Chapter 2: Basic Topology in Metric Spaces

Wissam Raji

Edited by Selena Nadjarian

Contents
1 Open and closed sets 1
1.1 Neighborhoods, balls, and open sets . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Limit points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Interior, closure, and boundary 3

2.1 Interior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 Relative openness (subspaces) 5

4 Compactness 5
4.1 Open covers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.2 Basic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.3 Sequential compactness in metric spaces . . . . . . . . . . . . . . . . . . . . . . . 7

5 Compactness in Euclidean spaces 7

6 Perfect sets and the Cantor set 13

6.1 The Cantor set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7 Connected sets 14

1 Open and closed sets

1.1 Neighborhoods, balls, and open sets
Definition 1.1 (Metric space). A metric space is a pair (X, d) where X is a nonempty set and
d : X × X → [0, ∞) satisfies, for all x, y, z ∈ X,
1. d(x, y) = 0 iff x = y;
2. d(x, y) = d(y, x);
3. d(x, z) ≤ d(x, y) + d(y, z).

Definition 1.2 (Open ball and neighborhood). For x ∈ X and r > 0, the open ball of radius r
centered at x is
Br (x) = {y ∈ X : d(x, y) < r}.
A neighborhood of x is any set N ⊆ X that contains an open ball around x: ∃r > 0 with
Br (x) ⊆ N .

1
Definition 1.3 (Open and closed sets). A set G ⊆ X is open if for every x ∈ G there exists
r > 0 such that Br (x) ⊆ G. A set F ⊆ X is closed if every limit point of the set is a point of
the set.

Remark 1.4. If a set is open, its complement is closed and vice versa.

Remark 1.5. It is important to note that the geometric shape of an open ballp depends entirely
2
on the chosen metric. In R , under the standard Euclidean metric d2 (x, y) = (x1 − y1 )2 + (x2 − y2 )2 ,
the open ball Br (x) is an open disk. However, if we equip R2 with the taxicab metric d1 (x, y) =
|x1 − y1 | + |x2 − y2 |, the open ball Br (x) takes the shape of a diamond.

Lemma 1.6 (Balls are open). For every x ∈ X and r > 0, the ball Br (x) is open.

x z y x+r

Figure 1: Visualization of the lemma that open balls are open.

Proof. If y ∈ Br (x), set δ = r − d(x, y) > 0. If z ∈ Bδ (y) then by the triangle inequality

d(x, z) ≤ d(x, y) + d(y, z) < d(x, y) + δ = r,

so z ∈ Br (x). Hence Bδ (y) ⊆ Br (x).

Example 1.7. In (R, |·|), Since open balls are open intervals (a, b) is open. Since the comple-
ment of [a, b] is (−∞, a) ∪ (b, +∞) is open then [a, b] closed.

Remark 1.8. Whether a set is open depends on the metric being used. Thus openness is
not purely a geometric notion coming from the picture alone; it depends on how distance is
measured. This is why the same subset of a set X may behave differently under different
metrics.

Theorem 1.9 (Algebra of open and closed sets). 1. Arbitrary unions of open sets are open.
2. Finite intersections of open sets are open.
3. Arbitrary intersections of closed sets are closed.
4. Finite unions of closed sets are closed.

Example 1.10. The intersection of infinitely many open sets need not be open. For example,
in R,
∞  
\ 1 1
− , = {0},
n n
n=1

and {0} is not open.

Likewise, the union of infinitely many closed sets need not be closed. For example,
∞  
[ 1
, 1 = (0, 1],
n
n=1

which is not closed in R.

2
1.2 Limit points
Definition 1.11 (Limit point). Let E ⊆ X. A point p ∈ X is a limit point (accumulation
point) of E if every neighborhood of p contains a point of E different from p. Equivalently,

∀r > 0 : (Br (p) \ {p}) ∩ E ̸= ∅.

Definition 1.12 (Isolated point). A point p ∈ E is isolated in E if p is not a limit point of E,

i.e. if some ball around p meets E only at p.

Example 1.13. Consider the set E = n1 : n ∈ N ⊂ R with the standard metric. The point


p = 0 is a limit point of E because every open interval (−ϵ, ϵ) contains points of E for sufficiently
large n. Notice that p = 0 is a limit point but 0 ∈ / E. Meanwhile, every point n1 ∈ E is an
isolated point. For any given n, we can choose a radius r small enough (for instance, r = n1 − n+11
)
1
such that Br ( n ) contains no other points of E.

