1: \magnification=\magstep1 \hoffset .65pt

2: \def\R{I\!\! R}

3: \def\Z{\it Z \!\!\! Z}

4: \def\H {I \!\! H}

5: \def\C{C \!\! \! \!  I\, }

6: \def\I{I\!\!I}

7: \def\J{J\!\!\!J}

8: \def\RP{I\!\! R I\!\! P}

9: \def\Ri{Riemannian }

10: 

11: \input epsf.tex

12: 

13: \font\eightrm=cmr8

14: \parindent=8pt

15: \font\eightrm=cmr8

16: \parindent=8pt

17: \hskip11cm {\bf DRAFT. April 24, 2004}

18: \bigskip

19: \centerline{\bf  Fitting Hyperbolic pants

20: to a three-body problem. }

21: \bigskip

22: \bigskip

23: 

24: {\bf  Abstract.}  Consider the three-body problem

25:  with an  attractive $1/r^2$

26: potential.  Modulo symmetries,

27: the dynamics of the    bounded zero-angular momentum solutions  

28: is equivalent to a  geodesic flow on the  thrice-punctured sphere,

29: or ``pair of pants''.  The sphere is the shape sphere.

30:   The punctures are the binary collisions.  The metric generating

31: the geodesics is the Jacobi-Maupertuis metric.  The metric is 

32: complete, has infinite area, and its ends, the neighborhoods

33: of the punctures,  are asymptotically cylindrical.

34: Our main result is  that when the three masses are  equal then the

35: metric   has negative curvature everywhere except at two

36: points (the Lagrange points).  A corollary

37: of this negativity is the uniqueness of the  $1/r^2$ figure eight,

38:  a complete symbolic dynamics for encoding   

39: the collision-free solutions, and the fact that collision solutions are

40: dense within the bound solutions. 

41:  

42: \vskip .3cm

43: 

44:  

45: 

46:  

47: {\bf 1. Introduction and Results.}

48: 

49: We   study  the  planar three-body problem

50:  with an  attractive $1/r^2$

51: potential.  According to the Lagrange-Jacobi

52: identity (eq. (3.7) below)

53:  every bounded  solution must have  

54: zero energy  and constant moment of interia $I$ , and conversely,

55: if an initial condition has zero energy and $\dot I (0) = 0$

56: then that solution is bounded.  

57: Setting  the moment of inertia $I$ equal to a constant  defines a three-sphere in

58: configuration space.  

59: Rotations act on this sphere according to the

60: Hopf flow so that the quotient of the three-sphere

61: by rotations is the two-sphere or {\it shape sphere}. See figure 1a. Points of this shape sphere 

62: represent oriented similarity classes of triangles.  Newton's equations,

63: for solutions with   $H = 0, \dot I = 0$, 

64: push down to  the shape sphere to yield a   

65: a family of second-order ODEs 

66: parameterized by the angular momentum.  These ODEs

67: have  singularities at the three points  representing

68: the three types of binary collisions.  

69: Upon deleting the collision points

70: we arrive  at dynamics on   the {\it pair of

71: pants}, -- the two-sphere minus three points.  

72: When the angular momentum is zero the resulting  dynamical 

73: system  is,   after a time

74: reparameterization, the geodesic flow for a certain \Ri metric on the

75: pair of pants.  This metric is the (reduced)  Jacobi-Maupertuis metric for energy

76: $0$.    

77: \proclaim Proposition 1. Endow

78: the pair of pants with the Jacobi-Maupertuis

79: metric (equations (3.9a,b) below).   

80:  Modulo rotations, translation, 

81: and scaling, the set of bounded 

82: zero-angular momentum solutions

83: for the $1/r^2$ potential three-body problem are in bijective correspondence 

84: with   geodesics

85: for this metric.   

86: The  metric is complete 

87: and its ends (the deleted neighborhoods of the three

88: binary  collisions) are asymptotic to Euclidean cylinders of positive radii. 

89: 

90: {\bf Proof:}  Section 3.

91: 

92: \vskip .3cm

93: 

94: See figure 1b for a depiction of the pair of pants.  

95: 

96: \eject

97: 

98: \vskip 0.1in

99: 

100: \epsfxsize=2.50in

101: 

102: \epsfbox{sphere4.eps}

103: \vskip 0.2in

104: 

105: {\bf Figure 1a.} The shape sphere 

106: \vskip .1in

107: \epsfxsize=3.00in

108: 

109: \epsfbox{pants4.eps}

110: 

111: \vskip 0.8in

112: 

113: {\bf Figure 1b.} The pair of pants

114: 

115:  

116: 

117:  The Jacobi-Maupertuis  metric depends parametrically on the   masses of the three bodies by way of

118: the potential (eq. 3.1).  Our main result is:

119: \proclaim Theorem 1.  If all three masses are

120: equal then  the

121:  Gaussian curvature for the Jacobi-Maupertuis metric on the pair of pants  is   negative 

122: everywhere  except at the two Lagrange points,  where it is zero.

123: 

124: {\bf Proof.}  Section 4. 

125: 

126: One might hope that  negativity   of 

127: the curvature persists for unequal masses.  It does not.  See section 7.

128: \vskip .44cm

129: 

130: 

131: {\bf 2. Motivation and  Dynamical Consequences.}  

132: 

133: {\bf 2.1. Periodic Orbits and their symbol sequences.}

134: 

135: This work began as  an attempt to give an analytic proof that  the

136: Newtonian ($1/r$ potential) figure eight solution of Moore-Chenciner-Montgomery

137: ([Moore], [ChMont])

138: is unique.   

139: I began with the   easier

140: case of the $1/r^2$ eight.

141:  The figure eight  

142: is a periodic solution   which realizes a certain free homotopy class 

143: on the pair of pants.  Figure eights  exists for all $1/r^a$ potentials, $a > 0$

144: ([CGMS] , [FerrTerr]).  For $a \ge 2$, 

145:  not only is the free homotopy class of the  eight realized,   but 

146: almost   every free homotopy class is realized by a solution. Combining these   facts

147: suggested the approach of this paper, and Theorem 1.

148: 

149: Our pair of pants metric from theorem 1 is neither compact, nor of negative curvature

150: everywhere.  But on a complete, noncompact surface

151: of negative curvature if a free homotopy class  has a geodesic representative,

152: then that representative is unique.  And  uniqueness

153: continues to hold if the curvature vanishes on a discrete set of points.

154: (This theorem is fairly well-known, and proved in  

155: in a more general context in section 6.4 below, and in particular

156: eq (6.4.3.).) We have proved

157:  

158: \proclaim Corollary.  For  the $1/r^2$ 

159: equal-mass zero-angular-momentum three-body problem,  if

160: a solution realizes a given  

161: free homotopy class on the pair of pants,

162: then that solution is unique   modulo rotation and scaling.

163: In particular the eight is unique modulo these

164: symmetries. 

165: 

166: In stating the corollary we begged the question of which classes 

167: are realized.  Every

168: class is realized  with the exception of  those classes which wind around a single end.

169: (See [MontN]).  

170: Gordon [1970] calls these  `bad' or unrealizable classes  `untied' 

171: while the complementary `good',  or realizable classes he  called `tied', being 

172: that they are 

173: `tied' to the collision singularities.  On the pants,

174: a bad class can be represented by  drawing a small circle,

175: or ``anklet''  around one

176: pants leg, and traversing it some number of times.

177: As this ``anklet''  is pushed   down towards the end of the leg

178: its length decreases.  As a result, any   minimizing sequence 

179: of curves realizing such a class ``falls off'' of the leg. (See theorem 3 below.)

180:       

181: 

182: We   follow [MontN] in using syzygies to describe the tied and untied classes.     

183: A {\it syzygy} 

184: is a collinear configuration of  the three bodies.

185: Syzygies come in three flavors,

186: 1,2, and 3, depending on which mass is  between the other two.

187: (We exclude collisions.) 

188: The collinear configurations form the equator of the shape sphere.

189: (Figure 1a.) The three collisions lie on the equator so

190: that deleting them divides the  equator into three arcs,

191: again labelled 1,2,3  according to the   mass in the middle. 

192:    A  curve  on the 

193: shape sphere has  an 

194:  associated syzygy sequence:     list the  syzygies in order. 

195: (Assume that the   syzygy times  are discrete.)

196: The syzygy sequence of a motion of the three bodies

197: is obtained by projecting the motion onto the

198: shape sphere and writing out the syzygy sequence of 

199: the curve resulting  on the shape sphere.   

200:  

201:  Periodic  curves

202: give rise to periodic sequences.    For example, the class

203: in which  1 and 2 circle about each other for ever  while 3 remains far

204: away  has  syzygy sequence $\ldots 121212 \ldots$. 

205: (This is a ``bad'' class as we see later.)   We   subject  syzygy sequences to   the {\it  no stutterring rule}:

206: if  $ij$ are   consecutive letters

207: of the sequence, then   $i \ne j$.  The

208: reason for imposing this rule is that a stutter   can be homotoped

209: away. See figure 2. 

210: 

211:  A letter $j$ with a plus superscript,

212: as in $j^+$, denotes that syzygy $j$  occurs by crossing from the upper

213: to the lower hemisphere of the sphere.  A $j^-$

214: means that  

215: syzygy $j$  occurs by crossing from the lower

216: to the upper hemisphere of the sphere.   Pluses and minuses must alternate, since 

217: the  path will alternate between hemispheres.   In topological

218: terms an arc segment with two consecutive

219: pluses, such as   $k^-i^+ j^+$ lies entirely in

220: the upper hemisphere and   can be homotoped to $k^- j^+$.

221: In dynamical terms such an arc can never occur since at $i^+$

222: it would have to be tangent to the collinear subspace (the equator), but 

223: if a solution is tangent to the collinear  subspace at a point then

224: it lies completely within the collinear subspace.   

225: 

226: It follows from the above considerations, and the topology of the

227: pair of pants, that there is a one-to-one onto correspondence between  

228:  free homotopy classes and  periodic signed non-stuttering syzygy sequences.

229: From now on we will drop the signing indications --the $+, -$ superscripts -- for simplicity. 

230: (Given an unsigned sequence there are only two ways to decorate it with

231: signs.) 

232: The untied (unrealizable)  classes are precisely  those with syzygy sequence

233:   $...1212...$, $2323...$ or $...3131....$.

234:  They correspond to a curve  winding around a single end.  We prove in theorem 3 

235: that there is no bounded zero angular momentum solution which realizes

236: them.    

237:   Excluding these classes is equivalent to  insisting that

238: all three letters occur in the sequence.   Thus  the corollary asserts that  {\bf

239: every periodic non-stuttering syzygy sequence in which all three letters occur

240: is realized by a unique (up to symmetry) relative periodic solution.} 

241: 

242: 

243: \vskip 0.2in

244: 

245: \epsfxsize=2.00in

246: 

247: \epsfbox{stutter.eps}

248: \vskip 0.4in

249: 

250: {\bf Figure 2.} Homotoping away a stutter.

251: 

252: \vskip .4cm

253: 

254: 

255:   

256: {\bf 2.2. Symbolic Dynamics;  aperiodic syzygy  sequences.}

257:  

258: We  move  on to infinite aperiodic syzygy sequences. 

259:  In the rest of this subsection, `the problem'  means 

260:  the 

261: $1/r^2$ 

262: equal-mass zero-angular-momentum three-body problem, and

263: `solution' means a   solution to the problem, i.e. this differential equation.

264: Many of the ideas and results here are adaptations of those

265: pioneered in [Morse] and [Hadamard] XX Ref: use Had? . 

266:  

267: 

268: \proclaim Theorem 2.  Every infinite nonstuttering syzygy sequence with

269: the exception of the untied classes $\ldots ijij \ldots$  is realized by

270: a solution. 

271: 

272: {\bf Proof.} Section 6.2.   The method   is the classical one 

273: [Morse] of approximation

274: by periodic solutions.   

275: 

276: \vskip .1cm

277: 

278: 

279: 

280: \proclaim Theorem 3.  If a  

281: syzygy  sequence ends (begins) with 

282:  $ijij\ldots$  then any bounded solution which realizes

283: this sequence must  end (begin)

284: in  the  $ij$ collision.  

285: The untied sequences $\ldots ijij \ldots$ are not

286: realized by   any solution. 

287: 

288: {\bf Proof.} Section 6.3.

289:  \vskip .1cm

290: 

291: It is perhaps worth remarking that if a solution suffers 

292:   collision 

293: then it does so in   finite Newtonian time, but 

294: infinite `Jacobi time''.  

295: 

296: Inspired by theorem 3, we call the sequences appearing there

297: ``collision sequences''.  In more detail:

298: 

299: \proclaim Definition.  A 

300:   bi-infinite nonstuttering  syzygy sequence 

301: $s = \{s_j\}_{j = -\infty} ^{+\infty}$ 

302: is a forward  {\it collision sequence}

303:  if 

304: one of  its forward tails

305: $\{s_j\}_{j> N}$ contains only two letters.

306: Similarly, we have backward collision sequences.  A

307: collision sequence is one which is either forward

308: or backward collision sequence.  In the contrary case,  all three letters occur

309: in every tail, and the sequence is called {\it collision-free}.

