Suppose that $q$ is a prime power and $\xi, \eta\in \mathbb{F}_q$ are nonzero. A computer calculation for $q<70$ suggests that the number $N$ of $4$-tuples $(a,b,c,d)\in\mathbb{F}_q^{4}$ satisfying $(ac-\xi bd)^2-(a^2-\xi b^2+1)(c^2-\xi d^2-\eta)=0$ is $q^3-q$.

Question. Is there some theory, or nice method, for computing $N$?

It may help, when $\xi$ is a non-square, to observe that the norm of $a+b\sqrt{\xi}\in\mathbb{F}_{q^2}$ is $a^2-\xi b^2$. The non-homogeneous polynomial above seems hard to me, but I am not an expert in such matters. I am not surprised that $N$ is a cubic in $q$.

  • $\begingroup$ Why are you interested in that polynomial in particular? $\endgroup$ – Daniel Hast Jun 15 at 13:06
  • 1
    $\begingroup$ @Danial Hast. This polynomial arose from a geometric problem regarding quadratic forms. The size of its fibres seem to be piecewise polynomial. Indeed, this is the first of a series of polynomials that arise essentially from the determinant of a Gram matrix. $\endgroup$ – Glasby Jun 15 at 14:16

I will concentrate on the case of odd $q$ and $\xi,-\eta$ being squares, but the solution should be extendable to the remaining cases.

It is convenient to use the language of characters sums in order to compute $N$. To reduce your point counting problem to a character sum problem, the following observation is useful: $$\#\{ x \in \mathbb{F}_q : ax^2+bx+c=0\} = 1 + \chi(b^2-4ac),$$ where $\chi$ is the unique non-trivial quadratic character of $\mathbb{F}_q^{\times}$, extended to give $0$ on $0$, and $a \neq 0$. This observation is proved by completing the square.

We will need the following well-known formula: $$(\sim) \, \sum_{x \in \mathbb{F}_q} \chi(x^2+t) = -1$$ for $t\neq 0$. This formula encodes the fact that the number of points on the genus-0 curve $y^2=x^2+t$ is $q-1$. In your case we are definitely lucky, as the computation of $N$ reduces to point counting on genus-0 curves, while for higher genus these point counts are not polynomial in general.

  1. (Cosmetic step) Replacing $(b,d)$ by $(b/\sqrt{\xi},d/\sqrt{\xi})$ we see that we may assume that $\xi = 1$. Further replacing $(c,d)$ by $(c\sqrt{-\eta},d\sqrt{-\eta})$ we see that we may assume that $-\eta = 1$ as well.

  2. Let us expand your defining hypersurface and express it as a quadratic polynomial in $a$:

$$(*) \, a^2(d^2-1) + a(-2bcd) + b^2c^2+b^2-c^2+d^2-1 = 0.$$

  1. The case $d^2 =1$ simplifies to $a(2bcd) = b^2c^2+b^2-c^2$. If $bc \neq 0$, this determines $a$ uniquely. If $bc=0$, we must have $b=c=0$, and $a$ can be arbitrary. This case yields $2((q-1)^2 + q)$ solutions.

  2. Let us assume $d^2 \neq 1$. The discriminant of $(*)$ factorizes as $4(c^2+1-d^2)(b^2+d^2-1)$, which is a lucky coincidence, and possibly the heart the matter. Hence we see that given $b,c,d$ contribute to $(*)$ $$1+\chi( (c^2+1-d^2)(b^2 +d^2-1) )$$ solutions. Summarizing, we have $$(**)\, N=2(q^2-q+1) + q^2(q-2) + \sum_{d^2 \neq 1, \, b,c,d \in \mathbb{F}_q} \chi( (c^2+1-d^2)(b^2 +d^2-1) ).$$

  3. By $(\sim)$, $$\sum_{b \in \mathbb{F}_q} \chi( (c^2+1-d^2)(b^2 +d^2-1) ) = \chi(c^2+1-d^2) \sum_{b \in \mathbb{F}_q} \chi(b^2 +d^2-1 ) = -\chi(c^2+1-d^2)$$ when $d^2 \neq 1$, and so $$\sum_{d^2 \neq 1, \, b,c,d \in \mathbb{F}_q} \chi( (c^2+1-d^2)(b^2 +d^2-1) )= -\sum_{d^2 \neq 1, \, c,d \in \mathbb{F}_q} \chi(c^2+1-d^2),$$ and applying $(\sim)$ once more this becomes $$-\sum_{d^2 \neq 1, \, c,d \in \mathbb{F}_q} \chi(c^2+1-d^2) = \sum_{d^2 \neq 1} 1 = q-2.$$ Plugging this character sum evaluation in $(**)$, $N=q^3-q$ is obtained, confirming your empirical observation.

For general $\xi$ and $\eta$, a very similar argument will work, because the discriminant of your defining equation (considered as a quadratic polynomial in $a$) still factorizes nicely, specifically it is $$4(c^2 - \eta-d^2\xi) (-b^2\xi\eta + d^2 \xi + \eta).$$

Even $q$ is easier. The defining equation can now be written as $$(a+1)^2 (\xi d^2 +\eta) = \xi b^2 (c^2+\eta) + c^2.$$ Recall that in $\mathbb{F}_q$ with even $q$, $x \mapsto x^2$ is a field automorphism.

  • If $\xi d^2+\eta \neq 0$ (happens for all but a unique $d$), $a+1$ is uniquely determined by $b,c$, giving $(q-1)q^2$ solutions.
  • If $\xi d^2+\eta = 0$, $d$ is determined uniquely, $a$ is arbitrary and it remains to count solutions $(b,c)$ to $c^2(\xi b^2+1) = \eta \xi b^2$. Specifying $b$ determines $c$ uniquely, unless $\xi b^2 + 1=0$ (happens for a unique $b$), in which case there are no solutions. So this case contributes $q(q-1)$ solutions.

All in all, $N = (q-1)q^2+q(q-1) = q^3-q$ for even $q$ as well.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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