Game Theory

Nash Equilibrium in Pure Strategies

Tetsuya Hoshino January 17, 2023

In this note, we study the notion of Nash equilibrium in a normal-form game (Nash, 1950).
We discuss the justifications based on a self-enforcing agreement.

Example 1. Consider the following prisoners’ dilemma:

C D
C −1, −1 −3, 0
D 0, −3 −2, −2

Table 1: the prisoners’ dilemma

Suppose that players 1 and 2 have agreed on playing a strategy profile a = (a1 , a2 ). We
examine whether this agreement is “stable” or not, by asking whether there exists at least one
player who wants to deviate from that strategy profile.
Suppose that players 1 and 2 have agreed to play strategy profile (D, D). Then, they are
willing to play it. This is because if (player 1 thinks) player 2 will play D, player 1 is willing to
play D, and similarly, if (player 2 thinks) player 1 will play D, player 2 is willing to play D. In
this sense, the agreement is self-enforcing; strategy profile (D, D) is said to be a (pure-strategy)
Nash equilibrium.
Suppose that players 1 and 2 have agreed to play strategy profile (D, C). Then, they are
not willing to play it. This is because if (player 2 thinks) player 1 will play D, he deviates to
playing D rather than C. In this sense, the agreement is not self-enforcing. □

1 Pure-Strategy Nash Equilibrium

The notion of Nash equilibrium captures the “self-enforcement” of the agreement of play. More
specifically, if players reach an agreement to play strategy profile a∗ before actually playing
game G, then no player will deviate from the agreement a∗ when actually playing it. That is,
the agreement a∗ is self-enforcing. We note that the Nash equilibrium does not attempt to
examine how players can reach such a stable agreement.

Definition 1. In a normal-form game G, a Nash equilibrium (in pure strategies) is a

strategy profile a∗ ∈ A such that for each i ∈ I,

ui (a∗i , a∗−i ) ≥ ui (ai , a∗−i ), ∀ai ∈ Ai . (1)

1
 We provide an equivalent definition of Nash equilibrium.

Definition 2. In a normal-form game G, player i’s best-response correspondence (in

pure strategies) is a correspondence BRi : A−i → 2Ai such that BRi (a−i ) is the set of player
i’s optimal strategies when players −i play strategies a−i . That is,

BRi (a−i ) = argmax ui (ai , a−i ).

ai ∈Ai

Definition 3. In a normal-form game G, a Nash equilibrium (in pure strategies) is a

strategy profile a∗ ∈ A such that for each i ∈ I,

a∗i ∈ BRi (a∗−i ).

The equivalence between Definitions 1 and 3 is straightforward. To see this, we note that
inequality (1) says that strategy a∗i is optimal for player i when players −i play strategies a∗−i .
Formally, a∗i ∈ argmaxai ui (ai , a∗i ) ≡ BRi (a∗−i ).
Although somewhat repetitive, we offer a memorable paraphrase of Nash equilibrium.

A strategy profile a∗ is a (pure-strategy) Nash equilibrium if every player i plays a best

response a∗i to players −i’s strategies a∗−i (that is, if no player has any unilateral profitable
deviation).

Exercise 1. Find a (pure-strategy) Nash equilibrium in the Cournot game described in the
previous note.

Exercise 2. Find a (pure-strategy) Nash equilibrium in the Bertrand game described in the
previous note.

Equilibrium Multiplicity We consider another simple example.

Example 2. Consider the following coordination game:

L R
U 1, 1 −2, 0
D 0, −2 0, 0

Table 2: a coordination game

Player 1 chooses either U or D, while player 2 chooses either L or R. This game has two
pure-strategy Nash equilibria, (U, L) and (D, R). Both players are strictly better-off at the
former than at the latter.
Suppose that players are playing the inefficient equilibrium (D, R). Can they move to the
efficient equilibrium (U, L)? In general, the answer is “no.” By the nature of Nash equilibrium,

2
if (player 1 thinks) player 2 will play R, player 1 does not deviate from playing D. Similarly,
if (player 2 thinks) player 1 will play D, player 2 does not deviate from playing L. That is,
neither player alone can escape from the (inefficient) equilibrium (D, R). □

It is worth summarizing the implication obtained from this example.

A normal-form game G may have multiple (pure-strategy) Nash equilibria, and it is possible
that an equilibrium is strictly better than another equilibrium for all players. It is, in
general, difficult that when players are stuck in a “bad” equilibrium, no player escapes
from the bad one to another one alone.

Exercise 3. Consider the Braess’s paradox in the previous note. Show that:
1. There are multiple (pure-strategy) Nash equilibria.
2. Every (pure-strategy) Nash equilibrium has five players choosing route SAG and five
players choosing route SBG.

Equilibrium Non-Existence We consider another simple example.

Example 3. Consider the following matching pennies game:

H T
H +1, −1 −1, +1
T −1, +1 +1, −1

Table 3: a matching pennies

Each player has a penny and secretly turns the penny to Heads (H) or Tails (T ). If the pennies
match then player 1 gets the both pennies and player 2 loses his penny; otherwise, player 2
gets the both pennies and player 1 loses her penny. This game has no pure-strategy Nash
equilibrium. □

2 How to Reach Nash Equilibrium

Rational Reasoning If players think rationally enough before playing a game, they may arrive
at a Nash equilibrium. Consider the prisoners’ dilemma of Example 1. In this game, it is always
in a player’s interest to play D, regardless of the opponent’s strategy; hence, rational players
can reach the Nash equilibrium (D, D).
However, rational reasoning alone cannot always lead to Nash equilibrium. Consider the
coordination game of Example 2. In this game, there are two (pure-strategy) Nash equilibria,
(U, L) and (D, R). If player 1 conjectures player 2 will play L, then player 1 will rationally
choose U . If, in addition, player 2 conjectures player 1 will play D, then player 2 will rationally
choose R. In the end, they will end up playing a non-equilibrium (U, R).

3
 Box (Leniency as Prisoner’s Dilemma): The prisoner’s dilemma is a fictional situation,
but we have such a situation in reality. A leniency system is a system that aims to
detect illegal activities by artificially creating a prisoners’ dilemma-like situation. It has
been introduced in Japan and has been successful. In many developed countries, collusive
behavior that restricts competition and increase prices (e.g., cartels) are illegal under the
Antitrust Law. In Japan, the Fair Trade Commission (FTC) is in charge of detecting them.
Firms engaging collusion may be subject to surcharges.
The leniency system in Japan is a system that reduces or exempts surcharges if a firm
confesses to having engaged in collusion prior to the FTC’s investigation. Specifically, the
first firm to report is exempted from the full surcharge, the second from 50%, and the third
from 30%.
There are two firms, 1 and 2, that are engaged in collusion. There is a probability of
10% that the collusion is discovered even if both firms remain silent. Assume that the
surcharge of 20 is imposed when the collusion is discovered. This situation is formalized as
follows:

Silent Confess
Silent −2, −2 −20, 0
Confess 0, −20 −5, −5

If both firms keep silent, the collusion is discovered with probability 0.1, and thus each firm
incurs the expected surcharge 0.1 × 20 = 2. If firm i alone confesses (but firm −i is silent),
it will be fully exempt from the surcharge. If firm i alone is silent (but firm −i confesses),
it will have to pay the full surcharge 20. If both firms confess, we assume that there is a
50-50 chance that each is judged as the first to report and as the second; in this case, firm
i’s expected penalty is 0.5 × (0% × 20) + 0.5 × (50% × 20) = 5.
This game has a unique Nash equilibrium, in which both confess. Moreover, this game
is exactly the prisoners’ dilemma in the sense that both will be better off if they keep silent.

That is, rational thinking (in each player’s brain) itself does not lead to a Nash equilibrium.
This is because rational thinking does not give a “right” conjecture about an opponent’s be-
havior. This observation suggests that Nash equilibrium requires more than rational thinking.
Indeed—as we shall see later—we can summarize as follows:

Roughly speaking, it holds that:a

rationality + right conjecture of opponents’ play =⇒ Nash equilibrium

a
More precisely, we require common knowledge of rationality and common knowledge of the conjecture,
but we do not discuss these requirements in detail.

This observation casts doubt on how players can reach right conjectures about opponents’ play.
In what follows, we discuss possible ways they arrive at the right conjectures.

4
Communication As suggested in Example 1, if players “agree” to play a Nash equilibrium
in advance, then they are willing to follow the agreement. In Example 2, if players can
communicate to decide, for example, to play strategy profile (U, L) before playing the game,
then they will play as agreed when actually playing the game. We note that this agreement
itself is non-binding. In other words, there is no external force (e.g., a law enforcement agency)
that compels players to follow the agreement or imposes any penalty on players who violate
the agreement.
It is worth summarizing this interpretation again:

A Nash equilibrium is a self-enforcing strategy profile. That is, players can implement it
by agreement (without any external device).

We note that a Nash equilibrium per se does not require any external device to enforce it,
but in reality, we use an external device to enforce a particular Nash equilibrium.

Example 4. Which side of the road do people drive? This situation is modeled as a coordi-
nation game that has two (pure-strategy) Nash equilibria: All drive on the left-hand side, and
all drive on the right-hand side. In Japan, people drive on the left-hand side, while in the US
and Mexico, people drive on the right-hand side.
Okinawa, a small island in Japan, was occupied by the US after World War II and returned
to Japan in 1972. Okinawa people drove on the right-hand side of the road because Okinawa
was occupied by the US when people started driving. This rule remained in effect for some time
after the reversion. However, Japan decided to switch from driving on the right-hand side of
the road to the left-hand side in 1978, because of the Geneva Convention on Road Traffic, which
requires uniform rules within the country. Such a change will never occur voluntarily, since
each person will drive on the right-hand side if others drive on the right-hand side (which is the
nature of Nash equilibrium). Therefore, Japan amended its Road Traffic Law and mobilized
the police etc. to force Okinawa people to drive on the left-side of the road. □

Learning What if players cannot communicate? In large communities, it is difficult for people
to communicate with others. Even in such communities, we observe Nash equilibrium behavior.
In Example 4, people are forced to follow a particular Nash equilibrium by laws, but it is also
possible that a Nash equilibrium spontaneously emerges without any such enforcement.

Example 5. Which side do people stand on escalators? Similar to Example 4, this situation
is modeled as a coordination game that has two (pure-strategy) Nash equilibria: All stand on
the left, and all stand on the right. Which conventions are prevalent varies from community to
community. In Japan, two conventions coexist: In Tokyo, people stand on the left; in Osaka,
people stand on the right.
This convention is considered as a spontaneous occurrence. When escalators were intro-
duced, people might have stood on a random side, but as they “play” this escalator game

5
every morning in a station, they might have learned the convention that they should stand on
a particular side. In Tokyo, some people, by chance, might have stood on the left; then, people
following them now have incentives to stand on the left. Similarly, in Osaka, some people,
by chance, might have stood on the right; then, people following them now have incentives to
stand on the right.1 □

Nash equilibrium may emerge as a result of trial-and-error learning, but in general this
trial-and-error process may take a long time until it converges, or it may not converge. That
being said, we can still say that:

Behavior patterns entrenched in society are likely to be Nash equilibrium.

Indeed, if a behavioral pattern is not a Nash equilibrium, then some people will deviate from
the pattern to taking different behavior, which breaks the pattern. Hence, if a behavioral
pattern is established in society, it should be a Nash equilibrium, so that no player alone will
not deviate.

References
Nash, J. F. (1950). Equilibrium points in n-person games. Proceedings of the National Academy
of Sciences, 36 (1), 48–49.

1
There have been recent campaigns to call for people to stand on both sides (for safety, etc.), but they have
not been successful at least not in Japan, as far as I know. It is generally difficult to change people’s Nash
equilibrium behavior.