2 GAME THEORY

1. Introduction

2. Strategies
criterion
3. The maximin and minimax
4. Existence of saddle point

5. Illustrative Examples
6. Games without a saddle point
7. Further consideration on mixed strategies
8. Solution of 2 x 2 games (Mixed strategies)
9. Solution of rectangular games with mixed strategy

10. Some useful theorems

11. Symmetric games

12. Concept of dominance

13. Some theorems on dominance
14. General rules for dominance
15. Two-person, zero-sum, 2x n or n x 2 games

16. Graphical method of solution

17. Algebraic method of solution of a game problem

18. Transformation of a game problem to an L.P.P.

19. Fundamental theorem (Minimar theorem)

20. Solution of a game by simplex method

21. Summary of the methods of solution

22. Illustrative Examples

23. Iterative method for approximate solution
24. Solution of a game problem by matrix method

25. Illustrative Examples

 AME THEORY 3
TROBUCTHON

is which two boys A and B

S they have in their bags. It was simultaneously
ruation
so decided thatdraw
Consider a situa

which if theeither
number
one

by B be the same the number of balls drawn

as
by A, then the
fwO
b a l l s d r aw n wn
A wins a n d gets one rupee from B. On the other hand, if the numbers be not
wins and gets one rupee from A. This situation is
en B wir
h esame, t h e n
expressed by
ble above. The possible courses of action of the players B and A are
table

by the columns and the rows outside the table. The

irst

the
spectively positive
the table are the payments from B to A and the negative entries are the
w'n

sho
enties in

B
B
1 2 3
1 BALL 2 BALLS
1 BALL 1 -1 I2-34
A 234-5
A 2 BALLS -1
4-5 6
the situation in which two boys A and B
again
nayments from A to B. Consider
one of the three numbers
play a game. They simultaneously and independently write
numbers written by them be even, then A wins and B
1, 2, 3. If the sum of the
to this sum in rupees. If the sum be odd,
then B wins
pays A an amount equal
This situation is expressed
and A pays B an amount equal to the sum in rupees.
of possible actions taken by
by the second matrix above, in which the courses
table associated with A and B
A and B by rows and columns outside the
are shown
second course) and B writes 1 (his
respectively. For example, if A writes (his
2
A will pay B rupees 3 and this is
first course), then the sum (2 + 1) being odd, first column
intersection of the second row and
indicated by the element (-3) at the
associated with other entries
of the above matrix. Similar meanings are

B when A chooses the i-th

course

The cell entry a is the payment of A by

The negative entries indicate
course of action.
of action and B chooses his j-th
the payments from A to B. between two opponents are
referred to
of conflict' and competition
These situations may be persons, countries,
or simply a game. The opponents
as a competitive game opponents are assumed not
versus a single
opponent. The
nature decisions under conflict
caused by
organisations or
deals with making later
to collude. Game theory w a s originated by
J. V. Neumann in 1928 and
This theory
opposing interests.
B. Dantzig. m a d e at the end of the game
is called
developed by G. how payments
should be
pavof to each
indicates the
A table showing matrix a s s o c i a t e d with the game
matrix. The ose number is, in general finite in a game. A strategy
a puyoff whose
called players, the number of rows and columns of
opponent, taken by the players;
of action
of strategies of the players A and B
is the
course
to the number
available to him and the player B
correspond
matrix
the payoff player A has m strategies
If the the payoff for various strategy combinations
respectively.
to him, then
available

strategies
x n payoff
matrix. The game is finite, if the number of
has n an m
by be finite; otherwise, it is an infinite game. A
to each player
r e p r e s e n t

s t r a t e g i e s a v ailable
action quite ignorant of the strategy used
IS ila

ayer ch00Ses o
ses
one
ne of his courses or
 4 GAME THEORY
indicates the gains to A (whose strata
by his opponent. The payoff matrix
Written on the left of the matrix) for each possible outcome of the gama are
A loss
is considered as a negative gain. Every combination of courses of action .
by A and B determines an outcome indicated by the corresponding element
payoff matrix. which determines the set of gains to each player. Thus
Thus payoff theis
the quantitative measure of satisfaction of a player at the end of the
the ni
play. This
Outcome as represented by a single payoff number may be positive (gain),
loss) or zero (no gain no loss). The outcome of the game is affected by the ch negative
made by the players.
oice
A game. in which the gain of one player is a loss to another, is called a
Sum game. In a zero-sum game, there is a re-distribution of initial monev ero-
and t
does not add anything to the total wealth of the players. If the number
of
be confined to two and if the losses of one player be layers
equivalent to the gains of
another. the sum of the gains to the players being zero after the
game, then it
is called a 'two-person zero-sum
game' or a 'rectangular game (because of the
rectangular form of the payoff matrix). It is also called a finite matrix game.
A game involving
players is called a n-person game.
n
In
zero-sum game, the payoff matrix of the
a
player B is just the negative of
the payoff matrix written for the
player A; because in such games the gains of
one are the
negatives of the gains of the other. In the given payoff matrices, A is
the maximizing (row)
player and B is the minimizing (column) player. Generally,
the payoff matrix is written for the
this matrix represents
maximizing player; for the minimizing player
the losses.
The course of the
game is determined by the desire of A to maximize his
and that of B to restrict his loss to a gain
minimum. The optimum strategy maximizes
and guarantees a payoff to one
player and minimizes the loss of the other.
2STRATEGIES
A strategy as defined earlier is a set of
which a player
pre-determined rules or programmes by
will decide which of the available
courses of action he will have
to
adopt at each play. The strategy of a player determines a
particular move itself depends on everything particular
the player but that move of

before that particular move that has happened

adopted by the player. The
game problem may represent
a deterministic situation in which the
represent
object is to maximize the
gain or it may
probabilistic situation in which the object of the maximizing
some
player
will be to maximize his
the courses of action that
expected gain. In the former, each
player knows in advance
may be taken by the other
a
particular course and in the latter, a player and he will then choosc
course of action to be player kept guessing as to the possi
is
Thus a pure
adopted by the opponent.
strategy is a decision making rule in which one
of action is
selected, while a mixed particular coul
hich
strategy is
decision making rule
a
wu in
a
player decides, in advance, to choose his
probability distribution. In the mixed courses of action with some init
ueessing
strategy, one player is always Kep
about the
opponent's choice. A player may have a finite 5 ies
but he has infinite
number
number of pure
sua
of mixed
selection is made of all strategies. To be specific, in a
mixed du
the pure
strategies with some definite probabilnacd
 GAME THEORY
5
f aplayer.
pure strategies, then the set of real numbers {x. x,. .
carresponds to the j-th strategy) will be a mixed strategy for him,
pobabh
if

, 2 0 ,j = 1, 2. .
M.
and

If, 1
and x, = 0 for all j except j then the
=
r, player is said to play
pure strategy.
r-th pure strateg Thus pure strategy is a particular case of mixed strategy.

THE MAX AND MINIMAX CRITERION

Let us consider the following payoff matrix :
For the player B, strategy lI is not good as he always loses more if he chooses
this plan of action. Hence his best strategy is always to choose I and the worst
that can happen is that he will lose 2 when A chooses his III. Now consider the
decision for the player A. The most that he can get is when he chooses II and
B chooses his II. But this is ulikely to happen as B is
B
assumed to be no less intelligent than A. Thus the most
A can make (if B chooses I) is by choosing II, in which

2 2 case he is to get 2. Thus II is the best strategy for A and

1
Iis the best strategy for B. The result of these choices is
A
II- 4
that B pays 2 to the player A. This amount is called the
III2 3 value of the game. Thus the value of the game is defined
to be the maximum guaranteed gain to the maximizing
player or the minimum possible loss to the minimizing player when best strategies
are used by both the
players. Solution of the game is to find the best strategies
for both the players and to find the value of the game. This value of the game
is unique.
If the value of the game be zero, then we call the game a fair game.
The game problem is viewed in such a manner that each player is interested in
determining his optimal strategy. Because of the conflicting nature of the problem
involving opposite interests, this optimality is based on a rather pessimistic principle.
As a matter of fact, such strategies are selected by the players which guarantees a
payoff that can never be worsened by the choice of his opponent. This principle
is known as "maximin" or "minimax principle and is stated as follows
Ifa player lists the worst possible outcomes of all his potential strategies, then
he will choose that strategy to be the mosSt Suilable jor him which corresponds to

the best of these worst outcomes

Such a strategy is the optimal strategy of the player. An optimal solution is
said to be reached, if neither player inas it beneficial to alter his pure strategy

state the game is said

to be stable.
and in this
If we apply this principle to the earlier payoft matrix, then we see that the worst

possible outcome of the strategies of A (row minima) are -2) for A - ) for Au

Ar According
to the maximin principle, A should choose
and 2 for strategy II
in this his guaranteea gain 1s maximum and is 2, since
be
Max -2, -1, 2) = 2.
 GAME THEORY
account the worst possible outcomes of the
we take into
On the other hand. if maxima) to be 2 for B, and 4 for

strategies of B, then we find

them (column
1s interested
B
in minimii
Now B
B.
maximum losses for ng
These are the possible the minimax principle,
choose his strategy 1, by nce
his loss. Hence he should
Min 12. 4) = 2.
II but whero
A may gain 4 by his strategy
It should be noted that although this choice of A. On he
lI alongwith the
is the guarantee that B will choose strategy instead.
he himself gets I
choose 1 so that
contrary. B will then
Thus the optimal solution of the game
is (III, 1) and the value of the game is 2
The first figure in the parenthesis indicates
the optimal strategy of A and the second
of B.
ngure in the parenthesis refers to the optimal strategy
It, in a game, maximin for A and minimax for B be equal, then the game is
said to have a saddle (or equilibrium) point. The gain at the saddle point, which is
thus the position in the matrix where the maximum of the row minima (maximin)
coincides with the minimum of the column maxima (minimax), is called the value

of the game. Thus

max. (row min.) = min. (column max.) = value of the game.
Let us apply this principle to solve the game whose payoff matrix is given
below on the left.
B B

B B B B B B B B
A 7|34 A 34
A A564
S A A2 5 6
As7 2 o 3
Putting squares around the row-minimums and circling the column-maximums,
we get the matrix on the right above.
Obviolusly, the matrix has a saddle point at the cell (A, B,) and hence A, and
B are the optimal strategies of A and B respectively and the value of the game
is 4, which is the entry at the saddle point. Obviously, the strategies used by both
the players are pure strategies.
The value of the game to A is the element at the saddle point and its negative
is the value of the game to B.
f a matrix involves more than one saddle point, then there exist more than
one solution of the game. B
If a game has a saddle point, then
the pure strategies corresponding to
B B B ROW MIN

the saddle point are the best strategies. A 48

But there are games for which the
A A-8 5-7-8
payoff matrix will have no saddle
point. Consider, for exemple, the A 6-7 6-7
adjacent payoff matrix. COL MAX6 8
Here max. (row min.) = -4 and

min. (col. max.) shown in circles are not equal. Determination of the
as
column maximums and maximum of the row minimums being
minimum of the
to expect that we should have
entirely different processes, there is no reason
two
 GAME THEORY
addle point. As a matter of fact, in general, these two quantities are no
always a sa

the following payoff

yual as will
ent from
B
matrix
la,)mxn
a 12 14 . 1 . . . an

a21 22 2q . 2s a2

A p 2 pa .. s. n
**°**°*°

1 2 rq TS

ml m2 ******* ma ****** ms ****** Gmn

mg

Here the maximin for A is max [min a]

and minimax for B is min [max a.

j
Let max [min a,] = a and min max a = a,

Then a is the minimum element of the p-th row; hence ap ap aps being
an element of the p-th row and s-th column.
But a, has been assumed to be the maximum element of the s-th column; hence
psSays
Combining these two, we have
pqSs
Thus max [min a,l min [max a.
i j J
s minimax for B.
In general, for A
maximin v
then the game possesses a
f, in particular, maximin a
minimax
a
are the optimal strategies for A and B
solution for which -th and q-th strategies
or the payoff matrix will be a saddle
respectively and thus the (, q)-th position
value of the game.
point and a, will be the
is denoted by
The maximin value of
the game, Caleu ne ower value, v and
1s denoted by
the minimax value, called
the upper value,
Y = v =
determinable game,
Thus, for a strictly Y = 0 = v.

and for a fair game,

be
a non-strictly-
a
non-strictly-determined matrix game, if
determined matri
said to
The game is V.

aid t to DE biased. If v
be Diased. >> 00, tha
then it is said
0, then the game
Is sad < 0, then i it
to be
and il
#
avet and il
If vv <0,
v
player IS said to be biased to
the maximising the
biased to
minimizing player.
in the solution of game
very important
relations
are

stablish in the
next section a
problems and as
These
we
to
tryrems, u
under w h i c h these
which
necessary
relations will hold.
and sufficient
condition
such nder
theorems,

two
with
 GAME THEORY
8

4EXISTENCE OF SADDLE
POINT
valued function of x and y suh

Theorem 1. Let fx, y) be a real that

E B, A and B being two sets, E
x, y) is a real number for x E A and y
exist,
rther,
and min mar Ax, y)
f both mar min fx. v) vEB xEA
EA EB

Then mar min fx. v) S min mar X, y).

