UNIT 12 TWO-PERSON ZERO-SUM GAMES

WITHOUT SADDLE POINT

Structure
12.1 Introduction
Objectives
12.2. Solution of 2  2 Two-Person Zero-Sum Games without Saddle Point
12.3 Dominance Rules for Reducing the Size of the Game
12.4 Graphical Method for Reducing the 2  n and m  2 Games
to 2  2 Games
Graphical Method for Reducing 2  n Games to 2  2 Games
Graphical Method for Reducing m  2 Games to 2  2 Games
12.5 Solution of 3  3 Two-Person Zero-Sum Game
12.6 Flow Chart
12.7 Summary
12.8 Solutions/Answers

12.1 INTRODUCTION
In Unit 11, you have learnt the definitions of key terms used in game theory.
We have also discussed the solution of two-person zero-sum games with saddle
point using the maximin-minimax principle. In this unit, we discuss some
methods of solving two-person zero-sum games without saddle point. In
Sec. 12.2, we describe the algebraic method for solving 2  2 two-person zero-
sum games without a saddle point. However, there are rectangular games of
sizes other than 2  2 and to obtain the solution for such games, we need to
know the dominance rules. These rules help in reducing the size of the games
and are described in Sec. 12.3. However, using the dominance ruels, we cannot
reduce all rectangular games into games of sizes 2  2 . So, we need methods to
solve games of sizes other than 2  2 . The graphical method discussed in Sec.
12.4 is one such method, which reduces the games of size 2  n and m  2 to
size 2  2 . Once the game is reduced to size 2  2, we can solve it by using the
algebraic method discussed in Sec. 12.2.
In Sec. 12.5, we solve a popular 3  3 two-person zero-sum game known as
Stone-Paper-Scissors game by extending the technique of solving 2  2 two-
person zero-sum games discussed in Sec. 12.2. A flow chart showing the
approach followed in this unit to solve the two-person zero-sum games is given
in Sec. 12.6.

Objectives
After studying this unit, you should be able to:
 solve 2  2 two-person zero-sum games by applying the algebraic method;
 apply dominance rules to reduce the size of a game;
 solve 2  n and m  2 two-person zero-sum games by applying the
graphical method; and
 solve 3  3 two-person zero-sum games.
12.2 SOLUTION OF 2  2 TWO-PERSON ZERO-
SUM GAMES WITHOUT SADDLE POINT
In Unit 11, you have learnt how to solve two-person zero-sum games with
saddle point using the maximin-minimax principle. You have also learnt that
the existence of a saddle point depends on the equality of maximin value and
minimax value, where
Maximin value = maximum among row minima = max min{a ij} … (1)
i j

Minimax value = minimum among column maxima = min max{a ij} … (2)
j i
th
and a ij is the (i, j) entry of the payoff matrix of the row player.
The operations for obtaining the maximin and minimax values from equations
(1) and (2) are different. So, the values obtained from equations (1) and (2)
may or may not be equal. If they are equal, the game has a saddle point and can
be solved using the maximin-minimax principle, which gives a pure strategy as
a solution. But if they are not equal, the game does not have a saddle point. So
we need methods to solve the games without saddle point. One such method
for solving 2  2 two-person zero-sum games without saddle point is discussed
in this section. But before describing the method, we state a theorem required
for solving such games.
Fundamental Theorem of Rectangular Games or the Minimax Theorem
It states that for every finite two-person zero-sum game:
(1) There exists a number v, called the value of the game,
(2) There exists at least one mixed strategy for player I such that the gain of
player I  v, irrespective of the strategy adopted by player II, and
(3) There exists at least one mixed strategy for player II such that the loss of
player II  v, irrespective of the strategy adopted by player I.
Note 1: Finite two-person zero-sum game means that the courses of action
available to both players should be finite in number.
This theorem simply states that there always exists a solution for every finite
two-person zero-sum game in terms of mixed strategies.
The solution of 2  2 two-person zero-sum games in terms of mixed strategies
is provided by the algebraic method. We first give a statement of the method
and then its proof.
Statement: Every 2  2 two-person zero-sum game without a saddle point
having the following payoff matrix for player A,
Player B
B1 B2
A1 a b
Player A
A2 c d
has a solution in terms of mixed strategies (p1 , p 2 ), (q1 , q 2 ) for player A and
player B, respectively. The value of the game is v, and p1 , p 2 , q1 , q 2 and v are
given by
dc
p1  ...(3a)
(a  b)  (d  c)
ab
p2  ...(3b)
(a  b)  (d  c)
 db
q1  ...(3c)
(a  b)  (d  c)
a c
q2  ...(3d)
(a  b)  (d  c)
ad  bc
v ...(3e)
(a  b)  (d  c)
such that
p1 , p 2 , q1 , q 2  0, p1  p 2  1 and q1  q 2  1 ...(3f )
Proof: We are given that the payoff matrix for player A has no saddle point. So
there is no solution in terms of pure optimal strategies. But from the
fundamental theorem of game theory, there exists a solution in terms of mixed
strategies. Let (p1 , p 2 ) and (q1 , q 2 ) be the optimal mixed strategies for player A
and player B, respectively, and v be the value of the game. We now rewrite the
payoff matrix indicating the above probabilities as follows:
Player B
q1 q2
B1 B2
p1 A1 a b
Player A
p2 A2 c d

The expected gain of player A is

ap1  cp 2 when player Bemploys strategy B1
bp1  dp 2 when player Bemploys strategy B2
Similarly, the expected loss of player B is
aq1  bq 2 when player A employs strategy A1
cq1  dq 2 when player A employs strategy A 2
Since v is the value of the game and (p1 , p 2 ) a mixed optimal strategy for
player A, we have
ap1  cp2  v  bp1  dp2
I II III
 ap1  cp 2  bp1  dp 2
 ap1  c(1  p1 )  bp1  d(1  p1 )  p1  p2  1
dc
 (a  c  b  d)p1  d  c  p1 
(a  b)  (d  c)
dc (a  b)  (d  c)  d  c a b
 p 2  1  p1  1   
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a(d  c) c(a  b)
The expected gain of player A  v  ap1  cp 2  
(a  b)  (d  c) (a  b)  (d  c)
ad  ac  ca  bc ad  bc
 
(a  b)  (d  c) (a  b)  (d  c)
Similarly, since v is the value of the game and (q1 , q 2 ) is a mixed optimal
strategy for player B, we have
aq1  bq 2  v  cq1  dq 2
I II III
 aq1  bq 2  cq1  dq 2  aq1  b(1  q1 )  cq1  d(1  q1 )
db
 (a  b  c  d)q1  d  b  q1 
(a  b)  (d  c)
db (a  b)  (d  c)  d  b a c
 q 2  1  q1  1   
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a(d  b) b(a  c)
The expected gain of player B  v  aq1  bq 2  
(a  b)  (d  c) (a  b)  (d  c)
ad  ab  ab  bc ad  bc
 
(a  b)  (d  c) (a  b)  (d  c)
Thus, we have proved that
dc a b
p1  , p2 
(a  b)  (d  c) (a  b)  (d  c)
db a c ad  bc
q1  , q2  and v 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
Now, it only remains to prove that p1 , p 2 , q1 , q 2  0.
Remember that we are given that the payoff matrix of player A has no saddle
point.
Therefore, either a  b or a  b.
First, suppose that a  b. This implies that b is the minimum element of the
first row
 b  d.
Otherwise, b will also become the maximum element of the second column and
so a saddle point.
Now, b  d  d is the maximum element of the second column.
 d  c.
Otherwise, d will also become the minimum element of the second row and so
a saddle point.
Again, d > c  c is the minimum element of the second row.
 c  a.
Otherwise, c will also become the maximum element of the first column and so
a saddle point.
Finally, c  a  a is the maximum element of the first column.
 a  b.
Otherwise, a will also become the minimum element of the first row and so a
saddle point.
Thus, we obtain the result that if a  b then b < d, d > c, c < a and a > b.
Due to symmetry, if a  b then b  d,d  c,c  a, and a  b.
Thus, in both cases, either both (a – b) and (d – c) are simultaneously positive
or both are simultaneously negative. Therefore, p1  0, p 2  0.
Similarly, we can show that q1  0, q 2  0.
Hence proved.
Note 2: For numerical problems you have to substitute the values of a, b, c, d
directly in equations (3a to e) for p1 , p 2 , q1 , q 2 and v to obtain the
solution.
Let us solve the following example using the algebraic method.
Example 1: Solve the two-person zero-sum game having the following payoff
matrix for player A:
Player B
B1 B2
A1 3 2
Player A
A2 1 5

Solution: We first check whether a saddle point exists or not.

