Louis Meunier

Analysis I MATH254

Course Outline:
Fundamentals of set theory. Properties of the reals. Limits, limsup, liminf. Continuity. Functions. Differentiation.
References:
Understanding Analysis, Abbott; Introduction to Real Analysis, Bartle; Analysis I, Tao
Based on Lectures from Fall, 2023 by Prof. Vojkan Jaksic.

Contents

1 Logic, Sets, and Functions 3

1.1 Mathematical Induction & The Naturals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Extensions: Integers, Rationals, Reals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 The Insufficiency of the Rationals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Sets & Set Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.1 Properties of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Reals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 Density of Rationals in Reals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.7 Cardinality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7.1 Power Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2 Sequences 27
2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Properties of Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Limit Superior, Inferior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4 Subsequences and Bolzano-Weirestrass Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 Cauchy Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6 Contractive Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7 Euler’s Number e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.8 Limit Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.9 Properly Divergent Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3 Functional Limits and Continuity 65

3.1 Sequential Characterization of Functional Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Left/Right Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3 Limits and Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
 3.3.1 Infinite Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3.2 Limits at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.3 Infinite Limits at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4.1 Extensions By Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.5 Continuity on Bounded & Closed Interval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.6 Intervals in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.7 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.8 Sequential Characterization of Non-Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.9 Monotone and Inverse Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.10 Continuous Inverse Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4 Differentiation 93
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2 The Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.3 Derivative of the Inverse Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5 Appendix 98
5.1 Interesting Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
 1 Logic, Sets, and Functions

1.1 Mathematical Induction & The Naturals

The natural numbers, N = {1, 2, 3, . . . }, are specified by the 5 Peano Axioms:

(1) 1 ∈ N 1 1
using 0 instead of 1 is also
valid, but we will use 1 here,
and throughout the rest of
(2) every natural number has a successor in N course.

(3) 1 is not the successor of any natural number

(4) if the successor of x is equal to the successor of y, then x is equal to y 2 2

axioms (2)-(4) can be
equivalently stated in terms
of a successor function s(n)
(5) the axiom of induction more rigorously, but won’t
here

The Axiom of Induction (AI), can be stated in a number of ways.

,→ Axiom 1.1: AI.i

Let S ⊆ N with the properties:

(a) 1 ∈ S

(b) if n ∈ S, then n + 1 ∈ S 3

then S = N.
3
(a) is called the inductive
base; (b) the inductive step.
⊛ Example 1.1 All AI restatements are
equivalent in having both of
Prove that, for every n ∈ N, 1 + 2 + · · · + n = n(n+1)
2
(≡ (1)) these, and only differentiate
on their specific values.

Proof (via AI.i). Let S be the subset of N for which (1) holds; thus, our goal is to
show S = N, and we must prove (a) and (b) of AI.i.

• by inspection, 1 ∈ S since 1 = 1(1+1)

2
= 1, proving (a)

• assume n ∈ S; then, 1 + 2 + · · · + n = n(n+1)

2
by definition of S. Adding n + 1

§1.1 Logic, Sets, and Functions: Mathematical Induction & The Naturals p. 3
 to both sides yields:

n(n + 1)
1 + 2 + · · · + n + (n + 1) = + (n + 1) (1)
2
n
= (n + 1)( + 1) (2)
2
(n + 1)(n + 2)
= (3)
2
(n + 1)((n + 1) + 1)
= (4)
2

Line (4) is equivalent to statement (1) (substituting n for n + 1), and thus if n ∈ S,
then n + 1 ∈ S and (b) holds. Thus, by AI.i, S = N and 1 + 2 + · · · + n = n(n+1)
2

holds ∀ n ∈ N. ■

⊛ Example 1.2
h i2
Prove (by induction), that for every n ∈ N, 13 + 23 + · · · + n3 = n(n+1)
2
.

Proof. Follows a similar structure to the previous example. Let S be the subset of N
for which the statement holds. 1 ∈ S by inspection ((a) holds), and we prove (b) by
assuming n ∈ S and showing n + 1 ∈ S (algebraically). Thus, by AI.i, S = N and
the statement holds ∀ n ∈ N. ■

This can also be proven directly (Gauss’ method).

Proof (Gauss’ method). Let A(n) = 1 + 2 + 3 + · · · + n. We can write 2 · A(n) =

1 + 2 + 3 + · · · + n + 1 + 2 + 3 + · · · + n. Rearranging terms (1 with n, 2 with n − 1,
etc.), we can say 2 · A(n) = (n + 1) + (n + 1) + · · · , where (n + 1) is repeated n
times; thus, 2 · A(n) = n(n + 1), and A(n) = n(n+1)
2
. ■

,→ Axiom 1.2: AI.ii

Let S ⊆ N s.t.

(a) m ∈ S

(b) n ∈ S =⇒ n + 1 ∈ S

then {m, m + 1, m + 2, . . . } ⊆ S.

§1.1 Logic, Sets, and Functions: Mathematical Induction & The Naturals p. 4
 ⊛ Example 1.3
Using AI.ii, prove that for n ≥ 2, n2 > n + 1.

Proof. Let S ⊆ N be the set of n for which the statement holds. n = 2 =⇒ 4 > 3,
so the base case holds. Consider n2 > n + 1 for some n ≥ 2. Then, (n + 1)2 =
n2 + 2n + 1 > n + 1 + 2n + 1 = 3n + 2 > 2n + 2 > n + 2, hence S = {2, 3, 4, · · · }
(all n ≥ 2). ■

,→ Axiom 1.3: Principle of Complete Induction, AI.iii

Let S ⊆ N s.t.

(a) 1 ∈ S

(b) if 1, 2, . . . , n − 1 ∈ S, then n ∈ S

then S = N.

Finally, combining AI.ii and AI.iii;

,→ Axiom 1.4: AI.iv

Let S ⊆ N s.t.:

(a) m ∈ S

(b) if m, m + 1, . . . , m + n ∈ S, then m + n + 1 ∈ S

then {m, m + 1, m + 2, . . . } ⊆ S.

,→ Theorem 1.1: Fundamental Theorem of Arithmetic

Every natural number n can be written as a product of one or more primes. 4
4
1 is not a prime number

Proof of theorem 1.1. Let S be the set of all natural numbers that can be written as a product
of one or more primes. We will use AI.iv to show S = {2, 3, . . . }.

• (a) holds; 2 is prime and thus 2 ∈ S

• suppose that 2, 3, . . . , 2 + n ∈ S. Consider 2 + (n + 1):

§1.1 Logic, Sets, and Functions: Mathematical Induction & The Naturals p. 5
 – if 2 + (n + 1) is prime, then 2 + (n + 1) ∈ S, as all primes are products of 1 and
themselves and are thus in S by definition.

– if 2 + (n + 1) is not prime, then it can be written as 2 + (n + 1) = a · b where

a, b ∈ N, and 1 < a < 2 + (n + 1) and 1 < b < 2 + (n + 1). By the definition of
S, a, b ∈ S, and can thus be written as the product of primes. Let a = p1 · · · · · pl
and b = q1 · · · · · qj , where the p’s and q’s are prime and l, j ≥ 1. Then, a · b is
a product of primes, and thus so is 2 + (n + 1). Thus, 2 + (n + 1) ∈ S, and by
AI.iv, S = {2, 3, 4, . . . }

1.2 Extensions: Integers, Rationals, Reals

Consider the set of naturals N = {1, 2, 3, . . . }. Adding 0 to N defines N0 = {0, 1, 2, . . . }.

We define the integers as the set Z = {. . . , −3, −2, −1, 0, 1, 2, 3, . . . }, or the set of all
positive and negative whole numbers.

Within Z, we can define multiplication, addition and subtraction, with the neutrals of
1 and 0, respectively. However, we cannot define division, as we are not guaranteed a
quotient in Z. This necessitates the rationals, Q. We define

p
Q = { : p ∈ Z, q ∈ Z, q ̸= 0}.
q

On Q, we have the familiar operations of multiplication, addition, subtraction and proper-

p
pq ′
ties of associativity, distributivity, etc. We can also define division, as q
p′ = qp′
.
q′

We can also define a relation < between fractions, such that

• x < y and y < z =⇒ x < z

• x < y =⇒ x + z < y + z

Q, together with its operations and relations above, is called an ordered field.

1.2.1 The Insufficiency of the Rationals

We can consider historical reasoning for the extension of Q to R. Consider a right triangle
of legs a, b and hypotenuse c. By the Pythagorean Theorem, a2 + b2 = c2 . Consider further
the case there a = b = 1, and thus c2 = 2. Does c exist in Q?

§1.2 Logic, Sets, and Functions: Extensions: Integers, Rationals, Reals p. 6

 ,→ Proposition 1.1
c2 = 2, c ∈
/ Q.

Proof of proposition 1.1. Suppose c ∈ Q. We can thus write c = pq , where5 p, q ∈ N, and p, q

share no common divisors, ie they are in “simplest form”. Notably, p and q cannot both be
even (under our initial assumption), as they would then share a divisor of 2. We write

p
c=
q
p2
c2 = 2 = 2
q
2q 2 = p2

p ∈ N =⇒ p2 ∈ N, and thus p2 , and therefore6 p, must be divisible by 2 ( =⇒ p even).

Therefore, we can write p = 2p1 , p1 ∈ N, and thus 2q 2 = (2p21 )2 =⇒ q 2 = 2p21 . By the
same reasoning, q must now be even as well, contradicting our initial assumption that p
and q share no common divisors. Thus, c ∈
/ Q. ■
5
Note that in the definition of
Q, p, q are defined to be in Z;
however, as we are using a
geometric argument, we can
1.3 Sets & Set Operations assume c > 0 =⇒
Sign(p) = Sign(q), and we
can just take p, q ∈ N for
• A ∪ B = {x : x ∈ A or x ∈ B} convenience and wlog.
6
√
even = even
• A ∩ B = {x : x ∈ A and x ∈ B}
S∞
• An = {x : x ∈ An for some n ∈ N}
S
i=1 An = n∈N

T∞
•
T
i=1 An = n∈N An = {x : x ∈ An ∀ n ∈ N}

/ A}7
• AC = {x : x ∈ X and x ∈ 7
X is often omitted if it is clear
from context.

,→ Theorem 1.2: De Morgan’s Theorem(s)

Let A, B be sets. Then,

(a) (A ∩ B)C = AC ∪ B C

and
(b) (A ∪ B)C = AC ∩ B C .

§1.3 Logic, Sets, and Functions: Sets & Set Operations p. 7

Proof of theorem 1.2. (b) (A similar argument follows. . . )

,→ Proposition 1.2

∞
!C ∞
\ [
(a) An = AC
n
n=1 n=1
∞
!C ∞
[ \
(b) An = AC
n
n=1 n=1

Proof of proposition 1.2. Consider Proposition (b). Working from the left-hand side, we have

∞
!C
[ [
An = {x : x ∈
/ An }
n=1

= {x : x ∈
/ An ∀ n ∈ N}
\
= {x : x ∈ / An }
\
= ACn

(a) can be logically deduced from this result. Consider the RHS, n . Taking the comple-
AC
S

ment:
C
via (b)
[ \ C
AC
n = AC
n
\
= An

Taking the complement of both sides, we have n = ( An ) , proving (a).

AC C
S T
■

1.4 Functions

,→ Definition 1.1
Let A, B be sets. A function f is a rule assigned to each x ∈ A a corresponding unique
element f (x) ∈ B. We denote
f : A → B.

,→ Definition 1.2
The domain of a function f : A → B, denoted Dom(f ) = A. The range of f , denoted

§1.4 Logic, Sets, and Functions: Functions p. 8

 Ran(f ) = {f (x) : x ∈ A}. Clearly, Ran(f ) ⊆ B, though equality is not necessary.

⊛ Example 1.4
The function f (x) = sin x, f : R → [−1, 1]. Here, Dom(f ) = R, and Ran(f ) =
[−1, 1].

⊛ Example 1.5: Dirichlet Function


1, x ∈ Q

f : R → R, f (x) = . Despite not having a true “explicit” formula, so
0, x ∈

/Q
to speak, this is still a valid function (under modern definitions).

1.4.1 Properties of Functions

,→ Proposition 1.3
Let f : A → B, C ⊆ A, f (C) = {f (x) : x ∈ C}. We claim f (C1 ∪ C2 ) =
f (C1 ) ∪ f (C2 ).

Proof. We will prove this by showing (1) ⊆ and (2) ⊇.

(1) y ∈ f (C1 ∪ C2 ) =⇒ for some x ∈ C1 ∪ C2 , y = f (x). This means that either

for some x ∈ C1 , y = f (x), or for some x ∈ C2 , y = f (x). This implies that either
y ∈ f (C1 ), or y ∈ f (C2 ), and thus y must be in their union, ie y ∈ C1 ∪ C2 .

(2) y ∈ f (C1 ) ∪ f (C2 ) =⇒ y ∈ f (C1 ) or y ∈ f (C2 ). This means that for some
x ∈ C1 , y = f (x), or for some x ∈ C2 , y = f (x). Thus, x must be in C1 ∪ C2 , and for
some x ∈ C1 ∪ C2 , y = f (x) =⇒ y ∈ f (C1 ∪ C2 ).

(1) and (2) together imply that f (C1 ∪ C2 ) = f (C1 ) ∪ f (C2 ). ■

⊛ Example 1.6
Let An = 1, 2, . . . be a sequence of sets. Prove that f ( ∞
S S∞
n=1 An ) = n=1 f (An ).

Proof. Let y ∈ f ( ∞
S∞
n=1 An ). This implies that ∃x ∈ n=1 An s.t. f (x) = y. This
S

implies that x ∈ An for some n, and y ∈ f (An ) for that same “some” n, and thus
y must be in the union of all possible f (An ), ie y ∈ f (An ). This shows ⊆, use
S

similar logic for the reverse. ■

§1.4 Logic, Sets, and Functions: Functions p. 9

 ,→ Proposition 1.4
f (C1 ∩ C2 ) ⊆ f (C1 ) ∩ f (C2 ) 8
8
NB: the reverse is not always
true, ie these sets are not
always equal; “lack” of
Proof. y ∈ f (C1 ∩ C2 ) =⇒ for some x ∈ C1 ∩ C2 , y = f (x). This implies that for some equality is more “common”
than not.
x ∈ C1 , y = f (x) and for some x ∈ C2 , y = f (x). Note that this does not imply that
these x’s are the same, ie this reasoning is not reversible as in the previous union case. This
implies that y ∈ f (C1 ) and y ∈ f (C2 ) =⇒ y ∈ f (C1 ) ∩ f (C2 ). ■

⊛ Example 1.7
T∞ T∞
Prove that if An , n = 1, 2, . . . , f ( n=1 An ) ⊆ n=1 f (An ).

Proof (Sketch). Use the same idea as in example 1.6, but, naturally, with intersec-
tions. ■

⊛ Example 1.8
Take f (x) = sin x, A = R, B = R, and take C1 = [0, 2π], C2 = [2π, 4π]. Then,
f (C1 ) = [−1, 1], and f (C2 ) = [−1, 1]. But C1 ∩ C2 = {2π}; f ({2π}) = {sin 2π} =
{0}, and thus f (C1 ∩ C2 ) = {0}, while f (C1 ) ∩ f (C2 ) = [−1, 1], as shown in
proposition 1.4.

,→ Definition 1.3: Inverse Image of a Set

Let f : A → B and D ⊆ B. The inverse image of D by F is denoted f −1 (D)9 and is
defined as
f −1 (D) = {x ∈ A : f (x) ∈ D}.
9
Note that this is not
equivalent to the typical
⊛ Example 1.9 definition of an inverse
function; f −1 may not exist
A = [0, 2π], B = R, f (x) = sin x, D = [0, 1].

f −1 (D) = {x ∈ A : f (x) ∈ D} = {x ∈ [0, 2π] : sin(x) ∈ [0, 1]} = [0, π].

,→ Proposition 1.5
Given function f and sets D1 , D2 ,

(a) f −1 (D1 ∪ D2 ) = f −1 (D1 ) ∪ f −1 (D2 )

§1.4 Logic, Sets, and Functions: Functions p. 10

 (b) f −1 (D1 ∩ D2 ) = f −1 (D1 ) ∩ f −1 (D2 )10
10
Just see next proposition; if
you really need convincing,
just use 2 rather than ∞ as
,→ Proposition 1.6: ⋆
the upper limit of the
Let An , n = 1, 2, 3 . . . . Then, unions/intersections and use
the same proof.

(a) f −1 ( ∞
S S∞ −1
n=1 An ) = n=1 f (An )

(b) f −1 ( ∞
T T∞ −1
n=1 A n ) = n=1 f (An )

11
Proof.

(a)

∞
[ ∞
[
x ∈ f −1 ( An ) ⇐⇒ f (x) ∈ An
n=1 n=1

⇐⇒ f (x) ∈ An for some n ∈ N

⇐⇒ x ∈ f −1 (An ) for some n ∈ N
∞
[
⇐⇒ x ∈ f −1 (An )
n=1

(b)

∞
\ ∞
\
x ∈ f −1 ( An ) ⇐⇒ f (x) ∈ An
n=1 n=1

⇐⇒ f (x) ∈ An for all n ∈ N

⇐⇒ x ∈ f −1 (An ) for all n ∈ N
∞
f −1 (An )12
\
⇐⇒ x ∈
n=1

■ 12
This is a “proof by
definitions” as I like to call it.
12
Remark 1.1. f : A → B, A1 ⊆ A. Given f (AC C Similar proof can be used to
1 ) and f (A1 ) , there is no general relation
prove proposition 1.5, less
between the two. generally.

For instance, take A = [0, 6π], B = [−1, 2], C = [0, 2π], and f (x) = sin x. Then,
f (C) = [−1, 1], and f (C C ) = f ([−1, 0)) = [−1, 1], but f (C)C = [−1, 1]C = (1, 2], and
f (C C ) ̸= f (C)C ; in fact, these sets are disjoint.

§1.4 Logic, Sets, and Functions: Functions p. 11

 ,→ Proposition 1.7
Let f : A → B and let D ⊆ B. Then f −1 (DC ) = [f −1 (D)]C .

Proof.

f −1 (DC ) = {x : f (x) ∈ DC } = {x : f (x) ∈

/ D}
[f −1 (D)]C = [{x : f (x) ∈ D}]C = {x : x ∈
/ f −1 (D)} = {x : f (x) ∈
/ D}

1.5 Reals

,→ Axiom 1.5: Of Completeness

Any non-empty subset of R that is bound from above has at least one upper bound
(also called the supremum).

In other words; let A ⊆ R and suppose A is bounded from above (A has at a least
upper bound). Then sup(A) exists.

Real numbers, algebraically, have the same properties as the rationals; we have addition,
multiplication, inverse of non-zero real numbers, and we have the relation <. All together,
R is an ordered field.

,→ Definition 1.4
Let A ⊆ R. A number b ∈ R is called an upper bound for A if for any x ∈ A, x ≤ B.

A number l ∈ R is called a lower bound for A if for any x ∈ A, x ≥ l.

,→ Definition 1.5: The Least Upper Bound

Let A ⊆ R. A real number s is called the least upper bound for A if the following
holds:

(a) s is an upper bound for A

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

The least upper bound of a set A is unique, if it exists; if s and s′ are two least upper
bounds, then by (a), s and s′ are upper bound for A, and by (b), s ≤ s′ and s′ ≤ s, and

§1.5 Logic, Sets, and Functions: Reals p. 12

 thus s = s′ .

This least upper bound is called the supremum of A, denoted sup(A).

,→ Definition 1.6: The Greatest Lower Bound

Let A ⊂ R. A number i ∈ R is called the greatest lower bound for A if the following
holds:

(a) i is a lower bound for A

(b) if l is any other lower bound for A, then i ≥ l.

If i exists, it is called the infimum of A and is denoted i = inf(A), and is unique by

the same argument used for sup(A).

,→ Proposition 1.8
Let13 A ⊆ R and let s be an upper bound for A. Then s = sup(A) iff for any ε > 0,
there exists x ∈ A s.t. s − ε < x.
13
Note that this, and
proposition 1.9 that follows,
are not definitions: they are
Proof. We have two statements: restatements, and do
technically require proof.

I. s = sup(A);

II. For any ε > 0, ∃x ∈ A s.t. s − ε < x;

and we desire to show that I ⇐⇒ II.

• I =⇒ II: Let ε > 0. Then, since s = sup(A), s − ε cannot be an upper bound for A
(as s is the least upper bound, and thus s − ε < s cannot be an upper bound at all).
Thus, there exists x ∈ A such that s − ε < x, and thus if I holds, II must hold.

• II =⇒ I: suppose that this does not hold, ie II holds for an upper bound s for A, but
s ̸= sup(A). Then, there exists some upper bound b of A s.t. b < s. Take ε = s − b.
ε > 0, and since II holds, there exists x ∈ A such that s − ε < x. But since s − ε = b
and thus b < x, then b cannot be an upper bound for A, contradicting our initial
condition. So, if II =⇒ I does not hold, we have a “impossibility”, ie a value b which
is an upper bound for A which cannot be an upper bound, and thus II =⇒ I.

§1.5 Logic, Sets, and Functions: Reals p. 13

 ,→ Proposition 1.9: ⋆
Let A ⊆ R and let i be a lower bound for A. Then i = inf(A) ⇐⇒ for every ε > 0
there exists x ∈ A s.t. x < i + ε.14
14
Use similar argument to proof
of previous proposition.
Remark 1.2. ?? 1.5 can also be expressed in terms of infimum. Define −A = {−x : x ∈ A}.
Then, if b is an upper bound for A, then b ≥ x ∀ x ∈ A, then −b ≤ −x ∀ x ∈ A, ie -b is a
lower bound of −A. Similarly, if l is a lower bound for A, −l is an upper bound for −A.

Thus, if A is bounded from above, then

− sup(A) = inf(−A),

and if A is bounded from below,

− inf(A) = sup(−A).

,→ Axiom 1.6: AC (infimum)

Let A ⊆ R; if A bounded from below, inf(A) exists.

,→ Definition 1.7: max, min

Let A ⊆ R. An M ∈ A is called a maximum of A if for any x ∈ A, x ≤ M . M is an
upper bound for A, but also M ∈ A.

If M exists, then M = sup(A); M is an upper bound, and if b any other upper

bound, then b ≥ M , because M ∈ A, and thus M = sup(A).

NB: M = max(A) need not exist, while sup(A) must exist. Consider A = [0, 1);
sup(A) = 1, but there exists no max(A).

The same logic exists for the existence of minimum vs infimum (consider (0, 1),
with no maximum nor minimum).

,→ Theorem 1.3: Nested interval property of R

Let In = [an , bn ] = {x : an ≤ x ≤ bn }, n = 1, 2, 3 . . . be an infinite sequence of
bounded, closed intervals s.t.

I1 ⊇ I2 ⊇ I3 ⊇ . . . In ⊇ In+1 ⊇ . . .

§1.5 Logic, Sets, and Functions: Reals p. 14

 T∞
Then, n=1 In ̸= ∅ (note that this does not hold in Q).

15
Proof. We have In = [an , bn ], In+1 = [an+1 , bn+1 ], . . . . And the inclusion In ⊇ In+1 . 15
Sketch: show that the left-end
points are increasing and the
an ≤ an+1 ≤ bn+1 ≤ bn , ∀ n ≥ 1. So, the sequence an (left-end) is increasing, and the right-end points are
sequence bn (right-end) is decreasing. decreasing. Show either that
all the left-end points are
bounded from above or that
We also have that for any n, k ≥ 1, an ≤ bk . We see this by considering two cases: all the right-end points are
bounded from below. As a
result, there exists a sup/inf
• Case 1: n ≤ k, then an ≤ ak (as an is increasing), and thus an ≤ ak ≤ bk . (depending on which end you
choose) of the set of all the
• Case 2: n > k, then an ≤ bn ≤ bk (again, as bn is decreasing). right/left points. For the sup
case, all upper bounds must
be ≥ sup, and thus the sup is
in all In , and thus in their
Let A = {an : n ∈ N}. Then, A is bounded from above by any bk (as in our inequality we
intersect, and thus the
showed above). Let x = sup(A), which must exist by ?? 1.5. intersect is not empty.

Note that as a result, x ≥ an for all n, and for all k, x ≤ bk , as x is the lowest upper
bound and must be ≤ all other upper bounds, and so for all n ≥ 1, an ≤ x ≤ bn , ie
x ∈ In ∀ n ≥ 1, and thus x ∈ ∞
T∞
n=1 In and so
T
n=1 ̸= ∅. ■

Remark 1.3. The proof above emphasized the left-end points; it can equivalently be proven
via the right-end points, and using y = inf({bn : n ∈ N}) = inf(B), rather than sup(A),
T
and showing that y ∈ In .
T∞
Remark 1.4 (⋆). Note too that, if x = sup(A) and y = inf(B), then x, y ∈ n=1 In ; in fact,
T∞
n=1 In = [x, y]. This can be done by

• Use the main proof to show x ∈

T
In

• Use the previous remark to show y ∈

T
In

• Show x ≤ y =⇒ [x, y] ⊆
T
In

• Show
T
In ⊆ [x, y] =⇒ equality.

Remark 1.5. The intervals In must be closed; if not, eg In = (0, n1 ), then ∞

T
n=1 In = ∅.

Say In ̸= ∅; take then some x ∈ In . Then, x ∈ (0, n1 ) ∀ n ∈ N. But by proposition 1.10,

T T

1
∀ x ∈ R, ∃N ∈ N s.t. N
< x. Clearly, x must be greater than 0 to exist in the intersec-
1
tion; hence, there will always exist some sufficiently large N such that N
< x =⇒ x ∈
/
(1, N1 ) =⇒ x ∈
T T
/ In =⇒ In = ∅.

1.6 Density of Rationals in Reals

§1.6 Logic, Sets, and Functions: Density of Rationals in Reals p. 15

 ,→ Proposition 1.10: Archimedean Property
(a) For any x ∈ R, there exists a natural number n s.t. n > x.

(b) For any y ∈ R satisfying y > 0, ∃n ∈ N such that 1

n
< y.

Remark 1.6. (a) states that N is not a bounded subset of R.

1 1 1
Remark 1.7. (b) follows from (a) by taking x = y
in (a), then ∃n ∈ N s.t. n > y
=⇒ n
< y,
and thus we need only prove (a).

Remark 1.8. Recall that Q is an ordered field (operations +, · and a relation <). Q can be
extended to a larger ordered field with extended definitions of these operations/relations, such
that it contains elements that are larger than any natural numbers (ie, not bounded above).
This is impossible in R due to AC.

Proof. Suppose (a) not true in R, ie N is bounded from above in R. Let α = sup N, which
exists by AC.

Consider α − 1; since α − 1 < α, α − 1 is not an upper bound of N. So, there exists

some n ∈ N s.t. α − 1 < n; then, α < n + 1 where n + 1 ∈ N, and thus α is also not an
upper bound, as there exists a natural number that is greater than α. This contradicts the
assumption that α = sup N, so (a) must be true. ■

,→ Theorem 1.4: Density

Let a, b ∈ R s.t. a < b. Then, ∃x ∈ Q s.t. a < x < b.

Remark 1.9. If you take a ∈ R and ε > 0, then by the theorem, ∃x ∈ Q where x ∈
(a − ε, a + ε). So any real number can be approximated arbitrarily closely (via choose of ε)
by a rational number.

Proof. Since b − a > 0, by (b) of proposition 1.10, ∃n ∈ N s.t. 1

n
< b − a, ie na + 1 < nb.

Let m ∈ Z s.t. m−1 ≤ na < m. Such an integer must exists since

S
m∈Z [m−1, m) = R,
the family [m − 1, m), m ∈ Z makes partitions of R. Then, na < m gives that a < m
n
. On
the other hand, m − 1 ≤ na gives m ≤ na + 1 < nb. So m
n
< b and it follows that m
n

satisfies a < m
n
< b. ■

In the proof, we used the claim:

§1.6 Logic, Sets, and Functions: Density of Rationals in Reals p. 16

 ,→ Proposition 1.11
If z ∈ R, then there exists m ∈ Z s.t. m − 1 ≤ z < m.

Proof. Let S be a non-empty subset of N. Then S has the least element; ∃m ∈ S s.t. m ≤
n, ∀ n ∈ S.

We can assume z ≥ 0; if 0 ≤ z < 1, then we are done (take m = 1), and assume that
z ≥ 1. Let now S = {n ∈ N : z < n}, ̸= ∅ by proposition 1.10, (a). Let m be the least
element of S. It exists by Well-Ordering Property; then, since m ∈ S, z < m. But, we also
have m − 1 ≤ z, otherwise, if z < m − 1 then m − 1 ∈ S and then m is not the least
element of S. Thus, we have m − 1 ≤ z < m, as required. ■

,→ Theorem 1.5
The set J of irrationals is also dense in R. That is, if a, b ∈ R, a < b, ∃ irrational y s.t.
a < y < b (noting that J = R \ Q).

Proof. Fix y0 ∈ J. Consider a − y0 , b − y0 . a − y0 < b − y0 , and by density of rationals,

∃x ∈ Q s.t. a − y0 < x < b − y0 . Then, a < y0 + x < b; let y = x + y0 , and we have
a < y < b.

Note that y cannot be rational; if y ∈ Q, y = x + y0 =⇒ y − x = y0 , and since x ∈ Q,

y − x ∈ Q =⇒ y0 ∈ Q, contradicting the original choice of y0 ∈
/ Q. Thus, y ∈ J. ■

,→ Theorem 1.6
∃ a unique positive real number α s.t. α2 = 2.

Proof. We show both uniqueness, existence:16

Uniqueness: if α2 = 2 and β 2 = 2, α ≥ 0, β ≥ 0, then 0 = α2 −β 2 = (α−β)(α+β) >

0, and so α − β = 0 =⇒ α = β.

• Existence: consider the set A = {x ∈ R : x ≥ 0 and x2 < 2}. A is not empty as

1 ∈ A. The set of A is bounded above by 2, since if x ≥ 2, then x2 ≥ 4 > 2, so x ∈
/ A.
So, by AC, sup A exists; let α = sup A. We will show that α2 = 2, by showing that
both α2 < 2 and α2 > 2 are contradictions.

α2 < 2

§1.6 Logic, Sets, and Functions: Density of Rationals in Reals p. 17

 For any n ∈ N we expand
 2
1 2α 1 2α + 1
α+ = α2 + + 2 ≤ α2 + ,
n n n n

noting that 1
n2
≤ 1
n
for n ≥ 1.
2−α2
Let y = 2α+1
, which is strictly positive. By proposition 1.10, ∃n0 ∈ N s.t.

1 2 − α2 2α + 1
< or < 2 − α2 .
n0 2α + 1 n0

Substituting this n0 into our inequality, we have

 2
1 2α + 1
α+ ≤ α2 + < α2 + 2 − α2 = 2.
n0 n0

Since α + 1
n0
is positive, α + 1
n0
∈ A. But, since α = sup A, α + 1
n0
≤ α, which
is impossible, so α2 < 2 cannot be true.

α2 > 2
Take n ∈ N;  2
1 2α 1 2α
α− = α2 − + 2 > α2 − .
n n n n
α2 −2
Now, let y = 2α
; y > 0, and by proposition 1.10, ∃n0 ∈ N s.t.

1 α2 − 2 2α
< , or < α2 − 2.
n0 2α n0

Substituting this n0 , we have

 2
1 2α
α− > α2 − > α2 + 2 − α2 = 2.
n0 n0
 2
So for any x ∈ A, we have α − n10 > 2 > x2 . α − 1
n0
> 0, and x > 0, since
 2
x ∈ A. Then, α − n0 > x2 gives that α − n10 > x.
1

So, α − 1
n0
> x for all x ∈ A. So α − 1
n0
is an upper bound for A, but since
α = sup A, α − 1
n0
≥ α ie α ≥ α + 1
n0
, which is impossible. So α2 > 2 cannot
be true.

Thus, α2 = 2.

■ 16
Proof sketch: uniqueness is
clear. Existence follows from
showing that α2 cannot be
either < or > 2. This is done
by contradiction, taking some
§1.6 Logic, Sets, and Functions: Density of Rationals in Reals p. 18 number slightly
Remark 1.10. A similar argument gives that for any x ∈ R, x ≥ 0, ∃!α ∈ R, α ≥ 0 such
√
that α2 = x. This x is called the square root of x, denoted α = x.

Remark 1.11. For any natural number m ≥ 2 and x ≥ 0, ∃!α ∈ R, α ≥ 0 s.t. αm = x. The
proof is similar, and we call α the m-th root of x.

Remark 1.12. Our last proof also gives that Q cannot satisfy AC. Suppose it does, ie any set in
Q bounded from above has a supremum ∈ Q. Then, consider B = {x ∈ Q : x ≥ 0 and x2 <
2}; set α = sup B. The exact same proof can be used, but we will not be able to find an upper
bound in Q.

1.7 Cardinality

,→ Definition 1.8
Let f : A → B.

1. f injective (one-to-one) if a1 ̸= a2 =⇒ f (a1 ) ̸= f (a2 )

2. f surjective (onto) if for any b ∈ B∃a ∈ A s.t. f (a) = b.

3. f bijective if both.

,→ Definition 1.9: Composition

If f : A → B, g : B → C, the composite map h = g ◦ f is define by h(x) = g(f (x)).
Note that h : A → C.

⊛ Example 1.10
Consider functions f, g.

1. If f, g injective, so is h = g ◦ f

2. If f, g bijective, then so is h

3. If ∃E ⊆ C, then h−1 (E) = f −1 (g −1 (E))

,→ Definition 1.10
The inverse function17 is defined only for bijective map f : A → B. y ∈ B, f −1 (y) = x
where x ∈ A s.t. f (x) = y.
17
Not the same as the inverse
image of a set by a function,
which is defined for any
§1.7 Logic, Sets, and Functions: Cardinality p. 19 function.
 ⊛ Example 1.11
1. A = R, B = (0, ∞), f (x) = ex . f is a bijection, and f −1 (y) = ln y, y ∈
(0, ∞).

2. A = (− π2 , π2 , B = R). f (x) = tan x, f −1 (y) = arctan y

,→ Definition 1.11: Equal Cardinalities

Let A, B be two sets. We say A, B have the same cardinality, denote A ∼ B if there
exists a bijective function f : A → B.

⊛ Example 1.12
Let E = {2, 4, 6, . . . } (even natural numbers). Define f : N → E by f (n) = 2n.
Thus, f is a bijection, and N ∼ E.18
18
See these independent notes
for more.
,→ Theorem 1.7
The relation ∼ is a relation of equivalence.

1. A ∼ A

2. if A ∼ B, then B ∼ A

3. if A ∼ B and B ∼ C, then A ∼ C

,→ Definition 1.12: Countable

A set A is countable if N ∼ A.

Remark 1.13. According to this, finite sets are not countable; this is just a convention. Some-
times, we say a set is countable if it is finite or to above definition holds, where we say that a
set is countably infinite if it is infinite and countable.

Other times, finite sets are treated separately than countable sets.

,→ Theorem 1.8
Suppose that A ⊆ B.

1. If B is finite or countable, then so is A

§1.7 Logic, Sets, and Functions: Cardinality p. 20

 2. If A is infinite and uncountable, then so is B

,→ Definition 1.13: Cartesian Product

If A, B sets, A × B = {(a, b) : a, b ∈ A, B}.

,→ Proposition 1.12
N × N ∼ N; there exists a bijection f : N × N → N.

,→ Proposition 1.13
Let A be a set. The following are equivalent statements:

(a) A is finite or a countable set;

(b) there exists a surjection from N onto A;

(c) there exists a injection from A into N.

Proof. We proceed by proving that each statement implies the next (and thus are equiva-
lent).

• (a) =⇒ (b): Suppose A is finite and has N elements. Then there exists a bijection
h : {1, 2, . . . n} → A. We now define a map f : N → A, by setting

h(m) if m ≤ n

f (m) = .
h(n) if m > n


f is surjective, and thus (b) holds. If (a) countable, ∃ bijection f : N → A, and any
bijection is a surjection, so (b) also holds.

• (b) =⇒ (c): Let h : N → A be a surjection, whose existence is guaranteed by (b).

Then, for any a ∈ A, the set

h−1 ({a}) = {m ∈ N : h(m) = n} =

̸ ∅,

since h is a surjection. Then, by the well-ordering property of N, the set h−1 ({a})
has a least element.
If n is the least element of h−1 ({a}), we set f (a) =. This defines a function

f : A → N,

§1.7 Logic, Sets, and Functions: Cardinality p. 21

 and we aim to show that f is injective, ie that f (a1 ) = f (a2 ) =⇒ a1 = a2 .
Suppose f (a1 ) = f (a2 ) = n. Then, n is the least element of h−1 ({a1 }) and of
h−1 ({a2 }), and in particular, h(n) = a1 and h(n) = a2 , and thus a1 = a2 and so
f is indeed injective.

• (c) =⇒ (a): Let f : A → N be an injection, whose existence is guaranteed by (c).

Consider the range of f , ie

f (A) = {f (a) : a ∈ A}.

Since f an injection, f is a bijection between A and f (A).

Otoh, f (A) ⊆ N, and so by theorem 1.8, f (A) is either finite or countable, and there
exists a bijection between A and some set that is either fininte or countable. Thus, A
must also be finite or countable, and so (a) holds.

,→ Theorem 1.9
Let An , n = 1, 2, . . . be a sequence of sets such that each An is either finite or count-
able. Then, their union
∞
[
A= An
n=1

is also either finite or countable.

Proof. We will use (a) ⇐⇒ (b) from proposition 1.13 to prove this.

Since each An finite or countable, by (a) =⇒ (b), there exists a surjection

φn : N → An .

Now, let h : N × N → A, (the union) by setting

h(n, m) = φn (m).

We aim to show that h is also surjective.

If a ∈ ∞n=1 An , then a ∈ An for some n ∈ N. Since φn : N → An is a surjection, there
S

exists an m ∈ N s.t. φn (m) = a. By definition of h, we have

h(n, m) = a,

§1.7 Logic, Sets, and Functions: Cardinality p. 22

and thus h is a surjection.

By proposition 1.12, there exists a bijection f : N → N × N, and we can define the

composite map
h ◦ f : N → A (= ∪∞
n=1 An ),

which is a surjection as both h, f are surjections. So, there exists a surjection from N → A,
and by proposition 1.13, (b) =⇒ (a), and thus A = ∞ n=1 An is also finite or countable.
S

■
S∞
Remark 1.14. If A = n=1 An , where each An is either finite or countable, and at least one
An is countable, then A is countable.

Remark 1.15. If A1 , . . . , An are finitely many finite or countable sets then their union A1 ∪
· · · ∪ An is also finite or countable (essentially just previous proof where we use n instead of
∞ for the upper limit of the union…).

,→ Theorem 1.10
The set Q of rational numbers is countable.

Proof. We write
Q = A0 ∪ A1 ∪ A2 ,

where A0 = {0}, A1 = { m
n
: m, n ∈ N}, and A2 = {− m
n
: m, n ∈ N}.
Let us show that A1 is countable; define

m
h : N × N → A1 , f (m, n) = .
n

h is clearly a surjection; if f : N → N × N is a bijection, then by proposition 1.12, h ◦ f :

N → A1 is a surjection. By proposition 1.13, A1 is countable.
We prove that A2 countable in essentially the same way.
Then, A0 ∪ A1 ∪ A2 is also countable, as it is the union of countable sets, and thus Q is also
countable. ■

§1.7 Logic, Sets, and Functions: Cardinality p. 23

 ,→ Theorem 1.11
The set R of real numbers is uncountable.19
19
Proof sketch: by
contradiction. Assume that a
bijection exists, and show
Proof. We will argue by contradiction; suppose R is countable, then show that the nested that it cannot be a surjection
by the previous props/thms.
interval property (theorem 1.3) of the real line fails. Specifically, carefully
construct nested intervals In ,
Let f : N → R be a bijection, setting f (1) = x1 , f (2) = x2 , . . . , f (n) = xn , . . . ; we can
for which xi ∈ / Ii , and then
then list the elements of R as R = {x1 , x2 , x3 , . . . , xn , . . . }. show that the intersection of
all these intervals is empty,
We can now construct a sequence In , n ∈ N of bounded, closed intervals, such that I1 does contradicting the nested
interval property of the real
not contain x1 .
line.
If x2 ∈
/ I1 , then I2 = I1 . If x2 ∈ I1 , then divide I1 into four equal closed intervals. See pg. 25 of Abbott’s
Analysis for a more concise
Call the leftmost/rightmost of these intervals I1′ and I1′′ respectively. We know that x2 ∈ I1 , proof in the same language.
so we must have that either x2 ∈
/ I1′ or x2 ∈
/ I1′′ If x2 ∈
/ I1′ , then I2 = I1′ . If x2 ∈
/ I1′′ , then
I2 = I1′′ .
Thus, we have constructed I1 , I2 s.t.

I1 ⊇ I2 and x1 ∈
/ I1 , x2 ∈
/ I2 .

Consider x3 ; if x3 ∈
/ I2 , then I3 = I2 . If x3 ∈ I2 , we repeat the “dividing” process as before.
/ I2′′ . If x3 ∈
/ I2′ or x3 ∈
Since x3 ∈ I2 , either x3 ∈ / I2′′ , I3 = I2′′ .
/ I2′ , I3 = I2′ . Else, if x3 ∈
We have now that
I1 ⊇ I2 ⊇ I3 and x1 ∈
/ I1 , x2 ∈
/ I2 , x3 ∈
/ I3 ,

and we can continue this construction to obtain an infinite sequence of bounded, closed
intervals In s.t.
I1 ⊇ I2 ⊇ · · · ⊇ In ⊇ In+1 ⊇ · · · ,

and for each n, xn ∈

/ In .
Consider the intersection of all these In ’s,

∞
\
In .
n=1

T∞
For every m, xm ∈
/ Im , so for every m ∈ N, xm ∈
/ n=1 In , and so R = {x1 , x2 , . . . xm , . . . }
has an empty intersection with this intersection, ie

∞
!
\
R∩ In = ∅.
n=1

T∞ T∞
Otoh, n=1 In ⊆ R, so we must have that n=1 In = ∅ contradicting the nested interval

§1.7 Logic, Sets, and Functions: Cardinality p. 24

property of the real line which states that this intersection must not be empty. We thus
have a contradiction, and our assumption that R countable fails. 20 ■
20
Note that theorem 1.3 is built
upon the Axiom of
Completeness, a “fact” of R
,→ Proposition 1.14
(what makes it “distinct” from
The set J of all irrational numbers in R is uncountable. Q, N, etc). Thus, we are really
just using AC, with some
abstractions sts.

Proof. We have that R = Q ∪ J. If J countable, then R would also be countable as the

union of two countable sets (as we showed Q countable in theorem 1.10). R uncountable,
so J is also uncountable. ■

,→ Proposition 1.15
The set (−1, 1) ⊆ R is uncountable.

S∞
Proof. We can write R = n=1 (−n, n). If each (−n, n) is countable, then R would also
be countable, as a countable union of countable sets. Thus, there must exist some n0 ∈
N s.t. (−n0 , n0 ) is not countable. The map

x
f : (−n0 , n0 ) → (−1, 1), f (x) =
n0

is a bijection, and so (−1, 1) is uncountable. ■

⊛ Example 1.13
Show that the map
x
f (x) =
1 − x2
is a bijection between (−1, 1) and R ie (−1, 1) ∼ R.

Proof. Surjection is fairly trivial (if stuck, consider the graph of the function).
Injection; given f (x) = f (y) where x, y ∈ (−1, 1),

x y
=
1 − x2 1 − y2
x − xy 2 = y − yx2
x − y = xy 2 − yx2 = xy(y − x)
x − y = −xy(x − y)
=⇒ −xy = 1 =⇒ xy = −1, or x − y = 0

xy = −1 is impossible given the domain of the function, hence x − y = 0 =⇒

§1.7 Logic, Sets, and Functions: Cardinality p. 25

 x = y, as desired. ■

,→ Proposition 1.16
Any bounded non-empty open interval (a, b) ∈ R is uncountable.

Proof. We will construct a bijection f : (a, b) → R so that (a, b) ∼ R. Since R is uncount-

able, so must (a, b).

The map
2(x − a)
f (x) = −1
b−a
is a bijection between (a, b) and (−1, 1), and we have shown that (−1, 1) ∼ R, so (a, b) ∼
R, and thus any open interval has the same cardinality as R. ■

⊛ Example 1.14
Prove that ∃ bijection between [0, 1) and (0, 1), and conclude that [0, 1) ∼ (0, 1) ∼
R. Then conclude for any a < b, [a, b) ∼ R.

1.7.1 Power Sets

,→ Definition 1.14: Power Set

Let A be a set. The power set of Am denoted P(A) is the collection of all subsets of A.

Generally, if A finite of size n, P(A) has 2n elements.

,→ Theorem 1.12: Cantor Power Set Theorem

Let A be any set. Then there exists no surjection from A onto P(A). 21
21
Certified Classic

Proof. Suppose that there exists a surjection,

f : A → P(A).

Let D ⊆ A defined as
D = {a ∈ A : a ∈
/ f (a)}.

Since D ⊆ P(A), and f is surjective, there must exist some a0 ∈ A s.t. f (a0 ) = D.
We have two cases:

§1.7 Logic, Sets, and Functions: Cardinality p. 26

 1. a0 ∈ D. But then, by definition of D, a0 ∈
/ f (a0 ) = D, so a0 ∈ D is not possible as it
implies a0 ∈
/ D.

2. a0 ∈
/ D. But then, since D = f (a0 ), a0 ∈
/ f (a0 ), and so by definition of D, a0 ∈ D,
which is again not possible.

So, the assumption of a surjection existing has led to a0 ∈ A such that neither a0 ∈ D nor
/ D, which is impossible. Thus there can be no surjective f .
a0 ∈
Notice, though, that there exists an injection A → P(A), a 7→ {a}, and thus there is an
injection but no bijection.
Thus, we can say that P(A) is strictly bigger than A.
■

2 Sequences

2.1 Definitions

,→ Definition 2.1
Let A be a set. An A-valued sequence indexed by R is a map

x : N → A.

The value x(n) is called the n-th element of the sequence. One writes x(n) = xn , or
lists its elements

{x1 , x2 , x3 , . . . } ≡ {xn }n∈N ≡ (xn )n∈N ≡ {xn }.

,→ Definition 2.2: Convergence

We say that a sequence (xn ) converges to a real number x if for every ε > 0, ∃N ∈
N s.t. for all n ≥ N we have
|xn − x| < ε.

If sequence (xn ) converges to x, we write limn→∞ xn = x.

⊛ Example 2.1
Let (xn ) be a sequence defined by xn = n1 , n ∈ N, then limn→∞ xn = 0.

§2.1 Sequences: Definitions p. 27

 Proof. Let ε > 0. Let N ∈ N s.t. N > 1ε . Then for n ≥ N , we have that

1 1
0< ≤ < ε.
n N

So, for n ≥ N, |xn − 0| < ε, and so the limit is 0. ■

,→ Definition 2.3: Quantifier of Limit ⋆

The limit can be written in terms of quantifiers.

lim xn = x
n→∞

means that
( ∀ ε > 0)(∃N ∈ N)( ∀ n ≥ N )(|xn − x| < ε).

⊛ Example 2.2
Prove22 that
n2 + 1
lim = 1.
n→∞ n2

Proof. Let ε > 0. Let N be a natural number such that N > √1 .

ε
Then, for n ≥ N ,

n2 + 1 n2 + 1 − n2 1 1
| 2
− 1| = | 2
| = 2 ≤ 2 < ε.
n n n N

■
22
Work backwards to start; how
can you simply the sequence
(that is, build a string of
,→ Definition 2.4: Divergent Sequences
inequalities) such that
If a sequence (xn ) does not converge to any real number x, we say that the sequence defining an N in terms of ε
becomes apparent?
is divergent. For instance, consider

xn = (−1)n , n ≥ 1.

The sequence alternates between 1 and −1 and so intuitively does not converge. How
do we prove it?

Proof. By contradiction; suppose that xn = (−1)n be a converging sequence. Let x =

limn→∞ xn . Take ε = 1, then ∃N ∈ N s.t. for all n ≥ N we have that |x − xn | < ε = 1.

§2.1 Sequences: Definitions p. 28

Consider indices n = N, n = N + 1. We have

|xN +1 − xN | = |xn+1 − x + x − xN | ≤ |xN +1 − x| + |x − xN | < 1 + 1 = 2.

| {z }
triangle inequality

But we also have that

|(−1)N +1 − (−1)N | = |(−1)N +1 + (−1)N +1 | = 2,

We thus have that 2 < 2, which is a contradiction. Thus, xn is not convergent. ■

⊛ Example 2.3
Evaluate the following examples using the ε definition:

1. limn→∞ sin
√
3n
n
=0

2. limn→∞ n!
nn
=0

(1+2+···+n)2
3. limn→∞ n4
= 1
4

Proof. 1. For all ε > 0; take 1

N
< ε3 =⇒ √
3
1
N
< ε. Then, ∀ n ≥ N ,

√ √
3 1 1
3
n ≥ N =⇒ √
n ≥ N =⇒ ≤ √3
3
n N
sin n 1 1
−1 ≤ sin n ≤ 1 =⇒ |sin n| ≤ 1 =⇒ √ ≤ √3
≤ √ 3
<ε
3
n N N
sin n
=⇒ lim √ =0
n→∞ 3 n

2. Take 1
N
≤ ε. Then, ∀ ε > 0, ∀ n ≥ N =⇒ 1
n
≤ 1
N
,

n! n! n! n(n − 1)(n − 2) · · · 1 n n−1 n−2 1

n
> 0 =⇒ n
= n = = · · ···
n n n n · n···n n n n n
1
≤ 1 · 1···1 ·
n
1 1
≤ ≤ <ε
n N
n!
=⇒ lim n = 0
n→∞ n

3. Note first that (1 + 2 + · · · + n)2 = ( n(n+1)

2
)2 (see example 1.1). Take 1
N
< 2ε ;

§2.1 Sequences: Definitions p. 29

 then, ∀ ε > 0, we have that ∀ n ≥ N ,

n2 (n+1)2
(1 + 2 + · · · + n)2 1 4 n4 n4 + 2n3 + n2 − n4
− = − =
n4 4 n4 n4 n4
2n3 + n2 2n + 1 2n 2 2
= 4
= 2
≤ 2 ≤ ≤ <ε
n n n n N
(1 + 2 + · · · + n)2 1
=⇒ lim =
n→∞ n4 4

2.2 Properties of Limits

,→ Lemma 2.1: Triangle Inequality

For x, y, z ∈ R,

(i) |x + y| ≤ |x| + |y|; (ii) |x − y| ≤ |x − z| + |z − y|23

23
Generally, proofs involving
limits will consist of 1)
picking/defining an ε based

x + y x+y ≥0 on given limit/series

Sketch proof. (i): |x + y| = . So if x + y ≥ 0, |x + y| = x + y ≤ definitions, and then 2) using
−(x + y) x + y ≤ 0

triangle inequality/related
techniques to reach the
|x| + |y|.
desired conclusion.
If x + y > 0, |x + y| = −(x + y) = (−x) + (−y) ≤ |x| + |y|.

(ii): |x − y| = |x − z + z − y| ≤ |x − z| + |z − y| (using (i)). ■

,→ Theorem 2.1: ⋆
A limit of a sequence is unique. In other words, if the sequence is converging, then
its limit is unique. The sequence cannot converge to two distinct numbers x and y.24
24
Proof sketch: contradiction,
assume two distinct limits,
and take ε as their midpoint.
Proof. By contradiction; suppose ∃(xn ) s.t. limn→∞ xn = x and limn→∞ xn = y, and that Arrive at a contradiction by
using triangle inequalities to
x ̸= y. show that |x − y| < |x − y|,
|x−y| and thus the limits cannot be
Take ε = . Since x ̸= y, we have that ε > 0. Since limn→∞ xn = x, ∃N1 ∈ N s.t. for
2 distinct.
n ≥ N1 , |xn − x| < ε.
Similarly, since lim xn = y, ∃N2 ∈ N s.t for g ≥ N2 , |xn − y| < ε.

§2.2 Sequences: Properties of Limits p. 30

Take some n ≥ max(N1 , N2 ); then

|x − y| = |x − xn + xn − y| ≤ |x − xn | + |xn − y|
< ε + ε = |x − y|
=⇒ |x − y| < |x − y|, ⊥

,→ Theorem 2.2
Any converging sequence is bounded.25
In other words, if (xn ) is a converging sequence,

∃M > 0 s.t. |xn | ≤ M ∀ n ≥ 1.

25
Take ε = 1, which is greater
than |xn − x| by limit
definition for n ≥ N for some
Proof. Let (xn ) be a converging sequence, and x = limn→∞ xn . Take ε = 1 in the definition N . We then use this to show
that |xn | < 1 + |x|, then
of the limit; then, ∃N ∈ N s.t. ∀ n ≥ N , |xn − x| < 1. construct a summation M
such that it bounds |xn |; it is
This gives that for n ≥ N , |xn | = |xn − x + x| ≤ |xn − x| + |x| < 1 + |x|.
equal to |x1 | + |x2 | + · · · up
Let now M = |x1 | + |x2 | + · · · + |xN −1 | + (1 + |x|). Then, for any n ≥ 1, |xn | ≤ M ; to |xN −1 |, then plus 1 + |x|.
We have finished.
If n ≤ N − 1, then |xn | is a summand in M , and thus |xn | ≤ M .
If n ≥ N , then we have by the choice of N that |xn | < 1 + |x| ≤ M .
Thus, for all n ≥ 1, |xn | ≤ M , and is thus bounded given (xn ) converges. ■

,→ Proposition 2.1: Algebraic Properties of Limits

Let (xn ), (yn ) be sequences such that26

lim xn = x, lim yn = y.

Then:

1. For any constant c, lim c · xn = c · lim xn = c · x

2. lim(xn + yn ) = lim xn + lim yn = x + y

3. lim xn · yn = (lim xn )(lim yn ) = x · y

4. Suppose y ̸= 0, yn ̸= 0 ∀ n ≥ 1. Then, lim xynn = lim xn

lim yn
= x
y
26
Note that the contrary of
these statements need not
hold; ie, if lim(xn · yn ) exists,
this does not imply the
existence of lim xn and
§2.2 Sequences: Properties of Limits p. 31 lim yn . Consider example 2.4
Remark 2.1. Let X be the collection of all sequences of real numbers, X = {(xn ) : xn is a sequence}.
If (xn ) ∈ X and c ∈ R, we can define c · (xn ) = (c · xn )27 ; this defines scalar multiplication
on X.
We can also define addition; if (xn ) and (yn ) are two sequences in X, then (xn ) + (yn ) =
(xn + yn ). Then, with these two operations X is a vector space.
27
NB: this denotes c multiplying
to each nth element in xn , ie
⊛ Example 2.4 c · x1 , c · x2 , etc

Take xn = (−1)n , yn = (−1)n+1 , n ≥ 1.

(xn ) + (yn ) = 0, xn · yn = −1, and so lim xn + yn = 0, lim xn · yn = −1, while
neither lim xn nor lim yn exist.

Proof (part 3. of proposition 2.1). Take28 lim xn = x, lim yn = y. Since (xn ) is converging, it
is bound by theorem 2.2, and there exists M > 0 s.t. ∀ n ≥ 1, |xn | ≤ M .
Now,

|xn yn − xy| = |xn yn − xn y + xn y − xy|

≤ |xn yn − xn y| + |xn y − xy|
= |xn | · |yn − y| + |y| · |xn − x|
≤ M · |yn − y| + |y| · |xn − x| (i)

Let ε > 0; since lim yn = y, there exists N1 ∈ N s.t. n ≥ N1 , |yn − y| < ε

2M
. Sim, since
lim xn = x, ∃N2 ∈ N s.t. |xn − x| < ε
2(|y|+1)

Let N = max(N1 , N2 ), n ≥ N . Then, we have, with (i),

(i) |xn yn − xy| ≤ M · |yn − y| + |y| · |xn | − x

ε ε
<M· + |y| ·
2M 2(|y| + 1)
ε ε
≤ + .
2 2

Thus, for n ≥ N , |xn yn − xy| < ε, and by definition of the limit, lim xn yn = xy. ■
28
Proof sketch: take an upper
bound of xn . Then, show that
|xn yn − xy| < ε, by using
,→ Theorem 2.3: Order Properties of Limits
triangle inequalities to show
Let (xn ), (yn ) be two sequences such that inequality to a combination of
M , arbitrarily small values
(by def of limits of xn , yn
lim xn = x, lim yn = y. resp,), and |y|.

1. xn ≥ 0 ∀ n =⇒ x ≥ 0.

§2.2 Sequences: Properties of Limits p. 32

 2. xn ≥ yn ∀ n =⇒ x ≥ y.

3. c is constant since c ≤ xn ∀ n ≥ 1 =⇒ c ≤ x. xn ≤ c ∀ n ≥ 1 =⇒ xn ≤ c.

Remark 2.2. 2., 3. follow from 1. Set zn = xn − yn ∀ n ≥ 1. Then, zn ≥ 0 ∀ b ≥ 1,

lim zn = lim(xn − yn ) = lim xn − lim yn (as these limits exist) = x − y. By 1., lim zn ≥ 0,
and so either x − y ≥ 0 or x ≥ y.

Proof of 1. Suppose 1. does not hold; suppose ∃(xn ) s.t. lim xn = x, xn ≥ 0 ∀ ≥, but
x < 0.
Let ε > 0 s.t. x < −2ε < 0. With this ε, lim xn = x gives that ∃N ∈ N s.t. ∀ n ≥
N, |xn − x| < ε, or particularly, xn − x < ε.
Then, xn < ε + x, and since x < −2ε, we have ∀ n ≥ N , xn < −ε, and thus ∀ n ≥ N ,
xn < 0, a contradiction. ■

,→ Theorem 2.4: The Squeeze Theorem

Let (xn ), (yn ), (zn ) be sequences such that xn ≤ yn ≤ zn , ∀ n ≥ 1, and limn→∞ xn =
limn→∞ zn = ℓ, then limn→∞ yn = ℓ.29
29
Sketch: This follows a similar
technique to many that
follow. Use the definitions of
Proof. Let ε > 0. Since lim xn = ℓ, there ∃N1 ∈ N s.t. ∀ n ≥ N1 , |xn − ℓ| < ε. the limits of xn , zn to take an
arbitrary ε, and an N for each
Since lim zn = ℓ, there ∃N2 ∈ N s.t. ∀ n ≥ N2 , |zn − ℓ| < ε. respective limit. Take the max
of these N ’s, and show that
Take N = max{N1 , N2 } and take n ≥ N . Then,
for all n ≥ max Ni , you can
show that f yn − l is less than
ε and greater than −ε. Really,
yn ≤ zn =⇒ yn − ℓ ≤ zn − ℓ ≤ |zn − ℓ| < ε, this is just a proof of applying
definitions correctly.
since n ≥ max{N1 , N2 } =⇒ n ≥ N2 .
Now, we have that
yn ≥ xn =⇒ yn − ℓ ≥ xn − ℓ > −ε,

since |xn − ℓ| < ε for n ≥ N1 , and our n is ≥ max{N1 , N2 }. Thus, for n ≥ N ,

−ε < yn − ℓ < ε =⇒ |yn − ℓ| < ε,

and thus lim yn = ℓ, by definition. ■

,→ Definition 2.5: Increasing/Decreasing

A sequence (xn ) is called increasing if xn+1 ≥ xn ∀ n ∈ N, and is decreasing if xn+1 ≤
xn ∀ n ∈ N.

§2.2 Sequences: Properties of Limits p. 33

 ,→ Definition 2.6: Bounded from above/below
A sequence (xn ) is called bounded from above if there exists some M ∈ R s.t. xn ≤
M ∀ n ≥ 1.
Sequence (xn ) is bounded from below if there exists some M ∈ R s.t. xn ≥ M ∀ n ≥
1.

,→ Theorem 2.5: Monotone Convergence Theorem

The following relate to bounded above/below and increasing/decreasing sequences:30

1. Let (xn ) be an increasing sequence that is bounded from above. Then (xn ) is
converging.

2. Let (xn ) be a decreasing sequence that is bounded from below. then (xn ) is
converging.
30
Sketch: 1,2 are proven very
similarly. For 1., take the set
of all xn in the given
Proof (of 1). Let A = {xn : n ≥ 1}. Since (xn ) is bounded above by M , the set A is bounded sequence. Since the sequence
is bounded, then so is the set,
from above. Let α = sup A, which exists by AC. and so we can take its
Let ε > 0. Since α is the least upper bound for A, α − ε is not an upper bound of A supremum. Use the ε
definition of sup to show that
(α − ε < α). Hence, there must exist some N ∈ N such that α − ε < xN (if it didn’t exist, this supremum is also the
limit of the sequence
then α wouldn’t be the supremum . . . ). Then, for n ≥ N , and since (xn ) increasing, (basically, a bunch of
inequalities, and being careful
with definitions). 2. follows
α − ε < xN ≤ xn ≤ α. identically but using the
infimum.

Then, for all n ≥ N ,

α − ε < xn ≤ α =⇒ −ε < xn − α ≤ 0,

and so |xn − α| < ε for n ≥ N . By definition, α = lim xn . ■

⊛ Example 2.5
A sequence (xn ) is called eventually increasing if there exists some N0 ∈ N s.t. ∀ n ≥
N0 , xn+1 ≥ xn . If (xn ) is eventually increasing and bounded from above, lim xn = α
exists.

Proof. (Sketch) If (xn ) eventually increasing, say ∀ n ≥ N0 , and bounded above,

then if we consider x′n as the sequence of xn where n′ = n − N0 , it must converge
by Monotone Convergence Theorem. Hence, taking N = N0 , xn too must converge.

§2.2 Sequences: Properties of Limits p. 34

 ■

⊛ Example 2.6
√ √
Let (xn ) be a sequence defined recursively by x1 = 2 and xn+1 = 2 + xn , n ≥ 1.
√ √
p q p
So x2 = 2 + 2, x3 = 2 + 2 + 2 · · · , xn = 2 cos 2n+1 π
, n ≥ 1. Show that
lim xn = 2.

Proof. We will prove this using the Monotone Convergence Thm by showing that
the xn is bounded from above and increasing, which implies that the limit exists.
We will then find the actual limit.
Recall that n ≥ 1, xn ≤ 2. We will prove this by induction. Let S ⊆ N be the set of
√
indices such that xn ≤ 2. Since x1 = 2 < 2, 1 ∈ S. Now suppose some n ∈ S, ie
√ √
xn ≤ 2. Then, we have that xn+1 = 2 + xn ≤ 2 + 2 = 2 =⇒ xn+1 ≤ 2. Thus,
by induction, n ∈ S =⇒ n + 1 ∈ S =⇒ S = N, ie xn ≤ 2 ∀ n ∈ N. Thus, our
sequence is bounded from above.
We now prove that (xn ) is increasing. Let S ⊆ N s.t. n ∈ S ⇐⇒ xn+1 ≤ xn .
p √ √
x2 = 2 + 2 ≥ 2 = x1 =⇒ x1 ≤ x2 =⇒ 1 ∈ S. Suppose n ∈ S =⇒
√ √
xn+1 ≥ xn . Then, xn+2 = 2 + xn+1 ≥ 2 + xn = xn+1 =⇒ n + 1 ∈ S. Thus,
S = N, so xn+1 ≥ xn ∀ n ∈ N.
So the sequence (xn ) is increasing and bounded from above, and thus ∃ lim xn = α.
√
To find the value of α, consider xn+1 = 2 + xn , or x2n+1 = 2 + xn . We can
also write that α = lim xn = lim xn+1 .31 We then have that lim xn+1 = α =⇒
lim x2n+1 = α2 , and thus x2n+1 = 2 + xn =⇒ lim x2n+1 = lim(2 + xn ) =⇒ α2 =
2 + α =⇒ α = 2, −1. xn ≥ 0 ∀ n, by Order Limit Theorem, and so α ≥ 0 and thus
α = 2. ■
31
Add proof

,→ Corollary 2.1
√
For a, b > 0, then 21 (a + b) ≥ ab

1 2 √
Proof. 2
(a + b) = 14 (a2 + 2ab + b2 ) ≥ ab =⇒ 1
2
(a + b) ≥ ab ■

⊛ Example 2.7
 
Let (xn ) be defined recursively by x1 = 2 and xn+1 = 1
2
xn + 2
xn
for n ≥ 1. Then,
√
(xn ) is converging and lim xn = 2.

Proof. We32 will show that (xn ) bounded from below and decreasing, implying the

§2.2 Sequences: Properties of Limits p. 35

 √ √
limit exists. We will show that for all n, xn ≥ 2. For n = 1, 2 ≥ 2. For n > 1,
√
we will use corollary 2.1. We then have that xn+1 = 21 (xn + x2n ) ≥ · · · ≥ 2 =⇒
√
xn ≥ 2 ∀ n ≥ 1, ie, it is bounded from below.
We will now show that the sequence is decreasing.

1 2 1 1 1 1 √ 2
xn − xn+1 = xn − (xn + ) = xn − = (x2n − 2) ≥ ( 2 − 2) ≥ 0,
2 xn 2 xn 2xn 2xn

where the second-to-last inequality holds from the lower bound we found on xn .
Having shown that xn decreasing and is bounded from below, we conclude that it
converges by Monotone Convergence Theorem. To find its limit, let L := lim xn .
Then,  
1 2 1 1
lim xn = lim xn−1 + = lim xn−1 + lim ,
2 xn−1 2 xn−1
and since the limit of xn is equal to the limit of xn−1 , we have that

1 1 √
L= L+ =⇒ L2 = 2 =⇒ L = ± 2,
2 L
√
but we know that xn ≥ 2 hence we can ignore the negative root, and thus xn
√
converges to 2. ■
32
This example, as well as the
more general one after it, rely
⊛ Example 2.8: ⋆ on applying 1) the monotone
convergence theorem, then 2)
Let a > 0 and let (xn ) be a sequence defined recursively by x1 is arbitrary (positive), using Algebraic Limit
Properties to turn the
and problem into an algebraic
1 a problem, using the given
xn+1 = (xn + ), n ≥ 1.
2 xn recursive relation.
√
Show that limn→∞ xn = a.

√
Proof. By corollary 2.1, xn+1 = 12 (xn + xan ) ≥ a
a, hence, xn is bounded
p
xn · xn
=
√
from below by a.
We also have that xn − xn+1 = xn − 12 xn − 2xan = x2n − 2xan = x1n (x2n − a). We
√
have that xn ≥ a =⇒ x2n ≥ a =⇒ x2n − a ≥ 0. Further, since the sequence
√
is bounded from below by a > 0( ⇐= a > 0), then x1n > 0 as well. Hence,
1
xn
(x2n − a) ≥ 0, and thus xn − xn+1 ≥ 0 =⇒ xn ≥ xn+1 and thus xn is decreasing.
Thus, by the Monotone Convergence Theorem, xn is convergent. Let X := limn→∞ xn .
 
We have from the recursive definition, lim xn = lim 2 (xn + xn ) . Since we know
1 a

§2.2 Sequences: Properties of Limits p. 36

 xn convergent, we can “split up” this limit using algebraic properties, hence

1 a 1 a 1
lim xn = lim xn + lim = lim xn + lim
2 2xn 2 2 xn
1 a
=⇒ X = X +
2 2X
X a √
=⇒ = =⇒ X 2 = a =⇒ X = a,
2 2X

which completes the proof. ■

⊛ Example 2.9
Evaluate33 the limit of xn given the recursive relation xn+1 = 1
4−xn
, x1 = 3.

Proof. We aim to show that (xn ) is bounded from below and decreasing.
Bounded from below: we claim xn > 0; we proceed by induction. x1 = 3 > 0
holds; say xn > 0 for some n ≥ 1. Then, we have

1 1 1
xn > 0 =⇒ −xn < 0 =⇒ 4−xn < 4 =⇒ > > 0 =⇒ xn+1 = > 0,
4 − xn 4 4 − xn

so the sequence is bounded from below by 0.

Decreasing: (xn ) decreasing iff xn+1 ≤ xn ∀ n. We have x2 = 1
4−3
= 1 =⇒ x1 =
3 ≥ 1 holds. Say xn−1 ≥ xn for some n ≥ 1. Then, we have

1 1
xn−1 ≥ xn =⇒ 4−xn−1 ≤ 4−xn =⇒ ≥ = xn+1 =⇒ xn ≥ xn+1
4 − xn−1 4 − xn

and thus the sequence decreases, and by theorem 2.5 the limit exists. Let X =
1 1
limn→∞ xn = limn→∞ =⇒ X = 4−X
4−xn−1
=⇒ 4X − X 2 = 1 =⇒ 0 =
√
X 2 − 4X + 1 =⇒ X = · · · = 2 ± 3. We must take the negative root, since X is
decreasing and thus must be less than 3. ■
33
Abbott, pg 54 exercise 2.4.2

2.3 Limit Superior, Inferior

,→ Definition 2.7: limsup, liminf

Recall theorem 2.2, stating that a convergence sequence is bounded. Let (xn ) be a
convergent sequence bounded by m and M from below/above resp, ie

m ≤ xn ≤ M, ∀ n

§2.3 Sequences: Limit Superior, Inferior p. 37

 and let An = {xk : k ≥ n} (the set of elements in the sequence “after” a particular
index).
Let yn = sup An ; by definition, yn ≤ M , and yn ≥ m, since yn ≥ xn ≥ m. Thus, we
have
A1 ⊇ A2 ⊇ · · · ⊇ An ⊇ An+1 ⊇ · · · ,

and further,
y1 ≥ y2 ≥ · · · ≥ yn ≥ yn+1 ≥ · · · ;

since A2 ⊆ A1 , y1 also an upper bound for A2 , and thus y2 ≤ y1 by definition of a

supremum.
So, the sequence (yn ) is decreasing, and bounded from below; by MCT, limn→∞ yn = y
exists. Note too that since m ≤ yn ≤ M , we have that m ≤ y ≤ M .
This y is called the limit superior of (xn ) denoted by

limn→∞ xn = lim sup xn .

n→∞

Now, similarly, note that An is bounded below by m and thus zn = inf An exists. We
further have that zn ≤ xn ≤ M , and that zn ≥ m ∀ n, and we have

z1 ≤ z2 ≤ · · · ≤ zn ≤ zn+1 ≤ · · · ,

by a similar argument as before. So, as before, the sequence (zn ) is increasing, and
bounded from above by M . Again, by MCT, limn→∞ zn = z exists. We call z the limit
inferior of (xn ), and denote

limn→∞ xn = lim inf xn .

n→∞

We note that yn ≥ zn , so limn→∞ xn ≥ limn→∞ xn (y ≥ z).

Further, lim inf and lim sup exist for any bounded sequence, regardless if whether or
not the limit itself exists.

⊛ Example 2.10
Let (xn ) = (−1)n , n ∈ N. We showed previously that this is a divergent sequence,
so the limit does not exist. However, the sequence is bounded, since −1 ≤ xn ≤
1 ∀ n. We have An = {(−1)k : k ≥ n} = {−1, 1}. So, yn = sup An = 1, and zn =
inf An = −1, ∀ n. Thus, lim sup xn = lim yn = 1, and lim inf xn = lim zn = −1,
despite lim xn not existing.

§2.3 Sequences: Limit Superior, Inferior p. 38

 More specifically, we have a divergent sequence, and lim inf ̸= lim sup.

,→ Theorem 2.6: lim inf, lim sup and convergence

Let (xn ) be a bounded sequence. The following are equivalent;

1. The sequence (xn ) is convergent, and limn→∞ xn = x.

2. limn→∞ xn = limn→∞ xn = x.

Proof. Let An , yn , zn be as in the definition of lim sup, lim inf.

(1) =⇒ (2): Suppose (xn ) is converging, and limn→∞ xn = x. Let ε > 0. Then, there
exists some N ∈ N s.t. ∀ n ≥ N ,
ε
|xn − x| < ,
2
or equivalently,
ε ε
x− < xn < x + , ∀ n ≥ N.
2 2
Since An = {xk : k ≥ n}, if n ≥ N , then x + ε
2
is an upper bound for An , and x − ε
2
is a
lower bound for An . This gives that

ε ε
yn = sup An ≤ x + ; zn = inf An ≥ x − .
2 2

This gives that for n ≥ N ,

ε ε
x− ≤ zn ≤ xn ≤ yn ≤ x + ,
2 2

ie zn , yn ∈ [x − 2ε , x + 2ε ]. So, for all n ≥ N , |zn − x| ≤ ε

2
< ε, and |yn − x| ≤ ε
2
< ε, so by
definition of the limit, this gives

lim yn = x and lim zn = x,

n→∞ n→∞

ie, limn→∞ xn = limn→∞ xn = x.

•

(2) =⇒ (1): Let ε > 0. Since limn→∞ yn = x, ∃N1 s.t. ∀ n ≥ N1 , |yn − x| < ε. Similarly,
since lim zn = x, ∃N2 s.t. ∀ n ≥ N2 , |zn − x| < ε.
Take N = max{N1 , N2 }. Then, for n ≥ N , we have

x − ε < zn ≤ xn ≤ yn < x + ε.

§2.3 Sequences: Limit Superior, Inferior p. 39

So, for n ≥ N, |xn − x| < ε, thus lim xn = x as desired. ■

⊛ Example 2.11
Let (xn ) be a bounded sequence. Then

lim sup(−xn ) = − lim inf xn .

n→∞ n→∞

Proof. Recall Remark 1.2; Let An := {xk : k ≥ n} as in the definition of lim sup, lim inf.
Let yn := sup An , zn := inf An . By theorem 2.6, lim yn = lim zn . Further, sup(−An ) =
− inf(An ), where −An = {−xk : k ≥ n}; hence, lim sup(−xn ) = − lim inf xn , as
desired. ■

Remark 2.3. Given (xn ) bounded and α ≥ 0, then the following holds:

limn→∞ (αxn ) = α limn→∞ (xn ) and limn→∞ (αxn ) = α limn→∞ xn .

,→ Proposition 2.2
Let (xn ) and (yn ) be bounded sequences. Then,

(1) limn→∞ (xn + yn ) ≤ limn→∞ xn + limn→∞ yn

and
(2) limn→∞ (xn + yn ) ≥ limn→∞ xn + limn→∞ yn

Proof. (1) Take An = {xk + yk : k ≥ n}, Bn = {xk : k ≥ n}, Cn = {yk : k ≥ n}. Then,
take
Bn + Cn = {xk + yj : k ≥ n, j ≥ n} ⊇ An

and so sup An ≤ sup(Bn + Cn ). We have shown previously (assignment question) that

sup(Bn + Cn ) = sup Bn + sup Cn . Let now

tn = sup An rn = sup Bn sn = sup Cn ,

so tn ≤ rn + sn , that is, lim tn ≤ lim rn + lim sn , and thus limn→∞ (xn + yn ) ≤ limn→∞ xn +
limn→∞ yn , proving (1).
(2) The same argument holds, replacing each instance of limn→∞ with limn→∞ and reversing
inequalities where necessary. Alternatively, it follows directly from (1) by negating the
sequences where appropriate. ■

§2.3 Sequences: Limit Superior, Inferior p. 40

 ,→ Proposition 2.3
Let (xn ) be a bounded sequence. Then

1. limn→∞ xn = inf{t : {n : xn > t} is either empty or finite }

2. limn→∞ xn = sup{t : {n : xn < t} is either empty or finite }

Remark 2.4. (2) follows from (1) by either repeating the argument used to prove (1) (changing
notation), or using the fact that limn→∞ xn = − limn→∞ (−xn ).

Remark 2.5. The set {n : xn > t} is empty or finite iff ∃ nt ∈ N s.t. ∀ n > nt , xn ≤ t. The
set is empty or finite if t is an eventual upper bound for (xn ); that is, starting with sufficiently
large nt , xn ≤ t ∀ n ≥ t.

In other words, t is an upper bound if we neglect finitely many elements. Hence, (1) states
equivalently states that limn→∞ xn is the infimum of the eventual upper bounds for (xn );

{n : xn > t} empty or finite ⇐⇒


empty xn ≤ t ∀ n ⇐⇒ t an upper bound of xn


finite

t upper bounds xn for an infinite number of n’s

Proof. (Of (1)) Let A = {t : {n : xn > t} is either empty or finite }. We note that this set
is non-empty and bounded from below, hence the inf is well-defined. We can see this by
recalling that (xn ) bounded, hence ∀ n ∃m, M s.t. m ≤ xn ≤ M . Then, {n : xn > M } is
empty, hence M ∈ A. Otoh, if t < m, then the set {n : xn > t} = N since xn ≥ m > t ∀ n.
So, if t < m, then t ∈
/ A and hence m is a lower bound for A.

We have now that limn→∞ xn is a lower bound for A, and hence limn→∞ xn ≥ inf A.
Let t ∈ A. We aim to show that limn→∞ xn ≤ t.

The set {n : xn > t} is finite by definition; assume t ∈ A. We can then let

nt = max{n : xn > t}.

Then, if k > nt , it must be that xk ≤ t. Consider now n > nt , then yn = sup{xk : k ≥ n}

and since xk ≤ t for k ≥ n, and t upper bounds {xk : k ≥ n}, we have that yn ≤ t for
n > nt . Hence, for sufficiently large n, yn ≤ t, thus lim yn ≤ t =⇒ limn→∞ xn ≤ t.

Thus, limn→∞ xn ≤ inf A. ■

Proof. (Of (1), More Concise) Let α = inf A and ε > 0. Then α − ε is not in A, α − ε is not

§2.3 Sequences: Limit Superior, Inferior p. 41

an eventual upper bound for xn . Since α − ε ̸∈ A, the set {n : xn > α − ε} is infinite.34 . 34
This is important; understand
why this is
Hence, for any m we can find n such that n ≥ m and xn > α − ε.

Let, now,
ym = sup{xn : n ≥ m}.

By our last observation, we have that ym > α − ε. By Order Properties of Limits,

limn→∞ xn = lim yn ≥ α − ε,
m→∞

so for any ε > 0, limn→∞ xn ≥ α − ε, and thus limn→∞ xn ≥ α = inf A, and the proof is
complete. ■

2.4 Subsequences and Bolzano-Weirestrass Theorem

,→ Definition 2.8: Subsequence

Let (xn ) be a sequence of real numbers, and let n1 < n2 < n3 < · · · < nk < nk+1 <
· · · be a strictly increasing sequence of natural numbers. Then, the sequence

(xn1 , xn2 , · · · , xnk , xnk+1 , · · · )

is called a subsequence of (xn ) and is denoted (xnk )k∈N .

Remark 2.6. k is the index of the subsequence, (xnk )k∈N , not n; xn1 is the 1st element, . . . ,
xnk is the k-th element.

⊛ Example 2.12
Let xn = n1 , ( n1 )n∈N , and let nk = 2k + 1, k ∈ N. n1 = 3, n2 = 5, n3 = 7, . . . , nk =
2k + 1. Our subsequence is then
   
1 1 1 1
(xn1 , xn2 , . . . , xnk , . . . ) = , ,..., ,... =
3 5 2k + 1 2k + 1 k∈N

is our subsequence of (xn ).

Remark 2.7. Note that for any k, nk ≥ k.

Let S = {k ∈ N : nk ≥ k}. Then, 1 ∈ S, since n1 ∈ N, n1 ≥ 1 . If k ∈ S, then nk ≥ k,
and so, since nk+1 > nk (increasing), we have that nk+1 > k =⇒ nk+1 ≥ k + 1. So,
k + 1 ∈ S, S = N.

Remark 2.8. limk→∞ xnk = x if ∀ ε > 0, ∃K ∈ N s.t. ∀ k ≥ K, |xnk − x| < ε.

§2.4 Sequences: Subsequences and Bolzano-Weirestrass Theorem p. 42

 ,→ Theorem 2.7
Let (xn ) be a sequence such that limn→∞ xn = x. Then, for any subsequence (xnk )k∈N ,
we have that limk→∞ xnk = x

Proof. Let ε > 0. Since limn→∞ xn = x, ∃N ∈ N s.t. ∀ n ≥ N , |xn − x| < ε. Take K = N

(from Remark 2.8). Then, for k ≥ K, we have from Remark 2.7 that

nk ≥ k ≥ K = N,

and hence |xnk − x| < ε =⇒ limk→∞ xnk = x. ■

,→ Theorem 2.8: Bolzano-Weirestrass Theorem

35
Any bounded sequence (xn ) has a convergent subsequence.
35
Fundamental property of the
real line; equivalent to AC.
⊛ Example 2.13
Take xn = (−1)n , n ∈ N. This sequence does not converge. However, if we take
a subsequence with nk = 2k, k ∈ N. xnk = (−1)2k = 1, so (xnk ) is a constant
sequence 1 and converges to 1.
Similarly, if nk = 2k + 1, k ∈ N, then xnk = (−1)2k+1 = −1, and the subsequence
converges to −1.

,→ Proposition 2.4
If 0 < b < 1, then limn→∞ bn = 0.

Proof. Let xn = bn . Then xn > 0, and xn+1 = bn+1 = bxn < xn , and since 0 < b < 1,
(xn ) is decreasing and bounded from below, (xn ) converges by the Monotone Convergence
Theorem. Let x = limn→∞ xn . Again, xn+1 = bxn , so limn→∞ xn+1 = limn→∞ bxn =
b limn→∞ xn , so x = bx =⇒ (1 − b)x = 0. 0 < b < 1 =⇒ x = 0. ■

36
BW Proof (1): using Nested Interval Property. Since (xn ) bounded, ∃M > 0 s.t. |xn | ≤ M ∀ n ∈
N. Let I1 = [−M, M ] and n1 = 1. We now construct I2 , n2 as follows.

Divide I1 into two intervals of the same size, I1′ = [−M, 0], I1′′ = [0, M ]. Now, consider
the sets

A1 = {n ∈ N : n > n1 (= 1), xn ∈ I1′ }, A2 = {n ∈ N : n > n1 , xn ∈ I1′′ }

§2.4 Sequences: Subsequences and Bolzano-Weirestrass Theorem p. 43

(ie, all the indices of all the elements in I1′ , I1′′ resp.).
Hence, A1 ∪ A2 = {n : n > n1 }, an infinite set, and hence, one of A1 , A2 must be infinite
(by theorem 1.9). If A1 infinite, set I2 = I1′ , n2 = min A1 . If A1 finite, then A2 infinite, and
set I2 = I1′′ , n2 = min A2 .

Suppose now that Ik , nk are chosen, and that Ik contains infinitely many elements of
the sequence (xn ). Divide Ik into two equal sub-intervals, Ik′ , Ik′′ . We now introduce

(k) (k)
A1 = {n ∈ N : n > nk and xn ∈ Ik′ }, A2 = {n ∈ N : n > nk and xn ∈ Ik′′ },

(k) (k)
(similar to our construction of A1 , A2 ). A1 ∪ A2 must be infinite, so one of the two
(k)
must be infinite. If A1 infinite, set Ik+1 = Ik′ , nk+1 = min A1 . If A2 infinite, set Ik+1 =
(k)
Ik′′ , nk+1 = min A2 .

This gives now that Ik+1 and nk+1 , where Ik+1 ⊆ Ik , Ik+1 contains infinitely many
elements of the sequence. Further, by construction, nk+1 > nk . This gives us a sequence
of closed intervals Ik = [ak , bk ], k ∈ N such that I1 ⊇ I2 ⊇ · · · ⊇ Ik ⊇ Ik+1 ⊇ · · · , such
that xnk ∈ Ik , and that nk is a strictly increasing sequence of natural numbers, defining
subsequence (xnk ).

Now, by construction, the length of Ik+1 is 1

2
of the length of Ik . Since Ik = [ak , bk ],
then
bk−1 − ak−1 b 1 − a1 2M M
b k − ak = = · · · k−1 = k−1 = k−2 .
2 2 2 2k
Since Ik , k ∈ N, is a nested sequence of closed intervals and by the nested interval property
of the real line (AC), ∃x ∈ ∞ k=1 Ik .
T

We claim now that our subsequence (xnk ) satisfies limk→∞ xnk = x. To see this, let
ε > 0. Since limk→∞ M
2k−2
= limk→∞ 4M
2k
= 0, by proposition 2.4, with b = 21 . There exists
K ∈ N such that ∀ k ≥ K, we have M
2k−2
= bk − ak < ε. So, since Ik is a nested sequence of
intervals, ∀ k ≥ K, xnk ∈ IK (xnk ∈ Ik ⊆ IK ). We also have that x ∈ IK , since x ∈ Ik .
T

So, x, xnk ∈ [aK , bK ] ∀ k ≥ K. So, for k ≥ K, |xnk − x| ≤ |bk − ak | < ε. So for ε > 0,
∃K ∈ N s.t. ∀ k ≥ K, |xnk − x| < ε, and so limk→∞ xnk = x, as desired. ■
36
Sketch: this proof is
somewhat diagonal in nature
(if one can say that); if you
,→ Definition 2.9: Peak
understand the proof of
Let (xn ) be a sequence of real numbers. An element xm is called a peak of this sequence Cantor’s Theorem using the
Nested Interval property, this
if xm ≥ xn ∀ n ≥ m. xm is bigger or equal then to any element of the sequence that should follow naturally. In
short, construct subsequences
follows it. such that the subsequence
If a sequence is decreasing, then any element of the sequence is a peak. has all its terms contained in
a “nest” of intervals, and
If a sequence is increasing, then there is no peak. show that the length (sts) of
these intervals converges to 0.
But these are subsets of R,
§2.4 Sequences: Subsequences and Bolzano-Weirestrass Theorem p. 44 their intersect must contain
BW Proof (2): using Peaks. Take sequence (xn ). Then,

• Case 1: (xn ) has infinitely many peaks; enumerate the indices of those peaks as
n1 < n2 < n3 < · · · , then xnk < xnk+1 ∀ k, since xnk is a peak, nk+1 > nk . This gives
a decreasing subsequence (xnk ).

• Case 2: (xn ) has finitely many peaks, with indices m1 < m2 < · · · < mr . Set
n1 = mr + 1. Then xn1 is not a peak, and so ∃ n2 > n1 s.t. xn2 > xn1 . Now, xn2 is
also not a peak, (n2 > n1 > mr ), and so there exists n3 > n2 such that xn3 > xn2 ,
and so on. In this way, we construct a subsequence (xnk ) that is strictly increasing,
that is, xnk+1 > xnk .

If in addition (xn ) is bounded, say |xn | ≤ M ∀ n, then the monotone subsequence con-
structed in Cases 1, 2 is also bounded; ie |xnk | ≤ M ∀ k. Thus, by Monotone Convergence
Theorem, (xnk ) is converging. ■

2.5 Cauchy Sequences

,→ Definition 2.10: Cauchy Sequence

A sequence (xn ) is called Cauchy if for every ε > 0, ∃N ∈ N s.t. ∀ n, m ≥ N, |xn − xm | <
ε.

,→ Theorem 2.9: Cauchy Criterion

A sequence (xn ) is convergent iff it is Cauchy.37
37
Sketch: Convergent =⇒
Cauchy; use definition of
Cauchy, add/subtract limit of
Remark 2.9. This is, again, an “equivalent” formulation of AC; at least, the direction (xn ) sequence, triangle inequality
Cauchy =⇒ convergent is. The other direction, convergent =⇒ Cauchy, does not rely on (and choose your ε to be ε/2,
optional).
AC. Convergent ⇐= Cauchy;
show that any Cauchy
sequence is bounded
Remark 2.10. AC ⇐⇒ BW, AC ⇐⇒ MCT, AC ⇐⇒ NIP; AC ⇐⇒ Cauchy Criterion +
(theorem 2.11), and thus has a
Archimedean Property converging subsequence
(Bolzano-Weirestrass
Theorem); finally, show that
Remark 2.11. Beyond the real line, AC (in terms of sup) cannot be formulated, because of the any Cauchy sequence that
lack of ordering. In this case, the Cauchy criterion can be used to extend AC to other spaces. has a converging
subsequence itself converges
(theorem 2.10), and you are
Proof. (theorem 2.9; (xn ) Convergent =⇒ Cauchy ) done.

Suppose limn→∞ xn = x. Let ε > 0, N ∈ N s.t. ∀ n ≥ N , |xn − x| < ε

2
. Then, for

§2.5 Sequences: Cauchy Sequences p. 45

n, m ≥ N ,

ε ε
|xn − xm | = |xn − x + x − xm | ≤ |xn − x| + |xm − x| < + =ε
2 2
=⇒ |xn − xm | < ε,

hence (xn ) is Cauchy. ■

Remark 2.12. To prove ⇐= , we first introduce the following theorem(s); see section 2.5 for
the remainder.

,→ Theorem 2.10
Let (xn ) be a Cauchy sequence and suppose that (xn ) has a convergent subsequence
(xnk ). Then (xn ) is also convergent.

Proof. Let x = limn→∞ xnk . Let ε > 0. Then, ∃K ∈ N such that ∀ k ≥ K, |xnk − x| < ε.
We have too that (xn ) Cauchy, ie ∃N ∈ N s.t. ∀ n, m ≥ N , |xn − xm | < 2ε .
Let now K0 ≥ max{K, N }. Recall that nK0 ≥ K0 ≥ N . Take now n ≥ N , and estimate

|xn − x| = xn − xnK0 + xnK0 − x ≤ xn − xnK0 + xnK0 − x

Since K0 ≥ K, xnK0 − x < 2ε . Since nK0 ≥ N , we also have xn − xnK0 < 2ε . Thus, we
have
ε ε
|xn − x| < + = ε,
2 2
hence limn→∞ xn = x. ■

Remark 2.13. This did not use AC.

,→ Theorem 2.11
Any Cauchy sequence is bounded.

Proof. Let (xn ) be Cauchy. We aim to show that ∃M > 0 s.t. ∀ n ∈ N, |xn | ≤ M .
Take ε = 1 in the definition of Cauchy sequence. Let N be such that ∀ n, m ≥ N ,
|xn − xm | < 1. We can take m = N , and so for all n ≥ N , |xn − xN | < 1, which gives that
for n ≥ N ,

|xn | = |xn − xN + xN | ≤ |xn − xN | + |xN | < 1 + |xN |

Let38 38
While this seems like an
arbitrary definition, this is a
common “trick” to find a
§2.5 Sequences: Cauchy Sequences p. 46
bound of a sequence based on
 M = |x1 | + |x2 | + . . . |xN −1 | + |xN | + 1.

Then, if n ≤ N , |xn | ≤ M ; if n ≥ M, |xn | ≤ M , so ∀ n ≥ 1, |xn | ≤ M , hence (xn ) is

bounded. ■

Remark 2.14. This did not use AC.

Proof. (theorem 2.9; (xn ) Convergent ⇐= Cauchy)

If (xn ) Cauchy, then (xn ) is bounded by theorem 2.11, and thus by Bolzano-Weirestrass
Theorem, (xn ) has a convergent subsequence (xnk ). Then, by theorem 2.10, (xn ) must
converge. ■

⊛ Example 2.14
Let39 (xn ) be a sequence defined recursively by x1 = 1, x2 = 2, xn+1 = 1
2
(xn +
xn−1 ), n ≥ 2. Prove that (xn ) is a convergence sequence, and find its limit.
39
Sketch: show xn Cauchy
=⇒ xn converges, then take
a subsequence of xn (spec,
Remark 2.15. Before solving, we establish a number of properties about the sequence. odd n) and find a closed form
of it which is nicer to
evaluate. Use then
,→ Proposition 2.5: Property I theorem 2.7 to conclude that
the limit of the subsequence
1 ≤ xn ≤ 2 ∀ n ≥ 1 is equal to the limit of the
sequence.

Proof. We proceed by induction. Let S ⊆ N be the set of all n such that 1 ≤ xn ≤ 2.

Base Case: 1 ∈ x, since x1 = 1.
Assumption: suppose {1, 2, . . . , n} ∈ S. We want to show that n + 1 ∈ S.
If n = 1, then x2 = 2, so x2 ∈ S. If n > 1, then

1
xn+1 = (xn + xn+1 ),
2

and by inductive assumption, 1 ≤ xn ≤ 2 and 1 ≤ xn−1 ≤ 2, hence

1 ≤ xn+1 ≤ 2,

hence n + 1 ∈ S, and thus S = N. ■

,→ Proposition 2.6: Property II

For all n ≥ 1, |xn+1 − xn | = 1
2n−1
.

§2.5 Sequences: Cauchy Sequences p. 47

Proof. We proceed by induction. Let S ⊆ N be the set of all n such that the statement holds
for xn .
Base Case: x2 = 2, x1 = 1, hence 2 − 1 = 1 = 1
20
= 1, holds.
Assumption: suppose n ∈ S, ie |xn+1 − xn | = 1
2n−1
holds for n. Then,

1
|xn+2 − xn+1 | = (xn+1 + xn ) − xn+1
2
1 1 1
= xn − xn+1 = |xn+1 − xn |
2 2 2
1 1 1
(assumption =⇒ ) = · n−1 = n ,
2 2 2

hence the statement holds for n + 1, and S = N. ■

,→ Corollary 2.2
1−rk+1
For any r ̸= 1, and any k ∈ N, 1 + r + r2 + · · · + rk = 1−r
.

1−r1
Proof. We proceed by induction. k = 1 =⇒ r0 = 1−r
= 1, holds. Suppose 1 + · · · rk−1 =
1−rk
1−r
holds for some k − 1 ∈ N. Then, we have that

1 − rk 1 − rk + (1 − r)rk
1 + · · · rk−1 + rk = + rk =
1−r 1−r
k k k+1
−r + r − r
1 

=
1−r
1 + 1 − rk+1
= ,
1−r

hence, the statement for k − 1 =⇒ the statement for k, hence it holds ∀ k ∈ N and the
proof is complete. ■

,→ Proposition 2.7: Property III

(xn ) a Cauchy sequence.

Proof. Let ε > 0. We need to find N ∈ N such that ∀ n, m ≥ N , |xn − xm | < ε. Let N be
such that40 2N1−2 = 4
2N
< ε. Let, now, n, m ≥ N , and suppose n > m (when n = m, we are
done; when n < m, simply switch the variables wlog). We can write

|xn − xm | = |xn − xn−1 + xn+1 − xn−2 + xn−2 + · · · − xm+1 + xm+1 − xm |

≤ |xn − xn−1 | + |xn−1 − xn−2 | + · · · + |xm+1 − xm |41

§2.5 Sequences: Cauchy Sequences p. 48

Using Property II we can write

1 1 1
|xn − xm | ≤ + + · · ·
2m−1  2m 2n−2 
1 1 1
= m−1 1 + + · · · + n−m−1
2 2 2

By corollary 2.2, with r = 1

2
and k = n − m − 1, we have

n−m !
1 − 12
 
1 1 1 1 1 1
1 + + · · · + n−m−1 = m−1 1 < m−2 ≤ N −2 .
2m−1 2 2 2 1− 2 2 2

We have chosen N so that 1

2N −2
< ε, hence for n, m ≥ N , |xn − xm | < ε, and thus our
sequence is Cauchy, so limn→∞ xn = X exists. ■
36
lim 21n = 0, so such an N
exists.
37
“Telescoping” the sequence;
Proof. (Of example 2.14) the inequality follows directly
By Property III, the limit lim xn = X exists. From the recursive definition, we can write from the triangle inequality.

1
X = lim xn = lim( (xn−1 + xn−2 ))
2
1
=⇒ X = (X + X) = X,
2

which, while true, is useless. Rather, consider the subsequence

(x2k+1 )k∈N

of (xn ). We claim, then, that

1 1 1
x2k+1 = 1 + + 3 + · · · + k−1 , k ≥ 1. ⋆
2 2 2

Note that ∀ n ≥ 1, x2n ≥ x2n−1 and x2n ≥ x2n+1 . We can argue by induction. Let S ⊆ N
for which the relation holds. Since x1 = 1, x2 = 2, x3 = 23 , we have that x2 ≥ x1 , x2 ≥ x3 ,
and so the relation holds, ie 1 ∈ S. Suppose that n ∈ S, ie x2n ≥ x2n−1 , x2n ≥ x2n+1 for
some n ≥ 1. We can write

1 1
x2k+2 = (x2k+1 + x2k ) ≥ (x2n+1 + x2n+1 ) ≥ x2n+1
2 2
1 1
=⇒ x2n+3 = (x2n+2 + x2n+1 ) ≤ (x2n+2 + x2n+2 ) = x2n+2
2 2

Hence x2n+2 ≥ x2n+1 and x2n+2 ≥ x2n+3 , n + 1 ∈ S, and hence S = N, and our relation
holds ∀ n ∈ N.