EA YEB
VEB rE A

and y, we have, by the definition of maximum,

For some fixed x

Ax. y) s max Ar. y) (1)

xEA
and by the definition of minimum, we have
min f (, y) S fx. y).
(2)
yEB
From (1) and (2). we have
min A y)
y¬ B
s
max fx. y).
xEA
(3)
Now, as we get min fx. y), this becomes independent of y, since fr, y) has
yE B
got its minimum value for some fixed value of y. Hence we can write
min fx y) s min max fx, y).
y¬B yEBx¬ A
4)
Again, the right hand member of (4) is independent of x; hence we say that
max min fx, y) min max
s fx, y).
xEA yEB yEB xE A
This proves the theorem.
Cor. The matrix
[a,mcan be regarded as a real valued function of the
variables i and j, such that
i . j) = a i = 1, 2 , . . m, j = 1, 2, n.

Further, if A and B be sets of the first and

then by the theorem, we have
m n
positive integers respectively,
max min aj S min max a
J j i

Analytical definition of the saddle point

Let fx, y) be
real valued function such that
a
and y B, where A and B are two
E fx, y) is defined for x E A
bEB is called a saddle sets; then a point (a,
b) where a E A and
point of the function fx. y), if the
be satisfied following conditions
i) fx, b) s ffa, b), for all x E A
(ii) ffa, b) s fa, y), for all
y e B.
Theorem 2. Let fx, y) be a
real valued
is real number for
a
x E A and
function of x and y such that Je
that both max min y e B, A and B being two sets.
flx, y) and min Further, suppoand
xEA yEB max fx, y) exist; then the
sufficient condition that
yE B xe A necessuy
max min fx, y) =
min max fx,
xEA y¬B y)
is that, the yEB xE A
function flx, y) possesses a saddle point.
 ME THEORY 9
maximum
min fr, y)
a
eA make
yEB
max fG, y) a minimum.
hE B make
xEA
n il e t

can
write
We
Thus min Aa y) = max min fix, y),
(1)
yEB xE A y¬ B
max fx, b) = min max fx, y). *

(2)
xEA yEB xEA

Assuming
the right hand side of (1) and (2) to be equal, we have
min fa, y) = max fx, b). (3
yEB xEA

definition of minimum, we have

From the
min Aa y) s fa, b).
y¬B
Hence, by (3), we have max fx, b) s fa, b),
xEA
be put, by the definition of maximum, as
which can
A.
fx, b) s fa, b), for all x e

function flx, y) to have a saddle point

This is one of the conditions for the
condition can similarly be deduced.
at (a, b). The other
Thus the given condition is necessary.
(a, that b) is a saddle
also suficient, we suppose
To prove that this condition is E B, we have, by definition
x E A and y
of the function fx, y); then, for
point
of a saddle point, (4)
fx. b) s fa b)
y).
**
(5)
Na b) s a
and as
and (5) can be written respectively
Now (4) min fa, y),
max fx b) s fa, b) and fia, b) yEB
cEA
conclude that
Irom which we can
s min fa y).
max fx, b) s fa, b) yeB
A
S max fr, b)
since min
max fx, y) xEA
NOw,
yEB xE A
max y¬ B min fr, y),
and min fla, y) xEA

yEB
b) s max min Ax, y).
we can write
max f , y) Aa, xEA ye B
min
xE A
B
yE
have
we
theorem 1,
Also
from
min max Ar, y).
min lx, y) yE B xEEA
max
B
xE A yE

(a.
if (a, b) be
b) a saddle point of flr. y), then
hat if
that
assert
max Ax, y) = Aa, b).
These
two
Ax, y)
=
mi
min
ax yEB E A
xEA yE B
 GAME THEORY
10
x n matrix, then the necessary and sufficiient
Cor. If A =la,.l. an m

condition that
max min a, = min max o

1s that. [a,) possesses a saddle pont; for, as shown carlier, this matrix may
be
real valued function f (i, j) such that
regarded as a

i. ) =

where i = ******* 1 = I. 2. .. (first m and n positive integers).

1LLUSTRATIVE EXAMPLES
Ex. 1Sove the foliowing game whose payoff matrix is given by
B
I

-53 20
A II
55 4 6

-4-2 0-5 Roorki, 19731

Show that the game is strictly determinable.
Let A be the maximizing player and B be the minimizing player.
B Put squares around the row minimums
II IV
and circles around the column maximums.
Then we the adjacent matrix. Now
5 20
seen that the element 4 of the second
is get it
4 II row
6 and third column has been marked around
III -4 2 -2 03 both by a square and by a circle. Then this
position is the saddle point of the payoff
matrix and the pure strategies II for A and M for B are
the optimal solution of
the game which is written as
(11, I). Thus 4 is the value of the game as
= 4 =V,
since y Max {-5, 4, -5}
=
4 and += Min {5, 5, 4,
20} = 4.
=

Here y 4 V. Hence the is strictly

=
game determinable.
Ex. 2. Show that whatever may be the value
of a, the game with the following
payoff matrix is strictBy determinable
B

37
3a
We first
ignore a. The piven payoff matrix ignoring a) gives the following
row
minimums and column maximums
B
Row Min

A
3 a

Col. Max. 3 7
 GAME THEORY
11
and i = 3, whatever may be the value of a.

is strictly determinable and v = 3.

The game
best strategies for A and B are (1, I), whatever may be the value of a.
The

For what value of a, the game with the following payoff matrix is
Trctly
determinable?

B
1 II
5 2
A I1a -8
III

We ignore a first and determine y and of the matrix by computing the rOw
minimums and the column maximums.
For this, we have the matrix below.
B
I II III ROWMIN
2 2
a5
AII -1 a -8 -8
-2
II23 a

COL. MAX -1 5 2

2 and minimax for B = -

1.
We have maximin for A =

= 2 and + = -1.
for B it is I for these values.
The strategy for A is II and
should be such that
with its value v, then v
If the game is to be determinable
-1Sv S 2.
-1 sa s 2.
Thus, for a strictly determinable game,
determined,
Show that the
2 x 2 game a b] is non-strictly
Ex. 4. b andd <c
asb, a < c, d <

therefore the row minimums

of the two rows are

< b and d C,
<
Since a

and d. Now
respectively a
= max (a, d). (1)
(row min)
Y = max
Uncielore ne column maximums of the two
< 0,
< c and a
since a
Again, and b. Now
c
are respectively
columns
= min (c, b). (2)
min (col.max)
V =

.
show that y *
Relations () and (2)
determined.
non-strictly
the game is
Hence
12 GAME THEORY
EXAMPLESA
problems (1 -2)
Set up the matrix for the following game
two players A and , the player A has o n
01 In a card game betweenwhile B has a black card 86, a red card 6r andred
card 8r and a black card 6b and a
red card 2r. The players expose one of their cards simultaneously. If the
cards
match in colour, then A gets trom B the positive difference between the numi
mbers
on the cards. If the cards do not match in colour, then B gets from A the nosit
e
difference between the numbers on the cards.
B
Ans 8b 6r 2r

8r
o2 6
65 4

0 Each of piayers A and B shows from one to five fingers. If the sum of the
fingers shown be exactly divisible by three, then there is exchange of any payoff,
But if there be a remainder of one then the player A gets from B an amount
of
rupees equai to the double of the fingers shown, while if the remainder be two
then B gets from A the sum of the fingers shown.
Ans B

2 5

0
-5 0
2 8 -5 14
A 3 5 0 4 -8

0 14
-8
0 14 - 8 0 20

03 Appiy the maximin and minimax

principleto solve the games whose
matrices are given below. Verify that (i) is biased towards A and
payoff
()
(i) is a fair game.
B () B
B, B2 Ba B B B
A. 15
2 3
A -12 8
5 7 A A 7 5
4 0
As 6 0 12
Anss () A B), V= 5.
(i) (A. B), V = 0.]
Solve the games
(4 6) by saddle point method
04 () B
(i) B
B, B
B B2 B

A A 1 2
A-5 2 6

A 8
A 6 4
A 2226 A,4 0 -3
Ans () A, B,) or
(Ag, B), v =
2.
(i) (Az, B). v = 4.]
 ME THEORY
B 06
13
B
B B B8 B. B B
73 B5
B

A A
2 17-
4
A A5 1 -4
|7 2 03
2
-7 -6 6
As
4.]
A A 323 4
-6
2

Anss (Ap B), 1-1

V =

As 0 0 60 0

Ans (A B), v = 2.]

Show that the following payoff matrices have no saddle point

B
(i) B
B B B BB B B
A, 1 3 6 A32
A A213 A
A 42 4

A 21
Delhi, 1977] A 04 0 8
In each case, show that min max> max min.

U8 Show that the following payoff matrices involve more than one saddle point.
Find the solution of the game and the value of the game.

) B i) B
B B2 B B 8, B BA
A -215 -2 A 42
A Az -6-6 - 4 A
A

A-5 18-8 Ay 2

Ans () A, B,), (A, B): V = -2 A 4 o0

B); 2.] [Calcutta Hons., 1990, 1993, 2002
(i) (A, B,). A v =

two-person zero-sum game is

09 The payoff matrix of a

B (i) B
(0)
B B B IV

A 21 9 2| 1 8 0
5 4 6 7
A Aa A

Ag 1 3 -2 4 3
IV

termine the number of saddle points and the corresponding optimal solutions.
5
62 2

1 Madurai, 1980)
(A,. B). (A. B);
V =

Two;
Ans ) 4.]
One; (1, ); v =

(i)

LPGT-F25
14 GAME THEORY
matrix
10 Consider the game Gwith the following payoffwhatever B

determinable,
) Show that G is strictly
0 ) may be.
A 2 7
Detemine the value of G.
(i) A1
Anss () 2]
the game with the following payoff matrix is strin
|iFor what value of , strictly
determinable?
(i) B
)
2 73
-1 | A-2 -8
-2 4 -34 :

Anss ) 1 s s 2. (i)-2 s s 3.]

J 1 (a) Find the range of values of p and q which will render the entry (2, 2)
a saddle point for the game
B B

2 45 2 6
10 7 9 A P 5 10
4 P 6
63 4
[If the cell (2, 2) be the saddle point,
then 7 and V = 7. Thus q 7 and p s 7]
Anss )7 s q and p s 7. (i) p2 5 and qs 5.]

(b) Find the range of values of p and q so that 2 is the value of the following
game

2 3
2P4
Ans p 2 5 and q 5.]
J13 In a game with the 2 x 2 payoff matrix b
a

where a <
d <b<, show that there is no saddle point. b
Show
and d>
further
C.
that the game is non-strictly determined, if a b , a > G d
in a
14 (a) Show that in a game with the 2 x 2 payoff matrix if two eleme
row or column be equal, then the game has a saddle point.
along
(b) In a 2 x 2 game if the largest and the second largest elemenis
a diagonal, then show that the game has no saddle point.
 GAME THEORY 15
Player
choose
A can choose his strategies from (A, Az A only, while player
ose from the set (B,, B} only. The rules of the
OOse le player B
be made in accordance with the selection of game state that
payments
Strategy selected
strategies
Payments to be made
(A. B,) A pays Re. 1 to B
(A. B) B pays Rs. 6 to A
(A B,) B pays Rs. 2 to A
(Ag B) B pays Rs. 4 to A
A B) A pays Rs. 2 to B
A B) A pays Rs. 6 to B
What strategies should the players take in the above
game in order to get the
naximum benefit?
Ans A B). v = 2]

A Az Ag are the pure strategies of the player A and

{B,, B, B are
nose of B. Payments are made according to the following table in accordance
ih the strategies chosen. What strategies the players should choose to
maximum benefit?
the get
Strategy selected Payments to be made
(A. B) B pays Rs. 3 to A
(A B) A pays Rs. 2 to B
(A, B) B pays Rs. 4 to A
(Ap B,) A pays Re. 1 to B
A B) B pays Rs. 4 to A
(Az B) B pays Rs. 2 to A

(4 B,) B pays Rs. 2 to A

B pays Rs. 2 to A
(A B)
B pays Rs. 6 to A
(A B)
Anss (A B, or B). V = 2.]

