Is it known whether the following version of the strong law of large numbers holds?

For each $k\in\mathbb{N}$, let $\Omega_k$ be a finite set and $\mu_k$ be a probability measure on $\Omega_k$. Let $(X_k)_{k\in\mathbb{N}}$ be a sequence of random variables, $X_k: \Omega_k \to [0,+\infty)$. Suppose in addition that $\sup_k \int X_k\,d\mu_k < \infty$.

Then with $\prod_{k=1}^\infty \mu_k$-probability one, we have $\lim_{N\to\infty}\frac{1}{N} \sum_{k=1}^N (X_k - \int X_k\,d\mu_k) = 0$.

  • $\begingroup$ What is a "Bernoulli probability space" and what do you mean by "Bernoulli random variables" in the title? Normally, "Bernoulli random variables" are ones taking only values $0$ and $1$ - but then your additional condition of uniformly bounded expectations would be trivial, so I imagine you mean something else. $\endgroup$ Sep 21 at 10:42
  • $\begingroup$ Tried to clarify the question, removed mentions of "Bernoulli probability spaces" by which I meant a finite probability space $\endgroup$
    – Aleksi
    Sep 21 at 11:41
  • 2
    $\begingroup$ Unless I misunderstood your question, no: math.stackexchange.com/questions/3871332/… $\endgroup$
    – unwissen
    Sep 21 at 12:43
  • $\begingroup$ But maybe you have use for a version of the strong law which holds when the variances of $(X_k)_k$ do not grow too fast (as mentioned in en.wikipedia.org/wiki/Law_of_large_numbers#Strong_law) $\endgroup$
    – unwissen
    Sep 21 at 12:51

1 Answer 1


Boundedness of the expectations is not enough.

You could consider the case where $X_k$ takes value $0$ with probability $1-1/(k+1)$ and value $k+1$ with probability $1/(k+1)$, so that $\mathbb{E} X_k=1$ for all $k$.

Then by Borel-Cantelli, with probability $1$ there are infinitely many $k$ with $X_k-\mathbb{E} X_k=k$, which means that $\frac{1}{N}\sum_{k=1}^N \left(X_k-\mathbb{E} X_k\right)$ does not converge to $0$ as $N\to\infty$.

By the way, giving all these details about the structure of the probability space is unusual and seems irrelevant. All that matters is the joint distribution of the random variables $X_k$. You could just as well say: let $X_k$ be a sequence of independent non-negative random variables, each one taking finitely many values.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.