Theorem 1.14 (Closed sets and limit points). A set F ⊆ X is closed if and only if it contains
all of its limit points.

Proof. If F is closed then X \ F is open. If p is a limit point of F and p ∈ / F , then p ∈ X \ F

and there is r > 0 with Br (p) ⊆ X \ F , contradicting the definition of limit point. Hence p ∈ F .
Conversely, assume F contains all its limit points. To prove F is closed it suffices to show
X \ F is open. If x ∈ X \ F and if every ball around x intersects F , then x would be a limit
point of F , hence x ∈ F , contradiction. Thus some Br (x) ⊆ X \ F .

2 Interior, closure, and boundary

2.1 Interior
Definition 2.1 (Interior). For E ⊆ X, the interior of E is

E ◦ = {x ∈ E : ∃r > 0 such that Br (x) ⊆ E}.

Remark 2.2. E is open iff E = E ◦ .

2.2 Closure
Definition 2.3 (Closure). For E ⊆ X, the closure of E is

E = {x ∈ X : ∀r > 0, Br (x) ∩ E ̸= ∅}.

Lemma 2.4 (Characterization by limit points). If E ′ ⊆ X denotes the set of limit points of E,
then
E = E ∪ E′.

Proof. If x ∈ E, then every Br (x) meets E (at least at x), hence x ∈ E. If x is a limit point,
then every Br (x) meets E in a point different from x, hence x ∈ E. Thus E ∪ E ′ ⊆ E.
Conversely, if x ∈ E and x ∈ / E, then every Br (x) meets E in a point different from x (since
x∈/ E), so x is a limit point. Hence E ⊆ E ∪ E ′ .

Theorem 2.5 (Basic properties of closure). For all E ⊆ X:

1. E ⊆ E.
2. E is closed.
3. E is closed iff E = E.

3
 4. If E ⊆ F then E ⊆ F .
5. E ∪ F = E ∪ F .
6. E ⊂ F ∀ closed sets F ⊂ X s.t. E ⊂ F

Proof. (1) is immediate from Definition 2.3. For (2), let x be a limit point of E. Then every
ball around x meets E, hence (by definition of E) meets E. Thus x ∈ E, so E contains its limit
points and is closed by Theorem 1.14. (3) follows from (2) and minimality (see (4)). (4) is clear
from the defining condition in Definition 2.3. (5) follows directly from (4) and the fact that a
ball meets E ∪ F iff it meets E or it meets F . (6) Let E ⊂ F , then E ′ ⊂ F ′ since F is closed.
So, E ′ ⊂ F ⊂ F . Therefore, E ⊂ F

Remark 2.6. For a set E ⊆ X, the interior E ◦ is the largest open set contained in E, while
the closure E is the smallest closed set containing E.

Theorem 2.7 (Sequential characterization (metric spaces)). Let (X, d) be a metric space and
E ⊆ X. Then x ∈ E if and only if there exists a sequence (xn ) ⊆ E such that xn → x.

Proof. If x ∈ E, then for each n the ball B1/n (x) meets E. Choose xn ∈ E ∩ B1/n (x). Then
d(xn , x) < 1/n, hence xn → x. Conversely, if xn ∈ E and xn → x, then every ball Br (x)
contains xn for all sufficiently large n, so it meets E, hence x ∈ E.

2.3 Boundary
Definition 2.8 (Boundary). For E ⊆ X, the boundary of E is

∂E = E \ E ◦ .

Lemma 2.9. A point x ∈ X lies in ∂E if and only if every ball Br (x) meets both E and X \ E.

/ E ◦ means no ball is contained in E,

Proof. x ∈ E means every ball meets E. And x ∈
equivalently every ball meets X \ E.

Remark 2.10. For every set E ⊆ X,

∂E = E \ E ◦ .

In particular, ∂E is always closed, since it is the difference of a closed set and an open set
relative to that closed set.

Example 2.11. To illustrate the concept of a boundary, consider the following subsets of R
with the standard metric:
1. Let E = (0, 1). The closure is E = [0, 1] and the interior is E ◦ = (0, 1). Therefore,
∂E = E \ E ◦ = {0, 1}. Note that the boundary points do not belong to E.
2. Let E = [0, 1]. Here, E = [0, 1] and E ◦ = (0, 1), so ∂E = {0, 1}. Here the boundary
points are contained in E.
3. Let E = Q, the set of rational numbers. Every open ball (interval) in R contains both
rational and irrational numbers because both are dense in R. Therefore, E ◦ = ∅ and
E = R. The boundary is ∂Q = R \ ∅ = R.

