GAME THEORY and APPLICATIONS

Mikhail Ivanov Krastanov

 Games in normal form

There are given n-players. The set of all strategies (possible

actions) of the i-th player is denoted by Σi . Each player choose an
element σi ∈ Σi , i = 1, 2, . . . , n. Then the mathematical
expectation of the payoff function for the i-th player is given by
πi (σ1 , σ2 , . . . , σn ), i = 1, 2, . . . , n.

Definition
We say that the sets Σi , i = 1, 2, . . . , n, and the functions
πi : Σ1 ×, Σ2 × · · · × Σn → R, i = 1, 2, . . . , n, determine a game Γ
in normal form.

Remark
Each player want to maximize his payoff function, but this depends
on the choice of all players!
 Nash equilibriom

Definition
Let the game Γ is determined by the sets Σi , i = 1, 2, . . . , n, and
the functions πi : Σ1 ×, Σ2 × · · · × Σn → R, i = 1, 2, . . . , n. It is
said that the n-tuple

Definition
We say that the sets Σi , i = 1, 2, . . . , n, and the functions
πi : Σ1 ×, Σ2 × · · · × Σn → R, i = 1, 2, . . . , n, determine a game Γ
in normal form.

Remark
Each player want to maximize his payoff function, but this depends
on the choice of all players!
 Nash equilibrium

Definition
Given a game Γ, a strategy n-tuple (σ̄1 , σ̄2 , . . . , σ̄n ) is said to be a
Nash equilibrium if for any index i = 1, 2, . . . , n, and any σi ∈ Σi ,
πi (σ̄1 , . . . , σ̄i−1 , σi , σ̄i+1 . . . , σ̄n ) ≤ π(σ̄1 , σ̄2 , . . . , σ̄n ).
In other words, a strategy n-tuple (σ̄1 , σ̄2 , . . . , σ̄n ) is said to be a
Nash equilibrium if no player has a reason to change his strategy,
assuming that none of the other players is going to change his
strategy.
 Zero-sum games

Definition
A game Γ is said to be zero-sum if
Σni=1 πi (σ1 , . . . , σi−1 , σi , σi+1 . . . , σn ) = 0
for any σi ∈ Σi , i = 1, . . . , n.

In general, a zero-sum games represents a closed system:

everything that someone wins must be lost by someone else.

Most parlor games are of the zero-sum type.

Two-person zero-sum games are called strictly competitive games.

 Two-person zero-sum games

Let (σ̄1 , σ̄2 ) ∈ Σ1 × Σ2 be a Nash equilibrium. Then

π1 (σ1 , σ̄2 ) ≤ π1 (σ̄1 , σ̄2 ) for each σ1 ∈ Σ1

and
π2 (σ̄1 , σ2 ) ≤ π2 (σ̄1 , σ̄2 ) for each σ2 ∈ Σ2 .

Because in this case π2 (σ1 , σ2 ) = −π1 (σ1 , σ2 ), we obtain that

π1 (σ1 , σ̄2 ) ≤ π1 (σ̄1 , σ̄2 ) ≤ π1 (σ̄1 , σ2 )

for each σ1 ∈ Σ1 and each σ2 ∈ Σ2 , i.e. (σ̄1 , σ̄2 ) is a saddle point

for the function π1 .
 Two-person zero-sum games
We have obtained that the first player want to maximize π1 (σ1 , σ̄2 )
over Σ1 , while the second player want to minimize π1 (σ̄1 , σ2 ) over
Σ2 .

Let us assume that first player is omniscient and he can guess

correctly the action σ2 ∈ Σ2 of the second player. Then the first
player will choose σ̄1 ∈ Σ1 so that

π1 (σ̄1 , σ2 ) = max π1 (σ1 , σ2 ).

σ1 ∈Σ1

For simplicity, we assume here and further that these “max” and
“min” exist.
But then it is natural the second player to choose σ̄2 ∈ Σ2 so that
 
max π1 (σ1 , σ̄2 ) = min max π1 (σ1 , σ2 )
σ1 ∈Σ1 σ2 ∈Σ2 σ1 ∈Σ1
 Two-person zero-sum games
Analogously it is natural for the first player to choose σ̄1 ∈ Σ1 so
that  
min π1 (σ̄1 , σ2 ) = max min π1 (σ1 , σ2 )
σ2 ∈Σ2 σ1 ∈Σ1 σ2 ∈Σ2

Definition
It is natural to define
 
“gain-floor” vI := max min π1 (σ1 , σ2 )
σ1 ∈Σ1 σ2 ∈Σ2

and  
“loss-ceiling” vII := min max π1 (σ1 , σ2 ) .
σ2 ∈Σ2 σ1 ∈Σ1

The meaning of vI and vII is: the first player should not win less
than vI and the second player should not loss more than vII .
 Two-person zero-sum games

Lemma 1.
We have that vI ≤ vII .

