1 Continuous Functions on Metric Spaces

1.1 The Definition of Continuity

Up to this point we have studied functions f : D → E where D and E are subsets of or n . We now
extend our discussion to functions f : M → N where M and N are metric spaces. Many concepts
and proofs that apply to Euclidean spaces can be easily transferred to general metric spaces.

Definition 1 A function f : M → N is continuous if f −1 (A) is open for all open sets A ⊂ N .

This definition is compatible with our previous definition of continuous functions on Euclidean
spaces, see Theorem ??.

Theorem 1 The following are equivalent.

1. f : M → N is continuous.
2. ∀ x0 ∈ M , and ∀ 1/m, ∃ 1/n such that
1 1
dM (x, x0 ) < ⇒ dN (f (x), f (x0 )) <
n m
3. ∀ {xj } ⊂ M , if xj → x0 in M , then f (xj ) → f (x0 ) in N .

1 ⇒ 2: Let x0 ∈ M and let 1/m be given. By assumption, the inverse image of an open ball is
open, so
1 1
f −1 (Bn (f (x0 ), )) = {x ∈ M | f (x) ∈ BN (f (x0 ), )}
m m
1
= {x ∈ M | dN (f (x), f (x0 )) < }
m
is open and contains x0 . Thus, ∃ 1/n such that
BM (x0 , 1/n) ⊂ f −1 (BN (f (x0 ), 1/m))
i.e., if dM (x, x0 ) < 1/n then dN (f (x), f (x0 )) < 1/m.

2 ⇒ 3: Let xj → x0 in M , and let 1/m be given. By assumption, ∃ 1/n such that Dn (f (xj ), f (x0 )) <
1/m whenever dM (xj , x0 ) < 1/n. However, since xj → x0 , ∃ J such that dM (xj , x0 ) < 1/n whenever
j ≥ J. So, if j ≥ J, then dN (f (xj ), f (x0 )) < 1/m showing that f (xj ) → f (x0 ).

3 ⇒ 1: Let A ⊂ N be open and suppose f −1 (A) is not open. Then ∃ x0 ∈ f −1 (A) such that
no open ball about x0 is contained in f −1 (A). Therefore, ∀ 1/j, ∃ xj ∈ BM (x0 , 1/j) such that
/ f −1 (A). Clearly, xj → x0 , so by assumption, f (xj ) → f (x0 ). Since A is open, ∃ 1/m such that
xj ∈
B(f (x0 ), 1/m) ⊂ A. But since f (xj ) → f (x0 ), ∃ J such that f (xj ) ∈ B(f (x0 ), 1/m) ⊂ A, ∀ j ≥ J,
and this implies xj ∈ f −1 (A), a contradiction.

1
1.1.1 Examples

Rb
1) Let M be C[a, b] with the sup-norm. Then I : M → defined by I(f ) = a f (x)dx is continuous.
To prove this, let fn → f with respect to the sup-norm. As we have seen this means that fn →
f uniformly. Therefore, f is integrable and I(fn ) → I(f ) by Theorem ??, showing that I is
continuous.

2) Let M = C 1 [a, b] with norm kf k∗ = kf k∞ + kf 0 k∞ and let N = C[a, b] with the sup-norm. Then
D : M → N defined by D(f ) = f 0 is continuous. To see this, note that if fn converges to f with
respect to the ∗-norm, then fn → f and fn0 → f 0 uniformly, see Theorems ?? and ??. Therefore,
D(fn ) → D(f ) with respect to the sup-norm, proving that D is continuous.

Recall that the uniform convergence of functions does not necessarily imply the convergence of
their derivatives. Hence, D would not be continuous with respect to the sup-norm on C 1 [a, b].
Continuity depends on the distance function under consideration.

3) Let M be a metric space and fix x0 ∈ M . Then f : M → defined by f (x) = d(x, x0 ) is

continuous: If xn → x in M , then, by definition, ∀ 1/m, ∃ N such that d(xn , x) < 1/m, ∀ n ≥ N .
Since d(xn , x0 ) ≤ d(xn , x) + d(x, x0 ) we find that
1
|f (xn ) − f (x)| = |d((xn , x0 ) − d(x, x0 )| ≤ d(xn , x) < , ∀n ≥ N
m
so f (xn ) → f (x0 ). Therefore, f is continuous.

