Fixed Point Theorems

Definition: Let X be a set and let f : X → X be a function that maps X into itself. (Such
a function is often called an operator, a transformation, or a transform on X, and the
notation T (x) or even T x is often used.) A fixed point of f is an element x ∈ X for which
f (x) = x.

Example 1: Let X be the two-element set {a, b}. The function f : X → X defined by
f (a) = b and f (b) = a has no fixed point, but the other three functions that map X into
itself each have one or two fixed points. More generally, let X be an arbitrary set; every
constant function f : X → X mapping X into itself has a unique fixed point; and for the
identity function f (x) = x, every point in X is a fixed point.

Remark: Note that the definition of a fixed point requires no structure on either the set X
or the function f .

Example 2: Let X be the unit interval [0, 1] in R. The graph of a function f : X → X

is a subset of the unit square X × X. If f is continuous, then its graph is a curve from the
left edge of the square to the right edge (see Figure 1). A fixed point of f is an element of
[0, 1] at which the graph of f intersects the 45◦ -line. Intuitively, it seems clear that if f is
continuous then it must have a fixed point (its graph must cross or touch the 45◦ -line), and
also that discontinuous functions f may not have a fixed point.

Fixed points show up in a number of contexts, but most prominently in the notion of equi-
librium. We typically represent a “system” by specifying the set of “states” the system can
be in, and also specifying how the system moves, or “transitions,” from one state to another.
For example, we can represent the state of a system of markets as a list p of prices and
a list z of net demand quantities. If we want to say this is the state at time t, we could
write st = (pt , zt ). We sometimes also include some kind of specification of how the economy
moves from state to state; let’s say st+1 = f (st ) for some transition function f : S → S,
where S is the set of all possible states. A stationary state would be defined as a state s for
which f (s) = s, so that if st = s then st+1 = st . We would also call this an equilibrium of
the system described by f : S → S — an equilibrium is a stationary state. This is exactly
the motivation for our definition of Walrasian equilibrium as a pair (p, z) at which z = 0
(markets clear): while we don’t specify a function f , we believe that whatever the true f
is, it satisfies f (s) 6= s when z 6= 0 and f (s) = s when z = 0 — that the fixed points of f
are the states at which markets clear. Similarly, we define a game-theoretic equilibrium to
be a list s of strategies or actions by the players that is a stationary state of any transition
function f that captures the idea that f (s) = s for strategy-lists s in which no player wants
to change his strategy.
 Contractions and The Banach Fixed Point Theorem

Definition: Let (X, d) be a metric space. A contraction of X (also called a contraction

mapping on X) is a function f : X → X that satisfies
∀x, x0 ∈ X : d f (x0 ), f (x) 5 βd(x0 , x)

for some real number β < 1. Such a β is called a contraction modulus of f . (Note that if
β is a contraction modulus of f and β < β 0 < 1, then β 0 is also a contraction modulus of f .)

In other words, a transformation is a contraction if the images of any pair of points are always
closer together than the points themselves, and if the ratio of these two distances is bounded
away from 1. In Example 3 their ratio is not bounded away from 1.

Example 3: Let I = (0, 1), the open unit interval, and let f : I → I be the function
f (x) = 21 + 12 x2 . Then |f (x0 ) − f (x)| = 12 (x0 + x)|x0 − x| < |x0 − x| for all x, x0 ∈ I, but f is
not a contraction because if β < 1 and x, x0 > β, then |f (x0 ) − f (x)| > β|x0 − x|. There is no
β < 1 that will satisfy the inequality in the definition of a contraction.

Example 4: Let f : R → R be a differentiable real function. If there is a real number

β < 1 for which the derivative f 0 satisfies |f 0 (x)| 5 β for all x ∈ R, then f is a contraction
with respect to the usual metric on R and β is a modulus of contraction for f . This is a
straightforward consequence of the Mean Value Theorem: let x, x0 ∈ R and wlog assume
x < x0 ; the MVT tells us there is a number ξ ∈ (x, x0 ) such that f (x0 ) − f (x) = f 0 (ξ)(x0 − x)
and therefore |f (x0 )−f (x)| = |f 0 (ξ)||x0 −x| 5 β|x0 −x|. The same MVT argument establishes
that if β < 1 and f : (a, b) → (a, b) satisfies |f 0 (x)| 5 β for all x ∈ (a, b), then f is a
contraction of (a, b).

