2nd Semester Real Analysis MMDSE 2.2

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 260

KARNATAKA STATE OPEN UNIVERSITY

Mukthagangothri, Mysuru – 570006

M.Sc. MATHEMATICS (CBCS)


(SECOND SEMESTER)

Course- MMDSC 2.2

Real Analysis II
M.Sc. MATHEMATICS
SECOND SEMESTER

Course: MMDSC 2.2


REAL ANALYSIS - II

i
Programme Name: M.Sc. Mathematics Year/Semester: II Semester
Course Code: MMDSC 2.2 Course Name: Real Analysis -II
Credit: 4 Unit Number : 1-16
COURSE DESIGN COMMITTEE
Dr. Sharanappa V. Halse Chairman
Vice Chancellor
Karnataka State Open University
Mukthagangothri, Mysuru-570006
Prof. Ashok Kamble Member
Dean (Academic)
Karnataka State Open University
Mukthagangothri, Mysuru-570006
Dr. Pavithra. M Course coordinator
Assistant Professor
DoS in Mathematics, KSOU, Mukthagangothri, Mysuru-06
EDITORIAL COMMITTEE
1. Dr. K. Shivashankara Chairman
BOS Chairman(PG), DoS in Mathematics, KSOU.
Professor, Yuvaraja College,
University of Mysore, Mysuru-06

2. Mr. S. V. Niranjan Member & Convener


Coordinator, (DoS in Mathematics)
Assistant Professor, DoS in Physics
KSOU, Mysuru-06

3. Dr.Pavithra. M Member
Assistant Professor
DoS in Mathematics, KSOU, Mysuru-06

4. Dr.Chandru Hegde Member


Assistant Professor, DoS in Mathematics,
Mangalagangotri, Mangaluru.

ii
COURSE WRITER

Dr. Nandeesh K. C Block 2.2A to Block 2.2 B


Assistant Professor (Unit 1 to Unit 08)
Department of Studies in Mathematics
KSOU, Mysuru-06

Dr. B. N. Dharmendra
Associate Professor Block 2.2C to Block 2.2 D
Department of Studies in Mathematics (Unit 9 to Unit 16)
Maharani Women’s Science College, Mysuru-06

COURSE EDITOR
Dr. K. Shivashankara
Professor
Department of Studies in Mathematics,
Yuvaraja College,
University of Mysore, Mysuru -06
COPYRIGHT
The Registrar
Karnataka State Open University
Mukthagangothri, Mysuru-570006

Developed by the Department of Studies in Mathematics under the guidance of Dean


(Academic), KSOU, Mysuru.
Karnataka State Open University, 2022.
All rights reserved. No part of this work may be reproduced in any form or any other means,
without permission in writing from the Karnataka State Open University.
Further information on the Karnataka State Open University Programmes may be obtained
from the University’s Office at Mukthagangothri, Mysuru – 570 006.

iii
PRELUDE

Real Analysis is the branch of mathematics that deals with inequalities and limits. The study

of real analysis is indispensable for a prospective post-graduate student of pure or applied

mathematics. It also has great value for any student who wishes to go beyond the repetitive

manipulations of formulas because it develops the ability to think deductively, analyze

mathematical situations and extend ideas to new contexts. Mathematics has become valuable

in many areas, including economics and management science as well as the physical sciences,

engineering, and computer science. This book was written to provide an accessible,

reasonably paced treatment of the basic concepts and techniques of real analysis for students

in these areas. While students will find this book challenging and demonstrative that makes

students capable in mastering the course.

iv
Contents
 BLOCK-I: THE RIEMANN-STIELTJES INTEGRAL

Unit 1: Definition and existence of integral 01-19


1.1 Main Objectives
1.2 Introduction
1.3 Definition, Riemann –Stieltjes Integrals
1.4 Existence of Riemann –Stieltjes Integrals
1.5 Examples
1.6 Summary
1.7 Key Words
1.8 Exercises
1.9 References
Unit 2: The Properties of Integral 20-35
2.1 Main Objectives
2.2 Introduction
2.3 The classes of Riemann-Steiltje’s Integral
2.4 The Properties of Riemann-Steiltje’s integral
2.5 Summary
2.6 Key Words
2.7 Exercises
2.8 References
Unit 3: Integration and Differentiation 36-47
3.1 Main Objectives
3.2 Introduction
3.3 Mean Value Theorems for Riemann- Stieltjes Integrals
3.4 Integration and differentiation
3.5 Summary
3.6 Key Words
3.7 Exercises
3.8 References
Unit 4: Integration of vector valued functions and Rectifiable curves 48-61
4.1 Main Objectives
4.2 Introduction
4.3 Integration of vector valued functions
4.4 Examples
4.5 Rectifiable curves
4.6 Summary
4.7 Key Words
4.8 Exercises
4.9 References
 BLOCK-II SEQUENCES AND SERIES OF FUNCTIONS

Unit 5: Point wise and Uniform Convergence 62-73


5.1 Objectives
5.2 Introduction
5.3 Pointwise Convergence
1.4 Uniform Convergence
5.5 Summary
5.6 Key Words
5.7 Exercises
5.8 References
Unit 6: Uniform Convergence and Integration 74-85
6.1 Main Objectives
6.2 Introduction
6.3 Uniform Convergence and Continuity
6.4Uniform Convergence and Differentiation
6.5Uniform Convergence and Integration
6.6Summary
6.7Key Words
6.8Exercises
6.9 References
Unit 7: Equicontinuous families of functions and Weierstrass theorem 86-94
7.1 Main Objectives
7.2 Introduction
7.3 Equicontinuous families of functions
7.4 Stone-Weierstrass Approximation Theorem
7.5 Summary
7.6 Key Words
7.7 Exercises
7.8 References
Unit 8: Power Series, the Exponential, Logarithmic and
Trigonometric Functions 95-104
8.1 Main Objectives
8.2 Introduction
8.3 Power Series
8.4 The Exponential Function
8.5 The Logarithmic function
8.6 Trigonometric functions
8.7 Summary
8.8 Key Words
8.9 Exercises
8.10 References

 BLOCK-III: Improper Integrals

Unit 9: Improper Integrals: Definition, Criteria for convergence,


Interchanging derivatives and integrals. 105-122
9.1 Main Objectives
9.2 Introduction
9.3 Definition of an Improper integral
9.4 Keywords
9.5 Terminal Problems
9.6 Books for reference
Unit 10: Test for Converfence of an improper Integral 123-145
10.1 Main Objectives
10.2 Test for Convergence of an Improper integral
10.3 Keywords
10.4Terminal Problems
10.5 Books for reference
Unit 11: Gamma Function and Beta Function 146-171
11.1 Main Objectives
11.2 Gamma Function
11.3 Beta Function
11.4 Keywords
11.5 Terminal Problems
11.6 Books for reference
Unit 12: Interchange of Differentiation and Integration 172-183
12.1 Main Objectives
12.2 Interchange of Differentiation and Integration
12.3 Keywords
12.4Books for reference

 BLOCK-IV: Functions of several variables

Unit 13: Functions of several variables 184-200


13.1 Main Objectives
13.2 Introduction
13.3 Limits
13.4 Continuity
13.5 Partial Derivatives
13.6 Keywords
13.7 Terminal Problems
13.8 Books for reference
Unit 14: Differentiation 201-225
14.1 Main Objectives
14.2 Introduction
14.3 Differentiability
14.4 Keywords
14.5 Terminal Problems
14.6 Books for reference
Unit 15: Taylor’s theorem for a function of n variable and
The contraction principle 226-240
15.1 Main Objectives
15.2 Introduction
15.3 Taylor’s theorem for a function of n variable
15.4 The contraction principle
15.5 Keywords
15.6 Books for reference
Unit 16: Inverse function theorem and the implicit Function theorem 241-251
16.1 Main Objectives
16.2The Inverse function theorem
16.3The implicit Function theorem
16.4 Keywords
16.5 Terminal Problems
16.6 Books for reference
Block – I
The Riemann-Stieltjes Integral

1
UNIT 1
DEFINITION AND EXISTENCE OF INTEGRALS

1.1 Main Objectives

1.2 Introduction

1.3 Definition, Riemann –Stieltjes Integrals

1.4 Existence of Riemann –Stieltjes Integrals

1.5 Examples

1.6 Summary

1.7 Key Words

1.8 Exercises

1.9 References

2
UNIT 1
DEFINITION AND EXISTENCE OF INTEGRALS

1.1 Main Objectives


The objective of the unit is to study and understand the following topics in detail.

 Riemann- Stieltjes integral.


 The existence of Riemann- Stieltjes integral.
 Evaluation of Riemann- Stieltjes integral.

1.2 Introduction

The Riemann-Stieltje integral is an important generalisation of the Riemann integral. The


Riemann integral is a particular case of a more general integral. The Riemann-Stieltjes sums are
given by ∑𝑛𝑘=1 𝑓(𝑥𝑘 )Δ𝛼𝑘 , where 𝑓 and 𝛼 are arbitrary functions on [𝑎, 𝑏] and under certain
conditions they approach a limiting value as the partition 𝑃 = {𝑥0 , 𝑥1 , . . , 𝑥𝑛 } becomes
sufficiently fine. The limit, when it exists, is the Riemann-Stieltjes integral of 𝑓 with respect to 𝛼
𝑏
on [𝑎, 𝑏] and is denoted by ∫𝑎 𝑓 𝑑𝛼 .
This integral is the Riemann integral in the case 𝛼(𝑥) = 𝑥. We shall see that, when 𝑓 is
continuous and 𝛼 is continuously differentiable, we have
𝑏 𝑏
∫ 𝑓(𝑥) 𝑑𝛼(𝑥) = ∫ 𝑓(𝑥)𝛼 ′ (𝑥).
𝑎 𝑎

If 𝛼 is not continuously differentiable, the Riemann-Stieltjes integral on the left can still
exist and be computed. The Riemann-Stieltjes integral is important in physics and probability
where moments of non-smooth distributions are to be computed.

1.3 Definition of Riemann – Stieltjes Integral

Definition-1.3.1: Let [𝑎, 𝑏] be a given interval. By a partition 𝑃 of [𝑎, 𝑏] we mean a finite set

of points 𝑥0 , 𝑥1 , … , 𝑥𝑛 , where 𝑎 = 𝑥0 ≤ 𝑥1 ≤ ⋯ ≤ 𝑥𝑛−1 ≤ 𝑥𝑛 = 𝑏.

3
We write,

Δxi = xi − xi−1 , 1 ≤ i ≤ n.

Now, let f be a bounded real function defined on [𝑎, 𝑏]. Corresponding to each partition P of

[𝑎, 𝑏]

We put

𝑀𝑖 = sup 𝑓(𝑥), 𝑚𝑖 = inf 𝑓(𝑥)

and U(P, f ) = ∑ni=1 Mi Δαi , L(P, f ) = ∑ni=1 mi Δαi .

Then we put

−𝑏

∫ 𝑓 𝑑𝑥 = inf {𝑈(𝑃, 𝑓) | 𝑝 𝜖 Ꝕ } (1)


𝑎

and

∫ 𝑓 𝑑𝑥 = sup {𝐿(𝑃, 𝑓) | 𝑝 𝜖 Ꝕ } (2)


−𝑎

Where Ꝕ denotes the set of all partitions of [𝑎, 𝑏].

The real numbers 𝑈(𝑃, 𝑓) and 𝐿(𝑃, 𝑓) are respectively called the Upper and lower

Rieman sums off with respect to the partition P of [𝑎, 𝑏].

−𝑏 𝑏
∫𝑎 𝑓 𝑑𝑥 is called the Upper Rieman integral of f over [𝑎, 𝑏] and ∫−𝑎 𝑓 𝑑𝑥 is called

the lower Rieman integral of f over [𝑎, 𝑏].

If the upper and lower integrals are equal, we say that f is Rieman integrable on [𝑎, 𝑏] and

we write 𝑓 ∈ 𝑅, where R denotes the set of all Rieman integrable functions.

𝑏 𝑏
We denote the common value of (1) and (2) by ∫𝑎 𝑓 𝑑𝑥 or by ∫𝑎 𝑓(𝑥) 𝑑𝑥.

This is called the Rieman integral of f over [𝑎, 𝑏].

4
Remarks: Since f is bounded, there exist two numbers m and M, such that

𝑚 ≤ 𝑓(𝑥) ≤ 𝑀, (𝑎 ≤ 𝑥 ≤ 𝑏)

Hence for any partition P, 𝑚(𝑏 − 𝑎) ≤ 𝐿(𝑃, 𝑓) ≤ 𝑈(𝑃, 𝑓) ≤ M(b − a).

Therefore, the sets on the right hand sides of (1) and (2) are bounded subsets of ℝ. So, the upper

and lower integrals are well defined for every bounded function 𝑓. The question of their equality

and hence the question of the integrable of 𝑓 is a more delicate one.

We shall study this in a more general situation.

Definition-1.3.2: Let α be monotonically increasing function on [a, b]. Then α (a) and α (b)

are finite.

Moreover 𝛼(𝑎) ≤ 𝛼(𝑥) ≤ 𝛼(𝑏), ∀ 𝑥 ∈ [𝑎, 𝑏].

Therfore, α is bounded on [𝑎, 𝑏]. Corresponding to each partition P of [𝑎, 𝑏], we write

Δαi = α(xi ) − α(xi−1 )

Since α is increasing, ∆𝛼𝑖 ≥ 0 for all i. For any bounded real function f on [𝑎, 𝑏],

we put

U(P, f, α ) = ∑ni=1 Mi Δαi ,

L(P, f, α ) = ∑ni=1 mi Δαi ,

Where,

𝑀𝑖 = sup 𝑓(𝑥) , 𝑚𝑖 = inf 𝑓(𝑥) , 1 ≤ 𝑖 ≤ 𝑛.


[𝑥𝑖−1 ≤ 𝑥 ≤𝑥𝑖 ] [𝑥𝑖−1 ≤𝑥 ≤ 𝑥𝑖 ]

Let P be the set of all partition P of [𝑎, 𝑏]. Then we define

5
−𝑏

∫ 𝑓 𝑑𝑥 = inf { 𝑈 (𝑃, 𝑓, α) | 𝑝 𝜖 Ꝕ } (1)


𝑎

and

∫ 𝑓 𝑑𝑥 = sup { 𝐿 (𝑃, 𝑓, α) | 𝑝 𝜖 Ꝕ } (2)


−𝑎

If the left members of (1) and (2) are equal, we denote the common value by

𝑏
∫𝑎 𝑓 𝑑𝛼 or

𝑏
by ∫𝑎 𝑓(𝑥) 𝑑𝛼(𝑥). (3)

This is called the Riemann – Stieltjes integral of f with respect to α over [a, b] .

If (1) and (2) are equal ( i.e, if (3) exists), we say that f is integrable ( or Riemann –

Stieltjes integrable) with respect to α in the Riemann sense.

We write 𝑓 ∈ 𝑅(𝛼), Where 𝑅(𝛼) stands for the class of all Riemann – Steiltje’s integrable

functions with respect to 𝛼.

Note: by taking α(x) = x, we see that the Riemann integral is a special case of the Riemann-

Stieltjes integral.

Definition 1.3.3: We say that the partitions P ∗ 𝑖𝑠 𝑎 refinement of P if 𝑃 ∗ ⊃ 𝑃.

i.e, if every point of P is a point of P ∗ , then the partition P ∗ is said to be a refinement of P.

Given two partitions P1 and P2 , we say that P ∗ is their common refinement if P * = P1 ∪ P2 .

Theorem 1.3.1: If P * is a refinement P , then

L(P , f, α) ≤ L(P * , f, α) (1)

and

𝑈(P ∗ , 𝑓, 𝛼) ≤ 𝑈(𝑃, 𝑓, 𝛼) (2)


6
Proof: (1) To prove the theorem it suffices to consider first the case when 𝑃∗ contains one point

more than P.

Let this extra point be 𝑥 ∗ and suppose 𝑥𝑖−1 ≤ 𝑥 ∗ ≤ 𝑥𝑖 . Where 𝑥𝑖−1 and 𝑥𝑖 . Are two

consecutive points of P

Put 𝑤1 = inf 𝑓(𝑥), 𝑥 ∈ [𝑥𝑖−1 , 𝑥 ∗ ] and 𝑤2 = inf 𝑓(𝑥), 𝑥 ∈ [ 𝑥 ∗ , 𝑥𝑖 ],

then , clearly 𝑤1 ≥ 𝑚𝑖 , and 𝑤2 ≥ 𝑚𝑖 , where

𝑚𝑖 = inf 𝑓(𝑥), infimum is taken over all 𝑥 for which 𝑥𝑖−1 ≤ 𝑥 ≤ 𝑥𝑖 .

Hence

𝐿(𝑃∗ , 𝑓, 𝛼) – 𝐿(𝑃, 𝑓, 𝛼)
𝑛
= ∑𝑗=1 𝑚𝑗 Δ𝛼𝑗 + 𝑤1 [𝛼(𝑥 ∗ ) − 𝛼(𝑥𝑖−1 )] + 𝑤2 [𝛼(𝑥𝑖 ) − 𝛼(𝑥 ∗ )] − ∑𝑛𝑗=1 𝑚𝑗 Δαj
𝑗≠𝑖

= 𝑤1 [𝛼(𝑥 ∗ ) − 𝛼(𝑥𝑖−1 )] + 𝑤2 [𝛼(𝑥𝑖 ) − 𝛼(𝑥 ∗ )] − 𝑚𝑖 [𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 )]

= (𝑤1 − 𝑚𝑖 )[𝛼(𝑥 ∗ ) − 𝛼(𝑥𝑖−1 )] + (𝑤2 − 𝑚𝑖 )[𝛼(𝑥𝑖 ) − 𝛼(𝑥 ∗ )]

≥ 0

Hence L(P, f, α) ≤ L(P * , f, α).

2. Put 𝑊1 = sup 𝑓(𝑥), 𝑥 ∈ [𝑥𝑖−1 , 𝑥 ∗ ] and 𝑊2 = sup 𝑓(𝑥), 𝑥 ∈ [ 𝑥 ∗ , 𝑥𝑖 ],

then clearly, 𝑊1 ≤ 𝑀𝑖 , and 𝑊2 ≤ 𝑀𝑖 , where

𝑀𝑖 = sup 𝑓(𝑥), supremum is taken over all 𝑥 for which 𝑥𝑖−1 ≤ 𝑥 ≤ 𝑥𝑖 .

Hence

𝑈(𝑃, 𝑓, 𝛼) – 𝑈(𝑃∗ , 𝑓, 𝛼)
𝑛
= ∑𝑗=1 𝑀𝑗 Δ𝛼𝑗 + 𝑤1 [𝛼(𝑥 ∗ ) − 𝛼(𝑥𝑖−1 )] + 𝑤2 [𝛼(𝑥𝑖 ) − 𝛼(𝑥 ∗ )] − ∑𝑛𝑗=1 𝑀𝑗 Δαj
𝑗≠𝑖

= 𝑊1 [𝛼(𝑥 ∗ ) − 𝛼(𝑥𝑖−1 )] + 𝑊2 [𝛼(𝑥𝑖 ) − 𝛼(𝑥 ∗ )] − 𝑀𝑖 [𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 )]

= (𝑊1 − 𝑀𝑖 )[𝛼(𝑥 ∗ ) − 𝛼(𝑥𝑖−1 )] + (𝑊2 − 𝑀𝑖 )[𝛼(𝑥𝑖 ) − 𝛼(𝑥 ∗ )]

7
≤ 0

Hence, U(P * , f, α) ≤ U(P, f, α).

If 𝑃∗ contains k points more than 𝑃, we need only to repeat the above step k times we arrive at

(1) or (2).

This completes the proof of the theorem.

Notes: 1.As partitions get finer the lower sums increase and the upper sums decrease.

2. Same is true for Riemann sums. That is L(P * , f) ≤ L(P, f, ) and 𝑈(P ∗ , 𝑓) ≤ 𝑈(𝑃, 𝑓)

Theorem 1.3.2: Let 𝑓 be a real bounded function defined on [𝑎, 𝑏] and let 𝛼 be monotonically

𝑏 −𝑏
increasing on[𝑎, 𝑏], then ∫−𝑎 𝑓 𝑑𝛼 ≤ ∫𝑎 𝑓 𝑑𝛼 .

Proof: Let 𝑃 ∗ be the common refinement of two partitions 𝑃1 and 𝑃2 of [𝑎, 𝑏].

Then, 𝑃∗ = 𝑃1 ∪ 𝑃2 . Since 𝑚𝑖 ≤ 𝑀𝑖 for each 𝑖, we have

𝐿(𝑃∗ , 𝑓, 𝛼) = ∑𝑛𝑖=1 𝑚𝑖 Δ𝛼𝑖 ≤ ∑𝑛𝑖=1 𝑀𝑖 Δαi = 𝑈(𝑃∗ , 𝑓, 𝛼).

Since 𝑃1 ⊂ 𝑃∗ and 𝑃2 ⊂ 𝑃 ∗ , by the previous theorem, it follows that

𝐿(𝑃1 , 𝑓, 𝛼) ≤ 𝐿(𝑃∗ , 𝑓, 𝛼) ≤ 𝑈(𝑃∗ , 𝑓, 𝛼) ≤ 𝑈(𝑃2 , 𝑓, 𝛼).

Hence, 𝐿(𝑃1 , 𝑓, 𝛼) ≤ 𝑈(𝑃2 , 𝑓, 𝛼).

Keeping 𝑃2 fixed , and taking the supremum on left hand side as 𝑃1 rums over all partition of

[𝑎, 𝑏], we obtain, from the definition,

𝑏
∫−𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃2 , 𝑓, 𝛼).

Now, taking the infimum on right hand side as 𝑃2 rums over all partition of of [𝑎, 𝑏], we

obtain

𝑏 −𝑏
∫−𝑎 𝑓 𝑑𝛼 ≤ ∫𝑎 𝑓 𝑑𝛼.

This completes the proof.

8
1.4 Existence of Riemann-Stieltjes integrals

Theorem 1.4.1: A function 𝑓 is Riemann- Steiltje’s integrable on [𝑎, 𝑏] if and only if for every

𝜖 > 0 there exists a partition 𝑃 such that

𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 .

Proof: Let 𝜖 > 0 be given and let 𝑃 be a partition of [𝑎, 𝑏] such that

𝑏
∫−𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃2 , 𝑓, 𝛼) holds.

𝑏 −𝑏
By theorem 1.3.2, we have 𝐿(𝑃, 𝑓, 𝛼) ≤ ∫−𝑎 𝑓 𝑑𝛼 ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼) for any partition

𝑃 of[𝑎, 𝑏].

𝑏 −𝑏
Hence 0 ≤ ∫−𝑎 𝑓 𝑑𝛼 − ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 follows from the given

𝑏 −𝑏
condition. Since 𝜖 > 0 is arbitrary, we have ∫−𝑎 𝑓 𝑑𝛼 = ∫𝑎 𝑓 𝑑𝛼 .

Hence 𝑓 ∈ ℜ(𝛼).

Conversely, suppose 𝑓 ∈ ℜ(𝛼) and let 𝜖 > 0 be given.



Then ∫− 𝑓 𝑑𝛼 = ∫ 𝑓 𝑑𝛼 = ∫ 𝑓 𝑑𝛼 .

Since ∫ 𝑓 𝑑𝛼 = ∫− 𝑓 𝑑𝛼 = sup 𝐿(𝑃, 𝑓, 𝛼) and


∫ 𝑓 𝑑𝛼 = ∫ 𝑓 𝑑𝛼 = inf 𝑈(𝑃, 𝑓, 𝛼) , there exist partitions 𝑃1 and 𝑃2 such that

𝜖
𝑈(𝑃1 , 𝑓, 𝛼) < ∫ 𝑓 𝑑𝛼 +
2
𝜖
and 𝐿 (𝑃2 , 𝑓, 𝛼) > ∫ 𝑓 𝑑𝛼 − or
2

𝜖
𝑈(𝑃1 , 𝑓, 𝛼) − ∫ 𝑓 𝑑𝛼 < (1.3.1)
2

𝜖
and ∫ 𝑓 𝑑𝛼 − 𝐿(𝑃2 , 𝑓, 𝛼) < 2 (1.3.2)

9
Let 𝑃 = 𝑃1 ∪ 𝑃2 , common refinement of 𝑃1 and 𝑃2 . Then by theorem 1.1.1 and from equations

(1.3.1) and (1.3.2), we have


𝜖
𝑈(𝑃, 𝑓, 𝛼) ≤ 𝑈(𝑃1 , 𝑓, 𝛼) < ∫ 𝑓 𝑑𝛼 + 2

< 𝐿(𝑃2 , 𝑓, 𝛼) + 𝜖

≤ 𝐿(𝑃, 𝑓, 𝛼) + 𝜖,

Hence 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 holds for partition 𝑃, which is as desired.

Theorem 1.4.2:

a) If for some 𝜖 > 0 there exists a partition P such that

𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 (1.4.1)

Then (1) holds for holds for every refinement of 𝑃, with the same 𝜖.

b) If (1) holds for 𝑃 = {𝑥0 , 𝑥1 , … , 𝑥𝑛 } and if 𝑠𝑖 , 𝑡𝑖 are arbitrary points in

[𝑥𝑖−1 , 𝑥𝑖 ], then ∑𝑛𝑖=1|𝑓(𝑠𝑖 ) − 𝑓(𝑡𝑖 )| Δ𝛼𝑖 < 𝜖 .

c) If 𝑓 ∈ ℜ(𝛼) and the hypothesis of (b) hold, then

𝑏
|∑𝑛𝑖=1 𝑓(𝑡𝑖 ) Δ𝛼𝑖 − ∫𝑎 𝑓 𝑑𝛼 | < 𝜖.

Proof:

If 𝑃∗ is any refinement of partition 𝑃 of [𝑎, 𝑏]. We have by theorem (1.4.1)

𝑈(P ∗ , 𝑓, 𝛼) ≤ 𝑈(𝑃, 𝑓, 𝛼), L(P , f, α) ≤ L(P * , f, α)

Since

L(P ∗ , f, α) ≤ 𝑈(P ∗ , 𝑓, 𝛼)

We have

L(P , f, α) ≤ L(P * , f, α) ≤ 𝑈(P ∗ , 𝑓, 𝛼) ≤ 𝑈(𝑃, 𝑓, 𝛼)

Hence

U(P * , f, α) - L(P * , f, α) ≤ 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼)

10
Therefore, if (1) holds, then

U(P * , f, α) - L(P * , f, α) < 𝜖

For every partition P ∗ that is a refinement of P

This proves (a).

b) Next, if 𝑠𝑖 , 𝑡𝑖 ∈ [𝑥𝑖−1, 𝑥𝑖 ] , we have 𝑓(𝑠𝑖 ) and 𝑓(𝑡𝑖 ) both lie in [𝑚𝑖 , 𝑀𝑖 ],

so that

|𝑓(𝑠𝑖 ) − 𝑓(𝑡𝑖 )| ≤ 𝑀𝑖 – 𝑚𝑖 .

Where,

𝑀𝑖 = sup 𝑓(𝑥) , 𝑚𝑖 = inf 𝑓(𝑥) , 1 ≤ 𝑖 ≤ 𝑛.


[𝑥𝑖−1 ≤ 𝑥 ≤𝑥𝑖 ] [𝑥𝑖−1 ≤𝑥 ≤ 𝑥𝑖 ]

Hence,

∑𝑛𝑖=1|𝑓(𝑠𝑖 ) − 𝑓(𝑡𝑖 )| Δ𝛼𝑖 ≤ ∑ni=1 Mi ΔαI , − ∑ni=1 mi Δαi .

= 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼)

If now (1) holds for P, then we have

∑𝑛𝑖=1|𝑓(𝑠𝑖 ) − 𝑓(𝑡𝑖 )| Δ𝛼𝑖 <𝜖

This proves (b).

c) Finally, if 𝑡𝑖 ∈ [𝑥𝑖−1, 𝑥𝑖 ] , we have mi ≤ f(𝑡𝑖 ) ≤ Mi , for all 𝑖, 1 ≤ 𝑖 ≤ 𝑛.

Therefore,

𝐿(𝑃, 𝑓, 𝛼) ≤ ∑𝑛𝑖=1 f(𝑡𝑖 ) Δ𝛼𝑖 ≤ 𝑈(𝑃, 𝑓, 𝛼) (2)

Also, we have

𝑏
𝐿(𝑃, 𝑓, 𝛼) ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼). (3)

From (1), (2) and (3) we have

𝑏
|∑𝑛𝑖=1 𝑓(𝑡𝑖 ) Δ𝛼𝑖 − ∫𝑎 𝑓 𝑑𝛼 | < 𝜖.

11
This proves (c). Hence the theorem.

1.5 EXAMPLES

1. Show that 𝑓(𝑥) = 𝑥 2 ∈ ℜ[𝑥 2 ] on [0,1].

Solution: Given 𝑓(𝑥) = 𝑥 2 and 𝛼(𝑥) = 𝑥 2 .

It is clear that 𝑓(𝑥) is bounded and 𝛼(𝑥) is monotonically increasing

1 𝑖 𝑛
Let 𝑃 = {0, 𝑛 , … , 𝑛 , … , 𝑛 = 1} be any partition of [0,1].

1 𝑖 (𝑖−1)2 1 𝑖 𝑖2
Then 𝑚𝑖 = inf [𝑖 − , ]= and 𝑀𝑖 = sup [𝑖 − , ]=
𝑛 𝑛 𝑛 𝑛 𝑛 𝑛

𝑖2 (𝑖−1)2 2𝑖−1
Since 𝛼(𝑥) = 𝑥 2 , Δ𝛼𝑖 = 𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ) = 𝑥𝑖2 − 𝑥𝑖−1
2
= 𝑛2 − = .
𝑛2 𝑛

The lower Riemann- Stieltjes sum


𝑛 𝑛
(𝑖 − 1)2 [2𝑖 − 1]
𝐿(𝑃, 𝑓, 𝛼) = ∑ 𝑚𝑖 Δ𝛼𝑖 = ∑
𝑛4
𝑖=1 𝑖=1

𝑛 𝑛 𝑛 𝑛
1
= [2 ∑ 𝑖 3 − 5 ∑ 𝑖 2 + 4 ∑ 𝑖 − ∑ 1]
𝑛4
1 1 1 1

1 𝑛2 (𝑛+1)2 𝑛(𝑛+1)(2𝑛+1) 𝑛(𝑛+1)


= 𝑛4 [2 −5 +4 − 𝑛]
4 6 2

1 1
lim 𝐿(𝑃, 𝑓, 𝛼) → 0 ⇒ lim 𝐿(𝑃, 𝑓, 𝛼) = [2 − 0 + 0 − 0] = 2.
||𝑃||→0 𝑛→∞

𝑖2 𝑖2 (𝑖−1)2
Now consider 𝑈(𝑃, 𝑓, 𝛼) = ∑𝑛1 𝑀𝑖 Δ𝛼𝑖 = ∑𝑛𝑖=1 𝑛2 [𝑛2 − ]
𝑛2

1
= 𝑛4 ∑𝑛𝑖=1[2𝑖 3 − 𝑖 2 ]

𝑛 𝑛
1
= 4 [∑ 2𝑖 3 − ∑ 𝑖 2 ]
𝑛
𝑖=1 𝑖=1

1 𝑛2 (𝑛+1)2 𝑛(𝑛+1)(2𝑛+1)
= 𝑛4
[2 4
− 6
]

12
1
Therefore lim 𝑈(𝑃, 𝑓, 𝛼) = 2 .
𝑛→∞

1 −1 1 1
Hence ∫−0 𝑥 2 𝑑[𝑥 2 ] = ∫0 𝑥 2 𝑑[𝑥 2 ] = ∫0 𝑥 2 𝑑[𝑥 2 ] = .
2

3
2. Evaluate ∫0 𝑥 𝑑([𝑥]), where [𝑥] is the greatest integer function.

Solution: Here 𝑓(𝑥) = 𝑥, 𝛼(𝑥) = [𝑥].

1 2 𝑖−1 𝑖 1 𝑖 1 𝑖
Let𝑃 = {0, 𝑛 , 𝑛 , … , , 𝑛 , … ,1, 1 + 𝑛 , … , 1 + 𝑛 , … , 2,2 + 𝑛 , … . ,2 + 𝑛 , … ,3} .
𝑛

𝑖−1
Consider 𝐿(𝑃, 𝑓, 𝛼)[0,1] = ∑𝑛𝑖=1 𝑚𝑖 Δ𝛼𝑖 = ∑𝑛𝑖=1 (𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ))
𝑛

1 2 𝑛−1
= 0 + 𝑛 (𝛼(𝑥2 ) − 𝛼(𝑥1 )) + 𝑛 ((𝛼(𝑥3 ) − 𝛼(𝑥2 )) + ⋯ + (𝛼(𝑥𝑛 ) − 𝛼(𝑥𝑛−1 ))
𝑛

1 2 1 2 3 2 𝑛−1 𝑛−1
([ ] − [ ]) + ([ ] − [ ]) + ⋯ + (1) = .
𝑛 𝑛 𝑛 𝑛 𝑛 𝑛 𝑛 𝑛
𝑛 𝑛
𝑖−1
𝐿(𝑃, 𝑓, 𝛼)[1,2] = ∑ 𝑚𝑖 Δ𝛼𝑖 = ∑ (1 + ) (𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ))
𝑛
𝑖=1 𝑖=1

1 2
= (𝛼(𝑥1 ) − 𝛼(𝑥0 )) + (1 + 𝑛) (𝛼(𝑥2 ) − 𝛼(𝑥1 )) + (1 + 𝑛) ((𝛼(𝑥3 ) − 𝛼(𝑥2 )) + ⋯ +

𝑛−1
(1 + ) (𝛼(𝑥𝑛 ) − 𝛼(𝑥𝑛−1 ))
𝑛

1 𝑛−1 𝑛−1
∴ 𝐿(𝑃, 𝑓, 𝛼)[1,2] = (1 − 1) + (1 + ) (1 − 1) + ⋯ + (1 + ) (2 − 1) = 1 + .
𝑛 𝑛 𝑛
𝑖−1
Now consider 𝐿(𝑃, 𝑓, 𝛼)[2,3] = ∑𝑛𝑖=1 𝑚𝑖 Δ𝛼𝑖 = ∑𝑛𝑖=1 (2 + ) (𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ))
𝑛

1 2
= 2(𝛼(𝑥1 ) − 𝛼(𝑥0 )) + (2 + 𝑛) (𝛼(𝑥2 ) − 𝛼(𝑥1 )) + (2 + 𝑛) ((𝛼(𝑥3 ) − 𝛼(𝑥2 )) + ⋯ +

𝑛−1
(2 + ) (𝛼(𝑥𝑛 ) − 𝛼(𝑥𝑛−1 ))
𝑛

1 𝑛−1 𝑛−1
∴ 𝐿(𝑃, 𝑓, 𝛼)[2,3] = 2(2 − 2) + (2 + ) (2 − 2) + ⋯ + (2 + ) (3 − 2) = 2 + .
𝑛 𝑛 𝑛

𝑛−1 𝑛−1 𝑛−1 1


∴ 𝐿(𝑃, 𝑓, 𝛼)[0,3] = +1+ +2+ = 3 [2 − ] → 6, 𝑎𝑠 𝑛 → ∞.
𝑛 𝑛 𝑛 𝑛

13
3
∴ ∫ 𝑓 𝑑𝛼 = 6.
−0

𝑖
Now consider, 𝑈(𝑃, 𝑓, 𝛼)[0,1] = ∑𝑛𝑖=1 𝑀𝑖 Δ𝛼𝑖 = ∑𝑛𝑖=1 𝑛 (𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ))

1 2 𝑛
= 𝑛 (𝛼(𝑥2 ) − 𝛼(𝑥1 )) + 𝑛 ((𝛼(𝑥3 ) − 𝛼(𝑥2 )) + ⋯ + 𝑛 (𝛼(𝑥𝑛 ) − 𝛼(𝑥𝑛−1 ))

= 0 + 0 + ⋯ + [1 − 0] = 1.
𝑛 𝑛
𝑖
𝑈(𝑃, 𝑓, 𝛼)[1,2] = ∑ 𝑀𝑖 Δ𝛼𝑖 = ∑ (1 + ) (𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ))
𝑛
𝑖=1 𝑖=1

𝑛
∴ U(𝑃, 𝑓, 𝛼)[1,2] = 0 + 0 + 0 + ⋯ + (1 + ) [2 − 1] = 2.
𝑛
𝑖
Now consider 𝑈(𝑃, 𝑓, 𝛼)[2,3] = ∑𝑛𝑖=1 𝑚𝑖 Δ𝛼𝑖 = ∑𝑛𝑖=1 (2 + 𝑛) (𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ))

𝑛
∴ 𝑈(𝑃, 𝑓, 𝛼)[1,2] = 0 + 0 + 0 + ⋯ + (2 + ) (3 − 2) = 3.
𝑛

∴ 𝑈(𝑃, 𝑓, 𝛼)[0,3] = 1 + 2 + 3 = 6.

3 −3
∴ ∫ 𝑓 𝑑𝛼 = ∫ 𝑓 𝑑𝛼 = 6 .
−0 0

3
∴ 𝑥 ∈ ℜ([𝑥]) and ∫0 𝑥 𝑑([𝑥]) = 6.

1 𝑖𝑓 𝑥 𝑖𝑠 𝑎 𝑟𝑎𝑡𝑖𝑜𝑛𝑎𝑙
3. Show that 𝑓(𝑥) = { is not Riemann- Stieltjes integrable on [𝑎, 𝑏] for
0 𝑖𝑓 𝑥 𝑖𝑠 𝑖𝑟𝑟𝑎𝑡𝑖𝑜𝑛𝑎𝑙

any monotonically increasing function 𝛼(𝑥).

Solution: Since the set of rationals and the set of irrationals are dense in 𝑅, on any subset of

𝑅, 𝑚 = 0 𝑎𝑛𝑑 𝑀 = 1.

For any partition 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} of [𝑎, 𝑏],

we have 𝑚𝑖 = 0 𝑎𝑛𝑑 𝑀𝑖 = 1 on each [𝑥𝑖−1 , 𝑥𝑖 ].

Hence 𝐿(𝑃, 𝑓, 𝛼) = ∑𝑛1 (0) Δ𝛼𝑖 = 0 and

𝑈(𝑃, 𝑓, 𝛼) = ∑𝑛1 (1) Δ𝛼𝑖 = 𝛼 (𝑏) − 𝛼(𝑎).

14
𝑏 𝑏
 ∫−𝑎 𝑓 𝑑𝛼 ≠ ∫−𝑎 𝑓 𝑑𝛼 and so 𝑓 ∉ ℜ(𝛼) on [𝑎, 𝑏] for any α.

4. Let 𝑓: [𝑎, 𝑏] → 𝑅: 𝑓(𝑥) = 𝑘 𝑎𝑛𝑑 𝛼: [𝑎, 𝑏] → 𝑅 be monotonically increasing function. Then

prove that 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [𝑎, 𝑏].

Solution: Let 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} be any partition of [𝑎, 𝑏].

Then 𝑚𝑖 = inf 𝑓(𝑥) = 𝑘 , 𝑀𝑖 = sup 𝑓(𝑥) = 𝑘.


[𝑥𝑖−1 ,𝑥𝑖 ] [𝑥𝑖−1 ,𝑥𝑖 ]

Now 𝑈(𝑃, 𝑓, 𝛼) = ∑𝑛1 𝑀𝑖 Δ𝛼𝑖 = 𝑘(Δ𝛼1 + Δ𝛼2 + ⋯ + Δ𝛼𝑛 ) = 𝑘(𝛼(𝑏) − 𝛼(𝑎),

𝑏
and so ∫−𝑎 𝑓 𝑑 𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

The lower R-S sum

𝐿(𝑃, 𝑓, 𝛼) = ∑𝑛1 𝑚𝑖 Δ𝛼𝑖 = 𝑘(Δ𝛼1 + Δ𝛼2 + ⋯ + Δ𝛼𝑛 ) = 𝑘(𝛼(𝑏) − 𝛼(𝑎),

−𝑏
and so ∫𝑎 𝑓 𝑑 𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

𝑏 −𝑏
 ∫−𝑎 𝑓 𝑑 𝛼=∫𝑎 𝑓 𝑑 𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [𝑎, 𝑏] 𝑎𝑛𝑑 ∫𝑓 𝑑𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

0 𝑖𝑓 𝑥 < 0 0 𝑖𝑓 𝑥 ≤ 0
5. Let 𝑓(𝑥) = { and 𝛼(𝑥) = { be two functions defined on [−1,1].
1 𝑖𝑓 𝑥 ≥ 0 1 𝑖𝑓 𝑥 > 0

Show that 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [−1, 1].


1
6. Evaluate ∫−1 𝑓 𝑑𝛼 .

Solution: Let 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} be any partition of [−1,1] not containing 0.

If 𝑥𝑖−1 < 0 < 𝑥𝑖 , then

Δ𝛼1 = 𝛼(𝑥1 ) − 𝛼(𝑥0 ) = 0 − 0 = 0,

Δ𝛼2 = 𝛼(𝑥2 ) − 𝛼(𝑥1 ) = 0 − 0 = 0,

Δ𝛼𝑖 = 𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ) = 1 − 0 = 1,

Δ𝛼𝑖+1 = 𝛼(𝑥𝑖+1 ) − 𝛼(𝑥𝑖 ) = 1 − 1 = 0,

15
… , Δ𝛼𝑛 = 𝛼(𝑥𝑛 ) − 𝛼(𝑥𝑛−1 ) = 1 − 1 = 0 and

therefore

𝑈(𝑃, 𝑓, 𝛼) = 𝑀1 .0 + 𝑀2 . 0 + ⋯ + 𝑀𝑖 . 1 + ⋯ + 𝑀𝑛 .0 = 𝑀𝑖

But 𝑀𝑖 = sup 𝑓(𝑥) = 1.


x∈[xi−1 ,xi ]

Similarly, 𝑚𝑖 = inf 𝑓(𝑥) = 0 and 𝐿(𝑃, 𝑓, 𝛼) = 0.


x∈[xi−1 ,xi ]

Hence 𝑈(𝑃, 𝑓, 𝛼) = 1 𝑎𝑛𝑑 𝐿(𝑃, 𝑓, 𝛼) = 0 for any partition 𝑃 of [𝑎, 𝑏] such that 0 ∉ 𝑃.

If 0 ∈ 𝑃 𝑎𝑛𝑑 𝑥𝑖 = 0 𝑓𝑜𝑟 𝑠𝑜𝑚𝑒 𝑖,

then Δ𝛼1 = 𝛼(𝑥1 ) − 𝛼(𝑥0 ) = 0,

Δ𝛼2 = 𝛼(𝑥2 ) − 𝛼(𝑥1 ) = 0, …, Δ𝛼𝑖 = 𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ) = 0,

Δ𝛼𝑖+1 = 𝛼(𝑥𝑖+1 ) − 𝛼(𝑥𝑖 ) = 1 − 0 = 1,…,

Δ𝛼𝑛 = 𝛼(𝑥𝑛 ) − 𝛼(𝑥𝑛−1 ) = 1 − 0 = 0,

 𝑈(𝑃, 𝑓, 𝛼) = ∑𝑛1 𝑀𝑖 Δ𝛼𝑖 = 𝑀𝑖+1 Δ𝛼𝑖+1 = 1 𝑎𝑛𝑑

𝐿(𝑃, 𝑓, 𝛼) = ∑𝑛1 𝑚𝑖 Δ𝛼𝑖 = 𝑚𝑖+1 Δ𝛼𝑖+1 = 1, for any partition 𝑃 of [−1,1] containing 0.

Thus we have for any partition 𝑃 of [−1,1], 𝑈(𝑃, 𝑓, 𝛼) = 1, where as

0 𝑖𝑓 0 ∉ 𝑃
𝐿(𝑃, 𝑓, 𝛼) = { .
1 𝑖𝑓 0 ∈ 𝑃

 sup𝑃 𝐿(𝑃, 𝑓, 𝛼) = 1 and inf𝑃 𝑈(𝑃, 𝑓, 𝛼) = 1 and they are equal.

1
Hence 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [−1, 1] and ∫−1 𝑓 𝑑𝛼 = 1.

7. Suppose 𝛼 increases on [𝑎, 𝑏] and 𝑎 ≤ 𝑐 ≤ 𝑏, 𝛼 is continuous at 𝑐 and 𝑓(𝑐) = 1 and 𝑓(𝑥) =

𝑏
0 when 𝑥 ≠ 𝑐. Show that 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [𝑎, 𝑏] and ∫𝑎 𝑓 𝑑𝛼 = 0.

Solution: Let 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} be any partition of [𝑎, 𝑏]

and let 𝑥𝑖−1 ≤ 𝑐 ≤ 𝑥𝑖 .

Consider 𝑈(𝑃, 𝑓, 𝛼) = 𝑀1 Δ𝛼1 + 𝑀2 Δ𝛼2 + … + 𝑀𝑛 Δ𝛼𝑛 = Δ𝛼𝑖


16
and 𝐿 (𝑃, 𝑓, 𝛼) = 𝑚1 Δ𝛼1 + 𝑚2 Δ𝛼2 + … + 𝑚𝑛 Δ𝛼𝑛 = 0.

 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) = Δ𝛼𝑖 = 𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ) (2.4.10)

Since 𝛼 is continuous at 𝑐, given 𝜖 > 0 there exists 𝛿 > 0 such that |𝛼(𝑥) − 𝛼(𝑐)| < 𝜖

whenever |𝑥 − 𝑐| < 𝛿 for all 𝑥.

If ||𝑃|| < 𝛿, then 𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ) = |𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 )|

=|𝛼(𝑥𝑖 ) − 𝛼(𝑐) + 𝛼(𝑐) − 𝛼(𝑥𝑖−1 )|

≤ |𝛼(𝑥𝑖 ) − 𝛼(𝑐)| + |𝛼(𝑐) − 𝛼(𝑥𝑖−1 )|

< 2𝜖, whenever |𝑥 − 𝑐| < 𝛿, for all 𝑥.

𝑏
Further ∫𝑎 𝑓 𝑑𝛼 = sup 𝐿(𝑃, 𝑓, 𝛼) = 0.

8. Compute U(P, f, α) and L(P, f, α) in [0,2] for the following:

Let f(x) = x − [x] and α(x) = [x].

Let f(x) = x and α(x) = [x] − x.

Furhter check whether f ∈ ℜ(α) on [0, 2].

1.6 Summary

1. If P and P * are partitions of [a, b] such that P * is a refinement P, then

L(P, f, α) ≤ L(P * , f, α) and 𝑈(𝑃, 𝑓, 𝛼) ≥ 𝑈(𝑃∗ , 𝑓, 𝛼), where f is a bounded function

and α is a monotonically increasing function on [a, b].

2. Let 𝑓 be a real bounded function defined on [𝑎, 𝑏] and let 𝛼 be monotonically increasing

𝑏 −𝑏
on[𝑎, 𝑏], then ∫−𝑎 𝑓 𝑑𝛼 ≤ ∫𝑎 𝑓 𝑑𝛼 .

3. A function 𝑓 is Riemann- Steiltje’s integrable on [𝑎, 𝑏] if and only if for every 𝜖 > 0 there

exists a partition 𝑃 of [𝑎, 𝑏] such that 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 .

17
1.7. Keywords

Riemann- Stieltjes integral, partition, bounded, monotonic, continuous, lower Riemann- Stieltjes
sum.

1.8. Exercises

1 𝑖𝑓 𝑥 𝑖𝑠 𝑎 𝑟𝑎𝑡𝑖𝑜𝑛𝑎𝑙
1. Show that 𝑓(𝑥) = { is not Riemann-Steiltje’s integrable on [𝑎, 𝑏] for
0 𝑖𝑓 𝑥 𝑖𝑠 𝑖𝑟𝑟𝑎𝑡𝑖𝑜𝑛𝑎𝑙

any monotonically increasing function 𝛼(𝑥).

2. Let 𝑓: [𝑎, 𝑏] → 𝑅: 𝑓(𝑥) = 𝑘 𝑎𝑛𝑑 𝛼: [𝑎, 𝑏] → 𝑅 be monotonically increasing function. Then

prove that 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [𝑎, 𝑏].

3. State and prove the necessary and sufficient condition for the Riemann-Steiltje’s integrability

of a bounded function.

4. With usual notations prove L(P, f, α) ≤ L(P * , f, α) and 𝑈(𝑃, 𝑓, 𝛼) ≥ 𝑈(𝑃∗ , 𝑓, 𝛼), where f
is a bounded function and α is a monotonically increasing function on [𝑎, 𝑏].

Solutions:

1. Since the set of rationals and and the set of irrationals are dense in 𝑅, on any subset of

𝑅, 𝑚 = 0 𝑎𝑛𝑑 𝑀 = 1.

For any partition 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} of [𝑎, 𝑏] , we have 𝑚𝑖 = 0 𝑎𝑛𝑑 𝑀𝑖 =

1 𝑜𝑛 𝑒𝑎𝑐ℎ [𝑥𝑖−1 , 𝑥𝑖 ]

 𝐿(𝑃, 𝑓, 𝛼) = ∑𝑛1(0)Δ𝛼𝑖 = 0 and 𝑈(𝑃, 𝑓, 𝛼) = ∑𝑛1(1)Δ𝛼𝑖 = 𝛼(𝑏) − 𝛼(𝑎).

𝑏 𝑏
 ∫−𝑎 𝑓 𝑑𝛼 ≠ ∫−𝑎 𝑓 𝑑𝛼 .

 𝑓 ∉ ℜ(𝛼) 𝑜𝑛 [𝑎, 𝑏] 𝑓𝑜𝑟 𝑎𝑛𝑦 𝛼.

18
2. Let 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} be any partition of [𝑎, 𝑏]. Then 𝑚𝑖 = inf 𝑓(𝑥) = 𝑘 ,
[𝑥𝑖−1 ,𝑥𝑖 ]

𝑀𝑖 = sup 𝑓(𝑥) = 𝑘 and 𝑈(𝑃, 𝑓, 𝛼) = ∑𝑛1 𝑀𝑖 Δ𝛼𝑖 = 𝑘(Δ𝛼1 + Δ𝛼2 + ⋯ + Δ𝛼𝑛 ) =


[𝑥𝑖−1 ,𝑥𝑖 ]

𝑘(𝛼(𝑏) − 𝛼(𝑎).

𝑏
And so ∫−𝑎 𝑓 𝑑 𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

The lower R-S sum

𝐿(𝑃, 𝑓, 𝛼) = ∑𝑛1 𝑚𝑖 Δ𝛼𝑖 = 𝑘(Δ𝛼1 + Δ𝛼2 + ⋯ + Δ𝛼𝑛 ) = 𝑘(𝛼(𝑏) − 𝛼(𝑎).

−𝑏
And so ∫𝑎 𝑓 𝑑 𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

𝑏 −𝑏
 ∫−𝑎 𝑓 𝑑 𝛼=∫𝑎 𝑓 𝑑 𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [𝑎, 𝑏] 𝑎𝑛𝑑 ∫𝑓 𝑑𝛼 = 𝑘(𝛼(𝑏) − 𝛼(𝑎)).

1.9. References

1. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing,1974.


2. Walter Rudin, Principles of Mathematical Analysis,McGraw-Hill 3rd edition, 1976.

19
UNIT 2
THE PROPERTIES OF INTEGRAL

2.1 Main Objectives

2.2 Introduction

2.3 The classes of Riemann-Steiltje’s Integral

2.4 The Properties of Riemann-Steiltje’s integral

2.5 Summary

2.6 Key Words

2.7 Exercises

2.8 References

20
UNIT 2
THE PROPERTIES OF INTEGRALS

2.1 Main Objectives


The key features that are introduced in this unit are,
 Algebraic properties of Riemann- Stieltjes integral.
 Different classes of Riemann- Stieltjes integrable functions.
 Apply the properties to evaluate the Riemann- Stieltjes integral of different
combinations of functions.

2.2 Introduction
In this unit we shall study the algebraic properties of the Riemann-Stieltje’s integral. i.e.

sum, difference and product of two R-S integrable functions are again R-S integrable . If 𝑓 is R-

S integrable then – 𝑓 , |𝑓| and 𝑓 2 are all R-S integrable. Furhter we shall study integral of

continuous functions with respect to monotonic function, integral of monotonic functions with

respect to continuous monotonic functions and integral of continuous function of a R-S

integrable function are all R-S integrable.

2.3 The Class of Riemann-Stieltjes Integrable Function

Definition2.3.1: Let 𝑓 be a bounded and 𝛼 be increasing function on [𝑎, 𝑏]. Let 𝑃 =

{𝑥0 , 𝑥1 , … , 𝑥𝑛 } and let 𝑄 = {𝜉0 , 𝜉1 , … , 𝜉𝑛 } an intermediate partition of 𝑃 such that 𝑥𝑖−1 ≤ 𝜉𝑖 ≤ 𝑥𝑖

for 𝑖 = 1,2, … , 𝑛. Then Riemann-Steiltje’s sum 𝑆(𝑃, 𝑄, 𝑓, 𝛼)is defined by ∑𝑛𝑖=1 𝑓(𝜉𝑖 )Δ𝛼𝑖 .

We say that lim 𝑆(𝑃, 𝑄, 𝑓, 𝛼) = 𝐿 if and only if for every 𝜖 > 0 there exists a 𝛿 > 0 such
||𝑃||→0

that ||𝑃|| < 𝛿 ⇒ |𝑆(𝑃, 𝑄, 𝑓, 𝛼) − 𝐿| < 𝜖, where ||𝑃|| denote the norm of 𝑃 defined by ||𝑃|| =

max 𝑙([𝑥𝑖−1 , 𝑥𝑖 ]), maximum of lengths of its subintervals.


21
Theorem 2.3.1: If lim 𝑆(𝑃, 𝑄, 𝑓, 𝛼) exists then 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and
||𝑃||→0

𝑏
lim 𝑆(𝑃, 𝑄, 𝑓, 𝛼) = ∫ 𝑓 𝑑 α.
||𝑃||→0 𝑎

Proof: Suppose that lim 𝑆(𝑃, 𝑄, 𝑓, 𝛼) = ∑𝑛𝑖=1 𝑓(𝜉𝑖 )Δ𝛼𝑖 exists and is 𝐿. Then for every 𝜖 > 0,
||𝑃||→0

𝜖
there exists a 𝛿 > 0 such that ||𝑃|| < 𝛿 ⇒ |𝑆(𝑃, 𝑄, 𝑓, 𝛼) − 𝐿| < 4

𝜖 𝜖
⇒ 𝐿 − 4 < 𝑆(𝑃, 𝑄, 𝑓, 𝛼) < 𝐿 + 4 (1.4.2)

Choosing one such 𝑃 and let the points 𝜉𝑖 range over the intervals [𝑥𝑖−1 , 𝑥𝑖 ] and taking infimum

and supremum of the numbers 𝑆(𝑃, 𝑄, 𝑓, 𝛼) so obtained, (1.) gives


𝜖 𝜖
𝐿 − 4 ≤ 𝐿(𝑃, 𝑓, 𝛼) ≤ 𝑈(𝑃, 𝑓, 𝛼) ≤ 𝐿 + 4 (2.3.1)

𝜖
 𝑈(𝑃, 𝑓, 𝛼) ≤ 𝐿 + 4 , (2.3.2)

𝜖
and −𝐿(𝑃, 𝑓, 𝛼) ≤ −𝐿 + 4 (2.3.3)

Adding (2.3.2) and (2.3.3), we get

𝜖
𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) ≤ < 𝜖.
4

It follows from theorem (1.4.1), 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏].

𝑏
We have 𝐿(𝑃, 𝑓, 𝛼) ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼) (2.3.4)

𝜖 𝑏 𝜖
From (2.3.1) and (2.3.4), we have 𝐿 − 4 ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝐿 + 4.

𝑏
Since 𝜖 > 0 is arbitrary, it follows that lim 𝑆(𝑃, 𝑄, 𝑓, 𝛼) = ∫𝑎 𝑓 𝑑𝛼.
||𝑃||⟶0

22
𝑏
Theorem 2.3.2: If 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] then show that −𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and ∫𝑎 (−𝑓)𝑑𝛼 =

𝑏
− ∫𝑎 𝑓 𝑑𝛼 .

Proof: It is clear that 𝑈(𝑃, −𝑓, 𝛼) = ∑𝑛𝑖=1 sup (−f(x))Δ𝛼𝑖


𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ]

= − ∑𝑛𝑖=1 inf (f(x))Δ𝛼𝑖


𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ]

=− ∑𝑛𝑖=1 mi Δ𝛼𝑖 = −𝐿(𝑃, 𝑓, 𝛼). (2.3.2)

Similarly, 𝐿(𝑃, −𝑓, 𝛼) = −𝑈(𝑃, 𝑓, 𝛼) . (2.3.3)

From (1) and (2), we have

𝑈(𝑃, −𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) = −𝐿(𝑃, 𝑓, 𝛼) − (−𝑈(𝑃, 𝑓, 𝛼))

= 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖, since 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏].

Therefore −𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏].

𝑏 𝑏
Next to prove ∫𝑎 (−𝑓)𝑑𝛼 = − ∫𝑎 𝑓 𝑑𝛼 .

𝑏
Since ∫𝑎 (−𝑓)𝑑𝛼 = inf 𝑈(𝑃, −𝑓, 𝛼) , for a given 𝜖 > 0, there exists a partition 𝑃 of [𝑎, 𝑏]
𝑃

𝑏
such that , ∫𝑎 −𝑓 𝑑𝛼 + 𝜖 > 𝑈(𝑃, −𝑓, 𝛼) = −𝐿(𝑃, 𝑓, 𝛼) . (2.3.4)

𝑏
But ∫𝑎 𝑓 𝑑𝛼 ≥ 𝐿(𝑃, 𝑓, 𝛼) and so

𝑏
− ∫𝑎 𝑓 𝑑𝛼 ≤ − 𝐿(𝑃, 𝑓, 𝛼) . (2.3.5)

From (2.3.4) and (2.4.5), we have,

𝑏 𝑏
− ∫𝑎 𝑓 𝑑𝛼 < ∫𝑎 −𝑓 𝑑𝛼 + 𝜖. (2.3.6)

𝑏 𝑏
Since 𝜖 > 0 is arbitrary, − ∫𝑎 𝑓 𝑑𝛼 ≤ ∫𝑎 −𝑓 𝑑𝛼 . (2.3.7)

𝑏
Since ∫𝑎 −𝑓 𝑑𝛼 = sup 𝐿(𝑃, −𝑓, 𝛼), we have for a given 𝜖 > 0, there exists a partition 𝑃

of [𝑎, 𝑏] such that

23
𝑏
∫𝑎 −𝑓 𝑑𝛼 − 𝜖 < L(𝑃, −𝑓, 𝛼) = −𝑈(𝑃, 𝑓, 𝛼). (2.3.8)

𝑏
The inequality ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼) implies

𝑏
− ∫𝑎 𝑓 𝑑𝛼 ≥ − 𝑈(𝑃, 𝑓, 𝛼) . (2.3.9)

From (2.3.8) and (2.3.9), we have,

𝑏 𝑏
∫𝑎 −𝑓 𝑑𝛼 − 𝜖 < 𝑈(𝑃, 𝑓, 𝛼) ≤ − ∫𝑎 𝑓 𝑑𝛼

𝑏 𝑏
⇒ ∫𝑎 −𝑓 𝑑𝛼 ≤ − ∫𝑎 𝑓 𝑑𝛼 , since 𝜖 > 0 is arbitrary,

𝑏 𝑏
− ∫𝑎 𝑓 𝑑𝛼 ≤ ∫𝑎 −𝑓 𝑑𝛼 . (2.3.10)

Combining (2.3.7) and (2.3.10), we get

𝑏 𝑏
∫𝑎 −𝑓 𝑑𝛼 = − ∫𝑎 𝑓 𝑑𝛼 .

Theorem 2.3.3: If 𝑓 is continuous function on [𝑎, 𝑏], then 𝑓 is Riemann-Steiltje’s integrable on

[𝑎, 𝑏].

Proof: Let 𝑓 be a continuous function and 𝛼 be a monotonically increasing function on [𝑎, 𝑏].

Given 𝜖 > 0, choose a positive integer 𝑛 such that

𝛼(𝑏)−𝛼(𝑎) 1
𝜖
< 𝑛
or 𝑛(𝛼(𝑏) − 𝛼(𝑎)) < 𝜖 . (1)

Since 𝑓 is uniformly continuous on [𝑎, 𝑏].

(⸪ A continuous function on [𝑎, 𝑏] is uniformly continuous) there exists a 𝛿 = 𝛿(𝜖) such that

|𝑓(𝑥) − 𝑓(𝑦)| < 𝑛 whenever |𝑥 − 𝑦| < 𝛿 .

Let 𝑃 = {𝑥0 , 𝑥1 , … , 𝑥𝑛 } be a partition of [𝑎, 𝑏] such that the length of the largest sub interval is

less than 𝛿. Then length of every subinterval is less than 𝛿.

If 𝑥, 𝑦 ∈ [𝑥𝑖−1 , 𝑥𝑖 ], then |𝑥 − 𝑦| < 𝛿 and |𝑓(𝑥) − 𝑓(𝑦)| < 𝑛 .

Taking sup over all ∈ [𝑥𝑖−1 , 𝑥𝑖 ] , we obtain |𝑀𝑖 − 𝑓(𝑦)| ≤ 𝑛.

