2
$\begingroup$

Let $f(x,y)=0$ be irreducible elliptic curve over the rationals.

Are there $f$ for which:

Both $x,y$ are arbitrary large powers infinitely often, i.e. infinitely many rational points $(u^k,v^m)$ with both $k,m$ arbitrary large?


For $x,y$ squares (asked in previous revision) this is possible. Take $f(x,y)=x^{6} - 2 x^{3} y^{3} + y^{6} - 72 x^{3} - 72 y^{3} + 1296$.

$f(x^2,y^2)$ is divisible by $x^3 + y^3 - 6$ which is genus $1$ of positive rank.

For large $k,m$ the n-conjecture implies the number of monomials can't be too small.

$\endgroup$
8
  • $\begingroup$ Squares of what? $x$ is a rational function on the curve with divisors of zeroes (or poles) of degree 3, it cannot be a square of a rational function. $\endgroup$
    – abx
    Jan 3, 2015 at 13:04
  • $\begingroup$ @abx I mean squares of rationals $x=u^2,y=v^2$. I don't ask about rational function at all. $\endgroup$
    – joro
    Jan 3, 2015 at 13:06
  • 1
    $\begingroup$ You are essentially looking for rational points on a curve $f(x^2,y)=0$ or $f(x^2,y^2)=0$ depending on your condition. So you have to work out if the curve has genus 1 or > 1, which depends on whether the obvious cover is unramified or not. This is not an MO question. $\endgroup$ Jan 3, 2015 at 13:15
  • $\begingroup$ @FelipeVoloch If you say so. The problem is I don't know $f$ and there are many of them. $\endgroup$
    – joro
    Jan 3, 2015 at 13:20
  • 1
    $\begingroup$ Usually the cover will be ramified, so there will be only finitely many solutions. For example, if you have a Weierstrass model $y^2 = x^3 + ax + b$, then $y^2 = x^6 + ax^2 + b$ defines a curve of genus 2 unless $b = 0$. $\endgroup$ Jan 3, 2015 at 13:29

2 Answers 2

9
+50
$\begingroup$

The method joro suggests works with little change for any $(k,m)$.

Fix an elliptic curve $E: P(X,Y)=1$ of positive rank. Some simple examples are $F(X,Y) = Y^2 - X^3 + 2X$ and $F(X,Y) = Y^2 - X^3 - 2$, each with generator $(X,Y)=(-1,1)$.

Let $u$ and $v$ be "random" rational functions on $E$ that generate the function field (almost any choice will work, see below). Then $u$ and $v$ satisfy a minimal equation $g(u,v)=0$ which defines an algebraic curve birational to $E$ and thus has infinitely many rational solutions.

Set $x = u^k$ and $y = v^m$. As long as these functions, too, generate the function field of $E$, the equation $f(x,y)=0$ that they satisfy gives an irreducible elliptic curve with infinitely many rational points at which $x$ is a $k$-th power and $y$ is an $m$-th power. To check this it is enough to verify that the map $(X,Y) \mapsto (x,y) = (u^k,v^m)$ is generically injective on $E$, and for that it's enough to find one point $(u,v)$ on $E$ that's not a zero or pole of either $u$ or $v$ and such that there's no other $(u',v') \in E$ with ${u'}^k=u^k$ and ${v'}^m = v^m$. For some $k$ and $m$ we can even take $(u,v) = (X,Y)$, and in general a pair of "random" translates $(X+u_0, Y+v_0)$ should suffice.

The polynomial equation $f(x,y) = 0$ can be computed by eliminating $u,v$ from the system $g(u,v) = u^k-x = v^m-y = 0$: take the resultant of the first two equations with respect to $u$ to obtain a relation between $v$ and $x$, and then take the resultant w.r.t. $v$ of that relation and $v^m-y$.