POINT
GAMES WITHOUT ASADDLE
We have seen earlier that there are games having no saddle point. In such
games, the principle of saddle point solution (maximin and minimax) breaks down
and the players do not have a single best plan as neir best strategy. Each player

can improve his payoff by selecting a arerent Straegy. The game in this case
is said to be unstable. In such cases, cac player nas to take recourse to some

maximize or nis gain to minimize his loss. Each nlavr

mixed strategy in order to manner so that his final pavoff OVer
random
would like to play in a
will be opumal. lt may so happen that ha.
number of plays of his game
individual play
of the game but the overall effect will be optimal for
more in any the indi
individual
vidual strategies in a random manner is to
to mix the
him. The simplest way recourse to chance
B takke mechanism, by introducing
rahabilities with choice and mathematical expectation with
proba
C C2
payoff.
6
make clear
R To a
understanding of the chance
procedure
A aS to obtain an
optimum mixture of probabilities for
R2 1et us consider the adjacent payotf matrix of a game.
strategies, let us
16 GAMETHEOAY
of A and B respectivelv
be the pure strategies
Let (R. R,) and (C C)
the time and R, half the time. If now B cho
Let A decide to play R, half
for A will be
C for all the time. then the expected gain
-2)+ (5)
-1 1.5.
=
+ 2.5
all the time, then A's expected gain will be
But. if B chooses C,
3 +.5 3.5.
(6)+ (1) =

On the other hand. if B also chooses a mixed strategy by choosing C, and

and C
time, then ' s expected gain will be
at random each for half of the
1.75 2.50.
(1.5) + (3.5) .75 = + =

In a similar manner, we can compute what A's expected gain will be for other
choice of mixed strategies. For example, A plays R one-fourth the time and R
three-fourth the time while B plays two-third time C and one-third time C. In
that case, the expected gain for A will be

Here the question is: what will be the best mixed strategies for the players
Our object will be to see if A can be sure of some minimum gain for him and B
can be sure that he will not have to lose more than a maximum amount. Under
such situation, we shall find out he gain for A.
Now, if A plays R, with probability x and R, with probability (1 - x) and if
alongwith it B plays C, all the time, then the expected gain of A is
-2) +(1 - x) 5 = 5 7x.
But, if B plays C, all the time, then the expected gain of A is
x(6) + (1 - x) 1 = 1 + 5x.

It will be shown later that if A chooses x such that these two expected gains
are equal, then this will lead to the best
strategy.
Thus, for the optimal solution, 5 -7x = 1 + 5x
or, x = and hence 1 - x =

Thus gain 8 (A) for A will be given by g(A) =

-2) +(5) =
Regardless of the probability with which B plays either his C, or C, strateg
A's gain will be For example, if B plays C, with probabilityand C, with

probability then the expected gain for A expressed symbolically gA. as

will be given by

t() #(3)-
Similar result will be obtained for
any other choice of probabilities oy
Thus, by the choice of
be
probabilityand
for R, and R, respectively,
can

sure of a gain
The same consideration can be made for the
player B.
 GAME THEORY
17
nrobability of choice of C, be denoted by y and that of
Let
the
C, by (1 -

y).
Expectedigain (loss)
gain (los of B for A choosing R, and R, are
respectively
-2)+(1 - y) 6 = 6 - 8y

y5)+(1 - y) 1 = 1 + 4y.
and
Then, for the best strategy of B, these are equal, that is,
6 8 y = 1 + 4y

Or.
y 2 and hence - y =

Thusst8) -2) + (6) =

Itshould be noted that g(A) = g(B), as is expected in a zero-sum game.
The optimal solution of the game is thus
A should play R, and R, with probabilities and B should play C, and C
withprobabilities and the value of the game is thus

7 FURTHER CONSIDERATION ON MIXED STRATEGIES

Consider the rectangular game whose payoff is given
B
by the adjacent matrix. I
It has no saddle point.
and (1 - x) be the probabilities with which A
2 6
A
chooses the strategies I and II. Then the expectation of A 51
for the choice of strategy I by B is
-2x+ 5 (1 - x) = 5 - 7x (1)
and for the choice of strategy II by B, the expectation of A is
6x + 1 (1 - x) = 1 + 5x. (2)
In (1), if x > then 5 7x <
and in (2), if x < then 1 Sx <
expectation, A will choose his I and II
Thus it appears that to have an optimal
strategies with probabilitiesandrespectively.

and (1 y) -
to be the probabilities of his choice
Similarly, for B, assuming y for the choice of strategy I of A, the
of the strategies I and II respectively,
expectation of B is
+6 (1
-

y) = 6 -

8y ..(3)
-2y
of B is
II of A, the expectation
and sor the choice of strategy
5y+I (1 -y) =l+ 4y.
then 6
-

8y>
Now, from (3), if y <

then 4y>
and from (4), if y >

stralegies will be and for I and II

Hence, as regards B, the optimal
respectively.

probiem in a slightly different way,

We may consider the
same
18 GAME THEORY
Let A play his strategies I and ll with probabilities x and (1 - x) and B nlau

his strategies I and II with probabilities y and (1 - y) respectively. Then the

mathematical expectation of A is
-21y+ 6(1 -y) x + 5(1 x) y+1 (1 - x) (1 - y)

-1215x + 4y + 1 = -12
(x - -
This shows that if A chooses x = then he can ensure that his expectation
will be at leastMoreover, he cannot be sure of more than for, by choosing
= B can ensure that the expectation of A will be exactlyand not more
than that. Thus A might play x =and be sure to get similarly B might as
well be satisfied by getting ) and play y = s o as to get it. Thus the
optimal probability for A to play his I and II will be and and the optimal
probability for B to play his I and II will beandthe value of the game
obviously will be
Hence the optimal solution of the game is for A. ( s and for B. (
and the value of the game is

8SOLUTION OF 22 GAMES MIKED STRATEGIES

Consider the game whose payoff is given by the adjacent matrix and for which
there is no saddle point. We are to determine the probabilities with which each
strategy of A and B will be played to get an optimal
B
solution.
Let
C C2
xand
y, be the probabilities with which A
chooses his i-th strategy and B chooses his j-th strategy R
respectively. Then, for this problem, the mixed strategies AR. a
of A and B are respectively,
X = (1, *) and Y = O1. y,) such that

+21,
+ 2 = 1| (1)
and
X220, y y2 2 0.
The expected gains for A when B chooses C, or C, are respectively
4 ta212 and az¥ + azza:
Similarly, the expected loss to B for the selections R, and R, by A are respectively
a1 +a12 and a1 + a22
Assuming that each rectangular game has a solution and assuming v to be the
expected value of the gane considered, we should have
a t21¥22 v,
(2)
as A expects to get at least v.
 GAME TH
R expects to lose at most 19
Similarly
v and hence
a1+a12 v,
a21+a S v.
(3)
ming the
ssuming above
the above
inequations as strict equations and assuming the existence
o f their s o l u lutions, we can say that the solutions ( x) and
s o l u t i o n s
for the game. ( y) are the
optimal
en (2) and (3) considered as equations together with
equations (1) will give
fay Now, y2 if these
satisfy the non-negativity restrictions given in
(1).
this
then this will be the optimal solution; otherwise, we shall have to take recourse
ta algebraic method for the solution of the game problem.
Considering the inequations (2) as strict equations, we get

a1+a
which gives (4)
12
Similarly, from the inequations (3), we have

2
(5)
From the equation (4) and the first equation of (1), we get

a
a221
+ a2 - (a12 + G21)

1 12 (6)
and
*2a +az (412 + a21
the second equation of (1), we get
Similarly, from the equation (5) and
a22 12
a + a22 - (a12 +a21
(7)
and a+az2 (G12 +a21)
as
value v of the game can be easily computed
Then the expected
122ai2421_ (8)
v a+22 a2+21)
non-negativity
requirements of x, y. * y
the
Now one is to verify Hence the largest and the
a saddle point.
does not possess of its diagonals.
matrix constitute one
The payoff the payor must matrix
elements of matrix will be as follows
second largest e l e m e n t s of the payoff
of the
orderings
Hence the possible

2a222 a 2 2
a12
a

2 a21
2 412
a22 a above are all positive for all
such
as oblained

values of x, X2 Y1 2 solution.
The Hence this is the optimal
orderings
of a . a12 21 22
 20 GAME THEORY
whose payoff matrix is given below and wh:.
ich
Consider. for example, the game
has no saddle point. we have
their usual meanings,
y , V, and v having
B
A
A 7- and v = 35= 23 20
Note. Incorect result may be obtained if these formulae for X. Xz Yi» Y2 and v be applied to a
2x2 game with saddle point.
9SOLUTION OF RECTANGULAR GAMES WITH MIXED STRATEGY
In a game problem, in which the payoff matrix has no saddle point, the plavers
are to play their strategies according to some probability distribution for each of
their pure strategies.
Consider a game with the payoff matrix [ a K SUch that the maximizing player
A has m pure strategies and the minimizing player B has n pure strategies. Let the
probabilities with which A and B select their pure strategies be given by
X (1 . m X 2 0, for all i
=
2 .. X
and Y =1 2 . n 20, for all j
which are called mixed strategies for A and B respectively. The set
S = (X = (1.
., 2 0, for all i and Ex, =1)}
is called the set of mixed strategies for the player A and the set
S = {Y = 1, y2» .y,)y 2 0, for all j and E », = 1}
is called the set of mixed strategies for the player B. Therefore A and B select their
i-th and j-th pure strategies A, and B, with probabilities x, and y, respectively. Here
XtX + (1)
and +2 + (2)
A pure strategy can be regarded as a special case of mixed strategy in which a
particular course of action is assigned with probability unity and all other probabilities
zero. Thus the m pure strategies of A are given by (i = 1, 2, . . . , m)
A, = (0, 0, ********
0, 1, 0, .., 0), 1 at the i-th place and the n pure
strategies o f B are given by (j = 1, 2, n)
B = (0, 0, O, 1, 0, . 0), I at the j-th place.
Now, if a, represents the (i, j-th entry of the game matrix, then x, and y, will
appear as follows
B
12
21 a22
A
2 a,
 GAME THEORY 21
is the payoff to player A when players A and B use the pure strategies
Here 4, is
Hee
and B, spectively. Then the expected payoff to player A, given that player B
r eure strategy B, is
s e sh i s
m
E (X, B)= i = 1
a,
expected payoff to player A (called the payoff function to A) when the
The
the strategy Y is given by
B
uses
ver
E (X. Y) = 2 , E X. B) = 4,4) (3)
The mixed strategy
problem is solved by applying the minimax criterion in which
maximize the smallest
he maximizing
the
player chooses the probabilities x which
and the minimizing player chooses his probabilities y,
nected gains in columns
gains in rows
which minimize the largest expected
Let us denote max min 2 2 4),
for A by
i j
and min max 2 2j 4 for B by v
i
so that = max [min {2 4 2 42 (4)
(5)
and min (max { a} a
These values are referred to as the maximin and the minimax expected payoffs
respectively.
case of pure strategy. A pair
of strategies (X°, Y),
In general, y =V, as in the
function
is called the saddle point of expectation
(X°ES Ye S,) for which y
=
chosen that this relation holds as pure equality,
E (X, Y). When x, and y, are of the problem and the expected value so obtained
so
then we get the optimal solution
value of the game. f X° and Y° be the optimal
will be equal to v, the expected
with the
players, each payoff element a will be associated
solutions for the
two
value of the game is
(x y) optimal expected
and the
probability
2 44
ECX,
P), X°eS Pes,
i = 1 j
add a fxed number
P to
each element of the payoff matrix,
Theorem. If we
uncnangea while the value of the game is
remain
then the optimal strategies
increased by P. to cach element of the payoff matrix, it
fixed quantity
On addition of a
becomes (a, + P).
and the new payor Tunctions by E, and E, respectively,
If we denote the original
then
and E, = (a, + P) xy
E, =2 a,y,
E, P, E, 1.
Hence E, =
2 ay +2 2 A
= + as
x, = =
J
22 GAME THEORY
element of the matrix does not cha.
Thus the addition of P to each
an be
hange the
nafure of the payoff function and
the new payoff function obtained from
the oniginal payoff function by simply adding the constant P to it.
it. Hence,
for
two matrix games with payoff matrices layand b where b = a, + p
the
optimal strategies for both the games are the same, only the value of the ao.
game
is increased by P.
The constant P does not play any role in the selection of an optimal strato
for the game with payoff matrix [b]. Hence, if P be suitably chosen, then
he
value of the game with matrix [b,) can be made positive. If P be chosen so large
that a + P> 0 for all i and j, then he value of the game will be positive,
Note. The optimal strategy remains the same even when each element of the payoff matriv
is
multiplied by a constant P. while the value of the game becomes P times the original value.
1O SOME USEFUL THEOREMS
Consider a finite matrix game given by the payoff matrix (a)mxn whose elements
a, are real numbers. The probability X = (%1, X is the mixed strategy
for player A and the probability Y 01. y = yn is the mixed strategy
**9
for player B. The m pure strategies A, i = 1, 2, .. m for A are
given by
..,
A, = (0, 0, 0,1, 0, . 0), 1 in the i-th position and the n pure strategies
B .j = 1,2, . , n for B are given by B, = (0, 0, .. 0, 1, 0, .
0),
1 in the j-th position. The payoff function E(X, Y is given by
where S and S, are the sets of mixed strategies for player A and B respectively
Let us study the necessary and sufficient condition for a given strategy to be
optimal in the above game.
Theorem 1. The necessary and sufficient condition for
X°eS"m to be an optimal
Strategy for the maximizing player A of a game with value v is
vSE X, all Y. for YeS,
Similarly, the necessary and suficient condition for YreS, to be an
optima
strategy for the minimizing player B of a game with value v is
E (X, Y) s v. for all
XeS
(We prove the first part of the theorem. The proof of the second part is similar.)
If X° be an optimal strategy for A, then there exists a reS, such that (X, r)
is a saddle point of the function E (X,
Y) and if v be the value of the game, then
V
=E (X°, r) s E (X°, ), for all YeS
Thus the condition is
necessary
To prove that this condition is also
sufficient, we assume
vSE X, Y, for all (1)
YES
We assume further that every matrix game has a solution which is (X, r) "
the present case. Hence this must be a saddle point of E(X, Y).
 ME THEDRY 23

