Prove or Disprove Almost Sure Convergence

60 Views Asked by At

Let $_1,_2, … , _$ be a sequence of random variables, with $(_) = 7$ and $(_) = 1/sqrt(^)$ for each . Prove or disprove that ${_}$ must converge to 7 with probability 1 if $ > 2$?

I'm thinking of a counter example here:

Let $Xn = 7 + Z(n^{-\frac{\alpha}{4}})$ where $Z$ is a standard normal random variable with mean 0 and variance 1

$E(X_n) = E(7 + Z(n^{-\frac{\alpha}{4}})) = 7 + E(Z)n^{-\frac{\alpha}{4}}= 7$

$Var(X_n) = Var(Z(n^{-\frac{\alpha}{4}})) = n^{-\frac{\alpha}{2}}Var(Z) = n^{-\frac{\alpha}{2}}$

The variance of Xn decreases at a rate of $n^{-\frac{\alpha}{2}}$, which is slower than $\frac{1}{n}$ for $a > 2$. This slower rate of convergence allows for the possibility of $X_n$ taking on values far from 7 with a non-zero probability, even as $n$ grows infinitely large.

From here, although it satisfies the condition for convergence in probability, the set of all possible outcomes where Xn remains close to 7 for all n has a probability of less than 1, violating the strict requirement of almost sure convergence.

Does this disprove correct? Is there any other ways to do this?

1

There are 1 best solutions below

0
On

Well, the counterexample is probably wrong since the statement is true.

In particular, Chebyshev's inequality gives that for any $\epsilon > 0$, $$ P(|X_n - 7| \geq \epsilon) \leq \frac{\operatorname{Var}(X_n)}{\epsilon^2} = \frac{n^{-a/2}}{\epsilon^2}. $$ Then, when $a > 2 \iff a/2 > 1$, $$ \sum_{n=1}^\infty P(|X_n - 7| \geq \epsilon) \leq \epsilon^{-2}\sum_{n=1}^\infty \frac{1}{n^{a/2}} < \infty $$ so Borel-Cantelli gives that, with probability 1, $|X_n - 7| \geq \epsilon$ only finitely often; since this holds for any positive $\epsilon$, $X_n \to 7$ almost surely.