§2.5 Sequences: Cauchy Sequences p. 49

 Recall now that ∀ n, |xn+1 − xn | = 1
2n−1
. We then have the following, given the relation
we proved above;

x2n+1 − x2n−1 = x2n+1 − x2n + x2n − x2n−1

| {z } | {z }
≤0 ≥0
1 1 1 2 1
=− + =− + =
22n−1 22n−2 22n−1 22n−1 22n−1

From here, we can prove the claim ⋆ by induction.

Summing up the RHS of ⋆, and factoring out a 21 , we have

 k−1 !
1 1 1
x2k+1 =1+ 1 + 2 + ··· + .
2 2 22

Recalling corollary 2.2, and taking r = 1

4
and ℓ = k − 1, we have

1 k

!
1 1 − 4
x2k+1 =1+
2 1 − 41
 k !
2 1
=1+ 1−
3 4
 k
5 2 1
= −
3 3 4

Thus, we have that limk→∞ x2k+1 = 53 , as the term ( 14 )k goes to zero.

Now, since limn→∞ xn = X and we showed xn convergent, then each of its subse-
quences converges to the same limit. Thus, X = 53 , ie,

5
lim xn = .
n→∞ 3

Remark 2.16. Generally, this type of approach is quite tedious. The next example(s) will try
to generalize it.

⊛ Example 2.15
Consider the recursive relation xn+1 = 12 xn + 21 xn−1 ⋆.

Proof. We have the following characteristic equation of the sequence:

1 1
x2 = x + ,
2 2

§2.5 Sequences: Cauchy Sequences p. 50

 with solutions a = 1, b = − 12 . We can now write the following sequence:

1
xn = C1 an + C2 bn = C1 + C2 (− )n , ⋆⋆
2

where C1 , C2 are arbitrary constants. We claim that this sequences satisfies our
recursive relation, ⋆; note that
 n+1  n−1     
1 1 1 1 1 1 1
− = − · = − − +
2 2 4 2 2 2 2
 n+1
1
=⇒ xn+1 = C1 + C2 −
2
 n−1   
C1 C1 1 1 1 1
= + + C2 − − +
2 2 2 2 2 2
 n  n
C1 C1 1 C2 1
= + + C2 − + −
2 2 2 2 2
 n  n−1
C1 C2 1 C1 C2 1
= + − + + −
2 2 2 2 2 2
xn xn−1
= +
2 2

Hence, our ⋆⋆ is our so-called general solution to ⋆. The only factor we must find,
then, are our C1 , C2 . Recall our initial x1 = 1, x2 = 2. Plugging these into ⋆⋆, then,
gives
   2
1 1
x1 = C 1 + C 2 − = 1; x2 = C 1 + C 2 − = 2,
2 2
which is simply a system of two equations for two unknowns, C1 , C2 . Solving
them42 , we have
5 4
C1 = , C2 = ,
3 3
hence we have the general formula
 n
5 4 1
xn = + −
3 3 2

The RHS of this sum goes to zero, and thus our limit is

5
lim xn = .
n→∞ 3

■
Remark 2.17. From this general form, we can conclude, as in example 2.14, that x2n ≥
x2n−1 , x2n ≥ x2n+1 , since x2n > 53 , x2n+1 < 35 , x2n−1 < 53 ; ie, the same property that we used
to prove the previous example holds here.

Remark 2.18. Any recursively defined sequence of the form xn+1 = Axn + Bxn−1 , n > 1

§2.5 Sequences: Cauchy Sequences p. 51

where A, B ∈ R, can be solved using the characteristic equation

x2 = Ax + B,

√ √
A+ A2 +4B A− A2 +4B
with solutions a = 2
,b = 2
. It may be that a, b ∈ C; we shall not consider
these cases. Indeed, we have:

xn+1 = C1 an+1 + C2 bn+1

= ···
= C1 an−1 (Aa + B) + C2 bn−1 (Ab + B)
= C1 Aan + C1 an−1 B + C2 Abn + C2 Bbn−1
= A(C1 an + C2 bn ) + B(C1 an−1 + C2 bn−1 )
= Axn + Bxn−1

Given initial x1 , x2 , then we have that

x1 = C1 a + C2 b, x2 = C2 a2 + C2 b2 .

C1 , C2 are uniquely determined by this relation, as long as the matrix of coefficients

a b
= ab2 − ba2 ̸= 0.
2 2
a b

In the case a = b, or a = 0 or b = 0, then the determinant is also equal to 0, and we thus have
to use a different method. As long as the determinant is nonzero, then we have a valid specific
definition.

Remark 2.19. Note that nothing in this derivation assumed xn convergent; this form can
indeed be found even if xn diverges. It will simply also diverge.

Remark 2.20. The recursive relation xn+1 = Axn +Bxn−1 is a discrete analog of a differential
equation.

2.6 Contractive Sequences

,→ Definition 2.11: Contractive Sequences

A sequence (xn ) of real numbers is called contractive with contractive constant K,
where 0 < K < 1, if |xn+2 − xn+1 | ≤ K|xn+1 − xn | ∀ n ≥ 1, ie, the distance between
successive elements of the sequence are contracted at least by a factor of K.

§2.6 Sequences: Contractive Sequences p. 52

 We have, by extension, that

|xn − xn−1 | ≤ K|xn−1 − xn−2 |

≤ K 2 |xn−2 − xn−3 |
≤ ···
≤ K n−2 |x2 − x1 |.

,→ Theorem 2.12
Let43 (xn ) be a contractive sequence with contractive constant K. Then, (xn ) is a
Cauchy sequence, and in particular, (xn ) converges.
43
Sketch: start with |xn − xm |,
and add/subtract each term
between xn and xm , use
Proof. Let n, m ∈ N such that n > m ≥ 2. Then, we have triangle inequality,
“substitute” in contractive
constant, collect like terms,
|xn − xm | = |xn − xn−1 + xn−1 − xn−2 + xn−2 − · · · − xm+1 + xm+1 − xm | and simplify. This creates an
upper bound for |xn − xm |,
≤ |xn − xn−1 | + |xn−1 − xn−2 | + · · · + |xm+1 − xm | which converges to 0, then
use this converges to define
≤ K n−2 |x2 − x1 | + K n−3 |x2 − x1 | + · · · + K m−1 |x2 − x1 | the epsilon to use in the
Cauchy definition.
= K m−1 |x2 − x1 | 1 + K + K 2 + · · · + K n−m−1

1 − K n−m
= K m−1 |x2 − x1 | by corollary 2.2
1−K
K m−1 |x2 − x1 |
<
1−K
K m−1 |x2 − x1 |
=⇒ |xn − xm | < ∀n > m ≥ 2
1−K

K m−1 K m−1
lim |x2 − x1 | = 0 =⇒ ∀ ε > 0, ∃N s.t. ∀ m > N, |x2 − x1 | < ε
1−K 1−K
K m−1
→ n > m ≥ N =⇒ |xn − xm | ≤ |x2 − x1 | < ε
1−K
K n−1
→ m > n ≥ N =⇒ |xn − xm | ≤ |x2 − x1 | < ε
1−K
=⇒ ∀ m, n ≥ N, |xm − xn | < ε, and (xn ) Cauchy

Remark 2.21. This proof also gives us a rate of convergence; we have

K m−1
|xn − xm | ≤ · |x2 − x1 |,
1−K

together with the fact that limn→∞ xn = X, whose convergence also implies by Algebraic

§2.6 Sequences: Contractive Sequences p. 53

Properties of Limits that
lim |xn − xm | = |X − xm |.

This implies, by Order Properties of Limits, that

K m−1
|X − xm | ≤ |x2 − x1 |,
1−K

that is, the sequence converges exponentially fast.

Remark 2.22. We have that limn→∞ |xn − xm | = |X − xm | where (xn ) → X, by the

inequality
||X − xm | − |xn − xm || ≤ |x − xn | < ε,

following from the more general fact that

||a| − |b|| ≤ |a − b| ∀ a, b ∈ R,

a direct consequence of the Triangle Inequality detailed in lemma 2.1.

Remark 2.23. The result that every contractive sequence is convergent is a simple example of
the more general “Fixed Point Theorems”; this proof can be generalized to the Banach Fixed
Point Theorem on arbitrary metric spaces. This is further used to establish the existence and
uniqueness of solutions of differential, integral equations.44 44
See the Picard-Lindelöf
Theorem
Remark 2.24. In the case of the recursively defined

1
xn+1 = (xn + xn−1 ),
2

we have that
1
|xn+2 − xn+1 | > |xn+1 − xn |,
2
that is, xn is a contractive sequence with K = 21 . The argument used to prove that this in-
equality implies (xn ) Cauchy is the same as the one we used to prove a general contractive
sequence is Cauchy.

⊛ Example 2.16
Let (xn ) be a sequence defined recursively by x1 = 2, xn+1 = 2 + 1
xn
. Prove that
the sequence converges and find its limit.

Proof. First, we note that xn ≥ 2 ∀ n. Now, we aim to show that (xn ) is contractive

§2.6 Sequences: Contractive Sequences p. 54

 with K = 14 :
 
1 1 1 1 xn − xn+1
xn+2 − xn+1 = 2 + − 2+ = − =
xn+1 xn xn+1 xn xn+1 · xn
|xn − xn+1 |
=⇒ |xn+2 − xn+1 | =
xn · xn+1
xn , xn+1 ≥ 2 =⇒ xn · xn · xn+1 ≥ 4
1
=⇒ ∀ n ≥ 1, |xn+2 − xn+1 | ≤ |xn+1 − xn |
4
theorem 2.12
=⇒ (xn ) contractive, hence convergent

We can now find the limit using the recursive definition; let X = limn→∞ xn . xn ≥
2, in particular, it is ̸= 0 for any n. Then, we have:
 
1 1
X = lim xn = lim 2 + =2+ =X
n→∞ n→∞ xn x
1
=⇒ X = 2 + =⇒ X 2 − 2X − 1 = 0
X
√
=⇒ X = 1 ± 2

√
1− 2 < 0, which can’t hold since xn ≥ 0 ∀ n, hence it must be that X = 1 +
√
2. ■

⊛ Example 2.17
Show that the sequence xn = 1 + 12 + · · · + n1 , n ≥ 1, diverges.

Proof. Note that

1 1 1 1 1
x2n − xn = + + ··· + ≥n· ≥ , ∗
|n + 1 n +{z2 2n} 2n 2
n terms, each ≥ 2n
1

which means that the sequence cannot be Cauchy hence it cannot be convergent
(see theorem 2.9).

More thoroughly, suppose (xn ) is convergent, that is, it is Cauchy. Take ε = 14 ;

since (xn ) Cauchy, there must exist some N such that ∀ n, m ≥ N ,

1
|xn − xm | < ε = .
4

§2.6 Sequences: Contractive Sequences p. 55

 But if we take, then, n = 2N and m = N , then

1
|x2N − xN | < ,
4

which is impossible, as we have shown in ∗ that |x2N − xN | ≥ 1

2
∀ N , hence we
have reached a contradiction. ■

2.7 Euler’s Number e

Remark 2.25. In the following section, we consider the sequences

 n
1
xn = 1 +
n

and  1+n
1
yn = 1+ .
n
We consider the following propositions regarding the sequences.

,→ Proposition 2.8: Step 1

xn is strictly increasing. 45
45
Proof sketch: lots of very ugly
algebra, starring Bernoulli’s
inequality.
,→ Proposition 2.9: Step 2
yn is strictly decreasing.46
46
Proof sketch: precisely the
same idea as Step 1, with just
ast much ugly algebra.
,→ Proposition 2.10: Step 3
For any n, k, xn < yk . 47
47
Proof sketch: very a-lá Nested
Interval Property. 3 cases.
,→ Proposition 2.11: Step 4
(xn ) is bounded from above and (yn ) is bounded from below. 48
48
Proof sketch: follows directly
from Step 3; any element of
yn , namely y1 , upper bounds
,→ Proposition 2.12: Step 5
xn , and any element of xn ,
(xn ) and (yn ) are converging sequences that namely x1 , lower bounds yn .

lim xn = lim yn ,
n→∞ n→∞

which we denote by the number e.49

49
Proof sketch: the sequences
converge by MCT (following
from the previous steps). yn
§2.7 Sequences: Euler’s Number e p. 56
is just x times a term (note
Remark 2.26. Step 3, Step 4, Step 5 are “easier”; the main parts of the proof deal with Step 1,
Step 2. We will prove it using Bernoulli’s Inequality.

,→ Proposition 2.13: Bernoulli’s Inequality

For all x > −1 and all n ∈ N,

(1 + x)n ≥ 1 + nx

Proof. We proceed by induction; fixing x > −1, let S ⊆ N the set for which the inequality
holds. n = 1 =⇒ (1 + x)1 ≥ 1 + x, which clearly holds, ie 1 ∈ S. Suppose n ∈ S, that is,

(1 + x)n ≥ 1 + nx

holds. Since 1 + x > 0, we can multiply both sides by 1 + x:

≥0
z}|{
(1 + x)n · (1 + x) = (1 + x)n+1 ≥ (1 + nx)(1 + x) = 1 + nx + x + nx2 ≥ 1 + (n + 1)x
=⇒ n + 1 ∈ S

Hence, by the axiom of induction, S = N. ■

Proof. (Of Step 1) We will show that xn+1

xn
> 1 ∀ n ∈ N. From our definition, we have

1

n+1 n+2 n+1


xn+1 1 + n+1 n+1 n + 2 (n + 2)n nn
=
n = n+1 n
= ·
1 + n1 n + 1 [(n + 1)2 ]n


xn n
n
n2 + 2n

n+2
=
n + 1 n2 + 2n + 1
n
n + 2 n2 + 2n + 1 − 1

=
n+1 n2 + 2n + 1
 n
n+2 1
= 1− 2
n+1 n + 2n + 1
 n
n+2 1
= 1−
n+1 (n + 1)2

By Bernoulli’s Inequality with x = − (n+1)

1
2 > −1, we have that

 n
1 n
1− ≥1− ,
(n + 1)2 (n + 1)2

§2.7 Sequences: Euler’s Number e p. 57

which gives with our results above

n + 2 n2 + n + 1
 
xn+1 n+2 n
≥ 1− = ·
xn n+1 (n + 1)2 n + 1 (n + 1)2
n3 + n2 + n + 2n2 + 2n + 2
=
n3 + 3n2 + 3n + 1
n3 + 3n2 + 3n + 2
= 3
n + 3n2 + 3n + 1
3 2
n + 3n + 3n + 1 1
= 3 2
+ 3 2
n + 3n + 3n + 1 n + 3n + 3n + 1
1
=1+ 3 >1
n + 3n2 + 3n + 1

hence, xn+1
xn
> 1 =⇒ xn+1 > xn ∀ n, ie it is strictly increasing. ■

Proof. (Of Step 2) We need to show yn

yn+1
> 1 ∀ n > 1. We have

(n+1)n+1

n+1 n+1 n+1
1 + n1


yn n+1
n nn+1
=
n+2 = = · (n+2)
yn+1 1 n+2 n+2
n+2 n+1


1 + n+1 n+1 (n+1)n+1
n+1 n+1
n + 1 [(n + 1)2 ] n + 1 n2 + 2n + 1

= · =
n + 2 nn+1 (n + 2)n+1 n+2 n2 + 2n
 n+1
n+1 1
= · 1+ 2
n+2 n + 2n
 
1 yn n+1 n+1
Bernoulli’s Inequalityx = 2 =⇒ ≥ 1+ 2
n + 2n yn+1 n+2 n + 2n
2
n + 1 n + 3n + 1
= ·
n+2 n2 + 2n
n3 + 3n2 + n + n2 + 3n + 1 n3 + 4n2 + 4n + 1
= =
n3 + 2n2 + 2n2 + 4n n3 + 4n2 + 4n
1
=1+ 3 >1
n + 4n2 + 4n

Hence, ∀ n, yn
yn+1
> 1 =⇒ yn > yn+1 , ie, it is strictly decreasing. ■

Proof. (Step 3) We aim to show that for all n, k, xn < yk .

- (Case 1) n = k:
 n   n  n+1
1 1 1 1
xn = 1 + < 1+ 1+ = 1+ = yn
n n n n

- (Case 2) n > k:

yk > yn > xn by Case 1, since (yn ) strictly decreasing.

§2.7 Sequences: Euler’s Number e p. 58

 - (Case 3) n < k:

xn < xk < yk by Case 1, since (xn ) strictly increasing.

Proof. (Of Step 4) Since xn < yk ∀ k, n, we have that

xn < y1 = 4 ∀ n,

and
2 = x1 < yk ∀ k,

hence (xn ) is bounded from above (by y1 , say) and (yn ) is bounded from below (by x1 ,
say). ■

Proof. (Of Step 5) Since (xn ) increasing and bounded from above, it is converging by Mono-
tone Convergence Theorem. Similarly, (yn ) is decreasing and bounded from below, hence
it too converges. We have too that
 n+1   n  
1 1 1 1
yn = 1 + = 1+ 1+ = 1+ xn
n n n n

Since limn→∞ 1 + 1
= 1, we have, from proposition 2.1, that


n

 
1
lim yn = lim 1 + · lim xn = lim xn ,
n→∞ n→∞ n n→∞

that is, (xn ) and (yn ) converge to the same limit, which we define as
 n  n+1
1 1
e ≡ lim 1+ = 1+ .
n→∞ n n

Remark 2.27. This proof naturally gives that ∀ n ∈ N,

 n  n+1
1 1
1+ <e< 1+ ,
n n

which we can use to estimate e arbitrarily.

§2.7 Sequences: Euler’s Number e p. 59

 ⊛ Example 2.18
Consider the sequence Sn = k=0 k! . Show that the sequence (Sn ) is Cauchy and
Pn 1

that limn→∞ Sn = e.

2.8 Limit Points

,→ Definition 2.12: Limit Point

Let (xn ) be a sequence. A number x ∈ R is called a limit point or accumulation point
if ∃ a subsequence (xnk ) of (xn ) such that limk→∞ xnk = x. We denote by L the set
of limit points.

Remark 2.28. Note that L could be an empty set; however, if xn bounded, then L ̸= 0 by
Bolzano-Weirestrass Theorem. Further, L is a bounded subset of R.

,→ Proposition 2.14
Let (xn ) be a sequence. Then, x ∈ L iff ∀ ε > 0 the set {n : |xn − x| < ε} is infinite.

Proof. Suppose first that x ∈ L and let (xnk ) be a subsequence such that limk→∞ xnk = x.
Let ε > 0. Then ∃K ∈ N s.t. ∀ k ≥ K, |xnk − x| < ε.

Then, the set

{nk : k ≥ K} ⊆ {n : |xn − x| < ε}.

Since the LHS set is infinite, the RHS must be too.

We now show the converse. Suppose that ∀ ε > 0, the set {n : |xn − x| < ε} is infinite.
Take ε = 1, then the set {n : |xn − x| < 1} is an infinite set. Take n1 = min{n : |xn − x| <
1}. We can now define nk , where k = 2, 3, . . . recursively. Suppose that some nk chosen.
Then, take ε = k1 , and consider

1
{n : n > nk , |xn − x| < }.
k

This set has infinitely many elements, since {n : |xn − x| < k1 } is also infinite. We then
set nk+1 = min{n : n > nk , |xn − x| < k1 }, which defines a strictly increasing sequence
(nk )k≥1 of natural numbers such that for any k, |xnk − x| < k1 . So, limk→∞ |xnk − x| = 0
which gives that limk→∞ xnk = x, so x ∈ L . ■

§2.8 Sequences: Limit Points p. 60

 ,→ Theorem 2.13
Let (xn ) be a bounded sequence. Then,

1. limn→∞ xn = sup L

2. limn→∞ xn = inf L

Remark 2.29. The following proof shows even more, that is, limn→∞ xn = sup L and
limn→∞ xn ∈ L (same for limn→∞ ).

Remark 2.30. 1. =⇒ 2. by changing the sign of xn , as “always”.

Proof. We50 will show first that limn→∞ xn ≥ sup L . 50

Sketch: show double
inequality. First, show that
Let x ∈ L and let (xnk ) be a subsequence such that limk→∞ xnk = x. Let yn = sup{xm : lim sup ≥ sup L , by using
the fact that xn bounded, and
m ≥ n}. We have, then, that yn ≥ xn , and that limn→∞ xn = lim yn , hence ∀ k, ynk ≥ xnk , so yn (the sequence that
converges to lim sup)
and that limn→∞ xn = limk→∞ ynk ((yn ) is a convergent sequence, so any subsequence converges and is ≥ xn ∀ n,
converges to the same limit.) So, we have that and so must be greater than
any subsequence.
To show lim sup ≤ sup L ,
show that lim sup ∈ L , and
limn→∞ xn = lim ynk ≥ lim xnk = x,
k→∞ k→∞ hence must be equal to
sup L . To do this, create to
two subsequences of yn (note
that is, for any x ∈ L , limn→∞ xn ≥ x =⇒ limn→∞ xn upper bounds L . Since sup least - yn NOT xn ) that both
upper bound, it must be that limn→∞ xn ≥ sup L . converge to xn . Note that
these exist since yn
converges. The real proof is in
We now show that limn→∞ xn ≤ sup L ; indeed, we will show that limn→∞ xn ∈ L constructing the sequence of
and thus must be ≤ sup L . We will show this by constructing a subsequence of (xn ) that indices nk such that ynk
“bounds” (so to speak) some
converges to limn→∞ xn . xnk . Then, using squeeze
theorem, xnk → lim sup xn ,
Set n1 = 1. Suppose nk defined. Then, so lim sup xn ∈ L and the
proof is complete.

ynk +1 = sup{xn : n ≥ nk + 1}.

If we consider ynk +1 − k+1

1
, then this number is smaller than ynk +1 and is thus not an upper
bound for the set {xn : n ≥ nk + 1}. Then there exists some nk+1 ≥ nk + 1 > nk such that

1
ynk +1 − ≤ xnk+1 ≤ ynk+1 = sup{xn : n ≥ nk+1 }.
k+1

So, we have constructed a strictly increasing sequence (nk ) of natural numbers such that
∀ k ≥ 1,
1
ynk +1 − ≤ xnk+1 ≤ ynk+1 . ⋆
k+1

§2.8 Sequences: Limit Points p. 61

Consider the subsequences (ynk +1 ) and (ynk+1 ) of (yn ), and a subsequence (xnk+1 ) of (xn ).
Since yn converges, and limn→∞ yn = limn→∞ xn , we have that

lim ynk +1 = limn→∞ xn ; and lim ynk+1 = limn→∞ xn ,

k→∞ k→∞

and so, given this and ⋆, by the The Squeeze Theorem, limk→∞ xnk+1 = limn→∞ xn , and so
limn→∞ xn ∈ L . ■

,→ Corollary 2.3
Let (xn ) be a bounded sequence and α = limn→∞ xn , β = limn→∞ xn . Then, α, β ∈ L
(that is, they are limit points of (xn )), and for any x ∈ L , α ≤ x ≤ β (that is, L is a
closed set).

2.9 Properly Divergent Sequences

,→ Definition 2.13: Properly Divergent Sequences

Let (xn ) be a sequence. We say that (xn ) properly diverges to ∞ if for any R ∃N ∈ N
such that ∀ n ≥ N, xn ≥ α. We write

lim xn = ∞.
n→∞

That is,
( ∀ α ∈ R)(∃N ∈ N)( ∀ n ≥ N )(xn ≥ α).

We analogously say (xn ) diverges to −∞ if ∀ α ∈ R∃N ∈ N s.t. ∀ n ≥ N, xn ≤ α.

⊛ Example 2.19
xn = n properly diverges to ∞; xn = −n properly diverges to −∞.

⊛ Example 2.20
Let C > 1. Then, limn→∞ C n = ∞.

Proof. Write C = 1 + x where x > 0. By Bernoulli’s Inequality, ∀ n ≥ 1,

C n = (1 + x)n ≥ 1 + nx.

§2.9 Sequences: Properly Divergent Sequences p. 62

 Let α ∈ R. If α ≤ 0, then ∀ n, C n > α. If α > 0, let N ∈ N, N ≥ αx . SO, ∀ n ≥ N,
C n ≥ 1 + nx > α and limn→∞ C n = ∞. ■

,→ Proposition 2.15
Let (xn ) be increasing. Then limn→∞ xn = ∞ iff xn not bounded from above.

Proof. ( =⇒ ) Let M ∈ R. Since (xn ) → ∞, ∃N ∈ N s.t. ∀ n ≥ N, xn ≥ M , that is, xn is

unbounded.
( ⇐= ) Suppose xn not bounded from above. Let α ∈ R, then, ∃N s.t. xN > α. (xn )
increasing =⇒ ∀ n ≥ N, xn ≥ xN > α =⇒ limn→∞ xn = ∞. ■

,→ Proposition 2.16
Let (xn ) be decreasing. Then limn→∞ xn = −∞ ⇐⇒ (xn ) not bounded from below.

Remark 2.31. Follows from proposition 2.15.

,→ Proposition 2.17
Let (xn ), (yn ) be sequences such that xn ≤ yn ∀ n. Then

1. limn→∞ xn = ∞ =⇒ limn→∞ yn = ∞

2. limn→∞ yn = −∞ =⇒ limn→∞ xn = −∞

Proof. (Of 1.) Let α ∈ R; since lim xn = ∞, ∃N s.t. ∀ n ≥ N, xn ≥ α =⇒ ∀ n ≥ yn ≥

xn ≥ α =⇒ lim yn = ∞. ■

,→ Proposition 2.18
Let (xn ) be a sequence and c > 0. Then lim xn = ∞ ⇐⇒ lim c · xn = ∞. The
converse follows for c < 0 and → −∞.

Proof. ■

,→ Proposition 2.19
Let (xn ) and (yn ) be strictly positive sequences. Suppose that for some L > 0,

xn
lim = L.
n→∞ yn

§2.9 Sequences: Properly Divergent Sequences p. 63

 Then, lim xn = ∞ ⇐⇒ lim yn = ∞.

Proof. Take ε = L
2
. Then, xn
yn
→ L =⇒ ∃N s.t. ∀ n ≥ N, L − ε < xn
yn
< L + ε ⇐⇒
L− L
2
< xn
yn
< L + L2 . So, ∀ n ≥ N, L2 < xn
yn
< 3L
2
=⇒ L
y
2 n
< xn < 32 Lyn . Hence,
if xn → ∞, it must be that yn → ∞, by the previous inequality. The other side of the
implication follows similarly. ■

,→ Proposition 2.20
Let (xn ), (yn ) be two sequences such that (xn ) is properly divergent and yn bounded.
Then their sum is also diverging.

Proof. ■

⊛ Example 2.21
xn = n, yn = −n
2
. xn → ∞, yn → −∞ while xn + yn → ∞.

,→ Definition 2.14: Limsup (Generalized)

Let (xn ) be a sequence bounded from above. Define, as previously, yn := sup{xk :
k ≥ n}; recall that this sequence is decreasing, and moreover, that limn→∞ yn exists.

This limit is finite, as seen previously, if yn bounded from below. If it is not, yn

diverges and as it is decreasing, lim yn = −∞. Recall that lim sup xn = lim yn , hence
if xn bounded from above, limn→∞ xn exists, and is either a real number, or −∞.

In the case xn not bounded above, then we define limn→∞ xn = ∞. In this way, we
define limn→∞ xn for all sequences, regardless of convergence or boundedness.

,→ Definition 2.15: Liminf (Generalized)

Let (xn ) be a sequence. If (xn ) bounded from below, let zn = inf{xk : k ≥ n}. This
is an increasing sequence. We define limn→∞ xn := lim zn ; lim zn finite ⇐⇒ xn
bounded from above, and infinite otherwise (ie, limn→∞ xn = ∞ ⇐⇒ xn un-
bounded from above).

If xn not bounded from below, then limn→∞ xn := −∞.

,→ Proposition 2.21
Practically all previously proven properties of limsup/liminf hold with these general-
izations:

§2.9 Sequences: Properly Divergent Sequences p. 64

 1. limn→∞ xn ≤ limn→∞ xn , −∞ < x < ∞ ∀ x ∈ R.

2. (xn ) converging or properly diverging and lim xn = a ⇐⇒ limn→∞ xn =

limn→∞ xn = a (noting that a ∈ R ∪ {−∞, ∞}51 )

3. limn→∞ (−xn ) = − limn→∞ xn 52

4. limn→∞ xn = inf{t : {n : xn > t} finite or empty} and limn→∞ xn = sup{t :

{xn < t} finite or empty}.53
53
See: Extended Real Line
53
We take, here, −(−∞) ≡ ∞
,→ Definition 2.16: Limit Set 53
We define inf ∅ = ∞ and
sup ∅ = −∞, as a
The limit set of a sequence (xn ) is the collection of all x ∈ R ∪ {−∞, ∞} s.t. for some convention. Moreover, if a set
subsequence (xnk ) of xn , limk→∞ xnk = x. Then, we have, as before, limn→∞ xn = A is not bounded from below,
then we have inf A = −∞,
sup L , limn→∞ xn = inf L . and if A not bounded from
above, sup A = ∞.

Remark 2.32. Not all concepts defined on convergent/bounded sequences extend easily to
properly divergent sequences. For instance, limn→∞ (xn + yn ) ≤ limn→∞ xn + limn→∞ yn
holds for bounded sequences xn , yn , but does not generally hold if limn→∞ xn = ∞, etc..

3 Functional Limits and Continuity

,→ Definition 3.1: Cluster Point

Let54 A ⊆ R. A point c ∈ R is called a cluster or limit point of A if ∀ ε > 0, ∃x ∈ A,
x ̸= c, s.t. 0 < |x − c| < ε.
54
Read: a point is a cluster
point if there exists points
Remark 3.1. Note that this definition does not require c ∈ A. (other than itself) in the set
that are arbitrarily close
(“epsilon close”) to it.

,→ Proposition 3.1
Let A ⊆ R, c ∈ R. Then, TFAE:

1. c is a cluster point of A

2. ∃ a sequence (xn ) s.t. ∀ xn ∈ A, xn ̸= c, and lim xn = c.

Proof. (1. =⇒ 2.) Let c be a cluster point of A, and take ε = 1

n
in the definition of a cluster
point. Then, by definition, ∃xn ∈ A, xn ̸= c, s.t. 0 < |xn − c| < n1 . This defines a sequence
xn ∈ A, xn ̸= c ∀ n, with the property that ∀ n, |xn − c| < 1
n
. Moreover, this gives, by
definition, that lim xn = c.

§?? Functional Limits and Continuity: p. 65

 (2. =⇒ 1.) Suppose there exists a sequence (xn ) in A, xn ̸= c, such that lim xn = c.
Take ε > 0, and let N be such that ∀ n ≥ N , |xn − c| < ε. Take x = xN ; then, we have
that x ∈ A, x ̸= c, and 0 < |x − c| < ε. By definition, then, c is a cluster point, and the
proof is complete. ■

⊛ Example 3.1
Let A = (0, 1). Then, 0 is a cluster point of A.

Proof. Consider the sequence xn = 1

n+1
. Then, since 0 < (xn ) < 1, xn ∈ A ∀ n,
moreover, xn ̸= 0. Hence, lim xn = 0, hence 0 is a cluster point of A. ■

⊛ Example 3.2
Let A = (0, 1) ∪ {5}. Is 5 a cluster point?

Proof. No; it is impossible to find arbitrarily (ε) close points to 5 in the set; ̸ ∃x ∈
A, x ̸= 5 such that 0 < |x − 5| < ε. Then, the set of all cluster points of A is equal
[0, 1]. ■

⊛ Example 3.3
Let A = { n1 : n ∈ N}. Then, c = 0 is the only cluster point of A.

Proof. We show first c = 0 is indeed a cluster point. Let xn = 1

n
; then, xn ∈
A ∀ n, xn ̸= 0, and moreover, lim xn = 0, hence c = 0 a cluster point of A.

We now show that 0 is the only cluster point of A. ■

⊛ Example 3.4
Let A = Q. Then, the set of cluster points is precisely R.

Proof. Take x ∈ R, ε > 0. Consider the interval (x, x + ε); by density of the
rationals, ∃q ∈ Q s.t. q ∈ (x, x + ε). Hence, ∃q ∈ Q, q ̸= x s.t. 0 < |x − q| < ε,
hence, x a cluster point of A. ■

,→ Definition 3.2: Functional Limits

Let A ⊆ R, f : A → R, and c a cluster point of A. Then, we say that the limit of f at
c is L, denoted
lim f (x) = L,
x→c

§?? Functional Limits and Continuity: p. 66

 if ∀ ε > 0, ∃δ > 0 s.t. ∀ x ∈ A satisfying 0 < |x − c| < δ, we have that |f (x) − L| <
ε.

Remark 3.2. “As x gets closer and closer to c, f (x) gets closer and closer to L”.

sin x
Remark 3.3. The point c may or may not be in A (for instance, limx→0 x
= 1). However,
it must be that c is a cluster point of A; this is what “allows” the arbitrary closeness to L in the
definition of a limit.

Remark 3.4. This definition is often called the “ε − δ” definition of functional limits. Quan-
tified, it states

( ∀ ε > 0)(∃δ > 0)( ∀ x ∈ A)(0 < |x − c| < δ =⇒ |f (x) − L| < ε).

⊛ Example 3.5
Let A = (0, ∞), let f (x) = x1 , x ∈ A, and let c ∈ A. Prove that limx→c f (x) = 1c .

Proof. Note: c a cluster point of A since for ε > 0, x = c + ε

2
∈ A, x ̸= c, 0 <
|x − c| = ε
2
< ε (hence the limit is indeed well-defined).

Fix ε > 0; take δ = min{ 12 c, 21 c2 ε}. Then,

1 1 c−x |x − c| δ
− = = < ,
x c xc |xc| |xc|

if x ∈ A is such that 0 < |x − c| < δ. Since |x − c| < δ, we have that x − c >

−δ =⇒ x > c − δ. We also have, by definition, δ ≤ 21 c, hence, x > 2c . This gives
that |xc|
1
= 1
xc
< c
1
c
= 2
c2
. We thus have that, for 0 < |x − c| < δ, that x
1
− 1
c
< 2
c2
δ.
2
c2 2 c2
But we also have that δ ≤ 2
ε, hence 1
x
− 1
c
< c2 2
ε =⇒ 1
x
− 1
c
< ε. Thus,
limx→c x1 = 1
c
. ■

3.1 Sequential Characterization of Functional Limits

,→ Theorem 3.1
Let A ⊆ R, f : A → R, and let c be a cluster point of A. Then, TFAE:

1. limx→c f (x) = L.

2. For any sequence (xn ) ∈ A, xn ̸= c, such that lim xn = c, we have that the
sequence (f (xn )) converges to L, that is, limn→∞ f (xn ) = L.

§3.1 Functional Limits and Continuity: Sequential Characterization of Functional Limits p. 67

Proof. (1. =⇒ 2.) By 1., ∀ ε > 0, ∃δ > 0 s.t. ∀ x ∈ A, x ̸= c such that 0 < |x − c| < δ,
we have |f (x) − L| < ε. Let (xn ) be a sequence in A s.t. xn ̸= c and limn→∞ xn = c. We
wish to show that limn→∞ f (xn ) = L. Take ε > 0; then, we have that ∃δ > 0 s.t. ∀ x ∈ A
satisfying 0 < |x − c| < δ, we have |f (x) − L| < ε. Fix such a δ; then, since limn→∞ xn =
c, ∃N s.t. ∀ n ≥ N, |xn − c| < δ. Then, it follows from the definition of δ that for n ≥ N ,
|f (xn ) − L| < ε, hence limn→∞ f (xn ) = L, and 2. holds.

(2. =⇒ 1.) Suppose not. Then, ∀ ε > 0, we can find δ > 0 such that ∀ x ∈ A s.t. 0 <
|x − c| < δ we have |f (x) − L| < ε. But then, this means that ∃ε0 > 0 s.t. ∀ δ > 0, ∃x ∈
A, x ̸= C s.t. 0 < |x − c| < δ and |f (x) − L| ≥ ε0 . So, for this ε0 > 0, we can take δ = n1 ,
which gives xn ∈ A, xn ̸= c, such that 0 < |xn − c| < 1
n
and |f (xn ) − L| ≥ ε0 . This gives
us a sequence (xn ) ∈ A, xn ̸= c, such that limn→∞ |xn − c| = 0 =⇒ limn→∞ xn = c, and
|f (xn ) − L| ≥ ε0 ∀ n. But this means that we have a sequence xn s.t. lim xn = c, and the
sequence (f (xn )) does not converge to L. But this contradicts 2.; hence, we have come to
a contradiction, and 1. holds. ■

,→ Proposition 3.2
A functional limit is unique. That is, if f : A → R and c a cluster point of A, if
limx→c f = L an limx→c f = M, L = M .

Proof. (Sequential) Let xn be a sequence in A such that xn ̸= c and limn→∞ xn = c. Then

by the sequential characterization, limx→c f (x) = L =⇒ limn→∞ f (xn ) = L, and
limx→c f (x) = M =⇒ limn→∞ f (xn ) = M . That is, the sequence f (xn ) converges
to both L and M , but the limit of a sequence is unique (if it exists), hence L = M . ■

(ε − δ)55 Take ε = |L−M |

2
, and suppose L ̸= M , hence ε > 0. Since limx→c f (x) = L, ∃δ1 > 55
Note the similarity of this
proof and that which we used
0 s.t. ∀ x ∈ A0 < |x − c| < δ1 , we have |f (x) − L| < ε. Similarly, since limx→c f (x) = to prove limits of sequences
M, ∃δ2 > 0 s.t. ∀ x ∈ A, 0 < |x − c| < δ2 , we have that |f (x) − M | < ε. Take δ = are unique (theorem 2.1).

min{δ1 , δ2 } and let x ∈ A s.t. 0 < |x − c| < δ. Then,

|L − M | = |L − f (x) + f (x) − M | ≤ |L − f (x)| + |f (x) − M |

< ε + ε = 2ε = |L − M |,

which implies |L − M | < |L − M |, a contradiction. Hence, L = M . ■

,→ Theorem 3.2: Algebraic Properties of Functional Limits

Let A ⊆ R, f, g : A → R, and let c be a cluster point of A. Suppose limx→c f (x) = L

§3.1 Functional Limits and Continuity: Sequential Characterization of Functional Limits p. 68

 and limx→c g(x) = M . Then,

1. For any constant k ∈ R, limx→c (k · f (x)) = k · limx→c f (x) = k · L.

2. limx→c (f (x) + g(x)) = L + M

3. limx→c (f (x) · g(x)) = L · M

4. If g(x) ̸= 0 ∀ x ∈ A, and M ̸= 0, limx→c f (x)

g(x)
= L
M
.

Proof. (Of 3.; Sequential) Let xn be sequence in A, xn ̸= c, and limn→∞ xn = c. Then,

limn→∞ f (xn ) = L and limn→∞ g(x) = M . But then, by product rule of converging se-
quences (proposition 2.1), limn→∞ (f (xn )g(xn )) = limn→∞ f (xn ) limn→∞ g(xn ) = L · M .
Moreover, by sequential characterization of functional limits, we have that limx→c (f (x)g(x)) =
L · M. ■

(ε − δ) Since limx→c f (x) = L, if we take ε = 1, we can find δ1 > 0 s.t. ∀ x ∈ A, x ̸=

c, 0 < |x − c| < δ1 , we have that |f (x) − L| < 1. For such an x, we have that |f (x)| =
|f (x) − L + L| ≤ |f (x) − L| + |L| < 1 + L. Take now ε > 0. Since limx→c f (x) = L and
limx→c g(x) = M , we can find δ2 > 0 s.t. ∀ 0 < |x − c| < δ2 , we have that |f (x) − L| <
ε
2(|M |+1)
, |g(x) − M| < ε
2(|L|+1)
. Take now δ = min{δ1 , δ2 }, and let x be s.t. 0 < |x − c| < δ.
Then,

|(f (x) · g(x)) − (L · M )| = |f (x)g(x) − f (x)M + f (x)M − LM |

≤ |f (x)g(x) − f (x)M | + |f (x)M − LM |
= |f (x)||g(x) − M | + |M ||f (x) − L|
< (1 + |L|)|g(x) − M | + |M ||f (x) − L|
ε ε ε ε
< (1 + |L|) + |M | ≤ + =ε
2(|L| + 1) 2(|M | + 1) 2 2

where the fourth line follows directly from δ ≤ δ1 as defined previously. ■

,→ Theorem 3.3: Functional Squeeze Theorem

Let A ⊆ R, f, g, h : A → R, and let c be a cluster point of A. Suppose that for all
x ∈ A, we have that f (x) ≤ g(x) ≤ h(x) , and that limx→c f (x) = limx→c h(x) = L,
then limx→c g(x) = L.

Proof. (Sequential) Let xn be a sequence in A, xn ̸= c such that limn→∞ xn = c. Then, we

have limn→∞ f (xn ) = limx→c f (x) = L and similarly limn→∞ h(xn ) = limx→c h(x) = L.

§3.1 Functional Limits and Continuity: Sequential Characterization of Functional Limits p. 69

Now, we have that ∀ n, f (xn ) ≤ g(xn ) ≤ h(xn ). By the squeeze theorem for sequences,
limn→∞ g(xn ) = L ■.

(ε − δ) Let ε > 0. Since limx→c f (x) = L, ∃δ1 > 0 s.t. ∀ x ∈ A s.t. 0 < |x − c| < δ1 we
have |f (x) − L| < ε. Since limx→c h(x) = L, we have that ∃δ2 > 0 s.t. ∀ x ∈ A s.t. 0 <
|x − c| < δ2 we have |h(x) − L| < ε. Let δ = min{δ1 , δ2 } and let x be such that x ∈ A, 0 <
|x − c| < δ. Then,

−ε < f (x) − L ≤ g(x) − L ≤ h(x) − L < ε,

thus, |g(x) − L| < ε whenever 0 < |x − c| < δ, hence limx→c g(x) = L. ■

Remark 3.5. Note the similarity between the ε − δ proofs above and the proofs of correspond-
ing properties for sequences.

,→ Definition 3.3: Divergence Criterion of a Function

Let f : A → R and let c be a cluster point of A. The following criterion state that the
limit of f at c does not exist:

1. Suppose there exists a sequence xn ∈ A, xn ̸= c, s.t. limn→∞ xn = c,

s.t. limn→∞ f (xn ) does not exist. Then, limx→c f (x) also does not exist.

2. Suppose there exist two sequences xn , yn ∈ A, xn , yn ̸= c, s.t. limn→∞ xn =

limn→∞ yn = c, and the limits limn→∞ f (xn ) and limn→∞ f (yn ) exist, but these
two limits are different, then limx→c f (x) does not exist.

⊛ Example 3.6: f (x) = sin x1

Let A = (0, ∞), f (x) = sin x1 and c = 0. Then, limx→0 sin x1 does not exist.

Proof. (Using divergence criterion 1.) Take xn = (2n+1)

1
π . Then, xn > 0, and
  2

limn→∞ xn = 0 = c. Moreover, f (xn ) = sin x1n = sin (2n + 1)( π2 ) = (−1)n .

This sequence does not converge, and so limx→0 sin x1 does not exist. ■

(Using divergence criterion 2.) Take xn = 1

2nπ
, yn = 1
2nπ+ π2
, noting that limn→∞ xn =
limn→∞ yn = 0. Then, f (xn ) = sin x1n = sin(2nπ) = ∀ n, while f (yn ) = sin y1n =
sin 2nπ + π2 = 1 ∀ n, hence, limn→∞ f (xn ) ̸= limn→∞ f (yn ), and thus limx→c f (x)

does not exist. ■

§3.1 Functional Limits and Continuity: Sequential Characterization of Functional Limits p. 70

 ⊛ Example 3.7: Abbott, 4.2E2
Let xn = 2
πn
, yn = 1
(2+n)π
. Then, we have that both (xn ) → 0 and (yn ) → 0, but
 πn 
f (xn ) = cos = 0 ∀ n; f (yn ) = cos((2 + n)π) = 1 ∀ n,
2

hence, limn→∞ f (xn ) = 1 ̸= limn→∞ f (yn ) = 0, so the limit does not exist.

Consider now limx→0 x cos x1 . Fix ε > 0, and take δ = ε, then, we have that
∀ x s.t. 0 < |x − 0| < δ. Then, we have by properties of cos,

1
x cos ≤ |x| < δ = ε,
x

hence the function converges to 0.

⊛ Example 3.8: Abbott, 4.2E14

Let f : A ⊆ R, f : A → R, c ∈ R be a cluster point of A, and f (x) ≥ 0 ∀ x ∈ A.
Prove that limx→c f (x) = limx→c f (x).
p p

Proof. (Seq’n) Define L := limx→c f (x). Then, we have that ∀ xn ∈ A\{c} s.t. (xn ) →
c, limn→∞ f (xn ) = L. We can write, then,

p p  p 2
L = lim f (xn ) = lim f (xn ) f (xn ) = lim f (xn ) ,
n→∞ n→∞ n→∞

and taking the square root of both sides, we have the desired result. Note that this
√
used the assumption that ∃ limn→∞ xn =⇒ ∃ limn→∞ xn .

(ε − δ)

3.2 Left/Right Limits

,→ Definition 3.4: Left/Right Limits

1. Let A ⊆ R, f : A → R, and suppose that c is a cluster point of the set

A ∩ (c, ∞) = {x ∈ A : x > c}.

§3.2 Functional Limits and Continuity: Left/Right Limits p. 71

 Then we say that a real number L is the right limit of f at c, denoted

lim f (x) = L,
x→c+

if ∀ ε > 0, ∃δ > 0 s.t. ∀ x ∈ A s.t. 0 < x − c < δ =⇒ |f (x) − L| < ε.

2. Let A ⊆ R, f : A → R, and suppose that c is a cluster point of

A ∩ (−∞, c) = {x ∈ A : x < c}.

Then we say that a real number L is the left limit of f at c, denoted

lim f (x) = L,
x→c−

if ∀ ε > 0, ∃δ > 0 s.t. ∀ x ∈ A s.t. − δ < x − c < 0 =⇒ |f (x) − L| < ε.

Remark 3.6. Sometimes, but not always, the right/left endpoints are equivalent to the “usual”
limit.

⊛ Example 3.9: The Heaviside Function


1 x ≥ 0

Let f : R → R defined f (x) = . We have
0 x < 0


lim f (x) = 1; lim f (x) = 0.

x→0+ x→0−

Let ε > 0. Take δ > 0. Then, ∀ x s.t. 0 < x < δ, |f (x) − 1| = |1 − 1| = 0 < ε,
hence limx→0+ f (x) = 0.

,→ Proposition 3.3
Let A ⊆ R, f : A → R, and let c be a cluster point of the sets A ∩ (c, ∞) and
A ∩ (−∞, c). TFAE:

1. limx→c f (x) = L

2. limx→c+ f (x) = limx→c− f (x) = L

Proof. (1. =⇒ 2.) Let ε > 0 and δ s.t. ∀ x ∈ A s.t. 0 < |x − c| < δ =⇒ |f (x) − L| < ε.
Then, we have that |f (x) − L| < ε ⇐= 0 < x − c < δ, and moreover, |f (x) − L| <
ε ⇐= −δ < x − c < 0, that is, limx→c+ f (x) = limx→c− f (x) = L, hence 2. holds.

§3.2 Functional Limits and Continuity: Left/Right Limits p. 72

(2. =⇒ 1.) Let ε > 0. Since limx→c+ f (x) = L, ∃δ1 > − s.t. 0 < x − c < δ1 =⇒
|f (x) − L| < ε. Since limx→c− f (x) = L, ∃δ2 > 0 s.t. ∀ x ∈ A s.t. − δ2 < x − c < 0 =⇒
|f (x) − L| < ε. Take δ = min{δ1 , δ2 }. Then, if 0 < |x − c| < δ, then we have that either
0 < x − c < δ ≤ δ1 , or −δ2 ≤ −δ < x − c < 0. In either case, |f (x) − L| < ε, so
limx→c f (x) = L and 1. holds. ■

,→ Theorem 3.4
Let56 A ⊆ R, f : A → R, and let c be a cluster point of the set A ∩ (c, ∞).Then, TFAE:

1. limx→c+ f (x) = L

2. For any sequence (xn ) ∈ A s.t. xn > c ∀ n, and limn→∞ xn = c, we have that
limn→∞ f (xn ) = L.

Proof. ( =⇒ )

( ⇐= ) ■
56
Abbott, 4.3E1 (Theorem 4.3.2)

3.3 Limits and Infinity

3.3.1 Infinite Limits

,→ Definition 3.5: Infinite Limits

Let A ⊆ R, f : A → R, c a cluster point of A.