Remark: The usual ways of combining continuous functions lead to new continuous functions.
For example, if f : M → N and g : N → P are continuous, then it is easy to prove that the
composition, f ◦ g : M → P , f ◦ g(x) = f (g(x)), is continuous. If f, g : M →n are continuous,
then f + g : M →N , (f + g)(x) = f (x) + g(x), is continuous. If f : M →n and g : M → are
continuous, then f · g : M →n , (f · g)(x) = f (x)g(x), and f /g : M →n , (f /g)(x) = f (x)/g(x)
(for x such that g(x) 6= 0), are continuous. If f1 , . . . , fn : M → are continuous, then f : M →n ,
f (x) = (f1 (x), . . . , fn (x)) is continuous and conversely.

1.2 Continuous Functions on Compact Domains

Let M and N be metric spaces with M compact. In this section we establish theorems for contin-
uous functions f : M → N that are analogous to familiar theorems about continuous real-valued
functions on compact intervals.

Definition 2 f : M → N is uniformly continuous if ∀ 1/m, ∃ 1/n such that ∀ x, y ∈ M

1 1
dM (x, y) < ⇒ dN (f (x), f (y)) <
n m

2
Theorem 2 If M is compact and f : M → N is continuous, then f is uniformly continuous.

Given 1/m and x ∈ M , ∃ 1/nx (depending on 1/m and x) such that

2 1
dM (x, y) < ⇒ dN (f (x), f (y)) < (1)
nx 2m
by the continuity of f . The open balls BM (x, 1/nx ), for x ∈ M , cover M , and since M is compact,
a finite number will cover M :

M = BM (x1 , 1/n1 ) ∪ · · · BM (xt , 1/nt )

Let n = max{n1 , . . . , nt }.

For any x, y ∈ M , x must lie in one of the open balls, x ∈ BM (xi , 1/ni ) for some i. If dM (x, y) < 1/n
then
1 1 2
dM (xi , y) ≤ dM (xi , x) + dM (x, y) < + ≤
ni n ni
so
1 1 1
dN (f (x), f (y)) ≤ dN (f (x), f (xi )) + dN (f (xi ), f (y)) < + =
2m 2m m
by (??).

Theorem 3 If M is compact and f : M → is continuous, then f has a maximum and minimum

in M , that is, there exists points u, v ∈ M such that

f (u) = sup f (x) and f (v) = inf f (x)

x∈M x∈M

There is a sequence {xn } ⊂ M such that lim f (xn ) = sup f (x) (or to ∞ if sup f (x) = ∞). Since M
is compact, there is a convergent subsequence, xn0 → b. Then sup f (x) = lim f (xn0 ) = f (b) by the
continuity of f , showing that the supremum is finite and achieved at a point b ∈ M . The proof for
the infimum is similar.

Theorem 4 Suppose f : M → N is continuous. If A ⊂ M is compact, then f (A) ⊂ N is compact.

S
The argument is word-for-word
S −1 the same as given in Theorem ??: Let f (A) ⊂ Uα be an open
cover of f (A). Then f (Uα ) is an open cover of A. Since A is compact, there exists a finite
subcover, A ⊂ f −1 (Uα1 ) ∪ . . . ∪ f −1 (Uαn ). But then f (A) ⊂ Uα1 ∪ . . . ∪ Uαn is a finite subcover
of f (A) proving that f (A) is compact.

1.3 Connectedness

Recall the Intermediate Value Theorem:

3
 If f : [a, b] → is continuous, then f ([a, b]) = [c, d] where c = min f (x) and d = max f (x).

The point of this theorem is that if there are no gaps in the domain of a continuous function, then
there are no gaps in the values of the function. To generalize this theorem to metric spaces we
need a way to specify that there are “no gaps.” There are two ways to define such a concept: being
“connected,” and being “arc-wise connected.” The first is the more general and is defined in terms
of open sets. The second concept is more restrictive, but is often more practical to apply.

Definition 3 Let M be a metric space.

1. M is connected if there does not exist a pair of disjoint, non-empty, open sets A and B with
M = A ∪ B.

2. M is arc-wise connected is there is a curve connecting any two points in M . More precisely,
given any two points x, y ∈ M , there exists a continuous function f : [a, b] → M such that
f (a) = x and f (b) = y.

Any interval of real numbers is connected.

Suppose I is an interval and I = A ∪ B with A and B disjoint, non-empty, open subsets of I. Fix
a ∈ A and b ∈ B. We may assume a < b. To produce a contradiction, we look for a dividing point
between A and B. Let
d = sup{x ∈ A | x ≤ b}
Let {xn } ⊂ A such that xn → d. Since A is the complement of the open set B in I, A is closed in
I. Therefore, d ∈ A unless d ∈/ I. The latter is only possible if d is an endpoint of the interval I,
which is not the case here since a ≤ d ≤ b. Therefore, d ∈ A. But A is also open, so there exists
an open interval around d contained in A, which contradicts d being the supremum of points in A
that are ≤ b.

