Qamrul Hasan Ansari Metric Spaces Page 1

Metric Spaces
Theory and Problems

Qamrul Hasan Ansari

Department of Mathematics
Aligarh Muslim University, Aligarh
E-mail: qhansari@gmail.com

Copyright@QH Ansari, Department of Mathematics, AMU, Aligarh

 SYLLABUS
B.A. / B.Sc. VI SEMESTER

METRIC SPACES

Course Title Metric Spaces

Course Number MMB-3019
Credits 4
Course Category Compulsory
Prerequisite Courses Real Analysis and Mathematical Analysis
Contact Course 4 Lecture + 1 Tutorial
Type of Course Theory
Course Assessment Sessional (1 hour) 30%
End Semester Examination (2:30 hours) 70%
Course Objectives To give the idea of distance between two elements in a set
and to extend the concepts, namely, open sets, closed sets,
convergence of sequences, compact sets, continuity of functions, etc,
from real line to a metric space.
The course focuses on basic notions of metric spaces
and their properties.
Course Outcomes Knowledge: A student will know:
• the basic idea of distance between two points in a set,
• the notions from metric spaces, namely, metric on a set,
completeness and compactness of a metric space,
• methods and techniques to prove basic results on
metric spaces, continuous mappings and fixed point theorems
Skills: A student can:
• check if a given function is a metric,
• check if a given function is continuous,
• check if a given set is open, closed, dense, compact
Final course output - social competence
A student will know the importance of metric spaces
in mathematics and its applications in different areas.

2
Qamrul Hasan Ansari Metric Spaces Page 3

Syllabus No. of Lectures

UNIT I: Some Concepts
Definition and examples of metric spaces,
Bounded and unbounded metric spaces,
Distance between sets, Diameter of a set,
Open and closed balls, Interior points and interior of a set, 12
Open set, Neighbourhood of a point, Limit point of a set,
Closure of a set, Closed set, Boundary points and boundary of a set,
Exterior points and exterior of a set, Subspace of a metric space
UNIT II: Completeness and Separability
Sequences and subsequence in a metric space,
Convergent and Cauchy sequences, Complete metric spaces,
Relation between completeness and closedness, 12
Cantor Intersection Theorem, Completion Theorem,
Dense sets, Separable spaces, Nowhere dense sets,
Categories and Baire Category Theorem
UNIT III: Compactness
Cover of a metric space, Compact metric spaces,
Compact sets and their criterion, Properties of compact sets,
Relation between compactness, completeness and closedness, 12
Finite Intersection property, Bolzano-Weierstrass property,
Sequential compactness, Totally bounded spaces
UNIT IV: Continuity and Fixed Points
Continuous functions between two metric spaces,
Characterizations of Continuous functions,
Continuous functions on compact spaces,
Uniform continuous functions, Homeomorphism and Isometry, 12
Equicontinuity and Ascoli-Arzela Theorem,
Fixed points, Banach contraction theorem.
Total 48
Qamrul Hasan Ansari Metric Spaces Page 4

Recommended Books:

1. Q. H. Ansari: Metric Spaces Including Fixed Point Theory and Set-valued Maps, Narosa Publishing
House, New Delhi. 2010.

2. E. T. Copson: Metric spaces, Cambridge University Press, 1968.

3. J. K. Hunter and B. Nachtergaele, Applied Analysis, World Scientific, New Jersey, London, Sin-
gapore, Hong Kong, 2001.

4. S. Kumaresan: Topology of Metric Spaces, Narosa Publishing House, 2nd Ed, 2011.

5. J. Lebl, Basic Analysis I & II: Introduction to Real Analysis, 2022,

https://www.jirka.org/ra/realanal.pdf

6. M. O. Searcoid: Metric spaces, Springer, 2007.

7. G. F. Simmons, Introduction to Topology and Modern Analysis, McGraw-Hill, 1963.

8. S. Shirali and H. Vasudeva, Metric Spaces, Springer-Verlag, London, 2006.

9. W. A. Sutherland: Introduction to Metric and Topological Spaces, Second Edition, Oxford Uni-
versity Press, Oxford, New York, 2009.
 Chapter 1

Basic Concepts

In our daily life, we observe that the distance be-

tween any two points is always positive, it van-
ishes if and only if two points lie on the same posi-
tion, the distance between two points is the same
from whichever point it is measured, and the dis-
tance from one point to another do not exceed
sum of the distances from these points to an in-
termediate point. By exploiting these fundamen-
tal properties of a distance, French mathemati-
cian René Maurice Fréchet (2 September 1878 -
4 June 1973) first introduced the concept of a
metric (distance function) between two elements
of a nonempty set in 1906 by unifying the work
on function spaces by Cantor, Volterra, Arzelà,
Hadamard, Ascoli, and others.
René Maurice Fréchet

5
Qamrul Hasan Ansari Metric Spaces Page 6

1.1 Definition and Examples of Metric Spaces

Definition 1.1.1

Let X be a nonempty set. A real-valued function d defined on X × X is said to be a

metric on X if it satisfies the following conditions:

(M1) d(x, y) ≥ 0, for all x, y ∈ X;

(M2) d(x, y) = 0 if and only if x = y;

(M3) d(x, y) = d(y, x), for all x, y ∈ X; (symmetry)

(M4) d(x, y) ≤ d(x, z) + d(z, y), for all x, y, z ∈ X. (triangle inequality)

The set X together with a given metric d on X is called a metric space and it is denoted
by (X, d). If there is no confusion likely to occur we, sometimes, denote the metric space
(X, d) by X.

Since d is the generalization of the distance function, we interpret d(x, y) as the distance
between two elements x and y of the set X. Sometimes d is called a distance function on X.

The triangle inequality may be interpreted as that “the length of one side of a triangle can
not exceed the sum of the length of the other two sides”. Equivalently, the distance from x
to y via any intermediate point z can not be shorter than the direct distance from x to y.

z
d(
, z)

z,
y)
d (x

x y
d(x, y)
Qamrul Hasan Ansari Metric Spaces Page 7

The triangle inequality can be extended for a finite number of elements as follows.

From the triangle inequality (M4), it follows by induction that for any x, y, z1 , z2 , . . ., zn ∈ X,

d(x, y) ≤ d(x, z1 ) + d(z1 , y)

≤ d(x, z1 ) + d(z1 , z2 ) + d(z2 , y)
≤ d(x, z1 ) + d(z1 , z2 ) + d(z2 , z3 ) + d(z3 , y)
······························
≤ d(x, z1 ) + d(z1 , z2 ) + · · · + d(zn , y).

Examples of Metric Spaces

Example 1.1.1

Let X = R, the set of all real numbers. For any x, y ∈ X, define

d(x, y) = |x − y|.

Then (X, d) is a metric space and the metric d is called an usual metric on R.

Verification. For all x, y, z ∈ X, we have

(M1) d(x, y) ≥ 0;
(M2) d(x, y) = |x − y| = 0 if and only if x = y;
(M3) d(x, y) = |x − y| = | − (x − y)| = |y − x| = d(y, x);
(M4) d(x, y) = |x − y| = |(x − z) + (z − y)|
≤ |x − z| + |z − y|
= d(x, z) + d(z, y).

In the verification of (M4) in Example 1.1.1, we used the fact that |a + b| ≤ |a| + |b| for real
numbers a and b. This property also holds for complex numbers a and b. Hence, we have
the following example.
Qamrul Hasan Ansari Metric Spaces Page 8

Example 1.1.2

Let X = C, the set of all complex numbers. For any x, y ∈ X, define

d(x, y) = |x − y| .

Then (X, d) is a metric space and the metric d is called an usual metric on C.

Example 1.1.3

Let X be any nonempty set. For any x, y ∈ X, define


0, if x = y,
d(x, y) =
1, if x 6= y.

That is, all points are equally distant from each other. Then (X, d) is a metric space.
The metric d is called a discrete metric and the space (X, d) is called a discrete metric
space.

1 1

e 1 c

1 1

1 1 1 1

a 1 b

The simple discrete metric space X = {a, b, c, d, e}, where the distance between any two
points is 1.
Qamrul Hasan Ansari Metric Spaces Page 9

Remark 1.1.1

Example 1.1.3 shows that on each nonempty set, we can always define at least one
metric that is a discrete metric.

Example 1.1.4

Let X = R, the set of all real numbers. For any x, y ∈ X, define

d(x, y) = |f (x) − f (y)|,

for any strictly monotonic function (say, increasing) f : R → R. Then d is a metric on

R.

Note that the strict monotonicity of f is needed to ensure the positivity of d.

Example 1.1.5

Let X = R2 , the set of all points in the coordinate plane. For any x = (x1 , x2 ),
y = (y1 , y2 ) in X, define
q
d(x, y) = (x1 − y1 )2 + (x2 − y2 )2 .

Then (X, d) is a metric space and d(x, y) is the natural distance between two points in
a plane. d is known as the usual metric on R2 .

R
(x1 , x2 )
y) •
x,
d( x2 − y2
(y
1,
y2•
)
x1 − y1 R

Verification. Let x = (x1 , x2 ), y = (y1 , y2 ) and z = (z1 , z2 ) be in X. Obviously,

d(x, y) ≥ 0 and so (M1) holds.
Qamrul Hasan Ansari Metric Spaces Page 10

(M2)
q
d(x, y) = 0 ⇔ (x1 − y1 )2 + (x2 − y2 )2 = 0
⇔ (x1 − y1 )2 = 0 and (x2 − y2 )2 = 0
⇔ x1 = y1 and x2 = y2
⇔ x = y.

(M3)
q
d(x, y) = (x1 − y1 )2 + (x2 − y2 )2
q
= (y1 − x1 )2 + (y2 − x2 )2
= d (y, x) .

(M4)

[d(x, y)]2 = (x1 − y1 )2 + (x2 − y2 )2

= [(x1 − z1 ) + (z1 − y1 )]2 + [(x2 − z2 ) + (z2 − y2 )]2
= (x1 − z1 )2 + (z1 − y1 )2
 

+2 (x1 − z1 )(z1 − y1 ) + (x2 − z2 )(z2 − y2 )

| {z } | {z } | {z } | {z }
a b c d
2 2
+ (x2 − z2 ) + (z2− y2 ) .

Taking x1 − z1 = a, z1 − y1 = b, x2 − z2 = c and z2 − y2 = d, and since

(ab + cd)2 ≤ (a2 + c2 )(b2 + d2 ),

we have

[d(x, y)]2 ≤ (x1 − z1 )2 + (x2 − z2 )2

p p
+2 (x1 − z1 )2 + (x2 − z2 )2 (z1 − y1 )2 + (z2 − y2 )2
+(z1 − y1 )2 + (z2 − y2 )2
= [d (x, z)]2 + 2d (x, z) d (z, y) + [d (z, y)]2
= [d (x, z) + d (z, y)]2 .

Therefore,
d (x, y) ≤ d (x, z) + d (z, y) .
Qamrul Hasan Ansari Metric Spaces Page 11

Example 1.1.6

Let X = R2 . For any x = (x1 , x2 ), y = (y1 , y2 ) in X, define

d(x, y) = |x1 − y1 | + |x2 − y2 | .

Then d is a metric on X and (X, d) is a metric space.

The distance between x = (x1 , x2 ) and y = (y1 , y2 ) with respect to the metric d(x, y) =
|x1 − y1 | + |x2 − y2 | is represented by the dark line in the following figure.

R (x1 , x2 )
•

|x2 − y2 |
2)
1, y

•
(y

|x1 − y1 |
R

Verification. The conditions (M1), (M2) and (M3) are obvious. We prove only condition
(M4). For any x = (x1 , x2 ), y = (y1 , y2 ) and z = (z1 , z2 ) in X, we have

d(x, y) = |x1 − y1 | + |x2 − y2 |

≤ |x1 − z1 | + |z1 − y1 | + |x2 − z2 | + |z2 − y2 |
= d(x, z) + d(z, y).

Example 1.1.7

Let X = R2 . For any x = (x1 , x2 ), y = (y1 , y2 ) in X, define

d(x, y) = max {|x1 − y1 | , |x2 − y2 |} .

Then d is a metric on X and (X, d) is a metric space.

Qamrul Hasan Ansari Metric Spaces Page 12

The distance between x = (x1 , x2 ) and y = (y1 , y2 ) with respect to the metric d(x, y) =
max {|x1 − y1 | , |x2 − y2 |} is represented by the dark line in the following figure.

R (x1 , x2 )
•

|x2 − y2 |

2)
1, y
•

(y
|x1 − y1 |
R

Remark 1.1.2

Examples 1.1.5, 1.1.6 and 1.1.7 show that more than one metric can be defined on a
nonempty set.

Example 1.1.8

Let X = Rn , the set of ordered n-tuples of real numbers. For any x = (x1 , x2 , . . . , xn ) ∈
X and y = (y1 , y2 , . . . , yn ) ∈ X, we define

n
X
(a) d1 (x, y) = |xi − yi |; (called taxicab metric)
i=1

n
! 12
X
(b) d2 (x, y) = (xi − yi )2 ; (called usual metric)
i=1

n
! p1
X p
(c) dp (x, y) = |xi − yi | , p ≥ 1;
i=1

(d) d∞ (x, y) = max {|xi − yi |}. (called max metric)

1≤i≤n

Then d1 , d2 , dp (p ≥ 1) and d∞ are metrics on X.

Verification. In view of Examples 1.1.5, 1.1.6 and 1.1.7, it is easy to verify that d1 , dp
and d∞ are metrics on X.
Qamrul Hasan Ansari Metric Spaces Page 13

The triangular inequality (M4) in the case of dp requires the use of Minkowski inequal-
ity (Theorem A.0.3). For x = (x1 , x2 , . . . , xn ) ∈ Rn , y = (y1 , y2 , . . . , yn ) ∈ Rn , and
z = (z1 , z2 , . . . , zn ) ∈ Rn , we have

n
! 1p n
! p1
X p
X p
dp (x, y) = |xi − yi | = |xi − zi + zi + yi |
i=1 i=1
n
! p1
X p
≤ (|xi − zi | + |zi − yi |)
i=1
n
! p1 n
! p1
X p
X p
≤ |xi − zi | + |zi − yi |
i=1 i=1
= dp (x, z) + dp (z, y).

Remark 1.1.3

Let X = Cn , the set of ordered n-tuples of complex numbers. We can define the metrics
d1 , dp and d∞ on X in the same way as in Example 1.1.8.

Example 1.1.9

Let ℓ∞ be the space of all bounded sequences of real or complex numbers, that is,
 
∞
ℓ = {xn } ⊆ R or C : sup |xn | < ∞ .
1≤n<∞

For any x = {xn } ∈ ℓ∞ and y = {yn } ∈ ℓ∞ , define

d∞ (x, y) = sup |xn − yn | .

1≤n<∞

Then it is easy to verify that d∞ is a metric on ℓ∞ and (ℓ∞ , d∞ ) is a metric space.

Qamrul Hasan Ansari Metric Spaces Page 14

Example 1.1.10

Let s be the space of all sequences of real or complex numbers, that is,

s = {{xn } ⊆ R or C} .

For any x = {xn } and y = {yn } in s, define

∞
X 1 |xn − yn |
d(x, y) = n 1 + |x − y |
.
n=1
2 n n

Then (s, d) is a metric space.

∞
X 1 |xn − yn |
Verification. The series is convergent since its nth term is less than
n=1
2n 1 + |xn − yn |
1
2n
. The conditions (M1), (M2) and (M3) can be easily verified.

Let x = {xn }, y = {yn } and z = {zn } be in s. Then by the triangular inequality, we have

|xn − yn | ≤ |xn − zn | + |zn − yn | ,

and hence1 ,

|xn − yn | |xn − zn | + |zn − yn |

≤
1 + |xn − yn | 1 + |xn − zn | + |zn − yn |
|xn − zn | |zn − yn |
= +
1 + |xn − zn | + |zn − yn | 1 + |xn − zn | + |zn − yn |
|xn − zn | |zn − yn |
≤ + .
1 + |xn − zn | 1 + |zn − yn |
1
Multiplying both the sides by 2n
and summing with respect to n, we obtain
∞ ∞ ∞
X 1 |xn − yn | X 1 |xn − zn | X 1 |zn − yn |
≤ + ,
n=1
2 1 + |xn − yn | n=1 2 1 + |xn − zn | n=1 2n 1 + |zn − yn |
n n

that is,
d(x, y) ≤ d(x, z) + d(z, y).

1
Let 0 ≤ α ≤ β. Then α + αβ ≤ β + αβ. Dividing both the sides by (1 + α)(1 + β), we have

α β
≤ .
1+α 1+β
Qamrul Hasan Ansari Metric Spaces Page 15

Example 1.1.11

Let 1 ≤ p < ∞. Consider the space ℓp of all sequences {xn } of real or complex numbers
X∞
such that |xn |p < ∞. Let x = {xn } and y = {yn } ∈ ℓp , we define
n=1

∞
! p1
X
d(x, y) = |xn − yn |p .
n=1

Then d is a metric on ℓp and (ℓp , d) is a metric space.

Verification. The conditions (M1), (M2) and (M3) can be easily verified. Let x = {xn },
y = {yn } and z = {zn } be sequences in ℓp . Then,

∞
! p1
X
d(x, y) = |xn − yn |p
n=1
∞
! p1
X
= |xn − zn + zn − yn |p
n=1
∞
! 1p ∞
! p1
X X
≤ |xn − zn |p + |zn − yn |p
n=1 n=1
(by Minkowski’s inequality)
= d(x, z) + d(z, y).

Example 1.1.12

Let B[a, b] be the space of all bounded real-valued functions defined on [a, b], that is,

B[a, b] = {f : [a, b] → R : f (t) ≤ k for all t ∈ [a, b] for some constant k ∈ R} .

For f, g ∈ B[a, b], we define

d(f, g) = sup |f (t) − g(t)| .

t∈[a,b]

Then (B[a, b], d) is a metric space.

Qamrul Hasan Ansari Metric Spaces Page 16

Example 1.1.13

Let C[a, b] be the space of all continuous real-valued functions defined on [a, b]. For
any f, g ∈ C[a, b], we define the real-valued functions d∞ and d1 on C[a, b] × C[a, b] as
follows:
d∞ (f, g) = sup |f (t) − g(t)| ,
t∈[a,b]

and Z b
d1 (f, g) = |f (t) − g(t)| dt,
a
where the integral is the Riemann integral which is possible because the functions f
and g are continuous on [a, b]. Then d∞ and d1 are metrics on C[a, b].

R
(t, g(t))
f

d∞ (f, g)

g
(t, f (t))

a t b R

The metric d∞ measures the distance from f to g as the maximum of vertical distances from
the points (t, f (t)) to (t, g(t)) on the graphs of f and g, respectively. d1 (f, g) represents as
a measure of the distance between the functions f and g to be the area enclosed between
their graphs from x = a to x = b.
Qamrul Hasan Ansari Metric Spaces Page 17

We can also define another metric on C[a, b]. Let f, g ∈ C[a, b], define
Z b  p1
d(f, g) = |f (t) − g(t)|p dt , for p ≥ 1.
a

Then (C[a, b], d) is a metric space.

d1 (f, g)

a b R

Remark 1.1.4

Let X be the set of all Riemann integrable functions on [a, b]. For f, g ∈ X, we define
Z b
d(f, g) = |f (x) − g(x)|dx.
a

Then d is not a metric on X. Indeed, if


0, if x = a,
f (x) =
1, if a < x ≤ b,

and g(x) = 1 for all x ∈ [a, b], then d(f, g) = 0. However, f 6= g. It may be noted
that all other properties of a metric do hold. Such real-valued functions are called
pseudometrics on X.
Qamrul Hasan Ansari Metric Spaces Page 18

Definition 1.1.2

Let X be a nonempty set. A real-valued function d : X × X → R+ is said to be a

pseudometric on X if it satisfies the following conditions:

(PM1) d(x, y) ≥ 0, for all x, y ∈ X;

(PM2) d(x, y) = 0 if x = y;

(PM3) d(x, y) = d(y, x), for all x, y ∈ X; (symmetry)

(PM4) d(x, y) ≤ d(x, z) + d(z, y), for all x, y, z ∈ X. (triangle inequality)

The set X together with a given pseudometric d on X is called a pseudometric space.

Example 1.1.14

Let X = R2 , x = (x1 , x2 ), y = (y1 , y2 ) ∈ X and d(x, y) = |x1 − y1 |. Then d is not a

metric on X; however, it is a pseudometric on X. Indeed, for x = (0, 0), y = (0, 1) ∈ X,
we have d(x, y) = 0 but x 6= y. Therefore, it is not a metric on X. It can be easily
checked that d satisfies the conditions (PM1) – (PM4).

Exercise 1.1.1. Let (X, d) be a metric space. Prove that

|d(x, z) − d(y, z)| ≤ d(x, y), for all x, y, z ∈ X.

Proof. It follows from (M3) and (M4) that

d(x, z) ≤ d(x, y) + d(y, z),

and
d(y, z) ≤ d(y, x) + d(x, z) = d(x, y) + d(x, z).
Thus,
−d(x, y) ≤ d(x, z) − d(y, z) ≤ d(x, y),
and therefore,
|d(x, z) − d(y, z)| ≤ d(x, y).

Exercise 1.1.2. Let (X, d) be a metric space. Prove that

|d(x, y) − d(z, w)| ≤ d(x, z) + d(y, w), for all x, y, z, w ∈ X.

Qamrul Hasan Ansari Metric Spaces Page 19

Proof. Since
|d(x, z) − d(z, y)| ≤ d(x, y),
we have
|d(x, y) − d(y, z)| ≤ d(x, z),
and
|d(y, z) − d(z, w)| ≤ d(y, w).
Hence,

|d(x, y) − d(z, w)| ≤ |d(x, y) − d(y, z)| + |d(y, z) − d(z, w)| ≤ d(x, z) + d(y, w).

Exercise 1.1.3. Let X be a nonempty set and d : X × X → R be a real-valued function

such that the following conditions hold: For all x, y, z ∈ X,

(i) d(x, y) = 0 if and only if x = y;

(ii) d(x, y) = d(y, x);

(iii) d(x, y) ≤ d(x, z) + d(z, y).

Prove that d is a metric on X.

Hint: Put y = x in (iii), we get 0 = d(x, x) ≤ d(x, z) + d(z, x). By (ii), 0 ≤ 2d(x, z) and so
d(x, z) ≥ 0 for all x, z ∈ X.

Exercise 1.1.4. Let X be a nonempty set and d : X × X → R be a real-valued function

such that the following conditions hold: For all x, y, z ∈ X,

(i) d(x, y) = 0 if and only if x = y;

(ii) d(x, y) ≤ d(z, x) + d(z, y).

Prove that d is a metric on X.

Exercise 1.1.5. Let X be a nonempty set and d : X × X → R be a real-valued function

such that the following conditions hold: For all x, y, z ∈ X,

(i) d(x, y) = 0 if and only if x = y;

(ii) d(x, y) ≤ d(x, z) + d(z, y).

Show that these conditions are not sufficient to make the function d a metric on the set X.
Qamrul Hasan Ansari Metric Spaces Page 20

Exercise 1.1.6. Let X = R2 and x = (0, 2), y = (3, 6) in X. Find the distance between x
and y by using the metrics of Examples 1.1.5, 1.1.6 and 1.1.7.
Exercise 1.1.7. Let c be the space of all convergent sequences of real or complex numbers.
For x = {xn } and y = {yn } in c, let

d(x, y) = sup |xn − yn | .

1≤n<∞

Prove that d is a metric on c.

Exercise 1.1.8. Let d be a metric on a set X. For all x, y ∈ X and α > 0, let

d1 (x, y) = αd(x, y),

d2 (x, y) = min{1, d(x, y)},
d3 (x, y) = (d(x, y))2 .

Prove that d1 and d2 are metrics on X, but d3 may or may not be a metric on X.
Exercise 1.1.9. Let K be the set of all real or complex numbers. Prove that for each
x, y ∈ K, 
0, if x = y,
d(x, y) =
|x| + |y|, if x 6= y,
is a metric on K.
Exercise 1.1.10. Let X = [0, 1) and d(x, y) = |x − y| for all x, y ∈ X. Prove that d is a
metric on X.
Exercise 1.1.11. Let X = Q, the set of all rational numbers. Show that

d(x, y) = |x − y|, for all x, y ∈ X,

is a metric on X.
Exercise 1.1.12. Let X = {x ∈ R : x > 0} and

1 1
d(x, y) = − , for all x, y ∈ X.
x y
Show that d is a metric on X.
Exercisep 1.1.13. Let X = R and d : X × X → R be a real-valued function defined by
d(x, y) = |x − y| for all x, y ∈ X. Show that d is metric on X.

Hint: For any x, y, z ∈ X, we have

[d(x, y)]2 = |x − y| ≤ |x − z| + |z − y|
= [d(x, z)]2 + [d(z, y)]2
≤ [d(x, z) + d(z, y)]2 .
Qamrul Hasan Ansari Metric Spaces Page 21

Exercise 1.1.14. Let X = Q, the set of all rational numbers. Let all pq , rs ∈ Q be written
 
p r
in their lowest terms, q, s > 0, and d q , s = 1q − 1s . Is d a metric on X?

Hint: Consider 31 , 32 ∈ Q in their lowest terms. Then d 1 2
,
3 3
= 1
3
− 1
3
= 0 but 1
3
6= 23 .
1 1
Exercise 1.1.15. Show that X = R \ Q with d(x, y) = x
− y
for all x, y ∈ X, is a metric
space on X.
Exercise 1.1.16. Show that C with

min{|z1 | + |z2 |, |z1 − 1| + |z2 − 1|}, if z1 6= z2 ,
d(z1 , z2 ) =
0, otherwise,
is a metric space.
Exercise 1.1.17. Determine whether the following real-valued functions defined on R × R
are metrics on R. For all (x, y) ∈ R × R,

(a) d1 (x, y) = [|x − y|], the greatest integer less than or equal to |x − y|;
(b) d2 (x, y) = ln |x − y|;
(c) d3 (x, y) = | sin(x − y)|;
(d) d4 (x, y) = e|x−y| ;
(e) d5 (x, y) = | cos(x − y)|;
(f) d6 (x, y) = |x2 − y 2|;
(g) d7 (x, y) = 2|x − y|;
(h) d8 (x, y) = (x − y)3;
(i) d9 (x, y) = |x − y|3;
(j) d10 (x, y) = |xy|.
Exercise 1.1.18. Let X = R and d : X × X → R be a real-valued function defined by

|x| + |x − y| + |y|, if x 6= y,
d(x, y) =
0, otherwise.

Show that (X, d) is a metric space.

Exercise 1.1.19. Let X = R2 and for each x = (x1 , x2 ), y = (y1 , y2 ) ∈ X, let

|y1 − y2 |, if x1 = x2 ,
d(x, y) =
|y1| + |y2 | + |x1 − x2 |, otherwise.

Prove that (X, d) is a metric space. It is known as a jungle river metric on R2 .

Qamrul Hasan Ansari Metric Spaces Page 22

Exercise 1.1.20. Let X = R2 and d : X × X → R be a real-valued function defined by


|x1 − y1 |, if x2 = y2 ,
d(x, y) = d((x1 , x2 ), (y1 , y2 )) =
|x1 | + |x2 − y2 | + |y1|, otherwise.

Show that (X, d) is a metric space.

Hint: Let x = (x1 , x2 ), y = (y1 , y2) and z = (z1 , z2 ) be in X. Observe that d(x, y) ≥ |x1 −y1 |.
If x2 = y2 , then d(x, y) = |x1 − y1 | ≤ |x1 − z1 | + |z1 − y1 | ≤ d(x, z) + d(z, y). If x2 6= y2 , then
z2 cannot be equal to both x2 and y2 ; so assume that z2 6= x2 . Then,

d(x, y) = |x1 | + |x2 − y2 | + |y1 | ≤ |x1 | + |x2 − z2 | + |z2 − y2 | + |y1|


(|x1 | + |x2 − z2 | + |z1 |) + |z1 − y1 |, if y2 = z2 ,
≤
(|x1 | + |x2 − z2 | + |z1 |) + (|z1 | + |z2 − y2 | + |y1|) , 6 z2
if y2 =
= d(x, z) + d(z, y).

Exercise 1.1.21. Let X = R2 and

d(x, y) = min{|x1 − y1 |, |x2 − y2 |}, for all x = (x1 , x2 ), y = (y1 , y2 ) ∈ X.

Is d a metric on X?

Hint: Choose x = (2, 3), y = (2, 4). Then x 6= y, but d(x, y) = min{0, 1} = 0. Hence d is
not a metric on X.
Exercise 1.1.22. Show that X = R2 with d(x, y) = d((x1 , y1 ), (x2 , y2 )) = |x2 − y2 | is not a
metric space, where x = (x1 , x2 ), y = (y1 , y2 ) ∈ X.
Exercise 1.1.23. Let X = C[0, 1] and d∞ : X × X → R be a real-valued function defined
by
d∞ (f, g) = sup |f (t) − g(t)|, for all f, g ∈ C[0, 1].
t∈[0,1]

Then calculate the distance between f (t) = t and g(t) = t2 .

1
Hint: f (t) − g(t) = t − t2 = −( 21 − t)2 + 1
4
and so sup |f (t) − g(t)| = .
t∈[0,1] 4
Exercise 1.1.24. Let X be the set of continuously differentiable functions on [0, 1]. Deter-
mine whether the following real-valued functions defined on X × X are metrics on X. For
all (f, g) ∈ X × X,

(a) d1 (f, g) = max |f ′ (x) − g ′(x)|;

x∈[0,1]

(b) d2 (f, g) = max |f (x) − g(x)| + max |f ′ (x) − g ′(x)|.

x∈[0,1] x∈[0,1]

Exercise 1.1.25. Let d1 and d2 be metrics on a set X. Check whether the following functions
are also metrics on X? Justify your answer.
Qamrul Hasan Ansari Metric Spaces Page 23

(a) d1 + d2 ;

(b) max{d1 , d2 };

(c) min{d1 , d2 };

(d) 14 d1 + 43 d2 ;

(e) d1 × d2 .

Hint: (a), (b) and (d) are metrics, however, (c) and (e) are not because R y triangle inequality
fails
Ry in these cases. For example, consider X = [0, 1] and d 1 (x, y) = x
t dt and d2 (x, y) =
1
x
(1 − t) dt are metrics on X. Then for x = 0, y = 2 , z = 1, we have
nR R1 o 
1
min{d1 , d2}(x, z) = min 0
t dt , 0
(1 − t) dt = min 21 , 21 = 12 .
nR R 1/2 o 1 3
1/2
min{d1 , d2 }(x, y) = min 0
t dt , 0
(1 − t) dt = min ,
8 8
= 81 .
nR R1 o 
1
min{d1 , d2}(y, z) = min 1/2
t dt , 1/2
(1 − t) dt = min 83 , 81 = 18 .

Hence the triangle inequality does not satisfy.

For (e), consider the metric space X = R with the usual metric and take d1 (x, y) = d2 (x, y) =
|x − y| for all x, y ∈ X. Then the triangle inequality does not satisfy for x = 0, y = 21 and
z = 1.
Exercise 1.1.26. Let (Xi , di ), i = 1, 2, . . . , n, be metric spaces and X = X1 × X2 × · · · × Xn .
Prove that for each x = (x1 , x2 , . . . , xn ) ∈ X and y = (y1 , y2 , . . . , yn ) ∈ X,
ˆ y) = max di (xi , yi ),
d(x,
1≤i≤n

and n
X
˜ y) =
d(x, di (xi , yi ),
i=1

are metrics on X.
Exercise 1.1.27. Let (X, d) be a metric space. Prove that

d(x, y)
d∗ (x, y) = , for all x, y ∈ X,
1 + d(x, y)
is also a metric on X.

a b a+b
Hint: Use + ≥ for all a ≥ 0, b ≥ 0.
1+a 1+b 1+a+b
Exercise 1.1.28. Let (X, d) be a metric space and t be a fixed positive number. Show that
d∗ (x, y) = min{t, d(x, y)} is also a metric on X.
Qamrul Hasan Ansari Metric Spaces Page 24

Exercise 1.1.29. Let S = {(cos θ, sin θ) : 0 ≤ θ ≤ 2π} be the unit circle in R2 . If pi =

(cos θi , sin θi ) ∈ S for i = 1, 2, define d(p1 , p2 ) = |θ1 − θ2 |. Show that d defines a metric on S.

Exercise 1.1.30. Let (X, d1 ) and (Y, d2 ) be two metric spaces such that X ∩ Y = ∅. Let
x0 ∈ X and y0 ∈ Y be any fixed elements. Define d : (X ∪ Y ) × (X ∪ Y ) → [0, ∞) by


 d1 (a, b), if a, b ∈ X,

d2 (a, b), if a, b ∈ Y,
d(a, b) =

 d1 (a, x0 ) + d2 (b, y0 ), if a ∈ X and b ∈ Y,

d1 (b, x0 ) + d2 (a, y0 ), if b ∈ X and a ∈ Y.

Prove that d is a metric on X ∪ Y which induces d1 on X and d2 on Y .

Exercise 1.1.31. Let (X, d1 ) and (Y, d2 ) be two metric spaces such that X ∩ Y = ∅. Define
d : (X ∪ Y ) × (X ∪ Y ) → [0, ∞) by


 1, if a ∈ X and b ∈ Y,

1, if a ∈ Y and b ∈ X,
d(a, b) =

 d1 (a, b), if a, b ∈ X,

d2 (a, b), if a, b ∈ Y.

Is d a metric on X ∪ Y which induces d1 on X and d2 on Y ?

Exercise 1.1.32. Let X = R and d : X × X → R be defined by

|x − y|
d(x, y) = √ p , for all x, y ∈ X.
1 + x2 1 + y 2

Prove that d is a metric on X.

Exercise 1.1.33. Let X be a nonempty set, and p : X × X → R be a function that satisfies

the following conditions. For all x, y, z ∈ X,

(i) p(x, y) ≥ 0;

(ii) p(x, y) = 0 if and only if x = y;

(iii) p(x, y) ≤ p(x, z) + p(z, y).

Show that the mapping d : X × X → R defined by

d(x, y) = max{p(x, y), p(y, x)}, for all x, y ∈ X,

is a metric on X.
Qamrul Hasan Ansari Metric Spaces Page 25

1.2 Distance Between Sets and Diameter of a Set

Definition 1.2.1

Let (X, d) be a metric space and let A and B be nonempty subsets of X. The distance
between the sets A and B, denoted by ρ (A, B), is given by

ρ (A, B) = inf {d(x, y) : x ∈ A, y ∈ B} .

Since d(x, y) = d(y, x), we have ρ (A, B) = ρ (B, A) .

If A is a singleton set {x}, then

ρ ({x} , B) = inf {d(x, y) : y ∈ B} .

It is called the distance of a point x ∈ X from the set B, and is denoted by ρ (x, B).

Remark 1.2.1

(a) The equation ρ (x, B) = 0 does not imply that x belongs to B.

(b) If ρ(A, B) = 0, then it does not imply that A and B have common points.

Example 1.2.1

Let A = {x ∈ R : x > 0} and B = {x ∈ R : x < 0} be subsets of R with the usual metric.

Then ρ (A, B) = 0, but A and B have no common point. If x = 0, then ρ (x, B) = 0;
but x ∈
/ B.

Definition 1.2.2

Let (X, d) be a metric space and A be a nonempty subset of X. The diameter of A,

denoted by δ(A), is given by

δ(A) = sup {d(x, y) : x, y ∈ A} .

The set A is called bounded if δ(A) ≤ k < ∞. In other words, A is bounded if its
diameter is finite, otherwise it is called unbounded.

In particular, the metric space (X, d) is bounded if the set X is bounded.

Qamrul Hasan Ansari Metric Spaces Page 26

Example 1.2.2

(a) The real line with the usual metric is an unbounded metric space.

(b) In R with the usual metric, the intervals [a, b], (a, b), [a, b) and (a, b] are bounded.
But [a, ∞) and (−∞, a] are not bounded.

(c) The space s of all sequences of real or complex numbers with the metric defined
k
X 1
in Example 1.1.10 is a bounded space since d(x, y) < .
n=1
2n

(d) Every set in a discrete metric space (X, d) is bounded and its diameter is 1.

Example 1.2.3

Consider the set S = {(x, y) ∈√R × R : 0 ≤ x ≤ 1, 0 ≤ y ≤ 1} in R2 . With the usual

metric d, the diameter of S is 2; with the taxicab metric, its diameter is 2; with the
max metric, its diameter is 1; and with the discrete metric its diameter is 1.

1
√
2

1 R

Remark 1.2.2

Let (X, d) be a metric space. We can define other metrics on X with the help of the
metric d in the following manner:

d(x, y)
d1 (x, y) = and d2 (x, y) = min {1, d(x, y)} .
1 + d(x, y)

Then d1 and d2 are metrics on X, and with these metrics, (X, d1 ) and (X, d2 ) are
bounded metric spaces irrespective of whether the metric space (X, d) is bounded or
not.

Exercise 1.2.1. Determine the distance from (3, 4) to the set [0, 1]×[0, 1] in R2 with respect
to the metrics (a) usual, (b) taxicab, (c) max, and (d) discrete.
Qamrul Hasan Ansari Metric Spaces Page 27

Exercise 1.2.2. Let A = {x = (x1 , x2 , x3 ) ∈ R3 : x21 + x22 + x23 ≤ 1} be a set in R3 . Compute

the diameter of A with respect to each of the following metrics: (a) usual, (b) taxicab, (c)
max, and (d) discrete.
Exercise 1.2.3. Let A = {x = (x1 , x2 ) ∈ R2 : x21 +x22 ≤ 1} and x = (1, 1). Find the distance
from x to A for the metrics (a) usual, (b) taxicab, (c) max, and (d) discrete.
Exercise 1.2.4. Let A and B be nonempty subsets of a metric space (X, d). Prove the
following statements.

(a) δ(A) = 0 if and only if A is a singleton set.

(b) For each x ∈ A, y ∈ B, ρ(A, B) ≤ d(x, y).
(c) If A ⊆ B, then δ(A) ≤ δ(B).
(d) For each x ∈ A, y ∈ B, d(x, y) ≤ δ(A ∪ B).
(e) δ(A ∪ B) ≤ δ(A) + ρ(A, B) + δ(B).

(f) If A ∩ B 6= ∅, then δ(A ∪ B) ≤ δ(A) + δ(B).

(g) d(x, A) ≤ d(x, y) + d(y, A) for all x, y ∈ X.

Proof of (e) and (f). Let a and b be arbitrary elements of A and B, respectively, and let
x, y ∈ A ∪ B. If both x and y are in A, then d(x, y) ≤ δ(A). If both x and y are in B, then
d(x, y) ≤ δ(B).

If x ∈ A and y ∈ B, then by the triangle inequality, we have

d(x, y) ≤ d(x, a) + d(a, y)

≤ d(x, a) + d(a, b) + d(b, y)
≤ δ(A) + d(a, b) + δ(B).

Similarly, if x ∈ B and y ∈ A, we have

d(x, y) ≤ δ(A) + d(a, b) + δ(B).

Thus,
d(x, y) ≤ δ(A) + d(a, b) + δ(B), for all x, y ∈ A ∪ B,
and therefore,

δ(A ∪ B) ≤ δ(A) + d(a, b) + δ(B), for all a ∈ A, b ∈ B.

Hence,
δ(A ∪ B) ≤ δ(A) + ρ(A, B) + δ(B).

Now, if A ∩ B 6= ∅, we have ρ(A, B) = 0, and hence δ(A ∪ B) ≤ δ(A) + δ(B).

Qamrul Hasan Ansari Metric Spaces Page 28

1.3 Open Sets and Interior Points

Definition 1.3.1

Let (X, d) be a metric space. Given a point x0 ∈ X and a real number r > 0, the sets

Sr (x0 ) = {y ∈ X : d(x0 , y) < r}

and
Sr [x0 ] = {y ∈ X : d(x0 , y) ≤ r}
are called open sphere (or open ball) and closed sphere (or closed ball), respectively, with
center at x0 and radius r.

Remark 1.3.1

(a) The open and closed spheres are always nonempty, since x0 ∈ Sr (x0 ) ⊆ Sr [x0 ].

(b) Every open (respectively, closed) sphere in R with the usual metric is an open
(respectively, closed) interval, but the converse is not true. For example, (−∞, ∞)
is an open interval in R, but not an open sphere.

(c) For every x, y ∈ X and every r > 0, we have

x ∈ Sr (y) ⇔ d(x, y) < r ⇔ y ∈ Sr (x),

therefore, x ∈ Sr (y) if and only if y ∈ Sr (x).

Example 1.3.1

(a) In the metric space R with the usual metric, the spheres Sr (x0 ) and Sr [x0 ] are
intervals
(x0 − r, x0 + r) and [x0 − r, x0 + r],
respectively.

(b) In the metric space C with the usual metric, the sphere Sr (z0 ) and Sr [z0 ] are
circular discs
|z − z0 | < r and |z − z0 | ≤ r,
respectively, where z0 ∈ C and r > 0.
Qamrul Hasan Ansari Metric Spaces Page 29

Example 1.3.2

Let X be a nonempty set with the discrete metric d. Then the open sphere Sr (x0 ) is

{x0 }, if 0 < r ≤ 1,
Sr (x0 ) =
X, if r > 1,

and the closed sphere Sr [x0 ] is


{x0 }, if 0 < r < 1,
Sr [x0 ] =
X, if r ≥ 1.

Example 1.3.3

Let X = [0, 1) be a metric space with the usual metric d(x, y) = |x − y| for all x, y ∈ X.
Then the open sphere Sr (0) is

[0, r), if r ≤ 1,
Sr (0) =
[0, 1), if r > 1,

and the closed sphere Sr [0] is


[0, r], if r < 1,
Sr [0] =
[0, 1), if r ≥ 1.

Example 1.3.4

In R2 , the open sphere with center at 0 = (0, 0) and radius 1 with respect to the metrics
d1 , d2 and d∞ , respectively, (in Example 1.1.8), are

S11 (0) = y = (y1 , y2 ) ∈ R2 : |y1 | + |y2 | < 1 ,
 q 
2 2 2 2
S1 (0) = y = (y1 , y2 ) ∈ R : y1 + y2 < 1 ,

and 
S1∞ (0) = y ∈ (y1 , y2 ) ∈ R2 : max (|y1 | , |y2 |) < 1 .
Similarly, we can define the closed spheres.
Qamrul Hasan Ansari Metric Spaces Page 30

S12 (0)

The sphere in R2 with respect to the metric

d1 (x, y) = d1 ((x1 , x2 ), (y1 , y2 )) = |x1 − y1 | + |x2 − y2 |

S12 (0)

The sphere in R2 with respect p to the metric

d2 (x, y) = d2 ((x1 , x2 ), (y1, y2 )) = (x1 − y1 )2 + (x2 − y2 )2
Qamrul Hasan Ansari Metric Spaces Page 31

S1∞ (0)

The sphere in R2 with respect to the metric

d∞ (x, y) = d∞ ((x1 , x2 ), (y1 , y2 )) = max{|x1 − y1 |, |x2 − y2 |}

Example 1.3.5

In the metric space C[a, b] with respect to the metric d∞ (f, g) = sup |f (t) − g(t)|, the
t∈[a,b]
open sphere Sr (f0 ) with center at f0 and radius r is the set of continuous functions g
such that
sup |f0 (t) − g(t)| < r,
t∈[a,b]

that is, the set of continuous functions g whose graph lies within the shaded band of
vertical width 2r centered on the graph of f0 .

R
g f0
f0 (
t) +
r
2r
r

r f0 (t
)−
r

a b R

The sphere in C[a, b]

Qamrul Hasan Ansari Metric Spaces Page 32

Definition 1.3.2

Let A be a nonempty subset of a metric space X.

• A point x ∈ A is said to be an interior point of A if x is the center of some open

sphere contained in A. In other words, x ∈ A is an interior point of A if there
exists r > 0 such that Sr (x) ⊆ A.

A
x
Sr (x)

• The set of all interior points of A is called interior of A and it is denoted by A◦ ,

that is,
A◦ = {x ∈ A : Sr (x) ⊆ A for some r > 0} .

• The set A is said to be open if each of its point is the center of some open sphere
contained entirely in A; that is, A is an open set if for each x ∈ A, there exists
r > 0 such that Sr (x) ⊆ A.

• Let x ∈ X. The set A is said to be a neighborhood of x if there exists an open

sphere centered at x and contained in A, that is, if Sr (x) ⊆ A for some r > 0. In
case, A is an open set, it is called an open neighborhood of x.

