I am reading Bredon's Introduction to compact transformation groups, and came across the following result and proof on page 34:

enter image description here

Even though he writes "Recall our standing assumption that $X$ is Hausdorff," I cannot see where any properties of Hausdorffness of $X$ are used in this proof. I suspect it would be used to ensure that convergent nets in $X$ converge uniquely, but I do not see why this fact would be needed for this proof.

Is it true that the continuous action of a compact topological group on a topological space $X$ is always closed, regardless of whether $X$ is Hausdorff? If not, then where in the above proof is Hausdorffness used?

  • $\begingroup$ There's also the question of whether one assumes a topological group to be $T_0$, and hence Hausdorff, by default. $\endgroup$ Jul 13, 2021 at 7:24
  • $\begingroup$ Compact/ Locally compact groups are usually assumed Hausdorff by default, even when topological groups are not. $\endgroup$
    – YCor
    Jul 13, 2021 at 8:03
  • $\begingroup$ Hausdorffness of $X$ doesn't seem to be used at any point in the proof. $\endgroup$
    – YCor
    Jul 13, 2021 at 8:05
  • $\begingroup$ The only critical point seems to be the existence of a subnet, but this is well known, see f.i. Kelley, General Topology. $\endgroup$ Jul 13, 2021 at 9:16

2 Answers 2


Yes, it is true, and the following argument will surely convince you so. Recall that a topological space $K$ is compact iff it is universally closed: for every topological space $X$, the coordinate projection map $X\times K \to X$ is closed. Now, use the obvious fact that $G\times X\to G\times X$, $(g,x)\mapsto (g,gx)$, is a homoemorphism and realize the action map $G\times X \to X$ as a composition of this homeo with the coordinate projection map.

  • $\begingroup$ Thanks for the great answer. It appears we do not even need $G$ to be Hausdorff (a question raised by David Roberts in the comments to my question). $\endgroup$
    – Ben
    Jul 13, 2021 at 12:30
  • 1
    $\begingroup$ @Ben, right, but this is not a big deal, really. If $G$ is not Hausdorff then, denoting $G_1=\overline{\{1\}}$, we have that $G_1$ is a closed normal subgroup of $G$ on which the topology is trivial, thus each $G_1$ orbit in $X$ is inseparable. Then the action map $G\times X \to X$ fibers naturally over the action map of $G/G_1$ on $X/G_1$, thus everything reduces to the separable case. That is, the theory of dynamical systems reduces easily to the Hausdorff world. $\endgroup$
    – Uri Bader
    Jul 13, 2021 at 16:34

Here is an elementary proof:

Let $C\subseteq G\times X$ be closed. Note that $x_0\notin \Theta(C)$ if and only if $\{(g,g^{-1}x_0)\mid g\in G\}\cap C = \emptyset$ which is if and only if $G\times \{(e,x_0)\}\subseteq \Phi^{-1}((G\times X)\setminus C)$ where $\Phi: (g,(h,x))\mapsto (hg^{-1}, gx)$. Now, let $x_0\notin \Theta(C)$. By our observation, $G\times \{(e,x_0)\}\subseteq \Phi^{-1}((G\times X)\setminus C)$ and since $C$ is closed, $\Phi^{-1}((G\times X)\setminus C)$ is a neighborhood of $G\times \{(e,x_0)\}$. Hence, (by the tube lemma), we can find neighborhoods $U$ of $e$ and $V$ of $x_0$ such that $G\times (U\times V) \subseteq \Phi^{-1}((G\times X)\setminus C)$. It is easy to see that $\Theta(U\times V)$ is a neighborhood of $x_0$ which does not intersect $\Theta(C)$.


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.