MATH 4301: Mathematical Analysis I

Tomoki Ohsawa
Updated on 2024/09/23 at 22:47:03
 Contents

Chapter 1. The Real Line and Euclidean Space 5

1.1. Ordered Fields and the Number Systems 5
1.2. Completeness and the Real Number System 7
1.3. Sequences 14
1.4. Summary of Completeness 19
1.5. lim inf and lim sup 20
1.6. Euclidean Spaces 24

Chapter 2. The Topology of Euclidean Spaces 29

2.1. Metric Spaces 29
2.2. Open Sets 31
2.3. Interior of a Set 35
2.4. Closed Sets 37
2.5. Accumulation Points 39
2.6. Closure of a Set 42
2.7. Boundary of a Set 45
2.8. Sequences 47

Chapter 3. Compact and Connected Sets 51

3.1. Compactness 51
3.2. The Heine–Borel Theorem 55
3.3. Nested Set Property 58

3
 The Real Line and Euclidean Space
1
1.1. Ordered Fields and the Number Systems

1.1.1. Ordered Fields. First recall that the R is an example of a field, that is, a F equipped with
addition “C” and multiplication “
” satisfying all the following 10 properties:

 Addition:
1. Commutativity: x C y D y C x 8x; y 2 F .
2. Associativity: x C .y C ´/ D .x C y/ C ´ 8x; y; ´ 2 F .
3. Zero: 90 2 F such that x C 0 D x 8x 2 F .
4. Additive inverse: 8x 2 F 9 x 2 F such that x C . x/ D 0.
 Multiplication:
5. Commutativity: x
y D y
x 8x; y 2 F .
6. Associativity: x
.y
´/ D .x
y/
´ 8x; y; ´ 2 F.
7. Unity: 91 2 F such that 1
x D x 8x 2 F .
8. Reciprocal: 8x 2 Fnf0g 9x 1 2 F such that x
x 1 D 1.
9. Distributive law: x
.y C ´/ D x
y C x
´ 8x; y; ´ 2 F .
10. Nontriviality: 1 ¤ 0

One quickly notices the set N D f1; 2; 3; : : : g of natural numbers is not a field; in particular, 3, 4, 8 are
not satisfied. Including zero and negative integers, one has the set Z of integers, which satisfy 3 and 4, but
do not satisfy 8. Now consider the set Q of rational numbers. Then one sees that Q constitutes a field. The
sec C of complex numbers is also a field.
In addition to being a field, both R and Q are equipped with an order in the following sense: Setting
F D R or Q,

 Order:
11. Reflexivity: 8x 2 F , one has x  x.
12. Antisymmetry: If x  y and y  x then x D y.
13. Tansitivity: If x  y and y  ´ then x  ´.
14. Linear ordering: For every pair x; y 2 F , either x  y or y  x.

5
Particularly,  satisfying the first three is called partial ordering, whereas if  satisfies all the four then it
is called total ordering.
By definition, x  y means y  x, and also x < y means that x  y and x ¤ y, and x > y means
that x  y and x ¤ y.
 Compatibility of Order with Operations:
15. Compatibility of  and C: If x  y then x C ´  y C ´ 8´ 2 F .
16. Compatibility of  and
: If 0  x and 0  y then 0  xy.

Definition 1.1.1: Ordered field

A set F equipped with addition “C” and multiplication “
” satisfying all the above 16 properties is
called an ordered field.

Clearly R and Q are ordered fields, but C is not an ordered field.

1.1.2. Number Systems. The natural numbers are the counting numbers 1; 2; 3; : : : . We define the
set of natural numbers as
N WD f1; 2; 3; : : : g:
As mentioned above, N is not a field: 3, 4, 8 are not satisfied. Note that N has total ordering by .
Let m; n 2 N, and consider the equation x C n D m for x. If n < m, then this equation has a solution
x D m n 2 N. However, it is not the case if n  m. In order for the equation to always have a solution in
the same number system, one extends the number system from N to the set of integers:
Z WD f: : : ; 3; 2; 1; 0; 1; 2; 3; : : : g:
Note that Z is not a field either: Now 3 and 4 are satisfied, but 8 is not.
Note that Property 8 is equivalent to saying that nx D 1 has a solution for any nonzero n. More
generally, let m; n 2 Z with n ¤ 0, and consider the equation nx D m. Then it does not always have a
solution in Z. To fix this issue, one extends the number system from Z to the set of rational numbers:
nm o
Q WD j m; n 2 Z; n ¤ 0 :
n
Then Q is a field, and particularly, an ordered field, as mentioned above.

6
 1.2. Completeness and the Real Number System

We have mentioned the set R of real numbers above. However, unlike N, Z, and Q, it is not straight-
forward to define R; one rather takes it for granted. The goal of this section is to give a precise definition of
R:

Goal
Construct the set R of real numbers from Q, and prove its fundamental properties.

1.2.1. Q is not rich enough. Although the rational number system Q is quite rich algebraically, one
sees that Q is not rich enough:

Proposition 1.2.1
p
There is no rational number r 2 Q such that r 2 D 2, i.e., 2 … Q.

We shall skip the proof as it is well known, and one can find it in many places.
In particular, if you consider a square whose side has length 1, then the length of the diagonal cannot be
represented by an element in Q. This necessitates to√consider the number system beyond Q.
1.1. Discussion: The Irrationality of 2 3

√
2
1

C
A 1 B
√
Figure 1.1: 2 exists as a geometric length.

The influential German mathematician Leopold Kronecker (1823–1891) once

1.2.2. Least Upperthat
asserted Bounds.
“The natural numbers are the work of God. All of the rest is
the work of mankind.” Debating the validity of this claim is an interesting
conversation
Definition 1.2.2: Upper for another
bound and time.
lower For the moment, it at least provides us with
bound
a place to start. If we restrict our attention to the natural numbers N, then
Let F be anwe can perform
ordered addition
field, and AF perfectly
. We say well,
A isbut we mustabove
bounded extendifour system
there existstouthe
2 F such that
integers
a  u for every a 2 A, and such an element u is called an upper bound for A. Similarly, we say
Z = {. . . , −3, −2, −1, 0, 1, 2, 3, . . .}
bounded below if there exists l 2 F such that l  a for every a 2 A, and such an element l is called
if wefor
a lower bound want
A. to
Wehave
sayan
A additive identity
is bounded if A(zero) and the
is bounded additive
both aboveinverses necessary
and below.
to define subtraction. The next issue is multiplication and division. The number
1 acts as the multiplicative identity, but in order to define division we need to
One realizeshave
thatmultiplicative
it is often theinverses.
case thatThus,
there we
are extend our one
more than system again to the
upper/lower rational
bound for a set A  F .
numbers
! "
p
Q = all fractions where p and q are integers with q ̸= 0 .
q
7
Taken together, the properties of Q discussed in the previous paragraph
essentially make up the definition of what is called a field . More formally stated,
a field is any set where addition and multiplication are well-defined operations
that are commutative, associative, and obey the familiar distributive property
 Definition 1.2.3: Least upper bound or sup

Let F be an ordered field, and A  F is non-empty. An element s 2 F is called the least upper
bound or supremum of A if it satisfies the following:
(i) s is an upper bound for A.
(ii) For every upper bound u for A, s  u.
In this case, we write s D sup A. Similarly, r 2 F is called the greatest lower bound or infimum of
A if it satisfies the following:
(i) r is a lower bound for A.
(ii) For every lower bound l for A, l  r.
In this case, we write r D inf A.

1.3. The Axiom of Completeness 15

inf A sup A
↓ ↓

lower bounds upper bounds

A

Figure 1.3: Definition of sup A and inf A.

Example 1.2.4: Consider

of the subject, we assume that R is an ordered  field , which contains Q as a
subfield. (A rigorous definition 1 of “ordered field” 1 1 is1 presented in Section 8.6.)
A WD j n 2 N D 1; ; ; ; : : :
This brings us to the final, n and most distinctive,
2 3 4 assumption about the real
as a number
subset of system. We
Q. We see must
that, e.g.,find
1, 2,some
3, areway to bounds
upper clearlyfor
articulate what
A, and also thatwe
anymean
upperby
bound
u2Q insisting that1Rdoes
must satisfy notsup
u. Thus contain
A D 1.the gaps that permeate Q. Because this is the
defining difference between the rational numbers and the real numbers, we will
be excessively precise about how we phrase this assumption, hereafter referred
This example is an instance of the following:
to as the Axiom of Completeness.
ExerciseAxiom
1.1. Let of  R. Prove that if Every
A Completeness. a 2 A nonempty
is an upper set
bound for A,
of real then a D
numbers supisAbounded
that (which is often
written as A). Similarly,
max has
above that if a 2 A is a lower bound for A, then a D inf A (which is often
provebound.
a least upper
written as min A). }
Now, what exactly does this mean?
The supremum does not always belong to the ordered field itself:
Least Upper Bounds and Greatest Lower Bounds
Example 1.2.5: Consider A WD f1; 1:4; 1:41; 1:414; 1:4142; : : : g, i.e., a sequence in Q converging
pLet’s first state the relevant definitions, and then look 2
at some examples. p
to 2. This is bounded above as each element a 2 A satisfies a  2. Intuitively, sup A D 2, but
Definition
this does to Q. A set A ⊆ R is bounded above if there exists a number b ∈ R
1.3.1.
not belong
such that a ≤ b for all a ∈ A. The number b is called an upper bound for A.
Similarly, the set A is bounded below if there exists a lower bound l ∈ R
1.2.3.
satisfying l ≤ a and
Completeness a ∈ A. System R. In the example above, we saw that a subset A in
Real Number
for every
Q that is bounded above may not have sup A in Q. This suggests that Q has some “gaps”.
Definition 1.3.2. A real number s is the least upper bound for a set A ⊆ R if
it meets the following two criteria:
8
(i) s is an upper bound for A;

(ii) if b is any upper bound for A, then s ≤ b.

 Definition 1.2.6: Completeness

An ordered field F is said to be complete if every non-empty set A  F that is bounded above has
the least upper bound sup A in F .

So Q is not complete, and thus we would like to define the real number system R by making it complete,
i.e., filling up the gaps:

Theorem 1.2.7: Construction of R from Q and Completeness of R

There exists a complete ordered field R containing Q such that Q is a subfield, i.e., a field itself
under the same addition and multiplication.

ROUGH S KETCH OF P ROOF. See, e.g., Abbott [1, Section 8.6] for details. The difficult part is to define
R in terms of Q: We have this collection of rational numbers Q that are almost filling up a line (which we
p
accept as R in Calculus) with some gaps (such as 2), and we must somehow define those possible gaps
using Q only. To do so, we define (Dedekind’s) cut as follows: A subset A  Q is called a cut if
(1) A ¤ ¿ and A ¤ Q.
(2) If r 2 A, then every q 2 Q with q < r is also in A.
(3) If r 2 A, then there exists q 2 A such that r < q.
For example, for every fixed r0 2 Q, the set Ar0 WD fr 2 Q j r < r0 g is a cut. So is, e.g., B WD
r 2 Q j r 2  2 or r < 0 . The idea is to identify those sets as “real numbers”. For example, Ar0 above is
˚
p
defined as r0 2 Q itself, and B would be identified with 2. More precisely, we define
R WD fall cuts in Qg:
What follows is to define operations between all cuts and order, e.g.,
A C B WD fa C b j a 2 A; b 2 Bg ; A  B is defined to be A  B:
One then show that this is indeed an ordered field (which is quite tedious).
Now, given A  R—which is a collection of cuts—that is bounded above, we define
S WD union of all cuts in A:
Then one shows that S is indeed sup A. □

So given an non-empty subset A  R, if A is bounded above, then sup A 2 R exists. If A is not

bounded above, then we define sup A D 1. Similarly, if A is bounded below, then inf A 2 R exists; if A is
not bounded below, then we define inf A D 1.

Proposition 1.2.8

Let A; B  R be both non-empty and bounded, and suppose that A  B. Then

inf B  inf A  sup A  sup B:

P ROOF. Exercise. □

9
 The definition of sup suggests that an upper bound s for A is sup A if and only if every number less than
s is not an upper bound for A. This is rephrased as the following useful characterization of sup:

Proposition 1.2.9: Characterization of inf and sup

(i) Let s 2 R be an upper bound for non-empty A  R. Then, s D sup A if and only if, for every
" > 0, there exists a 2 A such that s " < a.
(ii) Let r 2 R be a lower bound for non-empty A  R. Then, r D inf A if and only if, for every
" > 0, there exists a 2 A such that a < r C ".

