I have a question on the proof Durrett (p. $190$) gives for the uniqueness of the conditional expectation function.

If I understand his proof correctly, here is what I think it is saying:

Suppose $Y, Y'$ both satisfy the criteria to be a conditional expectation function for a random variable $X$ given a $\sigma$-algebra $\mathcal{A}$. Then if $A_{\epsilon} := \{ \omega \mid Y - Y' \geq \epsilon \}$, clearly $A_{\epsilon} \in \mathcal{A}$, and so by the criteria we have for conditional expectation, we have:

$$0 = \int \limits_{A_{\epsilon}} Y - Y' \,dP \geq \epsilon P(A_{\epsilon}) \geq 0$$ and this shows that $P(A_{\epsilon}) = 0$ for all $\epsilon > 0$.

Ok, then we note that $0 \leq P( \{\omega \mid Y - Y' > 0 \} ) = P(\bigcup \limits_{n =1}^{\infty} A_{\frac{1}{n}}) \leq \sum \limits_{n = 1}^{\infty} P(A_{\frac{1}{n}}) = \sum \limits_{n = 1}^{\infty} 0 = 0$. Thus, $P( \{\omega \mid Y - Y' \leq 0 \} ) = 1$, so $Y \leq Y'$ almost surely.

We can repeat the same proof switching $Y$ and $Y'$ to get $Y' \leq Y$ almost surely, thus they are equal almost surely.

My question:

I think the following proof is a much simpler one (using the fact that $\int \limits_{\Omega} X \,dP = 0 \implies X = 0$ almost surely). Is there any flaw in the argument below? If not, why did Durrett decide to do so many extra steps?

Since $\Omega \in \mathcal{A}$, we have $0 = \int \limits_{\Omega} Y - Y' \,dP$, and thus $Y - Y' = 0$ almost surely, so $Y = Y'$ a.s.


The problem with your argument is that $\int_{\Omega} X\,dP = 0$ does not imply that $X=0$ almost surely, unless you already know that $X \geq 0$ almost surely. Since you don't know that $Y-Y' \geq 0$ in your argument, your conclusion does not follow.

| cite | improve this answer | |
  • $\begingroup$ I am ridiculous for forgetting that it only holds when $X \geq 0$ a.s. Thank you for your sharp eye! $\endgroup$ – layman May 13 '15 at 18:58
  • $\begingroup$ Why is the limiting argument desirable? It should be enough to consider the events $A_1 = \{ Y \geq Y' \}$ and $A_2 = \{ Y \leq Y' \}$. We would have $Y_1 = Y_2$ a.s. on $A_1 \cup A_2$. The result follows since $\Omega \subset A_1 \cup A_2$. $\endgroup$ – Benjamin Bray Oct 15 '18 at 16:06
  • $\begingroup$ @BenBray Why does $Y_1 = Y_2$ a.s. on $A_i$ (for either $i$)? Why couldn't it be that $\int_{A_1} ( Y-Y') = \int_{A_2}(Y - Y') \neq 0$? $\endgroup$ – Tom Oct 19 '18 at 13:09

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.