One is (almost) normal in base π

See previous posts if you’re confused:

Part 1: Normal numbers in base ten
Part 2: Numbers in base φ and π
Part 3: One is normal in base π

This post is also available on LessWrong.

This wasn’t supposed to be a whole series of posts, so we’re going to speed through the punchline:

One is a normal number.

“No it’s not.”

In base π, I mean.

“No, it’s still just 1.“

Don’t be so greedy. Start with 0. and go from there.

“So it’s 0.222… or 0.333… or something?”

Nope, it’s 0.3011021110…

“That’s… bad. Does it repeat?”

I think it’s normal, but that seems hard to prove.

“So each digit occurs equally often?”

No, normal numbers in base π are about a 37%-30%-29%-4% split of 0, 1, 2, 3.

“Where did that distribution come from?”

Integrals of this weird function:

“And where did that come from?”

It’s the distribution of remainders when computing a random number in base π.

“What are the x coordinates of those discontinuities?”

Sequence of remainders when computing 0.3011021110…

“um”

which are dense in [0, 1], by the way.

“Can you prove that?”

No, and neither can you.

“…And the y coordinates?”

The discontinuities get smaller by a factor of π each time.

“And almost all numbers have this distribution of remainders?”

Yup.

“Including 1, if you use the 0.3011021110… representation.”

Yup.

“So the histogram of x-coordinates of the N largest discontinuities in this function approaches… this function itself, as N goes to infinity.”

Yup.

“Which has derivative zero almost everywhere, but has a dense set of discontinuities.”

Yup!

“Any other neat facts about it?”

It’s a fractal.

“How?”

Stretch out the colored regions by a factor of π horizontally, shrink by π vertically, and add them to each other. You get the original function back:

“And this is all because π is normal, or transcendental, or something?”

Nope, I think analogous statements are all true for base 3.5.