Professional Documents
Culture Documents
06-Handout-2x2 Game Theory University of Kent
06-Handout-2x2 Game Theory University of Kent
Stat155
Game Theory Zero sum games
Lecture 6: Solving two player zero-sum games Recall: payoff matrices, mixed strategies, safety strategies, Von
Neumann’s minimax theorem
Solving two player zero-sum games
Peter Bartlett Saddle points
Dominated pure strategies
Solving 2 × 2 games
Equalizing strategies
September 13, 2016
1 / 25 2 / 25
Definitions Definitions
Player I has m actions, 1, 2, . . . , m. A mixed strategy is a probability distribution over actions.
Player II has n actions, 1, 2, . . . , n. A mixed strategy for Player I is a vector
The payoff matrix A ∈ Rm×n
represents the payoff to Player I:
x1 ( )
a11 a12 · · · a1n x2 Xm
a21 a22 · · · a2n x = . ∈ ∆m := x ∈ Rm : xi ≥ 0, xi = 1 .
..
A= . . .. i=1
.. . . . xm
am1 am2 · · · amn
If Player I chooses i and Player II chooses j, the payoff to Player I is A mixed strategy for Player II is a vector y ∈ ∆n .
aij and the payoff to Player II is −aij . A pure strategy is a mixed strategy where one entry is 1 and the
The sum of the payoff to Player I and the payoff to Player II is 0. others 0. (This is a canonical basis vector ei .)
3 / 25 4 / 25
Recall: Two-player zero-sum games Recall: Two-player zero-sum games
This mixed strategy minimizes the worst case expected loss for
Player II.
5 / 25 6 / 25
7 / 25 8 / 25
Saddle points Saddle points
Example Definition
A pair (i ∗ , j ∗ ) ∈ {1, . . . , m} × {1, . . . , n} is a saddle point for a payoff
−1 1 5 matrix A ∈ Rm×n if
A = 5 3 4
6 2 1 max aij ∗ = ai ∗ j ∗ = min ai ∗ j .
i j
9 / 25 10 / 25
1 2 3 4 5 6 7 row min
1 -1 1 2 4 8 -2 7 -1
2 5 3 -4 7 3 1 3 1
3 6 2 -5 2 8 -1 2 1
4 3 1 3 -5 3 4 7 1 Theorem
5 1 2 3 5 4 -6 7 1 If (i ∗ , j ∗ ) is a saddle point for a payoff matrix A ∈ Rm×n , then
6 4 8 3 7 5 8 4 3 ei ∗ is an optimal strategy for Player I,
7 3 1 3 -5 3 4 7 1
ej ∗ is an optimal strategy for Player II, and
column max 6 8 3 7 8 8 7 the value of the game is ai ∗ j ∗ .
Proof
We have seen that we should always prefer to play last but with a
saddle point, the opposite inequality is also true: So if we find a saddle point, the game is easy to solve.
min max x > Ay ≥ max min x > Ay Another way to simplify a two-player zero-sum game is by removing
y ∈∆n x∈∆m x∈∆m y ∈∆n
dominated rows or columns.
≥ min ei>∗ Ay (∗)
y ∈∆n
15 / 25 16 / 25
Removing dominated pure strategies Removing dominated pure strategies
Outline 2 × 2 games
19 / 25 20 / 25
2 × 2 games 2 × 2 games
a a a a a a V = b + x1 (d − b),
L L L L L d L
b
cb c c d V = a + x1 (c − a).
R b R R d c
R d b R R
d c db cb
Equating gives
B T B T B T B T B T B T a−b
x1 = .
These represent all games (swap L ↔ R, T ↔ B, or both). a−b+d −c
Saddle points? First four games.
What happens if c → d?
Dominated rows? First, second, third.
Then x1 → 1; the game approaches the fourth case.
Dominated columns? First, third, fourth.
21 / 25 22 / 25
25 / 25