Question
Given a biased coin with probability of heads $p \neq \tfrac12$, how can we simulate a fair coin?
Solution
Naive method (Von Neumann trick).
Consider pairs of tosses. The outcomes $TH$ and $HT$ have equal probability: $$ \mathbb{P}[TH] = p(1-p), \qquad \mathbb{P}[HT] = p(1-p). $$
But $$ \mathbb{P}[HH] = p^2, \qquad \mathbb{P}[TT] = (1-p)^2, $$ which are not equal unless $p=\tfrac12$.
So we define:
- $TH \mapsto H$
- $HT \mapsto T$
- discard $HH$ and $TT$
This produces a fair coin.
Solution
Expected number of tosses.
One fair outcome occurs when the pair is either $TH$ or $HT$, whose total probability is $$ \mathbb{P}[TH] + \mathbb{P}[HT] = 2p(1-p). $$
Each attempt uses $2$ tosses, so the expected number of tosses is $$ 2 \times \frac{1}{2p(1-p)} = \frac{1}{p(1-p)}. $$