Chaotic dynamics near triple collision (Q911822)

Perhaps the simplest solutions of the three-body problem are the periodic orbits discovered by Lagrange in 1772. Three masses arranged in an equilateral triangle can rotate uniformly around their center of mass if the rate of rotation is chosen so that the centrifugal force just balances the gravitational attraction. These orbits are part of a family of periodic orbits in which the triangle formed by the masses remains equilateral but varies in size as each mass traverses an ellipse. For each fixed energy and angular momentum there are two such Lagrangian elliptical orbits (up to rotation) corresponding to the two distinct orderings of the masses around the triangle. Besides the equilateral triangles, there are three other configurations which admit such simple motions. For each of the three distinct orderings of the masses along a line there is a unique choice of spacing giving rise to uniformly rotating solutions. These also lie in a family of elliptical orbits called Eulerian since Euler discovered how to choose the spacing. As the angular momentum approaches zero the ellipses degenerate to line segments and the periodic solutions approach solutions exhibiting triple collisions in both forward and backward time. It is known that, for certain choices of the masses, the Lagrangian circular orbits are stable. It turns out, however, that for sufficiently small angular momenta the Lagrangian elliptical orbits are always unstable. This suggests the possibility of solutions homoclinic to one of the Lagrange orbits or heteroclinic between the two of them. The existence of such orbits is one of the main results of this paper. In addition to the Lagrangian orbits we can construct a variety of other periodic orbits exhibiting close approaches to triple collision. Among them are orbits which approximate all five Lagrangian and Eulerian elliptical motions in succession between their close approaches to collision. These new periodic orbits are linked with each other and with the Lagrangian orbits in a complicated network of homoclinic and heteroclinic connections. The whole network is described by use of symbolic dynamics. We obtained results similar to these in a special case of the three-body problem where a symmetry is used to lower the number of degrees of freedom: the isosceles three-body problem reduces to a flow on a three-dimensional manifold. The planar problem unfolds in five dimensions. The stable and unstable manifolds of the Lagrangian elliptical orbits are each three-dimensional and we are looking for transverse intersections. The strategy is to study first the case of zero angular momentum and then to treat the case of small nonzero angular momentum as a perturbation. Using the scaling technique of \textit{R. McGehee} [e.g.: Inventiones math. 27, 191-227 (1974; Zbl 0297.70011)] to tame the triple collision singularity we find that as the angular momentum tends to zero, the Lagrangiean orbits converge to a restpoint cycle. The stable and unstable manifolds of the restpoints patch together to form the stable and unstable sets of the cycle. These are again three-dimensional. We are able to produce intersections of these limiting sets by the following technique. The restpoints corresponding to the Lagrangian orbits are connected to restpoints corresponding to the Eulerian orbits. For certain choices of the masses, the latter exhibit complex eigenvalues. As the stable and unstable manifolds of the Lagrangian restpoints pass by they are wound into spirals around the Eulerian orbits and are thereby forced to intersect one another. Some of the details regarding spiralling of invariant manifolds in high dimensions are treated by \textit{R. Moeckel} [J. Differ. Equations 66, 189-207 (1987; Zbl 0634.57016)]. After constructing these intersections in the zero angular momentum case we study the effect of the perturbation to nonzero angular momenta, obtaining the results described above. Two interesting questions remain open. Firstly, we may wonder whether the homoclinic orbits to Lagrange's orbits persist for fairly large angular momenta. We mentioned above that for certain masses, the circular Lagrange orbits are stable: however, for other masses, including the case of exactly equal masses, these orbits are still hyperbolic. For such masses it is possible that homoclinic oribts exist over the whole range of angular momenta. Secondly, we may ask whether the invariant set described above is part of a larger invariant set which also includes oscillation and capture phenomena. This is so in the simpler isosceles case. In the planar problem the behaviour of orbits ``near infinity'' is more complicated and we have not been stable to incorporate them into the symbolic dynamics.