Nash Equilibria

in continuous games
Nash equilibria in continuous games
• So far: games with a finite number of strategies.

• Next up: Games that allow for “many” strategies, i.e. games with a
continuous strategy set

• A continuous strategy set is an interval of real numbers – no gaps and

no missing numbers.
𝑠𝜖[𝑠, 𝑠] ⊆ ℝ

• Examples:
• Pricing game – prices charged by firms
• Effort game – effort devoted to a project
Nash Equilibria in Continuous Games
How to find Nash Equilibria?

1. Identify and write down the best-response correspondence

• Use calculus if possible
• Otherwise use reasoning

2. Look for intersections of the best-response correspondences

• Eliminate strategies for which there is no well-defined best-
response
• A graph may help.
Example 1: Cournot duopoly
• Two firms, 𝐼 = 1, 2

• Linear cost function 𝐶𝑖 𝑞𝑖 = 𝑐 ∙ 𝑞𝑖

• Firms simultaneously choose which quantity to supply

→ quantities 𝑞1 , 𝑞2 ∈ ℝ+
0

→ total supply 𝑄 = 𝑞1 + 𝑞2

• Inverse demand function: p(Q) = 10 − 𝑄

How to solve for a Nash equilibrium:

1. Identify and write down the best-response correspondence
2. Look for intersections of the best-response correspondences
Example 1: Cournot duopoly
1. Identify and write down the best-response correspondence.
What is firm i’s best response if firm j chooses 𝑞𝑗 ?

Suppose firm j choose 𝑞𝑗 , then firm i’s profit function is

Π𝑖 𝑞𝑖 , 𝑞𝑗 = 𝑞𝑖 ∙ 10 − 𝑞𝑖 − 𝑞𝑗

As a rational player, firm i wants to maximize its profit.

FOC:
𝜕Π𝑖
= 10 − 𝑞𝑗 − 2𝑞𝑖 = 0
𝜕𝑞𝑖
𝐵𝑅 1
𝑞𝑖 𝑞𝑗 = 𝑚𝑎𝑥 ∙ 10 − 𝑞𝑗 − 𝑐 , 0
2
The second order condition is satisfied since
𝜕 2 Π𝑖
= −2 < 0.
𝜕𝑞𝑖2
Example 1: Cournot duopoly
2. Look for intersections of the best-response correspondences
1
∙ 10 − 𝑞𝑗 − 𝑐 𝑖𝑓 𝑞𝑗 ≤ 10 − 𝑐
BR-correspondences are 𝑞𝑖𝐵𝑅 𝑞𝑗 = ൝2
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

We can solve this graphically:

Example 1: Cournot duopoly
2. Look for intersections of the best-response correspondences
1
∙ 10 − 𝑞𝑗 − 𝑐 𝑖𝑓 𝑞𝑗 ≤ 10 − 𝑐
BR-correspondences are 𝑞𝑖𝐵𝑅 𝑞𝑗 = ൝2
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

To find an intersection we plug 𝑞𝑖𝐵𝑅 𝑞𝑗 into 𝑞𝑗𝐵𝑅 𝑞𝑖 and then solve for 𝑞𝑖 :
1 10 − 𝑞𝑖 − 𝑐
𝑞𝑖 = ⋅ 10 − −𝑐
2 2
∗ 10 − 𝑐
⇒ 𝑞𝑖 =
3
10−𝑐 10−𝑐
Since this is a symmetric game, we obtain that 𝑞1∗ , 𝑞2∗ = ( 3
, 3 ) is the
Nash equilibrium
Mixed Strategies
Mixed Strategies
• So far, we assumed that players make their choices with certainty…

A pure strategy for player i specifies a deterministic choice for her at

each of her information sets. Her set of pure strategies is denoted by Si.

• … but they may also randomize.

A mixed strategy for player 𝑖 is a randomization over pure strategies of

this player, 𝜎𝑖 𝜖 Δ𝑆𝑖 .
Mixed strategies
In any finite game, a mixed strategy of player 𝑖 is a probability vector.

Let 𝑆𝑖 = {𝑠1 , … , 𝑠𝑛 } be player 𝑖’s set of pure strategies, then a mixed

strategy is given by
𝜎𝑖 = 𝑝1 , … , 𝑝𝑛 𝑤𝑖𝑡ℎ 𝑝𝑗 ∈ 0,1 𝑠. 𝑡. σ𝑛𝑗=1 𝑝𝑗 = 1 𝜎𝑖 = 𝑝1 , … , 𝑝𝑛 ,
where 𝑝𝑖 is the probability assigned to playing pure strategy 𝑠𝑖 .

The set of player 𝑖’s mixed strategies is:

Δ𝑆𝑖 = { 𝑝1 , … , 𝑝𝑛 : 𝑝𝑗 ∈ 0,1 𝑠. 𝑡. σ𝑛𝑗=1 𝑝𝑗 = 1}

(!) The set of mixed strategies includes the set of pure strategies:
A mixed strategy that assigns all probability to one pure strategy, e.g.
𝜎𝑖 = (0,1,0, … , 0) corresponds to playing pure strategy 𝑠2 with
probability 1.
Mixed strategies

• The definition of a normal form game still applies – now extended to

allowing for mixed strategies:
(𝐼, ∆𝑆, (𝑢1 , … , 𝑢𝐼 ))

• The expected payoff from playing a mixed strategy is a weighted

average over the payoffs from each of the pure strategies in the mix.
For σi = p1 , … , pn , the expected payoff is:
𝐸𝑈𝑖 𝜎𝑖 , 𝜎−𝑖 = 𝑝1 ∙ 𝑢𝑖 𝑠1 , 𝜎−𝑖 + ⋯ + 𝑝𝑛 ∙ 𝑢𝑖 𝑠𝑛 , 𝜎−𝑖
Mixed Strategies - Example
Pl 2.
Consider the following game:
Player’s pure strategy sets: 𝑆1 = 𝑆2 = {𝑂, 𝐺} O G
O 3, 1 0, 0
Examples of mixed strategies for player 1: Pl. 1
G 0, 0 1, 3
𝜎1 = (1Τ2 , 1Τ2) or 𝜎1 = (3Τ4 , 1Τ4) or 𝜎1 = 1,0

The following are no well-defined mixed strategies:

𝜎1 = (3Τ4 , 1Τ2 ), or 𝜎1 = (3Τ5 , 1Τ4), or 𝜎1 = (3Τ4 , − 1Τ4)

Calculating the payoff of player 1 for strategy profile 𝜎 = ((1Τ2 , 1Τ2 ), (3Τ4 , 1Τ4 ))
Dominance and mixed strategies
We can generalize the definition of strictly and weakly dominated strategies to
games that allow for mixed strategies:
1. We say that for player i a strategy 𝜎𝑖 is strictly dominated by strategy 𝜎𝑖 ′ if
the payoff from strategy 𝜎𝑖 ′ is strictly greater than the payoff from strategy 𝜎𝑖
regardless of what the other players do:
𝑢𝑖 (𝜎𝑖 ′, 𝜎−𝑖 ) > 𝑢𝑖 (𝜎𝑖 , 𝜎−𝑖 ) for all 𝜎−𝑖 ∈ Δ𝑆−𝑖

2. We say that for player i a strategy 𝜎𝑖 is weakly dominated by strategy 𝜎𝑖 ′ if

(i) the payoff from strategy 𝜎𝑖 ′ is weakly greater than the payoff from
strategy 𝜎𝑖 for all strategy-profiles of the other players.
(ii) the payoff from strategy 𝜎𝑖 ′ is strictly greater than the payoff from
strategy 𝜎𝑖 for some strategy-profile of the other players.
𝑢𝑖 (𝜎𝑖 ′, 𝜎−𝑖 ) ≥ 𝑢𝑖 (𝜎𝑖 , 𝜎−𝑖 ) for all 𝜎−𝑖 ∈ Δ𝑆−𝑖 ,
∃𝜎−𝑖 ∈ Δ𝑆−𝑖 𝑠. 𝑡. 𝑢𝑖 𝜎𝑖 ′, 𝜎−𝑖 ) > 𝑢𝑖 (𝜎𝑖 , 𝜎−𝑖 .
Dominance and mixed strategies
Back to our definition:

We say that a strategy 𝜎𝑖 is strictly dominated by strategy 𝜎𝑖 ′ if the

payoff from strategy 𝜎𝑖 ′ is strictly greater than the payoff from
strategy 𝜎𝑖 regardless of what the other players do:
𝑢𝑖 (𝜎𝑖 ′, 𝜎−𝑖 ) > 𝑢𝑖 (𝜎𝑖 , 𝜎−𝑖 ) for all 𝜎−𝑖 ∈ Δ𝑆−𝑖 (1)

• Notice that is suffices to check inequality (1) for every pure strategy
profile of other players. That is, is suffices to check

𝑢𝑖 (𝜎𝑖 ′, 𝑠−𝑖 ) > 𝑢𝑖 (𝜎𝑖 , 𝑠−𝑖 ) for all 𝑠−𝑖 ∈ 𝑆−𝑖 (2)
Dominance and mixed strategies
We can now apply the solution concepts of (iterated) deletion of strictly
dominated strategies to games that allow for mixed strategies.

If we allow for mixed strategies, then a pure

strategy that is not (weakly or strictly) dominated
by a pure strategy may be (weakly or strictly)
dominated by a mixed strategy.
Mixed strategies in games
Pl. 2
L R
U 5, 1 0, 2
Pl. 1 M 0, 0 5, 1
D 1, 3 1, 2

• No strategy is strictly or weakly dominated by a pure strategy

• Are any pure strategies strictly dominated by a mixed strategy?

Yes, D is strictly dominated by (0.5, 0.5, 0), mixing between U and M.
• Notice that this allows us to eliminate strategy D for player 1
• Can we solve the game by IDSDS?
Yes, (M, R) is the only strategy profile that survives IDSDS.
Example Pl. 2
L R
• No strategy is strictly or weakly
dominated by a pure strategy U 6, 1 0, 2
Pl. 1 M 2, 0 5, 0
• Are any strategies strictly D 5, 3 1, 2
dominated by a mixed strategy?

• Yes, D is strictly dominated by mixing between U and M? But how do

we identify the mixture?
• Consider 𝜎𝑖 = 𝛼, 1 − 𝛼, 0 :

𝐸[𝑢𝑖 𝜎𝑖 , 𝐿 ] = 𝛼 ∗ 6 + 1 − 𝛼 ∗ 2 = 2 + 4𝛼
𝐸[𝑢𝑖 𝜎𝑖 , 𝑅 ] = 𝛼 ∗ 3 + 1 − 𝛼 ∗ 5 = 5 − 5𝛼
• For D to be strictly dominated by σi it must hold that
2 + 4𝛼 > 5 and 5 − 5𝛼 > 1,
3 4
which is satisfied for all 𝛼 ∈ ,
4 5
.
Mixed Strategy Nash Equilibrium
Two familiar games Pl. 2
• Penalty shots / rock-paper-scissors: R P S
R 0, 0 -1, 1 1, -1
Pl. 1 P 1, -1 0, 0 -1, 1
S -1, 1 1, -1 0, 0

• Matching pennies:
Pl 2.
H T
H 1, -1 -1, 1
Pl. 1
T -1, 1 1, -1

• Common feature: no pure strategy Nash equilibrium

What about mixed-strategy Nash equilibria?
Mixed strategy Nash equilibrium
Definition (Nash Equilibrium):
A (possibly mixed) strategy profile 𝜎 ∗ = 𝜎1∗, … 𝜎𝑛∗ ∈ ∆𝑆, with 𝜎𝑖∗ ∈ ∆𝑆𝑖
for each player 𝑖 is a (mixed-strategy) Nash equilibrium if for every
player 𝑖
𝐸𝑈𝑖 𝜎𝑖∗, 𝜎−𝑖
∗ ∗
≥ 𝐸𝑈𝑖 𝜎𝑖 , 𝜎−𝑖 ∀𝜎𝑖 ∈ ∆𝑆𝑖 .

• In other words, a strategy profile is a Nash equilibrium if every

player’s strategy 𝜎𝑖∗ is a best-response to the other players
∗
equilibrium strategies, 𝜎−𝑖 .

• A mixed-strategy Nash equilibrium profile has the property that no

player wants to deviate, given the other players’ strategies. That is, no
player can increase his/her payoff by switching to any other strategy.
Example 1: Matching pennies
• We know that this game has no Nash Pl 2.
equilibrium in pure strategies
• What about mixed-strategy Nash H T
equilibria? H 1, -1 -1, 1
1 1 1 1 Pl. 1
• Guess: 𝜎1∗ , 𝜎2∗ = , , , is a NE T -1, 1 1, -1
2 2 2 2

1 1 1
Suppose player 2 plays 𝜎2∗ = , , i.e., H with probability and T with
2 2 2
1
probability .
2
Expected payoff of player 1 from playing
1 1
• H: 𝐸[𝑢1 𝐻, 𝜎2∗ ] = ∙ 1 + ∙ −1 = 0
2 2
1 1
• T: 𝐸[𝑢1 𝑇, 𝜎2∗ ] = ∙ −1 + ∙ 1 = 0
2 2

Hence, player 1 obtains the same payoff from playing H and playing T. That is,
1 1
he cannot improve upon strategy 𝜎1∗ = , by playing either H or T with a
2 2
higher probability ⟹ player 1 has no incentive to deviate.
Example 1: Matching pennies
• We know that this game has no Nash Pl 2.
equilibrium in pure strategies
• What about mixed-strategy Nash 1ൗ H 1ൗ2 T 1ൗ2
2
equilibria? H 1, -1 -1, 1
1 1 1 1 Pl. 1
• Guess: 𝜎1∗ , 𝜎2∗ = , , , is a NE 1ൗ T -1, 1 1, -1
2 2 2 2 2
1 1
We can replicate the argument to show that if player 1 plays 𝜎1∗ = , , then
2 2
player 2 has no incentive to deviate.

