Tutorial 4 - Contraction Principle

May 1, 2019

Introduction
In today’s tutorial we’ll prove the contraction map theorem, that gives sufficient condi-
tions for existence of a unique fixed point in a mapping of a metric space to itself. For
more details, see (M) chapter 3, up to 3.3.

Definitions
Metric space A metric set is a set M equipped with a metric d on M, i.e. a function

d : M⇥M ! R

such that for every x, y, z 2 M the following holds:

1. d(x, y) = 0 () x = y

2. d(x, y) = d(y, x)

3. d(x, y)  d(x, z) + d(z, y) (triangle inequality)

Examples:
p
• M = Rn , d(x, y) = Â(xi yi )2

• M = R+ , d(x, y) = | log(y/x)|

• Any vector space V with a norm || · ||, with the metric defined as d(x, y) = ||x y||.
´b ´b
For example, M = { f : [a, b] ! R : a | f (x)|2 dx < •}, d( f , g) = a | f (x) g(x)|2 dx

Cauchy sequence (in a metric space) Given a metric space (M, d), a sequence {xn }• n=1 ⇢
M is Cauchy if, for any e > 0 there exists an integer N 2 N such that for all positive inte-
gers n, m > N,
d(xn , xm ) < e.
Property: Every convergent sequence is a Cauchy sequence.

Complete metric space A complete metric space is a metric space (M, d) in which all
Cauchy sequences converge to limits in M.
Examples: R, [0, 1] with d(x, y) = |x y|.
Counter-examples: Q, (0, 1] with d(x, y) = |x y|.

1
Contraction mapping Let (X, d) be a metric space. Then a map T : X ! X is called a
contraction map on X if there exists c 2 [0, 1) such that

d(T (x), T (y))  cd(x, y)

for all x, y 2 X.
Examples:

• X = R, d(x, y) = |x y|, T (x) = x/2. Who is c?

• The logistic map for r < 1: xn+1 = rxn (1 xn ).

Obvious property: A contraction map is Lipschitz continuous, i.e. f : X ! X with

d( f (x), f (y))  Kd(x, y) for K 0.

Contraction Map Theorem

AKA Banach-Caccioppoli fixed-point theorem.

Contraction Map Theorem Let (X, d) be a non-empty complete metric space with a
contraction mapping T : X ! X. Then T admits a unique fixed point x⇤ = T (x⇤ ) 2 X.

Proof: Choose some x0 2 X and define the following sequence:

{xn }•
n=1 : xn = T (xn 1 ).

Then, for all n 2 N, from the contraction mapping

d(xn , xn 1 )  qd(xn 1 , xn 2 )  q2 d(xn 2 , xn 3 )  ...  qn d(x1 , x0 ).

with 0  q < 1.

This sequence is a Cauchy sequence: For m, n 2 N with m > n,

d(xm , xn )  d(xm , xm 1 ) + ... + d(xn+1 , xn )

 (qm 1 + qm 2 + ... + qn )d(x1 , x0 )
m n 1
 qn d(x1 , x0 ) Â qk
k=0
•
1
 qn d(x1 , x0 ) Â qk = qn d(x1 , x0 )
k=0 1 q

Given e > 0, since q 2 [0, 1) there exists N 2 N large enough such that

e(1 q)
qN < .
d(x1 , x0 )
Thus, for n, m > N,
1 e(1 q) 1
d(xm , xn )  qn d(x1 , x0 ) < d(x1 , x0 ) = e.
1 q d(x1 , x0 ) 1 q

Thus this sequence is a Cauchy sequence, and by completeness of (X, d) the sequence has
a limit x⇤ 2 X.

2
 This limit is a fixed point of T :

x⇤ = lim xn = lim T (xn 1 ) = T ( lim xn 1 ) = T (x⇤ ),

n!• n!• n!•

where the third equality results from the continuity of T .

Another way to prove this: Suppose N is large enough such that d(xn , x⇤ ) < e for all
n > N, then:

d(T (x⇤ ), x⇤ )  d(T (x⇤ ), xn+1 ) + d(xn+1 , x⇤ )  d(T (x⇤ ), T (xn )) + d(xn+1 , x⇤ )  (q + 1)e.

