14
$\begingroup$

Let $p$ be a prime such that $2$ is a primitive root of $p$. We want to find a bijective function $f: (\mathbb{Z}/p\mathbb{Z})^× \to (\mathbb{Z}/p\mathbb{Z})^× $ s.t.

$$f(2k) = f(k) + f(f(k)) $$ $$f(-k) = -f(k)$$

As far as I could compute $f = Id$ seems to be the only solution.

Original motivation comes from doing some constructions on Coxeter groups. We didn't need $2$ to be a primitive root but there are non-trivial solution in that case (for $p=17$ we have a solution with $f(1) = f(3) = 1$ & some more) so I'll analyse them later. We also didn't need $f$ to be bijective but you can show the preimages have a coset-like behaviour and I have shown $|Im(f)|\neq 2,4,6$ for large $p$. I'm hoping we can show the bijectivity eventually but for now let's assume it. (see comments)

Such question doesn't seem to be available in literature. I could find 1. Functional Equations $f:G \to \mathbb{C}$ for some group $G$ and 2. Additive Combinatorial statements about sum of permutations, none of which seems to help yet.

I have gather some very small insignificant constraints for $p>5$ (let me add them from my memory. I'll add more when I get to my notes). Note that I'm looking for non-trivial solutions so I'm trying to avoid $f=Id$ too:

  1. If $f$ is a solution then so is $afa^{-1} \quad a \in (\mathbb{Z}/p\mathbb{Z})^×$
  2. $f(k) \neq k,-k,2k,-2k$ for any $k$

We can derive more using these, for instance,

  1. $f(2k) \neq -f(k), f(k), 2 f(k), 3f(k)$ for any $k$
  2. if there's no solution with $f(1)=t$ then there's no solution with $f(k)=tk$ for any $k$
  3. Let $m$ be s.t. $(m,p-1)=1$. If any non-trivial solutions exist then we can find one which satisfies $f(k2^m) \neq f(k)2^m $

(More from my notes)

  1. $f$ as a permutation has no $2$-cycles
  2. If for some $a$, $f(a) = 4a$ then $f(2a) = -a \ \& \ f(4a) = -5a$

Been stuck on this for weeks so I'd appreciate any kind of help, including small insights, your computations, your own literature search and attempts, etc.

Thanks in advance!

$\endgroup$
7
  • 5
    $\begingroup$ "As far as I could compute $f = Id$ seems to be the only solution." How far could you compute? $\endgroup$ Commented Dec 5 at 20:52
  • 1
    $\begingroup$ The Mace4 Counterexample finder seems to be very good at searching for solutions to this problem (though a specialized tool would be better). It found the function [0,1,2,1,4,13,2,8,8,9,9,15,4,13,16,15,16] mod 17 without the bijectivity constraint, and with the bijectivity constraint I was able to rule out any counterexample with $p < 600$ using it in under a minute, even before imposing that $2$ should be a primitive root. In my experience, when mace4 does this well it is because there is some easy reason why a counterexample is impossible, but maybe $p$ just needs to be very special. $\endgroup$ Commented Dec 6 at 23:44
  • 1
    $\begingroup$ yesterday I found a counterexample when $f$ is not bijective but still odd and 2 a primitive root. Put $f(k)=ak^n$, taking $n^2 \equiv n \ (\text{mod} \ p-1 )$ allows u to cancel terms and we get $a^n = 2^n -1$. One example is $(p,n,a)=(65,55,4)$ and there are more. If u drop the odd condition $p=19$ also gives a solution. My original conjecture was without the bijectivity so that's been dismissed. It seems bijectivity is a strong condition here even if u drop the odd and primitive root condition. Also I apologize for late interaction, I've been busy with some work. I will respond to comments $\endgroup$ Commented Dec 7 at 13:06
  • $\begingroup$ I'm confused, isn't $p$ not prime in that example? $\endgroup$ Commented Dec 7 at 13:12
  • 1
    $\begingroup$ Looks like it is $(p, n, a) = (19, 10, 3)$ or $(p, n, a) = (19, 10, 16)$. Either way I think I must have messed up something in my most recent run, so I guess you need to take my earlier claims about the bijective case with a grain of salt in case I misconfigured something there as well. $\endgroup$ Commented Dec 7 at 14:00

1 Answer 1

1
$\begingroup$

Claim. The one and only function $f$ that satisfies the conditions stated above is the identity function.

Proof. Put $N=p-1$. Since $2$ is a primitive root mod $p$, every $x\in\mathbb F_p^\times$ can be written uniquely as $x=2^n$ with $n\in\mathbb Z/N\mathbb Z$. Define $$ g:\mathbb Z/N\mathbb Z\to\mathbb Z/N\mathbb Z $$ by $f(2^n)=2^{g(n)}$. Because $f$ is a bijection on $\mathbb F_p^\times$, the map $g$ is a permutation of $\mathbb Z/N\mathbb Z$.

