8
$\begingroup$

This is a toy version of a problem I have posted recently.

Imagine playing the following game. You choose a polynomial $B$ over a finite field $\mathbb F_p$, of degree $\deg B\le p-1$ (where $p$ is a large prime). I can modify your polynomial any way I wish, except that I am not allowed to add or change monomials of degree exceeding $0.9p$. My goal is to make the resulting polynomial split completely into linear factors; if I fail to do so, you win.

  • What polynomial $B$ will you start the game with to win?
  • Is there a comprehensible classification of all winning (for the first player) polynomials?

As an illustration, if I can change only monomials of degree smaller than $(p-10)/4$, then you win by choosing $B(x):=x^{p-1}-nx^{p-3}$, where $n$ is a quadratic non-residue modulo $p$. To see this, suppose that $R$ is a polynomial of degree $d$ such that $Q(x):=x^{p-1}-nx^{p-3}+R(x)$ splits into linear factors, and show that $d\ge(p-10)/4$. Writing $D(x):=(Q(x),x^{p-1}-1)$, we have \begin{align*} D(x) &= (nx^{p-3}-R(x)+1,x^{p-1}-1) \\ &= (nx^{p-1}-x^2R(x)+x^2,x^{p-1}-1) \\ &= (x^2R(x)-x^2-n,x^{p-1}-1), \end{align*} so that $\deg D\le d+2$. On the other hand, since $Q(x)$ splits into linear factors, it is a divisor of the polynomial $$ (Q(x),x^p-x)^3(x^3Q(x))''' = (Q(x),x^p-x)^3(x^3R(x))'''. $$ Since the degree of $(Q(x),x^p-x)$ does not exceed $\deg D+1$, we conclude that either $$ p-1 = \deg Q \le 3(\deg D+1) + \deg R \le 4d+9, $$ or $(x^3R(x))'''=0$. In the former case we are done, in the latter case $R=0$ and $Q(x)=x^{p-1}-nx^{p-3}$, contrary to the choice of $n$ and the assumption that $Q$ splists into linear factors.

It should be possible to push this argument to replace $(p-10)/4$ with $p/2+O(1)$, but I cannot see how to get all the way up to $0.9p$.

$\endgroup$
7
  • $\begingroup$ Working symbolically write $B=\prod_i X_i x + Y_i$. After equating coefficients you have $2 \deg(B)$ unknowns and $\deg(B) - 0.9p$ equations. If the system is overdetermined likely it won't have solutions. $\endgroup$
    – joro
    Commented May 7, 2019 at 8:59
  • $\begingroup$ @joro: The system is actually greatly underdetermined, as the number of variables $2\deg B$ is much larger than the number of equations $\deg B-0.9p$ - or have you meant anything else? $\endgroup$
    – Seva
    Commented May 7, 2019 at 9:26
  • $\begingroup$ OK, I didn't examine it carefully. In general greatly underdetermined system will have solution over F_p, unless there is some hidden reason. $\endgroup$
    – joro
    Commented May 7, 2019 at 9:30
  • 1
    $\begingroup$ @joro: Exactly - my question is about these possible hidden reasons! $\endgroup$
    – Seva
    Commented May 7, 2019 at 9:33
  • 1
    $\begingroup$ One "hidden reason" is symmetry: you have only about $4^p$ ways to choose your roots because the permutations do not matter and I have about $p^{p/10}$ ways to choose my top coefficients, so I wouldn't even bother to think if $p$ is large, just roll a $p$-sided die. $\endgroup$
    – fedja
    Commented May 8, 2019 at 18:46

0

You must log in to answer this question.

Browse other questions tagged .