Measure Theory and Integration 1St Edition Ammar Khanfer Online Ebook Texxtbook Full Chapter PDF
Measure Theory and Integration 1St Edition Ammar Khanfer Online Ebook Texxtbook Full Chapter PDF
Measure Theory and Integration 1St Edition Ammar Khanfer Online Ebook Texxtbook Full Chapter PDF
https://ebookmeta.com/product/fundamentals-of-functional-
analysis-1st-edition-ammar-khanfer/
https://ebookmeta.com/product/basic-analysis-iv-measure-theory-
and-integration-1st-edition-james-k-peterson/
https://ebookmeta.com/product/measure-theory-probability-and-
stochastic-processes-1st-edition-jean-francois-le-gall/
https://ebookmeta.com/product/measure-integral-derivative-a-
course-on-lebesgue-s-theory-1st-edition-sergei-ovchinnikov/
Geometric Measure Theory and Free Boundary Problems
Cetraro Italy 2019 Matteo Focardi Emanuele Spadaro Eds
https://ebookmeta.com/product/geometric-measure-theory-and-free-
boundary-problems-cetraro-italy-2019-matteo-focardi-emanuele-
spadaro-eds/
https://ebookmeta.com/product/classical-and-discrete-functional-
analysis-with-measure-theory-london-mathematical-society-student-
texts-series-number-101-buntinas/
https://ebookmeta.com/product/relational-integration-of-
psychology-and-christian-theology-theory-research-and-
practice-1st-edition-steven-j-sandage-jeannine-k-brown-2/
Ammar Khanfer
Measure
Theory and
Integration
Measure Theory and Integration
Ammar Khanfer
Measure Theory
and Integration
Ammar Khanfer
Department of Mathematics and Sciences
Prince Sultan University
Riyadh, Saudi Arabia
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2023
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface
v
vi Preface
aspect deals with the notion of vector spaces of infinite dimensions. Meanwhile, the
third aspect deals with the notion of differentiation and the idea of employing the
preceding aspects to tackle many problems related to the theory of differential equa-
tions and its application. It can be said that real and functional analysis bridge the
gap between elementary mathematics to advanced mathematics, where more matu-
rity, knowledge, and skills are required to master the subject. This particular area of
mathematics is rich in history and deep in treatment. Above all, it is substantially
required in almost all graduate programs of mathematics. Usually, students study
this subject only when they become familiar with the basic concepts of set theory
(sets and their operations), linear algebra (vector spaces, norms, bases, linear trans-
formations), and classical real analysis (Cauchy sequence, series, the convergence of
sequences, differentiation and integration of functions and sequences of functions). A
good understanding of these concepts not only facilitates the study of the subject but
also provides motivation and a clear understanding of why one should get introduced
to these concepts, which will cast some beauty onto the subject.
One pitfall arises when reading this area of mathematics from books. The
complexity of the treatments of the books to this subject and the deepness of its
concepts which may end up giving the subject a negative reputation among some
weak and average students that the subject is too complicated and difficult to under-
stand. Many excellent books are available about the subject. However, the exposi-
tion in most of them might not be simple enough for students to fully grasp, and
thus, the gap between the student and this subject continues to exist and even widen.
Therefore, a typical book for the student should discuss these areas of analysis with a
comprehensive treatment and clear exposition and in a self-contained reader-friendly
text.
With this goal in mind, we have started this project to write a series of books
that is designed to serve this purpose for the field of mathematical analysis. This
project has brought forth a series of books consisting of three volumes on real anal-
ysis, functional analysis, and applied functional analysis. This multi-volume series
is intended to fulfill these guidelines and meet the expectations of students. It also
aims at introducing this subject to the reader as a bridge from classical analysis to
advanced analysis, wherein this aforementioned conceptually widening gap can be
closed by providing a detailed explanation and clear exposition of the ideas and main
results with a simple treatment. We present in this series the standard material that
should be pursued and taught at this level of courses. The purpose of this series is
to provide an excellent first step towards these goals and prepares the student for
the next study stage. Efforts have been made to make each volume a self-study and
reader-friendly text to allow the student to get acquainted with the basics of advanced
analysis and provide the basis for further studies that can be carried more deeply.
Preface vii
The present book is the first volume of the group of three books and is devoted to the
area of real analysis. It consists of five chapters, with eight sections in each chapter.
Chapter 1 introduces measure theory and the notions of inner and measure of sets,
and discusses Lebesgue’s measurability.
Chapter 2 deals with measurable functions as a generalization of continuous func-
tions by defining them in terms of measurable sets instead of intervals. It also covers
the sequences of measurable functions and their convergence and establishes approx-
imation theorems. Then we study some classes of measurable functions that will be
carried forward in Chap. 3.
Chapter 3 introduces Lebesgue’s theory of integration as a replacement for
Riemann’s theory of integration. The theory is based on measurable sets rather
than intervals. The integral of a bounded function is defined, then the concept is
extended to the integration of a general function. Convergence theorems are discussed
for Lebesgue’s integral. A comparison of the integrability of functions in terms of
Riemann’s and Lebesgue’s theories is discussed. Then the concept of double integrals
is extended to Lebesgue’s integral.
Chapter 4 introduces the infinite-dimensional spaces with a rigorous treatment.
These are normed vector spaces endowed with suitable norms. We investigate these
spaces discussing some classes of them, in particular the Lebesgue spaces. We estab-
lish some fundamental inequalities and obtain some convergence and approximation
results. Finally, we introduce linear operators and functionals on Lebesgue spaces,
followed by a discussion on the idea of representing members of conjugate Lebesgue
spaces, and the Riesz representation theorem.
Chapter 5 introduces measure theory and integration in more general settings. In
particular, we introduce abstract measure theory and general measure spaces other
than the Euclidean spaces, and with measures other than the Lebesgue measure.
The signed measure and decomposition of measures are introduced, and the Radon–
Nikodym theorem is presented and discussed thoroughly.
This textbook can be used as a reference for graduate courses in real analysis.
The book provides more than 210 problems distributed at the end of each chapter.
The questions aim to test whether the reader has absorbed the topic’s basics and
gained a complete understanding. Therefore, the levels of the questions range from
straightforward to challenging, and some of them are standard in their subjects that
can be found in many other books. Moreover, these questions help students prepare
for the Ph.D. comprehensive exams. It should be noted that some questions are
answered in detail, while others are left with some hints or key answers. Similarly,
some details in the proofs throughout the book are left to the reader to complete
independently.