Therefore
E (X. Y) SE (X, Y) s E (X, . (2)
V = E (X, Y). (3)
Rurthermore,
Fu
Y Y' in (1), we have from (1) and (2),
Putting
E (X, Y) s E (X, Y). (4)
atting X = X° in (2), we have
Again,
E (X, Y) SE (X, Y). (5)
and (5), we conclude that
Comparing (4)
V = E (X, Y) = E (X°, Y). (6)
Hence, from (2), (4) and (1), we have
E (X, Y) sE (X°, Y) s E (X°, )
and all YES
for all XES
Thus (X, Y) is a saddle (X, Y). point of E
Therefore X°ES is an optimal strategy for the maximizing player.
function of xn matrix game with value v, then
Cor. If E be the payoff an m

to be an optimal strategy for

A is
condition for X°E S
the necessary and suficient
vsE X, B), j = 1,2,
sufficient condition for Pes, to be an optimal
Similarly, the necessary and
strategy for B is
i = 1, 2, m.
E (A, P) S ,

These follow from the

theorem 1 by puting X A, and Y B, =
=

value of
condition for v to be the
Theorem 2. The necessary and sufficient
and Es, to be the optimal strategies
for
an m x n matrix game and x*ES
is
players A and B respectively
Sv sE(X, B),
E A, P) m; J
=
1, 2, n.
i = 1, 2, theorem.
from the corollary of the previous
the condition follows
The necessity of assume that (X, P) satisfy
the relation

the sufficiency, we
To prove
P) SvsEX", B).
E (A,
Then we have

,E A, P) s v
(2x)
E (X, Y)
=

all XeS (1)

s , for
Y)
that is, E (X,
yE, B)2
v
(Ey)
Again
E (X", n
=
2,
V, for all YeS, (2)
2
that is, E (X, Y)
and Y = in (2), we get
X in (1)
Putting
X = (3)
E (X°, P).
V =

(3), we have
(2) and
from (1),
Hence,
E (X, Y) s E
(X°, P) s E X°, Y).
hence X° and are the
saddle pOint of E (X, Y and
Thus (X°, Y) is a
A and B respectively.
stralegies lor
optimal
 GAME THEORY
24 each player in an mx
x .n
of optimal strategies for
matrix
Theorem 3. The set

game isS a conv'ex ser. the two optimal

Let v be the value of the game and X, Xz° be
rategies for
stras

have
the player A. Then we sS
v and E (X,°,
Y 2 v, for all Ye
E (X°. Y 2
A S 1, we have
Thus, for each Yand
0

ij

= AE (X°., Y) + (1 - ) E (X°, y

2 Av + (1 -

A) v = v.

A) X,°) is also an optimal strategy

for the player A
Hence (AX°+ (1
-

for the player A is a convex set.

Thus the set of optimal strategies
The proof for the player B is similar
hence the above convex set is closed.
Note. E (X. Y is a continuous function of X and

TRIG GAM
if the payoff matrix remains unchanged On
A game is said to be symmetric,
the two players. Hence the symmetric games will
interchanging the positions of
have skew-symmetric matrices as their payoff matrix. These are square matrices

in which a,, =
-aj and 0, since the gain of one player is loss to the other,
a =

in which both players have

We give a typical example of a symmetric game
known as game of Morra.
the same set of strategies and is
one or two fingers and at the same time
Two players simultaneously show either
the number announced by one of the players
each player announces a number. If
fingers the by both
shown players,
then he
be the same as the total number of
wins that number from his opponent. (Notice that if both players guess right, then
the net payment to each is zero).
It is obvious that four possible strategies are open to each player. A player may

show and guess the total to be

1 finger 2 or 3
or 2 fingers 3 or 4
and guessing four can never win and
[Other possibilities like showing one

hence are discarded].

Let sand 1denote the strategies of "showing i fingers and guessing the
number j" for the players A and B respectively.
Then the payoff matrix is
B
12 13 23 24
S202-3 0
A
2 2 oo 3
S23 3 0 -4

S24 o-34
 HEORY 25
I f A plays
S23 and B plays t24 then A has shown 2
iving a total of four, which was the fingers and B has shown
fingers, number announced by B (but not
that B
R wins 44 froOn
Wins from A as shown in the matrix. A)
so
Note that the payoff is a skew-symmetric matrix in this case.
A symmetric game is a fair
Theorem
game and the two
art identical.
optimal strategies
matrix game, the payoff function for the
For a
maximizing player is given by
E (X, Y =
2ij
where X =
( X . and r =
0, V2 ..yp are the mixed strategies
of A and B respectively and lamxn is the payoff matrix.
In the symmetric game, the payoff matrix is skew-symmetric and the players
have identical strategies. In this case (X ), it can be shown that the expected
payoff E (X, X) 0. =
(1)
If X and be the optimal strategies for A and B respectively, then
E (X°, Y°) = v = value of the game.
Now we have
V = max min E (X, ) = min E (X°, Y).
X Y

E (X, Y)2 v.

Putting Y (X°, X*) 2 v.

X°, we have E
But, from (1), E (X°, x°) =0; therefore v 0. (2)
V = min max E (X, Y) = max E (X, Y).
Again Y X

. E (X, r) [ v.

Putting X =Y, we have E (Y, r) s v.

But, from (1), E (Y, r)

=
0; therefore v 2 0. 3)
Thus, from (2) and (3), we have
V = 0

therefore X° = P
and
this identical.
optimal strategies in
case are
and thus the
MINANG
12 CONCRPT OP DO ...,a,) dominates (or strictly dominates) the n-tuple
An n-tuple a =
(a, a2
>
b) for all i 1, 2, ..., 1.
b, (or a,
=
z
B (b. b, . b,) it a,
action (pure strategy) player of one be
In
course or
problems, if
one
game Tor all possible courses of action of
better for him
than or as go0a as anOner,
dominate the second. The dominated (or
then the frst 1S Sala to
the opponent, aiscarded from the payoff matrix, since it
of action can simpiy De
inferior) course
ne, the optimal strategies for the reduced matrix
When this is done,
off
is o no value.
the original matrix with zero probability for the discarded
also optimal for
are
there is n o
saddle point in a payoff matrix, then we can reduce
strategies.
When
size of the game by this way, (which is called dominance), before we set to
the
problem.
solve the game
26 GAME THEQRY
payoff matrix
Consider, for example, the game with
B

2 7 2
A
5 26
6 37
lt is seen that the second row is inferior to the third row for the maximizino
player A. This means that we shall have to delete the second row and find the
Zing
optimal strategy for the reduced matrix
B

A
6 3 7
In this matrix, the minimizing player B will prefer the pure strategy given by
the first column to the pure strategy given by the third column, because his object
will be to lose less. Consequently, the reduced matrix becomes (after discarding
the inferior column)
B

A
27
6 3
Now, if ( x°) and y°) be the optimal strategies for this reduced
matrix, then (x, 0, x°) and (y. y' 0) are the optimal strategies for the original
payoff matrix.
It should be noted that for the row player the dominated row is inferior and
is to be discarded and for the column player the dominating column is inferior
and is to be discarded.

13 SOME THEOREMS ON DOMINANC

Theorem 1. If the i-th row of the payoff matrix of an m x n rectangular game
be dominated by its r-th row, then the deletion of the i-th row from the matrix
does not change the set of optimal strategies of the row player (maximizing player).
Let A and B be the row and column players respectively of a
rectangular game
whose payoff matrix is given by the matiix
lamxn
We are given that
a a , for j = 1, 2, n and i # r
(1)
and for at least one j, a,, t
a,
Let y2 Y b e an optimal mixed strategy for the player B.
Then we have from (1),

that is, E (A, Y) < E A, Y),

where A, and A, are the i-th and r-th
pure strategies of A.
 HEORY 27
we have, if v be the value of the game
this inequality,
v2 EA, ) > E (A, r).
From

2)
x)be an optimal strategy for the row
playerA.
Let X° ( , t ° .
= *******"

from we have
>0. Then, (2),
f nossible, let x
E A, P).

Now (X, r) being the optimal solution of the game problem, we have
m
v= E (X, r)= , E (A, )

E(A P) +
x E (A P
k#i

v.

ki
This is a contradiction to our assumption that x° > 0 and hence

0.
Thus deletion of the i-th row does not influence the optimal solution.
Theorem 2. If the j-th column of the payoff matrix of an m xn rectangular game
dominates its k-th column, then the deletion of the j-th column from
the matrix does
not change the set of optimal strategies of the column player (minimizing player).
The proof of this theorem is similar to that of theorem 1.
Theorem 3. If the i-th row of the payoff matrix of an m x n rectangular game
be strictly dominated by a convex combination of the other rows of the matrix,
then the deletion of the i-th row from the matrix does not affect the set of optimal
column of the matrix strictly dominates a
Strategies for the row player. If the j-th
convex combination of other columns,
then the deletion of the j-th column of the
matrix does not affect the optimal strategies of the column player.
Let A and B be the row and column players of an m x n rectangular game given
the given condition, we have scalars (probabilities)
by the payoff matrix lamxn By
Such that 0 [ < 1, X; = 0 and 2 x, = 1 such thatt
x X

k #

or, since =0, a , (1

holds for at least one j.
in which strict inequality
Let = (y°, y2°» nDe an oplimal strategy for the column player B.
have
Then, from (1),
we

or,
** (2)
where A. is the 1-th pure Suraegy Of A and v is the value of the gamne.
28 GAME THEORY
Let X =
(°, * x ) be an optimal strategy of the row plaver
A. If
* *

possible. let x,° * 0.

From (2). we have ,° E (A, YP) < x,°v, since x * 0.

Thus we have

= E (N". Y) =

j= 1 k = 1 , "y
=
E (A,. r)+2,1 4 k =

k i

=E A, P)+ E A, y)
m

k =

This is a contradiction to our assumption that x° # 0. Thus x = 0.

This proves that the deletion of the i-th row does not affect the optimal solution
of the row player.
The second part of the theorem can similarly be established.
Theorem 4. R, and Rz are the sub-sets of the rows and C, and C are the
Sub-sers of the columns of the payoff matrix lapmxn of a matrix game.
(1) 1f a convex combination of the rows in Ri strictly dominates a convex
combination of the then there exists a row in
rows in R, R which, if discarded,
does not change the optimal strategies of the row player.
() If a convex combination of the columns in Ci strictly dominates a comex
combination of the columns then there exists a column in C,
in C, which, if
discarded, does not chanige the optimal strategies of the column player.
Let I, and I, be the sets of indices of the rows in R and R, respectively. Then
we are given that there are scalars , and u, such that

1 , , 2 0 and 2 , , = 1, 4 2 0

such that 1, 2,
ie
4,0 j =
. l.

iEl
With the definition
n=0 for i e I, and , = 0 for i e I,

we have 2 A4 4,4 j = 1, 2, . .
i =
Let = y * . Y n ) b e an optimal strategy for the column player.