1 1 1 1
⟹ we have verified that 𝜎1∗ , 𝜎2∗ = , , , is a NE
2 2 2 2
Example: rock-paper-scissors
Pl. 2
R P S
R 0, 0 -1, 1 1, -1
Pl. 1 P 1, -1 0, 0 -1, 1
S -1, 1 1, -1 0, 0

• No Nash equilibrium in pure strategies

• What are Nash equilibria if we allow for mixed strategies?

• Guess:

1 1 1 1 1 1
𝜎1∗ , 𝜎2∗ = , , , , ,
3 3 3 3 3 3
Example 2: rock-paper-scissors Pl. 2
R P S
1 1 1 1 1 1
• Let’s verify that , , , , , R 0, 0 -1, 1 1, -1
3 3 3 3 3 3
is a Nash equilibrium:
Pl. 1 P 1, -1 0, 0 -1, 1
Consider player 1: S -1, 1 1, -1 0, 0
1 1 1
Payoff from playing “rock” if player 2 plays 𝜎2 ∗ = , ,
3 3 3
1 1 1
𝐸𝑈1 𝑅, 𝜎2 ∗ = ∙ 0 + ∙ −1 + ∙ 1 = 0
3 3 3
1 1 1
Payoff from playing “paper” if player 2 plays 𝜎2 ∗ = , ,
3 3 3
1 1 1
𝐸𝑈1 𝑃, 𝜎2 ∗ = ∙ 1 + ∙ 0 + ∙ (−1) = 0
3 3 3
1 1 1
Payoff from playing “scissors” if player 2 plays 𝜎2 ∗ = , ,
3 3 3
1 1 1
𝐸𝑈1 𝑆, 𝜎2 ∗ = ∙ −1 + ∙ 1 + ∙ 0 = 0
3 3 3
1 1 1
Expected payoff from playing 𝜎1 = 𝛼, 𝛽, 1 − 𝛼 − 𝛽 if player 2 plays 𝜎2 ∗ = , ,
3 3 3
𝐸𝑈1 𝜎1∗, 𝜎2 ∗ = 𝛼 ∙ 0 + 𝛽 ∙ 0 + (1 − 𝛼 − 𝛽) ∙ 0 = 0
⇒ Player 1 has not incentive to deviate (and by symmetry the same is true for player 2)
Mixed strategy Nash equilibrium
We have seen that it suffices to check that no player wants to deviate to
one of his pure strategies. We summarize this in the following fact:

Fact (Nash Equilibrium):

A (possibly mixed) strategy profile 𝜎 ∗ = 𝜎1∗, … 𝜎𝑛∗ ∈ ∆𝑆, with 𝜎𝑖∗ ∈ ∆𝑆𝑖
for each player 𝑖 is a (mixed-strategy) Nash equilibrium if for every
player 𝑖
𝐸𝑈𝑖 𝜎𝑖∗, 𝜎−𝑖
∗ ∗
≥ 𝐸𝑈𝑖 𝑠𝑖 , 𝜎−𝑖 ∀𝑠𝑖 ∈ 𝑆𝑖 .

We have seen how to verify that a strategy profile is a mixed-strategy

Nash equilibrium…

But how do we find mixed-strategy Nash equilibria?

Mixed strategies
Observation: The payoff from a mixed strategy must lie between the
minimal and maximal payoffs from the pure strategies in the mix.

Examples:
• Average height of a group
• Average scores in tests/exams
• Average marathon finishing times
• Batting averages in baseball
• Average points/rebounds in basketball
• ...
Mixed strategies & best responses
Important observation:
If a mixed strategy is a best response, then each of the pure strategies
involved in the mix must itself be a best response. In particular, each
must yield the same expected payoff.

