0
$\begingroup$

Let $p$ be a (large) prime.

How large can a set $C\subset\mathbb F_p$ be given that there is a function $f\colon\mathbb F_p^\times\to\mathbb F_p$ such that for every element $g\in \mathbb F_p$, there is at most one element $c\in C$ with the property that $$ g \notin \{f(z)-cz\colon z\in\mathbb F_p^\times\}\,? $$

In plain English: I want to find a set $C\subset\mathbb F_p$ and a function $f\colon\mathbb F_p^\times\to\mathbb F_p$ such that every element of $\mathbb F_p$ is contained in the images of all $|C|$ functions $z\mapsto f(z)-cz$ with at most one exception; how large can I make $C$?


Greg Martin's answer below along with Will Sawin's comment show that $|C|\approx\sqrt p$ is possible. Can this be further improved? Is $|C|\ge cp$ with an absolute constant $c>0$ possible?

$\endgroup$
0

1 Answer 1

1
$\begingroup$

Extending the idea from responses to your previous post (the case $k=2$), I think one can make $C$ as large as $\lfloor\sqrt[3]{p-1}\rfloor$ at least.

Let $C=\{c_1,\dots,c_k\}$. Without loss of generality we can assume $0\notin C$ (since we can add/subtract a linear function to $f(z)$). By a greedy algorithm, as long as $k^2\ell\le p-1$, one can choose a set $D=\{d_1,\dots,d_\ell\}\subset \Bbb F_p^\times$ such that $CD$ has size $k\ell$ and hence has no repeated products. In particular, if $k\le\sqrt[3]{p-1}$ then we can take $\ell=k$.

Now define $f(z)$ to be identically $0$ except that $f(d_j) = c_jd_j$ for each $1\le j\le k$. We have arranged for the value $0$ to be hit $k$ times (one more than needed, actually). Without the exceptions $f(d_j) = c_jd_j$, each nonzero value would have been hit $k$ times as well; we have removed one "hit" for each of the numbers $-c_id_j$, but since those are distinct in $\Bbb F_p$, every value is still hit at least $k-1$ times (not even counting the fact that those modified values also hit certain values).

$\endgroup$
1
  • 2
    $\begingroup$ It seems that we are allowed to choose $C$. Then it is easy to choose $C,D$ as long as $k\ell \leq p-1$, by choosing $C$ to be an interval of length $k$ in the multiplicative group and $D$ to consist of muliples of $k$. So in fact we win for $k \leq \sqrt{p-1}$. $\endgroup$
    – Will Sawin
    Nov 21, 2017 at 12:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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