8
$\begingroup$

Let $\varphi:\mathbb R^n \to \mathbb R^n$ be just some continuous function.

If the image of $\varphi$ happens to contain $\mathbb Q^n$, does it follow that in fact all of $\mathbb R^n$ is contained in the image as well?

No, it does not. For instance, the map given by $$(x,y)\overset{\varphi}{\longmapsto} (xy-1+\pi, x^2(xy-1)+y)$$ has as its image every point in $\mathbb R^2$ except for $(\pi,0)$.

But now suppose instead that $\mathbb Q^n$ is actually contained in the image of just $\mathbb Q^n$. That is, instead of requiring only that $$\mathbb Q^n \subseteq \varphi[\mathbb R^n]\phantom{.}$$ we require that in fact $$\mathbb Q^n \subseteq \varphi[\mathbb Q^n].$$

Question. Does it then follow that $\varphi[\mathbb R^n]=\mathbb R^n$, or is there a counterexample?

$\endgroup$
5
  • 7
    $\begingroup$ Yes for $n=1$. No for $n\ge 2$. For instance, for $n=2$ take the complex exponential plus an irrational ($f:z\mapsto\exp(z)+z_0$), and precompose with a self-homeomorphism of $\mathbf{R}^2$ mapping $\mathbf{Q}^2$ onto $f^{-1}(\mathbf{Q}^2)$. $\endgroup$
    – YCor
    Jun 25, 2021 at 12:15
  • $\begingroup$ @YCor : How is such a self-homeomorphism constructed? $\endgroup$ Jun 25, 2021 at 12:56
  • $\begingroup$ @IosifPinelis this inverse image is dense, and $\mathrm{Homeo}(\mathbf{R}^n)$ acts transitively on dense countable subsets of $\mathbf{R}^n$ (I think the latter fact appears somewhere on MO). $\endgroup$
    – YCor
    Jun 25, 2021 at 13:02
  • 2
    $\begingroup$ @YCor : Thank you. I have indeed found this answer: mathoverflow.net/a/281159/36721 . $\endgroup$ Jun 25, 2021 at 13:23
  • 2
    $\begingroup$ @YCor, since that (together with @‍IosifPinelis's reference) seems to be a complete answer, maybe it could be posted as such? $\endgroup$
    – LSpice
    Jun 25, 2021 at 13:47

1 Answer 1

13
$\begingroup$

A counterexample for $n=2$ is the map $\varphi(x,y) = (x,(x^2-2)y)$. Each point $(r,s)\in\mathbb{Q}^2$ is the image of $\left(r,\frac{s}{r^2-2}\right)\in\mathbb{Q}^2$, but e.g. $(\sqrt{2},1)\not\in\varphi(\mathbb{R}^2)$.

$\endgroup$
2
  • 4
    $\begingroup$ For $n\ge 2$ then take $(x_1,\dots,x_n)\mapsto (x_1(x_n^2-2),x_2,\dots,x_n)$, whose inverse as birational map is $(x_1,\dots,x_n)\mapsto (x_1/(x_n^2-2),x_2,\dots,x_n)$ $\endgroup$
    – YCor
    Jun 25, 2021 at 18:38
  • 1
    $\begingroup$ Fabulous! I was secretly wondering about polynomial maps, so this is extra useful for me. Thank you! $\endgroup$ Jun 26, 2021 at 0:17

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.