It is important to know that mastering all the topics in this book is not easily
possible without the time and effort needed to overcome all the difficulties that may
arise in terms of grasping the material and mastering the subject. This will prepare
students to think independently and test their maturity and background to accomplish
viii Preface
the task by filling in the gaps and unveiling all the hidden details whenever needed—a
common practice when one enrolls in a graduate program and inevitable when one
starts independent research.
I have to start thanking the God most of all, who gave me strength, health, knowledge,
and most importantly, patience to endure and complete this work successfully.
Writing a book is more exhausting than one can imagine and more gratifying than
one might expect. Before being rewarded, it is important in this regard to recall our
teachers. I am immensely thankful to everyone who taught me mathematics at various
stages of my life from childhood to the PhD level. Words cannot express gratitude to
them. First and foremost, I’m forever thankful and grateful to Prof. Kirk Lancaster
for being a great teacher, a PhD supervisor, a collaborator, and more importantly,
an inspirer. This is a moment where I need to express my sincere gratitude and pay
tribute to him. He is internationally recognized for his work on the analysis and
geometry of elliptic partial differential equations. He is also well-known for proving
the famous Concus–Finn conjecture. In 1996, together with Prof. David Siegel, they
established the existence of the central fans phenomena, which is a cornerstone in
understanding the behavior of capillary surfaces near convex corners. Recently, he,
together with myself, were successful in investigating connections between geometry
of hypersurfaces and generalized solutions of elliptic PDEs in higher dimensions.
Further, I would also like to express my deepest appreciation to all professors who
taught me mathematics in the past. In particular, I’m indebted to the following (in
chronological order starting from 1986): H.I. Karakas, Mashhoor Al-Refai, Bahaettin
Cengiz (passed away), Fuad Kittaneh, Roshdi Khalil, Hasan Al-Ezeh (passed away),
Yuri Ledyaev (the author of the book: Nonsmooth Analysis and Control Theory),
John Martino, John Petrovic (the author of the book: Advanced Calculus: Theory and
Practice), Qiji Jim Zhu (the author of the book: Techniques of Variational Analysis),
Alan Elcrat (passed away, the author of the book: Theory and Applications of Partial
Differential Equations), Alexander Bukhgeym (the author of the book: Introduction
to the Theory of Inverse Problems), Victor Isakov (passed away, the author of the
book: Inverse Problems for Partial Differential Equations, and the series editor of
the book: Sobolev Spaces in Mathematics). I’m fortunate to have had the chance to
be a student of those great mathematicians who influenced me the most.
I would like to express my sincere thanks to the staff at Springer who helped me
in publishing the book to fruition. In particular, I’m grateful and thankful to Shamim
ix
x Acknowledgments
Ahmad for his cooperation and being tremendously helpful and patient in dealing
with this project professionally from the very beginning to the end. My sincere thanks
extend to the wonderful production team for their amazing job in producing the book.
I also wish to express my heartfelt thanks to my lovely colleagues at the Depart-
ment of Mathematics in Prince Sultan University for their continuing encouragement
and support.
A special debt of gratitude is owed to my wife and family for their emotional and
gracious support. I deeply acknowledge the love and encouragement that I received
from them during the preparation of this book.
Lastly, I want to thank everyone who supported me with a positive word or feeling.
I heard it all .. and it meant quite something!
Contents
1 Measure Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Notion of Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Extended Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.3 Axiom of Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Outer Measure of a Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Definition of Outer Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Basic Properties of Outer Measure . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Additive Property of Outer Measure . . . . . . . . . . . . . . . . . . . . 6
1.3 Inner Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Definition of Inner Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Properties of Inner Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.3 The Inner Measure Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.4 Inner Measure As a Supremum . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Lebesgue Measurability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.1 Measurable Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.2 Characterization of Measurable Sets . . . . . . . . . . . . . . . . . . . . 14
1.4.3 Measurable Sets as Sigma-Algebra . . . . . . . . . . . . . . . . . . . . . 17
1.4.4 Borel Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4.5 Additivity of Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4.6 Continuity of Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . 21
1.5 Caratheodory Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.1 The Idea of Caratheodory’s Approach . . . . . . . . . . . . . . . . . . . 22
1.5.2 Characterization of Measurable Sets . . . . . . . . . . . . . . . . . . . . 23
1.5.3 Lebesgue Versus Caratheodory Approaches . . . . . . . . . . . . . . 24
1.5.4 Sigma-Algebra of Measurable Sets By Caratheodory
Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.5.5 Additivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6 Unusual Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
xi
xii Contents
3 Lebesgue Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1 The Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1.2 Riemann Integral of a Bounded Function . . . . . . . . . . . . . . . . 81
3.1.3 Deficiencies of Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . 83
3.2 Integral of Bounded Measurable Functions . . . . . . . . . . . . . . . . . . . . . 85
3.2.1 The Idea of Lebesgue Integral . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.2.2 Integral of Simple Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.2.3 Integral of Bounded Functions . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.3 Integral of General Measurable Functions . . . . . . . . . . . . . . . . . . . . . . 93
3.3.1 The Indeterminate Value Problem . . . . . . . . . . . . . . . . . . . . . . 93
3.3.2 Integral of Nonnegative Functions . . . . . . . . . . . . . . . . . . . . . . 94
3.3.3 The General Lebesgue Integral . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.4 Convergence Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4.1 Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4.2 Bounded Convergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . 101
3.4.3 Monotone Convergence Theorem . . . . . . . . . . . . . . . . . . . . . . 102
3.4.4 Fatou’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.4.5 Dominated Convergence Theorem . . . . . . . . . . . . . . . . . . . . . . 105
3.4.6 Historical Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5 Lebesgue Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5.1 Riemann Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5.2 Improper Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.5.3 Lebesgue Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.5.4 Improper Riemann Integral Versus Lebesgue Integral . . . . . . 110
3.5.5 Riemann–Lebesgue Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.6 Lebesgue Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . 116
3.6.1 The Recovering Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.6.2 Integrability of The Derivative . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.6.3 Differentiation of Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.6.4 Indefinite Integral of Derivative . . . . . . . . . . . . . . . . . . . . . . . . 122
3.7 Lebesgue Double Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.7.1 The Double Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.7.2 Sections of Sets and Functions . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.7.3 Fubini–Tonelli Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.7.4 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4 Lebesgue Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.1 Norms and Linear Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.1.1 Finite-Dimensional Linear Spaces . . . . . . . . . . . . . . . . . . . . . . 139
xiv Contents
Ammar Khanfer earned his Ph.D. from Wichita State University, USA. His area of
interest is analysis and partial differential equations (PDEs), focusing on the interface
and links between elliptic PDEs and hypergeometry. He has notably contributed to
the field by providing prototypes studying the behavior of generalized solutions of
elliptic PDEs in higher dimensions in connection to the behavior of hypersurfaces
near nonsmooth boundaries. He also works on the qualitative theory of differential
equations, and in the area of inverse problems of mathematical physics. He has
published articles of high quality in reputable journals.
