10

A paracompact space is a space in which every open cover has a locally finite refinement.

A compact space is a space in which every open cover has a finite subcover.

Why must the product of a compact and a paracompact space be paracompact?

I really have very little intuition about how to go about this question, so any hints or a proof would be greatly appreciated.

Mark
  • 6,038
  • @Mark: "two such spaces" is unclear; do you mean, as your title suggests, a product of a paracompact and a compact space? Or do you mean, as "two such spaces" suggests, the product of two paracompact, or of two compact spaces? Or all of the above? – Arturo Magidin Jan 02 '11 at 23:00
  • @Arturo: Sorry, edited for clarity. – Mark Jan 02 '11 at 23:13
  • 1
    @Mark: The result as stated is false, because you are asking the paracompact space to be Hausdorff but not the compact one. Take a product of a paracompact space with an indiscrete finite space; it will be non-Hausdorff, hence not paracompact by your definition. You should either ask for Hausdorff in both, or in neither (both are definitions that are used by some). – Arturo Magidin Jan 02 '11 at 23:31
  • @Arturo: Why would the product of a paracompact and indiscrete space be non-Hausdorff? (Intuitively it seems that the product of a Hausdorff space and any other space should be Hausdorff) – Mark Jan 02 '11 at 23:39
  • Haha, they named a topology after me: the indiscreet topology. – deeeez Jan 02 '11 at 23:48
  • 3
    @Mark: Take the underlying sets of $X$ and $Y$ to be ${x,y}$, give $X$ the discrete topology, $Y$ the indiscrete topology. The open sets of $X\times Y$ are $\emptyset$, ${(x,x),(x,y)}$, ${(y,x), (y,y)}$, and $X\times Y$. What are the disjoint neighborhoods of $(x,x)$ and $(x,y)$? The error in your intuition is that $(a,b)\neq(c,d)$ does not imply $a\neq c$. – Arturo Magidin Jan 02 '11 at 23:53
  • @Arturo: That is a good counter-example to the question as stated. It looks like X is paracompact and Y is compact, and yet their product is non-Hausdorff. I will remove the requirement that paracompactness -> Hausdorff. – Mark Jan 03 '11 at 00:07
  • @Mark: Yes, both $X$ and $Y$ are compact, hence $X$ is paracompact. A product of two spaces (with the product topology) is Hausdorff if and only if both spaces are Hausdorff, so you can either add the condition in both, or drop it in both. Many authors require Hausdorff-ness for compactness, and many authors do not require it for paracompactness, so it's six of one, half a dozen of the other. – Arturo Magidin Jan 03 '11 at 00:09

3 Answers3

8

The key idea is indeed to use the "tube lemma", as d.t. did in his solution, but there is a small gap in his solution.

(This proof does not assume prior knowledge of tube lemma)

Let $X$ be paracompact and $Y$ be compact. Let $\mathcal{A}$ be an open cover of $X\times Y$.

(Tube lemma part) First fix $x\in X$, and for each $y \in Y$, find $A\in\mathcal{A}$ and basis element $U\times V$ such that $(x,y)\in U\times V\subseteq A$. As $y$ ranges in $Y$, these various $U\times V$ cover $\{x\}\times Y$, which is compact. Thus there exists finitely many $U_1\times V_1 \subseteq A_1,\dots,U_n\times V_n\subseteq A_n$ that cover $\{x\}\times Y$. Let $U_x = U_1\cap \dots \cap U_n$. For later use, let $\mathcal{A}_x=\{A_1,...,A_n\}$.

Now, $\{U_x\}_{x\in X}$ is an open cover of $X$. Using paracompactness, there is an locally finite open refinement $\mathcal{B}$ that covers $X$. For purpose of notation, suppose $B_i,i\in I$ are the elements of $\mathcal{B}$. Using the refinement property, for each $i\in I$, pick $x_i\in X$ such that $B_i\subseteq U_{x_i}$.

Consider the open refinement $\mathcal{C}$ of $\mathcal{A}$ given by

$$\mathcal{C_{x_i}}:=\{A\cap (B_i\times Y)\}_{A\in \mathcal{A}_{x_i}},\quad \mathcal{C}:=\bigcup_{i\in I}\mathcal{C}_{x_i}$$

To prove that this is a cover, consider any $(x,y)\in X\times Y$. First $x$ is in some $B_i$. Since $\mathcal{C}_{x_i}$ covers $B_i \times Y$, $(x,y)$ is covered by $\mathcal{C}$.

