For a probability space $(\Omega, \mathcal{F}, \mathbb{P})$, we say $X: \Omega \rightarrow \mathbb{R}$ is a random variable if $X^{-1}(B) \in \mathcal{F} ~\forall B \in \mathcal{B}$, where $\mathcal{B}$ is the Borel $\sigma$-field. In other words, $X$ is a random variable if $X$ is measurable.

The intuition I've gathered so far is that since we can only talk about measurable subsets of $\Omega$, we want the preimage of any set in the image of X to be measurable. However, my question is, why are we only concerned with preimages of Borel sets, rather than all subsets of $\mathbb{R}$?

(For context, my experience with measure theory is only in the view of probability theory)

  • 1
    $\begingroup$ Measurability is always related to two sigma-algebras, one on the source set and the other on the target set. There is no reason to chose the power set as the latter (and actually one rarely does so). $\endgroup$ – Did Oct 31 '16 at 21:37
  • $\begingroup$ The equivalence of the two definitions seems to imply that the Borel $\sigma$-field is the largest collection of measurable subsets of $\mathbb{R}$, but I feel like that's not the case. $\endgroup$ – dkv Oct 31 '16 at 21:38
  • $\begingroup$ "The equivalence of the two definitions" What? If "the two definitions" are what I think, then they are not equivalent (and one of them is never used). $\endgroup$ – Did Oct 31 '16 at 21:43
  • $\begingroup$ $X^{-1}(B) \in \mathcal{F} ~\forall B \in \mathcal{B}$ vs "$X$ is measurable." Do these not mean the same thing? $\endgroup$ – dkv Oct 31 '16 at 21:47
  • $\begingroup$ You might want to check this $\endgroup$ – Ran Wang Feb 23 '17 at 20:23

In fact, it is enough to be concerned about sets $\{X > a\}$ with $a$ real.

For $X$ to be considered a "random variable" we need to be able to tell whether or not it is ${} > a$ for any given real number $a$.

You can try to work with "all subsets of $\mathbb R$", but you won't get very far. Why not use only minimal assumptions? We do not want to use the largest possible collection of sets; that is, we do not want to make it most difficult to be a "random variable". Instead we want to use the smallest possible collection of sets; that is, we want to make it very easy to be a "random variable".

  • $\begingroup$ Right, but that's exactly equivalent to the Borel-set definition above since $(a,\infty)$ is a generating set for $\mathcal{B}$. $\endgroup$ – dkv Oct 31 '16 at 21:55
  • $\begingroup$ So then I guess my question is, why use Borel sets, rather than some simpler sigma-field on R? Why specifically Borel? What makes those sets special? $\endgroup$ – dkv Oct 31 '16 at 21:56
  • $\begingroup$ The Borel sets are not special. Comparison with a constant, that is what is special. $\endgroup$ – GEdgar Oct 31 '16 at 21:57
  • $\begingroup$ Why only a constant? What about {a<X<b} or {a<X<b and c<X<d} or {c<X<d and X=4}? Sure, those can be turned into comparisons to a constant, but I can keep going and in the most general case ask about {X $\in$ S} for any S $\subset \mathbb{R}$? $\endgroup$ – dkv Oct 31 '16 at 22:11

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.