Given a prime $p > 2$ and positive integers $x,y \le \frac{p − 1}{2}$, prove that if $x(p − x)y(p − y)$ is a perfect square, then $x = y$.

This question was previously asked without context , and so was closed, but I did start towards a solution.

The primary insight is that the result depends on the unique representation of a prime as a sum of squares.

Obviously if that were not true we could have $m_0^2+n_0^2=m_1^2+n_1^2=p$. Then we could choose $x=m_0^2$ and $y=m_1^2$ as the lesser parts of those two different representations and with all contributing values being square, get a square that disproves the claim.

The other strand of proof that I did not succeed in completing was to eliminate the possibility that the various components might multiply together to give a square. Here are my thoughts on this so far:

Since $p$ prime, $x$ is coprime to $v:=(p-x)$ and $y$ is coprime to $w:=(p-y)$.

Find $g :=\gcd(x,y)$ and $h:=\gcd(v,w)$, and set $(a,b,c,d):=(x/g,y/g,v/h,w/h)$. If one of these terms is $1$ then all are, and $x=y$.

Assume that we have $a,b,c,d>1$. Then the expression of interest is $abcdg^2h^2$, which is square iff $abcd$ is square. The only possible common factors are in the pairs $(a,d)$ and $(b,c)$ so we would need $ad$ square and $bc$ square.

Can anyone either complete my proof or find a better way to demonstrate the truth of the opening claim?


This is the problem from the second stage of last year's Polish Mathematical Olympiad, see here. Below I present the official problem solution, from here.

Suppose the expression is a square, say $k^2$. Applying AM-GM inequality we quickly find $k<p^2/4$. On the other hand, from $x,y<p/2$, we find $k>xy$. Now we rewrite $$0=k^2-xy(p-x)(p-y)=(k-xy)(k+xy)-pxy(p-x-y),$$ from where $p\mid k-xy$ or $p\mid k+xy$.

Suppose first $k-xy=lp>0$. Then $$lp(lp+2xy)=lp(k+xy)=pxy(p-x-y),\\ p(xy-l^2)=xy(2l+x+y).$$ Since $x,y<p$, $p\mid 2l+x+y$. From $lp=k-xy<k<p^2/4$ we get $l<p/4$, and then $2l+x+y<2p/4+p/2+p/2=2p$, hence $2l+x+y=p$. However, then $p(xy-l^2)=pxy$, impossible given $l>0$.

Hence $k+xy=lp$ for some $l$. Like above, we get $p(xy-l^2)=xy(x+y-2l)$. From $lp=k+xy<p^2/4+(p/2)^2=p^2/2$ we get $l<p/2$. Hence $$p>x+y>x+y-2l>-2\cdot p/2=-p,$$ thus $x+y-2l=0$ (since it's divisible by $p$). Thus $p(xy-l^2)=0$, i.e. $xy=l^2$. Hence finally $$(x-y)^2=(x+y)^2-4xy=(2l)^2-4l^2=0$$ giving $x=y$.

  • $\begingroup$ Wow. Nice trackdown skills. And a solution that shows the uniqueness of prime decomposition into squares as a weaker byproduct. $\endgroup$ – Joffan Oct 24 '17 at 18:33

$gcd(x,y)=n_1,gcd(x,p-y)=n_2$, $y$ and $p-y$ coprime, so $gcd(x,y(p-y))=n_1n_2$

Let $x=n_1n_2x',y=n_1y',p-y=n_2y''$, we have $x'(p − x)y'y''$ is square, so $x'$ is square.

In similar fashion, we can get: $x=n_1n_2a^2,y=n_1n_3c^2,p-x=n_3n_4b^2,p-y=n_2n_4d^2$

Now we have:

$p = n_1n_2a^2+n_3n_4b^2=n_1n_3c^2+n_2n_4d^2\tag1$

This can be solved in similar fashion as this answer.

To be complete:





By(1), we have

$(p - n_1n_2a^2)n_1n_2d^2=(p - n_1n_3c^2)n_1n_3b^2$

which implies


From(4), $p$ divides $(n_2ad-n_3bc)$ or $(n_2ad+n_3bc)$. If $p$ divides $(n_2ad-n_3bc)$, then from (2) we get $(n_2ad-n_3bc)=0$, so from (4) $n_3b^2-n_2d^2=0$, so $p-x=p-y$. If $p$ divides $(n_2ad+n_3bc)$,then from (3) we get $(n_1ac-n_4bd)=0$, so $n_1ac=n_4bd$. $a$ and $b,d,n_4$ are coprime, so $a=1$.similarly $b=c=d=1$. Then $n_1=n_4=1$,$x=n_2,y=n_3, p=n_2+n3$, this contradicts $x,y\le\frac{p-1}{2}$.

  • $\begingroup$ How do you deduce that x' is square? $\endgroup$ – Matt Oct 24 '17 at 17:47
  • $\begingroup$ OK, an interesting way to proceed, but I don't quite see how it completes $\endgroup$ – Joffan Oct 24 '17 at 17:54
  • 2
    $\begingroup$ @MattS x' is square because it is coprime with the rest of the term $\endgroup$ – lion Oct 24 '17 at 17:57
  • $\begingroup$ But how is this a sum of two squares? How do you know that $n_1n_2a^2$ is a square? Is there any reason for $n_1n_2$ to be a square? Same for all the other summands. $\endgroup$ – Prasun Biswas Oct 24 '17 at 18:12
  • $\begingroup$ @PrasunBiswas no, I am not saying $n_1n_2$ to be a square. I am just saying if we follow the calculation in that answer by take $p^2$ etc, we can get the result in similar but not identical fashion. $\endgroup$ – lion Oct 24 '17 at 18:16

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.