MATH 362 Lecture Notes Notation

1 Notation

s.t. = “such that”

w.r.t.= “with respect to”
D = “there exists” (or “there exist”)
D! = “there exists uniquely”
@ = “for all” (or “for every”)
∅ = the empty set
A Ă B (or B Ą A) = “A is a subset of B”
A ( B = “A is a proper subset of B (a subset of, but not equal to it)”
WLOG = “without loss of generality”
TFAE = “the following are equivalent”
iff = ô = “if and only if”
ÑÐ = “a contradiction of an earlier fact or supposition”
xx, yy = inner or dot product between objects (vectors, functions, etc.) x and y

Definition 1.1: Given sets Ai , i “ 1, . . . , k, the Cartesian product of A1 , . . . , Ak , denoted

k
ź
A1 ˆ A2 ˆ ¨ ¨ ¨ ˆ Ak , or Ai ,
i“1

is the set of all k-tuples pa1 , a2 , . . . , ak q, where each ai P Ai .

k
ź
Note: Rk “ R.
i“1

2
MATH 362 Lecture Notes Sequences, lim sup, lim inf

2 Sequences, lim sup, lim inf

Theorem 2.1: Suppose pan q is a monotonic sequence in R. Then the sequence converges
ô pan q is bounded.

Proof:

ñ: pan q converges ñ pan q is Cauchy. Result follows from exercise in PS1.

ð: WLOG we assume pan q is monotone increasing. Let A :“ tan | n P Nu. By the l.u.b.
property, α :“ sup A exists. Let ε ą 0. Choose N s.t. α ´ ε ă aN ď α. (If no such
N existed, then α ´ ε would be an upper bound of A. ÑÐ) Since an is monotone
increasing, α ´ ε ă an ď α, @ n ě N.

Definition 2.2: The extended reals r´8, 8s, consists of the set R together with the symbols
p´8q and p8q. We order this set by defining ´8 ă x ă 8 @ x P R.

Notes:

1. r´8, 8s is ordered set, but not a field.

2. By convention,

• if x is real, then
x x
x ` 8 “ `8, x ´ 8 “ ´8, “ “ 0.
`8 ´8
• if x ą 0 then x ¨ p`8q “ `8, x ¨ p´8q “ ´8.
• if x ă 0 then x ¨ p`8q “ ´8, x ¨ p´8q “ `8.

Proposition 2.3: If A Ă r´8, 8s and A , ∅, then sup A, inf A both exist in r´8, 8s.

Proof: Exercise 

3
MATH 362 Lecture Notes Sequences, lim sup, lim inf

Definition 2.4: Let pan q be a sequence of reals with the property that @ M P R, DN P
N s.t. n ě N ñ an ě M. We then write an Ñ `8.

Similarly, if @ M P R, DN P N s.t. n ě N ñ an ď M, we write an Ñ ´8.

Proposition 2.5: If pan q is a monotone (real) sequence, then an Ñ x for some x P r´8, 8s.

Proof: (Outline) WLOG, a1 ď a2 ď ¨ ¨ ¨ . Show that if pan q is not bounded above, then
an Ñ `8. 

Definition 2.6: Let pan q be a sequence of real numbers. For n P N define

Mn :“ suptan , an`1 , an`2 , . . .u and mn :“ inftan , an`1 , an`2 , . . .u.

Note:

1. mn , Mn P r´8, 8s, @ n.

2. pMn q and pmn q are monotone decreasing and increasing sequences respectively.

Thus, Mn Ñ M for some M P r´8, 8s, and likewise mn Ñ some m. The numbers M, m are
called upper and lower limits of pan q; we use the notation

lim sup an “ M and lim inf an “ m,

nÑ8 nÑ8

or, alternatively,
lim sup an “ a˚ and lim inf an “ a˚ .
n n

Notes:

1. As we make precise in Thm. 2.9 (below) one can think of lim sup an as the largest number in
the extended reals that pan q “visits” infinitely often.

2. lim sup an and lim inf an exist for every (real) sequence pan q (Propositions 2.3 and 2.5).

4
MATH 362 Lecture Notes Sequences, lim sup, lim inf

Proposition 2.7: For any sequence pan q in r´8, 8s, lim inf an ď lim sup an .

Proof: For each n we have mn ď Mn in the definition. 

Theorem 2.8: Let pan q be a sequence of reals (i.e., in R), and suppose L P r´8, 8s. Then
an Ñ L ô lim inf an “ lim sup an “ L.

Proof: [for the case that L P R; the case L “ ˘8 is an exercise]

ñ: Let ε ą 0. Then D N P N s.t. n ě N ñ L ´ ε ă an ă L ` ε. Thus,

MN :“ suptaN , aN`1 , . . .u ď L ` ε. But lim sup an ď MN ď L ` ε. Since ε is
arbitrary, lim sup an ď L. It may similarly be shown that lim inf an ě L. Therefore,

L ď lim inf an ď lim sup an ď L,

which implies the result.

ð: Let ε ą 0. Since lim sup an “ L, DN1 P N s.t. n ě N1 ñ suptan , an`1 , . . .u ă L `
ε{2. Thus, an ă L`ε{2, @ n ě N1 . Similarly, D N2 P N s.t. n ě N2 ñ L´ε{2 ă an .
Take N “ maxtN1 , N2 u. Then
ε ε
L´ ă an ă L ` , @ n ě N,
2 2
which means limn an “ L.

Theorem 2.9: Let pan q be a real sequence and assume M “ lim sup an P R. Then @ ε ą 0,

(i) DN P N s.t. an ă M ` ε, @ n ě N.

(ii) an ą M ´ ε, for infinitely many n.

Thus, |an ´ M| ă ε, for infinitely many n. Similar (modified) statements may be made about
lim inf an .

5
MATH 362 Lecture Notes Sequences, lim sup, lim inf

Proof:

(i) If this were not so then there would be some ε ą 0 for which Mn ě M ` ε, @ n,
and this would imply M ě M ` ε. ÑÐ
(ii) If this were not so then there would be some ε ą 0 and some N P N for which
n ě N ñ Mn ď M ´ ε. This would imply M ď M ´ ε. ÑÐ

6
MATH 362 Lecture Notes Some Useful Inequalities

3 Some Useful Inequalities

Theorem 3.1 (Schwarz’s Inequality): Let u, v P Rk . Then

| xu, vy | ď |u| |v|, (1)

with equality holding if and only if one of u, v is a scalar multiple of the other.

Proof: The inequality (1) holds trivially if one of u, v is the zero vector. Thus, we turn
to the case where u, v are arbitrary (fixed) nonzero vectors.
For any real t,

0 ď |tu ` v|2 “ xtu ` v, tu ` vy “ |u|2 t2 ` 2 xu, vy t ` |v|2 .

As this is a quadratic polynomial in t which cannot have two real roots, we know the
discriminant

4 xu, vy2 ´ 4|u|2 |v|2 ď 0 ñ xu, vy2 ď |u|2 |v|2 ,

which implies (1).

To complete the proof, note if, say, u “ tv, then

| xu, vy | “ | xtv, vy | “ |t|| xv, vy | “ |t||v|2 “ |tv||v| “ |u||v|.

On the other hand, if | xu, vy | “ |u||v|, then the discriminant we calculated above is 0,
which means the quadratic polynomial |tu ` v|2 has precisely one root, t0 , so

|t0 u ` v|2 “ 0, or v “ ´t0 u.

Proposition 3.2: If a ą 0, b ą 0, and 0 ă λ ă 1, then

aλ b1´λ ď λa ` p1 ´ λqb. (2)

Proof: Since the function ln x, x ą 0 is concave, it follows that

lnpaλ b1´λ q “ λ ln a ` p1 ´ λq ln b ď lnpλa ` p1 ´ λqbq.

The result now follows, as c ď d implies ec ď ed . 

7
MATH 362 Lecture Notes Some Useful Inequalities

Proposition 3.2 allows us to prove a generalization of Schwarz’s Inequality. Note that, in the
statement of the next result, we may take p “ q “ 2, in which case we get the Schwarz Inequality
again.

Theorem 3.3 (Hölder’s Inequality): Let x, y P Rk have nonnegative components—i.e., each

xi , yi ě 0—and suppose p, q ą 1 satisfy 1{p ` 1{q “ 1. Then
˜ ¸1{p ˜ ¸1{q
n n n
p q
ÿ ÿ ÿ
xi yi ď xi yi . (3)
i“1 i“1 i“1

Proof: Note that if each xi “ 0 or each yi “ 0, then the result is trivial. Thus, we
assume xi ą 0 and y j ą 0 for at least one pi, jq P t1, . . . , ku2 . Set
˜ ¸1{p ˜ ¸1{q
n n
p q
ÿ ÿ
α“ xi , β“ yi ,
i“1 i“1

and note that α, β ą 0. Moreover, the vector x1 :“ α1 x with components x1i satisfies
n n ´ ¯p n
ÿ ÿ xi 1 ÿ p
px1i qp “ “ x “ 1.
i“1 i“1
α αp i“1 i

Similarly, y1 :“ β1 y with components y1i satisfies

n n ˆ ˙q
ÿ ÿ xi
py1i qq “ “ 1.
i“1 i“1
β

Thus, it suffices to show that, under the conditions

n n n
p q
ÿ ÿ ÿ
xi “ 1 “ yi , we have xi yi ď 1.
i“1 i“1 i“1

Taking λ “ 1{p and 1 ´ λ “ 1{q in Theorem 3.2, we have

p q 1 p 1 q
xi yi “ pxi q1{p pyi q1{q ď x ` yi ,
p i q

for each i “ 1, . . . , n. Summing over all such i, we obtain

n n n
ÿ 1 ÿ p 1 ÿ q 1 1
xi yi ď xi ` yi “ ¨ 1 ` ¨ 1 “ 1.
i“1
p i“1 q i“1 p q

8
MATH 362 Lecture Notes Some Useful Inequalities

Theorem 3.4 (Minkowski’s Inequality): Let x, y P Rk have nonnegative components—i.e.,

each xi , yi ě 0—and suppose p ě 1. Then
« ff1{p ˜ ¸1{p ˜ ¸1{p
n n n
p p
ÿ ÿ ÿ
pxi ` yi qp ď xi ` yi . (4)
i“1 i“1 i“1

Proof: In the case p “ 1 it is clear that the two sides of (4) are equal. Note, also, that
the result holds trivially in the case i pxi ` yi qp “ 0.
ř

Now, suppose p ą 1 and i pxi ` yi qp , 0. For the application of Hölder’s Inequality

ř

(Theorem 3.3) below, we assume 1{p ` 1{q “ 1 (so p “ pq ´ q). Then, we have
n
ÿ n
ÿ n
ÿ n
ÿ
pxi ` yi qp “ pxi ` yi qpxi ` yi qp´1 “ xi pxi ` yi qp´1 ` yi pxi ` yi qp´1
i“1 i“1 i“1 i“1
˜ ¸1{p « ff1{q ˜ ¸1{p « ff1{q
n
ÿ n
ÿ n
ÿ n
ÿ
p p
ď xi pxi ` yi qpp´1qq ` yi pxi ` yi qpp´1qq (by Thm. 3.3)
i“1 i“1 i“1 i“1
»˜ ¸1{p ˜ ¸1{p fi « ff1{q
n
ÿ n
ÿ n
ÿ
p p
“ – xi ` yi fl pxi ` yi q p
.
i“1 i“1 i“1

“řn ‰1{q
Now divide both ends of the expression by i“1 pxi ` yi qp to obtain (4). 

x
Proposition 3.5: The function f pxq “ , x ě 0, is monotonically increasing.
1`x

Proof: Suppose y ą x ě 0. Then

y 1 1 x
f pyq “ “ 1´ ą 1´ “ “ f pxq.
1`y 1`y 1`x 1`x


Proposition 3.6: For any two real numbers x, y, it is the case that

|x ` y| |x| |y|
ď ` .
1 ` |x ` y| 1 ` |x| 1 ` |y|

9
MATH 362 Lecture Notes Some Useful Inequalities

Proof: First, note that if either x “ 0 or y “ 0 then the result holds trivially. Next, let
assume x, y are both positive real numbers. Then

|x ` y| x`y x y
“ “ `
1 ` |x ` y| 1`x`y 1`x`y 1`x`y

x y |x| |y|
ď ` “ ` .
1`x 1`y 1 ` |x| 1 ` |y|

The case where both x, y are negative numbers follows immediately.

Now assume that xy ă 0 (i.e, they have opposite sign). WLOG, we may assume
|x| ą |y|, so that |x ` y| ď |x|. Then by Prop. 3.5,

|x ` y| |x| |x| |y|

ď ď ` .
1 ` |x ` y| 1 ` |x| 1 ` |x| 1 ` |y|


10
MATH 362 Lecture Notes Metric Spaces

4 Metric Spaces

Accompanying reading may be found in

• Elementary Analysis, the Theory of Calculus, 2nd Ed., by Kenneth A. Ross, Section 13

• Metric Spaces, by Satish Shirali and Harkrishan L. Vasudeva

Definition 4.1: A metric space is a set X along with an accompanying function d : X ˆ X Ñ

R (called a metric on X) s.t.

(i) dpx, xq “ 0, @ x P X

(ii) dpx, yq ą 0, @ x, y P X with x , y.

(iii) dpx, yq “ dpy, xq, @ x, y P X.

(iv) dpx, yq ď dpx, zq ` dpz, yq, @ x, y, z P X.

Examples: (Do carefully)

• R

• Rk
Here are several different choices of metrics:
« ff1{2
ÿk
d2 px, yq :“ pxi ´ yi q2 ,
i“1

k
ÿ
d1 px, yq :“ |xi ´ yi |,
i“1

d8 px, yq :“ max |xi ´ yi |,

1ďiďk

˜ ¸1{p
k
ÿ
dp px, yq :“ |xi ´ yi |p , p ě 1.
i“1

We claim that pRk , d8 q and pRk , dp q, p ě 1 are examples of metric spaces. To show this
requires showing that d8 , dp are metrics on Rk , having the properties described in Definition
4.1. A proof that d2 satisfies the triangle inequality is on pp. 28-29 in Shirali. It requires the
Cauchy-Schwarz inequality. For dp we need the Minkowski inequality.

11
MATH 362 Lecture Notes Metric Spaces

When we talk about Euclidean space Rk , the implied metric is d2 . In instances where we
wish to use some other metric on the vectors in Rk , we will be explicit about which metric
we employ.

• For any set X let d0,1 be the discrete metric defined by

#
0, if x “ y,
d0,1 px, yq :“
1, if x , y.

Then pX, d0,1 q is a metric space. In particular, this discrete metric is yet another suitable way
to define distance between points in Rk so as to make it into a metric space.

• Given a metric space pX, dq and a subset Y Ă X, pY, dq is a metric space.

• Given a metric space pX, dq, define d1 : X ˆ X Ñ r0, 8q via

dpx, yq
d1 px, yq :“ .
1 ` dpx, yq

Then pX, d1 q a metric space. (See Proposition 1.2.3 in Shirali.)

• Let S be a set, and consider the set BpSq of bounded functions f : S Ñ C—that is,

sup | f pxq| ă 8.
xPS

If, for functions f , g P BpSq, we define

dp f, gq :“ sup | f pxq ´ gpxq|,

xPS

then pBpSq, dq is a metric space.

• As a special case of the previous example, we might take S “ ra, bs Ă R. The collection
Cpra, bsq of continuous functions on S is a subset of BpSq (by the Extreme Value Theorem—a
continuous function defined on a closed interval attains minimum and maximum values),
so pCpSq, dq is a metric space, known as the space of continuous functions.

• In Ross, Definition 17.1, he defines f : S Ă R Ñ R to be continuous on S if, given any s0 P S,

for all sequences psn q in S converging to s0 , limn f psn q “ f ps0 q. If we take S “ t1, 2, . . . , ku,
then every vector x P Rk is a continuous mapping i ÞÑ xi from S to R. Thus, as another special
case, we have that Rk is a space of continuous functions on S “ t1, 2, . . . , ku under the metric

dpx, yq “ sup |xi ´ yi | “ max |xi ´ yi |.

1ďiďk i

This special case was encountered already, when we discussed the metric space pRk , d8 q
above.

12
MATH 362 Lecture Notes Metric Spaces

• Above we saw a family of metrics dp , p ě 1 on Rk . Now that we realize Rk is a set of

continuous functions, we might adapt the definition of dp to produces metrics on other sets
of continuous functions. That is, for functions f , g P Cpra, bsq, p ě 1, define
˜ż ¸1{p
b
dp p f, gq :“ | f pxq ´ gpxq|p dx .
a

It can be shown that dp is a metric on Cpra, bsq, so that pCpra, bsq, dp q is a metric space.

Definition 4.2: For the metric space pX, dq:

(i) Given a P X, r ą 0, the open ball of radius r centered at a is

Bpa, rq :“ tx P X | dpa, xq ă ru.

(ii) A subset E of X is bounded if E is contained in Bpa, Rq for some a P X and some R ą 0.

(iii) A sequence pan q in X converges to a (written an Ñ a) in X if @ ε ą 0, DN P N s.t. n ě

N ñ dpan , aq ă ε.

(iv) A sequence pan q in X is Cauchy if @ ε ą 0, DN s.t. m, n ě N ñ dpam , an q ă ε.

(v) X is complete if every Cauchy sequence in X converges to a point in X.

Theorem 4.3: Euclidean space Rk (k ě 1) is complete.

Proof: That this is so for k “ 1 The case k “ 1 is contained in Theorem 10.11 from Ross,
and is proved there. The case k ą 1 is left as an exercise. 

Definition 4.4: Let pX, dq be a metric space with subset E.

(i) A point a P X is limit point of E if D a sequence pan q of distinct points in E s.t. an Ñ a.

Notes:

(a) a need not be in E.

(b) The set of all limit points of E is denoted E1 .
(c) What might non-limit points look like?

13
MATH 362 Lecture Notes Metric Spaces

(ii) A point a P X is adherent point of E if D a sequence pan q in E s.t. an Ñ a. Note: Every

x P E is adherent point of E.

(iii) An adherent point which is not a limit point is called an isolated point.

(iv) E is closed in X if E contains all of its limit points.

(v) E is perfect if E is closed and every point of E is a limit point.

Definition 4.5: Suppose that A is a set with subset E. The set AzE (also denoted Ec , the
complement of E in A, when the set A is understood), is given by

AzE :“ ta P A | a < Eu.

Theorem 4.6: Suppose that pX, dq is a metric space, E Ă X. TFAE:

(i) a is a limit point of E.

(ii) Bpa, rq X E is infinite, @ r ą 0.

(iii) pBpa, rqztauq X E , ∅, @ r ą 0.

Proof: Exercise. 

Definition 4.7: Let pX, dq be a metric space. Ω Ă X is said to be open (in X) if @ a P Ω, Dr ą

0 s.t. Bpa, rq Ă Ω.

Examples:

1. r0, 1s and r1, 8q are closed in R (R with the usual metric).

2. N and Z are closed in R. (Neither set has any limit points.)

3. The empty set ∅ and X are always closed in pX, dq.

14
MATH 362 Lecture Notes Metric Spaces

4. p0, 1q is closed in p0, 1q, but not in p0, 1s.

5. A finite set is always closed.

6. Every point of pE, d0,1 q is isolated.

Theorem 4.8: Suppose that pX, dq is a metric space, Ω Ă X. Ω is open ô XzΩ is closed.

Proof: Denote XzΩ as E.

ñ: Assume Ω is open. By definition, for x P Ω, Dr ą 0 s.t. Bpx, rq Ă Ω — that is,

Bpx, rq X E “ ∅. By Theorem 4.6, x cannot be a limit point of E. Thus, the only
points which can be limit points of E reside in E :“ XzΩ.
ð: Assume E is closed. For x P Ω, Theorem 4.6 indicates that Dr ą 0 s.t. Bpx, rq X E “
∅. That is, Bpx, rq Ă Ω.

Proposition 4.9 (DeMorgan’s Laws): Suppose tAα u is a (finite or infinite) collection of sub-
sets of a set X. Then
˜ ¸c
ď č
(i) Aα “ Acα
α α
˜ ¸c
č ď
(ii) Aα “ Acα
α α

Proof:
˜ ¸c ˜ ¸c
ď ď č
(i) For x P Aα , x P Acα for each α, showing that Aα Ă Acα .
α α α¸
˜ c
č ď ď č
For x P Acα , x < Aα @ α ñ x < Aα , which shows that Aα Ą Acα .
α α α α
Taken together, these opposing set inclusions imply the result.
(ii) May be proved similarly; also follows from (i). (Verify)

15
MATH 362 Lecture Notes Metric Spaces

Definition 4.10: Let E be a subset of a metric space X.

(i) The closure of E, denoted E, is the set E Y E1 .

(ii) If E “ X, then E is said to be dense in X.

Theorem 4.11: Let pX, dq be a metric space with subset E.

(i) E is closed.

(ii) E “ E ô E is closed.

(iii) If E is a subset of closed F, then E Ă F.

Proof:

(i) Let x be a limit point of E. Then D a sequence pxn q in E s.t. xn Ñ x. @ n P N, Dyn P

E s.t. dpyn , xn q ă 1{n. The triangle inequality implies that yn Ñ x, which means
that x P E1 Ă E.
(ii) ñ: This follows immediately from (i).
ð: If E is closed, then E1 Ă E ñ E “ E Y E1 “ E.
(iii) The key observation: E Ă F ñ E1 Ă F1 .
F is closed ñ F1 Ă F ñ E “ E Y E1 Ă F Y F1 “ F “ F.

Note: (i) and (iii) imply E is smallest closed set containing E.

Examples (of closures): Let pX, dq “ pR, |x ´ y|q. Then

1. p0, 1q “ r0, 1s.

2. p0, 1q Y t2u “ r0, 1s Y t2u.

3. N “ N.

4. Q “ R. (So, Q is dense in R).

16
MATH 362 Lecture Notes Metric Spaces

Proof: Follows from Thm. 4.6, and the fact (likely proved in Math 361) that
between every two real numbers there is a rational. 