Theorem 5 Let M be a metric space.

1. M is arc-wise connected ⇒ M is connected.

2. M is connected ⇐⇒ the only subsets of M that are both open and closed are ∅ and M .

1) Suppose M is arc-wise connected but not connected. Then M = A ∪ B where A and B are
disjoint, non-empty and open. Let x ∈ A and y ∈ B and let f : [a.b] → M be a continuous function
such that f (a) = x and f (b) = y. Then

[a, b] = f −1 (A ∪ B) = f −1 (A) ∪ f −1 (B)

decomposes [a, b] into disjoint, non-empty, open sets, contradicting Proposition ??. Therefore, M
must be connected.

4
2) (⇒) Suppose M is connected and A ⊂ M is both open and closed. Then B = Ac is open and
closed by Theorem ??. Both A and B cannot be non-empty, since otherwise M = A ∪ B with
A and B disjoint, non-empty, open subsets. Therefore, either A = ∅ and B = M , or B = ∅ and
A = M.

(⇐) Conversely, suppose the only subsets of M that are both open and closed are ∅ and M . If
M = A ∪ B with A and B disjoint and open in M , then A = B c and B = Ac are also be closed
by Theorem ??. By assumption, either A = ∅ and B = M , or B = ∅ and A = M , proving that M
cannot be decomposed into a disjoint union of non-empty open sets. Therefore, M is connected.

Example: It is possible for a set to be connected but not arc-wise connected. The graph of the
function 
sin(1/x) x > 0
f (x) =
0 x=0
is a connected subset of 2 : Any decomposition of the graph, G = {(x, f (x) | x ≥ 0}, into disjoint,
non-empty, open subsets would give a similar decomposition of the part of the graph, G0 , for x > 0
which is impossible since G0 is clearly arc-wise connected. However, there is no curve from (0, 0)
to any other point on the graph. Suppose f : [0, 1] → G is continuous with f (0) = (0, 0) and
f (1) = (a, sin(1/a)) for some a > 0. Then f (t) = (x(t), sin(x(t))) → (0, 0) as t → 0. This implies
x(t) → 0 and sin(1/x(t)) → 0, which is impossible since limu→0 sin(1/u) does not exist.

Theorem 6 Let M and N be metric spaces and suppose f : M → N is continuous. If M is

connected then f (M ) is connected. If M is arc-wise connected, then f (M ) is arc-wise connected.

Suppose M is connected. If f (M ) = A ∪ B with A and B disjoint, non-empty, open subsets, then

M = f −1 (A ∪ B) = f −1 (A) ∪ f −1 (B)

decomposes M into disjoint, non-empty, open subsets, a contradiction. Therefore, no such decom-
position of f (M ) exists and f (M ) is connected.

Next assume M is arc-wise connected. Let x, y ∈ f (M ). Then there are u, v ∈ M such that
x = f (u) and y = f (v). Let g : [a, b] → M be a continuous function such that g(a) = u and
g(b) = v. Then f ◦ g : [a, b] → f (M ) is continuous with f ◦ g(a) = x and f ◦ g(b) = y, proving that
f (M ) is arc-wise connected.

1.4 The Contractive Mapping Principle

In this section we consider a continuous function of a complete metric space M to itself, f : M → M ,

that “shrinks” distance. We shall prove that any such function must have a unique fixed point, that
is, a point x0 ∈ M such that f (x0 ) = x0 . This theorem has many applications in analysis. In fact,
we shall use it to prove two fundamentally important theorems:

5
 • The existence of solutions to ordinary differential equations.
• The Implicit Function Theorem.

Definition 4 Let M and N be metric spaces.

1. A mapping of M to N is a continuous function f : M → N .

2. A mapping f : M → M is contractive if ∃ r, 0 ≤ r < 1, such that

d(f (x), f (y)) ≤ r d(x, y), ∀ x, y ∈ M

We introduce the notation f n for the iterated mapping f ◦ f ◦ · · · ◦ f (n times).

Theorem 7 (Contractive Mapping Principle) Let M be a complete metric space and assume
f : M → M is a contractive mapping. Then there exists a unique fixed point x0 ∈ M , f (x0 ) = x0 .
Moreover, ∀ x ∈ M , ∃ c > 0 such that

d(f n (x), x0 ) ≤ crn

In particular, limn→∞ f n (x) = x0 .