Remark 1.3.2

(a) In particular, an open sphere Sr (x) with center at x and radius r is a neighborhood
of x.

(b) The interior of A is the neighborhood of each of its points.

(c) Every open set is the neighborhood of each of its points.

(d) The set A is open if and only if each of its points is an interior point, that is,
A = A◦ .
Qamrul Hasan Ansari Metric Spaces Page 33

Example 1.3.6

Let R be the metric space with the usual metric and A be a subset of R.

(a) If A = (a, b), [a, b), [a, b], or (a, b], then A◦ = (a, b).

(b) If A = N, Z, Q or R \ Q, then A◦ = ∅.

(c) If A is a finite set, then A◦ = ∅.

(d) If A = C the Cantor set, then A◦ = ∅.

∞ h
\ 1
(e) If A = − 3, , then A◦ = (−3, 0).
n=1
n

(f) If A = (0, 1) ∩ Q, then A◦ = ∅.

(g) If A = ∅, then A◦ = ∅.

(h) If A = R, then A◦ = R.

Example 1.3.7

In R with the usual metric

(a) R is an open set;

(b) (a, b) is an open set;

(c) (a, b], [a, b) and [a, b] are not open sets;

(d) The set 1, 12 , 31 , · · · is not open;

(e) A set consists a singleton is not an open set;

(f) The set of all rational numbers Q is not open. But it is open with respect to the
metric d(x, y) = |x − y| defined on Q;

(g) The Cantor set C is not an open set.

Qamrul Hasan Ansari Metric Spaces Page 34

Example 1.3.8

Let X = [0, 1) with the usual metric d(x, y) = |x − y| for all x, y ∈ X. Then [0, α),
α ≤ 1, is an open set.

Example 1.3.9

Let A be a nonempty subset of a discrete metric space X. Then A◦ = A. Therefore,

every subset of a discrete metric space X is an open set.

Remark 1.3.3

(a) In a metric space X, the empty set ∅ and the whole space X are open sets.

(b) Whether a set is open or not depends upon the space in which it is considered.
For example, identify the real line R with horizontal axis {(x, 0) ∈ R2 : x ∈ R} in
R2 . R is not an open subset of R2 since R does not contain any open sphere in
R2 . However, R in the metric space R with the usual metric is an open set.

Theorem 1.3.1. Let A and B be two subsets of a metric space X. Then,

(a) A ⊆ B implies A◦ ⊆ B ◦ ;

(b) (A ∩ B)◦ = A◦ ∩ B ◦ ;

(c) (A ∪ B)◦ ⊇ A◦ ∪ B ◦ .

Proof. (a) Let x ∈ A◦ . Then there exists an open sphere Sr (x) ⊆ A for some r > 0. Since
A ⊆ B, Sr (x) ⊆ B, and hence, x ∈ B ◦ . Thus A◦ ⊆ B ◦ .

(b) Let x ∈ (A ∩ B)◦ . Then there exists an open sphere Sr (x) ⊆ A ∩ B for some r > 0.
Therefore, Sr (x) ⊆ A and Sr (x) ⊆ B, and hence, x ∈ A◦ and x ∈ B ◦ . So, x ∈ A◦ ∩ B ◦ and
thus (A ∩ B)◦ ⊆ A◦ ∩ B ◦ .

To prove the reverse inclusion, let us suppose that y ∈ A◦ ∩ B ◦ . Then y ∈ A◦ and y ∈ B ◦

and therefore, there exist open spheres Sr1 (y) ⊆ A and Sr2 (y) ⊆ B for some r1 , r2 > 0.
Set r = min {r1 , r2 }. Then Sr (y) ⊆ A ∩ B, and hence, y ∈ (A ∩ B)◦ . Consequently,
A◦ ∩ B ◦ ⊆ (A ∩ B)◦ .

(c) Let x ∈ A◦ ∪ B ◦ . Then either x ∈ A◦ or x ∈ B ◦ . This implies that there exists an open
sphere Sr (x) ⊆ A or Sr (x) ⊆ B for some r > 0. So, we have Sr (x) ⊆ A ∪ B, and therefore,
x ∈ (A ∪ B)◦ . Hence, A◦ ∪ B ◦ ⊆ (A ∪ B)◦ .
Qamrul Hasan Ansari Metric Spaces Page 35

Remark 1.3.4

Note that (A ∪ B)◦ 6⊆ A◦ ∪ B ◦ . For example, let X = R be the usual metric space and
A = [0, 1] and B = [1, 2]. Then A ∪ B = [0, 2]. Note that A◦ = (0, 1), B ◦ = (1, 2) and
(A ∪ B)◦ = (0, 2). This shows that A◦ ∪ B ◦ ⊆ (A ∪ B)◦ , but (A ∪ B)◦ 6⊆ A◦ ∪ B ◦ .

Theorem 1.3.2. Let (X, d) be a metric space. Then,

(a) each open sphere in X is an open set;

(b) a subset A of X is open if and only if it is the union of open spheres.

Sr (x0 ) Sr (x0 )

r r

x0 x0
y 0 r1
y 0 r1
S r1

Sr1 (y0 )
(y
0
)

r1 < d(x0 , y0 ) r1 > d(x0 , y0 )

Proof. (a) Let Sr (x0 ) = {x ∈ X : d(x, x0 ) < r} be an open sphere in X and let y0 ∈ Sr (x0 ).
We have to produce an open sphere centered at y0 and contained in Sr (x0 ). Since y0 ∈ Sr (x0 ),
we have d(x0 , y0 ) < r. Set
r1 = r − d(x0 , y0 ) > 0.

Consider
Sr1 (y0 ) = {y ∈ X : d(y, y0) < r1 } .

We have to show that Sr1 (y0 ) ⊆ Sr (x0 ). For this, let y ∈ Sr1 (y0 ) be arbitrary. Then
d(y, y0) < r1 , and therefore,

d(x0 , y) ≤ d(x0, y0 ) + d(y0, y) (by triangle inequality)

< d(x0 , y0 ) + r1 = r.

Thus, y ∈ Sr (x0 ), and consequently, Sr1 (y0 ) ⊆ Sr (x0 ).

Qamrul Hasan Ansari Metric Spaces Page 36

r Sr (x0 )
x0

x r1

)
(x
r1
S
(b) Suppose that A is an open set. Then each of its points is the center of some open sphere
contained in A. Hence, A is the union of all the open spheres contained in it.

To prove the converse part, let us assume that A is the union of a collection F of open
spheres. Let x ∈ A be arbitrary. Then, x belongs to some open sphere, say Sr (x0 ) ∈ F .
Since each open sphere is an open set, x is the center of an open sphere Sr1 (x) such that
Sr1 (x) ⊆ Sr (x0 ). But Sr (x0 ) ⊆ A, and hence, Sr1 (x) ⊆ A. Therefore, A is open

Theorem 1.3.3. Let (X, d) be a metric space. Then,

(a) arbitrary union of open sets in X is open;

(b) finite intersection of open sets in X is open.

S
Proof. (a) Let Λ be any index set, {Aα }α∈Λ be a family of open sets in X and A = Aα .
α∈Λ
Since each Aα is open, it is the union of open spheres for each α ∈ Λ. Then A is the union
of unions of open spheres. Hence, by Theorem 1.3.2, A is open.

n
T
(b) Let {Ai : i = 1, 2, . . . , n} be the finite family of open sets in X and let A = Ai . Let
i=1
x ∈ A. Then x is in each Ai . But each Ai is open, hence for each i, there exists ri > 0 such
that Sri (x) ⊆ Ai . Set r = min {r1 , r2 , . . . , rn }. Then,

Sr (x) ⊆ Sri (x) ⊆ Ai , for all i = 1, 2, . . . , n.

n
T
Therefore, Sr (x) ⊆ Ai = A, and hence, A is open.
i=1
Qamrul Hasan Ansari Metric Spaces Page 37

Remark 1.3.5

Arbitrary intersection of open sets need not be open.
For example, let X = R with
the usual metric. Consider the family An = − n1 , n1 : n ∈ N of open sets. Then
T∞
An = {0} is not open.
i=1

Theorem 1.3.4. Let A be a subset of a metric space X. Then A◦ is the largest open
subset of A.

Proof. First of all, we shall prove that A◦ is an open set. For that, let x ∈ A◦ be arbitrary.
Then, by the definition of an interior point, there exists an open sphere Sr (x) ⊆ A for some
r > 0. But Sr (x) is an open set, so each of its points is the center of some open sphere
contained in Sr (x) and hence contained in A. Therefore, each point of Sr (x) is the interior
point of A, that is, Sr (x) ⊆ A◦ . Thus, x is the center of an open sphere contained in A◦ .
Hence, A◦ is an open set.

Let B ⊆ A be an arbitrary open set, and let x ∈ B. Then there exists Sr (x) ⊆ B ⊆ A for
some r > 0. This implies that x ∈ A◦ , and hence, B ⊆ A◦ ⊆ A. Since A◦ is open, A◦ is the
largest open subset of A.

Remark 1.3.6

A◦ is the union of all open subsets of A.

Exercise 1.3.1. Suppose that, in a metric space X, we have Sr (x) = Ss (y) for some x, y ∈ X
and some positive real numbers r, s. Is x = y? Is r = s?

Hint: Consider the discrete metric.

Exercise 1.3.2. Find the open spheres with center 0 and radius 1 in the metric spaces with
respect to the metrics defined in Exercises 1.1.9 and 1.1.12.

Exercise 1.3.3. Let A be a subset of a metric space X. Prove that (A◦ )◦ = A◦ .

Exercise 1.3.4. In Rn , let R denote the set of points having only rational coordinates and
I its complements, that is, the set of points having at least one irrational coordinate. Then
prove that R◦ = I ◦ = ∅.

Exercise 1.3.5. Let (X, d) be a metric space, a ∈ X and 0 < r < r ′ . Prove that the set
{x ∈ X : r < d(x, a) < r ′ } is open in X.
Qamrul Hasan Ansari Metric Spaces Page 38

Exercise 1.3.6. Let (X, d) be a metric space and

d(x, y)
d∗ (x, y) = .
1 + d(x, y)

Prove that the family of open sets with respect to the metric d is same as the family of open
sets with respect to the metric d∗ .

Exercise 1.3.7. Let R be the same as in Exercise 1.3.4. Prove that

(a) every nonempty open set in Rn contains a member of R;

(b) every nonempty open set in Rn contains infinitely many members of R.

Exercise 1.3.8. Let (X, d) be a metric space and x, y be distinct points of X. Prove that
there exist disjoint open spheres centered on x and y.

Hint: Since x 6= y, d(x, y) > 0. Let d(x, y) = 3r for some r > 0, and let Sr (x) and Sr (y) be
open spheres centered on x and y, respectively. Then, clearly Sr (x) ∩ Sr (y) = ∅ because the
radius of Sr (x) and Sr (y) is r and the distance between x and y is 3r.

Exercise 1.3.9. Let X = R2 with the usual metric space. Prove that

(a) the set A = {(x, y) ∈ R2 : xy 6= 0} is open in R2 ;

(b) the set A = {(x, y) ∈ R2 : x2 + y 2 6= 1} is open in R2 .

Qamrul Hasan Ansari Metric Spaces Page 39

1.4 Closed Sets and Closure of Sets

Definition 1.4.1

Let A be a subset of a metric space X. A point x ∈ X is said to be a limit point

(accumulation point or cluster point) of A if each open sphere centered on x contains at
least one point of A different from x.

In other words, x ∈ X is a limit point of A if

(Sr (x) − {x}) ∩ A 6= ∅, for all r > 0.

The set of all limit points of A is called derived set and it is denoted by A′ .

x
(Sr (x) − {x}) ∩ A 6= ∅
Sr (x)
X

Example 1.4.1

1. In the usual metric space R,


(a) if A = 1, 21 , 13 , · · · , then A′ = {0};

(b) if A = N or Z, then A′ = ∅;

(c) if A is the set of all rational or irrational numbers, then A′ = R

(d) every point on the real line is a limit point, and therefore, R′ = R;

(e) if A is a Cantor set C, then A′ = C.

2. If A is a subset of a discrete metric space, then A has no limit point, since every open
sphere of radius 1 consists only the center. Thus A′ = ∅.
Qamrul Hasan Ansari Metric Spaces Page 40

Remark 1.4.1

By the definition of a limit point, we follow that any open sphere centered on a limit
point of A must contain infinitely many points of A, that is, a point x ∈ X is a limit
point of A if Sr (x) ∩ A is an infinite set for each r > 0.

Let Sr (x) contain a point x1 of A different from x. If d(x, x1 ) = r1 , then the open sphere
Sr1 (x) contains a point x2 of A different from x and x1 . Then by continuing in this way,
we see that Sr1 (x) contains infinitely may points of A different from x.

Note that a limit point of A is not necessarily

 a1 point of A. For example, in Example 1.4.1
1
(a), 0 is the only limit point of the set A = 1, 2 , 3 , · · · which is not in A.

In view of the above remark, we have the following definition.

Definition 1.4.2

A point x ∈ X is said to be an isolated point of A if there exists an open sphere centered

on x which contains no point of A other than x itself, that is, if Sr (x) ∩ A = {x} for
some r > 0.

Remark 1.4.2

If a point x ∈ X is not a limit point of A, then it is an isolated point. Hence every

point of a metric space X is either a limit point or an isolated point of X.

Example 1.4.2

Consider the metric space X = 0, 1, 12 , 31 , 41 , · · · with the usual metric given by the
absolute value. Then 0 is the only limit point of X while all other points are the isolated
points of X.
Qamrul Hasan Ansari Metric Spaces Page 41

Definition 1.4.3

Let A be a subset of a metric space X. The closure of A, denoted by A or clA, is the

union of A and the set of all limit points of A, that is, A = A ∪ A′ .

In other words, x ∈ A if every open sphere Sr (x) with center at x and radius r > 0
contains a point of A, that is, x ∈ A if and only if Sr (x) ∩ A 6= ∅ for every r > 0.

Remark 1.4.3

Let A and B be subsets of a metric space X. Then,

(a) ∅ = ∅;

(b) X = X;

(c) A ⊆ A;

(d) (A) = A;

(e) A ⊆ B implies A ⊆ B;

(f) A ∪ B = A ∪ B;

′
(g) A = A ;

(h) A ∩ B ⊆ A ∩ B, but A ∩ B + A ∩ B; For example, in the usual metric space R,

consider the sets A = (0, 1) and B = (1, 2). Then A ∩ B = [0, 1] ∩ [1, 2] = {1},
but A ∩ B = ∅, and hence, A ∩ B + A ∩ B.

Theorem 1.4.1. Let (X, d) be a metric space and A be a subset of X. Then x ∈ A if

and only if ρ(x, A) = 0.

Proof. Let x ∈ A. If x ∈ A, then obviously we have d(x, A) = 0. Assume x 6∈ A. Then x

is a limit point of A. Thus for any ε > 0, there exists a y ∈ Sε (x) ∩ A, that is, d(x, y) < ε.
Hence, ρ(x, A) = inf {d(x, y) : y ∈ A} = 0.

Conversely, suppose that ρ(x, A) = 0. If x ∈ A, then x ∈ A. Assume x 6∈ A. Then by

the property of the infimum, for any ε > 0, there is a y ∈ A such that d(x, y) < ε, that
is, y ∈ Sε (x) ∩ A. Since x 6∈ A, then y 6= x. Therefore, x is a limit point of A, and thus
x ∈ A.
Qamrul Hasan Ansari Metric Spaces Page 42

Definition 1.4.4

Let A be a subset of a metric space X. The set A is said to be closed if it contains all
its limit points, that is, A′ ⊆ A.

It is obvious that A is closed if and only if A = A.

Example 1.4.3

In the usual metric space R,

(a) the set of all rational numbers and the set of all irrational numbers are not closed;

(b) the set A = 1, 12 , 31 , · · · , is not closed, since A′ = {0} * A;
∞ h
\ 1
(c) if A = − 3, , then A = [−3, 0];
n=1
n

(d) if A = (0, 1) ∩ Q, then A = [0, 1];

(e) the Cantor set C is closed since A′ = A ⊆ A.

Remark 1.4.4

In a metric space X, every finite set, empty set and whole space are closed sets.

Theorem 1.4.2. A subset A of a metric space X is closed if and only if the complement
of A is an open set.

Proof. Let A be closed and x ∈ Ac be arbitrary. Then x ∈ / A and also x cannot be a

limit point of A since A is closed. Therefore, there exists an open sphere Sr (x) such that
Sr (x) ∩ A = ∅. This implies that Sr (x) ⊆ Ac for some r > 0. Since x ∈ Ac is arbitrary, each
point of Ac is the center of some open sphere which is contained in Ac . Hence Ac is open.

Conversely, assume that Ac is open. Let x ∈ X be a limit point of A. If x ∈ A, then A

contains all its limit points, and hence, A is closed. If x ∈/ A, then x ∈ Ac . Since Ac is open,
c
there exists an open sphere Sr (x) such that Sr (x) ⊆ A . Consequently, Sr (x) ∩ A = ∅ for
some r > 0. Hence, x cannot be a limit point of A, which contradicts to our assumption.
Therefore, x ∈ A. This proves that A is closed.
Qamrul Hasan Ansari Metric Spaces Page 43

Theorem 1.4.3. In a metric space (X, d), every closed sphere is a closed set.

Proof. Let Sr [x] be a closed sphere in X. Then it is sufficient to show that (Sr [x])c , the
complement of Sr [x], is an open set. Let y ∈ (Sr [x])c be arbitrary. Then y ∈/ Sr [x], and
therefore, d(x, y) > r.

X
d(x, y) > r

r
r1
x y
Sr1 (y)
z
Sr [x]

Set r1 = d(x, y) − r > 0. Let z ∈ Sr1 (y). Then d(z, y) < r1 . By the triangle inequality

d(x, y) ≤ d(x, z) + d(z, y),

we have
d(x, z) ≥ d(x, y) − d(z, y) > d(x, y) − r1 = r.
Therefore, z ∈ / Sr [x], and hence, z ∈ (Sr [x])c . Thus, Sr1 (y) ⊆ (Sr [x])c . But y ∈ (Sr [x])c
being arbitrary, each point of (Sr [x])c is the center of some open sphere contained in (Sr [x])c .
Hence, (Sr [x])c is an open set.

Remark 1.4.5

In a metric space (X, d), Sr (x) ⊆ Sr [x] for any x ∈ X and r > 0. By Remark 1.4.3 (e),
Sr (x) ⊆ Sr [x] = Sr [x] because Sr [x] is a closed set. But in general Sr (x) 6= Sr [x]. For
example, let R be a discrete metric space. Then S1 (0) = {x ∈ R : d(0, x) < 1} = {0}
which is open as well closed in X, while S1 [0] = {x ∈ R : d(0, x) ≤ 1} = R. Therefore,
S1 (x) 6= S1 [x].
Qamrul Hasan Ansari Metric Spaces Page 44

By using De Morgan’s law !c

\ [
(Acα ) = Aα
α∈Λ α∈Λ

and !c
n
[ n
\
Aci = Ai
i=1 i=1

and Theorem 1.3.3, we have the following result.

Theorem 1.4.4. In a metric space X,

(a) the arbitrary intersection of closed sets in X is closed; and

(b) the finite union of closed sets in X is closed.

Remark 1.4.6

The arbitrary union of closed sets need not be closed.

Example 1.4.4
 1 
Consider the family n
, 2 : n ∈ N of closed sets in the usual metric space R. Then,
[  1  
, 2 : n ∈ N = (0, 2],
n

which is not a closed set.

Theorem 1.4.5. Let (X, d) be a metric space and A be a subset of X. Then A is the
smallest closed subset of X containing A.

Proof. We first
prove that A is closed. Let x be a limit point of A. Then for every ǫ > 0,

Sǫ/2 (x) − {x} ∩ A 6= ∅. This implies that there exists y ∈ A such that y ∈ Sǫ/2 (x) − {x} ,
that is, d(x, y) < 2ǫ . But since y ∈ A, we have Sǫ/2 (y) ∩ A 6= ∅, that is, there exists z ∈ A
such that z ∈ Sǫ/2 (y). This implies that d(y, z) < 2ǫ . By the triangle inequality, we have

d(x, z) ≤ d(x, y) + d(y, z)

ǫ ǫ
< + = ǫ,
2 2
Qamrul Hasan Ansari Metric Spaces Page 45

that is, z ∈ Sǫ (x). This means that for every ǫ > 0, the open sphere Sǫ (x) contains a point
z of A. Hence, x is a limit point of A, and therefore, x ∈ A. This proves that A is a closed
set.

Now, we show that A is the smallest set containing A. Assume that B is any closed subset
of X such that A ⊆ B, then it is sufficient to prove that A ⊆ B. Let x ∈ A, then either
x ∈ A or x is a limit point of A. If x ∈ A, then x ∈ B and hence A ⊆ B. If x is a limit
point of A, then for every ǫ > 0, (Sǫ (x) − {x}) ∩ A 6= ∅, that is, there exists a point y ∈ A
such that y ∈ (Sǫ (x) − {x}). Then d(x, y) < ǫ. But since A ⊆ B and y ∈ A, we have y ∈ B.
Therefore, ρ(x, B) = inf d(x, y) = 0, and then by Theorem 1.4.1, x ∈ B. Since B is a closed
y∈B
set, x ∈ B and thus A ⊆ B.

Definition 1.4.5

Let A be a subset of a metric space X. A point x ∈ X is called a boundary point of A

/ (X \ A)◦ .
/ A◦ and x ∈
if it is neither an interior point of A nor of X \ A, that is, x ∈

In other words, x ∈ X is a boundary point of A if every open sphere centered on x

intersects both A and X \ A.

The set of all boundary points of A is called the boundary of A and it is denoted by
b(A).

Remark 1.4.7

It is clear that b(A) = A ∩ (X \ A) = A ∩ Ac .

Example 1.4.5

Let X = R be the usual metric space and A ⊆ X.

(a) If A = [a, b], [a, b), (a, b] or (a, b), then b(A) = {a, b}.

(b) If A = N (respectively, Z), then b (A) = N (respectively, Z).

 
(c) If A = 1, 21 , 31 , · · · , , then b(A) = 0, 1, 12 , 31 , · · · , .

(d) If A = Q, then b(A) = R.

(e) If A is a set of all irrational numbers, then b(A) = R.

Qamrul Hasan Ansari Metric Spaces Page 46

Example 1.4.6

Let X = R2 with the usual metric and A = {(x1 , x2 ) ∈ X : x21 + x22 < 1} be a subset
of X. Then b(A) = {(x1 , x2 ) ∈ X : x21 + x21 = 1} as every open sphere centered on the
point x ∈ b(A) intersects both A and Ac .

R
Ac
Sr (x)
x

The boundary of the set A

Example 1.4.7

Let (X, d) be a discrete metric space and A ⊆ X. Then b(A) = ∅.

Exercise 1.4.1. Verify that every subset of the discrete metric space is closed.

Exercise 1.4.2. Prove that every singleton set and every finite set in a metric space are
closed.

Exercise 1.4.3. Let A be a subset of a metric space X. Prove that A is the intersection of
all closed subsets of X containing A.
T
Hint: Let D = {B ⊆ X : B is closed and A ⊆ B}. Then clearly D is the smallest closed
subset of X containing A. From the proof of Theorem 1.4.5, A = D.

Exercise 1.4.4. Determine the derive set of the following sets.

(a) A finite set A = {1, 2, · · · , n}.

(b) R = {(x1 , x2 ) ∈ R2 : x1 , x2 are rational coordinates}.

Exercise 1.4.5. Let A be a subset of a metric space X. Prove that

Qamrul Hasan Ansari Metric Spaces Page 47

(a) (X \ A) = X \ A◦ , that is, (Ac ) = (A◦ )c ;

c
(b) (Ac )◦ = A .

Exercise 1.4.6. Let (X, d) be a metric space and A be a closed subset of X. Prove that
x ∈ A if and only if d(x, A) = 0, and hence,

x ∈ X \ A if and only if d(x, A) > 0.

Exercise 1.4.7. Let (X, d) be a metric space, x ∈ X and A ⊆ X be a nonempty set. Prove
that d(x, A) = 0 if and only if every neighborhood of x contains a point of A.

Exercise 1.4.8. Consider the set of rational numbers Q with usual metric d(x, y) = |x − y|.
Give an example of a nonempty closed subset A of Q and rational number x ∈ Q \ A such
that there is no y ∈ A for which d(x, y) = ρ(x, y) (Recall that ρ(x, y) = inf{d(x, y) : y ∈ A}).

Answer. Let x = 2 and A = {p ∈ Q : p2 < 2}. Then A can also be written as

2 2
{p ∈ Q : p ≤ √ 2} because there is no√p ∈ Q with p = 2. It can also be described as
{p ∈ Q : |p| < 2} or as√{p ∈ Q : |p| ≤ 2}. Then
√ A is both open as well as closed (verify!).
However, ρ(2, A) = 2 − 2, while d(2, p) > 2 − 2 for every p ∈ A.

Exercise 1.4.9. Let A be a subset of a metric space X. Prove that A = X if and only if
(X \ A)◦ = ∅, that is, (Ac )◦ = ∅.

Exercise 1.4.10. Let X be a metric space. Prove that every open sphere in X is included
in a closed sphere in X and that every closed sphere in X is included in an open sphere in
X.

Exercise 1.4.11. Prove that every open set in a metric space is the countable union of
closed sets.

Hint: Let A be an open set in a metric space X. If A is empty or if it is the whole space X,
so we are done. Now assume that A and its complement Ac are both
then A is also closed, S
nonempty. Let An = x∈Ac S1/n (x). The each An is open and its complement Acn (which is
closed) is contained in A. To finish off the problem show that A is the union of the Acn .

Exercise
1.4.12. Let A be a nonempty subset of a metric space X. Must (A◦ )c be equal
c

to cl A ?

Hint: Consider Q as a subset

c of the usual metric space R. Then Q◦ = ∅ and so (Q◦ )c = R,
whereas Q = R and Q = ∅.

Exercise 1.4.13. Let (X, d) be a metric space and A ⊆ X. Prove that

(a) b(A) = b(A);

(b) b (A) = b (X \ A) = A ∩ (X \ A);

Qamrul Hasan Ansari Metric Spaces Page 48

(c) b (A) = A \ A◦ = (X \ A) \ (X \ A)◦ ;

(d) X \ b(A) = A◦ ∪ (X \ A)◦ ;

(e) A = A◦ ∪ b(A);

(f) A = A ∪ b (A);

(g) A◦ = A \ b (A);

(h) A is closed if and only if b (A) ⊆ A;

(i) A is open if and only if A ∩ b (A) = ∅.

Exercise 1.4.14. Prove that any closed subset A of a metric space (X, d) is a countable
intersection of open sets.

[
Proof. For each n ∈ N, let On = S1/n (x). Then On is open and A ⊂ On for all n ∈ N.
\ x∈A \
Therefore, A ⊆ On . We claim that A = On . Assume contrary that there exists
\ n∈N n∈N
y∈ On \ A. Since A is closed, Ac is open and contains y. Hence there exists ε > 0 such
n∈N
that Sε (y) ⊆ Ac . Since ε > 0, there exists n ∈ N such that 1/n < ε. Since y ∈ On , there
exists x ∈ A such that d(x, y) < 1/n < ε. So Sε (y) ∩ A 6= ∅, which implies A ∩ Ac 6= ∅, a
contradiction.

Exercise 1.4.15. Let A be a nonempty subset of a metric space X. Prove or answer the
following statements.

(a) b(A) is a closed set.

(b) b(A) = b(X \ A).

(c) b(A) = A \ A◦ .

(d) If x ∈ b(A), does x have to be a limit point?

(e) x ∈ b(A) if and only if for every ε > 0. Sε (x) contains points of A and of X \ A.

Proof. (a) Since b(A) = A ∩ Ac and the intersection of two closed sets is closed, we have
b(A) is a closed set.

(b) Let x ∈ b(A) and Sr (x) be an open sphere centered on x for any r > 0. Then Sr ∩ A 6= ∅
and Sr ∩ Ac 6= ∅. But A = (Ac )c , so we also have Sr ∩ (Ac )c 6= ∅ and Sr ∩ Ac 6= ∅. This
implies that x ∈ b(Ac ). By reversing the argument, we obtain that x ∈ b(A) if x ∈ b(Ac ).
Qamrul Hasan Ansari Metric Spaces Page 49

(c) x ∈ b(A) if and only if every open sphere centered on x contains points of A, but not a
subset of A. Therefore, b(A) = A \ A◦ as x ∈ A◦ if and only if it is the center of some open
sphere contained in A, and x ∈ A if and only if every open sphere centered on x intersects
A.

(d) No. Indeed let A = {0} ⊆ R. Then A has no limit points, but b(A) = {0}.

(e) Let x ∈ b(A). Since b(A) = A ∩ (X \ A), we have x ∈ A and x ∈ (X \ A). Then x is
a limit point of A as well as of X \ A. Then for any ε > 0, we have Sε (x) ∩ A 6= ∅ and
Sε (x) ∩ (X \ A) 6= ∅.

Conversely let x ∈ X such that for any ε > 0, Sε (x) contains points of A and of X \ A. It is
enough to prove that x ∈ A. If x ∈ A, then we have nothing to prove. Assume x 6∈ A. Let
ε > 0. We know that Sε (x) ∩ A 6= ∅, since x 6∈ A, then Sε (x) ∩ A contains a point of A other
than x. Hence x is a limit point of A, that is, x ∈ A.

Exercise 1.4.16. Let A be a nonempty subset of a metric space X. Prove or answer the
following statements.

(a) A◦ ∩ b(A) = ∅ and A◦ ∪ b(A) = A.

(b) b(A) = ∅ if and only if A is both open and closed.

(c) A is open if and only if b(A) = A \ A.

◦
(d) If A is open, then is A = A ?

Proof. (a) The interior of A is the complement of (X \ A), while boundary is contained in
the latter, so the intersection is empty.

(b) If A is open, then X \ A is closed and so X \ A = X \ A, if A is closed, then A = A.

Therefore, b(A) = A ∩ (X \ A) = A ∩ (X \ A) = ∅.

The set bA is empty if and only if A and X \ A are disjoint. Since the latter contains X \ A,
it follows that A and X \ A are disjoint. Since their union is X, A must be contained in A,
which implies that A is closed. If one reverses the roles of A and X \ A in the preceding two
sentences, it follows that X \ A is also closed. Therefore, A is both open and closed in X.

(c) By Remark 1.4.7, b(A) = A ∩ (X \ A).

If A is open, then X \A is closed and so X \A = X \ A. Therefore, b(A) = A∩(X \A) = A\A.

Since b(A) = b(X \ A), b(X \ A) = A \ A ⊆ X \ A. By part (a), X \ A) ∪ b(X \ A) = (X \ A).

Since both summands of the left hand side are contained in X \ A, it follows that the right
hand side is contained in X \ A, which means that X \ A is closed in X.
Qamrul Hasan Ansari Metric Spaces Page 50

◦
(d) No. For example, consider the set A = (−1, 0) ∪ (0, 1) as a subset of R. Then A =

◦
◦
(−1, 1). However, we always have A ⊆ A because A ⊆ A implies A = A◦ A .
Exercise 1.4.17. Let X = R2 be a metric space with the usual metric and

(a) A = {(x, y) ∈ R2 : y = 0};

(b) B = {(x, y) ∈ R2 : x > 0 and y 6= 0};
(c) C = A ∪ B;
(d) D = {(x, y) ∈ R2 : x ∈
/ Q};
(e) E = {(x, y) ∈ R2 : 0 < x2 − y 2 < 1};
(f) F = {(x, y) ∈ R2 : x 6= 0 and y ≤ 1/x}.

Find the boundary of A, B, C, D, E and F .

Answer. (a) The complement of A is Ac = {(x, y) ∈ R2 : y 6= 0}. We claim that Ac is

open. But if (x, y) ∈ Ac , then y 6= 0 and the open sphere S|y| ((x, y)) with center at (x, y)
and radius |y| is contained in Ac . Therefore, A is closed. But Ac = R2 , which means that
every point of A is a limit point of Ac . Thus, b(A) = A ∩ R2 = A.

(b) Clearly, B = {(x, y) ∈ R2 : x ≥ 0} and X \ B = X \ B because the latter is open in X.

Therefore,
b(B) = B \ B = {(x, y) ∈ R2 : either x = 0 or both x > 0 and y = 0}.

(c) Clearly, C = {(x, y) ∈ R2 : x > 0 and y = 0}, C c = {(x, y) ∈ R2 : x ≤ 0 or y 6= 0},

C = {(x, y) ∈ R2 : x ≥ 0 and y = 0}, and C c = {(x, y) ∈ R2 : x ≤ 0}. Then,
b(C) = C ∩ C c = {(x, y) ∈ R2 : x = 0 or both x < 0 and y = 0}.

(d) Clearly, D = R2 , D c = {(x, y) ∈ R2 : x ∈ Q}, and D c = R2 . Therefore, b(D) = D ∩ D c =

R2 .

(e) E = {(x, y) ∈ R2 : 0 ≤ x2 − y 2 ≤ 1},

E c = {(x, y) ∈ R2 : either x2 − y 2 ≤ 0 or x2 − y 2 ≥ 1},
and
b(E) = E ∩ E c = {(x, y) ∈ R2 : x2 − y 2 = 0 or x2 − y 2 = 1}.

(f) F = {(x, y) ∈ R2 : x = 0 and y ≤ 1/x} and

F c = {(x, y) ∈ R2 : either x = 0 or both x 6= 0 and y ≥ 1/x}.
Then b(F ) = F ∩ F c = {(x, y) ∈ R2 : x = 0 or y > 1/x}.
Qamrul Hasan Ansari Metric Spaces Page 51

1.5 Subspaces

Let (X, d) be a metric space and Y be a subset of X. We may convert Y into a metric

d to Y × Y . In this manner each subset Y of X
space by restricting the distance function
can be made a metric space Y, d|Y ×Y . On the other hand, we may be given two metric
spaces (X, d) and (Y, d′ ). If Y is a subset of X, it makes sense to ask whether or not d′ is
the restriction of d.

Definition 1.5.1

Let (X, d) be a metric space and Y be a subset of X. The relative metric dY on Y is

the restriction of the metric function d on Y × Y , that is,

dY (x, y) = d(x, y), for all x, y ∈ Y.

It is easy to see that dY is a metric on Y . The space (Y, dY ) is called the metric subspace of
the metric space (X, d).

In other words, let (X, d) and (Y, d′ ) be metric spaces. We say that (Y, d′) is a subspace of
(X, d) if

(i) Y is a subset of X;

(ii) d′ = d|Y ×Y restriction of d on Y × Y .

Example 1.5.1

Let R be the usual metric space. If Y = [0, 1], (0, 1], [0, 1) or (0, 1) and

dY (x, y) = |x − y| = d(x, y), for all x, y ∈ Y.

Then (Y, dY ) is a subspace of (R, d).

Example 1.5.2

Let R be the usual metric space and Q be the set of rational numbers. Define dQ :
Q × Q → R by
dQ (x, y) = |x − y| = d(x, y), for all x, y ∈ Q.
Then (Q, dQ ) is a subspace of (R, d).
Qamrul Hasan Ansari Metric Spaces Page 52

Example 1.5.3

Let I n (the unit n cube) be the set of all n-tuples (x1 , x2 , · · · , xn ) of real numbers such
that 0 ≤ xi ≤ 1 for i = 1, 2, . . . , n. Define dc : I n × I n → R by

dc (x, y) = max |xi − yi | ,

1≤i≤n

for all x = (x1 , x2 , . . . , xn ) ∈ I n and y = (y1 , y2, · · · , yn ) ∈ I n . Then (I n , dc ) is a subspace

of (Rn , d∞ ), where d∞ is the max metric on Rn , that is, d∞ (x, y) = max |xi − yi | for all
1≤i≤n
x, y ∈ Rn .

Example 1.5.4

Let S n (the n-sphere) be the set of all n + 1-tuples (x1 , x2 , . . . , xn+1 ) of real numbers
such that x21 + x22 + · · · + x2n+1 = 1. Define dS : S n × S n → R by
v
u n+1
uX
dS (x, y) = t (xi − yi )2 = d2 (x, y),
i=1

n+1
!1/2
X
where d2 is a metric on Rn defined as d2 (x, y) = (xi − yi )2 for all x, y ∈ Rn .
i=1
Then (S n , dS ) is a subspace of (Rn+1 , d2 ) .

Example 1.5.5

Let A be the set of all (n+1)-tuples (x1 , x2 , . . . , xn+1 ) of real numbers such that xn+1 = 0.
Define dA : A × A → R by

dA (x, y) = max |xi − yi | = d∞ (x, y),

1≤i≤n

for all x = (x1 , x2 , . . . , xn , 0) ∈ A and y = (y1 , y2 , . . . , yn , 0) ∈ A, where d∞ is the max

metric on Rn+1 . Then (A, dA ) is a subspace of (Rn+1 , d∞ ).
Qamrul Hasan Ansari Metric Spaces Page 53

Example 1.5.6

Let P [a, b] be the set of all polynomials defined on [a, b]. Define dP : P [a, b]×P [a, b] → R
by
dP (f, g) = max |f (t) − g(t)| = d∞ (f, g),
t∈[a,b]

where d∞ is the max metric on C[a, b]. Then (P [a, b], dP ) is a subspace
Ra of (C[a, b], d∞ ).
But (P [a, b], dP ) is not a subspace of (C[a, b], d), where d(f, g) = b |f (t) − g(t)| dt.

Remark 1.5.1

If Y is a subspace of a metric space X, then a set which is open (respectively, closed)

in Y may
not be open (respectively, closed) in X. For example, if Y = [0, 1] then the
set 0, 21  is open in Y but not in R with the usual metric. Also, if Y = (0, 1) then the
set 0, 12 is closed in Y but not in R with the usual metric.

Lemma 1.5.1. Let (Y, dY ) be a subspace of a metric space (X, d). If a ∈ Y and r > 0,
then
Sr′ (a) = Y ∩ Sr (a),
where Sr (a) and Sr′ (a) are open spheres in (X, d) and (Y, dY ), respectively.

Proof. We have

Sr (a) ∩ Y = {x ∈ X : d(x, a) < r} ∩ Y

= {x ∈ Y : d(x, a) < r} because Y ⊆ X
= Sr′ (a).

Theorem 1.5.1. Let (Y, dY ) be a subspace of a metric space (X, d) and A be a subset
of Y . Then,

(a) A is open in Y if and only if there exists an open set G in X such that A = G ∩ Y ;

(b) A is closed in Y if and only if there exists a closed set F in X such that A = F ∩Y .

Proof. (a) Let Sr (x) and Sr′ (x) be the same as in Lemma 1.5.1. Suppose that A = G ∩ Y ,
where G is open in X. Let x ∈ A be arbitrary. Then we have to show that x is an interior
point of A, that is, x ∈ A◦ with respect to dY metric.
Qamrul Hasan Ansari Metric Spaces Page 54

Since A = G ∩ Y and x ∈ A, we have x ∈ G and x ∈ Y . Since G is open in X, there exists

r > 0 such that Sr (x) ⊆ G. Hence, by Lemma 1.5.1, we have

Sr′ (x) = Sr (x) ∩ Y ⊆ G ∩ Y = A.

It follows that x is an interior point of A as a subset of the metric space (Y, dY ). Hence
x ∈ A◦ with respect to dY metric, and hence A is open in Y .

Conversely, assume that A is an open set in Y and let x ∈ A be arbitrary. Then there exists
an open sphere Sr′ x (x) such that Sr′ x (x) ⊆ A. Now, we have
!
[ [ [
A = Sr′ x (x) = (Srx (x) ∩ Y ) = Srx (x) ∩ Y
x∈A x∈A x∈A
[
= G ∩ Y, where G = Srx (x).
x∈A

But G being an arbitrary union of open spheres in X is an open set in X. Hence A = G ∩ Y ,

where G is an open set in X.

(b) A is closed in Y ⇔ Y \ A is open in Y

⇔ Y \ A = G ∩ Y, (by part (a)) where G is open in X

⇔ A = Y \ (G ∩ Y )
⇔ A = (X ∩ Y ) \ (G ∩ Y )
⇔ A = (X \ G) ∩ Y
⇔ A = F ∩ Y, where F = X \ G is a closed set in X.

Corollary 1.5.1. Let (Y, dY ) be a subspace of a metric space (X, d) and A be a subset
of X.

(a) If A is open in Y and Y is open in X, then A is open in X.

(b) If A is closed in Y and Y is closed in X, then A is closed in X.

Qamrul Hasan Ansari Metric Spaces Page 55

Theorem 1.5.2. Let (Y, dY ) be a subspace of a metric space (X, d) and A be a subset
of Y . Then,

(a) x ∈ Y is a limit point of A in Y if and only if x is a limit point of A in X;

(b) the closure of A in Y , denoted by clA (Y), is clX (A) ∩ Y, where clX (A) is the
closure of A in X. In other words, clY (A) = clX (A) ∩ Y.

Proof. (a) Let x ∈ Y be a limit point of A in Y . Then for every open sphere Sr′ (x) in Y , we
have (Sr′ (x) − {x}) ∩ A 6= ∅. For any given r > 0, we have
(Sr (x) − {x}) ∩ A = (Sr (x) ∩ Y − {x}) ∩ A (since A ⊆ Y )
= (Sr′ (x) − {x}) ∩ A 6= ∅.
It follows that x is a limit point of A in X.

The converse can be established by retracting the above steps.

(b) Since clX (A) is closed in X, by Theorem 1.5.1, clX (A) ∩Y is closed in Y . Since clX (A) ∩Y
contains A and since clY (A) is the intersection of all closed subsets of Y containing A, we
must have
clY (A) ⊆ clX (A) ∩ Y.

On the other hand, clY (A) is closed in Y , then clY (A) = F ∩ Y, where F is a closed set in
X. Since A ⊆ clY (A), F is a closed set in X containing A. Since clX (A) is the intersection
of all closed sets containing A, we have
clX (A) ⊆ F.
Hence clX (A) ∩ Y ⊆ F ∩ Y = clY (A).
Exercise 1.5.1. Let (Y, dY ) be a subspace of a metric space (X, d). Prove that a subset M
of Y is a neighborhood of a point y ∈ Y if and only if there is a neighborhood N of y in
(X, d) such that M = Y ∩ N.

Proof. Let N be a neighborhood of a point y ∈ Y in (X, d) such that M = Y ∩ N. Then

there exists an open sphere Sr (y) such that Sr (y) ⊆ N. Since Sr′ (y) = Y ∩ Sr (y), we have
Sr′ (y) ⊆ Y ∩ N = M. Hence M is a neighborhood of y ∈ Y in (Y, dY ).

Conversely, suppose that M is a neighborhood of y in (Y, dY ). Then there exists an open

sphere Sr′ (y) ⊆ M. Let N = M ∪ Sr (y). Then,
Y ∩ N = Y ∩ (M ∪ Sr (y)) = (Y ∩ M) ∪ (Y ∩ Sr (y))
= M ∪ Sr′ (y) (since M ⊆ Y )
= M (because Sr′ (y) ⊆ M).
Since Sr (y) ⊆ N, N is a neighborhood of y in (X, d).
 Chapter 2

Complete Metric Spaces

Augustin-Louis Cauchy was born on August 21,

1789, Paris, France, and died on May 23, 1857,
Sceaux, France. He was a French mathemati-
cian, engineer, and physicist who made pioneer-
ing contributions to several branches of mathe-
matics, including mathematical analysis and con-
tinuum mechanics. He also researched in conver-
gence and divergence of infinite series, differential
equations, determinants, probability and mathe-
matical physics.
Augustin-Louis Cauchy
The concept of a convergent sequence plays an important role to investigate the closedness
of a set, the continuity of a function and several other properties. In this chapter, we give an
introduction to the convergence of sequences in arbitrary metric spaces. We investigate the
properties under which a sequence is convergent. In real analysis, Cauchy criteria provides
the necessary and sufficient conditions for a sequence to be convergent, that is, a sequence
is convergent if and only if it is Cauchy. This Cauchy criteria does not hold for an arbitrary
metric space. The metric spaces in which Cauchy criteria holds are called complete. Such
metric spaces are also considered and studied. A very important property of a complete
metric space is known as Cantor’s intersection theorem. Such theorem and its converse
version are presented. We also consider a metric space which contains an incomplete metric
space and some other points so that every Cauchy sequence is convergent in this larger space.
Such spaces are called completion of an incomplete metric space.