Example 2.12. In a discrete metric space, every subset is both open and closed. Hence for
every subset E,
E ◦ = E, E = E, ∂E = ∅.
This shows that boundary behavior depends strongly on the ambient metric space.

4
3 Relative openness (subspaces)
Definition 3.1 (Relative topology). Let (X, d) be a metric space and Y ⊆ X. A set E ⊆ Y
is open relative to Y if there exists an open set G ⊆ X such that E = G ∩ Y . A set F ⊆ Y is
closed relative to Y if F = K ∩ Y for some closed K ⊆ X.
Example 3.2. Let X = R with the standard metric, and consider the subspace Y = [0, 1]. The
set E = [0, 1/2) is open relative to Y . We can verify this using Definition 3.1 by choosing the
open set G = (−1/2, 1/2) in X. Then E = G ∩ Y . Note that E = [0, 1/2) is not open in X = R,
demonstrating that openness is a relative property depending on the ambient metric space.
Example 3.3. Let X = R and let Y = [0, 1]. Then the set (0, 1] is also open relative to Y ,
since
(0, 1] = (0, 2) ∩ [0, 1].
However, (0, 1] is not open in R. Thus a set may be open in a subspace without being open in
the ambient space.
Theorem 3.4. E ⊆ Y is open relative to Y if and only if for every x ∈ E there exists r > 0
such that
Br (x) ∩ Y ⊆ E.
Proof. If E = G ∩ Y with G open in X, then for x ∈ E ⊆ G there exists r > 0 with Br (x) ⊆ G,
hence Br (x) ∩ Y ⊆ G ∩ Y = E. Conversely, assume the S displayed condition holds. For each
x ∈ E choose rx > 0 with Brx (x) ∩ Y ⊆ E, and let G = x∈E Brx (x), which is open in X. Then
E =G∩Y.

4 Compactness
4.1 Open covers

S 4.1 (Open cover). Let K ⊆ X. A family of open sets {Gα }α∈A is an open
Definition S cover of
K if K ⊆ α∈A Gα . A finite subcover is a finite subfamily Gα1 , . . . , Gαn with K ⊆ nj=1 Gαj .
Definition 4.2 (Compact set). A set K ⊆ X is compact if every open cover of K has a finite
subcover.
Example 4.3. To understand compactness, it is helpful to look at a set that is not compact.
Consider the set K = (0, 1] ⊂ R. Let us define a family of open sets:
 
1
Gn = ,2 for n = 1, 2, 3, . . .
n
The collection {Gn }n∈N is an open cover for K because every x ∈ (0, 1] is greater than some
1/n, meaning x ∈ Gn . However, no finite subcollection can cover K. If we take any finite
number of these sets, there will be a maximum index N . The union of this finite subcollection
will simply be GN = (1/N, 2). This finite union fails to cover the points in K that are close to
0 (specifically, any 0 < x ≤ 1/N ). Therefore, K = (0, 1] is not compact.

4.2 Basic properties

Theorem 4.4. If F ⊂ K which is compact and F is closed, then F is compact
Proof. Let {Gα } be an open cover of F, assuming F is closed then Gα ∪ F c is an open cover
S
of K by theorem S 2.10. Since K is compact, there exists finitely man sets such and possible F c
n
such that K ⊆ Gα ∪ F c . Now if we intersect both side with F, we get that F α=1 Gα so F
S
has a finite subcover and hence it is compact.

5
Theorem 4.5 (Closed subsets of compact sets are compact). If K is compact and F is closed
in X, then K ∩ F is compact.
Proof. Let {Gα } cover K ∩ F with each Gα open in X. Then {Gα } ∪ {X \ F } is an open cover
of K since every x ∈ K either lies in F meaning it is in K ∩ F so in some Gα or it lies in the
complement of F . By compactness of K there is a finite subcover. Removing X \ F leaves a
finite cover of K ∩ F .

