It turns out that my previous proof of the optimal strategy for the M-door Monty Haul problem has a few errors in it. The most important was an assumption that switching doors on any round other than the final two-door round hurt one's chances of winnings. This must be false. To see this, let's suppose that at the start of the game you've picked door #1. Suppose also that, after making a series of switches or non-switches, you switch back to door #1. Can it really be said that your odds of winning the car are now lower than if you had never switched from door #1 at all?
In physical terms, this property is called the Markovian Postulate:
If K is any observational state and T is a trial whose preparation stage invariably ends with the system in the state K, then T is statistically regular. (O. Penrose, "Foundations of Statistical Mechanics", p.34)In particular, suppose you go into the two-door round having picked a certain door and with a certain other door remaining (call them #1 and #2). You can think of this as one long trial T which invariably end with you having picked door #1 with door #2 remaining at the end of a preparation stage of a certain length. The Markovian Postulate says that your probability of winning the car at this point will only depend upon whether or not you switch in this final two-door round, regardless of the past history of switches or non-switches in the previous rounds.
In the M-Door Monty Haul problem, the probability of winning the car by making no switches at all except for one final switch in the two-door round is (M-1)/M. So this must be the probability of winning the car by switching doors in the final round, regardless or whether or not you switch or not on any previous round. So the optimal strategy is simple: switch doors in the two-door round.
No comments:
Post a Comment