Player B Row Maximin
B1 B2 Minima Value
Player A1 3 2 2
max{2, 1}  2
A A2 1 5 1
Column Maxima 3 5
Minimax Value min{3,5}  3

Since the maximin value (=2)  minimax value (=3), there is no saddle point.
Hence, we have to solve the game in terms of mixed strategies. Let (p1 , p 2 ) and
(q1 , q 2 ) be the optimal mixed strategies for player A and player B, respectively
and v be the value of the game. From equations (3a to e), we have
dc ab db
p1  , p2  , q1 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a c ad  bc
q2  and v 
(a  b)  (d  c) (a  b)  (d  c)
In this game, a = 3, b = 2, c = 1 and d = 5.
5 1 4 3 2 1 52 3
 p1   , p2   , q1  
(3  2)  (5  1) 5 (3  2)  (5  1) 5 (3  2)  (5  1) 5
3 1 2 3  5  2 1 13
q2   and v  
(3  2)  (5  1) 5 (3  2)  (5  1) 5
Hence, the solution of the game is as follows:
The optimal strategy for player A is (4/5, 1/5), the optimal strategy for player B
is (3/5, 2/5) and the value of the game is 13/5.
Now, you can try the following exercises for solving a couple of games.
E1) Solve the two-person zero-sum game having the following payoff matrix
for player A:
Players B
B1 B2
A1 –2 –1
Player A
A2 4 –3

E2) Solve the coin matching game explained in Sec. 11.2 of Unit 11.
12.3 DOMINANCE RULES FOR REDUCING THE
SIZE OF THE GAME
In Sec. 12.2, we have derived and applied the algebraic method to solve
2  2 two-person zero-sum games without saddle point. So this method cannot
be applied directly to m  n two-person zero-sum games without a saddle point
if either m or n or both are greater than 2. But some of these games can be
reduced to 2  2 games by applying the dominance rules, if applicable. In this
section, we state these rules and explain how to apply them. But before you
learn the dominance rules, you should understand inferior and superior
strategies, which we explain below.
Inferior and Superior Strategies: Let a  (a1 , a 2 ,..., a n ) and b  (b1 , b 2 ,..., b n )
be two n-tuples. We say that n-tuple b dominates n-tuple a if
a i  bi for all i, 1  i  n
If a i  bi for all i, 1  i  n then we say that n-tuple b strictly dominates n-tuple
a. Suppose that aij ,  i  1, 2,..., m; j  1, 2,..., n  represents the payoff value of
payoff matrix of player A corresponding to the i th strategy of row player A and
th
jth strategy of column player B. Then we say that the r strategy of row player
A is inferior to his/her s th strategy (or s th strategy is superior to r th strategy for
row player A) if
a rj  a sj for all j, 1  j  n
that is, all the elements of the r th row are less than or equal to the
corresponding elements of the s th row.
Remember that in the payoff matrix of row player A, payoff values represent
gains of player A while for column player B, they represent losses. So, from
the point of view of player B, we say that the r th strategy of column player B is
inferior to his/her s th strategy (or s th strategy is superior to r th strategy for
column player B) if
 payoff values represent 
a ir  a is for all i, or1  i  m 
losses for player B 

that is, if all the elements of the r th column are greater than or equal to the
corresponding elements of the s th column.
So, in general, we define the dominant strategy as follows:
Dominant Strategy: A strategy is said to be the dominant strategy for a player
if it is better than another strategy available to that player irrespective of the
strategy adopted by the other player.
We now state the dominance rules to be used for reducing the size of the payoff
matrix.
Rule 1
If all elements in a row (say, the i th row) of a payoff matrix for row player are
less than or equal to the corresponding elements of the other row (say
jth row), then the row player A will never employ his/her i th strategy because it
is dominated by his/her jth strategy. So we can delete the i th row from the
payoff matrix. Remember that payoff values are gains for row player.
Rule 2
If all the elements in a column (say, the r th column) of a payoff matrix for row
player are greater than or equal to the corresponding elements of the other
column (say, the s th column), then the column player B will never employ
his/her r th strategy because it is dominated by his/her s th strategy. So we can
delete the r th column from the payoff matrix. Remember that payoff values are
losses for column player.
Note 3: It is important to note that in the case of row player A, the inferior
row is deleted but in the case of column player B, the superior column
is deleted. This is due to the fact that payoff values are gains for row
player A but losses for column player B.
Rule 3
A strategy can also be deleted if it is dominated by average (or any convex
combination) of two or more other pure strategies.
Note 4: A combination 1x1   2 x 2  ...   n x n of x1 , x 2 ,..., x n is said to be a
convex combination if i  0 for all i, 1  i  n and
1   2  ...   n  1.
If you want to know more about convex combination, refer to Unit 1 of the
Course MSTE-002.
Let us apply these rules to a couple of examples.
Example 2: Solve the two-person zero-sum game having the following payoff
matrix for player A:
Player B
B1 B2 B3
A1 –2 –4 6
Player A A2 5 4 1
A3 3 2 8

Solution: We first check whether a saddle point exists or not.

Player B Row Maximin Value
B1 B2 B3 Minima
A1 –2 –4 6 –4
Player A A2 5 4 1 1 max{ 4, 1, 2}  2
A3 3 2 8 2
Column Maxima 5 4 8
Minimax Value min{5, 4,8}  4

Since maximin value  minimax value, there is no saddle point. Hence, we

shall obtain the solution in terms of mixed strategies. Let us first reduce the
payoff matrix by applying the dominance rules.
Step 1: We note that each element of the first row is less than the
corresponding element of the third row. This implies that the first row
is dominated by the third row. So we can eliminate the first row from
the given payoff matrix. Thus, we write the reduced payoff matrix as:
Player B
B1 B2 B3
A2 5 4 1
Player A
A3 3 2 8
Step 2: Each element of the first column is greater than the corresponding
element of the second column. This implies that the first column is
dominated by the second column. So we can eliminate the first
column and write the reduced payoff matrix as:
Player B
B2 B3
A2 4 1
Player A
A3 2 8

In the reduced payoff matrix given above neither any row dominates
any other row nor any column dominates any other column. So this
matrix cannot be reduced any further using dominance rules. Since the
reduced payoff matrix is of order 2  2, we can apply the algebraic
method for 2  2 two-person zero-sum game.
Let (p1 , p 2 , p3 ) and (q1 , q 2 , q 3 ) be the mixed strategies for players A
and B, respectively, for the original payoff matrix. Therefore, from the
reduced payoff matrix, (p 2 , p3 ) and (q 2 , q 3 ) are the mixed strategies for
the players A and B, respectively. If v is the value of the game, then
from equations (3 a to e), we have
dc a b db
p2  , p3  , q2 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a c ad  bc
q3  and v 
(a  b)  (d  c) (a  b)  (d  c)
In this case a = 4, b = 1, c = 2, d = 8.
82 6 2 4 1 3 1
 p2    , p3   
(4  1)  (8  2) 9 3 (4  1)  (8  2) 9 3
8 1 7 42 2
q2   , q3  
(4  1)  (8  2) 9 (4  1)  (8  2) 9
4  8  1 2 30 10
v  
(4  1)  (8  2) 9 3
Hence, the solution of the game is:
The optimal strategy for player A is (0, 2/3, 1/3), the optimal strategy
for player B is (0, 7/9, 2/9) and the value of the game is 10/3.
Example 3: Solve the two-person zero-sum game having the following payoff
matrix for player A:
Player B
B1 B2 B3 B4 B5
A1 6 1 3 2 3
A2 5 7 –5 1 4
Player A
A3 8 4 –1 2 2
A4 3 3 –2 2 5

Solution: We first check whether a saddle point exists or not.

Player B Row Maximin Value
B1 B2 B3 B4 B5 Minima
A1 6 1 3 2 3 1
Player A2 5 7 –5 1 4 –5
max{1, 5, 1, 2}  1
A A3 8 4 –1 2 2 –1
A4 3 3 –2 2 5 –2
Column Maxima 8 7 3 2 5
Minimax Value min{8, 7, 3, 2, 5}  2
Since maximin value  minimax value, the saddle point does not exist. Hence,
we have to obtain the solution in terms of mixed strategies. Let us first reduce
the payoff matrix by applying the dominance rules.
Step 1: We note that each element of the first and fifth columns is greater than
or equal to the corresponding elements of the fourth column. This
implies that the fourth column dominates both first and fifth columns.
So we can eliminate the first and fifth columns from the given payoff
matrix and write the reduced payoff matrix as:
Player B
B2 B3 B4
A1 1 3 2
A2 7 –5 1
Player A
A3 4 –1 2
A4 3 –2 2

Step 2: Each element of the fourth row is less than or equal to the
corresponding element of the third row. This implies that the fourth
row is dominated by the third row. So we can eliminate the fourth row
from the reduced payoff matrix. The reduced payoff matrix after this
step is:
Player B
B2 B3 B4
A1 1 3 2
Player A A2 7 –5 1
A3 4 –1 2

Step 3: In this reduced payoff matrix, neither any row is directly dominated
by any other row nor any column is directly dominated by any other
column. But we observe that the average of the first and second
columns of the payoff matrix dominates the third column. So, we can
eliminate the third column from this reduced payoff matrix and write
the reduced payoff matrix as:
Player B
B2 B3
A1 1 3
Player A A2 7 –5
A3 4 –1

Step 4: Again, in this reduced payoff matrix neither any row is directly
dominated by any other row nor any column is directly dominated by
any other column. But we observe that the average of the first and
second rows of this reduced payoff matrix dominates the third row. So
we can eliminate the third row from this reduced payoff matrix. The
reduced payoff matrix after this step is:
Player B
B2 B3
A1 1 3
Player A
A2 7 –5

In this reduced payoff matrix, neither any row dominates any other
row nor any column dominates any other column. So it cannot be
reduced further using dominance rules. Since the reduced payoff
 matrix is of order 2  2, we can apply the algebraic method for
2  2 two-person zero-sum game. Let  p1 , p2 , p3 , p4  and
 q1, q2 , q3 , q4 , q5  be the mixed strategies for players A and B,
respectively, for the original payoff matrix. Thus, (p1 , p 2 ) and
(q 2 , q 3 ) are the mixed strategies from the reduced payoff matrix for
players A and B, respectively. If v is the value of the game, then from
equations (3 a to e) we have
dc a b db
p1  , p2  , q2 
(a  b)  (d  c)  a  b   (d  c) (a  b)  (d  c)
a c ad  bc
q3  and v 
 a  b   (d  c) (a  b)  (d  c)

In this case a = 1, b = 3, c = 7, d = –5.

5  7 12 6 1 3 2 1
 p1    , p2   
1  3   5  7  14 7 1  3   5  7  14 7
5  3 8 4 1 7 6 3
q2    , q3   
(1  3)  (5  7) 14 7 1  3  (5  7) 14 7
1  5  3  7 26 13
v  
1  3  (5  7) 14 7
Hence, the solution of the game is:
The optimal strategy for player A is (6/7, 1/7, 0, 0), the optimal
strategy for player B is (0, 4/7, 3/7, 0, 0) and the value of the game is
13/7.
You may like to try the following exercise for practice.
E3) Solve the two-person zero-sum game having the following payoff matrix
for player A:
Player B
B1 B2 B3 B4
A1 4 3 2 1
Player A A2 6 4 5 0
A3 1 2 0 3

Note 5: You have seen that dominance rules (if applicable) reduce the size of
the game. You can apply these rules to any game whether it has a
saddle point or not. So we can also apply these rules to the games
discussed in Unit 11. If we do so,the size of the game reduces to
11. And if the size of the game reduces to 11, it implies that the
game has been solved. For example, let us apply these rules to
Example 1 of Unit 11.
Step 1: We note that each element of the first row is greater than or
equal to the corresponding element of the second row. This
implies that the first row dominates the second row. So we
can eliminate the second row from the given payoff matrix
and write the reduced payoff matrix as:
 Player B
B1 B2 B3
Player A1 5 7 4
A A3 6 1 3

Step 2: Each element of the first column is greater than or equal to

the corresponding element of the third column. This implies
that the third column dominates the first column. So we can
eliminate the first column from the reduced payoff matrix.
The reduced payoff matrix after this step is:
Player B
B2 B3
A1 7 4
Player A
A3 1 3

Step 3: Each element of the first row is greater than or equal to the
corresponding element of the second row. This implies that
the first row dominates the second row. So we can eliminate
the second row from the reduced payoff matrix. The reduced
payoff matrix after this step is:
Player B
B2 B3
Player A A1 7 4

Step 4: The single element 7 of the first column is greater than the
corresponding single element of the second column. This
implies that the second column dominates the first column.
So we can eliminate the first column from the reduced payoff
matrix. The reduced payoff matrix after this step is:
Player B
B3
Player A A1 4

Thus payoff matrix reduces to order 11. Hence the solution

of the game is:
The optimal strategy for player A is (1, 0, 0), the optimal
strategy for player B is (0, 0, 1) and the value of the game is
4. You can match the solution obtained here with the one
obtained in Example 1 of Unit 11.

12.4 GRAPHICAL METHOD FOR REDUCING THE

2  n AND m  2 GAMES TO 2  2 GAMES
In Sec. 12.3, you have learnt the dominance rules for reducing the size of the
game if the payoff matrix has some inferior strategies. But many-a-times we
have to deal either directly with 2  n and m  2 games or they reduce to one of
these forms after applying dominance rules. Therefore, it becomes important to
learn how to solve 2  n and m  2 games. In this section, we discuss one such
method known as the graphical method, which reduces the 2  n and
m  2 games into 2  2 games. Once a game is reduced to size 2  2 , we can
solve it using the algebraic method discussed in Sec. 12.2. So, let us first
discuss the graphical method for 2  n games.
12.4.1 Graphical Method for Reducing 2  n Games to
2  2 Games
In 2  n games, the row player A has two moves and the column player B may
have any finite number of moves (say, n). Let the payoff matrix for player A
for a 2  n game be:
Player B
B1 B2 … Bj … Bn
A1 a 11 a 12 … a1 j … a1n
Player A
A2 a 21 a 22 … a2j … a 2n

If the game has a saddle point, it can be solved using the maximin-minimax
principle discussed in Unit 11, which gives pure strategies as a solution. If the
game does not have a saddle point, its solution will be in terms of mixed
strategies. Let  p1, p2  and  q1 ,q 2 ,...,q j ,...,q n  be the optimal mixed strategies
for player A and player B, respectively. Let us rewrite the payoff matrix with
these probabilities.
Player B
q1 q2 … qj … qn
B1 B2 … Bj … Bn
p1 A1 a 11 a 12 … a1 j … a1n
Player A
p2 A2 a 21 a 22 … a2j … a 2n

Now, if e1 , e 2 ,..., e n be the expected payoffs for player A corresponding to

different strategies of player B, then these are given by:
e1  a11p1  a 21p 2 if player B employs strategy B1
e2  a12 p1  a 22 p 2 if player B employs strategy B2
. .
. .
. .
en  a1n p1  a 2n p 2 if player B employs strategy Bn

But p1  p 2  1  p 2  1  p1. So writing the above expected payoffs for player

A in terms of p1 , we get
e1   a11  a 21  p1  a 21 if player B employs strategyB1
e2   a12  a 22  p1  a 22 if player B employs strategyB2
. .
. .
. .
en   a1n  a 2n  p1  a 2n if player B employs strategyBn

The above expected payoffs for player A are linear equations in p1. We know
from Sec. 2.5 of Unit 2 of MST-001 that the graph of a linear function given by
y  ax  b
is a straight line (here we have the expected payoff in place of y and p1 in place
of x). Hence, the n expected payoffs of row player A corresponding to different
strategies employed by column player B represent n straight lines. Let us now
explain how to plot these lines on a graph paper. First of all, we take values of
p1 along the horizontal axis. However, the values of p1 vary from 0 to 1
( 0  p1  1) because p1 represents probability. Therefore, we restrict the
horizontal axis at 1 unit, starting from p1  0 and ending at p1  1 (see Fig.
12.1a).