We that every of R, is associated with

assume row some positive probabily
in some optimal strategy. Then we have, if be the value of the
v
game,
n

j
2
= T
a , = v, for all i e 1
 THEORY 29
m m
Now v i = l j = 1 i = 1 j= 1|

)= 4=
. t this is a contradiction to our assumption. Hence our assumption of associating
row of R, with some positive probability is wrong, that is, some probability
every
h e zero with the result that the corresponding row or R, may be deleted.
This proves (i).
The proof of ii) can be achieved similarly.

CENERAL RULES FOR DOMINANCE

We give below a summary of rules for dominance as deduced in the previous
section.

) If all the elements of the i-th row be less than or equal to the corresponding
elements of any other row, say the r-th, then the r-th row dominates the i-th rowN
and we discard it.
(i) If all the elements of the j-th column be greater than or equal to the
corresponding elements of any other column, say the p-th, then the p-th column
is dominated by the j-th column and we discard it.
In the case of row player the inferior row is discarded, while in the case of
column player the dominating column is discarded.
(ii) If the i-th row be dominated by a convex combination of other rows,
then the i-th row is deleted from the payoff matrix. If the j-th column dominates
a convex combination of other columns, then the j-th column is deleted from the
payoff matrix.
iv) R. R are the sub-sets of the rows of an m xn payoff matrix and
C C are the sub-sets of its columns. If a convex combination of rows (columns)
in R, (C) dominates a convex combination of the rows (columns) in R, (C,), then
there exists a row (column) in R, (C) which is discarded.
B Let us use the concept of dominance to
B B B3 BA solve the game whose payoff matrix is given

A2
4
t
by the adjacent matrix.
4 Let A be the maximizing player and B be
A the minimizing player.
A 2 4
As By the rule (i), as all the elements of A
4 less than to the
A0 are or equal corresponding
here A, dominates A Hence
elements of Az, B
reduced B
we
discard A, and write the

associated with A.
matrix
B B B
is
A probability
we see
zero

that all elements of B, are less 3 4 24

Next
than or
equal to the corresponding
elements of A A 4
24
Hence B, is dominated by B Therefore A
B.
LPGT-F26
 GAME THEORY
30
matnx as given
get the next reduced
below
discarding the column B,, we
B
(Rule ii).
combination of B, and B B B
we see that the convex

Again strategies
B. for example. average
of these two pure
dominated
is by B,.
A424
that is, 3, 2. 4 2 4
i.c.(B + B,). reduced
A A3 |
discard the column B, to get the next
Hence we
A,4 0 8
matrix {Rule (i)].
hence we discard the pure strateos
A, and A, dominates Az; ategy
Also the average of on the right below. In
then comes to
be shown the
and the reduced matrix th e
A, zero with the
columns
B, and B, and also
process, we
associate a probability
row A.
B
B

B B B B
A A
Ag 40
A A 4 A08
A 08
No further reduction is possible.
If now x, and x, be the probabilities with which Az and A4 are played and

with which B, B, are played for the optimal solution,

'g Y be the probabilities
then we get, for this 2 x 2 rectangular game, as in the previous section,

and the value of the game is

Hence the optimal solution to the original problem is for A0, 0,

for B0, 0, 5and the value of the game is

Here zeros have been assigned for the discarded strategies.

15 TWO-PERSON, ZERO-SUM 2 x.a OR n 2 GAMES

If, in a game, one player has two pure strategies and the other has pure strategies
of any number n, then these games are mentioned as 2 x n games or n x 2 games.
If the game has a saddle point or it can be reduced, by dominance, to a 2 x
game, then the solution becomes easy. But if that does not happen, then, for the
players with more than two courses of action, a combination of just two out of his
n courses can be found which will give his optimal strategy. This follows directly
from
the duality theory of linear programming.**
game irom these 2 x n or n x 2 games will provide the optimal solution of the
Thus one particular 2 x 2
sub-
game. For the purpose, we shall have to form all possible 2 x 2 sub-games om

**
We shall prove in a linear
a
subsequent section thata game problem can be reducea
programming problem.
 ME THEORY 31
2 game and
x n or n x
compute their values. If the
plavermaximizing
eOurses
the
n co
of action, then the strategies of the sub-game
giving the maximum
value will provide the optimal solution and if the
has

minimizing
player has n courses
then the strategies of the sub-game with least value will
o f action,
solution of the game. The method is illustrated by an example.
provide the
optimal

Let us solsolve, for example, the game whose payoff matrix is given by the
followingmatrix
B
B B B
AA4 32
A64-
There is no saddle point and no course of action dominates any other.
Hence we compute the values v. v2» V, of the following three 2 x 2 sub-games
as obtained from the given matrix (vide Art. 8)
B B B
B B B B B B
A A43 A
A -4-2 A
A3-2
A 64 6- Az4-1

v = -1 (for saddle point solution),

= -
Now the minimizing player having three pure strategies B, B, and B, will have
to decide his course of action, and he will definitely select that pair of courses

which will give the least value of the 2 x 2 sub-games. v, being

the least, the
third sub-matrix will provide the optimal solution to the given game problem.

Theoptimal strategies of the players will

thus be and
(0. and

the value of the ame is (vide Art. 8)

These computed from the third sub-game,

are
a zero being assigned to the

discarded strategy B.
followed for the n x 2 games.
Similar procedure will be
n. For larger values.
for small values of
works good only
The above method 2
to compute values of
a larger number of 2 x sub-games. In such
we shall have
as explained in the next article.
the problem graphically
cases, we solve
APHICAMETHOD OF SOLUTION B
16
We illustrate this
method by examples. B, B B By
whose payoff
game problem
Consider the
matrix is given on the right
side. The game has no A 3o2
saddle point. Suppose
A chooses A, and A with A 30
 GAME THEORY
32
B choOses B, alongwith thie
and (1 -a) respectively. 3 21.
the
probabilities X+ 3
(1 -) = -

A is g =

expected gain g of against , we notice th

draw a graph of this function showing 8 by
If we to lie on the line
throughout, B can restrict A's gain
choosing B 8 3 - 2.

We compute similarly
the other gains of A, when B uses B, B and B. and

they are
8 3x, 8 = l -x, g = 3x - 1.
We draw two parallel axes one unit distance apart and mark a scale on each
The wo strategies A and A of A are represented by these two straight lines.
00 = I unit is the initial line.
The gain lines L, Lz Ly, L4 are then drawn
to represent the gains of A corresponding to
4
B, B B and Byy respectively of B by joining
1, 3, 0, 2 on the line representing A, with
3.0,1, -1) on the line representing A
Now, in order to find out the point which will
maximize the minimum expected gain of B, we
L -1
bound the figure so obtained from below by drawing
a thick line. The highest point M (maximum)) of
MAXIMIN 2 this bound refers to the maximum of the minimum
3 -3 gains. At this point, as is obvious, B has used his
A Au two courses of action and they are B and B
as it is the point of intersection of Lz and L

Therefore the 2 x 2 sub-matrix hich provides B

optimal solution of the given game is shown on the
right. Now, solving as usual this 2 x 2 sub-game, we
B By
get x =and 1 - x = (also indicated by the abscissa A A o2
of the point M). A1-
The probabilities for Bu and By are computed as } The value of the
game is (as indicated by the ordinate of the point M).

Thus the

Notice that
optimal strategies
and
are for A(andfor a(0.0. andv=
B, B are dominated strategies as is obvious from the graph
as also from the payoff matrix.

This method is equally applicable to n 2 games.

x
But, in this case, the graph
stands for B's loss against the probability
y with which B plays B. In this case,
to minimize the maximum loss to
B, we are to bound the figure from above. Ihe
lowest point of this bound will refer to
the two courses of action taken by A for
the purpose.
 E THEORY
Consider the game Oame whose payoff matrix is given below
33
alongwith
ith the graph.
the graph.

B
7 MINIMAX
6
B B
A o-2
Az 7-1 3
2
A A - 4

A-2
A5-3
3
B
In this graph, we have plotted the points of A's pure strategies on the corresponding
strate gies of B shown by two parallel lines. B, and B,on
B
the graph at unit distance 00 apart. The lowest point M
(minimax) of this figure, from above of the thick bound,
B B
coresponds to the pair of strategies A, and A, of A; hence A
A7
the 2 x 2 game which provides the optimal solution of A26
the given game problem is shown on the side and can be solved easily.
It may sometimes so happen that we get two pairs of strategies having the
same value for the game giving two pairs of optimal strategies.
Let us consider the game whose payoff matrix is given below.

B Graphical representation of the game as

represented by the given matrix is given in

B B B BV the diagram (given below), the manner of
computation being the same as in the previous
AA65 2 examples. From the graph, it is seen that the
A 126 3 player B has two pairs of supporting strategies,
the
for example, (B. By) and (B By) giving same

The player A has

infinite
value 3 for the sub-games.

number of mixed optimal

strategies (, 1 x), where - 6
be found either by 5 5
x varies from
two 2 x
to This

2 sub-games
or
can

from the abscissae Ls 4

solving the by solving the
P and Q. It can be shown
of the points formed by two different pairs
2 sub-games
two 2 optimal strategy for B is his
x
strategies
hat the
of B's
pure strategy By.
payoff matrix that three
happen in
o
some
so
t may minima or maximin point. In
meet at the
gain lines

case, to
mine the optimal strategy for the |A A
this
 GAME THEORY
34 to the gain lines
2 sub-games corresponding nes havingg
players. we solve the 2 x
will be explained
reason for the
same
The the
slopes with opposite sign. B
graph of such problem.
Consider, for exanmple, the game whose
B, B B BV
the adjacent matrix. The
payoff is given by
graphical representation
of the game is shown
that the
AA 32 2

in

three
the diagram given below.
gain
It is

lines of A corresponding to
seen

B,, B
Au6253
the maximin point M. The value of the game is seen to be 2
and B pass through
Solving the 2 x 2 sub-game involving B, and B..
solution
By
of B, we get the optimal as

for A, optimal strategy is ( i

for B, optimal strategy is ( 0.0 and the

value of the game is

M 3 The value of the game is for the 2 x 2 sub-

2 game with (B. By) also.

L But the value of the sub-game with (B,. B) is3
which is the saddle point of the corresponding sub-
matrix. Thus, by choosing (B, By), B is sure to lose
at most 3 while as shown earlier by using (B, B) or

A A (B By), B is sure to lose at most 2.5. Thus, for B,

the mixed strategy using (B, By) will not be optimal.
This is also obvious from the graph. Because, for the two strategies (B, B
the point M., the intersection of the lines, is not the maximin point but 3 is the
maximin point. But 3 cannot be the value of the original game; hence the 2 x2
sub-game with (B, B) will not provide the optimal solution for B.

17 ALGEBRAIC METHOD OF SOLUTION OFA GAME PROBLEM

Let v be the value of the game given by the payoff matrix lamxn and

x, (i 1, 2, .
=
m), y, (j = 1, 2, .., n) be the probabilities for the optimal
mixed strategies of the players A and B respectively, such that

x= I and
j
y, = 1, (1)

in which x, 2 0 for all i and y, 2 0 for all j.

