CHAPTER 1

METRIC SPACES

1 Fundamental Definitions n’ Stuff

Definition 1.1: Metric Space

A set X along with a function d : X × X → R+ ∪ {0} called distance, is said to be a
metric space if:

1. d(x, y) = 0 ⇐⇒ x = y (Positivity)

2. d(x, y) = d(y, x), ∀x, y ∈ X (Symmetric)

3. ∀x, y, z ∈ X we have, d(x, y) ≤ d(x, z) + d(y, z) (Triangle Inequality)

Example : Rn as a metric space

Note that Rn , the set of all n-tuples of R, is a metric space with
p
d(⃗x, ⃗y ) := |⃗x − ⃗y | = (x1 − y1 )2 + (x2 − y2 )+ · · · + (xn − yn )2

Definition 1.2: Open and Closed Balls around x in X

Open ball is defined as:

Br (x) := {y ∈ X : d(y, x) < r}

Closed ball is defined as:

B[r] (x) := {y ∈ X : d(y, x) ≤ r}

1
2 Metric Spaces

Definition 1.3: Convexity

A set S in Rn is said to be convex if ∀x, y ∈ S, t ∈ [0, 1], x + t(y − x) ∈ S

Example : Open and closed balls in Rn are convex

Consider Br (x) := {z ∈ Rn : |z − x| < r}. Consider arbitrary p and q in Br (x). We have
that d(p, x) < r and d(q, x) < r. Consider p + t(q − p) and consider d(p + t(q − p), x) =
|p + t(q − p) − x| = |tq + (1 − t)p − x + tx − tx| = |tq − tx + (1 − t)p − (1 − t)x| ≤
t|q − x| + (1 − t)|p − x| = td(q, x) + (1 − t)d(p, x) < r. Replacing < with ≤ in the above
proves the result for closed balls.
■

Definition 1.4: Sequences in Metric Spaces

A sequence {xn : xn ∈ X} is a mapping from the naturals to X, where order is implicit.
We say a sequence in a metric space X is convergent to x ∈ X if:

(∀ε > 0)(∃n0 ∈ N)(∀n ∈ N, n ≥ n0 )(d(xn , x) < ε)

Definition 1.5: Cauchy Sequence in X

A sequence {xn } in X is said to be cauchy if

(∀ε > 0)(∃n0 ∈ N)(∀n, m ∈ N)(n, m ≥ n0 =⇒ d(xn , xm ) < ε)

Theorem 1.6
Convergence =⇒ cauchy

Proof for Theorem.

Say xn converges to x in the metric space.

(∀ε > 0)(∃n0 )(∀n, m ≥ n0 )(d(xn , x) < ε, d(xm , x) < ε)

which implies (from triangle inequality)

(∀ε > 0)(∃n0 )(∀n, m ≥ n0 )(d(xn , x) < ε, d(xm , xn ) ≤ d(xm , x) < ε)

Which is the cauchy condition. ■

Metric Spaces 3

1.1 Inner products, Normz and some common metrics

Definition 1.7: Inner Product

Let F be either C or R, and V a vector space over F. An inner product on V is a
function that assigns to each ordered pair of vectors v, u ∈ V a scalar ⟨v|u⟩ ∈ F so that
the following holds:

1. ⟨v + cw|u⟩ = ⟨v|u⟩ + c⟨w|u⟩

2. ⟨u|v⟩ = ⟨v|u⟩ where · is just complex conjugation

3. ⟨u|u⟩ > 0 if u ̸= 0

A space V (F), ⟨·⟩ is called an inner product space.

Definition 1.8: Norm (From the inner product)

Given an inner product
p space V (F), ⟨·⟩, we define the norm or length of a vector x,
given by ||x|| as ⟨x|x⟩

Example : p-norm is a metric

The p-norm is defined on Rn as:
Xn
p p p 1/p
||⃗x||p := (|x1 | + |x2 | · · · |xn | ) =( |xi |p )1/p
i=1

We say dp (⃗x, ⃗y ), the distance between ⃗x and ⃗y to be ||x − y||p = ( ni=1 |xi − yi |p )1/p . From
P
this definition, unique 0 distance P property and reflexivity property are obvious. Consider
dp (x, z) and dp (z, y). dp (x, y) = ( ki=1 |(xi − zi ) + (zi − yi )|p )1/p whence from Minkowski
we see that
k
X Xn n
X
( |(xi − zi ) + (zi − yi )|p )1/p ≤ ( |xi − zi |p )1/p + ( |zi − yi |p )1/p = dp (x, z) + dp (y, z)
i=1 i=1 i=1

Therefore, p-norm is a metric ■

Example : The Max Norm
Suppose x ∈ Rn , the max norm ||x||∞ is defined as:
||⃗x||∞ := max (|xi |)
1≤i≤n

Clearly, d∞ (x, y) = 0 if and only if x = y. Moreover, from the properties of | · |, it is

reflexive. Consider d∞ (x, y) = max{|x1 −y1 |, |x2 −y2 | · · · , |xn −yn |}, d∞ (x, z) = max{|x1 −
z1 |, |x2 − z2 | · · · , |xn − zn |} and d∞ (y, z) = max{|y1 − z1 |, |y2 − z2 |, · · · |yn − zn |}.
d∞ (x, y) = max {|(xi −zi )+(zi −yi |)} ≤ max {|(xi −zi )}+ max {|(yi −zi )} = d∞ (x, z)+d∞ (z, y)
1≤1≤n 1≤1≤n 1≤1≤n
4 Metric Spaces

Hence the max norm is a metric. ■

Example : Norm on function spaces
Let X be a non empty set. Define B(X) as the space of all bounded, real functions. Then
||f ||∞ := supx∈X |f (x)| defines a norm on B(X) since |f (x) − g(x)| ≤ |f (x)| + |g(x)| ≤
supx∈X |f (x)| + supx∈X |g(x)| = ||f ||∞ + ||g||∞ which gives us ||f + g||∞ ≤ ||f ||∞ + ||g||∞
■
d
Example : the 1+d metric
d(x,y)
Say X, d is a metric space, then g(x, y) = 1+d(x,y)
is a metric.

Proof. exercise to the reader

■
Example : Constructing a metric from an injective function
Let Z be a set and (X, d) a metric space. Let f : Z → X be an injective function. Define
δ(x, y) = d(f (x), f (y)). This defines a metric on Z and makes (Z, δ) a metric space since
δ(x, x) = d(f (x), f (x)) = 0, δ(x, y) = 0 =⇒ d(f (x), f (y)) = 0 which gives x = y,
and δ(x, y) = d(f (x), f (y)) ≤ d(f (x), f (z)) + d(f (z), f (y)) = δ(x, z) + δ(y, z) which gives
triangle. ■
Example : The discrete metric
Let X be any set whatsoever. Define the metric
(
1 if x ̸= y
d(x, y) :=
0 if x = y

This is a metric due to positive definiteness and symmetry. The triangle inequality is,
also, not hard to see. ■
Example : Space of all bounded sequences
Let X be the set of all bounded sequences of reals, i.e, the set of all x such that x =
(x1 , x2 , · · · ) = {xn }∞ n
n=1 with supi=1 ({xi }) < ∞. If {xi } and {yi } are sequences (i,e,
elements in X) let the distance between two sequences be defined by
∞
d(x, y) := sup |xi − yi |
i=1

Clearly, it is positive definite and symmetric. Consider three sequences {xn }, {yn } and
{zn } in X. d(x, y) = sup{|xi − yi | : i ∈ N}, d(x, z) = sup{|xi − zi | : i ∈ N} and
d(y, z) = sup{|yi − zi | : i ∈ N}. Cosnider d(x, z) = sup{|xi − zi | = |xi − yi + yi − zi | : i ∈
N} ≤ sup{|xi − yi | + |yi − zi | : i ∈ N} =⇒ ≤ sup{|xi − yi | : i ∈ N} + sup{|yi − zi | : i ∈ N}
which proves the triangle. ■
Metric Spaces 5

Basics on Continuity (Read definition of sequence in metric space, and limit

points)

Definition 1.9: Continuous functions

A function f : X → Y that maps a metric space X, dx to a metric space Y, dy is said
to be continuous at a ∈ X (which is a limit point of X) if

(∀ε > 0)(∃δ > 0)(∀p ∈ X)(dx (p, a) < δ =⇒ dy (f (p), f (a)) < ε)

Theorem 1.10: Sequential Criteria for continuity

Let X, dx and Y, dy be metric spaces and f : X → Y . Let a be a limit point of X.
Then f is continuous at a if and only if for every sequence (qn ) ∈ X, qn ̸= a with
lim(qn ) = a, we have lim(f (qn )) = f (a)

Proof for Theorem.

=⇒ ) We have that for every ε > 0, there exists a δ > 0 so that for all z ∈ X such that
z ∈ Bδ (a), we have f (z) ∈ Bε (f (a)). Consider an arbitrary sequence in X that is so that
(qn ) ∈ X, qn ̸= a and lim(qn ) = a. Which reads ∀δ > 0, ∃n0 ∈ N so that ∀n ≥ n0 , we
have qn ∈ Bδ (a). Let ε > 0 be arbitrary. From the definition of continuity, there exists
a corresponding δ > 0 so that ∀z ∈ X, z ∈ Bδ (a) would imply f (z) ∈ Bε (f (a)). For
this δε , there exists an n0 (δ)(ε) ∈ N so that ∀n ≥ n0 , qn ∈ Bδ (a) which would imply
f (qn ) ∈ Bε (f (a)). This means that ∀ε > 0, there exists n0 ∈ N so that ∀n ≥ n0 , we have
f (qn ) ∈ Bε (f (a)) which means that lim(f (qn )) = f (a).

⇐= ) Suppose that every sequence qn ∈ X so that qn ̸= a with lim(qn ) = a, we have

that lim(f (qn )) = f (a) but that the function f is not continuous. i.e, ∃ε > 0 so that
∀δ > 0, ∃qδ ∈ X so that dx (qδ , a) < δ but dy (f (qδ ), a) ≥ ε. Choose δ = 1 and get the
corresponding x1 ∈ B1 (a) for which we have dy (f (x1 ), f (a)) ≥ ε. Choose δ = 1/2 and get
x2 ∈ X so that dx (x2 , a) < 1/2 but dy (f (x2 ), f (a)) ≥ ε. As such, keep going with δ = 1/n
to generate a sequence xn ∈ X so that d(xn , a) < 1/n but d(f (xn ), f (a)) ≥ ε. Note that
xn converges by definition, to a, but there exists an ε > 0 so that no matter what n0 we
take, there exists some n ≥ n0 so that dy (f (xn ), f (a)) ≥ ε which means that the sequence
f (xn ) does not converge to f (a), which contradicts hypothesis. ■

Corollary 1.11
Composition of continuous functions (at a point a) is finally a continuous function

Proof for Corollary.

Suppose X, dx ,Y, dy and Z, dz are metric spaces with f : X → Y and g : Y → Z two
continuous functions. g ◦ f : X → Z. Consider a sequence (xn ) ∈ X that converges to a.
From sequential criteria, f (xn ) → f (a). We have f (xn ), a sequence in Y , that converges
6 Metric Spaces

to f (a). From the continuity of g, we note that g ◦ (f (xn )) converges to g ◦ (f (a)), which
ultimately tells that for every sequence {xn } in X that converges to a, g ◦ f (xn ) is a
sequence in Z that converges to g ◦ f (a) whence we are done ■
Example : Dirichlet Function
Let f : R → R be defined by
(
1 if x is rational
f (x) =
0 if x is irrational

This function is discontinuous at every point of R. Bear in mind that to show that a
function is discontinuous at a point z, we need only show that one sequence in A which
converges to z is such that lim(f (xn )) ̸= f (lim(xn )). Consider a ∈ R an irrational point.
Consider the rational sequence xn that converges to a via density theorem. f (xn ) = 1 for
every n ∈ N which means lim f (xn ) = 1. But f (lim(xn )) = 0. Hence, discontinuous at
a ∈ R, irrational. In fact, the same argument can be re-used to show that f is irrational
at every point. ■
Example : Thomae Function
We define the Thomae function as follows: Let f : R+ → R,
(
0 if x is irrational
f (x) :=
1/n if x is rational, with x = m/n, gcd(m, n) = 1

This function is discontinuous at rational points quite obviously since, consider a rational
number m/n where f (m/n) = 1/n > 0. Consider an irrational sequence in R+ converging
to m/n. Clearly, f (xn ) = 0 for every n ∈ N whence, it is clear.

Consider z an irrational point where f (z) = 0. Consider a ε > 0 and an n0 so that

1/n0 < ε. Consider the subinterval (z − γ, z + γ) for some γ. We note that in this
subinterval, only finite rational points have their denominator smaller than n0 (because
(z − γ)n < m < (z + γ)n and this interval has size 2nγ. For a fixed n and γ, m can pick
values only from this interval. If n < n0 , then the interval size is smaller than 2n0 (γ). If we
make γ small enough, m would be restricted to just one value. say, m0 . With this γ (and
m0 fixed) the only values n can take that are less than n0 are certainly finite now since
there are only finite ns smaller than n0 . Summarising, for a given ε, we choose an n0 so
that 1/n0 < ε, where we concluded that for small enough γ, only finite rational points in
the interval (z − γ, z + γ) have their denominator smaller than n0 ). Now, we can eliminate
all these rational points by picking an appropriate δ so that no rational points with de-
nominator smaller than n0 exist in (z − δ, z + δ). Therefore, for a given ε (∃n0 : 1/n0 < ε)
there exists δ so that either a point q in (z − δ, z + δ) is irrational, whence f (q) = 0 < ε,
or it has its denominator larger than n0 , which means f (q) = 1/n < 1/n0 < ε which leads
to |f (q) − f (z)| = |f (q)| = f (q) < ε which is the "ε − δ" criterion for Continuity. Hence,
at all irrational points z, the Thomae function is continuous. ■
Now we deal with, primarily, continuous functions on closed, bounded intervals of the kind
I = [a, b] ⊂ R.
Metric Spaces 7

Theorem 1.12: Boundedness Theorem

Continuous functions on closed bounded intervals are bounded

Proof for Theorem.

Suppose f : I → R is actually unbounded, i.e, for any M ∈ R we take, there exists an
xM ∈ I so that f (xM ) > M . Let M1 = 1 and get the corresponding x1 so that f (x1 ) > 1.
Do the same for Mn = n to get xn so that f (xn ) > n. This sequence f (xn ) is divergent.
But the sequence {xn } is bounded, hence from Bolzano, there is a subsequence xnk that
is convergent, say to p. But f (xnk ) is a subsequence of f (xn ), which means it diverges.
But according to continuity, f (xnk ) converges to f (p), which is a contradiction. Hence, f
is bounded. ■

Theorem 1.13: Maxima-Minima Theorem

Let f : I → R be a continuous function on closed bounded interval I. Then f attains
its maxima and minima.

Proof for Theorem.

We know that f (I) is actually bounded, which means it has supremum and an infimum
U and L. Suppose it does not attain bounds, i.e, (talking about upper bound) ∀x ∈ I,
f (x) < U . Consider U − 1. There exists f (x1 ) so that U − 1 < f (x1 ) < U . Choose
x2 likewise so that U − 1/2 < f (x2 ) < U . Keep going as such, to find xn so that
U − 1/n < f (xn ) < U . This means from squeeze theorem that f (xn ) converges to U . Note
that xn that we have collected is a sequence that is in I, a bounded, closed interval. Hence,
it has a convergent subsequence xnk → x. Also, f (xnk ) is a subsequence of f (xn ) which
means f (xnk ) → U . But from continuity, f (xnk ) → f (x), which implies f (x) = U , where
x ∈ I. This means that the function actually does attain upper bound (similar argument
for the lower bound can be performed). ■

Theorem 1.14: Location of roots theorem

Let f : I → R be a continuous function on a closed bonunded interval I. Let a < b
with f (a) < 0 and f (b) > 0 (or the other way around). Then there exists c ∈ (a, b) so
that f (c) = 0

Proof for Theorem.

Let a0 = a and b0 = b with a < b. a0 , b0 forms an interval of size ξ = b0 − a0 . Look at
f (a0 + b0 /2). Is it greater than 0? If so, make a1 = a0 , and b1 = a + b/2, a new interval
a1 , b1 of size ξ/2. This new interval obeys the property that f (a1 ) < 0 and f (b1 ) > 0. Is
f (a0 + b0 /2) < 0? If so, make b1 = b0 and a1 = a0 + b0 /2 to make a1 , b1 a new interval of
size ξ/2. Again, f (a1 ) < 0 and f (b1 ) > 0 Suppose you have made the n − th interval an , bn
so that f (an ) < 0 and f (bn ) > 0 of size ξ/(2n ). Make an+1 , bn+1 by asking similar questions
as above. We thus have a sequence of nested intervals [an , bn ], for which f (an ) < 0 and
8 Metric Spaces

f (bn ) > 0. From Nested interval theorem, we have γ ∈ ∩n [an , bn ], moreover, since the size
of these intervals are converging to 0, this is a unique point in the intersection. an → γ
and bn → γ. From continuity, then, we have f (an ) → f (γ) ≤ 0 and f (bn ) → f (γ) ≥ 0
which means that f (γ) = 0. ■

Theorem 1.15: Bolzano’s IVT

Let I be a closed, bounded interval and let f : I → R be a continuous function. If
a, b ∈ I with k ∈ R satisfying f (a) < k < f (b), then there is a point c in I so that
f (c) = k

Proof for Theorem.

