This action might not be possible to undo. Are you sure you want to continue?
Series in Real Analysis  Volume 9
THEORIES OF INTEGRATION
The Integrals of Riemann, Lebesgue, HenstockKurzweil, and Mcshane
SERIES IN REAL ANALYSIS
VOl. 1: VOl. 2: VOl. 3:
Lectures on the Theory of Integration R Henstock Lanzhou Lectures on Henstock Integration Lee Peng Yee The Theory of the Denjoy Integral & Some Applications V G Celidze & A G Dzvarseisvili translated by P S Bullen Linear Functional Analysis W Orlicz Generalized ODE S Schwabik Uniqueness & Nonuniqueness Criteria in ODE R P Agawa/& V Lakshmikantham HenstockKurzweil Integration: Its Relation to Topological Vector Spaces Jaroslav Kurzweil Integration between the Lebesgue Integral and the HenstockKurzweil Integral: Its Relation to Local Convex Vector Spaces Jaroslav Kurzweil
VOl.
4:
VOl. 5: Vol. 6: VOl. 7: Vol. 8:
Series in Real Analysis  Volume 9
THEORIES OF INTEGRATION
The Integrals of Riemann, Lebesgue, HenstockKurzweil, and Mcshane
Douglas S Kurtz Charles W Swartz
New Mexico State University, USA
1 World Scientific :
N E W JERSEY LONDON SINGAPORE BElJlNG SHANGHAI HONG KONG TAIPEI CHENNAI
Published by
World Scientific Publishing Co. Re. Ltd.
5 Toh Tuck Link, Singapore 596224 USA ofice: Suite 202, 1060 Main Street, River Edge, NJ 07661 U K ofJice: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublicationData A catalogue record for this book is available from the British Library.
Series in Real Analysis Vol. 9 THEORIES OF INTEGRATION The Integrals of Riemann, Lebesgue, HenstockKurzweil,and McShane
Copyright 0 2004 by World Scientific Publishing Co. Re. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN 9812388435
Printed in Singapore by World Scientific Printers (S) Pte Ltd
.To Jessica and Nita. for supporting us during the long haul to bring this book to fruition.
This page intentionally left blank .
Preface This book introduces the reader to a broad collection of integration theories. we show the equivalence of the Riemann and Darboux definitions. The real content of the book begins with a chapter on the Riemann integral. HenstockKurzweil and McShane integrals. We conclude the chapter with a discussion of improper integrals. positivity and the Cauchy criterion. By studying classical problems in integration theory (such as convergence theorems and integration of derivatives). We begin the book with the problem of defining and computing the area of a region in the plane including the computation of the area of the region interior to a circle. including linearity. This leads to a discussion of the approximating sums that will be used throughout the book. We then discuss lattice properties and the Fundamental Theorem of Calculus. except for the sections which compare the various integrals. focusing on the Riemann. We give the definition of the Riemann integral and develop its basic properties. vii . After presenting Darboux’s definition of the integral and proving necessary and sufficient conditions for Darboux integrability. Several of the integrals receive detailed developments. we will follow a historical development to show how new theories of integration were developed to solve problems that earlier integration theories could not handle. The chapters of this book are written so that they may be read independently. We present necessary and sufficient conditions for Riemann integrability in terms of sets with Lebesgue measure 0. There should be sufficient exercises in each chapter to serve as a text. while problems and references directing the reader to future study are included. Lebesgue. This means that individual chapters of the book could be used to cover topics in integration theory in introductory real analysis courses. others are given a less complete discussion in the book.
Finally. the Fundamental Theorem of Calculus in full generality. We establish the basic properties of the McShane integral and discuss absolute integrability. Convergence theorems are used t o motivate the Lebesgue integral and the Fundamental Theorem of Calculus to motivate the HenstockKurzweil integral. We then give an example that points out the failure of the Bounded Convergence Theorem for the Riemann integral. and the Monotone and Dominated Convergence Theorems. In the following chapter.Vlll Theories of Jntegmtion We motivate the development of the Lebesgue and HenstockKurzweil integrals in the next two chapters by pointing out deficiencies in the Riemann integral. Next. obtained by slightly varying the definition of the HenstockKurzweil integral. we discuss versions of the Fundamental Theorem of Calculus for both the Riemann and Lebesgue integrals and give examples showing that the most general form of the Fundamental Theorem of Calculus does not hold for either integral. We show how Lebesgue’s descriptive definition leads in a natural way to the definitions of Lebesgue measure and the Lebesgue integral. We develop basic properties of the HenstockKurzweil integral. and Dominated). which these integrals address. and use this t o motivate Lebesgue’s descriptive definition of the Lebesgue integral. After comparing the HenstockKurzweil integral with the Lebesgue integral. and use the characterization to prove Fubini’s Theorem on the equality of multiple and iterated integrals. the McShane . we develop the basic properties of the Lebesgue integral. We then use the Fundamental Theorem to motivate the definition of the HenstockKurzweil integral. Monotone. We show that there are no improper integrals in the HenstockKurzweil theory. also know as the gauge integral and the generalized Riemann integral. We then show that the McShane integral is equivalent t o the Lebesgue integral and that a function is McShane integrable if and only if it is absolutely HenstockKurzweil integrable. we discuss the “gaugetype” integral of McShane. A discussion of the space of integrable functions concludes with the RieszFischer Theorem. We extend the Lebesgue integral t o ndimensional Euclidean space. including convergence theorems (Bounded. we conclude the chapter with a discussion of the space of HenstockKurzweil integrable functions and HenstockKurzweil integrals in R”. we compare the Riemann and Lebesgue integrals. We begin the discussion of the Lebesgue integral by establishing the standard convergence theorem for the Riemann integral concerning uniformly convergent sequences. Consequently. Following a discussion of Lebesgue measurable functions and the Lebesgue integral. give a characterization of the Lebesgue integral due to Mikusinski.
Preface i x integral could be used to give a presentation of the Lebesgue integral which does not require the development of measure theory. .
This page intentionally left blank .
.3 2. . . . . .1 Integration by parts and substitution . . . . .1 Areas . . . ... .Contents Preface 1. .5. . . . . . . .. . . . . Characterizations of integrability .. . . .. . . . .8 3 . . .6.3 Lattice properties ... . .. .1 2. . . . . . . . ... .2 Exercises vii 1 . Basic properties . . . .4.. . . .. . . 1.. .. 2. .. . . . . . . . . . . . . . . . . . . . . . 2. . .. .. ... . . ... . . Convergence theorems and the Lebesgue integral 3. . . . .. . .. . . . . ...2 Equivalence of the Riemann and Darboux definitions 2. ... . .. . . . ... .4 Riemann’s definition . . . . Exercises ..4. . . . ..5 Additivity of the integral over intervals ... . . . .. ..4. . . . . . . .. . . . . . . . . . . .. 2. .. . .1 Lebesgue measure zero . . . . . . . .. . . . . ...2 2. . 2. . 2. . . . . . . . .. . . . . . . .. ... 2. . . . . .6 2. .1 Necessary and sufficient conditions for Darboux integrability . . . . .. . . . . .. Darboux’s definition . . . .5 2. . . . Riemann integral 2. . . ... . . .. . . . 56 . . .. .. . 1 9 11 11 15 18 20 24 25 27 30 31 33 37 38 41 42 46 53 2 .7 2. .. . . . . . . . . . Cauchy criterion . . . . . .. . . . . . . . . .4 Integrable functions . Improper integrals . . . . . 2. .... . . . .. . . . . Fundamental Theorem of Calculus . . . . .4. . . . .. . . . . . . . .4. . . Introduction 1.1 Lebesgue’s descriptive definition of the integral xi . .
. . . . . .3 The Cantor set . 111 3. . . . . 5. .5 Henstock’s Lemma . . 78 3. . . .10 Exercises . . . . . . .4 Measurable functions . . . . . .10 Characterizations of indefinite integrals .2. . . . . . .2 A General Fundamental Theorem of Calculus . . . . . . . .3 Lebesgue measure in R” . . . . . . . . . . . . . . . . .1 Cauchy Criterion . . .6 Absolute integrability . . . . . . . . . . 125 4 . . . . . . . . . . . . . . . . 4. .7 Convergence theorems . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Measure . . . . . . . . .6. . 122 3. . . . . . . . . . . . . . . . .9 The space of Lebesgue integrable functions . . . . .11 The space of HenstockKurzweil integrable functions . . . . . . . . . . . . . . . . . 96 3. . . . . 4. . . . . .8 Fubini’s Theorem . . . . . 113 3. . . . . 4. . . . . . . . 4. . . . . . . . . . . 60 3. . . . . . . . . . . . . . .9 Differentiating indefinite integrals . . . 4. . . . . . . . . . . . . . . . . . . . 4. .1 Outer measure . . .1 Bounded variation . . . . . .2. . . . . . . . . . . . . . . .xii Theories of Integration 3. . . . . . . . . .1 Denjoy and Perron integrals . . . . . .2. . . . . . . . . . . . . . . . . . . 4.5 Lebesgue integral . . . . . . . . . 4. . . . . . . . 4. . . . . . . . . . . . . .7 Mikusinski’s characterization of the Lebesgue integral . . . .10. . .3. 4. . . . . . . . .1 Derivatives of monotone functions . .3 Basic properties . .13 Exercises . . . . . . . . . 85 3. . . . . . . . . . . . . . . . 4. . . . . . . . . . .6. . . . . . . . . . . . . . 79 3. . . . . . . . . . . 4. . .2 Indefinite Lebesgue integrals . . . .3 Lattice Properties . . . . . 4. . . . . . .8 HenstockKurzweil and Lebesgue integrals . . . . 117 3. . . . . 4. . .1 Functions with integral 0 . . . . . . . . . . . . . . 4. . .2 The integral as a set function . . . . . . . . . .3. 4.2 Absolute integrability and indefinite integrals . . . . . . . . . . . . . Absolute integrability and the McShane integral 133 135 137 145 150 151 154 162 172 172 175 178 180 189 190 195 195 198 203 204 205 206 214 223 . . . .6. . . . . . . . . . . . . .3 Indefinite Riemann integrals . . . . 4. . . . . . . . . . . . . . . . . . . 60 3. . . . . . . . . . . . . . .10. . . . . . . . . . 4. . 4. . . . . .2 Lebesgue Measure . Fundamental Theorem of Calculus and the HenstockKurzweil integral 4. . . . . . .12 HenstockKurzweil integrals on R” . . . . . . . . . . . . . . . .10. . . . . . . . . . . . . 4. . . .9. .4 Unbounded intervals . . . . . . 64 3. .6 Riemann and Lebesgue integrals .
. . .1 Definitions . . . . . . . . . . . . . . . . . 5. . . . . . . .3 Absolute integrability . . . .Contents Xlll 5. 5. . .6 The space of McShane integrable functions . . . . . . . . . . . . . . 5. . . . . . . . . . .7 McShane. . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . .2 Basic properties . . . . . . . HenstockKurzweil and Lebesgue integrals . . .5 The McShane integral as a set function . . . . . . . . . . 5. . . .11 Exercises . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.10 McShane. 5. . . . . . . .9 Fubini and Tonelli Theorems . . . . Bibliography Index 224 227 229 232 234 240 244 245 253 254 257 258 263 265 . . . . . . . HenstockKurzweil and Lebesgue integrals in R" 5. . . . . . . . . 5. 5.8 McShane integrals on R" . . .1 Fundamental Theorem of Calculus . . . . . 5. .4 Convergence theorems . .
This page intentionally left blank .
1 1 . and let P be the point on AB such that the line C P from C to P is perpendicular to AB. Figure 1.Chapter 1 Introduction 1. the sum of their areas is the area of A. using the second axiom. We begin with three axioms for areas: (1) the area of a rectangular region is the product of its length and width. From the first and third axioms. (3) congruent regions have equal areas. Then. Two regions are congruent if one can be converted into the other by a translation and a rotation. and C. suppose that A is a triangle with vertices A. (2) area is an additive function of disjoint regions. Now. by decomposing them into nonoverlapping triangles. it follows that the area of a right triangle is one half of the base times the height. In this way.1 Areas Modern integration theory is the culmination of centuries of refinements and extensions of ideas dating back to the Greeks. It evolved from the ancient problem of calculating the area of a plane figure. B . ACP and BCP are two right triangles and. one can determine the area of irregularly shaped areas. Assume that AB is the longest of the three sides.
the ancient Greeks calculated the area of a circle by approximating the circle by inscribed and circumscribed regular ngons whose areas were easily computed and then found the area of the circle by using the method of exhaustion. from the estimate A . The basic idea is to approximate a general region with simpler geometric regions whose areas are easy to calculate and then use a limiting process to find the area of the original region.u (I22+n+l) < 5 ( A . Consider the area inside of C and outside of I ~ z +This area is comprised of ~. the base inside the circle) and the opposite side touches the circle at one point.u (I22+")). Specifically. We will illustrate the method using modern not a t ion.a ( I 4 ) be the error in approximating A by the area of an inscribed 4gon. This same idea works for a starshaped region. by induction. such as the interior of a circle. Let n be a positive integer. For the pentagon. For example. Archimedes claimed that the area of a circle of radius r is equal to the area of the right triangle with one leg equal to the radius of the circle and the other leg equal to the circumference of the circle. Let C be a circle with radius r and area A . Let cup: be one such cap and let R: be the smallest rectangle that contains cup:. with In inscribed inside of C and On circumscribed outside of C. 22+n congruent caps.Tn be +I i 1 . Note that R7 shares a base with cap7 (that is. For more general regions in the plane. See the figure below. Let u represent the area function and let EI = A .2 Theories of Integration It is easy to see how this procedure would work for certain regularly shaped regions. though in this case. To see this. one connects both the points of the arms of the star and the points where two arms meet to the center of the region. such as a pentagon or a starshaped region. fix n _> 0 and let 122+* be inscribed in C. which is the midpoint of that side and a vertex of I ~ z + ~ Let . a more sophisticated method of computation is required. producing five triangles with disjoint interiors. The key estimate is which follows. any interior point will do). one merely joins each of the five vertices to the center (actually. We let I22+n+1 be the 22+n+1g0nwith vertices comprised of the vertices of I22+n and the 22+n midpoints of arcs between adjacent vertices of I22+n. and let In and On be regular ngons.
2 Suppose that cap. We can carry out a similar.$) = 2 [a (capy+') > 2a (cap. we get as we wished to show.)] . See the picture below. Then. 122+n+1 which implies a (cap?) = a (y) (cap.$. since capy++'Ucap"." and cap::.+: c R? \Ty. analysis with the circumscribed rectangles to prove .Introduction 3 the triangle with the same base and opposite vertex at the midpoint.+l +a u cap.$) + a (cap.+' u cap. Adding the areas in all the caps. but more complicated.! are the two caps inside of C and outside of that are contained in cap?. Figure 1.
By rotational invariance. enough to show that the area of the region bounded by the arc from E t o F and the segments D E and D F is greater than twice the area of the two regions bounded by the arc from E to F and the segments E N . is greater than the length of G H which is equal to the length of E H .a (S) To see this. Consider the lower right hand corner in the picture below. Again. since both are half the . let S' be the region bounded by the arc from E to G and the segments E H and GH and S be the region bounded by the arc from E to G and the segments DG and D E . respectively. More simply. so that 0 2 2 = 0 4 is a square.4 Theories of Integration where E o = a (04) A is the error from approximating A by the area of a circumscribed 4gon. ~ 1 As in the argument above. Note that the segment H J is one side of 0 2 2 . Let G be midpoint of the arc on C from E to F . it is . consider the case n = 0. so that the length of D H .( ~ ( 0 2 2 + . respectively. H J and F J . Figure 1. this estimate follows from the inequality a(O22+n+l) A < . ~ 2 1 For simplicity. We wish to show that a (S') < . where 0 4 and C meet. we may assume that 0 4 sits on one of its sides. which we denote IDHI. note that the triangle D H G is a right triangle with hypotenuse D H .3 Let D be the lower right hand vertex of O4 and let E and F be the points to the left of and above D .)A ) . and let H and J be the points where the tangent to C at G meets the segments D E and D F .
using rectangles instead of triangles to compute the approximating areas.( 2 2 s n ~ h. Suppose first that A > T . we made the tacit assumption that the circle had a notion of area associated with it. we see that a ( 0 2 z + n ) > T. A 5 T . Let s' be the length of the side joining the vertices and let h = T be the distance from this side t o the center.A > 0. We have made no attempt to define the area of a circle or. A . we see that is < T . which is a contradiction. ~ h ) 2 2 < r and 2 2 + n ~ less than the circumference of C. any other arbitrary region in the plane.1) we can choose an n so large that A . Since 2 2 + n ~ 'is greater than the circumference of C.~ = . A = T . Then. indeed.T . Let s be the length of the side joining the vertices and let h be the distance from this side to the center. For example. Then.A < T . The basic idea employed by the ancient Greeks leads in a very natural way to the modern theories of integration. if A < T.Introduction 5 length of a side of 022+l.T > 0.1) and (1. let f be a positive .a ( 1 2 2 + " ) < A . or a ( 0 2 2 + n ) < T. so that by (1. A 2 T. then T. With estimates (1. Then. + 1 1 u ( 1 2 2 + n ) = 2 2 S n (Ti)= 2 2 + n . Thus. or T < a ( 1 2 2 + n ) . We will discuss the problem of defining and computing the area of regions in the plane in Chapter 3.2). In the computation above.A .Consequently.2) follows as above. so that by (1. Thus. we can prove Archimedes claim that A is equal to the area of the right triangle with one leg equal to the radius of the circle and the other leg equal to the circumference of the circle. a ( S )< so that u ( S ) = u ( D H G ) u (S') > 2a (S') .2) we can choose an n so that a ( 0 2 2 + n ) . Let T be one of the 22+n congruent triangles comprising 122+n formed by i joining the center of C to two adjacent vertices of 1 2 2 + n . Call this area T . Let h be the distance from G to DE. Let Ti be one of the 22+n Since h a (122+n) congruent triangles comprising 0 2 2 + n formed by joining the center of C to two adjacent vertices of 0 2 2 + n . Similarly. Then. and the proof of (1. which is a contradiction.
He approximated the area of the region R by the Cauchy sum Cauchy used the value of the function at the left hand endpoint of each subinterval [xil.b].b] into subintervals and use these subintervals for the bases of the rectangles. that is. with xo = a and xn = b. xi] to generate rectangles with area f (xi1)(xi. . The French mathematician AugustinLouis Cauchy (17891857) studied the area of the region R for continuous functions.21. the area of the region R = ((2. .xn}. ordered set of points P = {ZO. t Figure 1. A partition of an interval [a. . The sum of the areas of the rectangles approximate the area of the region R. .6 Theories of Integration function defined on an interval [a. We divide the interval [a. 0 5 y 5 f (x)}. we consider approximating the area of the region R by the sums of the areas of rectangles. Consider the problem of computing the area of the region under the graph of the function f.b] is a finite..xi1). y) : a 5 x 5 b.4 Analogous to the calculation of the area of the circle.
The German mathematician Georg Friedrich Bernhard Riemann (18261866) was the first to consider the case of a general function f and region R. Riemann generated approximating rectangles by choosing an arbitrary point ti. (2) 4. were too restrictive. . however. Pesin [Pel and GrattanGuinness [Gr] for descriptions of Cauchy’s argument). ) have a limit. (See Cauchy [C.a concept not understood at that time. called a sampling point. since actually he assumed that the function was uniformly continuous on the interval [a.5 He then used the intermediate value property of continuous functions to P argue that the Cauchy sums C (f. ) satisfy a “Cauchy condition” as the mesh of the partition. which he defined to be the P integral of f over [a.b] and denoted by Jf f (x)dx. p ( P ) = maxl<isn (xi.b ] . Cauchy’s assumptions. in each subinterval xi^. He concluded that the sums C (f.xi~). xi] and forming the Riemann sum i=l to approximate the area of the region R.approaches 0.Introduction 7 b Figure 1. pages 1221271.
which we discuss in Chapter 3. leads to other. more general integration theories.b] into subsets other than intervals and we must develop a notion that generalizes the concept of length to these sets. The idea is then to multiply this value by the length of the set of points for which f is approximately equal to y. such as the Riemann sums. The construction of the approximating sums in both the Cauchy and Riemann theories is exactly the same.6 I Riemann defined the function f to be integrable if the sums S (f. These considerations led to the notion of Lebesgue measure and the Lebesgue integral. It is this seemingly small change that makes the Riemann integral so much more powerful than the Cauchy integral. have a limit as p ( P ) = rnaxlliSn (zi . It will be seen in subsequent chapters that using approximating sums. We will give a detailed exposition of the Riemann integral in Chapter 2.zi1) approaches 0. is chosen for each subinterval.P .8 Theories of Integration Y = f(X. 1 P fa I i Figure 1. the range of the function f is partitioned instead of the domain.) I “f I I I I I I 1 I I I I I I I . y. This means that we must consider “partitioning” the domain [a. but imposing different conditions on the subintervals or sampling points. or even a union of intervals. In the Lebesgue theory of integration. The problem is that this set of points need not be an interval. but Cauchy associated a single set of sampling points to each partition while Riemann associated an uncountable collection of sets of sampling points. A representative value. .
. the McShane integral is not as general as the HenstockKurzweil integral. 1. however. Take the limits of both these expressions to compute the area inside of C. x i ] . we can decompose either In or On into n congruent.l .b] + R by f (2) = x2 and let P be a partition of [a. but uses a different condition to control the size of the partition than that employed by Riemann.2 Exercises Exercise 1. except that the sampling points ti are not required to belong to the interval [ x i .P ) is the smallest Riemann sum associated to P for this function f . this gives a lower bound on the area inside of C.b]. nonoverlapping 27r isosceles triangles.3 Let 0 < a < b. Find the area of T . discussed in Chapter 5 . Define f : [a. By joining the vertices to P . The construction of the McShane integral is exactly the same as the HenstockKurzweil integral. It will be seen that this leads to a very powerful theory more general than the Riemann (or Lebesgue) theory. Exercise 1.Introduction 9 The HenstockKurzweil integral studied in Chapter 4 is obtained by using the Riemann sums as described above. D n Use this information to find the area of I n . Since more general sums are used in approximating the integral. Explain why the Cauchy sum C (f. Each of these 2n triangles will make an angle of .at I. Then.1 Let T be an isosceles triangle with base of length b and two equal sides of length s. Exercise 1.2 Let C be a circle with center P and radius T and let In and On be ngons inscribed and circumscribed about C. likewise uses Riemanntype sums. find the area of On to get an upper bound on the area of C. the McShane integral has some very interesting properties and it is actually equivalent to the Lebesgue integral. The McShane integral.
This page intentionally left blank .
Thus. x i ] : p ( P ) = max {xi .1 Riemann’s definition The Riemann integral.b] into contiguous subintervals.xi] . The groundwork for the Riemann integral of a function f over the interval [ a .xi1 : i = 1.xn} be a partition of [a. as we will see in subsequent chapters.b] c R. .l .was the first of the modern theories of integration and enjoys many of the desirable properties of an integration theory.X I .xl]) = xi . .b] is a finite set of numbers P = {xo. xn} form an increasing sequence of numbers in [a. ..b] begins with dividing the interval into smaller subintervals. Let f : [a. the points {xo. The mesh of the partition is then the length of the largest subinterval. ..xi1.1 Let [a.b] that divides the interval [a. . . X I . . While the most popular integral discussed in introductory analysis texts. A partition of [a.. . .. l ( [ x i . For each subinterval [ i l define its length to be z . and ti E [ i l for each i. .. the Riemann integral does have serious shortcomings which motivated mathematicians to seek more general integration theories to overcome them.Chapter 2 Riernann integral 2..x1. xn = b and xi1 < xi for i = 1. As noted in Chapter 1. xi^. the approximating (Riemann) sums defined with respect to the partition P and the set of sampling points 11 . P = ($0. defined in 1854 (see [Ril]. xn} such that xo = a .[Ri2]).n } . . .b] + R. .b]. Riemann began by considering z . .n..xi]. Definition 2.
b] with . dt This definition defines the integral as a limit of sums as the mesh of the partition approaches 0. Definition 2. the infimum m a y be 0. Then. xi] for all i . then w e write A = s. sI f . d E R with a 5 c < d xI be the characteristic function of I.. Thus.b] and both A and B satisfy Definition 2.b]. Since E was arbitrary. d] and let (4= { 1ifxEI OifzgI’ . we use (any positive number smaller than or equal to) the smaller oaf the two 6’s.5 Let a. 0 Remark 2. in this case.LA ( P )< 6 and ti E [xil. b. in the definition with E‘ = 5. Fix E > 0 and choose 6~ and 6 corresponding to .” f = s.b]to be a “limit” of in the following sense.2 A function f : [ a .” f (t> or. then the value of the integral is unique. Proof. Set I = [c. Example 2.4 The value of 6 is a measure of how small the subintervals must be s o that the Riemann sums closely approximate the integral. Let 6 = min ( & A . c. The following proposition justifies our definition of and notation for the integral. This works f o r a finite number of conditions b y choosing the minimum of all the 6’s) but m a y fail f o r infinitely m a n y conditions since.) ) and suppose that P is a partition with p ( P ) < 6. the sums S (f. W h e n we wish to satisfy two such conditions. Suppose that f is Riemann integrable over [a. it follows that A = B. and hence with mesh less than both 6~ and 6.12 Theories of Integration Riemann considered the integral of f over [a. Proposition 2.3 Iff i s Riemann integrable ouer [a.P .2. Let be any set of sampling points for P . if we set I = [ a .b] + R is Riemann integrable over [a.6. the value of the integral is unique. respectively. We consider now several examples.b] if there is an A E R such that for all E > 0 there is a 6 > 0 so that if P is any partition of [ a . A and B .b]. defined by XI b.
.c)I < 2s = E . (If c E P \ { a . xi] On the other hand.26 so that I S ( f .b]with mesh less than 6.6 Define f : [ O .1] and choose ti so that xi1 5 ti 5 xi. Write as a telescoping sum 4 . Let P = {xo. . Ja xI = d .xi] is either xi . x} be a partition of [0. Let P = {xo.=l) 2 = i= j + 1 c c k1 k i=j (Xi .xn} be a partition of [a. l ] + R by f ( z ) = x.Riemann antegral b 13 Then. if ti E [xil.c.xi1 or 0 depending on whether or not the sampling point is in I .) Then.b].(d .{ti}.. for each i. . then c is in two subintervals determined by P. Now. The contribution to the Riemann sum from [xil. fix e > 0.xj]and let lc be the largest index such that d E [xkl. P .. let b = e /2 and let P be a partition of [a. {ti).zi] be a subinterval determined by the partition. zk]. s (f.. Let j be the smallest index such that c E [xjl. b}.xj1) + (Zk  Zk1)) > (d . . Thus.xl.P . . Let [xil..c ) .xi1) .{(xj .xi1) (xi .=l) . x1 is Riemann integrable and Example 2.x1.
. See Corollary 2. Now.14 Theories of Integration Then. s (f. suppose f were .xi1) = i=l c n i=l 0 =0 n So.7 Define the Dirichlet . and can handle functions whose points of discontinuity form. given e > 0.1] + R by Let P = (20.xi1) = 1. X I . no matter how fine the partition. . x:.1]. xi] there is a rational number ri and an irrational number qi. However. we can always find a set of sampling points so that the corresponding Riemann sum equals 0 and another set so that the corresponding Riemann sum equals 1.42.xn} be a partition of [0. in some sense. . if ( P )< 6.) = while n c n f (Ti) (Xi . Since So.function f : [0. n i=l since (xi . Thus. {Ti>:==.1] and has integral .l 3 The Riemann integral is well suited for continuous functions. if the function has many discontinuities. Thus. this integral may fail to exist. In every subinterval [xil. Example 2. Then. . set 6 = e.P . f is Riemann integrable on [0. a small set.
g : [a. We can think of the Riemann integral in much the same way.8 (Linearity) Let f. Fix E > 0 and choose St > 0 so that if P is a partition of [a. p E R.b] with p ( P ) < Sf. positivity means that a nonnegative input produces a nonnegative output. Fix E < and choose a corresponding 6. then 0f pg is Riemann integrable and + Proof. so that the Riemann integral is an operator acting on Riemann integrable functions. If P is any partition with mesh less than 6. Two fundamental properties satisfied by the Riemann integral or any reasonable integral are known as linearity and positivity. then 3 This contradiction shows that f is not Riemann integrable.61 + R and let a.Riemann integral 15 Riemann integrable with integral A.2 Basic properties In the calculus. except now the input is a function and the output is either a number (in the case of definite integration) or a function (for indefinite integration). 2. . > 0 so that if P is a partition of [a. then . We call a function whose inputs are themselves functions an operator. then for any set of sampling points Similarly.b] with p ( P ) < S. Linearity means that scalars factor outside the operation and the operation distributes over sums. If f and g are Riemann integrable. choose 6. we study functions which associate one number (the input) to another number (the output). Proposition 2.
b] + R.16 Theories of Integration p ( P )< 6 and ti Now. for any positive 6." f 2 0.b] with p ( P ) < 6 and ti E [zil." Proof. Then. 2 0.b] with ) ..Bg is Riemann integrable and Proposition 2. .. f 2 0. P Consequently. Let E > 0 and choose a 6 > 0 according to Definition 2. It follows that s. s. .2. Then.zi]for i = 1. let 6 = min { 6 f . . E [zil. if P is a partition of [a. since S (f.xi]. Suppose that f is nonnegative and Riemann integrable. Then. it follows that af + .9 (Positivity) Let f : [a.n. 0 Applying this result to the difference g . . 6 and suppose that P is a partition of [a. Since E was arbitrary.f we have the following comparison result.
then f would be bounded on [u. x .xi]. Fix such a partition P and and let M .b]with a bound of max {MI. Then. if f were bounded on each subinterval [xil. with a bound of Mi. Then.b]. We formalized this result with the following proposition. Note that if p ( P ) < 6.M2. there is a subinterval [xjl.tn}.11 Suppose that f : [u. t j . . For.10 f (z) 5 g (x)for all x Suppose f and g are Riemann integrable on [u.~ 0 ~ x I2 . . .b]+ R is a Riemann integrable function. Thus. Set tj = yk. Note that as we vary Ic. Fix a set of sampling points ti E [xil. Suppose that f : [u. so that the sum xi] If is a fixed constant. Proposition 2. the smallest index such that x E [xjl. . a Riemann integrable function must be bounded. .xnl} > 0. Can such a function be Riemann integrable? Consider the following heuristic argument.Riernann integral 17 Corollary 2. xj]. t j + l . . . . . Then. Proof. Let x E [u. for i # j . xj]such ~ that (yk)l 2 Ic. .b] + R and f is unbounded on [u. Thus. x n . . .b]. . there is a sequence { y l ~ } r =c [zjl. Let P be a partition of [u.b] and let j be X . Let T be the set of sampling points {tl.b] and E [a. .b].l . Choose 6 > 0 so that sampling points and min(x1 .xj] on which f is unbounded. f is bounded. M n } . Then. the Riemann sums diverge and f is not Riemann integrable. .
18 Theories of Integration since the two Riemann sums contain the same addends except for the terms corresponding to the subinterval [ x j . < 1.2. It follows that If )I(.xj1) + 1 Since x was arbitrary. .b].l . we see that f is bounded. .4 (Zj . 0 2. . Let Pj = . { x ~ } : = ~ be satisfies a Cauchy condition. zj]. j = 1. that is. Suppose that f is Riemann integrable on [u. The proof of the boundedness of 1 Riemann integrable functions demonstrates that the Riemann sums of an integrable function satisfy an analogous estimate. Then. given E > 0 there is a natural number N such that x . Fix E > 0 and choose 6 corresponding to ~ / 2 in xy). m > N . x : } . or (Zj .xj1) < If ( t j > l (Zj .xj1) + 1 L A.3 Cauchy criterion Let { x ~ } : = ~ a convergent sequence. < E whenever n. Further.2.. be two partitions Definition 2.
Then. P 2 .b] with p ( P j ) < 6 and points relative to Pj. . Next.) s  I" /I" f /+ f  s ( f . So.{ti1'}.R i e r n a n n integral 19 p] with mesh less than 6 and let ti') E [x??. by min { S 1 .. Let A be the limit of this sequence. Thus. actually characterizes the integrability of f . P 2 . p ( P i ) < 6 j 5 6 k . { t i 2 ) i=l n n }n ) )I L 1 (f. . i= 1 { > k .b]+ R. and a set of sampling points ti")}nk Note that for j .{ t i 2 ) } n z < i=l Proof. We have already proved that the integrability of f implies the Cauchy criterion. which is known as the Cauchy criterion. . then 2=1 { t. z ) .Pl. P l . are partitions of [a.Sk}.{ti2)}"z ) i=l I < E. For each k E N.2. We will prove that f is Riemann integrable.ll are sets of sampling Is ( f . we may assume that 6 k 2 6k+1.12 Let f : [a. sequence in R. Theorem 2. choose a 61. the condition that for all partitions PI and P2 with mesh less that 6. and corresponding sampling points.Then 1 ( f .s ( f . and hence converges. for each k .. Analogous to the situation for realvalued sequences. we have )I Replacing 61.)}nl ) .{ttl)}nl ) . . P l . with mesh less than 6 k . fix a partition Pk with p ( P k ) < 61.{tl. > 0 so that for any two partitions PI and P2.'j'}.Pk. 6 2 .(ti2)}n2 s 2=1 i=l )I = l S ( f .b] iA and only i i for each E > 0 there is a 6 > 0 so that if Pj. P l . j = 1. which implies that the sequence { S (f.s ( f . P 2 . It i=l )}m is a Cauchy k=l . P 2 . assume the Cauchy criterion holds..~ f + S h f . f is Riemann integrable over [a.S ( f . { p } 2=11 .
2 if the value of the integral is not known. define Mi and mi by and We define the upper and lower Darboux sums associated to f and P by u (f. Gaston Darboux (18421917) developed a generalization of Riemann sums and used them to characterize Riemann integrability. .1 2=1 It remains to show that A satisfies Definition 2. .x n } be a partition of [a. 2.2. twentyone years after Riemann introduced his integral.i = 1. K K It now follows that f is Riemann integrable on [u.b] + R be a bounded function and let m = inf {f (2) : a 5 z I b} and M = sup {f (x): a 5 x 5 b}. ) = P c n M (xi .b].4 Darboux's definition In 1875. so that m 5 f (z) 5 M for all z E [a. Fix E > 0 and choose K > 2 / ~ .{p}"" A1 I i ' s ). . Let P be a partition with p ( P ) < SK and let be a set of sampling points for P .2 1 .b]. .. 0 In practice.xi1) i i=l . Then. Let P = {xg. ..b].n. (See [D]. . 1 1 <+<€. x i ] . the Cauchy criterion may be easier to verify than Definition 2.) Let f : [a. see also [Sm]. and for each subinterval [xil.20 Theories of Integration follows from the previous inequality that 1 ( f . P k .
.P)=O. 24 . since m 5 f (z) I When f 2 0.x i .. 5 3 7 6. i b Figure 2. ) P M .3].P ) .0 (: ) Y (:. P ) =C m a ( z i .Riemann integral 21 and n L ( f . Using calculus to find the extreme values of f on the three subintervals.1  Example 2.13 Let Consider the function f (z) = sin ~ T X the interval [0. we see that 29 and 7 L(f.. ..I ) .(3 . each upper Darboux sum provides an upper bound for the area under the graph of f and each lower Darboux sum gives a lower bound for this area. i=l Note that we always have L (f. on . In fact.:) 3 1. we have I U (f.
7) is not Darboux integrable on [0. Proof.. P'> u P>.1].b]. Definition 2. suppose say c. ) 5 L (f. every partition point of P is also a partition point of P'. so that 0 J'f x E =0 and Jof = 1. we do not introduce any special notation for the Darboux integral. u(f. E x a m p l e 2. P r o p o s i t i o n 2. I (f.} be a partition of [a. We say that a partition P' is a refinement of P if P implies x E P'.x1. then L (f. xj1 < c = < xj. we give an example of a function that is not Darboux integrable..since it is an infimum of upper bounds for this area. We say that f is Darboux integrable if J%f = J b f and define the Darboux integral of f to be equal to this common value: Our main goal in this section is to show that a bounded function is Darboux integrable if. Similarly.15 The Dirichlet function (see Example 2. Let to P.14 Let f : [a.b] + R be bounded and let P and P' be P P') and partitions of [a. In fact. 1 Let P be a partition.b] +R be bounded. Thus. . the lower integral yields a lower bound. and only if. Suppose defined as above. we define the upper and lower integrals of f by and both of which exist since the upper sums are bounded below and the lower sums are bounded above. that is. L (f. ) = 0 and U (f. and that the integrals are equal. Let Mi and Set Mj' S U P { ~ ( I I : ) : zj1 5 II: 5 c } .. mi be P = {xo.finement of P . It follows from the comment above that when f 2 0. The next result shows that passing to a refinement decreases the upper sum and increases the lower sum.b] and P' is the partition obtained by adding a single point.16 Let f : [u.x. ) = 1 for every partition P P P. Before pursuing that result. If P' is a re. the upper integral gives an upper bound for the area under the graph of f. it is Riemann integrable.22 Theories of Integration Next..
P Proof. L (f. P P'). Proposition 2. 2 L (f.b] which is a refinement of both PI and 732.18 Let f : [a. Finally. Let PI and P2 be two partitions of [a. my 2 mj. 0 An easy consequence of this result is that every lower sum is less than or equal t o every upper sum. we see that L (f.b]. Then. P ) . By the previous corollary. ) 5 U (f. By the previous proposition. P ' )5 U ( f . suppose that P' contains k more terms than P. is an upper bound for the set P') { L (f. Repeating the above argument k times. it follows from M. We can now prove that the lower integral is less than or equal to the upper integral.b]. Similarly.732). I ) 5 U (f. and : my = inf { f (z) : c 5 z 5 z j } .b] + R be bounded and let PI and P2 be partitions of [a. it follows that Since all the other terms in the lower sums are unchanged.17 Let f : [a. M y 5 Mj that P') P so that U ( f . mi = inf {f(z) xj1 5 x 5 c}. ) .b]. Then. L (f. Since mi. Corollary 2. Let P and P' be two partitions of [a.Riernann integral 23 M y = sup{f(x) : c 5 z 5 zj}. adding one point t o the refinement at each stage. P = PI U P2 is a partition of [a. completes the proof of the proposition.b ] } . Proof.b] + R be bounded.. Then. so that U (f. which implies that .P ) : P is a partition of [a.
) .L (f. P as we wished to show.L ( f .b ] } .4. We claim that f is Darboux P integrable.24 Theories of Integration Since this inequality holds for all partitions P'.b] iJ and only iJ for each E > 0 there is a partition P such that 7 : Proof. we see that U (f. ) : P is a partition of [a. f is Darboux integrable on [a.P ) < E . and.b] + R be bounded. . As the next result shows3. 0 2.1 Necessary and sumcient conditions f o r Darboux int egrabilit y Suppose that f : [a. ) . Theorem 2. We have already proved that Darboux integrability implies the existence of such partitions. consequently. So. we see that J b f is a lower a bound for the set {U (f. U Pu.ckis condition actually characterized Darboux integrability. Then.19 Let f : [a. Then. ) < E . P P Since J b f = f . There is a partition PL such that and a partition PIJsuch that Let P = P.b] + R is bounded and Darboux integrable and let E > 0 be fixed. assume that for any E > 0 there is a partition P such that U (f.
and since 7 4 E was arbitrary.1 ) I ( f . xn} be a partition of [a.l . ) . Let M be a bound for I f 1 on [a. . .1. we use Theorem 2. there i s a 6 > 0 so that U (f. yk]. f is Darboux integrable if.L ( f . It follows that c iEJ (Mz . . .2 Equivalence of the Riemann and Darboux definitions In this section. Then. Set X b] and m = inf {f (z) i : 5 x 5 xi}. Then.L ( f . ) < E f o r any partition P with p ( P ) < 6 . Thus. P ’ ) < . ~ i ] . xi] is contained in [ykl. . . P ’ )< U 5’ E . 0 2.yi. then there is a k such If that [ q . xi] which contain a point of P’ and J = {0.X i . i E J . .b] + R be a bounded function..20 Let f : [a.Riemann integral 25 Let E > 0 and choose P according to the hypothesis.4. .mi)(Xi . By Theorem 2. P ’ ). given E > 0.f = S’f. we will prove the equivalence of the Riemann and Darboux definitions.s*fl< E. there is a partition P‘ = E E {yo.L (f. Then. Theorem 2. P ’ ). Let I be the set of indices of all subintervals [zil. To begin.19 to prove a Cauchytype characterization of Darboux integrability. .l . Set 6 = and let 2 8Mm P = (20. with p (P)< 6. I .ym} such that U ( f ..19. f is Darboux integrable. n } \ I . where the second inequality follows from the fact that a point of Pr may be contained in two subintervals [ z i .. Separate P into two classes. and only if. P P Proof. we have b s. integrable and fix E > 0. It follows that a 1: f . Suppose that f is Darboux b].
P ) > we L (f. P Thus. L (f.P ) 5 A 5 U (f. while by construction. U (f. Thus. . it is enough to find a partition P such that U (f. 0 Theorem 2. then sb T:f. f is bounded.ti E [xi+ 4 such that Mi < f (Ti) ~ / (4 . f is Riemann integrable if. . ) and L (f.b] + R. . + n. A = f = Fix E > 0 and choose 6 by Theorem 2. ) < E .20. ) < E and f is Darboux integrable. and only if. Then. Then. By the definition of Mi and mi.P ) . Hence.5.P . using {ti}:==1.” f .A J < E for any set of sampling points {ti}:=l..a ) < mi. Another applicaP tion of Theorem 2.P ) . . Since f is P Riemann integrable. U (f. {ti}:=1) 5 U (f. f is bounded and Darboux integrable.19 shows the other implication and completes the proof of the theorem.P ) . Proof. there is a 6 so that if P is a partition with mesh less than 6.26 Theories of Integration Combining these estimate shows U (f.19. there are points Ti.ion with mesh less than 6 and let {ti}:=l be a set of sampling points for P. to show that f is Darboux integrable. P s.&.L (f. . Consequently. Similarly.X I . n. ) < E.. . b b for i = 1.~ / (4 .a ) and f (ti). By Proposition 2. x n } .L (f. Suppose f is Riemann integrable and E > 0.P )  0 . Fix such a partition P = {xo. Let P be a par2.P ) . ) < E.L (f. f is Riemann integrable with Riemann integral equal to A . ) 5 P P S (f.11. Suppose that f is bounded and Darboux integrable and let for any set of sampling points {ti}. by definition. {ti)E1) .P . p ( P ) < 6 implies IS ( f . see that L (f. By Theorem 2.21 Let f : [a.
x such that cp (x)= ai for xil < x < xi.”cp = L (f. for each E > 0 there are step functions cp and such that cp 5 f 5 and + + [ + s. there is a B > 0 such that (). we view cp and $J as step functions defined by the partition P . . Let f : [u.x n } .I . so that we can assume that cp and are defined by the same partition.22 Let f : [a. . f is Riernann integrable if.zi)(2) + ~ n x [ z . they assume a finite number of values. Step functions are clearly bounded.b]+ R and let P = { x o .xi1). . Fix E > 0 and choose cp € and such that (+ .b]. we will refer to Darboux integrable functions as being Riemann integrable.First. and s. . We need only show that the existence of such step functions for each E > 0 implies that f is Riemann integrable.. q . . . s.3 Lattice properties Fix an interval [a. x n } .” (+ .b] and scalars { a l . and set P = P9 UP$ Next.b] + R.”+ + are step functions.b] as follows. z ~(2) ] * = U (f.19.. Since cp _< f 5 and cp and are bounded. ai+l or any other value. .I 5 B + + + If . n } of (a. respectively. We call a function cp : [a. cp and n1 (2)= C i=l Mix[zil. We are not concerned with the definition of cp a t xi.2.cp) < .an) . Suppose that our fixed partition P equals {zo. cp 5 f 5 +.9 < 4 Proof.b] +R a step function if there is a partition P = {q. Then. 2 1 . . .zi) (x> + m n ~ [ z .. . Changing the value of p at a finite number of points has no effect on the integral.q. .P ) and Theorem 2. i = 1. and only i f . . we have the first half P of the following result.4. . . .1 and linearity. and define cp and by cy==. we partition [a. See Exercise 2. Let P9 2‘ and PQ be partitions defining cp and @. . it could be ai. + n1 CP (x>= and C i=l mix[zil . we see that step b functions are Riemann integrable with integral Ja cp = ai (xi .znl (x> ~ + Clearly.n. 2. As a consequence of Theorem 2. By Exercise 2. .Riemann integral 27 Consequently. ) . .
xi+l) < . by f+ = max{f.zol such that Iy: lxn .L (f.f.inf {f (2) : x E I } ) ! ( I ) 5 2 B l ( I ) < 2B=8Bn 4n‘ Since there are 2n such intervals. P’) PI) < completes the proof. Given functions f and 9. If1 = . b y f A g ( x ) = m i n { f ( z ) . ~ n ) such that is a refinement of P . z ~ ) such that \y&. . We consider two types of intervals: those of the form [& .O}. It followsthat themaximumand the minimum of two step functions is also a step function. inductively choose yi E ( ~ : .yi] P’) and the ones with an zi for an endpoint. Choose yb E ( z o . (sup {f (2): z E I } . See Exercise 2. P’) PI).l . by f V g (x)= rnax { f (2). g ( z ) } .28 Theories of Integration E for all x E [a.10. we get a contribution to U ( f .O} and f. . . since cp 5 f 5 on the interval. we see that f = f+ .= max{f. Then. denoted by f+ and f.respectively.. From these definitions.b].g (x)} and the minimum of f and g. denoted f V g . say. f A g .P’) that is less than Combining these two estimates shows that U (f.L (f. . choose 8Bn* The partition E Yn E ( y k .. .to the difference U (f. P ’ ) L (f. < E . n .and. and we are done if we can show that U ( f . Given a function f. 2 Next. the sum of these terms contribute less E than . E E + + Summing over all such intervals. equal to ai and bi.1. . consider an interval of the form Ji = [Y(~.~ . Suppose I is a subinterval determined by PI with an xi for an endpoint.yil < E <8Bn Finally. z and y: 6 (xi. Thus.ynl 8Bn i) for i = 1. P ‘ ) L (f. we define the positive and negative parts of f . On such an yi] interval. we define the maximum of f and 9. E and 0 It is easy to see that the sum and product of step functions are step functions. cp and are constant.
the set of Riemann integrable functions on [a.= w. it follows that f1 A f 2 is Riemann integrable. we may assume that each fi 2 0.. fS. Another application of the use of step functions allows us t o see that the product of Riemann integrable functions is Riemann integrable. for i = 1.b]. Since f1 A f2 = f1 f2 . then is Riemann integrable. Fix E > 0.b] + R are Riemann integrable. See Exercises 2. f . by linearity.b] is a vector space. Then.b]. then f l v f 2 f and fl A f2 are Riemann integrable.f1 V f 2 . Thus.2. f + = 7and f . If Corollary 2. f1 f 2 Proof. For example.nnn.b] + R are Riemann integrable.Then Pi V P 5 2 2' f1 v f2 5 $1 v $2. Note that I may be Riemann integrable while f is not.23 I f1.24 Suppose f is Riemann integrable o n [a.12. f2 : [a.25 If fl.Riem. Choose M > 0 so that fi ( 2 ) 5 M for i = 1 .g E S implies that f V 9.a n d f 1 are Riemann integrable o n [a. L Theorem 2. the set of Riemann integrable functions on [ a . 2 and 2 E [a. inteyral 29 We will now use step functions to show that these operations preserve integrability.22. (qi. there are step functions b E 'piand q i such that 'pi L fi 5 $i and S. follows by checking various cases. L lfl+f f + + f .PI .'p1 v ~2 $2 . we see that + Applying the corollary one more time.b] and We leave the proof as a n exercise. Since $l v $2 . 0 + A set of realvalued functions with a common domain is called a vector space if it contains all finite linear combinations of its elements. By Theorem 2. A vector space S of realvalued functions is called a vector lattice if f. There are . An immediate consequence of the previous theorem is the following corollary.Pi) < .~ 2 which . f2 : [a. By the previous corollary.f A g E S.b] is a vector lattice.11 and 2. Proof. Corollary 2. we have that f l V f 2 is Riemann integrable.
then f ) is constant and the result is a consequence of Example 2. 2. the condition of Theorem 2. (!I)/}. Without loss of generality. We now prove that monotone functions and continuous functions are Riemann integrable. 0 xy=l . (If f ( b ) = f (a). 0 Integrable functions The Darboux condition or.) Since f is increasing. Fix E > 0.b]. f is Riemann integrable on [a. {If If { f (xi) . Let P be a partition with mesh less than ~ / ( f ( b . Theorem 2. more correctly. Mi = f ( x i ) and mi = f (xi1).26 Suppose that f is a monotone function on [a.19 and 2.21. f is Riemann integrable.4.)I . It follows that Proof.22.19 makes it easy to show that certain collections of functions are Riemann integrable. Clearly. we may assume that f is increasing.b]. we may assume that 0 < qi and and observe that set and Hence. it is enough to Moreover.f (xi1)) is where the next to last equality uses the fact that a telescoping sum. f is bounded by max (. and and By Theorem 2.30 Theories of Integration and step functions and such that In fact.5 and linearity. Then.f ( a ) ) .4 f1f2 is Riemann integrable. By Theorems 2.
f as Riemann integrable on [a. Then.S (f.4.m i = Thus.ti1 5 p ( P ) < 6. then x E I f ( 4 . fix a function f : [a.P . .19.b]. To do this. . n n and the proof is completed as in the previous theorem.ti E [xil. Unfortunately. Let P be a partition of (a. Proof. Then. Ti.zi] such that Mi = f(Ti) and mi = f ( t i ) .R i e m a n n integral 31 Suppose that f is a continuous function on [a. a function acting on functions. y E [a. If P is a partition with sufficiently small mesh (depending on uniform continuity) and and {t:}y=l are sampling points for P . F may not be defined for many subsets of E . and we have Theorem 2. Since 1 i . then can be made as small as desired.. One of the recurring themes in developing an integration theory is to enlarge as much . Such problems can be avoided by using Theorem 2.it is uniformly continuous there.f (Y)l < ba. We say that f is Riemann integrable over E if the function fXE is Riemann integrable over [a. S (f.. f is uniformly continuous. for i = 1.b]. .n. Let E > 0 and choose a 6 so that if x .b] +R.b]. T Mi .y ( < 6.27 Suppose that f : [a.xi]. Since f is continuous on the :ompact interval [ i l there are points z . Thus.b] +R is continuous on [a. it seems likely that the Riemann sums for f will satisfy a Cauchy condition and f will be Riemann integrable.b] and I . .b] and we define the Riemann integral of f over E to be Unfortunately. which makes a proof along these lines complicated. the Cauchy condition must hold for Riemann sums defined by different partitions.b]. 2. and let E c [a. We can also view the integral as a function acting on sets. Since f is continuous on [a.b].5 Additivity of the integral over intervals We have observed that the integral is an operator. P .b] with mesh less than 6.
b].b]....then f is Riemann integrable on every subinterval of [a.b] with p (P[.c] and [c.] any partition of [a.28 Suppose that f : [a.. u P[c.]) < 6. < Since for any bounded function g and partition P .b].~ any partition be be of [ c .] of [ a . such that and a similar result for [c. let P[c. p (P) 6 and Then. ' that s.b] with p ( P ) < 6.P ) ." f < 5P E E .L (f.P [ c .20.c] and [c.1.c] and [c. we fx e > 0 and choose partitions PfU. then U (f.c] with p (PI. c ] ). ) P follows that 5 U (9. ) . To see that f JCb f = f.61. b ] ). it is enough to show that there is a partition P[. [ u .b~. L (9. For the Riemann integral." and P [ c . We first claim that f is Riemann integrable on [a.b] +R is Riemann integrable and c E ( a ."f < 5 and u (f. b ] such u (f.. Let PI...b].b]. there is a 6 > 0 so that if P is a partition of [a. As we will see below. Given E > 0. a natural collection of sets is the collection of finite unions of subintervals of [a.s.P ) < E.b]. it P and so that f is Riemann integrable on [a. By Theorem 2.b). and Proof.32 Theories of Integration as possible the collection of sets that are allowable as inputs.. Proposition 2. Then.. if f is Riemann integrable on [a.] i 1 + . f is Riemann integrable on [a.bl) and set P = P[u.c~ < 6.
~and [c.b].b] and [c. f is Riemann integrable on [c. I" n J" = 0.d] c [a.P ) . Then. if f is Riemann integrable on [a. < E.b] + R is Riemann integrable and [c. to be the set of x E I such that there is a 6 > 0 so that the 6neighborhood of z is contained in I .R i e m a n n integral 33 Set Then.b]. then I U J is an interval and this equality is an application of the previous proposition. 2. Suppose f is Riemann integrable on [a. then I U J is no longer an interval.b]are intervals with disjoint interiors.6 . Let I be an interval. (z . See Exercise 2. we have Corollary 2." f is the infimum of the U (f. we have + which is called an additivity condition. When I and J are at a positive distance.b] then f is Riemann integrable on [a. By applying the previous proposition twice.d] c [a. Since we can do this for any E > 0 and 6 we see that JaC f JC6 f = Ja f.b]idifferentiable R is .d ] . Then. Suppose that f : [a.5 Fundamental Theorem of Calculus The Fundamental Theorem of Calculus consists of two parts which relate the processes of differentiation and integration and show that in some sense these two operations are inverses of one another.b] (see ] Exercise 2.b] + R and I . + s. x 6) c I . denoted I " . When I and J are contiguous intervals. We begin by considering the integration of derivatives. Suppose that f : [a.15). J c [a. 0 We leave it as an exercise for the reader to show that if f is Riemann integrable on [ a .b]. if necessary.29 Suppose that f : [a.16. We define the interior of I .
we see that f‘ = f ( b ) . xi] such that f (xi) . Theorem 2.31 Define f : [0.30 is that f’ is Riemann integrable.. f is differentiable on [0. for any partition P .) . In fact. let P = {xo.b]. I] 4 R by Then. . Thus.n. Proof. xn} be any partition of [a. { tik)}:k ) = f (b) . for i = 1.f ( a ) .f ( a ) 2=1 i=l for all k.1) does not hold in general for the Riemann integral. there is a yi E [xil. ‘ n n which is a telescoping sum equal to f (2. The first part of the Fundamental Theorem of Calculus involves the familiar formula from calculus. n .b] with derivative f‘. 21.f ( a ) . s.. (2. Since f is differentiable on (a. Then. P k . Hence. x2 if 0 < x if 51 0 x=O . we are done if we can find a sequence of partitions {Pk}Eland corresponding sampling points { such that p (Pk) + 0 as k + 00 and S ( f ’ . . .34 Theories of Integration on [a. .b].b]. Since f‘ is Riemann integrable.b] + R and f’ is Riemann integrable on [ a .30 (findamental Theorem of Calculus: Part I) Suppose that f : [a. .b]. Example 2.1] with derivative 2xcos 2T + sinx 22 n .xi1).” The key hypothesis in Theorem 2.1) holds.b) and continuous on [a. Thus.f (20)= f ( b ) .f (xi1) = f (pi) (xi . The following example shows that (2. we may apply the Mean Value Theorem to any subinterval of [a.. there is a collection of sampling points { ~ i } y = ~ such that Taking any sequence of partitions with mesh approaching 0 and associating 0 sampling points as above.