Proof:
Let us fix arbitrary elements σ̂1 ∈ Σ1 and σ̂2 ∈ Σ2 . Clearly, we have
that

max π1 (σ1 , σ̂2 ) ≥ π1 (σ̂1 , σ̂2 ) ≥ min π1 (σ̂1 , σ2 ), i.e.

σ1 ∈Σ1 σ2 ∈Σ2

max π1 (σ1 , σ̂2 ) ≥ min π1 (σ̂1 , σ2 ),

σ1 ∈Σ1 σ2 ∈Σ2

Since the right-hand side does not depend on σ̄1 and the left-hand
side does not depend on σ̄2 , we obtain that
 Two-person zero-sum games

Proof: (continuation)
   
min max π1 (σ1 , σ̂2 ) ≥ max min π1 (σ̂1 , σ2 ) ,
σ2 ∈Σ2 σ1 ∈Σ1 σ1 ∈Σ1 σ2 ∈Σ2

i.e. vII ≥ vI . This completes the proof.

Definition.
If vI = vII , the common number v := vI = vII is said to be value of
the game Γ.
 Two-person zero-sum games

Lemma 2.
A game Γ has a value, if and only if, it has a saddle point.

Proof. (sufficiency)
Let the game Γ has a saddle point, i.e. there exists
(σ̄1 , σ̄2 ) ∈ Σ1 × Σ2 so that

π1 (σ1 , σ̄2 ) ≤ π1 (σ̄1 , σ̄2 ) ≤ π1 (σ̄1 , σ2 )

for each σ1 ∈ Σ1 and each σ2 ∈ Σ2 . Then

max π1 (σ1 , σ̄2 ) ≤ π1 (σ̄1 , σ̄2 ) ≤ min π1 (σ̄1 , σ2 )

σ1 ∈Σ1 σ2 ∈Σ2

and hence
 Two-person zero-sum games

Proof. (sufficiency: continuation)

   
min max π1 (σ1 , σ̄2 ) ≤ max min π1 (σ̄1 , σ2 )
σ2 ∈Σ2 σ1 ∈Σ1 σ1 ∈Σ1 σ2 ∈Σ2

i.e. vII ≤ vI . We have already prove that vI ≤ vII . Hence vI = vII

and the game Γ has a value. This completes the proof of the
sufficiency.

Proof. (necessity)
Let the game Γ has a value
   
v = min max π1 (σ1 , σ̄2 ) ≤ max min π1 (σ̄1 , σ2 ) .
σ2 ∈Σ2 σ1 ∈Σ1 σ1 ∈Σ1 σ2 ∈Σ2
 Two-person zero-sum games

Proof. (necessity: continuation)

According our simplifying assumption, there exist σ̄1 ∈ Σ1 and
σ̄2 ∈ Σ2 so that

v = min π1 (σ̄1 , σ2 ) = max π1 (σ1 , σ̄2 ),

σ2 ∈Σ2 σ1 ∈Σ1

and hence

π1 (σ1 , σ̄2 ) ≤ v ≤ π1 (σ̄1 , σ2 ) for each σ1 ∈ Σ1 and each σ2 ∈ Σ2 .

It follows from here that

π1 (σ1 , σ̄2 ) ≤ v = π1 (σ̄1 , σ̄2 ) ≤ π1 (σ̄1 , σ2 )

for each σ1 ∈ Σ1 and each σ2 ∈ Σ2 , i.e. (σ̄1 , σ̄2 ) is a saddle point

for the function π1 . This completes the proof of the necessity.
 Two-person zero-sum games
Definition
A mixed strategy is a probability distribution on the set of his pure
strategies.

Let the first player has m pure strategies. Then the set X of its
mixed strategies consists of all vectors x = (x1 , x2 , . . . , xm ) whose
components satisfy
m
X
xi = 1 and xi ≥ 0, i = 1, . . . , m.
i=1

Analogously, if the second player has n pure strategies. Then the

set Y of its mixed strategies consists of all vectors
y = (y1 , y2 , . . . , yn ) whose components satisfy
n
X
yj = 1 and yj ≥ 0, j = 1, . . . , n.
j=1
 Two-person zero-sum games

Let us assume that players I and II are playing a zero-sum game

determined by the matrix A. If I chooses the mixed strategy x, and
II chooses y , then the expected payoff is
   
Xm X n m
X n
X n
X n
X
P(x, y ) = aij xi yj = xi  aij yj  = yj  aij xi 
i=1 j=1 i=1 j=1 j=1 j=1

m
X n
X
i.e. P(x, y ) = xi P(i, y ) = yj P(x, j).
i=1 j=1

Here i in the expression P(i, y ) denotes the i-th pure strategy of I.

Analogously, j in the expression P(x, j) denotes the j-th pure
strategy of II.
 Two-person zero-sum games

Lemma 3.
Let (x1 , y1 ) ∈ X × Y 3 (x2 , y2 ) be saddle points of the
payoff-function P. Then (x1 , y2 ) and (x2 , y1 ) are also saddle points
ofthe payoff-function P.

Proof.
The definition of a saddle point implies that

P(x2 , y1 ) ≤ P(x1 , y1 ) ≤ P(x2 , y2 )

and
P(x1 , y2 ) ≤ P(x2 , y2 ) ≤ P(x2 , y1 ).
These inequalities imply that

P(x2 , y1 ) = P(x1 , y1 ) = P(x2 , y2 ) = P(x2 , y1 ).

 Two-person zero-sum games

Proof. (continuation)
Let x and y be arbitrary mixed strategies from X and Y ,
respectively. Then we have that

P(x, y1 ) ≤ P(x1 , y1 ) = P(x2 , y1 ) = P(x2 , y2 ) ≤ P(x2 , y ),

i.e. (x1 , y2 ) is a also saddle point of the payoff-function P.

Analogously, one can prove that (x2 , y1 ) is also a saddle point of
the payoff-function P.

Lemma 4.
Let (x̄, ȳ ) ∈ X × Y be a saddle point of the payoff-function P. If
x̄i0 > 0, then P(i0 , ȳ ) = P(x̄, ȳ ). Also, if ȳj0 > 0, then
P(x̄, j0 ) = P(x̄, ȳ ).
 Two-person zero-sum games

Proof.
Let us assume the contrary, i.e. P(i0 , ȳ ) < P(x̄, ȳ ). Since
(x̄, ȳ ) ∈ X × Y is a saddle point of the payoff-function P, we have
that

P(i, ȳ ) ≤ P(x̄, ȳ ) for each i = 1, 2, . . . , m, with i 6= i0 .

Multiplying the both sides of these inequalities by x̄i , we obtain that

xi P(i, ȳ ) ≤ xi P(x̄, ȳ ).

After adding of all these m inequalities, we obtain that

m
X m
X
x̄i P(i, ȳ ) < x̄i P(x̄, ȳ ), i.e. P(x̄, ȳ ) < P(x̄, ȳ ).
i=1 i=1
 Two-person zero-sum games

The obtained contradiction shows that our assumption is wrong,

and hence P(i0 , ȳ ) = P(x̄, ȳ ). Analogously, one can prove that
P(x̄, j0 ) = P(x̄, ȳ ). This completes the proof.

Remark.
Let (x̄, ȳ ) ∈ X × Y be a saddle point of the payoff-function P.
Then P(x̄, ȳ ) is a value of the corresponding zero-sum game.

Lemma 5.
Let (x̄, ȳ ) ∈ X × Y be a saddle point of the payoff-function P
determined by the matrix A = (aij )m×n . Then (x̄, ȳ ) is a saddle
point for the zero-sum game determined by the matrix
B = (bij )m×n with bij = αaij + β, where α > 0. Moreover,
αP(x̄, ȳ ) + β is the value of the zero-sum game determined by the
matrix B.
 Two-person zero-sum games
Proof.
Let us denote by PA and PB the payoff-functions of the zero-sum
games generated by the matrices A and B, respectively. Then
m X
X n m X
X n
PB (x, y ) = bij xi yj = (αaij + β)xi yj =
i=1 j=1 i=1 j=1

m X
X n m
X n
X
=α aij + β xi yj = αPA (x, y ) + β
i=1 j=1 i=1 j=1

Because (x̄, ȳ ) ∈ X × Y is a saddle point of the payoff-function PA ,

the following inequalities hold true:

PA (x, ȳ ) ≤ PA (x̄, ȳ ) ≤ PA (x̄, y )

for each x ∈ X and y ∈ Y .

 Two-person zero-sum games
Proof of Lemma 5 (continuation).
These inequalities imply that

αPA (x, ȳ ) + β ≤ αPA (x̄, ȳ ) + β ≤ αPA (x̄, y ) + β,

i.e.
PB (x, ȳ ) ≤ PB (x̄, ȳ ) ≤ PB (x̄, y )
for each x ∈ X and y ∈ Y . Hence (x̄, ȳ ) is a saddle point of the
payoff-function PB and PB (x̄, ȳ ) is its value.