Let a < b and define g(x) = f (x) − k. This is a continuous mapping on I. We then have
g(a) < 0 < g(b) whence from location or roots theorem, we find c ∈ I so that g(c) = 0 or
f (c) = k. ■

Theorem 1.16: Preservation of intervals Theorem

Let I be a closed, bounded interval. Let f : I → R be a continuous map. Then, f (I)
is also a closed bounded interval.

Proof for Theorem.

Consider f (I) := {z ∈ R : ∃x ∈ I such that f (x) = z}. We know that continuous functions
attain bounds, i.e, f (I) has a maxima and a minima that will (eventually) be the end points
of our closed bounded interval. If the function is constant, we are done. Consider a point
min(f (I)) < z < max(f (I)). By Bolzano’s IVT, we know that a pre-image exists for z.
This means that z is in the image. We are, therefore, done. The requisite interval is
[min(f ), max(f )]. ■

Definition 1.17: Uniform Continuity

A function f : A → R is said to be uniformly continuous on A if for every ε > 0,
∃δ > 0 so that for all x, u ∈ A such that 0 < |x − u| < δ, we have |f (x) − f (u)| < ε.

We see that this definition is similar to the definition of continuity on A except

for the fact that δ is independent of the x, the point at which we speak of continuity.
Metric Spaces 9

Theorem 1.18
Let f be a function defined on A ⊆ R. The following are equivalent:

1. f is not uniformly continuous on A

2. ∃ε > 0 such that ∀δ > 0, ∃xδ , uδ ∈ A so that 0 < |xδ − uδ | < δ but |f (xδ ) −
f (uδ )| ≥ ε

3. ∃ε and two sequences xn and yn so that lim(xn − un ) = 0 but |f (xn ) − f (un )| ≥ ε

for every n ∈ N

Proof for Theorem.

(1) =⇒ (2)) Obvious negation of the definition

(2) =⇒ (3)) Choose δ = 1, and get corresponding x1 , u1 so that 0 < |x1 − u1 | < 1
and |f (x1 ) − f (u1 )| ≥ ε. Choose δ = 1/2 and get the corresponding x2 and u2 so that
0 < |x2 − u2 | < 1/2 and |f (x2 ) − f (u2 )| ≥ ε. Keep going as such to get xn , un so that
0 < |xn − un | < 1/n ∀n ∈ N and |f (xn ) − f (un )| ≥ ε for every n ∈ N. This means that
lim(xn − un ) = 0 but |f (xn ) − f (un )| ≥ ε for every n ∈ N.

(3) =⇒ (2)) ∀δ > 0, ∃n0 (δ) so that ∀n ≥ n0 , we have 0 < |xn − un | < δ but
|f (xn ) − f (un )| ≥ ε , choose one of these xn −s for n greater than n0 (δ) as our xδ (and
likewise for uδ ). We are then done. ■

Theorem 1.19: Uniform Continuity Theorem

Let I be a closed bounded interval and let f : I → R be continuous on I. Then f is
uniformly continuous on I

Proof for Theorem.

Suppose f is not uniformly continuous on I. From the previous non uniform criteria, we
have that ∃ε and two sequences in I, xn , un so that lim(xn − un ) = 0 (or lim(xn ) = lim(un )
) but |f (xn ) − f (un )| ≥ ε for every n ∈ N. Since xn and un are sequences in I which is
closed and bounded, xn has a subsequence xnk converging to x. Since lim(xn − un ) = 0,
lim(xnk −unk ) = 0 Which means that lim(unk ) = x as well. We have xnk → x and unk → x,
which means from continuity that f (xnk ) → f (x) and f (unk ) → f (x), whence we find that
lim(f (xnk ) − f (unk )) = 0 necessarily. But this contradicts the assumption that f is not
uniform continuous. Hence, we are done. ■

Sequences of Functions
Given a set A ⊂ R, we primarily work with a (countably) infinite collection of functions
{fn } with fn : A → R. We generate a sequence of numbers by evaluating fn at a fixed
point x ∈ A. It could be, then, that {fn (x)}, treated as a sequence of numbers in R, either
converges or does not. For a subset A0 ⊆ A (which can possibly be empty), fn (x) converges
10 Metric Spaces

for every x ∈ A0 . If this happens, there is a uniquely determined value that we would like to
call "f (x)". Thus, there arises naturally a function f : A0 → R that we call the "limit" of
{fn }.

Definition 1.20: Convergence of sequence of functions

Let {fn } be a sequence of functions defined on A ⊆ R. Let A0 ⊂ A, and let
f : A0 → R. We say the sequence (fn ) converges to f pointwise if for each x ∈ A0 ,
The sequence (of real numbers) fn (x) converges to f (x).

In other words, {fn : A → R} converges to f pointwise if for every x ∈ A0 ,

∀ε > 0, ∃n0 (x, ε) ∈ N such that ∀n ≥ n0 (x, ε), we have |fn (x) − f (x)| < ε

If we remove the dependence of n0 above on the x, and leave it just to depend on ε, we

arrive at the definition of Uniform Convergence.

Definition 1.21: Uniform Convergence of a sequence of functions

Let {fn } be a sequence of functions defined on A ⊂ R. We say {fn } converges
uniformly to f : A0 → R if ∀x ∈ A0 , ∀ε > 0, ∃n0 (ε) ∈ N so that ∀n ≥ n0 (ε),
|fn (x) − f (x)| < ε.

Since the dependence of n on x is non-existent, we can rewrite the above defi-

nition to read: {fn } converges uniformly to f : A0 → R if ∀ε > 0, ∃n0 (ε) ∈ N so that
∀n ≥ n0 (ε), ∀x ∈ A0 , |fn (x) − f (x)| < ε.

Theorem 1.22: Non uniform convergence criteria

A sequence of functions {fn } defined on A does not converge uniformly to f : A0 → R
if and only if ∃ε > 0 such that ∀k ∈ N, ∃nk ≥ k and xk ∈ A so that |fnk (xk )−f (xk )| ≥ ε,
which means for some ε > 0, there exists a subsequence fnk of fn and a sequence in xk
in A0 so that
|fnk (xk ) − f (xk )| ≥ ε ∀k ∈ N

Proof for Theorem.

Obvious ■
Metric Spaces 11

Definition 1.23: The Uniform Norm

Say A ⊆ R and ϕ : A → R. We say ϕ is bounded on A if ϕ(A), the set, is bounded in
R. If it is bounded, we can define what is the uniform norm of ϕ on A by:

||ϕ||A := sup{|ϕ(x)| : x ∈ A}

It follows that ||ϕ||A ≤ ε ⇐⇒ |ϕ(x)| ≤ ε, ∀x ∈ A

Lemma 1.24
A sequence fn of bounded functions on A converges uniformly to f : A → R if and
only if ||fn − f ||A → 0

Proof for Lemma

=⇒ ) Say fn →f , this means that ∀ε > 0, ∃n0 ∈ N so that ∀n ≥ n0 , ∀x ∈ A, we have
|(fn − f )(x)| < ε, which means that for every n ≥ n0 , we have ||(fn − f )||A ≤ ε which
concludes the forward direction.

⇐= ) Say ||fn − f ||A → 0, which means that ∀ε > 0, ∃n0 ∈ N so that ∀n ≥ n0 , we

have ||fn − f ||A < ε which means sup{|(fn − f )(x)| : x ∈ A} < ε which means ∀x ∈ A,
|fn (x) − f (x)| < ε whence the back implication is also done.
■

Theorem 1.25: Cauchy criteria for uniform convergence

A sequence fn of bounded functions on A is uniformly convergent to f : A → R (a
bounded function) if and only if for every ε > 0, there is an n0 ∈ N so that ∀m, n ≥ n0 ,
||fm − fn ||A ≤ ε

Proof for Theorem.

=⇒ ) Say fn →f . This means that ∀ε > 0, ∃n0 so that ∀n, m ≥ n0 , we have, for every
x ∈ A, |fn (x) − f (x)| < ε/2 and |f (x) − fm (x)| < ε/2. Simply add them. We then have
|(fm − fn )(x)| < ε for every x ∈ A. In terms of the uniform norm. we see that ∀ε > 0
,∃n0 ∈ N so that for all m, n ≥ n0 , we have ||fm − fn ||A ≤ ε.

Suppose that ∀ε > 0, ∃n0 ∈ N so that ∀m, n ≥ n0 , ||fm − fn ||A ≤ ε which means
that for every x ∈ A, we have |fm (x) − fn (x)| < ε. This means that for every x, fn (x) is
a cauchy sequence, hence convergent to some f (x). More can be said for the sequence at
each x. Fix some x ∈ A. We have that ∀ε > 0, ∃n0 (ε) ∈ N so that ∀n, m ≥ n0 , we have
|fn (x) − fm (x)| < ε which means that ∀n ≥ n0 , |fn (x) − f (x)| < ε which means that, all
in all, ∀ε > 0, ∃n0 (ε) ∈ N so that ∀n ≥ n0 , ∀x ∈ A, we have |fn (x) − f (x)| < ε which is
12 Metric Spaces

the uniform convergence criteria. ■

Metric Spaces 13

lp Space

Definition 1.26: lp -Space

P∞
Define lp as the space of all sequences (an )∞
n=1 such that i=1 |ai |p exists. The set of
all sequences in Real line that are p summable

PIt is easily seen thet lp is a vector space over R. We defne the metric on lp as dp ({x}, {y}) =
( ∞i=1 |x i − y i |p 1/p
) . Positive definiteness is clear since the sum would not be 0 if even one
index is different. Symmetry is obvious, and triangle:
∞
X ∞
X
d(x, z) = ( |xi − zi |p )1/p ≤ ( |xi − yi + yi − zi |p )1/p
i=1 i=1

which gives
∞
X ∞
X
p 1/p
≤( |xi − yi | ) +( |yi − zi |p )1/p = d(x, y) + d(y, z)
i=1 i=1

(This must be obvious from the infinite case of minkowski inequality).

Example : Continuous functions on a closed bounded interval
C[a, b] is a metric space under the metric d(f, g) := sup{|f (x) − g(x)| : x ∈ [a, b]}.
Positive definiteness comes from | · |. Symmetry is aswell obvious. Look at d(f, g) =
sup{|f (x) − g(x)| : x ∈ [a, b]} = sup{|f (x) − h(x) + h(x) − g(x)|x ∈ [a, b]} which gives
≤ sup{|f (x) − h(x)| + |h(x) − g(x)| : x ∈ [a, b]} ≤ sup{|f (x) − h(x)| : x ∈ [a, b]} +
sup{|h(x) − g(x)| : x ∈ [a, b]} which gives us our desired d(f, g) ≤ d(f, h) + d(h, g) ■
Example : Space of all sequences in R
We consider the space of all sequences in R with the metric d({x}, {y}) := ∞ |xi −yi |
P
i=1 (1+|xi −yi |)2i .
1/2i . Positive definiteness is
P
This is quite well defined since the series is bounded by
clear since the sum is of positive numbers. Symmetry is aswell clear since d(x, y)/1 +
d(x, y) is a well defined metric as seen before. Consider d(x, y) := ∞ 1 |xi −yi |
P
i=1 2i ( 1+|xi −yi | ) ≤
P∞ 1 |xi −zi | P∞ 1 |zi −yi |
i=1 2i ( 1+|xi −zi | ) + i=1 2i ( 1+|zi −yi | ) = d(x, z) + d(z, y) (the inequality comes from the
metric structure of d/1 + d metric). ■

Definition 1.27: Norm

Given a Linear space V (R), a norm, ||.||, is a function mapping every vector to a real
number satisfying the following for every x, y ∈ V and every α ∈ R:

1. ||x|| ≥ 0 and equality of and only if x = 0.

2. ||αx|| = |α|||x||

3. ||x + y|| ≤ ||x|| + ||y|| (Triangle)

14 Metric Spaces

Definition 1.28: Psuedo Metrics

Let X be a non-empty set. A psuedometric is a function d : X × X → R that follows:

1. d(x, y) ≥ 0

2. d(x, y) = 0

3. d(x, y) = d(y, z)

4. d(x, y) ≤ d(x, z) + d(y, z)

The part that is different is that, for metric spaces, we require that if the distance is 0,
the points are the same. But for psuedometric spaces, the points can be different yet
of the same distance.

Theorem 1.29: Coordinate-wise convergence in Rn with dp metric

With the metric dp (x, y) = ( ni=1 (|xi − yi |p ))1/p on Rn , convergence implies coordinate
P
wise convergence

Proof for Theorem.

Suppose a sequence x⃗k converges to ⃗x, i.e, (xk1 , xk2 · · · xkn ) → (x1 , x2 , · · · xn ). This means
that for every ε > 0, there exists a n0 ∈ N so that ∀k ∈ N, k ≥ n0 would mean that
n
X
( (|xki − xi |p ))1/p < ε
i=1

From here, coordinate wise convergence is obvious.

Suppose that limk→∞ xkj = xj for every j ≤ n, i.e, coordinate wise convergence. That
means that for a given j,for every ε > 0, there exists n0 (ε, j) such that ∀k ∈ N, k ≥ n0 (ε, j)
ε
means that |xkj − xj | < n1/p . If we take the maximum of all the n0 (ε, j), and sum
εp 1/p
Pn k p 1/p n
= ε, which implies convergence in Rn . In a simi-
P
( i=1 |xj − xj | ) = i=1 ( n )
n
lar fashion, we can show that R , d∞ , the max norm, has equivalence between convergence
and coordinate wise convergence ■

Theorem 1.30
Rn with the metric d∞ (x, y) = max{|xj − yj | : j ≤ n} also has an equivalence between
convergence and coordinate wise convergence.

Example : Cauchy sequences need not converge R on C[0, 1]

Consider X = C[0, 1], with the metric d(f, g) := [0,1] |f (x) − g(x)|dx. Let fn (x) be a
Metric Spaces 15

sequence defined as follows:


1 1
0 , if 0 ≤ x ≤ 2 − n

fn (x) = n(x − 21 ) + 1 , if 12 − 1
n
<x≤ 1
2

1 , if 12 < x ≤ 1


. Is this sequence cauchy? Yes. Consider [0,1] |fm (x) − fn (x)|dx ≤ 21 (n + m/(mn)). For
R

any given ε, we can make m, n large enough to go below it. Suppose that fn converges
to some f , implying that for every R x ∈ [0, 1], we have that ∀ε, ∃n0 ∈ R N so that ∀n ∈
N, n ≥ n0 implies d(fn , f ) = [0,1] |fn (x) − f (x)| < ε. d(fn , f ) = [0, 1 − 1 ] |f (x)|dx +
R R 2 n
1 1 |fn (x) − f (x)|dx + |1 − f (x)|dx but since limn→∞ (d(fn , f )) → 0, we require
[ 21 − n ,2] [ 12 ,1]
R R
that [0, 1 − 1 ] |f (x)|dx = 0 and [ 1 ,1] |1 − f (x)|dx = 0. Since f is continuous (or supposed
2 n 2
to be), we see that f (x) = 0 from 0 to 12 , but 1 from 1
2
to 1, which is absurd. ■

Lemma 1.31: A useful lemma comparing norms on Rn

For Rn , we have
|| · ||p ≤ || · ||1 ≤ n|| · ||∞ ≤ n|| · ||p

Proof for Lemma

Pn
Consider
Pn any vector v in Rn , v = (x
P1p, x2 · · ·p, 1/p
xn ). We have ||v||p := ( i=1 |xi |p )1/p ≤
( i=1 |xi |) ≤ n max{xi : i ≤ n} ≤ n( i=1 |xi | ) ■

Definition 1.32: Complete Metric Space

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

Theorem 1.33
R with d(x, y) = |x − y| is a complete metric space

Theorem 1.34
Rn with dp , d∞ , is complete.

Proof for Theorem.

(k) (k) (k) (k)
Consider a cauchy sequence {x(k) } in Rn , i.e, x( k) = (x1 , x2 , x3 , · · · xn ) so that ∀ε > 0,
∃m0 (ε) so that ∀a, b ≥ m0 (ε) we have d(x(a) , x(b) ) < ε i.e:
Xn
a b (a) (b)
|xi − xi | ≤ ( |xi − xi |p )1/p < ε
i=1
which implies coordinate wise cauchy, which means coordinate wise convergence. That is
16 Metric Spaces

enough to ensure full convergence. This holds for both dp and d∞ . ■

Theorem 1.35
ℓp space, space of all p- summable sequences in R, is a complete metric space.