1. limx→c f (x) = ∞ if ∀ M ∈ R, ∃δ > 0 s.t. ∀ x ∈ A s.t. 0 < |x − c| < δ, f (x) ≥

M.

2. limx→c f (x) = −∞ if ∀ M ∈ R, ∃δ > 0 s.t. ∀ x ∈ A s.t. 0 < |x − c| <

δ, f (x) ≤ M .

⊛ Example 3.10
Let A = (−∞, 0) ∪ (0, ∞) and let f (x) = 1
x2
. Show that limx→0 f (x) = ∞.

Proof. Let M ∈ R, and take δ = √ 1

. Then, ∀ x ∈ A s.t. 0 < |x| < δ, we have
|M |+1
that
1 1
f (x) = > = |M | + 1 > M,
x2 δ2

§3.3 Functional Limits and Continuity: Limits and Infinity p. 73

 hence the limit holds. ■

⊛ Example 3.11
1. Give a sequential characterization of limx→c f (x) = ∞ and − ∞.

2. Give the definition of right/left hand limits going to infinity, limx→c+ f (x) =
∞ and − ∞, limx→c− f (x) = ∞ and − ∞.

3. Let A = (−∞, 0) ∪ (0, ∞), f (x) = x1 . Show that

lim f (x) = −∞, lim+ f (x) = ∞.

x→0− x→0

,→ Proposition 3.4: Order Properties of Infinite Limits

Let A ⊆ R, f, g : A → R, and suppose f (x) ≤ g(x) ∀ x ∈ A. Let c be a cluster point
of A. Then,

1. limx→c f (x) = ∞ =⇒ limx→c g(x) = ∞

2. limx→c g(x) = −∞ =⇒ limx→c f (x) = −∞

Proof. (1.) Let M ∈ R. Since limx→c f (x) = ∞, ∃δ > 0 s.t. ∀ x ∈ A s.t. 0 < |x − c| <
δ, f (x) ≥ M . But then g(x) ≥ f (x) ∀ x, hence g(x) ≥ M =⇒ limx→c g(x) = ∞. ■

3.3.2 Limits at Infinity

,→ Definition 3.6: Limit at ± Infinity

• Let A ⊆ R. Suppose that for some a ∈ R, (a, ∞) ⊆ A. Let f : A → R. We say
that a real number L is the limit of f at ∞ if ∀ ε > 0 ∃K > a s.t. ∀ x ≥ K, we
have |f (x) − L| < ε.

• Let A ⊆ R and suppose that for some a ∈ R, (−∞, a) ⊆ A. Let f : A → R. We

say that a real number L is the limit of f at −∞ if ∀ ε > 0, ∃K < a s.t. ∀ x ≤
K, |f (x) − L| < ε.

sin x
⊛ Example 3.12: x
at infinity
Let A = (0, ∞) and f (x) = sin x
x
. Prove that limx→∞ f (x) = 0.

§3.3 Functional Limits and Continuity: Limits and Infinity p. 74

 Proof. Let ε > 0. Take K = 2ε . Then, for x ≥ K, |f (x)| = sin x
x
≤ 1
x
≤ 1
K
= ε
2
<
ε. ■

⊛ Example 3.13
“Sequentialize” limits at infinity.

⊛ Example 3.14: Abbott, 4.4E9

Prove that if f : (a, ∞) → R is such that limx→∞ xf (x) = L ∈ R exists,

lim f (x) = 0.
x→∞

Proof.

lim xf (x) = L =⇒ ∀ ε > 0, ∃M s.t. ∀ x ≥ M > 0, |xf (x) − L| < ε

x→∞

=⇒ L − ε < xf (x) < L + ε

L−ε L+ε
=⇒ < f (x) <
x }
| {z x }
| {z
→0 →0
squeeze theorem
=⇒ lim f (x) = 0
x→∞

Noting that we take M > 0 wlog. ■

3.3.3 Infinite Limits at Infinity

,→ Definition 3.7: Infinite Limits at Infinity

3.4 Continuity

,→ Definition 3.8: Continuity

Let A ⊆ R, f : A → R, and c ∈ A. We say f is continuous at c if ∀ ε > 0, ∃δ >
0 s.t. ∀ x ∈ A s.t. |x − c| < δ, we have that |f (x) − f (c)| < ε. Quantified: f contin-
uous at c if

( ∀ ε > 0)(∃δ > 0)( ∀ x ∈ A)(|x − c| < δ =⇒ |f (x) − f (c)| < ε).

If f not continuous at some c, we say f discontinuous at c.

If c ∈ A a cluster point of A, f is continuous at c iff limx→c f (x) = f (c). If c not

§3.4 Functional Limits and Continuity: Continuity p. 75

 a cluster point, continuity at c still defined, while limx→c f (x) not.

,→ Theorem 3.5: Sequential Characterization of Continuity

Let f : A → R, c ∈ A. TFAE:

1. f continuous at c

2. for any sequence (xn ) ∈ A s.t. limn→∞ xn = c, limn→∞ f (xn ) = f (c)

Remark 3.7.

Remark 3.8. This theorem can be directly deduced from sequential characterization of func-
tional limits.

,→ Proposition 3.5: Algebraic Operations

Let f, g : A → R, c ∈ A. Suppose f, g continuous at c. Then:

1. ∀ k ∈ R, kf continuous at c

2. h = f + g continuous at c

3. h = f · g continuous at c

4. If g(x) ̸= 0 ∀ x ∈ A, h = f
g
continuous at c.

Proof. (Of 3.) Let (xn ) ∈ A s.t. limn→∞ xn = c. Since f, g continuous at c, we have that
limn→∞ f (xn ) = f (c) and limn→∞ g(xn ) = g(c). By algebraic properties of limits, then,
limn→∞ f (xn )g(xn ) = f (c)g(c) and so ∀ (xn ) ∈ A s.t. limn→∞ xn = c, we have that
limn→∞ h(xn ) = h(c) where h = f · g and thus h continuous at c. ■

,→ Theorem 3.6: Composition of Functions and Continuity

Let f : A → R, g : B → R be two functions such that

f (A) = {f (x) : x ∈ A} ⊆ B

so that the composite function h(x) = g ◦ f (x) = g(f (x)) is well defined on A.
Suppose c ∈ A such that f continuous at c and g continuous at f (c). Then, h also
continuous at c.

§3.4 Functional Limits and Continuity: Continuity p. 76

Proof. (Using sequential characterization) Let (xn ) ∈ A s.t. limn→∞ xn = c. f continuous
at c, so limn→∞ f (xn ) = f (c). Let (f (xn ))n≥1 is a sequence in B such that limn→∞ f (xn ) =
f (c) and so g is continuous at f (c). Then, limn→∞ g(f (xn )) = g(f (c)) so ∀ (xn ) ∈
A s.t. limn→∞ xn = c. We thus have that limn→∞ h(xn ) = h(c) where h = g ◦ f , hence h
is continuous at c.

(ε − δ) Fix ε > 0. Since g is continuous at f (c), ∃δ ′ > 0 s.t. ∀ y ∈ B s.t. |y − f (c)| < δ,
|g(y) − g(f (c))| < ε.

Since f continuous at c, ∃δ ′ > 0 s.t. ∀ x ∈ A s.t. |x − c| < δ ′ , |f (x) − f (c)| < δ ′ . Then,
for such x, |g(f (x)) − g(f (c))| < ε and the proof is complete, taking h = g ◦ f . ■

⊛ Example 3.15
f : R → R, f (x) = x.

Proof. ■

⊛ Example 3.16: f (x) = sin x

Show that f (x) = sin x continuous on any c ∈ R.

Proof. Fix ε > 0. Take δ = ε, and take x s.t. |x − c| < δ. Then,

  
x−c x+c
|sin x − sin c| = 2 sin cos
2 2
   
x−c x+c
= 2 sin cos
2 2
x−c
≤2 = |x − c| < δ = ε.
2

⊛ Example 3.17: Dirichlet Function


1 x ∈ Q

Let f : R → R, x 7→ . Show f discontinuous ∀ c ∈ R.
0 x ∈ R \ Q


Proof. Fix c ∈ R and let n ∈ N. Consider the interval (c − n1 , c + n1 ). By density of

the rationals, there must exist some xn ∈ Q s.t. xn ∈ (c − n1 , c + n1 ), and similarly,
by density of the irrationals, there must exists some yn ∈ J, yn ∈ (c − n1 , c + n1 ).

§3.4 Functional Limits and Continuity: Continuity p. 77

 We have, then,
1 1
|xn − c| < and |yn − c| < ,
n n
and moreover, f (xn ) = 1 and f (yn ) = 0 ∀ n. We also have limn→∞ xn = limn→∞ yn =
c, and so f cannot be continuous. ■

⊛ Example 3.18: Thomae’s Function


0 x ∈ J

Let f : R → R, x 7→
+
.
 1 x = m ∈ Q, gcd(m, n) = 1

n n

Show f discontinuous for any a ∈ Q and continuous for any a ∈ J.

Proof. Let a > 0 be rational. Then, f (a) > 0, by construction of the function. Let
(xn ) ∈ J s.t. (xn ) → a. Then, limn→∞ f (xn ) = 0, despite f (a) > 0, hence f is not
continuous at a. ■

3.4.1 Extensions By Continuity

,→ Definition 3.9: Extension by Continuity

Let f : A → R, c a cluster point of A s.t. c ∈
/ A. Since c ∈
/ A, we cannot say whether
f continuous or not at a, but we can extend f to A ∪ {c} by setting

f (x) x ∈ A

F (x) := .
L

x=c

Remark 3.9. Since c a cluster point of A∪{c}, we have that F continuous at c iff limx→c F (x) =
L ⇐⇒ limx→c f (x) = L. Hence, if f : A → R, c a cluster point of A, c ∈
/ A, and
limx→c = L, F is continuous at c. If limx→c f (x) DNE, f cannot be extended.

⊛ Example 3.19
f (x) = x sin x1 , defined on A = (−∞, 0) ∪ (0, ∞). 0 a cluster point of A. Note that
limx→0 f (x) = 0, since |f (x)| ≤ |x|, so if we extend f to 0 by setting f (0) = 0,
then the extended function is continuous at 0.

3.5 Continuity on Bounded & Closed Interval

§3.5 Functional Limits and Continuity: Continuity on Bounded & Closed Interval p. 78
 ,→ Definition 3.10: Bounded Function
Let A be a set. A function f : A → R is called bounded if ∃M > 0 s.t. |f (x)| ≤
M ∀ x ∈ A.

,→ Theorem 3.7: Closed Domain & Continuous Implies Bounded

Let f : [a, b] → R be a continuous function. Then, f is bounded.

Proof. We proceed by contradiction. Suppose there exists a continuous function f : [a, b] →

R that is not bound. Then, for any n ∈ N, it is not true that |f (x)| ≤ n ∀ x ∈ [a, b]
(otherwise, this n would be a bound).

So, for any n, ∃xn ∈ [a, b] s.t. |f (xn )| > n. Then, (xn ) is a sequence in [a, b] and by
the Bolzano-Weirestrass Theorem, this sequence has a subsequence (xnk ) that converges to
some x ∈ [a, b], that is,
lim xnk = x.
k→∞

So, by the sequential characterization of continuity, we have then that

lim f (xnk ) = f (x).

k→∞

Hence, (f (xnk )) is a converging sequence of real numbers. But by the construction of xn ,

we have
|f (xnk )| > nk ≥ k,

so (f (xnk )) is a converging sequence of real numbers that is not bounded, which contradicts
the fact that any converging sequence is bounded. ■

⊛ Example 3.20: “Not Closed” Domain

Consider the function f : (0, 1] → R, x 7→ x1 . This function is continuous, and the
interval (0, 1] is bounded but not closed. Hence, the function is not bounded on this
interval; for any M > 0, if we take x ∈ (0, 1] such that 0 < x < 1
M +1
, we have that
f (x) = 1
x
> M + 1 > M.

,→ Definition 3.11: Absolute Max/Min

Let f : A → R. We say that f has absolute maximum at x ∈ A if f (x) ≥ f (x) ∀ x ∈ A.
f has absolute minimum at x ∈ A if f (x) ≤ f (x) ∀ x ∈ A.

§3.5 Functional Limits and Continuity: Continuity on Bounded & Closed Interval p. 79
 ,→ Theorem 3.8
Let f : [a, b] → R be a continuous function. Then, f has an absolute maximum and
absolute minimum on [a, b].

Proof. (of absolute maximum) Consider the set 56

Absolute minimum case
follows by taking −f .

f ([a, b]) = {f (x) : x ∈ [a, b]}.

By theorem 3.7, f is bounded on [a, b], so there exists some M > 0 such that

f ([a, b]) ⊆ [−M, M ].

So, the set f ([a, b]) is bounded, and by Axiom Of Completeness, s = sup(f ([a, b])) exists.
Hence, s ≥ f (x) ∀ x ∈ [a, b]. We aim to show then that ∃x ∈ [a, b] s.t. s = f (x).

Let n ∈ N. Since s − 1
n
is not an upper bound of f ([a, b]),

1
∃xn ∈ [a, b] s.t. s − < f (xn ) ≤ s. ⊛
n

By Bolzano-Weirestrass Theorem, (xn ) has a converging subsequence (xnk ). Let x = limk→∞ xnk .
By the sequential characterization of continuity, then, we have that f (x) = limk→∞ f (xnk ).
By ⊛, we have
1
s− < f (xnk ) ≤ s.
nk
Moreover, we have that nk ≥ k =⇒ 1
nk
≤ 1
k
=⇒ − n1k ≥ − k1 . Hence,

1
s− < f (xnk ) ≤ s,
k

and so by the squeeze theorem, limk→∞ f (xnk ) = s = f (x), and the proof is complete. ■

,→ Theorem 3.9: Location of the Roots

Let f : [a, b] → R be a continuous function such that

f (a) < 0 < f (b).

Then, ∃c s.t. a < c < b and f (c) = 0.

Proof. Let S = {x ∈ [a, b] : f (x) ≤ 0}. S ̸= ∅ since a ∈ S. S also bounded (it is a subset
of a bounded interval). Let c = sup S (exists by AC). We claim this c is the point as defined

§3.5 Functional Limits and Continuity: Continuity on Bounded & Closed Interval p. 80
in the theorem; we aim to show that f (c) = 0.

Let ε = min{ |f (a)|

2
, f (b)
2
} > 0. Since f continuous at a, ∃δ ′ > 0 s.t. ∀ x ∈ [a, b] s.t. |x − a| <
δ ′ , we have |f (x) − f (a)| < ε. Since f is continuous at b, ∃δ ′′ > 0 s.t. ∀ x ∈ [a, b] s.t. |x − b| <
δ ′′ , we have |f (x) − f (b)| < ε. Let δ = min{δ ′ , δ ′′ , b−a
2
}. Then, ∀ x ∈ [a, a + δ), we have
that
|f (a)| |f (a)| f (a)
f (x) − f (a) < ε ≤ =⇒ f (x) < + f (a) = < 0.
2 2 2
So, ∀ x ∈ [a, a + δ), f (x) < 0 and thus [a, a + δ) ⊆ S. Hence, c, being the supremum of S,
must have that c ≥ a + δ > 0.

Since δ ≤ δ ′′ , we have that ∀ x ∈ (b − δ, b],

f (b) f (b) f (b)

f (x) − f (b) > −ε ≥ − =⇒ f (x) > f (b) − = > 0.
2 2 2

So, ∀ x ∈ (b − δ, b], we have that f (x) > 0. So, if we take [b − 2δ , b], then for every x ∈ this
interval f (x) > 0 and so S ⊆ [a, b − 2δ ), and thus c = sup S ≤ b − 2δ . Hence, ∃δ such that
a + δ ≤ c ≤ b − 2δ . So, c satisfies a < c < b.

We now show that f (c) ≤ 0 and f (c) ≥ 0 and so f (c) = 0.

(f (c) ≤ 0) Let n ∈ N and consider c − n1 ; this is not an upper bound of S, so ∃(xn ) ∈

S s.t. c − 1
n
< xn ≤ c. This gives us a sequence such that limn→∞ (xn ) = c. Since f is
continuous, by the sequential characterization of continuity, we have that limn→∞ f (xn ) =
f (c). Moreover, xn ∈ S and thus f (xn ) ≤ 0 (by construction of S), hence f (c) ≤ 0.

(f (c) ≥ 0) Since c < b, we can find (xn ) ∈ [a, b] s.t. xn > c, limn→∞ xn = c (xn = c + n1 ,
for instance). We must have that f (xn ) > 0; otherwise, f (xn ) ≤ 0 =⇒ xn ∈ S, and since
we have xn > c (by construction), this would contradict the fact that c an upper bound for
S. So, we have that limn→∞ xn = c =⇒ limn→∞ f (xn ) = f (c) ≥ 0, that is, f (c) ≥ 0.

Thus, having show both f (c) ≤ 0 and f (c) ≥ 0, we conclude that ∃c ∈ [a, b] s.t. a <
c < b, where f (c) = 0, and the proof is complete. ■

3.6 Intervals in R

,→ Definition 3.12: Types of Intervals in R

(Bounded Intervals)

• [a, b] = {x : a ≤ x ≤ b} ⊆ R

§3.6 Functional Limits and Continuity: Intervals in R p. 81

 • (a, b) = {x : a < x < b} ⊆ R

• [a, b) = {x : a ≤ x < b} ⊆ R

• (a, b] = {x : a < x ≤ b} ⊆ R

(Unbounded Intervals)

• [a, ∞) = {x : x ≥ a}

• (a, ∞) = {x : x > a}

• (−∞, a] = {x : x ≤ a}

• (−∞, a) = {x : x < a}

• R = (−∞, ∞)

Remark 3.10. If you take any interval and any two points x < y in the interval, then [x, y]
is completely contained within the given interval.

,→ Theorem 3.10
Let S ⊆ R that contains more than two points. Suppose S has the property that
∀ x, y ∈ S s.t. x < y, [x, y] ⊆ S. Then, S is an interval.

Proof. Suppose S bounded. Then, a = inf S, b = sup S exist. Then, for any x ∈ S, a ≤
x ≤ b, so S ⊆ [a, b]. Let now a < z < b. z < b =⇒ z not an upper bound of S so
∃y ∈ S s.t. z < y. z > a =⇒ z not a lower bound of S so ∃x ∈ S s.t. x < z. Then,
x < z < y, x, y ∈ S, so [x, y] ⊆ S =⇒ z ∈ S. So, (a, b) ⊆ S ⊆ [a, b] and thus S must be
a bounded interval (one of those types defined above). ■

,→ Theorem 3.11: Bolzano’s Intermediate Value Theorem

Let I be an interval and f : I → R a continuous function. Let a, b ∈ I and suppose
f (a) < f (b). Then, for any k s.t. f (a) < k < f (b), ∃c between a and b s.t. f (c) = k.

Proof. • (Case 1 : a < b) Consider h(x) = f (x) − k on the closed and bounded interval
[a, b]. Note that h(a) = f (a) − k < 0, and h(b) = f (b) − k > 0. By Location of the
Roots, there exists a a < c < b s.t. h(c) = 0 = f (c) − k =⇒ f (c) = k, as desired.

• (Case 2: a > b) Consider h(x) = k − f (x) on the closed and bounded interval [b, a].
The remainder of the proof follows identically to (Case 1).

§3.6 Functional Limits and Continuity: Intervals in R p. 82

 ■

,→ Theorem 3.12
Let I = [a, b] and f : I → R a continuous function. Let k be s.t. inf f (I) ≤ k ≤
sup f (I). Then, ∃c ∈ I s.t. f (c) = k.

Proof. Recall that m = inf f (I) is the absolute minimum of f on I and M = sup f (I) is
the absolute maximum of f on I. Moreover, ∃x, x ∈ I s.t. f (x) = M, f (x) = m. Hence,
we have that our k satisfies
f (x) ≤ k ≤ f (x).

If k = f (x), take c = x. If k = f (x), take c = x. Otherwise, the inequality is strict, and

we have f (x) < k < f (x). By Bolzano’s Intermediate Value Theorem, we have that ∃c
between x and x s.t. f (c) = k. Moreover, c ∈ [a, b]. ■

,→ Theorem 3.13
Let I = [a, b] and f : I → R a continuous function. Then, f (I) is also a bounded and
closed interval.

Proof. Let m = inf f (I), M = sup f (I). Then, for any x ∈ [a, b], m ≤ f (x) ≤ M , hence,
f (I) ⊆ [m, M ]. OTOH, by theorem 3.12, for any m ≤ k ≤ M , ∃c ∈ [a, b] s.t. f (c) = k,
hence, [m, M ] ⊆ f (I), and thus f (I) = [m, M ] and the proof is complete. Moreover, f (I)
is precisely [inf f (I), sup f (I)]. ■

,→ Theorem 3.14
Let I be an interval in R. Let f : I → R be a continuous function. Then, f (I) is also
an interval.

Proof. We assume f is not just a constant function. We aim to show that if α, β ∈ f (I), α <
β, then [α, β] ⊆ f (I), that is, f (I) an interval.

Let a, b ∈ I be such that f (a) = α, f (b) = β. We have that f (a) < f (b), so for any
k s.t. α ≤ k ≤ β, by Bolzano’s Intermediate Value Theorem, ∃c ∈ I s.t. f (c) = k. Hence,
[α, β] ⊆ f (I), and the proof is complete. ■

Remark 3.11. This argument does not specify the actual “shape” of the intervals f (I) look
like.

§3.6 Functional Limits and Continuity: Intervals in R p. 83

 • If I = R, can f (I) be bounded and closed? Yes; take f (x) = sin x; then, f (R) = [−1, 1]

• If I = (a, b), can f (I) = R? Yes; take I = (− π2 , π2 ), f (x) = tan x.

3.7 Uniform Continuity

Remark 3.12. Recall that in the definition of continuity, the “choice” of δ depended both on
c (the point in the domain) and ε. Uniform continuity defines a manner in which δ can be
chosen without relying on c; if this is the case for a function f : A → R, we say that f is
uniformly continuous on A.

,→ Definition 3.13: Uniform Continuity

Let f : A → R. We say f is uniformly continuous on A if

( ∀ ε > 0)(∃δ > 0)( ∀ c ∈ A)( ∀ x ∈ A)(|x − c| < δ =⇒ |f (x) − f (c)| < ε).

Remark 3.13. The difference, quantifiers-wise, is the position of the ( ∀ c ∈ A); since here the
“choice” of c comes after the choice of δ, δ is independent, in contrast with “local” continuity.

⊛ Example 3.21
Let f (x) = x. Then, f is uniformly continuous on R.

Proof. Let ε > 0, δ = ε. Then, ∀ c ∈ R, if x s.t. |x − c| < δ, then we have

|f (x) − f (c)| = |x − c| < ε. ■

⊛ Example 3.22
Let f (x) = x2 . Then, f is not uniformly continuous on R.

Proof. We proceed by contradiction. Suppose f uniformly continuous. Take ε = 1,

then, ∃δ > 0 s.t. ∀ x, c ∈ A s.t. |x − c| < δ, |f (x) − f (c)| = |x2 − c2 | < 1. Take
x= 1
δ
+ δ, c = 1
δ
+ 2δ . Then, we have

δ
|x − c| = < δ,
2

but

1 1 δ
x2 − c2 = ( + δ)2 − ( + )2
δ δ 2
3δ 2
= ··· = 1 + > 1,
4

§3.7 Functional Limits and Continuity: Uniform Continuity p. 84

 a contradiction. ■

⊛ Example 3.23
√
Let f (x) = x. Then, f is uniformly continuous on [0, ∞).

ε2
Proof. Let ε > 0. Take δ = 2
. Let x, c ≥ 0, and suppose |x − c| < δ. We consider
two cases:

2
• (Case 1) x, c ∈ [0, ε4 ). Then,

√ √ √ √
x− c ≤ x+ c
r r
ε2 ε2 ε
< + = 2 · = ε.
4 4 2

ε2
• (Case 2) Either x or c ≥ 4
. then
√ √
√ √ √ √ x+ c |x − c|
x − c = ( x − c) √ √ =√ √
x+ c x+ c
ε2
δ 2
< ε = ε = ε.
2 2

⊛ Example 3.24
Let f (x) = sin x1 is not uniformly continuous on (0, 1].

