It is well known that perfect sets in $\mathbb{R}^n$ are uncountable (e.g., baby Rudin states this). Recently I heard of this stronger result:

Every perfect set in $\mathbb{R}^n$ has cardinality $2^{\aleph_0}$.

This is easily proved if we assume the continuum hypothesis. However, this result does not rely on that. Is there a proof of this fact? And does this result hold for more general spaces (e.g., complete metric spaces)?

I understand that it's customary to show my effort here at math.SE, but honestly I have no idea how I should attempt at it...


Yes, it does not require the continuum hypothesis to prove.

Suppose $P$ is a perfect set. WLOG (if not, restrict to an appropriate closed interval) $P$ is bounded. Then we can find two disjoint closed subsets, $l(P)$ and $r(P)$, which are each also perfect. And we can iterate this process, building such things as $r(l(r(P)))$ and so on.

Now remember that the intersection of a descending sequence of closed and bounded sets is nonempty. This means that for every infinite binary sequence, the corresponding descending sequence of perfect sets has a point in the intersection. For example - identifying "$r$" with $1$ and "$l$" with $0$ - the sequence $$f=0,1,0,1,,...$$ yields the sequence $$l(P)\supseteq r(l(P))\supseteq l(r(l(P)))\supseteq r(l(r(l(P))))\supseteq...,$$ and we pick a point $p_f$ in the intersection of this chain of perfect sets.

Now just check that if $f\not=g$ we have $p_f\not=p_g$ (HINT: $r(X)\cap l(X)=\emptyset$ ...).

It might seem like we used the axiom of choice here, in two places:

  • Choosing $l(X)$ and $r(X)$, for a perfect set $X$.

  • Choosing a point in the intersection of a decreasing sequence of closed and bounded sets.

However, we don't actually need AC here:

  • If $X$ is perfect, we can in fact show that there is a pair of rationals $p<q$ such that $X\cap (-\infty, p]$ and $X\cap [q,\infty)$ are each perfect. But $\mathbb{Q}^2$ is well-orderable ...

  • There is in fact an easily-definable choice function for nonempty closed sets. (HINT: if the set is bounded from below, just pick the smallest element; now do you see how to deal with the not-bounded-from-below case?) We could also require that $l(X)$ and $r(X)$ have diameter at most one half of that of $X$ (where the diameter of a set is the supremum of the distances between any two points in the set).


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.