Proof for Theorem.

k
P∞{x }abe a bsequence
Let of sequences where x(k) = (xk1 , xk2 · · · ). The norm
P∞ here is d(xa , xb ) =
( i=1 |xi − xi | ) . ∀ε > 0, ∃m0 so that ∀a, b ≥ m0 , d(x , x ) = ( i=1 |xi − xbi |p )1/p < ε.
p 1/p a b a
(k)
Note that xki is point wise cauchy, hence pointwise convergent to say, xi → xi . Define
x := (x1 , x2 · · · ). For a given ε we have n0 so that
j ∞
X X
a b p 1/p
( |xi − xi | ) ≤ ( |xai − xbi |p )1/p < ε
i=1 i=1

for all a, b ≥ n0 , (for every j, too). We take the b → ∞ limit here, to get:
j
X
( |xai − xi |p )1/p < ε
i=1

for all a ≥ n0 and ∀j. From monotone convergence theorem, we can take the j limit as
well, to arrive at: ∀ε, ∃n0 so that ∀a ≥ n0 , we have
X∞
a
dp (x , x)( |xai − xi |p )1/p < ε
i=1

which lets us conclude xk → x. But the question is, is x ∈ ℓp ? Consider ( ji=1 |xi |p )1/p ≤
P

( ji=1 |xai − xi |p )1/p + ( ji=1 |xai |p )1/p (Minkowski). Both the summands on the right side
P P
are convergent (by MCT, and definition, respectively). Hence x ∈ ℓp , which makes ℓp a
complete space ■

Theorem 1.36
Set of all bounded sequences on R with d∞ norm is complete

Proof for Theorem.

Consider {xk } where xk = (xk1 , xk2 · · · ), a sequence that is cauchy. ∀ε > 0, ∃n0 so that
∀m, n ≥ n0 , we have |xni − xm m n
i | ≤ sup{|xi − xi | : i ∈ N} < ε (for every i ∈ N). This
ensures coordinate wise convergence. Let xni → xi , and let x = (x1 , x2 · · · ) so that ∀ε > 0,
∃m0 so that ∀n ≥ m0 , we have |xni − xi | < ε for all i ∈ N . This means for all n ≥ m0 ,
sup{|xni − xi | : i ∈ N} < ε, which is the condition for convergence. Hence, xn → x. Is
x a bounded sequence? We require that |xi | < M for some M , for all i ∈ N. Fix an ε,
and get a corresponding n and fix it so that |xni − xi | < ε for all i ∈ N, we then have:
|xi | ≤ |xni − xi | + |xni | ≤ ε + |xni | < some fixed number (since xn ∈ space of all bounded
Metric Spaces 17

sequences B(N)), hence we see that B(N) is cauchy complete. ■

Theorem 1.37
Set C[0, 1] of all continuous functions on closed, bounded interval [0, 1] is cauchy com-
plete under the supremum norm.

Proof for Theorem.

Consider {fn } a sequence of continuous functions on [0, 1] that is said to be cauchy. This
implies that, ∀ε > 0, ∃n0 so that ∀n, m ≥ n0 , sup{|fn (x) − fm (x)| : x ∈ [0, 1]} < ε.
This means that |fm (x) − fn (x)| < ε for all x ∈ X (beyond a certain n0 ). Hence, fm (x)
converges to say f (x). Moreover, note that for every point x, ∀ε > 0, ∃n0 (ε) (independent
of x) so that |fn (x) − f (x)| < ε for n ≥ n0 (ε). It is obvious then that f (x) is bounded on
[0, 1]. We also have that ∀ε > 0 ∃n0 (ε) so that sup{fn (x) − f (x) : x ∈ [0, 1]} < ε, which
tells that fn → f (uniformly). Now all that is left to see is that f is continuous.

Let ε > 0 be arbitrary. For every function fn , there exists δ(n)(ε) > 0 so that ∀a, b ∈ [0, 1],
0 < |a − b| < δ =⇒ |fn (a) − fn (b)| < ε/4. We also note that, there exists an n0 so that
for all m, n > n0 , we have for every x ∈ [0, 1], |fn (x) − fm (x)| < ε/4 for all m, n ≥ n0 .
Moreover, we also have that there exists n′ ∈ N so that |fn (x) − f (x)| < ε/4 for all x,
∀n ≥ n′ .

|f (a) − f (b)| = |f (a) − fn (a) + fn (a) − fm (a) + fm (a) − fm (b) + fm (b) − f (b)| ≤ |fn (a) −
f (a)|+|fm (a)−fm (b)|+|fn (a)−fm (a)|+|fm (b)−f (b)| There exists δ, minimum of the two
δn , δm that ensures that the first two modulii go lower than ε/4. There exists large enough
m, n to ensure the remaining terms go lower than ε/4 as well. All together, we have that
∃δ > 0 so that if a, b ∈ [0, 1] such that 0 < |a − b| < δ, we have |f (a) − f (b)| < 4(ε/4) = ε.
Hence, f is continuous. Therefore, C[0, 1] (and more generally, C[I]) is complete. ■
18 Metric Spaces

1.2 More basics- Closures, Relativeness, Interiors and Boundaries

Definition 1.38: Limit Point of a set E

We say p is a limit point of a set E if

(∀ε > 0)(∃qε ∈ E; qε ̸= p)(d(qε , p) < ε)

In other words, in every ε-ball around p, there would exist a point qε in E, which is
different from p.

Theorem 1.39
Every ball / neighbourhood of p which is a limit point of E, would contain infinitely
many points q such that q ∈ Bε (p) ∩ E \ {p}

Proof for Theorem.

Suppose for some neighbourhood, there only exists finite points q1 , q2 , · · · qk such that
qj ∈ Bε0 ∪ E \ {p}. Let δ < min{d(p, qj ) : j ∈ [1, 2, ...k]}. We then have that, there exists
no point q ∈ E such that its distancce from p is less than δ, making p a non-limit point.
Absurd. ■

Corollary 1.40
A finite set has no limit points

Theorem 1.41: Recharacterisation of Limit points

A point p ∈ X is a limit point of E ⊂ X if and only if there exists a sequence xn ∈ E,
xn ̸= p∀n ∈ N such that {xn } → p

Proof for Theorem.

=⇒ ) If p is a limit point, around every neighbourhood, there would exist a point qε ∈ E

such that 0 < d(qε , p) < ε. Choose ε1 = 1, and obtain x1 such that x1 ∈ E, x1 ̸= p
and 0 < d(x1 , p) < 1. Choose ε2 = 21 (d(x1 , p)). We find x2 ∈ E, x2 ̸= p such that
d(x2 , p) < d(x21 ,p) < 12 . Continue as such to obtain a sequence that converges to p.

⇐= ) Suppose there is a sequence xn such that xn ̸= p∀n ∈ N and ∀ε, ∃n0 (ε) such
that ∀n ≥ n0 we have d(xn , p) < ε which means for a given ε, there exists a point xn0 +1
in E such that it is not equal to p and it is in the ε-ball of p. Hence, p would be a limit
point. ■
Metric Spaces 19

Definition 1.42: Closed sets in X

A set E is closed in X if every limit point of E is contained in E

Definition 1.43: Equivalent definition of closed sets in X

A set E in X is closed if for every convergent sequence xn in x such that lim(xn ) ̸= xn
for any n, we have lim(xn ) ∈ E.

Definition 1.44: Open sets in X

A set E is said to be open if ∀x ∈ E, ∃ξx > 0 such that Bξx (x) ⊂ E

Theorem 1.45
Every open ball is an open set

Proof for Theorem.

Suppose a is a fixed point in X and δ > 0 is given. B = Bδ (a) := {y ∈ X : d(y, a) <
δ}.Consider arbitrary z ∈ B, for which we have d(z, a) = t < δ. Therefore δ − t > 0.
Consider 0 < ξz = r < δ − t from Density. Consider an arbitrary x such that d(x, z) <
ξz = r < δ − t. d(x, a) ≤ d(x, z) + d(a, z) = r + t ≤ δ − t + t = δ. We are done. ■

Definition 1.46: Compliment with respect to X

If E ⊆ X, we define compliment of E as

E C := {x ∈ X : x ̸∈ E}

Definition 1.47: Bounded

A set E ⊂ X is bounded if ∃ a positive number M > 0 and q ∈ E such that d(x, q) < M
∀x ∈ E. i.e, all the points of E gets contained in some ball in X.

Theorem 1.48: De Morgan’s Law

Let {Eα : α ∈ A} where A is some arbitrary indexing set represent a collection of sets
in X. Then
( ∪α Eα )C = ∩α EαC

Proof for Theorem.

20 Metric Spaces

Consider (∪α Eα )c = {x ∈ X : ∃α ∈ A : x ∈ Eα }c = {x ∈ X : ∀α ∈ A : x ̸∈ Eα } = {x ∈
X : ∀α ∈ A : x ∈ Eαc } = ∩α Eαc ■

Theorem 1.49: The Big Equivalence

E ⊂ X is open ⇐⇒ E c is closed.

Proof for Theorem.

=⇒ ) Suppose that E is open but E c is not closed. This means that there exists a limit
point of E c that falls in E, i.e, outside E c . Let this be q. This means for every ε-ball
around q, a point of E c exists. But since E is open and q ∈ E, we have for a particular
ε-ball, inside which, no point of E c resides. Contradiction.

⇐= ) Suppose E is closed but E c is not open. This means that there is a point in
E c , p, such that for every ε-ball around p, some point in E falls into this ball. But this
makes p a limit point of E, which is absurd since E is closed, limit points fall into the sets
themselves. ■

Theorem 1.50
For a collection of open sets {Gα : α ∈ A}, ∪α Gα is also an open set.

Proof for Theorem.

Consider x ∈ ∪α Gα which means ∃αx ∈ A such that x ∈ Gαx which means, there would
exist an ξ-ball around x that is contained in Gαx which is in turn contained in ∪α Gα . ■

Corollary 1.51
For any collection of closed sets Eα , ∩α Eα is also closed.

Proof for Corollary.

{Eαc } is a collection of open sets, and ∪α Eαc is an open set, which means ∪α Eαc = (∩α Eα )c
is an open set, from which we get that (∩α Eα ) is a closed set. ■

Theorem 1.52
For any finite collection of open sets {E1 , E2 , · · · Ek }, ∩ki=1 Ei is also open.

Proof for Theorem.

Suppose x ∈ ∩kj=1 Ej , which means ∀j ∈ [1, k], x ∈ Ej . We have ε1 , ε2 , · · · εk such that, the
εj -ball around x is fully contained in Ej . Choose 0 < δ < min{ε1 , ε2 , · · · , ε} (the minimum
exists by virtue of being a finite set). We see that the δ-ball around x is a subset of every
εj -ball around x, which means that the δ-ball around x is in every Ej , which proves the
theorem. ■
Metric Spaces 21

Corollary 1.53
For any finite collection of closed sets {G1 , G2 , · · · Gk } we have ∪kj=1 Gj to be closed

Remark.
In the above theorem and corollary, we require that the collection be finite. The reason
is that, we were able to get a minimal εj in the proof due to the finiteness of the set. It
may not be possible to find a number δ that is both larger than 0 but smaller than a given
infinite collection of ε-s. For example, consider the sequence of open sets (− n1 , n1 ). The
infinite intersection of these yields {0} which is a closed set by virtue of being finite.

Definition 1.54: Topology

Let X be a set and τ be a family of sets in X. X,τ is called a topology (with elements
of τ being called open sets) if:

1. Both ϕ and X are in τ

2. τ is closed under arbitrary unions

3. τ is closed under finite intersections

The complement of an element in τ with respect to X is called a closed set.

Definition 1.55: Closure of a set

Let E ′ be the set of all limit points of E. Then, the closure of E is :

Ē := E ∪ E ′

Theorem 1.56
Closure of a set is closed

Proof for Theorem.

Let p be a limit point of E ∪ E ′ . That means that ∀ε > 0 ∃qε ∈ (E ∪ E ′ ), qε ̸= p such that
qε ∈ Bε (p). If p is in E ∪ E ′ , we are done (especially if p is in E). Suppose p is not in E.
∀ε > 0 ∃qε ∈ (E ∪ E ′ ), qε ̸= p such that qε ∈ Bε (p). If the qε we recieve falls in E we are
ok. Suppose qε falls in E ′ . That means: ∀δ > 0, ∃rδ ∈ E, r ̸= qε such that d(rδ , qε ) < δ
=⇒ d(rδ , p) ≤ d(rδ , qε ) + d(qε , p) < δ + d(qε , p) < δ + ε If we choose δ0 < ε − d(qε , p) we
get: d(rδ , p) ≤ d(rδ , qε ) + d(qε , p) < δ + d(qε , p) < ε

Summarising we have: ∀ε > 0, ∃qε ∈ E or E ′ where: qε ∈ E and qε ∈ Bε (p)

22 Metric Spaces

or

∃δ(ε) > 0 such that ∃rδ ∈ E such that rδ ̸= p and rδ ∈ Bε (p). In either case, there
would exist a point dependent on ε, in E such that the point itself is different from p, and
exists in the ε-ball around p. Hence, we see that p is a limit point of E. Therefore, we see
that all the limit points of E either are points of E or points of E ′ . Hence, Ē is closed.
■

Theorem 1.57
Ē = E ⇐⇒ E is closed

Proof for Theorem.

=⇒ )Ē is closed, so E would be too.

⇐= ) if E is closed, E ′ ⊆ E =⇒ E ′ ∪ E = E = Ē ■

Theorem 1.58: Ē is the smallest closed set that contains E

If Fα is the collection of all closed sets such that E ⊆ Fα , then Ē ⊆ Fα for all α.

Proof for Theorem.

Consider an arbitrary closed set Fα that contains E. It would obviously contain all the limit
points of E among other things. Therefore, we can easily see that it contains E ∪ E ′ = Ē.
■

Fact 1.59
Topological Definition for Closure: The closure of A ⊆ X, denoted by A is defined as
the smallest closed set that contains A

Lemma 1.60: An equivalent definition for Closure.

An equivalent definition for closure is:

Ā := {x ∈ X : ∀ε > 0, Bε (x) ∩ A ̸= ϕ}

Proof for Lemma

We see that obviously, if x ∈ Ā, then either it is a point of A, or if not, it happens to be
a limit point of A. And the back implification: If q is a point of A or if it is a limit point
o A, it obviously falls into Ā. ■
Example : If E ⊆ R is bounded (and non empty), with s = sup(E), then s ∈ Ē
If s ∈ E we are done. If not, then ∀ε > 0, ∃ε > δ(ε) > 0, and a point xε ∈ E such that
s − ε < s − δ(ε) ≤ xε < s + δ < s + ε where xε ̸= s. Hence, s is a limit point of E and
Metric Spaces 23

hence, is a point in the closure. ■

Definition 1.61: Open Relative

Say E ⊆ Y ⊆ X, where X is a metric space. Y is also a metric space. We say E is
open relative to Y if ∀x ∈ E, ∃ε > 0 such that if y ∈ Y and y ∈ Bε (x) then y ∈ E.
Formally:
(∀x ∈ E)(∃εx > 0)((y ∈ Y ∩ Bε (x)) =⇒ y ∈ E)

Remark.
A set which is open relative to Y need not be open relative to X. For example, consider
R as a subset of Rn . An interval in R is open relative to R, but it is not open relative to
Rn .

Definition 1.62: Limit point relative

Say S ⊆ Y ⊆ X. x is said to be a limit point of S relative to Y if for every ε-ball
around x relative to Y (i.e, the set Bε (x) ∩ Y , or {y ∈ Y : d(y, x) < ε}) there exists a
point y ∈ S. In technical terms: x is a limit point of

(∀ε > 0)(∃y ∈ Y ∪ Bε (x))(y ∈ S)

Definition 1.63: Closed relative

A set S ⊂ Y ⊂ X is closed relative to Y if every limit point of S relative to Y is in S.

Remark.
Closed relative to Y needn’t imply closed relative to X. For example, a convergent se-
quence of S that converges in X needn’t converge in Y . As a result, if this sequence that
converges in X (but not in Y ) converges outside S, under the subspace topology of Y ,
this simply does not contribute as a limit point. Whereas, in X this contributes as a limit
point, and hence is not closed in X.

Remark.
Note that, we needn’t define the concept of "relative limit point" since every limit point
relative to Y , is a limit point of S relative to X. But the back direction needn’t be true.
i.e, you can have x a limit point of S relative to X, but not relative to Y . More concretely,
x is such that for every ε > 0, there exists z ∈ S so that z ̸= x and d(z, x) < ε. If z is in
X as well as Y , its a limit point of S with respect to both. Issue arises if x is not in Y .
Then it does not make sense.
24 Metric Spaces

Theorem 1.64
A set E ⊆ Y ⊆ X is open relative to Y ⇐⇒ ∃G ⊂ X that is open relative to X, such
that E = G ∩ Y

Proof for Theorem.

=⇒ ) Say E is open relative to Y . This means that ∀x ∈ E, ∃εx > 0 such that if
y ∈ Bεx (x) and y ∈ Y , then y ∈ E. Call G = ∪x∈E Bεx (x) which is an open set. If
z ∈ E, then z ∈ G obviously, and hence z ∈ G ∩ Y . Hence, E ⊆ G ∩ Y . Consider a point
z ∈ G ∩ Y which means z falls in one of the ε-balls around a point of E, and z is in Y .
From definition of open relativeness, we see that z ∈ E. Hence, E = G ∩ Y

⇐= ) Say E = G ∩ Y where G is an open set relative to X. Then, for every point

in G, there would exist an ε-ball around that point that is completely contained in G. Let
x ∈ E be arbitrary. ∃εx > 0 such that Bε (x) ⊂ G. Suppose y ∈ Y and y ∈ Bε (x). This
would mean that y ∈ G ∩ Y = E. Hence, ∀x ∈ E ∃ε > 0 such that if y ∈ Bε (x) and
y ∈ Y , then y ∈ E, which is the definition of open relativeness. ■

Fact 1.65
Topological definition of "open-relative": Let A ⊆ Y ⊆ X. A is said to be open relative
to Y if there exists an open set G, open in X so that A = G ∩ Y .

