Today a student ask me the following question regarding topological groups in the tutorial centre.

Let $H$ be a subgroup of a topological group $G$. Suppose that $U$ is an open containing the identity such that $\overline{U} \cap H$ is closed. Prove that $H$ is closed.

I am a bit rusty on point set topology so I couldn't figure out a proper solution. I tried a straightforward approach without success:

Let $x \in \overline{H}$, we want to show that $x \in H$. Since $e \in U$ then $x \in xU$, and we can take an element $y \in xU \cap H$, therefore $y = xa = h$ for some $a \in U$, $h\in H$, so $x = ha^{-1}$

Trying to prove that $a^{-1} \in H$, I want to use somehow that $\overline{U} \cap H$ is closed; but I'm stuck at this point. I wish I could dedicate more time to solve the problem and review the basic properties of topological groups, but unfortunely I am overhelmed with my own stuff.

  • $\begingroup$ Can you let $U = H - $ the limit points? $\endgroup$ – Larry B. Oct 31 '17 at 2:30
  • $\begingroup$ What do exactly you mean by $H$ - limit points? $\endgroup$ – C. Zhihao Oct 31 '17 at 2:45

Suppose $x\in\overline{H}$. Let $V$ be a neighborhood of $1$ such that $VV^{-1}\subseteq U$. Since $Vx$ is an open neighborhood of $X$, $x\in\overline{H\cap Vx}$. Now pick some $y\in H\cap Vx$ and multiply everything by $y^{-1}$ on the right. We conclude that $xy^{-1}\in\overline{Hy^{-1}\cap Vxy^{-1}}$. Since $H$ is a subgroup and $y\in H$, $Hy^{-1}=H$. Also, since $y\in Vx$, $xy^{-1}\in V^{-1}$ so $Vxy^{-1}\subseteq VV^{-1}\subseteq U$. Thus $$xy^{-1}\in\overline{H\cap U}\subseteq \overline{H\cap\overline{U}}=H\cap \overline{U}.$$ In particular, $xy^{-1}\in H$, so since $y\in H$ and $H$ is a subgroup, $x\in H$.


First, a bit of point set topology: $\bar U \cap H$ closed is equivalent to $U \cap H$ is closed in $U$.

Second, by homogeneity, for every $h\in H$, there exists $U_h$ neighborhood of $h$ so that $U_h \cap H$ is closed in $U_h$ ( take $U_h = h\cdot U$)

The above means that $H$ is locally closed in $G$. This is equivalent to : $H$ is the intersection between a closed set and an open set, or, what we are really after: $H$ is open in its closure $\bar H$.

Now, $\bar H$ is a topological group ( with the induced topology. Let's now recall: an open subgroup of a topological group is also closed. Indeed, any coset will be open, so any union of cosets. Therefore $H$ is also closed in $\bar H$, and so equal to $\bar H$, closed.

Take home message: a locally closed subgroup is closed.


Let $(x_n, n\in I)$ be a net where $A$ is a directed set. Suppose $x=lim_nx_n, x_n\in H$, write $y_n=x^{-1}x_n$, $lim_ny_n=e$, there exists $N$ such that $n>N$ implies that $y_n\in U$. Let $n_0>N$ $z_n=x_n^{-1}x_{n_0}$, $lim_nz_n=x^{-1}x_{n_0}\in U$ implies that there exists $M$ such that $n>M$ implies that $x_n^{-1}x_{n_0}\in U\cap H\subset\bar U\cap H$ we deduce that, $x^{-1}x_{n_0}=lim_{n\geq M}x_n^{-1}x_{n_0}\in \bar U\cap H$ since $\bar U\cap H$ is closed. This implies that $x^{-1}x_{n_0}\in H$ and $x\in H$.


  • 1
    $\begingroup$ I think you need first countability to deal with sequences; which I think in general is not true for any topological group. $\endgroup$ – C. Zhihao Oct 31 '17 at 0:12
  • 1
    $\begingroup$ Even though you added the condition of being Hausdorff (to asure uniqueness in the limits), being First countable is necessary to say that the closure of H is exactly the set of limit points of sequences with values in H. Here the statement is for general Topological groups and none of this properties should be assumed. $\endgroup$ – C. Zhihao Oct 31 '17 at 1:23
  • $\begingroup$ @C.Zhihao: The exact same argument works in general if you just use nets instead of sequences. The argument also doesn't actually assume limits are unique anywhere and so does not require $G$ to be Hausdorff. $\endgroup$ – Eric Wofsey Oct 31 '17 at 5:25
  • $\begingroup$ @EricWofsey I think first countability is still needed to use that $\lim f(x_n) = f(x)$ under a continuous function $f$. However, your solution is correct and I will take it as the solution to the OP. $\endgroup$ – C. Zhihao Oct 31 '17 at 16:43

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.