21-355: Real Analysis 1

Carnegie Mellon University

Professor Ian Tice - Fall 2013

Project LATEXd by Ivan Wang

Last Updated: November 15, 2013

CONTENTS

21-355 Notes

Contents
1 The Number Systems
1.1

1.2

1.3

1.4

1.5

1.6

The Natural Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.1

Positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.2

Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.3

Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.1

Properties of Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.2

Algebraic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Rationals and Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.3.1

Fields and Orders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problems with Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.4.1

Bounds (Infimum and Supremum) . . . . . . . . . . . . . . . . . . . . . . . .

The Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.5.1

Defining the Real Numbers: Dedekind Cuts . . . . . . . . . . . . . . . . . . .

1.5.2

Defining the Real Numbers: The Least Upper Bound Property . . . . . . . . 10

1.5.3

Defining the Real Numbers: Addition . . . . . . . . . . . . . . . . . . . . . . 10

1.5.4

Defining the Real Numbers: Multiplication . . . . . . . . . . . . . . . . . . . 11

1.5.5

Defining the Real Numbers: Distributivity . . . . . . . . . . . . . . . . . . . . 11

1.5.6

Defining the Real Numbers: Archimedean . . . . . . . . . . . . . . . . . . . . 12

Properties of R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6.1

Absolute Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2 Sequences
2.1

13

Convergence and Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.1.1

Squeeze Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2

Monotonicity and limsup, liminf

2.3

Subsequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.1

2.4

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Limsup Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Special Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Series
3.1

3.2

17

Convergence Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1

Cauchy Criterion Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.1.2

Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

The number e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

CONTENTS

21-355 Notes

3.3

More Convergence Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.4

Algebra of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.5

Absolute Convergence and Rearrangements . . . . . . . . . . . . . . . . . . . . . . . 22

4 Topology of R
4.1

4.2

Open and Closed Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.1.1

Open Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.1.2

Closed Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.1.3

Limit Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.1.4

Closure, Interior, and Boundary Sets . . . . . . . . . . . . . . . . . . . . . . . 26

Compact Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2.1

4.3

24

Heine-Borel Theorem

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

Connected Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5 Continuity
5.1

30

Limits of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.1.1

Divergence Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.2

Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

5.3

Compactness and Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

5.4

Continuity and Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

5.5

Discontinuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

5.6

Monotone Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6 Differentiation

35

6.1

The Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6.2

Mean Value Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

6.3

Darbouxs Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

6.4

LH
opitals Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

6.5

Higher Derivatives and Taylors Theorem . . . . . . . . . . . . . . . . . . . . . . . . 38

7 Riemann-Stieltjes Integration

39

7.1

The R-S Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

7.2

Integrability Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

21-355 Notes

1
1.1

The Number Systems

The Natural Numbers

Theorem (existence of N): There exists a set N satisfying the following properties, known as the
Peano Axioms:
PA1 0 N
PA2 There exists a function S : N N called the successor function. In particular,
S(n) N.
PA3 n N. S(n) 6= 0
PA4 S(n) = S(m) = n = m (S is injective, one-to-one)
PA5 [Axiom of Induction] Let P (n) be a property associated to each n N.
If P (0) is true, and P (n) = P (S(n)), then P (n) is true n N.
Definition: PA1 = 0 N. PA2 = S(0) N.
Define 1 = S(0), 2 = S(1), 3 = S(2), etc.
PA2 guarantees that {0, 1, 2, } N.
PA3 prevents wraparound: no successor can map to a negative number.
PA4 prevents stagnation: the cycle does not terminate.
Theorem: N = {0, 1, 2, }
Proof : We know that {0, 1, 2, } N, so it suffices to prove that N {0, 1, 2, }.
Let P (n) denote the proposition that n {0, 1, 2, }. Clearly P (0) is true.
Suppose P (n) is true; then n {0, 1, 2, } = S(n) {0, 1, 2, } by construction.
Hence, P (S(n)) is true. By induction, PA5 guarantees that P (n) is true n N.
It follows that N {0, 1, 2, }.
Definition: For any m N, we define 0 + m = m.
Then if n + m is defined for n N, we set S(n) + m = S(n + m).
Proposition (Properties of Addition):
1. n N. n + 0 = n

(0 is the additive identity)

2. m, n N. n + S(m) = S(n + m)
3. m, n N. m + n = n + m

(commutativity)

4. k, m, n N. k + (m + n) = (k + m) + n

(associativity)

5. k, m, n N. n + k = n + m = k = m

(cancelation)

Proof :
1. Let P (n) be n + 0 = n.
P (0) is true because 0 + 0 = 0 by definition.
Note P (n) = S(n) + 0 = S(n + 0) = S(n), so P (S(n)) is true. By induction,
(1) is true.
3

1.1

The Natural Numbers

21-355 Notes

2. Fix m N. Let P (n) denote n + S(m) = S(n + m).

P (0) is true because 0 + S(m) = S(m) = S(0 + m).
P (n) = S(n) + S(m) = S(n + S(m)) = S(S(n + m)) = S(S(n) + m), so P (S(n))
is true. By induction, since m N was arbitrary, (2) is true.
3. Let m be fixed and P (n) denote n + m = m + n.
P (0) is true since 0 + m = m by definition, and m + 0 = m by 1, so 0 + m = m =
m + 0.
Suppose P (n); then S(n) + m = S(n + m) = S(m + n) = m + S(n), so P (S(n)) is
true. By induction and arbitrary choice of m, (3) is true.
4. Fix k, m N and let P (n) denote k + (m + n) = (k + m) + n.
P (0) is true as k + (m + 0) = k + m = (k + m) + 0.
Suppose P (n); then k+(m+S(n)) = k+S(m+n) = S(k+(m+n)) = S(k+m)+n =
(k + m) + S(n) by (2). By induction and arbitrary choice, (4) is true.
5. Fix m, n N and let P (k) denote proposition 5.
P (0) is true because n + 0 = n = n + m = m = 0 = k = m.
Suppose P (k); also, suppose m + S(k) = n + S(k). Then S(m + k) = m + S(k) =
n + S(k) = S(n + k) = m + k = n + k = m = n (by 4). By the axiom of
induction, (5) is true.
1.1.1

Positivity

Definition: We say that n N is positive if n 6= 0.

Proposition (Properties of Positivity):
1. n, m N, if m is positive, then m + n is positive.
2. n, m N, if m + n = 0, then m = n = 0.
3. n N, if n is positive, then there exists a unique m N such that n = S(m).
1.1.2

Order

Definition: For all m, n N, m n or n m iff n = m + p for some p N.

m < n or n > m iff m n m 6= n. The relation provides what is called an order on N.
Proposition (Properties of Order):
Let j, k, m, n N. Then:
1. n n (reflexitivity)
2. m n k m = k n (transitivity)
3. m n m n = m = n (anti-symmetry)
4. j k m n = j + m k + n (order preservation)
5. m < n S(m) n
6. m < n n = m + p for some positive p N.
7. n m S(n) > m
8. n = 0 0 < n

1.2

The Integers

21-355 Notes

Theorem (Trichotomy of Order): Let m, n N. Then exactly one of the following is true:
m<n

m=n

m>n

Proof : Show that no two can be true simultaneously (by definition of < and >), and then at least
one must be true (by induction on n).
1.1.3

Multiplication

Definition: Fix m N. Define 0 m = 0. Now, if n m is defined for some n N, we define

S(n) m = n m + m.
Proposition (Properties of Multiplication):
Fix k, m, n N. Then:
1. m n = n m (commutativity)
2. m, n are positive = mn is positive
3. m n = 0 m = 0 n = 0 (no zero divisors)
4. k (m n) = (k m) n (associativity)
5. k m = k n k is positive = m = n (cancelation)
6. k (m + n) = (m + n) k = k m + k n (distributivity)
7. m < n k l k, l are positive = m k < n l

1.2

The Integers

Consider the following relation on the set N N:

(m, n) ' (m0 , n0 ) m + n0 = m0 + n
Lemma: ' is an equivalence relation.
Proof :
Reflexivity: m + n = m + n = (m, n) ' (m, n)
Symmetry: (m, n) ' (m0 , n0 ) = m + n0 = m0 + n = m0 + n = m + n0 =
(m0 , n0 ) ' (m, n)
Transitivity: Suppose (m, n) ' (m0 , n0 ) (m0 , n0 ) ' (m00 , n00 ). Then:
m + n0 = m0 + n m0 + n00 = m00 + n0
= m + n00 = m00 + n
= (m, n) ' (m00 , n00 )
Definition: Write the equivalence class of (m, n) as [(m, n)] = {(p, q) | (p, q) ' (m, n)}.
Define the integers Z = {[(m, n)]}.
Lemma: Suppose (m, n) ' (m0 , n0 ), (p, q) ' (p0 , q 0 ). Then:
1. (m + p, n + q) ' (m0 + p0 , n0 + q 0 )
5

1.2

The Integers

21-355 Notes

2. (mp + nq, mq + np) ' (m0 p0 + n0 q 0 , m0 q 0 + n0 p0 )

Proof : Consider equalities (a) : m + n0 = m0 + n and (b) : p + q 0 = p0 + q (by definition of ').
Using linear combinations of (a) and (b), we derive the two rules of the lemma:
1. (a) + (b)
2. (a)(p0 + q 0 ) + (b)(m + n)
Definition: Let [(m, n)], [(p, q)] Z. Then:
1. [(m, n)] + [(p, q)] = [(m + p, n + q)] (addition of integers)
2. [(m, n)] [(p, q)] = [(mp + nq, mq + np)] (multiplication of integers)
By the lemma, these are well-defined operations.
Note that for all m, n N:
[(m, 0)] = [(n, 0)] m + 0 = n + 0 m = n
[(m, 0)] + [(n, 0)] = [(m + n, 0)]
[(m, 0)] [(n, 0)] = [(mn, 0)]
As such, the set {[(n, 0)] | n N} Z behaves exactly like a copy of N.
Definition: For n N we set n Z to be n := [(n, 0)].
For x = [(m, n)] Z we define x = [(n, m)].
1.2.1

Properties of Integers

(We can see that every integer x Z can be represented as x := m n where x = [(m, n)].)
Theorem: Every x Z satisfies exactly one of the following:
1. x = n for some n N\{0}
2. x = 0
3. x = n for some n N\{0}
Proof : Write x = [(p, q)] for some p, q N. By trichotomy of order on N we know that p < q or
p = q or p > q. Each of these correlates to one of the three properties.
Corollary: Z = {0, 1, 2, . . .} {1, 2, 3, . . .}
1.2.2

Algebraic Properties

Proposition: Let x, y, z Z. Then the following hold:

1. x + y = y + x
2. x + (y + z) = (x + y) + z
3. x + 0 = 0 + x = x
4. x + (x) = (x) + x = 0
5. xy = yx
6

1.3

The Rationals and Ordered Fields

21-355 Notes

6. (xy)z = x(yz)
7. x 1 = 1 x = x
8. x(y + z) = xy + xz
Definition: Define x y = x + (y). The usual properties hold.
Definition: For x, y Z, we say x y or y x if y x = n for some n N.
We say x < y if x y x 6= y.

1.3

The Rationals and Ordered Fields

Let a relation on Z (Z\{0}) be given by (m, n) ' (m0 , n0 ) mn0 = m0 n.

Lemma: ' is an equivalence relation. Proof follows from properties of Z.
Definition: Q = {[(m, n)]}
1. [(m, n)] + [(p, q)] = [(mq + np, nq)] (addition)
2. [(m, n)] [(p, q)] = [(mp, nq)] (multiplication)
3. [(m, n)] = [(m, n)] (negation)
4. If m 6= 0 we set [(m, n)]1 = [(n, m)]
Remark: the heuristic here is that

m
n

= [(m, n)].

Definition: If m Z, we write m = [(m, 1)] Q; and thus N Z Q.

1. For x, y Q, we define x y = x + (y) Q
2. For x, y Q, y 6= 0 we define
y = [(0, n)].

x
y

= x(y)1 . This is well defined because y = 0

Proposition: Q = { m
n | m, n Z, n 6= 0}.
We define and propose the trichotomy of order on Q, as per the integers.
1.3.1

Fields and Orders

Definition: A field is a set F endowed with two binary operations, +, , satisfying the following
axioms:
(A1, M1) x, y F. x + y F, xy F (closure)
(A2, M2) x, y F. x + y = y + x, xy = yx (commutativity)
(A3, M3) x, y, z F. x + (y + z) = (x + y) + z, x(yz) = (xy)z (associativity)
(A4, M4) (0, 1) F. x F. 0 + x = x + 0 = x, 1 x = x 1 = x (identity)
(A5, M5) x F. (x). x + (x) = 0; x1 F. xx1 = x1 x = 1 (inverse)
(D1) x, y, z F. x(y + z) = xy + xz (distributivity)
Remark : Field must have at least 2 elements (0, 1) by (A/M4). To prove field, must prove 5
properties of addition and multiplication (closure, commutativity, associativity, identity, inverse)
as well as distributivity.
Definition: Let E be a set; an order on E is a relation < satisfying the following:
7

1.4

Problems with Q

21-355 Notes

1. x, y E exactly one of the following is true: x < y or x = y or y < x (trichotomy)

2. x, y, z E, x < y y < z = x < z (transitivity)
Definition: Let F be a field. Then we define x y = x + (y) and

x
y

= xy 1 (for y 6= 0).

Theorem: Q is an ordered field with order <.

Proof : Follows from definitions and properties of Z.

1.4

Problems with Q

Theorem: There does not exist a q Q such that q 2 = 2.

Proof : Suppose not; i.e. there does exist such a q Q.
Consider the set S(q) = {n N+ | q = m
n for some m Z}. Cleary |S(q)| > 0. Then
the well-ordering principle implies that !n S(q). n = min S(q).
2

m 2
m
2
Since n S(q), we know that q = m
n for some m Z. Then q = ( n ) = n2 =
2
2
2
m = 2n = m is even. We claim that m is also even (proof is exercise to reader).

Then l Z. m = 2l. Then 4l2 = (2l)2 = m2 = 2n2 = n2 = 2l2 = n2 is even

= n is even = n = 2p for some p N+ .
2l
l
Hence q = m
n = 2p = p = p S(q). But clearly p < n, which contradicts the fact
that n is the minimal element. By contradiction, the theorem must be true.

1.4.1

Bounds (Infimum and Supremum)

Informally, Q has holes:

Definition: Let E be an ordered set with order <.
1. We say A E is bounded above iff x E. a A. a x. We say x is an upper
bound of A.
2. We say A E is bounded below iff x E. a A. x a. We say x is a lower
bound of A.
3. We say A E is bounded iff its bounded above and below.
4. We say x is a minimum of A iff x A and x is a lower bound of A.
5. We say x is a maximum of A iff x A and x is an upper bound of A.
Remark : If a min or max exists, then it is unique.
Definition: Let E be an ordered set and A E.
1. We say x E is the least upper bound (supremum) of A, written x = sup A, iff x
is an upper bound of A and y E is an upper bound of A = x y.
2. We say x E is the greatest lower bound (infimum) of A, written x = inf A, iff x
is a lower bound of A and y E is a lower bound of A = y x.
Remark : If x = min(A), then x = inf(A). If x = max(A), then x = sup(A). But the converse is
false; some sets have a supremum but no maximum, others a infimum but no minimum.
Definition: Let F be an ordered field. We say that F has the least upper bound property iff every
6= A F that is bounded above has a least upper bound.
8

1.5

The Real Numbers

21-355 Notes

Theorem: Q does not satisfy the least upper bound property.

Proof : Consider the set A = {x Q | x > 0, x2 2}.
Note that 0 < 1 = 12 2 = 1 A, so A is non-empty. Also, 2 4 = 22 implies
(x A = 0 < x2 < 2 < 22 ) = x < 2. Then 2 is an upper bound of A.
Assume for sake of contradiction that Q has the least upper bound property. Then A
has a supremum. Let x = sup A Q and write x = pq for p, q Z.
By trichotomy, x2 < 2 or x2 = 2 or x2 > 2. We know x2 6= 2.
Case 1: Suppose x2 < 2. Then for any n N+ we have ( pq + n1 )2 =
p2
q2

1 2p+q
n ( q ).

From algebra, we derive

( pq

1 2
n)

< 2 for some n

Cleary x > 0 since otherwise x 0 < 1 A. Hence 0 < x =

x is not an upper bound = contradiction.

p
q

<

p2
q2

2p
qn

1
n2

N+ .
p
q

1
n

A. But then

Case 2: Suppose x2 > 2. Considering ( pq n1 )2 > 2 and using the same logic as
before, we can choose n large enough such that pq n1 is an upper bound of A. But
p
p
1
q n < q = x, which contradicts the fact that x = sup A.
As all cases are false, we contradict trichotomy, and hence Q cannot have the least
upper bound property.

1.5

The Real Numbers

We now construct an ordered field satisfying the least upper bound property using Q.
Definition: We say Q is Archimedean iff (x Q). x > 0 = (n N). x < n.
Lemma: If Q is Archimedean, then (p < q Q). (r Q). p < r < q.
(Proofs in HW 2.)
1.5.1

Defining the Real Numbers: Dedekind Cuts

Definition: We say that C P(Q) is a cut (Dedekind cut) iff the following hold:
(C1) 6= C, C 6= Q
(C2) If p C and q Q with q < p, then q C.
(C3) If p C, (r Q). p < r r C.
Lemma: Suppose C is a cut. Then:
1. p C, q
/ C = p < q
2. r
/ C, r < s = s
/C
3. C is bounded above
Lemma: Let q Q. Then {p Q | p < q} is a cut.
Proof : Call the set C. We prove the 3 properties of a cut:
(C1) q 1 C = C =
6 ; q + 1
/ C = C =
6 Q.
(C2) If p C and r Q such that r < p, then r < p < q = r < q = r C.
(C3) Let p C where p < q. Since Q is Archimedean, (r Q). p < r < q = r C.
9

1.5

The Real Numbers

21-355 Notes

Definition: Given q Q we write Cq = {p Q | p < q}. By the above lemma, Cq is a cut.

Definition: We write R = {C P(Q) | C is a cut} =
6 .
Lemma: The following hold:
1. A, B R, exactly one of the following holds: A B, A = B, B A.
2. A, B, C R, A B B C = A C.
Definition: If A, B R we say that A < B A B, and A B A B. This defines
an order on R by the above lemma.
1.5.2

Defining the Real Numbers: The Least Upper Bound Property

Lemma: Suppose 6= E R is bounded above. Then B :=

AE

A R.

Theorem: R satisfies the least upper bound property.

Proof : Let 6= E R be bounded above and set B =

AE

A R. We claim B = sup E.

First, we show that B is an upper bound of E. Let A E. Then A B = A B

(by definition). This is true for all A E, so B is an upper bound.
We claim that for C R. C < B = C is not an upper bound of E. If C < B, then
C B. This implies b B. b
/ C = (A E). b A b
/ C. Then A > C since
otherwise A C = b C, b
/ C. Hence C < A and C is not an upper bound of E.
By the contrapositive: if C is an upper bound, C B. Thus, B is the least upper bound,
and the theorem holds.
1.5.3

Defining the Real Numbers: Addition

Definition: Given A, B R, set A + B = {a + b | a A, b B}.

Lemma: If A, B R, then A + B R.
Theorem: Define A = {q Q | (p > q). p
/ A}. Then R, +, 0R = C0 = {p Q | p < 0}
satisfy the field axioms.
Proof :
(A1) A + B R by previous lemma.
(A2) A + B = {a + b} = {b + a} = B + A.
(A3) A + (B + C) = {a + (b + c)} = {(a + b) + c} = (A + B) + C.
(A4) Show A R. 0R + A = A.
(A5) Show that A R, then A + (A) = 0R using Archimedean property.
Theorem (Ordered Field): Let A, B, C R. If A < B then A + C < B + C.
Proof : Its trivial to see that A B = A + C B + C = A + C B + C.
If A + C = B + C, we can add C to both sides and use the last theorem to see that
A = B, a contradiction. Hence, A + C < B + C.

10

1.5

The Real Numbers

1.5.4

21-355 Notes

Defining the Real Numbers: Multiplication

Lemma: Let A, B R, A, B > 0R . Then C = {q Q | q 0} {a b | a A, b B, a, b > 0} R.

Proof :
(C1) 0 C = C 6= . A, B are bounded above by, say M1 , M2 , so M1 M2 + 1
/C
and C 6= Q.
(C2) Let p C and q < p. If q 0 then q C by definition. If q > 0 then 0 < q < p,
but then 0 < p = p = a b for a A, b B, a, b > 0. Then 0 < q < a b =
q
q
q
a < b = 0 < a B. Then q = a( a ) C.
(C3) Let p C. If p 0 then any a b with a A, b B, a, b > 0 satisfies p < a b C,
so r = a b is the desired element of C. However, if p > 0, then p = a b for
a A, b B, a, b > 0. Choose s A such that a < s, t B such that t > b. Then
p = a b < s t S, so r = s t proves the claim.
Definition of Multiplication: Let A, B R.
1. If A > 0, B > 0 we set A B = {q Q | q 0} {a b | a A, b B, a, b > 0} R.
2. If A = 0 or B = 0, we set A B = 0R .
3. If A > 0 and B < 0, let A B = (A (B)).
4. If A < 0 and B > 0, let A B = ((A) B).
5. If A < 0 and B < 0, let A B = (A) (B).
Theorem: R, satisfies (M1-M5) with 1R = C1 , and
A > 0 = A1 = {q Q | q 0} {q Q | q > 0, p > q. p1
/ A} R;
A < 0 = A1 = (A)1 .
Proof : HW3 (similar to addition).
Theorem: If A, B > 0, then A B > 0.
Proof : By definition C0 A B = 0 A B. Equality is impossible since A, B > 0.
1.5.5

Defining the Real Numbers: Distributivity

Theorem: Let A, B, C R. Then A (B + C) = A B + A C.

Proof : We prove the case where all are positive. The other cases are in HW.
Let p A(B + C). If p 0 then p A B + A B is trivial (both products contain the
interval less than 0).
If p > 0, p = a(b + c) for a A, b B, c C for a > 0, b + c > 0.
Regardless of sign of b or c, a b A B, a c A C. Hence p = a(b + c) = a b + a c
A B + A C. So A(B + C) A B + A C.
Finally, we show the converse is true; let p A B + A C = p = r + s for
r A B, s A C. Case on positivity of p, r, s to show p A(B + C).

11

1.6

Properties of R

1.5.6

21-355 Notes

Defining the Real Numbers: Archimedean

Theorem: For p, q Q, the following are true:

1. Cp+q = Cp + Cq
2. Cp = Cp
3. Cpq = Cp Cq
4. If p 6= 0 then Cp1 = (Cp )1
5. p < q Q Cp < Cq R
Proof : HW.
Definition: For q Q we say Cq R. Then Q R.
Theorem: There exists an ordered field satisfying the least upper bound property; R is unique (for
any ordered field F satisfying these properties, F = R up to isomorphism; and R is Archimedean.
Proof : The basic assertion is Steps (0)-(4). Step (5) proves 1, Step (6) proves 3.

1.6

Properties of R

Notation: think of R as numbers, not cut notation.

Proposition: R satisfies the following:
Theorem: For p, q Q, the following are true:
1. R is Archimedean: x R, x > 0. n N. x < n
2. N R is not bounded above
3. inf{ n1 | n N, n 1} = 0
4. x R the set B(x) = {m Z | x < m} has a minimum in Z.
5. x, y R, x < y. q Q. x < q < y
Remarks:
1. (5) is interpreted as the density of Q R. Any element x R can be approximated to arbitrary accuracy by elements of Q.
2. (4) allows us to define the integer part of any x R. We can set bxc = min B(x)
1 Z. Then bxc x < bxc + 1.
Next we show that R does not have the holes we saw in Q.
Theorem: Let x R satisfy x > 0 and n N, n 1. Then !y R. y > 0 y n = x.
Proof : The case n = 1 is trivial so assume n 2.
Set E = {z R | z > 0 z n < x}. We want to show E 6= and is bounded above. Set
x
t = 1+x
; then 0 < t < 1 and t < x. Hence 0 < tn < t < x, and so t E and E 6= .
Set s = 1 + x. Then 1 < s x < s = x < s < sn ; so if z E then z n < x < sn =
z < s. Then s is an upper bound of E.
By least upper bound property, y R. y = sup E. Since t E, 0 < t < y, so y > 0.
We claim that y n < x and y n > x are both impossible (proof is exercise), so y n = x.
1

Definition: Let n 1; for x R, x > 0, we write x n = y where y n = x. We set 0 n = 0.

12

21-355 Notes

1.6.1

Absolute Value

For x R, we define the function | | : R {r R | r 0}:

if x > 0
x
|x| = 0
if x = 0

x if x < 0
Proposition (Properties of | |):
1. x R. |x| 0 and |x| = 0 x = 0
2. x, y R. |x| < y y < x < y
3. x, y R. |xy| = |x||y|
4. x, y R. |x + y| |x| + |y| (Triangle Inequality)
5. x, y R. ||x| |y|| |x y|

Sequences

Let E be a set. Then we may define a sequence {an }

n=l E as the set of values an a(n) for
some l Z and some function a : {n Z | n l} E.

2.1

Convergence and Bounds

Definition: We say a sequence {an }

n=l R converges to a R, i.e. an a as n
or limn an = a, if for every 0 <  R, there exists N {m Z | m l} such that
n N = |an a| < .
Definition: We say a sequence {an }
n=l R is bounded iff. M R, M > 0. |an | < M (n l).
Lemma: If a sequence converges, then it is bounded.
Definition: Given {an }, {bn } R we define {an + bn } R to be the sequence whose elements are
an + bn . We similary define {can } for a fixed c R, {an bn }, and {an /bn } where bn 6= 0, n l.
Theorem (algebra of convergence): Let {an }, {bn } R, c R, and assume that an a, bn b
as n . Then the following hold:
1. an + bn a + b as n
2. can ca as n
3. an bn ab as n
4. If bn 6= 0 and b 6= 0, then an /bn a/b as n .
Proof : (1), (2) are in next weeks HW.
(3): Note that |an bn ab| = |an bn abn + abn ab| |an bn abn | + |abn ab| =
|bn ||an a| + |a||bn b|. Since bn b we know that M > 0. |bn | < M (n l).
Let  > 0. Since an a and bn b we may choose N1 such that n N1


; and N2 where n N2 = |bn b| < 2(1+|a|)
.
|an a| < 2M
13

2.1

Convergence and Bounds

21-355 Notes

Then set N = max(N1 , N2 ). So if n N we know that |an bn ab| |bn ||an a| +



|a||bn b| < M |an a| + |a||bn b| < M 2M
< 2 + 2 = .
+ |a| 2(1+|a|)
Since  was arbitrary, we deduce that an bn ab.
an bab+ababn
n
|
(4): We know | abnn ab | = | an bab
bn b | = |
bn b
|a|
|b||bn | |bn

|an bab|
|bn ||b|

|ababn |
|b||bn |

|an a|
|bn |

b|.

Let  > 0. Since bn b 6= 0 we know that N1 such that n N1 = |bn b| < |b|
2 .
|b|
|b|
Then n N = 0 < |b| = |b bn + bn | |b bn | + |bn | < 2 + |bn | = 0 < 2
2
|bn | = 0 < |b1n | < |b|
.
Similarly, an a = N2 . (n N2 = |an a| < 4 |b|; and
bn b = N3 . (n N3 = |bn b| <

|b|2
4(1+|a|) .

Set N = max(N1 , N2 , N3 ). Then n N = | abnn ab |

2
|b||an a|

2|a|
|b
|b|2 n

b| <


2

 |a|
2 1+|a|

|an a|
|bn |

|a|
|bn ||b| |bn

b| <

< .

Since  > 0 was arbitrary, we deduce

an
bn

a
b

as n .

Lemma: Let {an }

n=l converge to a R. Then  > 0. N. m, n N = |an am | < .
Definition: We say {an }
n=l R is Cauchy iff  > 0. N. m, n N = |an am | < .
Lemma: If {an } is Cauchy, then its bounded.
Proof : Let  = 1. Then N. m, n N = |am an | < 1. Then n N = |an aN | < 1 =
|an | < |an aN | + |aN | < 1 + |aN |. Set M = max(1 + |aN |, k), where k = max{|al |, . . . , |aN 1 |}.
Then |an | < M (n l), and {an } is bounded.
Theorem: Let {an } R. Then {an } converges {an } is Cauchy.
Proof : = is covered by 2nd-previous lemma. We show the converse:
Suppose {an } is Cauchy. Then |an | < M (n l) by the last lemma.
Set E = {x R | N. n N = x < an }. Note that M < an (n l), and so
M E and E 6= .
Also, x E = Nx . n Nx = x < an < M , and so M is an upper bound of E.
By the least upper bound property of R, a = sup E R. We claim that an a as
n .
Let  > 0. Then since {an } is Cauchy, N. m, n N = |an am | < 2 . In particular,
|an aN | < 2 when n N . Then n N = aN 2 < an = aN 2 E =
aN 2 a.
If x E, then Ex . (n Nx = x < an < aN + 2 ). Hence aN +
of E = a aN + 2 . Then |a aN | < 2 .
But if n N , then |an a| |an aN | + |aN a| <
2.1.1


2


2


2

is an upper bound

= . Hence an a.

Squeeze Lemma

Lemma: Let {an }

n=l , {bn }, {cn } R and suppose that an a, cn a as n . If k l such
that an bn cn (n k), then bn a as n .
Examples:
14

2.2

Monotonicity and limsup, liminf

21-355 Notes

1. Suppose an 0 and {bn } is bounded, i.e. |bn | M (n l). Then |an bn | = |an ||bn | |an |M .
But cn 0 |cn | 0. Then 0 |an bn | |an |M , both sides of which go to 0; and by
the squeeze lemma, |an bn | 0 = an bn 0.
2. Fix k N with k 1. Set an =

1
,n
nk

1. Then 0

1
nk

n1 , and by squeeze lemma

1
nk

0.

3. Fix k N with k 2. Let an = k1n , n 0. We know n N. n k n (proof by induction).

Then 0 k1n n1 , and by squeeze k1n 0.

2.2

Monotonicity and limsup, liminf

Definition: Let {an }

n=l R. We say {an } is:
1. increasing iff. an < an+1 (n l),
2. non-decreasing iff. an an+1 (n l),
3. decreasing iff. an+1 < an (n l),
4. non-increasing iff. an+1 an (n l).
We say {an } is monotone iff. it is either non-increasing or non-decreasing.
Remark : increasing = non-decreasing, decreasing = non-increasing.
Theorem: Suppose that {an }
n=l R is monotone. Then {an } is bounded iff {an } is convergent.
Proof : = is done in a previous lemma.
= : Well prove when the sequence is non-decreasing (other case handled by similar
argument).
Set E = {an | n l} R. Clearly E 6= . Also, since {an } is bounded, E is as well (in
particular above). By least upper bound property of R, a = sup(E) R. We claim
that a = limn an .
Let  > 0. Since a = sup(E) we know that a  is not an upper bound of E; hence
(N l). a  < aN . Also, since the sequence is non-decreasing, an an+1 (n l),
and so n N = aN an . Then n N = a  < aN an a because a is an
upper bound of E.
So n N =  < an a 0 = |an a| < . Since  > 0 was arbitrary, we
deduce that an a as n .
Lemma: Suppose that {an } is bounded. Set Sm = sup{an | n m} and Im = inf{an | n m}.
Then Sm , Im R are well-defined m l; {Sm } is non-increasing; and {Im } is non-decreasing.
Both sequences are bounded.
Definition: Suppose {an } R is bounded. We set lim supn an = limm Sm R and
lim inf n an = limm Im R. Both limits exist by the lemma and previous theorem. We know
that lim inf n an lim supn an from HW.

2.3

Subsequences

Definition: Let : {n Z | n l} {n Z | n l} be order preserving (increasing), i.e. m < n

then (m) < (n). Let {an }

l=k R be a sequence. We say {a(k) }k=l is a subsequence of {an }.
Remarks:
15

2.3

Subsequences

21-355 Notes

1. (k) = k is order preserving, so every sequence is a subsequence of itself.

2. Not every an has to be in the subsequence {a(k) }.
For example, if l = 0 then (k) = 2k is order preserving. In this case an , n odd
does not appear in the subsequence {a(k) }.
3. We will often write nk = (k) to simplify notation, so {ank } denotes a subsequence.
4. From HW1, we know k (k) (k l).
Proposition: Suppose {an } satisfies an a R as n . Then any subsequence of {an } also
converges to a.
Proof :
Let {a(k) } be a subsequence of {an }. Let  > 0. Since an a as n , we know
N l. n N = |an a| < . We claim K l. k K = (k) N .
If not, then (k) < N (k l); but k (k) < N (k l) is a contradiction. Then
the claim is true, and k K = (k) N = |a(k) a| < . Since  > 0 was
arbitrary, we deduce {a(k) } a as k .
Remark : Converse fails. Example: an = (1)n ; a2n = +1 +1, but a2n+1 = 1 1.
2.3.1

Limsup Theorem

Theorem: Let {an } R be bounded. The following hold:

1. Every subsequence of {an } is bounded.
2. If {ank } is a subsequence, then lim supk ank lim supn an .
3. If {ank } is a subsequence, then lim inf n an lim inf k ank .
4. There exists a subsequence {ank } such that limk ank = lim supn an .
5. There exists a subsequence {ank } such that limk ank = lim inf n an (6= (4)).
Proof :
1. Trivial.
2. Since k (k), {a(n) | n k} {an | n k} for every order-preserving . Hence
Sk = sup{a(n) } | n k} sup{an | n k} = Tk . But:
lim supn a(n) = lim supk {a(n) | n k} lim supk {an | n k} =
lim supn an .
3. Similar to (2); exercise to reader.
4. Too lazy to LATEX; exercise to reader.
5. Exercise to reader.
Theorem: Suppose {an } R; the following are equivalent:
1. an a as n
2. {an } is bounded, and every convergent subsequence converges to a.
3. {an } is bounded, and lim supn an = lim inf n an .
Proof : (1) = (2) proven already.
16

2.4

Special Sequences

21-355 Notes

(2) = (3)
Limsup theorem (4,5) = {a(k) }, {a(k) } subsequences such that a(k) lim supn an , a(k)
lim inf n an as k . By (2) the limits must agree.
(3) = (1)
Limsup theorem (1-3) = {a(k) }. lim inf n an lim inf k a(k) lim supk a(k)
lim supn an . As the first and last are equal, by transitivity it follows all subsequences
satisfy lim inf k a(k) = lim supk a(k) . As an is a subsequence of itself, it therefore
converges to some a as n .
Theorem (Bolzano-Weierstrass): If {an } R is bounded then there exists a convergent subsequence. Proof from (4) or (5) of Limsup Theorem.

2.4

Special Sequences

Definition: Given an R for 0 k n, n N we define

Pn

= a0 + a1 + + an .
n


P
n k nk
Lemma (Binomial Theorem): Let x, y R and n N. Then (x + y)n =
, where
k x y
k=0


n
n!
k := k!(nk)! N.
k=0 an

Theorem: In the following assuming that n 1:

1. Let x R, x > 0. Then an =
2. Let x R, x > 0. Then an =

1
nx 0
x1/n

as n .
1 as n .

3. Let an = n1/n ; then an 1 as n .

4. Let a, x R, x > 0. Then

na
(1+x)a

0 as n .

5. Let x R, |x| < 1. Then an = xn 0 as n .

Series

Definition: Let {an }

n=l R; for p < q we write
1. We define, for each n l, Sn =

n
P
k=l

Pq

n=p an

= (ap + + aq ).

ak R to be the nth partial sum of {an }

n=l .

2. P
If s R. Sn s as n , then

n=l an converges.

n=l

an = s. We say the infinite series

3. If the series does not converge, it diverges.

Examples
P
P
1. P
Let an = xn for
n 0, xP R. Then Sn = nk=0 xk . Notice that (1 x)Sn = nk=0 xk
P
n
n
n+1 k
k+1 =
k
n+1 .
k=0 x
k=0 x
k=1 x = 1 x
P
n+1
1
So Sn = nk=0 xk = ( 1x
1x ). If |x| < 1 then Sn 1x by special seq (5).
2. Suppose
{bn }
b as n . Set an = bn+1 bn for n 0. Then the series
n=0 R where bn P
P

a
converges
and
in
fact
n=0 n
n=0 = b b0 .

