« PreviousContinue »
From these equations it follows that di 12 = 1/2 and v2 = -V1. What this means is that all we can do is choose in the center-of-mass frame a new direction vector for one of the two colliding particles. Everything else is then determined. The other particle goes in the opposite direction, and the total kinetic energy in the center-ofmass frame is divided evenly between the two particles. Thus, the only element of randomness is in how the new direction vector is chosen. If all directions are assumed to be equiprobable, then it can be shown that no matter what the initial distribution of velocity is, the system tends under iteration to a limiting distribution that is the standard normal distribution in three-dimensional Euclidean space R3. We have thus rederived the Maxwell-Boltzmann distribution of velocities. Here again we can go further and consider more complicated redistribution laws.
Suppose one allows ternary collisions instead of binary collisions. Then there are more degrees of freedom, and the problem again becomes interesting mathematically. The results of our analysis show that the situation is much like the redistribution of energy in that the limiting distribution of velocity depends on the law of redistribution of velocity.
Problem 2. Geometry, Invariant Measures, and Dynamical Systems
The intimate relationship among geometry, measures, and dynamical systems that was elucidated in the last century continues to deepen and hold our attention today. Poincaré made several monumental contributions to this development in his treatise Les Méthodes Nouvelles de la Mécanique Céleste. One major issue he considered concerned the stability of motion in a gravitational field such as that of our solar system. Would small perturbations from any given set of initial orbits lead to a collision of the planets? A tremendous amount of work had been done on this dynamical system, but the governing system of differential equations remained unsolved. Faced with this situation, Poincaré made a wonderful flanking maneuver by introducing "qualitative" methods that involved measures.
For the setting consider the motion of N bodies and the corresponding phase space S, whose 6N coordinates code the position and momentum of each of the N bodies. The phase space is a subset of Euclidean 6N -space and each point of S corresponds to a state of the system. Consider T, the time-one map of S. That is, if s is the initial state of the system, then T(s) is the state of the system one time unit later. Now, various notions of stability can be given in terms of the properties of T. One of these is recurrence, or, as Poincaré said, "stabilité à la Poisson.” A state s is said to be recurrent provided that if the system is ever in s, then it will return arbitrarily close to s infinitely often. Formally, s is recurrent provided that for every open region U about s there are infinitely many positive integers n such that T"(s) is in U. Poisson had earlier attempted to show this kind of stability for the restricted three-body problem. Poincaré used the fundamental tenet of measure theory, countable additivity, to prove that the set of all points s in the phase space for which recurrence does not occur is of measure zero.
Recurrence Theorem: Let B = (s E Sis is not recurrent). Then B has measure zero.
Poincaré's proof of this theorem (see “The Essence of Poincaré's Proof of the Re
currence Theorem”) is a shining jewel that made clear to the mathematical world the importance of countable additivity in the development of measure.
But what measure did Poincaré have in mind here? After all, there is an entire grab bag of measures on the subsets of S. In the case of the N-body problem, since the system is a Hamiltonian system, the geometry of the phase space clearly indicates the correct measure. Let us see why. Liouville had proved the seminal result that if the map T that describes the time evolution of the system is a Hamiltonian, then T is volume-preserving in the phase space. That is, if U is an open set or region, then v(U) = v (T(U)), where v(E) is the volume of E. Poincaré carried out his analysis on a “surface of constant energy.” Since the N-body problem is a conservative system, the function T leaves the total energy invariant and therefore maps each such surface into itself. Moreover, since T is a Hamiltonian, it is volume-preserving on this surface. Consequently, the geometric structure of the surface determines the appropriate measure to use. Since the surface is a manifold, by definition there is a positive integer m such that each point of S lies in a region that is geometrically the same as a piece of Euclidean m-dimensional space. So, the measure to use on the manifold S is the one we naturally associate with Euclidean m-dimensional space, namely, m-dimensional volume.
