0
$\begingroup$

I want to identify the interval $[0, 1]$ with the Lebesgue measure to the probability space for infinite tossing a fair coin.

I know we can define a probability on cylinder sets (the sets can be represented by the outcomes of finite tosses) and then use Caratheodory extension theorem to extend the probability to the $\sigma$-algebra $\mathcal{F}$ generated by all the cylinder sets.

But I am not sure how to identify this probability measure with the Lebesuge measure on $[0,1]$. In particular, I don't know what $\mathcal{F}$ looks like and why it can be identified as $\sigma$-algebra where Lebesgue measure is defined. I just have a brief thought about identify $\omega\in[0,1]$ with a binary representation.

Could you explain how to identify these two measure space ? It will also be helpful if you can provide a reference about this topic.

$\endgroup$
1
$\begingroup$

Say $S=\{0,1\}^{\Bbb N}$ is the space of sequences $s=(s_1,\dots)$ of zeroes and ones. Define a map $f:S\to[0,1]$ by $$f(s)=\sum_{n=1}^\infty s_n2^{-n}.$$Then $f$ is "almost a bijection"; in fact it's a bijection except for countably many $x\in[0,1]$ that have two inverse images.

And $f$ is measure-preserving. For example, if $C$ is the cylinder in $S$ defined by $s_1=0$, $s_2=1$ then $f(C)=[1/4,1/2]$; both $C$ and $f(C)$ have measure $1/4$.


Or to do the same thing from the other direction: Regard $[0,1]$ with Lebesgue measure as a probability space. It's enough to show that this space supports a sequence $X_1,X_2,\dots$ of iid random variables with $P(X_j=0)=P(X_j=1)=1/2$. The Rademacher functions are such a sequence (except that they take the values $1,-1$ instead of $0,1$).


Ok, a detailed proof that $f$ is a bijection except for countably many points. Say $f(s)=f(t)$ but $s\ne t$. Choose $n$ so $s_j=t_j$ for all $j<n$ but $s_n\ne t_n$. WLOG $s_n=1$, $t_n=0$. So we have $$2^{-n}+\sum_{j=n+1}^\infty s_j2^{-j}=\sum_{j=n+1}^\infty t_j2^{-j}.$$

So $$\sum_{j=n+1}^\infty t_j2^{-j}\ge 2^{-n}.$$But$$\sum_{j=n+1}^\infty 2^{-j}=2^{-n}.$$So we must have $$t_j=1\quad(j\ge n+1);$$ now this shows that $$\sum_{j=n+1}^\infty s_j2^{-j}=0,$$so $$s_j=0\quad(j\ge n+1).$$

So both sequences $s$ and $t$ are eventually constant; there are only countably many such sequences. (The above also shows that there cannot be a third sequence with the same sum...)

$\endgroup$
  • $\begingroup$ Thanks for your reply. I understand the logic but there are several things I don't know how to prove. Q1, why there are only countably many $x$ has two binary representation? Q2, how to show $f$ preserve measure for all $C$ in $\sigma$-algebra generated by clylinder? Q3, to show the original question, do we need to show the inverse of $f$ is also measure-preserving? Thanks! $\endgroup$ – John Jan 20 '16 at 15:06
  • $\begingroup$ Q1 Think about it. Q2 Show the set of $C$ such that $C$ and $f(C)$ have the same measure is a $\sigma$-algebra. Q3 I don't know whether that's required, depends on exactly what the problem asks. But it follows, since $f$ is a bijection, except for a null set. $\endgroup$ – David C. Ullrich Jan 20 '16 at 15:10
  • 1
    $\begingroup$ Q1: Ok, see edit. $\endgroup$ – David C. Ullrich Jan 20 '16 at 15:46

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.