2

This answer nicely categorises which permutations of a finite set have square roots. This prompts the following question:

Does every permutation $\sigma:\mathbb{N}\to\mathbb{N}$ that contains no finite cycles (i.e. it is "finite-cycle-free") have a square root?

Finite-cycle-free permutations can be generated by using a double (two-way) greedy algorithm that avoids cycles.

For example, $\sigma(1) = 2,\ \sigma^{-1}(1) = 3,\ \sigma(2) = 4,\ \sigma^{-1} (3) = 5,\ \sigma(4) = 6,\ \sigma^{-1} (5) = 7,\ldots. $

You can generate other finite-cycle-free permutations by choosing larger numbers, in other words, the algorithm doesn't have to be greedy. For example, $\sigma(1) = 12,\ \sigma^{-1}(1) = 27,\ \sigma(2) = 400,\ \ldots. $ But you must avoid cycles.

Anyway, the question above is what I am interested in. I haven't found a square root for the greedy algorithm in the first example. So maybe there isn't one, but right now I don't see a reason why this would be the case.

Adam Rubinson
  • 20,052
  • 1
    Any cycle-free permutation of $\mathbb{N}$ decomposes $\mathbb{N}$ into a bunch of infinite chains, each of which "looks like" $\mathbb{Z}$ equipped with the successor map; whether or not a square root exists is appropriately isomorphism-invariant, so the only relevant variable is how many orbits the permutation has. So to start, does the successor map on $\mathbb{Z}$ itself have a square root? – Noah Schweber Nov 30 '23 at 15:48
  • @NoahSchweber I don't know if the successor map on $\mathbb{Z}$ has a square root. I guess you're hinting that it doesn't, but I can't see why this is so. – Adam Rubinson Nov 30 '23 at 15:51

2 Answers2

2

Let $\tau$ be a permutation of any set. Plainly, if $\tau^2$ is finite-cycle-free, then $\tau$ is finite-cycle-free. Moreover, if $\tau$ has $k$ infinite cycles, then $\tau^2$ has $2k$ infinite cycles, since each infinite cycle of $\tau$ decomposes into two infinite cycles upon squaring. Therefore, a finite-cycle-free permutation has a square root iff it has an even (or infinite) number of cycles.

More generally, for any permutation $\sigma$ and any $n\in\mathbb N\cup\{\infty\}$, let $\alpha_n(\sigma)$ denote the number of cycles of length $n$ in the cycle decomposition of $\sigma$. Note that $\alpha_n(\tau^2)=2\alpha_{2n}(\tau)$ if $n$ is even (or infinite), while $\alpha_n(\tau^2)=\alpha_n(\tau)+2\alpha_{2n}(\tau)$ if $n$ is odd. It follows that the permutation $\sigma$ has a square root iff $\alpha_n(\sigma)$ is even (or infinite) for every even $n\in\mathbb N$ as well as for $n=\infty$.

More generally still, let us characterize the permutations which have $k^\text{th}$ roots. Let $k\in\mathbb N$ and let $\sigma$ be a permutation of any set. For $n\in\mathbb N\cup\{\infty\}$ let $a_n$ be the number of cycles of length $n$ in the cycle decomposition of $\sigma$ (a finite or infinite cardinal number) and let $d_n$ be the smallest positive divisor $d$ of $k$ such that $\frac kd$ is relatively prime to $n$; in particular $d_\infty=k$. Observe that a cycle of length $m$, on being raised to the $k^\text{th}$ power, decomposes into $\operatorname{GCD}(k,m)$ cycles of length $\frac m{\operatorname{GCD}(k,m)}$. It follows that $\sigma$ has a $k^\text{th}$ root iff $a_n$ is a multiple of $d_n$ for every $n\in\mathbb N\cup\{\infty\}$.

bof
  • 78,265
1

Any cycle-free permutation of $\mathbb{N}$ decomposes $\mathbb{N}$ into a bunch of infinite chains, each of which "looks like" $\mathbb{Z}$ equipped with the successor map; whether or not a square root exists is appropriately isomorphism-invariant, so the only relevant variable is how many orbits the permutation has.

If $\sigma$ has an even number of orbits, then $\sigma$ has a square root. This is because we can just "bounce between" pairs of orbits. For instance, if $\sigma$ has two orbits, then - reconstruing $\sigma$ as acting on $\mathbb{Z}\times\{0,1\}$ via $\sigma(z,i)=(z+1, i)$ - we can get a square root via $$(z,0)\mapsto (z,1), \quad (z,1)\mapsto (z+1,0).$$ And this generalizes easily to any even number of orbits.

It remains to non-square-rootability of the permutations with an odd number of orbits. I strongly suspect this is true, but I don't immediately see a "clean" proof of this. For example, let's consider the case of a single orbit, for simplicity thinking of $\mathbb{Z}$ equipped with the successor map $\sigma: z\mapsto z+1$. Assume $\tau^2=\sigma$. We have for example $$\tau(0)=2\implies \tau(2)=1\implies \tau(1)=3\implies\tau(3)=2\implies\tau(2)=4,$$ so $\tau(0)$ can't be $2$. Similar arguments rule out $\tau(0)$ being any number whatsoever, so indeed the one-orbit cycle-free permutation on $\mathbb{N}$ has no square root. But generalizing this to arbitrarily many odd orbits seems messy.

Noah Schweber
  • 245,398
  • Would not your argument against the square-rootability of the one-orbit cycle-free permutation also show that if there are multiple orbits, any square root must carry one orbit of $\sigma$ to a different orbit, which itself must be carried back to the original orbit, since $\tau^2 = \sigma$. Thus the orbits can be paired up as in the even case. However this leaves one orbit without a partner, creating a contradiction. There is also the case of having an infinite number of orbits, but that can be treated as an even number of orbits. – Paul Sinclair Dec 01 '23 at 18:33
  • @PaulSinclair It's not obvious to me that $\tau$ would need to be orbit-preserving (e.g. maybe $\tau$ sends one element of orbit $A$ to an element of orbit $B$ and another element of orbit $A$ to an element of orbit $C$). But I might be missing easy. Certainly I think that the parity of the number of orbits is the determining factor. – Noah Schweber Dec 01 '23 at 18:35
  • If $\tau(x,i) = (y,j)$, then $\tau(y,j) = \sigma(x,i) =(x+1,i)$ and $\tau(x+1, i) = \sigma(y,j) = (y+1,j)$. And so on. By induction every member of orbit $i$ must be mapped to orbit $j$, and vice versa. – Paul Sinclair Dec 01 '23 at 18:40