The expected gain of A when he uses his i-th strategy and B uses his j-th
strategy S

As A expects at least v, the value of the game, we should have

(2)
4,v.j==1, 2.
i = l
 THEORY 35
imilarly, considering losses to B, we should have for his
choice
s v, i =1,2, *******, m.
1
(3)
This the problem will be to find the
values of x, (i =
1. 2.
g 1, 2, .. n) which will satisfy (1), (2) and (3). m ) and
Here, as we did in the case of 2 x 2
suppose games, we
all the
(3) equations
as and then try to solve
these equations inequations (2)
colution satisfying (1), (2) and (3) will be the alongwith the equations
(1).
Sometimes, some of the
required optimal strategy.
equations
found to be not consistent.
are
that one of the This implies
inequations is a strict inequality. Then
method to find the solution of a set of we
apply trial and error
explained in the following examples equations with a few strict
inequalities as
Here we state two theorems
although the method is
quite lengthy.
of the duality theory and (without proof) which are the direct
which will make the implications
Theorem 1. f. for some j k, computation easier.
there be a strict
=

a11a2*2 +.+mmV, then the inequality, say.

similarly, if corresponding y,
for some i =
p, there be
=
0;
a strict inequality, say,
p1+ay + .+
<V, then the corresponding
Theorem 2. If x, =

with non-ero
one
player's optimal strategy consists of
exactly r strategies
probabilities, then the optimal strategy of the other
involve r
pure strategies. player also will
Consider the game whose payoff matrix is B
by the adjacent matrix. Let (x, given
x x)
be the
B
probabilities of the
and (y, Y Y
optimal strategies of A and A
2 1
Tespectively of the given problem whose value is 2 2
supposed to be v.
For A, to expect at least a
34-3
gain v, we shall have -1.x+1.x,
and two other inequations and for B, to +
3.x 2
expect at most a loss v, we
-1.y+2y, + 1.y S shall have
v
and two other inequations.
Furthermore, + X 2 t+ X3 = 1,

Y+2+ Y3 1, (1)
(2)
First of all, we consider all the above (3)
inequations as equations
-t + 3x, = v,

2x 2x, + 4x = V, (4)
+2x - x = v,
(5)
t 2+ Y = vV, (6)
y-2 t+ 2y, = v,
(7)
3y,+4y2- 3y, = V. (8)
(9)
 36 GAME THEORY
and (4) to (9) in seven
cight equations (). (2)
unkna..
We have thus owns
find the can be solved to
. A Y y. y and v. linear equations
These
values n

of the variables. If this solution satisfy the non-negativity condition (3) also
then
this solution will give the optimal strategies and the value of the game.

We casily solve these equations and the optimal strategies for A and B are
re
found to be ( 6 and2 22 respectively which satisfy (3) also.

The value of the game is B

Let us again solve the game whose payoff is given I II III

by the adjacent matrix.

As in the previous example, we assume ( , Xzs X) I32|4
and (. y V)as the probabilities of the strategies
A II -1
4 2
of A and B respectively for optimal solution and v II22 6
as the value of the game.
Writing the obvious inequations as equations, we get a system of linear equations
to solve for the variables. They are

3x, -X2+ 2x =
V, (1)
-2x + 4x + 2x =
V, **
(2)
4x+2x, + 6r =
V, (3)
3y 2y + 4)3 = V, (4)
- + 4 + 2, = v, (5)
2y, +
22 +
6y3 =
(6)
together with (7)
and (8)
satisfying (9)
From (1) and (2), we have x = *2 and then from (2) and (3, we have
x= = V = 0 so
that x = 0 also. But then the equation (7) is not satisfied
and hence the system becomes inconsistent.
Then we try to find the solution by taking one or more of the inequations
(written as equations) as strict inequality and solve alongwith the others considered as
equations. If we now assume the third and the sixth of the inequations (corresponding
to the equations (3) and (6)) as strict inequalities, for instance.

4x,2x +
6x, > v and 2y, +2y, +6y, < v

then we have, by the Theorem I stated earlier, x = y, = 0. Solving the equations

with these values of x^ and ya we get

v and y, =v.
These together with y3 0 do not satisfy (8) and hence the system is again
inconsistent.
 GAME THEORY
37
we try a different set of
differ
equations as
Then
we take
inequations, for
example, for (3
nd (4),
4x+2x, 6x> and 3y, 2y, + 4y,<v
+ V -

nsider the remaining as equauons. Ihese two (10)

inequations imply
con:
and
0. With these values of x, and we solve
that x, =
0
and y» the
remaining equations to get
= 0 =
. y, =
and y =
0.
These values satisty all the
equations and (9), (10), giving v = 2.
Thius the optimal solution of the game is givenas
A 0, 0, 1), B 0 and v =
2.

STRANSFORMATION OF A GAME
PROBLEM TO AN L. P. P.
Every finite two-person zero-sum game be
can
expressed as a linear programme
and conversely, every linear programme can be represented
sum game.
as a
two-person zero-

Consider a game problem with the payoff matrix

The maximizing player A has m mixed
[ax
strategies which he chooses with
probabilities x. X2 , Xm respectively, such that
X+X + +Xm .
1, where x, X2
2 * ., m 0. (1)
We shall show that this problem can be transformed to a linear
programming
problem before it is solved. The expected gain to A when the minimizing player
n

B chooses his j-th course of action out of his n courses is 4

i=1|
If min (L4x,) = v, then A expects to get at least v. A will choose his strategies
j
with such probabilities as he maximizes this least guaranteed gain v. Now
m

2 a 2 v , for all j (2)

i=1
of the game is clearly the
as vis the minimum of all expected gains and the value
maximum value of v, if it exists.
Thus x, X2 Xm are to be so determined
maximize v.
as to satisfy (1) and
in a standard L. P. divide (1) and (2) by v, which
P, we
To put this problem
which will be indicated by the presence of some
is positive. If v be not positive,
matrix, then we add to all the elements of the
elements in the payoff
negative posiuve quantity c such that
v becomes positive.

matrix a sufficiently large

payoff the oplimal solution but only increases the value
operation does not change
This

of the game by c.
(>0) and setting
(1) and (2) by
v

Now, dividing
m), we write the problem as
X, (i 1, 2,
= =
1, j = 1, 2,
aX+ay*2 +
**'***

... +X
X+X+
and
*** *

with
38 GAME THEORY
Now the maximization of v is equivalent to the minimization of hence wve
can state the problem for A as

Minimize= X, + X, + +

subject to a X, + a*2 + m 1,
(3)
j= 1, 2,
and X, 0 . i= 1, 2,

This is an L P. P. ritten from the point of view of A.

Considering the problem from the point of view of B who will expect to
minimize all the expected loss for him, we can arive in a similar manner at the
following L P. P.:

Maximize==Y, +
}, +

subject to aY +4pl2 + +m'S1 ***

(4)
i l , 2, ***9

and j = 1, 2,
where y's are the mixed strategies of B
Having found X, and Y, we can find x, and y, from the relations x, = vX, and
v
Finally, the value of the original game is obtained by subtracting c, if needed.
It is seen that the problems (3) and (4) as stated above are dual to one another.
Thus the optimal solution of one problem will afford an optimal solution for the
other. Player B's problem can be solved by regular simplex method, while that of
A can be solved by the method of
duality.
19 TH REM
If mixed strategies be allowed, then there always exists a value of the game,
that is, Y = V.

Consider a game problem whose payoff matrix is [almxn Let the maximizing
player A and the minimizing player B use the mixed strategies with probabilities
mand y, Y2 s Ynrespectively.
We know that this game problem can be transformed into an L. P. P. from
the point of view of A as

Minimize X, +X2 +
X
subject to
a,X, +
az21 ****** T
m 1,
j = 1, 2,. *****ne.y

and X, = 20, i = 1, 2,
 ME THEORY 39
The
dual problem coresponding to this L P. P. is

Maximize = Y, + ¥, +
+Y
+Y
subject to aY, + apY, +..
+anS 1,
i =
1, 2, ... m

and Y 2 0, j 1, 2, .
** .

Rut this dual problem is the same . P, P. which can be written from the
siven game problem from the point of view of B.
gver

Without any loss of generality, we assume that

a 0 , i = 1, 2, . , m; j = 1, 2, ,

Written explicitly, the constraints of the L. P. P. from the point of view of A are
aX+X + m1X1
m221,

Tmn mnm1.

f 1
A Min (a ajg n
X Xg X =0,
then we see that X 2 0, i = 1, 2, ...
m and (X,, X7,
2 . , Xsatisfy
the above inequalities, So the L. P. P. from the point of view of A has a feasible
solution. Further, the objective function
X +X + + 20,
for every such feasible solution, so that the objective function of the minimization
problem cannot be unbounded. So the problem for A has a finite optimal solution.
Now the duality theory asserts that "f either the primal or the dual problem
has a finite optimal solution, then the other problem has a finite optimal solution
of the objective functions are equal"
and the optimal values
a finite optimal solution.
Hencethe problem of B has

Thus, in this case, Max. = Min.

that is, = V = v (value of the game).
Note, If the payoff matrix has no saddle point, then mixed strategies will be allowed and there
ert
optimal strategies of the two players.
will exist

2080LU0 GAMB BY 'SIMPLEX METHOD

Here we show how a game problem whose payoff is given by the matrix given in the

page can
nextWe
reducing
be solved by it to an
L P. P. and then
applying simplex method.
note that the game has no saddle point. The maximin for A is 2 and
nax for B is 4. Thus the value of the game v, lying between 2 and 4, is
tive (we need not add anything to the elements to make v> 0).
positiv
 40 GAME TH
lf y2» y y) be the probabilities with which the strategies for B are pla
W are to find these quantities which minimize v subject to layed,
B 3y,+ 2 + 4
3y,4y+ 23 + 4y S v,
III III IV
4y,+2y2 + 4 V,

4 4y2 +8y4 S,
A 34 and 2 0, j = 1, 2, 3, 4.
III 4 2 40 Dividing by v and putting
IV0 08 Y i = 1, 2, 3, 4, we can state
the L P. P. as
Maximize =Y, + Y,+ Y,+ Y
subject to 3 Y +2Y, + 4Y s1,
3Y4Y +
2Y, +4Y, 1,
4Y +2Y, +4Y S1,
4Y2 +8 1
and Y, 2 0, j = 1, 2, 3, 4.

Now adding slack variables, we can solve this problem by simplex method to
find Y, Y, Y, Y, and also to find the maximum value of . that is, the minimum
value of v.
Then we can find y, from the relations y, vY, and hence the optimal strategies
of B will be obtained.
We assume (x, iz X * to be the optimal strategies for A. Since A's problemis
a dual of the problem of B, the values X, X, X, X, where x, = vX, i = 1, 2, 3,4
can be found from the final tableau of the simplex iteration for the determination
of Y. Y Y, Y, from the (, - e) row under the columns of the slack vectors.
This is left as an exercise for the reader who can easily verify the optimal
strategies of A and B as0 0, (0. 0, and the value of the game is

2 M REMECHO
We summarise here the systematic method to be followed for solving a rectangular
game.
G) At first, search for a saddle point.
(i) If there be no saddle point, then use the concept of dominance by which
the game may be reduced in size. (Sometimes a solution may be obtained by
dominance only).
of the then
(i)
solve it
If the reduced size
the method of mixed
game be (2 x 2) with no saddle point,
by strategies.
(iv) If, on the other hand, the reduced size of the game be (n x or

(2 x n), then, by using graphical method, reduce it to a (2 x 2) game and then

apply the method of mixed strategies.
 GAME THEORY
()On failure of these methods, apply algebraic method.
41
(vi)On
On failure of other methods, one may reduce the game to an
simplex method. L P. P. and
olve it by
LLUSTRATIVE EXAMPLES
Ex. 1. Sotve the game with the
following payoff matrix

4 2
ICalcuta Hons., 1980; Kalyani Hons, 2008)
There is no saddle point. Let the mixed strategies be considered
nrobabilities
P X =
(1. x), Y =
0, y) for the
with the
respectively, so that maximizing and minimizing player
X+X2
1, y, + y2 =
I and
x, Xz y. Y2 2 0.
=

If v be the value of the

game, then the optimal solution is

T+2-(3+4)=
1-4
T+2-G*
V 1.2-34
1+2 (3 4)
Alternatively
The expected payoff function E (X, H is given by
E (X, ) = 22a, 7 1 x , * 4xy, + 3 x , + 2.xy.
where X = ( , X2). Y = 0 , V): X, + X2 = y + y2 = 1.
..
(1)
E (X, Y = xy, * 47y, + 3x,V2 + 2x,yj

=
xy, + 4y, (1 -

x) +3x, (1 -

») + 2 (1 -

x) (1 -

y), by (1)
=
Xy+4y -

4xy, +3x, -

3x,y, + 2 -

2 -

2y, +
2x
- 4,y +X+2y+ 2 = - 4 ( - ) O - +

faw if the row player chooses x 2. then he ensures that his expectation is

5 He cannot be sure of more than . because by choosing y, = the

column player keeps E (X, Y down to ~ So the row player might as well settle

5when the column player reconeiles to Hence the optimal strategies are
for
x - and =(
= v avlau
l ue
. of the game = E (X°, Y) =
=
v
Also
 GAME THEORY
42
and solve the follogino
reduce the payoff matrix
Ex.2 Use dominance to
game problem given by the payoff matri
B

|17 2
627
[Meerut, 1974
of A and B respectively. We
1. 2.
i 3 be the pure strategies
Let A, andB, =

dominated by the second row and the third

row is strictly
see here that the third
third row and the third
column. Thus when the
column strictly dominates the first
matrix becomes
column are deleted. the reduced payoff
B

A
7
6 2
No further reduction is possible.
In the process, we associate a probability zero with Ag and Bg.