Fig. 12.1: Horizontal and vertical axes for the graph of player A.
We take the expected payoff along the vertical axis (see Fig. 12.1b).
Now, consider the first line:
e1   a11  a 21  p1  a 21
Actually, two points are sufficient to draw the graph of a straight line (read the
margin remark). So we obtain two points corresponding to p1  0 and p1  1, Recall from Sec. 2.5 of
Unit 2 of MST-001,
respectively. how to draw the graph
of a straight line. We
At p1  0, e1  a 21 simply obtain any three
At p1  1, e1   a11  a 21 1  a 21  a11
points from the
equation of the straight
Thus, the two points on the first line are  0, a 21  and (1, a11 ). line, plot them on the
graph paper and finally
Similarly, two points on the second line are  0, a 22  , (1, a12 ). join the points. In that
. unit, we took three
. points just to keep an
.
Two points on the n th line are  0, a 2n  , (1, a1n ).
automatic check on our
calculations. Because if
the three points do not
Let us plot these lines: we first draw two vertical lines: the first along p1 = 0 fall on the line, it means
and the second along p1 = 1 (see Fig. 12.2a). that we have made
some mistake in
We observe that the second coordinates of all points having first coordinate 0 obtaining them. We do
are the same as the payoff values in the second row. Further, the second not get this advantage if
coordinate of the points having first coordinate 1 are the same as payoff values we work with two
in the first row. Thus, for numerical problems we need not obtain two points points.
for each line. We can simply take payoff values of the second row on the
vertical line along p1  0 and payoff values of the first row on the vertical line
along p1  1 (see Fig. 12.2a). Here we have made one assumption that
a 21  a 22  ...  a 2n and a12  a11  a13  a14  ...  a1n . The lines can be plotted
by simply joining the points a 21 to a11 , a 22 to a12 ,..., a 2n to a1n (see Fig. 12.2b).
 Fig. 12.2: Representation of expected payoff values and expected payoff line segments.
Note that these lines represent expected payoffs of row player A. So the lowest
boundary of these lines will give the minimum expected payoff as a function of
p1. Pay attention to the lowest boundary formed by the lower line segments of
expected payoff lines in Fig. 12.2b. It is known as the lower envelope and
indicated by bold line segments. But, obviously player A would like to
maximise his/her expected payoffs (gains). This can be achieved by selecting
the highest point on the lower envelope in Fig. 12.2b, it is indicated by the
point P. If there are only two straight lines passing through the highest point,
then we identify the two strategies corresponding to these lines and eliminate
all other strategies of player B from the payoff matrix. This reduces the game
to a 2  2 two-person zero-sum game and can be solved using the algebraic
method discussed in Sec 12.2.
But if there are more than two straight lines passing through the highest point,
then there will be alternative solutions. To obtain one solution, we select any
two straight lines from among the lines passing through the highest point and
identify the strategies corresponding to these two lines. This will again reduce
the game into 2  2 two-person zero-sum game and can be solved using the
algebraic method discussed in Sec. 12.2. In brief, we can summarise the above
procedure in the following steps:
Step 1: Draw a horizontal line of length one unit which extends from
p1  0 to p1  1.
Step 2: Draw two vertical lines passing through the two points
p1  0 and p1  1 on the horizontal axis.
Step 3: Represent the payoff values of the second row on the vertical line
p1  0 and payoff values of the first row on the vertical line p1  1.

Step 4: Join a 21 to a11 , a 22 to a12 ,..., a 2n to a1n . The n straight lines thus plotted
will represent the expected payoffs of row player A.
Step 5: Identify the lowest boundary of the expected payoff lines as obtained
in Step 4, called the lower envelope. Draw bold lines along the line
segments of the boundary.
Step 6: Identify the highest point on the lower envelope.
Step 7: Identify the strategies corresponding to the expected payoff lines,
which pass through the highest point. If there are only two such
strategies, then eliminate all other strategies. This reduces the game to
2  2 two-person zero-sum game. Solve it using the algebraic method.
If there are more than two such strategies, select any two among
 them and eliminate all other strategies. This will again reduce the
game into 2  2 two-person zero-sum game and can be solved using
the algebraic method.
Let us consider an example to explain the graphical method for solving
2  n games.
Example 4: Solve the two-person zero-sum game having the following payoff
matrix for player A:
Player B
B1 B2 B3
A1 1 2 7
Player A
A2 6 4 2

Solution: We first check if a saddle point exists or not.

Player B Row Maximin Value
B1 B2 B3 Minima
A1 1 2 7 1
Player A max 1, 2  2
A2 6 4 2 2
Column Maxima 6 4 7
Minimax Value min 6, 4, 7  4

Since maximin value  minimax value, there is no saddle point. Hence, we

have to obtain the solution in terms of mixed strategies. Moreover, this game
cannot be reduced by using dominance rules, because it has neither any inferior
row nor any inferior column. But this game is of the type 2  n where n = 3.
Hence, we can apply the graphical method for 2  n games which will reduce
the game into a 2  2 two-person zero-sum game. Let  p1 , p2  and  q1 ,q2 ,q3  be
the optimal mixed strategies for players A and B, respectively. Then the
expected payoffs of player A corresponding to the moves B1 , B2 , B3 of the
player B are given by

e1  1p1  6p 2  p1  6(1  p1 )  5p1  6 … (i)

e2  2p1  4p 2  2p1  4(1  p1 )  2p1  4 … (ii)

e3  7p1  2p2  7p1  2 1  p1   5p1  2 …(iii)

We now follow the steps given below to apply the graphical method for this
2  3 game. Refer to Fig. 12.3.
Step 1: We draw a horizontal line of length 1 unit extending from the point
R(p1  0) to point S(p1  1).

Step 2: We draw two vertical lines p1  0 and p1  1 passing through the

points R and S, respectively, on the horizontal axis.
Step 3: We represent the payoff values of the second row (6, 4, 2) along the
vertical line p1  0 and payoff values of the first row (1, 2, 7) along
the vertical line p1  1 .

Step 4: We plot the straight lines given by (i), (ii) and (iii) by joining the
payoff values of two rows with their corresponding values.
Here we plot the lines by joining 6 to 1, 4 to 2 and 2 to 7 as
shown in Fig. 12.3.
Step 5: We draw bold lines along the line segments, which form the lowest
boundary of expected payoff lines given by (i), (ii) and (iii) to obtain
the lower envelope (see Fig. 12.3).
Step 6: We identify the highest point on the lower envelope. In Fig. 12.3, the
highest point is P, which is the point of intersection of expected payoff
lines given by (ii) and (iii). Thus, the given game reduces to:
Player B
q2 q3
B2 B3
p1 A1 2 7
Player A
p2 A2 4 2

Fig. 12.3: Graph for player A.

