0

In this question I found out that the normal numbers in $[0,1]$ form a Borel set. Since every Borel set is Lebesgue measurable, I now wonder what would be the Lebesgue measure of the set of normal numbers?

Intuitively, normal numbers are what you would 'expect' to get when you pick a 'random number' in $[0,1]$. Therefore, I suspect that almost all numbers are normal. Then the Lebesgue measure would be 1.

In this question they prove it by using that the non-normal numbers have measure 0. However, how do you prove that?

Riemann
  • 929

1 Answers1

2

I will prove this for simply normal numbers in base $2$. Let $a$ be a uniformly random number in $[0,1]$. Write $a=\sum_{n=1}^{\infty}a_n2^{-n}$. Then the $a_n$ are i.i.d. random variables with mean $0.5$. By the Law of Large Numbers, it follows that the sample average $\frac{a_1+\dots+a_n}{n}$ converges to the expected value $0.5$ with probability $1$. With other words, the set of real numbers $x=\sum x_n2^{-n}\in[0,1]$ with $\lim_{n\rightarrow\infty}\frac{x_1+\dots+x_n}{n}=\frac12$ has Lebesgue measure $1$.

The case for higher bases can probably be proven in a similar way.

Now we can take the intersection for all bases. This is the intersection of countably many sets with measure $1$. Therefore, the interection also has measure $1$.

Riemann
  • 929