What is the minimax criterion with respect to a game ? Given below is the payoff matrix corresponding to a game :
| B₁ | B₂ | B₃ | B₄
A₁ | −2 | −1 | −2 | 8
A₂ | 1 | 0 | −1 | −1
A₃ | −3 | 1 | −3 | 1
Determine whether the game is stable or not. Also, obtain the optimal strategies and the value of the game.
15 marks · directive: Determine
The core of a good answer
Row minima are −2, −1, −3 so the maximin value is −1 (row A₂); column maxima are 1, 1, −1, 8 so the minimax value is −1 (column B₃); since maximin = minimax = −1 there is a saddle point at (A₂, B₃), the game is stable (strictly determined), the optimal pure strategies are A₂ for player A and B₃ for player B, and the value of the game is −1 (A pays B one unit).
An outline written by upsc.help, not an official UPSC answer; every fact checked by a separate review. How we work
What the question asks
Define the maximin and minimax principles for a two-person zero-sum game, then test for a saddle point by comparing the largest row minimum with the smallest column maximum. A stable game has a saddle point and is solved in pure strategies.
Answer outline
Introduction
Payoffs are to player A (rows). A, the maximising player, picks the row whose worst outcome is best (maximin); B, the minimising player, picks the column whose worst loss is smallest (minimax).
Step 1: Minimax criterion
- Maximin value v̲ = maxᵢ minⱼ aᵢⱼ: the gain A can guarantee.
- Minimax value v̄ = minⱼ maxᵢ aᵢⱼ: the loss B can hold A down to.
- Always v̲ ≤ v ≤ v̄. If v̲ = v̄ the common value is the value of the game and the cell is a saddle point (smallest in its row, largest in its column).
Step 2: Row minima
- A₁: min(−2, −1, −2, 8) = −2
- A₂: min(1, 0, −1, −1) = −1
- A₃: min(−3, 1, −3, 1) = −3
- Maximin = max(−2, −1, −3) = −1, attained at A₂.
Step 3: Column maxima
- B₁: max(−2, 1, −3) = 1; B₂: max(−1, 0, 1) = 1; B₃: max(−2, −1, −3) = −1; B₄: max(8, −1, 1) = 8
- Minimax = min(1, 1, −1, 8) = −1, attained at B₃.
Step 4: Saddle point and solution
- Maximin = minimax = −1, so a saddle point exists at (A₂, B₃): a₂₃ = −1 is the minimum of row 2 and the maximum of column 3.
- Game is stable (strictly determined); no mixed strategies are needed.
- Optimal strategies: A plays A₂ with probability 1, i.e. (0, 1, 0); B plays B₃ with probability 1, i.e. (0, 0, 1, 0).
- Value v = −1: the game is unfavourable to A, who loses one unit per play to B.
Step 5: Check
- If B plays B₃, A's payoffs are −2, −1, −3: A₂ is best. If A plays A₂, B's payoffs to A are 1, 0, −1, −1: B₃ (or B₄, also −1) gives the least. Neither player gains by deviating; B₄ is an equally good reply to A₂ but B₃ is the saddle point column because −1 is the maximum of column 3, while column 4 has maximum 8.
Conclusion
The game has a saddle point at (A₂, B₃), so it is stable; A should always play A₂, B should always play B₃, and the value of the game is −1.
Examples and data to cite
- Von Neumann's minimax theorem guarantees that every finite two-person zero-sum game has a value in mixed strategies; a saddle point is the special case where pure strategies suffice.
- A game is fair when its value is 0; this one is not fair, favouring B.
Facts to remember
- Maximin v̲ = maxᵢ minⱼ aᵢⱼ; minimax v̄ = minⱼ maxᵢ aᵢⱼ; v̲ ≤ v̄.
- Saddle point exists if v̲ = v̄; the game is then stable or strictly determined.
- Value of the game equals the saddle point payoff.
- Payoffs are to the row player in a zero-sum game; the column player's gain is the negative.