Formal statement:
If a mixed strategy 𝜎𝑖 for player 𝑖 is a best response to a strategy profile
𝜎−𝑖 then each pure strategies 𝑠𝑖 with 𝑝𝑖 𝑠𝑖 > 0 must itself be a best
response. That is, it must hold that 𝐸𝑈𝑖 𝑠𝑖 , 𝜎−𝑖 ≥ 𝐸𝑈𝑖 𝑠𝑖 ′, 𝜎−𝑖 ∀𝑠𝑖 ′ ∈ 𝑆𝑖 .
This implies that
𝐸𝑈𝑖 𝑠𝑖 , 𝜎−𝑖 = 𝐸𝑈𝑖 𝑠𝑖 ′, 𝜎−𝑖 ∀𝑠𝑖 , 𝑠𝑖′ with 𝑝𝑖 𝑠𝑖 , 𝑝𝑖 𝑠𝑖 ′ > 0
Example 2: Battle of the sexes
Pl 2.
• We know that this game has two Nash
equilibria in pure strategies (O, O), (G, G) O G
• Are there any additional mixed-strategy O 3, 1 0, 0
Nash equilibria? How to find them? Pl. 1
G 0, 0 1, 3

Suppose player 2 plays O with probability 𝑞 and G with probability 1 − 𝑞

Expected payoff of player 1 from playing
• O: 𝐸𝑈1 𝑂 = 𝑞 ∙ 3 + 1 − 𝑞 ∙ 0 = 3𝑞
• G: 𝐸𝑈1 𝐺 = 𝑞 ∙ 0 + 1 − 𝑞 ∙ 1 = 1 − 𝑞

In a mixed strategy equilibrium, any pure strategy in the mix must be a best-
response, which is the case if
1
𝐸𝑈1 𝑂 = 𝐸𝑈1 𝐺 ⟹ 𝑞 =
4
For player 1 to be indifferent between playing O and G, player 2 must mix
between O and G with probabilities 1Τ4 , 3Τ4 .
Example 2: Battle of the sexes Pl 2.
O G
• We know that in a Nash equilibrium, for
player 1 wanting to mix between O and G, O 3, 1 0, 0
player 2 must play 𝜎2 = 1Τ4 , 3Τ4 . Pl. 1
G 0, 0 1, 3

Suppose player 1 plays O with probability 𝑝 and G with probability 1 − 𝑝

Expected payoff of player 2 from playing
• O: 𝐸𝑈2 𝑂 = 𝑝 ∙ 1 + 1 − 𝑝 ∙ 0 = 𝑝
• G: 𝐸𝑈2 𝐺 = 𝑝 ∙ 0 + 1 − 𝑝 ∙ 3 = 3 − 3𝑝

In equilibrium, any pure strategy in the mix must be a best-response:

3
𝐸𝑈2 𝑂 = 𝐸𝑈2 𝐺 ⟹ 𝑝 =
4
For player 2 to be indifferent between playing O and G, player 1 must mix
between O and G with probabilities 3Τ4 , 1Τ4

⟹ Mixed-strategy Nash equilibrium in addition to (O, O) and (G, G):

(𝜎1∗ , 𝜎2∗ ) = ( 3ൗ4 , 1ൗ4 , 1ൗ4 , 3ൗ4 )
Example 2: Battle of the sexes
We can also depict the best response functions graphically to illustrate the
solution.
Suppose player 2 plays 𝜎2 = (q, 1 − q), then player 1’s best-response is:
1
𝑂 𝑖𝑓 𝑞 ≥
𝐵𝑅1 𝑞, 1 − 𝑞 = 4
1
𝐺 𝑖𝑓 𝑞 ≤
4

Suppose player 1 plays 𝜎1 = (𝑝, 1 − 𝑝), then player 2’s best-response is:
3
𝑂 𝑖𝑓 𝑝 ≥
𝐵𝑅2 𝑝, 1 − 𝑝 = 4
3
𝐺 𝑖𝑓 𝑝 ≤
4
We can illustrate this in a graph
Example 2: Battle of the sexes
Best-response functions:
𝐵𝑅2(𝑝)
We can see that this game has
three Nash equilibria equal to the
intersections points of the
best-response functions.
𝐵𝑅1 (𝑞)
(O, O): ( 1, 0 , (1, 0))
(G, G): ( 0, 1 , (0, 1))
and
𝜎1∗, 𝜎2∗ = ( .75, . 25 , .25, .75 )
Finding mixed-strategy Nash equilibria
• Use IDSDS first to reduce the strategies that you need to consider
when identifying mixed strategy Nash equilibria.

• Find all pure strategy Nash equilibria

⇒ If there are none, there must be a Nash equilibrium in (strictly)
mixed strategies (in finite games)

• To find ALL equilibria, you must consider every combination of

strategies that survive IDSDS