5

Definitions

For the purpose of this post, a (topological) dynamical system is a compact metric space $X$ equipped with a homemomorphism $T:X\to X$.

We say that a subset $S$ of $\mathbb Z$ is relatively dense in $\mathbb Z$ if there is a positive integer $N$ such that for all $a\in \mathbb Z$ the set $\{a+1, a+2, \ldots, a+N\}$ has non-empty intersection with $S$.

Let $x$ be a point in a dynamical system $(X, T)$.

$\bullet$ The orbit of $x$ is defined as $O_x=\{T^nx:\ n\in\mathbb Z\}$.

$\bullet$ We say that a point $x\in X$ is almost periodic if for all neighborhoods $U$ of $x$ in $X$, the set $\{n\in \mathbb Z:\ T^nx\in U\}$ is relatively dense in $\mathbb Z$.

$\bullet$ We say $x$ periodic if the orbit of $x$ is finite.

Clearly, any periodic point is almost periodic.

Question 1

Assuming $(X, T)$ is a dynamical system with $|X|=|\mathbb N|$, is it necessary that every almost periodic point is also periodic?

I do not know the answer to the above question. In fact, I do not know any "good" examples of a countable dynamical system. If you are aware of good examples then please feel free to share.

Question 2

Assuming $(X, T)$ is a dynamical system with $|X|=|\mathbb N|$, is it necessary that $X$ has a periodic point?

The answer to this question is in the affirmative. This is because we know that there is a $T$-invariant probability measure $\mu$ on $X$. Since $X$ is countable, there is a point $x$ in $X$ such that $\mu(x)>0$. Now the orbit of $x$ must be finite, for otherwise, by the $T$-invariance of $\mu$, we would have that $\mu(X)=\infty$.

Can we give an argument with does not go via measure theory and is purely topological in nature?

  • 1
    Just a heads up that a compact (hence complete) metric space with no isolated points must be uncountable. Hence when you are restricting yourself to this case, you are implicitly assuming that there are plenty of isolated points. This might help in answering the first question. – PStheman Feb 26 '20 at 13:21
  • @PStheman Yes. I considered that but could not get anywhere. – caffeinemachine Feb 26 '20 at 13:34
  • @caffeinemachine is it possible that if $T^{n_k}y \to x$, then $x$ is a periodic point (i.e. every limit point of an orbit is a periodic point)? The only examples I have in mind of topological dynamical systems with $|X| = \mathbb{N}$ are ${0}\cup{\frac{1}{1},\frac{1}{2},\dots}$ and countably many copies of those arranged in a compact way. – mathworker21 Feb 26 '20 at 21:50
  • @mathworker21 I do not know if the condition forces periodicity of $x$, but if true that will be great. – caffeinemachine Feb 27 '20 at 11:47
  • @caffeinemachine it does. see comments in answer – mathworker21 Feb 27 '20 at 13:02

1 Answers1

4

Here is the answer rewritten to avoid the transfinite induction.

Given a topological space $X$, let $X'$ denote the set of non-isolated points of $X$.

Definition. Given a homeomorphism $T: X\to X$, a point $x\in X$ is called recurrent (with respect to $(X,T)$) if for each neighborhood $U$ of $x$ there exists $n\ge 1$ such that $T^n(x)\in U$.

This condition is weaker than almost periodic.

Lemma 1. Let $X$ be a countable compact metrizable space, $T: X\to X$ a homeomorphism. Then every recurrent point $x\in X$ is periodic.

Proof. Consider the collection ${\mathcal I}_x$ of all compact $T$-invariant subsets of $X$ containing $x$. Let $C_x$ denote the intersection of all members of ${\mathcal I}_x$. Clearly, $C_x\in {\mathcal I}_x$. I claim that $x$ is an isolated point of $C_x$. Indeed, since $C_x$ is countable and compact metrizable, it has some isolated points, $C'_x\ne C_x$. If $x\in C'_x$ then $C'_x\in {\mathcal I}_x$ and $C'_x$ is strictly smaller than $C_x$, which is a contradiction. Hence, $x$ is isolated in $C_x$. The point $x$ is still recurrent with respect to $(C_x,T)$. Since $x$ is isolated in $C_x$, $\{x\}$ is a neighborhood of $x$ in $C_x$. Hence, by recurrence, there exists $n\ge 1$ such that $T^n(x)\in \{x\}$, i.e. $T^n(x)=x$, i.e. $x$ is $T$-periodic. qed

This answers Question 1. To answer Question 2, I will prove a stronger result:

Lemma 2. Let $X$ be a nonempty compact metrizable topological space, $T: X\to X$ is a homeomorphism. Then $X$ contains recurrent points. Equivalently, $X$ contains a $T$-invariant compact nonempty subset $X_0$ such that every $T$-orbit in $X_0$ is dense in $X_0$.