We know that for 2  2 two-person zero-sum games
p1 , p 2 , q 2 , q 3 and v are given from equations (3 a to e) as:
dc ab db
p1  , p2  , q2 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a c ad  bc
q3  and v 
(a  b)  (d  c) (a  b)  (d  c)
In this case, a = 2, b = 7, c = 4, d = 2.
24 2 2 27 5 5
 p1    , p2   
(2  7)  (2  4) 7 7 (2  7)  (2  4) 7 7
27 5 5 24 2 2
q2    , q3   
(2  7)  (2  4) 7 7 (2  7)  (2  4) 7 7
2 2  7 4 24 24
v  
(2  7)  (2  4) 7 7
Hence, the optimal strategies for original game for players A and B are
(2/7, 5/7) and (0, 5/7, 2/7), respectively, and the value of the game is
24/7.
We now discuss the graphical method for m  2 games.
12.4.2 Graphical Method for Reducing m  2 Games
to 2  2 Games
In an m  2 game, the column player B has two moves and the row player A
may have any finite number of moves (say, m). Let the payoff matrix for player
A for an m  2 game be:
 Player B
B1 B2
A1 a11 a12
A2 a 21 a 22
.. .. ..
. . .
Player A
Ai a i1 ai2
.. .. ..
. . .
Am a m1 a m2

If the game has a saddle point, it can be solved using the maximin-minimax
principle discussed in Unit 11, which gives pure strategies as a solution.
Suppose the game does not have a saddle point, which means that the solution
of the game will be in terms of mixed strategies. Let (p1 , p 2 ,..., p m ) and
(q1 , q 2 ) be the optimal mixed strategies for player A and player B, respectively.
Let us rewrite the payoff matrix with these probabilities as follows:
Player B
q1 q2
B1 B2
p1 A1 a11 a12
p2 A2 a 21 a 22
.. .. .. ..
. . . .
Player A
pi Ai a i1 ai2
.. .. .. ..
. . . .
pm Am a m1 a m2

If e1 , e2 ,..., em are the expected payoffs for player B corresponding to different

strategies of player A, then these are given by:
e1  a11q1  a12 q 2 if player A employs strategy A1
e2  a 21q1  a 22q 2 if player A employs strategy A 2
. .
. .
. .
em  a m1q1  a m2q 2 if player A employs strategy A m
But q1  q 2  1  q 2  1  q1.
So we can write the expected payoffs for player B in terms of q 1 as follows:
e1   a11  a12  q1  a12 if player A employs strategyA1
e2   a 21  a 22  q1  a 22 if player A employs strategyA2
. .
. .
. .
em   a m1  a m2  q1  a m2 if player A employs strategyAm

The expected payoffs for player B given above are linear equations in q 1 and
hence represent straight lines. We can plot these lines on a graph paper as
explained in Sec. 12.4.1. We take the values of q 1 along the horizontal axis
from q1  0 to q1  1  i.e., 0  q1  1 . So the horizontal axis is restricted at 1
unit starting from q1  0 and ending at q1  1 (see Fig. 12.4). We take expected
payoffs along the vertical axis.
Let us consider the first line:
e1   a11  a12  q1  a12
We know that a straight line can be plotted easily by joining two points on it.
So, we take the points q1  0 and q1  1 on the line:
At q1  0, e1  a12
At q1  1, e1   a11  a12 1  a12  a11
Thus, two points on the first line are  0, a12  and 1, a11  .
Similarly, two points on the second line are  0, a 22  and 1, a 21  .
.
.
.
Two points on the mth line are  0, a m2  and 1, a m1  .
Let us plot these lines. For convenience, we draw two vertical lines: the first
along q1  0 and the second along q1  1 as shown in Fig 12.4. We observe that
the second coordinates of all points having first coordinate 0 are the same as
the payoff values in second column. Moreover, the second coordinates of the
points having first coordinate 1 are the same as the payoff values in the first
column. Thus, for numerical problems we need not obtain two points for each
line. We can simply take payoff values of the second column along the vertical
line q1  0 and payoff values of the first column along the vertical line q1  1 .
The lines can be plotted by simply joining a11 to a12 , a 21 to a 22 ,..., a m1 to a m2 .
Note that these lines represent expected payoffs (losses) of column player B.
So the uppermost boundary of these lines will give the maximum expected
payoffs (losses) as a function of q1. The uppermost boundary formed by the
uppermost line segments of expected payoff lines is known as the upper
envelope and indicated by bold line segments.
Now, player A would like to minimise his/her expected payoffs (losses). This
can be achieved by selecting the lowest point on the upper envelope. If only
two straight lines pass through the lowest point, then we identify the strategies
corresponding to these lines and eliminate all other strategies of player A from
the payoff matrix. This reduces the game to a 2  2 two-person zero-sum game.
So it can be solved using the algebraic method.
However, if more than two straight lines pass through the lowest point, then
there will be alternative solutions. To obtain one solution, we select any two
straight lines from among the lines passing through the lowest point and
identify the strategies corresponding to these two lines. This will again reduce
the game to a 2  2 two-person zero-sum game, which can be solved using the
algebraic method. In brief, we can summarise the above procedure in the
following steps.
Step 1: Draw a horizontal line of length one unit, which extends from the
point R(q1  0) to the point S(q1  1).
Step 2: Draw two vertical lines passing through the points R and S,
respectively, on the horizontal axis.
Step 3: Represent the payoff values of the second column on the vertical line
along q1  0 and payoff values of the first column on the vertical line
along q1  1.

Step 4: Join a11 to a12 , a 21 to a 22 ,..., a m1 to a m2 . The m straight lines thus

plotted will represent the expected payoffs (losses) of column player
B.
Step 5: Identify the uppermost boundary of expected payoff lines obtained in
Step 4, called the upper envelope. Draw bold lines along the line
segments of the boundary.
Step 6: Identify the lowest point on the upper envelope.
Step 7: Identify the strategies corresponding to expected payoff lines, which
pass through the lowest point. If there are only two such strategies,
eliminate all other strategies. This will reduce the m  2 game to a
2  2 two-person zero-sum game. We can solve it using the algebraic
method. But if there are more than two such strategies, select any two
from them and eliminate all other strategies. This will again reduce the
game to a 2  2 two-person zero-sum game, which can be solved
using the algebraic method.
Let us consider an example to explain the graphical method for m  2 games.
Example 5: Solve the two-person zero-sum game having the following payoff
matrix for player A:
Player B
B1 B2
A1 5 –2
Player A A2 –3 6
A3 –2 3

Solution: We first check if a saddle point exists or not.

Player B Row Maximin Value
B1 B2 Minima
A1 5 –2 –2
Player A A2 –3 6 –3 max{2, 3, 2}  2
A3 –2 3 –2
Column Maxima 5 6
Minimax Value min{5,6}  5

Since maximin value  minimax value, there is no saddle point. Hence, we

have to obtain the solution in terms of mixed strategies. Also, we cannot reduce
this game using dominance rules, because there is neither any inferior row nor
any inferior column. But this game is of the type m  2 where m = 3. Hence, we
can apply the graphical method for m  2 games, which will reduce the game to
a 2  2 two-person zero-sum game. Let  p1, p2 , p3  and  q1,q2  be the optimal
mixed strategies for players A and B, respectively. The expected payoffs of
player B corresponding to the moves A1 , A 2 , A 3 of the player A are given by
e1  5q1  2q 2  5q1  2(1  q1 )  7q1  2 … (i)
e2  3q1  6q 2  3q1  6(1  q1 )  9q1  6 … (ii)
e3  2q1  3q 2  2q1  3(1  q1 )  5q1  3 … (iii)
We now follow the steps given below to apply the graphical method for
m  2 games.
Step 1: We draw a horizontal line, 1 unit in length, extending from the point
R  q1  0  to the point S  q1  1. (see Fig. 12.4).

