GAME THEORY

In a real life situation, there is always a conflict of interest between two or more
parties engaged in the same field of business. Consider an example where two
businessmen, say A and B, are working in the same field having some conflict of
interest between them regarding business matters. In game theory terminology, the
businessmen A and B are called players and the business is called a game. Suppose
A has three executives and B has four, namely A1,A2,A3 and B1,B2,B3,B4
respectively to control their business. We make a restriction that both players use
the service of their executives only one at a time, but the selection of the
executives entirely depends on their own discretion ignoring the selection done by
the other player. The selection of a particular executive (or a move) is called a
strategy taken by the players and selection of only one move or a strategy by a
player ignoring the strategy taken by his opponent is called a pure strategy.
We shall first consider problems relating to pure strategies taken by the players.
We make two assumptions:
(i) The player A is in a better position and is called the maximizing player or row
player and the player B is called a minimizing player or column player.
(ii) The total gain of one player is exactly equal to the total loss of the other player,
so that the sum total is zero.
In general, if the player A takes m pure strategies and the player B takes n, then the
game is called a two person zero sum game, or, mxn rectangular game. If m = n
then the game is called a square game.
We write down below a matrix called the pay-off matrix of the maximization
player A.
Player

Player B

From the table, if A assumes his pure move

and B assumes his pure move
then the gain of the player A is
. If
is negative, then it is a loss for A and if
, then there is no gain or no loss for both the players. If As gain is p for
some moves of A and B, then Bs loss is p for the same set of moves. So, if the
pay-off matrix of the maximization player A is C, then the pay-off matrix of the
minimization player B is C. A pay-off matrix is always written for the
maximization player. We now define a pay-off matrix in a formal manner.
Pay-off matrix: A pay-off matrix is a real rectangular matrix (

, where

indicates the gain of the maximization player (row player) for using the ith and the
jth moves of the row and the column players respectively.
From a practical point of view, As aim is to maximize his gain irrespective of the
moves taken by B, that is maximizing the minimum expected gain corresponding
to each move of B. In the same way, Bs aim is to minimizing the maximum
expected loss corresponding to each move of B. In mathematical language,
As problem is to maximize his minimum gains and
Bs problem is to minimize his maximum losses
whatever may be the strategies pure) taken by both players.
In general,
maximum of the minimum gains for A minimum of the maximum losses for B
When the equality holds, we say that the game problem is solved and the value of
the game is the equal value.
Assuming the existence of the value of the game, if the value is
, then
is
called the saddle point of the pay-off matrix. It is clear that there may exist more
than one saddle point in a pay-off matrix of a game with pure strategy. The
principle of determining a saddle point or points and the value of a game is called
the maximin minimax principle. If the value of the game is
for the pure
strategies
and of A and B respectively, then
and are called optimal
strategies for the respective players.
Determination of the value of a game using maximin minimax principle
Consider the following 3x4 pay-off matrix for A.

Player B
Row minima
4
6
-2 1
-2
Player A
3
3
4
2
2
4
5
5
1
1
Column maxima 4
6
5
2
2/2
As pure moves are ,
and Bs pure moves are ,
. Each player
has a right to select a pure move.
For As pure move
, As minimum gain is -2 , for
, they are 2 , 1
respectively. So if A selects
, his gain cannot be less than 2.
Now considering Bs point of view, reasoning in the same way, his loss cannot be
more than 2.
In case of A, we arrived at 2 by maximizing the minimum gain of A, and for B we
arrived at 2 by minimizing the maximum loss of B, which are equal. Hence the
value of the game exists and is equal to 2. The optimal strategies are
and
for
A and B respectively. The (2 ,4) position of the pay-off matrix is the saddle point
of the pay-off matrix.
There may exist more than one saddle point and in that case the optimal strategies
are not unique.
There may not exist a saddle point and in that case the value of the game cannot be
determined by the above principle even if it exists.
Fair game: If the value of the game is zero, that is, no loss or gain for any player,
the game is called a fair game.
Strictly determinate game: If the value of a game is a non-zero quantity, then the
game is called a strictly determinate game. If the value of the game is positive, then
it is in favour of the player A and if it is negative, then it is in favour of the player
B.
Theorem: Let (
Then prove that

be the pay-off matrix for a two person zero sum game.