Theorem 4.6 (Compact sets are closed and bounded (metric spaces)). If K is compact in a
metric space (X, d), then K is closed and bounded.
Proof. We first show that K is closed by showing that its complement is open.
Let x ∈ X \ K. For each y ∈ K, since x ̸= y, we have
d(x, y) > 0.
Set
d(x, y)
ry = > 0.
2
Then the open ball (B(y, ry )) = {z ∈ X : d(z, y) < ry } is an open neighborhood of y. As y
varies over K, the family
{B(y, ry ) : y ∈ K}
is an open cover of K.
Since K is compact, there exist finitely many points y1 , . . . , yn ∈ K such that
n
[
K⊆ B(yi , ryi ).
i=1

Now let
r = min{ry1 , . . . , ryn } > 0.
We claim that
B(x, r) ⊆ X \ K.
Suppose, for contradiction, z ∈ B(x, r). If z ∈ K, then z ∈ B(yi , ryi ) for some i. Hence
d(yi , z) < ryi .
Also, since z ∈ B(x, r) and r ≤ ryi ,
d(x, z) < r ≤ ryi .
Therefore, by the triangle inequality,
d(x, yi ) ≤ d(x, z) + d(z, yi ) < ryi + ryi = 2ryi = d(x, yi ),
which is impossible.
Thus z ∈/ K, and so
B(x, r) ⊆ X \ K.
This shows that X \ K is open. Therefore K is closed.
To show that K is bounded, fix a point x0 ∈ X. Then the family of open balls
{B(x0 , n) : n ∈ N}
is an open cover of K. By compactness, there exists N ∈ N such that
K ⊆ B(x0 , N ).
Hence K is bounded.

6
Theorem 4.7. If E is an infinite subset of a compact set K, then E has limit points in K
Proof. Assume E has no limit points ∈ K. Then, E ∪ Nr (q) = if q ∈ / E and E ∪ Nr (q) = q if
q ∈ E. So, no neighborhood can intersect E in more than one point. If E has no limit points,
points of E are “isolated” from each other. No cluster of points exists.
Since K is compact the, every open cover has a finite subcover. We can choose Sn a finite
number of neighborhoods to over K, and since E is in K then we can say E ⊂ K ⊂ i=1 Nr (qi )
.
But E is infinite so there exists at least one i such that E ∩ Nri (qi ) have infinite points so
qi is a limit point and hence we have a contradiction.

Theorem 4.8 (Finite unions). A finite union of compact subsets of X is compact.

Sm
Proof.
Sm Let K 1 , . . . , K m be compact with {Gα } as an open cover of i=1 Ki Since each Ki ⊆
i=1 Ki , then {Gα } covers each Ki . Since each Ki is compact then ∃ a finite subcover of Ki
from {Gα } . Taking the finite subcovers ∀ i = 1, ..., m and combining them gives use a finite
collection of open sets that cover m
S
i=1 i Thus, the union is compact.
K .

Theorem 4.9 (Nested compact T sets). If K1 ⊇ K2 ⊇ K3 ⊇ · · · is a decreasing sequence of

nonempty compact sets, then ∞ n=1 Kn ̸= ∅.

Proof. Assume for contradiction that ∞

T
n=1 Kn = ∅ . Then, we can cosntruct an open cover of
K1 by using the complements. Since the intersection is empty, ∀ x ∈ K1 ∃N2 s.t. x ∈ / KNx
So, x ∈ X\KNx . Hence {X\Kn } is an open cover of K1 .
Since K1 is compect then this open cover has a finite subcover. Let N = max(n1 , ..., nm ).
Since Kn are nested sets, the nesting of their complements reverses to X\Kn1 ∪ ... ∪ X\Knm ⊆
X\Kn . Since Kn is the smallest set, its complement is big enough to include all the other
complements.
So, the finite union of the open sets is just one set K1 ⊆ X\KN . Howver, this says that all
the points of K1 are outside KN while KN ⊆ K1 which leads to a contradiction,

4.3 Sequential compactness in metric spaces

Theorem 4.10 (Compact ⇒ sequentially compact). If K is compact in a metric space, then
every sequence in K has a convergent subsequence with limit in K.
Proof. Let (xn ) ⊆ K. If the set {xn } is finite, then some value occurs infinitely many times and
yields a constant convergent subsequence. Otherwise {xn } is infinite, hence has a limit point
p in K (since K is compact and thus closed; a standard argument shows every infinite subset
of a compact set has a limit point in K). Choose n1 with d(xn1 , p) < 1, then n2 > n1 with
d(xn2 , p) < 21 , etc. This produces a subsequence xnj → p ∈ K.

Remark 4.11. In Rk , compactness is equivalent to being closed and bounded (Theorem 5.7
below). However, this is a special property of Euclidean spaces and is not true in every metric
space. Thus, in a general metric space, one should always use the definition of compactness in
terms of open covers.

