The Borel sets of $[0;1]$ are defined to be the smallest $\sigma$-algebra which contains all the subintervals $\subseteq [0;1]$.

Is there a construction where we can obtain it as a limit of a countable sequence of finite algebra? Or as a limit of limits of countable sequences.

For instance something similar (in form) to : $$\lim_{n \to \infty} \bigcup_{k = 1}^{n} A_k \quad \text{where $A_k$ is a finite algebra}$$

  • $\begingroup$ You mean something like the following: Let $\mathcal{A}_n$ be the $\sigma-$algebras of the Borel subsets of $\left[0,\frac{n+1}{n}\right]$. Then $\mathcal{B}_n\to\mathcal{A}$ where $\mathcal{A}$ is the $\sigma-$algebra of all the Borel subsets of $[0,1]$? $\endgroup$ – Βασίλης Μάρκος May 9 '18 at 12:45
  • 1
    $\begingroup$ Hum, no, I will edit to be more clear. Edit: done -- I'm looking for a limit of algebra $\endgroup$ – Julien__ May 9 '18 at 12:49
  • $\begingroup$ Your edit is unreasonable, but transfinite induction could help. $\endgroup$ – Olivier May 9 '18 at 12:58
  • $\begingroup$ What does limit mean in this context? Are we just asking about $\bigcup_{k=1}^{\infty}A_k$, or does the limit mean something else? (Also, I think you have a mistake in your subscript in $A_n$) $\endgroup$ – Milo Brandt May 9 '18 at 12:59
  • $\begingroup$ @MiloBrandt subscript fixed thanks. I'm not sure I understand your question. The limit $\lim_{n \to \infty} u_n$ of any sequence of algebra ? $\endgroup$ – Julien__ May 9 '18 at 13:01

Yes there is a natural and useful construction of $\sigma$-algebras as a "limit". But not as the limit of a sequence, rather the limit of an uncountable collection, indexed by the countable ordinals (something logicians might call an $\omega_1$-sequence, but which the rest of us don't usually call a "sequence", since it's not countable).

First, given $A\subset\mathcal P(X)$, let $S(A)$ consist of all countable unions of elements of $A$ together with the complements of elements of $A$.

Now suppose $A\subset\mathcal P(X)$. By "transfinite recursion" there exists a family $(A_\alpha)_{\alpha<\omega_1}$ of subsets of $\mathcal P(X)$ with $$A_0=A$$and $$A_\alpha=S\left(\bigcup_{\beta<\alpha}A_\beta\right)$$for $\alpha>0$. Then it's not hard to show that$$\sigma(A)=\bigcup_{\alpha<\omega_1}A_\alpha$$is precisely the $\sigma$-algebra generated by $A$. (Of course we haven't defined what we mean by "limit" here, but it's natural to regard this as the limit of $A_\alpha$ as $\alpha\to\omega_1$.)

Proof. Say $\sigma(A)$ is defined as above and let $\sum(A)$ be the $\sigma$-algebra generated by $A$. Note first that if $F$ is a $\sigma$-algebra and $B\subset F$ then $S(B)\subset F$ by definition. So since $A\subset\sum(A)$ it follows by transfinite induction that $A_\alpha\subset\sum(A)$ for every $\alpha<\omega_1$, hence $\sigma(A)\subset\sum(A)$.

So we need only show that $\sigma(A)$ is a $\sigma$-algebra. This is clear. If $E\in\sigma(A)$ then there exists $\alpha < \omega_1$ with $E\in A_\alpha$; hence $X\setminus E\in A_{\alpha+1}$, so $X \setminus E \in \sigma(A)$.

Suppose that $E_1,E_2 \dots \in \sigma(A)$. For every $n$ there exists $\alpha_n<\omega_1$ with $E_n\in A_{\alpha_n}$. Since $\omega_1$ is uncountable there exists $\alpha<\omega_1$ with $\alpha>\alpha_N$ for every $n$; hence $\bigcup E_n\in A_\alpha\subset\sigma(A)$ $\square.$

This is useful, for example as far as I know it's the only way to show that the Borel algebra on $\Bbb R$ has cardinality $\mathfrak c$.


The answer is negative. Let $\mathcal{A_n}$, $n\in\mathbb{N}$ be a collection of finite algebras such that: $$\mathcal{B}=\bigcup_{n=1}^\infty\mathcal{A}_n$$ where $\mathcal{B}$ is the $\sigma-$algebra of all the Borel subsetes of $\mathbb{[0,1]}$. Now, on the one hand we have that $\#\mathcal{B}$ is uncountable, on the other hand $\#\bigcup\limits_{n=1}^\infty\mathcal{A}_n$ is obviously countable - countable union of finite sets.

As for your edit, it is also false that: $$\bigcup_{n=1}^\infty\mathcal{P}(\{1,\dots,n\})=\lim_{k\to\infty}\bigcup_{n=1}^k\mathcal{P}(\{1,\dots,n\})$$ is uncountable, since it is a countable union of finite sets. It is important to see that: $$\bigcup_{n=1}^\infty\mathcal{P}(\{1,\dots,n\})\underset{\neq}{\subset}\mathcal{P}(\mathbb{N})$$ since $\mathbb{N}\in\mathcal{P}(\mathbb{N})$ but $\mathbb{N}\not\in\bigcup\limits_{n=1}^\infty\mathcal{P}(\{1,\dots,n\})$ - obviously, since there is not $n\in\mathbb{N}$ such that $\mathbb{N}\in\mathcal{P}(\{1,2,\dots,n\})$.

Edit/Further Info: What is true is that: $$\bigcup_{n=1}^\infty\mathcal{P}(\{1,\dots,n\})=\{A\mid A\subset\mathbb{N},\ A\text{ is finite}\}.$$


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.