Currently, I am studying smooth manifolds and I want to solve some exercises. There is a question that says:
Show that the dimension of a connected topological manifold is defined without ambiguity! Meaning that if $\dim M=n$, with the change of charts, the dimension still is $n$. Then show that this is true for a $C^r$ connected manifold using Inverse Function Theorem.
To prove the first part, let $(U,\varphi)$ and $(V,\psi)$ be two charts with $\varphi(U) \subseteq \mathbb{R}^m$ and $\psi(V) \subseteq \mathbb{R}^n$ and suppose that $U \cap V \neq \emptyset$. Then, since $\psi \circ \varphi : \varphi (U \cap V) \rightarrow \psi (U \cap V) $ is a homeomorphism, $m=n$. Now, using this, if $M$ is a connected topological manifold with atlas $\mathcal{A}$, for all charts $(U_{\alpha}, \varphi_{\alpha})\in \mathcal{A}$, $\varphi_{\alpha}(U_{\alpha}) \subseteq \mathbb{R}^n$ and $n$ is constant for all the charts. Am I right?
But I have a problem showing the second part! I think that since every $C^r$ manifold is also a topological manifold, the answer to this question would be trivial! Why do we need "Inverse Function theorem" to answer it??
Any help is appreciated.
Your argument about overlapping chart sets uses invariance of domain: the fact that a homeomorphism between open sets of $\mathbb{R}^m$ and $\mathbb{R}^n$ exists implies $n=m$ is quite non-trivial. It's actually not quite invariance of domain, but the related fact that topological dimension (there are several equivalent definitions for separable metric spaces) is a topological invariant and $\dim(\mathbb{R}^n)=n$ (where $\dim$ is one of those equivalent topological dimension functions; topologically speaking the main fact we need for that is Brouwer's fixed point theorem.) This can be applied for simple topological manifolds, as all we need is homeomorphisms in the charts.
Now the chart open sets of the manifold form an open cover and so if $M$ is connected we can apply the chain characterisation of connectedness to conclude that the local dimension is constant: take any points $x$ of local dimension $m$ and $y$ of local dimension $n$. So there are finitely many chart open sets $U_1, \ldots, U_k$ such that $x \in U_1$, $y \in U_k$ and $U_i \cap U_{i+1} \neq \emptyset$ for all $i \in \{1,\ldots,k-1\}$. By the "overlapping chart sets implies equal dimension" argument you also gave and which is, as said, justified by non-trivial topological results on Euclidean spaces, we conclude that the local dimension of all $U_i$ is $m$: $U_1$ because of $x$, $U_2$ because it intersects $U_1$, $U_3$ because it intersects $U_2$ up to $U_k$ because it intersects $U_{k-1}$ (or make it a finite induction). So $m=n$ because local dimension is well-defined.
For $C^r$ manifolds we don't need the inverse function theorem, we know all charts maps are homeomorphisms by definition (or maybe you need the inverse function theorem to see that if it's not part of the definition?). You just fall back to the fact on topological manifolds.