8
$\begingroup$

Setup/Notation: Let $n,m\in \mathbb{N}$ and let $C(\mathbb{R}^n,\mathbb{R}^m)$ be the space of continuous functions from $\mathbb{R}^n$ to $\mathbb{R}^m$ equipped with the compact-open topology. Let $\mathcal{I}(\mathbb{R}^n,\mathbb{R}^m)$ be the subset of $C(\mathbb{R}^n,\mathbb{R}^m)$ consisting of injective functions.

Observations - Effect of Dimension:

  • If $n\leq m$: The subset $\mathcal{I}(\mathbb{R}^n,\mathbb{R}^m)$ need not be closed. To see this note that the family $$ f_{n}(x):=\frac1{n}\cdot x, $$ converges to $0$ (in the compact-open topology).
  • If $n>m$: Then Brouwer's Invariance Theorem implies that $\mathcal{I}(\mathbb{R}^n,\mathbb{R}^m)=\emptyset$. In particular, $\overline{\mathcal{I}(\mathbb{R}^n,\mathbb{R}^m)}=\emptyset\neq C(\mathbb{R}^n,\mathbb{R}^m)$.

Question: Is there a critical $m^{\star}$ (depending on $n$) such that if: $$ \begin{cases} \overline{\mathcal{I}(\mathbb{R}^n,\mathbb{R}^m)} = C(\mathbb{R}^n,\mathbb{R}^m) &: m\geq m^{\star}\\ \overline{\mathcal{I}(\mathbb{R}^n,\mathbb{R}^m)} \neq C(\mathbb{R}^n,\mathbb{R}^m) & : m<m^{\star} \end{cases}? $$ If so, more precisely, what is $m^{\star}$ and how does it depend on $n$?


Partial Thoughts/Observations'' - (Edit): If we embed $C(\mathbb{R}^n,\mathbb{R}^m)$ into $C(\mathbb{R}^n,\mathbb{R}^{m+n})$ via $f\mapsto \tilde{f}_{\infty}(x):=[x\mapsto (f(x),0)]$ then, for any $f\in C(\mathbb{R}^n,\mathbb{R}^{m})$ we can define the maps: $$ \tilde{f}_{n}(x):= (f(x),\frac1{n}\cdot x). $$ Moreover, for each $n\in \mathbb{N}$, the map $f\mapsto \tilde{f}_n$ is an embedding. So, then in this way... an affirmative pseudo-answer to the question below is ``kind of'' since: $\lim\limits_{n\to \infty}\,\tilde{f}_n =\tilde{f}_{\infty}$ and each $\tilde{f}_n \in \mathcal{I}(\mathbb{R}^n,\mathbb{R}^{m+n})$ and is the image of some $C(\mathbb{R}^n,\mathbb{R}^m)$.

$\endgroup$
12
  • 2
    $\begingroup$ @JochenWengenroth I don't follow. If a curve is injective, its components need not to be injective. $\endgroup$ Commented Sep 22, 2021 at 16:56
  • 2
    $\begingroup$ But note that a curve $\mathbb R\to\mathbb{R^2}$, whose image has a self-crossing point, can't be approximated by injective curves $\endgroup$ Commented Sep 22, 2021 at 17:04
  • 2
    $\begingroup$ just consider a nbd of the crossing point; say that there the curve to be approximated looks like this $\otimes$. Any approximating curve has to follow the two segments, and needs to cross itself. $\endgroup$ Commented Sep 22, 2021 at 17:15
  • 1
    $\begingroup$ @TomTheQuant: Sure, in 3 or more dimensions you can make it injective, but the point is not in 2. $\endgroup$ Commented Sep 22, 2021 at 17:35
  • 1
    $\begingroup$ If n<m/2 one can first approximate the map by a smooth one, and then try to approximate by injective ones. There should be enough room. $\endgroup$ Commented Sep 22, 2021 at 17:36

1 Answer 1

5
$\begingroup$

Kuratowski proved in Sur les théorèmes du „plongement" dans la théorie de la dimension. Fundamenta Mathematicae 28.1 (1937): 336-342 that the set of embeddings of an at most $n$-dimensional separable metrizable space into $\mathbb{R}^{2n+1}$ contains a dense $G_\delta$-set. Hence $m^*\le 2m+1$.

$\endgroup$
2
  • $\begingroup$ Thanks this is very interesting and basically exactly what I had hoped for! :) $\endgroup$
    – ABIM
    Commented Sep 24, 2021 at 8:31
  • 1
    $\begingroup$ A more tentative argument for the inequality $m^*\ge2m+1$: the union of all $n$-dimensional faces of a $2n$-simplex cannot be embedded in to $\mathbb{R}^{2n}$ (Flores, 1935, see Exercise 1.11.H in Engelkings Theory of Dimensions. Finite and Infinite). In case $n=1$ you get the complete graph $K_5$. If you draw that in the plane as a pentagram in a circle and if you parametrize that, say with a map of period $10$ then no map close enough to that parametrization is injective (hence $1^*\ge3$). I would expect that something similar can be done for larger $n$. $\endgroup$
    – KP Hart
    Commented Sep 25, 2021 at 16:38

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .