Showing that a differential is nonzero

159 Views Asked by At

Problem Statement: Let $g:U\rightarrow \mathbb{R}^{n}$ be a $C^{1}$ map ($U\subset \mathbb{R}^{n}$ open), with $dg(\mathbf{x})\in GL(\mathbb{R}^{n})$ for all $\mathbf{x}\in U$. Suppose $\mathbf{y}_{0}\notin g(U)$, and let $\psi(\mathbf{x})=\lvert g(\mathbf{x})-\mathbf{y}_{0}\rvert^{2}$. Show that $d\psi(\mathbf{x})\neq \mathbf{0}$, for any $\mathbf{x}\in U$.

I have been working through this problem (from Fleming) but I am kind of lost. First, I do not know if I am computing $d\psi(\mathbf{x})$ correctly. I thought by the chain rule, we have $$d\psi(\mathbf{x})=2\lvert g(\mathbf{x})-\mathbf{y}_{0}\rvert\circ dg(\mathbf{x}).$$ Then we must show that for any $\mathbf{x}\in U$ we may find $\mathbf{v}\in \mathbb{R}^{n}$ so that $d\psi(\mathbf{x})[\mathbf{v}]$.

We know that $dg(\mathbf{x})[\mathbf{v}]\neq \mathbf{0}$ if $\mathbf{v}\neq \mathbf{0}$ since $dg(\mathbf{x})\in GL(\mathbb{R}^{n})$. But, (if I even computed $d\psi$ correctly) I am not sure where $2\lvert g(\mathbf{x})-\mathbf{y}_{0}\rvert$ comes into play, or my choice for $\mathbf{x}$. Since $\mathbf{y}_{0}\notin g(U)$, then $\lvert g(\mathbf{x})-\mathbf{y}_{0}\rvert\neq 0$, but that seems too trivial, because then I would have $d\psi(\mathbf{x})[\mathbf{v}]=2\lvert g(\mathbf{x})-\mathbf{y}_{0}\rvert\circ dg(\mathbf{x})[\mathbf{v}]\neq\mathbf{0}$ for any $\mathbf{x}$ and $\mathbf{v}\neq \mathbf{0}$.

I was given a hint that if $A\in GL(\mathbb{R}^{n})$ then $A^{T}[\mathbf{w}]\neq \mathbf{0}$ if $\mathbf{w}\neq \mathbf{0}$, and so $\mathbb{v}=A^{T}[\mathbf{w}]$ satisfies $\langle \mathbf{w},A[\mathbf{v}]\rangle\neq 0$. I assume I should be applying this fact to $dg(\mathbf{x})$, but I don't understand how the inner product is being used relative to this problem.

Any suggestions are appreciated!

1

There are 1 best solutions below

3
On BEST ANSWER

It is sometimes better to write down the functions involved explicitly.

In this problem there are two functions, one is $g : U \to \mathbb{R}^n$ and the other is a function $f : U \to \mathbb{R}$ defined by $$f(u) = \lVert u - y_0 \rVert^2 = \sum_{i=1}^n (u^i - y_0^i)^2$$ where $u = (u^1,\ldots, u^n)$ and $y_0 = (y_0^1,\ldots,y_0^n)$. Then the derivative of $f$ is $$ df_u = 2(u^1-y_0^1,\ldots, u^n-y^n).$$

The function $\psi$ is just $\psi = f \circ g : U \to \mathbb{R}$. Applying chain rule we have $$d\psi_x = df_{g(x)}\circ dg_x $$ for all $x \in U$.

If there is $x \in U$ with $d\psi_x = df_{g(x)}\circ dg_x = 0$ we can conclude that $df_{g(x)} =0$ because $dg_x$, by hypothesis, is a non-singular linear map. Therefore, $$ df_{g(x)} = 2(g^1(x)-y_0^1,\ldots, g^n(x)-y^n_0)=0$$ which of course implies that $g(x) = (g^1(x),\ldots, g^n(x))=y_0$, a contradiction.