3
$\begingroup$

Given a positive definite integral binary quadratic form $f$, denote by $r_f(n)=\#\{(x,y)\in \mathbb{Z}^2~|~f(x,y)=n\}$, how can we obtain estimates for $$\sum_{~~~~~n\leq x\\ n\equiv k~\mathrm{mod}~q}r_f^2(n),$$ where $(k,q)=1$.

I'm aware of the results of Tolev here, but he treats the special case $f(x,y)=x^2+y^2$ and studies $\sum r_f(n)$ instead. Is there away to do this in more generality? Maybe the work of Blomer and Granville might help, as I need an explicit dependence on the form $f$ in the main term.

I know $$\sum_{~~~~~n\leq x\\ n\equiv k~\mathrm{mod}~q}r_f^2(n)=\frac{1}{q}\sum_{n\leq x} r_f^2(n)+\frac{1}{q}\sum_{1\leq a\leq q-1}e(-ak/q)\sum_{n\leq x} r_f^2(n)e(an/q).$$ The first term can be estimated by a result due to Muller down here, but how do I show that the error term coming from the other additive characters is small enough? say of order $\ll x^{3/4}$? I believe we have to examine the Dirchlet function $$L(r_f^2,a,s)=\sum \frac{r_f^2(n)e(an/q)}{n^s}$$ where $a\in \mathbb{Z}/q\mathbb{Z}$.

Any help would be appreciated.

Edit: I found this paper of Muller (see theorem 6.1) where he studies $\sum_{n\leq x}r_f^2(n)$ for any quadratic form $f$ ( doesn't have to be necessarily of fundamental discriminant), but his approach uses the mean-squared estimate which eliminates the usage of the characters.

$\endgroup$
12
  • $\begingroup$ How precise an estimate do you want? The main term is roughly of size $n/q$ and a very crude estimate could get an error term of size $\sqrt{n} q$. Is that enough or do you want better dependence on $n$ and/or $q$? $\endgroup$ Commented Mar 6 at 17:08
  • $\begingroup$ @WillSawin Yes! The estimates you mentioned work well with me. Could you kindly provide some details? I know the main term should be roughly $c_f\frac{n\log n}{\varphi(q)}$ but I wanted to know how we can get an error term as such? Thanks a lot! $\endgroup$ Commented Mar 6 at 20:21
  • $\begingroup$ @WillSawin Could you kindly provide a rough sketch? I'm not sure how one can obtain such an estimate as I'm not very much familiar with the function $L(s,\chi)=\sum_{n=1}^\infty \chi(n)r_f^2(n)n^{-s}$ for a Dirchlet character $\chi$ $\endgroup$ Commented Mar 22 at 17:19
  • $\begingroup$ My very crude estimate was to answer the original version with $r_f(n)$, not the current $r_f^2(n)$ version, by counting lattice points inside an ellipse. The method will not work at all for the $r_f^2(n)$ version. $\endgroup$ Commented Mar 22 at 17:48
  • 1
    $\begingroup$ The paper by Radu Toma that I linked for you (twice), extends the work of Blomer-Granville. In particular, it deals with non-fundamental discriminants: doi.org/10.1016/j.jnt.2020.03.005 Also, please don't ask questions from me directly. We come here to have fun, and being pressured is not fun. Whatever you want to know, ask in the main post, and we will see if we can contribute. At any rate, perhaps the best course of action for you would be to study the 4 works linked above (that is: read them), and then ask their authors if you still have questions closely related to those works. $\endgroup$ Commented Mar 23 at 0:11

0

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.