Step 2: We draw two vertical lines, q1  0 and q1  1, passing through the two
points R and S on the horizontal axis.
Step 3: We represent the payoff values of the second column (i.e.,
–2, 6, 3) along the vertical line q1  0 and payoff values of the first
column (i.e., 5, –3, –2) along the vertical line q1  1 (see Fig. 12.4).
Step 4: We plot the straight lines given by (i), (ii) and (iii) by
joining the payoff values of the two columns with their corresponding
values. Here we plot the line by joining 5 to –2, –3 to 6 and –2 to
3 as shown in Fig. 12.4.
Step 5: We draw bold lines along the line segments, which are the uppermost
boundary of expected payoff lines given by (i), (ii) and (iii), to obtain
the upper envelope.
Step 6: We identify the lowest point on the upper envelope. Here the lowest
point is P, which is the point of intersection of expected payoff lines
given by (i) and (ii). Thus, the given game reduces to:
Player B
q1 q2
B1 B2
p1 A1 5 –2
Player A
p2 A2 –3 6

We know that for 2  2 two-person zero-sum games,

p1 , p 2 , q1 , q 2 and the value of the game v are given from equations
(3a to e) as:
dc ab db
p1  , p2  , q1 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a c ad  bc
q2  and v 
(a  b)  (d  c) (a  b)  (d  c)
In this case a = 5, b = –2, c = –3, d = 6.
6  (3) 9 5  ( 2) 7
 p1   , p2  
 5  (2)    6  (3)  16  5  (2)    6  (3)  16
6  (2) 8 1 5  ( 3) 8 1
q1    , q2   
 5  (2)    6  (3)  16 2 5  (2)    6  ( 3)  16 2
5  6  (2)(3) 24 3
and v   
 5  (2)    6  (3)  16 2
Hence, the optimal strategies for the original game for players A and B
 9 7  1 1 3
are  , , 0  and  ,  , respectively, and the value of the game is .
 16 16  2 2 2
 Fig. 12.4: Graph for player B.
You may try the following exercises for applying the graphical methods to
solve 2  n and m  2 games.

E4) Solve the two-person zero-sum game having the following payoff
matrix for player A:
Player B
B1 B2 B3 B4 B5
A1 3 4 5 –2 2
Player A
A2 1 6 –3 3 7

E5) Solve the two-person zero-sum game having the following payoff
matrix for player A:
Player B
B1 B2
A1 8 –2
A2 5 1
Player A
A3 2 3
A4 4 7

12.5 SOLUTION OF 3  3 TWO-PERSON ZERO-

SUM GAME
In Sec. 12.2, we have derived a general result for solving 2  2 two-person
zero-sum games. In this section, we are not going to derive such a result for
3  3 two-person zero-sum games. We shall simply solve a 3  3 two-person
zero-sum game to give you an idea how such games can be solved. Let us
explain this idea with the help of a very popular game known as Stone-Paper-
Scissors game.
Stone-Paper-Scissors Game
Let us call the two players playing the game as player A and player B. The
rules of the game are as follows:
 In each turn, both players will simultaneously have to produce by a gesture
of their hands, either a stone, a paper or a pair of scissors (see Fig. 12.5).
 Stone beats scissors (since a stone can crush scissors), scissors beat paper
(since scissors can cut paper) and paper beats stone (since a stone can be
covered by paper).
 If both players produce the same gesture of their hands, it is a tie.
 In each turn, the loser has to pay Rs 1 to the winner, and in case of a tie both
get rupee zero as payoff from each other.
What are the optimal strategies for players A and B?
Solution: The above information can be represented by a payoff matrix as
given below. Note that the first entry of each of the nine cells represents the
payoff value of row player A and the second entry represents the payoff value
of column player B. You should also study Fig. 12.5 to understand the game
better.
Player B
Stone(B1 ) Paper (B2 ) Scissors(B3 )
Stone(A1 ) 0, 0 –1,1 1, –1
Player A Paper (A2 ) 1, –1 0, 0 –1, 1
Scissors(A3 ) –1, 1 1, –1 0, 0

Since it is a two-person zero-sum game, the payoff values of row player are
equal in magnitude but opposite in sign to the payoff values of the column
player in each play. So we can simply represent the game by the following
payoff matrix for the row player (refer to Note 2 in Unit 11 of this course):
Player B
Stone(B1 ) Paper (B2 ) Scissors(B3 )
Stone(A1 ) 0 –1 1
Player A Paper (A2 ) 1 0 –1
Scissors(A3 ) –1 1 0

Fig. 12.5: Gestures for the stone-paper-scissors game.

We first check for the saddle point as follows:
Player B Row Maximin Value
Minima
Stone(B1 ) Paper (B2 ) Scissors(B3 )
Stone(A1 ) 0 –1 1 –1
Player A

Paper (A2 ) 1 0 –1 –1 max{1, 1, 1}  1

Scissors(A3 ) –1 1 0 –1
Column Maxima 1 1 1
Minimax Value min{1,1,1}  1
Since the maximin value  minimax value, there is no saddle point. Hence, we
have to obtain the solution in terms of mixed strategies. Also, we cannot reduce
this game by using dominance rules because there is neither any inferior row
nor any inferior column. Further, it is neither a 2  n nor an m  2 game. So it
cannot be reduced to a 2  2 two-person zero-sum game using the graphical
method. Hence, we cannot obtain its solution using any of the methods
discussed so far. What do we do now? Let us find out.
We first apply the fundamental theorem of game theory (stated in Sec. 12.2) to
this game. According to the theorem, since this is a finite two-person zero-sum
game, it has a solution in terms of mixed strategies. Let
 p1, p2 , p3  and q1, q2 , q3  be the optimal mixed strategies for players A and
B, respectively, and v be the value of the game. Note that there is one
commonality between 2  2 games (discussed in Sec. 12.2) and the 3  3 game.
Both are square games. So let us proceed as in Sec. 12.2 and rewrite the payoff
matrix with their probabilities:
Player B
q1 q2 q3
Stone(B1 ) Paper (B2 ) Scissors(B3 )
p1 Stone(A1 ) 0 –1 1
Player A p2 Paper (A2 ) 1 0 –1
p3 Scissors(A3 ) –1 1 0

Now, the expected gain of player A is

0p1  1p 2  (1)p3  p 2  p3 when player B employs strategy B1

 1 p1  0p2 1p3  p1  p3 when player B employs strategyB2

1p1   1 p2  0p3  p1  p2 when player B employs strategyB3

Similarly, the expected loss of player B is
0q1  (1)q 2  1q 3  q 2  q 3 when player A employs strategy A1

1q1  0q 2  (1)q 3  q1  q 3 when player A employs strategy A 2

 1 q1 1q2  0q3  q1  q2 when player A employs strategyA3

Since v is the value of the game and  p1 , p2 , p3  is the optimal mixed strategy
for player A, we have
p2  p3  p1  p3  p1  p2  v
I II III IV
Now,
I & II  p1  p 2  2p3  0 … (i)
and II & III  2p1  p 2  p3  0 … (ii)

Also p1  p 2  p3  1 … (iii)
Subtracting (i) from (iii) gives
1
3p3  1  p3 
3
Subtracting (ii) from (iii) gives
1
3p1  1  p1 
3
Putting the values of p1 and p 3 in (iii), we get
1 1 1
 p2   1  p2 
3 3 3
1 1
Again I & IV  v  p 2  p3  v    v  0
3 3
Also, v is value of the game and  q1 ,q2 ,q3  is the optimal mixed strategy for
the player B.
q 2  q3  q1  q3  q1  q 2  v
I II III IV
I & II  q1  q 2  2q 3  0 … (iv)
II & III  2q1  q 2  q 3  0 … (v)

Also q1  q 2  q3  1 … (vi)
Adding (iv) and (vi) gives
1
3q 3  1  q 3 
3
Adding (v) and (vi) gives
1
3q1  1  q1 
3
Putting the values of q1 and q 3 in (vi), we get
1 1 1
 q2   1  q2 
3 3 3
1 1
I & IV  v  q 2  q3  v     v  0
3 3
Hence, the optimal strategy for both players A and B is (1/3, 1/3, 1/3) and the
value of the game is 0.