56
Qamrul Hasan Ansari Metric Spaces Page 57

2.1 Convergent Sequences

Definition 2.1.1

A sequence s in a set X is a mapping from the set of all natural numbers N into X.
The image under a sequence s of a natural number n is denoted by xn and is referred
as nth term of the sequence s.

Let {xn }n∈N be a given sequence and {nk }k∈N be a sequence of positive integers such
that n1 < n2 < · · · . Then the sequence {xnk }k∈N is called a subsequence of the sequence
{xn }.

Definition 2.1.2

Let {xn } be a sequence in a set X. If there exists a positive integer N such that xn = x
for all n > N, then the sequence {xn } is called eventually constant .

If xn = c for all n ∈ N, then the sequence {xn } is called a constant sequence .

Obviously, a constant sequence is a special case of an eventually constant sequence.

Definition 2.1.3

Let (X, d) be a metric space. A sequence {xn } of points of X is said to be convergent if

there is a point x ∈ X such that for each ε > 0, there exists a positive integer N such
that
d(xn , x) < ε, for all n > N.
The point x ∈ X is called a limit point of the sequence {xn }.

A sequence which is not convergent is said to be divergent.

More preciously, a sequence {xn } in a metric space X converges to a point x ∈ X if the

sequence {d(xn , x)} of real numbers converges to 0 as n → ∞.

Since d(xn , x) < ε is equivalent to xn ∈ Sε (x), the definition of convergent sequence can be
restated as follows:

A sequence {xn } in a metric space X converges to a point x ∈ X if and only if for each
ε > 0, there exists a positive integer N such that

xn ∈ Sε (x), for all n > N.

Qamrul Hasan Ansari Metric Spaces Page 58

x1
x2
x3
x4 x
5 x6 Sε (x)
ε
x
xN xN +1 x
N +2

A sequence {xn } is converging to a point x

We use the following symbols to write a convergent sequence

xn → x or lim xn = x,
n→∞

and we express it by saying that xn approaches x or that xn converges to x.

Remark 2.1.1

(a) An eventually constant sequence, and hence a constant sequence, is convergent.

(b) In a discrete metric space, a sequence can converge to a point only if it is an

eventually constant sequence, or it has an eventually constant subsequence.

Remark 2.1.2

The convergence of a sequence in a metric space (X, d) depends on the space X as well
as on the metric d.

Example 2.1.1

Let xn = n1 : n ∈ N be a sequence in the usual metric space R. Then xn → 0 as
n → ∞. However,if we consider the metric space X = (0, 1) with the usual metric,
then the sequence xn = n1 : n ∈ N converges to 0 ∈
/ X. Thus, the sequence {xn } is
not convergent in this case.
Qamrul Hasan Ansari Metric Spaces Page 59

Example 2.1.2

Let {fn }n∈N be a sequence in the space C[0, 1], where

fn (t) = e−nt , for all n ∈ N.

Then, fn → 0 with respect to the metric d1 on C[0, 1] as

Z 1
1

d1 (fn , 0) = e−nt dt = 1 − e−n → 0 as n → ∞.
0 n

On the other hand, the same sequence {fn }n∈N is not convergent with respect to the
metric d∞ on C[0, 1] as

d∞ (fn , 0) = sup e−nt = 1, for all n ∈ N.

t∈[0,1]

Hence, d∞ (fn , 0) 6→ 0 as n → ∞.

Theorem 2.1.1. A sequence in a metric space cannot converge to more than one limit
point. In other words, in a metric space, every convergent sequence has a unique limit.

Proof. Let (X, d) be a metric space and {xn } be a convergent sequence in X. Suppose to
the contrary that {xn } converges to two distinct points x and y. Then, for each ε > 0, there
exist positive integers N1 and N2 such that
ε
d(xn , x) < , for all n > N1 ,
2
and
ε
d(xn , y) < , for all n > N2 .
2
By setting N = max {N1 , N2 } and using the triangle inequality, we get

d(x, y) ≤ d(xn , x) + d(xn , y)

ε ε
< + = ε, for all n > N.
2 2
It follows that x = y, and hence the limit is unique.

Definition 2.1.4

A sequence in a metric space is said to be bounded if the range set of the sequence is
bounded.
Qamrul Hasan Ansari Metric Spaces Page 60

Theorem 2.1.2. In a metric space, every convergent sequence is bounded.

Proof. Let (X, d) be a metric space and {xn } be a sequence in X such that xn → x ∈ X as
n → ∞. Then there exists a positive integer N such that
1
d(xn , x) < , for all n > N.
2
1
Set r = max 2
, d(xn , x) for all 1 ≤ n ≤ N . Then,

d(xn , x) ≤ r, for all n ∈ N.

By the triangle inequality, we have

d(xn , xm ) ≤ d(xn , x) + d(x, xm ) ≤ 2r, for all n, m ∈ N.

Hence the diameter of the range set of the sequence {xn } is bounded by 2r. Therefore, the
sequence {xn } is bounded.

Remark 2.1.3

In a metric space, a bounded sequence need not be convergent. For example, consider
the usual metric space X = R and {xn } = {(−1)n }. Then {xn } is bounded but not
convergent.

Theorem 2.1.3. Let (X, d) be a metric space and A be a subset of X.

(a) A point x ∈ X is a limit point of A if and only if there exists a sequence {xn } of
points of A, none of which equals x, such that {xn } converges to x.

(b) The set A is closed if and only if every convergent sequence of points of A has its
limit in A.

Proof. (a) Let x ∈ X be a limit point of A. Construct a sequence {xn } by recursion as

follows:

Since x ∈ X is a limit point of A, we have (S1 (x) − {x}) ∩ A 6= ∅. So, we can choose
x1 ∈ (S1 (x) − {x}) ∩ A. Likewise the points x1 , x2 , . . . , xn can be chosen such that
 
xi ∈ S 1 (x) − {x} ∩ A, for i = 1, 2, . . . , n.
i

   
Still S 1 (x) − {x} ∩ A 6= ∅, we can always choose xn+1 ∈ S 1 (x) − {x} ∩ A. By
n+1 n+1
repeating this process infinitely many times, we construct the sequence {xn } by recursion
such that all the points of {xn } are in A and xn 6= x for all n.
Qamrul Hasan Ansari Metric Spaces Page 61

Let ε > 0 be given and N be a positive integer such that N > 1ε . Then,

xn ∈ S 1 (x) ⊂ Sε (x), for all n > N,

n

and hence, {xn } converges to x.

Conversely, assume that there is a sequence {xn } of points of A, none of which equals x,
such that {xn } converges to x. Then for every ε > 0, there exists a positive integer N such
that
xn ∈ Sε (x), for all n > N.
Therefore, (Sε (x) − {x}) ∩ A 6= ∅ which implies that x is a limit point of A.

(b) Suppose that A is closed and {xn } is a sequence of points of A which converges to a
point x in X. Then we have to show that x ∈ A.

If the range set of the sequence {xn } is infinite, then it follows from part (a) that x is a limit
point of this set. Since A is closed, we have x ∈ A.

On the other hand, if the range set of the sequence {xn } is finite, then xn = x for all n ≥ N,
since {xn } is a convergent sequence. Since each term of the sequence belongs to A, so is x.

Conversely, assume that every convergent sequence of points of A converges to a point of A.

We show that A is closed by showing that it contains all its limit points.

Let x be a limit point of A. Then by part (a), there is a sequence {xn } of points of A, none
of which equals x, such that xn → x. By hypothesis x ∈ A, and hence, A is closed.

Exercise 2.1.1. Show that the limit of a convergent sequence of distinct points in a metric
space is a limit of the range set of the sequence.

Proof. Let {xn } be a sequence in a metric space such that xn → x and let A be the range
set of the sequence {xn }. Then we have to show that x is a limit point of A.

Suppose that x is not a limit point of A. Then there exists an open sphere Sε (x) such that

(Sε (x) − {x}) ∩ A = ∅,

that is, Sε (x) contains no point of A other than x. Since x is a limit point of the sequence,
for each ε > 0, there exists a positive integer N such that

d(xn , x) < ε or xn ∈ Sε (x), for all n > N

which is a contradiction.

Exercise 2.1.2. Let (X, d) be a metric space. If {xn } and {yn } be sequences in X such
that xn → x and yn → y, then prove that d (xn , yn ) → d(x, y).
Qamrul Hasan Ansari Metric Spaces Page 62

Proof. Since xn → x and yn → y, for each ε > 0, there exist positive integers N1 and N2
such that
ε
d (xn , x) < , for all n > N1 ,
2

and
ε
d (xm , x) < , for all m > N2 .
2

Let N = max{N1 , N2 }. Then, for all n, m > N, we have

|d (xn , yn ) − d(x, y)| ≤ |d (xn , yn ) − d (xn , y)| + |d (xn , y) − d(x, y)|

≤ d (yn , y) + d (xn , x)
ε ε
< + = ε,
2 2

and hence, d (xn , yn ) → d(x, y).

Exercise 2.1.3. Let X = C[0, 1] be a metric space and {fn (x)}, where fn (x) = x1/n , be a
sequence in X. Show that {fn } converges to zero with respect to the metric d1 on X, while
it converges to 1 with respect to the metric d∞ on X.

Hint: Since x1/n ≤ 1 for each 0 ≤ x ≤ 1, we have

Z 1
1
d1 (fn , 0) = x1/n dx = 1 − → 0, as n → ∞,
0 1/n + 1

and
d∞ (fn , 1) = sup 1 − x1/n = 1.
x∈[0,1]

2.2 Cauchy Sequences

Definition 2.2.1

Let (X, d) be a metric space. A sequence {xn } in X is said to be a Cauchy sequence if

for each ε > 0, there exists a positive integer N such that

d (xm , xn ) < ε, for all m, n > N.

Qamrul Hasan Ansari Metric Spaces Page 63

f (n) = xn

| | | | | | | | | | | n
N
| | | | | |
N < m, n

Example 2.2.1

Let X = C[0, 1] be a metric space with the metric d∞ defined by

d∞ (f, g) = sup |f (t) − g(t)| , for all f, g ∈ X.

t∈[0,1]

The sequence {fn } in X given by

nt
fn (t) = , for all t ∈ [0, 1],
n+t
is a Cauchy sequence in X. Indeed, for m ≥ n, the function

mt nt (m − n)t2
t 7→ − =
m+t n+t (m + t)(n + t)

being continuous on [0, 1], assumes its supremum at some point t0 ∈ [0, 1]. So,

d∞ (fm , fn ) = sup |fm (t) − fn (t)|

t∈[0,1]

(m − n)t20
=
(m + t0 )(n + t0 )
t20 1
≤ ≤ → 0 for large n and m.
(n + t0 ) n

Moreover, the sequence {fn } converges to some limit. Indeed, let f (t) = t. Then,

nt t2 1
|fn (t) − f (t)| = −t = ≤ → 0 as n → ∞.
n+t n+t n

Therefore, {fn } converges to the limit f , where f (t) = t for all t ∈ [0, 1].
Qamrul Hasan Ansari Metric Spaces Page 64

Theorem 2.2.1. Every convergent sequence in a metric space is a Cauchy sequence.

Proof. Let (X, d) be a metric space and {xn } be a sequence in X such that xn → x as
n → ∞. Then for each ε > 0, there exists a positive integer N such that
ε
d (xn , x) < , for all n > N.
2
By the triangle inequality, we have

d (xm , xn ) ≤ d (xm , x) + d (xn , x)

ε ε
< + = ε, for all n, m > N.
2 2
Hence, {xn } is a Cauchy Sequence.

Remark 2.2.1

Every Cauchy sequence in a metric space need not be convergent.

Example 2.2.2

(a) Consider the sequence {xn } in the usual metric space Q, where

x1 =
1.4
x2 =
1.41
x3 =
1.414
x4 =
1.4142
x5 =
1.41421
... ..................
√
Then the sequence {xn } converges to 2. Also, {xn } is a Cauchy sequence.
However, it does not converge to a point of Q.
 
1
(b) Let X = (0, 1] be a metric space with the usual metric and xn = : n ∈ N be
n
a sequence in X. Then {xn } is a Cauchy sequence since for each ε > 0, we have

1 1 1
d (xm , xn ) = − < ε, for all m, n > .
m n ε

On other hand, xn → 0 ∈
/ X.
Qamrul Hasan Ansari Metric Spaces Page 65

Remark 2.2.2
1
In Example 2.2.2 (b), if we consider X = [0, 1], then the sequence , 1, 1, · · ·
2 3 4
is
Cauchy as well as convergent.

Theorem 2.2.2. Let (X, d) be a metric space and {xn } be a convergent sequence in X
such that xn → x as n → ∞. If {xnk } is any subsequence of {xn }, then xnk → x as
k → ∞.

Proof. Since xn → x as n → ∞, for each ε > 0, there exists a positive integer N such that
d (xn , x) < 2ε for all n > N. Also, since every convergent sequence is Cauchy, we have

d (xnk , x) ≤ d (xnk , xn ) + d (xn , x)

ε ε
< + , for all n, nk > N,
2 2
and hence xnk → x as k → ∞.

Remark 2.2.3

Theorem 2.2.2 says that every subsequence of a convergent sequence is convergent.

However, if a subsequence of a sequence in a metric space (X, d) is convergent, then the
sequence itself need not be convergent.

Example 2.2.3

Consider the sequence {xn = (−1)n : n ∈ N} in R with the usual metric. Let {x2n } be
a subsequence of the sequence {xn } given by

x2n = 1, for all n.

Then x2n → 1 as n → ∞. However, {xn } is not a convergent sequence.

Theorem 2.2.3. Let {xn } be a Cauchy sequence in a metric space (X, d). Then the
sequence {xn } is convergent if and only if it has a convergent subsequence.

Proof. Let {xnk } be a convergent subsequence of the sequence {xn }. Suppose that xnk → x
as k → ∞. Then for each ε > 0, there exists a positive integer N such that
ε
d (xnk , x) < , for all nk > N.
2
Qamrul Hasan Ansari Metric Spaces Page 66

Since {xn } is a Cauchy sequence, we have

ε
d (xnk , xn ) < , for all n, nk > N.
2
By the triangle inequality, we have

d (xn , x) ≤ d (xn , xnk ) + d (xnk , x)

ε ε
< + = ε, for all n > N.
2 2
Hence, the sequence {xn } is convergent.

The converse part follows from Theorem 2.2.2.

Exercise 2.2.1. Consider the sequence {xn } in the usual metric space Q, where

x1 = 0.1, x2 = 0.101, x3 = 0.101001, x4 = 0.1010010001, . . . .

Prove that the sequence {xn } is Cauchy which does not converge to a point of Q.

Exercise 2.2.2. Prove that every Cauchy sequence in a discrete metric space is convergent.

Proof. Let (X, d) be a discrete metric space and {xn } be a Cauchy sequence in X. Recall
that the discrete metric d is defined by

0, if x = y,
d(x, y) =
1, if x 6= y.

1
Let ε = . Since {xn } is a Cauchy sequence, there exists a positive integer N such that
2
1
d (xm , xn ) < , for all m, n > N.
2
From the definition of d, we have xm = xn for all m, n > N. In other words, {xn } is of the
form {x1 , x2 , . . . , xN , x, x, . . .}. Hence xn → x as n → ∞.

Exercise 2.2.3. Prove that every Cauchy sequence in a metric space is bounded.

Exercise 2.2.4. Let d and d∗ be two metrics on the same underlying set X and there exist
two real numbers α1 , α2 > 0 such that

α1 d(x, y) ≤ d∗ (x, y) ≤ α2 d(x, y), for all x, y ∈ X.

Prove that the sequence {xn } is Cauchy (respectively, convergent) in (X, d) if and only if it
is Cauchy (respectively, convergent) in (X, d∗ ).

Exercise 2.2.5. Let {xn } be a Cauchy sequence in a metric space (X, d) and {xnk } be a
subsequence of {xn }. Show that lim d (xn , xnk ) = 0.
n→∞
Qamrul Hasan Ansari Metric Spaces Page 67

Proof. Let ε > 0. Since {xn } is a Cauchy sequence in X, there exists a positive integer N
such that
d (xn , xm ) < ε, for all n, m > N − 1.
Now nN ≥ N > N − 1, and therefore,
d (xN , xnN ) < ε.
In other words, lim d (xn , xnk ) = 0.
n→∞

Exercise 2.2.6. Let {xn } and {yn } be sequences in a metric space (X, d) such that {yn } is
Cauchy and d(xn , yn ) → 0 as n → ∞. Prove that

(a) {xn } is a Cauchy sequence in X;

(b) {xn } converges to x ∈ X if and only if {yn } converges to x.

Proof. (a) Let ε > 0. Since {yn } is a Cauchy sequence, there exists a positive integer N1
such that
ε
d(ym , yn ) < , for all m, n > N1 .
3
1 ε
Since d(xn , yn ) → 0 as n → ∞, there exists a positive integer N2 such that < and
N2 3
ε
d(xn , yn ) < , for all n > N2 .
3
By the triangle inequality, we have
d(xm , xn ) ≤ d(xm , ym ) + d(ym , yn ) + d(yn , xn ).
Hence for all n, m > N2 , we have
ε ε
d (xm , xn ) < + d (ym , yn ) + .
3 3
Let N0 = max{N1 , N2 }. Then for all n, m > N0 , we have
ε ε ε
d (xm , xn ) < + + = ε.
3 3 3
Thus, {xn } is a Cauchy sequence.

(b) By the triangle inequality, we have

d (yn , x) ≤ d (yn , xn ) + d (xn , x) ,
and hence,
lim d (yn , x) ≤ lim d (yn , xn ) + lim d (xn , x) .
n→∞ n→∞ n→∞

But lim d (yn , xn ) = 0 and if lim d (xn , x) = 0, we have lim d (yn , x) = 0. Thus, yn → x as
n→∞ n→∞ n→∞
n → ∞.
Qamrul Hasan Ansari Metric Spaces Page 68

Exercise 2.2.7. Let (X, d) be a metric space and {xn } be a sequence in X such that
d(xn , xn+1 ) < 21n for all n. Prove that {xn } is a Cauchy sequence.

1
Proof. Let ε > 0 and choose a positive integer N such that 2N−1
< ε. Then for all n > m ≥
N, we have

d(xm , xn ) ≤ d(xm , xm+1 ) + d(xm+1 , xm+2 ) + · · · + d(xn−1 , xn )

1 1 1
< m + m+1 + · · · + n−1
2 2 2
∞
X 1 1 1
< k
= m−1 < N −1 < ε.
k=m
2 2 2

Exercise 2.2.8. Let {xn } and {yn } be Cauchy sequences in a metric space (X, d). Prove
that {d (xn , yn )} is a Cauchy sequence.

Proof. From Exercise 1.1.2, for any m, n ∈ N, we have

|d (xn , yn ) − d (xm , ym )| ≤ d (xn , xm ) + d (yn , ym ) .

Since {xn } and {yn } are Cauchy sequences, for any given ε > 0, there exist positive integers
N1 and N2 such that for any m, n ≥ N1 , we have
ε
d (xm , xn ) < ,
2
and for any m, n ≥ N2 , we have
ε
d (ym , yn ) < .
2
Set N = max {N1 , N2 }. Then for any m, n ≥ N, we have
ε ε
|d (xm , ym ) − d (xn , yn )| ≤ d (xm , xn ) + d (ym , yn ) < + = ε.
2 2
This implies that {d(xn , yn )} is a Cauchy sequence in R.

Exercise 2.2.9. Let (X, d) be a metric space and d∗ be another metric on X defined by

d∗ (x, y) = min{1, d(x, y)}.

Show that {xn } is a Cauchy sequence in (X, d) if and only if it is a Cauchy sequence in
(X, d∗ ).

Exercise 2.2.10. Let X = C[0, 1] be a metric space with the following metric
Z 1 1/2
2
d2 (f, g) = |f (t) − g(t)| , for all f, g ∈ X and for all t ∈ [0, 1].
0
Qamrul Hasan Ansari Metric Spaces Page 69

Show that the sequence {fn } defined by


1 − nt, if 0 ≤ t ≤ n1 ,
fn (t) =
0, if n1 ≤ t ≤ 1,

is Cauchy with respect to the metric d2 , but not Cauchy with respect to the metric d∞ .

Hint:
"Z 1 Z 1
#2
m n
d2 (fm , fn ) = (m − n)2 t2 dt + (1 − nt)2 dt
1
0 m
   12
1 1 2 n
= − +
3 n m m2

and
n
d∞ (fm , fn ) = sup |fm (t) − fn (t)| = 1 − .
0≤t≤1 m
Qamrul Hasan Ansari Metric Spaces Page 70

2.3 Complete Metric Spaces

Definition 2.3.1

A metric space (X, d) is said to be complete if every Cauchy sequence in X converges

to a point in X.

Remark 2.3.1

In view of Theorem 2.2.3, a metric space (X, d) is complete if and only if every Cauchy
sequence in X has a convergent subsequence.

Example 2.3.1

(a) The usual metric spaces R and C are complete.

(b) The set of integer Z with the usual metric is a complete metric space.

Verification. Let {xn } be a Cauchy sequence of integers, that is, each term of the
sequence belongs to Z = {. . . , −2, −1, 0, 1, 2, . . .}. Then the sequence must be of the form
{x1 , x2 , x3 , . . ., xn , x, x, x, . . .}. Indeed, if we choose ε = 12 , then
1
xn , xm ∈ Z and |xn − xm | < implies xn = xm .
2
Hence the sequence {x1 , x2 , . . . , xn , x, x, x, . . .} converges to x.

Example 2.3.2

The Euclidean space Rn with the usual metric

n
! 21
X
d(x, y) = (xk − yk )2 ,
k=1

for all x = (x1 , x2 , . . . , xn ) and y = (y1 , y2, . . . , yn ) in Rn , is a complete metric space.

 (m) 
n (m) (m) (m)
Verification. Let x be a Cauchy sequence in R , where x = x1 , x2 ,
    
(m) (1) (1) (1) (2) (2) (2)
. . . , xn , that is, x(1) = x1 , x2 , . . . , xn , x(2) = x1 , x2 , . . . , xn and so on. Then
for each ε > 0, there exists a positive integer N such that
" n #1

X  (m) 2 2
(p)
d x(m) , x(p) = xk − xk < ε, for all m, p > N.
k=1
Qamrul Hasan Ansari Metric Spaces Page 71

On squaring both the sides, we get

Xn  2  2
(m) (p) (m) (p) (m) (p)
xk − xk < ε2 ⇒ xk − xk < ε2 ⇒ xk − xk < ε,
k=1

for all m, p > N and all k = 1, 2, . . . , n.

n o
(m)
It follows that for each fixed k (1 ≤ k ≤ n), the sequence xk is a Cauchy sequence
m∈N
in the usual metric space R. Since R is complete, it converges to some point in R. Let
(m) (m) (m) (m)
xk → xk as m → ∞ for each k = 1, 2, . . . , n, that is, x1 → x1 , x2 → x2 , . . . , xn → xn .
(m)
Then for each k = 1, 2, . . . , n, xk − xk → 0 as m → ∞. Let x = (x1 , x2 , . . . , xn ). Then
clearly x ∈ Rn , and hence,
" n #1

X  (m) 2 2
d x(m) , x = xk − xk → 0 as m → ∞.
k=1

Therefore, x(m) → x as m → ∞, and thus the sequence x(m) is convergent. Hence, Rn is
complete.

Example 2.3.3

The unitary space Cn is a complete metric space.

Example 2.3.4

By Exercise 2.2.2, every Cauchy sequence is convergent in a discrete metric space and
hence every discrete metric space is complete.

Example 2.3.5

The space ℓp (1 ≤ p < ∞) of all sequences {xn } of real or complex numbers such that
∞
P
|xk |p < ∞ with the metric
k=1

∞
! p1
X p
dp (x, y) = |xk − yk | , for all x = {xn } , y = {yn } ∈ ℓp
k=1

is a complete metric space.

 n o
(m) (m)
Verification. Let x(m) m∈N be a Cauchy sequence in ℓp , where x(m) = x1 , x2 , . . . =
n o ∞
P p
(m) (m)
xk , such that xk < ∞ for all m = 1, 2, . . .. Then for each ε > 0, there exists
k∈N k=1
Qamrul Hasan Ansari Metric Spaces Page 72

a positive integer N such that

∞
! p1

X p
(m) (n)
dp x(m) , x(n) = xk − xk < ε, for all m, n > N, (2.1)
k=1

and thus,
(m) (n)
xk − xk < ε,
for all m, n > N and all k = 1, 2, . . . .
n o
(m)
This implies that for each fixed k (1 ≤ k < ∞), xk is a Cauchy sequence in K (=
m∈N
(m)
R or C). Since K is complete, it converges in K. Let xk as m → ∞ for all k. Using xk →
these limits, we define x = (x1 , x2 , . . .) and show that x ∈ ℓ and x(m) → x. p

From (2.1), we have

l
X p
(m) (n)
xk − xk < εp , for all m, n > N and all l = 1, 2, . . . .
k=1

Letting n → ∞, we obtain
l
X p
(m)
xk − xk ≤ εp , for all m > N and for any l = 1, 2, . . . .
k=1
 l

P p
(m)
The sequence xk − xk is a monotonically increasing sequence which is bounded
k=1 l≥1
∞
P p
(m)
above and therefore has a finite limit xk − xk , which is less than or equal to εp .
k=1
Therefore,
∞
X p
(m)
xk − xk ≤ εp , for all m > N. (2.2)
k=1

By Minkowski inequality (Theorem A.0.4), we have

∞
!1/p ∞
!1/p ∞
!1/p
X X p X p
p (m) (m)
|xk | ≤ xk − xk + xk .
k=1 k=1 k=1

Thus x ∈ ℓp . Moreover, from inequality (2.2), we obtain

dp x(m) , x < ε, for all m > N

which implies that x(m) → x in ℓp . Hence ℓp (1 ≤ p < ∞) is complete.

Qamrul Hasan Ansari Metric Spaces Page 73

Example 2.3.6
n o
∞
The space of all bounded sequences ℓ = {xk } ⊆ K = R or C : sup |xk | < ∞ with
1≤k<∞
the metric d(x, y) = sup |xk − yk |, where x = {xk }, y = {yk } ∈ ℓ∞ , is a complete
1≤k<∞
metric space.

 n
(m)
Verification. Let x(m) m∈N be a Cauchy sequence in ℓ∞ , where x(m) = x1 ,
o n o
(m) (m) (m)
x2 , . . . = xk , such that sup xk < ∞ for all m = 1, 2, . . .. Then for each
k∈N 1≤k<∞
ε > 0, there exists a positive integer N such that

(m) (n)
d x(m) , x(n) = sup xk − xk < ε, for all m, n > N,
1≤k<∞

and thus,
(m) (n)
xk − xk < ε,
for all m, n > N and all k = 1, 2, . . . . (2.3)
n o
(m)
This implies that for each fixed k (1 ≤ k < ∞), xk is a Cauchy sequence in K. Since
m∈N
(m)
K is complete, it converges in K. Let → xk as m → ∞ for all k. Using these limits, we
xk
define x = (x1 , x2 , . . .) and show that x ∈ ℓ∞ and x(m) → x.

Letting n → ∞ in (2.3), we obtain

(m)
xk − xk ≤ ε, for all m > N and all k = 1, 2, . . . . (2.4)
n o
(m) (m)
Since x(m) = xk ∈ ℓ∞ , there is a real number Nm such that xk ≤ Nm for all k.
k∈N
Therefore,
(m) (m) (m)
|xk | = xk − xk + xk ≤ xk − xk + |xm
k |

≤ ε + Nm , for all m > N and all k = 1, 2, . . . .

This inequality is true for each k and the right hand side is independent of k, it follows that
{xk } is a bounded sequence of numbers and hence x = {xk } ∈ ℓ∞ .

From (2.4), we obtain

(m)
d x(m) , x = sup xk − xk ≤ ε, for all m > N,
1≤k<∞

and therefore, x(m) → x in ℓ∞ . Hence ℓ∞ is complete.

Qamrul Hasan Ansari Metric Spaces Page 74

Example 2.3.7

The space C[a, b] of all continuous real-valued functions defined on [a, b] with the metric

d∞ (f, g) = sup |f (t) − g(t)|

t∈[a,b]

is a complete metric space.

Verification. Let {fm }m∈N be a Cauchy sequence in C[a, b]. Then for each ε > 0, there
exists a positive integer N such that

d∞ (fm , fn ) = sup |fm (t) − fn (t)| < ε, for all m, n > N. (2.5)
t∈[a,b]

Then for any fixed t0 ∈ [a, b], we have

|fm (t0 ) − fn (t0 )| < ε, for all m, n > N.

Therefore, {fm (t0 )} is a Cauchy sequence in R. Since R is complete, this sequence converges.
Let fm (t0 ) → f (t0 ) as m → ∞. In this way, we can associate to each t ∈ [a, b] a unique
real number f (t). Thus we have defined a function f with domain [a, b]. Now, we show that
f ∈ C[a, b] and fm → f .

From (2.5), we obtain

|fm (t) − fn (t)| < ε, for all m, n > N and all t ∈ [a, b].

Letting n → ∞, we obtain

|fm (t) − f (t)| ≤ ε, for all m > N and all t ∈ [a, b]. (2.6)

To show that f is continuous, we consider any t0 ∈ [a, b] and any η > 0. According to the
preceding paragraph, there exists a positive integer N1 (η) such that
η
|fm (t) − f (t)| < , for all m > N1 (η) and all t ∈ [a, b].
3
Let n ≥ N1 (η). Then
η
|fn (t) − f (t)| < , for all t ∈ [a, b]. (2.7)
3
By continuity of fn , we obtain δ > 0 such that
η
|fn (t) − fn (t0 )| < , whenever |t − t0 | < δ. (2.8)
3
Then from (2.7) and (2.8), we have

|f (t) − f (t0 )| ≤ |f (t) − fn (t)| + |fn (t) − fn (t0 )| + |fn (t0 ) − f (t0 )|
η η η
< + + = η,
3 3 3
Qamrul Hasan Ansari Metric Spaces Page 75

whenever |t − t0 | < δ. Therefore, f ∈ C[a, b].

Moreover, from (2.6), we have

sup |fm (t) − f (t)| < ε, for all m > N.

t∈[a,b]

Thus, d∞ (fm , f ) < ε for all m > N, and therefore, fm → f as m → ∞. Hence C[a, b] is
complete.

Example 2.3.8

The space C[0, 1] is not complete with respect to the metric

Z 1
d1 (f, g) = |f (t) − g(t)|dt, for all f, g ∈ C[0, 1].
0

Verification. Let {fn }n≥2 be a sequence in C[0, 1], defined by


 0,
 if 0 ≤ t ≤ 12 − n1 ,
fn (t) = nt − 21 n + 1, if 12 − n1 < t ≤ 21 ,

 1, if 12 < t ≤ 1.

We claim that {fn }n≥2 is a Cauchy sequence but does not converge in (C[0, 1], d1 ).

Indeed,

Z 1
d1 (fm , fn ) = |fm (t) − fn (t)| dt
0
Z 1 Z 1
2 2
≤ fm (t)dt + fn (t)dt
1 1 1
2
−m −1
2 n
 
1 1 1
= + .
2 m n

1 1
Thus, d1 (fm , fn ) < + < ε for all n, m > N, where N is a positive integer such that
m n
N > 2ε . This shows that {fn } is a Cauchy sequence. The measure d1 (fm , fn ) is the area of
the triangle in the following figure.
Qamrul Hasan Ansari Metric Spaces Page 76

1 1
n n

1 x
1 x

fn fn
fm

1
i

1 1
i
t 1
i

1 1
i

1 1
i
t
2
− n 2 2
− n 2
− m 2

d1 (fm , fn ) represents the area of the shaded triangle

Suppose that there is a function f ∈ C[0, 1] such that d1 (fn , f ) → 0. But

Z 1−1 Z 1 Z 1
2 n 2
d1 (fn , f ) = |f (t)| dt + |fn (t) − f (t)| dt + |1 − f (t)| dt. (2.9)
1
0 −1
2 n
1
2

Since the integrands are non-negative, so is the each integral on the right hand side of (2.9).
Since d1 (fn , f ) → 0 as n → ∞, we have
Z 1− 1 Z 1
2 n
lim |f (t)|dt = 0 and |1 − f (t)|dt = 0.
n→∞ 0 1
2

Since f is continuous, we have


0, if 0 ≤ t < 12 ,
f (t) =
1, if 12 ≤ t ≤ 1.
Therefore, f is not continuous which contradicts to our supposition that f is a continuous
function. Hence C[0, 1] is not complete.

Example 2.3.9

The space of all rational numbers Q with the usual metric of absolute value is not
complete.

Verification. Let X = Q be the set of all rational numbers and

d(x, y) = |x − y|, for all x, y ∈ X,
be the usual metric on X. Choose a sequence {xn }, represented in the decimal system, such
that xn = 1.a1 a2 · · · an is the largest rational number satisfying x2n < 2. Then we have the
sequence of numbers
x1 = 1.4, x2 = 1.41, x3 = 1.414, x4 = 1.4142, x5 = 1.41421, . . . .
Qamrul Hasan Ansari Metric Spaces Page 77

We find that
1
d (xm , xn ) = |xn − xm | = 0.00 · · · 0am+1 · · · an < , for all n ≥ m.
10m
Therefore, we conclude that d (xm , xn ) → 0 as m →
√ ∞. So, {xn } is a Cauchy sequence. But
this sequence converges to the irrational number 2 ∈ / Q. Thus {xn } does not have a limit
point in Q and so Q is not complete.

Example 2.3.10

Let X = (0, 1] and d be the usual metric on X. Then (X, d) is not complete as the
sequence xn = n1 : n ∈ N ⊆ X is Cauchy but xn → 0 ∈ / X.

Exercise 2.3.1. Prove or disprove that Rn is a complete metric space with respect to the
following metrics: For all x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) ∈ Rn ,

n
P
(a) d1 (x, y) = |xk − yk |;
k=1

 n
1/p
P p
(b) dp (x, y) = |xk − yk | for all p ≥ 1;
k=1

(c) d∞ (x, y) = max |xk − yk |.

1≤k≤n

Proof. (c) For any real numbers x1 , x2 , . . . , xn , we have

n
!1/2 n n
!1/2
X X X
x2k ≤ |xk | ≤ n max |xk | ≤ n x2k . (2.10)
1≤k≤n
k=1 k=1 k=1

Indeed, the first inequality follows on squaring both the sides. The second one is obvious.
The third inequality is a consequence of the fact that

n
!1/2
X
x2k ≥ |xk | , for k = 1, 2, . . . , n.
k=1

Let x = (x1 , x2 , . . . , xn ) ∈ Rn and y = (y1 , y2 , . . . , yn ) ∈ Rn . From inequality (2.10), we have

n
!1/2 n n
!1/2
X X X
(xk − yk )2 ≤ |xk − yk | ≤ n max |xk − yk | ≤ n (xk − yk )2 . (2.11)
1≤k≤n
k=1 k=1 k=1
Qamrul Hasan Ansari Metric Spaces Page 78

From inequality (2.11), we observe that a sequence in Rn is Cauchy (respectively, convergent)

with respect to one of these metrics if and only if it is Cauchy (respectively, convergent) with
respect to the other metric. Since Rn is complete with respect to the metric

n
!1/2
X
d(x, y) = (xk − yk )2 ,
k=1

it is also complete with respect to the metric d∞ .

Exercise 2.3.2. Prove that [0, 1) as a subspace of the discrete metric space R is complete.

Exercise 2.3.3. Prove that the space of all natural numbers N with the metric

1 1
d(x, y) = − , for all x, y ∈ N,
x y

is not complete.

Proof. Let {n}n≥1 be a sequence in N. Let ε > 0 and N be the least integer greater than 1ε .
If m, n > N, then  
1 1 1 1 1
d(m, n) = − ≤ max , < < ε.
m n m n N
Thus, the sequence {n}n≥1 is Cauchy. Suppose contrary that the sequence {n}n≥1 converges
to some point p ∈ N. Let N1 be any integer greater than 2p. Then n ≥ N1 implies that

1 1 1 1 1 1 1 1 1
d(p, n) = − = − ≥ − > − = .
p n p n p N1 p 2p 2p

This shows that the sequence {n}n≥1 cannot converge to p, a contradiction of our supposition.
Hence (N, d) is not complete.

Exercise 2.3.4. Let X = R and d : X × X → R be defined by

|x − y|
d(x, y) = √ p , for all x, y ∈ X.
1 + x2 1 + y 2

Show that (X, d) is a metric space but not complete.

Proof. We only give the proof of incompleteness. Consider the sequence {n}n≥1 of natural
numbers. Observe that
1
|n − m| n
− m1
d(n, m) = √ √ =q q
1 + n2 1 + m2 1
+ 1 m12 + 1
n2

1 1 1 1
≤ − ≤ + → 0, as n, m → ∞.
n m n m
Qamrul Hasan Ansari Metric Spaces Page 79

Thus {n}n≥1 is a Cauchy sequence in (X, d). Suppose contrary that the sequence {n}n≥1
converges to some point p ∈ R. Then,
|n − p| 1 − np
d(n, p) = √ p =q p
1 + n2 1 + p2 1 + n12 1 + p2
1
→p 6= 0, as n → ∞.
1 + p2
This shows that the sequence {n}n≥1 does not converge to p ∈ R, a contradiction. Hence
(X, d) is not complete.
Exercise 2.3.5. Show that the set A = {(x, y) ∈ R2 : y > 0} is not complete with respect
to the usual metric on R2 .

Hint: Consider a sequence {(xn , yn )} = (0, 1/n) ∈ A. Then (xn , yn ) → (0, 0) ∈

/ A.
Exercise 2.3.6. Prove that the metric space P [a, b] of all polynomials defined on [a, b] with
the uniform metric
d∞ (f, g) = sup |f (t) − g(t)|, for all f, g ∈ P [a, b]
t∈[a,b]

is not complete.

Proof. Let us take a = 0 and b = 1. Consider the following sequence

n  k
X t t tn
fn (t) = = 1 + + · · · + n , for all t ∈ [0, 1].
k=0
2 2 2
Clearly, fn (t) ∈ P [0, 1] for each n ∈ N. We show that the sequence {fn } is Cauchy. Taking
m < n and observe that
d (fn , fm ) = sup |fn (t) − fm (t)|
0≤t≤1
n  k m  k
X t X t
= sup −
0≤t≤1 2k=0 k=0
2
n
 k
t
X
= sup
0≤t≤1
k=m+1
2
n
X 1
≤ sup
0≤t≤1
k=m+1
2k
1 1
= m
− n.
2 2
This difference is arbitrarily small for large enough m and n, which implies that {fn } is
a Cauchy sequence in P [0, 1]. However, this sequence does not converge in (P [0, 1], d),
2
because lim fn (t) = for all 0 ≤ t ≤ 1, is not a polynomial. Therefore, (P [0, 1], d) is
n→∞ 2−t
not complete.
Qamrul Hasan Ansari Metric Spaces Page 80

Theorem 2.3.1. Let (Y, dY ) be a subspace of a metric space (X, d). If Y is complete,
then it is closed.

Proof. Suppose that Y is a complete subspace. To prove Y is closed, it is sufficient to show

that Y contains all its limit points.

Let x be a limit point of Y . Then by Theorem 2.1.3 (a), there exists a sequence {xn } of
distinct points of Y which converges to x. Since each convergent sequence is Cauchy, it is a
Cauchy sequence. Also since Y is complete, the limit point x of this sequence must lie in Y .
Thus, Y is closed.

Theorem 2.3.2. Let (X, d) be a complete metric space and (Y, dY ) be a subspace of
(X, d). Then Y is complete if and only if it is closed.

Proof. If Y is a complete subspace of (X, d), then by Theorem 2.3.1, it is closed.

Conversely, assume that Y is a closed subspace of a complete metric space X. Let {xn } be
a Cauchy sequence of points of Y . Since X is complete, this sequence converges to a point
x belonging to X. By Theorem 2.1.3 (b) and since Y is closed, x ∈ Y . Thus each Cauchy
sequence of points of Y converges to a point of Y . Hence Y is complete.

Theorem 2.3.3 (Cantor’s Intersection Theorem). Let (X, d) be a complete metric

space and let {Fn } be a decreasing sequence (that is, Fn+1 ⊆ Fn ) of nonempty closed
T∞
subsets of X such that δ(Fn ) → 0 as n → ∞. Then the intersection Fn contains
n=1
exactly one point.

Proof. Construct a sequence {xn } in X by selecting a point xn ∈ Fn for each n ∈ N. Since

Fn+1 ⊆ Fn for all n, we have xn ∈ Fn ⊆ Fm for all n > m. We claim that {xn } is a Cauchy
sequence.

Let ε > 0 be given. Since δ(Fn ) → 0, there exists a positive integer N such that δ(Fn ) < ε
for all n > N. Since {Fn } is a decreasing sequence, we have Fm , Fn ⊆ FN for all m, n ≥ N.
Therefore, xn , xm ∈ FN for all n, m ≥ N and thus, we have

d (xn , xm ) ≤ δ(Fn ) < ε, for all n, m ≥ N.

Hence {xn } is a Cauchy sequence. Since X is complete, there exists x ∈ X such that xn → x.
∞
T
We claim that x ∈ Fn .
n=1

Let n be fixed. Then the subsequence {xn , xn+1 , . . .} of the sequence {xn } is contained is
Fn and still converges to x, since every subsequence of a convergent sequence is convergent.
Qamrul Hasan Ansari Metric Spaces Page 81

But Fn being a closed subspace of the complete metric space (X, d), it is complete and so
∞
T ∞
T
x ∈ Fn . This is true for each n ∈ N. Hence x ∈ Fn , that is, Fn 6= ∅.
n=1 n=1

∞
T ∞
T
Finally, to establish that x is the only point in the intersection Fn . Let y ∈ Fn . Then
n=1 n=1
x and y both are in Fn for each n. Therefore,

0 ≤ d(x, y) ≤ δ(Fn ) → 0, as n → ∞.

Thus, d(x, y) = 0, and hence, x = y.

Remark 2.3.2

The assertion in Theorem 2.3.3 may not be true if either of the condition

(i) each Fn is closed, or

(ii) δ(Fn ) → 0 as n → ∞,

is dropped.

Example 2.3.11

(a) Consider the sequence {Fn }, where Fn = [n, ∞) for all n ∈ N, of nonempty closed
subsets of the usual metric space R which, of course, is complete. Then δ(Fn ) 6→ 0
∞
T
as n → ∞ and Fn = ∅.
n=1


(b) Consider the sequence {Fn }, where Fn = 0, n1 for all n ∈ N, of nonempty
subsets of the usual metric space R. Then for each n ∈ N, Fn is open but not
∞
T
closed, Fn+1 ⊆ Fn , δ(Fn ) → 0 as n → ∞ and Fn = ∅.
n=1

Now, we have the converse of Theorem 2.3.3.

Theorem 2.3.4. Let (X, d) be a metric space. If every decreasing sequence {Fn } of
nonempty closed sets in X with δ(Fn ) → 0 as n → ∞ has exactly one point in its
intersection, then (X, d) is complete.
Qamrul Hasan Ansari Metric Spaces Page 82

Proof. Let {xn } be a Cauchy sequence in X. Let

G1 = {x1 , x2 , . . .},
G2 = {x2 , x3 , . . .},
.. ..
. .
Gn = {xn , xn+1 , . . .}.

Then Gn+1 ⊆ Gn and so Gn+1 ⊆ Gn for all n ∈ N. Therefore, Gn is a decreasing sequence
of nonempty closed sets. Since {xn } is a Cauchy sequence, for a given ε > 0, there exists a
positive integer N such that

d (xm , xn ) < ε, for all m, n > N.

Since m, n > N, we have xm , xn ∈ GN and since d (xm , xn ) < ε, we have δ(GN ) < ε. For
n > N, we have Gn ⊆ GN , and thus, δ(Gn ) ≤ δ(GN ) < ε. Therefore, δ(Gn ) → 0 as n → ∞.

