Professional Documents
Culture Documents
Game Theory With Mixed Strategy
Game Theory With Mixed Strategy
STRATEGIES
Structure
14.1 Introduction
qbjectives
14.5 Summary'
- Garncs with Mixed Stntegics
l4J INTRODUCTION
SO far you have studied the solutions of games with Pure strategies where
you could solve a game only when the maxim~nvalue Coi the maximizing
player was equal to the minimax value for the minimizing player. In other
words the value of the game could be found out only when there was a
saddle or equilibrium point, and that game was called a game with Pure
strategies. But when there is no saddle point in a game, then it is called a
game with mixed strategies. In order to solve such a game, each player
adopts the concept of chance move. Each player starts to play in a random
manner and in such a way that his average pay-off over a large number of
plays of the game should be optimum, even though he may loose more in
any individual play of the game. In other words, the player adopts a certain
strategy for sometime and then shifts over to another strategy and then after
sometime to third strategy and so on SO that the average pay-off over a large
number of plays of the game becomes optimum (i.e. maximum for the
maximizing player and the minimum for the minimizing player).
Objectives
If a player does not adhere to the same strategy throughout the game but
, mixes the strategies in such a way that his opponent is not in a position to
guess it and moreover he himself does not know what his next move would
be, then the gaine is said to be played with mixed strategies. In other words,
a game with mixed strategies is the game where the saddle point does not
j exist.
I Player Y
I
Player X
."
am1 am2 am
. .,
Here the player X has m strategies (1,2,. m) and the player Y has n
strategies (1,2,.,., n). if the above matrix has a saddle point
i.e. Max Min a, = Min Max I+ = &
i j j i
then according to what you have learnt in unit 13, the best strategy for
player X is the pure strategy h, the best strategy for player Y is the pure
strategy k and the value of the game is q,k.
then there are no optimal pure strategies for the players and we have to
define a Mixed Strategy.
Mixed Strategy for player X Xis a vector x = (x,, ..., &) of m non-negative
m
real numbers that satisfies the condition x x i = 1 where
i=1
* may be
interpreted as the probability with which player X chooses the su-ategy i.
Similarly a Mixed Strategy for player Y is a vector y = (y,, . .., y,) of n non-
n
negative real numbers satisfying the condition yj = 1 where y may be
j-1
interpreted as the probability with which player y chooses the strategy j.
You have seen that the set of mixed strategies includes the set of pure
-
strategies. In fact, a pure strategy k for player X is a special cases of a
mixed strategy x = (0, 0, ..., I , ..., 0) where xk 1 and all other xi= 0. The
set of all mixed strategies for player X will be denoted by S and that of all,
mixed strategies for player Y by T. In other words I
I
Games witti Mixed
Strategies
You have seen in Unit 13 that for a game with pay-off matrix A = (aij), if
Player X adopts a pure strategy i and Player Y adopts a pure strategy j,
then outcome is aij. Player X wins if aij is positive and loses if aij is
negative. In the later case, we also say that player Y wins.
On the other hand if Player X uses the inixed strategy x = (x,,..., xi,... xm)
and player Y uses the mixed strategy y = (y ,,...yj,...,y,,), then we need to
know a way of evaluating the outcome.
For this we use the concept of 'expected value' from probability theory. In
probability theory the expected value of an event is defined to be the sum
of the values of each possible outcomes of the event times the probability
that the outcome occurs. Therefore if Player X chooses the mixed strategy
x = (x,,...,xi,...,x,,,) and player Y chooses the mixed strategy y =
(y,,...,yj,...,yn)then the outcome a,. will occur with probability xi y, since this
is the probability ofboth the Plaier X choosing strategy i and the Player Y
choosing strategy j. Thus the expected value for the game is the sum of all
the products xi aijyj, that is, the sum
I
i.e. (x~,.. ., xm)
... ... ... . ., I . .
m
Then the expected value = z x i a i j
i=l
You will miow see that we will follow the same steps in defining the value
of the game in the case of mixed strategies adopted by the two players as
we did in Unit 13 ir the case of puse strategies. Consider a game with pay-
off matrix A of order nn x n.
For each mixed strategy x = (x,,...,x ), Player X must determine the worst ,
possible outcome of the game allodGlg that Player Y also has available
inixed strategies y = (y,,...,yn).This would be then Min xAyT.
Y
This rnininluln represents expected minim~ungain of Player X il' he were
to choose strategy x. Since he wants to inslximize his gain, therefore Player
X should e c k a strategy that attains lhe maximum of these minima, that is,
strategy x* suc1-1 that
At this stage it is possible to define the strategic saddle point o f the game
with inixed strategy.
max min
If X E S ~ E T
xAyt - mill max xAyt = x*Ay*'
YET~ E S
then (x",y*) is called tlie strategic saddle point of the game wlierek*and y.
define the optimal strategies and v = x * ~ is~ *tlie' value of the game.
In this unit, we sllall discuss the algebraic method while the other two
methods will be dealt with in units 15 and-14.
Games with Mixed
14.3 ALGEBRAIC M E T A ~ D Strategies
Player B
Bl B2
I SOLUTION : I11 the given matrix, since there is no one value which is
II
smallest value in its row and largest in the column, there is no saddle point
I in it, and the game is not that of the Pure strategies i.e. noile of the players
A or B would play the same strategy tlkoughout the galne but they would
resort to what is know11 as inixed strategies i.e. the Player A would play
each of his rows A, and A, for a certaiil portion of time and the Player B
would play each of his columns 13, and I3, for a certain portiun of time.