31 ‘ is also not Lebesgue integrable so a general version of the Fundamental Theorem of Calculus for the Lebesgue integral also requires an integrability assumption on the derivative. called the gauge or HenstockKurzweil integral. ‘ There are also examples of bounded derivatives which are not Riemann integrable. that is. Thus.Lin(x. In Chapter 4 we will construct an integral.51 or [Swl.b]. for example. Theorem 2.I M for all x E [a. F . (See. If a 5 x < y 5 b. page 201.) We will see later in Chapter 4 that the derivative f in Example 2. Let f be Riemann integrable on [a. but these are more difficult to construct.1) holds. we define sy” f = .y E [a.” f.3.32 (Fundamental Theorem of Calculus: Part 11) Suppose that f : [a. Let x.b] with Lipschitz constant C. Choose M > 0 so that () 5 .b]. Section 1.F’(Y)l = = lLZ LY f ( t )dt f f ( t )dtl If @>I dt I IY . the HenstockKurzweil integral integrates all derivatives and (2. we have If IF (4 . Suppose that f : [a.b] and consider the difference F ( x ) . Using the additivity of the Riemann integral. Set F (2)= s. The second half of the Fundamental Theorem of Calculus concerns the differentiation of indefinite integrals. for which the Fundamental Theorem of Calculus holds in full generality. any indefinite integral satisfies a Lipschitz condition and any such function is uniformly continuous. Section 1.4.b] + R is Riemann integrable on [a. Section 3.4 * M 1.1].” f ( t )dt. [Be.1]. Then.” f = 0. [LV.b].3. We define the indefinite integral of f at x E [a.b] + R is Riemann integrable.s. f is not Riemann integrable on [0.y) A function g satisfying an inequality of the form is said to satisfy a Lipschitz condition on [a.b] by where F ( a ) = s.L’ m=(x.y) (t) .F (y).Riemann integral 35 Since f’ is not bounded on [0. page 981. The second part of the Fundamental Theorem of Calculus concerns the differentiation of indefinite integrals.
Suppose f is continuous a t and let E > 0. The indefinite integral of sgn is F (x) = 1x1. we only need show that the continuity of f implies the differentiability of F. if that (x) . This theorem only guarantees that F is differentiable a t points a t which f is continuous. 5 E [a. the indefinite integral is not differentiable a t the point where the function is not continuous. if # which is continuous at every x Next. So. We will discuss this in Chapter 4. If f is not continuous at a point. even though f is not continuous there. In fact.f (01 < If 0 < Ix . There is a 6 > 0 such € whenever x E [a. 0 The theorem tells us that F must be differentiable a t points where f is continuous.b]. then F is differen Proof.33 Define the signum function sgn:R +R by sgnx = ifx#O 1x1 { x ifx=O 0 The function sgn is continuous for x # 0 and is not continuous a t 0. Here.b]. then That is. F is differentiable at 5 and F’ (t)= f (t). In this case.36 Theories of Integration is continuous on [a. F ( 2 ) = 0 for all x is differentiable at 0. F may or may not be differentiable. which is continuous everywhere and differentiable except a t 0. consider g ( 2 ) = Iifx=O’ except 0. Example 2. we need only set S = €/Ad in the Lipschitz estimate on F above. Thus.(1 < 6 .51 < 6. .b] and Ix . If f is continuous at tiable at t and F’ ( E ) = f (t). F is differentiable at “most” points. To see that F is continuous.
b] +R be continuously diflerentiable. Consequently. By hypothesis. (fg)' = fg' fg is Riemann integrable and. are consequences of the Fundamental Theorem of Calculus. Proof.d ] + R i s continuous. f : [c.d] with 4 ( a ) = c and 4 (b) = d. Assume + ( [ a b ] ) = [c." f (4 ( t ) ) ( t )dt. By Corollary 2. F and H are differentiable.34 (Integration by parts) Suppose that f. by Theorem 2. by Theorem 2. Integration by parts. dt = Proof." f ( t ) d t and H ( y ) = s. H ( a ) = 0 and F (c) = F (4 ( a ) )= H ( a ) C. integration by parts and by substitution. fg' and f ' g are Riemann integrable. ( F O 4)' (Y> = F' ( 4 ( 9 ) )4' ( Y ) = f ( 4 (Y>> 4' ( 3 )= H' ( Y ) 1 so that ( F 0 4) ( y ) = H ( y ) C. Thus.f (49 (4 ' 0 We now consider integration by substitution.27.35 (Change of variables) Let 4 : [a. Theorem 2. C = 0. both these integrands are continuous 4' so that.30. 1 f' (4 (4 Jd b b 9 dx = (fg)' (4dJ: a = f ( b )9 ( b ) .Riemann integral 37 2.5. Since F ( c ) = 0. We now have + + . ' + f (49 (4dz + ' The result now follows. Define F and H by F (2) = s. F o 4 is differentiable on [a. Note that f and g are continuous and hence Riemann integrable by Theorem 2. + R and b] f ' and g' are Riemann integrable o n [a.1 Integration by parts and substitution Two of the most familiar results from the calculus. Theorem 2.b] and Ju f b 9 (2) = f ( b ) g (b) . If .25.b].b] and by the Chain Rule. or change of variables. is a kind of product rule for integration.g : [a.b] and [f ( 4 ( t ) )4' (t> ld f. then f ( 4 )4' is Riemann integrable o n [a.32. fg' and fg are Riemann ' integrable o n [a. Then.f (a> (a>' dx J b f' (z>dz. which follows from the product rule for differentiation.
set U (z) = {t E [a.b]) < E [ n (f. .}:=~ C S that converges to x.38 Theories of Integration as we wished to show. that S c 3. Note. for 6 > 0. . .b]4 R be bounded and let S be a nonempty subset of [a.b]. n .6 Characterizations of integrability We now characterize Riemann integrability in terms of the local behavior of the function. then 3 is the union of S with the set of its endpoints. .S ) = sup {f ( t ): t E S } . n. The set of endpoints of these . Then. 3321) L w (f. It follows immediately that if S c T then w (f. x) Lemma 2. we require the notion of the outer Jordan x1. intervals that lie in (a. P ) be the sum of the lengths of the closed intervals .l .x} be a partition of [a. Let x E [a.30.b] : It . .b) along with the points a and b yield a partition {xo.inf {f ( t ): t E S }.x) = 0. .b] such that w (f.x. there is a lc so that 0 [ x i 4xi1 c Hence.) Let S be a subset of R. It is easy to see that f is continuous at x if. . w (f. It. Since { I t : t E [ a . . If S is a bounded interval. We begin with a lemma.xl < S}.b]. x1.b] such that for each i = 1.x1.S ) 5 w (f.. content of a subset S of [a.b] and. Let P = (20. It2. 0 2. .[xil. The oscillation of f over S is defined to be w (f. Proof. x.zn K' For our first characterization. there is an open interval It centered a t t such that w [a. denoted 9. xi]) < E for i = 1.36 Suppose that w (f. is the set of all x E R for which there is a sequence {s. .b].} of [a. . .b]..b] and let J ( S . . For each t E [a.b]. .}. .} of [a.G [a.b ] ) < E ..T ) ... < E for every x E [a. and only if. in particular. b ] }is an open cover of [a. U s ) is a decreasing function of S. there is a finite subcover { I t l . Let f : [a. (See Exercise 2. The closure of S. . The oscillation of f s at x is defined to be Note that the limit exists since w ( f .. . X i . . w ( f . Our first characterization of Riemann integrability is in terms of the oscillation of the function f . there is a partition P = {TO.
w (f.) consequently. .XS. xi]) 2 E . P ) as P runs through all partitions of [a.I k } and label the remaining subintervals determined by P by { J1.2 1 . Then. (For a discussion of Jordan content. ( f ) # 0.xn} be a partition of [a. + Theorem 2.) is less than .b] if) and only if) f is bounded and for every E > 0 . i1 3 2n 2n .b] and let I be the set of all indices i such that the intersection of [xil. and D.b].a ) = E so that f is Riemann integrable by Theorem 2. . Let q > 0. (f). Let xi] I' c I be the set of indices i such that (xi1. (See Exercise 2. x i ] contain points of the closure of S.26. .37 Let f [a. b S.P ) = U (xs. l . By Exercise xi) 2.+ R.J j ) < for j = 1 . Suppose i E I \ I / . . at least one of the endpoints of [ x i . We then have 2 (b. I 2 . . see [Bar]. .and 1 : .b] : w (f. C(S) = A finite subset of [a.yi E (xi1.31. . (See Exercise 2. Choose M > 0 such that (t)l 5 M for a 5 t 5 b and let E > 0. P s. ( f ) is nonempty. . Iyi . Let P be the partition of [a. J l } .a) If k 1 j=l E i= 1 < 2M 4M + 2 (b E  a) (b.~ . is defined to be the infimum of J ( S . . .b]such that the sum of the lengths of the subintervals determined by P that contain E points of DEpp.27.1 < .xi)such that yi < y l . then C(S U T ) 5 E(S) E(T). f is Riemann integrable over [a.b]. xi] is in D. _> E } . . We will use Theorem 2.) For E > 0.Riemann integral 39 [ ~ i .[xil. Then. We characterize Riex) mann integrability in terms of the outer Jordan content of the sets D. Refine P by adding 7 r7 yi.z .T c [a.) The set function C is monotone in the sense that if S c T then C(S) 5 C(T and is also subadditive in the sense that if S. but an infinite set can also have outer Jordan content 0. n D. ( f ) has outer Jordan content 0. for i E I / . Proof. the set D. (f) has outer Jordan content 0. (f)= {x E [a. . The outer Jordan conwhich tent of denoted C(S). . ) and. Suppose first that f is bounded and for every E > 0. . the set D. Applying the previous lemma to each J j .19. . Let these subintervals be labeled 4M { I l . (f). . set D. For the converse.l . Note that J (S. J2. Let P = {xo.b]. we may assume E that w ( f.b]obviously has outer Jordan content 0.19 to show that f is not Riemann integrable.xi1 < . assume that there is an E > 0 such that C(D. (f)) = c > 0.
. n D.xi1) (f. (f). . is a kind of limit of D. then f o r 6 > 0. closed intervals { I I .. [Xi1. it follows that f is not Riemann integrable. . Note that [yi. .L (f. This will not change the union of the intervals and will decrease the s u m of their lengths. .so that D (f)= U E > ~ D .b] : f is discontinuous a t t } . it follows that iEI' Hence.In} such that S c Ur=lIi and Cz1 ( I i ) < 6. we can cover S b y nonoverlapping.38 If S is a subset of [a.bi]. + i=l n If two intervals in the set (Ji)yz1have a nonempty intersection. ) . the test involves an infinite number of conditions and.b] with outer Jordan content 0 .J . if E < E' then D. Our second characterization. w 4 )(Xi . yi] and [yi.b i + f 4I L ) .P ) = P 2 c c n i=l w (f. If Ii = [ai. u (f. the s u m of whose lengths is less than 6 . there is a finite number of nonoverlapping.30. . Thus.xi1) 2 EC. consequently. gives a characterization of . (f)= 0 so that yi] m iEI' k=l J1. S c Ur=lIi c U y . However.40 Theories of Integration m For i E I \ 1'. While this theorem gives a characterization of Riemann integrability. i E I' Since this is true for any partition P . open intervals. 4)(Xi . As a consequence (f) of Exercise 2. ( f ) as E decreases to 0. ' I 4n ' Then.' ( f ) C D. we see that D ( f ) = { t E [ a . set Ji = (ai . . is not practical to employ. . we can replace them by their union. [Xi1. 0 Remark 2. l Ji and xy=l n i=l El(&) = c n i=l {[(Ii) + $}= El(&)f < 6. xi] by 5 2n. where Since 7 > 0 is arbitrary. Set l '7 7 = 6 l ( I i ) > 0. label the intervals [zil. due to Lebesgue.
Proof. Since this is true for every 6. ( f ) : n E N}. The endpoints of { I I . Lemma 2. D1/. By Exercise 2. Since q V+E is a neighborhood Us ( x ) of x such that < E .6 > 0 and assume that there exist open intervals { I i } z l covering D (f) such that t ( I i ) < 6.40 in other terms. since D. (f). A subset E c R is said to have Lebesgue measure 0 or is called a null set if. so that D ( f ) can be covered by a countable number of intervals. the sum of whose lengths is less than 62". D.I n } in [a. .39 Proof.32.Iz. By the previous lemma.6.b]\ D. s so that the complement of D.b] such that the sum of the lengths of the intervals.b]. so if q+' 2 x1.(f) is less than 6.b] along with a and b comprise a partition of [a. It follows that D.b] and. ( f ) is closed in [a. there exist a finite number of open intervals { I I . D ( f ) = U { D1/. and only i f is bounded and. Us ( x ) ) < . For each E > 0 . for every 6 > 0. czl 2. the s u m of whose lengths is less than 6. . ( f ) is 0 closed in [a. . u (4n D€ (f) = 0. . ( f ) is open in [a. Since E is arbitrary. the set D.b].then If (x1) . the sum of whose lengths is less than c. determined by the partition. there w (f. Theorem 2. D ( f ) can be covered by a ' countable number of open intervals. s Thus. We will use the following lemma. Then.b] + R. by the previous 0 theorem. E can be covered by a countable number of open intervals the sum of whose length is less than 6.Riemann integral 41 Riemann integrability in terms of the single set D (f). By our first characterization and the previous remark. I n } which cover D. let c. for each n.< E .f ( 2 2 1 1 < 2. (f) c D (f). . which intersect D.1 Lebesgue measure zero We can use one of the basic ideas of Lebesgue measure to give a restatement of Theorem 2. Next. ( f ) can be covered by a finite number of open intervals. f is Riemann integrable. D.b]. ( f ) and set q = w ( f .. The following example shows that a countable set has measure zero. Suppose f is Riemann integrable. x ) . Let x E [a. .x2 E u ( x ) . ( f ) is closed in [a. . f o r every 4 > 0 .I2. ( f ) has outer Jordan content 0. f is Riemann integrable if.=l 42n = 6 .40 Let f : [a. .
> 0 and let Ii = (ci . 2.e. Thus.42 A bounded function f : [a. pages 1401501. ci + ~ 5 2 .t null set. we will give an example of an uncountable set that is null.a < c < b.43 Let f : [a. (See [C.b]+ R is equal to 0 a. and we define the CauchyRiemann integral of f over [a.b] + R is Riemann integrable i f . Note that this guarantees that f is bounded on [c. Thus. 62i1 = 6/2 < 6. For example. if the limit fails to exist. It follows that C c uzlIi and CE1 ( I i ) = CZ. Corollary 2.b]. Then.b] : g ( t )# 0) has Lebesgue measure 0. we can write 00 C = { ~ i } ~ Fix~S.7 Improper integrals Since the Riemann integral is restricted to bounded functions defined on bounded intervals. a function g : [a. in [a. Ii is = ~).e.b] + R and assume that f is Riemann integrable on every subinterval [c. it is necessary to make special definitions in order to allow unbounded functions or unbounded intervals. in [a. restates the previous theorem in terms of null sets.e.+ f exists. Let f : [a. Definition 2. The following corollary. We say that f is CauchyRiemann integrable over [a.b]. we consider the case of an unbounded function defined on a bounded interval..b]. we say the integral diverges. A statement about the points of a set E is said to hold almost everywhere (a.b] to be s. were first carried out by Cauchy and we will refer to the extensions as CawchyRiemann integrals. C is a .41 Let C c R be a countable set. every countable set is null. an open interval containing ci and having length l ( I i ) = 62i1.b) but not necessarily on all of [a. These extensions. sometimes called improper integrals. and only i i f is continuous a.b] for c E ( a .b] if lime. due to Lebesgue. In Chapter 3.62i2. ." When the limit exists.b] means that the set {t E [a. we say that the CauchyRiemann integral of f converges. (2) 4.) First.b] + R be as above.) in E if the points in E for which the statement fails to hold has Lebesgue measure 0.2 ~ Then.42 Theories of Integration Example 2.
Example 2." f < c < b. but f is not CauchyRiemann integrable over [1.b]and ] the integral over [a.b s.1] by Example 2.6). since f is an odd function (see Exercise 2.35 we see that if f is Riemann integrable over [a. then f is said to be CauchyRiemann integrable over [a.2) may exist and f may fail to be CauchyRiemann integrable over [a. JC1 tpdt = P+l 1 1 1 exists and equals if.o+ J t p d t '1 0 < t 5 1 and f (0) = 0. if f is Riemann integrable over eve.44 so Similarly.C P S 1 so limc. then this definition agrees with the original definition of the Riemann integral and. If a function f : [a. then f is defined to be CauchyRiemann integrable over [a.b].1] since f is not CauchyRiemann integrable over [0.45 Let f ( t )= t3 for 0 < I1 5 1 and f (0) = 0. For p # 1. and only if. gives an extension of the Riemann integral. ~and [c. exists (and equals 0).b] is defined to be s.Riemann integral 43 By Exercise 2. However.y subinterval [a. p > 1.z).b]. l ] +R by f ( t ) = t p . then exists and equals f . for 1.44.b] if = limc. and only if. thus. the limit in (2.b]if f is CauchyRiemann integrable over both [ a .c ] .In c which does not have a finite limit as c + O+. Thus. Let p E R and define f : [ O .b]." t Example 2." f + Note that if f is CauchyRiemann integrable over [a.. This definition follows by applying the previous definition to the function g (z) = f ( a b . as the following example shows. Jc t d t = . Then. p > 1. a exists. .b] +R has a singularity or becomes unbounded a t an interior point c of [a. s.b]. and then tpdt = P+l p+l' 11 If p = 1. t P is integrable if.
if f : (m. See Exercise 2. For p # 1. A similar definition is made for functions defined on intervals of the form Example 2.m) + R is CauchyRiemann integrable over (. we say the integral diverges. However. 1 1 so limb. if the limit fails to exist. b f.00.3). p bP+l < 1.” f exists. s.2) is called the Cauchy principal value of f over [a.m) + R. we say that the CauchyRiemann integral of f converges. then.00) if.+ca tpdt exists and equals if. j = lim bca S.46 Let f : [a.38 shows that the value of the integral is independent of the choice of a. We define the CauchyRiemann integral of f over [ a . If the limit exists. P+l If p = 1. We next define the CauchyRiemann integral for such functions.3) may exist and f may fail to be CauchyRiemann integrable over (m.00). J y t p d t = p+1’ If f : (m. J . 1 p < 1 and. Suppose now that f is defined on an unbounded interval such as [a. m ) to be Lca (00.c .00).47 J:tpdt = if. We say that f is CauchyRiemann integrable over [ a . m ) if f is Riemann integrable over [a. + .m) if. 00) is defined to be b + l t dt = In b which does not have a finite limit as S_. 00.f= 00 Ja ca f + Jca. and only if.for t 2 1.39. Definition 2.oo). the limit in (2. then the limit exists. then f is CauchyRiemann integrable over (m.44 Theories of Integration If f is Riemann integrable over [a. Exercise 2. b] for every small e > 0. As in the case of integrals over bounded intervals. the limit in (2. t P is CauchyRiemann integrable over [l.b] and is often denoted by p v J i f.00) + R. bl. f and Jam f both exist for some a and the : CauchyRiemann integral of f over (m. Let p E R and let f ( t ) = tP.E ] and [c + E. and only if.f. and only J : b l Thus.b]for every b > a and limb. The limit in (2.
. First. If Proof. 2 Now. Assume that f is Riemann integrable ouer [a. 00) and is often denoted pwJTm .m g exists and so its terms satisfy a Cauchy condition. First. 0 sin x Example 2. Jr I X which diverges to oo as k + oo.” I 1 I dx.48 and b2 cosx 1 Example 2. sinx dx X exists. we establish a preliminary result called a comparison test.47 since 1 2 5 22.c + 00.b] f o r every b > a and that g is CauchyRiemann integrable over [a.24 that if a function f : [a. To see that this integral does not exist.g : [a.00). it suffices to show that the Cauchy condition is satisfied for this limit. f We saw in Corollary 2. However.J: f exists. Proposition 2.oo). Then. then 1 f I is Riemann integrable over [a. then as b.is CauchyRiemann integrable over [ T )00) la b but 1 Ic1 sin x ’ A m is not.00) +R and suppose that If (t)l 5 g ( t ) f o r t >_ a. limb.Riemann integral 45 if it exists. To show that limb. We show in the next example that this property does not hold for the CauchyRiemann integral.49 The function . sin b 00 cosx ..b].b].48 (Comparison Test) Let f. if c > b > a. limb+mNext.0 and Jn 22dx exists by Proposition 2.b] + R is Riemann integrable over [a. Integration by parts gives sinx dx z  sinb b2 s. since. we show that J.* 7 cosx dx. by assumption. is called the Cauchy principal value of f over (00. we consider note that s. f (and I) is CauchyRiemann integrable over [a.
” Exercise 2.= f A 0 is CauchyRiemann integrable over 2.b] + R are Riemann integrable with s.4 If f : [a. However. Suppose that I is any interval with endpoints c and d .it follows that neither f S = X [ T . the last example shows that.1 In Example 2. s . d ) .8 Exercises Riemann’s definition Exercise 2. b .b] + R is continuous.b] + R is Riemann integrable.” f = s. For a more thorough discussion of the CauchyRiemann integral.oo). (c. From the fact that f (2)= sin x is conditionally integrable over [ T .48 is a test for absolute integrability. h : [a. Note that the comparison test in Proposition 2.5. or [c.3 Suppose that f . in contrast with the space of Riemann integrable functions.24 shows that there are no such functions for the Riemann integral.2 Suppose f : [a. It follows from standard limit theorems that S is a vector space of functions. Show that if f is altered a t a finite number of points. [CS].” h. If f is integrable over I but I f 1 is not integrable over I . prove that f E 0. then the altered function is Riemann integrable and that the value of the integral is unchanged. either prove the result or give a counterexample. We will see later that the HenstockKurzweil integral admits conditionally integrable functions whereas the Lebesgue integral does not. Exercise 2. that is. s. 4.00). Is continuity important? Is positivity? In each case. we assume that I is a closed interval. suppose I has one of the forms (c. see [Br]. f V 0 nor f . nonnegative and f = 0. f is said t o be conditionally integrable over I . Let S be the set of CauchyRiemann integrable functions. Suppose that f 5 g 5 h. [Fi] and [Fl]. Prove that g is Riemann integrable.c. d ) .46 Theories of Integration A function f defined on an interval I is said t o be absolutely integrable over I if both f and If1 are integrable over I . S is not a vector lattice of functions. The previous example shows that the CauchyRiemann integral admits conditionally integrable functions whereas Corollary 2. Prove that xI = d . Can this statement be changed to a countable number of points? Exercise 2.
b] such that f (c) > 0 . Exercise 2.b]+ R. = 0. Prove that cucp. If there is a c E [a. Show that the upper and lower indefinite integrals are differentiable a t x with derivatives equal to f ( x ) .7 Let f : [u.Riemann integral 47 Basic properties Exercise 2.9 Let f : [a.a] + R be Riemann integrable. Suppose there are partitions P and P‘ such that L (f. a < c < b. Exercise 2.P ) = U (f. (1) Prove the following results for upper and lower integrals: (a) (b) c() and and and (2) Give examples to show that strict inequalities can occur in part (1. g : [a.8 Suppose that f . .finite integrals of f by 7 x )= ( ( t )dt and (x)= f b f ( t )dt.b]+ R. cp V $J. (1) If f is an odd function.1] + R is continuous. Define the upper and lower inde. Prove 7:f that F and satisfy Lipschitz conditions.1] $ R such that is Riemann integrable but f is not Riemann integrable. prove that J_”. Exercise 2.u).5 Suppose that f is continuous and nonnegative on [a. Exercise 2.24.b]. Prove that f is Darboux integrable.P’). Show that lim 1200 If1 1’ f (xn)dx = f ( 0 ) . prove that f > 0. cpq. f (2) If f is an even function. Exercise 2.12 Give an example of a function f : [0. a + Q E R. u ] . We say that f is an odd function if f (x) = f (2) for all x E [a.6 Let f : [a.and cp A $J are step functions.10 Let cp and $Jbe step functions and cp $J. and Q 2 0. Exercise 2.” Exercise 2. Suppose that f is continuous a t z. Darboux’s definition Exercise 2.b] +R be bounded.11 Prove Corollary 2. prove that J f a f = 2 f. s. a] and we say that f is an even function if f (x) = f ( x ) for all x E [a.13 Suppose that f : [0.
differentiable on (0.b ] . If f is also nonnegative.17 Suppose that f : [a. give a geometric interpretation of this result. then there is a c E [a. prove there is a c E [a.16 Suppose f : [a.b]+ R and I .Prove that Exercise 2. Prove that c [a. If f : [a.b] + R is nonnegative and Riemann integrable on [a.a). I f f : [a.a ] .a].b] + R is continuous and g : [a.a].15 Let f : [a.b].b]+ R is continb uous.14 Suppose that f : [a.f + Fundamental Theorem of Calculus Exercise 2. Exercise 2.19 Prove the following version of the Mean Value Theorem.18 Prove the Mean Value Theorem. Exercise 2.b] such that Ja f = f ( c ) ( b .b]. and f (0) = 0.b] R is Riemann integrable and m 5 f(x) 5 M for all x E [a. Define g by for x E [O. J disjoint interiors.b] + R and suppose that f is Riemann integrable on [a.b] + R is continuous and nonnegative and set M = s u p { f ( t ) : a 5 t 5 b}.1 =Lf+S.b]are intervals with L". (1) Prove that g = 0 on [O.b] such that Exercise 2. a ) . . Prove that f is Riemann integrable on [a.b].20 Suppose that f is continuous and strictly increasing on [0. Show Exercise 2.c] and [c.48 Theories of Integration Exercise 2.
] Exercise 2. (3) Deduce Holder’s inequality: If a . and cp is Riemann integrable.b] + R be Riemann integrable and suppose m 5 f (x) 5 M for all x E [a. then (2) If If is Riemann integrable. n ) = 1 and cp (2)= Oif x = O 1 if O < x 5 1 ’ Note that f is continuous a.b] has outer Jordan content 0 and S C [ ~ . M Prove that cp o f is Riemann integrable. Give an example } of a countable subset of [0.is Riemann integrable.T c [a.= 1. Suppose cp : [m. ab 5 la f (x)dx + 1 b f’ (x)dx. then . show that E(S U T ) 5 C(S) E(T).b]. then ab<+P where p up bp’ P ” > 1 and . Exercise 2. b 2 0. b ]show E ( S U T ) = C(S). c Prove the following two statements.1] by x is irrational E Q and ( m .b] + R be Riemann integrable. show C(S) 5 F(T).b]. If S. . Exercise 2. x2 (2x3 + 16)1’2 dx.Riemann integral 49 (2) Use this result to prove Young’s inequality: for 0 < b 5 f ( a ) .26 that S = {5 Show that a finite set has outer Jordan content 0.25 Show that the composition of Riemann integrable functions is not necessarily Riemann integrable.+ .23 Let f : [a. > 0.1] has outer Jordan content 0.21 Exercise 2. Let p . (1) If f 2 0.22 Evaluate Evaluate :s cos 29 sin 39d9 and sf x2exdx. ] + R is continuous.If T c [a.e. 1 I 2 c > 0 .24 Exercise 2. Show :k E N c [0. Characterizations of integrability Exercise 2. fP f Let f : [a.1] with positive outer Jordan content. [HINT: define f and cp on [0.b]. P PI 1 1 Exercise 2. + .27 If S c T c [a.
b] + R be a bounded function. prove that b b g is Riemann integrable and g = f.28 Let f : [a.32 Let f : [a. (c.34 Determine whet her the following improper integrals converge or diverge: dx (2) (1) s l 4 (3) (5) (7) (9) (11) JiIn z d x xdx (X+2)”X+ 1) (4) (4) J ” : tan x d z (6) (8) (10) (12) .33 Prove that every subset of a null set is a null set.d ] .z) 2 E} and let ( c . D.b]+ R and let D (f)= {t E [a.b] be an interval.b]+ R be bounded. b] Prove that D ( f ) = U { D .d ) ) 2 E . Prove that a countable union of null sets is a null set.29 Suppose that f. : f is discontinuous a t t } . (f) : E > 0) = U {Di/n (f): n E N}. Prove that w (f. Show by example that we cannot replace the open interval (c. (f) = {z 6 [a. and only if. ( f ) f l ( c . sa Exercise 2.g : [a. Prove that f is continuous at x E [a.31 Let f : [a. w (f.x ) = 0.b] + R are bounded and f is Riemann integrable. Exercise 2.b]: w (f. d ) c [a. Improper integrals Exercise 2. Exercise 2.b] if. d ) # 8.30 Let f : [a.d ) by the closed interval [c. Prove that f is Riemann b integrable and f = 0.b] + R. If f = g except on a set of outer Jordan content 0. Suppose that D.50 Theories of Integration Exercise 2. Suppose that f = 0 except on a set of outer Jordan content 0. sa sa Exercise 2. Exercise 2.
00). Prove that for any a .b].36 Formulate and prove an analogue of the Comparison Test.b] + [u. ' (2) r (x 1) = x r (x). r ( n )= ( n .Prove that Iw is Riemann integrable over [f Exercise 2. (3) For n E W.oo) which is not CauchyRiemann integrable but such that the Cauchy principal value integral of f over (m. the CauchyRiemann integral of f is independent of the cutoff point a.l)!.Riemann integral 51 Exercise 2.35 Suppose that f : [u.38 Suppose that f is CauchyRiemann integrable over (00. Exercise 2.39 Give an example of a function f defined on (oo. Hence. . for improper integrals over [a.b].37 = )lT+ I' f. 00) exists. Define the gamma function for x > 0 by Prove the following results for r: (1) The improper integral defining I converges.48. Proposition 2. b E Iw. + Exercise 2. Exercise 2.
This page intentionally left blank .
A convergence theorem for an integral concerns a sequence of integrable functions { fk}E1 which converge in some sense. it also has several shortcomings. to a limit function f and involves sufficient conditions for interchanging the limit and the integral. The definition and derivation of the important properties of the Lebesgue integral on the real line. measure and measurable functions. including the Monotone and Dominated Convergence Theorems. in which the functions are uniformly bounded. that is to guarantee limk J f k = J limk f k . In modern integration theories. This example is used to motivate the presentation of Lebesgue’s descriptive definition of the integral that bears his name. are then carried out. we have the following basic convergence result.Chapter 3 Convergence t heorerns and the Lebesgue integral While the Riemann integral enjoys many desirable properties. As was pointed out in Chapter 2. and the Bounded Convergence Theorem. This leads to a discussion of outer measure. Another serious drawback which we will address in this chapter is the lack of ‘(good” convergence theorems for the Riemann integral. the standard convergence theorems are the Monotone Convergence Theorem. We begin the chapter by establishing a convergence theorem for the Riemann integral and then presenting an example that points out the deficiencies of the Riemann integral with respect to desirable convergence theorems. in which the functions converge monotonically. The Lebesgue integral on ndimensional Euclidean space is discussed and versions of the Fubini and Tonelli Theorems on the equality of multiple and iterated integrals are established . 53 . one of these shortcomings concerns the fact that a general form of the Fundamental Theorem of Calculus does not hold for Riemann integrable functions. such a pointwise. For the Riemann integral.
fk : [a. t p ) } i=l { { 1> nj :: ).b]. Since f~ is Riemann integrable.2.Pl.b].b] with mesh less than 6 and let ti') be corresponding sets of sampling points. j = 1.">"' i=l ) .for k E W.b] + R . Thus. S (9) = j S ( g .sj (fK)I 5 / \ r cIf nj ( t y ) . S (f). Pj.54 Theories of Jntegration Theorem 3. Suppose that each fk is Riemann integrable and that the sequence { k.fK ( t y ) (xp nj I  2Fl) i=l E 3(bE while the middle term is less that . < 3 ( b E.{t. <E .b] and b lk m l f k = [ f = l i limfk.{tz(2)}n')I s i=l so that f is Riemann integrable. it is enough to show that the partial sums for f satisfy the Cauchy criterion. by the uniform convergence. To prove that f is Riemann integrable. If are partitions of [a. f i s Riemann integrable over [a. P 2 . then { are sets of sampling Let PI and P2 be partitions of [a.U ) for k > N and all x E [a.s ( f .1 Let f.? } .f k ) I(.b] with p ( P j ) < S and points relative to P j . 3 1 (f. Then. Fix a K > N.by the choice of K . converges to f uniformly o n [a. the partial sums for f~ satisfy the Cauchy criterion. we have I. k (34 Proof. so that there is a S > 0 so that if Pj. Set Then. For the first and third terms. Fix E > 0 and choose an N E N such that (x). f .
” f = limk Jab f k .1]. The second assumption is to require that for each x. each f k is Riemann integrable.the sequence { f k is increasing and bounded by 1. However. The first is a uniform boundedness condition in which it is assumed that there exists an M > 0 such that lfk (z)1 5 M for all k and z.f k (z)I < for k > N ba E and x E [a. $ (z)}Z. so (3.7 which is not Riemann integrable.1) fails to hold.converges monotonically to f (2).1 is quite strong. . Jab f = limk s. + In addition to the assumption of pointwise convergence. l / k ] ( X ) . For each x E [0. pointwise convergence will not suffice for (3. Example 3. a theorem with this hypothesis is referred to as a Bounded Convergence Theorem. the sequence { f k (z)}T=. for all k >N. and it would be desirable to replace this assumption with a weaker hypothesis.42.define fk : [0. I] R by f k (rn) = 1 for 1 5 n 5 k and fk (2)= 0 otherwise.a theorem with this hypothesis is referred to as a Monotone Convergence Theorem.1].1] R by f k (z) = k X ( O . By Corollary 2. it should be noted that.1) to hold. The following example shows that neither the Bounded nor Monotone Convergence Theorem holds for the Riemann integral.3 Let { T ~ } : = ~ be an enumeration of the rational numbers in [0. there are two natural assumptions which can be imposed on a sequence of integrable functions as in Theorem 3.1. Note that the sequence in the previous example does not satisfy either of these hypotheses. s. Example 3. Then.” f k . 1 { f k } E 1 converges pointwise to o but Jo fk = 1 for every k . 0 The uniform convergence assumption in Theorem 3. For each k E N. fix E > 0 and.b]. by uniform convergence. We will see later in this chapter that both the Monotone and Bounded Convergence Theorems are valid for the Lebesgue integral. The sequence { f k } r = 1 converges to the Dirichlet function defined in Example 2.2 Define f k : [O. Thus.Lebesgue integral 55 To see that choose N E Then. N such that If (z) . in general. We will show in Chapter 4 that both theorems are also valid for the HenstockKurzweil integral.
) He wished to assign to each bounded function f defined on a finite interval [a.” g (2) dx. f (2) dx. s. that satisfied six conditions. 3.J. Lebesgue (18751941) defined Ze probkme d’intkgration (the problem of integration) as follows. denoted by S.1 Lebesgue’s descriptive definition of the integral H. in which an object is defined and then its properties are deduced from the definition. notice that setting cp = f in (3) shows that b (f) (x)dx = . If f 2 g. Hence. like Riemann’s.” s. (See [Leb. It is desirable that the integrability of the limit function be part of the conclusion of these results. we wish to determine other properties of this integral. To begin. He called this definition descriptive. b. 11. c. but both of them require one assume the Riemann integrability of the limit function. We begin by considering Lebesgue’s descriptive definition of the Lebesgue integral.” so that f (x)dx 2 ity condition. Vol. See [Lewl]. Then: (1) (2) (3) (4) If (5) (6) If and then increases pointwise to f then In other words. he described the properties he wanted this “integral” to possess and then attempted to deduce a definition for this integral from these properties. Suppose that a . then (3) and (4) imply that s.b] a number. In the remainder of this chapter. this integral satisfies a monotonic .”f (x) dx. we will construct and describe the fundamental properties of the Lebesgue integral. Assuming these six conditions. h E R. to contrast with the constructive definitions.56 Theories of Integration It should be pointed out that there are versions of the Monotone and Bounded Convergence Theorems for the Riemann integral.page 1141.
b E R. lr ldx 5 q . ~ I . for any interval [a.a = d . setting c = b.so that J/yb ldx + 1' ldx lc ldx. b E R. 1 ' so that ldx = r . by setting c = b = a . = Iterating this result shows l:' + lr ldx ldx + . ] 5 by lq lr 1" 1' ldx ldx 5 Idx 0 5 q ldx = q .s.Lebesgue integral 57 We next show that ldx = b . Letting p and q approach r . by iteration. we get s.b]of length 1. Thus.5 x[o." f (x)dx. by iteration. we conclude that for all real numbers r . F'rom (2). while setting ai = sd/" ldx = l/n.a for all a . Again. then.a for all a . if r E Iw. . + = lan 1. ldx = n1 ldx." f (x)dx = . F'rom (l). since x [ ~ ."f(x)dx = 0. from ( 5 ) . Setting ai = i yields ldx n.p . . Ids Setting 'p = f in (3). let p and q be rational numbers such that p < r < q.p . we see that n i shows that 6 ' 05 which implies ldx = q for any rational number q. s." sc [Idx = 1.we see b d that b .c implies la = ldx ldx. Finally. b la = b . Then. we see that J. .r~X [ ~ . monotonicity. we see that .
x1. Let P = {xo.” f (2) dx. . xn} be a partition of [a.b]. which in turn implies that this equality holds for any integer n.” Let Now. We have not made use of the crucial property b s . Setting f = cp = 0 shows that Ja Odx = 0. . from properties (3) and (4).a ’b and the integral is linear. (6)Suppose we have an integral satisfying properties (1)through (5) and let f be Riemann integrable on [a. note that since both f and f are bounded by monotonicity implies that f (z) dz I If\. see that this integral must be linear. .b].58 Theories of Integration b for every natural number n. To see that this equality holds for any real number. we By using properties (1) through ( 5 ) . Since it follows that for any rational number q. Let and choose a real number Then. Recalling the definitions and we see . we see that rf (2) dx = r s. such that Letting q approach r . Hence. . we have shown that J ldx = b . fix a real number r. s.
Then. Vol.d]. if f is Riemann integrable on [a.. Thus. b ] .. . . 2 .Lebesgue integral 59 which implies that Thus. by the linearity of the integral.n. then J b f = f and the middle integral must equal the Riemann integral. which is the length on the interval [c. . { ( P ~is }a sequence of simple functions that increase monotonically to f . Fix 1 and L such that 1 5 f < L.bxE( x ) d x = JCdldx. it is enough to be able to integrate the b functions cpk. .):=. . Consequently. it suffices to know how to calculate the integrals of functions 1c) which take only the values 0 and 1” [Leb. x E i (2) z d s.. page 1181.1 ) for k = 1 . Lebesgue reduced the problem of integration to that of extending the definition of length from intervals in R to arbitrary subsets su . Given a partition P = { l o . by (6) {(P~}T Thus.then S. (2) p ( P k ) 5 i p ( P k . . In fact.In} of the interval [1. 0 5 f . fix a partition Po and define a sequence of partitions {P.n. In = L and li < li+l for i = 1 .b] and. to evaluate the integral o f f . . . .cpk < p ( P k ) and p ( P k ) + 0. by construction.b] + R is bounded. let Ei = { X E [a. . 11.1 for k = 1 . Let cpk be the function associated to Pk as above. . which in turn depends on integrals of the form xE (x)d x . Suppose that f : [a.b].d] is an interval in [ a . As Lebesgue said. If E = [c. b li1 S. 2 .” (XId x = Now. .b]: li1 5 f ( x ) < l i } for i = 1 .. . and consider the simple function cp defined by n i=l 7 : It then follows that cp 5 f on [a. .L]with lo = 1. . such that: E= . We now investigate property (6). “To know how t o calculate the integral of any function. (1) Pk is a refinement of P k . so that converges to f uniformly on [a.b ] . T h i : any integral that satisfies properties (1) through (5) must agree with the Riemann integral for Riemann integrable functions. . 11. .
4 E . (3) the measure of the set [0. we use which m a y be finite or countably infinite. His goal was to assign to each bounded set E c R a nonnegative number m ( E ) satisfying the following conditions: (1) congruent sets (that is. I] is 1. c) for every E > 0. It follows that m* (0) = 0. m* ( E ) . 0 to represent a countable set. 3. with a function that preserves properties (1) through (3) of the problem of measure.1 Outer measure We first extend the length function by defining outer measure. to le probkme de la mesure des ensembles (the problem of the measure of sets). we see m* ( E ) 2 0. this problem has no solution. We define the (Lebesgue) outer measure of where the infimum is taken over all countable collections of open intervals { I j } j E . We show that m* extends the length function and establish the basic properties of outer measure.t ( I ) > 0 for every open interval I .10. As we shall see below in Remark 3. 3.such that E c ujEffIj.2 Measure Our goal is to extend the concept of length to sets other than intervals.5 Here and below. Definition 3. Given a set E c R and h E R.by Let E c R. that is.2. we define the . Since . and. Since 0 c (0. translations of a single set) have equal measure.60 Theories of Integration of R. Notation 3. 0 5 m* (0) 5 E for all E > 0. (2) the measure of a finite or countably infinite union of pairwise disjoint sets is equal to the sum of the measures of the individual sets (countable additivity). It follows immediately from the definition that my (0) = 0.
m* ( I ) 2 m* ( J ) = ! ( J ) > ! ( I ) . We claim that ! Ij)2 b . pages 3193221. let I c R be an interval with endpoints a and b. suppose that I is a bounded.b k ) : k = 1 . then [a. . the two are equal. ai < bi1 < bi and bn > b. For any E > 0. + Thus. Let { I j : J’ E a } be a countable cover of I by open intervals. Since ( I c U g l J i . .a . If F c E . We say a set function F is translation invariant if F ( E ) = F ( E whenever either side is defined. .a 2 ~ Hence. ( a . If bl <_ b.b] c Jil and since ! ( J i ) 2 P ( J i . If bl > b. + + xjEu EL. + h) Theorem 3. (See [BS. . Continuing this construction produces a finite number of intervals { J i . that is.a ( which will establish that m* ( I ) = ba. We leave (2) as an exercise. if 0 is a countable set and Ei f o r all i E a . = ( a k . a finite number of intervals from { I j : j E a} cover I ..n } from { Ji : i = 1. m}. (2) m* is translation invariant. Thus. Then.E . m* ( I ) = ! I ) . m* ( I ) 5 b . . b E ) is an open interval containing I so that m* ( I ) 5 b . n i=l k=l n k=2 as we wished to show. there is an i2 such that Ji2 = (a2. m } such that a1 < a . call this set { Ji : i = 1. m* ( I ) 2 ! ( I ) . which implies (1).. closed interval. See Exercise 3.) It suffices to show that EL1! J i ) 2 b . . then every cover of E by a countable collection of open intervals is a cover of F . ( If I is a bounded interval. ( (4) m* i s countably subadditive.Lebesgue integral 61 translation of E b y h to be the set E + h = {z E R : J: = y + h for some y E E } . . then m* (UiEaEi) CiEu (Ei). . there is an i l such that Jil = ( a 1 . . Since I is compact. i f F cR Proof. (3) i f I i s a n interval then m* ( I ) = ! I ) . that is. Thus. . b l ) with a1 < a < b l .1.E . .a.. and by the remark above. then for any E > 0 there is a closed interval J c I with [ ( I ) < [ ( J ) E . 5 m* (1) m* is monotone.a. To prove (3). we are done. Now. b2) and a2 < bl < b2. ) 2 b .6 T h e outer measure m* satigfies the following properties: c E c R then m* ( F ) 5 m* ( E ) .
(2) [0. it follows that m* ( U i E a E i ) 5 CiE0 (Ei). x + E ) for every E > 0.62 Theories of Integration Finally. 11. outer measure is countably subadditive.Ea.2.1]. + + T E Q o = . we get = 0 for all Corollary 3. we choose a set P c [0. We shall show in Section 3. we see that m* ((2)) x E R. uieo {Ii.y E [0.2.2 ~ . choose a countable collection of open interval {Ii. m* ( I ) 2 r for every r > 0 which implies m* ( I ) = 00. We say that x y if x . then m* ( E )= 0. where the union is taken over all Qn [I. As a consequence of this corollary.njnEai such that C n E a i t(Ii. Then. Since E was arbitrary.1] which contains exactly one point from each equivalence class determined by w . By the Axiom of Choice. By the subadditivity of outer measure.r E Q and q # r then (P q) n (P r ) = 0.3 that the converse of Corollary 3. Example 3. Let z. is a countable collection of open intervals whose union contains UiE. The outer measure we have defined above is defined for every (bounded) set E c R and satisfies conditions (1) and (3) listed under the problem of measure. Assume first that m* ( E i ) < 00 for all i E CT and let E > 0.E . then I is uncountable. Hence. For each i. We need to make two observations about P : N (1) if q. then EiEa ( E i ) = 00 and the inequality m* follows. m* if m* (Ei) = 00 for some i. 0 Since x E (x . as the following example shows. It remains to prove (4). Unfortunately.7 If E cR is a countable set.1] c U ( P + r ) .y E Q.1].n)< m* ( E i )+ 2 . if I is an unbounded interval. we obtain Corollary 3.8 If I is a nondegenerate interval.9 We begin by defining an equivalence relation on [0.Ei. there is a closed J c I with m* ( J ) = l? ( J ) = r. but not countably additive. Thus. then for every r > 0. by Exercise 3. Finally.n}.7 is false.
t E P . or infinity.2]) 5 3). if m* ( P ) = 0. and finite subadditivity (to show m* ([1.E Q and since x.s = T.t = r . Since monotonicity is a consequence of *finite subadditivity.1] and x E P r. Note that UrEQ". this violates the definition of P . So. N + + N and 0 < m* (U. so there is an s E P such that z s. which implies + + so that the sum is either 0. s t.fined o n all subsets of R that is translation invariant. and m* ([0. ( + + The #first follows f r o m translation invariance. m* ( P r ) = m* ( P ) for any r E R.Lebesgue integral 63 To see ( I ) .s E [0.q # 0. by translation invariance. z . z is in one of the equivalence classes determined by N .2]. I n the previous construction we have used the following facts to show that outer measure is not countably additive: ( I ) m* ( P r ) = m* ( P ) . The second uses m* ([0. proving (I). We will return to this example below. there is no function de. For (2). suppose that z E ( P q ) n ( P + r ) . Remark 3.10 This example shows that there i s no solution to Lebesgue's problem of measure. if m* ( P ) > 0.. so by monotonicity.( P + r ) c [1. In either case.1]) = 1. Then. r E [1. the only properties we used were translation invariance. and since r . monotonicity. *finitesubadditivity. countably additive and equals 1 o n [O. t E P such that z = q + s = r + t.q G Q. 1 . this example applies to any function satisfying these three properties.1]. Then. translation invariance and countable subadditivity. let x E [0. (2) O < m* (urEQo P r ) ) < 00. ( P r ) ) < 00. Thus. This implies that s . there exist s .=Q.1]. On the other hand. Thus. 1 . so that outer measure is not countably additive.1]) = 1. Since s .
bounded interval I = [a. Thus.11 are equivalent: Let I cR be a bounded interval. we define the distance from I to J by ~ ( I . The proof will be based on several preliminary results. Lebesgue defined a subset E c I to be measurable if m* ( E ) = m. Carathbodory’s idea was to test E with every subset of R. instead of just an interval containing E. + (2) m* ( A )= m* ( An E ) + m* ( A\ E ) .2 Lebesgue Measure Example 3. for all A c I . Given intervals I . . there is a characterization of Lebesgue measurable subsets of an interval I due to Constantin Carathkodory (18731950) that generalizes very nicely to arbitrary subsets of R. the following (1) m* ( I ) = m* ( E ) m* ( I \ E ) .a ) m* ( I \ E ) . y E J } . J= inf { ( z . Lebesgue worked on a closed. J c R. Unfortunately. he defined the inner measure of E to be m. and for E c I . ( E ) = ( b . The members of this subset were called measurable subsets by Lebesgue. we restrict the domain of m* to a suitable subset of the power set of R.9 shows that m* is not countably additive on the power set of R. that is.2. In order to obtain a countably additive set function which extends the length function.b]. this procedure is not meaningful if we want to consider arbitrary subsets of R since the length of R is infinite. so that E = E n I .64 Theories of Integration 3. ( E ) . the inner measure of E is the length of I minus the outer measure of the complement of E in I . Using the definition of inner measure and the fact that the outer measure of an interval is its length. we assume that E c I . If E c I . However. Theorem 3. he was led to consider the condition + m*(A)=m*(AnE)+m*(A\E) for every subset A c R.y l : z E I . A need not even be a measurable set! We now show that the two conditions are equivalent. ) We begin by proving that outer measure is additive over intervals that are a t a positive distance. Lebesgue’s condition is equivalent to m* ( I )= m* ( E ) m* ( I \ E ) . In the above equality.
fix 6 > 0 and choose a countable collection of open {Ii}iEosuch that IU J c UiEgIiand ! I i ) 5 m* ( I U J ) + E . Suppose a E Ii = (ai. If any of the intervals are closed. 0 Remark 3.Ui) = e (It:) + e (I:) . . ( I U J ) n Ii = ( I U J ) n (It: U I:).Lebesgue integral 65 Lemma 3. m* ( I U J ) 5 m* ( I ) m* ( J ) . iEU Since this inequality is true for any E > 0. J or I U J . Let a ( I ) = {i E a : Ii n J = 8 ) and a ( J ) = {i E a : Ii f l I = 8). and. The next result shows that condition (2) of Theorem 3. the proof is complete. It follows that I c UiEg(l)Ii and J c UiEU(j)Ii.13 This result is true f o r intervals whose interiors are disjoint. + Proof.12 Let I . By subadditivity. Let It: = (ai. we can replace them by their interiors. J c R be bounded intervals such that the distance f r o m I to J is positive. Then. If the intervals are open. opposite intervals Assume.a ) + ( a .ai = (bi . Lemma 3. ( without loss of generality. so that It: U I: covers the same part of I U J as Ii does. Assume that every interval Ii that contains Q is replaced by the two intervals It: and I:. that I lies to the left of J and let + xiEg 2 be the point midway between the two intervals. replacing Ii by It: and I: does not change the sum of the lengths of the intervals. since a= sup I + inf J e ( I i ) = bi . which does not change the measure of I .14 then If I cR is a bounded interval and J cI i s a n interval. r n * ( I ) + m * ( J )I iEU(I) C(Ii)+ i E u (J ) e(&) L x t ( I i ) L m * ( I U J ) + E .a ) and I: = ( a . Then. m* ( I U J ) = m* ( I ) m* ( J ). this proof works whether the intervals touch or not. since a $ I U J . m* ( A ) = m* ( A n J ) + m* ( A \ J ) for all A c I .11 holds when E is an interval.bi). To show the inequality. since the edge of a n interval i s a set of outer measure 0.bi) for some i E a. Thus.
Further. A and A n J are both intervals. the equality is merely the fact that the length function is additive over disjoint intervals.Therefore. Choose a countable collection of open intervals {Ii}iEasuch that A c uiE. with A1 and A2 intervals at positive distance and use the previous lemma. so their outer measures equal their lengths. we write A \ J = Al u A2. m* ( E ) = m* ( E n J ) Proof. A \ J is either an interval or a union of two disjoint intervals which are a t a positive distance S. the result follows by countable 0 In the following lemma.66 Theories of Integration Proof.Ii and CiEo ( I i ) 5 m* ( A ) E . and m* ( I \ E ) = m* ( ( I \ E ) n J ) + m* ( ( I \ E ) \ J ) . we show that condition (1) of Theorem 3. Note first the conclusion holds if A is an interval in I .11. In this case. In the second case. + m* ( A n J ) + m* ( A\ J ) 5 m* ((uiEoIi) J ) + m* ((UiEoIi) J ) n \ C [m*( A n J ) + m* = Cm* F i€U ( ~ i \ 41 ( ~ i ) 5 m* ( A )+ E . Let A c I and E > 0. As before. By condition (1) and subadditivity. then m* ( J ) = m* ( J n E ) for all intervals J + m* ( J \ E ) + m* ( E \ J ) cI. By the previous lemma. for any interval J C I . Since this is true for all subadditivity. E iEU > 0. we see m* ( I ) = m* ( E )+ m* ( I \ E ) = m* (En J ) +m* ( E \ J ) +m* ( ( I \ E ) n J ) +m* ( ( I \ E ) = {m*(En J ) +m* ( ( I \ E ) n J ) } \J) + {m*( E \ J >+ m* ((1\ E ) \ J ) ) . Lemma 3.11 implies condition (2) when A is an interval. In the first case. ! m* (Ii) = m* (Iin J ) + m* (Ii\ J ) for all i E 0.15 If E cI satisfies condition (1) of Theorem 3.
m* ( An E ) + m* ( A \ E ) 5 m* (A) and the proof is complete. we see that (2) implies (1).l(Ii) 5 m* (A) E . Let A c I and note that by subadditivity. + Thus.Ii and Ci. m* ( J ) m* ( I \ J ) = m* ( E n J ) + + m* ( E \ J ) +m*( ( I\ E ) n J ) + m* ( ( I \ E ) \ J ) . Setting A = I .. m* ( I \ J ) 5 m* ( E \ J ) +m*( ( I \ E ) \ J ) . and ( I \ E ) \ J = ( I \ J ) \ E . 0 Thus. E \ J = ( I \ J ) n E . it is enough to prove that m* ( A n E ) + m* (A \ E ) _< m* (A) . for subsets of bounded intervals. Then.rr. which implies m* ( J ) 2 m* ( E n J ) + m* ( ( I \ E ) n J ) = m* ( E n J ) + m* ( J \ E ) . Fix E > 0 and choose a countable collection of open intervals {Ii}iEasuch that A c UiE. assume that (1) holds. by the previous lemma. Proof. . In order to include unbounded sets.11. So.Le besy. it follows from subadditivity that Thus. 0 and the proof now follows by subadditivity. We can now prove Theorem 3. we adapt Carat heodory’s condition for our definition of measurable sets. measurability according to Lebesgue’s definition is equivalent to measurability according to Caratheodory’s definition. in t eyral e 67 Since ( I \ E ) n J = J \ E . By subadditivity.