P ROOF. We shall prove only (i), and leave (ii) as an Exercise.

()) If s D sup A, then any upper bound u for A satisfies s  u. Hence s " cannot be an upper bound,
i.e., there exists a 2 A such that s " < a.
(() The assumption says that any number less than s is not an upper bound for A. Thus any upper
bound u 2 R for A satisfies s  u. Therefore, s D sup A. □

Example 1.2.10: Given A D . 1; b/  R with b 2 R, sup A D b because b is clearly an upper

bound for A, and also, for every " > 0, we have b " < b 2" 2 A.

Exercise 1.2. Prove Proposition 1.2.9 (ii). }

Proposition 1.2.11: Archimedean Property

(i) For every x 2 R there exists n 2 N such that x < n.

(ii) For every " > 0 there exists n 2 N such that n1 < ".

P ROOF.

(i) By contradiction. Suppose the statement is false. Then there exists x 2 R such that n  x for every
n 2 N, i.e., N is bounded in R. So the completeness of R implies that N has the least upper bound
s WD sup N. By Proposition 1.2.9, this implies that there exist n 2 N such that s 1 < n, that is,
s < n C 1. But then this shows that n C 1 2 N is greater that s D sup N. Contradiction.
(ii) Set x WD 1=". Then the Archimedean Property implies that there exists n 2 N such that 1" D x < n,
which implies n1 < ".

Example 1.2.12: Given A WD 3 n1 j n 2 N , we claim that sup A D 3. First, clearly 3 is an

˚

upper bound for A. Also, for every " > 0 there exists n 2 N such that n1 < " by Proposition 1.2.11,
and thus 3 " < 3 n1 , where 3 n1 2 A. Therefore, by Proposition 1.2.9, sup A D 3.

One of the most important results in real analysis is the following:

10
 22 Chapter 1. The Real Numbers

Theorem
Theorem 1.4.3
1.2.13: Q (Density
is dense in R of Q in R). For every two real numbers a and b
with a < b, there exists a rational number r satisfying a < r < b.
For every pair a; b 2 R with a < b, there exists r 2 Q such that a < r < b.
Proof. A rational number is a quotient of integers, so we must produce m ∈ Z
and n ∈ N so that
P ROOF. We first assume that 0  a < b. Intuitively, we would like to first find a large enough n 2 N
to make a(1)“grid” of rational numbers m
a< < b.
nm n o
2Q j m2N
n
The first step is to choose the denominator n large enough so that consecutive
increments of size 1/n are too close together to “stepfall
that are precise enough so that at least one of those rational numbers over”
into the interval.a;(a,
the interval b/.b).

1 2 3 m−1 m
n n n ··· n n
• •
0 a b

Using
To that end, the
let us firstArchimedean Property
use Proposition 1.2.11 to (Theorem thatwe may pick n ∈ N large
1.4.2),
find n 2 N such
enough so that
1
1 < b a:
(2) n < b − a.
n
It suffices to find m 2 N such that
Inequality (1) (which we are trying to prove) is equivalent to na < m < nb.
With n already chosen, the idea m now is to choose m to be the smallest integer
a< < b ” na < m < nb:
greater than na. In other words, n pick m ∈ Z so that
The idea is to pick the smallest m 2 N such that (3)na <(4)m. (Note that, by the Archimedean Property
m − 1 ≤ na < m.
(Proposition 1.2.11), there exists at least one m 2 N such that na < m.) Then we have
Now, inequality (4) immediately yields a < m/n, which is half of the battle.
m 1  na < m
Keeping in mind that inequality (2) is equivalent to a < b − 1/n, we can use (3)
to write
The second inequality gives a < m 1
n as desired. On the other hand, the first inequality, along with n <
b a ” na C 1 < nb yields m ≤ na + 1
! "
1
m< nan Cb 1−< nb:+ 1 □
n
= nb.
If a < 0 < b, then the above argument applied to the interval .0; b/ gives r 2 Q such that r 2 .0; b/ 
.a; b/.
Because m < nb implies m/n < b, we have a < m/n < b, as desired.
If a < b  0, then 0  b < a, and thus there exists r 2 Q with r 2 . b; a/, i.e., r 2 .a; b/.
Theorem 1.4.3 is paraphrased by saying that Q is dense in R. Without
Corollary 1.2.14: Irrationals RnQ are dense in R
working too hard, we can use this result to show that the irrational numbers
are dense
For every pair a;in
b 2RRaswith
well.
a < b, there exists x 2 RnQ such that a < x < b.
Corollary 1.4.4. Given any two realpnumbers ap < b, there exists an irrational
number
P ROOF t satisfying
. Consider a<
the pair of realt < b.
numbers a 2 and b 2. By the above proposition, there exists
p p p p
r 2 Q such that a 2<r <b 2, and so a < r C 2 < b. Then r C 2 is irrational because it it
p
Proof. Exercise 1.4.5.
were rational then 2 would be rational too. □

11
 (b) Show that the implication goes the other way; that is, assume R possesses
the Cut Property and let E be a nonempty set that is bounded above.
Prove sup E exists.
(c) The punchline of parts (a) and (b) is that the Cut Property could be used
in place of the Axiom of Completeness as the fundamental axiom that
distinguishes
1.2.4. Countability. Twothe
setsreal numbers
A and B havefrom the rational
the same numbers.
cardinality To drive
if there exists f W A this
! B that
point home, give a concrete example showing that the Cut Property is not
is bijective or one-to-one correspondence, i.e., one-to-one and onto. Particularly, a set A is said to be
a valid statement when R is replaced by Q.
denumerable if it has the same cardinality as N. More generally, a set is said to be countable if it is either
finite or Exercise 1.3.11.
denumerable; Decide
otherwise if the
it is said following
to be statements about suprema and infima
uncountable.
are true or false. Give a short proof for those that are true. For any that are
Theorem
false,1.2.15:
supplyCountability
an exampleofwhere
Q the claim in question does not appear to hold.
The set(a) A and B are nonempty, bounded, and satisfy A ⊆ B, then sup A ≤
Q isIfcountable.
sup B.
P ROOF ( SKETCH
(b) If sup).A Let us define,
< inf B for for every
sets n 2 N,
A and B, then there exists a c ∈ R satisfying
a < c < b for pall a ∈ A and b ∈ B.
 
An WD ˙ 2 Q j p; q 2 N are in lowest terms with p C q D n ;
(c) If there existsq a c ∈ R satisfying a < c < b for all a ∈ A and b ∈ B, then
that is, sup A < inf B.
       
0 1 1 2 1 3
1.4A1 DConsequences
1
; A2 D ˙ ; A3 D ˙ ; ˙ ;
1 of Completeness 2 1
A4 D ˙ ; ˙ ;
3 1
 
1 2 3 4
A5 D
The first application of the Axiom ˙ ; of
˙ Completeness
; ˙ ; ˙ ; : : : is a result that may look
4 3 2 1
like a more natural way to mathematically express the sentiment that the real
lineAncontains
Then each no gaps.
is finite, and every rational number appears in exactly one of these sets. So Q is grouped into
those An ’s. One may then construct a bijection between N and Q. □
Theorem 1.4.1 (Nested Interval Property). For each n ∈ N, assume we
are given a closed interval In =interval
Next we would like to prove that the unit
[an , bn.0;
] = {x ∈ R : an ≤ x We
1/  R is uncountable.
≤ bshall
n }. prove
Assume
it using the
also that each In contains In+1 . Then, the resulting nested sequence of closed
following property of R that follows from the completeness:
intervals
Theorem 1.2.16: Nested Interval Property I1 ⊇ I2 (NIP)
⊇ I3 ⊇ I4 ⊇ · · ·
!∞
has a nonempty intersection;1that is, n=1 In ̸= ∅.
Consider a sequence fIn D Œan ; bn gnD1!of closed intervals that are nested such that In  InC1 for
∞
everyProof.
n 2 N.In order
theirtointersection
show that n=1 In is not empty, we are going to use the
T1
Then is non-empty, i.e., nD1 In ¤ ¿.
Axiom of Completeness (AoC) to produce a single real number x satisfying
x ∈ In for every n ∈ N. Now, AoC is a statement about bounded sets, and the
P ROOF
one. we
Define
wanttheto
setconsider
of all left is
endpoints
the set of the intervals:
A WDAfa=
n 2
{aRn j: nn ∈
2 Ng
N} R
Since anofleft-hand endpoints
bm for all n; m 2 N. Inofparticular,
the intervals.
we see that A is bounded in R.

A={an : n∈N}
" #$ %
[ [ [ [ ] ] ] ]
a1 a2 a3 · · · an · · · · · · bn · · · b3 b2 b1

Therefore, by completeness, there exists x WD sup A 2 R, and it satisfies an  x for every n 2 N. On the
other hand, we also know that, for every n 2 N, bn is an upper bound for A. Indeed, for 1  i < n, we
have ai  an  bn whereas for i  n, we have ai  bi  bn . Therefore, we have x  bn for every n 2 N.
Hence x 2 Œan ; bn  D In for every n 2 N, and thus x 2 1
T
nD1 In . □

12
(1) R = {x1 , x2 , x3 , x4 , . . .}

and be confident that every real number appears somewhere on the list. We
will now use the Nested Interval Property (Theorem 1.4.1) to produce a real
number that1.2.17:
Theorem is not there. of unit interval in R
Uncountability
Let I1 be a closed interval that does not contain x1 . Next, let I2 be a closed
The unit interval .0; 1/  R is uncountable.
interval, contained in I1 , which does not contain x2 . The existence of such an
I2 is Peasy to verify. Certainly I1 .0;
ROOF. By contradiction. Suppose that
contains two smaller disjoint closed intervals,
1/ is countable, i.e., one can write
and x2 can only be in one of these. In ˚ general, given an interval In , construct
.0; 1/ D xj j j 2 N :
In+1 to satisfy
We shall show that there exists x 2 .0; 1/ that is not any of the xj ’s—a contradiction.
(i) Let I1 ⊆.0;
In+1 In1/ and
be a closed interval (of non-zero length) that does not contain x1 , i.e., x1 … I1 . If
x2 … I1 then set I2 WD I1 . If x2 2 I1 , then pick I2 as a closed subinterval (of non-zero length) of I1 that
(ii) xn+1
does not
∈
/ I , i.e.,
contain xn+1
. I  I and x … I .
2 1 2 2 2

In
# $! "
[ [ ] • ] •
! "# $ xn+1 xn
In+1
%∞
WeRepeating
now consider
this process,the intersection
we have n=1 In .fInIfDxŒa
a nested closed intervals nn0 ; bisn gsome
1 real number from
nD1 with the property that xn … Ithe
n
T1
listforinevery
(1),n 2then we haveto xTheorem
N. According n0 ∈/ I , and
1.2.16
n0 it
(Nested follows
Interval that
Property), there exists x 2 I
nD1 n 
.0; 1/.
∞
However, x … xj j j 2 N because, for every&
˚
j 2 N, we have xj … Ij by construction, and so
T1
xj … nD1 In . x n0
∈
/ In . □
n=1

13
 this theme throughout the book, looking at sequences of functions and various
ideas regarding convergence.
Recall that a sequence in a set S is simply a function f : N -+ S. The
essence of the idea is that a sequence is a countable list of elements from S
arranged in a particular order. Subscript notation is often used to label the
elements f(n) = Xn, although sometimes other devices are used, such as Jn>
or x(n). The sequence is arranged as a list x 1,x2,X3, ...• Since the labeling is
somewhat arbitrary, we could relabel without loss of information, by putting
1.3. Sequences
Yk = Xk+~ and getting Yo,Y1 ,Y2,Y3, ... , and this might simplify the expression of
f. For example, the sequence
1.3.1. Sequences. A sequence in a set S is a map f W N ! S, i.e., a countable list of elements in S in
1 l I I
a particular order. We often write each f .n/ as, e.g., xn with subscripts, and fxn g1
shall16'""
2'4'8' nD1 for the sequence
as a whole.
can be represented as Xn = 1/2n for n =
1, 2, 3, ... or Yk =
I /2k+l for k =
0, 1, 2, 3, .... We typically write the sequence as Xn-
Definition
A sequence is said to1.3.1
converge to a limit x or to be convergent to x if we
can guarantee that the points in the sequence are as clo~e as we wish to x by
going far A sequence
enough put in the g1
nD1 This
fxnsequence. R isis said
made to converge
precise to a limit x 2 R if for every " > 0 there exists N 2 N
in the following
definition.
such that jxn xj < " whenever n  N . In this case, we write lim xn D x or xn ! x as n ! 1.
n!1
1.2.1 Definition A sequence Xn is said to converge to §1.2
a limitCompleteness
x iffor every and the Real Number System 37
e > 0 there is an integer N such that lxn - xi < e whenever n N. (The number
N may depend on e; smaller e may require larger N.) In this case we write
limn-+oo Xn = X or Xn -+ X as n -+ 00. /(n)

