3

Problem: Prove that $S^n$ the n-sphere is homeomorphic to $\partial I^{n+1}$ the unit cube. Show that the unit cube admits a smooth structure hence can be turned into a smooth manifold("even though it has corners"). Generalize this by proving that any topological manifold homeomorphic to a smooth manifold admits a smooth structure.

The unit cube is the equivalent of unit sphere for the "max metric" so it can be written as $\partial I^{n+1}=\{x\in \mathbb{R}^n:||x||_{\infty}\leqslant 1\}$ while the n-sphere can be written using the euclidean norm $S^n=\{x\in\mathbb{R}^n:||x||\leqslant1\}$.

For the first part I thought of the following map that is homeomorphic(the intuition came from the fact the two metrics generated by the norms are equivalent):

$f:S^{n}\to \partial I^{n+1}\\ \:\:\:\:\:\:x\longrightarrow x\frac{||x||_{\infty}}{||x||}$

f is bijective, continuous(once the norms are continuous and the product of continuous functions are continuous)

$f^{-1}=g:I^{n+1}\to S^n \\ \:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:y\longrightarrow y\frac{||y||}{||y||_{\infty}}$

By the same argument g is bijective and continuous hence f is a homeomorphism.

To prove that $\partial I^{n+1}$ is a smooth manifold I thought of defining a chart of the following kind $f\circ\phi$ where $\phi:U\subset S^n\to\mathbb{R}^n $ is an diffeomomrphism from a subset of $\partial I^{n+1}$. However for this chart $f\circ\phi$ to be smooth I would need $f$ to be smooth which is not necessarily true once it is only required for $f$ to be a homeomorphism.

Question:

How can a topological manifold homeomorphic to a smooth manifold gain a smooth structure? How would it be proved?

Thnaks in advance!

Pedro Gomes
  • 4,099
  • I think that this need not be true. Mainly, it is because homeomorphisms do not guarantee smoothness. Since smoothness of manifolds is defined in terms of transition maps, my first attempt at solving this question would be to use the known homeomorphism and get the whole setting into the smooth structure we have. However, while shifting this setting, we may destroy smoothness. I do believe, however, that a diffeomorphism might just work! I will have to work on it. – Aniruddha Deshmukh Mar 07 '20 at 16:19
  • @AniruddhaDeshmukh My thoughts are the same. Thanks for your comment! – Pedro Gomes Mar 07 '20 at 16:24
  • Your idea for defining charts of $\partial I^{n+1}$ to be of the form $f \circ \phi$ is correct. Where you went wrong is thinking that the map $f \circ \phi$ must be smooth respect to the coordinates on its domain $U \subset \partial I^{n+1} \subset \mathbb R^{n+1}$ and the coordinates on its domain $\mathbb R^n$; that's not what a smooth structure requires. See my answer for more details. – Lee Mosher Mar 07 '20 at 16:37

1 Answers1

8

Perhaps it is worthwhile recalling the definition of a smooth structure on a manifold: it is an atlas with smooth overlap maps. To be precise, a smooth structure on an $n$-dimensional manifold $N$ means an atlas of coordinate charts of the form $$\mathcal A = \{(\phi_i,U_i) \mid i \in I\} $$ such that for each $i \in I$, $\phi_i : U_i \to \mathbb R^n$ is a homeomorphism defined on an open subset $U_i \subset N$, and each overlap map $$\phi_j \circ \phi_i^{-1} : \phi_i(U_i \cap U_j) \to \phi_j(U_i \cap U_j) $$ is smooth. Notice that I did not say "$\phi_i$ is a smooth map" because that's not the requirement. The key requirement of a smooth atlas is not the smoothness of the maps defining the coordinate charts, but instead the requirement is smoothness of the overlap maps.

If $g : M \to N$ is any homeomorphism, then using $g$ together the smooth atlas $\mathcal A$ for $N$, we get the following smooth atlas for $M$, $$g^*\!\mathcal A = \{(\phi_i \circ g, g^{-1}(U_i) \mid i \in I\} $$ You can easily check that formally this does indeed satisfy the definition of a smooth atlas for $M$.

So yes, $M$ has a smooth structure. Furthermore, using this atlas, the homeomorphism $g$ becomes a diffeomorphism.

Lee Mosher
  • 135,265
  • I thought very hard about this insightful view and how to concile it with the fact that the unicity of differential structures of a manifold (up to diffeomorphism) holds only when the dimension is 3 or less. There has to be no conflict and I have tried to formalize this and clarify it for myself in this question . Would you be so kind as to have a look when you have time? thanks in advance – latelrn May 14 '21 at 23:25