Remark.
Let A = (aij )m×n be an arbitrary matrix and
M := 1 + max |aij |. We set B = (bij )m×n with
1≤i≤m,1≤j≤n
bij = aij + M, i = 1, . . . , m, j = 1, . . . , n. Then the all elements of
the matrix B are positive, and hence, its value is also positive.
 Two-person zero-sum games
Lemma 6.
Let (x̄, ȳ ) ∈ X × Y be a saddle point of the payoff-function P. If
P(i0 , ȳ ) < P(x̄, ȳ ), then x̄i0 = 0. Also, if P(x̄, j0 ) > P(x̄, ȳ ), then
ȳj0 = 0.

Proof.
Let us assume that x̄i0 > 0. According to Lemma 4 we obtain that
P(i0 , ȳ ) = P(x̄, ȳ ). This contradiction shows that x̄i0 = 0.
Analogously, one can prove that ȳj0 = 0.

Proposition 1.
Let S be a closed subset of R n and let x 6∈ S. Then there exists
point y ∈ S so that

0 < kx − y k = min(kx − sk : s ∈ S).

 Two-person zero-sum games

Proof of Proposition 1.
Let d := inf(kx − sk : s ∈ S) and let n be an arbitrary positive
1
integer. Because d + > d there exists point yn ∈ S so that
n
1
d ≤ kx − yn k < d + . Because
n
kyn k ≤ kx − yn k + kxk ≤ d + 1 + kxk,

the sequence {yn }∞

n=1 is bounded, and hence there exists a
convergent subsequence {ynk }∞
k=1 tending to y as k tends to
1
infinity. Taking a limit in the inequalities d ≤ kx − ynk k < d +
nk
as k → ∞, we obtain that d ≤ kx − y k ≤ d. This completes the
proof.
 Two-person zero-sum games
n
X
We denote by ha, bi := ai bi the scalar product of two vectors
i=1
a = (a1 , a2 , . . . , an ) and b = (b1 , b2 , . . . , bn ).

Separability theorem:
Let S be a closed convex subset of R n and let the point x does not
belong to S. Then there exist a non-zero vector p and a real
number q such that hp, xi = q and hp, si > q for each s ∈ S.

Proof.
Let y ∈ S be a point of S such that kx − y k =
= min(kx − sk : s ∈ S). We set p := y − x and q := hp, xi.
Then

hp, y i − q = hp, y i − hp, xi = hp, y − xi = kpk2 > 0.

 Two-person zero-sum games

Proof (continuation).
Let us assume that there exists a point s ∈ S so that hp, si ≤ q.
Then
r := hp, s − y i = hp, si − hp, y i < q − q = 0.
For each ε ∈ (0, 1) we set sε := (1 − ε)y + εs. The convexity of S
implies that xε ∈ S.
On can directly checked that

kx − xε k2 = k(x − y ) + ε(s − y )k2 =

kx − y k2 + 2ε hy − x, s − y i + ε2 ks − y k2 =
= kx − y k2 + ε 2r + εks − y k2 < kx − y k2

for each sufficiently small ε > 0 (because r < 0). The obtained
contradiction completes the proof.
 Two-person zero-sum games

Theorem of the Alternative for matrices

Let A = (aij )m×n be an arbitrary real matrix. Then either (i) or (ii)
must hold:
(i) The point 0 ∈ R m belongs to the convex hull C of the vectors
a1 , a2 , . . . , an , e1 , e2 , . . . , em , where aj := (a1j , a2j , . . . , amj ),
j = 1, 2, . . . , n, and e1 = (1, 0, . . . , 0), e2 = (0, 1, 0, . . . , 0), . . . ,
em = (0, 0, . . . , 1).
Xm
(ii) There exists a vector p = (p1 , p2 , . . . , pm ) such that pi = 1
i=1
and pi > 0 for each i = 1, 2, . . . , m, and hp, aj i > 0 for each
j = 1, 2, . . . , n.
 Two-person zero-sum games

Proof.
Let us assume that 0 6∈ C . Applying the Separability theorem, we
obtain that there exist a non-zero vector p and a number q such
that q = hp, 0i = 0, pi = hp, ei i > 0 for each i = 1, 2, . . . , m, and
hp, aj i > 0 for each j = 1, 2, . . . , n, i.e. (ii) holds true.

The min-max theorem (von Neuman and Morgenstern)

Let A = (aij )m×n be an arbitrary real matrix. Then the zero-sum
game determined by the matrix A has a value, i.e. vIA = vIIA .
 Two-person zero-sum games

Proof of the min-max theorem .

According to the Theorem of the Alternative for matrices, either (i)
or (ii) must hold.
If (i) is fulfilled, then there exists α1 ≥ 0, α2 ≥ 0, . . . , αn ≥ 0 and
β1 ≥ 0, β2 ≥ 0, . . . , βm ≥ 0 such that
n
X m
X
αj + βi = 1
j=1 i=1

and
n
X m
X
αj aj + βi ei = 0,
j=1 i=1
 Two-person zero-sum games

Proof (continuation)
i.e.
n
X
αj aij + βi = 0, i = 1, 2, . . . , m.
j=1

If all αj = 0, j = 1, 2, . . . , n, then all βi = 0, i = 1, 2, . . . , m, and

hence their sum can not be qual to 1. The obtained contradiction
shows that at least one αj must be positive. We set
αj
ȳj := Pn ≥ 0.
k=1 αk
 Two-person zero-sum games

Proof (continuation)
n
X
Clearly, each ȳj , j = 1, 2, . . . , n, is well defined, ȳj = 1 and
j=1

n
X
αj aij = −βi ≤ 0, i = 1, 2, . . . , m, i.e. P A (i, ȳ ) ≤ 0.
j=1

This implies that for each mixed strategy x of the first player
m
X
P A (x, ȳ ) = xi P A (i, ȳ ) ≤ 0, and hence max P A (x, ȳ ) ≤ 0.
x∈X
i=1

Thus we obtain that vII = min max P A (x, y ) ≤ 0.

y ∈Y x∈X
 Two-person zero-sum games

Proof (continuation)
Suppose, instead, that (ii) holds true, i.e. there exists a vector
Xm
x̄ = (x̄1 , x̄2 , . . . , x̄m ) such that x̄i = 1 and x̄i > 0 for each
i=1
i = 1, 2, . . . , m, and m
P
i=1 x̄i aij = hx̄, aj i > 0 for each
j = 1, 2, . . . , n. Then x̄ can be considered as a mixed strategy of
the I player and P A (x̄, j) > 0. This implies that for each mixed
strategy y of the second player
n
X
P A (x̄, y ) = yj P A (x̄, j) > 0, and hence min P A (x̄, y ) > 0.
y ∈Y
j=1

Thus we obtain that vI = max min P A (x, y ) > 0.

x∈X y ∈Y
 Two-person zero-sum games

Proof (continuation)
We know that vIA ≤ vIIA . Let us assume that vIA < vIIA . We consider
v A + vIIA
the matrix B := (bij )m×n with bij := aij − I . Then
2
v A + vIIA v A − vIIA
vIB = vIA − I = I < 0. Analogously,
2 2
A
v + vIIA A
v − vII A
vIIB = vIIA − I =− I > 0. But this is impossible. The
2 2
obtained contradiction shows that vIA = vIIA and completes the
proof of the Min-max theorem.
 Two-person zero-sum games
Lemma 6.
Let x̄ and ȳ be mixed strategies of players I and II, and v be a real
number such that

P(i, ȳ ) ≤ v ≤ P(x̄, j), i = 1, 2, . . . , m, j = 1, 2, . . . , n.

Then v is the value of the game and the couple (x̄, ȳ ) - a Nash
equilibrium.
Proof.
Let x and y are arbitrary mixed strategies of players I and II. Then
one can check that m m
X X
P(x, ȳ ) = xi P(i, ȳ ) ≤ xi v = v =
i=1 i=1
n
X n
X
= yj v = yj P(x̄, j) = P(x̄, y ).
j=1 j=1
 Two-person zero-sum games

Without loss of generality, we may think that the value v of a game

determined by the matrix A = (aij )m×n is positive (for, example, if
all elements aij > 0). Let v̄x and x̄ (v̄y and ȳ ) be solutions of the
following linear problems

vy → min vx → max
P(i, y ) ≤ vy , i = 1, 2, . . . , m P(x, j) ≥ vx , j = 1, 2, . . . , n
Xn m
X
yj = 1 xi = 1
j=1 i=1
yj ≥ 0, j = 1, 2, . . . , n xi ≥ 0, i = 1, 2, . . . , m

These problems can be written as follows

 Two-person zero-sum games

vy → min vx → max
Xn m
X
aij yj ≤ vy , i = 1, 2, . . . , m aij xi ≥ vx , j = 1, 2, . . . , n
j=1 i=1
n
X m
X
yj = 1 xi = 1
j=1 i=1
yj ≥ 0, j = 1, 2, . . . , n xi ≥ 0, i = 1, 2, . . . , m

These two linear problems are dual. If (vx , x̄) and (vy , ȳ ) are the
solutions of the first and second problem, respectively, then
vx = vy = v is the value of the game and the couple (x̄, ȳ ) is a
Nash equilibrium.
 Two-person zero-sum games

If we set pi = xi /v , i = 1, 2, . . . , m, and qj = yj /v , j = 1, 2, . . . , n,
then we can write the above problems as follows
m
m X
X pi → min
qj → max
i=1
i=1 m
n X
X aij pi ≥ 1, j = 1, 2, . . . , n
aij qj ≤ 1, i = 1, 2, . . . , m
i=1
j=1
qj ≥ 0, j = 1, 2, . . . , n
pi ≥ 0, i = 1, 2, . . . , m
 Two-person zero-sum games
Zero-sum games determined by a matrix of type 2 × 2
Let us consider a game determined by the following matrix
 
a b
c d

Let us assume that this game has no Nash equilibrium in pure

strategies. Let (x̄, ȳ ) be a Nash equilibrium with x̄ = (1, 0) and
ȳ = (ȳ1 , ȳ2 ), where ȳ1 > 0, ȳ2 > 0 and ȳ1 + ȳ2 = 1. Then
P(x, ȳ ) ≤ P(x̄, ȳ ) ≤ P(x̄, y ). Clearly, v = P(x̄, ȳ ) = aȳ1 + b ȳ2 .
We set y = (1, 0) and y = (0, 1), and obtain that
aȳ1 + b ȳ2 ≤ min(a, b). This implies that a = b = v . Also, we have
that v ≥ P(2, ȳ ) = c ȳ1 + d ȳ2 ≥ min(c, d). If c = min(c, d), we
obtain that ((1.0), (1, 0)) is a Nash equilibrium in pure strategies. If
d = min(c, d), we obtain that ((1.0), (0, 1)) is a Nash equilibrium
in pure strategies.
 Two-person zero-sum games

But, according to our assumption, the game has no Nash

equilibrium in pure strategies. Hence the assumption that (x̄, ȳ ) be
a Nash equilibrium with x̄ = (1, 0) and ȳ = (ȳ1 , ȳ2 ), where ȳ1 > 0,
ȳ2 > 0 and ȳ1 + ȳ2 = 1, is not possible. Analogously it can be
proved that it is not possible (x̄, ȳ ) to be a Nash equilibrium with
x̄ = (0, 1) and ȳ = (ȳ1 , ȳ2 ), where ȳ1 > 0, ȳ2 > 0 and ȳ1 + ȳ2 = 1.
This shows that whenever a zero-sum game has no a Nash
equilibrium in pure strategies, then both components of the Nash
equilibrium are mixed strategies.
 Two-person zero-sum games

Reducing zero-sum games

Let us consider a zero-sum game determined by the matrix
A = (aij )m×n . Let the i0 -pure strategy of I is dominated by a
convex combination of the remainder pure strategies of the I player,
i.e.
Xm
ai0 j ≤ αi ai,j for each j = 1, 2, . . . , n,
i=1,i6=i0

where all number αi are nonnegative and

m
X
αi = 1.
i=1,i6=i0
 Two-person zero-sum games

We denote by A0 the matrix obtained from the matrix A be deleting

the i0 -raw. Let (x̄ 0 , ȳ 0 ) with x̄ 0 = (x̄10 , x̄20 , . . . , x̄i00 −1 , x̄i00 +1 , . . . , x̄m
0 )

be a Nash equilibrium for the game determined by the matrix A0 .

Then (x̄, ȳ 0 ) is a Nash equilibrium for the game determined by the
matrix A, where x̄ = (x̄10 , x̄20 , . . . , x̄i00 −1 , 0, x̄i00 +1 , . . . , x̄m 0 ).

Because (x̄ 0 , ȳ 0 ) is a Nash equilibrium for the game determined by

the matrix A0 , the following inequalities hold true:
0 0 0
P A (x 0 , ȳ 0 ) ≤ v = P A (x̄ 0 , ȳ 0 ) ≤ P A (x̄ 0 , y 0 ), for x 0 ∈ X and y ∈ Y ,

where
m n
0
X X
P A (x 0 , y 0 ) = aij xi yj .
i=1,i6=i0 j=1
 Two-person zero-sum games

In particular, we have that

0 0
P A (i, ȳ 0 ) ≤ v ≤ P A (x̄ 0 , j), i = 1, . . . , i0 −1, i0 +1, . . . , m, j = 1, . . . , n.

i.e.
n
X
aij ȳj ≤ v, i = 1, . . . , i0 − 1, i0 + 1, . . . , m,
j=1
m
X (1)
aij x̄i ≥ v , j = 1, . . . , n
i=1,i6=i0

Since the vector α := (α1 , α2 , . . . , αi0 −1 , αi0 +1 , . . . , αm ) can be

0
considered as a mixed strategy of I, we have that P A (α, ȳ 0 ) ≤ v ,
 Two-person zero-sum games

i.e.
m
X n
X
aij αi ȳj ≤ v .
i=1,i6=i0 j=1

From here (taking into account that the i0 -pure strategy is

dominated by a convex combination of the remainder pure
strategies of the I player), it follows that
n
X n
X m
X
ai0 j ȳj ≤ aij αi ȳj ≤ v .
j=1 j=1 i=1,i6=i0

Applying this inequality and (1), we complete the proof taking into
account Lemma 6.
 Two-person zero-sum games

Analogously, let us assume that the j0 -pure strategy of II is

dominated by a convex combination of the remainder pure
strategies of the II player, i.e.
n
X
aij0 ≥ βj ai,j for each i = 1, 2, . . . , m,
j=1,j6=j0

where all number βj are nonnegative and

n
X
βj = 1.
j=1,j6=j0
 Two-person zero-sum games

We denote by A0 the matrix obtained from the matrix A be deleting

the j0 -column. Let (x̄ 0 , ȳ 0 ) with
ȳ 0 = (ȳ10 , ȳ20 , . . . , ȳj00 −1 , ȳj00 +1 , . . . , ȳn0 ) be a Nash equilibrium for the
game determined by the matrix A0 . Then (x̄, ȳ 0 ) is a Nash
equilibrium for the game determined by the matrix A, where
ȳ 0 = (ȳ10 , ȳ20 , . . . , ȳj00 −1 , 0, ȳj00 +1 , . . . , ȳn0 ).
 Two-person zero-sum games

2 × n and m × 2 games
We consider first a zero-sum game determined by a the 2 × n -
matrix A. If v is its value, then
2 X
X n 2 X
X n
v = max min aij xi yj = min max aij xi yj .
x∈X y ∈Y y ∈Y x∈X
i=1 j=1 i=1 j=1

Clearly,
2 X
X n
max min aij xi ≥ v .
x∈X j=1,2,...,n
i=1 j=1
 Two-person zero-sum games

On the other hand

X 2
n X n
X 2
X
aij xi yj = yj aij xi ≥
j=1 i=1 j=1 i=1

n
X 2
X 2
X
≥ yj min aij xi = min aij xi ,
j=1,2,...,n j=1,2,...,n
j=1 i=1 i=1

From here, it follows

2
n X
X 2
X
min aij xi yj ≥ min aij xi ,
y ∈Y j=1,2,...,n
j=1 i=1 i=1

and hence
 Two-person zero-sum games

2 X
X n 2
X
v = max min aij xi yj ≥ max min aij xi .
x∈X y ∈Y x∈X j=1,2,...,n
i=1 j=1 i=1

In such a way, we obtain that

2
X
v = max min aij xi .
x∈X j=1,2,...,n
i=1

Since X = {(x1 , x2 ) : x1 = α, x2 = 1 − α, α ∈ [0, 1]}, we obtain

that v is the maximum (with respect to α) of the minimum of n
linear functions (depending on α). So we can plot these function
and to find this maximum graphically.
 Two-person zero-sum games

2 × n and m × 2 games
We consider next a zero-sum game determined by a the m × 2 -
matrix A. If v is its value, then
2
X
v = min max aij yj .
y ∈Y i=1,2,...,m
j=1

Since Y {(y1 , y2 ) : y1 = β, y2 = 1 − β, β ∈ [0, 1]}, we obtain that

v is the minimum (with respect to β) of the maximum of m linear
functions (depending on β). So we can plot these function and to
find this minimum graphically.
 Two-person zero-sum games

Symmetric games
A square matrix A = (aij )n×n is said to be skew-symmetric if
aji = −aij for all i, j = 1, . . . , n. A zero-sum game is said to be
symmetric if its matrix is skew-symmetric.

Theorem.
The value of a symmetric game is zero. If (x̄, ȳ ) is a Nash
equilibrium, then (x̄, x̄) and (ȳ , ȳ ) are also Nash equilibriums.

Proof.
Because the matrix A is skew-symmetric, AT = −A. Let x be an
arbitrary mixed strategy. Then x T Ax = (x T Ax)T = x T AT x =
= −x T Ax, and hence x T Ax = 0.
 Two-person zero-sum games

Proof (continuation.
Then
0 ≤ max min x T Ay = v = min max x T Ay ≤ 0
x∈X y ∈Y y ∈Y x∈X

which implies that v = 0.

Let j be an arbitrary index. Then P(x̄, j) ≥ 0, i.e.
n
X n
X
aij x̄i ≥ 0, i.e. − aji x̄i ≥ 0, i.e. (2)
i=1 i=1

n
X
aij x̄j ≤ 0, i.e. P(i, x̄) ≤ 0.
j=1

The last inequality, (2) and Lemma 6 complete the proof.

 Two-person zero-sum games

Theorem.
A symmetric game has no a Nash equilibrium in pure strategies, if
and only if one of the following assertions holds true:
(i) each column has at least one positive element;
(ii) each raw has at least one negative element.