T
The definition requires that all terms beyond the Nth in the sequence be
x+t
closer than e to x. It is not enough to simply find one term close to x. The
----l----~-------------
1 I
Greek letter e (epsilon) can be thought of as standing for an allowed maximum I t I T •
error in approximation. You must be able lo assert that beyond some specific
x ----r-
1
• --+-----r-------
I I I • •
N, all terms of the sequence approximate x within this degree of predsion. A I I I I
feeling for this may be gained by plotting the points on a number line,x-£as in ----r-7--T-----r-------
Figure 1.2-1.
1 I I , I
, I I I I I
I I I I I I
_ _ _..____.2__3...___4_....5._.-.-.-N.____._I_N___,+_2__..._n
Xi X4
. ,.-
I e II 11•11 I I I e I
X - E X X + E N+ 1

FIGURE 1.2-1 Elements of a sequence Xn get within e of the limit for FIGURE
n large 1.2-2 Convergence viewed in terms of a graph

Proposition 1.3.2: Uniqueness of limit

Definition 1.2.1 requires that all but finitely many terms of the sequence lie
between
1
The limit of a convergent sequence fxn gnD1 x -Rcisand x + c. i.e.,
unique, More intuition
if x may be obtained from the graph of the
n ! x and xn ! y, then x D y.
function defining the sequence (Figure 1.2-2). Far enough to the right, all the
points must lie in the band of width 2£ around the horizontal line at height x. It
P ROOF. By contradiction. Supposeisxnot ¤ enough
y. Without lost3) of
that (3,x liesgenerality, wesince
in this band, may(4,x 4 ) doesx not.
suppose < y. Set
" D .y x/=2. Then there exists N1 ; N2 2 N such that
1.2.2 Sandwich Lemma Suppose Xn --+ L. Yn --+ L, and Xn Zn Yn for
jxn
xj < " 8nall n(itNis and tojxassume
1 enough n yj that 8nisanNNo
< " there 2 : such that Xn $ Zn $ Yn whenever
n > No). Then Zn --+ L
Setting N WD maxfN1 ; N2 g, the above two inequalities hold whenever n  N . Thus
y x D jy xj D jy xN CIntuitively,
xN xj  thisjxN
is so xj C jxNZn isyjtrapped
because < 2" D y and
above x: below by numbers
getting close to L. The complete proof is given at the end of the chapter. We
Contradiction. □ is that
will see further variations on this theme later. Another fact worth noting
the limit cannot exceed the bound of the sequence:
Since the limit is now guaranteed to be unique, we may not have to explicitly refer to the limit when a
sequence converges to it. So, in such cases, we shall often say a sequence fxn g1
nD1  R converges or is
1.2.3 Proposition If a $ Xn $ b for every n and Xn --+ x, then a $ x $ b.
convergent without explicitly referring to its limit.

Proposition 1.3.3: Squeeze Theorem The proof, which we leave to the reader, can be accomplished directly by a
method similar to that used for the sandwich lemma.
Suppose that sequences fxn g1 1 1
nD1 , fyn gnD1 , and f´n gnD1 in R satisfy xn  ´n  yn for every n 2 N
(or more generally there exists N0 21.2.4
N such that it holds
Example sequence nXn =N
Thewhenever (-),Jt,andn that
= 1, x2,n... ! L. several
; yn, illustrates
Then ´n ! L as well. things:

1. Some points of a sequence can be repeated; i.e., the function from N

defining it need not be one-to-one.
14
 P ROOF. Let " > 0 be arbitrary. Since xn ; yn ! L, there exists N1 ; N2 2 N such that

jxn Lj < " 8n  N1 ; jyn Lj < " 8n  N2 :

Set N WD maxfN0 ; N1 ; N2 g. Then we see that, whenever n  N ,

" < xn L  ´n L  yn L < ": 8n  N

Hence j´n Lj < " whenever n  N , i.e., ´n ! L. □

Proposition 1.3.4

Let fxn g1
nD1 be a sequence in R satisfying xn ! x as n ! 1.
(i) If a  xn for every n 2 N, then a  x.
(ii) If xn  b for every n 2 N, then x  b.

P ROOF. Exercise □

Definition 1.3.5: Bounded sequence

A sequence fxn g1
nD1  R is said to be bounded if there exists M > 0 such that jxn j  M for every
n 2 N.

Proposition 1.3.6

A convergent sequence in R is bounded.

P ROOF. Suppose that a sequence fxn g1

nD1 converges to x, i.e., xn ! x. Then, taking " D 1, there
exists N 2 N such that jxn xj < 1 whenever n  N , i.e.,

x 1 < xn < x C 1 8n  N;

which implies
jxj 1 < xn < jxj C 1 8n  N:
Thus, setting
M WD maxfjx1 j; : : : ; jxN 1 j; jxj C 1g;
we have jxn j  M for every n 2 N. □

Proposition 1.3.7: Limit theorems for sequences

Let fxn g1 1
nD1 and fyn gnD1 be sequences in R such that xn ! x and yn ! y as n ! 1, and  2 R
be a constant. Then, as n ! 1,
(i) xn C yn ! x C y.
(ii) xn ! x.
(iii) xn yn ! xy.
xn
(iv) If yn ¤ 0 for every n 2 N and y ¤ 0, then yn ! yx .

15
 P ROOF. Exercise. □

Proposition 1.3.8: Limits and order

Let fxn g1 1
nD1 and fyn gnD1 be sequences in R satisfying xn  yn for every n 2 N. If xn ! x and
yn ! y as n ! 1, then x  y.

P ROOF. Define a sequence f´n WD yn xn g1nD1 . Then ´n ! y x from the above Proposition. Then
´n  0 for every n 2 N, and thus y x  0 by Proposition 1.3.4 (i). □

1.3.2. Monotone Sequences. A sequence fxn g1 nD1 is said to be increasing (or non-decreasing) if xn 
xnC1 for every n 2 N, and strictly increasing if xn < xnC1 for every n 2 N. Likewise, it is said to be
decreasing (or non-increasing) if xn  xnC1 for every n 2 N, and strictly decreasing if xn > xnC1 for
every n 2 N. A sequence is said to be monotone if it is either increasing or decreasing.

Theorem 1.3.9: Monotone Convergence Theorem (MCT)

If a sequence in R is monotone and bounded, then it converges.

P ROOF. We shall assume that the sequence fan g1 nD1 is increasing. The proof is similar for the case
where it is decreasing.
2.4. The
The Monotone Convergence Theorem and Infinite Series
intuitive picture is that, since the sequence is increasing and bounded, the sequence clumps as
57
n ! 1.

aN s=sup{an :n∈N}

✛
a1 a2 a3 ··· ❄❄ ✲
• • • • • • • ••••
( •• )
s−ϵ s+ϵ

Now, let us consider the set of sequences

To prove this, let ϵ > 0. Because A WD fasn is

j nthe least
2 Ng ; upper bound for {an : n ∈ N},
s − ϵ is not an upper bound, so there exists a point in the sequence aN such
then it is bounded above and thus there exists s WD sup A. Intuitively, we conjecture that this would be the
that s − ϵ < aN . Now, the fact that (an ) is increasing implies that if n ≥ N ,
limit of the sequence.
aN ≤leta"n>. 0Hence,
thenIndeed, be arbitrary. Then, by Proposition 1.2.9, there exists a 2 A such that s " < a .
N N
However, since the sequence is increasing, this implies that s " < an whenever n  N ; but then, because
an  s for every n 2 N, we have s − ϵ < aN ≤ an ≤ s < s + ϵ,

jan sj D s an < " 8n  N: □

which implies |an − s| < ϵ, as desired.
1.3.3. Subsequences and the Bolzano–Weierstrass Theorem. Given a sequence fxn g1 nD1 , and fnk 2
NgkD1 be a strictly increasing sequence of natural numbers, i.e., n1 < n2 < n3 < : : : . Then fxnk g1
1
is
The Monotone Convergence
1
called a subsequence of fxn gnD1 .
Theorem is extremely useful
of for the study
kD1

infinite series, largely because it asserts the convergence of a sequence without

explicit mention of the actual limit. This is a good moment to do some prelimi-
nary investigations, so it is time to formalize
16 the relationship between sequences
and series.
 The Bolzano–Weierstrass Theorem
In the previous example, it was rather easy to spot a convergent subsequence
(or two) hiding
Proposition 1.3.10 in the original sequence. For bounded sequences, it turns out
that it is always possible to find at least one such convergent subsequence.
If fxn g1
nD1 is a convergent sequence in R, then every subsequence of it converges to the same limit.
Theorem 2.5.5 (Bolzano–Weierstrass Theorem). Every bounded sequence
contains a convergent subsequence.
P ROOF. Exercise. □

Theorem 1.3.11: Bolzano–Weierstrass Theorem

Proof. Let (an ) be a bounded sequence so that there exists M > 0 satisfying
n | ≤ bounded
|aEvery all n ∈ N.
M for sequence in R Bisect the closed
has a convergent interval [−M, M ] into the two closed
subsequence.
intervals [−M, 0] and [0, M ]. (The midpoint is included in both halves.) Now, it
1
must
P ROOF Let faat
be. that n gnD1
least
be aone of these
bounded closed
sequence, i.e., intervals M > 0 suchan
there exists contains jan j  M
thatinfinite number n2
for everyof
N.the
Ourterms
strategyinis the sequence
to extract (an ).subsequence
a convergent Select a halfusing for whichInterval
the Nested this isProperty
the case and label
(Theorem 1.2.16).
Firstinterval
that bisect the as
closed Then,Πlet
I1 . interval M; Man1tobe
closed
some term inΠthe
subintervals M; 0sequence
and Œ0; M .(a
Then at least one of
n ) satisfying
1
them ∈ Icontain
an1must 1.
infinitely many elements from fan gnD1 . Select such a subinterval and define it as I1 , and
pick n1 2 N so that an1 2 I1 . Next, we again bisect I1 into two closed subintervals, and again select one
that contains infinitely many elements from fan g1
nD1 . Then we pick n2 2 N so that n2 > n1 and an2 2 I2 .

an 2
I1
! "# $
❄
• • • • • • •••••••• • • • • • •
−M ✻ 0 M
# $! "
an 1 I2

Continuing this process to have a nested closed intervals fIk g1 kD1

with Ik  IkC1 for every k 2 N as
1
well as the subsequence fank gkD1 with ank 2 Ik for every k 2 N. Now, by the Nested Interval Property
Next, we there
bisect I1 ainto
2 closed
T1 intervals of equal length, and let I2 be a half
(Theorem 1.2.16), exists kD1 Ik .
that
So itagain contains
remains an infinite
to show that ank ! a number
as k ! 1. ofLet
terms
" > 0ofbethe original
arbitrary. sequence.
Notice Because
that the length of each
thereIkare
interval an kinfinite
is M=2 number
1 for every of terms
k 2 N. Hence there from (a2n )N to
exists K suchchoose from,
that M=2 K 1 <we can aselect
". Since nk 2 IK
an an2 kfrom
whenever thea original
 K, and we have with n2 > n1 and an2 ∈ I2 . In general, we
sequence
2 IK as well,
construct the closed interval Ik by taking M a half of Ik−1 containing an infinite
number of terms of (an ) and jank then
aj  select <
n " >8kn K: > · · · > n > n so that □
2K 1 k k−1 2 1
k
∈
an1.3.4. I k .
Cauchy Sequences.
We want to argue that (ank ) is a convergent subsequence, but we need a
Definition 1.3.12:
candidate Cauchy
for the limit.Sequence
The sets
A sequence fxn g1
nD1 is called a Cauchy sequence in R if, for every " > 0, there exists N 2 N such
that jxn xm j < " whenever n; m  N I1. ⊇ I2 ⊇ I3 ⊇ · · ·

form a nested
Proposition sequence
1.3.13: of closedsequence
Every convergent intervals, and by the Nested Interval Property
is Cauchy
there exists at least one point x ∈ R contained in every Ik . This provides us
Everythe
with convergent sequence
candidate in R islooking
we were a Cauchyfor.
sequence.
It just remains to show that (ank ) → x.
Let ϵ > 0. By construction, the length of Ik is M (1/2)k−1 which converges
toPzero. (This follows from Example 2.5.3 and the Algebraic Limit Theorem.) □
ROOF. Exercise.