5 Compactness in Euclidean spaces

Theorem 5.1. If we have
I1 ⊃ I2 ⊃ I3 ⊃ · · ·
where each In is an interval in R (i.e., a closed interval), then
∞
\
In ̸= ∅.
n=1

7
Proof. Let
In = [an , bn ]
and define
x = sup{an : n = 1, 2, . . . }.
Let
E = {an : n = 1, 2, . . . }.
We will show that x ∈ Im for every m = 1, 2, . . ..
We have, for any n, m,
an ≤ am+n ≤ bm+n ≤ bm . (1)
If we fix m and let n = 1, 2, . . ., then (1) implies that bm is an upper bound of E. Since
x = sup E is the smallest upper bound of E, we obtain
x ≤ bm . (2)
The previous argument holds for every m = 1, 2, . . ., and hence (2) holds for every m =
1, 2, . . ..
Also, since x = sup E, we have
am ≤ x ≤ b m , for all m = 1, 2, . . .
that is,
x ∈ [am , bm ] = Im , for all m = 1, 2, . . .
Therefore,
∞
\
x∈ Im .
m=1

Theorem 5.2. If
I1 ⊃ I2 ⊃ I3 ⊃ · · ·
where In ⊂ Rk and
In = [an,1 , bn,1 ] × [an,2 , bn,2 ] × · · · × [an,k , bn,k ],
then
∞
\
In ̸= ∅.
n=1

Proof. For every j = 1, . . . , k,

[a1,j , b1,j ] ⊃ [a2,j , b2,j ] ⊃ [a3,j , b3,j ] ⊃ · · ·
Theorem 5.1 implies that there exists x∗j , 1 ≤ j ≤ k, such that
an,j ≤ x∗j ≤ bn,j , 1 ≤ j ≤ k.
If we define
x∗ := (x∗1 , . . . , x∗k ),
then
x∗ ∈ In , n = 1, 2, . . .
Hence,
∞
\
x∗ ∈ In .
n=1

8
Definition 5.3. A k-cell I ⊂ Rk is defined by
n o
I = ⃗x = (x1 , . . . , xk ) ∈ Rk : ai ≤ xi ≤ bi , 1 ≤ i ≤ k

that is,
I = [a1 , b1 ] × [a2 , b2 ] × · · · × [ak , bk ].

Example 5.4.
[a, b]
is a 1-cell.
A rectangle in R2 is a 2-cell. [a1 , b1 ] × [a2 , b2 ]
A rectangular box in R3 is a 3-cell. [a1 , b1 ] × [a2 , b2 ] × [a3 , b3 ]

Theorem 5.5. Every k-cell I is compact.

Proof. Let
I = [a1 , b1 ] × · · · × [ak , bk ]
and define v
u k
uX
δ := t (bi − ai )2 .
i=1

(Note: δ is the length of the diagonal of the square or cube in R2 or R3 , respectively.)

We have
∥⃗x − ⃗y ∥ ≤ δ, ∀x, y ∈ I.
We proceed by contradiction. Suppose that there exists an open cover {Gα } of I which
contains no finite subcover of I. Let
ai + bi
ci := .
2
We can split I into 2k k-cells:
2 k
[
I= Ii .
i=1

This partition is given by the intervals

[ai , ci ] and [ci , bi ], i = 1, . . . , k.

At least one of these Ii , say I1 , cannot be covered by any finite subcollection of {Gα } (for
otherwise I could be covered by a finite subcollection). We subdivide I1 in the same way to
obtain a sequence such that:
1. I ⊃ I1 ⊃ I2 ⊃ I3 ⊃ · · ·,
2. In is not covered by any finite subcollection of {Gα },
3. if x, y ∈ In , then each time we split the cell we cute every coordinate interval into half
leading to:
δ
∥x − y∥ ≤ n .
2
By Theorem 5.2,
∞
\
∃x∗ ∈ In .
n=1

9
Since [
x∗ ∈ I ⊂ Gα ,
α

there exists α such that

x∗ ∈ Gα .
Since Gα is open, there exists Br (x∗ ) ⊂ Gα for some r > 0.
For N large enough,
δ
< r,
2N
which implies
IN ⊂ Br (x∗ ) ⊂ Gα ,
contradicting (2).This means one open set covers the whole cube IN but we had earlier con-
structed IN in such a way that it cannot be covered by any finite subcollection of the cover.
So, we reach a contradicition and we can conclude that I can be covered by a finite collection
in {Gα }. Hence I is compact.

Bounded Sets
Recall the definition of a bounded set in a metric space X.

