Zero-sum dynamic games and a stochastic variation of Ramsey's theorem (Q2485779)

For dynamic games with a bounded number of stages the existence of the value is usually proven by using a fixed-point argument. When the duration of the game lasts for countably many stages one cannot use standard fixed-point theorems. A new tool for proving the existence of the value in infinite-stage two-player zero-sum dynamic games is proposed. It is based on Ramsey's theorem which states that for every coloring of the complete infinite graph by finitely many colours there is a complete infinite monochromatic sub-graph. The following generalization of stopping games proposed by Dynkin is considered. At the outset, a state of the world \(\omega \in \Omega\) is chosen according to the probability measure \textbf{P}. At every stage \(n\) (\(n=1,2,3,\dots\)) the players independently and simultaneously choose actions \(a_n \in A\) and \(b_n \in B\), where \(A\) and \(B\) are measurable spaces of actions of the players. This choice is measurable with regard to filtration \(F_n\) (it describes information available to the players) and previously chosen actions. The game terminates with probability \(p_n(\omega,a_n,b_n) \in [0,1]\) and the terminal payoff is \(g_n(\omega,a_n,b_n) \in [-1,1]\). The game continues to the stage \(n+1\) with probability \(1-p_n(\omega,a_n,b_n)\). The main result of the article is the following: If \(A\), \(B\) are compact metric spaces, and the functions \(p_n(\omega,\cdot,\cdot)\), \(g_n(\omega,\cdot,\cdot)\) are continuous for each \(\omega \in \Omega\) and every \(n=1,2,3,\dots\), the game has a value. The \(\varepsilon\)-optimal strategies are constructed.