Suppose that $G$ is a compact abelian group. I denote by $G_0$ the connected component of the identity in $G$.

If $G_0$ is open in $G$ (equivalently $G/G_0$ is finite) is it true that $G\cong G_0\times G/G_0$?

I also assume that $G/G_0$ is equipped with the quotient topology and that everything is Hausdorff.

*If the title confuses you, the statement above is equivalent to whether the short exact sequence $0\rightarrow G_0\rightarrow G \rightarrow G/G_0\rightarrow 0$ splits.

  • $\begingroup$ If $G$ is a Lie group then the answer is yes, if it's any help. $\endgroup$ – Cronus Dec 19 '18 at 21:49
  • $\begingroup$ If $G$ is a compact abelian Lie group then it is a torus times a finite group. So of course it works. I'm reading your answer though $\endgroup$ – Yanko Dec 20 '18 at 18:21
  • $\begingroup$ But do you know of a simple proof for that (that a compact abelian Lie group is a torus times a finite group)? $\endgroup$ – Cronus Dec 21 '18 at 16:29
  • $\begingroup$ Actually, you're right, this fact is not hard to prove. $\endgroup$ – Cronus Dec 22 '18 at 12:35

Yes, this is true. The case for Lie groups is pretty easy (and I explained it here). For the general case, one can prove this using the fact compact groups are inverse limits of Lie groups.

Let $G$ be a compact abelian group, and denote $A=G/G^0$. By a corollary of Peter-Weyl, every identity neighbourhood contains a subgroup which is co-Lie (i.e. the quotient by it is a Lie group). Thus we can form a net of subgroups $N_\alpha$, ordered by reverse inclusion (so $\alpha \geq \beta$ if and only if $N_\beta \subseteq N_\alpha$), such that $N_\alpha\subseteq G^0$ for all $\alpha$ (sine $G^0$ is open) and such that $\bigcap N_\alpha=\{1\}$. Denote $G_\alpha=G/N_\alpha$ and denote by $p_\alpha:G\to G_\alpha$ the natural quotient map.

Since each $p_\alpha$ is open, continuous and surjective, the subgroup $p_\alpha(G^0)$ is open and connected, and hence is the connected component of $G_\alpha$. Since $N_\alpha\subseteq G^0$, we have by the third isomorphism theorem (for topological groups) that $G_\alpha/G_\alpha^0=(G/N_\alpha)/(G^0/N_\alpha)\cong G/G^0 = A$. Thus, by the case for Lie groups, we know $G_\alpha \cong G_\alpha^0\times A_\alpha$ for $A_\alpha\cong A$.

Denote $A'_\alpha = p_\alpha^{-1}(A_\alpha)$ and $A'=\bigcap A'_\alpha$. I will show that $A'$ is mapped isomorphically onto $A$ via the quotient $q:G\to G/G^0$, which shows the sequence does indeed split.

First let us see that it is mapped injectively into $A$ via $q$. We have, for every $\alpha$, $$p_\alpha(A'_\alpha\cap G^0)\subseteq p_\alpha(A'_\alpha)\cap p_\alpha(G^0)=A_\alpha\cap G_\alpha^0=\{1\}.$$ Thus $A'_\alpha\cap G^0\subseteq N_\alpha$ for every $\alpha$, so $\bigcap A'_\alpha\cap G^0\subseteq\bigcap N_\alpha=\{1\}$. This precisely mean that the intersection of $A'=\bigcap A'_\alpha$ with $\ker q=G^0$ is trivial, i.e. $q$ mapps $A'$ injectively into $A$.

It remains to show $q(A')=A$. I think this is pretty clear. For every $\alpha$ we know $q(A'_\alpha)=A$. To see this, first note that $p_\alpha(A'_\alpha\cdot G^0)$ contains both $p_\alpha(A'_\alpha)=A_\alpha$ and $p_\alpha(G^0)=G_\alpha^0$ and hence is all of $G_\alpha=G_\alpha^0\times A_\alpha$, and therefore $A'_\alpha\cdot G^0\cdot \ker p_\alpha = G$, but $\ker p_\alpha = N_\alpha \subseteq G^0$, so $A'_\alpha\cdot G^0=G$. Since $\ker q= G^0$ we have $q(A'_\alpha)=A$. But by compactness this implies $q(A')=A$ (since if $a\in A$ then for every $\alpha$ there is $a_\alpha\in A'_\alpha$ such that $q(a_\alpha)=a$; this net has some converging subnet which converges to some $a'\in A$, and by continuity $q(a')=a$).

This answer turned out a lot longer than I expected... Using the language of inverse limits it would have been a lot shorter, I think, if you know how to prove $G\cong \varprojlim G_\alpha$ in a natural way and not just that every neighbourhood contains a co-Lie subgroup.

  • $\begingroup$ Very nice thank you! $\endgroup$ – Yanko Dec 20 '18 at 18:27

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.