To prove that it is locally finite, consider any $(x,y)\in X\times Y$. First there exists an open neighbourhood $U\subseteq X$ of $x$ that intersects only finitely many elements of $\mathcal{B}$, say $B_1,...,B_m$. Then $U\times Y$ is the desired neighbourhood of $(x,y)$ that only intersects finitely many elements of $\mathcal{C}$ as it can only intersect elements from $\mathcal{C}_{x_1},...,\mathcal{C}_{x_m}$, each of which is a finite collection.

suncup224
  • 2,859
5

I think Trevor is right. Here are the details (this is an adaptation of the classical proof that the product of two compact spaces is a compact space that you can find in Munkres, for instance).

Let $X$ be a paracompact space, $Y$ a compact one and ${\cal U}$ an open cover of $X \times Y$.

For any $x \in X$, the slice $\left\{ x \right\} \times Y$ is a compact space and ${\cal U}$ an open cover of it (as a subspace). So it admits a finite subcover $U_{x,1}, \dots , U_{x,n_x} \in {\cal U}$. Let us call $N_x = U_{x,1} \cup \dots \cup U_{x,n_x}$ their union.

So $N_x$ is an open set that contains the slice $\left\{ x \right\} \times Y$. Because of the tube lemma, there exists an open set $W_x \subset X$ such that

$$ \left\{ x \right\} \times Y \quad \subset \quad W_x \times Y \quad \subset \quad N_x \ . $$

Now, those $W_x$ form an open cover ${\cal W}$ of $X$. By hypothesis, there is a locally finite refinement ${\cal W}' = \left\{ W_{x_i}\right\}$, $i\in I$ for some index set $I$.

Consider the following subcover of ${\cal U}$:

$$ {\cal U}' = \left\{ U_{x_i,j}\right\} \ , $$

with $i\in I$ and $j = 1, \dots , n_{x_i}$.

Let us show that ${\cal U}'$ is a locally finite subcover of ${\cal U}$: take any point $(x,y) \in X \times Y$. By hypothesis, there is a neighborhood $V \subset X$ of $x$ such that $V$ interesects only finitely many of the sets of ${\cal W}'$. Then $V\times Y$ is a neighborhood of $(x,y)$ that intersects only finitely many of the sets of ${\cal U}'$.

Agustí Roig
  • 18,489
  • You have the step:

    Now, those Wx form an open cover W of X. By hypothesis, there is a locally finite refinement W′={Wxi}

    Here, it seems that you are assuming a finite subcover, not a locally finite refinement.

    Also, for the last step, what if the set V contains infinitely many points? Then it seems it contains infinitely many slices of the form {x}*Y, and each slice intersects at least one neighborhood in your refinement.

    – Mark Jan 03 '11 at 16:24
  • Actually, I get it now, thanks! – Mark Jan 04 '11 at 01:16
  • @Mark. Thank you for saving my time! :-) – Agustí Roig Jan 04 '11 at 02:14
  • 2
    Are you sure the last sentence is correct? I fail to see how this holds true (might completely be on me, I'd just like to understand, then): From what I understand all we can conclude is that $V \times Y$ is contained in finitely many of the sets of $\mathcal{U}'$; however, I don't get how we can assume that other elements of $\mathcal{U}'$ have empty intersection with $V \times Y$. I'm thinking of $W_x \subsetneq N_x,,;x \in X$... – polynomial_donut Jun 01 '17 at 19:30
  • 3
    Just found a proof at https://topospaces.subwiki.org/wiki/Compact_times_paracompact_implies_paracompact ; there an important last step is included, which I would say is missing here. However, they are a bit sloppy in point 5, as they miss to state explicitly that they are only looking at finitely many intersections for each $P$. – polynomial_donut Jun 01 '17 at 21:15
  • I agree with polynomial_donut; I believe there is an issue with this proof. After the sentence "By hypothesis, there is a locally finite refinement $W′={W_{x_i}}, i\in I$ for some index set $I$", I believe you cannot just take the subcover $\mathcal{U}'$ that is in your proof. Rather, you should consider the refinement $\mathcal{U}' :={U_{x_i,j} \cap (W_{x_i}\times Y)}$. Without the additional intersection, the last sentence of your proof does not follow from the 2nd-to-last sentence. – suncup224 Sep 30 '17 at 03:28
  • 2
    The fix I proposed also misses a step, so I decided I should write up a solution instead – suncup224 Sep 30 '17 at 03:36
  • 1
    I think this solution is incorrect. You've shown that every open cover has a locally finite subcover, which is strong enough to imply compactness. But this is not always the case, as $\mathbb R \times [0,1]$ is paracompact but not compact. – YuiTo Cheng Apr 20 '19 at 13:39
3

You could prove that the product of a paracompact space and a compact space is paracompact by using the tube lemma.