Choose N so that k ≥ N implies that the length of Ik is less than ϵ. Because x

and ank are both in Ik , it follows that |a 17 nk
− x| < ϵ.
 The converse of the above in R is true as well:

Theorem 1.3.14: Cauchy Criterion

Every Cauchy sequence in R is convergent.

P ROOF. We first claim that every Cauchy sequence in R is bounded. Indeed, let fxn g1
nD1 be a Cauchy
sequence. Then, (taking " D 1) there exists N0 2 N such that jxn xm j < 1 whenever n; m  N0 . In
particular,
jxn xN0 j < 1 ” xN 1 < xn < xN C 1 8n  N0 :
Therefore, setting
M WD maxfjx1 j; : : : ; jxN0 1 j; jxN0 j C 1g;
we have jxn j  M for every n 2 N.
Since fxn g1
nD1 is bounded, by the Bolzano–Weierstrass Theorem (Theorem 1.3.11), there exists a sub-
sequence fxnk g1kD1
that converges to some x 2 R as k ! 1. Our claim is that the original sequence
converges to it as well, i.e., xn ! x as n ! 1. Let " > 0 be arbitrary. Then there exists N1 2 N such
that jxnk xj < "=2 whenever k  N1 . Also, since fxn g1 nD1 is Cauchy, there exist N2 2 N such that
jxn xm j < "=2 whenever n; m  N2 . Now set N WD maxfN1 ; N2 g. Then nN  nN1  N1 , and so
jxnN xj < "=2, whereas jxn xnN j < "=2 whenever n  N because nN  N  N2 as well. Therefore,
jxn xj  jxn xnN j C jxnN xj < " 8n  N:
As a result, xn ! x as n ! 1. □

Exercise 1.3. Suppose that sequence fxn g1

nD1 in R satisfies jxn xnC1 j < 1=3n . Prove that fxn g1
nD1
converges. }

18
 1.4. Summary of Completeness

We summarize the consequences of the completeness of R (Theorem 1.2.7) we have seen so far (plus
what is proved in the exercises below) as follows1:

Theorem 1.3.11 Theorem 1.3.14

NIP Bolzano–Weierstrass Cauchy Criterion
Theorem 1.2.16 Exercise 1.7

Exercise 1.5
Completeness Exercise 1.4
Theorem 1.2.7 Exercise 1.6
Theorem 1.3.9

Proposition 1.2.11 MCT

Archimedean
Property

Exercise 1.4. Assuming the statement of the Monotone Convergence Theorem (MCT; Theorem 1.3.9) prove
the Nested Interval Property (NIP; Theorem 1.2.16) directly, i.e., without assuming the Completeness (The-
orem 1.2.7). }

Exercise 1.5. Assuming the statement of the Nested Interval Property (NIP; Theorem 1.2.16), prove the
Completeness (Theorem 1.2.7). One may use the fact that 1=2n ! 0 as n ! 1 (since we cannot use the
Archimedean Property). }

Exercise 1.6. Assuming the statement of the Bolzano–Weierstrass Theorem (Theorem 1.3.11), prove the
Monotone Convergence Theorem (MCT; Theorem 1.3.9). }

Exercise 1.7. Assuming the statement of the Cauchy Criterion (Theorem 1.3.14) and the Archimedean Prop-
erty, prove the Bolzano–Weierstrass Theorem (Theorem 1.3.11). }

1The superscript  indicates that it requires the Archimedean Property as well.

19
 1.5. lim inf and lim sup

1.5.1. lim inf and lim sup. Let us start with a motivating example:

Example 1.5.1: Consider the sequence

1 1
  
xn WD 3 C . 1/n 1 C :
n nD1
This sequence does not converge as one can see in the plot.

5 10 15 20 25 30
However, we see that there are two subsequences converging to 2 and 4. Indeed, we notice that
8
<2 1 n odd;
n
xn D
:4 C 1 n even:
n

Therefore,
lim x2n 1 D 2; lim x2n D 4:
n!1 n!1

The goal of this section is to introduce those concepts that help us identify these “limits” of such a
(possibly) non-convergent sequence. Given a sequence fxn g1nD1 in R, define

an WD inf xk D inffxn ; xnC1 ; xnC2 ; : : : g; bn WD sup xk D supfxn ; xnC1 ; xnC2 ; : : : g;

kn kn

Observe that fan g1 1

nD1 is increasing whereas fbn gnD1 is decreasing, i.e., an  anC1 and bn  bnC1 for
every n 2 N. If fan g1 nD1 is bounded above then n!1lim an exists in R by the Monotone Convergence
Theorem (Theorem 1.3.9), whereas if it is not bounded above then lim an D 1. So lim an always
n!1 n!1
exists if one allows 1 as a limit. Similarly, if fbn g1
nD1 is bounded below then lim bn exists in R, whereas
n!1
if it is not bounded below then lim bn D 1; so lim bn always exists if one allows 1 as a limit. This
n!1 n!1
leads to the following:

20
 Definition 1.5.2: lim inf and lim sup

Given a sequence fxn g1

nD1 in R, we define its limit inferior and limit superior as
  !
lim inf xn WD lim inf xk ; lim sup xn WD lim sup xk
n!1 n!1 kn n!1 n!1 kn

respectively.

Example 1.5.3:
(a) Let fxn g1
nD1 D f 1; 0; 1; 1; 0; 1; 1; 0; 1; : : : g. Then lim inf xn D 1 and lim sup xn D 1.
n!1 n!1
(b) Going back to Example 1.5.1, consider the sequence
1 1
  
n
xn WD 3 C . 1/ 1 C :
n nD1
8 8
<x D 2 1 1
n nC1 D 4 C
odd; <x n odd;
n n nC1
inf xk D sup xk D
kn :xnC1 D 2 1 :xn D 4 C 1
nC1n even; kn n n even:
As a result, we have
lim inf xn D 2; lim sup xn D 4:
n!1 n!1

Proposition 1.5.4: Properties of lim inf and lim sup

Let fxn g1
nD1 be a sequence in R. Then
(i) lim inf xn  lim sup xn .
n!1 n!1
(ii) If there exists M 2 R such that xn  M for every n 2 N, then lim sup xn  M .
n!1
(iii) If there exists M 2 R such that M  xn for every n 2 N, then M  lim inf xn .
n!1
(iv) lim sup xn D 1 if and only if fxn g1
nD1 is not bounded above.
n!1
(v) lim inf xn D 1 if and only if fxn g1
nD1 is not bounded below.
n!1

P ROOF.
(i) Notice that
inf xk  sup xk :
kn kn

Then, taking the limit as n ! 1 of both sides and using Proposition 1.3.8 gives the desired result.

(ii) Since M is an upper bound for the set fxk j k  ng for every n 2 N,

sup xk D sup fxk j k  ng  M:

kn

21
 Then, taking the limit as n ! 1 of both sides of the inequality and using Proposition 1.3.4 gives the
desired result.

(iii) Similar to the previous one.

(iv) ()) Follows by taking the contrapositive of (ii).

(() If fxn g1
nD1 is not bounded above, then fxk j k  ng is not bounded above, i.e., supkn xk D 1
for every n 2 N. Thus lim supn!1 xn D 1.

(v) Similar to the previous one.

1.5.2. Subsequences and lim inf and lim sup. In one of the above examples, we have seen that, given
fxn g1
nD1 D f 1; 0; 1; 1; 0; 1; 1; 0; 1; : : : g, we have lim infn!1 xn D 1 and lim supn!1 xn D 1.
One can see that the lim inf and lim sup are the limits of subsequences of, e.g., f 1; 1; 1; : : : g and ,
f1; 1; 1; : : : g, respectively. Also, one can also find a subsequence, e.g., f0; 0; 0; : : : g that converges to 0 as
well, and 0 is between the lim inf and lim sup.
More generally, we have the following:

Proposition 1.5.5: Subsequence and lim inf and lim sup

Let fxn g1 1
nD1 be a sequence in R. If a subsequence fxnk gkD1 converges to x
N as k ! 1, then
lim inf xn  xN  lim sup xn :
n!1 n!1

Let us prove the following lemma before proving the above:

Lemma 1.5.6
Let fan g1
nD1 be a sequence in R. If lim an D a 2 R, then
n!1
inf ak  a  sup ak 8n 2 N:
kn kn

P ROOF. Let us just show the first inequality; one can prove the second in a similar manner.
Let n 2 N be arbitrary. We split it into two cases: (i) There exists n0  n such that an0 < a; (ii) ak  a
whenever k  n.
In the first case, clearly infkn ak  an0 , and so it follows that infkn ak < a.
In the second case, we claim that infkn ak D a; then the desired inequality follows. We first see that
the assumption implies that a is a lower bound for fak g1 kDn
. Also, since limn!1 an D a, for every " > 0
there exists K 2 N such that
jak aj < " 8k  K:
In particular, ak < a C " whenever k  K. So, for an arbitrary n 2 N, one can find k  maxfn; Kg so that
ak < a C ". Hence, by Proposition 1.2.9 (ii), infkn ak D a. □

22
 P ROOF OF P ROPOSITION 1.5.5. Since limj !1 xnj D x,
N the above lemma implies that
inf xnk  xN  sup xnk 8j 2 N:
kj kj

However, since fxnk g1

kD1
is a subsequence of fxn g1
nD1 , we have

inf xk  inf xnk ; sup xnk  sup xk ;

kj kj kj kj

and so
inf xk  xN  sup xk :
kj kj
Taking the limit as j ! 1, we obtain the desired result.
□

1.5.3. Relationship between lim, lim inf, and lim sup. Proposition 1.5.5 says that if there is a con-
vergent subsequence, then its limit falls between the (possible) gap between lim inf and lim sup. Intuitively,
this seems to imply that if there is no gap between them, then there is nowhere else for the sequence to go,
implying that the sequence converges. It turns out that this is the case, and moreover, the converse is true as
well:

Proposition 1.5.7: lim, lim inf, and lim sup

Let fxn g1
nD1 be a sequence in R and x 2 R. Then lim inf xn D lim sup xn D x if and only if
n!1 n!1
lim xn D x.
n!1

P ROOF. ()) First observe that

0  xn inf xk  sup xk inf xk 8k 2 N:
kn kn kn

The assumption implies that the right-hand side goes to 0 as n ! 1. Hence by the Squeeze Theorem
(Proposition 1.3.3), we have
lim xn D lim inf xk D x:
n!1 n!1 kn
(() Let " > 0 be arbitrary. Then there exists N 2 N such that
jxn xj < " ” x " < xn < x C " 8n  N:
This means that
x "  inf xk  sup xk  x C " 8n  N:
kn kn
But then this implies that
lim inf xk D lim sup xk D x: □
n!1 kn n!1 kn

23
 1.6. Euclidean Spaces

1.6.1. Euclidean Space and Vector Space. In this course, we shall study analysis not only in R but
also in higher-dimensional spaces such as R2 and R3 or those Euclidean spaces of higher dimensions too:

Definition 1.6.1: Euclidean n-space Rn

The Euclidean n-space Rn consists of all ordered n-tuples of real numbers, i.e.,
Rn WD f.x1 ; : : : ; xn / j x1 ; : : : ; xn 2 Rg :

We shall not go into the details of vector additions and scalar multiplications as the students should be
familiar with those concepts from Linear Algebra courses such as Math 2418.
We also expect the students to be familiar with the concept of basis for a vector space. In particular, one
often uses the following standard basis for Rn :

e1 D .1; 0; : : : ; 0/; e2 D .0; 1; 0; : : : ; 0/; :::; en D .0; : : : ; 0; 1/:

Then every x 2 R may be written as a linear combination of the standard basis:

n
X
xD xi ei :
i D1

One of the most fundamental facts about Rn is the following:

Theorem 1.6.2
The Euclidean n-space Rn with vector addition and scalar multiplication is a vector space of dimen-
sion n.

