5
$\begingroup$

For this question a manifold-with-boundary is a topological space which is Hausdorff and locally upper-Euclidean. Every metrizable manifold-with-boundary has a collared boundary, as shown in "Locally flat imbeddings of topological manifolds", Morton Brown, 1962. Let $M$ be a non-metrizable manifold-with-boundary. Does $M$ have a collared boundary?

$\endgroup$
0

2 Answers 2

8
$\begingroup$

A nice recent reference for questions about non-metrisable manifolds is David Gauld's book aptly named "non-metrisable manifolds". For instance it is shown that any metrisable component of the boundary of a manifold (metrisable or not) is collared (Corollary 3.11 on page 44, it is an almost immediate consequence of R. Connely proof of M. Brown's result).

This is not true if the component is non-metrisable: Example 1.29 on p. 16 (which is originally due to P. Nyikos) gives a description of a manifold whose interior is $\mathbb{R}^2$ and whose boundary is the open long ray $\mathbb{L}_{+}$. This boundary component cannot be collared because there is no embedding sending $\mathbb{L}_{+}\times[0,1]$ into the manifold.

It is not clear to me whether the usual definition of `collared' is only for connected components of the boundary or if there is a global one asking for an embedding from $\partial M\times[0,1]$ into the manifold $M$. In that case a simpler counter example is the Prüfer surface (dating back to Rado), which is also detailed in D. Gauld's book in Example 1.25.

EDIT: Comments by the OP (and silly ones by myself) made me realize that something was a bit unclear in the statement of Theorem 3.10 in Gauld's book, from which Corollary 3.11 follows. Here are some details.

Theorem 3.10 states that if a closed subset $B$ of a Hausdorff space $X$ is locally collared and strongly paracompact, then $B$ is collared. But actually, what is needed is a bit stronger: that $B$ is strongly paracompact in $X$, that is, given a cover $\mathcal{U}$ of $B$ by open sets of $X$, there is another cover $\mathcal{V}$ of $B$ by open sets of $X$ such that each member of $\mathcal{V}$ is contained in a member of $\mathcal{U}$ and $\mathcal{V}$ is star-finite in $X$. (Star-finite means each member intersects finitely many members.) I actually asked D. Gauld about it, and he agreed that he has been a little bit careless in the statement.

This requirement is stronger, since for instance any manifold whose components are metrisable is strongly paracompact. The boundary $\partial P$ of the Prüfer surface $P$ is a discrete union of continuum many real lines and hence is strongly paracompact (in itself, so to say). But $\partial P$ is not strongly paracompact in $P$, and actually $\partial P$ is not collared in $P$.

A consequence of Theorem 3.10 (amended) is that if the boundary $\partial M$ of a manifold $M$ is made of (at most) countably many metrisable components, then it is collared. The proof is by using the fact that Lindelöfness is equivalent to metrisability and strong paracompactness for connected manifolds to cover $\partial M$ by countably many euclidean sets whose union yields a strongly paracompact submanifold of $M$ also having $\partial M$ as a boundary.

$\endgroup$
10
  • $\begingroup$ Nice! I've been trying to see whether connectedness is used at all in the proofs, and it seems to me that the only reason the theorem mentions components is because Gauld defines a manifold as connected. So perhaps one could obtain a collar for the whole boundary (assuming metrizable boundary). But perhaps I'm wrong; I'll take a closer look. $\endgroup$
    – kaba
    Commented Oct 19, 2019 at 19:19
  • $\begingroup$ I think that he mentions components because the answer may depend on which ones or how many you take. For instance a manifold may have metrisable and non-metrisable boundary components. In the example of the Prüfer surface they are all metrisable, but if you take only the interior and countably many boundary components then the manifold is metrisable and the boundary (as a whole) is collared, and if you take all of them (that is, continuum many), the boundary as a whole is not collared though each boundary component, individually, is collared. I might not be very clear (cont) $\endgroup$ Commented Oct 19, 2019 at 19:54
  • $\begingroup$ (cont) but if you look at the definition of the Prüfer surface, you should see what I mean (I hope). $\endgroup$ Commented Oct 19, 2019 at 19:56
  • $\begingroup$ The trick is that the boundary of the Prüfer surface is a discrete collection of continuum many real lines, and hence as a topological space it is metrisable (points in different components may be thought to be at fixed distance $1$). So I believe that Gauld tries to avoid this difficulty by talking only about components. What is needed is the property that the boundary is strongly paracompact (see Theorem 3.10), which is equivalent to metrisability for connected manifolds but not in general. $\endgroup$ Commented Oct 19, 2019 at 20:04
  • $\begingroup$ According to this paper, also by Gould, a "paracompact locally compact space is strongly paracompact". If that holds, then I think that paracompactness, strong paracompactness, and metrizablity are equivalent even for non-connected manifolds, right? $\endgroup$
    – kaba
    Commented Oct 19, 2019 at 20:56
1
$\begingroup$

Mathieu Baillif provided an answer to the question. I'll leave here some more notes for future readers. The following strengthening of the theorem in Gauld's book "Non-metrizable manifolds" holds:

Let $X$ be a manifold-with-boundary. Then $\partial X$ is collared in $X$ and $\partial X$ is paracompact in $\partial X$ if and only if $\partial X$ is paracompact in $X$.

