2

I am stuck at the following exercise in Zorich.

Let $f,g\in C^{(k)}(D;\mathbb{R})$, and suppose that $f(x)=0\Rightarrow g(x)=0$ in the domain $D$. Show that if grad $f \neq 0$, then there is a decomposition $g=h\cdot f$ in $D$, where $h\in C^{(k-1)}(D;\mathbb{R})$.

If $D\subseteq \mathbb{R}$, then the result follows immediately from Hadamard's lemma. If $D\subseteq \mathbb{R^m}$ with $m>1$, then I can show that for $f(x_0)=0$, the limit $\lim_{x\rightarrow x_0} g(x)/f(x)$ exists. But how to prove that $g/f$ is $C^{(p-1)}$ at such points?

Thanks in advance!

Jiu
  • 1,545
  • Dear @Jiu, is $D$ a domain in $\mathbb{R}^n$? I'm confused about $h$, because the image of $f$ lies in $\mathbb{R}$, not in $D$. You should require $h(0)=0$ and perhaps some additional constraints on $h$. – user90189 Dec 05 '18 at 04:36
  • @user90189 thanks for your comment! Now that you make me realize this problem, I just checked and see that in the English version, the $\circ$ is replaced by $\cdot$... So the $\circ$ should be a typo in the German version! – Jiu Dec 05 '18 at 04:53
  • So, I think you can try to check that $g/f$ is well-defined :) – user90189 Dec 05 '18 at 04:56
  • @user90189 In fact when I try to write down the solution I realize that it is not correct :( – Jiu Dec 05 '18 at 08:57

1 Answers1

1

So, let $q_0 \in D$, we want to show that on an open neighborhood $U \subset D$ of $q_0$, there is a $C^{p-1}$ function $h$ such that $g_{|U}=hf_{|U}$.

If $f(x_0) \neq 0$, it is obvious.

So let’s assume $f(q_0)=0$. Assume eg that $\frac{\partial f}{\partial x_1}(q_0) \neq 0$.

Define $\psi(q)=(f(q),x_2(q)-x_2(q_0), \ldots, x_n(q)-x_n(q_0))$.

Then $\psi$ is defined and $C^p$ on $D$, vanishes at $q_0$ and its derivative at $q_0$ is invertible.

As a consequence, there is an open subset $q_0 \in U \subset D$ and a “box” open subset $0 \subset V \subset \mathbb{R}^m$ (ie a product of $m$ open intervals) such that $\psi: U \rightarrow V$ is a $C^p$-diffeomorphism with inverse $\sigma$.

Now, note that for $x \in V$, $f(\sigma(x))=x_1$.

So define $g_1=g \circ \sigma$ on $V$, $g(x)=0$ whenever $x_1=0$. Define now $h_1(x)=\int_0^1{\frac{\partial g_1}{\partial x_1}((1-t)p(x)+tx)\,dt}$, where $p(x_1, \ldots, x_n)=(0,x_2,\ldots,x_n)$. It is easy to see that $h_1$ is $C^{p-1}$ and $h_1x_1=g_1$.

So $U$ and $h_1 \circ \psi$ work around $q_0$.

Aphelli
  • 34,439