1.6.2. Norm and Inner Product on Rn . A standard way of measuring the distance in Rn is the fol-
lowing:

Definition 1.6.3: Standard norm on Rn

The (standard or Euclidean) norm or length of a vector x 2 Rn is defined as
n
!1=2
X
k
k W Rn ! R0 I kxk WD xi2 ;
i D1
where R0 WD fx 2 R j x  0g.

The norm also gives rise to the distance between every pair x; y 2 Rn :

n
!1=2
X
2
kx yk D .xi yi / :
i D1

24
 Proposition 1.6.4: Basic properties of norm in Rn

The norm k
k in Rn satisfies the following properties:

(i) kxk  0 8x 2 Rn .
(ii) kxk D 0 if and only if x D 0.
(iii) k˛xk D j˛j kxk 8˛ 2 R 8x 2 Rn .

P ROOF. Exercise. □

What about angles between vectors in Rn ? We first need to define an inner product to define the angle
between a pair of vectors:

Definition 1.6.5: Standard inner product on Rn

The standard inner product or dot product on Rn is defined as

n
X
h
;
iW Rn  Rn ! RI hx; yi WD xi yi :
i D1

In particular, one has hx; xi D kxk2 for every x 2 Rn .

Proposition 1.6.6: Basic properties of dot product in Rn

The standard inner product or dot product h
;
i in Rn satisfies the following properties:
(i) hx; xi  0 8x 2 Rn .
(ii) hx; xi D 0 if and only if x D 0.
(iii) h˛x; yi D ˛hx; yi 8˛ 2 R 8x; y 2 Rn .
(iv) hx; y C ´i D hx; yi C hx; ´i 8x; y; x 2 Rn .
(v) hx; yi D hy; xi 8x; y 2 Rn .

P ROOF. Exercise. □

We may define the angle between two vectors using the norm and the inner product: You may remember
that for vectors x; y in R2 or R3 , one has the relationship

hx; yi D kxk kyk cos ;

where  is the angle between the vectors x and y. In other words, if one projects x onto the line along y,
then the projected vector has the signed length kxk cos  D hx;yi
kyk
; see the figure.

25
 Thus we have llxll2 = {x,x}. In JR.3, the reader is familiar with another
expression for (x,y}, namely, {x,y} = llxll 11Yllcos0, where cos0 is the cosine
of the angle formed by x and y. See Figure l.6-3.

FIGURE
The idea extends this relationship 1.6-3thenLength
to Rn and define and
the inner
angle product
 2 Œ0;  between x and y in Rn by
setting
We now summarize the basic propenies of these operations:
hx; yi
cos  D :
kxk kyk
1.6.6 Theorem For vectors in IR.n, we have
However, in order for this definition to make sense, the right-hand side should always take values in Œ 1; 1.
The followingl.theorem
Properties of the
guarantees thatinner product:
this is the case:
i. (x, x} 2: 0. positivity
Theorem 1.6.7: The Cauchy–Schwarz Inequality
=
ii. (x,x) 0 iff x 0. = nondegeneracy
n
iii. k
(x,y
The standard norm k and+ the
w) =dot(x,y}
product
;
i in R satisfy the Cauchy–Schwarz
+ (x,hw). inequality:
distributivity
jhx; yij  kxk kyk 8x; y 2 Rn :
Also, the equality holds if and only if one of x and y is a scalar multiple of the other.

P ROOF. First observe that

p.t / WD kx C tyk2  0 8t 2 R:

However,

p.t / D hx C ty; x C tyi

D hx; xi C hty; xi C hx; tyi C hty; tyi
D hx; xi C t hy; xi C t hx; yi C t 2 hy; yi
D kxk2 C 2thx; yi C t 2 kyk2 :

Therefore, we have the following inequality for a quadratic function in t:

kyk2 t 2 C 2hx; yit C kxk2  0 8t 2 R:

26
The inequality implies that the quadratic function p.t / has either 0 or 1 real root. Hence its discriminant is
non-positive:
.2hx; yi/2 4 kyk2 kxk2  0 ” hx; yi2  kxk2 kyk2
” jhx; yij  kxk kyk :
Let us next prove the statement about the equality. The necessity (() is easy: If y D x (or x D y)
for some  2 R, then one sees that the equality is satisfied:
jhx; yij D jhx; xij D jhx; xij D jj kxk2 D kxk kxk D kxk kyk :
For sufficiency ()), if both x and y are zero, then the statement about the equality is trivial. So suppose,
without loss of generality, y ¤ 0. Notice that the equality jhx; yij D kxk kyk implies hx; yi D ˙ kxk kyk
p.t / D kxk2 ˙ 2t kxk kyk C t 2 kyk2 D .kxk ˙ t kyk/2
Setting t0 D  kxk = kyk, we have p.t0 / D 0, but then this implies
p.t0 / D kx C t0 yk2 D 0 H) x D t0 y: □

One of the consequences of the Cauchy–Schwarz inequality is:

Proposition 1.6.8: Triangle Inequality

The standard norm k
k in Rn satisfies the triangle inequality:

kx C yk  kxk C kyk 8x; y 2 Rn :

P ROOF. Following the same calculation as in the above proof with t D 1, and also using the Cauchy–
Schwarz inequality itself,
kx C yk2 D hx C y; x C yi
D kxk2 C 2hx; yi C kyk2
 kxk2 C 2 kxk kyk C kyk2
D .kxk C kyk/2 ;
which implies the desired inequality because both kx C yk and kxk C kyk are non-negative. □

Exercise 1.8. Prove that the standard norm k
k in Rn satisfies the following reverse triangle inequality:
ˇkxk kykˇ  kx yk 8x; y 2 Rn :
ˇ ˇ

27
 }

28
 The Topology of Euclidean Spaces
2
2.1. Metric Spaces

2.1.1. Metric spaces. Recall that, using the standard norm in Rn , one may define the distance between
every pair x; y 2 Rn as kx yk. We shall now define it as

n
!1=2
X
2
d.x; y/ WD kx yk D .xi yi / : (1)
i D1

Then it is starightforward to see that d satisfies the following properties:

(i) d.x; y/  0 8x; y 2 Rn .
(ii) d.x; y/ D 0 if and only if x D y.
(iii) d.x; y/ D d.y; x/ 8x; y 2 Rn .
(iv) d.x; y/  d.x; ´/ C d.´; y/ 8x; y; ´ 2 Rn (triangle inequality).

Definition 2.1.1: Metric space

A set M equipped with d W M M ! R satisfying the above four properties is called a metric space.
We shall often write such a metric space as a pair .M; d /, or just M for short, tacitly assuming that
it is equipped with such a function d .

In this course, we shall mainly concern the metric space .Rn ; d / with the above distance (1) called the
Euclidean metric (or distance) (given in terms of the standard norm on Rn ). When we refer to Rn , we
assume the Euclidean metric unless otherwise stated.
Here are some other examples:

Example 2.1.2:
(a) Let M WD fa; b; cgn D f.x1 ; : : : ; xn / j xi 2 fa; b; cg 8i 2 f1; : : : ; ngg. Given two sequences
x D .x1 ; : : : ; xn / and y D .y1 ; : : : ; yn / of words (or signals), we define their Hamming distance

29
 to be
dH .x; y/ WD number of places in which the words differ:
(b) Let M D R2 and define the taxicab metric as
dT .x; y/ WD jx1 y1 j C jx2 y2 j:

(c) Let M D C.Œa; b/, i.e., the set of real-valued continuous functions on Œa; b, and define
Z b
d.f; g/ WD jf .x/ g.x/j dx:
a

2.1.2. Main goal of this chapter. A good chunk of this section is devoted to those concepts that are
shown in the following picture:

Goal
Given a set A in a metric space M , we would like to define its interior int.A/, closure A, and
boundary @A.

30
 2.2. Open Sets

The building block for defining open sets on a metric space are open balls:

Definition 2.2.1: "-(open) ball

Let .M; d / be a metric space, and a 2 M and " > 0. Then the set
B" .a/ WD fx 2 M j d.x; a/ < "g
is called an "-(open) ball centered at a or "-neighborhood of a.

Definition 2.2.2: Open set

Given a metric space .M; d /, a subset U  M is said to be open or an open (sub)set if, for each
x 2 U , there exists " > 0 such that B" .x/  U .

Example 2.2.3: We shall assume the Euclidean metric (1) in Rn here.

(a) An open interval .a; b/  R is indeed open because for every x 2 .a; b/, one may take " WD
minfb x; x ag to have B" .x/  .a; b/.

31
 (b) Consider more or less the same set but this time in R2 :
A WD .x; 0/ 2 R2 j a < x < b  R2 :
˚

Then A is not open. Indeed, let x 2 A, and consider B" .x/ for an arbitrary " > 0. Then
B" .x/ 6 A.
(c) The closed ball with radius r > 0 centered at a 2 Rn , i.e.,
Br .a/ WD fx 2 Rn j d.x; a/  rg
is not open in Rn either, because if one takes x 2 Rn with d.x; a/ D r, then B" .x/ 6 B; see the
left figure below.

On the other hand, one can show that every "-ball is an open set:

Proposition 2.2.4: Every r-ball is open.

Let .M; d / be a metric space, a 2 M , and r > 0. Then the ball Br .a/ of radius r > 0 centered at
a 2 M is an open subset of M .

P ROOF. The main idea of the proof is in the right figure above. Let x 2 Br .a/ be arbitrary; then
d.x; a/ < r. Now, take " WD r d.x; a/. It suffices to show that B" .x/  Br .a/.
Indeed, let y 2 B" .x/ be arbitrary, i.e., d.y; x/ < ". Then, we have, using the triangle inequality,

d.y; a/  d.y; x/ C d.x; a/

< " C d.x; a/
Dr d.x; a/ C d.x; a/

32
 D r;

and hence y 2 Br .a/. □

Proposition 2.2.5: Topology in metric space

In a metric space .M; d /:

(i) The empty set ¿ and the whole space M are open.
(ii) The union of an arbitrary collection of open subsets of M is open.
(iii) The intersection of finite open subsets of M is open.

P ROOF.
(i) Since ¿ contains no points, it is vacuously satisfied that ¿ is open. Since every open ball in M centered
at a point in M is contained in M , the whole space M is also open.
S
(ii) Let fUi gi 2I be an arbitrary collection of open subsets in M , and we shall show that i 2I Ui is open.
S
Let x 2 i 2I Ui be arbitrary. Then there exists i0 2 I such that x 2 Ui0 , and since Ui0 is open, there
S S
exists " > 0 such that B" .x/  Ui0 . This implies that B" .x/  i 2I Ui , and thus i 2I Ui is open.
(iii) Let fUi gniD1 be an arbitrary collection of finite open subsets in M , and we shall show that niD1 Ui
T

is open. Let x 2 niD1 Ui be arbitrary. Then, for every i 2 f1; : : : ; ng, there exists ri > 0 such that
T

Bri .x/  Ui because Ui is open. Take " WD minfri gniD1 . Then B" .x/  Bri .x/  Ui for every
i 2 f1; : : : ; ng, and thus B" .x/  niD1 Ui .
T

Example 2.2.6:
(a) Let S1 WD .x; y/ 2 R2 j 1 < x < 2 . Show that S1 is open in R2 .
˚

Take an arbitrary .x; y/ 2 S1 . Then 1 < x < 2. Set " WD minfx 1; 2 xg. Then, we have
1 1
"  .x 1/; "  .2 x/;
2 2

33
 and so we have
1 1
x "x .x 1/ D .x C 1/ > 1;
2 2
and
1 1
x C "  x C .2 x/ D .x C 2/ < 2:
2 2
This shows that every point in the open ball B" ..x; y// is in S1 as well. Therefore, S is open.
(b) What if we include the vertical line x D 2 to the set? Let S2 WD .x; y/ 2 R2 j 1 < x  2 .
˚

Show that S2 is not open in R2 .

Take, e.g., .2; 0/ 2 S2 . For every " > 0, the ball B" ..2; 0// contains those points .x; 0/ with
2 < x < 2 C ", which are not in S2 . Hence S2 is not open.

Example 2.2.7: Let A  Rn be open and B  Rn . Prove that

A C B WD fa C b 2 Rn j a 2 A; b 2 Bg
is open.
Take an arbitrary a C b 2 A C B. Since a 2 A and A is open, there exists " > 0 such that B" .a/  A.
Now consider the ball B" .aCb/. We claim that B" .aCb/  ACB, which implies that ACB is open.
Indeed, let x 2 B" .a C b/ be arbitrary. Then, since kx .a C b/k < " ” k.x b/ ak < ",
we have x b 2 B" .a/  A. Therefore, we have
x D .x b/ C „ƒ‚…
b 2 A C B:
„ ƒ‚ …
2A 2B

Exercise 2.1. Let U  R be open and define A  R2 as

A WD .x; y/ 2 R2 j x 2 U :
˚

Prove that A is an open subset of R2 . }

34
 2.3. Interior of a Set
Definition 2.3.1: Interior of a set
Let M be a metric space and A  M . A point x 2 A is called an interior point of A if there is an
open set U such that x 2 U  A.

The interior of A is the collection of all interior points of A and is denoted by int.A/.

So an open set is a set in which every point of it is an interior point.

Example 2.3.2:
(a) Consider A WD .a; b  R with a < b. Then every point x 2 .a; b/ is an interior point (why?),
whereas b is not an interior point (why?). So int.A/ D .a; b/.
(b) Consider B WD x 2 R2 j kxk  1 . Then, since the unit disk B1 .0/  B is an open set, every
˚

point x 2 B1 .0/ (i.e., kxk < 1) is an interior point. However, those points x 2 B with kxk D 1
are not interior points (why?). Therefore, int.B/ D B1 .0/.

Proposition 2.3.3: Alternative characterization of the interior

Let M be a metric space and A  M . Then int.A/ is the union of all open sets contained in A, i.e.,
[
int.A/ D fU  M j U is open and U  Ag

P ROOF. Let us write

[
AV WD fU  M j U is open and U  Ag ;

and show that int.A/ D A.V

“” If a 2 int.A/, then there exists an open subset Ua such that a 2 Ua  A. But then, since Ua is an
open set contained in A, we have Ua  A. V Hence we have a 2 A. V
V V
“” If a 2 A then, from the definition of A, there exists U  A such that a 2 U . Hence a 2 int.A/. □

Corollary 2.3.4: The interior of a set is open.

For every subset A of a metric space M , its interior int.A/ is open.

P ROOF. It follows from the above proposition combined with Proposition 2.2.5 (ii). □

35
Exercise 2.2. Given a metric space M and A  M , prove that x is an interior point of A if and only if there
exists " > 0 such that B" .x/  A. }

Exercise 2.3. True of False: If True then prove it, and if False then provide a counterexample. Let n 2 N be
arbitray, and A and B be arbitrary pair of subsets in Rn .
(a) int.A/ [ int.B/ D int.A [ B/.
(b) int.A/ \ int.B/ D int.A \ B/.
}

36
 2.4. Closed Sets
Definition 2.4.1: Closed set
Given a metric space .M; d /, a subset C  M is said to be closed or a closed (sub)set if its
complement C c WD M nC is open.

Since ¿ and M itself are both open, they are also closed as well.

Example 2.4.2: A WD f.0; 0/g. Consider the complement Ac WD R2 nA. We shall show that Ac is
open, and hence A is closed. Indeed, take an arbitrary x 2 Ac . Then kxk > 0.

Take " WD kxk =2. Then, B" .x/  Ac , because for every y 2 B" .x/, we have ky xk < ", and so,
using the reverse triangle inequality from Exercise 1.8,
kxk kxk
kyk kxk > ky xk > "D H) kyk > > 0:
2 2
Hence B" .x/  Ac . Since x 2 Ac was taken arbitrary, this shows that Ac is open, and hence A is
closed.

More generally, every singple point in Rn is a closed set (assuming the standard Euclidean metric (1)):

Proposition 2.4.3: A single point in Rn is closed.

Every single point (more precisely every singleton subset) in Rn is closed.

P ROOF. Exercise. □

Example 2.4.4: Is the given subset of R2 closed?

(a) B WD x 2 R2 j kxk  1 .
˚

Its complement B c x 2 R2 j kxk < 1 is nothing but the unit ball B1 .0/, which is open
˚

due to Proposition 2.2.4. Hence B is closed.

37
 (b) C WD .x; y/ 2 R2 j 1 < x  2; 0  y  1 .
˚

Consider the complement C c , and take .1; y/ 2 C c with 0 < y < 1. Then, for every " > 0,
B" .0/ intersects C , i.e., B" .0/ 6 C c . Therefore, C is not open, and hence C c is not closed.

Exercise 2.4. Prove that, for every r > 0, Br .0/ WD fx 2 Rn j kxk  rg is a closed set. }

Proposition 2.4.5

In a metric space .M; d /,

(i) The empty set ¿ and the whole space M are closed.
(ii) The intersection of an arbitrary collection of closed subsets of M is open.
(iii) The union of finite closed subsets of M is closed.

P ROOF. Follows easily from Proposition 2.2.5:

(i) ¿c D M and M c D ¿ are both open.
T
(ii) Let fCi gi 2I be an arbitrary collection of closed subsets in M , and we shall show that i 2I Ci is close.
Indeed, !c
\ [
Ci D Cic
i 2I i 2I
is open because each Cic is open.
(iii) Let fCi gniD1 be an arbitrary collection of finite closed subsets in M , and we shall show that niD1 Ci
S

is closed. Indeed, !c
[n n
\
Ci D Cic
iD1 i D1
is open because each Cic is open.
□

Example 2.4.6: Prove that every finite set in Rn is closed.

We have seen above that every single point is closed. Since any finite set is a finite union of single
points, (iii) of the above proposition implies that the finite set is closed as well.

Exercise 2.5. Let I WD RnQ be the set of irrational numbers. Is I a closed subset of R? }

38
 2.5. Accumulation Points

In the previous section, we determined whether a given set is closed by examining whether its comple-
ment is open. Another way of doing so is via the concept of an accumulation point:

Definition 2.5.1: Accumulation points

Given a subset A of a metric space M , a point x 2 M is called an accumulation point (or limit
point) of A if every open set U containing x contains some point in A other than x, i.e., .U nfxg/ \
A ¤ ¿.

Exercise 2.6. Let A; B be subsets of a metric space M with A  B. Explain why every accumulation point
of A is an accumulation point of B as well. }

One may characterize it using only open balls (as opposed to open sets) as follows:

Proposition 2.5.2

Given a subset A of a metric space M , a point x 2 M is accumulation point of A if and only if, for
every " > 0, B" .x/ containts some point in A other than x, i.e., .B" .x/nfxg/ \ A ¤ ¿.

P ROOF. Exercise.
□

Example 2.5.3: For each given subset of R, find its accumulation points.
(a) A1 D Œ0; 1.
If x < 0 then x is not an accumulation point of A1 , because one sees that Bjxj .x/ \ A1 D ¿.
Similarly, if x > 1 then it is not an accumulation point of A1 either, because Bx 1 .x/ \ A1 D ¿.
Every point x 2 .0; 1/ is an accumulation point of A1 because for every " > 0, B" .x/ clearly
contains points in A1 other than x.
Finally, 0 and 1 are also accumulation points of A1 because for every " > 0, B" .0/ clearly
contains points in A1 other than x, and so does B" .1/. So the accumulation points of A1 are
Œ0; 1.
(b) A2 D Œ0; 1/. Following the same argument as above, one sees that the accumulation points of
A2 are again Œ0; 1.
(c) A3 WD Œ0; 1/ [ f2g.
One sees that none of the elements in B WD . 1; 0/ [ .1; 2/ [ .2; 1/ is an accumulation
point: Indeed, for every element x 2 B, one may find a small enough " > 0 so that B" .x/\A3 D
¿.
On the other hand, every element in Œ0; 1 is an accumulation point following a similar argu-
ment as the previous two examples.

39
 How about 2 2 A3 then? One finds that B1 .2/ contains no points in A3 other than 2. Hence
2 is not an accumulation point.
So the accumulation points of A3 are again Œ0; 1.

Notice that in the above examples, A1 is the only set that is closed, and also is the only set that contains
all accumulation points of it. This generalizes to the following:

Theorem 2.5.4: Closed set and accumulation points

A subset A of a metric space M is closed if and only if A contains all accumulation points of it.

P ROOF. ()) Suppose that A is closed, and let x 2 M be an accumulation point of A. We shall show
that x 2 A. Indeed, if x was not in A, then x 2 Ac , but then, since Ac is open, there exists " > 0 such that
B" .x/  Ac , and hence B" .x/ contains points no points in A. This contradicts that x is an accumulation
point of A.
(() Suppose that A contains all accumulation points of it. We shall show that Ac is open. Let x 2 Ac
be arbitrary. Then, by the assumption, x is not an accumulation point of A. Thus there exists " > 0 such
that B" .x/ contains no points of A other than x. Since x … A, this means that B" .x/ contains no points of
A, i.e., B" .x/  Ac . Therefore Ac is open. □

Example 2.5.5: Let A WD .x; y/ 2 R2 j 1  x  2 or x D 3 .

˚

Let us find the accumulation points of A. If we take a point .x; y/ 2 R2 with 1  x  2, then for
every " > 0 the open ball B" ..x; y// clearly contains points in A other than x. We also see that, for
an arbitrary point .2; y0 / 2 R2 with y0 2 R, for every " > 0, the open ball B" ..2; y0 // contains
those points .2; y/ with y0 " < y < y0 C " as well.
It is also clear that, for every .x; y/ 2 Ac , one can find " > 0 such that B" ..x; y// is contained in
Ac , and so no points in Ac are accumulation points.

40
 As a result, the accumulation points of A are A itself, and by the above theorem, A is closed. Indeed,
one can also directly see that A is closed because Ac is clearly open.

Exercise 2.7. Find the accumulation points of the set

   
1 2
A WD x; sin 2R j 0<x1
x
in R2 ; see the picture below.

1.0

0.5

0.2 0.4 0.6 0.8 1.0

- 0.5

- 1.0

41
 2.6. Closure of a Set
Definition 2.6.1: Closure of a set

Let M be a metric space and A  M . The closure of A, denoted A, is defined to be the intersection
of all closed sets containing A, i.e.,
\
A WD fC  M j C is closed and A  C g :

Note that A is closed for every A  M (why?). Therefore, A is the smallest closed set containing A.

Lemma 2.6.2
Let M be a metric space and A  M , and define
AO WD A [ faccumulation points of Ag:
Then
(i) every accumulation point of AO is also an accumulation point of A;
(ii) AO is closed.

P ROOF.
(i) Let x be an accumulation point of A. O Then, for every " > 0, B" .x/ contains a point y 2 Anfxg.
O Then
either y 2 A or y is an accumulation point of A. In order to show that x is an accumulation point of
A, it remains to show that the latter case implies that B" .x/ contains another point ´ 2 A that is not x.
So suppose that y is an accumulation point of A. Then, setting

r WD minfd.x; y/; " d.x; y/g;

the open ball Br .y/ contains a point ´ 2 A that is not y. But then we have

d.´; y/ < r  d.x; y/; d.´; y/ < r  " d.x; y/:

The first equality combined with the triangle inequality d.x; y/  d.x; ´/ C d.´; y/ gives

d.´; y/ < d.x; ´/ C d.´; y/ H) 0 < d.x; ´/:

42
 On the other hand, the second inequality gives

d.x; ´/  d.x; y/ C d.y; ´/ < ";

i.e., ´ 2 B" .x/nfxg. Hence B" .x/ contains a point ´ 2 A that is not x.

