I have a suspicion that the following is true: $\def\AAA {\mathcal{A}} \def\BBB {\mathcal{B}} \def\NN {\mathbb{N}}$
Theorem: let $(X,\AAA,\mu)$ be a complete measure space, and $(Y,\BBB)$ a measurable space where $\BBB$ is the Borel $\sigma$-algebra of some metrizable topology of $Y$. Then, the following are equivalent:
- $f:X\to Y$ is measurable.
- $f$ is almost everywhere the pointwise limit of measurable functions $f_n$.
What I do know.
- The result holds if $f_n\to f$ everywhere. For a proof, see here.
- The result does not hold if the domain space is not complete (a counterexample can be constructed from this post).
My attempt.
I have been trying to modify this proof so as to deal with the case where $f_n\to f$ only almost everywhere. I first partition $X$ into $$X_* := \{ x\in X : f_n(x)\to f\} \ \ \ \ \text{ and } \ \ \ \ X_0 := X\setminus X_*.$$ For any closed $C\subseteq Y$, completeness of $X$ allows us to claim $$\Bigg(f^{-1}(C)\Bigg) \triangle \Bigg(\bigcap_{n\in\NN}\bigcup_{N\in\NN}\bigcap_{k\ge N}f_k^{-1}(C_{2^{-n}})\Bigg)$$ is a null set, call it $C_0$. As the set on the right, call it $M$, is measurable by hypothesis, we have $$\big(f^{-1}(C)\setminus M\big)\cup\big(M\setminus f^{-1}(C)\big) = C_0$$ which hopefully, somehow, implies $f^{-1}(C)$ is measurable as well.
Is the 'Theorem' correct? If so, how could one prove it? If not, what is a counterexample?
Related.
- This post. I struggle to understand large part of the terminology, so it hasn't been of much help.