This sounds obvious (all balls of all rational radii on rational centres should work, I think) but I can't complete the proof simply from these two properties of $\mathbb R^n$.

  1. $\mathbb R^n$ has a basis of open balls (all balls of all radii on all centres)
  2. $\mathbb R^n$ is second countable and Lindelöf

So from that $\mathbb R^n$ has a countable basis; the cover of $\mathbb R^n$ by open balls has a countable subcover; any open set in $\mathbb R^n$ is the countable union of open balls. But does it easily follow that $\mathbb R^n$ has countable basis of open balls ?

If I take all open sets in $\mathbb R^n$ and express each as a countable union of open balls then I end up with an uncountable set of open balls that cover $\mathbb R^n$, and I can then get a countable subset that covers $\mathbb R^n$, but does this still generate each open set ?

Second thoughts after a hint in a comment. Does this work for a proof ?

  1. $\mathbb R^n$ is second countable so there is a countable basis of open sets {$C_i$}.
  2. $\mathbb R^n$ has a basis of open balls (all balls of all radii on all centres) so each $C_i$ is the union of open balls, and is "covered" by this union.
  3. $\mathbb R^n$ is Lindelöf so the cover of $C_i$ by open balls has a countable subcover.
  4. The set of all open balls in all countable subcovers of {$C_i$} is countable, generates {$C_i$} which generates $\mathbb R^n$ and is therefore a basis for $\mathbb R^n$ .

Thanks for all the feedback. I found a more general answer to the question (more or less in line with my own second thoughts) here: http://www.austinmohr.com/Work_files/hw2-1.pdf referenced in a question here: Bases having countable subfamilies which are bases in second countable space

  • $\begingroup$ No. You might end up with the balls of radius $k\in\Bbb N$ around the origin. A bit more care is required $\endgroup$ Dec 28 '15 at 13:13
  • $\begingroup$ @HagenvonEitzen Thanks - I think I got it, do you agree ? $\endgroup$ Dec 28 '15 at 13:37

Let $D$ be a dense subset of the metric space $X$ with metric $d.$ Let $B=\{B_d(p,q) :p\in D \land q\in Q^+\}.$ Then $B$ is a base for $X.$ To prove this it suffices to show that whenever $y\in U$ with $U$ open, there exists $ V\in B$ with $ y\in V\subset U.$ Consider that $B_d(y,r)\subset U$ for some $r>0$, and that some $p\in D\cap B_d(y,r/3).$ Take $q\in Q^+$ with $r/3<q<r/2.$ By the triangle inequality, $y\in B_d(p,q)\subset B_d(y,r)\subset U.$...... In particular if $D$ is countable then so is $B$. The set of members of $R^n$ with rational co-ordinates is countable, and dense in $R^n$.


I think you still end up in the same trap that Hagen von Eitzen points out in the comments. It's entirely possible that in step $3$ taking your countable subcover of $C_i$'s could leave you with $C_i = B_{i}(0)$, where each $C_i$ is also an open ball of radius $i$ centered at $0$ from your open-ball basis. If you are comfortable with what you have then ignore what I have to say. If you are looking for some critique and suggestions read on.

I think step $1$ is exactly what you want to do, as is everything you do in step $2$ up until you say "so". At this point I would explicitly show (or claim) that $C_i$ can be written as a union of open balls from your collection of all open balls, $\{\mathcal{B}_\alpha\}_{\alpha \in \lambda}$. More importantly, demonstrate that $C_i$ can be written as a countable union of open balls. That is, $$C_i = \bigcup_{j \in K_i}B_j$$ with $B_j \in \{\mathcal{B}_\alpha\}_{\alpha \in \lambda}$ and $K_i \subseteq \mathbb{N}$ is an index for $C_i$. This can be done by using balls with rational radii, and if need be, placing them carefully inside $C_i$. This will lead to a set that I'll define as $$\mathcal{B}_{C_{i}} = \left\{B_j \in \{\mathcal{B}_\alpha\}_{\alpha \in \lambda}: C_i = \bigcup_{j \in K_i}B_j \right\}$$ which is a subset of countably many open balls whose union gives $C_i$. Since we have countably many $C_i$'s we'll have countably many $\mathcal{B}_{C_{i}}$'s. A countable union of countably many things is still countable, so $\{\mathcal{B}_{C_{i}}\}$ contains countably many open balls. Lastly, any open set $U$ in the topology is such that $$U = \bigcup_{i \in I} C_i = \bigcup_{i \in I}\left(\bigcup_{j \in K_{i}}B_j \right)$$ where $I\subseteq \mathbb{N}$ is an index for $U$. This demonstrates that $\left\{\mathcal{B}_{C_{i}}\right\}$ is a countable basis of open balls for $\mathbb{R}^n$. I do not believe you need the Lindelöf property at any point.


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.