Since δ(Gn ) = δ(Gn ), we have δ(Gn ) → 0 as n → ∞. Taking Fn = Gn , then {Fn } is a

decreasing sequence of nonempty closed sets with δ(Fn ) → 0 as n → ∞. Then by hypothesis,
∞
T
there exists an x ∈ X such that x ∈ Fn . Therefore, d (x, xn ) ≤ δ(Fn ) for all n, and so
n=1
d (x, xn ) ≤ δ(Fn ) → 0 as n → ∞. Hence xn → x in X, and thus, (X, d) is complete.
Exercise 2.3.7. Let (X, d) be a complete metric space and {Gn }n∈N be a sequence of open
subsets of X. Prove that
! !
\ \
int cl Gn = int cl Gn . (2.12)
n∈N n∈N

!
\ \
Proof. Since cl Gn ⊃ cl Gn , it is obvious that
n∈N n∈N
! !
\ \
int cl Gn ⊃ int cl Gn . (2.13)
n∈N n∈N

!
\
We show the converse of inclusion (2.13). Let x ∈ int cl Gn and r > 0. Take
\ n∈N
x1 = x. Then there exists r1 ∈ (0, r) such that Sr1 [x1 ] ⊂ cl Gn, and therefore, Sr1 [x1 ] ⊂
n∈N
cl Gn for each n ∈ N. It follows that  Sr1 (x1 ) ∩ G1 is an open nonempty set. Therefore,
r1
there exist x2 ∈ X and r2 ∈ 0, 2 such that Sr2 [x2 ] ⊂ Sr1 (x1 ) ∩ G1 . It follows that
Sr2 [x2 ] ⊂ Sr1 [x1 ] ⊂ cl G2 , and so Sr2 (x2 ) ∩ G2 is an open nonempty set. Continuing in
this way, we find the sequences {xn }n∈N ⊂ X and {rn }n∈N ⊂ (0, ∞) such that rn → 0 as
n → ∞, and Srn+1 [xn+1 ] ⊂ Srn (xn ) ∩ Gn for all n. In particular, Srn+1 [xn+1 ] ⊂ Srn [xn ]
Qamrul Hasan Ansari Metric Spaces Page 83

for
\ all n. Since X is a complete metric space, by Cantor’s Intersection Theorem 2.3.3,
Srn [xn ] = {x̄} for some x̄ ∈ X. It follows that x̄ ∈ Gn for all n. Since x̄ ∈ Sr1 [x1 ] ⊂ Sr (x),
n∈N ! !
\ \
we have that Sr (x) ∩ Gn 6= ∅. As r > 0 is arbitrary, x ∈ cl Gn . Therefore,
! n∈N ! n∈N
\ \
int cl Gn ⊂ cl Gn .
n∈N n∈N

Exercise 2.3.8. Let (X, d) be a complete metric space and {Fn }n∈N be a sequence of closed
subsets of X. Prove that
! !
[ [
cl int Fn = cl int Fn . (2.14)
n∈N n∈N

Proof. It follows directly from Exercise 2.3.7 by taking Gn = X \ Fn for all n ∈ N.

Exercise 2.3.9. If {Fn }∞

n=1 is a sequence of closed subsets of a complete metric space X
∞
[
such that X = Fn , then prove that at least one of Fn ’s has nonempty interior.
n=1

Hint: Follows from Exercise 2.3.8.

Qamrul Hasan Ansari Metric Spaces Page 84

2.4 Completion

If a metric space (X, d) is not complete, then it is always possible to construct a complete
metric space which contains this incomplete metric space and some other points so that
every Cauchy sequence in X has a limit point in the larger space. In fact, we need to adjoin
new points to (X, d) and extend d to all these new points in such a way that the formerly
nonconvergent Cauchy sequences find limits among these new points and the new points are
limits of sequences in X.

Definition 2.4.1

Let (X, d) be a metric space. A complete metric space (X ∗ , d∗) is said to be a completion
of the metric space (X, d) if X is a subspace of X ∗ and every point of X ∗ is the limit of
some sequence in X.

Example 2.4.1

(a) The completion of the space Q of all rational numbers with the usual metric is
the space R of all real numbers with the usual metric.

(b) Let X = (0, 1), (0, 1] or [0, 1). The completion of the metric space X with the
usual metric is the metric space [0, 1] with the usual metric.

Example 2.4.2

The completion of the metric space P [a, b] of all polynomials defined on [a, b] with the
uniform metric

d∞ (f, g) = sup |f (t) − g(t)|, for all f, g ∈ P [a, b], (2.15)

t∈[a,b]

is the complete metric space C[a, b] of all continuous real-valued functions defined on
C[a, b] with the uniform metric (2.15).
Qamrul Hasan Ansari Metric Spaces Page 85

Definition 2.4.2

Let (X, d) and (X ∗ , d∗ ) be two metric spaces.

• A mapping f : X → X ∗ is said to be an isometry if

d∗ (f (x), f (y)) = d(x, y), for all x, y ∈ X.

The mapping f is also called an isometric embedding of X into X ∗ .

• The spaces X and X ∗ are said to be isometric if there exists a bijective isometry
from X to X ∗ .

Remark 2.4.1

We note that every isometry is always one-to-one.

Example 2.4.3

The map f : Q → R2 defined by f (x + iy) = (x, y) is onto and an isometry from the set
of complex numbers with the usual metric to the Euclidean plane with usual metric.

Example 2.4.4

Consider the set X = {a, b, c} with the discrete metric and define a function f : X → R2
by √
f (a) = (−1/2, 0), f (b) = (1/2, 0), f (c) = (0, 3/2).
These are the corners of an equilateral triangle with side length 1. If R2 is equipped
with the usual metric, then f is an isometry. But the two spaces are not isometric,
as there cannot be a bijection between these spaces because first space is finite while
second one is infinite.

(X, d) (R2 , k · k)
−
c f
1 • 1

1 √ 1
3
a• •b 2

d(a, b) = 1 p p
d(b, c) = 1 1 1
d(c, a) = 1
Qamrul Hasan Ansari Metric Spaces Page 86

The definition of a completion of a metric space in terms of isometry can be stated as follows:

A completion of a metric space (X, d) is a complete metric space (X ∗ , d∗) if there exists an
isometry f : X → X ∗ such that f (X) is dense in X ∗ .

Theorem 2.4.1. Every metric space has a completion.

Proof. Let (X, d) be a metric space and let C[X] denote the collection of all Cauchy sequences
in X. Define a relation ∼ on C[X] by
{xn } ∼ {yn } ⇔ lim d(xn , yn ) = 0.
n→∞

Then ∼ is an equivalence relation on C[X], that is, the relation ∼ is reflexive, symmetric
and transitive.1

Thus, C[X] splits into equivalence classes. Any two members of the same equivalence class
are equivalent, and no member of an equivalence class is equivalent to a member of any other
equivalence class. Let X ∗ denote the set of all equivalence classes for ∼, and the elements
of X ∗ will be denoted by [{xn }], [{yn }], etc, that is,
X ∗ = {[{xn }] : {xn } ∈ C[X]} .

Since d(yn , x) ≤ d(yn , xn ) + d(xn , x), we observe that if a Cauchy sequence {xn } has a limit
point x ∈ X, and if {yn } is equivalent to {xn }, then lim yn = x. Also, if {xn } and {yn } are
n→∞
2
nonequivalent sequences, then lim xn 6= lim yn .
n→∞ n→∞

Define d∗ : X ∗ × X ∗ → [0, ∞) by
d∗ ([{xn }], [{yn }]) = lim d(xn , yn ), for all [{xn }], [{yn }] ∈ X ∗ .
n→∞

To show that the limit on the right side exists, we consider any two sequences {xn } ∈ [{xn }]
and {yn } ∈ [{yn }] and observe that
|d(xn , yn ) − d(xm , ym )| ≤ d(xn , xm ) + d(yn , ym ).
1
Reflexivity: ({xn } ∼ {xn }): Since lim d(xn , xn ) = 0 for every n, and so lim d(xn , xn ) = 0 if and only
n→∞ n→∞
if {xn } ∼ {xn }.
Symmetry: ({xn } ∼ {yn } implies {yn } ∼ {xn }): If {xn } ∼ {yn }, then lim d(xn , yn ) = 0 = lim d(yn , xn ),
n→∞ n→∞
and so {yn } ∼ {xn }.
Transitivity: ({xn } ∼ {yn } and {yn } ∼ {zn } imply {xn } ∼ {zn }): If {xn } ∼ {yn } and {yn } ∼ {zn }, then
lim d(xn , yn ) = 0 and lim d(yn , zn ) = 0. Therefore, by the triangle inequality, we have
n→∞ n→∞

0 ≤ lim d(xn , zn ) ≤ lim d(xn , yn ) + lim d(yn , zn ) = 0,

n→∞ n→∞ n→∞

that is, lim d(xn , zn ) = 0, and hence {xn } ∼ {zn }.

n→∞
2
Indeed, if lim xn = x = lim yn , then by the triangle inequality, we have lim d(xn , yn ) = 0, contra-
n→∞ n→∞ n→∞
dicting the fact that {xn } and {yn } are nonequivalent sequences.
Qamrul Hasan Ansari Metric Spaces Page 87

Therefore, {d(xn , yn )}n≥1 is a Cauchy sequence of real numbers. Since R is complete,

lim d(xn , yn ) exists.
n→∞

To show that d∗ is well-defined, let {x′n } and {yn′ } be two Cauchy sequences in X such that
{x′n } ∼ {xn } and {yn′ } ∼ {yn }. Then,

lim d (xn , x′n ) = 0 and lim d (yn , yn′ ) = 0.

n→∞ n→∞

By the triangle inequality, we have

d(xn , yn ) ≤ d (xn , x′n ) + d (x′n , yn′ ) + d (yn′ , yn )

d (x′n , yn′ ) ≤ d (x′n , xn ) + d (xn , yn ) + d (yn , yn′ ) .

Therefore,
|d(xn , yn ) − d (x′n , yn′ )| ≤ d (xn , x′n ) + d (yn , yn′ ) → 0.
Since {d(xn , yn )} and {d (x′n , yn′ )} are convergent, we have

lim d(xn , yn ) = lim d (x′n , yn′ ) .

n→∞ n→∞

Thus, d∗ is well-defined.

Next, we show that d∗ is a metric on X ∗ .

Since d(xn , yn ) ≥ 0 for all n, we have lim d(xn , yn ) ≥ 0, and thus, d∗ ([{xn }], [{yn }]) ≥ 0.
n→∞

Let [{xn }], [{yn }] ∈ X ∗ . Then,

d∗ ([{xn }], [{yn }]) = 0 ⇔ lim d(xn , yn ) = 0

n→∞
⇔ {xn } ∼ {yn }
⇔ [{xn }] = [{yn }].

Also,
d∗ ([{xn }], [{yn }]) = lim d(xn , yn ) = lim d(yn , xn ) = d∗ ([{yn }], [{xn }]) .
n→∞ n→∞

Finally, let [{xn }], [{yn }], [{zn }] ∈ X ∗ . Since d(xn , zn ) ≤ d(xn , yn ) + d(yn , zn ), we have

lim d(xn , zn ) ≤ lim d(xn , yn ) + lim d(yn , zn ),

n→∞ n→∞ n→∞

and therefore,

d∗ ([{xn }], [{zn }]) ≤ d∗ ([{xn }], [{yn }]) + d∗ ([{yn }], [{zn }]) .

Thus, d∗ is a metric on X ∗ .
Qamrul Hasan Ansari Metric Spaces Page 88

For each x ∈ X, let x̂ = [{x, x, x, . . .}] ∈ X ∗ , the equivalence classes of the constant sequence
{x, x, x, . . .}. Define a mapping f : X → X ∗ by f (x) = x̂. Then the constant sequence
{x, x, x, . . . , x, . . .} is a representative of [{xn }], and therefore, the mapping f is one-to-one.
Now for any x, y ∈ X, we have

d∗ (f (x), f (y)) = d∗ (x̂, ŷ) = lim d(x, y) = d(x, y).

n→∞

Thus, f is an isometry from X into X ∗ .

We now show that every point x∗ = [{xn }] ∈ X ∗ is the limit of a sequence in f (X),
equivalently, f (X) is dense in X ∗ . For this purpose, let x∗ = [{xn }] ∈ X ∗ and let ε > 0. Since
{xn } is a Cauchy sequence, there exists a positive integer N such that for any m, n ≥ N,
d(xm , xn ) < 2ε for all m, n ≥ N. Let z = xN . Then ẑ ∈ f (X) and
ε
d∗ ([{xn }], f (xN )) = d∗ (x∗ , ẑ) = lim d(xn , z) = lim d(xn , xN ) ≤ < ε.
n→∞ n→∞ 2
∗
Thus, ẑ ∈ Sεd (x∗ ) ∩ f (X), and hence f (X) is dense in X ∗ .

Finally, we show that X ∗ is complete. Let {x∗n } be a Cauchy sequence in (X ∗ , d∗). Our goal
is to show that this sequence converges to some limit point x∗ ∈ X ∗ .

Since each x∗n is the limit of a sequence in f (X), we can find yn∗ ∈ f (X) such that d∗ (x∗n , yn∗ ) <
1
n
. Then {yn∗ } is a sequence in X ∗ as

d∗ (yn∗ , ym
∗
) ≤ d∗ (yn∗ , x∗n ) + d∗ (x∗n , x∗m ) + d∗ (x∗m , ym
∗
)
1 1
≤ + d∗ (x∗n , x∗m ) + → 0 as n, m → ∞.
n m
Since yn∗ is in f (X), there exists some yn ∈ X such that f (yn ) = yn∗ . Since {yn∗ } is a Cauchy
sequence in X ∗ and f is an isometry, {yn } in X must be Cauchy. Therefore, {yn } belongs
to some equivalence class x∗ ∈ X ∗ . We show that lim d∗ (x∗n , x∗ ) = 0. For this, take any
n→∞
ε > 0. Since {yn } is a Cauchy sequence in X, we have
1
d∗ (x∗n , x∗ ) ≤ d∗ (x∗n , yn∗ ) + d∗ (yn∗ , x∗ ) < + d∗ (yn∗ , x∗ )
n
and
d∗ (yn∗ , x∗ ) = d∗ (f (yn ), x∗ ) = lim d(yn , yp ) ≤ ε
n→∞
∗
for sufficiently large n. This implies that lim d (x∗n , x∗ ) = 0, thereby completing the proof
n→∞
that X ∗ is complete.

Theorem 2.4.2. All completions of a metric space are isometric.

Proof. Let (X ∗ , d∗ ) and (X ∗∗ , d∗∗ ) be any two completions of the metric space (X, d). Let
x∗ ∈ X ∗ be arbitrary. By the definition of completion, there exists a sequence {xn }n∈N in
Qamrul Hasan Ansari Metric Spaces Page 89

X such that xn → x∗ . The sequence {xn }n∈N may be assumed to belong to X ∗∗ . Since X ∗∗
is complete, {xn }n∈N converges in X ∗∗ to x∗∗ , say. Define f : X ∗ → X ∗∗ by f (x∗ ) = x∗∗ .
It is clear that the mapping f is one-to-one and does not depend on the choice of the
sequence {xn }n∈N converges to x∗ . Moreover, by construction, f (x) = x for all x ∈ X. If
xn → x∗ ∈ X ∗ and xn → x∗∗ ∈ X ∗∗ while yn → y ∗ ∈ X ∗ and yn → y ∗∗ ∈ X ∗∗ , then

d∗ (x∗ , y ∗) = lim d (xn , yn ) and d∗∗ (x∗∗ , y ∗∗) = lim d (xn , yn ) .

n→∞ n→∞

Thus,
d∗∗ (f (x∗ ), f (y ∗)) = d (x∗ , y ∗ ) , for all x∗ , y ∗ ∈ X ∗ ,
and hence, f is an isometry. Therefore, (X ∗ , d∗ ) and (X ∗∗ , d∗∗ ) are isometric.

Exercise 2.4.1. Let X and Y be metric spaces. If X is complete and isometric to Y , then
prove that Y is complete.
 Chapter 3

Separable Spaces

René-Louis Baire was born on January 21, 1874,

in Paris, Ile-de-France, France and was died on
July 5, 1932, at the age of 58 in Chambéry,
France. He is most famous for his Baire cat-
egory theorem, which helped to generalize and
prove future theorems. Among Baire’s other
most important works are Théorie des nombres
irrationels, des limites et de la continuité (Theory
of Irrational Numbers, Limits, and Continuity)
published in 1905 and both volumes of Leçons
sur les théories générales de l’analyse (Lessons
on the General Theory of Analysis) published in
1907–08.
René-Louis Baire
It is well known that the rational numbers or irrational numbers are densely packed along
the real line. This phenomenon can be characterized in terms of distance by saying that
every point of R is at zero distance from the set of rational numbers Q or from the set of
irrational numbers R \ Q. The concept of closure enabled us to write this density property
by the formula Q = R or R \ Q = R. In this chapter, we extend the idea of density to
an arbitrary metric space. We discuss some other properties of a metric space, known as
topological properties, such as countability, separability and Baire’s category theorem.

90
Qamrul Hasan Ansari Metric Spaces Page 91

3.1 Countability

Definition 3.1.1

Let O be the family of all open subsets of a metric space X. A subfamily B ⊆ O of

open subsets of X is said to be a base or basis for O if every open set G ∈ O is the
union of members of B.

Further, if the subfamily B is countable, then it is called countable base or countable

basis for O.

Before giving the examples of a base for a family of open sets, we mention the following
characterization of a base.

Theorem 3.1.1. Let O be the family of all open subsets of a metric space X. A
subfamily B ⊆ O is a base for O if and only if for every point x belonging to an open
set G, there exists Bx ∈ B such that x ∈ Bx ⊆ G.

Proof. Let B be a base for O and G be any open set in X, that is, G ∈ O. Then by the
definition a base, G is the union of members of B. Let x ∈ G. Since G is the union of
members of B, there exists a set Bx in B such that x ∈ Bx ⊆ G.

Conversely, let G be an arbitrary open set. Then by hypothesis,

S for any point x ∈ G, there
exists Bx in B such that x ∈ Bx ⊆ G. Clearly, G = {Bx ∈ B : x ∈ G}. Since G was
arbitrary, every open set is the union of members of B.

Example 3.1.1

The collection B = {(a, b) : a, b ∈ R} of all open intervals forms a base for the family of
all open sets in the usual metric space R.

If the endpoints a and b are rational, then the collection B = {(a, b) : a, b ∈ Q} of all
open intervals forms a countable base for the family of all open sets in the usual metric
space R.
Qamrul Hasan Ansari Metric Spaces Page 92

Example 3.1.2

Let X be a metric space. The collection {Srx (x) : x ∈ X and rx > 0} of all open spheres
forms a base for the family of all open sets in X.

Indeed, let G be a nonempty open subset of X and let x ∈ G. By the definition of an

open set, there exists rx > 0 (depending on x) such that x ∈ Srx (x) ⊆ G.

Example 3.1.3

Let X = Rn be a metric space with the metric

n
!1/p
X
dp (x, y) = |xi − yi |p , for all p ≥ 1,
i=1

where x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) in Rn . Let r > 0 be a rational number.

The collection {Sr (x) : x = (x1 , x2 , . . . , xn ) where xi is rational for all i = 1, 2, . . . , n} of
all open spheres with rational centers and rational radii is a countable base for the family
of all open sets of X.

Example 3.1.4

Let X be a discrete metric space. Then the collection B = {{x} : x ∈ X} forms a base
for the family of all open subsets of X since every subset of a discrete metric space is
open.

If X is countable, then the family B is also countable. In this case, B is a countable

base for the family of all open subsets of X.

Definition 3.1.2

A metric space X is said to be a second countable space (or second axiom space) if there
exists a countable base for the family of all open subsets of X.
Qamrul Hasan Ansari Metric Spaces Page 93

Example 3.1.5

In view of Example 3.1.1, the collection of all open intervals (a, b) with a and b as
rational points forms a countable base for the family of all open subsets of R. Hence
the usual metric space R is a second countable space.

Example 3.1.6

Let X = Rn be a metric space with the metric

n
!1/p
X
dp (x, y) = |xi − yi |p , for all p ≥ 1,
i=1

where x = (x1 , x2 , . . . , xn ) and y = (y1 , y2, . . . , yn ) in Rn . In view of Example 3.1.3,

(X, dp ) is a second countable space.

Example 3.1.7

In view of Example 3.1.4, the discrete metric space X is second countable if X is

countable.

Definition 3.1.3

A metric space X is said to be a first countable space (or first axiom space) if for every
point x ∈ X, there exists a countable family {Bn (x)} of open sets containing x such that
every open set G containing x also contains a member of {Bn (x)}, that is, Bn (x) ⊆ G
for some n.

Example 3.1.8

The usual metric

space R is a first countable space. Indeed, we may take Bn (x) =
1 1
x − n , x + n for each x ∈ R and n ∈ N.

Remark 3.1.1

Every second countable metric space is first countable but converse is not true.
Qamrul Hasan Ansari Metric Spaces Page 94

Example 3.1.9

Let X be a discrete metric space, where X is an uncountable set. Then X is first

countable but not second countable.

Indeed, any open sphere

 S1/2 (x) is a singleton set {x} because X is a discrete metric
∞
space. Therefore, S1/n (x) n=1 is the countable family of open sets containing x such
 ∞
that every open set G containing x also contains a member of S1/n (x) n=1 because
S1/n (x) = {x} for all n = 1, 2, . . .. Hence, X is a first countable space.

If X is uncountable, then the singleton sets {x} are also uncountable. Hence every base
is uncountable. Consequently, X is not second countable when it is uncountable.

Theorem 3.1.2. Every metric space X is a first countable space.


Proof. Let x ∈ X and n ∈ N. Set Bn (x) = S1/n (x). Then {Bn (x)} = B1 (x), B1/2 (x) ,
B1/3 (x), . . . is a countable collection of open subsets of X such that every member of this
collection contains x. Let G be an open set containing x. Then there exists Sε (x) such that
Sε (x) ⊆ G for some ε > 0. Therefore, Bn (x) = S1/n (x) ⊆ Sε (x) ⊆ G for each n > 1ε , and
hence, X is a first countable space.
Exercise 3.1.1. Let (X, d) be a metric space and Y be a nonempty subset of X. If (X, d)
is second countable, then prove that (Y, dY ) is also second countable.

Proof. Let B = {Bn : n ∈ N} be a countable base for the family O of all open subsets of X.
Then B∗ = {Bn ∩ Y : n ∈ N} is countable. We claim that B∗ is a base for the family G of
all open subsets of Y .

Let G be an open set in Y and let x ∈ G. Then there exists an open set O in X such that
G = O ∩ Y . Also, x ∈ O ∩ Y and so x ∈ O and x ∈ Y . Since O ∈ O, there exists a Bn0 ∈ B
such that x ∈ Bn0 ⊆ O. Since x ∈ Y , we have x ∈ Bn0 ∩ Y ⊆ O ∩ Y = G. But Bn0 ∩ Y ∈ B∗ .
Therefore, B∗ forms a base for G. Hence, (Y, dY ) is second countable.
Exercise 3.1.2. Let (X, d) be a metric space and Y be a nonempty subset of X. If (X, d)
is first countable, is so (Y, dY )? Justify your answer.
Exercise 3.1.3. Let X = R2 be the usual metric space, x = (x1 , x2 ) ∈ X be any arbitrary
element and Gλ (x) = {y = (y1 , y2 ) ∈ R2 : (y1 − x1 )2 + 3(y2 − x2 )2 < λ}, where λ > 0. Prove
that the family {Gλ (x) : λ > 0 and x ∈ X} is a base for the family of all open subsets of X.

Proof. Let G be any open set in X and let x ∈ G. Then, there exists r > 0 such that Sr (x) ⊆
G. We know that Sr (x) = {y = (y1 , y2 ) ∈ X : (y1 − x1 )2 + (y2 − x2 )2 < r}. Let λ = r 2 . If
y ∈ Gλ (x), then (y1 −x1 )2 +3(y2 −x2 )2 < λ which implies that (y1 −x1 )2 +(y2 −x2 )2 < λ = r 2 .
Therefore, y ∈ Sr (x) and so y ∈ Gλ (x) ⊆ Sr (x) ⊆ G. Hence, {Gλ (x) : λ > 0 and x ∈ X} is
a base for the family of all open subsets of X.
Qamrul Hasan Ansari Metric Spaces Page 95

3.2 Dense Sets

Definition 3.2.1

A nonempty subset A of a metric space X is said to be dense (or everywhere dense) in

X if A = X, that is, if every point of X is either a point or a limit point of A.

In other words, a set A is dense in X if for any given point x ∈ X, there exists a
sequence of points of A that converges to x.

Before giving the examples of dense sets, we provide some criterion for being dense.

Theorem 3.2.1. Let A be a nonempty subset of a metric space X. The following

statements are equivalent.

(a) For every x ∈ X, ρ(x, A) = 0.

(b) A = X.

(c) A has nonempty intersection with every nonempty open subset of X.

Proof. The equivalence of (a) and (b) directly follows from Theorem 1.4.1. We prove the
equivalence of (b) and (c).

(b) ⇒ (c). Suppose contrary that there exists a nonempty open subset O of X such that
A ∩ O = ∅. Then A ⊆ O c , and so, A ⊆ O c = O c because O c is closed. Therefore, A ⊆ O c
whence A ∩ O = ∅, and thus, A 6= X, a contradiction of (b). Hence, (c) holds.

c
(c) ⇒ (b). Suppose that (c) holds but (b) does not, that is, A 6= X. Then A 6= ∅. Since

c
c
A is a closed set, A is an open set in X and disjoint from A, that is, A ∩ A = ∅. Since

c
c
A is nonempty open subset of X, by hypothesis, A ∩ A 6= ∅, a contradiction. Hence,
A = X.

Remark 3.2.1

It can be easily seen that a subset A of X is dense if and only if Ac has empty interior.
Qamrul Hasan Ansari Metric Spaces Page 96

Example 3.2.1

(a) The set of all rational numbers Q is dense in the usual metric space R since Q = R.

(b) Since R \ Q = R, the set of all irrational numbers R \ Q is dense in the usual
metric space R.

(c) The set A = {a + ib ∈ C : a, b ∈ Q} is dense in C since A = C.

(d) The set Qn = Q × Q × · · · × Q is dense in Rn with the usual metric.

| {z }
n−times

Example 3.2.2

The set

A = {x = (a1 , a2 , . . . , an , 0, 0, . . .) : ai ∈ Q for all 1 ≤ i ≤ n and n ∈ N}

is dense in the space ℓp , 1 ≤ p < ∞, with the following metric

∞
!1/p
X
dp (x, y) = |xi − yi |p ,
i=1

where x = {x1 , x2 , . . .} and y = {y1 , y2, . . .} in ℓp .

Indeed, let x = {x1 , x2 , . . .} ∈ ℓp be an arbitrary element and let ε > 0 be given. Then
there exists a positive integer N such that
∞
X εp
|xi |p < .
i=N +1
2

Since Q is dense in R, for every real number, there is a rational number which is as
close to it as we wish. Consequently, we can choose an element y = (a1 , a2 , . . . , aN , 0, 0,
. . .) ∈ A such that
N
X εp
|xi − ai |p < .
i=1
2
Qamrul Hasan Ansari Metric Spaces Page 97

Therefore,
∞
X
p
[dp (x, y)] = |xi − ai |p
i=1
XN ∞
X
p
= |xi − ai | + |xi |p
i=1 i=N +1
p p
ε ε
< + = εp ,
2 2
and this implies that
dp (x, y) < ε.
It means that the distance from x ∈ ℓp to A is zero. By Theorem 3.2.1, A is dense in
ℓp .

Example 3.2.3

The set P [a, b] of all polynomials defined on [a, b] with rational coefficients is dense in
C[a, b].

Indeed, let f ∈ C[a, b]. By Weierstrass’s Theorem (Theorem E.0.2), there exists a
sequence {pn } of polynomials with real coefficients that converges uniformly to f on
[a, b], that is, for each ε > 0, there exists a positive integer N such that for all t ∈ [a, b]

|pn (t) − f (t)| < ε whenever n>N

and so d∞ (pn , f ) < ε for all n > N. But any polynomial can be uniformly approx-
imated by a polynomial with rational coefficients since Q is dense in R. Therefore,
corresponding to the sequence {pn }, there exists a sequence {qn } of polynomials with
rational coefficients (that is, {qn } ∈ P [a, b]) such that
1
|pn (t) − qn (t)| < , for all t ∈ [a, b].
n
This implies that
1
d∞ (pn , qn ) = sup |pn (t) − qn (t)| < , for all n = 1, 2, . . . .
t∈[a,b] n

By the triangle inequality, we have

d∞ (f, qn ) ≤ d∞ (f, pn ) + d∞ (pn , qn ) → 0 as n → ∞.

Hence, qn → f ∈ C[a, b] as n → ∞, and thus, P [a, b] = C[a, b].

Qamrul Hasan Ansari Metric Spaces Page 98

Example 3.2.4

(a) Let (X, d) be a discrete metric space. Since every subset of X is closed, the only
dense subset of X is itself.

(b) By the definition of a completion, every metric space is dense in its completion.

Theorem 3.2.2. Let {Gn } be a sequence of dense open sets in a complete metric space
∞
\
X. Then Gn is also dense in X.
n=1

Proof. Let
! S be any open sphere in X. In view of Theorem 3.2.1, it is sufficient to show that
∞
\
Gn ∩S 6= ∅. Since each Gn is dense in X, Gn = X and Gn ∩S is nonempty by Theorem
n=1
3.2.1 and open as the intersection of two open sets. In particular, G1 ∩ S is a nonempty open
set. Therefore, there exist x1 ∈ G1 ∩ S and r1 > 0 such that Sr1 [x1 ] ⊂ Sr1 +ε (x1 ) ⊆ G1 ∩ S
for ε > 0. Similarly, G2 ∩ Sr1 (x1 ) is nonempty and open, so there exist x2 ∈ G2 ∩ Sr1 (x1 )
and r2 > 0 with r2 ≤ r21 such that Sr2 [x2 ] ⊆ G2 ∩ Sr1 (x1 ). Continue in this way, we obtain a
sequence {xn } such that

Srn [xn ] ⊆ Gn ∩ Srn−1 (xn−1 ), for all n = 1, 2, . . . ,

rn−1 r1
with rn ≤ 2
= 2n−1
→ 0 as n → ∞.

Sr1 +ε (x1 )
Sr1 [x1 ]

Sr3 [x3 ]
x1
ε r1 •
•x
3 x
• 2
Sr2 [x2 ]

Observe that whenever m > n, we have

Sr1 [x1 ] ⊇ · · · ⊇ Srn [xn ] ⊇ · · · ⊇ Srm [xm ].

Qamrul Hasan Ansari Metric Spaces Page 99

Note that X is a complete metric space and {Srn [xn ]} is a decreasing sequence of nonempty
closed subsets of X such that δ (Srn [xn ]) → 0 as n → ∞. By Cantor’s Intersection Theorem
∞
\
2.3.3, Srn [xn ] contains exactly one point, say, x. Then,
n=1

∞
\ ∞
\
{x} = Srn [xn ] ⊆ Gn .
n=1 n=1

∞
! ∞
\ \
Since x is also in S, we have Gn ∩ S 6= ∅. Hence Gn is dense in X.
n=1 n=1

Definition 3.2.2

A metric space X is said to be separable if there exists a countable dense set in X.

A metric space which is not separable is called inseparable.

Example 3.2.5

(a) The usual metric space R is separable since the set of all rational numbers Q is
dense in R.

(b) The usual metric space C is separable since the set A = {a + ib ∈ C : a, b ∈ Q} is

dense in C.

(c) The Euclidean space Rn is separable since the set Qn = Q × Q × · · · × Q is count-

| {z }
n−times
able and dense in Rn .
Qamrul Hasan Ansari Metric Spaces Page 100

Example 3.2.6

The space ℓp , 1 ≤ p < ∞, is separable as the set

A = {x = (a1 , a2 , . . . , an , 0, 0, . . .) : ai ∈ Q, 1 ≤ i ≤ n and for all n ∈ N}

is countable and dense in the space ℓp .

Indeed, if An denotes the subset of all those elements x = {ai }i≥1 such that aj = 0 for
∞
[
all j ≥ n + 1, then An is countable, and so, A = An . By Example 3.2.2, A is dense
n=1
in the space ℓp .

Example 3.2.7

(a) The space C[a, b] is separable since the set P [a, b] of all polynomials defined on
[a, b] with rational coefficients is countable and dense in C[a, b] by Example 3.2.3.

(b) A discrete metric space X is separable if and only if the set X is countable.

Example 3.2.8

The space ℓ∞ of all bounded sequences of real or complex numbers with the metric

d(x, y) = sup |xn − yn |

1≤n<∞

where x = {xn } and y = {yn } in ℓ∞ , is not separable.

Consider a subset A = {x = {xn } ⊆ ℓ∞ : xn = 0 or 1} of ℓ∞ . We claim that A is

uncountable.

If B is any countable subset of A, then the elements of B can be arranged in a sequence

s1 , s2 , . . .. We construct a sequence s as follows. If the mth element of sm is 1, then the
mth element of s is 0, and vice versa. Then the element s of X differs from each sm in
the mth place and is therefore equal to none of them. So, s ∈ / B although s ∈ A. This
shows that any countable subset of A must be a proper subset of A. It follows that A
must be uncountable, for if it were to be countable, then it would have to be a proper
subset of itself, which is absurd. We proceed to use the uncountability of the subset A
to argue that ℓ∞ must be inseparable.
Qamrul Hasan Ansari Metric Spaces Page 101

The distance between two distinct elements x and y of A is

d(x, y) = sup |xn − yn | = 1.

1≤n<∞

Suppose contrary that E is a countable dense subset of ℓ∞ . Consider the spheres of radii
1
3
whose centers at the points of E. Then their union is the entire space ℓ∞ , because
E is dense, and in particular contains A. Since the spheres are countable in numbers
while A is not, in at least one sphere there must be two distinct elements x and y of A.
Let x0 denote the center of such a sphere. Then,
1 1
1 = d(x, y) ≤ d(x, x0 ) + d(x0 , y) < + < 1,
3 3
which is impossible. Hence, (ℓ∞ , d) is not separable.

Theorem 3.2.3. Let (X, d) be a metric space and Y be a nonempty subset of X. If X

is separable, then so is Y with respect to the induced metric.

Proof. Let S = {xn : n ∈ N} be a countable dense subset of X. If S is contained in Y ,

then there is nothing to prove. Otherwise, we construct a countable dense subset of Y whose
points are arbitrary close to those of S. For positive integers m and k, let Sm,k = S 1 (xm )
k
and choose ym,k ∈ Sm,k ∩ Y whenever this set is nonempty. Then d (xm , ym,k ) < k1 . We claim
that the countable subset A = {ym,k : m and k are positive integers} of Y is dense in Y .
1 ε
Let y ∈ Y and ε > 0 be given. Let k be so large that k
< 2
and find xm ∈ S 1 (y). Then
k
d(y, xm) < k1 and y ∈ Sm,k ∩ Y . Thus,

1 1 ε ε
d(y, ym,k ) ≤ d(y, xm ) + d(xm , ym,k ) < + < + = ε.
k k 2 2
Therefore, ym,k ∈ Sε (y). Since y ∈ Y and ε were arbitrary, it follows that y ∈ A, and hence,
A = Y . Thus, A is countable dense in Y .

Theorem 3.2.4. A metric space (X, d) is separable if and only if it is second countable.

Proof. Let (X, d) be a separable space and A = {an : n ∈ N}  be a countable dense subset of
X. Consider the countable collection of open spheres B = S1/k (aj ) : j, k ∈ N with centers
at aj , j = 1, 2, . . . and radii rk = k1 are rational for all k = 1, 2, . . ., that is,

B = {Sri (aj ) : aj ∈ A for j = 1, 2, . . . and ri is a rational number for i = 1, 2, . . .} .

Then we claim that B is a base for the family of all open sets in X. Let G be any open
set in X and x ∈ G. Then for some r > 0, Sr (x) ⊆ G. Choose a positive integer m such
Qamrul Hasan Ansari Metric Spaces Page 102

that m1 < 2r . Since A is dense in X, there exists aj ∈ A such that aj ∈ S1/m (x), that is,
d (x, aj ) < m1 . We now show that S1/m (aj ) ⊆ Sr (x). Let y ∈ S1/m (aj ). Then d (aj , y) < m1 .
By the triangle inequality, we have
1 1 2
d(x, y) ≤ d(x, aj ) + d(aj , y) < + = < r.
m m m
This implies that y ∈ Sr (x). Hence S1/m (aj ) ⊆ Sr (x) ⊆ G. Also, x ∈ S1/m (aj ). Thus,
for every x ∈ G, there exists S1/m (aj ) ∈ B containing x and contained in G. Hence B is a
countable base for the family of all open sets in X, and thus, X is a second countable space.

Conversely, assume that X is a second countable space. Let B = {Bn : n ∈ N} be a countable

base for the family of all open subsets of X. For each n ∈ N, we choose a point bn ∈ Bn .
Let A = {bn ∈ Bn : n ∈ N}. Then A is countable. We show that A = X. Let x ∈ X be
arbitrary and G be any open set containing x. Since B is a base, there exists at least one
Bn0 ∈ B such that x ∈ Bn0 ⊆ G. From the definition of A, we have bn0 ∈ A for all bn0 ∈ Bn0 .
Therefore, G contains a point of A other than x, and thus, x is a limit point of A. Hence,
x ∈ A, and so, A = X.
√
Exercise 3.2.1. Let S = {n + m 2 : n, m ∈ Z} and a, b ∈ R be such that a < b. Prove
that there exists an s ∈ S such that a < s < b. In other words, S is dense in R.

√
Proof. If x, y ∈ S and k ∈ Z,√then x ± y, kx ∈ √ S. Let n(m) := [m 2], the greatest
equal to m 2. Then, 0 < m 2 − n(m) <√1. It is easy to see that if
integer√less than or √
n + m 2 = n′ + m′ 2, then n = n′ and m = m′ . Let sm := m 2|n(m). Then 0 < sm < 1
and sm ∈ S. Also, if m 6= m′ , then sm 6= sm′ . Hence we conclude that {sm : m ∈ Z} is
an infinite subset of S ∩ [0, 1). Given ε > 0, we partition [0, 1) into k equal parts so that
each subinterval has length less than ε. At least one of these subintervals must contain two
distinct elements, say, sm , sm′ of S ∩ [0, 1). Without loss of generality, we may assume that
sm < sm′ . Then we have 0 < sm′ − sm < ε. Since sm′ − sm ∈ S, we have shown that given
ε > 0, there exists an element s ∈ S with 0 < s < ε. Now, let ε > 0 such that b − a > ε be
given. Then there exists n ∈ Z such that a < nε < b. For, choose n to be the least integer
k such that kε > a. Then (n − 1)ε ≤ a < nε. We claim that nε < b. For, otherwise, a
contradiction.

We take ε := (b|a)
2
. Then there exists s ∈ S such that 0 < s < ε. Hence there exists an
integer n such that a < ns < b. Since ns ∈ S, the result is proved.

Exercise 3.2.2. Let A be a nonempty subset of a metric space X. Prove that the following
statements are equivalent.

(a) A is dense in X.

(b) The only closed superset of A is X.

(c) The only open set disjoint from A is ∅.

Qamrul Hasan Ansari Metric Spaces Page 103

(d) A intersects every nonempty open set.

(e) A intersects every open sphere.

[ Let {Gn } be a sequence of dense open sets in a complete metric space X.

Exercise 3.2.3.
Prove that Gn is dense in X.
n∈N

Hint: The proof lies on the lines of the proof of Theorem 3.2.2.

Exercise 3.2.4. Let X be a metric space, A be a nonempty subset of X and x ∈ X. Prove

that x is a limit point of A if and only if A \ {x} is dense in A.

Proof. Suppose that x is a limit point of A. Then for every r > 0, (Sr (x) \ {x}) ∩ A 6= ∅,
that is, (Sr (x) ∩ (A \ {x}) 6= ∅. Then by Exercise 3.2.2 (e), A \ {x} is dense in A.

Conversely, assume that A \ {x} is dense in A. Then again by Exercise 3.2.2 (e), for any
r > 0, Sr (x) meets A \ {x}, that is, Sr (x) \ {x} meets A. Hence, x is a limit point of A.

Exercise 3.2.5. Let (X, d) be a metric space and Y be a subset of X. If Y is separable and
Y = X, then prove that X is separable.

Proof. Let A be a countable dense subset of Y . Let x ∈ X and ε > 0 be given. Since Y = X,
there exists y ∈ Y such that d(x, y) < 2ε . Also, since A is dense in Y , there exists z ∈ A such
that d(y, z) < 2ε . By the triangle inequality, we have
ε ε
d(x, z) ≤ d(x, y) + d(y, z) < + = ε.
2 2
It follows that A is dense in X, and hence, X is separable.
Qamrul Hasan Ansari Metric Spaces Page 104

3.3 Nowhere Dense Sets

Definition 3.3.1

◦
A subset A of a metric space X is said to be nowhere dense in X if A = ∅, that is,
A contains no interior point.

Example 3.3.1

(a) In the usual metric space R, any finite set is nowhere dense.

(b) The sets N and Z are nowhere dense in R since N = N and N◦ = ∅.

(c) In the usual metric space R2 , the set of points on a line is nowhere dense.

(d) In the discrete metric space, the empty set is the only nowhere dense set.

Remark 3.3.1

The notion of nowhere dense is not the opposite of everywhere dense, that is, if a set is
not nowhere dense then it does not imply that the set is everywhere dense. For example,
consider the usual metric space R and the set A = {x ∈ R : 1 < x < 2}. Since

◦
c
A = A 6= ∅ and A = {x ∈ R : x < 1 or x > 2} = (−∞, 1) ∪ (2, ∞),

A is not nowhere dense as well as not everywhere dense.

Remark 3.3.2

In a metric space,

(a) any subset of a nowhere dense set is nowhere dense;

(b) the closure of a nowhere dense set is nowhere dense.

Theorem 3.3.1. Let A be a subset of a metric space X. Then A is nowhere dense in

X if and only if X \ A is dense in X.

 
Proof. Since X \ A = X \ A◦ , we have A◦ = X \ X \ A . Replacing A by A, we get
(A)◦ = X \ (X \ A). Therefore, (A)◦ = ∅ if and only if X = (X \ A).
Qamrul Hasan Ansari Metric Spaces Page 105

Corollary 3.3.1. Let A be a closed subset of a metric space X. Then A is nowhere

dense in X if and only if X \ A is dense in X.

Remark 3.3.3

For any subset A of a metric space X, we have X \ A ⊆ X \ A. Therefore, from Theorem

3.3.1, it follows that if A is nowhere dense in X, then X \ A is dense in X. However,
the converse is not true. For example, the set of all irrational numbers R \ Q is dense
in the usual metric space R. However Q is not nowhere dense set in R because Q = R.

Definition 3.3.2

A subset A of a metric space X is said to be of first category if it is a countable union

of nowhere dense subsets. The set A is said to be of second category if it is not of first
category.

Remark 3.3.4

(a) Every subset of a metric space must be either of first category or of second category.
∞
[
(b) If {An }∞
n=1 be a countable family of first category sets, then An is of first
n=1
category because a countable union of countable sets is also countable.

(c) If X is a metric space of second category and A1 ⊆ X is of first category such

that X = A1 ∪ A2 , then A2 is of second category.

(d) A subset of a set of first category is of first category, because a subset of a nowhere
dense set is nowhere dense.

Example 3.3.2

(a) The empty set is of first category.

(b) The set of all rational numbers Q in R is of first category.

Indeed, if x1 , x2 , . . . is an enumeration of the rationalSnumbers, then each singleton
set {xi } is closed and {xi }◦ = ∅. It follows that {xi }, the set of all rational
numbers in R, is of first category.
Qamrul Hasan Ansari Metric Spaces Page 106

Theorem 3.3.2 (Baire’s Category Theorem). Every complete metric space is of second
category.

Proof. Let {An }∞

n=1 be a sequence of nowhere dense sets in a complete metric space X. Then
∞
[
we have to show that X 6= An , that is, there exists an element x ∈ X such that x ∈
/ An
n=1
∞
[
for all n ∈ N. Suppose that X = An . Since An is nowhere dense, An has no interior
n=1
point, so any open sphere in X must intersects Gn = X \ An for all n = 1, 2, . . .. Therefore,
\∞
{Gn } is a family of dense open sets in X. By Theorem 3.2.2, Gn 6= ∅. It follows that
n=1
∞
[
An 6= X, which is a contradiction.
n=1

Remark 3.3.5

Since R is complete, by Baire’s category theorem, R cannot be written as a countable

union of nowhere dense sets.

Corollary 3.3.2. If M is of first category subset of a complete metric space (X, d),
then M c is dense in X.

Corollary 3.3.3. A complete metric space is not a countable union of nowhere dense
sets.

