I have identified the most disturbing math theorem ever. What makes it the most disturbing is that it does not involve the Axiom of Choice in any way. I’ve seen the theorem many times before, but I never really noticed how disturbing it was until a couple of days ago.
The theorem is this: for any positive constant c, there is an open set U that contains every rational point, but has measure less than c. Think about what that means, for a minute. The rationals are dense in the reals. Here’s a set that contains an open interval around every rational. Naively I would have believed that the set would have to be the whole real line (except with maybe a finite or countable number of exceptions). At the worst, I would have at least expected the set itself to have infinite measure, and the set’s complement to be measure zero. Instead, not only can we construct such a U with finite measure, we can make that measure be arbitrarily close to zero.
The proof of this is pretty easy, and is a standard result in any real analysis book that covers Lebesgue measure. Not only have I seen it before, I’m pretty sure I’ve seen it pointed out before that the result is surprising. Somehow I never took in how strange it is until just this week.
Hmm… to me it only feels equivalently ‘strange’ to the fact that the reals have a countable dense subset.
Once you know that it doesn’t seem too surprising, you can just take, say, the union of open balls of radius epsilon/2^(n+1) around the nth rational, right?
I no longer really find this surprising, but when I first saw it in the study of Lebesgue theory, I found it somewhat pathological. I think I’ll spend some time before I go back to school next week trying to construct a covering of the rationals that excludes some fixed irrational (or, dually, finding an irrational not covered by a given covering of the rationals). That seems like an interesting challenge.
I agree. This one has always bothered me too. The proof is so short you can just see why it’s true. And yet it’s so counterintuitive.
The related result that disturbs me more is that the reals are the union of a set of measure 0 with a meager (in the sense of Baire) set. Just take the countable intersection of dense open sets of arbitrarily small measure containing all the rationals.
what about banach-tarski? chop an orange up into tiny bits – rearrange the bits and voila! two identical oranges.
Banach-Tarski depends on Axiom of Choice, so it doesn’t count.
This bothered me too, but a related fact that bugged me more was that the Cantor set has an uncountable number of connected components, even though you’ve cut out only a countable number of intervals. More generally, it’s curious that the structure of an arbitrary open set in R^1 is relatively simple (an at-most countable union of disjoint open intervals), but closed sets in R^1 can be so complex; even though a closed set is just the compliment of an open set.
–Rick Taylor