5
$\begingroup$

Let $\newcommand{\tn}{\{0,1\}^\mathbb{N}}\tn$ be the collection of all infinite binary sequences. For $s\in\tn$ and $k\in\mathbb{N}$ let the left-shift of $s$ by $k$ positions, $\ell_k(s)\in \tn$, be defined by $\ell_k(s)(n) = s(n+k)$ for all $n\in\mathbb{N}$.

The xor of $s,t\in\tn$, in symbols $s\oplus t \in \tn$ is defined by $(s\oplus t)(n) = s(n) + t(n)$ for all $n\in\mathbb{N}$, where $+$ denotes addition modulo $2$, which is the same as the binary XOR operation.

We say that $s\in\tn$ is irrational if for no $k\in\mathbb{N}$ the binary sequence $\ell_k(s)\in\tn$ is periodic. Otherwise we say that $s$ is rational. (Note that $s$ is rational if and only if the binary floating point number that $s$ represents is a rational number.)

Question. Is there an irrational $s\in\tn$ such that $s\oplus \ell_1(s)\in \tn$ is rational?

$\endgroup$
5
  • 3
    $\begingroup$ No, xor is a two-to-one cellular automaton. A periodic point generates a finite subshift under the shift, so the preimage subshift is finite, so has only periodic points. $\endgroup$ Commented Dec 9, 2024 at 14:40
  • 1
    $\begingroup$ (This argument is clearer for two-sided sequences, but the CA is one-sided so it works out) $\endgroup$ Commented Dec 9, 2024 at 14:47
  • $\begingroup$ Is ${\tt AND}$ , $\land$, a two-to-one cellular automaton? If yes, the argument would not work. If we take $s = {\tt 01\, 001\, 0001 \, 00001}\ldots$ then $s \land \ell_1(s)$ is the constant 0 sequence. A similar argument can be made for ${\tt OR}$, $\vee$. Maybe you can elaborate your argument in an answer - many thanks. $\endgroup$ Commented Dec 9, 2024 at 14:51
  • 2
    $\begingroup$ AND is not surjective, so not 2-to-1. I can write an answer at some point. A key point for doing one-sided is that xor is left-permutive, I guess that's what I meant when I said it's one-sided. $\endgroup$ Commented Dec 9, 2024 at 15:07
  • $\begingroup$ Thanks @VilleSalo, an answer would be very much appreciated! $\endgroup$ Commented Dec 9, 2024 at 15:26

2 Answers 2

16
$\begingroup$

No, there's no irrational $s$ with this property. Here's a concrete hands-on argument that complements Ville Salo's answer.

The sequence $d = s\oplus \ell_1(s)$ is just the sequence of differences (mod 2) between successive elements of $s$. Given one entry in the sequence, $s(j)$, we can reconstruct the rest of $s$ from $d$, since: $$s(j + 1) = s(j) \oplus d(j)$$ Suppose $d$ is rational, so $\ell_k(d)$ is periodic for some $k$. Since we can left-shift without changing rationality, we can shift both sequences left by $k$ and assume $d$ periodic. And if $d$ has period $n$, then $s$ must have period no more than $2n$, since $$s(j + 2n) = s(j) \oplus \bigoplus_{i=0}^{2n-1} d(i) = s(j) \oplus 2 \bigoplus_{i=0}^{n-1} d(i) = s(j)$$

Edit for transparency. This was part of an informal test of AI capabilities. The idea came from OpenAI's o1 model, although I rewrote its answer. See this meta thread for details.

$\endgroup$
3
  • 3
    $\begingroup$ My answer (which I'll add later, but not now) uses a bit of theory to get around actually giving any formulas. But of course this is the easy way. $\endgroup$ Commented Dec 9, 2024 at 16:14
  • $\begingroup$ I will accept Martin‘s answer while looking forward to Ville‘s answer $\endgroup$ Commented Dec 9, 2024 at 16:27
  • 2
    $\begingroup$ I read the thread on meta a I liked it a lot. Very balanced and informative, we'll need more and more things like that. $\endgroup$ Commented Jan 19 at 8:37
7
$\begingroup$

I'll prove a general theorem. Unlike in my comments, I won't use any specific theorems.

Let $A$ be a finite alphabet and $X \subset A^{\mathbb{N}}$ a subshift of finite type, or SFT, meaning $X$ is the set of sequences, or points, that avoid a finite set of forbidden words $W \subset A^*$, in the sense that $x \in A^{\mathbb{N}}$ belongs to $X$ if and only if for all $w \in W$ and $i \in \mathbb{N}$, we have $x_{[i, i+|w|-1]} \neq w$. We consider $X$ under the induced topology coming from the product topology. Note that $\sigma(X) \subset X$ where $\sigma(x)_i = x_{i+1}$ is the left shift.

We say an SFT is transitive if for every $u, v \in A^*$ that appear in its points, some word $uwv$ also does. This is equivalent to topological transitivity of the left shift map, and forces $\sigma(X) = X$.

An endomorphism is a $\sigma$-commuting continuous self-map $f : X \to X$. Equivalently, there is a radius $r \geq 0$ and a local rule $F : U \to A$ where $U \subset A^{1+r}$ is the set of words appearing in points of $X$, such that $f(x)_i = F(x_{[i, i+r]})$ for all $i \in \mathbb{N}, x \in X$. We say $f$ is left-injective if for some choice of $r, F$ (equivalently, any choice of $r, F$) we have $F(aw) \neq F(bw)$ whenever $a, b \in A, w \in A^r$ and $a \neq b$.

The poster is interested in the case $X = \{0,1\}^{\mathbb{N}}$, $r = 1$, $F(a, b) = a \oplus b$, which is clearly left-injective.

We say a point $x \in X$ is periodic if $\sigma^n(x) = x$ for some $n \geq 1$. Equivalently, $x = www...$ for some word $w \in A^*$. We say an endomorphism $f$ has property P if every preimage of a periodic point is itself necessarily periodic. This is what the poster asked about.

Theorem. An endomorphism $f : X \to X$ of a transitive SFT $X$ has property P if and only if it is left-injective.

Proof. Suppose first that $f$ is left-injective (this is the direction the poster is interested in). Let $z = wwww...$ be a periodic point, and let $x$ be any preimage. Then from $x_{[i, i+r-1]}$ and $z_{[0,i-1]}$ we can uniquely determine $x_{[0, i-1]}$ by using the left-injectivity property to determine symbols one by one from right to left (i.e. in order $i-1, i-2, ..., 0$).

These symbols are determined, in fact, by a finite state machine, which is fed with periodic input. The memory is the word $x_{[i-j, i-j+r-1]}$ determined so far, and the input is the reverse of the word $w$, repeatedly.

By the pigeonhole principle, the output of such a process must be eventually periodic, and thus $x_{[0, i-1]}$ must be eventually periodic. Furthermore, we can give an explicit bound on this period and the transient. The eventual period is seen on the left, so in fact $x$ begins with arbitrarily long periodic segments with a bounded period. In other words, $x$ is periodic.

Suppose then that $f$ is not left-injective. Then there exist $aw, bw$ with the same image under $F$. Since $X$ is a transitive SFT, there is a periodic point of the form $x = awuawuawuawu...$ whose image is necessarily periodic, since endomorphisms preserve periods. But clearly for $y = bwuawuawu...$ we have $f(y) = f(x)$, and $y$ is not periodic. Square.

$\endgroup$
1
  • $\begingroup$ +1 Very nice theorem and presentation! $\endgroup$ Commented Dec 10, 2024 at 9:40

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.