The restriction of m* to M is referred to as Lebesgue measure and denoted by m = m * J M . Proposition 3. Note that singleton sets are null. . Let A If m* ( E )= 0 then E is measurable. By subadditivity. and countable unions of null sets are null. By monotonicity.17 Let M be the collection of all Lebesgue measurable sets.18 The sets 0 and R are measurable. c R. Set EC= R \ IE. If A c R.2) is the same as m* ( A ) = m* ( A n E ) + m* ( An E") . We next study properties of m and M . then (3. . An immediate consequence of the definition is the following proposition. Note that condition (3.2) The set A is referred to as a test set for measurability. Thus.4.20 Let E l .68 Theories of Integration Definition 3.En be pairwise disjoint and measurable sets. we need only show that m* ( A n E ) + m* ( A \ E ) 5 m* (A) in order to prove that E is measurable. A subset E c R is Lebesgue measurable if for every set m* ( A ) = m* ( A n E ) + m* ( A \ E ) .16 A c R. then m ( E )= m* ( E ) . subsets of null sets are null.2) follows from subadditivity. See Exercise 3. Further. m* ( A ) 5 m* ( A n E ) and E is measurable. 0 5 m* ( A n E ) 5 m* ( E ) = 0. . We observe that we need only consider test sets with finite measure in (3. if E E M .19 Proof. Proposition 3. (3. Definition 3. then m* ( A n u ~ = ~ = ~ ) E j=1 Em*( A n E j ). n . sets of outer measure 0 are measurable. + m* ( A \ E ) = m* ( A \ E ) 5 m* ( A ) 0 We say that a set E is a null set if m ( E )= 0.2) since if m* ( A ) = 00. . Lemma 3. Thus.
Suppose that E and F are measurable. In fact. by the induction hypothesis. Note that p(X) is an algebra and so is the set ( 0 . Let A c R. Proof.Ei) n En = A 0 En and A r) (UYElEi)\ En Since En is measurable. Proceeding by induction. l ) is a countable union of singleton sets.Lebesgue integral 69 Proof. each of which is in A. l ) \ Q is finite. A is closed under finite unions and intersections. We call A an algebra if A .Ei)) n1 i=l =m*(AnE. and. note that Qn( 0 . X ) . We want to show that M is a aalgebra. Then. Then. =A n (U7g. l ) \ F is a finite or empty set}. first using the measurability of F then the .Ei). For A c R. every algebra contains 0 and X since A E A implies that X = A U A" E A and 0 = X" E A. Assume it is true for n. so if E is measurable then so is EC. Theorem 3. l ) : F or ( 0 . Since the Ei's are pairwise disjoint) A n (Uj"=. Example 3.)+~m*(AnEi) n i=l 0 Let X be a nonempty set and AAC p(X) a collection of subsets of X . A is a n algebra (see Exercise 3. we see that the Caratheodory condition is symmetric in E and E".5) which is not a aalgebra. we note that this statement is true for n = 1.1sets. To see that A is not a aalgebra. We need to prove two things: M is closed under complementation.22 The set M of Lebesgue measurable sets is a n algebra. A" = X \ A E A. M is closed under finite unions. An algebra A is called a aalgebra if it is closed under countable unions. l ) nor its complement (0. B E A implies that A U B . m* ( A n (uj"&i)) = m* ( A n En) + m* ( A n (u7. We first prove that M is an algebra.21 Let A = { F c ( 0 . but neither Q n ( 0 . Since A n E C= A n (R \ E ) = A \ E and A \ E" = A \ (R \ E ) = A n E . As a consequence of the definition and De Morgan's Laws.. write A n ( E U F ) = ( A n E ) U ( A n E" n F ) .
2 . We can now prove that M is a aalgebra. F n E M and F: 3 E". so for any index set a C N. Then. UiEOBi C UiEOAi.Ai. Proposition 3. . Then. Thus UZlAi C U z l B i so that the two unions are equal.1..P"i C M and set E = UZlEi. Bi E A. there is a sequence ( B i } z 1c M such that E = Ug1Bi and the Bi's are pairwise disjoint. . m* ( A n ( E u F ) ) m* ( A n ( E u F ) " ) = m* ( A n ( E u F ) ) +m*( An E" n FC) 5 m* ( An E ) + m* ( An E" n F ) +m*( An E" fl F") . Bi n Bj = 8.Bi. 0 4 Note that the proof actually produces a collection of sets { B i } z 1 satisfying UElAi = U c l B i for every N . By Lemma 3. If z E Bj.then x E A. Let x E UE. Thus. there is a collection (Bi}:. c A of pairwise disjoint sets so that UZlAi = U z 1Bi  + CI Proof.m* ( A n E ) +m* ( A n EC) < = m* ( A ) . . Since M is an algebra.24 Proof. Bi c Ai for all i.70 Theories of Integration measurability of E .20. x Ai for i = 1 . . Clearly. Finally. Therefore. Then. m* ( An (UrZlBi)) = E m * ( A nBi) . Let {Ei}. Choose the smallest j SO that x E A.. it satisfies the following proposition. By the previous proposition. Set B1 = A1 and for j > 1 set Bj = Aj \ UiZ. Let A c R. E U F is measurable and M is an algebra.23 Let A be a n algebra of sets and {Ai}E1 C A. SO that x @ Bi c Ai \ Aj. We need to show that M is closed under countable unions.Ai. The set M of Lebesgue measurable sets is a aalgebra.j . We want to show that E E M . which implies that x E Bj C U z l B i . Theorem 3. Set Fn = UY=. fix indices i and j and suppose that j < i. i=l n . Since A is an algebra.
for any n E N.h ) \ E ) + h ) + m* ( ( ( A. = Ij n [b. + Without loss of generality. i= 1 Since this is true for any n. Since m* ( A )5 m* ( An I ) it is enough to show + m* ( A\ I ) 5 m* ( A l )+ m* (A2)+ m* (A3) m* (Ai) + m* (A2)+ m* (A3)5 m* ( A ) . We now show Proposition 3.lIj and Cj"=. Set I. We saw above the m* ( I ) = l ( I ) for every interval I that every interval is measurable.Lebesgue integral 71 Thus. + m* ( A )= m* ( A. and l ( I j ) = {Ij}El . = Ij n (00. we may assume that m* ( A )< 00. that is.h ) = m* ( ( A. ( I j ) 5 m* ( A ) E . by subadditivity we see that m* ( A )2 Em*( A n B i )+m* ( A nE") 2 m* ( A nE ) + m* ( A nE " ) . I. To see this. 0 A consequence of the translation invariance of m* is that M is translation invariant. a ] .A2 = A n I and A3 = A n [b. Therefore.m). i=l 00 Thus. if E E M and h E R. Then. m* ( A )= m* ( Af l Fn) n + m* ( A nF.25 Every interval I c R as a measurable set. then E h E M .a ] .)+m*(AnE") = E m * ( A n Bi) m* ( A nE") .") + Lm*(AnF.h ) \ E ) + h ) = m* ( An ( E + h ) )+ m* ( A\ ( E + h ) ) = m* ( ( ( A h ) n E )  which shows that E h E M. b ) . Fix E > 0 and let be a collection of intervals such that A c Ujoo. M is a aalgebra. Each I. + c R.h ) n E ) +m* ( ( A.00). let E E M and A c R. Fix a set A A n(m. is either an interval or is empty. E is measurable. cR and set A1 = Proof. = Ij n I l and I. Assume first that I = ( a .
= I . and Iy By the definition of the intervals I. U Iyis an open interval contained in 0 and containing both I.n.) = m* (1. be the largest open interval contained in 0 that contains x. UxEoI. = Iyor I . 0 c U x E 0 I x .)+ e (I. Thus. and so are their complements (CQ. For each n. (I. by taking complements. u +n) and (00. ) } j=l 00 5 m* ( A )+ E . j=l m* ( A l ) + m* (A2) + m* (4 m* ( A ) I so that I is a measurable set.”=. y E 0. c 0 for every x E 0. 0 is a union of disjoint open intervals.c 0. Since I .we have A. Thus. c U g l I 7 . we see that Lebesgue measure extends the length function to the class of Lebesgue measurable sets. Clearly. We next study the open sets in R and the smallest aalgebra that contains these sets. Definition 3.Thus. then either I.26 Let X c R. The next result shows that we can realize 0 as a countable union of open intervals. we see that I. n Iy= 0. all intervals are measurable. l3 (X) contains all the closed subsets of X.00). Let 0 c R be an open set.) +m* ( I . note that if I. which implies that m* (An) I C. Proof.)+m* (I..) + e (1.72 Theories of Fntegmtion k‘ (1. ( b . we get m* m* ( A i ) + m* (A2) + m* (A3) 5 C {m*(I. let I . U Iy= Iy.l ( a . For each II: E 0.). Since D (X) is a aalgebra that contains the open subsets of X. Since E is arbitrary.27 Every open set in R is equal to the union of a countable collection of disjoint open intervals. Since each of the intervals contains . Set Exercise 3. If x . nIy# 0. Let 0 be an open subset of R. ) .00) = U.) + m* (I. (a.6. u] and b) [b.) + m* ( I . are measurable.t”. Since every interval is either the intersection or union of two such 0 infinite intervals. Theorem 3.b) = U?=. Since M is a aalgebra. The collection of Bore1 sets in X is the smallest aalgebra that contains all open subsets of X and is denoted l3 (X). To see this. so that 0 = UxEOIx. then I.
we get Corollary 3. for example. and [Swl.is contained in the Borel sets and contains all the open and closed sets. (uyZlEi))= cy=l ( An Ei). Clearly.31 below. page 671. page 531.29 The two sets B(R) and M are not equal.30 Let { Ei}cl M . Remark 3. Let Fu be the collection of all countable unions of closed sets. that is. FUcontains all the closed sets. Consequently. UglEi E M and the inequality follows since it is true for outer measure. that is. We need to show that i=l By Lemma 3.b) = UKl [u bSimilarly. since. If the sets Ei are pairwise disjoint. m is translation invariant. M . pages 148149]. Next. ( a . we defined a collection of sets. Exercise 6. there are countably many distinct maximal intervals . SS.28 Every Borel set is measurable.Lebesgue integral 73 a distinct rational number. Then. 0 Thus. Assume the sets are pairwise disjoint.9. for all n. as we saw in Example 3. +k. [Ru. then m (U& Ei) = m (Ei). Also. Then. This function does not satisfy property (2). We now show that m satisfies property (2). page 541. f o r example [Ha. B (R) c M . there are Lebesgue measurable sets that are not Borel sets. It also contains all the open sets. Since M is also a aalgebra that contains the open intervals. F g c B (R). we can view B(R) as the smallest aalgebra that contains the open intervals in R. Since M is a aalgebra. m* . we have defined a nonnegative function m* that is defined on all subsets of R and satisfies properties (1) and (3) of the problem of measure. m c Ei) 5 m (Ei). and called m the restriction of m* to M .20. m* ( A n Then. note that M is a proper subset of p ( R ) as we show in Example 3. [Mu. the collection of all countable intersections of open sets. czl Proof. and satisfies properties (1) and (3). So far. k]. (Uzl cz. Let A = R. Proposition 3. m is countably additive. See.
# is a measure on the aalgebra p(X) of X. which also shows that p is monotone. In either case. B E 13 with A c B . (2) p is countably additive. not all sets are measurable. We use this identity in the following proof.33 Define the counting measure. p (Ei) = 00 and p (Ei) = 00 for some i are allowed. p ( B )= p ( A ) p ( B \ A ) . m defined on 13 (W Example 3. m ( U & E i ) = czlm ( E i )as we wished to prove. But. Suppose p is a measure on l3 and A .9 that 1 5 m ( U . #. 0 Thus. by setting # ( A ) equal to the number of elements of A if A is a finite set and equal to 00 if A is an infinite set. Then P + r would be measurable for all T E R and m ( P r ) = m ( P ) . (P then the right hand side equals 0. i=I for all sequences of pairwise disjoint sets {Ei}zl c t3 . + We saw in Example 3. then the right hand side is infinite. also. m solves le probl2me de la mesure des ensembles for the collection of measurable sets. the equality fails. Thus. A nonnegative set function p defined for all A E 13 is called a measure if (1) PL0) = 0. Then. Note that both C z . P is not measurable. that is.74 Theories of Integration Thus. Suppose that {Ei}Z1c B. by countable additivity. Thus. E ~ o + r ) ) 5 3. for any set X. if m ( P ) > 0 . If m ( P ) = 0. Suppose P were measurable.9 is not measurable. .31 The set P defined in Example 3.32 Let t3 be a aalgebra of sets. Example 3.34 Let p be a measure on a aalgebra of sets t3. + Proposition 3. Definition 3. Examples of measures include m defined on M and. Then. as the next example shows.
assume that Ei 2 Ei+l and there is a K so that ~ ( E K )00.where the sets on the right hand side are pairwise U disjoint. proving the proposition. The second characterization is the classical definition in terms of open sets. So..35 Let Ei = [i. Suppose first that Ei c Ei+l. Set < E = nF1Ei. m ( E i ) = E = n Z l E i = 0 has measure zero. If p (Ei)= 00 for some i. p ( E )= limn. Then. EK \ E = ( E K \ E K + ~ ) u ( E K +\~E K + ~ ) . Since the sets are increasing.00p (En). The following theorem gives four alternate characterizations of measurability. E = (El \ Eo) U (E2 \ E l ) U (E3 \ E2) U . Now. .Lebesgue integral 75 (1) If Ei c Ei+l. Set E = UZ.. a+m Proof. then. A common definition of measurability in Euclidean spaces is in terms of open sets. The one we have chosen is useful for generalizing measurability to abstract settings. Since the sets are decreasing. 0 Notice that we cannot drop the assumption in (2) that one of the sets Ek has finite measure. Let Eo = 0. Thus. and there is a K so that p ( E K )< 00.. Example 3. 00 for all i while There are many ways to define Lebesgue measurability.. then the equality in (1) follows since both sides are infinite. p ( u & E ~ ) lim p ( E i ).oo). then. assume p ( E i ) < 00 for all i. It follows that Thus. which is a union of pairwise disjoint sets. p ( nglEi) = Jim p (Ei).Ei. . = 200 (2) If Ei 2 Ei+l.
Then. G = nT=lGk is the desired set. G . we show that (1) is a consequence of (4). there is a closed set F c E such that m* ( E \ F ) < E . The other implications are similar. Then. (2) . 3 E such that m* (GI.k . (3) for all E > 0 . We first show that (1)implies (2). The set F = G" is the desired set.for all E > 0. This implies that E = G \ (G \ E ) = G n (G \ E)" E M . G is an open set containing E such that + Therefore. m* (G \ 8) m ( G \ E ) = rn ( G ). the open set G = UF=. Assume that E is a measurable set of finite measure. (4) there is a G b set G 3 E such that m* (G \ E ) = 0. 0 i. = If m ( E )= 03.Gk 2 E and since we have that my (G\E) 5 Em*(Gk k=l \Ek) < E E ~ E . Proof. (5) there is a n F0 set F c E such that m* ( E \ F ) = 0. The sfollowing are equivalent: (1) E E M . there is an open set G 2 E" with m* (G \ E") < E . since E \ F = E \ G" = G \ E".' .76 Theories of Integration Theorem 3.36 Let E c R.\ E ) < Then. To show that (1) implies (3). there is an open set GI.' = k=l as we wished to show. Finally. k ) and choose open sets Gk 2 Ek so that k m* (GI. note that E E M implies E" E M . Let G E G h be such that E c G and m* ( G \ E ) = 0.(3) and ( 5 ) are equivalent. there is a n open set G 3 E such that m* ( G \ E ) < E .G \ E E M which implies that ( G \ E)" E M . Set such k'(Ii) G = U i E a I i . To show that (2) implies (4). It remains to show that (l). set E = E n ( . observe that for each k. Thus.m ( E ) < E . as we wished to show. .\ Ek)< ~ 2 . Fix E > 0 and choose a countable collection of open intervals {Ii}iEO that E c U i E a I i and CiEO < m ( E ) E . Then.
For each E > 0. Then. The only properties used to show that equality were translation invariance (l). In fact. By the corollary.Lebesgue integral 77 Suppose that E is a measurable set.38 outer regular if p ( E ) = inf { p (G) : E A measure p defined for all elements of B (R) is called c G. and our measures are translation invariant and finitely additive. If we wish to generalize the concept of length to general sets. by translation invariance and countable additivity. By this observation and translation invariance.b ] ) . which implies m ( G ) = m ( E ) m (G \ E ) 5 m ( E ) It follows that + +E. an interval with l ( I )= 1. m ( E )= inf { m(G) : E c G. . in fact. many more sets). We claim that p ( E ) = cm ( E ) for all E E B (R). there is an open set G 2 E such that m (G \ E ) = m* (G \ E ) < e. Theorem 3. p ( [ a b ] )= u ( ( a . G open} for all E E B(R). p ( [ a b ) ) = p ( ( a . a repetition of the argument proving s . l ) ) = c = c m ( I ) . every Borel measure on R is outer regular. finite additivity (2). Corollary 3. p ({z}) = 0 for all x E R. 1 and Ida: = 1 (3).37 Let E cR be a measurable set. page 641. then p = cm f o r some constant c. Since p is finite on bounded intervals.G open}. (See [Swl.b ) ) = . Remark 7. . In fact. we need a function that is defined on all of the Borel sets (and. It is b l d x = b . then p ( I )= p ( ( 0 .a for all a .39 If p is a translationinvariant outerregular Borel measure. Thus. b E R from Lebesgue’s descriptive properties of the integral. We will show that every translation invariant Borel measure is a multiple of Lebesgue measure. Lebesgue measure restricted to the Borel sets is an outer regular measure. we have already seen the proof of much of this theorem. so Proof. if I is . Definition 3.) We show next that a translation invariant Borel measure is a constant multiple of Lebesgue measure.l)). We call a measure p defined on B (R) that is finite valued for all bounded intervals a Borel measure. Set c = p ((0.
g] [g. p ( E ) = inf { p (G) : E c G. G open} = c m ( E ). b E R and p ( I ) = c m ( I ) for all open intervals I c R.1]. if G is an open set in R. Finally. . Thus. 3. [i. since p is outer regular.3 The Cantor set The Cantor set is an important example for understanding some of the concepts related to Lebesgue measure. a countable union of disjoint open intervals. . G open} = inf { c m (G) : E c G. Notice that C consists of two intervals and 1 has measure Next. Note that C2 consists of 4 = 22 intervals and has measure 4 = ( Continuing 3. (i) g)2. remove the open middle third interval of each piece of C1.2. . the Cantor set is an uncountable set with measure zero. we begin with the closed unit interval [0. Next. if r E R. if a0 < a1 < . Then. ] U 13. ) Setting ai = . q ) )= cq for any rational number q.shows that p ((0.27. (i.i) i] [g. Remove the open middle third of the interval. By countable additivity. In particular. we conclude that for all real numbers r .< an. 0 since Lebesgue measure is regular. iEo iEo Finally.78 Theories of Integration Since p is finitely additive. by Theorem 3.. and call the remainder of the set C1 = [ O . i] [i. G = U i E o I i . let p and q be rational numbers such that p < r < q. i n :)) = Letting p and q approach r . p (G) = C p (Ii)= C cm (Ii)= c m (G) . so that p ( ( u . ~C (O. U U U 11. Call the remainder C2 = [0. b ) ) for all a . p ) C ( 0 . if E E L? (R).q). since ( 0 . To create the Cantor set. p = cm. which in turn implies n p ( ( 0 . then 1 c. b ) ) = c m ( ( a . .
Prop. x E and C # 8.’ or ‘1222.a whose complement is dense in [0.’. Thus.instead of 3’”. If x is an endpoint of an interval in c k . Then. we are left with a closed set c k which is the union of 2‘“ closed and disjoint subintervals. if a1 = 1. Hence.. we showed how the natural length function in the real line could be extended to a translation invariant measure on the measurable subsets of R. C is measurable and m (C) = 0. c 3 c k + l for all k. Our procedure is very analogous to that employed in the onedimensional case. be its ternary expansion. Further. so that the Cantor set is uncountable. The resulting set is a closed set of measure 1 . and only if.. The rest of the con~ . (i. Finally. We begin .’ for ‘1000.’ and ‘2’ for ‘1222. .. . Dividing each term of this sequence by 2 defines a onetoone. after the kth division. it follows that m* (C) = 0. remove 2‘“l open intervals of length ~ 3 . we write our expansions so that they do not end with ‘1000. then it is also an endpoint of an interval in c k + j for all j E N. k Thus. The set k C = np?lCk is known as the C a n t o r set. page 121. ai # 1 for all i. Thus.. since (3) c m* ( C ) 5 m* ( C k ) = (5)k for all k . . I]. Thus. 8. m* ( c k ) = . these results extend the natural area function in the plane and the natural volume function in Euclidean 3space. we extend the result to Euclidean nspace.1]. ai E {0.1]. The Cantor set is an uncountable set of measure 0. we can think of the ternary decimal expansion of an element of C as a sequence of 0’s and 2’s. i).Le besgue integral 79 this process.2} for all i. this set of sequences is uncountable.1.’. struction is the same. we write ‘0222. let O. We define generalized C a n t o r sets as follows.. . As proved in [DS. 3.ala2a3. In this section. . then x E the first interval removed. At the kth step. We now make some observations about C. By construction.k .. each of length 3 . Fix an Q E (0. We next show that the Cantor set is uncountable.9. onto mapping from C to the set of all sequences of 0’s and 1’s.l). For x E [0. It is a closed set since it is an intersection of closed sets. See Exercise 3. To do this. x E C if. In particular.3 Lebesgue measure in R” In the previous section. For example. One can also prove that its complement is dense in [0. .
pages 7679]. Let { ~ k } r = r C Rn be a sequence in R". We define the n o m . d ( x .Yil Let B (20.80 Theories of Integration by defining Lebesgue outer measure for arbitrary subsets of R".b). A set G c R" is called open if for each x E G . . Gm} C {Gi}icA such that K C U z l G i . every open cover of K contains a finite subcover. if { x k } E 1C F and xk + 20. See [DS. We say I is open (closed) if each I is open i (closed). and only if.. We define the symmetric diflerence of sets E l .xn + yn) and t x = ( t x l . we define the volume of I to be with the convention that 0 oo = 0. ) = {x E R" : d (x. A set K c R" is compact if. is a finite there subset { G I . By the HeineBore1 Theorem. the edge of an interval is a degenerate interval and.. d. If each Ii is a halfclosed interval of the form [a. and then restricting the outer measure to a class of subsets of R" called the (Lebesgue) measurable sets to obtain Lebesgue measure on R". and only if. If x. . of x by lllcll = (Cy=l lxi12) . . i # i'. . converges to xo E R" if 1imkdm d (xk. That is. x"). is degenerate and has length 0. & c R". K 12 1 (ca. between points x . . is an interval in R.) 1/2 . G2.YII = 1% . . i = 1. . x In. that is. . then v ( I ) = 0. then xo E F . . . An important collection of subsets of Rn consists of the compact sets. we define x y and t x to be + x + y = ( 2 1 + 91. Finally. where each Ii. where xi E R. . a set H is called bounded if there is an M > 0 such that llxll 5 M for all x E H .xo) = 0.A such that K C L J ~ E A G ~ . IIII. . . y ) = . . .tx. < r } r xo) be the ball centered at xo with radius r. a set K c R" is compact if. The distance. A set F c R" is called closed if every convergent sequence in F converges to a point in F .n. Many of the statements and proofs of results for Rn are identical to those in R and will not be repeated. . y E R" is then the norm of their difference. ) c G.. We say that { x k } F = . when n = 2) function. to be the set ElAE2 = (El \ E2)U(E2 \ E l ) . If I c R" is an interval. An interval in R" is a set of the form I = I1 x . The space R" is the set of all realvalued ntuples z = ( 2 1 . given any collection of open sets {Gi}i. is closed and bounded. In particular. Below. denoted E1AE2. there is an r > 0 so that B (x. showing that Lebesgue outer measure extends the volume (area. . we call I a brick. so that if some interval I has infinite i length and another interval I p .y E R" and t E R. we will use the following characterization of compact sets. .
I ba I I b Figure 3. such that E c UjE. We define the (Lebesgue) outer measure of where the infimum is taken over all countable collections of open intervals { J j } j E . has volume 0. Finally.. .40 Let E E ) m.Jj. It is straightforward to extend results (l). For this result.(2) and (4) of Theorem 3.6 to m. by c R.6 also holds.1 Analogous to the case of outer measure in the line. . we need the observations that the intersection of two bricks is a brick and the difference of two bricks is a finite union of pairwise . we define the outer measure of a subset of R" by using covers of the subset by open intervals in Rn. Definition 3. We show that the analogue of property (3) of Theorem 3. In the figure below. bll .Lebesgue integral 81 hence. . the brick is the union of bricks b l ) . note that if B is a brick which is a union of pairwise disjoint bricks {Bi : 1 5 i 5 k}. then 21 ( B )= B1 c Ic i=l v (Bi) . ( E l .
. .. each of which is a finite union of pairwise disjoint bricks. then there is a finite family f . Then.2 We begin with a lemma. . In the following figure. Bm are unions of members of {Ci : 1 5 i 5 1 ) . . By the induction hypothesis. assume we have proved the result for sets of m bricks. . = { F l . . Suppose we have m 1 bricks B1. Fk} of pairwise disjoint bricks such that each Bi is a union F of members of F. .. . the difference of bricks B1 and B2 is the union of bricks bl . B1 fl B1 \ (B1 n B 2 ) and B2 \ B2. b4. . .82 Theories of Integration disjoint bricks. .Bm c R" are bricks. I I b . . . Figure 3. . Consequently. F Note that this result implies that the union of two bricks is a finite union of pairwise disjoint bricks. . Lemma 3. Proof. .41 I B1. . Note that B = Bm+1\ U z l Bi = ng. B is a finite union of disjoint bricks. (B1 n B2) are pairwise disjoint and since B1nB2 is a brick and the difference of bricks is a union of pairwise disjoint bricks. the existence of the family . . follows. there exist pairwise disjoint bricks C1. . . Assume that m = 2.Bm+1.+1 \ Bi) is an intersection of finite unions of disjoint bricks. Cl such that B1. (B. . . Proceeding by induction. B = U$LICi + . Since we can decompose B1 U B2 into three pairwise disjoint sets.
. . These bricks exist because I is a closed interval and each Ji is an open interval. It follows that v ( J i ) = v ( K i ) . The result follows by induction. Therefore.k .. Proof. + + xjEC P v(I)=v(K)=Cv j=l as desired. . 0 We now prove that the outer measure of an interval equals its volume.v ( I ) = v ( K )and K c U!=. . i = 1. . . ( I ) 5 v ( I ) . .42 If I c R" as a n interval. . Jk}.. and the members of this collection can be replaced by the pairwise disjoint sets Ci n Ci. I is covered by a finite number of the intervals { J1.16. there is a family F = {Fl. Then. . .. such that K and each Ki is a union of members of F.. Let Ki be the smallest brick containing Ji and K the largest brick contained in I .. .Ci. A subset E c R" is Lebesgue measurable if for every mi ( A ) = mi ( An E ) + mi ( A \ E ) . Theorem 3. Cl. It follows that m.43 set A c R".Lebesgue integral a3 where {Ci. x I: is an open set containing I and v ( I * ) = (1 E)" v ( I ) .. say.CL} is a collection of pairwise disjoint bricks. .6. . Suppose first that I = I1 x x In is a closed and bounded interval. then v ( I ) 5 v ( J i ) .let I: be an open interval with the same center as Ii such that l ( I t ) = (1 E) l ( I i ) . . .. Definition 3. I* = I. then rng ( I ) = v ( I ) .Suppose K = U?=lFj . CL}. we may replace the set {Bl. Since I is compact.Ki. The case of a general interval can be treated as in the proof of Theorem 3.. By the lemma.. ( I ) 5 v ( I ) . To complete the proof.. I and j = 1.. Ci \ (Ci f l Ci) and Ci \ (Ci Ci). 0 We note that since the edge of an interval is a degenerate interval. n each of which is a union of pairwise disjoint bricks. we need to know that if { Ji : i E a} is a countable cover of I by open intervals. .. . Fj E F . . Fl} of pairwise disjoint bricks . by {Cl. To see that rn. . x . . the outer measure of the surface of an interval is 0. . Then. . We define (Lebesgue) measurability for subsets of R" as in Definition 3.
which follow from the definitions of the sets Bk: (1) if x E Rn. such that x E B.45 If p is a translation invariant measure o n 2 (R") which 3 is finite o n compact sets.x [0.. 1)be the unit brick in R" and set c = u ( I ) . extending these concepts from intervals to more general sets. Theorem 3. Lemma 3. Proposition 3. we show that every open set G c R" can be decomposed into an countable collection of disjoint bricks. Let BI. it follows that G = U00. then there is a unique B E BI. Choose all the bricks in B1 (G). We will discuss computing these quantities using Fubini's Theorem below. Proof. with j < k . (2) if B E Bj and B' E BI. comprises the smallest aalgebra generated by 3 the open subsets of R". then G as the union of a countable collection of pairwise disjoint bricks.24. is a aalgebra containing all ndimensional intervals and mn is countably additive. then p = c m .36 and its corollary hold. choose all the . f3 (R") c M. Note that BI. 1)x . for large enough k . if x E G then x is contained in an open sphere contained in G. For any k E N. Thus. M . is a countable set. = ) bricks in B2 (G) that are not contained in any brick in B1 (G). restricted to M n . . Using Lemma 3. Since G is open. be the family of all bricks with edge length 2'" whose vertices are integral multiples of 2I. we keep all the bricks in Bj (G) that are not contained in any of the bricks chosen in the previous steps. : B C G}.44 below that. and the regularity conditions of Theorem 3.84 Theories of Integration We denote the collection of measurable subsets of R" by M n and define Lebesgue measure mn on Rn to be rn. As in the onedimensional case (Corollary 3.. Set BI. Since the Lebesgue measure of an interval in Rn is equal to its ndimensional volume. Before leaving this section. Continuing. U B E B l . Let I = [0.44 If G c R" i s a n open set. As in Theorem 3. ( ~B.(G) = { B E BI. Proof. f (R").39.44. Thus.30. f o r some constant c.28)) we see by using Lemma 3. The collection of Bore1 sets of R". there is a brick B E B k such that B c G and x E B .25 and Proposition 3. we can prove an extension of Theorem 3. This construction produces a countable family 0 of pairwise disjoint bricks whose union is G. We need the following observations. .I is the union of 2"'" pairwise disjoint bricks of side length . Next. we use Lebesgue measure to define area of planar regions in two dimensions and volume of solid regions in three dimensions. then either B' c B or B n B' = 0.
37. Recall that we follow the convention 00 . we have that p = cmn by the analog of Corollary 3. We recall their definition. In fact. then cp(x) = aixEi (x) is also integrable. which in turn forced us to consider a proper subset of the set of all subsets of R. Also. We already know that if E is an interval. By Lemma 3. To begin. ( B )for any such B . we extend the real numbers by adjoining two distinguished elements. p (G) = cmn ( G ) for any open set G c Rn. . (2) 0 0 + z = x + 0 0 . But. En c R are pairwise disjoint.4 Measurable functions Lebesgue's descriptive definition of the integral led us. both ~ + ( . For property (6) of Lebesgue's definition to hold. Our study of measurable functions will involve simple functions.m) the extended real numbers. If B is any brick with side length 2 . . measurable sets. (3) 00 x = x + (00). O = 0 00 = 0. . to consider the measure of sets. if E l . 0 (5) if a < 0 then 00 . (4) if a > 0 then 00. then. + While oo+m = 00 and co+(oo) = 00. by linearity. . so by translation invariance. Hence. each of these bricks has the same pmeasure.Lebesgue integral 85 2'". 0 = 00 and (00) . We call the set R* = RU { 00.p 1 2"k 1 ( I ) = cmn 2"k ( I ) = cmn ( B ). monotonic limits of such simple functions must also be integrable. in order for xE to have an integral. 3. page 64]). We now investigate such functions. Since p is outer regular ([Swl. we have p ( B )= .k . . a = a (00) = 00.a = a .0 0 < x < 0 0 . then J Cz1 x E d x = l ( E )= m ( E ). a = a . E must be a measurable set. 00 and 00. The extended real numbers satisfy the following properties for all x E R: (1) .o o )and (oo)+oo are undefined. (4) and (5) remain valid if a equals 00 or 00. 00 = m and (00) .44. Remark 7. (00) = 00. in a very natural way. p ( B )= cm.a = a .
then {x E E : cp (x)> a } = Em. i = 1.Em pairwise disjoint. . so the measurability of cp requires that Em E M n . .48 Let cp (x)= EL1aiXEi (x)be a simple function in canonical form. each Ei is measurable.am on the sets Ei = {x : p (x)= a i } .47 Let E be a measurable subset of R”. . . Example 3. . On the other hand. . Below. Continuing in this manner. a monotonic limit of simple functions is an extended realvalued function. and consequently cp is a measurable function. Then. unless explicitly stated otherwise.86 Theories of Integration Definition 3. then {Z E E ~ ( x>)a } = Em1 U Em. .. with E l . suppose that a1 < a2 < … < am. and only if. we will always assume that a simple function is measurable. If am1 5 a < a. we see that each Ei E M n . the limit exists. {(P~}T=~ Definition 3. and since Em E M n .49 Let f : R (2 + R be continuous. We say that an extended realvalued function f : E + R* is (Lebesgue) measurable if {Z E E : f (z)> a } E M n for all a E R. Since a } = f . each set Ei is measurable. Em1 = Em1 U Em \ Em E M n .. limk_. Thus.46 A simple function is a function which assumes a finite number of finite values. Then and it follows that cp is measurable if. {x E E : cp (x)> a } E M n for each a E R since it is a (finite) union of measurable sets. but may not be finite. Let cp be a simple function which takes on the distinct values a l . if each Ei E M n then. and only if. If am2 5 a < am1. We call f an extended realvalued function if f : E + R*. Suppose that is a monotonically increasing sequence of simple functions defined on some set E. We first observe that a simple function cp is measurable if. Example 3. .l ( ( a 00)) . then Em1 U Em is measurable.. . TO see this. for each x E E .m cpk (x) exists in R*. Then. Thus. E R : f (2)> . . the canonical form of cp is m i=l Let E c Rn. if cp is measurable.m.
l ) be a nonmeasurable set. Example 3. (2) and (3) are equivalent. (4) {X E E : f (x)5 a } E M n f o r all a E R.47 and Proposition 3.Lebesgue integral 87 is an open set.t i f t $ P { t iftEP * . (3) {X E E : f (x)< a } E M n f o r all a E R. {x E E : f (x)> a } E M n f o r all a E R. Proposition 3.1) by + Let P c ( 0 . It is a bit surprising that the converse to this corollary is false.n } . Proof.50. { z E E : f (x)2 a and {x E E : f (z) 2 a } = nT=. that is. Since {x E E : f (x)> a } = U E . E E : f(x) > n } = {X and {z E E : f (2) = 00) = n= =. similarly. it is measurable. Since {x E E : f(x) = a } = {Z E E : f ( ~ 5 a } n { z E E : f(x) 2 a } ) {x E E : f(z) 00) = nr==. completing the proof. See Exercise 3. { X E E : f (x)> a (1)and (2) are equivalent . we see that Corollary 3. The following are (1) f is measurable.51 W e can replace the condition “for all a E R ” b y “for every a in a dense subset of R ” in Definition 3. +i } Remark 3. ) + R* be measurable. Then.53 (0.52 {X E E : f Let E E M n and f : E ( ~ = a } E M n for all E R*. The measurability of a function.50 equivalent: Let E E M n and f : E + R*. Since {x E E : f (x)5 a } = E \ x E E : f ( ~ > a } and M n ) is a aalgebra. Thus. {x E E : f (x)< .20. 0 i}. . (1) and (4) are equivalent. every continuous function defined on R is measurable. Define f : f (4= . has several characterizations. like the measurability of a set. ( 2 ) {X E E : f (x)2 a } E M n f o r all E R.
l ) : f (x)> 0) = P . and f .88 Theories of Integration Then. f + g. Since {ZE E : f ( x ) + c > a } = { Z E E : f (z) >ac}. but since {x E ( 0 . then {x E E : cf (x)> a } = Thus. Then: + (1) {x E E : f (x)> g ( x ) } is a measurable set. Proof. f V g and f A g are measurable functions. by (l).notice that Since each of these sets is measurable. Proposition 3. cf is measurable function. g : E + R* be measurable and assume that f g is de*finedf o r all x E E . Note that Eifa<O 0 ifa2O' Since g is measurable. {x E E : f (x)> g (z)} is a rneasurable set. The next result contains some of the algebraic properties of measurable functions. Let c E R. Fix a E R. a .g is a measurable function by (2) and so. To prove (l). fg. (2) f + c. f is onetoone which implies that {z E ( 0 . f g is measurable. note that for a < 0. then E : cf (x) > a } = { z E E : f (x)> : } if c {x E {x E E : f (x)< : if c < 0 } If c = 0. cf . f is not measurable. Consider (2). l ) : f (2) = a } is Lebesgue measurable for all a E R*. If c # 0.54 Let E E M . + {x E E : f2 (x) > a } = E . To see that f g is measurable. . >0 the function f + c is measurable.
since {Z E E : (f V 9 ) (z) > a } = {z and E E : f (z) > a } U {z E E : (9) (z) > a } {Z E E : (f A 9 ) (x)> a } = {z € E : f (z) > a } n {z E E it follows that f V g and f A g are measurable. all of its subsets are measurable. a statement about the points of a measurable set E is said to hold almost everywhere in E if the set of points in E for which the statement fails to hold has Lebesgue measure 0.Lebesgu.. and f. Then. {z € E : g (2)> a } = {z E E \ 2 : g (z) > a } u {z E 2 : g (z) > a } = {z E E \ 2 : f (z) > a } u {z E 2 : g (z) > a } . See Exercise 3. 2 is measurable and mn (2)= 0. Thus.e. Iff is measurable. we use phrases like “almost every 2)’or LLalmost z” to mean that a property all holds almost everywhere in the set being considered. g is measurable. 0 Corollary 3. Then. then I is measurable. Fix a E R. 0 . Then. Proposition 3. f If The converse to the last statement is false. Finally. {X E E : f 2 (z) > a } = { z E E : f (z) > fi} { z E E : f U f2 (2) < f i Since all of these sets are measurable. : (9) (z) >a}. f is measurable iJ and only if.are measurable functions.55 Let E E M .56 Let E E M . Then. Let 2 = {z E E : f (z) # g (x)}. As in the case n = 1. Proof. we get the following result.d 2 4 we see that fg is a measurable function.e integral 89 while for a 2 0. and f : E R*. Writing (f + d 2. Since 2 has measure 0.g : E 4 R*. the measurability of f and the equality {z E E \ 2 : f (z) > a } = {z E E : f(z) a}\{. Additionally. Consequently. fg = is measurable.(f .21. f+ and f . Suppose that f is measurable and f = g a. E 2 : (z) > a } imply that {z E E : g(z) > a } E > f Mn.
57 The sequence (. lim supi xi and lim infi xi are equal. and lim supi xi is the largest accumulation point. Theorem 3. we first define some special limits. F. and.for each k . limi. the equality infk fk = . liminfi xi is the smallest accumulation point of ( x . When they are finite. is measurable. limk. The following result. limsupk fk and liminfk fk are measurable .OOxi exists if.e. We now consider limits of sequences of measurable functions. Fix a! E R..l)i} { 00 satisfies lim supi (.fk) proves that infk fk is measurable. infk f k . Finally. Proof. Further. it almost converges uniformly. fk : E +R* i s a measurable function . in which case limi.OO xi equals their common value. that is. } ~ . the sequence does not have a limit. To do this. I f lim f k exists a. the sequence converges uniformly except on a set of small measure. lim supk f k = infk supjzk f j and lim infk fk = supk infjlk fj. T h e n supk fk. which is due to D.. By definition.e.e.22.. consequently. then it is measurable.90 Theories of Integration We next investigate limits of measurable functions. we define the limit superior and the limit inferior of {xi}El by: lim sup xi = inf i i {: . by Exercise 3.59 (Egorof’s Theorem) Let m n ( E ) < 00. shows that when a sequence of measurable functions converges. which shows that limsupk fk and liminfk f k are measurable. Note that which implies that sup. Suppose that fk (x) = f (2 . Next. f k is measurable.58 Let E E M n and suppose f k : E +R* is a measurable function f o r all k E N. Egoroff (18691931).lZ = 1 and ) i=l lim infi ( l)i = 1. Example 3. Thus. xk} and We always have that 00 5 liminfi xi 5 limsup.functions. Given a sequence {xi}El c R. xk} = Jim 200 {: . xi 5 +m. then it equals 0 the limsup fk a.sup (. if limk f k exists a. Theorem 3. and only if.
for each m there is an im such that m.e. 0 The next two examples show that we cannot relax the conditions that E have finite measure or f be finite valued.Lebesgue integral 91 a. Emi and H are measurable sets and. since E has finite measure. choose m so that $ < q. 1 and f k (2) = k .61 Let E = [o.e.) < &". The function f is measurable since it is the pointwise limit a. we need E to have finite measure. . given q > 0.34 c UZlEmi.by Proposition 3. Since Emi c Em(i+l). I f k (2). Set F = nmoO. then by the definition of Emin.lEmi. Fix and. If k 2 im and x E F C Emi.60 Let E = Rn and let f k be the characteristic function of the ball centered at the origin and having radius k . Proof.. for all m.. For all m. Example 3. f k (x) +1 for all x but the convergence is not uniform on sets whose complements have finite measure.e.set and H = {x E E : lim f k k+co (x)= f (x) Then. SO that F is measurable and 00 00 m=l m=I Finally. { fk}p=l zl converges uniformly to f on F . 1imidm m n ( E \ Emi) = 0. f k converges to 1 the function which is identically 00 on [0. o n E .. of a sequence of measurable functions. H m.e. Then. Thus. Then. there is a measurable set F C E such that m n ( E \ F ) < E and { fk)p=1 converges unaforrnly t o f o n F . Then.. given a n y E > 0.iE N. (E\ Emi. given E > 0. so the convergence cannot be uniform.1]. We need f to be finite valued a. o n E and f as finite valued a. Therefore.f () < q. Example 3. Therefore.
and A? by Then. The third principle is Egoroff’s Theorem. There as a sequence o. since mn (G) = limktm m n ( U e . I f f is bounded. once k is greater than the bound on 1 If 1) If k (x). for k E N. and suppose f : E + R* is nonnegative and measurable. Eittlewood (18851977) summed up how nice Lebesgue measurable functions and sets are with his “three principles”: (1) Every measurable set is nearly a finite union of intervals. if f is bounded then. Given E E M n and E > 0. We can define a sequence of simple functions that converges pointwise to f . we can approximate E by the union of all the bricks. simple functions { f k } r = 1 which increases to f pointwise o n E.define measurable sets A. then the convergence is uniform o n E .which has the same measure as Bi. we can approximate E by a finite union of open intervals.f (x)l < 2” for all x E E . If CT is infinite. (3) Every convergent sequence of measurable functions is nearly uniformly convergent. By Lemma 3. Further. the function k2k . . If CT is finite. is nonnegative and simple and { fk}r=1 increases monotonically to f for all x E E. Thus. G = U i E u B i .62 Let E E M . (2) Every measurable function is nearly continuous.36. there is an open set G containing E such that m n (G \ E) < E . E. a countable union of disjoint bricks. We now turn our attention the second condition. To see this.44. we can approximate E by a finite set of the Bi’s (at least when the measure of E is finite).f nonnegative. Let f be a nonnegative and measurable function on E c R”. since the surface of a brick has measure 0. replacing Bi by the largest open interval contained inside of Bi. we have proved Theorem 3. B i ) .92 Theories of Fntegration The British mathematician J. Finally. The first principle follows from condition (2) of Theorem 3.
Then.where the ai’s are distinct. F is a closed set.36. ) Fix E > 0. we get the following corollary. f I F is continuous.E”=oFk. Since E = UZ. N.are nonnegative and measurable. Choose a sequence of simple functions { f k } & that converges pointwise to f . There is a sequence of sample functions { fk>r=1 which converges to f pointwise o n E . is continuous. (If f (x)= 0 for some 2. there is a closed set Fo C E such that mn(E\Fo) < ~ 2 . Lusin (18831950). suppose mn ( E ) = 00.l and {fk}r=1 converges uniformly to f on Fo. f l F is continuous.and both f+ and f. Then f k l F is continuous and { f k } z 1 converges uniformly to f on F . there is a closed set F c E such that mn ( E \ F ) < E: and f I F . Lit t lewood’s second principle is contained in the following theorem of N. Theorem 3. By Theorem 3. since the sets Fi are pairwise disjoint and f is constant on each of these sets.finite valued almost everywhere. Since a uniform limit of continuous functions is continuous.Ei.1 5 llxll < j} and write E = u Z l E n A j .36. Set F = UZiFi.Lebesgue integral 93 If f is a measurable function on E . Set F = n. there are closed sets Fi c Ei E such that mn (Ei \ Fi) < . ( E \ Fk) < ~ 2 . we have E\FF c U Z l (Ei \ F ) c U z l (Ei \ Fi) which implies that Proof. the Ei’s are measurable and pairwise disjoint. Given E > 0. m mn ( E \ F ) I mn ( u ~(Ei \ ~ i ) ) = C m n (Ei \ Fi) < E . and m.f. Choose closed sets Fk C E such that m. Finally. there is a closed set . Let Aj = {x E Rn : j . and E = UKlEi. Since mn ( A j ) < 00. Applying the theorem to each function separately. suppose that f is measurable and mn ( E ) < 00.63 Let E E M n and suppose f : E + R* i s measurable. then the convergence is unaforrn o n E . Corollary 3. then a j = 0 for some j . By Egoroff’s Theorem and Theorem 3. Assume that f is a simple function with canonical form f (2) = CE1 a i X E i (x).( ’ ” + ~ ) f k ] F k is conand tinuous. I f f is bounded. Further.64 (Egoroff’s Theorem) Let E E Mn and suppose f : E + R* is measurable and . then f = f + . l i=l Next. the restriction of f to F .
we set Gt = U equals the number of bricks in o k . ( E )+ &2k < 03. a is infinite.". m(R\F) = m(G) < E and since flF = 0. that m ( E ) < 03 and f is bounded. mn ( E \ F ) < E and f J Fis continuous. first. i . Set F = G". Then. Let f k be the simple function k2k f k (2) = i1 T X A f (x) i=l where A$ = {x E E : By construction. when considered as a function on R.e. We can define a step function on the entire real line by setting it equal to 0 on the complement of the union of these open intervals. there is an open set HF 3 At such that m. ( A t ) &2k 5 m.Bi.44. Let G be an open set containing Q with m(G) < 6. where l k . in E . f is not continuous on F . and scalars { u i } E 1 such that cp (x)= ai for x E Bi and 'p (x)= 0 for x $ UE. Moreover.f (x)l < 2'" for all x.65 The conclusion is not that f i s continuous o n F but that the restriction o f f to F is continuous.67 Let E E M n and suppose f : E + R* i s measurable. (Hi" \ A t ) < & 2 . We extend this idea to higher dimensions by calling 'p a step function if there are finite sets of pairwise disjoint bricks. af (). i . there as a sequence of step . F is closed.( 5 M f o r all x E E and k E W.+ (x). then I'pk I). By (2) 95 f < $}. Suppose. Note that by construction Fj and F are a t a positive distance L for j # 1.I 5 M f o r all x E E . (HF) < m. If o k . i such that + .. However. If g k .j of disjoint bricks.i. {&}El.Fj. j . is lk. we can choose l k .k .iBj where {BF'z}jEok. f l F is continuous. Then.i a countable union k. If Proof. since m. we have If.66 Let f be the Dirichlet function defined on all of R. Lemma 3. Let M be the bound on f and suppose k 2 M . j is finite.94 Theories of Integration Fj c E n A j such that f J F j is continuous and mn ( E n Aj \ F j ) < €2j. Set F = U. Example 3.i g k . Since each At is a measurable set. In the onedimensional case. Thus. Hf = LJjEuk. a step function is a finitevalued function which is constant on a finite number of open intervals of finite length. We now show that a measurable function defined on a set of finite measure can be approximated by a sequence of step functions.functions { ' p k ) E l that converges t o f a. Theorem 3. See the next example. 0 Remark 3.
suppose that f is a measurable function defined on a measurable set E . Set EN = E f l I N and define f~ by N if x E EN and f(x) 2 N f (2) if x E EN and If (x)1 < N . F k ) = 0 and {(Pk}T=1 converges to f a.. I'pk (x)1 5 M for all x E E and k E N.e. By construction.f N (x)1 < 2N for all x E EN \ FN and . 4 k=m k=m for all m. EN and { f ~ } . since :..f(x)I 2 2'}. there is a step function pN. Consequently. Finally. m ( E N ) < 00. then ' p k (x)+ f (x). Then. mn (n. supported in E N .N if x E EN and f (2) 5 N 0 if x @ EN ' Note that E = U F = ~ E NEN c EN+^.f ) I(. if x n = U = F k . = ~ converges to f for all x E E. F. F k C U z (GFAAf) so that mn ( F k ) 5 2'+'. f~ is bounded on . I'pk (2) .=. N]. Set so that ' p k is a step function and Further.e. Ur=. Let I N be the interval in R" that is the nfold product of the interval [N.Lebesgue integral 95 Then. then F. By the previous part of the proof. < 2k for all k > m so that ' p k (x)$ f (2). Now. 'pk (x)= f k (x)for all x @ UF:~ ( G ~ A A We now show that ' p k + f a. If z @ u E m F k . Let F k = {x : I'Pk (x). and a set FN C EN such that I ' P N (z) .
. we see ” k i=l k i=l as long as l ( I i ) < 00 for all i. Set 2 = n. We claim that for each fixed K . then mn ({z E E : ‘PN (2)+ f (z>}>L C mn ({zE E K K=l 00 ( P N (2)+ f (. For.. {(P~}. U Since m. Consequently. if x 4 n E = K U F = M F N . If f is a step function k (on R). 0 If If 3.>)> = 0 and {(P~}. . tends to 00 (or 00) so that cpN (2)+f (2).f ~ + 0.. if that were true.a2.This completes the proof of the proposition. . .96 Theories of Integration m. ( F N ) 2(N+1). It remains to show that pN (z) + f ( 3 .f~ (z)1 < 2N for all N > M so that cpN ( z ) .e. f (z) = aiXIi (z) where the li’s are pairwise disjoint intervals. ) for all x 4 Z K . then IpN (z) .5 Lebesgue integral Lebesgue’s descriptive definition of the integral led.M=K F = M F N . to a study of measurable functions.) = 0.=~ converges to f a. then cp has the canonical form cp (4 = . This equality will guide our definition of the Lebesgue integral. If ().(2). in a very natural way. via limits of simple functions. in E . If M 2 K and z 4 U F = M F N . to the development of the Lebesgue measure of sets in Rn and. then {pN (z)}. Recall that a simple function p takes on a finite number of distinct nonzero values al. fix K and argue as in the previous part. (2. then by using properties (l). if (z)] = 00. (z) then cpN (z) . (3) and (5) of Lebesgue’s descriptive definition.. then f~ (z) = f (z) for all sufficiently large N and p N (z) + f (2).=.arc.I < 00.=~ converges to f on EK except for a set of measure 0.f~ (z) + 0. If Ai = { z : p (z) = a i } . So.
We define the Lebesgue integral of cp to be If E Mn.70 Let 9 (z) = C:=. simple function with aiXAi (2). k i=l =ai m bj i=l j=l as we wished to show. Proposition 3. set Since we see that Remark 3. Proposition 3. simple functions and (1) 2() (3) (4) The mapping additive. Then.68 canonical form k Let cp : Rn +R be a nonnegative. j=1 Proof. Let cp and $ be nonnegative.71 QI 2 0. wise disjoint measurable sets. simple functions.e integral 97 Definition 3. we will use the Lebesgue integral. Then. Let and k J be the canonical form of Thus. Then. s to denote The definition of the Lebesgue integral of a simple function is independent of its representation. defined by is countably . b j X F j (2)where the sets F j are pair 1 c m = bjmn (Fj) . a The next result collects some of the basic properties of the Lebesgue integral of nonnegative.L e be sgni.69 For the remainder of this chapter.
j)EsE i j (x). to prove (4). bjX c(i. Let cp and Q be nonnegative.72 then J P I J$. By the proposition above. so that j For (3). ai 2 0 for all i. Corollary 3. we set Eij = Ai n Bj for 1 5 i 5 lc and 1 I 5 m. To prove (1). Let and be the canonical forms of and u. Thus. {Ej}jEu c Mn be a countable collection of k i=l jEu k j E u i=l Applying part (2) to the function Q .98 Theories of Integration Proof. we get the following corollary. ) : 1 5 i I.j)ES Finally. Let S = { ( i .let pairwise disjoint sets. I cp 5 $. since To prove (2). simple functions. we need only note that since cp is nonnegative. f . 1 I5 m } . Then. cp (x)= j k j a i x E i j (x) and $ (x)= C(i.cp.
()1 If h < f. even f o r bounded functions. simple function o n R”.73 If cp is a nonnegative. This is pointed out in Proposition 3. simple function on Rn.102 after we have developed some of the basic properties of the Lebesgue integral. we extend the definition of the Lebesgue integral to nonnegative. : Using simple functions. h : E measurable and cy 2 0 . Since we are considering functions which m a y be unbounded. Then Since Q is countably additive.76 Let E E M .w]defined b y Q ( E )= JE cp is a measure o n M n . measurable functions. However. then (2) If 0 < f.Lebesgue integral 99 In fact.74 Let E E M n and f : E . it is not necessary to compare upper and lower integrals. Define the Lebesgue integral of f over E by If A is a measurable subset of E . The next result shows that the Lebesgue integral is a positive operator on nonnegative measurable functions. we define Remark 3. and f .75 Equation (3. Definition 3.3) i s analogous to a “lower integral”. Suppose cp is a nonnegative. Proposition 3. the% m a y be no simple functions that dominate f ) so it would then be impossible to define a n “upper integral”. it is only necessary that ‘p 5 $ except on a null set. then M n [O. then (3) 3 R be nonnegative and . we see Corollary 3. Then. a we will s discuss below. R be nonnegative and t measurable.