Here the expected payoff function is given by

E (X. ) = xy +6r,y, + 7xy + 2x,y2
= x +6y, (1 - x,) + 7x, (1 - y ) + 2 (1 - x ) (1 - y )

+y 1.
since +X2 1
E (X Y = -10xy, + 5x +4 +2

- 10 - o- + 4.

Then proceeding as in the previous example, we get the optimal solutions with the
probabilities for A and B as A 5 } . B and the value of the game is 4

Hence the optimal solution of the complete game is for A ane

0 for B, the value of the game being 4.

Ex. 3. Use dominance to reduce the following game problem to 2 x 2 8ame

and hence find the optimal strategies ard the value of the game
PLAYER B
32
PLAYER A-1 4 2
226 [Visva-Bharati, 1988]
Let 2 X,) and ( Y2 Y3) be the probabilities with which B
A and B play the game.
We see that the first column is dominated by the third column
2
and hence the third column is deleted. The resulting 3 x 2 payoff A-
matrix is written.
 GAME THEORY 43
we see thatthe third row dominates a convex linear
Now
combinatiun of the
and second row; for example,
frst
the elements of (first row) + (second row)

areactually +2) or, 1,1

his dominated by the third row elements 2, 2. Hence either
first or second
Can be deleted (Rule iv). We delete the first row and the game reduces to
the 2 x 2 payoff matrix

|-14
2 2
Solving as in the first example, we get the optimal solution as

x= 0, Xg -1 2 (4
-1 - 4 2) _ 1,

- and v =2=2 2.
We associate
probability zero with the discarded row and column.
a
The solution
of the complete game is thus A (0, 0, 1), B . 0 and v = 2.
Note. Notice that the final 2 x 2 matrix may further be reduced, since the second
which is the dominating column, column,
may be deleted and then, from the reduced matrix, the first row of
that matrix being dominated
by the second, may be deleted. Thus the
optimal solution becomes for
A (0, 0, 1), for B (1, 0,
0) and the value of the game is 2.

Ex..4. Two players A and B match coins. If the coins match,

then A wins
one unit of value
and if the coins do not match, then B wins one unit of value.
Determine the optimum strategies for the players and the value of the
game.
Delhi M. B. A., 19721
The payoff matrix for the matching players
A and B in which H and T
denote
respectively the head and tail of a coin is given below
As can be seen, there is saddle point for the game
B
H T
and no reduction is possible by dominance. Let (x,, x,)
and (y ya,) be the probabilities with which A and B A
A HI-1
play their pure strategies. T-1 1
As in the first example, the optimum mixed strategies wil be

1+1
+ 1 -- l -1)
--
-

v =
value of the game =
0.
44 GAME THEORY

For the following payoff table, transform the zero-sum game into
kx.5. and solve it by simplex method
equivalent linear programming problem
We formulate the problem for the player Q.
PLAYER Q
Maximize = Y,+ Y, + Y

1.
subject to 9Y, +
Y, +
4Y, S P
4
6Y, +3Y,s 1, PLAYER P Pz0| 6 3
5Y, +2Y, + 87, s 1.

Y. Y, 2 0
and y, v¥, =
G =
1, 2, 3) P52|8
and is the maximum expected loss to
We introduce the slack variables Y Y, Y, respectively and then apply the
simplex method. We get the tables by iteration as shown below.

1 0 0 0
C Y b a 4 as
0 Y 0 0
0 1
Ys
0 Y6 2
-1 0

Y
0 0 0
0 Y6 13 0

0 0 0
9

0
54 18

0 0 0

0 1 0 91 13
54 18 54
4 0
27 9 27

0 0
13 13
40 0 0
273 91 2T
10 18
Y3 273
0
91 91
24 0 9
91 91 91

The expected value of the game is thus 91

 ORY
45
Hence , v = T5 = 24
Similarly, 2 =and y, =

The Optimum values of X, A2, K are obtained from the dual

solution 9 2
Hence vx, 1 97
Similarly, x 2and x^ =

Hence the optimum solution is given by P ad

F . 6.
payoff matrix for the first player in a game be
4e such that 0 a
a< a <
a a G

dhe vectors and Y be the optimal strategies for the first and second

player and v be the value

of the game, then
prove that
y2 0, ii) 4nd (ii) ag 4gt
Since a > as and
a^ aq, all the elements of second column are
>
than or equal to the greater
corresponding elements of the third coBumn and hence the
second strategy of the
minimizing player is dominated by his third strategy. Hence
Y = 0 and the matrix game reduces to

Since a,> a, the third strategy of the maximizing player is dominated by his
first strategy and hence xz =0 and the matrix game reduces to a, a

Now, by the formula, *1 (a -

a) +
(a, -

43
Since (a a) < (as - 41), we
get x, >

Now row minimums are a, and a

Hence v = max min a = dz as a ag.

column maximums are a and a.

Again,
Therefore v = min max a,j 4 as a < as

< V < V, we have

Hence, by y

LPGT-F27
46 GAME THEORY

EXAMPLESB
matrices
the games with the following payoff
01 Solve
5 1 (06-3
3 4 -3
[Bombay, 1975]
Visva-Bharati, 19877

determine the values of the games

02 Solve the following game problems and

02 5 (i)3-2 (i)-4 6
7 3 -2 3 2-3
[Madras, 19771 Punjab, 19771 [Bombay, 1974

03 In a rectangular game, the payoff matrix is given by

10 5 5 204
1115 10 17 25

7 12 8 98
5 13 9 10 5
State, giving reasons, whether the players will use pure or mixed strategies.
What is the value of the game? [Calcutta Hons., 1989]
Anss Pure strategy: v 10.]
0 Use dominance to reduce the payoff matrices and solve the games with the
following payoff matrices

B (ii) B

6 8 -5 3 120
A4 122 A 5 54 6
[Meerut, 1971] 4-2 0-5
[Calcutta Hons, 1985]
ii) B (iv) B

8 15-4-2 2 3
A 19 15 17 16 A2 0
o20 155
[Visva-Bharati Hons., 1995 [Bombay, 1975]
(v) B (vi) B

A 2
1727 8 8 65 58
5 1 6 45
[Calcutta M. Sc., 1975]
6 5 6
 GAME THEDRY 47
B
(vii) B
32 4
2 4 2 4

2 4 0
4 2o 2 1 1
A
4
32 2 3 1
408 4 3 75 1
Calcutta M. S., 1976;
Jadavpur, 1993 4 3 4-12 2
4 3 32 2 2
Anss)(A, B), v =6. (i) (A. B), v = 4. [Jadavpur, 1985)

( 0. o. }: v=-

( o, o). (o. . }: v-

(i) (o, o. 3. i: (o,. o. 3. i} v-

*
(it) (o, 0, 70. o. 0, o, . . o, o: v =
05 Reduce the following matrix games into 2 x 2 matrix
games
2 21-2 -3 (i) 4 2 0 2 11
4 3 4-2 o 43 13 2 2
5 12 5 4 3 7-5 12
[Calcutta Hons, 2006 4 3 4-1|2 2
4 3 3-2 2 2
[Shivaji, 1976
Anss ( 3 2 (1
06 Show that if dominance occurs in the payoff matrix of a 2 x 2 game, then
saddle point. Is the converse true?
there is a

Anss Yes.]
for the game problem the optimal strategies are
(07 Show that,
ab
aba6 and is the value
OS Show that,
for the game problem a-b

a are an E 0, the optimal strategies are

where a, b, C,

+ b +Ctdand v = ad E bc Calcutta Hons., 1984; Kalyani, 1985

E =
a
where

that tne game Is fair, if ad bc = 0.

now further
 48 GAME THEORY

09 A game has the payoff matrix

Assuming the mixed probabilities X = x, (1- X)) and Y {y, (1

.
show that E (X, Y) = 1 - 2x y-7and deduce that in the solution of the aarm
the first player follows a pure strategy while the second has infinite number of miver

strategies. Benaras Hindu University, 1968

10 By constructing the expected payoff function of the game given by the matrix

show that the game is fair.

46
2 3
| 1 A and B each takes out one or two matches and guesses how many matches
opponent has taken. If one of the players guesses correctly, then the opponent has
to pay him as many rupees as the sum of the numbers of matches held by both
the players. Otherwise, the payout is zero. Write down the payoff matrix and obtain
the optimal strategies for both the players. [Gujrat, 197

1 2 In a game of matching coins with two players, suppose A wins one unit of
value when there are two heads, wins nothing when there are two tails and loses
unit of value when there is one head and one tail. Determine the payoff matrix
and the optimal strategies for both the players. [Delhi M.B.A., 1971

Ans
13 Solve graphically or otherwise the games whose payoff matricess are given
below
(i) B B
(i)
B B B B
A
AA8 5
3 11
2 A A2 3 5
[Calcutta Hons., 2001 A 11
Jadavpur, 1984; Calcutta Hons., 1986
(ii) B (iv) B
B B2 B B B
A1-3 A, 3-3 4
A
A2 3 5
A-11-3
A As
6
[Madurai, 1977
4

As2 2
A-5 0
Kalyani, 1985; Calcutta Hons., 1984
 ORY 49
(vi) B
8, B B B B B B B
A,815-4 -2 4 A 2 2 3-1
A A 19 15 17
4 20 15 5
16
A4 3 2 6
Calcutta Hons., 1988, 1992
Jadavpur, 1993)

o..0. 0. o). 1 :v 3
o.
(m
.0. tva-
. (o. 0. v=
i. oo v
14 Find the value of the
game whose payoff matrix isgame and the optimal strategies for each
player of the
B
i) B

--1 2-3 4
A
-1-13 A-3 4 5
4-56
[Jadavpur, 1985)

1 Use algebraic method to solve the games

given by the following matrices
) B (i)
-5-1-1 -1 2 1
A4 0 2 A
1-2 2
-520 4-3
[Gauhati, 1976 [Kanpur, 1973

6 Transform the games with the following payoff matrices to the corresponding

PL. I

(i) B (0) B

A
2-2 3 1 0-2 |
-3 5 1 1 3 2
(Delhi, 1972 Delhi, 1970
To get the value of the game non-egaive, add 4 to all elements of ( l

s 8.. 6Y, *
2, +
7y, s 1,
Ans () Max. y,
+
Y2
y9y 33 1, y. Ya Ys 2 0, where y, = v

(i) Max. y, 2
+
Y, + Y 8.. 2, 2 - 2y, s 1,
y , 3 y +2y 1, Y. Y Ys Y 2 0 and y = VY]
50 GAME THEORY
17 Transform to L. P. P. and solve the game problems whose payoff mat.
are given below, by simplex method
ices
) B
B

L1-1 -1 -11 1
A -1-1 3 A
2-2 2
2-1 3 3-3
[Vidyasagar, 1988 [Visva-Bharati, 1986
(ii) B (iv) B

2 1-
1 -22 A
3 5 -3
3 43 6 2-2
[Jadavpur, 1984 [Vidyasagar Hons., 2002

(v) B (vi) B

3-1-3 1-4
A
-3 3|-1 A
-3
-4-3 3
[Calcutta Hons., 1986) [Calcutta M. Com., 1990

As va-

qin . v=
t3o). (0. : v = 1.
. :vs -
(vi) (0, 0, 1),. ( V = 2]

JIsTwo companies A and B are

competing for the same product. Their different
strategies are given by the following payoff matrix:
COMPANY A
A A A
cOMPANY B 22 3
Use linear
programming to
-35--
determine the best strategies for both the
players.
Delhi, 1979
A3A3 o). B ; v- ja
 AMETHEORY 51
nsider a modified form of "matching biased coins"
game
19 olaver is paid Rs. 8/- if the two coins turn both heads and problem. The
ching Re. 1/- if the
matc both tails. The non-matching player is paid Rs. 3/- when the two coins
od o nd match. Given the choice of being the matching or non-matching player,
o n e w o u l d you choose and what would be your strategy. which
Baroda, 1971)
A s For both players,75 15. v = -

5 As the
non-matching player always expects
tn gain, we would like to be non-matching player.]

0 In two-person game, each player simultaneously shows either one or

a
two
ngers. If the number of the fingers matches, then the
player A wins a rupee from
he player B, otherwise A pays a rupee to B. Show that the payoff matrix for this
game is

B
1 2

1 1 1
A

Solve this game

2-1 1

by reducing it to L. P. P.
[Calcutta Hons., 1982
Ass ). ( } v= 0
21 A and B play a game as follows:
They simultaneously and independently
write one of the three numbers 1, 2, 3.
If the sum of the numbers written be
even, then B pays to A this sum in rupees.
If it be odd, then A
pays the sum to B in rupees. Form the matrix of the game
for A and solve it.

