1
$\begingroup$

Consider a random walk with initial point zero. So we have $S_{n} = X_{1} + \dots + X_n$. And we have two fixed numbers $b > 0 > a$. Now we want to show that $(*) = \operatorname{P}(\{\forall n : a < S_{n} < b\}) \to 0$ , i.e. our random walk cross the bound with full measure.

My attempt : $\displaystyle (*) = \sum_{a < k < b} \binom{n}{(n+k)/2}p^{(n+k)/2} (1-p)^{(n-k)/2}$.

Consider a function $f(x) = x^{a}(1-x)^{n-a}$, it's easy to see that function has maximum at point $x = \frac{a}{n}$. So we have $\displaystyle (*) \le \sum_{a < k < b} \binom{n}{(n+k)/2} (\frac{n+k}{2n})^{(n+k)/2}(\frac{n-k}{2n})^{(n-k)/2} = \sum \binom{n}{(n+k)/2}\frac{1}{2^n} (1-\frac{k^2}{n^2})^{n/2} (\frac{1+\frac{k}{n}}{1-\frac{k}{n}})^{k/2}$. Unfourtunatly my estimates are bad , because of if we are taking limit from both sides we have that rhs goes to $1$. Any help with estimations will be good!

$\endgroup$
1
$\begingroup$

If at any point there are $(b-a)$ consecutive indices where $X_i=+1$, then the random walk will certainly cross the top boundary if it has not crossed any boundaries already. Therefore, letting $$ E_k=\{X_i=+1\text{ for all }i=k(b-a),k(b-a)+1,\dots,(k+1)(b-a)-1\} $$ then the events $E_k$ are independent, have probability less than $1$, and none of them must occur in order for $\{a<S_n<b\text{ for all }n\}$ to occur.

Can you conclude?

$\endgroup$
  • $\begingroup$ Then $\mathbb{P} (\{a < S_n < b\}) = 1 - \sum \mathbb{P}(E_{k})$ ? $\endgroup$ – openspace Feb 15 at 18:49
  • $\begingroup$ $\sum P(E_k)=\infty$, so not quite. Try $P(a<S_n<b)\le \prod_k (1-P(E_k))$. @openspace $\endgroup$ – Mike Earnest Feb 15 at 19:42
  • $\begingroup$ oh, yes. Then it will be something goes to zero on the right side $\endgroup$ – openspace Feb 15 at 20:40
  • $\begingroup$ But why probability of this events should be greater than zero? $\endgroup$ – openspace Feb 16 at 10:44
  • 1
    $\begingroup$ It's probability is $p^{(b-a)}$? $\endgroup$ – openspace Feb 16 at 17:18

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.