Let $(\Omega,\mathcal{F},\mathbb{P})$ be a probability space.

Let $p\ge 1$, $\left(X_n\right)_{n\ge 0} $ is a sequence of real random variables in $L^{p}(\Omega)$ such as : $$ \lim_{n} X_n = X \quad \text{in}\,\,L^{p}(\Omega) $$ Let $\mathcal{B}$ is a sub- $\sigma$ -algebra of $\mathcal{F}$ .

Show that : $$ \lim_{n} \mathbb{E}[X_n\mid \mathcal{B}]= \mathbb{E}[X\mid \mathcal{B}]\quad \text{in}\,\,L^{p}(\Omega) $$ With : $\mathbb{E}[X\mid \mathcal{B}]$ is the conditional expectation of $X$ given $\mathcal{B}$.

  • 1
    $\begingroup$ Dominated convergence theorem? $\endgroup$ – Ramanujan Nov 17 '19 at 15:02
  • $\begingroup$ (conditional) Expectation is an integral and the above theorem gives you $\int f_n d \mu \to \inf f d \mu$ if $f_n \to f$ point wise and $| f_n | \le g$ for some measurable $g$. That are the two things you have to check. $\endgroup$ – Ramanujan Nov 17 '19 at 15:19
  • $\begingroup$ @ViktorGlombik That's not true... $g$ needs to be integrable! In general there need not exist an integrable dominating function. $\endgroup$ – saz Nov 17 '19 at 15:22
  • $\begingroup$ @saz of course. You are correct. $\endgroup$ – Ramanujan Nov 17 '19 at 15:26

By Jensen's inequality for conditional expectations, we have

$$|\mathbb{E}(Y \mid \mathcal{B})|^p \leq \mathbb{E}(|Y|^p \mid \mathcal{B})\tag{1}$$

for any $p \geq 1$ and $Y \in L^p$.

Let $X_n \to X$ in $L^p$ for some $p \geq 1$. We need to check that

$$\lim_{n \to \infty} \mathbb{E}\bigg[ \bigg| \mathbb{E}(X_n \mid \mathcal{B}) - \mathbb{E}(X \mid \mathcal{B}) \bigg|^p \bigg]=0.$$


$$\mathbb{E}(X_n \mid \mathcal{B}) - \mathbb{E}(X \mid \mathcal{B})=\mathbb{E}(X_n-X \mid \mathcal{B}),$$

it follows from $(1)$ that

$$|\mathbb{E}(X_n \mid \mathcal{B}) - \mathbb{E}(X \mid \mathcal{B})|^p \leq \mathbb{E}(|X_n-X|^p \mid \mathcal{B}).$$

Taking expectation on both sides and using the tower property for conditional expectation, we obtain that

\begin{align*} \mathbb{E}\bigg[ \bigg| \mathbb{E}(X_n \mid \mathcal{B}) - \mathbb{E}(X \mid \mathcal{B}) \bigg|^p \bigg] &\leq \mathbb{E} \bigg[ \mathbb{E}(|X_n-X|^p \mid \mathcal{B}) \bigg] \\ &= \mathbb{E}(|X_n-X|^p) \xrightarrow[]{n \to \infty} 0.\end{align*}


You might want to use, that for any convex function $f$ we have that $$f(\mathbb{E}[ X| \mathcal{B}]) \leq \mathbb{E}[ f(X)| \mathcal{B}]\quad \text{(almost surely)} ,$$ therefore $$|\mathbb{E}[X_n \: | \: \mathcal{B}]-\mathbb{E}[X \: | \: \mathcal{B}]|^p =|\mathbb{E}[X_n-X \: | \: \mathcal{B}]|^p \leq \mathbb{E}[|X_n-X|^p \: | \: \mathcal{B}] $$ almost surely, since $x\mapsto |x|^p$ is convex for $p\geq 1$.


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.