Assume $X_1,\ldots,X_n$ are i.i.d exponential random variables with pdf $\lambda e^{\lambda x}$, and $Y_1,\ldots,Y_n$ are i.i.d exponential random variables, independent of $X_i$s, and with pdf $\lambda' e^{\lambda' x}$, where $\lambda'<\lambda$. Calculate $$p_n=\mathbb{P}\left(X_1<Y_1 \bigcap X_1+X_2<Y_1+Y_2 \bigcap \ldots \bigcap \sum_{i=1}^n X_i \leq \sum_{i=1}^n Y_i\right)$$ see if $\lim\limits_{n \to \infty} p_n =c>0$.

I can say: $$p_n = \mathbb{P}\left(X_1<Y_1\right) \mathbb{P}\left(X_1+X_2<Y_1+Y_2|X_1<Y_1\right) \ldots \mathbb{P}\left( \sum_{i=1}^n X_i \leq \sum_{i=1}^n Y_i\bigg|X_1<Y_1 \bigcap X_1+X_2<Y_1+Y_2 \bigcap \ldots \right)$$

But, I am not sure how to continue. Also, if I want to just consider the asymptotic behavior, I can use bounds: $$p_n\geq \mathbb{P}\left(X_1<Y_1 \bigcap X_2<Y_2 \bigcap \ldots \bigcap X_n<Y_n\right) = \mathbb{P}\left(X_1<Y_1\right)^n \to 0$$

which did not help. Any idea?

  • 1
    $\begingroup$ It looks like a generalization of the Bertrand's ballot problem (kurser.math.su.se/pluginfile.php/16103/mod_folder/content/0/…) $\endgroup$ – Jean Marie May 22 '17 at 20:23
  • $\begingroup$ From the problem statement you don't need to compute the probability, rather it seems, you need to bound it. Have you tried anything along these lines? $\endgroup$ – David May 22 '17 at 21:23
  • $\begingroup$ @david yes, but it didnt work since i achieved p>=0. $\endgroup$ – Susan_Math123 May 22 '17 at 21:24
  • $\begingroup$ @david Hi, I just updated the statement to show what I tried for bounds. $\endgroup$ – Susan_Math123 May 23 '17 at 0:42

You can think of $\sum_{i=1}^j X_i$ as the $j$th arrival of a Poisson process with rate $\lambda$; similarly for $\sum_{i=1}^j Y_i$ and a Poisson process with rate $\lambda'$.

Since these processes are independence, superposition implies that you can think of this as a combined Poisson process with rate $\lambda + \lambda'$, and then for each arrival, flip a coin with heads probability $q:=\frac{\lambda}{\lambda+\lambda'}$; if it is heads, label the arrival "heads," and otherwise label the arrival as "tails." The heads arrivals form the first Poisson process with rate $\lambda$, and the tails arrivals form the Poisson process with rate $\lambda'$.

Thus if you flip a coin with heads probability $q$ repeatedly, then $p_n$ is the probability that the $j$th heads is before the $j$th tails flip, for each $j=1,\ldots,n$.

However, I am not sure how to compute this. As noted in the comments, this is similar to Bertrand's ballot problem; see also this post.


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.