Ammar taught at several universities in the USA: Western Michigan University,
Wichita State University, and Southwestern College in Winfield. He was a member
of the Academy of Inquiry Based Learning (AIBL) in the USA. During the period
2008–2014, he participated in AIBL workshops and conferences on effective teaching
methodologies and strategies of creative thinking, which made an impact on his
engaging and motivational writing style. He then moved to Saudi Arabia to teach
at Imam Mohammad Ibn Saud Islamic University, where he taught and supervised
undergraduate and graduate students of mathematics. Furthermore, he was appointed
as coordinator of the PhD program establishment committee in the department of
mathematics. In 2020, he moved to Prince Sultan University in Riyadh, and has been
teaching there since then.
xvii
Chapter 1
Measure Theory
1.1.1 Motivation
There is a motivation to generalize the notion of length to another concept that could
capture length in a more general setting. In principle, the measure of a set A ⊆ R
should refer to the size of A, and it should agree with the natural properties of length if
the settings were reduced to length. Recall that a bounded set is a set that is contained
in a finite interval I, so it has an upper bound and a lower bound. Intuitively, the
measure of an interval should be nothing but its length. However, not all sets in R
are intervals, so two questions arise in this regard:
1. Does every set of R have a “sensible” size? In other words, can every subset of
R be measured somehow by a certain number?
2. If yes, then how to measure it? What tool should we use to represent it?
We may need at the beginning to classify the subsets of R into admissible sets, which
are sets that can be measured, and inadmissible sets, which are sets that cannot be
measured. Keep in mind that there is no information yet on these classes and how big
is each class, or even if one of them is empty. We need to develop a mathematical tool
that enables us to measure the size of subsets, and this tool will allow us to determine
if all sets are admissible (i.e., can be measured by that tool) or not.
Another observation led by intuition is that the measure of a set under the same
measuring tool must be unique since it represents the size of the set according to a
specific type of measure. This indicates that the measuring tool that we are consid-
ering should be a real-valued function, and it will be called a measur e, and will be
denoted by μ. The domain of μ is the collection Σ of all admissible sets of R, which
is indeed contained in the collection P(R) of all subsets of R (which is called power
set). Later, we will investigate this collection and see how big it is and if Σ = P(R)
or Σ P(R). Meanwhile, we will be dealing with Σ as a collection of subsets on
the understanding that it consists of all admissible sets that can be measured by μ.
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 1
A. Khanfer, Measure Theory and Integration,
https://doi.org/10.1007/978-981-99-2882-8_1
2 1 Measure Theory
In R, the two elements ±∞ play the role of a number in the following sense:
1. For any number x ∈ R, x + ∞ = ∞ and x − ∞ = −∞.
2. x · ∞ = ∞ for x ∈ R+ and x · (∞) = −∞ for x ∈ R− .
3. ∞ + ∞ = ∞, −∞ − ∞ = −∞.
4. ∞ · ∞ = ∞ and (−∞)∞ = −∞.
The new extended system of real numbers simplifies the description of the behavior
of functions. Under the new system, every subset of R has an upper bound and a
lower bound, and we can say for instance that lim x 2 = ∞, and lim+ ln x = −∞.
x→∞ x→0
μ : Σ → [0, ∞]
such that
∞
In general, for a collection of pairwise disjoint sets {Ai }i=1 we have:
∞
∞
μ( Ai ) = μ(Ai ).
i i=1
1.2 Outer Measure of a Set 3
The above axioms are necessary to hold for any measure on the set of real numbers,
and any set function μ on (R, Σ) satisfies the four axioms above is called a measure.
These axioms represent the natural properties that any measure μ should intuitively
satisfy, regardless of the size of the admissible sets in Σ measured by μ. The third
axiom can be immediately deduced from the fourth axiom as
B = A ∪ (B \ A),
so some authors ignore this axiom. We, however, prefer to mention it because it
implies two important facts: If a set in A ∈ Σ is contained in a bounded interval,
then μ(A) < ∞, and if A contains an unbounded interval, then μ(A) = ∞. As said
earlier, the collection Σ may be as big as the power set P(R), and may be strictly
smaller. Later, we will see that the fourth axiom makes it hard for Σ to be as big as
P(R).
There are several approaches to define a measure, but we will confine ourselves
to Lebesgue’s approach. There are two ways to measure a set using Lebesgue’s
approach: from outside the set and from inside. We will call the former: “outer
measure” and will be denoted by μ∗ (A), and we will call the latter: “inner measure”
and will be denoted by μ∗ (A). The main idea is to approximate the measure of a
set by other standard sets whose size is known to us, and use them as a device to
compute the measure of other sets. In Lebesgue’s approach, the interval is the device
to measure sets from outside, which is a good choice since we already know the
measure (or the length) of an interval. However, we need to keep in mind that μ
should define a nonnegative function, and each set in the domain of μ should have
a unique measure μ(A). The inner measure μ∗ (A) must be taken with extra care
since we have no idea of the set’s structure to be measured, and if the device to be
used is the interval, we do not know if our set contains intervals. So it is safer at the
beginning to use the outer measure to get a clue of what is going on.
given by
∗
μ (A) = inf (In ) , In ⊇ A,
In
4 1 Measure Theory
where (In ) is the length of the interval In , and the infimum is taken over all possible
collections of {In } that cover A.
The definition implies that for every > 0 there exists a collection {In } such that
(In ) < μ∗ (A) + .
Note that if a set A consists of disjoint subsets, we cannot obtain μ∗ (A) by simply
adding the measure of each subset using the additive property indicated in axiom
4 above. This observation justifies the importance of the additivity property and
its validity to guarantee correct measuring for all sets since, otherwise, we can’t
measure all sets. We will see, however, that this property is, in fact, the source of
many problems and difficulties the measure theory encountered and still encountering
since it emerged at the beginning of the twentieth century.
2
be a cover of {a}, then {In } = for all n. Taking the infimum, we obtain: μ∗ ({a}) =
n
0. Let A = {an }∞
n=1 , and consider the cover
I n = an − n+1
, an + n+1
.
2 2
Then
1.2 Outer Measure of a Set 5
∞
(In ) = = .
n
2n
μ∗ (I ) ≤ b − a + 2.
Let {In }∞
n=1 be collection of open intervals that cover [a, b], which is compact, so by
Heine–Borel Theorem, there exists a finite set of intervals I1 , I2 , . . . , Im such that
m
[a, b] ⊆ In .
n=1
Taking the infimum over all covers In , we obtain the other direction
μ∗ (I ) ≥ b − a.
Let I = [a, ∞), then we can find a closed interval [a, n] ⊂ I for all n. So
Since A ⊆ ∪Ii ,
A + x ⊆ ∪(In + x),
so by (1.2.1) we have
μ∗ (A + x) ≤ (In + x) = (In ) < μ∗ (A) + .
The preceding properties are natural and should be satisfied by any function repre-
senting a measure to demonstrate that it extends the notion of length naturally and
does not conflict with it.
∞
∞
μ∗ ( Ai ) ≤ μ∗ (Ai ).
i=1 i=1
Proof The result holds trivially if μ∗ (Ai ) = ∞ for at least one Ai , so we may assume
that μ∗ (Ai ) < ∞ for all set Ai . For > 0 and for each Ai there exists a cover {Ii,n }
such that
∞
(Ii,n ) < μ∗ (Ai ) + i .
n=1
2
Since
∞
∞
Ai ⊆ Ii,n ,
i=1 i=1
we have
1.2 Outer Measure of a Set 7
∞ ∞
∗
μ Ai ≤ (In )
i=1 i n=1
∗
≤ μ (Ai ) + i
i
2
= μ∗ (Ai ) + .
i
d(A, B) = inf{d(a, b) | a ∈ A, b ∈ B}
= inf{|a − b| | a ∈ A, b ∈ B}.
The property d(A, B) > 0 means the two sets are not only disjoint, but positively
separated. The positively separation property will be used in the following result to
prove additivity of μ∗ .
Proposition 1.2.4 If A and B are two subsets of R such that d(A, B) > 0, then
μ∗ (A ∪ B) = μ∗ (A) + μ∗ (B).
is already established by Proposition 1.2.3. To establish the other direction, let > 0.
Then we can find intervals {Ik } such that A ∪ B ⊆ ∪Ik and
(Ik ) ≤ μ∗ (A ∪ B) + .
Now, let d(A, B) = ρ. Then subdivide the intervals into smaller intervals such that
ρ
the length of each subinterval is no more than . Let {Ik } be the collection of
2
subintervals intersecting with A but not B, and {Ik } be the collection of subintervals
intersecting with B but not A. Then A ⊆ ∪Ik , B ⊆ ∪Ik , and
Hence
μ∗ (A) + μ∗ (B) ≤ (Ik ) + (Ik ) = (Ik ) ≤ μ∗ (A ∪ B) + .
It turns out that the condition d(A, B) > 0 can be used efficiently in proving
additivity property. It is obvious from the definition that if d(A, B) > 0 then A ∩
B = Ø, but the converse is not necessarily true. For example, take A = (0, 1) and
B = (1, 2). It suffices to require that both sets are closed and at least one of them
is compact. Let us take a simple case. Let x ∈ / A where A is closed. Then, d(x, A)
gives the distance between x and the boundary of A, which is indeed positive since
the boundary of A is included in A. In other words, d(x, A) = 0 if and only if x ∈ A
(the closure of A). Now, let x ∈ K where K is compact, and A ∩ K = Ø for some
closed set A. Then, d(x, A) = ρx > 0 by the argument above. Moreover, by the
compactness of K we can choose O1 , O2 , . . . , On covering K with d(xi , A) = ρxi
for each xi ∈ Oi . So we have
d(x, A) = ρ = min{ρxi }, i = 1, . . . , n .
for every x ∈ K , which implies that d(A, K ) > 0. Thus, we have the following fact:
since the other direction is established by Proposition 1.2.4 and the previous Lemma.
Taking into account the inclusion
n ∞
Ai ⊆ Ai ,
i=1 i=1
monotonicity of μ∗ implies
1.3 Inner Measure 9
n
n
∞
∗ ∗ ∗
μ (Ai ) = μ Ai ≤μ Ai .
i=1 i=1 i=1
Proposition 1.2.3 proves subadditivity for general sets not necessarily disjoint. But
∞
what if the sets {Ai }i=1 are pairwise disjoint? Does μ∗ satisfy the countably additive
property in this case? Well, taking into account that countable additivity is one of
the axioms of measure, we need to establish this property. Unfortunately, there is
no direct way to prove it for this measure, but the aid will come from the inner
measure. Let us look at the additivity property from the following simple way: Let
X = A ∪ (X \ A) for some A ⊆ X and μ∗ (X \ A) < ∞. Proposition 1.2.3 implies
that
μ∗ (X ) ≤ μ∗ (A) + μ∗ (X \ A),
that is,
μ∗ (X ) − μ∗ (X \ A) ≤ μ∗ (A). (1.3.1)
This will not lead to any problem since we assume μ∗ (X \ A) < ∞. But μ∗ (X \ A)
is the outer measure of the complement of A under X , so the LHS of (1.3.1) can
be thought of as the measure of A from inside, i.e., the inner measure of A. This
motivates us1 to introduce the following.
Definition 1.3.1 (Inner Measure) Let X ⊂ R with μ∗ (X ) < ∞, and let A ⊆ X .
Then, the inner measure of A, denoted μ∗ (A), is defined as
μ∗ (A) = μ∗ (X ) − μ∗ (X \ A).
1 In fact, this is what motivated Lebesgue in the first place to define the inner measure.
10 1 Measure Theory
1. 0 ≤ μ∗ (A) ≤ μ∗ (A).
2. If B ⊆ A, then μ∗ (B) ≤ μ∗ (A).
3. If A is countable then μ∗ (A) = 0. Consequently, μ∗ (Ø) = 0.
4. If A is a bounded interval, then μ∗ (A) = (A).
5. μ∗ (A) = μ∗ (A + x) for any x ∈ R.
Proof Statement (1) is immediate, and (2) follows from definition since μ∗ (Ac ) ⊆
μ∗ (B c ).
If A is countable then by (1) we have
0 ≤ μ∗ (B) ≤ μ∗ (B) = 0.
If the structure of the set is too bad in such a way that the inner measure differs from
the outer measure, then the set will have two various measures, which is absurd. In
this case, we shall say that this set cannot be measured (or nonmeasurable) in the
sense of Lebesgue. If it happens that μ∗ (A) = μ∗ (A), then we say that the set A has
a unique Lebesgue measure
We need to show in this case that μ is a measure. Another point to consider is that the
definition of inner measure does not generally work for unbounded sets. So we need
to restrict our attention to bounded sets, then we use it to define measurability for
unbounded sets. Now, let us see what the measurability of the bounded set means. If
μ∗ (A) = μ∗ (A), then
μ∗ (A) = μ∗ (X ) − μ∗ (X \ A)
1.3 Inner Measure 11
for X = A ∪ (X \ A), or
μ∗ (X \ A) = μ∗ (X ) − μ∗ (A). (1.3.2)
The reader should realize now that we have some troubles in using the definition of
inner measure. In view of (1.3.2), saying that the inner measure of a set agrees with
its outer measure means countable additive property of outer measure is valid for
subsets and their complements if both are inside a given set. This, however, might
lead to a problem, as the availability of the superset X makes it hard to prove the
measurability of sets. It is not clear if the complement set of a measurable set is
measurable or not. If X is chosen to contain A, then Ac = R \ A is unbounded and
the outer measure of any set containing Ac is infinite. We can get around this way by
two approaches: 1. Generalizing (1.3.2). 2. Formulating another equivalent definition
for the inner measure. We postpone the second approach until Sect. 1.5. Let us now
explain the first approach.
Let E be a bounded set and let A be a set of finite outer measure such that E ⊆ A.
Then
μ∗ (E) = μ∗ (A) − μ∗ (A ∩ E c )
= μ∗ (A) − inf c μ∗ (A ∩ O).
O⊇E
(− inf(X )) = sup(−X ).
Then
μ∗ (E) = sup (μ∗ (A) − μ∗ (A ∩ F c )) = sup μ∗ (F). (1.3.3)
F⊆E F⊆E
One can use a similar argument for the reverse direction (verify). The existence of
the open set O ⊇ E c guarantees the existence of a closed set contained in E, and
singleton sets and the empty set are closed sets; therefore, we can talk about the
supremum of the closed sets contained in E.
We will adopt this approach to formulate the following form of inner measure.
Definition 1.3.3 (Inner Measure of a Set) The inner measure of subsets of R, denoted
μ∗ , is a nonnegative function
defined by
μ∗ (A) = sup{μ∗ (F) : for all closed sets F ⊆ A}.
While the definition may appear to have some advantages, it has two drawbacks:
1. The definition only serves bounded sets A, but not any arbitrary set in R. This is
because compact sets are bounded in R.
2. For bounded sets, all closed subsets are compact, so replacing a closed set F
with a compact set K is valid, but considering the outer measure of this set is not
valid because, from (1.3.3), this would imply that μ∗ (K ) = μ∗ (K ), which means
that compact sets by definition have the same inner measure and outer measure.
However, the definition should not automatically grant this ultimate result because
the main purpose of defining a set’s inner measure is to allow characterizing sets
that have one measure.
These arguments led some mathematicians to prefer Definition 1.3.1, and oth-
ers to search for alternative measurability options other than this approach (e.g.,
Caratheodory definition which will be discussed in Sect. 1.5).
Note that one can also use Definition 1.3.3 to prove Proposition 1.3.2(4) in a more
general setting.
μ∗ (F) ≤ μ∗ (F) ≤ b − a.
μ∗ (J ) ≤ b − a.
On the other hand, for any > 0 the set A = [a + , b − ] is closed, and
2 2
(A) = b − a − ≤ μ∗ (J )
μ∗ (J ) = b − a = (J ),
μ∗ (Jn ) = n − a ≤ μ∗ (J ),
In the preceding two sections, we defined the measure of a set from outside and
from inside. Combining the two concepts, we are ready to define measurability in
the sense of “Lebesgue”.
Definition 1.4.1 (Lebesgue Measurability) Let A be a bounded set in R, (i.e.,
μ∗ (A) < ∞). If μ∗ (A) = μ∗ (A), then we say that A is measurable, and that its
measure is
μ(A) = μ∗ (A) = μ∗ (A).
which implies that A is measurable provided that A ∩ Ik is measurable for each k and
the countable union of measurable sets is measurable. It turns out that the treatment
of measure for unbounded sets will not be complete until we determine whether the
countable union of measurable sets is measurable. In lights of Definition 1.4.1, the
following central questions arise:
1. What type of examples are there for measurable sets?
2. How large is the collection of measurable sets?
3. If μ∗ = μ∗ , does this define a measure satisfying countable additive property?
14 1 Measure Theory
Proof (1) follows from Proposition 1.3.2(1) and Definition 1.4.1. Proof of (2) follows
from propositions: 1.2.2(3) and 1.3.4, noting that for any bounded interval I, the set
I ∩ [a, ∞) will either be a singleton, I, a subinterval of I, or the empty set. All of
them are measurable sets, hence by Definition 1.4.1 [a, ∞) is measurable. Similarly,
we can prove all unbounded intervals are measurable.
In view of Proposition 1.4.2, we see that all types of intervals and all finite and
countable sets, in addition to the empty set, are measurable sets. Is this all? To be
able to characterize measurable sets efficiently we need the following important
result.
Proposition 1.4.3 (Approximation Criterion) Let E ⊆ R. Then:
1. E is measurable if and only if for every > 0, there exists an open set O ⊇ E
such that μ∗ (O \ E) < .
2. E is measurable if and only if for every > 0, there exists a closed set F ⊆ E
such that μ∗ (E \ F) < .
Proof The proof of (1) is split into two cases: for the case when E is bounded and
when E is unbounded. Let E be a bounded measurable set, so μ∗ (E) < ∞. For
> 0, there exists a collection of intervals {In } such that E ⊆ In and
n
(In ) < μ∗ (E) + .
Let O = In then O is an open set and
n
μ∗ (O) ≤ (In ) < μ∗ (E) + .
1
μ∗ (G \ E) ≤ μ∗ (On \ E) < .
n
Letting n → ∞ gives
μ∗ (G \ E) = 0.
Since E ⊆ G, we write G = (G \ E) ∪ E. So
Hence
μ∗ (E) = μ∗ (G). (1.4.1)
1
μ∗ (On \ E) < .
n
Define G = On , and using the same argument in the first case we have
μ∗ (G \ E) = 0.
μ∗ ((G ∩ I ) \ E ∩ I )) = μ∗ (G \ E) = 0.
μ∗ (G ∩ I )) ≤ μ∗ (E ∩ I ) + μ∗ ((G ∩ I ) \ E ∩ I )) = μ∗ (E ∩ I ).
Hence
μ∗ (E ∩ I ) = μ∗ (G ∩ I ). (1.4.3)
μ∗ (E ∩ I ) = μ∗ (G ∩ I )) − μ∗ ((G ∩ I ) \ E ∩ I )) = μ∗ (G ∩ I )).
Hence
μ∗ (E ∩ I ) = μ∗ (E ∩ I ). (1.4.4)
For the reverse direction, let E be bounded set. Then for each n, there exists a closed
set Fn ⊆ E such that
1
μ∗ (E \ Fn ) < .
n
Define K = Fn . Then using similar argument as above,
μ∗ (E \ K ) = 0.
μ∗ (K ) = μ∗ (E) − μ∗ (E \ K ) = μ∗ (E).
It follows that
μ∗ (E) ≤ μ∗ (K ) ≤ μ∗ (E),
Proof Let {E i } be a collection of measurable sets, and let > 0. Then for every E i
there exists Oi ⊇ E i such that
μ∗ (Oi \ E i ) ≤ .
2i
Let E = E i . Then
E⊆O= Oi
By monotonicity,
μ∗ (O \ E) ≤ μ∗ ( (Oi \ E i )) ≤ μ∗ (Oi \ E i ) = = .
2i
18 1 Measure Theory
1
μ∗ (On \ E) < .
n
Then Onc = Fn is closed set, and E c \ Fn = On \ E, and so
1
μ∗ (E c \ Fn ) < ,
n
Let F = Fn , then by previous argument it is easy to see that
μ∗ (E c \ F) = 0.
μ∗ (E c ∩ I \ (F ∩ I )) = 0.
μ∗ (F ∩ I )) = μ∗ (E c ∩ I ). (1.4.6)
μ∗ (F ∩ I )) = μ∗ (E c ∩ I ) − μ∗ (E c ∩ I \ (F ∩ I ) = μ∗ (E c ∩ I ). (1.4.7)
μ∗ (F ∩ I )) = μ∗ (F ∩ I )),
to conclude that the countable intersection of bounded measurable sets and differ-
ences between bounded measurable sets are measurable.
One of the immediate consequences of the preceding theorem is that once open
sets are proven to be measurable, we can immediately conclude that closed sets
1.4 Lebesgue Measurability 19
(and hence all clopen sets) are measurable as well. Nevertheless, Proposition 1.4.3
automatically implies the measurability of open sets and closed sets. It characterizes
measurable sets by approximating them to open and closed sets, which means that
open sets and closed sets are themselves measurable sets. Hence, Proposition 1.4.3
and Theorem 1.4.4 lead to the following essential corollary:
Corollary 1.4.5 Open sets, closed sets, and all countable intersections of open sets
and countable unions of closed sets are measurable.
In light of the preceding results, all countable unions and intersections of intervals and
measurable sets and their complements and differences are measurable, in addition
to the set of irrational numbers and even the whole set R. The collection of Lebesgue
measurable subsets of R forms a special type of algebra of sets, known as: “σ -
algebra”.
Borel σ -algebra is the smallest σ -algebra containing open sets. The intersection
of open sets is a Borel set and is denoted by G δ , and the union of closed sets is
a Borel set and is denoted by Fσ . Here, the symbols σ and δ refer to unions and
intersections, respectively, i.e., a set in G δσ can be written as a countable union of
sets in G δ . Similarly, a set in Fσ δ can be written as a countable intersection of sets
in Fσ . In lights of the previous definition and all preceding results, we can say that
the Lebesgue measurable sets forms σ -algebra of R, and we have the following
F ⊂ Fσ ⊂ Fσ δ ⊂ Fσ δσ ⊂ · · · ,
G ⊂ G δ ⊂ G δσ ⊂ G δσ δ ⊂ · · · .
∞
1 1
{a} = a − ,a + .
n=1
n n
Hence {a} ∈ B, i.e., a Borel set, and hence the set of rational numbers is a Borel set.
Moreover,
[a, b) = {a} ∪ (a, b),
and
∞
1 1
[a, b] = a − ,a + .
n=1
n n
Hence all closed intervals and half-open intervals are Borel sets.
The tools are handy now to answer the third question: Does μ = μ∗ = μ∗ define a
measure on R? The only axiom remaining is the countable additivity. We first prove
the following result.
Proposition 1.4.7 Let {An }∞
n=1 be a collection of subsets of R that are mutually
disjoint. Then
∞ ∞
μ∗ (An ) ≤ μ∗ ( An ).
n=1 n
Proof The result is trivial if μ∗ (An ) = ∞ for some n. Hence we may assume
μ∗ (An ) < ∞ for every n ∈ N. For > 0 and each n, there exists compact set K n
such that K n ⊆ An and
μ(K n ) ≥ μ∗ (An ) − n .
2
But since An are disjoint, so are K n and d(K i , K j ) > 0 for each i, j ∈ N. Fix a
number m ∈ N. Using monotonicity, Proposition 1.2.6, and Corollary 1.4.5 (since
compact sets are closed) we obtain
m
m m
m
μ∗ An ≥μ Kn = μ(K n ) ≥ μ∗ (An ) − .
n=1 n=1 n=1 n=1
∞ ∞
μ Ei = μ(E i ).
i=1 i=1
Proof By Proposition 1.4.7, Proposition 1.3.2 (1), and Proposition 1.2.3, we have
∞
∞
∞ ∞
μ∗ (E i ) ≤ μ∗ Ei ≤ μ∗ Ei ≤ μ∗ (E i ).
i=1 i=1 i=1 i=1
μ∗ (E i ) = μ∗ (E i ) = μ(E i ).