Thus $) s. by the previous proposition. Note first that f is nonnegative and measurable since it is a limit of measurable functions.f k (z). For (3). so the Lebesgue integral of f is the supremum over a bigger set. Further. Proof. measurable functions defined o n E. Theorem 3. fix 0 < a < 1 and let cp be a simple function with 0 5 cp 5 f.100 Theories of Integration Proof. Set E k = {z E E : f k (2)2 acp(z)}. the Monotone Convergence Theorem. If a > 0. then a f = 0 and by our convention that 0 . Then.Ek. is simple Note that (2) is the statement that the Lebesgue integral is a positive operator on nonnegative measurable functions We now come to our first convergence theorem for the Lebesgue integral. then simple function and 0 5 . d$ af = sup { $ : 0 5 $ 5 a f and 1c. { JE fk}fLo.+.l is a monotonic sequence and limk JE fk 5 JE f . if $ is a simple function and c 0 5 $ 5 clf. Thus. and { f k } z l be a n increasing sequence of nonnegative. Setting h = 0 in (1) proves (2). then a p is a c simple function and 0 5 a p 5 a f .77 (Monotone Convergence Theorem) Let E E M . Set f (z) = limk. Since 0 5 f k 5 f k + l 5 f . To prove ( l ) . we see that if cp is a simple function and 0 5 cp 5 f .I 5 f. To prove the reverse inequality. Since f k (z) increases to f (2)pointwise. note first that if a = 0.00 = 0.note that if cp 5 h then cp 5 f. . it follows that E k C E k + l for all k and E = UP=.
increases to f g. so by a s ... Since all the functions are nonnegative. cp = a 2 . 0 5 ‘pi qi and {cpi qi}Z. JE’p defines a measure. that The proof is almost done. T h e n . measurable functions. Corollary 3. We use linearity and induction to show for all N E N. simple functions {‘pi}. we see that limk JE simple functions cp 5 f .78 Let E E M . t s. we can easily show that the Lebesgue integral is countably additive. fk cp L If we let a + 1. we get 2 I sE p. By Theorem 3. @ ( E )= Proposition 3. and {qi}Z1which increase to f and g. By Proposition 3. Proof. respectively. and { fk}E1 be a sequence of nonnegative. Since this is true for all which completes the proof.71. measurable functions defined o n E . we can apply the Monotone Convergence Theorem to complete the proof.71.34.62.Lebesgue integral 101 By part (4) of Proposition 3. Suppose that f and g are nonnegative and measurable. + + + so by the Monotone Convergence Theorem. 0 . there are sequences of nonnegative. we see that the Lebesgue integral is linear when restricted to nonnegative. Thus. Thus. Using this result.
_< )f(x) f o r all x E E . The same is true for all nonnegative. Suppose f (x)= k uixAi (x) is simple function. Pick a sequence of nonnegative simple functions { ( P k } p = l that increases to f. f E f = 0. measurable function o n Rn. nonnegative functions f . X X ( ~ ~ ~ : ~ ( ~ ) . Proposition 3. If f = 0 a. Then. Suppose that f is a nonnegative. measurable functions. f E p k + f E f . . We saw above that the Lebesgue integral of a nonnegative. f = 0 i f . 0 Consequently. and only i f . the mapping @ : M . ( A ) = 0 and f = 0 a. Now. Proof. Then. measurable function defines a measure. measurable function o n Rn. in E .79 Let f be a nonnegative. By part (4)of Proposition 3.81 The previous proof uses a very important inequality in analysis.e. Remark 3. ( A ) > 0 .102 Theories of Integration In fact. then m .80 Let E E M n and f be a nonnegative. the result follows by approximating f by simple functions.+ R* defined by @ ( E )= f E f is countably additive. then f E f = 0. f = 0 a. This contradiction shows that m. if f = 0 a.e. Tchebyshev (18211894).Ak. then mn ( & ) > 0 for some Ic which implies i}.e. Suppose { E j } j E ais a countable collection of pairwise disjoint measurable sets and E = U j E a E j . By the Monotone Convergence Theorem. in E . This will follow from the next two results. Thus. L. (Ai n E ) = 0. sE Proof. measurable function o n a measurable set E . (2) ~ Thus. suppose that f E f = 0. Set Ak = {x E E : f (x)2 so that A = {X E E : f (2) > 0 ) = Up=. . Thus. in E . simple function defines a measure. If m . in E and ui > 0. Proposition 3.3. Let X > 0 . this corollary is equivalent to the Monotone Convergence Theorem.71 and Exercise 3. See Exercise 3. For general. know as Tchebyshev's inequality after P. Then.36. @ (0) = Ja f = 0 and the Lebesgue integral of a nonnegative.e.
. When this is the case and cp (x)= k Further. Definition 3. For a general measurable function. mn ( E f Ai) < 00 l for all i = 1. the set E . Then. = {t E E : 1 f (t)l > a } has finite measure. We say that f has a Lebesgue integral over E if at least one of f+ and f is finite and in this case we define the Lebesgue integral of f over E to be sE sE We say that f is Lebesgue integrable over E if the Lebesgue integral of f over E exists and is finite.. Proposition 3. .84 Iff has a Lebesgue integral over E .. I n order f o r f to be Lebesgue integrable.Lebesgue integral 103 f r o m which we get Tchebyshev's inequality. Let E E M n and f be a Lebesgue integrable function ( I ) for all a! > 0 .e. the integral must exist and be . whenever we can make sense of their difference.to define the Lebesgue integral of f . (at least) one of the sets { t E E : cp(t) > 0 ) and { t E E : cp ( t )< 0 ) has finite measure. in E . and f be a measurable function on IE. Remark 3.85 o n E .finite. sE Note that if cp is a simple function. the measure <p defined by <p ( E ) = JE x A = mn ( A n E ) is the restriction of Lebesgue measure mn to A. An important consequence of Tchebyshev's inequality is that a Lebesgue integrable function is finite almost everywhere.finite valued a. we can use the Lebesgue integrals of f+ and f.82 Let A E M n and set f ( x ) = X A ( x ) . Then. then f m a y equal f o o . and only if. 1 r mn ({x € E : f (2)> A}) I /E " Example 3.k . (2) f is . then cp has a Lebesgue integral over E if. . cp is Lebesgue integrable over E if.83 Let E E M . and only if.
in E . f is . the Monotone Convergence Theorem implies that the Dirichlet function is Lebesgue integrable with integral 0. We see that Ea = { t E E : f S ( t )> a} U { t E E : f .104 Theories o f Integration a Proof.e. which shows that the Riemann integral does not satisfy the Monotone Convergence Theorem. f S and f . See Exercise 3.we defined a sequence of simple functions that increase pointwise to the Dirichlet function on [0. Since { t E E : f ( t ) = 0 0 ) C { t E E : f ( t ) > a} for all a > 0. again. If 1. + R* IJE I If 1 be measurable. we can get an improvement of the Monotone Convergence Theorem. so letting a tend to finite a. To show the second part. and only i f . by Tchebyshev ' s inequality.are Lebesgue integrable. Then. f is Lebesgue integrable ouer . Then. and {fk}T=l be a n increasing sequence of nonnegative.. it is. Proposition 3. we get the following result which shows that Lebesgue integrable functions are absolutely integrable . f 5 JE 1 f l . enough to show that f S and fare finite valued a. Lebesgue integrable .. Note that each f k is also Riemann integrable while the limit function is not.e.finite a.86 Let E E M . so it is enough to prove (1) for nonnegative functions f . so we assume that f is nonnegative. By hypothesis.( t )> a} . In this case.1].e. Using the relationships between f . sE Example 3. Corollary 3. In this case. in E . 00 shows that mn ( { t E E : f ( t )= 0 0 ) ) = 0 and f is 0 For Lebesgue integrable functions. Then.88 Let E E M n and f : E is Lebesgue integrable over E iJ and only i f E .functions defined o n E . Set f ( x ) = 1imkdm f k ( x ) .3. both f + and f .87 In Example 3. Fix > 0. SUPk fk < 00. Since each f k is Lebesgue integrable.37. f is Lebesgue integrable i f .
On the other hand. = {x E E : f (x)2 Then. Proposition 3.a. sLl. i}.1])# 0.ll # 0.5 1. If f = 0 f o r all measurable sets A c E .89 Suppose that E E M n and f : E + R* as measurable. an E .1] and m ([1. Proposition 3. sEf+ sEf= =0 and the 0 On the other hand. that is. and so is f . SA proof0 Since . then both f+ and f. in E .= JAn{zEE:f(z)<O} ' f f. As above. f = 0 while f # 0 a.f+s. Thus.I IS. Finally.Lebesgue integral 105 Proof. then it follows that f = 0 a. or that f = 0 a. If mn ( E )= 0 . ( E )= 0.e.f++++Sf=/ f = E E (f++f)= s.are finite..91 Suppose that f has a Lebesgue integral over E . Proof. sE for x Example 3. The null sets. then f is Lebesgue integrable over E and f = 0. the integrals f+ and f .e. by Tchebyshev's inequality. sets of measure 0.90 Let f : [1. so that = f+ f.1]$ R be defined by f (x)= Then. we may assume that f is nonnegative. play an important role in integration theory. I f f is Lebesgue integrable.are finite. if the Lebesgue integral of f is 0 over enough subsets of E .< 00 and I is Lebesgue integrable. set AI. we see that JA f+ = 0 and JA f. for general measurable functions. > 0)) 5 Cr=l ( A k ) = 0 mn 0 .fA f = JAn{zEE:f(z)>O} and ] A f .e. If1 0 and the proof is complete. in E . in [1. mn ({z E E : f (z) which implies f = 0 a. sE f. sEIf1 sE + sE If1 If sE s' sE sE If sE ls. then f = 0 a. in E . by the positivity of the integral and the pointwise inequalities f + 5 1 f 1 and f . it is not enough to assume that f = 0 to derive m.e.e. Thus. Since 0 = f = f+ = result follows. The next few results examine some of the properties of null sets.= 0 for all measurable subsets A c E . in E . SO that mn ( A k ) = 0. However.e. if is Lebesgue integrable.
( A ) < 6. where the limit is taken over measurable sets A c E. then g is Lebesgue integrable over E and f =sEg.93 Iff 2 0 is Lebesgue integrable. then @ ( E )= f defines a measure.fk) < 5.g : E +R* are measur (1) If I f l 5 g a. 0 < 6 < &. this this happens in a continuous way. then f is Lebesgue integrable over E .e. Suppose that E E M n and f . By the Monotone Convergence Theorem.92 Then. Theorem 3. then fg = 0 a. so is the other and their integrals are equal. in E so JE (f . then JE (f .106 Theories of lntegration If f is measurable. in E and g i s Lebesgue integrable over E . W h e n this condition i s satisfied. in E .. Let E E M . Proposition 3. When f is Lebesgue integrable. Proof.e. (2) If f is Lebesgue integrable over E and f = g a. we say that the measure @ is absolutely continuous with respect to m. 0 Remark 3. When one of the functions is Lebesgue integrable. sE .e.e.g) = 0. Since f 5 JE I for Lebesgue integrable functions f . I If Fix E > 0 and choose k such that m. then JE f = 0 for every set E of measure 0. Fix 6. k } so that f k 2 0 and {fk}p=1 increases to f. This theorem says that given any E > 0 . in E . there i s a 6 > 0 so that i f m. it is enough to prove the result for f 2 0.94 able. ( E ) < 6 then @ ( E ) < E . If as we wished to show. and f : E + R* be Lebesgue integrable. sE If f and g are equal a. Set f k (2)= min { f (2).
e. in E and g and h are Lebesgue integrable over E . in E and f . since that is not always the case.a. note that by hypothesis.8 E R. Then.e. we may assume that the Lebesgue integral of the sum is well defined. set 2 = {x E E : f (x)> g ( z ) } and note that ( 2 ) = 0 so that If I = s z g = 0. g . .since g+ _< f + and g. sz Thus. .e. f is Lebesgue integrable over E . Then.e. Moreover. Since If I and g are nonnegative and measurable functions.96 (Linearity) Suppose f and g are Lebesgue integrable over a measurable set E . The next result shows that the Lebesgue integral is linear for Lebesgue integrable functions. Thus. a f + pg is Lebesgue integrable and .Lebesgue integral 107 mn To prove (l). By part (l).= f .Thus. + Corollary 3.for all a . in E and both f + and f . sEg+ 5 f + and g. if a function is Lebesgue integrable.are Lebesgue integrable over E . Reversing the roles of f and g . the sum of Lebesgue integrable functions is defined almost everywhere and. we conclude that JE g+ = f + and 9. If I is Lebesgue integrable over E and.a. f + = g+ a. g is Lebesgue integrable over E . g+ and g. It follows that sE sE sE sE sE sE + Suppose that h 5 f 5 g a. then we have seen that it is finite almost everywhere. since sets of measure 0 do not effect the value of the Lebesgue integral. Proof. in E .95 Suppose that E E M n and f . If I 5 191 Ihl. The sum of measurable functions is measurable if the sum is defined but. If h 5 f 5 g a.5 f .= g. However. If g and h are Lebesgue integrable over E . Thus. consequently. Theorem 3.e. h : E 4 R* are measurable. then so is 191 IhJ. in general we cannot integrate the sum of measurable functions. then f is integrable over E .5 f . in E .are Lebesgue integrable. we have the following corollary. For the second part.
in many situations. If f and g are nonnegative and h = f .the set of Lebesgue integrable functions on a measurable set E is a vector lattice.so that Thus. This result replaces the monotonicity condition of the Monotone Convergence Theorem by the requirement that the convergent sequence of functions be bounded by a Lebesgue integrable function. s.+ 9+ .b].91) and f A g = (f .g.and ( a f ) . Consequently.91) are Lebesgue integrable over E .= af+. analogous to the set of Riemann integrable functions on an interval [a. f v g = (f g If . If a < 0. Consequently. See Theorem 2. However. (f + g ) = s. the monotonicity condition is not satisfied by a convergent sequence and other conditions which guarantee the exchange of the limit and the integral are desirable. It follows that f . the Monotone Convergence Theorem is a very useful tool in analysis. As we have seen. If .9+ + f  SEy s.a f .eories of Integration Proof.9 7 + =J.= g so that since h is defined and finite almost everywhere. we only need consider the sum of Lebesgue integrable functions. Thus.f.If . s.g are Lebesgue integrable. (f .108 Th. then h+ = f and h.( f + = +g+  (f +g)) = 1 E (f+ + 9 + ) = f + s. We next consider Lebesgue's Dominated Convergence Theorem. As a corollary of the Dominated Conver a + + 4 + .+ 9 . hence.g and. then (af)' = . f 4 1 E (f. for Lebesgue integrable functions f and g. We have already proved this result when f and g are nonnegative and a.. 0 Suppose f and g are Lebesgue integrable functions over a measurable 1 set E .) since all the integrals are finite.23 and the following paragraph.P 2 0.
I g as Lebesgue integrable over E . See Exercise 3.+. Corollary 3.g : E + R* are measurable and f k 5 g for all k . s lim sup f k k+oo 2 lim sup k+oo E .linnf fk 5 lim inf koo s. hk 5 liminf kwo S. ! J and the proof is complete. x E E and all k E N.so that h k is nonnegative and measurable. Then.9 ) = lim inf k+oo LfkLg' + Since g is Lebesgue integrable.Lebesgue integral 109 gence Theorem. By the Monotone Convergence k+oo Theorem.98. Fatou (18781929). Lemma 3. we will get the Bounded Convergence Theorem. then. f S. known a Fatou's Lemma. Since hk 5 fk for all z E E .g : E R* are measurable and g 5 f k f o r all k . Then f k . fk. l i y k f f k P r 5 lim inf koo / E ~ ~ ' Proof. There is also a result dual to Corollary 3. increases to liminffk. fk : E + R* J. and {hk}p=.g is a nonnegative function and we can apply Fatou's Lemma to get that k+oo (f.99 Suppose that E E M n and fk. Suppose that g is a Lebesgue integrable function and each f k is a measurable function such that f k 2 g for a. If g as Lebesgue integrable over E . We begin with a result due to P. then. we have Corollary 3.97 (Fatou's Lemma) Suppose that E E M n and as nonnegative and measurable for all k . Set hk (z) = infjzk f j (z). fk. koo lim S.e.38.98 Suppose that E E M n and fk.
Suppose there is a number M so that lfk (x)1 5 M for all k and for almost all x E E . then We defined the Lebesgue integral by approximating nonnegative. Then. Thus. Proof.101 (Bounded Convergence Theorem) Let { f k } E 1 be a sequence of measurable functions on a set E of . and (x). 9 5 f k 5 g a.e. f = lirnS. From the Dominated Convergence Theorem we get the Bounded Convergence Theorem. Pages 9192]. Theorem 3.f k l = 0. If the measure of E is finite. measurable functions from below by simple functions. see [Ro. By hypothesis.oo f k (x) almost everywhere.e.fk. To complete the proof. 1 j= E k00 lim inf f k 5 lim inf k+oo s.99 apply.finite measure. f is Lebesgue integrable and s.. we mentioned that for bounded functions we could also consider approximation from above by simple functions. note that If . f = hmJE f k . If f ( 5 )= limk. If 0 sE S.>I 5 29 (x)for all Ic and almost every x.fkI converges to 0 pointwise a. then constant functions are Lebesgue integrable over E . Corollary 3.f k (. lim lE . We now show that the two constructions .100 (Dominated Convergence Theorem) Let { fk}r=1 be a sequence of measurable functions defined o n a measurable set E . " ' If For a more traditional proof of the Dominated Convergence Theorem. Moreover. At the time. Since { fk}F=l converges to f pointwise almost everywhere.fkI = 0 and the proof is complete. by the first part of the theorem.98 and 3. Suppose that { f k ) E l converges to f pointwise almost everywhere and there is a Lebesgue integrable function g such that Ifk ) I(. l i m L If . 5 g (x)f o r all k and almost every x E E . 1k+oos u p L f k 5 im / E limsupfk = koo Thus. so Corollaries 3.110 Theories of Integration We can now prove Lebesgue's Dominated Convergence Theorem.
. we do not need to show that an upper integral equals a lower integral to conclude that a function is Lebesgue integrable. . Thus..m.cp) = 0. in E and it follows that f is measurable. Choose 1 and L such that 1 I f (z) < L for all z E E .. Then. for all k . q < f < u and which implies (3. Further. Then. which is 0 on the irrationals. . f is measurable o n E a i and only if. Then. .6 Riemann and Lebesgue integrals The Dirichlet function.102 Let E be a measurable set with finite measure and f . and define simple functions cp and $ by m i=l m i=l Then. there exist simple functions cpk and q k such that cpk 5 f L $k on E and fE ($k .Lebesgu.c p k ) < Define cp and $ by cp (z) = supk cpk (z) and $ (z) = infk $k (x). suppose (3. Let E > 0 and P = {yo.cp = 0 a. Thus.e. . 3. cp and ‘$ are measurable and cp I f 5 $ on E .L]with mesh p ( P ) < e. y1. for i = 1. in E . 0 $ = f = cp a. On the other hand.4) holds. Thus.e integral 111 lead to the same value for the integral. Therefore.Ym} be a partition of [Z. E + R* be bounded. $ . Set Ei = {x E E : yi1 5 f ( z ) < l i } . Lebesgue integrability does not imply Riemann integrability. Proposition 3.4). Suppose that f is measurable. provides an example of a function that is Lebesgue integrable but is not Riemann integrable on any interval. Proof.e. so that fE ($ .. The next result shows that the Lebesgue integral is a proper .
P k ) .bl f = L ( f . f is Lebesgue integrable and the two integrals are equal. f .bl Uk = ( f . Z 1 .a . z j ] ) i=l > . z i(x>+ m j x [ z j . P k ) and Jla.a . Since the partitions ! k are nested.b] k . so that b b . decreases. b ] lk = lim L (f.b]3 R be Riemann integrable. (1) 2() (3) Let { }= &: I be a sequence of partitions of [a. .112 Theories of Isntegration extension of the Riemann integral.=lQj. In the proof below. that satisfy conditions ( ) (2). a s f = s. . we use Riemann integral. and (3).103 Let f : [a. Set mi = inf {f ( t ): xi1 5 t 5 xi} and Mi = sup {f (t) : xi1 5 t 5 xi}. then {Pk}E1 is a sequence of nested partitions.) = limk uk ( 2 ) . P k C P k + l . . Fix k and suppose P k = { Z o . By the Monotone Convergence Theorem.l . Then. u 1 = lim la. Rs for the Theorem 3. b ] uk = lim '*O0 u ( f . P k )= Since f is Riemann integrable. Proof.+ mi . Define 1 and u by 1 (z) = limk lk (x) and u (12.l . ( so that J[a. I. k ) = P ktm a Sb b f and u = lim Ja. and define simple functions lk and uk by j1 lk and (2) = ~ m i x [ z i .)Xj}.. it follows that lk 5 f 5 uk and the sequence {lk}E1 increases and {uk}.b] ktw l .b] such that: If we then set P k = U.
in [a. we refer to the notation used in the proof of Theorem 3. 1 = f = u a.b] and There is no direct comparison of the Lebesgue and CauchyRiemann integrals. . Again. Thus. Mikusinski (see [Mil]. We saw in Theorem 3.e. n } .e.7 Mikusinski's characterization of the Lebesgue integral We next give a characterization of the Lebesgue integral due t o J. The characterization is of interest because it involves no mention of Lebesgue measure or the measurability of functions. % 3. By considering f+ and f . Without loss of generality.b]. f is finite valued a.104 Let f : R" + R* be Lebesgue integrable. . since f is Lebesgue integrable. . The characterization will be utilized in the next section where we discuss F'ubini's Theorem on the equality of integrals on Rn for n 2 2 and iterated integrals. the sets are measurable with finite measure and increase to R". . When the function is Lebesgue integrable. Also.e.e. Theorem 3. The function f (2)= of Example 2. f is Lebesgue integrable over [a.67. we may assume that f is finite valued on all of R". Then. k ) . by step functions on a set of finite measure.Lebesgue integral 113 Thus.separately. as shown in that example.49 is CauchyRiemann integrable but. it is not Lebesgue integrable. because 1 5 f 5 u. Then. see also [MacN]). Let Bk = {z E Rn : f (z) Bk < k and xi E [lc. is not absolutely integrable. i = 1 . we may assume that f 2 0.67 that a measurable function can be approximated a. the Dirichlet function is an example of a Lebesgue integrable function that is not CauchyRiemann integrable. Hence. such that Proof. . that the Lebesgue integrals of the step functions converge in a very strong sense. we can say more. In the following proof. there is a sequence of step functions ( ( P k ) F = 1 that converges t o f a.
which implies that j. we see that so that m n ( z )= 0. suppose that x 4 2 and f ( 2 ) is finite. I‘pk (x).f I). Since 2 1 c urZmFk for all m. H t and Gf relative to the function f X B k .114 Theories of Integration Using the notation of Theorem 3. For. ufzi (GfAAf). 1 as we wished to show.. Then.( < Thus.. Since Fk = Since Fk c it follows that so that limk.(Pkl = 0.f I = 0. for k 2 N . there is an m such that x Uk.converges to f ( x )for almost every x $ 2. Up=m Fk. I f k . the support of f k is contained in B k and { f k } E 1 increases to f a. we see that mn ( p k ) 5 2‘+’. 0 . we define the function f k and the sets Af.x $ Fk and x E B k .1 converges to f ( x ) for almost every x 4 2.+OO 1 f k . Define step functions (Pk by k2“ i=l i1 2k xG: and set { x E BI.”=.67. } Then. limk. Set 2 = ng=. w e claim that {(Pk(x)}:.e.Fk and a j such that x E Bj. in Rn. Then. Set N = max {m. : ) ( P k ( x ) f(x)1 2 2”). {(Pk (x)}. By the Monotone Convergence Theorem. so that 4 +.
. Lemma 3.‘ P k j . O r . where we define qko= 0. there is a sequence of bricks { B k } p = l such that cp=lm n ( B k ) < 00 and X B (t)= ~ f o r all t E E . for each i E N.. Arrange the doublyindexed sequence m 2 each {Kij}ieN.”=. we say that the series is absolutely convergent to f a.jeui into a sequence { B k } E 1 . Since t E E C UjEciKij for all i. we will use the following two lemmas.Lebesgue integral 115 To prove the Mikusinski characterization. ( E )= 0. c:=l Thus. Then. and / ((Pk . the sum of the measures of the bricks is finite but each t E E belongs to infinitely many of the bricks.e. Proof. 1 2 I 3 1.104.e. there exists a sequence of step functions {$k).”=. that is. Then. Proof. Finally. cr=l Lemma 3. there is a sequence of step functions { ( p k } r = 1 which converges to f a.q k j < 5 .105 Let E be a null set and E > 0.106 Suppose f : R” + R* is Lebesgue integrable.=1 such that the series $k converges to f absolutely a.e. Since m. By Theorem 3. if (z)l is finite for almost every z. + f a. xEl 00 00 k=l i=l 0 Suppose that a series of functions C.f I + 0. and c. If the series converges absolutely.. converges to a function f ?+hk pointwise (almost everywhere). = f a.e.there is a countable collection of open intervals {Iij : j E ai} covering E such that CjEai n (Iij) < ~ / 2 m Let Kij be the smallest brick containing Iij. t belongs to infinitely many bricks B k so that xBk( t )= 00.. Set Gj = ‘ P k j .. Then.e. Since &=l$j K = ‘Pk.?+h. there is a subsequence 00 of {(Pk}r=1 such that / l‘Pkj+l . SO that E c UjEaiKij and CjEain(Kij) < ~ / for ~ i. Thus.i 1 { ‘ ~ k j j=1 for j c.
" = .78. a. I'pjl is Lebesgue integrable and the series lpj converges in R for almost all x E R". By Corollary 3. Then. By (Z). Choose { $ k } .=.e. f is Lebesgue integrable zf. Thus. Since If I 5 Cj"=.e.116 Theories of Integration by cFl (. Theorem 3.1 . then x @ E and )I(. /pk < 00. Suppose first that such a sequence of functions exists. hence. the almost everywhere limit of a sequence of measurable functions. xB. C z . is Lebesgue integrable and hence converges in R for almost all x E R". equal to 0. Corollary 3.. Let {Bk}:=1 be the bricks corresponding to E in Lemma 3. the series 0 + j is absolutely convergent to f a. by Lemma 3.106 and let E be the null set of points at which l$k ) I(. 'Pk= { if k = 31 . the Dominated Converlqj gence Theorem implies that f is Lebesgue integrable and f = (Pk. diverges.105. is finite and.=. then the series I'pk (x)I = 00 by construction.=.)I xi"=. Moreover. f (x)= C:. by ~ x. Proof. Define a sequence of step functions { ( P ~ } . c g l I+jl Mikusinski characterized Lebesgue integrable functions as absolutely convergent series of step functions.107 Let f : R" + R*. and only iJ there is a sequence { ( P k } T = 1 of step functions satisfying: (1) (2) I n either case. Now. c."=. is measurable. . ? .) I s cT=l s EL.I(. xBI if k = 31 $l If If x E E . suppose that f is Lebesgue integrable. 'pk(x)at such points and f . so that (Pk (x)= + k (x)= f (x)a.e.=l c.78.2 xBl if k = 31 . c. c..
then I = I1 x 12.egral on the left is a Lebesgue integral in R2 and the expression on the right is an iterated integral. for each y E R. we make the agreement that if a function g is defined almost everywhere. We are interested in equalities of the form in which the int.Lebesgue integral 117 by Lemmas 3. These conditions can be utilized to gave a development of the Lebesgue integral in R" which depends only o n properties of step functions and not o n a development of Lebesgue measure.R* by . For each z E R. f Y (2) = f (z. then g is defined to be . Similarly. the results remain valid in = R" x R". For convenience. by linearity. or [MM]. i = 1. it holds for Lebesgue integrable step functions. For the remainder of this section. Remark 3.we can view f as a function of two real variables. where Ii is an interval in R.2. y). Since Fubini's Theorem holds for characteristic functions of intervals and. If f is a function on R2. Theorems which assert the equality of integrals in R" with iterated integrals are often referred to as "Fubini Theorems" after G.108 Note that i f (1) and ( 2 ) hold. we will treat the case n = 2.105 and 3.y).106.8 Fubini's Theorem The most efficient way to evaluate integrals in R" for n 2 2 is to calculate iterated integrals. f (z. we will use Mikusinski's characterization of the Lebesgue integral in Rn to establish a very general form of Fubini's Theorem. where z.e. Note also that the conditions (1) and ( 2 ) contain no statements involving Lebesgue measure. define a function fz : R +R* by fz (y) = f (z. then f = C:=. Fubini (18791943).y E R. 3. we define f Y : R . y) . Suppose f : R2 = 118 x R + R*. In this section. cpk a. If f = xI is the characteristic function of an interval in R2. For such a n exposition. see [DM]? [Ma2]. and f is measurable.
e.y>. By Mikusinski's Theorem."=. y) dy is Lebesgue integrable over R. if cr=. c.y). if {gk}F=l is a sequence of measurable functions which converges a. (3) the following equality holds: Proof. which we now prove. n o 0 n 00 n n which implies that I'pk (x.Y) by (ii) 4 . < 00. there is a sequence of step functions on R x R such that: J R x R l'pkl l'pk i. 1By Corollary 3. Theorem 3.for almost every x E R. 11. Corollary so that C:=. (2) the function x w J fx = JR R f (x. 'pk (x. we may assume that there exists a measurable function g : R" + R* such that {gk}& converges t o g a.78 implies there CzlJR < 0 0 is a null set E c R such that for all x $ E . where the null set may 0 depend on II: E. 111. I'pk (x.78.y)l dy 3. f (2. For such a pair (x. Y ) l < 0 then f 0 'pk. Now. JRxR f = cr=1 JRx. for x $ E . y)I < 0 for almost all y E R. y) = 'pk (x. the fact that Fubini's Theorem holds for step functions. This situation is encountered several times in the proof of Fubini's Theorem. Then: (fibini's Theorem) Let f : R x R +R* be Lebesgue (1) fx is Lebesgue integrable in R .e. = CE.. ( X .118 Theories of Integration equal t o 0 on the null set where g fails to be defined. Thus. and (i). in R".109 integrab le.
e. Theorem 3.5). when f is nonnegative. Tonelli’s Theorem. and The main difficulty in applying F’ubini’s Theorem is establishing the integrability of the function f on R2. then f Y is Lebesgue integrable on R for almost every y E R. for x 4 E . dy. we have so that (3) holds. fz = xrz1 (‘pk)z a. dy is measurable o n R. named after L. (2) holds. then IN c k = l (cpk)zl 5 and the function on the right hand side of the inequality is Lebesgue integrable over R by (3. Then: (1) fz + R* be nonnegative is measurable o n R for almost every x E R. the function y I+ JR f (z. in particular. guarantees the equality of multiple integrals and iterated integrals for nonnegative functions. y) If x 4 E .110 (Tonelli’s Theorem) Let f : R x R and measurable. if f is Lebesgue integrable on R2.. 0 Fubini’s Theorem could also be stated in terms of f Y . we get the equality of the double integral with the iterated integral. Tonelli (18851946). y) dx in Lebesgue integrable over R. in this case.and Jw fz = C z l Jw cp (x. and by (iii). y) . By the Dominated Convergence Theorem. l . I (2) the function x JR fz = JR f (x. the Dominated Convergence Theorem implies that fz is I Lebesgue integrable over R. However. Thus. Since EL1( ( P k ) .Lebesgue integral 119 and. proving (1). Thus. f is Lebesgue integrable if either the integral of f or the iterated integral is finite.
is Lebesgue integrable for almost all x and since { ( f k ) . JRz fk = JR Jwfk (x. f. If1 If . dy is measurable and the Monotone Convergence Theorem applied to (3. = Jw f (x. For alternate proofs of the Fubini and Tonelli Theorems. E E } . y) As an application of Fubini’s Theorem. the function x I+ Jw f. By Fubini’s Theorem. y) y). Tonelli’s Theorem can be used to check the integrability of a measurable function f : R2 + R*. JRz f = JR JR f (x.. (x.k } ) xI. for y E R. For each k . on R. set fk (2. so that fk is Lebesgue integrable over R2. Jwz f = limk JRz fk. If the iterated integral JR JR (x. the ysection of E at y is y) defined to be E = {x : (x. is measurable for almost every x.k . k ] so that U r = l I k = R2. y)I then and. dydx and since {fk}rZl increases to f pointwise.120 Theories of Integration (3) the following equality holds: Proof.47. .58.6) yields y) By F’ubini’s Theorem. by F’ubini’s Theorem.k . (fk). Similarly. See Exercise 3. y) = (max { f (x. } z l increases to f. k ] x [ . by the Monotone Convergence Theorem. = {y : (x. Let I k = [ .7) implies Note that we cannot drop the nonnegativity condition in Tonelli’s Theorem. By the Monotone Convergence Theorem y) By Theorem 3. the xsection of E at x is defined t o be E. we show how the area of a bounded subset of R2 can be calculated as a onedimensional integral. see [Ro. dydx is finite. E E } . Combining this with (3. If E c R2 and x E R. dydx. pages 3033091. f are Lebesgue integrable by Tonelli’s Theorem and then. consequently. We have the following elementary Y y) observations.
b] + R be nonnegative and continuous. Thus. (2) the function x I+ m ( E x )is Lebesgue integrable over R.113 Let f : [a. Then. xE (x.Lebesgue integral 121 Proposition 3. we can use this result to compute the area under the graph of f. From this proposition and Tonelli’s Theorem. By considering the points where x < a .48 and 3. one sees that the complement of E is an open set.b]. Then. (3) m2 ( E )= J m (E. c R2. (1) f o r almost every x E R. Example 3. E. the sections Ex are measurable.111 Let E . .112 Let E c R2 be measurable. by the previous theorem. x > b.49. y < 0 and y > f ( x ) separately. (x. R When f is a continuous function on an interval [a. See Exercises 3. the region under the graph of f is the set E = {(x. m ( E ) = JR m ( E x ) 2 dx. and (y) only if. Then. a E A. and x E R. (1) (2) (3) (4) For example.y) E E . = xE. y) = ( x E ) .b] and 0 L: y 5 f (x)}. (y) since all three equal 1 if. Notice that which implies that This result can be used to compute the area and volume of familiar regions.) dx. y) : x E [a. so that E is closed and hence measurable. we have Theorem 3.
x = 0. then 11 11 induces a (semi)metric d. If 11 11 is a (semi)norm on X.y / in z R and d (z. For a proof of the triangle inequality in 1 R" . z E X : (1) (2) [symmetry]. It is common for a (semi)metric in vector space to be induced by a function called a (semi)norm.y[I (see Exercise 3. A seminorm 11 11 is called a n o r m if.12. The metric d is called the discrete metric or the distance1 metric.52). often denoted dl 11. Example 3. z = y. d )is called a (semi)metric space. If X is a nonempty set. o o ) which satisfies for all z. [triangle inequality].9 The space of Lebesgue integrable functions The space of Lebesgue integrable functions possesses a natural distance function which we will study in this section and use to contrast the Lebesgue and Riemann integrals. Standard examples of metrics are the function d (z. y) = 1 . [0. a semimetric on X is a function d : X x X . y. a seminorm on X is a function 11 11 : X +[ O .122 Theories o f Integration 3. If X is a real vector. then the pair ( X . x E X: (1) (2) (3) for all Inequality (3) is known as the triangle inequality. For example. the function d : S XS + [0.yII in R". y) = 0 if.00) Oifx=y lifxfy defines a metric on S. l y) 1 . 00) which satisfies for all z.114 defined by If S is any nonempty set. and only if. defined by d (x. = 12 . and only if 11 (4) llzll = 0 if. y) = 12 . From (2) it follows that 101 = 0. If d is a (semi)metric on X . and only if. y. see Exercise 3. A semimetric d is a metric if (3) d (z.
xk) < E whenever k 2 N . while the subset Q of rational numbers is not complete under this metric. R" is complete under its natural metric. in E . See Exercise 3. it suffices to show that a sequence is Cauchy in order to assert that the sequence converges. The RieszFischer Theorem asserts that L1 ( E ) is complete under the semimetric d l .e. f = 0 a. We first show that there is a subsequence { f k j } j " . Thus.called the L1norm. Then. there is a n f E L1 ( E ) such ( that { . and only if. there is an N E N such that j . be a Cauchy sequence in (L1 E ). We call x the limit of the sequence { x k } p = 1 . k 2 N implies that d ( x j . Then.13. f k ) F = l converges to f in the metric d l .Le besgu e integral 123 the standard distances in R and R" are induced by the norms lltll = I1 for t t E R and 1x1 = (Cy=l q l 2 ) 11 l for x E R".g ) = . Completeness is an important property of a space since in a complete space. Fischer (18751954).116 (RieszFischer Theorem) Let E E M . By the triangle inequality. Proof. We define an integral seminorm 11 Ill on L1 ( E ) . A sequence { x k } r = 1 C X is said to converge to x E X if for every E: > 0. For example. Theorem 3. sEIf Example 3. every discrete metric space is complete. by setting 1/2 llflll = / If1 E * The semimetric d l induced by 1) )Il is then d l (f. every Cauchy sequence {xk}El c S converges to an element of S since the sequence must eventually be constant. for all f . g E L1 ( E ) . C { f k } z l which converges to f E L1 ( E ) _ . Let E c R" be a measurable set and let L1 ( E ) be the space of all realvalued Lebesgue integrable functions on E . Let { f k } r = 1 C L1 ( E ) be a Cauchy sequence. R is complete under its natural metric. A sequence { x k } z 1 C X is called a Cauchy sequence in X if for every E: > 0. A semimetric space is said to be complete if every Cauchy sequence in (X. every convergent sequence is Cauchy. d l ) . We establish a theorem due to F."=. then 1) Ill is a norm and d l is a metric on L1 ( E ) . x k ) < E .d ) converges to a point in X. and let { fk). Since llflll = 0 if. d ) be a discrete metric space. However.91. Similarly. Let d be a semimetric on X. 11 Ill is not a norm (and d l is not a metric). there is an N E N such that d (2.115 Let ( S . Riesz (18801956) and E. if we identify functions which are equal almost everywhere.
{ f k j } p l converges to f f k l .. { ~ j } ~ .fkj (x)} if the series converges absolutely 0 otherwise for almost every 2. Since f and f k l are Lebesgue integrable.f k j ) L 1. By the Dominated Convergence Theorem.e. by the Monotone Convergence Theorem. ~ increases 00 to the function g j=l Since s E g j 5 1.+l . we can pick a subsequence 1 {f k j } E l such dl (fkj+lj fkj) < 3’ It follows that 00 C l ( f k j + l . Define f by f (4 = Then. It remains to show that { f k } E l converges to f + f k l . d j=l Set gj = C’i=l defined by Ifk. g is Lebesgue integrable. { c” 1 . + + . f f k l E L1( E ) . f is absolutely Lebesgue integrable and as j + 00. Since that {fk}El is Cauchy. Thus.124 Theories of Integration a.fkiI.= {fkj+l (z) . Then.
3.1] + R by setting It is easily checked that { f k } E l is a Cauchy sequence in ( R([0.1]) since f is unbounded. then which implies that {fk}p=l converges to f +fkl in the metric dl. further justifying that the Lebesgue integral is superior to the Riemann integral.e.Lebesgue zntegml 125 f + fkl. If k > K . we show that the space of Riemann integrable functions is not complete under the natural semimetric d l . fkk + Fix > K .1] + R defined by This implies that f = g a. Note that another counterexample is provided by the functions in Example 3. Let R ( [ a b ] ) be the space of Riemann integrable func.b]. tions on [a. in [0.1]. However. .l]).117 Define f k : [0.10 Exercises Measure Exercise 3. Example 3. It follows from the Monotone Convergence Theorem that { f k } Z l converges in dl to the function g : [0. d l ) ) . For. suppose that { f k } z 1 converges to f with respect to d l . Since there is a K {fk}El > 0 such that is a Cauchy sequence in L1 ( E ) and for k j . { f k } : = l does not converge to a function in R((O.1 Prove that outer measure is translation invariant. kj Fix E > 0.1] so that the function f does not belong to R ([0. 0 In contrast to the case of the Lebesgue integral. 3. k > K .
7 Show that we can replace “there is a closed set F” in Theorem 3. l ) \ F is a finite or empty set} is an algebra. Exercise 3. . is Exercise 3.3 Let { a j k } T k . Prove that Lebesgue measure restricted to the Borel sets is an inner regular measure.j}E. l be a doublyindexed sequence of nonnegative terms such that a j k 5 ” j ( k + l ) for all J’ and k .5 Prove that A = { F C ( 0 .10 Prove that the complement of the Cantor set is dense in Show that every countable set is a Borel set.126 Theories of Integration Exercise 3.=i be a doubly indexed collection of intervals.4 Prove that every subset of a null set is a null set and that a countable union of null sets is a null set. l ) : F or ( 0 .6 Let X be a set and S c P ( X ) .K compact} for all E E l3 (R). = { B : S C B and B is a aalgebra} F .8 regular if A measure p defined for all elements of B (R) is called inner p ( E ) = sup { p ( K ) : K c E . Exercise 3.2 Prove that Let {Ii. 00 0 0 0 0 Exercise 3.9 [O. Let . 1 Exercise 3. 1 . Prove that Exercise 3.36 part (3) by “there is a compact set F”. Prove that C = nBEFx3 the smallest aalgebra that contains S . Exercise 3. Exercise 3.
Lebesgue integral 127 Lebesgue measure in R" Exercise 3.14 Prove that m.21 Give an example of a nonmeasurable function f on [0. that is. and only if.16 Let E C R". lim sup xi = lim inf xi. Exercise 3.50 is valid. E + h is measurable for all h E Rn. Exercise 3. given E C R" and u > 0..13 Prove that (R". and only . (1) Prove that E is measurable if.l(uE) = anrnt(E). c R.17 Suppose that E E x F is a null set in Rj+lC. Prove that limi400xi exists if.1] such that If1 is measurable. Exercise 3. Exercise 3. c Exercise 3. where aE = {y E R" : y = a x for some x E E } .22 Let xi}^. = y) . if.yll defines a metric on R".11 Prove the CauchySchwarz inequality. m: ( E + h ) = m: ( E ) .20 c R is a measurable set if. Exercise 3. n n Exercise 3. then E must contain a nondegenerate interval. y E R". and only if. That is. and only if.12 Use the CauchySchwarz inequality to prove that d (x. Measurable functions Exercise 3. aE is measurable for all a > 0. xE is Prove that the remark following Proposition 3. c R j is a null set and F c Rk.15 Prove that rn: is homogeneous of degree n. that is. d ) is a complete metric space. show that 1 yI = x xiyil 5 IIxII llyll by expanding the sum Izy. Exercise 3.l is translation invariant. if x.19 Prove that E a measurable function. (2) Prove that E is measurable if.18 Either prove or give a counterexample to the following statement: if E c R is measurable and m ( E ) > 0.Prove that Exercise 3. given E R" and h E R". m. Exercise 3.
Exercise 3. If E has finite measure and f is bounded.1] + R be continuous..28 Suppose that f : E + R is measurable. Show that the functions x + f (P) Lebesgue integrable for all m and f (x:") dx + f (0).128 Theories of Integration Exercise 3. that is.e.) = F (Ei)for all sequences of pairwise disjoint sets czl {&}& c M.29 Suppose that f is a bounded.32 Evaluate limn (1 + :)" e22dx.". See [Mu..25 Show that any derivative is measurable by showing that a derivative is the pointwise limit of a sequence of continuous functions.a ) . Prove that f / g is a measurable function if it is defined a. Exercise 3.33 Let f be Lebesgue integrable on R" and define F by F ( E ) = JE f d m n for all E E M.y) = f (x)g(y). are Jl Exercise 3. Prove that foris ( Lebesgue integrable and satisfies the linear change of variables + Exercise 3. Lebesgue integral Exercise 3.b].24 Suppose that f : E + R* is measurable. That is. Show that there is a sequence of bounded measurable functions { fk}T=l which converges to f pointwise on E.b]. then f ' is measurable on [a.30 Let f : E c R" R* be Lebesgue integrable and let a E R". Exercise 3.23 Suppose that f and g are measurable functions. Define f @ g : Rn+' = R" x Rk + R by f @g(z.26 Suppose that f : R" + R and g : Rk + R are measurable. F (U. if f : [a. Show that F is countably additive. measurable function on E and g is Lebesgue integrable over E . Exercise 3. Prove that f g is Lebesgue integrable over E. .E. Define f a ! : E + a + R* by for t ) = f ( t . Prove that If la is measurable for all a > 0. Prove that f @ g is a measurable function on Itn+'. Exercise 3.. Suppose that f : E + R is Lebesgue measurable and g : R + R is continuous. pages 1481491. Exercise 3. Note that we cannot conclude that f o g is measurable. Prove that g o f is a Lebesgue measurable function. Exercise 3.27 Let E c R" be a Lebesgue measurable set. show that f is Lebesgue integrable.31 Let f : [0.b] + R is differentiable on [a..
Can the boundedness condition be deleted? Exercise 3. Prove that infk fk is Lebesgue integrable over E . R* is Lebesgue integrable and + {x : () > k } .43 Suppose { is a sequence of Lebesgue integrable functions such that lfkl 5 M for all k . h : E c R” + R* are Lebesgue integrable over E and h 5 fk a.38 Prove Corollary 3.78 implies Theorem 3. ( A k ) + 0 as k +00. Show that f = 0.34 Suppose that f : R z+oo + lim lx+’ R is Lebesgue integrable. {&}El Riemann and Lebesgue integrals Exercise 3. y) dy.37 Exercise 3. zl Ak = If Exercise 3. prove that m. Exercise 3. Prove Corollary 2. sB Exercise 3. Show that F is continuous.39 Let f : [0.77. Show that the functions z + xkf (z) are Lebesgue integrable for all k and zkf(z)dz + $ 0. sE fk}E. show that limlzl+m (z) = 0. Suppose that f : A x B + R is continuous.1] : x A is Riemann integrable}.35 Suppose that fk.42 If f : R + R is Lebesgue integrable over R and uniformly continuous on R.45 Prove that a function which is absolutely CauchyRiemann integrable is Lebesgue integrable and the integrals agree. for all k .44 Let R = { A c [0.40 If f : E c Rn . Prove Exercise 3. f Exercise 3. Define F : A + R by F (z) = f (2.86. Exercise 3.46 Define f : R2 + R by .41 Let A c R j and B c Rk be compact sets. t k f k (z) is absolutely convergent for almost all z E E . F’ubini’s Theorem Exercise 3.e.1] + R be Lebesgue integrable. that R is an algebra which is not a aalgebra. Show that if satisfies Cr=l I t k ( < 00. Exercise 3.Lebesgue integral 129 Exercise 3. and hence show that the two are equivalent.36 Prove that Corollary 3.23. Exercise 3. then the series C:=.
Exercise 3. Exercise 3. show that the distance1 metric on X is not induced by a norm. is a Lebesgue integrable function on Rn+' and J.yil : 1 5 i 5 n } z i define metrics on Rn. Hint: Consider the integral over the set [0.+ .48 and Find the area inside of a circle of radius r and of an ellime Exercise 3.1].52 Let 11 11 be a (semi)norm on a vector space X .1]x [1.49 Find the volume inside of the ellipsoid . Prove that f @ g .130 Theories of Integration Show that but f is not integrable over [1. Prove that d (z.53 Show that and d.= 1. a2 b2 c2 x2 y2 z2 Exercise 3.+ . defined in Exercise 3.1]. y) = JIz.47 Define f : R2 + R by Compare Exercise 3. Exercise 3.50 Suppose that f : Rn + R and g : Rk + R are Lebesgue integrable.yII defines a (semi)metric.1] x [0. n+k f C gdmn+k 3 = JRn fdmn IRk gdmk* The Space of Lebesgue integrable functions Exercise 3. y) = max { I . .51 If X # (0) is a vector space.26. (z.
58 Suppose that f is Lebesgue integrable on Rn. 27rI. Suppose that cp is a step function on [0.55 L1 ( [ a bl).56 Exercise 3.27r]+ R is a Lebesgue integrable function. consider first the case where f is a step function. Show that the step functions on E are dense in L1( E ) .54 A set D in a semimetric space (S. Exercise 3. Prove that 2n 'p ( x )cos nxdx = 0.d ) is called dense if D = S. : Deduce from this the RiemannLebesgue Lemma: Suppose that f [0. Exercise 3.57 Prove that the continuous functions on [ q b ] are dense in Suppose that f E L1 (R). Show that limbtoo Jboo If\ = 0. where D is the union of D with the set of all of its limit points. . Show that To prove this. Exercise 3. Then noo lim fn f ( x )cos nxdx = 0. .Lebesgue integral 131 Exercise 3.
This page intentionally left blank .
we consider Part I of the Fundamental Theorem of Calculus for the Lebesgue integral and show that the Lebesgue integral suffers from the same defect with respect to Part I of the Fundamental Theorem of Calculus as does the Riemann integral. We show that f is not Lebesgue integrable. That is. Recall that Part I of the Fundamental Theorem of Calculus involves the integration of the derivative of a function f and the formula In Example 2. Unfortunately. not Riemann integrable. we would like to have an integral which integrates all derivatives and satisfies (4. and we showed in Theorem 2. we gave an example of a derivative which is unbounded and is. In the first part of this chapter. The function f is differentiable with derivative f satisfying f (0) = 0 and f (x)= 22 cos $+ ' ' ' 2n sin 5 for 0 < x 5 1. in order for ' (4. ' 133 . then (4. the example below shows that the general form of Part I of the Fundamental Theorem of Calculus does not hold for the Lebesgue integral. the assumption that the derivative f ' is Riemann integrable is required. we considered the function f defined by f (0) = 0 and f (z) = z2cos$ for 0 < x 5 1. It would be desirable t o have an integration theory for which Part I of the Fundamental Theorem of Calculus holds in full generality.1) holds. we gave a brief discussion of the Fundamental Theorem of Calculus for the Riemann integral.Chapter 4 Fundamental Theorem of Calculus and the HenstockKurzweil integral In Chapter 2.31. therefore. We then use this result to motivate the discussion of the HenstockKurzweil integral for which Part I of the Fundamental Theorem of Calculus holds in full generality.1 In Example 2.1) to hold. That is.31. Example 4.1).30 that if f is Riemann integrable.
134
Theories of Integration
If 0 < a < b < 1, then f is continuous on [a,b] and is, therefore, ' Riemann integrable with
r f = b2cos   a 2 cos.r '
b2
a2
and a k = ,/ , we see that Setting bk = I / & Since the intervals [ak,bk] are pairwise disjoint,
:: f J '
= 1/2k.
' Hence, f is not absolutely integrable on [0,1]and, therefore, not Lebesgue integrable there.
The most general form of Part I of the Fundamental Theorem of Calculus for the Lebesgue integral is analogous to the result for the Riemann integral; it requires the assumption that the derivative f be Lebesgue in' tegrable. This result is somewhat difficult to prove, and we do not have the requisite machinery in place at this time to prove it. In order to have a version of the Fundamental Theorem of Calculus for the Lebesgue integral, we prove a special case and later establish the general version for the Lebesgue integral in Theorem 4.81 after we discuss the HenstockKurzweil integral and show that it is more general than the Lebesgue integral.
Theorem 4.2 (Fundamental Theorem of Calculus: Part I) Let f : [a,b] +R be differentiable o n [a,b] and suppose that f' is bounded. Then, f ' is Lebesgue integrable on [a,b] and satisfies (4.i).
Proof. Note first that f is Lebesgue integrable since it is bounded by ' assumption and measurable by Exercise 3.25. For convenience, extend f to [a,b 11 by setting f ( t )= f ( b ) for b < t 5 b 1. Define f n : [a,b] + R by
+
+
n
By the Mean Value Theorem, for every n, n > and t E [a,b  ]; , there exists an S n , t E [a,b] such that f n ( t ) = f (sn,t). For t E [b  b ] , ' there is an S n , t E [a,b] such that
A,
i,
= n (b t )
(b)
bt
( t ) = n ( b  t )f ( S n , t ) , '
HenstockKurzweil integral
135
where n ( b  t ) 5 1. Since f is bounded, it follows that { f n } z = 1 is uni' formly bounded. Since { f n } r = l converges to f everywhere in [a,b], except ' possibly b, the Bounded Convergence Theorem shows that
By Exercise 3.30, the linear change of variables s = t last integral above shows that
+
in the next to
(4.2)
= noo lim
{1
n
j
b+
$
j
(s) cis  n
la+'
j ( t )d t }
The function f is continuous and, therefore, Riemann integrable so from the Mean Value Theorem (Exercise 2.18), for every n there are bn and an, b 5 bn 5 b + $ and u 5 an 5 U + $, such that nJbf'f = f (bn) and
Jaa+k f = f (an). Since bn from (4.2) we obtain
b, a,
+ a,
and f is continuous on [a,b
+ 11,
as we wished to show.
0
4.1
Denjoy and Perron integrals
Upon noting that the general form of Part I of the Fundamental Theorem of Calculus failed to hold for the Lebesgue integral, mathematicians sought a theory of integration for which Part I of the Fundamental Theorem of Calculus holds in full generality, i.e., an integral for which all derivatives are integrable. In 1912, A. Denjoy (18841974) introduced such an integration theory. His integral is very technical, and we will make no attempt to define or describe the Denjoy integral. Lusin later gave a more elementary characterization of the Denjoy integral, but this is still quite technical. For a description of the Denjoy integral and references, the reader may consult the text of Gordon [Go]. Later, in 1914, 0. Perron (18801975) gave another integration theory for which the Part I of the Fundamental Theorem of Calculus holds in full
136
Theories of Integration
generality. The definition of the Perron integral is quite different from that of the Denjoy integral although later Alexandrov and Looman [Pe, Chap. 91 showed that, in fact, the two integrals are equivalent. We will give a very brief description of the Perron integral since some of the basic ideas will be used later when we show the equivalence of absolute HenstockKurzweil integrability and McShane integrability.
Definition 4.3 Let f : [a,b] + R and x E [a,b]. The upper derivative of f at x is defined to be
Df
(2)
= limsup f
t+x
( t ) f tx
(4
f(t)f(x). tx
Similarly, the lower derivative is defined to b e D f (x)= liminft.+x Thus, f is differentiable a t x if, and only if, upper and lower derivatives are finite.
Bf (x) = of ( x ) and
both
Definition 4.4 Let f : [a,b] + R*. A function U : [a,b] +R is called a major function for f if U is continuous on [a,b], U ( a ) = 0, DU (x) > m and Q U ( x ) 2 f (x)for all x E [a,b]. A function u : [a,b]+R is called a m i n o r function for f if u is continuous on [a,b],u ( a ) = 0 , Du ( x ) < 00 and Du ( 2 ) 5 f ( 2 ) for all x E [a,b].
It follows that if f is differentiable on [a,b] and has finitevalued derivative, then f  f ( a ) is both a major and a minor function for f’. If U is a major function for f and u is a minor function for f , then it can be shown that U  u is increasing. Therefore,
00
< sup {u( b ) : u is a minor function for f }
5 inf {U ( b ) : U is a major function for f} < 00.
Definition 4.5 A function f : [a,b]+R is called Perron integrable over [a,b] if, and only if, f has a t least one major and one minor function on [a,b]and
sup {u ( b ) : u is a minor function for f }
= inf
{U ( b ) : U is a major function for
f} .
(4.3)
The Perron integral of f over [a,b] is defined to be the common value in (4.3).
HenstockKurzrueil integral
137
If a function f : [a,b] + R has a finite derivative on [a,b],it then follows from the definition that f’ is Perron integrable over [a,b] with Perron integral equal to f ( b )  f ( a ) . That is, Part I of the Fundamental Theorem of Calculus holds in full generality for the Perron integral. For a description and development of the Perron integral, see [Go]and [N]. Both the Denjoy and Perron integrals are somewhat technical to define and develop, but in the next section we will use Part I of the Fundamental Theorem of Calculus as motivation to define another integral, called the HenstockKurzweil integral, which is just a slight variant of the Riemann integral and for which Part I of the Fundamental Theorem of Calculus holds in full generality. It can be shown that the HenstockKurzweil integral is equivalent to the Denjoy and Perron integrals (see [Go]).
4.2
A General Fundamental Theorem of Calculus
Suppose that f : [a,b] + J is a differentiable function and we are interested R in proving equality (4.1). Let P = {zo, 21,. . . , xn} be a partition of [a,b]. By the Mean Value Theorem, there is a yi E (zi1, xi) such that f (xi)f (xi1) = f’ (yi) (xi xi1). Thus, given any partition P , there is a set of sampling points {yl, . . . , g n } such that
The problem is that given a partition P , there may be only one such set b of sampling points. However, if we want to show that f is equal to ‘ f ( b )  f ( a ) , we do not need the Riemann sums to equal f ( b )  f ( a ) , but rather be within some prescribed margin of error. Thus, we are led to consider more closely the relationship between f‘ (yi) (zi+l  xi) and
s ,
f (Xis1)

f (Xi).
Fix an E > 0 and let y E [a,b]. Since f is differentiable a t y, there is a 6 (y) > 0 so that if z E [a,b] and 0 < ( z  y ( < S (y) then
138
Theories of Integration
Multiplying through by Iz  yI, we get
which is also valid for z = y. Now, suppose that u, E [a,b] and y  S (y) v u 5 y 5 < y 6 (y). Then,
+
<
So, if y  S(y) < u 5 y 5 v < y + 6 (y), then f‘(y) (v  u) is a good approximation to f ( u )  f (u). This result, known as the Straddle Lemma, will be useful to us below.
Lemma 4.6
y E [a,b]. For each
(Straddle Lemma) Let f : [a,b] + R be differentiable at E > 0 , there is a S > 0, depending o n y, such that
whenever u,v E [a,b] and y  6 < u 5 y 5 v
< y+6.
The geometric interpretation of the Straddle Lemma is that the slope of the chord between ( u , f ( u ) ) and ( v , f ( v ) ) is a good approximation to the slope of the tangent line at ( y , f ( y ) ) . It is important that the values u and v “straddle” y, that is, occur on different sides of y. Consider the function f equal to x2 cos ( n / z )for z # 0 and f (0) = 0. This function has derivative 0 for z = 0, but for u = 4 and v = the slope of the chord 2n+ 5 joining (u,f (u)) ( v , f ( v ) ) is and
&,

cos 2nn 
((*)
1 1
2
cos (2nn
+ $)
)
(39;
1 
>2.
2n
2n
2n+4
Thus, if u and u do not straddle 0, then the slope of the chord is not a good approximation to the slope of the tangent line. This lemma already gives us a hint of how to proceed. When studying Riemann integrals, we chose partitions based on the length of their largest subinterval. This condition does not take into account any of the properties of the function being considered. The Straddle Lemma, on the other hand, assigns a 6 to each point where a function is differentiable based on how
we would expect 6 to be small. There is another point that must be resolved. . if the function oscillates wildly near the point.b] (0. x n } and sampling points {yi}y=i such that . a tagged partition is a finite set of ordered pairs D = { ( t i .b] and the intervals have disjoint interiors.. xi] C ( y i . .. x m } is a partition of [a. This simple change to varying the size of 6 from point to point is the key idea behind the HenstockKurzweil integral. The point ti is called the tag associated to the interval Ii. where 6 : [a. [ X i . xi]. .q . if P is .. we consider Riemann sums for every set of sampling points { y i } z 1 such that yi E [ x i . Let Ii = [ Z i .HenstockKurzweil integral 139 the function acts near that point. P = (xo. + + + + Definition 4. In the general case. Remark 4. then [ X i . such that yi E [xil.l .oo) is a positive function. ~ i ]However. yi S) for every sampling point yi E [ Z i . Suppose P = { Q .8 B y the preceding argument. .l . given a partition P = ( 2 0 . : i = 1.. . Thus. . xi] c (yi . If the function acts smoothly near the point. so that ti E Ii. x i ] . Given a tagged partition D . . This can happen when a tag is an endpoint for two adjacent intervals and is used as the tag for both intervals.b] and t is a point in I . . xn}.. we will be interested in partitions P = { x o .7 Given an interval I = [a.m } . ti E Ii.b] c R.6 ( y i ) . renumIi) ber the pairs so the right endpoint of Ii1 equals the kft endpoint of Ii and set Ii = [xil. a tagged partition generates a partition and a set of associated sampling points.6 . in which 6 is a positive function of y. : i = 1.. . In other words.b].l . .l . a partition with a set of sampling points generates a tagged partition (by setting Ii = [ X i . For the HenstockKurzweil integral.xi].b] and is a set of sampling points associated to P .l . x n } is a partition of [a. x ~. we will consider only partitions P and sampling points {yi}y==. Similarly. xi]). . where I is a subinterval of [a.. Given a tagged partition 2) = {(ti. U z l I i = [a. In the Riemann theory. . We use this idea to determine which pairs of partitions and sampling points to consider. . namely the relationship between the partition and the sampling points. xi] C ( y i . we can view a partition together with a set of sampling points as a set of ordered pairs ( t . I: n Ij” = 8 if i # j .a point can be a tag for at most two intervals. m } such that Ii is a closed Ii) subinterval of [a. we would expect the associated 6 to be large. . .b].I ) . Fix [a. a partition with mesh at most S. Then.b] and .l .pi 6 ( Y i ) ) . .6 ( Y i ) . x ~. a tagged partition is a partition with a distinguished point (the tag) in each interval.x1. yi 6 (yi)).
140
Theories of Integration
E Ii. Note that while a partition is a n ordered set of numbers, the intervals in a tagged partition are not ordered ( f r o m left to right), so we must .first reorder the intervals so that their endpoints create a partition of [a,b].
ti
Next, we need a way to measure and control the size of a tagged partition. Based on the discussion leading to the Straddle Lemma, we will do this using a positive function, 6, o f t .
Definition 4.9 Given an interval I = [u,b],an intervalvalued function y defined on I is called a gauge if there is a function S : [a,b] +(0,oo) such that y ( t )= ( t  6 ( t ),t S ( t ) ) .If 2) = {(ti, : i = 1 , . . . ,m } is a tagged Ii) partition of I and y is a gauge on I , we say that D is yfine if Ii c y(t;) for all i. We denote this by writing D is a ?.fine tagged partition of I .
+
Let P = { x o , x ~ ., . , xn} be a partition of [a,b] with mesh less than S and let {yi}El be any set of sampling points such that yi E [xi1,xi]. If y ( t ) = (t  6, t 6 ) for all t E I , then [xil, xi] c y (yi) so that D = {(yi, [xil, xi]) : i = 1,.. . ,m } is a yfine tagged partition of [a,b]. This is the gauge used for the Riemann integral. Consequently, the constructions used for the Riemann integral are compatible with gauges. The value of changing from a mesh to a gauge is that points where a function behaves nicely can be accentuated by being associated to a large interval, and points where a function acts poorly can be associated to a small interval.
+
Example 4.10
The Dirichlet function
f : [0,1]+R,
defined in Example 2.7, is not Riemann integrable. This function is equal to 0 most of the time, so we want a gauge that associates larger intervals to irrational numbers than it does to rational numbers. Let {Ti}& be an enumeration of the rational numbers in Qn [0,1]. Let c > 0 and define 6 : [0,1] + (0,oo) by 6(x) = Then, c if x $ Q 2Zc if x = ri E Q
‘
HenstockKurzweil integral
141
and every irrational number is associated to an interval of length 2c while the rational number ri is associated to an interval of length 21ic. After introducing the HenstockKurzweil integral, we will use this construction to prove that the Dirichlet function is HenstockKurzweil integrable. If D = { ( t i ,Ii) : i = 1,.. . , m } is a tagged partition of I , we call
rn i=l
the Riemann s u m with respect to V. Let us restate the definition of the Riemann integral in terms of tagged divisions.
Definition 4.11 A function f : [a,b] + R is Riemann integrable over [a,b]if there is an A E R such that for all E > 0 there is a 6 > 0 so that if 2) = { ( t i ,[zi~, : 1 5 i 5 rn} is any tagged partition of [a,b] satisfying xi]) [xi+ xi] c (ti  6 , ti S), then IS (f,D) A < E . 1
+
Note that the mesh of this partition is at most 26. For the Riemann integral, the partitions are chosen independent of f. Thus, this definition fails to take into account the particular function involved. A major advantage of the HenstockKurzweil integral is one only need consider partitions that take the behavior of the function into account.
Definition 4.12 Let f : [a,b] + R. We call the function f HenstockKurzweil integrable on I = [a,b]if there is an A E R so that for all E > 0 there is a gauge y on I so that for every yfine tagged partition D of [a,b],
The number A is called the HenstockKurzweil integral of f over [a,b], and b we write A = f = f .
s ,
sI
The HenstockKurzweil integral is also called the gauge integral and the generalized Riemann integral,
Notation 4.13 For the remainder of this section, we will use the symbols f and f to represent the HenstockKurzweil integral o f f .
s,"
sI
The first question that arises is whether this definition is meaningful. We need to know that, given a gauge y, there is an associated yfine tagged partition, so that we have Riemann sums to define the HenstockKurzweil
142
Theories of Integration
integral, and also that the HenstockKurzweil integral is well defined. We will return to both issues at the end of this section. Observe that every Riemann integrable function is HenstockKurzweil integrable. For, suppose that f is Riemann integrable. Let 6 correspond to a given E in the definition of the Riemann integral. Set y ( t )= (t  t $) , Then, any tagged partition that is yfine has mesh less than 6. Thus, we have proved
4, +
Theorem 4.14 If f : [a,b] 3 R is Riemann integrable then f is HenstockKurzweil integrable and the two integrals agree.
However, there are HenstockKurzweil integrable functions that are not Riemann integrable. In fact, the Dirichlet function is one such example.
Example 4.15 Let f : [0, I] + R be the Dirichlet function. We will show that f = 0. Let E > 0 and let y be the gauge defined in Example 4.10 with c = 6. Let V = { ( t i , i ) : i = 1 , . . . , m } be a yfine tagged partition of I [ O , l ] and note that
#
To The sum for ti 4 DnQ equals 0 since f ( t )= 0 whenever t # Q. estimate the sum for ti E VnQ, note that j ( t i ) = 1 since ti E Q and recall that each tag ti can be a tag for at most two intervals. Since ti E Qn[O,l], there is an j so that ti = r j . Thus, if ( t i , I i ) E D,then Ii c y(ti), so that ! ( I i ) 5 ! ( y ( t i ) ) = l ( y ( r j ) )= 2lj:. Thus,
We have shown that given any E > 0, there is a gauge y so that for any yfine tagged partition, D,IS (f,27) 0) < E. In other words, the Dirichlet 1 function is HenstockKurzweil integrable over [0, I] with Jo f = 0.
Hen,stockKu.rziiteil integral
143
Notice the use of the variable length intervals in the definition of the gauge. We will give a generalization of this result in Example 4.38; see also Exercise 4.7. Let us return to the Fundamental Theorem of Calculus. The proof is an easy consequence of the Straddle Lemma.
Theorem 4.16 (Fundamental Theorem of Calculus: Part I ) Suppose that f : [a,b] + R is diflerentiable o n [a,b]. Then, f is HenstockKurzweil ' integrable o n [a,b] and
Proof. Fix an c > 0. For each t E: [a,b],we choose a 6 ( t ) > 0 by the Straddle Lemma (Lemma 4.6) and define a gauge y on [a,b] by y ( t ) = (t  6 ( t ),t S ( t ) ) . Suppose that D = { ( t i , i ) : i = 1,.. . , m } is a yfine I tagged partition of [a,b]. We reorder the intervals Ii so that the right endpoint of Ii1 equals the left endpoint of Ii, and set Ii = [zi1,zi] for each i. Then,
+
f (b)  f ( a )=
so, by the Straddle Lemma,
I m
c[f
m
i=l
(Xi) 
f (.idl
5
c
i=l
€
( i  .) Z & I
= € (ba ) .
Thus, f is HenstockKurzweil integrable and satisfies equation (4.1). '
0
Thus, every derivative is HenstockKurzweil integrable. This is not a surprising coincidence. Kurzweil [K] initiated his study leading to the HenstockKurzweil integral in order to study ordinary differential equations. A few years later, working independently, Henstock [He] developed many of the properties of this integral. We will establish a more general version of Theorem 4.16 later in Theorem 4.24. An immediate consequence of the Fundamental Theorem of Calculus is
144
Theories of Integration
that the unbounded derivative
IT 2r 7 IT 22 cos  + sin  if 0 < x 5 1 22 x 22 if x = O
9
defined in Example 2.31, is HenstockKurzweil integrable on [0,1] with integral equal to 1. Since f ' is unbounded, it is not Riemann integrable and, as we saw in Example 4.1, f is not Lebesgue integrable. ' Before concluding this section, we prove two results which guarantee that the HenstockKurzweil integral is well defined. We prove that given a gauge y, there is a related yfine tagged partition, and that the value of the integral is unique.
Theorem 4.17 Let y be a gauge on I = [a,b]. Then, there is a yfine tagged partition of I .
Proof. Let E = {t E ( a ,b] : [a,t] has a yfine tagged partition}. We want to show b E E . First observe that E # 8 since if x E ? ( a ) n ( a , b ) , then { ( a ,[a, is a yfine tagged partition of [a,4. Thus, x E E and E # 8. We next claim that y = sup E is an element of E . By definition, y 6 [a,b], so y is defined at y. Choose x E y (y) so that x < y and x E E , and let D be a yfine tagged partition of [a, Then, D = D U { (y, [x, ' y])} is a yfine tagged partition of [a,y]. Therefore, y E E . Finally, we show y = b. Suppose y < b. Choose w E y (y) n (y, b). Let D be a yfine tagged partition of [ a , y ] . Then, D' = DU ((9, [y, w])} is a yfine tagged partition of [a,201. Since y < w , this contradicts the definition of y. Thus, y = b. 0
XI)}
XI.
Thus, there is a yfine tagged partition associated to every gauge y. In fact, there are many, as we can see by varying the choice of x in the first step of the proof above. Finally, we prove that the HenstockKurzweil integral is unique, justifying our notation in Definition 4.12. The proof employs a very useful technique for working with gauges. Suppose that y1 and y2 are two gauges defined on an interval [u,b]. Then the (intervalvalued) function y defined by y ( t ) = y1 ( t )n y2 ( t ) is also a gauge on [a,b]. In fact, if 61 and 6 2 are the positive functions used to define y1 and y2, respectively, and S ( t )= min (61 ( t ), 6 2 ( t ) } , then y ( t ) = (t  6 ( t ), t 6 ( t ) ) .Further, if D is a yfine tagged partition, then D is also a ylfine tagged partition and a y2fine tagged partition, since for ( t ,I ) E V ,I c y ( t )c yi ( t ) ,for i = 1,2.
+
we will need to employ more sophisticated proofs. 6. . as we progress with this more advanced theory. we will begin with proofs that directly mimic the Riemann proofs.and partitions and sampling points.HenstockKurzweil integral 145 Theorem 4. P and with tagged partitions.B < E' D S 2) I + E' = €. the value of the HenstockKurzweil integral is unique. with gauges. then af f Pg is HenstockKurzweil integrable and Proof.b] +R and let a. Fix E > 0 and choose yf > 0 so that if D is a yffine tagged partition of [ a . and hence 2) is both ylfine and y2fine.12. respectively. review the proof of Proposition 2. Proposition 4. Of course. replacing positive numbers. 5.19 (Linearity) Let f. V.18 The HenstockKurzweil integral of a function is unique. 4. Then. in which we establish the basic properties of the HenstockKurzweil integral. 7 .then . in the definition with E' = Let y ( t )= y1 ( t )n y2 ( t )and suppose that V be a yfine tagged partition. it follows that A = B. Suppose that f is HenstockKurzweil integrable over [a. linearity and positivity. 0 Now. You will notice that the proof is exactly the same as the one above.8 E R. ( A.3 Basic properties We begin with the two most fundamental properties of an integral.3.b] and both A and B satisfy Definition 4.s (f. Proof. Thus.g : [a. Since E was arbitrary. ) l + I (f.B ( 5 IA . Iff and g are HenstockKurzweil integrable. y. b ] . Fix E > 0 and choose YA and yB corresponding to A and B . In the following section.
Consequently. if V is a yfine tagged partition of [a. [f > for any positive E. let y ( t ) = yf ( t )n yg ( t ) and suppose that D is a yfine tagged partition of [a. Suppose that f is nonnegative and HenstockKurzweil integrable."f 2 0.b]. S (f. . s.12. since S (f. Then. (dbf tP/b9) Since E was arbitrary.€ D > € It follows that s.b]. choose yg then > 0 so that if V is a 7. Now. Let E > 0 and choose a gauge y according to Definition 4.20 (Positivity) Let f : [a.146 Theories of h t e y m t i o n Similarly. Then. it follows that a f +Pg is HenstockKurzweil integrable and 0 Proposition 4. ) .b] + R. f 2 0. Then." Proof.b].fine tagged partition of [a.D)2 0.
b] : f (2) > so there must be a k such that m ({x E [u. We will see in Section 4. then [a. Suppose that f 5 g .b]is called absolutely integrable are HenstockKurzweil integrable over [a.b]. i}) i}. However.22 Suppose f and g are HenstockKurzweil integrable over [a. and a function is Lebesgue integrable if. Remark 4. If f i s HenstockKurzweil integrable. The Riemann integral has the same defect. { x E [a.b].b]. if both f and If1 Corollary 4.If1 5 f 5 I f l . But. and only if. then the best we can conclude i s that f 2 0. f r o m our results so f a r . it is absolutely (Lebesgue) integrable.4 that a HenstockKurzweil integrable function need not be absolutely integrable. Then.HenstockKurzweil integral 147 A comparison of the last two proofs with the corresponding proofs for the Riemann integral immediately shows their similarity. then. Applying the previous result to g . . For absolutely integrable functions.b] and f (x)5 g (x)f o r all x E [a. Let 12 Jab f be the Lebesgue integral o f f . libom Tchebyshev's inequality we have s.f yields Corollary 4.b]. A function f defined on an interval [u. A Riemann integrable function is absolutely (Riemann) integrable.21 Suppose that f i s a positive function o n [a. i f f is Lebesgue integrable.b]: f (x)> > 0.b]. we have the following result. we cannot conclude that the integral is positive. then Proof. then the Lebesgue integral o f f is strictly positive. the previous corollary implies that The result follows." A ({x E [a.b] : f m (2) > A}) 5 L If f is strictly positive o n [a.23 If f is absolutely integrable over [a. Since .b] = Ur=.
148 Theories o f Integration While the ability to integrate every derivative is a main feature of the HenstockKurzweil integral. by (1) and ( 2 ) . Suppose that F is continuous and F’ = f except f o r possibly a countable number of points in [a.b] + R.m } is a yfine tagged partition of Ii) [a. ~ be the points where either F‘ fails to exist or ~ F‘ exists but is not equal to f .. Choose 6 ( t )= 6 ( c k ) > 0 so that I . By the Straddle Lemma. then there is a j so that ai = bj. with a similar statement for each right endpoint bi # b. f : [a.b] and 1x . then t = ck for some k .24 (Generalized Fbndamental Theorem of Calculus: Part I. choose 6 ( t )> 0 for this E by the Straddle Lemma. If t E C . for Suppose that D = {(ti. Define a gauge y on [a.bi] for each i.b] \ C and D be the set of 2 elements of .ckl < S ( c k ) implies: x (1) (2) We can define such a 6 since F is continuous on [a. + If ti = ck for some k . The derivative can fail t o exist at a countable number of points and still satisfy equation (4.b] and Proof.6 ( t ).b] \ C . } . the HenstockKurzweil integral satisfies an even stronger result. .1). .b] by setting y ( t )= (t . Let D1 be the set of elements of D with tags in [a.b].b].) Let F. Let C = { c .. If t E [a. Note that if ai # a .b]. f is HenstockKurzweil integrable over [a. Let E > 0. Theorem 4. : i = 1. Then.t 6 ( t ) ) all t E [a. where Ii = [ai.D with tags in C.ckl can be made as small as desired by choosing x sufficiently close t o ck.
Therefore. occurs as both a left and right endpoint.5 that there are no improper integrals for the HenstockKurzweil integral.24. g : [a.We will show in Section 4. other than a and b. l ] and F' = f except at x = 0. (4. a this case. except for at most a countable number of points. f G is HenstockKurzweil integrable and. Suppose that F and G are continuous and F' = f and G' = g .26 (Integration b y Parts) Let F.13 F ( 2 ) = 2fi. Using Theorem 4. Then. Then.1] $ f = F (1) . f . and the result is established.24.24is important.5) . and Note that f is an improper integral in the Riemann sense since f is unbounded. F is continuous on [ O .16. Fg+ f G is HenstockKurzweil integrable and Jt Moreover. by Theorem 4.24.HenstockKurzweil integral 149 Therefore. see Exercise 4. since each ck can be a tag for at most two subintervals of D.F (0)= 2. n [Fg+lfG=F(b)G(b)F(o)G(o).f is HenstockKurzweil integrable over [0.b] + R. and f (0)= 0 by and f ( x ) = otherwise. 0 The continuity of F in Theorem 4.25 Define F and f on [0. Example 4. Theorem 4. and only i f . Fg is HenstockKurzweil integrable if. Since each endpoint. G . but we were able to show that f is HenstockKurzweil integrable directly from Theorem 4. can prove a general form of the familiar intewe gration by parts formula from calculus.
Note that for j > k . Fg = ( F g f G ) . D) f < 5 . For each k E N.4) valid. Since (FG)’ = Fg f G except possibly a t a countable number of points.4) since.150 Theories of Integration Proof. Then. by Theorem 4. fix a ykfine tagged partition Vk. is 4. So.61.24. we have l b] IS(f31). assume the Cauchy criterion holds. we give an example in which neither Fg nor f G is HenstockKurzweil integrable so that (4. s.” / Is(f. then Proof.4) holds.b] + IR is HenstockKurzweil integrable over [a. [a. 0 + + In Example 4. for example.f G. Theorem 4.choose a gauge yk > 0 so that for any two ykfine tagged partitions D and V2 of [a. even though (4.S ( f 9 2 ) l < Z’ Replacing yk by n$=lyj. Then. As in the case of the Riemann integral. since yj c yk.5) makes no sense.53 below. We will prove that f is HenstockKurzweil integrable. then IS (f.b] iJ and only if. b]. there is a gauge y so that if 2) is a yfine tagged partition of [a. the HenstockKurzweil integral is characterized by the Cauchy condition.n) S ( f 9 2 ) I I which is the Cauchy criterion. D is a j ykfine tagged partition of [a.27 A function f . (FG)’ is HenstockKurzweil integrable and (4. We have already shown that the integrability of f implies the Cauchy criterion. we may assume that yk+l c y k . For each k . Let D1 and D2 be two yfine tagged o partitions of [u.1 Cauchy Criterion Suppose that f : [a.b] and E > 0. 1 1 .b].3. Thus. f o r every E > 0 there i s a gauge y so that i f V1 and V2 are two yfine tagged partitions of [a. b]. The last statement follows immediately from (4.b] + IR is HenstockKurzweil integrable over [a.
Then 2' = V U (VI V2) ) U and E' = E U (V1 u V 2 ) are yfine tagged partitions of I . By the Cauchy criterion.b]. if f is HenstockKurzweil integrable over every closed subinterval . and hence converges. Of course. Let A be the limit of this sequence. We will use the Cauchy criterion in the following section. Suppose that D and I are 7fine tagged partitions of J . Then. and C D i is a 7. Theorem 4.b].Let V be a yKfine tagged partition of [a. Fix E > 0 and choose K > 2 / ~ . we need not consider it further. ~and J2 = [ d .HenstockKurzweil integral 151 which implies that the sequence {S ( f . Since V' and I' contain the same pairs ( z j .b] + R is HenstockKurzweil integrable over I and J is a subinterval of I . V l ) .3. Proof.b].)I < E. Let 7 = y l and ~ y i = yJJi.28 Let f : [a.12.fine tagged partition of J i .b] so that if V1 and V2 are two yfine tagged partitions of [a. is a Cauchy sequence in EX. If J c [a. then it is HenstockKurzweil integrable over every subinterval of I and the set function F ( J ) = JJ f is defined for all closed subintervals J c I .b]. then f is HenstockKurzweil integrable ouer J . i = 1 . Set J = [ a .b] + R be HenstockKurzweil integrable over [a. It is reasonable to expect that the HenstockKurzweil integral of f over J exists.d] be a closed subinterval of [a. Let J = [c. 4.V. f is HenstockKurzweil integrable over J .Vk)}T=.2 The integral as a set function Suppose that f : I = [ a . Let E > 0 and y be a gauge on [a.I j ) off of J . 0 Thus. It now follows that f is HenstockKurzweil integrable on [a.S(f. b ] 1 ] if either is degenerate. 2 .then I S ( f .b] is a closed subinterval.b]. if f is HenstockKurzweil integrable over an interval I . It follows from the previous inequality that It remains to show that A satisfies Definition 4.
C ) if x E [a. Write D as D1 U D2. and only if. then f is HenstockKurzweil integrable over I . the largest x such interval is ( c .c] and 1 2 = [c.b] .x (11: .c I ) if x E (c. If ( c .b] + R and let {Ij}YZl be a *finiteset of closed intervals with disjoint interiors such that [a. then l S ( f .x 111: . f ( c ) l ( J )= f ( c ) l ( J . Fix E > 0 and. a much stronger result is true. if necessary.b] = Uy=lIj. Actually. c is a tag for every yfine tagged partition. If x < c. D + + . for i = 1 .c( .cl). if x > c .b ] . and f is HenstockKuriweil integrable over both intervals.731) S (f.I . If f i s HenstockKurzweil integrable over each I j .(z . Suppose that 23 is a yfine tagged partition of [a.D is a yifine tagged i partition of Ii.JIi < 5 . i = 1 . By the construction of y.D ) = S (f. since I is a subinterval of itself.c ( ) = (x . Define a gauge on all of I as follows: fl + + y1 (z) n (Z. In order for f to be HenstockKurzweil integrable over I . 2 ) . Theorem 4. ) f ( c ) l ( J 2 ) .C ( . 2 .J ) E D : J c I i } . x 1 .c( . divide J into two intervals Ji = J n Ii. we have that S ( f .. with Ji c yi ( c ) . I1 = [a. Suppose first that [a. Thus.C ) y2 (z) n ( c .J ) E D and J has a nonempty intersection with both I1 and 1 2 .152 Theories of Integorntion J c I .29 Let f : [a. then the largest interval centered at x that does not contain c is (x . choose a gauge yi on Ii so that if D is a yifine tagged partition of Ii. D ). 2 . Then. After dividing the interval associated to the tag c.b]. x y 1 ( c ) n y 2( c ) if x = c + IZ Since c E y ( x ) if. c ) .b] and Proof.b] is divided into two subintervals. x = c. which is a consequence of the next theorem. it is enough to know that f is HenstockKurzweil integrable over a finite number of closed intervals whose union is I . similarly. where Di = {(x. then f is HenstockKurzweil integrable over [a.
Proof. 0 + A key point in the previous proof is defining a gauge in which a particular point ( c ) is always a tag. Let E > 0 and choose corresponding functions cpl and cp2. Sr Lemma 4. every step function defined on an interval is HenstockKurzweil integrable there. f o r every e > 0. if D1 and D are yfine tagged partitions of I then 2 This implies that By the Cauchy criterion. There are gauges y1 and y2 on I so that if D is a yifine tagged partition of I . one can design a gauge y that forces a finite set of points to be tags for every yfine tagged partition.2.HenstockKurzweil integral b 153 Thus.~.D) .l ~ t ~ z(it ) x. cp.. suppose that f is a continuous function on [a. .b] with canonical form aiXI. there are HenstockKurzweil integrable functions cpl and p2 such that p1 5 f 5 9 2 o n I and S I (P2 5 sr 91 + E . f is HenstockKurzweil integrable over [a. is the same as the value of the Riemann and Lebesgue integrals of cp. See Exercise 4. 0 Now.b] and recall rni = i n f z i . f .JI cpil < E for i = 1. . and the value of the HenstockKurzweil integral. Therefore.b] and Ja f = JIl f J12 f . HenstockKurzweil integrable. Set y ( z ) = y1 ( z ) n y2 (2).b] + R. Suppose that.30 Let f : I = [a. Let D be a yfine tagged partition of I .. Then. By iteration. then IS (pi. Let cp be a step function defined on [a. f is HenstockKurzweil integrable on I.17. The proof is now completed by an induction argument.b]. Then. f is HenstockKurzweil integrable. Let P = { x ~ J . So. by linearity EL.z m } be a partition of [a. hence. Since the characteristic function of an interval is Riemann and.
we have proved that every continuous function defined on a closed interval is HenstockKurzweil integrable.m. there is a S > 0 so that (f(x). So. It then follows that (p2 (x)5 p1 (x)+& so that By the previous lemma.f (y)J < for all Z. ‘pl 5 f 5 (p2 and cpl and (p2 are HenstockKurzweil integrable. By Theorem 2.4 Unbounded intervals We would like to extend the definition of the HenstockKurzweil integral to unbounded intervals. we need to define a partition of R. Riernann integrable functions are HenstockKurzweil integrable. This ‘extension’ of f to R should have the same integral as the original function defined on I . z .b]. x. this result is not surprising. extend our definition of the HenstockKurzweil integral directly to R. since f is uniformly continuous on [a. Then x [Mi mil 5 & for i = 1..154 Theories of Integration and Define step functions 01 and 02 by m j=2 and (P2 ( t )= ~ 1 X [ Z O . Theorem 4. if we 1.31 Let f : [a. . To extend the definition of the HenstockKurzweil integral to functions on R.27.b] + R be continuous on [a. . Of course. .b] such that I . continuous functions are Riemann integrable and. . it is easy to extend f to all of R by defining f to equal 0 off of I . A partition of R is a finite. Still. Then.. we run into problems immediately since any tagged partition of R will have . Further.b]. given E > 0. ordered set of points in R*.b]. Z l(]t )+ j =2 c m Mjx(zjl.14. by Theorem 4. clearly. we may assume that our function f is defined on R. y E [a. f is HenstockKurzweil integrable over [a. 4. Given a function f defined on an interval I c R.zj]( t ) * Then. . P = (00 = ZO. = 0 0 ) . Suppose we choose a partition P with mesh less than 6.yl < 6.
For the remainder of this chapter. m. and we define f : R* + R by setting U f (m) = f (00) = 0. The Riemann sum I of f with respect to V is defined to be rn i=l If m and m are the tags for any intervals of infinite length... we wall always assume that f is extended to R* b y setting f (x)= 0 f o r x $ I . We define a partition of I to be a finite collection of nonoverlapping closed intervals { I I .then we avoid the problem above.f course. m } such that { I i : i = 1. To handle intervals of infinite length. . . Let V = { ( t i .oo] and [00. The point ti is called the tag associated to the interval Ii.. Such a situation arises if we consider a positive function defined on all of R. Even with the convention that 0 00 = 0.m } be a tagged partition of I . Let P # 0 be a partition of R.b] and ti E I i . implies that f is equal to 0 at 00 and m. . A tagged partition of I is a finite set of ordered pairs V = { ( t i . . we will often choose gauges so that the only tag for an interval containing 00 (00) will be 00 (00). R* = R (00. then If f (z) = 4 + ~ and a1 < 0 < an. i = 1. Let I c R* be a closed interval. if the value of the function at the tag associated with an interval of infinite length is not 0 then the Riemann sum would not be a finite number. .&y. E R are the tags.I i ) : i = 1. ones of infinite length. . then this sum is well defined and finite. . If a l . .Kurzw eil integral 155 a t least one (and generally two) subintervals of infinite length since. Then. We call intervals of the form [a. o. . Let I be a closed subinterval of R* and suppose that f : I + R. ..m } is a partition of [a.. we set [ ( I ) = 00..33 Suppose that f .32 Define f : R + R by f (z) = .. P has two unbounded intervals. if I is an unbounded interval. say 1 1 and In. a] closed intervals containing 00 and 00. Remark 4. we consider f to be defined on the extended real line.. ) open inter0 a vals containing 00 and 00. i ) : i = 1. oo}. a .01 and [oo. and ( a .Hens tock. . If a1 = oo and a. . = oo. . I c R +R. I m } such that I = U z l I i . this expression is not even well defined. Example 4. To get around this problem. this.
This definition does not make sense when t = 00. a gauge at t was defined to be an open interval centered a t t . We denote this by writing D is a yfinetagged partition of I . Proof. . some a . let Do be a yfine tagged division of [a. Since (t . are equivalent. b E R. If b < a . similar. ~ ] The other cases are . Then.. [b 1.fine tagged partition of I . If D = { ( t i . there is a y. one defined in terms of a positive function 6 ( t )and the other in terms of an open interval containing t . y ( t ) is an open interval containing t . an intervalvalued function y defined on I is called a gauge if.18. There is a b E R such that y(00) = (b.fineif Ii c y (ti)for all i. Definition 4. 001 and y (00) = [a. Thus. for all t E I . If b 2 a. y ( t )is an open interval in R* containing t . We show first that for any gauge y .b 1 . . D = Do U ( ( 0 0 . then it satisfies the Definition 4. Theorem 4. we say that D is y. This new definition extends to elements of R* by setting y ( 0 0 ) = ( a .m } is a tagged partition of I and Ii) y is a gauge on I . then 'D = ( ( 0 0 . an intervalvalued function y defined on I is called a gauge if.001)) is a yfine 1 tagged partition of I . the two definitions of a gauge.6 ( t ). Then.37 Let I be a closed subinterval of R* and f : I +R.b]. if a function y satisfies the Definition 4. I ) } is a yfine tagged partition of I .36 Let y be a gauge on a closed interval I = [a.34 Given an interval I = [a.b].6 ( t ). In fact. so we need to revise our definition. b) for Definition 4. It turns out that the important feature of a gauge is that the gauge associates to t an open interval containing t . there exists a yfine tagged partition. we can revise the definition of a gauge.34. . not that the interval is centered a t t.156 T h e o r i e s of Integration For real numbers t . We will prove the result for I = [ a .t 6 ( t ) ) .00]..t + S ( t ) )is an open interval containing t . See Exercise 4.35 Given an interval I c R*. for all t E I .9. + Definition 4.b] c R*. 0 + + We can now define the HenstockKurzweil integral over arbitrary closed subintervals of R*. : i = 1. We call the function f HenstockKurzweil integrable over I if there is an A E R so that for all E > 0 there is a gauge y on I so that for every yfine tagged partition D of [u. ( t .
We show that f is HenstockKurzweil integrable over R and JRf = 0.say. Finally.38 Let E c R be a null set. Thus.@ R. It is easy t o adapt that proof t o show that a function which is 0 except on a countable set has HenstockKurzweil integral 0. That is.1] with an integral of 0. It is also positive. for every finite set {Ij>j”_l of closed intervals with disjoint interiors such that I = U Y l l I j . f is HenstockKurzweil integrable over each I j . Note that if I is an interval of infinite length. Suppose that f : R +R and f = 0 except in E . the HenstockKurzweil integral is additive over disjoint intervals. we proved that the Dirichlet function is HenstockKurzweil integrable over [0.HenstockKurzmeil integral 157 Note that the basic properties of integrals. the notation A = J I f is well defined. so that f 2 0 implies that f 2 0. E that is the HenstockKurzweil integral in linear.co]. Example 4. such as linearity and positivity. using the notation of that proof. In either case.19 and 4.e.e. Fix E > 0. These results generalize Propositions 4. we can replace the gauge in the proof by Earlier. The proof that the value of A is unique is the same as above. f : I + R is HenstockKurzweil integrable over I if. Set . and satisfies a Cauchy condition. in R. since we can use interval gauges. f = 0 a. We now prove a much stronger result. sI The proof of this result is a little easier than before. i. and only if. Suppose that f and g are HenstockKurzweil integrable on I . I = [a. are valid for the HenstockKurzweil integral.. Let I c R be an arbitrary interval.29. as in Theorem 4. namely that any function which is 0 except on a null set (recall that a set E is null if m ( E )= 0) is HenstockKurzweil integrable with integral 0. Thus. we write J I f = Jam f.20 and Theorem 4.27 and follow from the same proofs. For all scalars a.
Let Do = {(ti. that is.Ji) E Dm} are nonoverlapping and U ( t i . Then.158 Theories of Integration Em = { t E R : m . Ji c UjEa. we will use a covering lemma.. there is a countable family { ( t k . This covering lemma will be used to pick a subset of C so that the union of the members of the subset still cover E and have additional useful properties. so each Em is a null set. we see that if f : R +R and f = 0 except on a null set E .IT. and E c UkEa& c I . since the intervals { ( t i . Suppose we have a set E c R and a collection of sets {S}sEc such that E c UsE:cS.. Lemma 4. i ) E D : ti E Em}. i ) E D : ti 4 E } and.. As a consequence of this example. y ( t )= for t E Em.1 < (t)l _< m}. J i ) E D . If + Thus.. so f is HenstockKurzweil integrable over R and JR f = 0. then JI f = 0 for every interval I c R. . for m E N.If t E Em.0). Jk c y ( t k ) . Then. that if JI f = 0 for every interval I c R then f = 0 a.) Suppose that D = { ( t i .1. let m ( t )be < the smallest integer j such that t E IT. if E is a null set. Let y be a gauge o n I . there are countably many open intervals { Ij” : j E a m } such that Em C U j E u m I T and CjEb. Note that the sets {Em}z=i are pairwise disjoint and Em c E since f equals 0 off of E. C (IT) ~/2”m. if E c R is a null set. S (f. . For all m E N.39 Let I c R be a closed and bounded interval and E c I be nonempty. then xE is HenstockKurzweil integrable with J1xE = 0 for any interval I c R. (The choice of 0 for an endpoint is arbitrary. We show next that the converse to this statement is true. In order to prove this result.e. i ) : i = 1. We will show later. J { J i : (ti.J k ) k E a ) such that the intervals in {Jk : k E a } are nonoverlapping and closed subintervals of I . then xE is HenstockKurxweil integrable with JRxE = 0. after discussing Part I1 of the Fundamental Theorem of Calculus.Define a gauge y on R by setting y ( t )= (t .t 1) for t 4 E . if JRxE= 0. In particular. k } is a yfine tagged partition J of R*. then E is measurable with m ( E ) = 0.001 and y (00) = [00. let Dm = J Do) = 0 and. In particular. y ( 0 0 ) = (0. . t k E J k n E .
Let D k be the set of closed subintervals of I obtained by dividing I into 2k equal subintervals. D 1 contains the two intervals obtained by bisecting I into two equal parts. let €2 be the family of intervals J E V2 such that there is a t E E n J with J c y ( t ) and J is not contained in any element of E l . We are now ready to prove Theorem 4. and let Vi be a yfine tagged partition of Ki. in general. and.40 Let E c R. By construction. The sufficiency is proved in Example 4. it follows that x ~ ~ hasintegral 0. The set & = u&&k is a countable collection of nonoverlapping. By Exercise 4. .1. J k ) : k E a} be the countable family given by Lemma 4. D consists of the four 2 intervals obtained by bisecting the two intervals in 2 3 1 . E as a null set zf) and only i f ) x E is HenstockKurzweal integrable and xE = 0. sR Proof. Thus. n] is a null set for all n E N. D = { ( t k . D ) [ < for every yfine tagged partition 2) of I . Thus. assume that xE is HenstockKurzweil integrable with integral 0. one gets a sequence of collections of closed subintervals of I . then J k ( t ) C y(t). . there is an integer K so that for k 2 K .9. Thus. and continue the process. is then either J' and J" are nonoverlapping or one is contained in the other.38. 5 . closed intervals in I . Suppose t E E. J k ) : k E a'} ~ f pi is~ a yfine tagged partition of I .Kurzweil integral 159 Proof. Notice that U r = l D ~ a countable set and if J' E D k and J" E Dl.. then there is a t E E n J such that J c y ( t ) . some of which may be empty. Fix E > 0 and choose a gauge y such that I S ( x E . Then. Let I be a bounded interval containing E . if J k ( t ) E 2>k is the subinterval that contains t . Let K I . Let &1 consist of the elements J E D 1 for which there is a t E E n J with J c y ( t ) . D k is comprised of the 2'" intervals created when the intervals in D k . .1 are bisected.39. 0 we wished to prove.& such that J K C J . Either J K E EK or there is a J E ufG.. { & k } T = l . ~ ~ that E is a bounded set. .it follows that E is a null set. It remains to show that E c U J ~ & J . Let { ( t k . Next. Then. The set I \ U k E g ' J k is a union of a finite ' set of nonoverlapping intervals. Let a C a be a finite subset. t E U J ~ E Jas . i = 1. In other words. we may assume [ ~ . since if we can show that E n [n. 0. To prove the necessity. Since xE 2 . .. if J E &. Then. Kl be the closure of these intervals.Henstock.
Let E > 0. . Since this holds for all E > 0. a k X [ k . k must be a tag. SO that b > M and f ( t m )t( I m ) = 0. . t + m i n ( & .160 Theories of Integration Since this is true for every finite subset of 0. Let V N = {(ti. For t E ( k .00) and 00 czl k=l Since the series is convergent. l ) . 0 In the following example. E V : ti E N}. let y (00) = ( M . Since E c U k E a J k c U k E u I k .m i n ( & . let I k be an open interval containing J k with c ( I k ) = a ( J k )+E2k. for each k E 0 .001. k r ( t ) =( I c . K 2 M . it follows that Finally. Suppose that V = { ( t i . let lajl < E for j 2 M . Then. IcF=j I + . we see that E is a null set. For k E N. there is a B > 0 so that l a k l 5 B for all k E N. Further.. k + l ) . Define a gauge y as follows. Pick a natural number M so that a k < E and l). l e t y ( t ) = ( t . i ) : i = 1.41 Suppose that a k is a convergent sequence and set f (x) = cF=. { I k } k E a is a countable collection of open intervals containing E .m} is a yfine tagged partition of [l. . We claim that f is HenstockKurzweil integrable over [I.I i ) E DN and ti = k } c y ( k ) . we relate HenstockKurzweil integrals to infinite series. we may assume that t m = 00 and Im = [b. Note that for k E N and k 5 b. f o r t = k .001. Without loss of generality. l ) ) . k + l ) (z). Example 4. Ii) Thus.001. I and.U {Ii : (ti. Let K be the largest integer less than or equal to b.
See Exercise 4.00). k=l It follows that f is HenstockKurzweil integrable over [l. Therefore. There is no need t o view this as an improper integral. and + Thus. I i ) E D k I iis a ) subinterval of ( k . Using this example. & ) E D : ti E ( k .& . if the function f (2)= CFx1 k X [ k . by the definition of ~ ( jfor j E W. Note that we have evaluated the integral of a function defined on an interval of infinite length directly from the definition of the HenstockKurzweil integral.HenstockKurzweil integral 161 Set 2)k = { ( t i . Moreover.&. If . We say a function f is conditionally integrable if f is HenstockKurzweil integrable but I is not HenstockKurzweil integrable. then the series CEl a k converges.k 1) with length & 2 1 .23. I 0 0 1 00 0 0 1 I k=l I k=l K1 k=l I <  I < c& + 00 E + ~ E +E =6 ~ . k + l ) (2)is HenstockKurzweil a integrable. k + 1)). Note first that For 1 5 k < K . We will discuss this issue in the following section. This example highlights two of the important properties of the HenstockKurzweil integral. U ( t i .
6) holds due to cancellation in the expression on the left hand side. Equation (4. partition of I . f is a conditionally integrable function. 00). there is a gauge y so that if V = { ( t i I i ) : i = 1.7) . Thus. for which integrability implies absolute integrability. more generally. one can simultaneously control the estimate for a single interval (ti)e ( I i ) . we can rewrite Equation (4. one might expect that However. in addition to controlling the difference of sums. one is led to consider if. . if 27' c 27.162 Theories of In. This is in contrast to the Riemann and Lebesgue integrals.k+l) is a condi(x) 4.5 Henstock's Lemma If f is HenstockKurzweil integrable over an interval I .JIi < E or. in general. an estimate of part of the sum. .42 The function f (z) = CF=l tionally integrable function on [l.tegration one can now easily construct conditionally integrable functions. . it is not at all clear that Equation (4. If fl that is.m } is a yfine tagged . given any E > 0.41 while = Czllakl (z) is not HenstockKurzweil integrable by Exercise 4. If Cgl a k is a conditionally convergent series.22.6) as Thus. I I)f . then f (2) = EL1k X ( k .. then Since JI f = Czl JIi f . ( (1)k Example 4. x(k. k + l ) (z) is a a HenstockKurzweil integrable function by Example 4. Since the cancellation from one interval may help the estimate for another interval.
. i ) : i = 1 . We will now prove Henstock’s Lemma. k } such that {Ji}i=l a subpartition of I and ti E Ii.J 1 ) . let y be a gauge such that if V is a yfine tagged partition of I . . . . . Let I c R be an interval.. Let K1. V is a yfine tagged partition of I . Note that a yfine tagged partition of I is also a yfine tagged subpartition of I . The set I \ J is a finite union of disjoint intervals. Ic. . Since f is HenstockKurzweil integrable over each K j . Proof. . . . . .HenstockKurzweil integral 163 will hold. Lemma 4. (xk. Km be the closure i of these intervals. . For E > 0.7) (with < replaced by 5 ) follows from Equation (4. Since g.J k ) } is y. . Set D = D’UV1 .. ..fine tagged subpartition of I . there is a yfine tagged partition Vj of Kj such that j One can find such a partition by choosing a gauge yj for the interval K and the margin of error and then choosing a partition which is y n yjfine. J is We say that a tagged subpartition is yfine if Ii c y (ti)for all i. which is a valuable tool for deriving results about the HenstockKurzweil integral. even if V‘ contains a single pair ( t . . we will show that Equation (4. UfZl Let E > 0 and y a gauge satisfying the hypothesis. Fix q > 0. In this section.I ) .6). A subpartition of I is a finite set k of nonoverlapping closed intervals {Ji}i=l such that Ji c I for i = 1.43 (Henstock’s Lemma) Let f : I C R + R be HenstockKurzweil integrable over I . A tagged subpartition of I is a finite set of ordered pairs S = k { ( t i . then I JI I Suppose Then 2)’ = {(XI. Then.UV. We will apply Henstock’s Lemma t o the study of improper integrals and convergence theorems.
0 Suppose that I is a subinterval of R and a E I .and V + are yfine tagged subpartitions of I . Theorem 4. set and D." f for all x E I . so they satisfy the previous estimate.44 I f f : I F as continuous o n I . m Since 7 > 0 was arbitrary. so that f is integrable over every subinterval of I .= D'\V+. Suppose that f : I + R is HenstockKurzweil integrable.164 Theories of Integration we have ' I <E+m=e++. + R is HenstockKurzweil integrable ouer I .finite integral F of f by F (x)= s. This completes the proof of the theorem. Thus. then . it follows that To prove the other estimate. Note that both V . Define the inde.
25 that an unbounded function can be HenstockKurzweil integrable.b ] .s. )..x. Applying { +} Henstock’s Lemma to the yfine tagged subpartition { (x.” f = 0.45 Let f : I = [a. 0 We have seen above in Example 4.JI D < E for every yfine tagged partition D of I . Fix a E I and x E I . If y (x) = fl ( a . ’ Proof. . then If I is HenstockKurzweil integrable with 1 If I = 0. m P I which implies that If I is HenstockKurzweil integrable with fI 1 f 1 = 0.a. Choose a gauge y so that IS (f. Let E > 0.HenstockKu. p ) . . By hypothesis.. Let D = { ( t i . Henstock’s Lemma implies that the indefinite integral of a HenstockKurzweil integrable function is continuous. F is continuous on I. Let J be the subinterval of I with endpoints x and y. and in Example 4. Using Henstock’s lemma.41 that a function defined on an unbounded interval can be HenstockKurzweil integrable. so that f J f = 0 for every interval J c I . we show that there are no improper integrals for the .I i ) : i = 1. Corollary 4. F is continuous at x.xl < 6.x . We apply the second inequality in Henstock’s Lemma in the proof of the following corollary. and suppose that y E I and ly . ) } shows that J This implies that Thus. 0 Thus. If J. By Henstock’s Lemma. if a 5 c < d 5 b. Since z E I was arbitrary. set 6 = min p . Let E > 0 and choose a gauge y such that d d for every yfine tagged partition D .m } be a yfine tagged partition. then f c f = Ja f .“f = 0 for every c E [a.b]+ R be HenstockKurzweil integrable over I .rzweil integral 165 Proof.
+ f exists.a ) < c/3. Fix s E ( a . we may assume that y.co].c ) and let E be a ?. and only i f . then For each c E ( a .fine tagged partition of [c.46 Let f : [a. = [a. Next. lim.. by replacing y.b].b]. Suppose first that f is HenstockKurzweil integrable over [a.b]. D is a yfine tagged partition of [a. by y." .b] then Without loss of generality. . n y if necessary.b ) . 3 3 3 Thus. and E E € <++=€. Let E > 0 and choose a gauge y so that if D is a yfine tagged partition of [a.s])} U 1.166 Theories of Integration HenstockKurzweil integral. I n either case. Choose c E y ( a ) such that If ()I ( c . so that if D is a ylfine cg] ' tagged partition of [ c l . c k > c k + l and c k + a. Set 'D = { ( a . there is a gauge yc defined on [c. then s. Define a gauge y1 on [q. Choose { C ~ } T C ~ b] so that co = b.. c y.b]4 R be HenstockKurzweil integrable over [ c . suppose the limit exists.b]. Theorem 4.+ f = s. Then.b] so that if E is a 7." f . f is HenstockKurzweil integrable over [a.b] f o r every a < c < b..[a. We begin by considering a function defined on a bounded interval. scb Proof.b] <f. lim.fine tagged partition of [s. Then..
ck11. instead of a t a. c k .+ f . By the definition of y."" f .b] for all a < c < b and has an improper Riemann integral over [a.c] and [c. then [ c k .b] separately. only finitely 3 many Dk # 0 and D nDj = 0 for i # j . . and 5 C (c1.rzweil integral 167 For k > 1. Since 2 has a finite number of elements.b]and f = A. for a singularity at an interior point c E ( a . Then. b Ja Thus. one may consider the integrals over [a.b]. and only if. b ) . t = a ..b] + R is Riemann integrable over [c. f is HenstockKurzweil integrable over [a.. Let (x.JJk f. 0 This proof can be modified to handle a singularity a t b. d ] ) E D. b so that x = a. Since S (f. in which both sums have finitely many nonzero terms..ck21. Define a gauge y on [a. for k 2 1. Choose K so that Is. f is HenstockKurzweil integrable over [c.D k ) and f = CF=.2 ) . [a. Dk is ykfine on J k .b].+ JCb f exists." < E: for a < s 5 C K Let D be a yfine tagged partition of [a. and 2)k be the subset of 2)with tags in ( c k .HenstockKu.A 1 and If (a)l ( C K .a ) CE.a E y ( t ) if. Let J k be the union of subintervals i in Dk. S (f.b] by s. Then. Thus. f is HenstockKurzweil integrable over [a. define a gauge yk on partition of [ck.b].D ) = f ( a ) ( d . ck21 so that if D is a ykfine tagged Set A = lim.b] and lim.. Suppose that f : [a.cg] and J k C ( c k . + sd <€+ k=l c$ +E = 3€. Further. By 1 Henstock's Lemma.a ) < E .
47 Let p E Iw and define f : [0. Let E > 0 and choose a gauge y so that if V is a yfine tagged partition of I .+ € = 2 2 E7 since 1 f (0l 0) l ( [ c . P+l Suppose.. Fix c > max { T . treating the difficulty at 00 as the one at a was handled above.01) Then.001) = 0 by convention. ~then ] ] and such that yc ( z ) c y (x) for all x E [a. Thus. For each c > max { T . Then.a } . there is a gauge yc defined on [ a . Set 2) = I U ( ( 0 0 . 1 b f = a sIf . s. that f is defined on an unbounded interval I = [ a . V is a yfine tagged partition of I . and 0). and only if.001.0 1 0. then Suppose y (00) = ( T .b] for every a < b < 00. [c.00] + R be HenstockKurzweil integrable over [a. for 0 < t 5 1 and f (0) = 0. demonstrating that there are no CauchyRiemann integrals in the HenstockKurzweil theory. . Suppose first that f is HenstockKurzweil integrable over I .1] + R by f ( t ) = t P .c ] .003 iJ and only if. we see that f is Henstock1 1 Kurzweil integrable over [0.168 Theories of Integration Example 4. limb.fine tagged partition of [ a . Proof.fine tagged partition of [ a .a } and let I be a 7. f is HenstockKurzweil integrable over [ a . We show that integrals over I exist in the HenstockKurzweil sense as proper integrals.48 Let f : I = [a. next.c]. limb.44. € < .” f exists. Theorem 4. I n either case.. ~so that if & is a 7. By Example 2. so p > 1. I] with integral t p d t = if. The proof is similar to the previous one.
0 .c 1 ] so that for every yofine tagged partition 2) of [ C O .l . c ~ ) and J k C ( C k . ck+1] so that if D is a ykfine tagged partition of [ C k . 2)k is ?. Choose K so that Define a gauge y on I by s. For k 2 0. If Ii = [a. By Henstock’s Lemma. for k 2 0. + 00. then ti = oo and Q > C K . As above. f is HenstockKurzweil integrable over I and fIf = A. Define a gauge yo on [co. Then. choose a gauge y k on [ c k . c k + l ] . For k 2 1. let v k be the subset of 2) with tags in [ c k . only finitely many D # 0 and k 2)i f l 2)j = 8 for z # j . and JO C [ c o . Since Q > C K . c k + l ) . c k + l ) .” I f ab f A < ~ / 2 b 2 for I CK.00) so that co = a . Choose { c k } E l C [a.fine on J k .l .HenstockKurzw eil integral 169 ck < c k + l and c k Next.it follows that k=O I Thus. then Set A = limb. c l ] .m f . is the unbounded 003 interval of V . suppose the limit exists. Let V be a yfine tagged partition of I .l . Let J k be the union of subintervals in 2 ) k .
