Suppose a coin is tossed repeatedly until either two successive heads appear or two successive tails appear. Then, assume that the first coin toss results in a head. I would like to find the probability of this tossing ending on two successive tails. One way I know of doing this is by creating a Markov chain and then doing first step analysis. I was wondering if there was perhaps an easier way to do this type of problem using some other principles. thank you.

  • $\begingroup$ @David How do you know it is fair? $\endgroup$
    – BCLC
    Feb 9 '16 at 0:39
  • 1
    $\begingroup$ I assume it is fair, as did @lulu, but I suppose it isn't specified and hence should remain arbitrary. $\endgroup$ Feb 9 '16 at 0:41
  • 1
    $\begingroup$ Sorry, forgot to mention that the coin is fair, thanks! $\endgroup$
    – user136503
    Feb 9 '16 at 0:45

Let $p$ be the probability you want.

Look at the possible sequences of three tosses (including the initial $H$). We have $\{HH*,HTT,HTH\}$ Conditional on the first being $H$, the probability of $HH*$ is $\frac 12$, and the probability of the other two are each $\frac 14$. The first is a win for $H$, the second is a win for $T$, and the third restarts the game. Thus $$p=\frac 12\times 0+\frac 14 \times 1 +\frac 14\times p\implies p=\frac 13$$

Note: if you want to allow for a weighted coin, the same calculation works but we need to correct for those initial probabilities. Thus assume that the coin comes up $H$ with probability $\phi$. Then $HH*$ has probability $\phi$, $HTT$ has probability $(1-\phi)^2$, and $HTH$ has probability $(1-\phi)\phi$. So the recursion is now $$p=\phi \times 0+ (1-\phi)^2 \times 1 + (1-\phi)\phi \times p\implies p=\frac {(1-\phi)^2}{1-(1-\phi)\phi}$$ Comparison shows that this matches the result obtained by looking at infinite sums, as in the posted solution of @BCLC (with different notation, unfortunately).

  • $\begingroup$ (i'll read later so assuming this is right) how elegant! $\endgroup$
    – BCLC
    Feb 9 '16 at 1:27

Let $p$ be the probability of heads.

It seems to me that we have the following possible scenarios with the following probabilities:

HH --> $p^2$

H T HH --> $p(1-p)p^2$

H T HT HH --> $p(1-p)[p(1-p)]p^2$

H T HT HT HH --> $p(1-p)[p(1-p)]^2p^2$

H T HT HT ... HT HH --> $p(1-p)[p(1-p)]^np^2$

H TT --> $p(1-p)^2$

H TH TT --> $p[(1-p)(p)](1-p)^2$

H TH TH TT --> $p[(1-p)(p)]^2(1-p)^2$

H TH TH ... TH TT --> $p[(1-p)(p)]^n(1-p)^2$

So we have:

$$P(H...TT) = \sum_{n=0}^{\infty} p[(1-p)(p)]^n(1-p)^2$$

$$ = p(1-p)^2 \sum_{n=0}^{\infty} [(1-p)(p)]^n$$

$$ = p(1-p)^2 \frac{1}{1-(1-p)(p)}$$

If we have $p=1/2$, then $P(H...TT) = 1/6$.

Similarly $P(H...HH or HH) = 1/3$

Of course $P(T) = 1/2$.

Check if it adds up: $P(T) + P(H...HH or HH) + P(H...TT) = 1/6 + 1/3 + 1/2 = 1$.

Just realised (before lulu pointed out) that I didn't condition on $H...$ so just divide all (or most? eg $P(T)$?) above by p or something. Anyway:

$P(T|H...) = 0$

$P(H...HH or HH|H...) = \frac{1/3}{1/2} = 2/3$

$P(H...TT|H...) = \frac{1/6}{1/2} = 1/3$.

Check if it adds up: $P(T|H...) + P(H...HH or HH|H...) + P(H...TT|H...) = 0 + 2/3 + 1/3 = 1$.

  • 2
    $\begingroup$ Note: you should not be factoring in the probability that the first one comes up $H$, that is a given of the problem. Thus the factor of $p$ in your final answer shouldn't be there, and if $p=\frac 12$ the answer is $\frac 13$, not $\frac 16$. $\endgroup$
    – lulu
    Feb 9 '16 at 0:54
  • $\begingroup$ @lulu Oh yeah, noticed that. Edited. Thanks $\endgroup$
    – BCLC
    Feb 10 '16 at 11:08

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.