Theorem 1.66
A set S is open relative to Y ⊆ X if and only if Y \ S is closed relative to Y

Proof for Theorem.

Once we notice that replacing Bε (x) ∩ Y with BεY (x), i.e, the corresponding ε ball of x
with respect to Y , the proof falls in much the same way as the earlier version, done for
X. ■

Theorem 1.67
S is closed relative to Y ⊆ X if and only if S = H ∩ Y for a closed set H closed in X

Proof for Theorem.

Say S is closed, that means Y \ S = K is open relative to Y . That means there exists G
open relative to X so that G∩Y = K. Y \(K) = Y ∩K C = Y ∩(G∩Y )C = Y ∩(GC ∪Y C ) =
(Y ∩ GC ) ∩ (Y ∪ Y C ) = Y ∩ Gc . We have that Y \ K = Y \ (Y \ S) = S = Y ∩ GC .

Let S = H ∩ Y where H is closed in X. Y \ S = Y \ (H ∩ Y ) = Y ∩ (H C ∪ Y C ) = Y ∩ H C ,

we done. ■
Metric Spaces 25

Theorem 1.68
Suppose A1 , A2 · · · , An · · · ∈ X. Then,

1. If Bn = ∪ni=1 Ai , then B¯n = ∪ni=1 Āi

2. B = ∪∞ ∞
i=1 Ai , B̄ ⊇ ∪i=1 Āi with possibility of strict inequality.

and as a result we have for open sets Gi :

1. (∩ni=1 Gi )◦ = ∩ni=1 (Gi )◦

Proof for Theorem.

Suppose x ∈ ∪ni=1 Āi . Which means x ∈ Āi for some i. It is clear that x is either a point
of Ai or a limit point of Ai . We have that, whatever maybe the case either x is a point in
Bn or a limit point of Bn . Therfore B¯n ⊃ ∪ni=1 Āi .

Suppose x is a point in B¯n . If it is a point of Bn , we are done. Suppose it is the limit point
of Bn , but not a point. Also suppose that x is not a limit point of any Ai : i = 1 → n.
This means that,

∃ε1 such that ∀q ∈ A1 , q ̸= x, we have d(q, x) ≥ ε1

∃ε2 such that ∀q ∈ A2 , q ̸= x, we have d(q, x) ≥ ε2

..
.
∃εn such that ∀q ∈ An , q ̸= x, we have d(q, x) ≥ εn
If we choose 0 < ε0 < min{εi : i = 1 → n}, we would have that, for every point q in
A1 ∪ A2 · · · An , q ̸= x(which is needless to say), we have d(q, x) ≥ ε0 which makes x a
non-limit point of Bn , which is absurd. Hence we see that if x is point of Bn or a limit
point of Bn , then it is a point or the limit point of some Aj .

For a good counterexample, we look to Q := {q : q is rational}. This set is countable.

Let {q1 , q2 , · · · } be the enumeration of Q. Consider Q := ∪nj=1 {qj }. The closure of Q is
R but since these singleton sets are by definition closed, the union of them only gives you Q.

Proof for topological space:

if Bn := ∪ni=1 Ai , is it true in topological spaces (and not just metric spaces) that Bn =
∪ni=1 Ai ? the direction ∪ni=1 Ai ⊆ ∪ni=1 Ai can be shown from definition as the smallest
closed set that contains Bn . But Ai ⊆ ∪ni=1 Ai for every i, so if we union them we must
get ∪ni=1 Ai ⊆ ∪ni=1 Ai ■
26 Metric Spaces

Definition 1.69: Interior of a Set

Given S ∈ X, the interior S is defined as:

S := {x ∈ X : ∃εx > 0 such that Bεx (x) ⊂ S}

Theorem 1.70
The Interior is an open set

Proof for Theorem.

Consider (S)C := {x ∈ X : ∀ε > 0, Bε (x) ∩ S C ̸= ϕ}. It is possible that x is a point of S C ,

if it is in SC . Suppose it is a point of SC but not a point of C . From the definition, we
see that ∀ε > 0, ∃q ∈ S C , q ̸= x such that d(q, x) < ε. This makes x a limit point of S C ,
which means, for every x in SC , x is either a point of S C or a limit point of S C . Hence,

SC ⊆ S¯C

Suppose x is a point of S¯C . Say it is a point of S C , then obviously, it is a point of (S)C .

Suppose x is not a point of S C , but a limit point of S C . This means ∀ε > 0, ∃q ∈ S C , q ̸= x
so that d(q, x) < ε. This is precisely the condition for which x is a point of (S)C . Hence
we see (S)C ⊇ (S¯C ). Therefore, (S)C = (S¯C ). From here we see that S is an open set.

Alternate Argument (similar): Consider (S)C := {x ∈ X : ∀ε > 0, Bε (x) ∩ S C ̸= ϕ}

Consider a limit point p of (S)C . ∀ε > 0, ∃qε ∈ SC such that d(qε , p) < 2ε . Since qε is in
(S)C , we have that: ∀δ > 0, ∃rδ ∈ S C , rδ ̸= qε such that d(rδ , qε ) < δ
Combining these we have:

(∀ε > 0)(∃qε ∈ SC )(∃δ > 0)(∃qδ ∈ S C )

ε
(d(qδ , p) ≤ d(qδ , qε ) + d(qε , p) < + δ < ε)
2
This means p is a limit point of S C . Hence, SC is closed. ■

Theorem 1.71
S = S ⇐⇒ S is open

Proof for Theorem.

=⇒ ) if S = S, obviously S is open.

⇐= ) If S is open, then by definition S = set of all points in S so that there’s an ε-

ball of x in S. But that is every point of S. ■
Metric Spaces 27

Theorem 1.72
S is the largest open set contained in S

Proof for Theorem.

Consider an open subset of S. These are subsets of S whose each point has an ε-ball
around it so that the ball is contained in the subset, which is contained in S. So by
definition, these points in these subsets are contained in S. ■

Theorem 1.73
(S)C = (S C )
"The compliment of the interior is the closure of the compliment"

Proof for Theorem.

Refer to the proof of "Interiors of sets are Open", to see this construction.

Alternate method(slicker):

S ⊆ S =⇒ S C ⊂ (S)C where (S)C is a closed set containing S C . Since S C is the

smallest closed set that contains S C , we have S C ⊆ (S)C .

Note that S C ⊆ S C =⇒ (S C )C ⊆ S where (S C )C is an open set inside S. Since S

is the largest open set containing S, we have that (S C )C ⊆ S =⇒ S C ⊆ S C . Combining
these two set inequalities, we are done. ■
28 Metric Spaces

Definition 1.74: Isometry

(X, dX ) is said to be isometric to (Y, dY ) if there exists a function ϕ : X → Y reffered
to as the isomoetry, such that dY (ϕ(x), ϕ(y)) = dX (x, y) for all x, y ∈ X

Theorem 1.75
Let X, dX and Y, dY be isometric under the map φ : X → Y . Then the map φ is
injective, and its inverse φ−1 : ϕ(X) → X is also also an isometry.

Proof for Theorem.

Suppose φ(x) = φ(y). Then dY (φ(x), φ(y)) = dX (x, y) = 0 =⇒ x = y.

dX (φ−1 (φ(a)), φ−1 (φ(a))) = dX (a, b) = dY (φ(a), φ(b)) which makes φ−1 an isometry
from φ(X) → X. ■

Theorem 1.76
Let Z be a set and f : Z → (X, d) be any injective map from Z to X. Then δ(x, y) :=
d(f (x), f (y)) would form a metric on Z and make it a metric space, and this would in
turn make f an isometry from (Z, δ) to (X, d)

Proof for Theorem.

δ(x, y) = d(f (x), f (y)) would obey all the properties of a metric space, and moreover,
δ(x, y) = d(f (x), f (y)) making it by definition an isometry. ■

Definition 1.77: Diam

Diam(A ⊆ X) := sup{d(x, y) : x ∈ X, y ∈ X}

Definition 1.78: Distance between a point and a set

d(x, A) := inf{d(x, a) : a ∈ A}

Theorem 1.79
x ∈ A ⇐⇒ d(x, A) = 0

Proof for Theorem.

Obviously, if d(x, A) = 0, either x ∈ A or ∀ε > 0, ∃aε ∈ A, aε ̸= x such that d(x, aε ) < ε,

making x a limit point of A. The other direction is just as trivial. ■
Metric Spaces 29

Definition 1.80: Isolated point

x ∈ X is said to be an isolated point of S ⊆ X if x ∈ S and d(x, S \ {x}) ̸= 0 or rather,
there is a ball around x such that Bε (x) \ {x} is completely disjoint from S.

Remark.
In some sense, being an isolated point of a set is the opposite of being a limit point. x is
a limit point of S if for every ε > 0, there exists a point sε ∈ S \ {x} so that sε ∈ Bε \ {x}
or rather for every ε > 0, Bε (x) \ {x} ∩ S ̸= ∅. The negation of this statement is that
there exists an ε > 0 so that Bε (x) \ {x} ∩ S = ∅

Definition 1.81: Nearest Point

Let S ⊆ X and z ∈ X. If there exists s ∈ S so that dist(z, S) = d(z, s) then s is said
to be the nearest point of S to z.

Definition 1.82: Boundary Point

A point x in X is said to be a boundary point of S ⊆ X if dist(x, S) = dist(x, S c ) = 0.
Right from the definition, a point x is a boundary point of S if and only if its a
boundary point of S C . The Boundary of a set is defined as the set of all boundary
points of S denoted ∂(S).

Theorem 1.83
∂(S) = S \ S ◦

Proof for Theorem.

∂S was defined as the set of all points x so that dist(S, x) = dist(x, S C ) = 0. By virtue
of that, x ∈ ∂S implies x ∈ A. Can x ∈ S ◦ ? That would mean that there is an epsilon
ball around x ∈ S ◦ which means that the distance between x and S c would be non zero.
Hence, ∂S ⊆ S \ {S ◦ }

Now let x ∈ S \ S ◦ . Clear that dist(x, S) = 0 Obviously. S ∩ (S ◦ )C = S ∩ (S C ). This tells

us that x is in S C as well, hence dist(x, S C ) = 0, making x ∈ ∂(S). This concludes the
proof that ∂(S) = S \ S ◦ ■

Fact 1.84
Boundary of a set is closed (∂(S) = S ∩ S C ).
30 Metric Spaces

1.3 Relative closures, interiors and boundary

Definition 1.85: Relative closure (clsZ (S))

The closure of S ⊂ Z ⊂ X relative to Z, denoted as clsZ (S) is defined either as the
set S along with the set of all limit points of S in Z, or equivalently as the smallest
closed set, closed relative to Z, that contains S.

Definition 1.86: Relative Interior (intZ (S))

The interior of S ⊂ Z ⊂ X relative to Z, denoted as intZ (S) is defined either as the
set of all points z ∈ Z so that there exists an ε > 0 ball so that Bε (x) ∩ Z is fully
contained in S, or as the largest open set, open relative to Z, that is inside S.

Theorem 1.87
Let S ⊆ Z ⊆ X. Then,
clsZ (S) = clsX (S) ∩ Z

Proof for Theorem.

Note that clsX (S)∩Z is a closed set, closed relative to Z that contains S. Hence, clsZ (S) ⊆
clsX (S) ∩ Z. clsZ (S) is the smallest closed set relative to Z that contains S, which means
clsZ (S) = H ∩ Z for some closed set H. Note that this closed set must too contain S.
Hence, we have clsX (S) ⊆ H which gives clsX (S) ∩ Z ⊆ H ∩ Z = clsZ (S). Hence we have
clsZ (S) = clsX (S) ∩ Z. ■

Theorem 1.88
Let S ⊆ Y ⊆ X. Then intY (S) = intX (S ∪ (X \ Y )) ∩ Y

Proof for Theorem.

intY (S) = ∪ G open (G ∩ Y ) = (∪ G open G) ∩ Y which gives intY (S) = (∪ G open G) ∩Y =

G∩Y ⊂S G∩Y ⊂S G⊂S∪Y C
C
intX (S ∪ Y ) ∩ Y ■

1.4 Denseness, topologies and more on open / closed sets

Definition 1.89: Dense sets

A set A ⊆ X is said to be dense in X if A = X
Metric Spaces 31

Theorem 1.90
A set A is dense in X if and only if for every open set G ∈ X, A ∩ G ̸= ∅

Proof for Theorem.

=⇒ ) Suppose G0 ∩ A = ∅. That means A ⊆ GC C
0 ̸= X which means, since G0 is open, we
C
have A ⊆ G0 ̸= X.

⇐= ) if A ⊂ X properly, then (A)C is an open set disjoint from A. ■

Question: Suppose metric γ : X × X → R+ induces a topology U. Can the topology tell
us the metric? i.e, can we get back γ : X × X → R+ from the topology U?

Theorem 1.91
If U is a topology arising from δ(x, y), then the same topology can also be achieved by
the metric d(x, y) = δ(x, y)/(1 + δ(x, y)), as well as e(x, y) = δ(x, y)/2

Proof for Theorem.

Let S be an open r ball in the δ metric. Sr := {x ∈ X : δ(x, y) < r}. Let Ts be the open
s ball in the d metric. T := {x ∈ X : δ(x, y)/(1 + δ(x, y)) < s}. Now let E be the a ball
in the e metric. Ea := {x ∈ X : δ(x, y)/2 < a}. Easy to see that Er/2 = Sr and Er = S2r .
Moreover, Tr/1+r = Sr and Tr = Sr/1−r ■
The previous theorem tells us that the answer to the prior question is: no. Different
metrics can give rise to the same topologies.

Topologies on Subspaces and Superspaces

Given a topology of X, we would like to classify, or have the structure of the topology
generated in a subspace Y ⊂ X.

Theorem 1.92
If U is the topology of X, then U ∩ Y , that is {U ∩ Y : U ∈ U} is the topology of Y.

Proof for Theorem.

We know by definition (and in metric spaces, by virtue of a theorem) that S ⊆ Y ⊆ X
is open relative to Y if and only if S = G ∩ Y for some open set G relative to X. Let U
be an open set in X. That means that U ∩ Y is open relative to Y , making it part of the
topology. Let K be an open set relative to Y . That means K = G ∩ Y for some G open
relative to X. Hence, the topology of Y is such that every element in it is of the form
G ∩ Y for open G in X, and for every open G ⊆ X, G ∩ Y is part of the topology of Y .
Hence, T = {U ∩ Y : U ∈ U} ■
32 Metric Spaces

Theorem 1.93
Suppose Z is a metric space and Z ⊆ X. Topology of X includes that of Z if and only
if Z is open in X.

This essentially means that every set open relative to Z is open relative to X
if and only if Z itself is open in X.

Proof for Theorem.

⇐= ) Suppose Z is open in X. That means Z is part of the topology of X. Moreover, If
S ⊆ Z ⊆ X is open relative to Z (i.e, is part of the topology of Z) then it can be written
as G ∩ Z which is also again an open set in X. Hence, topology of Z is contained in the
topology of X.

=⇒ ) Suppose the topology of Z is contained in X, that means that every open set
relative to Z is actually an open set relative to X. That makes Z an open set relative to
X. We are done. ■
Question: Are there non empty metric spaces such that they are open relative to every
superspace?

Answer: No. Consider X to be a non empty metric space with distance d. Consider
X × {0} = Y . It is obvious that Y is bijective to X. Consider f : X → Y given by
f (x) = (x, 0). Under the obvious metric k[(x, 0), (y, 0)] = k[f (x), f (y)] = d(x, y), we can
clearly see that Y, k forms a metric space isometric to X, d. We can therefore identify X, d
with Y, k. Now look at Z = X × [0, 1] with the metric δ[(x, rx ), (y, ry )] := d(x, y) + |rx − ry |.
Consider distZ [(x0 , 0), Z \ Y ]. This is distZ [(x0 , 0), X × (0, 1]] which is 0. Hence, Y is not
open wrt Z. Its over.

Definition 1.94: Complete Metric Space

A metric space is said to be complete if X is closed in every metric superspace

Lemma 1.95: Cantor’s Intersection Lemma

X is a cauchy complete metric space if and only if for every sequence of nested, non
empty closed sets Fn ⊆ Fn−1 ⊆ Fn−2 · · · such that diam(Fn ) → 0, ∩n∈N Fn is a single-
ton.

Proof for Lemma

=⇒ ) Let X be cauchy complete. Consider x1 ∈ F1 , x2 ∈ F2 · · · xn ∈ Fn and so on. We
are told diam(Fn ) → 0, i.e ∀ε > 0, ∃n0 such that ∀n ≥ n0 , diam(Fn ) < ε. This means
that ∀n, m ≥ n0 , we have d(xm , xn ) < ε. Hence, (xn ) is a cauchy sequence. It converges
to x. Note that for any Fk , only finite points of (xn ) lie outside of Fk . Suppose x is not in
some Fk . That means it is in FkC , an open set. There is a ball around x that is completely
Metric Spaces 33

inside FkC . Around this ball, though, only finite points of xn exists. This contradicts the
fact that x is a limit of (xn ). Hence, x ∈ Fk for all k ∈ N. Hence x ∈ ∩n∈N Fn . Since
diam(Fn ) → 0, its obvious that two points cannot exist in the final intersection.