2-34
AsA
L4-5 6
4-5.v-5, A ) . B(
22 If G be a p x q matrix game with optimal strategies p and q and the value
v, what can you say about the optimal strategies and the value of
the matrix game
(mG nE)? Prove your result.
Here m> 0, n is a real number and E is a P x q matrix with all elements
unity.
23 A and B play a game in which each has three coins, a 5p., a 10p. and
a 200. Each selects a coin without the knowledge of the other's choice. If the sum
of the coins be an odd amount, then A wins B's coin; if the sum be even, then
B wins A's coin. Find the best strategy for each player and the value of the game.

Ana A O). Bl O); v -o

onTwo plavers independently select one of stone, paper and scissors and
simultaneously disclose their cholces. The rules of the game are that the stone beats
the Scissors, the scissors Deat the paper and the paper beats the stone. Suppose
hat there is a score of one point for winning9. (-1) point for losing and o if both
olavers select the same object. Find the payoff matrix and show that the game
Vmmetric. Find also the optimal strategies for each player by reducing it to an
L. P. P.
[Calcutta Hons., 1991]

Anss -1 0
O.)
 GAME THEORY
52
APPROXIMATESOLUTION
23 ITERATIVE METHOD FOR
demand an approxim
matrices, situation may nate
In case large
of games with payoff
close enough
to the value of the game and nnot
solution giving an average gain
Iterative method ror approximate solution. s
the actual and exact optimal solution.
strategies as well as an approximatee
explained here. is used in such cases. Optimal of
this to any desired aegree accuracy. The
value of the game is obtained by
is
principle behind this iterative method
best suited to him (or worst to hic
Ar each play. the plavers select a strategy
has done upto that iteration.
oPponent) in view of what the opponent
Let A and B be respectively the maximizing and minimizing players. A arbitrarily
selects any strategy. say A, and places it under the matrix. Then B chooses his
strategy B, comesponding to the smallest number (most unfavourable to A) in the

the right of the matrix. Now.

places this column to
row already chosen by A and
for this selection of B. A plays that strategy which will maximize his average
gain. This selection of strategy will naturally correspond to the row with the
largerst number in the column B, The elements of this row, say A, are added to
the corresponding elements of the row A, and placed beneath the row A, Now, B,
in order to minimize his average loss, selects a strategy B, corresponding to the
smallest number in the new row and adds the elements of this column to the
corresponding elements of the column already placed on the right of the matrix.
Then A chooses a new strategy which corresponds to the largest element of this
second column placed on the right. In case of a tie, the player will select the row
or column different from his last choice. This procedure is continued for several
iterations. Finally, at any iteration for any player, a mixed strategy will be obtained
by dividing the number of times the respective pure strategies are used by the total
number of iterations upto that stage. In other words,
the probability of any pure strategy = number of times it is used
total number of iterations
To get the approximate value of the game, we have
minimum element of the last row
total number of iterations value of the game

largest element in the last column

total number of iterations
Note. The method,
although used to solve large game problems, is
and closer very slow. This method gives
closer
optimal strategies with increase in the number of
iterations.
We give here an example to exhibit the iterative
We encircle the smallest method as explained earlier.
element in the A's
to indicate the next strategy at the bottom of the matrnx
strategy to be adopted by B.
B's strategy on the Similarly, the greatest element or
nght of the matrix is encircled to
indicate A's next sud6:
 GAME THEORY
53
we solved earlier algebraically.
ecider the game
Consia

-1 2 11 2 4 6 5 6
A -2 2 O 2 4 5 3 4 6 4
8
3 4-3-3 -6 2 -5 -2 2 2
3 4 3
4 2 E
5
4 2
4 3
4 5
4 6
6 8 3

We have completed eight iterations.

During these eight iterations, A has used his first strategy thrice, second strategy
four times and the third strategy only once as is seen from the
corresponding
number of circles on the right of the payoff matrix. Here A's best strategies based
on eight iterations are

By similar consideration, we see that B's best strategies are

(3.i)
Ifvbe the value of the game, then s vsie, s v s since
3 is the minimum element of the last row and 6 is the maximum element of the
last column.

and
Thus the solution of the game for
v s
eight iterations is A
B(
Notice that the strategies determined above are not too far from the optimal
strategies obtained previously by solving the problem by algebraic method.
Note, By increasing the number of iterations, we can approach to the actual optimal strategies as
the game.
actual value of
also the

N SOLUTION OF A GAME PROBLEM BY MATRIX METHOD

Tet A = l4 be the payo matrix and B be any square sub-matrix of A of

orderr > | i and = 1, 2,

, . , m, J = 1, 2, . .

be the (1 X T) malrix given by [1, 1,

Le t T.
I

Let X = , X2 ** * and 2 = 1, x, 2 0
**
. . , , a n d }y, = 1, y, 2 0.
and
Y i Y2
 GAME THEORY
54
Let X =x, A the clements corresno
a (l x r) matrix obtained
from X by dcleting B and
deleted from A to get
to the rows that
were

deleting the clements correspondine

a (1 r) matrix obtained from Yby
x
deleted from A to get B.
to the columns that were
Then. for a square sub-matrix
B of order r 2

T,adj.B Y
ladj.B]"

and =

X T,(adj.B) TT T,(adj.B)T
where superscript T denotes the transpose. If any one of X; or y, be negative,
then reject this B and repeat the operation with another B.
If again all x, and y, 2 0, for i = 1, 2, ; j = 1, 2, r, then
calculate the value v of the game as

V
B
T(adj.B) 7
Notice that the denominator of the right hand side expression for each of
X. Y and v is a scalar matrix which may be used as a scalar multiplier by rule.
To get the optimal strategies for the two players, construct X and Y by adding
zeros to the appropriate positions of X and Y.
(The proof of this theorem is beyond the scope of this book.)
Note that x, and y, so obtained must satisfy
m

2 ,42 v, for every j =

1, 2, *****°***)

and
2,a[
=
, for every i = 1, 2, ********9 I.

Note. Though highly systematic, this method involves laborious calculations for large payoff
matrices. We shall describe a short-cut matrix procedure that solves any n x n game quite efficiently
in the following illustrative examples 3 and 4. This method is known as the method of oddments
or arithmetic method

25) ILLUSTRATIVE EXAMPLES

Ex. 1. In a well known children's game, each player says stone or scissors
or paper f one says 'stone and the other
a rupee. Similarly,
scissors then the former wins
"seissors beat paper and paper beats 'stone, te, the player
calling the former word wins a rupee. f the two players name the same item,
then there is a tie, ie, there is no payoff. Write down the
the value of the game as well as the payoff matrix and fnd
optimal strategies of both the players.
[Bombay, 19771
If A (maximizing) and B B
(minimizing) be the S P
players, then the payoff matrix of A is the adjacent
one in which S, p and indicate the 0
s
strategies of
saying scissors, paper and stone respectively. We A p 0
solve the game by iteration.
 GAME THEORY
55
B

0
A-I0 0 Oo 0
- o- 0

For three iterations,

optimal strategies for A and B
and the value the
Ex. 2.
game is 0.
are
)and
Solve the game whose payoff matrix is given
method below, by the matrix
Y

X8 52
Clearly the matrix has no saddle
We have point.
only three possible square sub-matrices
B, 8 , 3 111 and T, =[1, 1]
Consider B,
adj. B, =

1. 11 53
X 1, 11S3 3, -2]
3, -2] .. -[}
1, 115 -8
-3 .3
Y
Since y, =
-a <0, hence we
reject B,

Consider B,
ady B, =[ 2-11lad. B,l
B.
=[ 2
8
56 GAME THEORY

. 11 2- -6, -10] 610 =

-16
-6, -101[11
-1
.

Y=
-16
- -16

All the elements of X and Y are positive.

Now B, has been obtained by deleting the second
column of A.

Hence

-* -16
We now check the fundamental relations

+8r =8 =v,

This relation contradicts the requirement. So B, is not suitable and we try B

Consider B,

1,1 2-
-5 3J
-1

-11 -11

All the elements of X and Y are positive. Now B, has been obtained by deleting
the first column of A. Hence

.
V
-11
 GAME THEORY 57
to check the fundamental relations
We are now

3x+51 + ==
1lx+2 =3 ,49
+3y,+ 1ly, =+ = v
8y,+5y,+ 2» + V,
The optimal solution is therefore

Ex. 3.Solvethe following 2 x 2 gme by the method of oddnients

There is no saddle point of the game. Dominance is not applicable here. We

try the matrix method to solve it. First we find the difference of the two numbers
in the first column and write the difference below the second column. Thenwe
find the difference of the two numbers of the second column and write it below
the first column omitting the negative sign.
Next we perform these operations for the rows. These calculated numbers are
called the oddments of the various pure strategies and indicate their frequencies in
the optimal strategies of the players.
The oddments are written as in the following table :
B
II

A
I-6 3
II
2 7
6 10

The sum of the two sets of oddments are the same, (here 10). This is a necessary
the application of oddment method.
condition for
The optimal strategies of the players are

A
(to o).»( o) that is,
(
of the game is v
3 x (-1)+ 7 x5 16
The value 10

4x 5 + 6 x 2
10
 58 GAME THEORY
Ex. 4 Solve the following 3 x 3 game by the niethod of oddments

3-
A

The game has no saddle point. Dominance is not applicable. We then use short
cut matrix method.
Each row is subtracted froom the row above and the differences so obtained are
written in the form of two rows below the payoff matrix. Similarly, each column
is subtracted from the column to its left and the differences so obtained are written
in the form of two columns to the right of the payoff matrix. The following table
isobtained by these operations
B

I II
3 -1-3 4 2
A II 3 3-1 -6 4
II4-3 3-1
-4

6 -4
Now we compute the oddments for the strategies of A and B as follows
Oddment of A's first strategy is-6 4= 40,
-1 -6
oddment of A's second strategy is 4 2- 22.

Similarly, oddment of A's third strategy is 28.

Oddment of B's first strategy is-4-2 = 28.
4

Oddments of B's second and third strategies, on similar computation, are (-22)
and 40 respectively.
Now, neglecting the negative signs, we write the corresponding oddments in
the following table :
B
I II II
We see that the sum of the oddments
3- 40 of both the players is the same (here 90).
A II3 22
Thus the game is conformable for oddment
III 4-3 3 28 method. (If this condition be not fulfilled,
28 22 40 90 then the oddment method fails.)
 GAME THEORY 59
Now
net the optimal strategies of the players, we divide the odd1ments hv
the sum of
the oddments. Thus the optimal strategies of A and B are respectivelv

40 22 28 that is, A

and t h a t is, B5
The value of the game is
40 x 3 + 22 x (-3) + 28 (-4)
1
90
40 x (-1) +22 x 3 28 x (-3) 29
90 45
28 x 3 22 x (-1) +40 x (-3) -29
90

EXAMPLES C
01 Solve the following matrix game approximately : (Ten iterations)
PLAYER B

1-1
PLAYER A -1 3
1
2
[Meerut, 1975]

AaA - i svs gi
02 Obtain an approximate solution of the following game (Eight iterations)
B

0 2
-31
A -1 2
12
2 0
|1
AnEs A0. . B o: svs
O3 Solve the following matrix game approximately (Ten iterations)
PLAYER B

PLAYERA 5 4 2-2
1 3 8 2

1 Sc., 1975]

AA 0. .Bo. 0, 0. o: sv
60 GAME THEORY i

whose payoft matrices

Davoff matrices are given below
given below.
(4-8)
Solve, by matrix method, the games

3 -3 4
J04 4 05
3
2 -11 -3

Anss (o. v - CEs ) (0. T. Tr): v= -

06 07 6 7 15
20 12 10
-3
Ans ( T o); v- -

A i. (0} v-
11.]

0S
3 0 0

o2 1
Ans 3: v= 1]
Solve, by the method of oddments, the games whose payoff matrices are given

beiow (9-11)
09 B | 10 B

A 51 3
|04
A 5 6 1

2 83

o12
A 2 0 1
2

12 Show that the games whose payoff matrices are given below
) B B
(n)
-1 3
1-1
A 3 5-3 A -6
3-2
6
2-2 8-5 2
cannot be solved by arithmetic method.