Proof that Borel's normal number theorem implies the weak law of large numbers

368 Views Asked by At

This is from Billingsley's Probability and Measure, Chapter I. The following theorems are proved:

The Weak Law of Large Numbers: For every $\epsilon > 0$, $$\lim_{n \rightarrow \infty} P\left[\omega \in (0, 1]: \left|\frac{1}{n}s_n(\omega)\right|\geq \epsilon \right] = 0$$ where $s_n(\omega) = \sum_{i=1}^n r_i(w)$. The Rademacher functions $r_i$ are defined as $r_n(\omega) = \begin{cases}+1 &d_n(\omega) = 1 \\ -1 &d_n(\omega) = 0\end{cases}$ where $d_n(\omega)$ is the $n$th digit in the binary expansion of $\omega$ (non terminating representation).

And the following special case of the Strong Law of Large Numbers is proved.

Borel's Normal Number Theorem: $$\lim_{n \rightarrow \infty} \frac{1}{n}s_n(\omega) = 0 \quad \text{almost everywhere on } (0, 1] $$ where $s_n(\omega)$ is defined as above.

The book then mentions (without proof) that if $f_n$ is a step function on $(0, 1]$ for each $n$ then if $\lim_{n \rightarrow \infty} f_n(\omega) = 0$ almost everywhere on $(0, 1]$ we have $\lim_{n \rightarrow \infty} P\left[\omega \in (0, 1]: \left|f_n(\omega)\right|\geq \epsilon \right] = 0$ for every $\epsilon > 0$. In particular, taking $s_n$ to be our step functions $f_n$, we can conclude that Borel's normal number theorem implies the weak law of large numbers.

I have been trying to prove this general result without much luck.

If for every $n \in \mathbb{N}$, $f_n$ is a step function on $(0, 1]$ and if $\lim_{n \rightarrow \infty} f_n(\omega) = 0$ almost everywhere on $(0, 1]$, then for every $\epsilon > 0$, $\lim_{n \rightarrow \infty} P\left[\omega \in (0, 1]: \left|f_n(\omega)\right|\geq \epsilon \right] = 0$.

Fix $\epsilon > 0$. If $A_n = \left\{\omega: \left|f_n(\omega)\right|\geq \epsilon \right\}$, we need to show that given any $\epsilon_0 > 0$, there exists some $N \in \mathbb{N}$ such that for $n \geq N$, $ |A_n| < \epsilon_0$. As $f_n$ is a step function, each $A_n$ is the union of a finite number of intervals, so its length is well defined.

If $A = \left\{\omega: \lim_{n \rightarrow \infty} f_n(\omega) = 0\right\}$, then by hypothesis, the complement of $A$ has measure zero. If $\omega \in A$, then there exists some $N_\omega$ such that for $n \geq N_\omega$, $f_n(\omega) < \epsilon$. That is, $\omega \notin A_n$ for $n \geq N_\omega$. However, I'm struggling to see where to go from here -- how do we choose an $N$ large enough so that $A_n$ does not contain "most" of the interval $(0, 1]$ for $n \geq N$? I feel like the fact that the $f_n$'s are step functions is key here.

Any hints appreciated, I feel quite stuck.