Exercise 3.3.1. Let A be a subset of a metric space X. Prove that the following statements
are equivalent.

(a) A is nowhere dense in X.

(b) A does not contain any nonempty open set.

(c) Every nonempty open set has a nonempty open subset disjoint from A.

(d) Every nonempty open set contains a nonempty open subset disjoint from A.

(e) Every nonempty open set contains an open sphere disjoint from A.
Exercise 3.3.2. Prove that the Cantor set is nowhere dense.
Exercise 3.3.3. Let X be a metric space. Prove that a closed set F is nowhere dense if and
only if it does not contain any open sphere.
Qamrul Hasan Ansari Metric Spaces Page 107

Proof. If F contains an open sphere, then obviously it is not nowhere dense.

Assume that F does not contain any open sphere. Given a nonempty open set G, we know
that F cannot be contained in G as G contains open spheres and F does not. Pick an
element x in G that is not in F . Since F is closed, there is r1 > 0 such that Sr1 (x) does
not intersect F . Since F is open, there is r2 > 0 such that Sr2 (x) is contained in G. Choose
r = min{r1 , r2 }. Then Sr (x) is contained in G and does not intersect F , and hence, F is
nowhere dense.

Exercise 3.3.4. Let A be an open subset of a metric space X. Prove that A is dense in X
if and only if X \ A is nowhere dense in X.

Exercise 3.3.5. Prove that a finite union of nowhere dense sets in a metric space is a
nowhere dense set.

Proof. Let A1 and A2 be nowhere dense subsets of a metric space X and let G be an
arbitrary open set in X. Then by Exercise 3.3.1 (d), there exists an open set G1 ⊆ G such
that G1 ∩ A1 = ∅. Similarly, there exists an open set G2 ⊆ G1 ⊆ G such that G2 ∩ A2 = ∅.
Since
G2 ∩ (A1 ∪ A2 ) = (G2 ∩ A1 ) ∪ (G2 ∩ A2 ) = ∅,
G contains a nonempty open set G2 such that G and A1 ∪A2 are disjoint. Again, by Exercise
3.3.1 (d), A1 ∪ A2 is nowhere dense. By induction, one can easily prove that the finite union
of nowhere dense sets is nowhere dense.

Exercise 3.3.6. Give an example to show that a countable (infinite) union of nowhere dense
sets in a metric space X need not be a nowhere dense set in X.

Solution. Let X = R be a metric space with the usual metric. Consider the set Q of all
rational numbers as a subset of X. We can write
[  p  
Q= : p ∈ Z, q ∈ N .
q
n o
Then Q has been written as a countable union of nowhere dense sets pq in R. It is well
known that Q is not nowhere dense set in R as Q is everywhere dense set in R.

Exercise 3.3.7. Let X be a metric space. If X has no isolated point, then prove that every
finite set is nowhere dense.

Hint: A singleton set {x} is open if and only if x is an isolated point of X, and then use
Exercise 3.3.5.

Exercise 3.3.8. Let (X, d) be a metric space, and A and B be nonempty subset of M.

(a) If the derived set A′ of A is nowhere dense, then prove that A is nowhere dense.
Qamrul Hasan Ansari Metric Spaces Page 108

(b) If B is open, then prove that the boundary set b(B) of B is nowhere dense.

Exercise 3.3.9. Prove that the subset R \ Q, the set of all irrational numbers, of R is of
second category.

Proof. Since R is a complete metric space, by Baire’s category Theorem 3.3.2, R is of second
category. By Example 3.3.2 (b), the set Q of all rational numbers is of first category. Since
R = Q ∪ (R \ Q), by Remark 3.3.4 (iii), R \ Q is of second category.

Exercise 3.3.10. Prove that any nonempty open interval is of second category.

Proof. Suppose contrary that a nonempty open interval is of first category. Then each of its
translate would also be of first category. Since R is the countable union of such translates,
it follows that R is of first category, which contradicts that R is of second category.

Exercise 3.3.11. Use the Baire’s category theorem to show that the set of all rational
numbers Q is not the intersection of a countable collection of open sets. Also, show that the
set of all irrational numbers is not the union of a countable collection of closed sets.

Proof. Suppose contrary that there exists a sequence {Gn }n∈N of open sets such that Q =
\
Gn . Since Q is countable, we may write Q = {rn : n ∈ N}. Set On = Gn \ {rn } for all
n∈N
n ∈ N. Then clearly, On is open as intersection of two open sets. Since Q ⊆ Gn , Gn is dense
in R and consequently
\ On is also dense in R for all n ∈ N. By Baire’s category theorem, we
have that On is nonempty and dense in R. But this contradicts with
n∈N

\ \
On = Gn \ Q = ∅.
n∈N n∈N

To prove the second part, we again assume to the contrary that the set of all irrational
numbers is the union of a countable collection of closed sets. Then by taking the complement,
we can easily prove that Q is the intersection of a countable collection of open sets which is
a contradiction of first part.
 Chapter 4

Compact Spaces

Karl Theodor Wilhelm Weierstrass was born

on 31 October 1815, Ostenfelde, Westphalia
(now Germany) and he was died on 19 February
1897 Berlin, Germany. He is often cited as the
“father of modern analysis” for his contribution
to modern analysis and to theory of complex
functions by means of power series.

Karl Theodor
Wilhelm Weierstrass
It is well known that every sequence in a closed and bounded subset of R has a subsequence
converges in it. This result is not true in general metric spaces. The metric spaces in
which every sequence has a convergent subsequence are called sequentially compact. We
present the definition of a compact set in a metric space and its characterizations. The
Bolzano-Weierstrass property is given in the setting of a metric space. We prove that three
concepts, namely, compactness, sequentially compactness and Bolzano-Weierstrass property
are equivalent in the setting of a metric space.

109
Qamrul Hasan Ansari Metric Spaces Page 110

4.1 Definitions and Basic Concepts

Definition 4.1.1

Let X be a metric space and Λ be any index set.

S
• A collection F = {Gα }α∈Λ of subsets of X is called a cover of X if α∈Λ Gα = X,
that is, every element of X belongs to at least one member of F . If each member
of F is an open set in X, then it is called an open cover of X.

• A subcollection C of a cover F of X is called a subcover if C is itself a cover of

X. C is called a finite subcover if it consists only a finite number of members.
S
In other words, if there exist Gα1 , Gα2 , . . . , Gαn ∈ F such that nk=1 Gαk = X,
then the subcollection C = {Gα1 , Gα2 , . . . , Gαn } is called a finite subcover of X.
In this case, F is said to be reducible to a finite cover or contains a finite subcover.

Definition 4.1.2

Let X be a metric space and Y Sbe a subset of X. A collection F = {Gα }α∈Λ of subsets
of X is said to cover Y if Y ⊆ α∈Λ Gα .

Example 4.1.1

(a) Let R be the usual metric space, F = {(−n, n) : n ∈ N} and C = {(−2n, 2n) : n ∈ N}.
Then F is an open cover of R and C is a subcover of F .

   
( ( | ) ) R
−4 −3 −2 −1 0 1 2 3 4

(b) Let (0, 1) be a metric space with the usual metric. Let
     
1 1
F = 0, 1 − : n ∈ N and C = 0, 1 − :n∈N .
n+1 4(n + 1)

Then F is an open cover of (0, 1) and C is a subcover of F .

Qamrul Hasan Ansari Metric Spaces Page 111

Example 4.1.2

Let R2 be the usual metric space. Consider the family F = {S1 (x) : x ∈ Z × Z} of
open spheres S1 (x) in the plane R2 with radius 1 and center x = (m, n), where m and
n are integers. Then F is a cover of R2 , that is, every point in R2 belongs to at least
one member of F .


On the other hand, the family C = S1/2 (x) : x ∈ Z × Z of open spheres S1/2 (x) in
the plane R2 with radius 1/2 and center x = (m, n), where m and n are integers, is not
a cover of R2 because the point (1/2, 1/2) ∈ R2 does not belong to any member of C .

R y = (1/2, 1/2)

•
R
Qamrul Hasan Ansari Metric Spaces Page 112

Definition 4.1.3

A metric space X is said to be compact if every open cover of X has a finite subcover.

A subspace (Y, dY ) of a metric space (X, d) is called compact if it is compact as a metric

space in its own right.

A subset Y of a metric space X is called compact if it is compact as a metric subspace.

In other words, a nonempty subset Y of a metric space (X, d) is compact if it is a
compact metric space with the metric induced on it by d.

Example 4.1.3

(a) The usual metric space R is not compact.

Consider an open cover F = {(−n, n) : n ∈ N} of R by open intervals. Let
1 ≤ i ≤ k} be any finite subcollection of F and let n∗ = max {n1 , n2 , . . . , nk }.
{(−ni , ni ) : S
Then n∗ ∈ / ki=1 (−ni , ni ). Thus no finite subcollection of F covers R. Hence R
is not compact.

(b) Any finite subset Y of a metric space X is compact, because in this case every
open cover of Y is finite.

(c) The set of integers Z with the usual metric d(x, y) = |x − y| for all x, y ∈ Z is not
a compact metric space.


Indeed, {n} = Z ∩ n − 21 , n + 12 is an open cover of Z. But this open cover does
not have any finite subcover. Thus Z is not compact.

(d) The subspace X = {0} ∪ n1 : n ∈ N of R is compact.
Indeed, if F is a given open cover of X, then there is an element G of F containing
0. The set G contains all but finitely many of the points n1 . For each point of X
not in G, choose a member of F containing it. The collection consisting of these
members of F along with the member G, is a finite subcollection of F that covers
X.
Qamrul Hasan Ansari Metric Spaces Page 113

Example 4.1.4

Let R be the usual metric space and Y = (0, 1) be the subspace of R. Then Y is not
compact.

G4
◦ ◦
G3
◦ ◦
G2
◦ ◦
G1
◦ ◦

| | | | | | |
1 1 1 1 1
0 6 5 4 3 2 1

Consider the class of open intervals

        
1 1 1 1 1 1 1
F = ,1 , , , , , , ,... .
3 4 2 5 3 6 4
S

Then Y = ∞ 1 1
n=1 Gn , where Gn = n+2 , n for all n ∈ N. Hence F is an open cover of
Y . But F contains no finite subcover.

Indeed, let F ∗ = {(a1 , b1 ) , (a2 , b2 ) , . . . , (an , bn )} be any finite subcover of F . If ε =

min {a1 , a2 , . . . , an }, then ε > 0 and (a1 , b1 ) ∪ (a2 , b2 ) ∪ · · · ∪ (an , bn ) ⊂ (ε, 1). But (0, ε]
and (ε, 1) are disjoint. Hence F ∗ is not a cover of Y and so Y is not compact.

Example 4.1.5

Let R be the usual metric space and Y = (0, 1] be the subspace of R. Then, Y is not
1
compact because the open cover F = n
, 1 : n ∈ N contains no finite subcover of
(0, 1].

On the other hand, the interval [0, 1] is compact.

Qamrul Hasan Ansari Metric Spaces Page 114

Example 4.1.6

(a) The open sphere S1 (0) with center at origin and radius 1 is not compact.
S
Indeed, S1 (0) ⊆ ∞ n=2 S1− n 1 (0). However, no finite subcollection of the open cover
n o
S1− 1 (0) : n = 2, 3, . . . covers S1 (0).
n

(b) Let (X, d) be a discrete metric space and X be an infinite set. Then X is not
compact, but it is closed and bounded.
Consider the family F = {{x} : x ∈ X} of subsets of X which is an open cover
of X since in the discrete metric space, every subset is an open set and so the
singleton set. But there is no finite subcollection of F that covers X.
Further, any infinite subset of a discrete metric space is not compact.

Lemma 4.1.1. Let Y be a subspace of a metric space X. Then Y is compact if and

only if every cover of Y by sets open in X contains a finite subcover of Y .

Proof. Suppose that Y is compact and F = {Gα }α∈Λ is an open cover of Y , where each
Gα is open in X. Then the collection F ∗ = {Gα ∩ Y : α ∈ Λ} is an open cover of Y by
sets Gα ∩ Y open in Y . Since Y is compact, F ∗ has S a finite subcover C ∗ of Y , that is,
C ∗ =S{Gα1 ∩ Y, Gα2 ∩ Y, . . . , Gαn ∩ Y } such that Y = nk=1 (Gαk ∩ Y ). This implies that
Y ⊆ nk=1 Gαk , and hence, C = {Gα1 , Gα2 , . . . , Gαn } is a finite subcover of Y .

Conversely, assume that every cover of Y by sets open in X contains a finite subcover of Y .
Then we claim that Y is compact. Let F ∗ = {G∗α }α∈Λ be an open cover of Y , where each
G∗α is open in Y . For each α ∈ Λ, choose a set Gα open in X such that G∗α = Gα ∩ Y . Since
F ∗ is an open cover of Y , we have
!
[ [ [
∗
Y = Gα = (Gα ∩ Y ) = Gα ∩ Y.
α∈Λ α∈Λ α∈Λ
S
This implies that Y ⊆ α∈Λ Gα , and hence, F = {Gα }α∈Λ is an open cover of Y by sets Gα
open Sin X. By hypothesis, some finite subcollection {Gα1 , Gα2 , . . . , Gαn } covers Y , that is,
Y ⊆ nk=1 Gαk . Thus,
n
! n n
[ [ [
Y = Gαk ∩ Y = (Gαk ∩ Y ) = G∗αk .
k=1 k=1 k=1

Therefore, G∗α1 , G∗α2 , . . . , G∗αn is a finite subcover of Y , and hence, Y is compact.
Qamrul Hasan Ansari Metric Spaces Page 115

Theorem 4.1.1. Every closed subset of a compact metric space is compact.

Proof. Let Y be a closed subset of a compact metric S space X and let F = {Gα }α∈Λ be an
open cover of Y by sets open in X, that is, Y ⊆ α∈Λ Gα , where each Gα is open in X.
Then F ∗ = F ∪ Y c is an open cover of X, because Y c is open in X. Since X is compact,
F ∗ has a finite subcover of X, say C ∗ . If this finite subcover C ∗ contains Y c , that is, if

X = Gα1 ∪ Gα2 ∪ · · · ∪ Gαn ∪ Y c , where Gαk ∈ F , k = 1, 2, . . . , n,

then discard Y c since it covers no part of Y ; otherwise leave the subcollection alone, that is,
C ∗ = {Gα1 , Gα2 , . . . Gαn }. Then,

Y ⊆ Gα1 ∪ Gα2 ∪ · · · ∪ Gαn , where Gαk ∈ F , k = 1, 2, . . . , n.

X
Y
closed

Y c is open

Thus, C ∗ is a finite subcollection of F that covers Y . Then by Lemma 4.1.1, Y is compact.

Remark 4.1.1

(a) Any intersection of compact sets is compact.

(b) The finite union of compact sets is compact.

Definition 4.1.4

A collection C = {C1 , C2 , . . .} of subsets of a metric space X is said to have the finite

intersection property if every finite subcollection of C has nonemptyTn intersection, that
is, for every finite collection {C1 , C2 , . . . , Cn } of C, we have i=1 Ci 6= ∅.
Qamrul Hasan Ansari Metric Spaces Page 116

Example 4.1.7

(a) Consider the following collection of open intervals

       
1 1 1
C = (0, 1) , 0, , 0, , 0, ,... .
2 3 4

Then, C has the finite intersection property.

Indeed, (0, a1 )∩(0, a2 )∩(0, a3 )∩· · ·∩(0, am ) = (0, b), where b = min{a1 , a2 , . . . , am } > 0.
Therefore, any subclass C has nonempty intersection even then C itself has an empty
intersection. Thus C satisfies the finite intersection property.

(b) Consider the following collection of closed infinite intervals in R

C = {. . . , (−∞, −2], (−∞, −1], (−∞, 0], (−∞, 1], (−∞, 2], . . .} .
T
Then C has an empty intersection, that is, n∈Z Cn = ∅, where Cn = (−∞, n]. But any
finite subclass of C has a nonempty intersection. So, C satisfies the finite intersection
property.

Theorem 4.1.2. A metric space X is compact if and only if every collection of closed
sets in X having finite intersection property has nonempty intersection.

Proof. Let X be a compact metric space and C = {Cα }α∈Λ be

T any collection of closed sets in
T finite intersection property. Then we claim that α∈Λ Cα 6= ∅. Assume contrary
X having
that α∈Λ Cα = ∅. Then by De Morgan’s law, we have
!
[ \
(X \ Cα ) = X \ Cα = X.
α∈Λ α∈Λ

This implies that F = {X \ Cα }α∈Λ is an open cover of X since each X \ CαSis open. Since
n n
X is compact, F has a Tfinite subcover C ∗ = {X \ CT αi }i=1 , say. Then X = i=1 (X \ Cαi ).
n n
This implies that X \ ( i=1 Cαi ) = X, and hence, i=1 Cαi = ∅. This shows that C does
not
T have the finite intersection property, which is a contradiction of our hypothesis. Hence
α∈Λ Cα 6= ∅.

Conversely, assume that X is not compact. Then there exists an open cover F = {Gα }α∈Λ
of X which does not have any finite subcover, that is, for every finite subcollection C =
{Gα1 , Gα2 , . . . , Gαn } of F , we have
n n
! n
[ [ \
Gαi 6= X ⇒ X \ Gαi 6= ∅ ⇒ (X \ Gαi ) 6= ∅.
i=1 i=1 i=1
Qamrul Hasan Ansari Metric Spaces Page 117

For each α ∈ Λ, let Cα = X \ Gα , then Cα is closed. Since

n
\ n
\
Cαi = (X \ Gαi ) 6= ∅,
i=1 i=1

{Cα }α∈Λ is a collection of closed sets having finite intersection property. By the hypothesis,
T
α∈Λ Cα 6= ∅.
S
On the other hand, since F = {Gα }α∈Λ is a cover of X, we have α∈Λ Gα = X, and so,
!
\ \ [
Cα = (X \ Gα ) = X \ Gα = ∅,
α∈Λ α∈Λ α∈Λ

which is a contradiction. Hence, X is compact.

Corollary 4.1.1. A metric space X is compact if and only if every collection T C =

{F1 , F2 , . . .} of subsets of X having finite intersection property, the intersection F ∈C F
of their closures is nonempty.

Tn
Proof. Let {F1 , F2 , . . . , Fn } be a finite subcollection of C such that i=1 Fi 6= ∅. Since for
each i, Fi ⊆ F i , we have
\n \n
∅=6 Fi ⊆ F i.
i=1 i=1
Tn 
Therefore, i=1 F i 6= ∅, and hence, C ′ = F 1 , F 2 , . . . be a collection of closed sets having
T
finite intersection property. By Theorem 4.1.2, F ∈C F 6= ∅.

Exercise 4.1.1. Let (X, d) be a metric space such that every subset of X is compact. Prove
that X must be finite.

Proof. Let x ∈ X. Since X \ {x} is compact, it is closed and therefore, {x} is open.
Consider the open covering F = {{x} : x ∈ X} of X. Since X is compact as a subset
S there must be finite subcover C = {{x1 }, {x2 }, . . . , {xn }} of the cover F . Then
of itself,
X ⊆ ni=1 {xi } = {x1 , x2 , . . . , xn }. We conclude that X has cardinality at most n, and in
particular X is finite.

Exercise 4.1.2. (a) Prove that any intersection of compact sets is compact.

(b) Prove that any finite union of compact sets is compact.

Proof. We prove only part (b). Part (a) is left for the readers.
Qamrul Hasan Ansari Metric Spaces Page 118

m
Let {Yk }S
k=1 be a family of compact sets in a metric space X. Let F = {Gα }α∈Λ be an open
cover of m k=1 Yk by sets open in X. Since each Yk is compact, there exists finite subset Jk
of Λ such that [
Yk ⊆ Gαi , for each k = 1, 2, . . . m.
αi ∈Jk
S S Sm
Therefore, m SmYk ⊆ αi ∈J Gαi , where J = k=1 Jk is a finite subset of Λ. Hence every
k=1

Sm cover of k=1 Yk by sets open in X has a finite subcover, and thus, by Lemma 4.1.1,
open
k=1 Yk is compact.

Exercise 4.1.3. Prove that a metric space X is compact if and only if every family of closed
sets with empty intersection has a finite subfamily with empty intersection.

Proof. LetTX be a compact metric space and let {Cα }α∈Λ be a family of closed sets in X
such that α∈Λ Cα = ∅. Taking the complements, we have
" #c
\ [
X = [∅]c = Cα = Cαc ,
α∈Λ α∈Λ
S
that is, X = α∈Λ Cαc , where each Cαc is an open set in X since Cαc is the complement of
a closed set Cα . Thus F = {Cαc }α∈Λ is an open cover of X. Since X is compact, there
c c c
exists Tαn1 , Cα2 , . . . , Cαn ∈ F such that
Sn ac finite subcover C of cover F , that is, there exist C
i=1 Cαi = X. Again, by taking the complement, we have i=1 Cαi = ∅.

Conversely, assume that the hypothesis holds.

S Then we claim that X is compact. Let
F = {Gα }α∈Λ be an open cover of X. Then α∈Λ Gα = X. Taking the complements, we
have " #c
[ \
∅ = [X]c = Gα = Gcα ,
α∈Λ α∈Λ
T c
that is, α∈Λ Gα= ∅. Thus, {Gcα }α∈Λ
is a family of closedsets with empty intersection.
c c c
Tn by cthe given hypothesis, there exists a finite subfamily Gα1 , Gα2 , . . . , Gαn such that
So,
i=1 Gαi = ∅. Taking the complements, we have
" n
#c n
\ [
c
X = [∅] = Gcαi = Gαi .
i=1 i=1

Thus, {Gα1 , Gα2 , . . . , Gαn } is a finite subcover of F , and hence, X is compact.

Exercise 4.1.4. Let {Cn }∞

n=1 be a sequence of nonempty closed
T subsets of a compact metric
space X such that Cn+1 ⊂ Cn for each n ∈ N. Prove that ∞ n=1 n 6= ∅.
C

Proof. Since C1 ⊃ C2 ⊃ C3 ⊃ · · · ⊃ Cn ⊃ Cn+1 ⊃ · · · is a nested sequence of nonempty

C = {Cn }∞
closed sets in a compact metric space X. Then the collection T n=1 satisfies the finite
intersection property (prove it). By Theorem 4.1.2, we have ∞ C
n=1 n 6
= ∅.
Qamrul Hasan Ansari Metric Spaces Page 119

Exercise 4.1.5. Let X be a compact metric space and X = X × X \ {(x, x) : x ∈ X} where

the diagonal is removed. Prove that X is not compact.

Hint: Consider X = [0, 1] and check the properties of the set X × X \ {(x, x) : x ∈ X}.

4.2 Sequentially Compact Spaces

In the real line R, we have the following well known result.

Theorem 4.2.1 (Bolzano-Weierstrass Theorem). If A is a closed and bounded subset
of R, then every infinite subset of A has a limit point in A.

This result is no longer valid in the setting of general metric spaces. For example, consider
the metric space (X, d), 
where X = (0, 1] and d is the usual metric. Then X itself is closed
and bounded. Let A = 1, 12 , 31 , 41 , . . . be an infinite subset of X. We observe that A has
exactly one limit point 0 ∈
/ X. This motivates the following concept.

Definition 4.2.1

A metric space X is said to have the Bolzano-Weierstrass property if every infinite

subset of X has a limit point.

Example 4.2.1

By Bolzano-Weierstrass Theorem 4.2.1, every closed and bounded interval A = [a, b]

has the Bolzano-Weierstrass property.

Theorem 4.2.2. Every compact metric space has the Bolzano-Weierstrass property.

Proof. Let A be an infinite subset of a compact metric space X. Assume that A has no limit
point. Then each point x of X is not a limit point of A, and so, there exists rx > 0 such
that Srx (x) does not contain any point of A other than x. Therefore, Srx (x) ∩ A contains no
points of A other than x, that is,

∅, if x ∈
/ A,
Srx (x) ∩ A = (4.1)
{x}, if x ∈ A.
Since at each x ∈ X, we have an open sphere Srx (x) which contains no point of A other
than its center, the family F = {Srx (x) : x ∈ X} of these open spheres forms an open cover
of X. Since X is compact, there exists a finite subcover of F , say

Srxi (xi ) : xi ∈ X and i = 1, 2, . . . , n
Qamrul Hasan Ansari Metric Spaces Page 120

Sn
such that X = i=1 Srxi (xi ). Then,
n
!
[
A=A∩X = A∩ Srxi (xi )
i=1
n
[

= A ∩ Srxi (xi )
i=1
[n
⊆ {xi } by (4.1)
i=1
= {x1 , x2 , . . . xn } .
Consequently, A is a finite set which is a contradiction of our assumption. Hence A has a
limit point.

Definition 4.2.2

A metric space X is said to be sequentially compact if every sequence in X has a

convergent subsequence.

A subset A of a metric space X is said to be sequentially compact if every sequence in

A contains a subsequence which converges to a point in A.

Example 4.2.2

(a) Every finite subset of a metric space X is sequentially compact.

Indeed, if A is a finite subset of a metric space X and {x1 , x2 , . . . , xn , . . .} is a
sequence in A, then at least one of the elements in A, say xn0 must appear in
infinite number of times in the sequence. Hence {xn0 , xn0 , . . .} is a subsequence of
the sequence {x1 , x2 , . . . , xn , . . .} that converges to the point xn0 ∈ A.

(b) The open interval A = (0, 1) on the real line R with the usual metric is not
sequentially compact.

Consider the sequence {xn } = 12 , 31 , 14 , . . . in A. Then xn → 0, and therefore,
every subsequence will also converges to 0. But 0 ∈ / A. In other words, the
sequence {xn } in A does not contain a subsequence that converges to a point in
A, that is, A is not sequentially compact.

In order to prove the following relation

compactness ⇔ Bolzano-Weierstrass property ⇔ sequentially compactness
we first establish the equivalence between sequentially compactness and Bolzano-Weierstrass
property.
Qamrul Hasan Ansari Metric Spaces Page 121

Theorem 4.2.3. A metric space X is sequentially compact if and only if it has the
Bolzano-Weierstrass property.

Proof. Let X be a sequentially compact metric space. Then we claim that every infinite
subset of X has a limit point. Let A be an infinite subset of X. Then we can extract a
sequence {xn } with infinitely many distinct points from A. Since X is sequentially compact,
the sequence {xn } has a convergent subsequence, say, {xnk }. Let xnk → x in X. Then we
prove that x is a limit point of A.

Let ε > 0 be given. Since xnk → x as k → ∞, there exists a positive integer N such that
xnk ∈ Sε (x), for all k > N.
Since {xn } consists infinitely many distinct points from A, we have xnk 6= x for all k > N.
Thus Sε (x) contains the element xnk , for all k > N, of A different from x. Hence x is a
limit point of A. Thus every infinite subset of X has a limit point, and hence, X has the
Bolzano-Weierstrass property.

Conversely, assume that X has the Bolzano-Weierstrass property. Then every infinite subset
of X has a limit point. Let {xn } be an arbitrary sequence in X. Then we have to prove that
{xn } has a convergent subsequence. Let A be the range set of the sequence {xn }, that is,
A = {x1 , x2 , x3 , . . . , xn , . . .} is a set of points x1 , x2 , x3 , . . . , xn , . . .. Then there are two cases:

Case 1. If A is a finite set, then the sequence {xn } has a point x which appears infinitely
many times in the sequence {xn } (that is, {xn } = {x1 , x2 , . . . , xn , x, x, . . .}). Then we have
a constant subsequence {x, x, x, . . . , x . . .} of {xn } which certainly converges to x.

Case 2. If A is an infinite set, then the sequence {xn } has infinitely many distinct points.
By our assumption, the set A has a limit point, say, x ∈ X. Then S1 (x) must have infinitely
many points from A. Let us choose one such element, say, xn1 other than x. Again S1/2 (x)
must have infinitely many points from A, and in the same way, we get xn2 other than x,
n2 > n1 . Similarly, S1/3 (x) must have infinitely many points from A and we get xn3 6= x,
n3 > n2 , and so on. In this way, we obtain a subsequence {xnk } of the sequence {xn } such
that
xnk ∈ S1/k (x), for all k = 1, 2, . . . ,
that is,
1
→ 0 as k → ∞,
d (xnk , x) <
k
and hence, xnk → x. Thus the sequence {xn } has a convergent subsequence {xnk }, and
therefore, X is sequentially compact.

Corollary 4.2.1. Every compact metric space is sequentially compact.

Proof. It follows from Theorems 4.2.3 and 4.2.2.

Qamrul Hasan Ansari Metric Spaces Page 122

Theorem 4.2.4 (Heine-Borel Theorem). A subset of R is closed and bounded if and

only if it is compact.

As an application of Corollary 4.2.1, we present a general form of the Heine-Borel theorem

in the setting of metric spaces.

Theorem 4.2.5. Every compact subset of a metric space is closed and bounded.

Proof. Let (X, d) be a metric space and A be a compact subset of X. If A is finite, then
there is nothing to prove because every finite set is closed and bounded. So, we assume that
A is an infinite set.

Let x ∈ A be arbitrary. Then by Theorem 2.1.3, there exists a sequence {xn } ⊂ A such that
xn → x. Since A is compact, by Corollary 4.2.1, it is sequentially compact, and therefore,
{xn } has a subsequence that converges in A. Let {xnk } be a convergent subsequence of {xn }.
Then xnk → x ∈ A, and hence, A ⊂ A. Therefore, A = A, and thus, A is a closed set.

Further, let A be unbounded. Then we can find pairs of points of A at arbitrary large
distances apart, that is, there would exist x and y in A such that d(x, y) > M for some given
M > 0. Consider the family F = {S1 (x) : x ∈ A} of open spheres with centered at a point
of A and radius 1. Then clearly, [
A⊆ S1 (x).
x∈A

This implies that F is an open cover of A. Since A is compact, F has a finite subcover
C = {S1 (xi ) : i = 1, 2, . . . , n}, say. Then,
n
[
A⊆ S1 (xi ).
i=1

Let k = max {d (xi , xj ) : i, j = 1, 2, . . . , n, i 6= j}. Also, let M = k + 2, then

d(x, y) > M = k + 2. (4.1)

Since x and y are in A, there exist xi and xj such that x ∈ S1 (xi ) and y ∈ S1 (xj ). So,

d(x, y) ≤ d(x, xi ) + d(xi , xj ) + d(xj , y) < k + 2

which contradicts (4.1). Hence A is bounded.

Qamrul Hasan Ansari Metric Spaces Page 123

Remark 4.2.1

The converse of the above theorem need not be true. For example, consider an infinite
subset A of a discrete metric space X. Then A is closed and bounded because the open
sphere S 1 (x) is the set {x} alone and d(x, y) ≤ 1 for all x, y ∈ A. But the open cover
2
{{x} : x ∈ A} of A has no finite subcover. Thus A is not compact.
√ √

Also, the set − 2, 2 ∩ Q is closed and bounded in Q, but not compact.

In fact, a subset of Rn with the usual metric is compact if and only if it is closed and
bounded.
Qamrul Hasan Ansari Metric Spaces Page 124

4.3 Totally Bounded Spaces

Definition 4.3.1

Let (X, d) be a metric

S space and ε > 0 be given. A subset A of X is called an ε-net if A
is finite and X = x∈A Sε (x), that is, if A is finite and its points are scattered through
X in such a way that each point of X is distant by less than ε from at least one point
of A.

In other words, a finite subset A = {x1 , x2 , . . . , xn } of X is an ε-net for X if for every

point y ∈ X, there exists an xi0 ∈ A such that d (y, xi0 ) < ε.

Example 4.3.1

Let X = {(x, y) ∈ R × R : x2 + y 2 < 4}, that is, X is the open sphere centered at the
origin and radius 2. If ε = 32 , then the set

A = {(1, −1), (1, 0), (1, 1), (0, −1), (0, 0), (0, 1), (−1, −1), (−1, 0), (−1, 1)}

is an ε-net for X.

1
• • • y = (1/2, 1/2)
−2 −1 •
1 2
• • • R

• −1• •

−2

On the other

hand, if ε = 1/2, then A is not an ε-net for X. For example, the point
y = 21 , 12 belongs to X but the distance between y and any point in A is greater than
1
2
.
Qamrul Hasan Ansari Metric Spaces Page 125

Definition 4.3.2

A metric space (X, d) is said to be totally bounded if it has an ε-net for each ε > 0.

Remark 4.3.1

Every totally bounded metric space X is bounded but converse is not true in general.

X is totally bounded, it has an ε-net A = {x1 , x2 , . . . , xn } for each ε > 0. Then

Since S
X = ni=1 Sε (xi ). Since finite union of bounded sets is bounded, it follows that X is
bounded.

Example 4.3.2

Under the usual metric d(x, y) = |x − y|, the real line R is neither bounded nor totally
bounded. However, a bounded subset of the real line with the usual metric is totally
bounded.

Under the metric d∗ (x, y) = min {|x − y|, 1}, the real line R is bounded but not totally
bounded.

Example 4.3.3

An infinite set X with the discrete metric is bounded, but not totally bounded.

Theorem 4.3.1. Every sequentially compact metric space is totally bounded and com-
plete.

Proof. Let (X, d) be a sequentially compact metric space. Assume to the contrary that X
is not totally bounded. Then there exists an ε > 0 such that there does not exist any ε-net.
Choose a point x1 ∈ X and form the open sphere Sε (x1 ). Then, clearly Sε (x1 ) 6= X for
otherwise {x1 } would be an ε-net for X. Let x2 ∈ X be an element such that x2 ∈ / Sε (x1 ),
that is, d(x1 , x2 ) ≥ ε. Form the open sphere Sε (x2 ) and the set Sε (x1 ) ∪ Sε (x2 ). Then,
Sε (x1 ) ∪ Sε (x2 ) 6= X since in other case {x1 , x2 } would be an ε-net for X. Let x3 ∈ X be
such that x3 ∈ / Sε (x1 ) ∪ Sε (x2 ), that is,

d(x1 , x3 ) ≥ ε and d(x2 , x3 ) ≥ ε.

Form an open sphere Sε (x3 ) and a set Sε (x1 )∪Sε (x2 )∪Sε (x3 ). Then Sε (x1 )∪Sε (x2 )∪Sε (x3 ) 6=
X, otherwise the set {x1 , x2 , x3 } would be an ε-net for X.
Qamrul Hasan Ansari Metric Spaces Page 126

Continuing in this way, we obtain a sequence {xn } of elements of X such that

n−1
[
xn ∈
/ Sε (xi ), for n = 2, 3, . . . ,
i=1

that is,
d(xi , xn ) ≥ ε, for i = 1, 2, . . . , n − 1 and n = 2, 3, . . . with n 6= i.
Consequently,
d(xn , xm ) ≥ ε, for all n, m and n 6= m.
Thus {xn } is a sequence with no Cauchy subsequence, and hence, {xn } cannot have any
convergent subsequence, contradicting our hypothesis that X is sequentially compact. Hence
X is totally bounded.

Now, we prove that X is complete. For that, we have to show that every Cauchy sequence
converges to a point of X.

Let {xn } be a Cauchy sequence. Then for each ε > 0, there exists a positive integer N such
that
d (xm , xn ) < ε, whenever m > n > N.
Since X is sequentially compact, every sequence, in particular, a Cauchy sequence, has
a convergent subsequence. Therefore, this Cauchy sequence {xn } contains a subsequence
{xkn }, say, which converges to a point x ∈ X, that is,

lim d (xkn , x) = 0.
n→∞

Since {kn } is an increasing sequence of integers, km ≥ m. We now have

0 ≤ d (xn , x) ≤ d (xn , xkm ) + d (xkm , x) < ε + d (xkm , x) ,

whenever m > n > N. Taking m → ∞, then we have

0 ≤ d(xn , x) < ε, whenever n ≥ N.

Therefore, the Cauchy sequence {xn } converges to x ∈ X.

Remark 4.3.2

A complete metric space may not be sequentially compact. For example, consider the
metric subspace A = (−1, 1) of R under the usual metric d(x, y) = |x−y| for all x, y ∈ A.
Then A = (−1, 1) is totally bounded but not sequentially compact.
Qamrul Hasan Ansari Metric Spaces Page 127

Remark 4.3.3

Every sequentially compact metric space may not be a complete space. For example,
as we have seen in Example 2.3.7 that the space C[a, b] of all continuous real-valued
functions defined on [a, b] with the metric

d∞ (f, g) = max |f (t) − g(t)|

t∈[a,b]

is a complete metric space, but it is not sequentially compact. Needed, for each n ∈ N,
define dn = a + (b − a)2−n . Consider the piecewise linear function

1 − 2n (t − a), if a ≤ t ≤ dn ,
fn (t) =
0, if dn ≤ t ≤ b.

By construction, fn (t) is continuous, fn (dn ) = 0, and if m < n, then fm (dn ) ≥ 12 . Thus

if m < n, we have
1
d∞ (fn , fm ) = max |fn (t) − fm (t)| ≥ |fn (dn ) − fm (dn )| ≥ .
t∈[a,b] 2

Therefore, if m < n, d∞ (fn , fm ) ≥ 12 . Thus we conclude that the sequence {fn } defined
as above does not have a convergent subsequence. Hence C[a, b] with respective to the
metric d∞ not sequentially compact.

But every totally bounded complete metric space is sequentially compact.

Theorem 4.3.2. Every totally bounded complete metric space is sequentially compact.

Proof. Let (X, d) be a totally bounded and complete metric space. Let {xn } be any sequence
in X. Since X is totally bounded, it can be covered by a finite number of open spheres of
radius 1, and at least one of these open spheres will contain an infinite number of members
of the sequence {xn }. Let N1 be a such open sphere and let {xn1 } be a subsequence of the
sequence {xn } in N1 .

Since X is covered by a finite number of open spheres and N1 is contained in X, therefore, N1

can also be covered by a finite number of open spheres of radius 12 . By the same argument,
there would be an open sphere N2 of radius 12 which contains a subsequence {xn2 } of {xn1 }.
Since N2 ⊂ N1 , it follows that xn2 ∈ N1 . Continuing in this way, for any positive integer
k, we obtain an opensphere Nk of radius k1 which is contained in Nk−1 and contains a
subsequence {xnk } of xnk−1 , that is, the diameter of Nk , δ(Nk ) < k2 and xnk ∈ Nk , where
nk > nk−1 . Since xnk , xnk+1 , . . . all lie in Nk and the diameter δ(Nk ) < k2 , it follows that

2
d xnk , xnk+m < .
k
Qamrul Hasan Ansari Metric Spaces Page 128

For any ε > 0, choose k > 2ε . Then,

d xnk , xnk+m < ε,

and hence, {xnk } is a Cauchy subsequence of the sequence {xn }. Since X is complete, this
subsequence converges to a point of X. Hence X is sequentially compact.

Corollary 4.3.1. A closed subspace of a complete metric space is sequentially compact

if and only if it is totally bounded.

Theorem 4.3.3. Every totally bounded metric space is separable.

1
Proof. Let (X, d) be a totally bounded metric space. Then for each S∞n ∈ N, X has a n -net
which is denoted by An . Since each An is a finite set, the set A = n=1 An is countable. We
further prove that A = X, that is, if x ∈ X then we claim that x ∈ A.
1
Let x ∈ X and r > 0 be given. Choose a positive integer n such that n
< r. Let An =
{a1 , a2 , . . . , ak } since each An is finite. Then,
k
[
X= S 1 (aj ),
n
j=1

and so, x ∈ S 1 (aj ) for some j = 1, 2, . . . , k. Therefore,

n

1
d(x, aj ) < < r,
n
and thus, aj ∈ Sr (x), that is, Sr (x) ∩ A 6= ∅ because aj ∈ An ⊂ A. This implies that x ∈ A
and hence A = X.

Remark 4.3.4

The converse of above theorem does not hold, that is, a separable metric space may
not be totally bounded. For example, consider the set of integers Z which is countable
dense subset of the usual metric space R, and hence, R is separable. But R with the
usual metric is not totally bounded.
Qamrul Hasan Ansari Metric Spaces Page 129

Definition 4.3.3

Let F = {Gα }α∈Λ be an open cover of a metric space X. A real number λ > 0 is called
a Lebesgue number for the given open cover F if for each subset of X whose diameter is
less than λ is contained in at least one Gα , that is, if every subset A ⊂ X with δ(A) < λ
then A ⊂ Gα for at least one α ∈ Λ.

Theorem 4.3.4 (Lebesgue’s Covering Lemma). In a sequentially compact metric space,

every open cover has a Lebesgue number.

Proof. Let (X, d) be a sequentially compact metric space and let F = {Gα }α∈Λ be an open
cover of X. Assume that F does not have any Lebesgue number. Then for each n ∈ N,
there exists a subset Bn of X such that

1
0 < δ(Bn ) < and Bn 6⊂ Gα , for all α ∈ Λ. (4.2)
n

For each n ∈ N, choose a point xn ∈ Bn . Then {xn } is a sequence of points of X. Since X

is sequentially compact, {xn } has a subsequence {xnk } converges to a point x ∈ X.

Since x ∈ X and F is an open cover of X, x ∈ Gα0 for some α0 ∈ Λ. But Gα0 being an
open set containing x, there exists an open sphere Sε (x) such that x ∈ Sε (x) ⊆ Gα0 . Since
{xnk } converges to x, there exists a positive integer N such that for all k ≥ N, we have

ε
d (xnk , x) < .
2

1
Let n0 ≥ N be such that n0
< 2ε . Then from the above inequality, we have

ε
d (xnk , x) < , that is, xnk ∈ S 2ε (x), for all nk ≥ n0 .
2

In particular, xn0 ∈ S 2ε (x). In view of (4.2), we have

1 ε
xn0 ∈ Bn0 and 0 < δ (Bn0 ) < < .
n0 2
Qamrul Hasan Ansari Metric Spaces Page 130

xn0 Bn0

Sε (x) ε x •
2
• Gα0

Since xn0 belongs to Bn0 as well as to S 2ε (x), Bn0 ∩ S 2ε (x) 6= ∅. This implies that Bn0 will
also intersect with Sε (x). Since δ (Bn0 ) < 2ε and Bn0 intersects with Sε (x), we have
Bn0 ⊂ Sε (x) ⊆ Gα0 .
But this contradicts with the fact that Bn0 6⊂ Gα for all α ∈ Λ. Hence, F has a Lebesgue
number.

Theorem 4.3.5. Every sequentially compact metric space is compact.

Proof. Let X be a sequentially compact metric space. Then by Theorem 4.3.1, it is totally
bounded. Then for a given ε > 0, X has an ε-net A, say. Then A = {x1 , x2 , . . . , xn } is finite
subset of X and n
[
X= Sε (xk ).
k=1
Let F = {Gα }α∈Λ be an open cover of X. Since X is sequentially compact, by Lebesgue’s
covering lemma, this open cover F has a Lebesgue number λ, say. We put ε = λ3 . For each
k = 1, 2, . . . , n, we have
2λ
δ (Sε (xk )) < 2ε = < λ.
3
By the definition of a Lebesgue number, for each k = 1, 2, . . . , n, we can find an open set
Gαk ∈ F such that Sε (xk ) ⊆ Gαk . This implies that
n
[ n
[
Sε (xk ) ⊆ Gαk .
k=1 k=1
n
[ n
[
Since X = Sε (xk ), we have X = Gαk . Hence, the class {Gα1 , Gα2 , . . . , Gαn } is a finite
k=1 k=1
subcover of X, and therefore, X is compact.
Qamrul Hasan Ansari Metric Spaces Page 131

Corollary 4.3.2. A closed subspace of a complete metric space is compact if and only
if it is totally bounded.

Finally, we have the following relations.

Compact

By
3.
4.

Th
em

eo
r

r
eo

em
Th

4.
2.
By

2
Sequentially Compact Bolzano-Weierstrass Property
By Theorem 4.2.3

Exercise 4.3.1. Let A be a compact subset of a metric space X. Show that the derived set
A′ of A is compact.

Exercise 4.3.2. Which of the following sets are compact? Justify your answer.

(a) A = [0, 2) ⊂ R.

(b) A = [0, 1] ∪ [2, 3] ⊂ R.

(c) A = {x ∈ R : x ≥ 0} ⊂ R.

(d) A = Qc ∩ [0, 1] = {x ∈ R : 0 ≤ x ≤ 1 and x is irrational}.


(e) A = 1, 21 , 13 , . . . , n1 , . . . ∪ {0}.

(f) A = {(x, y) ∈ R2 : y = 0}.

(g) A = {(x, y) ∈ R2 : x2 + y 2 ≤ 4} ∪ {(1, 2)}.