17

3.1

Convergence Results

3.1

21-355 Notes

Convergence Results

We develop tools that will let us deduce the convergence of a series without knowing its value.
P
Theorem: Suppose
n=l an converges. Then an 0 as n .
Proof : Notice that an = Sn Sn1 and so limn an = limn (Sn Sn1 ) = S S = 0.
P
P
n
Corollary:
n=0 (1) and
n=0 n diverge, as neither sequences converge to 0.
P n
Corollary: The series n=0 x converges |x| < 1.
Proof : |x| 1 = |xn | = |x|n 1(n N). The converse was proved last time.
Next, we provide a characterization of convergence in terms of the size of the tails of the series.
P
Pm
Theorem:
n=l an converges  > 0. N l. m k N = |
n=k an | < .
P
Pk
Proof : n=l an converges Sk = n=l an converges {Sk } is Cauchy.
This is useful in practice because we can guarantee a series converges without knowing its value.
Theorem:
1. If n k. |an | bn for some k l, and

bn converges, then

n=l

2. If n k. 0 an bn for some k l, and

an converges.

n=l

an diverges, then

n=l

bn diverges.

n=l

Proof : (1) Let  > 0 and prove with previous theorem and induction on triangle inequality. (2)
follows from contrapositive.
Examples:
P
1.
n=0

(1)n
2n

converges because | (1)

2n | =

1
2n

and

1
n=0 2n

converges ( 12 < 1).

2. Suppose
n=0 an converges and an 0 n 0. Let {bn } R be
Pbounded,
i.e.
= |an ||bn | M an . Then M Sn = M nk=0 an =
Pn |bn | M n. Then |an bn | P

n=0 an bn converges.
k=0 M an , so by the theorem,
P (1)n n!
2
3n
3.
n=0 2n nn 4n2 +2 converges because the product is bounded.
P

Theorem: Suppose n l. an 0. Then

n=l an converges {Sn }n=l is bounded.
Pn
Proof : Since an 0, the sequence Sn = k=l ak is non-decreasing: Sn+1 = an+1 + Sn Sn . Since
Sn is monotone and converges, it is bounded.
3.1.1

Cauchy Criterion Theorem

Theorem: Suppose that {an }

n=1 R satisfies n l. an 0 and n 1. an+1 an . Then

P
P
n
an converges
2 a2n converges.
n=1

n=0

Proof :
P
P
n
k
Let Sn = nk=1 ak and Tn = m
n=0 2 a2n . Notice that if m 2 then Sm = a1 + a2 +
+ a2k a1 + (a2 + a3 ) + + (a2k + + a2k+1 1 ) a1 + 2a2 + + 2k a2k = Tk .

18

3.1

Convergence Results

21-355 Notes

On the other hand, if m 2k , Sm a1 + + a2k = a1 + a2 + (a3 + a4 ) + + (a2k1 1 +

+ a2k ) 21 a1 + a2 + + 2k1 a2k = 12 Tk .
P
n
Now, if
Tn T as n and so Sm limn Tm = T ,
n=0 2 a2n converges, thenP
which means {Sm } is bounded and
n=1 an converges.
P
Similarly,
P n if n=1 an converges, then Tk 2 limn Sn = {Tk } is bounded =
n=0 2 a2n converges.
Theorem: Let p R. Then

1
n=1 np

converges p > 1.

Proof :
If p 0 the result is trivial since n1p 1 (the sequences converges to 0). Assume that
1
1
p > 0. Then (n+1)
p np , so we can apply the Cauchy criterion:

X
X
1
2n
converges

converges.
np
(2n )p

n=1

But

2n
n=0 (2n )p

n=0

1
n=0 (2p1 )n ,

and this series converges

P
P
1
Notice
n=1 n is divergent, but
n=1
we need the logarithm.
3.1.2

1
n1+r

1
2p1

< 1 p > 1.

converges r > 0. To try to find intermediate series,

Logarithm

Definition: From Supplemental Reading 3, for every 1 < b R, we define a function

logb : {x R | x > 0} R such that
1. blogb x = x (x > 0)
2. logb (1) = 0, logb b = 1
3. 0 < x < y logb x < logb y
4. logb (xz ) = z logb (x) (x > 0, z R)
5. logb is a bijection
6. lim

logb n
nr

= 0 (r R, r > 0)

Then from (6), for large n and p > 0 we know:

1
n(log1
n n(logb n)p n np = n1+p = n1+p

So

1
n(logb n)p

P
n=2

n=2

P
n=1

n1 .

is such an intermediate series.

Theorem: Let b > 1.

Proof :

n)p

1
n(logb n)p

converges

1
(logb 2)p np

In particular,

P
n=2

1
n(logb n)p

1
(logb 2)p

1
n logb n

P
n=1

converges p > 1. (n 2 = logb n > 0)

P
n=1

1
np

2n
2n (logb 2n )p

converges by Cauchy criterion, but

converges p > 1.

is divergent.
19

3.2

3.2

The number e

21-355 Notes

The number e

Lemma:

1
n=0 n!

converges.

Proof : If n 2 then:
n
X
1
1
1
=1+1+
+ +
Sn =
k!
21
n(n 1) 2 1
k=0

1
1
1
+
+ + n1
2 22
2

X
1
=1+2=3
1+
2k

1+1+

k=0

Since Sn is increasing and bounded, we know that

Definition: We set e =

P
n=0

1
n! .

1
n=0 n!

converges.

Note that e > 1.

Theorem: e = limn (1 + n1 )n .
P
1
Proof : Let Sn = nk=0 k!
, Tn = (1 + n1 )n . Then by the Binomial Theorem:
n

X
n!
1
1
Tn = (1 + )n =
n
k!(n k)! nk
k=0

1 n(n 1) 1
1 n(n 1)
=1+1+
+ +
2
2!
n
n!
nn
1
1
1
2
1
1
n1
1
)
= 1 + 1 + (1 ) + (1 )(1 ) + + (1 ) (1
2!
n
3!
n
n
n!
n
n
1
1
= Sn
1 + 1 + + +
2!
n!
Hence, lim supn Tn lim supn Sn = limn Sn = e.
OTOH, fix m N. Then for n m:

1
1
1
1
m1
(1 ) + +
(1 ) (1
)
2!
n
m!
n
n
1
1
1
1
m1
= lim inf Tn lim inf RHS 1 + 1 + lim inf (1 ) + +
lim inf (1 (1
) =1+1+
n
n
2! n
n
m! n
n
n
Tn 1 + 1 +

Then, letting m , e = limm Sm lim inf n Tn .

Thus, e lim inf n Tn lim supn Tn e = limn Tn = e.
Theorem: n 1. 0 < e Sn <

1
nn! .

Also, e R\Q is irrational.

Proof : Since Sn is increasing, 0 < e Sn is clear. The other side can be seen from algebra.
Now, suppose e Q; then e =

p
q

for p, q N, p, q 1.

Then 0 < q!(e Sq ) < 1q (q 1). Notice that q!e = q! pq = (q 1)!p N and
q!(1 + 2!1 + + q!1 ) N.
Hence q!(e Sq ) Z; but this yields an integer between 0 and 1, a contradiction. So e
is irrational.
Remark : In fact, e is transcendental.
20

3.3

More Convergence Results

3.3

21-355 Notes

More Convergence Results

1/n } is bounded. Let 0 = lim sup

1/n .
Theorem (Root Test): Suppose {an }
n |an |
n=l R and {|an |
Then the following holds:
P
1. If < 1, then
n=l an converges.
P
2. If > 1, then n=l an diverges.

3. if = 1, both convergence and divergence are possible.

Theorem (Ratio Test): Let {an }
n=l R. Then

n=l

an :

1. converges if {| an+1
an |}n=l is bounded and lim supn

|an+1 |
|an |

< 1.

2. diverges if k l. |ak | =
6 0 and |an+1 | |an |(n k).
Lemma (Summation of Parts): Let {an }
n=0 R and define:
(P
n
k=0 ak if n 0
An =
0
if n = 1
Then if 0 p < q:
q
X

an bn =

n=p

q1
X

An (bn bn+1 ) + Aq bq Ap1 bp

n=p

Theorem (Dirichlet Test): Suppose {an }

n=0 , {bn }n=0 R satisfy:
P
1. The sequence An = nk=0 ak is bounded.

2. 0 bn+1 bn (n N)
3. limn bn = 0
Then

n=0 an bn

converges.

P
n
Corollary (Alternating Series): Suppose 0 an+1 an , an 0 as n . Then
n=l (1) an
converges. Proof follows from Dirichlet Test.
P
Corollary
(Abels
Test):
Suppose
n=l an converges, bn+1 bn (n l) and bn b as n .
P
Then n=l an bn converges.

3.4

Algebra of Series

Theorem: If A =

n=l

an , B =

(1)A + B =

n=l

B N , then

(an + bn )

(2)cA =

n=l

can (c R)

n=l

Theorem: Suppose {an }

n=0 , {bn }n=0 R satisfy:

(1)

|an | converges

n=0

Then

n=0 cn

(2)

bn = B

n=0

(3) cn =

n
X
k=0

= A B converges.
21

ak bnk for n 0

3.5

Absolute Convergence and Rearrangements

Definition: The series

of the series

an ,

n=0

n
P

cn , where cn =

n=0

21-355 Notes

ak bnk , is called the Cauchy product

k=0

bn .

n=0

P
P
P
Remark : If an , bn converge, cn does not necessarily converge if neither series has convergent
absolute values.

3.5

Absolute Convergence and Rearrangements

P
|an | converges, then
n=l an converges. Proof is trivial.
P
P
Definition:
Suppose n=l an converges. If n=l |an | converges, the series converges absolutely. If
P
|an | diverges, the series is conditionally convergent.
P
P
(1)n
(1)n
Example:
is conditionally convergent, while
n=1
n=1 n2 is absolutely convergent.
n
Proposition: If

n=l

Lets try to manipulate the series without being careful.

=

X
(1)n+1

n=1

= lim (Sk =
k

=1

1 1 1
+ +
2 3 4

k
X
(1)n+1

n=0

) = lim (S2k =
k

2k
X
(1)n+1
n=0

1
1 1
1
1
but: S2k = (1 ) + ( + + (
)>0
2
3 4
2k 1 2k
Hence, > 0. But the next step is questionable:
2 =
?

X
(2)(1)n+1
n=1

X
k=0

X
k=0

X 2
2

2k + 1
2k
k=1

X
X
2
1 X 1
1

=
2k + 1
k
2k + 1
2k
k=1

= 2 = > 0

k=0

k=1

a contradiction!

Problem: rearrangement is a delicate issue.

P
Definition: Let : {m P
Z | m l} {m Z | m l} be a bijection. The series
n=l a(n) is
called a rearrangement of
a
.
n=l n
P
P
Theorem: If n=l an is absolutely convergent, then every rearrangement converges to
n=l an .
Proof : Let  > 0.
P

Pk

an converges absolutely, N l. k m N =
n=m |an | < 2 .
P

Let k :
n=m |an | 2 < .
Now choose MP N suchPthat {l, l + 1,
. . , N } {(l), (l + 1), . . . , (M )}. Then
P.
m
m
m M = | n=l an n=l a(n) | n=N |an | < .
Pm
P
Hence lim
and from this we deduce
m ( n=l an
n=l
P
Pam(n) ) = 0P

limm m
a
=
lim
a
=
m
n=l (n)
n=l n
n=l an .
Since

n=l

22

3.5

Absolute Convergence and Rearrangements

21-355 Notes

When a series is only conditionally convergent, the situation is vastly worse.

P
Theorem: Suppose
c R.
n=0 an is conditionally convergent. Let P
There exists a rearrangement (bijection) : N N such that
n=0 a(n) = c.
P
Lemma: Suppose n=0 an is conditionally convergent and set:
(
(
an if an > 0
an if an < 0
bn =
cn =
0
if an 0
0
if an 0
Then

n=0 bn

and

n=0 cn

both diverge.

P
Proof
IfP bn converges,
then cn = bn an =
P : Suppose
P
Pnot; one of the series is convergent.
P
P
cn = bn an ; but |anP
| = bn +cn and so |an | = bn + cn is convergent, a contradiction.
A similar argument holds if
cn converges.
Rearrangement Theorem Proof :

Let {a+
n }n=0 denote the subsequence of {bn | bn > 0 or bn = 0 an = 0}. Let {an }n=0
denote the subsequence of {cn | cn > 0} (from last lemma). Note:

1. a+
n 0, an 0 since an 0 = bn 0, cn 0.
P +
P
P
P
2.
an and a
bn , cn respectively.
n both diverge because they differ by 0 from

P +
m1 =
Set m0 = n0 =
Since
an diverges we may use the well-ordering
Pk
Pm1 + principle:
P1.
< c}.
a
a

>
c}.
Similarly,
n
=
min{k

N
|
min{k N | kn=0 a+
1
n
n=0 n
n=0 n
Next, if mp and np are known, we set:

p1 X
p1 X
nl
ml
k

X
X
X
+

a
>
c
a
+
mp+1 = min k N |
a+

j
j
j

j=1+mp
l=0 j=1+nl
l=0 j=1+ml

mp+1
p1 X
p1 X
nl
ml
k

X
X
X
X

+
a
<
c
np+1 = min k N |
a+

a
+
a

j
j
j
j

l=0 j=1+ml

l=0 j=1+nl

j=1+mp

j=1+np

+
) + (a+
Consider the series (a+
1+m1 + + am+2 ) (a1+n1 +
1 + + am1 ) (a1 + + an+1P

+ a
n2 ) + . This is clearly a rearrangement of
n=0 an .
Pmp+1
P
np+1
+

Write Ap = l=1+mp al , Ap = l=1+np al , and let Sj denote the j th partial sum of

the rearrangement.
P
P
By construction, lim supj Sj = lim supp ( p+1
A+
pl=0 A
l=0
l
l ) and
Pp
Pp
+

lim inf j Sj = lim inf p ( l=0 Al + l=0 Al ).

P
Pp
Pp+1 + Pp+1
+

Also, c < p+1

l=0 Al
l=0 Al < c + amp+1 and c anp+1 <
l=0 Al
l=0 Al < c.
Pp+1 + Pp
Pp
+
Thus,
by the squeeze lemma, limp ( l=0
Al l=0 A
l ) = limp ( l=0 Al
Pp
P

n=0 a(n) = c.
l=0 Al ) = c, and so limj Sj = c =

Remark : One can also rearrange such that

a(n) = .

23

21-355 Notes

Topology of R

Our goal in Section 4 is to develop some tools for understanding the topology of R, which is a
sort of generalized qualitative geometry.

4.1
4.1.1

Open and Closed Sets

Open Sets

Definition:
1. For a, b R with a b, we define:
(a, b) = {x R | a < x < b}

[a, b) = {x R | a x < b}

(a, b] = {x R | a < x b}

[a, b] = {x R | a x b}

2. For x R and  > 0, we set B(x, ) = (x , x + ) and B[x, ] = [x , x + ].

We call the set B(x, ) a neighborhood of x or a ball of radius  centered at x.
3. A set E R is open if x E.  > 0. B(x, ) E.
In other words, every point in E has a neighborhood contained in E.
Examples:
1. is vacuously open.
2. R is open because x R. B(x, 1) R.
3. If a < b then (a, b) is open.
Proof : Fix x (a, b) and let  = min{x a, b x} > 0. Then a x  < x <
x +  b by construction, and B(x, ) (a, b).
4. If a < b then [a, b) is not open.
Proof : For x = a we know that  > 0. a 
/ [a, b) and hence B(a, ) 6 [a, b).
5. [a, b] is not open, nor is (a, b] by previous argument.
6. E = {a} is not open.
7. E = { n1 | n N, n 1} is not open:  > 0. B(1, ) 6 E.
S
Lemma: If E R is open A (some index set), then A E is open.
S
Proof : SLet x A E . Then x E0 for some 0 A. Since E0 is open,  > 0. B(x, )
E0 A E .
T
Lemma: If Ei R is open for i [n], n N, then ni=1 Ei is open.
1
Remark
: Infinite intersections of open sets need not be open. Let En = ( 1
n , n ), n 1.
T
Then n=1 En = {0} which is closed.

4.1.2

Closed Sets

Definition: We say E R is closed iff E c = R\E is open.

Lemma: E is open E c is closed (by definition).
Examples:
24

4.1

Open and Closed Sets

21-355 Notes

1. is closed because c = R is open.

2. R is closed because Rc = is open.
3. [a, b] is closed because [a, b]c = (, a) (b, ) is the union of open sets, and thus
open.
4. [a, b) and (a, b] are not closed because [a, b)c = (, a) [b, ) and B(b, ) 6
[a, b)c ( > 0).
5. {a} is closed since {a}c = (, a) (a, ), both open sets.
6. Suppose E R is finite. Write E = {ai | i [n]} where a1 < a2 < . . . < an . Then
E c = (, a1 ) (a1 , a2 ) (an1 , an ) (an , ), all of which are open.
S
1
1
7. E = { n1 | n N, n 1} is not closed. E c = (, 0]
n=1 ( n+1 , n ) (1, ) is not
open because B(0, ) E = { n1 | 1 < n} =
6 = B(0, )
/ Ec.
S
1
1
8. E = {0} { n1 | n 1} is closed, as E c = (, 0)
n=1 ( n+1 , n ) (1, ) is open.
Lemma:
T
1. If E R is closed A, then A E is closed.
S
2. If Ei R is closed i [n] then ni=1 Ei is closed.
Proof : The complement is the union of Ec (open by claim), which is open by previous lemma.
Remark : Example (7) shows that infinite unions of closed sets need not be closed.
4.1.3

Limit Points

Definition: Let E R.
1. A point x R is a limit point of E iff.  > 0. (B(x, ) E)\{x} =
6 .
2. A point x E is called isolated if it is not a limit point.
Example: E = { n1 | n 1}. 0 is a limit point, but

1
n

1
E is isolated, since B( n1 , n(n+1)
) E = { n1 }.

Theorem: Let E R. E is closed every limit point of E is contained in E.

Proof :
= :
Assume E is closed and x R is a limit point of E. If x E c then, since E c is open,
 > 0. B(x, ) E c = B(x, ) E = . But this contradicts the fact that x is a
limit point of E; thus x E.
=:
Suppose E is not closed; then E c is not open and so  > 0. x E c . B(x, ) E 6= .
Since x E c , (B(x, ) E)\{x} = B(x, ) E 6= and hence x is a limit point of E.
Then x E E c , a contradiction; and E is closed.
Definition: Let {xn }
n=l S for some set S. We say {xn } is eventually constant if
N l. xn = xN (n N ).
Proposition: Let E R. Then x is a limit point of E {xn }
n=1 E such that the sequence
is not eventually constant and xn x as n .
Proof :
25

4.1

Open and Closed Sets

21-355 Notes

= :
Suppose x is a limit point of E, i.e.  > 0. (B(x, ) E)\{x} 6= . Set r1 = 1 and
choose x1 E such that x1 (B(x, r) E)\{x}.
Set rn = min( n1 , |x xn1 |) and choose xn (B(x1 , rn ) E)\{x}.
1
Then n 1. {xn }
n=1 E and |x xn1 | < |x xn | and |x xn | < n . It follows {xn }
is not eventually constant, and xn x as n .

=:
Let  > 0. N 1. n N = |x xn | < . Then {xn | n N } B(x, ) E.
If {xn | n N } = {x} then {xn } is eventually constant, a contradiction. Hence
6= {xn | n N }\{x} (B(x, ) E)\{x} = (B(x, ) E)\{x} 6= , and hence x
is a limit point.
Corollary: Let E R. The following are equivalent (proof follows from last theorem):
1. E is closed.
2. If x R is a limit point of E, x E.
3. If {xn }
n=l E is such that xn x as n , then x E.
Corollary: Let E R and E 6= . Suppose E is closed.
1. If E is bounded above, then sup E E, i.e. sup E = max E.
2. If E is bounded below, then inf E E, i.e. inf E = min E.
4.1.4

Closure, Interior, and Boundary Sets

Definition: Let E R.
1. Let O(E) = {V R | V E and V is open} P(R)
C(E) = {C R | E C and C is closed} P(R).
Note that O(E) and R C(E).
S
2. We define E 0 = V O(E) V , and call this set the interior of E.
=T
We define E
CC(E) C, and call this set the closure of E.
3. We define E = E\E 0 to be the boundary of E.
Theorem: Let E R. The following hold:

1. E 0 E E
E are closed.
2. E 0 is open and E,
3. For every x E, x E 0 x E.
4. E = {x R |  > 0. B(x, ) E 6= and B(x, ) E c 6= }.

5. E is open E = E 0 , E is closed E = E.
Proof :
1. Trivial.
is an arbitrary intersection
2. E 0 is an arbitrary union of open sets and thus open; E
0

of closed sets, so its closed. E = E\E = E (R\E 0 ) is the intersection of two

closed sets, so its closed.
26

4.2

Compact Sets

21-355 Notes

3. Trivial.
4. Suppose x E. Show the two properties of the set are satisfied via contradiction.
Next, assume x in the set, and show that x E.
5. Trivial.
Corollary: Let E R. Then E is closed E E.
= E E
E. On the other hand, if E E then
Proof : E is closed = E = E
0

E E = E E E, so E = E.
Theorem (Bolzano-Weierstass, Part 2): Let E R be infinite and bounded. Then E has a
limit point.
Proof : Since E is infinite we may construct a non-eventually-constant sequence {xn }
n=0 E. We
+
do so by choosing x0 E arbitrarily, and xn E\{x0 , . . . , xn1 } for any n N . Since E is
bounded, the sequence is too, so B-W implies there exists a convergent subsequence {xnk }
k=0 E.
This subsequence is not eventually constant by construction, so its limit is a limit point.

4.2

Compact Sets

Definition:
1. Let A be some index set and assume A. V R. We write V = {V }A for
the collection of all of these subsets.
S
2. If E R and E A V , then we say V is a cover of E.
3. If V R is open A and V is a cover of E, we say V is an open cover.
4. Let V be a cover of E. We say W = {V }A0 is a subcover of E if A0 A and W
is a cover of E.
5. Let V be a cover of E. If A is finite, then W = {V }A0 is a finite subcover of E,
if W is a subcover of E.
Examples:
1. Every E R admits a cover: E =

xE {x}.

2. Every E R admits an open cover: E

xE

B(x, ) for  > 0.

3. If E is finite and V is an open cover, we claim there is a finite open subcover.

Indeed,
S write E = {ai | 1 i n} and choose Vi such that ai Vi . Then
E ni=1 Vi and {Vi }ni=1 {V }A . Hence every open cover of a finite set
admits a finite open subcover.
1
4. E = { n1 | n 1}. V = {B( n1 , n(n+1)
}
n=1 is an open cover of E. Note that
1
1
1
B( n , n(n+1) ) E = n , so there does not exist a finite subcover.

5. E = {0}{ n1 | n 1}. Suppose V is an open cover of E. Since 0 E, 0 A. 0

V0 . Since V0 is open,  > 0. B(0, ) V0 . Then B(0, ) E = { n1 ||; n N }
where N = min{n N | n 1 }. Hence E\B(0, ) = { n1 | 1 n N }. There
S
exist Vn for n [N ] such that n1 Vn . Then E N
n=0 Vn and E has a finite
subcover.
1
1
6. Let a < b and E = (a, b). Then V = {(a + n+1
, b n+1
)}nN is an open cover of
E. Since these intervals are nested, there cannot be a finite subcover.

27

4.2

Compact Sets

21-355 Notes

Definition: Let E R. We say that E is compact if every open cover of E admits a finite
subcover.
Examples:
1. is trivially compact.
2. R is not compact because V = {B(0, n)}nN is an open cover that clearly does not
admit a finite subcover of R.
3. Any finite set E R is compact.
4. (a, b) for a < b is not compact.
5. { n1 | n 1} is not compact.
6. {0} { n1 | n 1} is compact.
Notice in each of our examples of compact sets that the set is closed and bounded.
4.2.1

Heine-Borel Theorem

Theorem: Let K R. Then K is compact K is closed and bounded.