48. Then.001. We now give another example of a function that has a conditionally convergent integral.." f is increasing f.b] c R* t R. See Exercise 4. satisfies a Cauchy condition near b. a] U [a.G ( X ) ' Thus." I W Y ) . Suppose that f is absolutely integrable over [a. For (2). The first result includes a comparison test. and only if. Applying the same argument to H ( x ) = If I s.b] + R such that If (t)I 5 g ( t ). By Example 2.b] if.001 with integral S tpdt = P 11 if.and the result follows on [a. note that for a<x<y<b. for + t 2 1. sup { [. p > 1. sup f : a 5 c < b } < 00. We now use these theorems to obtain several useful results for guaranteeing absolute integrability. (1) Suppose f is nonnegative. define F as above and set G ( x ) = g." f : a c < b } = lim.42.50 It was shown in Example 2. Corollary 4.p .but that If 1 is not CauchyRiemann integrable there. too. f is HenstockKurzweil integrable over [a.49 that f ( 2 )= vergent CauchyRiemann integral over [I. By Theorem 4. Example 4.28. Thus.47." Note that b may be finite or infinite. by either Theorem 4. and only if.170 T h e o r i e s of Integration An analogous result holds for intervals of the form [m." .51 Let f : [a. G satisfies a Cauchy condition near b. we saw that f ( x ) = Cp=l ( x ) is a conditionally integrable function on [l.c] f o r every a 5 c < b.b]. s.46 or 4.oo] R by f ( t ) = t . F is Cauchy near b. and f is HenstockKurzweil integrable.F ( 4 l = 1lYf1 5 LY If1 L slg = G ( Y ). 001 follows by writing [00. Since g is HenstockKurzweil integrable.46 or 4.48. f is HenstockKurzweil integrable and 1 f I is not. has a conExample 4.b from either Theorem 4. The value of the integral so obtained does not depend on the choice of a. oo). To prove (1))note that the function F ( x ) = s. 001 = [00. {s. We claim that F . so f has a conditionally convergent integral. b]." s. we see that f is HenstockKurzweil integrable over [l. To see this.49 Let p E R and define f : [l. (2) If there is a HenstockKurzweil integrable function g : [a. A version of this result for [00. Proof. 001. then f is absolutely integrable over I .48.for all t E I . T (1)k Following Example 4.
every bounded. for example. F g + Gf is HenstockKurzweil integrable by Theorem 4. Proposition 4. the series converges..94. JY c:=. Letting n + 00 in rn] 0 (4. F is continuous and g is HenstockKurzweil integrable.16. However. the series C:=. + 2 A function cp is a multiplier for the HenstockKurzweil integral if. so that f is absolutely 0 As a consequence of the corollary. T h e integral S. Summing in i yields f + + + 1. I n either case.H e n s tock. f ( k ) I f f (1).. we see that continuous functions need not be multipliers for the HenstockKurzweil integral. f 5 f (2). If\ is HenstockKurzweil integrable. See Theorem 4. f (i 1) _< f (x)5 f (i) for i _< x 5 i which implies that f (i 1) 5 ' .8) shows that sp" f 5 CF=.26. This implies that neither F g nor f G is HenstockKurzweil integrable over [0." f exists i f . for the HenstockKurzweil integral.Kurzw eil integral 171 shows that integrable. Theorem 12. and only if. we derive the integral test for convergence of series. Example 4. Since. and only if.91. is not HenstockKurzweil integrable over [0.1]. fL f (lc) 5 J. Surprisingly.53 Define F . For if cp is measurable and bounded by B .f (z) G (z) = for z # 0. decreasing and HenstockKurzweil integrable over [l. measurable function is a multiplier." f + f (1).rn] + R be positive. F ( x )g (z) . i=l i= 1 By the previous corollary.52 Let f : [l. 11. then for any Lebesgue integrable function f. so cp f is Lebesgue integrable by Proposition 3. f (lc) converges. even on intervals of finite length. (See [Lee. For the Lebesgue integral. pf is measurable and pf is bounded by the Lebesgue integrable function B If I. continuous functions need not be multipliers. it is equal almost everywhere to a function of bounded variation.) .b] f o r all 1 < b < 00. Since f is decreasing. which we define in the next section. it now follows that f is HenstockKurzweil integrable over [l. G : [0. if.13 + R by F (0) = G ( 0 ) = 0 and F (z) = x2 sin (z*) and G (z) = x2 cos ( x W 4 )for 0 < 2 5 1 and let f = F' and g = G'. JY + A function cp is called a multiplier if the product cpf is integrable for every integrable function f . and only if. Proof. Since ( F G ) ' = F g f G.
a continuous function need not have bounded variation. . we do not know whether or not is HenstockKurzweil integrable. A function can have a jump discontinuity and still have bounded variation.172 Theories of Integration 4. which follows immediately from the definition.. In A constant function has 0 variation.6.54 Let cp : [a. define the variation of cp with respect to P by m i=l and the variation of cp over [a. Somewhat surprisingly. A function with bounded variation has finite oscillation. .6 A b s o l u t e integrability Let f be HenstockKurzweil integrable over I . the function f defined on [0. Since f need not be absolutely integrable. we will use the concept of bounded variation. 00.b ] ) < this case. .b].1 Bounded variation The variation of a function is a measure of its oscillation. [a.b] if V a r ( c p . Given a partition P = (20.b ] } .1] + Iw defined by . Definition 4. E x a m p l e 4.55 The function cp : [0. For example. We now turn our attention to characterizing when a HenstockKurzweil integrable function is absolutely integrable.2] by f (x) = 0 for 0 5 x < 1 and f(x) = 1 for 1 5 x 5 2 has a variation of 1. For this characterization. xm} of [a. we write cp E BY ( [ a b]). . We say that cp has bounded variation over [a. If[ 4.b]by : P is a partition of [a.b] + R. equal t o the jump at x = 1.
.b]. P').. . (m++)n 1 if m is even .cp (4 {a h 5 Icp (4 . xi1. ~. [0. Then.P (4+ Icp ( b ) . . Then.  (m++)n ifmisodd Since ' . Proof. I]. P ) I(cp. z ..HenstockKurzweil integral 173 is continuous on [0.57 Let cp : [a. there is an 2 such that P = { z o .b] I and P is a refinement of P .. We will use this result to prove that variation is additive. all the terms in the sum for . If P and P are partitions of [a. 0 0 00 and cp does not We next develop some of the basic properties of functions with bounded variation. . .~i.b]. Set x = . 1 . Proposition 4.. . xn} and P = . . c .xn}. I]. then v (cp. If a < c < b. (m++ . For x E [a.i?. We first show that a function with bounded variation is bounded and that the variation of a function is additive over disjoint intervals.b } ) = lcp ( b ) .56 If cp E BV ( [ a b]) then cp is bounded on [a.b] + R. it follows that Var (9. I v Proof..1]) = nm have bounded variation on [0.consider the partition P = { a .Then. that is. I { Z O ..cp (c>l = (9.b]. which implies Thus. cp is bounded on [a. Proposition 4. b } ) This inequality is the basic point in the proof that the variation of a function increases as one passes from a partition to one of its refinements.diverges.b]. by the triangle inequality (9. Z I.b } of [a. Suppose first that PI has one more element than P.
As above.58 Let cp : [a. moreover.b]). Thus.b]. Then.b] and set P‘ = P U { c } . 0 + We can now show that the variation of a function is additive over disjoint intervals. V a r ( c p . It follows that for any partition P of [a.c] and P2 = {z E P‘ : x 2 c} is a partition of [c. P I ) + v (9.b] + R and suppose that a < c < b. Proof.cp (2)l Icp ( 2 ) . P’) = v (cp.cp (zil)l which is bounded by Icp (xi). ] On the other hand.c]yields Then.c ] ) V a r ( c p . Let P be a partition of [a. One can argue . if PI is a partition of [ a . Proposition 4. If a 5 z then 0 < x < y 5 b. v (cp. P’). by the previous proposition. Another easy consequence of the definition is that . p similarly for a decreasing function. + Taking the supremum over all partitions PI of [a. ~and 732 is a partition of [c.b]. [a.b]) = I ( b ) .b]shows that which completes the proof.b].then P = PI u 772 is a partition of [a.b].P) = Ip(b) . P’) except for Icp (xi). P ) are the same as those for v ( c p . v (cp. taking the supremum over all partitions P2 of [c.cp ( i l l Thus.174 T h e o r i e s o f Integration v ( c p .b]. It follows that Var (cp. PI = {z E P‘ : z 5 c} is a partition of [a. [a.cp (a)l.cp(a)l and cp E BV ( [ a b ] ) .b ] ) 5 V a r (p.v (cp. Then. The proof now follows by an induction argument. Suppose that cp is an increasing function on [a.Pz).[a. [c. so that every monotone function on a bounded interval has bounded variation there. and v (cp. P ) 5 z).
b ] ) . So. Since Proposition 4.6. Let V = V a r ( F . there are in. 1 f 1 is HenstockKurzweil integrable ouer I if.b] + R be HenstockKurzweil integrable over I . Proof.b].From [a. s . Theorem 4. If p and q are increasing functions on [a.58 implies that V a r (cp. Theorem 4.b]. its indefinite integral has bounded variation.9. Proof. then Thus. 4.F (z)l = f Suppose first that If\ is HenstockKurzweil in 1: 1. Define p by p (x)= . x]). I n either case.59 A function cp E t3V ( [ a b ] ) i f . YI).2 0 Absolute integrability and indefinite integrals We are now ready to prove that a HenstockKurzweil integrable function is absolutely integrable if. Proposition 4.V a r (cp. Therefore.a ] )= 0 by definition. [ a . q is increasing and the proof is complete.” f. and only if. by the observations above. and q = p . [a.q E t3V ( [ a b]). A surprising fact about functions of bounded variation is that all such functions can be written as the difference of increasing functions. Then.60 Let f : I = [a.where V a r (cp.HenstockKurzweil integral 175 which implies that linear combinations of functions of bounded variation have bounded variation. p . creasing functions p and q so that cp = p . suppose that cp E BV [a.38. V a r (9. See Exercise 4. Note that for a 5 x < y 5 b. ( F (y) . [a.y]) . If a 5 x < y 5 b.58. and only iJ the indefinite integral o f f has bounded variation over I . and only i f . p is increasing. Recall that we define the indefinite integral of f by F (x)= s.q. XI) = V a r (9. [a.
. I:) and (xj. V 5 < 00. .yi]. then x E P. Thus. K } in which all such terms (xj. set z1 = cm and zm+1 = 00. Define a gauge y on I by: fl = { if x E (xi1.P)5 ( F . . then xi is the tag for Ii. k } be a yfine tagged partition of I and. it then follows that z = xj . the interior of Ii. ..b ] ) . D'). . . .yi].E < w ( F . E .. Choose a partition v ." If[ E > 0. .JI < E. k. suppose that F E BU ( [ a b ] ) and let . X m } of I . If xj E I f .. by s If s If + If v E < 21 ( F . assume that max Ii1 = min Ii for i = 1.P)5 v. Since f is HenstockKurzweil integrable over I . D Ii) are replaced by the two terms (zj.176 Theories of Integration tegrable. one gets a new tagged partition D'= { (x:. since (zj)l e ( ~ i = (xj)le (I:) ) (zj)l e (I:) for all xj E P.xi) ~ ( zn (zi1. if (x. Proposition 4.. y (z) is an open interval that does not contain any elements of P. I.I:) : i = 1. I:). Let D = {(xi. without loss of generality. P = ( 2 0 . I i ) : i = 1. and a refinement P = P U Q of P.: . (If1 7 D )= (If1. Next.. By the definition of y for elements of P. Let Q = {yo.. We replace Ii by the pair of intervals I t = [yil. For any partition P = ( 2 0 .J ) E D and there is an xj E P such that xj E J c y ( x ) . . Since P' is a refinement of P.yk} be the partition defined by D so that Ii = [yi1. . . we can choose a gauge 7 on I so that if D is a ?fine tagged partition of I then IS (f.xj] and I: = [zj. so F E BU ( [ a .P') = 2) c// K i=l 5 v. Note that D' is yfine since I/ c Ii c y(xj). x m } of I such that s. . . For convenience.. Thus.57. . Repeating this for all the terms in P as necessary. . .xi) ) zi+l) if x = xi 7 (z) n Note that for z $ P.P) . Fi' nally..
42 and 4.b]. Let P = {Q. Therefore. . .b] + R be HenstockKurzweil integrable over I and suppose that 1 f (t)l 5 g ( t ) f o r all t E I . Corollary 4. 0 Since BV ( [ a b ] ) is a linear space.62 Let f . . g : I = [a. Then.43. If I is HenstockKurzweil integrable and JI If I = V . g : I = [a.HenstockKurzweil integral 177 Since D' is a yfine tagged partition of I . the following corollary is immediate. f is absolutely integrable over I and Proof. xcm}be a partition of I . . so by Theorem 4. Corollary 4. Then Thus.b] + R are absolutely integrable over I . then f g is absolutely integrable over I . .60 If 1 is integrable over I and Extensions of the three results in this section to functions f : R + R are given in Exercises 4. it follows from Henstock's Lemma that 55 i=l Thus.. + As a consequence of Theorem 4. we obtain the following comparison result for integrals.60. the indefinite integral F of f has bounded variation over [a.61 If f .
To prove (l). and f . f+ = a + + + If we only assume that f and g are HenstockKurzweil integrable.1. (2) If f and g are absolutely integrable over I .1] whose we derivative f’ is not Lebesgue integrable. exhibited a function f on [0. We now study the lattice properties of HenstockKurzweil integrable functions. d m .The result now follows from the linearity of the 2 2 integral. for example. If1 = f+ + f.f . the proof is complete.3 Lattice Properties We have seen that the sets of Riemann and Lebesgue integrable functions satisfy lattice properties so that.178 Theories of Integration The examples above of HenstockKurzweil integrable functions that are not absolutely integrable involved functions defined on infinite intervals. the maximum and minimum of Riemann integrable functions are Riemann integrable. Thus. we need an additional assumption in order to guarantee that the maximum and the minimum of HenstockKurzweil integrable functions are HenstockKurzweil integrable.recall that f = f + . if f’ is defined as in Example 4. We conclude this section with an example of such a function on [0. f + and f .1]. By the linearity of the integral and the fact that the sum of absolutely integrable functions is absolutely integrable.1]) so that [0111. For example. 0 lfl + f If’ ’. p k ] are pairwise disjoint.63 In Example 4. we observe that f V g = [f g If . For (2).= .are HenstockKurzweil integrable over I .911. I f’l is not HenstockKurzweil integrable over 4.gl] and f A g = [f g . Proposition 4. Since the intervals [ a k .64 Suppose that f .6. The key estimate in that proof is J b k f‘ = I/%. then . f @ BV ([0. then f V g and f A g are HenstockKurzweil integrable over I . and only if.If . g : I + R.. Proof. where bk = 1/and a k = By the F’undaak mental Theorem of Calculus. Example 4.1. (1) The function f is absolutely integrable over I if. f’ is HenstockKurzweil integrable.
f ) V ( h .b] + R be continuous o n ( a . is not HenstockKurzweil integrable while both f ' and 0 are HenstockKurzweil integrable.b]. By the previous proposition.66 (Dedekind's Test) Let f . b].+ f g as HenstockKurzweil integrable over [a. Fg' is absolutely integrable over [c.f and h .f 1 2 =hs[f+gIf 1 +g1] 41 =hfAg. Since F is bounded. there is a B > 0 so that IFg'I 5 B )g'1. is bounded g (x)= 0. Since g' is absolutely integrable. even on an bounded interval. (Fg)' = f g Fg' on the interval [c. then f V g and f A g are HenstockKurzweil integrable over I . in contrast t o the Riemann and Lebesgue integrals." f f o r a < x 5 b.b]. it follows that f A g is HenstockKurzweil integrable.b] for all + . Proposition 4. For a < c 5 b. then f f Proof.HenstockKurzweil integral 179 (f')'. (1) Suppose that f . Proof. h : I + R are HenstockKurzweil V g and f A g are HenstockKurzweil integrable over I . (h f) = [2h. the maximum of f' and 0. defined b y F (x)= s. We saw in Example 4. Proposition 4. I f 5 h and g 5 h. Suppose that F . ( h .65 integrable over I .f . by Corollary 4. Suppose the conditions of (1)hold. (2) If h 5 f and h 5 g . We conclude this section with conditions that guarantee the integrability of the product of two functions. they are absolutely integrable. g . Fg' is HenstockKurzweil integrable over [c.53 that the product of a continuous function and a HenstockKurzweil integrable function need not be HenstockKurzweil integrable. b]. Since. The remaining proofs 0 are similar. so that Fg' = (Fg)'+f g and. g' is absolutely integrable over [a. g : [a. Then. Since h . by the Fundamental Theorem of Calculus and the fact that f g is continuous.g are nonnegative and HenstockKurzweil integrable.b].. o n ( a .62.b]. and lim..g + l .g ) is HenstockKurzweil integrable.
the simplest condition that allows the interchange of limit and integral on bounded intervals is uniform convergence. c4a+ by Theorem 4. and 4. 00) + defined by f k (z) = x ( k . Jo fk = 1 while So f = 0.+ g (2)= 0.(Fg)' b]. so that f g is HenstockKurzweil integrable over [a.35 for additional examples of integrable products. by the Fundamental Theorem of Calculus. 0 See Exercises 4. Since fg = Fg' . A similar problem arises when one considers integrals over unbounded intervals.180 Theories of Integration a < c < b.9) does not hold. 1 + IR by f k (2)= k X ( o ..31. Example 467 Define f k : [O.9) does not hold. from above. by Corollary 4. All the functions are HenstockKurzweil integrable. Then. 4. so equation (4. Since F is bounded and lim. but equation (4. 4. 1 1 Thus.7 Convergence theorems The Lebesgue integral is noted for the powerful convergence theorems it satisfies.b]. Thus. i / k ) (Z).51.33. Example 4. As we saw in the previous chapter.1]. . some restrictions are required for the equation to hold. Like the Riemann integral. 3 {fk}:l converges pointwise to the function f which is identically 0 on [0. k + l ) (2) converge pointwise to the 0 function.68 The functions f k : [O. but each f k has HenstockKurzweil integral equal to 1.. We now consider their analogs for the HenstockKurzweil integral. Fg' is (absolutely) integrable over [a.46.
k) (Z). Then. fk ( t )5 f ( t )for all t E I . . Since the sequence { f k (t)}FZ is monotonic for all t E I . Since f k 5 JI f so that SUpk J” f k < 00. it follows that { J I fk}r=l is monotonic and converges to A = supk JI f k .V) JR < for any y. cp is HenstockKurzweil integrable over R and JR cp = Let yp be a gauge such that IS (cp. Define cp : R + R by p ( t ) = Cl 2kX{t:kl<ltl<k).70 (Monotone Convergence Theorem) Let f k . by positivity. f is HenstockKurzweil integrable over I if. By the pointwise convergence of f k to f . Each f k is HenstockKurzweil integrable and JR f k = 1. and only i f ) supk f k < 00. Theorem 4. { fk}F=1 converges uniformly to the function f which is identically 0 on R so that equation (4. there is a gauge yk on I such that I s ( f k . sI First. Now. for each t E I . .fine tagged partition D of R.HenstockKurzweil integral 181 See Exercise 4.fine. suppose that SUpk JI f k < 00. Fix E: > 0 and choose a K E N such that Proof. which is finite by assumption. . f k ( t i ) (ti)e (&)D) = . Example 4. Then. JI (4. f : I C R* + R be HenstockKurzweil integrable over I and suppose that (fk}T=1 increases monotonically to f o n I . V ) . assume that f is HenstockKurzweil integrable. such a result does not hold in full generality for the HenstockKurzweil integral over unbounded intervals. : i = 1 . 0 5 S (cp.10) For each k . V )5 JR cp+ = 1 whenever D is y.41. = : Repeating the proof of Example 4. In either case. Further.9) does not hold. choose a k ( t )E N such that k ( t )2 K and SI 6 + 3.m } be a yfine tagged partition of I and consider Ii) the difference IS (f. .f k l < for every ykfine tagged partition V of I .A [ . 3 V Define a gauge y on I by setting y ( t )= yk(t) ( t ) f l y . However. The first convergence result will be an analog of the Monotone Convergence Theorem for the HenstockKurzweil integral. . Adding and subtracting CZ. ( t )for all t E I .45. Let { ( t i .69 Define f k : R + R by f k (Z) = i&(k.
. I . . Then.which implies 5 Summing over i.eories of Integration we see that I m m I By (4. f~ 5 fs. . by (4.f k ( t i ) (ti) I e (12) < i= 1 c m €9(ti)e (12)= E (cp.182 Th. k ( t m ) } K . set S = max { k ( t l ). D> L 6 .11) and the definition of p. by monotonicity and the definitions of k ( t )and S. 2 in which we have grouped together all terms corresponding to fk for a fixed k .10) we see . fk(ti) Finally. I L cIf m (ti) . II L k=K c2" S 2€ < 2€. so that Henstock's Lemma implies Summing over k . S i=l To estimate 11. . Note that the set { ( t i I i ) : k (ti)= k } is a ykfine tagged subpartition of .
Thus.then it holds almost everywhere in I . and f. A = SUpk f k = limk+m f k . By the Monotone Convergence Theorem (for a decreasing sequence of functions). since I = UF=l ( I n [n. { is a bounded interval.+m fk (x)= 00). By replacing f k by fk . we needed t o assume that the limit function was finite on I . we get an analogous version of the Monotone Convergence Theorem for a decreasing sequence of functions. Then. we have IS (f. under the assumption that infk JI f k > 00. sup. we may assume that each f k is nonnegative.HenstockKu. Set M = supk f k and let E = {x E I : limk. sI Proof. Let f: = 1A (+) f k and define hi : I +R by sr increases to hi pointwise as k + 00. Further.A < 4 ~ . and the proof is complete. In the proof of the Monotone Convergence Theorem above. f is HenstockKurzweil integrable with integral A . Combining these estimates.71 Let f k : I C R* + R be HenstockKurzweil integrable over I and suppose that { f k ( x ) } ~increases monotonically f o r each x E I and . for any yfine tagged partition D of I . sI sI {sr Suppose that { fk}r=1 is a sequence of HenstockKurzweil integrable functions that decreases monotonically for each x E I . as a consequence of the monotonicity of the sequence of functions and the condition SUPk f k < 00.}r==. Applying the theorem above to {.n])and if the conclusion holds on I n [n. hi is HenstockKurzweil integrable and . f.f k } r = l . Since I For each fixed i. 1 is HenstockKurzweil integrable over I .f1.n]. In fact. HenstockKurzweil integrable. f k < 00.is a monotonic se0 quence. the limit is finite almost everywhere. sI Lemma 4. since fk}rl .rzweil integral 183 so that 111 5 E. limk.m f k (x) exists and is finate f o r almost every x EI. . we may assume that I is a bounded interval. D) 1 Since E was arbitrary. since the minimum of absolutely integrable functions is. Then.
By the Monotone Convergence Theorem. increases pointwise to f on I . Suppose each f k as HenstockKurzweil integrable over I and SUPk f k < 00.finite otherwise is HenstockKurzweil integrable over I with Proof. so limk+m k exists and is finite f almost everywhere in I .40.m f k (x)= m}. we can improve the statement of the Monotone Convergence Theorem by removing the assumption that the pointwise limit is finite everywhere.+m (x) is .38 Define by Kurzweil integrable and Since is Henstock Further. E is a null set. by Theorem 4. .72 (Monotone Convergence Theorem) Let f k : I c R* + and suppose that { f k (x)}T=. f is HenstockKurzweil integrable and The Monotone Convergence Theorem is equivalent to the following result about infinite series of nonnegative functions. 0 Using this lemma.184 Theories o f Integration Thus. increases monotonically f o r each x E I . by Example 4. fk de. Since E is a null set. Then.finite for almost every x E I and the function f .fined by sI limk+mk f (2) zf the limit is . {gk ( x ) } ~ . limk. the function f is defined almost everywhere in I . By the previous lemma. Corollary 4. Let E = {x E I : limk.
73 Let f k : I C R* + R be nonnegative and HenstockKurzweil integrable over I for each k and de*fine f b y cr=l fk (x)if the series converges otherwise Then. czl Finally. the proof will be based on Fatou's Lemma. Let sm = f k . k=l k=l 0 Our next goal is to prove a version of the Dominated Convergence Theorem for the HenstockKurzweil integral. JI f k < 00. forms an increasing sequence for each x E I and cy=l By Lemma 4. Then. suppose that f is HenstockKurzweil integrable and E = {x E I : cFZ1 (x)= 0 0 } has measure 0. f k = hm400 S m is finite almost everywhere and. by the Monotone Convergence Theorem (Corollary 472))f is HenstockKurzweil integrable over I . and only if.71. and suppose that CY 5 f k in I . czl sI Proof. In either case. Suppose first that cF=. the series converges for almost all x E I and f is HenstockKurzweil integrable over I if.H e n s to&. Lemma 4. Then. We begin with a lemma.74 Let f k . by the linearity of fk the integral and the nonnegativity of the functions fk. { S m (x)}:=. On the other hand. a 1 I c R* + R be HenstockKurzweil integrable f o r all k . in either case. fk < 00. As in the case for the Lebesgue integral. Since each f k 2 0. . infk f k is HenstockKurzweil integrable over I .K U ~ Z W in t egml ei1 185 Theorem 4.
by the Monotone Convergence Theorem. fk (x)af the limit i s finite otherwise 0 is HenstockKurzweil integrable over I with S.. Thus. infk f k is finite valued everywhere on I . the function @k defined by (x)= inf { f j (x): j 2 k } @k for each k E N is HenstockKurzweil integrable over I . Note that since a 5 infk f k 5 f l .74.71. Lemma 4. f k z finite almost everywhere s in I and the function f defined by m={ lim infk.12) that @ k } E 1 converges and is hence bounded. {@k}T=1is an increasing sequence which.. a : I c R* + R be HenstockKurzweil integrable for all k .75 (Fatou’s Lemma) Let f k . lim infk. Since a 5 @k 5 f k on I for all k . we can apply the Monotone Convergence Theorem to the decreasing sequence of functions {gk}El which converges to infk f k . Thus. by Lemma 4. @k liminf k+m 1 fk. Since a 5 gk for all k . Since { J I @k}r=l is monotonic. converges pointwise to j almost everywhere in I .70. by the comment in the paragraph following the proof of Theorem 4.. by definition.186 Theories of Integration Proof. We can now prove Fatou’s Lemma. infk JI gk 2 jIa > 00. By Lemma 4. the function gk = infl<j<k f k is HenstockKurzweil integrable over I by Proposition 4.12) {sI . $I f k < 00. it then follows from (4. Then. Since a 5 f k . it follows that each function finite valued on I and is which implies Further.65. and suppose that a 5 f k a I and n liminfk. f is HenstockKurzweil integrable and by (4. f 5 Proof.
Since a 5 f k .finite almost everywhere a I and the function f de. we may assume all the hypotheses hold for all 2 E I . Fatou’s Lemma shows that liminfk. the Set has measure zero. limsup. If z $ E .finite otherwise is HenstockKurzweil integrable over I with / I f 2 l ik mosou p i f k .. Let Ek = {X E I fk ( 2 ) < Ct or fk ( 2 ) > p ( X ) } . Define f by limk. Then.d in I and lim Supk. Then. Corollary 4. then f k ( 2 ) + f ( 2 ) and a ( 2 ) 5 f k (x)5 p (x) for all Ic E N. f k as . the result dual t o Fatou’s Lemma also holds. fk (x)i f the limit is finite otherwise Suppose that there are HenstockKurzweil integrable functions a. f is HenstockKurzweil integrable over I and Proof. f k is finite almost everywhere in I and .E f k and JE f = 0.f the limit is .P : I +R such that a 5 f k 5 . Then.. 0 As in the case of the Lebesgue integral. h We are now prepared to prove the Dominated Convergence Theorem. p : I C R* + R be HenstockKurzweil integrable f o r all k . and suppose that f k 5 . Theorem 4. f o r all k E N.76 Let f k . JI f k > 00. I C R* +R be HenstockKurzweil integrable over I and suppose that { fk)E1converges pointwise almost everywhere o n I ..d almost everywhere in I ..fined n by lim supkrw f k ( 2 ) i.HenstockKurzweil integral 187 which completes the proof. Since JI f k = JI.77 (Dominated Convergence Theorem) Let f k .
Thus. De.. . We conclude this section with the Bounded Convergence Theorem. Note that if f k and g are HenstockKurzweil integrable and fkI 5 g. almost everywhere in I and fk is finite Combining these results. absolutely integrable and f k is absolutely integrable by Corollary 4. we see so that For the Lebesgue integral. and . since f k 5 p. the condition of Theorem 4. I < I Corollary 4. The importance of the condition a 5 f k 5 . then One need only observe that the function g (x)= M for all x E I is HenstockKurzweil integrable over I . hence.8 for the HenstockKurzweil integral is that the functions f k .6 may be conditionally integrable.eories of Integration Similarly. where g is a vergence Theorem employs the condition that I fkI Lebesgue integrable function.76 implies that limsupk.! be HenstockKurzweil integrable over a bounded interval I and suppose that { fk}F=1 converges pointwise almost everywhere o n I .62.for all k and all x E I . the usual statement of the Dominated Cong .188 Th. a . then g is nonnegative and. such an hypothesis implies the hypothesis above.100. Since fkI 5 g is equivalent to g 5 f k 5 g.fine f by limk+f k (2) if the limit i s finite o therwis e If there is a number M so that 1fk (x)l 5 M .78 (Bounded Convergence Theorem) Let f k : I C R* + & .77 is more general than the condition of Theorem 3. Corollary 4.
2 L Ik fk = 2t i Lk k fk = klim +m fk = 1fb If f is Lebesgue integrable. Theorem 4. We now show that the HenstockKurzweil integral is more general than the Lebesgue integral. Since { f k ) & increases to f pointwise. L s. Each f k is nonnegative. f is Lebesgue integrable iJ and only $i is HenstockKurzweil intef grable. we will use L JI f to denote the Lebesgue integral of f.b]." f = S. realvalued function defined on an arbitrary interval in R. We now consider the relationship between Lebesgue integrability and HenstockKurzweil integrability. by Theorem 3. Since (pk is a step function. and bounded so. again. holds for both integrals) to conclude that f is Lebesgue integrable if.H e n s tock. a function is Lebesgue integrable if. Since the Lebesgue integral is an absolute integral (that is. and I = [a. Then. there is a sequence of step functions {(.ok}F=l such that (pk + f pointwise a.b] is a bounded interval. it is absolutely Lebesgue integrable) and the HenstockKurzweil integral is a conditional integral. Next.are HenstockKurzweil integrable. suppose that f is an arbitrary nonnegative. measurable.1]. and only if. J!k f k = Jk f k . Further. When either of the integrals is finite.e. k ] (z)." ( p k . f+ and f .Kurzw eil integral 189 4. L Ja (Pk = s. with a bound of M .8 HenstockKurzweil and Lebesgue integrals We saw earlier that every Riemann integrable function is HenstockKurzweil integrable. so that by the Bounded Convergence Theorem (which holds for both b integrals). f is HenstockKurzweil integrable.k . . and I'pk (z)l 5 b M for all k E N and z E [a. and only if. consequently. First suppose that f is also bounded. both defined on (0. provide examples. then f+ and f.1 is HenstockKurzweil integrable and not Lebesgue integrable. f. the function in Example 4. By linearity. The unbounded function l / f i and the Dirichlet function. by the previous case. k } X [ . we see that fk k = : . we can apply the Monotone Convergence Theorem (which. As above. Define a sequence of functions { f k } r = 1 by f k (2) = min { f (z) . Further. measurable.are Lebesgue integrable and. sI sI Proof. Then.67.79 Suppose that f : I + R is nonnegative and measurable. by defining the gauge to have constant length. I n either case. there are HenstockKurzweil integrable functions which are not Riemann integrable. L f = f . the conditions cannot be equivalent.
by linearity.f ( a ) .” f ( t )d t .b] + R be HenstockKurzweil integrable o n [a. 4. We now have the necessary background to prove a general version of Part I of the Fundamental Theorem of Calculus for the Lebesgue integral. Then. This is the content of Part I of the Fundamental Theorem of Calculus (Theorem 4. Theorem 4. f iand f .3. I n either case. Thus.3. F (z) = s. the indefinite integral of f. so Corollary 4. is differentiable at x. By Theorem 4.9 Differentiating indefinite integrals One of the most valuable features of the HenstockKurzweil integral is its ability to integrate every derivative. Then. page 1581. f is Lebesgue integrable. but the converse is not valid.11 and [Swl. that of differentiating integrals. IX. we have the following corollary.80 can be dropped. Vol.b]. Corollary 4.are nonnegative and HenstockKurzweil integrable. see [N. f is absolutely HenstockKurzweil integrable. Theorem 4.” s.81 (Fundamental Theorem of Calculus: Part I ) Suppose that f : [a. 4. .82 Let f : [a. and only i f .completing the proof. We will show in Corollary 4.b] and f is Lebesgue integrable o n [a.b] + R is differentiable o n [a. I.86 that every HenstockKurzweil integrable function is measurable. F .b]. We now turn our attention to the second part of the Fundamental Theorem of Calculus.b] and continuous at x E [a. Thus. so the measurability condition in Corollary 4. Proof. Then.16). the integrals agree. suppose that f is absolutely HenstockKurzweil integrable. f is Lebesgue integrable ij. f f’ = f ( b ) . We first observe that if f is continuous a t x then its indefinite integral F .16.”‘ = f‘. 0 s.80 Suppose that f : I + R is measurable. By assumption. is diflerentiable at x and F‘ (2) = f (2). On the other hand.80 im‘ plies that LS.17. ‘ Then.” For a proof that does not use the HenstockKurzweil integral.190 Theories of Integration f is absolutely HenstockKurzweil integrable. Lebesgue integrability implies HenstockKurzweil integrability.
is equal to the measure of I . and consequently also F ' . I F'(z)= f (2).1] be a null set and consider f = xE.84 Let C = ( I i : i = 1. F' (x)# f (z) if x E E . Thus. F is diflerentiable at almost all z E [a. the indefinite integral is still differentiable almost everywhere. then so that Similarly..F ( x ) I  f(x)l L E . . Since f is continuous at x.zI < 6.b].b] and I . then the length of I .b] + R is HenstockKurzweil integrable. m ( I ) .b]. In order to prove this theorem. Observe that we cannot do better than a statement which holds almost everywhere. 1F ( x + hh . [a. I f O < h < 6 is such that x + h E [a.finite set 0.b] and F' (z) = f (x). . f is equal to 0 for almost all x E [0.f intervals in R. N ) be a . we need another covering lemma. Jk E C such that . for lhl 5 S and x ) + h E [a. It follows that F . let 31 be the interval concentric with I and having three times the length of I .f (2)< E . for h < 0. Let E c [0. is identically 0 on [0. Then. . I]. there exists a pairwise disjoint collection J 1 . Then. Theorem 4. for t E [a..HenstockKurzweil integral 191 Proof. Recall that if I is an interval in Iw. then t E: E > 0 there is a 6 > 0 so that if < f ( t ).Thus.83 (Fundamental Theorem of Calculus: Part 11) Suppose that f . k Lemma 4. . I7 When f is merely HenstockKurzweil integrable. Then. Thus. ' ( I ) . . Given an interval I . .1].
we may assume that Set 5 = I . then Ii c 3J1. . there is nothing to prove. we let Jz be the element of C1 with the smallest index (and hence the greatest length). Thus.then F is differentiable at x and F'(x)= 4 It suffices to show that E p is null for any p > 0. there is a neighborhood U. # 8. by Henstock's Lemma there is a gauge y on [a. w) and F(u) vu f(x)l > p.14) f (4. the intervals in (51. be the set of all x E (a. ~ J ~Ii c i #~ so that We are now ready to prove Theorem 4. Then.b) that satisfy condition (*. Suppose that x E .w] such that x E (u. Fix E > 0.192 Theories of Integration Proof. Proof. we say that x E ( a .E1/.. with x E (u.. for all n 2 1.w). say k . By construction. . Since f is HenstockKurzweil integrable. By reordering the intervals if necessary.44). the selection of intervals Jj ends after finitely many steps. this inequality holds when u is replaced by x. so assume that E. of x such that for any interval [u. If E. Set C2 = {IE C1 : 1 2 n I = 8) and continue as above. if x E. is a j so that I C 3 J j . w] one has 4 By the continuity of F (Theorem 4. UEIIi = Ujk=l U I .83.) if every neighborhood of x contains an interval [u. Let C1 = { I E C : 11n I = 8) and note that if I E C and 1 1 i Ii 4 C1. J k } are pairwise disjoint. Next.b] such that (4. Thus.. and if Ii E C is not selected. (4. = 0. For a fixed p > 0. Since C is a finite set. has measure 0.EI/. ) and set E = u:=. c U. since then E = U:=.b) satisfies condition (*. then there .13) Let E.
.HenstockKurzweil integral 193 for any yfine tagged subpartition D = {(xi. such that y1 (x) c (u.b + 1 by setting F ( t )= F ( b ) for b < t 5 b + l . . 5.vz) for all x E E.. and [Ro. choose a gauge y1 on E.] that z E [u. Then. Jk c 7 (zk)c ( u Z k r v Z k ) . 4.b] f R be Lebesgue integrable. vi]) For x E E. T h e n . 3 Since the extended function is continuous on [a.. e vZi]) : i = 1.. .vy.b]. Since E.~. Vol..N } is a yfine tagged subpartition of [a.. . IX.13) holds.b]. C U&OJk C ( J k ) = a < E .13) and (4.e. I.]C y ( x ) and such (4. Let f : [a. and E.4. ( u y KvyK) } such that .14)) It now follows from (4.. it now follows that E..b].. Corollary 4. . is null. in [a. c U k E O J k c [a.39. : k = 1. Z } of [a. xkEo c. page 1071. .a < 00 and pick N such that (Jk) > 2' uZk) Apply Lemma 4.21. N } to get a set of nonoverlapping intervals { (uyl. it follows that the sequence of functions { f k } E 1 defined by + . Suppose f is HenstockKurzweil integrable over [a. F is continuous on [a. page 1501. xkEO and 0 Since every Lebesgue integrable function is (absolutely) HenstockKurzweil integrable. ) .[ui.b].15) that E > a.10. by (4. .b].. Next.9.. . choose an interval [u. [Swl. . Since {(xi. .~. Part I1 of the Fundamental Theorem of Calculus of Lebesgue integrals follows as an immediate corollary. there exist countably many nonoverlapping closed intervals { J k : k E a } and points {xk : k e a } such that x k E Jk n E.==.b]* 1 Let a a= e ( J k ) 5 b . .b 13.1.85 a. see [N. F' = f For a proof that does not use the HenstockKurzweil integral.[uZi.84 to ((ulk..extend F to [a. : i = 1.3. By Lemma 4.
89 Suppose that f : I + R is HenstockKurzweil integrable over I . which implies that f is mea Corollary 4. f is measurable since I . set In = I n [n. If I is unbounded. By Theorem 4. F(t+ k By Theorem 4. Due to this corollary. in Theorem 4. Then. The proof when I is a bounded interval is contained in the previous paragraph.6. and since f is the pointwise 00 0 f is measurable.. Then. we can drop the assumption that f is measurable. Proof. In either case.87 Corollary 4.86. Then. f is Lebesgue integrable over I if) and only if) f is HenstockKurzweil integrable over every measurable subset E c I . limk.79 and Corollary 4. first. We say that f is HenstockKurzweil integrable over E if xE f is HenstockKurzweil integrable over I and we set The next result follows from Theorem 4. For the converse. we define the HenstockKurzweil integral of f over a set E in terms of the function xEf.101 and [Pf. the integrals agree. 5. we have Theorem 4. that f is Lebesgue integrable.31. f is Lebesgue integrable if) and only if) f is absolutely HenstockKurzweil integrable. By Corollary 4. As in the Lebesgue integral case. Then. E+ and E . .83. surable. f is (Lebesgue) measurable over I .86 Let f : I c R R be HenstockKurzweil integrable over I. n].87 Suppose that f : I +R. Definition 4.88 Let f : I + R and E c I . is a bounded interval.80.= { t E I : f ( t ) < 0).The function X I . x Ef is (absolutely) integrable over I . so that f is HenstockKurzweil integrable over E . Let E be a measurable subset of I .194 Theories of Integration is measurable on [a. since both HenstockKurzweil and Lebesgue integrability imply (Lebesgue) measurability.are measurable sets. put E+ = {t E I : f ( t )2 0) and E. f Other proofs of the measurability of HenstockKurzweil integrable functions can be found in [Lee.87. X E f is Lebesgue integrable over I .b]. Suppose.3. f (t) = F'(t) = 3 )F(t) for almost all t f E I . Proof. Thus. limit of { X I .
.91 Let E c R. X E is absolutely HenstockKurzweil integrable.HenstockKurzweil integral 195 so that sE+ and sEf both exist. O Let E c R be a null set.83.n].. this is just Theorem 4. I n either case. we will characterize indefinite integrals for the three integrals considered so far. xE is HenstockKurzweil integrable.90 Let f : R + R be HenstockKurzweil integrable and suppose that f = 0 f o r every bounded interval I .are Lebesgue integrable over I .e. so that xE is Lebesgue integrable and m (E) L xE = 0. Then. f (x) = F'(x)= 0 for almost all x E [n. We have already seen that sets with measure 0 play a n important role in integration. Then.40.It follows that f = 0 a. Thus. sI Proof.10 C h a r a c t e r i z a t i o n s of indefinite integrals In this section. E is a null set. We begin with the HenstockKurzweil integral. E is a Lebesgue measurable set with finite measure if.fis both f+ and f. On the other hand.n].e.92 Let E c R. Corollary 4. E is a null set if. in R. However. Then. We now investigate some properties of functions with integral 0. Thus. Then. s 4. Suppose that f is a HenstockKurzweil integrable function on an interval I c R. and only if. F . proved in Section 4. suppose that xE is HenstockKurzweil integrable and J x E = 0. 0 4 9 1 Functions with integral 0 . Let F (2) = s". We have proved = C o r o l l a r y 4. Thus. Of course.4. f = f+ Lebesgue integrable. xE = m ( E ) . xE is HenstockKurzweil integrable and s xE = 0. xE is HenstockKurzweil integrable and s xE = 0. C o r o l l a r y 4. Then. F (z) = 0 for all x E [n. By definition. and only if. Then. f for all x E [n. the argument above can easily be modified to prove the following result.n]. By Theorem 4. the indefinite integral of f . Since f . is differentiable and F' = f almost . E is measurable and L s xE = 0. f = 0 a.
. m. . suppose that a function F is differentiable almost everywhere in an interval I .l xi]): i = 1. . Define f1 : I +R by fl ( t ) = f ( t ) for t E I \ 2 and fl ( t ) = 0 for t E 2. with [ . Then. Does it then follow that the derivative F‘ is HenstockKurzweil integrable? In general the answer is no. F is the indefinite integral of a HenstockKurzweil integrable function f : I + R iJ. we assume that F‘ = f on I \ 2 and F has negligible variation over 2. consequently. ..[xil. and only iJ there is a null set Z c I such that F‘ = f o n I\ Z and F has negligible variation over 2.94 Let F : I = [a. As we shall see. in order for F‘ to be HenstockKurzweil integrable. m } . Then. For the sufficiency.” f l . . Extend f to all of I by setting .F (xil))l L 2E. F’ = f almost everywhere on I . xi]): i = 1. . Let 2 be the set where the equality fails.xi1)  ( F (Xi) . Theorem 4.. By Henstock’s Lemma. we need to know more about how F acts on the set where it is not differentiable in order to conclude that its derivative is integrable. Given E > 0. D = { ( t i .F (b)l < E for every yfine tagged partition 2) of I . Proof. since ti E 2 . by Theorem 4. f (ti)= 0 for all i = 1. F (b) = s.. assume F is the indefinite integral of a HenstockKurzweil integrable function f . m i=l Note that the ti’s must be elements of E .m. ~ i .93 Let f : I + R and E c I .” f l and. Definition 4.196 Theories of Integration everywhere. F (x)= s. We say that f has negligible variation over E if for every E > 0. Then. there is a gauge y so that for every yfine tagged subpartition of I .b] +R. in addition to being contained in [zil.83. there is a gauge y such that (S(fi. i=1 But. D). . For the necessity..for some null set 2.m} is a yfine tagged subpartition of I with tags ti E 2.. so that m i=l as we wished to show. . . Suppose D = { ( t i . On the other hand. . c m If1 (ti)(Xi . xi]. t i E E f o r i = l . .
.F ( a )= Jdb f. By the i=l tiEI\Z Therefore. Suppose that V = { ( t i . Since F has negligible variation over 2.6 ( t ). E. tition of I .xi]) : i = 1. Let E > 0.. where 6 ( t ) is the value corresponding to E > 0 and the function f in the Straddle Lemma (Lemma 4.zi1) L: E (1 + b .6).F ( a ) c m i=l f (ti) (xi . F (b) . .HenstockKurzweil integral 197 f ( t ) = 0 for t E 2 . I 5 Straddle Lemma.93.. Define a gauge y on I by setting y ( t )= yz ( t )for t E 2 and y ( t )= (t .” f for all z E I . z .there is a gauge yz satisfying Definition 4.t 6 ( t ) )for t E I \ 2. We claim that the extended function f is HenstockKurzweil integrable and F (z) = s. which shows that f is HenstockKurzweil integrable over I with F (b). . + = I iI I .m } is a yfine tagged par[il .a ) .  Since F has negligible variation over 2 and f = 0 on 2 . Then.
we call V a Vitali cover of E .1. bounded subintervals of R" covers E in the Vztali sense if for all x E E and for all E > 0.b] and E > 0. Example 4.96 Let f : [a. The existence of the intervals I X . sax (2)  F (a) = 0 We now turn our attention to characterizing indefinite integrals of Lebesgue integrable functions.95 Let E c R". A family V of closed. For later use. To do this. Given an interval I in R". we need to know that every increasing function is differentiable almost everywhere. € is guarann teed by the differentiability of f . we will discuss Vitali covers in ndimensions. recall that v ( I ) represents the volume (and measure) of I . For each x E [a.b] + R be differentiable over [a. . Definition 4. Recall the upper and lower derivatives. it follows that < V = { I x .b]. To prove that an increasing function is differentiable almost everywhere. the set 1 : (x.198 Theories of Integration Applying the same argument to the interval [a. 00 < D f(x)= Qf(z)< co. containing x in its interior.x] yields F f . E y) E and n E N is a Vitali cover of E . €be a closed interval of length less than E . we first study monotone functions and their derivatives. so that F is an indefinite integral of f. 4. €: x E [a. there is an interval I E V so that x E I and the v ( I ) < E . Given a set E c R2.1 Derivatives of monotone functions In order to characterize indefinite integrals of Lebesgue integrable functions.b]. and that f is differentiable at x if. If V covers E in the Vitali sense. and discussed in Section 4. nf of.b] and E > 0) is a Vitali cover of [a.10. Since l(IxlE) E .b]. and only if. A typical cover that arises in applications is given in the following example. such that for all y E IzlE [a. let I X . we will use Vztali covers.
shows that a finite set of elements from V covers all of E except a set of small measure. + ~ Q. This implies that i=l 5" * It remains to show that m.Qi) = 0} . Let j be the smallest such index and let Q. Since e (Q) > 0. = 0.Qi) < E . QI. + ~2 Sk.Kurzn)eil integral 199 Suppose a set E of finite outer measure is covered in the Vitali sense by a family of cubes V . ( E \ U z l Q i ) < E .97 (Vitali Covering Lemma) Let E c Rn have finite outer measure.Qi.Hens t o ck. .Qi. known as the Vitali Covering Lemma. Suppose that x E E \ ULlQi.Qi) = 0.==. < 00. Suppose that a family 0. then e (Q) 5 Sk. let e ( Q ) be the length of an edge of Q and note that v (Q) = mn (Q) = [e (Q)ln. We need consider only the Q E V such that Q c J . we get a sequence of disjoint cubes such that U g 1 Q i c J .set N = k and {Qi}El is the desired cover.f cubes V is a Vitali cover of E . By construction. SI.it follows ) that limk+m SI. Since m n ( J ) < 00. Otherwise.= sup { e (Q) : Q E V and Q n (U:=. . Since UElQi is a closed set. The following proof is due to Banach [Ban]. be the cube concentric with Q j and having edge length 5 times as long. j > N and e (Q) 5 Sj1 5 2e (Qj). Proof.) . mn (Qi) < 4 (ULIQ. there is a Q E V such that z E Q and Qn = 0. Since Czl mn (Qi) < 00 and 2e ( Q I .. Assume that Q1. Since + SUP {V (Q) : Q E V } 5 mn ( J ) < (1 + ~ ) m( E ) < t 00. The next result. ( E \ UiN. Choose an N so that C&+. have been chosen. Let J be an open set containing E such that mn ( J ) < (1 E) mz ( E ) . Given E > 0 . Set Q I .. = When E \ Ut==. Since E @ Ut==.there is an Q E V such that e (Q) > S k / 2 and Q f~ (u. let SI. there must be an i such that Q n Qi # 0. we can choose Q1 so that e ( Q 1 ) > asup{e(Q) : Q E V } . Define a sequence of cubes by induction. If E c Ut==. Lemma 4. If Q n Qi = 0 for all i 5 k . . Given a cube Q. there i s a finite collection of pairwise disjoint cubes {Qi}Li c V such that m.Qi# 0 for all k .
EU. this would imply that (x)and f (x) are equal almost everywhere. m. Proof. We now return to the differentiation of monotone functions. Then. The proof above . ( E ) .b] + I be a n increasing function. ci=l + + Remark 4. then x E Qj*for some j > N . which has fixed eccentricity. v E Q.mn ( Q l )= 5nmn(Qi) and we have 00 00 i=N+1 i=N+1 as we wished to prove.yi]) < E . We claim that E = {x E [a.99 Let f : [a.for cubes applies to this situation since the key geometric estimate. f ' exists almost everywhere in [a. = Let r = m* (EU.b]: Df (x)> D (x)} has measure J zero. we will apply the Vitali Covering Lemma to a collection oaf intervals that are obtained by repeated bisection a . replacing E by 2 ..113. (x) < v. and choose an open set I 2 EUlw that fix such m ( I ) < r E. D (x)> u > v > Q f ( x ) } so that E = Set EuIw = {x E [a. 0 We have actually proved more. which implies that E \ U L I Q i c U&+lQ:. . Since ( 2 ) 2 Qf ( x ) for all x.98 I n Theorem 4.u).w) 0 for all u . containing x such that f (p). . Q c Qg.u.fixed interval in R". By iterating the argument.' ~ a t the kth iteration.200 Theories o f Integration Therefore. there are pairwise disjoint intervals {[xi. if V covers E in the Vitali sense. Since intervals [a. Thus.f ( a )< v ( p . if x E E \ UzlQi.a ) .~ U E l [ x i . Since U z l Q i c J and the Qi's are N pairwise disjoint. R Theorem 4. since Q n Qj # 0 . E > 0.2) p] is covered in the Vitali sense by the collection nf of of + of By the Vitali Covering Lemma. This implies that \ . there are arbitrarily short closed Let x E EU. namely Q n Qj # 0 implies Q c Q. w E ~ E UItwis enough to show that m* (EU. (for the smallest j such that Q n Qi # 0 ) continues to hold f o r such a collection of intervals. Since e ( Q T ) = 5e ( Q i ) .b]: U U . then there is a sequence of pairwise disjoint cubes { Q i } E 1such that m n ( E \ U P l Q i ) = 0 and CF1mn ( Q i ) < (1 E ) m: ( E ) . yi]}E1 C V such that m* (JT. (Qi) < (1 E ) m. Thus.b].
it follows m* ({x E [a. there is a collection of pairwise disjoint intervals { [ ~ j .yi]. Thus. Eu. and since u > v. Since Df (s) > u.. Since E > 0 was arbitrary. It remains to show that (z) is finite almost everywhere.pi] for some i and f ( p ) . Fix k .. then f' exists almost everywhere and the proof is complete. E is measurable with measure 0. Finally. This implies that u ( r .e.b] : nj (z) > k } . p] c [zi.f (A) > u ( p . N i=l i=l Suppose s E A is not an endpoint of any [xi. since m* ({z E [a.b] : Df(z) = oo}) = 0 so that Df(z) is finite a.E .26) < v ( r + e). 0 . t j ) such~that m* ( A n } ~ [ s j . Further.f (4 k for all k > 0. i = 1. .p] containing s such that [A. we have ur 5 u r .yi] for some i. by the Vitali Covering Lemma. .. and since f is increasing.HenstockKvrziueil integral 20 1 the set A = E. Repeating the nf argument in the previous paragraph yields Since E > 0 is arbitrary. As above. For if this were the case.n N (UL1[xi. . t j ] ) ) > r . Hence. consequently.pi]) has outer measure m* ( A ) > 7 . in [a. m* (E.v is measurable with measure 0 and. ..there are arbitrarily short intervals [A.b].A). Thus. E > 0 and set Ek = {z E [a.bl.b] : Df(z) = oo}) 5 m* (nr=lEk) 5 m* ( E k ) 5 f ( b ) ..2~ and (ug.) = 0. f ' exists and is finite almost everywhere in [a.f ( a ) 2 km* (Ek). we see that r = 0. f ( b ) . M M j=l Since each [ s j t j ] is i= 1 contained in [xi. N .
