MA651 Topology. Lecture 12. Mapping in Metric Spaces.

This text is based on the following books:

• ”Introduction to Real Analysis” by A.N. Kolmogorov and S.V. Fomin

• ”Linear Algebra and Analysis” by Marc Zamansky

I have intentionally made several mistakes in this text. The first homework assignment is to find
them.

66 Uniform continuity
The definition of continuity of a mapping f of a matric space (E, d) into a metric space (F, δ) is
the same as that given in the case of two arbitrary topological spaces. However we can now define
the concept of uniform continuity.

Definition 66.1. Let f be a mapping of a metric space (E, d) into a metric space (F, δ). f is
said to be uniformly continuous if for each ε > 0 we can find α(ε) > 0 such that if d(x, x0 ) < α
then δ(f (x), f (x0 )) < ε.

Example 66.1. If f is uniformly continuous it is continuous, but the converse is false. For
example, the continuous mapping of R+ into R+ defined by x → 1/x.

There are cases where continuity implies uniform continuity. These depend on topological prop-
erties of the space (E, d). For example, let us prove the following result:

Theorem 66.1. Let f be a mapping of a compact metric space (E, d) into a metric space (F, δ).
if f is continuous it is uniformly continuous.

Proof. Let ε > 0. To each x ∈ E we assign an open ball B(x, rx ) of center x and radius rx such
that if x0 ∈ B(x, rx ), δ(f (x0 ), f (x)) < ε/2. This may be done since f is continuous.

Consider the open balls B(x, rx /2). They cover E and, since E is compact, include a finite cover
B(xi , rxi /2).

1
Let
m = inf(rxi /2)
and consider two points x, x0 of E such that d(x, x0 ) < m. The point x is contained in a certain
ball B(xi , rxi /2) and we have

d(x0 , xi ) 6 d(x0 , x) + d(x, x0 ) < m + rxi /2 6 rxi

It follows that x0 ∈ B(xi , rxi ). We now have

δ(f (x0 ), f (x)) < δ(f (x0 ), f (xi )) + δ(f (x), f (xi )) < ε
since x0 and x belong to B(xi , rxi ).

Example 66.2. Let us consider a continuous function defined by distance. The distance d(x, A)
of a point x from a subset A of a metric space E is defined by

d(x, A) = inf d(x, y)

y∈A

We will prove that the function x → d(x, A) is uniformly continuous on E for every set A.

d(x, A) being the lower bound of the d(x, u) for u ∈ A, given ε > 0 there exists a u0 ∈ A such that

d(x, A) 6 d(x, u0 ) < d(x, A) + ε

Let y be another point of E. We have

d(y, u0 ) 6 d(x, y) + d(x, u0 ) < d(y, x) + d(x, A) + ε

and since

d(x, A) = inf d(y, u) 6 d(y, u0 )

u∈A

it follows that

d(y, A) 6 d(y, x) + d(x, A) + ε

Since ε is arbitrary there results

d(y, A) 6 d(y, x) + d(x, A)

and interchanging x and y

d(x, A) 6 d(x, y) + d(y, A)

2
Thus for each subset A and arbitrary points x, y of E we have

|d(x, A) − d(y, A)| < d(x, y)

which establishes the uniform continuity of the function x → d(x, A).

67 Extension by continuity
The following question arises in a natural way. If E and F are two spaces, A a dense subspace of
E, and φ a continuous mapping of A into F , is there a mapping f of E into F which is continuous
in E and whose restriction to A is φ?

This question can be posed more graphically as follows: if f is a continuous mapping of E into F
and A is dense in E can f be reconstituted from its restriction φ to A?

The solution of this problem is called the extension of φ from A to E by continuity. This, in
fact, can be done if we impose some quite general conditions satisfied by metric spaces (which are
separated and normal).

We first prove the following statement:

Proposition 67.1. If f and g are two continuous mappings of a space E into a separated space
F , and are equal at the points of a dense subset A of E, then they are equal everywhere in E.

Proof. For, if x ∈ E is adherent to A, f (x), the limit of f (ξ) when ξ tends to x, is also the limit
when ξ tends to x in A. Since f (ξ) = g(ξ) for ξ ∈ A, and F is separated, the limits of f and g at
each point x ∈ E are equal.

