I want to prove a particular case of Sard's Theorem, obviously without using the main Sard's Theorem.

Let $f:M\to N$ be a differentiable ($C^{\infty}$) function. If $m=\dim M<\dim N=n$, then $f(M)$ has measure zero in $N$.

Hint: It is enough to show that if $g:U\subseteq\mathbb{R}^m\to\mathbb{R}^n$ is differentiable, where $U$ is open, then $g(U)$ has measure zero in $\mathbb{R}^n$.

I proved the hint:

Just consider the projection $\pi:U\times\mathbb{R}^{n-m}\to\mathbb{R}^m$ over $U$ and $g\circ\pi:U\times\mathbb{R}^{n-m}\to\mathbb{R}^n$. Because $U\times\mathbb{R}^{n-m}$ and $\mathbb{R}^n$ are in the "same dimention" and $g\circ\pi$ is differentiable, this function sends null sets to null sets. So that $(g\circ\pi)(U\times\{0\}^{n-m})=g(U)$ has measure zero.

But I don't understand why this is enough.

In order to show $f(M)$ has measure zero in $N$, we need to show that if $(V,\phi)$ is a chart in $N$ then $\phi(f(M)\cap V)$ has measure zero in $\mathbb{R}^n$. As the hint suggests, how can I construct some $g:U\subseteq\mathbb{R}^m\to\mathbb{R}^n$ differentiable from here?

  • 1
    $\begingroup$ Use coordinate charts in both $M$ and $N$ to reduce to the hint. If $f(U)$ has measure zero for every open $U\subset M$, then ... $\endgroup$ – Neal Mar 21 '16 at 22:03

Let $(\psi,U)$ be a chart on $M$ and $\varphi$ a local chart on $N$. The overlap map given by;

$$g=\varphi \circ f \circ \psi^{-1}: \psi(U) \subset \mathbb{R}^m \to \mathbb{R}^n$$

is differentiable. Here treat $\psi(U)$ as $U$ since $\psi$ is a local homeomorphism. Therefore, from what you've shown, it follows that $g(\psi(U))=\varphi(f(U))$ has measure zero. So we have a covering of ball $\{B_{\alpha}\}$ of $\varphi(f(U))$ such that $\sum vol(B_{\alpha}) < \delta$. Hence, $\{\varphi^{-1}(B_{\alpha})\}$ is a covering of $f(U)$ which has measure zero.

Now recall that there exists a countable basis $\{U_j\}$ of $M$ and therefore $f(M) \subset \bigcup_j f(U_j)$. However, we know know that each $f(U_j)$ has measure zero; say $vol(f(U_j))< \frac{\delta}{2^j}$ and so,

$$vol(f(M)) \leq \sum_j vol(f(U_j))< \sum_j\frac{\delta}{2^j} = \delta$$

  • $\begingroup$ Thank to you and Neal. It was all about that overlap map what I couldn't think. $\endgroup$ – JonSK Mar 22 '16 at 0:01

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.