.

Proof: We have,
for any , and
for any .
Let the above maximum be attained at
and the minimum be attained at
Then
for any
From this we have
for all
.
Hence,
.

Formation of the pay-off matrix

Example 1: Each of two players A and B shows one or two fingers simultaneously.
The player B pays to A an amount equal to the total number of fingers shown. On
the other hand A pays to B an amount equal to the product of the number of fingers
shown. Form the pay-off matrix.
Solution:

1+1=2, 1x1=1,
2-1=1
2+1=3, 2x1=2,
3-2=1

So, the pay-off matrix is:

A

1+2=3, 1x2=2,
3-2=1
2+2=4, 2x2=4,
4-4=0

B
1
1

1
0

Problems
1. Solve the game by using maximin minimax principle.
Player B

4
-2 -1
1
5
-2
5
6
2
1
-2 -3
2. Solve the following 4x4 game and show that there are two saddle points.
Player A

Player B

Player A

4
-2
5
4

2
-1
2
0

3
4
3
0

5
-3
3
1

3. Prove that the following pay-off matrix do not possess a saddle point (i.e. the
value of the game cannot be determined) and establish that
.
Player B

Player A

6
-3
4

0
2
-3

4
-1
5

4. Solve the following game stating the optimal strategies and the saddle
points.
[

Let us now try to solve a very simple 2x2 game problem by maximin minimax
principle. Let the pay-off matrix be
B
4
-1
2
3
From the pay-off matrix, it can easily be determined that
and
. Hence the problem cannot be solved
and the value of the game cannot be determined by this method. To solve the game
we introduce another technique called the principle of mixed strategies.
A

Mixed strategy
Mixed strategy is used to solve a game where there does not exist a saddle point.
The property of the existence of the value of a game is established by a theorem
known as the fundamental theorem of rectangular games.
In the case of pure strategy, both players A and B select only a single move at a
time say and at their discretion, irrespective of the move taken by the other.
In mixed strategy both players A and B select their m and n moves simultaneously.
Let us define a set of m quantities
such that
for all i and

. To obtain a least possible gain, A may utilize the services of the

moves
in such a way that performs times the total work so that
the total work will be done by
. Similarly if
be
quantities such that
for all i and
, then B may utilize the
services of the moves
in such a way that performs times the
total work so that the total work will be done by
, ignoring all about
the moves of A.
The quantities and associated with the ith and jth moves of A and B
respectively are called the probabilities of the respective moves.
We now define two variable vectors
in
and
in . It is always possible to determine some particular value
of and say
such that the value of the game can be determined, i.e., it
is possible to determine the same value of maximum of the minimum expected
gain for A and minimum of the maximum expected loss for B. In particular, if
for
and
for all other ,
for
and
for all
other , then the mixed strategy is reduced to a pure strategy with pure moves
and of A and B respectively.
Pay-off function: Let
. Then the
( ) be any pay-off matrix of order
pay-off function or mathematical expectation of a game which is denoted by

is defined as
, where and are the mixed
strategies for A and B respectively.
In matrix notation
can be written as
.
In particular, if B takes his then the expected gain of A is given by

Similarly, for particular ith pure move of A only, the expected loss of B is given by

.
For any , A is sure that his expected winning will be at least
. He
then maximizes the expression over , so that his expected winnings will be at
least
.
Theorem: Let

be such that both

exist, then

and
.

