Inverse function theorem on smooth manifolds

256 Views Asked by At

Following this thread, I'm trying to prove in detail this theorem. Could you please check if my proof is fine or contains logical mistakes?

Theorem: Let $X \subseteq \mathbb R^M$ and $Y \subseteq \mathbb R^N$ be $n$-dimensional smooth manifolds, $f:X \to Y$ smooth, and $x \in X$. If $\mathrm d f_x: T_x X \to T_{f(x)}Y$ is an isomorphism, then $f$ is a local diffeomorphism around $x$.


Proof: Clearly, $T_xX$ and $T_{f(x)}Y$ are $n$-dimensional vector subspace of $\mathbb R^M$ and $\mathbb R^N$ respectively. Let $\varphi:U \to V$ and $\psi:A \to B$ be local parameterizations around $x$ and $f(x)$ respectively.

Clearly, $f \circ \varphi$ is smooth. There exists an open neighborhood $U'$ of $\varphi^{-1} (x)$ in $U$ such that $h:= \psi^{-1} \circ f \circ \varphi_{\restriction U'}$ is well-defined, i.e., the domains and images of functions in the composition are compatible.

Clearly, $h$ is smooth and we write $$\mathrm d h_{\varphi^{-1} (x)}: \mathbb R^n \to \mathbb R^n.$$

Also, $$\mathrm d f_x := \mathrm d \psi_{\psi^{-1} \circ f (x)} \circ \mathrm d h_{\varphi^{-1} (x)} \circ (\mathrm d \varphi_{\varphi^{-1} (x)})^{-1}.$$

Because $\varphi, \psi$ are diffeomorphisms, $$\mathrm d \psi_{\psi^{-1} \circ f (x)}: \mathbb R^n \to T_{f(x)}Y \quad \text{and} \quad (\mathrm d \varphi_{\varphi^{-1} (x)})^{-1}: T_x X \to \mathbb R^n$$ are isomorphisms. This combining with the fact that $\mathrm d f_x$ is an isomorphism implies $\mathrm d h_{\varphi^{-1} (x)}$ is also an isomorphism.

Notice that $h$ is a smooth function whose domain is an open subset of a Euclidean spaces, so we can apply IFT for Banach spaces. This means there exists an open neighborhood $U''$ of $\varphi^{-1} (x)$ in $U'$ such that $A' := h(U'')$ is open in $\mathbb R^n$, and $h_{\restriction U''}: U'' \to A'$ is a diffeomorphism. Wlog, we assume $U' = U''$.

There exists some open neighborhood $V'$ of $x$ in $V$ such that $f_{\restriction V'} = \psi \circ h \circ \varphi^{-1}_{\restriction V'}$ is well-defined, i.e., the domains and images of functions in the composition are compatible. Diffeomorphism is open, so we can take, for example, $V' := \varphi (U')$.

We have $\varphi^{-1}_{\restriction V'}$, $h$, and $\psi$ are all diffeomorphisms, so $f_{\restriction V'}: V' \to B'$ with $B' := f(V')$ is the composition of diffeomorphisms and thus is also a diffeomorphism.

We have $V'$ is open in $V$ and $V$ is open in $X$, so $V'$ is open in $X$ by this lemma. Let's show that $B'$ is indeed open in $Y$.

  • First, $V'$ is open in $V$ and $\varphi^{-1}: V \to U$ is a diffeomorphism implies $\varphi^{-1} (V')$ is open in $U$. This combining with this lemma and $\varphi^{-1} (V') \subseteq U' \subseteq U$ implies $\varphi^{-1} (V')$ is open in $U'$.

  • Second, $h:U' \to A'$ is a diffeomorphism, so $h \circ \varphi^{-1} (V')$ is open in $A'$. On the other hand, $A'$ is open in $A$, so $h \circ \varphi^{-1} (V')$ is open in $A$ by this lemma.

  • Third, $\psi: A \to B$ is a diffeomorphism, so $B'$ is open in $B$. On the other hand, $B$ is open in $Y$, $B'$ is also open in $Y$ be this lemma.