Let $k<m$ and $f:\mathbb{R}^k \to \mathbb{R}^m$ be continuously differentiable around $x_0$. with $\operatorname{rank}Df(x_0) = k$.

Then $\forall r>0$ small enough $f(B(x_0,r))$ doesn't contain a ball around $f(x_0)$.

My attempt is the following:

Since $f$ is onto its image $\operatorname{Img}(f) \subset \mathbb{R}^m$ of degree at most $k$. Since $\operatorname{rank}Df(x_0) = k$, around $x_0$ the image of $f$ is a subset of $\mathbb{R}^m$ which is of dimension $k$. That is, around $x_0$ $f$ is injective (by the inverse function theorem).

The inverse function theorem applies for functions from $\mathbb{R}^k$ to itself, is this a legitimate use of this theorem, thinking of the image as a $k$ dimensional subset of $\mathbb{R}^m$? If so how do we show this?

Letting $U$ be the environment in which $f$ is injective, we take $r>0$ so that $B(x_0,r) \subset U$. If $\exists \delta > 0$ s.t $B(f(x_0),\delta) \subset f(B(x_0,r))$ then since $\dim(B(f(x_0),\delta)) = m$ we get $m \leq k$, in contradiction.

I'm not sure the use of the inverse function theorem is correct (due to the fact the image is in $\mathbb{R}^m$). And also the second part, couldn't I have concluded that without $f$ being injective in $U$?

  • 1
    $\begingroup$ If you write $k<m$ then why write $\mathbb {R^k}$ and $\mathbb{R^m}$ instead of $\mathbb R^k$ and $\mathbb R^m$? That doesn't make sense so I changed it along with some other corrections. $\endgroup$ – Michael Hardy Jan 23 '17 at 15:09
  • $\begingroup$ @MichaelHardy thanks! $\endgroup$ – Mariah Jan 23 '17 at 15:11
  • $\begingroup$ @MichaelHardy any ideas on the question itself? $\endgroup$ – Mariah Jan 23 '17 at 15:45

HINT: Suppose $df(x_0)=\begin{bmatrix} I_k \\ 0 \end{bmatrix}$. Consider $g\colon\Bbb R^k\times \Bbb R^{m-k}\to\Bbb R^m$ given by $g(x,y) = f(x)+(0,y)$.

  • $\begingroup$ Which part of my question is your hint addressed for? For my attempt to $f$ is injective around $x_0$? Also, do you mean to think of $\Bbb R^k\times \Bbb R^{m-k}$ as $\mathbb{R}^m$ and use the inverse function theorem? That may achieve $g$ as injective on some subset of $\mathbb{R}^m$, but why would it imply $f$ is too? $\endgroup$ – Mariah Jan 23 '17 at 22:48
  • $\begingroup$ Yes to everything. Write out the definition of injective :) $\endgroup$ – Ted Shifrin Jan 23 '17 at 23:02
  • $\begingroup$ Ok, with that information, I'd like to understand if the last part of the proof is correct (my ball dimension argument) (where I claim this cannot happen since $f$ being injective implies the ball is of dimension $k$ and can't contain a subset of dimension $m$). I also don't understand how $B(f(x_0),\delta) \subset f(B(x_0,r))$ in any case since $k < m$. Do you have an idea for this? $\endgroup$ – Mariah Jan 23 '17 at 23:08
  • $\begingroup$ Intuitively, the image of $f$ will (at least locally) be a $k$-dimensional submanifold in $\Bbb R^m$. But you now have $g^{-1}$ (on a small ball) to use. Could $f$ cover a whole ball in $\Bbb R^m$? $\endgroup$ – Ted Shifrin Jan 23 '17 at 23:16
  • $\begingroup$ I still don't understand why not; If I differentiate around $(x_0,0)$, for example, I get a non singular Jacobian for $g$ and then $g$ is injective and onto from and on some open sets. Then $f$ is also injective on the restriction to $\mathbb{R}^k$. If we assume $f$ covers a ball, then the image of $g(x,0)$ covers that ball as well - I don't see a way to continue from here yet. $\endgroup$ – Mariah Jan 25 '17 at 12:18

Fix an $x$ where derivative is nonsingular. Notice that $Df(x)(\mathbb{R}^k)$ is a linear $k$-dimensional subspace of $\mathbb{R}^m$. Near the point $x$, $f$ is very close (little $o$ - close) to the derivative map. Thus, projection onto the affine plane $f(x) + Df(x) (\mathbb{R}^k)$, which is a copy of $\mathbb{R}^k$, we end up with a map between same dimensional Euclideans. Derivative agrees with the original derivative, hence is nonsingular. Now, we are in a position to impose IFT. If projection of $f$ is injective then we deduce that image of $f$ is (locally) $k$-dimensional.

For detailed argument refer to https://mathoverflow.net/a/323429/121665 .


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.