For example, applying this recipe to $Y^2 = X^3 + 2$ with $(k,m)=(11,7)$ and $(u,v)=(X,Y)$ can be done with the gp command

polresultant(polresultant(Y^2-X^3-2, X^11-x, X), Y^7-y, Y)

which yields the irreducible polynomial

x^21 + (-36960*y^2 + 14336)*x^18 + (154*y^6 + 396506880*y^4 + 19029491712*y^2 + 88080384)*x^15 + (12029248*y^8 - 1216366964736*y^6 + 284451691560960*y^4 - 391801278038016*y^2 + 300647710720)*x^12 + (5236*y^12 + 41292382208*y^10 + 649374955536384*y^8 + 261188560760078336*y^6 + 4032263773430480896*y^4 + 1338427807910330368*y^2 + 615726511554560)*x^9 + (-61949888*y^14 + 5483890384896*y^12 - 20842181162958848*y^10 + 8440505090430730240*y^8 - 456851090435200778240*y^6 + 2810028717064564244480*y^4 - 898304588698945060864*y^2 + 756604737398243328)*x^6 + (16016*y^18 + 1301780480*y^16 + 5890099511296*y^14 + 5689217111818240*y^12 + 1688064484691673088*y^10 + 172250320914033410048*y^8 + 5873494991874845310976*y^6 + 55086593993714775883776*y^4 + 78432673497651496353792*y^2 + 516508834063867445248)*x^3 + (-y^22 + 1408*y^20 - 901120*y^18 + 346030080*y^16 - 88583700480*y^14 + 15874199126016*y^12 - 2031897488130048*y^10 + 185773484629032960*y^8 - 11889503016258109440*y^6 + 507285462027012669440*y^4 - 12986507827891524337664*y^2 + 151115727451828646838272)

vanishing on infinitely many pairs $(x,y) = (X^{11},Y^7)$.

$\endgroup$
8
  • $\begingroup$ Is typo in the resultants possible? You don't appear to use $F$, might be wrong. $\endgroup$
    – joro
    Jan 10, 2015 at 7:06
  • $\begingroup$ You're welcome. I think the resultant is right. I just edited to add the gp code. I also checked that a few multiples of $(-1,1)$ yield rational multiples whose coordinates are the expected $11$th and $7$th powers. $\endgroup$ Jan 10, 2015 at 16:53
  • $\begingroup$ This indeed works. Better proof than checking is to factor q(x^11,y^7) and see it is divisible by the curve. Is it coincidence that in the factorization the degree 18 factor is genus 1 too? $\endgroup$
    – joro
    Jan 11, 2015 at 7:08
  • $\begingroup$ By the way, for $k=m=2$ the resultant is square and is genus 0. $\endgroup$
    – joro
    Jan 11, 2015 at 8:49
  • $\begingroup$ That's covered by the $(u',v')$ test I gave. When $m$ is even you can't take $u,v$ to be the $X,Y$ of the classical Weierstrass equation because then the map $(u,v) \mapsto (u^k,v^m)$ is not generically $1:1$ (try $(u',v') = (u,-v)$). But $(X,Y+1)$ works. (Also for $Y^2=X^3+2$ if $3 \mid k$ you must tweak $u$ to avoid triplication on the $X$ side.) $\endgroup$ Jan 11, 2015 at 15:30
3
$\begingroup$

Partial answer, someone possibly might hopefully complete it.

Solutions exists for $k=m \in \{2,3,4,5\}$.

The case $k=4$ was found with rational surface and $5$ using very similar approach.

Fix $k,m$ large.

Consider the rational surface given by parametrization: $x=u^k, y=v^m, z=h(u,v)$ where $h(u,v)=0$ is genus $1$, positive rank (possibly better cubic model).

The surface must satisfy some polynomial $g(x,y,z)=0$.

Take $u,v$ on $z=h(u,v)=0$ and $f(x,y)=g(x,y,0)$.

If $f(x,y)=0$ is irreducible, all $x,y$ are large perfect powers and the genus is zero or one.

Experimentally genus zero is possible, though haven't seen it on $h$ cubic model.

So to complete the proof, $f(x,y)=0$ must be irreducible and genus one.

$\endgroup$

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.