-
Accordingly, if the Player A plays the stritegy A, for certain time t,, tllen
he would play his second strategy A, for the remaining tilnc (1 - t,).
Similarly if the Player B plays the strategy B, for time t,, then he would
play his second strategy B, for the remaining time (1 - t,). (The total time
for the play being taken as unity). This could also be interpreted in terms
of the probabilities as : If the Player A selects strategy A, with a
probability p then he would play slrategy A, with a probability (1 - p),
I
since the sum of the probabililies is one. Similarly, if the Player R selects
I
' (his strategy. B, with probability q, then he would play his strategy B, with
I a probability (1 - q). I11 the given pay-off matrix, Pet thc Player A select the
/ strategy A, with a probability p, then he would select the strategy A, wit11 a
probability (1 -p). Suppose now that the Player B selects the strategy B,
I
I
and adheres to it throughout the game, then the expected pay-off or
1 expected gain to Player A for this game would be the algebraic sum of the
expe-:ed values of the two events, i.e.,
But if the Player B selects the strategy B, and adheres to it throughout the
game, then the expected pay-off or expected gain to Player A for the game
would be
. 5p+15(1-p) .
, Now in order that the Player A may be indifferent to the strategy B selects,
the optimal plan for the player A should be such that
15~+10=15-10p
1
Hence the Player A would select the strategy A, with probability -and the
5
4
5 .
strategy A, with a probability -
and the expected pay-off or expected loss to Player B. when the Player A ,
adopts the strategy A2 throughout the game would be
and
7
r #
Hence the Player B would select the strategies B, and B, with probabilities Games with Mixed I
Strategies
2 3
- and - respectively.
5 5
The value of the game is detennined by substituting the value of p or q in
any of the expected value as given below:
ii) 5~.2+15x,8=13
Thus the value of the game is equal to the expected gain to the Player A or
the expected loss to Player B, Thus the value. of the game = v = 13.
Step I. Substract the smaller pay-off in each row from the larger one and the
'
smaller pay-off in each column from the larger one as
1 B
. Step TI. Interchange each of these pairs of subtracted numbers found in Step
I, as
Game Theory \ Step III. Put each of the interchanged numbers over the sum of the pair of
numbers, as
-
25 5 5
A [la IS]\ -
4
Step V. Value of the game can be determined from the view point, of the
Player by finding his expected value. For Player A - the expected value
(expected gain) is obtained by (i) or (ii) as
1 4
ii) (5 x -) + (15 x -)= 13
5 5
For Player I3 - The expected value (the expected loss) is obtaii~cdby (iii)
or (iv) as .
I Note : The short-cut method can be applied only when the pay-off matrix of
a game is of order 2 x 2 and there'is no saddle point in it.
4
1 ) I
24 .
EXAMPLE 2 : Solve the game wt~osepay-on matrix is given by Games with Mixed I
Strategies 1
I '
Step I. Subtracting the pay-offs,
I
Step III. Putting the pairs over their sum and on simplification
10 7
Therefor., the strategies (A,, A,) for Player A are (- -) and for the
I 17 ' 17
7
Player B, the strategies (B,, B,) are (- -
10
) respectively.
17' 17
EXERCISE 2 : Assuming the following pay-off matrix for A and B and the
amounts being the utilities gained by A and lost by B
Player B I
B, B2
Player A
ii) If both A and B follow mixed strategies, what is the value of the game?
. .
. . 1 , .1
' ', : show that E(x, y) = 1 - 2x(y - -), where x and y are the probabiljries of A
2
and B for the strategies A, and B, respectively. Show also that 111 the , )
solution of the games, Player A follows the pure strategy while the Player . j
B may adopt pure or mixed strategies. I
I I
I
If the pay-off matrix is not of order 2 x 2 and there is no saddle point in it,
then it can be solved with the graphical method (discussed in Unit 15). 1I /
However a game of m x 2 or 2 x n order can be solved by considering the i
I
game as k i n g sub-games, each of order 2 x 2, because the players X or Y
can choose to play any two of his rows or columns as the case may be. The
method of sub-games is illustrated in the following example :
I
I
(0 B (ii) B
.
A, -3 4 A, 2 -9
I
(iii) B (iv) B 1 II
,
BI B, B, 13, i
215 315 ,I
I
II(
Sub-game (iii).
Sub-game (iv).
I3
B I
B, B2
(6) (g)
I
65 Strategies
-
"
I
if we look at the values of all the six sub-games, we find that all of them
are negative i.e. the Player B wins and Player A looses in all the six sub-
games. Thus the Player A would like to play the safest and accordingly, he
would choose that sub-game which gives the maximum value of all the six I
1
values, -3.6, -3.87, -6, -4, -3.57 and -3.82. Accordingly Player A would I
play the sub-game (v) whose value is -3.57 and the strategies for A and B
would be
(Q 67 0 7 and 1) (- -)
3 4
7 7
I
'EXERCISE 4 : Solve the game whose pay-off matrix is
Player Y
-
14.5 SUMMARY
1) a game without a saddle point and hence the game with mixed
strategies.
and v = -2517.
I
, I ) and (-
1 1 3 1
E2) The strategies for player A and B are (,T 4 '4
) and I
I
I v = 9.
!
E3) The strategy for player A is (0,1). The strategies for player B are
1 0 ( , :). The value v of the gake = 1.
B would play the sub game 2) that minimizes his losses. Strategies
for A and B are (3/5, 215) and (415, 0, 115).
UNIT 15 GRAPHICAL METHOD AND
DOMINANCE
Structure
15.1 Introduction
Objectives
15.6 Summary
15.7 Answers/M[ints/Solutions