Plugging $x=2^n$ into $f(2x)=f(x)+f(f(x))$ gives, for all $n\in\mathbb Z/N\mathbb Z$,

$$ 2^{g(n+1)} = 2^{g(n)} + 2^{g(g(n))}. \tag{1} $$

If $f$ has a fixed point, then so does $g$: $$ f(2^n)=2^n \iff g(n)=n. $$ Moreover, if $g(n_0)=n_0$ for some $n_0$, then $(1)$ evaluated at $n=n_0$ gives $$ 2^{g(n_0+1)}=2^{n_0}+2^{n_0}=2^{n_0+1}, $$ so $g(n_0+1)=n_0+1$, and iterating yields $g(n)=n$ for all $n$, hence $f(x)=x$ for all $x$. So, if $g$ has a fixed point, then so does $f$.

Define vectors $u,v\in\mathbb F_p^N$ (indexed by $\mathbb Z/N\mathbb Z$) by $u_n:=2^n$ and $v_n:=2^{g(n)}$. Let $S$ be the cyclic shift operator $(Sw)_n:=w_{n+1}$, and let $P_g$ be the permutation operator $(P_gw)_n:=w_{g(n)}$. We can easily see that $(1)$ is exactly $$ (Sv)_n = v_n + (P_g v)_n $$ for all $n$, i.e. $$(S-I)v = P_g v.$$

Substitute $v=P_g u$: $$ (S-I)P_g u = P_g^2 u $$

Left-multiply by $P_g^{-1}$ and set $C:=P_g^{-1}SP_g$: $$ (C-I)u = P_g u $$

If we look at the indices (which is only natural, because these operators act, after all, on the indices of some vectors), $S$ corresponds to the $N$-cycle $s(n)=n+1$, hence $C$ corresponds to the conjugate permutation $c := g^{-1}\circ s\circ g$, i.e. $c(n)=g^{-1}(g(n)+1)$. The relevant formulation of this identity being $$ g(c(n))=g(n)+1 \tag{2} $$ for all $n$.

Now, the equation $$ (C-I)u=P_g u $$ may be written as: $$ u_{c(n)}-u_n = u_{g(n)}. $$ Since $u_n=2^n$, this becomes $$ 2^{c(n)} - 2^n = 2^{g(n)} \tag{3} $$ for all $n$.

Fix $n_0\in\mathbb Z/N\mathbb Z$ and define the $c$-orbit $n_k:=c^k(n_0)$ for $k=0,1,\dots,N-1$. Because $c$ is an $N$-cycle, the residues $n_0,n_1,\dots,n_{N-1}$ are all distinct and run through all of $\mathbb Z/N\mathbb Z$. From $(2)$, applied repeatedly along the orbit, we get

$$ g(n_k)=g(n_0)+k \pmod N \tag{4} $$ for all $k$.

Look at $(3)$ and apply $(4)$ with $n=n_k$: $$ 2^{n_{k+1}} - 2^{n_k} = 2^{g(n_k)} = 2^{g(n_0)+k} \tag{5} $$ for all $k$.

It is trivial to see that if we sum $(5)$ from $k=0$ to $t-1$: $$ 2^{n_t}-2^{n_0} = \sum_{k=0}^{t-1}2^{g(n_0)+k} = 2^{g(n_0)}(2^t-1). $$

Thus $$ 2^{n_t} = A\cdot 2^t + B, \tag{6} $$ where $A:=2^{g(n_0)}\neq 0$ and $B:=2^{n_0}-2^{g(n_0)}$.

Now we use the fact that $2$ is a primitive root mod $p$ once more: as $t$ runs through $0,1,\dots,N-1$, the values $2^t$ run through all of $\mathbb F_p^\times$; and since $n_t$ runs through all residues mod $N$, the values $2^{n_t}$ also run through all of $\mathbb F_p^\times$. So the affine map $\varphi(x)=Ax+B$ satisfies $\varphi(\mathbb F_p^\times)=\mathbb F_p^\times$.

If $B\neq 0$, then $\varphi(x)=0$ has the solution $x=-B/A$, and since $A\neq 0$ this solution is valid and nonzero, i.e. $x\in\mathbb F_p^\times$. That would force $0\in\varphi(\mathbb F_p^\times)$, which is contradictory. Hence $B=0$ in $(6)$, so $2^{n_0}=2^{g(n_0)}$, hence $g(n_0)=n_0$.

Thus $g$ has a fixed point, hence $g=\mathrm{id}$, hence $f=\mathrm{id}$ on $\mathbb F_p^\times$. $$\tag*{$\Box$}$$


Remark: Curiously, we did not use the fact that $f$ is an odd function, but, as it has been discussed in the comments, bijectivity was necessary for our proof.

New contributor
Sergiu Bisceanu is a new contributor to this site. Take care in asking for clarification, commenting, and answering. Check out our Code of Conduct.
$\endgroup$
2
  • $\begingroup$ Plz help me with something. I had got to the equation (3) but with $c(n)=g(g^{-1}(n) + 1)$ rather than $c(n)=g^{-1}(g(n) + 1)$, since the linear operators are defined such that multiplication of operator on left corresponds to composition on right (correct? also composition only distribute over $+$ from right). But I can also see this directly from $f\circ 2 \circ f^{-1} = Id + f$ on $2^n$. If I'm missing something plz help me. $\endgroup$ Commented Dec 15 at 14:40
  • $\begingroup$ There doesn't seem to be way around to save the proof when (2) is $$g^{-1](c(n)) = g^{-1}(n) + 1$$ $\endgroup$ Commented yesterday

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.