5. E “ tx P X | D a sequence pxn q in E s.t. xn Ñ xu “ tx P X | xis adherent pt. of Eu.

Theorem 4.12:
ď
(i) If tΩα u is a collection of open sets, then Ωα is open.
α

n
č
(ii) If Ω1 , Ω2 , ..., Ωn are open, then Ωm is open.
m“1
č
(iii) If tFα u is a collection of closed sets, then Fα is closed.
α

n
ď
(iv) If F1 , F2 , ..., Fn are closed, then Fm is closed.
m“1

Proof:
ď
(i) Let x P Ωα . Then x P Ωα0 for some (fixed) α0 and, since Ωα0 is open, Dr ą
α ď
0 s.t. Bpx, rq Ă Ωα0 Ă Ωα .
α
n
č
(ii) Let x P Ωm . Then x P Ωm for each m “ 1, . . . , n, and Drm ą 0 s.t. Bpx, rm q Ă Ωm .
m“1
Let r :“ mintrm | 1 ď m ď nu. By construction, r ą 0 and Bpx, rq Ă Ωm for each m,
n
č
and hence, Bpx, rq Ă Ωm .
m“1
(iii) This follows from (i), Theorem 4.8 and Proposition 4.9 (ii). (Verify)
(iv) This follows from (ii), Theorem 4.8 and Proposition 4.9 (i). (Verify)

Examples:

1. Finite intersections of sets p´1{n, 1{nq are open, but the intersection of all such sets n “ 1, 2, . . .
is t0u (a closed set).

2. Each finite union of singletons (single element subsets) of Q is closed, but Q is open (in Q).

17
MATH 362 Lecture Notes Metric Spaces

Theorem 4.13: Let E Ă R be nonempty and bounded above. If E is closed then sup E P E.

Proof: Exercise. 

18
MATH 362 Lecture Notes Compactness

5 Compactness

Definition 5.1: Let pX, dq be a metric space.

(i) An open cover of a set E in X is a collection tGα u of open subsets of X such that E is
ď
contained in Gα .
α

(ii) A subset K of X is said to be compact if every open cover of K contains a finite subcover
ď n
ď
— that is, whenever K Ă Gα , Dα1 , . . . , αn s.t. K Ă Gα j .
α j“1

Examples:

1. Any finite set is compact.

2. N is not compact in R
8
ď
Proof: N Ă pn ´ 1{2, n ` 1{2q, which has no finite subcover. 
n“1

3. p0, 1q is not compact in R.

8
ď
Proof: p0, 1q “ p1{n, 1q, which has no finite subcover. 
n“1

4. pE, d0,1 q is not compact if E is an infinite set.

Theorem 5.2: Compact subsets of metric spaces are closed.

Proof: Let K be a compact subset of a metric space X. We will prove that Kc is open. To
that end, let x P Kc . By the compactness of K, there exists an open cover of K consisting
of finitely many open balls Bpy1 , r1 q, . . . , Bpyn , rn q with each r j ă dpx, y j q{2. Now choose
r ă mintr j | 1 ď j ď nu. Clearly Bpx, rq has empty intersection with K—that is, it is
contained in Kc . 

or, alternatively
Proof: Let K be a compact subset of a metric space X. Let a P K1 , and suppose
a < K. Define sets Ωn :“ tx P X | dpa, xq ą 1{nu “ Bpa, 1{nq c . Each Ωn is open, and

19
MATH 362 Lecture Notes Compactness

8
ď
K Ă Ωn “ Xztau, making tΩn u an open cover of K. Since K is compact, there is
1
n´1
ď
a finite subcover. But, @ n P N, Ωm Ă Ωn , so D some ΩN which covers K. Thus
m“1
K X Bpa, 1{Nq “ ∅. ÑÐ 

Definition 5.3: Let A, B be subsets of a metric space.

(i) We take the diameter of A, denoted diampAq, to be

diampAq :“ suptdpx, yq | x, y P Au.

(ii) We take the distance between A and B, denoted distpA, Bq, to be

distpA, Bq :“ inftdpa, bq | a P A, b P Bu.

Corollary 5.4: Suppose K is a compact subset of a metric space. Then distptau, Kq “ 0 iff
a P K.

Theorem 5.5: Let pX, dq be a metric space, Y Ă X.

(i) Ω Ă Y is open in Y ô Ω “ Ω̃ X Y for some Ω̃ open in X.

(ii) F Ă Y is closed in Y ô F “ F̃ X Y for some F̃ closed in X.

Proof:

(i) The key idea here is that for a P Y, BY pa, rq “ pBX pa, rq X Yq.
ñ: So, let Ω Ă Y be open in Y. For each a P Ω, Dra ą 0 s.t. BY pa, ra q Ă Ω. So,
˜ ¸
ď ď ď
Ω “ BY pa, ra q “ pBX pa, ra q X Yq “ BX pa, ra q X Y.
aPΩ aPΩ aPΩ
ď
Set Ω̃ “ BX pa, ra q, which is open since it is a union of open balls in X.
aPΩ

20
MATH 362 Lecture Notes Compactness

ð: Now suppose Ω̃ is open in X, and that Ω “ Ω̃ Y Y. a P Ω ñ a P Ω̃. Thus,

Dr ą 0 s.t. BX pa, rq Ă Ω̃.
ñ BY pa, rq “ pBX pa, rq X Yq Ă Ω̃ X Y “ Ω.
` ˘

Thus, Ω is open in Y.
(ii) The key idea here is that, for F Ă Y, F1Y “ F1X X Y.
ñ:
` ˘
F closed inY ñ F “ F Y F1Y “ F Y F1X X Y
` ˘
“ F Y F1X X Y “ FX X Y.

So, take F̃ “ FX .
ð: Assume F̃ is closed in X and F “ F̃XY. Let a P F1Y . Then a P F̃1X Ă F̃. Moreover,
a P Y. Thus, a P F̃ X Y “ F, showing that F is closed in Y.

Theorem 5.6: Let X be a metric space with K Ă Y Ă X. K is compact in Y ô K is compact

in X.

Proof:

ñ: Let tΩα u be an open cover of K in X. K Ă Y ñ tΩα X Yu is an open cover of K in

Y. K is compact in Y ñ D a finite subcover of tΩα X Yu — that is,

K Ă pΩα1 X Yq Y ¨ ¨ ¨ Y pΩαn X Yq .

Thus,
K Ă Ωα1 Y ¨ ¨ ¨ Y Ωαn ,

showing that K is compact in X.

ð: Let tΩα u be an open cover of K in Y. Each Ωα “ Ω̃α X Y for some Ω̃ open in X. So,
tΩ̃α u is an open cover of K in X. As before, there is a subcover
K Ă Ω̃α1 Y ¨ ¨ ¨ Y Ω̃αn
ñ K X Y Ă Ω̃α1 X Y Y ¨ ¨ ¨ Y Ω̃αn X Y
` ˘ ` ˘

ñ K Ă Ωα1 Y ¨ ¨ ¨ Y Ωαn .
Thus, K is compact in Y.

Remarks:

21
MATH 362 Lecture Notes Compactness

• The word “compact” could not be replaced by the words “open” or “closed” for the previous
theorem to be true.
Example: If X “ R and Y “ p0, 1s, with d “ |x ´ y|, then p0, 1s is closed in Y, but neither
closed nor open in X.

• An immediate corollary to the theorem is:

K is compact relative to itself iff compact in X.

Proposition 5.7: If a ă b, then the interval ra, bs is compact.

Proof: Exercise.
Hint: If tΩα u is open cover of ra, bs, let

E “ tx P ra, bs | ra, xs is covered by finitely many Ωα u.

Show sup E P E and sup E “ b. 

Theorem 5.8: Closed subsets of compact sets are compact.

Proof: Suppose F Ă X, with X compact and F closed. Let tΩα u be an open cover of F.
Then tΩα u Y pXzFq is an open cover of X. X compact ñ X Ă Ωα1 Y ¨ ¨ ¨ Y Ωαn Y pXzFq
ñ F Ă Ωα1 Y ¨ ¨ ¨ Y Ωαn .
Therefore, F is compact. 

Corollary 5.9: Suppose F and K are subsets of X. If F is closed and K is compact, then F X K
is compact.

Proof: Exercise. 

n
č
Definition 5.10: A collection C of sets has the finite intersection property if C j , ∅ for
j“1
every finite subcollection tC1 , C2 , . . . , Cn u Ă C.

22
MATH 362 Lecture Notes Compactness

Theorem 5.11: Let tKα u be a collection of compact subsets of a metric space X. If tKα u has
č
the finite intersection property, then Kα , ∅.
α

č
Proof: Suppose Kα “ ∅. Then by DeMorgan’s laws,
α
˜ ¸c
č ď
X “ Kα “ Kαc .
α α

Since each Kαc is open (Thm. 5.2), tKαc u is an open cover of X. Choose any Kα0 . Since
compact, there are α1 , . . . , αn for which
Kα0 Ă Kαc 1 Y ¨ ¨ ¨ Y Kαc n
ñ Kαc 0 Ą Kα1 X ¨ ¨ ¨ X Kαn
ñ Kα0 X Kα1 X ¨ ¨ ¨ X Kαn “ ∅. ÑÐ 

Corollary 5.12: Suppose tKn u is a sequence of nonempty compact subsets of a metric space
8
č
X s.t. Kn Ą Kn`1 for n “ 1, 2, . . .. Then Kn , ∅.
n“1

Theorem 5.13: Suppose tIn u is a sequence of closed intervals in R, nested so that I1 Ą I2 Ą

8
č
I3 Ą ¨ ¨ ¨ . Then In , ∅.
n“1

Proof: This follows immediately from 5.12 and the fact that closed intervals are
compact (5.7). 

Theorem 5.14 (Nested Interval Theorem): Suppose that In “ ran , bn s, that In`1 Ă In , @ n P
N, and that the quantity `pIn q :“ bn ´ an Ñ 0 as n Ñ 8. Then D! c P R s.t.
Ş
n In “ tcu.

Proof: Exercise. 

23
MATH 362 Lecture Notes Compactness

Definition 5.15: Suppose ai ă bi for each i “ 1, . . . , k. The set

k
ź
rai , bi s “ tpx1 , . . . , xk q P Rk | ai ď xi ď bi , 1 ď i ď ku
i“1

is called a k-cell. A k-cell is the k-dimensional generalization of a rectangle.

Theorem 5.16: Let k ą 0 be an integer. If tJn u is a sequence of k-cells, nested so that

8
č
J1 Ą J2 Ą ¨ ¨ ¨ , then Jn , ∅.
n“1

k
ź
Proof: For each n “ 1, 2, . . ., write Jn “ ran,i , bn,i s. For each i “ 1, . . . , k the intervals
i“1
8
č
ran,i , bn,i s satisfy the hypotheses of Theorem 5.13, and hence there exists x˚i P ran,i , bn,i s.
n“1
8
č
Thus, x˚ :“ px˚1 , x˚2 , . . . , x˚k q P Jn . 
n“1

Theorem 5.17: Every k-cell is compact.

Proof: We will prove the case k “ 2, but the ideas are easily adapted to a different
dimension k. To this end, assume J “ ra1 , b1 s ˆ ra2 , b2 s. Set
b
δ “ diampJq “ pb1 ´ a1 q2 ` pb2 ´ a2 q2 .

Suppose that J is not compact. Then D an open cover tGα u of J which contains no finite
subcover of J. Let c1 “ pa1 `b1 q{2, c2 “ pa2 `b2 q{2. Then J is the union of four rectangles

ra1 , c1 s ˆ ra2 , c2 s, ra1 , c1 s ˆ rc2 , b2 s, rc1 , b1 s ˆ ra2 , c2 s, and rc1 , b1 s ˆ rc2 , b2 s,

all the same size, all of diameter 2´1 δ. By the pigeonhole principle, at least one of these
four rectangles cannot be covered by a finite subcollection of tGα u; call this rectangle
J1 . We may once again subdivide J1 into four rectangles, at least one of which cannot
be covered by a finite subcollection of tGα u; this rectangle we call J2 . Proceeding in like
fashion, we have rectangles J, J1 , J2 , . . . such that

24
MATH 362 Lecture Notes Compactness

(i) J Ą J1 Ą J2 Ą ¨ ¨ ¨ ,
(ii) each Ji cannot be covered by any finite subcollection of tGα u, and
(iii) diampJi q “ 2´i δ.
8
č
(i) and Theorem 5.16 imply that some x˚ exists in Jn . This x˚ P J, so x˚ P Gα0 for some
n“1
α0 . Since Gα0 is open, there is some r ą 0 for which Bpx˚ , rq Ă Gα0 . Let N P N be large
enough so that n ě N implies 2´n δ ă r{4. Then for n ě N, Jn Ă Bpx˚ , rq Ă Gα0 . ÑÐ


Theorem 5.18: Let E Ă Rk . TFAE:

(i) E is closed and bounded.

(ii) E is compact.

(iii) Every infinite subset of E has a limit point in E.

Proof:

(i) ñ (ii): Since E is bounded, it is contained in a k-cell of sufficiently large size. Since closed,
E is compact, by Theorems 5.8 and 5.17.
(ii) ñ (iii): Suppose A Ă E is infinite but has no limit point. By definition, A is closed,
and hence compact (by Theorem 5.8). For each a P A, D rpaq ą 0 such that
pBpa, rpaqqztauq X A “ ∅. The collection tBpa, rpaqq | a P Au is an open cover of A
which has no finite subcover. ÑÐ
(iii) ñ (i): We assume (iii), and suppose that E is not bounded. Thus, E must be an infinite
set, and we may choose, for each n P N, xn P E s.t. |xn | ą n. The sequence pxn q
has no limit point in Rk , let alone in E. ÑÐ
Now, let x0 P E1 . By definition, D a sequence pxn q of distinct points in E for which
xn Ñ x0 . Since S :“ txn | n P Nu is an infinite set, it has a limit point x˚ P E.
We claim that x˚ “ x0 . To prove this, suppose that x˚ , x0 . Then D radii r0 ,
r˚ ą 0 s.t. Bpx˚ , r˚ qXBpx0 , r0 q “ ∅. We select N P N s.t. , for n ě N, xn P Bpx0 , r0 q.
Then Bpx˚ , r˚ q X S Ă Bpx˚ , r˚ q X pSzBpx0 , r0 qq Ă SzBpx0 , r0 q Ă tx1 , . . . , xN´1 u. This
shows that Bpx˚ , r˚ q X S is a finite set. ÑÐ
By the claim, x0 P E. Since x0 is an arbitrarily-chosen limit point of E, E is closed.

25
MATH 362 Lecture Notes Compactness

Notes:

• The equivalence (i) ô (ii) in the previous theorem is usually called the Heine-Borel Theorem.

• “Closed and bounded” was the first defn of compactness (W. Ramey). Let A be an infinite
set and consider metric space pA, d0,1 q. Note: pA, d0,1 q is closed (there are no limit points)
and bounded. Yet, tBpa, 1{2q | a P Au is an open cover of A which contains no finite subcover.
Thus, “closed and bounded” is not an equivalent characterization of compact metric spaces
in general, only compact Euclidean metric spaces.
Another example: The metric space pp0, 1q, |x ´ y|q is closed in p0, 1q and bounded, but not
compact.

• (iii) suggests another possible characterization of compact spaces. This characterization does
hold for general metric spaces, as stated in the following theorem:

Theorem 5.19: A subset E of a metric space pX, dq is compact ô every infinite subset
of E has a limit point contained in E.

There is yet another characterization of compact metric spaces.

Definition 5.20: A subset E of a metric space pX, dq is said to be totally bounded if

n
ď
@ ε ą 0, D finitely-many sets E1 , E2 , . . . , En Ă X s.t. E Ă Em and diampEm q ă ε
m“1
for m “ 1, . . . , n.

Employing this definition, there is the result:

Theorem 5.21: A subset E of a metric space pX, dq is compact ô E is complete and

totally bounded.

We will prove neither of these characterizations. For a reference, see Section 5.1 of Metric
Spaces, by Satish Shirali and Harkrishan L. Vasudeva. Neither of these characterizations
extends to general topological spaces, and so neither serves as an appropriate alternate
definition for the term compact.

26
MATH 362 Lecture Notes Compactness

Theorem 5.22 (Bolzano-Weierstrass): Every bounded infinite subset of Rk has a limit point
in Rk .

Proof: Easy corollary to Thm. 5.18. 

27
MATH 362 Lecture Notes The Cantor Set

6 The Cantor Set

Theorem 6.1: Every nonempty perfect set in Rk is uncountable.

Proof: Let P Ă Rk be perfect. By defn., P has limit points, and must therefore be an
infinite set. Suppose P “ tx1 , x2 , . . .u is countable. For some r ą 0, let B1 “ Bpx1 , rq.
Having chosen balls B2 , . . . , Bn , let Bn`1 be a ball satisfying these properties:

(i) Bn`1 Ă Bn ,
(ii) xn < Bn`1 , and
(iii) Bn`1 X P , ∅.

Note that these assumptions (assumption (iii), in particular), applied to Bn guarantee

that such a Bn`1 may be found. Now, for each n P N, set Kn “ Bn X P. P is closed (by
definition) and Bn is compact (Heine-Borel), so each Kn is compact (Thm. 5.9). Also,
each Kn is nonempty, and Kn`1 Ă Kn , @ n. By construction, no xn P j K j , which means
Ş
Ş Ş
n Kn “ n pKn X Pq “ ∅, contradicting Coro. 5.12. ÑÐ 

The Cantor middle-thirds set

(i) How it is constructed — call En the points remaining in r0, 1s after the nth stage of middle-
thirds intervals is removed.

(ii) Note that

(a) Every number x P r0, 1s has a ternary expansion:

8
ÿ
x “ x1 ¨ 3 ´1 ´2
` x2 ¨ 3 ` x3 ¨ 3 ´3
` ¨¨¨ “ x j 3´ j ,
j“1

where each x j P t0, 1, 2u (not necessarily unique). E1 contains only those x P r0, 1s
expressible with x1 , 1, E2 contains those numbers in E1 expressible with x2 , 1, and so
on.
(b) r0, 1s “ E0 Ą E1 Ą E2 Ą ¨ ¨ ¨ ,
(c) the En have the finite intersection property, and
(d) each En is the union of 2n disjoint closed intervals, all of length 3´n . Hence, each En is
compact.
8
č
(iii) Cantor set is C :“ En . By Theorem 5.11, C , ∅ and it is compact (because closed). No
n“1
number in C requires a 1 in its ternary expansion (by (a)).

28
MATH 362 Lecture Notes The Cantor Set

(iv) Between any two numbers in r0, 1s, D x whose ternary expansion requires some xk “ 1. Thus,
C contains no interval.

(v) Given any x P C, ε ą 0, Bpx, εq X C is infinite. Thus, x is a limit point (ñ C is perfect).

29
MATH 362 Lecture Notes Series

7 Series

Definition 7.1: Given a sequence pa j q of complex numbers, consider the sequence of partial
sums psn q where
ÿn
sn :“ a j.
j“1

As with any sequence, psn q may converge to a complex number s, in which case we call s
ř
the sum of the infinite series j a j , writing

8
ÿ
s “ a j.
j“1
ř
If psn q diverges, then we say the series j aj diverges.

Theorem 7.2: Let pan q be a sequence of complex numbers, with an “ αn ` iβn for each n.
The series n an converges ô both of the real series n αn and n βn converge.
ř ř ř

Proof: Let
n
ÿ n
ÿ n
ÿ
sn “ aj “ αj ` i β j.
j“1 j“1 j“1

Now psn q is a sequence in C which, as a metric space, is indistinguishable from the

Euclidean space R2 . We know that a sequence in R2 converges iff it converges in each
of its components (here the real and imaginary parts). 

Theorem 7.3 (nth Term Test): If the series

ř
j aj converges, then limn an “ 0.

Proof: Let psn q denote the sequence of partial sums. Set tn “ sn`1 for all n. We have
that sn Ñ some L, and so tn Ñ L. Thus

an`1 “ sn`1 ´ sn “ tn ´ sn Ñ L ´ L “ 0.

30
MATH 362 Lecture Notes Series

ř8 n
Theorem 7.4 (Geometric series): For 0 ď x ă 1, n“0 x “ p1 ´ xq´1 .

Proof: Show
1 ´ xn`1 1 xn`1
1 ` x ` ¨ ¨ ¨ ` xn “ “ ´ .
1´x 1´x 1´x
But, for 0 ď x ă 1, limn xn “ 0 ñ result. 

ř
Theorem 7.5: Suppose an ě 0, @ n P N. Set sn :“ a1 ` ¨ ¨ ¨ ` an . The n an converges ô psn q
is bounded.

Proof: Note that psn q is monotone. The result follows from Theorem 2.1. 

Corollary 7.6 (Comparison test): If 0 ď an ď bn , @ n, then

ř ř
n an converges if n bn con-
verges.

n
ÿ
Proof: Let sn :“ ak . By supposition,
k“1

n
ÿ 8
ÿ
sn ď bk ď bk ,
k“1 k“1

where the latter expression is a fixed number. Thus, psn q is a bounded sequence, and
converges by Thm. 7.5. 

Theorem 7.7: Suppose pan q8

n“1
is a monotone decreasing sequence of positive numbers.
ř
Then the series n an converges iff the series
8
ÿ
2k a2k “ a1 ` 2a2 ` 4a4 ` 8a8 ` ¨ ¨ ¨
k“0

converges.

Proof:

31
MATH 362 Lecture Notes Series

ř
ñ: Assume that n an converges. Notice that

a2 ď a2 ,
2a4 ď a3 ` a4 ,
4a8 ď a5 ` a6 ` a7 ` a8 ,

n´1 a n
ř ř
etc. By the Comparison Test (Coro. 7.6), n2 2n ă 8. Hence n 2 a2n ă 8 as
well.
ð: This time

a1 ď a1 ,
a2 ` a3 ď 2a2 ,
a4 ` a5 ` a6 ` a7 ď 4a4 ,
..
.
a2n ` a1`2n ` ¨ ¨ ¨ ` a´1`2n`1 ď 2n a2n .

Since n 2n a2n converges, the sequence of partial sums of the series

ř ř
an are
ř
bounded, and hence the series n an converges.

8
ÿ 1
Corollary 7.8 (p-series): The series converges if p ą 1 and diverges if p ď 1.
n“1
np

Proof: For each n P N let an :“ n´p . Notice that

8 8 ˆ ˙n
ÿ ÿ 1 ÿ 1
n
2 a2n “ 2 ¨ n p “n
.
n n“1
p2 q n“1
2p´1

For p ą 1, this last series is a convergent geometric series ( n rn with |r| ă 1), and the
ř
ř
series n n´p converges by Thm. 7.7.
Now let p “ 1. In this case,
8 8
ÿ ÿ 1
2n a2n “ 2n ¨ “ 1 ` 1 ` 1 ` ¨ ¨ ¨ “ 8.
n“1 n“1
2n

The result follows again from Thm. 7.7.

Now let p ă 1. Then for each fixed n P N, np ă n, which implies n´p ą 1{n. Suppose
ř ´p ř
nn converged. Then by the Comparison Test, n 1{n converges as well. ÑÐ


32
MATH 362 Lecture Notes Series

ř
Definition 7.9: Let n an be a complex series.

ř ř
1. If n |an | converges, n an is said to converge absolutely.
ř ř ř
2. If an converges but |an | “ 8, we say that an is conditionally convergent.

converges ô @ ε ą 0, D N P N s.t.
ř
Theorem 7.10: The series j aj
ˇ ˇ
ˇm ˇ
ˇÿ ˇ
ˇ a jˇ ă ε whenever m ě n ě N.
ˇ ˇ
ˇ j“n ˇ

Proof: This is Cauchy criterion for convergence of the sequence of partial sums. 

ř
Theorem 7.11 (Absolute convergence test): Let pan q be a complex sequence. If n an con-
ř
verges absolutely, then n an converges.

Proof: Let sn “ a1 ` a2 ` ¨ ¨ ¨ ` an . By Thm. 7.10, it is enough to show that the sequence

psn q is Cauchy. Let ε ą 0. By assumption, D N s.t. k“N |ak | ă ε. Suppose n ą m ě N.
ř8

Then
8
ÿ
|sm ´ sn | “ |am`1 ` am`2 ` ¨ ¨ ¨ ` an | ď |am`1 | ` |am`2 | ` ¨ ¨ ¨ ` |an | ď |ak | ď ε.
k“N

Thus, psn q is a Cauchy sequence. 

Remarks:

n
ř
1. By the Absolute Convergence Test (Thm. 7.11), if z P C with |z| ă 1, then nz converges.
n diverges by the nth Term Test (Thm. 7.3).
ř
2. Note that, if |z| ą 1 in the above, then nz

Some practice with series:

Example 1:

33
MATH 362 Lecture Notes Series

2
ř ř
Suppose an ą 0 and n an ă 8. Prove n an ă 8.

Proof: By the convergence of the series, an Ñ 0, so D N P N s.t. 0 ă an ă 1, whenever

n ě N. Thus, 0 ă a2n ă an for n ě N. By the Comparison Test, n a2n ă 8.
ř


Example 2:
ř ř
Suppose an ą 0 and n an ă 8. Prove n 1{p1 ` an q “ 8.

Proof: As above, an Ñ 0, and so 1{p1 ` an q Ñ 1. By the nth Term Test,

ř
n 1{p1 ` an q
diverges. 

Example 3:
ř ř
Suppose n an converges absolutely, and let psn q denote the sequence of partial sums of n an .
ř
Prove n sn an converges.

ř8
Proof: Let L “ n“1 |an |. We have

|sn | “ |a1 ` ¨ ¨ ¨ ` an | ď |a1 | ` ¨ ¨ ¨ ` |an | ď L, @ n P N,

ř ř
and hence |sn an | “ |sn ||an | ď L|an |. Now, by comparison to the series n L|an |, n |sn an |
ř
(and hence n sn an ) converges (Absolute Convergence Test). 

Example 4:
ř
Suppose an ą 0 and n an ă 8. Let p be a polynomial of degree ą 0, with pp0q “ 0. Prove
ř
n ppan q converges.

Proof:
Case: ppxq “ xk , for k ě 1.
This case may be argued in the same way the case k “ 2 was (see above).

Case: ppxq “ c1 x ` c2 x2 ` ¨ ¨ ¨ ` ck xk , k ě 1.
Here, we have
|ppan q| ď |c1 |an ` |c2 |a2n ` ¨ ¨ ¨ ` |ck |akn ,

34
MATH 362 Lecture Notes Series

which implies

k 8 8
j
ÿ ÿ ÿ ÿ ÿ
|ppan q| ď |c j | an “ |c1 | an ` ¨ ¨ ¨ ` |ck | akn .
j“1 n“1 n n

ř
Since each of these series converges, n ppan q converges absolutely, and hence con-
verges (Absolute Convergence Test). 

b
n an , set α :“ lim sup
n
ř
Theorem 7.12 (Root test): For the series |an |.
n

(i) If α ă 1,
ř
n |an | converges.

(ii) If α ą 1,
ř
n an diverges.

(iii) If α “ 1, no conclusion.

Proof:

(i) Choose β so that α ă β ă 1. By supposition, D N P N s.t. |an |1{n ă β whenever

n ě N. Thus, |an | ă βn . But n βn ă 8 (geometric series), and so n |an | ă 8
ř ř

(Comparison Test).
(ii) Choose β s.t. 1 ă β ă α. Then |an | ą βn for infinitely many n. But limn βn “ 8,
and so an 9 0, and n an diverges by the nth Term Test.
ř
?n
(iii) In the case an “ 1, @ n, n an “ 8 but α “ 1 “ 1. Similarly, if an “ 1{n, then
ř
ř ?
1{n “ 1{ n n “ 1 (Theorem 9.7(c) in Ross). But if a “ 1{n2 ,
n 1{n “ 8 and p1{nq n
ř 2 2 1{n
?n 2
then n 1{n ă 8 despite the fact that p1{n q “ p1{ nq “ 1.

Theorem 7.13: For any sequence pan q of positive numbers,

an`1 ? ? an`1
lim inf ď lim inf n an ď lim sup n an ď lim sup .
n an n n n an

Proof: This is Theorem 12.2 in Ross. 

35
 MATH 362 Lecture Notes Series

ř
Corollary 7.14 (Ratio test): Let n an be a complex series with an , 0 @ n.

ˇ ˇ
ˇ an`1 ˇ ř
(i) If lim sup ˇ
ˇ ˇ ă 1, then
n |an | ă 8.
an ˇ
ˇ ˇ
ˇ an`1 ˇ
(ii) If DN P N s.t. @ n ě N, ˇ ˇ ˇ ě 1, then an 9 0 (and ř an diverges).
n
an ˇ

Note: There are no series for which the Ratio Test indicates convergence while the Root Test fails (see
Thm. 7.13). However, for the series
1 1 1 1 1 1 1 1
` ` 2 ` 2 ` 3 ` 3 ` ¨¨¨ ` n ` n ` ¨¨¨ ,
2 3 2 3 2 3 2 3
ˆ ˙n
an`1 3 ? ?
2n 1
lim sup “ lim “ `8, while lim sup n an “ lim 2´n “ ? .
n an nÑ8 2 n nÑ8 2

Proof: (of the Ratio Test)

(i) This follows immediately from Thm. 7.13 and the Root Test.
(ii) Notice that the supposition of (ii) implies that

0 ă |aN | ď |aN`1 | ď . . . ,

and thus Bp0, |aN |{2q contains at most finitely many points of the sequence pan q.
Thus an 9 0.

Theorem 7.15 (Ratio test from calculus): Suppose an , 0, @ n, and assume α :“

limn |an`1 {an | exists.

(i) α ă 1 ñ
ř
|an | ă 8.

(ii) α ą 1 ñ pan q is unbounded.

(iii) α “ 1 ñ no conclusion.

Proof: (i), (ii) follow from Theorem 2.8 and the Root Test. For (iii), consider the series
1{n and 1{n2 .
ř ř


36
MATH 362 Lecture Notes Series

Theorem 7.16 (Alternating Series Test): Suppose pan q is a monotone decreasing sequence
of positive reals with lim an “ 0. Then
8
ÿ
p´1qn`1 an converges.
n“1

řn k`1 a .
Proof: For each n P N let sn :“ k“1 p´1q k Notice

s2 “ a1 ´ a2 ě 0
s4 “ pa1 ´ a2 q ` pa3 ´ a4 q ě s2
..
.

That is, ps2n q is a monotone increasing. Also

s2 “ a1 ´ a2 ď a1
s4 “ a1 ` pa3 ´ a2 q´ ď a1
..
.

That is, ps2n q is bounded above by a1 . Thus s2n Ñ some L P R.

Let ε ą 0. Then D N P N s.t. n ě N implies
ε ε
|s2n ´ L| ă and an ă .
2 2
Thus, for n ě N,

|s2n`1 ´ L| “ |s2n ´ L ` a2n`1 |

ď |s2n ´ L| ` |a2n`1 |
ε ε
ă `
2 2
“ ε.
ř8 n`1 a
Therefore, s2n`1 Ñ L which, coupled with s2n Ñ L, gives sn Ñ L. Thus, n“1 p´1q n
converges. 

Example 5:
8
ÿ p´1qn`1
If p ą 0, then converges. For 0 ă p ď 1, this convergence is conditional.
n“1
np

Example 6:

37
MATH 362 Lecture Notes Series

Consider the conditionally convergent series

1 1 1
L :“ 1 ´ ` ´ ` ¨¨¨ .
2 3 4
By inspection, L satisfies 1{2 ă L ă 1 (in fact, L “ ln 2). Note that

1 1 1 1 1
L “ ´ ` ´ ` ¨¨¨
2 2 4 6 8
1 1 1 1
“ 0 ` ` 0 ´ ` 0 ` ` 0 ´ ` ¨¨¨ , (0 included between terms)
2 4 6 8
and if we add this to the series whose sum is L, we get

3 1 1 1 1 1 1 1 1
L “ 1` ´ ` ` ´ ` ` ´ ` ¨¨¨ ,
2 3 2 5 7 4 9 11 6
the right-hand side of which is simply a rearrangement of the original series. Thus, it is
possible for a rearrangement
ÿ ÿ
ank , an .
k n

Definition 7.17: Let n1 , n2 , . . . be a sequence of positive integers in which each n P N

ř
appears once and only once. Given a series an , the series
8
ÿ
ank
k“1
ř
is a rearrangement of an .

Theorem 7.18 (Riemann): Let x P R. There exists a rearrangement of

1 1 1
1´ ` ´ ` ¨¨¨
2 3 4
that converges to x.

Proof: Both of the series

1 1 1
1` ` ` ` ¨¨¨
3 5 7
and
1 1 1 1
` ` ` ` ¨¨¨
2 4 6 8

38
MATH 362 Lecture Notes Series

diverge. Choose N1 to be the smallest positive odd integer s.t.

1 1 1
S1 :“ 1 ` ` ` ¨¨¨ ` ą x.
3 5 N1

Then S1 ´ x ď 1{N1 . Now, let N2 be the smallest even positive integer s.t.
ˆ ˙
1 1 1
S2 :“ S1 ´ ` ` ¨¨¨ ` ă x.
2 4 N2

Then x ´ S2 ď 1{N2 . Now, let N3 be the smallest odd positive integer s.t.
ˆ ˙
1 1 1
S3 :“ S2 ` ` ` ¨¨¨ ` ą x.
2 ` N1 4 ` N1 N3

Then S3 ´ x ď 1{N3 . Continuing in this manner, we get sequences pN j q, pS j q. Notice

that N j ě j, and so 1{N j Ñ 0 as j Ñ 8. Thus, S j Ñ x.
Claim: The rearrangement

1 1 1 1 1 1 1 1
1` ` ¨¨¨ ` ´ ´ ´ ¨¨¨ ´ ` ` ¨¨¨ ` ´ ´ ¨ ¨ ¨ “ x.
3 N1 2 4 N2 2 ` N1 N3 2 ` N2
At issue here is that we need the sequence of partial sums of the above rearrangement
to converge to x. All we currently have is that the S j Ñ x, which is a subsequence
of that sequence of partial sums. We finish, however, by observing that each of the
intermediate partial sums is trapped between consecutive S j ’s. 

Proposition 7.19: Suppose an ě 0 @ n. Then

# +
ÿ ÿ
an “ sup an | F is a finite subset of N .
nPF

Proof: Exercise. 

ř
Corollary 7.20: If an ě 0 then all rearrangements of an yield the same value.

8
ÿ 8
ÿ
Proof: an and ank are both supremums over the same set. 
n“1 k“1

39
MATH 362 Lecture Notes Series

ř
Theorem 7.21 (Riemann): Let an be a conditionally convergent series of real numbers,
ř
and x P R. There exists a rearrangement of an that converges to x.

Proof: (Sketch)
ř ř
By assumption, an converges but n |an | diverges. For each n, set
# #
0, if an ă 0, ´an , if an ă 0,
pn :“ and qn :“
an , if an ě 0, 0, if an ě 0,

Notice that
an “ pn ´ qn , and |an | “ pn ` qn .
ř ř
Claim: pn “ 8 and qn “ 8.
ř ř
Suppose this were not so. Then pn ă 8. Since an converges, we would have
ÿ ÿ ÿ ÿ
qn “ ppn ´ an q “ pn ´ an

converges as well. But then

ÿ ÿ ÿ ÿ
|an | “ ppn ` qn q “ pn ` qn

converges. ÑÐ.
Also,
ÿ
an (convergent) ñ an Ñ 0 ñ pn , qn Ñ 0.

Now, mimic the proof given for Thm. 7.18. 

ř ř
Theorem 7.22 (Riemann): Let an be a complex series with |an | ă 8. Then every
rearrangement of the series converges to the same value.

Proof: Note that an “ αn ` iβn for real sequences pαn q, pβn q, and |αn | ď |an | ď |αn | ` |βn |,
ř
|βn | ď |an | ď |αn | ` |βn | for each n. Thus, by the Comparison Test, the series an is
absolutely convergent if and only if each of αn , βn are absolutely convergent. It is,
ř ř

thus, WLOG that we assume each an P R. (Verify this.)

Define pn , qn ě 0 as in the proof of Thm. 7.21.
ÿ ÿ ÿ
|an | ă 8 ñ pn , qn ă 8 (Comparison test).

40
MATH 362 Lecture Notes Series

ř ř
Let ank be a rearrangement of an .
ÿ ÿ ÿ ÿ ÿ ÿ
ank “ ppnk ´ qnk q “ pnk ´ qnk “ pn ´ qn (Coro. 7.20)
ÿ ÿ
“ ppn ´ qn q “ an .

Summary of rearrangements of series:

ř
1. If |an | ă 8, then order of terms doesn’t matter.
ř
2. If an conditionally convergent, then order always matters. It is useful to explore what
kinds of order changes are significant.

41
MATH 362 Lecture Notes Power Series

8 Power Series

Definition 8.1: Let pcn q be a complex sequence. The series

8
ÿ
cn zn
n“0

is called a power series.

Main question: For which z P C does a power series converge?

Theorem 8.2: Let α :“ lim sup |cn |1{n , and set R “ α´1 . (If α “ 0, set R “ `8; if α “ `8,
n
set R “ 0.) The power series 8 n
ř
n“0 cn z converges absolutely if |z| ă R and diverges if
|z| ą R.

Lemma 8.3: If psn q is a sequence in r0, 8q, lim sup sn ă 8, and c ě 0, then lim suppcsn q “
cplim sup sn q.

Proof: (of the lemma) Verify. 

Proof: (of Thm. 8.2)

Case: R “ 0
Let z P Czt0u. By supposition, lim sup |cn |1{n “ 8. Using the previous lemma, we thus
have
lim sup |z||cn |1{n “ 8 ñ lim sup |cn zn |1{n “ 8 ą 1.

Therefore, cn zn diverges by the Root Test.

ř

Case: R “ 8
Let z P C. Here

lim sup |cn |1{n “ 0 ñ lim |cn zn |1{n “ 0 ă 1

ÿ
ñ cn zn converges absolutely (Root Test).

42
MATH 362 Lecture Notes Power Series

Case: R P p0, 8q
For any z P C,
|z|
lim sup |cn zn |1{n “ |z| lim sup |cn |1{n “ .
R
The result follows from the Root Test. 

Examples:

1. For 8 n 1{n “ 1 ñ R “ 1. For z “ 1, get harmonic series (divergent). For

ř
n“1 z {n, limn p1{nq
any other z with |z| “ 1, the series converges (will not be proved).

2. For 8 n 2 2 1{n “ 1 ñ R “ 1. For |z| “ 1, series converges absolutely.

ř
n“1 z {n , limn p1{n q

Theorem 8.4 (Taylor’s Theorem with Remainder): Suppose f : ra, bs Ñ R, and let n P N.
Suppose also that f pn´1q is continuous on ra, bs and f pnq pxq exists @ x P pa, bq. Then Dc P
pa, bq s.t.
n´1
ÿ f pmq paq f pnq pcq
f pbq “ pb ´ aqm ` pb ´ aqn .
m“0
m! n!

Notes:

1. When n “ 1, this is the MVT.

2. Thm. holds on rb, as if b ă a.

n´1
ÿ f pmq paq
Proof: Define ppxq :“ px ´ aqm . Note: ppaq “ f paq, p1 paq “ f 1 paq, . . . ,
m“0
m!
ppn´1q paq “ f pn´1q paq. Consider

gpxq :“ f pxq ´ ppxq ´ Apx ´ aqn ,

where A is chosen so that gpbq “ 0. More specifically,

f pbq ´ ppbq
A :“ .
pb ´ aqn
By construction,

gpaq “ 0 “ gpbq ñ D c1 P pa, bq s.t. g1 pc1 q “ 0, (Rolle’s Thm.)

1 1 2
g paq “ 0 “ g pc1 q ñ D c2 P pa, c1 q s.t. g pc2 q “ 0,
..
.
gpn´1q paq “ 0 “ gpn´1q pcn´1 q ñ D c P pa, cn´1 q s.t. gpnq pcq “ 0.

43
MATH 362 Lecture Notes Power Series

Now,

f pbq ´ ppbq f pnq pcq

gpnq pcq “ f pnq pcq ´ An! “ 0 ñ “
pb ´ aqn n!
f pnq pcq
ñ f pbq “ ppbq ` pb ´ aqn .
n!


Example 7:

x3 x5
sin x “ x ´ ` ´ ¨¨¨ , @ x P R.
3! 5!

Proof: Think of a “ 0 and b “ x (fixed) in the theorem. By Taylor’s Thm.,

D cn P p0, xq s.t.
n´1
ÿ sinpmq p0q
sin x “ xm ` εn pxq,
m“0
m!
where ˇ ˇ
ˇ sinpnq pc q ˇ
n nˇ |x|n
|εn pxq| :“ ˇ x ˇ ď Ñ 0 as n Ñ 8.
ˇ
ˇ n! ˇ n!
So, our claim holds by the definition of the sum of an infinite series. 

44
 MATH 362 Lecture Notes Limits and Continuity in Metric Space

9 Limits and Continuity in Metric Space

Definition 9.1: Let X, Y be metric spaces, E Ă X, f : E Ñ Y, and p P E1 . We write f pxq Ñ q

as x Ñ p, or
lim f pxq “ q,
xÑp

if there is a point q P Y with the property that, @ ε ą 0, D δ ą 0 s.t.

dY p f pxq, qq ă ε for all x P E satisfying 0 ă dX px, pq ă δ.

Note: As in Calc I, the point p need not be in E (the domain of f ) and, even if

(i) p P E, and

(ii) lim f pxq exists,

xÑp

it need not be the case that lim f pxq “ f ppq.

xÑp

Theorem 9.2: Let X, Y, E, f , and p be as in Defn. 9.1. Then

lim f pxq “ q ô lim f ppn q “ q

xÑp nÑ8

for every sequence ppn q in E with pn , p (@ n) and lim pn “ p.

nÑ8

Proof:

ñ: Let ppn q be a sequence of points in E with pn , p and pn Ñ p, and let ε ą 0. Since

lim f pxq “ q, D δ ą 0 s.t.
xÑp

dY p f pxq, qq ă ε whenever dX px, pq ă δ px P Eq.

Since pn Ñ p, choose N P N s.t. n ě N ñ dX ppn , pq ă δ. Then dY p f ppn q, qq ă ε.

ð: Suppose lim f pxq , q. Then D ε ą 0 s.t. @ δ ą 0, the conditions of Defn. 9.1 fail.
xÑp
In particular, for each n P N D pn P E s.t. 0 ă dX ppn , pq ă 1{n but dY p f ppn q, qq ě ε.
Then ppn q is a sequence in E, with each p , pn Ñ p, and yet f ppn q 9 q ÑÐ.

45
 MATH 362 Lecture Notes Limits and Continuity in Metric Space

Definition 9.3: Suppose f : X Ñ Y, where X, Y are metric spaces, and let a P X. We say f
is continuous at a if @ ε ą 0 D δ ą 0 s.t.

dY p f pxq, f paqq ă ε whenever dX px, aq ă δ.

We may also write lim f pxq “ f paq.

xÑa

If f is continuous at every point of X, then f is said to be continuous on X.

Note: f continuous at a in the previous definition is equivalent to saying f pBpa, δqq Ă Bp f paq, εq.

Theorem 9.4: Suppose f : X Ñ Y, g : Y Ñ Z. If f is continuous at a, and g is continuous at

f paq, then g ˝ f is continuous at a.

Proof: Let ε ą 0. g cont. at f paq ñ D δ ą 0 s.t. gpBp f paq, δqq Ă Bpgp f paqq, εq.
f cont. at a ñ D γ ą 0 s.t. f pBpa, δqq Ă Bp f paq, δq. Together, the above imply
pg ˝ f qpBpa, γqq Ă Bppg ˝ f qpaq, εq, showing that g ˝ f is cont. at a. 

Definition 9.5: Let X, Y be sets, and f : X Ñ Y. For E Ă Y, the preimage of E is the set

f ´1 pEq :“ tx P X | f pxq P Eu.

Theorem 9.6: Suppose f : X Ñ Y. Then f is continuous on X ô f ´1 pΩq is open in

X @ Ω Ă Y open.

Proof:

ñ: Let Ω Ă Y be open (in Y). Let a P f ´1 pΩq. Ω open ñ D ε ą 0 s.t. Bp f paq, εq Ă Ω.

f is continuous at a ñ D Bpa, δq s.t. f pBpa, δqq Ă Bp f paq, εq Ă Ω. Thus, Bpa, δq Ă
f ´1 pΩq. a P f ´1 pΩq is arbitrary, so f ´1 pΩq is open.
ð: Let a P X, and let ε ą 0. Bp f paq, εq is open in Y ñ U :“ f ´1 pBp f paq, εqq is
open in X. Clearly a P U. Since U is open, D δ ą 0 s.t. Bpa, δq Ă U. Hence,
f pBpa, δqq Ă f pUq Ă Bp f paq, εq.

46
MATH 362 Lecture Notes Limits and Continuity in Metric Space

Theorem 9.7: Suppose f : X Ñ Y. Then f is continuous on X ô f ´1 pEq is closed in

X @ E Ă Y closed.

Proof: Hinges on the fact that f ´1 pYzEq “ Xz f ´1 pEq. 

Theorem 9.8: Suppose f : X Ñ Y, a P X. Then f is continuous at a ô f pan q Ñ f paq, @

sequences pan q in X converging to a.

Proof:

ñ: Let an Ñ a, and let ε ą 0. f is cont. at a ñ D δ ą 0 s.t. f pBX pa, δqq Ă BY p f paq, εq.
Choose N P N so that an P BX pa, δq whenever n ě N. Thus, f pan q P BY p f paq, εq if
n ě N.
ð: Suppose f is not cont. at a. Then D ε ą 0 s.t. f pBX pa, δqq 1 BY p f paq, εq, @ δ ą 0.
Thus, @ n D an P BX pa, 1{nq s.t. f pan q < BY p f paq, εq. Then an Ñ a, but f pan q 9 f paq.
ÑÐ

Theorem 9.9: Suppose f : X Ñ Y is continuous on X. If X is compact, then f pXq is compact.

Proof: Let tΩα u be an open cover of f pXq. By Thm. 9.6, t f ´1 pΩα qu is an open cover of
X. X is compact ñ D a finite subcover for X—that is,

X Ă f ´1 pΩα1 q Y f ´1 pΩα2 q Y ¨ ¨ ¨ Y f ´1 pΩαn q.

Thus,
f pXq Ă Ωα1 Y Ωα2 Y ¨ ¨ ¨ Y Ωαn .

47
MATH 362 Lecture Notes Limits and Continuity in Metric Space

Definition 9.10: Let pX, dq, pY, ρq be metric spaces, and f : X Ñ Y. We say that f is uniformly
continuous on X if and only if given any ε ą 0, Dδ ą 0 s.t. ρp f px1 q, f px2 qq ă ε for every x1 ,
x2 P X satisfying dpx1 , x2 q ă δ.

Example 8:

The function f pxq “ x2 mapping r1, 3s to R is uniformly continuous on r1, 3s.

To see this, let ε ą 0, and set δ “ ε{6. For x1 , x2 P r1, 3s, note that

|x1 ` x2 | ď |x1 | ` |x2 | ď 6.

Thus, assuming |x1 ´ x2 | ă δ, we have

| f px1 q ´ f px2 q| “ |x21 ´ x22 | “ |x1 ´ x2 ||x1 ` x2 | ă 6δ “ ε.

Example 9:

The function f pxq “ x2 mapping R to R is not uniformly continuous on R.

To see this, we first consider what the negation of uniform continuity looks like. Knowing
that the negation of a statement @x, Ppxq employing the universal quantifer takes the form
Dx, Ppxq, we see that we must find an ε ą 0 such that, for every choice δ ą 0, there exist at
least two points x1 “ x1 pδq, x2 “ x2 pδq in R for which |x1 ´ x2 | ă δ, but | f px1 q ´ f px2 q| ě ε.
δ 1 1 δ
Let’s try this for ε “ 1. If we take x1 pδq “ ` and x2 pδq “ , then clearly |x1 ´ x2 | “ ă δ,
2 δ δ 2
yet
δ δ 2 δ2
ˆ ˙ˆ ˙
| f px1 q ´ f px2 q| “ |x1 ´ x2 ||x1 ` x2 | “ ` “ ` 1 ě ε.
2 2 δ 4

The previous example exploited the fact that as x increases, x2 increases at an increasing rate. As
the domain of f extended to ˘8, a gap of 1 (or of any size) between f px1 q and f px2 q could be
despite a shrinking gap between x1 and x2 . As the next theorem indicates, this is impossible for a
continuous function f when the domain is compact.

48
MATH 362 Lecture Notes Limits and Continuity in Metric Space

Theorem 9.11: Let X, Y be metric spaces, and f : X Ñ Y. If X is compact and f is continuous

on X, then f is uniformly continuous on X.

Proof: Let d, ρ be the metrics in X, Y respectively. Suppose f is not uniformly

continuous. Then Dε ą 0 and sequences pxi q, pξi q in X such that

dpxi , ξi q Ñ 0 as i Ñ 8 but ρp f pxi q, f pξi qq ě ε.

Since X is compact, the sequence pxi q has a subsequence pxi j q that converges to a P X.
Since dpxi , ξi q Ñ 0, it follows that ξi j Ñ a, too.
Since f is continuous on X, it is continuous at x. Thus, Dδ ą 0 such that
ε
|x ´ a| ă δ ñ | f pxq ´ f paq| ă .
3
Further, DJ ě 1 (integer) such that

j ě J ñ dpxi j , aq ă δ and dpξi j , aq ă δ.

Thus, for j ě J we have

ε ε
ρp f pxi j q, f pξi j qq ď ρp f pxi j q, f paqq ` ρp f paq, f pξi j qq ď ` ă ε.
3 3


Definition 9.12: Suppose T : Rn Ñ Rm is a linear mapping; that is, for each x1 , x2 P Rn and
each pair of real scalars c1 , c2 , T has the property that Tpc1 x1 ` c2 x2 q “ c1 Tpx1 q ` c2 Tpx2 q.
Define the norm of the mapping T, denoted as }T}, to be

|Tpxq|
}T} :“ sup “ max |Tpxq|. (5)
xPRn zt0u |x| |x|“1

Theorem 9.13: Let T : Rn Ñ Rm be a linear map. Then T is uniformly continuous.

ε
Proof: Let ε ą 0, and take δ “ , where the norm }T} is as in Definition 9.12.
1 ` }T}
Then for x1 , x2 P Rn satisfying |x2 ´ x1 | ă δ we have
}T}
|Tpx2 q ´ Tpx1 q| “ |Tpx2 ´ x1 q| ď }T}|x2 ´ x1 | ď ε ă ε.
1 ` }T}


49
MATH 362 Lecture Notes Limits and Continuity in Metric Space

50
MATH 362 Lecture Notes Sequences of Functions

10 Sequences of Functions

Definition 10.1: Suppose fn : E Ñ C, n “ 1, 2, . . ., E some set. At each fixed x P E the

complex sequence p fn pxqq either converges or diverges. If it converges at x, we may define
f pxq “ limn fn pxq. If p fn pxqq converges for each x P E, we say that the sequence p fn q converges
pointwise to f on E, and write
fn Ñ f pointwise on E.

ř8
We may similarly define f pxq “ n“1 fn pxq, assuming that the series converges @ x P E.

Questions one might ask: Suppose fn Ñ f pointwise on r0, 1s.

1. Does fn1 Ñ f 1 pointwise on r0, 1s?

ş1 ş1
2. Does 0 fn pxq dx Ñ 0 f pxq dx pointwise on r0, 1s?

3. If each fn is continuous, is f continuous?

Notes:

(a) Another way to phrase Q3: Is it true for each a P r0, 1s that

lim lim fn pxq “ lim lim fn pxq?

xÑa nÑ8 nÑ8 xÑa

In fact, all three questions are about the validity of interchanging limit processes.

(b) The answer to Q3 is “Usually not!” As a counterexample, let fn pxq “ xn , n “ 1, 2, . . ., with

common domain r0, 1s. Certainly each fn is continuous, and the sequence p fn pxqq converges
@ x P r0, 1s. But the pointwise limit function is
#
0, if x P r0, 1q,
f pxq :“ lim fn pxq “ ,
nÑ8 1, if x “ 1.

which is discontinuous at x “ 1.
Pointwise limits of continuous functions need not be continuous.

(c) Suppose fn is defined on r0, 1s for n “ 1, 2, . . . by

$ „ 
1 1
& npn ` 1q, if x P , ,
fn pxq :“ n`1 n .
0, elsewhere.
%

51
MATH 362 Lecture Notes Sequences of Functions

Notice that fn Ñ 0 pointwise on r0, 1s (write limn fn pxq “: f pxq ” 0). But
ż1 ż1
fn pxq dx “ 1, @ n, while f pxq dx “ 0.
0 0

The limit of a sequence of integrals need not be the same as the integral of the pointwise limit of the
sequence of integrands.
?
(d) Suppose fn : R Ñ R is given by fn pxq “ sinpnxq{ n for n “ 1, 2, . . .. Then fn Ñ 0 pointwise
on R. (This convergence is uniform, as we will shortly define.) But
n cospnxq ?
fn1 pxq “ ? “ n cospnxq.
n
?
Thus fn1 p0q “ n Ñ `8 , 01 “ 0.
The pointwise limit of a sequence of functions need not have derivative that is the pointwise limit of
the sequence of derivatives.

(e) “Double sequences”

Let am,n :“ m{pm ` nq. Note that

lim lim am,n “ 0, but lim lim am,n “ 1.

mÑ8 nÑ8 nÑ8 mÑ8

Definition 10.2: Suppose fn : E Ñ C, n “ 1, 2, . . ., and f : E Ñ C. We say that p fn q converges

uniformly to f on E if @ ε ą 0 D N P N s.t.

| fn pxq ´ f pxq| ă ε, @ x P E,

whenever n ě N.

Examples:

?
1. sinpnxq{ n Ñ 0 uniformly on R.

2. xn Ñ 0 pointwise on r0, 1q, but not uniformly.

3. xn Ñ 0 uniformly on r0, bs, @ b P p0, 1q.

Proposition 10.3: If fn Ñ f uniformly on E, then fn Ñ f pointwise on E.

Proof: Verify. 

52
MATH 362 Lecture Notes Sequences of Functions

Definition 10.4: A sequence p fn q of complex functions on E is uniformly Cauchy on E if

@ ε ą 0 D N P N s.t. m, n ě N ñ

| fm pxq ´ fn pxq| ă ε, @ x P E.

Theorem 10.5: The sequence p fn q converges uniformly on E ô p fn q is uniformly Cauchy

on E.

Proof:

ñ: Suppose fn Ñ f uniformly on E. Let ε ą 0. Choose N P N s.t. n ě N ñ

| fn pxq ´ f pxq| ă ε{2, @ x P E. Then m, n ě N implies
ε ε
| fm pxq ´ fn pxq| ď | fm pxq ´ f pxq| ` | f pxq ´ fn pxq| ă ` , @ x P E.
2 2
Thus, p fn q is uniformly Cauchy.
ð: Assume p fn q is uniformly Cauchy. By implication, for each x P E, p fn pxqq8n“1
is a
8
Cauchy sequence in C. C is complete ñ p fn pxqqn“1 converges in C @ x P E. Thus,
define f pxq :“ limn fn pxq, making f : E Ñ C the pointwise limit function of p fn q.
Claim: fn Ñ f uniformly on E.
To prove this, let ε ą 0. Choose N P N as in the Defn. 10.4. Let n ě N. Then, if
x P E, we have
| fn pxq ´ f pxq| “ lim | fn pxq ´ fm pxq| ď ε,
mÑ8
since | fn pxq ´ fm pxq| ă ε for m ě N. This is true @ x P E.

Definition 10.6: Let fn : E Ñ C, and set sn pxq :“ nm“1 fm pxq. We say that 8
ř ř
n“1 fn pxq
converges uniformly on E if psn q converges uniformly on E.

Theorem 10.7 (Weierstrass M-test): Suppose that | fn pxq| ď Mn , @ x P E, @ n. If

ř8
n“1 Mn ă
8, then 8
ř
n“1 fn converges uniformly on E.

53
MATH 362 Lecture Notes Sequences of Functions

n
ÿ
Proof: For each n let sn pxq :“ fm pxq. It is enough to show that psn q is uniformly
m“1
8
ÿ
Cauchy on E (Thm. 10.5). Let ε ą 0. D N P N s.t. Mn ă ε. If n ą m ě N, then
n“N

|sn pxq ´ sm pxq| “ | fm`1 pxq ` ¨ ¨ ¨ ` fn pxq| ď Mm`1 ` ¨ ¨ ¨ ` Mn , @x P E

ÿ8
ď Mk ă ε.
k“N

Thus, psn q is uniformly Cauchy. 

Theorem 10.8: Suppose X is a metric space with limit point a P X, and that fn : X Ñ C, @ n.
Assume that fn Ñ f uniformly on Xztau. If lim fn pxq “ Ln P C for n “ 1, 2, . . ., then pLn q
xÑa
converges. Moreover,
lim f pxq “ lim Ln
xÑa nÑ8

(that is,
lim lim fn pxq “ lim lim fn pxqq.
xÑa nÑ8 nÑ8 xÑa

Proof: p fn q converges uniformly ñ p fn q uniformly Cauchy. And, since

|Lm ´ Ln | “ lim | fm pxq ´ fn pxq|,

xÑa

it follows that pLn q is a Cauchy sequence in C. C is complete ñ Ln Ñ some L P C.

Now, let ε ą 0. Choose N P N s.t. n ě N ñ

(i) |Ln ´ L| ă ε{3, and

(ii) | fn pxq ´ f pxq| ă ε{3, @ x P pXztauq.

D δ ą 0 s.t. x P Bpa, δqztau ñ | fN pxq ´ LN | ă ε{3. Then, for x P Bpa, δqztau,

| f pxq ´ L| ď | f pxq ´ fN pxq| ` | fN pxq ´ LN | ` |Ln ´ L| ă ε.

Corollary 10.9: The uniform limit of continuous functions is continuous.

54
MATH 362 Lecture Notes Sequences of Functions

Proof: We assume fn : X Ñ C is continuous on X @ n, with fn Ñ f uniformly on X. We

must show that f is continuous on X. To this end, let a P X be a limit point of X.

lim f pxq “ lim lim fn pxq (by the theorem)

xÑa nÑ8 xÑa
“ lim fn paq (by the continuity of fn )
nÑ8
“ f paq.

Since a P X is an arbitrary limit point, the result holds. 

Definition 10.10: Let X be a metric space. We denote

C pXq “ t f : X Ñ C | f is continuous on Xu,

Cb pXq “ t f P C pXq | f is bounded on Xu.

For f P Cb pXq, set } f } :“ supxPX | f pxq| ă 8.

Notes:

1. When X is compact, then C pXq “ Cb pXq.

2. For f, g P Cb pXq, dp f, gq :“ } f ´ g} defines a metric on Cb pXq.

Proposition 10.11: Let p fn q be a sequence in Cb pXq and let f P Cb pXq. Then

(i) p fn q is Cauchy in Cb pXq ô p fn q is uniformly Cauchy in X.

(ii) fn Ñ f in Cb pXq ô fn Ñ f uniformly on X.

Proof:

(i) dp fm , fn q “ supxPX | fm pxq ´ fn pxq|.

(ii) dp fn , f q “ supxPX | fn pxq ´ f pxq|.

Theorem 10.12: Cb pXq is complete.

55
MATH 362 Lecture Notes Sequences of Functions

Proof: Suppose p fn q is Cauchy in Cb pXq. Then p fn q is uniformly Cauchy on X, which

implies fn Ñ some f uniformly on X. Each fn is continuous on X, and since the
uniformly limit of continuous functions is continuous, so is f . One can show that the
uniform limit of bounded, continuous functions is bounded as well, and so we have
f P Cb pXq. By Prop. 10.11, fn Ñ f in Cb pXq. 

Theorem 10.13: Suppose fn Ñ f uniformly on ra, bs, and each fn P R on ra, bs. Then f P R
on ra, bs, and
żb żb
f “ lim fn .
a nÑ8 a

Proof: First, since each fn is bounded, f is bounded. Let ε ą 0. Choose n s.t. | fn ´ f | ă ε

on ra, bs, and let P be a partition of ra, bs s.t. UpP, fn q ´ LpP, fn q ă ε. Then

UpP, f q ď UpP, fn q ` εpb ´ aq and LpP, f q ě LpP, fn q ´ εpb ´ aq.

Therefore,
UpP, f q ´ LpP, f q ă ε ` 2εpb ´ aq ñ f P Rra, bs.

Now
ˇż ˇ ˇż ˇ
ˇ b żb ˇ ˇ b ˇ
f f ˇ “ ˇ p fn ´ f q ˇ
ˇ ˇ ˇ ˇ
´
ˇ a n
ˇ
a ˇ ˇ a ˇ
żb
ď | fn ´ f |
a
˜ ¸
ď sup | fn ´ f | pb ´ aq,
ra,bs

where the first of these factors goes to zero with n by uniform convergence. Thus,
żb żb
f “ lim fn .
a nÑ8 a

ř8
Corollary 10.14: If n“1 fn converges uniformly on ra, bs and each fn P R on ra, bs, then
ż b ˜ÿ
8
¸
8
ÿ
˜ż
b
¸
fn “ fn .
a n“1 n“1 a

56
 MATH 362 Lecture Notes Sequences of Functions

Proof: Let sn pxq :“ nj“1 f j pxq. Then, by the previous theorem,

ř

żb ÿ 8 żb żb ÿn żb 8 żb
ÿ
fn “ lim sn “ lim sn “ lim fj “ fn .
a n“1 a n n a n
j“1 a n“1 a

?
Note: We have seen (with fn “ sinpnxq{ n) that fn Ñ f uniformly on E with each fn differentiable does
not necessarily imply fn1 Ñ f 1 uniformly (or otherwise) on E. (In fact, it need not even be the case
that f 1 exists.)

Theorem 10.15: Suppose p fn q is a sequence of functions on ra, bs and that fn1 is continuous
on ra, bs @ n. Suppose, also, that

(I) p fn1 q converges uniformly on ra, bs.

(II) D x0 P ra, bs s.t. p fn px0 qq converges.

Then fn Ñ some f uniformly on ra, bs. Moreover,

f 1 pxq “ lim fn1 pxq, @ x P ra, bs.

nÑ8

Proof: Suppose fn1 Ñ g uniformly on ra, bs, and fn px0 q Ñ L. By Coro. 10.9, g P C pra, bsq.
By FTC, żx
fn pxq “ fn px0 q ` fn1 .
x0
Define the limit function ż x0
f pxq “ L ` g.
x
Then
ˇż x ˇ żb
ˇ 1
ˇ
| fn pxq ´ f pxq| ď | fn px0 q ´ L| ` ˇˇ p fn ´ gqˇˇ ď | fn px0 q ´ L| ` | fn1 ´ g|.
x0 a

This is a bound that does not depend on x P ra, bs. Moreover, the first term in this last
expression may be made arbitrarily small (since fn px0 q Ñ L), and so can the integral
term, by Thm. 10.13. In fact, we may choose N P N s.t.
żb
ε ε
| fn px0 q ´ L| ` | fn1 ´ g| ď ` , whenever n ě N, @ x P ra, bs.
a 2 2
Thus, fn Ñ f uniformly on ra, bs. Again, by FTC,

f 1 pxq “ gpxq “ lim fn1 pxq.

nÑ8

57
MATH 362 Lecture Notes Sequences of Functions

Theorem 10.16 (Weierstrass): D continuous f : R Ñ R s.t. f 1 pxq does not exist @ x P R.

Proof: Define φpxq “ |x|, for |x| ď 1, and extend φ periodically to all of R (picture).
Observe that

(i) |φpsq ´ φptq| ď |s ´ t|, @ s, t P R.

(ii) If the interval ps, tq contains no integer, then |φpsq ´ φptq| “ |s ´ t|.

Now, for x P R, define

8 ˆ ˙n
ÿ 3
f pxq :“ fn pxq, where fn pxq :“ φp4n xq.
n“1
4

Notice: Each fn p¨q is continuous on R. Also, | fn pxq| ď p3{4qn , @ x P R. By the Weierstrass

ř
M-test, fn converges uniformly on R, which means f is continuous on R.
ˆ ˙
1 1
Now, fix x P R. For each m, let δm :“ ˘ ¨ m , where the sign is chosen so that the
2 4
open interval with endpoints 4m px ` δm q and 4m x does not contain an integer.
ˇ f px ` δm q ´ f pxq ˇ
ˇ ˇ
Claim: ˇˇ ˇ Ñ 8 as m Ñ 8.
δm ˇ
1
(Note: If so, then f pxq does not exist.) Fix m for the moment.
ˆ ˙n
f px ` δm q ´ f pxq ÿ8
3 φp4n px ` δm qq ´ φp4n xq
“ γn , where γn :“ .
δm n“1
4 δm
If n ą m, then
4n
4n δm “ ˘ (an even integer) ñ γn “ 0 (φ has period 2).
p2qp4m q
If n “ m, then ˆ ˙m m
3 |4 δm |
|γn | “ |γm | “ “ 3m .
4 |δm |
If n ă m, then ˆ ˙n n
3 |4 δm |
|γn | ď “ 3n .
4 |δm |
Therefore,
ˇ ˇ
ˇ f px ` δm q ´ f pxq ˇ
ˇ ˇ ˇÿ m ˇ m´1
ÿ
γ
ˇ
ˇ “ ˇ ˇ
ˇ n ě |γm | ´ |γn |
δm
ˇ ˇ ˇ
ˇ ˇ
n“1 n“1

m´1
m
ÿ 3m ´ 1
ě 3 ´ 3n “ 3m ´
n“0
3´1

3m 3m
ě 3m ´ “ Ñ 8 as m Ñ 8.
2 2


58
MATH 362 Lecture Notes Sequences of Functions

Theorem 10.17 (Weierstrass approximation theorem): If f : ra, bs Ñ C is continuous, there

exists a sequence pPn q of polynomials such that Pn Ñ f uniformly on ra, bs. If f is real, then
the Pn may all be taken real as well.

Proof: WLOG, ra, bs “ r0, 1s, and f p0q “ f p1q “ 0 (Verify). Let
#
f pxq, if 0 ď x ď 1,
f˜pxq :“
0, otherwise.

f˜ is a continuous extension of f to all of R. Notice that f is uniformly continuous (since

continuous on compact set r0, 1s, and that this uniform continuity is maintained by f˜.
If we prove that there is a sequence of polynomials converging uniformly to f˜ on R,
then the result follows. In what follows, we identify f˜ with f .
For each n let Qn pxq :“ cn p1 ´ x2 qn , where
˜ż ¸´1
1
cn :“ 2 n
p1 ´ x q dx .
´1

ş1
By construction, Qn pxq ě 0 for x P r´1, 1s, and ´1 Qn pxq “ 1, @ n. Furthermore, we
have
ż1 ż1 ż 1{ ?n
1
:“ p1 ´ x2 qn dx “ 2 p1 ´ x2 qn dx ě 2 p1 ´ x2 qn dx
cn ´1 0 0
ż 1{ ?n
4 1
ě 2 p1 ´ nx2 q dx “ ? ą ? .
0 3 n n

(Note that gpxq :“ p1 ´ x2 qn ´ 1 ` nx2 satisfies gp0q “ 0 and g1 pxq ą 0 for 0 ă x ă 1, so

?
the 2nd inequality is justified.) Since cn ă n, it is the case for any δ ą 0 that
?
Qn pxq ď n p1 ´ δ2 qn , δ ď x ď 1,

so that Qn Ñ 0 uniformly in rδ, 1s.

For x P r0, 1s, define
ż1 ż 1´x
Pn pxq :“ f px ` tqQn ptq dt “ f px ` tqQn ptq dt
´1 ´x
ż1
“ f ptqQn pt ´ xq dt,
0

where, because t is only the dummy variable of integration (and is integrated out), this
last expression is a polynomial in x. So, pPn q is a sequence of polynomials, and is clearly
real if f is real.

59
MATH 362 Lecture Notes Sequences of Functions

Let ε ą 0, and choose δ ą 0 s.t. |y ´ x| ă δ ñ | f pyq ´ f pxq| ă ε{2 (possible by

uniform continuity). Let M :“ sup | f pxq|. Then,
ˇż ˇ
ˇ 1 ˇ
|Pn pxq ´ f pxq| “ ˇ r f px ` tq ´ f pxqsQn ptq dtˇ
ˇ ˇ
ˇ ´1 ˇ
ż1
ď | f px ` tq ´ f pxq|Qn ptq dt
´1
żδ
ε
ż ´δ ż1
ď 2M Qn ptq dt ` Qn ptq dt ` 2M Qn ptq dt
´1 2 ´δ δ
? ε
ď 4M np1 ´ δ2 qn ` (a bound independent of x)
2
ă ε,

for all large enough n. 

Some notes on power series:

1. Suppose f : p´R, Rq Ñ R is differentiable up to all orders. We might hope that

8
ÿ f pnq p0q n
f pxq “ x .
n“0
n!

This occurs if the remainder in Taylor’s Theorem (Thm. 8.4) goes to 0.

ř8 n,
2. Instead, one might start with power series n“0 cn x define f to be the series sum, and ask
what properties the true f has?

ř8 n.
Theorem 10.18: Suppose the radius of convergence R ą 0 for n“0 cn x Define f pxq :“
ř8 n
n“0 cn x for ´R ă x ă R. Then

(i) the series defining f converges uniformly on every compact subset of p´R, Rq.
8
ÿ
(ii) f 1 pxq “ cn nxn´1 , @ x P p´R, Rq.
n“1

Proof:

(i) Let ε ą 0. Then |x| ď R ´ ε ñ |cn xn | ď |cn |pR ´ εqn . But since a power
series converges absolutely within its (open) interval of convergence, we take
Mn “ |cn |pR ´ εqn in the Weierstrass M-test to get the result.

60
MATH 362 Lecture Notes Sequences of Functions

(ii) We know n1{n Ñ 1 as n Ñ 8 ñ lim sup |ncn |1{n “ lim sup |cn |1{n ñ the radii of
convergence for the two series 8
ř n
ř8 n´1 are the same. By part
n“0 cn x and n“1 cn nx
(i), the latter series converges uniformly on each rR ´ ε, R ` εs. By Theorem 10.15,
f 1 pxq equals this series on each rR ´ ε, R ` εs.

Corollary 10.19: If f is as stated in the previous theorem, then

8
ÿ
f pkq pxq “ npn ´ 1q ¨ ¨ ¨ pn ´ k ` 1qcn xn´k , @ x P p´R, Rq, k “ 0, 1, 2, . . . .
n“k

As a result, ck “ f pkq p0q{k!, @ k.

an xn “ bn xn @ x P p´δ, δq, then an “ bn @ n.

ř ř
Corollary 10.20: If

Proof: Both series define the same function f pxq as above. Thus

f pkq p0q
ak “ “ bk .
k!


Example 10:

Consider the series

1
1 ` x ` x2 ` ¨ ¨ ¨ “ , x P p´1, 1q.
1´x
First, we claim that this series does not converge uniformly on p´1, 1q

Proof: Each partial sum of the series is bounded on p´1, 1q. Since the uniform limit
of a bounded sequence of functions is bounded, if the series did converge uniformly
we would have that 1{p1 ´ xq was bounded on p´1, 1q, and this is not so. 

By the previous results we have that

ˆ ˙1
1
“ 1 ` 2x ` 3x2 ` ¨ ¨ ¨ , on p´1, 1q.
1´x

61
MATH 362 Lecture Notes Sequences of Functions

Theorem 10.21 (Summation by parts): Let a0 , . . . , an , b0 , . . . , bn P C. Set sm “ a0 ` ¨ ¨ ¨ ` am ,

with s´1 “ 0. Then
n
ÿ n´1
ÿ
am bm “ sn bn ` sm pbm ´ bm`1 q.
m“0 m“0

Proof:
n
ÿ n
ÿ n
ÿ n
ÿ
a m bm “ psm ´ sm´1 qbm “ sm bm ´ sm´1 bm
m“0 m“0 m“0 m“1

n
ÿ n´1
ÿ n´1
ÿ
“ sm bm ´ sm bm`1 “ sn bn ` sm pbm ´ bm`1 q.
m“1 m“0 m“0


Theorem 10.22 (Abel’s theorem): Suppose 8

ř
cn converges (and, as a result, we know
ř8 n“0 n
the radius of convergence of the series n“0 cn x is at least 1). Then
8
ÿ 8
ÿ
lim cn xn “ cn .
xÑ1
n“0 n“0

řn
Proof: Let sn :“ m“0 cm , s´1 :“ 0. By Thm. 10.21 (with bm “ xm ), we have
n
ÿ n´1
ÿ n´1
ÿ
cm xm “ sn xn ` sm pxm ´ xm`1 q “ sn xn ` p1 ´ xq sm xm , @ x P p´1, 1q.
m“0 m“0 m“0
ř
cn converges, so the sequence psn q is bounded. Since |x| ă 1, when we let n Ñ 8 we
have sn xn Ñ 0. Thus,
8
ÿ ÿ8
f pxq :“ cn xn “ p1 ´ xq sn xn , @ x P p´1, 1q.
n“0 n“0
ř8
Define s :“ n“0 cn , so sn Ñ s. Intuition: For large n (i.e., for n ě some N), sn « s. Then
˜ ¸
N
ÿ 8
ÿ
f pxq « p1 ´ xq sn xn ` sxn
n“0 n“N`1
˜ ¸
N
ÿ
n N`1 1
« p1 ´ xq sn x ` sx
n“0
1´x
N
ÿ
“ p1 ´ xq sn xn ` sxN`1 .
n“0

62
MATH 362 Lecture Notes Sequences of Functions

Taking the limit as x Ñ 1, we get lim f pxq “ s. Finish this out: Choose N and a
xÑ1
neighborhood p1 ´ δ, 1s to make it all work. (See Baby Rudin, Section 8.2) 

Example 11:

Manipulating the geometric series gives us

1 1
“ “ 1 ´ x ` x2 ´ x3 ` x4 ´ ¨ ¨ ¨ , @ x P p´1, 1q.
1`x 1 ´ p´xq
Thus, by the Corollary 10.14,

x2 x3 x4
lnp1 ` xq “ x ´ ` ´ ` ¨¨¨ , in p´1, 1q.
2 3 4
Since
1 1 1
1´ ` ´ ` ¨¨¨
2 3 4
converges (Thm. 7.16, which is the Alternating Series Test), Abel’s Thm. says that it converges
to lnp2q.

Definition 10.23: A function f which has k derivatives in an open connected set Ω is

denoted by f P C k pΩq. If f P C 8 pΩq then f is said to be smooth.

Lemma 10.24: Suppose amn ě 0, @ pm, nq P N2 . Then

ÿ 8
8 ÿ 8 ÿ
ÿ 8
amn “ amn .
m“1 n“1 n“1 m“1

Proof: Set s “ suppm,nqPF amn , where this sup is taken over all finite F Ă N2 . Note
ř

that both
ÿ8 ÿ 8 ÿ8 ÿ 8
amn ě s and amn ě s.
m“1 n“1 n“1 m“1
If s “ `8 then we are done.
ř ř
Suppose s ă 8. We show that m n amn ď s as well. (The proof that the sum with
indices reversed is ď s is similar.) Suppose that D M P N s.t.
M ÿ
ÿ 8
amn ą s.
m“1 n“1

63
MATH 362 Lecture Notes Sequences of Functions

Then we may choose N1 , N2 , . . . , NM P N s.t.

Nm
M ÿ
ÿ
amn ą s. ÑÐ
m“1 n“1

Thus,
M ÿ
ÿ 8 8 ÿ
ÿ 8
amn ď s, for M “ 1, 2, . . . . ñ amn ď s.
m“1 n“1 m“1 n“1


ř ř ř ř ř ř
Corollary 10.25: Suppose amn P C and m n |amn | ă 8. Then m n amn “ n m amn .

Proof: WLOG, each amn is real. Set pmn “ maxtamn , 0u and qmn “ maxt´amn , 0u, @ m, n P
N, so that we have pmn ` qmn “ |amn | and pmn ´ qmn “ amn . The convergence of
ř ř ř ř ř ř
m n |amn | implies the convergence of both m n pmn and m n qmn . Thus
ÿÿ ÿÿ ÿÿ ÿÿ
amn “ ppmn ´ qmn q “ pmn ´ qmn
m n m n m n m n
ÿÿ ÿÿ
“ pmn ´ qmn (by Lemma 10.24)
n m n m
ÿÿ ÿÿ
“ ppmn ´ qmn q “ amn .
n m n m


Theorem 10.26 (Taylor’s Theorem): Suppose f pxq “ 8 n

ř
n“0 cn x in p´R, Rq. If a P p´R, Rq,
then
8
ÿ f pnq paq
f pxq “ px ´ aqn , @ x P pa ´ δ, a ` δq, δ “ R ´ |a|.
n“0
n!

Proof: Let x P pa ´ δ, a ` δq.

8 8n ˆ ˙
ÿ
n
ÿ
ÿ n
f pxq “ cn px ´ a ` aq “ cn px ´ aqm an´m
n“0 n“0 m“0
m
n
8 ÿ ˆ ˙
ÿ n
“ cn px ´ aqm an´m (6)
n“0 m“0
m
8 ÿ 8 ˆ ˙
ÿ n
“ cn px ´ aqm an´m (7)
m“0 n“m
m
« ff
8 8 ˆ ˙
ÿ ÿ n n´m
“ cn a px ´ aqm .
m“0 n“m
m

64
MATH 362 Lecture Notes Sequences of Functions

The bracketed expression is just a constant (depending on m), which proves that f has
a power series expansion centered at x “ a, which then must be the Taylor series. The
only detail left to us is the justification for the exchange of summation in going from
(6) to (7). This switch is permissible if
n
8 ÿ ˆ ˙
ÿ n
|cn | |x ´ a|m |a|n´m
n“0 m“0
m

is finite. But
n
8 ÿ ˆ ˙ 8 n ˆ ˙ 8
ÿ n ÿ ÿ n ÿ n
|cn | m n´m
|x ´ a| |a| “ |cn | |x ´ a|m |a|n´m “ |cn | p|x ´ a| ` |a|q ,
n“0 m“0
m n“0 m“0
m n“0

and |x ´ a| ă δ “ R ´ |a|, so |x ´ a| ` |a| ă R. Thus, this power series converges

absolutely. 

Definition 10.27: Suppose Ω Ă R is open, f : Ω Ñ R. If @ a P Ω D δ ą 0 s.t.

8
ÿ f pnq paq
f pxq :“ px ´ aqn , for |x ´ a| ă δ,
n“0
n!

then f is said to be analytic in Ω, or real analytic.

ř8 n
Proposition 10.28: If f pxq “ n“0 cn x for x P p´R, Rq, then f is analytic in p´R, Rq.

Proof: This is just Theorem 10.26 rephrased so as to use the new definition. 

Question: Let f pxq “ 1

1`x2
, x P R. Notice that f P C 8 pRq. Moreover,

f pxq “ 1 ´ x2 ` x4 ´ x6 ` ¨ ¨ ¨ , if |x| ă 1.

Why does this series stop converging for |x| ě 1?

65
MATH 362 Lecture Notes Sequences of Functions

Answer: Complex analysis. We have

1
“ 1 ´ z2 ` z4 ´ z6 ` ¨ ¨ ¨ , if |z| ă 1, z P C.
1 ` z2
But f pzq is undefined for z “ ˘i. Now f is analytic on R. In fact, the radius of
?
convergence of the Taylor series of f at 1 is 2. To find the Taylor series at 1, we
write
1 a b
2
“ ` (partial fractions),
1`z z`i z´i
and
ˆ ˙˜ ¸
8 ˆ ˙n
1 1 1 1 1 ÿ z´1
“ “ z´1
“ ,
z´i z´1`1´i 1´i 1` 1 ´ i n“0 1 ´ i
1´i
?
which is OK if |pz ´ 1q{p1 ´ iq| ă 1 ñ |z ´ 1| ă 2.

ř
Definition 10.29: Suppose I is countable. If αPI |aα | ă 8 (here aα P C), define

ÿ 8
ÿ
aα :“ aαpnq ,
αPI n“1

where αp1q, αp2q, αp3q, . . . is any ordering of I. (Note: The right hand side is well-defined
by our “absolute convergence–rearrangement” theorem.)

Theorem 10.30: Let I be a set. For each α P I, let aα ě 0. Define

ÿ ÿ
aα “ sup aα ,
αPI αPF

where this sup is taken over all finite sets F Ă I.

ř
(i) If α aα ă 8, then only countably-many aα ą 0.

(ii) If I1 Ă I2 Ă I3 Ă ¨ ¨ ¨ , and Y8 I “ I, then

n“1 n
ÿ ÿ
lim aα “ aα .
nÑ8
αPIn αPI

Proof: Exercise. 

66
MATH 362 Lecture Notes Sequences of Functions

ř
Corollary 10.31: If I1 Ă I2 Ă ¨ ¨ ¨ , Y8 I “ I with I countable, and
n“1 n αPI |aα | ă 8 with
aα P C, then
ÿ ÿ
lim aα “ aα .
nÑ8
αPIn αPI

Theorem 10.32: Suppose f is analytic in p´R, Rq. Set the zero set of f Zp f q :“ tx P
p´R, Rq | f pxq “ 0u. If f ı 0, then Zp f q has no limit point in p´R, Rq. (And, this implies
Zp f q is countable.)

ˇ (
Proof: f is continuous ñ Zp f q is closed in p´R, Rq. Let E :“ Z1 p f q “ x P p´R, Rq ˇ x is limit point of Zp f q .
E is closed in p´R, Rq. Note: E Ă Zp f q.
Claim: E is open in p´R, Rq.
Let a P E. D distinct a1 , a2 , . . . in Zp f q with an Ñ a. By Rolle’s Thm., D a sequence pbn q
with bn Ñ a s.t. f 1 pbn q “ 0, @ n. We have f paq “ 0. f 1 is continuous (since it is analytic),
and f 1 pbn q “ 0 with bn Ñ a, so f 1 paq “ 0. Proceeding similarly, we have f pkq paq “ 0,
@ k P N. But, since f is analytic,
8
ÿ f pkq paq
f pxq “ px ´ aqk “ 0
k“0
k!

for x near a. Thus, the entire interval pa ´ δ, a ` δq Ă E. This proves the claim.

So, E is both closed and open, yet a subset of p´R, Rq, a connected set. This means that
either E “ ∅ or E “ p´R, Rq. Since f ı 0, so E “ ∅. 

67
 MATH 362 Lecture Notes Measure Theory

11 Measure Theory

11.1 Rings of sets and additive set functions

Definition 11.1: A family R of sets is called a ring (of sets) if

AYBPR and AzB P R, @ A, B P R.

If all of the sets in R are subsets of a set X, and if X P R, then R is called an algebra (of
sets). A ring/algebra R is called a σ-ring/σ-algebra if
8
ď
An P R, whenever An P R, n “ 1, 2, . . . .
n“1

That is, a σ-ring/algebra is closed under countable unions.

Note:

(i) For A, B P R a ring, A X B “ AzpAzBq. Thus, A X B P R as well.

(ii) By repeated application of (i) and Defn. 11.1, it is clear that rings are closed under finite
unions and intersections.

(iii) Given sets An , n “ 1, 2, . . ., ˜ ¸

8
č 8
ď
An “ A1 z pA1 zAn q .
n“1 n“1

Thus, if R is a σ-ring, then it is closed under countable intersections.

(iv) Given any collection C of sets, there is always a σ-algebra that contains them. Simply let
ď
X“ A, and then take the σ-algebra to be the power set of X. Moreover, it can be shown
APC
that the intersection over all σ-algebras containing C is once again a σ-algebra, the smallest
one containing C.

Definition 11.2: We say that φ is a set function on R if φ assigns to every A P R a number

φpAq P p´8, 8s. φ is additive if

φpA Y Bq “ φpAq ` φpBq, whenever A X B “ ∅,

68
 MATH 362 Lecture Notes 11.1 Rings of sets and additive set functions

and countably additive if

˜ ¸
8
ď 8
ÿ
φ An “ φpAn q, whenever each Ai X A j “ ∅ for i , j
n“1 n“1

(that is, whenever the Ai are (pairwise) disjoint).

A set function φ : R Ñ r0, 8s is said to be nonnegative.

Note:

(i) If φ is additive, then

• φp∅q “ 0.
řn
• φpA1 Y ¨ ¨ ¨ Y An q “ i“1 φpAi q, if the Ai are disjoint.
• φpA Y Bq ` φpA X Bq “ φpAq ` φpBq.
• If φ is a nonnegative set function, then φpAq ď φpBq, if A Ă B.
• If B Ă A with φpBq ă 8, then φpAzBq “ φpAq ´ φpBq.

(ii) Countable additivity of φ requires that 8 n“1 φpAn q converge absolutely whenever it con-
ř

verges, since unions are the same for all “rearrangements”. (This assumption naturally
applies to all nonnegative set functions φ.)
It also implies that, regardless of whether R is just a ring or a σ-ring, φ is defined on the
smallest σ-ring containing R.

(iii) Countable additivity of φ implies additivity, since every ring R contains the empty set.

Theorem 11.3: Suppose φ is countably additive on ring R. Let An P R for n “ 1, 2, . . .,

with A1 Ă A2 Ă ¨ ¨ ¨ , and define A :“ 8n“1 An . If A P R, then φpAn q Ñ φpAq as n Ñ 8.
Ť

Proof: Let
B1 :“ A1 , B2 :“ A2 zA1 , . . . , B j :“ A j zA j´1 , . . . ,
Notice that the B j are disjoint, that An “ B1 Y ¨ ¨ ¨ Y Bn , @ n, and A “ Yn Bn . By the
additivity and σ-additivity of φ, we have
n
ÿ 8
ÿ
φpAn q “ φpB j q and φpAq “ φpBn q.
j“1 n“1

69
 MATH 362 Lecture Notes 11.1 Rings of sets and additive set functions

Definition 11.4: An interval in Rp is the set of x “ px1 , . . . , xp q P Rp satisfying

ai ď xi ď bi , i “ 1, 2, . . . , p,

where either of these “ď” symbols may be replaced with “ă”.

The finite union of intervals is known as an elementary set. We denote the collection of
elementary sets in Rp by E .

For an interval I in Rp , we define

p
ź
mpIq :“ pbi ´ ai q “ pb1 ´ a1 qpb2 ´ a2 q ¨ ¨ ¨ pbp ´ ap q.
i“1

When A P E is the union of disjoint intervals I1 , . . . , In in Rp , we set

n
ÿ
mpAq :“ mpIi q.
i“1

Note:

1. Every p-cell in Rp is an interval.

2. The possibility that some ai “ bi in the above is not excluded. Thus, even ∅ is considered an
interval.

3. E is a ring, but not a σ-ring.

Proof: That E is closed under unions should be clear. To show that E is closed
under complementation, suppose A, B P E and write A “ A1 Y ¨ ¨ ¨ Y An , B “
B1 Y ¨ ¨ ¨ Y Bm , with each Ai , B j an interval. Notice that
˜ ¸
ď n
ď
AzB “ Ai zB “ pAi zBq,
i i“1

which reduces the question to whether or not IzB P E whenever I is an interval

and B P E . But ¨ ˛
ď čm
IzB “ Iz ˝ B j ‚ “ pIzB j q,
j j“1

which further reduces the question. Now we must show that

(i) IzJ P E whenever I, J are intervals, and

70
MATH 362 Lecture Notes 11.1 Rings of sets and additive set functions

(ii) E is closed under finite intersections.

We prove (i) by considering the different ways two intervals (in R2 ) can overlap
(the higher-dimensional case can be similarly considered):
To prove (ii), it is enough to show that à X B̃ P E whenever Ã, B̃ P E . Write
à “ Ã1 Y ¨ ¨ ¨ Y Ãñ and B̃ “ B̃1 Y ¨ ¨ ¨ Y B̃m̃ , where each Ãi , B̃ j is an interval. Then
˜ ¸ ¨ ˛
č ď ñ č ď m̃ ď m̃
ñ ď
à B̃ “ Ãi ˝ B̃ j ‚ “ pÃi X B̃ j q.
i“1 j“1 i“1 j“1

Thus, (ii) comes down to whether the intersection of two intervals is in E . The
pictures above once again should convince.
To see that E is not a σ-ring it is enough to give a counterexample. In R1 , each
8
ď
interval r2n, 2n ` 1s P E , but r2n, 2n ` 1s < E . 
n“1

4. For each A P E , A is the finite union of disjoint intervals.

Proof: A P E ñ A “ A1 Y ¨ ¨ ¨ Y An with each A j an interval. We carry out a

similar “disjunctification” process as in the proof of Thm. 11.3: Set

B1 :“ A1 , (already an interval)
B2 :“ A2 zA1 , (look at pictures of cases—union of disjoint intervals)
B3 :“ A3 zpA1 Y A2 q, (etc.)
..
.
B j :“ A j zpA1 Y ¨ ¨ ¨ Y A j´1 q
..
.

Furthermore, each B j P E , and the

Ť Ť
Clearly the B j are disjoint, and j Bj “ k Ak .
union of disjoint intervals. 

5. Our expression for mpAq is well-defined. That is, given any A P E and any two decomposi-
tions of A into disjoint intervals, the expression for mpAq arising from either decomposition
yields the same result.

Proof: Read excerpt pp. 147–159 from “Excursions into Mathematics”. This deals
only with the p “ 2 case. 

6. m is additive on E by virtue of how we defined it.

71
 MATH 362 Lecture Notes 11.1 Rings of sets and additive set functions

Definition 11.5: A nonnegative additive set function φ defined on E is regular if the fol-
lowing holds: To each A P E and to every ε ą 0 there exist sets F, G P E with F closed, G
open, and F Ă A Ă G and

φpGq ´ ε ď φpAq ď φpFq ` ε.

Example 12:

1. Our set function m is regular on E .

2. Let α : R Ñ R be monotone increasing. By the monotone-ness of α,

αpx´q :“ suptαpyq | y ă xu “ lim αpyq

yÑx´

and
αpx`q :“ inftαpyq | x ă yu “ lim αpyq
yÑx`

are defined for each x P R. (In fact, αpx´q “ αpx`q “ αpxq for all but countably many x.)
For the four types of intervals in R1 , define the set function µ in the following way: Set

µpra, bqq :“ αpb´q ´ αpa´q,

µpra, bsq :“ αpb`q ´ αpa´q,
µppa, bsq :“ αpb`q ´ αpa`q,
µppa, bqq :“ αpb´q ´ αpa`q.