Definition 5.6. Let E ⊂ X, where X is a metric space. We say that E is bounded if there
exist M > 0 and q ∈ X such that

d(p, q) < M, ∀p ∈ E.

Example 5.7. If X = Rk , then E ⊂ Rk is bounded if there exist M > 0 and x̄ ∈ Rk such that

∥ȳ − x̄∥ < M, ∀ȳ ∈ E.

Remark 5.8. This means that E is contained in the open ball of radius M , centered at x̄:

BM (x̄) = {z̄ ∈ Rk : ∥z̄ − x̄∥ < M }.

Theorem 5.9. Let E ⊂ Rk . Then the following are equivalent:

(a) E is closed and bounded.
(b) E is compact.
(c) Every infinite subset of E has a limit point in E.

The equivalence of (a) and (b) is called the Heine–Borel Theorem.

Proof. (a) ⇒ (b).

If E is bounded, then E is contained in some k-cell I:

I = [a1 , b1 ] × · · · × [ak , bk ].

Since I is compact and E ⊂ I, and since E is closed, theorem 4.4 implies that E is compact.
(b) ⇒ (c).
This implication is proven in Theorem 4.7.
(c) ⇒ (a).
We assume that every infinite subset of E has a limit point in E.
We proceed by contradiction.

10
 First, we show that E is bounded. Suppose E is not bounded. Then E is not contained in
B1 (0̄), hence there exists x̄1 ∈ E such that

∥x̄1 ∥ > 1,

i.e. x̄1 ∈
/ B1 (0̄).
Since E is not contained in B2 (0̄), there exists x̄2 ∈ E such that

∥x̄2 ∥ > 2,

i.e. x̄2 ∈
/ B2 (0̄).
We can choose x̄2 ̸= x̄1 ; otherwise E would be bounded. Proceeding in this way, since E is
not contained in Bn (0̄), there exists x̄n ∈ E, with

x̄n ̸= x̄i , i = 1, . . . , n − 1,

such that
∥x̄n ∥ > n,
i.e. x̄n ∈
/ Bn (0̄).
The set
S = {x̄n : n = 1, 2, . . . }
is infinite, but clearly has no limit point in Rk (and hence no limit point in E). Indeed, if ȳ were
a limit point of S, then given any N , BN (ȳ) would have to contain infinitely many elements of
S, but this is not possible since
BN (ȳ) ⊂ BN +∥ȳ∥ (0̄),
and for all n > N + ∥ȳ∥,
x̄n ∈
/ BN +∥ȳ∥ (0̄).
Since this contradicts assumption (c), we conclude that E is bounded.
Next, we show that E is closed. Suppose E is not closed. Then there exists x̄0 ∈ E ′ such
that
x̄0 ∈
/ E.
By definition of limit point, there exists x̄1 ∈ B1 (x̄0 ) such that

x̄1 ∈ E, x̄1 ̸= x̄0 ,

hence
∥x̄1 − x̄0 ∥ < 1.
Again, since x̄0 is a limit point of E, there exists x̄2 ∈ B1/2 (x̄0 ) such that

x̄2 ∈ E, x̄2 ̸= x̄0 , x̄2 ̸= x̄1 ,

and
1
∥x̄2 − x̄0 ∥ < .
2
Proceeding in this way, for any n, since x̄0 is a limit point of E, there exists x̄n ∈ B1/n (x̄0 )
such that
x̄n ∈ E,
and we can choose x̄n so that

x̄n ̸= x̄i , i = 1, . . . , n − 1,

11
for otherwise x̄0 would not be a limit point (recall that any neighborhood of x̄0 must contain
infinitely many elements of E). Thus
1
∥x̄n − x̄0 ∥ < , n = 1, 2, . . .
n
The set
S = {x̄n : n = 1, 2, . . . }
is an infinite set. We show next that x̄0 is the only limit point of S in Rk . For if ȳ ∈ Rk , ȳ ̸= x̄0 ,
then
∥ȳ − x̄0 ∥ ≤ ∥ȳ − x̄n ∥ + ∥x̄n − x̄0 ∥,
so
1
∥ȳ − x̄n ∥ ≥ ∥ȳ − x̄0 ∥ − ∥x̄n − x̄0 ∥ ≥ ∥x̄0 − ȳ∥ − .
n
Hence, for n ≥ N , with N large enough,
1
∥ȳ − x̄n ∥ ≥ ∥x̄0 − ȳ∥.
2
This says that ȳ cannot be a limit point of S.
Therefore,
S ′ ∩ E = ∅,
which contradicts hypothesis (c), since S is an infinite subset of E and must have a limit point
in E.
From this contradiction, we conclude that E is closed.

