3
$\begingroup$

Let $\{X_n\}$ be a collection of positive random variable with $X_n \rightarrow X$ in probability. Prove that if $E(X_n) \rightarrow E(X)$, then $X_n \rightarrow X$ in $L^1$.

My partial answer: Let $(\Omega,B,P)$ be the probability space. For any $\varepsilon>0$, we have \begin{align*} \int_{\Omega} |X_n-X| dP &= \int_{\{|X_n-X|<\varepsilon\}} |X_n-X| dP+\int_{\{|X_n-X|\geq \varepsilon\}} |X_n-X| dP \\ &<\varepsilon P(|X_n=X|<\varepsilon)+\int_{\{|X_n-X|\geq \varepsilon\}} |X_n-X| dP \\ &<\varepsilon +\int_{\{|X_n-X|\geq \varepsilon\}} |X_n-X| dP \end{align*} The problem is I don't know how to estimate the second term in RHS with information convergence in probability and expectation convergence.

Thanks for your help.

$\endgroup$
  • 1
    $\begingroup$ I'm not sure your approach will easily work. But, one way to prove your statement is to apply the Dominated Convergence Theorem to the sequence $(g_n)$, where $g_n=|X_n-X|+X-X_n$. Note $0\le g_n\le 2X$ for each $n$. $\endgroup$ – David Mitra May 2 '13 at 1:17
  • $\begingroup$ $X_n$ doesn't converge pointwise to $X$ but only converge in probability. Meanwhile, the hypothesis of DCT need $X_n$ converge pointwise to $X$ $\endgroup$ – tes May 2 '13 at 1:40
  • 1
    $\begingroup$ DCT holds also for convergence in measure. See here. $\endgroup$ – David Mitra May 2 '13 at 1:41
  • $\begingroup$ Also $E(X)$ need not be finite. $\endgroup$ – tes May 2 '13 at 1:51
3
$\begingroup$

Let $Y_n:=|X-X_n|+X-X_n$. Then $0\leqslant Y_n\leqslant 2X$. As $E(X_n)\to E(X)$, we just need to show that $E(Y_n)\to 0$. As $X_n\to X$ in measure, we have $Y_n\to 0$ in measure. We can conclude by dominated convergence.

$\endgroup$

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.