Theorem: Every contraction mapping is continuous.

Proof: Let T : X → X be a contraction on a metric space (X, d), with modulus β, and let
x ∈ X. Let  > 0, and let δ = . Then d(x, x) < δ ⇒ d(T x, T x) 5 βδ < . Therefore T is
continuous at x. Since x was arbitrary, T is continuous on X. 

The above proof actually establishes that a contraction mapping is uniformly continuous:

Definition: Let (X, dX ) and (Y, dY ) be metric spaces. A function f : X → Y is uniformly

continuous if for every  > 0 there is a δ > 0 such that
∀x, x0 ∈ X : dX (x, x0 ) < δ ⇒ dY f (x), f (x0 ) < .

Notice how this definition differs from the definition of continuity: uniform continuity requires
that, for a given , a single δ will work across the entire domain of f , but continuity allows
that the δ may depend upon the point x at which continuity of f is being evaluated.

2
Theorem: Every contraction mapping is uniformly continuous.

Banach Fixed Point Theorem: Every contraction mapping on a complete metric space
has a unique fixed point. (This is also called the Contraction Mapping Theorem.)
Proof: Let T : X → X be a contraction on the complete metric space (X, d), and let β be
a contraction modulus of T . First we show that T can have at most one fixed point. Then
we construct a sequence which converges and we show that its limit is a fixed point of T .

(a) Suppose x and x0 are fixed points of T . Then d(x, x0 ) = d(T x, T x0 ) 5 βd(x, x0 ); since
β < 1, this implies that d(x, x0 ) = 0, i.e., x = x0 .

(b) Let x0 ∈ X, and define a sequence {xn } as follows:

x1 = T x0 , x2 = T x1 = T 2 x0 , ... , xn = T xn−1 = T n x0 , ...
We first show that adjacent terms of {xn } grow arbitrarily close to one another — specifically,
that d(xn , xn+1 ) 5 β n d(x0 , x1 ):
d(x1 , x2 ) 5 βd(x0 , x1 )
d(x2 , x3 ) 5 βd(x1 , x2 ) 5 β 2 d(x0 , x1 )
...
d(xn , xn+1 ) 5 βd(xn−1 , xn ) 5 β n d(x0 , x1 ).

1
Next we show that if n < m then d(xn , xm ) < β n 1−β d(x0 , x1 ):
d(xn , xn+1 ) 5 β n d(x0 , x1 )
d(xn , xn+2 ) 5 d(xn , xn+1 ) + d(xn+1 , xn+2 )
5 β n d(x0 , x1 ) + β n+1 d(x0 , x1 ) = (β n + β n+1 )d(x0 , x1 )
...
d(xn , xm ) 5 (β n + β n+1 + · · · + β m−1 )d(x0 , x1 )
= β n (1 + β + β 2 + · · · + β m−1−n )d(x0 , x1 )
< β n (1 + β + β 2 + · · · )d(x0 , x1 )
1
= βn d(x0 , x1 ).
1−β
1
Therefore {xn } is Cauchy: for  > 0, let N be large enough that β N 1−β d(x0 , x1 ) < , which
ensures that n, m > N ⇒ d(xn , xm ) < . Since the metric space (X, d) is complete, the
Cauchy sequence {xn } converges to a point x∗ ∈ X. We show that x∗ is a fixed point of
T : since xn → x∗ and T is continuous, we have T xn → T x∗ — i.e., xn+1 → T x∗ . Since
xn+1 → x∗ and xn+1 → T x∗ , we have T x∗ = x∗ . 

Note that the proof of uniqueness did not require that the space be complete.