Now let φ be a mapping of a set A which is dense in E, into a separated space F . In order to
be able to extend φ to E we must suppose that when ξ ∈ A tends to x ∈ E, φ(ξ) has a limit,
which we shall denote by f (x). (If E is a metric space we can say that for every sequence (ξn ) of
elements of A converging to x ∈ E, φ(ξn ) must have a limit, and this limit must be the same for
every such sequence (ξn ).) We now prove the following theorem:

Theorem 67.1. Let A be a dense subspace of a space E, F a normal space, and φ a mapping of
A into F such that for every x ∈ E, φ(ξ) has a limit f (x) in F when ξ ∈ A tends to x. Then the
function f is continuous in E.

Proof. See the proof of Theorem 38.1

3
68 Contraction mapping
68.1 The fixed point theorem
Let A be a mapping of a metric space R into itself. Then x is called a fixed point of A if Ax = x,
i.e. A carries x into itself. Suppose there exists a number α < 1 such that
ρ(Ax, Ay) 6 αρ(x, y)
for every pair of points x, y ∈ R. Then A is said to be a contraction mapping. Every con-
traction mapping is automatically continuous, since it follows from the ”contraction condition”
(ρ(Ax, Ay) 6 αρ(x, y)) that Axn → Ax whenever xn → x.
Theorem 68.1. (Fixed point theorem). Every contraction mapping A defined on a complete
metric space R has a unique fixed point.
Proof. Given an arbitrary point x0 ∈ R, let

x1 = Ax0 , x2 = Ax1 = A2 x0 , . . . , xn = Axn−1 = An x0 , . . .

where A2 x = A(A(x)), A3 x = A(A2 x) = A(A(A(x))), etc.
Then the sequence {xn } is fundamental. In fact, assuming to be explicit that n 6 n0 , we have

0
ρ(xn , xn0 ) = ρ(An x0 , An x0 ) 6 αn ρ(x0 , xn0 −n )
6 αn [ρ(x0 , x1 ) + ρ(x1 , x2 ) + · · · + ρ(xn0 −n−1 , xn0 −n )]
0
6 αn ρ(x0 , x1 )[1 + α + α2 + · · · + αn −n−1 ]
1
< αn ρ(x0 , x1 )
1−α
But the expression on the right can be made arbitrary small for sufficiently large n, since α < 1.
Since R is complete, the sequence {xn }, being fundamental, has a limit
x = lim xn
n→∞

Then, by continuity of A,
Ax = A lim xn = lim Axn = lim xn+1 = x
n→∞ n→∞ n→∞

This proves the existence of a fixed point x. To prove the uniqueness of x we note that is

Ax = x, Ay = y
then by definition of contraction mapping

ρ(x, y) 6 αρ(x, y)
But then ρ(x, y) = 0 since α < 1, and hence x = y

4
Remark. The fixed point theorem can be used to prove existence and uniqueness theorems for
solutions of equations of various types. Besides showing that an equation of the form Ax = x has
a unique solution, the fixed point theorem also gives a practical method for finding the solution,
i.e. calculation of the ”successive approximations” (xn = An x0 ). In fact, as shown in the proof,
the approximations actually converge to the solution of the equation Ax = x. For this reason, the
fixed point theorem if often called the method of successive approximations.

Example 68.1. Let f be a function defined on the closed interval [a, b] which maps [a, b] into
itself and satisfies a Lipschitz condition

(1) |f (x1 ) − f (x2 )| 6 K|x1 − x2 |

with constant K < 1. Then f is a contraction map, and hence, by Theorem (68.1), the sequence

(2) x0 , x1 = f (x0 ), x2 = f (x1 ), . . .

converges to the unique root of the equation f (x) = x. In particular, ”contraction condition” (1)
holds if f has a continuous derivative f 0 on [a, b] such that

|f 0 (x)| 6 K < 1

Example 68.2. Consider the mapping A of n−dimensional space into itself given by the system
of linear equations

n
X
(3) yi = aij xj + bi (i = 1, . . . , n)
j=1

If A is a contraction mapping, we can use the method of successive approximations to solve the
equation Ax = x. The condition under which A is a contraction mapping depend on the choice of
metric. We now examine three cases:

1. The space R0n with metric

ρ(x, y) = max |xi − yi |
16i6n

In this case,
X
ρ(y, ye) = max |yi − yei | = max | aij (xj − x
ej )|
i i
j
X
6 max |aij ||(xj − x
ej )|
i
j
X X
6 max |aij | max |(xj − x
ej )| = (max |aij |)ρ(x, x
e)
i j i
j j