Corollary 4.b] and set F ( z ) = Ls. Lebesgue integrable function on [a. We use this condition to extend the idea of absolute continuity to functions. lapping subintervals of [a. Suppose that f is a nonnegative.d] c [a. then f is digerentiable a.b] if for every E > 0. 0 . Choose N E W such that N > .b]. Fix E > 0 and choose 6 > 0 by absolute continuity. Clearly.bil}f=l of nonoverk (bi .b] is less than or equal to N . every absolutely continuous function is uniformly continuous. Definition 4. It then follows that cf=.b] into N nonoverlapping intervals each of length .< 6.bi]}f=l be a finite set of nonoverlapping intervals in [a.F (ai>l< E: for every finite collection { [ a i . there is a 6 > 0 so that I F ( b i ) .b].a < 6. We say that F is absolutely continuous on [a.ai) < 6.”f. p] every such function also has bounded variation.d] is at most 1.101 Let F : [a.b] such that x.102 Suppose that F : [a.b] + R.=. It N follows that the variation of F over [a.b]. Let { [ai. Thus. we observed that F is absolutely continuous with respect to Lebesgue measure.100 Iff : [a. then the variation of F ba over [c. we can .59 that every function of bounded variation is the difference of two increasing functions.b] +R has bounded variation o n [a. 6 ba divide [a.a i ) < 6. in [a. there is a 6 > 0 so that m ( E ) < 6 implies @ ( E ) < E .e. Proof.b]. we defined a measure @ to be absolutely continuous with respect to Lebesgue measure if given any E > 0. which is seen by considering a single interval [a.c < 6. It then follows from Theorem 4. In the same remark.Then. In Remark 3.202 Theories of Integration We saw in Theorem 4.b] and d . Then. Choose 6 > 0 corresponding to E = 1 in the definition of absolute continuity. if [c. with p . F has bounded variation o n [a.99 that a function of bounded variation is differentiable almost everywhere.93. Proposition 4. b] and suppose that (bi . Further.b] + R is absolutely continuous o n [a.
To show that (1) implies (2). Proof. (2) F i s absolutely continuous o n [a.. if t E: 2 . t ( J k ) < 6.94. note that so that the absolute continuity of F follows from the comments above. where Z is the null set where F' . i ] ): i = 1. Define a gauge on [a. for both integrals.2 Indefinite Lebesgue integrals A wellknown result for the Lebesgue integral relates absolute continuity and indefinite integration. xi] C J k t i SO that m i=l kEo . If t E I \ Z . In the following theorem. m } is a yfine tagged subpartition of [a. To see this. [xil. there exists a countable collection of open intervals { J k } k E c r such that 2 c U k E o J k and C & .b] + Iw..100.94. The . (3) F has bounded variation o n [a. Theorem 4. fix E.b]and F' exists almost everywhere.following statements are equiv (1) F is the indeafinateintegral of a Lebesgue integrable function f : [a.102 and Corollary 4. Condition (3) should be compared with the condition for the HenstockKurzweil integral given in Theorem 4. [zi~. Then. the indefinite integral has negligible variation over the null set where its derivative fails t o exist. let kt be the smallest integer such that t E. Let 2 be the null set where F' fails to exist. By Proposition 4.b] as follows.b] + R.b] and F has negligible variation ouer Z. Suppose that D = {(ti. J k t and set y ( t ) = J k t .b] x with tags ti E 2. > 0 and choose 6 > 0 by the definition of absolute continuity.HenstockKurzweil integral 203 4.10.fails to exist.103 alent: Let F : [a. In particular. F has bounded variation on [a. . We claim that F has negligible variation over Z. Since 2 is null.b]. set y ( t ) = R. . Suppose (2) holds. we show that these conditions are also equivalent to a condition similar to that of Theorem 4.
Thus. f is absolutely HenstockKurzweil integrable by Theorem 4. We observed above. note that if A is the Lipschitz constant for F .” f for all x E [a.b]. $ Proof. f is continuous almost everywhere. F‘ (z) exists and equals f (z) for almost every z E [a.F ( a ) = s.e. 5 that If for all x.b]. m i=l so that 2 has negligible variation over E . it is enough to prove the necessity of the result.F (x) everywhere limit of the continuous difference quotients 1 > n F‘ is measurable if we define F’ to be 0 where F’ fails to exist.b]3 R if.b]. Theorem 4.2. By the previous remarks.3 Indefinite Riernann integrals We conclude this section by considering indefinite integrals of Riemann integrable functions.42. 4 then F ’ ( z ) = limg+% (3) YX .b].b]. f ( t )= F‘ ( t )for t E I \ 2 and f ( t )= set 0 for t E 2. Further. Thus.b]. in Section 2. 4 such that (x)/ M for all z E [a.104 Let F : [a. F’ exists almost everywhere. Since F has bounded variation.b] R. F is the indefinite integral of a Riemann integrable function f : [a. 0 4.204 Theories of Integration By absolute continuity.10. F satisfies a Lipschitz condition o n [a. By Theorem 4. and only if.y E [a. F‘ is . Further. Then.( I 5 M imply that IF’ (x)1 5 M whenever the limit exists. To show that (3) implies (l).b] + Since f is bounded. Thus. F satisfies a Lipschitz condition on [a.. by Corollary 2.87. F (x’+ $) . F is continuous and F‘ is bounded whenever it exists. since F‘ (2) is the almost . To see this.60.32). f is Lebesgue integrable and F is the indefinite integral of a Lebesgue integrable function. By the Lipschitz condition. there is an A > 0 R. and F’ is equal almost everywhere to a bounded . By Theorem 4.function f which is continuous a. (2) implies (3). so that by Part I1 of the Fundamental Theorem of Calculus for the Riemann integral (Theorem 2. (x> and I If).94. Suppose F is an indefinite integral of a Riemann integrable function f : [a.5. f is HenstockKurzweil integrable and F (z) . .
e.105 Let I = [a.11 The space of HenstockKurzweil integrable functions In Section 3.” f1. By Theorem 4. the Alexiewicz seminorm of f is defined to be .where M is the Lipschitz constant for F . f Since F satisfies a Lipschitz condition. both of which require more sophisticated constructions than the Riemann integral. Since f is continuous a. F (z) . is not meaningful. we consider the space of HenstockKurzweil integrable functions. where f1 = F’ almost everywhere. due to Alexiewicz [A].31 is HenstockKurzweil integrable while 1 f’l is not.rzw eil integral 205 Lebesgue integrable and since F’ = f almost everywhere..b] c R and let ‘FIX ( I ) be the vector space of all HenstockKurzweil integrable functions defined on I .Ku.9 of Chapter 3.” F’ = L s. the function f’ defined in Example 2. L s. F ( 2 ) .one need only choose 6 5 E / M .Henstock.b].F ( U ) = C ~ x f l = C ~ x F f’ = ~ ~ I . If f E 7K ( I ) . 4. For example. F is the indefinite integral of a Lebesgue integrable function. As we saw in the proof of that theorem. it is absolutely continuous on [a. the L1norm defined on L1( E ). In this section.b].. which we now define. 0 There is something troubling about this proof.103. Definition 4. Thus.F ( a ) = L s. However. which in turn are consequences of results for the HenstockKurzweil integral. It relies on results for the Lebesgue integral.of Lebesgue integrable functions on a measurable set E and showed that L1( E ) had a natural norm under which the space was complete. Thus. L 1 ( E ) . we considered the vector space. so that RsUx = Lsuxf = CsuxF’. F is the indefinite integral of the Riemann integrable function f. the space of HenstockKurzweil integrable functions does have a natural seminorm. Since the HenstockKurzweil integral is a conditional integral. f is Riemann integrable.” f for all x E [a.
we have that l l f l l = 0 if.116) asserts that the space L1 ( E ) is complete under the L1semimetric.e. Example 4. . the space does have other desirable properties. is a Cauchy Since { p k } E l converges uniformly to p . for example.. are identified.1])with respect to the Alexiewicz seminorm. then 1111 is actually a norm on 'FIX ( I ) .1]. [DS. pk ( t )= s i p : for every t E [0.106 Let p : [0. if functions in ' I ( I ) which FX are equal a. 4. see Exercise 4. which is a contradiction to the definition of p .e. that 'FIX ( I ) is not complete under the semimetric generated by the Alexiewicz seminorm.) By the Weierstrass Approximation Theorem ([DS. {pk}El so so Although the space 'FIK ( I ) is not complete under the Alexiewicz seminorm. 1 1 defines a seminorm. pk ( t )= pL + f uniformly in t E [o. [ S W ~ . (with derivative j ) . there is a sequence of polynomials {pk}zl that converges uniformly to p such that pk (0) = 0 for all k . Suppose that there is an f E 7K ([0. 1 By Part I1 of the Fundamental Theorem of Calculus.206 Th. For a discussion of the properties of 'FIX ([0. p is differentiable a.1])is not complete. The RieszFischer Theorem (Theorem 3.f l l +0 as k + t oo.58.60. Since many of the higher dimensional results follow from proofs analogous to their onedimensional versions.e. page 239]). Then. By Part I of the Fundamental Theorem of Calculus. We begin by laying the groundwork necessary to define the integral.) From Corollary 4. page 1371. Thus.eories of In.1])such that IIpk .tegration (To see that 11. (See. f = 0 a. it follows that sequence in 'FIX ([0. our presentation will be brief.1] + R be a continuous and nowhere differentiable function with p ( 0 ) = 0.90 (which is also valid for intervals I c R) and Exercise 4.12 HenstockKurzweil integrals on Rn We conclude this chapter by extending the HenstockKurzweil integral to functions defined on ndimensional Euclidean space. 1 so that p ( t )= t f . and only if. however. see Chapter 71. We show. l]). Hence 'FIX ([0. Thus.
109 Let I be a closed interval in (a*)" y be a gauge o n I .107 A partition of a closed interval I C nite collection of closed. The volume of an interval I C ( a * ) " is defined to be v ( I ) = n t ( I j ) .. .. . . and only if. k } of I with I = Usz1J j . j=1 is a fiDefinition 4. a gauge on I intervals to points in I .k . Then. As in the onedimensional case. A tagged partition of I is a finite set of ordered pairs V = { (xj. Assume that there is no yfine tagged partition of I . . for if each of the Zn subintervals . At least one of these subintervals must not have any yfine tagged partitions... The point xj is called the tag associated to the interval J j . the Riemann sum of f with respect to D is defined to be S(f... that the function f has value 0 at all infinite points (that is. In order to use these sums to define a multidimensional integral. each Ij is open (closed) in R*. is closed and bounded. Bisect each Ij and consider all the products of the n bisected intervals. . .Do> = Cf(+(Jj). containing x. nonoverlapping subintervals { J j : j = 1... k } is called yfine if Xj E ( J j ) for j = 1. k } such that { J j : j = 1. . First.. where each j=l Ij is an interval in It*. and Theorem 4.J j ) : j = 1. We say that I is open (closed) in (It*)" if. k } is a partition of I and xj E J j for all j . as before. suppose that I = Il x . there i s a yfine tagged partition of I . . A tagged partition V = {(xj.x I .J j ) : j = l .with the convention 0 n 00 = 0. any point with at least one coordinate equal to 00) and 0 . Jj c + R and D = { ( z j . . This partitions I into 2n nonoverlapping closed subintervals. we need to know that every gauge y has at least one yfine tagged partition associated to it. c associates open is a mapping Definition 4. .HenstockKurzw eil integral n 207 We define an interual I in (R*)n to be a product I = n I j . J j ): j = 1..00 = 0... j=1 k We assume.108 A gauge y on an interval I C defined on I that associates to each x E I an open interval J . k } is a tagged If f : I c partition of the interval I .. Proof. . .
L ( I ) is a closed and bounded interval and h' o y is a gauge on h' ( I ) . . k } . . and the Cauchy condition. j ) : j = 1. suppose that is a closed. x n ) = (h( X I ) .. Let J1 be one of the subintervals without a yfine tagged partition. by = t I ' if x = oo 2 arctanx if . undounded interval. and we write A = fI f. This contradiction shows that I has a yfine tagged partition. Consequently. It then follows that i j ( D ) = J { ( i j ( x j ) ij(J j ) ) : j = 1.Ji. .208 Theories of Integration had a yfine tagged partition. We call the function f HenstockKurzweil integrable on I if there is an A E R so that for all E > 0 there is a gauge y on I so that for every yfine tagged partition D of I . 0 We can now define the HenstockKurzweil integral for functions defined on intervals in Definition 4. Let { x } = n g l Ji. . . . and i j map closed intervals onto closed intervals and open intervals onto open intervals. Next. Since the diameters decrease to 0. . Then.5] h (z) h' ( X I . . k } is a yfine tagged partition of I . carry over to subintervals of as before. there is a io such that Jio c y ( x ) . Define {Ji}E1 by h(x) = nnn and h' : (R*)n +  [2. we do not repeat the statements or proofs. . each tag in a tagged partition in (R*)" can appear as the tag for up to 2n different subintervals in the partition.. which implies that D = { ( x . h' ( I ) has an h' o yfine tagged partition 2) = {(zj. .. . In R*.110 Let f : I c + R. a tag can be associated to one or two intervals in a tagged partition. The basic properties of the integral. Continuing this bisection procedure produces a decreasing sequence of subintervals of I such that the diameters of the Ji's approach 0 and no Ji has a yfine tagged partition. By the previous case.. The number A is called the HenstockKurzweil integral of f over I .)} is a yfine tagged partition of Jio. . such as linearity. then the union of these partitions would be a yfine tagged partition of I . . . h ( x n ) ) . o o < x < o o 7r x=oo if 2 7  r k Note that h' is onetoone and let i j be the inverse function of h'. positivity and additivity.
assume n = 2.6 . Fix E > 0 and choose 6 > 0 so that the sum of the areas of the four strips surrounding the boundary of I . if k cp (x)= a i x I . Further.6 . for x E R ~ set f (x) = . Since the series is convergent. b + 6 ) x ( ~ .Henst o c k .d]. ( a .6 ..111 Let I c R" be a bounded interval.J)ED. there is a B > 0 so that l a k l 5 B for all k E N. c k k ai XIi = aiv (Ii) * 2=1 i=l The following example generalizes Example 4. and y (x)n ? = 8 for x ?. Define a gauge y on R2 so that y (x)= I' for x E I o . J C SJ . Let I c R2 and I = [u. If V is a yfine tagged partition of R2. Example 4. We claim that f is HenstockKurzweil integrable over R2 and + + xp=.41 to higher dimensions. is less than E . it follows from Example 4. Example 4. Let E > 0. d + S ) . J ) E D . ~ + 6 and ( a . IC I) x ( k . For each k E N.b]x [c. b + 6) x ( d . k I) and. d + 6 ) .K urzw e il sin t e 209 We first show that the HenstockKurzweil integral of the characteristic function of a bounded interval equals its volume. = W" i=l c s. ( ~ . then / Lnc k cp = UiXI. (b6. Let S be the union of these four intervals. Set J k = ( k .Z€IO J c U ( Z .6. Then. Since the HenstockKurzweil integral is linear.6 .6 .the boundary of I . Without loss of generality.112 Suppose that x F = l a k is a convergent sequence.6 . JRn x I = v ( I ) .let o k be an open interval containing al .then 4 since I\ U(Z.111 that step functions are HenstockKurzweil integrable. a + 6) x ( ~ . d + S). (x)is a step function. a k X J k (2).b+6) x ( ~ . Pick a natural number M so that and l j < E for j 2 M . y (x)C_ S for x E 81.
Without loss of generality.001. For j = K . Note that v ( J j ) = 1 and.m } is a yfine tagged partition I of R2. so that a . .001 x [b. Then.00) and Im = [a. i ) : i = 1.m] if x = (00.. by the definition of y.I i ) E 'D : ti E J j } . ( M .00) o k . Set 'Dj = { ( t i . K 2 M . we may assume that t. b > M and f (tm) ( I m )= 0. such that v (01. ) < min \ Jk I . Suppose that 2) = {(ti. b}. . Let K be the w largest integer less than or equal to max {a. = (00. be an open interval disjoint from as follows: Uri1Jk U { (00. . we have Ii while for 1 5 j <K. U ( t i .210 Theories of Integration JI.001 x ( M .. 0 ) . 0) let Define a gauge y Y(x) = if X € J k IX if 12: $ u ~ =u~ X 00)) ((00. l i ) E ~ j c Oj.
We now prove this latter result.:. which implies a generalization of Corollary 4.~ 2 2'] ) x [a2 + ( k . hence.b l ] x [a2.1)(b2 . . I 00 I loo I IK1 k=l I +I c a*j l o o k=K+1 k=l I k=l I I <xG+E+€=3€.~ = 1. the Monotone Convergence Theorem.. Corollary 4. Consider first the case in which I = [ a l .87. Theorem 4..1 . Given the validity of the Dominated Convergence Theorem.a l ) 2'] + k (b2 . f is measurable. k . Without loss of generality.1) (bl  a l ) 2 .80 extends to R" and we have that absolute HenstockKurzweil integrability in R" is equivalent to Lebesgue integrability. 2 2'. a2 ) and let (1) (2) (3) I. It follows that f is HenstockKurzweil integrable over R2 with integral equal to c F = 1 a k Henstock's Lemma holds for functions defined on intervals in (It*)" and. .[ul + ( j .2'>. : j . once we know that every HenstockKurzweil integrable function on R" is measurable. c R" + R is HenstockKurzweil Proof. al + j (bl . Then. Fatou's Lemma and the Dominated Convergence Theorem are also valid for the HenstockKurzweil integral in (It*)".113 Suppose that f : I integrable.HenstockKurzweil integral 211 Therefore. we may assume n = 2. k Then.b2] is a bounded interval. Set (1) I j . .
We next show that {fi (z)}p"=. Let X = {x E AI : {fl (x)}.'S).:?)) E El(. Then. Since m dI!') = 0 (see the comment on page 83 Ilk) preceding Definition 3. .O.J l ( 2 K ) such that Jl &. < E. Fix e > 0. (4.16) Let X M = {x E A I : (fi(.converges to f (x)a.) (2). m. Further. where d J represents the boundary 2 of the interval J .e. we can choose a finite set of pairwise disjoint intervals Jl(zl).* ( x K . Thus.f(x)I > for all 2 (z). Let Jl(.tegration Define fl : I + R by Let dr. if {fl (x)}. for every yfine tagged partition 2)) D of I .) that contains x in its interior. . so that X = U. in I . by the definition of f m . 2 Thus.XM. ( z z ) .) be the interval I. by (4.e. it follows that m2 ( A I ) = m ( I " ) = m2 ( I ) .=. By the Vitali Covering Lemma. then {fl ( x ) } z .S (f.. in AI.enough to show that m ( X M ) 0 for all M E W to prove It is 2 = the claim.e. If x E X . .16) and Henstock's Lemma.does not converge to f (x)}."=. ( X M j i=l K +€ i=l 5 2M€ t ..43). converges to f (x)a. Let V M = {JlcZ): x E X M and Jl(. 2' = { (xlr J l ( z 1 ) ) 7 . there is a gauge y on I such that f .then there is a M E W and a sequence { l ( z ) }c N such that l f l ( 5 ) (x). . in A I .) c y(x)} and note that V M is a Vitali cover of X M ."=.212 Theories of In.converges to f(x) a. Since f is HenstockKurzweil integrable over I . I \ uEIJl(zi)) < c.f ( x ) 1 > for all Z(x)}. ( &. l ( z ~ ) ) J is a yfine partial tagged subpartition of I .
11 and [ S W ~ . and only if. We refer the reader to [Ma]. Let f k = fxI. f is measurable on I . converges pointwise to f on I .e.HenstockKurzweil kntegml 213 Since E: > 0 is arbitrary. Then: (1) fx is absolutely HenstockKurzweil integrable in R f o r almost every x E R. m2 ( X ) = 0 and { fi (x)}1"=. it follows that f k is measurable on I for all k.k] x [k.. (3) the following equality holds: s . this implies that the Fubini and Tonelli Theorems (Theorems 3. Then: j R be nonnegative (1) fz is measurable o n R for almost every x E R. and hence measurable on I k by the first part of the proof.109 and 3.28. By the ndimensional analog of Theorem 4. Since every step function is measurable and the pointwise (a.110) hold for absolutely HenstockKurzweil integrable functions in R". 0 Since a function is absolutely HenstockKurzweil integrable if.y ) dy is measurable o n R.) limit of measurable functions is measurable. converges to f (x)a. y ) d y is absolutely Henstock Theorem 4. Consequently. it follows that X M is measurable with measure 0.114 ( h b i n i ' s Theorem) Let f : R x R + R be absolutely HenstockKurzweil integrable. Theorem 4. (2) the function x I+ JR fz = f ( 2 . f is integrable over I . 8. Suppose I is an unbounded interval. Since I \ I k is a measurable set. we do not give them. since { f k } z . (2) the function x I+ fx = Kurzweil integrable over R. . It should be pointed out that there are versions of the Fubini Theorem for (HenstockKurzweil) conditionally integrable functions in R". and set I k = I n ([k.115 (Tonelli's Theorem) Let f : R x R and measurable. it is Lebesgue integrable. Thus. 6. in I . (3) the following equality holds: Sw swf ( x . [McL. but as the proofs are somewhat long and technical. it follows that f is measurable.e.131. k]).
2 R is increasing. that is. sI .2.b].6 to I construct intervals I0 = I I I1 II I2 II .1] defined by hi.3 Define positive functions 6 1 . Let yi be the gauge on [0. Exercise 4. such that l ( I k ) 5 l ( I k . 1 . Exercise 4.17: Assume that the theorem is false. Exercise 4. Fix c E (a. Exercise 4.214 Theories of Integration 4. (0) + Suppose that f : [a. Show that any 7.b] c R +R. Suppose that f (z) = 0 except for 5 E C.1  Let f (z) = 1 1 Find 5.1] and 62 (0) = 62 (1) = and 6 2 ( t ) = t for 0 < t < 1.t 6i ( t ) ) . Exercise 4. g is nonnegative and HenstockKurzweil integrable.1] + (0.6 Let 1 = [a. for i = 1.9 Suppose that f.6i ( t ).5 y1 ( t )c y2 ( t )for all Suppose that y 1 and y2 are gauges on an interval I such that t E I . show that f is HenstockKurzweil integrable over I and f = 0.13 Exercises Denjoy and Perron integrals Exercise 4. 1 + Exercise 4. for i = 1. a) 61 ( t )= for all t E [0.b) and set I1 = [a.g = 0..4 also y2fine. of and Qf(0). Show that D = D1 U D2 is a yfine tagged partition of I . Use bisection and Exercise 4. Show that y is a gauge on I such that any yfine tagged partition of I is also yifine. for i = 1. If s. Suppose that Di is a yfine tagged partition of Ii.b] + R and let C = { ~ iC [a. & : [0.Give examples of yifine tagged partitions of [O.b]and let y be a gauge on I ..fine tagged partition of I is Suppose that y1 and yz are gauges on an interval I and set y ( t ) = y1( t )n y2 (t).c] and I2 = [c. Note that f prove that f is HenstockKurzweil integrable and may take on a different value at each ci E C.b]. f = 0.g : I = [a.b] Df (z) 2 Qf(z) 2 0 for all z E [a.2.8 Use the following outline to give an alternate proof of Theorem 4. Prove that A General Fundamental Theorem of Calculus by Exercise 4. yi ( t ) = (t .b] be~a countable } ~ ~ set.7 Let f : [a. and If(t)l 5 g ( t ) for all t E I .2. Use the fact that nr?lIk {z} = to obtain a contradiction. Using only the definition. Exercise 4.l ) / 2 and no yfine tagged partition of I k exists.
16 Give an example which shows the importance of the continuity assumption in the Generalized Fundamental Theorem of Calculus. then t = zo.10 Let f : I + R. b] + R is HenstockKurzweil integrable over [a. s : Exercise 4.14 Suppose that j. (1) (Translation) Let h E R. there are HenstockKurzweil integrable functions g and h satisfying g 5 f 5 h and A . b + h] + R by fh ( t ) = f (t .E < g 5 h < A + E . prove that . Show that g is HenstockKurzweil integrable with g = f. + + s . s . Prove that f is HenstockKurzweil integrable with f = A . Basic properties Exercise 4. .13 Let f l g : I + R. Exercise 4.h).18 Prove that Definitions 4.9 and 4.b]. Prove that f is HenstockKurzweil integrable over I if. Exercise 4. Suppose that f is HenstockKurzweil integrable over I and g is equal to f except at countably many points in I .~ . Show that g is HenstockKurzweil integrable over [a/c. Unbounded intervals Exercise 4. Exercise 4.b/c] + R by g ( t ) = f (ct).b] and (t. sI Exercise 4.11 Let a 5 zo 5 b. Define fh : [ a + h . (2) (Dilation) Let c > 0 and define g : [a/c.g ) = 0. b h] with:: : J fh = f.34 of a gauge are equivalent.b]such that if D is a yfine tagged partition of [a. sI sI sI Exercise 4.HenstockKurzweil integral 215 Exercise 4.15 This example studies the relationships between the HenstockKurzweil integral and translations or dilations. that is.29. fl SI f = 0. Theorem 4. Show that there is a gauge y on [a.24. Generalize this result to a finite number of points { z .J ) E V with zo E J. That is.b/c] with c Jab/: g ( t )d t = f (t) d t .12 Let f : I + R. show that f is HenstockKurzweil integrable over I and s .34.17 Complete the induction proof of Theorem 4.Suppose there is a real number A such that for every E > 0. s . s. zn}.. 1 f . . Assume that f : [a. Show that fh is HenstockKurzweil integrable over b [a h. given a gauge y satisfying the Definition 4. zo must be the tag for J . and only if. If 1 is HenstockKurzweil integrable over I and If1 = 0. g is HenstockKurzweil integrable over I and f = 9.
25 Suppose that f : [a. cr=. show that g is HenstockKurzweil integrable over I R with JR f = JR g.24 show that Using the notation of Henstock’s Lemma (Lemma 4. and only if.22 Suppose that a k 2 0 for all k and ak = 00. and only if. V ). Show that f is HenstockKurzweil inteExercise 4. g : R +R.b].b]if.D).20 Let f : R +R. cT=. If f is HenstockKurzweil integrable over R and g = f a.b] and HenstockKurzweil integrable over [c.b] such that IS ( f . then IS (f. see Example 4.19 Suppose f .e.b] + R is bounded on [a. Show that if f is HenstockKurzweil integrable over [l. Exercise 4.=1 c R and set f (z) = ZE.47 to show that the product of HenstockKurzweil integrable functions need not be HenstockKurzweil integrable. 1 f Exercise 4.c . Exercise 4.26 Use Example 4.43).A < E for every yfine tagged partition D of [a. ~ ) . Prove that the function f defined by f (z) = a k X [ k . Exercise 4. there is an A E R such that for every E > 0 there exist r > 0 and a gauge y on R such that if a 5 r and b 2 r .E] and [ c E. Show that f is HenstockKurzweil integrable over R if.27 Recall that a function f has a Cauchy principal value integral over [u. k + l ) (x).23 Suppose {ak}. oo).216 Theories of Integ*ration there is a gauge y‘ satisfying Definition 4.b]. For the converse. R. for some a < c < b. Show that f is HenstockKurzweil integrable over [a. a < b.then the series 00 ak converges. and the limit + .b] for every a < c 5 b. and a gauge y on [a. f is HenstockKurzweil integrable over [a. xk=l Henstock’s Lemma Exercise 4.b].41. Exercise 4.21 Let f : R grable over R if. there is an A E R such that for every E > 0 there exist a . k + l ) (2)is not HenstockKurzweil integrable over [1 . b] for every (sufficiently small) E > 0. This implies that every 7’fine tagged partition is also a yfine tagged partition. Exercise 4. Exercise 4.9 so that y’(t) c y(t). a k x [ k . b E R.A < E 1 for every yfine tagged partition V of [a..
” dt tP exists for p > 0. Give an example of a function f whose principal value integral over [a.b.31 Let f.b]. + Exercise 4.b]. exists in the Exercise 4. (2) If 0 < L < 00. differentiable and decreasing.00). (3) If L = 00 and f is HenstockKurzweil integrable over [a.50.b] if.28 Suppose that f : [00. 001.] Exercise 4. sin (z2) d z .b].] saw sin t Exercise 4. [Hint: integrate by parts. Is the integral absolutely convergent? [Hint: try the substitution t = x2. Prove that f g is HenstockKurzweil integrable over [a.b].29 Let f : [a. If either (a) limtm g ( t ) = 0 or (b) Jam f exists. Set F ( t ) = s. Exercise 4.] Exercise 4.then g is HenstockKurzweil integrable over [a. f = s_” f s. [Hint: integrate by parts.F ( t )exists. It may help to review Example 4.then f is HenstockKurzweil integrable over [a.b]. and only if.00) t R be differentiable.32 Prove the following limit form of the Comparison Test: Suppose that f . g : [a.b] . . ~for every a 5 c < b. Show that the integral is conditionally convergent for 0 < p 5 1.34 Use Abel’s Test in Exercise 4.33 to show that s.30 Show that the Fresnel integral. Assume that F ( t ) = s. 00) t R. Suppose that f is continuous on [a.” f for every choice of .HenstockKurzweil integral 217 exists and is finite. Exercise 4. g : [a.b]. then f g exists.” f for a 5 t < b and assume that limt.b] t R are HenstockKurzweil integrable over for all Assume and and for all (1) If L = 0 and g is HenstockKurzweil integrable over [a. 001 + R is HenstockKurzweil integrable over [00. Give necessary and sufficient conditions for f’ to be HenstockKurzweil integrable over [a.” f is bounded and assume that g is nonnegative.g : I +R. Prove s”.00).b]. then g is HenstockKurzweil integrable over [a. a E R. f is HenstockKurzweil integrable over [a. HenstockKurzweil sense.b] exists but such that f is not HenstockKurzweil integrable over [a. Suppose that f g and f are HenstockKurzweil integrable over [ a .33 (Abel’s Test) Prove the following result: Let f . g is differentiable and ] 9’ is absolutely integrable over [a.
..8 E R.40 on [a. + Exercise 4.44 Suppose that f : I let c > 0.b] +R are continuous on ( a .60 for I = R. Define fc by R is absolutely integrable over I and Show that fc is absolutely integrable over I . Show that f g is HenstockKurzweil integrable over [a.43 Prove Theorem 4. t s sint dt logt exists. a ] )exists and is finite. Prove that s. lime. (2) Give an example of a function cp E BV ( [ a b ] )for all [a. cp is constant Exercise 4. b ] ) .00) + R with limt. . g (t) = 0 and that g' is nonpositive and continuous on [a. Exercise 4. (3) Prove that if cp. Exercise 4.00). a p P.218 Theories of Integration Exercise 4.b]. Assume F (t) = 1 f is bounded.62 to I = R.42 Exercise 4. Prove that V a r (cp..p E R. . (1) Prove that cp E DV (R) implies cp E BU ( [ a b ] ) for all [a.b]). .37 Suppose that f . g : [a.R) = lima+mV a r (cp.b] c R such that . Extend Corollaries 4.00). and only if. then a p + . Exercise 4. cp 4 W R > .b]) and c + + E BV ( [ a b ] ) and a .35 to show that s .41 We say a function cp E BU (R) if cp E DV ([a.b] if.39 Suppose that cp E B V ( [ a .b] and g' is absolutely integrable over [a.f3+ E BV (R). a ] ) for all a > 0 and V a r (9.b) c R. c + Exercise 4. Assume g : [a.IcI E BV ( [ a . Is the converse true? Either prove or give a counterexample." f g exists.+ F (c) g (c) exists. [a.b].00) +R is continuous and F ( t ) = f is bounded on [a. .Prove that 11 E BU ([a.b]) = 0 if. [a. Absolute integrability Exercise 4.36 Use Exercise 4.35 Suppose f : [a. Prove that 9 Exercise 4. and only if. E DV (R) and a .38 Suppose that cp.61 and 4.
h : I +R are HenstockKurzweil integrable.f l + 0. p : I + R are HenstockKurzweil integrable over I and fk 5 . In particular. fk is finite a.48 Suppose that f k .51 Let f : R + R and suppose A c B and both sets are measurable. f is HenstockKurzweil integrable over I and g is HenstockKurzweil integrable over every bounded subinterval of I .L? for all k and limsupk f r fk > m. Show that the conclusion of the Dominated Convergence Theorem can be improved to include fI lfk .HenstockKurzweil integral 219 Convergence theorems Exercise 4.47 Suppose that f. If If . Show that limsup.e. {fk}El Exercise 4. + Exercise 4.46 Let fk : I IR be HenstockKurzweil integrable over I . [Hint: Use the Monotone Convergence Theorem.] . Prove that supk fh is HenstockKurzweil integrable over I .50 Suppose that fk : I + IR is HenstockKurzweil integrable over I and converges to f pointwise.f A k is HenstockKurzweil integrable over I . prove that f is conditionally integrable. for every k E N. there are HenstockKurzweil integrable functions hl and h2 satisfying hl 5 fk 5 hz for all k E N. Show that there is a HenstockKurzweil integrable function g : I + R such that lfk . j E N if.49 (Dual to Fatou’s Lemma) Suppose that f k . Show that the result fails if we replace “absolutely integrable” with “HenstockKurzweil integrable”. and k k Exercise 4. Exercise 4. Exercise 4. and only if. o o ) . Exercise 4. Show that if f is absolutely integrable over B then f is absolutely integrable over A . Suppose there exists a HenstockKurzweil integrable function g : I + IR such that lfkl 5 g for all k N.h( 5 g and h is conditionally integrable. p : I + R are HenstockKurzweil integrable over I and fk 5 p for all k . Exercise 4.9. Show that f A g is HenstockKurzweil integrable over I . [ O .45 State and prove a uniform convergence theorem for the HenstockKurzweil integral.fj I 5 g for all k .g : I .52 Suppose f.
is a Vitali cover of [0.k i f f (t) < k (t)l 5 k k if f ( t )> k ‘ Show that f ’” is absolutely integrable over I .g E 7 K( I ) and X E R. and only if. define f k . xEf is HenstockKurzweil integrable over I for all measurable E c I. Prove that l l f l l 5 shows that the imbedding L1( I ) c) ‘FIX ( I ) is continuous. Q T ] .)Idefined in Definition 4. Exercise 4.1]n Q and n E N} :x Exercise 4. For k E N.r] x [r2.k ) V 9. 3.Q T . prove that 91 I l l f l ) 1 1g1 and )lAfll I 1x1 l l f l l 11 for all f. the truncation of f at height k . show that V = {x Fr : ))xllm5 1 and 0 < r 5 1) is a Vitali cover of E .b] and define = sup Ilflll. show that V = {x Er : ))z)). by f ( t )if If . (2) Fix Q > 0.1] and set (1) If Er [r.1] ´ [0.60 Let I = [a.56 Show that the set of intervals with rational endpoints is a Vitali cover of R. + + The space of HenstockKurzweil integrable functions Exercise 4.53 Let f : I + R be absolutely integrable over I . I].1 HenstockKurzweil and Lebesgue integrals Exercise 4. Exercise 4. That is. Characterizations of indefinite integrals Exercise 4.59 Let I c Iw and f E L1 ( I ) . This Ilfll‘ Prove that { IJ’ )lf)\’ by f /: J c I is a closed subinterval J llfll’ is a seminorm and l l f l l 5 Ilfll’ 5 2 Ilfll.58 Show that the function /).54 Let f : I +R. + i] E [0.55 Show that V = { [x .57 Let E = [0.220 Theories of Integration Exercise 4.  [If + + Exercise 4.105 is a seminorm. .$. [Hint: consider g = f A k and h = ( . Prove that f is absolutely integrable over I if.r] x [ . 5 1 and 0 < r 5 1) is a Vitali cover of E .72]. If Fr = [r.
Show there is a x2) gauge y on I such that if 2) is a yfine tagged partition of I . Write the multiple integral in Example 4. defined by Ilfll’ = sup { IJ’ J fl :J cI is a closed subinterval defines a seminorm on 7% ( I ) such that Ilfll‘ 5 Ilfll1 for all f E L1 ( I ) . x must be the tag for any subinterval from 2) that contains x.K u r z w e i l integral 22 1 Exercise 4. (x. Exercise 4.63 integral. Prove that llfll’. Let ‘FIX ( I ) be the vector space of HenstockKurzweil integrable functions on I .1] x [0.1] and x = ( X I .J ) E 2 and 7 x E J .112 as an iterated .62 Let I = [0.H e n s t o c k . unbounded interval. In other words. HenstockKurzweil integrals on Rn Exercise 4. then x = x.61 Let I c R* be a closed. E I .
This page intentionally left blank .
Suppose we drop the requirement that if ( t . We will use the word “free” to denote that the tag need not be an element of its associated interval.J ) E D. This is exactly what E. Further. However.J ) E D} be a partition of I and that J c y ( t ) but now require only that t E I . Thus.then t E J . J. every yfine tagged partition of I will satisfy this new definition.Chapter 5 Absolute integrability and the McShane integral Imagine the following change in the definition of the HenstockKurzweil integral. but so might some other sets D. there are HenstockKurzweil integrable functions which are not McShane integrable. every McShane integrable function is also HenstockKurzweil integrable. 223 . McShane (19041989) did (see [McSl] and [McS2]) and we next study the integral that bears his name. we have seen that absolutely HenstockKurzweil integrable functions are Lebesgue integrable and we will conclude this chapter by proving the equivalence of Lebesgue and McShane integrability. This is a consequence of the fact that the McShane integral is an absolute integral. Thus. Not surprisingly. in other words.Thus. every McShane integrable function is absolutely integrable. Clearly. which is a conditional integral. we still require that { J : ( t . any HenstockKurzweil integral proof that does not rely on any geometric constructions will carry over to prove a corresponding McShane integral result. Let y be a gauge on an interval I and D be a yfine tagged partition of I . suppose we allow the tag to lie outside of J . the McShane integral is based on yfine free tagged partitions. This result is in sharp contrast to the HenstockKurzweil integral.
. i ) : i = 1. .1 Definitions Let I c R* be a closed interval (possibly unbounded) and let f : I + R.. the intervals have disjoint interiors.2 Let 2) = { ( t i . For a tagged partition. m } is a yfine free tagged partition of [0. . Ii) : i = 1. in fact. The point ti is called the tug associated to the interval Ii. Definition 5. )= 0 is a good approximation of the expected McShane integral D off. A free tagged partition is a finite set of ordered pairs 2) = { ( t i . 2) = ((7.1]..m } such that Ii is a closed subinterval of I .1] + R in Example 4.. We denote this by writing D is a yfine free tagged partition of I . Note that S (f. The Riemann sum of a function f : I defined to be + R and a free tagged partition V is Definition 5. Then.1 Let I c R* be a closed interval.10 with c = 1. .3 Consider the gauge defined for the Dirichlet function f : [0. . If r is an irrational number in [0... . there are yfine free tagged partitions associated to every gauge y and interval I . Example 5. and ti E I . .1]. Let { I i } E l be a partition of [0. . the requirement that the tag lie in the associated interval meant that a number could be a tag for at most two intervals.224 Theories of Integration 5.1] c y(7). Since every tagged partition is a free tagged partition.0 0 ) = 0. We shall always assume that f is extended to all of R* by defining it to be 0 off of I and that f (00) = f ( .4 Let f : I c R* + R. We saw in Theorem 4. then there is a yfine tagged partition V of I . a single number could be a tag for every interval. uEIIi = I .17 that if y is a gauge on an interval I . We say that V is ?. This is no longer the case in a free tagged partition. Definition 5. then [0.m } be a free tagged partition I of I and y be a gauge on I .Ii) : i = 1.1].fine if Ii c y(ti) for all i. We call the function f McShane integrable over I if there is an A E R so that for all E > 0 there is a gauge y on I so that for every yfine free tagged partition 2) of I .
and we write A= JJ.00) and cy=l To prove this result..41. there is a B > 0 so that l a k l 5 B for all k E N. b + $ ) . . ~ ( b = ( b . In fact.m } is a yfine free tagged partition of [ l . we use an argument analogous to that in Example 4.t 1) min (&. K 2 M . k l).2. a < b.5 Suppose that al. which we repeat here to allow the reader to more easily identify the differences. set y ( t ) = (u. 1 ) ) . Since we are guaranteed that yfine free tagged partitions exist. we consider the analog of Example 4. let y ( t ) = ( k . b ) . sI Example 5. See Exercise 5. Let E > 0. Then. in this case.b]. a k must be absolutely convergent.: . f is McShane integrable over [l. We leave it to the reader to complete the details. Suppose that V = { ( t i . and.min (A. See the comments following the example for a discussion of the difference between the two examples. k + 1). Note that. let y ( t ) = (t .a)l < E. is an absolutely convergent series and set f (x)= C.( b .. l? Let K be the largest integer less than or equal to b. Pick a natural number A so that 4 lakl < E. In the next example. if I has endpoints a and b. since every tagged partition is a free tagged partition. Several observations are immediate or follow from corresponding results for the HenstockKurzweil integral. Without loss of generality. For t E (k. We will use the symbol f to represent the McShane integral in this chapter.001.41.b]. this definition makes sense.D ).i . &.. k + l ) (2) for x 2 1. the series c r = . . for t = k . Since the series is absolutely convergent..I i ) : i = 1. Then. 27. ? ( a ) = ( u . every McShane integrable function is HenstockKurzweil integrable and the integrals agree.McShane integral 225 The number A is called the McShane integral of f over I .IS ( f . so that b > A4 and f (tm) ( I m )= 0. we may assume that t m = 00 and Im = [b. ~ ] . Using the proof of Theorem 4. It is not hard to prove that the characteristic function of a bounded interval I is McShane integrable with JRxI = [ ( I ) .18. a + : ) .let ~ ( ) ! be an interval disjoint with [a.b) for t E ( a . for every yfine free tagged partition. First.001. Then. and E > 0. and for t $ [u. + + cr=M . a k X [ k . let y ( 0 0 ) = ( M . Define a gauge y as in Example 4. one sees that the McShane integral of a function is unique.41 for the McShane integral.
I i ) E D k I is . Not all natural numbers less than or equal to b need to be tags. i a finite union of subintervals of ( k . ~ + + Set Vk = { ( t i I i ) E 2) : ti E ( k .Vk). and + + Thus. then an interval to the left of I . one could have 2)k = 8. For k E N. If k E N is a tag.a k l for M 5 k 5 K .k l). U ( t i . k 1)). If e k is the sum of the lengths of these subintervals.  UKI + + I . U {Ii : (ti. I i ) E D k I i which is a finite union of subin. and 2)k. in which case IS (f. Let V N = { ( t i .tegration 1. 1 5 k < M .226 Theories of In. (2. then l?k 2 1 . For 1 5 k < M .k 1) c y (00). (k. Thus.DO. are the same as before. and b E y ( k ) implies k 5 K 1.& ' B . tervals of ( k . for k 2 M .& .B+ m) 2'"1' < Note that the arguments for V. k 1). U ( t i . as was the case for the HenstockKurzweil integral. One the same estimate obtained for IS (f. Thus. we have S e E E ) in Example 4.could be a set of intervals the sum of whose lengths is small. then k 5 K if b E y ( k ) . To estimate I (f. In fact. could be tagged by k . I i ) E V : t i E N}.41.2 ) ~ cannot obtain a better estimate for these terms since. because an integer k between M and b is an element of y ( 0 0 ) .I i ) E 2 ) and ti = k } c y ( k ) .)ak = J a kI.
D k ) . then the tag associated to I could be 00. that is. This is related to the fact that the McShane integral is an absolute integral. the point at 00 may be a tag for more than one interval. then the series is ak is absolutely convergent.00). this term is not a summand in (f.) .fzkl for h 5 k 5 K . iff (z) = a k X [ k .41 and 5. XE. so if f is McShane integrable then so is See Theorem 5. Combining all these estimates. for the McShane integral.McShane integral 227 Finally. For the McShane integral. further. the series must converge absolutely. In that case f (00)l ( I ) = 0 and. Hence.D. 1 s s xfzM If]. k + l). l i ) E D k l ( I i ) could be much less than one. with arbitrarily many terms.4. S(f. As for the HenstockKurzweil integral. we list some of the fundamental properties satisfied by the McShane integral. When estimating ( f . See Exercise 5. Since f ( o o ) = 0. 5. the converse of this example holds. one needs to address the fact that if f (t. For the so HenstockKurzweil integral. it can happen for arbitrarily many intervals. This leads to the sum ] a k l in the estimate above.5 provide an illustrative comparison between the HenstockKurzweil and McShane integrals.2 Basic properties In this section.11 below. we have I 0 0 1 = 0. XE.I ) E V and I c ( k . k + l ) (2) McShane integrable. <  k=l 00 k=M 00 k=K+1 k=l k=M It follows that f is McShane integrable over [l. let D = {(00. Examples 4.Dk)) that & t i . this situation could arise for a t most one interval. that is.Ii)E D}. .
b] i i and only if.19 and 4.27 for a proof of this result. ( I ) (Linearity) I a.. g : I c R* 4 R be McShane integrable over I .7 A function f : I + R is McShane integrable over I if.8 Let I be a closed. The next result now follows. f is McShane integrable over [a.28 and 4.6 Let f . Theorem 5. one can prove that if f is McShane integrable over an interval I and J is a closed subinterval of I . one has Proposition 5. McShane integrability is characterized by a Cauchy criterion. See Theorems 4. Using the Cauchy condition. JII = p +[f. Similar to the Riemann and HenstockKurzweil integrals. . Iff : I + R is continuous over I .20 for proofs of these results. then f is McShane integrable over J . then QIf f . + Pg is McShane integrable and (2) (Positivity) If f < g on I. Corollary 5. Note that by induction.B E R. A free tagged subpartition of an interval I c R* is k a finite set of ordered pairs S = { ( t i .b].5.c] and [c.9 Let cm 5 a < c < b 5 00. Using the fact that continuous functions on closed and bounded intervals are uniformly continuous there. and only if.b]. .43).. One of the key results for the HenstockKurzweil integral is Henstock’s Lemma (Lemma 4. Ic} such that { Ji}i=lis a subpartition of I and ti E I . Further.29 for details of the proof. then See Theorem 4. bounded subinterval ofR.J i ) : i = 1. We say that a free tagged subpartition is yfine if Ii c y (ti)for all i. Then. f is McShane integrable over I = [a. then See Propositions 4. . See Exercise 5.228 Theories of fnteymtion Proposition 5. then f is McShane integrable over I . the result extends to finite partitions of [u. f o r every E: > 0 there is a gauge y so that i f D1 and D are two 2 yfine free tagged partitions of I .
.fine.12 Let V = { ( t i .j = 1.3 A b s o l u t e integrability The previous section documented the similarity between the HenstockKurzweil and McShane integrals. Then.. . . ..n.fine.10 (Henstock's Lemma) Let f : I c R*+ R be McShane integrable over I ...m} be a free tagged parI tition of a n interval I and let 3 = ( J j : j = 1. each Kij E . J k ) } is a y. . . i f y is a D) gauge o n I and D is ?.. we will use a couple of preliminary results. Consequently.51). We will prove that every McShane integrable function is absolutely integrable Theorem 5. For E > 0 .Fi. n . m and F = U2.fine free tagged subpartition of I .I:n J: # S} is a free tagged partition of I and S (f. Since the intersection of two closed intervals (in R*) is a closed interval. I f n J: # S} for i = 1. . . .Ii n J j ) : i = 1.F is a closed interval. Then The proof is the same as before..m. i ) : i = 1.. .McShane integral 229 Lemma 5. . P r o p o s i t i o n 5. = S (f.. then D' is ?. 2)'= {(ti.Further. is McShane integrable over I and If 1 To prove this theorem. Then.D'). . Let Fi = {Kij = Ii n Jj : j = l .. Proof. . . . then Suppose V' = {(XI.11 Let f : I + R be McShane integrable over I . . ( x k . . 5. n ) be a partition of I ... let y be a gauge such that i f 2) is a yfine free tagged partition of I .. We now turn our attention to their fundamental difference.
. J j ) : j = 1. 1111.) .2/31 . . . .) n (J.n. Thus. [2/3.. set tij = ti and sij = s j . .F= {Kij = Ii n Jj : i = 1 . .f ( s j ) l . The intervals are nonoverlapping since K.so that 0 Notice that this result fails for tagged partitions.) n (I. j = 1 . [o. then 2)' = { (c. which is empty unless i = ' and j = j'. and 3 = {[o. . then (ti. .13 Let f : I + R be McShane integrable over I . D ' ) .f ( s i j ) = If ( t i ). by definition. [0. if f (ti) < f ( s j ) . ) V' implies Kij c Ii c y ( t i ) .) = (I: n I. if D is yfine. D = { ( c . ( c . if c E I = [0. (c. it is the first result we see that distinguishes the McShane integral from the HenstockKurzweil integral. then sI I m n Proof.n) are yfine free tagged partitions of I . partitions that are not free. .I i ) : i = I. note that [ ( I . [2/3. The proof of the following lemma makes crucial use of free tagged partitions. Define tags t i j and S i j as follows. Thus. K i j ): Kij E F} and . set t i j = S j and sij = ti. Lemma 5.1])} is a free tagged partition (for any choice of c).2/3]) . .111. n J. ~ = (I: n J.f < 6 .K i j ) E 2' is a yfine free tagged partition. Let E > 0 and suppose y is a gauge on I such that IS (f. D ) = S ( f . TO see that S ( f . 2' is a free i ) tagged partition of I .If n Jj" # S}. that is.D ).. m i=l m i=l m Finally. n J. for every yfine free tagged partition V of I . f ( t i j ) . ) = C K i j E F i l ( K i Thus. In fact. . Set . but it cannot be a tagged partition because c can be an element of at most two of the intervals. 1/31 [1/3.. If f ( t i ) 2 f ( s j ) . Since ti E I for all i ..230 Theories of Integration decomposes I into a finite set of closed intervals. m . .. n K . m ] and & = { ( s j . .I]. If V = { ( t i . Let V' = { ( t i j . [1/3.1/3]) .
is one example of such a function. (See also Examples 4. ) In the proof above.13.11. 4.42 and 4. { ( s j . n } be yfine free tagged partitions of I . Let D = { ( t i I i ) : i = 1.50.. Let E E > 0 and choose a gauge y on I such that ( S(f. it is easy to find examples of HenstockKurzweil integrable functions that are not McShane integrable.McShane integral 231 By Proposition 5. . D') s (f. The proof does not work if they cannot be free. 0 which completes the proof. Proof. &')I . d and e are yfine free tagged partitions if I.) Since the McShane integral is an absolute integral. IS(f. 1 I m n I m n i=l j = 1 The integral inequality follows from part (2) of Proposition 5.12.41. + Proposition 5.. It is enough to show that If1 satisfies the Cauchy condition.11. we make use of the fact that 2' and I' are free tagged partitions.J j ) : j = 1. . which was introduced in Example 2.. . 0 Due to Theorem 5. one merely needs a conditionally (HenstockKurzweil) integrable function. .31.g : I + R be McShane integrable ouer I .S(f.&')I I On the other hand. I = IS (f. We can now prove Theorem 5. . The function f : [0. By Lemma 5. f V g and f A g are McShane integrable ouer I .&')< 2€.6.JI f < . m } and & = . 5 1.14 Let f.S(f. m n f .1] + R defined by f (0) = 0 and f (2) = 2x cos 5 sin 5 for 0 < x. D).. Then.D') .for every 2 yfine free tagged partition D. so by assumption. it satisfies stronger lattice properties than the HenstockKurzweil integral.
We have the following version of Part I of the Fundamental Theorem of Calculus for the McShane integral.3.b] and continuous at x E [a. 0 As for the Riemann and Lebesgue integrals. The example above.1]and hence it is HenstockKurzweil integrable.232 Theories of Integration Proof.I f . the statement and proof of Theorem 4.b] + R is diflerentiable on [a. Since f ' is McShane integrable. Since f V g = [f g If . Proof. in which f (2) = 2xcos $ sin $ for 0 < x 5 1 and f (0) = 0. the assumption that f be ' McShane integrable is necessary for Part I of the Fundamental Theorem of Calculus.1 Fundamental Theorem of Calculus The beauty of the HenstockKurzweil integral is that it can integrate every derivative.b] and assume that f is McShane ' integrable over [a.b]. one needs to assume that both f and g are bounded above by a HenstockKurzweil integrable function.15 (Fundamental Theorem of Calculus: Part I) Suppose that f : [a. provides such an example.16. 0 Recall that for the HenstockKurzweil integral. . the indefinite integral o f f . The function f is a derivative on [0. not every derivative is McShane integrable. so it cannot be McShane integrable. By Theorem 4. Such a result cannot hold for the McShane integral.16 Let f : [a. F .911 and f A g = the result follows from linearity and Theorem 5. In other words. (See Proposition 4. [f' = XK Jllb f' = f (b) .65. +? Theorem 5.b]. Then. it is HenstockKurzweil integrable and the two integrals are equal. is differentiable at x and F' (x)= f (x). fr + + 3 [f + g . or bounded below by one.911.f (a). But it is not absolutely integrable. Concerning the differentiation of indefinite integrals.b] +R be McShane integrable on [a. Then. Theorem 5.) 5.11.82 yield the following result for the McShane integral.
J I . so that . the set . For each k .. . Note that by construction.9 < E . Theorem 5. Then. Define a step function cp : I + R by cp ( t )= f ( t i )xIi ( t ) . Also.n}. By linearity and Theorem 5. Assume that & = {(sI.b] + 118 is McShane integrable.83.17 (Fundamental Theorem of Calculus: Part 11) Suppose that f : [u. While the result follows from previously established relationships between the McShane. ( t )is a bounded interval for all t E I n R and IS ( f . Let V = { ( t i .fine. Set & = U&&i. . . . by Lemma F 5..such that JI.J I < ~ / for every 7.JI . Since Jk c Iik c y1 (ti. .. ) : k = 1. so that & is a yfine free tagged partition of I . .fine.. n y2. . 00 (or 00) must be a tag for any unbounded interval and f (00) = f ( . Set y = y.McShane integral 233 In fact.13 we have fl czl If (If If I k=l j=1 However ?! (Jj U n J k ) = 0 if j # k . ) : k = 1. HenstockKurzweil and Lebesgue integrals.cpI .cpI < ~ / for 3 every y2fine free tagged partition & of I .b] and F’ (2)= f (z).18 Let f : I +R be McShane integrable over I and There exists a step function g : I +R such that 1 f . the McShane integral satisfies the same version of Part I1 of the Fundamental Theorem of Calculus that is valid for the HenstockKurzweil integral. We conclude this section by showing that every McShane integrable function can be approximated by step functions in the appropriate norm.cpI is McShane integrable over I ..8.n } is ylfine.I i ) : i = 1.11. so there is a gauge y2 on I such that IS .. c Ii. m } be 7. I . cp is McShane integrable by Exercise 5. D ) . we use a more direct proof to establish the theorem. Choose a gauge y1of I such that y. Proof.fine free tagged partition 3 V of I . = {(ti. For each subinterval Ii (from D). Theorem 4.0 0 ) = 0.. &) . 1 sI E > 0. . there J is a unique i ~ . 1 5 k 5 n. &i be a yfine free tagged partition let of Ii.). Since & is also ?. F as daflerentiable at almost all x E [a. Theorem 5. .