Proof. (sufficiency)
We shall consider only (i). The case (ii) can be studied in the same
way. Let (k, k) be a Nash equilibrium in pure strategies and let
aik > 0. Then
0 < aik = P(i, k) ≤ v = 0.
The obtained contradiction shows that there is no Nash equilibrium
in pure strategies.
 Two-person zero-sum games

Proof. (necessity)
Let us assume that there is no Nash equilibrium in pure strategies
and there exists a column k such that aik ≤ 0 for each i = 1, 2, . . . ,
X n
n. This implies that aik xi ≤ 0 for each mixed strategy
i=1
n
X n
X
x = (x1 , x2 , . . . , xn ), i.e. −aki xi ≤ 0, and hence akj xj ≥ 0.
i=1 j=1
So, we obtain that P(x, k) ≤ 0 and P(k, x) ≥ 0. This means that
(k, k) is a Nash equilibrium in pure strategies.
 Two-person zero-sum games
Symmetric games
Let us consider the matrix
 
a11 a12 ... a1n
 a21 a22 ... a2n 
A=  ... ...

... ... 
am1 am2 ... amn

and let us define the following m + n + 1 × m + n + 1

skew-symmetric matrix:
 
Om×m Am×n −Im×1
B =  −AT n×m On×n In×1 
T
I1×m T
−I1×n O1×1
T
where Ok×k is a k × k zero matrix and I1×m is a vector-raw with m
components equal to 1.
 Two-person zero-sum games
Theorem.
Let the elements of the matrix A are positive. Let
((x̄, ȳ , λ), (x̄, ȳ , λ)) be a Nash equilibrium of the game determined
of the matrix B. Then
X m X n
(i) x̄i = ȳj = µ > 0;
i=1 j=1
(ii) the couple (x̄/µ, ȳ /µ) is a Nash equilibrium of the game
determined by the matrix A;
(iii) the value of the game determined by the matrix A is λ/µ.

Proof.
We have that
m
X n
X
x̄i + ȳj + λ = 1,
i=1 j=1

with x̄i ≥ 0, i = 1, . . . , m, ȳj ≥ 0, j = 1, . . . , n, and λ ≥ 0.

 Two-person zero-sum games

Proof (continuation).
Since ((x̄, ȳ , λ), (x̄, ȳ , λ)) is a Nash equilibrium of the game
determined of the matrix B, the following inequalities hold true:
n
X
aij ȳj − λ ≤ 0, i = 1, . . . , m, (3)
j=1

m
X
− aij x̄i + λ ≤ 0, j = 1, . . . , n, (4)
i=1
m
X n
X
x̄i − ȳj ≤ 0. (5)
i=1 j=1
 Two-person zero-sum games

Let us assume that λ = 0. This contradicts to (3).

Let us assume that λ = 1. Then x̄ = 0 and ȳ = 0. But this
contradicts to (4).
Hence λ ∈ (0, 1). Then
m
X n
X
x̄i + ȳj = 1 − λ =: 2µ
i=1 j=1

and
m
X n
X m
X n
X
x̄i − ȳj = 0, i.e. x̄i = ȳj = µ.
i=1 j=1 i=1 j=1
 Two-person zero-sum games

Then
m n
X x̄i X ȳj
= =1
µ µ
i=1 j=1
n
X ȳj λ x̄i
aij ≤ , ≥ 0, i = 1, . . . , m,
µ µ µ
j=1
m
X x̄i λ ȳi
aij ≥ , ≥ 0, j = 1, . . . , n.
µ µ µ
i=1

Applying Lemma 6, we obtain that the couple (x̄/µ, ȳ /µ) is a Nash

equilibrium of the game determined by the matrix A.
 Two-person zero-sum games

Theorem.
Let the elements of the matrix A are positive. Let the couple (x̄, ȳ )
be a Nash equilibrium of the game determined by the matrix A with
value v . Then  
x̄ ȳ v
, , .
2+v 2+v 2+v
is a Nash equilibrium of the game determined of the matrix B.

Proof.
Since (x̄, ȳ ) is a Nash equilibrium of the game determined by the
matrix A, we have that
m
X n
X
x̄i = ȳj = 1
i=1 j=1
 Two-person zero-sum games

Proof (continuation).
n
X
aij ȳj ≤ v , x̄i ≥ 0, i = 1, . . . , m,
j=1
m
X
aij x̄i ≥ v , ȳj ≥ 0, j = 1, . . . , n.
i=1
 Two-person zero-sum games

Proof (continuation).
These inequalities imply that
n
X ȳj v
aij − ≤ 0, x̄i ≥ 0, i = 1, . . . , m,
v +2 v +2
j=1

m
X x̄i v
− aij + ≥ 0, ȳj ≥ 0, j = 1, . . . , n.
v +2 v +2
i=1

Applying Lemma 6, we obtain that the couple

x̄i ȳj v
( , , )
v +2 v +2 v +2
is a Nash equilibrium of the game determined by the matrix B.