310: 

311:  

312: \vskip .1cm 

313: 

314: Does every solution have a syzygy sequence?

315: If so, is this sequence unique?  In  [MontI]  I showed that

316: every bounded noncollinear zero-angular momentum solution to the Newtonian

317: three-body problem  suffers infinitely many

318: syzygies, provided the solution does not tend to triple collision,

319: {\bf and provided}  

320: binary collisions are counted as syzygies.  (See [Fuji] for another proof.)

321:   That proof  works verbatim for any  $1/r^a$

322: potential, $a > 0$, with the exception that

323: we must exclude   binary collisions. (They cannot

324: be regularized.)   Thus

325: every  bounded  {\bf collision-free} solution  has a syzygy sequence.  

326: That sequence must be nonstuttering in our equal mass case.

327: To prove that there is no stuttering,   use the fact that on a surface of negative

328: curvature any compact geodesic arc is the unique length minimizing curve among all

329: homotopic curves which share its endpoints. (See equation 6.3 and its derivation.)

330: Consequently, an application of  the method of reflection as exposed in

331: [ChM]  rids us of solution arcs representing stutters, i.e. solution arcs which

332:  hit  the same equatorial   arc

333: twice in a row.   These considerations allow us

334: to define a syzygy map from collision-free  sequences to 

335: infinite nonstuttering syzygy sequences. 

336: 

337: \proclaim  Theorem 4.   The

338: syzygy map from bounded solutions to   syzygy sequences

339: is a bijection between the set of   collision-free  solutions,   modulo

340: symmetry and time-translation,  and  the set of  bi-infinite nonstuttering collision-free syzygy sequences,

341: modulo shift.  

342: 

343: {\bf Proof.} Section 6.4.

344:  \vskip .1cm

345: 

346: Finally, we would like to know how much of phase space (the unit tangent

347: bundle of the pair of pants) is taken up

348: by the noncollision solutions.  Not much: 

349: 

350: \proclaim Theorem 5. Solutions tending to   binary collision are dense

351: within the space of all bounded solutions.  Thus the 

352: collision-free solutions have empty interior. 

353: 

354: {\bf Proof.} Section 6.5. 

355:  

356: \vskip .3cm 

357: {\bf Summary.} Putting  the theorems together gives a rather 

358:  complete symbolic dynamical

359: picture of the dyanmics of our problem -- the  zero-angular momentum

360: equal-mass $1/r^2$ three body problem restricted to the

361: bound orbits -- those with $I = const.$.  There are no linearly stable periodic orbits, by theorem 1.  

362: We will use the word ``bounded''

363: in the rest of this  paragraph to mean solutions  which tend to   to collision,

364: as these orbits are precisely the geodesics on the pair of pants which tend to infinity.

365:   

366:  Theorem 4 provides a  complete symbolic dynamics  

367: picture for the    bound  orbits : they are precisely the orbits none

368: of whose tails agree with the tied sequences $ \ldots ijij \ldots$.

369: The  closure of this set of orbits is the recurrent set.  The recurrence

370: set also   coincides  with  the closure of the

371: set of   periodic orbits.  There are unbounded orbits

372: on the frontier of this closure.   The situation is  similar to that of  

373: the recurrent set for the Kepler problem:  

374:   the space of periodic orbits is the recurrent set

375: and  contains

376: the unbounded parabolic orbits.    These unbounded  recurrent orbits   are ``just

377: barely unbound''  in that the collision condition $J_1 ^2 - (m_1 + m_2)  \le 0$  occuring

378: in  the appendix A, inequality (12A) is an equality on these orbits.    The complement of the

379: recurrent set   consists 

380: of orbits tending ``strongly'' to a binary collision. These

381: strongly colliding  orbits  form an open set. (Appendix

382: A.)  Finally,  the density result, theorem 6,  is an analogue of 

383:  what one would like to prove for the honest $1/r$ three-body problem:

384: that the set of solutions tending to infinity (via tight binary pairs)

385: is dense, for fixed energy and angular momentum.  M. Hermann

386: calls this density question  ``the oldest problem in dynamical systems

387: [Hermann] XX.  

388: 

389: 

390: {\bf Loose ends.} 

391: There are some collision orbits

392: which we have  left out of symbol sequence considerations.

393: The collinear solutions are not accounted for.  (Collinear

394: solutions should either  have no syzygy sequence or a continuum

395: of `$i$'s as their sequence, depending on one's taste.)  There are exactly six collinear solutions, two for each of the

396: three collision arcs, the two being related by reversing orientation.    

397: There are also collision orbits which end in collision but without the bodies winding around  infinitely often.

398: They `head straight in' to infinity  down one of the pants legs and their 

399: corresponding syzygy sequences will  truncate in the forward direction, for

400: forward time collision.

401: The simplest of these truncated solutions are the isosceles solutions.  

402: Again,  there are six of these by the same

403:   counting as for colliner solutions.  The

404: isosceles solution  $r_{ij} = r_{ik}$   begins and ends  at the $jk$

405: collision, and has exactly one syzygy in between,  the Euler

406: point in which $i$ is at the midpoint of $j$ and $k$. 

407: Its syzygy sequence is the single letter 

408:  `$i$'. Interpolating between collinear and isosceles is a  

409: one-parameter family of  solutions whose  syzygy sequences truncate.    

410: These interpolating solutions are the $\lambda$-curves

411: of  eq. (3.13), the   curves of constant $\chi$, in

412: the $\lambda, \chi$ coordinate system  there.  

413: I do not know if the syzygy sequences of these  

414: solutions are finite, or one-sided infinite.  

415: 

416: 

417:  

418: \vskip .3cm 

419: 

420: 

421: 

422: {\bf Open questions.} {\bf 1.} Can two distinct collision orbits

423: share the same  syzygy sequence?  

424: 

425: {\bf 2.} Are there any solutions  besides isosceles which have a finite syzygy sequence?

426: If so, can any finite syzygy  sequence occur as the syzygy sequence of

427: some collision orbit?  

428: 

429: \vskip 1cm

430: 

431: \vskip 1cm

432: 

433:  

434: 

435: 

436: \vskip 1cm

437:  

438:  {\bf 3. Set-Up and Proof of Prop. 1} 

439: 

440: Write $x = (x_1, x_2, x_3) \in \R^6$ with $x_i \in \R^2$

441: for the positions of the three bodies,

442: and $r_{ij} = \|x_i - x_j\|$

443: for the distances between them.  The

444: potential is $-U$ where  

445: $$U = \Sigma m_i m_j/ r_{ij} ^2 \hskip 1cm (3.1).$$

446: The $m_i$ are the masses. 

447: Set

448: $$

449: \eqalign{ K  &= \Sigma m_i \| \dot x_i\|^2  \cr

450: &  = \langle \dot x , \dot x \rangle, \hskip 1cm (3.2) 

451: }$$

452: for twice the kinetic energy. The last

453: equality of (3.2) defines  the ``mass inner product'' on the

454: three-body configuration space $\R^6$.  The total energy 

455: $$H = K/2 - U \hskip 1cm (3.3)$$

456:  is constant along solutions.   The 

457: equations of motion, 

458:  $  \ddot x_i = -2 \Sigma_{i \ne j} m_j (x_i -x_j)/

459: r_{ij} ^4$, $i = 1,2,3$,  can be written

460: as the single vector equation

461: $$\ddot x = \nabla U \hskip 1cm (3.4)$$

462: where $\nabla U$ is defined using the

463: mass inner-product: 

464: $dU(x)(v) = \langle \nabla U(x), v \rangle$.

465: 

466: By the standard method of freshman physics, we can,

467: without loss of generality restrict our

468: considerations to motions for which

469:  $$\Sigma m_i x_i = 0 \hskip 1cm (3.5)$$

470: throughout.

471: This constraint defines a four-dimensional

472: real vector space which can be identified

473: with the two-dimensional complex space

474: $\C^2$ in such a way that counterclockwise rotation

475: of a triangle $(x_1, x_2, x_3)$

476: by $\theta$ radians  turns into scalar

477: multiplication of the corresponding complex vector by

478: $exp(i \theta)$. 

479:  Set  

480: $$ 

481: \eqalign{I & = \Sigma m_i m_j r_{ij}^2/ \Sigma m_i \cr

482: & = \langle x , x \rangle 

483: } \hskip 1cm (3.6)

484: $$

485: where the last equality is only true when 

486: the center of mass constraint (3.5) is in place.  

487: Using $\dot I = 2 \langle x ,\dot x \rangle$,

488: $\ddot I = 2 \langle \dot x, \dot x \rangle + 

489: 2 \langle x , \ddot  x \rangle$ and

490: $\langle x, \nabla U (x) \rangle = - 2 U(x)$

491: (by $U$'s homogeneity) 

492: we obtain  the Lagrange-Jacobi identity: 

493:  $$\ddot I = 4 H \hskip 1cm (3.7), $$

494: valid along any solution. 

495: Thus 

496: $I(t) = const.$  along the solution

497: if and only if   $H = 0$  and $\dot I (0) = 0$

498: for that solution.    

499: 

500: 

501: We will call a solution ``bounded'' if

502: the $r_{ij}$ are bounded as functions of time,

503: and  

504: do not simultaneously tend to zero, i.e. to triple collison.    

505: Now $I \to \infty$ if and only if one

506:  of   the $r_{ij}$ tend to infinity,

507: and   $I \to 0$ if all $r_{ij} \to 0$.

508: It follows from (3.7)  that

509: every bounded solution must satisfy $H = 0$,

510: $\dot I (0) = 0$, and $I(t) = const.$.

511:   

512:  The scaling symmetry

513: $x(t) \mapsto \lambda ^{-1/2}x (\lambda t)$

514: takes solutions to solutions, preserves zero energy,

515: and takes $I$ to $I/\lambda$.  Using this scaling, 

516: we may, without loss of generality,

517: assume that  $I = 1$ in studying   bounded solutions.

518: The set $I = 1$, $\Sigma m_i x_i = 0$

519: forms a three-sphere in the $\C^2$.  

520: {\bf We have reduced 

521: the study of the bounded solutions to the $1/r^2$ problem

522: to 

523: a second order dynamics on this three-sphere.}

524: It is well-known 

525: ([Arn] or [AbMar]) that a 

526: dynamics on a constant energy surface 

527: $H  = E$

528: level set is equivalent to   

529: geodesic flow for the Jacobi-Maupertuis metric 

530: $( E + U) ds^2$

531: where $ds^2$ is the kinetic energy metric.

532: In our case, $E = 0$, and 

533: we restrict the kinetic energy to the sphere $I = 1$.  Consequently, 

534: the study of bounded solutions is

535: equivalent to the study of geodesics on

536: the three-sphere under the metric 

537: $ds^2_J =  U ds^2$

538: conformal to the standard metric $ds^2$ on

539: that three-sphere.   

540:  

541: To obtain a metric

542: on the shape sphere, we quotient

543: by rotations.   We review the discussion

544: in [MontN], [ChMont], or [MontR] on

545: this metric. See especially the appendix

546: of [MontR] for explicit computations 

547: and derivations.  The group of rigid rotations acts  on the three-body

548: configuration space   

549: according to scalar multiplication on $\C^2$ by unit modulus complex scalars.

550: Restricting ourselves to   

551: the three-sphere $S^3:= \{ I =1 \}   \subset \C^2$ and forming 

552: the quotient by this rotational action yields the

553: famous Hopf fibration 

554: $$S^3 \to S^2 = S^3/ S^1 \hskip 1cm (3.8) .$$

555: The quotient two-sphere is

556: the  shape sphere ([ChM], [MontR] esp. the appendix).

557: Its points represent oriented similarity classes

558: of triangles. 

559: Both the  dynamics and  the  Jacobi metric $U ds^2$

560: on the three-sphere   descend under the projection (3.8)

561: to the shape sphere  

562: once we fix the value of the total angular momentum.   

563: The total angular momentum of a solution is zero if and only if

564: that solution is orthogonal to the rotational orbits,

565: i.e. orthogonal to the fibers of (3.8).  The projection of 

566: such a zero-angular momentum  

567:  solution under (3.8)

568: is  a geodesic for the quotient metric.

569: (See [Hermann], lemma 4.1. His situation is  more general

570: than ours. Our circle  bundle (3.8) is replaced

571: by a general Riemannian submersion.) 

572: We can write the Jacobi-Maupertuis metric on the

573: shape sphere as  

574: $$ds^2_J = U ds_{shape}^2 \hskip 1cm (3.9a)$$

575:  where $ds_{shape}^2$ is

576: the kinetic-energy induced metric on shape sphere,

577: and $U$ is the (negative) potential (3.1) restricted

578: to $I =1$ and then viewed as a fucntion on the shape-sphere

579: (possible because of its rotation invariance.

580: The shape sphere metric is the   round metric on a sphere of radius $1/2$:

581: $$ds^2 _{shape} = ({1 \over 2})^2 [d \phi ^2 + \cos^2 (\phi) d

582: \theta ^2) ] \hskip 1cm (3.9b)$$

583: where

584:  $\phi$ is the colatitude --

585: the angle from the equator, and $\theta$

586: is labels longitudinal circles on the sphere.