Take infimum of the above over all 𝑦 ∈ [𝑥𝑖−1 , 𝑥𝑖 ] to get

24
𝑀𝑖 − 𝑚𝑖 ≤ 𝑛 . for all i (2)

Now consider

𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) = ∑𝑛𝑖=1 𝑀𝑖 Δ𝛼𝑖 − ∑𝑛𝑖=1 𝑚𝑖 Δ𝛼𝑖 = ∑𝑛𝑖=1(𝑀𝑖 − 𝑚𝑖 )Δ𝛼𝑖 ≤ ∑𝑛𝑖=1 𝑛 Δ𝛼𝑖 =

𝑛 ∑𝑛𝑖=1 Δ𝛼𝑖 = 𝑛(𝛼(𝑏) − 𝛼(𝑎)) < 𝜖.

Hence by theorem 1.4.1, it follows that 𝑓 ∈ ℜ(𝛼).

This completes the proof.

Theorem 2.3.4: Let 𝑓 be monotonic and 𝛼 be monotonic and continuous on [𝑎, 𝑏]. Then 𝑓 ∈

ℜ(𝛼).

Proof: Let 𝜖 > 0 be given. Suppose 𝛼 is increasing on [𝑎, 𝑏]. Since 𝛼 is continuous, for any

positive number n , we can choose a partition 𝑃 = {𝑥0 , 𝑥1 , … , 𝑥𝑛 } of [𝑎, 𝑏] such that

𝛼(𝑏)−𝛼(𝑎)
Δ𝛼𝑖 = , 𝑖 = 1,2, … , 𝑛 (3)
𝑛

Since 𝛼 is monotonic increasing on the closed interval [𝑎, 𝑏], it assumes every value between its

bounds 𝛼(𝑎) and 𝛼(𝑏).

Since 𝑓 is monotonic increasing on [𝑎, 𝑏], its lower and the upper bounds 𝑚𝑖 and 𝑀𝑖 are given

by

𝑚𝑖 = 𝑓(𝑥𝑖−1 ), 𝑀𝑖 = 𝑓(𝑥𝑖 ), 𝑖 = 1,2, … , 𝑛.

So that 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) = ∑𝑛𝑖=1 (𝑀𝑖 − 𝑚𝑖 )Δ𝛼𝑖

(𝛼(𝑏)−𝛼(𝑎))
= ∑𝑛𝑖=1 (𝑀𝑖 − 𝑚𝑖 ) , follows from 3.
𝑛

(𝛼(𝑏)−𝛼(𝑎))
= ∑𝑛𝑖=1(𝑀𝑖 − 𝑚𝑖 )
𝑛

(𝛼(𝑏)−𝛼(𝑎))
= ∑𝑛𝑖=1 (𝑓(𝑥𝑖 ) − 𝑓(𝑥𝑖−1 )
𝑛

(𝛼(𝑏)−𝛼(𝑎))
= (𝑓(𝑏) − 𝑓(𝑎)) .
𝑛

25
Given any positive number 𝜖, we can choose a partition 𝑃 such that

(𝛼(𝑏)−𝛼(𝑎))
(𝑓(𝑏) − 𝑓(𝑎)) < 𝜖.
𝑛

Therefore U(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 .

Hence 𝑓 ∈ ℜ(𝛼) 𝑜𝑛 [𝑎, 𝑏].

Theorem 2.3.5: Suppose 𝑓 is bounded on [𝑎, 𝑏] , 𝑓 has only finitely many discontinuities on

[𝑎, 𝑏], and 𝛼 is continuous at every point at which 𝑓 is discontinuous. Then 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏].

Proof: Let 𝜖 > 0 be given and let 𝑀 = sup|𝑓(𝑥)|. Let 𝐸 be the set of discontinuity of 𝑓. Since 𝐸

is finite and 𝛼 is continuous at every point of 𝐸 there exist finitely many disjoint intervals[𝑥𝑖 , 𝑦𝑗 ]

in [𝑎, 𝑏] such that ∑ (𝛼(𝑥𝑖 ) − 𝛼(𝑦𝑗 )) < 𝜖. These intervals can be so arranged that 𝐸 ∩ (𝑎, 𝑏) ⊆

𝑖𝑛𝑡 (𝑥𝑖 , 𝑦𝑗 ), for some 𝑖 and 𝑗. The set 𝐶 obtained from [𝑎, 𝑏] by removing the segments

(𝑥𝑖 , 𝑦𝑗 )is compact. Hence the continuous function 𝑓 on the compact set 𝐶 is uniformaly

continuous. Therefore there exists a 𝛿 > 0 such that for 𝑥, 𝑦 ∈ 𝐶, |𝑥 − 𝑦| < 𝛿  |𝑓(𝑥) −

𝑓(𝑦)| < 𝜖.

Let 𝑃 = {𝑎 = 𝑡0 , 𝑡1 , … , 𝑡𝑛 = 𝑏} be a partition of [𝑎, 𝑏] such that only the endpoints of each

segment (𝑥𝑖 , 𝑦𝑗 ) lie in 𝑃. If 𝑡𝑖−1 ∉ 𝑃 then Δ𝑡𝑖 < 𝛿.

Since 𝑀 = sup|𝑓(𝑥)|, for each 𝑖, 𝑀𝑖 − 𝑚𝑖 ≤ 2𝑀 and 𝑀𝑖 − 𝑚𝑖 ≤ 𝜖 provided 𝑡𝑖−1 ≠ 𝑥𝑗 for j.

Thus we have 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) ≤ [𝛼(𝑏) − 𝛼(𝑎)]𝜖 + 2𝑀𝜖. Since 𝜖 > 0 is arbitrary, it

follows that 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏].

Theorem 2.3.6: Let 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and 𝑚 ≤ 𝑓(𝑥) ≤ 𝑀, ∀ 𝑥 ∈ [𝑎, 𝑏]. If 𝜙 is a continuous

function on [𝑚, 𝑀] and ℎ(𝑥) = 𝜙 (𝑓(𝑥)) ∀ 𝑥 ∈ [𝑎, 𝑏] then ℎ ∈ ℜ(𝛼) on [𝑎, 𝑏].

26
Proof: Since a continuous function on a closed and bounded interval is uniformly continuous,

the function 𝜙 is uniformaly continuous on [𝑚, 𝑀]. Given any 𝜖 > 0 there exists a 𝛿 > 0 such

that |𝜙(𝑥) − 𝜙(𝑦)| < 𝜖 whenever |𝑥 − 𝑦| < 𝛿. (4)

Given 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏], for 𝛿 > 0 there exists a partition 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} of

[𝑎, 𝑏] such that 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝛿 2 .

i.e. ∑(𝑀𝑖 − 𝑚𝑖 )Δ𝛼𝑖 < 𝛿 2 , (5)

where 𝑀𝑖 = sup 𝑓(𝑥), 𝑚𝑖 = inf 𝑓(𝑥).


𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ] (𝑥𝑖−1 ≤𝑥≤𝑥𝑖 )

Define 𝑀𝑖∗ = sup ℎ(𝑥) , 𝑚𝑖∗ = inf h(x).


𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ] (𝑥𝑖−1 ≤𝑥≤𝑥𝑖 )

If 𝑀𝑖 − 𝑚𝑖 < 𝛿 then put 𝑖 ∈ 𝐴 and if 𝑀𝑖 − 𝑚𝑖 ≥ 𝛿 then put 𝑖 ∈ 𝐵.

Thus {1,2, … , 𝑛} is divided into two sets 𝐴 and 𝐵 such that 𝐴 ∩ 𝐵 = 𝜙.

Let 𝑓(𝑡) = 𝑀𝑖 , 𝑓(𝑠) = 𝑚𝑖 , 𝑡, 𝑠 ∈ [𝑥𝑖−1 , 𝑥𝑖 ].

Then for 𝑖 ∈ 𝐴, |𝑓(𝑡) − 𝑓(𝑠)| < 𝛿 which from (2.2.4) implies

|ℎ(𝑡) − ℎ(𝑠)| = |𝜙(𝑓(𝑡) − 𝜙(𝑓(𝑠)| < 𝜖 , 𝑡, 𝑠, [𝑥𝑖−1 , 𝑥𝑖 ]. (6)

By definition of 𝑀𝑖∗ 𝑎𝑛𝑑 𝑚𝑖∗ , there exist 𝑡1 , 𝑠1 ∈ [𝑥𝑖−1 , 𝑥𝑖 ] such that ℎ(𝑡1 ) = 𝑀𝑖∗ and ℎ(𝑠1 ) =

𝑚𝑖∗ and so (2.2.6), implies

|𝑀𝑖∗ − 𝑚𝑖∗ | < 𝜖.

Thus we have proved if 𝑖 ∈ 𝐴 then 𝑀𝑖∗ − 𝑚𝑖∗ < 𝜖. (7)

If 𝑖 ∈ 𝐵 then 𝑀𝑖 − 𝑚𝑖 ≥ 𝛿 , so multiplying the inequality throught by Δ𝛼𝑖 and taking

summation over 𝑖 = 1, … , 𝑛, we obtain

∑𝑖∈𝐵 𝛿Δ𝛼𝑖 < ∑𝑛𝑖=1(𝑀𝑖 − 𝑚𝑖 )Δ𝛼𝑖 < 𝛿 2 ∑i∈B Δαi < 𝛿 .

Now consider,

𝑈(𝑃, ℎ, 𝛼) − 𝐿 (𝑃, ℎ, 𝛼) = ∑𝑛𝑖=1(𝑀𝑖∗ − 𝑚𝑖∗ )Δ𝛼𝑖

27
= ∑𝑖∈𝐴 (𝑀𝑖∗ − 𝑚𝑖∗ )Δ𝛼𝑖 + ∑𝑖∈𝐵 (𝑀𝑖∗ − 𝑚𝑖∗ )Δ𝛼𝑖

< ∑𝑖∈𝐴 𝜖Δ𝛼𝑖 + ∑𝑖∈𝐵(𝑀𝑖∗ − 𝑚𝑖∗ )Δ𝛼𝑖

From (7), if we let |ℎ(𝑥)| ≤ 𝑀, ∀ 𝑥 ∈ [𝑎, 𝑏], then

𝑀𝑖∗ ≤ 𝑀 and 𝑚𝑖∗ ≤ 𝑀, so that 𝑀𝑖∗ − 𝑚𝑖∗ ≤ 𝑀𝑖∗ + (−𝑚𝑖∗ ) ≤ 2𝑀 .

Therefore U(P, h, α) − L(P, h, α) < ∑i∈A ϵ Δαi + ∑i∈B 2M Δαi

< 𝜖[𝛼(𝑏) − 𝛼(𝑎)] + 2 𝑀 𝛿.

We choose 𝛿 such that 𝛿 < 𝜖. Then

𝑈(𝑃, ℎ, 𝛼) − 𝐿(𝑃, ℎ, 𝛼) < 𝜖 [𝛼(𝑏) − 𝛼(𝑎) + 2 𝑀],

Since 𝜖 is arbitrary, we have ℎ ∈ ℜ(𝛼) on [𝑎, 𝑏].

2.4 The properties of Riemann-Stieltjes Integral

Theorem 2.4.1: If 𝑓1 ∈ ℜ(𝛼) and 𝑓2 ∈ ℜ(𝛼) on [𝑎, 𝑏], then 𝑓1 + 𝑓2 ∈ ℜ(𝛼) on [𝑎, 𝑏] and

𝑏 𝑏 𝑏
∫𝑎 (𝑓1 + 𝑓2 )𝑑𝛼 = ∫𝑎 𝑓1 𝑑𝛼 + ∫𝑎 𝑓2 𝑑𝛼 .

Proof: Let 𝑔 = 𝑓1 + 𝑓2 . Then 𝑈(𝑃, 𝑔, 𝛼) = ∑ Sup 𝑔(𝑥) Δ𝛼𝑖 , where supremum is taken over all

𝑥 ∈ [𝑥𝑖−1 , 𝑥𝑖 ].
𝑛

𝑈(𝑃, 𝑔, 𝛼) = ∑ sup(𝑓1 + 𝑓2 ) (𝑥)Δ𝛼𝑖


𝑖=1

≤ ∑𝑛𝑖=1 ( sup f1 (x) + sup 𝑓2 (𝑥)) Δ𝛼𝑖


𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ] 𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ]

≤ ∑𝑛𝑖=1( sup f1 (x) + ∑𝑛𝑖=1 sup 𝑓2 (𝑥)) Δ𝛼𝑖


𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ] 𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ]

 𝑈(𝑃, 𝑔, 𝛼) ≤ 𝑈(𝑃, 𝑓1 , 𝛼) + 𝑈(𝑃, 𝑓2 , 𝛼). (2.4.1)

Similarly, 𝐿(𝑃, 𝑔, 𝛼) ≥ 𝐿(𝑃, 𝑓1 , 𝛼) + 𝐿(𝑃, 𝑓2 , 𝛼), from which we get

−𝐿(𝑃, 𝑔, 𝛼) ≤ − 𝐿(𝑃, 𝑓1 , 𝛼) − 𝐿(𝑃, 𝑓2 , 𝛼), (2.4.2)

From (2.4.1) and (2.4.2), we have

28
𝑈(𝑃, 𝑔, 𝛼) − 𝐿(𝑃, 𝑔, 𝛼) ≤ 𝑈(𝑃, 𝑓1 , 𝛼) − 𝐿(𝑃, 𝑓1 , 𝛼) + 𝑈(𝑃, 𝑓2 , 𝛼) − 𝐿(𝑃, 𝑓2 , 𝛼) (2.4.3)

Let ϵ > 0 be given. Since f1 ∈ ℜ(α) and f1 ∈ ℜ(α) on [a, b], there exist partitions, P1 and P2 of

[a, b] such that


ϵ
U(P1 , f1 , α) − L(P1 , f1 , α) < 2
(2.4.4)

ϵ
and U(P2 , f2 , α) − L(P2 , f2 , α) < 2
, (2.4.5)

Let P = P1 ∪ P2 .

Then U(P, f1 , α) ≤ U(P1 , f1 , α) and L(P, f1 , α) ≥ L(P1 , f1 , α) .

U(P, f1 , α) − L(P, f1 , α) ≤ U(P1 , f1 , α) − L(P1 , f1 , α) (2.4.6)

U(P, f2 , α) − L(P, f2 , α) ≤ U(P2 , f2 , α) − L(P2 , f2 , α) (2.4.7)

From (2.4.3), (2.4.4), (2.4.5), (2.4.6) and (2.4.7), we have

𝑈(𝑃, 𝑔, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 which by theorem 1.1.3, proves the first part .

𝑏 𝑏 𝑏
Next to prove ∫𝑎 (𝑓1 + 𝑓2 )𝑑𝛼 = ∫𝑎 𝑓1 𝑑𝛼 + ∫𝑎 𝑓2 𝑑𝛼 .

Given any 𝜖 > 0, there exists a partition 𝑃 such that

𝑏
∫ 𝑔 𝑑𝛼 + 𝜖 > 𝑈(𝑃, 𝑔, 𝛼) ≥ 𝐿(𝑃, 𝑔, 𝛼) > 𝐿(𝑃, 𝑓1 , 𝛼) + 𝐿(𝑃, 𝑓2 , 𝛼).
𝑎

𝑏 𝑏 𝑏
This implies ∫𝑎 𝑓1 𝑑𝛼 + ∫𝑎 𝑓2 𝑑𝛼 < ∫𝑎 𝑔 𝑑𝛼 + 𝜖.

𝑏 𝑏 𝑏
Since 𝜖 is arbitrary, we must have ∫𝑎 𝑓1 𝑑𝛼 + ∫𝑎 𝑓2 𝑑𝛼 ≤ ∫𝑎 𝑔 𝑑𝛼.

If we replace 𝑓1 and 𝑓2 by −𝑓1 and −𝑓2 , the inequality is reversed.

𝑏 𝑏 𝑏
i.e. we get ∫𝑎 𝑓1 𝑑𝛼 + ∫𝑎 𝑓2 𝑑𝛼 ≥ ∫𝑎 𝑔 𝑑𝛼.

𝑏 𝑏 𝑏
Thus we have ∫𝑎 (𝑓1 + 𝑓2 ) 𝑑𝛼 = ∫𝑎 𝑓1 𝑑𝛼 + ∫𝑎 𝑓2 𝑑𝛼 .

29
𝑏
Theorem 2.4.2: If 𝑓1 ∈ ℜ(𝛼), 𝑓2 ∈ ℜ(𝛼) and 𝑓1 (𝑥) ≤ 𝑓2 (𝑥) on [𝑎, 𝑏], then ∫𝑎 𝑓1 𝑑𝛼 ≤

𝑏
∫𝑎 𝑓2 𝑑𝛼 .

Proof: For any partition 𝑃 of [𝑎, 𝑏], clearly 𝐿(𝑃, 𝑓1 , 𝛼) ≤ 𝑈(𝑃, 𝑓1 , 𝛼) ≤ 𝑈(𝑃, 𝑓2 , 𝛼),

⇒ sup {𝐿(𝑃, 𝑓1 , 𝛼) − 𝑈(𝑃, 𝑓2 , 𝛼)} ≤ 0


𝑃

⇒ sup 𝐿(𝑃, 𝑓1 , 𝛼) − inf 𝑈(𝑃, 𝑓2 , 𝛼)} ≤ 0


𝑃 𝑃

𝑏 𝑏
⇒ ∫𝑎 𝑓1 𝑑𝛼 − ∫𝑎 𝑓2 𝑑𝛼 ≤ 0

𝑏 𝑏
⇒ ∫𝑎 𝑓1 𝑑𝛼 ≤ ∫𝑎 𝑓2 𝑑𝛼 .

Theorem 2.4.3: If 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏], and if 𝑎 < 𝑐 < 𝑏 then 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑐] and 𝑓 ∈

ℜ(𝛼) on [𝑐, 𝑏].

𝑏 𝑐 𝑏
Further ∫𝑎 𝑓 𝑑𝛼 = ∫𝑎 𝑓 𝑑𝛼 + ∫𝑐 𝑓 𝑑𝛼 .

Proof: If 𝑐 ∈ (𝑎, 𝑏), it is clear that 𝑓 is bounded on [𝑎, 𝑏] if and only if it is bounded on [𝑎, 𝑐]

and on [𝑐, 𝑏]. If 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏]. For every 𝜖 > 0, there exists a partition 𝑃 such that

𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖. Let 𝑃∗ = 𝑃 ∪ {𝑐}. Then 𝑃∗ is also a partition of [𝑎, 𝑏] and

𝑈(𝑃∗ , 𝑓, 𝛼) − 𝐿(𝑃∗ , 𝑓, 𝛼) ≤ 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖.

Let us break the partition 𝑃 ∗ into two partitions 𝑃1 on [𝑎, 𝑐] and 𝑃2 on [𝑐, 𝑏]. Then we get

𝑈(𝑃1 , 𝑓, 𝛼) − 𝐿(𝑃1 , 𝑓, 𝛼) + 𝑈(𝑃2 , 𝑓, 𝛼) − 𝐿(𝑃2 , 𝑓, 𝛼)

= 𝑈(𝑃1 , 𝑓, 𝛼) + 𝑈(𝑃2 , 𝑓, 𝛼) − 𝐿(𝑃1 , 𝑓, 𝛼) − 𝐿(𝑃2 , 𝑓, 𝛼)

= 𝑈(𝑃∗ , 𝑓, 𝛼) − 𝐿(𝑃∗ , 𝑓, 𝛼) < 𝜖.

Since each of 𝑈(𝑃1 , 𝑓, 𝛼) − 𝐿(𝑃1 , 𝑓, 𝛼) and 𝑈(𝑃2 , 𝑓, 𝛼) − 𝐿(𝑃2 , 𝑓, 𝛼) is nonnegative, each of

these is less than 𝜖, therefore 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑐] and 𝑓 ∈ ℜ(𝛼)on [𝑐, 𝑏].

Next 𝑈(𝑃∗ , 𝑓, 𝛼) = 𝑈(𝑃1 , 𝑓, 𝛼) + 𝑈(𝑃2 , 𝑓, 𝛼)

⇒ inf 𝑈(𝑃∗ , 𝑓, 𝛼) = inf 𝑈(𝑃1 , 𝑓, 𝛼) + inf 𝑈(𝑃2 , 𝑓, 𝛼)


30
𝑏 𝑐 𝑏
⇒ ∫𝑎 𝑓 𝑑𝛼 = ∫𝑎 𝑓 𝑑𝛼 + ∫𝑐 𝑓 𝑑𝛼 .

Theorem 2.4.4: If 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and if |𝑓(𝑥)| < 𝑀 for all 𝑥 ∈ [𝑎, 𝑏], then

𝑏
|∫𝑎 𝑓 𝑑𝛼 | ≤ 𝑀(𝛼(𝑏) − 𝛼(𝑎)).

Proof: Since |𝑓(𝑥)| < 𝑀 , i.e. −𝑀 < 𝑓(𝑥) < 𝑀 and so, as 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] , we have
𝑏
𝐿(𝑃, 𝑓, 𝛼) ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼) ,
𝑏
⇒ ∑𝑛𝑖=1 mi Δ𝛼𝑖 ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ ∑𝑛𝑖=1 Mi Δ𝛼𝑖 ,
𝑏
 −𝑀 (𝛼(𝑏) − 𝛼(𝑎)) ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑀(𝛼(𝑏) − 𝛼(𝑎)),
𝑏
i.e. |∫𝑎 𝑓 𝑑𝛼 | ≤ 𝑀(𝛼(𝑏) − 𝛼(𝑎)).

Theorem 2.4.5: If 𝑓 ∈ ℜ(𝛼1 ) on [𝑎, 𝑏] and 𝑓 ∈ ℜ(𝛼2 ) on [𝑎, 𝑏] then𝑓 ∈ ℜ(𝛼1 + 𝛼2 ) on

𝑏 𝑏 𝑏
[𝑎, 𝑏] and ∫𝑎 𝑓 𝑑(𝛼1 + 𝛼2 ) = ∫𝑎 𝑓 𝑑𝛼1 + ∫𝑎 𝑓 𝑑𝛼2 .

Proof: Since Δ(𝛼1 + 𝛼2 )𝑖 = (𝛼1 + 𝛼2 )(𝑥𝑖 ) − (𝛼1 + 𝛼2 )(𝑥𝑖−1 )

= 𝛼1 (𝑥𝑖 ) − 𝛼1 (𝑥𝑖−1 ) + α2 (xi ) − α2 (xi−1 )

= (Δα1 )i + (Δα2 )i ,
𝑛

𝑈(𝑃, 𝑓, 𝛼1 + 𝛼2 ) = ∑ 𝑀𝑖 Δ(𝛼1 + 𝛼2 )𝑖
𝑖=1

= ∑𝑛𝑖=1 𝑀𝑖 ((Δα1 )i + (Δα2 )i )

= ∑𝑛𝑖=1 𝑀𝑖 (Δα1 )i + ∑ni=1 𝑀𝑖 (Δα2 )i


= 𝑈(𝑃, 𝑓, 𝛼1 ) + 𝑈(𝑃, 𝑓, 𝛼2 ).
Since 𝑓 ∈ ℜ(𝛼1 ) and 𝑓 ∈ ℜ(𝛼2 ) on [𝑎, 𝑏], for every 𝜖 > 0 , there exist partitions 𝑃1 and 𝑃2 of
[𝑎, 𝑏] such that
𝜖
𝑈(𝑃1 , 𝑓, 𝛼1 ) − 𝐿(𝑃1 , 𝑓, 𝛼1 ) < 2
𝜖
𝑈(𝑃2 , 𝑓, 𝛼2 ) − 𝐿(𝑃2 , 𝑓, 𝛼2 ) < 2
.
Let 𝑃 = 𝑃1 ∪ 𝑃2 . Then we have
𝑈(𝑃, 𝑓, 𝛼1 ) ≤ 𝑈(𝑃1 , 𝑓, 𝛼1 ) and 𝐿(𝑃, 𝑓, 𝛼1 ) ≥ 𝐿(𝑃1 , 𝑓, 𝛼1 ).
𝜖
Therefore, 𝑈(𝑃, 𝑓, 𝛼1 ) − 𝐿(𝑃, 𝑓, 𝛼1 ) ≤ 𝑈(𝑃1 , 𝑓, 𝛼1 ) - 𝐿(𝑃1 , 𝑓, 𝛼1 ) < 2
.

31
𝜖
Similarly, (𝑃, 𝑓, 𝛼2 ) − 𝐿(𝑃, 𝑓, 𝛼2 ) < 2
.
𝜖
Hence 𝑈(𝑃, 𝑓, 𝛼1 + 𝛼2 ) − 𝐿(𝑃, 𝑓, 𝛼1 + 𝛼2 ) < 2
+ 2𝜖 = 𝜖.
By theorem 1.4.1., we have 𝑓 ∈ ℜ(𝛼1 + 𝛼2 ) on [𝑎, 𝑏].
𝑏
Further ∫𝑎 𝑓 𝑑(𝛼1 + 𝛼2 ) = inf 𝑈(𝑃, 𝑓, 𝛼1 + 𝛼2 )
P

= inf(𝑈(𝑃, 𝑓, 𝛼1 ) + 𝑈(𝑃, 𝑓, 𝛼2 ))
P

≥ inf 𝑈(𝑃, 𝑓, 𝛼1 ) + inf 𝑈(𝑃, 𝑓, 𝛼2 )


P P

𝑏 𝑏
≥ ∫𝑎 𝑓 𝑑𝛼1 + ∫𝑎 𝑓 𝑑𝛼2 (2.4.8)

Next to prove the reverse inequality

𝑏
Consider ∫𝑎 𝑓 𝑑(𝛼1 + 𝛼2 ) = sup 𝐿(𝑃, 𝑓, 𝛼1 + 𝛼2 )
P

= sup(𝐿(𝑃, 𝑓, 𝛼1 ) + 𝐿(𝑃, 𝑓, 𝛼2 ))
P

≤ sup 𝐿(𝑃, 𝑓, 𝛼1 ) + sup 𝐿(𝑃, 𝑓, 𝛼2 )


P P

𝑏 𝑏
≤ ∫𝑎 𝑓 𝑑𝛼1 + ∫𝑎 𝑓 𝑑𝛼2 (2.4.9)

Combining (2.2.18) and (2.2.19), we have

𝑏 𝑏 𝑏
∫𝑎 𝑓 𝑑(𝛼1 + 𝛼2 ) = ∫𝑎 𝑓 𝑑𝛼1 + ∫𝑎 𝑓 𝑑𝛼2 .

This completes the proof of the theorem.

Theorem 2.4.6: If 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and c is any constant then show that 𝑐𝑓 ∈ ℜ(𝛼) on

𝑏 𝑏
[𝑎, 𝑏] and ∫𝑎 (𝑐𝑓)𝑑𝛼 = 𝑐 ∫𝑎 𝑓 𝑑𝛼 .

Proof: If 𝑐 = 0, then there is nothing to prove. Let 𝑐 > 0. Then


𝑛
𝑈(𝑃, 𝑐𝑓, 𝛼) = ∑ sup (𝑐𝑓)(𝑥)Δ𝛼𝑖
𝑖=1 x∈[xi−1 ,xi ]

= 𝑐 ∑𝑛𝑖=1 sup 𝑓(𝑥)Δ𝛼𝑖


x∈[xi−1 ,xi ]

= 𝑐 𝑈(𝑃, 𝑓, 𝛼),
32
Similarly, 𝐿(𝑃, 𝑐𝑓, 𝛼) = 𝑐 𝐿(𝑃, 𝑓, 𝛼) and −𝐿(𝑃, 𝑐𝑓, 𝛼) = − 𝑐 𝐿(𝑃, 𝑓, 𝛼).

Therefore 𝑈(𝑃, 𝑐𝑓, 𝛼) − 𝐿(𝑃, 𝑐𝑓, 𝛼) = 𝑐 ( 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼)).

Since 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏], given a 𝜖 > 0, there exists a partition 𝑃 of [𝑎, 𝑏] such that

𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖𝑐 , and therefore 𝑈(𝑃, 𝑐𝑓, 𝛼) − 𝐿(𝑃, 𝑐𝑓, 𝛼) < 𝜖. Hence 𝑐𝑓 ∈ ℜ(𝛼) on

[𝑎, 𝑏].

𝑏 𝑏
Therefore ∫𝑎 (𝑐𝑓)𝑑𝛼 = inf 𝑈(𝑃, 𝑐𝑓, 𝛼) = inf(𝑐 𝑈(𝑃, 𝑓, 𝛼)) = 𝑐 ∫𝑎 𝑓 𝑑𝛼 .

𝑏 𝑏
Thus for any 𝑐 > 0, ∫𝑎 (𝑐𝑓)𝑑𝛼 = 𝑐 ∫𝑎 𝑓 𝑑𝛼 holds . (2.3.11)

If 𝑐 < 0, put 𝑐 = −𝑝, where 𝑝 is positive. Therefore from the above, 𝑝𝑓 ∈ ℜ(𝛼) and

𝑏 𝑏
∫𝑎 (𝑝𝑓)𝑑𝛼 = 𝑝 ∫𝑎 𝑓 𝑑𝛼 .

From the previous result (a), we have 𝑐𝑓 = (−𝑝)𝑓 = −(𝑝𝑓) ∈ ℜ(𝛼)

𝑏 𝑏 𝑏
and ∫𝑎 (𝑐𝑓)𝑑𝛼 = ∫𝑎 (−𝑝𝑓)𝑑𝛼 = ∫𝑎 −(𝑝𝑓)𝑑𝛼

𝑏
= − ∫𝑎 (𝑝𝑓)𝑑𝛼 , follows from part (a),

𝑏
= −𝑝 ∫𝑎 𝑓𝑑𝛼 , follows from (2.3.11),

𝑏
= 𝑐 ∫𝑎 𝑓 𝑑𝛼 , holds for any 𝑐 < 0.

2.5 Summary

1. If 𝑓 is R-S integrable then – 𝑓 , 𝑐𝑓, |𝑓| and 𝑓 2 are all R-S integrable.And

𝑏 𝑏 𝑏 𝑏
∫𝑎 (−𝑓)𝑑𝛼 = − ∫𝑎 𝑓 𝑑𝛼 , ∫𝑎 (𝑐𝑓)𝑑𝛼 = 𝑐 ∫𝑎 𝑓 𝑑𝛼 ,

𝑏 𝑏 𝑏
∫𝑎 (𝑓1 + 𝑓2 )𝑑𝛼 = ∫𝑎 𝑓1 𝑑𝛼 + ∫𝑎 𝑓2 𝑑𝛼 .

2. A continuous function 𝑓 on [𝑎, 𝑏] is Riemann-Steiltje’s integrable on [𝑎, 𝑏].

3. Let 𝑓 be monotonic and 𝛼 be monotonic and continuous on [𝑎, 𝑏]. Then 𝑓 ∈ ℜ(𝛼).

33
4. Suppose 𝑓 is bounded on [𝑎, 𝑏] , 𝑓 has only finitely many discontinuities on [𝑎, 𝑏],

and 𝛼 is continuous at every point at which 𝑓 is discontinuous. Then 𝑓 ∈ ℜ(𝛼) on

[𝑎, 𝑏].

5. Let 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and 𝑚 ≤ 𝑓(𝑥) ≤ 𝑀, ∀ 𝑥 ∈ [𝑎, 𝑏]. Let 𝜙 be a continuous

function on [𝑚, 𝑀] and put ℎ(𝑥) = 𝜙 (𝑓(𝑥)) ∀ 𝑥 ∈ [𝑎, 𝑏] then ℎ ∈ ℜ(𝛼) on [𝑎, 𝑏].

6. Let 𝑓 be a bounded real function on [𝑎, 𝑏] and let 𝛼 be a monotonically increasing on

[𝑎, 𝑏] such that 𝛼 ′ is Riemann integrable on [𝑎, 𝑏]. Then 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] if and

𝑏
only if 𝑓𝛼 ′ is Riemann integrable function on [𝑎, 𝑏] and ∫𝑎 𝑓 𝑑𝑥 =

𝑏
∫𝑎 𝑓(𝑥)𝛼 ′ (𝑥) 𝑑𝑥.

7. Let ϕ be strictly increasing continuous function from [a, b] onto [c, d]. Suppose α is

monotonically increasing function on [c, d] and f ∈ ℜ(α) on [c, d]. If β and g are

defined on [a, b] by g(x) = f(ϕ(x)) and β(y) = α(ϕ(y)) on [a, b]. Then g ∈ ℜ(β)

b d
and ∫a g dβ = ∫c f dα.

2.6 Keywords

Riemann integrable, strictly increasing, bounded, monotonic, continuous, discontinuities

2.7 Exercises

1. Prove that the sum, product, scalar multiple of R-S integrable functions are R-S integrable.

2. Prove that a monotonic function is R-S integrable with respect to a monotonic continuous
function.

3. Prove that a bounded function which has only finitely many discontinuities is R-S integrable.

4. Show that a continuous function of an R-S integrable function is R-S integrable.

34
2.8 References

1. Walter Rudin, Principles of Mathematical Analysis,McGraw-Hill 3rd edition, 1976.


2. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing,1974.

35
UNIT 3
INTEGRATION AND DIFFERENTIATION

3.1 Main Objectives

3.2 Introduction

3.3 Mean Value Theorems for Riemann- Stieltjes Integrals

3.4 Integration and differentiation

3.5 Summary

3.6 Key Words

3.7 Exercises

3.8 References

36
UNIT 3
INTEGRATION AND DIFFERENTIATION

3.1 Main Objectives


The objective of this unit is to study,

 The mean value theorems and their use of estimating integrals.


 Integration and differentiation.

3.2 Introduction

Riemann- Stieltjes integrals occur in a wide variety of problems but the explicit value of the

integrals are obtained in very few cases. In this unit we shall study the mean value theorems

which are useful in obtaining estimate value of the R-S integrals. Also we shall show that

integration and differentiation are inverse operations in a certain sense.

3.3 Mean Value Theorems for Riemann- Stieltjes Integrals


Theorem 3.3.1: If 𝑓, 𝑔 ∈ ℜ(𝛼) on [𝑎, 𝑏], then 𝑓𝑔 ∈ ℜ(𝛼) on [𝑎, 𝑏].

Proof: Define 𝜙(𝑡) = 𝑡 2 on ℝ. And let ℎ(𝑥) = 𝜙(𝑓(𝑥)) = 𝑓 2 (𝑥).

It is clear that 𝜙 is continuous on ℝ.

By previous theorem, 𝑓 2 = ℎ ∈ ℜ(𝛼).

Thus we have 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏]  𝑓 2 ∈ ℜ(𝛼) on [𝑎, 𝑏].

1
Since 𝑓, 𝑔 ∈ ℜ(𝛼) on [𝑎, 𝑏] , (𝑓 + 𝑔)2 and (𝑓 − 𝑔)2 and hence 𝑓𝑔 = 4 {(𝑓 + 𝑔)2 − (𝑓 − 𝑔)2 }

are in ℜ(𝛼) on [𝑎, 𝑏].

37
𝑏 𝑏
Theorem 3.3.2: If 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏], then |𝑓| ∈ ℜ(𝛼) on [𝑎, 𝑏] and |∫𝑎 𝑓 𝑑𝛼 | ≤ ∫𝑎 |𝑓| 𝑑𝛼.

Proof: Let 𝜙(𝑡) = |𝑡| ∀ 𝑡 ∈ ℝ and ℎ(𝑥) = 𝜙(𝑓(𝑥)) = |𝑓(𝑥)|.

𝑏
By theorem, |𝑓| ∈ ℜ(𝛼) on [𝑎, 𝑏]. Choose 𝑐 = ±1 such that 𝑐 ∫𝑎 𝑓 𝑑𝛼 ≥ 0.

𝑏 𝑏 𝑏
Then |∫𝑎 𝑓 𝑑𝛼 | = 𝑐 ∫𝑎 𝑓 𝑑𝛼 = ∫𝑎 𝑐 𝑓 𝑑𝛼. But 𝑐𝑓 ≤ |𝑓|,

𝑏 𝑏 𝑏
therefore |∫𝑎 𝑓 𝑑𝛼 | = ∫𝑎 𝑐 𝑓 𝑑𝛼 ≤ ∫𝑎 |𝑓|𝑑𝛼 .

The following theorem illustrates when Riemann-Stieltjes integral reduces to Riemann integral.

Theorem 3.3.3: Let 𝑓 be a bounded real function on [𝑎, 𝑏] and let 𝛼 be monotonically

increasing on [𝑎, 𝑏] such that 𝛼 ′ is Riemann integrable on [𝑎, 𝑏]. Then 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] if

𝑏 𝑏
and only if 𝑓𝛼 ′ is Riemann integrable function on [𝑎, 𝑏] and ∫𝑎 𝑓 𝑑𝑥 = ∫𝑎 𝑓(𝑥)𝛼 ′ (𝑥) 𝑑𝑥.

Proof: Given that 𝛼 ′ is Riemann integrable on [𝑎, 𝑏].

Then for every 𝜖 > 0, there exists a partition 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} of [𝑎, 𝑏]

such that 𝑈(𝑃, 𝛼 ′ ) − 𝐿(𝑃, 𝛼 ′ ) < 𝜖 (1)

Since 𝛼 is differentiable on [𝑎, 𝑏] , it is differentiable on the subinterval [𝑥𝑖−1 , 𝑥𝑖 ] of [𝑎, 𝑏].

Then by the mean value theorem, we have points 𝑠𝑖 𝜖 [𝑥𝑖−1 , 𝑥𝑖 ]

such that Δ𝛼𝑖 = 𝛼(𝑥𝑖 ) − 𝛼(𝑥𝑖−1 ) = 𝛼 ′ (𝑠𝑖 )Δ𝑥𝑖 , for 𝑖 = 1, … , 𝑛. If 𝑡𝑖 , 𝑠𝑖 ∈ [𝑥𝑖−1 , 𝑥𝑖 ], then

|𝛼 ′ (𝑠𝑖 ) − 𝛼 ′ (𝑡𝑖 )| ≤ |𝑀𝑖 − 𝑚𝑖 | = 𝑀𝑖 − 𝑚𝑖 ,

where 𝑀𝑖 = sup α′ (𝑡), 𝑚𝑖 = inf 𝛼 ′ (𝑡), 𝑡 ∈ [𝑥𝑖−1 , 𝑥𝑖 ].

 |𝛼 ′ (𝑠𝑖 ) − 𝛼 ′ (𝑡𝑖 )| Δ𝑥𝑖 ≤ Δ𝑥𝑖 ( 𝑀𝑖 − 𝑚𝑖 ) .

Putting 𝑖 = 1,2, … , 𝑛 and adding all the n inequalities ,

we obtain

∑𝑛𝑖=1|𝛼 ′ (𝑠𝑖 ) − 𝛼 ′ (𝑡𝑖 )| Δ𝑥𝑖 ≤ ∑𝑛𝑖=1(𝑀𝑖 − 𝑚𝑖 )Δ𝑥𝑖 < 𝜖 (2)

38
Put 𝑀 = sup |𝑓(𝑥)|. Since ∑𝑛𝑖=1 𝑓(𝑠𝑖 )Δ𝛼𝑖 = ∑𝑛𝑖=1 𝑓(𝑠𝑖 )𝛼 ′ (𝑡𝑖 )Δ𝑥𝑖 ,
(𝑥𝑖−1 ≤𝑥≤𝑥𝑖 )

it follows from (3.3.2) that | ∑𝑛𝑖=1 𝑓(𝑠𝑖 ) Δ𝛼𝑖 − ∑𝑛𝑖=1 𝑓(𝑠𝑖 )𝛼 ′ (𝑡𝑖 )Δ𝑥𝑖 | ≤ 𝑀 𝜖. (3)

In ∑𝑛𝑖=1 𝑓(𝑠𝑖 ) Δ𝛼𝑖 ≤ 𝑈(𝑃, 𝑓𝛼 ′ ) + 𝑀𝜖, for all points 𝑠𝑖 ∈ [𝑥𝑖=1 − 𝑥𝑖 ],

so that

𝑈(𝑃, 𝑓, 𝛼) ≤ 𝑈(𝑃, 𝑓𝛼 ′ ) + 𝑀 𝜖.

If we proceed the same way, we have from (3) , that

𝑈(𝑃, 𝑓𝛼 ′ ) ≤ 𝑈(𝑃, 𝑓, 𝛼) + 𝑀 𝜖. Thus we have

𝑈(𝑃, 𝑓𝛼 ′ ) − 𝑈(𝑃, 𝑓, 𝛼) ≤ 𝑀 𝜖. (4)

Equation (1) remains true even if replace 𝑃 by any refinement of 𝑃.

Hence (4) also remains true for any refinement of 𝑃.

Thus taking infimum over all refinements of,

we have

−𝑏 −𝑏
|∫ 𝑓 𝑑𝛼 − ∫ 𝑓(𝑥)𝛼 ′ (𝑥)𝑑𝑥| ≤ 𝑀 𝜖.
𝑎 𝑎

−𝑏 −𝑏
Since 𝜖 is arbitrary, we have ∫𝑎 𝑓 𝑑𝛼 = ∫𝑎 𝑓(𝑥)𝛼 ′ (𝑥)𝑑𝑥 . on similar lines.

𝑏 𝑏
we have ∫−𝑎 𝑓 𝑑𝛼 = ∫−𝑎 𝑓(𝑥)𝛼 ′ (𝑥)𝑑𝑥.

𝑏
Thus we have 𝑓𝛼 ′ is Riemann integrable if and only if 𝑓 ∈ ℜ(𝛼) and ∫𝑎 𝑓 𝑑𝛼 =

𝑏
∫𝑎 𝑓(𝑥)𝛼 ′ (𝑥)𝑑𝑥.

Theorem 3.3.4 (Change of variable): Let 𝜙 be strictly increasing continuous function from

[a, b] onto [c, d]. Suppose α is monotonically increasing function on [c, d] and f ∈ ℜ(α) on

39
[c, d]. If β and g are defined on [a, b] by g(x) = f(𝜙(x)) and β(y) = α(𝜙(y)) on [a, b]. Then

b d
g ∈ ℜ(β) and ∫a g dβ = ∫c f dα.

Proof: For each partition P = {c = y0 , y1 , … , yn = d} of [c, d] there corresponds a partition

Q = {a = x0 , x1 , … , xn = b}of [a, b] such that yi = 𝜙(xi ).

All partitions of [a, b] are obtained in this way. Since the values of g on [xi−1 , xi ] are same as

the values of f on [yi−1 , yi ], we have


𝑛 𝑛

𝑈(𝑄, 𝑔, 𝛽) = ∑ sup g(t) Δ𝛽𝑖 = ∑ sup f(𝜙(t)) (β(yi ) − β(yi−1 ))


𝑡∈[𝑥𝑖−1 ,𝑥𝑖 ] 𝑡∈[𝑥𝑖−1 ,𝑥𝑖 ]
𝑖=1 𝑖=1

= ∑ sup f(𝜙(t)) (α(𝜙(yi )) − α(𝜙(yi−1 )))


𝑠=𝜙(𝑡)∈[𝜙(𝑦𝑖−1 ) ,𝜙(𝑦𝑖 )]
𝑖=1

= 𝑈(𝑃, 𝑓, 𝛼) ,

Similarly 𝐿(𝑄, 𝑔, 𝛽) = 𝐿(𝑃, 𝑓, 𝛼).

Since 𝑓 ∈ ℜ(𝛼), given any 𝜖 > 0 , we can choose 𝑃 such that 𝑈(𝑃, 𝑓, 𝛼) − 𝐿(𝑃, 𝑓, 𝛼) < 𝜖 .

Therefore we have 𝑈(𝑄, 𝑔, 𝛽) − 𝐿(𝑄, 𝑔, 𝛽) < 𝜖.

Hence 𝑔 ∈ ℜ(𝛽) on [𝑎, 𝑏]. Also we have 𝑈(𝑄, 𝑔, 𝛽) = 𝑈(𝑃, 𝑓, 𝛼).

𝑏 𝑑
Therefore inf 𝑈(𝑄, 𝑔, 𝛽) = inf 𝑈(𝑃, 𝑓, 𝛼) or ∫𝑎 𝑔 𝑑𝛽 = ∫𝑐 𝑓 𝑑𝛼.
𝑃 𝑃

Note: Take 𝛼(𝑥) = 𝑥 in the above theorem. Then 𝛽(𝑦) = 𝛼(𝜙(𝑦)) = 𝜙(𝑦).

And so 𝛽 = 𝜙. If we assume 𝜙 ′ is Riemann integrable on [𝑎, 𝑏], then applying theorem (3.3.3)

𝑏 𝑑 𝑏 𝑏
to LHS of ∫𝑎 𝑔 𝑑𝛽 = ∫𝑐 𝑓 𝑑𝛼., we obtain ∫𝑎 𝑔(𝑦)𝛽 ′ (𝑦) 𝑑𝑦 = ∫𝑎 𝑓 𝑑𝛼.

𝑑 𝑏
But 𝛽 ′ = 𝜙 ′ , we have ∫𝑐 𝑓(𝑥) 𝑑𝑥 = ∫𝑎 𝑓(𝜙(𝑦))𝜙 ′ (𝑦).

40
Theorem 3.3.5: (First Mean Value theorem for Riemann-Stieltjes Integrals )

Let α be monotonically increasing and let 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏]. Let 𝑀 and 𝑚 denote the sup and

inf of the set {𝑓(𝑥): 𝑥 ∈ [𝑎, 𝑏]}. Then there exists a real number 𝜆, with 𝑎 ≤ 𝜆 ≤ 𝑏 , such that
𝑐
∫𝑎 𝑓(𝑥)𝑑𝛼 = 𝜆 [𝛼(𝑏) − 𝛼(𝑎)]. Further if 𝑓 is continuous on [𝑎, 𝑏] , then there exists a 𝑥0 ∈

[𝑎, 𝑏] such that 𝜆 = 𝑓(𝑥0 ).

Proof: Since 𝑀 = sup 𝑓(𝑥) and 𝑚= inf 𝑓(𝑥) , it follows that


𝑥∈[𝑎,𝑏] 𝑥∈[𝑎,𝑏]

𝑚 ≤ 𝑓(𝑥) ≤ 𝑀. Multiplying the inequality throughout by 𝛼(𝑏) − 𝛼(𝑎),

we get

𝑚 (𝛼(𝑏) − 𝛼(𝑎)) ≤ 𝑓(𝑥)(𝛼(𝑏) − 𝛼(𝑎)) ≤ 𝑀 (𝛼(𝑏) − 𝛼(𝑎))

Let 𝑃 = {𝑎 = 𝑥0 , 𝑥1 , … , 𝑥𝑛 = 𝑏} be a partition of [𝑎, 𝑏], then 𝑚 ≤ 𝑚𝑖 ≤ 𝑀𝑖 ≤ 𝑀,

where 𝑀𝑖 = sup 𝑓(𝑥) and 𝑚𝑖 = inf 𝑓(𝑥).


𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ] 𝑥∈[𝑥𝑖−1 ,𝑥𝑖 ]

Multiplying throughout by Δ𝛼𝑖 , we get

𝑚 Δ𝛼𝑖 ≤ 𝑚𝑖 Δ𝛼𝑖 ≤ 𝑀𝑖 Δ𝛼𝑖 ≤ 𝑀 Δ𝛼𝑖 , putting 𝑖 = 1,2, … , 𝑛, and adding all the inequalities,

we get

𝑚 (𝛼(𝑏) − 𝛼(𝑎)) ≤ 𝐿(𝑃, 𝑓, 𝛼) ≤ 𝑈(𝑃, 𝑓, 𝛼) ≤ 𝑀 (𝛼(𝑏) − 𝛼(𝑎)), but

𝑏 −𝑏
𝐿(𝑃, 𝑓, 𝛼) ≤ ∫−𝑎 𝑓 𝑑𝛼 ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼), the above inequality becomes

𝑏 −𝑏
𝑚(𝛼(𝑏) − 𝛼(𝑎)) ≤ 𝐿(𝑃, 𝑓, 𝛼) ≤ ∫ 𝑓 𝑑𝛼 ≤ ∫ 𝑓 𝑑𝛼 ≤ 𝑈(𝑃, 𝑓, 𝛼) ≤ 𝑀(𝛼(𝑏) − 𝛼(𝑎)).
−𝑎 𝑎

𝑏 −𝑏
Since 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏], ∫−𝑎 𝑓 𝑑𝛼 = ∫𝑎 𝑓 𝑑𝛼 ,

𝑏
we have 𝑚(𝛼(𝑏) − 𝛼(𝑎)) ≤ ∫𝑎 𝑓 𝑑𝛼 ≤ 𝑀(𝛼(𝑏) − 𝛼(𝑎)).

𝑏
Therefore there is a number 𝜆, 𝑚 ≤ 𝜆 ≤ 𝑀, such that ∫𝑎 𝑓 𝑑𝛼 = 𝜆(𝛼(𝑏) − 𝛼(𝑎)).

Since 𝑓 is continuous , 𝑓 assumes every value between its bounds 𝑚 and 𝑀.

41
Therefore there exists a number 𝑥0 ∈ [𝑎, 𝑏] such that 𝑓(𝑥0 ) = 𝜆.

Hence we have,

𝑏
∫𝑎 𝑓 𝑑𝛼 = 𝑓(𝑥0 )(𝛼(𝑏) − 𝛼(𝑎)). This completes the proof.

It may not always be possible to choose 𝑥0 such that 𝑎 < 𝑥0 < 𝑏.

0 𝑖𝑓 𝑥 = 𝑎
For example, consider a function 𝑓 continuous on [𝑎, 𝑏] and(𝑥) = { , then
1 𝑤ℎ𝑒𝑛 𝑥 > 𝑎

Let 𝑃 = {𝑎 = 𝑥0 < 𝑥1 < ⋯ < 𝑥𝑛 = 𝑏} 𝑏𝑒 𝑎𝑛𝑦 𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 𝑜𝑓 [𝑎, 𝑏].

Then since Δ𝛼1 = 𝛼(𝑥1 ) − 𝛼(𝑥0 ) = 1 − 0 = 1, Δ𝛼2 = 0, Δ𝛼3 = 0, … , Δ𝛼𝑛 = 0, 𝑤𝑒 ℎ𝑎𝑣𝑒


n

S(P, f, α) = ∑ f(xi )Δαi = 𝑓(𝑥1 ) = 𝑓(𝑎).


i=1

 lim S(P, f, α) = 𝑓(𝑎) = 𝑓(𝑎)(𝛼(𝑏) − 𝛼(𝑎)).


‖P‖→0

b
i.e. there does not exist c ∈ (a, b) such that ∫a f dα = f(a)(α(b) − α(a)).

Theorem 3.3.6: (Second Mean Value theorem for Riemann- Stieltjes Integrals)

Let 𝛼 be continuous and that 𝑓 be monotonically increasing on [𝑎, 𝑏].

𝑏
Then there exists a point 𝑥0 in [𝑎, 𝑏] such that ∫𝑎 𝑓(𝑥)𝑑𝛼 = 𝑓(𝑎) [𝛼(𝑥0 ) − 𝛼(𝑎)] +

𝑓(𝑏)[𝛼(𝑏) − 𝛼(𝑥0 )].

Proof: We have

𝑏 𝑏
∫𝑎 𝑓 𝑑𝛼 = 𝑓(𝑏)𝛼(𝑏) − 𝑓(𝑎)𝛼(𝑎) − ∫𝑎 𝛼 𝑑𝑓 .

By the first mean value theorem, there exists a point 𝑥0 ∈ [𝑎, 𝑏] such that

𝑏
∫ 𝛼 𝑑𝑓 = 𝛼(𝑥0 )[𝑓(𝑏) − 𝑓(𝑎)].
𝑎

Hence we have,

42
𝑏
∫ 𝑓 𝑑𝛼 = 𝑓(𝑏)𝛼(𝑏) − 𝑓(𝑎)𝛼(𝑎) − 𝛼(𝑥0 )[𝑓(𝑏) − 𝑓(𝑎)]
𝑎

𝑏
 ∫𝑎 𝑓 𝑑𝛼 = 𝑓(𝑎) [𝛼(𝑥0 ) − 𝛼(𝑎)] + 𝑓(𝑏)[𝛼(𝑏) − 𝛼(𝑥0 )].

This proves the result.

Theorem 3.3.7.(Second mean value theorem for Riemann integrals)


Let 𝑔 be continuous and let 𝑓 be increasing on [𝑎, 𝑏]. Let 𝑓(𝑎 +) ≥ 𝐴 and 𝑓(𝑏 −) ≤ 𝐵, for some
real numbers 𝐴 and 𝐵. Then there exists a point 𝑐 ∈ [𝑎, 𝑏] such that
𝑏 𝑐 𝑏
𝑖) ∫ 𝑓(𝑥)𝑔(𝑥)𝑑𝑥 = 𝐴 ∫ 𝑔(𝑥)𝑑𝑥 + 𝑏 ∫ 𝑔(𝑥) 𝑑𝑥.
𝑎 𝑎 𝑐

In particular if 𝑓(𝑥) ≥ 0 for all 𝑥 ∈ [𝑎, 𝑏], we have


𝑏 𝑏
𝑖𝑖) ∫𝑎 𝑓(𝑥)𝑔(𝑥)𝑑𝑥 = 𝐵 ∫𝑐 𝑔(𝑥)𝑑𝑥, where 𝑐 ∈ [𝑎, 𝑏].
Proof:
𝑥
If 𝛼(𝑥) = ∫𝑎 𝑔(𝑡)𝑑𝑡 , then 𝛼 ′ = 𝑔.
By second mean value theorem for Riemann Stieltjes integrals,
𝑏 𝑐 𝑏
we get ∫𝑎 𝑓(𝑥)𝑔(𝑥)𝑑𝑥 = 𝐴 ∫𝑎 𝑔(𝑥)𝑑𝑥 + 𝐵 ∫𝑐 𝑔(𝑥)𝑑𝑥, where 𝐴 = 𝑓(𝑎) and 𝐵 = 𝑓(𝑏).
This proves the assertion (𝑖).
To prove (𝑖𝑖), If 𝑓(𝑎 +) ≥ 𝐴 and 𝑓(𝑏 −) ≤ 𝐵, set 𝑓(𝑎) = 𝐴 and 𝑓(𝑏) = 𝐵.
Then 𝑓 is still increasing on [𝑎, 𝑏] and changing the value of 𝑓 at a finite number of points does
not affect the value of a Riemann integral.
𝑏 𝑏
Taking 𝐴 = 0, from (𝑖), it follows that ∫𝑎 𝑓(𝑥)𝑔(𝑥)𝑑𝑥 = 𝐵 ∫𝑐 𝑔(𝑥)𝑑𝑥, for some 𝑐 ∈ [𝑎, 𝑏].

3.4 Integration and Differentiation


𝑥
Theorem 3.4.1: Let 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and let 𝑔(𝑥) = ∫𝑎 𝑓(𝑡)𝑑𝑡 ∀ 𝑥 ∈ [𝑎, 𝑡]. Then 𝑔(𝑥) is

continuous on [𝑎, 𝑏]. Further if 𝑓 is continuous at 𝑥0 ∈ [𝑎, 𝑏], then 𝑔 is differentiable at 𝑥0 and

𝑔′ (𝑥0 ) = 𝑓(𝑥0 ).

Proof: Let 𝑥 ∈ [𝑎, 𝑏]. We show that 𝑔 is continuous at 𝑥. Let 𝜖 > 0 be given

43
𝑦 𝑥
and let 𝑎 ≤ 𝑥 ≤ 𝑦 ≤ 𝑏. Then |𝑔(𝑦) − 𝑔(𝑥)| = |∫𝑎 𝑓(𝑡) 𝑑𝑡 − ∫𝑎 𝑓(𝑡) 𝑑𝑡|

𝑥 𝑦 𝑥
= |∫𝑎 𝑓(𝑡)𝑑𝑡 + ∫𝑥 𝑓(𝑡)𝑑𝑡 − ∫𝑎 𝑓(𝑡)𝑑𝑡|

𝑦 𝑦
= |∫𝑥 𝑓(𝑡)𝑑𝑡| ≤ ∫𝑥 𝑓(𝑡)𝑑𝑡 .

Let 𝑀 = max|𝑓(𝑡)|, ∀ 𝑡 ∈ [𝑎, 𝑏].


𝑦 𝑦
Then |𝑔(𝑦) − 𝑔(𝑥)| ≤ ∫𝑥 𝑀 𝑑𝑡 = 𝑀 ∫𝑥 𝑑𝑡 = 𝑀(𝑦 − 𝑥).

𝜖
Choose 𝛿 such that 𝛿 = .
𝑀

Whenever |𝑦 − 𝑥| < 𝛿, we have |𝑔(𝑦) − 𝑔(𝑥)| < 𝑀(𝑦 − 𝑥) < 𝑀𝛿 = 𝜖.

Therefore 𝑔 is continuous at 𝑥. Since 𝑥 is arbitrary, 𝑔 is continuous on [𝑎, 𝑏].

Suppose 𝑓 is continuous at 𝑥0 . To show that 𝑔′ (𝑥0 ) = 𝑓(𝑥0 ).

Let 𝑎 ≤ x0 < 𝑦 ≤ 𝑏.
𝑦 𝑥
𝑔(𝑦)−𝑔(𝑥0 ) ∫ 𝑓(𝑡)𝑑𝑡− ∫𝑎 0 𝑓(𝑡)𝑑𝑡))
Consider | − 𝑓(𝑥0 )| = | 𝑎 − 𝑓(𝑥0 )|
𝑦−𝑥0 𝑦−𝑥0

𝑦 𝑦 𝑥
∫𝑎 0 𝑓(𝑡)𝑑𝑡 + ∫𝑥 𝑓(𝑡)𝑑𝑡− ∫𝑎 0 𝑓(𝑡)𝑑𝑡−𝑓(𝑥0 )(𝑦−𝑥0 )
0
=| |
𝑦−𝑥0

𝑦
∫𝑥 𝑓(𝑡)𝑑𝑡−𝑓(𝑥0 )(𝑦−𝑥0 )
0
=| |
𝑦−𝑥0

𝑦 𝑦
∫𝑥 𝑓(𝑡)𝑑𝑡−𝑓(𝑥0 ) ∫𝑥 𝑑𝑡
0 0
=| |
𝑦−𝑥0

𝑦 [𝑓(𝑡)𝑑𝑡−𝑓(𝑥0 )]
=|∫𝑥 𝑑𝑡|
0 (𝑦−𝑥0 )

𝑦 |𝑓(𝑡)𝑑𝑡−𝑓(𝑥0 )|
≤ ∫𝑥 |𝑦−𝑥0 |
𝑑𝑡.
0

Given 𝑓 is continuous at 𝑥0 . For any 𝜖 > 0 there exists 𝛿 > 0 such that |𝑓(𝑡) − 𝑓(𝑥0 )| < 𝜖

whenever |𝑡 − 𝑥0 | < 𝛿.

𝑔(𝑦)−𝑔(𝑥0 ) 𝑦 𝜖
⟹ | − 𝑓(𝑥0 )| ≤ ∫𝑥 |𝑦−𝑥0 |
𝑑𝑡 = 𝜖 whenever |𝑦 − 𝑥0 | < 𝛿.
𝑦−𝑥0 0

44
𝑔(𝑦)−𝑔(𝑥0 )
Similarly for 𝑎 ≤ 𝑦 < 𝑥0 < 𝑏, | − 𝑓(𝑥0 )| < 𝜖 whenever |𝑦 − 𝑥0 | < 𝛿.
𝑦−𝑥0

𝑔(𝑦)−𝑔(𝑥0 )
Therefore lim = 𝑓(𝑥0 ) or 𝑔′ (𝑥0 ) = f(x0 ).
𝑦→𝑥0 𝑦−𝑥0

The following theorem tells that how to integrate a derivative.

Theorem 3.4.2: (Fundamental theorem of Calculus)

Let 𝑓 be Riemann integrable function on [𝑎, 𝑏] and if there is a differentiable function 𝑔 on [𝑎, 𝑏]

𝑏
such that 𝑔′ = 𝑓, then ∫𝑎 𝑓(𝑥)𝑑𝑥 = 𝑔(𝑏) − 𝑔(𝑎).

Proof: Let P = {a = x0 ≤ x1 ≤ ⋯ ≤ xn } be any partition of [a, b] .

Since g is differentiable on [a, b], it is differentiable on each subinterval[xi−1 , xi ].

By mean value theorem of differential calculus,

g(xi ) − g(xi−1 ) = g ′ (t i )(xi − xi−1 ), t i ∈ [xi−1 , xi ]. Put i = 1, … , n and add to get


n n

∑[g(xi ) − g(xi−1 )] = ∑ g ′ (t i )(xi − xi−1 ), t i ∈ [xi−1 , xi ].


i=1 i=1

 g(b) − g(a) = ∑ni=1 f(t i )Δxi , t i ∈ [xi−1 , xi ]

Since 𝑓 is Riemann integrable on [𝑎, 𝑏], from theorem 1.4.2, we have

𝑏
|𝑔(𝑏) − 𝑔(𝑎) − ∫𝑎 𝑓 (𝑥)𝑑𝑥| < 𝜖, since this holds for every 𝜖 > 0, the result follows.

Theorem 3.4.3. (Integration by Parts): Let 𝐹 and 𝐺 be two differentiable functions on [𝑎, 𝑏]

such that their derivatives 𝐹 ′ = 𝑓 and G′ = 𝑔 are Riemann integrable on [𝑎, 𝑏]. Then

𝑏 𝑏
∫𝑎 𝐹(𝑥)𝑔(𝑥)𝑑𝑥 = 𝐹(𝑏)𝐺(𝑏) − 𝐹(𝑎)𝐺(𝑎) − ∫𝑎 𝑓(𝑥)𝐺(𝑥)𝑑𝑥.

Proof: Let us put ℎ(𝑥) = 𝐹(𝑥)𝐺(𝑥). Since 𝐹 and 𝐺 are differentiable on [𝑎, 𝑏], it follows from

theorem 3.3.1 ℎ is differentiable and ℎ′ = 𝐹 ′ 𝐺 + 𝐹 𝐺 ′ .