This is quite a nice theorem. However, it leaves open the characterization of of non-paracompact collared boundaries. I would be interested to hear if someone can complete the characterization.

The following example shows that there exists non-paracompact collared boundaries. The open long ray is a non-paracompact manifold. Therefore, its cartesian product with the interval $[0, 1]$ is a manifold-with-boundary whose boundary is non-paracompact and clearly collared.

Some background

An $X$-collar of $X' \subset X$ is a function $h : X' \times [0, 1] \to X$, such that

  • $h$ is an embedding,
  • $h(x, 0) = x$, for each $x \in X'$,
  • $h(X' \times [0, 1))$ is open in $X$

A subset $X' \subset X$ is $X$-collared, if there exists an $X$-collar of $X'$.

A local collar of $X' \subset X$ in $X$ is a pair $(U, h)$, where $U \subset X'$ is open in $X'$, and $h : \overline{U}(X') \times [0, 1] \to X$ is an embedding such that

  • $h^{-1}(X') = \overline{U}(X') \times \{0\}$,
  • $h(x, 0) = x$, for each $x \in \overline{U}(X')$,
  • $h(U \times [0, 1))$ is open in $X$.

A subset $X' \subset X$ is locally collared in $X$, if for each $x \in X'$ there exists a local collar $(U, h)$ of $X'$ in $X$ such that $x \in U$.

Here are some properties of collared subsets, which I think to have proved:

  • If $X' \subset X$ is $X$-collared, then $X' \times Y$ is $(X \times Y)$-collared. Applied to manifolds-with-boundary, if $X$ is a manifold-with-boundary, $Y$ is a (boundaryless) manifold, and $\partial X$ is $X$-collared, then $X \times Y$ is a manifold-with-boundary, and $\partial (X \times Y) = \partial X \times Y$ is $(X \times Y)$-collared.
  • If $X'_i \subset X_i$ is $X_i$-collared for each $i \in I$, then $\sqcup X'_I$ is $\sqcup X_I$-collared, where $\sqcup$ denotes disjoint sum.
  • Let $X' \subset X$, $\mathcal{U} \subset \mathcal{P}(X')$ be an $X'$-open cover of $X'$, $V_U = \overline{U}(X') \times [0, 1]$ for each $U \in \mathcal{U}$, and $h : X' \times [0, 1] \to X$. Then $h$ is a collar of $X'$ in $X$ and $\mathcal{U}$ is $X'$-locally finite if and only if $h$ is injective, $(U, h|V_U)$ is a local collar of $X'$ in $X$ for each $U \in \mathcal{U}$, and $\{h(V_U) : U \in \mathcal{U}\}$ is $X$-locally finite.
$\endgroup$
14
  • $\begingroup$ I'll add that your question got me thinking because I noticed as well that there are cases where we cannot ensure that the collar is a closed subset of the manifold in the non-metrisable case. Connelly's and Brown's definitions of collars differ on this point, as the former includes closedness while the latter does not. I found these subtle points interesting enough to prepare a small note (for myself mainly) on the subject summarizing what I know. I might publish it when I am done writing it if some people are interested. (Nothing in the note is really new, just compiled in one place.) $\endgroup$ Commented Nov 19, 2019 at 19:28
  • $\begingroup$ So, thanks for asking the question in the first place. $\endgroup$ Commented Nov 19, 2019 at 19:28
  • $\begingroup$ Do you mean that there is a Brown-collar (collar being the embedding function, rather than its image) which cannot be extended to a Connelly-collar? (By definition, if I've understood correctly, the image of a Brown-collar is open, and the image of a Connelly-collar is closed.) What would be an example of this? Do link the notes if you end up writing them. $\endgroup$
    – kaba
    Commented Nov 19, 2019 at 23:07
  • $\begingroup$ Yes, that's exactly what happens: you can have an embedding of $\partial M\times [0,1)$ which gives a neighborhood of the boundary, but none that is a closed embedding. Such an example is given by the product of the longray with $[0,1)$ with $\omega_1\times\{0\}$ removed. If you know what these spaces are, it is not difficult to check that it is the case. If not, then maybe I'll try to finish writing up these notes and make them available. Or you can check Gauld's book where the longray (and longline) are discussed in detail. $\endgroup$ Commented Nov 20, 2019 at 8:48
  • $\begingroup$ @MathieuBaillif Thanks for the example; I could only now get back to this. Can you spell out the product in more detail? The sets don't seem to match up. Say $\mathbb{L}_{\geq 0} = \omega_1 \times [0, 1)$, and $A = \omega_1 \times \{0\} \subset \mathbb{L}_{\geq 0}$. Is it $(\mathbb{L}_{\geq 0} \setminus A) \times [0, 1)$ or something else? $\endgroup$
    – kaba
    Commented Nov 27, 2019 at 7:45

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .