Probability - Noisy message

[Robin04]

Homework Statement

We try to send a message in a very noisy environment. The messages are coded with a four-letter alphabet. By convention, only the messages without lonely letters are considered to be meaningful. A letter is lonely if the next and previous letters are different (the first letter is lonely if the next is different and the last letter is lonely if the previous is different). At least how long does the message have to be if we want to have the chance of having a meaningful message to be less then $10^{-10}$ because of the noise? The noise changes the letters to other letters in the alphabet. Every letter appears independently with an equal chance.

Relevant Equations

-

The phrasing of the problem is a bit messy but here's how I understood it so far:
We make random character strings out of a four-letter alphabet. Every letter appears independently with an equal chance. The chance of having at least two identical letters next to each other in a string of length $N$ is $p(N)$. We are looking for $N$ given that $1-p(N) \leq 10^{-10}$
So the key is to find $p(N)$. Can you help me a bit with that?

 

[StoneTemplePython]

Interesting...

I think your take on how to interpret the problem is helpful and I'd run with it -- the problem statement itself is a bit ambiguous so your interpretation seems to fill gaps. I presume this loneliness tag doesn't apply to the first and last letters -- i.e. it applies to interrior letters only.

What is missing is relevant equations-- there are many ways to tackle something like this, some of which are in reach and many (I suspect) are out of reach.

In essence it appears to be theory of runs problem -- what is the probability that the maximal run size is $\leq 2$ (i.e. event : no runs at all aka the degenerate run length of 1, union event: run of size 2) or equivalently, the probability that there is never a run of length 3. Markov Chain and poisson approximations comes to mind. There's going to be an ordinary generating function approach as well. Many other possibilities... The problem is small enough that I think a direct calculation is doable without too much trouble as well --- this requires inclusion-exclusion and carefully working through events.

edit:
my vote is for markov chains -- it boils down a 2x2 matrix... which is very easy to work with.
my second favorite here would be to set this up as a linear recurrence.