A renewal theorem question

117 Views Asked by At

Here's the question: on each round, Ann and Bob each simultaneously toss a fair coin. Let $X_n$ be the number of heads tossed in the $2n$ flips which occur during the first $n$ rounds. For each integer $m>0$, let $r_m$ denote the probability that there exists an $n$ such that $X_n = m$. Use a renewal theorem to find $\lim_{m\to\infty}r_m.$

My thought is, we can think of $X_n$ as summation of $n$ inter-arrival times of a renew process, each is independent and has mean 1; think of $m$ as a time point. Let $N(m)$ denote max number of hits (in this case each hit is 1 round of coin toss) before time $m$. We have $\lim_{m\to\infty} r_m = \lim_{m\to\infty} P(\exists \ n \ s.t. X_n = m) = \lim_{m\to\infty} P(\exists \ n \ s.t. X_n = m | N(m) = m-1)P(N(m) = m-1)+P(\exists \ n \ s.t. X_n = m | N(m) = m-2)P(N(m)=m-2) = \lim_{m\to\infty} \frac{1}{2} P(N(m) = m-1) + \frac{1}{4}P(N(m)=m-2).$

This is where I got stuck. Now by renew theorem, we have: $\lim_{m\to\infty} \frac{X_{N(m)}}{m} = \lim_{m\to\infty}\frac{X_{N(m)}}{N(m)}\frac{N(m)}{m} = 1 \ w.p.1$. Does this mean $\lim_{m\to\infty} P(\frac{X_{N(m)}}{m-1} = 1) = \lim_{m\to\infty} P(\frac{X_{N(m)}}{m-2} =1) = 1?$ This doesn't make intuitive sense to me since $X_{N(m)}$ can either be $m-1$ or $m-2$, and the two probabilities should sum up to 1, not each equal to 1. Any help on how to proceed is greatly appreciated!

1

There are 1 best solutions below

3
On

I'm not sure how to apply the renewal theorem to solve this problem (I tried), but you can certainly solve it using a recurrence relation.

Take $A_m$ as the event that $X_1+\dots+X_n=m$ for some $n\geq 1$. Conditioning on $X_1$ yields $$\begin{eqnarray*}r_1 &=& \mathbb{P}(A_1) \\ &=& \frac{1}{4}\mathbb{P}(A_1|X_1=0)+\frac{1}{2}\mathbb{P}(A_1|X_1=1)+\frac{1}{4}\mathbb{P}(A_1|X_1=2) \\ &=& \frac{1}{4}\mathbb{P}(A_1)+ \frac{1}{2} \cdot 1 + \frac{1}{4}\cdot 0 \\&=& \frac{r_1}{4}+\frac{1}{2}\end{eqnarray*}$$ Solving for $r_1$ yields $r_1=\frac{2}{3}$. Similarly, $$\begin{eqnarray*}r_2 &=& \mathbb{P}(A_2) \\ &=& \frac{1}{4}\mathbb{P}(A_2|X_1=0)+\frac{1}{2}\mathbb{P}(A_2|X_1=1)+\frac{1}{4}\mathbb{P}(A_2|X_1=2) \\&=& \frac{1}{4}\mathbb{P}(A_2)+\frac{1}{2}\mathbb{P}(A_1)+\frac{1}{4}\cdot 1 \\ &=& \frac{r_2}{4}+\frac{r_1}{2}+\frac{1}{4}\end{eqnarray*}$$ Solve for $r_2$ and get $r_2=\frac{7}{9}$. Since $\mathbb{P}(A_m|X_1=j)=r_{m-j}$ for $m\geq 3$ and $j\in \{0,1,2\}$ we get $$\begin{eqnarray*}r_m&=& \mathbb{P}(A_m) \\ &=& \frac{1}{4}\mathbb{P}(A_m|X_1=0)+\frac{1}{2}\mathbb{P}(A_m|X_1=1)+\frac{1}{4}\mathbb{P}(A_m|X_1=2) \\ &=& \frac{r_m}{4}+\frac{r_{m-1}}{2}+\frac{r_{m-2}}{4}\end{eqnarray*}$$ This unveils the recurrence relation $$r_m=\frac{2}{3}r_{m-1}+\frac{1}{3}r_{m-2};r_1=\frac{2}{3},r_2=\frac{7}{9}$$ which has unique solution $$r_m=\frac{1}{4}\left(-\frac{1}{3}\right)^m+\frac{3}{4}$$ Taking $m\mapsto +\infty$ gives a limit of $\frac{3}{4}$. I would be very interested to see an approach using a renewal theorem.