\em Invent. Math. 27\/} (1974), 191--227.

2391: 

2392: \bibitem{Moeckel81}

2393: {\sc Moeckel, R.}

2394: \newblock Orbits of the three-body problem which pass infinitely close to

2395:   triple collision.

2396: \newblock {\em Amer. J. Math. 103}, 6 (1981), 1323--1341.

2397: 

2398: \bibitem{Moeckel83}

2399: {\sc Moeckel, R.}

2400: \newblock Orbits near triple collision in the three-body problem.

2401: \newblock {\em Indiana Univ. Math. J. 32}, 2 (1983), 221--240.

2402: 

2403: \bibitem{SalomoneXia}

2404: {\sc Salomone, M., and Xia, Z.}

2405: \newblock Non-planar minimizers and rotational symmetry in the {$N$}-body

2406:   problem.

2407: \newblock {\em J. Differential Equations 215}, 1 (2005), 1--18.

2408: 

2409: \bibitem{SimoLacomba82}

2410: {\sc Sim{\'o}, C., and Lacomba, E.}

2411: \newblock Analysis of some degenerate quadruple collisions.

2412: \newblock {\em Celestial Mech. 28}, 1-2 (1982), 49--62.

2413: 

2414: \bibitem{sundman}

2415: {\sc Sundman, K.~F.}

2416: \newblock Nouvelles recherches sur le probleme des trois corps.

2417: \newblock {\em Acta Soc. Sci. Fenn. 35}, 9 (1909).

2418: 

2419: \bibitem{Tisserand}

2420: {\sc Tisserand, F.-F.}

2421: \newblock {\em Trait\'e de m\'ecanique c\'eleste}.

2422: \newblock Gauthiers-Villars, Paris, 1889.

2423: \newblock Tome I (Reprinted by Jacques Gabay in 1990).

2424: 

2425: \bibitem{Vidal99}

2426: {\sc Vidal, C.}

2427: \newblock The tetrahedral {$4$}-body problem with rotation.

2428: \newblock {\em Celestial Mech. Dynam. Astronom. 71}, 1 (1998/99), 15--33.

2429: 

2430: \end{thebibliography}

2431: 

2432: \end{document}

2433: 

2434: \endinput

2435: