Handouts RealAnalysis II

COMSATS University Islamabad, Islamabad
Pakistan
Real Analysis II (MTH631)
Salman Amin Malik
Department of Mathematics, CUI Islamabad

Fall 2019
Email: [email protected],
[email protected]
To my unknown students
ii
About the instructor

Dr. Malik did his MS and PhD (Mathematics) from University of La Rochelle, La
Rochelle, France in 2009 and 2012, respectively. Prior to MS and PhD, Dr. Malik
completed his MPhil and MSc (Mathematics) from Department of Mathematics,
University of the Punjab, Lahore, Pakistan. He has been aliated with several
universities in Pakistan and abroad. He has the experience of teaching a wide range
of mathematics courses at undergraduate and graduate level.
Dr. Malik has published several research articles in international journals and
conferences. His area of research includes the study of dierential equations with
nonlocal operators and their applications to image processing. He is also interested
in inverse problems related to reaction-diusion equations with nonlocal integro-
dierential operators and boundary conditions. These models have numerous appli-
cations in anomalous diusion/transport, biomedical imaging and non-destructive
testing.
About the handouts

The books followed during this course are: W. Rudin, Principles of Mathemat-
ical Analysis, Third Edition, McGraw-Hill, 1976. ISBN: 9780070542358. and
W. F. Trench, Introduction to Real Analysis, Pearson Education, 2013. Con-
sequently, the most of the examples considered in these notes are from the above
mentioned books and their exercises, but not restricted to those books only. If
you nd any typing error in the text kindly report to me by writing an email to
[email protected].
iii
Course Information
Title and Course Code: Real Analysis II (MTH631)
Number of Credit Hours: 3 credits
Course Objective: Real Analysis II is the follow up course of Real Analysis I and in
general an advanced course related to mathematical analysis. The topics of the Real
Analysis II are linked with its rst course namely Real Analysis I, indeed, we will
extend the ideas of Real Analysis I to Euclidean space Rn , we will discuss sequences
and series of functions, limits and continuity of functions of several variables, partial
derivatives their applications, multiple integrals etc. Upon completion of this course
students will be able to
• Understand the convergence of sequence of functions (LO1).
• Understand the pointwise convergence, uniform convergence, several tests for

convergence (LO2).
• Apply the interchange of limit and integration, derivative of sequence of func-

tions (LO3).
• Understand the innite series of functions, convergence, Weierstrass's test and

some other results about the convergence (LO4).
• Apply Dirichlet's test for uniform convergence, series of product of two func-
tions, interchange of sum and intgeration (LO5).
• Represent and study the function which could be written as power series,term
by term integral and derivative of a power series, (LO6). item Understand the
concept of equicontinuous function, The Stone-Weierstrass Theorem (LO7).
• Understand and nd the Fourier series, Fourier coecients, convergence of

Fourier series (LO8).
• Apply the best approximation theorem and understand the Euler gamma func-
tion and the beta function and their properties (LO9)
• Understand the functions of several variables, Heine-Borel Theorem, limits

and continuity of functions of several variables (LO10)
• Vector valued functions and their calculus, Bounded functions and several
results about vector valued functions (LO11)
• Dierentiability in Rn , Dierentials, Directional derivatives, Partial deriva-

tives, Maxima and minima (LO12)
• Improper integrals, Multiple integrals, Functions of bounded variation (LO15)

iv
Prerequisites: Real Analysis I (MTH621)
The textbooks for this course:

[1] W. Rudin, Principles of Mathematical Analysis, Third Edition, McGraw-Hill,
1976. ISBN: 9780070542358.
[2] W. F. Trench, Introduction to Real Analysis, Pearson Education, 2013.
[3] S. Ponnusamy, Foundations of Mathematical Analysis, Birkhauser, 2012.
Reference books:
[4] A. N. Kolmogorov and S. V. Fomin, Introductory Real Analysis, Revised English
Edition Translated and Edited by R. A. Silverman, Dover Publication, Inc. New
York.
[5] R. G. Bartle and D. R. Sherbert, Introduction to Real Analysis, Third Edition,

2000, John Wiley & Sons Inc.
• Sequences and Series of functions
• Functions of several variables
• Vector valued functions
• Integral Calculus
Contents
1 Sequences and Series of Functions 2
1.1 Informal way . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Pointwise Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Norm Dened Over a Set . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Cauchy's Uniform Convergence Criterion . . . . . . . . . . . . . . . . 8
1.6 Properties Preserved by Uniform Convergence . . . . . . . . . . . . . 10
1.6.1 Continuity of the Limit Function at a Point . . . . . . . . . . 10
1.6.2 Interchange of Limit and Integration . . . . . . . . . . . . . . 12
1.6.3 Under What Conditions We May Have F ′ = limn→∞ Fn′ . . . 14
1.7 Series of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.8 Convergence of Series of Functions . . . . . . . . . . . . . . . . . . . 15
1.8.1 Cauchy's criterion for functional series . . . . . . . . . . . . . 17
1.8.2 Dominated Series of Real Numbers for Series of Functions . . 18
1.8.3 Weierstrass M-test/dominated Convergence Test . . . . . . . 18
1.9 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.10 The Taylor's Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.11 The Abel's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.12 Pointwise and Uniform Bounded Functions . . . . . . . . . . . . . . . 47
1.13 Equicontinuous Functions on a Set . . . . . . . . . . . . . . . . . . . 48
1.14 The Stone-Weierstrass Theorem . . . . . . . . . . . . . . . . . . . . . 51
1.15 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.15.1 Periodic Functions . . . . . . . . . . . . . . . . . . . . . . . . 53
1.15.2 Periodic Extension . . . . . . . . . . . . . . . . . . . . . . . . 55
1.15.3 Trigonometric Polynomials . . . . . . . . . . . . . . . . . . . 56
1.16 The space E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.16.1 Fourier Series of Even and Odd Functions . . . . . . . . . . . 61
1.17 Fourier Series for Arbitrary Periodic Function . . . . . . . . . . . . . 66
1.18 Best Approximation Theorem . . . . . . . . . . . . . . . . . . . . . . 69
2 Functions of Several Variables 71

2.1 Euclidean Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.2 Schwarz's Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Line Segment in R
2.2.1
n . . . . . . . . . . . . . . . . . . . . . . . 75
2.3 Neighbourhoods and Open Sets in Rn . . . . . . . . . . . . . . . . . 75
Convergence of a Sequence in R
2.4
n . . . . . . . . . . . . . . . . . . . . 77
2.5 Principle of nested sets . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.6 Heine-Borel Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.7 Connected Sets in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Contents vi
2.7.1 Polygonal Path . . . . . . . . . . . . . . . . . . . . . . . . . . 81

2.8 Polygonally Connected Set . . . . . . . . . . . . . . . . . . . . . . . . 81
2.9 Sequences in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.10 Domain of Function of n Variable . . . . . . . . . . . . . . . . . . . . 85
2.11 Limit at a Point of a Function of n Variables . . . . . . . . . . . . . 85
2.12 Innite Limits and Limits at X→∞ . . . . . . . . . . . . . . . . . . 88
2.12.1 Limit at Innity . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.13 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.14 Vector Valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.14.1 Composite Function . . . . . . . . . . . . . . . . . . . . . . . 93
2.15 Bounded Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.16 Directional Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.16.1 Partial Derivative . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.16.2 Equality of Mixed Partial Derivatives . . . . . . . . . . . . . . 100
2.16.3 Generalization of Equality of Mixed Partial Derivative . . . . 101
2.17 Dierentiability of Functions of Several Variables . . . . . . . . . . . 102
2.17.1 Linear Function . . . . . . . . . . . . . . . . . . . . . . . . . . 104
2.17.2 Dierential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
2.17.3 A sucient Condition for Dierentiability . . . . . . . . . . . 107
2.17.4 Continuously Dierentiable Function . . . . . . . . . . . . . . 108
2.17.5 Geometric Interpretation of Dierentiability . . . . . . . . . . 110
2.18 Maxima and Minima . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
2.19 Dierentiable Vector Valued Function . . . . . . . . . . . . . . . . . 114
2.20 The Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
2.21 Higher derivatives of composite functions . . . . . . . . . . . . . . . . 118
2.22 rth Dierential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
2.23 Taylor's Theorem for Functions of n Variables . . . . . . . . . . . . . 124
2.23.1 Positive Denite . . . . . . . . . . . . . . . . . . . . . . . . . 126
3 Integral Calculus 129

3.1 Locally Integrable Functions . . . . . . . . . . . . . . . . . . . . . . . 129
3.1.1 The Comparison Test . . . . . . . . . . . . . . . . . . . . . . 134
3.2 Absolute integrability . . . . . . . . . . . . . . . . . . . . . . . . . . 137
3.3 Nonoscillatory and Oscillatory Functions . . . . . . . . . . . . . . . . 138
3.4 Conditional convergence . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.5 Dirichlet's Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Rectangles in R
3.6
n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
3.7 Riemann Sum in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
Riemann Integral in R
3.8
n . . . . . . . . . . . . . . . . . . . . . . . . . 143
3.9 Upper and Lower Integrals . . . . . . . . . . . . . . . . . . . . . . . . 146
3.10 Sets with Zero Content . . . . . . . . . . . . . . . . . . . . . . . . . . 151
3.11 Integral Over Bounded Set . . . . . . . . . . . . . . . . . . . . . . . . 152
3.12 Dierentiable Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 153
3.13 Iterated Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
List of Figures
( )
1.1 Plot of Fn (x) = 1
, n ≥ 1, for n = 1, 2, 4, 8, 100 . . . . . . . . 2
( n+x
)
1.2 Plot of Fn (x) = x
n+x , n ≥ 1, for n = 1, 2, 4, 8, 20, 100 . . . . . . . 3
1.3 Uniform convergence graphically . . . . . . . . . . . . . . . . . . . . 5

1.4 Periodic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.5 Geometric proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
1.6 Periodic extension example 1 . . . . . . . . . . . . . . . . . . . . . . 55
1.7 Periodic extension example 2 . . . . . . . . . . . . . . . . . . . . . . 56
1.8 At discontinuous points . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.9 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.10 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.1 The set S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

2.2 Heine-Borel Theorem for n=2 . . . . . . . . . . . . . . . . . . . . . 79
2.3 Disconnected set which is not a region . . . . . . . . . . . . . . . . . 83
2.4 A connected set which is not a region . . . . . . . . . . . . . . . . . . 83
2.5 A region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.6 Domain of the function . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.7 Composite of vector valued functions . . . . . . . . . . . . . . . . . . 94
2.8 Domain of the function . . . . . . . . . . . . . . . . . . . . . . . . . . 110
2.9 Geometric interpretation of dierentiability . . . . . . . . . . . . . . 111
3.1 Rectangle in R2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

Rectangular parallelepiped in R
3.2
3 . . . . . . . . . . . . . . . . . . . . 141
Partitioning of a rectangle in R
3.3
2 . . . . . . . . . . . . . . . . . . . . 142
3.4 S1 and S2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Chapter 1
Sequences and Series of Functions
1.1 Informal way

If Fk , Fk+1 , ...,Fn , . . . are real-valued functions dened on a subset D of the real
numbers, we say that {Fn } is an innite sequence or (simply a sequence ) of functions
on D. For each x0 ∈ D, we have a sequence of real numbers and we can talk about
the convergence of that sequence of real numbers.
If the sequence of values {Fn (x)} converges for each x in some subset S of D, then
{Fn } denes a limit function on S.
Example: The functions

( )
1
Fn (x) = , n ≥ 1,
n+x
dene a sequence on D = [0, ∞).
( )
Figure 1.1: Plot of Fn (x) = 1
n+x , n ≥ 1, for n = 1, 2, 4, 8, 100

( )
x
Fn (x) = , n ≥ 1,
n+x
dene a sequence on D = [0, ∞).

1.2. Pointwise Convergence 3
( )
Figure 1.2: Plot of Fn (x) = x
n+x , n ≥ 1, for n = 1, 2, 4, 8, 20, 100
( )
nx n/2
Fn (x) = 1 − , n ≥ 1,
n+1
dene a sequence on D = (−∞, 1].
1.2 Pointwise Convergence

Suppose that {Fn } is a sequence of functions on D and the sequence of values
{Fn (x)} converges for each x in some subset S of D . Then we say that {Fn }
converges pointwise on S to the limit function F , dened by
F (x) = lim Fn (x), x ∈ S.

n→∞
Example: The sequence of functions dened by

 ∞, x < 0,
lim Fn (x) = 1, x = 0,
n→∞ 
0, 0 < x ≤ 1.
Therefore, {Fn } converges pointwise on S = [0, 1] to the limit function F dened by
{
1, x = 0,
F (x) =
0, 0 < x ≤ 1.
Example: Consider the functions
Fn (x) = xn e−nx , x ≥ 0, n ≥ 1.
1.2. Pointwise Convergence 4
Equating the derivative

Fn′ (x) = nxn−1 e−nx (1 − x)
to zero shows that the maximum value of Fn (x) on [0, ∞) is e−n , attained at x = 1.
Therefore,
|Fn (x)| ≤ e−n , x ≥ 0,
solimn→∞ Fn (x) = 0 for all x ≥ 0. The limit function in this case is identically zero
on [0, ∞).
Example: For n ≥ 1, let be dened on (−∞, ∞) by

Fn


 0, x < − n2 ,





 −n(2 + nx), − n2 ≤ x < − n1 ,



Fn (x) = n2 x, − n1 ≤ x < n1 ,




 n(2 − nx),
 n ≤ x < n,
1 2





0, x ≥ n2
Since Fn (0) = 0 for all n, limn→∞ Fn (0) = 0. If x ̸= 0, then Fn (x) = 0 if

n ≥ 2/|x|. Therefore,
lim Fn (x) = 0, −∞ < x < ∞,

n→∞
so the limit function is identically zero on (−∞, ∞).
Example: Show that the sequence of functions
( )
1
Fn (x) = , n ≥ 1,
n+x
dene a sequence on D = [0, ∞), converges to 0.
Example: For each positive integer n, let Sn be the set of numbers of the form
x = p/q , where p and q are integers with no common factors and 1 ≤ q ≤ n.
Dene {
1, x ∈ Sn ,
Fn (x) =
0, x ∈
̸ Sn .
Ifx is irrational, then x ̸∈ Sn for any n, so Fn (x) = 0, n ≥ 1. If x is rational, then
x ∈ Sn and Fn (x) = 1 for all suciently large n.
Therefore,
{
1 if x is rational,
lim Fn (x) = F (x) =
n→∞ 0 if x is irrational.
1.3. Norm Dened Over a Set 5
1.3 Norm Dened Over a Set

Let us introduce the notation
∥g∥S = sup |g(x)|.

x∈S
Lemma: If g and h are dened on S, then
∥g + h∥S ≤ ∥g∥S + ∥h∥S

∥gh∥S ≤ ∥g∥S ∥h∥S .
Moreover, if either g or h is bounded on S, then
∥g − h∥S ≥ |∥g∥S − ∥h∥S ∥| .
1.4 Uniform Convergence

A sequence {Fn } of functions dened on a set S converges uniformly to the limit
function F on S if
lim ||Fn − F ∥S = 0.
n→∞
Thus, {Fn } converges uniformly to F on S if for each ε>0 there is an integer N
such that
∥Fn − F ∥S < ε if n ≥ N. (1.1)
y =F (x) +
y =F (x)
y =F (x) −
x
a b
Figure 1.3: Uniform convergence graphically
A sequence {Fn } of functions dened on a set S converges uniformly to the limit

function F on S if
lim ||Fn − F ∥S = 0.
n→∞
1.4. Uniform Convergence 6
Thus, {Fn } converges uniformly to F on S if for each ε>0 there is an integer N

such that
∥Fn − F ∥S < ε if n ≥ N. (1.2)
If S = [a, b] and F is the function with graph shown in then (1.2) implies that the
graph of
y = Fn (x), a ≤ x ≤ b,
lies in the shaded band
F (x) − ε < y < F (x) + ε, a ≤ x ≤ b, n≥N
.
Example: The sequence {Fn } dened by
Fn (x) = xn e−nx , n ≥ 1,
converges uniformly to F ≡ 0.
We have
∥Fn − F ∥S = ∥Fn ∥S = e−n ,
so
∥Fn − F ∥S < ε
if n > − log ε. For these values of n, the graph of
y = Fn (x), 0 ≤ x < ∞,
lies in the strip

−ε ≤ y ≤ ε, x≥0
Theorem: Let {Fn } be dened on S. Then
1. {Fn } converges pointwise to F on S if and only if there is, for each ε > 0 and
x ∈ S , an integer N (which may depend on x as well as ε) such that
|Fn (x) − F (x)| < ε if n ≥ N (ε, x).
2. {Fn } converges uniformly to F on S if and only if there is for each ε > 0 an

integer N (which depends only on ε and not on any particular x in S) such
that
|Fn (x) − F (x)| < ε for all x in S if n ≥ N (ε).
Theorem: If {Fn } converges uniformly to F on S, then {Fn } converges pointwise

to F on S.
The converse is false; that is, pointwise convergence does not imply uniform conver-
gence.
1.4. Uniform Convergence 7
Counter example: n ≥ 1, let Fn be

For dened on (−∞, ∞) by


 0, x < − n2 ,





 −n(2 + nx), − n2 ≤ x < − n1 ,



Fn (x) = n2 x, − n1 ≤ x < n1 ,





 n(2 − nx), 1
≤ x < n2 ,

 n



0, x≥ 2
n
The sequence {Fn } of converges pointwise to F ≡ 0 on (−∞, ∞), but not uniformly.
Because ( ) ( )
1 −1
∥Fn − F ∥(−∞,∞) = Fn = Fn = n,
n n
so
lim ∥Fn − F ∥(−∞,∞) = ∞.
n→∞
Counter example: For n ≥ 1, let Fn be dened on (−∞, ∞) by



 0, x < − n2 ,





 −n(2 + nx), − n2 ≤ x < − n1 ,



Fn (x) = n2 x, − n1 ≤ x < n1 ,




 n(2 − nx),
 n ≤ x < n,
1 2





0, x ≥ n2
However, the convergence is uniform on
Sρ = (−∞, ρ] ∪ [ρ, ∞)
for any ρ > 0, since

2
∥Fn − F ∥Sρ = 0 if n> .
ρ
How to show that a sequence of functions is not uniformly convergent?

Suppose that a sequence of function Fn is point wise convergent on the set S.
Then the convergence of Fn is not uniform, if there exists an ε > 0 such that to
each integer N there correspond and integer n>N and a point xn ∈ S for which
we have
|Fn (xn ) − F (xn )| ≥ ε.
Example: If Fn (x) = xn , n ≥ 1, then {Fn } converges pointwise on S = [0, 1] to
{
1, x = 1,
F (x) =
0, 0 ≤ x < 1.
1.5. Cauchy's Uniform Convergence Criterion 8
The convergence is not uniform on S. To see this, suppose that 0 < ε < 1. Then
|Fn (x) − F (x)| > 1 − ε if (1 − ε)1/n < x < 1.
Therefore,
1 − ε ≤ ∥Fn − F ∥S ≤ 1
for all n ≥ 1. Since ε can be arbitrarily small, it follows that
∥Fn − F ∥S = 1 for all n ≥ 1.
Example: If Fn (x) = xn , n ≥ 1, then {Fn } converges pointwise on S = [0, 1] to
{
1, x = 1,
F (x) =
0, 0 ≤ x < 1.
However, the convergence is uniform on [0, ρ] if 0 < ρ < 1, since then
∥Fn − F ∥[0,ρ] = ρn
and limn→∞ ρn = 0. Another way to say the same thing: {Fn } converges uniformly
on every closed subset of [0, 1).
1.5 Cauchy's Uniform Convergence Criterion

Theorem: A sequence of functions {Fn } converges uniformly on a set S if and
only if for each ε > 0 there is an integer N such that
∥Fn − Fm ∥S < ε if n, m ≥ N. (1.3)
Proof : For necessity, suppose that {Fn } converges uniformly to F on S.

Then, if ε > 0, there is an integer N such that
ε
∥Fk − F ∥S < if k ≥ N.
2
Therefore,
∥Fn − Fm ∥S = ∥(Fn − F ) + (F − Fm )∥S

≤ ∥Fn − F ∥S + ∥F − Fm ∥S
ε ε
< + = ε if m, n ≥ N.
2 2
For suciency, we rst observe that (1.17) implies that
|Fn (x) − Fm (x)| < ε if n, m ≥ N,
for any xed x in S.

1.5. Cauchy's Uniform Convergence Criterion 9
Therefore, Cauchy's convergence criterion for sequences of constants implies that

{Fn (x)} converges for each x in S; that is, {Fn } converges pointwise to a limit
function F on S.
To see that the convergence is uniform, we write
|Fm (x) − F (x)| = |[Fm (x) − Fn (x)] + [Fn (x) − F (x)]|

≤ |Fm (x) − Fn (x)| + |Fn (x) − F (x)|
≤ ∥Fm − Fn ∥S + |Fn (x) − F (x)|.
This and (1.17) imply that
|Fm (x) − F (x)| < ε + |Fn (x) − F (x)| if n, m ≥ N. (1.4)
Since limn→∞ Fn (x) = F (x),
|Fn (x) − F (x)| < ε
for some n ≥ N, so (1.4) implies that
|Fm (x) − F (x)| < 2ε if m ≥ N.
But this inequality holds for all x in S, so
∥Fm − F ∥S ≤ 2ε if m ≥ N.
Since ε is an arbitrary positive number, this implies that {Fn } converges uniformly
to F on S.
Example: Suppose that g is dierentiable on S = (−∞, ∞) and
|g ′ (x)| ≤ r < 1, −∞ < x < ∞. (1.5)
Let F0 be bounded on S and dene
Fn (x) = g(Fn−1 (x)), n ≥ 1. (1.6)
Show that {Fn } converges uniformly on S.
Solution: We rst note that if u and v are any two real numbers, then (1.5) and
the mean value theorem imply that
|g(u) − g(v)| ≤ r|u − v|. (1.7)
Recalling (1.6) and applying this inequality with u = Fn−1 (x) and v = 0 shows that
|Fn (x)| = |g(0) + (g(Fn−1 (x)) − g(0))|

≤ |g(0)| + |g(Fn−1 (x)) − g(0)|
≤ |g(0)| + r|Fn−1 (x)|.
1.6. Properties Preserved by Uniform Convergence 10
Therefore, since F0 is bounded on S, it follows by induction that Fn is bounded on

S for n ≥ 1.
Moreover, if n ≥ 1, then (1.6) and (1.7) with u = Fn (x) and v = Fn−1 (x) imply
that
|Fn+1 (x) − Fn (x)| = |g(Fn (x)) − g(Fn−1 (x))|

≤ r|Fn (x) − Fn−1 (x)|, −∞ < x < ∞,
so
∥Fn+1 − Fn ∥S ≤ r∥Fn − Fn−1 ∥S .
By induction, this implies that
∥Fn+1 − Fn ∥S ≤ rn ∥F1 − F0 ∥S . (1.8)
If n > m, then
∥Fn − Fm ∥S = ∥(Fn − Fn−1 ) + (Fn−1 − Fn−2 ) + · · ·

+(Fm+1 − Fm )∥S
≤ ∥Fn − Fn−1 ∥S + ∥Fn−1 − Fn−2 ∥S + · · ·
+∥Fm+1 − Fm ∥S .
Now (1.8) implies that
∥Fn − Fm ∥S ≤ ∥F1 − F0 ∥S (1 + r + r2 + · · · + rn−m−1 )rm

rm
< ∥F1 − F0 ∥S .
1−r
rN
Therefore, if ∥F1 − F0 ∥S < ε,
1−r
then ∥Fn − Fm ∥S < ε if n, m ≥ N .
1.6 Properties Preserved by Uniform Convergence
1.6.1 Continuity of the Limit Function at a Point

Theorem: If {Fn } converges uniformly to F on S and each Fn is continuous at a
point x0 in S, then so is F. Similar statements hold for continuity from the right
and left.
Proof : Suppose that each Fn is continuous at x0 . If x∈S and n ≥ 1, then
|F (x) − F (x0 )| ≤ |F (x) − Fn (x)| + |Fn (x) − Fn (x0 )| + |Fn (x0 ) − F (x0 )|
≤ |Fn (x) − Fn (x0 )| + 2∥Fn − F ∥S .
(1.9)
Suppose that ε > 0. Since {Fn } converges uniformly to F on S, we can choose n so

that ∥Fn − F ∥S < ε. For this xed n, (1.9) implies that
|F (x) − F (x0 )| < |Fn (x) − Fn (x0 )| + 2ε, x ∈ S. (1.10)
Since Fn is continuous at x0 , there is a δ>0 such that
|Fn (x) − Fn (x0 )| < ε if |x − x0 | < δ.
So, from (1.10),

|F (x) − F (x0 )| < 3ε, if |x − x0 | < δ.
Therefore, F is continuous at x0 .
Similar arguments apply to the assertions on continuity from the right
and left.
Corollary: If {Fn } converges uniformly to F on S and each Fn is continuous on S,
then so is F; that is, a uniform limit of continuous functions is continuous.
Proof : See video lectures.
Remark: If {Fn } converges uniformly to F on S. Is the following
∫ b ∫ b
F (x) dx = lim Fn (x) dx,
a n→∞ a
is true?
∫b ∫b
Example: a F (x) dx = limn→∞ a Fn (x) dx, is not true generally.
Consider the sequence of functions dened on S = [0, 1]



 0, x = 0,


Fn (x) = n, 0 ≤ x ≤ n1 ,




0, n1 < x < 1.
Then the sequence {Fn } converges pointwise to F (x) = 0 on [0, 1] and it is not
uniformly convergent. We have
∫ 1 ∫ 1/n ∫ 1 ∫ 1
Fn (x) dx = n dx + 0dx = 1 But F (x) dx = 0
0 0 1/n 0
∫ b ∫ b
lim Fn (x) dx ̸= lim Fn (x) dx,
a n→∞ n→∞ a
1.6.2 Interchange of Limit and Integration

Theorem: Suppose that {Fn } converges uniformly to F on S = [a, b]. Assume that
F and all Fn are integrable on [a, b]. Then
∫ b ∫ b
F (x) dx = lim Fn (x) dx. (1.11)
a n→∞ a
Proof : Consider ∫ b ∫ b

Fn (x) dx − F (x) dx

a a
∫ b ∫ b ∫ b

Fn (x) dx − F (x) dx ≤ |Fn (x) − F (x)| dx

a a a
≤ (b − a)∥Fn − F ∥S
and limn→∞ ∥Fn − F ∥S = 0, the conclusion follows.
Remark: Recall the theorem we have just proved; i.e.,
Theorem: Suppose that {Fn } converges uniformly to F on S = [a, b]. Assume that
F and all Fn are integrable on [a, b].
Then ∫ b ∫ b
F (x) dx = lim Fn (x) dx.
a n→∞ a
The hypotheses of Theorem are stronger than necessary.

Theorem: Suppose that {Fn } converges pointwise to F and each Fn is integrable
on [a, b].
1. If the convergence is uniform, then F is integrable on [a, b] and
∫ b ∫ b
a n→∞ a
holds.
2. If the sequence {∥Fn ∥[a,b] } is bounded and F is integrable on [a, b], then
∫ b ∫ b
a n→∞ a
holds.
Remark: Part (1) of this theorem shows that it is not necessary to assume that F
is integrable on [a, b], since this follows from the uniform convergence. Part (2) is
known as the bounded convergence theorem. Neither of the assumptions of (2) can
be omitted.
Example (Unbounded sequence of functions): For n ≥ 1, let Fn be dened

on (−∞, ∞) by


 0, x < − n2 ,





 −n(2 + nx), − n2 ≤ x < − n1 ,



Fn (x) = n2 x, − n1 ≤ x < n1 ,





 n(2 − nx), 1
≤ x < n2 ,

 n



0, x≥ 2
n
{∥Fn ∥[0,1] } is unbounded while F is integrable on [0, 1],
∫ 1 ∫ 1
Fn (x) dx = 1, n ≥ 1, but F (x) dx = 0.
0 0
Example (Bounded sequence of functions but limit is not integrable): For

each positive integer n, let Sn be the set of numbers of the form x = p/q , where p
and q are integers with no common factors and 1 ≤ q ≤ n.
Dene {
1, x ∈ Sn ,
Fn (x) =
0, x ∈
̸ Sn .
Ifx is irrational, then x ̸∈ Sn for any n, so Fn (x) = 0, n ≥ 1. If x is rational, then
x ∈ Sn and Fn (x) = 1 for all suciently large n. Therefore,
{
1 if x is rational,
lim Fn (x) = F (x) =
n→∞ 0 if x is irrational.
In this example it is clear that ∥Fn ∥[a,b] = 1 for every nite interval [a, b], Fn is
integrable for all n ≥ 1, and F is nonintegrable on every interval.
Example: The sequence {Fn } dened by
1
Fn (x) = xn sin .
xn−1
The sequence of functions converges {Fn } converges uniformly to F ≡0 on [r1 , r2 ]
if 0 < r1 < r2 < 1 (or, equivalently, on every compact subset of (0, 1)).
However,
1 1
Fn′ (x) = nxn−1 sin − (n − 1) cos ,
xn−1 xn−1
so {Fn′ (x)} does not converge for any x in (0, 1).
1.6.3 Under What Conditions We May Have F ′ = limn→∞ Fn′

Theorem: Suppose that Fn′ is continuous on [a, b] for all n ≥ 1 and {Fn′ } converges
uniformly on [a, b]. Suppose also that {Fn (x0 )} converges for some x0 in [a, b].
Then {Fn } converges uniformly on [a, b] to a dierentiable limit function F, and
F ′ (x) = lim Fn′ (x), a < x < b, (1.12)

n→∞
while
F+′ (a) = lim Fn′ (a+) and F−′ (b) = lim Fn′ (b−). (1.13)
n→∞ n→∞
Proof : Since Fn′ is continuous on [a, b], due to fundamental theorem of calculus, we
can write ∫ x
Fn (x) = Fn (x0 ) + Fn′ (t) dt, a ≤ x ≤ b. (1.14)
x0
Let L = lim Fn (x0 ), G(x) = lim Fn′ (x). (1.15)

n→∞ n→∞
Since Fn′ is continuous and {Fn′ } converges uniformly to G on [a, b], G is contin-
uous on [a, b].
∫b
Therefore,
∫b
(1.14) and using the fact we have proved
a F (x) dx =
limn→∞ Fn (x) dx (with F and Fn replaced by G ′
and Fn ) imply that {Fn } con-
a
verges pointwise on [a, b] to the limit function
∫ x
F (x) = L + G(t) dt.
x0
∫ x
F (x) = L + G(t) dt. (1.16)
x0
The convergence is actually uniform on [a, b], since subtracting (1.14) from (1.16)
yields
∫
x
|F (x) − Fn (x)| ≤ |L − Fn (x0 )| + |G(t) − Fn′ (t)| dt
x0
≤ |L − Fn (x0 )| + |x − x0 | ∥G − Fn′ ∥[a,b] .
Consequently,
∥F − Fn ∥[a,b] ≤ |L − Fn (x0 )| + (b − a)∥G − Fn′ ∥[a,b] ,
where the right side approaches zero as n → ∞.

Since G is continuous on [a, b], (1.15), (1.16), Denition ??, and Theorem ??
imply (1.12) and (1.13).
1.7. Series of Functions 15
1.7 Series of Functions

If {fj }∞ is a sequence of real-valued functions dened on a set D of real numbers,
∑k∞
then j=k fj is an innite series (or simply a series ) of functions on D.
∑∞
The partial sums of , j=k fj are dened by
∑
n
Fn = fj , n ≥ k.
j=k
If{Fn }∞
k converges pointwise to a function F on a subset S of D, we say that
∑∞
f
j=k j converges pointwise to the sum F on S , and write
∞
∑
F = fj , x ∈ S.
j=k
∑∞
{Fn } converges uniformly to F
If on S , we say that j=k fj converges uniformly
to F on S .
1.8 Convergence of Series of Functions

∑∞
The innite series of functions j=k fj on D is said to be uniformly convergent if
the sequence of partial sum {Fn } dened by
∑
n
Fn = fj , n ≥ k.
j=k
converges uniformly to F (x) on D.
Example: For the functions
fj (x) = xj , j ≥ 0,
dene the innite series of functions
∞
∑
xj
j=0
on D = (−∞, ∞).
Pointwise convergence: The nth partial sum of the series is
Fn (x) = 1 + x + x2 + · · · + xn ,
or, in closed form,

{ 1−xn+1
1−x , x ̸= 1,
Fn (x) =
n + 1, x = 1.
1.8. Convergence of Series of Functions 16
We have seen earlier that {Fn } converges pointwise to
1
F (x) =
1−x
if |x| < 1 and diverges if |x| ≥ 1.
Hence, we write
∞
∑ 1
xj = , −1 < x < 1.
1−x
j=0
Since the dierence

xn+1
F (x) − Fn (x) =
1−x
can be made arbitrarily large by taking x close to 1,
∥F − Fn ∥(−1,1) = ∞,
so the convergence is not uniform on (−1, 1).

We have seen earlier that {Fn } converges pointwise to
1
F (x) =
1−x
if |x| < 1 and diverges if |x| ≥ 1.
Neither is it uniform on any interval (−1, r] with −1 < r < 1, since
1
∥F − Fn ∥(−1,r) ≥
2
for every n on every such interval.
Example: For the functions fj (x) = xj , j ≥ 0,

∑∞
discuss the uniform convergence
of the innite series of functions j=0 fj (x).
Uniform convergence: The series does converge uniformly on any interval [−r, r]
with 0 < r < 1, since
rn+1
∥F − Fn ∥[−r,r] =
1−r
and limn→∞ rn = 0. Put another way, the series converges uniformly on closed
subsets of (−1, 1).
Uniform convergence (using ε): See video lectures.
∑∞
Remark: A necessary condition for j=0 fj (x) to converge on S is that fj (x) → 0
for each x ∈ S.
Remark: As for series of constants, the convergence, pointwise or uniform, of a

series of functions is not changed by altering or omitting nitely many terms. This
justies adopting the convention that we used for series of constants: when we are
interested only in whether a series of functions converges, and not in its sum, we
∑
will omit the limits on the summation sign and write simply fn .
1.8.1 Cauchy's criterion for functional series

Recall the following Theorem knows as Cauchy's convergence criterion
Theorem: A sequence of functions {Fn } converges uniformly on a set S if and

only if for each ε > 0 there is an integer N such that
∥Fn − Fm ∥S < ε if n, m ≥ N. (1.17)
∑
Theorem: A series fn converges uniformly on a set S if and only if for each
ε>0 there is an integer N such that
∥fn + fn+1 + · · · + fm ∥S < ε if m ≥ n ≥ N. (1.18)
∑
Proof : Apply Cauchy's convergence criterion to the partial sums of fn , observing
that
fn + fn+1 + · · · + fm = Fm − Fn−1 .
∑
Theorem: A series fn converges uniformly on a set S if and only if for each
ε>0 there is an integer N such that
∥fn + fn+1 + · · · + fm ∥S < ε if m ≥ n ≥ N. (1.19)
∑
Corollary: If fn converges uniformly on S, then limn→∞ ∥fn ∥S = 0. Setting
m = n.
Remark: The above conditions is necessary but not sucient.
∑∞
Example: We have proved that the series j=0 fj (x), where
fj (x) = xj , j ≥ 0,
is uniformly convergent on any compact subset of (−1, 1) say [−r, r], where 0<r<
1.
Let us apply Cauchy's criterion for functional series, recall that we have
1 − xn+1
Fn (x) = 1 + x + x2 + ... + xn = .
1−x
Consider
1 − xn+1 1 − xm+1 xm+1 − xn+1

|Fm − Fn | = | − |=| |
1−x 1−x 1−x
2|xn+1 |
≤
1 − |x|
2|rn+1 |
≤ .
1 − |r|
We have
2|rn+1 |
∥Fm − Fn ∥[−r,r] ≤ .
1 − |r|
Since
2|rn+1 |
→0 as n → ∞,
1 − |r|
there is an integer N (ε) can be found for which
2|rn+1 |
< ε, when n > N (ε).
1 − |r|
We have
∥Fm − Fn ∥[−r,r] ≤ ε,
∑∞ j
hence by Cauchy's criterion the series j=0 x , is uniformly convergent on [−r, r].
1.8.2 Dominated Series of Real Numbers for Series of Functions

Let {Mn } be a sequence of nonnegative real numbers, and {Fn (x)} a sequence of
functions dened on the set S such that
|Fn (x)| ≤ Mn , ∀x ∈ S and n ∈ N.

∑∞
Then the series of
∑∞
functions n=1 Fn (x) is said to be dominated on S by the series
n=1 Mn .
∑∞
Example: Consider
∑
Fn = 1
x2 +n2
and the series of functions n=1 Fn is dominated
by the series 1/n2 because
1
|Fn | < =: Mn .
n2
∑
We know that 1/n2 < ∞.
1.8.3 Weierstrass M-test/dominated Convergence Test

∑
Theorem The series fn converges uniformly on S if
∥fn ∥S ≤ Mn , n ≥ k, (1.20)
∑
where Mn < ∞.
Proof : From Cauchy's convergence criterion for series of constants, there is for each
ε>0 an integer N such that
Mn + Mn+1 + · · · + Mm < ε if m ≥ n ≥ N.
which, because of (1.20), implies that
∥fn ∥S + ∥fn+1 ∥S + · · · + ∥fm ∥S < ε if m, n ≥ N.

∥fn + fn+1 + · · · + fm ∥S < ε if m, n ≥ N.

∑
Due to Cauchy's criterion, we conclude that fn converges uniformly on S.
Recall the following necessary condition for uniform convergence:

∑
If fn converges uniformly on S, then limn→∞ ∥fn ∥S = 0.
Example: Check the uniform convergence of the following series of functions
∑
• 1
x2 +n2
.
∑ sin nx
• n2
.
Solution: We have
1 1 sin nx 1
≤ 2,
x2 + n2 n n2 ≤ n2 .
Taking Mn = 1/n2 and recalling that
∑ 1
< ∞.
n2
Due to Weierstrass M-test, we can conclude
∑ 1 ∑ sin nx
and
x + n2
2 n2
converge uniformly on (−∞, ∞).
Example: Check the uniform convergence of the series
∑ ∑ ( x )n
fn (x) = .
1+x
Solution: The given series converges uniformly on any set S such that

x

1 + x ≤ r < 1, x ∈ S. (1.21)
For such a set S, we have ∥fn ∥S ≤ rn .

∑ ∑
By Weierstrass's test applies, with Mn = rn < ∞.
Since (1.21) is equivalent to
−r r
≤x≤ , x ∈ S,
1+r 1−r
this means that the series converges uniformly on any compact subset of (−1/2, ∞).
Example: Check the uniform convergence of the series
∑ ∑ ( x )n
fn (x) = .
1+x
Solution: See the solution in video lecture.
∑
Recall: If fn converges uniformly on S, then limn→∞ ∥fn ∥S = 0. The series
does not converge uniformly on S = (−1/2, b) with b < ∞ or on S = [a, ∞) with
a > −1/2, because in these cases ∥fn ∥S = 1 for all n.
∑
Absolute
∑
convergence: A series of functions fn is said toconverge absolutely
on S
∑
if |fn | converges pointwise on S, and converges absolutely uniformly on S if
|fn | converges uniformly on S.
Remarks:
• The condition of absolutely convergence (pointwise or uniform) is stronger
than the usual convergence (pointwise or uniform).
∑
• In our proof of Weierstrass's M-test, we actually proved that fn converges
absolutely uniformly on S.
• Show that if a series converges absolutely uniformly on S, then it converges

uniformly on S.
Theorem: The series

∞
∑
fn gn
n=k
∑
converges uniformly on S if {fn } converges uniformly to zero on S, (fn+1 − fn )
converges absolutely uniformly on S, and
∥gk + gk+1 + · · · + gn ∥S ≤ M, n ≥ k, (1.22)
for some constant M.
Proof : Let
Gn = gk + gk+1 + · · · + gn ,
∑∞
and consider the partial sums of n=k fn gn :
Hn = fk gk + fk+1 gk+1 + · · · + fn gn . (1.23)
By substituting g k = Gk and gn = Gn − Gn−1 , n ≥ k + 1,

into (1.23), we obtain
Hn = fk Gk + fk+1 (Gk+1 − Gk ) + · · · + fn (Gn − Gn−1 ).
Which we rewrite as
Hn = (fk − fk+1 )Gk + (fk+1 − fk+2 )Gk+1 + · · · + (fn−1 − fn )Gn−1 + fn Gn ,

or
Hn = Jn−1 + fn Gn , (1.24)
where
Jn−1 = (fk − fk+1 )Gk + (fk+1 − fk+2 )Gk+1 + · · · + (fn−1 − fn )Gn−1 . (1.25)
That is, {Jn } is the sequence of partial sums of the series
∞
∑
(fj − fj+1 )Gj . (1.26)
j=k
From (1.22) and the denition of Gj ,

m
∑ ∑m
[fj (x) − fj+1 (x)]Gj (x) ≤ M |fj (x) − fj+1 (x)|, x ∈ S,

j=n j=n
so
∑ ∑
m m
(fj − fj+1 )Gj ≤ M |fj − fj+1 |
.
j=n j=n
S S
∑
Now suppose that ε > 0. Since (fj − fj+1 ) converges absolutely uniformly on S ,
Cauchy's convergence criterion implies that there is an integer N such that the right
side of the last inequality is less than ε if m ≥ n ≥ N . The same is then true of the
left side, so Cauchy's convergence criterion implies that (1.26) converges uniformly
on S.
We have now shown that {Jn } as dened in (1.25) converges uniformly to a limit
function J on S. Returning to (1.24), we see that
Hn − J = Jn−1 − J + fn Gn .
Hence, we have
∥Hn − J∥S ≤ ∥Jn−1 − J∥S + ∥fn ∥S ∥Gn ∥S

≤ ∥Jn−1 − J∥S + M ∥fn ∥S .
Since{Jn−1 − J} and {fn } converge uniformly to zero on S , it now follows that

limn→∞ ∥Hn − J∥S = 0. Therefore, {Hn } converges uniformly on S .
∑∞
Corollary: The series n=k fn gn converges uniformly on S if
fn+1 (x) ≤ fn (x), x ∈ S, n ≥ k,
{fn } converges uniformly to zero on S, and
∥gk + gk+1 + · · · + gn ∥S ≤ M, n ≥ k,
for some constant M.
Example: Consider the series

∞
∑ sin nx
n
n=1
with fn = 1/n (constant), gn (x) = sin nx, and
Gn (x) = sin x + sin 2x + · · · + sin nx.
We have
1
|Gn (x)| ≤ , n ≥ 1, n ̸= 2kπ (k = integer).
| sin(x/2)|
Therefore, {∥Gn ∥S } is bounded, and the series converges uniformly on any set S on
which sin x/2 is bounded away from zero.
Example: For example, if 0 < δ < π , then

x
δ
sin ≥ sin
2 2
if x is at least δ away from any multiple of 2π ; hence, the series converges uniformly
on
∞
∪
S= [2kπ + δ, 2(k + 1)π − δ].
k=−∞
Since
∑ sin nx

n = ∞, x ̸= kπ.
This result cannot be obtained from Weierstrass's test.
Example: The series

∞
∑ (−1)n
n + x2
n=1
satises the hypotheses of Corollary on (−∞, ∞), with
1
fn (x) = , gn = (−1)n , G2m = 0, and G2m+1 = −1.
n + x2
Therefore, the series converges uniformly on (−∞, ∞). This result cannot be ob-
tained by Weierstrass's test, since
∑ 1
=∞
n + x2
for all x.
Recall the following result:

Theorem: If {Fn } converges uniformly to F on S and each Fn is continuous at a

point x0 in S, then so is F. Similar statements hold for continuity from the right
and left.
∑∞
Theorem: If n=k fn converges uniformly to F on S and each fn is continuous at
a point x0 in S, then so is F. Similar statements hold for continuity from the right
and left.
Proof : See Lecture.
∑∞
Recall the following: Theorem: If n=k fn converges uniformly to F on S and
each fn is continuous at a point x0 in S, then so is F. Similar statements hold for
continuity from the right and left.
Example: Recall, we have proved that the series
∞ (
∑ )n
x
F (x) =
1+x
n=0
converges uniformly on every compact subset of (−1/2, ∞).

Since the terms of the series are continuous on every such subset, implies that
F is also.
In fact, we can state a stronger result: F is continuous on (−1/2, ∞), since

every point in (−1/2, ∞) lies in a compact subinterval of (−1/2, ∞).
Example: Show that the function
∞
∑ sin nx
G(x) =
n
n=1
is continuous except perhaps at xk = 2kπ (k = integer).
∑∞ sin nx
We have seen that the series n=1 is uniformly convergent by applying Dirich-
n
let's Test for Uniform Convergence except at xk = 2kπ (k = integer).
Example: The function

∞
∑ 1
H(x) = (−1)n
n + x2
n=1
is continuous for all x.
Theorem: Suppose that {Fn } is a sequence of Riemann integrable functions dened

on an interval [a, b]. If {Fn } converges uniformly on [a, b] to F, then F is Riemann
integrable on [a, b], and
∫b ∫b
lim Fn (x)dx = F (x)dx.
n→∞
a a
For each t ∈ [a, b]

∫t
Fn (x)dx,
a
converges uniformly on [a, b] to

∫t
F (x)dx.
a
Proof : We need to show that the function F is integrable on [a, b].

Observe that the following statements holds:
• Fn is bounded, because each Fn is integrable on [a, b].
• F is bounded, because
|F (x)| ≤ |Fn (x) − F (x)| + |Fn (x)| ≤ δn + |Fn (x)|,
where δn = supx∈[a,b] |Fn (x) − F (x)|.
• Since Fn converges uniformly to F, for every ε > 0, there exists an N such

that
ε
|Fn (x) − F (x)| < , for all x ∈ [a, b], n > N.
3(b − a)
Also, Fn is integrable, there exists a partition P of [a, b] such that
ε
S(P, Fn ) − s(P, Fn ) < .
3
For each x ∈ [a, b] with n=N
ε
|Fn (x) − F (x)| < , for all x ∈ [a, b], n > N,
3(b − a)
implies that
ε ε
Fn (x) − < F (x) < Fn (x) + .
3(b − a) 3(b − a)
Therefore,
ε ε
s(P, Fn ) − < s(P, F ) ≤ S(P, F ) < S(P, Fn ) +
3 3
Hence F is integrable. Finally, for n ≥ N and for each t ∈ [a, b], we have
∫ t ∫t ∫t

Fn (x)dx − F (x)dx ≤ |Fn (x) − F (x)|dx

a a a
ε(b − a)
≤ , for all x ∈ [a, b], n > N.
3(b − a)
Remark: The limit of a uniformly convergent series of integrable functions is inte-

grable, and so term-by-term integration is permissible for such a series.
∑∞
Theorem: Suppose that n=k fn converges uniformly to F on S = [a, b]. Assume
that F and fn , n ≥ k, are integrable on [a, b]. Then
∫ b ∞ ∫
∑ b
F (x) dx = fn (x) dx.
a n=k a
∑∞
We say in this case that n=k fn can be integrated term by term over [a, b].
Example: Consider the {Fn } dened by
x
Fn (x) = , x ∈ [a, b] ⊂ R.
1 + nx2
∑
Then Weieretrass's M-test shows that Fn converges uniformly on [a, b]
Consequently, term-by-term integration is permissible in this series.
Example: Consider the following
∑ ∞
1
= xn , −1 < x < 1.
1−x
n=0
The series converges uniformly, and the limit function is integrable on any closed
subinterval [a, b] of (−1, 1).
Hence,
∫ b ∑ ∞ ∫ b
dx
= xn dx.
a 1−x
n=0 a
Consequently,
∞ n+1
∑ b − an+1
log(1 − a) − log(1 − b) = .
n+1
n=0
Remark: We have seen that
∞ n+1
∑ b − an+1
log(1 − a) − log(1 − b) = .
n+1
n=0
Letting a=0 and b=x yields
∞
∑ xn+1
log(1 − x) = − , −1 < x < 1.
n+1
n=0
Example: Evaluate the

∞ ∫
∑
1
x(ex − 1)e−nx dx.

n=1 0
Solution: The sequence of partial sum is
∑
n
Fn = x(ex − 1)e−kx dx.
k=1
Observe that Fn (0) = 0 and for x>0
e−x (1 − e−nx )
Fn (x) = x(ex − 1) ,
1 − e−x
Fn (x) = x(1 − e−nx ).

∞ ∫
∑
1

n=1 0
Solution: For the function xe−nx , we have seen that it attains its maximum at
x = 1/n, we have
∥Fn (x) − x∥ = sup |Fn (x) − x|
x≥0
1
∥Fn (x) − x∥ = sup |xe−nx | = .
x≥0 en
So, as n → ∞, we have ∥Fn (x) − x∥ → 0.

∞ ∫
∑
1

n=1 0
Solution: The series of functions
∞
∑
x(ex − 1)e−nx dx,
n=1
converges uniformly to F (x) = x.

Applying the theorem of interchange of sum and integral sign, we can conclude
that
∞ ∫
∑
1 ∫1 ∑
∞ ∫1
−nx −nx
x(e − 1)e
x
dx = x(e − 1)e
x
dx = xdx.
n=1 0 0 n=1 0
Example: Consider
x
Fn (x) = , x ∈ R.
1 + nx2
x |x| 1
|Fn (x)| = ≤ √ = √ .
1 + nx2 2 n|x| 2 n
Fn (x) is uniformly convergent to F (x) = 0 on R. We have
1 − nx2
Fn′ (x) =
(1 + nx2 )2
When x = 0, we have limn→∞ Fn′ (x) = 0 and for x ̸= 0 limn→∞ Fn′ (x) = 1.
Remark: What we have observed in this example is:
• We have a sequence of dierentiable functions {Fn } dened on S.
• Fn converges uniformly to F on S.
• F is dierentiable on S.
• There exists x∈S with F ′ (x) ̸= limn→∞ Fn′ (x), because Fn′ (0) → 1 ̸= F ′ (0).
Thus, even if the limit of a uniformly convergent sequence (respec-

tively series) of dierentiable functions on S is dierentiable on S , it may
happen that the derivative of the limit is not the limit of the sequence
(respectively sequence of partial sums) of derivatives of the dierentiable
functions.
Theorem: Suppose that fn is a sequence of functions such that:
• fn is continuously dierentiable on [a, b] for each n ≥ k, i.e., fn ∈ C 1 [a, b].

∑∞
• n=k fn (x0 ) converges for some x0 in [a, b].
∑∞ ′
• n=k fn converges uniformly on [a, b].
∑∞
Then n=k fn converges uniformly on [a, b] to a dierentiable function F, such
that
∞
∑
F ′ (x) = fn′ (x), a < x < b,
n=k
∞
∑ ∞
∑
while F ′ (a+) = fn′ (a+) and F ′ (b−) = fn′ (b−).
n=k n=k
Proof : Since fn′ is uniformly convergent to g on any closed interval contained in

[a, b], say in an interval with endpoints x0 and x, x ∈ [a, b]. Thus, for all x ∈ [a, b],
we have
∫x ∫x
g(t)dt = lim fn′ (t)dt.
n→∞
x0 x0
Recall the fundamental theorem of calculus, we have
∫x
g(t)dt = lim (fn (x) − fn (x0 )).
n→∞
x0
Recall the limn→∞ fn (x0 ) exists (given hypothesis), we can obtain
∫x
g(t)dt + lim fn (x0 ) = lim fn (x), on [a, b].
n→∞ n→∞
x0
The above convergence is uniform. By setting F (x) = limn→∞ fn (x), we have
∫x
g(t)dt + lim fn (x0 ) = F (x), on [a, b].
n→∞
x0
Now, g , being the limit of a uniformly convergent sequence of continuous functions

on [a, b], is continuous on [a, b].
∫x
Recall the second fundamental theorem of calculus with G(x) = g(t)dt is
x0
dierentiable and G′ (x) = g(x) on [a, b].
Therefore, we have
F ′ (x) = g(x), F ′ (x) = lim fn′ (x), on [a, b].

n→∞
∑∞
Remark: The series n=k fn can be dierentiated term by term on [a, b].
How to apply this result?

∑∞
• We rst verify that n=k fn (x0 ) converges for some x0 in [a, b].
∑∞
• Then dierentiate n=k fn term by term. If the resulting series converges
uniformly. Then term by term dierentiation was legitimate.
Example: The series

∞
∑ 1 x
(−1)n cos (1.27)
n n
n=1
converges at x0 = 0. Dierentiating term by term yields the series
∞
∑ 1 x
(−1)n+1 2
sin (1.28)
n n
n=1
of continuous functions. This series converges uniformly on (−∞, ∞), by Weier-

strass's test. Consequently, the series (1.27) converges uniformly on every nite
interval to the dierentiable function
∞
∑ 1 x
F (x) = (−1)n cos , −∞ < x < ∞,
n n
n=1
∑∞
1 x
F ′ (x) = (−1)n+1 2
sin , −∞ < x < ∞.
n n
n=1
∞
∑ xn x2 x3
E(x) = =1+x+ + + ··· . (1.29)
n! 2! 3!
n=0
The series converges uniformly on every interval [−r, r] by Weierstrass's test, because
|x|n rn
≤ , |x| ≤ r,
n! n!
∑ rn
< ∞
n!
for all r, by the ratio test.
Dierentiating the right side of (1.30) term by term yields the series
∞
∑ ∞
∑
xn−1 xn
= ,
(n − 1)! n!
n=1 n=0
which is the same as (1.30).
∞
∑ xn x2 x3
E(x) = =1+x+ + + ··· . (1.30)
n! 2! 3!
n=0
1.9. Power Series 30
Therefore, the dierentiated series is also uniformly convergent on [−r, r] for every
r, so the term by term dierentiation is legitimate and
E ′ (x) = E(x), −∞ < x < ∞.
This is not surprising if you recognize that E(x) = ex .
Remark: Failure to verify that the given series converges at some point can lead
to erroneous conclusions.
Example: For example, dierentiating
∞
∑ x
cos (1.31)
n
n=1
term by term.
We have
∞
∑ 1 x
− sin .
n n
n=1
Since
1
sin ≤ |x| ≤ r ,
x
|x| ≤ r,
n n n2 n2
∑
and 1/n2 < ∞. which converges uniformly on [−r, r] for every r,
We cannot conclude from this that (1.31) converges uniformly on [−r, r]. In fact,
it diverges for every x.
1.9 Power Series

An innite series of the form
∞
∑
an (x − x0 )n , (1.32)
n=0
where x0 and a0 , a1 , . . . , are constants, is called a power series in x − x0 . If x0 = 0

then power series becomes
∞
∑
an xn .
n=0
∑
Theorem: The radius of convergence of an (x − x0 )n is given by

1 an+1
= lim
R n→∞ an
if the limit exists in the extended real number system.

∑∞
Theorem: For the power series n=0 an (x − x0 )n , dene R in the extended real
numbers by
1
= lim sup |an |1/n . (1.33)
R n→∞
Theorem: A power series
∞
∑
f (x) = an (x − x0 )n
n=0
with positive radius of convergence R is continuous and dierentiable in its interval

of convergence, and its derivative can be obtained by dierentiating term by term;
that is,
∞
∑
f ′ (x) = nan (x − x0 )n−1 , (1.34)
n=1
which can also be written as
∞
∑
f ′ (x) = (n + 1)an+1 (x − x0 )n . (1.35)
n=0
This series also has radius of convergence R.
Proof : Since
lim sup((n + 1)|an |)1/n = lim sup(n + 1)1/n |an |1/n

n→∞ n→∞
( )( )
= lim (n + 1) 1/n
lim sup |an |1/n
n→∞ n→∞
[ ( )]
log(n + 1)
= lim exp
n→∞ n
( )
e0 1
lim sup |an |1/n = = ,
n→∞ R R
the radius of convergence of the power series obtained by term by term dierentiation
is R. Therefore, the power series in
∞
∑
f ′ (x) = (n + 1)an+1 (x − x0 )n ,
n=0
converges uniformly in every interval [x0 − r, x0 + r] such that 0 < r < R.

The term by term dierentiation is valid for the power series and the series
∞
∑
f ′ (x) = (n + 1)an+1 (x − x0 )n ,
n=0
converges uniformly for all x in (x0 − R, x0 + R).

∞
∑
f (x) = an (x − x0 )n
n=0
with positive radius of convergence R has derivatives of all orders in its interval of
convergence, which can be obtained by repeated term by term dierentiation. That
is,
∞
∑
f (k) (x) = n(n − 1) · · · (n − k + 1)an (x − x0 )n−k . (1.36)
n=k
The radius of convergence of each of these series is R.
Proof : ∞
∑
f (k)
(x) = n(n − 1) · · · (n − k + 1)an (x − x0 )n−k .
n=k
The proof is by induction. The assertion is true for k = 1, by the Theorem we

proved in previous module.
Suppose that it is true for some k ≥ 1. By shifting the index of summation, we
can write
∞
∑
f (k) (x) = (n + k)(n + k − 1) · · · (n + 1)an+k (x − x0 )n , |x − x0 | < R.
n=0
Dening
bn = (n + k)(n + k − 1) · · · (n + 1)an+k . (1.37)
We rewrite this as
∞
∑
f (k)
(x) = bn (x − x0 )n , |x − x0 | < R.
n=0
By Theorem of term by term dierentiation of power series, we can dierentiate

this series term by term to obtain
∞
∑
f (k+1)
(x) = nbn (x − x0 )n−1 , |x − x0 | < R.
n=1
Substituting from (1.37) for bn for |x − x0 | < R yields
∞
∑
f (k+1)
(x) = (n + k)(n + k − 1) · · · (n + 1)nan+k (x − x0 )n−1 .
n=1
Shifting the summation index yields
∞
∑
f (k+1)
(x) = n(n − 1) · · · (n − k)an (x − x0 )n−k−1 , |x − x0 | < R,
n=k+1
which is (1.42) with k replaced by k + 1. This completes the induction.
Example: We have proved that
∑ ∞
1
= xn , |x| < 1.
1−x
n=0
Repeated dierentiation yields
∞
∑
k!
= n(n − 1) · · · (n − k + 1)xn−k
(1 − x)k+1
n=k
∑∞
= (n + k)(n + k − 1) · · · (n + 1)xn , |x| < 1,
n=0
∑∞ ( )
1 n+k n
= x , |x| < 1.
(1 − x)k+1 k
n=0
Example: Show that the series
∞
∑ ∞
∑
x2n+1
n x2n
S(x) = (−1) and C(x) = (−1)n
(2n + 1)! (2n)!
n=0 n=0
converges for all x.

Dierentiating yields
∞
∑
′ xn
S (x) = (−1)n = C(x)
(2n)!
n=0
and
∞
∑ ∑ ∞
′ x2n−1 x2n+1
C (x) = (−1) n
=− (−1)n = −S(x).
(2n − 1)! (2n + 1)!
n=1 n=0
These results should not surprise you if you recall that
S(x) = sin x and C(x) = cos x.
Theorem: If
∞
∑
f (x) = an (x − x0 )n , |x − x0 | < R,
n=0
then
f (n) (x0 )
an = .
n!
Proof : We have
∞
∑
f (k)
(x) = n(n − 1) · · · (n − k + 1)an (x − x0 )n−k .
n=k
Setting x = x0 in the above equation yields
f (k) (x0 ) = k!ak .
Theorem: If
∞ ∞
∑ ∑
an (x − x0 )n = bn (x − x0 )n (1.38)
n=0 n=0
for all x in some interval (x0 − r, x0 + r), then
an = bn , n ≥ 0. (1.39)
∑∞ ∑∞
Proof : Let f (x) = n=0 an (x − x0 )n and g(x) = n=0 bn (x − x0 )n .
From previous result, we have
f (n) (x0 ) g (n) (x0 )

an = and bn = . (1.40)
n! n!
From (1.38), f =g in (x0 − r, x0 + r). Therefore,
f (n) (x0 ) = g (n) (x0 ), n ≥ 0.
This and (1.40) imply (1.39).
Theorem (Recall the following): For the power series, dene R in the extended
real numbers by
1
= lim sup |an |1/n . (1.41)
R n→∞
In particular, R = 0 if lim supn→∞ |an |1/n = ∞, and R = ∞ if lim supn→∞ |an |1/n =
0.
Then the power series converges
1. only for x = x0 if R = 0;
2. for all x if R = ∞, and absolutely uniformly in every bounded set;
3. for x in (x0 − R, x0 + R) if 0 < R < ∞, and absolutely uniformly in every

closed subset of this interval.
Remark: |x − x0 | > R. No general statement can be made

The series diverges if
concerning convergence at the endpoints x = x0 + R and x = x0 − R : the series
may converge absolutely or conditionally at both, converge conditionally at one and
diverge at the other, or diverge at both.
∑∞
Theorem (Recall the following): Suppose that n=k fn converges uniformly to F
on S = [a, b]. Assume that F and fn , n ≥ k, are integrable on [a, b].
Then
∫ b ∞ ∫
∑ b
F (x) dx = fn (x) dx.
a n=k a
Theorem: If x1 and x2 are in the interval of convergence of
∞
∑
f (x) = an (x − x0 )n ,
n=0
then
∫ ∞
∑
x2
an [ ]
f (x) dx = (x2 − x0 )n+1 − (x1 − x0 )n+1 ;
x1 n+1
n=0
that is, a power series may be integrated term by term between any two points in
its interval of convergence.
Proof : See Lecture.
Some questions related to Power Series.
• For what values of x a given power series converges.
• We discussed, what are the properties of its sum.
• What properties guarantee that a given function f can be represented as the

sum of a convergent power series in x − x0 ?
Recall the following:
∞
∑
f (x) = an (x − x0 )n
n=0
with positive radius of convergence R has derivatives of all orders in its interval of
convergence, which can be obtained by repeated term by term dierentiation; thus,
∞
∑
f (k) (x) = n(n − 1) · · · (n − k + 1)an (x − x0 )n−k . (1.42)
n=k
1.10. The Taylor's Series 36
The radius of convergence of each of these series is R. If
∞
∑
f (x) = an (x − x0 )n , |x − x0 | < R,
n=0
then
f (n) (x0 )
an = .
n!
1.10 The Taylor's Series

The only power series in x − x0 that can possibly converge to f in such a neighbor-
hood is
∞
∑ f (n) (x0 )
(x − x0 )n . (1.43)
n!
n=0
This is called the Taylor series of f about x0 (also, the Maclaurin series of f, if
x0 = 0). The mth partial sum of (1.43) is the Taylor polynomial
∑
m
f (n) (x0 )
Tm (x) = (x − x0 )n ,
n!
n=0
Remark: The Taylor series of an innitely dierentiable function f may converge

to a sum dierent from f.
Example: Consider the function

{
e−1/x , x ̸= 0,
2
f (x) =
0, x = 0.
the function f is innitely times dierentiable on (−∞, ∞) and f (n) (0) = 0 for
n ≥ 0. So its Maclaurin series is identically zero.
Taylor's theorem: If f is innitely dierentiable on (a, b) and x and x0 are in

(a, b) then, for every integer n ≥ 0,
f (n+1) (cn )
f (x) − Tn (x) = (x − x0 )n−1 , (1.44)
(n + 1)!
where cn is between x and x0 .
Therefore,
∞
∑ f (n) (x0 )
f (x) = (x − x0 )n
n!
n=0
for an x in (a, b) if and only if
f (n+1) (cn )
lim (x − x0 )n+1 = 0.
n→∞ (n + 1)!
Remark: It is not always easy to check this condition, because the sequence {cn }
is usually not precisely known, or even uniquely dened; however, the next theorem
is suciently general to be useful.
Theorem: Suppose that f is innitely dierentiable on an interval I and
rn (n)
lim ∥f ∥I = 0. (1.45)
n→∞ n!
Then, if x0 ∈ I 0 , the Taylor series
∞
∑ f (n) (x0 )
(x − x0 )n
n!
n=0
converges uniformly to f on
Ir = I ∩ [x0 − r, x0 + r].
Proof : We know that
f (n+1) (cn )
f (x) − Tn (x) = (x − x0 )n−1 ,
(n + 1)!
rn+1 rn+1
∥f − Tn ∥Ir ≤ ∥f (n+1) ∥Ir ≤ ∥f (n+1) ∥I ,
(n + 1)! (n + 1)!
so (1.45) implies the conclusion.
Example: If f (x) = sin x, then ∥f (k) ∥(−∞,∞) = 1, k ≥ 0. We know that
rn
lim = 0, 0<r<∞
n→∞ n!
holds for all r.

Since
f (2m) (0) = 0 and f (2m+1) (0) = (−1)m , m ≥ 0,
Apply the previous theorem, with I = (−∞, ∞), x0 = 0, and r arbitrary. We have
the the well known series expansion of sin x, that is,
∞
∑ x2n+1
sin x = (−1)n , −∞ < x < ∞,
(2n + 1)!
n=0
and the convergence is uniform on bounded sets.

Example: A similar argument shows that
∞
∑ x2n
cos x = (−1)n , −∞ < x < ∞,
(2n)!
n=0
with uniform convergence on bounded sets.
Example: If f (x) = ex , then f (k) (x) = ex and ∥f (k) ∥I = er , k ≥ 0, if I = [−r, r].

Since
rn r
lim e = 0.
n→∞ n!
we conclude that
∞
∑ xn
ex = , −∞ < x < ∞,
n!
n=0
with uniform convergence on bounded sets.
Example: If f (x) = (1 + x)q , then
( ) ( )
f (n) (x) q f (n) (0) q
= (1 + x)q−n , so = . (1.46)
n! n n! n
The Maclaurin series

∞ ( )
∑ q n
x
n
n=0
is called the binomial series . We saw in Analysis I that this series equals (1 + x)q
for all x if q is a nonnegative integer.
Example: We will now show that if q is an arbitrary real number, then
∞ ( )
∑ q n
x = f (x) = (1 + x)q , 0 ≤ x < 1. (1.47)
n
n=0
Since ( )/( )
q q q − n
lim = lim = 1,
n→∞ n + 1 n n→∞ n + 1
the radius of convergence of the series in (1.47) is 1.
From (1.46),
( )
∥f (n) ∥[0,1]
q q

≤ [max(1, 2 )] , n ≥ 0.
n! n
Example: Therefore, if 0 < r < 1,

( )
rn (n) q n
lim sup ∥f ∥[0,1] ≤ [max(1, 2 )] lim
q r = 0,
n→∞ n! n→∞ n
where the last equality follows from the absolute convergence of the series in (1.47)
on (−1, 1).
Theorem: If
∞
∑
f (x) = an (x − x0 )n , |x − x0 | < R1 , (1.48)
n=0
∞
∑
g(x) = bn (x − x0 )n , |x − x0 | < R2 , (1.49)
n=0
and α and β are constants, then
∞
∑
αf (x) + βg(x) = (αan + βbn )(x − x0 )n , |x − x0 | < R,
n=0
where R ≥ min{R1 , R2 }.
Proof : See the video lectures.
Recall the following theorem:

∑∞ ∑∞
Theorem: n=0 an∑
If and n=0 bn∑
converge absolutely to sums A and B, then
∞ ∞
the Cauchy product of a
n=0 n and n=0 bn converges absolutely to AB.
Theorem: If f and g are given by power series
∞
∑
f (x) = an (x − x0 )n , |x − x0 | < R1 ,
n=0
∞
∑
g(x) = bn (x − x0 )n , |x − x0 | < R2 ,
n=0
then
∞
∑
f (x)g(x) = cn (x − x0 )n , |x − x0 | < R, (1.50)
n=0
∑n ∑
n
cn = ar bn−r = an−r br
r=0 r=0
and R ≥ min{R1 , R2 }.
Proof : Suppose that R1 ≤ R2 .

Since the series
∞
∑
f (x) = an (x − x0 )n , |x − x0 | < R1 ,
n=0
∞
∑
g(x) = bn (x − x0 )n , |x − x0 | < R2 ,
n=0
converge absolutely to f (x) and g(x).

If |x − x0 | < R1 , their Cauchy product converges to f (x)g(x) if |x − x0 | < R1 ,
by product of series.
The nth term of this product is
( )
∑
n ∑
n
ar (x − x0 ) bn−r (x − x0 )
r n−r
= ar bn−r (x − x0 )n = cn (x − x0 )n .
r=0 r=0
Example: If
∑ ∞
1
f (x) = = xn , |x| < 1,
1−x
n=0
∞
∑
g(x) = b n xn , |x| < R,
n=0
∑∞
g(x)
= sn xn , |x| < min{1, R},
1−x
n=0
where
sn = (1)b0 + (1)b1 + · · · + (1)bn

= b0 + b1 + · · · + bn .
Example: We have already discussed
∞ ( )
∑ p n
(1 + x)p = x , |x| < 1.
n
n=0
Also
∞ ( )
∑ q n
(1 + x) = q
x , |x| < 1.
n
n=0
Since
∞ (
∑ )
p q p+q p+q n
(1 + x) (1 + x) = (1 + x) = x ,
n
n=0
∑∞ n
while the Cauchy product is n=0 cn x , with
n ( )(
∑ )
p q
cn = .
r n−r
r=0
Product of power series implies that

( )
p+q
cn = .
n
This yields the identity
( ) ∑n ( )( )
p+q p q
= ,
n r n−r
r=0
valid for all p and q. The quotient
h(x)
f (x) = (1.51)
g(x)
of two power series
∞
∑
h(x) = cn (x − x0 )n , |x − x0 | < R1 ,
n=0
∑∞
g(x) = bn (x − x0 )n , |x − x0 | < R2 ,
n=0
can be represented as a power series
∞
∑
f (x) = an (x − x0 )n (1.52)
n=0
with a positive radius of convergence, provided that
b0 = g(x0 ) ̸= 0.
This is surely plausible. Since g(x0 ) ̸= 0 and g is continuous near x0 , the denomina-
tor of (1.51) diers from zero on an interval about x0 . Therefore,f has derivatives
of all orders on this interval, because g and h do.
Since
f (x)g(x) = h(x),
The result about the product of Power series implies that
∑
n
ar bn−r = cn , n ≥ 0.
r=0
Solving these equations successively yields
c0
a0 = ,
b0
( )
1 ∑
n−1
an = cn − bn−r ar , n ≥ 1.
b0
r=0
Remark: It is not worthwhile to memorize these formulas. Rather, it is usually

better to view the procedure as follows: Multiply the series f (with unknown co-
ecients) and g according to the procedure of Theorem ??, equate the resulting
coecients with those of h, and solve the resulting equations successively for a0 , a1 ,
....
Example: Suppose that we wish to nd the coecients in the Maclaurin series
tan x = a0 + a1 x + a2 x2 + · · · .
We rst observe that since tan x is an odd function, its derivatives of even order
vanish at x0 = 0, so a2m = 0, m ≥ 0. Therefore,
tan x = a1 x + a3 x3 + a5 x5 + · · · .
Since
sin x
tan x = ,
cos x
it follows from series of sin x and cos x that
x3 x5
x− + 120 + · · ·
a1 x + a3 x + a5 x + · · · =
3 5 6
2 4
1 − x2 + x24 + · · ·
so
( )
x2 x4 x3 x5
(a1 x + a3 x + a5 x + · · · ) 1 −
3 5
+ + ··· = x − + + ··· ,
2 24 6 120
or
( a1 ) 3 ( a3 a1 ) 5 x3 x5
a1 x + a3 − x + a5 − + x + ··· = x − + + ··· .
2 2 24 6 120
Comparing coecients of like powers of x on the two sides of this equation must be
equal; hence,
a1 = 1, a3 − a1
2 = − 16 , a5 − a3
2 + a1
24 = 1
120 ,
(1)
a1 = 1, a3 = − 16 + 21 (1) = 1
3, a5 = 1
120 + 1
2 3 − 1
24 (1) = 2
15 .
Therefore,
x3 2
tan x = x + + x5 + · · · .
3 15
Example: To nd the reciprocal of the power series
∞
∑ xn
g(x) = 1 + ex = 2 + ,
n!
n=1
we let h=1 in (1.51). If

∑ ∞
1
= an x n ,
g(x)
n=0
then
( )
x2 x3
1 = (a0 + a1 x + a2 x + a3 x + · · · ) 2 + x +
2 3
+ + ···
2 6
(a )
0
= 2a0 + (a0 + 2a1 )x + + a1 + 2a2 x2
(a 2)
0 a1
+ + + a2 + 2a3 x3 + · · · .
6 2
From Corollary,
2a0 = 1,
a0 + 2a1 = 0,
a0
+ a1 + 2a2 = 0,
2
a0 a1
+ + a2 + 2a3 = 0.
6 2
Solving these equations successively yields
1
a0 = ,
2
a0 1
a1 = − =− ,
2 4 ( )
1 ( a0 ) 1 1 1
a2 = − + a1 = − − = 0,
2 2 2 4 4
( )
1 ( a0 a1 ) 1 1 1 1
a3 = − + + a2 = − − +0 = ,
2 6 2 2 12 8 48
so
1 1 x x3
= − + + ··· .
1 + ex 2 4 48
Example: To nd the reciprocal of
∞
∑ xn
g(x) = ex = , (1.53)
n!
n=0
we again let h=1 in (1.51). If
∞
∑
(ex )−1 = an x n ,
n=0
then ( ∞
)( ∞
) ∞
∑ ∑ xn ∑
n
1= an x = cn xn ,
n!
n=0 n=0 n=0
1.11. The Abel's Theorem 44
where
∑
n
ar
cn = .
(n − r)!
r=0
We have, c0 = a0 = 1 and cn = 0 if n ≥ 1; hence,
∑
n−1
ar
an = − , n ≥ 1. (1.54)
(n − r)!
r=0
Solving these equations successively for a0 , a1 , . . . yields
1
a1 = − (1.32) = −1,
1!
[ ]
1 1 1
a2 = − (1) + (−1) = ,
2! 1! 2
[ ( )]
1 1 1 1 1
a3 = − (1) + (−1) + =−
3! 2! 1! 2 6
[ ( ) ( )]
1 1 1 1 1 1 1
a4 = − (1) + (−1) + + − = .
4! 3! 2! 2 1! 6 24
From this, we see that

(−1)k
ak =
k!
for 0≤k≤4 and are led to conjecture that this holds for all k. To prove this by
induction, we assume that it is so for 0≤k ≤n−1 and compute from (1.54):
∑n−1 (−1)r
an = − 1
r=0 (n−r)! r!
∑n−1 ( )
= − n!
1
r=0 (−1)
r n
r
(−1)n
= n! .
Thus, we have shown that
∞
∑ xn
(ex )−1 = (−1)n .
n!
n=0
Since this is precisely the series that results if x is replaced by −x in (1.53), we have
veried a fundamental property of the exponential function: that
(ex )−1 = e−x .
This also follows from Example ??.
1.11 The Abel's Theorem

Theorem: Let f be dened by a power series with nite radius of convergence R.
∑∞
• If n=0 an R
n converges, then
∞
∑
lim f (x) = an R n .
x→(x0 +R)−
n=0
∑∞
• If n=0 (−1)
na
nR
n converges, then
∞
∑
lim f (x) = (−1)n an Rn .
x→(x0 −R)+
n=0
Proof : Let
∞ ∞
∑ ∑
n
g(y) = bn y , bn = s (nite).
n=0 n=0
We will show that
lim g(y) = s. (1.55)
y→1−
We have
∞
∑
g(y) = (1 − y) sn y n , (1.56)
n=0
where
sn = b0 + b1 + · · · + bn .
Since
∑ ∞ ∞
∑
1
= yn and therefore 1 = (1 − y) yn, |y| < 1, (1.57)
1−y
n=0 n=0
we can multiply through by s and write
∞
∑
s = (1 − y) sy n , |y| < 1.
n=0
Subtracting this from (1.56) yields

∞
∑
g(y) − s = (1 − y) (sn − s)y n , |y| < 1.
n=0
If ε > 0, choose N so that
|sn − s| < ε if n ≥ N + 1.
Then, if 0 < y < 1,
∑
N ∞
∑
|g(y) − s| ≤ (1 − y) |sn − s|y + (1 − y)
n
|sn − s|y n
n=0 n=N +1
∑
N ∞
∑
< (1 − y) |sn − s|y + (1 − y)εy
n N +1
yn
n=0 n=0
∑N
< (1 − y) |sn − s| + ε,
n=0
because of the second equality in (1.57).

Therefore,
|g(y) − s| < 2ε
if
∑
N
(1 − y) |sn − s| < ε.
n=0
To obtain rst part of the theorem from this, let bn = an Rn and g(y) = f (x0 + Ry);
to obtain second part, let bn = (−1)n a n
n R and g(y) = f (x0 − Ry).
Example: The series

∑ ∞
1
f (x) = = (−1)n xn
1+x
n=0
diverges at x = 1, while limx→1− f (x) = 1/2.
This shows that the converse of Abel's theorem is false.
Integrating the series term by term yields
∞
∑ xn+1
log(1 + x) = (−1)n , |x| < 1,
n+1
n=0
where the power series converges at x = 1. The Abel's theorem implies that
∞
∑ (−1)n+1
log 2 = .
n+1
n=0
Example: If q ≥ 0, the binomial series
∞ ( )
∑ q n
x
n
n=0
converges absolutely for x = ±1. This is obvious if q is a nonnegative integer, and

it follows from Raabe's test for other positive values of q, since
( )/( )
an+1 q q n − q
= = , n > q,
an n + 1 n n+1
and
( ) ( )
an+1 n−q
lim n −1 = lim n −1
n→∞ an n→∞ n+1
n
= lim (−q − 1) = −q − 1.
n→∞ n + 1
Therefore, Abel's theorem imply that
∞ ( )
∑ ∞
∑ ( )
q q
= 2q and (−1)n = 0, q ≥ 0.
n n
n=0 n=0
1.12. Pointwise and Uniform Bounded Functions 47
1.12 Pointwise and Uniform Bounded Functions

A sequence of functions {Fn } on the set S is said to be pointwise bounded on S if
the sequence of functions is bounded for every x ∈ S, that is, if there exists a nite
valued function ϕ(x) dened on S such that
|Fn (x)| < ϕ(x), x ∈ S, n = 1, 2, 3, ....
We say that {Fn } is uniformly bounded on S if there exist a number M such

that
|Fn (x)| < M, x ∈ S, n = 1, 2, 3, ....
Remark: If {Fn } is pointwise bounded on S and S1 is countable subset of S, it is

always possible to nd a subsequence {Fnk } such that subsequence is convergent.
However, even if {Fn } is uniformly bounded sequence of continuous functions on a
compact set S, there need not exist a subsequence which converges pointwise on S.
Example: Consider the sequence of functions
Fn (x) = sin nx, x ∈ [0, 2π].
Suppose there exists a sequence {nk } such that {sin nk x} converges, for every
x ∈ [0, 2π]. Then we must have
lim (sin nk x − sin nk+1 x) = 0, x ∈ [0, 2π].

k→∞
Hence
lim (sin nk x − sin nk+1 x)2 = 0, x ∈ [0, 2π].
k→∞
By Lebesgue's theorem concerning integration of bounded convergent sequences,
we have
∫2π
lim lim (sin nk x − sin nk+1 x)2 = 0.
k→∞ k→∞
0
But we have
lim (sin nk x − sin nk+1 x)2 = 2π.
k→∞
which is a contradiction.
Example: Consider the sequence of functions
x2
Fn (x) = , S = [0, 1].
x2 + (1 − nx)2
1.13. Equicontinuous Functions on a Set 48
Then |Fn | ≤ 1, so that {Fn (x)} is uniformly bounded on [0, 1]. Also
lim Fn (x) = 0, x ∈ [0, 1].

n→∞
But
1
Fn ( ) = 1,
n
so no subsequence can converge uniformly on [0, 1].
1.13 Equicontinuous Functions on a Set

A family of functions F dened on the set S is equicontinuous if for all f ∈F and
for each ε>0 there is a δ>0 such that
|f (x1 ) − f (x2 )| ≤ ε if x1 , x2 ∈ [a, b], |x1 − x2 | < δ. (1.58)
Remark: It is clear that every member of F is uniformly continuous.
Theorem: If{Fn } is a pointwise bounded sequence of functions on a countable

set S , then {Fn } has a subsequence {Fnk } such that subsequence converges for all
x ∈ S.
Proof : Let{xi }, i = 1, 2, 3, ... be the points of S arranged in a sequence.

Since {Fn (xi )} is bounded, there exists a subsequence, which we shall denote by
{Fi,k }, such that {Fi,k (xi )} converges as k → ∞.
Consider the sequences S1 , S2 , ..., dened by
S1 : F1,1 F1,2 F1,3 F1,4 ...
S2 : F2,1 F2,2 F2,3 F2,4 ...

S3 : F3,1 F3,2 F3,3 F3,4 ...
...........................
Consider the sequences S1 , S2 , ..., dened by
S1 : F1,1 F1,2 F1,3 F1,4 ...
S2 : F2,1 F2,2 F2,3 F2,4 ...

S3 : F3,1 F3,2 F3,3 F3,4 ...
...........................
The sequence has the following properties
• Sn is a subsequence of Sn−1 , for n = 2, 3, 4, ...

• Due to the boundedness of {Fn (xn )}, we can say that Fn,k (xn ) converges, as
k → ∞.
• The order in which the functions appear is the same in each sequence, i.e., if
one function precedes another in S1 , they are in the same relation in every
Sn , until one or the other is deleted. Hence, when going from one row in the
above array to the next below, functions may move to the left but never to
the right.
We consider the sequence
E : F1,1 F2,2 F3,3 ...
By (3) property E is a subsequence of Sn , for n = 1, 2, 3, .... The order in which the

functions appear is the same in each sequence, i.e., if one function precedes another
in S1 , they are in the same relation in every Sn , until one or the other is deleted.
Hence, when going from one row in the above array to the next below, functions
may move to the left but never to the right. The (2) property of the sequence
ensures that {Fn,n (xi ) } converges as n→∞ for every x ∈ S.
Theorem: If K is a compact subset and if {Fn } is a sequence of continuous functions

dened on K and {Fn } converges uniformly then {Fn } is equicontinuous on K.
Proof : Since the sequence of functions {Fn } is uniformly convergent, for every
ε > 0, there is an integer N such that
∥Fn − FN ∥K < ε, n > N.
We know that continuous functions on compact sets are uniformly continuous, there
is a δ>0 such that
|Fi (x) − Fi (y)| < ε, |x − y| < δ, 1 ≤ i ≤ N.
Theorem: If K is a compact subset and if {Fn } is a sequence of continuous functions

dened on K and {Fn } converges uniformly then {Fn } is equicontinuous on K.
For n > N and |x − y| < δ , we have
|Fn (x) − Fn (y)| ≤ |Fn (x) − FN (x)| + |FN (x) − FN (y)|

+|FN (y) − Fn (y)|
< 3ε.
Theorem: If {Fn } is a sequence of continuous functions dened on a compact set

S and if {Fn } is a pointwise bounded and equicontinuous on S, then
1. {Fn } is uniformly bounded on S,
2. {Fn } contains a uniformly convergent subsequence.
Proof : Since {Fn } is equicontinuous then by denition for every ε > 0, we have
|Fn (x) − Fn (y)| < ε, |x − y| < δ.
From Analysis I, we know that S is compact then there are nitely many points
p1 , p2 , ...pr in S such that to every x ∈ S corresponds at least one p1 such that
|x − p1 | < δ .
Since {Fn } is pointwise bounded, there exists Mi < ∞ such that
|Fn (pi )| < Mi , n ∈ N.
If we take
M = max{M1 , ..., Mr },
then |Fn (x)| < M + ε for every x ∈ S. This proves the rst part of the theorem.
Theorem: If {Fn } is a sequence of continuous functions dened on a compact set

S and if {Fn } is a pointwise bounded and equicontinuous on S, then
1. {Fn } contains a uniformly convergent subsequence.
Proof : Let E be a countable dense subset of S. Then from previous theorem we

have a subsequence {Fni (x)} such that the subsequence {Fni (x)} converges for every
x ∈ E.
Fix the notationFni (x) = gi , we shall prove that {gi } converges uniformly on S .
Letε > 0, and choose δ as before. Let V (x, δ) be the set of all y ∈ S such that
|x − y| < δ .
Since E is dense in S , and S is compact, there are nitely many points x1 , ..., xm
in E such that
S ⊂ V (x1 , δ) ∪ ... ∪ V (xm , δ) (∗).
Since {gi (x)} converges for every x ∈ E, there is an integer N such that
|gi (xs ) − gj (xs )| < ε, whenever i, j ≥ N, 1 ≤ s ≤ m.
If x ∈ S, from (*) shows that x ∈ V (xs , δ) for some s, so that
|gi (x) − gi (xs )| < ε
for every i.
If i≥N and j ≥ N, it follows that
|gi (x) − gj (x)| ≤ |gi (x) − gi (xs )| + |gi (xs ) − gj (xs )| + |gj (xs ) − gj (x)|
|gi (x) − gj (x)| ≤ 3ε.

1.14. The Stone-Weierstrass Theorem 51
1.14 The Stone-Weierstrass Theorem

Theorem: If f is continuous function on [a, b], there exists a sequence of polyno-
mials Pn such that
lim Pn (x) = f (x),
x→∞
uniformly on [a, b].
Proof : Without any loss of generality, we may assume that [a, b] = [0, 1].
We may also assume that f (0) = f (1) = 0. As we can consider
g(x) = f (x) − f (0) − x[f (1) − f (0)], x ∈ [0, 1].
If g can be obtained as the limit of uniformly convergent sequence of polynomials,

it is clear that the same is true for f , since f − g is a polynomial.
Furthermore, we dene f (x) to be zero for x outside [0, 1]. Then f is uniformly
continuous on the whole line.
We take
Qn (x) = cn (1 − x2 )n , n = 1, 2, ...,
where cn is chosen so that
∫1
Qn (x)dx = 1, n = 1, 2, ....
−1
Consider the function

(1 − x2 )n − 1 + nx2 ,
which is zero at x=0 and whose derivative is positive in (0, 1).
Since
∫1 ∫1
(1 − x ) dx = 2
2 n
(1 − x2 )n dx
−1 0
√
∫ n
1/
≥ 2 (1 − x2 )n dx
0
√
∫ n
1/
≥ 2 (1 − nx2 )dx
0
4
= √
3 n
1
= √
n
1.14. The Stone-Weierstrass Theorem 52
It follows from
∫1
Qn (x)dx = 1, n = 1, 2, ....
−1
√
that cn < n.
For any δ > 0, we have
√
Qn (x) ≤ n(1 − δ 2 )n , δ ≤ |x| ≤ 1.
So that Qn → 0 uniformly in δ ≤ |x| ≤ 1.

Now set
∫1
Pn (x) = f (x + t)Qn (t)dt, x ∈ [0, 1].
−1
By change of variable and assumption on f implies that
∫
1−x ∫1
Pn (x) = f (x + t)Qn (t)dt = f (t)Qn (t − x)dt,
−x 0
x.
and the last integral is clearly a polynomial in
Thus {Pn } is a sequence of polynomials.
Given ε > 0, we chose δ > 0 such that |y − x| < δ implies
ε
|f (y) − f (x)| < .
3
Let M = sup |f (x)|, we see that for x ∈ [a, b], we have
∫1
|Pn (x) − f (x)| = | [f (x + t) − f (x)]Qn (t)dt|
−1
∫1
≤ |f (x + t) − f (x)|Qn (t)dt
−1
∫−δ ∫δ
ε
≤ 2M Qn (t)dt + Qn (t)dt
2
−1 −δ
∫1
+2M Qn (t)dt
δ
√ ε
≤ 4M n(1 − δ 2 )n +
2
1.15. Fourier Series 53
1.15 Fourier Series

One of the fundamental methods of solving many problems in engineering elds
is to represent the behavior of a system by a combination of simple behaviors.
Mathematically, this is related to representing a function f(x) in the form of a
functional series
∞
∑
f (x) = ck ϕk (x).
k=1
Here the functions ϕk (x) are suitable elementary functions, also called the base
set of functions, and the ck are called the coecients of the expansion.
For the Taylor series
∞
∑
f (x) = ck xk , |x| < R,
k=0
the set {1, x, ..., xn , ...} is a base set of functions.

Fourier Series: A Fourier series expansion of a function is a representation of
the function as a linear combination of sines and cosines, that is, the base set of the
representation is
{1, cos nx, sin nx}∞
n=1 .
1.15.1 Periodic Functions

A function f : Ω ⊂ R → R is said to be periodic if there exists a nonzero real
number ω such that
f (x) = f (x + ω), x ∈ Ω.
The simplest examples of periodic functions from R into R include the well known
sine and cosine functions, since for each k ∈ Z\{0}.
Figure 1.4: Periodic functions
Remark: If ω1 and ω2 are such that
f (x + ω1 ) = f (x), f (x + ω2 ) = f (x).
Then so is ω1 ± ω2 .
f (x + (ω1 ± ω2 ))
There is a smallest positive value ω of a periodic function f called the primitive

period (or the basic period or the fundamental period) of f (x).
The reciprocal of the primitive period is called the frequency of the periodic function.
Lemma: If f : R → R is a periodic functions with period ω , then the period of

f (cx) is ω/c. If f (x) and g(x) are periodic with the same period ω , then h(x) =
af (x) + bg(x) is also a periodic function with period ω . Here ω is not necessarily a
primitive period.
Proof : Let ϕ(x) = f (cx), then
ϕ(x) = f (cx) = f (cx + ω) = f (c(x + ω/c)) = ϕ(x + ω/c), x ∈ R.
This shows that ω/c is a period.

For the second part, we consider
h(x + ω) = af (x + ω) + bg(x + ω) = af (x) + bg(x) = h(x).
Example: sin(cx) and cos(cx) are periodic functions with period 2π/c.
The function
∞
∑
(an cos nx + bn sin nx),
n=1
is a periodic function with period 2π .

Although, individual functions, cos x, cos 2x, cos 3x, ..., have periods
2π, π, 2π/3, ..., respectively.
Lemma: If f (x) is a periodic function with period ω, then
∫ c+ω ∫ ω
f (x)dx = f (x)dx,
c 0
whenever f is integrable on [0, ω].
Proof : Geometrically, it is obvious
Figure 1.5: Geometric proof

Consider
∫
c+ω ∫0 ∫ω ∫
c+ω
f (x)dx = f (x)dx + f (x)dx + f (x)dx

c c 0 ω
∫c ∫ω ∫c
= − f (x)dx + f (x)dx + f (s)ds
0 0 0
∫ω
= f (x)dx.
0
showing that the integral of a periodic function with period ω taken over an arbitrary
interval of length ω always has the same value.
1.15.2 Periodic Extension

Suppose that f is a function dened on [a, a + ω]. Then the periodic extension of f
over the innite interval (−∞, ∞) is dened by the formula
{
f (x), a ≤ x < a + ω,
f˜( x) =
f (x − nω), a + nω ≤ x < a + (n + 1)ω,
where n is an integer.
Figure 1.6: Periodic extension example 1

Figure 1.7: Periodic extension example 2
1.15.3 Trigonometric Polynomials

Any linear combination of the trigonometric functions sin kx, cos kx, given by
n ( )
a0 ∑
sn (x) = + ak cos kx + bk sin kx , x ∈ R,
2
k=1
where ak and bk are real numbers, is known as trigonometric polynomials.
Recall the Stone and Weierstrass theorem stating that the trigonometric polynomials
are dense in C[a, b] for any closed interval [a, b], provided that b − a < 2π .
n ( )
a0 ∑
sn (x) = + ak cos kx + bk sin kx , x ∈ R,
2
k=1
The sequence {sn }, converges on a set E, then we may dene a function f :E→R
by
∞ ( )
a0 ∑
f (x) = lim sn (x) = + ak cos kx + bk sin kx , x ∈ E.
n→∞ 2
k=1
The series on the right is called a trigonometric series. The constants a0 , ak , bk
(k ∈ N) are called coecients of the trigonometric series.
We have taken the constant term in series as a0 /2 rather than a0 so that we can
make a0 /2 t in a general formula later.
We observe that if the series on the right converges for all real t[0, 2π], then the
sum f must satisfy
f (x) = f (x + 2π), x ∈ R.
Vector Space: A vector space is a nonempty set V of objects, called vectors, on

which are dened two operations, called addition and multiplication by scalars (real
numbers), subject to the ten axioms (or rules). The axioms must hold for all vectors
u, v, and w in V and for all scalars c and d.
1. The sum of u and v, denoted by u + v, is in V .

1.16. The space E 57
2. u + v = v + u.
3. ( u + v ) + w = u + ( v + w)
4. There is a zero vector 0 in V such that u + 0 = u.

5. For each u in V , there is a vector −u in V such that u + (−u) = 0.
6. The scalar multiple of u by c, denoted by cu, is in V .
7. c(u + v) = cu + cv.
8. (c + d)u = cu + du.
9. c(du) = (cd)u.
10. 1u = u.
Remark: Using only these axioms, one can show that the zero vector in Axiom 4 is
unique, and the vector −u, called the negative of u, in Axiom 5 is unique for each
u in V .
The Inner Product: Let u, v, and w be vectors in vector space V , and let c be
a scalar. Then an inner product is a function < ., . >: V × V → F such that
1. < v, u >=< u, v >
2. < (v + u), w >=< v, w > + < u, w >
3. < cu, v >=< u, cv >= c < v, u >
4. < u, u >≥ 0, and < u, u >= 0 if and only if u = 0.
1.16 The space E

Let us dene the space E be the set of all real valued piecewise dened periodic
function f on the interval [−π, π].
Theorem: The space E is a linear space, that is, a vector space. Moreover, E an
inner product space with respect to the inner product
∫π
1
< f, g >= f (x)g(x)dx.
π
−π
The trigonometric functions: The set of functions
1
Φ = { √ , cos(nx), sin(nx) : n ∈ N}
2
is an innite orthonormal system in E with respect to the inner product dened
∫π
1
< f, g >= f (x)g(x)dx.
π
−π
Let Φ = {ϕ1 , ϕ2 , ..., ϕn , ...} be an orthonormal basis of an innite dimensional

inner product space X , and let f ∈ X . Then the innite series
∞
∑ ∞
∑
< f, ϕk > ϕk (x) := ck ϕk (x),
k=1 k=1
is called the Fourier series of f (relative to Φ), and the coecients ck =< f, ϕk >
are called the kth Fourier coecient of f.
We introduce ∫ π
1
∥f ∥ =< f, f >=
2
|f (x)|2 dx.
π −π
Suppose that we are given a trigonometric series of the form
∞ ( )
a0 ∑
f (x) = + ak cos kx + bk sin kx , x ∈ E.
2
k=1
Clearly, since each term of the series has period 2π , if it converges to a function
f (x), then f (x) must be a periodic function with period 2π .
Thus, only 2π -periodic functions are expected to have trigonometric series of the
above form.
Problem: Suppose that f is a 2π -periodic function. Under what conditions does

the function have a representation of the form
∞ ( )
a0 ∑
f (x) = + an cos nx + bn sin nx .
2
n=1
When it does, what should be an , bn ?

Assume for the moment that the series
∞ ( )
a0 ∑
f (x) = + an cos nx + bn sin nx , (∗)
2
n=1
converges uniformly on R. This is the case if
∞ ( )
|a0 | ∑
+ |an | + |bn |
2
n=1
converges, so that the series (*) is dominated by the convergent series in R.

∫ ∞ ∫ { }
1 π
a0 ∑ 1 π
f (x)dx = + (an cos nx + bn sin nx)
π −π −π 2 π
n=1
{ ∫ } ∑ ∞ { ∫
a0 1 π an π
= dx + cos nxdx
2 π −π π −π
n=1
∫ }
bn π
+ sin nx)
π −π
= a0
Recall:
∫ π ∫ π
1 1
cos nx cos kxdx = δnk = sin nx sin kxdx
π −π π −π
∫ π
cos nx sin kxdx = 0
−π
and
2 cos α cos β = cos(α + β) + cos(α − β)
2 sin α sin β = cos(α − β) − cos(α + β)
2 sin α cos β = sin(α + β) + sin(α − β).
∞ ( )
a0 ∑
f (x) = + an cos nx + bn sin nx , (∗)
2
n=1
Multiply by cos kx and the series forf (x) cos kx can be integrated term by term for
each xed k, we can determine ak and bn .
∫ π ∫
1 a0 1 π
f (x) cos kxdx = cos kxdx +
π −π 2 π −π
∞ (
∑ ∫ π
an cos kx cos nxdx
n=1 −π
∫ π )
+bn sin nx cos kxdx .
−π
∫ π
1
ak = f (x) cos kxdx.
π −π
Multiply by sin kx and the series forf (x) sin kx can be integrated term by term
for each xed k, we can determine bk and bn .
∫ π ∫
1 a0 1 π
f (x) sin kxdx = sin kxdx +
π −π 2 π −π
∞ (
∑ ∫ π
an sin kx cos nxdx
n=1 −π
∫ π )
+bn sin nx sin kxdx .
−π
∫ π
1
bk = f (x) sin kxdx.
π −π
Fourier Series: For any integrable function f on [−π, π], the numbers ak and bk
dened by
∫ π ∫ π
1 1
ak = f (x) cos kxdx, k ≥ 0, bk = f (x) sin kxdx, k ≥ 1.
π −π π −π
are called the Fourier coecients of f. The corresponding trigonometric series
∞ ( )
a0 ∑
+ ak cos kx + bk sin kx ,
2
k=1
is called the Fourier series of f. We express this association by writing
∞ ( )
∼ a0 ∑
f (x) = + ak cos kx + bk sin kx ,
2
k=1
to indicate that the Fourier series on the right may or may not converge to f at
some point t ∈ [−π, π].
Theorem: If the trigonometric series of the form
∞ ( )
a0 ∑
+ ak cos kx + bk sin kx , (∗)
2
k=1
converges uniformly on [−π, π], then it is the Fourier series of its sum.
More precisely, if the trigonometric series (*) converges uniformly to f on [−π, π],
then the ak bk are given by
and
∫ π ∫
1 1 π
ak = f (x) cos kxdx, k ≥ 0, bk = f (x) sin kxdx, k ≥ 1.
π −π π −π
Remark: We have no idea what happens if the series
∞ ( )
a0 ∑
+ ak cos kx + bk sin kx , (∗)
2
k=1
doesn't converge uniformly on [−π, π].
However, since
|ak cos kx + bk sin kx| ≤ |ak | + |bk |,
Weierstrass M-test shows that the trigonometric series (*) converges absolutely and
uniformly on every closed interval [a, b] whenever
∞
∑
(|ak | + |bk |)
k=1
is convergent.
1.16.1 Fourier Series of Even and Odd Functions

Even and odd functions possess certain simple but useful properties:
• The product of two even (or odd) functions is an even function.
• The sum of two even (or odd) functions is an even (or odd) function.
• The product of an even and an odd function is an odd function.
• For a Riemann integrable function f dened on [−c, c] (c > 0), it is evident

that ∫ c ∫ c
f (x)dx = 2 f (x)dx, if f is even
−c 0
∫ c
f (x)dx = 0, if f is odd
−c
Fourier series of even function: Suppose that f (x) is a periodic function of

period 2π . Let us further assume that f is even on (−π, π), i.e., f (x) = f (−x) for
all x ∈ (−π, π).
Then the product function f (x) sin kx is odd, which means that bk = 0 for all k ≥ 1,
and hence we have the Fourier cosine series
∞ ∫
a0 ∑ 1 π
f (x) ∼
= + ak cos kx, ak = f (x) cos kxdx.
2 π −π
k=1
Fourier series of odd function: Suppose that f (x) is a periodic function of

period 2π . Let us further assume that f is odd on (−π, π), i.e., f (x) = −f (−x) for
all x ∈ (−π, π).
Then the product function f (x) cos kx is odd, which means that ak = 0 for all k ≥ 0,
and hence we have the Fourier cosine series
∞
∑ ∫ π
1
f (x) ∼
= ak sin kx, bk = f (x) sin kxdx.
π −π
k=1
Example: Consider f (x) = |x| on [−π, π].

Then f is even and continuous on [−π, π].
2(1 − (−1)n )
an = − .
n2 π
We have
∞
π 4 ∑ cos(2k + 1)x
|x| = − .
2 π (2k + 1)2
k=0
∞
π 4 ∑ cos(2k + 1)x
|x| = − .
2 π (2k + 1)2
k=0
Remark: Note that the Fourier series here converges uniformly to |x| on [−π, π]
but not on the whole interval (−∞, ∞), and so outside the interval (−∞, ∞), f (x)
is determined by the periodicity condition f (x) = f (x + 2π).
we can make use of this series to nd the values of some numerical series. For instant
x=0 gives
∞
π2 ∑ 1
= .
8 (2k + 1)2
k=1
Some natural questions arise:
• For what values of x does the Fourier series of f converge? Does it converge
for all x in [−π, π]? If it converges on [−π, π] but not to f , what will be its
sum?
• If the Fourier series of f converges at x, does it converge to f?
• If the Fourier series of f converges to f on [−π, π], does it converge uniformly

to f on [−π, π]?
Is the continuity of f is sucient to guarantee convergence of the Fourier series

of f on [−π, π]?
In 1876, Paul du Bois-Reymond constructed a continuous function f : [−π, π] →
R whose Fourier series failed to converge to f at each point in a dense subset of
[−π, π].
Indeed, the following are true statements
• There exists a continuous function whose Fourier series diverges at a point.
• There exists a continuous function whose Fourier series converges everywhere

on [−π, π], but not uniformly.
• There exists a continuous function whose Fourier series diverges for points in
some set S and converges on (−π, π) \ S .
The space E: Let us dene the space E be the set of all real valued piecewise
dened periodic function f on the interval [−π, π].
Dene
{
f (x + h) − f (x+)
E′ = f ∈E : lim exists x ∈ [−π, π)
h→0+ h
}
f (x + h) − f (x−)
lim exists x ∈ (−π, π]
h→0− h
Theorem: Let f ∈ E ′. Then for each x ∈ (−π, π), the Fourier series of f (x)
converges to the value
f (x−) + f (x+)
.
2
At the end points x = ±π , the series converges to
f (π−) + f (−π+)
.
2
Remark: If f ∈ E′ is continuous at x, then f (x−) = f (x+) = f (x), and so at such

points
f (x−) + f (x+)
= f (x).
2
Thus, the Fourier series of f converges to f(x) at the point x where it is continuous.
At the point of discontinuity x, the Fourier series of f assumes the mean of the
one-sided limits of f.
Corollary: f : [−π, π] → R is continuous, and if f (−π) = f (π), f ′ (x)
If exists and
is piecewise continuous on [−π, π], then the Fourier series of f converges to f (x) at
every point x ∈ [−π, π].
Theorem: Suppose that f : [−π, π] → R [−π, π] and

is piecewise continuous on
piecewise monotone,that is, there exists a partition P = x0 , x1 , ..., xn of [−π, π] such
that the restriction f |[x ,k = 1, 2..., n, is either increasing or decreasing.
k−1 ,xk ]
Let f (x) be dened for other values of x by the periodicity condition f (x) = f (x +
2π). Then the Fourier series of f on [−π, π] converges to
• f (x) if f is continuous at x ∈ (−π, π).
• (f (x+) + f (x−))/2 if f is discontinuous at x.
• (f (π−) + f ((−π)−))/2 if f is discontinuous at x = ±π .

Figure 1.8: At discontinuous points
Example:If f (x) = x on [−π, π) and f (π) = −π. Find the Fourier sine series
of f.
Figure 1.9: Example
• f is odd function, hence an = 0.

∫π ∫
2 π 2(−1)n−1
• bn = 1
π −π x sin nxdx = π 0 x sin nxdx = n .
∞
∑ (−1)k−1
x∼
=2 sin kx.
k
k=1
Remarks: Note that the Fourier series does not necessarily agree with f (x) = x
at every point in [−π, π].
The Fourier series vanishes at both endpoints x = ±π , whereas the function
does not vanish at either endpoint.
However, the Dirichlet's theorem states that series converges to f (x) at every
interior point of (−π, π).
For example at x = π/2 ∼
= could be replaced by =
the symbol and so
( )
π 0 (−1) 0 1
=2 1− + − + + ... .
2 2 3 4 5
∞
∑ (−1)k−1
x∼
=2 sin kx.
k
k=1
Remarks: Finally, we remark that at the endpoints x = ±π, the series converges
to
f (π−) + f ((−π)−) π + (−π)
= = 0.
2 2
we could also consider f as follows: f (x) = x on (−π, π) and f (−π) = f (π) = 0.
Figure 1.10: Example
Example: If f (x) = ex on [−π, π) and f (x + 2π) = f (x) for x ∈ R. Determine

the Fourier series of the function f.
Some facts about complex numbers.
Example: If f (x) = ex [−π, π) and f (x + 2π) = f (x)
on for x ∈ R. Determine
the Fourier series of the function f .
∫ ∫ ∫
einx = cos nxdx + i sin nxdx.
According to this, the Fourier coecients are easy to derive quickly by writing
∫ π
1
an − ibn = e−inx ex dx
π −π
1 e (1−in)x
= |π
π 1 − in −π
( )
1 e(1−in)π − e−(1−in)π
=
π 1 − in
(−1) (e − e−π )
n π
=
π(1 − in)
2(−1)n sinh π 2(−1)n−1 n sinh π

an = , bn = .
π(1 + n2 ) π(1 + n2 )
1.17. Fourier Series for Arbitrary Periodic Function 66
We have
∞
x ∼ sinh π 2 sinh π ∑ (−1)n
e = + cos nx
π π (1 + n2 )
n=1
∞
2 sinh π ∑ (−1)n−1 n
+ sin nx.
π (1 + n2 )
n=1
Remark: In particular, at the point of continuity x = 0, it follows that
∞
sinh π 2 sinh π ∑ (−1)n
1= + .
π π (1 + n2 )
n=1
Which can be written as
∞
π csc π − 1 ∑ (−1)n
= .
2 (1 + n2 )
n=1
Remark: According to Dirichlet's theorem, at the endpoint x = π, we have
∞
eπ + e−π sinh π 2 sinh π ∑ 1
= + ,
2 π π (1 + n2 )
n=1
∞
∑ 1
π coth π = 1 + 2 .
(1 + n2 )
n=1
Which reduces to
∞
π coth π − 1 ∑ 1
= .
2 (1 + n2 )
n=1
1.17 Fourier Series for Arbitrary Periodic Function

Suppose that f 2L-periodic
is a and Riemann integrable function. The function
f (at) has period 2L/a.
In particular, f ((L/π)t) is 2π -periodic, and so the Fourier series expansion has the
following in terms of the variable t:
∞ ( )
L ∼ a0 ∑
f ( t) = + an cos nt + bn sin nt , t ∈ [−π, π],
π 2
n=1
where ∫ π ∫
1 L 1 L kπ
an = f ( t) cos ntdt = f (x) cos( x)dx.
π −π π L −L L
∞ ( )
L a0 ∑
f ( t) ∼
= + an cos nt + bn sin nt , t ∈ [−π, π],
π 2
n=1
where ∫ π ∫ L
1 L 1 kπ
an = f ( t) cos ntdt = f (x) cos( x)dx,
π −π π L −L L
and similarly,
∫ L
1 kπ
bn = f (x) sin( x)dx.
L −L L
We remark that the interval of integration in the last two formulas for the Fourier
coecients can be replaced with an arbitrary interval [c, c + 2L], of length 2L.
Changing the variable t, by setting t = (π/L)x.
Theorem: Let f be a periodic function with period 2L. Then the Fourier expansion
of f is given by
∞ ( )
a0 ∑ nπ nπ
f (x) ∼
= + an cos( x) + bn sin( x) , x ∈ [−L, L],
2 L L
n=1
where ∫ L
1 nπ
an = f (x) cos( x)dx,
L −L L
and ∫ L
1 nπ
L −L L
Remark: The interval of integration in the last formulas for the Fourier coecients
can be replaced with the interval [c, c + 2L], where c is any real number; we usually
let c = −L. Notice that
nπ nπ
cos( (x + 2L)) = cos( x)
L L
nπ nπ
sin( (x + 2L)) = sin( x).
L L
Corollary: The Fourier series of an even function f with period 2L is a Fourier

cosine series
∞
a0 ∑ kπ
f (x) ∼
= + an cos( x), x ∈ [c, c + 2L],
2 L
n=1
where ∫ c+2L
1 nπ
an = f (x) cos( x)dx.
L c L
and the Fourier series of an odd function f with period 2L is a Fourier sine series
∞
∑ nπ
f (x) ∼
= bn sin( x), x ∈ [c, c + 2L],
L
n=1
where ∫ c+2L
1 nπ
bn = f (x) sin( x)dx
L c L
where c is any real number.

{
0, −2 ≤ x < 0,
f (x) =
1, 0 ≤ x ≤ 2
Here, we have L = 2, and the function is even. We have
∫ L
1 kπ
an = f (x) cos( x)dx,
L −L L
and ∫ L
1 kπ
L −L L
we obtain
1
a0 = , an = 0.
2
1 + (−1)n−1
bn = , n ≥ 1.
nπ
Example: Consider the function f (x) = | sin x|. The function is dened for all x
and the function has period π.
Clearly, f represents a continuous, piecewise smooth, even function of period π,
and therefore it is everywhere equal to its Fourier series, consisting of cosine terms
only.
We have c = 0, and L = π/2, then we have
∫
2 π
ak = f (x) cos(2kx)dx
π 0
∫
2 π
= sin x cos(2kx)dx
π 0
∫ π[ ]
1
= sin(1 + 2k)x − sin(2k − 1)x dx
π 0
∫ [ ]
1 π
= sin(1 + 2k)x − sin(2k − 1)x dx
π 0
( )
1 cos(1 + 2k)x cos(2k − 1)x π
= − +
π 2k + 1 2k − 1 0
( )
1 (−1)2k+1 − 1 (−1)2k−1 − 1
= − −
π 2k + 1 2k − 1
4
ak = − .
π(4k 2 − 1)
1.18. Best Approximation Theorem 69
Thus, the Fourier series expansion of | sin x| is
∞
2 4 ∑ cos 2kx
| sin x| = − , x ∈ [−π, π].
π π 4k 2 − 1
k=1
1.18 Best Approximation Theorem

Theorem: Let Φ = ϕ1 , ..., ϕn be an orthonormal set of functions in the inner
product space E , and let ck be the Fourier coecients of f relative to ϕk :
∫
1 π
ck = f (x)ϕk (x)dx :=< f, ϕk > .
π −π
T (x) is an arbitrary Fourier polynomial relative to ϕk ,
If that is, Tn (x) =
∑n n
k=1 dk ϕk (x) for some constants d1 , ..., dn , then we have
2
∑n

f − c ϕ (x) ≤ ∥f − Tn ∥2 ,
k k
k=1
with equality if and only if ck = dk for each k = 1, ..., n. Moreover,
∑
n ∫π
1
|ck | ≤
2
|f (x)|2 dx.
π
k=1 −π
∑n
Proof : Setting Sn = k=1 ck ϕk (x). Then we have
∫π
1
∥f − Tn ∥2 = |f (x) − Tn (x)|2 dx
π
−π
∫π ∫π
1 1
= |f (x)| dx +
2
|Tn (x)|2 dx
π π
−π −π
∫π
1
−2 dk f (x)ϕk (x)dx
π
−π
∫π ∑
n ∑
n
1
= |f (x)|2 dx + |dk |2 − 2 ck dk
π
−π k=1 k=1
∫π ∑
n ∑
n
1
= |f (x)|2 dx + |dk |2 − 2 ck dk
π
−π k=1 k=1
∫π ∑
n ∑
n
1
= |f (x)|2 dx + |ck − dk |2 − 2 ck
π
−π k=1 k=1
∑
n
= ∥f − Sn ∥2 + |ck − dk |2 .
k=1
1.18. Best Approximation Theorem 70
Therefore,
∥f − Tn ∥2 ≥ ∥f − Sn ∥2 ,
with equality if and only if ck = dk for each k = 1, ..., n.
∥f − Tn ∥2 ≥ ∥f − Sn ∥2 ,
Note that f and ϕk are xed, while the dk are allowed to vary.
In particular, setting dk = ck , shows that the minimum value of ∥f − Tn ∥2 ≥
∥f − Sn ∥2 , is given by
∫π ∑
n ∑
n
1
min ∥f − Tn ∥ = 2
|f (x)| dx −
2
|ck | = ∥f ∥ −
2 2
|ck |2 ,
Tn π
−π k=1 k=1
which has to be nonnegative. This gives
∑
n ∫π
1
|ck | ≤
2
|f (x)|2 dx for all n.
π
k=1 −π
Chapter 2
Functions of Several Variables
2.1 Euclidean Spaces

The vector sum of
X = (x1 , x2 , . . . , xn ) and Y = (y1 , y2 , . . . , yn )
is
X + Y = (x1 + y1 , x2 + y2 , . . . , xn + yn ). (2.1)
If a is a real number, the scalar multiple of X by a is

aX = (ax1 , ax2 , . . . , axn ). (2.2)
In R4 , let
( )
X = (1, −2, 6, 5) and Y = 3, −5, 4, 12 .
Then
( )
X + Y = 4, −7, 10, 11
2
and
6X = (6, −12, 36, 30).
Theorem: If X, Y, and Z are in Rn and a and b are real numbers, then
• X + Y = Y + X (vector addition is commutative).
• (X + Y) + Z = X + (Y + Z) (vector addition is associative).
• There is a unique vector 0, called the zero vector, such that X + 0 = X for all
in R .
X n
• For each X in Rn there is a unique vector −X such that X + (−X) = 0.
• a(bX) = (ab)X.
• (a + b)X = aX + bX.
• a(X + Y) = aX + aY.
• 1X = X.
2.2. Schwarz's Inequality 72
Remark: Clearly, 0 = (0, 0, . . . , 0) and, if X = (x1 , x2 , . . . , xn ), then
−X = (−x1 , −x2 , . . . , −xn ).
We write X + (−Y) as X − Y. The point 0 is called the origin .

Length, distance: The length of the vector X = (x1 , x2 , . . . , xn ) is
|X| = (x21 + x22 + · · · + x2n )1/2 .
The distance between points X and Y is |X − Y|.

In particular, |X| is the distance between X and the origin. If |X| = 1, then X is a
unit vector .
Example: The lengths of the vectors
( )
X = (1, −2, 6, 5) and Y = 3, −5, 4, 12
are
√
|X| = (12 + (−2)2 + 62 + 52 )1/2 = 66
and √
201
|Y| = (3 + (−5) + 4 +
2 2 2
( 12 )2 )1/2 = .
2
The distance between X and Y is
√
149
|X − Y| = ((1 − 3) + (−2 + 5) + (6 − 4) + (5 −
2 2 2 1 2 1/2
2) ) = .
2
The inner product X · Y of X = (x1 , x2 , . . . , xn ) and Y = (y1 , y2 , . . . , yn ) is
X · Y = x1 y1 + x2 y2 + · · · + xn yn .
2.2 Schwarz's Inequality

Lemma: If X and Y are any two vectors in Rn , then
|X · Y| ≤ |X| |Y|, (2.3)
with equality if and only if one of the vectors is a scalar multiple of the other.
Proof : Suppose that Y ̸= 0 and t is any real number. Then
∑
n
0 ≤ (xi − tyi )2
i=1
∑n ∑n ∑n (2.4)
= x2i − 2t i=1 xi yi + t2 2
i=1 yi
i=1
= |X|2 − 2(X · Y)t + t2 |Y|2 .
The last expression is a second-degree polynomial p in t. From the quadratic formula,

the zeros of p are
√
(X · Y) ± (X · Y)2 − |X|2 |Y|2
t= .
|Y|2
Hence,
(X · Y)2 ≤ |X|2 |Y|2 . (2.5)
because if not, then p would have two distinct real zeros and therefore be negative
between them, contradicting the inequality (2.4).
Proof :
(X · Y)2 ≤ |X|2 |Y|2 , (2.6)
Taking square roots in (2.6) yields (2.3) if Y ̸= 0. If X = tY, then |X · Y| =

|X||Y| = |t||Y|2 (verify), so equality holds in (2.3).
Conversely, if equality holds in (2.3), then p has the real zero t0 = (X · Y)/|Y|2 ,
and
∑
n
(xi − t0 yi )2 = 0
i=1
from (2.4); therefore, X = t0 Y .
Theorem: If X and Y are in Rn , then
|X + Y| ≤ |X| + |Y|, (2.7)
with equality if and only if one of the vectors is a nonnegative multiple of the other.
Proof : By denition,
∑n
|X + Y|2 = (xi + yi )2
∑i=1
n 2
∑n ∑n 2
= i=1 xi + 2 i=1 xi yi + i=1 yi
= |X|2 + 2(X · Y) + |Y|2 (2.8)
≤ |X|2 + 2|X| |Y| + |Y|2 (by Schwarz's inequality)
= (|X| + |Y|)2 .
Hence,
|X + Y|2 ≤ (|X| + |Y|)2 .
Taking square roots yields (2.7).
From the third line of (2.8), equality holds in (2.7) if and only if X · Y = |X||Y|,
which is true if and only if one of the vectors X and Y is a nonnegative scalar
multiple of the other.
Corollary: If X, Y, and Z are in Rn , then
|X − Z| ≤ |X − Y| + |Y − Z|.
Proof : Write
X − Z = (X − Y) + (Y − Z),
and apply triangle inequality with X and Y replaced by X−Y and Y − Z.
Corollary: If X and Y are in Rn , then
|X − Y| ≥ ||X| − |Y|| .
Proof : Since
X = Y + (X − Y),
Triangle inequality implies that
|X| ≤ |Y| + |X − Y|,
which is equivalent to |X| − |Y| ≤ |X − Y|.

Interchanging X and Y yields
|Y| − |X| ≤ |Y − X|.
Since |X − Y| = |Y − X|, the last two inequalities imply the stated conclusion.
Theorem: If X, Y, and Z are members of Rn and a is a scalar, then
• |aX| = |a| |X|.
• |X| ≥ 0, with equality if and only if X = 0.
• |X − Y| ≥ 0, with equality if and only if X = Y.
• X · Y = Y · X.
• X · (Y + Z) = X · Y + X · Z.
• (cX) · Y = X · (cY) = c(X · Y).

2.3. Neighbourhoods and Open Sets in Rn 75
2.2.1 Line Segment in Rn

The equation of a line through a point X0 = (x0 , y0 , z0 ) in R3 can be written
parametrically as
x = x0 + u1 t, y = y0 + u2 t, z = z0 + u3 t, −∞ < t < ∞,
where u1 , u2 , and u3 are not all zero. We write this in vector form as
X = X0 + tU, −∞ < t < ∞, (2.9)
with U = (u1 , u2 , u3 ), and we say that the line is through X0 in the direction of U.
There are many ways to represent a given line parametrically.
For example,
X = X0 + sV, −∞ < s < ∞, (2.10)
represents the same line as (2.9) if and only if V = aU for some nonzero real number
a.
Then the line is traversed in the same direction as s and t vary from −∞ to ∞ if
a > 0, or in opposite directions if a < 0. To write the parametric equation of a line
through two points X0 and X1 in R3 .
We take U = X1 − X0 in (2.9), which yields
X = X0 + t(X1 − X0 ) = tX1 + (1 − t)X0 , −∞ < t < ∞.
The line segment from X0 to X1 consists of those points for which 0 ≤ t ≤ 1.

Suppose that X0 U are in Rn and U ̸= 0.
and
Then the line through X0 in the direction of U is the set of all points in Rn of the
form
X = X0 + tU, −∞ < t < ∞.
A set of points of the form
X = X0 + tU, t 1 ≤ t ≤ t2 ,
is called a line segment . The line segment from X0 to X1 is the set of points of the
form
X = X0 + t(X1 − X0 ) = tX1 + (1 − t)X0 , 0 ≤ t ≤ 1.
2.3 Neighbourhoods and Open Sets in Rn

If ε > 0, the ε-neighborhood of a point X0 in Rn is the set
Nε (X0 ) = {X|X − X0 | < ε}.
Nε (X0 ) in R2
2.3. Neighbourhoods and Open Sets in Rn 76
We are going to dene neighborhood, interior point, interior of a set, open set,
closed set,limit point, boundary point, boundary of a set, closure of a set, isolated
point, exterior point, and exterior of a set.
Example: S be the set of points in R2 in the square bounded by the lines

Let
x = ±1, y = ±1, except for the origin and the points on the vertical lines x = ±1
thus,
S = {(x, y) : (x, y) ̸= (0, 0), −1 < x < 1, −1 ≤ y ≤ 1}.
Every point of S not on the lines y = ±1 is an interior point.
(−1, 1) (1, 1)
x x
(−1, −1) (1, −1)
Figure 2.1: The set S
S 0 = {(x, y) : (x, y) ̸= (0, 0), −1 < x, y < 1}.
S is a deleted neighborhood of (0, 0) and is neither open nor closed.

The closure of S is
S = {(x, y) : −1 ≤ x, y ≤ 1},
and every point of S is a limit point of S.
The origin and the perimeter of S form ∂S , the boundary of S . The exterior of
S consists of all points (x, y) such that |x| > 1 or |y| > 1. The origin is an isolated
point of Sc.
Example: X0 is
If a point in Rn and r is a positive number, the open n-ball of
radius r about X0 is the set
Br (X0 ) = {X : |X − X0 | < r}.
Thus, ε-neighborhoods are open n-balls. If X1 is in Sr (X0 ) and
|X − X1 | < ε = r − |X − X0 |,
then X is in Sr (X0 ). Thus, Sr (X0 ) contains an ε-neighborhood of each of its points,

and is therefore open.
2.4. Convergence of a Sequence in Rn 77
We can show that the closure of Br (X0 ) is the closed n-ball of radius r about X0 ,
dened by
S r (X0 ) = {X : |X − X0 | ≤ r}.
Remark: Open and closed n-balls are generalizations to Rn of open and closed
intervals.
Lemma: If X1 andX2 are in Sr (X0 ) for some r > 0, then so is every point on the
line segment from X1 to X2 .
Proof : The line segment is given by
X = tX2 + (1 − t)X1 , 0 < t < 1.
Suppose that r > 0. If
|X1 − X0 | < r, |X2 − X0 | < r,
and 0 < t < 1, then
|X − X0 | = |tX2 + (1 − t)X1 − tX0 − (1 − t)X0 |

= |t(X2 − X0 ) + (1 − t)X1 − X0 )|
< tr + (1 − t)r = r.
2.4 Convergence of a Sequence in Rn

A sequence of points {Xr } in Rn converges to the limit X if
lim |Xr − X| = 0.
r→∞
In this case we write

lim Xr = X.
r→∞
Theorem: Let
X = (x1 , x2 , . . . , xn ) and Xr = (x1r , x2r , . . . , xnr ), r ≥ 1.
Then limr→∞ Xr = X if and only if
lim xir = xi , 1 ≤ i ≤ n;
r→∞
that is, a sequence {Xr } of points in Rn converges to a limit X if and only if the
sequences of components of {Xr } converge to the respective components of X.
2.5. Principle of nested sets 78
Theorem (Cauchy's Convergence Criterion): A sequence {Xr } in Rn con-

verges if and only if for each ε>0 there is an integer K such that
|Xr − Xs | < ε if r, s ≥ K.
Diameter of a Set: If S is a nonempty subset of Rn , then
d(S) = sup{|X − Y| : X, Y ∈ S}
is the diameter of S.
If d(S) < ∞, S is bounded ; if d(S) = ∞, S is unbounded .
2.5 Principle of nested sets

Theorem: If S1 , S 2 , . . . are closed nonempty subsets of Rn such that
S1 ⊃ S2 ⊃ · · · ⊃ Sr ⊃ · · · (2.11)
and
lim d(Sr ) = 0, (2.12)
r→∞
then the intersection

∞
∩
I= Sr
r=1
contains exactly one point.
Proof : Let {Xr } be a sequence such that Xr ∈ Sr (r ≥ 1).

Because of S1 ⊃ S2 ⊃ · · · ⊃ Sr ⊃ · · · , Xr ∈ Sk if r ≥ k , so
|Xr − Xs | < d(Sk ) if r, s ≥ k.
From limr→∞ d(Sr ) = 0 and Cauchy's convergence theorem, Xr converges to a

limit X. Since X is a limit point of every Sk and every Sk is closed, X is in every
Sk (A set is closed if and only if it contains all its limit points). Therefore,
X ∈ I , so I ̸= ∅. Moreover, X is the only point in I, since if Y ∈ I, then
|X − Y| ≤ d(Sk ), k ≥ 1,
and (2.12) implies that Y = X.

2.6. Heine-Borel Theorem 79
2.6 Heine-Borel Theorem

We are going to state and prove the Heine-Borel theorem for Rn .
This theorem concerns compact sets. As in R, a compact set in Rn is a closed and

bounded set.
Recall that a collection H of open sets is an open covering of a set S if
S ⊂ ∪{H : H ∈ H}.
Theorem: If H is an open covering of a compact subset S, then S can be covered

by nitely many sets from H.
Proof : The proof is by contradiction. We rst consider the case where n = 2, so

that you can visualize the method.
Suppose that there is a covering H for S from which it is impossible to select a
nite subcovering.
Since S is bounded, S is contained in a closed square
T = {(x, y)|a1 ≤ x ≤ a1 + L, a2 ≤ x ≤ a2 + L}
with sides of length L
T (1) T (2)
S (1) S (2)
S (4) S (3)
T (4) T (3)
Figure 2.2: Heine-Borel Theorem for n=2
Bisecting the sides of T leads to four closed squares, T (1) , T (2) , T (3) , and T (4) ,
with sides of length L/2. Let
S (i) = S ∩ T (i) , 1 ≤ i ≤ 4.
Each S (i) , being the intersection of closed sets, is closed, and
∪
4
S= S (i) .
i=1
2.6. Heine-Borel Theorem 80
Moreover, H covers each S (i) , but at least one S (i) cannot be covered by any nite
subcollection of H, since if all the S
(i) could be, then so could S. Let S1 be a set
with this property, chosen from S
(1) , S (2) , S (3) , and S (4) .
We are now back to the situation we started from: a compact set S1 covered by H,
but not by any nite subcollection of H. However, S1 is contained in a square T1
with sides of length L/2 L. Bisecting the sides of T1 and repeating the
instead of
argument, we obtain a subset S2 of S1 that has the same properties as S , except that
it is contained in a square with sides of length L/4. Continuing in this way produces
a sequence of nonempty closed sets S0 (= S), S1 , S2 , . . . , such that Sk ⊃ Sk+1 and
d(Sk ) ≤ L/2k−1/2 (k ≥ 0).
∩∞
From Principle of Nested Sets Theorem, there is a point X in k=1 Sk .
Since X ∈ S , there is an open set H in H that contains X, and this H must also
contain some ε-neighborhood of X. Since every X in Sk satises the inequality
|X − X| ≤ 2−k+1/2 L,
it follows that Sk ⊂ H for k suciently large.
This contradicts our assumption on H, which led us to believe that no Sk could be

covered by a nite number of sets from H.
Consequently, this assumption must be false: H must have a nite subcollection

that covers S . This completes the proof for n = 2.
The idea of the proof is the same for n > 2. The counterpart of the square T is
the hypercube with sides of length L:
T = {(x1 , x2 , . . . , xn ) : ai ≤ xi ≤ ai + L, i = 1, 2, . . . , n}.
Halving the intervals of variation of the n coordinates x1 , x2 , ..., xn divides T into

2n closed hypercubes with sides of length L/2:
T (i) = {(x1 , x2 , . . . , xn ) : bi ≤ xi ≤ bi + L/2, 1 ≤ i ≤ n},
where bi = ai or bi = ai + L/2. If no nite subcollection of H covers S , then at least

one of these smaller hypercubes must contain a subset of S that is not covered by
any nite subcollection of S. Now the proof proceeds as for n = 2.
Remark: The BolzanoWeierstrass theorem is valid in Rn ; its proof is the same as

in R.
2.7. Connected Sets in Rn 81
2.7 Connected Sets in Rn

A subset S of Rn is connected if it is impossible to represent S as the union of two
disjoint nonempty sets such that neither contains a limit point of the other.
If S cannot be expressed as S = A ∪ B, where
A ̸= ∅, B ̸= ∅, A ∩ B = ∅, and A ∩ B = ∅. (2.13)
If S can be expressed in this way, then S is disconnected .

Example: The empty set and singleton sets are connected, because they cannot be
represented as the union of two disjoint nonempty sets.
Example: Rn is connected.
The space
If R
= A ∪ B with A ∩ B = ∅ and A ∩ B = ∅, then A ⊂ A and B ⊂ B .
n
That is, A and B are both closed and therefore are both open.
Since the only nonempty subset of R that is both open and closed is R
n n itself,
one of A and B is R and the other is empty.
n
2.7.1 Polygonal Path

If X1 , X2 , . . . , Xk are points in Rn
Let Li is the line segment from Xi to Xi+1 , 1 ≤ i ≤ k − 1, we say that L1 , L2 ,

..., Lk−1 form a polygonal path from X1 to Xk .
We say that X1 and Xk are connected by the polygonal path.
2.8 Polygonally Connected Set

A set S is polygonally connected if every pair of points in S can be connected by a
polygonal path lying entirely in S.
Theorem: An open set S in Rn is connected if and only if it is polygonally

connected.
Proof : For suciency, we will show that if S is disconnected, then S is not poly-
gonally connected.
Let S = A ∪ B, where A and B satisfy
A ̸= ∅, B ̸= ∅, A ∩ B = ∅, and A ∩ B = ∅.
2.8. Polygonally Connected Set 82
.
Suppose that X1 ∈ A and X2 ∈ B , and assume that there is a polygonal path
in S connecting X1 to X2 . Then some line segment L in this path must contain a
point Y1 in A and a point Y2 in B .
The line segment
X = tY2 + (1 − t)Y1 , 0 ≤ t ≤ 1,
is part of L and therefore in S. Now dene
ρ = sup{τ : tY2 + (1 − t)Y1 ∈ A, 0 ≤ t ≤ τ ≤ 1}.
Let Xρ = ρY2 + (1 − ρ)Y1 . Then Xρ ∈ A ∩ B .
However, since Xρ ∈ A ∪ B and A ∩ B = A ∩ B = ∅, this is impossible.
Therefore, the assumption that there is a polygonal path in S from X1 to X2 must

be false.
For necessity, suppose that S X0 ∈ S . Let A be the set
is a connected open set and
consisting of X0 and the points in S can be connected to X0 by polygonal paths in
S . Let B be set of points in S that cannot be connected to X0 by polygonal paths.
If Y0 ∈ S , then S contains an ε-neighborhood Nε (Y0 ) of Y0 , since S is open. Any
point Y1 in Nε (Y0 can be connected to Y0 by the line segment
X = tY1 + (1 − t)Y0 , 0 ≤ t ≤ 1,
which lies in Nε (Y0 ) and therefore in S. This implies that Y0 can be connected
to X0 by a polygonal path in S Nε (Y0 ) can also.
if and only if every member of
Thus, Nε (Y0 ) ⊂ A if Y0 ∈ A, and Nε (Y0 ) ∈ B if Y0 ∈ B . Therefore, A and B are
open. Since A ∩ B = ∅, this implies that A ∩ B = A ∩ B = ∅. Since A is nonempty
(X0 ∈ A), it now follows that B = ∅, since if B ̸= ∅, S would be disconnected.
Therefore, A = S , which completes the proof of necessity.
Remark: Any polygonally connected set, open or not, is connected. The converse
is false. A set (not open) may be connected but not polygonally connected.
Regions in Rn : A region S in Rn is the union of an open connected set with some,

all, or none of its boundary; thus, S0 is connected, and every point of S is a limit
0
point of S .
Example: Intervals are the only regions in R. The n-ball Br (X0 ) is a region in Rn ,
S r (X0 ). The set
as is its closure S = {(x, y) : x2 + y 2 ≤ 1 or x2 + y 2 ≥ 4} is not
a region in R , since it is not connected.
2
The set S1 obtained by adding the line segment
L1 : X = t(0, 2) + (1 − t)(0, 1), 0 < t < 1,

2.8. Polygonally Connected Set 83
Figure 2.3: Disconnected set which is not a region
Figure 2.4: A connected set which is not a region

2.9. Sequences in Rn 84
to S is connected but is not a region, since points on the line segment are not limit
points of S10 . The set S2 obtained by adding to S1 the points in the rst quadrant
bounded by the circles x2 + y 2 = 1 and x2 + y 2 = 4 and the line segments L1 and
L2 : X = t(2, 0) + (1 − t)(1, 0), 0 < t < 1,
is a region.
Figure 2.5: A region
2.9 Sequences in Rn
A sequence {Xr } of points in Rn converges to a limit X if and only if for every ε > 0
there is an integer K such that
|Xr − X| < ε if r ≥ K.
The Rn denitions of divergence, boundedness, subsequence, and sums, dierences,

and constant multiples of sequences are analogous to those we discussed in Analysis
I.
Since Rn is not ordered for n > 1, monotonicity, limits inferior and superior of
sequences in R , and divergence to ±∞ are undened for n > 1.
n
Products and quotients of members of Rn n > 1.

are also undened if
Several theorems from Analysis I remain valid for sequences in R ,
n with proofs
unchanged, provided that | | is interpreted as distance in R .
n
1. uniqueness of the limit.
2. Boundedness of a convergent sequence.

2.10. Domain of Function of n Variable 85
3. Concerning limits of sums, dierences, and constant multiples of convergent

sequences.
4. Every subsequence of a convergent sequence converges to the limit of the

sequence.
2.10 Domain of Function of n Variable

We denote the domain of a function f by Df and the value of f at a point X=
(x1 , x2 , . . . , xn ) by f (X) or f (x1 , x2 , . . . , xn ).
If a function is dened by a formula such as
( )1/2
f (X) = 1 − x21 − x22 − · · · − x2n (2.14)
( )
2 −1
g(X) = 1 − x21 − x22 − · · · − xn (2.15)
without specication of its domain, it is to be understood that its domain is the

largest subset of Rn for which the formula denes a unique real number.
2.11 Limit at a Point of a Function of n Variables

A function f (X) approaches the limit L as X approaches X0 and write
lim f (X) = L,
X→X0
if X0 is a limit point of Df and, for every ε > 0, there is a δ>0 such that
|f (X) − L| < ε
for all X in Df such that

0 < |X − X0 | < δ.
Example: If g(x, y) = 1 − x2 − 2y 2 , then
lim g(x, y) = 1 − x20 − 2y02 (2.16)

(x,y)→(x0 ,y0 )
for every (x0 , y0 ).

To see this, we write
|g(x, y) − g(x0 − y0 )| = |(1 − x2 − 2y 2 ) − (1 − x20 − 2y02 )|

≤ |x2 − x20 | + 2|y 2 − y02 |
= |(x + x0 )(x − x0 )| (2.17)
+2|(y + y0 )(y − y0 )|
≤ |X − X0 |(|x + x0 | + 2|y + y0 )|),
2.11. Limit at a Point of a Function of n Variables 86
since
|x − x0 | ≤ |X − X0 | and |y − y0 | ≤ |X − X0 |.
If |X − X0 | < 1, then |x| < |x0 | + 1 and |y| < |y0 | + 1.
|g(x, y) − g(x0 − y0 )| < K|X − X0 | if |X − X0 | < 1,
where
K = (2|x0 | + 1) + 2(2|y0 | + 1).
Therefore, if ε>0 and
|X − X0 | < δ = min{1, ε/K},
then

g(x, y) − (1 − x20 − 2y02 ) < ε.

√
sin 1 − x2 − 2y 2
h(x, y) = √
1 − x2 − 2y 2
is dened only on the interior of the region bounded by the ellipse
x2 + 2y 2 = 1.
It is not dened at any point of the ellipse itself or on any deleted neighborhood of
y y
X0
X − X0 = δ
x x
x 2 + 2y 2 = 1 x 2 + 2y 2 = 1
(a) (b)
Figure 2.6: Domain of the function
such a point. Nevertheless,
lim h(x, y) = 1 (2.18)

(x,y)→(x0 ,y0 )
if
x20 + 2y02 = 1. (2.19)
To see this, let

√
u(x, y) = 1 − x2 − 2y 2 .
2.11. Limit at a Point of a Function of n Variables 87
Then
sin u(x, y)
h(x, y) = . (2.20)
u(x, y)
Recall that
sin r
lim = 1.
r→0r
Therefore, if ε > 0, there is a δ1 > 0 such that

sin u

u − 1 < ε if 0 < |u| < δ1 . (2.21)
From previous example, we have
lim (1 − x2 − 2y 2 ) = 0.
(x,y)→(x0 ,y0 )
If (2.19) holds, so there is a δ>0 such that
0 < u2 (x, y) = (1 − x2 − 2y 2 ) < δ12 .
if X = (x, y) is in the interior of the ellipse and |X − X0 | < δ ; that is, if X is in the
shaded region.
Therefore,
√
0<u= 1 − x2 − 2y 2 < δ1 (2.22)
if X is in the interior of the ellipse and |X − X0 | < δ ; that is, if X is in the shaded
region. This, (2.20), and (2.21) imply that
|h(x, y) − 1| < ε
for such X, which is the required result.
Theorem: If limX→X0 f (X) exists, then it is unique.
Proof : See lecture.

xy
f (x, y) =
x2 + y2
is dened everywhere in R2 except at (0, 0). Does lim(x,y)→(0,0) f (x, y) exist?
If we try to answer this question by letting (x, y) approach (0, 0) along the line
y = x, we see the functional values
x2 1
f (x, x) = =
2x2 2
and conclude that the limit is 1/2.
2.12. Innite Limits and Limits at X → ∞ 88
However, if we let (x, y) approach (0, 0) along the line y = −x, we see the
functional values
x2 1
f (x, −x) = − 2
=−
2x 2
and conclude that the limit equals −1/2.
In fact, they are both incorrect. What we have shown is that
1 1
lim f (x, x) = and lim f (x, −x) = − .
x→0 2 x→0 2
Since limx→0 f (x, x) and limx→0 f (x, −x) must both equal lim(x,y)→(0,0) f (x, y).
Theorem: Suppose that f and g are dened on a set D, X0 is a limit point of D,

and
lim f (X) = L1 , lim g(X) = L2 .
X→X0 X→X0
Then
lim (f + g)(X) = L1 + L2 , (2.23)

X→X0
lim (f − g)(X) = L1 − L2 , (2.24)
X→X0
lim (f g)(X) = L1 L2 , (2.25)
X→X0
if L2 ̸= 0,
( )
f L1
lim (X) = . (2.26)
X→X0 g L2
2.12 Innite Limits and Limits at X → ∞

We say that f (X) approaches ∞ as X approaches X0
lim f (X) = ∞
X→X0
if X0 is a limit point of Df and, M, there is a δ>0 such that
f (X) > M whenever 0 < |X − X0 | < δ and X ∈ Df .
We say that
lim f (X) = −∞
X→X0
if
lim (−f )(X) = ∞.
X→X0
Example: If
f (X) = (1 − x21 − x22 − · · · − x2n )−1/2 ,
then
lim f (X) = ∞
X→X0
if |X0 | = 1, because
1
f (X) = ,
|X − X0 |
so
1
f (X) > M if 0 < |X − X0 | < δ = .
M
Example: If
1
f (x, y) = ,
x + 2y + 1
then lim(x,y)→(1,−1) f (x, y) does not exist (why not?).
But
lim |f (x, y)| = ∞.
(x,y)→(1,−1)
To see this, we observe that
|x + 2y + 1| = |(x − 1) + 2(y + 1)|

√
≤ 5|X − X0 | (by Schwarz's inequality),
where X0 = (1, −1). So
1 1
|f (x, y)| = ≥√ .
|x + 2y + 1| 5|X − X0 |
Therefore,
1
|f (x, y)| > M if 0 < |X − X0 | < √ .
M 5

( )

sin x2 +y12 +z 2
f (x, y, z) =
x2 + y 2 + z 2
assumes arbitrarily large values in every neighborhood of (0, 0, 0).
For example, if Xk = (xk , yk , zk ), where
1
xk = yk = zk = √ ( ) ,
3 k + 12 π
then ( )
1
f (Xk ) = k + π.
2
However, this does not imply that limX→0 f (X) = ∞. Since, for example, every
neighborhood of (0, 0, 0) also contains points
( )
1 1 1
Xk = √ ,√ ,√ .
3kπ 3kπ 3kπ
For which f (Xk ) = 0.
2.12.1 Limit at Innity

If Df is unbounded, we say that
lim f (X) = L (nite)

|X|→∞
if for every ε > 0, there is a number R such that
|f (X) − L| < ε whenever |X| ≥ R and X ∈ Df .
Example: If ( )
1
f (x, y, z) = cos ,
x + 2y 2 + z 2
2
then
lim f (X) = 1. (2.27)
|X|→∞
To see this, we recall that the continuity of cos u at u = 0 implies that for each
ε>0 there is a δ>0 such that
| cos u − 1| < ε if |u| < δ.
Since
1 1
≤ .
x2 + 2y 2 + z 2 |X|2
√
It follows that if |X| > 1/ δ , then
1
< δ.
x2 + 2y 2 + z 2
Therefore,
|f (X) − 1| < ε.
Example: Consider the function dened only on the domain
D = {(x, y) : 0 < y ≤ ax}, 0 < a < 1,

by
1
f (x, y) = .
x−y
We will show that
lim f (x, y) = 0. (2.28)
|X|→∞
It is important to keep in mind that we need only consider (x, y) in D, since f is

not dened elsewhere.
In D,
x − y ≥ x(1 − a) (2.29)
and
|X|2 = x2 + y 2 ≤ x2 (1 + a2 ).
So
|X|
x≥ √ .
1 + a2
1−a
x−y ≥ √ |X|, X ∈ D.
1 + a2
So √
1 + a2 1
|f (x, y)| ≤ , X ∈ D.
1 − a |X|
1−a
x−y ≥ √ |X|, X ∈ D.
1 + a2
So √
1 + a2 1
|f (x, y)| ≤ , X ∈ D.
1 − a |X|
Therefore,
|f (x, y)| < ε
if X∈D and √
1 + a2 1
|X| > .
1−a ε
Remarks: In the same manner we can dene lim|X|→∞ f (X) = ∞ and

lim|X|→∞ f (X) = −∞. We will have the following notion limX→X0 f (X) exists
means that limX→X0 f (X) = L, where L is nite; to leave open the possibility that
L = ±∞.
We will say that limX→X0 f (X) exists in the extended reals. A similar convention
applies to limits as |X| → ∞.
2.13. Continuity 92
2.13 Continuity
If X0 is in Df and is a limit point of Df , then we say that f is continuous at X0 if
lim f (X) = f (X0 ).

X→X0
Theorem: Suppose that X0 is in Df and is a limit point of Df . Then f is continuous

at X0 if and only if for each ε>0 there is a δ>0 such that
|f (X) − f (X0 )| < ε
whenever
|X − X0 | < δ and X ∈ Df .
f (x, y) = 1 − x2 − 2y 2
is continuous on R2 .
Solution: See lecture.
 √
 √ 1−x −2y , x2 + 2y 2 < 1,
sin 2 2
h(x, y) = 21−x −2y

2
 1, x2 + 2y 2 = 1,
then it follows from the example we have discussed that h is continuous on the
ellipse
x2 + 2y 2 = 1.
Example: Can we redene the function
xy
f (x, y) = ,
x2 + y 2
to make it continuous at (0, 0).
The limit
lim f (x, y)
(x,y)→(0,0)
does not exist.
Consequently, it is impossible to dene the function at origin to make it contin-

uous.
Theorem: If f and g are continuous on a set S in Rn , then so are f + g, f − g, and

f g. Also, f /g is continuous at each X0 in S such that g(X0 ) ̸= 0.
2.14. Vector Valued Functions 93
2.14 Vector Valued Functions

Suppose that g1 , g2 , ..., gn are real-valued functions dened on a subset T of Rm .
We dene the vector-valued function G on T by
G(U) = (g1 (U), g2 (U), . . . , gn (U)) , U ∈ T.
Then g1 , g2 , ..., gn are the component functions of G = (g1 , g2 , . . . , gn ). We say

that
lim G(U) = L = (L1 , L2 , . . . , Ln )
U→U0
if
lim gi (U) = Li , 1 ≤ i ≤ n,
U→U0
and that G is continuous at U0 if g1 , g2 , ..., gn are each continuous at U0 .
Theorem: For a vector-valued function G,
lim G(U) = L
U→U0
if and only if for each ε>0 there is a δ>0 such that
|G(U) − L| < ε whenever 0 < |U − U0 | < δ and U ∈ DG .
Similarly, G is continuous at U0 if and only if for each ε>0 there is a δ>0 such
that
|G(U) − G(U0 )| < ε whenever |U − U0 | < δ and U ∈ DG .
2.14.1 Composite Function

Let f be a real-valued function dened on a subset of Rn , and let the vector-valued
function G = (g1 , g2 , . . . , gn ) be dened on a domain DG in Rm .
Let the set
T = {U : U ∈ DG and G(U) ∈ Df },
be nonempty.
Composite function: Dene the real-valued composite function
h=f ◦G
on T by
h(U) = f (G(U)), U ∈ T.
T = {U : U ∈ DG and G(U) ∈ Df },
2.14. Vector Valued Functions 94
R(G) = range of G
n
m
G
DG
Df
Figure 2.7: Composite of vector valued functions
Theorem: Suppose that U0 is in T and is a limit point of T, G is continuous at

U0 , and f is continuous at X0 = G(U0 ). Then h = f ◦ G is continuous at U0 .
Proof : Suppose that ε > 0. Since f is continuous at X0 = G(U0 ), there is an

ε1 > 0 such that
|f (X) − f (G(U0 ))| < ε (2.30)
if
|X − G(U0 )| < ε1 and X ∈ Df . (2.31)
Since G is continuous at U0 , there is a δ>0 such that
|G(U) − G(U0 )| < ε1 if |U − U0 | < δ and U ∈ DG .
By taking X = G(U) in (2.30) and (2.31), we see that
|h(U) − h(U0 )| = |f (G(U) − f (G(U0 ))| < ε
if
|U − U0 | < δ and U ∈ T.
Example: If
√
f (s) = s
and
g(x, y) = 1 − x2 − 2y 2 ,
then Df = [0, ∞], Dg = R2 , and
T = {(x, y) : x2 + 2y 2 ≤ 1}.
We have proved that g is continuous on R2 .

We can obtain the same conclusion by observing that the functions p1 (x, y) = x
are continuous on R .
and p2 (x, y) = y 2
2.15. Bounded Functions 95
Theorem: Suppose thatU0 is in T and is a limit point of T, G is continuous at

U0 , and f is continuous at X0 = G(U0 ).
Then h = f ◦ G is continuous at U0 . Since f is continuous on Df , the function
√
h(x, y) = f (g(x, y)) = 1 − x2 − 2y 2
is continuous on T.
Example: If
√
g(x, y) = 1 − x2 − 2y 2
and { sin s
s , s ̸= 0,
f (s) =
1, s = 0,
then Df = (−∞, ∞) and
Dg = T = {(x, y) : x2 + 2y 2 ≤ 1}.
We have proved that g is continuous on T . Since f is continuous on Df , the

composite function h = f ◦ g dened by
 √
 sin√ 1−x2 −2y2 , x2 + 2y 2 < 1,
h(x, y) = 1−x2 −2y 2

1, x2 + 2y 2 = 1,
is continuous on
Dg = T = {(x, y) : x2 + 2y 2 ≤ 1}.
2.15 Bounded Functions

The denitions of bounded above, bounded below , and bounded on a set S are the
same for functions of n variables as for functions of one variable, as are the denitions
of supremum and inmum of a function on a set S .
Theorem: If f is continuous on a compact set S in Rn , then f is bounded on S.
Theorem: Let f be continuous on a compact set S in Rn and
α = inf f (X), β = sup f (X).

X∈S X∈S
Then
f (X1 ) = α and f (X2 ) = β
for some X1 and X2 in S.

2.16. Directional Derivative 96
Theorem: Let f be continuous on a region S in Rn .

Suppose that A and B are in S and
f (A) < u < f (B).
Then f (C) = u for some C in S.
Proof : If there is no such C, then S = R ∪ T, where
R = {X : X ∈ S and f (X) < u}

T = {X : X ∈ S and f (X) > u}.
If X0 ∈ R, the continuity of f implies that there is a δ>0 such that
f (X) < u if |X − X0 | < δ
and X ∈ S.
This means that X0 ̸∈ T . Therefore, R ∩T = ∅. Similarly, R ∩T = ∅. Therefore,
S is disconnected, which contradicts the assumption that S is a region. Hence, we
conclude that f (C) = u for some C in S.
Theorem: A function f is uniformly continuous on a subset S of its domain in Rn

if for every ε>0 there is a δ>0 such that
|f (X) − f (X′ )| < ε
whenever
|X − X′ | < δ
and X, X′ ∈ S .
Remark: We emphasize that δ must depend only on ε and S, and not on the
particular points X ′
and X .
Theorem: If f is continuous on a compact set S in Rn , then f is uniformly contin-

uous on S.
2.16 Directional Derivative

Let Φ be a unit vector and X a point in Rn .
The directional derivative of f at X in the direction of Φ is dened by
∂f (X) f (X + tΦ) − f (X)

= lim
∂Φ t→0 t
if the limit exists.
That is, ∂f (X)/∂Φ is the ordinary derivative of the function
h(t) = f (X + tΦ)
at t = 0, if h′ (0) exists.
Example: Let Φ = (ϕ1 , ϕ2 , ϕ3 ) and
f (x, y, z) = 3xyz + 2x2 + z 2 .
Then
h(t) = f (x + tϕ1 , y + tϕ2 , z + tϕ3 ),

= 3(x + tϕ1 )(y + tϕ2 )(z + tϕ3 ) + 2(x + tϕ1 )2
+(z + tϕ3 )2 .
h(t) = 3(x + tϕ1 )(y + tϕ2 )(z + tϕ3 ) + 2(x + tϕ1 )2 + (z + tϕ3 )2
Then we have
h′ (t) = 3ϕ1 (y + tϕ2 )(z + tϕ3 ) + 3ϕ2 (x + tϕ1 )(z + tϕ3 )

+ 3ϕ3 (x + tϕ1 )(y + tϕ2 ) + 4ϕ1 (x + tϕ1 )
+2ϕ3 (z + tϕ3 ).
Therefore,
∂f (X)
= h′ (0) = (3yz + 4x)ϕ1 + 3xzϕ2 + (3xy + 2z)ϕ3 .
∂Φ
2.16.1 Partial Derivative

Consider the unit vectors
E1 = (1, 0, . . . , 0), E2 = (0, 1, 0, . . . , 0), . . . , En = (0, . . . , 0, 1).
Since X and X + tEi dier only in the ith coordinate, ∂f (X)/∂Ei is called the
partial derivative of f with respect to xi at X.
It is also denoted by ∂f (X)/∂xi or fxi (X); thus,
∂f (X) f (x1 + t, x2 , . . . , xn ) − f (x1 , x2 , . . . , xn )

= fx1 (X) = lim ,
∂x1 t→0 t
f (x1 , . . . , xi−1 , xi + t, xi+1 , . . . , xn ) − f (x1 , x2 , . . . , xn )
fxi (X) = lim
t→0 t
if 2 ≤ i ≤ n, and
∂f (X) f (x1 , . . . , xn−1 , xn + t) − f (x1 , . . . , xn−1 , xn )

= fxn (X) = lim ,
∂xn t→0 t
if the limits exist. If we write X = (x, y), then we denote the partial derivatives
accordingly; thus,
∂f (x, y) f (x + h, y) − f (x, y)
= fx (x, y) = lim
∂x h→0 h
∂f (x, y) f (x, y + h) − f (x, y)
= fy (x, y) = lim .
∂y h→0 h
It can be seen from these denitions that to compute fxi (X) we simply dierentiate
f with respect to xi according to the rules for ordinary dierentiation, while treating
the other variables as constants.
Example: Let
f (x, y, z) = 3xyz + 2x2 + z 2 .
Taking Φ = E1 (that is, setting ϕ1 = 1 and ϕ2 = ϕ3 = 0), we nd that
∂f (X) ∂f (X)
= = 3yz + 4x,
∂x ∂E1
which is the result obtained by regarding y and z as constants in and taking the
ordinary derivative with respect to x. Similarly,
∂f (X) ∂f (X)
= = 3xz
∂y ∂E2
∂f (X) ∂f (X)
= = 3xy + 2z.
∂z ∂E3
Theorem: If fxi (X) and gxi (X) exist, then
∂(f + g)(X)
= fxi (X) + gxi (X),
∂xi
∂(f g)(X)
= fxi (X)g(X) + f (X)gxi (X),
∂xi
and, if g(X) ̸= 0,
∂(f /g)(X) g(X)fxi (X) − f (X)gxi (X)
= .
∂xi [g(X)]2
If fxi (X) exists at every point of a set D, then it denes a function fxi on D.
If this function has a partial derivative with respect to xj on a subset of D, we
denote the partial derivative by
( )
∂ ∂f ∂2f
= = fxi xj .
∂xj ∂xi ∂xj ∂xi
Similarly,
( )
∂ ∂2f ∂3f
= = fxi xj xk .
∂xk ∂xj ∂xi ∂xk ∂xj ∂xi
The function obtained by dierentiating f successively with respect to

xi1 , xi2 , . . . , xir is denoted by
∂rf
= fxi1 · · · xir−1 xir ;
∂xir ∂xir−1 · · · ∂xi1
it is an rth-order partial derivative of f . The function
f (x, y) = 3x2 y 3 + xy
has partial derivatives everywhere. Its rst-order partial derivatives are
fx (x, y) = 6xy 3 + y, fy (x, y) = 9x2 y 2 + x.
Its second-order partial derivatives are
fxx (x, y) = 6y 3 , fyy (x, y) = 18x2 y,

fxy (x, y) = 18xy 2 + 1, fyx (x, y) = 18xy 2 + 1.
There are eight third-order partial derivatives. Some examples are
fxxy (x, y) = 18y 2 , fxyx (x, y) = 18y 2 , fyxx (x, y) = 18y 2 .
Compute fxx (0, 0), fyy (0, 0), fxy (0, 0), and fyx (0, 0) if
{ 2
(x y+xy 2 ) sin(x−y)
x2 +y 2
, (x, y) ̸= (0, 0),
f (x, y) =
0, (x, y) = (0, 0).
If (x, y) ̸= (0, 0), the ordinary rules for dierentiation, applied separately to x
and y , yield
(2xy+y 2 ) sin(x−y)+(x2 y+xy 2 ) cos(x−y)
fx (x, y) = x2 +y 2
(2.32)
2x(x2 y+xy 2 ) sin(x−y)
− (x2 +y 2 )2
, (x, y) ̸= (0, 0),
and
(x2 +2xy) sin(x−y)−(x2 y+xy 2 ) cos(x−y)
fy (x, y) = x2 +y 2
(2.33)
2y(x2 y+xy 2 ) sin(x−y)
− (x2 +y 2 )2
, (x, y) ̸= (0, 0).
These formulas do not apply if (x, y) = (0, 0), so we nd fx (0, 0) and fy (0, 0) from
their denitions as dierence quotients:
f (x, 0) − f (0, 0) 0−0

fx (0, 0) = lim = lim = 0,
x→0 x x→0 x
f (0, y) − f (0, 0) 0−0
fy (0, 0) = lim = lim = 0.
y→0 y y→0 y
Setting y=0 in (2.32) and (2.33) yields
fx (x, 0) = 0, fy (x, 0) = sin x, x ̸= 0,

so
fx (x, 0) − fx (0, 0) 0−0
fxx (0, 0) = lim = lim = 0.
x→0 x x→0 x
fy (x, 0) − fy (0, 0) sin x − 0
fyx (0, 0) = lim = lim = 1.
x→0 x x→0 x
Setting x=0 in (2.32) and (2.33) yields
fx (0, y) = − sin y, fy (0, y) = 0, y ̸= 0,
so
fx (0, y) − fx (0, 0) − sin y − 0

fxy (0, 0) = lim = lim = −1,
y→0 y y→0 y
fy (0, y) − fy (0, 0) 0−0
fyy (0, 0) = lim = lim = 0.
y→0 y y→0 y
2.16.2 Equality of Mixed Partial Derivatives

Theorem: Suppose that f, fx , fy , and fxy exist on a neighborhood N of (x0 , y0 ),
and fxy is continuous at (x0 , y0 ).
Then fyx (x0 , y0 ) exists, and
fyx (x0 , y0 ) = fxy (x0 , y0 ). (2.34)
Proof : Suppose that ε > 0. Choose δ>0 so that the open square
Sδ = {(x, y) : |x − x0 | < δ, |y − y0 | < δ}
is in N.
|fxy (b
x, yb) − fxy (x0 , y0 )| < ε if x, yb) ∈ Sδ .
(b (2.35)
This is possible because of the continuity of fxy at (x0 , y0 ). The function
A(h, k) = f (x0 + h, y0 + k) − f (x0 + h, y0 ) − f (x0 , y0 + k) + f (x0 , y0 ) (2.36)
is dened if −δ < h, k < δ .

Moreover,
A(h, k) = ϕ(x0 + h) − ϕ(x0 ), (2.37)
where
ϕ(x) = f (x, y0 + k) − f (x, y0 ).
Since
ϕ′ (x) = fx (x, y0 + k) − fx (x, y0 ), |x − x0 | < δ,
(2.37) and the mean value theorem imply that
x, y0 + k) − fx (b
A(h, k) = [fx (b x, y0 )] h. (2.38)
where b
x is between x0 and x0 + h.
The mean value theorem, applied to fx (b
x, y) (where b
x is regarded as constant),
also implies that
x, y0 + k) − fx (b
fx (b x, yb)k,
x, y0 ) = fxy (b
where yb is between y0 and y0 + k .
From this and (2.38),
x, yb)hk.
A(h, k) = fxy (b
Now (2.35) implies that

A(h, k)
− f (x , y ) = |fxy (b
x, yb) − fxy (x0 , y0 )| < ε
hk xy 0 0
if 0 < |h|, |k| < δ.
Since (2.36) implies that
A(h, k) f (x0 + h, y0 + k) − f (x0 + h, y0 )

lim = lim
k→0 hk k→0 hk
f (x0 , y0 + k) − f (x0 , y0 )
− lim
k→0 hk
fy (x0 + h, y0 ) − fy (x0 , y0 )
= .
h
It follows from (2.39) that

fy (x0 + h, y0 ) − fy (x0 , y0 )
− f (x , y )≤ε if 0 < |h| < δ.
h
xy 0 0
Taking the limit as h→0 yields
|fyx (x0 , y0 ) − fxy (x0 , y0 )| ≤ ε.
Since ε is an arbitrary positive number, this proves (2.34).
2.16.3 Generalization of Equality of Mixed Partial Derivative

Theorem: Suppose that f and all its partial derivatives of order ≤ r are continuous
on an open subset S of Rn .
Then
fxi1 xi2 ,...,xir (X) = fxj1 xj2 ,...,xjr (X), X ∈ S. (2.39)
If each of the variables x1 , x2 , ..., xn appears the same number of times in
{xi1 , xi2 , . . . , xir } and {xj1 , xj2 , . . . , xjr }.

2.17. Dierentiability of Functions of Several Variables 102
If this number is rk , we denote the common value of the two sides of (2.39) by
∂ r f (X)
. (2.40)
∂xr11 ∂xr22 · · · ∂xrnn
It being understood that
0 ≤ rk ≤ r, 1 ≤ k ≤ n, (2.41)
r1 + r2 + · · · + rn = r, (2.42)
and, if rk = 0, we omit the symbol ∂x0k from the denominator of (2.40).
Remark: A function of several variables may have rst-order partial derivatives at

a point X0 but fail to be continuous at X0 .

{ xy
x2 +y 2
, (x, y) ̸= (0, 0),
f (x, y) = (2.43)
0, (x, y) = (0, 0).
Then
f (h, 0) − f (0, 0) 0−0

fx (0, 0) = lim = lim =0
h→0 h h→0 h
f (0, k) − f (0, 0) 0−0
fy (0, 0) = lim = lim = 0,
k→0 k k→0 k
but f is not continuous at (0, 0).
Remark: If dierentiability of a function of several variables is to be a stronger

property than continuity, as it is for functions of one variable, the denition of
dierentiability must require more than the existence of rst partial derivatives.
A function f is dierentiable at x0 if and only if
f (x) − f (x0 ) − m(x − x0 )

lim =0
x→x0 x − x0
for some constant m, in which case m = f ′ (x0 ).
2.17 Dierentiability of Functions of Several Variables

A function f is dierentiable at
X0 = (x10 , x20 , . . . , xn0 )
if X0 ∈ Df0 and there are constants m1 , m2 , . . . , mn such that

∑
f (X) − f (X0 ) − ni=1 mi (xi − xi0 )
lim = 0. (2.44)
X→X0 |X − X0 |
Example: Show that the following function f (x, y) = x2 + 2xy, is dierentiable at

any point (x0 , y0 ).
f (x, y) − f (x0 , y0 ) = x2 + 2xy − x20 − 2x0 y0
= x2 − x20 + 2(xy − x0 y0 )
= (x − x0 )(x + x0 ) + 2(xy − x0 y)
+2(x0 y − x0 y0 )
= (x + x0 + 2y)(x − x0 ) + 2x0 (y − y0 )
= 2(x0 + y0 )(x − x0 ) + 2x0 (y − y0 )
+ (x − x0 )(x − x0 + 2y − 2y0 )
= m1 (x − x0 ) + m2 (y − y0 ) + (x − x0 )(x − x0 + 2y − 2y0 ),
where
m1 = 2(x0 + y0 ) = fx (x0 , y0 ) and m2 = 2x0 = fy (x0 , y0 ). (2.45)
Therefore,
|f (x, y) − f (x0 , y0 ) − m1 (x − x0 ) − m2 (y − y0 )|
|X − X0 |
|x − x0 ||(x − x0 ) + 2(y − y0 )|
=
|X − X0 |
√
≤ 5|X − X0 |,
by Schwarz's inequality. This implies that
f (x, y) − f (x0 , y0 ) − m1 (x − x0 ) − m2 (y − y0 )
lim = 0,
X→X0 |X − X0 |
so f is dierentiable at (x0 , y0 ).
Theorem: If f X0 = (x10 , x20 , . . . , xn0 ), then fx1 (X0 ), fx2 (X0 ),

is dierentiable at
. . . , fxn (X0 ) exist and the constants m1 , m2 , . . . , mn in
∑
f (X) − f (X0 ) − ni=1 mi (xi − xi0 )
lim = 0,
X→X0 |X − X0 |
are given by
mi = fxi (X0 ), 1 ≤ i ≤ n; (2.46)
that is, ∑
f (X) − f (X0 ) − ni=1 fxi (X0 )(xi − xi0 )
lim = 0.
X→X0 |X − X0 |
Proof : Let i be a given integer in {1, 2, . . . , n}. Let X = X0 + tEi , so that

xi = xi0 + t, xj = xj0 if j ̸= i, and |X − X0 | = |t|.
Then ∑
f (X) − f (X0 ) − ni=1 mi (xi − xi0 )
lim = 0.
X→X0 |X − X0 |
and the dierentiability of f at X0 imply that
f (X0 + tEi ) − f (X0 ) − mi t

lim = 0.
t→0 t
Hence,
f (X0 + tEi ) − f (X0 )
lim = mi .
t→0 t
This proves (2.46), since the limit on the left is fxi (X0 ), by denition.
2.17.1 Linear Function

A linear function is a function of the form
L(X) = m1 x1 + m2 x2 + · · · + mn xn , (2.47)
where m1 , m2 , ..., mn are constants. f is

From denition of dierentiability,
dierentiable at X0 if and only if there is a linear function L such that f (X)−f (X0 )
can be approximated so well near X0 by
L(X) − L(X0 ) = L(X − X0 )

that
f (X) − f (X0 ) = L(X − X0 ) + E(X)(|X − X0 |), (2.48)
where
lim E(X) = 0. (2.49)
X→X0
Theorem: If f is dierentiable at X0 , then f is continuous at X0 .
Proof : From L(X) = m1 x1 + m2 x2 + · · · + mn xn , and Schwarz's inequality,
|L(X − X0 )| ≤ M |X − X0 |,
where
M = (m21 + m22 + · · · + m2n )1/2 .
This and f (X) − f (X0 ) = L(X − X0 ) + E(X)(|X − X0 |), imply that
|f (X) − f (X0 )| ≤ (M + |E(X)|)|X − X0 |.

which, with (2.49), implies that f is continuous at X0 .
2.17.2 Dierential
The linear function
L(X) = fx1 (X0 )x1 + fx2 (X0 )x2 + · · · + fxn (X0 )xn .
This function is called the dierential of f at X0 . We will denote it by d X0 f

and its value by (dX0 f )(X).
Thus,
(dX0 f )(X) = fx1 (X0 )x1 + fx2 (X0 )x2 + · · · + fxn (X0 )xn . (2.50)
In terms of the dierential, dierentiability can be rewritten as
f (X) − f (X0 ) − (dX0 f )(X − X0 )

lim = 0.
X→X0 |X − X0 |
For convenience in writing dX0 f , and to conform with standard notation, we intro-
duce the function dxi , dened by
dxi (X) = xi ;
that is, dxi is the function whose value at a point in Rn is the ith coordinate of the
point.
It is the dierential of the function gi (X) = xi . From (2.50),
dX0 f = fx1 (X0 ) dx1 + fx2 (X0 dx2 + · · · + fxn (X0 ) dxn . (2.51)
If we write X = (x, y, . . . , ), then we write
dX0 f = fx (X0 ) dx + fy (X0 ) dy + · · · ,
where dx, dy , . . . are the functions dened by
dx(X) = x, dy(X) = y, . . .
When it is not necessary to emphasize the specic point X0 , (2.51) can be written
more simply as
df = fx1 dx1 + fx2 dx2 + · · · + fxn dxn .
When dealing with a specic function at an arbitrary point of its domain, we may
use the hybrid notation
df = fx1 (X) dx1 + fx2 (X) dx2 + · · · + fxn (X) dxn .

f (x, y) = x2 + 2xy
is dierentiable at every X in Rn .
The dierential of the functions is
df = (2x + 2y) dx + 2x dy.
To nd d X0 f with X0 = (1, 2), we set x0 = 1 and y0 = 2; thus,
d X0 f = 6 dx + 2 dy
(dX0 f )(X − X0 ) = 6(x − 1) + 2(y − 2).
Since f (1, 2) = 5, the dierentiability of f at (1, 2) implies that
f (x, y) − 5 − 6(x − 1) − 2(y − 2)

lim √ = 0.
(x,y)→(1,2) (x − 1)2 + (y − 2)2
Example: The dierential of a function f = f (x) of one variable is given by
dx0 f = f ′ (x0 ) dx,
where dx is the identity function; that is,
dx(t) = t.
For example, if
f (x) = 3x2 + 5x3 ,
then
df = (6x + 15x2 ) dx.
If x0 = −1, then
dx0 f = 9 dx, (dx0 f )(x − x0 ) = 9(x + 1),
and, since f (−1) = −2,

f (x) + 2 − 9(x + 1)
lim = 0.
x→−1 x+1
Remark: Unfortunately, the notation for the dierential is so complicated that it

obscures the simplicity of the concept. The peculiar symbols df , dx, dy , etc., were
introduced in the early stages of the development of calculus to represent very small
(innitesimal) increments in the variables. However, in modern usage they are not
quantities at all, but linear functions. This meaning of the symbol
∫b dx diers from its
meaning in f (x) dx, where it serves merely to identify the variable of integration;
a ∫b
indeed, some authors omit it in the latter context and write simply
a f.
Lemma: If f is dierentiable at X0 , then
f (X) − f (X0 ) = (dX0 f )(X − X0 ) + E(X)|X − X0 |,

where E is dened in a neighborhood of X0 and
lim E(X) = E(X0 ) = 0.

X→X0
Theorem: If f andg are dierentiable at X0 , then so are f + g and f g. The same

is true of f /g if g(X0 ) ̸= 0. The dierentials are given by
dX0 (f + g) = dX0 f + dX0 g,

dX0 (f g) = f (X0 )dX0 g + g(X0 )dX0 f,
and ( )
f g(X0 )dX0 f − f (X0 )dX0 g
dX0 = .
g [g(X0 )]2
2.17.3 A sucient Condition for Dierentiability

Theorem: If fx1 , fx2 , ..., fxn exist on a neighborhood of X0 and are continuous
at X0 , then f is dierentiable at X0 .
Proof : Let X0 = (x10 , x20 , . . . , xn0 ) and suppose that ε > 0. Our assumptions
imply that there is a δ > 0 such that fx1 , fx2 , . . . , fxn are dened in the n-ball
Sδ (X0 ) = {X : |X − X0 | < δ}
and
|fxj (X) − fxj (X0 )| < ε if |X − X0 | < δ, 1 ≤ j ≤ n. (2.52)
Let X = (x1 , x, . . . , xn ) be in Sδ (X0 ). Dene
Xj = (x1 , . . . , xj , xj+1,0 , . . . , xn0 ), 1 ≤ j ≤ n − 1,
and Xn = X. Thus, for 1 ≤ j ≤ n, Xj diers from Xj−1 in the j th component only,

and the line segment from Xj−1 to Xj is in Sδ (X0 ). Now write
∑
n
f (X) − f (X0 ) = f (Xn ) − f (X0 ) = [f (Xj ) − f (Xj−1 )], (2.53)
j=1
and consider the auxiliary functions
g1 (t) = f (t, x20 , . . . , xn0 ),

gj (t) = f (x1 , . . . , xj−1 , t, xj+1,0 , . . . , xn0 ), 2 ≤ j ≤ n − 1, (2.54)
gn (t) = f (x1 , . . . , xn−1 , t),
where, in each case, all variables except t are temporarily regarded as constants.
Since
f (Xj ) − f (Xj−1 ) = gj (xj ) − gj (xj0 ),
the mean value theorem implies that
f (Xj ) − f (Xj−1 ) = gj′ (τj )(xj − xj0 ),
where τj is between xj and xj0 . From (2.54),
b j ),
gj′ (τj ) = fxj (X
where bj
X is on the line segment from Xj−1 to Xj . Therefore,
b j )(xj − xj0 ),
f (Xj ) − f (Xj−1 ) = fxj (X
and (2.53) implies that
∑
n
f (X) − f (X0 ) = b j )(xj − xj0 )
fxj (X
j=1
∑n ∑
n
= fxj (X0 )(xj − xj0 ) + b j ) − fx (X0 )](xj − xj0 ).
[fxj (X j
j=1 j=1
From this and (2.52),

∑ ∑
n
n
f (X) − f (X0 ) − f (X )(x − x ) ≤ ε |xj − xj0 | ≤ nε|X − X0 |,
xj 0 j j0
j=1 j=1
which implies that f is dierentiable at X0 .
2.17.4 Continuously Dierentiable Function

We say that f is continuously dierentiable on a subset S of Rn if S is contained in
an open set on which fx1 , fx2 , ..., f xn are continuous.
The above theorem implies that such a function is dierentiable at each X0 inS .
Example: If
x2 + y 2
f (x, y) = ,
x−y
then
2x x2 + y 2
fx (x, y) = −
x − y (x − y)2
2y x2 + y 2
fy (x, y) = + .
x − y (x − y)2
Since fx and fy are continuous on
S = {(x, y) : x ̸= y},
f is continuously dierentiable on S.
Remark: If fx1 , fx2 , . . . , fxn exist on a neighborhood of X0 and are continuous

at X0 , then f is dierentiable at X0 . These conditions are not necessary for dier-
entiability; that is, a function may be dierentiable at a point X0 even if its rst
partial derivatives are not continuous at X0 .
Example: let {
(x − y)2 sin x−y
1
, x ̸= y,
f (x, y) =
0, x = y.
Then
1 1
fx (x, y) = 2(x − y) sin − cos , x ̸= y,
x−y x−y
and
f (x + h, x) − f (x, x) h2 sin(1/h) − 0
fx (x, x) = lim = lim = 0,
h→0 h h→0 h
so fx exists for all (x, y), but is not continuous on the line y = x.
Example: Let {
(x − y)2 sin x−y
1
, x ̸= y,
f (x, y) =
0, x = y.
The same is true of fy , since
1 1
fy (x, y) = −2(x − y) sin + cos , x ̸= y,
x−y x−y
and
f (x, x + k) − f (x, x) k 2 sin(−1/k) − 0
fy (x, x) = lim = lim = 0.
k→0 k k→0 k
Now,
f (x, y) − f (0, 0) − fx (0, 0)x − fy (0, 0)y
√
x2 + y 2
{ (x−y)2
√ 1
sin x−y , x ̸= y,
2 +y 2
= x
0, x = y,
and Schwarz's inequality implies that

(x − y)2 2(x2 + y 2 ) √
1
√ sin ≤ √ = 2 x2 + y 2 , x ̸= y.
x2 + y 2 x − y x2 + y 2
Therefore,
f (x, y) − f (0, 0) − fx (0, 0)x − fy (0, 0)y

lim √ = 0,
(x,y)→(0,0) x2 + y 2
so f is dierentiable at (0, 0), but fx and fy are not continuous at (0, 0).
2.17.5 Geometric Interpretation of Dierentiability

If a function f of one variable is dierentiable at x0 , then the curve y = f (x) has a
tangent line
y = T (x) = f (x0 ) + f ′ (x0 )(x − x0 ).
The tangent line approximates it so well near x0 that
f (x) − T (x)
lim = 0.
x→x0 x − x0
Moreover, the tangent line is the limit of the secant line through the points
(x1 , f (x0 )) and (x0 , f (x0 )) as x1 approaches x0 . Dierentiability of a function of n
variables has an analogous geometric interpretation. We will illustrate it for n = 2.
If f is dened in a region D in R2 , then the set of points (x, y, z) such that
z = f (x, y), (x, y) ∈ D, (2.55)
is a surface in R3 Geometric interpretation of dierentiability:

z
z = f (x, y)
y
D
Figure 2.8: Domain of the function
If f is dierentiable at X0 = (x0 , y0 ), then the plane
z = T (x, y) = f (X0 ) + fx (X0 )(x − x0 ) + fy (X0 )(y − y0 ) (2.56)
intersects the surface z = f (x, y) at (x0 , y0 , f (x0 , y0 )) and approximates the surface
so well near (x0 , y0 ) that
f (x, y) − T (x, y)
lim √ = 0.
(x,y)→(x0 ,y0 ) (x − x0 )2 + (y − y0 )2
Moreover, (2.56) is the only plane in R3 with these properties.

We say that this plane is tangent to the surface z = f (x, y) at the point
(x0 , y0 , f (x0 , y0 )).
Ta
n ge
nt
pla
z = f (x,y) ne
y
(x 0 , y0 )
Figure 2.9: Geometric interpretation of dierentiability
Show that the tangent plane to the surface z = f (x, y) is the limit of the
secant planes.
Let Xi = (xi , yi ) (i = 1, 2, 3). The equation of the secant plane through the
points (xi , yi , f (xi , yi )) (i = 1, 2, 3) on the surface z = f (x, y) is of the form
z = f (X0 ) + A(x − x0 ) + B(y − y0 ), (2.57)
where A and B satisfy the system
f (X1 ) = f (X0 ) + A(x1 − x0 ) + B(y1 − y0 ),

f (X2 ) = f (X0 ) + A(x2 − x0 ) + B(y2 − y0 ).
Solving for A and B yields
(f (X1 ) − f (X0 ))(y2 − y0 ) − (f (X2 ) − f (X0 ))(y1 − y0 )

A = (2.58)
(x1 − x0 )(y2 − y0 ) − (x2 − x0 )(y1 − y0 )
(f (X2 ) − f (X0 ))(x1 − x0 ) − (f (X1 ) − f (X0 ))(x2 − x0 )
B = (2.59)
(x1 − x0 )(y2 − y0 ) − (x2 − x0 )(y1 − y0 )
if
(x1 − x0 )(y2 − y0 ) − (x2 − x0 )(y1 − y0 ) ̸= 0, (2.60)
which is equivalent to the requirement that X0 , X1 , and X2 do not lie on a line. If

we write
X1 = X0 + tU and X2 = X0 + tV,
where U = (u1 , u2 ) and V = (v1 , v2 ) are xed nonzero vectors, then (2.58), (2.59),
and (2.60) take the more convenient forms
f (X0 +tU)−f (X0 )

t v2 − f (X0 +tV)−f
t
(X0 )
u2
A = , (2.61)
u1 v2 − u2 v1
f (X0 +tV)−f (X0 )
t u1 − f (X0 +tU)−f
t
(X0 )
v1
B = , (2.62)
u1 v2 − u2 v1
and
u1 v2 − u2 v1 ̸= 0.
If f is dierentiable at X0 , then
f (X) − f (X0 ) = fx (X0 )(x − x0 ) + fy (X0 )(y − y0 ) + ε(X)|X − X0 |, (2.63)
where
lim ε(X) = 0. (2.64)
X→X0
Substituting rst X = X0 + tU and then X = X0 + tV in (2.63) and dividing by t

yields
f (X0 + tU) − f (X0 )
= fx (X0 )u1 + fy (X0 )u2 + E1 (t)|U| (2.65)
t
and
f (X0 + tV) − f (X0 )
= fx (X0 )v1 + fy (X0 )v2 + E2 (t)|V|, (2.66)
t
where
E1 (t) = ε(X0 + tU)|t|/t and E2 (t) = ε(X0 + tV)|t|/t,
so
lim Ei (t) = 0, i = 1, 2, (2.67)
t→0
because of (2.64). Substituting (2.65) and (2.66) into (2.61) and (2.62) yields
A = fx (X0 ) + ∆1 (t), B = fy (X0 ) + ∆2 (t), (2.68)
where
v2 |U|E1 (t) − u2 |V|E2 (t)
∆1 (t) =
u 1 v2 − u 2 v1
and
u1 |V|E2 (t) − v1 |U|E1 (t)
∆2 (t) = ,
u1 v2 − u2 v1
so
lim ∆i (t) = 0, i = 1, 2, (2.69)
t→0
because of (2.67).
From (2.57) and (2.68), the equation of the secant plane is
z = f (X0 ) + [fx (X0 ) + ∆1 (t)](x − x0 ) + [fy (X0 ) + ∆2 (t)](y − y0 ).
Therefore, because of (2.69), the secant plane approaches the tangent plane (2.56)
as t approaches zero.
2.18. Maxima and Minima 113
2.18 Maxima and Minima

We say that X0 is a local extreme point of f if there is a δ>0 such that
f (X) − f (X0 )
does not change sign in Sδ (X0 ) ∩ Df .

More specically, X0 is a local maximum point if
f (X) ≤ f (X0 )
or a local minimum point if

f (X) ≥ f (X0 )
for all X in Sδ (X0 ) ∩ Df .
Theorem: Suppose that f is dened in a neighborhood of X0 in Rn and fx1 (X0 ),

fx2 (X0 ), . . . , fxn (X0 ) exist.
Let X0 be a local extreme point of f. Then
fxi (X0 ) = 0, 1 ≤ i ≤ n. (2.70)
Proof : Let E1 = (1, 0, . . . , 0), E2 = (0, 1, 0, . . . , 0), . . . , En = (0, 0, . . . , 1), and
gi (t) = f (X0 + tEi ), 1 ≤ i ≤ n.
Then gi is dierentiable at t = 0, with
gi′ (0) = fxi (X0 ).
Since X0 is a local extreme point of f , t0 = 0 is a local extreme point of gi .
Remark: The converse of theorem is false, since (2.70) fxi (X0 ) = 0, 1 ≤ i ≤ n.

may hold at a point X0 that is not a local extreme point of f.
For example, let X0 = (0, 0) and
f (x, y) = x3 + y 3 .
We say that a point X0 where (2.70) holds is a critical point of f. Thus, if f is

dened in a neighborhood of a local extreme point X0 , then X0 is a critical point
of f; however, a critical point need not be a local extreme point of f.
2.19. Dierentiable Vector Valued Function 114
2.19 Dierentiable Vector Valued Function

A vector-valued function G = (g1 , g2 , . . . , gn ) is dierentiable at
U0 = (u10 , u20 , . . . , um0 )
if its component functions g1 , g2 , ..., gn are dierentiable at U0 .
Lemma: Suppose that G = (g1 , g2 , . . . , gn ) is dierentiable at
U0 = (u10 , u20 , . . . , um0 ),
and dene
 1/2
∑ m (
n ∑ )
∂gi (U0 2
M =  .
∂uj
i=1 j=1
Then, if ε > 0, there is a δ>0 such that
|G(U) − G(U0 )|
<M +ε if 0 < |U − U0 | < δ.
|U − U0 |
Proof : Since g1 , g2 , ..., gn are dierentiable at U0 to gi shows that
gi (U) − gi (U0 ) = (dU0 gi )(U − U0 ) + Ei (U)|(U − U0 |

∑m ∂gi (U0 ) (2.71)
= j=1 ∂uj (uj − uj0 ) + Ei (U)|(U − U0 |,
where
lim Ei (U) = 0, 1 ≤ i ≤ n. (2.72)
U→U0
From Schwarz's inequality,
|gi (U) − gi (U0 )| ≤ (Mi + |Ei (U)|)|U − U0 |,
where
 
∑m ( )2 1/2
∂g (U )
Mi =   .
i 0
∂uj
j=1
Therefore,
( n )1/2
|G(U) − G(U0 )| ∑
≤ (Mi + |Ei (U)|) 2
.
|U − U0 |
i=1
From (2.72),
( n )1/2 ( )1/2
∑ ∑
n
lim (Mi + |Ei (U)|) 2
= Mi2 = M,
U→U0
i=1 i=1
which implies the conclusion.

2.20. The Chain Rule 115
2.20 The Chain Rule

Theorem: Suppose that the real-valued function f is dierentiable at X0 in Rn .
The vector-valued function G = (g1 , g2 , . . . , gn ) is dierentiable at U0 in Rm , and
X0 = G(U0 ).
Then the real-valued composite function h=f ◦G dened by
h(U) = f (G(U)) (2.73)
is dierentiable at U0 , and
dU0 h = fx1 (X0 )dU0 g1 + fx2 (X0 )dU0 g2 + · · · + fxn (X0 )dU0 gn . (2.74)
Proof : First we will show that U0 is an interior point of the domain of h. It is

legitimate to ask if h is dierentiable at U0 . Let X0 = (x10 , x20 , . . . , xn0 ). Note that
xi0 = gi (U0 ), 1 ≤ i ≤ n,
by assumption.
Since f is dierentiable at X0 , which implies that
∑
n
f (X) − f (X0 ) = fxi (X0 )(xi − xi0 ) + E(X)|X − X0 |, (2.75)
i=1
where
lim E(X) = 0.
X→X0
Substituting X = G(U) and X0 = G(U0 ) in (2.75) and recalling (2.73) yields
∑
n
h(U) − h(U0 ) = fxi (X0 )(gi (U) − gi (U0 ))
i=1
+E(G(U))|G(U) − G(U0 )|. (2.76)
Substituting gi (U) − gi (U0 ) = dU0 gi )(U − U0 ) + Ei (U)|U − U0 | into (2.76) yields
∑n
h(U) − h(U0 ) = i=1 fxi (X0 )(dU0 gi )(U − U0 )
∑n
+( i=1 fxi (X0 )Ei (U)) |U − U0 |
+ E(G(U))|G(U) − G(U0 |.
Since
lim E(G(U)) = lim E(X) = 0.
U→U0 X→X0
Due to Lemma we proved in previous module, imply that

∑n
h(U) − h(U0 ) − i=1 fxi (X0 dU0 gi (U − U0 )
= 0.
|U − U0 |
Therefore, h is dierentiable at U0 , and d U0 h is given by (2.74).
Example: Let
f (x, y, z) = 2x2 + 4xy + 3yz,
g1 (u, v) = u2 + v 2 , g2 (u, v) = u2 − 2v 2 , g3 (u, v) = uv,
and
h(u, v) = f (g1 (u, v), g2 (u, v), g3 (u, v)).
Let U0 = (1, −1) and
X0 = (g1 (U0 ), g2 (U0 ), g3 (U0 )) = (2, −1, −1).
Then
fx (X0 ) = 4, fy (X0 ) = 5, fz (X0 ) = −3,
Since
g1 (u, v) = u2 + v 2 , g2 (u, v) = u2 − 2v 2 , g3 (u, v) = uv,
∂g1 (U0 ) ∂g1 (U0 )

= 2, ∂v = −2,
∂u
∂g2 (U0 ) ∂g2 (U0 )
= 2, ∂v = 4,
∂u
∂g3 (U0 ) ∂g3 (U0 )
= −1, ∂v = 1.
∂u
Therefore,
dU0 g1 = 2 du − 2 dv, dU0 g2 = 2 du + 4 dv, dU0 g3 = −du + dv.
According to chain rule we have
dU0 h = fx1 (X0 )dU0 g1 + fx2 (X0 )dU0 g2 + · · · + fxn (X0 )dU0 gn .
dU0 h = fx (X0 ) dU0 g1 + fy (X0 ) dU0 g2 + fz (X0 ) dU0 g3
= 4(2 du − 2 dv) + 5(2 du + 4 dv) − 3(−du + dv)
= 21 du + 9 dv.
Since
dU0 h = hu (U0 ) du + hv (U0 ) dv
we conclude that
hu (U0 ) = 21 and hv (U0 ) = 9. (2.77)
Alternatively: This can also be obtained by writing h explicitly in terms of

(u, v) and dierentiating; thus,
h(u, v) = 2[g1 (u, v)]2 + 4g1 (u, v)g2 (u, v) + 3g2 (u, v)g3 (u, v)
= 2(u2 + v 2 )2 + 4(u2 + v 2 )(u2 − 2v 2 ) + 3(u2 − 2v 2 )uv
= 6u4 + 3u3 v − 6uv 3 − 6v 4 .
Hence,
hu (u, v) = 24u3 + 9u2 v − 6v 3 and hv (u, v) = 3u3 − 18uv 2 − 24v 3 ,
so hu (1, −1) = 21 and hv (1, −1) = 9, consistent with (2.77).
Corollary: Under the assumptions of the chain rule theorem
∂h(U0 ) ∑ ∂f (X0 ) ∂gj (U0 )

n
= , 1 ≤ i ≤ m. (2.78)
∂ui ∂xj ∂ui
j=1
Proof :Substituting
∂gi (U0 ) ∂gi (U0 ) ∂gi (U0 )
dU0 gi = du1 + du2 + · · · + dum , 1 ≤ i ≤ n,
∂u1 ∂u2 ∂um
into (2.74) and collecting multipliers of du1 , du2 ,
dum yields
...,
 
∑
m ∑
n
∂f (X ) ∂g (U )
d U0 h =  0 j 0 
dui .
∂xj ∂ui
i=1 j=1
However, from Theorem ??,

∑
m
∂h(U0 )
d U0 h = dui .
∂ui
i=1
Comparing the last two equations yields (2.78).
Remark: When it is not important to emphasize the particular point X0 , we write
∂h ∑ ∂f ∂gj
n
= , 1 ≤ i ≤ m, (2.79)
∂ui ∂xj ∂ui
j=1
with the understanding that in calculating ∂h(U0 )/∂ui , ∂gj /∂ui is evaluated at U0
and ∂f /∂xj at X0 = G(U0 ).
∂h ∑ ∂f ∂gj
n
= , 1 ≤ i ≤ m, (2.80)
∂ui ∂xj ∂ui
j=1
2.21. Higher derivatives of composite functions 118
with the understanding that in calculating ∂h(U0 )/∂ui , ∂gj /∂ui is evaluated at U0
and ∂f /∂xj at X0 = G(U0 ). By replacing the symbol G with X = X(U); then we
write
h(U) = f (X(U))
and
∂h(U0 ) ∑ ∂f (X0 ) ∂xj (U0 )
n
= ,
∂ui ∂xj ∂ui
j=1
∂h ∑ ∂f ∂xj n
or simply = . (2.81)
∂ui ∂xj ∂ui
j=1
2.21 Higher derivatives of composite functions

Higher derivatives of composite functions can be computed by repeatedly applying
the chain rule.
For example, dierentiating (2.81) with respect to uk yields
∑n ( )
∂2h ∂ ∂f ∂xj
∂uk ∂ui = j=1 ∂uk ∂xj ∂ui
∑n ∑n ( ) (2.82)
∂f ∂ 2 xj ∂xj ∂ ∂f
= j=1 ∂xj ∂uk ∂ui + j=1 ∂ui ∂uk ∂xj .
We must be careful nding ( )

∂ ∂f
,
∂uk ∂xj
which really stands here for
( )
∂ ∂f (X(U))
. (2.83)
∂uk ∂xj
The safest procedure is to write temporarily
∂f (X)
g(X) = ;
∂xj
then (2.83) becomes
∂g(X(U)) ∑ ∂g(X(U)) ∂xs (U)

n
= .
∂uk ∂xs ∂uk
s=1
Since
∂g ∂2f
= ,
∂xs ∂xs ∂xj
this yields
( ) ∑
n
∂ ∂f ∂ 2 f ∂xs
= .
∂uk ∂xk ∂xs ∂xj ∂uk
s=1
Substituting this into (2.82) yields
∂2h ∑
n
∂f ∂ 2 xj ∑
n
∂xj ∑ ∂ 2 f ∂xs
n
= + ∂uk . (2.84)
∂uk ∂ui ∂xj ∂uk ∂ui ∂ui ∂xs ∂xj
j=1 j=1 s=1
To compute hui uk (U0 ) from this formula, we evaluate the partial derivatives of x1 ,
x2 , ..., xn U0 and those of f at X0 = X(U0 ). The formula is valid if x1 , x2 ,
at
..., xn and their rst partial derivatives are dierentiable at U0 and f , fxi , fx2 ,
..., fxn and their rst partial derivatives are dierentiable at X0 .
Example: Let (r, θ) be polar coordinates in the xy -plane; that is,
x = r cos θ, y = r sin θ.
Suppose that f = f (x, y) is dierentiable on a set S, and let
h(r, θ) = f (r cos θ, r sin θ).

We have
∂h ∂f ∂x ∂f ∂y ∂f ∂f
= + = cos θ + sin θ (2.85)
∂r ∂x ∂r ∂y ∂r ∂x ∂y
∂h ∂f ∂x ∂f ∂y ∂f ∂f
= + = −r sin θ + r cos θ ,
∂θ ∂x ∂θ ∂y ∂θ ∂x ∂y
where fx and fy are evaluated at (x, y) = (r cos θ, r sin θ).
Example: Suppose that fx and fy just calculated are dierentiable on an open set
in R . Dierentiating (2.85) with respect to r yields
S 2
( ) ( )
∂2h ∂ ∂f ∂ ∂f
∂r 2
= cos θ ∂r ∂x + sin θ ∂r ∂y
(2.86)
( ) ( )
∂ 2 f ∂x ∂ 2 f ∂y ∂ 2 f ∂x ∂ 2 f ∂y
= cos θ ∂x2 ∂r
+ ∂y ∂x ∂r + sin θ ∂x ∂y ∂r + ∂y 2 ∂r
.
if (x, y) ∈ S . Since
∂x ∂y ∂2f ∂2f
= cos θ, = sin θ, and =
∂r ∂r ∂x ∂y ∂y ∂x
if (x, y) ∈ S . The equation (2.86) yields
∂2h 2
2 ∂ f ∂2f 2
2 ∂ f
= cos θ + 2 sin θ cos θ + sin θ .
∂r2 ∂x2 ∂x ∂y ∂y 2
Dierentiating (2.85) with respect to θ yields
( ) ( )
∂2h ∂f ∂f ∂ ∂f ∂ ∂f
= − sin θ + cos θ + cos θ + sin θ
∂θ ∂r ∂x ∂y ∂θ ∂x ∂θ ∂y
( 2 2
)
∂f ∂f ∂ f ∂x ∂ f ∂y
= − sin θ + cos θ + cos θ +
∂x ∂y ∂x2 ∂θ ∂y ∂x ∂θ
( 2 2
)
∂ f ∂x ∂ f ∂y
+ sin θ + 2 .
∂x ∂y ∂θ ∂y ∂θ
Since
∂x ∂y
= −r sin θ and = r cos θ,
∂θ ∂θ
it follows that
( )
∂2h ∂f ∂f ∂2f ∂2f
= − sin θ + cos θ − r sin θ cos θ −
∂θ ∂r ∂x ∂y ∂x2 ∂y 2
∂2f
+ r(cos2 θ − sin2 θ) .
∂x∂y
Remark: For a composite function of the form
h(t) = f (x1 (t), x2 (t), . . . , xn (t))
where t is a real variable, x1 , x2 , . . . , xn are dierentiable at t0 , and f is dierentiable

at X0 = X(t0 ). We have
∑
n
h′ (t0 ) = fxj (X(t0 ))x′j (t0 ). (2.87)
j=1
Theorem: Let f be continuous at X1 = (x11 , x21 , . . . , xn1 ) and X2 =

(x12 , x22 , . . . , xn2 ) and dierentiable on the line segment L from X1 to X2 .
Then
∑
n
f (X2 ) − f (X1 ) = fxi (X0 )(xi2 − xi1 ) = (dX0 f )(X2 − X1 ) (2.88)
i=1
for some X0 on L distinct from X1 and X2 .
Proof : An equation of L is
X = X(t) = tX2 + (1 − t)X1 , 0 ≤ t ≤ 1.
Our hypotheses imply that the function
h(t) = f (X(t))
is continuous on [0, 1] and dierentiable on (0, 1).

Since
xi (t) = txi2 + (1 − t)xi1 ,
We have
∑
n
h′ (t) = fxi (X(t))(xi2 − xi1 ), 0 < t < 1.
i=1
From the mean value theorem for functions of one variable
h(1) − h(0) = h′ (t0 )
t0 ∈ (0, 1).
for some Since h(1) = f (X2 ) and h(0) = f (X1 ), this implies (2.88) with
X0 = X(t0 ), i.e.,
∑
n
f (X2 ) − f (X1 ) = fxi (X0 )(xi2 − xi1 ) = (dX0 f )(X2 − X1 ).
i=1
Theorem: If fx1 , fx2 , ..., f xn are identically zero in an open region S of Rn , then
f is constant in S.
Proof : We will show that if X0 and X are in S , then f (X) = f (X0 ).

Since S is an open region, S is polygonally connected.
Therefore, there are points
X0 , X1 , . . . , Xn = X
such that the line segment Li from Xi−1 to Xi is in S , 1 ≤ i ≤ n. From mean value
theorem
∑
n
f (Xi ) − f (Xi−1 ) = e i f )(Xi − Xi−1 ),
(dX
i=1
where e
X is on Li and therefore in S.
Therefore,
e i ) = fx (X
fxi (X e i ) = · · · = fxn (X
e i ) = 0,
2
which means that e i f ≡ 0.

dX Hence,
f (X0 ) = f (X1 ) = · · · = f (Xn );
that is, f (X) = f (X0 ) for every X in S .

Motivation: Suppose that f is dened in an n-ball Bρ (X0 ), with ρ > 0.
If X ∈ Bρ (X0 ), then
X(t) = X0 + t(X − X0 ) ∈ Bρ (X), 0 ≤ t ≤ 1,
so the function
h(t) = f (X(t))
is dened for 0 ≤ t ≤ 1.
We know that
∑
n
h′ (t) = fxi (X(t)(xi − xi0 ).
i=1
2.22. rth Dierential 122
If f is dierentiable in Bρ (X0 ), and
( n )
∑
n
∂ ∑ ∂f (X(t))
h′′ (t) = (xi − xi0 ) (xj − xj0 )
∂xj ∂xi
j=1 i=1
∑n
∂ 2 f (X(t))
= (xi − xi0 )(xj − xj0 )
∂xj ∂xi
i,j=1
If fx1 , fx2 , ..., fxn are dierentiable in Bρ (X0 ). Continuing in this way, we see
that
∑
n
∂ r f (X(t))
h(r) (t) = (xi − xi1 ,0 )(xi2 − xi2 ,0 )
∂xir ∂xir−1 · · · ∂xi1 1
i1 ,i2 ,...,ir =1
· · · (xir − xir ,0 )
if all partial derivatives of f of order ≤r−1 are dierentiable in Bρ (X0 ).
2.22 rth Dierential

Suppose that r ≥ 1 and all partial derivatives of f of order ≤ r − 1 are dierentiable
in a neighborhood of X0 .
rth dierential of f at X0 , denoted by dX0 f , is dened by
(r)
Then the
(r)
∑
n
∂ r f (X0 )
d X0 f = dxi1 dxi2 · · · dxir , (2.89)
∂xir ∂xir−1 · · · ∂xi1
i1 ,i2 ,...,ir =1
where dx1 , dx2 , . . . , dxn are the dierentials, that is, dxi is the function whose value
point in R is the ith coordinate of the point. For convenience, we dene
at a
n
(0)
(dX0 f ) = f (X0 ).
(1)
Notice that d X0 f = d X0 f .
Remark: Suppose that r≥1 and all partial derivatives of f of order ≤ r−1 are
dierentiable in a neighborhood of X0 , the value of
∂ r f (X0 )
∂xir ∂xir−1 · · · ∂xi1
depends only on the number of times f is dierentiated with respect to each variable,
and not on the order in which the dierentiations are performed.
Remark: The dierential can be rewritten as
(r)
∑ r! ∂ r f (X0 )
d X0 f = (dx1 )r1 (dx2 )r2 · · · (dxn )rn , (2.90)
r
r1 !r2 ! · · · rn ! ∂x1 ∂xr22 · · · ∂xrnn
r1
2.22. rth Dierential 123
∑
where r indicates summation over all ordered n-tuples (r1 , r2 , . . . , rn ) of nonneg-
ative integers such that
r1 + r2 + · · · + rn = r
and ∂xri i is omitted from the denominators of all terms in (2.90) for which ri = 0.
In particular, if n = 2,
r ( ) r
∑
(r) r ∂ f (x0 , y0 )
d X0 f = (dx)j (dy)r−j .
j ∂xj ∂y r−j
j=0
Example: Let
1
f (x, y) = ,
1 + ax + by
where a and b are constants.
Then
∂ r f (x, y) r aj br−j
= (−1) r! ,
∂xj ∂y r−j (1 + ax + by)r+1
so
r ( )
∑
(r) (−1)r r! r j r−j
d X0 f = r+1
a b (dx)j (dy)r−j
(1 + ax0 + by0 ) j
j=0
(−1)r r!
= (a dx + b dy)r
(1 + ax0 + by0 )r+1
if 1 + ax0 + by0 ̸= 0.
Let  
∑
n
f (X) = exp − aj xj  ,
j=1
where a1 , a2 , ..., an are constants. Then

 
∂ r f (X) ∑
n
= (−1)r ar11 ar22 · · · arnn exp − aj xj  .
∂xr11 ∂xr22 · · · ∂xrnn
j=1
Therefore,
( )
(r) r! ∑
(dX0 f )(Φ) = (−1) r
ar11 ar22 · · · arnn (dx1 )r1 (dx2 )r2 · · · (dxn )rn
r
r !r
1 2 ! · · · rn !
 
∑
n
× exp − aj xj0 
j=1
 
∑
n
= (−1)r (a1 dx1 + a2 dx2 + · · · + an dxn )r exp − aj xj0 
j=1
2.23. Taylor's Theorem for Functions of n Variables 124
2.23 Taylor's Theorem for Functions of n Variables

Theorem: Suppose that f and its partial derivatives of order ≤k are dierentiable
at X0 and X in Rn and on the line segment L connecting them.
Then
∑
k
1 (r) 1 (k+1)
f (X) = (d f )(X − X) + (d e f )(X − X0 ) (2.91)
r! X0 (k + 1)! X
r=0
for some e
X on L distinct from X0 and X.
Proof : Dene
h(t) = f (X0 + t(X − X0 )). (2.92)
With Φ = X − X0 , our assumptions and the discussion preceding Denition of

dierentials imply that h, h′ , ..., h(k+1) exist on [0, 1].
From Taylor's theorem for functions of one variable,
∑
k
h(r) (0) h(k+1) (τ )
h(1) = + , (2.93)
r! (k + 1)!
r=0
for some τ ∈ (0, 1).

From (2.92),
h(0) = f (X0 ) and h(1) = f (X). (2.94)
We have Φ = X − X0 ,
(r)
h(r) (0) = (dX0 f )(X − X0 ), 1 ≤ r ≤ k, (2.95)
( )
h(k+1) (τ ) = dk+1
e f (X − X0 ) (2.96)
X
where
e = X0 + τ (X − X0 )
X
is on L and distinct from X0 and X.
Substituting (2.94), (2.95), and (2.96) into (2.93) yields (2.91).
Let
1
f (x, y) = ,
1 + ax + by
where a and b are constants.
Then
∂ r f (x, y) r aj br−j
= (−1) r! ,
∂xj ∂y r−j (1 + ax + by)r+1
so
r ( )
∑
(r) (−1)r r! r j r−j
d X0 f = r+1
a b (dx)j (dy)r−j
(1 + ax0 + by0 ) j
j=0
(−1)r r!
= (a dx + b dy)r
(1 + ax0 + by0 )r+1
if 1 + ax0 + by0 ̸= 0.
Example: The Taylor series with X0 = (0, 0) and Φ = (x, y) imply that if 1 + ax +
by > 0, then
1 ∑ k
(ax + by)k+1
= (−1)r (ax + by)r + (−1)k+1
1 + ax + by (1 + aτ x + bτ y)k+2
r=0
for some τ ∈ (0, 1). (Note that τ depends on k as well as (x, y).)
Remark: By analogy with the situation for functions of one variable, we dene the
k th Taylor polynomial of f about X0 by
∑
k
1 (r)
Tk (X) = (d f )(X − X0 ).
r! X0
r=0
If the dierentials exist; then we have
1 (k+1)
f (X) = Tk (X) + (d e f )(X − X0 ).
(k + 1)! X
Theorem: Suppose that f and its partial derivatives of order ≤ k − 1 are dieren-
tiable in a neighborhood N of a point X0 in Rn and all k th-order partial derivatives
of f are continuous at X0 . Then
f (X) − Tk (X)
lim = 0. (2.97)
X→X0 |X − X0 |k
Proof : If ε > 0, there is a δ > 0 such that Bδ (X0 ) ⊂ N and all k th-order partial
derivatives of f satisfy the inequality

e
∂ k f (X) ∂ k f (X0 )
e ∈ Bδ (X0 ).
− < ε, X (2.98)
∂xik ∂xik−1 · · · ∂xi1 ∂xik ∂xik−1 · · · ∂xi1
Now suppose that X ∈ Bδ (X0 ). From Taylor series expansion, with k replaced by
k − 1,
1 (k)
f (X) = Tk−1 (X) + (d e f )(X − X0 ), (2.99)
k! X
where e is some point on the line segment from X0
X to X and is therefore in Bδ (X0 ).
We can rewrite (2.99) as
1 [ (k) (k)
]
f (X) = Tk (X) + (d e f )(X − X0 ) − (dX0 f )(X − X0 ) . (2.100)
k! X
But denition of dierential and (2.98) imply that

(k) (k)
(dX
e f )(X − X0 ) − (d X0 f )(X − X0 ) < nk ε|X − X0 |k (2.101)
which implies that
|f (X) − Tk (X)| nk ε
< , X ∈ Bδ (X0 ),
|X − X0 |k k!
from (2.100). This implies (2.97).
2.23.1 Positive Denite

Let r be a positive integer and X0 = (x10 , x20 , . . . , xn0 ). A function of the form
∑
p(X) = ar1 r2 ...rn (x1 − x10 )r1 (x2 − x20 )r2 · · · (xn − xn0 )rn , (2.102)
r
where the coecients {ar1 r2 ...rn } are constants and the summation is over all n-tuples
of nonnegative integers (r1 , r2 , . . . , rn ) such that
r1 + r2 + · · · + rn = r,
is a homogeneous polynomial of degree r in X − X0 , provided that at least one of the

coecients is nonzero. For example, if f satises the conditions of rth dierential,
then the function
(r)
p(X) = (dX0 f )(X − X0 )
is such a polynomial if at least one of the rth-order mixed partial derivatives of f
at X0 is nonzero. Clearly, p(X0 ) = 0 if p is a homogeneous polynomial of degree
r ≥ 1 in X − X0 .
If p(X) ≥ 0 for all X, we say that p is positive semidenite ; if p(X) > 0 except
when X = X0 , p is positive denite . Similarly, p is negative semidenite if p(X) ≤ 0
or negative denite if p(X) < 0 for all X ̸= X0 . In all these cases, p is semidenite .
With p as in (2.102),
p(−X + 2X0 ) = (−1)r p(X),
so p cannot be semidenite if r is odd.
Example: The polynomial
p(x, y, z) = x2 + y 2 + z 2 + xy + xz + yz
is homogeneous of degree 2 in X = (x, y, z). We can rewrite p as
1[ ]
p(x, y, z) = (x + y)2 + (y + z)2 + (z + x)2 .
2
so p is nonnegative, and p(x, y, z) = 0 if and only if
x + y = y + z = z + x = 0,
which is equivalent to (x, y, z) = (0, 0, 0). Therefore, p is positive denite and −p is

negative denite.
Example: The polynomial
p1 (x, y, z) = x2 + y 2 + z 2 + 2xy
p1 (x, y, z) = (x + y)2 + z 2 ,
so p1 is nonnegative. Since p1 (1, −1, 0) = 0, p1 is positive semidenite and −p1 is

negative semidenite.
The polynomial
p2 (x, y, z) = x2 − y 2 + z 2
is not semidenite, since, for example,
p2 (1, 0, 0) = 1 and p2 (0, 1, 0) = 1.
Theorem: f and its partial derivatives of order ≤ k − 1 are dieren-

Suppose that
tiable in a neighborhood N of a point X0 in R and all k th-order partial derivatives
n
of f are continuous at X0 . with k ≥ 2, and
(r) (k)
dX0 f ≡ 0 (1 ≤ r ≤ k − 1), dX0 f ̸≡ 0. (2.103)
Then
(k)
• X0 is not a local extreme point of f unless d X0 f is semi-denite as a polynomial
in X − X0 .
In particular, X0 is not a local extreme point of f if k is odd.
(k)
• X0 is a local minimum point of f if d X0 f is positive denite, or a local maxi-
(k)
mum point if dX f is negative denite.
0
(k)
• Ifd X0 f is semidenite, then X0 may be a local extreme point of f, but it need
not be.
Corollary: Suppose that f, fx , and fy are dierentiable in a neighborhood of a

critical point X0 = (x0 , y0 ) of f and fxx , fyy , and fxy are continuous at (x0 , y0 ).
Let
D = fxx (x0 , y0 )fxy (x0 , y0 ) − fxy
2
(x0 , y0 ).
Then
• (x0 , y0 ) is a local extreme point of f

D > 0; (x0 , y0 ) is a local minimum
if
point if fxx (x0 , y0 ) > 0, or a local maximum point if fxx (x0 , y0 ) < 0.
• (x0 , y0 ) is not a local extreme point of f if D < 0.

Proof : Write (x − x0 , y − y0 ) = (u, v) and
(2)
p(u, v) = (dX0 f )(u, v) = Au2 + 2Buv + Cv 2 ,
where A = fxx (x0 , y0 ), B = fxy (x0 , y0 ), and C = fyy (x0 , y0 ), so
D = AC − B 2 .
If D > 0, then A ̸= 0, and we can write
( ) ( )
2B B2 2 B2
2
p(u, v) = A u + uv + 2 v + C − v2
A A A
( )
B 2 D 2
= A u+ v + v .
A A
(2)
This cannot vanish unless u = v = 0. Hence, dX0 f is positive denite if A>0 or
negative denite if A < 0, and Theorem implies the rst part of the corollary.
If D < 0, there are three possibilities:
( )
1. A ̸= 0; then p(1, 0) = A and p −B
A,1 =
D
A.
( )
2. C ̸= 0; then p(0, 1) = C and p 1, − B
C =
D
C.
3. A = C = 0; then B ̸= 0 and p(1, 1) = 2B and p(1, −1) = −2B .

In each case the two given values of p dier in sign, so X0 is not a local extreme
point of f, from Theorem part I.
Example: If
2 +by 2
f (x, y) = eax .
We have
fx (x, y) = 2axf (x, y), fy (x, y) = 2byf (x, y),
so
fx (0, 0) = fy (0, 0) = 0,
and (0, 0) is a critical point of f.
To apply Corollary, we calculate
fxx (x, y) = (2a + 4a2 x2 )f (x, y),
fyy (x, y) = (2b + 4b2 y 2 )f (x, y),
fxy (x, y) = 4abxyf (x, y).

Therefore,
D = fxx (0, 0)fyy (0, 0) − fxy

2
(0, 0) = (2a)(2b) − (0)(0) = 4ab.
Corollary implies that (0, 0) is a local minimum point if a and b are positive. A
local maximum if a and b are negative. Neither if one is positive and the other is
negative. Corollary does not apply if a or b is zero.
Chapter 3
Integral Calculus
Attempting to formulate denition of Riemann integral for a function dened on an

innite or semi-innite interval would introduce questions concerning convergence
of the resulting Riemann sums, which would be innite series.
3.1 Locally Integrable Functions

We say f is locally integrable on an interval I if f is integrable on every nite closed
subinterval of I.
For example,
f (x) = sin x
is locally integrable on (−∞, ∞).
1
g(x) =
x(x − 1)
is locally integrable on (−∞, 0), (0, 1), and (1, ∞).
The function
√
h(x) = x
is locally integrable on [0, ∞).
If f is locally integrable on [a, b), we dene
∫ b ∫ c
f (x) dx = lim f (x) dx (3.1)
a c→b− a
if the limit exists (nite). To include the case where b = ∞, we adopt the convention
that ∞− = ∞.
Remarks:
• The limit in (3.1) always exists if [a, b) is nite and f is locally integrable and
bounded on [a, b).
• In this case, the denition of Riemann integral and locally integrable function
∫b
assign the same value to
a f (x) dx no matter how f (b) is dened. However,
the limit may also exist in cases where b=∞ or b < ∞ and f is unbounded
as x approaches b from the left.
3.1. Locally Integrable Functions 130
• In these cases, the denition of locally integrable assigns a value to an integral

∫b
that does not exist in the sense of Riemann integral, and
a f (x) dx is said to
be an improper integral that converges to the limit in (3.1).
Remarks:
∫b
• We also say in this case that f is integrable on [a, b) and that a f (x) dx exists .
If the limit in (3.1) does not exist (nite), we say that the improper integral
∫b
a f (x) dx diverges , and f is nonintegrable on [a, b).
∫c ∫b
• In particular, if limc→b− a f (x) dx = ±∞, we say that
a f (x) dx diverges to
±∞, and we write
∫ b ∫ b
f (x) dx = ∞ or f (x) dx = −∞,
a a
whichever the case may be.
If f is locally integrable on (a, b], we dene
∫ b ∫ b
f (x) dx = lim f (x) dx
a c→a+ c
provided that the limit exists (nite).

To include the case where a = −∞, we adopt the convention that −∞+ = −∞.
If f is locally integrable on (a, b), we dene
∫ b ∫ α ∫ b
f (x) dx = f (x) dx + f (x) dx,
a a α
where a < α < b, provided that both improper integrals on the right exist (nite).
∫b
Remarks: The existence and value of
a f (x) dx according to the above denition
do not depend on the particular choice of α in (a, b).
When we wish to distinguish between improper integrals and integrals in the

sense of denition of Riemann integral, we will call the latter proper integrals .
1 1
f (x) = 2x sin − cos
x x
is locally integrable and the derivative of
1
F (x) = x2 sin
x
on [−2/π, 0).
Hence,
∫
c
1 c 1 4 2
f (x) dx = x sin = c2 sin + 2
−2/π x −2/π c π
∫ 0 ( )
2 1 4 4
f (x) dx = lim c sin + 2 = 2 .
−2/π c→0− c π π
However, this is not an improper integral, even though f (0) is not dened and
cannot be dened so as to make f continuous at 0. If we dene f (0) arbitrarily
(say f (0) = 10), then f ∫is bounded on the closed interval [−2/π, 0] and continuous
0
except at 0. Therefore, −2/π f (x) dx exists and equals 4/π 2 as a proper integral, in
the sense of denition of improper integral.

f (x) = (1 − x)−p
is locally integrable on [0, 1).
If p ̸= 1 and 0 < c < 1,
∫ c
−p (1 − x)−p+1 c (1 − c)−p+1 − 1
(1 − x) dx = = .
0 p−1 0 p−1
Hence,
∫ {
c
−p (1 − p)−1 , p < 1,
lim (1 − x) dx =
c→1− 0 ∞, p > 1.
For p = 1, ∫ c
lim (1 − x)−1 dx = − lim log(1 − c) = ∞.
c→1− 0 c→1−
Hence,
∫ {
1
−p (1 − p)−1 , p < 1,
(1 − x) dx =
0 ∞, p ≥ 1.

f (x) = x−p
is locally integrable on [1, ∞).
If p ̸= 1 and c > 1,
∫
c
−p x−p+1 c c−p+1 − 1
x dx = = .
1 −p + 1 1 −p + 1
Hence,
∫ {
c
−p (p − 1)−1 , p > 1,
lim x dx =
c→∞ 1 ∞, p < 1.
For p = 1, ∫ c
lim x−1 dx = lim log c = ∞.
c→∞ 1 c→∞
Hence,
∫ ∞ {
(p − 1)−1 , p > 1,
x−p dx =
1 ∞, p ≤ 1.
Example: If 1 < c < ∞, then
∫ ∫ c
c
1 1 c
1 1
2 1
log dx = − log x dx = − (log x) = − (log c)2 .
1 x x 1 x 2 1 2
Hence, ∫ c
1 1
lim log dx = −∞,
c→∞ 1 x x
so ∫ ∞
1 1
log dx = −∞.
1 x x
The function f (x) = log x is locally integrable on (0, 1], but unbounded as
x → 0+. Since
∫ 1
1
lim log x dx = lim (x log x − x) = −1 − lim (c log c − c) = −1,
c→0+ c c→0+ c c→0+
Denition ?? yields ∫ 1
log x dx = −1.
0
The function f (x) = cos x is locally integrable on [0, ∞) and
∫ c
lim cos x dx = lim sin c
c→∞ 0 c→∞
∫∞
does not exist; thus,
0 cos x dx diverges, but not to ±∞.
In connection with Denition
∫α ∫b ??, it is important to recognize ∫that the improper
integrals
a f (x) dx and
α f (x) dx must converge separately for
b
a f (x) dx to con-
verge. For example, the existence of the symmetric limit
∫ R
lim f (x) dx,
R→∞ −R
∫∞ ∫∞
which is called the principal value of −∞ f (x) dx, does not imply that −∞ f (x) dx
converges; thus,
∫ R
lim x dx = lim 0 = 0,
R→∞ −R R→∞
∫∞ ∫0 ∫∞
but
0 x dx and
−∞ x dx diverge and therefore so does −∞ x dx.
Theorem: Suppose that

∫b f1 , f∫2 , . . . , fn are locally
∫ b integrable on [a, b).
b
The integrals f
a 1 (x) dx, a 2 f (x) dx, . . . , a fn (x) dx converge.
∫b
Let c1 , c2 , . . . , cn be constants. Then
a (c1 f + c2 f1 + · · · + cn fn )(x) dx converges.
Furthermore,
∫ b ∫ b
(c1 f1 + c2 f2 + · · · + cn fn )(x) dx = c1 f1 (x) dx
a a
∫ b
+c2 f2 (x) dx
a
∫ b
+ · · · + cn fn (x) dx.
a
Proof : Ifa < c < b, then

∫ c ∫ c
(c1 f1 + c2 f2 + · · · + cn fn )(x) dx = c1 f1 (x) dx
a a
∫ c
+c2 f2 (x) dx
a
∫ c
+ · · · + cn fn (x) dx.
a
Letting c → b− yields the stated result.
∫b
Theorem: If f is nonnegative and locally integrable on [a, b), then
a f (x) dx con-
verges if the function ∫ x
F (x) = f (t) dt
a
∫b
is bounded on [a, b), and
a f (x) dx = ∞ if it is not.
These are the only possibilities, and
∫ b
f (t) dt = sup F (x)
a a≤x<b
in either case.
Proof : The function ∫ x

F (x) = f (t) dt
a
is nondecreasing on [a, b).
Recall: Suppose that f is monotonic on (a, b) and dene
α = inf f (x), β = sup f (x).

a<x<b a<x<b
If f is nondecreasing, then f (a+) = α and f (b−) = β .
Remarks: We often write ∫ b

f (x) dx < ∞
a
to indicate that an improper integral of a nonnegative function converges.

∫b
Similarly, if f is nonpositive and
a f (x) dx converges, we write
∫ b
f (x) dx > −∞
a
because a divergent integral of this kind can only diverge to −∞.
• These conventions do not apply to improper integrals of functions that assume

both positive and negative values in (a, b), since they may diverge without
diverging to ±∞.
3.1.1 The Comparison Test

Theorem: If f and g are locally integrable on [a, b) and
0 ≤ f (x) ≤ g(x), a ≤ x < b, (3.2)
then
∫b ∫b
1.
a f (x) dx < ∞ if
a g(x) dx < ∞
∫b ∫b
2.
a g(x) dx = ∞ if
a f (x) dx = ∞.
Proof : Since
0 ≤ f (x) ≤ g(x), a ≤ x < b,
we have ∫ x ∫ x
f (t) dt ≤ g(t) dt, a ≤ x < b.
a a
So ∫ x ∫ x
sup f (t) dt ≤ sup g(t) dt.
a≤x<b a a≤x≤b a
∫b
If
a g(x) dx < ∞, the right side of this inequality is nite by the previous Theorem,
so the left side is also.
∫b
This implies that f (x) dx < ∞.
a ∫b
The proof is by contradiction. g(x) dx < ∞, then (1)
If implies that
∫b ∫b a
a f (x) dx < ∞, contradicting the assumption that

a f (x) dx = ∞.
Example: Determine the convergence of the improper integral
∫ 1
2 + sin πx
I= dx.
0 (1 − x)p
Solution: We are going to show that the improper integral converges if p < 1.
Since
2 + sin πx 3
0< ≤ , 0 ≤ x < 1.
(1 − x) p (1 − x)p
We have ∫ 1
3 dx
< ∞, p < 1.
0 (1 − x)p
Example: Determine the convergence of the improper integral
∫ 1
2 + sin πx
I= dx.
0 (1 − x)p
Solution: However, I diverges if p ≥ 1, since
1 2 + sin πx
0< ≤ , 0 ≤ x < 1,
(1 − x)p (1 − x)p
and ∫ 1
dx
= ∞, p ≥ 1.
0 (1 − x)p
Remark: If f is any function (not necessarily nonnegative) locally integrable on

[a, b). If a1 and c are in [a, b), then
∫ c ∫ a1 ∫ c
f (x) dx = f (x) dx + f (x) dx.
a a a1
∫ a1
Since f (x) dx is a proper integral, on letting c → b− we conclude that if
a ∫b ∫b
either of the improper integrals
a f (x) dx and
a1 f (x) dx converges then so does
the other, and in this case
∫ b ∫ a1 ∫ b
f (x) dx = f (x) dx + f (x) dx.
a a a1
Remark: This means that any theorem implying convergence or divergence of

∫b
an improper integral
af (x) dx remains valid if its hypotheses are satised on a
subinterval [a1 , b) of [a, b) rather than on all of [a, b).
For example, the comparison test remains valid if we have
0 ≤ f (x) ≤ g(x), a1 ≤ x < b,
where a1 is any point in [a, b).

From this, you can see that if f (x) ≥ 0 on some subinterval [a1 , b) of [a, b), but
not necessarily for all x in [a, b), we can still use the convention introduced earlier
∫b
for positive functions; that is, we can write f (x) dx < ∞ if the improper integral
∫b a
converges or
a f (x) dx = ∞ if it diverges.
Theorem: Suppose that f and g are locally integrable on [a, b), g(x) > 0 and
f (x) ≥ 0 on some subinterval [a1 , b) of [a, b), and
f (x)
lim = M. (3.3)
x→b− g(x)
∫b ∫b
• If 0 < M < ∞, then
a f (x) dx and
a g(x) dx converge or diverge together.
∫b ∫b
• If M =∞ and
a g(x) dx = ∞, then
a f (x) dx = ∞.
∫b ∫b
• If M =0 and
a g(x) dx < ∞, then
a f (x) dx < ∞.
Proof : From (3.3), there is a point a2 in [a1 , b) such that
M f (x) 3M
0< < < , a2 ≤ x < b,
2 g(x) 2
and therefore
M 3M
g(x) < f (x) < g(x), a2 ≤ x < b. (3.4)
2 2
The rst inequality in (3.4) imply that
∫ b ∫ b
g(x) dx < ∞ if f (x) dx < ∞.
a2 a2
The second inequality in (3.4) imply that
∫ b ∫ b
f (x) dx < ∞ if g(x) dx < ∞.
a2 a2
∫b ∫b
Therefore,
a2 f (x) dx and a2 g(x) dx converge or diverge together, and in the latter
case they must diverge to ∞, since their integrands are nonnegative. If M = ∞,
there is a point a2 in [a1 , b) such that
f (x) ≥ g(x), a2 ≤ x ≤ b,
∫b
We have
a f (x) dx = ∞. If M = 0, there is a point a2 in [a1 , b) such that
f (x) ≤ g(x), a2 ≤ x ≤ b,
∫b
so we have
a f (x) dx < ∞.
3.2. Absolute integrability 137
3.2 Absolute integrability

We say that
∫b f is absolutely integrable on [a, b) if∫ f is locally integrable on [a, b) and
|f (x)| dx < ∞. In this case we also say that a f (x) dx converges absolutely or is
b
a
absolutely convergent .
Remark: If f is nonnegative and integrable on [a, b), then f is absolutely integrable
on [a, b), since |f | = f .
Example: Since
sin x 1

xp ≤ xp
∫∞
and
1 x−p dx < ∞ if p > 1.
The comparison theorem implies that
∫ ∞
| sin x|
dx < ∞, p > 1.
1 xp
The function
sin x
f (x) =
xp
is absolutely integrable on [1, ∞) if p > 1.
Example: It is not absolutely integrable on [1, ∞) if p ≤ 1.

To see this, we rst consider the case where p = 1.
Let k be an integer greater than 3. Then
∫ kπ | sin x| ∫ kπ | sin x|
1 x dx > π x dx
∑k−1 ∫ (j+1)π | sin x|
= j=1 jπ x dx (3.5)
∑k−1 ∫ (j+1)π
> 1
j=1 (j+1)π jπ | sin x| dx.
But
∫ (j+1)π ∫ π
| sin x| dx = sin x dx = 2,
jπ 0
so (3.5) implies that
∫
2∑ 1
k−1
kπ
| sin x|
dx > . (3.6)
1 x π j+1
j=1
However,
∫ j+2
1 dx
≥ , j = 1, 2, . . . ,
j+1 j+1 x
so (3.6) implies that
∫ k−1 ∫
kπ
| sin x| 2 ∑ j+2 dx
>
1 x π x
j=1 j+1
∫
2 k+1 dx 2 k+1
= = log .
π 2 x π 2
3.3. Nonoscillatory and Oscillatory Functions 138
Since limk→∞ log[(k + 1)/2] = ∞, implies that

∫ ∞
| sin x|
dx = ∞.
1 x
Now implies that
∫ ∞
| sin x|
dx = ∞, p ≤ 1. (3.7)
1 xp
∫b ∫b
Theorem: If f is locally integrable on [a, b) and |f (x)| dx < ∞,
a then
a f (x) dx
converges; that is, an absolutely convergent integral is convergent.
Proof : If
g(x) = |f (x)| − f (x).
Then
0 ≤ g(x) ≤ 2|f (x)|
∫b
and
a g(x) dx < ∞, because of comparison theorem and the absolute integrability
of f. Since
f = |f | − g,
∫b
Due to comparison test, we can conclude that
a f (x) dx converges.
3.3 Nonoscillatory and Oscillatory Functions

A function f is nonoscillatory at b− (= ∞ if b = ∞) if f is dened on [a, b) and
does not change sign on some subinterval [a1 , b) of [a, b).
If f changes sign on every such subinterval, f is oscillatory at b−.
Remark: For a function that is locally integrable on

∫b [a, b) and nonoscillatory at b−,
convergence and absolute convergence of
a f (x) dx amount to the same thing, so
absolute convergence is not an interesting concept in connection with such functions.
However, an oscillatory function may be integrable, but not absolutely inte-
grable, on [a, b), as the next example shows. We then say that
∫b f is conditionally
integrable on [a, b), and that
a f (x) dx converges conditionally .
3.4 Conditional convergence

An oscillatory function may be integrable, but not absolutely integrable, on [a, b),
as the next example shows. We then say that
∫b f is conditionally integrable on [a, b),
and that
a f (x) dx converges conditionally .
Example: The integral ∫ ∞
sin x
I(p) = dx
1 xp
3.5. Dirichlet's Test 139
is not absolutely convergent if 0 < p ≤ 1.

We will show that it converges conditionally for these values of p.
Integration by parts yields
∫ ∫
c
sin x − cos c c
cos x
p
dx = + cos 1 − p dx. (3.8)
1 x cp 1 xp+1
Since cos x
1
p+1 ≤ p+1
x x
∫∞
and x−p−1 dx < ∞ if p > 0, the comparison
1 theorem implies that x−p−1 cos x is
absolutely integrable [1, ∞) if p > 0.
Therefore, we have an absolutely convergent integral, this implies that
x−p−1 cos x is integrable [1, ∞) if p > 0.
Letting c→∞ in (3.8), we nd that I(p) converges, and
∫ ∞
cos x
I(p) = cos 1 − p dx if p > 0.
1 xp+1
∫∞ | sin x|
This and
1 xp dx = ∞, p ≤ 1, imply that I(p) converges conditionally if
0 < p ≤ 1.
3.5 Dirichlet's Test

∫x
Theorem: Suppose that f is continuous and its antiderivative F (x) = a f (t) dt is
bounded on [a, b).
′
Let g be absolutely integrable on [a, b), and suppose that
lim g(x) = 0. (3.9)

x→b−
∫b
Then
a f (x)g(x) dx converges.
Proof : The continuous function fg is locally integrable on [a, b). Integration by

parts yields
∫ c ∫ c
f (x)g(x) dx = F (c)g(c) − F (x)g ′ (x) dx, a ≤ c < b. (3.10)
a a
The comparison test implies that the integral on the right converges absolutely
∫b
as c → b−, since
a |g ′ (x)| dx < ∞ by assumption, and
|F (x)g ′ (x)| ≤ M |g ′ (x)|,
where M is an upper bound for |F | on [a, b). Moreover, (3.9) and the boundedness
of F imply that limc→b− F (c)g(c) = 0.
Letting c → b− in (3.10) yields

∫ b ∫ b
f (x)g(x) dx = − F (x)g ′ (x) dx,
a a
3.6. Rectangles in Rn 140
where the integral on the right converges absolutely.
Remark: Dirichlet's test is useful only if f is oscillatory at b−, since

∫ b it can be shown
that if f is nonoscillatory at b− and F is bounded on [a, b), then a |f (x)g(x)| dx <
∞ if only g is locally integrable and bounded on [a, b).
Remark: Dirichlet's test can also be used to show that certain integrals diverge.
Example: For example, ∫ ∞

xq sin x dx
1
diverges if q > 0, but none of the other tests that we have studied so far implies
this. It is not enough to argue that the integrand does not approach zero as x→∞
(a common mistake), since this does not imply divergence. To see that the integral
∫x
diverges, we observe that if it converged for some q > 0, then F (x) = 1 xq sin x dx
would be bounded on [1, ∞).
We could let
f (x) = xq sin x and g(x) = x−q
in Dirichlet's test and conclude that
∫ ∞
sin x dx
1
also converges. This is false.
3.6 Rectangles in Rn
The
S1 × S2 × · · · × Sn
of subsets S1 , S2 , . . . , Sn of R is the set of points (x1 , x2 , . . . , xn ) in Rn such that
x1 ∈ S1 , x2 ∈ S2 , . . . , xn ∈ Sn . For example, the Cartesian product of the two closed
intervals
[a1 , b1 ] × [a2 , b2 ] = {(x, y) : a1 ≤ x ≤ b1 , a2 ≤ y ≤ b2 }
is a rectangle in R2 with sides parallel to the x- and y -axes.
The Cartesian product of three closed intervals
[a1 , b1 ] × [a2 , b2 ] × [a3 , b3 ] = {(x, y, z) : a1 ≤ x ≤ b1 ,

a2 ≤ y ≤ b2 , a3 ≤ z ≤ b3 }
is a rectangular parallelepiped in R3 with faces parallel to the coordinate axes. A

coordinate rectangle R in R is the Cartesian product of
n n closed intervals; that is,
R = [a1 , b1 ] × [a2 , b2 ] × · · · × [an , bn ].

3.6. Rectangles in Rn 141
b2
a2
x
a1 b1
Figure 3.1: Rectangle in R2

z
Figure 3.2: Rectangular parallelepiped in R3
The content of R is
V (R) = (b1 − a1 )(b2 − a2 ) · · · (bn − an ).
The numbers b1 − a1 , b2 − a2 , ..., bn − an are the edge lengths of R. If they are

equal, then R is a coordinate cube . If ar = br for some r , then V (R) = 0 and we say
that R is degenerate ; otherwise, R is nondegenerate .
If n = 1, 2, or 3, then V (R) is, respectively, the length of an interval, the area of
a rectangle, or the volume of a rectangular parallelepiped. Henceforth, rectangle
or cube will always mean coordinate rectangle or coordinate cube unless it is
stated otherwise. If
R = [a1 , b1 ] × [a2 , b2 ] × · · · × [an , bn ]
and
Pr : ar = ar0 < ar1 < · · · < armr = br
is a partition of [ar , br ], 1 ≤ r ≤ n, then the set of all rectangles in Rn that can be
written as
[a1,j1 −1 , a1j1 ] × [a2,j2 −1 , a2j2 ] × · · · × [an,jn −1 , anjn ],

1 ≤ jr ≤ mr , 1 ≤ r ≤ n,
is a partition of R. We denote this partition by
P = P1 × P2 × · · · × Pn . (3.11)
3.7. Riemann Sum in Rn 142
We dene its norm to be the maximum of the norms of P1 , P2 , ..., Pn , thus,
∥P∥ = max{∥P1 ∥, ∥P2 ∥, . . . , ∥Pn ∥}.
Put another way, ∥P∥ is the largest of the edge lengths of all the subrectangles
in P. Geometrically, a rectangle in R2 is partitioned by drawing horizontal and ver-
tical lines through it; in R , by drawing planes through it parallel to the coordinate
3
axes. Partitioning divides a rectangle R into nitely many subrectangles that we

can number in arbitrary order as R1 , R2 , . . . , Rk . Sometimes it is convenient to
write
P = {R1 , R2 , . . . , Rk }
rather than (3.11).
b2
a2
x
a1 b1
Figure 3.3: Partitioning of a rectangle in R2
• IfP = P1 ×P2 ×· · ·×Pn and P′ = P1′ ×P2′ ×· · ·×Pn′ are partitions of the same
rectangle, then P is a renement of P if Pi is a renement of Pi , 1 ≤ i ≤ n.
′ ′
3.7 Riemann Sum in Rn

Suppose that f is a real-valued function dened on a rectangle R in Rn , P =
{R1 , R2 , . . . , Rk } is a partition of R.
Let Xj is an arbitrary point in Rj , 1 ≤ j ≤ k .
Then
∑
k
σ= f (Xj )V (Rj )
j=1
is a Riemann sum of f over P.

Since Xj can be chosen arbitrarily in Rj , there are innitely many Riemann
sums for a given function f over any partition P of R.
3.8. Riemann Integral in Rn 143
3.8 Riemann Integral in Rn

: Let f be a real-valued function dened on a rectangle R in Rn .
We say that f is Riemann integrable on R if there is a number L with the
following property:
For every ε > 0, there is a δ>0 such that
|σ − L| < ε.
If σ is any Riemann sum of f over a partition P of R such that ∥P∥ < δ .

In this case, we say that L is the Riemann integral of f over R, and write
∫
f (X) dX = L.
R
∫
Remarks: The integral
R f (X)dX is also written as
∫ ∫
f (x, y) d(x, y) (n = 2), f (x, y, z) d(x, y, z) (n = 3),
R R
or ∫
f (x1 , x2 , . . . , xn ) d(x1 , x2 , . . . , xn ) (n arbitrary).
R
Here dX does not stand for the dierential of X.
It merely identies x1 , x2 , ..., xn , the components of X, as the variables of
∫
integration.
∫ To avoid this minor inconsistency, some authors write simply
Rf
rather than
R f (X) dX.
As in the case where n = 1, we will say simply integrable or integral when
we mean Riemann integrable or Riemann integral. If n ≥ 2, we call the integral
of above denition a multiple integral ; for n = 2 and n = 3 we also call them double
and triple integrals , respectively. When we wish to distinguish between multiple
integrals and the integral we studied in Chapter (n = 1), we will call the latter an
ordinary integral.
∫
Example: Find
R f (x, y) d(x, y), where
R = [a, b] × [c, d]
and
f (x, y) = x + y.
Solution: Let P1 and P2 be partitions of [a, b] and [c, d]; thus,
P1 : a = x0 < x1 < · · · < xr = b

and
P2 : c = y0 < y1 < · · · < ys = d.
A typical Riemann sum of f over P = P1 × P2 is given by
∑
r ∑
s
σ= (ξij + ηij )(xi − xi−1 )(yj − yj−1 ), (3.12)
i=1 j=1
where xi−1 ≤ ξij ≤ xi and yj−1 ≤ ηij ≤ yj . (3.13)
The midpoints of [xi−1 , xi ] and [yj−1 , yj ] are
xi + xi−1 yj + yj−1
xi = and yj = , (3.14)
2 2
and (3.13) implies that
xi − xi−1 ∥P1 ∥ ∥P∥

|ξij − xi | ≤ ≤ ≤ (3.15)
2 2 2
yj − yj−1 ∥P2 ∥ ∥P∥
|ηij − y j | ≤ ≤ ≤ . (3.16)
2 2 2
Now we rewrite (3.12) as
∑r ∑s
σ = i=1 j=1 (xi + y j )(xi − xi−1 )(yj − yj−1 )
∑r ∑s [ ]
+ i=1 j=1 (ξij − xi ) + (ηij − y j ) (3.17)
(xi − xi−1 )(yj − yj−1 ).

∫
To nd
R f (x, y) d(x, y) from (3.17), we recall that
∑
r ∑
s
(xi − xi−1 ) = b − a, (yj − yj−1 ) = d − c (3.18)
i=1 j=1
and
∑
r ∑
s
(x2i − x2i−1 ) = b2 − a2 , (yj2 − yj−1
2
) = d2 − c2 . (3.19)
i=1 j=1
Because of (3.15) and (3.16) the absolute value of the second sum in (3.17) does not
exceed
[ r ]
∑
r ∑
s ∑
∥P∥ (xi − xi−1 )(yj − yj−1 ) = ∥P∥ (xi − xi−1 )
j=1 j=1 i=1
 
∑s
 (yj − yj−1 )
j=1
= ∥P∥(b − a)(d − c)
(see (3.18)), so (3.17) implies that

∑r ∑
s

σ − (x + y )(x − x )(y − y ) ≤ ∥P∥(b − a)(d − c). (3.20)
i j i i−1 j j−1
i=1 j=1
It now follows that

∑r ∑s
i=1 j=1 xi (xi − x[i−1 )(yj − yj−1 ) ]
∑ ∑s
= [ ri=1 xi (xi − xi−1 )] j=1 (yj − yj−1 )
∑
= (d − c) i=1r xi (xi − xi−1 ) (from (3.18))
∑ r
= d−c i=1 (xi − xi−1 )
2 2 (from (3.14))
2
= 2 (b − a )
d−c 2 2 (from (3.19)).
Similarly,
∑
r ∑
s
b−a 2
y j (xi − xi−1 )(yj − yj−1 ) = (d − c2 ).
2
i=1 j=1
Therefore, (3.20) can be written as

d − c b − a
σ − (b − a ) −
2 2
(d − c ) ≤ ∥P∥(b − a)(d − c).
2 2
2 2
Since the right side can be made as small as we wish by choosing ∥P∥ suciently
small, ∫
1[ ]
(x + y) d(x, y) = (d − c)(b2 − a2 ) + (b − a)(d2 − c2 ) .
R 2
Theorem: If f is unbounded on the nondegenerate rectangle R in Rn , then f is

not integrable on R.
Proof : We will show that if f is unbounded on R, P = {R1 , R2 , . . . , Rk } is any

partition of R, and M > 0, then there are Riemann sums σ and σ ′ of f over P such
that
|σ − σ ′ | ≥ M. (3.21)
This implies that f cannot satisfy denition of Riemann integral. (Why?)

Let
∑
k
σ= f (Xj )V (Rj )
j=1
be a Riemann sum of f over P. Let
∑
k
σ= f (Xj )V (Rj )
j=1
be a Riemann sum of f over P.

3.9. Upper and Lower Integrals 146
There must be an integer i in {1, 2, . . . , k} such that
M
|f (X) − f (Xi )| ≥ (3.22)
V (Ri )
for some X in Ri , because if this were not so, we would have
M
|f (X) − f (Xj )| < , X ∈ Rj , 1 ≤ j ≤ k.
V (Rj )
If this is so, then
|f (X)| = |f (Xj ) + f (X) − f (Xj )| ≤ |f (Xj )| + |f (X) − f (Xj )|

M
≤ |f (Xj )| + , X ∈ Rj , 1 ≤ j ≤ k.
V (Rj )
However, this implies that
M
|f (X)| ≤ max |f (Xj )| + 1 ≤ j ≤ k, X ∈ R,
V (Rj )
which contradicts the assumption that f is unbounded on R.

Now suppose that X satises (3.22).
Consider the Riemann sum
∑
n
σ′ = f (X′j )V (Rj )
j=1
over the same partition P, where
{
Xj , j ̸= i,
X′j =
X, j = i.
Since
|σ − σ ′ | = |f (X) − f (Xi )|V (Ri ),
(3.22) implies (3.21).
3.9 Upper and Lower Integrals

If f is bounded on a rectangle R in Rn and P = {R1 , R2 , . . . , Rk } is a partition of
R.
Let
Mj = sup f (X), mj = inf f (X).
X∈Rj X∈Rj
The upper sum of f over P is
∑
k
S(P) = Mj V (Rj ).
j=1
The upper integral of f over R, denoted by

∫
f (X) dX,
R
is the inmum of all upper sums.
Upper and Lower Integrals: The lower sum of f over P is

∑
k
s(P) = mj V (Rj ).
j=1
The lower integral of f over R, denoted by

∫
f (X) dX,
R
is the supremum of all lower sums.
Theorem: Let f be bounded on a rectangle R and let P be a partition of R.
Then
1. The upper sum S(P) of f over P is the supremum of the set of all Riemann
sums of f over P.
2. The lower sum s(P) of f over P is the inmum of the set of all Riemann sums
of f over P.
Remarks: If
m ≤ f (X) ≤ M for X in R,
then
mV (R) ≤ s(P) ≤ S(P) ≤ M V (R);
∫ ∫
therefore,
R f (X) dX and R f (X) dX exist, are unique, and satisfy the inequalities
∫
mV (R) ≤ f (X) dX ≤ M V (R)
R
and ∫
mV (R) ≤ f (X) dX ≤ M V (R).
R
Remarks: The upper and lower integrals are also written as
∫ ∫
f (x, y) d(x, y) and f (x, y) d(x, y) (n = 2),
R R
∫ ∫
f (x, y, z) d(x, y, z) and f (x, y, z) d(x, y, z) (n = 3),
R R
or ∫
f (x1 , x2 , . . . , xn ) d(x1 , x2 , . . . , xn )
R
and ∫
f (x1 , x2 , . . . , xn ) d(x1 , x2 , . . . , xn ) (n arbitrary).
R
∫ ∫
Example: Find
R f (x, y) d(x, y) and R f (x, y) d(x, y), with R = [a, b] × [c, d] and
f (x, y) = x + y.
Solution: Let P1 and P2 be partitions of [a, b] and [c, d]; thus,
P1 : a = x0 < x1 < · · · < xr = b
P2 : c = y0 < y1 < · · · < ys = d.

The maximum and minimum values of f on the rectangle [xi−1 , xi ] × [yj−1 , yj ] are
xi + yj and xi−1 + yj−1 , respectively.
Therefore,
r ∑
∑ s
S(P) = (xi + yj )(xi − xi−1 )(yj − yj−1 ) (3.23)
i=1 j=1
∑r ∑ s
s(P) = (xi−1 + yj−1 )(xi − xi−1 )(yj − yj−1 ). (3.24)
i=1 j=1
By substituting
1
xi + yj = [(xi + xi−1 ) + (yj + yj−1 ) + (xi − xi−1 ) + (yj − yj−1 )]
2
into (3.23). We nd that
1
S(P) = (Σ1 + Σ2 + Σ3 + Σ4 ), (3.25)
2
where
∑r ∑s
i=1 (xi − xi−1 ) − yj−1 ) = (b2 − a2 )(d − c),
Σ1 = 2 2
j=1 (yj
∑r ∑s
i=1 (xi − xi−1 ) − yj−1 = (b − a)(d2 − c2 ),
Σ2 = 2 2 )
j=1 (yj
∑r ∑
Σ3 = − xi−1 )2 sj=1 (yj − yj−1 ) ≤ ∥P∥(b − a)(d − c),
i=1 (xi
∑r ∑s
i=1 (xi − xi−1 ) j=1 (yj − yj−1 )
Σ4 = 2 ≤ ∥P∥(b − a)(d − c).
Substituting these four results into (3.25) shows that
I < S(P) < I + ∥P∥(b − a)(d − c),
where
(d − c)(b2 − a2 ) + (b − a)(d2 − c2 )
I= .
2
From this, we see that
∫
(x + y) d(x, y) = I.
R
After substituting
1
xi−1 + yj−1 = [(xi + xi−1 ) + (yj + yj−1 ) − (xi − xi−1 ) − (yj − yj−1 )]
2
into (3.24), a similar argument shows that
I − ∥P∥(b − a)(d − c) < s(P) < I.
So ∫
(x + y) d(x, y) = I.
R
Theorem: Suppose that |f (X)| ≤ M if X is in the rectangle
R = [a1 , b1 ] × [a2 , b2 ] × · · · × [an , bn ].
Let P = P1 × P2 × · · · × Pn and P′ = P1′ × P2′ × · · · × Pn′ be partitions of R, where

Pj′ is obtained by adding rj partition points to Pj , 1 ≤ j ≤ n. Then
 
∑n
rj 
S(P) ≥ S(P′ ) ≥ S(P) − 2M V (R)  ∥P∥ (3.26)
bj − aj
j=1
and  
∑
n
rj 
s(P) ≤ s(P′ ) ≤ s(P) + 2M V (R)  ∥P∥. (3.27)
bj − aj
j=1
Theorem: If f is bounded on a rectangle R, then
∫ ∫
f (X) dX ≤ f (X) dX.
R R
Theorem: If f is integrable on a rectangle R, then
∫ ∫ ∫
f (X) dX = f (X) dX = f (X) dX.
R R R
Theorem: If f is bounded on a rectangle R and ε > 0, there is a δ>0 such that
∫ ∫
f (X) dX ≤ S(P) < f (X) dX + ε
R R
and ∫ ∫
f (X) dX ≥ s(P) > f (X) dX − ε
R R
if ∥P∥ < δ.
Theorem: A bounded function f is integrable on a rectangle R if and only if
∫ ∫
f (X) dX = f (X) dX.
R R
Theorem: If f is bounded on a rectangle R, then f is integrable on R if and only

if for every ε > 0 there is a partition P of R such that
S(P) − s(P) < ε.
Theorem: If f is bounded on a rectangle R and
∫ ∫
f (X) dX = f (X) dX = L,
R R
then f is integrable on R, and
∫
f (X) dX = L.
R
Theorem: If f is continuous on a rectangle R in Rn , then f is integrable on R.

3.10. Sets with Zero Content 151
3.10 Sets with Zero Content

A subset E of Rn has zero content if for each ε>0 there is a nite set of rectangles
T1 , T2 , ..., Tm such that
∪
m
E⊂ Tj (3.28)
j=1
and
∑
m
V (Tj ) < ε. (3.29)
j=1
Example: Since the empty set is contained in every rectangle, the empty set has
zero content.
If E consists of nitely many points X1 , X2 , ..., Xm , then Xj can be enclosed
in a rectangle Tj such that
ε
V (Tj ) < , 1 ≤ j ≤ m.
m
∪m ∑m
Then E⊂ j=1 Tj and j=1 V (Tj ) < ε hold, so E has zero content.
Example: Any bounded set E with only nitely many limit points has zero content.
To see this, we rst observe that if E has no limit points, then it must be nite,
by the BolzanoWeierstrass theorem, and therefore must have zero content.
Now suppose that the limit points of E are X1 , X2 , ..., Xm . Let R1 , R2 , ...,
Rm be rectangles such that Xi ∈ Ri0 and
ε
V (Ri ) < , 1 ≤ i ≤ m. (3.30)
2m
The set of points of E that are not in ∪m
j=1 Rj has no limit points (why?) and, being
bounded, must be nite (again by the BolzanoWeierstrass theorem).
If this set contains p points, then it can be covered by rectangles R1′ , R2′ , ...,
Rp′ with
ε
V (Rj′ ) < , 1 ≤ j ≤ p. (3.31)
2p
Now,  
(m )
∪ ∪ ∪
p
E⊂ Ri  Rj′ 
i=1 j=1
From (3.30) and (3.31),
∑
m ∑
p
V (Ri ) + V (Rj′ ) < ε.
i=1 j=1
3.11. Integral Over Bounded Set 152
Example: If f is continuous on [a, b], then the curve
y = f (x), a≤x≤b (3.32)
(that is, the set {(x, y) : y = f (x), a ≤ x ≤ b}), has zero content in R2 .
Lemma: The union of nitely many sets with zero content has zero content.
Theorem: Suppose that f is bounded on a rectangle
R = [a1 , b1 ] × [a2 , b2 ] × · · · × [an , bn ] (3.33)
and continuous except on a subset E of R with zero content. Then f is integrable

on R.

{
x + y, 0 ≤ x < y ≤ 1,
f (x, y) =
5, 0 ≤ y ≤ x ≤ 1,
is continuous on R = [0, 1] × [0, 1] except on the line segment
y = x, 0≤x≤1
Since the line segment has zero content, f is integrable on R.
3.11 Integral Over Bounded Set

Suppose that f is bounded on a bounded subset of S of Rn . Let
{
f (X), X ∈ S,
fS (X) = (3.34)
0, X ̸∈ S.
Let R be a rectangle containing S. Then the integral of f over S is dened to be
∫ ∫
f (X) dX = fS (X) dX
S R
∫
if
R fS (X) dX exists.
Area
∫ and volume as integrals: If S is a bounded subset of Rn and the integral
S dX (with integrand f ≡ 1) exists.

∫
We call dX the
S content (also, area if n=2 or volume if n = 3) of S, and
denote it by V (S).
Thus, ∫
V (S) = dX.
S
Theorem: Suppose that f is bounded on a bounded set S and continuous except

on a subset E of S with zero content.
Suppose also that ∂S has zero content. Then f is integrable on S.

3.12. Dierentiable Surfaces 153
3.12 Dierentiable Surfaces

A dierentiable surface S in Rn (n > 1) is the image of a compact subset D of Rm ,
where m < n, under a continuously dierentiable transformation G: Rm → Rn . If
m = 1, S is also called a dierentiable curve .
Example: The circle

{(x, y) : x2 + y 2 = 9}
is a dierentiable curve in R2 .
Since it is the image of D = [0, 2π] under the continuously dierentiable trans-
formation G:R→ R2 dened by
[ ]
3 cos θ
X = G(θ) = .
3 sin θ
Example: The sphere

{(x, y, z) : x2 + y 2 + z 2 = 4}
is a dierentiable surface in R3 .
Since it is the image of
D = {(θ, ϕ) : 0 ≤ θ ≤ 2π, −π/2 ≤ ϕ ≤ π/2}
under the continuously dierentiable transformation G : R2 → R3 dened by
 
2 cos θ cos ϕ
X = G(θ, ϕ) =  2 sin θ cos ϕ  .
2 sin ϕ
Theorem: A dierentiable surface in Rn has zero content.
Let S , D, and G be as in Denition ??. From Lemma ??, there is a constant

M such that
|G(X) − G(Y)| ≤ M |X − Y| if X, Y ∈ D. (3.35)
Since D is bounded, D is contained in a cube
C = [a1 , b1 ] × [a2 , b2 ] × · · · × [am , bm ],
where
bi − ai = L, 1 ≤ i ≤ m.
Suppose that we partition C into Nm smaller cubes by partitioning each of the
intervals [ai , bi ] into N equal subintervals. Let R1 , R2 , ...,Rk be the smaller cubes
so produced that contain points of D, and select points X1 , X2 , . . . , Xk such that
Xi ∈ D ∩ Ri , 1 ≤ i ≤ k . If Y ∈ D ∩ Ri , then (3.35) implies that
|G(Xi ) − G(Y)| ≤ M |Xi − Y|. (3.36)

Since Xi and Y are both in the cube Ri with edge length L/N ,
√
L m
|Xi − Y| ≤ .
N
√
ML m
|G(Xi ) − G(Y)| ≤ ,
N
which in turn implies that G(Y) lies in a cube ei
R in Rn centered at G(Xi ), with
√
sides of length 2M L m/N . Now
∑
k ( √ )n ( √ )n
2M L m m 2M L m √
ei ) = k
V (R ≤N = (2M L m)n N m−n .
N N
i=1
Since n > m, we can make the sum on the left arbitrarily small by taking N
suciently large. Therefore, S has zero content.
Theorem: Suppose that S is a bounded set in Rn , with boundary consisting of a

nite number of dierentiable surfaces.
Let f be bounded on S and continuous except on a set of zero content. Then f
is integrable on S.
Example: Let
S = {(x, y) : x2 + y 2 = 1, x ≥ 0}.
The set S is bounded by a semicircle and a line segment, both dierentiable curves
in R .
2
Let {
(1 − x2 − y 2 )1/2 , (x, y) ∈ S, y ≥ 0,
f (x, y) =
−(1 − x2 − y 2 )1/2 , (x, y) ∈ S, y < 0.
Then f is continuous on S except on the line segment
y = 0, 0 ≤ x < 1,
which has zero content.

Hence, from the theorem we just stated implies that f is integrable on S.
Theorem: If f and g are integrable on S, then so is f + g, and

∫ ∫ ∫
(f + g)(X) dX = f (X) dX + g(X) dX.
S S S
Theorem: If f is integrable on S and c is a constant, then cf is integrable on S,

and ∫ ∫
(cf )(X) dX = c f (X) dX.
S S
Theorem: If f and g are integrable on S and f (X) ≤ g(X) for X in S, then

∫ ∫
f (X) dX ≤ g(X) dX.
S S
Theorem: If f is integrable on S, then so is |f |, and

∫ ∫

f (X) dX ≤ |f (X)| dX.

S S
Theorem: If f and g are integrable on S, then so is the product f g.
Theorem: Suppose that u is continuous and v is integrable and nonnegative on a

rectangle R.
Then ∫ ∫
u(X)v(X) dX = u(X0 ) v(X) dX
R R
for some X0 in R.
Theorem: Suppose that S is contained in a bounded set T and f is integrable on

S.
Then fS is integrable on T, and
∫ ∫
fS (X) dX = f (X) dX.
T S
Theorem: If f is integrable on disjoint sets S1 and S2 , then f is integrable on

S1 ∪ S2 , and ∫ ∫ ∫
f (X) dX = f (X) dX + f (X) dX. (3.37)
S1 ∪S2 S1 S2
Theorem: Suppose that f is integrable on sets S1 and S2 such that S1 ∩ S2 has

zero content. Then f is integrable on S1 ∪ S2 , and
∫ ∫ ∫
f (X) dX = f (X) dX + f (X) dX.
S1 ∪S2 S1 S2
Example: Let
S1 = {(x, y) : 0 ≤ x ≤ 1, 0 ≤ y ≤ 1 + x}
S2 = {(x, y) : −1 ≤ x ≤ 0, 0 ≤ y ≤ 1 − x}
3.13. Iterated Integrals 156
y=1−x y=1+x
x
−1 1
Figure 3.4: S1 and S2
Then
S1 ∩ S2 = {(0, y) : 0 ≤ y ≤ 1}
has zero content.
Hence, by using corollary implies that if f is integrable on S1 and S2 , then f is
also integrable over
S = S1 ∪ S2 = {(x, y) : −1 ≤ x ≤ 1, 0 ≤ y ≤ 1 + |x|}
and ∫ ∫ ∫
f (X) dX = f (X) dX + f (X) dX.
S1 ∪S2 S1 S2
3.13 Iterated Integrals

Let us rst assume that f is continuous on R = [a, b] × [c, d].
Then, for each y in [c, d], f (x, y) is continuous with respect to x on [a, b], so the
integral
∫ b
F (y) = f (x, y) dx
a
exists.
Moreover, the uniform continuity of f on R implies that F is continuous and
therefore integrable on [c, d].
We say that
∫ d ∫ d (∫ b )
I1 = F (y) dy = f (x, y) dx dy
c c a
is an iterated integral of f over R.
Iterated integrals: We will usually write it as
∫ d ∫ b
I1 = dy f (x, y) dx.
c a
Another iterated integral can be dened by writing
∫ d
G(x) = f (x, y) dy, a ≤ x ≤ b,
c
Dening
∫ b ∫ b (∫ d )
I2 = G(x) dx = f (x, y) dy dx,
a a c
which we usually write as
∫ b ∫ d
I2 = dx f (x, y) dy.
a c
Example: Let
f (x, y) = x + y
and R = [0, 1] × [1, 2]. Then
∫ 1 ∫ 1
1
F (y) = (x + y) dx = + y
f (x, y) dx =
0 0 2
∫ 2 ∫ 2( ) ( )
1 y y 2 2
I1 = F (y) dy = + y dy = + = 2.
1 1 2 2 2 1
Also,
∫ ( )
2
y 2 2 3
G(x) = (x + y) dy =
xy + =x+ ,
1 2 y=1 2
∫ 1 ∫ 1( ) ( 2 )
3 x 3x 1
I2 = G(x) dx = x+ dx = + = 2.
0 0 2 2 2 0
Theorem: Suppose that f is integrable on R = [a, b] × [c, d] and
∫ b
F (y) = f (x, y) dx
a
exists for each y in [c, d].

Then F is integrable on [c, d], and
∫ d ∫
F (y) dy = f (x, y) d(x, y); (3.38)
c R
that is,
∫ d ∫ b ∫
dy f (x, y) dx = f (x, y) d(x, y). (3.39)
c a R
Theorem: If f is integrable on [a, b] × [c, d], then
∫ b ∫ d ∫ d ∫ b
dx f (x, y) dy = dy f (x, y) dx,
a c c a
∫d ∫b
provided that f (x, y) dy exists for a ≤ x ≤ b and a f (x, y) dx exists for c ≤ y ≤ d.
c
In particular, these hypotheses hold if f is continuous on [a, b] × [c, d].

f (x, y) = x + y
is continuous everywhere.
For example, let R = [0, 1] × [1, 2].
Then we have
∫ ∫ 2∫ 1
(x + y) d(x, y) = dy (x + y) dx
∫ 2 [( 2 ) 1 ]
R 1 0
x
= + xy dy
1 2 x=0
∫ 2( ) ( )
1 y y 2 2
= + y dy = + = 2.
1 2 2 2 1
Since f also satises the hypotheses of Fubini's Theorem with x and y interchanged,
we can calculate the double integral from the iterated integral in which the integra-
tions are performed in the opposite order.
Thus,
∫ ∫ 1∫ 2
(x + y) d(x, y) = dx (x + y) dy
∫ 1 [( ) ]
R 0 1
y 2 2
= xy + dx
0 2 y=1
∫ 1( ) ( 2 )
3 x 3x 1
= x+ dx = + = 2.
0 2 2 2 0
∫d ∫b ∫
Remark: If
c dy a f (x, y) dx exists then so does
R f (x, y) d(x, y). However, this
need not to be true.
Example: If f is dened on R = [0, 1] × [0, 1] by

{
2xy if y is rational,
f (x, y) =
y if y is irrational,
then ∫ 1
f (x, y) dx = y, 0 ≤ y ≤ 1,
0
and ∫ 1 ∫ 1 ∫ 1
1
dy f (x, y) dx = y dy = .
0 0 0 2
However, f is not integrable on R.
Theorem: Let I1 , I2 , . . . , In be closed intervals and suppose that f is integrable

on R = I1 × I2 × · · · × In .
Suppose that there is an integer p in {1, 2, . . . , n − 1} such that
∫
Fp (xp+1 , xp+2 , . . . , xn ) = f (x1 , x2 , . . . , xn ) d(x1 , x2 , . . . , xp )
I1 ×I2 ×···×Ip
exists for each (xp+1 , xp+2 , . . . , xn ) in Ip+1 × Ip+2 × · · · × In .
Then
∫
Fp (xp+1 , xp+2 , . . . , xn ) d(xp+1 , xp+2 , . . . , xn )
Ip+1 ×Ip+2 ×···×In
∫
exists and equals
R f (X) dX.
Theorem: Ij = [aj , bj ], 1 ≤ j ≤ n,
Let and suppose that f is integrable on
R = I1 × I2 × · · · × In .
Suppose also that the integrals
∫
Fp (xp+1 , . . . , xn ) = f (X) d(x1 , x2 , . . . , xp ), 1 ≤ p ≤ n − 1,
I1 ×I2 ···×Ip
exist for all

(xp+1 , . . . , xn ) in Ip+1 × · · · × In .
Then the iterated integral
∫ bn ∫ bn−1 ∫ b2 ∫ b1
dxn dxn−1 · · · dx2 f (X) dx1
an an−1 a2 a1
∫
exists and equals
R f (X) dX.

Handouts RealAnalysis II

Uploaded by

Copyright:

Available Formats

Handouts RealAnalysis II

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Handouts RealAnalysis II

Uploaded by

Copyright:

Available Formats

COMSATS University Islamabad, Islamabad

Real Analysis II (MTH631)

Salman Amin Malik

Department of Mathematics, CUI Islamabad

About the instructor

About the handouts

Number of Credit Hours: 3 credits

• Understand the convergence of sequence of functions (LO1).

• Understand the pointwise convergence, uniform convergence, several tests for

• Apply the interchange of limit and integration, derivative of sequence of func-

• Understand the innite series of functions, convergence, Weierstrass's test and

• Understand and nd the Fourier series, Fourier coecients, convergence of

• Understand the functions of several variables, Heine-Borel Theorem, limits

• Dierentiability in Rn , Dierentials, Directional derivatives, Partial deriva-

• Improper integrals, Multiple integrals, Functions of bounded variation (LO15)

Prerequisites: Real Analysis I (MTH621)

The textbooks for this course:

[2] W. F. Trench, Introduction to Real Analysis, Pearson Education, 2013.

[3] S. Ponnusamy, Foundations of Mathematical Analysis, Birkhauser, 2012.

[5] R. G. Bartle and D. R. Sherbert, Introduction to Real Analysis, Third Edition,

• Sequences and Series of functions

• Functions of several variables

• Vector valued functions

2 Functions of Several Variables 71

2.7.1 Polygonal Path . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3 Integral Calculus 129

1.3 Uniform convergence graphically . . . . . . . . . . . . . . . . . . . . 5

2.1 The set S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3.1 Rectangle in R2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

Sequences and Series of Functions

1.1 Informal way

Example: The functions

dene a sequence on D = [0, ∞).

Example: The functions

dene a sequence on D = [0, ∞).

Example: The functions

dene a sequence on D = (−∞, 1].

1.2 Pointwise Convergence

F (x) = lim Fn (x), x ∈ S.

Example: The sequence of functions dened by

Therefore, {Fn } converges pointwise on S = [0, 1] to the limit function F dened by

Example: Consider the functions

Equating the derivative

Example: For n ≥ 1, let be dened on (−∞, ∞) by

Since Fn (0) = 0 for all n, limn→∞ Fn (0) = 0. If x ̸= 0, then Fn (x) = 0 if

lim Fn (x) = 0, −∞ < x < ∞,

so the limit function is identically zero on (−∞, ∞).

Example: Show that the sequence of functions

dene a sequence on D = [0, ∞), converges to 0.

1.3 Norm Dened Over a Set

∥g∥S = sup |g(x)|.

Lemma: If g and h are dened on S, then

∥g + h∥S ≤ ∥g∥S + ∥h∥S

Moreover, if either g or h is bounded on S, then

∥g − h∥S ≥ |∥g∥S − ∥h∥S ∥| .

1.4 Uniform Convergence

Figure 1.3: Uniform convergence graphically

A sequence {Fn } of functions dened on a set S converges uniformly to the limit

Thus, {Fn } converges uniformly to F on S if for each ε>0 there is an integer N

F (x) − ε < y < F (x) + ε, a ≤ x ≤ b, n≥N

• Understand the innite series of functions, convergence, Weierstrass's test and

• Understand and nd the Fourier series, Fourier coecients, convergence of

• Dierentiability in Rn , Dierentials, Directional derivatives, Partial deriva-

dene a sequence on D = [0, ∞).

dene a sequence on D = [0, ∞).

dene a sequence on D = (−∞, 1].

Example: The sequence of functions dened by

Therefore, {Fn } converges pointwise on S = [0, 1] to the limit function F dened by

Example: For n ≥ 1, let be dened on (−∞, ∞) by

dene a sequence on D = [0, ∞), converges to 0.

1.3 Norm Dened Over a Set

Lemma: If g and h are dened on S, then

A sequence {Fn } of functions dened on a set S converges uniformly to the limit

Counter example: For n ≥ 1, let Fn be dened on (−∞, ∞) by

for any xed x in S.

Example: Suppose that g is dierentiable on S = (−∞, ∞) and

Let F0 be bounded on S and dene

Consider the sequence of functions dened on S = [0, 1]

Example (Unbounded sequence of functions): For n ≥ 1, let Fn be dened

Example: The sequence {Fn } dened by

Then {Fn } converges uniformly on [a, b] to a dierentiable limit function F, and

dene the innite series of functions