(h) A = {(x, y) ∈ R2 : 2x2 + y 2 = 1}.

(i) A = {(x, y) ∈ R2 : 0 ≤ x ≤ 1, 0 ≤ y ≤ 1}.

Solution. (b) The set A = [0, 1] ∪ [2, 3] ⊂ R is compact because it is closed and bounded
(Heine-Borel Theorem).

(c) The set A = {x ∈ R : x ≥ 0} ⊂ R is not compact because it is unbounded.

Qamrul Hasan Ansari Metric Spaces Page 132

(d) The set A = Qc ∩ [0, 1] = {x ∈ R : 0 ≤ x ≤ 1 and x is irrational} is not compact

because x = 41 ∈
/ A, but every interval around it contains irrational numbers which are in A.
c
Therefore, A is not open so A is not closed. Therefore, A can not be compact.

(e) Let {Gα } be an arbitrary open cover of A. Then the point 0 lies in one of these open
sets. Assume that 0 ∈ Gα0 for some α0 . Since Gα0 is open and n1 → 0, there is an
N such that N1 , N 1+1 , . . . all lie in Gα0 . Since {Gα } is an open cover of A, there must exist
Gα1 , Gα2 , . . . , GαN (not necessarily all distinct from one another, but we can rename and give
multiple names to the sets as is needed) such that 1, 1/2, . . . , 1/N ∈ Gα1 ∪ Gα2 ∪ · · · ∪ GαN .
Then {Gα0 , Gα1 , Gα2 , . . . , GαN } is a finite subcover of A. Therefore, every open cover has a
finite subcover and so A is compact.

Exercise 4.3.3. Which of the following sets are compact? Justify your answer.

(a) A = {(x, y) ∈ R2 : x2 − y 2 = 1}.

(b) A = {(x, y) ∈ R2 : y 2 = x}.

n 2 2
o
(c) A = (x, y) ∈ R2 : xa2 + yb2 = 1 for all a, b ∈ R.

(d) A = {x = (x1 , x2 , . . . , xn ∈ Rn : x21 + 2x22 + · · · + nx2n ≤ (n + 1)2 }.

Exercise 4.3.4. Let A and B be subsets of a metric space X such that A is closed and B
is compact. Prove that A ∩ B is compact.

Proof. Let {Gα }α∈Λ be an open cover of A ∩ B by sets open in X. Let Gβ = X \ A.

Then Gβ is open in X since A is closed. Let us denote by J = Λ ∪ {β}. We claim that
{Gα }α∈J is anS open cover of B. For that, let b ∈ B. Then either b ∈ Gβ (if b ∈ / A), or else
b ∈ A ∩ B ⊆ α∈Λ Gα . This implies S that b ∈ Gα for some α ∈ Λ. In either case b ∈ Gα
for some α ∈ J. Therefore, B ⊆ α∈J Gα and so {Gα }α∈J is an open cover of B. Since B
is compact, thisSopen cover has a finite subcover, that is, there exists a finite subset L of J
such that B ⊆ αi ∈LSGαi . The set Lβ = L − {β} is a finite subset of J − {β} = Λ.S Now we
show that A ∩ B ⊆ αi ∈Lβ Gαi . For this, suppose that b ∈ A ∩ B. Then b ∈ B ⊆ αi ∈L Gαi
and so b ∈ Gαi for some αi ∈ L. But b ∈ A and so b ∈ / X \ A = Gβ . Thus b ∈ Gαi for some
αi ∈ Lβ . Thus the open cover {Gα }α∈Λ of A ∩ B has a finite subcover, namely {Gαi }αi ∈Lβ .
Hence A ∩ B is compact.

4.3.5. Let {Aα }α∈Λ be anySfamily of compact subsets of a metric space X. Show
Exercise T
that B = α∈Λ Aα is compact, while α∈Λ Aα is not necessarily compact.

T
Proof. Choose a fixed α0 ∈ Λ and put A = α6=α0 Aα and take B = Aα0 . Since every
compact set is closed, each Aα is closed and so A is closed being an intersection\
of closed
sets. Since B is compact, by Exercise 4.3.4, A ∩ B is compact and so is A ∩ B = Aα .
α∈Λ
Qamrul Hasan Ansari Metric Spaces Page 133

Arbitrary union of compact sets need not be compact. For example, consider the sets
S∞ usual metric space R. Then each An is
An = [−n, n] for all positive integer n, in the
compact since it is closed and bounded. But n=1 An = R which is not compact.
Exercise 4.3.6. Given an example of a metric space X (other than R) having Bolzano-
Weierstrass property.
Exercise 4.3.7. Show that no finite subcollection of the open cover F = {S1− 1 (0) : n =
n
2, 3, . . .} of S1 (0) covers S1 (0).
Exercise 4.3.8. Show that the closed unit sphere S1 [0] in the space
( ∞
)
X
ℓ2 = {xn } ⊆ R or C : |xn |2 < ∞
n=1
P∞ 1/2
with the metric d(x, y) = ( n=1 |xn − yn |2 ) for all x = {x1 , x2 , . . .}, y = {y1, y2 , . . .} in ℓ2 ,
is not compact.

Proof. Consider the sequence {en } in ℓ2 defined by en (k) = δnk for all n.k ∈ N, where δnk
denotes the Kronecker delta symbol defined by δnk = 1 if n = k and 0 otherwise. Then {en }
is a sequence of all sequences whose terms are zero except the nth term.
√ n Clearly, en ∈ So1 [0]
and d(em , en ) = 2 if m 6= n. We claim that the open cover F = S 1 (x) : x ∈ S1 [0] of
2
n o
S1 [0] admits no finite subcover. For, if S 1 (xj ) : 1 ≤ j ≤ n is a finite subcover, then there
2
exists j (1 ≤ j ≤ n) and m 6= n such that em , en ∈ S 1 (xj ). In particular, d(em , en ) ≤
2
d(em , xj ) + d(xj , en ) < 1, a contradiction.
Exercise 4.3.9. Prove that the real line R under the metric d∗ (x, y) = min {|x − y|, 1} for
all x, y ∈ R is bounded but not totally bounded.
Exercise 4.3.10. Let R be
 the usual
 metric space and Y = (0, 1] be a subspace of R. Prove
that the open cover F = n1 , 1 : n ∈ N of Y contains no finite subcover.
Exercise 4.3.11. Prove that a subset A of the usual metric space Rn is compact if and only
if it is closed and bounded.

Proof. Let A be a compact subset of Rn . Then by Theorem 4.2.5, it is closed and bounded.

Conversely, assume that A is closed and bounded. Fix ε > 0. The set
  
εk1 εk2 εkn
C̃ = , ,..., : ki ∈ Z, 1 ≤ i ≤ n
n n n
is an ε-net for Rn . Since A is bounded, there exists r > 0 such that A ⊂ Sr (0). Then the set
C = C̃ ∩ Sr (0)
is a finite ε-net for A. So, A is totally bounded. Then by Theorem 4.3.2, A is sequentially
compact and Theorem 4.3.5 implies that A is compact.
Qamrul Hasan Ansari Metric Spaces Page 134

Exercise 4.3.12. Let X be a metric space and {xn } be a sequence in X such that xn → x
as n → ∞. Show that the subset {x} ∪ {xn : n ∈ N} of X is compact.

Proof. Let F be an open cover for {x} ∪ {xn : n ∈ N}. Then there exists an open set G ∈ F
such that x ∈ G. Since xn → x and G is open, there exists a positive integer N such that
G contains all points of the set {xn : n ∈ N} for all n > N. For each i = 1, 2, . . . , N, let
Gi ∈ F such that xi ∈ Gi . Then G ∪ G1 ∪ · · · ∪ GN is finite subcover of the cover F . Hence
the set {x} ∪ {xn : n ∈ N} of X is compact.

Exercise 4.3.13. Let A be a compact subset of a metric space (X, d). Show that for any
subset B of X, there exists a point x ∈ A such that ρ(x, B) = ρ(A, B).

Proof. Let ρ(A, B) = ε. Then ε = ρ(A, B) = inf{d(x, y) : x ∈ A, y ∈ B} ≤ d(x, y) for all

x ∈ A and y ∈ B. Therefore, for any positive integer N, there exist xn ∈ A, yn ∈ B such
that
1
ε ≤ d (xn , yn ) < ε + , for all n > N.
n
Since A is compact, it is sequentially compact, and therefore, the sequence {xn } ⊆ A has
a subsequence {xnk } converges to a point x ∈ A. We claim that ρ(x, B) = ρ(A, B) = ε.
Suppose contrary that ρ(x, B) > ε. Then for some δ > 0, ε + δ = ρ(x, B) = inf d(x, y), and
y∈B
so, d(x, yn ) ≥ ε + δ for all yn ∈ B. Since xnk → x as k → ∞, there exists a positive integer
N such that
δ
d (xnk , x) < , for all k > N,
2
and
1 δ
d (xnk , ynk ) < ε + < ε + , for all k > N.
nk 2
Then,

δ δ
d (x, xnk ) + d xnk ,ynk < + ε + = ε + δ = ρ(A, B) ≤ d (x, ynk ) , for all nk > N.
2 2
But this contradicts the triangle inequality. Hence ρ(x, B) = ρ(A, B).

Exercise 4.3.14. Let A be a compact subset of a metric space (X, d) and let B be a closed
subset of X such that A ∩ B = ∅. Show that ρ(A, B) > 0.

Proof. Since ρ(A, B) = inf{d(x, y) : x ∈ A, y ∈ B} ≥ 0, it is sufficient to show that

ρ(A, B) 6= 0. Suppose contrary that ρ(A, B) = 0. Then by Exercise 4.3.13, there exists
x ∈ A such that ρ(x, B) = ρ(A, B) = 0. By Theorem 1.4.1, B = B = {x ∈ X : ρ(x, B) = 0}.
Therefore, x ∈ B contradicting to our assumption that A ∩ B = ∅.

Exercise 4.3.15. Let X be a metric space. Prove that X is complete if and only if every
infinite totally bounded subset of X has a limit point
Qamrul Hasan Ansari Metric Spaces Page 135

Exercise 4.3.16. Let X be a metric space. Prove that X is totally bounded if and only if
every sequence in X has a Cauchy subsequence.

Hint: See the proof of Theorem 4.3.2.

 Chapter 5

Continuous Functions

Bernard Placidus Johann Nepomuk Bolzano was

born on October 5, 1781 in Prague, Czech Re-
public and was died on December 18, 1848. He
was a mathematician, theologian, philosopher,
and logician. His logical analysis of mathemat-
ical problems made him a pioneer in geometry
and calculus. The definition of continuity in
terms of the epsilon–delta was first given by
Bolzano in 1817. Augustin-Louis Cauchy defined
continuity of a function y = f (x) as follows: an
infinitely small increment α of the independent
variable x always produces an infinitely small
change f (x + α) − f (x) of the dependent variable
y.
Bernard Placidus
Johann Nepomuk Bolzano
This chapter deals with the concepts of continuity and uniform continuity of functions defined
on a metric space. Several characterizations and properties are presented. The concepts of
homeomorphism and equivalent metrics are given. The uniform convergence of the sequences
of functions defined on a metric space X, equicontinuity of the space C(X, Rn ) and Arzelà-
Ascoli theorem are discussed.

136
Qamrul Hasan Ansari Metric Spaces Page 137

5.1 Definition and Characterizations

Definition 5.1.1

Let (X, dX ) and (Y, dY ) be metric spaces. A function f : X → Y is said to be continuous

at a point x0 ∈ X if for every ε > 0, there exists a δ > 0 such that for all x ∈ X,

dX (x, x0 ) < δ implies dY (f (x), f (x0 )) < ε,

that is,
x ∈ Sδ (x0 ) implies f (x) ∈ Sε (f (x0 )) .
In other words, f is continuous at a point x0 ∈ X if for every ε > 0, there exists a δ > 0
such that
f (Sδ (x0 )) ⊆ Sε (f (x0 )) .

The function f is said to be continuous on X if it is continuous at every point of X.

f
>
f (x)
x
δ > ε
f
x0 f (x0 )

Sδ (x0 )
Sε (f (x0 ))

Example 5.1.1

(a) Let (X, d) be a metric space. The identity function I : X → X is continuous on

X.

(b) Let X = R be the set of all real numbers with the usual metric d(x, y) = |x − y|.
Then every constant function defined on X is continuous.

(c) Let X be a discrete metric space and Y be a metric space. Then every function
f : X → Y is continuous on X.
Qamrul Hasan Ansari Metric Spaces Page 138

Example 5.1.2

Let X = Y = C[0, 1] with the uniform metric d∞ (f, g) = sup |f (t) − g(t)|. The
t∈[0,1]
mapping φ : X → Y defined by
Z t
(φ(f )) (t) = f (s)ds, for all f ∈ C[0, 1],
0

is continuous. In fact,
Z t Z t
|(φ(f )) (t) − (φ(g)) (t)| = f (s)ds − g(s)ds
0 0
Z t
≤ |f (s) − g(s)| ds
0
≤ d∞ (f, g).

Hence,
d∞ (φ(f ), φ(g)) ≤ d∞ (f, g), for all f, g ∈ C[0, 1].
So, the assertion follows if we choose δ = ε.

Example 5.1.3

Let A be a nonempty subset of a metric space (X, d). Then the mapping x 7→ ρ(x, A),
defined by ρ(x, A) := inf{d(x, a) : a ∈ A} for all x ∈ X, is continuous.

Indeed, let x, y ∈ X and a ∈ A be arbitrary. From the triangle inequality

d(a, x) ≤ d(a, y) + d(y, x),

we have
d(a, y) ≥ d(a, x) − d(y, x),
that is,
d(a, y) ≥ ρ(x, A) − d(y, x), (5.1)
since d(a, x) ≥ inf{d(b, x) : b ∈ A} := ρ(x, A). The inequality (5.1) says that ρ(x, A) −
d(y, x) is a lower bound for the set {d(a, y) : a ∈ A}. Hence the greatest lower bound of
this set, namely, inf{d(a, y) : a ∈ A} is greater than or equal to this lower bound, that
is,
ρ(y, A) ≥ ρ(x, A) − d(y, x).
Qamrul Hasan Ansari Metric Spaces Page 139

Therefore, ρ(y, A) − ρ(x, A) ≥ −d(y, x), that is, ρ(x, A) − ρ(y, A) ≤ d(y, x). By inter-
changing x and y in this inequality, we see that

−d(x, y) ≤ ρ(x, A) − ρ(y, A) ≤ d(x, y),

that is,
|ρ(x, A) − ρ(y, A)| ≤ d(x, y).
Hence, the continuity of x 7→ ρ(x, A) follows.

Definition 5.1.2

Let X be a set. A real-valued function f : X → R is said to be bounded if there exists

a constant M ∈ R such that |f (x)| ≤ M for all x ∈ X.

Remark 5.1.1

A continuous function is not necessarily bounded. For example, consider the function
f (x) = 1/x defined on the set X = (0, ∞). Then clearly f is continuous, but not
bounded.

Theorem 5.1.1. Let (X, dX ) and (Y, dY ) be metric spaces. A function f : X → Y is

continuous at a point x0 ∈ X if and only if for every sequence {xn } in X,

xn → x0 implies f (xn ) → f (x0 ).

Proof. Let f be continuous at a point x0 ∈ X. Then for any given ε > 0, there exists a
δ > 0 such that for all x ∈ X,

dX (x, x0 ) < δ implies dY (f (x), f (x0 )) < ε.

Let {xn } be a sequence in X such that xn → x0 as n → ∞. Then there exists a positive

integer N such that
dX (xn , x0 ) < δ, for all n > N.
Hence for all n > N, we have
dY (f (xn ), f (x0 )) < ε,
and therefore, f (xn ) → f (x0 ).

Conversely, assume that for every sequence {xn } in X,

xn → x0 implies f (xn ) → f (x0 ).

Qamrul Hasan Ansari Metric Spaces Page 140

Suppose that f is not continuous at x0 . Then there exists an ε > 0 such that for every δ > 0,
there is an x 6= x0 satisfying
dX (x, x0 ) < δ but dY (f (x), f (x0 )) ≥ ε.
In particular, for δ = n1 , there is an xn satisfying
1
dX (xn , x0 ) < but dY (f (xn ), f (x0 )) ≥ ε.
n
Then clearly xn → x0 , but {f (xn )} does not converge to f (x0 ). This contradicts to our
hypothesis that f (xn ) → f (x0 ). Hence, f is continuous at x0 .

Theorem 5.1.2. Let (X, dX ) and (Y, dY ) be metric spaces. A function f : X → Y is

continuous on X if and only if for each x ∈ X and for every sequence {xn } in X, we
have
xn → x implies f (xn ) → f (x).

Remark 5.1.2

If f : X → Y is continuous and f (xn ) → f (x) for any sequence {xn } in X, then the
assertion xn → x may not be true.

For example, let X = R be a metric space with the usual metric d(x, y) = |x − y| and
f : X → X be a function defined by

f (x) = x2 , for all x ∈ X.

Consider a sequence {xn } in X, where xn = (−1)n for all n ∈ N. Then,

f (xn ) = 1 → f (1) as n → ∞,

but the sequence xn does not converge to 1 as n → ∞.

Theorem 5.1.3. Let X and Y be metric spaces. A function f : X → Y is continuous

on X if and only if f −1 (G) is open in X whenever G is open in Y .

Proof. Assume that f is continuous on X and G is an open set in Y . Then we prove that
f −1 (G) is open in X. If f −1 (G) = ∅, then we are done. So, we assume that f −1 (G) 6= ∅.
Let x ∈ f −1 (G), then f (x) ∈ G. Since G is open, Sε (f (x)) ⊆ G for some ε > 0. By the
continuity of f , there exists a δ > 0 such that
f (Sδ (x)) ⊆ Sε (f (x)) ,
and since Sε (f (x)) ⊆ G, it follows that f (Sδ (x)) ⊆ G. Thus, Sδ (x) ⊆ f −1 (G), and hence,
f −1 (G) is open.
Qamrul Hasan Ansari Metric Spaces Page 141

f −1 (G) G

> ε
δ f
x f (x)
Sδ (x) Sε (f (x))

Space X Space Y

Conversely, assume that f −1 (G) is open in X whenever G is open in Y . Let x ∈ X be

arbitrary and ε > 0 be given. Then f (x) ∈ Y and Sε (f (x)) is an open set. Therefore, by
assumption f −1 (Sε (f (x))) is open in X and x ∈ f −1 (Sε (f (x))). Consequently, there exists
a δ > 0 such that Sδ (x) ⊆ f −1 (Sε (f (x))), and thus, f (Sδ (x)) ⊆ Sε (f (x)). Hence, f is
continuous at x. Since x ∈ X was arbitrary, f is continuous on X.

Theorem 5.1.4. Let X and Y be metric spaces. A function f : X → Y is continuous

on X if and only if f −1 (F ) is closed in X whenever F is closed in Y .

Proof. Let f be a continuous function and F be a closed in Y . Then Y \ F is open in Y ,

and therefore, f −1 (Y \ F ) is open in X. Since


f −1 (F ) = X \ f −1 (Y \ F )

and f −1 (Y \ F ) is open, it follows that f −1 (F ) is closed.

Conversely, assume that f −1 (F ) is closed in X whenever F is closed in Y . Then we show

that f is continuous. Let G be an open subset of Y . Then Y \ G is closed in Y , and by
hypothesis, f −1 (Y \ G) is closed in X. Since


f −1 (G) = X \ f −1 (Y \ G)

and f −1 (Y \ G) is closed , we have f −1 (G) is open. Hence, f is continuous on X.

Qamrul Hasan Ansari Metric Spaces Page 142

Remark 5.1.3

If f is a continuous function from a metric space X to another metric space Y , then

the image f (G) of an open set G in X need not be open in Y and the image f (F ) of a
closed set F in X need not be closed in Y .

For example, consider the function f : R → R defined as f (x) = x2 . Then, of course, f

is continuous on R. Let G = (−1, 1) be an open set in R but f (G) = [0, 1) is not open
in R.

Consider another function f : [1, +∞) → R defined by f (x) = x1 . Then f is continuous

on [1, +∞). Let A = [1, +∞), then A is closed as a whole space with the usual metric
but f (A) = (0, 1] is not closed in R.

Theorem 5.1.5. Let
X and Y be metric spaces. A function f : X → Y is continuous

if and only if f A ⊆ f (A) for every subset A of X.

 
−1
Proof. Let f be a continuous function. Then, f f (A) is closed in X, since f (A) is closed
in Y . Now we have
   
f (A) ⊆ f (A) ⇒ A ⊆ f −1 f (A) ⇒ A ⊆ f −1 f (A) ,
   

and thus, A ⊆ f −1 f (A) because f −1 f (A) is closed. Hence f A ⊆ f (A).


Conversely, let f A ⊆ f (A) for every subset A of X. We prove that f is continuous. Let
F be any closed set in Y . Then F = F . Now we have
 
f f −1 (F ) ⊆ f (f −1 (F )) = F = F.

This implies that f −1 (F ) ⊆ f −1 (F ). But f −1 (F ) ⊆ f −1 (F ), therefore, f −1 (F ) = f −1 (F ).

Hence, f −1 (F ) is closed in X, and thus, f is continuous on X.
Qamrul Hasan Ansari Metric Spaces Page 143

By combining Theorems 5.1.2 – 5.1.5, we have the following result.

Theorem 5.1.6. Let X and Y be metric spaces and f : X → Y be a function. The

following statements are equivalent:

(a) f is continuous on X.

(b) For each x ∈ X and for every sequence {xn } in X,

xn → x implies f (xn ) → f (x).

(c) f −1 (G) is open in X wherever G is open in Y .

(d) f −1 (F ) is closed in X wherever F is closed in Y .

(e) f (A) ⊆ f (A) for every subset A of X.

Theorem 5.1.7. Let X, Y and Z be metric spaces and f : X → Y and g : Y → Z be

continuous functions. Then the composite function g ◦ f : X → Z is continuous on X.

Proof. Let G be an open set in Z. Then by continuity of g, g −1 (G) is open in Y . The

continuity of f implies that f −1 (g −1 (G)) is open in X, that is, (f −1 ◦ g −1 ) (G) = (g ◦ f )−1 (G)
is open in X. Hence, g ◦ f is continuous.
Exercise 5.1.1. Let (X, d) be a metric space. If f, g : X → R are continuous on X,
then prove that f + g, f g, αf , for any α ∈ R, are so, where f + g, f g, αf are defined as
(f + g)(x) = f (x) + g(x), (f g)(x) = f (x)g(x), (αf ) = αf (x) for all x ∈ X.
Exercise 5.1.2. Let f : X → Y be a continuous function from a metric space X to another
metric space Y . Give an example to show that f (A◦ ) ⊆ (f (A))◦ does not hold.

Hint: Consider the function f : R → R defined by f (x) = sin x for all x ∈ R and take
A = (0, 3π).
Exercise 5.1.3. Give an example to show that the following statement does not hold:

f : X → Y is continuous if and only if (f (A))◦ ⊆ f (A◦ ) for all subsets A of X.

Solution. The function f : R → R defined by

(
0, if x is rational,
f (x) =
1, if x is irrational,

is not continuous. However, f (A) ⊆ {0, 1} for all A ⊆ X and (f (A))◦ = ∅. So, (f (A))◦ ⊆
f (A◦ ) holds for all subsets A of X.
Qamrul Hasan Ansari Metric Spaces Page 144

For the converse, define a function f : R → R as f (x) = |x| for all x ∈ R and consider the
set
A = ([−2, −1] ∩ Q) ∪ (−1, 1) ∪ ([1, 2] ∩ (R \ Q)) .
Then A◦ = (−1, 1), and therefore, f (A◦ ) = [0, 1). But f (A) = [0, 2], and so, (f (A))◦ =
(0, 2).
Exercise 5.1.4. Let X and Y be metric spaces. Prove that the function
f : X → Y is
continuous on X if and only if for every subset B of Y , f −1 (B) ⊆ f −1 B .

Proof. Let f be continuous and A = f −1 (B). Then by Theorem 5.1.5, we have f A ⊆ f (A)
and

f (A) ⊆ B ⇒ f (A) ⊆ B ⇒ f A ⊆ B.
Therefore, A ⊆ f −1 (B), and thus, f −1 (B) ⊆ f −1 (B).


Conversely, let f −1 (B) ⊆ f −1 B for every subset B of Y . Let F be a closed set in Y . Then
F = F , and by the hypothesis, we have


f −1 (F ) ⊆ f −1 F = f −1 (F ).

But f −1 (F ) ⊆ f −1 (F ), and therefore, f −1 (F ) = f −1 (F ). Thus f −1 (F ) is closed and hence f

is continuous on X.
Exercise 5.1.5. Let X and Y be metric spaces and f : X → Y be a function. Prove that
◦
f is continuous on X if and only if for every subset B of Y , f −1 (B ◦ ) ⊆ [f −1 (B)] .

Proof. Suppose that f is a continuous function on X. Let B be any arbitrary subset of

Y . Then B ◦ is open in Y , and by continuity of f , f −1 (B ◦ ) is open in X. Therefore,
◦
[f −1 (B ◦ )] = f −1 (B ◦ ). But
B ◦ ⊆ B ⇒ f −1 (B ◦ ) ⊆ f −1 (B),
and thus,  ◦  ◦
f −1 (B ◦ ) ⊆ f −1 (B) .
◦
This implies that f −1 (B ◦ ) ⊆ [f −1 (B)] .

Conversely, let G be an open subset of Y . Then G◦ = G. By the hypothesis

 −1 ◦
f (G) ⊇ f −1 (G◦ ) = f −1 (G).
◦ ◦
But [f −1 (G)] ⊆ f −1 (G), therefore, f −1 (G) = [f −1 (G)] and thus f −1 (G) is open in X.
Hence f is continuous on X.
Exercise 5.1.6. Let (X, d) be a metric space and for each i = 1, 2, . . . , n, the function
fi : X → R be continuous. Prove that the function f : X → Rn defined by
f (x) = (f1 (x), f2 (x), . . . , fn (x)) , for all x ∈ X,
is continuous.
Qamrul Hasan Ansari Metric Spaces Page 145

Proof. Let ε > 0 be given, and x0 be an arbitrary point in X. Since each fi is continuous,
there exists δi > 0 for each i = 1, 2, . . . , n such that for all x ∈ X and for each i = 1, 2, . . . , n
ε
d(x, x0 ) < δi implies |fi (x) − fi (x0 )| < √ .
n

Take δ = min {δ1 , δ2 , . . . , δn }. Then, d(x, x0 ) < δ implies that

 1/2
d (f (x), f (x0 )) = (f1 (x) − f1 (x0 ))2 + · · · + (fn (x) − fn (x0 ))2 < ε.

Thus, f is continuous at x0 ∈ X. Since x0 was arbitrary, f is continuous on X.

Exercise 5.1.7. Let (X, d) be a metric space and f : X → Rn be a function defined as

f (x) = (f1 (x), f2 (x), . . . , fn (x)) , for all x ∈ X,

where fi : X → R is a function for each i = 1, 2, . . . , n. If f is continuous on X, then prove

that each fi is also continuous on X.

Hint: In fact, for each i = 1, 2, . . . , n, we have

|fi (x) − fi (y)| ≤ d (f (x), f (y)) ,

n
!1/2
X
where d (f (x), f (y)) = |fi (x) − fi (y)|2 .
i=1

Exercise 5.1.8. Let Rn and R be equipped with the usual metric. Show that the projection
maps pi : Rn → R given by pi (x) = xi , where x = (x1 , x2 . . . , xn ), are continuous.

Exercise 5.1.9. Let X and Y be metric spaces and f : X → Y be a function. The graph
of f , denoted by graph (f ), is defined as graph (f ) = {(x, y) ∈ X × Y : y = f (x)}. If f is
continuous, then prove that graph (f ) is closed. Also, provide an example to show that the
closedness of the graph of f does not guarantee the continuity of f .

Exercise 5.1.10. Let X and Y be metric spaces and f, g : X → Y be continuous functions.

Prove that the set A = {x ∈ X : f (x) 6= g(x)} is open in X.

Exercise 5.1.11. Let (X, dX ) and (Y, dY ) be metric spaces and f, g : X → Y be continuous
functions. Prove that the set {x ∈ X : f (x) = g(x)} is closed. In particular, the set
{x ∈ X : f (x) = x} is closed.

Proof. Let {xn } be a sequence in A = {x ∈ X : f (x) = g(x)} such that xn → x0 . Then,

f (xn ) = g(xn ). By continuity of f and g, we have f (xn ) → f (x0 ) and g(xn ) → g(x0 ). Then,

dY (f (x0 ), g(x0 )) = dY (f (xn ), f (x0 )) + (g(xn ), g(x0)) → 0 as n → ∞.

Therefore, f (x0 ) = g(x0 ), and hence, x0 ∈ A. Thus, A is closed.

Qamrul Hasan Ansari Metric Spaces Page 146

Exercise 5.1.12. Let X and Y be metric spaces, D be a nonempty dense subset of X, and
f, g : X → Y be continuous functions such that f (x) = g(x) for all x ∈ D. Prove that f ≡ g
on X, that is, f (x) = g(x) for all x ∈ X.

Proof. Let A = {x ∈ X : f (x) = g(x)}. Then clearly, D ⊆ A. By Exercise 5.1.11, A is

closed, and so, X = D ⊆ A = A as D is dense in X. That is, X = A, and hence result
follows.

Exercise 5.1.13. Let (X, d) be a metric space. If f : X → R is continuous at x ∈ X and

if f (x) 6= 0, then there exists r > 0 such that f (y) 6= 0 for all y ∈ Sr (x), and the function
1
g(y) := f (y) from Sr (x) to R is continuous at x.
Qamrul Hasan Ansari Metric Spaces Page 147

5.2 Urysohn’s Lemma

Definition 5.2.1

Let A and B be subsets of a metric space X. If there exists a continuous function

f : X → [0, 1] such that

0, for x ∈ A,
f (x) =
1, for x ∈ B,
then we say that A and B can be separated by a continuous function.

The following result, which is known as Urysohn’s lemma, provides the existence of a con-
tinuous function that separates two nonempty disjoint closed sets in a metric space X.

Theorem 5.2.1 (Urysohn’s Lemma). Let A, B be two disjoint closed subsets of a metric
space (X, d). Then there exists a continuous function f : X → [0, 1] that separates A
and B, that is, 
0, for x ∈ A,
f (x) =
1, for x ∈ B,

Proof. Let ρ(x, A) = inf{d(x, y) : y ∈ A} be the distance of a point x ∈ X from the set A.
Define a function f : X → [0, 1] by

ρ(x, A)
f (x) = , for all x ∈ X.
ρ(x, A) + ρ(x, B)

Then clearly, f (x) = 0 for x ∈ A and f (x) = 1 for x ∈ B as ρ(x, A) = 0 if x ∈ A by Theorem

1.4.1.

We first prove that f is well defined, that is, ρ(x, A) + ρ(x, B) 6= 0 for any x ∈ X. If
ρ(x, A) + ρ(x, B) = 0, then each of the term, being nonnegative, must be zero. But then
x ∈ A and x ∈ B by Theorem 1.4.1, a contradiction since A and B are disjoint. Hence we
conclude that f (x) makes sense for all x ∈ X.

By Example 5.1.3, x 7→ d(x, A) and x 7→ d(x, B) are continuous. Since the sum and quotient
of continuous functions is a continuous function, f is also continuous.

If x ∈ A, then ρ(x, A) = 0 so that f (x) = 0 for x ∈ A. Also, if x ∈ B, then the denominator

of f (x) is ρ(x, A) + ρ(x, B) = ρ(x, A), and hence, f (x) = 1. Clearly, 0 ≤ f (x) ≤ 1 for all
x ∈ X.
Qamrul Hasan Ansari Metric Spaces Page 148

5.3 Continuous Functions and Compact Spaces

Theorem 5.3.1. Let X and Y be metric spaces and f : X → Y be a continuous
function. If A is a compact subset of X, then f (A) is compact in Y .

Proof. Let F = {Gα }α∈Λ be an open cover of f (A). Then by Theorem 5.1.3, f −1 (Gα ) is
open in X for each α ∈ Λ. Hence {A ∩ f −1 (Gα )}α∈Λ forms an open cover of A. Since A is
compact, there exists a finite set J = {1, 2, . . . , n} of Λ such that
n n
! n
!
[
[ [
A= A ∩ f −1 (Gk ) = A ∩ f −1 (Gk ) = A ∩ f −1 Gk .
k=1 k=1 k=1

Thus, !
n
[
A ⊆ f −1 Gk ,
k=1

and so,
n
[
f (A) ⊆ Gk .
k=1

Hence, {G1 , G2 , . . . , Gn } is a finite subcover of F . Thus, f (A) is compact.

Corollary 5.3.1. Let X and Y be metric spaces and f : X → Y be a continuous

function. If X is compact, then f (X) is bounded.

Theorem 5.3.2. Let X and Y be metric spaces and f : X → Y be a continuous

function. If X is compact, then f (F ) is closed in Y whenever F is closed in X.

Proof. Let F be a closed subset of X. Since every closed subset of a compact set is compact,
by Theorem 5.3.1, we have f (F ) is compact and hence it is closed.

Theorem 5.3.3. Let X and Y be metric spaces and f : X → Y be a continuous

function. If f is bijective and X is compact, then f −1 is continuous on Y .

Proof. Since f is bijective, f −1 : Y → X exists and also bijective. Let F be a closed set in
−1
X. Then (f −1 ) (F ) = f (F ) and by Theorem 5.3.2, f (F ) is closed in Y . Thus, the inverse
image of a closed set is closed, and hence, f −1 is continuous.
Qamrul Hasan Ansari Metric Spaces Page 149

Remark 5.3.1

In Theorem 5.3.3, if X is not compact, then f −1 need not be continuous. For example,
consider and identity function I : (R, d) → (R, | · |) from R with the discrete metric to
R with the usual metric. Then I is continuous but I −1 is not.

Exercise 5.3.1. Give an example to show that Theorem 5.3.2 does not hold if X is not
compact.

Exercise 5.3.2. Let X and Y be metric spaces and f : X → Y be a bijective and continuous
function. If A is a nonempty dense subset of X, then f (A) is dense in Y .

Proof. Let A be a dense subset of X. Then we need to show that f (A) is dense in Y ,
equivalently, f (A) = Y = f (X) because f is bijective.

Let y ∈ f (X) be any point and consider an open sphere Sr (y) for r > 0. Since Sr (y) is open
and f is continuous,
f −1 (Sr (y)) = {x ∈ X : f (x) ∈ Sr (y)}
is open in X. Since A is dense in X, by Theorem 3.2.1, f −1 (Sr (y)) ∩ A is nonempty. Let
x ∈ f −1 (Sr (y)) ∩ A. Then,


f (x) ∈ f f −1 (Sr (y)) ∩ A = Sr (y) ∩ f (A),

that is, Sr (y) ∩ f (A) 6= ∅. Therefore, y is a limit point of f (A), that is, y ∈ f (A). Thus,
f (X) ⊆ f (A), and hence, f (A) = f (X) = Y .

Exercise 5.3.3. Let X and Y be metric spaces and f : X → Y be a continuous bijective

mapping. If A ⊆ X is nowhere dense in X, then prove that f (A) may not be nowhere dense.

Hint: Consider the usual metric space R and Q = {rn: n ∈
N} be the set of all rational
numbers. Let A = {(x, 0) : x ∈ R \ Q}. Let M = A ∪ rn , n1 : n ∈ N ⊆ R × R with the
Euclidean (usual) metric.

The projection map f (x, y) = x is a continuous bijection from M to R. A is a closed nowhere

dense subset of M, while f (A) = R \ Q is a dense subset of R.

Exercise 5.3.4 (The Extreme Value Theorem). Prove that a continuous real-valued function
defined on a compact set is bounded and it assumes maximum and minimum values.

Proof. Let f : X → R be continuous and A be a compact subset of a metric space X. By

Theorem 5.3.1, f (A) is a compact subset of R. By Heine-Borel Theorem 4.2.4, f (A) is closed
and bounded. Thus sup f (A) and inf f (A) exist and belong to f (A). Therefore, there exist
x̂, x̃ ∈ A such that for all y ∈ A, inf f (A) = f (x̂) ≤ f (y) ≤ f (x̃) = sup f (A).
Qamrul Hasan Ansari Metric Spaces Page 150

Exercise 5.3.5. Let (X, d) be a metric space and A be a nonempty compact subset of X.
Prove that for every x0 ∈ X, there exists a y0 ∈ A such that

d(x0 , y0) = d(x0 , A) = inf d(x0 , y).

y∈A

Proof. Consider the function f : A → R+ defined by f (x) = d(x, x0 ) for all x ∈ A. Now
|f (x) − f (y)| = |d(x, x0 ) − d(y, x0 )| ≤ d(x, y), so f is continuous on A. But A is compact, so
f has a minimum on A by Exercise 5.3.4. That is, there exists a y0 ∈ A such that

f (y0) = d(x0 , y0 ) = inf d(x0 , y) = d(x0 , A).

y∈A

Exercise 5.3.6. Let (X, d) be a metric space and f : X → R be a continuous function.

Prove that the set A = {x ∈ X : |f (x)| < r} is open in X for all r > 0.

Hint: A = {x ∈ X : −r < f (x) < r} = f −1 (−r, r). Since (−r, r) is open in R, its inverse
image must be open in X by continuity of f .
Qamrul Hasan Ansari Metric Spaces Page 151

5.4 Uniform Continuous Functions

Before giving the definition of uniform continuity, we examine the following examples.

Consider a real-valued function f : [−1, 1] → R defined as f (x) = x2 . Let x, x0 be any points

of [−1, 1]. Then,

d (f (x), f (x0 )) = |f (x) − f (x0 )| = x2 − x20

= |x − x0 | · |x + x0 | < ε,

whenever |x − x0 | < 21 ε = δ, where δ is independent on the choice of x and x0 .

Thus for any ε > 0, there exists a δ = 12 ε such that for any x, x0 ∈ [−1, 1], we have

d (f (x), f (x0 )) < ε whenever d(x, x0 ) < δ.

Now, if we consider the same function f (x) = x2 defined on R, that is, f : R → R such that
f (x) = x2 . Then for every real numbers x, x0 , we have

d (f (x), f (x0 )) = x2 − x20 = |x − x0 | · |x + x0 | < ε

ε
whenever |x − x0 | < |x+x0 |
= δ, where δ depends on ε and x0 .

In this way, we see that δ may depend not only on ε but also on x0 . Uniform continuity is
essentially continuity plus the added condition that for each ε we can find a δ which works
uniformly over the entire space, in the sense that it does not depend on x0 .

Definition 5.4.1

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

continuous if for each ε > 0, there exists a δ > 0 (depends only on ε) such that for every
x, y ∈ X,
dX (x, y) < δ implies dY (f (x), f (y)) < ε.

Remark 5.4.1

Every uniform continuous function is continuous but converse need not be true in gen-
eral. For example, in the first example mentioned above, δ is independent on the choice
of x and x0 , and therefore, it is uniformity continuous. But in the latter example, δ
depends on ε and x0 , and hence, it is only continuous but not uniformly continuous.
Qamrul Hasan Ansari Metric Spaces Page 152

Example 5.4.1

(a) Let X be a discrete metric space and Y be any metric space. Then any function
f : X → Y is uniformly continuous.

(b) Let X = (0, 1) be a metric space with the metric induced by the usual metric on
R and Y = R with the usual metric. The function f : X → Y defined as f (x) = x1
for all x ∈ X, is not uniformly continuous.
Let ε = 21 and δ be any positive number. Choose x = 1
n
and y = 1
n+1
, where n is
a positive integer such that n > 1δ . Then,

1 1 1 1
|x − y| = − = < < δ,
n n+1 n(n + 1) n

but |f (x) − f (y)| = |n − (n + 1)| = 1 > ε. Thus whatever δ > 0 may be, there
exist x and y such that |x − y| < δ but |f (x) − f (y)| > 21 .

(c) No polynomial function of degree greater than 1 is uniformly continuous on the

usual metric space R. Note that any polynomial function is continuous.

(d) The logarithmic function is not uniformly continuous on the usual metric space
X = (0, ∞).

Theorem 5.4.1. Let (X, dX ) and (Y, dY ) be metric spaces and f : X → Y be a contin-
uous function. If X is compact, then f is uniformly continuous.

Proof. Let ε > 0 and x ∈ X be arbitrary. Consider the image f (x) of x and the open sphere
Sε (f (x)). Since f is continuous, f −1 (Sε (f (x))) is an open set in X. Consider the family
F = {f −1 (Sε (f (x))) : x ∈ X} of these open sets in X. Then clearly F is an open cover of
X. Since X is compact, it is sequentially compact, and therefore, by Lebesgue’s covering
lemma (Theorem 4.3.4) there exists a Lebesgue number δ > 0 for F . Thus every open sphere
of diameter less than δ will be contained in at least one member of F and, consequently, we
have

Sδ/2 (x) ⊆ f −1 (Sε (f (x))) ⇒ f Sδ/2 (x) ⊆ Sε (f (x)).
Hence for each ε > 0, there exists a δ̃ > 0 (independent of x) such that

δ
dX (x, y) < = δ̃ ⇒ dY (f (x), f (y)) < ε.
2
Hence, f is uniformly continuous.
Qamrul Hasan Ansari Metric Spaces Page 153

Remark 5.4.2

Consider the space C[a, b] of all continuous real-valued functions defined on [a, b] with
the metric
d∞ (f, g) = max |f (t) − g(t)|, for all f, g ∈ C[a, b].
t∈[a,b]

Note that every f ∈ C[a, b] is a function from a compact set [a, b] to R. Then from the
above theorem, every f ∈ C[a, b] is uniformly continuous. However, as we have seen in
Remark 4.3.3 that C[a, b] is not sequentially compact, and hence not compact.

Theorem 5.4.2. Composition of two uniformly continuous functions is a uniformly

continuous function.

Theorem 5.4.3. Let (X, dX ) and (Y, dY ) be metric spaces and f : X → Y be an

uniformly continuous function. If {xn } is a Cauchy sequence in X, then {f (xn )} is also
a Cauchy sequence in Y .

Proof. Since f is uniformly continuous, for each ε > 0, there exists a δ > 0 (depends only
on ε) such that for all x1 , x2 ∈ X,

dX (x1 , x2 ) < δ implies dY (f (x1 ), f (x2 )) < ε.

In particular, we have

dX (xn , xm ) < δ implies dY (f (xn ), f (xm )) < ε. (5.2)

Since {xn } is a Cauchy sequence in X, for a given δ > 0, there exists a positive integer N
such that
dX (xn , xm ) < δ, for all n, m ≥ N. (5.3)
From (5.2) and (5.3), we obtain

dY (f (xn ), f (xm )) < ε, for all n, m ≥ N.

Hence {f (xn )} is a Cauchy sequence in Y .

Qamrul Hasan Ansari Metric Spaces Page 154

The following example shows that a continuous function may not map a Cauchy sequence
into a Cauchy sequence.
Example 5.4.2

Let X = (0, ∞) with the induced usual metric on R and Y = R with the usual metric.
The function
 f : X → Y defined by f (x) = x1 for all x ∈ X,
is continuous on X.
Clearly, xn : xn = n1 n∈N is a Cauchy sequence in X. But f n1 n∈N = {n}∞ n=1 is not
a Cauchy sequence in Y . Indeed, the absolute difference of any two distinct points is at
least as large as 1.

Exercise 5.4.1. Show that the function f (x) = ex defined on the usual metric space R is
not uniformly continuous.

Hint: There is no δ > 0 that guarantees d(x, y) = |ex − ey | < 1 for all x, y ∈ R with
ey − ex
|x − y| < δ. Specially, for all x, y ∈ R with x < y, we have > ex . Then for any
y−x
γ ∈ (0, 1), pick x = − ln γ and y = γ − ln γ, so that ey − ex > (y − x)ex = 1.
Exercise 5.4.2. Let (X, d) be a metric space and A be a subset of X. Prove that the
function f : X → R defined by
f (x) = d(x, A), for all x ∈ X
is uniformly continuous.

