Proving Sard's theorem

499 Views Asked by At

Theorem: (Sard) Let $f:U\to \mathbb R^p$ be a smooth map with $U$ open in $\mathbb R^n$, and let $C$ be the set of all critical points of $f$. Then $m([f(C)]=0$ where $m$ is the Lebesgue measure of $\mathbb R^p$.

A proof goes by induction on $n\leq 0$, while $p>0$. The base case is obvious. Let $C_k$ denote the elements of $U$ for which the partial derivatives of $f$ up to the $k$-th order vanish at the said element. We may brake the proof in three steps. We will be focusing on the first step.

1) $m([f(C-C_1)])=0$

2) $m([f(C_k-C_{k+1})])=0, \quad k>0$

3) $m([f(C_k)])=0, \quad\forall i>n_0$, for some natural number $n_0$.

Step 1) Assume $p>1$, since $C=C_1$ when $p=1$. For each $y\in C-C_1$ we will find a an open neighborhood $V$ containing $y$ for which $m([f(V\cap C)])=0$.

Up to an isometry, $y\not\in C_1\Rightarrow \frac{\partial f_1}{\partial x_1}(y)\not=0$. Consider $h(x):=(f_1(x),x_2,x_3,\dots,x_n)$. Since $dh_y$ is nonsingular, $h$ maps some neighborhood of $y$ diffeomorphicaly onto an open set $V'$. The composition $g=f\circ h^{-1}$ maps $V'$ to $\mathbb R^p$, etc.

Questions: How can we see that $g(t,x_2,\dots,x_n)\in t\times\mathbb R^{p-1}$, for each $(t,x_2,\dots,x_n)\in V'$? How can we conclude from the preceding that $g$ caries hyperplanes to hyperplanes?

Comment: Obviously, the first component of $g(t,x_2,\dots,x_n)$ is $t$, but there are many ways to include $\mathbb R^{p-1}$ in $\mathbb R^p$. I guess that the meaning of the statement is that some suitable inclusion exists but it varies with $t$.

Continuing: A point is critical for $g^t:=g|_{t\times\mathbb R^{n-1}\cap V'}$ if and only if it is critical for $g$. Therefore, the set $C'$ of critical values of $g$ intersects intersects each hyperplane $t\times\mathbb R^{p-1}$ in a set of measure zero. By a theorem of Fubini, $m([g(C')])=0$ but $f(V\cap C)=g(C')$. Hence, $m([f(C-C_1)])=0$ because it is a countable union of sets of measure zero. q.e.d.

Questions: Again many ways exist to restict $g$ to a hyperplane with a fixed first coordinate. Which restriction may we mean? (I think that we must mean every possible restriction.)

Moreover, the theorem of Fubini states that if a set $A\subset\mathbb R^n=\mathbb R\times\mathbb R^{n-1}$ intersects each hyperplane $\text{const}\times\mathbb R^{n-1}$ in a set of measure zero, then $A$ must also have measure zero. It seems that we have not satisfied the conditions of the theorem. In order to satisfy them we need to argue about every possible hyperplane.

1

There are 1 best solutions below

0
On

In John Milnor's book, "Topology from the Differentiable Viewpoint", there is a proof of Sard's theorem.

We have a map $g:f\circ h^{-1}$

Where $f:U \to R^p$ where U is an open set in $R^n$

$h:U\to R^n$ where $h(x)=(f_1(x),x_2,...,x_n)$

$h$ is a diffeomorphism which means its a bijection

For each point $(t,x_2,...x_n)\in V'$,

Let $h^{-1}:V' \to R^n$ where $h^{-1} = (t,x_2,...,x_n)$

Then $f(t,x_2,...,x_n)=(t,x_2,...,x_{p-1})$ which is a hyperplane

Therefore the composition $g:f\circ h^{-1}$ maps hyperplanes to hyperplanes