This must be known or easy for some of you, but here goes:

Suppose $f_0,f_1:[n]\to [n]$ are invertible functions, where $[n]=\{0,\dots,n-1\}$ is a set of $n$ elements. For a word $w=w_1\dots w_m\in\{0,1\}^m$ we define $f_w=f_{w_m}\circ f_{w_{m-1}}\circ\dots\circ f_{w_1}$ (or make it the opposite order, if you prefer). Suppose $c, d$ and $m$ are such that there is exactly one word $w\in\{0,1\}^m$ with $f_w(c)=d$. Does it follow that $n\ge m+1$?

I've checked that it does follow for all $m\le 11$ (eleven).


Fix $c\in [n]$. Let $\mathcal R_m(c)$ be $\{f_w(c)\colon |w|=m\}$ and $r_m(c)=|\mathcal R_m(c)|$. We define $\mathcal R_0(c)$ to be $\{c\}$.

Claim: Let $m\ge 0$. If $r_m(c)=r_{m+1}(c)$, then for all $M>m$ and for all $d\in\mathcal R_M(c)$, there are two $w$'s in $\{0,1\}^M$ with $f_w(c)=d$.

Proof: Let $S=\mathcal R_m(c)$. Then $\mathcal R_{m+1}(c)=f_0(S)\cup f_1(S)$. By invertibility if the $f_i$, $f_0(S)$ and $f_1(S)$ each have cardinality $r_m(c)=r_{m+1}(c)$, and their union also has cardinality $r_{m+1}(c)$. It follows that $f_0(S)=f_1(S)$. Now let $w\in \{0,1\}^M$ and let $f_w(c)=d$. Let $w$ be the concatenation of $u$ (of length $m+1$) and $v$ of length $M-(m+1)\ge 0$. By the above, there exists a $u'$ also of length $m+1$ with the opposite $(m+1)$st symbol such that $f_{u'}(c)=f_{u}(c)$. Now $f_v\circ f_{u'}(c)=f_v\circ f_u(c)$. $\square$

It follows that if $c$ and $d$ are such that there is a unique $w$ of length $m$ with $f_w(c)=d$, then $r_{j+1}(c)>r_j(c)$ for each $j<m$, so that $n\ge r_m(c)\ge m+r_0(c)=m+1$.

I should comment that this proof somewhat resembles that of the Morse-Hedlund theorem in symbolic dynamics.

| cite | improve this answer | |
  • $\begingroup$ Yes: I used $|f_i(S)|=|S|$. Also there are easy counter-examples if you don’t have invertibility. $\endgroup$ – Anthony Quas Sep 27 at 15:10
  • $\begingroup$ I edited the answer to make the role of invertibility explicit. $\endgroup$ – Anthony Quas Sep 27 at 15:17
  • $\begingroup$ Hey Bjørn, it would feed my ego to be a co-author on a logic paper, but I don’t think this is much of a contribution. $\endgroup$ – Anthony Quas Sep 27 at 15:54
  • 1
    $\begingroup$ The standard practice in this situation would be to acknowledge Anthony Quas's help in your paper, and possibly cite this question using a weblink. $\endgroup$ – Derek Holt Sep 27 at 16:18
  • $\begingroup$ I agree with @DerekHolt. That would be just fine. $\endgroup$ – Anthony Quas Sep 27 at 16:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.