587: \vskip .4cm

588: \vskip .4cm

589: {\bf Proof of Proposition 1.} The 

590: discussion of the last two paragraphs shows

591: that,  modulo rotations, translation, 

592: and scaling, the set of bounded 

593: zero-angular momentum solutions

594: for the negative potential (3.1) are in bijective correspondence 

595: with   geodesics

596: for the Jacobi-Maupertuis metric on the shape sphere

597: minus the three binary collision.   

598:  Under this correspondence the

599: geodesic flow for the metric corresponds,

600: after a time reparameterization, to the

601: the flow defined by Newton's equations. 

602: It remains to verify the claims about the completeness

603: and  that the ends asymptote to cylinders.

604:   

605: \vskip .4cm

606: 

607: {\bf Completeness.}

608: 

609: Let $\rho = \rho_{ij}$ be the spherical

610: distance from  the $ij$ collision point $C_{ij}$,

611: as measured in the spherical metric  $ds^2_{shape}$.

612: Then as $\rho \to 0$ we will show that 

613: $$U = {C^2 \over \rho^2} + O (1)  \hskip 1cm

614: (3.10a) $$

615: for some positive constant $C^2$, while 

616: $$ds^2 _{shape} = d \rho^2 + (\rho^2 + O(\rho^4)) d \chi^2

617:  \hskip 1cm (3.10b)$$

618: where  $\chi$ is  the  angular coordinate based at $\Sigma_{ij}$ 

619: so that $(\rho, \chi)$ are geometric polar

620: coordinates.   It follows that the Jacobi metric has

621: the expansion:

622: $$ ds^2_J =   {C^2 \over \rho^2} (d \rho^2 + \rho^2 d \chi^2)) + O(1) \hskip 1cm

623: (3.10c).$$ It follows

624: that if we  approach the collision $\rho = 0$ along any curve then the

625: length of that curve diverges at least as fast as the integral

626: of $C\sqrt{d \rho^2/ \rho^2} = Cd\rho/\rho$, that is,  it  

627: diverges logarithmically as $C|\log(\rho)|$ as $\rho \to 0$.

628: Consequently any  curve tending towards ``infinity''

629: i.e. to one of the binary collisions,   has infinite length,

630: which proves completeness.  

631:  

632: 

633: To establish (3.10a), it suffices to establish

634: $$r_{ij} = {1 \over {\sqrt{\mu_{ij}}}}  \sin(\rho_{ij})

635: \hskip 1cm (3.11)$$

636: where $\mu_{ij} = m_i m_j/(m_i + m_j )$ is the reduced mass.

637: The other two distances $r_{ik}, r_{jk}$ are bounded

638: away from zero as $r_{ij} \to 0$, due to 

639: the constraint $I =1$

640: (see (3.6)).  Then (3.10a) follows from (3.1)

641:   and

642: the Taylor expansion of $\sin(\rho)$.  The constant

643: $C$ in (3.10a) is $m_i m_j / \sqrt{\mu_{ij}}$.

644: 

645:  To establish (3.11) we work in 

646: the full three dimensional shape space which is the space whose points are

647: oriented congruence classes of planar triangles.  The full shape space is 

648: isometric to the   cone over the shape sphere and consequently

649: distances $d$ in the full shape space can be obtained 

650: from spherical distances together with knowldge of

651: the distance $R = \sqrt{I}$  from the cone point. 

652: Write $d_{ij}$ for the distance in the full shape space between

653: an arbitrary point and   the $ij$ binary collision {\it ray}.

654: Then we have  

655: $$r_{ij}  = {1 \over {\sqrt{\mu_{ij}}}} d_{ij} \hskip 1cm (3.12a) $$

656: and

657: $$d_{ij} =  R \sin(\rho_{ij}) \hskip 1cm (3.12b)  .$$

658: Upon setting $R= 1$, (3.11) follows immediately.

659: Equation (3.12a,b) can be 

660:  found in section 4, equations (4.3.15a,b) of [MontN]. However, note that  there

661: is a typo in eq 4.3.15a. The $\mu_{ij}$ in that equation must be replaced by

662: $\sqrt{\mu_{ij}}$.)   

663:  

664: To get (3.10b)  use the fact that the shape sphere is isometric

665: to the sphere of radius $1/2$ and that the metric on such a sphere is given by 

666: $$ds^2_{shape} = d \rho^2 + [(1/2) \sin(2\rho))]^2 d\chi^2 \hskip 1cm (3.13).$$

667: in spherical-polar coordinates. Then use the Taylor expansion of

668: $(1/2) \sin(2 \rho)$. 

669: 

670: \vskip .4cm 

671: 

672: {\bf Asymptotes to Cylinders.}  We 

673: use a more precise version of the expansion (3.10c).

674: Set 

675: $$d\lambda = - \sqrt{U} d \rho \hskip 1cm (3.14).$$

676: Integrating (3.14) defines   a  function 

677: $\lambda = \lambda(\rho, \chi)$ such thta 

678:  $\lambda \to \infty$

679: as  the collision $\rho = 0$ is approached. 

680: From (3.10a) we have  $d\lambda = - C d \rho / \rho  + O(1)$

681: from which it follows that 

682: $$\rho = e ^{ -C \lambda} + o(\rho).$$

683: The  $(\lambda, \chi)$ are coordinates for the end $\rho = 0$,

684: and in these coordinates

685: $$d\bar s^2 = d \lambda^2 + f(\lambda, \chi)^2 d \chi^2

686: \hskip 1cm (3.15)  $$

687: where, from (3.13) and (3.9a) we have  

688: $$f^2  = ({1 \over 2} \sin (\rho) )^2 U.$$

689: Now $({1 \over 2} \sin  ( 2\rho) )^2 =  \sin^2 \rho \cos^2 \rho$

690: so that from (3.11)  

691: $ ({1 \over 2} \sin (2 \rho) )^2 =  \mu_{ij}

692: r_{ij}^2 \cos^2 (\rho )$

693:  and

694: $$f^2 =  \mu_{ij}  \cos^2 (\rho)\{ m_i m_j  + m_i m_k {{r_{ij}^2} \over

695: {r_{ik}^2}} 

696:  +m_j m_k {{r_{ij}^2} \over {r_{jk}^2}} \} \hskip 1cm (3.16)$$

697: where $ijk$ is a permutation of $123$. 

698: As we approach the collision $\lambda = \infty$

699: we have $r_{ij} \to 0$ while $r_{ik}, r_{jk}$ remain bounded

700: since we are constrained to $I =1$.  Thus

701: $$\lim_{\lambda \to \infty} f = \sqrt{\mu_{ij}  m_i m_j}: = K_{ij} > 0 \hskip 1cm (3.17).$$ 

702: Summarizing:  

703: $$d   s^2 _J  = d \lambda^2 + (K_{ij} ^2 + O( e^{-2 C  \lambda})) d \chi^2 

704: \hskip 1cm (3.18)$$ 

705: which says the metric asymptotes to a Euclidean cylinder of

706: radius

707: $K_{ij}$ as we approach the $ij$ end.  

708: 

709:  QED 

710: \vskip .1cm

711: {\bf Remark.}  It follows from equations (7.12, .13)

712: that   the Gaussian curvature near the end is

713: negative.  But for any metric

714: of the form (3.15) this curvature is equal to $ -{1 \over f}{{\partial ^2

715: f}

716: \over {\partial \lambda ^2}}$.  Thus, for 

717: fixed $\chi$,   the 

718: function $f(\lambda, \chi)$ is a  strictly convex

719: of $\lambda$, for all $\lambda$ from some point on, 

720: and from this point on,   

721: $f(\lambda, \chi)$  monotonically decreases to $K_{ij}$.

722: \vskip .4cm

723:  

724: \vskip 1cm 

725: 

726: {\bf 4.  Curvature. Proof of theorem 1.}

727: 

728: We proceed to the proof of our main result, Theorem 1,

729: the negativity of the Gaussian curvature   when the masses are equal.

730: The computation   proceeds 

731: through a series of lemmas.  The first  is standard and we will

732: not provide the proof. 

733: 

734: \proclaim Lemma 4.1.  Let a  surface 

735: be endowed with   conformally related  metrics

736: $ds^2$ and $d \bar s^2  = U ds^2$.

737: Then their curvatures $K, \bar K$ 

738: are related by

739: $$\bar K = U^{-1} (K - {1 \over 2} \Delta \log(U))$$

740: where the Laplacian $\Delta$ is with respect to the

741: $ds^2$ metric.

742: 

743: 

744: The curvature $K$ of the standard shape metric  $ds^2 = ds^2 _{shape}$  of (3.9b) is 

745: $K = 4$. According to lemma 4.1

746: $$\bar K = 4 - {1 \over 2} \Delta (\log(U) \hskip 1cm (4.1)$$

747: is the desired curvature, the curvature

748:  of  the  Jacobi-Maupertuis metric (3.9a) 

749: of proposition 1 and Theorem 1.

750: A routine computation yields

751: $$\Delta (\log(U)) = {{ U \Delta U - \|\nabla U \|^2} \over

752: {U^2}} \hskip 1cm (4.2) .$$

753: Here, and throughout this section,  $U$ is considered  

754:   as a function on the shape sphere,  

755: $\Delta U$ is its Laplacian with respect to the standard

756: shape space metric metric $ds^2$  

757:    and  

758: $\| \nabla U \|^2$  is the squared length of its gradient with

759: respect to the same metric. 

760: 

761: A key to the subsequent computations

762: is to use the squared length coordinates as in [AlbCh] 

763: $$s_k = r_{ij}^2,  ijk \hbox{ a permutation of } 123,  \hskip 1cm (4.3)$$

764: rather than the lengths $r_{ij}$ themselves. 

765: Write 

766: $$U_{2n} = \Sigma 1/r_{ij}^{2n} = \Sigma 1/s_k^n \hskip 1cm (4.4)$$

767: so that $U = U_2$.

768: \vskip .2cm

769: 

770: \proclaim Lemma 2.

771:   $$\Delta U = 8U_4 \hskip 1cm (4.5) .$$

772: 

773: {\bf Proof. } Section 5.2.

774: 

775: \vskip .2cm

776: 

777: \proclaim Lemma 3.

778: $$\| \nabla U \|^2 = 4 S \hskip 1cm (4.6a)$$

779: where

780: $$S = 2U_6 -  U_4 -{3/2} \Sigma^{\prime} 1/s_i ^2 s_j ^2 +  2 \Sigma^{\prime} 1/s_i  s_j ^2

781: - \Sigma^{\prime} 1/s_i  s_j  

782: \hskip 1cm (4.6a)$$

783: and where ``$ \Sigma^{\prime}$ '' means to sum over all indices 

784: $i, j$ with $i \ne j$. (For example  $ \Sigma^{\prime} s_i s_j = 2s_1 s_2 + 2 s_2 s_3 + 2 s_3 s_1$,

785:  twice the second symmetric polynomial in the $s_i$.)

786: 

787: {\bf Proof. } Section 5.3.

788: 

789: \vskip .2cm

790: 

791: {\bf Proof of the Negativity of the curvature.}

792: Combining the equations (4.1), (4.2), (4.5) and (4.6a) 

793: we find 

794: $$ -  \bar K U^3 =  4 U U_4 -  4 U ^2  -  2 S \hskip 1cm (4.7).$$

795: Expand out the first two terms on the right hand side:

796: $$\eqalign{ U U_4 & = \Sigma 1/s_i \Sigma 1/s_j^2 \cr

797: &= \Sigma 1/s_i ^3 + \Sigma^{\prime} 1/s_i s_j ^2 \cr

798: & = U_6 + \Sigma^{\prime} 1/s_i s_j ^2 .

799: } \hskip 1cm (4.8) 

800: $$

801: while

802: $$\eqalign{ U ^2 & = \Sigma 1/s_i \Sigma 1/s_j \cr

803: &= \Sigma 1/s_i ^2 + \Sigma^{\prime} 1/s_i s_j  \cr

804: & = U_4 + \Sigma^{\prime} 1/s_i s_j  .

805: } \hskip 1cm (4.9) 

806: $$ 

807: Plugging  (4.8), (4.9) and equation (4.6b)  

808:  back into equation  (4.7) yields :  

809: $$ 

810: -  \bar K U^3   = -2 U_4  - 2 \Sigma^{\prime} 1/s_i  s_j  + 3  \Sigma^{\prime} 1/s_i ^2  s_j ^2 

811: \hskip 1cm (4.10a).$$

812: Use the fact that $U_4 + \Sigma^{\prime} 1/s_i  s_j = (\Sigma 1/s_i)^2 = U^2$ 

813: to rewrite the right hand side of (4.10), and divide the resulting equation

814:  by two in order to  obtain

815: $$ 

816: - \bar K U^3   = 3  \Sigma^{\prime} 1/s_i ^2  s_j ^2  - 2 U^2  \hskip 1cm (4.10b)

817: $$

818: Consequently, $\bar K \le  0$ if and only if  

819: $$3(\Sigma^{\prime} 1/s_i ^2 s_j ^2) \ge 2 ( \Sigma 1/s_i)^2 \hskip 1cm (4.11)$$

820: To prove (4.11), multiply both sides of it by $s_1 ^2 s_2 ^3 s_3 ^2$

821: thus arriving at $6 (\Sigma s_i ^2) \ge 2 \sigma_2 ^2$ or

822: $$3 (\Sigma s_i ^2) \ge  \sigma_2 ^2 \hskip 1cm (4.12)$$

823: where $\sigma_2 = s_1 s_2 + s_2 s_3 + s_3 s_1$

824: is the second elementary symmetric polynomial in the $s_i$.

