2
$\begingroup$

I have done the problem, but I'm confused about why the local path-connectedness of $Z$ is necessary.

My solution: Let $p:X\to Y$ and $q:Y\to Z$ be such that $q$ and $q\circ p$ are covering spaces. Let $y\in Y$ and let $W$ be a path-connected neighborhood of $z=q(y)\in Z$ which is evenly covered by $q$ and $q\circ p.$ Let $V$ be the path component of $q^{-1}(W)$ which contains $y$ (observe that $V$ is then a sheet of $W$ by $q$). Any sheet $U$ of $p^{-1}(V)$ is mapped homeomorphically to $W$ by $q\circ p.$ Since $q:V\to W$ is a homeomorphism, we conclude that $p:U\to V$ is a homeomorphism, so $p$ is a covering space.

It seems like I used the local path-connectedness of $Z$ in my solution, letting $W$ be a path-connected neighborhood of $q(y),$ but this use seems artificial. It isn't clear to me why $W$ must be path-connected, other than when one tries to isolate an evenly covered neighborhood $V$ of $y.$ However, it seems like local connectedness is enough to do this.

So my question is this: is local path-connectedness of $Z$ necessary (and if so, is there some instructive counterexample when $Z$ is not locally path-connected) or can this hypothesis be weakened?

$\endgroup$
  • $\begingroup$ You need some additional assumptions. Probably the best setting is to require that $X,Y,Z$ are connected. Otherwise you get trivial counterexamples like $X = Z$ and $Y= Z \times \{0,1 \}$ with $p(z) = (z,0)$ and $q(z,i) = z$. $\endgroup$ – Paul Frost Dec 29 '18 at 14:56
0
$\begingroup$

It can be weakened. We shall use the following well-known fact about covering projections:

Given $p,q$ as in your question (i.e. $q \circ p$ and $q$ are assumed to be covering projections). If $Z$ is locally connected and $p$ is a surjection, then $p$ is a covering projection.

In this result no further connectedness assumptions on $X,Y,Z$ are needed.

Hence we have to look for assumptions assuring that $p$ is surjective. To avoid trivial counterexamples, it seems that we should have the minimal requirement that $Y$ is connected (otherwise we may take $X = Z$, $Y = Z \times F$ with a discrete $F$ having more than one point and $p(z) = (z,f_0)$, $q(z,f) = z$). This implies that also $Z$ is connected. For the sake of homogeneity we may moreover asume that $X$ is connected, but this is not really needed.

Therefore, let us assume that $X,Y,Z$ are connected and locally connected (it suffices to assume one of these spaces to be locally connected because they are locally homeomorphic). We shall prove

If $Y$ is path connected, then $p$ is surjective, i.e. a covering projection.

This is somewhat weaker than requiring $Z$ locally path connected. Note that if the latter is satisfied, then also $Y$ is locally path connected, hence $Y$ is path connected.

Let $y \in Y$. Choose any $x \in X$ and any path $v$ in $Y$ from $p(x)$ to $y$. Then $w = q \circ v$ is a path in $Z$ beginning at $(q \circ p)(x)$. It can be lifted to path $u$ in $X$ beginning at $x$. Both paths $v$ and $p \circ u$ are lifts of $w$ beginning at $p(x)$. By unique path lifting we see that $v = p \circ u$, hence $y = v(1) = p(u(1)) \in p(X)$.

Update:

Given $p,q$ as in your question (i.e. $q \circ p$ and $q$ are assumed to be covering projections), let us first observe that if if one of $X, Y, Z$ is locally connected, then so are the other two because the three spaces are locally homeomorphic. Let us verify the following more general results:

(1) If $Z$ is locally connected, then each $y \in Y$ has an open neighborhood $U$ which is evenly covered by $p$. Note that this is trivially true if $p^{-1}(U) = \emptyset$. See my answer to Exercise 1.3.16 in Hatcher. However, $p$ is in general not a covering projection because surjectivity is not guaranteed.

(2) In $Z$ is locally connected, then $p(X)$ is open and closed in $Y$ and $p : X \to p(X)$ is a covering projection.

In fact, for $y \in Y$ let $U$ be an open evenly covered neighborhood of $y$. If $y \in p(E)$, then $p^{-1}(U) \ne \emptyset$ since it contains $p^{-1}(y)$. Hence $U = p(p^{-1}(U)) \subset p(X)$, i.e. $p(X)$ is open. If $y \notin p(X)$, then $p^{-1}(U)$ must be empty because otherwise we would get $y \in U = p(p^{-1}(U)) \subset p(X)$. Hence $U \cap p(E) = \emptyset$, i.e. $Y \setminus p(X)$ is open. That $p : X \to p(X)$ is a covering projection is now clear.

(3) If $Y$ is connected and locally connected, then $p$ is a covering projection.

$p(X)$ is open and closed, thus $p(X) = Y$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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