If we extend µ to E as before so as to make it additive on E , then µ is regular.

Note:

1. Have defined m on A P E only. How about mpCq, where C < E is the Cantor middle-thirds
set? C does belong to any σ-ring containing all intervals in R1 . So, extending m to be countably
additive on countable unions of disjoint intervals yields mpCq (see Exercise P29 on Problem
Set 9.)

2. But what of ‘weirder’ sets, like the rationals inside r0, 1s? One attempt (Jordan): extend m to
such sets in the following way:

For a set A Ă Rp , Jordan defined the upper content of A

# +
n
ÿ n
ď
µpAq :“ inf mpAi q | each Ai is an interval, A Ă Ai , Ai X A j “ ∅ for i , j ,
i“1 i“1

72
 MATH 362 Lecture Notes 11.2 Outer measure

and the lower content

# +
ÿn n
ď
µpAq :“ sup mpAi q | each Ai is an interval, A Ą Ai , Ai X A j “ ∅ for i , j .
i“1 i“1

When A has the property that µpaq “ µpAq, Jordan defined the content µpAq of A to
be this common value.

Indeed, µ extends m (i.e., collection of sets having content is a ring that properly contains E ,
and µ and m agree on E ; see HW). This extension of m is still unsatisfactory:

(a) The set of rational points inside a bounded interval does not have content.
(b) There exist open bounded sets which do not have content.
(c) While the sets which have content form a ring, µ is is not countably additive on this
ring.

11.2 Outer measure

Definition 11.6: Let µ be additive, regular, nonnegative and finite on E . Let E Ă Rp , and
define the outer measure of E (corresponding to µ) to be
# +
8
ÿ 8
ď
µ pEq :“ inf
˚
µpAn q | each An P E , with An open, E Ă An .
n“1 n“1

Note: m is one such set function which satisfies the requirements of µ in Defn. 11.6. We will use µ˚
when speaking generally, and m˚ (called Lebesgue outer measure) when speaking about the outer
measure corresponding to m.

Theorem 11.7: µ˚ has the following properties:

(i) 0 ď µ˚ pEq ď 8, @ E Ă Rk .

(ii) µ˚ p∅q “ 0.

(iii) µ˚ is monotone. That is, µ˚ pAq ď µ˚ pBq whenever A Ă B Ă Rp .

(iv) µ˚ is extension of µ that generated it per Defn. 11.6 (that is, µ˚ pAq “ µpAq, @ A P E ).
˜ ¸
8
ď ÿ8
(v) µ˚ is σ-subadditive. That is, µ˚ En ď µ˚ pEn q.
n“1 n“1

73
MATH 362 Lecture Notes 11.2 Outer measure

Proof:

(i) Part of what is being asserted is that µ˚ pEq is defined for every subset E of Rp .
This follows from the fact that every subset is eventually contained in the union
of expanding open intervals Jn in Rp of diameter n, centered at the origin. The rest
of the assertion is clear.
(ii) Take each An “ ∅ as one covering of ∅ satisfying Defn. 11.6. Recall that µp∅q “ 0.
(iii) For A Ă B, every countable open covering of B in Defn. 11.6 is also a covering of
A. Since the opposite is not true, µ˚ pAq is an infimum over a larger set than µ˚ pBq.
(iv) Let A P E . For each ε ą 0, the regularity of µ implies that D open G P E s.t. A Ă G
and µpGq ď µpAq ` ε. But µ˚ pAq ď µpGq, showing that µ˚ pAq is a lower bound for
the set pµpAq, 8s. Thus, µ˚ pAq ď µpAq.
Fix ε ą 0. By Defn. 11.6, D a sequence pAn q of open sets in E s.t. A Ă n An
Ť

and 8 n“1 µpAn q ď µ pAq ` ε. Again employing the regularity of µ, we have that
ř ˚

D closed F P E s.t. F Ă A and µpFq ě µpAq ´ ε. F is compact (all elementary sets

Ť
are bounded), and F Ă n An . Thus, D N P N s.t.

F Ă A1 Y A2 Y ¨ ¨ ¨ Y AN .

So,
N
ÿ 8
ÿ
µpAq ď µpFq`ε ď µpA1 Y¨ ¨ ¨YAN q`ε ď ε` µpAn q ď ε` µpAn q ď µ˚ pAq`2ε.
n“1 n“1

Thus, µpAq ď µ˚ pAq, showing that µ˚ pAq “ µpAq for A P E .

(v) If µ˚ pEn q “ 8 for some n P N, then (v) holds trivially. So, WLOG we assume
µ˚ pEn q ă 8, @ n. Let ε ą 0. For each n P N, let pAnk q8
k“1
be a sequence of open
sets in E covering En satisfying
8
ÿ
µpAnk q ď µ˚ pEn q ` 2´n ε.
k“1

Thus,
˜ ¸
ď 8 ÿ
ÿ 8 8
ÿ 8
ÿ
µ µpAnk q ď µ pEn q ` 2 ε ď ε` µ˚ pEn q.
˚
` ˚ ´n
˘
En ď
n n“1 k“1 n“1 n“1

The result follows.

74
MATH 362 Lecture Notes 11.2 Outer measure

Definition 11.8: For sets A, B we define the symmetric difference A a B to be the set of all
elements contained in either A or B, but not both. That is,

A a B :“ tx P A Y B | x < A X Bu.

Using this, we define the function d : PpRp q ˆ PpRp q Ñ r0, 8s by

dpA, Bq :“ µ˚ pA a Bq, A, B Ă Rp .

When pAn q is a sequence of sets in E (elementary sets) and lim dpA, An q “ 0, we write
nÑ8
An Ñ A, and say that A is finitely µ-measurable. The collection of finitely µ-measurable sets
is denoted by MF pµq.

If A is the union of a countable collection of sets in MF pµq, we say that A is µ-measurable,

and write A P Mpµq.

Proposition 11.9 (Properties of a, d): For sets A, B, C, A1 , A2 , B1 , and B2 , we have

(i) A a B “ B a A.

(ii) A a A “ ∅.

(iii) A a B Ă pA a Cq Y pC a Bq.

(iv) pA1 a B1 q Y pA2 a B2 q contains ppA1 Y A2 q a pB1 Y B2 qq, ppA1 X A2 q a pB1 X B2 qq, and
ppA1 zA2 q a pB1 zB2 qq.

(v) dpA, Bq “ dpB, Aq.

(vi) dpA, Aq “ 0.

(vii) dpA, Bq ď dpA, Cq ` dpC, Bq.

(viii) dpA1 , B1 q ` dpA2 , B2 q is larger than each of dpA1 Y A2 , B1 Y B2 q, dpA1 X A2 , B1 X B2 q, and

dpA1 zA2 , B1 zB2 q.

Proof:

(i) and (ii) These follow immediately from the definition of a.

(iii) Note that another way to write A a B is pAzBq Y pBzAq. Thus, since

AzB Ă pAzCq Y pCzBq and BzA Ă pBzCq Y pCzAq,

75
MATH 362 Lecture Notes 11.2 Outer measure

it follows that

A a B Ă pAzCq Y pCzAq Y pBzCq Y pCzBq “ pAzCq Y pBzCq.

(iv) First, note that

pA1 Y A2 qzpB1 Y B2 q Ă pA1 zB1 q Y pA2 zB2 q,

from which it follows that

pA1 Y A2 q a pB1 Y B2 q Ă pA1 a B1 q Y pA2 a B2 q.

Next, note that AzB “ Bc zAc , which implies A a B “ Ac a Bc . Thus, by Demorgan’s

Laws, we have

pA1 X A2 q a pB1 X B2 q “ pAc1 Y Ac2 q a pBc1 Y Bc2 q

Ă pAc1 a Bc1 q Y pAc2 a Bc2 q
“ pA1 a B1 q Y pA2 a B2 q.

Finally, since AzB “ A X Bc , we have

pA1 zA2 q a pB1 zB2 q “ pA1 X Ac2 q a pB1 X Bc2 q

Ă pA1 a B1 q Y pAc2 a Bc2 q (by the previous inclusion result)
“ pA1 a B1 q Y pA2 a B2 q.

(v) This follows immediately from part (i).

(vi) This follows from (ii) and part (ii) of Theorem 11.7.
(vii) We have, from part (v) of Theorem 11.7, that µ˚ is σ-subadditive, which means that
¨ ˛
8
ď 8
ÿ
µ˚ ˝ E j ‚ ď µ˚ pE j q.
j“1 j“1

If we fix n and take En`1 “ En`2 “ ¨ ¨ ¨ “ ∅, this expression reduces to

¨ ˛
ďn ÿn
µ˚ ˝ E j ‚ ď µ˚ pE j q,
j“1 j“1

a property we might call subadditivity (or finite subadditivity). Thus, the sigma-
subadditivity of µ˚ implies it is also finitely subadditive. We then have

dpA, Bq “ µ˚ pA a Bq ď µ˚ ppA a Cq Y pC a Bqq (by (iii), and Thm. 11.7 part (iii))
ď µ˚ pA a Cq ` µ˚ pC a Bq (by subadditivity)
“ dpA, Cq ` dpC, Bq.

76
 MATH 362 Lecture Notes 11.2 Outer measure

(viii) By part (iv) and monotonicity of µ˚ (Thm. 11.7 part (iii)), each one of the expres-
sions on the left
,
dpA1 Y A2 , B1 Y B2 q /
.
dpA1 X A2 , B1 X B2 q ď µ˚ ppA1 a B1 q Y pA2 a B2 qq
/
dpA1 zA2 , B1 zB2 q
-

ď µ˚ pA1 a B1 q ` µ˚ pA2 a B2 q “ dpA1 , B1 q ` dpA2 , B2 q.

Proposition 11.10: In the case of Lebesgue outer measure, D distinct sets A, B Ă

Rp s.t. dpA, Bq “ 0.

Proof: Let A “ tx1 , x2 , . . .u be a countable set and B “ ∅. Let ε ą 0. For each n P N

choose an interval In in Rp s.t. xn P In and mpIn q ă 2´n ε. Then
˜ ¸
8
ď ÿ8
˚
dpA, Bq “ m pA a Bq ď m ˚
In ď ε 2´n “ ε.
n“1 n“1

Since dpA, Bq ď ε, @ ε ą 0, dpA, Bq “ 0. 

Note: Were it not for Proposition 11.10, d would be a metric on PpRp q. To make d a metric on something,
we partition up PpRp q into equivalence classes, calling two sets A and B equivalent if dpA, Bq “ 0.
Then d is a metric on the equivalence classes of PpRp q.

Proposition 11.11: If A, B Ă Rp and at least one of µ˚ pAq, µ˚ pBq is finite, then

|µ˚ pAq ´ µ˚ pBq| ď dpA, Bq.

Proof: We have

µ˚ pAq “ dpA, ∅q ď dpA, Bq`dpB, ∅q “ dpA, Bq`µ˚ pBq ñ µ˚ pAq´µ˚ pBq ď dpA, Bq.

Apply a symmetric argument to show that

µ˚ pBq ´ µ˚ pAq ď dpA, Bq,

and get the result. 

77
MATH 362 Lecture Notes 11.2 Outer measure

Theorem 11.12: Mpµq is a σ-algebra, and µ˚ is countably additive on Mpµq.

Proof: We break the proof up into major sub-steps:

I. MF pµq is a ring.
Let A, B P MF pµq. We must show that A Y B, AzB P MF pµq. By definition, D
sequences of elementary sets pAn q, pBn q s.t. An Ñ A and Bn Ñ B. To prove MF pµq
is closed under unions, note that

dpAn Y Bn , A Y Bq ď dpAn , Aq ` dpBn , Bq Ñ 0 as n Ñ 8.

(The inequality comes from Prop. 11.9 (viii).) This shows that An Y Bn Ñ A Y B,
and since each An Y Bn P E , this proves A Y B P MF pµq. Similarly, we have that

dpAn zBn , AzBq ď dpAn , Aq ` dpBn , Bq Ñ 0,

showing that pAn zBn q Ñ pAzBq, and AzB P MF pµq.

II. If A P MF pµq, An P E and An Ñ A, then µ˚ pAn q Ñ µ˚ pAq.
This follows directly from Prop. 11.11, since each An P E implies µ˚ pAn q “ µpAn q ă
8. (Note that this means µ˚ pAq ă 8, @ A P MF pµq.)
III. µ˚ is additive on MF pµq.
Let A, B, pAn q and pBn q be as in the proof of I, but now assume that A X B “ ∅. By
Prop. 11.11 again,

µ˚ pAq ` µ˚ pBq “ lim rµpAn q ` µpBn qs “ lim rµpAn Y Bn q ` µpAn X Bn qs

nÑ8 nÑ8
“ µ pA Y Bq ` µ˚ pA X Bq
˚

“ µ˚ pA Y Bq.

IV. µ˚ is countably additive on MF pµq. Let pAn q be a disjoint sequence in MF pµq. By

Thm. 11.7 (v) (σ-subadditivity),
˜ ¸
8
ď ÿ8
µ ˚
An ď µ˚ pAn q.
n“1 n“1

But, by the additivity (III above)/monotonicity (Thm. 11.7 (iii)) of µ˚ ,

¨ ˛ ˜ ¸
ÿn n
ď ď8
µ˚ pA j q “ µ˚ ˝ A j ‚ ď µ˚ An ,
j“1 j“1 n“1

@ n P N, which means ˜ ¸
8
ÿ 8
ď
µ˚ pA j q ď µ˚ An .
j“1 n“1

78
 MATH 362 Lecture Notes 11.2 Outer measure

V. If A P Mpµq and µ˚ pAq ă 8, then A P MF pµq.

A P Mpµq ñ D a sequence pAn q in MF pµq with A “ Yn An . WLOG, the An are
disjoint. (Otherwise, take A11 “ A1 and A1n “ An zpA1 Y ¨ ¨ ¨ Y An´1 q, n ą 1.) By IV,
8
ÿ
µ˚ pAq “ µ˚ pAn q.
n“1

Suppose µ˚ pAq ă 8. For n P N, set Bn :“ A1 Y ¨ ¨ ¨ Y An .

Claim: Bn Ñ A.
¨ ˛
ď8 ÿ8
dpBn , Aq “ µ˚˝
Aj “
‚ µ˚ pA j q Ñ 0 as n Ñ 8.
j“n`1 j“n`1

By the claim and P57 on Problem Set 20, A P MF pµq.

VI. µ˚ is countably-additive on Mpµq.
Let A “ Yn An with the An disjoint in Mpµq. If some µ˚ pAn q “ 8, then µ˚ pAq “ 8
and the result holds. On the other hand, if each µ˚ pAn q ă 8, then each An P MF pµq
(by V) and the result follows from IV.
VII. Mpµq is a σ-ring (in fact, a σ-algebra, for Rp P Mpµq).
To see that Mpµq is closed under countable unions, let pAn q be a sequence in Mpµq.
By definition, each An is a countable union of sets in MF pµq. But countable unions
Ť
of countable unions results in a countable union, so n An P Mpµq.
Now suppose A, B P Mpµq with
8
ď 8
ď
A “ An , B “ Bn ,
n“1 n“1

and each An , Bn P MF pµq. Notice that

8
ď
An X B “ pAn X B j q,
j“1

with each pAn X B j q P MF pµq (by I). Since MF pµq is contained in Mpµq, which is
closed under countable unions, each pAn X Bq P Mpµq. Moreover, since

µ˚ pAn X Bq ď µ˚ pAn q ă 8 (see II),

each pAn X Bq P MF pµq. Thus, pAn zBq “ An zpAn X Bq P MF pµq, and

8
ď
AzB “ pAn zBq P Mpµq.
n“1

Note:

79
MATH 362 Lecture Notes 11.2 Outer measure

(i) For those A P Mpµq we drop the asterisk notation, denoting the measure of A by µpAq.

(ii) For every µ, the sets of measure zero form a σ-ring. (Note: What is not being said, and not
true, is that the sets of measure zero are the same for each measure µ.)

(iii) For m (Lebesgue measure), every countable set has measure zero. Likewise, mpCq “ 0, even
though the Cantor middle-thirds set is uncountable.

Definition 11.13: A collection tVα u of open subsets of a set X is called a base for X if @ x P X
and every open set G Ă X containing x, D α0 s.t. x P Vα0 and Vα0 Ă G. That is, every open
G is the union of a subcollection of tVα u.

Definition 11.14: A metric space X is said to be separable if it contains a dense subset

which is countable.

Example 13:

The collection of all open balls

tBpx, rq | x P Rp , r ą 0u

is a base of Rp . Since Rp is, in fact, separable (for instance, the set of all points in Rp with
rational components is dense in Rp and countable) this implies that it has a countable base. The
collection of open balls
tBpq, rq | q P Qp , r P Q X p0, 8qu

is such a countable base.

Proposition 11.15: Every open set Ω Ă Rp P Mpµq.

Proof: The proof rests on the fact that Rp has a countable base. Instead of the countable
base presented in the last example (open balls centered at points in Qp with rational
radii) we choose a base B consisting of open p-cubes—intervals in Rp whose side-lengths
are all equal—where such a cube is in B iff it is centered on a point of Qd and has rational

80
MATH 362 Lecture Notes 11.2 Outer measure

side length. (Convince yourself that (1) every open G Ă Rp is the union of sets in B,
and that (2) B is countable.) Since every open set in Rp is thus the union of sets in E ,
the claim follows. 

Theorem 11.16: Suppose A P Mpµq. Then

(i) @ ε ą 0, D an open G Ą A (G P Mpµq, by Prop. 11.15) such that µpGzAq ă ε.

and µ pp
Ş Ş
(ii) D a sequence of open sets pGn q with A Ă n Gn n Gn q zAq “ 0.

(iii) @ ε ą 0, D a closed F Ă A (F P Mpµq since Mpµq is a ring) such that µpAzFq ă ε.

Ă A and µ pAz p n Fn qq “ 0.
Ť Ť
(iv) D a sequence of closed sets pFn q with n Fn

Proof:

(i) We divide the proof into two cases.

Case: µpAq ă 8.
This is true by definition of µpAq :“ µ˚ pAq.
Case: µpAq “ 8.
Let ε ą 0, and set An :“ A X Bp0, nq P Mpµq, @ n P N. Since each An is contained
in an interval, µ˚ pAn q ă 8, @ n. By what we have already proved, D an open
Gn s.t. µpGn zAn q ă 2´n ε. Let G “ n Gn . Since A “ n An ,
Ť Ť

ď 8
ÿ 8
ÿ
pGzAq Ă pGn zAn q and µpGzAq ď µpGn zAn q ď 2´n ε “ ε.
n n“1 n“1

(ii) By (i), @ n P N, D an open Gn Ą A s.t. µpGn zAq ă 1{n. And, since p n Gn q zA Ă

Ş

pGn zAq, @ n, monotonicity of µ gives us that

˜˜ ¸ ¸
č 1
0 ď µ Gn zA ď , @n.
n
n

Thus, µ p n Gn q zA “ 0.
Ş

(iii) This follows directly from (i). For, since A P Mpµq, we have that Ac P Mpµq. Thus,
we may apply (i) to get an open set G Ą Ac s.t. µpGzAc q ă ε. Now take F “ Gc .
(iv) Much as (iii) follows from (i), this assertion follows from (ii).

81
MATH 362 Lecture Notes 11.3 Existence of a non-measurable set

Proposition 11.17: Let X be a nonempty set, C Ă PpXq. Then D a smallest σ-algebra

containing C (in the sense that all other σ-algebras containing C contain this one.)

Proof: The proof is HW.

Hint: Consider the set

A :“ tL | L is a σ-algebra and C Ă Lu.

Show that A is nonempty. Then consider the set

č
Σ :“ L.
LPA

Show that Σ is a σ-algebra, that C Ă Σ, and that Σ is contained in every other σ-algebra
that contains C. 

Definition 11.18: Take C in Prop. 11.17 to be the collection of all open sets in X. The smallest
σ-algebra containing C is called the Borel algebra, and is denoted BpXq, or simply B when
the underlying space (for us, usually Rp ) is understood. The elements of B are called Borel
sets.

11.3 Existence of a non-measurable set

Question: We have three σ-algebras of subsets of Rp we have been discussing: B Ă Mpµq Ă

PpRp q. Are they one and the same?

82
MATH 362 Lecture Notes 11.3 Existence of a non-measurable set

Answer: If we accept the Axiom of Choice:

Axiom of Choice: If E is any collection of nonempty sets, then D a choice

function C : E Ñ S such that, @ E P E, S X E is a single element.

then Mpµq , PpRp q.

Most mathematicians do accept the Axiom of Choice. It is needed to prove things
like

a. The cardinality of any two sets is comparable—i.e., one set has cardinality less
than or equal to that of the other set; one set is in one-to-one correspondence
with some subset of the other.

b. Any vector space over a field F has a basis.

c. Any product of compact topological spaces is compact. (This is now known as

Tychonoff’s Theorem.)

But some mathematicians do not accept it, primarily because, despite its seemingly-
innocuous assertion, it implies some bizarre-sounding statements. These include

a. The Well-Ordering Property (equivalent to AC): If E , ∅, then D a relation ď

on E which is a linear order and such that, if S Ă E, then S contains a smallest
element.

b. Zorn’s Lemma (equivalent to AC): Every partially ordered set in which every
chain (i.e. totally ordered subset) has an upper bound contains at least one
maximal element.

c. The Banach-Tarski Paradox: It is possible to take the 3-dimensional closed

unit ball B, partition it into finitely many pieces, and move those pieces in rigid
motions (i.e., rotations and translations, with pieces permitted to move through
one another) and reassemble them to form two copies of B.

Proposition 11.19: Let A Ă Rp . TFAE:

(i) A P Mpµq.

(ii) @ ε ą 0, D an open G Ą A (G P Mpµq, by Prop. 11.15) such that µ˚ pGzAq ă ε.

(iii) D a sequence of open sets pGn q with A Ă n Gn and µ˚ pp n Gn q zAq “ 0.

Ş Ş

(iv) @ ε ą 0, D a closed F Ă A (F P Mpµq since Mpµq is a ring) such that µ˚ pAzFq ă ε.

(v) D a sequence of closed sets pFn q with n Fn Ă A and µ˚ pAz p n Fn qq “ 0.

Ť Ť

83
MATH 362 Lecture Notes 11.3 Existence of a non-measurable set

Proof: This proof is HW.

Hint: First show that any set whose outer measure is zero must be µ-measurable. Also,
take another look at the proof of Thm. 11.16 to see what parts of the current statement
have not yet been proved. 