€ is also yzfine.69.68. k=l Finally.PI 4 + < 5 €7 0 Convergence theorems Since every McShane integrable function is HenstockKurzweil integrable. Then. Set S n ( x ) = f k ( x ) . f k (x)pointwise o n I . However.PI < S(lf . JI Ifkl < (2) (3) . and 4.4.67.7.234 Theories of Integration Since sk E Iik implies that ‘ p ( s k )= f(tik). f (x) = C:=. Further. Thus. fk a McShane s CE. when considering convergence results for the McShane integral we will need to avoid the same problems that arise for the HenstockKurzweil integral.4 If . in this chapter we will present a slightly different one that highlights the importance of series of functions. Theorem 5. 5. our conditions must eliminate the pathologies demonstrated in Examples 4. and 00.19 Let f k . Suppose each integrable over I . f : I C R* + R. 4. which implies JI as we wished to show. We could easily follow the approach in Section 4. it will satisfy convergence theorems stronger than the ones satisfied by the HenstockKurzweil integral. (1) f is McShane integrable over I . since the McShane integral is an absolute integral.
M c S h a n e integral
23 5
Proof. esis, so
v = cTz1f k is finite. Choose K E N such that sI
Let
E
> 0.
The series c:=,
sI
fk
converges absolutely by hypoth
For each n, s is McShane integrable (since it is a finite sum of , McShane integrable functions) and there is a gauge y on I such that , IS (, D ) s < & for every 7,fine free tagged partition V of I . s , By modifying the proof of Example 5.5, the function ‘p : R + R defined by cp ( t ) = 2  n ~ ~ t : n  1 ~ l t l is n ~ < McShane integrable over R* and JR* ‘p = . Let y be a gauge such that IS (9,D) JR cpI < , for any 7,fine free tagged partition V of R. Then, 0 5 S(cp,D) 5 JR‘p =1 whenever 2) is a 7,fine free tagged partition of I . By the pointwise convergence of s to f , for each t E I , choose an , n ( t )E N such that n ( t )2 K and, for n 2 n ( t ) ,
sI
I
fr
cr=l
+4
Define a gauge y on I by setting y ( t )= Yn(t) (t)ny, ( t )for all t E I . Let
D = {(ti,I i ) : i = 1,.. . ,m } be a yfine free tagged partition of I . Then, by
the absolute convergence,
=I
+
I i=l k=n(t;)+l
11+ I I I .
‘p,
By (5.2) and the definition of
236
Theories of Integration
Next,
by (5.1). To estimate I , set S = max { n ( t l ), . , . ,n ( t m ) }2 K . By Henstock's Lemma,
D) Thus, (S(f,  Vl
< 3 ~ It follows that f is McShane integrable with .
s,f
=vFinally, since
for n proof.
2 K,
JI
f
=
limntooJI
Sn
and
JI
Is,  f l
+
0, completing the
0
As a corollary of this theorem, we prove a preliminary version of the Monotone Convergence Theorem.
Theorem 5.20 (Monotone Convergence Theorem) Let f k , f : I C R* + be McShane integrable over I and suppose that { f k } T = 1 increases monotonically to f on I . I f s u p , JI f k < 00, then f as McShane integrable over I and
Proof.
Set
and
for
Then,
McShane integral
237
C;=,g k = f,
+
f pointwise on I and
Thus,
{gk}r=1
satisfies Theorem 5.19 so that
We next pursue a more general form of the Monotone Convergence Theorem and then use this general version t o obtain Fatou’s Lemma and the Dominated Convergence Theorem for the McShane integral. We begin with a series of three preliminary results. The first two results are analogs of Theorem 4.40 and Lemma 4.71.
Theorem 5.21 Let E McShane integrable and
SRxE
c R.
Then, E is a null set zf, and only if)xE is = 0.
Proof. Suppose first that E is null and let E > 0. Let {Gj};, be a sequence of open intervals covering E and such that C,”=,(Gj) < 5 . Since the characteristic function of an interval is McShane integrable, by Proposition 5.14, s, = xG1V..VxG, is McShane integrable. Since {s,}:=~ increases monotonically, h = limns, exists. Since s, is a maximum of characteristic functions and E c UZlGj, we see that 0 5 h 5 1 and xE 5 h. Note that s, 5 Cj”=, which implies that xGj,
n
00
By the Monotone Convergence Theorem, h is McShane integrable and
&h
< 5.
Now, choose a gauge y so that if V is a yfine free tagged partition of R*, then IS (h, ) h < V l Then, for any yfine free tagged partition,
v,
sR
5.
Since E > 0 is arbitrary, xE is McShane integrable with JRxE= 0. To prove the necessity, we argue as in the proof Theorem 4.40.
0
238
Th,eories of Inte!wation
Using the fact that every McShane integrable function is HenstockKurzweil integrable, Lemma 4.71 yields the following result.
Lemma 5.22 Let f k : I C R* + [0,00) be McShane integrable over I and suppose that { f k (x)>El increases monotonically f o r each x E I and sup, JI f k < 00. Then, limk,, f k ( 2 ) exists and i s finite for almost every x E I.
Suppose that f is McShane integrable and g is equal to f almost everywhere. Then, E = {x : f (x)# g (%)} is a null set and hence JR x E = 0. Employing this fact and the Monotone Convergence Theorem allows us to prove the next lemma.
Lemma 5.23 Let f : I c R* + R be McShane integrable ouer I and suppose that g : I + R is such that g = f a.e. in I . Then, g is McShane integrable over I with
Proof. The function h = f  g equals 0 a.e. in I . By linearity, it suffices to show that h is McShane integrable and JI h = 0. Let E = ( t E I : h ( t )# 0). Fix K E Z, for n E W, set and
hn = (lhl A n> I ~ ( K , K + ~ I . X
Then, hn 5 nXEn(K,K+l]. By the argument in the proof of Theorem 5.21, hn is McShane integrable over I f l ( K ,K 1 with JIn(K,K+ll hn = 0. Since 1 { hn}T=1 increases to lhl pointwise, the Monotone Convergence Theorem implies that Jhlis McShane integrable over I n ( K ,K + 1 and JIn(K,K+ll Ihl = 1 0. It now follows that h is McShane integrable over I n ( K ,K 1 and 1 SI" ( KK +1] h = 0. (See Exercise 5.3.) Since h = CkEZ , hXm(K,K+1] on I, Theorem 5.19 shows that h is McShane integrable over I with h = 0. 0
+
+
sI
We now have the necessary tools to prove a more general form of the Monotone Convergence Theorem.
Theorem 5.24 (Monotone convergence Theorem) Let f k : I C R* + [O, 00) and suppose that { fk ( ) ; , increases monotonically f o r each x E I . x}= Suppose each f k as McShane integrable over I and sup, SI f k < 00. Then, limk,, fk (x) is finite for almost every x E I and the function f, defined by
limk4w
fk
(x)if the limit is ,finite
otherwise
McShane integral
239
is McShane integrable over I with
Proof. By Lemma 5.22, limk,,m f k ( t ) is finite a.e. in I . Let be the null set where the limit equals 00. Set f ( t ) = limk,,m f k ( t ) if t !$ E and f ( t )= 0 if t E E . Set gk = fkXI\E. Then, by Lemma 5.23, gk is McShane integrable over I with JI gk = JI f k and {gk}& increases to f pointwise (everywhere in I ) . Thus, by Theorem 5.20, f is McShane integrable over I and
Recall that the proofs of Fatou’s Lemma and the Dominated Convergence Theorem (Lemma 4.75 and Theorem 4.77) rely on the Monotone Convergence Theorem. Thus, those arguments imply corresponding versions for the McShane integral.
Lemma 5.25 (Fatou’s Lemma) Let f k : I C R* + [ O , c o ) be McShane integrable f o r all k , and suppose that liminfk,,, JI f k < 00. Then, liminfk+m f k is *finite almost everywhere a I and the function f defined n
by
lim infk
fk
(x) zf the limit is .finite otherwise
is McShane integrable over I with
Theorem 5.26 (Dominated Convergence Theorem) Let f k : I c R* +R be McShane integrable over I and suppose that { fk}y=1 converges pointwise almost everywhere o n I . De.fine f by
m={
limk,,
fk
(x)
af
0
the limit exists and is ,finite otherwise
Suppose that there is a McShane integrable function g : I + R such that If k (x)l 5 g (x).for all k E N and almost all x E I . Then, f as McShane integrable over I and
I + R is McShane integrable. Then. then the set function f : M I + R is countably additive and absolutely continuous with respect to Lebesgue measure. Our main result in this section is the following theorem. the indefinite integral of f is countably additive.77). If 5. Lemma 5. The proof is a consequence of three results: f is McShane integrable over every Lebesgue measurable subset of I . The comparison condition for the Dominated Convergence Theorem (I fk (z)1 5 g (2)) is the same as for the Lebesgue integral (Theorem 3. unlike the condition for the HenstockKurzweil integral (Theorem 4. Extensions of Fatou’s Lemma analogous to Corollaries 3.27 I f f .98 and 3. f is McShane integrable over every Lebesgue measurable subset E c I. that limJI .99 hold for the McShane integral.100). The absolute integrability is also the reason why the Dominated Convergence Theorem for the McShane integral includes a stronger conclusion. . than one obtains for the Henst ockKurz weil integral . while the HenstockKurzweil integral is a conditional integral.fkI = 0. This is because the Lebesgue and McShane integrals are absolute integrals. s As an immediate consequence. the indefinite integral of f is absolutely continuous. we show that f is McShane integrable over every measurable set in M I and that f is countably additive. then the set function f : M I + R is a measure o n M I . sE sI s Theorem 5. we see that when f is nonnegative.j k l = 0.240 Theories of Jntegration Moreover. We say that f is McShane integrable over a set E c I if xE f is McShane integrable over I and define f = xE f .28 If f : I 4R is nonnegative and McShane integrable. f is a Corollary 5.29 Suppose that f : I 4 R i s McShane integrable over I . measure on MI. and. l i m l If . If f is McShane integrable over I .5 The McShane integral as a set function Let f : I c R*+R be McShane integrable and let M I be the set of Lebesgue measurable subsets of I .
Then. . m . J j ) : j = 1.. n } are ?‘fine free tagged partitionsofE.. S (f.D)... Fix E > 0 and let y be a gauge such that (S( f . Note that D and E‘ use and &) ‘ the same subintervals but have different tags. D‘= { ( t i . For each k E N. .N } . j = 1. . ... I i n J j ) : i = 1.. n } are 7’fine free tagged partitions.. . ... we may assume that D‘ = {(ti.. . = S (f. ..& I ) . n} and I’ = { ( s j .m .f < E for every yfine free tagged partition D of I. Ii): i = 1. .. I i n J j ) : i = l. .. .DO’) S (f. Then. . Define a gauge y’ on I by: s’ 1 Suppose that D = {(ti.m} and & = { ( s j .. Let E be a Lebesgue measurable subset of I .D)= S (f.McShane integral 24 1 Proof. .choose an open set O k 3 E and a closed set Fk C E such that m (Ok \ F k ) < & . Relabelling to avoid the use of multiple subscripts... Kl) : I = 1.N } and E‘ = { ( s i . j = l .. ..Kl) : I = 1.
29. then A similar argument shows that R 3 I so that E. Let { E j } E l c M I be a collection of pairwise disjoint sets and let E = U Z l E j . then the set function is countably additive. 0 . Since E E M I . si and Consequently. by Lemma 5.30 sf :M I + Iw I f f : I + R is McShane integrable. set ok = (1 : t! E E . Since the sets { E j } g l are pairwise disjoint. f X E satisfies a Cauchy condition and is McShane integrable. 00 4 E . the result follows.242 Theories of Integration By Henstock's Lemma. c = . If I x E is McShane integrable. Therefore. 0 We show next that the indefinite integral of a McShane integrable function is countably additive. Since E was an arbitrary measurable subset of I . 5 2E. Lemma 5. k . we see that ". Thus.15 1f ( t i ) \ < k } . Proof. By the Dominated Convergence Theorem for the McShane integral. f X E j 5 If I x E . and 00 If 1 E so that ok. Next. f x E j+ f xE as k + 00 and ICf_. I which shows that the indefinite integral is countable additive.
When f is nonnegative..d j ] : j = 1. the indefinite integral of a McShane integrable function f is defined on M I and countably additive. Lemma 5. we may assume that for each j. . F ( t ) = f f. . the set function F i s absolutely continuous over I with respect to Lebesgue measure.18. .m } such that [ c j .. co < a < b < co. . By (d Henstock’s Lemma..b]. s’ . m } be such a partition and set M = max{l f ( t i ) [ i = 1. We have shown that the point function F : I 3 R is absolutely continuous. I m I 5 E + M6 = 2 E .101.bi]. . Let S D) V’ = { ( t i . There is a gauge y on I such that I (f..31 Let I = [a. . is ‘ absolutely continuous. we show the indefinite integral is absolutely continuous as a point function in the sense of Definition 4.. Proof. uzl EL. It is also true that the set function f f satisfies the definition of absolute continuity given in Remark 3.c j ) < 6 . 0 Thus.b i ] } and set E = {(ti.93. Then.”=.[ai.d j ] c [ai. This result is an easy consequence of Theorem 5.McShane integral 243 Thus. the indefinite integral of f . . By subdividing these intervals. : Suppose that { [ c j . . this implies that the indefinite integral defines a measure on M I . .32 Let f : I +Iw be McShane integrable over I and define F by F ( E ) = f for E E M I .d j ] c [ai. m } 1 and 6 = E / M . + c. F is absolutely continuous. Let sI I E > 0. and f : I + R be McShane integrable.[ c j .d j ] ) : j E ai}. For each i. . E is a yfine free partial tagged partition of I with CjEoi j . . We conclude by showing that the indefinite integral is absolutely continuous both as a point function and as a set function. Then.p } is a collection of nonoverlapping closed subintervals of I such that ( d j . set a = i { j : [ c j . if necessary.f < E for every yfine free tagged partition 2) of I . Theorem 5.. First.bi]): i = 1 . Then.c j ) < 6 . there is an i E (1.
For a proof of this result.1}. be the canonical representation of s and set ikf = max {lull . there is a step function s such that JI 1 f .18. We saw in Sections 3. Therefore. That the space of McShane integrable functions is complete is a consequence of the Dominated Convergence Theorem (Theorem 5.11 that the space of Lebesgue integrable functions is complete in the (semi) metric dl (see the RieszFischer Theorem.23 that l l f l l l = 0 if. Let ci=lakxA. 0 5. and a corresponding semimetric d l by setting d l (f. It follows from (the proof of) Lemma 5. .g(I1= JI . see Theorem 3.g E M1 ( I ) . in I .26). Identifying functions which are equal almost everywhere makes (1 a norm and d l a metric on M 1 (I). .e. . . Suppose that f is McShane integrable over I and fix E > 0. Then. From Theorem 5.33 (RieszFischer Theorem) Let I c R* be a n interval and let { fk}F=1 be a Cauchy sequence in (M1 ( I ).s ( 5 5 .=l converges to f in the metric dl.18. f = 0 a. Theorem 3. dl is not a metric on M1 ( I ) .116. Set 6 = & and suppose that E is a measurable subset of I with m ( E ) < 6. We now observe that the RieszFischer Theorem holds for the McShane integral.116) while the space of Riemann integrable functions and the space of HenstockKurzweil integrable functions are not complete. for all f. so that F is absolutely continuous with respect to Lebesgue measure.4. consequently.6 The space of McShane integrable functions Let I c R* be an interval and let M1 ( I ) be the space of all McShane integrable functions on I . the step functions are dense in M1 ( I ) . so that 11 [I1 is not a norm and. By Theorem 5. . We define a seminorm 11 /I1 on M1 ( I ) by l l f l l l = 1 f 1. d l ) .244 Theories of Fntegmtion Proof. Then. there is a n f E M 1 ( I ) such that { fk}.91.9 and 4. lajl . in the appropriate (semi) metrics. SI If Theorem 5.g ) = Ilf . and only if.3.
~ )that is.4. In Section 4. we will employ major and minor functions. all x E [a.1.McShane integral 245 5. and hence that the McShane and Lebesgue integrals are equivalent.7 McShane.4. : I + R* by If sI XI. for example. we can also view y as a gauge on [a. Let y be a gauge on I . the McShane integral of f will be denoted M f. x S (x)).b]. we may assume that the gauge y is defined by a positive function 6 : I + ( 0 . Define m y . Let I = [a. Let 7r. The function hrir. On the other hand. . is called a major function for f . In order to prove this result. Since we will be dealing with three integrals in this section. 7X. We summarize our results for my and M.b] be a finite interval and suppose f : I + R. variants of the ones defined in conjunction with the Perron integral in Section 4. Consequently. + .for . and It is clear that my (x)5 M. For a < x 5 b.8. in the following lemma. y (2) = (x .b]. HenstockKurzweil and Lebesgue integrals Suppose that f : I c R*+R is McShane integrable over I . By Exercise 4. (z) for all x E [a. The crux of the matter is to prove that absolute HenstockKurzweil integrability implies McShane integrability. there are HenstockKurzweil integrable functions that are not McShane integrable. XI. and f is absolutely (HenstockKurzweil) integrable over I . and L ) to identify the integral being used. I is McShane integrable so that both f and If I are HenstockKurzweil integrable over I .S (x). we saw that Lebesgue and absolute HenstockKurzweil integrability are equivalent.my is called a manor function for f . we will identify the type of integral by letters (M. ( [ a x])be the set of all yfine tagged partitions of [a.18. In this section.5.M. We extend these results to higher dimensions in Section 5. we prove that in the onedimensional case McShane integrability is equivalent to absolute HenstockKurzweil integrability.10.
fined b y 6 : I (1) + Suppose that f : I = [a. (01 +  M . (3) is proved.4. Taking the supremum over all D E 7ry ( [ a u ])shows that My (v) 2 My (u) .4. . If x  d(x) < u L x I o < x (‘2) I .m.m. ( [ a . Suppose that y i s a gauge o n I (defined by 6 ) such that f (4(v . + f (x)(v . then M .m. ( [ a u ] ) . (4) I f f 2 0 . ( u ) . Part (4) follows from the fact that the nonnegativity of f implies that if u < v then SO . so that .26. See Exercise 5. For (3). = D‘ U .(u) 2 f (4(v . we can find V.246 Theories of Integration Lemma 5. x . Then. then both M . Thus. which proves (1).4) E 7ry ( [ a V I ) . E 7ry ( [ uv]). then my (v) .b] + R and y is a gauge o n I (0. . (v) . and my are nonnegative and increasing functions o n I . + S ( x ) . fix c > 0 and a 5 u < v 5 b.fix u and v and let D {(x. (u> 2 for every Then. (5) Let f be HenstockKurzweil integrable over I and E > 0. D U . €‘ F 7ry ( [ a v]). E F .my ( u ) 5 M. is increasing.D‘E T .34 de. To prove (l).u ) . By definition. (3) My .so that E = D U F. Since it is clearly nonnegative.my is a nonnegative and increasing function o n I . (v) and M .[u. Proof.6 ( x ) < u 5 x 5 v < x S ( x ) . that M . The proof of (2) is similar. E 7r.m). u ]such that ) Fix .
. j . f is McShane integrable. ( t ) . for V . The result now follows from the definitions of M. Zk]. Example 5. . sI Therefore. There are gauges y1 and y2 on I so that if D is a yifine free tagged partition of I .35 Let f : I c R* + R.Then. and 3 I k = [ Z k . f is McShane integrable on I . . there are McShane integrable functions 91 and g2 such that g1 5 f 5 92 on I and M g2 5 M g1 i E .V ).2.for g2 ( t )= c”.36 Let f : I = [a.M giI < E for i = 1. we collect a few other results. Then. . we show that increasing functions are McShane integrable. hence.30 on HenstockKurzweil integration.a ) . Let D be a yfine free tagged partition of I . Suppose that.i . are yfine free tagged partitions of I then This implies that By the Cauchy criterion.. Then. if V1 and V. note that the hypothesis implies E ([a.b]into j equal subintervals by setting xk = a + 5 ( b . As a consequence of this lemma. Divide [a. Set 91 ( t ) = f (Xk1) X I .  sI sI Proof. xI. g1 and f (xk) k = 1.McShane integral 247 for every D E 7rT ( [ a . . ( 5 ) . then IS (gi. . 0 This result is an analog of Lemma 4. xi=. j . and my.b] + R be increasing. 0 Before considering the equivalence of McShane and absolute HenstockKurzweil integrability. ( t ) and 92 are step functions and. Set y (x) = y1 ( z )n y2 (2). . . for every E > 0 . 1 . E rlTY b]). ~ ]To prove ). the proofs are the same. for k = 0 . Lemma 5. Let E > 0 and choose corresponding McShane integrable functions g1 and g2...
h Proof. and extend my 1 and M . Then. By Lemma 5. Let S correspond to y.@). Extend f to [a. Hn and h = lim supn4.” f < E whenever D is a yfine tagged partition of [a. (t .b 1 by defining my ( t ) = my ( b ) and My ( t )= M. there is a realvalued b la (M ( a+ 1/ n M.hn.  . f is McShane integrable over I a and only if.. M Hn < 00 so that liminf. Fix c > 0 and choose a gauge y on I such that S ( f . by considering f + and f.. ( t ) )and hn ( t ) = n (my (t $) .M.M. By a linear change of variable (Exercise 5. and my are increasing so that Example 5. By Lemma 5. t o [a.. Set H = lim inf. ( b ) for 1 b < t 5 b+ 1. Thus.my ( t ) ) .D).36 implies Hn and h. it is enough to show the result when f is nonnegative and HenstockKurzweil integrable.( t [+ljr + i) M. f is absolutely HenstockKurzweil integrable over I . ( t ) )d t I n M L + l i n M y ) =Ad.34 (4).7Ks. sI sI Theorem 5. We have already observed that McShane integrability implies absolute HenstockKurzweil integrability.M g1 < E by choosing j sufficiently large.37 Let f : I = [a. observe that I I + + + i) 0IM =n [ Hn = M [ n (M. We are now ready to prove the equivalence of McShane and absolute HenstockKurzweil integrability..6). we can make M g2 . are nonnegative and McShane integrable. f is McShane integrable.248 Theories of Integmtion McShane integrable.35.. Define functions Hn and hn by Hn ( t ) = n ( M . For the converse.b]. lim inf.25). by Fatou’s Lemma (Lemma 5. M. Hn is finite almost everywhere and. Since e ( I k ) = ba j ’ Given E > 0.) ..b]+ R.b + 1 by setting f ( t )= 0 for b < t 5 b + 1.
Fatou’s Lemma applied to (p hn)+. and Consequently. If E l = { t ~ [ a . .lim sup hn ( t )= H ( t ).hn)+ ( t )= H ( t ). these new functions are McShane integrable with the same integral as before. Let E2 = { t E [a. Since this only changes the functions on a set of measure 0.there is a realvalued function F which is equal to H . (b). by Lemma 5.e.b] and suppose that n > Then. M ( g. by Lemma 5. h then E2 is null and F = 0 on Ez.b].23.e.34 (1) and (2) 7 &. Define f by  f (4= { f ( a : ) ifa: f E 0 ifzEE’ so that = f almost everywhere and We claim that 7 5 5 75 2 on [a.b]. h Fix t E [a.b].McShane integral 249 function H which is equal to H a. lim inf n+co (H . Note that the function = H . b ] : H ( t ) # l ? ( t )t } .e n & i s n u l l a n d f S = O o n E l . so that s. Arguing as above shows that 0 5 M my ( b ) .h a. ( t ) 2 f ( t )2 hn ( t )for large n. and such that M s. 5 By .) ( t )# F ( t ) } .” h.h) 5 2 ~.b]: (p.h ( t ) n+co for almost every t E [a.e.”a 5 M. Let E = El U E2 and redefine H and E to be 0 on E . h ( t ) 5 f ( t ) 5 H ( t ) for all t E [a.7is McShane integrable and equal to h 8. Since H.. and is McShane integrable.
=M [ n (My( t + i) . It now follows from Lemma 5. so Thus.23 shows that f is McShane integrable over I and. of course.my ( t + k) + m y ( t ) )d t by Lemma 5. 0 ( H n ..h n ) ( x ) if lim inf. 0 .34 (3). 5 4 almost everywhere and 4 is finite everywhere.hn) ( t ). By Fatou’s as we wished to show.b].e.. F ( t )= lim inf n+oo (B. (Hn .hn) ( x ) is finite otherwise so that Lemma..35 that 7 is McShane integrable over I .34 (5). n o +O nmo = lim inf n+w + Define4 by = { lirn inf. for almost every t E [a.. My .hn) ( t )= H ( t )+ lim inf (hn ( t ) ) n+co Hn ( t ) lim inf (hn ( t ) )5 lim inf ( H n .my is increasing..My ( t ). Since f = f a. the McShane and HenstockKurzweil integrals are equal. Lemma 5.250 Theories of Integration In fact. once f is McShane integrable over I . Now. by Lemma 5.
McShane integral
25 1
By Theorem 5.37 and Corollary 4.80, it follows that McShane and Lebesgue integrability are equivalent. We conclude this section by giving a direct proof of this result, which uses arguments more like those found in the Lebesgue theory.
Theorem 5.38 Let f : [a,b] + R. Then, f is McShane integrable i f , and only i f , f is Lebesgue integrable. The value of the two integrals are the
same.
Proof. Assume first that f is Lebesgue integrable over [u,b]. Without loss of generality, we may assume that f is nonnegative. Let E > 0 and by absolute continuity (see Remark 3.93) choose 6 > 0 so that L f <E whenever A c [a,b] is measurable and m (A) < 6. Set A = min 6). Let a = min 1,& Set Ek = { t E [ U , b ] : ( k  1) 5 f ( t )< ka} for k E N. Then, Ek f l Ej = 8 if k # j and [a,b] = U ~  , E I , . For each k , choose an open set Gk such that EI, C GI,and m (GI,\ EI,)< Define a gauge y on [a,b] as follows. If t E Ek, then choose an open interval y ( t )c GI,that contains t. Suppose that D = { ( t i ,I i ) : i = 1, , . . , I } is a yfine free tagged partition of [a,b]. We will show that
sA
( E )
{
}.
&.
which implies that f is McShane integrable with integral equal t o L fa f. For i = 1,.. . , I , choose ki so that ti E EI,;.Then,
b
= Ri+
R 2
iR3. 
If ti,t E Eki, then both f (ti) and f ( t ) belong to the interval
252
Theories of Integration
[ ( k i  1)a , kia), SO that
If
(ti) f (t)l < a. Thus,
To estimate
R2,
since ti E Ek, and Ii C y ( t i ) C Gk,, we have
Finally, let
Since
Thus, by the choice of 6, R3 5 L / A f
23) shows that IS (f,  C
s,” f 1 < 3 ~ proving that f is McShane integrable ,
< E . Combining all these estimates
andMS,bf=CS,bf. For the remainder of the proof, assume that f is McShane integrable over [a,b] and let F ( t ) = M s,” f . By Theorem 4.103, it is enough to show that F is absolutely continuous on [ q b ] to conclude that f is Lebesgue integrable there. Fix E > 0 and let y be a gauge on [a,b] such that I ( f , D ) M S < E for every yfine free tagged partition D of [a,b].
s,”
fl
Let VZ>O ( t i , : i = 1,.. . , I} be a yfine free tagged partition of [a,b], = { Ii) let M = max(If (ti)l,t = I,.. . , l } , and set 7 = Suppose that { [yj z j ] : j = 1,. . . , lc} is a finite collection of nonoverlapping subintervals of [a,b] such that
)
Replacing [yj, z j ] by the nondegenerate intervals in { [yj, z j ] n Ii}f=l, we may assume that for each j there. is an i so that [yj, zj] c Ii. Set Di=
McShane integral
{(ti, [ ~ jz , ] ) : [ ~ jz , ] C Ii}, for i = 1,.. . ,1. Then, j j free tagged subpartition of [a,b]. Since
253
D
= Uf,,Di
is a yfine
by Henstock's Lemma,
Thus, F is adsolutely continuous with respect to Lebesgue measure and f is Lebesgue integrable. 0
Remark 5.39
Theorems 5.37 and 5.38 are valid ,for unbounded intervals
I C R. See Exercises 5.27 and 5.28.
5.8
McShane integrals on Rn
The McShane integral can be extended to functions defined on intervals in (It*)" in the same manner as the HenstockKurzweil integral. If f is defined on an interval I c (R*)n, we assume that f vanishes at all infinite points and extend the definition of f to all of by setting f equal to 0 off of I . (See Sections 4.4.4 and 4.4.12 ). In fact, the only change needed to define the McShane integral over I is to extend the definition of a free tagged partition (Definition 5.1) to the interval I in the obvious way.
254
Theories of Integration
Definition 5.40 Let I be a closed subinterval of (R*)" and f : I + R. We call the function f McShane integrable over I if there is an A E R so that for all E > 0 there is a gauge y on I so that for every yfine free tagged partition D of I ,
Since every gauge y has at least one corresponding yfine tagged partition, and hence a yfine free tagged partition, this definition makes sense. The number A , called the McShane integral of f over I and denoted by A = / I f , is unique. The proof of this statement is the same as before. Recall that every McShane integrable function is HenstockKurzweil integrable. Since the value of the McShane integral is unique, it must equal the HenstockKurzweil integral. Thus, the basic properties of the McShane integral, such as linearity, positivity and the Cauchy criterion, carry over to this setting without further proof. By Example 4,111, the characteristic function of a brick is McShane integrable; by linearity, step functions are McShane integrable. Again, the McShane integral is an absolute integral in this setting. Finally, the Monotone Convergence Theorem, Dominated Convergence Theorem and Fatou's Lemma hold for the McShane integral in R". 5.9
Fubini and Tonelli Theorems
One of the main points of interest in the study of multiple integrals concerns the equality of multiple and iterated integrals. In Chapter 3, we gave conditions for the equality of these integrals for the Lebesgue integral in the Fubini and Tonelli Theorems (Theorems 3.109 and 3.110). We now establish versions of these two results for the McShane integral. These results are used later to establish the connection between the Lebesgue and McShane integrals on R". In proving the Fubini Theorem for the Lebesgue integral, we used Mikusinski's characterization of the Lebesgue integral. Since we do not have such a characterization for the McShane integral, our method of proof will be quite different and more in line with the usual proofs of the Fubini and Tonelli theorems for the Lebesgue integral. (See [Ro, 12.41.) For simplicity, we consider the case n = 2. We will use the notation for sections and iterated integrals that was employed in Section 3.3.8. In particular, it is enough for a function to be defined almost everywhere. We begin with a lemma which establishes the connection between
McSh,ane integral
255
Lebesgue measure and the McShane integral. Lemma 5.41 Suppose that E Then, m2 ( E ) = h 2 X E .
c R2
is measurable with
m 2
(E) <
00.
Proof. First, assume that E is a brick in R2. The gauge defined in Example 4.111 for the HenstockKurzweil integral also proves that xE is McShane integrable and JR2 xE = v ( E )= m ( E ) . 2 Next, assume that E is open. Then, by Lemma 3.44, E is a union of a countable collection of pairwise disjoint bricks, {Bi}iEo.Since m2 is countably additive, the Monotone Convergence Theorem implies
so the result holds for open sets. Now assume that E is a G d set. Then, E = nzlGi with Gi open, m (Gi) < 00, and Gi c Gi+l. By Proposition 3.34, the Monotone Conver2 gence Theorem, and the previous result, we have
We proved in Theorem 5.21 that if E C R is a null set, then & x E = 0. The same proofs works for subsets of R", so the conclusion holds for null sets in R ~ . Finally, assume that E is measurable and m ( E ) < 00. Then, E = 2 G \ B , where G is a Ss set, B is a null set, and B c G. This follows from Theorem 3.36, which is valid in higher dimensions, by setting B = G \ E, which is a null set. From the previous results, we have
m2
= m2
(G) =
k2 k2 L,
XG = XG 
XB =
S,,
(XG  X B ) =
l,
XE.
This completes the proof of the lemma.
0
From the equivalence of the Lebesgue and McShane integrals in R (Theorem 5.38) and Theorem 3.112, we derive Lemma 5.42
Let E
c R2 be measurable with m2 ( E ) < 00.
Then:
(1) f o r almost every x E R, the sections Ex are measurable; (2) the function x c)m (E,) is McShane integrable over R;
(3)
m 2
( E ) = J R m ( E x ) dx.
the Monotone Convergence Theorem implies that { JR f k (x. y) dy for almost every x. fk = 1f. . simple functions { f k } r = 1 which increases pointwise to f . (2) the function x I+ fx = f (x. y (3) the following equality holds: Iw Iw Proof.. Jwz f = limk. By Exercise 5.increases y) to JR f (2. increases to jx for every x E R. assume that f is nonnegative and McShane integrable. pairwise disjoint. there is a sequence of nonnegative. (1) and (2) hold and Next. k First. } ~ . = i ~ J . ~ f k = R2 J f <00* Thus. ) d y = limkw JR f k (2. we obtain /’ w w f (z. Jw f (z. From Lemmas 5. 9 Consequently.JRz f k .43 tegrable. ) dydx = :. Then: (Fubini’s Theorem) Let f : R x R + R be McShane in (1) fx is McShane integrable in R for almost every x E R. assume that f is a simple function with f (z) = aixAi (z).% 3 S. each f k is McShane integrable. and from the Monotone Convergence Theorem. By Theorem 3. where the Ai’s are measurable.62. note that by the Monotone Convergence Theorem.42.41 and 5.30. ) dy is finite for y almost every z.. y ) dy is finite for almost every 2. .256 Theories of Integration We now have the machinery in place to establish a Fubini Theorem for the McShane integral. y) dydz = lim IC+Oo Jw /w fk (x. from our previous work and two applications of the Monotone Convergence Theorem. Since { ( f k ) . dy}T==. To see that JR f (x. >d y is McShane integrable over R. and m 2 (Ai) < 00. Theorem 5.
Theorem 5.measurable and McShane integrable. A IC) x [ . f is Lebesgue integrable i f .f.McShane integral 257 Finally. Of course. f is absolutely HenstockKurzweil integrable. each fk is bounded.k . Define f k by f k (x. If JR integrable and sR The assumption in Tonelli's Theorem is that the iterated integral exists and is finite. each f k is McShane integrable. Y) Y) 9).45 Let f : Rn +R.k . Theorem 5. so f = fS . then f is McShane y and measurable. ) dydx exists and is finite. = (f (x.10 McShane. Then. . 0 As was the case with the Lebesgue integral. we showed that in R the McShane and Lebesgue integrals are equivalent and that a function is Lebesgue (McShane) integrable if. By Exercise 5.5.7. By a second application of these y) two results. proof. Then. k ] x [ . from which one can conclude that the double integral is finite. k ] (x. f is also HenstockKurzweil integrable so f is measurable by Theorem 4. f is absolutely HenstockKurzweil integrable. and only af. it absolutely HenstockKurzweil integrable. assume that f is McShane integrable. the roles of x and y can be interchanged.43 and the Monotone Convergence Theorem. dy for almost every x. we have sR 5. HenstockKurzweil and Lebesgue integrals in R" In Section 5. From Theorem 5. This result is contained in the Tonelli Theorem. Further.29.with both f+ and f. d.44 (Tonelli's Theorem) Let f : R x R +R be nonnegative f (x. In this section we extend these results to R". The result now follows from the case just proved. we can use the Fubini Theorem to obtain a criterion for integrability from the existence of iterated integrals. we have that { f k (x. Then. and only if. y) y } : increases to JR f (x. measurable and nonzero on a set of finite measure.113.
i ) . (fly) Exercise 5.7 hold in R". . We may assume that f is nonnegative and. Suppose that 1 f I 5 h on I and that h is McShane integrable over I with JI h = 0.46 Let f : R" 4 R. then f is absolutely HenstockKurzweil integrable and.5. Suppose that f is Lebesgue integrable.f. CJRz f = M JR M JR f ( x . 0 Theorem 5. C JR2 f = C JR C JR f (x.79 applies to R" since bounded step functions which vanish outside bounded intervals in R" are HenstockKurzweil integrable. that n = 2. the proof of Theorem 4. as before. 5.i . By Fubini's Theorem for the Lebesgue integral (Theorem 3. If f is nonnegative and measurable. by Tonelli's Theorem for the McShane integral.80.1 Letybeagaugeon[O. Give an example of a yfine free tagged partition tagged partition of [0. Let C J and M J denote the Lebesgue and McShane integrals.2 Prove that the characteristic function of a bounded interval I is McShane integrable and JR xI = l ( I ) .h : I c R* + R. y) Since the Lebesgue and McShane integrals coincide in R.109).45. the results of Section 5. If f is McShane integrable. y ( l ) = and y ( t ) = for 0 < t < 1. and only f is McShane integrable. therefore.3 Let f.y ) d y d x .l]definedbyy(O) = ( .:).113. Since any HenstockKurzweil integrable function is measurable by Theorem 4.as in the proof of Corollary 4. for convenience.11 Exercises Definitions Exercise5. Lebesgue integrable by Theorem 5. Prove that f is McShane integrable over I and J I f = 0. Now. (z. Then. f is McShane integrable and Thus. f as Lebesgue integrable if. Exercise 5. Proof.258 Theories of fntegration Proof. and hence absolutely McShane integrable.1] which is not a yfine tagged part it ion. the result follows by considering f = f+ . dydx.
Exercise 5.10 Let f : R + R be bounded and McShane integrable.13 Let f : I + R be McShane integrable over I . [Hint: Pick y such that IS sJ (If/.'.11 Let f.8 integrable. b ] with 7 f ' = r f.9.s. the Lipschitz condition and the Cauchy criterion.o If1 = 0. Basic properties Exercise 5. For p E N. I1 = [b." If1 for c > b.f2 . prove that f is McShane integrable over I .5 If I is a closed and bounded interval and f is continuous on I .6 (Translation) Let f : [a. Define f' : [TO.9 Let I c R* and J c R be intervals. Prove that g o f is McShane integrable over I .McShane integral 259 Exercise 5. [Hint: Suppose that If(t)l 5 M . Jbm If1 = 0. : i = 1.12 Let f : [u. Absolute integrability Exercise 5.b] + R be McShane integrable over [a. s. EL.b] and h E R. Let 'p : I c R* + R be a step function. M ] + R defined by g (y) = yP in Exercise 5. .] Exercise 5. Suppose that g : J + R satisfies a Lipschitz condition (see page 35) on J and f : I + J ." (5).] I Exercise 5. Use the function g : [M.b] + R be McShane integrable over [a.b] and h > 0.oo] . sI If11 .1 Exercise 5. b h] with fh = fa f. [Hint: Use the proof of Theorem 5.g : R + R be bounded and McShane integrable. Ii) Consider s. show that f P is McShane integrable.4 Suppose {uk}T=. . Show that D)< E: lirne(J).oo). rb] + R by ' ( t )= f f Show that ' f is McShane integrable over [m.7 (Dilation) Let f : [u. b h] + R by fh ( t )= f (t . Show that if f is McShane integrable over [l. Prove that f g is McShane integrable. see Example 5.5. [Hint: Recall that f g = [V + g12 .. k + l ) (x). + + + + sabfhh Exercise 5. Prove that limb.hb. [Hint: Pick y such that y ( t ) is bounded for t E R and (S(If1 ..I Exercise 5. ~ ] Fix such a 2) = {(ti.g2] /2." If1 < E whenever D is yfine free tagged partition of [ u . c R and set f (x)= a k X ( k . For the converse. Show that fh is McShane integrable over [u h. Define fh : [u h.11.D).then the series ak converges absolutely. Prove 'p is McShane Exercise 5.. cr=." s.oo) + R be McShane integrable. k } with tl = 00.
.17 State and prove the analog of Theorem 5. by fk ( t )= { k if f ( t ) < k f ( t ) if If(t)I I k k if f ( t ) > k * Show that each fk is McShane integrable and / I fk + f .g : I + R are bounded and McShane integrable. I i ): i = 1. Further.. then f g is McShane integrable. k } .Ii) : i & = { ( s i .] for for Exercise 5. Theorem 4.Ii) : i = 1. . .k } and use HenIi stock’s Lemma to see how to choose 6 so that ‘ ( J ) < 6 implies k If1 < 26.. define fk. .. Exercise 5.S (f. the truncation of f at k . i ) : i = 1..20 Let f : I c R* + R be McShane integrable over I . Let J be a subinterval of I . which employ the same subintervals of I .. For k E N. .. [Hint: Consider fk = x[o. ) 1 < 6 E for all yfine free tagged partitions V = {(ti. k } and set M = max{lf(ti)l : i = l . show that such a result fails for the HenstockKurzweil integral.3] k even.. ..18 Use Exercise 5. for all E > 0 there is a gauge y such that IS (f.15 Use Exercise 5. ..m } . ..12 to prove the following variant of the Cauchy criterion.m } and E = I { ( s i .70. sI . . for the HenstockKurzweil integral. Consider & = { ( t i .17 to prove the Monotone Convergence Theorem. . .15 to show that if f.Ii) : i = I . .14 to show that f : I + R is McShane integrable if. . m } and Exercise 5.19 Show that strict inequality can hold in Fatou’s Lemma. and only if. .14 Use Proposition 5.D).19 for the HenstockKurzweil integral Exercise 5. Fix such a partition D = { ( t i . for all e > 0 there is a gauge y such that 5 i=l for all yfine free tagged partitions 2) = { ( t i . m } = 1.2] k odd and fk = x[1. .260 Theories of Integration whenever D is yfine free tagged partition of I . .16 Use Exercise 5. and only if. The function f : I + Iw is McShane integrable if. Exercise 5. . Convergence theorems Exercise 5. . n J ) : i = 1.1 sJ Exercise 5.
6 to R” using the Fubini theorem. and 0 5 f g 5 M . + Exercise 5. 00).26 Exercise 5.21 Suppose that f .24 Suppose that f : R +R is continuous and the function z I+ x2f (2) is bounded. and 0 5 cp 1 f . that p is McShane integrable. and M are nonnegative and McShane integrable.22 Suppose that f and g are McShane integrable and g is bounded.] Exercise 5.32 Extend Exercise 5.with f McShane inteExercise 5. Prove that z I+ e’lf (2) is McShane integrable over [ O .41 to show .20 and 511.28 Exercise 5. If prove that f is McShane integrable.34.oo). [O. HenstockKurzweil and Lebesgue integrals Exercise 5.38 to unbounded intervals. Use Lemma 5. F’ubini and Tonelli Theorems Exercise 5. Prove that f g is McShane integrable. Exercise 5. o o ) for some a E R.25 Let f : I = [u.oo)+ R and suppose that the function x I+ easf (z) is McShane integrable over [ O . o o ) for every b > a and the function F defined by F ( b ) = e’lf (z)dx is continuous on [a.23 Let f : [O. Exercise 5. Suppose f : R2 + R is measurable and bounded.29 Prove part (2) of Lemma 5.10 and the Weierstrass Approximation Theorem to show that g o f is McShane integrable.30 Prove that a nonnegative. McShane.31 Suppose that f . c p : R2 grable.27 Exercise 5.McShane integral 26 1 Exercise 5. Extend Theorem 5. Use Exercise 5. Extend Theorem 5. Prove that f g is McShane integrable.37 to unbounded intervals.b] + R be McShane integrable and If1 5 c. simple function with support of finite measure on R” is McShane integrable. 9. . Exercise 5. cp simple and measurable. Suppose that g : [c. Show that f is McShane integrable over R.c] + R is continuous. [Hint: Use Exercises 5.
This page intentionally left blank .
. SOC. Real Variable and Integration. 289293. [D] G. Geneva.. Perron and Henstock. Advanced Calculus. Bartle. [Gr] I. Princeton. SUP. Mathematical Analysis. J. Banach. Amer. 1990. MIT Press. Ecole Norm. [DM] L. “Definitions of Riemann type of variational integral. Brand. McGrawHill. Introduction t o Real Analysis. M. “MBmoire sur les fonctions discontinues. T h e Integrals of Lebesgue. New York.4 (1875) NO. [Be] J. [Br] L. W . Ma. Fischer. Heath.” Proc. Lexington. Intermediate Real Analysis. Van Nostrand. New York. SOC. 1899. Gordon. 5 (1924)) 130136. (1961)) 402418. G.” Czech. Darboux. L’Enseignement Mathkmatique. Introduction to Real Analysis. Mikusinski. 1955. Swartz. 1976. Lebesgue. Debnath and P. Math. SpringerVerlag. 57112. Cambridge. Advanced Calculus. Benedetto. Cauchy. Academic Press. 1 (1948). Stuttgart. Denjoy. 82 (1957)) 418449. 1950. CroninScanlon. Henstock. 11 London Math. Math. 1970. Volumes I and 11. 2. New York. [Ban] S. RI. Wiley. Lanzhou Lectures o n Henstock Integration. 2000. Flett.. [Lee] Lee PengYee. Wiley. “Sur le thkor6me de M. [Bar] R. G. Math. [Leb] H.” Ann. DePree and C. [Fi] E. Kurzweil. Wiley. Alexiewicz. 1976. 1988. G. B. [BS] R. Teubner. 263 . GrattanGuinness. Measure Theory. K. Halmos. revised ed. Sherbert. [CS] J.Bibliography [A] A. 1972. Introduction to Hilbert Spaces with Applications. GauthierVillars. [Go] R.” Colloq. [C] A . T h e Elements of Real Analysis. New York. J. [K] J. D. “Linear functionals on Denjoy integrable functions. Math. B u v r e s Scientifiques. New York. Wiley. 1994.” Fund. Paris. [Ha] P. [Fl] T. Bartle and D. [He] R. Oeuvres Complete. Providence. New York. 1969. Vitali. T h e Development of the Foundations of Mathematical Analysis f r o m Euclid to Riemann. “Generalized ordinary differential equations and continuous dependence on a parameter. [DS] J. 1983. New York. World Scientific. 1966.
“Some applications of the Bounded Convergence Theorem for ar introductory course in analysis. 395397. USL 27 (1941). Academic Press. Wiley.’’ Amer. F.93 (1986). I and 11. Lewin. [Swl] C. New York. M. MacNeille. [McL] R. L. Acad.. Acad. Partielle Diflerentialgleichungen und deren Anwendung auf physikalische Fragen. Swartz. Academic Press. 3rd ed. Vol.” Amer. World Scientific. 1988. New York. McLeod. Wheeden and A . 1970. McShane. Introduction to Gauge Integrals. Vyborny. New York. New York. Mahwin. 988993. Marcel Dekker New York. [Sm] H. F. [Mi21 3. J .264 Theories of Integration Singapore. 1997. 1983. New York. Press. [Pel I. McShane. New York. New York. Sci. McGrawHill. B. 1953. Cambridge. Theory of Functions of a Real Variable. Swartz. Polon. Frederick Ungar. [N] I. Zygmund. 3rd ed. Pfeffer. Prentice Hall. Acad. Classical and Modern Theories of Integration. Analyse. 1989. 1. The Integral: An Easy Approach afte: Kurzweil and Henstock. [McSl] E. [Ru] W. Hattendorff (ed. The Riemann approach to integration. RI. SOC. Reading. The Generalized Riemann Integral. A n Introduction to Analysis. [Mu] M. 1960. [MacN] H. Ma. [LV] Lee PengYee and R. 1987. J . P. Gesammelte Mathematische Werke. Mathematical Association of America. Math. H. AddisonWesley. SOC. Paris.” Natl. 1980. N. Riemann.6 (1875) No. [McS2] E. 1993. 7176. Cambridge. “A unified theory of integration. 1993. Cambridge University Press. Real Analysis. Introduction to Measure and Integration. 1869. Rudin. Mikusinski. [Ri2] G. Braunschweig. [Mil] J. Math. “A truly elementary approach to the Bounded Convergence Theorem.. [MM] J . [Pf] W.. “On the integration of discontinuous functions. 349359. Weber (ed. SOC. [Lewl] J. Measure and Integral. Integration and Function Spaces. Sci. [ S W ~C. Mikusinski and P. “Sur une definition de l’integrale de Lebesgue. [Ro] H. . Smith.).). The Bochner Integral. Englewood Cliffs. [Lew2] J. F. Real and Complex Analysis. K . World Scientific. Singapore: 1994. Dover Publications. 12 (1964). Royden. Lond. Riemann. [Ma] J.” Bull. Sr. Math.” Amer. 2000. B. Unified Integration. Singapore ] 2001* [WZ] R. Mikusinski. NJ.94 (1987).. Pesin. M. L. “A Unified Theory of Integration. Math. Mikusinski. 1953. J. 203204. Vieweg. Lewin. De Boeck and Larcier. Providence. Cambridge University Press. 1977. [Rill G. Natanson.” Proc. Munroe. 1978. Monthl! 80 (1973). 148160.
4 . 11 II 111.38 L 1 ( E ) .e. 90 a L (f. 84 m. V. 61 28 28 117 117 73 73 51 205 12. 68 M I . 28 mi. 141..81 M . 120 E + h. 21 PI. 208. 12. 141 31.Xl]). (E).J). (f).64 m* (E). 20 M 1 ( I ) .60 m. (f). 45 265 . 22 22 97 I”. 240 M n . 225 a Jbf. 84 A. 99 J’f.S ) .244 m. 39 D. (f). 39 d(I. 38 PVJ: PUS: f .33 J ( S . 44 f . (f). 20 P ( P > . 68 mn.infzi. 120 Ex.Index a. 38 w (f. 1237 244 w (f. 64 123. ( E ) . P ) . 28 Mi. 39 D.123 t([Z&l. 42 80 73 84 72 218 172 6 12 38 68 39 74 D. 39 D. 244 Ex. 90 a lim supzi . 11 lim.7.
64 distance1 metric. 46 CauchySchwarz inequality. 187. 123 conditionally integrable. 37 characteristic function. 80 Bounded Convergence Theorem. 131 derivative lower. 186. 150. 89 almost everywhere. 74 Darboux. see Lebesgue integrable. 80 comparison test. 224 7. 109. 73. 85 extended realvalued function. 224 . 19. 6 CauchyRiemann integrable. 207. 110. 69 almost all. 11 80 20 172 172 80 Abel's Test. 20 Darboux integrable. 122 diverge. 6 Cauchy criterion. 80 closure. 64 Cauchy. 42. 89 Archimedes. 42. 170 complete. 205 algebra. 123 countable additivity. 188 bounded variation. 136 Dirichlet function. 44 conditionally. 20 Dedekind's Test. 79 CarathBodory. 202 absolutely convergent. 179 Denjoy. 80. 61 counting measure. 86 Fatou. 90 even function.266 Theories of Integration 85 80 60 36 141. 12 closed. 135 dense. 72 bounded. 136 upper. 161 converge. 74 countably subadditive. 2 ball. 239 Fischer. 80 Borel measure. 123 free tagged partition. 127 change of variables. see CauchyRiemann integrable. 38 compact. 115 absolutely integrable. see HenstockKurzweil integrable. 33 Alexiewicz seminorm. 20 upper. 60 countably additive. 45. 110. see McShane integrable additivity condition. 47 extended real numbers. 217 absolutely continuous. 106. 42 Dominated Convergence Theorem. 22 Darboux sum lower. 89 almost every. 80 distance from I to J . 42 . 172 brick. 86 Cantor set. see Riemann integrable. 80 canonical form. 122 distance. 79 generalized. 228 Cauchy principal value. 45 Cauchy sequence. 239 Egoroff. 109 Fatou's Lemma. 14 discrete metric. 123 Cauchy sum. 44. 77 Borel sets.
56 Lebesgue integrable. 208 absolutely. 38 Lebesgue.Index 267 interior. see Lebesgue integrable McShane. see Generalized Fundamental Theorem of Calculus: Part I. 245 maximum. 147 HenstockKurzweil integral indefinite. 238 multiplier. 48 measurable function. 80. 118. 171 norm. 136. 156. 84 Lebesgue measure 0. see McShane integrable Riemann. 213. 233 yfine free tagged partition. 64 measure. 74 mesh. 37. 207 gamma function. 195 improper integral. 149 integration by substitution. 64 integrable Darboux. 113 minimum. 181. 51 gauge. 92 Lusin. 41 limit. 141 Henstock’s Lemma. 97. 140. 92 Littlewood’s three principles. 123 inferior. 99 indefinite. 236. 7. 90 linearity. see Riemann integrable integrable over E HenstockKurzweil. 86 set. 103 absolutely. 229 HenstockKurzweil integrable. 122 metric space. 42 indefinite integral. 34. see HenstockKurzweil integral. 31 integration by parts. 141. 240 Riemann. 80. 256 Fundamental Theorem of Calculus: Part I. 193. 11 metric. 104 Lebesgue integral. see HenstockKurzweil integrable Lebesgue. 134. 224 yfine tagged partition. 122 free tagged subpartition. 190. 61 Monotone Convergence Theorem. 175. 224. 141 Generalized Fundamental Theorem of Calculus: Part I. 68. 164. 37 . 194 Lebesgue. 33 interval. 232 Fundamental Theorem of Calculus: Part 11. see Riemann integral. 148 generalized Riemann integral. 103 McShane. 90 superior. 140. 100. 28 McShane integrable. 245 monotone. 28 minor function. 117 Fubini’s Theorem. 122 Mikusinski. 143. 229 McShane integral. 15 Lipschitz condition. 104. 93 major function. 156. see Darboux integrable HenstockKurzweil. 68. 254 Mean Value Theorem. 190. 254 absolutely. see Lebesgue integral inner measure. 156. 203 Lebesgue measurable. 184. 136. 207 gauge integral. 35 Lipschitz constant. 35. 207 Jordan content outer. 191. 35 Littlewood. 83 Lebesgue measure. 228 Fubini. 163.
29 Vitali cover. 60. 16. 68 Tonelli. 22 regular inner. 139. 122 aalgebra. 59. 47 open. 198 Vitali Covering Lemma. 138 subpartit ion. 135 Perron integrable. 68 odd function. 163 symmetric difference. 7 semimetric. 15. 35. 36 simple function. 38 outer measure. 122 seminorm. 228 probkme d’intkgration. 61 translation invariant. 102 test set. 196 vector lattice. 7 Riemann integrable. 172 negligible. 244 sampling point. 207 2section. 224 tagged partition. 207. 123 RieszFischer Theorem. 80 tag. 81 partition. 61 variation. 86 step function. 22 Riemann sum. 94 Straddle Lemma. 204 lower. 123. 207 Perron. 207 free. 155. 41. 119 Tonelli’s Theorem. 11. 136 positivity. 163 free. 29 vector space. 7. 22 upper. 207. 119. 131 Riesz. 6. 146. 224 tagged subpartition. 213. 12 absolutely. 46 Riemann integral indefinite. 199 volume. 80 operator. 56 probkme de la mesure des ensembles. 80. 77 Riemann. 155. 155. 120 ysection. 102 Tchebyshev’s inequality. 139. 257 translation. 27. 120 . 224 RiemannLebesgue Lemma. 126 outer. 228 Tchebyshev. 154. 60 refinement. 69 signum function. 141.268 Theories of h t e g m t i o n null set. 15 oscillation.