Let x ∈ M . We first show the sequence {f n (x)} is Cauchy. The definition of a contractive mapping
implies
d(f n+1 (x), f n (x)) ≤ r d(f n (x), f n−1 (x)) ≤ · · · ≤ rn d(f (x), x)
For any j > n, the above inequality, along with the triangle inequality, implies

d(f j (x), f n (x)) ≤ d(f j (x), f j−1 (x)) + · · · + d(f n+1 (x), f n (x))
≤ (rj−1 + · · · + rn )d(f (x), x)
rn
≤ d(f (x), x)
1−r
Since r < 1, rn d(f (x), x)/(1 − r) → 0 as n → ∞, proving that the sequence {f n (x)} is Cauchy.

Because M is complete, the sequence {f n (x)} must converge to some point x0 ∈ M . This point is
a fixed point of f :
f (x0 ) = f ( lim f n (x)) = lim f n+1 (x) = x0
n→∞ n→∞
The previous inequality gives us the rate of convergence:

d(x0 , f n (x)) = lim d(f j (x), f n (x)) ≤ crn

j→∞

where c = d(f (x), x)/(1 − r).

The fixed point is unique since if f (x1 ) = x1 then

d(x0 , x1 ) = d(f (x0 ), f (x1 )) ≤ rd(x0 , x1 )

6
But r < 1, so d(x0 , x1 ) = 0 and hence x1 = x0 .

It can be difficult to apply the Contractive Mapping Principle. In order for a mapping f : M → M
to have the strong property of shrinking distances, one must often first restrict the domain to a
smaller subset M0 ⊂ M . But then one must verify that f (M0 ) ⊂ M0 in order to apply the theorem.
For example, if f (x) = x2 + 1, then |f (x) − f (y)| = |x2 − y 2 | = |x + y| · |x − y|. If we restrict
the domain to [−1/3, 1/3], then |f (x) − f (y)| ≤ (2/3)|x − y| which looks like the condition for a
contractive mapping. However, f ([−1/3, 1/3]) = [1, 10/9], so f does not map [−1/3, 1/3] to itself.

Example: Suppose f : [a, b] → [a, b] satisfies |f 0 (x)| < 1 for all x ∈ [a, b]. Then f is contractive
because ∀ x, y ∈ [a, b], the Mean Value Theorem implies there is a z between x and y such that

|f (x) − f (y)| = |f 0 (z)| · |x − y| ≤ r · |x − y|, r = max |f 0 (z)| < 1

z∈[a,b]

The Contractive Mapping Principle tells us that f has a fixed point in [a, b] that can be found by
repeatedly applying f to any starting number in [a, b]. To get a numerical approximation to the
fixed point, one could program f into a calculator, evaluate f on a, then repeatedly evaluate f on
the previous output until the numbers on the display do not change.

For example, consider f (x) = e−x on the interval [.3, .8]. Since f is decreasing, f ([.3, .8]) =
[f (.8), f (.3)] = [e−.8 , e−.3 ] ⊂ [.3, .8]. Moreover, |f 0 (x)| ≤ e−.3 < 1 for x ≥ .3, so f is contractive.
The Contractive Mapping Principle implies that there exists a unique number x0 ∈ [.3, .8] such that
f (x0 ) = x0 , i.e., such that e−x0 = x0 . If we start with 0.5 and repeatedly press the “button” for
the function e−x on the calculator, we obtain the sequence: 0.500000, 0.606531, 0.545239, 0.579703,
0.560065, 0.571172, 0.564863, 0.568438, 0.566409, 0.567560, 0.566907, 0.567277, 0.567067, 0.567186,
0.567119, 0.567157, 0.567135, 0.567148, 0.567141, 0.567145, 0.567142, 0.567144, 0.567143, . . . . The
digits stabilize at 0.567143, and we find that e−0.567143 = 0.567143.

Actually, we do not need a contractive mapping f : [a, b] → [a, b] in order to find a fixed point.

Any mapping f : [a, b] → [a, b] has a fixed point.

To prove this, consider g(x) = f (x) − x. Since g(a) = f (a) − a ≥ 0 and g(b) = f (b) − b ≤ 0,
the Intermediate Value Theorem implies that there exists c ∈ [a, b] such that g(c) = 0 and hence
f (c) = c. This is a special case of a more general fact known as the Brouwer Fixed Point Theorem:

Any mapping of a closed ball in n into itself has a fixed point.

It is necessary to have a closed ball for this theorem, even in the case of intervals. The function
f (x) = x/2 maps (0, 1) to itself, but does not have a fixed point in (0, 1). The disadvantage of
Brouwer’s theorem over the Contractive Mapping Principle is that it is not constructive. We know
the fixed point must exist, but the theorem does not tell us how to find it.