Proof: Let
be some arbitrarily chosen point in their given specified
ranges. Then we must have
and
Combining these two, we get
But is any arbitrarily chosen point in their given specified range and hence the
inequality holds for all values of . Hence, if be such a point for which
has the minimum value, the inequality remains true.
Therefore,
.
Again, since is any point in its given range, the inequality holds even if we
select that which gives the maximum value of
.
Therefore
Note: Denote

and

, then

There exists a point (

in their given specified range of and such that
which is the value of the game in the case of mixed strategy and
the point (
is called the strategic saddle point.

Solution of a 2x2 game using mixed strategy (Problems without a saddle point in
the case of pure strategy in the pay-off matrix)
Let the pay-off matrix for A be

Player B

Player A

+ with mixed strategies for A and B as

and

respectively such that

.
For the moves
and
of the player B, the net expected gain of the player A are
respectively.
Similarly Bs expected loss will be

If

be the value of the game, then

.
Assuming the existence of the value of the game, and as both
and
, then by the complementary slackness theorem, all the inequalities
strictly reduce to equalities (which will be shown later in LPP method) and so we
have
.
Therefore
,
and so
Also since

, we get

which are the optimal strategies for A .

Similarly,
and
strategies for B and the value of the game will be
. It is easy to see that

are the optimal

all lie in the open

interval (0,1). Also the value of the game will be

.
Note: Whatever be the strategies taken by B, the value of the pay-off function will
remain same with the assumption
. For example, if B uses the
mixed strategies
with
,

the pay-off function

(t
Similarly, for the assumption
the value of the pay-off
function will remain same, whatever be the strategies taken by A.
Example: Find the value of the following 2x2 game algebraically by using mixed
strategies. Also find the optimal strategies of the two players.
Player B

Player A

Solution: The problem has no saddle point. So we solve it by using mixed

strategies for A and B as
and
respectively such that
. Assuming the existence of the value
of the game, we have
,
To determine the optimal values of
and
we have
, the value of the game.
Solving we get
. So
and the value of the game is
( )

( )

( ).

Similarly, from Bs point of view we have

which gives

Hence the optimal strategies are

.
(

) and

Problems
Solve the following 2x2 games (without saddle point) algebraically by using mixed
strategies. Find the optimal strategies of the two players and also the value of the
game.

1.

Player B

Player A
2.

Player B

Player A

Graphical solution of 2 x n or m x 2 game

The mixed strategies discussed above can only be applied to a 2x2 game. By using
graph it is possible to reduce a 2 x n or m x 2 game to a 2x2 game and then it can
be solved by the previous algebraic method.
Let a particular 2 x n game without a saddle point be
Player B

Player A

Let the mixed strategies used by A and B be

and
Then the net expected gain of A when B plays his pure strategy

is given by
as
Both and must lie in the open interval (0,1) , because if any one is equal to
one, the game reduces to a game of pure strategy. Hence
is a linear function
of
or . Considering
as a linear function of , we have from the
limiting values 0 and 1 of ,

for

for
Hence
represents a line segment joining the points
and
for
.
Draw two parallel vertical lines with distance between them being one unit, the
first one represents the line
and the second one represents the line
.
Now draw n line segments joining the points
and
,
.
The lower envelope of these line segments will give the minimum expected gain of
A as a function of
. The highest point of the lower envelope will give the
maximum of minimum gain of A. The line segments passing through the point
corresponding to Bs two pure moves say
and
are the critical moves of B
which will maximize the minimum expected gain of A. The 2x2 pay-off matrix
corresponding to As moves
and
and Bs moves
and will produce the
required result.
Using similar techniques we can solve any m x 2 game.
Note1: If there are more than two lines passing through the highest point of the
lower envelope, then there exists an alternative optimal solution. Choosing any two
lines with opposite slopes will give a solution.
Note2: If there are more than one highest point of the lower envelope, then there
exists an optimal solution corresponding to each of these points.
Problems
1. Solve the following 2x4 game graphically.
Player B
Player A

