6
$\begingroup$

Looks like there is counterexample to Proposition related to abc conjecture. Confusion is likely.

From RATIONAL AND INTEGRAL POINTS ON QUADRATIC TWISTS OF A GIVEN HYPERELLIPTIC CURVE, Andrew Granville


p. 11, Proposition 2 b

Suppose that $G(x,y) \in \mathbb{C}[x,y]$ is homogeneous without any repeated factors. For any coprime polynomials $r(t),s(t) \in \mathbb{C}[t]$, we have

$$ \#\{\alpha \in \mathbb{C}: G(r(\alpha),s(\alpha))=0\} \ge \max\{\deg(r),\deg(s)\}(\deg(G)-2) + 2. $$


$\#\{\alpha \in \mathbb{C}: G(r(\alpha),s(\alpha))=0\}$ counts the distinct zeros and equals the degree of the radical of $G(r(t),s(t))$.

Explicit counterexample.

Let $G(x,y)=x^4+xy^3,r(t)=8t^3 + 64,s(t)=t^4 - 64t$

We have:

$$ G(r(t),s(t))=\left(8\right) \cdot (t + 2) \cdot (t^{2} - 2 t + 4) \cdot (t^{2} + 4 t - 8)^{2} \cdot (t^{4} - 4 t^{3} + 24 t^{2} + 32 t + 64)^{2} $$

So $G(r(t),s(t))$ have $9$ distinct zeros.

By the Proposition $9 \ge (( (4\cdot(4-2)+2)=10)$ which is false.

Q1 Is this really counterexample?

The Proposition is unconditional and this doesn't appear to contradict abc.

The errata of the paper doesn't address this.


Andrew Granville ask for other $G$. There are constructions.

Here is example in computer readable form with t=x:

G=x^3*y + x*y^3 + 8*y^4
r=x^16 - 40*x^14 - 4352*x^13 + 348*x^12 + 1024*x^11 + 189416*x^10 + 14080*x^9 + 622022*x^8 + 4485120*x^7 + 910312*x^6 + 13647104*x^5 + 65163612*x^4 + 3943424*x^3 + 46235608*x^2 + 134216960*x - 1050623
s=16*x^15 + 176*x^13 + 5248*x^12 + 400*x^11 + 30976*x^10 + 433584*x^9 - 4224*x^8 + 343472*x^7 + 486912*x^6 - 392816*x^5 - 4060288*x^4 - 16662352*x^3 + 1313024*x^2 + 8413200*x + 33685632
$\endgroup$

2 Answers 2

12
$\begingroup$

I checked the proof of Granville. The proof only yields the bound $\max \{ \deg(r),\deg(s)\}(\deg(G)-2)\}+1$ which covers your counterexample.

To be more detailed: The polynomials $r,s$ yield a morphism $\mathbb{P}^1\to \mathbb{P}^1$. The set $G=0$ consists of $\deg(G)$ points. The set of $\alpha$ with $G(r(\alpha),s(\alpha)=0$ is the set of points in the preimage of this set with finite $t$-coordinate. The complete set consists of at least $\max \{ \deg(r),\deg(s)\}(\deg(G)-2)\}+2$ points, but one of them may be the point at infinity and hence you have to subtract one. This is precisely what is happening in your example.

$\endgroup$
5
  • $\begingroup$ Thank you. What about r,s in C[u,v] and both r,s depend on both u,v. Granville's bound is attainable. Is there a counterexample to this? $\endgroup$
    – joro
    Apr 12, 2015 at 8:47
  • $\begingroup$ If you take $r,s$ in $\mathbb{C}[u,v]$ homogeneous and of the same degree then Granville's bound seems correct. $\endgroup$ Apr 12, 2015 at 8:55
  • $\begingroup$ Thanks. What about general r,s with the only restriction to depend on both u,v? $\endgroup$
    – joro
    Apr 12, 2015 at 9:30
  • $\begingroup$ Asked the same: mathoverflow.net/questions/202695/… $\endgroup$
    – joro
    Apr 12, 2015 at 10:07
  • $\begingroup$ The bivariate case is explicitly false as per the linked question. $\endgroup$
    – joro
    Apr 14, 2015 at 8:20
16
$\begingroup$

Thanks to Remke for telling me about this. The proposition should have read

$$ \#\{\alpha \in \mathbb{C}\cup \{ \infty\}: G(r(\alpha),s(\alpha))=0\} \ge \max\{\deg(r),\deg(s)\}(\deg(G)-2) + 2. $$

ie you have to include the possible "root at infinity" (appropriately interpreted). However, it is probably easiest to change the statement of the result the way Remke suggests, so as to avoid worrying about the correct interpretation of a root at infinity. Sorry for the confusion.

It is a challenge to construct such examples in order to obtain equality for other $G$ (or even arbitrary $G$).

$\endgroup$
1
  • 1
    $\begingroup$ I added another $G$ per your request. Contact me via email if you need the construction. $\endgroup$
    – joro
    Apr 12, 2015 at 9:29

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.