I am reading Lee's Introduction to Smooth Manifolds, 2nd edition, and happen to read a sentence in a proof for Proposition 5.2 goes like this:

With this smooth structure on $S$, the map $F$ is a diffeomorphism onto its image, and this is obviously the only smooth structure with this property.

Perhaps I missed something from previous chapters, but this is not obvious to me at first sight. Maybe I would prove the following:

Say $F:M\to N$ is a homeomorphism between two topological manifolds. We endow $M$ with only one smooth structure, and $N$ with two smooth structures $C_1$ and $C_2$, such that $F$ is a diffeomorphism in both cases. We would show that $C_1=C_2$. Due to symmetric footing, it suffices to show $C_1\subseteq C_2$.

Observe $id=F\circ F^{-1}:(S,C_1)\to (S,C_2)$ is a diffeomorphism. To show $C_1\subseteq C_2$, I think this could be stated and proved more generally:

For a diffeomorphism $f:X\to Y$ and any smooth chart $(U,\varphi)$ on $X$, we have that $(f(U),\varphi\circ f^{-1})$ is a smooth chart on $Y$.

Now let $(U,\varphi)$ be a smooth chart on $X$ and $(V,\psi)$ be a smooth chart on $Y$. We are to show $\psi\circ f\circ\varphi^{-1}:\varphi\circ f^{-1}(f(U)\cap V)\to\psi(f(U)\cap V)$ and $\varphi\circ f^{-1}\circ\psi^{-1}:\psi(f(U)\cap V)\to\varphi\circ f^{-1}(f(U)\cap V)$ are smooth as maps between subsets of $\Bbb R^n,\Bbb R^m$, where $f(U)\cap V$ is nonempty for non-triviality.

For a point $p\in f^{-1}(f(U)\cap V)$, by smoothness of $f$, pick a chart $(U_p,\varphi_p)$ containing $p$, a chart $(V_{f(p)},\psi_{f(p)})$ containing $f(p)$, such that $f(U_p)\subseteq V_{f(p)}$, and $\psi_{f(p)}\circ f\circ\varphi_p^{-1}:\varphi_p(U_p)\to\psi_{f(p)}(V_{f(p)})$ is smooth. Impose also that $U_p\subseteq f^{-1}(f(U)\cap V)$; if not, we can use the chart $(U_p\cap f^{-1}(f(U)\cap V),\varphi|_{U_p\cap f^{-1}(f(U)\cap V)})$ in place of $(U_p,\varphi_p)$, which still satisfies the above conditions.

Now observe that the composition of smooth maps $(\psi\circ\psi_{f(p)}^{-1})\circ(\psi_{f(p)}\circ f\circ\varphi_p^{-1})\circ(\varphi_p\circ\varphi^{-1})$ is smooth. Cancellation gives that the composition is smooth on the domain of definition. We are to check that the domain is a neighbourhood of $\varphi(p)$.

To see this, observe the condition $U_p\subseteq f^{-1}(f(U)\cap V)$ gives $U_p\subseteq U$, so that the transition map $\varphi_p\circ\varphi^{-1}$ has domain and codomain $\varphi(U_p)$ and $\varphi_p(U_p)$ respectively; the conditions $U_p\subseteq f^{-1}(f(U)\cap V)$ and $f(U_p)\subseteq V_p$ gives $f(U_p)\subseteq V_{f(p)}\cap V$, so that the map $\psi_{f(p)}\circ f\circ\varphi_p^{-1}$ has range being a subset of $\psi_{f(p)}(V_{f(p)}\cap V)$; the transition map $\psi\circ\psi_{f(p)}^{-1}$ has domain $\psi_{f(p)}(V_{f(p)}\cap V)$. Hence, $(\psi\circ\psi_{f(p)}^{-1})\circ(\psi_{f(p)}\circ f\circ\varphi_p^{-1})\circ(\varphi_p\circ\varphi^{-1})$ is defined on the whole neighbourhood $\varphi(U_p)$ of $\varphi(p)$. Since this function $\psi\circ f\circ\varphi^{-1}$ is locally smooth around each point $\varphi(p)\in\varphi\circ f^{-1}(f(U)\cap V)$, it is smooth on the whole set $\varphi\circ f^{-1}(f(U)\cap V)$.

Smoothness of $\varphi\circ f^{-1}\circ\psi^{-1}$ is proved similarly: for each point $p'\in f(U)\cap V$, choose a chart $(V_{p'},\psi_{p'})$ containing $p'$, a chart $(U_{f^{-1}(p')},\varphi_{f^{-1}(p')})$ containing $f^{-1}(p')$, such that $f^{-1}(V_{p'})\subseteq U_{f^{-1}(p')}$, and $\varphi_{f^{-1}(p')}\circ f^{-1}\circ\psi_{p'}^{-1}$ is smooth. Impose $V_{p'}\subseteq f(U)\cap V$. We can then prove $\varphi\circ f^{-1}\circ\psi^{-1}$ is locally smooth around $\psi(p')$ by the same arguments as above.

Is the above proof correct? Is there a shorter proof for the above claim? Or is this proved in the book but I overlooked?

Edit: I will include the whole statement of Proposition 5.2 here.

Suppose $M$ is a smooth manifold with or without boundary, $N$ is a smooth manifold, and $F:N\to M$ is a smooth embedding. Let $S=F(N)$. With the subspace topology, $S$ is a topological manifold, and it has a unique smooth structure making it into an embedded submanifold of $M$ with the property that $F$ is a diffeomorphism onto its image.

