Abstract:We explore a version of the minimax theorem for two-person win-lose games with infinitely many pure strategies. In the countable case, we give a combinatorial condition on the game which implies the minimax property. In the general case, we prove that a game satisfies the minimax property along with all its subgames if and only if none of its subgames is isomorphic to the "larger number game." This generalizes a recent theorem of Hanneke, Livni and Moran. We also propose several applications of our results outside of game theory.
What problem does this paper attempt to address?
### The problems the paper attempts to solve
This paper aims to explore the applicable conditions of the minimax theorem in two - person win - loss games with infinite pure strategies. Specifically, the author attempts to solve the following problems:
1. **Applicability of the minimax property in infinite games**: The traditional minimax theorem is mainly applicable to finite two - person zero - sum games. This paper studies whether the minimax theorem still holds when there are infinitely many pure strategies in the game.
2. **The relationship between combinatorial conditions and the minimax property**: The author provides combinatorial conditions for the countable case, and these conditions can ensure the establishment of the minimax property. For more general cases, the author proves that a necessary and sufficient condition for a game and all its sub - games to satisfy the minimax property is that they do not contain the "larger number game" (LNG) as a sub - game.
3. **Generalize existing results**: The results of this paper generalize the recent theorems of Hanneke, Livni and Moran, who obtained similar conclusions under the additional assumption of a finite VC dimension. This paper relaxes this restriction, making the results more general.
4. **Scope of application**: In addition to game theory itself, the author also explores the applications of these results in other fields, such as sequence convergence, row - and - column densities of infinite binary matrices, fractional matching and covering in infinite hypergraphs, stability in first - order logic, and online learning theory.
### Key concepts and formulas
- **Two - person win - loss game**: Defined as \(G=(S, T, \pi)\), where \(S\) and \(T\) are the pure strategy sets of player 1 and player 2 respectively, and \(\pi: S\times T\rightarrow\{0, 1\}\) is the payoff function.
- **Mixed - extension game**: \(G_{\text{mix}} = (\Delta(S),\Delta(T),\pi_{\text{mix}})\), where \(\Delta(X)\) represents the set of probability distributions on \(X\), and \(\pi_{\text{mix}}(p, q)=E_{(s,t)\sim p\times q}[\pi(s,t)]\)
- **Minimax property**: A game \(G\) satisfies the minimax property if the following equation holds:
\[
\sup_{p\in\Delta(S)}\inf_{q\in\Delta(T)}\pi_{\text{mix}}(p, q)=\inf_{q\in\Delta(T)}\sup_{p\in\Delta(S)}\pi_{\text{mix}}(p, q)
\]
- **Larger number game (LNG)**: Each player chooses a positive integer, and the player with the larger number wins. This game does not satisfy the minimax property because having prior knowledge of the opponent's strategy is a significant advantage.
### Main conclusions
- **Sufficient condition**: For a countable game \(G\), if \(B_{1}^{\text{down}}(G)\) is ascending - union - closed, then \(G\) satisfies the minimax property.
- **Completely minimax game**: A necessary and sufficient condition for a game \(G\) to be completely minimax is that it does not contain any LNG sub - games.
Through these studies, the author not only solves the problem of the applicability of the minimax theorem in games with infinite pure strategies but also reveals the combinatorial structure behind it and shows its broad application potential in multiple fields.