Proof. By the triangular inequality, we have

d(x, a) ≤ d(x, y) + d(y, a), for all a ∈ A, x ∈ X.
By taking infimum, we obtain
inf d(x, a) ≤ d(x, y) + inf d(y, a).
a∈A a∈A

Therefore,
d(x, A) ≤ d(x, y) + d(y, A),
and so,
d(x, A) − d(y, A) ≤ d(x, y), for all x, y ∈ X.
By interchanging x and y, we obtain
d(y, A) − d(x, A) ≤ d(y, x) = d(x, y).
Thus,
|d(x, A) − d(y, A)| ≤ d(x, y).
Therefore, for a given ε > 0, choosing a δ such that 0 < δ < ε, we have
|f (x) − f (y)| = |d(x, A) − d(y, A)| ≤ d(x, y) < δ < ε,
that is,
|f (x) − f (y)| < ε, whenever d(x, y) < δ.
Hence, f is uniformly continuous on X.
Qamrul Hasan Ansari Metric Spaces Page 155

5.5 Homeomorphism and Equivalent Metrics

Definition 5.5.1

Let X and Y be two metric spaces. A function f : X → Y is said to be a homeomorphism

if

(i) f is bijective, that is, f is one-to-one and onto

(ii) f and f −1 both are continuous.

Two metric spaces X and Y are said to be homeomorphic if there exists a homeomor-
phism from X onto Y .

Example 5.5.1

(a) Let [0, 1] and [0, 2] be metric spaces with the usual metric and f : [0, 1] → [0, 2]
be a function defined as f (x) = 2x for all x ∈ [0, 1]. Then f is a homeomorphism
and the spaces [0, 1] and [0, 2] are homeomorphic.

(b) Let X = (0, ∞) and Y = R be metric spaces such that Y is equipped with the
usual metric and X with the usual metric induced from R. Let f : X → Y be
a function defined as f (x) = ln x for all x ∈ X. Then f is a homeomorphism
between X and Y .

(c) The metric space R with the usual metric is not homeomorphic to the metric space
R with the discrete metric.

Recall that a mapping f : X → Y from a metric space (X, dX ) to a metric space (Y, dY ) is
an isometry if
dY (f (x), f (y)) = dX (x, y), for all x, y ∈ X.
It is obvious that an isometry is one-to-one and uniformly continuous.

Recall also that X and Y are said to be isometric if there exists an isometry between them
that is onto.

An isometry is necessarily a homeomorphism, but the converse need not be true. In Example
5.5.1 (a) and (b), the mappings are homeomorphism but not isometry.
Qamrul Hasan Ansari Metric Spaces Page 156

Theorem 5.5.1. Let X and Y be two metric spaces and f : X → Y be a bijective

function. Then the following statements are equivalent.

(a) f is a homeomorphism.

(b) The set G ⊆ X is open if and only if its image f (G) ⊆ Y is open.

(c) The set F ⊆ X is closed if and only if its image f (F ) ⊆ Y is closed.

Proof. (a) ⇒ (b). Let (a) hold. Then f and f −1 both are continuous. Therefor, for any
−1
open set G ⊆ X, we have (f −1 ) (G) = f (G) is open in Y by using continuity of f −1 .

Conversely, let f (G) ⊆ Y be open. Then by the continuity of f , f −1 (f (G)) = G is open in

X (f −1 (f (G)) = G because f is injective).

(b) ⇒ (c). Let F ⊆ X be a closed set. Then X \ F is open in X. By (b), f (X \ F ) is open

in Y . But f (F ) = Y \ f (X \ F ). Hence f (F ) is closed in Y .

Conversely, let f (F ) is closed in Y . Then Y \ f (F ) is open in Y . Then f (X \ F ) = Y \ f (F )

is open. By (b), X \ F is open, and hence, F is closed.

(c) ⇒ (a). Let F be a closed subset of X. Then by hypothesis, f (F ) is closed in Y . Since

f −1 (f (F )) = F , the inverse image of closed set is closed, and therefore, f is continuous.
−1
Also, since (f −1 ) (F ) = f (F ), by hypothesis F is closed in X and f (F ) is closed in Y , we
have f −1 is continuous on Y . Therefore, f is a homeomorphism.

Remark 5.5.1

Let X, Y and Z be metric spaces. If X is homeomorphic to Y and Y is homeomorphic

to Z, then X is homeomorphic to Z.

Definition 5.5.2

Let X and Y be metric spaces. A mapping f : X → Y is said to be open (respectively,

closed) if f (G) is open (respectively, closed) in Y for every open (respectively, closed)
set G in X.
Qamrul Hasan Ansari Metric Spaces Page 157

Remark 5.5.2

A bijection function f : X → Y is closed if and only if it is open. Indeed, if f is a closed

bijection and G is an open set in X, then the complement Gc of G is closed in X, and
therefore the image f (Gc ) is closed in Y since f is closed. But the bijectivity means
that f (Gc ) = (f (G))c , which implies that f (G) is open in Y , being the complement of
a closed set. Thus, f is an open map as well.

Remark 5.5.3

Let X and Y be metric spaces.

(a) A continuous and open mapping f : X → Y may not be a homeomorphism.

(b) A continuous and closed mapping f : X → Y may not be a homeomorphism.

(c) A continuous and closed mapping f : X → Y may not be open.

(d) A continuous and open mapping f : X → Y may not be closed.

Exercise 5.5.1. Let X and Y be metric spaces and f : X → Y be a bijective function.

Prove that the following conditions are equivalent:

(a) f is homeomorphism;

(b) f (A) = f (A) for all A ⊆ X.

(c) f (A◦ ) = [f (A)]◦ for all A ⊆ X.

(d) f −1 (B) = f −1 (B) for all B ⊆ Y .

(e) f −1 (B ◦ ) = [f −1 (B)]◦ for all B ⊆ Y .

(f) For every sequence {xn } in X and any x ∈ X, the sequence {f (xn )} in Y converges to
f (x) if and only if xn converges to x.

Exercise 5.5.2. Let S1 := {(x, y) ∈ R2 + x2 + y 2 = 1} be a unit circle in the usual metric

space R2 . Define a map f : [0, 2π) → S1 by f (θ) = (cos θ, sin θ). Prove that f is continuous
and bijective but not a homeomorphism.

Exercise 5.5.3. Let X, Y and Z be metric spaces. If X is homeomorphic to Y and Y is

homeomorphic to Z, then prove that X is homeomorphic to Z.

Proof. Let f : X → Y and g : Y → Z be homeomorphisms. Then f , g are bijective,

continuous and f −1 , g 1 are also continuous. Consider the function g ◦ f : X → Z.
Qamrul Hasan Ansari Metric Spaces Page 158

Since composition of two bijective and continuous functions is bijective and continuous,
respectively, g ◦ f is bijective and continuous. Also since (g ◦ f )−1 = f −1 ◦ g −1 is continuous,
we have that g ◦ f is a homeomorphism, and hence X is homeomorphic to Z.
Exercise 5.5.4. Show that the function f : R → (−1, 1) defined by
x
f (x) = , for all x ∈ R
1 + |x|
is a homeomorphism. Also, show that f is uniformly continuous on R.

Hint: The given function f can be restated as

(
x
(1+x)
, if x ≥ 0,
f (x) = x
(1−x)
, if x < 0.

Observe that the inverse function g : (−1, 1) → R of the function f is given by

( y
(1−y)
, if y ≥ 0,
g(y) = y
(1+y)
, if y < 0.
x
In fact, for x ≥ 0, we have (1+x)
≥ 0 and hence,
  x
x (1+x)
(g ◦ f )(x) = g = x = x,
1+x 1 − (1+x)
x
whereas, for x < 0, we have (1−x)
< 0 and hence,
  x
x (1−x)
(g ◦ f ) (x) = g = x = x.
1−x 1 − (1−x)
Similar computations for f ◦ g is left for the readers. Since the function f and its inverse g
are both continuous, f is a homeomorphism.

Now we check the uniform continuity of f . For x, y > 0,

x y |x − y|
|f (x) − f (y)| = − = < |x − y|.
1+x 1+y (1 + x)(1 + y)
Similarly, for x < 0 and y < 0, we have |f (x) − f (y)| < |x − y|. For x > 0 and y < 0,
x y |x − y − 2xy|
|f (x) − f (y)| = − =
1+x 1−y (1 + x)(1 − y)
x − y − 2xy + x + y 2
2
≤
(1 + x)(1 − y)
(x − y)(1 + x − y)
=
(1 + x)(1 − y)
≤ x − y = |x − y|.
The case when x or y is 0 is left to the readers.
Qamrul Hasan Ansari Metric Spaces Page 159

Exercise 5.5.5. (a) Show that the image of a complete metric space under homeomor-
phism need not be complete.

(b) Show that the image of a complete metric space under a one-to-one uniformly contin-
uous mapping need not be complete.

Hint: The function f in Exercise 5.5.4 is both homeomorphism and uniformly continuous.
R is complete but f (R) = (−1, 1) is not.

Exercise 5.5.6. Let X and Y be metric spaces. Given an example of a function f : X → Y

that is

(a) continuous and open, but is not a homeomorphism;

(b) continuous and closed, but is not a homeomorphism;

(c) continuous and closed, but is not open;

(d) continuous and open, but is not closed.

Hint: (a) Consider S = {(x, y) ∈ R2 : x2 + y 2 = 1} and define a map f : R → S as f (t) =

(cos(2πt), sin(2πt)). Since the components of f are continuous, and therefore, f is continuous.
Observe that it is also open. Indeed, let (a, b) = I and ℓ(I) = b−a. If ℓ(I) > 1, then f (I) = S
which is open in S. If ℓ(I) < 1, then f (I) is an arc without endpoints and so is open in S,
being the intersection of S with an open disc with the missing endpoints on its edge. The
function is not one-to-one and thus, cannot be a homeomorphism.

(b) and (c) Consider any constant function f : R2 → R and observe that it is continuous
and closed, but not a homeomorphism, as it is not one-to-one.

(d) Consider the function f : R2 → R defined by f (x1 , x2 ) = x1 for all (x1 , x2 ) ∈ R2 and
observe that it is continuous and open but not closed, as it maps the hyperbola x1 x2 = 1
onto the set R \ {0}, which is not closed.

Exercise 5.5.7. Let (X, dX ) and (Y, dY ) be metric spaces. Let X × Y be equipped with the
product metric defined by

d((x1 , y1 ), (x2 , y2 )) := max{dX (x1 , x2 ), dY (y1 , y2)}, for all (x1 , y1), (x2 , y2 ) ∈ X × Y.

Let px and py denote the projection of X ×Y onto X, respectively onto Y given by px (x, y) =
x, respectively py (x, y) = y. Show that px and py are continuous and open but not closed.

Exercise 5.5.8. Show that the identity mapping I : (R, | · |) → (R, d) from the usual metric
space (R, | · |) to the discrete metric space (R, d) is continuous but neither open nor closed.

Exercise 5.5.9. Let X, Y and Z be metric spaces.

Qamrul Hasan Ansari Metric Spaces Page 160

(a) If f : X → Y is a continuous and onto and g ◦ f is open, then prove that the map
g : Y → Z is open.

(b) If g : Y → Z is a continuous and injective and g ◦ f is open, then prove that the map
f : X → Y is open.

Exercise 5.5.10. An indicator function (also known as characteristic function) of a set A

is defined by (also denoted by iA )

1, if x ∈ A,
χA (x) =
0, otherwise.

Let A be a nonempty subset of a metric space (X, d). Prove that the indicator function χA
of the set A is continuous if and only if A is closed as well as open.

Proof. Let χA be continuous. Observe that χA : X → {0, 1} is indeed a function between

metric spaces when each subset of R is taken together with the relative metric. Clearly,

χ−1 −1 c
A ({1}) = A and χA ({0}) = A .

As {1} is closed in {0, 1}, A is closed in X (by continuity of χA ). Also, as {0} is closed in
{0, 1}, Ac is closed in X (by continuity of χA ), and so A is open in X. Therefore, A is closed
as well as open.

Conversely, assume that A is both open and closed set in X and consider an arbitrary open
subset O of R. Then we show that χ−1
A (B ∩ O) is open for any subset B of {0, 1}. Note that

χ−1
A (∅) = ∅, open set
χ−1
A (0) = Ac , open set, since A is closed
−1
χA (1) = A, open set
−1
χA ({0, 1}) = X, open set.

Therefore, χA is continuous.

Exercise 5.5.11. Assume that the metric space (X, d) is not compact. Prove that there
exists a function f : X → R which is continuous but not bounded.
Qamrul Hasan Ansari Metric Spaces Page 161

Definition 5.5.3

Two metrics d1 and d2 on the same underlying set X are said to be equivalent if for
every sequence {xn } in X and x ∈ X,

lim d1 (xn , x) = 0 if and only if lim d2 (xn , x) = 0,

n→∞ n→∞

that is, a sequence converges to x with respect to the metric d1 if and only if it converges
to x with respect to the metric d2 .

The metric spaces (X, d1 ) and (X, d2 ) are said to be equivalent if the metrics d1 and d2
are equivalent.

Remark 5.5.4

(a) Two metrics d1 and d2 on a nonempty set X are equivalent if and only if the
identity map I : (X, d1 ) → (X, d2 ) together with its inverse map I −1 : (X, d2 ) →
(X, d1 ) are continuous, that is, if and only if the identity map I : (X, d1 ) → (X, d2 )
is a homeomorphism.

(b) If two metrics are equivalent, then the families of open sets are same in (X, d1 )
and (X, d2 ).

The following result provides a sufficient condition for two metrics on a set to be equivalent.
Theorem 5.5.2. Two metrics d1 and d2 on a nonempty set X are equivalent if there
exist constants k1 , k2 > 0 such that

k1 d2 (x, y) ≤ d1 (x, y) ≤ k2 d2 (x, y), for all x, y ∈ X. (5.4)

Proof. Let {xn } be a sequence in X such that d1 (xn , x) → 0 as n → ∞. From the first
inequality in (5.4), we have
k1 d2 (xn , x) ≤ d1 (xn , x) → 0, as n → ∞
and thus, d2 (xn , x) → 0 as n → ∞.

If d2 (xn , x) → 0 as n → ∞, then in view of the second inequality in (5.4), we have

d1 (xn , x) ≤ k2 d2 (xn , x) → 0, as n → ∞
and thus, d1 (xn , x) → 0 as n → ∞.

The following example shows that the converse of the above theorem does not hold, that is,
if d1 and d2 are equivalent metrics then the inequality (5.4) need not be true.
Qamrul Hasan Ansari Metric Spaces Page 162

Example 5.5.2

Let (X, d) be a metric space and ρ be anther metric on X defined by

d(x, y)
ρ(x, y) = , for all x, y ∈ X.
1 + d(x, y)

Then d and ρ are equivalent metrics on X.

Indeed, let d(xn , x) → 0 as n → ∞. Then obviously, ρ(xn , x) → 0 as n → ∞. On the

other hand, if ρ(xn , x) → 0 as n → ∞, then for 0 < ε < 1, we have 2 − ε > 1, and also,
there exists a positive integer N such that ρ(xn , x) < ε/2 for all n > N. So

d(xn , x) ε
= ρ(xn , x) < , for all n > N
1 + d(xn , x) 2

which is equivalent to
ε
d(xn , x) < < ε, for all n > N.
2−ε
Thus, d(xn , x) → 0 as n → ∞.

On the other hand, we observe that there is no constant k1 > 0 such that

k1 d(x, y) ≤ ρ(x, y), for all x, y ∈ X.

This establishes that the condition (5.4) in Theorem 5.5.2 is not necessary for the metrics
d and ρ to be equivalent.

This example shows that there always exists an equivalent metric ρ on a metric space
such that (X, ρ) is bounded.
Qamrul Hasan Ansari Metric Spaces Page 163

Now we present some examples of equivalent and non-equivalent metrics.

Example 5.5.3

The metrics d1 , d2 and d∞ defined on Rn by

n
X
d1 (x, y) = |xi − yi |,
i=1
n
!1/2
X
d2 (x, y) = (xi − yi )2 ,
i=1
d∞ (x, y) = max |xi − yi | ,
1≤i≤n

for all x = (x1 , . . . , xn ) ∈ Rn and y = (y1 , . . . , yn ) ∈ Rn , are equivalent. In fact, any real
numbers α1 , α2 , . . . , αn , we have

n
!1/2 n n
!1/2
X X X
αi2 ≤ |αi | ≤ n · max {|αi | : 1 ≤ i ≤ n} ≤ n αi2 .
i=1 i=1 i=1

Example 5.5.4

Let m be the space of all bounded sequences with the following two metrics

d(x, y) = sup |xn − yn | ,

1≤n<∞

and ∞
X 1 |xn − yn |
d1 (x, y) = ,
n=1
2n 1 + |xn − yn |
where the sequences x = (x1 , x2 , . . .) and y = (y1 , y2 , . . .) are elements of m.

Consider the sequence {en } in m whose respective terms are the sequences

e1 = (1, 0, 0, . . .), e2 = (0, 1, 0, . . .), . . . , en = (0, 0, . . . , 0, 1, 0, . . .), . . . ,

where 1 is in the nth place. Let e0 = (0, 0, 0, . . .). Then,

1 1 1
d (en , e0 ) = 1 while d1 (en , e0 ) = n
= n+1 → 0, as n → ∞.
2 1+1 2
Thus the sequence {en } converges with respect to the metric d1 , but not with respect
to the metric d. Hence the metrics d and d1 are not equivalent.
Qamrul Hasan Ansari Metric Spaces Page 164

Example 5.5.5

Let C[0, 1] be the space of all continuous linear functionals defined on [0, 1]. Consider
the metrics d∞ and d defined on C[0, 1] by

d∞ (f, g) = sup |f (x) − g(x)| ,

x∈[0,1]

and Z 1
d(f, g) = |f (x) − g(x)| dx,
0

for all f, g ∈ C[0, 1]. Then these metrics d∞ and d are not equivalent.

Indeed, let {fn } be a sequence in C[0, 1] defined by

fn (x) = xn , for all x ∈ [0, 1].

Also, let f (x) = 0 for all x ∈ [0, 1]. Then d (fn , f ) → 0 as n → ∞. In fact,
Z 1
1
d (fn , f ) = xn dx = → 0, as n → ∞.
0 n+1
However,
d∞ (fn , f ) = sup |fn (x) − f (x)| = 1, for all n.
x∈[0,1]

So, lim d∞ (fn , f ) 6= 0. Thus the sequence {fn } converges to f with respect to the
n→∞
metric d, but not with respect to the metric d∞ . Therefore, the metrics d∞ and d are
not equivalent.

However, for any sequence {fn } in C[0, 1] and any f ∈ C[0, 1], when d∞ (fn , f ) → 0 as
n → ∞, it does follow that d (fn , f ) → 0 as n → ∞. Observe that
Z 1
d (fn , f ) = |fn (x) − f (x)| dx ≤ sup |fn (x) − f (x)| = d∞ (fn , f ) .
0 x∈[0,1]

Exercise 5.5.12. Give an example to show that the metric spaces (X, d1 ) and (X, d2 ) are
homeomorphic but their respective metrics d1 and d2 are not equivalent.

Hint: Consider X = R and define a function f : X → R by


−x, when |x| < 1,
f (x) =
x, otherwise.

Consider the following metrics d1 and d2 on X:

d1 (x, y) = |x − y| and d2 (x, y) = |f (x) − f (y)|, for all x, y ∈ X.

Qamrul Hasan Ansari Metric Spaces Page 165

Then f is a homeomorphism from (R, d1 ) onto (R, d2 ) but the metrics are not equivalent,
because the set of positive reals is open under d1 but not under d2 .

Exercise 5.5.13. Let X and Y be metric spaces. If X is complete and f : X → Y is an

isometry, then prove that f (X) is closed.

Proof. Let {yn } be a sequence in f (X) such that yn → y. Then there are xn ∈ X such
that f (xn ) = yn . Since {yn } is convergent, it is Cauchy, and hence {xn } is Cauchy as
f is an isometry. Since X is complete, xn → x ∈ X, and hence f (xn ) → f (x). Thus,
y = f (x) ∈ f (X), and therefore, f (X) is closed.

Exercise 5.5.14. Let X and Y be metric spaces such that X is complete and Y is compact.
If f : X → Y is an isometry, then prove that f (X) is compact.
Qamrul Hasan Ansari Metric Spaces Page 166

5.6 Uniform Convergence of Sequences of Functions

In this section, we study two ways in which a sequence {fn } of continuous functions can
converge to a function f : pointwise convergence and uniform convergence. The distinction is
rather similar to the distinction between pointwise and uniform continuity – in the pointwise
case, a condition can be satisfied in different ways for different x’s, however, in the uniform
case, it must be satisfied in the same for all x.

Definition 5.6.1

Let {fn } be a sequence of functions from a metric space (X, dX ) to another metric
space (Y, dY ) and f : X → Y be a function. We say that the sequence {fn } converges
pointwise to the function f if for any ε > 0 and for all x ∈ X, there exists a positive
integer N such that
dY (fn (x), f (x)) < ε, for all n > N. (5.5)

In general, the integer N depends on ε as well as on x. It is not always possible to find an

integer N such that (5.4) holds for all x ∈ X simultaneously.

If N depends only on ε, then we say that the sequence {fn } converges uniformly on X.

Definition 5.6.2

Let {fn } be a sequence of functions from a metric space (X, dX ) to another metric
space (Y, dY ) and f : X → Y be a function. We say that the sequence {fn } converges
uniformly to the function f if for any ε > 0, there exists a positive integer N (depending
only on ε) such that

dY (fn (x), f (x)) < ε, for all n > N and all x ∈ X. (5.6)

It is clear that uniform convergence implies pointwise convergence, but the converse need
not be true.
Qamrul Hasan Ansari Metric Spaces Page 167

Example 5.6.1

Let X = [0, 1] and Y = R with the usual metric. Let {fn } be a sequence of functions
fn : X → Y defined by

fn (x) = xn , for all x ∈ [0, 1] and all n ∈ N.

Then {fn } converges pointwise to the function f : X → Y defined by


0, if 0 ≤ x < 1,
f (x) =
1, if x = 1.

Remark 5.6.1

(a) Observe that the function f is not continuous even though each of the function fn
is continuous. That is, a sequence of continuous functions may converge pointwise
to a non-continuous function.

(b) Since f is not continuous and {fn } is a sequence of continuous functions, by

Theorem 5.6.1, {fn } does not converge uniformly to f .

Example 5.6.2

Let X = Y = R with the usual metric. Let {fn } be a sequence of functions fn : X → Y

defined by

1 − n1 |x|, if |x| < n,
fn (x) =
0, if |x| ≥ n.

Then {fn } converges pointwise to the function f : X → Y defined by

f (x) = 1, for all x ∈ X.

But {fn } does not converge uniformly to f .

Indeed, let ε = 12 . Note that for every n ∈ N, there exists x0 ∈ X with fn (x0 ) = 0 and
so
|fn (x0 ) − f (x0 )| = 1 > ε.

The above examples show that the pointwise limit of a sequence of continuous functions may
not be continuous. However, the following theorem says that if the convergence is uniform,
then the limit function is also continuous.
Qamrul Hasan Ansari Metric Spaces Page 168

Theorem 5.6.1. Let (X, dX ) and (Y, dY ) be metric spaces, {fn } be a sequence of con-
tinuous functions each defined on X to Y and f : X → Y be a function. If fn → f
uniformly on X, then f is continuous on X. In other words, an uniform limit of a
sequence of continuous functions is continuous.

Proof. Let ε > 0 be given and x0 ∈ X be arbitrary. Since fn → f uniformly on X, there

exists a positive integer N (depending on ε only) such that for each x ∈ X,
ε
dY (fn (x), f (x)) < , for all n > N. (5.7)
3
Let n0 > N. Since fn0 is continuous at x0 , we can choose δ > 0 such that for all x ∈ Sδ (x0 ),
that is, d(x, x0 ) < δ, we have
ε
dY (fn0 (x), fn0 (x0 )) < . (5.8)
3
By using (5.7) and (5.8), for all x ∈ Sδ (x0 ), we get
dY (f (x), f (x0 )) ≤ dY (f (x), fn0 (x)) + dY (fn0 (x), fn0 (x0 )) + dY (fn0 (x0 ), f (x0 ))
ε ε ε
< + + = ε,
3 3 3
and therefore, f is continuous at x0 .

Proposition 5.6.1. Let (X, dX ) and (Y, dY ) be metric spaces and {fn } be a sequence
of functions fn : X → Y . Then for any function f : X → Y , {fn } converges uniformly
to f if and only if sup {dY (fn (x), f (x))} → 0 as n → ∞.
x∈X

Proof. Assume that {fn } converges uniformly to f . Then for any ε > 0, there exists a
positive integer N (depending only on ε) such that
dY (fn (x), f (x)) < ε, for all n > N and all x ∈ X,
that is,
sup {dY (fn (x), f (x))} ≤ ε, for all n > N.
x∈X
Since ε was arbitrary, we have
sup dY (fn (x), f (x)) → 0, as n → ∞.
x∈X

Conversely, assume that sup {dY (fn (x), f (x))} → 0 as n → ∞. Then for a given ε > 0, there
x∈X
exists a positive integer N (depending only on ε) such that
sup {dY (fn (x), f (x))} < ε, for all n > N.
x∈X

Therefore,
dY (fn (x), f (x)) < ε, for all n > N and all x ∈ X.
Hence, {fn } converges uniformly to f .
Qamrul Hasan Ansari Metric Spaces Page 169

Remark 5.6.2

The above proposition says that the uniform convergence means that the “maximal”
distance between fn and f goes to zero.

Example 5.6.3

Let X = [0, 1] and Y = R with the usual metric. Let {fn } be a sequence of functions
defined by 
1
 2nx,
 if 0 ≤ x ≤ 2n ,
1 1
fn (x) = −2nx + 2, if 2n ≤ x < n ,

 0, if n1 ≤ x ≤ 1,
and f (x) = 0 for all x ∈ R. Then fn → f pointwise on [0, 1], because at every point
x ∈ [0, 1] the value of fn (x) eventually becomes 0 (for x = 0, the value is always 0,
and for x > 0 the “tent” will eventually pass to the left of x). However, fn does
not converge uniformly to f on [0, 1] as the maximum value of all fn is 1, that is,
sup |fn (x) − f (x)| = 1 for all n.
x∈[0,1]

1 −

| |1 R
1
n

Theorem 5.6.2 (Cauchy Criterion). Let {fn } be a sequence of functions defined on a

metric space (X, dX ) to a complete metric space (Y, dY ). Then there exists a continuous
function f : X → Y such that fn → f uniformly on X if and only if the following
condition is satisfied: For every ε > 0, there exists a positive integer N such that

dY (fm (x), fn (x)) < ε, for all m, n > N and all x ∈ X.

Proof. Suppose that fn → f uniformly on X. Then for any given ε > 0, there exists a
positive integer N such that
ε
dY (fn (x), f (x)) < , for all n > N and all x ∈ X.
2
Qamrul Hasan Ansari Metric Spaces Page 170

Thus for all m, n > N and all x ∈ X,

ε ε
dY (fm (x), f (x)) < and dY (fn (x), f (x)) < .
2 2
Hence for all m, n > N and all x ∈ X, we have
ε ε
dY (fm (x), fn (x)) < dY (fm (x), f (x)) + dY (fn (x), f (x)) < + = ε.
2 2

Conversely, assume that

dY (fm (x), fn (x)) < ε, for all m, n > N and all x ∈ X.

Then for each x ∈ X, the sequence {fn (x)} is a Cauchy sequence in a complete metric space
Y , and therefore, converges to some point f (x), say. We claim that fn → f uniformly on X.
If ε > 0 is given, we can choose N such that for all n > N, we have
ε
dY (fn (x), fn+k (x)) < , for all k = 1, 2, . . . and all x ∈ X.
2
Letting k → ∞, we obtain
ε
dY (fn (x), f (x)) = lim dY (fn (x), fn+k (x)) < .
n→∞ 2
Hence,
dY (fn (x), f (x)) < ε, for all n > N and all x ∈ X,
and so, fn → f uniformly on X.

It is known that the uniform convergence implies pointwise convergence. However, the
following theorem, known as Dini’s theorem, provides the conditions under which converse
holds.

Theorem 5.6.3 (Dini’s Theorem). Let (X, d) be a compact metric space, f : X → R

be a continuous function and fn : X → R, n ∈ N, be a decreasing (fn+1 (x) ≤ fn (x) for
all n ∈ N and all x ∈ X) sequence of continuous functions. If {fn } converges pointwise
to f , then the convergence is uniform.

Proof. For each n ∈ N, define gn (x) = fn (x) − f (x) for all x ∈ X. Then {gn } is a sequence
of continuous functions on the compact metric space X that converges pointwise to 0. Also,
for each n ∈ N, 0 ≤ gn+1(x) ≤ gn (x) for all x ∈ X. If we show that {gn } converges uniformly
to 0, then {fn } converges uniformly to f .

Let ε > 0 be given. Consider the set

On = {x ∈ X : gn (x) < ε} = gn−1 (−∞, ε) .
Qamrul Hasan Ansari Metric Spaces Page 171

Then for each n ∈ N, On is an open set as it is the inverse image of an open set by a
continuous function. Since {gn } converges pointwise to 0, for each x ∈ X, there exists a
positive integer N such that gn (x) < ε for all n ≥ N. ThisS implies that x ∈ On for all
n ≥ N. Therefore, {On } is an open cover of X, that is, X = ∞ n=1 On . Since X is compact,
there is a finite collection of {On } that also covers X, that is,

X = On1 ∪ On2 ∪ · · · ∪ Onk ,

for some n1 < n2 < · · · < nk . Since 0 ≤ gn+1 (x) ≤ gn (x) for all x ∈ X, we have On ⊆ On+1 ,
and so
X = On1 ∪ On2 ∪ · · · ∪ Onk = Onk .
Write N = nk . Then gN (x) < ε for all x ∈ X. As {gn (x)} is decreasing, we have

|gn (x) − 0| = gn (x) ≤ gN (x) < ε, for all n > N and all x ∈ X.

Hence {gn } converges uniformly to 0.

The following examples show that the hypotheses:

(i) X is compact

(ii) fn are continuous

(iii) f is continuous

are each necessary in the above theorem.

Example 5.6.4

(a) Consider the space X = (0, 1) with the usual metric and the sequence {fn } of
continuous functions on X defined by fn (x) = xn . Then it is well known that X is not
compact, and {fn } decreases pointwise to zero. However, in view of Proposition 5.6.1,
the convergence is not uniform as

sup{fn (x)} = sup{xn : x ∈ X} = 1, for all n ∈ N.

x∈X

(b) Consider the sequence {fn } of functions defined on R by fn (x) = nx . Then {fn } is a
sequence of continuous functions and monotonically converges to 0, but the convergence
is not uniform.
Qamrul Hasan Ansari Metric Spaces Page 172

Example 5.6.5

Consider the sequence {fn } of functions fn : [0, 1] → R defined by

 2
x, if x < 1,
fn (x) =
0, else.

Then {fn } is monotonically decreasing pointwise to 0, and the domain is compact, but
the convergence is not uniform.

Example 5.6.6

Consider the sequence {fn } of functions on R defined by

nx
fn (x) = , for all x ∈ R and all n ∈ N.
1 + n2 x2
Then each fn is continuous and goes to zero pointwise (but not monotonically). If we
take X = [0, 1], then the domain is compact, yet the convergence is not uniform since
fn (1/n) = 1/2 for all n.

Exercise 5.6.1. Let (X, dX ) and (Y, dY ) be metric spaces, {fn } be a sequence of continuous
functions each defined on X to Y and f : X → Y be a function. If {fn } converges uniformly
to f on X, and {xn } is a sequence in X that converges to a point x. Show that {fn (xn )}
converges to f (x). Show also by an example that this is not true in case of pointwise
convergence of {fn }.

Proof. Let ε > 0 be given. Then there exists a positive integer N1 such that
dY (fn (x), f (x)) < ε, for all n ≥ N1 and all x ∈ X.
Since fn → f uniformly, by Theorem 5.6.1, f is continuous. Also since xn → x, we must
have f (xn ) → f (x), that is, there exists a positive integer N2 such that
dY (f (xn ), f (x)) < ε, for all n ≥ N2 .
For n ≥ max{N1 , N2 }, we have
dY (fn (xn ), f (x)) ≤ dY (fn (xn ), f (xn )) + dY (f (xn ), f (x)) < ε.

Exercise 5.6.2. Let X = Y = R with the usual metric. Let {fn } be a sequence of functions
defined by 

 0, if x ≤ 0,

 nx, if 0 ≤ x ≤ n1 ,
fn (x) =

 −nx + 2, if n1 ≤ x ≤ n2 ,


0, if x ≥ n2 ,
Qamrul Hasan Ansari Metric Spaces Page 173

and f (x) = 0 for all x ∈ R. Show that {fn } converges pointwise to f on R, but does not
converge uniformly to f on R.

(0, 1)•

(1, 0)
• •
• R
2
n
,0

( n1 , 0)

Proof. If x is close to 0 and positive, then there exists a positive integer N such that 2/N < x
and so fn (x) = 0 for all n > N. However, there exists no N such that (5.6) holds for ε = 21 ,
for all n > N and for all x ∈ R. For, if such an N were to exist, we would have
1
fn (x) < , for all n > N and all x ∈ R.
2
We would then have Nx < 2 for 0 ≤ x ≤ N1 ; but for x = 32 N, we obtain the contradiction
1
2
3
< 21 .

Since sup |fn (x) − f (x)| = 1 for all n, by Proposition 5.6.1, fn does not converge uniformly
x∈[0,1]
to f on [0, 1] as the maximum value of all fn is 1,
Exercise 5.6.3. Let X = [0, 1] and Y = R with the usual metric. Let {fn } be a sequence
of functions defined by

2 1
 4n x,
 if 0 ≤ x ≤ 2n ,
2 1 1
fn (x) = −4n x + 4n, if 2n < x < n ,

 0, if n1 ≤ x ≤ 1.
Show that the sequence {fn } converges pointwise to the constant function f (x) = 0, but
does not converge uniformly to f on [0, 1].
Exercise 5.6.4. Let X = Y = R with the usual metric. Let {fn } be a sequence of functions
defined by 
 −1,
 if x ≤ − n1 ,
fn (x) = nx, if − n1 < x < n1 ,

 1, if n1 ≤ x,
and f be a function defined by

 −1,
 if x < 0,
f (x) = 0, if x = 0,

 1, if x > 0.
Qamrul Hasan Ansari Metric Spaces Page 174

Show that the sequence {fn } converges pointwise to the function f , but not uniformly.

Hint: Note that {fn } is a sequence of continuous functions and converges pointwise to f
which is not continuous. Use Theorem 5.6.1.
Exercise 5.6.5. Prove that the sequence {fn } defined by
fn (x) = tan−1 (nx), for all x ≥ 0,
is uniformly convergent on [α, ∞) when α > 0, but is not uniformly convergent on [0, ∞).

Proof. The pointwise limit function is

 π
2
, if x > 0,
f (x) = lim fn (x) =
n→∞ 0, if x = 0.
We show that fn → f uniformly on [α, ∞) when α > 0.

For x > 0, we claim that

π
|fn (x) − f (x)| = tan−1 (nx) − = cot−1 (nx).
2
Since 0 < tan−1 θ < π/2 for any θ > 0, therefore, when x > 0, we have 0 < tan−1 (nx) < π/2,
and hence,
π π
0 < − tan−1 (nx) < . (5.9)
2 2
Also, π 
cot − tan−1 (nx) = nx. (5.10)
2
It follows from (5.9) and (5.10) that π/2 − tan−1 (nx) = cot−1 (nx). The first inequality in
(5.9) implies that
π π
tan−1 (nx) − = − tan−1 (nx), for all x > 0.
2 2
Thus,
π
tan−1 (nx) − = cot−1 (nx).
2

Let ε > 0 be arbitrary. When x ≥ α, the inequality n > cotα ε implies n > cotx ε , so that
nx > cot ε, and hence, cot−1 nx < ε in view of the fact that cot−1 is a decreasing function.
It follows that if N is an integer greater than or equal to cotα ε , then
π
|fn (x) − f (x)| = tan−1 (nx) − = cot−1 nx < ε
2
whenever n > N and x ≥ α. However, cotx ε → ∞ as n → 0, so that no integer N exists for
which
|fn (x) − f (x)| < ε for all n > N and all x ∈ [0, ∞).
Actually this proves that the convergence fails to be uniform even on the smaller set (0, ∞).
Qamrul Hasan Ansari Metric Spaces Page 175

Exercise 5.6.6. Consider the sequence {fn } of functions defined by

1
fn (x) = , for all x ∈ [0, 1].
1 + xn

(a) Find f (x) = lim fn (x).

n→∞

(b) Show that for 0 < a < 1, {fn } converges uniformly to f on [0, a].
(c) Show that {fn } does not converge uniformly to f on [0, 1].

Proof. (a) Since fn (1) = 1/2 for all n ∈ N, we obtain f (1) = 1/2 and
1
lim = 1, for 0 ≤ x < 1,
n→∞ 1 + xn

which yields f (x) = 1 for 0 ≤ x < 1.

(b) If x ∈ [0, a], then

1 xn an
− 1 = ≤ .
1 + xn 1 + xn 1 + an
Since 0 < a < 1,
an
lim
= 0,
n→∞ 1 + an

so {fn } converges uniformly to f on [0, a].

r
1 1
< x < 1, then < xn < 1. Therefore,
n
(c) Given n ∈ N, let x be such that
2 2
1
1 xn 2 1
n
−1 = n
> = ,
1+x 1+x 1+1 4
which implies {fn } does not converge uniformly to f on [0, 1].
Exercise 5.6.7. Let {fn } be a sequence of functions fn : R → R defined by

1 − n1 |x|, if |x| < n,
fn (x) =
0, if |x| ≥ n.
Prove that {fn } converges uniformly on each compact subset of R to the function f (x) = 1.

Proof. Let E be any compact subset of R and 0 < ε < 1. Since E is compact, it is bounded,
and therefore, E ⊆ (M − M) for M > 0. Now there exists a positive integer N such that
N > M/ε, that is, M/N < ε. Thus, for n > N, we have
1 M
|fn (x) − f (x)| = |x| < < ε, for all x ∈ E.
n N
Hence {fn } converges uniformly to f on E.
Qamrul Hasan Ansari Metric Spaces Page 176

Exercise 5.6.8. Consider the sequence {fn } of functions defined by

( √
1
n
n2 − x2 , if |x| < n,
fn (x) =
0, if |x| ≥ n.

(a) Show that {fn } does not converge uniformly to the constant function f (x) = 1.

(b) Show that {fn } converges uniformly on every compact subset of R to the constant
function f (x) = 1.
Exercise 5.6.9. Let {fn } be a sequence of real-valued functions defined on [0, 1] by fn (x) =
nx(1 − x)n . Show that {fn } converges pointwise to the constant function f (x) = 0, but not
uniformly.
Exercise 5.6.10. Let {fn } be a sequence of functions fn : R → R defined by fn (x) = n+1
n
x.
Show that {fn } converges uniformly on each compact subset of R to the function f (x) = x,
but not on the whole space R.
Exercise 5.6.11. Let {fn } be a sequence of functions defined by
nx
fn (x) = , for all x ∈ [0, 2].
enx

(a) Show that lim fn (x) = 0 for x ∈ (0, 2].

n→∞

(b) Show that the convergence is not uniform on [0, 2].

Proof. (a) Since

fn+1 (x) (n + 1)xenx (n + 1) −x
= (n+1)x
= e ,
fn (x) nxe n
fn+1 (x)
we have lim = e−x < 1. Thus lim fn (x) = 0 for x ∈ (0, 2].
n→∞ fn (x) n→∞

nenx − nenx nx
(b) Let us find the maximum of fn in [0, 2]. Since fn′ (x) = = 0, we have
(enx )2
1
enx (n − n2 x) = 0 ⇔ n2 x = n ⇔ x = .
n
 
1 1 1
It can be easily seen that is the maximum with fn
n
= . Since
n e
 
1 1
lim sup {|fn (x) − 0| : x ∈ [0, 2]} = lim fn = ,
n→∞ n→∞ n e

the convergence is not uniform on [0, 2].

Qamrul Hasan Ansari Metric Spaces Page 177

x
Exercise 5.6.12. Show that the sequence {fn } defined by fn (x) = x+n
does not converge
uniformly on [0, ∞).

x
Proof. We can easily see that the sequence {fn } defined by fn (x) = x+n on [0, ∞) converges
pointwise to a function f (x) = 0. To show that the convergence is not uniform, we consider
x
sup |fn (x) − f (x)| = sup |fn (x)| = sup .
x≥0 x≥0 x≥0 x+n
n
Since f ′ (x) = (x+n)2
is positive, so fn (x) is increasing and

x x
sup |fn (x) − f (x)| = sup = lim = 1.
x≥0 x≥0 x + n x→∞ x + n

This expression does not go to zero as n → ∞, so the convergence is not uniform.

x
Exercise 5.6.13. Show that the sequence {fn } defined by fn (x) = x2 +n2
does not converge
uniformly on [0, ∞).

x
Proof. We can easily see that the sequence {fn } defined by fn (x) = x2 +n 2 on [0, ∞) converges

pointwise to a function f (x) = 0. To show that the convergence is uniform, we need to show
that
x
sup |fn (x) − f (x)| = sup |fn (x)| = sup 2 2
x≥0 x≥0 x≥0 x + n

x2 +n2 −2x2 n2 −x2

converges to zero as n → ∞. Since f ′ (x) = (x2 +n2 )2
= (x2 +n2 )2
, fn (x) is increasing when
x < n and decreasing when x > n.
x x
sup |fn (x) − f (x)| = sup = lim = 1.
x≥0 x≥0 x + n x→∞ x + n

This expression does not go to zero as n → ∞, so the convergence is not uniform.

Exercise 5.6.14. Let fn : R → R be defined by fn (x) = nx . Show that the sequence {fn }
converges pointwise, not uniformly to f (x) = 0.

ne
Exercise 5.6.15. Let fn : [0, ∞) → R be defined by fn (x) = e−x nx . Show that the
sequence {fn } converges pointwise. Find the maximum value of fn . Does {fn } converge
uniformly?


Exercise 5.6.16. Let fn : (0, ∞) → R be defined by fn (x) = n x1/n − 1 . Show that the
sequence {fn } converges

pointwise to f (x) = ln x. Also show that the convergence is uniform
1
on each interval n , n , n ∈ N, but not on (0, ∞).

Exercise 5.6.17. Let fn , gn : X → R be functions from a metric space X to R. If {fn }

and {gn } converge uniformly to f and g, respectively, then prove that {fn + gn } converges
uniformly to f + g.
Qamrul Hasan Ansari Metric Spaces Page 178

Exercise 5.6.18. Let fn : [a, b] → R be continuous functions such that {fn } converges
uniformly to a function f : [a, b] → R. Prove that
Z b Z b
fn (x)dx → f (x)dx.
a a

Show by an example that this is not true in case of pointwise convergence.

Proof. Since {fn } converges uniformly to a function f , for any ε > 0, there exists a positive
integer N such that
ε
|fn (x) − f (x)| < , for all n > N and all x ∈ [a, b].
b−a
Therefore, if n > N, then we have
Z b Z b Z b
fn (x)dx − g(x)dx = (fn (x) − g(x)) dx
a a a
Z b
≤ |fn (x) − g(x)| dx
a
Z b 
ε
< dx = ε.
a b−a

Exercise 5.6.19. Let fn : R → R be defined by fn (x) = n1 sin(nx). Prove that {fn }

converges uniformly to f (x) = 0, but the sequence {fn′ } of derivatives does not converge.

Exercise 5.6.20. Let

ℓ∞ = {x = {xn }n≥1 : xn ∈ C, sup |xn | < ∞} with the metric d∞ (x, y) = sup |xn − yn |,
1≤n<∞

( ∞
) ∞
!1/2
X X
ℓ2 = x = {xn }n≥1 : xn ∈ C, |xn |2 < ∞ with the metric d2 (x, y) = |xn − yn |2 .
n=1 n=1
 x x
Prove that the mapping T : ℓ∞ → ℓ2 defined by T (x) = x1 , 22 , 33 , . . . is linear and
uniformly continuous.
Qamrul Hasan Ansari Metric Spaces Page 179

5.7 The Spaces B(X, Y ) and C(X, Y )

Definition 5.7.1

Let (X, dX ) and (Y, dY ) be metric spaces. A function f : X → Y is said to be bounded

if f (X) is bounded in Y . In other words, there exists a constant M ∈ R such that
dY (y1 , y2 ) ≤ M for all y1 , y2 ∈ f (X).

The set of all bounded functions f : X → Y is denoted by B(X, Y ).

Example 5.7.1

Every continuous function defined on a compact metric space is bounded.