The following theorem discusses the continuity of the measure function in the fol-
lowing sense:
Theorem 1.4.9 Let {E i } be a collection of measurable sets in R. Then we have the
following:
1. Continuity of measure from below: If {E i } is ascending (i.e., E n ⊆ E n+1 for
all n), then
∞
μ( E i ) = lim μ(E i ).
i
Proof Let {E i } be ascending, i.e., E n ⊆ E n+1 . Note that the sets E i+1 \ E i are dis-
joint, hence
∞
∞
μ( E i ) = μ(E 1 ( E i+1 \ E i )
i=1 i=1
∞
= μ(E 1 ) + μ(E i+1 \ E i )
i=1
k
= lim [μ(E 1 ) + μ(E i+1 ) − μ(E i )]
k→∞
i=1
= lim μ(E k ),
k→∞
22 1 Measure Theory
but
lim μ(E 1 \ E i ) = lim μ(E 1 ) − μ(E i ) = μ(E 1 ) − lim μ(E i ). (1.4.9)
Earlier in the chapter, we illustrated two approaches to get around the problem of
(1.3.2). Lebesgue measurability was explained as the first approach, and now it is
time to explain the second approach. Let E be a given measurable set. Let I be
bounded interval; then E ∩ I is measurable, i.e.,
μ∗ (E ∩ I ) = μ∗ (E ∩ I ) = μ∗ (I ) − μ∗ (I ∩ E c ),
or we can write it as
μ∗ (I ) = μ∗ (E ∩ I ) + μ∗ (I ∩ E c ). (1.5.1)
In other words, every bounded interval should satisfy (1.5.1). Caratheodory’s idea is
to generalize the treatment from a bounded interval to any set A. It is known that
μ∗ (A) ≤ μ∗ (A ∩ E) + μ∗ (A ∩ E c ). (1.5.2)
1.5 Caratheodory Criterion 23
The following criterion provides another, but equivalent, way of characterizing mea-
surable sets.
Proposition 1.5.1 (Caratheodory Criterion) Let E ⊆ R be a set. Then E is measur-
able iff for any set A we have
μ∗ (A) = μ∗ (A ∩ E) + μ∗ (A ∩ E c ). (1.5.3)
μ∗ (O ∩ E) = μ∗ (O ∩ E).
μ∗ (A) ≤ μ∗ (O ∩ E) + μ∗ (O \ E)
≤ μ∗ (O) − μ∗ (O \ E) + μ∗ (O \ E),
which implies
μ∗ (A) ≤ μ(O).
Now, since O is open and contains A, and knowing that any open set can be written
as a union of countable intervals, we can take the infimum over all open set O ⊇ A
and substitute back in (1.5.4) to obtain the result.
For the reverse direction, assume E satisfies (1.5.3), and let I be any bounded
interval. Then
μ∗ (I ) = μ∗ (I ∩ E) + μ∗ (I \ E),
or
μ∗ (I ) − μ∗ (I \ E) = μ∗ (I ∩ E). (1.5.5)
But from Definition 1.2.1, the LHS of (1.5.5) is μ∗ (I ∩ E), and therefore I ∩ E is
measurable, and since I is arbitrary bounded, measurability follows.
A = (A ∩ E) ∪ (A ∩ E c ),
24 1 Measure Theory
by subadditivity of μ∗ , we have
μ∗ (A) ≤ μ∗ (A ∩ E) + μ∗ (A ∩ E c ).
μ∗ (A) ≥ μ∗ (A ∩ E) + μ∗ (A ∩ E c ).
Under this clever criterion, a measurable set E can split any set A into two disjoint
parts such that the measure of A is the total measure of both parts. This “split”
condition characterizes measurable sets and gives the notion of measurability more
flexibility in dealing with all sets. This condition is named after Caratheodory, who
proposed his definition in 1914. Due to its popularity, many textbooks now introduce
this condition as the definition of measurable sets, and the previous proposition
proves the legitimacy of this adoption since they are equivalent. Both definitions are
common and widespread in the literature, and they both treat the measurability of
sets in the sense of Lebesgue theory. Still, the inner–outer approach was the idea of
Lebesgue, while the splitting approach was the idea of Caratheodory. The former
is more natural and appealing. It also gives credit to Lebesgue, who proposed this
definition, and to appreciate his precursor work. However, the latter is more efficient
and flexible in dealing with measurable sets and studying abstract spaces other than
Euclidean spaces Rn because the topological structure of Euclidean spaces is clear
and well understood and so it becomes simple to determine open and closed sets to
perform outer and inner measures. This may be hard in general abstract spaces, if
impossible, which explains why Caratheodory’s criterion is favorable for researchers
working on probability theory, for instance.
Another advantage of the Caratheodory condition is that it enables us to prove that
the collection Σ of measurable sets is σ -algebra in less work and efforts that were
needed using the inner–outer Lebesgue’s approach. For example, it is immediate from
(1.5.3) that the sets R and Ø are measurable. Moreover, if E ⊆ R is measurable,
then E c is measurable.
The next result will help us in establishing our main result about the collection of all
measurable sets.
1.5 Caratheodory Criterion 25
Proof We will establish the result in two steps. The first step is to prove
n
n
∗
μ A∩ Ek = μ∗ (A ∩ E k ) (1.5.6)
by induction. In the second step, we extend the result to infinity. Equation (1.5.6) is
obviously true for n = 1. Assume the statement is true for n − 1, that is,
n−1
n−1
μ∗ A ∩ Ek = μ∗ (A ∩ E k ). (1.5.7)
n
S = A∩( E k ).
But
n
S ∩ En = A∩ Ek ∩ En = A ∩ En
n
n−1
S ∩ E nc = A∩ Ek ∩ E nc = A ∩ Ek .
n
= μ∗ (A ∩ E k ).
n
n
∞
μ∗ (A ∩ E k ) = μ∗ A ∩ Ek ≤ μ∗ A ∩ Ek .
Taking n → ∞ gives
∞
∞
μ∗ (A ∩ E k ) ≤ μ∗ (A ∩ ( E k )).
It follows that
∞ ∞
μ∗ A ∩ Ek = μ∗ (A ∩ E k ).
We’ll use the preceding lemma to prove the following theorem which establishes the
same result of Theorem 1.4.4, using the Caratheodory condition.
Theorem 1.5.3 All differences, complements, countable unions, and countable
intersections of measurable sets are measurable.
Proof Let E 1 and E 2 be two measurable sets. Let A be arbitrary set in R. First,
we show E 1 \ E 2 is measurable. The idea of the proof is to split A by E 1 then split
A ∩ E 1 by E 2 . So we have
μ∗ (A) = μ∗ (A ∩ E 1 ) + μ∗ (A ∩ E 1c )
= μ∗ (A ∩ E 1c ) + μ∗ (A ∩ E 1 ∩ E 2 ) + μ∗ (A ∩ E 1 ∩ E 2c ).