  • $\begingroup$ It seems like you may be establishing that there is only one smooth structure in a diffeomorphism class on a manifold, which is, I believe, false. .. btw is there any more info on $F $ and $S $ in your example $\endgroup$
    – user403337
    Aug 21, 2017 at 7:17
  • $\begingroup$ @ChrisCuster $F$ is simply required to be a fixed diffeomorphism. $\endgroup$
    – edm
    Aug 21, 2017 at 7:20
  • $\begingroup$ What about $S $? $\endgroup$
    – user403337
    Aug 21, 2017 at 7:30
  • $\begingroup$ @ChrisCuster do you mean the $S$ from the book? $\endgroup$
    – edm
    Aug 21, 2017 at 7:31
  • $\begingroup$ yeah. Is it an arbitrary smooth manifold; or anything more specific? $\endgroup$
    – user403337
    Aug 21, 2017 at 7:34

2 Answers 2


Here is a possible reason why it could be okay to call it 'obvious'. It is a straight forward combination of three basic observations. Let me use the provisory definition that a topological chart on a topological space $X$ is a pair $(U,\varphi)$ of an open subset $U\subset X$ and a homeomorphism $\varphi\colon U\to U'$ onto any open subset $U'$ of some $\mathbb{R}^n$.

  1. Let $U,V,W$ be smooth manifolds. If $f\colon U\to V$ is a diffeomorphism, then a map $g\colon V\to W$ is a diffeomorphism if and only if $g\circ f\colon U\to W$ is a diffeomorphism.

  2. The maximal atlas of a smooth manifold $X$ is the set of topological charts $(U,\varphi)$ on $X$ where $\varphi\colon U\to U'$ is a diffeomorphism. (Note that this is somehow recursive since we're taking $U$ with the differentiable structure naturally inherited from $X$; this is exactly the trick here.)

  3. Let $f\colon X\to Y$ be a homeomorphism between topological manifolds. For each topological chart $(U,\varphi)$ on $X$ we get a topological chart $f_*(U,\varphi):=(f(U),\varphi\circ f^{-1}|_{f(U)})$ on $Y$ and each topological chart on $Y$ is of this form since the process can be inverted by mapping a topological chart on $Y$, $(V,\psi)$, to the topological chart $f^*(V,\psi) := (f^{-1}(V),\psi\circ f|_{f^{-1}(V)})$ on $X$.

In combination, we see that if $f\colon X\to Y$ is a diffeomorphism, then we can recover the maximal atlas $A_Y$ (hence, the smooth structure) of $Y$ from the maximal atlas $A_X$ of $X$ as follows. A topological chart $(V,\psi)$ on $Y$ is actually a chart, i.e., $(V,\psi)\in A_Y$, if and only if $\psi$ is a diffeomorphism; but this is the case if and only if $\psi\circ f|_{f^{-1}V}$ is a diffeomorphism and this in fact precisely means $f^*(V,\psi)\in A_X$. Thus, we get a pair of inverse bijections $f_*\colon A_X\to A_Y$ and $f^*\colon A_Y\to A_X$. In particular, the smooth structure on $Y$, that is, $A_Y =\{f_*(U,\varphi)\mid(U,\varphi)\in A_X\}$, is uniquely determined by $(X,A_X)$ and $f$.


The embedded submanifold $S $ inherits a unique smooth structure from the manifold $M $. With this structure the inclusion $\mathcal i:S \to M $ is an embedding. Note, however, there certainly can be more than one atlas defining this structure. These atlases are called compatible...

It's clear that if we took a different smooth structure, that is, one defined by atlases not smoothly equivalent to the above, $F $ would not be a diffeomorphism anymore. ..

So, say $\phi_\alpha \circ \psi_\beta^{-1} $ isn't a diffeomorphism .. then $F\circ \psi_\beta^{-1} $ isn't either. .. because, say $g=F\circ \psi_\beta^{-1} $ is a diffeomorphism. .. then $\phi_\alpha \circ F^{-1}\circ g=\phi_\alpha \circ \psi_\beta^{-1}$ would be a diffeomorphism. ..

And I now think you're right, he didn't mean up to diffeomorphism. Incompatible maps can define diffeomorphic smooth structures, as in @Ben's example. .. ($(\mathbb R, id)$ and $(\mathbb R, x\to x^3)$)

  • $\begingroup$ I am aware that the uniqueness of smooth structure is under the assumption that $F$ has to be a diffeomorphism. What is not obvious is the statement "all smooth structures on the image $F(S)$ that make $F$ a diffeomorphism must be equal to each other". $\endgroup$
    – edm
    Aug 8, 2017 at 5:03
  • $\begingroup$ I don't think the author means "up to diffeomorphism". He wants to make a fixed mapping $F$ to be a diffeomorphism. $\endgroup$
    – edm
    Aug 21, 2017 at 7:32
  • $\begingroup$ I'm aware of the function $F $. I meant diffeomorphic atlases (smooth structures), which would make $F $ a diffeomorphism. .. $\endgroup$
    – user403337
    Aug 21, 2017 at 9:41
  • 1
    $\begingroup$ An easy example for the claim that there can be incompatible but diffeomorphic atlases: Take $\mathbb{R}$ once with the most obvious chart $(\mathbb{R},\mathrm{id})$ and once with the chart $(\mathbb{R},\varphi)$ where $\varphi\colon\mathbb{R}\to \mathbb{R}$ is the map $x\mapsto x^3$. The two charts are incompatible but there is still a diffeomorphism between them, namely, $\varphi\colon (\mathbb{R},\varphi)\to (\mathbb{R},\mathrm{id})$. $\endgroup$
    – Ben
    Aug 21, 2017 at 10:16
  • $\begingroup$ @Ben This is getting further and further away from what I am asking... To respond to ChrisCuster's addendum, I don't think I want to use results that comes from latter pages because the author does not seem to postpone the proof as he writes "...is obviously the only smooth structure...". $\endgroup$
    – edm
    Aug 21, 2017 at 10:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.