Let $n$ be a positive integer. We will call a divisor $d(<\sqrt{n})$ of $n$ special if there exists no perfect squares between $d$ and $\frac{n}{d}$. Prove that $n$ can have at-most one special divisor.

My progress: I boiled down the problem to the following: Suppose $k^2\le a,b,c,d\le (k+1)^2$, then $ab=cd\implies \{a,b\}=\{c,d\}$. But I can't seem to prove this.

Arriving here isn't difficult so I am omitting any further details(one more reason being I am not sure if I am on the correct path).

  • 1
    $\begingroup$ This MO website is for questions of math research. It's not clear to me that this question qualifies. How did you come across it? (Also, you are overloading the symbol $n$.) $\endgroup$ – Gerry Myerson Mar 24 at 11:38
  • $\begingroup$ @Gerry Myerson My math professor says that this problem came up while he was trying to solve some open problem. I didn't bother asking him the details though. I would be very grateful if you can help me solve this problem. Or even any good idea would do. $\endgroup$ – user154024 Mar 24 at 12:07
  • 2
    $\begingroup$ Simulposted to m.se, math.stackexchange.com/questions/3592817/… without notice to either site. That's an abuse. $\endgroup$ – Gerry Myerson Mar 24 at 12:12
  • $\begingroup$ Doesn't $d$ being special imply $\sqrt{n/d}-\sqrt{d}<1$? $\endgroup$ – Sylvain JULIEN Mar 24 at 12:13

This follows from a result I have asked about a few years ago, namely:

For any $n\in\Bbb N$ there is at most one divisor of $n$ in the interval $[\sqrt{n},\sqrt{n}+\sqrt[4]{n}]$.

I claim that if $d<\sqrt{n}$ is special, then $n/d\in[\sqrt{n},\sqrt{n}+\sqrt[4]{n}]$. Indeed, if this were not the case, we would have $n/d>\sqrt{n}+\sqrt[4]{n}$ and $$d<\frac{n}{\sqrt{n}+\sqrt[4]{n}}=\frac{n-\sqrt{n}}{\sqrt{n}+\sqrt[4]{n}}+\frac{\sqrt{n}}{\sqrt{n}+\sqrt[4]{n}}<\sqrt{n}-\sqrt[4]{n}+1.$$ Letting $x=\sqrt[4]{n}$, it remains to show that for any $x>2$ there is a perfect square between $x^2-x+1$ and $x^2+x$. By monotonicity, it is enough to show that if $x^2-x+1=k^2$ for some $k\in\mathbb N$, then $x^2+x\geq (k+1)^2$. The first equality resolves to $x=\frac{1}{2}(\sqrt{4k^2-3}+1)$, and we have $$x^2+x=x^2-x+1+2x-1=k^2+\sqrt{4k^2-3}<k^2+2k<(k+1)^2.$$ I'm sure the last part of the argument can be carried out more cleanly, but it checks out.

  • 1
    $\begingroup$ Thanks for answering this question, so that I can stop racking my brain on it! $\endgroup$ – Sylvain JULIEN Mar 24 at 14:23

Here is an elementary approach. We want to rule out finding distinct $a,b,c,d$ in short interval (specifically $[n^2+1,n^2+2n]$) with $ad=bc.$ We may assume that $a$ is the smallest and that $b<c$. Then $a<b<c<d.$ I will show that $a+2\sqrt{a}+1\leq d$ so that if $n^2 \leq a$ then $(n+1)^2 \leq d.$

Claim: There are integers $u<v$ and $x<y$ with $$a,b,c,d=ux,uy,vx,vy.$$

Proof: Let $u=\gcd(a,b)$ so $a=ux$ and $b=uy$ with $\gcd(x,y)=1.$ Then $uxd=uyc$ so $xd=yc$ and thus (since $x$ and $y$ are co-prime) there is $v$ with $c=xv$ and $d=yv.$

ASIDE We might as well use $v=u+1$ and $y=x+1$ since $$a,b',c',d'=ux,u(x+1),(u+1)x,(u+1)(x+1)$$ give $ad'=b'c'$ with $a <b',c',d'\leq d.$ I'll comment a bit more about this at the end.

So we want to show

if $ad=bc$ with $n^2<a<b<c<d$ then $d>(n+1)^2.$

From the claim above, $n^2<a=ux$ and $d\geq (u+1)(x+1)=ux+u+x+1.$ But given $ux=a>n^2,$ we know $u+x\geq 2\sqrt{a}>2n.$ Thus $ux+u+x+1>n^2+2n+1$ as desired.

Consider this problem: Given $a$, find $a<b<c<d$ with $d$ minimal such that $ad=bc.$ The work above shows that the solution is to have $$a,b,c,d=ux,u(x+1),(u+1)x,(u+1)(x+1)$$ with $|u-x|$ minimal and that $d>a+2\sqrt{a}+1.$ If we allow $b=c$ then $$n^2\cdot (n+1)^2=(n^2+n)\cdot (n^2+n).$$ If we want $b < c$ then $(n^2-n)\cdot(n^2+n)=(n^2-1)\cdot(n^2).$ With $d-a\approx 2\sqrt{a}.$

If $ad=bc$ then $abcd$ is a perfect square. However this property is weaker and there are solutions such as $$a,b,c,d=2\cdot 120^2,3\cdot 98^2,30\cdot 31^2,5\cdot 76^2=$$ $$28800, 28812, 28830, 28880$$ with $d<a+\frac12\sqrt{a}$ and all four factors between $28561=169^2$ and $28900=170^2.$


Only a partial answer for now. Let $d_{-1}$ the largest divisor of $n$ below $\sqrt{n}$, $d_{-(k+1)}$ the largest divisor of $n$ below $d_{-k}$ and $d_{k}:=n/d_{-k}$. Let $r_{k}:=d_{k}-\sqrt{n}$ and $l_{k}:=\sqrt{n}-d_{-k}$.

Define the $k$-th "square root divisor span" of $n$ as $s_{k}(n):=\sqrt{d_{k}}-\sqrt{d_{-k}}$. The sequence $(s_{k}(n))_k$ is strictly increasing, and its general term equals $\sqrt{\sqrt{n}+r_{k}}-\sqrt{\sqrt{n}-l_{k}}$ which is greater or equal than $\sqrt{\sqrt{n}+k}-\sqrt{\sqrt{n}-k}$.

I think the condition in my comment, namely $s_{k}(n)<1$, is fulfilled only when $\max(l_{k},r_{k})<n^{1/4}$, so for $m(n)=O(1)$ values of $k$, with $\displaystyle{\lim_{n\to\infty}m(n)=0}$.


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.