3

In p. 176 of John M. Lee's Introduction to smooth manifolds, 2nd edition, Lee defines vector fields along subsets $A \subseteq M$ of a smooth manifold with boundary $M$. The part that intrigues me is the fact that he requires a smooth local extension at each point by vector fields, not just any smooth maps. That makes you think there is a separate definition for smooth vector fields along arbitrary subsets. I've been thinking about it and I think the requirement of local extension at each point by vector fields is unnecessary.

I'll write briefly about that here in the more general context of vector bundles. I want to know (preferably from @Jack Lee) if I'm doing this correctly.

This is the definition that Lee suggest of smooth sections along arbitrary subsets.

Definition. Let $M$ be a smooth manifold with or without boundary, $\pi: E \to M$ a vector bundle of rank $k$, $A \subseteq M$ an arbitrary subset and $\sigma: A \to E$ a rough section, i.e. a map (not necessarily continuous) such that $\pi\circ\sigma = Id_A$. We say $\sigma$ is smooth if $\sigma$ extends to a smooth local section of E in a neighborhood of each point.

See Definition of smoothness to have some context about the definition of smoothness along arbitrary subsets.

And this is what I'm thinking.

Theorem 1. Let $M$ be a smooth manifold with or without boundary, $A \subseteq M$ an arbitrary subset and $k \in \mathbb{N}$. If $f: A \to \mathbb{R}^k$ is smooth then there exist an open neighborhood $U$ of $A$ and a smooth extension $\overline{f}: U \to \mathbb{R}^k$ of $f$.

Proof. Use partitions of unity.

Theorem 2. Let $M$ be a smooth manifold with or without boundary, $\pi: E \to M$ a vector bundle of rank $k$, $(\sigma_1, \dots, \sigma_k)$ a local frame for $E$ over the open subset $U \subseteq M$, $A \subseteq U$ an arbitrary subset and $\sigma: A \to E$ a rough section. Write $\sigma = \tau^i\sigma_i$ for some component functions $\tau^i: A \to \mathbb{R}$. Then $\sigma$ is smooth if and only if all $\tau^i$ are smooth. Furthermore, if $\sigma$ is smooth then there exists a neighborhood $W$ of $A$ and a smooth extension $\overline{\sigma}: W \to E$ of $\sigma$ which is a section such that $W \subseteq U$.

Proof. The first part is just following the definitions (but being careful of using the right definition of smoothness for arbitrary subsets of smooth manifolds with boundary). For the second part, assume $\tau = (\tau^1, \dots, \tau^k): A \to \mathbb{R}^k$ is smooth. By Theorem 1., there exists a neighborhood $W$ of $A$ and a smooth extension $\overline{\tau}: W \to \mathbb{R}^k$ of $\tau$ such that $W \subseteq U$. Now simply define $\overline{\sigma} = \overline{\tau}^i\sigma_i$. This is a smooth section and extends $\sigma$.

Theorem 3. Let $M$ be a smooth manifold with or without boundary, $\pi: E \to M$ a vector bundle of rank $k$, $A \subseteq M$ an arbitrary subset and $\sigma: A \to E$ a rough section. If $\sigma$ is smooth then there exists an open neighborhood $U$ of $A$ and a smooth extension $\overline{\sigma}: U \to E$ of $\sigma$ which is a section.

Proof. By Theorem 2., we can pick for each $p \in A$, a neighborhood $U_p$ of $p$ and a smooth extension $\overline{\sigma}_p: U_p \to E$ of $\sigma|_{A \cap U_p}$ which is a section. Now pick a smooth partitions of unity $(\psi_p)_{p \in A}$ subordinated to the open cover $(U_p)_{p \in A}$ of $U = \bigcup_{p \in A}U_p$. Now define \begin{equation*} \overline{\sigma}(x) = \sum_{p \in A}\psi_p(x)\overline{\sigma}_p(x), \end{equation*} for $x \in U$. This is a smooth section and extends $\sigma$.

Zero
  • 3,643
  • 26
  • 52
  • Can you write out the two definitions you want to compare? – Elchanan Solomon Mar 22 '23 at 01:42
  • @ElchananSolomon I just added it. I don't add any extra definitions. – Zero Mar 22 '23 at 01:49
  • I don't see your proposed alternative definition for a smooth section of the vector bundle. Can you disentangle it from your Theorem 1? – Elchanan Solomon Mar 22 '23 at 01:54
  • @ElchananSolomon It's just the usual definition for smoothness along arbitrary subsets. I'll add a link. – Zero Mar 22 '23 at 01:56
  • Do you want to work in the special case that $E$ is Euclidean space? – Elchanan Solomon Mar 22 '23 at 02:01
  • To elaborate on the prior point, and please tell me if I'm over-simplifying things, but the partition of unity trick will patch together the neighborhood sections using a sum weighted by the partition functions -- this sum makes sense in Euclidean space but not necessarily for arbitrary $E$. – Elchanan Solomon Mar 22 '23 at 02:06
  • @ElchananSolomon $E$ can be any vector bundle. You can multiply sections by real functions, it's just scalar multiplication. See p. 257 of Lee's Introduction to smooth manifolds. – Zero Mar 22 '23 at 02:09
  • Oh right, oops, this is a vector bundle not just a covering space. Hmm! Thanks for clarifying. – Elchanan Solomon Mar 22 '23 at 02:11
  • 3
    Yes, this looks fine. I just gave the definition I did in order to bypass this kind of construction for simplicity, because the existence of smooth local sections extending $\sigma$ is what's actually used in practice. – Jack Lee Mar 24 '23 at 19:42

1 Answers1

1

As confirmed by John M. Lee, what I've wrote in the question is fine, though maybe a bit wordy. Therefore, I will mark this question as answered.

Zero
  • 3,643
  • 26
  • 52