Proof :
= Suppose K is compact.
S
S
Notice that
n=1 B(0, n).
n=1 B(0, n) = R (since R is Archimedean) and so K R =
ThenS{B(0, n)}
is
an
open
cover
of
K.
Since
K
is
compact,

a
finite
subcover :
n=1
B(0,
n
)
for
some
m

N.
K m
i
i=1
S
Set r = maxi[m] ni . Then K m
i=1 B(0, ni ) B(0, r) = K is bounded.
Now we show K is closed. Let x K C . For each y K we set ry = 21 |x y| > 0. Then
B(y, ry ) B(x, ry ) = (y K). Also, {B(y, ry )}yK is an open cover.
S
B(yi , ryi ). Set r = mini[n] ri > 0
K compact = a finite subcover: K ni=1S
and notice that B(yi , ryi ) B(y, r) = . Hence ni=1 B(yi , ryi ) B(x, r) = =
K B(x, r) = = B(x, r) K C . This means that K C is open and so K is closed.
= (Heine-Borel) Suppose K is closed and bounded. If K = were done, so suppose K 6= .
Notice that K bounded = inf K, sup K R, and K closed = inf K, sup K K.
In particular, sup K = max K, inf = min K. Let V be an open cover of K.
Let E = {x K | V admits a finite subcover of K [inf K, x]} K. Notice that
K [inf K, inf K] = {inf K} is a finite set and hence compact; thus V admits a finite
subcover of K [inf K, inf K]. Hence inf K E, and so E 6= . Clearly E is bounded
above by sup K. By LUB property, sup E R and sup E sup K.
We want to show sup E = sup K = max E. Notice that n 1. xn E K such
that sup E n1 < xn sup E. Then xn sup E as n , and so sup E K (since
K is closed).
Write V = {V }A . Since sup E K, 0 A such that sup E V0 . But V0 is
open so  > 0. B(sup E, ) V0 . By definition, x E. sup E S< x sup E.
n
Hence V admits a finite
Sn subcoverof K [inf K, x], i.e. K [inf K, x] i=1 Vi . Then
K [inf K, sup E] i=0 Vi = sup E E = sup E = max E.
28

4.3

Connected Sets

21-355 Notes

S
Assume for sake of contradiction that max E < max K. Let K 0 = K\ ni=0 Vi . K 0
0
is
Snclosed since its the intersection of closed sets. K 6= since otherwise K
i=0 Vi = max E = max K.
Let y = inf K 0 = min K 0 (since S
K 0 is closed) and note that y > max E. Then K
0
[inf K, y] = K[inf K, min K ] ni=0 Vi {y}. But since y K 0 K, Vn+1 V such
S
that y Vn+1 . Hence K [inf K, y] n+1
i=0 Vi = y E = max E < y max E,
a contradiction. We then deduce that max E = max K = K = K [min K, max K]
is covered by a finite subcover of V; thus, K is compact.
Corollary:
1. If K R is compact and E R is closed, then E K is compact.
2. If K R is compact and E K is closed, then E is compact.
S
3. If Ki R is compact for i [n], then ni=1 Ki is compact.
T
4. If K R is compact A, then A K is compact.

4.3

Connected Sets

= A B = .
Definition: We say two sets A, B R are separated if A B
A set E R is disconnected if E = A B such that a 6= , B 6= and A, B are separated.
If a set is E R is not disconnected, we say its connected.
Examples:
1. (0, 1) and [1, 2) are not separated, though they are disjoint, since (0, 1) [1, 2) =
[0, 1] [1, 2) = {1} =
6 .
2. (a, b) and (b, c) for a < b < c are separated, since (a, b) (b, c) = = (a, b) (b, c).
Then (a, c)\{b} is disconnected, since (a, c)\{b} = (a, b) (b, c).
3. Similarly, a R. (, a) an (a, ) are separated.
Then R\{a} = (, a) (a, ) is disconnected.
Theorem: Let E R. Then E is connected (x, y E and x < z < y = z E).
Proof :
2 = 1:
If (2) is false then x, y E and z (x, y) such that z
/ E. Then E = Lz Rz for
Lz = E (, z) and Rz = E (z, ). Since x Lz , y Rz , and Lz (, z) and
Rz (z, ), it follows that Lz and Rz are separated. Hence E is disconnected.
1 = 2
= .
Suppose E is disconnected. Write E = A B with A, B 6= and A B = A B
Let x A and y B. Without loss of generality, we assume x < y.
Let z = sup(A [x, y]). Clearly z A and so z
/ B = z 6= y = x z y. If

z
/ A then z 6= x = x < z < y and z
/ A B = E. Otherwise, if z A, then z
/ B.
C
is closed, so B
is open; and hence we can find w such that z < w < y, w
B
/ B, and
w
/ A. Then x < w < y and w
/ A B = E. In all cases, then, 2 is true.
Corollary: R, (, a), (, a], (a, ), [a, ), (a, b), (a, b], [a, b), and[a, b] are all connected.
29

21-355 Notes

Continuity

5.1

Limits of Functions

Definition: Let E R, f : E R, and p R be a limit point. Let q R.

We say limxp f (x) = q or f (x) q as x p iff  > 0. > 0. x E 0 < |x p| < =
|f (x) q| < .
Examples:
1. E = [0, 1], f (x) = x. Let p = 12 . limx 1 f (x) = 12 .
2

Proof : Let  > 0; choose =  > 0. Then x [0, 1] and 0 < |x 21 | < S =
|f (x) 12 | < .
2. E = [0, 1], f (x) = x (for x 6= 21 ), f (x) = 37 (for x = 12 ).
By the proof of (1), the claim still holds.
3. f (x) = xn on E = (0, 1) for 2 n N. 0 is a limit point of E; we claim
limx0 xn = 0.
Proof : Let  > 0; choose = 1/n > 0. Then x (0, 1) and 0 < |x 0| < =
0 < x < = 0 < xn < n =  = |f (x) 0| = xn < .
4. limxp x = p whenever p is a limit point of E.
5. If x E. f (x) = 1 then limxp f (x) = 1 whenever p is a limit point of E.
2

6. Let E = R and f (x) = cos(x). From HW6, | cos(x) 1| x2 ex . We claim

limx0 cos(x) = 1.
p
Proof : Let  > 0. Choose

=
min(1,
/e) > 0. Then for x R, 0 < |x 0| <
p
= |x| < min(1, /e) = | cos(x) 1| x2 e1 (since |x|2 < 1 =
p
2
e|x| e1 ) = | cos(x) 1| < 2 e ( /e)2 e = .
7. E = { n1 | n 1}, p = 0. Let f (x) = x1 for x E. We claim limx0 f (x) does not
exist.
Proof : Suppose not. Then for  = 1. > 0. x E, 0 < |x 0| < =
1
|f (x) q| < 1. But x E, |x| < = x = n1 , 1 < n, and |f (x) q| = | 1/n
q| =
|n q| < 1, which is a contradiction.
Definition: Let f : E R for some E R. If A E we define f (A) = {f (x) | x A} R as
the image of A under f . If B R we define f 1 (B) = {x E | f (x) B} as the pre-image of B
under f .
Lemma: Suppose f : E R. Then A B E = f (A) f (B), and
A B R = f 1 (A) f 1 (B) E.
5.1.1

Divergence Criteria

Theorem (Divergence Criteria): Let E R, f : E R, p be a limit point of E, q R. The

following are equivalent:
1. limxp f (x) = q
2. For every open set V R such that q V , an open set U R with p U such
that f (U E\{p}) V . (Topological characterization)

30

5.2

Continuous Functions

21-355 Notes

3. If {xn }
n=l E satisfies xn 6= p (n l) and xn p as n , the sequence
{f (xn )}
n=l R converges and f (xn ) q as n . (Sequential characterization)
Proof :
(1) = (2) :
Assume (1) and let V R be open with q V . Since V is open,  > 0. B(q, ) V .
Since limxp f (x) = q, > 0. x E 0 < |x p| < = |f (x) q| < . Let
U = B(p, ) (an open set). Then x U E\{p} = x E |x p| < =
|f (x) q| <  = f (x) B(q, ) V . So f (U E\{p}) V as desired.
(2) = (3):
Assume (2) and let {xn }
n=l E satisfy xn 6= p, xn p. Let  > 0 and set V = B(q, )
(open). From (2), open U such that f (U E\{p}) V and p U . Since U is open,
> 0. B(p, ) U . Since xn p as n , N l. n N = ) < |xn p| <
= xn U E\{p} = f (xn ) V = B(q, ). Hence n N = |f (xn ) q| < ,
and f (x) q as n .
(1) = (3):
Suppose (1) is false; then  > 0. > 0. x E with 0 < |x p| < such that
|f (x) q| . For n N, n 1, set = n1 to find xn E such that 0 < |xn p| < n1
and |f (xn ) q| . Clearly, {xn }
n=1 E satisfies xn 6= p, xn p. But f (xn ) does
not converge to q. Hence (3) fails.
Corollary: If E R, f : E R, p is a limit point of E, and limxp f (x) = q, then q is unique.
Proof : Limits of sequences are unique, so this follows from (3) in Divergent Criteria theorem.
Corollary (Algebra of limits): Let E R, f, g : E R, p be a limit point of E. Assume
limxp f (x) = q1 , limxp g(x) = q2 . The following hold:
1. If , R then limxp (f (x) + g(x)) = q1 + q2
2. limxp f (x)g(x) = q1 q2
3. If q2 = limxp g(x) 6= 0, then

f
g

point of E\g 1 ({0}), and limxp

: E\g 1 ({0}) R is well-defined, p is a limit

f (x)
g(x)

q1
q2

Proof : All follow from the algebra of sequential limits and (3) in the Theorem.
As an application of this, we get a large class of limit examples.
Corollary: Let P : E R be a polynomial, i.e. P (x) = a0 +a1 x+ +an xn for some n N, ai R
for i [n]. If p is a limit point of E, then limxp P (x) = P (p).
Proof : We know limxp 1 = 1, limxp x = p. Algebra of limits (2) and simple induction show
limxp xk = pk (k N+ ). Then algebra of limits (1) and another induction argument prove
limxp P (x) = limxp (a0 + a1 x + + an xn ) = limxp (a0 + a1 p + + an pn ) = P (p).

5.2

Continuous Functions

Definition: Let E R, f : E R, p E. We say f is continuous at p iff:

 > 0. > 0. x E |x p| < = |f (x) f (p)| < 
31

5.2

Continuous Functions

21-355 Notes

If f : E R is continuous at each p E we say f is continuous on E.

Remarks:
1. In order to be continuous at p E, f must be defined at p. Contrast this to
limxp f (x), in which case p need only be a limit point of E.
2. Informally one can think of continuous functions as those approximated well near
p by f (p), i.e. f (x) f (p) when x p.
3. In the definition, the value of may depend on the point p. If a function is
continuous on E then for a given  > 0 the = (p) may vary greatly as p varies.
4. If p E is isolated (not a limit point of E), then f is vacuously continuous at p:
x E, |x p| < for small enough = x = p.
Example:
We saw last time that limxp P (x) = P (p) for all polynomials P : R R. Hence
 > 0. > 0. x R, 0 < |x p| < = |P (x) P (p)| < . Hence P is continuous
at p.
Theorem: Let E R, f : E R, p E be a limit point of E. Then:
f is continuous at p lim f (x) = f (p)
xp

Corollary (Algebra of Continuity): Let E R, f, g : E R, and p E. Assume that f, g are

continuous at p. Then the following hold:
1. If , R then f + g is continuous at p.
2. f g is continuous at p.
3. If g(p) 6= 0 then

f
g

: E\g 1 ({0}) R is well-defined and continuous at p.

Proof : If p is isolated, the claim is vacuously true. Assume p is not isolated, i.e. p is a limit point
of E. Then the last theorem and algebra of limits gives the result.
Corollary: Let E R, f, g : E R. If f, g are continuous on E, then:
1. If , R then f + g is continuous on E.
2. f g is continuous on E.
3. If g(x) 6= 0 (x E), then

f
g

is continuous on E.

Theorem: Let E, F R, f : E R, g : F R. Assume f (E) F , f is continuous at p E,

and g is continuous at f (p) F . Then g f : E R (where (g f )(x) = g(f (x))) is continuous at
p. Moreover, if f is continuous on E and g is continuous on F , then g f is continuous on E.
Proof : Let  > 0.
Since g is continuous at f (p), > 0. y F and |y f (p)| < = |g(y)g(f (p))| < .
Since f is continuous at p, > 0. x E, |x p| < = |f (x) f (p)| < .
Since f (E) F we know that x E, |x p| < = f (x) F, |f (x) f (p)| < =
|g(f (x)) g(f (p))| < . Hence, g f is continuous by definition.
Examples:
32

5.3

Compactness and Continuity

21-355 Notes

1. exp, cos, sin : R R are continuous on R (proof in HW). YAlso, log : (0, ) R
is continuous on (0, ).
2. Let R and set f : (0, ) R via f (x) = x . Notice that f (x) = exp( log x).
Since log and exp are continuous, f (x) = x is continuous.
Definition: Let E R and A E. We say A is relatively open in E iff A = U E for some open
set U R. Similarly, we say A is relatively closed in E iff A = C E for some closed C R.
Proposition: Let A E R. The following hold:
1. A is relatively open in E x A.  > 0. B(x, ) A E.
2. A is relatively closed in E A = B C E for some relatively open B E.
Theorem (Continuity Criteria): Let E R, f : E R. The following are equivalent:
1. f is continuous on E.
2. If p E is a limit point of E, then limxp f (x) = f (p).
3. If p E is a limit point of E and {xn }
n=l E satisfies xn p as n , then
f (xn ) f (p) as n .
4. If V R is open, then f 1 (V ) E is relatively open in E.
5. If C R is closed, then f 1 (C) E is relatively closed in E.
Proof :
(1) (2) (3) follows from the sequential criterion of limits, previous theorem.
(4) (5) follows since f 1 (V C ) = (f 1 (V ))C E.
(1) = (4):
Let V R be open and choose p f 1 (V ). Since V is open,  > 0. B(f (p), ) V .
It suffices to show, via previous proposition, that > 0. B(p, ) E f 1 (V ). Since
f is continuous on E, > 0. x E, |x p| < = |f (x) f (p)| < . That
is, x B(p, ) E = |f (x) f (p)| <  = f (x) B(f (p), ) V . Hence
B(p, ) E f 1 (V ).
(4) = (1):
Let p E,  > 0, and V = B(f (p), ). Then f 1 (B(f (p), )) E is relatively open in
E = (by previous proposition) > 0. B(p, ) E f 1 (B(f (p), )). Then x E
and |x p| < = f (x) B(f (p), ) = |f (x) f (p)| < . Since , p were arbitrary,
we deduce f is continuous on E.

5.3

Compactness and Continuity

Theorem: Suppose K R is compact and f : K R is continuous on K. Then f (K) is compact.

Proof :
Note that for E R, f (f 1 (E)) E and E f 1 (f (E)). Let {V }A be an open
cover of f (K). Since f is continuous and V is open, f 1 (V ) is relatively open in
K = f 1 (V ) = U K for some open U R.
Since {V }A cover f (K), we see that {f 1 (V )}A is a cover of K. Then {UT }A is
an open cover
compact, there exists a finite
subcover: K Sni=1 Ui .
Sn of K. Since KSis
S
Then K i=1 Ui K = ni=1 f 1 (Vi ) = f (K) ni=1 f (f 1 (Vi )) ni=1 Vi .
As we have extracted a finite open subcover of f (K), f (K) is compact.
33

5.4

Continuity and Connectedness

21-355 Notes

Extreme Value Theorem: Let K R be compact and f : K R be continuous. Then

x0 , x1 K such that f (x0 ) f (x) f (x1 ) (x K). That is, f (x0 ) = minxK f (x) = min f (K)
and f (x1 ) = maxxK f (x) = max f (K).
Proof : From last theorem, we know f (K) is compact, so its closed and bounded. From a previous theorem, closed and bounded sets contain their infimum and supremum (and thus min, max).
Definition: Let E R and f : E R. We say f is uniformly continuous on E iff:
 > 0. > 0. x, y E |x y| < = |f (x) f (y)| < 
Remarks:
1. f is uniformly continuous on E = f is continuous on E.
2. The key difference is that for uniform continuity, > 0 works for all points in E.
Examples:
1. Let E = (0, 1) and f (x) = x1 . Its trivial that f is continuous on E, but it is not
uniformly continuous.
Proof : Suppose it is; then for  = 12 , > 0. x, y (0, 1) |x y| < =
1
|f (x) f (y)| < 21 . Choose N N such that 1 < N . Then x = n1 , y = n+1
satisfy
1
|x y| = n(n+1)

contradiction.

1
n2

< if n N . Then

1
2

> |f (x) f (y)| = |n (n + 1)| = 1, a

Definition: A function f : E R is Lipschitz if x, y E. k > 0. |f (x) f (y)| k|x y|.

Claim: If f is Lipschitz, it is uniformly continuous. Proof: let = k .
Theorem: Let K R be compact and f : K R be continuous. Then f is uniformly continuous
on K.

5.4

Continuity and Connectedness

Theorem: Let E R be connected and f : E R be continuous on E. If X E is connected,

then f (X) is connected.
Intermediate Value Theorem: Let a < b R. Suppose f : [a, b] R is continuous. If
f (a) < c < f (b) or f (b) > c > f (a) for some c R, then x (a, b). f (x) = c.

5.5

Discontinuities

Lemma: If p is a limit point of E R then p is a limit point of Ep+ = E (p, ) or Ep =

E (, p).
Definition: Let E R, f : E R, p be a limit point of E, q R.
1. If p is a limit point of Ep , we say limxp f (x) = q  > 0. > 0. x Ep ,
0 < p = |f (x) q| < .
2. If p is a limit point of Ep+ , then limxp+ f (x) = q  > 0. > 0. x Ep+ ,
0 < x p < = |f (x) q| < .

34

5.6

Monotone Functions

21-355 Notes

Proposition: If p is not a limit point of Ep+ then limxp f (x) = limxp f (x). If p is not a limit
point of Ep then limxp f (x) = limxp+ f (x).
Proposition: If p is both a limit point of either Ep+ or Ep , then
lim f (x) = q lim f (x) = lim f (x) = q

xp

xp

xp+

Definition: Suppose E R, f : E R, p E is a limit point of E. Suppose further that p is not

a point of continuity of f .
1. We say f has a simple discontinuity of p if
p is not a limit point of Ep+ and limxp f (x) exists,
p is not a limit point of Ep and limxp+ f (x) exists, or
p is a limit point of Ep+ and Ep and limxp+ f (x), limxp f (x) both exist.
2. Otherwise, we say f has an essential discontinuity of p.

5.6

Monotone Functions

Definition: Let E R and f : E R. We say:

f is non-decreasing (increasing) if x, y E and x < y = f (x) f (y) (f (x) < f (y)), and
f is non-increasing (decreasing) if x, y E and x < y = f (y) f (x) (f (y) < f (x)).
If f is non-increasing or non-decreasing, f is monotone.
Theorem: Suppose f : (a, b) R is monotone, and let p (a, b). Then limxp f (x) and
limxp+ f (x) both exist. Moreover, if f is non-decreasing, then
lim f (x) = sup f ((a, p)) f (p) inf f ((p, b)) = lim f (x)

xp

xp+

Corollary: If f : (a, b) R is monotone, then f has no essential discontinuities.

Example: f (x) = bxc is non-decreasing and f has countably many simple discontinuities.
Theorem: If f : (a, b) R is monotone, then f has at most countably many simple discontinuities.

6
6.1

Differentiation
The Derivative

Definition: Assume f : [a, b] R for a < b R. For all x [a, b], the function : (a, b)\{x} R
(x)
via (t) = f (t)f
is well-defined, and x is a limit point of (a, b)\{x}. If limtx (t) exists we
tx
0
write f (x) = limtx (t) and say that f is differentiable at x.
We define f 0 : {x [a, b] | x is differentiable at x} R to be the derivative of f . If f is differentiable x E [a, b], we say f is differentiable on E.
Definition (General): Let E R, f : E R, and x E be a limit point of E. Define
(x)
: E\{x} R by (t) = f (t)f
. If limtx (t) exists we say f is differentiable at x, and write
tx
0
f (x) = limtx (t).
Proposition (locality of derivative): Suppose f : E R, g : F R, x E F is a limiit
point of E F , and that f and g are differentiable at x. If f = g on E F then f 0 (x) = g 0 (x).
This shows that f 0 (x) only depends on the value of f near x.
35

6.2

Mean Value Theorems

21-355 Notes

Proposition (Newtonian approximation): Let f : E R, x E be a limit point of E, and

L R. Then the following are equivalent:
1. f is differentiable at x and f 0 (x) = L
2.  > 0. > 0. t E |x t| < = |f (t) (f (x) + L(t x))| < |t x|
Proof follows from definition of limtx (t). Newtons approximation says differentiable
functions are those that can be well-approximated by affine functions +x. Continuous functions are those well-approximated by constants, while differentiable functions
are well-approximated by the next simplest function.
Theorem: Suppose f : E R, x E is a limit point of E, and f is differentiable at x. Then f is
continuous at x.
Proof : By definition, if t E\{x} then f (t)f (x) = (t)(tx). Then f (t) = f (x)+(t)(tx) and
hence limtx f (t) = f (x) + limtx (t)(t x) = f (x) + f 0 (x)0 = f (x). By the limit chracterization
of continuity, we deduce that f is continuous at x.
Remark : The converse fails. Let f (x) = |x| on R. Since ||x| |y|| |x y|, f is Lipschitz
and hence uniformly continuous. However, for x = 0, t > 0 = (t) = |t|0
t0 = 1 and
t0
t < 0 = (t) = t0 = 1. Then limt0 (t) = 1 6= limt0+ (t) = 1, so f 0 (0) does
not exist.
Theorem (Algebra of Derivatives): Let f, g : E R be differentiable at x E. Then:
1. f + g : E R is differentiable at x and (f + g)0 (x) = f 0 (x) + g 0 (x)
2. f g : E R is differentiable at x and (f g)0 (x) = f (x)g 0 (x) + f 0 (x)g(x)
3. If g(x) 6= 0 then
g(x)f 0 (x)f (x)g 0 (x)
g(x)2

f
g

: E\g 1 ({0}) R is differentiable at x and ( fg )0 (x) =

Examples:
1. f (x) = + x on R = f 0 (x) = limtx

f (t)f (x)
tx

= (x R).

2. f (x) = xn for n N = f 0 (x) = nxn1 . Proof by induction.

PN
P
n1 .
n
0
3. Every polynomial P (x) = N
n=0 nan x
n=0 an x is differentiable, and P (x) =
P (x)
4. R(x) = Q(x)
is dfiferentiable when P, Q are polynomials at points p R where
Q(p) 6= 0.

Theorem (Chain Rule): Suppose f : E R is differentiable at x E, f (E) F , and g : F R

is differentiable at f (x) F . Then gf : E R is differentiable at x and (gf )0 (x) = g 0 (f (x))f 0 (x).

6.2

Mean Value Theorems

Definition: Let f : E R. We say that f has a local maximum at x E if > 0. t E and

|x t| < = f (t) f (x). We say f has a local minimum at x E if f has a local maximum.
If f has either a local max or min at x E, we say f has a local extremum at x.
Theorem: Suppose f : E R is differentiable at x E and x is a limit point of both Ex+ and
Ex . If f has a local extremum at x, then f 0 (x) = 0.

36

6.3

Darbouxs Theorem

21-355 Notes

Proof : It suffices to assume that f has a local max at x. Let > 0 such that t E and
(x)
|x t| < = f (t) f (x). Then t E, 0 < x t < = f (t)f
0 and 0 < t x < =
tx
f (t)f (x)
tx

0. So limtx

f (t)f (x)
tx

= f 0 (x) 0, limtx

f (t)f (x)
tx

= f 0 (x) 0, and thus f 0 (x) = 0.