N + ∥y∥

0 ∥y∥ y

Figure 2: The ball BN (y) is contained in BN +∥y∥ (0).

Example 5.10. The interval [0, 1] is compact in R because it is closed and bounded. By
contrast, the interval (0, 1) is bounded but not closed, so it is not compact. Likewise, the ray
[0, ∞) is closed but not bounded, so it is not compact.
Remark 5.11. Note that the previous theorem is proved for X = Rk . For any metric space
X, (b) ⇔ (c) is still true, but in general,

(a) ⇏ (b) and (a) ⇏ (c).

Example 5.12. The converse of Theorem 5.7 is false in general metric spaces. For example,
let X = N with the discrete metric
(
0, m = n,
d(m, n) =
1, m ̸= n.

12
Then X is bounded, since d(m, n) ≤ 1 for all m, n ∈ X, and every subset of X is closed.
However, X is not compact, because the open cover

{n} : n ∈ N

has no finite subcover.

Theorem 5.13 (Bolzano–Weierstrass). Every bounded infinite subset of Rk has a limit point
in Rk . Equivalently, every bounded sequence in Rk has a convergent subsequence.
Proof. Let E ⊆ Rk be bounded and infinite. Then E is closed and bounded, hence compact
by Heine–Borel (Theorem 5.7). Any infinite subset of a compact set has a limit point in that
compact set. Thus E has a limit point in E ⊆ Rk .
For sequences, apply the set version to E = {xn : n ∈ N} if this set is infinite; if it is finite,
some value repeats infinitely often and yields a constant subsequence.

6 Perfect sets and the Cantor set

Definition 6.1 (Perfect set). A set P ⊆ X is perfect if it is closed and every point of P is a
limit point of P .
Example 6.2. In R, the sets R and any closed interval [a, b] are perfect. A finite set is not
perfect.

6.1 The Cantor set

Definition 6.3 (Cantor set). Let C0 = [0, 1]. Remove the open middle third ( 13 , 23 ) to obtain
C1 = [0, 13 ] ∪ [ 23 , 1]. Inductively, obtain Cn+1 from Cn by removing the open middle third from
each interval component of Cn . The Cantor set is
∞
\
C= Cn .
n=0

C0 = [0, 1]
1 2
C1 = [0, ] ∪ [ , 1]
3 3
1 2 3 6 7 8
C2 = [0, ] ∪ [ , ] ∪ [ , ] ∪ [ , 1]
9 9 9 9 9 9
Illustration 6.3b. Visually, the construction begins with a solid line segment C0 . In the
first step, the middle third is removed, leaving two shorter closed segments (C1 ). In the second
step, the middle thirds of those two segments are removed, leaving four even shorter segments
(C2 ). As n → ∞, the set fractures into an infinite number of points that contain no contiguous
intervals, yet still form a non-empty, uncountably infinite set.
Remark 6.4. The Cantor set has several deeply counterintuitive properties. Despite being
uncountably infinite (as proven in Theorem 6.6), it contains no intervals. If we take the limit
of the sum the lengths of all the open intervals removed during its construction, we get:
∞
2n
   
1 1 1 X 1/3
+2 +4 + ··· = n+1
= = 1.
3 9 27 3 1 − 2/3
n=0

Since the original interval [0, 1] had length 1, and we removed intervals of total length 1, the
Cantor set has a total ”length” (or Lebesgue measure) of 0. It is a mathematical dust: un-
countable, yet taking up no space.

13
Remark 6.5 (Ternary description). Every number x ∈ [0, 1] can be written as x = 0.d1 d2 d3 ...(base3)
where each di ∈ {0, 1, 2}. Let’s start with [0, 1] and remove the middle third ( 31 , 23 ). Any number
in this interval has a ternary expansion starting with 0.1 since 0.03 = 0, 0.13 = 13 , and 0.23 = 32 .
So the first ternary digit is d1 = 1 for every number in the middle third. The remianing intervals
have numbers starting with 0.0 in [0, 13 ] and 0.2 in [ 23 , 1]. Once we remove the midle third of
those two intervals, we will remove numbers with 1 in the second ternary digit d2 and so on
with leads to numbers with only 0s and 2s in every ternary digit.

Theorem 6.6 (Cantor set is compact and perfect). The Cantor set C is compact and perfect.

