I've seen the law of large numbers stated mainly in two (or three) forms: $S_n/n$ converges in probability (weak law) and converges almost surely (strong law). Also, there is convergence in the $L^2$-norm for uncorrelated random variables ($L^2$ weak law).

However there is a backwards martingale proof of the strong law of large numbers (any graduate level probability theory book, for example Durrett, should have it). The important thing is that $M_{-n}:=S_n/n$ is a backward martingale, and backward martingales converge both a.e. and in the $L^1$-norm. Then, in particular, $S_n/n$ converges in the $L^1$-norm.

Does $S_n/n$ really converge in the $L^1$-norm?

  • If yes, why is this never mentioned?
  • If no, what is wrong with my above proof?

1 Answer 1


Yes, $\bar X_n = S_n / n$ converges to $\mu$ in $\mathscr{L}_1$. It suffices to show that $\bar X_n$ is uniformly integrable since we already know $\bar X \to \mu$ in probability. This follows very quickly from the fact that the $X_i$ are trivially uniformly integrable (just fix $\epsilon > 0$ and choose $\delta > 0$ so that if $P(A) < \delta$ then $\int_A |X_n| \ dP < \epsilon$ for all $n$ and show this $\delta$ also works for $\bar X_n$). You also need to show $\sup E|\bar X_n| < \infty$ but this follows from the triangle inequality since $X_1 \in \mathscr L_1$.

I don't know that I would say that this isn't mentioned. I know it is an exercise in Chung's book, and I would be surprised if it wasn't also in Billingsley somewhere. People don't seem to care that much about $\mathscr L_p$ convergence around these parts because a.s. convergence and convergence in $P$ seem to come up more in practical settings, but that's just the way it seems to me as a statistics student.

  • $\begingroup$ It's also an exercise in Resnick's A Probability Path. And I assigned it as homework in a course last year :) $\endgroup$ Nov 3, 2011 at 3:16
  • $\begingroup$ Thanks! I probably didn't look hard enough. I was mostly looking in Durrett and on the Internet. (Maybe it was even an exercise Durrett and I missed it.) $\endgroup$
    – Jason Rute
    Nov 3, 2011 at 3:27
  • $\begingroup$ L1 boundedness is not sufficient for uniform integrability. This answer is false. $\endgroup$
    – Daniel Li
    Apr 3, 2019 at 16:25
  • $\begingroup$ The correct way to do it is to use the lemma that $E|X|<\infty$ implies that there exists a convex function $\phi$ such that $\phi(x)/x \to \infty$. And then show that $\sup_n E\phi|\bar{X}_n| < \infty$ $\endgroup$
    – Daniel Li
    Apr 3, 2019 at 16:33
  • $\begingroup$ @DanielLi the $X_i$'s are uniformly integrable because they are iid; we are using more than $L_1$-boundedness. You then use that to show the average $\bar X_n$ is uniformly integrable. $\endgroup$
    – guy
    Apr 3, 2019 at 19:12

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.