12.6 FLOWCHART
In Unit 11, we have introduced some key terms involved in the game theory.
You have studied the maximin-minimax principle to solve two-person zero-
sum games with saddle point. In Sec. 12.2, we have derived and applied the
algebraic method to solve 2  2 two-person zero-sum games. In Sec. 12.3, you
have learnt the dominance rules and applied them to reduce the size of the
payoff matrix by eliminating inferior row(s) or column(s) or both. In Sec. 12.4,
we have discussed the graphical method, which reduces 2  n and m  2 games
into 2  2 games. We have also solved some games to explain the steps
involved in this method. Finally, in Sec. 12.5, you have seen how 3  3 two-
person zero-sum games can be solved by extending the technique for 2  2 two-
person zero-sum games discussed in Sec 12.2. We have also solved a
commonly played game known as Stone-Paper-Scissors game. At this stage,
you may wish to know: How do we solve the m  n games, which are not
reducible to any of the games discussed in this unit? The answer is
that m  n two-person zero-sum games can be solved by converting them into
linear programming problems. But this discussion is beyond the scope of this
course. However, if you are keen to learn this technique, you should first learn
how to solve linear programming problems. This is discussed in detail in the
course MSTE-002. Then you may refer to any book recommended at the end of
this block or any other book available to you. We end the discussion by giving
a flow chart in Fig. 12.6, which shows all the information that we have
explained in this unit sequentially.

Fig. 12.6: Flowchart of game theory approach.

Let us now present a summary of the main points discussed in this unit.

12.7 SUMMARY
1) The solution of a 2  2 two-person zero-sum games in terms of mixed
strategies is provided by the algebraic method:
 dc
p1 
(a  b)  (d  c)
ab
p2 
(a  b)  (d  c)
db
q1 
 a  b   (d  c)
a c
q2 
(a  b)  (d  c)
ad  bc
v
(a  b)  (d  c)
2) Dominant Strategy: A strategy is said to be the dominant strategy for a
player if it is better than another strategy available to that player
irrespective of the strategy adopted by the other player.
3) The dominance rules for reducing the size of the payoff matrix are:
Rule 1
If all the elements in a row (say, the i th row) of a payoff matrix for row
player are less than or equal to the corresponding elements of the other
row (say, the jth row), then the row player A will never employ his/her
i th strategy because it is dominated by his/her jth strategy. So we can
delete the i th row from the payoff matrix.
Rule 2
If all the elements in a column (say, the r th column) of a payoff matrix
for column player are greater than or equal to the corresponding
elements of the other column (say, the s th column), then the column
player B will never employ his/her r th strategy because it is dominated
by his/her s th strategy. So we can delete the r th column from the payoff
matrix.
Rule 3
A strategy can also be deleted if it is dominated by average (or any
convex combination) of two or more other pure strategies.
4) 2  n and m  2 games are solved by using the graphical method. The
graphical method reduces 2  n and m  2 games to 2  2 games.
Once a game is reduced to size 2  2 it can be solved by the algebraic
method.

12.8 SOLUTIONS/ANSWERS
E1) We first check for the saddle point:
Player B Row
B1 B2 Maximin Value
Minima
A1 –2 –1 –2
Player A max{2,  3}  2
A2 4 –3 –3
Column Maxima 4 –1
Minimax Value min{4,  1}  1
 Since the maximin value (–2)  minimax value (–1), there is no saddle
point. Hence, we have to obtain the solution in terms of mixed strategies.
Let  p1 , p2  and  q1 ,q2  be the optimal mixed strategies for player A and
player B, respectively, and v be the value of the game. Then from
equations (3a to e), we have
dc ab db
p1  , p2  , q1 
(a  b)  (d  c)  a  b   (d  c) (a  b)  (d  c)

a c ad  bc
q2  and v 
 a  b   (d  c) (a  b)  (d  c)
In this case, a = –2, b = –1, c = 4, d = –3.
3  4 7 7 2  (1) 1 1
 p1    , p2   
2  (1)  (3)  4 8 8 2  (1)  (3)  4 8 8
3  (1) 2 1 2  4 6 3
q1    , q2   
2  (1)  (3)  4 8 4 2  (1)  (3)  4 8 4
(2)(3)  (1)  4 10 5
v  
2  (1)  (3)  4 8 4
Hence, the solution of the game is:
The optimal strategy for player A is (7/8, 1/8), the optimal strategy for
player B is (1/4, 3/4) and the value of the game is  5/4.
E2) In Note 2 of Unit 11, we have explained that the coin matching game is a
two-person zero-sum game between child X (player I) and child Y
(player II). The payoff matrix for player I is given below:
Child Y (player II)
Head Tail
Child X Head 1 –1
(Player I) Tail –1 1
We first check for the saddle point.
Player II
Row Minima Maximin Value
B1 B2
A1 1 –1 –1
Player I max{1,  1}  1
A2 –1 1 –1
Column Maxima 1 1
Minimax Value min{1, 1}  1

Since maximin value (–1)  minimax value (1), there is no saddle point.
Hence, we obtain the solution in terms of mixed strategies.
Let  p1 , p2  and  q1 ,q2  be the optimal mixed strategies for player A and
player B, respectively, and v be the value of the game.
From equations (3a to e),
dc a b db
p1  , p2  , q1 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a c ad  bc
q2  , and v 
(a  b)  (d  c) (a  b)  (d  c)
In this case, a = 1, b = –1, c = –1, d = 1
 1  (1) 2 1 1  (1) 2 1
 p1    , p2   
1  (1)  1  (1) 4 2 1  (1)  1  (1) 4 2
1  (1) 2 1 1  (1) 2 1
q1    , q2   
1  (1)  1  (1) 4 2 1  (1)  1  (1) 4 2
11  (1)  (1) 0
v  0
1  (1)  1  (1) 4
Hence, the solution of the game is:
The optimal strategy for both child X (player I) and child Y (player II)
is (1/2, 1/2) and the value of the game is 0.
E3) We first check for the saddle point.
Player B Row
Maximin Value
Minima
B1 B2 B3 B4
A1 4 3 2 1 1
Player A A2 6 4 5 0 0 max{1,0,0}  1
A3 1 2 0 3 0
Column Maxima 6 4 5 3
Minimax Value min{6, 4, 5, 3}  3

Since the maximin value  minimax value, there is no saddle point.

Hence, we obtain the solution in terms of mixed strategies. Let us first
reduce the payoff matrix by applying the dominance rules.
Step 1: We note that each element of the first column is greater than or
equal to the corresponding element of the third column. This
implies that the third column dominates the first column. So, we
eliminate the first column from the given payoff matrix. Thus,
the reduced payoff matrix can be written as follows:
Player B
B2 B3 B4
A1 3 2 1
Player A A2 4 5 0
A3 2 0 3

Step 2: In the reduced payoff matrix given above, neither any row is
directly dominated by any other row nor any column is directly
dominated by any other column. But we observe that the
average of the second and the third columns of the reduced
payoff matrix dominates the first column. So, we can eliminate
the first column from the reduced payoff matrix and obtain a
further reduced payoff matrix as:
Player B
B3 B4
A1 2 1
Player A A2 5 0
A3 0 3

Step 3: In the reduced payoff matrix given above, neither any row is
directly dominated by any other row nor any column is directly
dominated by any other column. But we observe that the
average of the second and third rows of the reduced payoff
 matrix dominates the first row. So, we can eliminate the first
row from the reduced payoff matrix and obtain a further reduced
payoff matrix as:
Player B
B3 B4
A2 5 0
Player A
A3 0 3
Now, neither any row dominates any other row nor any column
dominates any other column. So, this reduced payoff matrix
cannot be reduced further by using dominance rules. But this
reduced payoff matrix is of the order 2  2. So, we can apply the
algebraic method for 2  2 two-person zero-sum games. Let
(p1 , p 2 , p3 ), (q1 , q 2 , q 3 , q 4 ) be the mixed strategies for players A
and B, respectively, for the original payoff matrix. The (2  2)
reduced payoff matrix tells us that (p 2 , p3 ) and (q 3 , q 4 ) are the
mixed strategies for players A and B, respectively. If v is the
value of the game, then from equations (3a to e), we have
dc ab db
p2  , p3  , q3 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a c ad  bc
q4  and v 
(a  b)  (d  c) (a  b)  (d  c)
In this case, a = 5, b = 0, c = 0, d = 3.
30 3 50 5
 p2   , p3  
(5  0)  (3  0) 8 (5  0)  (3  0) 8
30 3 50 5
q3   , q4  
(5  0)  (3  0) 8 (5  0)  (3  0) 8
5 3  0 0 15
v 
(5  0)  (3  0) 8
Hence, the solution of the game is:
The optimal strategy for player A is (0, 3/8, 5/8), the optimal
strategy for player B is (0, 0, 3/8, 5/8) and the value of the game
is 15/8.
E4) We first check for the saddle point.
Player B Row Maximin Value
B1 B2 B3 B4 B5 Minima
A1 3 4 5 –2 2 –2
Player A max 2, 3  2
A2 1 6 –3 3 7 –3
Column Maxima 3 6 5 3 7
Minimax Value min 3, 6, 5, 3, 7  3

Since maximin value  minimax value, there is no saddle point.