Solution: There is no saddle point. Let the mixed strategies used by A be

,
,
and lie in the open interval (0,1) .
We now draw two vertical lines
and
, unit distance apart. Mark
both lines using the same scale. Draw 4 line segments joining the points

and
,
: joining the points
: joining the points

. The line segments are

and
, : joining the points
and
, : joining the points

and
and

6
5
4
3
2

Max
0

0
-1

Lower
envelope

-2

The maximum point of the lower envelope lies on the point of intersection of the
lie segments
and
. Therefore the game can be solved by solving the 2x2
pay-off matrix

+ with mixed strategies

for A and
method

for B. Then using the previous

which gives
and the value of the game is

. So
( )

( )

( ).

Similarly we can show that

,
Therefore the final solution to the problem is

and the value of the game is

.
2. Solve the following 2x4 game graphically.
Player B
*

Player A

3. Solve the following 3x2 game graphically.

Player B

Player A

Dominance property
To understand dominance property, we consider a particular pay-off matrix.
Player B

Player A

In the above matrix we see that all elements in the third row are less than or equal
to the corresponding elements of the second row. Hence A will definitely not play
his pure strategy
. So the pay-off matrix reduces to

Again we see that the elements in the second and the fourth columns are greater
than or equal to the corresponding elements of the third column. Therefore B will
definitely not play his pure strategies
and
. So the pay-off matrix again
reduces to

Again elements in the first row are less than or equal to the corresponding elements
of the second row. Hence A will definitely not play his pure strategy
. So the
pay-off matrix reduces to

[ 5 4 ] from which it is clear that B will select his pure move

and the
value of the game is 4 and the optimal strategies are
and
for A and B
respectively.
The above example shows that a pay-off matrix can be reduced only by mere
observation. The property used is called dominance property. In some cases the
problem can be solved only by adopting this method (as shown in the above
example), in other cases it may be reduced to a 2 x 2 or 2 x n or m x 2 game so that
it can be solved by methods discussed earlier.
General rule for dominance
(i)

(ii)

If all the elements of the kth row are less than or equal to the
corresponding elements of the rth row, then we say that the kth row is
dominated by the rth row and we reject the kth row.
If all the elements of the lth column are greater than or equal to the
corresponding elements of the sth column, then we say that the lth column
is dominated by the sth column and we reject the lth column.

Modified dominance property

A pure move or strategy of a maximizing (minimizing) player may be dominated if
it be inferior (superior) to a convex combination of two or more strategies of that
player. This property is called the modified dominance property.
Example: Reduce the following game to a 2x2 game by using dominance and
modified dominance property and then solve the game.

Table 1
Player B

Player A

All the elements of the fourth row are less than the corresponding elements of the
third row. Thus
is dominated by
and thus we drop the fourth row.
Table 2
Player B

Player A

All the elements of the fourth column are greater than the corresponding elements
of the first column. Thus
is dominated by
and thus we drop the fourth
column.
Table 3
Player B

Player A

By using a convex combination of the 2nd and 3rd row we have

[

] . All the elements are greater than the corresponding

elements of the first row, so we drop the first row.

Table 4
Player B

Player A

The convex combination of

and

* + are less than or equal to

the corresponding elements of the 3rd column and thus drop the 3rd column.
Table 5
Player B

Player A

Now we solve the 2x2 game as usual.

Example 2: Reduce the following game by using dominance property and then
solve the game.
Player B

Player A

Solution: 3rd column is dominated by the 1st column, so we drop the 3rd column.
Player B

Player A

] Then we solve the 3x2 game graphically.

Reduction of a game problem to a L.P.P

We first state and prove a lemma.
Lemma: If a quantity has been added to each element of the matrix to make
for all
then the value of the original game will be
, but the
optimal strategies remain unaltered.
Proof: Let the pay-off matrix be (

and let the mixed strategies taken by A

and B be
),

for all i and

and

for all j and

respectively.

