Optimal coin flipping

Abstract

This paper studies the problem of simulating a coin of arbitrary real bias q with a coin of arbitrary real bias p with minimum loss of entropy. We establish a lower bound that is strictly greater than the information-theoretic bound. We show that as a function of q, it is an everywhere-discontinuous self-similar fractal. We provide efficient protocols that achieve the lower bound to within any desired accuracy for