3
First Cournot Equilibrium Example: Two firms compete in a market, producing at
output levels q1 and q2 . Each firm responds to the other firm’s production level when choosing
its own level of output. Specifically (with a1 , a2 , b1 , b2 all positive),
q1 = r1 (q2 ) = a1 − b1 q2
q2 = r2 (q1 ) = a2 − b2 q1

but qi = 0 if the above expression for qi is negative. See Figure 2. The function ri : R+ → R+


is firm i’s reaction function. Define r : R2+ → R2+ by r(q1 , q2 ) = r1 (q2 ), r2 (q1 ) . The function
r is a contraction with respect to the city-block metric if b1 , b2 < 1:

d r(q), r(q0 ) = |r1 (q) − r1 (q0 )| + |r2 (q) − r2 (q0 )|

= |(a1 − b1 q2 ) − (a1 − b1 q20 )| + |(a2 − b2 q1 ) − (a2 − b2 q10 )|

= b1 |q20 − q2 | + b2 |q10 − q1 |
5 max{b1 , b2 }(|q1 − q10 | + |q2 − q20 |)
= max{b1 , b2 } d(q, q0 ).

Therefore we have an “existence and uniqueness result” for Cournot equilibrium in this
example: r has a unique fixed point q∗ — a unique Cournot equilibrium — if each bi < 1.

There are several things to note about this example. First, note that while the condition
b1 , b2 < 1 is sufficient to guarantee the existence of an equilibrium, it is not necessary. Second,
note that we could have obtained the same result, and actually calculated the equilibrium
production levels, by simply solving the two “response” equations simultaneously. The con-
dition b1 , b2 < 1 is easily seen to be sufficient (and again, not necessary) to guarantee that
the two-equation system has a solution. (Note that b1 b2 6= 1 is in fact sufficient.) However,
things might not be so simple if the response functions are not linear, or if there are more
firms (and therefore more equations). We’ll consider a second, nonlinear example shortly.

A third thing to note about the example is that we used the city-block metric instead of the
Euclidean metric. This highlights an important and useful fact: a given function mapping a
set X into itself may be a contraction according to one metric on X but not be a contraction
according to other metrics. Recall that the definition of a fixed point, and therefore whether
a given point in the function’s domain is a fixed point, does not depend on the metric we’re
using, and in fact does not even require that X be endowed with any metric structure.
We’ve earlier seen that a judicious choice of metric can make a proof easier; here we see that
a judicious choice of metric can make a method of proof available that would not work with
a different metric.

4
Exercise: Let A be the matrix " #
1 1
4 2
A= 1 2
,
6 3

and let T : R2 → R2 be the transformation defined by T (x) = Ax. Let e1 and e2 denote the
unit vectors in R2 , " # " #
1 0
e1 = and e2 = .
0 1

(a) Plot the locus of all points x ∈ R2 that satisfy kxk∞ = 1 and the locus of all points that
satisfy kxk1 = 1. In the same diagram, plot the points T (e1 ) and T (e2 ).
(b) Is the transformation T a contraction? Does this depend on which norm you’re using?
(The geometry in (a) should provide you with some help in answering this question.) For
each of the norms k·k∞ and k·k1 , determine whether T is or is not a contraction with respect
to the norm.

5
 The Brouwer Fixed Point Theorem

In Example 2 in the preceding section the Banach Theorem seems somewhat limited: it
seems intuitively clear that any continuous function mapping the unit interval into itself will
have a fixed point, but the Banach Theorem applies only to functions that are contractions.
An elementary example is the function f (x) = 1 − x, which has an obvious fixed point at
x = 1/2. But for every x and x0 in the interval [0, 1], d f (x), f (x0 ) = d(x, x0 ), so f is not

a contraction and the Banach Fixed Point Theorem doesn’t apply to f . The fixed point
theorem due to Brouwer covers this case as well as a great many others that the Banach
Theorem fails to cover because the relevant functions aren’t contractions.

Brouwer Fixed Point Theorem: Let S be a nonempty, compact, convex subset of Rn .

Every continuous function f : S → S mapping S into itself has a fixed point.

The Brouwer Theorem requires only that f be continuous, not that it be a contraction, so
there are lots of situations in which the Brouwer Theorem applies but the Banach Theorem
doesn’t. In particular, Brouwer’s Theorem confirms our intuition that any continuous func-
tion mapping [0, 1] into itself has a fixed point, not just the functions that satisfy |f 0 (x)| 5 β
for some β < 1. But conversely, the Banach Theorem doesn’t require compactness or con-
vexity — in fact, it doesn’t require that the domain of f be a subset of a vector space, as
this version of Brouwer’s Theorem does. So there are also lots of situations where Banach’s
Theorem applies and Brouwer’s doesn’t.

Proofs of Brouwer’s Theorem require some highly specialized mathematical ideas. For us,
the benefit of developing these ideas in order to work through a proof of the theorem doesn’t
come close to justifying the time cost it would require, so we won’t go there.

Here’s a generalization of Brouwer’s Theorem to normed vector spaces (which, in particular,

don’t have to be finite-dimensional, as Brouwer’s Theorem requires):

Schauder Fixed Point Theorem: Let S be a nonempty, compact, convex subset of a

normed vector space. Every continuous function f : S → S mapping S into itself has a fixed
point.

This theorem would apply, for example, to any compact convex subset of C[0, 1], the vector
space of continuous functions on the unit interval, with the max norm.

6
 The Method of Successive Approximations

The Banach and Brouwer Theorems are existence theorems: when a function satisfies the
assumptions of one of the theorems, the theorem tells us that the function has a fixed point.
We’ve described how economic and game theoretic equilibria can generally be represented as
fixed points; therefore fixed point theorems can tell us when an economic or strategic model
has an equilibrium, which is important information.

Often we also want to find the equilibrium, or equilibria. For example, if we want to know
what will be the result of some policy change or other exogenous change in the economy, we
typically want to find the new equilibrium for the new economic parameter values. So what
we need are methods for computing equilibria, i.e., fixed points. Here we’ll present only the
most elementary version of this problem. In subsequent courses you’ll study more powerful
methods for dealing with particular classes of problems.

Let’s start by going back and taking a look at our proof of the Banach Fixed Point Theorem.
We began by using the given function f to recursively define a sequence of points in the
space X via the recursion formula xn+1 = f (xn ). Then everything we did in the proof prior
to the last two sentences was in the service of proving that the sequence converges: in order
to get that result we used the facts that the space is complete and that the function f is a
contraction. Then, once we knew that the sequence converges, in the last two sentences of
the proof we simply showed that the sequence’s limit is a fixed point of f . The argument in
these last two sentences used only the fact that f is continuous and that lim xn exists — we
no longer needed to use either completeness of the space or the fact that f is a contraction.

The following theorem and its proof repeat the result and the argument in those last two
sentences of the Banach Theorem’s proof: if we have a sequence that’s defined recursively
from a continuous function f and the sequence converges, then the sequence’s limit is a fixed
point of f .

Theorem: Let (X, d) be a metric space, let f : X → X, let x0 ∈ X, and let {xn } be the
sequence defined recursively from f and x0 by xn+1 = f (xn ). If f is continuous and {xn }
converges, then lim xn is a fixed point of f .
Proof: Let x∗ = lim xn . Then
f (x∗ ) = f (lim xn ) = lim f (xn ) = lim xn+1 = lim xn = x∗ ,
where f (lim xn ) = lim f (xn ) follows from continutiy of f . 

This theorem is an example of The Method of Successive Approximations — recursively

constructing a sequence that will converge to the fixed point (or other value) we’re trying to

7
find. The sequence’s terms can be thought of as approximations to the fixed point, and —
if the sequence converges! — the terms eventually become arbitrarily close approximations
to the fixed point. The Banach Fixed Point Theorem tells us that a contraction mapping
defined on a complete metric space will have a fixed point, because in that case any sequence
defined by recursively applying the function will in fact converge.

Ideally, the sequence we construct in implementing the method of successive approximations

would be one in which successive terms of the sequence are successively closer approximations
to the fixed point. The above theorem doesn’t guarantee that. But the proof of the Banach
Theorem works precisely because the terms are successively closer to the the function’s
fixed point, as the following theorem guarantees for any sequence defined recursively from a
contraction mapping. (Note that the theorem does not assume that the space is complete,
and it does not guarantee the existence of a fixed point. We’ve shown earlier that if a
contraction does have a fixed point, it will be unique.)

Theorem: If T is a contraction with contraction modulus β on a metric space (X, d), and
if T has a fixed point x∗ , then for any x ∈ X,
∀n ∈ N : d(T n x, x∗ ) 5 β n d(x, x∗ ).

Proof:
d(T n x, x∗ ) = d(T T n−1 x, T x∗ ), because x∗ is a fixed point of T
5 βd(T n−1 x, x∗ ), because T is a contraction
5 β 2 d(T n−2 x, x∗ )
5 ··· 5 β n−1 d(T x, x∗ )
= β n d(x, x∗ ). 

The Cournot Equilibrium Example Again: Suppose the current “state” of the market
is q(0) = (q1 (0), q2 (0)) — Firm i is producing qi (0) units. Suppose further that “tomorrow”
(at time t = 1) each firm chooses its production level qi (1) by responding to the amount its
rival firm produced today (at time t = 0), and similarly at each subsequent date:
q1 (t + 1) = r1 (q2 (t)) = a1 − b1 q2 (t) and q2 (t + 1) = r2 (q1 (t)) = a2 − b2 q1 (t),
or more concisely, q(t + 1) = r(q(t)). We’ve already proved that the function r(·) is a
contraction if b1 , b2 < 1, in which case the above theorem guarantees that the sequence
{q(t)} will converge monotonically (in distance) to the Cournot equilibrium q∗ . Try it
yourself by choosing values for the parameters a1 , a2 , b1 , b2 — for example, a1 = 25, a2 =
30, b1 = 3/5, b2 = 1/2 — and any starting state q(0). Plot the sequence along with the
reaction curves in the q1 -q2 -space.

8
Second Cournot Equilibrium Example: Suppose the response functions in our Cournot
example are
1 1
q1 = r1 (q2 ) = and q2 = r2 (q1 ) = e−q1 .
2(1 + q2 ) 2
Now it’s not so easy to calculate the equilibrium, or even to tell whether an equilibrium exists,
as it was in the linear example, where we could simply solve the two response equations
simultaneously. But it’s possible to show that the function r : R2+ → R2+ defined as before is
a contraction (so a Cournot equilibrium exists and is unique), and one can use the Method
of Successive Approximations to compute an approximation to the equilibrium.

Exercise: Use Excel (or any other computational program) and the Method of Successive
Approximations to compute the Cournot equilibrium to three decimal places in the Second
Cournot Example.

Example: Let a be a positive real number and define the function f : R++ → R++ as
1 a
f (x) = x + .
2 2x

Does this function have a fixed point? Neither the Banach Theorem nor the Brouwer Theorem
gives us the answer, because the space R++ is neither complete nor compact. And we
can’t extend the domain of f to all of R+ (which would make the space complete) because
limx→0 f (x) = ∞. But the method of successive approximations leads us to the (unique)
fixed point of f , as follows.

Let’s choose some (arbitrary) positive real number as x0 and then recursively define the
sequence {xn } by
1 a
xn+1 = f (xn ) = xn + .
2 2xn
√
We’ve seen this sequence before: we showed that it converges to a. Therefore, since f is
√
continuous, we know from the above theorem that a is a fixed point of f .
√
It’s also straightforward to show that a is the unique fixed point of f : it’s easy to show
that
√ √ √ √
(1) if x < a then f (x) > a, and (2) if x > a then a < f (x) < x;
√
therefore no x 6= a can be a fixed point of f .

9
Figure 1

Figure 2

10