I already know that this is impossible for $(\Bbb Q, +, \mathcal T)$ to be a compact Hausdorff topological group (notice that the trivial topology does not work because it is not Hausdorff).

Indeed, this follows from Baire theorem: for if $(\Bbb Q, +, \mathcal T)$ were a compact Hausdorff topological group, then $\Bbb Q$ would be the union of the countable collection of closed sets $\{r\}$ (with $r \in \Bbb Q$). As we have a locally compact Hausdorff space, Baire theorem tells us that at least one of the closed set has non empty interior, i.e. some $\{r\}$ is open. Since we have a topological group, it follows that $(\Bbb Q, +, \mathcal T)$ is discrete and hence non compact.

But what about $(\Bbb R, +, \mathcal T)$ ? My first idea was to use the group (and even vector spaces) isomorphism $\Bbb R \cong \Bbb Q^{(\Bbb N)}$. Transporting the topology $\mathcal T$ on $\Bbb Q^{(\Bbb N)}$ preserves compactness, and we could try to use the projection $\Bbb Q^{(\Bbb N)} \to \Bbb Q$. But I was not sure what to do then.

Any comment would be appreciated!


A good way to think about this is in terms of Pontryagin duality. Since we only care about the abelian group structure of $\mathbb{R}$, let's first get a nice characterization of this structure. As an abelian group, $\mathbb{R}$ is the unique $\mathbb{Q}$-vector space of its cardinality (up to isomorphism). An abelian group $A$ is a $\mathbb{Q}$-vector space iff for each nonzero $n\in \mathbb{Z}$, the multiplication by $n$ map $n:A\to A$ is an isomorphism.

Now the neat thing about this is that this condition is self-dual under Pontryagin duality. If $A$ is a locally compact abelian group, then the dual of the map $n:A\to A$ is just the map $n:\hat{A}\to\hat{A}$ on the dual group. So this says that a locally compact abelian group is a $\mathbb{Q}$-vector space iff its dual is a $\mathbb{Q}$-vector space.

In particular, let us use this to classify the compact abelian groups which are isomorphic (as groups) to $\mathbb{R}$. These are just the Pontryagin duals $\hat{V}$ of all $\mathbb{Q}$-vector spaces $V$ (with the discrete topology) for which $\hat{V}$ has cardinality $2^{\aleph_0}$. It is not hard to show that $\hat{\mathbb{Q}}$ has cardinality $2^{\aleph_0}$. If $V$ is a $\kappa$-dimensional $\mathbb{Q}$-vector space then $\hat{V}$ is a product of $\kappa$ copies of $\hat{\mathbb{Q}}$, which has cardinality $2^{\aleph_0\cdot\kappa}$.

So to sum up, there are indeed compact group topologies on $\mathbb{R}$. Up to continuous isomorphism, there is one such topology for each cardinal $\kappa$ such that $2^{\aleph_0\cdot\kappa}=2^{\aleph_0}$ (in particular, this includes all $\kappa$ such that $0<\kappa\leq\aleph_0$). The Pontryagin dual of this compact group is a $\mathbb{Q}$-vector space of dimension $\kappa$.

The case $\kappa=1$ gives $\hat{\mathbb{Q}}$, which is a solenoid. Explicitly, $\hat{\mathbb{Q}}$ is the inverse limit of the sequence $\dots\to S^1\stackrel{4}{\to}S^1\stackrel{3}{\to}S^1\stackrel{2}{\to}S^1$, since $\mathbb{Q}$ is the direct limit of the sequence $\mathbb{Z}\stackrel{2}{\to}\mathbb{Z}\stackrel{3}\to\mathbb{Z}\stackrel{4}{\to}\mathbb{Z}\to\dots$. For general $\kappa$, you just have a product of $\kappa$ copies of $\hat{\mathbb{Q}}$.

  • $\begingroup$ This is a great answer, because I love Pontryagin duality! I will read it carefully. Thank you! $\endgroup$ – Watson Jul 21 '16 at 20:35
  • $\begingroup$ Just a little question: since $\widehat{\mathbb{Q}}$ is a compact topological group of cardinality $2^{\aleph_0}$, we can transport that structure on $\Bbb R$, but then the addition is not necessarily the usual addition on the reals. Am I right? $\endgroup$ – Watson Jul 21 '16 at 20:52
  • 1
    $\begingroup$ $\hat{\mathbb{Q}}$ is not just a compact topological group of cardinality $2^{\aleph_0}$ but also a $\mathbb{Q}$-vector space under its group structure, so it is isomorphic as a group to $\mathbb{R}$. $\endgroup$ – Eric Wofsey Jul 21 '16 at 21:09
  • $\begingroup$ Is there a way to construct such a topology on $\mathbb{R}$, though? $\endgroup$ – Cronus Aug 19 '16 at 1:28
  • 1
    $\begingroup$ @Cronus: Probably not without the axiom of choice. With the axiom of choice, it's "easy": just pick a basis for $\mathbb{R}$ over $\mathbb{Q}$ and a basis for $\hat{\mathbb{Q}}$ over $\mathbb{Q}$ and a bijection between them, and transport the topology along the induced bijection $\mathbb{R}\to\hat{\mathbb{Q}}$. $\endgroup$ – Eric Wofsey Aug 19 '16 at 1:33

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.