Definition 11.20: Let A, B Ă Rp . We denote by A ` B the set

A ` B :“ ta ` b | a P A, b P Bu.

When B “ tbu (a singleton), we are often sloppy and write A ` b in place of A ` tbu.

Lemma 11.21: If E Ă Rp , then m˚ pE ` cq “ m˚ pEq, @ c P Rp . If E P Mpmq, then so is E ` c.

Proof: Let pRn q be a sequence of intervals in Rp s.t. E Ă n Rn . Clearly E ` c Ă

Ť
Ť ˚ ˚
n pRn ` cq, and mpRn ` cq “ mpRn q, @ n. Thus, m pE ` cq and m pEq are infimums taken
over the same set of numbers, making them equal.

Now, suppose E P Mpmq. By Thm. 11.16 (i), @ ε ą 0, D an open G Ă Rp s.t. E Ă G and

m˚ pGzEq ă ε.
Claim: pGzEq ` c “ pG ` cqzpE ` cq.

x P pGzEq ` c ô x ´ c P GzE
ô x ´ c P G and x ´ c < E
ô x P G ` c and x < E ` c
ô x P pG ` cqzpE ` cq.

Clearly pE ` cq Ă pG ` cq, G ` c is open, and

m˚ ppG ` cqzpE ` cqq “ m˚ ppGzEq ` cq “ m˚ pGzEq ď ε.

So, E ` c P Mpmq (Prop. 11.19 (ii) implies (i)). 

Definition 11.22: A measure µ defined for a σ-algebra of subsets of Rp which has the
property asserted in Prop. 11.21 (there, asserted for m, the measure arising from restricting
the Lebesgue outer measure m˚ to Mpmq) is said to be translation invariant.

84
MATH 362 Lecture Notes 11.3 Existence of a non-measurable set

Theorem 11.23: Assuming AC, there exists a non-measurable subset of p0, 1q (i.e., an S Ă
p0, 1q for which S < Mpmq.

Proof: For each x, y P p0, 1q, define a relation „ in the following way: say x „ y if
x ´ y P Q. Then „ is an equivalence relation (verify) and, as such, it partitions p0, 1q
into equivalence classes. Write rxs :“ ty | y „ xu, and let E :“ trxs | 0 ă x ă 1u. Then
rx1 s , rx2 s ñ rx1 s X rx2 s “ ∅. We apply the choice function (from AC) to E, and let
S be the resulting set; that is, S contains exactly one member of each equivalence class
rxs. Let pqn q be an enumeration of Q X p´1, 1q. For each i P N, let Si :“ S ` qi .

Claim 1: Si X S j “ ∅ whenever i , j.
Suppose x P Si X S j . Then x “ ti ` qi “ t j ` q j for some ti , t j P S.

ñ t i ´ t j “ q j ´ qi P Q
ñ ti „ t j
ñ ti “ t j (from our application of the choice function)
ñ qi “ q j
ñ i “ j.

This proves Claim 1.

Ť
Claim 2: p0, 1q Ă i Si Ă p´1, 2q.
Ť
That i Si Ă p´1, 2q is obvious. Let x P p0, 1q. Then x P rxs.

ñ rxs X S “ ttu, for some t „ x

ñ x ´ t “ qi , for some i
ñ x P S ` qi “ S i ,

which proves the other inclusion in Claim 2.

Now, suppose S P Mpmq. By Prop. 11.21, Si P Mpmq and mpSi q “ mpSq, @ i. By Claim 2,
1 ď m p i Si q ď 3. Since m “ m˚ |M is σ-additive,
Ť

˜ ¸
8
ď ÿ8 8
ÿ
m Si “ mpSi q “ mpSq 1.
i“1 i“1 i“1
`Ť8 ˘ `Ť8 ˘
So, if mpSq ą 0 then m i“1 Si “ 8, and if mpSq “ 0, then m i“1 Si “ 0. ÑÐ 

Corollary 11.24: Assuming the AC, there exist disjoint E1 , E2 Ă R s.t. m˚ pE1 Y E2 q ă
m˚ pE1 q ` m˚ pE2 q.

85
 MATH 362 Lecture Notes 11.4 Lebesgue integration

Proof: Suppose no such disjoint E1 , E2 exist. Then m˚ pE1 YE2 q “ m˚ pE1 q`m˚ pE2 q, @ E1 ,
E2 Ă R with E1 X E2 “ ∅. Using induction, m˚ is finitely additive on the subsets of R.
Since m˚ is an outer measure, it is monotone and σ-subadditive (Thm. 11.7 (iii) and (v)).
Thus, m˚ is σ-additive on PpRq. (The argument used to establish IV in Thm. 11.12 goes
through perfectly well here, assuming that the disjoint pAn q come from PpRq instead
of MF pmq.)
Ť
Taking S, Si as in Thm. 11.23, we have p0, 1q Ă i Si Ă p´1, 2q. But

m˚ pp0, 1qq “ 1
˜ ¸
8
ď
ď m˚ Si (by monotonicity)
i“1
8
ÿ
“ m˚ pSi q (by σ-additivity)
i“1
8
ÿ
“ m˚ pSq 1 (Prop. 11.21)
i“1
ď m˚ pp´1, 2qq
“ 3. ÑÐ

Note: There is no countably-additive, nonnegative, translation-invariant set function φ defined for

Ppr0, 1sq for which φp∅q “ 0 and φpr0, 1sq “ 1.

11.4 Lebesgue integration

Definition 11.25: Let X , ∅ be a set. The combination pX, Mq is called a measurable space if
M is a σ-algebra consisting of subsets of X.

pX, M, µq is called a measure space if pX, Mq is a measurable space and µ is a nonnegative,

countably additive set function (called a measure) on M.

Definition 11.26: Let f : X Ñ r´8, 8s, where pX, Mq is a measurable space. The function
f is called measurable if the set tx | f pxq ą au P M, @ a P R.

Note:

86
MATH 362 Lecture Notes 11.4 Lebesgue integration

(i) There is no reference to a measure in the definition of measurable functions. Whether a

function is measurable or not depends only upon the σ-algebra (i.e., on whether certain sets
are measurable or not).

(ii) If X in Defn. 11.26 is a metric space, then every continuous f : X Ñ r´8, 8s is measurable.
If M “ B (i.e., if our σ-algebra is the Borel algebra), then a measurable f is called Borel
measurable. So, every continuous f is Borel measurable.

Theorem 11.27: TFAE:

(i) tx | f pxq ą au P M, @ a P R.

(ii) tx | f pxq ě au P M, @ a P R.

(iii) tx | f pxq ă au P M, @ a P R.

(iv) tx | f pxq ď au P M, @ a P R.

8 " *
č 1
Proof: tx | f pxq ě au “ x | f pxq ą a ´ ,
n“1
n
so (i) implies (ii), since this is a countable intersection of measurable sets. Likewise,

tx | f pxq ă au “ tx | f pxq ě auc ,

so (ii) implies (iii), since M is closed under complementation. Next

8 " *
č 1
tx | f pxq ď au “ x | f pxq ă a ` ,
n“1
n

showing that (iii) implies (iv). Finally, (iv) implies (i), since

tx | f pxq ą au “ tx | f pxq ď auc ,

Theorem 11.28: If f is measurable, then | f | is measurable as well.

Proof: This result is stated in the context of a measurable space pX, Mq, with f : X Ñ
r´8, 8s assumed to be measurable in the sense of Definition 11.26. Let a P R be fixed.
We have
ˇ ˇ
tx : | f pxq| ă au “ x ˇ f pxq ă a X x ˇ f pxq ą ´a .
( (

87
MATH 362 Lecture Notes 11.4 Lebesgue integration

ˇ ( ˇ (
Since f is measurable, the sets x ˇ f pxq ă a and x ˇ f pxq ą ´a are in M. Since rings
are closed under intersections, the set tx : | f pxq| ă au is in M as well. The choice of a is
arbitrary, so by Theorem 11.27, | f | is measurable. 

Theorem 11.29: Suppose p fn q is a sequence of measurable functions. For x P X, n P N, set

gpxq :“ sup fn pxq and hpxq :“ lim sup fn pxq.

n nÑ8

Then g and h are measurable. Similar statements may be made about inf fn pxq and
n
lim inf fn pxq.

Proof: Note that

8
ď
tx | gpxq ą au “ tx | fn pxq ą au,
n“1
while hpxq “ inf gm pxq, where gm pxq :“ sup fn pxq. 
m něm

Corollary 11.30: (i) If f , g are measurable, then maxt f, gu and mint f, gu are measurable.
In particular, the functions defined by

f ` :“ maxt f, 0u and f ´ :“ ´ mint f, 0u

are measurable if f is.

(ii) The (pointwise) limit of a convergent sequence of measurable functions is measurable.

Proof:

(i) Take f1 “ f and fn “ g for n “ 2, 3, . . . in Thm. 11.29. Then mint f, gu “ inf fn and
maxt f, gu “ sup fn .
(ii) Under the assumption of convergence, lim sup fn “ lim fn .

Theorem 11.31: Let f , g be measurable real-valued functions defined on X, let F be real

and continuous on R2 , and set

hpxq :“ Fp f pxq, gpxqq, x P X.

Then h is measurable. In particular, f ` g and f g are both measurable if f and g are.

88
 MATH 362 Lecture Notes 11.4 Lebesgue integration

Proof: Define Ga :“ tpu, vq | Fpu, vq ą au. Since F is continuous, Ga Ă R2 is an open set

@ a P R. Since R2 has a countable base of open rectangles, we may write
8
ď
Ga “ In , where In “ tpu, vq | an ă u ă bn , cn ă v ă dn u, n “ 1, 2, . . . .
n“1

Now sets of the form

tx | an ă f pxq ă bn u “ tx | f pxq ą an u X tx | f pxq ă bn u

are measurable, and so is

tx | p f pxq, gpxqq P In u “ tx | an ă f pxq ă bn u X tx | cn ă gpxq ă dn u.

Thus,
8
ď
tx | hpxq ą au “ tx | p f pxq, gpxqq P Ga u “ tx | p f pxq, gpxqq P In u.
n“1


Definition 11.32: Suppose pX, Mq is a measurable space and s : X Ñ R. We say that s is a

simple function if ranpsq contains finitely many (distinct) elements.

Let A Ă X. We define the characteristic function χA to be the simple function

#
1, if x P E,
χA pxq :“
0, otherwise.

If s is a simple function, it may be written as a sum

n
ÿ
s “ yi χAi , (8)
i“1

where ranpsq “ ty1 , y2 , . . . , yn u. This decomposition of s is generally not unique. However,

it is unique if we require yi , y j for i , j. When this is the case, then we call (8) the canonical
decomposition of s.

Note:

ˇ (
1. In the canonical decomposition (8) of simple s, one has yi , y j for i , j, Ai “ x ˇ spxq “ yi “
s´1 ptyi uq, and Ai XA j “ ∅ for i , j. Unless otherwise mentioned, you may assume expressions
used in the following of the form (8) are canonical.

89
MATH 362 Lecture Notes 11.4 Lebesgue integration

2. One may verify that a simple s expressed in canonical form (8) is measurable if and only if
Ai P M for i “ 1, 2, . . . , m.

3. If we take as our measure space pR, Mpmqq (the reals with the Lebesgue measurable sets
Mpmq) and define s “ χS , where S is the nonmeasurable set of Theorem 11.23, then s is a
simple function which is not measurable.

Theorem 11.33: Let pX, Mq be a measurable space, and f : X Ñ r´8, 8s. There exists a
ptwise
sequence psn q of simple functions such that sn ÝÑ f as n Ñ 8. If f is measurable, then
the simple functions sn may be chosen to all be measurable as well. If f ě 0, then psn q may
be chosen to be a monotone increasing sequence.

Proof: First, suppose ranp f q Ă r0, 8s, and for each n “ 0, 1, 2, 3, . . . and each k “
1, 2, . . . , n2n ` 1, define sets
" * ˆ„ ˙˙ ˆ„ n ˙
k´1 k k´1 k n2 ` 1
Enk :“ x | n ď f pxq ă n “ f ´1
, , and Fn :“ f ´1
, `8 .
2 2 2n 2n 2n

For each n “ 0, 1, 2, . . . let

k´1
$ ,
’
’ , x P Enk , /
/ n
& 2n
’ /
. n2n ` 1
n2ÿ`1
k´1
sn pxq :“ “ n
χ Fn ` χEnk .
n 2 2n
% n2 ` 1 , x P Fn
’ /
’
’ /
/
- k“1
2n
Clearly sn is simple (has finitely many values, all of which are finite).

Claim 1: f measurable implies sn measurable.

rpk ´ 1q{2n , k{2n q “ r´8, 8szpr´8, pk ´ 1q{2n q Y rk{2n , `8sq, so Enk “ f ´1 prpk ´
1q{2n , k{2n qq P M. (Verify the measurability of Enk .) Similarly, Fn P M. Thus,
s´1 ppc, `8qq is either ∅, or is comprised of a finite union of measurable sets, and
s is measurable by definition of measurable functions. By construction, notice that
sn pxq ď f pxq, @ x.

Claim 2: sn pxq ď sn`1 pxq, @ x P X. ˜ n `1

n2ď
¸
Fix n P t0, 1, 2, . . .u and let x P X “ Fn Y Enk . By construction, this union is over
k“1
pairwise disjoint sets. We consider two cases:

I. x is in Enk for some k “ 1, . . . , n2n ` 1.

90
MATH 362 Lecture Notes 11.4 Lebesgue integration

Let k be fixed so that x P Enk . We have

" * " *
ˇk´1 k ˇ 2k ´ 2 2k
Enk “ x ˇ n ď f pxq ă n “ x ˇ n`1 ď f pxq ă n`1
2 2 2 2
" * " *
ˇ 2k ´ 2 2k ´ 1 ˇ 2k ´ 2 2k
“ x ˇ ď f pxq ă n`1 Y x ˇ ď f pxq ă n`1
2n`1 2 2n`1 2
“ En`1,2k´1 Y En`1,2k .

Note that, since k P 1, 2, . . . , n2n ` 1, we have 2k P t1, 2, . . . , pn ` 1q2n`1 ` 1u, and

so En`1,2k is a set we have defined. Thus,
$ ,
2k ´ 2
& n`1 , if x P En`1,2k´1 , /
’
’ /
2
. k´1
sn`1 pxq “ ě “ sn pxq .
2k ´ 1 2n
, if x P En`1,2k
’
’ /
/
% -
2n`1

II. x is in Fn .
Note that
n2n ` 1 n2n`1 ` 2
" * " *
ˇ ˇ
Fn “ x f pxq ě
ˇ “ x f pxq ě
ˇ
2n 2n`1
# + ¨pn`1q2n`1 `1 " ˛
pn ` 1q2 n`1 `1 ˇm´1
*
ˇ ď m
“ x ˇ f pxq ě n`1
Y˝ x ˇ n`1 ď f pxq ă n`1 ‚
2 2 2
m“n2n`1 `3
¨ n`1
˛
pn`1q2
ď `1
“ Fn`1 Y ˝ En`1,m ‚ , (9)
m“n2n`1 `3

where the sets in (9) are among those used to define sn`1 ; on these sets we have
$
n2n`1 ` 2
, x P En`1,n2n`1 `3 ,
’
’
’
’
’ 2n`1
..
’
’
’
.
’
’
&
sn`1 pxq “ pn ` 1q2n`1
’
’ , x P En`1,pn`1q2n`1 `1 ,
’
’
’
’ 2n`1
pn ` 1q2n`1 ` 1
’
’
, x P Fn`1 .
’
’
%
2n`1
Thus, for x P Fn ,

n2n`1 ` 2 n2n ` 1
sn`1 pxq ě “ “ sn pxq .
2n`1 2n
Thus, Claim 2 is proved.

Now, if f pxq ă 8, then D N P N s.t. n ě N ñ x P Enk for some k. Thus,

1
| f pxq ´ sn pxq| ă Ñ 0 as n Ñ 8.
2n

91
 MATH 362 Lecture Notes 11.4 Lebesgue integration

On the other hand, if f pxq “ `8, then x < Enk @ n and k. Thus, for each n,
n2n ` 1
sn pxq “ ą n Ñ 8 as n Ñ 8.
2n
To complete the proof (for f as general as stated in the theorem), we note that we
have proved the existence of a sequences of simple functions psn q converging to f `
(measurable if f is measurable), and ptn q converging to f ´ . The sequence psn ´ tn q is
made up of simple functions, too, and converges to f “ f ` ´ f ´ . 

Definition 11.34: Let pX, M, µq be a measure space. Suppose s : X Ñ r0, 8q is a measurable

simple function; that is,
n
ÿ
spxq :“ ci χEi pxq, with each Ei P M.
i“1

For any A P M, we define an integral for s over A as

n
ÿ
IA psq :“ ci µpA X Ei q.
i“1

If f : X Ñ r0, 8s is measurable, we define

ż
f dµ :“ sup tIA psq | s is a measurable, simple function with 0 ď spxq ď f pxq, @ x P Xu .
A

Note:

1. It is easy enough to see that

IA psq “ suptIA ptq | 0 ď t ď su,

which shows that ż

s dµ “ IA psq.
A
ş
This means ¨ dµ is an extension of IA p¨q, defined for a larger class of functions, just as the
outer measure µ˚ was an extension of µ, defined for a larger class of sets.

2. There is the possibility in Defn. 11.34 of encountering expressions such as 0 ¨ 8, when ci “ 0

and µpA X Ei q “ 8. We follow the convention that this expression equals 0. Quoting Rudin
(from Real & Complex Analysis, p. 18),

It may seem strange to define 0¨8 “ 0. However, one verifies without difficulty
that with this definition the commutative, associative and distributive laws hold in
r0, 8s without restriction.

92
MATH 362 Lecture Notes 11.4 Lebesgue integration

The cancellation laws have to be treated with some care: a ` b “ a ` c implies

b “ c when a ă 8, and ab “ ac implies b “ c only when 0 ă a ă 8.
Observe that the following useful proposition holds: If 0 ď a1 ď a2 ď ¨ ¨ ¨ , 0 ď
b1 ď b2 ď ¨ ¨ ¨ , an Ñ a, and bn Ñ b, then an bn Ñ ab.

3. For measurable f : X Ñ r0, 8s, the integral in Defn. 11.34 is called the Lebesgue integral of f
ş
over A. This number exists for all such f , though it is entirely possible that A f dµ “ 8.

Theorem 11.35: Suppose pX, M, µq is a measure space, f, g : X Ñ r0, 8s, A, B P M, and

c P r0, 8q. Then
ż ż
(i) If f ď g, then f dµ ď g dµ.
A A
ż ż
(ii) If A Ă B, then f dµ ď f dµ.
A B
ż ż
(iii) c f dµ “ c f dµ.
A A
ż
(iv) If f pxq “ 0, @ x P A, then f dµ “ 0, even if µpAq “ 8.
A
ż
(v) If µpAq “ 0, then f dµ “ 0, even if f pxq “ 8, @ x P A.
A
ż ż
(vi) f dµ “ χA f dµ.
A X

Proof:
ş
(i) The set of numbers over which A f dµ is the supremum is contained in the set
ş
over which A g dµ is the supremum.
(ii) If s is a simple function with 0 ď s ď f , then by the monotonicity of µ, IA psq ď IB psq.
ş
The result follows from the definition of A f dµ.
(iii) For c “ 0, the result follows from the fact that only s ” 0 is simple between zero
and c f . For c , 0 we note that cs is simple for each simple s, and that IA pcsq “ cIA psq.

93
MATH 362 Lecture Notes 11.4 Lebesgue integration

Thus,
ż
c f dµ “ c suptIA psq | s simple, 0 ď s ď f u
A
“ suptcIA psq | s simple, 0 ď s ď f u
“ suptIA pcsq | s simple, 0 ď s ď f u
“ suptIA ptq | t simple, 0 ď t ď c f u
ż
“ c f dµ.
A

(iv) Let s be simple with 0 ď s ď f . Let ranpsq “ ty1 , . . . , yn u where the yi are distinct,
and Ei :“ tx | spxq “ yi u. If A “ ∅, then
n
ÿ
IA psq “ yi µp∅q “ 0.
i“1

If A , ∅, then 0 P ranpsq, and WLOG we take y1 “ 0. Since f pxq “ 0 for x P A and

0 ď s ď f , it follows that A Ă E1 . Thus,

IA psq “ 0 ¨ µpA X E1 q “ 0 ¨ µpAq “ 0.

ş
(See Note 2 in ??.) Since IA psq “ 0 for each simple s satisfying 0 ď s ď f , A f dµ “ 0.
(v) Simple functions, by definition, have ranges in R. Therefore, it is easily seen from
the definition of IA psq (with s ě 0) that IA psq “ 0 whenever µpAq “ 0. Thus, the
ş
supremum in Defn. 11.34 that defines A f dµ is zero.
(vi) Notice that χAXB “ χA ¨ χB . Thus,
ż n
ÿ
f dµ “ sup IA psq “ sup yi µpA X Ei q
A yi χEi ď f i“1
ř
0ďsď f 0ď
˜ ¸ ˜ ¸
n
ÿ n
ÿ
“ sup IA yi χAXEi “ sup IA yi χEi ¨ χA
yi χEi ď f yi χEi ď f
ř ř
0ď i“1 0ď i“1
˜ ¸
n
ÿ
“ sup IA χA ¨ yi χEi “ sup IA pχA ¨ sq
yi χEi ď f
ř
0ď i“1 0ďsď f

“ sup IX pχA ¨ sq “ sup IX psq .

0ďsď f 0ďsď f χA

Proposition 11.36: Let pX, M, µq be a measure space, and s, t be nonnegative measurable

simple functions on X. For E P M, define
ż
φpEq “ s dµ.
E

94
MATH 362 Lecture Notes 11.4 Lebesgue integration

Then φ is countably additive on M (i.e., φ is a measure on M). Moreover,

ż ż ż
ps ` tq dµ “ s dµ ` t dµ.
X X X

Proof: Assume that

n
ÿ m
ÿ
s “ yi χAi , and t “ z j χB j ,
i“1 j“1

with the yi distinct and Ai “ tx | spxq “ yi u, and similarly for z j , B j . We first prove the
countable additivity of φ. Let pEk q be a sequence of disjoint sets in M, with E “ k Ek .
Ť

Then
ÿn ÿn 8
ÿ
φpEq “ yi µpE X Ai q “ yi µpEk X Ai q (by σ-additivity of µ)
i“1 i“1 k“1

ÿ n
8 ÿ 8
ÿ
“ yi µpEk X Ai q “ φpEk q.
k“1 i“1 k“1

Moreover, φp∅q “ 0, by Theorem 11.35 (v), since µp∅q “ 0.

