Games & Strategies

RK Jana

1
 Outline

 Introduction
 Types of games
 Maximin and minimax principles
 Mixed strategies in games
 Solution of 2×2 games
 Graphical method of solving games
 Dominance rules
 Linear programming method for solving games
 Introduction
 Game theory is a decision making technique under conflicting
situations.
 Originated by von Neumann in 1928
 von Neumann and Morganstern, 1944
 Zero-sum games

 John Nash
 Nonzero-sum games

 It is applicable only when each player in the game has the choice
of finite number of moves and a finite choices at each move.
 Applicable to human beings, individuals, groups, organizations etc
 All the participants in the game are independent decision makers
and the outcome of the game depends on the outcome of all
strategies used by participating players.
 Types of Games

 Two-person Game
 Zero Sum Game
 Two-person zero-sum game
 Two-person non-zero sum game
 N-person Game

4
 Assumptions

 The players act rationally and intelligently.

 Each player has all relevant information available.
 Each player can use the information in finite number of
moves with finite choices for each move.
 Players make independent decisions of course of
action without consultation.
 Players play the game for optimization.
 The pay-off is fixed and known in advance.

5
 An Example

Two players A & B are playing a game.

Game: A & B will simultaneously & independently write one of the
numbers from 1, 2 & 3.
Rule: If the sum of numbers is even then A wins & B pays the amount to
A. If the sum of numbers is odd then B wins & A pays the amount to B.

B
1 2 3
1
2 -3 4
A 2
-3 4 -5
3
4 -5 6
6
Rules & Payoffs

 The rules of the game state who can do what, and when they
can do it.

 A player's payoff is the amount that the player wins or loses in

a particular situation in a game. The matrix associate with the
game indicates the payoffs and is known as the payoff matrix.

7
 Strategy

 A player's strategy is a course of actions in each possible

situation of the game.
 Pure Strategy: Pure strategy is a decision in making rule in which
one particular course of action is selected.
 Mixed Strategy: Mixed strategy is a decision making rule in which
a player selects his course of action from all the pure strategies
with some definite probability.
 Optimal Strategy: It is the course of action in which the player
optimizes the possible loss or gain.
 A player has a dominant strategy if that player's best strategy
does not depend on what other players do.

8
 Nash Equilibrium

A Nash equilibrium is a situation in which each player

makes his or her best response. So, in Nash
equilibrium, each player's strategy is optimal, given the
strategies of the other players.

9
 Maximin & Minimax Criteria

B
I II
I -2 2
A
II -1 4
III 2 3

Maximizing player: Maximum{Row minimum} = Maximin

Minimizing player: Minimum{Column maximum} = Minimax

Maximin = Minimax = Value of the game

10
 Saddle Point

If in a game the maximin for player A is equal to the

minimax for player B then the game is said to have a
saddle point.

Saddle point is the position in the game matrix where

maximum of the row minimum (maximin) coincides
with the minimum of the column maximum (minimax).

Maximum of the row minimum = Minimum of the

column maximum = Value of the game
11
 Game with a Saddle Point:
Example 1
Solve the following game:

1 2 3 4

I
-5 3 1 20
A
II
5 5 4 6
III
-4 -2 0 -5

12
Continued…

-5 3 1 20

5 5 4 6

-4 -2 0 -5

13
 Example 3

Solve the following game:

B

B1 B2 B3 B4 B5
A1
4 2 1 7 -1
A Value of the
A2
1 -4 -6 -7 6 Game = 2
A3
3 2 3 4 2
A4
-6 1 -1 0 4
A5
0 0 6 0 0
14
 Example 4

For the following game, determine the number of

saddle points and the value of the game.
B

I II III

I 1 2 1 # saddle points = 2
A (I, I), (I, III)
II 0 -4 -1 value of the game = 1

III 1 3 -2
15
 Games without a Saddle Point

 Maximin-minimax criteria do not satisfy.

 Players do not have a single best strategy.
 Each player can improve his payoff by selecting a
different strategy.
 Players have to adopt mixed strategy.
 The game becomes unstable.

16
 Methods for Solving Games
without Saddle Point
 Solution of 2×2 Games
 Method of dominance
 Graphical method
 Algebraic method
 Linear programming method
 Solving 2×2 Games

B
B1 B2

A1 a11 a12
x1 : probability of choosing strategy A1 by A
A a21 a22 x2 : probability of choosing strategy A2 by A
A2
y1 : probability of choosing strategy B1 by B
y2 : probability of choosing strategy B2 by B
v : value of the game

x1 + x2 = 1
y1 + y2 = 1
18
 Continued…

a22  a21 a22  a12

x1  y1 
a11  a22  ( a12  a21 ) a11  a22  (a12  a21 )
a11  a12 a11  a21
x2  y2 
a11  a22  (a12  a21 ) a11  a22  (a12  a21 )

a11a22  a12 a21

v
a11  a22  ( a12  a21 )
19
 Example 5

Solve the following 2×2 game:

B1 B2

A1 -2 5

A2 7 -6

20
Continued…

13 11 23
x1  y1  v
20 20 20
7 9
x2  y2 
20 20

21
 Method of Dominance

 If all the elements of the i-th row be less than or equal to the
corresponding elements of any other row, say r-th, then the
r-th row dominates the i-th row and we discard it.
 If all the elements of the j-th column be greater than or
equal to the corresponding elements of any other column,
say p-th column, then the p-th column dominates the j-th
column and we discard it.
 If the i-th row be dominated by a convex combination of
other rows, then the i-th is deleted from the payoff matrix.
Similarly, if the j-th column is dominated by a convex
combination of other columns then the j-th column is
deleted from the payoff matrix.
22
 Example 6

 Use dominance to solve the following game:

B1 B2 B3 B4

A1 3 2 4 0
A2 3 4 2 4
A3 4 2 4 0
A4 0 4 0 8

23
 Continued…

All the elements of A1 are less than or equal to the

corresponding elements of A3. So, A3 dominates A1 &
hence A1 is deleted.

B1 B2 B3 B4

A2 3 4 2 4
A3 4 2 4 0
A4 0 4 0 8
24
 Continued…

All the elements of B3 are less than or equal to the

corresponding elements of B1. So, B3 dominates B1 &
hence B1 is deleted.

B2 B3 B4

A2 4 2 4
A3 2 4 0
A4 4 0 8

25
 Continued…

The convex combination of B3 & B4 (B3 + B4)/2

dominates B2. So, B2 is deleted.

B3 B4

A2 2 4
A3 4 0
A4 0 8

26
 Continued…

Again, the convex combination of A3 & A4 (A3 + A4)/2

dominates A2. So, A2 is deleted.

B3 B4

A3 4 0

A4 0 8

27
Continued…

2 2 8
x3  y3  v
3 3 3
1 1
x4  y4 
3 3
Final Solution:
A (0, 0, 2/3, 1/3)
B (0, 0, 2/3, 1/3)
v = 8/3 28
 Example

Use dominance to solve the game:

B1 B2 B3 B4

A1 8 15 -4 -2
Solution:
A2 19 15 17 16 A (0, 15/16, 1/16)
B (0, 11/16, 0, 5/16)
A3 10 20 15 5 v = 245/16

29
 Example

Use dominance to solve the following games:

B1 B2 B3 B1 B2 B3

A1 1 7 2 A1 1 7 2
A2 6 2 7 A2 0 2 7
A3 5 1 6 A3 5 1 6
Solution: Solution:
A (2/5, 0, 3/5) A (2/5, 3/5, 0)
B (3/5, 2/5, 0) B (1/2, 1/2, 0)
v = 17/5 v=4 30
 Example

Use dominance to solve the game:

B1 B2 B3 B4

A1 5 -10 9 0
A2 6 7 8 1
A3 8 7 15 1
A4 3 4 -1 4

31
 Example

Use dominance to solve the game:

B1 B2 B3 B4

A1 5 -10 9 0
A2 6 7 8 1 Solution:
A (0, 0, 5/19, 14/19)
A3 8 7 15 1 B (0, 0, 3/19, 16/19)
v = 61/19
A4 3 4 -1 4

32
 Example

Use dominance to solve the game:

B1 B2 B3 B4 B5 B6

A1 0 0 0 0 0 0
A2 4 2 0 2 1 1 Solution:
A (0, 0, 6/7, 1/7, 0, 0)
A3 4 3 1 3 2 2
B (0, 0, 4/7, 3/7, 0, 0)
A4 4 3 7 -5 1 2 v = 13/7
A5 4 3 4 -1 2 2
A6 4 3 3 -2 2 2 33
Graphical Method:
Solution of 2×n Game

B
B1 B2 B3 B4

A1 1 3 0 2

A A2 3 0 1 -1

34
 Continued…

4 4
B3 B4
3 3
2 2 A1 0 2
1 1
0 A2 1 -1
0
-1 -1
-2 -2 Solution:
-3 Maximin
-3 A (1/2, 1/2)
-4 -4 B (0, 0, 3/4, 1/4)
A1 A2 v = 1/2

35
Graphical Method:
Solution of n×2 Game

B1 B2

A1 0 -2

A2 7 -1

A3 -1 4

A4 -2 6

A5 5 -3

36
 Continued…
7 Minimax 7
6 6
B1 B2
5 5
4 4
A2 7 -1
3 3
2 2
1 1
A4 -2 6
0 0

-1 -1 Solution:
-2 -2
A (0, 1/2, 0, 1/2,0)
-3 -3
-4
B (7/16, 9/16)
-4
v = 5/2
B1 B2
37
 Home Work: 3
Solve the following games using graphical method:

B1 B2 B3 B4
1. 2. B1 B2
A1 2 2 3 -1 A1 1 3
A2 4 3 2 6 A2 3 5
A3 -1 6
Solution: 1
A (1/2, 1/2), B (0, 1/2, 1/2, 0) A4 4 1
v = 5/2
Solution: 2
A5 2 2
A (0, 3/5, 0, 2/5, 0, 0)
B (4/5, 1/5), v = 17/5
A6 -5 0 38
 Algebraic Method

y1 y2 ... yn

B1 B2 … Bn Expected gain of A:
a11 x1 + a21 x2 + … + am1 xm
x1 A1 a11 a12 … a1n a12 x1 + a22 x2 + … + am2 xm
……………………………….
x2 A2 a21 a22 … a2n a1n x1 + a2n x2 + … + amn xm