(ii) Since A  A,O the above result implies that every accumulation point of AO belongs to A.
O Hence the
claim follows from Theorem 2.5.4.
□

It turns out that A and AO are the same thing:

Proposition 2.6.3

Let M be a metric space and A  M . Then A consists of A and the accumulation points of A, i.e.,
A D A [ faccumulation points of Ag:

P ROOF. We would like to show that A D A. O

Let us first show that AO  A. First recall from Theorem 2.5.4 that a subset (of M ) is closed if and only
if it contains all accumulation points of it. Therefore, if C is a closed set such that A  C , then C must
contain all accumulation points of C and hence those of A as well; this in turn implies that AO  C , and
hence it follows that AO  A because A is the intersection of all such C ’s.
Next we shall show that A  A. O Indeed, by the above lemma, AO is a closed set such that A  A, O and
O
hence A  A because A is the intersection of all those closed C ’s with A  C .
□

Exercise 2.8. Let A; B be arbitrary subspaces in a metric space M . Is the given statement True or False? If
True then prove it, and if False then provide a counterexample.
(a) A \ B D A \ B.
(b) A [ B D A [ B.
}

Example 2.6.4: Let A be a subset of a metric space M . Show that x 2 A if and only if
inf fd.x; y/ j y 2 Ag D 0.
()) Suppose x 2 A. Then either x is in A or an accumulatin point of A. If x 2 A, then
d.x; x/ D 0 and so clearly inf fd.x; y/ j y 2 Ag D 0 because 0 is clearly a lower bound for
the set fd.x; y/ j y 2 Ag and 0 is in the set. If x is an accumulatin point of A, then for every
" > 0, B" .0/ contains a point y 2 A that is not x, i.e., d.x; y/ < ". Since 0 is a lower bound for
fd.x; y/ j y 2 Ag, this implies that 0 is inf fd.x; y/ j y 2 Ag due to Proposition 1.2.9 (ii).
c
(() We prove the contrapositive. Suppose x 2 A , which is open. Then there exists " > 0 such
c
that B" .x/  A  Ac . Thus A  B" .x/c , i.e., d.x; a/  " for every y 2 A. Therefore 0 < " 
inf fd.x; y/ j y 2 Ag.

43
Exercise 2.9. Consider A WD Œ0; 1 \ Q as a subset of R. Prove that A D Œ0; 1. }

44
 2.7. Boundary of a Set

Let us go back to the picture from the beginning of the chapter.

This picture suggest us to define the boundary @A of a set A in a metric space as follows:

Definition 2.7.1: Boundary of a set

Let A be a subset of a metric space .M; d /. Then the boundary of A is defined as

@A WD A nint.A/:

The following characterization is also intuitively natural too:

Proposition 2.7.2

Let A be a subset of a metric space M . Then x 2 @A if and only if, for every " > 0, B" .x/ contains
points of A and of Ac (which could be x itself).

P ROOF. ()) Since x 2 A, either x 2 A or x 2 Ac but is an accumulation point of A. Suppose x 2 A.

Since x … int.A/, for every " > 0, B" .x/ contains a point in Ac . But then x 2 A as well, and so B" .x/
contains points of A and of Ac . On the other hand, if x 2 Ac , then x is an accumulation point of A. Thus,
for every " > 0, B" .x/ contains a point in A other than x. Since x 2 Ac , B" .x/ contains points of A and of
Ac .
(() Let x 2 M and suppose that, for every " > 0, B" .x/ contains points of A and of Ac . If x 2 A,
then clearly x 2 A, but since B" .x/ contains a point in Ac , x … int.A/. Thus x 2 @A. If x 2 Ac then clearly
x … int.A/. Also, since B" .x/ contains a point in A (which clearly cannot be x) for every " > 0, x is an
accumulation point of A. Thus x 2 A; hence x 2 @A. □

Example 2.7.3: Let A be a subset of a metric space M .

True or False: Every x 2 @A is an accumulation point of A.
False. Consider M D R and A D f0g. Then A has no accumulation point, and so A D A. Also,
int.A/ D ¿ because 0 is not an interior point of A. Thus @A D f0g. So 0 2 @A but 0 is not an
accumulation point of A.

Example 2.7.4: Consider A D Œ0; 1 \ Q as a subset of R. Find @A.

Let us first show that no points in Œ0; 1c D . 1; 0/ [ .1; 1/ is in @A. Indeed, if x < 0 then
B x=2 .x/ does not intersect A, whereas if x > 1 then B.1 x/=2 .x/ does not intersect A.

45
 On the other hand, if x 2 Œ0; 1 then, for every " > 0, B" .x/ contains both rational in Œ0; 1 (which is
A) and irrationals in Œ0; 1 (which is in Ac ) because of the density of rationals and irrationals in R;
see Theorem 1.2.13 and Corollary 1.2.14
As a result, we have A D Œ0; 1.

Exercise 2.10. In the above example, use the definition of @A to show that @A D Œ0; 1. }

Exercise 2.11. True of False: If True then prove it, and if False then provide a counterexample.
(i) If A and B are subsets of a metric space M and A  B, then @A  @B.
(ii) For every subset A of a metric space M , @A D @.int.A//
}

46
 2.8. Sequences

2.8.1. Convergence of sequence in metric space.

Definition 2.8.1: Convergent sequence in metric space

Let .M; d / be a metric space. A sequence fxk g1

kD1
in M is said to coverge to x 2 M if, for every
" > 0, there exists N 2 N such that
d.xk ; x/ < " 8k  N:
In this case we write lim xk D x or xk ! x as k ! 1.
k!1

Example 2.8.2: Let A be a subset of metric space M , and consider a sequence fxk g1 kD1
in A that
converges in M . Is its limit x in A?
Consider the sequecne fxk D 1=kg1 kD1
in R, and let A D Œ0; 1 and B D .0; 1. Clearly fxk g1
kD1
is
a sequence in both A and B, and xk ! 0. Although 0 is in A, it is not in B.

In the above example, notice that A is closed but B is not. This suggests the following:

Theorem 2.8.3: A is closed ” every convergent sequence in A converges to a point in A

Let M be a metric space. A subset A  M is closed if and only if, for every sequence fxk g1
kD1
A
that converges in M , its limit x is in A.

P ROOF. ()) Suppose that A is closed, and that fxk g1 kD1

is a sequence in A that converges to x 2 M .
Suppose, on the contrary, x 2 A . Since A is open by assumption, there exists " > 0 such that B" .x/  Ac .
c c

But then this implies that B" .x/ contains no elements from the sequence fxk g1
kD1
; this contradicts xk ! x.
(() By Theorem 2.5.4, it suffices to show that every accumulation point of A is in A. Let x 2 M be an
accumulation point of A. Then, for every k 2 N, there exists xk 2 A such that d.xk ; x/ < k1 . This implies
that xk ! x because for every " > 0 one can find N 2 N such that N1 < " (Proposition 1.2.11) and thus
1 1
d.xk ; x/ <  < " 8k  N:
k N
By assumption, we then have x 2 A. □

Going back to the above example, we see that 0 2 B D Œ0; 1 although 0 … B. But then there exists a
sequence in B that converges to 0. This generalizes to the following:

Theorem 2.8.4: x 2 A ” 9 sequence in A converging to x

Let M be a metric space and A  M . Then x 2 A if and only if there exists a sequence fxk g1
kD1
in
A that converges to x.

P ROOF. Exercise. □

47
 2.8.2. Sequences in Rn . Our main interest is the case with M D Rn with the standard (Euclidean)
metric:

Example 2.8.5: In Rn , a sequence fxk D .xk1 ; : : : ; xkn /g1 kD1

of vectors in Rn is said to converge to
x D .x 1 ; : : : ; x n / if, for every " > 0, there exists N 2 N such that
kxn xk < " 8n  N:

Proposition 2.8.6: Convergence in Rn

A sequence fxk D .xk1 ; : : : ; xkn /g1

kD1
of vectors in Rn converges to x D .x 1 ; : : : ; x n / if and only
if each sequence of coordinates converges to the corresponding coordinates of x, i.e., xki ! x i as
k ! 1 for every i 2 f1; : : : ; ng.

P ROOF. Exercise. □

Proposition 2.8.7: Properties of convergent sequences in Rn

Suppose that fxk g1

kD1
and fyk g1
kD1
be sequences in Rn , and fk g1
kD1
be a sequence in R satisfying
xk ! x 2 M; yk ! y 2 M; k !  2 R
as k ! 1. Then:
(i) xk C yk ! x C y.
(ii) cxk ! cx 8c 2 R.
(iii) k ´ ! ´ 8´ 2 M .
(iv) k xk ! x.

P ROOF. We leave the first three as an exercise, and prove only the last one. First observe that, for every
k 2 N,

kk xk xk D kk xk k x C k x xk

 kk xk k xk C k.k /xk
D jk j kxk xk C jk j kxk :

Now, since fk g1

kD1
is convergent, it is bounded. Thus there exists M > 0 such that jk j  M for
every k 2 N. Let " > 0 be arbitrary. Then there exists N1 ; N2 2 N such that
" "
kxk xk < 8k  N1 ; jk j < : 8k  N2 :
2M 2.kxk C 1/
Then, setting N WD maxfN1 ; N2 g, we have, whenever k  N ,

kk xk xk  M kxk xk C jk j kxk

" kxk
<M
C "
2M 2.kxk C 1/

48
 " "
< C
2 2
D ": □

Proposition 2.8.8: Bolzano–Weierstrass Theorem in Rn

Every bounded sequence in Rn has a convergent subsequence.

P ROOF. Let fxk g1 kD1

be a bounded sequence in Rn . Then there exists R > 0 such that fxk gk2N 
Œ R; Rn , i.e., jxki j  R for every i 2 f1; : : : ; ng.
Specifically, we have a bounded sequence fxk1 gk2N in R. Thus, by the Bolzano–Weierstrass Theorem
in R (Theorem 1.3.11), there exists a convergent subsequence fxk1 gk2K1 with some countably infinite subset
K1  N. Let x 1 2 R be its limit. Next we consider fxk2 gk2K1 and follow the same argument to find a
convergent subsequence fxk2 gk2K2 with some countably infinite subset K2  K1 , and le x 2 2 R be its
limit.
Continue this process up to the n-th element to have the nested countably infinite subsets Kn  Kn 1 



 K2  K1  N so that, for every i 2 f1; : : : ; ng, fxki gk2Ki converges to some x i 2 R. Notice that, for
every i 2 f1; : : : ; ng, fxki gk2Kn is a subsequence of the convergent sequence fxki gk2Ki , and thus fxki gk2Kn
converges to x i as well. Hence we have subsequence fxk gk2Kn that converges to x WD .x 1 ; : : : ; x n /. □

Exercise 2.12. Consider the sequence

1
.sin n/n 1
 
xn WD ; 2
n n nD1
in R2 . Find its limit, and prove that the sequence indeed converges to it. }

Exercise 2.13. Let n 2 N be arbitrary. Prove that a sequence in Rn conveges if and only if it is Cauchy. }

Exercise 2.14. Let n 2 N be arbitrary.

(1) Prove that if a sequence fxk g1kD1
in Rn satisfies xk ! x as k ! 1 for some x 2 Rn , then
kxk k ! kxk as k ! 1.
(2) Use the above result to give another proof that B1 .0/ D fx 2 Rn j kxk  1g is closed (instead of
proving that its complement is open).
}

49
 Compact and Connected Sets
3
3.1. Compactness

3.1.1. Open covers.

Definition 3.1.1: Open cover

Let M be a metric space. Given a subset A  M , an open cover of A is a (not necessarily countable)
collection O WD fUi gi 2I of open subsets in M such that
[
A Ui :
i 2I
A finite subcollection of O is called a finite subcover of A if it is (still) an open cover of A.

Example 3.1.2: Let M D R with the standard metric.

(a) Let C D Œ0; 1. The collection
O1 WD fB1 .r/ D .r 1; r C 1/ j r 2 Œ0; 1 \ Qg
is an open cover of C because
[
C  B1 .r/:
r2Œ0;1\Q

The subcollection fB1 .0/ D . 1; 1/; B1 .1/ D .0; 2/g of O1 is a finite subcover because it is a
finite subcollection of O1 and
C  B1 .0/ [ B1 .1/:
(b) Let A D .0; 1/. Find an open cover of A with no finite subcover.
Consider the collection
1 1
  
1
O2 WD Un WD ;1 :
n n nD1

51
 of open intervals. It is an open cover of A because
1
[
AD Un :
nD1
However, no finite subcollection of O2 is an open cover.

3.1.2. Compact sets. Notice that, in the above example, no finite subcollection of A gives an open
cover.

Definition 3.1.3: Compact set in metric space

Let M be a metric space. A subset A  M is said to be compact if every open cover of A has a
finite subcover.