Proof. Consider the poset ${\mathcal I}$ of all nonempty $T$-invariant compact subsets of $X$ (with the partial order given by inclusion). Clearly, the intersection of members of each totally ordered (nonempty) subset in ${\mathcal I}$ belongs to ${\mathcal I}$. Hence, by Zorn's Lemma, ${\mathcal I}$ contains a minimal element $X_0$. By the minimality, every $T$-orbit in $X_0$ is dense (otherwise, take the closure of a non-dense $T$-orbit in $X_0$). qed

Combining the two lemmata, we see that if $X$ is countable, compact, metrizable, nonempty, then for each homeomorphism $T: X\to X$, there exists a $T$-periodic point.

Moishe Kohan
  • 111,854
  • "so it answers Question 2 as well". can you explain this? – mathworker21 Feb 27 '20 at 00:31
  • also, do you use the actual definition of "almost periodic" anywhere? – mathworker21 Feb 27 '20 at 00:32
  • @mathworker21: As I said (last sentence), I am using a weaker notion: Every almost periodic point satisfies my condition. – Moishe Kohan Feb 27 '20 at 00:41
  • ok I missed that last sentence. I asked in a comment above whether $T^{n_k}y \to x$ for some $y$ and $(n_k)_k$ implies $x$ is a periodic point. I think the existence of such a $y$ and $(n_k)_k$ is equivalent to "for all $U$ there is $n$ with $T^x \in U$". I'll try to work through your answer soon. I need to work through it since I don't know what transfinite induction is; I assume I can just prove what needs to be proven, using a Zorn's lemma argument. – mathworker21 Feb 27 '20 at 01:02
  • @mathworker21: Yes, you can use Zorn's lemma instead. I am not sure about the condition $T^{n_k}\to x$ for some $y$ and $n_k$. What I am using is the case $y=x$, which is the notion of recurrence and it is weaker than the condition of "almost periodic." – Moishe Kohan Feb 27 '20 at 01:39
  • I understand what you are using. I said "is equivalent to". Of course, your thing implies my thing (by taking $y=x$). To see that my thing implies your thing: If $T^{n_k}y, T^{n_{k+1}}y \approx x$, then $T^{n_{k+1}-n_k}x \approx x$; this can be made rigorous I'm pretty sure. – mathworker21 Feb 27 '20 at 02:15
  • @MoisheKohan Thank you for the answer. I am not familiar with transfinite induction but I want to learn it. I read triple_sec's answer here: https://math.stackexchange.com/questions/1343527/principle-of-transfinite-induction which gives a formulation without mentioning ordinals. Can you give me a hint as to how to make use of this formulation in the situation at hand? Thanks. – caffeinemachine Feb 27 '20 at 11:22
  • @caffeinemachine you don't know Zorn's lemma? – mathworker21 Feb 27 '20 at 13:02
  • @mathworker21: According to wikipedia (https://en.wikipedia.org/wiki/Limit_set) your definition (which says that $x\in \omega_T(y)$) does not imply recurrence of $x$. But they do not give an example. I suggest, you post a separate question about this. – Moishe Kohan Feb 27 '20 at 16:36
  • @mathworker21 I know Zorn's lemma but somehow I am not able to apply it here. I will try for some more time and will ask for help if I still do not get it. – caffeinemachine Feb 27 '20 at 19:34
  • @caffeinemachine: Did you understand the proof? – Moishe Kohan Mar 01 '20 at 04:49
  • @MoisheKohan I understand the new proof. I still (shamefully) do not understand the previous one using Zorn's lemma. – caffeinemachine Mar 01 '20 at 09:09
  • I think I get the Zorn's lemon's argument now. Thanks. Great answer. – caffeinemachine Mar 01 '20 at 17:05
  • 1
    @caffeinemachine you said you don't want to involve measure theory, but it might be worth mentioning that if you know that there is a $T-$ invariant probability measure $\mu$ on $X$, then lemma $2$ of this answer can actually be strengthened to say that the set of recurrent points has measure $1$ which by lemma $1$ implies the set of periodic points also has measure $1$. – Aryaman Jal Mar 19 '20 at 18:12
  • @AryamanJal Yes I think you are right. Thanks. – caffeinemachine Mar 19 '20 at 19:18