⇐= ) Let (xn ) be a cauchy sequence. Note that, if {xn : n ∈ N} has a limit point,
say x, then choose ε1 = 1 and pick xk1 in this 1-ball. Then, look at the set {xn : n ≥ k1 }.
This set still has x as its limit point, wherein you choose ε2 = 1/2 and get the corre-
sponding xk2 in the 1/2−ball of x. As such keep going to get a subsequence xkj that
converges to x. We know that ∀ε > 0, ∃n0 so that ∀n, m ≥ n0 , d(xn , xm ) ≤ ε/2, more
specifically, ∀n, kj ≥ n0 , d(xn , xkj ) < ε/2. Moreover, ∀ε > 0, ∃n′ so that ∀kj ≥ n′ we
have d(xkj , x) < ε/2. Adding these two gives us that xn → x. Therefore, we assume that
{xn : n ∈ N} has no limit point. Let Fm := {xn : n ≥ m}. Each of these sets, by virtue
of having no limit points, are closed. They are, as well, nested and obviously non empty.
Moreover, diam(Fl ) := sup{d(xn , xm ) : m, n ≥ l}. If ε > 0 is given, there exists a l0 so
that ∀n, m ≥ l0 , we have d(xm , xn ) < ε which means sup{d(xm , xn ) : n, m ≥ l0 } ≤ ε.
Hence, diam(Fn ) → 0. Hence, from hypothesis, ∩n∈N Fn = {x}. d(xn , x) ≤ diam(Fn ),
which means that xn → x. Hence, cauchy complete. ■

Theorem 1.96
Cauchy completeness ⇐⇒ completeness

Proof for Theorem.

=⇒ )Let X be a cauchy complete space. That is, every cauchy sequence in X converges
in X. Let X ⊆ Y . If (zn ) is a sequence in X, then if it converges, it must converge in X.
This is because, if it converges outside of X, that makes (zn ) a cauchy sequence, which
by definition must converge in X. Hence, X is closed with respect to Y . Since Y was
arbitrary, it is closed in every subspace.

⇐= ) Suppose X is closed with respect to every superspace. Suppose (xn ) is a cauchy

sequence in X. Consider the completion X ′ of X, that is, that unique superspace of X
such that clsX ′ (X) = X ′ and X ,→ X ′ is an isometric embedding. But since X is closed
with respect to X ′ , we have that X = X ′ , and the cauchy sequence in X converges in
X = X ′ . Hence X is cauchy complete. ■
An alternate version using an intermediate lemma:

Proof. Let (xn ) be a cauchy sequence in X. Let x ̸∈ X. Define δ(a, b) = d(a, b) if a, b ∈ X.

Define δ(a, x) = limn→∞ d(a, xn ). This limit exists since d(a, xn ) ≤ d(a, xm ) + d(xn , xm ) =⇒
d(a, xn ) − d(a, xm ) ≤ d(xm , xn ). The far right can be made arbitrarily small for large values
of m, n, hence d(a, xn ) is cauchy, and hence convergent. Now, look at δ(xk , x) = lk :=
limn→∞ d(xk , xn ). For arbitrarily large vaules of k, by virtue of xn being a cauchy sequence,
we can make lk arbitrarily small. Hence, δ(xk , x) → 0 meaning, xk → x (under the metric
space X ∪{x}, δ). All that is left to show is that this function δ is actually a metric. Let x ̸∈ X
and y ∈ X. We define δ(x, y) := limn→∞ d(y, xn ). δ(x, y ′ ) := limn→∞ d(y ′ , xn ). Positivity is
obvious since we assert that xn is actually not convergent in X. Let a, b ∈ X. We check
34 Metric Spaces

triangle inequality for these. Is δ(a, b) = d(a, b) ≤ δ(a, x) + δ(b, x) = limn→∞ d(a, xn ) +
limn→∞ d(b, xn )? We have the RHS limn→∞ (d(a, xn ) + d(b, xn )) which would be larger than
d(a, b). Now let a ∈ X and x = x. Let b ∈ X. We want to show δ(a, x) = limn→∞ d(a, xn ) ≤
δ(a, b) + δ(b, x) = d(a, b) + limn→∞ (b, xn ) In essence we want to show that limn→∞ d(a, xn ) ≤
limn→∞ (d(a, b) + d(b, xn )). Notice that the right hand side is always larger than d(a, xn ),
which is still preserved in the n → ∞ limit. Hence, triangle inequailty is verified. Hence, δ
is a metric.
Example : Two disjoint closed subsets X, Y of R that have d(X, Y ) = 0
Consider the sequence of irrational numbers x1 , x2 , · · · that is monotonically increasing
and divergent (to +∞). Let this be X. Consider another sequence, likewise monotonic
increasing and likewise divergent, yn of rational numbers obeying the property d(xk , yk ) <
1/k. We are done. ■

Lemma 1.97
If A is a set in a metric space X, and A is closed, then A can be written as the union
of the set of isolated points of A and the set of limit points of A (disjoint union).

Proof for Lemma

That it would be disjoint union is obvious, since isolated points of A are precisely not
limit points of A. Since A is closed, A′ (derived set) is a subset of A. A = (A \ A′ ) ∪ A′ .
(A \ A′ ) := {x ∈ A : ∃εx such that Bε (x) \ {x} ⊆ AC } is precisely the set of isolated points
of A. We are done. ■

Definition 1.98: Nowhere dense sets

◦
D ⊆ X is said to be nowhere dense if (D) = ∅

Lemma 1.99
D is nowhere dense =⇒ its complement is dense. Converse needn’t be true

Proof for Lemma

◦
=⇒ ) Let D = ∅. That also means that A◦ = ∅. Consider AC , AC = (A◦ )C = X. Done.

̸ ⇐= ) Consider Q which is everywhere dense. Its compliment is unfortunately not nowhere

dense. ■
Metric Spaces 35

2 Compactness

Definition 2.1: Open Conver

A collection of open sets Gα ⊂ X is an open cover of a set E if E ⊂ ∪α Gα

Definition 2.2: Compact Set

A set E ⊂ X is said to be Compact if Every open cover has a finite subcover. i.e,
for every collection of open sets Gα , if E ⊆ ∪α Gα , then there would exist a finite sub
collection {Gα1 , Gα2 · · · Gαk } of {Gα } such that E ⊆ ∪ki=1 Gαi

Remark.
The notion of Being open depends largely on the metric space one is talking about. For
example, we see that certain sets may be open relative to Y ⊂ X, but not X in itself.
This is not the case for compactness though, as shall be seen.

Theorem 2.3: "Compact Relativeness" is conserved.

Definition: We say E ⊆ Y ⊆ X is compact relative to Y if for every open cover
Gα open relative to Y we have a finite sub collection Gαk of Gα such that E ⊆ ∪kj=1 Gαj .

Theorem: E ⊆ Y ⊆ X is compact relative to Y ⇐⇒ E is compact relative

to X

Proof for Theorem.

=⇒ )Suppose E is compact relative to Y . This means that, for any collection of sets Fα
which are open relative to Y (i.e, Fα = Gα ∩ Y where Gα is an open set in X), there exists
a finite sub collection Fα1 , Fα2 · · · Fαk such that E ⊆ ∪ki=1 Fαi . Consider an open cover Hα
of E open relative to X. E ⊆ ∪α Hα , but also, E ⊆ (∪α Hα ) ∩ (Y ) since E is subset of Y
as well. This implies E ⊆ ∪α (Hα ∩ (Y )). {Hα ∩ Y } is an open cover of E open relative
to Y which means there would be a finite sub collection {Hαj ∩ Y : j ∈ [1, k]} such that
E ⊆ ∪kj=1 (Hαj ∩Y ) = (∪kj=1 Hαj )∩Y . Since E is a subset of Y , we then have E ⊆ (∪kj=1 Hαj )
which proves that for an arbitrary open cover open relative to X, we have a finite subcover.

⇐= ) Suppose E is open relative to X. Consider an open cover of E open relative to

Y , which is {Fα }. This means that Fα = Gα ∩ Y for Gα open relative to X. E ⊆ ∪α Fα =
(∪α Gα ) ∩ Y . Since E is a subset of Y , we have E ⊆ (∪α Gα ). Therefore, there would be
a finite subcollection of {Gα }, {Gα1 , Gα2 · · · Gαk } such that E ⊆ ∪ki=1 Gαi . This means,
E ⊆ ∪ki=1 Gαi ∩ Y = ∪ki=1 Fαi . Hence, for every open cover open relative to Y , there exists
a finite subcover.
■
36 Metric Spaces

Remark.
Note that the above theorem "compact-relativeness is conserved" did not use any property
of metric spaces, like distances or balls, so this theorem holds in any general Topological
space.

Fact 2.4
Every finite set in X is compact

Proof. Consider an open cover Gα for finite set E. This means that, for every point
x1 , x2 , · · · , xk in E, there would exist some {α1 , α2 , · · · } collection of "α-s" that is utmost
finite, such that xj ∈ Gαj . Simply take the union of Gαj to get a finite subcover.

Theorem 2.5: Alternate definition for compactness

A set E is compact if for every closed collection of sets Kα such that ∩α Kα ⊂ E c , we
have a finite subcollection {Kα1 , Kα2 · · · Kαp } such that ∩pi=1 Kαi ⊂ E c

Theorem 2.6
Closed balls in X are closed

Proof for Theorem.

Consider B[ε] (p) := {x ∈ X : d(x, p) ≤ ε}. B c = C := {x ∈ X, d(x, p) = txp > ε}.
Consider an arbitrary point x ∈ C. We have d(x, p) = txp > ε. Find, from density, a
δ such that txp > δ > ε. Let d(x, y) < δ − ε. We then have from triangle, d(y, p) ≥
d(x, p) − d(x, y) > txp − (δ − ε) > ε. Hence, y is also in C. Therefore, C is open, which
means B is closed. ■

Definition 2.7: diam(A)

diam(A) := sup{d(x, y) : x, y ∈ A}
It is finite if and only if it is bounded. It would be infinite if unbounded.

Lemma 2.8
If xn is a cauchy sequence and {xn } the set contains a limit point, then the cauchy
sequence converges.

Proof for Lemma

∀ε/2 > 0, ∃n0 so that ∀n, m ≥ n0 , d(xn , xm ) < ε/2. Suppose it has a limit point x.
Around every ε/2 ball of x, there exists infinite points of {xn }. Choose x1 , and then make
Metric Spaces 37

Figure 1.1: Figure for the proof: Closed balls are closed

δ < min{d(x1 , x)} and get a corresponding xn2 . Then choose δ < min{xn2 , x1 } and keep
going to create a subsequence of {xn } that converges to x. ∀ε/2 > 0 ∃n′ so that ∀nk ≥ n′
we have d(xnk , x) < ε/2. Therefore, ∀ε > 0, ∃q = max n′ , n0 so that ∀n, nr ≥ q, we have
d(xn , xns ) < ε/2 and d(xns , x) < ε/2, Adding these two we get the desired result. ■

Theorem 2.9
A metric space X is complete if and only if for every sequence of nested non-empty
closed sets {Fn : n ∈ N} such that limn→∞ diam(Fn )=0, there exists one point x ∈
∩∞i=1 Fi

Proof for Theorem.

=⇒ ) Say X is complete. Every cauchy sequence converges. Consider an arbitrary nested
sequence of closed sets so that Fn ⊆ Fn−1 . Choose xn1 = x1 ∈ F1 . Choose x2 ∈ F2 \ (F1 )
(if it exists). If not, move on to the next set, and so on. If every set contains only one
point, then we are done. If not, we are able to find xn2 ∈ Fn2 such that xn2 ̸= xn1 . As
such create a sequence xnk ∈ Fnk where nk < nk+1 (which gives Fnk+1 ⊂ Fnk ). Is xnk a
cauchy sequence? Let ε > 0. Since diam(Fnk ) → 0, we have that there exists some k ∈ N
such that for every nk ≥ k, we have 0 ≤ diam(Fnk ) < ε. Let q, p ≥ this k. That means
xnq and xnp are both in Fnk which means d(xnq , xnp ) ≤ diam(Fnk ) < ε which means this
sequence is cauchy, hence convergent. Hencce a limit x exists. Is x ∈ ∩∞ i=1 Fi ? If so, we
∞
are done. If not, it must exist in ∪i=1 Fi . This is a countable union of open sets Fic which
C

means it is finally in an open set. Outside any Fj , only finite points of the sequence exists.
This immediately contradicts the previous result, since x is a limit point of the sequence,
but it is in some Fjc which contains only finite points of the sequence so it is impossible
for a particular (chosen) ε−ball to contain infinite points of {xnk }. Hence, we are done.
x ∈ ∩∞ i=1 Fi . Moreover, since diam(Fn ) → 0, it is trivial to see that only one point survives
38 Metric Spaces

the intersection. (If there are two points,we can always make epsilon smaller than their
fixed distance).

⇐= ) Suppose {xn } is a cauchy sequence, i.e, ∀ε > 0 ∃n0 so that for all m, n ≥ n0 ,
d(xn , xm ) < ε. Look at Sm := {xn : n ≥ m}. If any of these sets has a limit point, then
we would be done from the previous lemma. So we assume these all dont have a limit
point. Hence they are closed, nested sets. Moreover, ∀ε > 0, ∃n0 so that ∀n, m ≥ n0 (or
rather, for any two points x and y in Sn0 ) we have d(xn , xm ) < ε(or d(x, y) < ε) which
means diam(Sn0 ) < ε. This means diam(Sn ) → 0. This means that ∩∞ i=1 Si = {x} for
some element x. x is in every Si . Let ε > 0. There exists Snε so that diam(Snε ) < ε which
means d(xnε , x) < ε which means xnε forms a convergent subsequence, which means the
sequence is finally convergent from previous lemma. ■

Theorem 2.10
Compact sets are closed.

Proof for Theorem.

Method 1:

Let E be compact. Consider a point p ∈ E c . Let εx be the "half" distance between

a point x ∈ E and p. Therefore, Bεx (x) is completely outside Bεx (p). Consider ∪x∈E Bε (x)
which is an open cover for E. This means there is a finite subcover
{Bεx1 (x1 ), Bεx2 (x2 ), Bεx3 (x3 ) · · · , Bεxk (xk )} such that E ⊂ ∪ki=1 Bεxi (xi ). Bεxi (p) does
not intersect with Bεxi (xi ). Therefore, ∩ki=1 Bεxi (p) does not intersect with any Bεxi (xi )
for any i. Hence, it does not intersect with ∪ki=1 Bεxi (xi ) which means ∩ki=1 Bεxi (p) lies
completely outside E. If we choose δ < min{εx1 , εx2 · · · εxk }, we would have Bδ (p) ⊆
∩ki=1 Bεxi (p). This means that, for p outside E, there would exist a δ such that the δ-ball
around p is fully contained in E c . This means that E c is open, hence, E is closed.

Method 2:

Consider E to be compact, i.e, for every closed collection {Fα } such that ∩α Fα ⊂ E c ,
there exists a finite sub collection {Fα1 , Fα2 · · · Fαk } such that ∩kj=1 Fαj ⊂ E c . Consider a
point p outside E, i.e, in E c . Notice that ∩ε∈R+ B[ε] (p) = {p} which is in E c . This would
be a collection of closed sets whose intersection falls completely inside E c . Hence, there
would exist a finite subcollection such that ∩kj=1 B[εj ] (p) ⊂ E c which means there would
exist a neighbourhood around p which is completely in E c . Hence, E c is open, and E is
closed.
■

Fact 2.11
ϕ and X are both open and closed.
Metric Spaces 39

Theorem 2.12
Closed subsets of compact sets are compact

Proof for Theorem.

Consider K ⊂ E where E is compact and K is closed. K C is, therefore, open. Consider an

arbitrary open cover {Fα } for K. since K ⊆ ∪α Fα , and K c is open, we have X = ∪α Fα ∪K c
which means E ⊂ ∪α Fα ∪ K c . Since E is compact, there would exist a finite subcover
such that E ⊂ ∪nj=1 Fαj ∪ K c . We then have K ⊂ ∪nj=1 Fαj ∪ K c , which would mean
K ⊂ ∪nj=1 Fαj , whence, we see that K is compact. ■

Corollary 2.13
If F is closed, and K is compact, then F ∩ K is compact

Fact 2.14
A compact set is bounded

Proof. Consider (WLOG, a non empty compact set E) and an arbitrary point q in X. Bε (x)
for every ε > 0 forms an open cover for E (since it is basically X). Which means there is a
finite subcover, i.e, a number ε0 > 0 such that E ⊂ Bε0 (p) which makes E bounded.

Theorem 2.15
Finite union of compact sets is compact

Proof for Theorem.

