2

Suppose $X, Y$ are smooth manifolds, $Z \subset X$ a compact submanifold and $f: X \rightarrow Y$ a smooth map such that the restriction of $f$ to $Z$ is an injective smooth immersion. As $Z$ is compact, the restriction is actually a smooth embedding.

Is it true that $f$ is a smooth embedding in some open neighbourhood of $Z$, and if so, how can this be seen?

Paul
  • 1,232
  • 2
    Not really, let $Z \subset X = Z\times \mathbb R$ (as zero section), and $f:Z\times \mathbb R \to Y=Z$ be the projection. –  Nov 11 '13 at 08:44
  • True if and only if $dim(Y)\ge dim X$. – Moishe Kohan Nov 11 '13 at 09:43
  • Thanks for the counterexample, I didn't catch that. The dimension condition actually holds for my purposes - so could you give me a pointer in the right direction? – Paul Nov 12 '13 at 06:31
  • Paul: I do not know a reference, but I will write a proof when I have time. However, I realized that there is one more obstruction, coming from the normal bundles. – Moishe Kohan Nov 12 '13 at 07:05
  • The extra condition is that the map $Z\to Y$ should extend to a monomorphism of normal bundles. Otherwise, there is again a counter-example. – Moishe Kohan Nov 12 '13 at 07:19

1 Answers1

2

First, an example: Let $X$ denote the (open) Moebius band, $Y=S^1\times {\mathbb R}$ and let $Z$ denote the "core curve'' in $X$, i.e., if we think of $X$ as the total space of a nontrivial line bundle over $S^1$, then $Z$ is the image of the zero section. Now, let $p: X\to Z$ denote the natural projection and $i: Z\to Y$ the embedding $i: S^1\to S^1\times (0) \subset S^1\times {\mathbb R}$. Then $f=i\circ f$ is a smooth map which is the embedding $i$ when restricted to $Z$. However, the map $i$, clearly, does not extend to an embedding of a neighborhood of $Z$ in $X$.

Now, the correct statement is:

Lemma. An embedding $i: Z\to Y$ extends to an embedding $h$ of a neighborhood of $Z$ into $Y$ if and only if the map $i$ extends to a monomorphism $\mu$ of normal bundles $\nu_Z\to \nu_{i(Z)}$ where the first normal bundle is in $X$ and the second is in $Y$.

Proof. If $h$ exists, then its differential $dh$ induces a monomorphism of normal bundles. Conversely, suppose that $\mu$ exists. I will assume that $X, Y$ are equipped with Riemannian metrics. We need the following standard fact:

Let $M\subset N$ be a compact submanifold and $N$ is Riemannian, let $\nu_M$ denote the normal bundle. Let $exp_M: \nu_M\to N$ denote the normal exponential map: It sends the normal vector $v$ to the point $exp(v)$. Then there exists $r>0$ such that the restriction of $\exp_M$ to $B_r(\nu_M))$ is a diffeomorphism to its image. Here $B_r(\nu_M))$ consists of normal vectors of lengths $<r$.

Now, define $h$ to be the composition $\exp_{i(Z)}\circ \mu \circ (\exp_{Z})^{-1}$, defined on a sufficiently small open neighborhood $U$ of $Z$ in $X$. Then $h$ is a diffeomorphic extension of $i$ (provided that $U=exp_Z(B_r(\nu_Z))$ with $r$ sufficiently small. qed

Edit: Assume that $$ dim(Z) + dim(X)\le dim(Y). $$ Then there exists a monomorphism of normal bundles $$ \nu_Z\to \nu_{i(Z)}. $$ This follows immediately from the Existence Theorem in this paper, since the inequality of dimensions implies vanishing of the bordism group where the obstruction to existence of a monomorphism lies. In particular, in this range of dimensions, the map $i$ extends to an embedding of a tubular neighborhood of $Z$. Note that in my example the above inequality fails by 1, so it is sharp.

Moishe Kohan
  • 97,719
  • Neat, thank you very much. About the formulation of the lemma: by "extends to a monomorphism of normal bundles", wouldn't this be satisfied by any map with an injective differential sending normal vectors to normal vectors? Maybe I am completely off track here - I am completely new to bundles - but this condition would hold for what I'm trying to do. In the counterexample, the normal vectors are sent to zero, so this would be violated. (I could also apply the dimension criterion - which is interesting on it's own - but I'm trying to keep it as much down to earth as I can). – Paul Nov 13 '13 at 20:31
  • @Paul: Yes, this is correct, if $df$ is injective on $TX$ along $Z$ then monomorphism of normal bundles is automatic. Hence, you do not need the dimension assumptions in this case. – Moishe Kohan Nov 13 '13 at 21:30