Mixed strategies

G = (N , (Si) (i) , Si-finite Sato) (purel strategies

Mi : S- R -

Mixed strotegies for player i

Ei =
[5i :
Sitfan &vilsi) =
.
13

SitSi

·
i(si) -

probability of playing He pure

strategy Bi .
Mixed extension of a Game.

G =
(N (Si) (Milien).
, ,
Si is finite .

the mixed extension of G is T (N

,
=
Kilien (i)ieN
,

Ei-set of mixed strategies of player i

[ = [, XEzX--XEn.

Hi : -IR .

Ui(2) = &Mi (b) 2

, 181) G2(92) --
Un (en)
SES
-

: : 5 +R .

Example
52

E 1-g
Y Y

P A
2
, -

1 -

Uc (2 22) ,,
=
4 Pq + 1 P (1 g) +
. . .
-
2 . (1 p(q + 3(1 p)(1-g)
- -

= 3-
2p- g +
4pq .
-

Bilinear function .
Theorem . Ui is a multiline function n
in variable.

For
every ifN ,
Y Wi ,
Wie Ei and for every XE[O]

Hi (x5i + (1 -

x)aj , i) -
=
XUi(vi ,
2 -
i) +

(ix) Hi (wi -i) Kwiti.

,

Consequence 1 .
Ui(i -i) ,
= [0 : (si) Hi (si -i)
,

Sie Si

Consequence 2
. Hi is a continuous
function in the

extension to mixed strologies /for all finite strategic

games) .
Nash Equilibrium
mixed o is Nash equilibrium
A
strology profile a

if VIEN , VeeEi ,

Ui( )
*
= Mi(v :) ,

Example
12 1/2 2

H T

1/2 + 1, + -
t )
,
5
1/2 T -

111 .
-

U2(a ,
22) =
0 He (t val ,
=
.
0
Theorem .
Let G =
(N (Si) (i)
.
,
and let
T be

He mixed extension of G . &

*
ENE(T) iff

FiEN , VSitSi
,
Hil =i (si 5) .
*
,

Proof .

#) Follows from the definition

#) Suppose of
Satisfie Mile ) Hi (si ei)
*
>
,

F iEN , SiESi-

Then for each mixed Stroksy GitEi ,

Ui(ei , ) =
Esi(si) Mili ai) ,

SitSi

=>(i) Ni (e, w ) .

=
Mi(2 )
*
[Wilsi) = Mila
*
) .

Sit Si
Theorem [Nesh] ·

Every finite game has a

Nash equilibrium in mixed strotegies.

Zero-sum

Corollary [Minmox Theorem]. Every finite two

player game
has a value in mixed strategies
.

V= max min U(x2) = min max u ( (2)

,

eiEE , Gathe 22tE2 EE

,
 *
Theorem [The be Nash equilibrium
indifference principle] Let o a

and Si , : be two pure strategies of player . If

i

(Si) >0 and (5) >0 Iten

Ui(bi , ) = Ni ( ,
5)

Proof · Suppose Mi(Si ,

2) > Mili ,
e-z)

Define ori as
follows :

if titEBi Bi}
[t)
,

vict =
if ti = 5:
if ties :

Mi(vi 0-E) Milk )

*
To argue ,
>
 Milvis) Milti wil
=Es Tilti) ,

= [ silti) Hilti 2) + Wilsi) Milsi =)

, ,
e +

ti[si ,:3
,w
Mi
)
·
*
= E 5 (ti) Mi (ti =) ,
+
ti [Si ,:]

f (bi) + = (5)i (Si we))

*
(2 ,

= + (Si) Mi(si 5) + w(i) Mi (Si , ei)

+
,

↓
> + + of()ui( , wi)
W(ti) Hi (ti , ) Ui(
=
=
Strict dominance by a mixed strategy .

strictly do minded SitE: if

A
strategy BitSi is by
SiESi Milli bi)
,
> MilBi Si) ,

Example L R

112 T 2, 1 ,0
0

12 in 0, 1 2, 0

01 0, 2
B

B is
strictly dominated by the mixed strategy
[12(T) ,
/2 (M)·

Theorem . Let G (N (Silien

=
, ,
(Milien) .
if SitSi

is strictly dominated by a mixedStrategy WitEi ,

then in every (mixed) Nosh equilibrium in G

,

Si is chosen will probability Zero .

Proof. Let that
Si be a
pure strategy is

strictly dominated by ei

Let E where 5 (si) 30. To show o is not a NE

.

Suffices to show : E; is not a best

response
to i

Define Gi :

(Si)
&(ti) +
wi(Si) if ti =
si

· Hti)
(Si) Wiltil if tisi
=
·

Claims .

Fes Filtil =
.
1

S
=
iii(ti) + :(si) [Esivi (til]
Ei(til + (Si) Filti]) + ·
: (bil di (bi)
tiSi

= siFi (ti) + (i) Wilti) + (i) will

=
Esi(ti) + (i) [Eiesi-i(til]
Proof. Let that
Si be a
pure strategy is

strictly dominated by ei

Let E where 5 (si) 30. To show o is not a NE

.

Suffices to show : E; is not a best

response
to i

Define Gi :

(Si)
&(ti) +
wi(Si) if ti =
si

· Hti)
(Si) Wiltil if tisi
=
·

claims .
E 5 ; (bi) =
1 .

Sit Si

= i(ti) + :(si) [EsiVi(til] =

1

tiSi

Claim 2 GI is better to : (compared to

. a
response
i)

That is , :( E i) < Miloi , (i)

 Hi(i , i) = E Wiltil Mi (ti ,
i
tieSi

= [i(ti) Hi (ti, i) + Ei (Si) Hi (Si ,

Ei) .

tifsi

[
↓ + ;(si) Hi (wi , i)

=
Fi(ti) Hi (ti , w i) +; (i) Wi(si) Kilsi , i)

+ (bi) [ di (ti) Mi (ti Ei) .

,
tifsi

=
r(si) Hi /Si, i) + 25 ; (ti) Hilti & i) ,

tiSi

= [ (ti) Milti , i) =
Mi (wi , Ei)
tiESi
.