Let K1 , K2 , · · · Kr be r compact sets. Let K = ∪ri=1 Ki . Consider an open cover Fα whose

union subsumes K. We have that, for every i ≤ r, Ki ⊂ ∪α Fα . Since each Ki is compact,
there exists a finite of Fα whose union subsumes Ki . For each i = 1, to r, we have a finite
subcollection, therefore, taking the union of all these finite subcollections gives us a finite
subcollection which subsumes whole of K. Hence K is compact.
■
Remark.
Note that finiteness in the above theorem is important. This is because, each compact
may have finite subcollection, but at the end, the union of all these finite collections will
be countable, not finite.
40 Metric Spaces

Theorem 2.16
If {Kα } is a collection of compact sets such that for every finite subcollection
{Kαj : 1 ≤ j ≤ k} we have that ∩kj=1 Kαj ̸= ϕ. Then ∩α Kα ̸= ϕ. In pithy words:

"If you have a collection of compact sets for which every finite subcollection’s
intersection is non-empty, the intersection of the whole collection is non empty"
-Krishna, to Arjuna

Proof for Theorem.

Suppose, on the contrary, let ∩α Kα = ϕ which means ∪α Kαc = X which means for every
for some α0 , we have Kα0 ⊂ ∪α Kαc , where ∪α Kαc is an open cover of Kα0 . This implies
that there exists a finite subcollection {KαC1 , KαC2 · · · KαCr } such that Kα0 ⊂ ∪rj=1 KαCj =⇒
∩rj=1 Kαj ⊂ KαC0 . But this means ∩rj=1 Kαj ∩ Kα0 = ϕ, which is absurd since all finite
intersection is non empty.
■

Corollary 2.17
If K1 , K2 , · · · is a sequence of non-empty compact sets such that · · · Kn ⊂
Kn−1 · · · K3 ⊂ K2 ⊂ K1 , then ∩∞i=1 Ki is non empty.

Theorem 2.18: Compactness =⇒ Limit point Compact

If K is a compact set and E is an infinite subset of K, then E has a limit point in K

Proof for Theorem.

Suppose that E has no limit point in K. Since K is closed, E must have no limit points.
Hence, E is closed. Since closed subsets of compact sets are compact, E is compact. If no
point of E is a limit point of E, then ∀x ∈ E, ∃εx > 0 such that no point of E apart form
x itself falls into the εx -ball of x. Consider the open cover {Bεx (x) : x ∈ E} of E. This has
a finite subcover {Bεx1 (x1 ), Bεx2 (x2 ), · · · Bεxl (xl )}. We see that E ⊆ ∪lj=1 Bεxj (xj ). But
since for every εxj -ball around xj , no point in E except xj resides, ∪lj=1 Bεxj (xj ) will have
utmost finite points. Since an infinite set E cannot be the subset of a finite set, we have
a contradiction. ■

Definition 2.19: k-cell

a k-cell, E is a set in Rk such that E := {⃗x = (x1 , x2 , · · · , xk ) ∈ Rn : aj ≤ xj ≤
bj for given aj and bj for every 1 ≤ j ≤ k}

A k-cell is basically a k dimensional cuboid.

Metric Spaces 41

Theorem 2.20
k-cells are closed

Proof for Theorem.

Consider a k-cell E. Consider a point z not in E, i.e, ∃j0 such that either zj < aj or
zj > bj . WLOG, take the case of zj < p aj . Let 0 < δ < (aj − zj ). Consider a point q in the
δ-ball around z. i.e, d(z, q) < δ =⇒ (z1 − q1 )2 + (z2 − q2 )2 + · · · + (zk − qk )2 < δ =⇒
(z1 − q1 )2 + (z2 − q2 )2 + · · · + (zk − qk )2 < δ 2 < (aj − zj )2 =⇒ 0 < (qj − zj )2 < (aj − zj )2
=⇒ qj < aj . Hence q ̸∈ E, which implies there exists, for every x in E c , a δ for which the
δ-ball around x is fully contained in E c which means E c is open. This implies E is closed.
Same argument applies for the case where zj > bj . ■

Theorem 2.21
Closed intervals in R are compact

Proof for Theorem.

Let I be, WLOG, [−a, a]. Suppose it is not compact. i.e, There is an open cover Gα of I
such that there exists no finite subcover. ∀x ∈ I, ∃αx such that x ∈ Gαx and ∃εx such that
Bεx (x) ⊂ Gαx . ∪x Bεx (x) ⊂ ∪α Gα is an open cover for I. note that, if no finite subcover
for Gα exists, then no finite subcover for Bεx (x) exists either. So we can safely work with
Bεx (x). Split the interval into two halves, [−a, 0] and [0, a]. One of these intervals is not
finitely covered by Bεx (x), for if not, the whole thing would be finitely covered. let that
interval which is not finitely covered be I1 . This interval’s size is a. Split this interval
into two again. Yet again, one of the halves must not be finitely covered, for if not, I1
would be finitely covered, which is contradictory. Let this interval be I2 . This is of size
a
2
. Yet again, keep doing this process to obtain a sequence of intervals Ij , sized aj , which
are not finitely covered. These are nested intervals, non empty, and closed. From nested
intervals theorem, we see that a point ξ exists in ∩∞ j=1 Ij . ξ is a point in I, and there is a
a εξ
corresponding εξ . Consider that j0 for which j0 < 2 . We know from Archimedean such a
j0 exists. This means that the interval Ij0 containing ξ, sized ja0 , is completely inside the
εξ -ball around ξ, which means it is finitely covered. Contradiction. Hence, I is compact.
■

Corollary 2.22
Since intervals of the form [−a, a] are compact, every closed interval of the form [a, b]
is compact since it would be a closed subset of an interval of the form [−x, x].

Generalisation:
42 Metric Spaces

Theorem 2.23
n-cells are compact

Proof for Theorem.

Consider K := {⃗x ∈ Rn : −a ≤ xj ≤ a; ∀j ≤ n} to be non-compact. There is an open
cover Gα of K such that there exists no finite subcover. ∀x ∈ K, ∃αx such that x ∈ Gαx
and ∃εx such that Bεx (x) ⊂ Gαx . ∪x Bεx (x) ⊂ ∪α Gα is an open cover for K. note that,
if no finite subcover for Gα exists, then no finite subcover for Bεx (x) exists either. So
we can safely work with Bεx (x). Till here, everything is the same as the 1-d case. Note
that here, the n-cell is constructed by taking the cartesian product of n- intervals in R of
the kind [−a, a]. Construct 2n subdivisions of K by halving each interval [−a, a] in the
construction of K. The total number of subdivisions we make would be 2 × 2 × · · · × 2,
n times (simple combinatorial argument: for each i, there exists 2 choices, the two half
intervals, for crossing. From i = 1, you have 2 choices, likewise, j = 2, 3, · · · n). We assume
that atleast one of these 2n subdivisions are not finitely covered by {Bεx (x)}. We let this
one be K1 , whose each interval size is now a. Subdivide this yet again into 2n subsets,
and assert that one of these subdivisions is not finitely covered. Call this K2 , whose each
interval is of size a2 . Construct a sequence of sets Kj , each of whose intervals are sized aj .
Each Kj is closed and non empty, hence compact, and are nested. Therefore, ∩∞ j=1 Kj ̸= ϕ.
∞
Let p ∈ ∩j=1 Kj . For this p, there would exist a εp and the corresponding ball Bεp (p). We
require one of our Kj n-cell to fall into this εp ball. Let δ be smaller than ε2p . Let p be the
centre of the δ−ball. Let p be the centre of the n-cube H in the following√construction:
Consider w to be the side length of H. We require the diagonal length w n = δ, which
gives us w = √δn . Consider H such that each side is the interval [pj − w2 , pj + w2 ]. This
would force p to fall in the centre of the n-cube H. This cube is fully contained in the δ
ball of p which is contained in the εp ball of p. We consider that n cell Kj for which each
side aj < w. This can be found, and hence, the this Kj cell is finitely covered by Bεp (p),
which is absurd. Hence, K is compact. ■

Remark.
We proved the result for k-cells of the kind [−a, a]k , but it is easily generalised by noting
that arbitrary k-cells are contained in some k-cell of the above kind. By virtue of being
closed, they also are compact.

Theorem 2.24: Heine-Borel

Given a set E ⊂ Rn , the following are equivalent:

1. E is closed and bounded

2. E is compact

Proof for Theorem.

Metric Spaces 43

Figure 1.2: Figure for the proof: n-cells are compact.(The εp -ball around p, and the n-cell
construction)

⇐= ) We know that all compact sets are closed.

=⇒ ) If E ⊂ Rn is closed and bounded, it is contained in some n-cell, which is com-

pact. By virtue of being a closed subset of a compact set, E is also compact. ■

Theorem 2.25
If {xn } is a sequence in X convergent to x ∈ X, then the set {xn } has only one limit
point, which is x.

Proof for Theorem.

That x is a limit point is clear. We note that ∀ε > 0, ∃n0 such that ∀n ∈ N, n ≥ n0 we
have d(xn , x) < ε. i.e, beyond a particular n0 , every point of {Xn } falls in the ε-ball of
x. Therefore, only finite points lie outside this ε-ball of x. Suppose it has another limit
point y, other than x. Therefore, there would exist a δ such that the δ-ball around y lies
completely outside the δ-ball around x. This means that, only finite points of xn lie in the
δ ball of y, making it unviable to be a limit point. ■
Extending Heine Borel we have:
44 Metric Spaces

Theorem 2.26: (Extension)

For a subset E ⊂ Rn , the following are equivalent:

1. E is closed and bounded

2. E is compact

3. every infinite subset K of E has a limit point in E

Proof for Theorem.

(1) =⇒ (2)) Heine Borel

(2) =⇒ (3)) Already seen

(3) =⇒ (1)) Let us assume that E is either not closed, or not bounded. We start
by assuming it is not closed. Which means that ∃q outside E such that there exists a
sequence in E that converges to q. We take this sequence {xn } as our infinite set, and we
see that, from the previous theorem, this has only one limit point q, which lies outside E.
Hence, there exists an infinite set {xn } which has no limit point in E.

Suppose that E is unbounded. We then have that, ∀x ∈ X, ∀ε > 0, ∃y ∈ E such

that d(x, y) > ε. Fix some x0 in X. Choose some y0 that is a distance z00 = d(y0 , x0 )
away from x0 . Look if there is a point y1 so that its distance from x0 is more than z00
but less than 2(z00 ). If it doesn’t exist, check for less than 3(z00 ). Find some k1 (z00 ) so
that distance of y1 to x0 is more than z00 but less than k1 (z00 ). Same way, for y2 , find
y2 so that its distance from x0 is more than k1 (z00 ) but less than some other k2 (z00 ).
Inductively, find yj whose distance is more than kj−1 (z00 ) but less than kj (z00 ). Note that
1 < k1 < k2 < · · · . Hence, for any ε-ball around x0 , only finite yj exists in that ball, since
there would exist some kq (z00 ) and kq−1 (z00 ) between which ε lies. And inside kq−1 (z00 )
ball around x0 , utmost finite points yj exists. Hence, x0 is clearly not a limit point for
the set of yj -s. Consider any other point a ∈ X. For some every ε-ball around x0 , only
finite points exists. For some, perhaps larger δ-ball around a, a chosen ε-ball around x0
gets subsumed into the δ-ball around a. This implies that only finite points of yj -s exists
in the δ-ball around a as well, making a a non viable limit point. We see that, for this
infinite subset {yj } of E, no limit point exists. ■

Remark.
In the previous proof, we note that (3), which is called Limit Point Compactness, implies
(1) Closed and Bounded, in any metric space, not just Rn , as we see in the proof, no
property of Rn was used.

Spoiler Alert: In any metric space, Limit Point Compact ⇐⇒ Compact

Metric Spaces 45

Figure 1.3: Figure for proof: Lim point compact =⇒ closed+bounded. (The construction
of an unbounded sequence)

Theorem 2.27: Weierstrass Theorem

Every Bounded, infinite set in Rn has a limit point in Rn .

Proof for Theorem.

If a set is bounded in Rn , it is the subset of a compact set (i.e, a closed and bounded set).
From the previous equivalence, an infinite subset E of a compact set has a limit point in
the compact set, which means the bounded, infinite set we have has a limit point in the
compact set that contains it, hence, it has a limit point in Rn . ■

Remark.
The above "Weierstrass Theorem" is just the "Bolzano-Weierstrass" Theorem we saw in
sequences. Actually, the "Bolzano-Weierstrass" Theorem is a direct corollary of the more
general "Weierstrass Theorem". Let {xn } be any sequence in R that is bounded. This
means that this sequence is the subset of a compact set, hence, has a limit point in R.
This implies, a subsequence of {xn } converges in R. Hence, every bounded sequence has
a convergent subsequence.

Fact 2.28
Let X = Rn . The closure of any open ball is the corresponding closed ball.

Proof. Consider B := Bδ (x0 ) := {y ∈ Rn : ||y − x0 || < δ}. Let z0 be a point on the rim
of B, i.e d(x0 , z0 ) = δ. Such a point obviously exists. Consider ⃗γ (t) = t⃗z0 + (1 − t)⃗x0 with
46 Metric Spaces

t ∈ (0, 1). For every t ∈ (0, 1), ⃗γ (t) belongs in B. To see this, consider ||⃗γ (t) − ⃗x0 || =
||t⃗z0 + (1 − t)⃗x0 − ⃗x0 || = ||t(⃗z0 − ⃗x0 )|| = t||⃗z0 − ⃗x0 || < t(δ) < δ. Suppose we are given
an arbitrary η > 0. Does there exist a t ∈ (0, 1) so that ⃗γ (t) belongs in the η-ball of ⃗z0 ?
i.e, we need a t so that ||⃗γ (t) − ⃗z0 || = ||t⃗z0 + (1 − t)⃗x0 − ⃗z0 || = ||(1 − t)(⃗z0 − ⃗x0 )|| < η
=⇒ (1 − t)||⃗z0 − ⃗x0 || < η =⇒ 1 − t < ηδ =⇒ 1 − ηδ < t < 1. Such a t exists for
every η. Hence, ⃗z0 is a limit point of B (by virtue of there existing a sequence of ⃗γ (tj ) that
converges to ⃗z0 ). Hence, every point on the rim is a limit point. Moreover, no point w so
that d(w, x0 ) > δ is a limit point of B, since there would exist an ε-ball around w so that no
point of B falls into it (from openness). Hence, closure of B is the corresponding closed ball,
in Rn .

Definition 2.29: Dense sets

A set A ⊆ X is said to be dense in X if every point of x is a limit point of A.

Theorem 2.30
A is dense in X if and only if A = X if and only if for every open set U of X, A ∩ U ̸= ∅

Proof for Theorem.

=⇒ ) Let A be dense in X. Let U be an open set such that A ∩ U = ∅ which means A is
completely contained in U C which is a closed proper subset of X, which is a contradiction
since A ⊆ U C since it is the smallest closed set that contains A. Contradiction.

⇐= )Suppose A is not dense in X, this means A ⊂ X which means (A)C is a non

empty open set in X that necessarily does not intersect with A. Contradiction. ■

3 Perfect Sets

Definition 3.1: Perfect Set

A set E ⊂ X is perfect if every point of E is a limit point of E, and E is closed

Theorem 3.2
Perfect subsets in Rn are uncountable.

Proof for Theorem.

Suppose E is a perfect set in Rn but is countable. i.e, it can be enumerated as E =
{x1 , x2 , · · · }.

Choose x1 , and ε0 = 1. Let V0 denote the ε0 -ball around x1 . This ball is non empty,
Metric Spaces 47

moreover, V¯0 ∩ E(which is the corresponding closed ball of V0 ) is non empty, and is com-
pact by virtue of being closed and bounded. Inside, V0 ∩ E, there exists infinite points of
E, since x1 is a limit point of E.

Choose an arbitrary point z1 in V0 that is not x1 . Now let ε1 < d(x1 , z1 ). Let V1 be
the ε1 -ball around z1 . Notice the following: z1 is a limit point of E, hence, there are
infinite points of E in V1 . x1 is not in V¯1 . V¯1 ∩ E is closed, bounded and non empty, hence
Compact.

Choose a point z2 in V1 that is not x2 , and let ε2 < min{ε1 , d(x2 , z2 )}. Let V2 be the
ε2 -ball around z2 . Note that, x2 is not in V¯2 . Also note yet again that there are infinitely
many points of E in V2 . It is crucial to note now that V¯2 ∩ E ⊂ V¯1 ∩ E ⊂ V¯0 ∩ E.

Suppose you have already constructed Vk by finding zk in Vk−1 that is not xk and an
εk < mind(zk , xk ), εk−1 such that xk ̸∈ V¯k , V¯k ∩ E is compact, non empty and V¯k ∩ E ⊂
¯ ∩ E ···.
Vk−1

Now, choose zk+1 ̸= xk+1 , inside Vk . Choose εk+1 < min{d(zk+1 , xk+1 ), εk }. Let Vk+1
¯ ∩ E is non empty, xk+1 is not in
be the εk+1 -ball around zk+1 . Yet again, we see that Vk+1
¯ , and Vk+1
Vk+1 ¯ ∩ E ⊂ V¯k ∩ E. Hence, we have a sequence of non empty, nested compact
sets. This implies that ∃ξ ∈ E ⊂ Rn such that ξ ∈ ∩∞ i=1 (V̄i ∩ E). Is ξ any one of xj
enumerated? No, because if it was, from the construction, xj would not belong in Vj .
Hence, ξ is not in the enumeration of E. Contradiction. ■
48 Metric Spaces

Figure 1.4: Figure: Perfect sets are uncountable. Construction of the nested sequence of
compact sets by choosing zk ̸= xk ∈ Vk−1 .