Proof. Suppose that f is indeed uniformly continuous. Take ε = 1

2
. Then, ∃δ >
0 s.t. ∀ x, c ∈ (0, 1] s.t. |x − c| < δ, |f (x) − f (c)| = |sin x − sin c| < 21 . Take n ∈
N s.t. 1
n
< δ. Take x = 1
nπ
,c = 1
(2n+1) π2
= 1
nπ+ π2
. Then,

π
1 1 2
|x − c| = − π = <δ
nπ nπ + 2
nπ(nπ + π2 )

Then, we have

1 1
f (x) − f (c) = sin 1 − sin 1
nπ (2n+1) π2
1
= |(−1)n | = 1 > ,
2

a contradiction. ■

§3.7 Functional Limits and Continuity: Uniform Continuity p. 85

3.8 Sequential Characterization of Non-Uniform Continuity

,→ Theorem 3.15
Let f : A → R be a continuous function. TFAE:

1. f is not uniformly continuous on A;

2. ∃ε0 > 0 and two sequences (xn ), (yn ) ∈ A s.t. limn→∞ (xn − yn ) = 0 and
|f (xn ) − f (yn )| ≥ ε0 ∀ n.

Proof. (1. =⇒ 2.) For f to be not uniformly continuous, then it is not true that ∀ ε >
0∃δ > 0 s.t. ∀ x, y ∈ A, if |x − y| < δ, |f (x) − f (y)| < ε. That is, ∃ε0 > 0 s.t. ∀ δ > 0, one
can find x, y ∈ A s.t. |x − y| < δ and |f (x) − f (y)| ≥ ε0 .

Take this ε0 and let δ = n1 . Then, ∃xn , yn ∈ A s.t. |xn − yn | < 1

n
and |f (xn ) − f (yn )| ≥
ε0 . This defines sequences (xn ), (yn ) ∈ A s.t. limn→∞ (xn − yn ) = 0 and |f (xn ) − f (yn )| ≥
ε0 ∀ n, hence 2. holds.

(2. =⇒ 1.) We argue by contradiction. Suppose there ∃f continuous, f : [a, b] → R s.t.

2. holds but 1. does not; that is, f uniformly continuous and ∃ε0 > 0 and (xn ), (yn ) ∈
A s.t. lim(xn − yn ) = 0 and |f (xn ) − f (yn )| ≥ ε0 ∀ n.

Take this ε0 in the definition of uniform continuity; then, if f uniformly continuous,

∃δ > 0 s.t. ∀ x, y ∈ A s.t. |x − y| < δ, we have |f (x) − f (y)| < ε0 . Consider our
(xn ), (yn ). Since limn→∞ (xn − yn ) = 0, ∃N s.t. ∀ n ≥ N, |xn − yn | < δ. But then, this
implies that ∀ n ≥ N , we have that |f (xn ) − f (yn )| < ε0 . But this contradicts our original
assumption in 2., and hence 1. must hold and the proof is complete. ■

⊛ Example 3.25: f (x) = x2

Show that f (x) = x2 not uniformly continuous on R.

Proof. Take xn = n + n1 , yn = n. Then, xn − yn = 1

n
hence limn→∞ (xn − yn ) = 0.
OTOH,

1 1
f (xn ) − f (yn ) = n2 + 2 + 2
− n2 = 2 + 2 ≥ 2,
n n

hence, by the sequential characterization, taking ε0 = 2, f is not uniformly contin-

uous on R. ■

§3.8 Functional Limits and Continuity: Sequential Characterization of Non-Uniform Continuity p. 86

 ⊛ Example 3.26: f (x) = sin x1
Show that f (x) = sin x1 not uniformly continuous on (0, 1]

Proof. Let xn = 1
nπ+ π2
, yn = 1
nπ
. Both of these converge to 0, hence their differences
do as well. OTOH, |f (xn ) − f (yn )| = |−1 − 0| = 1 ≥ 1, hence, f is not uniformly
continuous with ε0 = 1.s ■

,→ Theorem 3.16
Let f : [a, b] → R be a continuous function. Then, f is uniformly continuous on [a, b].

Proof. We proceed by contradiction. Suppose ∃f : [a, b] → R that is continuous but not

uniformly continuous on [a, b]. Then, by the sequential characterization, ∃ε0 > 0 and
xn , yn ∈ [a, b] s.t. limn→∞ (xn − yn ) = 0 and |f (xn ) − f (yn )| ≥ ε0 ∀ n.

Since [a, b] bounded, by Bolzano-Weirestrass Theorem, the sequence (xn ) has a conver-
gent subsequence (xnk ) that converges to z ∈ [a, b] (since [a, b] closed). We can write

|ynk − z| = |ynk − xnk + xnk + z|

≤ |ynk − xnk | + |xnk − z| .
| {z } | {z }
→0 →0

Hence, by the The Squeeze Theorem, |ynk − z| converges to 0 so (ynk ) also converges to z,
that is
lim xnk = lim ynk = z.
k→∞ k→∞

Since f continuous on [a, b], we have that limk→∞ f (xnk ) = f (z) and limk→∞ f (ynk ) =
f (z), and so
lim (f (xnk ) − f (ynk )) = z − z = 0,
k→∞

By definition, then, ∃K s.t. ∀ k ≥ K, |f (xnk ) − f (ynk )| < ε0 . This is a contradiction,

hence, f uniformly continuous and the proof is complete. ■

,→ Theorem 3.17: Preservation of Cauchy Criterion by Uniformly Continuous

Functions
Let f : A → R be a uniformly continuous function. Let (xn ) ∈ A, and assume xn
Cauchy. Then, (f (xn )) is also a Cauchy sequence.

Proof. Let ε > 0. Since f uniformly continuous on A, there is δ > 0 s.t. ∀ x, y ∈ A, |x − y| <

§3.8 Functional Limits and Continuity: Sequential Characterization of Non-Uniform Continuity p. 87

δ =⇒ |f (x) − f (y)| < ε. Since (xn ) Cauchy, ∃N s.t. ∀ n, m ≥ N , |xn − xm | < δ. But
then, ∀ n, m ≥ N, we have |f (xn ) − f (yn )| < ε, and hence (f (xn )) is Cauchy. ■

,→ Theorem 3.18: Continuous Extension Theorem

Let (a, b) be an bounded, open interval and f : (a, b) → R a continuous function.
TFAE:

1. f is uniformly continuous on (a, b);

2. f can be at the end points a, b such that it is continuous on the closed interval
[a, b].

Proof. (2. =⇒ 1.) If f can be extended to a, b so that it is continuous on [a, b], then it is
also uniformly continuous by theorem 3.16. Then, f is also uniformly continuous on any
subset [a, b], in particular, on (a, b) ⊆ [a, b].

(1. =⇒ 2.) Let (xn ) ∈ (a, b) that converges to a. (xn ) Cauchy, and by theorem 3.17, (f (xn ))
also Cauchy, hence L = limn→∞ f (xn ) exists; define (“extend”) f (a) = L. It remains to
show that f continuous with this extension.

Let (un ) be an arbitrary sequence in (a, b) such that limn→∞ un = a. Let ε > 0. Since
f is uniformly continuous on (a, b), then ∃δ > 0 s.t. ∀ x, y ∈ (a, b), |x − y| < δ =⇒
|f (x) − f (y)| < 2ε . Now, we have that limn→∞ un = a, limn→∞ xn = a, so limn→∞ (un −
xn ) = 0. Hence, ∃N1 s.t. ∀ n ≥ N1 , |un − xn | < δ. This implies, then, that ∀ n ≥ N1 , we
have |f (un ) − f (xn )| < 2ε .

We have, by our extension, that limn→∞ f (xn ) = L, hence, ∃N2 s.t. ∀ n ≥ N2 , |f (xn ) − L| <
ε
2
. Let, now, N = max{N1 , N2 }. Then, ∀ n ≥ N ,

|f (un ) − L| = |f (un ) − f (xn ) + f (xn ) − L|

≤ |f (un ) − f (xn )| + |f (xn ) − L|
ε ε
< + = ε,
2 2

that is, ∀ n ≥ N, |f (un ) − L| < ε. Hence, for any arbitrary (un ) ∈ (a, b) such that (un ) →
a, lim n → ∞un = L, hence, as we have set f (a) = L, by sequential characterization of
continuity, f is continuous at a.

The proof for b, the RHS endpoint, follows identically. ■

§3.8 Functional Limits and Continuity: Sequential Characterization of Non-Uniform Continuity p. 88

3.9 Monotone and Inverse Functions

,→ Definition 3.14: Increasing/Decreasing Function

Let f : A → R. We say f is:

• increasing on A if ∀ x, y ∈ A, x ≤ y =⇒ f (x) ≤ f (y);

• strictly increasing on A if ∀ x, y ∈ A, x < y =⇒ f (x) < f (y);

• decreasing on A if ∀ x, y ∈ A, x ≤ y =⇒ f (x) ≥ f (x);

• strictly decreasing on A if ∀ x, y ∈ A, x < y =⇒ f (x) > f (y).

A function that is either increasing or decreasing is called monotone. If this increasing

or decreasing is strict, the function is called strictly monotone.

,→ Proposition 3.6
f : A → R increasing on A ⇐⇒ g = −f decreasing on A.

Remark 3.14. Analogous statements hold for decreasing/strictly increasing/decreasing etc.

The remaining theorems/propositions will be discussed with respect to increasing functions;
the same concepts apply (with reversed inequalities, etc) to decreasing functions.

,→ Theorem 3.19
Let I ⊆ R, f : I → R be increasing. Let c ∈ I, where c not an endpoint of I. Then:

1. limc→c− f (x) = sup{f (x) : x ∈ I, x < c}

2. limc→c+ f (x) = inf{f (x) : x ∈ I, x > c}

Proof. We prove for 2.; 1. follows identically. Let A := {f (x) : x ∈ I, x > c}. Note that
A ̸= ∅, since c not an endpoint of I by construction and hence ∃x ∈ I s.t. x > c.

Since f increasing, we have that x > c =⇒ f (x) ≥ f (c) hence A bounded below by
f (c), and thus L := inf A exists. Let ε > 0; since L + ε not a lower bound for A, there
exists some xε ∈ I s.t. L + ε > f (xε ) ≥ L. Take δ = xε − c. Since f increasing, we have
that
c < x < c + δ = xε =⇒ |f (x) − L| = f (x) − L ≤ f (xε) − L < ε.

§3.9 Functional Limits and Continuity: Monotone and Inverse Functions p. 89

But this is just the definition of the right hand limit, hence

lim f (x) = L.
x→c+

,→ Corollary 3.1
Let I ⊆ R and f : I → R be increasing on I. Take c ∈ I such that c not an endpoint
of I. TFAE:

1. f continuous at c

2. limx→c− f (x) = f (c) = limx→c+ f (x)

3. sup{f (x) : x ∈ I, x < c} = f (c) = inf{f (x) : x ∈ I, x > c}.

Proof. Note first that 1. ⇐⇒ 2. does not relate to f increasing; rather, it follows from the
left-hand limit equals right-hand limit iff limit holds; this holds if f continuous at c.

2. ⇐⇒ 3. follows from theorem 3.19. ■

,→ Definition 3.15: Jump

Let f : I → R be increasing on I. If c ∈ I not an endpoint of I, the jump of f at c is
defined
jf (c) = lim+ f (x) − lim− f (x).
x→c x→c

If c the left endpoint of I, then we define

jf (c) = lim+ f (x) − f (c),

x→c

and if c the right endpoint of I,

jf (c) = f (c) − lim− f (x).

x→c

Remark 3.15. It follows naturally that f continuous at c ∈ I ⇐⇒ jf (c) = 0.

,→ Theorem 3.20
Let I ⊆ R be an interval and f : I → R be increasing. Then the set D ⊆ I of points
at which f is discontinuous is either finite or countable.

§3.9 Functional Limits and Continuity: Monotone and Inverse Functions p. 90

Proof. We will prove this result in the case that I = [a, b], and deduce the remaining cases.

Note first that jf (c) ≥ 0 ∀ c ∈ I. Consider some n points in I,

a ≤ x1 < x2 < · · · xn ≤ b.

We claim that the following inequality holds:

jf (x1 ) + jf (x2 ) + · · · jf (xn ) ≤ f (b) − f (a).

Indeed, we have that

jf (x1 ) + · · · + jf (xn ) = lim+ f (x) − lim− f (x) + · · · lim+ f (x) − lim− f (x)
x→x1 x→x1 x→xn x→xn
n−1
!
X
= lim+ f (x) − lim− f (x) + lim f (x) − lim
−
f (x)
x→xn x→x1 x→x+
k x→xk+1
k=1
| {z }
≤0

≤ lim+ f (x) − lim− f (x)

x→xk x→x1

≤ f (b) − f (a) ⊛

From this, we have that for any k ∈ N, there are at most k points in I such that jf (x) ≥
f (b)−f (a)
k
; suppose there were k + 1 points; then,

k+1
f (b) − f (a) ≥ jf (x1 ) + · · · + jf (xk+1 ) ≥ (f (b − f (a))) > f (b) − f (a)⊥.
k
S∞
Let D := {x ∈ I : f discontinuous at x} = {x ∈ I : jf (x) > 0} = k=1 {x ∈ I : jf (x) ≥
f (b)−f (a)
k
}. This is a countable union of finite sets, hence D itself is finite or countable given
I = [a, b].

We now prove for general I. Any interval I can be written as

∞
[
I= [ak , bk ],
k=1

for some sequences an , bn , that is, as a countable union of bounded and closed intervals ⊖.
Hence, we can write our set DI defined above as

[
DI = DS[an ,bn ] = D[an ,bn ] ,

which is again a union of finite/countable sets, and the proof is complete. ■

§3.9 Functional Limits and Continuity: Monotone and Inverse Functions p. 91

Remark 3.16. To be more explicit about the statement ⊖:

S∞
• R= k=1 [−k, k]

S∞
• (a, b) = k=1 [a + b−a
3k
,b − b−a
3k
]
S∞
• (−∞, b] = k=1 [−k − |b|, b]
S∞
• (−∞, b) = k=1 [−k − |b|, b − 1
2k
]

• ...

3.10 Continuous Inverse Theorem

,→ Theorem 3.21
Let I ⊆ R be an interval and let f : I → R be a continuous function. Let S :=
f (I). Suppose f strictly increasing. Then, for any y ∈ S, there is precisely one
x ∈ I s.t. f (x) = y.

Proof. Suppose x1 , x2 s.t. f (x1 ) = f (x2 ) = y. f strictly increasing, hence both x1 > x2
and x1 < x2 are impossible, hence x1 = x2 . ■

,→ Definition 3.16: Inverse

et I ⊆ R be an interval and let f : I → R be a continuous function. Let S := f (I).
∀ y ∈ S, we set g(y) = x ∈ I s.t. f (x) = y. This defines a function g : S →
I s.t. g(S) = I. This gives

(f ◦ g)(y) = y ∀ y ∈ S; (g ◦ f )(x) = x ∀ x ∈ I.

g is call the inverse of f ; we often denote g = f −1 .

,→ Proposition 3.7
If f strictly increasing, so is f −1 .

,→ Theorem 3.22: Continuous Inverse Theorem

Let I ⊆ R be an interval, and let f : I → R be strictly increasing and continuous.
Then, g = f −1 is also strictly increasing and continuous, on S = f (I).

§3.10 Functional Limits and Continuity: Continuous Inverse Theorem p. 92

Proof. We show only continuous. Suppose g not continuous at some point c ∈ S; assume
c not an endpoint, for now. Since g not continuous at c, we have that

jg (c) = lim+ g(y) − lim− g(y) > 0.

y→c y→c

Let x ∈ I s.t. x ̸= g(c) and s.t.

lim g(y) < x < lim+ g(y).

y→c− y→c

Then, there is no y ∈ S s.t. g(y) = x, by our construction. But this contradicts the fact that
g(S) = I, and hence g must be continuous on S, ■

4 Differentiation

4.1 Introduction

,→ Definition 4.1: Differentiability

Let I ⊆ R be an interval, f : I → R and c ∈ I. We say that f is differentiable at c if
the limit
f (x) − f (c)
lim
x→c x−c
exists. If this limit exists, we denote it f ′ (c) and call it the derivative of f at c.

,→ Theorem 4.1
If f : I → R has a derivative at c ∈ I, then f is continuous at c.

Proof. We have for x ∈ I \ {c},

f (x) − f (c)
f (x) − f (c) = (x − c).
x−c

f being differentiable at c gives that

f (x) − f (c)
lim = f ′ (c),
x→c x−c

so be algebraic properties of limits,

f (x) − f (c)
lim(f (x) − f (c)) = (lim lim(x − c)) = f ′ (c) · 0 = 0,
x→c x→c x−c x→c

§4.1 Differentiation: Introduction p. 93

hence, limx→c f (x) = f (c), and thus f continuous at c. ■

Remark 4.1. The converse of this theorem does not hold.

⊛ Example 4.1: Continuous ⇏ differentiable

Consider f (x) = |x|. This function is continuous on R but not differentiable at
c = 0;

f (x) − f (c) |x| 1

x>0
= =
x−c x −1 x < 0


f (x) − f (c) f (x) − f (c)

=⇒ lim+ = 1, lim− = −1
x→c x−c x→c x−c
f (x) − f (c)
=⇒ lim DNE =⇒ f not differentiable atc = 0.
x→c x−c

,→ Theorem 4.2: Algebraic Properties of the Derivative

Let I ⊆ R be an interval and c ∈ I. Let f : I → R and g : I → R be differentiable at
c. Then

1. For any k ∈ R, kf differentiable at c, and moreover,

(kf )′ (c) = k · f ′ (c).

2. f + g is differentiable at c;

(f + g)′ (c) = f ′ (c) + g ′ (c).

3. (Product Rule) f · g is differentiable at c and

(f g)′ (c) = f ′ (c) · g(c) + f (c) · g ′ (c)

4. (Quotient Rule) If g(x) ̸= 0 ∀ x ∈ I, then the quotient function f

g
is differen-
tiable at c;  ′
f f ′ (c)g(c) − f (c)g ′ (c)
(c) =
g [g(c)]2

Proof. 1.

2.

§4.1 Differentiation: Introduction p. 94

 3.

4. Let h(x) = f (x)

g(x)
. Then,

f (x)
h(x) − h(c) g(x)
− fg(c)
(c)

=
x−c x−c
f (x)g(c) − f (c)g(x)
=
(x − c)(g(x)g(c))
z }| { z }| {
f (x)g(c) − f (c)g(c) + f (c)g(c) − f (c)g(x)
=
(x − c)g(x)g(c)
(f (x) − f (c))g(c) (g(x) − g(c))f (c)
= − ⊛
(x − c)g(x)g(c) (x − c)g(x)g(c)
f ′ (c)g(c) g ′ (c)f (c)
lim ⊛ = lim −
x→c x→c g(x)g(c) g(x)g(c)
f (c)g(c) − g ′ (c)f (c)
′
=
[g(c)]2

,→ Definition 4.2
If f ′ exists on every point c ∈ I, then we say that f is differentiable on I. This gives
a function
f ′ : I → R.

,→ Proposition 4.1: Power Rule

Let f : I → R, f (x) = xn , n ∈ N. We have that f ′ (x) = nxn−1 .

Proof. If n = 1, then f = x, and so f (x)−f (c)

x−c
= x−c
x−c
= 1. Suppose the rule holds up to some
n ∈ N. Consider f = xn+1 . Then,

f (x) = xn+1 = xn x
power rule
=⇒ f ′ (x) = nx n−1
| {z } ·x + x
n

assumption

= (n + 1)xn

⊛ Example 4.2
Prove that d
dx
sin x = cos x and d
dx
cos x = − sin x.

§4.1 Differentiation: Introduction p. 95

4.2 The Chain Rule

,→ Theorem 4.3: Caratheodory Theorem

Let I be an interval, f : I → R, and c ∈ I. TFAE:

1. f is differentiable at c;

2. ∃ a function φ : I → R, continuous at c, such that

f (x) = f (c) + φ(x)(x − c), ∀ x ∈ I.

Remark 4.2. From 2. =⇒ 1., we have, moreover, that f ′ (c) = φ(c).

Proof. (1. =⇒ 2.) Let


 f (x)−f (c)

x ̸= c
x−c
φ : I → R, x 7→ .
f ′ (c)

x=c

We have, then,
f (x) − f (c)
lim φ(x) = lim = f ′ (c) = φ(c),
x→c x→c x−c
hence, φ is continuous at c. For x ̸= c, the desired relation f (x) = f (c) + φ(x)(x − c) holds
by definition.

(1. ⇐= 2.) If x ̸= c, we have that f (x)−f (c)

x−c
= φ(x). Moreover, φ continuous at c, hence
limx→c φ(x) = φ(c), and thus limx→c x−c exists, and moreover, is equal to φ(c). Thus,
f (x)−f (c)

f differentiable at c and f ′ (c) = φ(c). ■

,→ Theorem 4.4: Chain Rule

Let I, J be intervals in R, and let g : I → R, f : J → R be s.t. f (J) ⊆ I. Let c ∈ J;
then, if f differentiable at c and g differentiable at f (c), then the composite function

h = g ◦ f, h : J → R,

is differentiable at c, and moreover,

h′ (c) = (g ◦ f )′ (c) = g ′ (f (c)) · f ′ (c)

Proof. Given f ′ (c) exists, the Caratheodory theorem gives that there exists a function φ :

§4.2 Differentiation: The Chain Rule p. 96

J → R which is continuous at c such that

f (x) − f (c) = φ(x)(x − c) ∀ x ∈ J.

Similarly, since g is differentiable at f (c) ∈ I, there exists a function Ψ : I → R continuous

at f (c), such that
g(y) − g(f (c)) = Ψ (y)(y − c).

Letting y = f (x), this yields

g(f (x)) − g(f (c)) = Ψ (f (x))(f (x) − c)

= Ψ (f (x))φ(x)(x − c).

Letting h = g ◦ f and r(x) = Ψ (f (x))φ(x) gives us

h(x) − h(c) = r(x)(x − c) ∀ x ∈ J.

By compositions, r is continuous at c, and moreover, r(c) = Ψ (f (c))φ(c) = g ′ (f (c))f ′ (c),

and hence,
h′ (c) = g ′ (f (c)) · f ′ (c).

4.3 Derivative of the Inverse Function

,→ Theorem 4.5
Let I be an interval and f : I → R be a strictly increasing continuous function. Let
J = f (I) and g : J → R be the inverse of f . Suppose f differentiable at c, f ′ (c) ̸= 0.
Then, g is differentiable at f (c), and g ′ (f (c)) = 1
f ′ (c)
.

Proof. By the Caratheodory theorem, we have some φ : I → R continuous at c s.t. f (x) −

f (c) = φ(x)(x − c), where φ(c) = f ′ (c). Since f ′ (c) ̸= 0 and φ continuous at c, we have
that there exists δ > 0 s.t. φ(x) ̸= 0 ∀ x ∈ (c − δ, c + δ) ∩ I. ■

§4.3 Differentiation: Derivative of the Inverse Function p. 97

 5 Appendix

5.1 Interesting Results

A summary of theorems or results that stemmed from assignments, tutorials, etc..

,→ Theorem 5.1: Cesàro Summation

Consider a convergent sequence (xn ). Then, the sequence defined

n
x1 + x2 + · · · + xn 1X
yn = = xk
n n k=1

is also convergent, and we have that

lim xn = lim yn .
n→∞ n→∞

,→ Theorem 5.2: Stolz-Cesàro

Let (yn ) be a strictly monotone sequence of positive numbers. Consider some other
sequence (xn ). We have, then, if

xn+1 − xn
lim =L
n→∞ yn+1 − yn

exists, then the limit

xn
lim =L
n→∞ yn
as well.

,→ Lemma 5.1: Fekete’s Subadditive Lemma

A sequence (xn ) is called subadditive if ∀ n, m ∈ N,

xn+m ≤ xn + xm

holds. For any subadditive sequence (xn ), its limit exists, and moreover,

xn
lim xn = inf{ : n ∈ N}.
n→∞ n

§5.1 Appendix: Interesting Results p. 98

 ,→ Definition 5.1: Lacunary Sequence
A sequence xn is called lacunary if there exists some real number q such that ∀ n ∈ N,

xn+1
≥ q > 1.
xn

§5.1 Appendix: Interesting Results p. 99