... … … … … …
Expected loss of B:
xm Am am1 am2 … amn a11 y1 + a12 y2 + … + a1n yn
a21 y1 + a22 y2 + … + a2n yn
x1 + x2 + … + xm = 1 ………………………………
am1 y1 + am2 y2 + … + amn yn
y1 + y2 + … + yn = 1 All xi , yj ≥ 0 39
 Continued…

If v is the value of the game then, the original game

problem can be reduced equivalently to the following
form:
a11 x1 + a21 x2 + … + am1 xm = v
a12 x1 + a22 x2 + … + am2 xm = v
……………………………
a1n x1 + a2n x2 + … + amn xm = v
a11 y1 + a12 y2 + … + a1n yn = v
a21 y1 + a22 y2 + … + a2n yn = v
……………………………
am1 y1 + am2 y2 + … + amn yn = v
x1 + x2 + … + xm = 1
y1 + y2 + … + yn = 1 All xi , yj ≥0 40
 Example

Solve the following game by algebraic method:

B1 B2 B3

A1 -1 2 1
A2 1 -2 2
A3 3 4 -3

41
Continued…

- x1 + x2 + 3 x3 = v
2 x1 - 2 x2 + 4 x3 = v
x1 + 2 x2 - 3 x3 = v
Solution:
A (17/46, 10/23, 9/46)
-y1 + 2 y2 + y3 = v B (7/23, 6/23, 10/23)
y1 - 2 y2 + 2 y3 = v v = 15/23
3 y1 + 4 y2 - 3 y3 = v

x1 + x2 + x3 = 1
y1 + y2 + y3 = 1

x1 , x2 , x3 , y1 , y2 , y3 ≥ 0

42
 Linear Programming Method

y1 y2 ... yn

B1 B2 … Bn Assume that the maximizing

player A uses his strategies as:
x1 A1 a11 a12 … a1n A(x1, x2, …, xm) and the
minimizing player B uses his
x2 A2 a21 a22 … a2n strategies as:
B(y1, y2, …, yn)
... … … … … …
xm Am am1 am2 … amn

x1 + x2 + … + xm = 1
y1 + y2 + … + yn = 1 43
 Continued…

If v is the value of the game then, then using the

maximin criterion, we get:
a11 x1 + a21 x2 + … + am1 xm ≥ v
a12 x1 + a22 x2 + … + am2 xm ≥ v
……………………………
a1n x1 + a2n x2 + … + amn xm ≥ v
x1 + x2 + … + xm = 1

All xi ≥ 0

44
Continued…
a11 x1/v+ a21 x2/v + … + am1 xm/v ≥ 1
a12 x1/v + a22 x2/v + … + am2 xm/v ≥ 1
………………………………………….
a1n x1/v + a2n x2/v + … + amn xm/v ≥ 1
x1/v + x2/v + … + xm/v = 1/v
All xi ≥ 0

Denote xi/v = ti
Corresponding LPP :
Min: 1/v = Z = t1 + t2 + … + tm
s. t.
a11 t1+ a21 t2 + … + am1 tm ≥ 1
a12 t1 + a22 t2 + … + am2 tm ≥ 1
………………………………………….
a1n t1 + a2n t2 + … + amn tm ≥ 1
45
All ti ≥ 0.
Continued…
Similarly, considering minimax criterion for B, we get the
corresponding LPP as :

Max: 1/v = W = u1 + u2 + … + un
s. t.
a11 u1+ a21 u2 + … + an1 un ≤ 1
a12 u1 + a22 u2 + … + an2 un ≤ 1
………………………………………….
am1 u1 + am2 u2 + … + amn un ≤ 1
All uj ≥ 0.

46
 Example

Solve the following game using LP method:

B1 B2 B3

A1 9 2 4
A2 0 6 3
A3 5 2 8

47
 Continued…

 Assume that the maximizing player A uses his strategies

A(x1, x2, x3) and the minimizing player B uses his strategies
B(y1, y2, y3).
Considering the pay-off for A, we get (P-I)
Min: 1/v = Z = t1 + t2 + t3
s. t.
9 t1+ 5 t3 ≥ 1
t1 + 6 t2 + 2 t3 ≥ 1
4 t1 + 3 t2 + 8 t3 ≥ 1
t1 , t2 , t3 ≥ 0.
 Continued…

Considering the pay-off for B, we get (P-II)

Max: 1/v = Z = u1 + u2 + u3
s. t.
9 u1+ u2 + 4 u3 ≤ 1
6 u2 + 3 u3 ≤ 1
5 u1 + 2 u2 + 8 u3 ≤ 1
u1 , u2 , u3 ≥ 0.
Now, (P-II) is the dual of (P-I). Solve the dual.
Final solution: A(3/8, 13/24, 1/12), B(7/24, 5/9, 11/72), Value of
the game = 91/24.