Then the pay-off function of the game is given by

Now let a number (positive or negative) be added to each element

pay-off function of the new game is



as

and

.
. Hence the

Without any loss of generality, we can assume that (

.
Then the value of the second game exists and is unique and if the optimal strategies
be
for the second game. Then the value of the game is
.
Again,
.
Combining the two results above, we get,
.
Similarly, we get,
.
Hence, from the two relations above we can say that the value of the original game
exists and is unique, the optimal strategies for both games remain the same.
Also, from the relation
, we can say that the value of the
second game is only increased by the number k.
Note: The relation is also valid if (

Theorem: Every game problem (two person zero sum game) can be converted to a
L.P.P.

Proof: Let the pay-off matrix of a game be ( )

, and
with proper adjustment (by adding a suitable constant to each element of the
matrix, using the previous lemma) we may assume that
for all . So, if
the value of the game exists then it will be a positive quantity. Let the mixed
strategies used by A and B be
),

for all i and

for all j and

respectively.

The net expected gain of A when B plays his pure move

is

Now A will expect a least possible gain . Hence

As all

.
, then

is essentially a positive quantity.

As problem is to select
in such a way that they will satisfy the
conditions stated above and attains its maximum.
Since
, maximization of is equivalent to minimization of
. Now
dividing
by , As problem can be written as
Minimize ,
Subject to

where

Using the same logic, Bs problem can be written as

Maximize ,
Subject to

where

Thus considering from A and Bs point of view, we find that a game problem can
be reduced to a L.P.P. in the two forms given above.

Fundamental theorem of rectangular games

Theorem: If mixed strategies are allowed, the value of a game exists and is unique
(optimal strategies may be different).
Proof: We have already proved that each rectangular game problem can be reduced
to a L.P.P. both from A and Bs point of view.
As problem is
Minimize
Subject to

where

and Bs problem is
Maximize
Subject to

where

We note that the two problems are such that one is the dual of the other.
Using a pure strategy
condition
have

when

of A, which gives
) and taking

(using the
, we
,

which indicates that for a pure strategy

of A , the first
L.P.P. has at least one feasible solution set.
Similarly, it can be shown that the second problem also has a feasible solution set.
Since the two problems are dual of each other, we can say from duality theory that
both problems have finite optimal solutions and
Min

, say, i.e.,

which is the unique value

of the game and the optimal strategies are

, where
,

Note: To solve a game problem by converting it to a L.P.P., it is easier to solve Bs

problem as no artificial variables will be required.
Problems
1. Convert the following game problem to a L.P.P.

Solution: Add 3 to each element of the pay-off matrix to make the value of the
game positive and the new pay-off matrix is

Considering Bs problem, the problem is

Maximize
subject to

,
If max

, then the value of the original game will be

optimal strategy for B is

and the

2. Solve the following game problem by converting it to a L.P.P.

Solution: Add 2 to each element of the pay-off matrix to make the value of the
game positive and the new pay-off matrix is
*

Let the optimal strategies for A and B be

Considering Bs problem, it can be reduced to

Maximize
subject to

Now if max

, the value of the original game will be

optimal solution is

Simplex tables
c
B

Basis
0

0
Min ratio

0
0

1
0

1
0

1
1

1/17

Here
max

at

0
4/17 -3/17
0
1
-1/17 5/17
0
3/17
2/17
, so the solution is optimal.
.

and the

Then the value of the original game is

at

Using duality theory

. So,

,
(

Thus the optimal strategies are

),

3. Solve the following game problem using L.P. method.

*

Player A

4. In a two person game, each player simultaneously shows either one or two
fingers. If the number of fingers match then the player A wins a rupee from
the player B, otherwise A pays a rupee to B. Show that the pay-off matrix
for the game is
B
A *

+ . Solve the game by reducing it to a L.P.P.