Note that the “every” is critical in the above definition; in other words, just because there is some open
cover of A that has a finite subcover doesn’t mean that A is compact. For example, consider A D .0; 1/
from above example. Then the collection f. 1; 2/g is an open cover of A, and is a finite subcover itself.
However, A is not compact because the open cover we saw in the example does not have a finite subcover.

Exercise 3.1. Let

A WD .x; y/ 2 R2 j 0  x < 1; 0  y  1 :
˚

Prove that there exists an open cover of A that does not have a finite subcover (and hence A is not compact).
}

Exercise 3.2. Let fxk g1

kD1
be a sequence in R that converges to x 2 R, and define
A WD fxk j k 2 Ng [ fxg:
Prove that every open cover of A has a finite subcover (and hence A is compact). }

Proposition 3.1.4: Compactness of closed interval in R

Let a; b 2 R with a < b. Then the closed interval Œa; b is compact.

P ROOF. Let O WD fUi gi 2I be an open cover of Œa; b, and set

C WD fx 2 Œa; b j Œa; x can be covered by a finite subcollection of Og :
We would like to show that b 2 C .
To that end, let us first show that c WD sup C exists and c 2 Œa; b. Since the singleton set fag is compact
(why?), a 2 C , and so C ¤ ¿. Also, b is clearly an upper bound for C . Thus C is a non-empty set that is
bounded above, and hence c WD sup C exists, and c  b because b is an upper bound for C . Clearly, a  c
because a 2 C and c is an upper bound for C . Therefore, a  c  b.
Next we shall show that c 2 C . Since c 2 Œa; b, there exists i0 2 I such that c 2 Ui0 . However, Ui0
being open, there exists " > 0 such that .c "; c C "/  Ui0 . On the other hand, since c D sup C , there
exists x 2 C such that c " < x  c. But then that x 2 C implies that there exists a finite subcollection

52
fUin gN N
nD1 that covers Œa; x. Notice that then fUin gnD1 [ fUi0 g gives a finite collection that covers Œa; c.
Hence c 2 C .
Finally, we shall show that c D b, which implies b 2 C as desired. By contradiction. If c < b, then
Œa; c is covered by a finite subcollection of O, but one can append it by Ui0 from above so that Œa; c C "=2
is also covered by a finite subcollection of O, implying c C "=2 2 C . This contradicts that c D sup C . □

3.1.3. Implication of Compactness.

Definition 3.1.5: Bounded set in metric space

Let .M; d / be a metric space. A subset A  M is said to be bounded if there exists R > 0 such that
d.x; y/  R for all x; y 2 A.

Exercise 3.3. Let .M; d / be a metric space. Prove that a subset A  M is bounded if and only if there exist
x0 2 M and R > 0 such that d.x0 ; y/  R for every y 2 A. }

Proposition 3.1.6: Compact ) Closed and Bounded

In a metric space, every compact set is closed and bounded.

P ROOF. Let .M; d / be a metric space and A  M be compact.

We shall first show that A is closed by showing that Ac is open. Let x 2 Ac and we shall show that
there exists an open set U such that x 2 U  Ac . Let a 2 A be arbitrary, and pick r.a/ 2 .0; d.x; a/=2/.
Then we have Br.a/ .x/ \ Br.a/ .a/ D ¿. Now consider the collection of open balls
˚
O WD Br.a/ j a 2 A

Then O is an open cover of the compact set A. Hence there exists a finite subcover, i.e., there exist finite
points fai gN N
i D1 in A along the radii fri WD r.ai /gi D1 such that
˚
ON WD Bri .ai / j i 2 f1; : : : ; N g

gives an open cover of A. However, since Bri .x/ \ Bri .ai / D ¿ for every i 2 f1; : : : ; N g, So if we set
r WD min ri , then Br .x/ \ Bri .ai / D ¿ for every i 2 f1; : : : ; N g, and thus Br .x/ \ A D ¿ because ON
1i N
is an open cover of A. Therefore, x 2 Br .x/  Ac .
It remains to show that A is bounded. Consider the collection of unit balls centered at every point of A,
i.e.,
fB1 .a/ j a 2 Ag :
It clearly gives an open cover of A, and thus there exist a finite points fai gN
i D1 (not the same as the above
ones) in A such that
fB1 .ai / j i 2 f1; : : : ; N gg
gives an open cover of A. Let x 2 A be arbitrary. Then there exists m 2 f1; : : : ; N g such that x 2 B1 .am /,
i.e., d.am ; x/ < 1, and so

d.a1 ; x/  d.a1 ; am / C d.am ; x/

53
 m
X1
 d.ai ; ai C1 / C d.am ; x/
i D1
N
X1
 d.ai ; ai C1 / C 1 DW R;
i D1
where R > 0 is a constant independent of x. Therefore, A is bounded. □

Proposition 3.1.7: Closed subset of a compact set is compact.

In a metric space, every closed subset of a compact set is compact.

P ROOF. Let M be a metric space, C  M be compact, and B  C be a closed subset. Let O be an

arbitrary open cover of B. Then, since B c D M nB is open, O [ fB c g is an open cover of M and hence of
C . So there exists a finite subcollection O0  O such that O0 [ fB c g is an open cover of C . However, this
implies that O0 gives a finite cover of B. □

54
 3.2. The Heine–Borel Theorem

One can see that it is very difficult to directly check if a given set is compact because it seems impossible
to list all possible open covers of a given set in the first place, let alone checking if every open cover has a
finite subcover.
However, if M D Rn then compactness may be characterized in terms of those concepts that are familiar
to us:

Theorem 3.2.1: The Heine–Borel Theorem

A subset A  Rn is compact if and only if it is closed and bounded.

The sufficiency ()) is a special case of Proposition 3.1.6, and so it suffices to prove the necessity (().
We break it down to a few lemmas.

Lemma 3.2.2
Let m; n 2 N be arbitrary. If A  Rm is compact and y0 2 Rn , then A  fy0 g  Rm  Rn is
compact.

P ROOF. Let fVi gi 2I a collection of open sets in Rm  Rn that gives an open cover of A  fy0 g. Let us
define, for each i 2 I ,
Ui WD fx 2 Rm j .x; y0 / 2 Vi g :
Then it is open: For every x 2 Ui , one has .x; y0 / 2 Vi and there exists " > 0 such that B" ..x; y0 //  Vi
because Vi is open. But then this implies that B" .x/  fy0 g D .x "; x C "/  fy0 g 2 B" ..x; y0 //  Vi ,
and hence B" .x/  Ui .
Now consider the collection fUi gi 2I of open sets in Rm . Then it is an over cover of A. Indeed, let
a 2 A be arbitrary. Then .a; y0 / 2 A  fy0 g, and so .a; y0 / 2 Uia for some ia 2 I . Then a 2 Uia .
Since A is compact, there exists a finite subcover fUij gjJD1 of A. Then the corresponding fVij gjJD1
gives a finite subcover of A  fy0 g. □

Lemma 3.2.3

Let R > 0 and n 2 N be arbitrary. If Œ R; Rn  Rn is compact, then so is Œ R; RnC1  RnC1 .

P ROOF. Suppose Œ R; Rn is compact, and O be an arbitrary open cover of Œ R; RnC1 . Define

Y WD y 2 Œ R; R j Œ R; Rn  Œ R; y  RnC1 has a finite subcover in O :

˚

Notice that R 2 Y because of the above lemma implies that Œ R; R  f Rg is compact. Our goal is to
show that R 2 Y regardless of the choice of open cover O.
Since Y is bounded above by R, one sees that y0 WD sup Y exists. We would like to show that y0 2 Y .
By the above lemma, Œ R; Rn  fy0 g is compact, and thus there exists a finite subcover O0  O of it.
We shall show that there exists " > 0 such that Œ R; Rn  B" .y0 / is covered by O0 . Since Œ R; Rn  fy0 g
is covered by O0 (whose union is open), for every .x; y0 / 2 Œ R; Rn  fy0 g, there exists "x > 0 such that

55
B"x ..x; y0 // is covered by O0 . Now define

Vx WD B"x .x/  B"x .y0 /  Bp2 "x ..x; y0 //;

which is covered by O0 , and consider the collection

V WD Vx j x 2 Œ R; Rn 1
˚
:

Clearly V is an open cover of Œ R; Rn  fy0 g, and thus by the above lemma, there is a finite subcover
fVxi gN
i D1 . Let " WD min1i N "xi . Then
N
[
n
Œ R; R  B" .y0 /  Vyi ;
i D1

which is covered by O0 .
Now, since y0 D sup Y , there exists y 2 Y such that y0 " < y  y0 . Then, since y 2 Y ,
there exists a finite subcover O00  O of Œ R; Rn  Œ R; y. Hence O0 [ O00 gives a finite cover of
Œ R; Rn  Œ R; y0 C "/. Therefore, y0 2 Y .
Finally, we shall show that y0 D R. First, clearly y0  R because R is an upper bound for Y . So
suppose, on the contrary, that y0 < R. By setting "0 WD 21 minfR y0 ; "g, we have y0 C "0 < R and
y0 C "0 < y0 C ". Thus Œ R; Rn  Œ R; y0 C "0  is covered by O0 [ O00 , and so y0 C "0 2 Y . This
contradicts that y0 D sup Y . Hence y0 D R.
Note that the above argument applies to every choice of O. Therefore, we conclude that Œ R; RnC1 is
compact. □

Lemma 3.2.4
Let R > 0 and n 2 N be arbitrary. Then Œ R; Rn  Rn is compact.

P ROOF. This follows by induction from the compactness of closed interval in R (Proposition 3.1.4) and
the above lemma. □

P ROOF OF T HEOREM 3.2.1. As mentioned above, we need to just prove the necessity (().
Suppose A is closed and bounded. The boundedness implies that there exists R > 0 such that A 
Œ R; Rn . By the above lemma, Œ R; Rn is compact, whereas A is closed. Therefore, by Proposition 3.1.7,
A is compact as well. □

Example 3.2.5: Determine whether the given subset is compact:

(a) .x; y/ 2 R2 j x  0  R2
˚

It is closed but not bounded, and so it is not compact.

(b) B1 .0/ D fx 2 Rn j kxk < 1g  Rn
It is bounded but not closed, and so B1 .0/ is not compact for every n 2 N.
(c) Sn WD x 2 RnC1 j kxk D 1  RnC1 (called the n-dimensional sphere)
˚

It is bounded and closed (why?), and so the n-dimensional sphere Sn is compact for every n 2 N.

56
Exercise 3.4. Let A be a bounded set in Rn with an arbitrary n 2 N. Prove that its closure A is compact. }

57
 3.3. Nested Set Property

We may generalize the Nested Interval Property (NIP; Theorem 1.2.16) in R to metric spaces as follows:

Theorem 3.3.1: Nested Set Property

Let M be a metric space, and fFk g1kD1

be a sequence of nonempty compact sets satisfying Fk 
FkC1 for every k 2 N. Then their intersection contains at least one point, i.e.,
1
\
Fk ¤ ¿:
kD1

T1
P ROOF. By contradiction. Suppose that kD1 Fk D ¿, i.e.,
1
!
\
F1 \ Fk D ¿;
kD2
which implies that !c
1
\ 1
[
F1  Fk D Fkc :
kD2 kD2
So fFk g1
kD2
gives an open cover of F1 . However, since F1 is compact, there exists a finite subcover
c N
fFk gj D1 , where k1  2 and j 7! kj is strictly increasing and so Fkj  Fkj C1 for each j . Thus
j
0 1c
[N \N
F1  Fkcj D @ Fkj A ;
j D1 j D1

and so 0 1
N
\
F1 \ @ Fkj A D ¿:
j D1
However, since F1  Fk1 


 FkN , the intersection on the left is FkN . As a result, we have FkN D ¿,
which is a contradiction. □

58
 Bibliography

[1] S. Abbott. Understanding Analysis. Undergraduate Texts in Mathematics. Springer New York, 2015.
[2] J. E. Marsden and M. J. Hoffman. Elementary Classical Analysis. W. H. Freeman, New York, 1993.
[3] W. Rudin. Principles of Mathematical Analysis. International series in pure and applied mathematics.
McGraw-Hill, 1976.

59