825: (The coefficient $6$ arose because for each pair $ij$ there

826: are two terms in the sum $\Sigma^{\prime}$.  See the parenthetical

827: remark in  lemma 3.)

828: To prove (4.12), remember that we are restricting ourselves

829: to the sphere $I = 1$ and that $I = \Sigma s_i /3$.  Thus we 

830: can homogenize the equation by using that $3 =  (\Sigma s_i )^2 /3$ 

831: on the sphere.  So, the desired inequality now reads:

832: $${{(\Sigma s_i )^2} \over 3}(\Sigma s_i ^2) \ge  \sigma_2 ^2.$$

833: The  two inequalities:

834: $$\Sigma s_i ^2 \ge \sigma_2 \hskip 1cm  \hskip 1cm (4.13A)$$

835: and 

836: $${{(\Sigma s_i )^2} \over 3} \ge \sigma_2 \hskip 1cm (4.13B) $$

837: are classical, with 

838: equality  in {\bf either case} if and only

839: if all the $s_i$ are equal.   Here are the proofs. 

840:  Inequality (4.13A) follows simply upon 

841: rearranging  the inequality 

842: $(s_1 - s_2)^2 + (s_2 - s_3)^2 + (s_3 - s_1)^2 \ge 0$.

843:  Inequality (4.13B) is a special case of a general

844: inequality among the elementary symmetric polynomials evaluated at 

845: positive arguments $s_i$.  See for example the 

846: Encyclopaedia [Math],  

847: App. A, Table 8, inequality (4).  Alternatively, expand

848: out $(\Sigma s_i)^2 = \Sigma s_i^2 + 2 \sigma_2$

849: and use  inequality (4.13A).

850: Multiplying the two inequalities

851: yields the desired inequality (4.12),

852: with equality if and only if we are at the Lagrange points

853: $s_1 =s_2 = s_3$ of the shape sphere.

854: Since (4.12) is equivalent to the curvature inequality, 

855: Theorem 1 is proved, modulo the proofs of lemmas 2 and 3

856: which follow in the next section.   QED

857: 

858: \vskip 1cm

859:   

860: 

861: {\bf 5. Proofs of lemmas 2 and 3.}

862: 

863: {\bf 5.1. Notation.}

864:   

865: 

866: To compute $\Delta U$ and

867: $\|\nabla U \|^2$ we must express

868: the squared distances $s_k = r_{ij}^2$ of (4.3) 

869: in terms of the  spherical coordinates $(\phi, \theta)$ of (3.9b). 

870: In [MontI] I prove that upon restriction to the sphere $I = 1$ 

871: $$s_k = 1 - \cos(\phi) \gamma_k (\theta) \hskip 1cm (5.1.1b)$$

872:  where 

873: $$\gamma_1 (\theta) = \cos(\theta), \gamma_2 (\theta) =

874: \cos(\theta + 2 \pi/3), \gamma_3 (\theta) = \cos(\theta + 4

875: \pi/3) \hskip 1cm (5.1.2).$$ The special angles $\theta = 0, 2 \pi/3, 4 \pi/3$

876: mark the locations of the three binary collision

877: on the equator $\phi = 0$ of collinear triangles. 

878: Later on we will use the 

879: fact that the three planar vectors

880: $(\gamma_k, \gamma_k^{\prime})$, $k =1,2,3$ form the vertices of an equilateral

881: triangle inscribed in the unit circle.  

882: \def\dphi{\partial_{\phi}}

883: \def\dtheta{\partial_{\theta}}

884: {\bf Here and throughout we   write  $\gamma_i ^{\prime}$

885: for the derivative $\dtheta \gamma_i$

886: of $\gamma_i$ with respect to $\theta$.}

887: 

888: \vskip .2cm  

889: 

890: {\bf 5.2. Proof of Lemma 2.} 

891: Write $c = \cos(\phi)$,

892: $\dphi$ for the partial derivative with respect

893: to $\phi$ and $\dtheta$ for the partial derivative

894: with respect to $\theta$.  Then 

895: $$\Delta U = {4 \over c} \dphi (c \dphi U) + {4 \over c^2} (\dtheta

896: ^2 U). \hskip 1cm (5.2.1)$$

897: And

898: $$\dphi (1/s_i) = -s\gamma_i /s_i ^2 \hskip 1cm (5.2.2a)$$

899: $$\dtheta(1/s_i) = c\gamma_i ^{\prime}/s_i ^2 \hskip 1cm (5.2.2b). $$

900: Since  

901: $U = \Sigma 1/s_i$

902: we have

903: $$

904: \eqalign{

905: \dphi(c \dphi U)  & = \dphi c \Sigma (-s\gamma_i)/s_i ^2

906: \cr

907: &= \dphi(-cs \Sigma \gamma_i)/s_i ^2  \cr

908:      &= (-c^2 + s^2)\Sigma \gamma_i /s_i ^2 

909:  + 2c s ^2 \Sigma \gamma_i ^2/ s_i^3}.

910:  $$

911: Thus

912: $$

913: {1 \over c}

914: \dphi(c \dphi U) = ((s^2 - c^2)/c)\Sigma \gamma_i /s_i ^2 

915:  + 2 s^2 \Sigma \gamma_i ^2/ s_i^3 .\hskip 1cm

916: \hskip 1cm (5.2.3)$$

917: And

918: $$\eqalign{{1 \over c^2} \dtheta \dtheta U 

919: & = {1 \over c^2} \dtheta \Sigma c \gamma_i ^{\prime}/s_i ^2

920: \cr

921: & = {1 \over c^2} \Sigma c [\gamma_i ^{\prime \prime}/s_i ^2

922: + 2 (c \gamma_i )^2/ s_i ^3

923: \cr

924: & = {1 \over c} \Sigma \gamma_i ^{\prime \prime}/s_i ^2

925: + 2 \Sigma (\gamma_i ^{\prime})^2/s_i ^3

926: }

927: \hskip 1cm (5.2.4)$$

928: Now $\gamma_i^{\prime \prime} = - \gamma_i$

929: so that

930: $${1 \over c^2} \dtheta \dtheta U =

931: -{1 \over c} \Sigma \gamma_i /s_i ^2 

932:  + 2 \Sigma (\gamma_i ^{\prime})^2/s_i ^3 \hskip 1cm (5.2.5)

933: $$

934: 

935: 

936: Adding 4 times (5.2.4) to 4 times (5.2.5) we get

937: $$\Delta U = 

938: 4 ((s^2 - c^2 - 1)/c ) \Sigma \gamma_i /s_i ^2  

939: +  8 s^2 \Sigma \gamma_i ^2/ s_i^3   + 

940: 8 \Sigma (\gamma_i ^{\prime})^2/s_i ^3 \hskip 1cm (5.2.6) $$

941: Now

942: $s^2 -c^2 - 1 = -2c^2$ so that

943: $$\Delta U = -8 \Sigma c \gamma_i /s_i ^2  

944: + 8 \Sigma [s^2 \gamma_i ^2 + (\gamma_i ^{\prime})^2]/s_i ^3

945: $$

946: Recalling that  $\gamma_i ^2 + (\gamma_i ^{\prime})^2 =1$

947: (the vectors $(\gamma_i,  \gamma_i ^{\prime})$

948: define an equilateral triangle inscribed in the unit circle)

949: we see that we can replace  $(\gamma_i^{\prime})^2$ by $1 - \gamma_i ^2$

950: in order to  obtain

951: $$\eqalign{

952: s^2 \gamma_i ^2 + (\gamma_i ^{\prime})^2

953:  & = (s^2 -1) \gamma_i ^2  +1 \cr

954: & = - c^2 \gamma_i ^2 +1   \cr

955: & = -(1-s_i)^2 +1  \cr

956: & = 2s_i -s_i^2 \cr

957: & = s_i (2 -s_i)

958: }

959: \hskip 1cm (5.2.7) $$

960: Then 

961: $[s^2 \gamma_i ^2 + (\gamma_i ^{\prime})^2]/s_i ^3

962: = (2 -s_i)/s_i^2 = (c \gamma_i +1 )/s_i^2$

963: where I used $1 -s_i = c \gamma_i$

964: It follows that

965: $$

966: \Delta U = -8  \Sigma c \gamma_i /s_i ^2  

967: + 8 \Sigma c \gamma_i /s_i ^2 +  8 \Sigma (1/s_i ^2)

968: = 8U_4

969: $$

970: as claimed.

971:  

972: \vskip .4cm

973: {\bf 5.3. Proof of Lemma 3.}

974: 

975:  

976: We have  

977: $$\eqalign{

978: dU  &= \dphi U d \phi + \dtheta U d \theta \cr

979:    & = ( \Sigma (-s \gamma_i /s_i ^2  )d\phi + ( \Sigma (c \gamma_i ^{\prime})/s_i^2) ) d \theta

980: } \hskip 1cm (5.3.1) 

981: $$

982: Now $\| \nabla U \|^2 = \|d U\|^2$.  The length squared of

983: the covector $dU$ is computed relative to the metric `$g^{ij}$ '

984: induced on covectors, which, from (3.9) is 

985: given at the  point $(\phi, \theta) $

986: by  $\|a d\phi + b d \theta \|^2 = 4( a ^2 + {1 \over c^2} b^2)$,

987: with $c = \cos(\phi)$.

988: It follows that

989: $$

990: \eqalign{

991: \|\nabla U \|^2  & =  4(\Sigma s \gamma_i/s_i^2)^2 +  4 (\Sigma \gamma_i ^{\prime}/s_i ^2 )^2 

992: \cr

993: & =  4 \big( \Sigma s^2 \gamma_i ^2 /s_i^4 + \Sigma^{\prime} s \gamma_i s \gamma_j / s_i^2 s_j^2 

994: + \Sigma \gamma_i ^{\prime ^2} /s_i^4 + \Sigma^{\prime} \gamma_i ^{\prime} \gamma_j ^{\prime} / s_i ^2 s_j ^2 

995: \big)

996: \cr

997: & = 4 ( \Sigma (s^2 \gamma_i ^2 + (\gamma_i ^{\prime 2} ) /s_i ^4 +

998: \Sigma^{\prime} (s^2 \gamma_i  \gamma_j + \gamma_i ^{\prime} \gamma_j ^{\prime})/ s_i^2 s_j^2 ).  

999: } \hskip 1cm (5.3.2) 

1000: $$

1001: Simplify the numerator

1002: in the first summand of the last equation

1003: by  using (5.2.7). 

1004:  We will simplify the numerator

1005: of the second summand  by using  an analogous identity for

1006: $s^2 \gamma_i  \gamma_j + \gamma_i ^{\prime} \gamma_j ^{\prime}$, $i \ne j$.

1007: Indeed, since the vectors $(\gamma_i, \gamma_i ^{\prime})$

1008: form the vertices of an equilateral triangle inscribed within

1009: the unit circle,we have that 

1010: $\gamma_i \gamma_j + \gamma_i ^{\prime} \gamma_j ^{\prime} = -1/2$

1011: for $i \ne j$, since  $-1/2 = \cos(2\pi/3)$ is the cosine

1012: of the central angle defined by any two vertices of an equilateral triangle. Thus 

1013: $$\eqalign{

1014: s^2 \gamma_i  \gamma_j + \gamma_i ^{\prime} \gamma_j ^{\prime}

1015: &= \gamma_i  \gamma_j + \gamma_i ^{\prime} \gamma_j ^{\prime} - c^2 \gamma_i \gamma_j

1016: \cr

1017: &=  -1/2 - (1- s_i)(1-s_j) \cr

1018: &= -3/2 + s_i + s_j - s_i s_j \hskip 1cm  

1019: }. \hskip 1cm (5.3.3)

1020: $$

1021: Plugging (5.3.2) and (5.3.3) into (5.3.1) we get

1022: $$

1023: \eqalign{

1024: \|\nabla U \|^2  & =  4 ( 2 \Sigma 1 /s_i ^3 - \Sigma 1/s_i ^2 

1025: -{3 \over 2} \Sigma^{\prime} 1/ s_i ^2 s_j^2 + 2 \Sigma^{\prime} 1/s_i s_j ^2  - \Sigma^{\prime} 1/s_i s_j )  

1026: \cr

1027: &= 4 S}

1028: $$

1029: as claimed. 

1030: QED

1031: 

1032: 

1033: 

1034: \vskip .4cm

1035: 

1036: {\bf 6.  Dynamical Consequences.} 

1037: 

1038: {\bf 6.1}  

1039: 

1040: 

1041: 

1042: Let $P$ denote the pair of pants.  

1043: Using the spherical shape metric (3.9a),

1044: construct three disjoint circles with centers at

1045: the three binary collision points.  Delete the open discs

1046: bounded by these circles to obtain a   compact region $R \subset P$ 

1047: having for its   boundary  the three disjoint circles.   Refer back to  figure 1.

