8
$\begingroup$

By Definition, smooth manifolds are assumed to be Hausdorff and to satisfy the second countability axiom. I have heard (but never seen written) that these assumptions imply paracompactness (and thus the existence of a Riemannian metric by the well-known construction using Partition of unity). Does anybody know a reference or Proof for paracompactness?

$\endgroup$
6
  • 4
    $\begingroup$ The proof is in most introductory manifold theory textbooks, usually immediately before the construction of partitions of unity. Try Conlon's Differentiable Manifolds for example. $\endgroup$ May 12, 2012 at 16:09
  • $\begingroup$ More generally, a regular Lindelöf space is paracompact. This should be proved in general topology texts. $\endgroup$ May 12, 2012 at 16:39
  • 7
    $\begingroup$ Oh come on guys, it's all too easy to click on the close button instead of answering the question. I thought it was useful to give a self-contained answer. Yes, the answer in Conlon is similar, but even there, more spread out. $\endgroup$ May 12, 2012 at 16:53
  • 3
    $\begingroup$ @Greg, that seems like a rather narrow reading of what happened here. I'm having a hard time thinking of an introductory manifold theory textbook that does not cover this topic, one way or another. $\endgroup$ May 13, 2012 at 17:31
  • 3
    $\begingroup$ @Ryan Fair enough. Still, if MathOverflow builds up a Wikipedia-like library of answers, even those that appear in textbooks, that's not such a bad thing. $\endgroup$ May 20, 2012 at 2:24

1 Answer 1

28
$\begingroup$

Theorem: A countable atlas of charts for a Hausdorff $n$-manifold $M$ can be refined to a locally finite atlas. In fact, each chart only needs to be trimmed.

Proof: Let $U_1,U_2,\ldots$ be the charts. Each $U_i$, as a subset of $\mathbb{R}^n$, is the limit of a nested sequence of compact subsets $K_{i,1} \subseteq K_{i,2} \subseteq \ldots$. Since $M$ is Hausdorff, each $K_{i,j}$ is closed in $M$. So it suffices to delete $K_{1,i} \cup \cdots \cup K_{i-1,i}$ from $U_i$ to make a new chart $V_i$. Some of the $V_i$ might be empty, but this is no problem.

$\endgroup$
7
  • 1
    $\begingroup$ This does not seem to work. Take $M = \mathbb{R}$; let $U_i = \mathbb{R}$ for all $i$, mapped identically into $M$; let $K_{i, j} = K_j = [-j, -\frac{1}{j}] \cup [\frac{1}{j}, j]$, independently of $i$. Then $V_i$ is simply the complement of $K_i$, but the family formed by the $V_i$ is not locally finite, as they all contain $0$. Or did I miss something? $\endgroup$ Oct 29, 2019 at 10:48
  • 8
    $\begingroup$ The condition on the K's means that each U is the union of the corresponding K's; yours don't satisfy this requirement. Your revised example does clearly show that an emendation to the original argument is needed; the following seems to suffice: Require of the sets K that the interiors of those associated with any U exhaust the corresponding U. One easily sees that any given K--and so, a fortiori, the interior of any given K--meets only finitely many V's. Since any point belongs to the interior of some K, local finiteness of the covering by the V's is thus established. $\endgroup$ Oct 31, 2019 at 17:56
  • 1
    $\begingroup$ @IliaSmilga You need to take compact sets whose union is $U_i$, not just whose union is dense in $U_i$. $\endgroup$
    – Will Sawin
    Oct 31, 2019 at 18:11
  • 1
    $\begingroup$ Oh, sorry, my mistake. So let's take $K_j = [-j,0] \cup [\frac{1}{j},j]$. Then they do indeed exhaust $\mathbb{R}$; but the family of their complements $V_i$ is still not locally finite, since any neighborhood of $0$ intersects infinitely many of them. $\endgroup$ Oct 31, 2019 at 20:12
  • 2
    $\begingroup$ Equivalently, since we are working on $\mathbb{R}^n$, we can just change one symbol in the proof: instead of $\subset$ use $\Subset$ for the relation between the $K$s. $\endgroup$ Nov 1, 2019 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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