-1
$\begingroup$

Let $\{X_{n}\}$ be a sequence of r.v. such that $$ X_{n} \stackrel{a.s.}{\to} X $$ According to wikipedia https://en.wikipedia.org/wiki/Convergence_of_random_variables $$ P[\omega\in\Omega: X_{n}(\omega) \to X(\omega)] = 1 $$ is equivalent to $$ P[\underset{n\to\infty}{\liminf} \{\omega\in\Omega:|X_{n}(\omega) - X(\omega)|<\varepsilon \}] = 1 $$ for any $\varepsilon$. Could someone prove it?

$\endgroup$
  • 1
    $\begingroup$ You have a notation and a definition of a concept. What's there to prove? $\endgroup$ – Eric Towers Apr 14 '18 at 20:11
  • 2
    $\begingroup$ @EricTowers I think they're probably looking for a proof that this is equivalent to $P(X_n\to X) = 1.$ $\endgroup$ – spaceisdarkgreen Apr 14 '18 at 20:16
  • $\begingroup$ I have edited the post $\endgroup$ – Mr.M Apr 15 '18 at 15:14
2
$\begingroup$

There is indeed something to prove here.

Claim 1:

If $\{B_k\}_{k=1}^{\infty}$ is an infinite sequence of events such that $P[B_k]=1$ for all $k \in \{1, 2, 3,...\}$, then $$P[\cap_{k=1}^{\infty} B_k] = 1$$

Proof: Computing the probability of the complement of $\cap_{k=1}^{\infty} B_k$ gives:
$$ P[(\cap_{k=1}^{\infty} B_k)^c] = P[\cup_{k=1}^{\infty} B_k^c] \leq \sum_{k=1}^{\infty} P[B_k^c] = \sum_{k=1}^{\infty} 0 = 0 $$ where we have used the union bound. $\Box$


The definition of $\liminf$ of a sequence of sets $\{A_n\}_{n=1}^{\infty}$ is: $$ \liminf_{n\rightarrow\infty} A_n = \cup_{M=1}^{\infty} \cap_{n=M}^{\infty} A_n$$ For each $\epsilon>0$ and each $n \in \{1, 2, 3, ...\}$, define $A_n(\epsilon)$ as the following event: $$ A_n(\epsilon) = \{|X-X_n|< \epsilon\}$$ Define events $B(\epsilon)$ by $$ B(\epsilon) =\liminf_{n\rightarrow\infty} A_n(\epsilon) = \cup_{M=1}^{\infty} \cap_{n=M}^{\infty} \{|X-X_n|<\epsilon \} $$ The event $B(\epsilon)$ is equivalent to the event that there exists an integer $M>0$ such that $|X-X_n|<\epsilon$ for all $n \geq M$.

We want to prove that: $$\boxed{P[X_n\rightarrow X]=1 \quad \mbox{ if and only if } \quad P[B(\epsilon)]=1 \quad \forall \epsilon>0} $$


Claim 2:

$\{X_n\rightarrow X\} = \cap_{k=1}^{\infty} B(1/k)$.

Proof: This claim says that $X_n\rightarrow X$ if and only if for each integer $k$ there is an integer $M>0$ such that $|X_n-X|<1/k$ for all $n \geq M$. This is true by definition of a limit. $\Box$

Claim 3

a) $P[X_n\rightarrow X] = P[\cap_{k=1}^{\infty} B(1/k)] \leq P[B(1/i)]$ for all $i \in \{1, 2, 3, ...\}$.

b) $P[X_n\rightarrow X] = 1$ if and only if $P[B(1/i)]=1$ for all $i \in \{1, 2, 3, ...\}$.

c) $P[X_n\rightarrow X] = 1$ if and only if $P[B(\epsilon)]=1$ for all $\epsilon>0$.

Proof: Part (a) follows immediately from Claim 2 and the fact that for every $i \in \{1, 2, 3, ...\}$ we have $\cap_{k=1}^{\infty} B(1/k) \subseteq B(1/i)$.

Part (b) follows from (a) together with Claim 1, specifically, if $P[B(1/i)]=1$ for all $i \in \{1, 2, 3, ...\}$ then $P[\cap_{i=1}^{\infty} B(1/i)]=1$.

To prove part (c), assume $P[B(\epsilon)]=1$ for all $\epsilon>0$. Then $P[B(1/i)]=1$ for all positive integers $i$, and by part (b) we conclude $P[X_n\rightarrow X]=1$. Now supose $P[X_n\rightarrow X] = 1$. Then from part (b) we know $P[B(1/i)]=1$ for all $i \in \{1, 2, 3, ...\}$. Fix $\epsilon>0$. Choose $i \in \{1, 2, 3, ...\}$ such that $1/i\leq \epsilon$. Then $$ B(1/i)\subseteq B(\epsilon) \implies \underbrace{P[B(1/i)]}_{1} \leq P[B(\epsilon)]$$ and so $P[B(\epsilon)]=1$. $\Box$

$\endgroup$
  • $\begingroup$ Note: An additional equivalence that I find more useful is: $P[X_n\rightarrow X]=1$ if and only if for all $\epsilon>0$ we have $$ \lim_{n\rightarrow\infty} P[\cup_{m=n}^{\infty} \{|X_m-X|>\epsilon\}]=0$$ since from this we see that "with probability 1" convergence is more stringent than "in probability" convergence, which means $\lim_{n\rightarrow\infty} P[|X_n-X|>\epsilon]=0$. $\endgroup$ – Michael Apr 15 '18 at 23:32

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.