1048: 

1049:  

1050:  \proclaim Lemma 6.1. Any  arc in $P$ whose (finite) syzygy sequence 

1051: contains all  three letters   must cut through  $R$.

1052: 

1053: {\bf Proof.} Let $c$ be such a `123'  arc.

1054: If one of $c$'s syzygies is within $R$, we are done.

1055: Otherwise, all three syzygies lie within the three excised discs.

1056:  But  all three syzygies cannot be   in the same disc,

1057: since each disc contains exactly  two syzygy types. Thus

1058: $c$ must   must travel  from one disc to the other, and in 

1059:  so doing it cross into $R$.  

1060:  

1061: QED.

1062: 

1063: \vskip .3cm

1064: 

1065: 

1066: {\bf 6.2. Proof of Theorem 2.} Let $s$ be 

1067: syzygy sequence containing all three letters.

1068: Approximate $s$ by a sequence $w^N$, $N = 1,2, 3$ 

1069: of periodic sequences as follows.   

1070: Truncate   $s$ to form the finite even length

1071: subword 

1072: $w_N = s_{-N +1} s_{-N +1} \ldots s_N$.   

1073: Turn this subword into a periodic sequence

1074: $ \ldots w_N w_N w_N \ldots$ by repeating it

1075: in blocks.  If the resulting word

1076: has stutters at the join,  shift the ``window''

1077: we used to form $w^N$ so as to form the word 

1078: $w_{N, j} = s_{-N + j +1} s_{-N + j} \ldots s_{N + j}$

1079: along with its  corresponding  periodic word.  

1080: We can always find a $j$ so that the resulting periodic word,

1081: call it $w^N$,  is non-stuttering.

1082: The sequence 

1083: $w_N$ 

1084: contain all three letters $123$ for all $N$

1085: sufficiently large, since $s$ itself contains all three letters.  

1086: Theorem 2 implies that  the $w_N$, for $N$ large,  

1087: are  represented by a unique geodesic

1088: $\gamma_N$. By lemma 6.1 the $\gamma_N$  must cut through $R$.  

1089: Shifting time (and thus shifting the sequence) if necessary, we may assume that

1090: $\gamma_N (0) \in R$.  Since $R$ is compact, 

1091: so is  the unit tangent bundle to $P$   

1092: (relative to the Jacobi-Maupertuis metric) over $R$.   

1093: The pairs

1094: $(\gamma_N (0), \dot \gamma_N (0)$ lie

1095: in this compact space, so we  

1096: we can find a subsequence of them   which 

1097:   converge to some initial condition $(q,v)$.

1098: The geodesic with initial condition $(q,v)$

1099: realizes the   infinite sequence

1100: $s$. This establishes the existence of a  solution 

1101: realizing the   syzygy sequence $s$.

1102: 

1103: QED

1104: 

1105: \vskip .2 cm

1106: 

1107: FIGURE, SECTION AND EQ RELABELLING NEEDED  BELOW   XX

1108: 

1109: 

1110: {\bf 6.3. Collision sequences. Proof of theorem 3.} 

1111: 

1112: Let $s$ be a collision sequence.  

1113: We may  assume, 

1114: without loss of generality,  that it is a forward

1115: collision sequence, and that the two letters in its  

1116: forward tail are  $1$ and $2$.  The first part of the theorem

1117: asserts that  any solution  which realizes $s$

1118: must satisfy $r_{12} \to 0$ as

1119: the Jacobi time  $t \to \infty$.

1120: (In Newtonian time  the  collision  occurs in finite time.

1121: As the     two bodies get closer,

1122: the third body  effects them less and less.

1123: A straightforward  analysis of the  two-body $1/r^2$

1124: problem shows that 

1125: $\lim \inf r_{12} = 0$ if and only if

1126: $\lim r_{12} =0$.  The perturbation

1127: of the third body does not affect this assertion.

1128: Thus in order to prove the solution realizing $s$ suffers

1129: collision it suffices to show that  

1130: $$\lim \inf r_{12} = 0 \hskip 1cm (6.3.1) .$$

1131: along  the solution. 

1132: 

1133: 

1134: Our proof of (6.3.1)  relies on the fact that on a simply connected complete surface

1135: of  non-negative curvature any compact geodesic arc is the

1136:  unique minimizer between

1137: its endpoints.   On  a complete non-simply connected surface such

1138: as the pair of pants   this implies

1139: that if we have a geodesic arc, then there is no shorter curve

1140: which share  endpoints with that arc and which is homotopic to it

1141: through endpoint-fixing homotopies.    

1142: 

1143: We argue by contradiction.  Suppose that some

1144: solution $\gamma \subset P$ realizes the collision sequence

1145: $s$ but satisfies 

1146: $\lim \inf r_{12} = \delta > 0$.

1147:   We will construct a comparison curve  $c$ which

1148: has the same endpoints as a (long) arc of $\gamma$,  is homotopic

1149: to this arc 

1150: through end-point fixing homotopies, 

1151: but which is shorter than $\gamma$.  The  arc will

1152: be one whose syzygy sequence is $1212 \ldots 12$ with

1153: $N$ repeats of $12$, and $N$ large.  

1154:  See figure 6 for the picture of this arc

1155: and the  shorter comparison curve $c$.  The existence

1156: of $c$ contradicts the minimality of $\gamma$ described in  the previous 

1157: paragraph.

1158:  

1159: 

1160: To  construct $c$ we will use 

1161: the cylindrical  coordinates

1162: $(\lambda, \chi)$ 

1163: of  (3.14a,b), (3.15) associated

1164: to the  $12$-collision end.  We have    

1165: $$\lambda (\rho, \chi) = \int_{\rho_0} ^{\rho} \sqrt{U(s,

1166: \chi)} ds$$

1167: where $(\rho, \chi)$ are spherical-polar coordinates

1168: centered at the collision. 

1169:  The $(\lambda, \chi)$ coordinates are valid 

1170: on the entire sphere minus the `$3$' equatorial arc

1171: and the collision points.

1172:  The coordinate $\lambda$ satifies 

1173:   $\lambda  \to \infty$  as collision

1174: is approached.  The Jacobi metric in these coordinates is 

1175: $$ds^2 = d \lambda^2 + f(\lambda, \chi)^2 d \chi^2

1176: \hskip 1cm (6.3.2)  $$

1177: where  

1178: $$f^2 = {1 \over 2}   \cos^2 (\rho) \{ 1   +

1179:  {{s_{12}} \over

1180: {s_{13}}} + {{s_{12}} \over

1181: {s_{23}}} \} \hskip 1cm  $$

1182: The curvature of any metric of the form (6.3.2) 

1183: is  

1184: $ -{1 \over f}{{\partial ^2 f} \over

1185: {\partial \ \lambda ^2}}$.  Since this

1186: curvature is negative (theorem 2) and since 

1187: $\lim_{\lambda \to \infty} f(\lambda, \chi) = 1/\sqrt{2}$

1188: (eq. 3.18) 

1189: we have that   for each fixed

1190: $\chi$ the function    

1191: $f(\chi, \lambda)$ decreases monotonically to its infimum $1 / \sqrt{2}$

1192: as $\lambda \to \infty$.  

1193: It follows that 

1194: $F(\lambda) = \min_{\chi} f(\lambda, \chi)$

1195: also   decreases monotonically to $1/\sqrt{2}$.

1196: 

1197: Any geodesic arc $\gamma$ on the pair of pants which

1198: realizes the syzygy sequence $12$  

1199: cannot cross either isosceles circle

1200: $r_{12} = r_{23}$ or

1201: $r_{12} = r_{13}$.  This follows from the minimality 

1202: property of the arc, discussed above, and   the reflection principle 

1203: (as in [ChM] and the proof of no stuttering in section 2, between

1204: theorems 3 and 4).

1205: Reflections about the isosceles circles are isometries of the Jacobi metric,

1206: so that any segment of $\gamma$  which crosses, then crosses back, can be reflected,

1207: so as to form a new arc with the same endpoints as $\gamma$, and the same homotopy

1208: type, contradicting uniqueness.   Thus, without loss of generality,

1209: we may assume that our long subarc of $\gamma$ lies entirely in

1210: the union of the regions $r_{12} < r_{13}$ and

1211: $r_{12} < r_{23}$.   In particular,

1212: the $(\lambda, \chi)$ coordinates are valid all along our

1213: arc.  

1214: 

1215: To construct the desired comparison curve $c$ we will  

1216: use the fact that $F$ is monotone

1217: decreasing in $\lambda$  and that the number $N$ of $12$ crossings of  a subarc of $\gamma$

1218: can be taken arbitrarily

1219: large.      Since $\inf r_{12} > 0$, by assumption, we have

1220: that $\Lambda = \sup \lambda < \infty$ along our curve.

1221: Let $\epsilon > 0$ small be given. Choose  points $\gamma(s_N), \gamma (t_N)$

1222: $s_N < t_N$, along the arc 

1223: for which $\lambda (t_n), \lambda (s_n)  >  \Lambda - \epsilon$ 

1224: and such that  the arc $\gamma[s_n, t_n]$ in between realizes

1225: the syzygy sequence $12 \ldots 12$ with $N$ copies of $12$.

1226: Our comparison curve $c$ will go ``straight in'' to collision

1227: until some point with $\lambda_* > \Lambda$, to be determined momentarily,

1228: winds around the $12$ collision point in the same sense as $\gamma$ for the same number of

1229: syzygies at this fixed value $\lambda = \lambda_*$, and then return headed ``straight out''

1230: from collision to the point $\gamma(t_n)$.  See figure 6.  ``Straight in'' and ``straight out''

1231: means that $\chi$ is fixed, and only $\lambda$ varies. During

1232: its `winding around'  journey, $\lambda = \lambda_*$ is

1233: fixed and $\chi$ varies, starting at  $\chi =

1234: \chi(\gamma(s_n))$, increasing so that $c$ 

1235: suffers the   syzygy sequence $1212 \ldots 12$ (N

1236: times), and then stopping   at $\chi = \chi(\gamma(t_n)$ in time for the ``straight out''

1237: return segment.  

1238: 

1239: 

1240: \vskip 0.1in

1241: 

1242: \epsfxsize=3in

1243: 

1244: \epsfbox{collisionfig.eps}

1245: \vskip 0.2in

1246: 

1247: {\bf Figure 6.} Orbit surgery to shorten length of

1248: a collision sequence path

1249: 

1250: FFXX LABEL FIG as per next to last par of this sec.

1251: 

1252: \vskip .1in

1253:  

1254: \vskip 0.8in 

1255: 

1256:  

1257: 

1258: By construction, the curve  $c$ shares endpoints with our arc of $\gamma$,

1259: and is homotopic to it.

1260: It remains to  show that $c$ is shorter  than our arc of $\gamma$.

1261: By the monotonicity and the limiting properties of $f$ we can 

1262: choose $\lambda_* > \Lambda$ so that 

1263: $$max_{\chi}f(\lambda_*, \chi) < F(\Lambda):  = min_{\chi} f(\Lambda, \chi).$$

1264: Choose $N$ so large that

1265: $${{2(\lambda_* - \Lambda) + \epsilon + \pi F(\Lambda) } \over N} < \pi (F(\Lambda) -

1266: max_{\chi}f(\lambda_*,

1267: \chi) ).$$  It follows that 

1268: $$ \pi N  max_{\chi} f(\lambda_*, \chi) + \pi max_{\chi} f(\lambda_*, \chi)   + 2(\lambda_*

1269: - \Lambda) + 2 \epsilon <  N\pi (F(\Lambda)$$

1270: Now consider the cartoon in figure XX or the

1271: construction of $c$.   The `in and out' arcs of

1272: $c$ have length less than $2(\lambda_* - \Lambda) + 2 \epsilon$.

1273: The `around arc' between the $N$ syzyygies has length  less than $\pi N  max_{\chi}

1274: f(\lambda_*,

1275: \chi)$.  (Refer again to figure 6.) The additional $\pi max_{\chi} f(\lambda_*, \chi)$

1276: accounts for the fact that $\chi(s_n)$ and $\chi(t_n)$ need

1277: not be equal.      Thus the  left hand

1278: side of the inequality is greater than the length of $c$.

1279: A similar but simpler analysis shows that the right hand side is smaller than

1280: the length of our arc of $\gamma$.   Thus $c$ is shorter than the arc of $\gamma$,

1281: completing  the proof of first assertion of  theorem 3. 

1282: 

1283: The same analysis shows that the infinite sequence $\ldots 1212 \ldots$

1284: of all $12$'s is never realized.  For such a realization must

1285: be a local minimizer, and the above orbit surgery shows we can always

1286: decrease the length of a path by making it closer to collision.

1287: 

1288: 

1289: 

1290: QED

1291: 

1292: \vskip .3cm 

1293: 

1294: {\bf 6.4. Proof of theorem 4.} 

1295: 

1296: To establish the   uniqueness of the realizing solutions, we 

1297: work on the

1298: universal cover $D$ of the pair of pants $P$. Topologically, $D$ is  

1299:  the Poincare disc and the  

1300: fundamental group  $\Gamma = \pi_1 (P)$ (the free group on two letters) 

1301: acts on $D$ as a Fuchsian group.    See figure 3,

1302:   and also the

1303: book Indra's Pearls [Mumf].  

