Mn if
k>0, so that no repetitions occur. These sequences,
which form a nondenumerable set, may conceivably approach
periodic sequences asymptotically and be quasi-periodic,
or they may be nonperiodic.
Finally, consider two sequences
Mo, M1,… and M0',
M1',… " where Mo'=Mo+ε.
Then for a given k, if ε is sufficiently small,
Mk'=Mk ±
2kε.
All sequences are therefore unstable
with respect to small modifications. In particular, all
periodic sequences are unstable, and no other sequences
can approach them asymptotically. All sequences except a
set of measure zero are therefore nonperiodic, and correspond
to nonperiodic trajectories.
Returning to Fig, 4, we see that
periodic sequences analogous to those tabulated above can
be found. They are given approximately by
398,…
377, 410, …
369, 391, 414, …
362, 380, 419, …
The trajectories possessing these or
other periodic sequences of maxima are presumably periodic
or quasi-periodic themselves.
The above sequences are temporarily
approached in the numerical solution by sequences beginning
at iterations 5340, 4881, 3625, and 3926. Since the numerical
solution eventually departs from each of these sequences,
each is presumably unstable.
More generally, if Mn'
= Mn+ε, and if E is sufficiently
small, Mn+k'=Mn+k+Λε,
where Λ is the product of the slopes of
the curve in Fig. 4 at the points whose abscissas are Mn,...
Mn+k-1. Since the curve apparently has
a slope whose magnitude exceeds unity everywhere, all sequences
of maxima, and hence all trajectories, are unstable. In
particular, the periodic trajectories, whose sequences of
maxima form a denumerable set, are unstable, and only exceptional
trajectories, having the same sequences of maxima, can approach
them asymptotically. The remaining trajectories, whose sequences
of maxima form a nondenumerable set, therefore represent
deterministic nonperiodic flow.
These conclusions have been based
upon a finite segment of a numerically determined solution.
They cannot be regarded as mathematically proven, even though
the evidence for them is strong. One apparent contradiction
requires further examination.
It is difficult to reconcile the
merging of two surfaces, one containing each spiral, with
the inability of two trajectories to merge. It is not difficult,
however, to explain the apparent merging of the surfaces.
At two times τ0 and τ1,
the volumes occupied by a specified set of particles satisfy
the relation
according to (30). A typical circuit
about C or C' requires about 70 iterations, so
that, for such a circuit,
|
τ2= τ1+0.7,
and, since σ+b+1=41/3,
Two particles separated from each other
in a suitable direction can therefore come together very
rapidly, and appear to merge.
It would seem, then, that the
two surfaces merely appear to merge, and remain distinct
surfaces. Following these surfaces along a path parallel
to a trajectory, and circling C or C', we
see that each surface is really a pair of surfaces, so that,
where they appear to merge, there are really four surfaces.
Continuing this process for another circuit, we see that
there are really eight surfaces, etc., and we finally conclude
that there is an infinite complex of surfaces, each extremely
close to one or the other of two merging surfaces.
The infinite set of values at
which a line parallel to the X-axis intersects these
surfaces may be likened to the set of all numbers between
zero and one whose decimal expansions (or some other expansions
besides binary) contain only zeros and ones. This set is
plainly nondenumerable, in view of its correspondence to
the set of all numbers between zero and one, expressed in
binary. Nevertheless it forms a set of measure zero. The
sequence of ones and zeros corresponding to a particular
surface contains a history of the trajectories lying in
that surface, a one or zero immediately to the right of
the decimal point indicating that the last circuit was about
C or C', respectively, a one or zero in second
place giving the same information about the next to the
last circuit, etc. Repeating decimal expansions represent
periodic or quasi-periodic trajectories, and, since they
define rational fractions, they form a denumerable set.
If one first visualizes this infinite
complex of surfaces, it should not be difficult to picture
nonperiodic deterministic trajectories embedded in these
surfaces.
8. Conclusion
Certain mechanically or thermally forced
nonconservative hydrodynamical systems may exhibit either
periodic or irregular behavior when there is no obviously
related periodicity or irregularity in the forcing process.
Both periodic and nonperiodic flow are observed in some
experimental models when the forcing process is held constant,
within the limits of experimental control. Some finite systems
of ordinary differential equations designed to represent
these hydrodynamical systems possess periodic analytic solutions
when the forcing is strictly constant. Other such systems
have yielded nonperiodic numerical solutions.
A finite system of ordinary differential
equations representing forced dissipative flow often has
the property that all of its solutions are ultimately confined
within the same bounds. We have studied in detail the properties
of solutions of systems of this sort. Our principal results
concern the instability of nonperiodic solutions. A nonperiodic
solution with no transient component
|