Geometry and Dynamical Systems
To summarize, the N-body problem is a classical dynamical system in which the time-one map T is a continuous one-to-one map of the phase space X onto itself. The inverse map, T-1, is also continuous. Thus, T is a homeomorphism. There is a natural measure on the phase space X that is invariant under T. From one point of view, this measure is the volume element corresponding to the dimension of the phase space. From another viewpoint the natural invariant measure expresses the fact that the system is a Hamiltonian system. In the phase space X a surface S of constant energy forms an invariant set, and again there is an invariant measure on S corresponding to our ordinary notion of volume. The set B of all points that are not recurrent is also an invariant set with respect to T. However, it is not at all clear that we can define some natural invariant measure on B that is both nonzero and invariant under T. Many dynamical systems being studied today “live” on invariant sets that, like B, are not manifolds. Instead they are “pathological” sets, sets that at one time were thought to be the private domain of the purest and most abstract mathematicians. The examples range from Cantor sets to nowhere-differentiable curves to indecomposable continua. Many of these pathological invariant sets are "strange attractors" of dynamical systems; the system is "attracted" in the sense that it will eventually end up on the set from any starting point. (The discovery of one of the first strange attractors is described in the section Cubic Maps and Chaos of the article "Iteration of Maps, Strange Attractors, and Number Theory-An Ulamian Potpourri.”)
Properties of Invariant Sets. Let us now indicate some of the problems and techniques used in studying such sets in the context of dynamical systems. We will consider discrete dynamical systems, that is, systems in which the time evolution is described by discrete steps. We consider a function T that maps a space X into itself and the iterates of T, that is, T',12,13, where Tn+(x) = T (T"(x)). We are interested in an invariant set—a subset M of X such that T(M) CM. The simplest invariant set consists of a fixed point x such that T(x) = x; a more complicated invariant set is a periodic orbit, a set consisting of the points x,T(x),...,7"-'(x), and T"(x) = x. Invariant sets are further classified according to how points near the invariant set behave under T. An invariant set M is called an attractor if there is a region U surrounding M such that if x EU, then T"(x) gets closer and closer to M as n increases. On the other hand, M is called a repeller if there is a region U surrounding M such that if XE (U -M), then T"(x) is not in M for n sufficiently large. For example, if X is the real number line, then 0 is an attracting fixed point for T(x) = x/2 and a repelling fixed point for T -'(x) = 2x. The intrinsic properties of an invariant set are also of interest. For example, one might want to know whether there is a point x of M such that the orbit of x, that is, x,T(x), 12(x),..., is dense in M. If I is an irrational rotation of the plane, then the unit circle is invariant and the orbit of every point on the circle is dense in the circle. Another possibility is that T is topologically mixing on M; that is, for every region U of M there is some n such that M CT"(U).
One central problem we will look at in some depth is the construction of “natural” or useful invariant measures for the sets M. In particular we want a measure y such that u(X – M) = 0 and u (T-'(B)) = u(B) for each measurable subset B of M. That is, the measure is zero for points outside the invariant set M and is invariant with respect to the inverse of T.
Cantor's Set as an Invariant Set. Let us consider a simple example of a map whose invariant set is Cantor's middle-third set. Let X be the real number line and let T(x) = (3/2)(1 – 12x – 11). Then T is a two-to-one map of X into itself, the "triangle function" whose graph is shown in Fig. 7. This transformation can also be written in the following form:
if x < 1/2 T(x) =
3(1 – x) if x > 1/2.
Now consider what happens to x under the iterates of T. If x < 0, then T"(x) = 3" x
Hausdorff Measure and Dimension. If we think of T as an analog of a dynamical system whose motion in phase space is restricted to a Cantor set we might like to find a natural measure on this set. Our problem is: Which one of the many possible invariant measures is useful? One clue for determining the appropriate measure for the N-body problem was the fact that the phase space is a manifold and we therefore know the dimension of the space. We could then use the corresponding volume in the Euclidean space of that dimension to guide us to the correct measure. But what do we do with the Cantor set of our example? What is its dimension? In the early part of this century Felix Hausdorff developed an approach for determining the dimension of a general metric space (a space with a notion of a metric, or distance, between points) in terms of measures associated with the metric. It is perhaps surprising at first that the dimension of a space may not be an integer. Such spaces have been christened fractals by Mandelbrot, and he has provided many examples of their occurrence in physical phenomena. The idea behind Hausdorff's generalization of dimension is very simple