1304: In  figure 3b   we have drawn in a fundamental domain $P_0$

1305: for $P$ and some of its images   under $\Gamma$. The

1306: boundary of $P_0$ consists of four circular arcs which are lines 

1307: lines relative to the Poincare metric on $D$.  These bounding 

1308: arcs are labelled $1_+, 1_-$ and $2_+, 2_-$. To form $P$

1309: out of $P_0$ glue 

1310: arcs $1_+$ and $1_-$ to form syzygy arc $1$, and glue

1311: arcs $2_+$ and $2_-$ to form syzygy arc $2$.

1312: Syzygy arc $3$ is internal

1313: to the fundamental domain.  In the figure we have dropped the

1314: $+, -$ subscripts on the arcs. 

1315: 

1316: \eject

1317: 

1318: \vskip 0.1in

1319: 

1320: \epsfxsize=3.00in

1321: 

1322: \epsfbox{funddomain.eps}

1323: \vskip 0.1in

1324: 

1325: {\bf Figure 3a.} The fundamental domain

1326: \vskip .2in

1327: \epsfxsize=3.00in

1328: 

1329: \epsfbox{tiling.eps}

1330: 

1331: {\bf Figure 3b.} Some of the tiles.

1332: 

1333: \vskip 0.4in

1334: 

1335: Write $$\pi: D \to P$$

1336: for the covering map, so that the fibers of $\pi$

1337: are copies of $\Gamma$.   We will use the hyperbolic, or   constant 

1338: negative curvature metric on $D$ 

1339: in order to understand the fundamental group $\Gamma$ and its action on $D$.

1340: (With   respect to this   metric  $\pi$ is not a local isometry.)  

1341: $\Gamma$ acts on $D$ with respect to hyperbolic

1342: isometries, i.e. M\"obius transformations. 

1343: $\Gamma$ is freely generated by two elements $a$ and $b$,

1344: one of which, say $a$,  interchanges $1_+$ and $1_-$, and 

1345: the other of which, $b$, 

1346: interchanges $2_+$ and $2_-$. These elements act on $D$

1347: as Mobius transformations.  (Viewed

1348: as acting on the Riemann sphere, they  interchange exteriors with interiors of 

1349: their respective circles. )

1350:  In order to form $P$ out of $P_0$ glue arc $1^+$ to $1_-$ 

1351: by $a$  

1352: and glue  $2_+$ to $2_-$ by $b$.  

1353: The projection under $\pi$ of these boundary arcs form the arcs $1,2$ of the equator

1354: of $P$.

1355: The third arc $3$ is internal to $P_0$, and separates it into

1356: two halves, the northern (+) hemisphere, and southern (-) hemisphere.

1357:       

1358:  The images of $\gamma P_0$ of the fundamental domain 

1359: $P_0$ under elements of $\gamma \in \Gamma$ tile all of $D$. 

1360: Each of the four  boundary arcs of such a tile $\gamma P$  

1361:  is the image under $\gamma$ of a unique

1362: boundary arc of $P_0$ and we continue to  label the tile's boundary arcs  

1363: by the corresponding $P_0$ lables  $1_+, 1_-, 2_+$, or  $2_-$. 

1364: Two  tiles intersect, if at all,   along 

1365: a common boundary arc.  This common

1366: arc must be   a   $j_-$ arc of one tile

1367: and a $j_+$ of the other,   $j =1, 2$. 

1368:    

1369: Suppose  now  that 

1370:  two geodesics   realize the same symbol sequence $s$.

1371: Denote by $\gamma,   c$   the lifts of these geodesics to 

1372:   the universal cover $D$.  After  translating these  curves  by   elements of $\Gamma$ 

1373: we may suppose that  both  begin in the  reference fundamental domain

1374: $P_0$.   

1375: I claim that the syzygy sequence $s$ uniquely specificies

1376: a sequence of contiguous tiles  $\ldots P_{-2}P_{-1} P_0 P_1 P_2 \ldots $

1377: through which   $c$ and $\gamma$ must pass.  

1378: To see this fact, we first note that  each sequence of

1379: three contiguous tiles $P_{i-1} P_i P_{i+1}$

1380: represents either two or three letters of a signed syzygy sequence. See figure 4. 

1381: The sequence is obtained by drawing a curve which crosses from $P_{i-1}$

1382: through $P_i$ and into $P_{i+1}$ in the `most direct'' way.

1383: The curve must enter into $P_i$ across one

1384: of its bounding arcs $1_+, 1_-, 2_+,

1385: 2_-$. The choice of $P_{i-1}$ uniquely specifies

1386: which arc.  It must leave   

1387: across another such arc and the choice of  $P_{i+1}$

1388: uniquely specifies this exit arc.  Along the way it must either

1389: cross $3$ or not.   If no   internal syzygy with arc 3 occurs then the  sequence has two letters

1390: and no `$3$', and otherwise the sequence

1391: does contain the  letter $3$ as the middle letter.    

1392:     Consequently,

1393: both $\gamma, c$ pass through an identical

1394: list of tiling domains, as claimed.  

1395: 

1396: Each tiling domain 

1397: $P_j$ has within

1398: it an inverse image $R_j = \pi^{-1} (R) \cap P_j$  of our  compact domain $R$.  

1399: By lemma 6.1 both $c$ and $\gamma$

1400: must have the property that for infinitely many $j$ we

1401: have that both  $\gamma$ and $c$ 

1402: lie in $R_j$. 

1403: 

1404: \vskip 0.2in

1405: 

1406: \epsfxsize=3.00in

1407: 

1408: \epsfbox{crossings.eps}

1409: \vskip 0.4in

1410: 

1411: {\bf Figure 4.} Syzygies and Tile Crossings.

1412: \vskip .2in

1413:  

1414: 

1415:  Now  lift the Jacobi-Maupertuis metric

1416: from $P$ up to $D$, using the

1417: projection $\pi: D \to P$, thus arriving at a complete

1418: non-negatively curved $\Gamma$-invariant metric

1419: on $D$ for which our two curves

1420: $\gamma$ and $c$ are geodesics.

1421: The curvature of this metric is   zero only

1422: at the discrete set of points $\pi^{-1}(L_{\pm})$,

1423: where $L_{\pm}$ are the Lagrange points of $P$.

1424: And the map $\pi$ is  a local isometry for this metric.   

1425: Write 

1426: $$h(t) = dist(\gamma, c(t)) \hskip 1cm (6.4.1)$$

1427: for the distance between the variable point $c(t)$ on

1428: the curve $c$ 

1429: and the entire geodesic $\gamma$.  

1430: See figure 8.  Now

1431: each $R_j$ has finite diameter $\delta$ because

1432: $R$ is compact, and the   two curves pass

1433: through the $R_j$'s infinitely often, indeed every time 

1434: the letters  $123$ occurs contiguously in $s$. It follows that 

1435:   

1436: $$\lim \inf_{t \to +\infty} h(t) \le \delta 

1437: \hbox{ and } 

1438: \lim \inf_{t \to -\infty} h(t) \le \delta. \hskip 1cm (6.4.2)$$ 

1439: 

1440: We will now show that inequality (6.4.2) is impossible unless  

1441:  the two geodesics are in fact the same, in which case $h(t) =0$ everywhere.

1442: We use the formula

1443: $$d^2 h /dt^2 = -sin(A(t)) \int_{d_t} K ds \hskip 1cm (6.4.3)$$

1444: proved in the following paragraph.

1445: In this formula, 

1446: $d_t $ is  the geodesic realizing the distance

1447: $h(t)$.  It  has one endpoint at the point $c(t)$ on $c$

1448: and the other endpoint on $\gamma$ which it intersects

1449: perpindicularly.  See figure 5. The angle $A(t)$ is the

1450: angle of intersection

1451: between the geodesics $c$ and

1452: $d_t$ at  $c(t)$.  See figure 5. 

1453: This angle satisfies   $0 < A(t) < \pi$, so that $\sin(A(t) > 0$  

1454: The negativity of   $K$ 

1455: (except at a discrete  point set) implies that 

1456: $h(t)$ is strictly convex: $d^2 h /dt^2 > 0$.  

1457:    But any strictly convex function defined

1458: on the real line  tends to infinity  in

1459: one direction or the other. This  contradicts   (6.4.2).

1460: Our two geodesics must be the same.  

1461: 

1462: 

1463: \vskip 0.2in

1464: 

1465: \epsfxsize=3.00in

1466: 

1467: \epsfbox{variationarc.eps}

1468: \vskip 0.4in

1469: 

1470: {\bf Figure 5.} Variation of Distance

1471: \vskip .2in

1472:  

1473: {\bf Derivation of (6.4.3). }

1474: 

1475: The first variation of arclength implies

1476:  $$dh/dt = \cos(A(t)) \hskip 1cm (6.4.4) .$$

1477:  Let

1478: $M(t) \subset D$ denote the quadrilateral whose  edges 

1479: consist of the geodesic arcs $d_0, d_t$

1480: together with  the arcs of $\gamma, c$ which connect

1481: $d_0$ to  $d_t$.  According to  the Gauss-Bonnet theorem,

1482: for any such geodesic quadrilateral $Q$ we have

1483: $$ 2 \pi - (\Sigma \hbox{ interior angles }) = - \int \int_Q  K dA$$

1484: In  the case of $M(t)$ the interior angles are $\pi/2, \pi/2, \pi - A(0)$ and $A(t)$.  

1485: See figure 5 again.  

1486: Thus  

1487: $$A(0)  - A(t)= - \int \int_{M(t)} K(t) dA \hskip 1cm (6.4.5) $$

1488: Differentiating (6.4.4) with respect to $t$ yields  

1489: $$dA/dt = \int_{d_t} K ds. \hskip 1cm (6.4.6) $$

1490: Now differentiate (6.4.4) with respect to $t$, using (6.4.6)  to obtain (6.4.3).

1491:  

1492: \vskip 1cm

1493: 

1494: \magnification=\magstep1 \hoffset .65pt

1495: \def\R{I\!\! R}

1496: \def\Z{\it Z \!\!\! Z}

1497: \def\H {I \!\! H}

1498: \def\C{C \!\! \! \!  I\, }

1499: \def\I{I\!\!I}

1500: \def\J{J\!\!\!J}

1501: \def\RP{I\!\! R I\!\! P}

1502: \def\Ri{Riemannian }

1503: 

1504: 

1505: 

1506: {\bf Proof of theorem 5.}  Let $c$ be a bound orbit and

1507: $s = \{ s_j \}_{j = - \infty} ^{\infty}$ its syzygy sequence.

1508: Approximate $s$ by a family $w^n$, $n = 1,2, 3, \ldots$  

1509: of forward collision sequences by replacing the tails $s_j, j > n$ of $s$

1510: by that of the $12$ collision sequence. Collapse two letters if neccessary when

1511: stutters appear at the `join' $j = n$ of   the replacement. 

1512: The backward tails of

1513: the $w^n$ contain

1514: all three letters because $s$ is collision-free, so  we can apply theorem 2 to realize the $w^n$ by

1515: solutions $\gamma_n$ (not necessarily unique).  By lemma 6.1

1516: the $\gamma_n$ all pass through $R$, so, shifting time if necessary,

1517: we have that the tangent vectors   $v_n = (\gamma_n (0), \dot \gamma_n (0))$ 

1518: are unit vectors with position $\gamma_n (0)$ in   $R$.  

1519: We now argue as in the proof of theorem 3.  By compactness

1520: of the  set of unit

1521: tangent vectors over $R$, we can form a convergent

1522: subsequence of the $v_n$, which we  relabel as  $v_n$, so that $v_n \to v$.

1523: Let $\gamma$ be the curve with initial condition $v$.  

1524: The curves $\gamma_n$  converge, over compact

1525: sets,  to  $\gamma$, so that the syzygy sequences $w^n$ must converge to

1526: the syzygy sequence of $\gamma$.  Thus $\gamma$ and $c$ share the same

1527: syzygy sequence.  By   theorem 4,  $\gamma = c$.  Consequently the

1528: unboounded curves $\gamma_n$ converge to our initial bound curve $c$.

1529: 

1530: QED

1531:  

1532: 

1533: \vskip .3cm 

1534: 

1535: 

1536: {\bf 8. Curvature for  unequal masses.} 

1537: 

1538: In this section is to prove that Theorem 1 is

1539: a special case:  for

1540: most mass distributions the curvature takes on both signs.  

1541: Take the masses to be general positive numbers

1542: $m_i$.  Writing  $$p_i = m_j m_k, s_i = r_{jk}^2$$  for

1543: $ijk$ a   permutation of $123$ we have

1544: $$U = \Sigma p_i /s_i, \hskip .3cm  I = {1 \over M} \Sigma p_i s_i,

1545: \hbox{  

1546: where } M = \Sigma m_i.$$

1547: The Jacobi metric is obtained by multiplying the

1548: shape metric $ds^2_m$ on the shape sphere $I =1$  

1549: by $U$ and restricting it   to the sphere $I =1$.

1550: (The subscript `m' indicates dependence on the masses.)

