Nice probability question. (1 Viewer)

RealiseNothing

what is that?It is Cowpea
Joined
Jul 10, 2011
Messages
4,591
Location
Sydney
Gender
Male
HSC
2013
Maybe try thinking of "reversing" your thought process.

What do you think it could possibly converge to?

Might be easier to work from that.
Nah my original working had a slight flaw (forgot to do 1-P(m)).

Re-doing my working gives:



Ain't gonna lie though, I'm really not confident with this answer.
 

anomalousdecay

Premium Member
Joined
Jan 26, 2013
Messages
5,766
Gender
Male
HSC
2013
Nah my original working had a slight flaw (forgot to do 1-P(m)).

Re-doing my working gives:



Ain't gonna lie though, I'm really not confident with this answer.
I would think for a question like this, the way you start is the most crucial factor in getting the end result.

Its because with probability you need to understand the case you are trying to address.

Though I can't even begin to start a question like this so I have no clue.
 

RealiseNothing

what is that?It is Cowpea
Joined
Jul 10, 2011
Messages
4,591
Location
Sydney
Gender
Male
HSC
2013
Anyway here's my working out/thought process. Some one might be able to see what's going on.

To find we will find the amount of ways we can have a string of 'n' heads after exactly 'm' throws, then divide by the total amount of possibilities.

Now the total amount of possibilities is just as there are 'm' throws, two outcomes.

Now for the other bit. The last n+1 throws must be T, H, H, ..., H, H to have a run of 'n' heads after 'm' throws. So this can only happen in one way, and thus we only need to find the amount of ways to arrange the first m-n-1 throws without having a string of 'n' heads.

To do this, we will find the amount of possibilities where we do have a string of 'n' heads, and subtract this from the total ways of arranging the first m-n-1 throws.

So assume we have a run of 'n' heads, and call this element A. Now we have to arrange the element A amongst m-2n-1 throws. Element A can be placed in (m-2n) positions, and the rest of the throws have two possibilities each - T or H. So we end up with the total ways being:



The logic behind this is that it also contains the cases where we have a string of 'n+k' heads as we have a minimum of 'n' heads, and by "randomising" the remaining throws, it is possible for us to end up with a larger string of heads.

So the total ways where we don't get a string of 'n' heads is:



Dividing this by the total ways gives:



After some simplifying:

 

seanieg89

Well-Known Member
Joined
Aug 8, 2006
Messages
2,662
Gender
Male
HSC
2007
Without latex rendering properly atm, I won't look carefully at your work yet realise...but something that might help all of you is a quick and dirty heuristic for why we should expect convergence:

What is the probability of an arbitrary string of length m not containing n consecutive heads somewhere?

Well let's call strings containing n consecutive heads good and let's count those that are bad.

We do this crudely by breaking the m character string into "blocks" of n. There are roughly m/n of these.

Every bad string cannot have any block comprised entirely of H's. So the probability of an arbitrary m-string being bad is roughly bounded by (1-1/2^n)^(m/n), which clearly tends to zero as m->inf.

This also tells us that the average should be finite (which isn't obvious), because the decay in m is exponential and thus beats the linear factor m that occurs in the sum for the average.
 
Last edited:

JJ345

Member
Joined
Apr 27, 2013
Messages
78
Gender
Female
HSC
N/A
How nice is the answer? I'm getting like 2^(n+1)-2
I think that's a bit too nice :/
 

seanieg89

Well-Known Member
Joined
Aug 8, 2006
Messages
2,662
Gender
Male
HSC
2007
Is anyone still working on this or should I post a solution today?
 

seanieg89

Well-Known Member
Joined
Aug 8, 2006
Messages
2,662
Gender
Male
HSC
2007
Haven't thought about this sort of stuff (and in particular this problem) for ages. I suspect it would be quite hard if not impossible to get a closed form for the pdf. Otherwise summing a series involving the pdf would be the natural way to find the mean.

Given that the exploitation of symmetry was the only was I could originally find for computing the mean, this doesn't bode well for closed form computations.

(It is entirely possible I missed something, this is not an area of strength for me.)
 

Users Who Are Viewing This Thread (Users: 0, Guests: 1)

Top