I was studying the Inverse function theorem when I came across the following problems :

(Let the closed set $V$ i.e the range have non-empty interior)

  1. Does there exist a continuous onto function from an open set $U$ in $\mathbb{R}^n $ to a closed set $V$ in $\mathbb{R}^m$ such that some points in the interior of $U$ get mapped to the boundary of $V$?

  2. Does there exist a continuous $1-1$ map from an open set $U$ in $\mathbb{R}^n $ to a closed set $V$ in $\mathbb{R}^m$ such that some points in the interior of $U$ get mapped to the boundary of $V$?

If there are examples in $C(\mathbb{R})$ i.e continuous functions from $\mathbb{R}$ to $\mathbb{R}$, then that would be great too! Though I do need some example in the general case too.

Simpler examples will be really appreciated.

Thanks in advance.


The case (1) can be dealt with using any "cut-off" function. e.g let $U,V$ two balls around $0$ in $\mathbb{R}^n $ with radius $r(>1)$ and $1$, and be open and closed respectively.

Let $f: U \rightarrow V $ such that $x \in V \implies f(x)=x$ and $x \in U-V \implies f(x)= x/||x|| $.


I am assuming you want $V$ to actually be the image of $U$. In this case, there is no such map satisfying your second condition.

If $m = n$, this follows from invariance of domain, since the image of $U$ will necessarily be open.

If $m < n$, there is no continuous injective map from $\mathbb{R}^n$ to $\mathbb{R}^m$ (let alone to a closed subset). You can find some more elementary arguments here, or you can again apply invariance of domain. In particular, if $f : U \to V$ is continuous and injective, and $\iota : \mathbb{R}^m \to \mathbb{R}^n$ is an inclusion map, then $\iota \circ f : U \to \mathbb{R}^n$ is an open map, but the image of $U$ is not open in $\mathbb{R}^n$.

For $m > n$, similar logic: we cannot have a continuous injective map from $U$ onto a set with non-empty interior in $\mathbb{R}^m$ (since $\mathbb{R}^n$ and $\mathbb{R}^m$ are not homeomorphic).

Since we didn't even use the fact that an interior point maps to the boundary of $V$, I suspect there is an easier argument. (Maybe take an interior point $u$ which maps to the boundary, restrict to a small compact neighborhood of $u$, and use the fact that the map on the compact neighborhood is a homeomorphism and must preserve the boundary).

Also, I guess it's possible that you never intended for $V$ to actually be the image of $U$. In this case, we still cannot find such a function when $m \leq n$, based on the same arguments as above, but we can for $m > n$. For example, take $V$ to be the unit disk in $\mathbb{R}^2$ and consider the map $(-1/2,1/2) \to \mathbb{R}^2 : x \mapsto (x,1-4x^2)$, or something similar. This maps $0$ to the boundary of $V$.

  • 2
    $\begingroup$ In the $m>n$ case, you have to say a bit more, because the continuous injection might not be a homeomorphism. One thing you can do is write $U$ as a countable union of open subsets whose closures in $U$ are compact, and then by the Baire category theorem, the image of one of these sets must have nonempty interior. By compactness, our continuous injection is actually a homeomorphism onto an image when restricted to this set. $\endgroup$ – Eric Wofsey Feb 13 '16 at 4:25
  • $\begingroup$ Ah yeah, good point, and nice argument $\endgroup$ – Michael Harrison Feb 13 '16 at 4:35

The sine/cosine functions map $\mathbb{R}$ to the closed interval $[-1,1]$.

(THIS IS WRONG: Moreover, the embedding $\mathbb{R} \to \mathbb{R}^2, x \mapsto (x, 0)$ satisfies both assertions.)

  • 2
    $\begingroup$ Doesn't the embedding renders the real line to be a closed set? ($\mathbb{R}$ is a closed subset of $\mathbb{R}^2$) $\endgroup$ – NeerajKumar Feb 6 '16 at 14:04
  • 1
    $\begingroup$ A set being open is relative to the space in which it is embedded. When defining the embedding as above, it is defined on $\mathbb{R}$ which is open in $\mathbb{R}$. If we were seeing the domain as a closed subset of $\mathbb{R}^2$, we have already embedded it and there is no need of such a function. $\endgroup$ – ChaPi Feb 6 '16 at 21:56
  • $\begingroup$ @ChandrahasPiduri - That's true but the claim by OP that this will satisfy the 1-1 case is false. The embedding maps $\mathbb{R}$, which is by itself open, to $\mathbb{R}^2$ which is closed. But where are the interior points getting mapped to the boundary. There is no boundary of $\mathbb{R}^2$, unless see it as a subset of $\mathbb{R}^3$, in which case the closed set $V=\mathbb{R}^2$ has all the point as boundary point but then it has empty interior! The assumption about non-empty interior precisely removes all such cases. $\endgroup$ – NeerajKumar Feb 7 '16 at 9:45
  • $\begingroup$ Oh, you are right. The image does not have non-empty interior... $\endgroup$ – Paul K Feb 7 '16 at 9:47

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.