3
$\begingroup$

Let $F$ be the finite field ${\rm GF}(p^n)$ of order $p^n$. The automorphism $\theta(x)=x^{p^k}$ of $F$ has order $n/d$ where $d=\gcd(n,k)$. Consider the function $f(x)=\theta(x)-x^{-1}$ from $X:=F\setminus\{0\}$ to $F$ which depends on $p,n,k$.

Claim 1. For each $y\in X$ the size of the fibre $|f^{-1}(y)|$ is 0, 1, 2 or $p^d+1$ (i.e. depending on $p, d$, and not $k$).

Remark 1. The function $g\colon X\to X, g(x)=\theta(x)x^{-1}$ has fibres of size $0,p^d-1$, and the function $h\colon F\to F, h(x)=\theta(x)-x$ has fibres of size $0,p^d$. Our function $f(x)$ is a hybrid of the functions $g(x)$, $h(x)$.

If Claim 1 is correct, then there is a partition of the image $Y=f(X)$ into subsets $Y_1, Y_2, Y_3$ where each $y\in Y_1$ has 1 preimage in $X$, each $y\in Y_2$ has 2 preimages in $X$, and each $y\in Y_3$ has $p^d+1$ preimages in $X$. I am interested in the size of the subsets $|Y_1|,|Y_2|,|Y_3|$ especially when $p=2$, and $n/d$ is odd. (The set $Y$ arises with Suzuki 2-groups.) Clearly $|Y|=|Y_1|+|Y_2|+|Y_3|$ and $|X|=p^n-1=|Y_1|+2|Y_2|+(p^d+1)|Y_3|$ hold.

Claim 2. If $p=2$ and $n/d$ is odd, then $$ |Y_1|=2^{n-d},\quad |Y_2|=\frac{(2^{d-1}-1)(2^n-1)}{2^d-1},\quad |Y_3|=\frac{2^{n-d}-1}{4^d-1}. $$

Remark 2. Claim 2 is correct for $n\leqslant 24$ by Magma. The subsets $Y_i$ can be empty, for example, $Y_3=\emptyset$ if $k=n$ as $y=x^{-1}-x$ has at most 2 solutions for $x$. The conjectured size of the set $Y_1$ is the same size as the kernel of the trace map ${\rm GF}(2^n)\twoheadrightarrow{\rm GF}(2^d)$, namely $2^{n-d}$. However, $Y_1$ is not closed under addition. Surprisingly, if $p=2$, $k=1$ and $y\in Y_1$ is nonzero, then $y+y'\in Y_1$ holds for precisely $|Y_1|/2$ choices of $y'\in Y_1$ (for $2\leqslant n\leqslant 11$).

$\endgroup$
3
  • $\begingroup$ Do you mean for the codomain of $f$ to be all of $F$ rather than $X$? If $x = \pm 1$ we have $f(x) = x^{p^k} - x^{-1} = 0$. $\endgroup$ Commented Jul 8 at 8:44
  • $\begingroup$ Thanks Justin. Yes, $f$ maps from $X$ to $F$. I corrected the typo. $\endgroup$ Commented Jul 8 at 8:52
  • $\begingroup$ It may help to note that $n/d=2m+1$ odd implies $n-d=2md$ and so $|Y_3|=(4^{md}-1)/(4^d-1)$ is obviously an integer in Claim 2. Different formulas for $|Y_1|, |Y_2|, |Y_3|$ hold when $n/d$ is even. $\endgroup$ Commented Jul 9 at 0:51

1 Answer 1

2
$\begingroup$

$\newcommand{\fQ}{\mathbb F_Q}\newcommand{\fQs}{\fQ\setminus\{0\}}$Claim 1 is answered at this related question, so we care about Claim 2. In addition to the notation from the question, we set $q=p^k$, $Q=p^n$, $n_i=|Y_i|$. As noted there, we have \begin{equation} Q-1=n_1+2n_2+(p^d+1)n_3. \end{equation} With $f:\fQs\mapsto\fQ$, $x\mapsto x^q-1/x$, let $v_r$ be the number of subsets $\{x_1,\dots,x_r\}\subseteq\fQs$ of size $r$ with $f(x_i)=f(x_j)$ for all $i,j$. Then clearly \begin{align*} v_2 &= 2n_2+p^d(p^d+1)n_3\\ v_3 &= (p^d-1)p^d(p^d+1)n_3. \end{align*} Note that once we know the values of $v_2$ and $v_3$, we get $n_1,n_2,n_3$ from these three linear equations in triangular form.

Lemma 1. For $x\in\fQs$ and $z\in\fQ\setminus\{0,1\}$, the following are equivalent:

  1. $f(x)=f((1-z)x)$.
  2. $z^q-z^{q-1}=1/x^{q+1}$.

Proof. A straightforward computation, using $(1-z)^{q+1}=(1-z)^q(1-z)=(1-z^q)(1-z)$.

From now on we add the assumptions from Claim 2, namely that $p=2$ and that $n/d$ is odd.

Lemma 2. $v_2=2^n-2$.

Proof. We need to count the number of pairs $(x,y)$ where $x,y\in\fQs$, $x\ne y$, and $f(x)=f(y)$. Write $y=(1-z)x$, where $z\in\fQ\setminus\{0,1\}$. By Lemma 1, this is equivalent to $z^q-z^{q-1}=1/x^{q+1}$. As $q+1$ is relatively prime to $Q-1$ (see the lemma at related question), the map $X\mapsto X^{q+1}$ is bijective on $\fQ$. Therefore we get the same count if we replace $1/x^{q+1}$ with $x$. Thus, as any $z\in\fQ\setminus\{0,1\}$ gives a nonzero $x$, we obtain $v_2=Q-2$.

Lemma 3. $v_3=2^n-2^d$.

Proof. Now let $x$, $(1-z_1)x$, and $(1-z_2)x$ be three distinct nonzero elements from $\fQ$ which are mapped to the same element under $X\mapsto X^q-1/X$. Again, by Lemma 1, this is equivalent to $z_i^q-z_i^{q-1}=1/x^{q+1}$ for $i=1,2$.

As $q+1$ is relatively prime to $Q-1$ (see above), we may replace $1/x^{q+1}$ by $x$. For any $z_1\not\in\{0,1\}$ we get a unique nonzero $x$. Thus, we need to count the pairs $(z_1,z_2)$ where $z_1,z_2\not\in\{0,1\}$, $z_1\ne z_2$, and $z_1^q-z_1^{q-1}=z_2^q-z_2^{q-1}$. Write $z_2=\delta z_1$ (so $\delta\not\in\{0, 1\}$.) This yields (similarly as in the proof of Lemma 1) $z_1=\frac{\delta^{q-1}-1}{(\delta-1)^q}$. This gives a nonzero $z_1$ if and only if $\delta^{q-1}\ne1$. As $\gcd(q-1,Q-1)=p^d-1$, the number of choices for $\delta$ is $Q-1-(p^d-1)=2^n-2^d$.

From the three linear equations (which are already in triangular form) we immediately get the expected numbers $|Y_i|$.

Remark. A slight modification allows to cover the case $p>2$ and $n/d$ odd. Then $\gcd(q+1,Q-1)=2$. So similarly as in the proof of Lemma 2 we now get $z^q-z^{q-1}=1/x^2$. As $q-1$ is even, $z^{q-1}$ is a square. So the condition on $z$ is that $z-1$ is a square. The condition $z\ne0,1$ gives $(Q-3)/2$ or $(Q-1)/2$ possibilities for $z$, depending on whether $-1$ is a square in $\fQ$ or not. For each admissible $z$ there are two possibilities for $x$. Thus $v_2=Q-3$ if $Q\equiv 1\pmod 4$, and $v_2=Q-1$ if $Q\equiv -1\pmod 4$.

Similarly, one can compute $v_3$. The calculation as in the proof of Lemma 3 yields $z_1-1=-\left(\frac{\delta}{\delta-1}\right)^{q-1}$. Recall that $z_1-1$ need to be a square. So $v_3=0$ if $Q\equiv -1\pmod 4$. If, however, $Q\equiv 1\pmod 4$, then this condition on $z_1$ is automatically fulfilled. As there are two possibilities for $x$ for each $z_1$, we obtain $v_3=2(p^n-p^d)$ in this case.

$\endgroup$

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.