Regarding the second line of the first proof by @JDL in Altenate definitions of almost sure convergence

Are the following equivalent?

$P(\omega:\exists n\in\mathbf{N}, \forall m\in\mathbf{N}, \exists i>m \,\,\, \text{s.t.} \,\, |X_i(\omega) - X(\omega)| < 1/n)$


$P(\omega:\forall n\in\mathbf{N}, \exists m\in\mathbf{N},\ \text{s.t.}\ \forall i>m \,\, |X_i(\omega) - X(\omega)| \ge 1/n)$

  • 1
    $\begingroup$ FYI, the answer you refer to is certainly not the most direct nor the clearest one that one can imagine, and it confuses several times the union of events $\{Y_n\geqslant y\}$ with the event $\{\sup Y_n\geqslant y\}$. Even if one denotes unions as supremums, it is not true that $$\sup_n\{Y_n\geqslant y\}=\{\sup_nY_n\geqslant y\}$$ $\endgroup$ – Did Oct 11 '16 at 6:43
  • $\begingroup$ Re the current page, I mention that there is no question in your post since the highlighted formulas are not statements. $\endgroup$ – Did Oct 11 '16 at 6:45
  • $\begingroup$ Finally, if what you are interested in is the result on the other page, not the suboptimal approaches proposed there to prove it, you should say so and I might go as far as posting a two-lines complete, purely real-analysis proof (since there is not an ounce of probability involved in the result, actually). $\endgroup$ – Did Oct 11 '16 at 6:51
  • $\begingroup$ @Did That would be the most helpful. Should I post a new question? I am actually concerned about the equivalence of the two definitions. $\endgroup$ – Qwerty Oct 11 '16 at 9:00
  • $\begingroup$ @Did Can you please help? $\endgroup$ – Qwerty Oct 11 '16 at 12:30

The original question asks to prove the equivalence: $$P[\omega:\lim_{n\to\infty}X_n(\omega) = X(\omega)] = 1\iff\lim_{n\to\infty}P[\omega:\sup_{k>n}|X_k(\omega) - X(\omega)|>\epsilon] = 0$$

Adding the missing "for every positive $\epsilon$" on the RHS and correcting the notations on both sides, one should show the equivalence of the properties:

  • $P(A)=1$ where $A=[\lim\limits_{n\to\infty}X_n=X]$
  • $\lim\limits_{n\to\infty}P(A_n^\epsilon)=1$ for every positive $\epsilon$, where $A_n^\epsilon=[Y_n\leqslant\epsilon]$ and $Y_n=\sup\limits_{k>n}|X_k-X|$

To show this equivalence, first note that the sequence of random variables $(Y_n)$ is nonincreasing, hence $$A=\bigcap_{\epsilon>0}[\lim\limits_{n\to\infty}Y_n\leqslant\epsilon]=\bigcap_{\epsilon>0}[\exists n,Y_n\leqslant\epsilon]=\bigcap_{\epsilon>0}\bigcup_nA_n^\epsilon$$ Now, for every positive $\epsilon$, the sequence of events $(A_n^\epsilon)$ is nondecreasing, hence $$P(A)=\inf_{\epsilon>0}P\left(\bigcup_nA_n^\epsilon\right)=\inf_{\epsilon>0}\left(\lim_{n\to\infty}P(A_n^\epsilon)\right)$$ which proves the desired equivalence. Note that the last identity above nowhere uses the hypothesis that $P(A)=1$, thus, for every $p$ in $[0,1]$, $$P(A)\geqslant p\iff\forall\epsilon>0,\lim\limits_{n\to\infty}P(A_n^\epsilon)\geqslant p$$

Nota bene: The step $$\bigcap_{\epsilon>0}[\lim\limits_{n\to\infty}Y_n\leqslant\epsilon]=\bigcap_{\epsilon>0}[\exists n,Y_n\leqslant\epsilon]$$ is trickier than may appear at first sight, since the identity $$[\lim\limits_{n\to\infty}Y_n\leqslant\epsilon]=[\exists n,Y_n\leqslant\epsilon]$$ is false for some fixed $\epsilon$ (consider $Y_n=\epsilon+\frac1n$). But the following inclusions save the day: $$[\lim\limits_{n\to\infty}Y_n\leqslant\epsilon]\subseteq[\exists n,Y_n\leqslant2\epsilon]\subseteq[\lim\limits_{n\to\infty}Y_n\leqslant2\epsilon]$$ since indeed they imply that $$\bigcap_{\epsilon>0}[\lim\limits_{n\to\infty}Y_n\leqslant\epsilon]\subseteq\bigcap_{\epsilon>0}[\exists n,Y_n\leqslant2\epsilon]\subseteq\bigcap_{\epsilon>0}[\lim\limits_{n\to\infty}Y_n\leqslant2\epsilon]=\bigcap_{\epsilon>0}[\lim\limits_{n\to\infty}Y_n\leqslant\epsilon]$$ Another option is to use strict inequalities, since $$[\lim\limits_{n\to\infty}Y_n<\epsilon]=[\exists n,Y_n<\epsilon]$$

  • $\begingroup$ Can you help me understand the followoing things better? 1>$P(A)=\inf_{\epsilon>0}P\left(\bigcup_nA_n^\epsilon\right)$ 2>$[\lim\limits_{n\to\infty}Y_n\leqslant\epsilon]\subseteq[\exists n,Y_n\leqslant2\epsilon]$. I am not much experienced in handling limits (sups and infs) with inclusions/exclusions. $\endgroup$ – Qwerty Oct 12 '16 at 4:32
  • $\begingroup$ Not sure what you are asking really, partly because of your vague formulations. $\endgroup$ – Did Oct 12 '16 at 5:41
  • $\begingroup$ I was asking, for e.g. how to get that $[\lim_{n\to \infty} Y_n\le \epsilon]$ a subset of $[\exists n, Y_n\le 2\epsilon]$. I am not getting the intuition. $\endgroup$ – Qwerty Oct 12 '16 at 6:00
  • $\begingroup$ Well, if $Y_n>2\epsilon$ for every $n$, do you think that $\lim Y_n\leqslant\epsilon$ can happen? Nothing probabilistic there, just plain real analysis. $\endgroup$ – Did Oct 12 '16 at 7:45
  • $\begingroup$ I need more help. Can you please show the steps you jumped in going from $[\lim\limits_{n\to\infty}X_n=X]$ to $\bigcap\limits_{\epsilon>0}[\lim\limits_{n\to\infty}\sup_{k>n}|X_k-X|\le\epsilon]$ ? $\endgroup$ – Qwerty Oct 12 '16 at 10:07

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.