Remark.
It is easily seen that, closed intervals in R are perfect: From density theorem, for every
point in I, there would exist a sequence of rationals converging to that point. Moreover,
closed intervals in R are closed since closed balls in metric spaces are closed. Therefore,
we see that intervals are uncountable.

3.1 The Cantor Set

The following is the construction of an uncountable, perfect set that contains no intervals:
The Cantor Set.

Let I0 = [0, 1]. size of the interval(s) in I0 is 1, and there are 20 = 1 intervals.

Let I1 = [0, 31 ] ∪ [ 23 , 1] be constructed by trisecting I0 and tossing the middle one. Here,
we have each interval sized 311 , and there are 21 = 2 intervals total.

Let I2 = [0, 19 ] ∪ [ 29 , 39 ] ∪ [ 96 , 79 ] ∪ [ 98 , 1] be generated by taking each of the two sub inter-

Metric Spaces 49

vals in I1 , trisecting them, and tossing the middle one, and joining them finally. We have
each interval sized 312 and there are 22 = 4 intervals total.

Inductively keep making these trisections+tossings to make a sequence of closed, nested

intervals (Compact, too) Ik , each containing 2k intervals each of size 31k .

Finally, define the Cantor set P as

P := ∩∞
i=1 Ii

Note that P is compact since it is the closed subset of a compact set. It is also non empty
by virtue of being the intersection of a sequence of nested, non empty, compact sets.

Note that, no interval of the kind [a, b] exists in the Cantor Set. The size of each inter-
val in Ij is 31j . We can find j so that 31j < b − a =⇒ b−a 1
< 3j =⇒ log3 ( b−a 1
) < j. For
such Ij , we notice that [a, b] has "inbetween" points that doesn’t exist in any of Ij ’s intervals.
Hence, taking the intersection, these "inbetween" terms don’t survive. Hence, no intervals
exist.

Theorem 3.3
The Cantor set P is perfect.

Proof for Theorem.

We already know that the Cantor set is closed. We need to show that every point in the
cantor set is a limit point. First, observe that, for any Ik , if z is the end point of any of
the sub interval of Ik , it survives the ∞-intersection. This is because, after Ik -s trisection,
the end points still stay endpoints. Let ξ be any point in the cantor set, which means
it is a point in every Ik . Let δ > 0 be given. Consider the interval (ξ − δ, ξ + δ). This
interval is sized 2δ. ξ exists in one of the sub intervals of Ik for all K ≥ k0 for some k0 .
Choose j so that 31j < δ. Then, the interval in Ij containing ξ would fall completely inside
(ξ − δ, ξ + δ). Choose q as one of the end points of this sub interval of Ij . Therefore,
∀ξ ∈ P , ∀δ > 0, ∃q ∈ P, q ̸= ξ so that q ∈ (ξ − δ, ξ + δ). Therefore, every ξ ∈ P is a limit
point of P . Hence, P is perfect. ■

4 Connected Sets

Definition 4.1: Separated Sets

A ⊂ X and B ⊂ X are said to be separated if Ā ∩ B = B̄ ∩ A = ϕ, i.e, they are disjoint
and no point of one, is the limit point of the other.
50 Metric Spaces

Definition 4.2: Connected Set

A set E ⊂ X is said to be connected if it is not the union of two non-empty separated
sets. In other words, for every "split" of E into two non empty sets, none of them are
separated. Even if one split of E is separated, then E is not connected.

Example : Separated =⇒ Disjoint, but Disjoint ̸ =⇒ Separated.

[0, 1]and (1, 2) are disjoint, but are not connected since a sequence in (1, 2) converges to 1
in [0, 1]. ■

Theorem 4.3
E ⊂ R is connected ⇐⇒ ∀x, y ∈ E, x < z < y =⇒ z ∈ E.

Proof for Theorem.

=⇒ ) Suppose ∃x0 , y0 ∈ E so that ∃z, x0 < z < y0 , but z ̸∈ E. Consider A := (−∞, z)
and B := (z, ∞). A and B are seen to be separated, and E is a subset of A ∪ B, which
makes it disconnected.

⇐= ) Suppose that we have E disconnected, which means it is the union of two separated
sets A and B that are non-empty. x0 ∈ A and y0 ∈ B. Consider z(t) = x0 + t(y0 − x0 ) for
t ∈ [0, 1]. Note that z(0) = x0 and z(1) = y0 .

Conjecture: There exists a tB ∈ (0, 1) so that for every t < tB , z(t) does not belong
in B. If it is not true, then for every t ∈ (0, 1), there exists a point tB < t so that z(tB )
is in B. Choose t = 1 to get z(t1 ) in B. Chooe t = t21 to get z(t2 ) in B with t2 < t1 and
t2 < 12 . Keep going with t = 2tn−1 1
n−1 to get z(tn ) in B with tn < tn−1 and tn < 2n−1 . This

gives us a sequence z(tk ) which we can see is monotone decreasing assuming x0 < y0 . This
sequence converges to z(0) which is in A which means that there exists a sequence in B,
z(tn ) that converges to A. Absurd.

In a similar vein, we can show that there exists tA ∈ (0, 1) so that for every t > tA ,
z(t) is not in A. Consider

SA := {t ∈ [0, 1] : z(t) ∈ A ∩ [x0 , y0 ]}

and
SB := {t ∈ [0, 1] : z(t) ∈ B ∩ [x0 , y0 ]}

It is easy to see that SA and SB are disjoint. If a sequence in one converges in another,
say tn ∈ SB converges to t0 ∈ SA . Then z(tn ) ∈ B by definition, for every n. But then
by definition, z(tn ) = x0 + tn (y0 − x0 ) ∈ B such that lim(z(tn )) = x0 + t0 (y0 − x0 ) ∈ A,
which means a sequence in B converges in A. Absurd. So SA and SB are separated.
Metric Spaces 51

Note that, for t > tA , no z(t) is in A. Hence, we see that for every t so that z(t)
falls in A, there is an upperbound. Likewise, for every t such that z(t) falls in B, there is
a lowerbound. Hence, SA has a supremum sup(SA ) and SB has an infimum inf (SB ).

At this point, we may as well assume that for every t < inf (SB ), t ∈ SA for if not,
what we wanted to prove would get proved. Suppose then, for argument sake, that for
every t > Sup(SA ), t ∈ SB , and likewise, for every t < Inf (SB ). t ∈ SA . Now then, does
Sup(SA ) belong in SA ? we see that for every t > sup(SA ), t ∈ SB which means we can
construct a sequence in SB using those ts, which converge to Sup(SA ) in SA . So that is
ruled out. So is Sup(SA ) in SB ? That is not possible either, since SA is a bounded, infinite
set (mainly because supremum isn’t in the set), we know that there is a monotone sub-
sequence in SA converging to Sup(SA ) which is in SB . We therefore conclude that, there
exists a point t ∈ (0, 1) so that t is neither in SA , nor in SB . This translates to there being
a point z = z(t), between x0 and y0 so that z(t) ̸∈ A∪B =⇒ ̸∈ E. We are, therefore, done.

Slicker Argument: Suppose E = A ∪ B with Ā ∩ B = B̄ ∩ A = ϕ. Consider x0 ∈ A

and y0 ∈ B and WLOG assume x0 < y0 . Define z = sup(A ∩ [x0 , y0 ]). There would be a
sequence in A that converges to z, by virtue of being the supremum. z ∈ Ā =⇒ z ̸∈ B.
This means x0 ≤ z < y0 . If z ̸∈ A, we would be done. If z ∈ A, then z ̸∈ B̄. Therefore, z
is in an open set B̄ C . There would exist an εz -ball around z so that it is fully contained
outside B̄. Choose z + ε2z as your z ′ . Note that z ′ is greater than the supremum of A. We
see that z ′ is not in B, and not in A either. Hence, we are done. ■

Figure 1.5: Figure: Proof for the equivalence for connectedness for sets in R. A look at
A ∩ [x0 , y0 ] and B ∩ [x0 , y0 ].

5 Misc Knowledge

Theorem 5.1
1. If A and B are closed, disjoint subsets of X, then A and B are separated.

2. If A and B are open, disjoint sets, then A and B are separated.

Proof for Theorem.

52 Metric Spaces

(1) A B closed implies A = A and B = B which are disjoint. From here it is obvious.

(2) A and B are open disjoint sets, then we see that A ⊆ B C and B ⊆ AC where AC and
B C are closed by definition. Since closure is the smallest closed set containing A (and B),
we see that A ⊆ B C and B ⊆ AC . It is now trivial to see that A∩B ⊆ B C ∩B = ϕ = B ∩A
which is the definition of separated. ■

Corollary 5.2
Let p ∈ X and δ > 0. Define A := Bδ (p) and B = (B[δ] (p))C . A and B are, then,
separeted.

Proof for Corollary.

Easy to see that they are both open sets that are disjoint. ■

Theorem 5.3
Every connected metric space with atleast two points is uncountable.

Proof for Theorem.

Let a and b be in X. Let ξ ≤ d(a, b). Note that, P = Bξ (a) and Q = (B[ξ] (a))C are non
empty, separated sets (from the previous corollary). If X is not the union of P and Q,
then there is a point zξ in X so that it is neither in P not in Q. That means that it is
exactly ξ distance away from p. For every ξ < d(a, b), there exists a point zξ so that its
distance from p is exactly ξ. Therefore, every zξ is unique (from positivity property of
metric spaces) which means there are uncountable zξ -s. ■

Theorem 5.4
If P and Q are connected such that P ∩ Q ̸= ϕ, then P ∪ Q is also connected.

Proof for Theorem.

Suppose P ∪ Q is actually not connected. This means P ∪ Q = A ∪ B for non empty,

separated sets A and B. Suppose P is fully contained in A. This means that Q has
intersection with A and intersection with B which are non empty. Obviously Q ⊆ A ∪ B
which means Q = (A ∩ Q) ∪ (B ∩ Q) where (A ∩ Q) and (B ∩ Q) are separated and non
empty. Since Q is connected, this is absurd. Suppose then that P is not fully contained in
A. This means that P ∩A and P ∩B is non empty each. This means P = (P ∩A)∪(P ∩B).
From the same reasoning, this is absurd. ■
Metric Spaces 53

Lemma 5.5
Given two balls B1 and B2 in Rn that are closed, with B1 ∩ B2 = {z} with z ∈ Rn ,
then the interior of B1 ∪ B2 , i.e, B1 ∪ B2 , is B1 ∪ B2 , which are their respective open
ball counterparts.

Proof for Lemma

We understand that B1 ∪ B2 ⊆ B1 ∪ B2 . Note that none of the "rim" points of B1 or B2 ,
are in the interior. This would conclude the result. ■

Corollary 5.6
If A ⊂ X is connected, it needn’t be true that A is connected.

Proof for Corollary.

Consider the set B1 ∪B2 from the previous lemma. We note that, its interior is the disjoint
union of two non empty open balls. These two sets are separated, which makes B1 ∪ B2 a
separated set. ■

Theorem 5.7
Let E be the set of all x ∈ [0, 1] ⊆ R so that the decimal expansion of x only contains
4 and 7. Then:

1. E is uncountable

2. E is not dense in [0, 1]

3. E is compact

4. E is prefect

Proof for Theorem.

(1) E is uncountable via the diagonal argument. If {x1 , x2 , · · · } is the enumeration of E,
simply take the first decimal place of x1 and flip it (i.e. to 4 if 7 or vice versa). Likewise
for x2 and so on to get a new decimal expansion that is unlike all x1 , x2 · · · xn · · · which is
a contradiction.

(2) Obviously, since 0.4 ≤ x ≤ 0.8 for any x ∈ E.

(3) We already know E is bounded. Consider E c . This is the set of all numbers in
[0, 1] so that not all points in the decimal expansion is 4 or 7. i.e, there would be a point
in the expansion that is neither 4, nor 7. Suppose we take one such arbitrary x ∈ E c . Let
the first non 4, 7 number occur at the j−th place. 0.z1 z2 z3 · · · xj · · · . Look for another
non 4, 7 after the j−th place (if it exists). If it doesn’t exist, then all the numbers after xj
54 Metric Spaces

would be 4 or 7 so safely add 10−(j+1) as our ε. This ε range around x would contain only
points of E c . Suppose another point exists that is non 4, 7 after j, perhaps at k > j-th
index. Then it would look something like: 0.z1 z2 · · · zj xj+1 xj+2 · · · zk · · · . Here we simply
take ε = 10−(k+1) so that all points in the ε-neighbourhood of x is in E c . Therefore, E c is
open, which means E is closed. Closed and bounded implies compact in R.

We saw that E is closed. We need only show that every point of E is a limit point
of E. This can be done easily for any ε-ball, using the technique that follows: Choose a
k so that 101k < ε. Look at the interval x − 101k , x + 101k that is contained in E. Just find
some n > k, and flip the 4 to a 7 or vice versa to land in a "different" element from x, yet
within the neighbourhood in consideration. Hence, every point is a limit point, making E
a perfect set.
■

Theorem 5.8: Existence of a compact set in R with countable

limit points.
Title

Proof for Theorem.

Consider the points x1 = 1, x0 = 0, x2 = 12 , · · · xn = n1 · · · . Let x11 < x12 · · · < x1n be
a sequence that converges to x1 = 1. Let x21 < x22 < x23 · · · < x2n · · · be a sequence
convergent to x2 = 21 , with the added condition that x2j < x1k for every j, k ∈ N. Likewise
for every xn , create a sequence xnk that converges to xn . Make sure that xan < xbm if
a > b, for every m, n. We therefore have:

x11 < x12 · · · → x1

x21 < x22 · · · → x2

..
.
We claim that the set Φ = x0 , x1 , x2 , · · · along with x11 , x12 · · · , x21 , x22 · · · xn1 etc. forms
a Compact set in R that has countable limit points.

Suppose that q ∈ R ̸= 0 and q ̸= 1j for any j be a limit point of Φ. This means a

subsequence zn in Φ converges to q. Does infinite points of x1j exist in the subsequence
zn convergent to q? Obviously not, since that would make a subsequence of zn convergent
to one of 1. So only utmost finite elements of x1j are in zn . Same way one can argue that
utmost finite elements of xkj are in zn for every k. If we establish a subsequence of zn
that converges to 0, then the only limit point of {zn } would be 0 which would mean 0 is
where zn would converge. Enough wishful thinking; Does any point of x1j exist in zn ? If
yes, choose that point. If not, find the next n1 so that a point in xn1 j is in zn . Find, then
n2 > n1 so that some point in xn2 j is in zn . As such keep going, making n1 < n2 < · · ·
and a sequence that is monotone decreasing by construction, that converges to 0. Hence,
this is a subsequence in {zn } that converges to 0. If zn converged to q, then the only limit
Metric Spaces 55

point would be q. Hence, q = 0. This means that the only limit points of Φ other than
1
j
is 0. Hence, we have a countable limit point. Hence, Φ is a closed set and bounded
obviously, with countable limit points. ■

Theorem 5.9: technique weve already seen

If A and B are separated sets in Rn (that are non empty), and x⃗0 ∈ A and y⃗0 ∈ B,
define p(t) = x⃗0 + t(⃗y0 − ⃗x0 ) for t ∈ [−∞, ∞] and

SA := {t ∈ R : p(t) ∈ A}

SB := {t ∈ R : p(t) ∈ B}
Then:

1. SA and SB are separated sets

2. ∃t0 ∈ (0, 1) so that t0 ̸∈ SA ∪ SB

Proof for Theorem.

(1) If SA and SB weren’t disjoint, then obviously A and B wont be. If there is a sequencei
n SA converging in SB or vice-versa, it is easy to see that this would lead to there ex-
iting a sequence in A converging to B (or vice-versa). Therefore, SA and SB are separated.

(2) Since SA and SB are separated and non-empty, there are two points x0 and y0 in
SA and SB respectively. Define z(l) = x0 + l(y0 − x0 ) for l ∈ [0, 1]. Now, for some lA , we
have that for every l > lA , z(l) ̸∈ SA , the set GA := {l : z(l) ∈ SA } is therefore bounded
above with supremum uA . Likewise for some lB , we have that for every l < lB , z(l) ̸∈ SB ,
the set GB := {l : z(l) ∈ SB } is therefore bounded below with infimum vB . Note that GA
and GB are separated sets. We may as well assume that for every l < lB , l is in GA , or
z(l) falls in SA . If not, we would be done. We now ask: does vB fall in GA or GB ? If
it falls in GB , there exists a sequence in GA that would converge to vB , which is absurd.
If it falls in GA , then by virtue of being the infimum of GB , there is a sequence in GB
converging in GA . Hence, there exists a point l between x0 and y0 l ̸∈ GA or GB which
means z(l) ̸∈ SA or SB . This again means that p(z(l)) ̸∈ A or B. Phew. ■

Corollary 5.10
Every convex set in Rn is connected.

Proof for Corollary.

If they were not connected, then there would exist sets A and B, non empty, disjoint and
separated so that our convex set C would be A ∪ B. Suppose x0 ∈ A and y0 ∈ B. From
the previous theorem we see that, if z(t) = x0 + t(y0 − x0 ), then there would exist t′ so
that z(t′ ) ̸∈ A or B, which means it wont be in C. But if x0 and y0 are in C, by definition
of convexity, z(t) for any t must exist in C. Contradiction. ■
56 Metric Spaces

Example : A Pedagogical Example.

