Exercise 5.3.1 in Durrett's "Probability Theory and Examples" states

Let $X_n$, $n\ge 0$, be a submartingale with $\sup X_n < \infty$. Let $\xi_n=X_n-X_{n-1}$, and suppose $E(\sup \xi_n^+)<\infty$. Show that $X_n$ converges almost surely.

Here are my thoughts: It looks like we need to use theorem 5.2.8 (Martingale convergence theorem), which states that $X_n$ converges almost surely if $\sup EX_n^+<\infty$. I am trying to understand why we can just do the following (or at least I'm guessing we can't):

Suppose $X_n > 0$ (edit: on some set with positive measure), then $0< \sup X_n < \infty$, and therefore $EX_n^+ \le \sup X_n < \infty$. So we're done.

Otherwise, if $X_n \le 0$ a.e., then $\sup X_n^+ = 0$ a.e. Once again, we're done.

Am I missing something? Since I didn't use all the information, I probably am.

Thanks for the help.

  • $\begingroup$ Actually I think you can, but I also think the point of the problem was to get you to try to use Theorem 5.3.1 and deduce that $P(D) = 0$. $\endgroup$
    – gogurt
    Dec 16 '15 at 17:47

You've stepped off a cliff in deducing that the expectation $E(X^+_n)$ is bounded above by the random variable $\sup_n X_n$.

Suggestion: Employ the argument used by Durrett in the proof of his Theorem 5.3.1. Fix a positive real $K$, define the stopping time $T=T_K$ to be the first time $n$ that $X_n$ is larger than $K$, and observe that the stopped process satisfies $$ X_{n\wedge T}\le K+\sup_m\xi_m^+, $$ so that $$ E(X_{n\wedge T})\le K+E(\sup_m\xi_m^+)<\infty,\qquad\forall n. $$ Now apply the submartingale convergence theorem to the stopped process. This yields a.s. convergence of $X_n$ on the event $\{T_K=\infty\}=\{\sup_mX_m\le K\}$. Finally, vary $K$.


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.