5
$\begingroup$

I am interested in proving an upper bound (expressed as a power of $N$, with $N\rightarrow\infty$ ) for the number of elements of the set $$ A_N=\{(k,l,m,n)\in([N,2N]\cap\mathbb{Z})^4: |k^2+l^2-m^2-n^2|\le|k+l-m-n|\}. $$ My intuition is that the inequality defining this set cannot hold too often unless the quadruple exhibits some form of diagonal behavior, which would suggest a bound of the form $N^{2+\epsilon}$, with arbitrarily small $\epsilon>0$. I can't quite formalize this idea, though.

$\endgroup$
5
  • 1
    $\begingroup$ I tried a few different ways to apply the divisor bound (see for example, terrytao.wordpress.com/2008/09/23/the-divisor-bound ), but I wasn't able to do better than $N^{3+\epsilon}$ no matter how I rearranged it. Which does seem to make sense; if you let $(k,l,m)$ vary, I don't see why you couldn't have some finite set values of $n$ that works which are not essentially diagonal. Do you mind expanding on your intuition for essentially diagonal solutions only? $\endgroup$ Commented Sep 22, 2021 at 3:59
  • 1
    $\begingroup$ Empirically, considering $50 \le N < 100$, the exponent seems to be above 2.3 and climbing slowly. The sequence is not in OEIS. $\endgroup$ Commented Sep 22, 2021 at 9:09
  • $\begingroup$ Thanks, for your answers! @AnuragSahay, you may be right that one can't do better than $N^3$. It is more of wishful thinking than intuition on my side; in 2 dimensions it is clear that $|k^2-l^2|\le |k-l|$ can only hold if $k=l$, so I hoped something similar would happen in $4d$. $\endgroup$
    – Tony419
    Commented Sep 22, 2021 at 13:55
  • 1
    $\begingroup$ Well, $1$ is both $2 - 1$ and $2/2$, so I'm not sure you can expect diagonal behavior to persist for more variables. I'd wager that in $2k$ variables, the correct exponent is closer to $2k-1$ than $k$. $\endgroup$ Commented Sep 22, 2021 at 20:11
  • $\begingroup$ I totally agree. $\endgroup$
    – Tony419
    Commented Sep 22, 2021 at 20:26

1 Answer 1

9
$\begingroup$

$|A_N|$ has order of magnitude $N^3$: We use the change of variables $x=k-m$, $y=k+m$, $u=n-l$, $v=n+l$, so that the inequality becomes $|xy-uv|\le |x-u|$. If $x=0$ or $u=0$, there are clearly $\ll N^3$ solutions. Given $x\ge u \ge 1$ and $y$, and assuming $xy\ge uv$, the variable $v$ must satisfy $$ v \in [2N+u,4N-u] \cap \left[ \frac{xy}{u}-\frac{x-u}{u}, \frac{xy}{u}\right]. $$ The other cases are similar. For this intersection to be non-empty, we must have $x/u \asymp 1$, since $y \asymp N$. This implies that the width of the second interval is $O(1)$. Thus there are $O(1)$ choices for $v$, which means that the number of solutions is $O(N^3)$.

To show that there are actually $\gg N^3$ solutions, choose $$ x \in [0.5 N, 0.55 N], \ y\in [2.6N, 2.8N], \ y\equiv x \bmod 2, \ u\in [0.45N,0.5N], \ u\equiv 0 \bmod 2. $$ Then the second interval for $v$ is contained in the first. We need to count the number of $v$ in the second interval such that $v\equiv u \bmod 2$. The number of such $v$ is $$ \ge \left\lfloor \frac{xy}{2u} \right\rfloor - \left\lfloor \frac{xy}{2u}-\frac{x-u}{2u} \right\rfloor = \frac{x-u}{2u} -\psi\left( \frac{xy}{2u} \right) + \psi\left( \frac{xy}{2u} -\frac{x-u}{2u}\right) , $$ where $\psi(t)=t-\lfloor t \rfloor -1/2$. Summing the $\psi$'s over $u$ results in $o(N)$, so their total contribution (after also summing over $x$ and $y$) is $o(N^3)$. Summing the term $(x-u)/(2u)$ over $u,x,y$ contributes $\asymp N^3$.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .