1
$\begingroup$

First of all, I'm aware of an existing question with a similar title, but my problem is fundamentally different.

I'm trying to solve this: in the first try, there is a 5% probability of obtaining a success. In the second try, if the first one was a success, the probability resets to 5%, but if the first one was a failure, now there is 5% more probability (10%). This goes on an on for an arbitrary number of trials, what I'm trying to calculate is the expected number of trials to obtain a single success.

What I got is the probability of obtaining a success in try number $n$, which obviously depends on the probability of the previous try: $$p(n) = p(n-1)\times 0.05 + (1-p(n-1))\times(p(n-1)+0.05),$$ with $p(1)=0.05$. Is it possible to calculate what I want?

In a more general way, what I'm asking is how to calculate the expected number of trials when the probability of success in a single trial depends on the probability of the previous trial, $p(n)=f(p(n-1))$.

$\endgroup$
  • $\begingroup$ Model this with a Markov model. The rest is straightforward. $\endgroup$ – msm Aug 12 '18 at 9:25
  • $\begingroup$ The probability on try $n$ does not depend only on the probability on the previous try. It really depends on the actual outcome! Perhaps you should make a tree of the first few steps. $\endgroup$ – Arnaud Mortier Aug 12 '18 at 9:40
  • $\begingroup$ Also note that $n$ can't be arbitrarily large. $20$ at most. $\endgroup$ – Arnaud Mortier Aug 12 '18 at 9:43
  • $\begingroup$ You are right, so in my case I can just calculate the probabilities of the first 20 steps and from that get the expected value that I want. I guess I was thinking in the general case in which $E(n)$ could be arbitrarily large. Thank you. $\endgroup$ – Julian L Aug 12 '18 at 9:50
1
$\begingroup$

Let $p(n)$ be the probability of having a success within the first $n$ trials, and let $q(n)$ be the probability of having the FIRST success in trial $n$. Then it holds that $p(1)=q(1)=0.05$, $$q(n)=0.05n[1-p(n-1)],$$ and $$p(n)=p(n-1)+q(n).$$ Generate the sequences $p(n)$ and $q(n)$ for as long as they satisfy $0\leq p(n)\leq1$ and $0\leq q(n)\leq1$. It turns out that $p(21)=1$ and this is where you can stop. Then calculate $$\sum_{n=1}^{21}q(n)n.$$

$\endgroup$
0
$\begingroup$

Not a complete answer. Let the probability of success on the first try be $1/m$, where $m$ is a positive integer. If the first try fails, the probability of success on the second try is $2/m$, and so on, as explained by the OP for $m = 20$. Success requires at most $m$ tries, since the probability of success on the $m$th try is $1$. The probability of success on the $r$th try and not before is $p(r)$ where $$p(r) = {{m-1}\over{m}} \times {{m-2}\over{m}} \times \cdots \times {{m-r+1}\over{m}} \times {{r}\over{m}}.$$ The expected number of tries is $$p(1) + 2p(2) + 3p(3) + \cdots + mp(m).$$ The values for $m = 1, \dots\ 6$ work out to $$1,\quad3/2^1,\quad17/3^2,\quad142/4^3,\quad1569/5^4,\quad21576/6^5.$$ The numerators can be found in the OEIS at https://oeis.org/search?q=1%2C3%2C17%2C142%2C1569&language=english&go=Search Unfortunately the table doesn't go as far as $m = 20$ but the references may help.

$\endgroup$

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.