A Dichotomous Search for a Geometric Random Variable

We are given a two-state system that starts in state 0, ends in state 1, and makes a single transition from state 0 to state 1 during N periods. If the system is in state 0, it moves to state 1 in the next period with a known positive time-independent probability. Once it reaches state 1, it remains there. By observing the state of the system at some intermediate period, we can learn whether this transition occurred earlier or not. An optimal search strategy minimizes the expected number of observations needed to locate the exact transition time. In this paper we show how to compute efficiently and how to approximate the optimal strategy. Applications to the problem arise in the areas of quality control and maintenance of communication and supply lines.