3
$\begingroup$

Let $(X_n)$ be an increasing sequence of real valued integrable rvs on a probability space $(\Omega,\mathcal{F},P)$, such that $(X_n)$ converges ae to some rv $X$. Is it true that $E(X_n)\rightarrow E(X)$?

Some thoughts: Ι guess not. If $X_n$'s were positive then we could apply the monotone convergence theorem, but now we are looking for a counterexample. I am working on $[0,1]$, equipped with the Lebesgue measure and looking for discrete rvs that do the job, but no matter how I change their formula, I seem not to be getting the desired non-convergence.

Thanks a lot in advance!

$\endgroup$
7
  • $\begingroup$ split the random variable up into positive and negative parts and apply the monotone convergence theorem to each. $\endgroup$ – user159517 Feb 10 '16 at 1:12
  • 1
    $\begingroup$ @user159517 which could give you $\infty-\infty$ $\endgroup$ – snarfblaat Feb 10 '16 at 1:20
  • $\begingroup$ @user159517, you claim that since $X_n \rightarrow X$ ae, then $X_n^{+}\rightarrow X^{+}$ and $X_{n}^{-}\rightarrow X^{-}$? (if that holds, then things are a bit straightforward, since $X_n^{+}\nearrow X^{+},~X_n^{-}\nearrow X^{-}$ and all the rvs are positive valued). $\endgroup$ – Nikolaos Skout Feb 10 '16 at 1:22
  • 5
    $\begingroup$ The sequence $X_n-X_1$ of non-negative, integrable random variables increases to $X-X_1.$ $\endgroup$ – user940 Feb 10 '16 at 1:30
  • $\begingroup$ Not an answer, but shouldn't this be true for weaker condition ($X_n\rightarrow X$ in probability/measure), by absolute continuity of integral w.r.t. measure? $\endgroup$ – user175968 Feb 10 '16 at 2:00
4
$\begingroup$

You said that the variables $X_n$ are integrable. This means that $E |X_n| < \infty$ for all $n$. Then $X_n - X_1$ make an increasing sequence of nonnegative random variables. By monotone convergence theorem, $$ E(X_n - X_1) \to E \lim_{n\to\infty}(X_n - X_1) = E X - E X_1\qquad (1) $$ with finite $E X_1$. ($E X$ is either finite or is equal to $+\infty$). The desired convergence is obtained by adding $E X_1$ to both sides of (1).

If $X_n$ are measurable, but not necessarily integrable, then we can construct a counterexample. Let $U$ be a random variable uniformly distributed on $(0, 1)$. Let $$ X_n = \begin{cases} -1/U & \text{if}~ U < 1/n, \\ 0 & \text{if}~ U \ge 1/n. \end{cases} $$ Then the sequence of random variables $(X_n)$ is increasing in $n$ (surely) and is converging to $0$ (surely). But $E X_n = -\infty$ for all $n$, and $E 0 = 0$. Thus, $E X_n$ do not tend to $E \lim_{n\to\infty} X_n$.

$\endgroup$

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.