Suppose $f$ and $g$ are polynomials with integral coefficients and $f(\mathbb Z)\subset g(\mathbb Z)$. Is there any relation between $f$ and $g$?

For instance, this happens if $f=g\circ h$ for some third integral polynomial $h$. In the particular case that $f$ takes on only square values, we could deduce that the converse is true: $f$ is a square, so $f=g\circ h$ where $h$ is some polynomial and $g(x)=x^2$. Does something like this hold more generally?

  • 6
    $\begingroup$ What about $f(x) = x(x+1)$ and $g(y) = 2y$? $\endgroup$ – Jason Starr Jul 20 '15 at 19:53
  • 6
    $\begingroup$ The curve $f(x)=g(y)$ will have infinitely many integral points so will be reducible or have genus zero. Google irreducibility of $f(x)-g(y)$ for many papers on this. $\endgroup$ – Felipe Voloch Jul 20 '15 at 20:41
  • 6
    $\begingroup$ In particular, a paper of Bilu and Tichy (Acta Arith, XCV 2000) is relevant. $\endgroup$ – Felipe Voloch Jul 20 '15 at 20:43
  • 1
    $\begingroup$ See mathoverflow.net/questions/105304/… for more on Voloch's comment. $\endgroup$ – David E Speyer Jul 21 '15 at 3:51
  • 1
    $\begingroup$ I think that using the suggested results will only lead to complicated issues of rationality and a lengthy case-by-case analysis on the pairs in the paper of Bilu and Tichy. See Theorem 1 in: H. Davenport, D. J. Lewis, A. Schinzel, Polynomials of certain special types. Acta Arith. 9 (1964), 107-116 for a result better suited for this problem. $\endgroup$ – Pasten Jul 21 '15 at 17:32

The suggestion from the comments to use Siegel's Theorem about integral points on algebraic curves is vast overkill. The by far easier Hilbert's Irreducibility Theorem is sufficiently strong here:

Suppose that $f(\mathbb Z)\subseteq g(\mathbb Z)$. Write $f(X)-g(Y)=A_1(X,Y)A_2(X,Y)\cdots A_r(X,Y)$ with polynomials $A_i(X,Y)\in\mathbb Q[X,Y]$ which are irreducible over $\mathbb Q$. By Hilbert's Irreducibility Theorem, there is an infinite set $H$ of integers such that $A_i(h,Y)$ is irreducible for each $h\in H$ and each $i$. On the other hand, for each integer $h$ there is an integer $u$ such that $f(h)=g(u)$. So $f(h)-g(Y)$ and therefore some $A_i(h,Y)$ has an integral root. Running through $h\in H$, we see that there is some $i$ such that $A_i(h,Y)$ has an integral root for infinitely many $h\in H$.

So for this index $i$ there are infinitely many $h\in H$ such that $A_i(h,Y)$ is irreducible and has a rational root. So $A_i(X,Y)$ has degree $1$ in $Y$.

We obtain $f(X)-g(Y)=A(X,Y)(b(X)-Yc(X))$ for $A\in\mathbb Q[X,Y]$ and $b,c\in\mathbb Q[X]$. Setting $Y=b(X)/c(X)$ yields $f(X)=g(b(X)/c(X))$. Looking at poles shows that $c(X)$ actually is a constant. So $f(X)=g(b(X))$ for a polynomial $b(X)$ with rational coefficients.

  • $\begingroup$ Great answer! I think you need to say something a little more sophisticated that "$A_i(h,Y)$ has an integral root for infinitely many $h \in H$ ... this implies that $A_i(h,Y)$ has degree $1$ in $Y$." The polynomial $h^2 - 2 y^2 - 1$ has an integer root for infinitely many $h$. But I agree that making some statement about positive upper density should make the argument work. $\endgroup$ – David E Speyer Jul 22 '15 at 14:53
  • $\begingroup$ I should mention that the idea in this argument is a particular case of the previously mentioned paper of Davenport-Lewis-Schinzel, where a more general result is proved using HIT. $\endgroup$ – Pasten Jul 22 '15 at 15:09
  • $\begingroup$ @David: I tried to clarify my argument. The point is that for some fixed $i$, $A_i(h,Y)$ is irreducible and at the same time has a rational root for infinitely many integers $h$. And this of course implies $Y$-degree $1$. $\endgroup$ – Peter Mueller Jul 22 '15 at 15:10
  • $\begingroup$ Ah, I see. Yes, that works. $\endgroup$ – David E Speyer Jul 22 '15 at 15:26

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.