5
 and the contraction condition is now

X
(4) |aij | 6 α < 1 (j = 1, . . . , n)
j

2. The space R1n with metric

n
X
ρ(x, y) = |xi − yi |
i=1

Here

X X X
ρ(y, ye) = |xi − yei | = | aij (xj − x
ej )|
i i j
XX
6 |aij ||(xj − x
ej )|
i j
X
6 (max |aij |)ρ(x, x
e)
j
i

and the contraction condition is now

X
(5) |aij | 6 α < 1 (j = 1, . . . , n)
j

3. Ordinary Euclidean space Rn with metric

v
u n
uX
ρ(x, y) = t (xi − yi )2
i=1

Using the Cauchy-Schwarz inequality, we have

XX XX
ρ2 (y, ye) = ( ej ))2 6 (
aij (xj − x a2ij )ρ2 (x, x
e)
i j i j

and the contraction condition becomes

XX
(6) a2ij 6 α < 1
i j

6
Thus, if at least one of conditions (4-6) holds, there exist a unique point x = (x1 , x2 , . . . , xn ) such
that
n
X
xi = aij xj + bi (i = 1, . . . , n)
i=1
The sequence of successive approximations to this solution of the equation x = Ax are of the form

x0 = (x01 , x02 , . . . , x0n )

x1 = (x11 , x12 , . . . , x1n )

xk = (xk1 , xk2 , . . . , xkn )

where
n
X
xki = aij xk−1
j + bi
i=1
0
and we can choose any point x as the ”zeroth approximation”.
Each of the conditions (4-6) is sufficient for applicability of the method of successive approxima-
tions, but none of them is necessary. In fact, examples can be constructed in which each of the
conditions (4-6) is satisfied, but not the other two.

68.2 Contraction mapping and differential equations

The most interesting applications of Theorem (68.1) arise when the space R is a function space.
We can use this theorem to prove a number of existence and uniqueness theorems for differential
and integral equations.
Theorem 68.2. (Picard). Given a function f (x, y) defined and continuous on a plane domain
G containing the point (x0 , y0 ) suppose f satisfies a Lipschitz condition of the form
|f (x, y) − f (x, ye)| 6 M |y − ye|
in the variable y. Then there is an interval |x − x0 | 6 δ in which the differential equation

dy
(7) = f (x, y)
dx
has a unique solution
y = ϕ(x)
satisfying the initial condition
(8) ϕ(x0 ) = y0

7
Remark. By an n−dimensional domain we mean an open connected set in Euclidean n−space.

Proof. Together the differential equation (7) and the initial condition (8) are equivalent to the
integral equation

Z x
(9) ϕ(x) = y0 + f (t, ϕ(t))dt
x0

By the continuity of f , we have

(10) |f (x, y)| 6 K

in some domain G0 ⊂ G (in fact f is bounded on G0 ⊂ G) containing the point (x0 , y0 ). Choose
δ > 0 such that

1. (x, y) ∈ G0 if |x − x0 | 6 δ, |y − y0 | 6 Kδ

2. M δ < 1

and let C ∗ be the space of continuous functions ϕ defined on the interval |x − x0 | 6 δ and such
that |ϕ(x) − y0 | 6 Kδ, equipped with the metric

e = max |ϕ(x) − ϕ(x)|

ρ(ϕ, ϕ) e
x
∗
The space C is complete, since it is closed subspace of the space of all continuous functions on
[x0 − δ, x0 + δ]. Consider the mapping ψ = Aϕ defined by the integral equation
Z x
ψ(x) = y0 + f (t, ϕ(t))dt (|x − x0 | 6 δ)
x0

Clearly A is a contraction mapping carrying C ∗ into itself. In fact, if ϕ ∈ C ∗ , |x − x0 | 6 δ then

Z x
|ψ(x) − y0 | = | f (t, ϕ(t))dt| 6
x0
Z x
6 |f (t, ϕ(t))|dt 6
x0

6 K|x − x0 | 6 Kδ

by (10), and hence ψ = Aϕ also belongs to C ∗ . Moreover,

Z x
|ψ(x) − ψ(x)|
e 6 |f (t, ϕ(t)) − f (t, ϕ(t))|dt
e
x0

6 M δ|ϕ(t) − ϕ(t)|
e

8
and hence
ρ(ψ, ψ)
e 6 M δρ(ψ, ψ)
e

after maximizing with respect to x. But M δ < 1, so that A is a contraction mapping. It follows
from Theorem (68.1) that the equation ϕ = Aϕ, i.e. the integral equation (9), has a unique
solution in the space C ∗ .