Remark : The result is false if x is not a limit point of either Ex+ or Ex . Consider f (x) = x on
E = [0, 1]; f has a local min at x = 0, local max at x = 1, but f 0 (x) = 1x [0, 1].
Theorem (Monotonicity part 1): Let f : E R be differentiable at x E.
1. If f is non-decreasing on E, then f 0 (x) 0.
2. If f is non-increasing on E, then f 0 (x) 0.
Cauchys Mean Value Theorem: Suppose that f, g : [a, b] R are continuous on [a, b], differentiable on (a, b). Then x (a, b). (g(b) g(a))f 0 (x) = (f (b) f (a))g 0 (x).
Proof :
Consider h : [a, b] R via h(x) = (g(b) g(a))f (x) (f (b) f (a))g(x). To prove the
result, it suffices to find x (a, b) such that h0 (x) = 0 (since h is cont, diff on [a, b] and
(a, b)).
Notice that h(a) = g(b)f (a) g(a)f (b) = h(b).
If h is constant, then h0 (x) = 0 trivially and were done. Assume h is not constant; then
t (a, b). h(t) > h(a) or h(t) < h(a).
If h(t) h(a), then Extreme Value Theorem guarantees that x (a, b). h(x) =
max h([a, b]) and Local Extremum Theorem = h0 (x) = 0.
If h(t) < h(a) then EVT guarantees x (a, b). h(x) = min h([a, b]) and LET =
h0 (x) = 0.
Corollary (Mean Value Theorem): If f : [a, b] R is cont and diff on [a, b], (a, b) then
x (a, b). f (b) f (a) = f 0 (x)(b a). Proof : Set g(x) = x.
Corollary (Monotonicity part 2): Suppose f : (a, b) R is differentiable on (a, b). The
following hold:
1. x (a, b). f 0 (x) > 0 = f is increasing.
2. x (a, b). f 0 (x) 0 = f is non-decreasing.
3. x (a, b). f 0 (x) = 0 = f is constant.
4. x (a, b). f 0 (x) 0 = f is non-increasing.
5. x (a, b). f 0 (x) < 0 = f is decreasing.
Proof : by MVT, if a < x1 < x2 < b, then f (x2 ) f (x1 ) = f 0 (x)(x2 x1 ).

6.3

Darbouxs Theorem

Definition: We say a function g : R R is periodic with period p > 0 if g(x + p) = g(x) (x R).
Theorem (Darboux): Suppose f : [a, b] R is differentiable on [a, b] and f 0 (a) < < f 0 (b).
Then x (a, b). f 0 (x) = .
Corollary: If f : [a, b] R is differentiable on [a, b], then f 0 has no simple discontinuities.
37

6.4

6.4

LHopitals Rule

21-355 Notes

LH
opitals Rule

Theorem: Suppose f, g : [a, b] R are continuous on [a, b], differentiable on (a, b), and g 0 (x) 6= 0
0 (x)
(x)
(x (a, b)). Assume that limxa fg0 (x)
= L. If f (a) = g(a) = 0, then limxa fg(x)
= L.
Proof :
We claim first that g(x) 6= 0 for x (a, b]. Otherwise, g(x) = 0 for some x (a, b] =
0 = g(x)g(a)
= g 0 (z) for some z (a, x), a contradiction. So fg : (a, b] R is wellxa
defined.
f (xn )
Let {xn }
n=l (a, b] satisfy xn a as n . We claim that limn g(xn ) = L. Once
this is established, the sequential characterization of limits yields the desired result.

To prove the claim, we apply Cauchys Mean Value Theorem on [a, xn ]: yn (a, xn )
such that f 0 (yn )g(xn ) = f 0 (yn )(g(xn ) g(a)) = g 0 (xn )(f (xn ) f (a)) = g 0 (xn )f (xn ).
0 (x )
(xn )
n
Then n l. fg(x
= fg0 (x
. Since a < yn < xn , the squeeze lemma implies yn a.
n)
n)
Hence limn

f (xn )
g(xn )

= limn

f 0 (xn )
g 0 (xn )

= limxa

f 0 (x)
g 0 (x)

= L.

Remarks:
1. The theorem is also true if we take limits at t.
2. If f, g : (a, b] R and limxa f (x) = limxa g(x) = 0, then the theorem still
works.

6.5

Higher Derivatives and Taylors Theorem

Definition: Suppose f : E R is differentiable at x E, and x is a limit point of {y

E | f 0 (x) exists}. We say f is twice differentiable at x if f 0 : {y E | f 0 (y) exists} R is
differentiable at x; and f 00 (x) = f (2) (x) = (f 0 )0 (x). Similarly, for n N with n > 2, we say f is ntimes differentiable at x if x is a limit point of {y E | f (n1) (y) exists} and f (n1) is differentiable
at x, in which case f (n) (x) = (f (n1) )0 (x).
If f (n) exists n N, n 1 we say f is infinitely differentiable at x.
Theorem (Taylor): Suppose f : [a, b] R. Assume f (n1) is continuous on [a, b] and f ( n) exists
on (a, b). Let x, y [a, b] with x 6= y. Then z (min{x, y}, max{x, y}) such that
f (y) =

n1
X
k=0

f (k) (x)
f (n) (z)
(y x)k +
(y x)n
k!
n!

(called the Taylor polynomial or Taylor approximation).

Proof :
Suppose x < y (y < x is handled without loss of generality). Let P (t) =
x)k , and set M =

f (y)P (y)
(yx)n .

It suffices to prove that M =

f (n) (z)
n!

Pn1
k=0

f (k) (x)
k! (t

for some z (x, y).

Define g(t) = f (t) P (t) M (t x)n , and notice that g (n) (t) = f (n) (t) n!M . As such,
it suffices to show that g (n) (z) = 0 for some z (x, y).
By construction, g (k) (x) = 0 (k = 0, . . . , n 1), and g(y) = 0 (by choice of M ).
By Mean Value Theorem, xi (x, y). g 0 (x1 ) = g(y)g(x)
= 0. Similarly, x2
yx
g 0 (x1 )g 0 (x)
= 0. Iterating,
x1x
(n1)
(xn1 )g (n1) (x)
g (n) (z) = g
xn1 x

(x, x1 ). g 00 (x2 ) =

we eventually find xn1 (x, y). g (n1) (xn1 ) =

0. Then 0 =

= 0 for some z (x, xn1 ).

38

21-355 Notes

7
7.1

Riemann-Stieltjes Integration
The R-S Integral

Definition: Let a, b R with a b. A partition of [a, b] is a finite ordered set P = {x0 , . . . , xn }

such that a = x0 x1 xn = b. Write [a, b] = {P | P is a partition of [a, b]}. For brevity
well write = [a, b].
Universal Assumptions: Throughout 7 we will always assume that:
1. f : [a, b] R is bounded: x [a, b]. m f (x) M , where m = inf f ([a, b]), M =
sup f ([a, b])
2. : [a, b] R (the integrator or weight function) is non-decreasing (in particular,
is also bounded)
Definition: For each P [a, b] we associate to f the following quantities (P = {x0 , . . . , xn }):
1. mi = inf{f (x) | x [xi1 , xi ]} for i [n]
2. Mi = sup{f (x) | x [xi1 , xi ]} for i [n]
3. i = (xi ) (xi1 ) 0 for i [n]
P
P
We write U (P, f, ) = ni=1 Mi i , L(P, f, ) = ni=1 mi i .
U is the upper Riemann-Stieltjes sum, and L is the lower R-S sum.
Pn
Pn
Pn
Remark
Pn : Clearly m((b) (a)) = i=1 m((xi ) (xi1 )) = i=1 mi i=1 Mi i
M i=1 i = M ((b) (a)). Hence P [a, b]. m((b) (a)) L(P, f, ) U (P, f, )
M ((b) (a)).
Definition of Integral: We define
Rb
Rb
a f d = sup{L(P, f, ) | P [a, b]}, and a f d = inf{U (P, f, ) | P [a, b]}.
Both are well-defined by the remark.
Rb
Rb
If a f d = a f d then we say f is R-S integrable with respect to , and write
Rb
Rb
Rb
a f d = a f d = a f d.
We write R([a, b]; ) = {f : [a, b] R | f is bounded, f is R-S integrable with respect to }.
Rb
When (x) = x, a f dx is the Riemann integral and we write R([a, b]).
Heuristics: The function assigns different weights to different points in [a, b]. The
Rb
intuition is that a f d is a weighted Riemann integral. If is continuous then we
Rb
have a geometric interpretation of a f d: consider the curve in R2 parameterized by
Rb
(x(t), y(t)) = ((t), f (t)); a f d = area under this curve.
Lemma: Let f (x) = C (x [a, b]). Then f R([a, b]; ) and

Rb

f d = C((b) (a)).
P
Proof : For any P [a, b] we have mi = Mi = C. Hence U (P, f, ) = L(P, f, ) = ni=1 Ci =
Rb
Rb
C((b) (a)). So a f d = sup{(L(P, f, )} = C((b) (a)) = inf{U (P, f, } = a f d.
a

Definition: If P, P 0 [a, b] and every point in P is in P 0 , we say P 0 is a refinement of P .

If P1 , P2 [a, b] we define the common refinement P1 #P2 [a, b] by P1 #P2 = P1 P2 , ordered
appropriately.
39

7.2

Integrability Criteria

21-355 Notes

Proposition: If P, P 0 [a, b] and P 0 is a refinement of P , then L(P, f, ) L(P 0 , f )

U (P 0 , f, ) U (P, f, ).
Theorem:

Rb

a f d

Rb

a f d.

Proof : Let P1 , P2 [a, b]. Then L(P1 , f, ) L(P1 #P2 , f, ) U (P1 #P2 , f, ) U (P 2, f, )
Rb
Rb
by last proposition. Hence a f d = sup{L(P1 , f, ) | P1 [a, b]} U (P2 , f, ). Then a f d
Rb
inf{U (P2 , f, ) | P2 [a, b]} = a f d.

7.2

Integrability Criteria

Theorem (Riemann)*: f R([a, b]; )  > 0. P [a, b]. U (P, f, ) L(P, f, ) < .
Proof :
(1) = (2):
Rb
Let  > 0. By definition, P1 , P2 [a, b]. a f d 2 < L(P1 , f, ), and U (P2 , f, )
Rb
Rb


f
d
+
.
Let
P
=
P
#P
.
Then
U
(P
,
f,
)

U
(P
,
f,
)
<
1
2
1
2
2
a
a f d + 2 . Hence
U (P, f, ) L(P, f, ) < .
(2) = (1):
Rb
Rb
For any partition P [a, b] we know a f d U (P1 , f, ) and L(P, f, ) a f d.
Rb
Rb
We also know a f d a f d. Then (2) implies that for  > 0 we have P
Rb
Rb
[a, b]. U (P1 , f, ) L(P1 , f, ) < . But then 0 a f d a f d <  ( > 0)
Rb
Rb
and so a f d = a f d = f R([a, b]; ).
Lemma: Let P [a, b]. The following are true:
1. If P 0 is a refinement of P , then U (P 0 , f, ) L(P 0 , f, ) U (P, f, ) L(P, f, ).
P
2. If si , ti [xi1 , xi ] (i [n]), then 0 ni=1 |f (si ) f (ti )|i U (P, f, )
L(P, f, ).
Rb
P
3. If f R([a, b]; ), then ti [xi1 , xi ] = | ni=1 f (ti )i a f d| U (P, f, )
L(P, f, ).
Theorem: If f is continuous on [a, b], then f R([a, b]; ).
Proof :
Note that EVT implies f is bounded. Let  > 0 and choose k > 0 so k((b) (a)) < .
Since f is cont on compact [a, b], f is uniformly continuous. Then > 0. x, y [a, b]
and |x y| < = |f (x) f (y)| < k.
Choose a partition P [a, b] such that xi xi1 < (i [n]). Then by EVT,
s
that mi = f (si ), Mi = f P
(ti ). Then U (P,P
f, ) L(P, f, ) =
Pin, ti [xi1 , xi ] suchP
n
n
n
i=1 (Mi mi )i =
i=1 (f (ti ) f (si ))i
i=1 ki = k
k=1 i =
k((b) (a)) < . Since  > 0 was arbitrary, we deduce Riemanns Theorem that
f R([a, b]; ).
Theorem: Suppose that f is monotone and is continuous on [a, b]; then f R([a, b]; ).
Remarks:
40

7.2

Integrability Criteria

21-355 Notes

1. If = x, then all monotone functions f are in R([a, b]), i.e. all monotone functions
are Riemann integrable.
2. Monotone functions dont have to be continuous, so f with simple discontinuities
in R([a, b]; ) when is continuous.
3. Monotone functions can have countably infinite sets of discontinuity; R-S integrals
can handle infinite discontinuities.

41