Hence, we have to obtain the solution in terms of mixed strategies. But
column four dominates the second and fifth columns. So, by dominance
rules, both second and fifth columns can be eliminated from this payoff
matrix and the reduced payoff matrix can be written as:
 Player B
B1 B3 B4
A1 3 5 –2
Player A
A2 1 –3 3

This payoff matrix cannot be reduced further by using dominance rules.

But this game is of the type 2  n where n = 3. Hence, we can
apply the graphical method for 2  n games, which will reduce the game
to a 2  2 two-person zero-sum game. Let  p1 , p2  and
 q1, q2 , q3 , q4 , q5  be the optimal mixed strategies for players A and B,
respectively, for the original payoff matrix. Then the expected payoffs of
player A corresponding to the moves B1 , B3 , B4 of player B are given by:
e1  3p1  p2  3p1  1  p1   2p1  1 … (i)

e3  5p1  3p2  5p1  31  p1   8p1  3 … (ii)

e4  2p1  3p2  2p1  31  p1   5p1  3 … (iii)

We now follow the steps given below to apply the graphical method
for 2  n games as shown in Fig. 12.7:
Step 1: We draw a horizontal line, 1 unit in length, extending from the
point R  p1  0  to point S  p1  1 .
Step 2: We draw two vertical lines p1  0 and p1  1 passing through the
two points R and S on the horizontal axis.
Step 3: We represent the payoff values of the second row (i.e., 1, –3, 3)
along the vertical line p1  0 and payoff values of the first row
(i.e., 3, 5, –2) along the vertical line p1  1 .
Step 4: We plot the straight lines given by (i), (ii) and (iii) by joining the
payoff values of two rows with their corresponding values.
Here we plot the lines by joining 3 to 1, 5 to –3 and –2 to 3 as
shown in Fig. 12.7.
Step 5: We draw bold lines along the line segments which form the
lowest boundary of expected payoff lines given by (i), (ii), (iii).
It is the lower envelope.
Step 6: We identify the highest point P on the lower envelope. Here it is
the point of intersection of expected payoff lines given by (ii)
and (iii). Thus, the given game reduces to:
Player B
q3 q4
B3 B4
p1 A1 5 –2
Player A
p2 A2 –3 3
From equations (3a to e), we have
dc ab db
p1  , p2  , q2 
(a  b)  (d  c)  a  b   (d  c) (a  b)  (d  c)
a c ad  bc
q3  and v 
 a  b   (d  c) (a  b)  (d  c)
 Here a = 5, b = –2, c = –3, d = 3.
3  (3) 6 5  (2) 7
 p1   , p2  
 5  (2)    3  (3)  13  5  (2)    3  (3)  13
3  (2) 5 5  (3) 8
q2   , q3  
 5  (2)    3  (3)  13  5  (2)    3  (3)  13
5  3  (2)  (3) 9
v 
 5  (2)    3  (3)  13

Hence, the optimal strategies for original game for the players A and B
6 7  5 8 
are  ,  and  0, 0, , , 0  , respectively, and the value of the
 13 13   13 13 
9
game is .
13

Fig. 12.7: Graph for player A in E4.

E 5) We first check for the saddle point.
Player B Row Maximin Value
B1 B2 Minima
A1 8 –2 –2
A2 5 1 1
Player A max 2, 1, 2, 4  4
A3 2 3 2
A4 4 7 4
Column Maxima 8 7
Minimax Value min{8, 7}  7

Since maximin value  minimax value, there is no saddle point. Hence,

we have to obtain the solution in terms of mixed strategies. Since the
third row is dominated by the fourth row, using dominance rules, we can
eliminate the third row from this payoff matrix. The reduced payoff
matrix can be written as:
Player B
B1 B2
A1 8 –2
Player A A2 5 1
A4 4 7

This payoff matrix cannot be reduced further by using dominance rules.

But this game is of the type m  2 , where m = 3. Hence, we can apply the
graphical method for m  2 games, which will reduce the game to a 2  2
two-person zero-sum game. Let  p1, p2 , p3 , p4  and  q1,q2  be the optimal
mixed strategies for players A and B, respectively, for the original payoff
matrix. The expected payoffs of player B corresponding to the moves
A1 , A 2 , A 4 of player A are given by
e1  8q1  2q2  8q1  2 1  q1   10q1  2 … (i)
e2  5q1  1q2  5q1  11  q1   4q1  1 … (ii)
e4  4q1  7q 2  4q1  7(1  q1 )  3q1  7 … (iii)
We follow the steps given below to apply the graphical method for
solving m  2 games as shown in Fig.12.8:
Step 1: We draw a horizontal line, 1 unit in length, extending from the
point R  q1  0 to point S(q1  1) (see Fig. 12.8).

Step 2: We draw two vertical lines q1  0 and q1  1 passing through the

two points R and S on the horizontal axis.
Step 3: We represent the payoff values of the second column
(i.e.,–2, 1, 7) along the vertical line q1  0 and the payoff values
of the first column (i.e., 8, 5, 4) along the vertical line q1  1 as
shown in Fig. 12.8.
Step 4: We plot the straight lines given by (i), (ii) and (iii) by joining the
payoff values of two columns with their corresponding values.
Here we plot the line by joining 8 to –2, 5 to 1 and 4 to 7 as
shown in Fig. 12.8.
Step 5: We draw bold lines along the line segments, which form the
uppermost boundary of the expected payoff lines given by (i),
(ii), (iii). This is the upper envelope.
Step 6: We identify the lowest point P on the upper envelope. Here it is
the point of intersection of expected payoff lines given by (i)
and (iii). Thus, the given game reduces to:
Player B
q1 q2
B1 B2
p1 A1 8 –2
Player A
p4 A4 4 7

From equations (3a to e), we have

dc a b db
p1  , p4  , q1 
(a  b)  (d  c) (a  b)  (d  c) (a  b)  (d  c)
a c ad  bc
q2  and v 
(a  b)  (d  c) (a  b)  (d  c)
In this case, a = 8, b = –2, c = 4, d = 7.

74 3 8  (2) 10
 p1   , p4  
 8  (2)    7  4  13 8  (2)    7  4  13
 7  (2) 9 84 4
q1   , q2  
 8  (2)   (7  4) 13 8  (2)   (7  4) 13
8  7  ( 2)  4 64
v 
 8  (2)    7  4  13
Hence, the optimal strategies for original game for players A and B are
 3 10  9 4
 , 0, 0,  and  ,  , respectively, and the value of the game is
 13 13   13 13 
64
.
13

Fig. 12.8: Graph for player B.

FURTHER READING
1) Operations Research by Kanti Swarup, P.K. Gupta and Man
Mohan; Sultan Chand & Sons Education Publishers, New
Delhi (2006) [Chapters 16 and 17].

2) Basic Statistics by B. L. Agarwal; New Age International (P)

Limited, Publishers, New Delhi (2009) [Chapter 20].

3) Mathematics Programming Techniques by N. S. Kambo;

Affiliated East-West Press Pvt. Ltd., New Delhi (1984)
[Chapter 16].