If E := {q ∈ Q : 2 < q 2 < 3} is considered a set in the metric subspace of Q with the
usual distance, then we have:

1. E is closed and bounded (wrt Q obviously)

2. E is non compact.

3. E is open with respect to Q

Proof. Consider an arbitrary convergent sequence pn in E. We see that if pn is contained

in E, then 2 < p2n < 3. If we pass to the limit, we would have 2 ≤ q 2 ≤ 3 but the limit q
would either not exist in Q (whence the sequence pn wouldn’t be convergent anyway) or
it does, in which case it follows 2 < q 2 < 3 (since no rational number has its square as 2
or 3). Therefore, we can see that every convergent sequence in E converges in E. Hence,
E is closed. Boundedness is obvious.

One way to see that E is non compact is simply by making use of the "conservation"
of compactness going from one space to a bigger space or vice-versa. Since E is clearly
not compact in R, it wont be compact in the metric subspace Q. Another way to see
that it is non compact√is to√consider the√ following
√ construction:
√ √ Look at the union of the
two split intervals (− 3, − 2) and ( 2, 3). Let h = 3 − 2. We construct √ it for the
right side
√ interval, and simply copy it to the left one. Choose the point 3 and a h/2-ball
around
√ 3 and call it V0 . This would intersect the right interval at a distance h/2 from
3. Choose this intersection point and a h/4-ball around this called V2 . Both V1 and V2
together covers 3h/4 of the interval. Choose this intersection point and let the h/8-ball
around this point be V3 . V1 , V2 and V3 together cover 7h/8 of the interval. Keep going as
such to construct Vn so that (Vi )ni=1 covers (n−1)h
n
of the whole interval. As n tends to ∞,
the "coverage" converges to h. This means that for every ε distance away, there exists a
finite n0 so that V1 → Vn0 covers up to that ε distance. Every point e in the strict interval
would therefore be covered up by some Vk by our construction. But obviously, this "open
cover" has no finite subcover, for no finite "coverage" covers all the way till h distance of
the interval. Some ε-gap is always left, hence missing points.
√ √
E can also be written as the√disjoint union
√ of the set A of all p so that 2 <
√ p < 3√
and
so that − √
the set B of all p √ 3 < p < − 2. WLOG say q ∈ A. Obviously 2 < q < 3.
Choose δ < min{ 3 − q, q − 2} and the δ−ball called V around q. Easy to see that this
ball V is fully contained in A. Likewise, it can be shown for B as well. Hence, E is open
with respect to Q.
■

Theorem 5.11: Existence of a non-empty, perfect set in R that

contains no rational points
Title
Metric Spaces 57

Proof for Theorem.

Definition 5.12: Separability

A metric space X is said to be separable if it has a countable dense subset S.

Example : Rn is separable
Consider the set of all points z ∈ Rn with coordinates in Q. Basically Qn . This would be
a countable set, and every point x ∈ Rn is a limit point of this set (since coordinate we
can have a rational sequence converging to that coordinate) ■

Definition 5.13: Basis

A collection of open sets {Vα : α ∈ A} is said to be a basis for X, a metric space, if
for every point x ∈ X, and for every open set G so that x ∈ G, we have a Vα in the
basis so that x ∈ Vα ⊂ G. In other words, every open set G can be "covered" by a
subcollection of {Vα }

Theorem 5.14
For a metric space X, separability ⇐⇒ countable basis

Proof for Theorem.

=⇒ )Suppose X is separable, i.e, has a countably dense set {x1 , x2 , · · · , xn , · · · }. Choose

all the 1−balls of each of these points to form a countable set. Choose all the 1/2−ball
of each for another countable set. As such keep choosing 1/n-balls for each xj in the
countably dense set. We claim that this collection of open sets are my basis. Say z is an
arbitrary point in X, and G an arbitrary open set containing z. There would exist εG z so
G εG
that εz -ball of z would be contained in G. Choose δ < 4 . There would be an n so that
z

1
n
< δ. In this 1/n ball around z, there would exist a point xk in the dense set. Choose
now, the 1/n ball around xk which contains z, and would be completely contained in the
εG
z ball, which would be contained in G. Hence, we are done.

⇐= ) Say X has a countable basis {V1 , V2 , · · · } that are non empty. Choose x1 ∈ V1 ,
x2 ∈ V2 · · · . We claim that this set {x1 , x2 · · · } forms our countable dense set. Say a point
z ∈ X, choose any arbitrary ε > 0. The ε−ball around z would be an open set, which
means a set Vk is inside ε−ball around z. This would mean that xk would be in the set,
which makes z a limit point of this supposed to be dense set {x1 , x2 · · · }. We are done.
■
58 Metric Spaces

Theorem 5.15

”(∃{Vα : α ∈ A})(∀x ∈ X)(∀G ⊂ X : G open : x ∈ G)(∃α ∈ A)(x ∈ Vα ⊂ G)”

⇐⇒
∃{Vα : α ∈ A}(∀G ⊂ X : G open in X)(∃A′ ⊆ A)(G = ∪α Vα : α ∈ A′ )
Or in pithy words,

Vα is basis if and only if every open set is the union of a subcollection of {Vα }

Proof for Theorem.

=⇒ ) Consider an open set G (arbitrary) and an arbitrary point x ∈ G. There exists,
for every point x in G a corresponding base Vαx so that x ∈ Vαx ⊂ G. This means, if we
collect all the x in G and its corresponding bases, the union of that basis would give us G.

⇐= ) Suppose for any given G ⊂ X that is open, there is a subcollection of a (fixed)

{Vα } so that G = ∪α′ Vα′ . Consider an arbitrary point x ∈ X and an arbitrary G : x ∈ G.
This means there is an epsilon ball around x that is fully contained in G. But since open
balls are open, there is a subcollection {Vα′′ } so that "this ball"= ∪Vα′′ . Since x ∈ this
ball, x ∈ Vk for some k. But this vk is in G. Hence, we are done. ■

Fact 5.16

”(∃{Vα : α ∈ A})(∀x ∈ X)(∀G ⊂ X : G open : x ∈ G)(∃α ∈ A)(x ∈ Vα ⊂ G)”

is the same thing as

”(∃{Vα : α ∈ A})(∀G ⊂ X : G open)(∀x ∈ X : x ∈ G)(∃α ∈ A)(x ∈ Vα ⊂ G)”

Definition 5.17
2nd countable

A space X is said to be second countable if it has a countable basis.

Theorem 5.18
Let X be a metric space in which every infinite subset E has a limit point in X. Then
X has countable basis (or equivalently is separable).

Proof for Theorem.

Metric Spaces 59

Note that no infinite subset of X can be "unbounded" since we have seen before that Limit
point compact implies closed and bounded. Consider an arbitrary δ0 . Choose an initial
point x1 . Is there a point x2 that is outside the δ−ball around x1 ? If no, the only δ−ball
of x1 covers the whole space. If yes, then again ask the question, is there a point x3 that
is both outside the δ−ball of x1 and the δ−ball of x2 ? If not, then these two balls will
cover the whole space, if yes, again ask the question, is there a point x3 so that it is not
in x1 , x2 and x3 ’s δ−balls. Keep going, as such. We claim that it must terminate after a
finite number of steps. Suppose not, i.e for every xn you can find, you can find an xn+1 to
form an infinite set of xj such that xj is not in the δ-ball of any of its preceeding elements
xj−1 , xj−2 · · · . This is true for every xj . From the hypothesis, this set has a limit point
q. Choose ε < δ/4, within which exists a point xk in this infinite set. But since no point
exists within the δ−ball of xk apart from itself, no other point exists in the ε−ball around
xk as well. This would contradict the proposition that q is a limit point. Hence, this
procedure must terminate in a finite amount of steps. So for a given δ = d, there exists
nδ balls V1 , V2 · · · Vnd with centres xd1 , xd2 · · · xdnd . Choose δ = 1, and create a collection of
1-balls (by the procedure mentioned above) around the points x11 , x12 · · · x1n1 . Likewise, for
δ = 1/2, find points x12 , x22 · · · x2n2 , and for δ = 1/k, find points xk2 , xk2 · · · xknk . Let z be a
point in X. Choose a ε ball around z. Let 1/n < ε. There must be some xnj so that z is
in the 1/n-ball of xnj . Hence, {xkj : j ∈ N, k ∈ N} is the dense set we need. ■

Theorem 5.19
Separable ⇐⇒ countable basis ⇐⇒ every open cover of X has a countable subcover

Proof for Theorem.

1 ⇐⇒ 2) already seen

2 =⇒ 3) Suppose we have an open cover {Gα : α ∈ A} of X which has a count-

able basis {Vn }. Let x ∈ X be arbitrary, which means it is in some Gα , which is the
union of some subfamily of {Vn } which means x is in some Vj which is inside Gα . Let
P := {n ∈ N : ∃αn such that Vn ⊆ Gαn }. Since every x is in some Vj we have that
X ⊆ ∪∞ i=1 Vni ⊆ ∪i∈P Gαi which ultimately means that Gαi forms a countable subcover of
{Gα : α ∈ A}.

3 =⇒ 1) Suppose every open cover has a countable subcover. Consider ε = 1 and

all 1−balls in X. This is an open cover that has a countable subcover of 1-balls centred
at {x11 , x12 , · · · }. Likewise, consider 1/2−balls in X which would have a countable sub-
collection of 1/2−balls centred at {x21 , x22 , x23 · · · }. As such keep going for ε = 1/n to
get a countable collection of 1/nballs. Take all these balls (which gives us a countable
collection, from axiom of choice) and call this set S. Let z ∈ X. let ε > 0 and 1/n < ε/4.
Consider the ε− ball around z. For δ = 1/n we have a countable collection of points
{xn1 , xn2 · · · } whose 1/n balls cover the whole space, which means there is a point in S,
say x, whose 1/n ball contains z. But this ball is in turn in the ε−ball around z, which
makes x a point in ε ball of z, hence every point of X is a limit point of S (or a point of
S). ■
60 Metric Spaces

Theorem 5.20
A compact metric space has a countable base (or equivalently, is separable- has a
countable dense set in it).

Proof for Theorem.

Choose δ1 = 1 and all the δ1 -balls in X (centred at every point of X). This makes an open
cover of X that has a finite subcover. i.e, finite points x11 , x12 · · · x1n1 whose δ1 = 1 balls
cover the whole space X. Next, choose δ2 = 1/2 and every δ2 ball in X. This likewise,
would have a finite subcover i.e, δ2 -balls around x21 , x22 · · · x2n2 . As such, for every δk = 1/k,
we have points xk1 , xk2 , · · · xknk whose δk balls cover the whole space.

That this space is separable is obvious now. Choose xij : i, j ∈ N . Let z be a point
in X. Choose any arbitrary ε. Find an n so that n1 < ε. From the result we have, we can
always find a xkj in our set so that it is in the n1 ball of z. This means that in every ε ball
of z, there exists a point from our stipulated set. Hence, our set is dense in X. ■

Theorem 5.21
If X is a metric space so that every infinite set in X has a limit point in X, then every
open cover of X (open relative to X) has a finite subcover. Hence:

"If X is a limit point compact metric space, then X is a compact metric space
(compact relative to itself, but it also applies to all spaces it lives in, due to
preservation of compactness)"
-Krishna, to Arjuna, at Kurukshetra

Proof for Theorem.

Suppose that X is limit point compact. This means that it has a countable basis. Consider
an open cover {Gα : α ∈ A} so that X = ∪α∈A Gα . Let {V1 , V2 · · · } be the countable basis
for X. This means that for every open set G and every point x ∈ X so that x ∈ G, we
have an n ∈ N so that x ∈ Vn ⊂ G. Let P := {n ∈ N : ∃α ∈ A such that Vn ⊆ Gα }.
Consider any point x ∈ X. For this x, there exists some Gα so that x ∈ Gα which would
imply from the conutable bases property that ∃Vn so that x ∈ Vn ⊆ Gα . For every point,
there exists a Vn and a corresponding Gα so that x ∈ Vn ⊂ Gα . Therefore, X is countably
"subcovered" by Gαj (those Gα that correspond to the Vn , n ∈ P ).

Let this countable subcover be G1 , G2 · · · Gn · · · . Suppose to this, there does not exist
a finite subcover. i.e,
∃x1 ∈ GC 1

,
∃x2 ∈ (G1 ∪ G2 )C = C1c ∩ GC
2

.
..
.
Metric Spaces 61

∃xn ∈ (G1 ∪ G2 ∪ G3 ∪ · · · Gn )C = ∩ni=1 GC

i

, for every n ∈ N. Collect all such xn s to form an infinite set in X. By hypothesis, this has
a limit point, call it z. Note that, outside a given GC k , only finite points of our stipulated
set exists, which are possibly x1 , x2 · · · xn−1 . Suppose there exists a GC C
j so that z ̸∈ Gj .
This means that z is in Gj which is an open set. Therefore, there exists an ε ball that is
completely inside Gj , or rather completely outside GC j . This means that only finitely many
x1 , x2 · · · make into this ε-ball of z which is absurd since z is supposedly a limit point.
Hence, we cannot have a countable cover that has no finite subcover. Hence, Gj : j ∈ N
has a finite subcover, which means X is a compact metric space. ■

Remark.
We showed that every limit point compact metric space is compact, i.e, every open cover,
open relative to X, has a finite subcover. Suppose X is actually a subset of a bigger set
H and X inherits the metric from H. From conservation of compact relativeness, we see
that X must be compact relative to H as well. Hence, every limit point compact set in X
is compact in X.

Definition 5.22: Haussdrof Space

A space X in which two distinct points x and y have disjoing neighbourhoods (open
sets) Vx and Vy so that x ∈ Vx and y ∈ Vy

Theorem 5.23: Equivalent statements pertaining to connected-

ness
The following are equivalent:

1. only clopen sets in X are X and ∅

2. X is not the disjoint union of two non empty open sets

3. X is not the union of disjoint non empty closed sets

4. X is not the union of separated sets A ∩ B = B ∩ A = ∅

5. Every proper non-empty subset of X has a non empty boundary in X

6. X does not have a continuous onto function mapping X to {−1, 1} with discrete
topology

Proof for Theorem.

1 =⇒ 2) Say X is the disjoint union of two non empty open sets, X = A ∪ B where
A and B are open. X \ (A) = B which, by definition, is closed. But B by hypothesis is
open, hence B is a clopen set in X that is not X or ∅.
62 Metric Spaces

2 =⇒ 1) Suppose Z ⊂ X is a non empty clopen, proper subset of X. W = X \ (Z)

is also, by definition, clopen. So we have W and Z, disjoint, non empty open sets whose
union gives us X.

1 =⇒ 3) Suppose X is the disjoint union of two closed sets. X = A ∪ B. X \ (A) = B

which is open, but also by hypothesis, closed. Hence, there is a clopen set in X.

3 =⇒ 1) Suppose there exists clopen set Z ⊂ X. Then W = X \ (Z) is also clopen by

definition. Therefore, W and Z form disjoint closed sets whose union gives us X.

1 =⇒ 4) Say X is the union of disjoint separated sets A and B so that A∩B = B ∩A = ∅.

We then have B ⊆ (A)C and likewise A ⊆ (B)C . X = A ∪ B. (A) ⊆ X and B ⊆ X which
means that A ∪ B = X. A ∪ B \ (A) = X \ (A) = B tells us that B is a open set. Doing
the same to B ∪ A = X tells us that A, too, is an open set. Hence, X is the disjoint union
of two open sets.

4 =⇒ 1) Say there exists a clopen set Z ∈ X. Consider Z C which is also clopen,

and it is disjoint from Z. This means immediately that Z ∩ Z C = Z ∩ Z C = Z ∩ Z C = ∅,
hence, X is the union of two separated sets.

1 =⇒ 5) Suppose there exists a non empty proper subset of X, call it W that has
an empty boundary. i.e, W \ (W 0 ) = ∅ which means W ∩ (W 0 )C = ∅. W C ⊆ (W 0 )C which
means W ∪ (W o )C = X which makes X the disjoint union of two closed sets

5 =⇒ 1) Suppose there exists a clopen set W in X. that is neither X nor ∅. W C

is, then, also clopen. Boundary of W would be W \ (W o ) = W \ (W ) = ∅. Hence, W is a
proper set, non empty such that its boundary is empty.

1 =⇒ 6) Let f be a continuous onto function from X to the discrete set {−1, 1}

with the discrete topology. Consider α = f −1 (−1) and β = f −1 (1). Of course, α ∪ β = X.
Is α and β disjoint? they must be since {−1} and {1} are disjoint. Moreover, since f is
continuous, and −1 and 1 are open sets in the discrete topology, we have that α and β
are open sets. Hence, X is the disjoint union of two open sets.

6 =⇒ 1) Suppose X = A ∪ B where A and B are non empty, disjoint open sets.

Define the function f : X → {−1, 1} given by f (x) = −1 if x ∈ A and f (x) = 1 if x ∈ B.
f −1 (−1) would be A and f −1 (1) would be B, which means every open set inverse maps to
an open set, making f a continuous onto function from X to the discrete topology {−1, 1}
■