1551: It is conceptually and computationally more straightforward to identify

1552: the shape sphere  as the space of rays in shape space

1553: and to make $U$ a function on the space of rays by

1554: making it  homogeneous of degree  0  by multiplying it by $I$. Thus, setting

1555: $$\tilde U = I U , $$  our Jacobi metric is

1556: $$ds^2_J =  \tilde U ds^2_m.$$

1557: 

1558: 

1559: In order to compute, we will use the  coordinates  $\phi, \theta, R$  of   [Mont2] 

1560: for  shape space. 

1561: (See, in particular, the notation and computations of 

1562: section 7 there.)

1563: Write $I_1$ for the moment of inertial when

1564: all masses are equal to one:

1565: $$I_1 = \Sigma r_{ij}^2/ 3. \hskip 1cm (7.1) $$

1566: The   $\phi, \theta$ are spherical coordinates

1567: for the   $I_1$ shape sphere, while the radial coordinate  

1568: $R = \sqrt{I}$ for the mass-dependent   $I$ of (3.6). 

1569:  With respect to these coordinates we have that  

1570: $$s_k  =   I_1 (1 - \gamma_k (\theta) \cos(\phi)  \hskip 1cm (7.2).$$

1571: as before

1572: Note that  the coordinates $\theta, \phi$ and the functions $s_k$

1573: do not vary as the masses are changes.  

1574:  The metric $ds^2_m$ when expressed in our coordinates is mass-dependent and

1575: is given by 

1576: $$ds^2 _m = \lambda^2 ds^2_1 \hskip 1cm (7.3a) $$

1577: where

1578: $$ds^2_1 ={1 \over 4}(d \phi ^2 + \cos(\phi)^2 d \theta^2) \hskip 1cm (7.3b)$$

1579: (see [MontI], eq (5.6)  and Prop. 2) is the shape sphere metric when all the

1580: masses are $1$,  and where the conformal factor

1581: $\lambda$ is given  by 

1582: $$\lambda = d(m)I_1/I ; \hskip .2cm d(m) = \sqrt{ 3 m_1 m_2 m_3/ M}  \hskip 1cm (7.3c)$$

1583: (See eq (5.7) of [MontI]).

1584: The metric we are to work with is thus 

1585: $$ds^2 _J  = \tilde U\lambda^2 ds^2_1 \hskip 1cm (7.4).$$

1586: Its curvature is given by lemma 4.1 :

1587: $$\bar K = {{ 1} \over { \tilde U  \lambda^2}} \{ 4 - {1 \over 2} \Delta \log (\tilde U \lambda^2) \} \hskip 1cm (7.5).$$

1588: where the

1589: Laplacian $\Delta$ is with respect to the metric

1590: $ds^2_1$.   The conformal factor is 

1591: $$\eqalign{

1592: \tilde U \lambda^2 & = d(m)^2 I U I_1^2/I^2 \cr

1593: & = d(m) I_1 U (d(m) I_1/I) \cr

1594: & = d(m) \hat U \lambda

1595: }$$

1596: where 

1597: $$\hat U = I_1 U.$$

1598: Then $$\Delta \log \tilde U \lambda^2 = \Delta \log \hat U + \Delta \log \lambda.$$

1599: Set 

1600: $$\hat s_i = s_i/I_1 = (1 - \gamma_k (\theta) \cos(\phi)  $$

1601: so that

1602: $$\hat U = \Sigma p_i/ \hat s_i $$

1603: Since $\hat s_i$ is equal to the variable

1604: $s_i$ as given by eq. (5.1.1b), and since the Laplacian

1605: is the Laplacian of that section, we can continue

1606: to   compute  as per section 5.

1607: We have

1608: $$ \hat U \lambda \bar K =  4 - {1 \over 2} \Delta \log (\hat U) - {1 \over 2} \Delta \log (\lambda) .$$

1609: 

1610: 

1611: To compute the last term $- {1 \over 2} \Delta \log (\lambda)$ of (7.5)

1612: we use the fact that  the metrics   $ds^2_m$ are 

1613: $ds^2_1$  through $\lambda^2$, and that 

1614: both have  curvature $4$. From lemma 4.1 it follows that

1615: $$4 = {1 \over \lambda^2} (4 -  {1 \over 2} \Delta \log \lambda^2 \}$$

1616: or

1617: $$4\lambda^2 - 4 = - \Delta \log \lambda.$$

1618: Equation (7.5) can then be rewritten 

1619: $$ -\hat U^3 \lambda \bar K =  -\hat U^2(4 - {1 \over 2} \Delta \log (\hat U))

1620:  + \hat U^2 (2 - 2 \lambda^2)  \hskip 1cm

1621: (7.6).$$

1622: 

1623: {\bf Computation of $\Delta \hat U$.}

1624:  To ease  notation, we drop the hats for this subsection,

1625: so that   $U$ means the function $\hat U$ and

1626: the   $s_k$ means the function 

1627: $\hat s_k$. 

1628: The   structural form of all the formulae and calculations of section 5

1629: remains intact provided we  insert   the weightings

1630: $p_i$ in the correct places.  

1631: 

1632: We proceed to the computation of $\Delta \log (\hat U))$.  Equations (4.2)

1633: and (5.2.1) continue to hold with $\hat U$ in place of $U$, and equations

1634: (5.2.2a,b) hold with $\hat s_i$ in place of $s_i$.  The analogues of

1635: (5.2.3), (5.2.4) are

1636: $$

1637: {1 \over c}

1638: \dphi(c \dphi U)   = ((s^2 - c^2)/c)\Sigma p_i \gamma_i /s_i ^2 

1639:  + 2 s^2 \Sigma p_i\gamma_i ^2/ s_i^3 \hskip 1cm

1640: \hskip 1cm (7.7a)$$

1641: And

1642: $$ {1 \over c^2} \dtheta \dtheta U 

1643:   = {1 \over c} \Sigma p_i \gamma_i {\prime \prime}/s_i ^2

1644: + 2 \Sigma p_i (\gamma_i ^{\prime})^2/s_i ^3

1645: \hskip 1cm (7.7b)$$

1646: Now note that  algebraic steps going from (5.2.5) to (5.2.7) 

1647: apply term by term, so can be carried over verbatim except that

1648: the $i$th term must be multiplied by $p_i$. Thus (reverting to hats)

1649: $$\Delta \hat U = 8 \hat U_4 \hskip 1cm (7.8a)$$

1650: where $$\hat U_4 = \Sigma p_i/\hat s_i ^2 \hskip 1cm (7.8b).$$

1651: 

1652: The first line of  (5.3.2) becomes 

1653: $$\|\nabla U \|^2  =  4(\Sigma s p_i\gamma_i/s_i^2)^2 +  4 (\Sigma p_i\gamma_i ^{\prime}/s_i ^2 )^2$$

1654: The algebra which follows is essentially the same, leading to

1655: $$\| \nabla U \|^2 = 4 S \hskip 1cm (7.9a)$$

1656: where

1657: $$S = 2 \Sigma p_i ^2/ s_i ^3 - \Sigma p_i ^2/s_i ^3 

1658: -{3/2} \Sigma^{\prime} p_i p_j /s_i ^2 s_j ^2 +  2 \Sigma^{\prime} p_i p_j/s_i  s_j ^2

1659: - \Sigma^{\prime} p_i p_j /s_i  s_j \hskip 1cm (7.9b).$$

1660: Combining (7.8) and (7.9)  according to (4.2) (see also the steps (4.7)-(4.9)) yields the formula: 

1661: $$- U^2 (4 - {1 \over 2} \Delta \log U)  = 3 \Sigma^{\prime} p_i p_j /s_i ^2 s_j ^2 - 2(\Sigma p_i /s_i)^2 

1662: \hskip 1cm (7.10).$$

1663: 

1664: \vskip 1cm 

1665: From earlier, we have   

1666: $$ U^2 (2 - 2 \lambda^2) = 

1667:  2 (\Sigma p_i /s_i)^2 - 2(\Sigma p_i /s_i)^2 d(m)^2 ({1 \over 3} \Sigma

1668: s_i)^2/(\Sigma p_i s_i /M)^2, \hskip 1cm (7.11).$$

1669: (We continue to use $s_i$ in place of $\hat s_i$.) Upon adding (7.10) and (7.11) there

1670: is a   cancellation yielding:  

1671: $$ - U^3 \lambda^2 \bar K =  3 \Sigma^{\prime} p_i p_j /s_i ^2 s_j ^2   - 2(\Sigma p_i /s_i)^2 d(m)^2 ({1 \over 3} \Sigma

1672: s_i)^2/(\Sigma p_i s_i /M)^2.$$

1673: A computation shows that

1674: $$d(m)^2 M^2 = {3 \over 2}\Sigma^{\prime} p_i p_j$$

1675: Recall that the $s_i$ (the previous $\hat s_i$'s) satisfy

1676: ${1 \over 3} \Sigma s_i =1$.  We finally obtain:

1677: $$ - U^3 \lambda^2 \bar K =  3 \{ \Sigma^{\prime} p_i p_j /s_i ^2 s_j ^2  

1678:  - \Sigma^{\prime} p_i p_j {{ (\Sigma p_i /s_i)^2 } \over {(\Sigma p_i s_i )^2}} \} 

1679: \hskip 1cm (7.12).$$

1680: Consequently 

1681: $$\kappa = \sqrt{\Sigma^{\prime} {{p_i p_j} \over{ s_i ^2 s_j ^2}}}  

1682: {{ (\Sigma p_i s_i)} \over {(\Sigma p_i /s_i )}} - \sqrt{  \Sigma^{\prime} p_i p_j }

1683: \hskip 1cm (7.13)$$

1684: governs the sign of the curvature,

1685: with the curvature   $\bar K$   negative if $\kappa$

1686: is positive, positive if $\kappa$ is negative, and zero

1687: if $\kappa$ is zero.  We note that 

1688: $\kappa$, and hence the curvature is zero 

1689: at the Lagrange point $s_1 = s_2 =s_3 =1$,

1690: and that this is true for all choices of the masses $p_i$.

1691: 

1692: \proclaim Theorem 6.  

1693: For a Zariski-dense set of  mass distributions,

1694: the sign of the curvature changes in a neighborhood of the

1695: Lagrange point.

1696: 

1697: Proof. It suffices

1698: to show that for a Zariski-dense set of  mass distributions

1699: the differential  $d \kappa  \ne 0$ at the Lagrange point.

1700: A differential form $\Sigma a_i ds_i$

1701: represents zero on the shape sphere if and only if it is proportional to

1702: $\Sigma ds_i$, the latter being the differential of the constraint

1703: $\Sigma s_i = 3$ satisfied by the $s_i$ (which are the old $\hat s_i$'s). 

1704: A computation shows

1705: that  at  the Lagrange point $s_i =1$ we have  

1706: $$\beta d \kappa = p_1 (p_2 ^2 + p_3 ^2) ds_1 + p_2 (p_1 ^2 + p_3 ^2) ds_2 + p_3 (p_1 ^2 + p_2 ^2) ds_3

1707: \hbox{ mod }  \Sigma ds_i$$

1708: where $\beta$ is a nonzero constant.

1709: We thus want to know whether or not

1710: the equality

1711: $$(p_1 (p_2 ^2 + p_3 ^2), p_2 (p_1 ^2 + p_3 ^2) , p_3 (p_1 ^2 + p_2 ^2)) 

1712: = (\lambda, \lambda, \lambda) \hskip 1cm (**) $$

1713: can be satisfied for some $\lambda$.   

1714: The right hand side of equation (**),

1715: being homogeneous of degree $3$, 

1716:   defines a polynomial map $\RP^2 \to \RP^2$

1717: and we want to know if it is equal to the

1718: constant map   $[1,1,1]$.  Because the map is  polynomial,

1719: if we can exhibit a single point where the  inequality fails

1720: then it must fail on a Zariski-dense set.    Plugging in

1721: $p_1 = p_2 = 1, p_3 = a$ yields

1722: $(p_1 (p_2 ^2 + p_3 ^2), p_2 (p_1 ^2 + p_3 ^2) , p_3 (p_1 ^2 + p_2 ^2))

1723: = (1 + a^2, 1 + a^2, 2a)$

1724: which is not proportional to $(1,1,1)$ unless $a = 1$.

1725: 

1726: QED 

1727: 

1728: \vskip .3cm 

1729: {\bf 

1730: Appendix A}

1731: 

1732: We prove

1733: \proclaim Theorem A.  The set of initial conditions

1734: within $H = 0,  I = 1 , c = 0$

1735: whose solutions tend to a binary collision

1736: of type $ij$ has nonempty interior. This fact

1737: holds for all positive mass distributions. 

1738: 

1739: {\bf Proof of theorem A.}We use Newtonian time and Jacobi

1740: coordinates For notational simplicity,  take $ij = 12$.

1741: The   Jacobi coordinates are

1742: $\zeta_1 = x_1 - x_2$, $\zeta_2 = x_3 - (m_1 x_1 +m_2 x_2)/(m_1 + m_2)$.

1743: The distance to binary collision is

1744: $$r = | \zeta_1| \hskip 1cm (1A).$$

1745: We will exhibit a nonempty open set of

1746: initial conditions at time $t =0$

1747:  for which $r(t) = 0$

1748: for some time $t < O(r(0))$. 

