Derandomization in game-theoretic probability (Q468727): Difference between revisions

The authors give a general method for constructing a deterministic strategy of reality from a randomized strategy in game-theoretic probability -- a probability theory based on a betting game between two players, skeptic and reality. In this theory, skeptic has winning strategy, if he can increase his capital to infinity without risking bankruptcy given some almost sure event does not happen. A number of skeptic winning strategies have been constructed so far corresponding to well known proofs in measure-theoretic probability that a certain event happens with probability one. There is no counterpart of reality's strategy in measure-theoretic probability, because in measure-theoretic probability reality is simply generating random variables under a given probability distribution without any specific strategy. Reality's strategies correspond to the notion of de-randomization, since if reality is following a strategy she is not random in the measure-theoretic sense. The construction of reality's strategy proposed in the paper goes as follows: {\parindent=6mm \begin{itemize} \item[1.] Take a randomized strategy. \item [2.] Construct a strategy of skeptic that forces the random event. \item [3.] Construct a strategy of reality using it. \end{itemize}} Each step is straightforward and does not require coming up with a new strategy. Since this approach constructs a deterministic strategy from a randomized one, it is called de-randomization in game-theoretic probability. The generated strategy complies with the success and the failure of the strong law of large numbers. General theory of the notion of compliance as well as some examples are also presented.