Proof. Each Cn is a finite union of closed intervals, hence closed, and C ⊆ [0, 1] is bounded.
Since C is an intersection of closed sets, it is closed, hence compact by Heine–Borel.
To see C is perfect, let x ∈ C and r > 0. Choose n large enough so that the length of each
interval in Cn is 3−n < r. Since x ∈ C ⊆ Cn , x lies in some closed interval I ⊆ Cn of length
< r. The construction of C ensures that I ∩ C contains more than one point (indeed it contains
a scaled copy of C), so we can pick y ∈ (I ∩ C) \ {x}. Then y ∈ Br (x) ∩ C, proving x is a limit
point of C.

Theorem 6.7 (Cantor set is uncountable). The Cantor set C is uncountable.

Proof. Using the ternary description, associate to each sequence (an )n≥1 ∈ {0, 2}N the real
number
∞
X an
x= ∈ [0, 1].
3n
n=1

This x has a ternary expansion with digits only 0 or 2, hence x ∈ C. Different sequences
give different numbers except for the usual ambiguity of terminating expansions, which can be
resolved by choosing the non-terminating representation (e.g. avoiding tails of all 2). Thus there
is an injection {0, 2}N ,→ C. Since {0, 2}N is uncountable (diagonal argument), so is C.

7 Connected sets
Definition 7.1 (Separated sets). Non-empty sets A, B ⊆ X are separated if

A∩B =∅ and A ∩ B = ∅.

Definition 7.2 (Connectedness). A set E ⊆ X is connected if there do not exist nonempty

separated sets A, B ⊆ E such that E = A ∪ B. Otherwise E is disconnected.

Example 7.3. Let E = [0, 1) ∪ (1, 2] ⊂ R. Let A = [0, 1) and B = (1, 2]. We can verify that
A and B are separated: the closure of A in R is [0, 1], which does not intersect B, and the
closure of B is [1, 2], which does not intersect A. Because E can be written as the union of two
non-empty separated sets, E is disconnected. Note that A and B are not open in R, but they
are open relative to the subspace E.

Lemma 7.4. If E is connected and E ⊆ F ⊆ E, then F is connected. In particular, E is

connected.

Proof. If F = A ∪ B with A, B separated in F , then E = (A ∩ E) ∪ (B ∩ E) is a separation of

E unless one of the intersections is empty. Thus E ⊆ A or E ⊆ B. But then F ⊆ E ⊆ A or B,
contradicting separation unless the other set is empty.

Theorem
T S connected sets). If {Eλ }λ∈Λ is a family of connected subsets of X
7.5 (Unions of
with λ∈Λ Eλ ̸= ∅, then λ∈Λ Eλ is connected.

14
Proof. Let p lie in the common intersection. If the union were disconnected, it would be A ∪ B
with A, B separated and nonempty. Choose λ with p ∈ Eλ . Then Eλ = (Eλ ∩ A) ∪ (Eλ ∩ B) is
a separation of Eλ unless one of these intersections is empty. Since p ∈ Eλ lies in either A or
B, we get Eλ ⊆ A or Eλ ⊆ B. Running this over all λ forces the whole union to lie in A or in
B, contradicting that both are nonempty.

Theorem 7.6 (Connected subsets of R are intervals). A set E ⊆ R is connected (with the usual
metric) if and only if it is an interval (possibly a single point). Equivalently, E is connected iff
for all a < b in E we have [a, b] ⊆ E.

Proof. Assume E is connected and pick a < b in E. If there existed c ∈ (a, b) with c ∈
/ E, then

A = E ∩ (−∞, c), B = E ∩ (c, ∞)

are nonempty, disjoint, and separated in E, and E = A∪B, contradicting connectedness. Hence
[a, b] ⊆ E and E is an interval.
Conversely, assume E is an interval. If E = A∪B with A, B separated and nonempty, choose
a ∈ A and b ∈ B with a < b. Since E is an interval, [a, b] ⊆ E. Let c = sup(A ∩ [a, b]). Then
c ∈ [a, b] ⊆ E. One checks that every neighborhood of c meets both A and B, contradicting
separation. Thus E is connected.

A B
a c = sup(A ∩ [a, b]) b R

Figure 3: Visualization of intervals A, B, points a, b, and c = sup(A ∩ [a, b]) along R.

Example 7.7. The set (0, 1) is connected because it is an interval. The set (0, 1) ∪ (2, 3) is
disconnected because it is not an interval. The set Q ⊆ R is also disconnected, since between
any two rational numbers there is an irrational number, so Q contains no nontrivial interval.

15