1749: 

1750:  

1751: The Hamiltonian is 

1752: $$H = {1 \over 2} ( \mu_1 | \dot \zeta_1|^2 + 

1753: \mu_2 | \dot \zeta_2 |^2)  -{{m_1 m_2} \over {r^2}} 

1754: - W(\zeta_1, \zeta_2) \hskip 1cm (2A)$$

1755: where $\mu_1 = m_1 m_2 /(m_1 + m_2)$,

1756: $\mu_2 = m_3(m_1 + m_2)/M$, 

1757: $W = m_1 m_3 /s_2 + m_2 m_3/s_1 $

1758: and the squared distances $s_2$, $s_3$  can be expressed 

1759: $|\zeta_2 + a_i

1760: \zeta_1 |^2$ in Jacobi coordinates,  with

1761:  mass-dependent

1762: nonzero constants

1763: $a_i$.   

1764: The interaction term  $W$  satisfies the estimates  

1765: $$|W| \le C_1 + C_2 \epsilon ,\hskip .3cm \hbox{ for } r < \epsilon \hskip 1cm 

1766: (3A), $$ 

1767: $$|{{\partial W} \over {\partial

1768: \zeta_1}}|

1769: \le C_1 + C_2 \epsilon 

1770: \hskip .1cm ,  \hskip .1cm  |{{\partial W} \over {\partial

1771: \zeta_2}}| \le C_1 + C_2 \epsilon , \hskip .3cm 

1772: \hbox{ for  } r  \le \epsilon \hskip 1cm (4A)$$

1773: ($\epsilon$ sufficiently small), where $C_1, C_2$

1774: are constants depending only on the masses.   

1775: 

1776: The equations of motion are 

1777: $$\ddot \zeta_1 = - (m_1 + m_2){{\zeta_1} \over {r^4}}  + {1

1778: \over

1779: \mu_1}{{\partial W} \over {\partial \zeta_1}} \hskip 1cm (5A)$$

1780: and

1781: $$\ddot \zeta_2 =   {1 \over

1782: \mu_2}{{\partial W} \over {\partial \zeta_2}} \hskip 1cm (6A)$$

1783: Write 

1784: $$J_1 = \zeta_1 \wedge \dot \zeta_1 \hskip 1cm (7A)$$

1785: for the angular momentum (up to a factor of $\mu_1$)

1786: of the 12 system.  We compute 

1787: $$\eqalign{\dot J_1 & = \zeta_1 \wedge \ddot \zeta_1 \cr

1788: &= \zeta_1 \wedge{1 \over

1789: \mu_1}{{\partial W} \over {\partial \zeta_1}}}

1790: $$

1791: so that

1792: $$|\dot J_1| \le Cr \hskip 1cm (8A).$$

1793: 

1794: Because $| \dot \zeta_1|^2 = \dot r^2 + J_1^2/ r^2$ we

1795: have that 

1796: $$r^2 H = {1 \over 2} (\mu_1 r^2 \dot r^2 + \mu_1 J_1^2 )

1797: - m_1 m_2  + r^2 ( {1 \over 2} \mu_2 | \dot \zeta_2 |^2

1798: - W ) \hskip 1cm (9A) $$

1799: 

1800: 

1801: Now let $\zeta(t) = (\zeta_1 (t), \zeta_2 (t))$

1802: be a solution satisfying the initial conditions

1803: $r(0) < \epsilon,  H = 0, I =1, J = 0$.

1804: From (6A) and  (4A)

1805: we have that 

1806: $$| \dot \zeta_2 (t)|^2 \le | \dot \zeta_2 (0)|^2 + Ct \hskip 1cm (10A),$$

1807: for $t = O(1)$, provided $r(0) < \epsilon$.

1808: Here $C$ depends only on the masses and $\epsilon$.  

1809:  Letting $r \to 0$, we see from (9A) that if our solution

1810: is to have a collision then

1811: we must have

1812: $$\lim r^2 \dot r^2 + \lim J_1 ^2 - 2 (m_1 + m_2) = 0,  \hskip 1cm (11A)$$

1813:  where we have used $m_1 m_2 /\mu_1 = m_1 + m_2$.

1814: But $r^2 \dot r^2 \ge 0$, so we must have

1815: $$ 2 (m_1 + m_2) -\lim J_1 ^2  \ge 0 \hskip 1cm (12A)$$

1816:  

1817: We  argue   in the reverse. Suppose that   $ 2 (m_1 + m_2) - J_1 (0) ^2 $ is

1818: sufficiently positive at the initial time $t=0$, and that $\dot r (0) < 0$ 

1819: then (11A) forces $r^2 \dot r^2$ to be positive over

1820: a finite  time interval. We will show that, upon  integration, 

1821: this  will  force

1822: $r(t) = 0$ in some finite time  $t = O(\sqrt{r(0)})$.

1823: Note from the bounds (10A)   and the fact that $H= 0$

1824: we have

1825: $$|\mu_1  r^2 \dot r^2 + \mu_1 J_1 (0)^2 )

1826: - 2 m_1 m_2  |  \le  K r(0) \hskip 1cm (13A) $$

1827: for $0 \le t \le 1$ and for as long as $\dot r(t) < 0$. 

1828: Here the constant $K$ depends only on the masses

1829: and $\dot \zeta_2 (0)$.   Dividing by $\mu_1$ and using

1830: $m_1 m_2 /\mu_1 = m_1 + m_2$ we arrive at 

1831: $$|  r^2 \dot r^2 + J_1 (0)^2 )

1832: - 2(m_1  + m_1 ) |  \le  K^* r(0)  \hskip 1cm (14A) $$ 

1833: where $K^* = K/\mu_1$. 

1834: 

1835: We now impose the 

1836: open condition

1837: $$2(m_1 +  m_2)  - J_1 (0)^2 - K^* r(0) > \delta^2 \hskip 1cm 

1838: (15A)

1839: $$

1840: on our initial conditions.  This will be the open condition

1841: of theorem A.  

1842: The positive constant    $\delta$ will

1843: be constrained further

1844: below.     

1845: It follows from (15A) and (14A) that 

1846: $$\delta^2  <  2(m_1 +  m_2) - J_1 (0)^2) - K^* r(0)  \le  r^2 \dot

1847: r^2  \hskip 1cm 

1848: (16A)$$

1849: (16A) together with $\dot r(0) < 0$  forces   $\dot r < 0$ throughout the time interval in

1850: question.   Thus $ - r \dot r > 0 $, and so we can take

1851: square roots of  inequality  (16A) to obtain

1852: $$\delta \le  - r \dot r  \hskip 1cm 

1853: (17A) .$$  Taking negatives and integrating we find

1854: that

1855: $ -  \delta t \ge {1 \over 2} r(t)^2 - {1 \over 2}

1856: r(0)^2$ or 

1857: $$r(0)^2 - 2  \delta t  \ge r(t)^2 \hskip 1cm 

1858: (18A).$$

1859: This forces $r(t) = 0$ for some time

1860: $t$ with $t \le r(0)^2 / 2 \delta $. 

1861: In order that the collision time $t$ is $o(1)$ 

1862: it is sufficient to take $\delta = O(r(0))$.   

1863: 

1864: We have proved that a 12 collision occurs

1865: within a time $t = r(0)^2/ 2 \delta $ 

1866: for all initial conditions satisfying (15A), 

1867: $\dot r(0) < 0$,  and $r(0) < \epsilon$, 

1868: where $\epsilon$ is small enough so that the inequalities  (3A, 4A) are in force.

1869: This set of initial conditions is clearly open.

1870: It remains to show that this set is nonempty.

1871: Consider  the {\it collinear}  solution

1872: having  $H =

1873: 0 = J$ and $I =1$.  (There are precisely two

1874: such solutions, up to   time translation

1875: and rotation, one for each arc of the equator which

1876: ends in the 12 collision.)   These solutions  satisfy $J_1 = 0 $.

1877:    In this case

1878: (15A) reads

1879: $m_1 + m_2  > K r(0) + \delta$.

1880: and so will hold for $r (0)$ small provided

1881: only that $\delta < m_1 + m_2$.

1882: Since the solution tend to collision

1883: it follows that (15A) is eventually in force

1884: along the collinear solution, and hence that

1885: our set of of initial conditions is nonempty.  

1886: 

1887: QED

1888: 

1889: 

1890: \vskip 1cm

1891: 

1892: 

1893: {\bf Acknowledgements.}  

1894: I   dedicate this paper to the memory of my father.

1895: I    acknowledge useful correspondences

1896: with  Toshiaki Fujiwara, Alain Chenciner, Alain Albouy,

1897: and   conversations with  Anatole Katok, Rafe Mazzeo, and with 

1898:   Jeff Xia for pointing out that theorem 1

1899: combined with an earlier

1900: version of theorems 2, 3 and 4 ought to imply theorem 5.  

1901: 

1902: \medskip

1903: 

1904: {\bf References}

1905: 

1906: [AbMar] R. Abraham and J. Marsden, {\bf Foundations of Mechanics}, 

1907: Benjamin-Cummings, [1978]. 

1908: 

1909: [AlbCh]  A. Albouy and A. Chenciner, {\it Le probl\'eme des $n$ corps et les

1910: distances mutuelles}, Invent Math. 131 (1998), no. 1, 151--184.

1911:  

1912: [Arn]  V.I. Arnol'd, {\bf Mathematical Methods of Classical Mechanics},

1913: Springer-Verlag, [1989]. 

1914: 

1915: [Ban] T. Banachiewitz,

1916: {\it Sur un cas particulier du probleme

1917: des trois corps}, CRAS, Paris, 142, (1906), pp 510-512.

1918: 

1919: [CGMS] A. Chenciner, J. Gerver, R. Montgomery R. and C. Sim\'o{\it Simple

1920: choreographies of $N$ bodies: a preliminary study}

1921:  in {\bf Geometry, Mechanics

1922: and Dynamics}, 287--308, Springer,New York, 2002.

1923: 

1924: [ChMont] Chenciner A. and Montgomery R. {\it A remarkable periodic solution of the three-body

1925: problem in the case of equal masses, Annals of Math., 152, pp. 881-901 (2000)}

1926:  

1927: 

1928: [FerrTerr] Davide Ferrario and Susanna Terracini  

1929: {\it  On the Existence of Collisionless Equivariant Minimizers for the Classical

1930: n-body Problem.} Math ArXivs,         math-ph 0302022 [2003]. 

1931: 

1932: 

1933: [Gordon] W. B. Gordon, {\it A minimizing property of keplerian orbits}, American

1934: Journal of Mathematics, vol. 99,  $n^05$,   961-971, (1970).

1935: 

1936: [Fuji]  T. Fujiwara, H. Fukuda,  A. Kameyama,  H. Ozaki, M. Yamada,

1937: {\it  Synchronised Similar Triangles for Three-Body Orbit

1938: with Zero Angular Momentum}, arxiv.org/abs/math-ph/0404056.

1939: 

1940: [Hermann] R. Hermann, {\it On the differential geometry of foliations}, Ann. of

1941: Math. (2), (1959), 445-457 

1942: 

1943: [Math] Mathematical Society of Japan, {\bf Encyclopedic Dictionary of

1944: Mathematics},  by the Mathematical Society of Japan, ed. by S Iyanga and Y Kawada,

1945: translated by K. O. May, The MIT Press, Cambridge, Massachussets, and

1946: London, England, [1977] 

1947: 

1948:  

1949: [MontI]Richard Montgomery, {\it Infinitely Many Syzygies}, 

1950: Archives for  Rational Mechanics and Analysis, 

1951: v. 164 (2002), no. 4, 311--340, 2002. 

1952: 

1953: 

1954: [MontN] Richard Montgomery,  {\it The N-body problem, the braid group, and

1955: action-minimizing periodic orbits}, Nonlinearity, vol. 11, no. 2, 363-376, 1998.

1956: 

1957: [MontR] Richard Montgomery, 

1958: {\it  Geometric Phase of the Three-Body Problem}, Nonlinearity,

1959: vol. 9, no. 5, 1341-1360, 1996.

1960: 

1961:   

1962: [Moore] Cris Moore,  {\it Braids in Classical Gravity},  Physical Review

1963: Letters 70, pp. 3675--3679, (1993).

1964: 

1965: [Morse] H. M. Morse, {\it A one-to-one representation of geodesics on a surface of

1966: negative curvature}, Am. J. Math, {\bf 43}, no. 1, 33-51, 1921.

1967: 

1968: [Mumf] David Mumford, Caroline Series,David Wright, David  

1969: {\bf Indra's pearls.} 

1970: The vision of Felix Klein. 

1971: Cambridge University Press, New York, 2002. 

1972: 

1973: 

1974: [Poin] Poincar\'e, ~H. [1896], {\it Sur les solutions p\'eriodiques et le

1975: principe de moindre action}, .

1976:  {\it C.R.A.S. Paris} {\bf 123}, 915--918.

1977: 

1978: 

1979: 

1980: 

1981: 

1982: \medskip

1983:  

1984: \smallskip

1985: 

1986: %\insert bib

1987: 

1988: 

1989: \bigskip

1990: 

1991: \vfill\eject

1992: \end

1993: 

1994: 

1995: \end

1996: 

1997: 

1998: 

1999: