2
$\begingroup$

Let $$\displaystyle f(x_1,x_2,x_3) = a_1 x_1^2 + a_2 x_2^2 + a_3 x_3^2,$$ $$\displaystyle g(x_1, x_2, x_3) = b_1 x_1^2 + b_2 x_2^2 + b_3 x_3^2$$

be two integral ternary quadratic forms with $f$ positive definite. Let $p$ be an odd prime and suppose that $f$ and $g$ are not equivalent modulo $p$, i.e., there does not exist an integer $k$ such that

$$\displaystyle f(x_1, x_2, x_3) \equiv k g(x_1, x_2, x_3) \pmod{p}$$

for all $x_1, x_2, x_3 \in \{0, 1, \cdots, p-1\}$. We can further assume that $f,g$ are irreducible modulo $p$. Let $B$ be a positive number, and define

$$\displaystyle N_{f,g,p}(B) = \#\{(x,y,z) \in \mathbb{Z}^3 : f(x,y,z) \equiv g(x,y,z) \equiv 0 \pmod{p}, f(x,y,z) \leq B \}.$$

Put $N_f(B) = \#\{(x,y,z) \in \mathbb{Z}^3 : f(x,y,z) \leq B\}$. Is it true that for some $l \geq 2$ we have

$$\displaystyle \lim_{B \rightarrow \infty} \frac{N_{f,g,p}(B)}{N_f(B)} = \frac{1}{p^l}?$$

If so, what error term can be proved for the expression

$$\displaystyle N_{f,g,p}(B) = \frac{1}{p^l} N_f(B) + o(N_f(B))?$$

$\endgroup$

1 Answer 1

2
$\begingroup$

Consider the example $f=x_1^2+x_2^2+x_3^2$, $g=x_1^2$. For any prime $p\equiv 3\mod 4$, we have $f\equiv g\equiv 0\mod p\Leftrightarrow x_1\equiv x_2\equiv x_3\equiv 0\mod p$. So for this example $$ \lim_{B \to\infty} \frac{N_{f,g,p}(B)}{N_f(B)} = \frac{1}{p^3}. $$


Edit: Write $\overline{f}$ and $\overline{g}$ for the reductions of $f$ and $g$ modulo $p$. The value of the limit is $$ \frac{\#\{x,y,z\in\mathbb{F}_p:\overline{f}(x_1,x_2,x_3)=\overline{g}(x_1,x_2,x_3)=0\}}{p^3}. $$ Define curves $X=\mathbb{V}(\overline{f})$ and $Y=\mathbb{V}(\overline{g})\subset\mathbb{P}^2_{\mathbb{F}_p}$. These are irreducible plane curves of degree $2$, and we can also compute the value of the limit as $$ \frac{1+(p-1)\cdot\#(X\cap Y)(\mathbb{F}_p)}{p^2}. $$ Bezout's theorem implies there are four points in the intersection (over $\overline{\mathbb{F}}_p$, counting multiplicity), and it is possible to have $0$, $1$, $2$, $3$, or $4$ distinct points over $\mathbb{F}_p$. The values of the limit can be $$ \frac{1}{p^3},\hspace{8mm} \frac{1}{p^2},\hspace{8mm} \frac{2}{p^2}-\frac{1}{p^3},\hspace{8mm} \frac{3}{p^2}-\frac{2}{p^3}. $$

We should be able to get an error term $$ N_{f,g,p}(B)=L\cdot N_f(B)+O\left(\frac{N_f(B)}{\sqrt{B}}\right), $$ where $L$ is the value of the limit. This is closely related to counting lattice points inside an ellipsoid. I don't know much about this, but a better bound might be possible.

$\endgroup$
4
  • $\begingroup$ Thanks Julian, I hadn't considered the possibility of the exponent exceeding 2. In any case any exponent at least 2 with an acceptable error term is fine for what I am trying to do, so I will edit the question to reflect this. Thanks! $\endgroup$ May 23, 2016 at 23:49
  • $\begingroup$ Two comments. First, you probably mean Bezout's theorem, not Bertini's theorem. Second, if there are four intersection points over $\overline{\mathbb F}_p$, then it is not possible for exactly 3 of them to be defined over $\mathbb F_p$. (The four points form a Galois invariant set, so if three elements are fixed by Galois, then so is the fourth.) $\endgroup$ May 24, 2016 at 2:16
  • $\begingroup$ @JoeSilverman Thanks, I definitely meant Bezout's theorem. There could be two intersection points with multiplicity $1$ and another with multiplicity $2$. $\endgroup$ May 24, 2016 at 2:19
  • $\begingroup$ @JulianRosen That's true, but it's worth pointing out that unless the multiplicty 2 intersection happens over $\mathbb C$, then it will happen mod $p$ for only finitely many primes, so won't happen for most $p$. So the various possibilities for the limit depend on the geometry of the complex solutions (and more precisely, on the field extension of $\mathbb Q$ generated by the $\overline{\mathbb Q}$ intersection points, via the splitting of $p$ in this field). $\endgroup$ May 24, 2016 at 11:46

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.