Since this is true for any e > 0, the distance is 0 and T (x⇤ ) = x⇤ .

The limit is unique: Assume two fixed points p1 = T (p1 ) and p2 = T (p2 ). From
the contraction property of T ,

d(T (p1 ), T (p2 ))  q · d(p1 , p2 ) = q · d(T (p1 ), T (p2 )).

Since q < 1, this means that d(p1 , p2 ) = 0 and from the definition of a metric space this
means p1 = p2 .

Examples:

• M = [a, b], d(x, y) = |x y|. This example can be analyzed graphically:

Consider a mapping T : [0, 1] ! [0, 1] satisfying the contraction condition. This
means that its slope is never above 1, and therefore it must cross the y = x line once
and only once:

x=y
1

T(x)
x*
0 1 x

Figure 1: A general 1D contraction map.

• Importance of completeness of the metric space: Consider M = (0, 1], d(x, y) =

|x y|, T (x) = x/2. Then all of the conditions for the contraction map theorem
hold except for completeness of the metric space, and indeed the map does not have
a fixed point in M.

• M = C0 (S1 ) is the space of continuous functions on the unit circle that are periodic
with period 1: f (x) = f (x + 1). The distance between two functions is defined by
the sup norm: d( f , g) = supx | f (x) g(x)|.

3
 Consider the mapping operator
1
T ( f )(x) = f (2x).
2
Then T ( f ) 2 C0 (S1 ), and d(T ( f ), T (g)) = 12 d( f , g). Therefore T is a contraction
map and hence it has a fixed point, which in this case is a function f ⇤ (x) : [0, 1] ! R
that satisfies T ( f ⇤ )(x) = 12 f ⇤ (2x) for all values of x.
What is f ⇤ ? According to the theorem, any initial function will converge to f ⇤ , so it
suffices to check one. For example, take f0 (x) = sin 2px. Then f1 (x) = T ( f0 )(x) =
1 1 n+1 px. In the sup norm, this sequence converges to 0:
2 sin 4px, and f n (x) = 2n sin 2

1 1
sup{ fn (x) 0} = sup{ fn (x)} = n
sup{sin 2n+1 px} = n ! 0.
x x 2 x 2 n!•

• A more interesting example: Consider the same M and d as the previous example,
with the following mapping:
1
T ( f )(x) = cos 2px + f (2x).
2
This is still a contraction mapping (check!), so the theorem holds and the map has
a single fixed point f ⇤ (x). Again, to find the fixed point let’s consider the initial
function f0 = sin(2px). Then
1
f1 (x) = cos 2px + sin(4px),
2
1 1
f2 (x) = cos 2px + cos 4px + sin(8px)),
2 4
j 1 n+1
cos(2 px) 1
f j (x) = Â n
+ j sin (2 j+1 px)
n=1 2 2

The last term goes to zero in the sup-norm, and by the contraction-mapping theorem,
the result is guaranteed to be unique and continuous. The fixed point is not an elementary
function; it is an example of a Weierstrass function (continuous everywhere, differentiable
nowhere), see Fig. 2.

Figure 2: The Weierstrass function (from (M) chapter 3).

4
Uses:

• Proof of the theorem about persistence of period-m maps given small perturbations.

• Proof of existence and uniqueness of solutions to ODEs:

Theorem - Picard-Lindelof Existence and Uniqueness Consider the initial

value problem
ẋ = f (x), x(t0 ) = x0
for x : R ! Rn , f : Rn ! Rn .
Suppose that for x0 2 Rn , there is a b such that f : Bb (x0 ) ! Rn is Lipschitz with
constant K. Then the initial value problem has a unique solution x(t) for q 2 J =
[t0 a,t0 + a], provided that a = b/M where M = maxx2Bb (x0 ) | f (x)|.
Proof idea:
´t
Look at the integrated problem x(t) = x0 + t0 f (x(t))dt, and define the map:
ˆ t
T (u)(t) = x0 + f (u(t))dt.
t0

Then x⇤ (t) is a solution to the initial value problem iff it is a fixed point of T . One
can show that T is a contraction for short enough times and therefore a solution
x⇤ (t) exists and is unique. The full proof appears in (M), chapter 3.