By subadditivity of μ∗ , we write
μ∗ (A) ≥ μ∗ (A ∩ E 1c ) ∪ (A ∩ E 1 ∩ E 2 ) + μ∗ (A ∩ E 1 ∩ E 2c ). (1.5.9)
(A ∩ E 1c ) ∪ (A ∩ E 1 ∩ E 2 ) = A ∩ (E 1 ∩ E 2c )c .
(A ∩ B) = (A ∩ E 1 ) ∪ (A ∩ (E 2 \ E 1 )),
1.5 Caratheodory Criterion 27
we obtain
μ∗ (A ∩ B) ≤ μ∗ (A ∩ E 1 ) + μ∗ A ∩ (E 2 ∩ E 1c ) (1.5.10)
μ∗ (A ∩ B) + μ∗ A ∩ B c ≤ μ∗ (A ∩ E 1 ) + μ∗ (A ∩ E 1c ). (1.5.11)
But writing A as
A = A ∩ R = A ∩ (B ∪ B c ) = (A ∩ B) ∪ (A ∩ B c )
μ∗ (A) ≤ μ∗ (A ∩ (E 1 ∪ E 2 )) + μ∗ (A ∩ (E 1 ∪ E 2 )c ). (1.5.12)
Equations (1.5.11) and (1.5.12) imply that E 1 ∪ E 2 , hence by induction and the fact
that
n
n−1
Ei = Ei ∪ En
i=1 i=1
n
we conclude that i=1 E i is measurable.
Now, we prove the result for a countable union. Construct the following sequence
of sets:
n−1
S1 = E 1 & Sn = E n \ Ei .
i=1
Then it is easy to see that {Sn } is a collection of mutually disjoint sets with the fact
that Sn ⊆ E n and
∞ ∞
Si = E i = E,
i=1 i=1
c
n
n
∗ ∗ ∗
μ (A) ≥ μ A∩ Si +μ A∩ Si
i=1 i=1
n
≥ μ∗ A∩ Si + μ∗ (A ∩ E c ).
i=1
Using Lemma 1.5.2, and taking the limit of the sum as n → ∞, then using Lemma
1.5.2 again, we can extend the above inequality to
∞
μ∗ (A) ≥ μ∗ (A ∩ Si ) + μ∗ (A ∩ E c )
i=1
= μ∗ (A ∩ E) + μ∗ (A ∩ E c ) .
∞
Therefore, we conclude that E = i=1 E i is measurable.
Now, intersections can be deduced directly from unions and complements.
The fact that the collection of all measurable sets forms σ -algebra can be concluded
now. One can begin with proving that all intervals of the form (−∞, a), (−∞, a],
(a, ∞), and [a, ∞) are measurable (see Problems 24, 25), then using intersections
to show that all intervals are measurable, then the previous theorem to show that all
Borel sets are measurable as predicted in Sect. 1.4.
1.5.5 Additivity
Let A = R in Lemma 1.5.2. Then from the hypothesis and the previous the-
Proof
orem, ∞ E k is also measurable; hence the result follows.
We’ve seen that the collection of measurable sets contains a multiple type of sets,
such as finite sets, countable sets, open sets, closed sets, intersections of open sets,
1.6 Unusual Sets 29
unions of closed sets, and any combinations of unions and intersections of these
sets, in addition to the empty set (which can be expressed as the intersection of two
disjoint open sets). In fact, it is hard (mostly impossible) to visualize a nonmeasurable
set, and this gives birth to the fourth question: “Does there exist such sets that are
nonmeasurable in Lebesgue sense?” In other words, does Σ equals P(R), the power
set of R?
In 1905, B. Vitali, an Italian mathematician, proposed a set that cannot be measured
by Lebesgue measure. Before we start defining the set, it is well known that a coset
of Q in R is any set of the form
x + Q ={x + q : q ∈ Q, x ∈ R}.
{x + q : q ∈ Q, x ∈ [0, 1]}.
It is easy to see that the collection of cosets of Q in [0, 1] form a partition of [0, 1].
That is, if x − y ∈ Q then
x + Q =y + Q
and if x − y ∈
/ Q then
(x + Q) ∩ (y + Q) = Ø.
— Selittäkää sananne.
— Hellé!
— Vaikenemalla teitte itsenne myötärikolliseksi panetteluun. Tekö
menettelitte niinkuin hyvinkasvatetun miehen tulee? Olisin
mieluummin nähnyt, että olisitte menetellyt niinkuin mies, vaikkakin
se olisi sotinut sovinnaisuutta, varovaisuutta ja etujanne vastaan.
— En pelkää totuutta.
— No siis?
— Mitä tarkoitatte?
— Se on mahdotonta.
— Minä?
Hän kertasi:
Minä sopersin:
— Ulos!
Talvi, joka näillä seuduin oli leuto, läheni loppuaan. Istuin eräässä
huoneessa ensimmäisessä kerroksessa vaalenneilla
musliiniuutimilla verhotun ikkunan ääressä ja katselin, miten
taivaanranta peittyi sateen harsoon tai sumuun. Puutarhassa ei ollut
kukkia ja ainoastaan puksipuut, muratit ja punakuusamat säilyttivät
vielä tumman vehreytensä, surullisina ja vakavina kuin haudat, joita
ne koristavat. Väliin kun sade lakkasi, saatoin istua tuntikausia
liikkumatta, puhumatta, tuijottaen taivaan vaihtelevain valaistusten
värittämiin maisemiin. Edessäni ei ollut enää kesän loistavaa
väriasteikkoa: vivahdukset olivat hienompia, käsittäen kaikki
harmaan, sinipunaisen, sinisen herkät yhtymät, jotka sulivat utuisen
ilman vienoon valohämytaustaan. Tämän laajan näyttämön etualalla
muodostivat viljellyt vainiot eloisia, likaisen ruskeita, raikkaan ja
kostean vihreitä väritäpliä. Mutta kauneinta maisemassa oli taivas,
— sininen, ikäänkuin maitoon kastettu taivas, jossa utuhaituvat uivat
valkeina ja pehmeinä kuin joutsenet, — harmaa taivas, joka vaihteli,
lyijynharmaasta helmenharmaaseen ja sinipunervan harmaasta
hopean harmaaseen, — noiden hallavien, raskaiden sumupilvien
täyttämä taivas, jotka tuulen kiidättäminä lentävät yhdessä
muuttolintujen kanssa, läikehtivä taivas, levoton kuin ihmiselämä.
— Niin.
— Runoilijako?
— Näyttelijätär?
— Kuuluisan Grannis'in?
— En.
— On, rouva.