Indeed, if X is a compact metric space and f : X → Y is a continuous function, then

f (X) is compact and therefore bounded, so f is bounded.

Exercise 5.7.1. Let (X, dX ) and (Y, dY ) be metric spaces, {fn } be a sequence of bounded
functions each defined on X to Y and f : X → Y be a function. If {fn } converges uniformly
to f on X, then prove that f is bounded on X.

Proof. By the uniform convergence, there exists a positive integer N such that

dY (fn (x), f (x)) ≤ 1, for all x ∈ X and all n > N.

Since fn is bounded, there exist y ∈ Y and M ∈ R such that

dY (fn (x), y) ≤ M, for all x ∈ X and all n ∈ N.

It follows that

dY (f (x), y) ≤ dY (f (x), fn (x)) + dY (fn (x), y) ≤ M + 1, for all x ∈ X and all n > N,

that is, f is bounded.

Let (X, dX ) and (Y, dY ) be metric spaces. Define a mapping ρ : (X, dX ) → (Y, dY ) by

ρ(f, g) = sup{dY (f (x), g(x)) : x ∈ X}, for all f, g ∈ B(X, Y ).

Then ρ forms a metric on B(X, Y ), and therefore, (B(X, Y ), ρ) is a metric space.

Qamrul Hasan Ansari Metric Spaces Page 180

Theorem 5.7.1. Let (X, dX ) and (Y, dY ) be metric spaces and {fn } be a sequence of
functions in B(X, Y ). Then {fn } converges to f ∈ B(X, Y ) with respect to the metric
ρ if and only if {fn } converges uniformly to f .

Proof. Let ε > 0. Assume that {fn } converges to f ∈ B(X, Y ) with respect to the metric ρ,
then there exists a positive integer N such that

ρ(fn , f ) < ε, for all n > N.

Therefore, for n > N, we have

dY (fn (x), f (x)) ≤ sup{dY (fn (x), f (x)) : x ∈ X} = ρ(fn , f ) < ε, for all x ∈ X.

Hence, {fn } converges uniformly to f .

Conversely, let ε > 0 and suppose that {fn } converges uniformly to f . Then there exists a
positive integer N such that
ε
dY (fn (x), f (x)) < , for all x ∈ X and all n > N.
2
Therefore, for n > N, we have
ε
ρ(fn , f ) = sup{dY (fn (x), f (x)) : x ∈ X} ≤ < ε, for all x ∈ X.
2
Hence, {fn } converges to f ∈ B(X, Y ) with respect to the metric ρ.

Let (X, dX ) and (Y, dY ) be metric spaces and C(X, Y ) denote the set of all continuous
functions f : X → Y from X to Y . We further assume that X is compact. Define ρ :
C(X, Y ) × C(X, Y ) → [0, ∞) by

ρ(f, g) = sup{dY (f (x), g(y)) : x ∈ X}, for all f, g ∈ C(X, Y ). (5.11)

To prove that ρ is a metric on C(X, Y ), we first establish the following lemma.

Lemma 5.7.1. Let (X, dX ) and (Y, dY ) be metric spaces, and assume further that X is
compact. Then for any f, g ∈ C(X, Y ),

ρ(f, g) = sup{dY (f (x), g(y)) : x ∈ X}

is finite, and there is an element x ∈ X such that dY (f (x), g(x)) = ρ(f, g).

Proof. If we prove that the function

h(x) := dY (f (x), g(x))

Qamrul Hasan Ansari Metric Spaces Page 181

is continuous, then by Extreme Value Theorem (Exercise 5.3.4) the result follows.

By the triangle inequality, we have

dY (f (x), g(x)) ≤ dY (f (x), f (y)) + dY (f (y), g(y)) + dY (g(y), g(x)),
that is,
dY (f (x), g(x)) − dY (f (y), g(y)) ≤ dY (f (x), f (y)) + dY (g(y), g(x)).
By symmetry, we also have
dY (f (y), g(y)) − dY (f (x), g(x)) ≤ dY (f (x), f (y)) + dY (g(y), g(x)),
and therefore,
|h(x) − h(y)| = |dY (f (x), g(x)) − dY (f (y), g(y))|
≤ dY (f (x), f (y)) + dY (g(y), g(x)).
Since f and g are continuous, for any given ε > 0, there exists δ > 0 such that
ε ε
dY (f (x), f (y)) < and dY (g(x), g(y)) < ,
2 2
whenever dX (x, y) < δ. But then
ε ε
|h(x) − h(y)| ≤ dY (f (x), f (y)) + dY (g(y), g(x)) < + = ε,
2 2
whenever dX (x, y) < δ. Therefore, h is continuous.

We are ready to prove that ρ is a metric on C(X, Y ).

Theorem 5.7.2. Let (X, dX ) and (Y, dY ) be metric spaces, and assume further that X
is compact. Then,

ρ(f, g) = sup{dY (f (x), g(y)) : x ∈ X}, for any f, g ∈ C(X, Y )

is a metric on C(X, Y ).

Proof. By Lemma 5.7.1, ρ(f, g) ≥ 0 for all f, g ∈ C(X, Y ). Since other properties of a metric
are obvious, we only concentrate on the triangle inequality.

Let f, g, h ∈ C(X, Y ). By Lemma 5.7.1, there is a point x ∈ X such that ρ(f, g) =

dY (f (x), g(x)). But then
ρ(f, g) = dY (f (x), g(x))
≤ dY (f (x), h(x)) + dY (h(x), g(x))
≤ ρ(f, h) + ρ(h, g),
and the result follows.
Qamrul Hasan Ansari Metric Spaces Page 182

Theorem 5.7.3. Let (X, dX ) be a compact metric space and (Y, dY ) be a complete
metric space. Then (C(X, Y ), ρ) is complete.

Proof. Let {fn } be a Cauchy sequence in the metric space (C(X, Y ), ρ) and x ∈ X be any
fixed element. Since dY (fn (x), fm (x)) ≤ ρ(fn , fm ), the function values {fn (x)} is a Cauchy
sequence in the complete metric space (Y, dY ). Therefore, {fn (x)} converges to a point f (x)
in Y . This means that {fn } converges pointwise to a function f : X → Y . We prove that
f ∈ C(X, Y ) and that {fn } converges to f with respect to the metric ρ.

Since {fn } is a Cauchy sequence, for any ε > 0, there exists a positive integer N such that
ρ(fn , fm ) < 2ε whenever n, m > N, that is, for all x ∈ X and all n, m > N, dY (fn (x), fm (x)) <
ε
2
. Taking m → ∞. Then for all x ∈ X and all n > N, we have
ε
dY (fn (x), f (x)) = lim dY (fn (x), fm (x)) ≤ < ε,
m→∞ 2
and hence {fn } converges uniformly to f . By Theorem 5.6.1, f is continuous and belongs
to C(X, Y ). Finally, by Proposition 5.7.1, {fn } converges to f in (C(X, Y ), ρ).

The following proposition shows that the convergence in C(X, Y ) is exactly the same as
uniform convergence.

Proposition 5.7.1. Let (X, dX ) and (Y, dY ) be metric spaces, and assume further that
X is compact. Let {fn } be a sequence of functions in the metric space (C(X, Y ), ρ). Then
{fn } converges to f with respect to the metric ρ if and only if it converges uniformly to
f.

Proof. By Proposition 5.6.1, {fn } converges uniformly to f if and only if

sup {dY (fn (x), f (x))} → 0 as n → ∞,

x∈X

that is, ρ(fn , f ) → 0 which means that {fn } converges to f in (C(X, Y ), ρ).
Qamrul Hasan Ansari Metric Spaces Page 183

5.8 Equicontinuity and the Arzelà-Ascoli Theorem

It is known that a subset of Rn is compact if and only if it is closed and bounded. This
characterization is very much useful to check the compactness of a set in Rn . Does a closed
and bounded subset of C(X, Rn ) compact? The answer is “no”. However, a subset of
C(X, Rn ) is compact if and only if it is closed, bounded and equicontinuous. Such a result is
known as the Arzelà-Ascoli theorem.

Definition 5.8.1

Let X be a set. The collection B of functions fn : X → R is said to be pointwise

bounded if for every x ∈ X, there exists a constant Mx ∈ R such that

|fn (x)| ≤ Mx , for all n ∈ N.

The collection B of functions fn : X → R is said to be uniformly bounded (or simply

bounded) if there exists a constant M ∈ R such that

|fn (x)| ≤ M, for all n ∈ N and all x ∈ X.

Remark 5.8.1

In the above definition of pointwise bounded collection of functions, Mx depends on the

point x. It may vary from point to point. For example, consider F = {f : R → R :
f (x) = αx for α ∈ [−1, 1]} with Mx = |x|.

Remark 5.8.2

If X is a compact metric space, then the uniform boundedness is same as boundedness

in the metric space (C(X, R), ρ), where ρ is defined by

ρ(f, g) = sup{|f (x) − g(y)| : x ∈ X}, for all f, g ∈ C(X, R).

Qamrul Hasan Ansari Metric Spaces Page 184

Does every (uniformly) bounded sequence of functions has a convergent subsequence? In

general, the answer is “no”.

Example 5.8.1

Consider the sequence {fn } of continuous functions defined by fn (x) = xn on [0, 1].
Then {fn } is uniformly bounded, but has no subsequence that converges uniformly,
although the sequence converges pointwise (to a discontinuous function).

Example 5.8.2

n x 3
Let {fn } be a sequence of functions in C([0, 1], R) defined by fn (x) = 1+n 4 x2 . Then {fn }

converges pointwise to the zero function f (f (x) = 0 for all x) as fn (x) = 0 at x = 0

n3 x 1
and for x > 0, we have 1+n 4 x2 ≤ nx . As for each x, {fn (x)} converges, so it is bounded.

Therefore, {fn } is pointwise bounded.

We maximize fn over [0, 1], and we find the maximum occurs at the critical point
x = 1/n2 . Since ρ(fn , f ) = sup{|fn (x) − f (x)| : x ∈ [0, 1]} = fn (1/n2 ) = n/2,
limn→∞ ρ(fn , f ) = ∞. Therefore, the sequence {fn } is not uniformly bounded.

Proposition 5.8.1. Let (X, d) be a metric space and F be a pointwise bounded family
of continuous functions f : X → R. Then there exist a nonempty open set G and a
constant M ∈ R such that |f (x)| ≤ M for all f ∈ F and all x ∈ G.

Proof. For all n ∈ N and all f ∈ F , the set f −1 ([−n, n]) is closed as it is the inverse image
of a closed set under continuous function. Then the set
\
An = f −1 ([−n, n])
f ∈F
S
is closed as it is intersection of closed sets. Since F is pointwise bounded, R = n∈N An . By
Corollary 3.3.3, all An can not be nowhere dense. If An0 is not nowhere dense, then there
exists an open set G such that every open sphere inside G contains elements from An0 . Since
An0 is closed, G ⊆ An0 . Then by the definition of An0 , we have |f (x)| ≤ n0 for all f ∈ F
and all x ∈ G.

Proposition 5.8.2. Let X be a countable set and {fn } be a pointwise bounded sequence
of functions fn : X → R. Then {fn } has a pointwise convergent subsequence.

Proof. Let x1 , x2 , x3 , . . . be an enumeration of the elements of X. The sequence {fn (x1 )}∞
n=1 is
bounded, and hence we have a subsequence of {fn }, which we denote by {f1,k }∞ k=1 , such that
Qamrul Hasan Ansari Metric Spaces Page 185

{f1,k (x1 )}∞ ∞ ∞

k=1 converges. Next, {f1,k (x2 )}k=1 is bounded, and so {f1,k }k=1 has a subsequence
{f2,k }∞ ∞ ∞
k=1 such that{f2,k (x2 )}k=1 converges. Note that {f2,k (x1 )}k=1 is still convergent. In
general, we have a sequence {fm,k }∞ ∞
k=1 , which is a subsequence of {fm−1,k }k=1 , such that
{fm,k (xj )}k=1 converges for j = 1, 2, . . . , m. Let {fm+1,k }k=1 be a subsequence of {fm,k }∞
∞ ∞
k=1
such that {fm+1,k (xm+1 )}∞k=1 converges (and hence it converges for all xj for j = 1, 2, . . . , m).
Rinse and repeat.

If X is finite, we are done as the process stops at some point. If X is countably infinite,
we pick the sequence {fk,k }∞ ∞
k=1 . This is a subsequence of the original sequence {fn }n=1 . For
∞ ∞
every m, the tail {fk,k }k=m is a subsequence of {fm,k }k=1, and hence for any m the sequence
{fk,k (xm }∞
k=1 converges.

Definition 5.8.2

Let (X, dX ) and (Y, dY ) be metric spaces. The collection F of functions f : X → Y is

said to be equicontinuous if for every ε > 0, there exists δ > 0 such that for all f ∈ F
and all x, y ∈ X,
dX (x, y) < δ implies dY (f (x), f (y)) < ε.

Note that the same δ works for all points x, y ∈ X as well as for all functions f ∈ F.

Example 5.8.3

Let R be a metric space with the usual metric, and {fn } be a sequence of functions
defined by 
0, if x < n,
fn (x) =
x − n, if n ≤ x.
Then the collection {fn } is equicontinuous.

Example 5.8.4

Let R be a metric space with the usual metric, and {fn } be a sequence of functions
defined on [0, 1] by fn (x) = xn for all x ∈ [0, 1]. Then the collection {fn } is not
equicontinuous.
1
Indeed, let xn = 1 − n
and yn = 1. Then, |xn − yn | = n1 → 0 as n → ∞. Also,
 n
1 n→∞
|fn (yn ) − fn (xn )| = 1 − 1 − −→ 1 − e−1 > 0.
n

So if we take ε > 0 such that ε < 1 − e−1 (for instance we can take ε = (1 − e−1 )/2),
Qamrul Hasan Ansari Metric Spaces Page 186

then there exists a positive integer N such that

|fn (yn ) − fn (xn )| > ε, for all n ≥ N.

Therefore, {fn } is not an equicontinuous collection.

Proposition 5.8.3. Let X be a compact metric space and {fn } be a sequence in C(X, R)
that converges uniformly. Then {fn } is equicontinuous.

Proof. Let ε > 0 be given. Since {fn } converges uniformly, there exists a positive integer N
such that
ε
|fn (x) − fN (x)| < , for all n > N and all x ∈ X.
3
Now {f1 , f2 , . . . , fN } is a finite set of uniformly continuous functions and so an equicontinuous
family. Therefore, there exists a δ > 0 such that
ε
|fj (x) − fj (y)| < < ε, whenever d(x, y) < δ and 1 ≤ j ≤ N.
3
Now take n > N. Then for d(x, y) < δ, we have
|fn (x) − fn (y)| ≤ |fn (x) − fN (x)| + |fN (x) − fN (y)| + |fN (y) − fn (y)|
ε ε ε
< + + = ε.
3 3 3
This completes the proof.

Note that a uniformly equicontinuous sequence in the metric space C(X, R) that is pointwise
bounded is bounded in C(X, R) and furthermore it contains a convergent subsequence in
C(X, R).

We now present Arzelà-Ascoli theorem about the existence of convergent subsequences.

Theorem 5.8.1. Let (X, d) be a compact metric space, and {fn } be a sequence of
functions fn ∈ C(X, R) which is pointwise bounded and equicontinuous. Then {fn } is
uniformly bounded and it contains a uniformly convergent subsequence.

Proof. We first prove that the sequence {fn } is uniformly bounded.

By equicontinuity, there exists δ > 0 such that

Sδ (x) ⊆ fn−1 S1 (fn (x)) , for all x ∈ X and all n ∈ N.

Since
Sk the space X is compact, there exists a finite set {x1 , x2 , . . . , xk } such that X =
j=1 δ (xj ). As {fn } is pointwise bounded, there exist M1 , M2 , . . . , Mk such that
S

|fn (xj | ≤ Mj , for all n and for j = 1, 2, . . . , k.

Qamrul Hasan Ansari Metric Spaces Page 187

Let M = 1 + max{M1 , M2 , . . . , Mk }. Now for any given x ∈ X, there is a j such that
−1


x ∈ Sδ (xj ), and therefore, x ∈ fn S1 fn (xj ) for all n, that is, |fn (x) − fn (xj )| < 1 for all
n. By reverse triangle inequality, we have

|fn (x)| < 1 + |fn (xj | ≤ 1 + Mj ≤ M, for all n.

Since x was arbitrary, {fn } is uniformly bounded.

Now we shall prove that {fn } contains a convergent subsequence. Pick a countable dense
subset D of X. Then by Proposition 5.8.4, we can find a subsequence {fnj } that converges
pointwise on D. For simplicity, we write gj := fnj . Note that the sequence {gn } is equicon-
tinuous.

Let ε > 0 be given. Then there exists δ > 0 such that



Sδ (x) ⊆ gn−1 Sε/3 gn (x) , for all x ∈ X and all n ∈ N.

Since D is dense, every x ∈ X lies in Sδ (y) for some y ∈SD. By compactness of X, there
exists a finite set {x1 , x2 , . . . , xk } of D such that X = kj=1 Sδ (xj ). As there are finitely
many points and {gn } converges pointwise on D, there exists a single N such that
ε
|gn (xj ) − gm (xj )| < , for all n, m > N and all j = 1, 2, . . . , k.
3

Let x ∈ X be arbitrary. There is some j such that x ∈ Sδ (xj ) and so for all i ∈ N,
ε
|gi (x) − gi (xj )| < .
3
Therefore, for all n, m > N, we have

|gn (x) − gm (x)| ≤ |gn (x) − gn (xj )| + |gn (xj ) − gm (xj )| + |gm(xj ) − gm (x)|
ε ε ε
< + + = ε.
3 3 3
Hence, the sequence is Cauchy, and by completeness of R, it is uniformly convergent.

Corollary 5.8.1. Let (X, d) be a compact metric space and D ⊆ C(X, R) be a closed,
bounded and equicontinuous set. Then D is compact.

Proposition 5.8.4. Let (X, dX ) be a compact metric space and {fn } be a bounded
and equicontinuous sequence in C(X, Rn ). Then {fn } has a subsequence converging in
C(X, Rn ).
Qamrul Hasan Ansari Metric Spaces Page 188

Proof. Since X is compact, by Theorem 4.3.1, it is totally bounded, and then by Theorem
4.3.3, it is separable. Therefore, X has a countable dense subset A = {a1 , a2 , . . . , an , . . .}.
Since {fn } is bounded and A is countable, by Proposition 5.8.4, we can find a subsequence
{gk } of {fn } such that {gk (a)} converges for all a ∈ A.

Since C(X, Rn ) is complete, it suffices to prove that {gk } is a Cauchy sequence in C(X, Rn ),
that is, for any given ε > 0, we have to find a positive integer N such that ρ(gn , gm ) =
sup{|gn (x) − gm (x)| : x ∈ X} < ε whenever n, m ≥ N.

Since {fn } is equicontinuous, there exists δ > 0 such that

ε
dR (fn (x), fn (y)) < , for all n,
4
whenever dX (x, y) < δ. Since {gk } is a subsequence of {fn }, we clearly have
ε
dR (gk (x), gk (y)) < , for all k.
4
Choose a finite subset Aδ of A such that any element in X is within less than δ of an element
in Aδ . Since the sequences {gk (a)}, a ∈ Aδ , converge, they are all Cauchy sequences, and so
we can find a positive integer N such that
ε
dRn (gn (a), gm (a)) < , for all a ∈ Aδ ,
4
whenever n, m > N (here we are using that Aδ is finite).

For any x ∈ X, we can find an a ∈ Aδ such that dX (x, a) < δ. But then for all n, m > N,
we have

dRn (gn (x), gm (x)) ≤ dRn (gn (a), gn (a)) + dRn (gn (a), gm (a)) + dRn (gm (a), gm (x))
ε ε ε 3ε
< + + = .
4 4 4 4
Since above inequality holds for all x ∈ X, we must have
3ε
ρ(gn , gm ) < < ε, for all n, m > N,
4
and hence, {gk } is a Cauchy sequence in a complete metric space, and thus converges.

It is remain to prove that the original sequence {fn } really has a subsequence {gk } such that
{gk (a)} converges for
n allo a ∈ A. Wen begin aolittle less ambitiously by showing that {fn }
(1) (1)
has a subsequence fn such that fn (a1 ) converges (recall that a1 is the first element
n o n o
(1) (2)
in our listing of countable set A). Next we show that fn has a subsequence fn
n o n o
(2) (2)
such that both fn (a1 ) and fn (a2 ) converges. Continuing taking subsequences in
n o n o
(j) (j)
this way, for each j ∈ N, we find a subsequence fn such that fn (a) converges for
Qamrul Hasan Ansari Metric Spaces Page 189

n = ao1 , a2 , . . . , aj . Finally, we construct the sequence {gk } by combining all the sequences
a
(j)
fn in a clever way.
n o
(1)
Let us construct . Since the sequence {fn } is bounded, {fn (a1 )} is a bounded sequence
fn
n
in Rn, ando by Bolzano-Weierstrass Theorem 4.2.1, it has a convergent subsequence {f n nk (ao1 )}.
(1) (1)
Let fn consist of the functions appearing in this subsequence. If we now apply fn to
n o
(1)
a2 , we get a new bounded sequence fn (a2 ) in Rn with a convergent subsequence. We let
n o n o
(2) (2)
fn be the functions appearing in this subsequence. Note that fn (a1 ) still converges
n o n o
(2) (1)
as fn is a subsequence of fn . Continuing in this way, we see that for each j ∈ N,
n o n o
(j) (j)
we have a sequence fn such that fn (a) converges for a = a1 , a2 , . . . , aj . In addition,
n o
(j)
each sequence fn is a subsequence of the previous ones.

We are now ready to construct a sequence {gk } such that {gk (a)} converges for all a ∈ A.
We do
n o it by a diagonal argument, putting g1 equal to the first element
n o in the first sequence
(1) (2)
fn , g2 equal to the second element in the second sequence fn , etc. In general, the
n o
(k)
k-th term in the g-sequence equals the k-th term in the k-th f -sequence fn , that is,
(k)
gnk = o fk . Note that except for the first few elements, {gk } is a subsequence of any sequence
(j)
fn . This mean that {gk (a)} converges for all a ∈ A, and the proof is complete.

As a simple consequence of the above result, we have the following result.

Corollary 5.8.2. Let (X, d) be a compact metric space. Then every closed, bounded
and equicontinuous set K in C(X, Rn ) is compact.

Proof. In view of Proposition 5.8.4, any sequence in K has a convergent subsequence. Since
K is closed, the limit must be in K, and hence K is compact.

We now finally ready to state and prove the following Arzelà-Ascoli’s theorem.

Theorem 5.8.2 (Arzelà-Ascoli’s Theorem). Let (X, dX ) be a compact metric space. A

subset K of C(X, Rn ) is compact if and only if it is closed, bounded and equicontinuous.

Proof. In view of Corollary 5.8.2, it remains to show that the compact set K is closed,
bounded and equicontinuous. Since K is compact, it is closed and bounded. So we only left
to prove the equicontinuity of K.
Qamrul Hasan Ansari Metric Spaces Page 190

Suppose that K is not equicontinuous. Then there exists an ε > 0 such that for any δ > 0,
there is a function f ∈ K and points x, y ∈ X such that

dX (x, y) < δ but dRn (f (x), f (y)) ≥ ε.

Since K is compact, there is a subsequence {fnk } of {fn } which converges (uniformly) to a

function f ∈ K. Also,
n since
o X is compact, the corresponding subsequence
 {xnk } of {xn },
has a subsequence xnkj converging to a point a ∈ X. Since dX xnkj , ynkj < n1k , the
n o j

corresponding sequence ynkj of y’s also converges to a.

n o n o n o
Since fnkj converges to f , and xnkj , ynkj both converge to a, by Exercise 5.6.1, we
have    
fnkj xnkj → f (a) and fnkj ynkj → f (a).
    
But this is not possible as dRn f xnkj , f ynkj ≥ ε for all j. So we reach to a contra-
diction and the proof is complete.

Exercise 5.8.1. Let {fn } be a sequence of functions fn : [−1, 1] → R defined by fn (x) :=

nx
1+n2 x2
. Prove that the sequence {fn } is uniformly bounded, converges pointwise to 0, but
there is no subsequence that converges uniformly. Which hypothesis of Arzelà-Ascoli Theo-
rem is not satisfied? Prove your assertion.
1
Exercise 5.8.2. Let {fn } be a sequence of functions fn : R → R defined by fn (x) := (x−n)2 +1 .

Prove that this sequence {fn } is uniformly bounded, equicontinuous, converges pointwise to
zero, but there is no subsequence that converges uniformly. Which hypothesis of Arzelà-
Ascoli Theorem is not satisfied? Prove your assertion.

Exercise 5.8.3. If {fn } is an equicontinuous family of functions fn : [0, 1] → R, and

g : [0, 1] → R is continuous, then prove that {fn + g} is equicontinuous.

Exercise 5.8.4. Prove that a compact subset K of C([0, 1], R) is nowhere dense.

Proof. Since every compact set is closed, in view of Exercise 3.3.3, it suffices to show that
each sphere Sε (f ) contains elements that are not in K. By Arzela-Ascoli’s Theorem 5.8.2,
compact sets are equicontinuous and hence we need only to prove that Sε (f ) contains a
family of functions that is not equicontinuous.

For each n ∈ N, define gn by

 ε
nx, for x ≤ 2n
gn (x) = ε ε
2
, for x ≤ 2n .

Then gn is not equicontinuous and f + gn is in Sε (f ), but since {f + gn } is not equicontinuous

(it can be prove by using Exercise 5.8.3), all these functions cannot be in K, and hence Sε (f )
contains elements that are not in K.
Qamrul Hasan Ansari Metric Spaces Page 191

Exercise 5.8.5. Prove that the space C([0, 1], R) is not a countable union of compact sets.

Proof. Since C([0, 1], R) is complete, by Baire’s Category Theorem 3.3.2, it cannot be a
countable union of nowhere dense sets. By Exercise 5.8.4, we obtain the desired result.

Exercise 5.8.6. Let (X, d) be a compact metric space and {fn } be a equicontinuous sequence
of functions in C(X, R). If {fn } converges pointwise, then prove that it converges uniformly.

Exercise 5.8.7. Let {fn } be an equicontinuous and uniformly bounded sequence of 2π-
periodic functions fn : R → R. Show that there is a uniformly convergent subsequence.
 Chapter A

Some Basic Inequalities

x
Proposition A.0.1. For any x ≥ 0, the function f (x) = 1+x
is monotonically increas-
ing.

Proof. Let y > x ≥ 0. Then,

1 1 1 1
< and so 1 − >1− ,
1+y 1+x 1+y 1+x
that is,
y x
> .
1+y 1+x

Theorem A.0.1. For any two real numbers x and y, the following inequality holds:

|x + y| |x| |y|
≤ + . (A.1)
1 + |x + y| 1 + |x| 1 + |y|

Proof. Let x and y have the same sign. Without loss of generality, we may assume that
x ≥ 0 and y ≥ 0. Then,
|x + y| x+y x y
= = +
1 + |x + y| 1+x+y 1+x+y 1+x+y
x y |x| |y|
≤ + = + .
1+x 1+y 1 + |x| 1 + |y|
If x and y have different signs, then we may assume that |x| > |y|. Then |x + y| ≤ |x|. From
Proposition A.0.1, we have
|x + y| |x| |x| |y|
≤ ≤ + .
1 + |x + y| 1 + |x| 1 + |x| 1 + |y|

192
Qamrul Hasan Ansari Metric Spaces Page 193

Theorem A.0.2 (Hölder’s Inequality). Let xi ≥ 0 and yi ≥ 0 for all i = 1, 2, . . . , n and

let p > 1 and q > 1 such that p1 + 1q = 1. Then,

n n
!1/p n
!1/q
X X X
xi yi ≤ xpi yiq . (A.2)
i=1 i=1 i=1

When p = q = 2, then the above inequality becomes

n n
!1/2 n
!1/2
X X X
xi yi ≤ x2i yi2 . (A.3)
i=1 i=1 i=1

This inequality is known as Cauchy-Schwarz inequality.

Theorem A.0.3 (Minkowski’s Inequality). Let xi ≥ 0 and yi ≥ 0 for all i = 1, 2, . . . , n

and let p > 1. Then,

n
!1/p n
!1/p n
!1/p
X p
X X
(xi + yi ) ≤ xpi + yip . (A.4)
i=1 i=1 i=1

Proof. If p = 1, then there is nothing to prove. So, we assume that p > 1. Note that
n
X n
X n
X
p p−1
(xi + yi ) = xi (xi + yi ) + yi (xi + yi )p−1 . (A.5)
i=1 i=1 i=1

1 1
Let q > 1 be such that + = 1. From equation (A.5) and Hölder’s inequality, we obtain
p q
n n
!1/p n
!1/q
X p
X p
X (p−1)q
(xi + yi ) ≤ xi (xi + yi )
i=1 i=1 i=1
n
!1/p n
!1/q
X X
+ yip (xi + yi )(p−1)q
 i=1 !1/p
i=1
!1/p  n !1/q
 Xn n
X  X
= xpi + yip
(xi + yi )p .
 
i=1 i=1 i=1

n
X n
X
p
If (xi + yi ) = 0, then obviously (A.4) holds. If (xi + yi )p 6= 0, then by dividing the
i=1 i=1
n
!1/q
X
above inequality by (xi + yi )p , we obtain (A.4).
i=1
Qamrul Hasan Ansari Metric Spaces Page 194

Theorem A.0.4 (Minkowski’s Inequality for Infinite Sums). Let p > 1 and let {xn }
∞
X ∞
X
p
and {yn } be sequences of nonnegative terms such that xn and ynp are convergent.
n=1 n=1
∞
X
Then (xn + yn )p is convergent. Moreover,
n=1

∞
!1/p ∞
!1/p ∞
!1/p
X p
X X
(xn + yn ) ≤ xpn + ynp . (A.6)
n=1 n=1 n=1

Proof. For any positive integer m, from Theorem A.0.3, we have

m
!1/p m
!1/p m
!1/p
X p
X X
(xn + yn ) ≤ xpn + ynp
n=1 n=1 n=1
∞
!1/p ∞
!1/p
X X
≤ xpn + ynp .
n=1 n=1
nP o
p 1/p
Since ( mn=1 (xn + y n ) ) is an increasing sequence of nonnegative real numbers, it is
bounded above by the sum

∞
!1/p ∞
!1/p
X X
xpn + ynp .
n=1 n=1
P∞
It follows that n=1 (xn + yn )p is convergent and that the inequality (A.6) holds.

Theorem A.0.5. Let p > 1. For any a ≥ 0 and b ≥ 0, we have

(a + b)p ≤ 2p−1 (ap + bp ) . (A.7)

Proof. If either a or b is 0, then nothing to prove. Suppose that a > 0 and b > 0. Since the
function x 7→ xp defined on the set of all positive numbers is convex when p > 1, we have
 p
a+b ap + bp
≤ equivalently (a + b)p ≤ 2p−1 (ap + bp ) .
2 2
 Chapter B

Upper and Lower Bounds

Definition B.0.1

Let A be a subset of R. We say that

• M ∈ R is an upper bound of A if x ≤ M for all x ∈ A;

• m ∈ R is a lower bound of A if m ≤ x for all x ∈ A;

• A is bounded from above if there exists M ∈ R such that x ≤ M for all x ∈ A;

• A is bounded from below if there exists m ∈ R such that m ≤ x for all x ∈ A;

• A is bounded if it is both bounded below as well as bounded above.

If A has an upper bound M (lower bound m), then A may have many upper (lower) bounds.
Namely, all numbers greater (less) than or equal to M (m) is an upper (lower) bound.

Definition B.0.2

Let A be a subset of R.

• A number M ∈ R is said to be a supremum or least upper bound of A if M is an

upper bound of A and M ≤ M ′ for all upper bounds M |prime of A.

• A number m ∈ R is said to be an infimum or greatest lower bound of A if m is a

lower bound of A and m ≥ m′ for all lower bounds m|prime of A.

The supremum and infimum of A are denoted by sup A and inf A, respectively.

195
Qamrul Hasan Ansari Metric Spaces Page 196

Note that the supremum and infimum are unique if they exist. Also, supremum
√ and infimum
of a set may not lie in that set. For example,
√ the set A = {x ∈ Q : x < 2} of rational
numbers Q is bounded from above by 2, but has no supremum in Q.

If A does not have an upper (lower) bound, then we define sup A = ∞ (inf A = −∞). By
convention, sup ∅ = −∞ and inf ∅ = −∞.

If the supremum (infimum) of a set A belongs to it, then sup A (inf A) is called maximum
(infimum) of A.

Let {xn } be an arbitrary sequence of real numbers. We construct a new sequence {yn } by
taking the supremum of successively truncated ‘tails’ of the original sequence {xn }, yn =
sup{xk : k ≥ n}. The sequence {yn } is monotonically decreasing (xn ≥ xn+1 for all n)
because the supremum is taken over smaller sets for large n’s. Therefore, the sequence {yn }
has a limit, and such a limit is called lim sup of the sequence {xn } and it is denoted by
lim sup xn . Similarly, by taking the infimum of successively truncated ‘tails’ of the original
sequence {xn }, yn = inf{xk : k ≥ n}. Then we get a monotonically increasing(xn ≤ xn+1 for
all n) sequence {yn }. The limit of that sequence is called lim inf of {xn } and it is denoted
by lim inf xn .

Definition B.0.3

Let {xn } be a sequence of real numbers. Then,

lim sup xn = lim [sup{xn : k ≥ n}] ,

n→∞ n→∞

lim inf xn = lim [inf{xn : k ≥ n}] ,

n→∞ n→∞

It follows from the definition that

lim inf xn ≤ lim sup xn .

n→∞ n→∞

Moreover, a sequence {xn } converges if and only if

lim inf xn = lim sup xn .

n→∞ n→∞

Example B.0.1

If xn = (−1)n , then
lim inf xn = −1 and lim sup xn = 1.
n→∞ n→∞
 Chapter C

Partial Ordering

Definition C.0.1: Partial Ordering

A binary relation 4 on a nonempty set X is called partial ordering if the following

conditions hold:

(i) For all a ∈ X, a 4 a; (Reflexivity)

(ii) If a 4 b and b 4 a, then a = b; (Antisymmetry)

(iii) If a 4 b and b 4 c, then a 4 c; (Transitivity)

Definition C.0.2: Partially Ordered Set

A nonempty set X is called partially ordered if there is a partial ordering on X.

Remark C.0.1

The word ‘partially’ emphasizes that X may contain elements a and b for which neither
a 4 b nor b 4 a holds. In this case, a and b are called incomparable elements. If a 4 b
or b 4 a (or both), then a and b are called comparable elements.

197
Qamrul Hasan Ansari Metric Spaces Page 198

Definition C.0.3: Totally Ordered Set

A partially ordered set X is said to be totally ordered if every two elements of X are
comparable.

In other word, partially ordered set X is totally ordered if it has no incomparable

elements.

Definition C.0.4

Let M be a nonempty subset of a partially ordered set X.

• The element x ∈ X is called an upper bound of M if a 4 x for all a ∈ M.

• The element x ∈ X is called a lower bound of M if x 4 a for all a ∈ M.

• The element x ∈ M is called a maximal element of M if x 4 a implies x = a.

Remark C.0.2

A subset of a partially ordered set M may or may not have an upper or lower bound.
Also, M may or may not have maximal elements. Note that a maximal element need
not be an upper bound.

Example C.0.1

(a) Let X be the set of all real numbers and let a ≤ b have its usual meaning. Then
X is totally ordered and it has no maximal element.

(b) Let X = N, the set of all natural numbers. Let m 4 n mean that m divides n.
Then 4 is a partial ordering on X.

(c) Let P(X) be the power set, that is, set of all subsets of a set X. Let A 4 B mean
A ⊂ B. Then P(X) is a partially ordered set and the only maximal element of
P(X) is X.

(d) Let X be the set of all ordered n-tuples x = (x1 , x2 , . . . xn ), y = (y1 , y2 , . . . yn ), . . .

of real numbers and let x 4 y mean xi ≤ yi for all i = 1, 2, . . . , n. Then 4 is a
partial ordering on X.
Qamrul Hasan Ansari Metric Spaces Page 199

Lemma C.0.1 (Zorn’s Lemma). Let X be a nonempty partially ordered set in which
every totally ordered set has an upper bound. Then X has at least one maximal element.
 Chapter D

Nested Interval Property

Least Upper Bound Axiom. If A is a set of real numbers bounded from above, then A
has a least upper bound.

Theorem D.0.1 (Nested Interval Property). Let I1 = [a1 , b1 ], I2 = [a2 , b2 ], . . . be a

sequence of nested closed and bounded intervals, that is, I1 ⊇ I2 ⊇ I3 ⊇ · · · ⊇ In ⊇
ITn+1 ⊇ · · · . Then there exists at least one point common to every interval, that is,
∞
i=1 Ii 6= ∅.

Proof. Since I1 ⊇ I2 ⊇ I3 ⊇ · · · ⊇ In ⊇ In+1 ⊇ · · · , we have

a1 ≤ a2 ≤ · · · and · · · ≤ b2 ≤ b1 .

We claim that am < bn for every m, n ∈ N

If m > n then am < bm ≤ bn , and if m ≤ n then am ≤ an < bn . Thus each bn is an upper

bound for the set A = {a1 , a2 , a3 , . . .} of left end points. By the Least Upper Bound Axiom
of R, A has a least upper bound, that is, sup(A) exists, say, p = sup(A).

Since each bn is an upper bound for A and p is the least upper bound, we have p ≤ bn for
each n ∈ N. Furthermore, since p is an upper bound for A = {a1 , a2 , . . .}, we have an ≤ p
for each n ∈ N. But
an ≤ p ≤ bn implies p ∈ In = [an , bn ].
Hence p is common to every interval.

200
Qamrul Hasan Ansari Metric Spaces Page 201

Remark D.0.1

It is necessary that the intervals in the above theorem be closed and bounded, otherwise
the theorem is not true.

Example D.0.1

Let A1 = (0, 1], A2 = (0, 21 ], . . . , An = (0, n1 ], . . ., then {An } is the sequence of open-
closed intervals. It is clear that the sequence {An } is nested, T∞that is, A1 ⊇ A2 ⊇ A3 ⊇
· · · . But the intersection of the intervals is empty, that is, n=1 An = ∅.

Example D.0.2

Let {An } be the following sequence of closed infinite intervals

A1 = [1, ∞), A2 = [2, ∞), . . . , An = [n, ∞), . . . .

T∞ clearly A1 ⊃ A2 ⊃ A3 ⊃ · · · , that is, the sequence of intervals is nested. But

Then
n=1 An = ∅.

Theorem D.0.2 (Heine-Borel Theorem). Every closed and bounded interval of the real
line R is compact.

Proof. Let I1 = [a1 , b1 ] be a closed and bounded subset of R. Assume that I1 is not compact.
Then there exists an open cover F = {Gα }α∈Λ of I1 which does have  a finite subcover.
 a +b  Divide
a1 +b1
I1 = [a1 , b1 ] into two closed subintervals of equal length a1 , 2 and 1
2
1
, b1 .

Then, by assumption, at least one of these two intervals will not be covered by any finite
subcover of the cover F . Let us denote  such interval by I2 and write I2 = [a2 , b2 ]. Then
[a2 , b2 ] is either a1 , a1 +b
2
1
or a1 +b1
2
, b1 .

   
Divide I2 into two closed subintervals of equal length a2 , a2 +b 2
2
and a2 +b
2
2
, b2 . Again, by
assumption, at least one of these two intervals will not be covered by any finite subcover of
the cover F . Let us denote such interval by I3 and write I3 = [a3 , b3 ].

Continuing in this way, we obtain a sequence of closed intervals having the following prop-
erties:

(i) In ⊃ In+1 for all n ∈ N,

(ii) In is closed for all n ∈ N,

Qamrul Hasan Ansari Metric Spaces Page 202

(iii) In is not covered by any finite subclass of F ,

(iv) δ(In ) → 0 as n → ∞, where δ(In ) denotes the length of the interval In .

Hence by the nest closed interval property,

∞
\
In 6= ∅.
n=1
T∞
So there exists a number x ∈ n=1 In which implies that x ∈ In for each n ∈ N. In particular,
x ∈ I1 .

Since F is an open cover of I1 , there exists Gα0 ∈ F such that x ∈ Gα0 . Since Gα0 is an
open set, there exists an open neighborhood (x − ε, x + ε) such that

x ∈ (x − ε, x + ε) ⊆ Gα0 .

Now, δ(In ) → 0 as n → ∞, we can choose a positive integer n0 large enough that δ(In0 ) < ε.
This implies that
In0 ⊆ (x − ε, x + ε) ⊆ Gα0 .
This contradicts to our assumption that no In is covered by a finite number of members of
F . Hence [a1 , b1 ] is compact.

Definition D.0.1: Cantor Set

Let C0 denote the interval [0, 1]. Remove from C0 in succession:

1 2


(i) the open interval ,
3 3
, the middle third of the interval in C0 , leaving behind the
set    
1 2
C1 = 0, ∪ ,1 .
3 3
1 2

7 8


(ii) the
 1 open intervals
  ,
9 9
and ,
9 9
, the middle thirds of the two closed intervals
0, 3 and 32 , 1 in C1 , leaving behind the set
       
1 2 1 2 7 8
C2 = 0, ∪ , ∪ , ∪ ,1 .
9 9 3 3 9 9
Qamrul Hasan Ansari Metric Spaces Page 203

0 1
C0
1 2
C1 3 3 1
1 2 1 2 7 8
C2 9 9 3 3 9 9 1

.. .. .. .. ..
. . . . .

Continue in this way, we get a sequence {Cn } of sets, where Cn is obtained from Cn−1
by removing the (open) middle thirds of the 2(n−1) disjoint closed intervals of which
Cn−1 is composed of, that is,
∞  
[ 1 + 3k 2 + 3k
Cn = Cn−1 − , .
k=0
3n 3n

The intersection \
C= Cn
n∈N

is called the Cantor set.

Each of the sets Cn is nonempty, closed and bounded. Also, Cn ⊂ Cn−1 for each n. Hence
the set C is nonempty, closed and bounded.

Exercise D.0.1. Prove that the Cantor set C is compact.

Proof. Since each Cn is the union of 2n disjoint closed intervals, so that each Cn is closed.
Also, arbitrary intersection of closed sets is closed, we have
∞
\
C= Cn is closed.
n=1

Also, C is bounded, because C ⊂ [0, 1]. Hence by Heine-Borel Theorem D.0.2, C is compact.
 Chapter E

Some Miscellaneous Results

Lemma E.0.1. Let f : [0, 1] → R be a continuous function such that f (x) ≥ 0 for all
R1
x ∈ [0, 1]. Then 0 f (x)dx = 0 if and only if f (x) = 0 for all x ∈ [0, 1].

Remark E.0.1
R1
If f : [0, 1] → R is nonnegative and Riemann integrable on [0, 1] such that 0 f (x)dx = 0,
then we cannot conclude that f ≡ 0 on [0, 1]. For example, consider the function

0, if x 6= 12
f (x) =
2, if x = 21 .
R1
Then f is Riemann integrable on [0, 1] and 0
f (x)dx = 0.

Theorem E.0.1 (Young’s inequality). Let a, b be nonnegative real numbers. Let p > 1
and q be such that p1 + 1q = 1. Then,

ap bq
ab ≤ + . (E.1)
p q

The equality holds in (E.1) if and only if ap = bq .

204
Qamrul Hasan Ansari Metric Spaces Page 205

Theorem E.0.2 (Weierstrass’s Theorem). Let f : [a, b] → R be a continuous function.

Then there exists a sequence {pn } of polynomials with real coefficients that converges
uniformly to f on [a, b], that is, for all ε > 0, there exists a positive integer N such that
for all t ∈ [a, b]
|pn (t) − f (t)| < ε, whenever n > N.

We now mention some basic properties of complex numbers.

p
For z ∈ C, we write z = x + iy for x, y ∈ R and define |z| = x2 + y 2 . The conjugate z̄ of
z ∈ C is x − iy. The real and imaginary parts of the complex number z are denoted by Rez
and Imz, respectively.

Some facts about the absolute value function on C.

(a) |z| = |z̄| for all z ∈ C.

(b) |z|2 = z z̄ for all z ∈ C.

(c) Rez ≤ |z| and Imz ≤ |z| for all z ∈ C.

(d) |zw| = |z| |w| for all z, w ∈ C.

(e) |z + w| ≤ |z| + |w|.