Let $X_1,\ldots,X_n$ be $\{0,1\}$-valued random variables drawn from some joint distribution. Let $\tilde X_1,\ldots,\tilde X_n$ be their independent version: $\mathbb{E}X_i=\mathbb{E}\tilde X_i$ for each $1\le i\le n$ and the $\tilde X_i$ are mutually independent (as well as being independent of the $X_i$'s). I would like to compare $M:=\mathbb{E}\max_{1\le i\le n}X_i$ and $\tilde M:=\mathbb{E}\max_{1\le i\le n}\tilde X_i$. It's obvious that $M$ can be made arbitrarily small, while $\tilde M$ is arbitrarily close to $1$, so in general, there can be no bound on $\tilde M/M$. But it seems that in the reverse situation, one should be able to say something. Is some bound of the form $$ M \le c\tilde M $$ known, where $c$ is an absolute constant? If such a thing is impossible (what's the counterexample?), what's the best dependence on $n$ in the bound?


1 Answer 1


Yes, this inequality holds with $$c:=\frac e{e-1}.$$

Indeed, let $A_i:=\{X_i=1\}$. Then $$M=P\Big(\bigcup_i A_i\Big)\le\min(s,1)\le c(1-e^{-s}),$$ where $s:=\sum_i P(A_i)$. On the other hand, $$\tilde M=1-\prod_i(1-P(A_i)) \ge1-\prod_i\exp(-P(A_i))=1-e^{-s}.$$ So, $M\le c\tilde M$.

  • $\begingroup$ Beautiful! Was anything like this known before (for reference)? $\endgroup$ May 16, 2022 at 5:39
  • 1
    $\begingroup$ Thank you Aryeh for your appreciation. I think this may be somewhere in the existing literature, but I don't recall seeing it anywhere. $\endgroup$ May 16, 2022 at 10:56

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.