8
$\begingroup$

For given integers $m$,$n$, what bounds are known on the number of positive integers $x$ such that $m-x^2$ and $n-x^2$ are both perfect squares? In particular, is the number of such $x$ bounded above by a constant independent of $m,n$?

A search for $m,n \le 5000$ gives three pairs for which there are 4 such x: (1370,2210),(2210,3050), and (3485,3965).

$\endgroup$
4
  • 8
    $\begingroup$ The number of $x$ will almost certainly not be bounded by a constant independent of $m,n$. Here's why. The equations you're trying to solve are two quadrics; the intersection of two quadrics is a genus 1 curve and you have a lot of genus 1 curves to choose from. Find $m_0$ and $n_0$ such that this genus 1 curve is an elliptic curve of positive rank. Then you have infinitely many rational $x$ such that $m_0-x^2$ and $n_0-x^2$ are perfect rational squares. The denominators will be growing -- but now choose $m=m_0D^2$ and $n=n_0D^2$ for some large $D$ to get as many solutions as you want. $\endgroup$ Apr 25, 2012 at 6:50
  • $\begingroup$ That's a comment not an answer, because you now need to find the appropriate $m_0$ and $n_0$. It wouldn't surprise me if you'd already found them in your computer search but I don't have time to check -- it's the school run. More interesting would be restricting to the case $m$ and $n$ coprime. Then things get much thorny. Here's a related question: mathoverflow.net/questions/50661/… which remains unanswered at this point and is probably open... $\endgroup$ Apr 25, 2012 at 6:53
  • 1
    $\begingroup$ Are you saying that for fixed $m_0,n_0$ the set of $(x,y,z)$ with $m_0=x^2+y^2$ and $n_0=x^2+z^2$ is an elliptic curve? Tell me more. $\endgroup$ Apr 27, 2012 at 16:56
  • $\begingroup$ Aaron -- the generic intersection of two quadrics in projective 3-space is a genus 1 curve. This is a standard cohomological calculation -- but if you want to see an explicit low-level argument then the birational transformation that does the job is in, for example, Cassels' little blue book on elliptic curves. $\endgroup$ Apr 28, 2012 at 9:40

2 Answers 2

5
$\begingroup$

In comments Aaron asked about an example of Kevin's construction with $10$ points.

In Kevin's comment the rational points come from the group law on the elliptic curve and $D$ is the lcm of the denominators.

Here is magma online code and example with $10$ points -- as Kevin wrote this way you get as many solutions as you want.

Starting with the OP's $m,n=1370,2210$ got a Weierstrass model of EC.

Found $4$ generators with mwrank (could have used some of OP's points instead of finding generators).

Multiples of the $4$ generators gave additional solutions to the OP and here is the result:

D= 12390849713183581790836556709545874255316431121037842112055747871712302486846209203045
m= 210340424562141262047595910887493791274649532004925386992592745031243269297527125123770927661074947583428727448136984975928253274592074678864394207689104340429534031062674250
n= 339308276118490648996486834351358597603631726810865040331116763882516514706229887973382299365675645371808385153564041457519299077991594919919935181746657366678299422371175250
x1= 12390849713183581790836556709545874255316431121037842112055747871712302486846209203045
x2= 284989543403222381189240804319555107872277915783870368577282201049382957197462811670035
x3= 458461439387792526260952598253197347446707951478400158146062671253355192013309740512665
x4= 359334641682323871934260144576830353404176502510097421249616688279656772118540066888305
x5= 37114042644837641401487345470535116987974416718966693064558863256140865315458375412615
x6= 61662771334048589214986191564717597484363845941296621783693700780051335795784841331249
x7= 136028531722798363505361591190162668220641503334832494123018223476468013307956416186455
x8= 244444823900679257002213529250900630227602064446780336065278693335401699985373414615885
x9= 301450648169146711879386180421262383173184752443718928831863273736903739700469518290993
x10= 437274569420552490793563721414855158314534791058269868589191046763361083229670113418305

Magma online code

m:=1370;
n:=2210;
aa<x,y,z>:=AffineSpace(Rationals(),3);
C:=Curve(aa,[m-x^2-y^2,n-x^2-z^2]);
P:=C!([1,37,47]);
pc:=ProjectiveClosure(C);
E,m1:=EllipticCurve(pc,pc!(P));
m2:=Inverse(m1);
aInvariants(E);
Ep:=E!([-323231734744697104/27633477663066497041,585299700649024 /27633477663066497041]);
m2(Ep);
m2(2*Ep);
$\endgroup$
1
  • 1
    $\begingroup$ Thanks for doing the dirty work joro! So this is a proof that there is no bound independent of $m,n$. $\endgroup$ Apr 28, 2012 at 9:42
4
$\begingroup$

(revised) I suspected that there is no bound, but now I am not as sure . In an example $m,n$ with many such $x$ it will have to be the case that $m$ can be written in many ways as a sum of two squares (and the same for $n$.) Also, $m^2-n^2$ will have to be expressed in many ways as a difference of $2$ squares. We know that the number of ways to do these three things can be derived from the prime factorizations of $m,n$ and $n-m$. So this tells us where to look for possible examples. The number of positive solutions of $m=x^2+y^2$ is $2^{k-1}$ when $m$ is a product of $k$ distinct primes all congruent to $1\mod{4}$ (This gives $2^k$ possible $x$. Also, the general formula is not much harder). Consider the $126$ products of $5$ out of the $9$ primes $5, 13, 17, 29, 37, 41, 53, 61, 73.$ It turns out that $m=2401165, n=4254445$ allow for 7 such $x.$

Going out to $101$ and using products of $4$ primes one finds that there are $5$ choices of $x$ in the case of the relatively prime integers $m=1662965=5\cdot37\cdot89\cdot101$ and $n=3068117=13\cdot53\cdot61\cdot73.$ Note that $n-m=2^5\cdot3^2\cdot7\cdot17\cdot41$ which allows $48$ ways to have $n-m=y^2-z^2.$

Here is why I wonder if we can be confident of getting even larger numbers: The points $(x,y)$ with $x^2+y^2=m$ are the points of $\mathbb{Z}^2$ which are on the circle of radius $m$ (and are in the first quadrant.) So we are looking for two circle of integer radii $m \lt n$, centered at the origin, and having a large number of pairs lattice points one on the larger circle and the other vertically below and on the smaller circle. To have a large number of lattice points on the circle of radius $m$ will require $m$ to have many factors and hence be large. As $m$ grows the number of lattice points can grow but the circumference grows more rapidly leading to a sparser distributin of points. $m=13\cdot17\cdot29$ allows for $8$ possible $x$ values along a quarter circle of circumference $m\pi/2 \approx 10067.$ Now $m=5\cdot13\cdot17\cdot29$ would allow for twice as many points, but along a quarter circle with $5$ times the circumference (and this is the most favorable example.)

$\endgroup$
8
  • 1
    $\begingroup$ The first comment explains how to "get as many solutions as you want". Doesn't your last revision contradict it? $\endgroup$
    – joro
    Apr 27, 2012 at 15:01
  • $\begingroup$ The first comment tells you how to get as any solutions as you want for $m=s^2+t^2$ and for $n=u^2+v^2$ but not how to get lots of solutions of $m=x^2+t^2$ and $n=x^2+v^2$. Still using that I looked at $m_1 \cdots m_{126}$ each with $32$ possible values for $x$ and did find one pair with $7$ shared $x$ values . In my first version I suggested that there was no limit how high you could go. Later I was less sure. I didn't figure out Kevin's comment yet. $\endgroup$ Apr 27, 2012 at 16:48
  • 1
    $\begingroup$ I meant Kevin's comment. $m_0=x^2+y^2$ and $n_0=x^2+z^2$ for fixed $m_0,n_0$ is genus 1 curve. One can check in a CAS like sage or magma. If it has a rational point it is birationally equivalent to a Weierstrass model of elliptic curve. The first equation is genus 0 and it can parametrized over $\mathbb{Q}$ or $\bar{\mathbb{Q}}$. Plugging the parametrization for $x$ gives quartic = square which is only 1 equation and a quartic model of an EC. If the EC has a point of infinite order using the group law one can find many rational points. To make the points integral, scale by $D$... $\endgroup$
    – joro
    Apr 28, 2012 at 5:40
  • 1
    $\begingroup$ ...which is the lcm of the denominators. This changes $m_0,n_0$ though $\endgroup$
    – joro
    Apr 28, 2012 at 5:42
  • $\begingroup$ That may well be. I can believe that there may be an example with as many as $10$ common $x$ values, but I'd like to see one. I found $7$ very easily. Looking at most of the $2^{15}$ products of some or all of $2, 5, 13, 17, 29, 37, 41, 53, 61, 73, 89, 97, 101, 109, 113$ turned up a few more cases of $7$ but never $8$ or more. $\endgroup$ Apr 28, 2012 at 7:31

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.