Applying previous theorem to ℎ and its derivative ℎ′ , we obtain

45
𝑏
∫ ℎ′ (𝑥) 𝑑𝑥 = ℎ(𝑏) − ℎ(𝑎),
𝑎

𝑏
 ∫𝑎 (𝐹 ′ (𝑥)𝐺(𝑥) + 𝐹(𝑥)𝐺 ′ (𝑥)) 𝑑𝑥 = 𝐹(𝑏)𝐺(𝑏) − 𝐹(𝑎)𝐺(𝑎)

𝑏 𝑏
 ∫𝑎 𝐹(𝑥)𝑔(𝑥) 𝑑𝑥 = 𝐹(𝑏)𝐺(𝑏) − 𝐹(𝑎)𝐺(𝑎) − ∫𝑎 𝑓(𝑥)𝐺(𝑥) 𝑑𝑥.

3.5 Summary

1. Let α be monotonically increasing and let 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏]. Let 𝑀 and 𝑚 denote the sup

and inf of the set {𝑓(𝑥): 𝑥 ∈ [𝑎, 𝑏]}. Then there exists a real number 𝜆, with 𝑎 ≤ 𝜆 ≤ 𝑏 ,
𝑐
such that ∫𝑎 𝑓(𝑥)𝑑𝛼 = 𝜆 [𝛼(𝑏) − 𝛼(𝑎)].

2. Let 𝛼 be continuous and that 𝑓 be monotonically increasing on [𝑎, 𝑏]. Then there exists a

𝑏
point 𝑥0 in [𝑎, 𝑏] such that ∫𝑎 𝑓(𝑥)𝑑𝛼 = 𝑓(𝑎) [𝛼(𝑥0 ) − 𝛼(𝑎)] + 𝑓(𝑏)[𝛼(𝑏) − 𝛼(𝑥0 )].

𝑥
3. Let 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and let 𝑔(𝑥) = ∫𝑎 𝑓(𝑡)𝑑𝑡 ∀ 𝑥 ∈ [𝑎, 𝑡]. Then 𝑔(𝑥) is continuous on

[𝑎, 𝑏]. Further if 𝑓 is continuous at 𝑥0 ∈ [𝑎, 𝑏], then 𝑔 is differentiable at 𝑥0 and 𝑔′ (𝑥0 ) =

𝑓(𝑥0 ).

𝑏 𝑏
4. The formula ∫𝑎 𝐹(𝑥)𝑔(𝑥)𝑑𝑥 = 𝐹(𝑏)𝐺(𝑏) − 𝐹(𝑎)𝐺(𝑎) − ∫𝑎 𝑓(𝑥)𝐺(𝑥)𝑑𝑥 gives the way to

integrate by parts

5. For a Riemann integrable function 𝑓 on [𝑎, 𝑏] and a differentiable function 𝑔 on [𝑎, 𝑏] such

𝑏
that 𝑔′ = 𝑓, we have ∫𝑎 𝑓(𝑥)𝑑𝑥 = 𝑔(𝑏) − 𝑔(𝑎).

46
3.6 Keywords

Differentiable Monotonic,
Continuous Integration,
Mean value theorem Integration by Parts,
Fundamental theorem of Calculus Riemann integrals

3.7 Exercises

1. State and prove first mean value theorem for R-S integrals.

2. State and prove second mean value theorem for R-S integrals.
𝑥
3. If 𝑓 ∈ ℜ(𝛼) on [𝑎, 𝑏] and 𝑔(𝑥) = ∫𝑎 𝑓(𝑡)𝑑𝑡 ∀ 𝑥 ∈ [𝑎, 𝑡] is continuous on [𝑎, 𝑏]. and if 𝑓 is

continuous at 𝑥0 ∈ [𝑎, 𝑏], then prove that 𝑔 is differentiable at 𝑥0 and 𝑔′ (𝑥0 ) = 𝑓(𝑥0 ).

3.8. References

1. Walter Rudin, Principles of Mathematical Analysis,McGraw-Hill 3rd edition, 1976.

2. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing,1974.

47
UNIT 4
INTEGRATION OF VECTOR VALUED FUNCTIONS AND
RECTIFIABLE CURVES

4.1 Main Objectives

4.2 Introduction

4.3 Integration of vector valued functions

4.4 Examples

4.5 Rectifiable curves

4.6 Summary

4.7 Key Words

4.8 Exercises

4.9 References

48
UNIT 4
INTEGRATION OF VECTOR VALUED FUNCTIONS AND
RECTIFIABLE CURVES

4.1 Main Objectives

After going through this unit, the student should be able to,
 Define a function of variation function.
 Understand some basic properties of functions of bounded variation.
𝑏 𝑏
 Prove |∫𝑎 𝑓 𝑑𝛼 | ≤ ∫𝑎 |𝑓| 𝑑𝑣𝛼 under certain conditions.

4.2 Introduction

The concept of bounded variation helps in extending the theories of integration and

differentiation. The definition of Riemann-Steiltje’s integral can be extended for the cases of

monotonically non decreasing functions α on [a, b] by employing bounded variation notion. In

this unit we shall study some properties of functions of bounded variation. Further we shall study

b
under conditions either f or α or both are of bounded variation then the inequality |∫a f dα | ≤

b
∫a |f| dvα holds .

4.3 Integration of vector valued functions

Definition 4.3.1: Let f: [a, b] ⟶ ℝ be any function and let P = {a = x0 , x1 , … , xn } of [a, b].

Then Vab (f) = ∑ni=1 |f(xi ) − f(xi−1 | and

49
Tab (f) = sup Vab (f) are respectively called the variation function and the total variation of f on
P

[a, b]. Tab (f) may be finite or infinite. If Tab (f) is finite then we say that f is of bounded variation

on [a, b]. Equivalently a function f on [a, b] is said to be of bounded variation on [a, b] if there

exists K > 0 such that for every partition P of [a, b] , ∑ni=1 |f(xi ) − f(xi−1 | ≤ K.

Note : 1) A function of bounded variation is always bounded.

Since f is of bounded variation on [a, b] there exists a positive number M such that V(f) ≤ M.

Also |f(x) − f(a)| ≤ V(f) for all x ∈ [a, b].

Then it follows that |f(x)| − |f(a)| ≤ |f(x) − f(a)| ≤ M.

Hence |f(x)| ≤ M + f(a), for all x ∈ [a, b] or f is bounded.

The converse need not be true.

But if the function is monotonic and bounded then it is of bounded variation.

2.) A monotonic and bounded function f on [a, b] is a function of bounded variation on [a, b] and

𝑇𝑎𝑏 (f) = |f(b) − f(a)|.

Proof: For any partition P = {a = x0 , x1 , … , xn = b} of [a, b] , we have


n n

∑|f(xi ) − f(xi−1 )| = ∑(f(xi ) − f(xi−1 )) = f(b) − f(a)


i=1 i=1

n n

Tab (f) = sup ∑|f(xi ) − f(xi−1 )| = sup ∑(f(xi ) − f(xi−1 ))


1 1

= f(b) − f(a), supremum being taken over all partitions of [a, b].

Hence f is a function of bounded variation on [a, b].

This is true for decreasing functions also.

3) Product of two functions of bounded variation functions is also of bounded variation.

50
Proof: Suppose f and g are two functions of bounded variation on [a, b]. Since Tab (f) < ∞ and

Tab (g) < ∞, we have Tab (fg) = sup ∑n1 |(fg)(xi ) − (fg)(xi−1 )|
P

=sup ∑n1|f(xi )g(xi ) − f(xi )g(xi−1 ) + f(xi )g(xi−1 ) − f(xi−1 )g(xi−1 )|


P

=sup ∑n1|f(xi )(g(xi ) − g(xi−1 )) + (f(xi ) − f(xi−1 ))g(xi−1)| (4.3.1)


P

Since f and g are of bounded variation on [a, b], the functions f and g are bounded.

There exist positive numbers M and N such that |f(x)| ≤ M and |g(x)| ≤ N, ∀ x ∈ [a, b].

We have
n n

Tab (fg) ≤ supM ∑(g(xi ) − g(xi−1 )) + sup N ∑(f(xi ) − f(xi−1 ))


1 1

≤ M Tab (g) + N Tab (f) < ∞, 𝑠𝑖𝑛𝑐𝑒 Tab (f) < ∞, and Tab (g) < ∞.

Hence Tab (fg) < ∞ or fg is of bounded variation on [a, b].

This establishes the theorem.

A continuous function need not be a function of bounded variation.

𝜋
𝑥 cos {2𝑥} 𝑖𝑓 𝑥 ≠ 0
For example, let 𝑓(𝑥) = { 0 𝑖𝑓 𝑥 = 0 .

1 1 1 1
Then the function is continuous on [0,1], but for the the partition 𝑃 = {0, 2𝑛 , 2𝑛−1 , … , 3 , 2 , 1},

1 1 1 1 1 1 1
we have ∑2𝑛
𝑘=1|Δ𝑓𝑘 | = + 2𝑛 + 2𝑛−1 + … + 2 + 2 = 1 + 2 + ⋯ + 𝑛.
2𝑛

1
Since the series ∑∞
1 𝑛 diverges, the above sum is not bounded for all 𝑛. Hence the function 𝑓

Is not of bounded variation on [0,1].

Functions of bounded variation as a difference of two monotonic increasing functions.

51
Theorem 4.3.1 (Jordan Theorem): A function of bounded variation 𝑓 on [𝑎, 𝑏] can be

expressed as a difference of two monotonically increasing functions on [𝑎, 𝑏], and conversely.

Proof: Let 𝑓 be a function of bounded variation on [𝑎, 𝑏].

Let 𝑔(𝑥) = 𝑇𝑎𝑥 (𝑓). Then 𝑔(𝑥) is monotonically increasing function on [𝑎, 𝑏].

also

ℎ =𝑔−𝑓 (4.3.2)

for 𝑎 ≤ 𝑥 < 𝑦 < 𝑏 gives that

ℎ(𝑦) − ℎ(𝑥) = 𝑔(𝑦) − 𝑓(𝑦) − 𝑔(𝑥)


𝑦
= 𝑇𝑎 (𝑓) − 𝑇𝑎𝑥 (𝑓) − [𝑓(𝑦) − 𝑓(𝑥)]
𝑦
= 𝑇𝑥 (𝑓) − {𝑓(𝑦) − 𝑓(𝑥)} ≥ 0.

Thus ℎ is also monotonically increasing on [𝑎, 𝑏]. Hence from (4.1.2)

we have 𝑓 = 𝑔 − ℎ, where 𝑔 and ℎ are monotonically increasing on [𝑎, 𝑏].

Conversely, suppose 𝑓 = 𝑔 − ℎ, where 𝑔 and ℎ are monotonically increasing on [𝑎, 𝑏], then

𝑇𝑎𝑏 (𝑓) ≤ |𝑔(𝑏) − 𝑔(𝑎)| + |ℎ(𝑏) − ℎ(𝑎)| < ∞.

Hence 𝑓 is of bounded variation on [𝑎, 𝑏].

Definition 4.3.2: Let f: [a, b] → R be a function of bounded variation on [a, b]. Then the mapping

vf : [a, b] → R defined by vf (x) = Tax (f).

Theorem:4.3.3: Let f be a function of bounded variation on [a, b] then its variation function vf

is continuous at c ∈ [a, b] if and only if f is continuous at c ∈ [a, b].

Proof: Suppose vf is continuous at c ∈ [a, b] . then for ϵ > 0 there exists δ > 0 such that for x ∈

[a, b] and |x − c| < 𝛿,

52
we have |vf (x) − vf (c)| < 𝜖. Since |f(x) − f(c)| ≤ |vf (x) − vf (c)|, it follows that f is

continuous at x0 . Conversely suppose f is continuous at x0 . then for ϵ > 0 there exist δ > 0 such

that for x ∈ [a, b] and |x − c| < 𝛿, we have

ϵ
|f(x) − f(c)| < .
2

Also there exists a partition P = {c = x0 , x1 , … , xn = b} when c ≠ b, so as

ϵ
vf (x) > Tcb (f) − .
2
ϵ
And 0 < x1 − c < 𝛿  |f(x1 ) − f(c)| < 2.

ϵ ϵ
Therefore Tcb (f) − 2 < vf (P) < 2 + ∑ni=2 |f(xi ) − f(xi−1 )|

ϵ
≤ 2 + Txb1 (f).

i.e. Tcb (f) − Txb1 (f) < 𝜖, therefore 0 < vf (x1 ) − vf (c) < 𝜖 whenever |x1 − c| < 𝛿.

Hence vf is continuous at c.

Theorem 4.3.4: Let f and α be complex valued functions defined on [a, b] satisfying one of the

following conditions:

i. f is continuous and α is a function of bounded variation on [a, b]

ii. f and α are of bounded variation on [a, b] and α is continuous.

If vf is the total variation function of α, then

b b
|∫a f dα | ≤ ∫a |f| dvα .

𝐏𝐫𝐨𝐨𝐟: If (1) holds, then f is continuous and so |f| is continuous.

If (2) holds, vf is continuous by previous theorem.

b
Since f is of bounded variation, so is |f|. Therefore in either case, ∫a |f|dvα exists.

If P = {x0 , x1 , … , xn } is any partition of [a, b] then

53
| ∑ni=1 f(ξi )Δαi | ≤ ∑ni=1 |f(ξi ||Δαi | ≤ ∑ni=1 |f(ξi )|Δvα , (4.3.3)

where xi−1 ≤ ξi ≤ xi for 1 ≤ i ≤ n.

b
It follows that right side of the above inequality tends to ∫a |f|dvf .

b b
Hence we have |∫a f dα | ≤ ∫a |f| dvα .

Theorem 4.3.5: Let f and α be complex valued functions of bounded variation on [a, b] and let

be also continuous on [a, b].

b b
Then ∫a fd α = f(b)α(b) − f(a)α(a) − ∫a α d f.

Proof: Let P = {a = x0 ≤ x1 ≤ ⋯ ≤ xn = b} be any partition of [a, b]

and let Q = {ξ0 = a, ξ1 , … , ξn = b} be an intermediate partition of P so that xi−1 ≤ ξ ≤ xi for i =

1,2, … , n.

Then we have

S(P, Q, f, α) = ∑ni=1 f(ξi )Δαi = f(ξ1 )α(x1 ) − f(ξ1 )α(x0 ) + f(ξ2 )α(x2 ) − f(ξ2 )α(x1 ) +

⋯ + f(ξn )α(xn ) − f(ξn )α(xn−1 ) = f(x0 )α(x0 ) − α(x0 )[f(ξ1 ) − f(ξ0 )] − α(x1 )[f(ξ2 ) −

f(ξ1 )] − α(xn−1 )[f(ξn ) − f(ξn−1 )] + f(xn )α(xn ).

Adding and subtracting the term f(x0 )α(x0 ) and rearranging,

we obtain
n

S(P, Q, f, α) = f(b)α(b) − f(a)α(a) − ∑ α(xi−1 )[f(xi ) − f(xi−1 )]


i=1

⇒ S(P, Q, f, α) = f(b)α(b) − f(a)α(a) − S(Q, P, α, f) (4.3.6)

b
If ||P|| → 0 ⇒ ||Q|| → 0 and therefore S(P, Q, f, α) → ∫a f dα

b
and S(Q, P,′ α, f) → ∫a α df . It follows from (4.3.6) that,

b b
∫a fd α = f(b)α(b) − f(a)α(a) − ∫a α d f.

54
Theorem 4.3.6: Let α be of bounded variation on [a, b] and let V(x)be the toal variation of

α on [a, x] for a < 𝑥 ≤ 𝑏.Let f be bounded on [a, b]. If f ∈ R[α]on [a, b], then f ∈ R[V]on [a, b].

Proof: The result is trivial for the case of V(b) = 0, since V is constant.

So we assume that V(b) > 0. Suppose |f(x)| ≤ M for all x ∈ [a, b].

Since V is increasing, we need only to prove the condition for integrability.

Let ϵ > 0 be given and let P ∗ be any refinement of partition Pϵ corresponding to ϵ.

Then for all choices of the points sk , sk′ ∈ [xi−1 , xi ],

we have
ϵ
|∑ni=1[f(si ) − f(si′ )]Δαi | < and
4

ϵ
V(b) < ∑ni=1|Δαi | + 4M . Since ΔVi − |Δαi | ≥ 0,

we have

∑ni=1[Mi (f) − mi (f)](ΔVi − |Δαi |) ≤ 2M ∑ni=1(ΔVi − |Δαi |)


ϵ
= 2M(V(b) − ∑ni=1|Δαi |) < 2. (4.3.7)

1
( ϵ)
∗)
Let A(P = {i: Δαi ≥ 0} and B(P ∗ ) = {i: Δαi < 0} and let h = V(b)
4
.

If i ∈ A(P ∗ ), then choose si and si′ such that

f(si ) − f(si′ ) > Mi (f) − mi (f) − h.

If i ∈ B(P ∗ ), then choose si and si′ such that f(si′ ) − f(si ) > Mi (f) − mi (f) − h.

Then ∑ni=1[Mi (f) − mi (f)]|Δαi | < ∑i∈B(P∗)[f(si ) − f(si′ )]|Δαi |

+ ∑i∈A(P∗)[f(si′ ) − f(si )]|Δαi |+ h ∑ni=1|Δαi |

= ∑ni=1[f(si ) − f(si′ )]Δαi + h ∑ni=1|Δαi |


ϵ ϵ ϵ ϵ
< 4 + h V(b) = 4 + 4 = 2. (4.3.8)

Therefore by adding (4.3.7) and (4.3.8)

55
we obtain that U(P ∗ , f, V) − L(P ∗ , f, V) < 𝜖.

Theorem4.3.7: Let 𝛼 be of bounded variation on [𝑎, 𝑏] and let 𝑓 ∈ ℝ[(𝛼)] with 𝑓 on [𝑎, 𝑏]. If
𝑥
we define 𝐹(𝑥) = ∫𝑎 𝑓(𝑡) 𝑑𝛼(𝑡) for 𝑎 ≤ 𝑥 ≤ 𝑏, we have
a) 𝐹 is of bounded variation on [𝑎, 𝑏]
b) every point of continuity of 𝛼 is a point of continuity of 𝐹
c) If 𝛼 is increasing then 𝐹 ′ (𝑥) exists at every point where 𝛼 ′ (𝑥) exists and 𝑓(𝑥) is
continuous for such 𝑥 we have 𝐹 ′ (𝑥) = 𝑓(𝑥)𝛼 ′ (𝑥).
Proof: Without loss of generality we can assume 𝛼 is increasing. If 𝑥 ≠ 𝑦 then by the first mean
value theorem for R-S integrals,
we have
𝑥
𝐹(𝑦) − 𝐹(𝑥) = ∫𝑎 𝑓(𝑡) 𝑑𝛼(𝑡) = 𝑐(𝛼(𝑦) − 𝛼(𝑥)) (4.3.9)
where 𝑚 = inf 𝑓 ≤ 𝑐 ≤ 𝑠𝑢𝑓 𝑓 = 𝑀 this yields (a) and (b).
To prove (c), dividing (4.3.9) by 𝑦 − 𝑥 > 0,
we obtain
𝐹(𝑦) − 𝐹(𝑥) 𝑐(𝛼(𝑦) − 𝛼(𝑥))
=
𝑦−𝑥 𝑦−𝑥
Taking limit as 𝑦 ⟶ 𝑥, we obtain
𝐹(𝑦) − 𝐹(𝑥) (𝛼(𝑦) − 𝛼(𝑥))
lim = 𝑐 lim
𝑦→𝑥 𝑦−𝑥 𝑛→∞ 𝑦−𝑥
as 𝑦 ⟶ 𝑥 𝑐 = 𝑓(𝜉) ⟶ 𝑓(𝑥)
Hence we have 𝐹 ′ (𝑥) = 𝑓(𝑥)𝛼 ′ (𝑥).

4.4 Example
3
Example 4.4.1: Evaluate ∫0 𝑥 𝑑([𝑥] − 𝑥).

Solution: here 𝑥 and [𝑥] − 𝑥 are of bounded variation on [0,3] and 𝑥 is also continuous.

Therefore from the previous result,

we have

56
3 3
∫ 𝑥 𝑑([𝑥] − 𝑥) = {𝑥([𝑥] − 𝑥)}30 − ∫ ([𝑥] − 𝑥))𝑑𝑥
0 0

1 2 3
= − ∫0 (−𝑥)𝑑𝑥 − ∫1 (1 − 𝑥)𝑑𝑥 − ∫2 (2 − 𝑥)𝑑𝑥

1 22 1 32 22 3
= 2 − 1 + ( 2 − 2) − 2.1 + ( 2 − ) = 2.
2

4.5 Rectifiable Curves

Here we shall study an application of the theory of integration

Definition 4.5.1: A continuous mapping 𝛾 of an interval [a, b] into ℝ𝑘 is called a curve in ℝ𝑘 .

Sometimes we say that 𝛾 is a curve on [a, b].

If 𝛾 is one-to-one, 𝛾 is called an arc..

If 𝛾(𝑎) = 𝛾(𝑏), 𝛾 is said to be a closed curve . Let f: [a, b] ⟶ ℝ be any function

and let P = {a = x0 , x1 , … , xn } of [a, b]. Then Vab (f) = ∑ni=1 |f(xi ) − f(xi−1 | and

Tab (f) = sup Vab (f) are respectively called the variation function and the total variation of f on
P

[a, b]. Tab (f) may be finite or infinite. If Tab (f) is finite then we say that f is of bounded variation

on [a, b]. Equivalently a function f on [a, b] is said to be of bounded variation on [a, b] if there

exists K > 0 such that for every partition P of [a, b] , ∑ni=1 |f(xi ) − f(xi−1 | ≤ K.

Remarks: 1) A curve is defined as a mapping. Hence with each curve 𝛾 in ℝ𝑘 there is

associated a subset of ℝ𝑘 , namely the range of 𝛾.

2) Different curves may have the same range.

Now we associate to each partition P = {a = x0 , x1 , … , xn } of [a, b]. and to each curve 𝛾

on [a, b]. the number

𝛬(𝑃, 𝛾) = ∑ni=1 ‖f(xi ) − f(xi−1 )||

Note that, the ith term

57
‖𝛾(xi ) − 𝛾(xi−1 )||

Is the distance (in ℝ𝑘 ) between the points 𝛾(xi−1 ) and 𝛾(xi ). Hence 𝛬(𝑃, 𝛾) is the length of a

polygonal path with vertices at 𝛾(x0 ), 𝛾(x1 ), … , 𝛾(xn ), in this order. As our partition becomes

finer and finer, this polygon approaches the range of 𝛾 more and more closely.

Hence we define the length of 𝛾 as 𝛬( 𝛾)= sup 𝛬(𝑃, 𝛾),

Where the supremum is taken over all partitions of [a, b].

If 𝛬( 𝛾) < ∞ , we say that 𝛾 is rectifiable.

A curve 𝛾 is called continuously differentiable if 𝛾′ is continuous.

Theorem 4.5.1: If 𝛾 ′ is continuous function on [a, b], then 𝛾 is rectifiable, and

𝛬(𝛾) = ∫||𝛾 ′ (𝑡)|| 𝑑𝑡


𝑎

Proof: If a ≤ xi−1 < xi ≤ 𝑏, then


x
‖𝛾(xi ) − 𝛾(xi−1 )||= ∫x i ||𝛾 ′ (𝑡)|| 𝑑𝑡.
i−1

x
≤ ∫x i ||𝛾 ′ (𝑡)|| 𝑑𝑡.
i−1

Hence,
x
𝛬(𝛾) ≤ ∫x i ||𝛾 ′ (𝑡)|| 𝑑𝑡.
i−1

𝑏
For every partition P of [a, b] consequently, 𝛬(𝛾) = ∫𝑎 ||𝛾 ′ (𝑡)|| 𝑑𝑡. (1)

To prove the opposite inequality, let ↋ > 0 be given , Since 𝛾′ is continuous on [a, b],

it is uniformly continuous on [a, b].so there exists a 𝜕 > 0 such that

‖𝛾′(s) − 𝛾(t)|| < ↋, if │s-t│< 𝜕.

Let P = {a = x0 , x1 , … , xn } be a partition of [a, b] with ∆𝑥𝑖 < 𝜕 for all i,

58
if xi−1 ≤ 𝑡 ≤ xi , it follows that

||𝛾 ′ (𝑡)|| ≤ ‖𝛾′(xi )‖ + ↋.

Hence
x
∫x i ||𝛾 ′ (𝑡)|| 𝑑𝑡. ≤ ‖𝛾′(xi )‖ ∆𝑥𝑖 + ↋∆𝑥𝑖
i−1

x
= ‖ ∫x i [𝛾 ′ (𝑡) + 𝛾′(xi ) − 𝛾 ′ (𝑡)]𝑑𝑡|| + ↋∆𝑥𝑖
i−1

x x
≤ ∫x i ||𝛾 ′ (𝑡)𝑑𝑡 || + ‖∫x i [𝛾 ′ (xi ) − 𝛾 ′ (𝑡)𝑑𝑡‖ + ↋ ∆𝑥𝑖
i−1 i−1

≤ ‖𝛾(xi ) − 𝛾(xi−1 )||+2↋ ∆𝑥𝑖 .

If we add these inequalities, we obtain,

𝑏
∫𝑎 ||𝛾 ′ (𝑡)|| 𝑑𝑡 ≤ 𝛬(𝑃, 𝛾)+ 2↋(b − a).

≤ 𝛬( 𝛾)+ 2↋(b − a).

Since ↋ is arbitrary, we have

𝑏
∫𝑎 ||𝛾 ′ (𝑡)|| 𝑑𝑡 ≤ 𝛬( 𝛾). (2)

From (1) and (2) it follows that

𝛬(𝛾) = ∫||𝛾 ′ (𝑡)|| 𝑑𝑡


𝑎

This completes the proof.

⃗⃗⃗ ∈ ℝ𝑘 , then the distance between them is given by,


⃗ ,𝒚
Notes: If 𝒙

𝑑(𝑥,
⃗⃗⃗ 𝑦 ) = || 𝑥 - 𝑦 ||, the norm of 𝑥 - 𝑦.

If x, y ∈ ℝ, then the distance between them is given by

𝑑(𝑥, 𝑦) = |𝑥 − 𝑦 |, the modulus of 𝑥 − 𝑦.

59
4.6 Summary

1. A function f on [a, b] is said to be of bounded variation on [a, b] if there exists K > 0 such

that for every partition 𝑃 of [𝑎, 𝑏] , ∑𝑛𝑖=1 |𝑓(𝑥𝑖 ) − 𝑓(𝑥𝑖−1 | ≤ 𝐾.

2. A function of bounded variation is always bounded.

3. A monotonic and bounded function 𝑓 on [𝑎, 𝑏] is a function of bounded variation on [𝑎, 𝑏]

and 𝑉(𝑓) = |𝑓(𝑏) − 𝑓(𝑎)|.

4. Let 𝑓 be a function of bounded variation on [𝑎, 𝑏] then its variation function 𝑣𝑓 is continuous

at 𝑐 ∈ [𝑎, 𝑏] if and only if 𝑓 is continuous at 𝑐 ∈ [𝑎, 𝑏].

5. Let 𝑓 and 𝛼 be complex valued functions defined on [𝑎, 𝑏] satisfying one of the following

conditions:

i. 𝑓 is continuous and α is a function of bounded variation on [𝑎, 𝑏]

ii. 𝑓 and α are of bounded variation on [𝑎, 𝑏] and α is continuous.

𝑏 𝑏
If 𝑣𝑓 is the total variation function of 𝛼, then|∫𝑎 𝑓 𝑑𝛼 | ≤ ∫𝑎 |𝑓| 𝑑𝑣𝛼 .

4.7 Keywords

Function of bounded variation, variation function, total variation, monotonic, continuous

4.8 Exercises

1. Show that the variation function 𝑣𝑓 of 𝑓 is continuous at 𝑐 ∈ [𝑎, 𝑏] if and only if 𝑓 is


continuous at 𝑐 ∈ [𝑎, 𝑏]

2. Prove that if 𝑓 and 𝛼 are complex valued functions of bounded variation on [𝑎, 𝑏] and also
𝑏 𝑏
continuous on [𝑎, 𝑏] then ∫𝑎 𝑓𝑑 𝛼 = 𝑓(𝑏)𝛼(𝑏) − 𝑓(𝑎)𝛼(𝑎) − ∫𝑎 𝛼 𝑑 𝑓.

60
2
3. Evaluate ∫0 𝑥 3 𝑑[𝑥 2 ], where [𝑥] is the largest integer not greater than 𝑥.

Solutions:

1. See theorem 4.3.1.

2. See theorem 4.3.3


2 2
3. By theorem 4.3.3 we have ∫0 𝑥 3 𝑑[𝑥 2 ] = 23 [22 ] − 0 − ∫0 [𝑥 2 }𝑑𝑥 3

1 √2 √3 2
= 32 − {∫0 [𝑥 2 ]𝑑𝑥 3 + ∫1 [𝑥 2 ]𝑑𝑥 3 + ∫√2 [𝑥 2 ]𝑑𝑥 3 + ∫√3[𝑥 2 ]𝑑𝑥 3 }

= 32 − 0 − 1(2√2 − 1) − 2(3√3 − 2√2) − 3(263 − 3√3)

= 9 + 2√2 + 3√3.

4.9 References

1. Walter Rudin, Principles of Mathematical Analysis, McGraw-Hill 3rd edition, 1976.


2. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing, 1974.

61
Block – II
Sequences and Series of Functions

62
UNIT 5

UNIFORM CONVERGENCE AND CONTINUITY

5.1 Objectives

5.2 Introduction

5.3 Pointwise Convergence

5.4 Uniform Convergence

5.5 Summary

5.6 Key Words

5.7 Exercises

5.8 References

63
UNIT 5

UNIFORM CONVERGENCE AND CONTINUITY

5.1 Main Objective

After going through this unit, student should be able to,


(i) Distinguish pointwise convergence and uniform convergence.
(ii) To examine whether the given sequence is uniformly convergent or not.
(iii) Understand criterion for uniform convergence of sequences and of series.
(iv) Apply the Cauchy and Weierstrass criterions to check the uniform convergence of the
sequences or series.

5.2 Introduction

Sequences of functions are of great importance in many areas of pure and applied

mathematics, and their properties can be studied in the context of metric spaces. Sequences of

functions are useful in approximating a given function and to define new functions from the

given ones. Pointwise convergence and uniform convergence are of importance when we look at

sequences of functions. A type of convergence that preserves at least some of the properties of a

function sequence is always preferred. Such a concept is uniform convergence.

For pointwise convergence we could first fix a value for x and then choose N.

Consequently, N depends on 𝛜 and 𝒙 both . For uniform convergence 𝐟𝐧 (𝐱) must

be uniformly close to 𝐟(𝐱) for all 𝐱 in the domain. Thus 𝐍 only depends on 𝛜 but not on 𝐱.

Uniform convergence clearly implies pointwise convergence, but the converse is false.

64
5.3 Pointwise convergence

Definition 5.3.1: Let 𝑨 ⊆ 𝑹. Let {𝒇𝒏 } be a sequence of real valued functions defined on 𝑨. The

sequence {𝒇𝒏 } converges pointwise to 𝒇 on 𝑨, if

𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝒇(𝒙), 𝒙 ∈ 𝑨. (5.3.1)


𝒏→∞

If {𝒇𝒏 } converges to 𝒇, then 𝒇 is called the limit of the sequence {𝒇𝒏 }.

Remark.1: According to the definition 5.3.1, {𝒇𝒏 } converges to 𝒇, on 𝑨 if for each 𝒙 ∈ 𝑨 and

given 𝝐 > 𝟎 there exists a positive integer 𝑵 such that |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 for all 𝒏 ≥ 𝑵. In

general the number 𝑵 depends on both 𝝐 and 𝒙.

5.3.1 Example

1. Let 𝒇𝒏 (𝒙) = 𝒙𝒏 , 𝟎 ≤ 𝒙 ≤ 𝟏 and 𝒏 ∈ 𝑵. Then prove that {𝒇𝒏 } converges on [𝟎, 𝟏].

If 𝒙 = 𝟏, then {𝒇𝒏 (𝟏)} = {𝟏} converges to 1.

And 𝐥𝐢𝐦 𝒙𝒏 = 𝟎 for 𝟎 ≤ 𝒙 < 𝟏. Thus {𝒇𝒏 } converges on [𝟎, 𝟏] to 𝒇, where 𝒇(𝒙) =
𝒏→∞

𝟎 𝒇𝒐𝒓 𝟎 ≤ 𝒙 < 𝟏
{ .
𝟏 𝒇𝒐𝒓 𝒙 = 𝟏
𝒙
2. Let 𝒇𝒏 (𝒙) = 𝒏 , 𝒇𝒐𝒓 𝒏 ∈ 𝑵 𝒂𝒏𝒅 𝒙 ∈ 𝑹.

𝒙 𝟏
Since 𝐥𝐢𝐦 𝟏/𝒏 = 𝟎 , it follows that 𝐥𝐢𝐦 = 𝒙 𝐥𝐢𝐦 = 𝒙. 𝟎 = 𝟎, ∀ 𝒙 ∈ 𝑹.
𝒏→∞ 𝒏→∞ 𝒏 𝒏→∞ 𝒏

𝒙
3. Let 𝒇𝒏 (𝒙) = , 𝟎 ≤ 𝒙 < ∞, for 𝒙 ∈ 𝑹.
𝟏+𝒏𝒙

𝒙 𝟏
If 𝒙 > 𝟎, then 𝟎 < 𝒇𝒏 (𝒙) ≤ 𝒏𝒙 = 𝒏  𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝟎 𝒇𝒐𝒓 𝒙 > 𝟎.
𝒏→∞

Also, since 𝒇𝒏 (𝟎) = 𝟎 for each 𝒏 ∈ 𝑵, it is clear that {𝒇𝒏 }∞


𝟏 converges to 𝒇, where 𝒇 =

𝟎 identically on [𝟎, ∞).

65
𝒏𝒙
4. Let 𝒇𝒏 (𝒙) = 𝟏+𝒏𝟐 𝒙𝟐 , (−∞ < 𝒙 < ∞), 𝒏 ∈ 𝑵.

𝟏
( 𝟐 𝟐)
𝒏 𝒙
If 𝒙 > 𝟎, then we have 𝒇𝒏 (𝒙) = 𝟏 and therefore 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝟎.
( 𝟐 𝟐 )+𝟏 𝒏→∞
𝒏 𝒙

Since 𝒇𝒏 (𝟎) = 𝟎, for each 𝒏, we have {𝒇𝒏 }∞


𝟏 converges 𝒇 = 𝟎 on (−∞, ∞).

Definition 5.3.2: Let 𝑨 ⊆ 𝑹. Let {𝒇𝒏 } be a sequence of real valued functions defined on 𝑨. Let

𝑺𝒏 = ∑𝒏𝒊=𝟏 𝒇𝒊 . The series ∑∞


𝒏=𝟏 𝒇𝒏 converges to 𝒇 pointwise if the sequence of functions {𝑺𝒏 }

converges to 𝒇 on 𝑨. In this case we write ∑∞


𝒏=𝟏 𝒇𝒏 = 𝒇 or ∑∞
𝒏=𝟏 𝒇𝒏 (𝒙) = 𝒇(𝒙), 𝒙 ∈ 𝑨.

5.4 Uniform convergence

In the definition of pointwise convergence, we have seen that {𝒇𝒏 }∞


𝟏 converges

pointwise to 𝒇 on 𝑨 if for each 𝒙 ∈ 𝑨, given 𝝐 > 𝟎 there exists a positive integer 𝑵 such that

|𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 for all 𝒏 ≥ 𝑵. (5.4.1)

There are several examples in which it is possible to find a 𝑵 such that equation 5.4.1 holds for

all 𝒙 ∈ 𝑨. In this case we say that the convergence is uniform.

Definition 5.4.1: Let {𝒇𝒏 }∞


𝟏 be a sequence of real valued functions defined on a set ⊂ 𝑹 . Then

{𝒇𝒏 }∞
𝟏 is said to converge to 𝒇 uniformly on 𝑨 if given 𝝐 > 𝟎 there exists a positive integer 𝑵

such that |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 for all 𝒏 ≥ 𝑵 and for all 𝒙 ∈ 𝑨.

Note that the number 𝑵 depends only on 𝝐, but not on 𝒙.

It is clear that uniform convergence implies pointwise convergence.

Uniform convergence for series of functions is defined in a similar fashion.

Definition 5.4.2: Let {𝒇𝒏 }∞


𝟏 be a sequence of real valued functions defined on a set ⊂ 𝑹 . The

series ∑∞
𝒏=𝟏 𝒇𝒏 converges uniformly to 𝒇 on 𝑨 if the sequence {𝑺𝒏 } of partial sums converges

uniformly to 𝒇 on 𝑨. In this case we write ∑∞ ∞


𝒏=𝟏 𝒇𝒏 = 𝒇 uniformly or ∑𝒏=𝟏 𝒇𝒏 (𝒙) = 𝒇 (𝒙)

uniformly.

66
5.4.1 Examples

1. Let 𝒇𝒏 (𝒙) = 𝒙𝒏 , (𝟎 ≤ 𝒙 ≤ 𝟏) for each 𝒏 ∈ 𝑵. Show that the sequence {𝒇𝒏 } is not

uniformly convergent on [𝟎, 𝟏) but converges uniformly on 𝟎 ≤ 𝒙 ≤ 𝒌, for some 𝒌

satisfying 𝟎 < 𝒌 < 𝟏.

Solution: For 𝟎 ≤ 𝒙 < 𝟏, 𝐥𝐢𝐦 𝒙𝒏 = 𝟎 and 𝒍𝒊𝒎𝒏→∞ 𝒙𝒏 = 𝟏 if 𝒙 = 𝟏.


𝒏→∞

𝟎 𝒊𝒇 𝟎 ≤ 𝒙 < 𝟏
Thus if we define 𝒇(𝒙) = { , then we have 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝒇(𝒙).
𝟏 𝒊𝒇 𝒙 = 𝟏 𝒏→∞

i.e. the sequence {𝒇𝒏 (𝒙)} converges to 𝒇(𝒙) on [𝟎, 𝟏). Let us check whether this convergence is

uniform or not.

Let 𝝐 > 𝟎 be given. If |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐, then |𝒙𝒏 − 𝟎| < 𝝐  𝒙𝒏 < 𝝐

𝟏
𝟏 𝟏 𝟏 𝟏 𝒍𝒐𝒈( )
 𝒙𝒏 >  𝒏 𝒍𝒐𝒈 (𝒙) > 𝒍𝒐𝒈 (𝝐 )  𝒏 > 𝝐
𝟏 . (5.4.3)
𝝐 𝒍𝒐𝒈( )
𝒙

Now from (5.4.3) , as 𝒙 ranges over [𝟎, 𝟏) 𝒏 approaches ∞. Therefore it is not possible to find a

positive integer 𝒎 such that 𝒏 ≥ 𝒎  |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 for all 𝒙 ∈ [𝟎, 𝟏).

Therefore {𝒇𝒏 } is not uniformly convergent on [𝟎, 𝟏).

Let us consider the interval 𝟎 ≤ 𝒙 ≤ 𝒌 for some 𝒌 such that 𝟎 < 𝒌 < 𝟏.
𝟏 𝟏 𝟏
𝒍𝒐𝒈( ) 𝒍𝒐𝒈( ) 𝒍𝒐𝒈( )
𝝐 𝝐 𝝐
Then 𝐦𝐚𝐱 𝟏 = 𝟏 and so for > 𝟏 , we have
𝒍𝒐𝒈( ) 𝒍𝒐𝒈( ) 𝒍𝒐𝒈( )
𝒙 𝒌 𝒌

𝒏 ≥ 𝒎  |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 for all 𝒙 ∈ [𝟎, 𝒌]. Thus {𝒇𝒏 (𝒙)} is uniformly convergent on

[𝟎, 𝒌].

The following theorem gives a necessary and sufficient condition for uniform convergence.

67
Theorem5.4.1.(Cauchy criterion for Uniform convergence):

Let {𝒇𝒏 }∞ ∞
𝒏=𝟏 be a sequence of real valued functions on a set 𝑨. Then {𝒇𝒏 }𝒏=𝟏 converges

uniformly to 𝒇 on 𝑨 if and only if given 𝝐 > 𝟎 there exists a positive integer 𝑵 such that

|𝒇𝒎 (𝒙) − 𝒇𝒏 (𝒙)| < 𝝐, for all 𝒎, 𝒏 ≥ 𝑵 and 𝒙 ∈ 𝑨. (5.4.4)

Proof: Suppose {𝒇𝒏 }∞


𝒏=𝟏 converges to 𝒇 on 𝑨. Then by definition, given 𝝐 > 𝟎 there exists a

positive integer 𝑵 such that |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 , for all 𝒏 ≥ 𝑵 and 𝒙 ∈ 𝑨.

Thus if 𝒎, 𝒏 ≥ 𝑵 we have for any 𝒙 ∈ 𝑨,


𝝐 𝝐
|𝒇𝒎 (𝒙) − 𝒇𝒏 (𝒙)| ≤ |𝒇𝒎 (𝒙) − 𝒇(𝒙)| + |𝒇𝒏(𝒙) − 𝒇(𝒙)| < + = 𝝐, and hence the given
𝟐 𝟐

condition holds for this 𝑵.

Conversely, suppose the given condition (5.4.4) holds for any sequence {𝒇𝒏 }∞
𝒏=𝟏 of functions on

𝑨. We must show that {𝒇𝒏 }∞


𝒏=𝟏 converges uniformly to some function 𝒇 on 𝑨. From the given

condition (5.4.4) it follows that, for each 𝒙 ∈ 𝑨, the sequence {𝒇𝒏 (𝒙)}∞
𝒏=𝟏 is a Cauchy sequence

of real numbers and hence 𝐥𝐢𝐦 𝒇𝒏 (𝒙) exists for each 𝒙 ∈ 𝑨. Set 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝒇(𝒙) for 𝒙 ∈ 𝑨.
𝒏→∞ 𝒏→∞

We have to prove that the convergence is uniform. Let 𝝐 > 𝟎 be given. Keeping 𝒎 fixed and

letting 𝒏 → ∞ we obtain |𝒇𝒎 (𝒙) − 𝒇(𝒙)| ≤ 𝝐 for all 𝒎 ≥ 𝑵 and 𝒙 ∈ 𝑨.

Since 𝝐 > 𝟎 was arbitrary, it follows that {𝒇𝒏 }∞


𝒏=𝟏 converges uniformly to 𝒇 on 𝑨. This

completes proof of the theorem.

We have another criterion for uniform convergence.

Theorem 5.4.2. (M-Test).

Let {𝒇𝒏 }∞
𝒏=𝟏 be a sequence of real valued functions on a set 𝑨 ⊂ 𝑹 and suppose

𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝒇(𝒙) for all 𝒙 ∈ 𝑨. Set 𝑴𝒏 = 𝐬𝐮𝐩 |𝒇𝒏 (𝒙) − 𝒇(𝒙)| . Then
𝒏→∞ 𝐱∈𝐀

68
{𝒇𝒏 }∞
𝒏=𝟏 converges uniformly to 𝒇 on 𝑨 if and ony if 𝑴𝒏 → 𝟎 𝒂𝒔 𝒏 → ∞.

Proof: Suppose first that {𝒇𝒏 }∞


𝒏=𝟏 converges uniformly to 𝒇 on 𝑨. Then for every 𝝐 > 𝟎, there

exists a positive integer 𝑵 such that for all 𝒏 ≥ 𝑵 and 𝒙 ∈ 𝑨, we have

|𝒇𝒏 (𝒙) − 𝒇(𝒙| < 𝝐

Since 𝑴𝒏 = 𝐬𝐮𝐩 |𝒇𝒏 (𝒙) − 𝒇(𝒙)| it follows from (1.2.5) that for all 𝒏 ≥ 𝑵, we have 𝑴𝒏 < 𝝐.
𝐱∈𝐀

Hence 𝑴𝒏 → 𝟎 as 𝒏 → ∞.

Conversely suppose that 𝑴𝒏 → 𝟎 as 𝒏 → ∞. Then given 𝝐 > 𝟎, there exist a positive integer

𝑵 such that 𝑴𝒏 < 𝝐 for all 𝒏 ≥ 𝑵.

Since 𝑴𝒏 is the supremum of |𝒇𝒏 (𝒙) − 𝒇(𝒙)| , we have |𝒇𝒏 (𝒙) − 𝒇(𝒙) ≤ 𝑴𝒏 < 𝝐 for all 𝒙 ∈ 𝑨

and all 𝒏 ≥ 𝑵. Hence {𝒇𝒏 }∞


𝒏=𝟏 converges uniformly to 𝒇 on 𝑨.

For series there is a very convenient test for uniform convergence due to Weierstrass.

Theorem 5.4.3: (Weierstrass M-test for uniform convergence of series)

Let {𝒇𝒏 }∞
𝒏=𝟏 be a sequence of real valued functions on a set 𝑨 ⊂ 𝑹 and suppose |𝒇𝒏 (𝒙)| ≤ 𝑴𝒏

for all 𝒙 ∈ 𝑨 and 𝒏 = 𝟏, 𝟐, …,Then ∑∞ ∞


𝟏 𝒇𝒏 converges uniformly on 𝑨 if ∑𝟏 𝑴𝒏 converges.

Proof: If ∑∞
𝟏 𝑴𝒏 converges then for arbitrary 𝝐 > 𝟎, we can find a positive integer 𝑵 such that

for all 𝒏 ≥ 𝑵, we have ∑𝒎


𝒊=𝒏 𝑴𝒊 < 𝝐. Since |𝒇𝒏 (𝒙)| ≤ 𝑴𝒏 for all 𝒙 ∈ 𝑨 and 𝒏 = 𝟏, 𝟐, … , it

follows that |∑𝒎 𝒎 𝒎


𝒊=𝒏 𝒇𝒊 (𝒙)| ≤ ∑𝒊=𝒏|𝒇𝒊 (𝒙)| ≤ ∑𝒊=𝒏 𝑴𝒊 < 𝝐 for all 𝒙 ∈ 𝑨 and 𝒏 ≥ 𝒎. Hence the

series ∑∞
𝟏 𝒇𝒏 converges uniformly and absolutely on 𝑨.

5.4.2 Examples

1. The sequence {𝒇𝒏 }∞ 𝒏


𝒏=𝟏 ,where 𝒇𝒏 (𝒙) = 𝒏𝒙(𝟏 − 𝒙) is not uniformly convergent on [0,1].

(𝒏𝒙) ∞
Solution: Consider 𝐥𝐢𝐦𝒇𝒏 (𝒙) = 𝐥𝐢𝐦(𝒏𝒙)(𝟏 − 𝒙)𝒏 = 𝐥𝐢𝐦 (𝟏−𝒙)−𝒏 (= ∞ 𝒇𝒐𝒓𝒎)
𝒏→∞ 𝒏→∞ 𝒏→∞

69
Hence by L’Hospital’s rule

(𝒏𝒙) 𝒙
𝐥𝐢𝐦𝒇𝒏 (𝒙) = 𝐥𝐢𝐦 (𝟏−𝒙)−𝒏 = 𝐥𝐢𝐦 −(𝟏−𝒙)−𝒏 𝐥𝐨𝐠(𝟏−𝒙)
𝒏→∞ 𝒏→∞ 𝒏→∞

−𝒙(𝟏−𝒙)𝒏
= 𝐥𝐢𝐦 = 𝟎, 𝒔𝒊𝒏𝒄𝒆 (𝟏 − 𝒙)𝒏 → 𝟎 𝒂𝒔 𝒏 → ∞.
𝒏→∞ 𝐥𝐨𝐠(𝟏−𝒙)

Hence 𝒇(𝒙) = 𝐥𝐢𝐦𝒇𝒏 (𝒙) = 𝟎, ∀ 𝒙 ∈ [𝟎, 𝟏].


𝒏→∞

Now 𝑴𝒏 = 𝐬𝐮𝐩 |𝒇𝒏 (𝒙) − 𝒇(𝒙)| = 𝐬𝐮𝐩 |𝒏𝒙(𝟏 − 𝒙)𝒏 |


𝒙∈[𝟎,𝟏] 𝒙∈[𝟎,𝟏]

𝟏 𝟏 𝒏 𝟏 𝒏
≥ 𝒏. 𝒏 (𝟏 − 𝒏) = (𝟏 − 𝒏) → 𝒆 𝒂𝒔 𝒏 → ∞.

(𝐜𝐨𝐬 𝒏𝒙)
2. The series ∑∞
𝒏=𝟏 converges uniformly on R.
𝒏𝟐

(𝐜𝐨𝐬 𝒏𝒙) 𝟏 𝟏
Solution: Since | | ≤ , ∀ 𝒙 ∈ 𝑹, and the series ∑∞
𝒏=𝟏 𝒏𝟐 is convergent, it follows from
𝒏𝟐 𝒏𝟐

(𝐜𝐨𝐬 𝒏𝒙)
Weierstrass M-test that, the series ∑∞
𝒏=𝟏 is uniformly convergent.
𝒏𝟐

𝒏𝒙
3. Show that the series ∑∞
𝟏 𝟏+𝒏𝟐 𝒙𝟐 does not converge uniformly on [𝟎, 𝟏].

𝒏𝒙
Solution: Now 𝒇(𝒙) = 𝐥𝐢𝐦 𝟏+𝒏𝟐 𝒙𝟐 = 𝟎, ∀ 𝒙 ∈ 𝑹.
𝒏→∞

To check this convergence is uniform, if the convergence is uniform, then for a given 𝝐 > 𝟎

there exists a positive integer 𝒏𝟎 such that

𝒏|𝒙|
𝒏 ≥ 𝒏𝟎  |𝒇𝒏 (𝒙) − 𝒇(𝒙)| = 𝟏+𝒏𝟐 𝒙𝟐 < 𝝐.

𝟏
𝟏 𝒏. 𝟏
𝒏
If 𝒙 = 𝒏 for 𝒏 = 𝟏, 𝟐, … then |𝒇𝒏 (𝒙) − 𝒇(𝒙)| = 𝟏 = 𝟐.
𝒏𝟐 . 𝟐
𝒏

𝟏
Hence if there exists a positive integer 𝒎 such that |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐, then for 𝒙 = 𝒎 , we

𝟏 𝟏 𝒏𝒙
obtain 𝟐 < 𝟐, a contradiction. Hence the series ∑∞
𝟏 𝟏+𝒏𝟐 𝒙𝟐 does not converge uniformly on [𝟎, 𝟏].

𝟏
4. Test for uniform convergence of the series ∑∞
𝟏 𝒙+𝒏 is uniformly convergent in any interval

[𝟎, 𝒃], 𝒃 > 𝟎.

70
Solution: Here the sum function 𝒇(𝒙) = 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝟎, ∀ 𝒙 ∈ [𝟎, 𝒃]. So that the sequence
𝒏→∞

𝟏
converges pointwise to 𝟎. For any 𝝐 > 𝟎, |𝒇𝒏 (𝒙) − 𝒇(𝒙)| = 𝒙+𝒏 < 𝝐.

𝟏 𝟏
If 𝒏 > 𝝐 − 𝒙, which decreases with 𝒙 and the maximum value being 𝝐 .

𝟏
Let 𝑵 be the positive integer ≥ 𝝐 , then |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐, ∀ 𝒏 ≥ 𝑵.

Hence the sequence is uniformly convergent in [𝟎, 𝒃], 𝒃 > 𝟎.


𝒙
5. Prove that the sequence {𝒇𝒏 }, where 𝒇𝒏 (𝒙) = 𝟏+𝒏𝒙𝟐 , 𝒙 ∈ 𝑹 is uniformly convergent on any

closed interval [𝒂, 𝒃].

Solution: Here the pointwise limit is 𝒇(𝒙) = 𝐥𝐢𝐦𝒇𝒏 (𝒙) = 𝟎, ∀ 𝒙.


𝒏→∞

𝒙 𝟏 𝒙
Now 𝑴𝒏 = 𝒔𝒖𝒑𝒙∈[𝒂,𝒃] |𝒇𝒏 (𝒙) − 𝒇(𝒙)| = 𝐬𝐮𝐩 |𝟏+𝒏𝒙𝟐 | = 𝟐 → 𝟎 , since attains its
𝒙∈[𝒂,𝒃] √𝒏 𝟏+𝒏𝒙𝟐

𝟏
maximum value 𝟐 at 𝒙 = 𝟏/(√𝒏 ) at the origin.
√𝒏

Hence {𝒇𝒏 } converges uniformly on [𝒂, 𝒃].


𝒏𝒙
6. Prove that the sequence {𝒇𝒏 }, where 𝒇𝒏 (𝒙) = 𝟏+𝒏𝟐 𝒙𝟐 is not uniformly convergent on any

interval containing zero.


𝒙
𝒏
Solution: Here 𝒇(𝒙) = 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝐥𝐢𝐦 𝟏 = 𝟎, ∀ 𝒙.
𝒏→∞ 𝒏→∞ +𝒙𝟐
𝒏

𝒏𝒙 𝟏 𝟏
Now 𝟏+𝒏𝟐 𝒙𝟐 attains the maximum value 𝟐 at 𝒙 = 𝒏 → 𝟎, 𝒂𝒔 𝒏 → ∞.

Let [𝒂, 𝒃] be any closed interval containing 0.

𝒏𝒙 𝟏
Then 𝑴𝒏 = 𝐬𝐮𝐩|𝒇𝒏 (𝒙) − 𝒇(𝒙)| = 𝐬𝐮𝐩 |𝟏+𝒏𝟐 𝒙𝟐 | = 𝟐 ↛ 𝟎 𝒂𝒔 𝒏 → ∞.

Hence {𝒇𝒏 } is not uniformly convergent in any interval containing 0.

𝒄𝒐𝒔𝒏𝜽
7. Prove that the series ∑∞
𝟏 is uniformly convergent for all real values of 𝜽 and 𝒑 > 𝟏.
𝒏𝒑

71
𝟏 𝒄𝒐𝒔𝒏𝜽
Solution: Let 𝑴𝒏 = 𝒏𝒑 , 𝒇𝒐𝒓 𝒑 > 𝟏. Since | | ≤ 𝑴𝒏 , ∀ 𝒏 𝒂𝒏𝒅 𝜽 and ∑𝑴𝒏 converges, by
𝒏𝒑

𝒄𝒐𝒔𝒏𝜽
Weierstrass M test the series ∑∞
𝟏 converges uniformly and absolutely for all 𝜽.
𝒏𝒑

5.5 Summary

1. Pointwise Convergence : {𝑓𝑛 } converges pointwise to 𝑓, on 𝐴 if for each 𝑥 ∈ 𝐴 and given

𝜖 > 0 there exists a positive integer 𝑁 such that |𝑓𝑛 (𝑥) − 𝑓(𝑥)| < 𝜖 for all 𝑛 ≥ 𝑁. In

general the number 𝑁 depends on both 𝜖 and 𝑥.

2. Uniform convergence: {𝒇𝒏 }∞


𝟏 is said to converge to 𝒇 uniformly on 𝑨 if given 𝝐 > 𝟎 there

exists a positive integer 𝑵 such that |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 for all 𝒏 ≥ 𝑵 and for all 𝒙 ∈ 𝑨.

Here the number 𝑵 depends only on 𝝐, but not on 𝒙.

3. Uniform convergence implies point wise convergence.

4. Cauchy criterion : A sequence {𝒇𝒏 }∞


𝒏=𝟏 of real valued functions on a set 𝑨 converges

uniformly to 𝒇 on 𝑨 if and only if given 𝝐 > 𝟎 there exists a positive integer 𝑵 such that

|𝒇𝒎 (𝒙) − 𝒇𝒏 (𝒙)| < 𝝐 for all 𝒎, 𝒏 ≥ 𝑵 and 𝒙 ∈ 𝑨.

5. M-test: A sequence {𝒇𝒏 }∞


𝒏=𝟏 of real valued functions on a set 𝑨 ⊂ 𝑹 is such

that 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝒇(𝒙) for all 𝒙 ∈ 𝑨. Set 𝑴𝒏 = 𝐬𝐮𝐩 |𝒇𝒏 (𝒙) − 𝒇(𝒙)| . Then {𝒇𝒏 }∞
𝒏=𝟏
𝒏→∞ 𝐱∈𝐀

converges uniformly to 𝒇 on 𝑨 if and ony if 𝑴𝒏 → 𝟎 𝒂𝒔 𝒏 → ∞.

6. Weierstrass M-test: Let {𝒇𝒏 }∞


𝒏=𝟏 be a sequence of real valued functions on a set 𝑨 ⊂ 𝑹

and suppose |𝒇𝒏 (𝒙)| ≤ 𝑴𝒏 for all 𝒙 ∈ 𝑨 and 𝒏 = 𝟏, 𝟐, …. Then ∑∞


𝟏 𝒇𝒏 converges uniformly

on 𝑨 if ∑∞
𝟏 𝑴𝒏 converges.

5.6 Keywords

Sequence, series, Point wise, uniform, M-test, Weierstrass M test.

72
5.7 Exercises

𝒙𝟐 + 𝒏𝒙
1. Check for the convergence of 𝒇𝒏 (𝒙) = , (−∞ < 𝒙 < ∞), 𝒏 ∈ 𝑵.
𝒏

2. Check for the convergence of 𝒇𝒏 (𝒙) = 𝒙𝒏 , (−𝟏 < 𝒙 < 𝟏) .

𝟏
3. Show that the series ∑∞
𝟏 𝟏+𝒏𝟐 𝒙 converges uniformly on [𝟏, ∞).

Solutions:
x2
1. Since fn (x) = ( n ) + x , it follows from 1.1.1example 4, that lim fn (x) = x, ∀ x ∈ R .
n→∞

𝟏 𝟏
2. Since |𝐱| < 𝟏, ∑∞ ∞ 𝐧
𝐧=𝟏 𝐟𝐧 (𝐱) = ∑𝐧=𝟏 𝐱 = 𝟏−𝐱 . Here 𝐟(𝐱) = 𝟏−𝐱.

𝟏 𝟏 𝟏
3. Set 𝐟𝐧 (𝐱) = 𝟏+𝐧𝟐 𝐱. Then |𝐟𝐧 (𝐱)| ≤ 𝟏+𝐧𝟐 ≤ 𝐧𝟐 , ∀ 𝐱 ∈ [𝟏, ∞).

𝟏 𝟏
Since ∑∞ ∞
𝟏 𝐧𝟐 is convergent, by Weierstrass M-test , the series ∑𝟏 𝟏+𝐧𝟐 𝐱 is uniformly convergent.

5.8 References

1. Richard R Goldberg, Methods of Real analysis, IBH publishing,New Delhi,1970


2. Walter Rudin, Principles of Mathematical Analysis,McGraw-Hill 3rd edition, 1976.
3. Robert G. Bartle, Donald R Sherbert, Introduction to Real Analysis, Wiley India, Third
edition, 2005.
4. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing,1974.

73
UNIT 6
UNIFORM CONVERGENCE AND INTEGRATION

6.1 Main Objectives

6.2 Introduction

6.3 Uniform Convergence And Continuity

6.4 Uniform Convergence And Differentiation

6.5 Uniform Convergence And Integration

6.6 Summary

6.7 Key Words

6.8 Exercises

6.9 References

74
UNIT 6
UNIFORM CONVERGENCE AND INTEGRATION

6.1 Main Objective

After going through this unit, student should be able to,


 Distinguish pointwise convergence and uniform convergence.
 To examine whether the given sequence is uniformly convergent or not.
 Understand criterion for uniform convergence of sequences and of series.
 Apply the Cauchy and Weierstrass criterions to check the uniform convergence of the
sequences or series.

6.2 Introduction

Sequences of functions are of great importance in many areas of pure and applied

mathematics, and their properties can be studied in the context of metric spaces. Sequences of

functions are useful in approximating a given function and to define new functions from the

given ones. Pointwise convergence and uniform convergence are of importance when we look at

sequences of functions. A type of convergence that preserves at least some of the properties of a

function sequence is always preferred. Such a concept is uniform convergence.

For pointwise convergence we could first fix a value for x and then choose N.

Consequently, N depends on 𝛜 and 𝒙 both . For uniform convergence 𝐟𝐧 (𝐱) must

be uniformly close to 𝐟(𝐱) for all 𝐱 in the domain. Thus 𝐍 only depends on 𝛜 but not on 𝐱.

Uniform convergence clearly implies pointwise convergence, but the converse is false.

75
6.3 Uniform Convergence and Continuity

Theorem 6.3.1: Let {𝐟𝐧 } be a sequence of real valued functions defined on a set 𝐀 ⊂ 𝐑 and

let the sequence {𝐟𝐧 } converges uniformly to 𝐟 on 𝐀. Let 𝒄 be a limit point of 𝐀, and suppose that

𝐥𝐢𝐦 𝐟𝐧 (𝐱) = 𝐀 𝐧 , (𝐧 = 𝟏, 𝟐, 𝟑, … )then {𝐀 𝐧 } converges and 𝐥𝐢𝐦 𝐟(𝐱) = 𝐥𝐢𝐦 𝐀 𝐧 .


𝐱→𝐜 𝐱→𝐜 𝐧→∞

In otherwords, 𝐥𝐢𝐦 𝐥𝐢𝐦 𝐟𝐧 (𝐭) = 𝐥𝐢𝐦 𝐥𝐢𝐦 𝐀 𝐧 .


𝐱→𝐜 𝐧→∞ 𝐧→∞ 𝐱→𝐜

Proof: Let 𝝐 > 𝟎 be given. Since {𝒇𝒏 } converges uniformly on 𝑨, there exists a positive integer

𝑵 such that 𝒏 ≥ 𝑵, 𝒎 ≥ 𝑵, 𝒙 ∈ 𝑨  |𝒇𝒏 (𝒙) − 𝒇𝒎 (𝒙)| < 𝝐. (6.3.1)

Letting → 𝒄 , we get

|𝑨𝒏 − 𝑨𝒎 | < 𝝐 ∀ 𝒏 ≥ 𝑵, 𝒎 ≥ 𝑵. (6.3.2)

Therefore by Cauchy’s general principle of real sequences {𝑨𝒏 } converges to say 𝑨.

i.e. 𝐥𝐢𝐦 𝑨𝒏 = 𝑨.
𝒏→∞

Since {𝑨𝒏 } converges to 𝑨 and {𝒇𝒏 } converges uniformly to 𝒇, there exists a positive integer 𝑵

such that
𝝐
|𝑨𝒏 − 𝑨| < (6.3.3)
𝟑

𝝐
and |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < , for all 𝒙 ∈ 𝑨. (6.3.4)
𝟑

Again since 𝐥𝐢𝐦 𝐟𝐧 (𝐱) = 𝐀 𝐧 , there exists 𝜹 > 𝟎 such that


𝐱→𝐜

𝝐
𝒅(𝒙, 𝒄) < 𝜹  |𝒇𝒏 (𝒙) − 𝑨𝒏 | < . (6.3.5)
𝟑

Hence for all 𝒙 satisfying, 𝒅(𝒙, 𝒄) < 𝜹, we have

|𝒇(𝒙) − 𝑨| = |𝒇(𝒙) − 𝒇𝒏 (𝒙) + 𝒇𝒏 (𝒙) − 𝑨𝒏 + 𝑨𝒏 − 𝑨|

≤ |𝒇(𝒙) − 𝒇𝒏 (𝒙)| + |𝒇𝒏 (𝒙) − 𝑨𝒏 | + |𝑨𝒏 − 𝑨|


𝝐 𝝐 𝝐
< 𝟑 + 𝟑 + 𝟑 = 𝝐, by (6.3.3), (6.3.4) and (6.3.5).

76
Thus we have proved that for a given 𝝐 > 𝟎, there exists a 𝜹 > 𝟎 such that

|𝒙 − 𝒄| < 𝜹  |𝒇(𝒙) − 𝑨| < 𝝐 .

 𝐥𝐢𝐦 𝒇(𝒙) = 𝑨 = 𝐥𝐢𝐦 𝑨𝒏 or 𝐥𝐢𝐦 𝐥𝐢𝐦 𝐟𝐧 (𝐱) = 𝐥𝐢𝐦 𝐥𝐢𝐦 𝐀 𝐧 , which proves the theorem.
𝐱→𝐜 𝐧→∞ 𝐱→𝐜 𝐧→∞ 𝐧→∞ 𝐱→𝐜

Theorem 6.3.2: Let {𝒇𝒏 } be a sequence of continuous functions on 𝑨 ⊂ 𝐑, and if 𝒇𝒏 → 𝒇

uniformly on 𝑨, then 𝒇 is continuous on 𝑨.

Proof: This is an immediate corollary of the previous theorem.

Let 𝒄 ∈ 𝑨. to prove the theorem it suffices to prove that 𝒇 is continuous at 𝒄.

Since each 𝒇𝒏 is continuous on 𝑨, it follows that each 𝒇𝒏 is continuous at 𝒄.

Let 𝝐 > 𝟎 be given. Since {𝒇𝒏 } converges uniformly to 𝒇 on 𝑨, there exists a positive integer 𝑵
𝝐
such that 𝒏 ≥ 𝑵 , 𝒙 ∈ 𝑨 ⟹ |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝟑. (6.3.6)

In particular for 𝒙 = 𝒄 , 𝒏 = 𝑵, we have


𝝐
|𝒇𝑵 (𝒄) − 𝒇(𝒄)| < . (6.3.7)
𝟑

Again since 𝒇𝑵 is continuous at 𝒄, there exists 𝜹 > 𝟎 such that


𝝐
|𝒙 − 𝒄| < 𝜹  |𝒇𝑵 (𝒙) − 𝒇𝑵 (𝒄)| < . (6.3.8)
𝟑

 if |𝒙 − 𝒄| < 𝜹 we have

|𝒇(𝒙) − 𝒇(𝒄)| = |𝒇(𝒙) − 𝒇𝑵 (𝒙) + 𝒇𝑵 (𝒙) − 𝒇𝑵 (𝒄) + 𝒇𝑵 (𝒄) − 𝒇(𝒄)|

𝝐 𝝐 𝝐
≤ |𝒇(𝒙) − 𝒇𝑵 (𝒙)| + |𝒇𝑵 (𝒙) − 𝒇𝑵 (𝒄)| + |𝒇𝑵 (𝒄) − 𝒇(𝒄)| < + + = 𝝐,
𝟑 𝟑 𝟑

by (6.3.6), (6.3.7) and (6.3.8). This proves that 𝒇 is continuous at 𝒄.

The converse of the above theorem is not true in general but there is a case in which converse is

true.

77
Theorem 6.3.3: Let 𝑨 be any compact set in 𝑹. Let {𝒇𝒏 } be a sequence of continuous functions

converges point wise to a continuous function 𝒇 on 𝑨. If 𝒇𝒏 (𝒙) ≥ 𝒇𝒏+𝟏 (𝒙) for 𝒏 = 𝟏, 𝟐, 𝟑 … and

for every 𝒙 ∈ 𝑨, then 𝒇𝒏 → 𝒇 converges uniformly on 𝑨.

Proof: Let 𝑭𝒏 (𝒙) = 𝒇𝒏 (𝒙) − 𝒇(𝒙). Since 𝒇𝒏 and 𝒇 are continuous, 𝑭 is also continuous. Also

since 𝒇𝒏 → 𝒇, we have 𝑭𝒏 = 𝒇𝒏 − 𝒇 → 𝟎.

Furhter 𝑭𝒏 − 𝑭𝒏+𝟏 = (𝒇𝒏 − 𝒇) − (𝒇𝒏+𝟏 − 𝒇) = 𝒇𝒏 − 𝒇𝒏+𝟏 and so

𝒇𝒏 ≥ 𝒇𝒏+𝟏 ⇒ 𝑭𝒏 ≥ 𝑭𝒏+𝟏 .

We claim that 𝑭𝒏 ⟶ 𝟎 uniformly on 𝑨.

Let 𝝐 > 𝟎 be given. Since 𝑭𝒏 ⟶ 𝟎 on 𝑨, therefore for each 𝒙 ∈ 𝑨, here exists a positive integer
𝝐
𝑵𝒙 such that |𝑭𝑵𝒙 (𝒙)| < 𝟐. Since 𝑭𝑵𝒙 is continuous , there exists an open set 𝑼(𝒙) containing 𝒙

such that for every 𝒄 ∈ 𝑼𝒙 , we have


𝝐
|𝑭𝑵𝒙 (𝒄) − 𝑭𝑵𝒙 (𝒙)| < 𝟐.

𝝐
Then |𝑭𝑵𝒙 (𝒄)| − |𝑭𝑵𝒙 (𝒙)| ≤ |𝑭𝑵𝒙 (𝒙) − 𝑭𝑵𝒙 (𝒄)| < 𝟐 or

𝝐 𝝐 𝝐
|𝑭𝑵𝒙 (𝒙)| < |𝑭𝑵𝒙 (𝒄)| + 𝟐 < 𝟐 + 𝟐 = 𝝐. (6.3.9)

Now 𝑭𝒏 ≥ 𝑭𝒏+𝟏 for each 𝒏 

𝑭𝑵𝒙 (𝒙) ≥ 𝑭𝒏 (𝒙) 𝒇𝒐𝒓 𝒏 ≥ 𝑵𝒙 (6.3.10)

From (6.3.9) and (6.3.10), we have

|𝑭𝒏 (𝒄)| < 𝝐, (6.3.11)

for every 𝒄 ∈ 𝑼𝒙 and 𝒏 ≥ 𝑵𝒙 .

Since 𝑨 is compact, the collection {𝑼𝒙 }𝒙∈𝑨 which forms an open cover of 𝑨 has a finite

subcover. So there exist points 𝒙𝟎 , 𝒙𝟏 , … , 𝒙𝒌 such that 𝑨 ⊂ 𝑼𝒙𝟏 ∪ 𝑼𝒙𝟐 ∪ … ∪ 𝑼𝒙𝒌 .

Let 𝑵𝟎 = 𝐦𝐚𝐱{𝑵𝒙𝟎 , 𝑵𝒙𝟏 , … , 𝑵𝒙𝒌 }. Then from (6.3.9) and (6.3.11), we have

78
|𝑭𝒏 (𝒄)| < 𝝐 for every 𝒄 ∈ 𝑨 and 𝒏 ≥ 𝑵𝟎 . Hence 𝑭𝒏 → 𝟎 uniformly on 𝑨.

i.e. 𝒇𝒏 → 𝒇 uniformly on 𝑨.

This completes the proof of the theorem.

In the above theorem compactness is necessary.

𝟏
For example: if 𝒇𝒏 (𝒙) = 𝒏𝒙+𝟏 (𝟎 < 𝒙 < 𝟏) , then the sequence {𝒇𝒏 } is not uniformly

𝟏
convergent on (𝟎, 𝟏). Since 𝒇𝒏+𝟏 (𝒙) = (𝒏+𝟏)(𝒙+𝟏) , we have 𝒇𝒏 ≥ 𝒇𝒏+𝟏 for n=1,2,3,…

𝟏
Also 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝐥𝐢𝐦 𝒏𝒙+𝟏 = 𝟎, for every 𝒙 ∈ (𝟎, 𝟏). Thus 𝒇𝒏 → 𝟎 on (𝟎, 𝟏).
𝒏→∞ 𝒏→∞

𝟏 𝟏 𝟏
If we choose 𝝐 = 𝟐, and if there is a positive integer 𝑵 such that |𝒇𝒏 (𝒙) − 𝟎| < 𝒏𝒙+𝟏 < 𝟐 for

𝟏 𝟏 𝟏
every 𝒙 ∈ (𝟎, 𝟏) and 𝒏 ≥ 𝑵 then for 𝒙 = 𝒏, we have 𝟐 < 𝟐 which is not true.

Thus we have {𝒇𝒏 } is not uniformly convergent on (𝟎, 𝟏).

Note: Let us consider example, 𝒇𝒏 (𝒙) = 𝒏𝒙(𝟏 − 𝒙)𝒏 , 𝒙 ∈ [𝟎, 𝟏].

Here 𝒇(𝒙) = 𝐥𝐢𝐦 𝒇𝒏 (𝒙) = 𝟎, when 𝟎 < 𝒙 < 𝟏. Also 𝒇𝒏 (𝒙) = 𝟎 when 𝒙 = 𝟎 𝒐𝒓 𝟏.
𝐧→∞

Hence 𝒇(𝒙) = 𝟎 for all values of 𝒙 ∈ [𝟎, 𝟏]. The function 𝒇(𝒙) is continuous for all values of

𝒙 ∈ [𝟎, 𝟏]. But the sequence {𝒇𝒏 (𝒙)} is not uniformly convergent on [𝟎, 𝟏] .

6.4 Uniform Convergence and Differentiation

Theorem 6.4.1: Let {𝒇𝒏 } be a sequence of differentiable functions defined on [𝒂, 𝒃] and such

that {𝒇𝒏 (𝒙)} converges for some point 𝒙 ∈ [𝒂, 𝒃]. If {𝒇′𝒏 } converges uniformly on [𝒂, 𝒃], then

{𝒇𝒏 } converges uniformly on [𝒂, 𝒃] to a function 𝒇 and 𝒇′ (𝒙) = 𝐥𝐢𝐦 𝒇′𝒏 (𝒙) for all 𝒙 ∈ [𝒂, 𝒃].
𝒏→∞

Proof: Let 𝝐 > 𝟎 be given. Since {𝒇𝒏 } converges and {𝒇′𝒏 } converges uniformly to 𝒇on [𝒂, 𝒃],

there exists a positive integer 𝑵 such that for all 𝒏 ≥ 𝑵, 𝒎 ≥ 𝑵 ,we have

79
𝝐
|𝒇𝒏 (𝒄) − 𝒇𝒎 (𝒄)| < (6.4.1)
𝟐

𝝐
and |𝒇′𝒏 (𝒙) − 𝒇′𝒎 (𝒄)| < (6.4.2)
𝟐(𝒃−𝒂)

for all 𝒙 ∈ [𝒂, 𝒃].

Applying the mean value theorem of differential calculus to function 𝒇𝒏 − 𝒇𝒎 , we obtain

𝒇𝒏 (𝒙) − 𝒇𝒎 (𝒙) − 𝒇𝒏 (𝒚) + 𝒇𝒎 (𝒚) = (𝒙 − 𝒚)[𝒇′𝒏 (𝝃) − 𝒇′𝒎 (𝝃)], for all 𝒙, 𝒚 ∈ [𝒂, 𝒃]

And for some 𝝃, 𝒙 < 𝝃 < 𝒚 and 𝒏 ≥ 𝑵, 𝒎 ≥ 𝑵.

Hence |𝒇𝒏 (𝒙) − 𝒇𝒎 (𝒙) − 𝒇𝒏 (𝒚) + 𝒇𝒎 (𝒚)| = |𝒙 − 𝒚||𝒇′𝒏 (𝝃) − 𝒇′𝒎 (𝝃)|

|𝒙−𝒚|𝝐 𝝐
< 𝟐(𝒃−𝒂) < 𝟐, (6.4.3)

for all 𝒏, 𝒎 ≥ 𝑵 and 𝒙, 𝒚 ∈ [𝒂, 𝒃].

The above inequality follows from (6.4.2) and the fact that |𝒙 − 𝒚| ≤ (𝒃 − 𝒂).

Now consider |𝒇𝒏 (𝒙) − 𝒇𝒎 (𝒙)| = |𝒇𝒏 (𝒙) − 𝒇𝒎 (𝒙) − 𝒇𝒏 (𝒄) + 𝒇𝒎 (𝒄) + 𝒇𝒏 (𝒄) − 𝒇𝒎 (𝒄)|

≤ |𝒇𝒏 (𝒙) − 𝒇𝒎 (𝒙) − 𝒇𝒏 (𝒄) + 𝒇𝒎 (𝒄)| + |𝒇𝒏 (𝒄) − 𝒇𝒎 (𝒄)|


𝝐 𝝐
 |𝒇𝒏 (𝒙) − 𝒇𝒎 (𝒙)| < + = 𝝐, follows from (6.4.1) and (6.4.3).
𝟐 𝟐

Thus {𝒇𝒏 } converges uniformly to a function 𝒇on [𝒂, 𝒃] and so 𝒇(𝒙) = 𝐥𝐢𝐦 𝒇𝒏 (𝒙), 𝒂 ≤ 𝒙 ≤ 𝒃,
𝒏→∞

hence the first part is proved.

Let 𝒙 ∈ [𝒂, 𝒃] be fixed.

𝒇𝒏 (𝒚)−𝒇𝒏 (𝒙) 𝒇(𝒚)−𝒇(𝒙)


Set 𝝓𝒏 (𝒙) = and 𝝓(𝒚) = , for 𝒂 ≤ 𝒚 ≤ 𝒃, 𝒚 ≠ 𝒙. (6.4.4)
𝒚−𝒙 𝒚−𝒙

𝒇𝒏 (𝒚)−𝒇𝒏 (𝒙)
Then 𝐥𝐢𝐦 𝝓𝒏 (𝒚) = 𝐥𝐢𝐦 = 𝒇′𝒏 (𝒙), for 𝒏 = 𝟏, 𝟐, 𝟑, … (6.4.5)
𝒚→𝒙 𝒚→𝒙 𝒚−𝒙

Now for 𝒏 ≥ 𝑵, 𝒎 ≥ 𝑵, we have

|𝒇𝒏 (𝒚)−𝒇𝒏 (𝒙)+𝒇𝒎 (𝒚)−𝒇𝒎 (𝒙)| 𝝐


|𝝓𝒏 (𝒚) − 𝝓𝒎 (𝒚)| = < 𝟐(𝒃−𝒂) , follows from (6.4.3).
|𝒚−𝒙|

80
i.e. {𝝓𝒏 } converges uniformly for 𝒚 ≠ 𝒙. Since {𝒇𝒏 } converges to 𝒇, we conclude from (6.4.4)

𝒇𝒏 (𝒚)−𝒇𝒏 (𝒙) 𝒇(𝒚)−𝒇(𝒙)


that 𝐥𝐢𝐦 𝝓𝒏 (𝒚) = 𝐥𝐢𝐦 = = 𝝓(𝒚), (6.4.6)
𝒏→∞ 𝒏→∞ 𝒚−𝒙 𝒚−𝒙

uniformly for 𝐚 ≤ 𝐲 ≤ 𝐛, 𝐲 ≠ 𝐱.

Finally applying theorem 2.2.1 to {𝝓𝒏 } with 𝑨𝒏 = 𝒇′𝒏 (𝒙), from (6.4.5) and (6.4.6), it follows that

𝐥𝐢𝐦 𝝓(𝒚) = 𝐥𝐢𝐦 𝒇′𝒏 (𝒙)


𝒚→𝒙 𝒏→∞

𝒇(𝒚)−𝒇(𝒙)
 𝐥𝐢𝐦 = 𝐥𝐢𝐦 𝒇′𝒏 (𝒙) or 𝒇′ (𝒙) = 𝐥𝐢𝐦 𝒇′𝒏 (𝒙) , for every 𝒙 ∈ [𝒂, 𝒃].
𝒚→𝒙 𝒚−𝒙 𝒏→∞ 𝒏→∞

This completes the proof of the theorem.

6.5 Uniform Convergence and Integration

Theorem 6.5.1:Let 𝜶 be monotonically increasing on [𝒂, 𝒃]. Suppose 𝒇𝒏 ∈ 𝕽(𝜶) on [𝒂, 𝒃], for

n=1,2,3,…, and suppose 𝒇𝒏 ⟶ 𝒇 uniformly on [𝒂, 𝒃]. Then 𝒇 ∈ 𝕽(𝜶) on [𝒂, 𝒃]

𝒃 𝒃
and ∫𝒂 𝒇 𝒅𝜶 = 𝐥𝐢𝐦 ∫𝒂 𝒇𝒏 𝒅𝜶.
𝒏→∞

Proof: Let 𝝐 > 𝟎 be given. Let us choose 𝜼 > 𝟎 such that


𝝐
𝜼[𝜶(𝒃) − 𝜶(𝒂)] < 𝟑 . (6.5.1)

Since {𝒇𝒏 } converges uniformly to 𝒇 on [𝒂, 𝒃], there exists an integer 𝑵 > 𝟎 such that

𝒏 ≥ 𝑵 and 𝒙 ∈ [𝒂, 𝒃] ⟹

|𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝜼. (6.5.2)

Since 𝒇𝒏 ∈ 𝕽(𝜶) on [𝒂, 𝒃], there exists a partition 𝑷 = {𝒙𝟎 = 𝒂, 𝒙𝟏 , … , 𝒙𝒏 = 𝒃} of [𝒂, 𝒃] such

that
𝝐
𝑼(𝑷, 𝒇𝒏 , 𝜶) − 𝑳(𝑷, 𝒇𝒏 , 𝜶) < 𝟑. (6.5.3)

The inequality (6.5.2) may be written in the form

81
𝒇𝒏 (𝒙) − 𝜼 < 𝒇(𝒙) < 𝒇𝒏 (𝒙) + 𝜼.

Hence we have, 𝒇𝒏 (𝒙) < 𝒇(𝒙) + 𝜼 ⟹ 𝑳(𝑷, 𝒇𝒏 , 𝜶) < 𝑳(𝑷, 𝒇, 𝜶) + 𝜼[𝜶(𝒃) − 𝜶(𝒂)].

Then by (6.4.1) that


𝝐
𝑳(𝑷, 𝒇𝒏 , 𝜶) < 𝑳(𝑷, 𝒇, 𝜶) + 𝟑. (6.5.4)

𝝐
Similarly, 𝒇(𝒙) < 𝒇𝒏 (𝒙) + 𝜼 ⟹ 𝑼(𝑷, 𝒇, 𝜶) ≤ 𝑼(𝑷, 𝒇𝒏 , 𝜶) + 𝟑. (6.5.5)

Adding (6.5..4) and (6.5.5), we get

𝟐𝝐
𝑼(𝑷, 𝒇, 𝜶) + 𝑳(𝑷, 𝒇𝒏 , 𝜶) ≤ 𝑼(𝑷, 𝒇𝒏 , 𝜶) + 𝑳(𝑷, 𝒇, 𝜶 ) + .
𝟑
𝟐𝝐
Or 𝑼(𝑷, 𝒇, 𝜶) − 𝑳(𝑷, 𝒇, 𝜶 ) ≤ 𝑼(𝑷, 𝒇𝒏 , 𝜶) − 𝑳(𝑷, 𝒇𝒏 , 𝜶) + 𝟑

𝝐 𝟐𝝐
< 𝟑+ = 𝝐, follows from (2.3.3).
𝟑

Hence we have 𝒇 ∈ 𝕽(𝜶) on [𝒂, 𝒃].

To prove the second assertion, let 𝝐 > 𝟎 be given. Since {𝒇𝒏 } converges uniformly to 𝒇 on

[𝒂, 𝒃], there exists a positive integer 𝑵 such that

𝒏 ≥ 𝑵 𝒂𝒏𝒅 𝒙 ∈ [𝒂, 𝒃] ⟹ |𝒇𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐. (6.5.6)

Hence for 𝒏 ≥ 𝑵, we have

𝒃 𝒃 𝒃
|∫ 𝒇 𝒅𝜶 − ∫ 𝒇𝒏 𝒅𝜶| = |∫ (𝒇 − 𝒇𝒏 )𝒅𝜶 |
𝒂 𝒂 𝒂

𝒃 𝒃
≤ ∫ |𝒇 − 𝒇𝒏 | 𝒅𝜶 < 𝝐 ∫ 𝒅𝜶 = 𝝐[𝜶(𝒃) − 𝜶(𝒂)].
𝒂 𝒂

𝒃 𝒃
Since 𝝐 was arbitrary, ∫𝒂 𝒇 𝒅𝜶 = 𝐥𝐢𝐦 ∫𝒂 𝒇𝒏 𝒅𝜶 .
𝒏→∞

Corollary 6.5.1(term by term integration):

If 𝒇𝒏 ∈ 𝕽(𝜶) on [𝒂, 𝒃] and if 𝒇(𝒙) = ∑∞


𝒏=𝟏 𝒇𝒏 (𝒙) (𝒂 ≤ 𝒙 ≤ 𝒃), the series converging

𝒃 𝒃
uniformly on [𝒂, 𝒃], then ∫𝒂 𝒇 𝒅𝜶 = ∑∞
𝒏=𝟏 ∫𝒂 𝒇𝒏 𝒅𝜶.

82
Proof: Let 𝑺𝒏 = 𝒇𝟏 + 𝒇𝟐 + ⋯ + 𝒇𝒏 , for each 𝒏. Since the sum of a finite number of R-S

integrable functions is R-S integrable function, it follows that 𝑺𝒏 ∈ 𝕽(𝜶) on [𝒂, 𝒃] for each

fixed 𝒏. Since the series ∑∞


𝒏=𝟏 𝒇𝒏 (𝒙) converges uniformly to (𝒙) , the sequence {𝑺𝒏 } converges

uniformly to 𝒇 on [𝒂, 𝒃]. Hence by the above theorem, 𝒇 ∈ 𝕽(𝜶) on [𝒂, 𝒃].

𝒃 𝒃 𝒃
∫𝒂 [∑∞
𝒏=𝟏 𝒇𝒏 ] 𝒅𝜶 = ∫𝒂 𝒇 𝒅𝜶 = 𝐥𝐢𝐦 ∫𝒂 𝑺𝒏 𝒅𝜶
𝒏→∞

𝒃 𝒃 𝒃
= 𝒍𝒊𝒎𝒏→∞ ∫𝒂 [∑𝒏𝒊=𝟏 𝒇𝒊 ] 𝒅𝜶 = 𝒍𝒊𝒎𝒏→∞ ∑𝒏𝒊=𝟏 ∫𝒂 𝒇𝒊 𝒅𝜶 = ∑∞
𝒏=𝟏[∫𝒂 𝒇𝒏 𝒅𝜶] .

6.6 Summary

1. If {𝒇𝒏 } is a sequence of continuous functions and 𝒇𝒏 → 𝒇 uniformly on 𝑨, then 𝒇 is

continuous on 𝑨.

2. Let {𝒇𝒏 } be a sequence of continuous functions on 𝑨 converges point wise to a continuous

function 𝒇 on 𝑨. If 𝒇𝒏 (𝒙) ≥ 𝒇𝒏+𝟏 (𝒙) for 𝒏 = 𝟏, 𝟐, 𝟑 … and for every 𝒙 ∈ 𝑨, then 𝒇𝒏 → 𝒇

converges uniformly on 𝑨.

3. Let {𝒇𝒏 } be a sequence of differentiable functions defined on [𝒂, 𝒃] and such that {𝒇𝒏 (𝒙)}

converges for some point 𝒄 ∈ [𝒂, 𝒃]. If {𝒇′𝒏 } converges uniformly on [𝒂, 𝒃], then {𝒇𝒏 }

converges uniformly on [𝒂, 𝒃] to a function 𝒇 and 𝒇′ (𝒙) = 𝐥𝐢𝐦 𝒇′𝒏 (𝒙) for all 𝒙 ∈ [𝒂, 𝒃].
𝒏→∞

4. If 𝒇𝒏 ∈ 𝕽(𝜶) on [𝒂, 𝒃], for n=1,2,3,…, and suppose 𝒇𝒏 ⟶ 𝒇 uniformly on [𝒂, 𝒃]. Then 𝒇 ∈

𝕽(𝜶) on [𝒂, 𝒃] and limit and integral can be interchanged.

83
5. If 𝒇𝒏 ∈ 𝕽(𝜶) on [𝒂, 𝒃] and if 𝒇(𝒙) = ∑∞
𝒏=𝟏 𝒇𝒏 (𝒙) (𝒂 ≤ 𝒙 ≤ 𝒃), the series converging

𝒃 𝒃
uniformly on [𝒂, 𝒃], then ∫𝒂 𝒇 𝒅𝜶 = ∑∞
𝒏=𝟏 ∫𝒂 𝒇𝒏 𝒅𝜶.

6.7 Keywords

Uniform convergence, continuity, differentiability, integrability, term by term integration

6.8 Exercises

1. Show that uniform limit of a sequence of continuous is continuos.

2. Show that for a decreasing sequence of continuous functions on a compact set converges

point wise to a continuous function, point wise convergence is the uniform convergence.

3. Show that uniform limit of a sequence of R-S integrable functions is R-S integrable.

Further show that limit process and R-S integral can be interchanged.

4. Show that a uniformly convergent series of R-S integrable functions can be integrated term

by term.

Solutions:
1. See theorem 6.3.2.

2. See theorem 6.3.3.

3. See theorem 6.5.1.

4. See corollary 6.5.1.

6.9 References

1. Richard R Goldberg, Methods of Real analysis, IBH publishing,New Delhi,1970

84
2. Walter Rudin, Principles of Mathematical Analysis,McGraw-Hill 3rd edition, 1976.
3. Robert G. Bartle, Donald R Sherbert, Introduction to Real Analysis, Wiley India, Third
edition, 2005.
4. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing,1974.

85
UNIT 7
EQUICONTINUOUS FAMILIES OF FUNCTIONS,
STONE- WEIERSTRASS THEOREM

7.1 Main Objectives

7.2 Introduction

7.3 Equicontinuous families of functions

7.4 Stone-Weierstrass Approximation Theorem

7.5 Summary

7.6 Key Words

7.7 Exercises

7.8 References

86
UNIT 7
EQUICONTINUOUS FAMILIES OF FUNCTIONS,
STONE WEIERSTRASS THEOREM

7.1 Main Objectives


After going through this unit, students will be able to;
(i) Know there exist functions which are continuous everywhere but
differentiable nowhere.
(ii) Know that a continuous function is uniformly approximated by a sequence of
polynomials.

7.2 Introduction
In the early nineteenth century most mathematicians believed that a continuous

function has derivatives at a significant set of points. But in 1872, in his presentation

Berlin academy, Karl Weierstrass shocked the mathematics world by proving this

conjecture is false. He presented a function which is continuous everywhere but

differentiable nowhere. Because of the contribution of several mathematicians (after the

invention of Weierstrass function), we have a large class of continuous nowhere

differentiable functions which is of second category according to Baire. In this unit we

shall study a function which is continuous but nowhere differentiable. Later we shall

study the famous Weierstrass approximation theorem, which says that any continuous

function on a closed and bounded interval is uniformly approximated by a sequence of

polynomials.

87
7.3. Equicontinuous families of functions
Theorem 7.3.1: There exists a real continuous function on the real line which is nowhere
differentiable.
𝒙 𝒇𝒐𝒓 𝟎 ≤ 𝒙 ≤ 𝟏
Proof: Let us define 𝒇(𝒙) = { (7.3.1)
𝟐 − 𝒙 𝒇𝒐𝒓 𝟏 ≤ 𝒙 ≤ 𝟐
and let 𝒇(𝒙) be extended to all real numbers such that 𝒇(𝒙 + 𝟐) = 𝒇(𝒙).
It is clear that 𝒇 is continuous on 𝑹𝟏 .
𝟑 𝒏
Set 𝑭(𝒙) = ∑∞ 𝒏
𝒏=𝟏 (𝟒) 𝒇(𝟒 𝒙). (7.3.2)
𝟑 𝒏 𝟑 𝐧
The definition of 𝒇 shows that 𝟎 ≤ 𝒇(𝒙) ≤ 𝟏, for all 𝒙 ∈ 𝑹. Hence |(𝟒) 𝒇(𝟒𝒏 𝒙)| ≤ ∑∞
𝟏 (𝟒) .

𝟑 𝒏 𝟑
But the series ∑∞
𝟏 (𝟒) is convergent, being a geometric series with constant ratio < 𝟏. It
𝟒

𝟑 𝒏
follows by theorem 5.4.3 that the series ∑∞ 𝒏 𝟏
𝒏=𝟏 (𝟒) 𝒇(𝟒 𝒙) converges uniformly to 𝑭 on 𝑹 . By

theorem 6.3.2, 𝑭 is continuous on 𝑹𝟏 .


Let 𝒄 be a fixed real number and let k be a fixed positive integer. Then by property of real
numbers, there exists an integer 𝒎 such that
𝒎 ≤ 𝟒𝒌 𝒄 ≤ 𝒎 + 𝟏. (7.3.3)
Set 𝜶𝒌 = 𝟒−𝒌 𝒎, 𝜷𝒌 = 𝟒−𝒌 (𝒎 + 𝟏). (7.3.4)
Consider the numbers 𝟒𝒏 𝜷𝒌 and 𝟒𝒏 𝜶𝒌 . If 𝒏 > 𝒌, then 𝟒𝒏 𝜷𝒌 − 𝟒𝒏 𝜶𝒌 is an even integer. If 𝒏 =
𝒌, then 𝟒𝒏 𝜷𝒌 = 𝒎 + 𝟏 and 𝟒𝒏 𝜶𝒌 = 𝒎 so that their difference is 𝟏. If 𝒏 < 𝒌, then there is no
integer between 𝟒𝒏 𝜷𝒌 and 𝟒𝒏 𝜶𝒌 . It follows that
𝟎 𝒊𝒇 𝒏 > 𝒌
|𝒇(𝟒𝒏 𝜷𝒌 ) − 𝒇(𝟒𝒏 𝜶𝒌 )| = { . (7.3.5)
𝟒𝒏−𝒌 𝒊𝒇 𝒏 ≤ 𝒌
Then by (7.3.2) and (7.3.5), we have
𝟑 𝒏
𝑭(𝜷𝒌 ) − 𝑭(𝜶𝒌 ) = ∑𝒌𝒏=𝟎 (𝟒) [𝒇(𝟒𝒏 𝜷𝒌 ) − 𝒇(𝟒𝒏 𝜶𝒌 )], since all terms of the series are zero for
𝟑 𝒏 𝟑 𝒏
𝒏 > 𝒌 by (3.3.5). Hence |𝑭(𝜷𝒌 ) − 𝑭(𝜶𝒌 )| ≥ (𝟒) − ∑𝒌−𝟏
𝒏=𝟎 (𝟒) 𝟒
𝒏−𝒎

𝟑 𝒌 𝟑𝒌 − 𝟏 𝟏 𝟑 𝒌 𝟏
=( ) − = ( ) +
𝟒 𝟐. 𝟒𝒌 𝟐 𝟒 𝟐. 𝟒𝒌
𝑭(𝜷𝒌 )−𝑭(𝜶𝒌 ) 𝟏
or | | > 𝟐 𝟑𝒌 , (7.3.6)
𝜷𝒌 −𝜶𝒌

88
since|𝜷𝒌 − 𝜶𝒌 | = 𝟒−𝒌 .
 𝜷𝒌 − 𝜶𝒌 → 𝟎 𝒂𝒔 𝒌 → ∞ and 𝜶𝒌 ≤ 𝒄 ≤ 𝜷𝒌 , it follows that 𝒇 is not differentiable at 𝒄.

7.4. Stone-Weierstrass Approximation Theorem


Definition 7.4.1: A polynomial function is a function 𝑷 defined by 𝑷(𝒙) = 𝒂𝟎 𝒙𝒏 + 𝒂𝟏 𝒙𝒏−𝟏 +
⋯ + 𝒂𝒏 , ∀ 𝒙 ∈ 𝑹, where 𝒏 is a non-negative integer and 𝒂𝟎 , 𝒂𝟏 , … , 𝒂𝒏 are real numbers , 𝒏 is
called the degree of the polynomial.
Definition 7.4.2: For any 𝒇 ∈ 𝑪[𝟎, 𝟏], we define a sequence of polynomials {𝑩𝒏 }∞
𝒏=𝟏 as follows:
𝒏
𝒏 𝒏
𝑩𝒏 (𝒙) = ∑ ( ) 𝒙𝒌 (𝟏 − 𝒙)𝒏−𝒌 𝒇 ( ) (𝟎 ≤ 𝒙 ≤ 𝟏; 𝒏 𝒊𝒔 𝒂𝒏 𝒊𝒏𝒕𝒆𝒈𝒆𝒓),
𝒌 𝒌
𝒌=𝟎
𝒏 𝒏!
where ( ) = (𝒌! . The polynomial 𝑩𝒏 (𝒙) is called the Bernstein polynomial for 𝒇.
𝒌 (𝒏−𝒌)!)

The following theorem due to Weierstrass proves that every function in 𝑪[𝒂, 𝒃] is uniformly
approximated by polynomials.
Theorem 7.4.1(The Weierstrass approximation theorem):
Let 𝒇 be a continuous complex function on [𝒂, 𝒃], then there exists a sequence of polynomials
𝑷𝒏 such that 𝐥𝐢𝐦 𝑷𝒏 (𝒙) = 𝒇(𝒙) uniformly on [𝒂, 𝒃]. If 𝒇 is real then 𝑷𝒏 may be taken real.
𝒏→∞

Proof: Part I: It is sufficient to prove the theorem for [𝒂, 𝒃] = [𝟎, 𝟏]. For if 𝒇 ∈ 𝑪[𝒂, 𝒃],
𝒂𝒏𝒅 𝝐 > 𝟎, we must find a polynomial 𝑷𝒏 such that
|𝑷𝒏 (𝒙) − 𝒇(𝒙)| < 𝝐 (𝒂 ≤ 𝒙 ≤ 𝒃). (7.4.1)
Let 𝒈(𝒙) = 𝒇(𝒂 + (𝒃 − 𝒂)𝒙) (𝟎 ≤ 𝒙 ≤ 𝟏).
Then 𝒈(𝟎) = 𝒇(𝒂) 𝒂𝒏𝒅 𝒈(𝟏) = 𝒇(𝒃). And 𝒈 is continuous on [𝟎, 𝟏]. Thus by our assumption
there is a polynomial 𝑷𝒏 (𝒙) such that |𝒈(𝒙) − 𝑷𝒏 (𝒙)| < 𝝐 (𝟎 ≤ 𝒙 ≤ 𝟏).
𝒙−𝒂 𝒙−𝒂 (𝒃−𝒂)(𝒙−𝒂)
Set 𝒕 = , then 𝒈(𝒕) = 𝒈 ( ) = 𝒇 (𝒂 + ) = 𝒇(𝒙).
𝒃−𝒂 𝒃−𝒂 𝒃−𝒂
(𝒙−𝒂)
We have |𝒇(𝒙) − 𝑷𝒏 ( 𝒃−𝒂 )| < 𝝐 (𝒂 ≤ 𝒙 ≤ 𝒃).
𝒙−𝒂
If we set 𝑸𝒏 (𝒙) = 𝑷𝒏 (𝒃−𝒂), then by Binomial theorem, 𝑸𝒏 is a polynomial because

𝑷𝒏 is so. Therefore we have |𝒇(𝒙) − 𝑸𝒏 (𝒙)| < 𝝐.


PART II: We assume that 𝒇(𝟎) = 𝒇(𝟏) = 𝟎.

89
For if the theorem is proved for this case then for 𝒈(𝒙) = 𝒇(𝒙) − 𝒇(𝟎) − 𝒙[𝒇(𝟏) −
𝒇(𝟎)] (𝟎 ≤ 𝒙 ≤ 𝟏), we have 𝒈(𝟎) = 𝒈(𝟏) = 𝟎 and if 𝒈 is the uniform limit of a sequence of
polynomials then 𝒇 is also the uniform limit of the sequence of polynomials, since 𝒇 − 𝒈 is a
polynomial.
Let 𝒇(𝒙) = 𝟎 for 𝒙 ∉ [𝟎, 𝟏]. Then 𝒇 is continuous on 𝑹𝟏 .
Set 𝝓𝒏 (𝒙) = 𝒂𝒏 (𝟏 − 𝒙𝟐 )𝒏 (𝒏 = 𝟏, 𝟐, … , ), (7.4.2)
Where, 𝒂𝒏 are so chosen that
𝟏
∫−𝟏 𝝓𝒏 (𝒙)𝒅𝒙 = 𝟏 (𝒏 = 𝟏, 𝟐, 𝟑, …). (7.4.3)
𝟏
𝟏 𝟐 )𝒏 𝟏 𝟐 )𝒏
Since ∫−𝟏(𝟏 −𝒙 𝒅𝒙 = 𝟐 ∫𝟎 (𝟏 −𝒙 𝒅𝒙 ≥ 𝟐 ∫𝟎 (𝟏 − 𝒙𝟐 )𝒏 𝒅𝒙
√𝒏

𝟏
√𝒏 𝟒 𝟏
≥ 𝟐 ∫ (𝟏 − 𝒏𝒙𝟐 )𝒅𝒙 = > ,
𝟎 𝟑√ 𝒏 √𝒏
It follows that the bound for 𝒂𝒏 is
𝒂𝒏 < √𝒏. (7.4.4)
In the above we have used the inequality (𝟏 − 𝒙𝟐 )𝒏 ≥ (𝟏 − 𝒏𝒙𝟐 ). This is true because the
function (𝟏 − 𝒙𝟐 )𝒏 − 𝟏 + 𝒏𝒙𝟐 is zero at 𝒙 = 𝟎 and whose derivative is positive in (𝟎, 𝟏).
Let 𝜹 > 𝟎 be given. Then (3.2.4) 
𝝓𝒏 (𝒙) ≤ √𝒏(𝟏 − 𝜹𝟐 )𝒏 (𝜹 ≤ |𝒙| ≤ 𝟏), (7.4.5)
 𝝓𝒏 ⟶ 𝟎 uniformly in 𝜹 ≤ |𝒙| ≤ 𝟏.
𝟏
Set 𝑷𝒏 (𝒙) = ∫−𝟏 𝒇(𝒙 + 𝒚)𝝓𝒏 (𝒚)𝒅𝒚 (𝟎 ≤ 𝒙 ≤ 𝟏). (7.4.6)
By ou assumption about 𝒇, changing the variable, we obtain
𝟏−𝒙 𝟏
𝑷𝒏 (𝒙) = ∫ 𝒇(𝒙 + 𝒚)𝝓𝒏 (𝒚)𝒅𝒚 = ∫ 𝒇(𝒚)𝝓𝒏 (𝒚 − 𝒙)𝒅𝒚,
−𝒙 𝟎

And the integral on rhs is clearly a polynomial in 𝒙. Thus we have a sequence {𝑷𝒏 } of
polynomials, which are real when 𝒇 is real.
Since 𝒇 is continuous, given a positive number 𝝐, there exists a 𝜹 > 𝟎 such that |𝒚 − 𝒙| < 𝜹 
𝝐
|𝒇(𝒚) − 𝒇(𝒙)| < .
𝟐

If 𝑴 = 𝒍𝒖𝒃|𝒇(𝒙)|, the using (3.2.3), (3.2.5) and from the inequality 𝝓𝒏 (𝒙) ≥ 𝟎, we get for 𝟎 ≤
𝒙 ≤ 𝟏,

90
𝟏
|𝑷𝒏 (𝒙) − 𝒇(𝒙)| = | ∫ [𝒇(𝒙 + 𝒚) − 𝒇(𝒙)] 𝝓𝒏 (𝒚)𝒅𝒚 |
−𝟏
𝟏
≤ ∫ |𝒇(𝒙 + 𝒚) − 𝒇(𝒙)| 𝝓𝒏 (𝒚)𝒅𝒚
−𝟏
−𝜹
𝝐 𝜹 𝟏
≤ 𝟐𝑴 ∫ 𝝓𝒏 (𝒚)𝒅𝒚 + ∫ 𝝓𝒏 (𝒚)𝒅𝒚 + 𝟐 𝑴 ∫ 𝝓𝒏 (𝒚)𝒅𝒚
−𝟏 𝟐 −𝜹 𝜹
𝝐
≤ 𝟒𝑴√𝒏(𝟏 − 𝜹𝟐 )𝒏 + 𝟐 < 𝝐, for large 𝒏.

This completes the proof of the theorem.

Corollary 7.4.1: The polynomials are dense in ([𝑎, 𝑏]).


Proof: Let 𝑓: [𝑎, 𝑏] ⟶ ℝ be continuous and let ℎ: [0, 1] ⟶ [𝑎, 𝑏] be linear and onto.
Then 𝑓 ∘ ℎ: [0, 1] ⟶ ℝ is continuous and so by the previous theorem there exists a polynomial
𝑃𝑛 such that |𝑓(ℎ(𝑥)) − 𝑃𝑛 | < 𝜖 for all 𝑥 ∈ [0, 1].
Since ℎ is linear, 𝑃𝑛 ∘ ℎ−1 is a polynomial and therefore for all 𝑦 ∈ [𝑎, 𝑏],
|𝑓(𝑦) − 𝑃𝑛 (ℎ−1 (𝑦))| < 𝜖.
This proves the corollary.

Corollary 7.4.2: For every interval [−ℎ, ℎ] there is a sequence of real polynomials 𝑃𝑛 such that
𝑃𝑛 (0) = 0 and such that lim 𝑃𝑛 (𝑥) = |𝑥| uniformly on [−ℎ, ℎ].
𝑛→∞

Proof: Since |𝑥| is continuous on [−ℎ, ℎ], by previous theorem there exists a sequence {𝑃𝑛∗ } of
real polynomials such that {𝑃𝑛∗ } converges uniformly to |𝑥| on [−ℎ, ℎ].
In particular 𝑃𝑛∗ (0) ⟶ |0| = 0 as 𝑛 ⟶ ∞.
Set 𝑃𝑛 (𝑥) = 𝑃𝑛∗ (𝑥) − 𝑃𝑛∗ (0) (𝑛 = 1, 2, 3, ⋯ ).
Then 𝑃𝑛 is a real polynomial such that lim 𝑃𝑛 (𝑥) = lim (𝑃𝑛∗ (𝑥) − 𝑃𝑛∗ (0))
𝑛→∞ 𝑛→∞

= |𝑥| − 0 = |𝑥|
And 𝑃𝑛 (0) = 𝑃𝑛∗ (𝑥) − 𝑃𝑛∗ (0) = 0.
This proves the corollary.
The following theorem generalizes the Weirstrass theorem and is known as Stone- Weierstrass
approximation theorem.
First we define algebra of functions 𝒜, uniform closure of 𝒜.

91
Definition7.4.3: Let 𝐸 ⊆ ℝ. A family 𝒜 of complex functions defined on 𝐸 is said to be an
algebra if 𝒜 is closed under addition, multiplication and scalar multiplication. i.e., for every
𝑓, 𝑔 ∈ 𝒜 and 𝑐 ∈ ℂ, 𝑓 + 𝑔, 𝑓𝑔, 𝑎𝑛𝑑 𝑐𝑓 are in 𝒜.

Definition7.4.4: Let 𝒜 is an algebra of functions on 𝐸 ⊆ ℝ. 𝐴 is said to be uniformly closed if


for every sequence {𝑓𝑛 } in 𝒜 such that {𝑓𝑛 } ⟶ 𝑓 converges uniformly on 𝐸, we have 𝑓 ∈ 𝒜.

Definition7.4.5: The set of all limits of uniformly convergent sequences in an algebra 𝒜 on 𝐸 ⊆


ℝ is called the uniform closure of 𝒜.

Definition7.4.6: We say that the algebra separates points if whenever 𝑥 ≠ 𝑦 in 𝐸 then ∃ 𝑓 ∈ 𝒜


such that 𝑓(𝑥) ≠ 𝑓(𝑦).

Theorem 7.4.2 (Stone Weierstrass Theorem).


Let 𝐶 be a compact set and let 𝒜 be an algebra of real continuous functions on 𝐶. If 𝒜separates
points on 𝐶 and annihilates no point, then the uniform closure 𝐵 of 𝒜 consists of all real
continuous functions on 𝐶.
Proof: We begin by proving simple lemma.
Lemma 7.4.1: If 𝑐1 and 𝑐2 are two real numbers and if 𝑥 and 𝑦 are two distinct points of 𝐶 then
∃ a function 𝑓 ∈ 𝒜 such that 𝑓(𝑥) = 𝑐1 and 𝑓(𝑦) = 𝑐2 .
Proof of the lemma: Let 𝑔 ∈ 𝐴 satisfy 𝑔(𝑥) ≠ 𝑔(𝑦). Such a 𝑔 exists because 𝒜 separates
points. Since 𝒜annihilates no point, ∃ functions ℎ and 𝑘 such that ℎ(𝑥) ≠ 0, 𝑘(𝑥) ≠ 0.
Then let 𝑢 = 𝑔ℎ − 𝑔(𝑦)ℎ, 𝑣 = 𝑔𝑘 − 𝑔(𝑦)𝑘.
It follows that 𝑢(𝑥) = 0 and 𝑢(𝑦) = 0 while 𝑣(𝑦) ≠ 0 and 𝑣(𝑥) ≠ 0.
𝑐 𝑢
1 𝑐 𝑣
2
Then 𝑓 = 𝑢(𝑥) + 𝑣(𝑦) satisfy 𝑓(𝑥) = 𝑐1 and 𝑓(𝑦) = 𝑐2 .

Now we continue with the proof of the theorem.


Proof of the Theorem: We shall prove the theorem in 4 steps.
Step 1: We claim that 𝑓 ∈ ℬ ⟹ |𝑓| ∈ ℬ.
Suppose ℎ = sup|𝑓(𝑥)| , 𝑥 ∈ 𝐶, (7.4.7)
and let 𝜖 > 0 be given.

92
By corollary 2, ∃ real numbers 𝑎𝑖′ 𝑠 𝑖 = 1, 2, ⋯ , 𝑛 such that
|∑𝑛𝑖=1 𝑎𝑖 𝑥 𝑖 − |𝑥|| < 𝜖 (7.4.8)
for all 𝑥 ∈ (−ℎ, ℎ).
Since 𝑔 = ∑𝑛𝑖=1 𝑎𝑖 𝑓 𝑖 ∈ 𝐵, by (1) and (2) it follows that
|𝑔(𝑥) − |𝑓(𝑥)|| < 𝜖 when ever 𝑥 ∈ 𝐶.
⟹ |𝑓| ∈ ℬ. Since ℬis uniformly closed.
Step 2: If 𝑓, 𝑔 ∈ 𝐵 then max{𝑓, 𝑔} and min{𝑓, 𝑔} are also in ℬ.
(𝑓+𝑔) (𝑓−𝑔)
Set ℎ = max{𝑓, 𝑔}. Since ℎ = + and ℬ is an algebra, from step 1 it follows that ℎ =
2 2

max{𝑓, 𝑔} ∈ ℬ. Similarly min{𝑓, 𝑔} ∈ ℬ.


By induction, the result can be extended to any finite set of functions.
Step 3: For every continuous real function 𝑓 on 𝐶, a point 𝑥 ∈ 𝐶, and 𝜖 > 0, there exists a
function 𝑔𝑥 ∈ ℬ such that 𝑔𝑥 (𝑥) = 𝑓(𝑥) and
𝑔𝑥 (𝑦) > 𝑓(𝑦) − 𝜖 (7.4.9)
for all 𝑦 ∈ 𝐶.
Since 𝐴 ⊂ ℬ and 𝐴 satisfies the hypothesis of the lemma, ℬ also does the same.
Hence for every 𝑦 ∈ 𝐶 we can find a function ℎ𝑦 ∈ ℬ such that
ℎ𝑦 (𝑥) = 𝑓(𝑥) and ℎ𝑦 (𝑦) = 𝑓(𝑦). (7.4.10)
Since ℎ𝑦 is continuous there exists an open set 𝐺(𝑦) such that
ℎ𝑦 (𝑠) > 𝑓(𝑠) − 𝜖 (7.4.11)
for all 𝑠 ∈ 𝐺(𝑦).
Since 𝐶 is compact, the open cover 𝐶 ⊆ ⋃𝑦𝑖 ∈𝐶 𝐺(𝑦𝑖 ) has a finite sub cover.

i.e., 𝐶 ⊆ ⋃𝑛𝑖=1 𝑔(𝑦𝑖 ). Put 𝑔𝑥 = max {ℎ𝑦1 , ℎ𝑦2 , ⋯ , ℎ𝑦 𝑛 }. Then 𝑔𝑥 ∈ ℬ (by step 2).

And 𝑔𝑥 (𝑥) = 𝑓(𝑥), 𝑔𝑦 (𝑦) = 𝑓(𝑦), 𝑔𝑥 (𝑠) > 𝑓(𝑠) − 𝜖 for all 𝑠 ∈ 𝐶.

Step 4: Let 𝑔𝑥 = min {ℎ𝑦1 , ℎ𝑦2 , ⋯ , ℎ𝑦 𝑛 }. Since 𝑔𝑥 for each 𝑥 ∈ 𝐶 is continuous there exists

open set 𝐺(𝑥) with 𝑥 ∈ 𝐺(𝑥) such that


𝑔𝑥 (𝑦) < 𝑓(𝑦) + 𝜖 for all 𝑦 ∈ 𝐺(𝑥). (7.4.12)
Since 𝐶 is compact, 𝐶 ⊆ ⋃𝑥∈𝐶 𝐺(𝑥) ⟹
𝐶 ⊆ ⋃𝑘𝑖=1 𝐺(𝑥𝑖 ), 𝑥𝑖 ∈ 𝐶 . (7.4.13)
If ℎ = min{𝑔𝑥 1 , ⋯ , 𝑔𝑥 𝑘 }, then by step 2 ℎ ∈ ℬ and by (3) we have

93
ℎ(𝑦) > 𝑓(𝑦) − 𝜖 for all 𝑦 ∈ 𝐶. (7.4.14)
Also from (3.2.12) and (3.2.13), we have
ℎ(𝑦) < 𝑓(𝑦) + 𝜖 for all 𝑦 ∈ 𝐶. (7.4.15)
Finally from (3.2.14) and (3.2.15), we have
|ℎ(𝑥) − 𝑓(𝑥)| < 𝜖 for all 𝑥 ∈ 𝐶. (7.4.16)

Since ℬ is uniformly closed, the inequality (7.4.16) establishes the conclusion of Stone
Weierstrass theorem.

7.5 Summary
1. There exists real valued continuous nowhere differentiable function.
2. A continuous function is uniformly approximated by a sequence of polynomials.

7.6 Keywords
Continuous, nowhere differentiable, uniform approximation, polynomial

7.7 Exercises
1. Show that there exists a continuous nowhre differentiable function.
2. State and prove Weierstrass approximation theorem
Solutions:

1. See theorem 7.3.1.

2. See theorem 7.4.1.

7.8 References
1. Richard R Goldberg, Methods of Real analysis, IBH publishing,New Delhi,1970
2. Walter Rudin, Principles of Mathematical Analysis,McGraw-Hill 3rd edition, 1976.
3. Robert G. Bartle, Donald R Sherbert, Introduction to Real Analysis, Wiley India, Third
edition, 2005.
4. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing,1974.

94
UNIT 8
POWER SERIES, THE EXPONENTIAL, LOGARITHMIC
AND TRIGONOMETRIC FUNCTIONS

8.1 Main Objectives

8.2 Introduction

8.3 Power Series

8.4 The Exponential Function

8.5 The Logarithmic function

8.6 Trigonometric functions

8.7 Summary

8.8 Key Words

8.9 Exercises

8.10 References

95
UNIT 8
POWER SERIES, THE EXPONENTIAL, LOGARITHMIC
AND TRIGONOMETRIC FUNCTIONS

8.1 Main Objectives


After going through this unit, students will be able to
 Find circle of convergence of a power series.
 Establish a formula for radius of convergence of a power series.
 Know how power series differentiated term by term.
 Know the properties of exponential, logarithmic, and trigonometric functions.

8.2 Introduction

In this unit we shall study an important class of series of functions called power series

that possess properties that are not valid for general series of functions. At each point interior to

the circle of convergence, the power series not only converges but converges absolutely. What is

very important about power series is that, in each circle concentric with the circle of convergence

but of smaller radius, the power series converges uniformly.

Also we shall study logarithmic, exponential functions along with an important collection of

transcendental functions called trigonometric functions such as sine, cosine etc.,

8.3. Power Series

Definition8.3.1: A series of the form ∑∞ 𝒏


𝒏=𝟎 𝒂𝒏 𝒙 , where {𝒂𝒏 } is a sequence of real numbers and

𝒙 is a real number is called a power series. The numbers 𝒂𝒏 are called the coefficients of the

power series.

96
Note: A power series may converge or diverge. Its convergence depends on the choice of 𝒛.

Circle of convergence is the circle, interior of which is the region of convergence of the series. In

otherwords, at every point of circle of convergence the power series converges and the series

diverges at every point outside the circle of convergence.

8.3.1 Examples
𝐱𝐧
1. Find the radius of convergence of the series ∑∞
𝐧=𝟎 . 𝐧!

𝟏 𝒂𝒏+𝟏 𝒏! 𝟏
Solution: Here 𝒂𝒏 = 𝒏! , 𝜶 = 𝐥𝐢𝐦 𝐬𝐮𝐩 ( ) = 𝐥𝐢𝐦 𝐬𝐮𝐩 (𝒏+𝟏)! = 𝐥𝐢𝐦 𝐬𝐮𝐩 𝐧 = 𝟎. Hence
𝒏→∞ 𝒂𝒏 𝒏→∞ 𝒏→∞

𝟏
𝑹 = 𝜶 = ∞. The given power series converges for all values of 𝒛.

1 Find the radius of convergence of the series ∑∞ 𝒏


𝒏=𝟎 𝒙 .

𝒏 𝟏
Solution: Here 𝒂𝒏 = 𝟏. Hence 𝜶 = 𝐥𝐢𝐦 𝐬𝐮𝐩 √𝟏 = 𝟏 and so 𝑹 = 𝜶 = 𝟏. If |𝒙| = 𝟏, then
𝒏→∞

𝒙𝒏 ↛ 𝟎 𝒂𝒔 𝒏 ⟶ ∞ and hence the series diverges and converges for all other values of 𝒙.

Theorem 8.3.1: A power series ∑∞ 𝒏


𝒏=𝟎 𝒂𝒏 𝒙 converges if |𝒙| < 𝑹 and diverges if |𝒙| > 𝑹, where

𝟏 𝒏
𝑹 = 𝜶, and 𝜶 = 𝐥𝐢𝐦 𝐬𝐮𝐩 √|𝒂𝒏 |.
𝒏→∞

Proof: Let 𝒃𝒏 = 𝒂𝒏 𝒙𝒏 𝒂nd applying the root test for convergence of series, we have

𝒏 𝒏 |𝒙|
𝐥𝐢𝐦 𝐬𝐮𝐩 √|𝒃𝒏 | = |𝒙| 𝐥𝐢𝐦 𝐬𝐮𝐩 √|𝒂𝒏 | = .
𝒏→∞ 𝒏→∞ 𝑹
|𝒙|
Thus ∑∞ ∞ 𝒏
𝒏=𝟎 𝒃𝒏 = ∑𝒏=𝟎 𝒂𝒏 𝒛 converges if < 𝟏 or |𝒙| < 𝑹 and diverges for |𝒙| > 𝑹.
𝑹

We next show that a power series can be differentiated term by term.

Theorem8.3.2: If the series ∑∞


𝒏=𝟎 𝒂𝒏 𝒙
𝒏
converges for |𝒙| < 𝑹 and if 𝒇(𝒙) =

∑∞
𝒏=𝟎 𝒂𝒏 𝒙
𝒏
(|𝒙| < 𝑹), then the series converges uniformly on [−𝑹 + 𝝐, 𝑹 − 𝝐] for every 𝝐 >

𝟎 . Also the function is continuous and differentiable in (−𝑹, 𝑹), and

97
𝒇′ (𝒙) = ∑∞
𝒏=𝟎 𝒏𝒂𝒏 𝒙
𝒏−𝟏
for all |𝒙| < 𝑹.

Proof: Given 𝝐 > 𝟎, for every 𝒙 ∈ [−𝑹 + 𝝐, 𝑹 − 𝝐], since |𝒙| < 𝑹 − 𝝐 , we have

|𝒂𝒏 𝒙𝒏 | ≤ |𝒂𝒏 (𝑹 − 𝝐)𝒏 | and so ∑∞ 𝒏 ∞ 𝒏


𝒏=𝟎|𝒂𝒏 𝒙 | ≤ ∑𝒏=𝟎 |𝒂𝒏 (𝑹 − 𝝐) |. Since the series on the right

side is a power series which converges absolutely in the interior of circle of convergence, by the

theorem the given series ∑∞ 𝒏


𝒏=𝟎 𝒂𝒏 𝒙 converges uniformly on [−𝑹 + 𝝐, 𝑹 − 𝝐] for every 𝝐 > 𝟎.

𝒏 𝒏 𝒏
Since √𝒏 ⟶ 𝟏 as 𝒏 ⟶ ∞, we have 𝐥𝐢𝐦 𝐬𝐮𝐩 √𝒏|𝒂𝒏 | = 𝐥𝐢𝐦 𝐬𝐮𝐩 √|𝒂𝒏 | .
𝒏→∞ 𝒏→∞

Hence the series ∑∞


𝒏=𝟎 𝒂𝒏 𝒙
𝒏
and its derived series ∑∞
𝒏=𝟎 𝒏𝒂𝒏 𝒙
𝒏−𝟏
have the same radius of

convergence.

Since the series ∑∞


𝒏=𝟎 𝒏𝒂𝒏 𝒙
𝒏−𝟏
is also a power series, it converges uniformly in [−𝑹 + 𝝐, 𝑹 −

𝝐] for every 𝝐 > 𝟎 , hence by theorem 3.2.4, 𝒇′ (𝒙) = ∑∞


𝒏=𝟎 𝒏𝒂𝒏 𝒙
𝒏−𝟏
for every |𝒙| ≤ 𝑹 − 𝝐.

Since for 𝒙 satisfying |𝒙| < 𝑹, there is a 𝝐 > 𝟎 such that |𝒙| < 𝑹 − 𝝐, we have

𝒇′ (𝒙) = ∑∞
𝒏=𝟎 𝒏𝒂𝒏 𝒙
𝒏−𝟏
for every 𝒙 satisfying |𝒙| < 𝑹.

Since 𝒇′ exists, it follows that 𝒇 is continuous.

By repeated application of the above theorem we have, a power series ∑∞ 𝒏


𝒏=𝟎 𝒂𝒏 𝒙 can be

differentiated term by term k-times.

Corollary 8.3.1: If the series ∑∞


𝒏=𝟎 𝒂𝒏 𝒙
𝒏
converges for |𝒙| < 𝑹 and if the function 𝒇(𝒙)

defined by 𝒇(𝒙) = ∑∞
𝒏=𝟎 𝒂𝒏 𝒙
𝒏
(|𝒙| < 𝑹) has derivatives of all orders in (−𝑹, 𝑹), which are

given by 𝒇𝒌 (𝒙) = ∑∞
𝒌=𝟎 𝒏(𝒏 − 𝟏) … (𝒏 − 𝟏) … (𝒏 − 𝒌 + 𝟏)𝒂𝒏 𝒙
𝒏−𝟏
.

In particular, 𝒇𝒌 (𝟎) = 𝒌! 𝒂𝒌 (𝒌 = 𝟎, 𝟏, 𝟐 … ).

Proof: Applying the above theorem to 𝒇(𝒙) = ∑∞


𝒏=𝟎 𝒂𝒏 𝒙
𝒏 (|𝒙| < 𝑹),

we obtain 𝒇′ (𝒙) = ∑∞
𝒏=𝟎 𝒏𝒂𝒏 𝒙
𝒏−𝟏
for every 𝒙 satisfying |𝒙| < 𝑹 and again applying the

theorem to 𝒇′ (𝒙) = ∑∞
𝒏=𝟎 𝒏𝒂𝒏 𝒙
𝒏−𝟏
and its derivatives 𝒇′′ , 𝒇′′′ , …, we get the desired result.

98
8.4 The Exponential Function

Definition8.4.1: The exponential function 𝑬: ℝ → ℝ defined by

𝒙𝒏
𝑬(𝒙) = ∑∞
𝒏=𝟎 𝒏! . (8.4.1)

It is clear that the series in 8.4.1, converges for every real number 𝒙, and 𝑬(𝒙) is continuous on

(−∞, ∞).

8.4.1. Properties of 𝐄(𝐱):

Theorem8.4.1: The exponential function 𝑬(𝒙) satisfies the following properties:

1. 𝑬(𝒙)𝑬(𝒚) = 𝑬(𝒙 + 𝒚).

2. 𝑬(𝒙)𝑬(−𝒙) = 𝑬(𝒙 − 𝒙) = 𝑬(𝟎) = 𝟏.

3. 𝑬(𝒒) = 𝒆𝒒 , for every rational number 𝒒.

4. 𝑬′ (𝒙) = 𝑬(𝒙).

Proof:

1. By multiplication of absolutely convergent series, we have

𝑬(𝒙)𝑬(𝒚) = 𝑬(𝒙 + 𝒚).

2. This is an immediate consequence 𝐨𝐟 (𝟏). 𝑬(𝒙)𝑬(−𝒙) = 𝑬(𝒙 − 𝒙) = 𝑬(𝟎) = 𝟏.

3. Repeated application of (1) yields,

𝑬(𝒙𝟏 + 𝒙𝟐 + ⋯ + 𝒙𝒏 ) = 𝑬(𝒙𝟏 ) 𝑬(𝒙𝟐 ) … 𝑬(𝒙𝒏 ).

If we take 𝒙𝟏 = 𝒙𝟐 = ⋯ + 𝒙𝒏 = 𝟏 in the above relation, we get 𝑬(𝒏) = 𝒆𝒏 , for 𝒏 =

𝟏 𝒏
𝟏, 𝟐, 𝟑 …, where 𝒆 = ∑∞
𝒏=𝟎 𝒏! . If 𝒑 = 𝒎, where 𝒏 and 𝒎 are positive integers, then

[𝑬(𝒑)]𝒎 = 𝑬(𝒎𝒑) = 𝑬(𝒏) = 𝒆𝒏 , so that 𝑬(𝒒) = 𝒆𝒒 for all positive rational numbers

and since 𝑬(𝒙)𝑬(−𝒙) = 𝟏  𝑬(−𝒙) = 𝟏/𝑬(𝒙), we have 𝑬(−𝒒) = 𝒆−𝒒 and hence

𝑬(𝒒) = 𝒆𝒒 , for every rational number 𝒒.

99
𝑬(𝒙+𝒉)−𝑬(𝒙) 𝑬(𝒉)−𝟏
4. 𝑬′ (𝒙) = 𝐥𝐢𝐦 = 𝑬(𝒙) 𝐥𝐢𝐦 = 𝑬(𝒙).
𝒉→𝟎 𝒉 𝒉→𝟎 𝒉

The exponential function 𝑬(𝒙) is frequently written 𝒆𝒙 .

Theorem8.4.2: The exponential function 𝑬 is strictly increasing on ℝ and has range equal to

{𝒙 ∈ ℝ: 𝒙 > 𝟎}. Further we have i) 𝐥𝐢𝐦 𝑬(𝒙) = 𝟎 𝒊𝒊) 𝐥𝐢𝐦 𝑬(𝒙) = ∞.


𝒙→∞ 𝒙→∞

Proof: We know that 𝑬(𝟎) = 𝟏 > 𝟎 and 𝑬(𝒙) ≠ 𝟎 for all real 𝒙. Since 𝑬 is continuous on 𝑹, it

follows from Intermediate value property that 𝑬(𝒙) > 𝟎 for all 𝒙 ∈ ℝ.

Therefore 𝑬′ (𝒙) = 𝑬(𝒙) > 𝟎 for 𝒙 ∈ ℝ, so that 𝑬 is strictly increasing on ℝ.

𝒙 𝒙𝟐 𝒙𝒏
We know that 𝑬(𝒙) = 𝐥𝐢𝐦 𝑬𝒏 (𝒙), where 𝑬𝒏 (𝒙) = 𝟏 + 𝟏! + + ⋯+ , for 𝒙 ∈ ℝ.
𝒙→∞ 𝟐! 𝒏!

And so if 𝒙 > 𝟎, {𝑬𝒏 } is strictly increasing and in particular 𝟏 + 𝒙 = 𝑬𝟏 (𝒙) < 𝑬(𝒙).

Hence 𝟐 < 𝒆 and that 𝐥𝐢𝐦 𝑬(𝒙) = ∞. Also if 𝒙 > 𝟎 then since 𝟎 < 𝑬(−𝒙) = 𝟏/𝑬(𝒙), it follows
𝒙→∞

that 𝐥𝐢𝐦 𝑬(𝒙) = 𝟎. By the intermediate value theorem, range of 𝑬={𝒙 ∈ ℝ: 𝒙 > 𝟎}.
𝒙→−∞

8.5 The Logarithmic Function

Definition 8.5.1: The function 𝑳: ℝ → ℝ which is the inverse of 𝑬: ℝ → ℝ is the logarithm.

Note: Since 𝑬 and 𝑳 are inverse functions, 𝑳(𝑬(𝒚)) = 𝑬(𝑳(𝒚)) = 𝒚, ∀ 𝒚 ∈ ℝ.

8.5.1: Properties of Logarithm:

Theorem8.5.1: The logarithm function 𝑳(𝒙) has the following properties:

1. Logarithm is strictly increasing function with domain {𝒙 ∈: 𝒙 > 𝟎} and range ℝ.

2. 𝑳(𝒙𝒚) = 𝑳(𝒙)𝑳(𝒚) , 𝒙 > 𝟎, 𝒚 > 𝟎.

3. 𝑳(𝟏) = 𝟎 and 𝑳(𝒆) = 𝟏.

4. 𝑳(𝒙𝒏 ) = 𝒏 𝑳(𝒙), 𝒙 > 𝟎, 𝒓𝒂𝒕𝒊𝒐𝒂𝒏𝒍 𝒏.

100
5. 𝐥𝐢𝐦𝑳(𝒙) = −∞, 𝒂𝒏𝒅 𝐥𝐢𝐦 𝑳(𝒙) = ∞.
𝒙→𝟎 𝒙→∞

Proof:

1. Since 𝑬(𝒙) is strictly increasing with domain 𝑹 and range 𝑬={𝒙 ∈ 𝑹: 𝒙 > 𝟎}, and 𝑳 is the

inverse of 𝑬, it follows that 𝑳 is strictly increasing with domain {𝐱 ∈ 𝐑: 𝐱 > 𝟎} and range 𝐑.

2. If 𝑳(𝒙) = 𝒂 𝒂𝒏𝒅 𝑳(𝒚) = 𝒃, then 𝒙 = 𝑬(𝒂) and 𝒚 = 𝑬(𝒃).

Hence by 𝑬(𝒙 + 𝒚) = 𝑬(𝒙)𝑬(𝒚), we have 𝒙𝒚 = 𝑬(𝒂)𝑬(𝒃) = 𝑬(𝒂 + 𝒃).

Thus 𝑳(𝒙𝒚) = 𝑳(𝑬(𝒂 + 𝒃)) = 𝒂 + 𝒃 = 𝑳(𝒙) + 𝑳(𝒚) , 𝒙 > 𝟎, 𝒚 > 𝟎.

3. Since 𝑬(𝟎) = 𝟏, 𝒂𝒏𝒅 𝑬(𝟏) = 𝒆, it follows that 𝑳(𝟏) = 𝟎 and 𝑳(𝒆) = 𝟏.

4. From (2), 𝒚 = 𝒙 ⟹ 𝑳(𝒙𝟐 ) = 𝟐𝑳(𝒙), 𝑳(𝒙𝟑 ) = 𝑳(𝒙𝟐 . 𝒙)

= 𝑳(𝒙𝟐 ) + 𝑳(𝒙) = 𝟐𝑳(𝒙) + 𝑳(𝒙) = 𝟑𝑳(𝒙).

By induction we have, 𝑳(𝒙𝒏 ) = 𝒏 𝑳(𝒙). And so 𝒙𝒏 = 𝑬(𝒏𝑳(𝒙)), if 𝒙 > 𝟎 and 𝒏 is an integer.

𝟏
𝟏
Similarly if 𝒎 is a positive integer , then we have 𝒙𝒎 = 𝑬 ( 𝑳(𝒙)) .
𝒎

Raising to nth power the above equation becomes 𝑳(𝒙𝒒 ) = 𝒒 𝑳(𝒙), for every rational 𝒒.

5. Since 𝟐 < 𝒆, 𝐥𝐢𝐦(𝒆𝒏 ) = ∞ and 𝐥𝐢𝐦(𝒆−𝒏 ) = 𝟎.


𝒙→𝟎 𝒙→𝟎

Since 𝑳(𝒆𝒏 ) = 𝒏 and 𝑳(𝒆−𝒏 ) = −𝒏 and since 𝑳 is strictly increasing , it follows that

𝐥𝐢𝐦 𝑳(𝒙) = 𝐥𝐢𝐦 𝑳(𝒆𝒏 ) = ∞ and 𝐥𝐢𝐦 𝑳(𝒙) = 𝐥𝐢𝐦 𝑳(𝒆−𝒏 ) = −∞.
𝐱→∞ 𝐱→∞ 𝒙→𝟎 𝒙→𝟎

8.6. Trigonometric Functions

Along with the exponential and logarithmic functions there is another very important

class of transcendental functions called the trigonometric functions. These are the sine, cosine,

101
tangent, cotangent, secant and cosecant functions. It suffices to deal with sine and cosine

functions, since the other four trigonometric functions derived from these two.

𝟏 𝟏
Definition 8.6.1: We define 𝑪(𝒙) = 𝟐 [𝑬(𝒊𝒙) + 𝑬(−𝒊𝒙)], 𝑺(𝒙) = 𝟐𝒊 [𝑬(𝒊𝒙) − 𝑬(−𝒊𝒙)].

We show that these functions coincide with sinx and cosx. It is clear that for real 𝒙, 𝑪(𝒙)and

𝑺(𝒙) are real. Also Since 𝑬(𝒊𝒙) = 𝑪(𝒙) + 𝒊 𝑺(𝒙). Thus 𝑪(𝒙) and 𝑺(𝒙) are real and imaginary

parts of 𝐄(𝐢𝐱) if 𝐱 is real.

8.6.1 Properties of Trigonometric Functions:

̅̅̅̅̅̅̅̅ = 𝑬(𝒊𝒙)𝑬(−𝒊𝒙) = 𝟏, so that |𝑬(𝒊𝒙)| = 𝟏, 𝒙 real.


1. |𝑬(𝒊𝒙)|𝟐 = 𝑬(𝒊𝒙)𝑬(𝒊𝒙)

2. There exist positive numbers 𝒙 such that 𝑪(𝒙) = 𝟎.

Proof: Suppose 𝑪(𝒙) ≠ 𝟎 , for any 𝒙. Since 𝑪(𝟎) = 𝟏, 𝑪(𝒙) > 𝟎 ∀𝒙 > 𝟎, hence it follows from

2 that 𝑺′ (𝒙) > 𝟎. Hence 𝑺 is strictly increasing.

From the identities, 𝑬(𝒊𝒙) = 𝑪(𝒊𝒙) + 𝒊𝑺(𝒊𝒙) and |𝑬(𝒊𝒙)| = 𝟏, we have 𝑪(𝒙) − 𝑪(𝒚) ≤ 𝟐.
𝒚
For 𝟎 < 𝒙 < 𝒚, we have 𝑺(𝒙)(𝒚 − 𝒙) < ∫𝒙 𝑺(𝒕)𝒅𝒕 = 𝑪(𝒙) − 𝑺(𝒙).

𝒚
Thus we have 𝑺(𝒙)(𝒚 − 𝒙) < ∫𝒙 𝑺(𝒕)𝒅𝒕 = 𝑪(𝒙) − 𝑺(𝒙) ≤ 𝟐.

Since (𝒙) > 𝟎 , the above inequality does not hold for large 𝒚, a contradiction.

Hence there exist numbers such that 𝑪(𝒙) = 𝟎.

Theorem 8.6.1. The functions 𝑪 and 𝑺 are periodic with period 𝟐𝝅, where 𝝅 = 𝟐𝒓𝟎 , where 𝒓𝟎 is

the smallest positive integer such that 𝑪(𝒓𝟎 ) = 𝟎.

𝝅 𝝅
Proof: We have 𝑪 (𝟐 ) = 𝟎 and therefore 𝑺 ( 𝟐 ) = ±𝟏, since |𝑬(𝒊𝒙)| = |𝑪(𝒙) + 𝒊𝑺(𝒙)| = 𝟏.

𝝅 𝝅 𝝅 𝝅
𝑺′ (𝒙) = 𝑪(𝒙) > 𝟎 in (𝟎, 𝟐 ), 𝑺 is increasing in (𝟎, 𝟐 ), hence 𝑺 ( 𝟐 ) = 𝟏. Thus from 𝑪 ( 𝟐 ) = 𝟎 ,

𝝅 𝝅𝒊
𝑺 (𝟐 ) = 𝟏 and 𝑬(𝒊𝒙) = 𝑪(𝒙) + 𝒊 𝑺(𝒙), we have 𝑬 ( 𝟐 ) = 𝒊.

102
Now from 𝑬(𝒙 + 𝒚) = 𝑬(𝒙)𝑬(𝒚), we have 𝑬(𝝅𝒊) = −𝟏, 𝑬(𝟐𝝅𝒊) = 𝟏 and 𝑬(𝒛 + 𝟐𝝅𝒊) =

𝑬(𝒛), for complex number 𝒛. Hence 𝑬 is periodic with period 𝟐𝝅𝒊.

𝟏 𝟏
Since 𝑺(𝒛 + 𝟐𝝅) = 𝟐𝒊 [𝑬(𝒊𝒛 + 𝟐𝝅𝒊) − 𝑬(−𝒊𝒛 − 𝟐𝝅𝒊)] = 𝟐𝒊 [𝑬(𝒊𝒛) − 𝑬(−𝒊𝒛))] = 𝑺(𝒛),

𝟏 𝟏
and 𝑪(𝒛 + 𝟐𝝅) = 𝟐 [𝑬(𝒊𝒛 + 𝟐𝝅𝒊) + 𝑬(−𝒊𝒛 − 𝟐𝝅𝒊)] = 𝟐 [𝑬(𝒊𝒛) + 𝑬(−𝒊𝒛)] = 𝑪(𝒛), we have

that 𝑪(𝒙) and 𝑺(𝒙) are periodic with period 𝟐𝝅 .

This establishes theorem 8.6.1.

8.7. Summary

1. A power series may converge or diverge. Its convergence depends on the choice of 𝒛. Circle

of convergence is the circle, interior of which is the region of convergence of the series..

2. The exponential function 𝑬 is strictly increasing on 𝑹 and has range equal to {𝒙 ∈ 𝑹: 𝒙 > 𝟎}.

Further we have i) 𝐥𝐢𝐦 𝑬(𝒙) = 𝟎 𝒊𝒊) 𝐥𝐢𝐦 𝑬(𝒙) = ∞.


𝒙→∞ 𝒙→∞

3. Logarithm is strictly increasing function with domain {𝒙 ∈: 𝒙 > 𝟎} and range 𝑹.

4. 𝐥𝐢𝐦𝑳(𝒙) = −∞, 𝒂𝒏𝒅 𝐥𝐢𝐦 𝑳(𝒙) = ∞.


𝒙→𝟎 𝒙→∞

5. The functions 𝑪 and 𝑺 are periodic with period 𝟐𝝅.

8.8. Keywords

Power series, converge, diverge, exponential, trigonometric, logarithmic

103
8.9. Exercises

1. Find the radius of convergence of the series ∑∞ 𝒏


𝒏=𝟎 𝒏! 𝒙 .

2. Show that the functions 𝑪 and 𝑺 satisfy 𝑪(𝟎) = 𝟏, 𝑺(𝟎) = 𝟎 and 𝑪′ (𝒙) = −𝑺(𝒙) and

𝑺′ (𝒙) = 𝑪(𝒙).

𝟏
3. Show that the derivative of 𝑳 is given by 𝑳′ (𝒙) = 𝒙 for 𝒙 > 𝟎.

Solutions:
𝟏 𝒂𝒏+𝟏
1. Here 𝒂𝒏 = 𝒏!. By ratio test, we have 𝑹 = 𝐥𝐢𝐦 𝒔𝒖𝒑 = 𝐥𝐢𝐦 𝐬𝐮𝐩( 𝒏 + 𝟏) = ∞
𝒏→∞ 𝒂𝒏 𝒏→∞

or 𝑹 = 𝟎.

2. The results follow by definition of 𝐶(𝑥) and 𝑆(𝑥).


𝟏
3. If 𝑳(𝒙) = 𝒚, then 𝑬(𝒚) = 𝒙. Since 𝑳 is the inverse of 𝑬, we have, 𝑳′ (𝒙) = 𝑬′ (𝒙) , but

𝟏 𝟏 𝟏
𝑬′ (𝒙) = 𝑬(𝒙), hence 𝑳′ (𝒙) = = 𝑬(𝒙) = 𝒙 .
𝑬′ (𝒙)

8.10 References

1. Richard R Goldberg, Methods of Real analysis, IBH publishing,New Delhi,1970


2. Walter Rudin, Principles of Mathematical Analysis,McGraw-Hill 3rd edition, 1976.
3. Robert G. Bartle, Donald R Sherbert, Introduction to Real Analysis, Wiley India, Third
edition, 2005.
4. Apostol, Mathematical Analysis, Second Edition, Narosa Publishing, 1974.

104
Block III: Improper Integrals

Block III
Unit 9: Improper integrals: Definition, Criteria for conver-
gence, Interchanging derivatives and integrals.
9.1.1 Main Objectives
9.1.2 Introduction
9.1.3 Definition of an Improper Integral
9.1.4 Keywords
9.1.5 Terminal Problems
9.1.6 Books for Reference

1
105
Block III
Unit 9: Improper integrals: Definition, Criteria for
convergence, Interchanging derivatives and integrals.

9.1.1 Main Objectives


The objective of the unit is to study and understand the following topics in detail.

• Definite integrals that are improper either by virtue of an infinite limit of integration.

• Evaluate an integral over an infinite interval.

• Evaluate an integral over a closed interval with an infinite discontinuity within the
interval.

• Use the comparison theorem to determine whether a definite integral is convergent.

9.1.2 Introduction
The goal of this chapter is to meaningfully extend our theory of integrals to improper in-
tegrals. There are two types of so-called improper integrals: the first involves integrating
a function over an infinite domain and the second involves integrands that are undefined
at points within the domain of integration. In order to integrate over an infinite domain,
Z b
we consider limits of the form lim f (x) dx. If the integrand is not defined at c(a <
b→∞ a Z b Z b
c < b) then we split the integral and consider the limits f (x) dx = lim f (x) dx +
a →0 c+
Z c−
lim f (x) dx. The latter is sometimes also referred to as improper integrals of the sec-
→0 a
ond kind. Such situations occur, for example, for rational functions f (x) = p(x)/q(x)
whenever q(x) has zeroes in the domain of integration.

The notions of convergence and divergence as discussed in the context of sequences and
series will be very important to determine these limits. Improper integrals (of both types)
arise frequently in applications and in probability. By relating improper integrals to infinite
series we derive the last convergence test: the Integral Comparison test. As an application
X∞
we finally prove that the p−series k −p converges for p > 1 and diverges otherwise.
k=1

9.1.3 Definition of an Improper Integral


Classification of improper integrals:
For convenience we classify all improper integrals into four types as follows.

106
Definition 1. Type I
Z ∞
f (x) dx, where f is bounded and integrable over every closed subinterval of [a, ∞).
a
In this case we define
Z ∞ Z λ
f (x) dx = lim f (x) dx, when the limit exists.
a λ→∞ a

Z λ
If lim f (x) dx = A exists. We say that the improper integral converges and A is the
λ→∞ a
value of the improper integral.
Z ∞
dx
Example 1. Evaluate 2
.
1 x
Solution: This is the improper integral of type I, so
Z ∞ Z λ
dx dx
= lim
1 x2 λ→∞ 1 x2
 λ
1
= lim −
λ→∞ x 1
 
1
= lim 1 −
λ→∞ λ
Z ∞
dx
= 1.
1 x2

Therefore the improper integral converges and has the value is 1.

Definition 2. Type II
Z b
f (x) dx, where f is bounded and integrable over every closed subinterval of (−∞, b].
−∞
In this case we define
Z b Z b
f (x) dx = lim f (x) dx, when the limit exists.
−∞ λ→−∞ λ

Z b
If lim f (x) dx = B exists. We say that the improper integral converges and B is the
λ→−∞ λ
value of the improper integral.
Z 0
Example 2. Evaluate sin xdx.
−∞
Solution: This is the improper integral of type II, so
Z 0 Z 0
cos xdx = lim sin xdx
−∞ λ→−∞ −∞
= lim [− cos x]0λ
λ→∞
Z 0
cos xdx = lim [−1 + cos λ]
−∞ λ→∞

which does not exit. Therefore the improper integral diverges.

107
Z ∞
Note: If the improper integral f (x) dx is given, then we divide the integral into two
−∞
parts and we write
Z ∞ Z a Z ∞
f (x) dx = f (x) dx + f (x) dx.
−∞ −∞ a

On the right side we have improper integrals of type II and type I, which can be evaluated
using the above definitions.
Z ∞
Example 3. Evaluate ex dx.
−∞
Solution: This is the improper integral of type I and II, so
Z ∞ Z 0 Z ∞
x x
e dx = e dx + ex dx
−∞ −∞ 0
Z 0 Z δ
x
= lim
e dx + lim ex dx
λ→−∞ λ δ→∞ 0

= lim 1 − eλ + lim eδ − 1
   
λ→−∞ δ→∞
= 1+∞
Z ∞
ex dx = ∞.
−∞

Therefore the improper integral diverges.

Definition 3. Type III


Z b
f (x) dx, where f is bounded and integrable over every closed subinterval of (a, b] and
a+
the lim f (x) does not exists.
x→a+
In this case we define
Z b Z b
f (x) dx = lim f (x) dx, where δ > 0.
a+ δ→0 a+δ

Z b
If lim f (x) dx = A exists. We say that the improper integral converges and A is the
δ→0 a+δ
value of the improper integral.

Z 1
dx
Example 4. Evaluate .
0 x
Solution: This is the improper integral of type III, so
Z 1 Z 1
dx dx
= lim
0 x λ→0 λ x

= lim [log(x)]1λ
λ→0
= lim [log(1) − log(λ)]
λ→0
Z 1
dx
= ∞.
0 x

108
Therefore the improper integral diverges.

Definition 4. Type IV
Z b−
f (x) dx, where f is bounded and integrable over every closed subinterval of (a, b] and
a
the lim f (x) does not exists.
x→a+
In this case we define
Z b− Z b−δ
f (x) dx = lim f (x) dx, where δ > 0.
a δ→0 a

Z b−δ
If lim f (x) dx = B exists. We say that the improper integral converges and A is the
δ→0 a
value of the improper integral.
Z 1
dx
Example 5. Evaluate √ .
0 1 − x2
Solution: This is the improper integral of type IV, so
Z 1 Z 1−λ
dx dx
√ = lim √
0 1 − x2 λ→0 0 1 − x2
 −1 1−λ
= lim sin (x) 0
λ→0
= lim sin−1 (1 − λ) − sin−1 (0)
 
λ→0
Z 1
dx π
√ = .
0 1−x 2 2
π
Therefore the improper integral converges and has the value is .
2
Z b
Note: If the improper integral f (x) dx is given, then we divide the integral into two
a
parts and we write Z b Z c− Z b
f (x) dx = f (x) dx + f (x) dx.
a a c+

On the right side we have improper integrals of type IV and type III, which can be evaluated
using the above definitions.
Z b
dx
Example 6. Evaluate 1/3
, a,b>0.
−a x
Solution: This is the improper integral of type III and IV, so
Z b Z 0 Z b
dx dx dx
= + [sum of integrals of types III and IV]
−a x1/3 −a x
1/3
0 x
1/3
Z −λ Z b
dx dx
= lim 1/3
+ lim
λ→0 −a x λ1 →0 λ x1/3
1
3  2/3 2/3
 3 h
2/3 2/3
i
= lim λ − a + lim b − λ1
λ→0 2 2 λ1 →0
Z b
dx 3 2/3
b − a2/3 .
 
=
−a x1/3 2

109
3  2/3
b − a2/3 .

Therefore the improper integral converges and has the value
2
Note:

If a limit of integration a+ or b− appears, it is apparent that the integral is improper.


However the sings + and − are not always used. In such a case the integral has to be
recognized as improper byZthe discontinuities of the integrand.

dx
For example, the integral is improper and can be written as the sum of the
−∞ x(x − 1)
four types of improper integrals as follows:

Z ∞ Z −1 Z 0 Z 1/2
dx dx dx dx
= + +
−∞ x(x − 1) −∞ x(x − 1) −1 x(x − 1) 0 x(x − 1)
Z 1 Z 2 Z ∞
dx dx dx
+ + + .
1/2 x(x − 1) 1 x(x − 1) 2 x(x − 1)

When an improper integral is expressed as the sum of two or more improper integrals.
We say that the improper integral converges if every improper integral on the right side is
convergent.
Remark: If the integrand tends to infinity at a point c within the range of integration.
We usually define the integral by the equation
Z b Z c−δ Z c+δ1
f (x) dx = lim f (x) dx + lim f (x) dx.
a δ→0 a δ1 →0 b

But in certain problems the two limits in the last equation are both infinite, while the sum
δ1
of two integrals tends to a finite limit if tends to a finite limit. We then define the principal
δ
value (sometimes called the Cauchy’s principle value) of the integral by the equation
Z b Z c−δ Z c+δ 
P f (x) dx = lim f (x) dx + f (x) dx .
a δ→0 a b

Note that the more existence of the principal value doesnot imply the convergence of the
improper integral.
Z 1
dx
Example 7. Evaluate .
−1 x
Solution:
Z 1 Z 0 Z 1
dx dx dx
= +
−1 x −1 x 0 x
Z λ+ Z 1
dx dx
= lim + lim +
λ→0 −1 x λ1 →0 λ x
1

= lim [log(x)]λ−1 + lim [log(x)]1λ1


λ→0 λ→0
= lim [log(λ) − log(−1)] + lim [log(1) − log(λ1 )]
λ→0 λ1 →0
= lim [log(λ)] + lim [log(λ1 )]
λ→0 λ1 →0

110
λ1
which does not exist unless tends to finite limit. Therefore the improper integral does not
λ
converges.
However, the principal value of the integral is given by
Z 1 Z −λ Z 1 
dx dx dx
P = lim +
−1 x λ→0 −1 x λ x
  
λ
= lim log
λ→0 λ
Z 1
dx
P = 0.
−1 x

Remark:
It must not be supposed that the four types of improper integrals are fundamentally differ-
ent. An improper integral of one type can always be the transformed so as to belong to other
type. More precisely an improper integral with unbounded integrand can be transformed
into one with indefinite limit of integration.
Z b
For consider the improper integral f (x) dx, where f (x) → ∞ as x → b and f is contin-
a
uous elsewhere in the interval (a, b). Put

x−a a + bξ
ξ= or x = , (9.1)
b−x 1+ξ

When x → b, then ξ → ∞ and when x = a, then ξ = 0.


Then the improper integral becomes

b ∞  
(b − a)
Z Z
a + bξ
f (x) dx = f dξ.
a 0 1+ξ (1 + ξ)2

Here the integrand in ξ is everywhere finite. Z


1
dx
In particularly, consider the improper integral √ (Type IV)
0 1 − x2
Put
x ξ
ξ= or x = ,
1−x 1+ξ
When x → 1, then ξ → ∞ and when x = 0, then ξ = 0.
Then the improper integral becomes
Z 1 Z ∞
dx dξ
√ = ,
0 1 − x2 0 (1 + ξ)(1 + 2ξ)1/2

which is of the type I.


Caution:We should take care in applying this kind of transformation when the infinity
of f (x) is inside the range of integration. In such a case, it is safer to divide the integral into
two using the definition and then use the type of transformation given in (9.1).
Remark:It is possible to express an integral with infinite limit of integration in terms of an
integral with finite limits.

111
Z ∞
dx
For example, consider the improper integral
1 xs
1 1
Put x = or ξ = , then dx = − ξ12 dξ
ξ x
when x = 1, then ξ = 1 and when x → ∞, then ξ → 0, then the integral takes the form
Z ∞ Z 0   Z 1
dx s 1
= ξ − 2 dξ = ξ s−2 dξ.
1 xs 1 ξ 0

[both then integrals converges if 1 < s < 2]


Remark: It is also possible in any cases to express a convergent improper integral of the
types III and IV as
Z 1a finite√integral by a change of variables. For example, consider the
improper integral f (x) 1 − x2 dx
0
π
Put x = sin θ then dx = cos θ dθ, when x = 0 then θ = 0 and when x = 1 and θ = .
2
Thus the integral becomes,
Z 1 Z π/2
f (x)
√ dx = f (sin θ) dθ.
0 1 − x2 0

Caution: Although such a transformation is always possible and also theoretically true, it
is not effectively practicable in all cases.
Comparison tests of Type I
Z ∞
Lemma 9.1. Let f (x) > 0 for all x > a, then f (x) dx converges if and only if there
Z t a

exists a number M > 0, such that f (x) dx ≤ M , for all t ≥ a.[or t ∈ [a, ∞)]
a
Z t
Proof. Put φ(t) = f (x) dx,
a

a t

since f (x) > 0 for all x > a φ is monotonically increasing on [a, ∞). Hence lim φ(t)
Z ∞ t→∞

exists if and only if φ(t) is bounded above. That is f (x) dx exists if and only if there
a
exists a number M such that, φ(x) ≤ M for all t ≥ a.

Theorem 9.1. If 0 < f (x) ≤ g(x), f and g are both continuous for a ≤ x < ∞ and if
Z b Z ∞
g(x) dx converges then so does f (x) dx
a a

Proof. Since 0 < f (x) ≤ g(x) for a ≤ x < ∞, we have


Z λ Z λ
f (x) dx = g(x) dx, for all λ > a (9.2)
a a

112
Z ∞
Since g(x) dx converges, we have
a

Z ∞
g(x) dx = M < ∞ (9.3)
a

From (9.2) and (9.3), we have


Z λ Z λ Z ∞
f (x) dx = g(x) dx ≤ g(x) dx = M .
a a a

Z λ
Hence f (x) dx ≤ M , for all λ > a.
a Z ∞
Therefore by Lemma 1, f (x) dx converges.
a

Note:Sometimes Theorem (9.1) is called "Comparison test for convergence".

Theorem 9.2. If 0 < g(x) ≤ f (x), f and g are both continuous for a ≤ x < ∞ and if
Z b Z ∞
g(x) dx diverges then so does f (x) dx
a a
Z ∞ Z ∞ Z ∞
Proof. If f (x) dx converges, then g(x) dx converges by Theorem 1, therefore f (x) dx
a
Z ∞ a a

diverges whenever g(x) dx diverges.


a

Note:Sometimes Theorem (9.2) is called "Comparison test for divergence".

Theorem 9.3. Let f (x) > 0 and g(x) > 0 for all x ≥ a,
Z ∞
f (x)
1. If lim (= l exists) is finite and g(x) dx converges,
x→a g(x) a
Z ∞
then f (x) dx converges.
a
Z ∞
f (x)
2. If lim = ∞ or 6= 0 and g(x) dx diverges,
x→a g(x) a
Z ∞
then f (x) dx diverges.
a

f (x)
Proof. First, let lim = l < ∞.
g(x) x→∞
Since f and g are positive. We must have l ≥ 0. Now let  > 0 be fixed.
f (x)
Since lim = l, there exists c > a such that
x→∞ g(x)

f (x)
−l < , for all x ≥ c.
g(x)
f (x)
l− < < l + , for all x ∈ (b, +∞)
g(x)
(l − ) g(x) < f (x) < (l + ) g(x), for all x ≥ c. (9.4)

113
Thus, we have by (9.4),

f (x) < (l + ) g(x), for all x ≥ c (9.5)


Z ∞ Z ∞
If now g(x) dx converges, then so does (l + ) g(x) dx.
a a
Since (l + ) g(x) > 0, we have
Z ∞ Z ∞
g(x) dx < g(x) dx converges
c a
Z ∞ Z ∞
(l + ) g(x) dx < (l + ) g(x) dx < +∞
c a
Z ∞
Therefore (l + ) g(x) dx converges.
a

Z ∞
By Theorem 1, f (x) dx converges. [by using (9.5)]
c
Since Z ∞ Z c Z ∞
f (x) dx = f (x) dx + f (x) dx
a a c
Z ∞
f (x) dx converges.
a
Again from (9.4), we have

(l − ) g(x) < f (x), for all x ≥ c (9.6)


Z ∞
If l 6= 0 and also l −  6= 0, then the divergence of g(x) dx implies the divergence of
Z ∞ a

(1 − ) g(x) dx.
a Z ∞
Therefore (1 − ) g(x) dx also diverges.
a Z ∞
In view of (9.6) and Theorem 2 f (x) dx diverges.
a
Finally, let l = +∞, then given m > 0 there exits a c > a such that

f (x)
> m, for all x ≥ c
g(x)
m g(x) < f (x) for all x ≥ c (9.7)
Z ∞ Z ∞
If now g(x) dx diverges, then so does m g(x) dx.
a Z a
∞ Z ∞
Therefore m g(x) dx diverges. By (9.7) and Theorem 2, it follows that f (x) dx
c Z ∞ c

diverges. Hence f (x) dx diverges.


a

Exercise:
Z ∞
1
1. Show that converges if and only if p > 1, where a > 0.
a xpdx

114
Solution: If p = 1, then
Z ∞ Z δ
1 1
dx = lim dx
a x δ→∞ a x

= lim [log x]δa


δ→∞
= lim [log δ − log a]
δ→∞
= log ∞ − log a
= ∞.

∴ The improper integral is diverges if p = 1.


If p 6= 1, then
Z ∞ Z δ
1 1
dx = lim dx
a xp δ→∞ a xp
Z δ
= lim x−p dx
δ→∞ a
 −p+1 δ
x
= lim
δ→∞ −p + 1
 −p+1 δ
δ − a−p+1
= lim
δ→∞ −p + 1
1

 if p > 1,
= (p − 1)ap−1
∞ if p < 1.

Test the following improper integral for convergence.


Z ∞
x2
1. √ dx
2 x7 + 1
1 1
Solution: Let f (x) = √ and choose g(x) = 3/2
x3/2 1 + x−7 x

1

f (x) 1 + x−7 x3/2
lim = lim
x→∞ g(x) x→∞ 1
x 3/2
1
= lim √
x→∞ 1 + x−7
= 1.
Z ∞
x2
Z
∴ g(x) dx convergent, then √ dx is convergent.
2 x7 + 1
Z ∞
x3
2. √ dx
2 x7 + 1

115
1 1
Solution: Let f (x) = √ and choose g(x) = 1/2
x1/2 1 + x−7 x

1

f (x) 1 + x−7
x1/2
lim = lim
x→∞ g(x) x→∞ 1
x 1/2
1
= lim √
x→∞ 1 + x−7
= 1.
Z ∞ Z ∞
x3
∴ g(x) dx divergent, then √ dx is divergent.
2 2 x7 + 1
Z ∞
log x
3. dx
1 x2
log x 1
Solution: Let f (x) = 2 and choose g(x) = 3/2
x x

log x
f (x) 2
lim = lim x
x→∞ g(x) x→∞ 1
x 3/2
log x
= lim 1/2
x→∞ x
1
h ∞ i
= lim x ∵ form
x→∞ 1 −1/2 ∞
x
2
= 0.
Z ∞ Z ∞
log x
∴ g(x) dx convergent, then dx is convergent.
1 1 x2
Z ∞ 2
sin x
4. dx
1 x
 2
sin x 1
Solution: Let f (x) = and choose g(x) = 2
x x
Z ∞ Z ∞ 2
sin x
By comparison test g(x) dx convergent, then dx is convergent.
1 1 x2
Z ∞
x tan−1 x
5. √
3
dx
1 1 + x4
tan−1 x 1
Solution: Let f (x) = √
3
and choose g(x) = 1/3
x1/3 1 + x−4 x

tan−1 x

f (x) x1/3 3 1 + x−4
lim = lim
x→∞ g(x) x→∞ 1
x 1/2
π
= .
2

116
∞ ∞
x tan−1 x
Z Z
∴ g(x) dx divergent, then √3
dx is divergent.
1 1 1 + x4
Remark:The improper integral of type I resembles in many respects an infinite series. It is
interesting to consider the analogies between the results of the chapter on series and those
of this chapter.
In fact, we can see the following similarities:
Z ∞ ∞
X
f (x)dx, corresponding to gk
a a

i.e., x corresponding to k and f (x) corresponding to gk .


Z λ X λ
f (x) dx corresponding to sn = gk .
a a X
However the analogue of the theorem ”If gn converges, then lim gn = 0.”
Z ∞ n→∞

Namely "If f (x) dx converges, then lim f (x) = 0" does not hold.
a x→∞
For example, set (
1 − |x|, if 0 ≤ |x| ≤ 1
h(x) =
0 if 1 ≤ |x| ≤ ∞
and

n − 1/n2 n n + 1/n2

inf ty
X
g n2 [x − n] , then

f (x) =
2


X
g n2 [x − n]

f (n) =
(2
1, , if k = n
=
0 if k 6= n.

The graph of f in the neighborhood of x = n is shows in the figure.

∴ lim f (x) 6= 0. [∵ f (n) = 1, ∀n = 1, 2, 3, . . .]


x→∞

But
Z ∞ Z ∞
∞X
g n2 [x − n] dx

f (x) dx =
0 0 2

!
X Z n Z n+1/n2
g n2 [x − n] dx + g n2 [x − n] dx
 
=
2 n−1/n2 n

117
∞  
X 1 1
= +
2
2n2 2n2

X 1
=
2
n2
= ξ(2) − 1

Here the improper integral convergence. Absolute Convergence of Type I:


Z ∞ Z ∞
Definition 5. 1. We say that f (x) dx converges absolutely, if |f (x)| dx is con-
a a
vergent.
Z ∞ Z ∞
2. If f (x) dx converges but not absolutely, then we say that f (x) dx is con-
a a
verges conditionally.
Z ∞
Theorem 9.4. If f is continuous on [a, +∞) and f (x) dx converges absolutely, then
Z ∞ a

f (x) dx is convergent.
a

Proof. We have,
0 ≤ |f (x)| − f (x) ≤ 2|f (x)|, a ≤ x ≤ ∞.
Z∞ Z ∞
Since f (x) dx converges absolutely, then |f (x)| dx is convergent.
a Z a

Therefore 2 |f (x)| dx is convergent.
a Z ∞
Hence by comparison test (|f (x)| − f (x)) dx is convergent. Now,
a

Z ∞ Z ∞
f (x) dx = (f (x) + |f (x)| − |f (x)|) dx
a a
Z ∞
(|f (x)| − {|f (x)| − f (x)}) dx
=
a
Z ∞ Z ∞ Z ∞
f (x) dx = |f (x)| dx − {|f (x)| − f (x)} dx
a a a

Z ∞
Hence f (x) dx is convergent. [∵ both the integrals on the right side are convergent]
a

Comparison tests of Type III:

Theorem 9.5. If f and g are continuous on (a, b] and are unbounded at a and if 0 ≤ f (x) ≤
Z b Z b
g(x) (a, b] and g(x) dx converges, then f (x) dx converges.
a+ a+

Proof. Let  > 0 be given, then


Z b Z b Z b
f (x) dx ≤ g(x) dx ≤ g(x) dx
a+ a+ a+

118
. As  → a+ the integral on left side increases (∵ f (x) ≥ 0) and remains bounded by
Z b
g(x) dx.
a+ Z b
Hence lim+ f (x) dx exists.
→a a+
Z b
∴ f (x) dx converges.
a+

Note: The above theorem can also be proved by using Theorem 0.9 as follows put x =
1
a + , then
t Z b
Z ∞
f (x) dx = f (a + t−1 ) t−2 dt
a+ (b−a)−1

Therefore the integral on left side converges if the integral on right side converges, since

0 ≤ f (a + t−1 ) t−2 ≤ g(a + t−1 ) t−2 , 0 < (b − a)−1 ≤ t < ∞


Z ∞
the integral on right side converges if g(a + t−1 ) t−2 dt converges by Theorem 0.1.
(b−a)−1
This is true since Z b Z ∞
g(x) dx = g(a + t−1 ) t−2 dt
a+ (b−a)−1

converges.

Theorem 9.6. If f and g are continuous on (a, b] and are unbounded at a and if 0 ≤ g(x) ≤
Z b Z b
f (x) (a, b] and g(x) dx diverges, then f (x) dx diverges.
a+ a+
Z b Z b Z b
Proof. If f (x) dx converges, then so does g(x) dx, by Theorem 0.9 so f (x) dx
a+ a+ a+
must diverges.
Z b
dx
Example: Consider the integral p
dx
a+ (x − a)
First, if p ≤ 0, then the integral becomes proper.
Next, if 0 < p < 1 we have
Z b Z b
dx dx
p
dx = lim+ p
dx
a+ (x − a) →0 a+ (x − a)
b
(x − a)−p+1

= lim+
→0 −p + 1 a+
1 
(b − a)1−p − 1−p

= lim+
→0 1 − p
1
= .
(1 − p) (b − a)1−p

∴ The improper integral converges if 0 < p < 1.


If p = 1,

119
Z b Z b
dx dx
dx = lim+ dx
a+ (x − a) →0 a+ (x − a)
1
= lim+ [log(b − a) − log ]
→0 1 − p
= +∞.

∴ The improper integral diverges, if p = 1


If p > 1,
Z b Z b
dx dx
p
dx = lim+ p
dx
a+ (x − a) →0 a+ (x − a)
b
(x − a)−p+1

= lim+
→0 −p + 1 a+
1  1−p
− 1−p

= lim+ (b − a)
→0 1 − p
= −∞.
Z b
dx
∴ The improper integral diverges if p > 1. Thus p
dx converges if and only if
a+ (x − a)
(b − a)1−p
−∞ < p < 1. In this case its value is .
1−p
Theorem 9.7. Let f (x) ≥ 0, g(x) ≥ 0 on (a, ∞] and both be unbounded at a.
f (x)
Suppose lim+ = A.
x→a g(x)
Z b Z b
1. If A is finite and g(x) dx converges, then so f (x) dx.
a+ a+
Z b Z b
2. If A 6= 0 or + ∞ and g(x) dx diverges, then so f (x) dx.
a+ a+

Proof. (i) Since f (x) ≥ 0, g(x) ≥ 0. We have A ≥ 0 for  > 0, there exists c ∈ (a, b] such
that

f (x)
− A < , ∀ x ∈ (a, c)
g(x)
(A − ) g(x) < f (x) < (A + ) g(x), ∀ x ∈ (a, c) (9.8)
Z c Z c
∴ (A + )g(x) dx converges. Hence by Theorem 0.9 it follows that f (x) dx con-
a+ a+
verges. Since Z b Z c Z b
f (x) dx = f (x) dx + f (x) dx
a+ a+ c
Z b
∴ f (x) dx converges.
a+
(ii) If A 6= 0, then we can choose  so that A −  > 1, then by first part of the inequality
Z b Z b
in (9.8) and Theorem 0.10, it follows that f (x) dx diverges. Whenever ∴ f (x) dx
a+ a+

120
diverges if A = +∞, then for ω > 0 (however large). We can find c ∈ (a, b) such that

f (x)
> ω, ∀ x ∈ (a, c)
g(x)
i.e., f (x) > ω g(x), ∀ x ∈ (a, c)
Z b Z b
Again from Theorem 0.10 and the divergence of g(x) dx, it follows that f (x) dx
a+ a+
diverges.

Example:
Z π/2
1. Consider the integral log(sin x) dx.
0
Here f (x) = log(sin x) clearly f is continuous in (0, π/2) and f (x) ≤ 0 on (0, π/2].
1
Also f (x) → −∞ as x → 0+ . We can consider −f (x) and take g(x) = p , p > 1,
x
then
f (x)
lim+ − = lim −x−p log(sin x) = 0
x→0 g(x) x→0+
Hence by Theorem (9.7), it follows that
Z π/2 Z π/2
− f (x) dx = − log(sin x) dx converges,
0 0

Z π/2
since x−p dx converges,
0
Z π/2
∴ log(sin x) dx converges.
0

Z 1
log x
2. Consider the integral √ dx.
0
4
x
log x
Here f (x) = √ clearly f is continuous in (0, 1) and f (x) ≤ 0 on (0, 1]. Also f (x)
4
x
1
is unbounded at x = 0.We can consider −f (x) and take g(x) = 1/2 , p > 1, then
x

f (x)
lim+ − = lim −x1/4 log(sin x) = 0
x→0 g(x) x→0+
Z 1 1 Z
1
Since g(x) dx = 1/2
dx converges. Hence by Theorem (9.7), it follows that
Z 1 0 0 x Z 1
log x
− f (x) dx converges, hence √ dx converges.
0 0
4
x

Absolute Convergence of Type III:


Z b Z b
Definition 6. 1. If |f (x)| dx converges, we say that f (x) dx converges absolutely.
a+ a+

121
Z b Z b Z b
2. If f (x) dx converges, but |f (x)| dx diverges we say that f (x) dx condi-
a+ a+ a+
tional convergent.
Z b Z b
Theorem 9.8. If f is continuous on a < x ≤ b and |f (x)| dx converges, then f (x) dx
a+ a+
converges.
Proof. The proof is same as the proof of Theorem 0.4, except for a change in the limits of
integration.

9.1.4 Keywords
Bounded, Convergent and Divergent.

9.1.5 Terminal Problems


Z ∞
1. e−αx dx, where α > 0.
0
Z ∞
2. sin x dx.
0
Z ∞
1
3. dx.
1 xp
Z 1
1
4. dx, where 0 < k < 1.
0 xk
Z 1  
1 1
5. 2
sin dx.
0 x x
Z 1
1
6. dx.
−1 x2

9.1.6 Books for reference


1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd


Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

122
Block III: Test for Convergence of an
Improper Integral:

19
123
Block III
Unit 10: Test for Convergence of an Improper Integral:
10.1.1 Main Objectives
10.1.2 Test for Convergence of an Improper Integral
10.1.3 Keywords
10.1.4 Terminal Problems
10.1.5 Books for reference

124
10.1.1 Main Objectives
If the limit exists and is a finite number, we say the improper integral converges. If the limit
is ±∞ or does not exist, we say the improper integral diverges.

1. To test the improper integrals of type I is convergent or divergent.

2. To test the improper integrals of type II is convergent or divergent.

3. To test the improper integrals of type III is convergent or divergent.

4. To test the improper integrals of type IV is convergent or divergent.

10.1.2 Test for Convergence of an Improper Integral


Limit Test of Type I:

Theorem 10.9. [Limit test for convergence] Z ∞


p
If f is continuous on [a, ∞) and lim x f (x) = A, where p > 1, then f (x) dx converges
x→∞ a
absolutely.

Proof. Since lim xp f (x) = A,, corresponding to  = 1, there exists a number b > 0 such
x→∞
that

|xp f (x) − A| < 1, ∀ x ∈ [b, +∞)


but ||xp f (x)| − |A|| ≤ |xp f (x) − A|
∴ ||xp f (x)| − |A|| < 1, ∀ x ∈ [b, +∞)
xp |f (x)| < |A| + 1, ∀ x ∈ [b, +∞)
|A| + 1
∴ |f (x)| < , ∀ x ∈ [b, +∞)
xp

∞ ∞
|A| + 1
Z Z
1
Now, since dx converges (∵ p > 1), so does dx
0 xp 0 xp
Z ∞ Z b
Hence by comparison test |f (x)| dx converges, since f is continuous |f (x)| dx is
a a
proper integral. Thus
Z ∞ Z b Z ∞
|f (x)| dx = |f (x)| dx + |f (x)| dx
a a b

Z ∞
Therefore |f (x)| dx converges.
a

Theorem 10.10. [Limit test for divergence] Z ∞


If f is continuous on [a, ∞) and lim x f (x) = A 6= 0 or ± ∞, then f (x) dx diverges.
x→∞ a
The test fails if A = 0.

125
Proof. Case(i) Let A > 0 (or A = +∞), since lim x f (x) = A, there exists a number b
x→∞
such that
A
x f (x) > , ∀ b < x < ∞ (10.9)
2
[If A = +∞ the right hand side of (10.9) may be taken as any number in particular 1]

A 1
∴ f (x) > , ∀b<x<∞
2 x
Z ∞ Z ∞
1
Since dx diverges (to +∞), by comparison test f (x) dx diverges and since f is
b xZ b
b
continuous f (x) dx is proper integral. Thus
a

Z ∞ Z b Z ∞
f (x) dx = f (x) dx + f (x) dx
a a b

Z ∞
Therefore f (x) dx diverges.
a
Case(ii) Let A < 0 (or A = −∞). Then by Case(i) the integral lim − f (x) dx diverges.
x→∞
R ∞ dx
Case(iii) Let us consider the improper integral 1 2 .
x
1 1
Here f (x) = 2 , then lim x → ∞x f (x) = lim x → ∞ = 0
x x
R ∞ dx
Next consider the improper integral 1 .
x log x
1 1
Here f (x) = , then lim x → ∞x f (x) = lim x → ∞ =0
x log x log x
Thus in either case A = 0. But in the first integral converges, whereas in the second integral
diverges.

Exercises:
Examine the following integrals for convergence using limit tests:
Z ∞
2
1. e−x dx
0 Z ∞
p
Solution: We know that lim x f (x) dx = A, p > 1 & f ∈ [a, ∞), then f (x) dx
x→∞ a
converges absolutely. Choose

2
lim x2 f (x) = lim x2 e−x
x→∞ x→∞
x2 h ∞ i
= lim x2
x→∞ e ∞
2x
= lim 2
x→∞ 2 x ex
lim x2 f (x) dx = 0
x→∞

∴ The given improper integral converges absolutely.


Z ∞
cos x
2. √ dx
0 1 + x3

126
Solution:

cos x xp cos x
lim xp √ = lim √
x→∞ 1 + x3 x→∞ x3/2 1 + x−3
cos x
= lim √
x→∞ x3/2−p 1 + x−3

put p = 5/4
cos x
= lim √
x→∞ x3/2−5/4 1 + x−3
cos x
= lim √
x→∞ x1/4 1 + x−3
= 0.

∴ The given improper integral converges absolutely.


Z ∞
1
3. √ dx
0 1 + 2x2 1/2
Solution:Here x2 = x and f ∈ [a, ∞).
By second theorem of limit, we know that

lim x f (x) = A 6= 0 or ± ∞
x→∞

Z ∞
f (x) dx diverges
0

. Consider

1
lim x f (x) = lim x √
x→∞ x→∞ 1 + 2x2
1
= lim √
x→∞ 2 + x−2
1
= √ 6= 0.
2

∴ The given improper integral diverges.


Z ∞
7 e−x − 1
4. √3
dx
0 1 + 2x2
Solution: By second theorem of limit, we know that

(7 e−x − 1)
lim x f (x) = lim x
x→∞ x→∞ (1 + 2x2 )1/3
x (7 e−x − 1)
= lim
x2/3 (2 + x−2 )1/3
x→∞

7 e−x − 1
= lim −1/3
x→∞ x (2 + x−2 )1/3
= −∞

127
∴ The given improper integral diverges.
Z ∞
log x
5. Show that √p
dx converges absolutely for 0 < p < 3 and diverges else-
1/2 1 + x3
where except p = 0.
Solution:Case(i): We know that

(log x)
lim xa f (x) = lim xa
x→∞ x→∞ (1 + x3 )1/p
xa (log x)
= lim
x→∞ x3/p (1 + x−3 )1/p
log x
= lim
x3/p−a (1 + x−3 )1/p
x→∞

Choose a = 3/q
log x h∞i
= lim
x→∞ x3/p−3/q (1 + x−3 )1/p ∞
1
x
= lim h i  
x→∞ 3
p
− 3
q
x3/p−3/q−1 (1 + x−3 )1/p + x3/p−3/q 1
p
(1 + x−3 )1/p−1 (−3x)
= 0.

∴ The given improper integral converges absolutely for 0 < p < 3.


Case(ii): If p ≥ 3, then

(log x)
lim x f (x) = lim x
x→∞ x→∞ (1 + x3 )1/p
x (log x)
= lim
x→∞ x3/p (1 + x−3 )1/p
log x
= lim
x→∞ x3/p−1 (1 + x−3 )1/p
= ∞.

∴ The given improper integral diverges.

Conditional Convergence of Type I: In this section we develop a result analogous to


Leibnitz’s theorem concerning the convergence of alternating series. In the present case a
trigonometric factor such as sin x or cos x in the integrand takes the place of the factor (−1)n
in the general term of the alternating series.
Integrand with the oscillating sign:-
We shall prove the following

Theorem 10.11. If g is continuous and decreasing for a ≤ x < ∞ and lim g(x) = 0, then
Z ∞ x→∞

g(x) sin x dx convergence.


a

Proof. Since g is decreasing and lim g(x) = 0, we must have


x→∞

g(x) ≥ 0, for a ≤ x < ∞.

128
Let a < mπ < nπ < λ ≤ (n + 1)π, where m and n are integer, then
Z ∞ Z mπ n−1 Z
X (k+1) π Z λ
g(x) sin x dx = g(x) sin x dx + g(x) sin x dx + g(x) sin x dx
a a k=m kπ nπ
(10.10)
Keeping m fixed. Let λ → ∞ then n → ∞, since
Z λ Z (n+1) π
g(x) sin x dx ≤ g(x) | sin x| dx = 2 g(n π)
nπ nπ

and since g(x) → 0 as x → +∞, the last term on the right hand side of (10.10) approaches
to 0, when λ → ∞. Also the first term on the right hand side of (10.10) is proper integral
for each fixed m. So to prove the theorem it suffices the prove that the series
∞ Z
X (k+1) π
g(x) sin x dx (10.11)
k=m kπ

converges. we put
Z (k+1) π
bk = g(x) sin x dx

Z (k+1) π
= g(x) |sin x| dx

Since sin x does not change sign for k π ≤ x ≤ (k + 1) π and g(x) ≥ 0. Because g(x) is
decreasing, we have
Z (k+1) π Z (k+1) π
g(k π + π) |sin x| dx ≤ bk ≤ g(k π) |sin x| dx
kπ kπ
i.e., g(k π + π) ≤ bk ≤ g(k π) (10.12)
Replacing k by k − 1 in (10.12), we get
g(k π) ≤ bk ≤ g(k π − π) (10.13)

Combining (10.12) and (10.13), we obtain

≤ bk ≤ bk−1 ≤ 2 g(k π − π)

Hence {bk } is decreasing and lim bk = 0. (∵ g(x) → 0 as x → 0)


k→∞
Thus all condition for the Leibnitz’s theorem are satisfied. Hence by Leibnitz’s theorem it
X
follows that (−1)k bk converges, thus is equivalent to saying that the series (10.11) is
convergent.
Z ∞
sin x
Example: Consider dx
0 x
1
We have already shown that the integral is not convergent absolutely. Since is not contin-
x

129
uous at x = 0, we shall write the integral as
Z ∞ Z 1 Z ∞
sin x sin x sin x
dx = dx + dx.
0 x 0 x 1 x

sin x
Since lim+ = 1, the first inter=gral on the right hand side is proper. Next put g(x) =
x→0 x
1
, then g satisfies all the conditions of the above theorem.
x
i.e., g(x) is continuous on [1, +∞), decreasing on [1, +∞) and lim g(x) = 0. Therefore
x→∞
by the above theorem
Z ∞ Z ∞
sin x
g(x) sin x dx = dx converges.
1 1 x
Z ∞
sin x
Hence dx converges conditionally.
0 x
Sufficient Conditions for Conditional Convergence of Type I:
Z ∞
Theorem 10.12. If g(x) is continuous and decreasing on [a, +∞), lim g(x) = 0 and g(x) dx
x→∞ a
Z ∞
diverges (converges), then g(x) | sin x| dx diverges (converges).
a

Proof. Since g is decreasing and approaches to 0, it is necessary non-negative . Let a <


mπ < nπ < λ ≤ (n + 1)π, where m and n are integer, then
Z ∞ Z mπ n−1 Z
X (k+1) π Z λ
g(x) sin x dx = g(x) sin x dx + g(x) sin x dx + g(x) sin x dx
a a k=m kπ nπ
(10.14)
Keeping m fixed. Let λ → ∞ then n → ∞, since lim g(x) = 0 then the last term in (10.14)
x→∞
approaches to 0 as λ → ∞ (as in the proof of the previous theorem).
Also since g is decreasing, we have
Z (k+1) π Z (k+1) π
bk−1 ≥ 2 g(k π) ≥ 2 g(x) dx, where bk = g(x) | sin x| dx
kπ kπ
n
X Z (n+1) π
∴ bk ≥ 2 g(x) dx → ∞ as λ → ∞
k=m+1 (m+1) π

Z mπ
Since g(x) dx diverges. Hence the second term on the right hand side of (10.14) di-
a
verges.Z

Hence g(x) | sin x| dx diverges
aZ
b
Next, let g(x) dx converges, since | sin x| ≤ 1 and g(x) is non-negative. We have
a

|g(x) sin x| ≤ |g(x)|


Z ∞ Z ∞
∴ g(x) |sin x| dx ≤ g(x) dx
a a

130
Z ∞
By comparison test, it follows that g(x) | sin x| dx converges.
a

Corollary 10.1. If g is continuous and decreasing for x ∈ [a, ∞) and if lim g(x) = 0,
x→inf ty
Z ∞ Z ∞
then g(x) sin(αx + β) dx and g(x) cos(αx + β) dx converges (α 6= 0).
a a

Proof. Using the Theorem 0.8 and changing of variable.

Corollary 10.2. If g is continuous and decreasing for x ∈ [a, ∞) and if lim g(x) = 0. If
x→inf ty
Z ∞
a
n is an integer such that n > , then g(x) sin x dx ≤ 2 g(n π).
π nπ

Z (k+1) π
Proof. Put bk = g(x) sin x dx , then we have

Z ∞ ∞
X
g(x) sin x dx = ± (−1)k bk
nπ k=n+1

. Also, bn ≤ 2 g(n π). Now using the result.


" ∞
X ∞
X
k
|A − sn | = (−1) bn ≤ bn+1 , where A = (−1)n bn and sn is the nth partial sum
k=n+1 n=1
of the series. This is because

lim s2n = A = lim s2n+1 ,


n→∞ n→∞

and s2n is decreasing, s2n+1 is increasing.


∴ A lies between two consecutive terms of {sn }

∴ |A − sn | ≤ |sn+1 − sn = bn+1 |]

Example:
Z ∞
1. Consider sin x2 dx
0
2 1
Set x = t, then 2 x dx = dt or dx = √ dt. Also when x = 0, then t = 0 and
2 t
when x = ∞, then t = ∞.

1 ∞ sin t
Z Z
2
sin x dx = √ dt
0 2 0 t
1 ∞ sin t
Z 1 Z
1 sin t
= √ dt + √ dt
2 0 t 2 1 t

The first integral on the right hand side is a proper integral to examine the second
1
integral we take g(t) = √ , then g is continuous and decreasing on [1, ∞) and
t
lim g(t) = 0.
t→∞

131
Z ∞ Z ∞
1
Further g(t) dt = √ dt is divergent.
1 1 Zt ∞
Therefore by Theorem 0.8 g(t) | sin x| dt diverges
Z ∞ 1
sin t
Therefore the integral √ dt is not absolutely convergent.
1 Z t

sin t
However by Theorem 0.7 √ dt converges.
1 t
Therefore the given integral is conditionally convergent.
Z ∞
sin x
2. Consider dx
1 x
1
 
By using Theorem 0.7,Z ∞ the integral is convergent ∵ with g(x) = x
.
sin x π
Latter, we show that dx = .
0 x 2

Z nπ Z ∞
π sin x sin x
− dx = dx
2 0 x nπ x
≤ 2 g(n π)
2
=

on using Corollary 0.2.

Limits Tests of Type III:

Theorem 10.13. {For Convergence} Z b


p
If f is continuous on a < x ≤ b and (x − a) f (x) = A, where 0 < p < 1, then |f (x)| dx
a+
converges.

Proof. Since (x − a)p f (x) = A corresponding to  = 1 then there exists c > a, such that

|(x − a)p f (x) − A| < 1, ∀ x ∈ (a, c)


Since ||(x − a)p f (x)| − |A|| ≤ |(x − a)p f (x) − A|
we have |(x − a)p f (x)| < 1 + |A|, ∀ x ∈ (a, c)
1 + |A|
∴ |f (x)| <
(x − a)p
Z b Z c
dx dx
Since p
dx converges, then p
dx converges.
Z c a+ (x − a) Z c a+ (x − a)
1 + |A|
∴ converges, so |f (x)| dx converges by comparison test.
(x − a)p dx
a+ Z a+
b
Hence |f (x)| dx converges.
a+

Theorem 10.14. {For Divergence} Z b


If f is continuous on a < x ≤ b and (x − a) f (x) = A(6= 0 or ± ∞), then f (x) dx
a+
diverges.

132
A
Proof. First let A > 0, then corresponding to  = > 0 then there exists c, such that
2

A
|(x − a) f (x) − A| < , ∀ x ∈ (a, c)
2
A
i.e., (x − a) f (x) >
2
A
f (x) >
2 (x − a)

Z b Z c
1 1
Since dx diverges, so dx diverges by comparison test.
a +x − a a+ x − a
Z b
Hence f (x) dx diverges.
a+
If A = +∞, we can find c, such that

i.e., (x − a) f (x) > 1, ∀ x ∈ (a, c)


1
f (x) > , ∀ x ∈ (a, c)
(x − a)

Z b Z b
dx
Since dx divergence, the divergence of f (x) dx
a+ (x − a) a+
If A < 0 or A = −∞ we agree with −f.

Remark: Consider the integral Z 1


dx
√ (10.15)
0+ x
Z 1/2
dx
(10.16)
x log x1

0+

then we have
1 √
lim+ x f (x) = lim+ x √ = lim+ x = 0
x→0 x→0 x x→0
and
1 1
lim+ x f (x) = lim+ x 1
 = lim  =0
x→0 x→0 x log x
+
x→0 log x1
But the first integral is convergent (∵ p = 12 < 1).
−1
For second integral put x1 = t, then dx = 2 dt and also when x → 0+ , then t → ∞ :

t
1
when x = , then t = 2
2
1/2 2 
−1
Z Z
dx 1
 = t dt
0+ x log x1 ∞ t2 log t
Z ∞
1
= dt
2 t log t

which is divergent.

133
Example:

Z 1/2
 α
1
1. Consider log dx
0+  x
α
√ 1
Now, lim+ x log = 0, if α > 0.
x→0 x
Therefore by limit test
 the improper
α integral is convergent if α > 0.
1 1
If α ≤ 0, then lim+ log = lim+ −α = 0
x→0 x x→0 log x1
∴ The integral is proper if α ≤ 0.
Z 1
2. Consider tx−1 e−t dt
0+
We have (
0, if x > 1
lim+ f (t) =
t→0 1, if x = 1
Therefore the integral is proper if x ≥ 1.
So, let x < 1 then

lim t1−x f (t) = lim t1−x tx−1 e−t


t→0+ t→0+
= 0 if 0 < x < 1

Therefore by limit test the improper integral is convergent if 0 < x < 1 (0 < 1 − x <
1), finally

lim t f (t) = lim t tx−1 e−t


t→0+ t→0+
= lim tx e−t
t→0+
(
1, if x = 0
=
∞, if x < 0
Z 1
Therefore by limit test, the improper integral is divergent. Thus tx−1 e−t dt con-
0+
vergent for x > 0 and divergence for x ≤ 0.

Oscillating Integrand of Type III:

Theorem 10.15. If g is continuous on a < x ≤ b,g(x) (x− a)2 is increasing on a < x ≤ b


Z b
2 1
and if lim+ g(x) (x − a) = 0, then g(x) sin dx converges.
x→a a+ x−a
1 1 1
Proof. Put or x = + a, then dx = − 2 dt.
x−a t t
1
When x → a+ , then t → ∞ and when x = b, then t = . Hence
b−a
Z b   Z ∞
1
g(x) sin dx = g(a + t−1 ) t−2 sin t dt (10.17)
a+ x−a (b−a)−1

134
 
1 1
The function g a + is continuous on [(b − a)−1 , ∞), because g(x) is continuous on
t t2
(a, b].
lim g(a + t−1 ) t−2 = lim+ g(x) (x − a)2 = 0
t→+∞ a→0
 
1 1
As x increases t decreases and hence g a + decreases on [(b − a)−1 , ∞), thus all
t t2
the conditions of Theorem 0.7 areZ satisfied. Hence
 by Theorem 0.7 the integral on right side
b
1
of (10.17) converges. Therefore g(x) sin dx converges.
a+ x−a
Note:If the improper integral of type II and type IV are given, then one can reduce these
integrals to type I and type III respectively by change of variables. Hence to discuss their
converges the theorems proved aboveZ can be employed. For instance, if f is continuous of
Z b b
(−∞, b], then f (x) dx becomes f (−t) dt when we set x = −t.
−∞ −∞
Z ∞ Z b
Hence f (−t) dt = f (x) dx converges absolutely, if
−b −∞

lim tp f (−t) = lim (−x)p f (x) = A (c, +∞), p > 1


t→+∞ x→−∞

and the integral diverges, if

lim t f (−t) = lim (−x) f (x) = A 6= 0 (or = ±∞)


t→+∞ x→−∞
Z b−
Similarly, if f is continuous on [a, b) and lim− f (x) does not exits, then f (x) dx be-
x→b a
Z b−a
comes f (b − t) dt, when we set x = b − t (change x = −t will also do). Hence the
0+
integral converges absolutely, if

lim tp f (b − t) = lim− (b − x)p f (x) = A, (0 < p < 1)


t→0+ x→b

and diverges, if

lim t f (b − t) = lim− (b − x) f (x) = A 6= 0 (or = ±∞).


t→0+ x→b

Uniform Convergence:
Z ∞
Definition 7. 1. Suppose f (x, t) dt converges to F (x) for each x ∈ [A, B].
Z R a Z ∞
Let f (x, t) dt = SR (x), we say that f (x, t) dt converges uniformly to F (x)
a a
in the interval [A, B], if for every  > 0, there exists a number Q independent of
x ∈ [A, B], such that

|F (x) − SR (x)| <  whenever R > Q, x ∈ [A, B].

135
Z b
2. Suppose f (x, t) dt converges to F (x) for each x ∈ [A, B].
Z b a+ Z ∞
Let f (x, t) dt = Sr (x), we say that f (x, t) dt converges uniformly to F (x)
r a+
in the interval [A, B], if for every  > 0, there exists a number q independent of
x ∈ [A, B], such that

|F (x) − Sr (x)| <  whenever a < r < q, x ∈ [A, B].

Example:

Z ∞
1. Consider the integral e−x t dt, 1 ≤ x ≤ 2.
0
We have
Z R
SR (x) = e−x t dt
0
 −x t R
e
= −
x 0
−x R
e 1
SR (x) = − +
x x
1
lim SR (x) =
R→∞ x

1
i.e., F (x) = . Let  > 0 be given, then
x
 −x R 
1 e 1
|F (x) − SR (x)| = − − +
x x x
−x R
e
=
x
−R
≤ e , ∀ x ∈ [1, 2]
< e−Q , ∀ R > Q
 
1
< , if Q = log

Thus |F (x) − SR (x)| < , ∀ R > Q, x ∈ [1, 2].
Z ∞
1
Hence e−x t dt, converges uniformly to in [1, 2].
0 x
Z ∞
2. Consider the integral x e−x t dt, where 0 ≤ x ≤ 1.
0
We have
Z R
SR (x) = x e−x t dt
0
 −x t R
xe
=
−x 0

136
= 1(− e−x R
1 − e−x R , if 0 < x ≤ 1
SR (x) =
0, if x = 0.
(
1, if 0 < x ≤ 1
So F (x) = lim SR (x) =
R→∞ 0, if x = 0.

The improper integral converges to F (x) for each x ∈ [0, 1]. Now
(
e−x R if 0 < x ≤ 1
|F (x) − SR (x)| =
0 if x = 0.

1
If the convergence is uniform, then for  = there exits a number Q independent of
2
x, such that
|F (x) − SR (x)| = e−x R for R > Q, 0 < x ≤ 1

But this is false, since for each R > 0.

1
lim e−x R = 1 (≮ )
x→0 2
The Weierstrass M-test

Theorem 10.16. Suppose

1. f (x, t) is continuous for a ≤ t < ∞, A ≤ x ≤ B.

2. M (t) is continuous for a ≤ t < ∞.

3. |f (x, t)| ≤ M (t), for a ≤ t < ∞, A ≤ x ≤ B.


Z ∞
4. M (t) dt converges, then
a
Z ∞
f (x, t) dt converges uniformly in [A, B].
a
Z ∞
Proof. Since 0 ≤ |f (x, t)| ≤ M (t), a ≤ t < ∞, A ≤ x ≤ B and M (t) dt con-
Z ∞ a Z

verges. By comparison test it follows that |f (x, t)| dt converges and hence f (x, t) dt
a a
converges say to F (x), A ≤ x ≤ B we have
Z ∞ Z R
|F (x) − SR (x)| = f (x, t) dt − f (x, t) dt
a a
Z ∞
= f (x, t) dt
R
Z ∞
≤ |f (x, t)| dt
ZR∞
≤ M (t) dt
R

137
Z ∞ Z ∞
Now, since M (t) dt converges M (t) dt → 0 as R → ∞.
R R
Therefore, given  > 0 there is a number Q, such that
Z ∞ Z ∞
M (t) dt = M (t) dt < , whenever R > Q
R R

Hence
|F (x) − SR (x)| < , whenever R > Q, A ≤ x ≤ B.

This proves that Z ∞


f (x, t) dt converges uniformly to F (x).
a

Theorem 10.17. Suppose

1. f (x, t) is continuous for a < t ≤ b, A ≤ x ≤ B.

2. M (t) is continuous for a < t ≤ b.

3. |f (x, t)| ≤ M (t), for a < t ≤ b, A ≤ x ≤ B.


Z b
4. M (t) dt converges, then
a+
Z b
f (x, t) dt converges uniformly in [A, B].
a+

Proof. Similar to the above proof of Theorem (10.16).

Application of Uniform Convergence:

1. Uniform Convergence and Continuity

Theorem 10.18. Suppose

(a) f (x, t) is continuous for a ≤ t < ∞, A ≤ x ≤ B.


Z ∞
(b) f (x, t) dt converges uniformly to F (x) in [A, B], then
a

F is continuous on [A, B].


Z ∞
Proof. Let  > 0 be given, since f (x, t) dt converges uniformly to F (x), in
a
[A, B] then there exists a number Q independent of x, such that
Z ∞
|F (x) − SR (x)| = f (x, t) dt < , (10.18)
a

138
whenever R > Q and A ≤ x ≤ B. If x0 ∈ [A, B], then
Z ∞ Z ∞
|F (x) − F (x0 )| = f (x, t) dt − f (x0 , t) dt
a a
Z R Z ∞  Z R Z ∞ 
= f (x, t) dt + f (x, t) dt − f (x0 , t) dt + f (x0 , t) dt
a R a R
Z R Z R Z ∞ Z ∞
= f (x, t) dt − f (x0 , t) dt + f (x, t) dt − f (x0 , t) dt
a a R R
Z R Z R Z ∞ Z ∞
≤ f (x, t) dt − f (x0 , t) dt + f (x, t) dt + f (x0 , t) dt
a a R R
Z R Z R
< f (x, t) dt − f (x0 , t) dt + 2 
a a

letting x → x0 , we obtain

0 ≤ lim |F (x) − F (x0 )| ≤ 


x→x0

Since  was arbitrary,

lim F (x) = F (x0 )


x→x0

Therefore F is continuous at x0 , since x0 is an arbitrary point in [a, b], it follows that


F is continuous on [A, B].

Examples:

Z ∞
(a) Consider the improper integral e−x t dt, 0 < x < ∞. We know that this
0
integral is uniform convergent in [A, B], where 0 < A ≤ x ≤ B. So the value
of the improper integral F (x) is continuous on [A, B]. Since
Z ∞
1
F (x) = e−x t dt =
0 x

and since given any x0 ∈ (0, ∞), we can find A and B such that 0 < A < B
1
and x0 ∈ [A, B] the continuity of at x0 ∈ (0, ∞) follows the above theorem
x
1
continuous on (0, ∞).
x
Z ∞
(b) Consider the improper integral x2 t e−x t dt, if x 6= 0, then for any R > 0,
0

139
we have
Z R
x2 t e−x t dt = −x R e−x R − e−x R + 1
0
Z R (
2 −x t 1, if x 6= 0
∴ lim x te dt =
R→∞ 0 0 if x = 0

Therefore if F (x) is the value of the improper integral, then we have


(
1, if x 6= 0
F (x) =
0 if x = 0

Therefore F is not continuous at x = 0. Hence the given improper integral does


not converges uniformly in any interval containing the origin.

2. Uniform Convergence and Integration

Theorem 10.19. Suppose

(a) f (x, t) is continuous for a ≤ t < ∞, A ≤ x ≤ B.


Z ∞
(b) f (x, t) dt converges uniformly to F (x) in [A, B], then
a
Z B Z ∞ Z B 
F (x) dx = f (x, t) dx dt
A a A
or Z B Z ∞  Z ∞ Z B 
F (x) dt dx = f (x, t) dx dt
A a a A
Z ∞
Proof. Let  > 0 be given, since f (x, t) dt converges uniformly to F (x), in
a
[A, B] then there exists a number Q independent of x, such that
Z ∞

f (x, t) dt < , (10.19)
R B−A

whenever R > Q and A ≤ x ≤ B. If x0 ∈ [A, B], then


Z B Z ∞  Z B Z ∞
f (x, t) dt dx ≤ f (x, t) dt dx
A R A R
Z B

<
A B−A
= 
Z B Z ∞ 
i.e., f (x, t) dt dx < 
A a

140
This implies,
Z B Z ∞ 
lim f (x, t) dt dx = 0
R→∞ A R
Z B Z R 
i.e., lim F (x) − f (x, t) dt dx = 0
R→∞ A a
Z B Z B Z R 
i.e., F (x) dx = lim f (x, t) dt dx
A R→∞ A a
Z R Z B 
= lim f (x, t) dx dt
R→∞ a A
Z B Z ∞ Z B 
F (x) dx = f (x, t) dx dt
A a A

This completes the proof.



e−p t − e−r t
Z  
q
Example: Show that dt = log , 0 < p < q.
0 t p
Z ∞
1
Solution: We know that this integral e−x,t dt is uniform convergent to F (x) =
0 x
in [p, q]. Since 0 < p < q, we have
  Z p
q 1
log = dx
p q x
Z p Z ∞ 
−x t
= e dt dx
q 0
Z ∞ Z p 
−x t
= e dx dt
0 q
Z ∞  −x t q
e
= − dt
0 t p
Z ∞ −p t
− e−r t
 
q e
log = dt.
p 0 t

as desired.

3. Uniform Convergence and Differentiation

Theorem 10.20. Suppose


∂[f (x, t)]
(a) f (x, t), is continuous for a < t ≤ b, A ≤ x ≤ B.
∂x
Z ∞
(b) f (x, t) dt converges to F (x) for A ≤ x ≤ B and
a
Z ∞
∂[f (x, t)]
(c) dt converges uniformly on x ∈ [A, B], then
a ∂x
Z ∞
0 ∂[f (x, t)]
F (x) = dt
a ∂x

141
or Z ∞  Z ∞
d ∂[f (x, t)]
f (x, t) dt = dt
dx a a ∂x
Z ∞
∂[f (x, t)]
Proof. Put dt = φ(x), A ≤ x ≤ B.
Z ∞ a ∂x
∂[f (x, t)]
Since dt converges uniformly in [A, B], φ is continuous on [A, B] and
a ∂x
Z h Z h Z ∞ 
∂[f (x, t)]
φ(x) dx = dt dx
A A a ∂x
Z ∞ Z h 
∂[f (x, t)]
= dx dt
∂x
Za ∞ A

= [f (h, t) − f (A, t)] dt


a
Z h
φ(x) dx = F (h) − F (A)
A
∴ F 0 (h) = φ(h) for A ≤ h ≤ B
Z ∞
0 ∂[f (x, t)]
F (h) = dt
a ∂x

Z ∞
sin x
Example:Evaluate the integral e−x y dx, y > 0.
0 x
Solution: Let
Z ∞
sin x
F (y) = e−x y dx
x
Z0 ∞
d[F (y)]
then, = e−x y sin x dx
dy 0
 −x y R
−e
= lim (−y sin x − cos x)
R→∞ 1 + y 2
0
 −R y 
e (y sin R + cos R) 1
= lim −
R→∞ 1 + y2 1 + y2
d[F (y)] 1
= − , y>0
dy 1 + y2
⇒ F (y) = −arc tan y + c.

π π
As y → ∞, F (y) → 0 and arc tan y → , so c = .
2 2
π
Hence F (y) = − arc tan y, y > 0.
2

10.1.3 Keywords
Comparison Test, Weierstrass M-test, Absolute Convergent, Conditional Convergent and
Uniform Convergence.

142
10.1.4 Terminal Problems
Exercises:
Examine the following improper integrals for convergence
Z ∞
sin x
1. dx
0 x2
Z ∞
sin x
2. dx
0 x
Z ∞
x
3. 2
dx
0 x +1
Z ∞ 2
x −1
4. 2
dx
−7 x + 1
Z ∞ 2
x −1
5. dx
2 x4 − 9
Z ∞
f (x)
6. If f and g are continuous for x ≥ a, lim = A and if |g(x)| dx converges,
x→∞ g(x) a
Z ∞
then prove that f (x) dx converges absolutely.
a
Z ∞
7. If f is continuous and decreasing for x ≥ a and f (x) dx converges, then prove
a
that lim f (x) = 0
x→∞

Exercises:

Z ∞
1. Prove that if f continuous on [1, ∞) and |x f (x)| > 1, for x > 1, then f (x) dx
1
diverges.

2. If if f continuous on [a, ∞) and lim x (log x)p f (x) = A, for p > 1, then prove that
Z ∞ x→∞

f (x) dx converges absolutely.


a

3. If if f continuous on [a, ∞) and lim x (log x) f (x) = A 6= 0 or = ±∞, then prove


Z ∞ x→∞

that f (x) dx diverges.


a

Exercise:

1. Examine the following integrals for absolute and conditional convergence


Z ∞
sin x
• √3
dx
0 x2 + x + 1
Z ∞
cos(1 − 2x)
• √ √ dx
1 x 3 x2 + 1
Z ∞
sin 2x
• p dx
0 x log(x + 1)

143

e−x − 1
Z
• dx
0 e−x + 1
Z ∞
2. Discuss the integral g(x) cos x dx on the theorem 0.7 and 0.8, Corollaries 0.1 and
0
0.2

sin x1
Z 1 
1.
0 x3/2
Z 1
log x
2. √
0 x
Z 2 √
x
3.
1 log x
Z 1
4. x2 e1/x
0
Z    α
1
1
5. log log
0 x

sin x1
Z 1 
6. 3/2 log(1 + x−1 )
0 x

cos x1
Z 1 
7. 5/4 + x2 sin 1

0 x x

Show that the following integral are uniformly convergent in the intervals given
Z ∞
1. ex t dt, 1 ≤ x ≤ 2.
0

sin xt
Z 1

2. √ dt, 0 < A ≤ x ≤ B.
0+ t
Z ∞
x
3. dt 1 ≤ x ≤ 2.
0 x + t2
2

Z ∞
2 2
4. e−x t dt 1 ≤ x ≤ 10.
0
Z ∞
sin (x t)
5. dt, −10 ≤ x ≤ 10.
1 t2

ea x
Z
1. Show that eax sin(bx) dx = 2 [a sin(b x) − b cos(b x)].
a + b2
Z ∞
tan−1 (a x) − tan−1 (b x) π a
2. Show that dx = log .
0 x 2 b
Z ∞ −a2 x2 2 2
e − e−b x
3. Evaluate dx, 0 < a < b.
0 x
Z ∞ Z ∞
d −x t
4. Show that e dt = − e−x t t dt, 0 < x < ∞.
dx 0 0

144
10.1.5 Books for reference
1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd


Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

145
Block III: Gamma Function and Beta
Function

42
146
Block III
Unit 11: Gamma Function and Beta Function
11.1 Main Objectives
11.2 Gamma Function
11.3 Beta Function
11.4 Keywords
11.5 Terminal Problems
11.6 Books for reference

147
11.1.1 Main Objectives
Beta and Gamma are the two most popular functions in mathematics. Gamma is a single
variable function, whereas Beta is a two-variable function. The relation between beta and
gamma function will help to solve many problems in physics and mathematics.

11.1.2 Gamma Function


In this chapter we shall define a function know as the Gamma function. It will be defined in
terms of an improper integral. It has great importance in analysis and in the applications.

Definition 8. The Gamma function is denoted by Γ(x) and it is defined by


Z ∞
Γ(x) = e−t tx−1 dt, where 0 < x < ∞.
0+

Z ∞
Since the improper integral e−t tx−1 dt converges for x > 0 the Gamma function is
0+
well defined.

Theorem 11.21.
Γ(x + 1) = x Γ(x), 0 < x < ∞.

Proof. Integration by parts, gives

R R
tx −t 1 R −t x
Z Z
−t x−1
e t dt = e + e t dt
 x  x 
Now letting R → ∞ and  → 0, we obtain
Z ∞
1 ∞ −t x
Z
−t x−1
e t dt = e t dt
0+ x 0
1
i.e., Γ(x) = Γ(x + 1)
x
or Γ(x + 1) = x Γ(x).

Hence the theorem.

Corollary 11.3. For p = 1, 2, 3, . . ., we have

Γ(x + p) = (x + p − 1) (x + p − 2) · · · x Γ(x), x > 0.

Proof. By the above theorem, we have

Γ(x + p) = (x + p − 1) Γ(x + p − 1)
= (x + p − 1) (x + p − 2) Γ(x + p − 2)
..
.
Γ(x + p) = (x + p − 1) (x + p − 2) · · · x Γ(x).

148
Corollary 11.4.
Γ(n + 1) = n!.

Proof. By Corollary (11.3), we have

Γ(n + 1) = n (n − 1) (n − 1) · · · 1 Γ(1)
Γ(n + 1) = n! Γ(1). (11.20)

Now,
Z ∞
Γ(1) = e−t dt
0
Z R
= lim e−t dt
R→∞ 0
 R
= lim e−t 0
R→∞
Γ(1) = 1.

Therefore equation (11.20) becomes

Γ(n + 1) = n!.

Theorem 11.22.
lim Γ(x) = ∞.
x→0+
Z ∞
Proof. Since the integrand of the integral e−t tx−1 dt is positive, we have
0

Z ∞
Γ(x) = e−t tx−1 dt
Z0 1
> e−t tx−1 dt
0
Z 1
−1
> e tx−1 dt
0
1
=
ex
1
Now, lim+ =∞
x→0 ex
Hence lim+ Γ(x) = ∞.
x→0

Theorem 11.23. Γ(x) is continuous for 0 < x < ∞.

149
Proof. Let x0 be any positive real number, then we can find A and B such that 0 < A <
x0 < B.
Now if A ≤ x ≤ B, we have
|e−t tx−1 | ≤ e−t tB−1
Z ∞
and e−t tB−1 dt converges. Hence by Weierstrass-M test
Z ∞ 1
e−t tx−1 dt converges uniformly in [A, B].
1
Next, if A ≤ x ≤ B we have
|e−t tx−1 | ≤ tB−1
Z 1
and tB−1 dt converges. Hence by Weierstrass-M test
Z 1 0

e−t tx−1 dt converges uniformly in [A, B].


0 Z ∞
Thus e−t tx−1 dt converges uniformly in [A, B].
0
So, Γ(x) is continuous i A ≤ x ≤ B, since x0 ∈ [A, B]. Γ(x) is continuous at x0 , since x0
was arbitrary.

Theorem 11.24. lim+ x Γ(x) = 1.


x→0

Proof. We have,
x Γ(x) = Γ(x + 1)

Since Γ(x) is continuous for 0 < x < ∞, we have

lim x Γ(x) = lim x Γ(x + 1)


x→0+ x→0+
= Γ(1)
lim x Γ(x) = 1.
x→0+

Extension of definition of Gamma function:


For n = 1, 2, . . . , . we define

Γ(x + n)
Γ(x) = , −n < x < −n + 1
x (x + 1) (x + 2) · · · (x + n − 1)

Thus, we have defined Γ(x) for all x except x = 0, −1, −2, . . . , .

Theorem 11.25. {Euler-Poisson Integral}


Z ∞ √
−x2 π
e dx = .
0 2
2 −y 2
Proof. Consider the double integral of e−x over the two circular sectors D1 , D2 and the
square S as shown in the figure.

150
Y

(0, 2R)

(0, R) (R, R)
S

D1 D2


(R, 0) ( 2R, 0) X

2 −y 2
Since e−x is positive. We have
Z Z Z Z Z Z
−x2 −y 2 −x2 −y 2 2 −y 2
e dx dy ≤ e dx dy ≤ e−x dx dy
D1 S D2

Now we evaluate these integral by integration the center one in rectangular co-ordinates the
other two in polar co-ordinates.

Z R Z π/2 Z R Z R Z 2R Z π/2
−r2 −x2 −y 2 2
e r dr dθ ≤ e dx dy ≤ e−r r dr dθ
0 0 0 0 0 0
Z R 2
π  2
 2 π  2

i.e., 1 − eR ≤ e−x dx ≤ 1 − e2 R
4 0 4
Letting R → ∞, we obtain
Z R 2
−x2 π
e dx =
0 4
Z ∞ √
2 π
e−x dx = .
0 2

Problems:


 
1
1. Prove that Γ = π.
2
Solution: We have
Z ∞
Γ(x) = e−t tx−1 dt, 0 < x < ∞
0
1
Put x =
  Z2 ∞
1
Γ = e−t t−1/2 dt
2 0

151
Letting t = x2 , then, dt = 2 x dx
Z ∞ −1/2
2
= e−x x2 2 x dx
0
Z ∞
2
= 2 e−x dx
√0
π
= 2
2

 
1
Γ = π.
2

Z ∞
2. Show that Γ(x) = r x
e−r t tx−1 dt
0
Solution: We have
Z ∞
Γ(x) = e−t tx−1 dt, 0 < x < ∞
0
Letting t = r t, then
Z ∞
= e−r t rx−1 tx−1 r dt
Z0 ∞
= e−r t rx tx−1 dt
0
Z ∞
Γ(x) = r x
e−r t tx−1 dt.
0

Z ∞
3. Show that Γ(x) = 2 e−t t2x−1 dt
0
Solution:We have Z ∞
Γ(x) = e−t tx−1 dt
0
2
Putting x = s , then dt = 2 s ds. When t = 0, then s = 0 and When t → ∞, then
s → ∞.
Z ∞
Γ(x) = e−t tx−1 dt
Z0 ∞
2 x−1
= e−s s2 2 s ds
0
Z ∞
2
= 2 e−s s2x−1 ds
Z0 ∞
Γ(x) = 2 e−t t2x−1 dt.
0

 
1
4. Compute Γ −
2

152
Solution: We have

Γ(x + 1) = x Γ(x), ∀ x 6= 0, −1, −2, . . . , .


Γ(x + 1)
∴ Γ(x) =
x
1
Putting x = − , we get
2  
1
  Γ
1 2
Γ − =
2 1

2 
1
= −2 Γ
2

 
1
Γ − = −2 π.
2

5. Find lim + (x + n) Γ(x).


x→−n
Solution: We have by definition,

Γ(x + n)
Γ(x) = , −n < x < −n + 1 and n = 1, 2, 3, . . . , .
x (x + 1) · · · (x + n − 1)

Consider

(x + n) Γ(x + n)
lim + (x + n) Γ(x) = lim +
x→−n x→−n x (x + 1) · · · (x + n − 1)

Γ(x + n + 1)
= lim +
x→−n x (x + 1) · · · (x + n − 1)

Γ(1)
=
−n (−n + 1) · · · (−1)
1
lim + (x + n) Γ(x) = .
x→−n (−1)n n!

  √
1 (2 n)! π
6. Show that Γ n + = .
2 4n n!
Solution: On using Γ(x + 1) = x Γ(x) repeatedly, we have
   
1 1
Γ n+ = Γ n− +1
2 2
   
1 1
= n− Γ n−
2 2
    
1 3 3
= n− n− Γ n−
2 2 2
..
.     
1 3 31 1
= n− n− ··· Γ
2 2 22 2

153
3 1√
  
2n − 1 2n − 3
= ··· π
2 2 22

(2 n − 1) (2 n − 3) · · · 3, 1 π
=
2n √
1 · 3 · · · (2 n − 3) · (2 n − 1) π (2 · 4 · · · 2 n)
= n
×
2 (2 · 4 · · · 2 n)

1 · 2 · 3 · · · (2 n − 1) · (2 n) π
=
2n 2n (1 · 2 · · · n)
  √
1 (2 n)! π
Γ n+ = .
2 4n n!

Z 1−   x−1
1
7. Show that Γ(x) = log dt.
0+ t
Solution: We have
Z ∞
Γ(x) = e−t tx−1 dx
0+
 
1 1
Putting t = − log s or t = log , then dt = − ds
t s
When t → 0+ then s → 1− and when t → ∞ then s → 0+
Z 0+   x−1  
log s 1 1
Γ(x) = e log − ds
1− s s
Z 1−   x−1  
1 1
= s log ds
0+ s s
Z 1−   x−1
1
= log ds
0+ s
Z 1−   x−1
1
Γ(x) = log dt.
0+ t

Z 1  − 21

1
8. Evaluate log dt
0 t
Solution: We have
Z 1   x−1
1
Γ(x) = log dt
0 t
1
put x = , we get
2
  Z 1   −1/2
1 1
Γ = log dt
2 0 t
Z 1   −1/2
√ √
 
1 1
log dt = π, [∵ Γ = π]
0 t 2

Z 1
9. Evaluate (loge x)n dx
0

154
Solution: We know that
Z 1   x−1
1
Γ(x) = log dt
0 t

Consider
Z 1   n Z 1
n 1
[log x] dx = − log dx
0 0 x
Z 1   n
n 1
= (−1) log dx
0 x
= (−1)n Γ(n + 1)
Z 1
(loge x)n dx = (−1)n n!.
0

Z ∞
10. Evaluate e−t tx−1 (log t)2 dt.
0
Solution: We have
Z ∞
Γ(x) = e−t tx−1 dt
0
Differentiating the above equation with respect to x
Z ∞
d d
Γ(x) = e−t tx−1 dt
dx dx 0
Z ∞
0 d −t x−1 
i.e., Γ (x) = e t dt
0 dx
Z ∞
0
Γ (x) = e−t tx−1 log tdt
0
Differentiating the above equation with respect to x
Z ∞
00 d
Γ (x) = e−t tx−1 log tdt
dx 0
Z ∞
00
∴ Γ (x) = e−t tx−1 (log t)2 dt.
0

11.1.3 Beta Function


In this section we shall introduce a useful function of two variable known as the Beta func-
tion.
First we shall consider the integral

Z 1− Z 1/2 Z 1−
x−1 y−1 x−1 y−1
t (1 − t) dt = t (1 − t) dt + tx−1 (1 − t)y−1 dt. (11.21)
0+ 0+ 1/2

Now, the first integral on right hand side of equation (11.21) is proper, if x ≥ 1 (No matter
what the value of y may be).

155
If 0 < x < 1, then

lim tx−1 t1−x (1 − t)y−1 = lim+ (1 − t)y−1 = 1


t→0+ t→0

By limit test, the first integral on right hand side of equation (11.21) is absolutely convergent
and hence is convergent (∵ 0 < 1 − x < 1) for 0 < x < 1.
If x ≤ 0, then

(1 − t)y−1
lim+ t tx−1 (1 − t)y−1 = lim+
t→0 t→0
( t−x
1, if x = 0
=
0, if x < 0.

By limit test (for divergence), the first integral on right hand side of equation (11.21) is
divergent for x ≤ 0.
The second integral on right hand side of equation (11.21) is proper, if y ≥ 1 (for all x).
If 0 < y < 1, then

lim (1 − t)1−y tx−1 (1 − t)y−1 = lim− (1 − t)x−1 = 1


t→1− t→1

By limit test, the second integral on right hand side of equation (11.21) is absolutely conver-
gent and hence is convergent for 0 < y < 1. If y ≥ 1 (y − 1 ≥ 0) proper finally.

t1−x
lim+ (1 − t) tx−1 (1 − t)y−1 = lim+
t→0 t→0 (1 − t)−y
(
1, if y = 0
=
∞, if y < 0.

By limit test (for divergent), the second integral on right hand side of equation (11.21) is
divergent.
Thus the integral
Z 1−
tx−1 (1 − t)y−1 dt
0+

is convergent for all x > 0 and y > 0.


Therefore we define the Beta function as follows

Definition 9. Z 1−
β(x, y) = tx−1 (1 − t)y−1 dt, x > 0, y > 0.
0+

Theorem 11.26.
β(x, y) = β(y, x), x > 0, y > 0.

156
Proof. We have
Z 1
β(x, y) = tx−1 (1 − t)y−1 dt
0
Changing t by t − 1, we get
Z 1
= (1 − t)x−1 ty−1 dt
0
β(x, y) = β(y, x).

Theorem 11.27.
Z π/2
β(x, y) = 2 (sin t)2x−1 (cos t)2y−1 dt x > 0, y > 0.
0

Proof. We have
Z 1
β(x, y) = tx−1 (1 − t)y−1 dt
0
Putting t = (sin u)2 , then dt = 2 sin u cos u du
When t → 0, then u → 0 and when t → 1, then u → π/2
Z π/2
x−1 y−1
β(x, y) = sin2 u 1 − sin2 u 2 sin u cos u du
0
Z π/2
x−1 y−1
= 2 sin2 u cos2 u sin u cos u du
0
Z π/2
= 2 (sin u)2x−2 (cos u)2y−2 sin u cos u du
0
Z π/2
= 2 (sin u)2x−1 (cos u)2y−1 du
0
Z π/2
∴ β(x, y) = 2 (sin t)2x−1 (cos t)2y−1 dt.
0

Theorem 11.28. ∞
tx−1
Z
β(x, y) = dt x > 0, y > 0.
0 (1 + t)x+y

157
Proof. We have
Z 1
β(x, y) = tx−1 (1 − t)y−1 dt
0
u 1
Putting t = , then dt = du
1+u (1 + u)2
When t → 0, then u → 0 and when t → 1, then u → ∞
Z ∞ x−1  y−1
u 1 1
β(x, y) = du
0 1+u 1+u (1 + u)2
Z ∞
ux−1
= du
0 (1 + u)x+y
Z ∞
tx−1
∴ β(x, y) = dt.
0 (1 + t)x+y

Theorem 11.29.
Γ(x) Γ(y)
β(x, y) = x > 0, y > 0.
Γ(x + y)
Proof. By Gamma function definition, we have
Z ∞
Γ(x) = e−t tx−1 dt, x > 0.
0
changing t to t2 , we obtain
Z ∞
2 x−1
Γ(x) = e−t t2 2 t dt
0
Z ∞
2
= 2 e−t t2x−2 t dt
Z0 ∞
2
Γ(x) = 2 e−t t2x−1 dt
Z0 ∞ Z ∞
2 2
So, Γ(x) Γ(y) = 4 e−t −t1 t2x−1 t2y−1
1 dt dt1 .
0 0

Y

(0, 2R)

(0, R) (R, R)
S

D1 D2


(R, 0) ( 2R, 0) X

2 −t2
Now consider the double integral of e−t 1 t2x−1 t2y−1
1 circular sector D1 and D2 the

158
square S as shown in the adjoining figure. Since the integrand is positive, we have
Z Z Z Z Z Z
2 2 2 2 2 −t2
4 e−t −t1 t2x−1 t2y−1
1 dt dt1 ≤ 4 e−t −t1 t2x−1 t2y−1
1 dt dt1 ≤ 4 e−t 1 t2x−1 t2y−1
1 dt dt1
D1 S D2

Now evaluating these integral by iteration, the center one in rectangular co-ordinates and the
other two in polar co-ordinates, we obtain
Z R Z π/2
2
4 e−r (cos θ)2x−1 (sin θ)2y−1 r2 x+2 y−2 r dr dθ
0 0
Z R Z R
2 −t2
≤ 4 e−t 1 t2x−1 t2y−1
1 dt dt1
0 0

Z 2 R Z π/2
2
≤ 4 e−r (cos θ)2x−1 (sin θ)2y−1 r2 x+2 y−2 r dr dθ
0 0


Z R Z R Z R Z 2R
β(x, y) −r2 2 x+2 y−1 −t2 −t21 β(x, y) −r2 2 x+2 y−1
4 e r dr ≤ 4 e t2x−1 t2y−1
1 dt dt1 ≤ 4 e r dr.
0 2 0 0 0 2

Letting R → ∞, we obtain

β(x, y) Γ(x + y) β(x, y) Γ(x + y)


4 ≤ Γ(x) Γ(y) ≤ 4
2 2 2 2
Γ(x) Γ(y)
i.e., β(x, y) =
Γ(x + y)

Corollary 11.5.
2 2 4 4 2k 2k π
· · · ··· ··· ··· = . (11.22)
1 3 3 5 2k − 1 2k + 1 2
Proof. Let Pn denotes the nth partial product of the infinite product on left side of equation
(11.22).
By Theorem (Gamma theorem 2 and 4) and , we have
Z π/2
Γ(s) Γ(t)
(sin x)2 s−1 (cos x)2 t−1 dx = (11.23)
0 2 Γ(s + t)
1 1
Putting s = n + and t = in (11.23), we get
2 2
Γ n + 12 Γ 21
Z π/2  
2n
(sin x) dx =
0 2 Γ(n + 1)
√
Γ n + 12
Z π/2
2n π
(sin x) dx = , n = 0, 1, 2, . . . , . (11.24)
0 2 n!
1
Putting s = n + 1 and t = in (11.23), we get
2
Γ (n + 1) Γ 12
Z π/2 
2 n+1
(sin x) dx =
2 Γ n + 1 + 12

0

159
Z π/2 √
2 n+1 n! π
(sin x) dx =  , n = 0, 1, 2, . . . , . (11.25)
0 2 Γ n + 32
Dividing the above equations (11.24) and (11.25), we obtain
Z π/2 √
Γ n + 12 π
(sin x)2 n dx
0
= 2√n!
Z π/2
2 n+1 n! π
(sin x) dx
2 Γ n + 32

0
Γ n + 12 Γ n + 32
 
=
n! n!
2n + 1 2n − 1 2n − 1 3 3 1 π
= · · ··· · · ·
2n 2n 2n − 2 4 2 2 2
Z π/2
(sin x)2 n dx
0 1 π
π/2
= · (11.26)
P2 n 2
Z
2 n+1
(sin x) dx
0

Now by (11.25), we have


Z π/2 Z π/2
2 n+1 2n + 1
(sin x) dx = (sin x)2 n−1 dx (11.27)
0 2n 0

h πi
Since 0 ≤ sin x ≤ 1 in the interval 0, , we have
2
Z π/2 Z π/2 Z π/2
2 n+1 2n
0< (sin x) dx < (sin x) dx < (sin x)2 n−1 dx
0 0 0

Dividing the above equation by the first of the integrals and using (11.27), we obtain
Z π/2
(sin x)2 n dx
0 2n + 1 1
0 < Z π/2
< = 1+ (11.28)
2 n+1 2n 2n
(sin x) dx
0
Letting n → ∞ in (11.28), we obtain
Z π/2
(sin x)2 n dx
0
lim Z π/2
= 1. (11.29)
n→∞
2 n+1
(sin x) dx
0

Letting n → ∞ in (11.26) and using (11.29), we get

π
lim P2 n = .
n→∞ 2

And also
2n + 2 π
lim P2 n+1 = lim P2 n = .
n→∞ n→∞ 2 n + 1 2
π
Hence lim Pn = .
n→∞ 2

160
This completes the proof.

Corollary 11.6.
(n!)2 22 n √
lim √ = π.
n→∞ (2 n)! n

Proof. We have, by the proof of Corollary (11.5)

2 2 4 4 2n 2n
P2 n = · · · ··· ·
1 3 3 5 2n − 1 2n + 1
2 n! 2n n!
n
2 · 2 · 4 · 4 · · · (2 n) · (2 n)
= ×
1 · 3 · 3 · 5 · 5 · · · (2 n − 1) · (2 n + 1) 2 · 2 · 4 · 4 · · · (2 n) · (2 n)
24 n (n!)4
=
(1 · 2 · 3 · 4 · · · (2 n − 1) · (2 n)) × (1 · 2 · 3 · 4 · · · (2 n − 1) · (2 n)) × (2 n + 1)
24 n (n!)4
=
(2 n)! (2 n)! (2 n + 1)
24 n (n!)4
P2 n =
[(2 n)!]2 (2 n + 1)
Taking square root on both side, we get
p 22 n (n!)2
P2 n = √ q (11.30)
1
[(2 n)!] n 2 + n

We know that
π
lim P2 n = ,
n→∞ 2
then
r
p π
lim P2 n =
n→∞ 2
Taking the limit as n → ∞ in the equation (11.30), we get

(n!)2 22 n √
lim √ = π.
n→∞ (2 n)! n

Problems:

   
n+1 m+1
Z π/2 Γ Γ
n m 2 2
1. Show that sin t cos t dt =  
0 n+m+2

2
Solution: We have
Z π/2
1
sin2 x−1 t cos2 y−1 t dt =
βx, y
0 2
Z π/2
Γ(x) Γ(y)
sin2 x−1 t cos2 y−1 t dt = (11.31)
0 2 Γ(x + y)

161
Put 2 x − 1 = n and 2 y − 1 = m in the above equation (11.31), we obtain
   
n+1 m+1
Z π/2 Γ Γ
n m 2 2
sin t cos t dt =  
0 n+1 m+1
2Γ +
2 2

   
n+1 m+1
Z π/2 Γ Γ
n m 2 2
sin t cos t dt =  
0 n+m+2

2

 
m+1
Z 1 √ Γ
tm π n
2. Show that √ dt =  
0 1 − tn n m+1 1
Γ +
n 2
Z 1 m
t
Solution:Consider √ dt
0 1 − tn
1−n
Put tn = x, or t = x1/2 then dt = n1 x n . When t = 0, then x = 0 and t → 1,


then x → 1
1 1
tm xm/n 1 1−n
Z Z
√ dt = √ x n dx
0 1 − tn 0 1−x n
1 1 m+1 −1
Z
1
= x n (1 − x) 2 −1 dx
n 0
 
1 m+1 1
= β ,
n n 2
   
m+1 1
Γ Γ
1 n 2
=  
n m + 1 1
Γ +
n 2
 
m+1
1 √ Γ
tm
Z
π n
√ dt =  .
0 1 − tn n m+1 1
Γ +
n 2

1 x−1
+ ty−1
Z
t
3. Show that β(x, y) = x+y
dt
0 (1 + t)
Solution: We have by Theorem (11.28)

tx−1
Z
β(x, y) = dt x > 0, y > 0.
0 (1 + t)x+y
1 Z ∞
tx−1 tx−1
Z
β(x, y) = dt + dt. (11.32)
0 (1 + t)x+y 1 (1 + t)x+y

tx−1
Z
Consider dt
1 (1 + t)x+y

162
 
1 1 1
Put t = or u = , then dt = − 2 du. When t → ∞, then u → 0 and t → 1,
u u u
then u → 1. Above equation becomes

∞ 1
tx−1
Z Z  
1 1 1
dt = 1 x+y
− 2 du
(1 + t)x+y ux−1 1 + u

1 0 u
Z 1 x+y
u
= du
0 ux−1 u2(1 + u)x+y
1
uy−1
Z
= x+y
du
0 (1 + u)
1 Z ∞
tx−1 ty−1
Z
dt = dt
1 (1 + t)x+y 0 (1 + t)x+y

Substituting the above equation in the equation (11.32), we obtain


1 1
tx−1 ty−1
Z Z
β(x, y) = x+y
dt + dt
0 (1 + t) 0 (1 + t)x+y
Z 1 x−1
t + ty−1
β(x, y) = x+y
dt.
0 (1 + t)

Z π/2
π n π 
n
4. Show that (tan t) dt = sec .
0 2 2
Solution: Consider
Z π/2 Z π/2
n
(tan t) dt = sinn t cos−n t dt
0 0
 
1 n + 1 −n + 1
= β ,
2 2 2
n+1 −n+1
 
1Γ 2 Γ 2
=
2 Γ n+1−n+1

n+1
 2
Γ 1 − n+1

1 Γ 2 2
=
2 Γ (1)
   
1 n+1 n+1
= Γ Γ 1−
2 2 2
 
1 π π
= ∵ Γ(x) Γ(1 − x) =
2 sin n+1

2
π sin(π x)
1 π
=
2 cos n2π

Z π/2
π n π 
(tan t)n dt = sec .
0 2 2


5. Show that π Γ(2 x) = 22 x−1 Γ(x) Γ x + 12 , x > 0.


[This is also called as Legendre’s Duplication Formula]

163
Solution: We have
Z π/2
Γ(x) Γ(y)
(sin t)2 x−1 (cos t)2 y−1 dt = (11.33)
0 2 Γ (x + y)

Putting y = 1/2 in the above equation (11.33), we get


Z π/2 √
2 x−1 π Γ(x)
(sin t) dt = (11.34)
2 Γ x + 12

0

Putting y = x in the equation (11.33), we get

π/2
Γ2 (x)
Z
2 x−1
(sin t cos t) dt =
0 2 Γ (2 x)
Z π/2  2 x−1
sin 2 t Γ2 (x)
dt =
0 2 2 Γ (2 x)
Z π/2
22 x−1 Γ2 (x)
(sin 2 t)2 x−1 dt =
0 2 Γ (2 x)
Z π/2 2 x−1 2
2 Γ (x)
i.e., (sin t)2 x−1 dt = (11.35)
0 2 Γ (2 x)

Employing equations (11.34) and (11.35), we get



π Γ(x) 22 x−1 Γ2 (x)
=
2 Γ x + 21

2 Γ (2 x)

 
2 x−1 1
∴ π Γ(2 x) = 2 Γ(x) Γ x + .
2

Z π/2
6. Compute sin3 t cos2 t dt
0
Solution: Consider
Z π/2 Z π/2
3
3
sin t cos t dt = 2
sin2 2−1 t cos2 2 −1 t dt
0 0
3
Put x = 2 and y = , we get
2
Z π/2
= sin2 x−1 t cos2 y−1 t dt
0
1
= β(x, y)
2  
1 3
= β 2,
2 2
 
3
Γ(2) Γ
1 2
=  
2 3
Γ 2+
2

164
 
1 1
1! · · Γ
1 2 2
=  
2 5 3 1 1
· · ·Γ
2 2 2 2
Z π/2
2
sin3 t cos2 t dt = .
0 15


t3
Z
7. Compute 7
0 (1 + t) dt
Solution: Consider
∞ Z ∞
t3 t4 − 1
Z
7
= 4
0 (1 + t) dt 0 (1 + t) + 3 dt
= β(4, 3)
Z ∞
tx−1
 
Γ(4) Γ(3) Γ(m) Γ(n)
= ∵ β(m, n) = dt =
Γ(7) 0 (1 − t)x+y Γ(m + n)
3! 2!
=
Z ∞ 6!
t3 1
∴ 7
=
0 (1 + t) dt 60

 
1
8. Compute Γ .
2
Solution: We know that
Z π/2
Γ(m) Γ(n)
= β(m, n) = 2 sin2 x−1 t cos2 y−1 t dt
Γ(m + n) 0
1
Put x = y = , we get
2

  2
1
Γ Z π/2
2
= 2 dt
Γ(1) 0
π
= 2
2
  2
1
Γ = π
2

 
1
Γ = π.
2

Problems:

Z ∞
2
1. Evaluate e−t cos(x t) dt.
0
Solution: Clearly the integral converges absolutely for all x.

165
So put
Z ∞
2
f (x) = e−t cos(x t) dt,
0
Z ∞
0 2
then f (x) = − e−t t sin(x t) dt,
0
Integration by partsyields
(" 2
!#∞ 2
)

e−t e−t
Z
f 0 (x) = − sin(x t) − + x cos(x t) dt
2 0 2
0
Z ∞
x 2
e−t cos(x t) dt
= −
2 0
x
∴ f 0 (x) = − f (x)
2

Solving this differential equation, we get

x2
log(f (x)) = − + log c
4
−x2 /4
f (x) = c e
√ √
π π
Since f (0) = , we have c =
2 2

π −x2 /4
∴ f (x) = e .
2
Z ∞
2 −x2 t2
2. Evaluate e−t dt, x ∈ (−∞, ∞).
0
Solution: Let
Z ∞
2 −x2 t2
f (x) = e−t dt
0
Differentiating the above equation with respect to x, we get
Z ∞
0 2 2 2
f (x) = −2 e−t −x t x t2 dt
0
x x
Put t = , then dt = −
du
u u2
when t → ∞ then u = 0 and when t = 0 thenu →∞
Z 0
u2

0 −x2 u−2 −u2 −x
∴ f (x) = −2 e x du
∞ x2 u2
Z ∞
2 2 −2
= −2 e−u −x u du
0
i.e., f 0 (x) = −2 f (x), 0 < x < ∞
Solving the differential equation, we obtain
f (x) = c e−2 x , 0 < x < ∞.

+ π
Letting x → 0 , we get c = and f (−x) = f (x), we have
Z ∞ 2 √
2 2 2 π −2 |x|
e−t −x t dt = e , −∞ < x < ∞.
0 2

166
Z π/2
3. Evaluate log(sin x) dx
0
Solution: First we shall show that the improper integral converges, so we consider
Z π/2 Z π/2 
sin x
log(sin x) dx = log x dx
0 0 x
Z π/2   Z π/2
sin x
= log dx + log(x) dx
0 x 0
Z π/2   Z 1 Z π/2
sin x
= log dx + log(x) dx + log(x) dx
0 x 0 1

sin x
Since lim = 1, and log 1 = 0 the first and second integral are proper. Since
x→0 x
Z 1

lim+ x log(x) = 0. The improper integral log(x) dx converges by limit test.
x→0 0
Therefore the improper integral converges.
Z π/2
Let I = log(sin x) dx
0
Z π/2  hπ i
= log sin −x dx
0 2
Z π/2
I = log(cos x) dx
0
Z π/2 Z π/2
∴ 2I = log(sin x) dx + log(cos x) dx
0 0
Z π/2
= log(sin x cos x) dx
0
Z π/2  
sin 2x
= log dx
0 2
Z π/2
= [log (sin 2x) − log 2] dx
0
Z π/2 Z π/2
= log (sin 2x) dx − log 2 dx
0 0
Z π
1 π/2
= log (sin x) dx − log 2 [∵ changing 2x by x]
2 0 2
"Z #
π/2 Z π
1 π
= log (sin x) dx + log (sin x) dx − log 2
2 0 π/2 2
"Z #
π/2 Z π/2
1 π
= log (sin x) dx + log (sin x) dx − log 2
2 0 0 2
Z π/2
1 π
= 2 log (sin x) dx − log 2
2 0 2
π
2I = I − log 2
2

167
π
∴ I = − log 2
2

Z 1
4. Evaluate log (Γ(x)) dx.
0
Solution:
Z 1 Z 1/2 Z 1
log (Γ(x)) dx = log (Γ(x)) dx + log (Γ(x)) dx
0 0 1/2
Z 1/2 Z 0
= log (Γ(x)) dx + log (Γ(1 − x)) (−dx) [changing x by 1 − x in
0 1/2
Z 1/2 Z 1/2
= log (Γ(x)) dx + log (Γ(1 − x)) dx
0 0
Z 1/2
= log (Γ(x) Γ(1 − x)) dx
0
Z 1/2    
π π
= log dx ∵ Γ(x) Γ(1 − x) =
0 sin(π x) sin(π x)
Z 1/2 Z 1/2
= log(π) dx − log (sin(π x)) dx
0 0
Put π x = y, then π dx = dy
when x = 0 then y = 0 and when x = 1/2 then y = 1/2
Z 1 Z 1/2
1 dy
i.e., log (Γ(x)) dx = log(π) − log (sin(y))
0 2 0 π
1 1  π 
= log(π) − − log(2)
2 π 2
1
= [log(π) + log(2)]
2
1
= log(2 π)
2
Z 1 √ 
∴ log (Γ(x)) dx = log 2π .
0

Z ∞
2 x2
5. Evaluate e−a cos(2 b x) dx
0
Solution: We know that
Z ∞ √
−t2 π − x2
e cos(x t) dt = e 4 (11.36)
0 2
2b
Changing x by in (11.36), we get
Z ∞  a √
2 2 2b π − b22
e−a x cos t dt = e a (11.37)
0 a 2
Changing t by a x in (11.37), we obtain
Z ∞ √
2 2 π − b22
e−a x cos(2 b x) a dx = e a
0 2
Z ∞ √
2 2 π − b22
e−a x cos(2 b x) dx = e a .
0 2a

168
Z ∞
cos(α x)
6. Evaluate dx.
0 a2 + x 2
Solution: Now consider
Z ∞ Z ∞  
−(a2 +x2 )z 2 dt −t 2 2
 2 dt
e 2 z dz = e ∵ Put a + x z = t ⇒ 2 z dz = 2
0 0 a2 + x 2 a + x2
Z R
1
= 2 lim e−t dt
a + x2 R→∞ 0
 −t R
1 e
= 2 2
lim
a +x R→∞ −1 0

Z ∞
1
e−(a +x )z 2 z dz = 2
2 2 2

0 a + x2
Z ∞ Z ∞ Z ∞ 
cos(α x) −(a2 +x2 )z 2
∴ dx = cos(α x) e 2 z dz dx
0 a2 + x 2 0 0
Z ∞Z ∞
2 2 2 2
= 2 z cos(α x) e−a z e−x z dx dz
Z ∞ Z0 ∞ 0 Z ∞ 
cos(α x) −a2 z 2 −x2 z 2
dx = 2ze cos(α x) e dx dz.
(11.38)
0 a2 + x 2 0 0

We know that Z ∞ √
−t2 π − x2
e cos(x t) dt = e 4 (11.39)
0 2
α
Changing x by in the above equation (11.39), we get
z
Z ∞ α  √
−t2 π − α22
e cos t dt = e 4z
0 z 2
changing t by x z in the above equation, we obtain
Z ∞ √
−x2 z 2 π − α22
e cos (α x) z dx = e 4z
0 2
Z ∞ √
−x2 z 2 π − α22
∴ e cos (α x) dx = e 4z (11.40)
0 2z

Substituting (11.40) in (11.38), we get


Z ∞ Z ∞ √ 
cos(α x) −a2 z 2 π − α22
dx = 2ze e 4z dz.
a2 + x 2 2z
Z0 ∞ 0
Z ∞
cos(α x) √ 2
−a2 z 2 −− α 2
dx = π e 4 z dz. (11.41)
0 a2 + x 2 0

We know that Z ∞ √
−t2 −x2 t2 π −2 x
e dt = e . (11.42)
0 2

169
αa
Changing x by in the above equation (11.42), we get
2
Z ∞ √
−t2 − α2a π −α a
e dt = e
0 2
changing t by a z in the above equation, we obtain
Z ∞ √
−a2 z 2 − α 2
2
π −α a
e 4 z a dz = e
0 2
Z ∞ √
−a2 z 2 − α 2
2
π −α a
e 4 z dz = e (11.43)
0 2a

Substituting (11.43) in (11.41), we get


∞ √

Z
cos(α x) π −α a
dx = π e
a2 + x 2 2a
Z0 ∞
cos(α x) π −α a
∴ dx = e .
0 a2 + x 2 2a

11.1.4 Keywords
Gamma Function, Beta Function and Legendre’s Duplication Formula.

11.1.5 Terminal Problems



tan−1 (a x)
Z
1. Evaluate dx.
0 x (1 + x2 )
Z ∞
sin(α x)
2. Evaluate dx.
0 x (a2 + x2 )
Z ∞
3. Evaluate e−a x sin(b x) dx.
0
Z ∞
sin(x)
4. Evaluate dx.
0 x
Z ∞
sin(x) cos(x)
5. Evaluate dx.
0 x
Z ∞
sin(2 x)
6. Evaluate dx.
0 x2

11.1.6 Books for reference


1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd


Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

170
4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

171
Block III: Interchange of Differentiation
and Integration

68
172
Block III
Unit 12: Interchange of Differentiation and Integration
12.1 Main Objectives
12.2 Introduction
12.3 Interchange of Differentiation and Integration
12.4 Keywords
12.5 Books for reference

173
12.1.1 Main Objectives
The theme of this course is about various limiting processes. We have learnt the limits of
sequences of numbers and functions, continuity of functions, limits of difference quotients
(derivatives), and even integrals are limits of Riemann sums.
The theme of this course is about various limiting processes. We have learnt the limits
of sequences of numbers and functions, continuity of functions, limits of difference quo-
tients (derivatives), and even integrals are limits of Riemann sums. As often encountered in
applications, exchangeability of limiting processes is an important topic. For example, we
learnt
Z x Z x
d df
f= , f (a) = 0,
dx a a dx
df
whenever is integrable; also
dx

d d
lim fn = lim fn ,
n→∞ dx dx n→∞

if {fn } and {fn0 } converge uniformly.


Here we consider the following situation. Let f (x, y) be a function defined in [a, b] ×
[c, d] and
Z b
φ(y) = f (x, y) dx.
a

12.1.2 Interchange of Differentiation and Integration


It is natural to ask if continuity and differentiability are preserved under integration.

Theorem 12.30. Let f (x, y) be continuous in [a, b] × [c, d]. Then φ defined above is a
continuous function on [c, d].

Proof. Since f is continuous in [a, b] × [c, d], it is bounded and uniformly continuous. In
other words, for any ε > 0, there exist δ such that
Z b
0
φ(y) − φ(y ) ≤ |f (x, y) − f (x, y 0 )| dx
a
< ε (b − a), ∀y, |y − y 0 | < δ.

which shows that φ is uniformly continuous on [c, d].


∂f
Theorem 12.31. Let f and be continuous in [a, b] × [c, d]. Then f is differentiable and
∂y
Z b
d ∂f (x, y)
φ(y) = dx
dx a ∂y

holds.

174
Proof. Fix y ∈ (c, d), y + h ∈ (c, d) for small h ∈ R,

b
φ(y + h) − φ(y)
Z
1
= [f (x, y + h) − f (x, y)] dx
h h a
Z b
∂f (x, y)
= = dx
a ∂y

where z is a point between y and y + h which depends on x. In any case,

b b b
φ(y + h) − φ(y)
Z Z Z
∂f (x, y) ∂f (x, y) ∂f (x, y)
− dx ≤ − dx.
h a ∂y a ∂z a ∂y

df
Since is uniformly continuous on [a, b] × [c, d], for ε > 0, there exits δ such that
dx

∂f (x, y 0 ) ∂f (x, y)
− < ε, ∀ |y 0 − y| < δ and ∀ x.
∂y ∂y

Taking h ≤ δ, we get

b
φ(y + h) − φ(y)
Z
∂f (x, y)
− dx < ε,
h a ∂y

whence the condition follows.


When y = c or d, the same proof works with some trivial changes.

In many applications, the rectangle is replaced by an unbounded region. When this


happens, we need to consider improper integrals. As a typical case, lets assume f is defined
in [a, ∞) × [c, d] and set Z ∞
φ(y) = f (x, y) dx.
a
Z ∞
The function f(y) makes sense if the improper integral f (x) dx is well-defined for each
a
y. Recall that this means Z b
lim f (x, y) dx
b→∞ a

exists. We introduce the following definition: The improper integral


Z ∞
f (x, y) dx
a

is uniformly convergent if ∀ε, there exist b0 > 0 such that


Z b0
f (x, y) dx < ε, ∀b0 , b ≥ b0 .
a
Z ∞
Notice that in particular, this implies that f (x, y) dx exists for every y.
a

Uniform convergence of an improper integral may be studied parallel to the uniform

175
convergence of infinite series. In fact, if we let
Z n
φn (y) = f (x, y) dx,
a

it is not hard to see that the improper integral converges uniformly if and only if the infinite
X∞
series φn (y) converges uniformly when f (x, y) ≥ 0. When f changes sign, the equiv-
n=n0
alence does not always hold. Nevertheless, techniques in establishing uniform convergence
can be borrowed and applied to the present situation. As a sample, we have the following
version of M −test, whose proof is omitted.

Z ∞ 12.32. Suppose that |f (x, y)| ≤ h(x)| and h has an improper integral on [a, ∞).
Theorem
Then f (x, y) dx converges uniformly and absolutely
a

Theorem 12.33. LetZ ∞f be continuous in [a, ∞) × [c, d]. Then f is continuous in [c, d] if the
improper integral f (x, y) dx converges uniformly.
a

Proof. By Theorem (12.30), the function


Z n
φn (y) = f (x, y) dx
a

is continuous on [c, d] for every n. By assumption, ∀ ε > 0, there exits b0 such that
Z m
|φn (y) − φm (y)| = f (x, y) dx < ε, ∀n, m ≥ b0 .
n

Hence {φn } is a Cauchy sequence in sup-norm. Since any Cauchy sequence in sup-norm
converges, φn converges uniformly to some continuous function ψ. As φn converges point
wisely to φ, φ and ψ coincide, so ψ is continuous.
∂f
Theorem 12.34. Let f and be continuous in [a, ∞) × [c, d]. Suppose that the improper
Z ∞ Z ∞ ∂y
∂f
integrals f and are uniformly convergent. Then φ is differentiable, and
a a ∂y
Z ∞
dφ(x) ∂f (x, y)
= dy
dy a ∂y

holds.

Proof. Applying the mean-value theorem to φm − φm ,

φn (y) − φm (y) − [φn (y0 ) − φm (y0 )] = (y − y0 )[φ0n (z) − φ0m (z)]

forZsome z between y and y0 . According to Theorem (12.31) and the uniform convergence

∂f
of ,
a ∂y Z n
0 0 ∂f (x, y)
|φn (z) − φm (z)| = dy → 0
m ∂y

176
as n, m → ∞. This shows that ∀ε > 0, there exists b0 such that

φn (y) − φn (y0 ) φm (y) − φ(y0 )


− < ε, n, m ≥ b0.
y − y0 y − y0

Letting m → ∞,

φn (y) − φn (y0 ) φ(y) − φ(y0 )


− ≤ ε, n ≥ b0.
y − y0 y − y0

By triangle inequality,
Z ∞
φ(y) − φ(y0 ) ∂f (x, y)
− dx
y − y0 a ∂y
Z ∞
φ(y) − φ(y0 ) φn (y) − φn (y0 ) φn (y) − φn (y0 ) ∂f (x, y)
≤ − + − dx
y − y0 y − y0 y − y0 a ∂y
Z n Z ∞
∂f (x, y) ∂f (x, y)
+ dx − dx .
a ∂y a ∂y

Fix a large n = b0 such that Z ∞


∂f (x, y)
dx < ε
n ∂y
and Theorem (12.31), we can also find δ > 0 such that
n
φn (y) − φn (y0 )
Z
∂f (x, y)
− dx < ε, |y − y0 | < δ.
y − y0 a ∂y

Putting things together, we conclude



φ(y) − φ(y0 )
Z
∂f (x, y)
− dx < ε, |y − y0 | ≤ ε + ε + ε < 4ε.
y − y0 a ∂y

One may appreciate these results when considering its relevance in partial differential
equations. Consider the Laplace equation

uxx + uyy = 0

on the disk D = (x, y) : x2 + y 2 < 1. Expressed in polar coordinates, the equation becomes

ur uθθ
urr + + 2 = 0, (r, θ) ∈ [0, 1) × [0, 2π].
r r

To solve this equation means to find a function u = u(r, ∞) which satisfies this equation,
and, moreover, u is periodic in θ for r ∈ [0, 1). This is because when returning to the rect-
angular coordinates, u is continuous in D.
We observe that the Laplace equation is rotationally invariant. More precisely, for any so-
lution u(r, θ), the function v(r, θ) = u(r, θ + θ0 ) is a solution for each θ0 . From linearity it

177
Pn
follows that j=1 cj u(r, θ + θj ) is again a solution. In limit form, the function
Z 2π
ũ(r, θ) = g(α) u(r, θ + α) dα
0

should also be a solution for any continuous g. Indeed, define f (r, θ, α) = g(α) u(r, θ + α).
∂f 2 ∂f ∂f 2
The functions f, 2 , , , are continuous in [0, d] × [0, 2π], d < 1. It follows from
∂θ ∂r ∂r2
Theorem (12.31) that ũ is also harmonic. Noting that g is arbitrary, in this way we have
found many many harmonic functions from a single one. In fact, taking the special harmonic
function to be
1
u(r, θ) = ,
1 − r cos(θ) + r2
one can show that every harmonic function in D which is continuous in (x, y) : x2 + y 2 = 1
arises in this way.
We shall prove a more sophisticated criterion for uniform convergence. Indeed, recall that
the comparison test is only effective in proving absolute convergence of infinite series. We
need Abel’s and Dirichlet’s criteria to handle the convergence of alternating series. Here the
situation is similar. We shall establish a version of Abel’s criterion. The following lemma,
which is usually called the second mean value theorem, is an integral analog of the Abel’s
lemma.

Theorem 12.35. Let f be integrable and g be non-negative, decreasing and continuous on


[a, b]. Then there exists c ∈ [a, b] such that
Z b Z b
f g = g(c) f.
a a

Proof. Divide [a, b] equally by the partition a = x0 < x1 < · · · < xn = b. We have
Z b n Z
X xj Z n Z xj n
X
fg= fg= g(xj−1 ) f+ [g(x) − g(xj )] f (x) dx.
a j=1 xj−1 j=1 xj−1 j=1

As g is continuous and f is bounded on [a, b], the


Z xsecond term on Right hand side of this
equation tends to 0 as n → ∞. Writing F (x) = f , the second term
a

n
X Z xj n
X
g(xj−1 ) f = g(xj−1 ) [F (xj ) − F (xj−1 )]
j=1 xj−1 j=1
n+1
X n
X
= g(xj−2 )F (xj−1 ) − g(xj−1 )F (xj−1 )
j=2 j=1
n
X
= g(xj−2 )F (b) + [g(xj−2 ) − g(xj−1 )F (xj−1 )] − g(a)F (a).
j=1

178
As g is decreasing,
" n
#
X
g(a)m = g(xj−1 ) + [g(xj−2 ) − g(xj−1 )] m
j=1
n
X Z xj
≤ g(xj−1 ) f
j=1 xj−1
" n
#
X
= g(xj−1 ) + [g(xj−2 ) − g(xj−1 )] M
j=1
= g(a)M

for M = supF and m = inf F . By mean-value theorem then there exists ξn ∈ [a, b] such
that Z xin n Z xj
X
g(a) f= g(xj−1 ) f.
a j=1 xj−1

Taking n → ∞, by passing to a convergent subsequence of ξn , we get


Z b Z b
g(c) f= f g.
a a

Now, we can prove the criterion of Abel.


Z ∞
Theorem 12.36. Suppose that f (x, y) dx converges uniformly for y ∈ [c, d], and g(x, y)
a Z ∞
is decreasing for each fixed y and is bounded. Then f (x, y) g(x, y) dx converges uni-
a
formly on [c, d].

Proof. By Theorem (12.35), there exists ξ ∈ [b, b0 ] such that


Z b0 Z b0
f (x, y) g(x, y) dx = g(ξ, y) f (x, y) dx
b b

< (sup|g|) ε,

for large b and b0 , the conclusion follows.


The following application is of technical nature.
Lets evaluate the Dirichlet integral
Z ∞
sin x
I= dx.
0 x

The trick is to consider the integral


Z ∞
sin x
ϕ(y) = e−yx dx, y ≥ 0.
0 x

179
Letting (
1, x=0
f (x, y) =
e−yx sinx x , x 6= 0.
Z ∞
sin x
We showed that dx converges, so it is uniformly convergent in y trivially. By
Z a∞ x
sin x
Abel’s criterion, e−yx dx converges uniformly. The y-derivative of f is given by
a x

∂f
= −e−yx sin x.
∂y
Z ∞
∂f (x, y)
For each y ≥ δ, dx is clearly uniformly convergent. By Theorem (12.31), we
δ ∂y
conclude that

d e−yx sinx x
Z 
ϕ0 (y) = dx
0 dy
Z ∞
= − e−yx sin x dx
0
1
= − ,
1 + y2

which holds for y ≥ δ > 0. By integration we get

ϕ(y) = − tan−1 y + C.

As Z ∞ Z ∞
−yx sin x 1
|ϕ(y)| − e ≤ e−yx = →0
0 x 0 y
as y → ∞, C = tan−1 ∞ = π2 . So

π
I = lim+ ϕ(y) = .
y→0 2

Definition: A real valued function f defined on an interval I is said to be convex, if

f (λ x + (1 − λ) y) ≤ λ f (x) + (1 − λ) f (y),

whenever x, y ∈ I and 0 < λ < 1.


Note:

1. Every convex function defined on an open interval is "Continuous".

2. If f is convex and if a < x1 < x2 < x3 < b, then

f (x2 ) − f (x1 ) f (x3 ) − f (x1 ) f (x3 ) − f (x2 )


≤ ≤ .
x2 − x1 x3 − x1 x3 − x2

180
3. log Γ(x) is convex on (0, ∞).

Proof. We have by Holder’s inequality


Z b Z b 1/p Z b 1/q
p q
f g dα ≤ |f | |g| , (12.44)
a a a

1 1
where f g ∈ R(α) and p > 0, q > 0, + = 1.
p q
Now consider
  Z ∞
x y x y
Γ + = t p + q e−t dt
p q
Z0 ∞   
x−1
−t/p
 y−1
−t/q
 1 1
= t p e t q e dt ∵ + =1
0 p q
Z ∞  x−1
1/p Z ∞  y−1
1/q
−t/p −t/q
≤ t p e t q e
0 0
1/p 1/q
= [Γ(x)] [Γ(y)]
 
x y 1 1
∴ log Γ + ≤ log Γ(x) + log Γ(y)
p q p q
i.e., log Γ (λ x + (1 − λ) y) ≤ λ log Γ(x) + (1 − λ) log Γ(y).

Theorem 12.37. If f is a positive function on on (0, ∞), such that


(a) f (x + 1) = x f (x), (b) f (1) = 1 and (c) log f is convex, then f (x) = Γ(x).

Proof. Since Γ satisfies (a), (b), and (c), it is enough to prove that f (x) is uniquely deter-
mined by (a), (b), and (c), for all x > 0. By (a) it is enough to do this for x ∈ (0, 1).
Put φ = log f, then

φ(x + 1) = log(f (x + 1))


= log(x f (x))
= log(x) + log(f (x))
φ(x + 1) = log(x) + φ(x). (0 < x < ∞)

and φ(1) = 0, φ is convex. Suppose 0 < x < 1 and n is a positive by using equation (12.45).

φ(n + 1) = log(n!). (12.45)

Since n < n + 1 < n + 1 + x < n + 2 and φ is convex, we have

φ(n + 1) − φ(n) φ(n + 1 + x) − φ(n + 1) φ(n + 2) − φ(n + 1)


≤ ≤ (12.46)
1 x 1

181
Using (12.45) and (12.46), we obtain

φ(n + 1 + x) − φ(n + 1)
log(n) ≤ ≤ log(n + 1). (12.47)
x

Repeated application of (12.45) gives

φ(n + 1 + x) = φ(x) + log[x · (x + 1) · · · (x + n)]

Thus we have by (12.47) and (12.45)

φ(x) + log[x · (x + 1) · · · (x + n)] − log(n!)


log(n) ≤ ≤ log(n + 1)
x x
nx n!
 
n+1
0 ≤ φ(x) − log ≤
x · (x + 1) · · · (x + n) n
 x
n+1
Since →0 or n → ∞, we obtain
n
nx n!
 
lim log = φ(x) = log(f (x))
n→∞ x · (x + 1) · · · (x + n)
nx n!
 
i.e., lim = f (x). (12.48)
n→∞ x · (x + 1) · · · (x + n)

Hence f (x) is uniquely determined.

Corollary 12.7.
nx n!
 
lim = Γ(x).
n→∞ x · (x + 1) · · · (x + n)
Proof. Follows from equation (12.48) for 0 < x < 1. Since

Γ(x + 1) = x Γ(x)

is holds for all x > 0.

12.1.3 Keywords
Continuous Function, Integral Sign, Finite Interval, Integral Theorem, Dirichlet’s criterion
and Abel’s criterion.

12.1.4 Books for reference


1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd


Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

182
4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

8. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

183
Block IV: Functions of several variables

Unit 13: Functions of several variables


13.1.1 Main Objectives
13.1.2 Introduction
13.1.3 Keywords
13.1.4 Terminal Problems
13.1.5 Books for Reference

80
184
Block IV
Unit 4: Functions of several variables

13.1.1 Main Objectives


The objective of the unit is to study and understand the following topics in detail.
• Recognize a function of two variables and identify its domain and range.

• Sketch a graph of a function of two variables.

• Sketch several traces or level curves of a function of two variables.

• Recognize a function of three or more variables and identify its level surfaces.

13.1.2 Introduction
We know by definition

Rn = {(x1 , x2 , ..., xn )|xi ∈ R, i = 1, 2, 3...n}

Thus an element of Rn is an n-tuple, often refereed to as a Vector or a point of Rn .

Given x = (x1 , x2 , ..., xn ), y = (y1 , y2 , ..., yn ) α ∈ R, we define


(i)x = y, if xi = yi , for i = 1, 2, 3.., n [Equality]
(ii)x + y=(x1 + y1 , x2 + y2 , ...xn + yn ) [Addition]
(iii) α x=(αx1 + αx2 + ... + αxn ) [Scalar multiplication]
we know that
1. With to respect to addition addition and scalar multiplication defined above Rn becomes
a vector space over the field R.
2. The set of vectors {e1 , e2 , .., en },
where e1 = (1, 0, 0, ..., 0), e1 = (0, 1, 0, ..., 0), . . . , en = (0, 0, 0, ..., 1) is a basis of Rn .
3. Rn is a normed linear space , where a norm is defined by
v
u n
uX
kxk = t x2i , x ∈ Rn .
i=1

4. Rn is also an inner product space, where the inner product is defined by


n
X
< x, y >= xi yi , x, y ∈ Rn .
i=1

5. Rn is a metric space , the metric (Euclidian) d is defined by

d(x, y) = kx − yk, , x, y ∈ Rn .

185
the square metric defined by ρ is defined by

ρ(x, y) = max{|xi − yi |; 1 ≤ i ≤ n}, x, y ∈ Rn .

Topology on Rn :

Definition 10. Let a = (a1 , a2 , ..., an ) ∈ Rn , by an ε-neighborhood of a we mean set Nε (a)


defined by

Nε (a) = {x ∈ Rn ; kx − ak < ε}
 v 
 u n 
u X
Nε (a) = x ∈ Rn ; t (x2i − a2i ) < ε
 
i=1

or Nε (a) = {x ∈ Rn ; ρ(x, a) < ε}


Nε (a) = {x ∈ Rn ; max{|x1 − a1 |, |x2 − a2 |, ..., |xn − an |} < ε} .

Definition 11. Let E ⊂ Rn . A point a ∈ Rn is called a limit point of E, if every neighbor-


hood of a contains a point of E other than a.

Definition 12. Let E ⊂ Rn . A point a ∈ Rn is called an interior point of E, if there is a


neighborhood Nε (a) of a, such that Nε (a) ⊂ E.

Definition 13. Let E ⊂ Rn . E is called closed, if it contains all of its limit points.

Definition 14. Let E ⊂ Rn . E is called open, if every point of E is an interior point of E.

Examples:

1. Consider R2 and (a, b) ∈ R2 then a neighborhood of a, b is given by

Nε (a, b) = {(x, y) ∈ R2 |(x − a)2 + (y − b)2 < ε2 }

i.e., The ε- neighborhood of (a, b) is the set of all points inside a circle centered at
(a, b) and radius ε.

(x, y) ∈ R2 /max|x − a|, |y − a| < ε



or Nε (a, b) =
(x, y) ∈ R2 /x ∈ (a − ε, a + ε), y ∈ (b − ε, b + ε)

=
Nε (a, b) = (a − ε, a + ε) × (b − ε, b + ε).

i.e., The ε- neighborhood of (a, b) is the set of all points inside rectangle with sides
parallel to the co-ordinate axes.

2. The set E = {(x, y)/x2 + y 2 < 1} ⊂ R2 is an open set.

3. The set E = {(x, y)/x2 + y 2 ≤ 1} ⊂ R2 is an closed set.

186
4. The set E = {(x, y)/x > a, a ∈ R} is open in R2 .

5. The set E = {(x, y)/a ≤ x ≤ b, c ≤ y ≤ d} is closed in R2 . In fact E is a rectangular


region in R2 .

6. The set E = {(x, y)/x > 0, y > 0, 3x + 4y < 5} is open in R2 .

Definition 15. Let n ≥ 2 and m > 0 be integers. E ⊂ Rn . A map f : E → Rn is called a


function of several variables.
In fact, f is a function of n variables.

For example

1. f : R2 → R defined by f (x, y) = x + y is a function of two variables.

2. f : R2 → R3 defined by f (x, y) = (x + y, 2x, x − 3y) is a function of two variables.

13.1.3 Limits
Limits: We shall begin with a function of two variables and define the limit of function of
two variables.

Definition 16. Let E be a non empty and subset of R2 , (a, b) be a limit point of E and
f : E → R be a map. Then we say that f (x, y) tends to A as (x, y) to (a, b), if for every
ε > 0 there exist a δ > 0, such that
p
|f (x, y) − A| < ε whenever (x, y) ∈ E and 0 < (x − a)2 + (y − b)2 < δ.

Equivalently, we say that f (x, y) → A as (x, y) → (a, b), if for every ε > 0, there exist a
δ > 0, such that

|f (x, y) − A| < ε whenever (x, y) ∈ E and 0 < |x − a| < δ, 0 < |y − b| < δ.

We denote this by
lim f (x, y) = A
(x,y)→(a,b)

and read as ”limit of f (x, y) as (x, y) tends to (a, b) is A”.

REMARKS:

1. If the limit of a function f (x, y) exists the it is unique.

2. The limiting value of f (x, y) is unchanged as a variable point (x, y) approaches (a, b)
along any curve what so ever. Hence to establish the non-existence of a limit we must
find two roots of approach to limiting point which give different values.

Examples:

187
1. Let f (x, y) = x2 + y 2 . Then clearly f (x, y) defined for all (x, y) and (0, 0) is a limit
point of the domain of f (x, y) moreover

lim f (x, y) = 0.
(x,y)→(a,b)

For ε > 0 be given, then consider

|f (x, y) − 0| = |x2 + y 2 − 0| = |x2 + y 2 |



so, we choose δ = ε, then
p
(x − 0)2 + (y − 0)2 < δ
⇒ x2 + y 2 < δ 2 = ε
p
∴ |f (x, y) − 0| < ε, whenever 0 < (x − 0)2 + (y − 0)2 < δ as desired.

xy 3
2. Show that lim does not exist.
(x,y)→(0,0) x2 + y 6
Solution: First we shoe that two different paths for (x, y) to approaches (0, 0)
first let (x, y) approach (0, 0) along the line y = x then

xy 3 xx3
lim = lim
(x,y)→(0,0) x2 + y 6 x→0 x2 + x6

x2
= lim
x→0 1 + x4
xy 3
lim = 0.
(x,y)→(0,0) x2 + y 6


next let (x, y) approach (0, 0) along the curve y = 3
x then
√ 3
xy 3 x3x
lim = lim √
(x,y)→(0,0) x2 + y 6 y→0 x2 + 3 x6

x2
= lim 2
x→0 x + x2
1
= lim
x→0 2
3
xy 1
lim 2 6
= .
(x,y)→(0,0) x + y 2

xy 3
Thus lim does not exist.
(x,y)→(0,0) x2 + y 6
 
1
3. Show that lim y sin = 0.
(x,y)→(0,0) x
Solution: Let ε > 0 be given, consider
   
1 1
y sin −0 = y sin ≤ |y|
x x

we choose δ = ε, then

188
0 < |x − 0| < δ, 0 < |y − 0| < δ

yield.
 
1
y sin −0 < ε
x
 
1
∴ lim y sin = 0.
(x,y)→(0,0) x

4. Let f : R2 → R defined by,


 2xy

(x, y) 6= (0, 0)
2 2
f (x, y) = x + y
0,

(x, y) = (0, 0)

Show that limit does not exist.


Solution: As usual we shall allow (x, y) to approach (0, 0) along two different paths.
First, let (x, y) approach (0, 0) along the line y = x, then

2xx 2 x2
lim f (x, y) = lim f (x, x) = lim = lim =1
(x,y)→(0,0) x→0 x→0 x2 + x2 x→0 2 x2

Now, let (x, y) approach (0, 0) along the curve y = 2x, then

2x.2x 4x2 4
lim f (x, y) = lim f (x, x) = lim 2 2
= lim 2
=
(x,y)→(0,0) x→0 x→0 x + (2x) x→0 5x 5

Thus limit does not exist.

5. Show that lim (x2 + 2y) = 5.


(x,y)→(1,2)
Solution: Let ε > 0 be given, we have to find a δ, such that

|x2 + 2y − 5| < ε whenever |x − 1| < δ, |y − 2| < δ.

Now if,

|x − 1| < δ, then
1−δ < x < 1+δ
(1 − δ)2 < x2 < (1 + δ)2
i.e., 1 + δ 2 − 2δ < x2 < 1 + δ 2 + 2δ (13.49)

189
Similarly if,

|y − 2| < δ, then
2−δ < y < 2+δ
i.e., 4 − 2δ < 2y < 4 + 2δ (13.50)

Adding equation (13.49) and (13.50), we obtain

5 − 4δ + δ 2 < x2 + 2y < 5 + 4δ + δ 2
i.e., δ 2 − 4δ < x2 + 2y − 5 < δ 2 + 4δ
Now, δ 2 + 4δ < 5δ, if 0 < δ < 1 and − 5δ < δ 2 − 4δ (∵ −δ < δ 2 )
Thus, −5δ < x2 + 2y − 5 < 5δ, if 0 < δ < 1. (13.51)
n εo
So, we choose δ = min 1, , then we Then, we have from (13.51)
5

|x2 + 2y − 5| < ε, whenever |x − 1| < δ and |y − 2| < δ.

∴ lim (x2 + 2y) = 5.


(x,y)→(1,2)

xy


p (x, y) 6= (0, 0)
6. Let f (x, y) = x2 + y 2 , then show that

0, (x, y) = (0, 0)

lim f (x, y) = 0.
(x,y)→(0,0)

Solution: Let ε > 0 be given, we have to show that

xy
|f (x, y) − 0| = p < ε, whenever |x − 1| < δ and |y − 0| < δ.
x2 + y 2
Now ,if

|x − 0| = |x| < δ and |y − 0| = |y| < δ


x2 < δ 2 and y 2 < δ 2
1 1 1 1
∴ 2 > 2 and > 2 (∵ δ 6= 0)
x δ y2 δ
1 1 2
∴ 2+ 2 > 2
x y δ
x + y2
2
2
> 2
x2 y 2 δ
x2 y 2 δ2
<
x2 + y 2 2

190
Taking square root on both sides, we obtain
xy δ
i.e., p < √
2
x +y 2 2

so we choose δ = 2 ε, then we have

|f (x, y) < ε|, whenever |x − 1| < δ and |y − 0| < δ.

∴ lim f (x, y) = 0.
(x,y)→(0,0)

13.1.4 Continuity
Let f be a real valued function defined on an open set E of R2 and (x0 , y0 ) ∈ E. We say
that f is continuous at (x0 , y0 ), if for every ε > 0 there exist a δ > 0 such that

|f (x, y) − f (x0 , y0 )| < ε,


p
whenever (x, y) ∈ E and (x − x0 )2 + (y − y0 )2 < δ

. Equivalently, we say that f is continuous at (x0 , y0 ), if for every ε > 0 there exist a δ > 0,
such that
|f (x, y) − f (x0 , y0 )| < ε,

whenever (x, y) ∈ E and |x − x0 | < δ and |y − y0 | < δ.

If f is continuous at every point of E we say that f is continuous on E.

Example:
x − y
 , if x 6= y
1. Let f : R → R be defined by f (x, y) = x + y
2

1, if x = y

then show that f is not continuous at (0, 0).
Solution: We shall compute the limits of f (x, y) as (x, y) approaches (0,0) along two
different paths
First along the line y = 0

lim f (x, y) = lim f (x, 0)


(x,y)→(0,0) x→0
x
= lim
x→0 x
lim f (x, y) = 1.
(x,y)→(0,0)

191
Along the line x = 0

lim f (x, y) = lim f (0, y)


(x,y)→(0,0) y→0
−y
= lim
y→0 y
= −1.

Thus lim f (x, y)does not exist.


(x,y)→(0,0)

∴ f is not continuous at the origin.


xy


p , if (x, y) 6= (0, 0)
2. Let f : R2 → R be defined by f (x, y) = x2 + y 2

0, if (x, y) = (0, 0)
then show that f is continuous at (0, 0).
Solution: Let  > 0 be given, consider

xy
|f (x, y) − f (0, 0)| = p −0
x2 + y 2
xy
= p
x2 + y 2
r cos(θ) r sin(θ)
= √ [∵ x = r cos(θ) and y = r sin(θ)]
r2
≤ r
p
= x2 + y 2 .

So, we choose δ = , then


p
|f (x, y) − f (0, 0)| < , whenever x2 + y 2 < δ.

∴ f is continuous at the origin.



x + y, if x = 0 or y = 0
3. Let f : R2 → R be defined by f (x, y) =
1, otherwise.
then show that f is not continuous at (0, 0).
Solution: First we note that f (0, 0) = 0.
Next, we shall compute the limits of f (x, y) as (x, y) approaches (0,0) along two
different paths
Let (x, y) approaches to (0, 0) along the line y = x, then

lim f (x, y) = lim f (x, x)


(x,y)→(0,0) x→0(x6=0)

= lim 1
x→0(x6=0)

lim f (x, y) = 1.
(x,y)→(0,0)

192
Let (x, y) approaches to (0, 0) along the line x−axis, then

lim f (x, y) = lim f (x, 0)


(x,y)→(0,0) x→0

= lim x
x→0
= 0.

Thus lim f (x, y)does not exist.


(x,y)→(0,0)

∴ f is not continuous at the origin.


 2 2
x y x − y , if (x, y) 6= (0, 0)

4. Let f : R2 → R be defined by f (x, y) = x2 + y 2
0,

if (x, y) = (0, 0)
then show that f is continuous at (0, 0).
Solution: Let  > 0 be given, consider

x2 − y 2
|f (x, y) − f (0, 0)| = xy −0
x2 + y 2
x2 − y 2
= xy 2
x + y2
x2 − y 2
 
= |x| |y| ∵ 2 ≤ 1, ∀(x, y) 6= (0, 0)
x + y2


So, we choose δ = , then |x| < δ, |y| < δ yield |x| |y| < δ 2 = 

|f (x, y) − f (0, 0)| < , whenever |x| < δ, |y| < δ.

∴ f is continuous at the origin.

193
13.1.5 Partial Derivatives
Let E be a non-empty subset of R2 and(x0 , y0 ) ∈ E. Let f be real valued function defined
on E, if
f (x0 + h, y0 ) − f (x0 , y0 )
lim
h→0 h
exists, then it is called the partial derivative of f with respect to x at (x0 , y0 ).
Similarly, if
f (x0 + k, y0 ) − f (x0 , y0 )
lim
k→0 k
exists, then it is called the partial derivative of f with respect to y at (x0 , y0 ).

Notations:
The partial derivative of f w.r.t x at (x0 , y0 ) is denoted by

δf (x0 , y0 )
fx (x0 , y0 ) or D1 f (x0 , y0 ) or .
δx

Similarly, The partial derivative of f w.r.t y at (x0 , y0 ) is denoted by

δf (x0 , y0 )
fy (x0 , y0 ) or D2 f (x0 , y0 ) or .
δy

Note:
To find the partial derivative of f w.r.t a particular variable, we shall just differentiate the
function with respect to that particular variable by treating the other variables as constants.
The continuity of f at a point need not to imply the existence of partial derivative at that
point.

Examples

1. Let f : R2 → R be defined by

x + y, if x = 0 or y = 0
f (x, y) =
1, otherwise

then f is not continuous at the origin.


Solution: First we shall note that f (0, 0) = 0.
Next we shall compute the limit of f (x, y) as (x, y) → (0, 0) along two paths.
Let (x, y) approach (0, 0) along the line y = x (x 6= 0), then

lim = lim f (x, x) = lim 1 = 1.


(x,y)→(0,0) x→0 x→0

194
Let (x, y) approach (0, 0) along the line x-axis (y = 0), then

lim = lim f (x, 0) = lim = 0.


(x,y)→(0,0) y→0 x→0

Thus the limit does not exist.

2. Let f : R2 → R be defined by
 2 2
xy x y

if (x, y) 6= (0, 0)
f (x, y) = x2 + y 2
0,

if (x, y) = (0, 0)

then f is continuous at (0, 0).


Solution: Let ε > 0 be given,

Consider

x2 y 2
|f (x, y) − f (0, 0)| = −0
xy
x2 + y 2
x2 y 2
= xy 2
x + y2
x2 y 2
 
≤ |x||y| ∵ 2 ≤ 1∀(x, y) 6= (0, 0)
x + y2

So, we choose δ = ε, then |x| < δ, |y| < δ, yields |x y| < δ 2 = ε
∴ |f (x, y) − f (0, 0)| < ε

thus f is continuous at (0, 0).

3. Let f : R2 → R be defined by
    
1 1
x sin + y sin 6 0
, if x y =

f (x, y) = y x

0, if x y = 0

then f is continuous at (0, 0).


Solution: Let ε > 0 be given
consider
   
1 1
|f (x, y) − f (0, 0)| = x sin + y sin − 0 ≤ |x| + |y|
y x

ε
so we choose δ = , then |x| < δ |y| < δ yield. Clearly f is continuous at (0, 0).
2
But,
f (0 + h, 0) − f (0, 0) |h|
lim = lim
h→0 h h→0 h

195
f (0 + k, 0) − f (0, 0) |k|
and lim = lim .
k→0 k k→0 k

Since these limits do not exist, the function has no partial derivatives at (0, 0).

Remark: If a function has partial derivatives at a point then it need not to be continuous
at that point.
For example let f : R2 → R be defined by

x + y if x = 0 or y = 0
f (x, y) =
1, otherwise

then
f (0 + h, 0) − f (0, 0) h
fx (0, 0) = lim = lim = 1
h→0 h h→0 h

f (0 + k, 0) − f (0, 0) k
fy (0, 0) = lim = lim = 1
k→0 k k→0 k

thus f has partial derivatives at (0, 0). But f is not continuous at (0, 0).

Some more examples:

1. Consider the function f defined by,


 xy
 , if (x, y) 6= (0, 0)
f (x, y) = x2
+ y2
0, if (x, y) = (0, 0)

then clearly f is not continuous at (0, 0). But,

f (0 + h, 0) − f (0, 0) 0
fx (0, 0) = lim = lim = 0.
h→0 h h→0 h

f (0 + k, 0) − f (0, 0) 0
fy (0, 0) = lim = lim = 0.
k→0 k k→0 k

Thus f has partial derivatives at (0, 0).

2. Consider the function f defined by,


x − y
 if x 6= −y
f (x, y) = x + y
0, if x = −y

then clearly f is not continuous at (0, 0)


Consider
h
f (0 + h, 0) − f (0, 0) h
−0 1
fx (0, 0) = lim = lim = lim
h→0 h h→0 h h→0 h

196
does not exist.
−k
f (0 + k, 0) − f (0, 0) k
−0 −1
fy (0, 0) = lim = lim = lim
k→0 k k→0 k k→0 k

does not exist. Thus fx (0, 0) and fy (0, 0) do not exist.

3. Consider the function f defined by,


 3 3
x − y

(x, y) 6= (0, 0)
f (x, y) = x2 + y 2
0,

(x, y) = (0, 0)

First, let ε > 0 be given, consider

x3 − y 3
|f (x, y) − f (0, 0)| =
x2 + y 2
r3 cos3 θ − r3 sin3 θ
= , [x = r cos(θ) and y = r sin(θ)]
r2 cos2 θ + r2 sin2 θ
= r cos3 (θ) − sin3 (θ)
≤ 2r
p
= 2 x2 + y 2
ε
choose δ = √ , then |x| < δ, |y| < δ and
2 2
2 2 2
p √ ε
x + y < 2δ ⇒ x2 + y 2 < 2δ = .
2
Hence |f (x, y) − f (0, 0)| < ε, whenever |x − 0| < δ, |y − 0| < δ.

∴ f is continues at (0, 0)

f (0 + h, 0) − f (0, 0)
fx (0, 0) = lim
h→0 h
h3
2
= lim h
h→0 h
fx (0, 0) = 1

f (0 + k, 0) − f (0, 0)
fy (0, 0) = lim
k→0 k
3
−k
2
= lim k
k→0 k
fy (0, 0) = −1

Thus fx (0, 0) and fy (0, 0) exist. Hence f has partial derivatives at (0, 0).

197
4.
x y tan y , if (x, y) 6= (0, 0)
  
If f (x, y) = x
0, if (x, y) = (0, 0)
Show that x fx + y fy = 2 f.
Solution: First Let (x, y) 6= (0, 0), then

f (x + h, y) − f (x, y)
fx (x, y) = lim
h→0 h  
y y
(x + h) y tan − x y tan
x+h x
= lim
h→0
  h  y   
y y
x y tan − tan + h y tan
x+h x x+h
= lim
h→0 h
2
−y  y   y 
fx (x, y) = sec2 + y tan . (13.52)
x x x

Similarly, we have

f (x, y + k) − f (x, y)
fy (x, y) = lim
k→0 k  
y+k y
x (y + k) tan − x y tan
x x
= lim
k→0
  k  y   
y+k y+k
x y tan − tan + k y tan
x x x
= lim
h→0 k
2 y
  y
fy (x, y) = y sec + x tan . (13.53)
x x

Employing equation (13.52) and (13.53), it follows that x fx + y fy = 2 f.


The case (x, y) = (0, 0) follows similarly.

13.1.6 Keywords
Topology, Limits, Continuity and Partial Derivatives.

13.1.7 Terminal Problems


Problems on Limits:
    
xsin 1 + ysin 1 , (x, y) 6= (0, 0)

1. Find the lim f (x, y), where f (x, y) = y x
(x,y)→(0,0) 
0, (x, y) = (0, 0)

2. Evaluate
sin(x2 + y 2 )
(a) lim p
(x,y)→(0,0) x2 + y 2

198
 
1
(b) lim x y sin
(x,y)→(0,0) y
 2 2
 xy p x y

, if (x, y) 6= (0, 0)
3. 3 Let f (x, y) = x2 + y 2

 0, if (x, y) = (0, 0).
then show that lim f (x, y) = 0.
(x,y)→(0,0)

1 − cos(x2 + y 2 )
4. Show that lim does not exist.
(x,y)→(0,0) x2 y 2 (x2 + y 2 )

5. Show that lim (5x + 6y) = 28.


(x,y)→(2,3)

x2 + y 2
6. Show that lim p = 2.
(x,y)→(0,0) x2 + y 2 − 1
Problems on Continuity:
 2 2
 xy

(x, y) 6= (0, 0)
1. Examine the function f (x, y) = x2 + y 2
1,

(x, y) = (0, 0)
for continuity at (0, 0).

 2xy
 if (x, y) 6= (0, 0)
2 2 n
2. Let f (x, y) = (x + y )
1,

if (x, y) = (0, 0)
Show that f is continuous (0,0) if n = 21 , but not continuous at (0, 0) if n = 1.

3. Test the continuity of the following functions at the given point:



x2 y 2
. if (x, y) 6= (0, 0)


(a) f (x, y) = x2 y 2 + (x + y)2 at (0, 0).

1, if (x, y) = (0, 0)

sin(xy)

p , if (x, y) 6= (0, 0)
(b) f (x, y) = x2 + y 2 at (0, 0).

1, if (x, y) = (0, 0)

 |x| e y2 , if y 6= 0
|x|

(c) f (x, y) = y 2 at (0, 0).
0,

if y = 0
    
1 1
x sin + y sin , if xy 6= 0

(d) f (x, y) = y x at (0, 0).

0, if xy = 0

13.1.8 Books for reference


1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

199
2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd
Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

200
Block IV: Functions of several variables

Unit 14:Differentiation
14.1.1 Main Objectives
14.1.2 Introduction
14.1.3 Keywords
14.1.4 Terminal Problems
14.1.5 Books for Reference

201
Block IV
Unit 4: Differentiation.

14.1.1 Main Objectives


δz δz
A function z = f (x, y) has two partial derivatives: δx and δy . These derivatives correspond
to each of the independent variables and can be interpreted as instantaneous rates of change
(that is, as slopes of a tangent line).But in more than one variable, the lack of a unique
independent variable makes this more complicated. In particular, the rates of change may
differ, depending upon the direction in which we move.

14.1.2 Differentiability
Definition 17. Let E be a non-empty open subset of R2 and (x0 , y0 ) ∈ E.
A function f : E → R is said to be differentiable at (x0 , y0 ), if there exist real numbers A
and B such that for sufficiently small h and k, we have

f (x0 + h, y0 + k) = f (x0 , y0 ) + h A + k B + ||(h, k)|| (h, k), (14.54)

where (h, k) → 0 as (h, k) → (0, 0)


The function f is said to be differentiable, if it is differentiable at every point of E.

Note: If we put x0 + h = x and y0 + k = y, then equation (14.54) takes the form


p
f (x, y) = f (x0 , y0 ) + A(x − x0 ) + B(y − y0 ) + (x − x0 )2 + (y − y0 )2 (x − x0 , y − y0 ),

where lim (x − x0 , y − y0 ) = 0


(x,y)→(x0 ,y0 )
the function
f (x, y) = z = z0 + A(x − x0 ) + B(y − y0 ),

where z0 = f (x0 , y0 ) is a linear function of two variables the graph of which is a plane.
[recall linear approximation theorem]

Theorem 14.38. If the function f : E → R is differentiable at (x0 , y0 ) ∈ E, then f is


continuous at (x0 , y0 ) and has partial derivatives at (x0 , y0 ).

Proof. If f is differentiable at (x0 , y0 ) then there exist real numbers A and B such that for
sufficiently small h and k

f (x0 + h, y0 + k) = f (x0 , y0 ) + hA + kB + k|(h, k)k| ∈ (h, k), (14.55)

where (h, k) → 0 as (h, k) → (0, 0)

202
putting k = 0 in the equation (14.55), we get

f (x0 + h, y0 ) = f (x0 , y0 ) + h A + k(h, 0)k(h, 0)


f (x0 + h, y0 ) − f (x0 , y0 ) |h|
∴ = A+ (h, 0)
h h
taking limits as h → 0, we obtain
 
f (x0 + h, y0 ) − f (x0 , y0 ) |h|
lim = lim A + (h, 0)
h→0 h h→0 h
 
|h|
D1 f (x0 , y0 ) = A. ∵ = ±1 textand lim (h, 0) = 0
h h→0

Similarly, putting h = 0 in the equation (14.55), we obtain

f (x0 , y0 + k) = f (x0 , y0 ) + k B + k(0, k)k(0, k)


f (x0 , y0 + h) − f (x0 , y0 ) |k|
∴ = B+ (k, 0)
k k
taking limits as k → 0, we obtain
 
f (x0 , y0 + k) − f (x0 , y0 ) |k|
lim = lim B + (0, k)
k→0 k k→0 k
 
|k|
D2 f (x0 , y0 ) = k. ∵ = ±1 textand lim (k, 0) = 0
k k→0

Hence D1 f (x0 , y0 ) and D2 f (x0 , y0 exist.


Finally taking limits as (h, k) → (0, 0) in equation (14.55), we get

lim f (x0 + h, y0 + k) = f (x0 , y0 )


(h,k)→(0,0)

This proves the continuity of f at (x0 , y0 ).

Remark: The converse of the above Theorem need not to be true.


for example consider the function f defined by

xy


p (x, y) 6= (0, 0)
f (x, y) = x2 + y 2

0, (x, y) = (0, 0)

then clearly f is continuous at (0, 0). Also,

f (0 + h, 0) − f (0, 0) 0
D1 f (0, 0) = lim = lim = 0
h→0 h h→0 h

similarly,

f (0, 0 + k) − f (0, 0) 0
D2 f (0, 0) = lim = lim = 0
k→0 k k→0 k
Therefore f has partial derivatives at (0,0)

203
Next, if f is differentiable at (0, 0) then for sufficiently small h and k, we have

f (x0 + h, y0 + k) = f (x0 , y0 ) + h A + k B + k|(h, k)k| (h, k), (14.56)

where (h, k) → 0 as (h, k) → (0, 0)


since D1 f (0, 0) = 0 = D2 f (0, 0), equation (14.56) yields

hk √
√ = 0 + 0 + 0 + h2 + k 2 (h, k)
h2 + k 2
hk
∴ (h, k) = 2 .
h + k2

Now, let (h, k) → (0, 0) along the line k = h, then

h2 1
lim (h, k) = lim (h, h) = lim 2
=
(h,k)→(0,0) h→0 h→0 2h 2

This contradicts the fact that


lim (h, k) = 0.
(h,k)→(0,0)

Hence f is not differentiable at (0, 0).

Exercise:

1. Let f : R2 → R defined by

x3


p (x, y) 6= (0, 0)
f (x, y) = x2 + y 2

0, (x, y) = (0, 0)

then show that f is continuous at (0, 0) has partial derivatives at (0, 0), but not differ-
entiable at (0, 0). Solution: First, let  be given, consider

x3
|f (x, y) − f (0, 0)| = p −0
x2 + y 2
r3 cos3 (θ)
= 2 2
+ r2 sin2 (θ) [∵ x = r cos(θ), and y = r sin(θ)]
r cos (θ)
≤ r
p
= x2 + y 2

So, we can choose δ = , then


p
|f (x, y) − f (0, 0)| < , whenever k|(x, y)k| = x2 + y 2 < δ.

204
∴ f is continuous at (0, 0). Next

h3
f (0 + h, 0) − f (0, 0) 2
fx (0, 0) = lim = lim h = 1.
h→0 h h→0 h

0
f (0, 0 + k) − f (0, 0) 2
fy (0, 0) = lim = lim k = 0.
k→0 k k→0 k

Therefore, f has partial derivatives at (0, 0).


Finally, if f were differentiable at (0, 0), then for sufficiently small h and k. We would
have

f (h, k) = f (0, 0) + h fx (0, 0) + k fy (0, 0) + h2 + k 2 (h, k). (14.57)

where lim (h, k) = 0. Since fx (0, 0) = 1 and fy (0, 0) = 0, we have by equation


(h,k)→(0,0)
(14.57)

h3 √
= 0 + h + h2 + k 2 (h, k)
h2 + k 2
h3 h
(h, k) = 2 2 3/2
−√
(h + k ) h2 + k 2

If (h, k) tends to (0, 0) along the line k = h, we have

lim (h, k) = lim (h, h)


(h,k)→(0,0) h→0

h3
 
h
= lim −√
h→0 (h2 + h2 )3/2 h2 + h2
h3
 
h
= lim √ −√
h→0 2 2 h 3 2h
 
1 1
= lim √ −√
h→0 2 2 2
1
= − √
2 2

This is contradict the fact that lim (h, k) = 0.


(h,k)→(0,0)
Hence f is not differentiable at (0, 0).

2. Let f : R2 → R defined by
 2 2
 y(x + y )

(x, y) 6= (0, 0)
f (x, y) = x2 + y 2
0,

(x, y) = (0, 0)

then show that f is continuous at (0, 0), find the partial derivatives and examine
whether it is differentiable or not at (0, 0).

205
p
3. Show that f (x, y) = |xy| is continuous at (0, 0), but not differentiable at (0, 0).

4. Examine the function f (x, y) = |x| + |y| for differentiability at (0, 0).

Mean value theorem (Basic)

Now we establish a result of fundamental importance in the theory of partial differenti-


ation. It is analogous to the corresponding theorem of function of one variable.

Theorem 14.39. Suppose

1. Let f be real valued continuous function defined on an open disc D.

2. D1 f and D2 f exist on D.

3. (x0 , y0 ) ∈ D,

then for sufficiently small h and k, we have

f (x0 + h, y0 + k) − f (x0 , y0 ) = hD1 f (x0 + θ1 h, y0 + k) + kD2 f (x0 , y0 θ2 k),

for some θ1 , θ2 ∈ (0, 1).

Proof. We have

f (x0 +h, y0 +k)−f (x0 , y0 ) = [f (x0 + h, y0 + k) − f (x0 , y0 + k)]+[f (x0 , y0 + k) − f (x0 , y0 )]


(14.58)
Define φ : [0, 1] →, by

φ(t) = f (x0 + th, y0 + k)

Since D1 f exists, φ is differentiable. In fact

φ(s) − φ(t)
φ0 (t) = lim
s→t s−t
f (x0 + sh, y0 + k) − f (x0 + t h, y0 + k)
= lim
s→t hs − ht
f (x0 + sh, y0 + k) − f (x0 + t h, y0 + k)
= lim h
s→t hs − ht
0
φ (t) = h D1 f (x0 + t h, y0 + k)

Thus, by the mean value theorem for a function of one variable, we have

φ(1) − φ(0) = φ0 (θ1 ), for some θ1 ∈ (0, 1)


f (x0 + h, y0 + k) − f (x0 , y0 + k) = h D1 f (x0 + θ1 h, y0 + k), 0 < θ1 < (14.59)
1.

206
Next, define
ψ : [0, 1] → R by,
ψ(t) = f (x0 , y0 + tk)

Since D2 f exists, ψ is differentiable. In fact

ψ 0 (t) = k D2 f (x0 , y0 + t k)

Again by the mean value theorem for a function of one variable, we have

ψ(1) − ψ(0) = ψ 0 (θ2 ), for some θ2 ∈ (0, 1)


f (x0 , y0 + k) − f (x0 , y0 ) = k D2 f (x0 , y0 + θ2 k), 0 < θ2 < 1. (14.60)

Employing equation (14.59) and (14.60) in (14.58),we obtain

f (x0 + h, y0 + k) − f (x0 , y0 ) = hD1 f (x0 + θ1 h, y0 + k) + kD2 f (x0 , y0 θ2 k),

where θ1 , θ2 ∈ (0, 1). This completes the proof.

Theorem 14.40. Let f be a real valued continuous function defined on an open disc D
containing (x0 , y0 ). If D1 f and D2 f exists and are continuous at (x0 , y0 ), then f is differ-
entiable at (x0 , y0 ).

Proof. By the mean value theorem, we have

f (x0 + h, y0 + k) − f (x0 , y0 ) = hD1 f (x0 + θ1 h, y0 + k) + kD2 f (x0 , y0 θ2 k), (14.61)

where θ1 , θ2 ∈ (0, 1).


Now, since D1 f is continuous at (x0 , y0 ), we have

lim D1 f (x0 + θ1 h, y0 + k) = D1 f (x0 , y0 )


(h,k)→(0,0)

Also since D2 f is continuous at (x0 , y0 ), we have

lim D2 f (x0 , y0 + θ2 k) = D2 f (x0 , y0 )


k→0

Define
1 (h, k) = D1 f (x0 + θ1 h, y0 + k) − D1 f (x0 , y0 ) (14.62)

and
2 (h, k) = D2 f (x0 , y0 + θ2 k) − D2 f (x0 , y0 ) (14.63)

note that both 1 (h, k) and 2 (h, k) tend to 0 as (h, k) → (0, 0).

207
Using equations (14.62) and (14.63) in (14.61). We obtain

f (x0 + h, y0 + k) = f (x0 , y0 ) + h D1 f (x0 , y0 ) + k D2 f (x0 , y0 ) + h 1 (h, k) + k 2 (h, k)


f (x0 + h, y0 + k) = f (x0 , y0 ) + h D1 f (x0 , y0 ) + k D2 f (x0 , y0 ) + k(h, k)k (h, k)

h1 (h, k) + k2 (h, k)


where (h, k) = √ .
h2 + k 2
Finally, we have

h1 (h, k) + k2 (h, k)


0 ≤ |(h, k)| = √
h2 + k 2
h 1 (h, k) k 2 (h, k)
≤ √ + √
2
h +k 2 h2 + k 2
≤ |1 (h, k)| + |2 (h, k)|
→ 0 as (h, k) → (0, 0)
hence lim (h, k) = 0.
(h,k)→(0,0)

This proves that f is differentiable at (x0 , y0 ).

Example:

1. Find the θ1 and θ2 from the function f (x, y) = x2 + y 2 + x3 at (x0 , y0 ) = (0, 0).
Solution: Given (x0 , y0 ) = (0, 0), then

f (1, 2) = 12 + 22 + 13 = 1 + 4 + 1 = 6.

f (1 + h, 2 + k) − f (1, 2) = (1 + h)2 + (2 + k)2 + (1 + h)3 − 6


= 1 + 2 h + h2 + 4 + 4 k + k 2 + 1 + 3 h + 3 h2 + h3 − 6
= 5 h + 4 h2 + 4 k + k 3 + h3
= h [5 + 4 h + h2 ] + k [4 + k]
= h [2(1 + θ1 h) + 3(1 + θ1 h)2 ] + k [2(2 + θ2 k)]
= h D1 f (1 + θ1 h, 2 + θ2 k) + k D2 f (1, 2 + θ2 k),

where

2(1 + θ1 h) + 3(1 + θ1 h)2 = 5 + 4 h + h2


2 + 2 θ1 h + 3 + 6 θ1 h + 3 θ12 h2 = 5 + 4 h + h2
i.e., 3 θ12 h2 + 8 θ1 h − (4 h + h2 ) = 0

−4 + 64 + 12 h + 3 h2
∴ θ1 =
3h

208
and

2 k(2 + θ2 k) = 4 k + k 2
1
i.e., θ2 = .
2

where θ1 , θ2 ∈ (0, 1).

Partial derivatives of composite function


Here we shall derive the chain rule for finding the partial derivatives of composite functions.

Theorem 14.41. [Chain rule]


Let x and y be two real valued continuous functions defined on a non-empty subset G of R2 .
Assume that x have partial derivative of first order at the point (r0 , s0 ) ∈ G
Suppose F is a real valued continuous function defined on an open subset D of R2 which
contains the set x(G) × y(G). If F has continuous partial derivatives of first order in D then
the function f : G → R defined by

f (r, s) = F (x(r, s), y(r, s))

also possess partial derivative of first order of first order at the point (r0 , s0 ). In fact at
(r0 , s0 ) we have

∂f ∂F ∂x ∂F ∂y ∂f ∂F ∂x ∂F ∂y
= + and = +
∂r ∂x ∂r ∂y ∂r ∂s ∂x ∂s ∂y ∂s

Proof. Consider

f (r0 + h, s0 ) − f (r0 , s0 ) = F (x(r0 + h, s0 ), y(r0 + h, s0 )) − F (x(r0 , s0 ), y(r0 , s0 )) (14.64)

Since x and y are continuous at (r0 , s0 ), we can write

x(r0 + h, s0 ) = x(r0 , s0 ) + δx

y(r0 + h, s0 ) = y(r0 , s0 ) + δy,

where δx → 0, δy → 0 as h → 0 substituting these in the equation (14.64), we get

f (r0 + h, s0 ) − f (r0 , s0 ) = F (x + δx, y + δy) − F (x, y)


f (r0 + h, s0 ) − f (r0 , s0 ) = δx D1 F (x + θ1 δx, y + δy) + δy D2 F (x + δx, y + θ2 δy),

for some θ1 , θ2 ∈ (0, 1) on using the mean value theorem, thus


 
∂F f (r0 + h, s0 ) − f (r0 , s0 )
= lim
∂r (r0 ,s0 )
h→0 h
 
δx δy
= lim D1 F (x + θ1 δx, y + δy) + D2 F (x + δx, y + θ2 δy)
h→0 h h

209
x(r0 + h, s0 ) − x(r0 , s0 )
= lim D1 F (x + θ1 δx, y + δy)
h→0 h
y(r0 + h, s0 ) − y(r0 , s0 )
+ lim D2 F (x + δx, y + θ2 δy)
 h→0  h
∂F ∂x ∂F ∂y
= +
∂x ∂r ∂y ∂r (r0 ,s0 )

similarly, we can prove


   
∂F ∂F ∂x ∂F ∂y
= +
∂s (r0 ,s0 ) ∂x ∂r ∂y ∂r (r0 ,s0 )

Hence the theorem.

Note: This can be generalized further


If u = f (x1 , x2 , x3 , ..., xm ) and xi = φi (r1 , r2 , r3 , ..., rn ), 1 ≤ i ≤ m
then
∂u ∂f ∂x1 ∂f ∂x2 ∂f ∂xm
= + + ... + , for j = 1, 2, . . . , n.
∂rj ∂x1 ∂rj ∂x2 ∂rj ∂xm ∂rj

Homogeneous functions
Let E be an open subset of R2 , such that (x, y) ∈ E implies (tx, ty) ∈ E, for all t > 0. We
say that, a real valued function f defined on E is a homogeneous function of degree α, if

f (tx, ty) = tα f (x, y), ∀ t > 0 and (x, y) ∈ E.

Examples:

1. The function f (x, y) = x2 + 2 x y + 3y 2 is homogeneous function of degree 2, for

F (t x, t y) = (t x)2 + 2 (t x)(t y) + 3 (t y)2


= t2 x2 + 2 x y + 3 y 2


= t2 f (x, y), ∀ t > 0, (x, y) ∈ R2

Hence f is a homogeneous function of degree 2.

y

2. Consider the function f (x, y) = 3 + log x
, x, y 6= 0 then
 
ty
f (t x, t y) = 3 + log
tx
= t0 f (x, y)

Hence f is a homogeneous function of degree 0.

210
3. Let f (x, y) = x1/3 y −2/3 + x2/3 y −1/3
then

f (t x, t y) = (tx)1/3 (t y)−2/3 + (tx)2/3 (t y)−1/3


= t1/3 x1/3 t−2/3 y −2/3 + t2/3 x2/3 t−1/3 y −1/3

this is not homogeneous function.

Theorem 14.42. [Euler’s theorem for homogeneous function]


Let E be a non-empty open subset of R2 , such that (x, y) ∈ E implies (tx, ty) ∀t > 0.
Let f be a real valued differentiable function defined on E, then it is homogeneous of degree
α if and only if
x D1 f (x, y) + y D2 f (x, y) = α f (x, y). (14.65)

Proof. First, suppose f is a homogeneous function of degree α defined on E.


Then, define φ : R+ → R by,

φ(t) = f (t x, t y) = tα f (x, y). (14.66)

Since f is differentiable on E, φ is differentiable on R+ and we have on other hand

φ0 (t) = x D1 f (t x, t y) + y D2 f (t x, t y)
and also φ0 (t) = αtα−1 f (x, y)
∴ x D1 f (t x, t y) + y D2 f (t x, t y) = α tα−1 f (x, y)
putting t = 1, we get
x D1 f (x, y) + y D2 f (x, y) = α f (x, y).

This proves equation (14.65).


Conversely, Suppose equation (14.65) holds.
Define ψ : R+ → R by
ψ(t) = t−α f (t x, t y). (14.67)

Since f is differentiable on E, ψ is differentiable on R+ and

ψ 0 (t) = −α t−α−1 f (t x, t y) + t−α [x D1 f (t x, t y) + y D2 f (t x, t y)] (14.68)

In the equation(14.65) replace x by t x and y by t y, we get

t x D1 f (t x, t y) + t y D2 f (t x, t y) = α f (t x, t y)
i.e., x D1 f (x, y) + y D2 f (x, y) = α t−1 f (x, y). (14.69)

211
Employing equation (14.68) and (14.69), we obtain

ψ 0 (t) = −α t−α−1 f (t x, t y) + α t−α−1 f (t x, t y)


ψ 0 (t) = 0

Hence ψ is constant. In particular, we have

ψ(t) = ψ(1)
∴ t−α f (t x, t y) = f (x, y)
i.e., f (t x, t y) = tα f (x, y).

Thus f is homogeneous function of degree α. Hence the proof.

Exercise:
x3 + y 3
 
−1
1. Let f (x, y) = tan x 6= y.
x−y
Show that x D1 f (x, y) + y D2 f (x, y) = sin(2 f (x, y)).
Solution: we have
 3
x + y3

−1
f (x, y) = tan
x−y
3 3
x +y
tan(f (x, y)) =
x−y
x3 + y 3
Let F (x, y) = ,
x−y

then

(t x)3 + (t y)3
F (t x, t y) =
tx − ty
 3
x + y3

2
= t
x−y
2
F (t x, t y) = t F (x, y).

By Euler theorem

x D1 F (x, y) + y D2 F (x, y) = 2 F (x, y)


x D1 tan(f (x, y)) + y D2 tan(f (x, y)) = 2 tan(f (x, y))
x sec2 ((f (x, y)) D1 (f (x, y)) + y sec2 (f (x, y)) D2 (f (x, y)) = 2 tan(f (x, y))
2 tan(f (x, y))
i.e., x D1 f (x, y) + y D2 f (x, y) =
sec2 (f (x, y))
x D1 f (x, y) + y D2 f (x, y) = 2 sin(f (x, y)) cos(f (x, y))
∴ x D1 F (x, y) + y D2 F (x, y) = sin(2f (x, y)).

212
x5 + y 5
 
−1
2. Suppose f (x, y) = tan , x 6= y,
x−y
then show that x D1 f (x, y) + y D2 f (x, y) = 2 sin(2f (x, y)).

x4 + y 4
 
−1
3. Suppose f (x, y) = tan , x 6= y
x−y
3
then show that x D1 f (x, y) + y D2 f (x, y) = sin(2f (x, y)).
2

xk + y k
 
−1
4. In general, If f (x, y) = tan , x 6= y
x−y
k−1
then show that x D1 f (x, y) + y D2 f (x, y) = sin(2f (x, y)).
2

Higher order Partial derivatives


Let f be real valued function defined on an open subset E of R2 . Suppose D1 f and
D2 f exist on E. In general D1 f and D2 f are also functions of x and y and hence may have
partial derivatives. The partial derivatives of D1 f and D2 f are called second order partial
derivatives of f .
Thus the partial derivatives of second order are defined by

D1 f (x + h, y) − D1 f (x, y)
D1 (D1 f (x, y)) = lim
h→0 h
D2 f (x, y + k) − D2 f (x, y)
D2 (D2 f (x, y)) = lim
k→0 k
D2 f (x + h, y) − D2 f (x, y)
D1 (D2 f (x, y)) = lim
h→0 h
D1 f (x, y + k) − D1 f (x, y)
and D2 (D1 f (x, y)) = lim ,
k→0 k

or it is also represented as fxx , fyy , fxy and fy,x . Among these second partial derivatives
fxy and fy,x are called mixed partial derivatives.
In general fxy need not to be equal tofy,x .

Exercise:

1. Let f : R2 → R defined by
 2 2
 x y (x − y ) , if (x, y) 6= (0, 0)

f (x, y) = x2 + y 2
0,

if (x, y) = (0, 0)

213
Show that D12 f (0, 0) 6= D21 f (0, 0).
Solution:First

f (0 + h, 0) − f (0, 0)
D1 f (0, 0) = lim
h→0 h
f (h, 0) − f (0, 0)
= lim
h→0 h
D1 f (0, 0) = 0.

f (0 + h, k) − f (0, k)
D1 f (0, k) = lim
h→0 h
f (h, k) − f (0, k)
= lim
h→0 h
h k (h2 −k2 )
2 2 −0
= lim h +k
h→0 h
k (h − k 2 )
2
= lim
h→0 h2 + k 2
D1 f (0, k) = −k.
f (0, 0 + k) − f (0, 0)
D2 f (0, 0) = lim
k→0 k
f (0, k) − f (0, 0)
= lim
k→0 k
D2 f (0, 0) = 0.

f (h, 0 + k) − f (h, 0)
D2 f (h, 0) = lim
k→0 k
f (h, k) − f (h, 0)
= lim
k→0 k
h k (h − k 2 )
2
−0
= lim h2 + k 2
k→0 k
h (h − k 2 )
2
= lim
k→0 h2 + k 2
D1 f (0, k) = h.
So, D12 f (0, 0) = D1 (D2 f (0, 0))
D2 f (0 + h, 0) − D2 f (0, 0)
= lim
h→0 h
D2 f (h, 0) − D2 f (0, 0)
= lim
h→0 h
h−0
= lim
h→0 h
D12 f (0, 0) = 1.

214
and D21 f (0, 0) = D2 (D1 f (0, 0))
D1 f (0, 0 + k) − D1 f (0, 0)
= lim
k→0 k
D1 f (0, k) − D1 f (0, 0)
= lim
k→0 k
−k − 0
= lim
k→0 k
D21 f (0, 0) = −1.
∴ D12 f (0, 0) 6= D21 f (0, 0).

Theorem 14.43. [Young’s Theorem]


Let f be a real valued continuous function defined on an open subset E of R2 . If D1 f and
D2 f exists on E and are differentiable at (x0 , y0 ) ∈ E, then

D12 f (x0 , y0 ) = D21 f (x0 , y0 ).

Proof. We may assume that (x0 , y0 ) = (0, 0), then D1 f and D2 f exists on E and E is an
open subset of R2 . So there exists a real number r > 0, such that D1 f and D2 f are defined
on the open square (−r, r) × (−r, r) ⊂ E. By hypothesis, D1 f and D2 f are differentiable
at (0, 0).
Define φ : [0, 1] → R, by
φ(t) = f (t h, h) − f (t h, 0),

then φ is differentiable and

φ0 (t) = h [D1 f (t h, h) − D1 f (t h, 0)] .

By the mean value theorem, we have

φ(1) − φ(0) = φ0 (θ), where 0 < θ < 1. (14.70)

Put Ω(h) = φ(1) − φ(0), then

Ω(h) = f (h, h) − f (h, 0) − f (0, h) + f (0, 0)


Ω(h) = h [D1 f (θ h, h) − D1 f (θ h, 0)] (14.71)

Now, since D1 f is differentiable at (0, 0), we have

D1 f (θ h, h) = D1 f (0, 0) + θ h D11 f (0, 0) + h D21 f (0, 0) + k(θ h, h)k 1 (θ h, h), (14.72)

where 1 (θ h, h) → 0 as h → 0. Similarly,

D1 f (θ h, 0) = D1 f (0, 0) + θ h D11 f (0, 0) + 0 + k(θ h, 0)k 2 (θ h, 0), (14.73)

215
where 2 (θ h, 0) → 0 as h → 0. Using equation (14.72) and (14.73) in (14.71), we get
h √ √ i
Ω(h) = h h D21 f (0, 0) + θ2 h2 + h2 1 (θ h, h) − θ2 h2 2 (θ h, 0)
Ω(h) √ √
i.e., = D f (0, 0) + θ 2 + 1  (θ h, h) − θ2 2 (θ h, 0)
21 1
h2
Ω(h)
Thus lim 2 = D21 f (0, 0). (14.74)
h→0 h

and the limit exists by hypothesis.


Similarly, by considering the function ψ : [0, 1] → R defined by

ψ(t) = f (h, th) − f (0, th),

then ψ is differentiable and

ψ 0 (t) = h D2 f (h, t h) − h D2 f (0, t h)


ψ 0 (t) = h [D2 f (h, t h) − D2 f (0, t h)]

We can show as before that

Ω(h)
lim = D12 f (0, 0). (14.75)
h→0 h2

From equations (14.74) and (14.75), it follows that

D12 f (x0 , y0 ) = D21 f (x0 , y0 ).

Hence the theorem.

Theorem 14.44. Let f be a real valued continuous function defined on an non-empty open
subset E of R2 and D1 f is continuous at a point (x0 , y0 ) ∈ E,, then f is differentiable at
(x0 , y0 ).

Proof. We may assume that (x0 , y0 ) = (0, 0). Since (x0 , y0 ) ∈ E and E is open, there exits
a positive real number r1 , such that

S = (−r, r) × (−r, r) ⊂ E

Now we can choose r and k, such that

0 < r < r1 and 0 < k < r1

then the rectangle (−r, r) × (−r, r) ⊂ E.


Now, we choose h such that 0 < h < r and Define φ : [0, 1] → R by

φ(t) = f (t, k),

216
then φ is differentiable and φ0 (t) = D1 f (t, k)
By the mean value theorem, we have

φ(h) − φ(0) = h φ0 (θh), 0 < θ < 1


= h D1 f (θh, k)
i.e., f (h, k) − f (0, k) = h D1 f (θh, k)

for some θ ∈ (0, 1).


Since D1 f is continuous as (0, 0), we have

D1 f (θh, k) = D1 f (0, 0) + ε1 (h, k), (14.76)

where ε1 (h, k) → 0 as (h,k)→(0,0).


Since D2 f (0, 0) exists, we can write

f (0, k) − f (0, 0) = k D2 f (0, 0) + k ε2 (0, k), (14.77)

where ε2 (0, k) → 0 as k→0


By (14.76),(14.76) and (14.77), we have

f (h, k) − f (0, 0) = [f (h, k) − f (0, k)] + [f (0, k) − f (0, 0)]


= h D1 f (θh, k) + [f (0, k) − f (0, 0)]
= h [D1 f (0, 0) + ε1 (h, k)] + k D2 f (0, 0) + k ε2 (0, k)
= h D1 f (0, 0) + k D2 f (0, 0) + h ε1 (h, k) + k ε2 (h, k)
f (h, k) − f (0, 0) = h D1 f (0, 0) + k D2 f (0, 0) + k(h, k)kε(h, k)

where

h ε1 (h, k) + k ε2 (0, k)
ε(h, k) = → 0 as (h, k) → (0, 0)
k(h, k)k

Hence,

f (h, k) = f (0, 0) + h D1 f (0, 0) + k D2 f (0, 0) + k(h, k)k ε(h, k)

where ε(h, k) → 0, as (h,k)→(0,0)


∴ f is differentiable at (0, 0). This completes the proof.

Exercises

1. Suppose f is homogeneous function of degree α, then show that

x2 D11 f (x, y) + xyD12 f (x, y) + yxD21 f (x, y) + y 2 D22 f (x, y) = α(α − 1)f (x, y)

217
. What continuity assumption are you making?
Solution Define

F (x, y) = x D1 f (x, y) + y D2 f (x, y) = α f (x, y)

then

F (t x, t y) = α f (t x, t y) = α tα f (x, y) = tα F (x, y).

∴ F is a homogeneous function of degree α. Now,

α(α − 1) f (x, y) = (α − 1) F (x, y)


= α F (x, y) − F (x, y)
= [x D1 F (x, y) + y D2 F (x, y)] − F (x, y) [By Euler’s Theorem]
= {x [D1 f (x, y) + x D11 f (x, y) + y D12 f (x, y)]}
+{y [xD21 f (x, y) + D2 f (x, y) + y D22 f (x, y)]}
−x D1 f (x, y) − y D2 f (x, y)
= x2 D11 f (x, y) + x y D12 f (x, y) + y x D21 f (x, y) + y 2 D22 f (x, y).

as desired. The second partial derivatives of f exist and are continuous and hence F
is Differentiable.

JACOBIANS

Definition 18. If f1 , f2 , . . . , fn are functions of n variables x1 , x2 , . . . , xn then, the Jacobian


of f1 , f2 , . . . , fn with respect to x1 , x2 , . . . , xn is denoted by

∂(f1 , f2 , . . . , fn )
∂(x1 , x2 , . . . , xn )
and is defined as the determinant

∂f1 ∂f1 ∂f1


∂x1 ∂x2
··· ∂xn
 
∂f2 ∂f2 ∂f2 ∂fi
∂x1 ∂x2
··· ∂xn = det
.. .. .. .. ∂xj
. . . .
∂fn ∂fn ∂fn
∂x1 ∂x2
··· ∂xn

some times the jacobian is denoted by


 
f1 , f2 , . . . , fn
J
x1 , x2 , . . . , xn

Examples:-

218
1. If x = r cos θ,and y = r sin θ, then

  ∂x ∂x
x, y ∂r ∂θ cos θ −r sin θ
= r cos2 θ + sin2 θ = r

J = ∂y ∂y =
r, θ sin θ r cos θ
∂r ∂θ

2. If x = ρ cos θ sin φ, y = ρ sin θ sin φ, and z = ρ cos φ, then

∂x ∂x ∂x
∂ρ ∂θ ∂φ cos θ sin φ −ρ sin θ sin φ ρ cos θ cos φ
∂(x, y, z) ∂y ∂y ∂x
= = sin θ sin φ ρ cos θ sin φ ρ sin θ cos φ
∂(ρ, θ, φ) ∂ρ ∂θ ∂φ
∂z ∂z ∂x cos φ 0 −ρ sin φ
∂ρ ∂θ ∂φ
cos φ −ρ2 sin2 θ sin φ cos φ − ρ2 cos2 θ sin φ cos φ
 
=
− ρ sin φ ρ cos2 θ sin2 φ + ρ sin2 θ sin2 φ
 

−ρ2 sin φ cos2 φ sin2 θ + cos2 θ − ρ2 sin3 φ cos2 θ + sin2 θ


   
=
−ρ2 sin φ cos2 φ + sin2 φ
 
=
= −ρ2 sin φ.

Exercise:-

 
x2 x3 x1 x3 x1 x2 f1 , f2 , f3
1. If f1 = , f2 = and f3 = , then find J
x1 x2 x3 x1 , x2 , x3

Solution:

∂f1 ∂f1 ∂f1


  ∂x1 ∂x2 ∂x3
f1 , f2 , f3 ∂f2 ∂f2 ∂f2
J =
x1 , x2 , x3 ∂x1 ∂x2 ∂x3
∂f3 ∂f3 ∂f3
∂x1 ∂x2 ∂x3
−x2 x3 x3 x2
2
x1 x1 x1
x3 −x1 x3 x1
= 2
x2 x2 x2
x2 x1 −x1 x2
x3 x3 x23

219
−x2 x3 x21 x2 x3 x21
     
x3 −x1 x2 x3 x1 x2 x 2 x1 x3 x1 x2 x3
= − − − + + 2
x21 x22 x23 x 2 x3 x1 x2 x23 x2 x3 x1 x2 x3 x2 x3
−x2 x3 x3 2x1 x2 2x1
= ·0+ · + ·
x21 x1 x3 x 1 x2
= 4

2.

If y1 = 1 − x1
y2 = x1 (1 − x2 )
y3 = x1 x2 (1 − x3 )
..
.
yn = x1 x2 · · · xn−1 (1 − xn ),

 
y1 , y2 , · · · , yn
find J =
x1 , x2 , . . . , xn
 
y1 , y2 , . . . , yn
Answer:-J = = (−1)n (x1 )n−1 (x2 )n−2 . . . (xn−1 )
x1 , x2 , . . . , xn
 
y1 , y2 , . . . , yn
3. If yi + yi+1 + . . . + yn = x1 x2 . . . xn , i = 1, 2, . . . , n find J =
x1 , x2 , . . . , xn

Answer:- xn−1
1 xn−2
2 . . . xn−1
[Hint:First find y1 , y2 , . . . , yn , then find the Jacobian]

 
x, y, z
4. If x = cos u, y = sin u cos v, z = sin u sin v cos w, find J
u, v, w

5. If u3 + v 3 = x + y , u2 + v 2 = x2 + y 2 , then prove that


 
u, v 1 y−x
J =
x, y 3 uv(u − v)

Theorem 14.45. If y1 , y2 , . . . , yn are functions of u1 , u2 , . . . , un and u1 , u2 , . . . , un are func-


tions of x1 , x2 , . . . , xn , then
     
y1 , y2 , . . . , yn y1 , y2 , . . . , yn u1 , u2 , . . . , um
J =J ·J .
x1 , x2 , . . . , xn u1 , u2 , . . . , um x1 , x2 , . . . , xn
Proof. We have, by chain rule

220
∂yi ∂yi ∂u1 ∂yi ∂u2 ∂yi ∂un
= · + · + ... + ·
∂xj ∂u1 ∂xj ∂u2 ∂xj ∂un ∂xj
n
∂yi X ∂yi ∂ur
= · , 1 ≤ i, j ≤ n
∂xj r=1
∂ur ∂xj

       
y1 , y2 , . . . , yn u1 , u2 , . . . , un ∂yi ∂ur
∴ J ·J = det · det
u1 , u2 , . . . , un x1 , x2 , . . . , xn ∂ur n×n ∂xj n×n
n
!
X ∂yi ∂ur
= det ·
r=1
∂ur ∂xj
n×n
 
∂yi
= det
∂xj n×n
 
y1 , y2 , . . . , yn
= J .
x1 , x2 , . . . , xn

Hence the proof.


Corollary 14.8. If u1 , u2 , . . . , un are functions of x1 , x2 , . . . , xn , then

∂ (u1 , u2 , . . . , un ) ∂ (x1 , x2 , . . . , xn )
· = 1.
∂ (x1 , x2 , . . . , xn ) ∂ (u1 , u2 , . . . , un )

Proof. Put y1 = x1 , y2 = x2 , . . . , yn = xn then by above theorem, we have

∂ (x1 , x2 , . . . , xn ) ∂ (u1 , u2 , . . . , un ) ∂ (x1 , x2 , . . . , xn )


· =
∂ (u1 , u2 , . . . , un ) ∂ (x1 , x2 , . . . , xn ) ∂ (x1 , x2 , . . . , xn )
10 ···
0 0
00 ···
1 0
= ..
.. .. ..
.. ···
. .
0 0 0 ··· 1
= 1.

Theorem 14.46. Let f = (f1 , f2 , . . . , fn ) be a map from G to Rn , where G is an open


subset of Rn . Let E be any open subset of R2n such that G × f (G) ⊂ E suppose Fi , where
i = 1, 2, ..., n are mappings from E to R such that Fi vanish on the set G × f (G), such that

Fi (x1 , x2 , . . . , xn , f1 (x1 , x2 , . . . , xn ), f2 (x1 , x2 , . . . , xn ), . . . , fn (x1 , x2 , . . . , xn )) = 0, ∀i = 1, 2, ..., n,


(14.78)
where fi , Fi , (1 ≤ i ≤ n) are all continuously differentiable functions, then

∂ (F1 , F2 , . . . , Fn ) ∂ (f1 , f2 , . . . , fn ) ∂ (F1 , F2 , . . . , Fn )


· = (−1)n .
∂ (f1 , f2 , . . . , fn ) ∂ (x1 , x2 , . . . , xn ) ∂ (x1 , x2 , . . . , xn )

221
Proof. Differentiating the equation (14.78) given equation partially with respect to xj , we
obtain

∂Fi ∂Fi ∂f1 ∂Fi ∂f2 ∂Fi ∂fn


+ · + · + ... + · = 0
∂xj ∂f1 ∂xj ∂f2 ∂xj ∂fn ∂xj
n
X ∂Fi ∂fr ∂Fi
i.e., · = − , for j = 1, 2, . . .(14.79)
,n
r=1
∂fr ∂xj ∂xj

Therefore,
   
∂ (F1 , F2 , . . . , Fn ) ∂ (f1 , f2 , . . . , fn ) ∂Fi ∂fr
· = det · det
∂ (f1 , f2 , . . . , fn ) ∂ (x1 , x2 , . . . , xn ) ∂fr n×n ∂xjn×n
n
!
X ∂Fi ∂fr
= det
r=1
∂fr ∂xj
n×n
 
∂Fi
= det − [∵ from the equation (14.79)]
∂xj
 
n ∂Fi
= (−1) det
∂xj
∂ (F1 , F2 , . . . , Fn )
= (−1)n
∂ (x1 , x2 , . . . , xn )

Hence the theorem.

Exercise:

1. If u, v and w are the roots of the equation (t − x)3 + (t − y)3 + (t − z)3 = 0


∂(u, v, w)
find .
∂(x, y, z)

Solution: Since u, v and w are the roots of the given equation, we have

(t − x)3 + (t − y)3 + (t − z)3 = 3(t − u)(t − v)(t − w)

i.e., 3t3 − 3t2 (x + y + z) + 3t(x2 + y 2 + z 2 ) − (x3 + y 3 + z 3 )

= 3[t3 − t2 (u + v + w) + t(uv + uw + vw) − uvw]

comparing the co-efficient, we get

x + y + z = u + v + w, x2 + y 2 + z 2 = uv + vw + wu, x3 + y 3 + z 3 = 3uvw

Define F1 , F2 , F3 by

F1 (x, y, z, u, v, w) = u + v + w − (x + y + z) = 0.

222
F2 (x, y, z, u, v, w) = uv + vw + wu − (x2 + y 2 + z 2 ) = 0.

F3 (x, y, z, u, v, w) = 3uvw − (x3 + y 3 + z 3 ) = 0.

Then by the above theorem, we have

∂ (F1 , F2 , F3 ) ∂ (u, v, w) ∂ (F1 , F2 , F3 )


· = (−1)3
∂ (u, v, w) ∂ (x, y, z) ∂ (x, y, z)
∂ (F1 , F2 , F3 )
∂ (u, v, w) ∂ (x, y, z)
= (−1)3 (14.80)
∂ (x, y, z) ∂ (F1 , F2 , F3 )
∂ (u, v, w)

Therefore we shall find,

∂F1 ∂F1 ∂F1


∂x ∂y ∂z

∂(F1 , F2 , F3 ) ∂F2 ∂F2 ∂F2


=
(x, y, z) ∂x ∂y ∂z

∂F3 ∂F3 ∂F3


∂x ∂y ∂z
−1 −1 −1

= −2x −2y −2z

−3x2 −3y 2 −3z 2

−1 0 0
= −2x −2(y − x) −2(z − x) [∵ c02 = c2 − c1 , c03 = c3 − c1 ]
−3x2 −3(y 2 − x2 ) −3(z 2 − x2 )
= −6(x − y)(y − z)(z − x)

and also

∂F1 ∂F1 ∂F1


∂u ∂v ∂w 1 1 1

∂(F1 , F2 , F3 ) ∂F2 ∂F2 ∂F2


= = v+w u+w u+v = −3(u − v)(v − w)(w − u)
(u, v, w) ∂u ∂v ∂w

∂F3 ∂F3 ∂F3 3vw 3uw 3uv


∂u ∂v ∂w

223
Therefore equation (14.80) becomes

∂(u, v, w) −6(x − y)(y − z)(z − x) (x − y)(y − z)(z − x)


= (−1)3 = −2
∂(x, y, z) −3(u − v)(v − w)(w − u) (u − v)(v − w)(w − u)

2. Prove that the functions u = x + y − z, v = x − y + z, w = x2 + y 2 + z 2 − 2yz are


not independent of one another. Find a relation between them.
Solution We compute the Jacobian

∂u ∂u ∂u
∂x ∂y ∂z

∂(u, v, w) ∂v ∂v ∂v
=
∂(x, y, z) ∂x ∂y ∂z

∂w ∂w ∂w
∂x ∂y ∂z
1 1 −1

= 1 −1 1

2x 2(y − z) 2(z − y)
= 1[−2(z − y) − 2(y − z)] − 1[2(z − y) − 2x] − 1[2(y − z) + 2x]
= 0.

Therefore, the functions are not independent of one another.


Next, we observe that
u + v = 2x, u − v = 2(y − z) and w = x2 + (y − z)2

4w = (u + v)2 + (u − v)2 = 2(u2 + v 2 )


or u2 + v 2 = 2w.

x y z
3. Prove that the functions u = ,v= ,w= are not independent of
y−z z−x x−y
one another.

xj
4. If uj = q , j = 1, 2, ..., n. Show that
2 2 2
1 − xj − xj+1 − . . . − xn

∂(u1 , u2 , . . . , un ) 1
=p .
∂(x1 , x2 , . . . , xn ) 1 − x1 − x22 − . . . − x2n
2

224
5. If u3 + v + w = x + y 2 + z 2 , u + v 3 + w = x2 + y + z 2 , u + v + w3 = x2 + y 2 + z.
Show that
∂(u, v, w) 1 − 4(xy + yz + zx) + 16xyz
= .
∂(x, y, z) 2 − 3(u2 + v 2 + w2 ) + 27u2 v 2 w2

14.1.3 Keywords
Homogenous Function, Critical Point, Maxima, Minima and Jacobian.

14.1.4 Terminal Problems


Find, for each of the following functions, all critical points and classify them
1. f (x, y) = x4 + y 4 − x2 − y 2 + 10
2. f (x, y) = x3 + y 3 − 3x − 12y + 20
3. f (x, y) = x3 + y 3 − 63x − 63y + 12xy
4. f (x, y) = 2x2 + 2y 2 − xy − 20x
5. f (x, y) = 2(x − y)2 − x4 − y 4
6. f (x, y) = x2 + 2xy + 2y 2 + 4x
7. f (x, y) = x3 − y 2 + 3x2 + 3y 2 − 9x
8. f (x, y) = x2 − xy + y 4
9. f (x, y) = (x + y)3 + (x − y)2 − 12(x + y)
10.Show that f (x, y) = y 2 + x2 y + x4 has a minimum at (0,0)

14.1.5 Books for reference


1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd


Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

225
Block IV: Functions of several variables

Unit 15:Taylor’s Theorem for a function of n variable and


The contraction principle theorem
15.1.1 Main Objectives
15.1.2 Taylor’s Theorem for a function of n variables
15.1.4 inverse function theorem
15.1.5 Keywords
15.1.6 Books for Reference

122
226
Block IV
Unit 4: Taylor’s Theorem for a function of n variables and
The contraction principle theorem

15.1.1 Main Objectives


Taylor’s theorem gives an approximation of a n-times differentiable function around a given
point by a polynomial of degree k, called the nth-order Taylor polynomial. For a smooth
function, the Taylor polynomial is the truncation at the order n of the Taylor series of the
function. The first-order Taylor polynomial is the linear approximation of the function,
and the second-order Taylor polynomial is often referred to as the quadratic approximation.
There are several versions of Taylor’s theorem, some giving explicit estimates of the approx-
imation error of the function by its Taylor polynomial.

The contraction principle theorem is an important tool in the theory of metric spaces;
it guarantees the existence and uniqueness of fixed points of certain self-maps of metric
spaces, and provides a constructive method to find those fixed points. It can be understood
as an abstract formulation of Picard’s method of successive approximations.

15.2 Taylor’s Theorem for a function of n variables


:

Let f be a real valued function defined on an open subset E of Rn . If

x = (x1 , x2 , . . . , xn ) ∈ E, t = (t1 , t2 , . . . , tn ) ∈ E,

then we write
n
X
(1)
f (x, t) = Di f (x) ti
i=1
Xn Xn
f (2) (x, t) = Dij f (x) ti tj
i=1 j=1
Xn X n Xn
f (3) (x, t) = Dijk f (x) ti tj tk and so on
i=1 j=1 k=1

Let a, b ∈ E, we say that E contains the line segment L(a, b), if it contains the set

L(a, b) = {a + λ(b − a)\0 ≤ λ ≤ 1}

Theorem 15.47. Taylor’s Theorem

227
Suppose f is a real valued function defined on an open subset E of Rn .
Assume that partial derivatives of f of all orders less than m exist and are differentiable.
Let a, b ∈ E, such that L(a, b) ⊂ E, then there exists a point z ∈ L(a, b) such that

m−1
X f (k) (a, b − a) f (m) (z, b − a)
f (b) − f (a) = + .
k=1
k! m!

Proof. Since L(a, b) is a subset of E and E is open, it follows that all points which are
sufficiently near to the points of L(a, b) belong to E. That is there exists a δ > 0 such that

a + t(b − a) ∈ E if t ∈ (−δ, 1 + δ)

Define a real valued function g : (−δ, 1 + δ) → R by

g(t) = f (a + t(b − a)).

Now, it follows that all derivatives of g up to (m − 1)th order exist and are continuous
and mth derivative of g exists. Hence by the Taylors theorem for a function of one variable.
We have m
X g (k) θ g (m) θ
g(1) − g(0) = + , where θ ∈ (0, 1). (15.81)
k=1
k! m!

Note that

g(t) = f (a + t(b − a))


= f (a1 + t(b1 − a1 ), a2 + t(b2 − a2 ), . . . , an + t(bn − an ))
∴ g 0 (t) = D1 f [a + t(b − a)](b1 − a1 ) + D2 f [a + t(b − a)](b2 − a2 ) +
. . . + Dn f [a + t(b − a)](bn − an )
Xn
= Di f [a + t(b − a)](bi − ai )
i=1
(1)
= f (a + t(b − a), (b − a))
∴ g (1) (0) = f (1) (a, b − a).

Similarly, we can show that

g (k) (0) = f (k) (a, b − a) and


g (m) (θ) = f (m) (a + θ(b − a), b − a)
= f (m) (z, b − a), where z = a + θ(b − a) ∈ L(a, b).

Substituting these values in (15.81), we obtain

m−1
X f (k) (a, b − a) f (m) (z, b − a)
f (b) − f (a) = +
k=1
k! m!

228
for some z = a + θ(b − a) ∈ L(a, b). This completes the proof of the theorem.

The Taylor’s theorem for a function of two variables can be stated and

Theorem 15.48. Taylor’s theorem for a function of two variables


If f (x, y) possesses continuous partial derivatives up to the nth order in a neighborhood of
(a, b) in its domain and (a + h, b + k) is a point in this neighborhood, then
   2
∂ ∂ 1 ∂ ∂
f (a + h, b + k) = f (a, b) + h +k f (a, b) + h +k f (a, b) + . . .
∂x ∂y 2! ∂x ∂y
 n−1  n
1 ∂ ∂ 1 ∂ ∂
+ h +k f (a, b) + h +k f (a + θh, b + θk),
(n − 1)! ∂x ∂y n! ∂x ∂y

Where 0 < θ < 1.

Proof. Define φ : [0, 1] → R0 by

φ(t) = f (a + th, b + tk), x = a + th and y = b + tk.

Since f possesses continuous partial derivatives up to the nth order. φ possesses continuous
derivatives up to the nth order. Hence

φ(t) = f (a + th, b + tk), φ(0) = f (a, b)


∂f ∂x ∂f ∂y
φ0 (t) = · + ·
∂x ∂t ∂y ∂t
 
0 ∂f ∂f ∂ ∂
φ (t) = h +k = h +k f (a + th, b + tk)
∂x ∂y ∂x ∂y
 
0 ∂ ∂
∴ φ (0) = h +k f (a, b)
∂x ∂y
Similarly,
 2
00 ∂ ∂
φ = h +k f (a + th, b + tk)
∂x ∂y
 2
00 ∂ ∂
∴ φ (0) = h +k f (a, b)
∂x ∂y
and in general,
 k
k ∂ ∂
φ (0) = h +k f (a, b) 1 ≤ k ≤ n
∂x ∂y

Therefore, by Taylor’s theorem for a function of one-variable, we have


n−1
t t2 tn−1 (n − 1)! tn n
φ(t) = φ(0) + φ0 (0) + φ00 (0) + . . . + (0) + φ (θt), where 0 < θ < 1.
1! 2! φ n!

229
Putting t = 1, we get

1 0 1 1 1
φ(1) = φ(0) + φ (0) + φ00 (0) + . . . + φn−1 (0) + φn (θ), where 0 < θ < 1.
1! 2! (n − 1)! n!

Substituting the values of φ(0), φ0 (0), . . . , φn (θ),


   2
∂ ∂ 1 ∂ ∂
f (a + h, b + k) = f (a, b) + h +k f (a, b) + h +k f (a, b) +
∂x ∂y 2! ∂x ∂y
 n−1  n
1 ∂ ∂ 1 ∂ ∂
... + h +k f (a, b) + h +k f (a + θh, b + θk),
(n − 1)! ∂x ∂y n! ∂x ∂y

where 0 < θ < 1.

This completes the proof.

Note: Taylor’s series is also called as "Power series" with (x − a) and (y − b).

Exercise

1. Express f (x, y) = x2 y + 3y − 2 in powers of (x − 1) and (y + 2).


Solution:Here a = 1 and b = −2, given

f (x, y) = x2 y + 3y − 2.

We know that
   2
1 ∂ ∂ 1 ∂ ∂
f (a + h, b + k) = f (a, b) + h +k f (a, b) + h +k f (a, b)
1! ∂x ∂y 2! ∂x ∂y
 3
1 ∂ ∂
+ h +k f (a, b).
3! ∂x ∂y

fx (x, y) = 2xy ∴ fx (1, −2) = 2.1(−2) = −4


fy (x, y) = x2 + 3 ∴ fy (1, −2) = 1 + 3 = 4
fxx (x, y) = 2y ∴ fxx (1, −2) = 2(−2) = −4
fxy (x, y) = 2x = fyx (x, y) ∴ fxy (1, −2) = 2.1 = 2
fyy (x, y) = 0 ∴ fyy (1, −2) = 0
fxxx (x, y) = 0 ∴ fxxx (1, −2) = 0
fxxy (x, y) = 2 = fxyx (x, y) = fyyx (x, y) ∴ fxxy (1, −2) = 2
fyyy (x, y) = 0 ∴ fyyy (1, −2) = 0
fxyy (x, y) = 0 = fxxy (x, y) = fyyx (x, y) ∴ fxyy (1, −2) = 0

230
All higher partial derivatives are zero.

1
f (x, y) = f (1, −2) + (x − 1)fx (1, −2) + (y + 2)fy (1, −2) + (x − 1)2 fxx (1, −2)
2!
1 1 1
+ 2(x − 1)(y + 2)fxy (1, −2) + (y + 2) fyy (1, −2) + (x − 1)3 fxxx (1, −2)
2
2! 2! 3!
1 1 1
+ 3(x − 1) (y + 2)fxxy (1, −2) + 3(x − 1)(y + 2) fxyy (1, −2) + (y + 2)3 fyyy (1, −
2 2
3! 3! 3!
= −10 − 4(x − 1) + 4(y + 2) − 2(x − 1)2 + 2(x − 1)(y + 2) + (x − 1)2 (y + 2).

2. Expand f (x, y) = sin x sin y at the origin up to 4th degree.

a2 x2 − b2 y 2 a3 x3 − 3a2 bxy 2
3. Prove that eax cos by = 1 + ax + + + . . ..
2! 3!

4. Express the following functions in powers of (x − 1) and (y − 2).

(a) f (x, y) = x2 + xy + y 2 ,
(b) f (x, y) = x3 + y 3 + xy 2 .

5. Express f (x, y) = ex tan−1 y in powers of (x − 1) and (y − 1).

MAXIMA AND MINIMA

Definition 19. Let f be a real valued function defined on a subset E of R2 and (x0 , y0 ) ∈ E.

If for all (x, y) in some neighborhood of (x0 , y0 ). We have f (x, y) ≤ f (x0 , y0 ). We say
that f has a Local maximum at (x0 , y0 ) and f (x0 , y0 ) is called a Local maximum of f .

If for all (x, y) in some neighborhood of (x0 , y0 ). We have f (x0 , y0 ) ≤ f (x, y). We say
that f has a Local minimum at (x0 , y0 ) and f (x0 , y0 ) is called a Local minimum of f .

Theorem 15.49. Suppose f is continuous and has first order partial derivatives on E. If f
has a Local maximum (or Local minimum) at a point (x0 , y0 ) ∈ E, then

D1 f (x0 , y0 ) = 0, D2 f (x0 , y0 ) = 0.

Proof. First, Suppose that f has a Local maximum at (x0 , y0 ), then for all sufficiently small
|h| and |k|, we have
f (x0 + h, y0 + k) ≤ f (x0 , y0 ). (15.82)

Letting k → 0, we obtain
f (x0 + h, y0 ) ≤ f (x0 , y0 )

231
Since f is continuous on E. Also, since D1 f exists at (x0 , y0 ) we have

f (x0 + h, y0 ) − f (x0 , y0 )
D1 f (x0 , y0 ) = lim
h→0
 h
≤ 0, if h > 0
=
≥ 0, if h < 0.

Therefore, We have
D1 f (x0 , y0 ) = 0.

Similarly, letting h → 0 in (15.82), we have

f (x0 , y0 + k) ≤ f (x0 , y0 ).

Since f is continuous on E. Also, since D2 f exists at (x0 , y0 ), we have

f (x0 , y0 + k) − f (x0 , y0 )
D2 f (x0 , y0 ) = lim
k→0
 h
≤ 0, if k > 0
=
≥ 0, if k < 0.

Therefore, we have
D2 f (x0 , y0 ) = 0.

Next, if f has a local minimum at (x0 , y0 ), then the proof follows similarly.

Remark

1. The above condition is only a necessary condition for the existence of an Extreme
value (Maximum or Minimum value) of a function having partial derivatives.
In fact a function can have an extreme value without possessing partial derivatives.
For example, consider the function

f (x, y) = |x| + |y| ,

then f (0, 0) = 0 < f (x, y) = |x| + |y| , ∀(x, y) near (0, 0), but (x, y) 6= (0, 0). Thus
f has a local minimum at (0, 0). But f has no partial derivatives at (0, 0).

2. A function can have vanishing partial derivatives at a point, but not having a Local
maximum or Local minimum at that point.
For example, consider the function

f (x, y) = x y,

then we have
D1 f (x, y) = y, D2 f (x, y) = x

232
D1 f (x, y) = 0, D2 f (x, y) = 0, yields x = 0, y = 0.

That is at (0, 0), both the partial derivatives vanishes


In every neighborhoods of (0, 0), we can find points (t, t) and also the points (t, −t)
at which have

f (t, t) = t2 > 0 = f (0, 0); f (t, −t) = −t2 < 0 = f (0, 0)

When t 6=0. Therefore, f has neither a local maximum nor a local minimum at (0, 0).

Definition 20. Let f be a real valued function defined on an open subset E of R2 . If


(x0 , y0 ) ∈ E is such that D1 f (x0 , y0 ) = 0, D2 f (x0 , y0 ) = 0, then the point (x0 , y0 ) is
called a Critical Point of f.
∴ Sometimes Critical Points are called Stationary points.

Definition 21. A Critical Point at which f has neither a maximum nor a minimum is called
a Saddle Point.

Examples:

1. Consider the function f (x, y) = xy.


As seen above, (0, 0) is a Saddle Point of f .

2. Consider the function f (x, y) = x2 + y 2 − 6x − 8y + 26, then fx = 0, fy = 0 yield


2x − 6 = 0 and 2y − 8 = 0

i.e., x = 3, y = 4

∴ (3, 4) is a Critical Point of f .

f (3, 4) = 32 + 42 − 6 · 3 − 8 · 4 + 26
= 1

for |h| and |h| sufficiently small, we have

f (3 + h, 4 + k) = (3 + h)2 + (4 − k)2 − 6(3 + h) − 8(4 + k) + 26


= 9 + 6h + h2 + 16 + 8k + k 2 − 18 − 6h − 32 − 8k + 26
= h2 + k 2 + 1
> 1 = f (3, 4)
i.e., f (3 + h, 4 + k) > f (3, 4)

233
∴ f has a local minimum at (3, 4).

3. Consider the function f (x, y) = x2 + y 2 + 4xy, then fx = 0, fy = 0 yield 2x + 4y = 0


and 2y + 4x = 0
Solving these equations simultaneously, we get x = 0 and y = 0
∴ (0, 0) is a critical point of f .

f (0, 0) = 0

In every neighborhood of (0, 0), we can find points (t, t) and (t, −t), where 0 < t < 1
such that

f (t, t) = t2 + t2 + 4t2 = 6t2 > 0 = f (0, 0).


f (t, −t) = t2 + t2 − 4t2 = −2t2 < 0 = f (0, 0).

Therefore, f attains neither a maximum nor a minimum at (0, 0). Hence (0, 0) is a
Saddle point of f.
We shall now prove a theorem which gives a sufficient condition for a function f of
two variables to have a maximum or a minimum.

Theorem 15.50. Suppose f is a real valued function defined on an open set G ⊂ R2 and
has continuous partial derivatives of second order.
Let (x0 , y0 ) ∈ G be a critical point of f .
If at (x0 , y0 )

D11 f · D22 f − (D12 f )2 > 0 and D11 f > 0 or D22 f > 0,

then f has a maximum at (x0 , y0 ).


If at (x0 , y0 )
D11 f · D22 f − (D12 f )2 < 0,

then (x0 , y0 ) is a saddle point of f .


If at (x0 , y0 )
D11 f · D22 f − (D12 f )2 = 0,

then the theorem gives no information.

Proof. Let (x0 + h, y0 + k) be a point in a neighborhood of (x0 , y0 ). Since f has continuous


partial derivatives of second order. We have by Taylor’s theorem

1 2
f (x0 + h, y0 + k) − f (x0 , y0 ) = hD1 f (x0 , y0 ) + kD2 f (x0 , y0 ) +
h D11 f (x0 + θh, y0 + θk)
2
+ 2hkD12 f (x0 + θh, y0 + θk) + k 2 D22 f (x0 + θh, y0 + θk) ,


234
where 0 < θ < 1.
Since D1 f (x0 , y0 ) = 0 = D2 f (x0 , y0 ), we have

1 2
h D11 f (α, β) + 2hkD12 f (α, β) + k 2 D22 f (α, β)(15.83)

f (x0 + h, y0 + k) − f (x0 , y0 ) = ,
2

where x0 + θh = α, y0 + θk = β,
[Note that D12 f = D21 f , since the second partial derivatives of f are continuous].
Suppose that at (x0 , y0 ),

∆ = D11 f · D22 f − (D12 f )2 > 0, and D11 f > 0 (or D22 f > 0)

then both D11 f (x0 , y0 ) and D22 f (x0 , y0 ) are distinct from 0, for if

D11 f (x0 , y0 ) = 0, D22 f (x0 , y0 ) = 0,

then we would have


∆ = −(D12 f )2 ≤ 0

a contradiction to the hypothesis.


Also, D11 f (x0 , y0 ) and D22 f (x0 , y0 ) must be of the same sign, Otherwise we get a contra-
diction to the hypothesis as above.
Therefore,if D11 f (x0 , y0 ) > 0, then D22 f (x0 , y0 ) > 0. Since the second order partial
derivatives of f are continuous there exists a neighborhood of (x0 , y0 ) in which

D11 f (x, y)D22 f (x, y) − [D12 f (x, y)]2 > and D11 f (x, y) > 0

So, We can choose (h,k) such that (x0 + h, y0 + k) belongs to this neighborhood. Since
0 < θ < 1 in (9.6)
We have D11 f (α, β)D22 f (α, β) − [D12 f (α, β)]2 > 0 and D11 f (α, β) > 0

1 2
h D11 f (α, β) + 2hkD12 f (α, β) + k 2 D22 f (α, β)

∴ f (x0 + h, y0 + k) − f (x0 , y0 ) =
2
1
(hD11 f (α, β) + kD12 f (α, β))2 + k 2 (D11 f (α, β)D22 f (α, β) − (D12 f (α, β))2 ) > 0
 
=
2D11 f (α, β)

Therefore, we have
f (x0 , y0 ) < f (x0 + h, y0 + k) (15.84)

for all sufficiently small (h, k), such that (x0 + h, y0 + k) is in the neighborhood of (x0 , y0 ).
Hence f (x0 , y0 ) is a minimum of f or f has a minimum at (x0 , y0 ).
This proves (15.84)
Next, suppose that at (x0 , y0 ),

D11 f · D22 f − (D12 f )2 > 0 and D11 f < 0 (or D22 f < 0)

235
then, as before, it follows that

D11 f (α, β)D22 f (α, β) − [D12 f (α, β)]2 > 0 and D11 f (α, β) < 0

Hence by (15.83),

f (x0 + h, y0 + k) − f (x0 , y0 ) < 0


f (x0 + h, y0 + k) < f (x0 , y0 ),

for all sufficiently small |h| and |k|. Therefore, f has a maximum at (x0 , y0 ) or f (x0 , y0 ) is
a maximum of f .
More specifically, let D11 f > 0. Since the second order partial derivatives are continuous,
there exists a neighborhood of (x0 , y0 ) in which

D11 f (α, β)D22 f (α, β) − [D12 f (α, β)]2 < 0 and D11 f (α, β) > 0

But then, from (15.83) we have

f (x0 + h, y0 + k) − f (x0 , y0 )
1
(hD11 f (α, β) + kD12 f (α, β))2 + k 2 (D11 f (α, β)D22 f (α, β) − (D12 f (α, β))2 )
 
=
2D11 f (α, β)
> 0, if k = 0, h 6= 0
D12 f (α, β)
< 0, if k 6= 0, h = −k .
D11 f (α, β)

Thus, there exists a neighborhood of (x0 , y0 ) in which there are points (x, y) for which
f (x, y) > f (x0 , y0 ) and also the points (x, y) for which f (x, y) < f (x0 , y0 )
Hence f has neither maximum nor minimum at (x0 , y0 ).

Similarly, if
D11 f (x0 , y0 ) < 0 and D11 f · D22 f − (D12 f )2 < 0,

it follows, as above that f has neither maximum nor minimum at (x0 , y0 )

Next, if
D11 f (x0 , y0 ) = 0, but D22 f (x0 , y0 ) 6= 0

then by interchanging the roles of the two in above discussion we see that f has neither
maximum nor minimum at (x0 , y0 ).

Finally, if
D11 f (x0 , y0 ) = 0 and D22 f (x0 , y0 ) = 0

then D12 f (x0 , y0 ) 6= 0,, since D11 f · D22 f − (D12 f )2 < 0. Since the second partial deriva-

236
tives of f are continuous, there exists a neighborhood of (x0 , y0 ) in which, we have

D11 f (α, β) = 0, D22 f (α, β) = 0 and D12 f (α, β) 6= 0.

from (15.83), we have

f (x0 + h, y0 + k) − f (x0 , y0 ) = hkD12 f (α, β) (15.85)


for k = h in the equation(15.87) we have
f (x0 + h, y0 + h) − f (x0 , y0 )
lim = D12 f (x0 , y0 ) (15.86)
h→0 h2
and for k = −h in the equation(15.87), we obtain
f (x0 + h, y0 − h) − f (x0 , y0 )
lim = −D12 f (x0 , y0 ) (15.87)
h→0 k2

Since D12 f (x0 , y0 ) 6= 0, and D12 f is continuous, it follows that D12 f maintains its sign in
every neighborhood of (x0 , y0 ).
Hence the equations (15.86) and (15.87) imply that

f (x0 + h, y0 + k) − f (x0 , y0 ) textchangesitssignnear (x0 , y0 )

. Hence f has neither maximum nor minimum at (x0 , y0 ). Therefore (x0 , y0 ) is a Saddle
point.
Finally, we consider the three functions

f (x, y) = x4 + y 4 , g(x, y) = −(x4 + y 4 ) and h(x, y) = x4 − y 4 .

Clearly, (0, 0) is a Critical Point of all the three functions and at (0, 0), we have

D11 f · D22 f − (D12 f )2 = 0


D11 g · D22 g − (D12 g)2 = 0 and
D11 h · D22 h − (D12 h)2 = 0

But (0, 0) is a minimum point of f , a maximum point of g and a saddle point of h.


Therefore the theorem gives no information, if

D11 f · D22 f − (D12 f )2 = 0 at the Critical Point.

This completes the proof.

Exercise:

1. Examine the following function for extreme values

f (x, y) = y 2 + 4xy + 3x2 + x3 .

237
Solution: We have f (x, y) = y 2 + 4xy + 3x2 + x3 .

∴ D1 f = 4y + 6x + 3x2
D2 f = 2y + 4x

D1 f = 0, D2 f = 0 yield

∴ 3x2 + 6x + 4y = 0, (15.88)
2x + y = 0 ⇒ y = −2x (15.89)

Substituting the equation (15.89) in (15.88), we get

3x2 + 6x − 8x = 0
x(3x − 2) = 0

D22 f = 2 , D12 f = 4
 
2 2 −4
∴ D22 f − (D12 f ) = 12 − 16 = −4 < 0, at , ,
3 3
D11 f = 10, D22 f = 2, D12 f = 4

∴ D11 f · D22 f − (D12 f )2 = 20 − 16 = 4 > 0

and D11 f = 10 > 0 (or D22 f = 2 >  0) 


2 −4
∴ (0, 0) is a saddle point of f and , is a minimum point of f .
3 3  
2 −4
In other words, f has no extremum at (0, 0), but has a minimum at , .
3 3

2. Show that f (x, y) = 2x4 − 3x2 y + y 2 has neither a maximum nor a minimum at (0, 0).
Solution: We have

f (x, y) = 2x4 − 3x2 y + y 2


∴ D1 f = 8x3 − 6xy, D2 f = −3x2 + 2y
D11 f = 24x2 − 6y, D22 f = 2, D12 f = −6x
∴ D11 f · D22 f − (D12 f )2 = 0 at (0, 0).

Therefore, the above theorem cannot be applied.


Consider for small |h| and |k|,

f (h, k) − f (0, 0) = 2h4 − 3h2 k + k 2


= (h2 − k)(2h2 − k)

238
Now,
 
2 2 2 k 2
(h − k)(2h − k) > 0, if k < h or if h < min k, or if k < 0.
2
k
(h2 − k)(2h2 − k) < 0, if 0 < < h2 < k.
2

∴ f (h, k) − f (0, 0) takes both positive and negative values for sufficiently small |h|
and |k|.
∴ (0, 0) is a saddle point.
In other words f has neither a maximum nor a minimum at (0, 0).

15.2.1 The contraction principle


We now interrupt our discussion of differentiation to insert a fixed point theorem that is
valid in arbitrary complete metric spaces. It will be used in the proof of the inverse function
theorem.

Definition 22. Let X be a metric space, with metric d. If ϕ maps X into X and if there is a
number c < 1 such that
d (ϕ(x), ϕ(y)) ≤ c d(x, y) (15.90)

for all x, y ∈ X, then ϕ is said to be a contraction of X into X.

Theorem 15.51. If X is a complete metric space, and if ϕ is a contraction of X into X,


then there exists one and only one x ∈ X such that ϕ(x) = x.
In other words, ϕ has a unique fixed point. The uniqueness is a triviality, for if ϕ(x) = x and
ϕ(y) = y, then (15.90) gives d(x, y) ≤ cd(x, y), which can only happen when d(x, y) = 0.
The existence of a fixed point of ϕ is the essential part of the theorem. The proof actually
furnishes a constructive method for locating the fixed point.

Proof. Pick x0 ∈ X arbitrarily, and define {xn } recursively, by setting

xn+1 = φ(xn ), ∀ n = 0, 1, 2, · · · . (15.91)

Choose c < 1 so that (15.90) holds. For n ≥ 1 we then have

d (xn+1 , xn ) = d (ϕ(xn ), ϕ(xn−1 )) ≤ c d (xn , xn−1 ) .

Hence induction gives

d (xn+1 , xn ) ≤ cn d (xn , xn−1 ) , ∀ n = 0, 1, 2, · · · . (15.92)

239
If n ≤ m, it follows that
m
X
d (xn , xm ) ≤ d (xi , xi−1 )
i=n+1

c + cn+1 + · · · + cm−1 d(x1 , x0 )


n


≤ (1 − c)−1 d(x1 , x0 ) cn .
 

Thus {xn } is a Cauchy sequence. Since X is complete, lim xn = x for some x ∈ X.


n→∞

Since ϕ is a contraction, ϕ is continuous (in fact, uniformly continuous) on X. Hence

ϕ(x) = lim ϕ(xn ) = lim xn+1 = x.


n→∞ n→∞

15.2.2 Keywords
Cauchy Sequence, Metric Space and Linear Property.

15.2.3 Books for reference


1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd


Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

240
Block IV: The inverse function theorem
and Functions of several variables

Unit 16:Inverse function theorem and Implicit function the-


orem
16.1.1 Main Objectives
16.1.2 The inverse function theorem
16.1.3 Implicit function theorem
16.1.4 Keywords
16.1.5 Terminal Problems
16.1.6 Books for Reference

137
241
Block IV
Unit 4:Inverse function theorem and Implicit function
theorem

16.1.1 Main Objectives


Inverse function theorem gives a sufficient condition for a function to be invertible in a
neighborhood of a point in its domain: namely, that its derivative is continuous and non-
zero at the point. The theorem also gives a formula for the derivative of the inverse function.
In multivariable calculus, this theorem can be generalized to any continuously differentiable,
vector-valued function whose Jacobian determinant is nonzero at a point in its domain, giv-
ing a formula for the Jacobian matrix of the inverse. There are also versions of the inverse
function theorem for complex holomorphic functions, for differentiable maps between man-
ifolds, for differentiable functions between Banach spaces, and so forth.

The technique of implicit differentiation allows you to find the derivative of y with re-
spect to x without having to solve the given equation for y. The chain rule must be used
whenever the function y is being differentiated because of our assumption that y may be
expressed as a function of x.

16.1.2 The Inverse Function Theorem


The inverse function theorem states, roughly speaking, that a continuously differentiable
mapping f is invertible in a neighborhood of any point x at which the linear transformation
f 0 (x) is invertible:

Theorem 16.52. Suppose f is a `− mapping of ana open set E ⊂ Rn into Rn , f’(a) is


invertible for some a ∈ E, and b = f(a). Then

1. there exist open sets U and V in Rn such that a ∈ U, b ∈ V, f is one to one on U, and
f(U)=V;

2. if g is the inverse of f [which exists, by (1)], define in V by

g(f (x)) = x, x∈U

then g ∈ `0 (V ).

Writing the equation y=f(x) in component form, we arrive at the following interpretation of
the conclusion of the theorem: The system of n equations

yi = fi (x1 , . . . , xn ) , 1 ≤ i ≤ n

can be solved for x1 , . . . , xn in terms of y1 , . . . , yn , if we restrict x and y to small enough


neighborhoods of a and b, the solutions are unique and continuously differentiable.

242
Proof. 1. Put f 0 (a) = A, and choose λ so that

2 λ kA−1 k = 1. (16.93)

Since f 0 is continuous at a, there is an open ball U ⊂ E, with center at a, such that

kf 0 (x) − Ak < λ, x ∈ U. (16.94)

We associate to each y ∈ Rn a function ϕ, defined by

ϕ(x) = x + A−1 (y − f (x)) , x ∈ E. (16.95)

Note that f (x) = y if and only if x is a fixed point of ϕ.


Since ϕ0 (x) = 1 − A−1 f 0 (x) = A−1 (A − f 0 (x)) , (16.93) and (16.94) imply that

1
kϕ0 (x)k < , x ∈ U. (16.96)
2

Hence
1
|ϕ(x1 ) − ϕ(x2 )| ≤ |x1 − x2 | , x1 , x2 ∈ U, (16.97)
2
by Theorem
(Suppose f maps a convex open set E ⊂ Rn into Rm , f is differentiable in E, and
there is a real number M such that

kf 0 (x)k ≤ M

for every x ∈ E. Then


|f (b) − f (a)| ≤ M |b − a|

for all a, b ∈ E.) it follows that ϕ has at most one fixed point in U, so that f (x) = y
for at most one x ∈ U. Thus f is 1 − 1 in U.
Next, put V = f (U ), and pick y0 ∈ V. Then y0 = f (x0 ) for some x0 ∈ U. Let B be
an open ball with center at x0 and radius r >), so small that its closure B̄ lies in U .
We will show that y ∈ V whenever |y − y0 | < λ r. This proves, of course, that V is
open. Fix y, |y − y0 | < λ r. With ϕ as in (16.94),

r
|ϕ(x0 ) − x0 | = A−1 (y − y0 ) < kA−1 k λ r = .
2

If x ∈ B̄, it therefore follows from (16.97) that

|ϕ(x) − x0 | ≤ |ϕ(x) − ϕ(x0 )| + |ϕ(x0 ) − x0 |


1 r
< |x − x0 | +
2 2
≤ r;

243
hence ϕ(x) ∈ B. Note that (16.97) holds if x1 ∈ B̄, x2 ∈ B̄.

Thus ϕ is a contraction of B̄ into B̄. Being a closed subset of Rn , B̄ is complete.


Theorem 2.14 implies therefore that ϕ has a fixed point x ∈ B̄. For x, f (x) = y.
Thus y ∈ f (B̄) ⊂ f (U ) = V.
This proves part (1) of the theorem.

2. Pick y ∈ V, y + k ∈ V. Then there exist x ∈ U, x + h ∈ U, so that y = f (x), y + k =


f (x + h). With ϕ as in (16.95),

ϕ(x + h) − ϕ(x) = h + A−1 [f (x) − f (x + h)] = h − A−1 k.

By (16.97),
1
|h − A−1 k| ≤ |h|
2
Hence |A−1 k| ≥ 21 |h|, and

|h| ≤ 2kA−1 k |k| = λ−1 |k|. (16.98)

By (16.93), (16.94), and Theorem ∗ (Theorem ∗ Let Ω be the set of all invertible linear
operators on Rn .

(a) If A ∈ Ω, B ∈ L (Rn ) , and

kB − Ak · kA−1 k < 1,

then B ∈ Ω.
(b) Ω is an open subset of L (Rn ) , and the mapping A → A−1 is continuous on Ω.

), f 0 (x) has an inverse, say T. Since

g(y + k) − g(y) − T k = h − T k = −T [f (x + h) − f (x) − f 0 (x) h]

(16.98)

|g(y + k) − g(y) − T k| kT k |f (x + h) − f (x) − f 0 (x) h|


≤ ·
|k| λ |h|

As k → 0, (16.98) shows that h →). The right side of the last inequality thus tends to
0. Hence the same is true of the left. We have thus proved that g 0 (t) = T. But T was
chosen to be the inverse of f 0 (x) = f 0 (g(y)). Thus

−1
g 0 (y) = {f 0 (g(y))} , y ∈ V. (16.99)

Finally, note that g is a continuous mapping of g is a continuous mapping of V onto


U (since g is differentiable), that f 0 is a continuous mapping of U into the set Ω of all

244
invertible elements of L(Rn ), and that inversion is a continuous mapping of Ω onto Ω,
by Theorem ∗ . If we combine these facts with (16.99), we see g ∈ `0 (V ).
This completes the proof.

16.1.3 The implicit function theorem

The equation of a curve in the plane can be expressed either in an explicit form y = f (x)
or in an implicit form such as f (x, y) = 0. However, we are given an equation of the form
f (x, y) = 0, it does not necessarily represent the function. So it is natural to ask "When is
the relation defined by f (x, y) = 0 also a function?". The implicit function theorem deals
with this question locally.

Theorem 16.53. Let f be a real valued continuously differentiable function defined on an


open subset G of R2 . Suppose (x0 , y0 ) ∈ G, such that

1. f (x0 , y0 ) = 0,

2. D2 f (x0 , y0 ) 6= 0

and φ is differential on (x0 − h, x0 + h), then

D1 f (x, φ(x))
φ0 (x) = − .
D2 f (x, φ(x))

Proof. Since D2 f (x0 , y0 ) 6= 0. We may assume that D2 f (x0 , y0 ) > 0.


Since D2 f is continuous, there exists a neighborhood (x0 − h1 , x0 + h1 ) of (x0 , y0 ), such
that D2 f (x, y) > 0, if
|x − x0 | < h1 and |y − y0 | < k1 .

Hence, for x ∈ (x0 − h1 , x0 + h1 ) the function f (x, y) is strictly increasing for y ∈ (y0 −
k1 , y0 + k1 ).
Taking x = x0 , we have

f (x0 , y0 −k) < f (x0 , y0 ) < f (x0 , y0 +k), where 0 < k < k1 f (x0 , y0 −k) < 0 < f (x0 , y0 +k)

Since f is continuous as a function of first variable from (16.100) it follows that there exists
a real number h > 0 and h ≤ h1 , such that

f (x, y0 − k) < 0 < f (x, y0 + k), if x ∈ (x0 − h, x0 + h).

Hence by the intermediate value theorem there is a unique point [∵ f is increasing]


y ∈ (y0 − k, y0 + k), such that
f (x, y) = 0.

245
Define a map
φ : (x0 − h, x0 + h) → (y0 − h, y0 + h)

by setting φ(x) to be the unique point y ∈ (y0 − k, y0 + k), such that

f (x, y) = 0.

By the definition of φ, we have φ(x0 ) = y0 and f (x, φ(x)) = 0, ∀ x ∈ (x0 − h, x0 + h).


Now, we show that φ is differentiable on (x0 − h, x0 + h).
So, let x ∈ (x0 − h, x0 + h) Choose δx small enough, such that

x + δx ∈ (x0 − h, x0 + h).

Put φ(x) = y and φ(x + δx) = y + δy, then we have

f (x, y) = 0 and f (x + δx, y + δy) = 0.

∴ f (x + δx, y + δy) − f (x, y) = 0.

Applying basic mean value theorem, we obtain

δx D1 f (x + θ1 δx, y + δy) + δy D2 f (x, y + θ2 δy) = 0.


δy D2 f (x, y + θ2 δy) = −δx D1 f (x + θ1 δx, y + δy)
(16.100)
δy D1 f (x + θ1 δx, y + δy)
= − ,
(16.101)
δx D2 f (x, y + θ2 δy)

where 0 < θ1 < 1, 0 < θ2 < 1


From (16.101), it is clear that δy → 0 as δx → 0
Now,

φ(x + δx) − φ(x) δy


lim = lim
δx→0 δx δx
δx→0
D1 f (x + θ1 δx, y + δy)
φ0 (x) = lim −
δx→0 D2 f (x, y + θ2 δy)
D1 f (x, y)
= −
D2 f (x, y)
D1 f (x, φ(x))
φ0 (x) = − .
D2 f (x, φ(x))

Thus, φ is differentiable on (x0 − h, x0 + h) and is continuous, as it is the ratio of two


continuous functions.
Hence the proof.

Exercise:

1. Find the derivative of y = φ(x) at x = 1 defined implicitly by the equation x2 y 3 +

246
2x2 y + 3x4 − y = 1.
Solution: Define f by

f (x, y) = x2 y 3 + 2x2 y + 3x4 − y − 1. (16.102)

Here x0 = 1, we have to find y0 such that f (x0 , y0 ) = 0. Substituting x = 1, we get


on equating to 0,

y 3 + 2y + 3 − y − 1 = 0,
i.e., y 3 + y + 2 = 0

Clearly, y = −1, satisfies this equation. We can apply implicit function theorem.

D1 f (x, y) = 2xy 3 + 4xy + 12x3


D1 f (1, −1) = 2 · 1 · (−1) + 4 · 1 · (−1) + 12 · 1 = 6
D2 f f (x, y) = 3x2 y 2 + 2x2 − 1
D2 f (1, −1) = 3 · 1 · 1 + 2 · 1 − 1 = 4
D1 f (1, −1) 6 3
∴ φ0 (1) = − = − = − .
D2 f (1, −1) 4 2

2. Find the derivative of the function y = φ(x) at x = 1 defined implicitly by the


equation e−x cos y + sin y = e.
Solution:Define f by
f (x, y) = e−x cos y + sin y − e.

Clearly for (x0 , y0 ) = (1, π), we have

f (x0 , y0 ) = f (1, π) = e−1 cos π + sin π − e = 0


D2 f (x, y) = x sin y e−x cos y + cos y and
D2 f (1, π) = 1 sin π e− cos π + cos π
= −1
D2 f (1, π) 6= 0

So, we can apply implicit function theorem

D1 f (x, y) = − cos y e−x cos y


∴ D1 f (1, π) = − cos(π) e− cos π
= e.
 
dy D1 f (1, π) e
= φ0 (1) = − =− = e.
dx x=1 D2 f (1, π) −1

247
3. Show that the functional relation x + y + z − xyz = 0 defines on implicit function
z = φ(x, y) near (0,0,0). Hence deduce φx and φy .
Solution:Let f (x, y, z) = x + y + z − xyz, then f (0, 0, 0) = 0 and fx = 1 − yz,fy =
1 − zx,fz = 1 − xy.
Clearly fx , fy and fz are continuous in a neighborhood of (0, 0, 0) and fz 6= 0 at
(0, 0, 0).

x+y
Hence, f (x, y, z) = 0 defines a function z = φ(x, y), given by z = by implicit
xy − 1
function theorem.
Further,

fx 1 − yz
φx = − = − and
fz 1 − xy
fy 1 − zx
φy = − = − .
fz 1 − xy

4. Find the partial derivatives D1 φ(x, y), D2 φ(x, y) at x = 1, y = −1, defined implicitly
by the equation
x2 + y 2 + z 2 − 6 = 0.

Solution:Let f (x, y, z) = x2 + y 2 + z 2 − 6,
then clearly f (1, −1, 2) = 0
fx = 2x, fy = 2y and fz = 2z are all continuous in a neighborhood of (1, −1, 2).
Moreover f (1, −1, 2) = 4 6= 0.

Note: The implicit function theorem can be generalized as follows:

Suppose f (x, y, z) is a function with (x0 , y0 , z0 ) in its domain such that

1. f (x0 , y0 , z0 ) = 0

2. fx ,fy and fz are continuous in a neighborhood of (x0 , y0 , z0 ) and

3. fz ((x0 , y0 , z0 ) 6= 0.

Then there exists a neighborhood R of (x0 , y0 , z0 ), given by

R = {(x, y, z)/ |x − x0 | < h, |y − y0 | < k, |z − z0 | < l}

such that for every (x, y) ∈ G = {(x, y)/ |x − x0 | ≤ h, |y − y0 | ≤ k} there is a unique


z ∈ [z0 − l, z0 + l], thus defining a function z = φ(x, y) which satisfies

248
i) z0 = φ(x0 , y0 )
ii) f (x, y, φ(x, y)) = 0, for every (x, y) ∈ G.
∂z D1 f ∂z D2 f
iii) =− and =− , at (x0 , y0 , z0 ).
∂x D3 f ∂y D3 f

MAXIMA AND MINIMA CONSTRAINTS

Lagrange’s Multipliers:

Lagrange prescribed a method of determining extreme values of a function subjected to


certain constraints.

Theorem 16.54. Suppose that f (x, y) and g(x, y) are continuously differentiable functions
in a domain D and that (D1 g)2 + (D2 g)2 > 0, then the set of points (x, y) on the curve
g(x, y) = 0 where f (x, y) has maxima or minima are included in the set of simultaneous
solutions (x, y, λ) of the equations

D1 f (x, y) + λD1 g(x, y) = 0,


D2 f (x, y) + λD2 g(x, y) = 0 and
g(x, y) = 0.

Proof. Put
u = f (x, y) (16.103)

where x and y are connected by the equation

g(x, y) = 0. (16.104)

By hypothesis f and g are continuously differentiable and (D1 g)2 +(D2 g)2 > 0. If D2 g =
6 0.
We can solve for y in (10.13) treating x as the independent variable and substitute it in
equation (10.12).
du
Then a necessary condition for f to have a maximum or minimum is .
dx

du dy
i.e., = D1 f + D2 f · = 0
dx  dx

D1 g
D1 f + D2 f − = 0, by Implicit function theorem
D2 g
D1 f · D2 g − D2 f · D1 g = 0
∂(f, g)
∴ = 0
∂(x, y)

Thus the desired points will then be included among the simultaneous solutions of the equa-

249
tion 
 ∂(f, g) = 0;

∂(x, y) (16.105)
g(x, y) = 0.

On the other hand, if D1 g 6= 0, we may solve equation (10.13) for x and substitute in
equation (10.14). But in this case also, we are led to the same pair of equations.
To solve the same problem by the method of Lagrange, introduce the multiplier λ and form
the equation
v = f (x, y) + λg(x, y)

Then set

D1 v = D1 f + λD1 g = 0, and (16.106)


D2 v = D2 f + λD2 g = 0. (16.107)

If D2 g 6= 0, then from equation (10.15),we have

D2 f
λ=−
D2 g

Substituting this in equation (9.8), we obtain

D2 f
D1 f − D1 g = 0
D2 g
i.e., D1 f · D2 g − D2 f · D1 g = 0
∂(f, g)
∴ = 0
∂(x, y)

Combining this with (16.104), we arrive at equation (16.105). Thus, instead of solving the
two equations in (16.105) for x and y. We must now solve the three equations (16.105),
(16.106) and (16.107). This completes the proof.

16.1.4 Keywords
Continuity, Differentiability, Maxima, Minima and Lagrange’s Multipliers.

16.1.5 Terminal Problems


1. Find the derivative of the function y = φ(x) and x = 2, defined implicitly by the
relation x2 + xy + y 2 = 7.

dy
2. Examine the following relations for defining implicit functions and then find
dx
(a) sin xy − exy − x2 y = 0

250
(b) xy = y x

(c) xy − logy = a

(d) (tan x)y + y cot x = a

(e) x3 + y 3 − exy = 0
du
(f) If u = f (x, u), find
dx

∂u ∂u
(g) If u = f (g(x, u), h(y, u)), find , .
∂x ∂y

(h) Define f by f (x, y, z) = x2 y+ex +z. Show that f (0, 1, −1) = 0,D1 f (0, 1, −1) 6=
0 and by implicit function theorem further show that there exists a differen-
tiable function g in some neighborhood of (1, −1), such that g(1, −1) = 0 and
f (g(y, z) − y, z) = 0. Find D1 g(1, −1) and D2 g(1, −1)

16.1.6 Books for reference


1. W. Rudin − Principles of Mathematical Analysis, International Student edition, Mc-
Graw Hill, 3rd Edition.

2. T. M. Apostal − Mathematical Analysis, Addison Wesley, Narosa, New Delhi, 2nd


Edition.

3. R. R. Goldberg − Methods of Real Analysis, Oxford and IBH, New Delhi.

4. Torence Too − Analysis I, Hindustan Book Agency, India, 2006.

5. Torence Too − Analysis II, Hindustan Book Agency, India, 2006.

6. Kenneth A. Ross − Elementary Analysis: The Theory of Calculus, Springer Interna-


tional Edition, 2004.

7. R. G. Bartle − The Elements of Real Analysis, Wiley International Edition, New


York, 2nd Edition.

251

You might also like