We will give an indication of why period 3 seems to occur at the end of all the occurrences of finite attractors for quadratic maps. This is just a very small piece of a fantastic result called Sarkovskii's theorem. Our discussion is a rehashing of more detailed material found in [Dev89b].
LetIn short, period 3 implies periodic points of all periods. This doesn't say anything about the stability of these periodic points.be a continuous map. Suppose f has a 3-cycle. Then f has a periodic point of prime period n for all integers
.
The proof of this theorem begins by assuming we have a 3-cycle for f(x). This would consist of three distinct points a,b,c such that f(a)=b, f(b)=c, and f(c)=a. By symmetry, we may assume that a is the smallest of the three: a< b,c. Then there are two cases b<c and c<b. We shall assume b<c and leave the second case to the reader to carry out.
Assuming a<b<c, we define the intervals A=[a,b] and B=[b,c]. Since f(a)=b and f(b)=c and f is continuous, we have
(The Intermediate Value Theorem says that on a closed interval [a,b] a continuous function assumes all values between f(a) and f(b).) Similarly, since f(b)=c and f(c)=a, we have
Question: What do these relations remind you of? See here.
The key idea is to iterate these subset relations to prove the existence of periodic points of a given period. To do this, we require two lemmas.
Lemma 1: Let I be a closed interval such that. Then f has a fixed point in I.
Let the interval be I=[c,d]. Let S be the subset of
such
that f(x)=c, and let T be the subset of
such that f(x)=
d. Since f is continuous, both S and T are closed subsets of
I, and hence compact. Since
, S and T are
disjoint. Therefore, we can consider the distance between these two sets
Since S and T are compact, this minimum is achieved for some
points
and
. If
, then f(c)=c and we
have found our fixed point. Similarly, if
, d is a fixed
point. Now assume
and
. Then
,
while
. By the Intermediate Value Theorem, there
is a point x between
and
such that f(x)-x=0.
This is our fixed point.
Lemma 2: Suppose I and J are closed intervals such that. Then there is a closed interval
such that
.
This is proved by basically the same proof as the first lemma. Let
I=[a,b] and J=[c,d]. Let S be the set of
such that
f(x)=c, and T the set of x with f(x)=d. Once again we find
,
in I such that
. Let
be the
interval with endpoints equal to
and
. We claim that
. The Intermediate Value Theorem (IVT) certainly implies
that
. If equality doesn't happen, then there is a
point
such that f(x)>d or f(x)<c. Suppose the former
happens (the latter case may be argued similarly). Then again by the
IVT there is a point
between
and x such that
.
However,
. This is a
contradiction since
and
. Therefore, no such
point x can exist, and we have proved our lemma.
Now we may return to the 3-cycle a<b<c and the proof of our theorem.
Since
, by Lemma 1, f has a fixed point in B.
That takes care of the theorem for n=1.
Since
, there is an interval
such that
, by Lemma 2. Since
, there is an interval
such that
. Then
. By Lemma 1,
has a fixed point p in
. Since
,
we cannot have f(p)=p, unless p=b the common boundary of A and
B. But b is clearly not a fixed point of f, since f(b)=c.
Thus, p is a periodic point of prime period 2.
Having been supplied with a periodic point of prime period 3, we suppose now that n>3. We define a sequence of intervals as follows:
Suppose
. Then the iterates of f on p cycle through
the points
forever. By our construction, these points are all contained in
, and so all the iterates are forever contained in B. On the
other hand,
(this was the
clever part of our construction). That can only happen if
, the endpoint shared by A and B. Then
, and
, a contradiction. This proves that the
period of p is no smaller than n.
This theorem is dramatically visualized in the bifurcation diagram, although that's a story for another day.