6
$\begingroup$

This is an extension of the $n+1$ vs n problem here: Probability of $5$ fair coin flips having strictly more heads than $4$ fair coin flips

So a common way to think about this is to say that after $n$ tries, both players have the same expected number of heads. So player $A$ has a $50\%$ chance of winning because he gets a head with $50\%$ chance on his last try. But if he gets $x$ more tries than player $B$, does that still stand? He wins with probability $1 - \frac{1}{2^x}$? I did the calculations for a few $n$'s and $x$'s and it doesn't seem to be the case.

What's the answer here? Thanks!

$\endgroup$
2
  • 1
    $\begingroup$ It isn't the same because A could be losing at the n flip mark and then win after the final x flips. This doesn't happen with only one extra flip. $\endgroup$ Jun 15 '13 at 16:01
  • $\begingroup$ That's what I thought. Is there a way to get a simple equation for all $n$ and $x$ though? How can you solve this for general $n$ and $x$? $\endgroup$
    – narcissa
    Jun 15 '13 at 16:11
1
$\begingroup$

A is flipping $n+x$ coins and will get $u$ heads with probability $C(n+x,u)/2^{n+x}.$ Also B is flipping $n$ coins and will get $v$ heads with probability $C(n,v)/2^n.$ The range on $v$ is from $0$ to $n$, and in favorable cases for A to win, the range on $u$ is from $v+1$ to $n+x$. This gives the probability for A to win as $$\frac{1}{2^{2n+x}} \sum_{v=0}^n \sum_{u=v+1}^{n+x} C(n,v)C(n+x,u).$$ Maple was unable to get a closed form for the double sum of binomials here.

In the case where $x=2$, wherein A has two more coins than B, the calculation gives the following numerators for $n=1,2,3,4,5,6$: $$11,\ \ 42,\ \ 163,\ \ 638,\ \ 2510,\ \ 9908.$$ In each case for the probabilities one has to divide by $2^{2n+2}=4^{n+1}.$

$\endgroup$

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.