To prove the other assertion, for each i “ 1, . . . , n, j “ 1, . . . , m, let Ei j “ Ai X B j . Then

ż
ps ` tq dµ “ pyi ` z j q µpEi j q,
Eij

while ż ż
s dµ ` t dµ “ yi µpEi j q ` z j µpEi j q,
Eij Eij

which shows that ż ż ż

ps ` tq dµ “ s dµ ` t dµ.
Eij Eij Eij

Since X is the disjoint union of the sets Ei j (over i “ 1, . . . , n, j “ 1, . . . , m), and ps ` tq is

a simple function, the first half of our proposition (and the fact that σ-additivity along
with φp∅q “ 0 implies additivity) yield the result. 

Proposition 11.37 (Lebesgue’s Monotone Convergence theorem): Let pX, M, µq be a mea-

sure space, and p fn q be a sequence of measurable functions on X. Suppose that
0 ď f1 pxq ď f2 pxq ď ¨ ¨ ¨ ď 8, @ x P X, and take f pxq :“ limn fn pxq, the pointwise limit
function. (Note that f pxq is a real number when p fn pxqq8
n“1
is a bounded sequence, and is
8
`8 when p fn pxqqn“1 is unbounded.) Then f is measurable, and
ż ż
fn dµ Ñ f dµ, as n Ñ 8.
X X

95
 MATH 362 Lecture Notes 11.4 Lebesgue integration

Proof: Note that f is measurable by Coro. 11.30(ii). Next, Thm. 11.35(i) and the
supposition that fn pxq ď fn`1 pxq at each x P X together imply that the sequence of
˘8
numbers X fn dµ n“1 is monotone increasing, and so approaches some limit α P r0, 8s.
`ş
ş ş
Moreover, fn pxq ď f pxq for each x, implying that X fn dµ ď X f dµ for each n. Thus,
α ď X f dµ.
ş

To get the other inequality, let s be simple with 0 ď s ď f , c P p0, 1q, and define

En :“ tx | fn pxq ě cspxqu, n “ 1, 2, . . . .
Ť
Note that each En P M, E1 Ă E2 Ă E3 Ă ¨ ¨ ¨ , and X “ n En . Also,
ż ż ż
fn dµ ě fn dµ ě IEn pcsq “ c s dµ, (10)
X En En

n “ 1, 2, . . .. But, by Prop. 11.36 and Thm. 11.3,

ż ż
s dµ Ñ s dµ as n Ñ 8.
En X

So, taking the limit as n Ñ 8 of both sides of (10), we get

ż
α ě c s dµ.
X

The choice of c P p0, 1q is arbitrary, so α ě IX psq. Since true for each simple s with
0 ď s ď f , it follows that α ě X f dµ.
ş


Note: By Thm. 11.35 (vi), the above theorem is equally true if the region X over which we are integrating
is replaced by E P M. That is, we have the following corollary (which some texts call Lebesgue’s
Monotone Convergence Theorem):

Corollary 11.38: Let pX, M, µq be a measure space with E P M. Let p fn q be a sequence of

measurable functions on X such that

• 0 ď f1 pxq ď f2 pxq ď ¨ ¨ ¨ ď 8, @ x P E, and

ptwise
• fn pxq ÝÑ f pxq as n Ñ 8, @ x P E.

Then f is measurable, and

ż ż
fn dµ Ñ f dµ, as n Ñ 8.
E E

96
MATH 362 Lecture Notes 11.4 Lebesgue integration

There is a similar ability to replace, in the next result and many others which follow it, integrals
over the entire measure space X with integrals over a more limited region E P M. We will not
routinely point out instances in which similar replacements can be made, but leave it to the reader
to check for specific theorems of interest.

Proposition 11.39: Let pX, M, µq be a measure space, and suppose f , g are functions from
X into r0, 8s. Then ż ż ż
p f ` gq dµ “ f dµ ` g dµ.
X X X

Proof: Let psn q, ptn q be sequences of nonnegative measurable simple functions con-
verging pointwise, respectively, to f and g, and such that 0 ď sn ď sn`1 ď f ,
0 ď tn ď tn`1 ď g. For x P X define rn pxq “ sn pxq ` tn pxq. Then the sequence of
functions rn converges pointwise to p f ` gq, with 0 ď sn ` tn ď sn`1 ` tn`1 ď f ` g. We
have
ż ż ż
p f ` gq dµ “ lim rn dµ “ lim rn dµ (by Prop. 11.37)
X X n n X
ż ˆż ż ˙
“ lim psn ` tn q dµ “ lim sn dµ ` tn dµ (by Prop. 11.36)
n X n X X
ż ż
“ lim sn dµ ` lim tn dµ (again by the Monotone Convergence Thm.)
n X n
żX ż
“ f dµ ` g dµ.
X X


Using induction with this proposition serving as the base case, we can deduce that for measurable
functions f1 , f2 , . . . , fn (finitely many of them) mapping X into r0, 8s,
¨ ˛
ż ÿn n ż
ÿ
˝ f j dµ “
‚ f j dµ. (11)
X j“1 j“1 X

The following theorem extends this result to countable sums of nonnegative measurable functions.

Theorem 11.40: Suppose pX, M, µq is a measure space, and fn : X Ñ r0, 8s is measurable

for n “ 1, 2, 3, . . .. Let
ÿ8
f pxq “ fn pxq, for x P X.
n“1
Then ż 8 ż
ÿ
f dµ “ fn dµ.
X n“1 X

97
MATH 362 Lecture Notes 11.4 Lebesgue integration

Proof: Let Fn “ f1 ` ¨ ¨ ¨ ` fn . Since each fn ě 0, we have Fn ě 0. Moreover,

ptwise ř8
Fn ď Fn`1 , @ n, and Fn ÝÑ n“1 fn . Thus, the result follows from the application of
the Monotone Convergence Theorem (Thm. 11.37) to Equation (11). 

Corollary 11.41: If ai j ě 0 for each i, j “ 1, 2, 3, . . ., then

ÿ 8
8 ÿ 8 ÿ
ÿ 8
ai j “ ai j .
i“1 j“1 j“1 i“1

Proof: To apply the Theorem 11.40, we take X “ N, M “ PpNq, and µ to be counting

measure. That is,
#
# of elements in A, if A is a finite set,
µpAq :“
`8 if A is an infinite set.
One should verify that µ is σ-additive on PpNq, that functions on X into r0, 8s are
nonnegative sequences, and that integrals over such functions with respect to µ are
infinite sums (series). 

Theorem 11.42 (Fatou’s lemma): Suppose pX, M, µq is a measure space, and fn : X Ñ r0, 8s
is measurable for each n P N. Then
ż ´ ¯ ż
lim inf fn dµ ď lim inf fn dµ.
X nÑ8 nÑ8 X

Proof: For each x P X, k P N, define gk pxq :“ infiěk fi pxq. Observe that

• each gk is measurable,
• 0 ď g1 ď g2 ď ¨ ¨ ¨ ,
• gk pxq Ñ lim inf f pxq as k Ñ 8, and
nÑ8
ż ż
• gk ď fk , so gk dµ ď fk dµ.
X X

By the Monotone Convergence Theorem (Thm. 11.37),

ż ż ´ ¯ ż
gk dµ Ñ lim inf fn dµ ď lim inf fn dµ.
X X nÑ8 nÑ8 X


98
MATH 362 Lecture Notes 11.4 Lebesgue integration

Theorem 11.43: Suppose pX, M, µq is a measure space, f : X Ñ r0, 8s is measurable, and

ż
φpEq :“ f dµ, for E P M.
E

Then φ is a measure on M, and

ż ż
g dφ “ g f dµ, for any measurable g : X Ñ r0, 8s.
X X

Ť
Proof: Let pEn q be a disjoint sequence of sets in M, and let E “ n En . Then
ż ż ˜ÿ 8
¸
ÿ8 ż ÿ8
φpEq “ f χE dµ “ f χEn dµ “ f χEn dµ “ φpEn q,
X X n“1 n“1 X n“1

where the third equality holds by Thm. 11.40.

Having established that φ is a measure, suppose

m
ÿ
s “ y j χA j
j“1

is measurable and simple, then

¨ ˛
ż m
ÿ m
ÿ ż ż m
ÿ ż
s dφ “ y j φpA j q “ yj f χA j dµ “ f˝ y j χA j ‚ dµ “ s f dµ.
X j“1 j“1 X X j“1 X

(12)
For general measurable g : X Ñ r0, 8s, we apply Thm. 11.33 to get a sequence of
ptwise
measurable simple functions psn q with 0 ď s1 ď s2 ď ¨ ¨ ¨ ď sn ÝÑ g. We note that
ptwise
0 ď s1 f ď s2 f ď ¨ ¨ ¨ ď sn f ÝÑ g f , and hence the Monotone Convergence Theorem
(Thm. 11.37) applied to both ends of (12) gives that
ż ż
g dφ “ g f dµ.
X X


Definition 11.44: Let pX, M, µq be a measure space. We say that a measurable function
f : X Ñ r´8, 8s is Lebesgue integrable (or summable) with respect to µ if
ż
| f | dµ ă 8.
X

The collection of all Lebesgue integrable functions is denoted L pµq, or L 1 pµq. We some-
times write just L in place of L pmq.

99
 MATH 362 Lecture Notes 11.4 Lebesgue integration

Definition 11.45: Let pX, M, µq be a measure space, f : X Ñ r´8, 8s is measurable. Defin-

ing f ` and f ´ as in 11.30, we consider the separate integrals
ż ż
`
f dµ and f ´ dµ.
X X

If at least one of these two integrals is finite, we define

ż ż ż
`
f dµ :“ f dµ ´ f ´ dµ. (13)
X X X

Note: If both of the integrals on the right-hand side in (13) are finite, then
ż ż ż
`
| f | dµ “ f dµ ` f ´ dµ,
X X X

by Thm. 11.40, and f P L pµq. On the other hand, if f P L pµq, then the inequalities f ` ď | f | and
f ´ ď | f | show that each of f ` , f ´ P L pµq, by Thm. 11.35 (i).

Theorem 11.46: Suppose pX, M, µq is a measure space, and f , g : X Ñ r´8, 8s. If α, β P R

and f , g P L pµq, then α f ` βg P L pµq, with
ż ż ż
pα f ` βgq dµ “ α f dµ ` β g dµ.
X X X

Proof: That pα f ` βgq is measurable follows from Thm. 11.31. To see that it is in L pµq,
we have
ż ż ż ż
|α f ` βg| dµ ď p|α|| f | ` |β||g|q dµ “ |α| | f | dµ ` |β| |g| dµ ď 8,
X X X X

using Thm. 11.35 and Prop. 11.39, and the assumption that f , g P L pµq.
ż ż ż
Claim 1: p f ` gq dµ “ f dµ ` g dµ.
X X X

Let h :“ f ` g, and note that

h` ´ h´ “ h “ f ` g “ f ` ´ f ´ ` g` ´ g´ ,

or
h` ` f ´ ` g´ “ f ` ` g` ` h´ .

100
MATH 362 Lecture Notes 11.4 Lebesgue integration

By Thm. 11.40,
ż ż ż ż ż ż
h` ` f´ ` g´ “ f` ` g` ` h´ ,
X X X X X X

and since each of these integrals is finite, we may rewrite this expression as
ż ż ż ż ż ż
h` ´ h´ “ f` ´ f´ ` g` ´ g´ .
X X X X X X

The claim follows from Defn. 11.45.

ż ż
Claim 2: α f dµ “ α f dµ.
X X

The case where α “ 0 holds trivially. Next, we assume α ą 0. Then

ż ż ż „ż ż  ż
α f dµ “ pα f q dµ ´ pα f q dµ “ α
` ´ `
f dµ ´ f dµ “ α
´
f dµ.
X X X X X X

For α ă 0, we have similarly that

ż ż ż ż ż
α f dµ “ ` ´
pα f q dµ ´ pα f q dµ “ p´αq f dµ ´ p´αq f ` dµ
´
X X X X X
„ż ż  ż ż
“ α f ` dµ ´ ´
f dµ “ α f dµ.
X X X X

Since the Claim 2 holds for β, g in place of α, f , and Claim 1 holds for pα f q, pβgq in place
of f , g, the theorem is proved. 

Theorem 11.47: Let pX, M, µq be a measure space, f : X Ñ r´8, 8s. If f P L pµq, then
ˇż ˇ ż
ˇ ˇ
ˇ f dµˇ ď | f | dµ.
ˇ ˇ
X X

Proof: We have
ż ż ż ż ż ż
` ´ ` ´
f dµ “ f dµ ´ f dµ ď f dµ ` f dµ ď | f | dµ.
X X X X X X

Similarly,
ż ż ż ż ż ż
´ ` ` ´
´ f dµ “ f dµ ´ f dµ ď f dµ ` f dµ ď | f | dµ.
X X X X X X

The result follows. 

101
MATH 362 Lecture Notes 11.4 Lebesgue integration

Theorem 11.48 (Lebesgue’s Dominated Convergence theorem): Suppose pX, M, µq is a

ptwise
measure space, and fn : X Ñ r´8, 8s is measurable for each n P N, with fn ÝÑ f.
If there exists a function g P L pµq such that

| fn pxq| ď gpxq, @ x P X, n “ 1, 2, . . . ,

then f P L pµq,
ż ż ż
lim | fn ´ f | dµ “ 0, and lim fn dµ “ f dµ.
nÑ8 X nÑ8 X X

Proof: f is measurable and | f | ď g (since it is the pointwise limit of the fn ’s, and each
| fn | ď g). Thus, f P L pµq. Since | fn ´ f | ď 2g, we may apply Fatou’s Lemma (11.42) to
the sequence of functions p2g ´ | fn ´ f |q to get
ż ż
lim inf p2g ´ | fn ´ f |q dµ ď lim inf p2g ´ | fn ´ f |q dµ.
X nÑ8 nÑ8 X

But lim infp2g ´ | fn ´ f |q “ limn p2g ´ | fn ´ f |q “ 2g, so

nÑ8
ż ż
2g dµ ď lim inf p2g ´ | fn ´ f |q dµ
X nÑ8 X
ż ˆ ż ˙
“ 2g dµ ` lim inf ´ | fn ´ f | dµ
X nÑ8
ż ż X
“ 2g dµ ´ lim sup | fn ´ f | dµ.
X nÑ8 X
ş
Since X 2g dµ is finite, we subtract it from both sides above to get
ż
lim sup | fn ´ f | dµ ď 0.
nÑ8 X
ş
Since X | fn ´ f | dµ ě 0, @ n, we must have equality (not inequality) above, and it must
be a limit (not an upper limit). The other result follows from Thm. 11.47. 

Theorem 11.49: Suppose f is Riemann-integrable on ra, bs, with (Riemann) integral

şb
a f pxq dx (how we will denote the value of the Riemann integral of f in this ş
theorem).
Then f is Lebesgue-integrable on ra, bs, and the value of its Lebesgue integral ra,bs f pxq dm
is equal to that of its Reimann integral.

102
MATH 362 Lecture Notes 11.4 Lebesgue integration

Proof: WLOG, we assume that a “ 0. Since f is Riemann-integrable, it is bounded—

that is, for each x P ra, bs, | f pxq| ď M for some M ą 0. For a fixed n P N we build a
uniform partition of the interval r0, bs

0 “ x0 ă x1 ă x2 ă ¨ ¨ ¨ ă xk “ b,

with xk “ pk{2n qb for k “ 1, . . . , 2n . Let Mnk , mnk denote the least upper bound and
greatest lower bound, respectively, of f on the kth subinterval xk´1 ď x ď xk . Note
that maxt|mnk |, |Mnk |u ď M, for each n, k. We obtain corresponding upper and lower
(Darboux) sums
n
2 2 n 2 n
ÿ b b ÿ b ÿ
Un “ Mnk n “ n Mnk and Ln “ n mnk .
k“1
2 2 k“1 2 k“1

Since f is assumed to be Riemann-integrable, we have

żb
f pxq dx “ lim Un “ lim Ln .
a n n

Returning for the moment to a fixed n with its corresponding partition, let us define
functions
f n pxq “ Mnk , when xk´1 ď x ă xk ,

taking f n pbq “ f pbq, and

f pxq “ mnk , when xk´1 ď x ă xk ,

n

again with f pbq “ f pbq. These f n , f are simple functions, and it is easy to see that
n n
their (Lebesgue) integrals coincide with the upper/lower Darboux sums:
ż ż
f n dm “ Un and f dm “ Ln .
r0,bs r0,bs n

It is also the case that the f n are monotone decreasing,

f 1 ě f 2 ě f 3 ě ¨ ¨ ¨ ě f,

and, hence, there is a pointwise limit function f “ limn f n , with f pxq ě f pxq, for each
x P ra, bs. Similarly, there is a limit function f “ limn f with f pxq ď f pxq, for a ď x ď b.
n
If we take gpxq “ M for x P ra, bs, then g is Lebesgue-integrable on ra, bs (with integral
Mpb ´ aq), and | f n pxq| ď gpxq on ra, bs. So,
ż ż ´ ¯ ż ż
lim Un “ lim f n dm “ lim f n dm “ f dm ě f dm,
n n r0,bs r0,bs n r0,bs r0,bs

where the 2nd equality holds by the LDCT (Thm. 11.48). Similarly,
ż ż ´ ¯ ż ż
lim Ln “ lim f dm “ lim f dm “ f dm ď f dm.
n n r0,bs n r0,bs n n r0,bs r0,bs

103
MATH 362 Lecture Notes 11.4 Lebesgue integration

We may distill from these last two strings of inequalities that

żb ż żb
f pxq dx “ lim Un ě f dm ě lim Ln “ f pxq dx,
a n r0,bs n a

and since the same (Riemann) integral value is at both ends of the inequality, every
ş
value in between (including the Lebesgue integral r0,bs f dm of f ) is the same. 

Question: What’s wrong with the Riemann theory of integration?

Answers:

ptwise
(i) One defect we have already seen. When fn ÝÑ f on E “ ra, bs, it need not be the case that
şb şb
a fn Ñ a f . We’ve already seen the counterexample
$ „ 
1 1
& npn ` 1q, if x P , ,
fn pxq :“ n`1 n ,
0, elsewhere.
%

ş1
which converges pointwise to the zero function on r0, 1s, despite 0 fn “ 1, @ n. But this
sequence p fn q is not uniformly bounded, and one might wonder if we restrict our attention
şb şb
to uniformly bounded sequences whether a fn Ñ a f . The answer is “no”, as seen by letting
pxn q be an enumeration of the rational numbers in r0, 1s, and defining
#
1, if x P tx1 , x2 , . . . , xn u,
fn pxq :“
0, otherwise.

Then #
ptwise 1, if x P r0, 1s with x rational,
fn ÝÑ f, where f pxq :“
0, otherwise.
But here the problem is somewhat different. It’s not that f has an integral which is unequal
şb
to limn a fn . Rather, it seems that the definition of the Riemann integral itself is to blame, not
being sufficiently general to allow f to be integrable in the Riemann sense of this term.

(ii) When first investigating metric spaces, we discussed several concrete examples. Several took
Euclidean space Rp as the collection of objects, but differed in their choice of metric. Vectors
in Rp are really functions with discrete (finite) domains. To extend these ideas to functions
defined on intervals ra, bs, we took (or, at least, could have taken)
ˆż ˙1{p
dp p f, gq :“ p
| f ´ g| dµ . (14)
ra,bs

At the time we asserted this is a metric on Cpa, bq, for each p ě 1, where we assumed
the integral was a Riemann integral. If we employ the Lebesgue integral in Equation (14)
and allow f , g to be measurable functions (not simply continuous ones) which are Lebesue-
integrable, then dp is a metric except for the fact that there are unequal functions f , g for which

104
MATH 362 Lecture Notes 11.5 Littlewood’s Principles

ˇ
dp p f, gq “ 0. So, we define the equivalence relation f g precisely when µ x ˇ f pxq , gpxq “ 0
(

and, taking r f s to denote the equivalence class of f , we define

" ż *
p
L pµq :“ r f s : f : X Ñ r´8, 8s is measurable, with | f | dµ ă 8 .
p
X

It turns out that these Lp -spaces are complete, for each p ą 1.

(iii) We next obtain a continuous-domain analog for Euclidean space with the 8-norm. For
f : X Ñ r´8, 8s we define the essential supremum of f to be
ˇ
} f }8 :“ inf c : µp x ˇ | f pxq| ą c q “ 0 .
( (

We then set L8 pµq to be the collection

ˇ
L8 pµq “ r f s ˇ f : X Ñ r´8, 8s is measurable, and }r f s}8 ă 8 .
(

Like the Lp pµq spaces with µ ă 8, this space is complete.

11.5 Littlewood’s Principles

Definition 11.50: Suppose f : Rp Ñ R. We define the support of f , denoted suppp f q, to be

ˇ (
suppp f q :“ x ˇ f pxq , 0 , a closed set.

We say that f has compact support if suppp f q is a compact set. For a subset S Ă Rp , we
denote the collection of continuous functions on S with compact support by Cc pSq.

We first state, without proof, several important theorems.

Theorem 11.51 (Lusin): Let pRp , M, µq be a measure space, f : Rp Ñ p´8, 8q is measurable,

ˇ
and x ˇ f pxq , 0 Ă A P M with µpAq ă 8. Then for each δ ą 0 there exists a function
(
ˇ
g P Cc pRp q such that µp x ˇ f pxq , gpxq q ă δ. Furthermore, we may arrange it so that
(

supxPX |gpxq| ď supxPX | f pxq|.

Theorem 11.52 (Egorov): Let pX, M, µq be a measure space with µpXq ă 8. If p fn q is a

sequence of measurable functions from X into r´8, 8s converging pointwise to a real-
valued function f on X, and if δ ą 0, then there is a measurable set Eδ Ă X with µpXzEδ q ă δ,
such that p fn q converges uniformly on Eδ .

105
MATH 362 Lecture Notes 11.5 Littlewood’s Principles

In the book Lectures on the Theory of Functions, from 1944, Littlewood stated the three principles of
measure theory named for him. He said

The extent of knowledge required is nothing like so great as is sometimes supposed.

There are three principles, roughly expressible in the following terms:

1. Every measurable set is nearly a finite union of intervals.

2. Every measurable function is nearly continuous.
3. Every convergent seq. of measurable fns is nearly uniformly convergent.

Most of the results of [the theory] are fairly intuitive applications of these ideas, and
the student armed with them should be equal to most occasions when real variable
theory is called for. If one of the principles would be the obvious means to settle the
problem if it were ’quite’ true, it is natural to ask if the ’nearly’ is near enough, and for
a problem that is actually solvable it generally is.

106