Disclaimer: sorry for my poor english and edition.

Claim: If $M\subseteq \mathcal{P}(X)$, then $\Sigma(M)=M_3$, where:

  • $\Sigma(M)$ is the sigma-algebra generated by $M$
  • $M_1=\{A\subseteq X:(A\in M) \lor(A^c\in M)\}$,
  • $M_2=\{\cap_{i\in I} A_i:A_i\in M_1,|I|\leq |\mathbb{N}|\}$
  • $M_3=\{\cup_{j\in J} A_j:A_j\in M_2,|J|\leq |\mathbb{N}|\}$
  • $|I|\leq |\mathbb{N}|$ means $I$ is countable

(Note: Is easy to prove that my claim is equivalent to say that $M_3$ is a sigma-algebra)

I proved that $\mathcal{A}(M)=M'_3$, where:

  • $\mathcal{A}(M)$ is the algebra generated by $M$
  • $M'_2=\{\cap_{i\in I} A_i:A_i\in M_1,|I|< |\mathbb{N}|\}$
  • $M'_3=\{\cup_{j\in J} A_j:A_j\in M'_2,|J|< |\mathbb{N}|\}$
  • $|I|< |\mathbb{N}|$ means $I$ is finite

But my proof uses induction (on $|J|$ to iterate "intersections distributes over unions") to guarantee that $M'_3$ is close under complements,* and therefore fails to prove my claim. However, this fact allows me to reformulate my claim like this: $\mathcal{M}(M'_3)=M_3$, where $\mathcal{M}(M'_3)$ is the monotone class generated by $M'_3$.

*More explicitly: $$A^c=\cap_{j\in J}\left(\cup_{i_{j}\in I_{j}}A_{i_{j},j}\right)=\cup{}_{i_{1}\in I_{1}}\left(\dots\left(\cup{}_{i_{n}\in I_{n}}\left(\cap_{j\in J}A_{i_{j},j}\right)\right)\right)$$ In fact, $$A^c=\cup{}_{\left(i_{1}\dots i_{n}\right)\in I_{1}\times\dots\times I_{n}}\left(\cap_{j\in J}A_{i_{j},j}\right)=\cup{}_{i\in I}\left(\cap_{j\in J}A_{i,j}\right)$$ with $I=\prod_i^{|J|}I_i$, where $J$ needs to be finite (I think, because my proof of this is by induction on $|J|$ via "intersections distributes unions"), but $I_i$ can be countable.

(Note that, even if I admit $J$ infinite and then $I=\prod_i^\infty I_i$, then $I$ is uncountable, link, and therefore $A^c$ is not necessarily of $M_3$)

Then I saw a description of "sigma-algebra generated by collection of subsets" here that suggested me that I was wrong, and then exists $X,M$ such that $\Sigma(M)\neq M_3$ (or $\mathcal{M}(M'_3)\neq M_3$).

Because my above cited proof fails when $J$ is infinite, I thought to search "big" $M$ for a counterexample, e.g., open sets (and then $\Sigma(M)$ are the borel sets). I found that exists borel sets that are not in $(G_\delta)_\sigma$, but $(G_\delta)_\sigma\subseteq M_3$ and I'm not sure that the other inclusion holds.

I try and find more things, but the above are (I think) the more relevant. Any help?

P.S. (related questions)

1) In 1 and 2 exists proofs by transfinite induction that $\Sigma(M)=M_{\omega_1}$ (similar to the above cited proof of Asaf Karagila), but I insist: ¿is this the finest construction? 3 examine this question but not solve my problem (if I understood correct: my set theory is also poor).

2) A very close question (without desired answer) is 5. Point of view of Victor in the conversation with Matt is similar to mine (up to myside bias): if $X$ is finite (equivalent, $\mathcal{P}(X)$ is finite), then the process stop somewhere, but otherwise it can escape of $M_3$; in other words, the same idea that (1).

3) A similar but different question is 4.


For an easier example, try letting $X = \mathbb{R}$ and $M = \{(a, +\infty) : a \in \mathbb{R}\}$ the set of all open intervals that extend to infinity on the right.

Then $M_1 = \{(a, +\infty), (-\infty, a] : a \in \mathbb{R}\}$ consists of open and closed intervals.

Any intersection of intervals is again an interval (easy exercise) so $M_2$ consists only of (open, closed, half-open) intervals.

Thus every set in $M_3$ is a countable union of intervals.

Let $E$ be the irrational numbers. $E$ doesn't contain any nontrivial intervals, so if we are to write $E$ as a union of intervals, each of them must be a single point (like $[a,a] = \{a\}$). But since $E$ is uncountable we cannot write it as a countable union of points.

However, since $M_2$ contains all open intervals, $M_3$ contains all open sets, so $\Sigma(M) = \Sigma(M_3)$ is the Borel $\sigma$-algebra, and $E$ is certainly a Borel set.

| cite | improve this answer | |

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.