Let $(\Omega, \mathcal{A}, P)$ be a probability space and consider a square-integrable random variable $X$ and a sub-sigma-Algebra $\mathcal{F}$.

Denote $Z:=E[X \mid \mathcal{F}]$.

Conditional expectation is the unique minimizer (orthogonal projection), meaning that

$E[(X-Z')^2] \geq E[(X-Z)^2]$ for all $Z' \in L^2(\Omega, \mathcal{F},P)$.

Now we have seen the proof of this, deriving:

$E[(X-Z')^2] \geq E[(X-Z)^2] + E[(Z-Z')^2]$

Now it is written that we have equality iff $Z=Z'$. But why? Couldn't it be that $E[(Z-Z')^2]=0$ but $Z\neq Z'$?


Since $(Z-Z')^2 \geq 0$, its expectation is zero if and only if $Z=Z'$ with probability $1$. We sometimes just write this as $Z=Z'$; in particular, $Z$ and $Z'$ are indeed equal if you identify them with their $L^2$ (or any $L^p$) equivalence classes.


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.