Minimax

A decision rule that chooses the action that minimizes the worst-case (maximum) loss; central in zero-sum game theory.

Minimax is a rule for decision-making under uncertainty: choose the action that makes your worst possible outcome as good as possible, i.e., minimize the maximum loss.

Minimax in decision theory

Suppose you choose an action (a) and the state of the world is (s). If your loss is (L(a,s)), the minimax decision rule is:

[ a^* \in \arg\min_a \max_s L(a,s). ]

It is “pessimistic” in the sense that it evaluates each action by its worst-case scenario.

Minimax in zero-sum games

In a two-player zero-sum game (one player’s gain is the other’s loss), minimax is part of the equilibrium logic.

Let (u(a,b)) be player 1’s payoff when player 1 chooses (a) and player 2 chooses (b). Player 1’s maximin value is:

[ \max_a \min_b u(a,b), ]

and player 2’s minimax value (from player 1’s perspective) is:

[ \min_b \max_a u(a,b). ]

The minimax theorem (von Neumann) says that in finite zero-sum games, when players are allowed to use mixed strategies, these two values coincide. That is why mixed-strategy Nash equilibria exist and can be computed by minimax reasoning.

Practical example

In games like matching pennies, there is no best pure strategy: any predictable move can be exploited. The minimax/mixed-strategy solution randomizes in a way that prevents the opponent from earning a systematic advantage.

Knowledge Check

### In decision theory, a minimax rule chooses the action that: - [x] Minimizes the maximum possible loss - [ ] Maximizes the maximum possible gain - [ ] Minimizes the average loss across states - [ ] Maximizes expected utility using probabilities > **Explanation:** Minimax evaluates each action by its worst-case outcome and picks the action with the best worst case. ### In finite zero-sum games, the minimax theorem implies that: - [x] With mixed strategies, the maximin value equals the minimax value - [ ] Pure strategies always exist and are unique - [ ] Arbitrage opportunities persist in equilibrium - [ ] The game has no equilibrium concept > **Explanation:** Allowing randomization (mixed strategies) closes the gap between “maximize the minimum payoff” and “minimize the opponent’s maximum payoff.” ### Why can randomizing be optimal in games like matching pennies? - [x] Any predictable pure strategy can be exploited, so mixed strategies prevent systematic advantage - [ ] Randomizing guarantees the highest possible payoff in every state - [ ] Randomizing eliminates uncertainty - [ ] Randomizing makes both players cooperate > **Explanation:** In some games there is no best pure action; the equilibrium requires making the opponent indifferent by randomizing.