Sie sind auf Seite 1von 235
RICHARD R. GOLDBERG The University of lowe YSIS ‘SECOND EDITION JOHN WILEY & SONS, Ine. London Sydoey ‘Copyright © 1976, by John Wiley & Sons, Inc. Copyright 1968, Ginn and Company (Xerox Corporation) All sights reserved. Published simultaneously in Canada [No part ofthis book may be reproduced by any means, or tranfmitted, nor translated into machine language wih: ‘out the writen permision ofthe publisher, Library of Congress Cataloging in Publication Data: Goldberg, Richard R. “Methods of rel analysis, Includes index 1. Funetone of real variables, 2, Mathemati cal analysis. Tie, QAS3I-5.058 1976 SIS'8 75.30615 ISBN 0-471-31065-4 Printed inthe United States of America Wo9s76s4321 PREFACE This book is intended as a one-year course for students who have completed an ordinary sequence of courses in elementary calculus. It presents in rigorous fashion basic material on the fundamental concepts and tools of analysis—functions, limits, continu- ity, derivatives and integrals, sequences, and series. Most of the difficult points usually lossed over in elementary courses are dealt with in detail, as well as many more advanced topics designed to give a good background for (and, hopefully, a taste of) modern analysis and topology. In particular, there are treatments of metric spaces and Lebesgue integration, topics that are often reserved for more advanced courses. Also included are many smaller but interesting topics not usually presented in courses at this level; these topics include Baire category and discontinuous functions, summability of series, the Weierstrass theorem on approximation of continuous functions by poly- nomials, and a proof of the standard existence theorem for differential equations from the point of view of fixed-point theory. The book is written at the same level as texts for traditional “advanced calculus” courses, but does not consider topics in “several variables.” Material on differentials and. vector calculus, in our opinion, can be understood best from the point of view of modern differential geometry and belong in a separate course. REMARKS ON THE SECOND EDITION Many changes in, additions to, and some deletions from the first edition have been made in accordance with thoughtful criticism from many colleagues at large and small institutions, ‘A major feature of this new edition is the addition of sections called “Notes and Additional Exercises,” which include a variety of material. There are famous theorems related to the material in the body of the text—for example, the Schrder-Bernstein theorem from set theory, the Tietze extension theorem from topology, and Stone's generalization of the Weierstrass approximation theorem. The proofs are given in outline only, with a great deal left to the student as exercises. In these new sections there are also we | prerace miscellaneous exercises (many of which are quite challenging) and an occasional histori- cal note. An instructor's solutions manual for the problems in the new material can be obtained from the author. T also added an appendix that contains an axiomatic treatment of the real number system, This was a compromise between no treatment at all in the first edition and a lengthy development from basic principles that | think would retard the reader's progress into the core of the book. All the assumptions about the real numbers and the necessary results that can be derived from these assumptions are carefully presented. There are a number of pictorial illustrations—also a departure from the first edition— ‘and new exercises in many of the chapters, and new proofs. Richard R. Goldberg CONTENTS Introduction; Assumptions and Notations 1 Sets and functions 3 1.1. Sets and elements 3 1.2. Operations on sets 4 1.3. Functions 8 1.4, Reol-valued functions “4 1.5. Equivalence, countability 6 1.6, Real numbers 20 1.7. Least upper bounds a Sequences of real numbers 7 2.1. Definition of sequence and subsequence 7 2.2. Limit of a sequence 2» 2.3. Convergent sequences 33 2.4, Divergent sequences 35 2.5. Bounded sequences 7 2.6. Monotone sequences 38 2.7. Operations on convergent sequences a 2.8. Operations on divergent sequences “7 2.9. Limit superior end limit inferior 48 2.10. Cauchy sequences 54 2.11. Summabilty of sequences 58 wii CONTENTS. 2.12, Limit superior and limit inferior for sequences of sets Series of real numbers 3.1, Convergence and divergence 3.2, Series with nonnegative terms 3.3, Aliernoting series 3.4, Conditional convergence and absolute convergence 3.5, Rearrangements of series 3.6, Tests for absolute convergence 3.7, Series whose terms form a noninereasing sequence 3.8, Summation by parts 3.9, (C,1) summability of series 3.10. The class /? 3.11, Real numbers and de al expansions 3.12. Notes ond additional exercises for Chopters 1, 2, ond 3 Limits and metric spaces 4.1, Limit of function on the real ine 4.2. Mette spaces 4.3, Limits in metre spaces Continuous functions on metric spaces 5.1. Functions continuous at a point on the real line 5.2. Reformulation 5.3, Functions continuous on a metric space 5.4, Open sets 5.5. Closed sels 5.8, Discontinuous functions on R 5.7. The distance from @ point fo 0 set Connectedness, completeness, and compactness 6.1. More about open sets 6.2. Connected sets 6.3. Bounded sets and totally bounded sets 6.4. Complete metric spaces, 6.5. Compact metric spaces 338 74 Sees e 108. 108, 108 ne 121 126 126 129 132 134 138 43 146 148 148 149 153 156 160 10. CONTENTS, 6.6, Continuous functions on compact metric spaces 8.7. Continuity ofthe inverse function 6.8. Uniform continuity 6.9. Notes and additional exercises for Chapters 4, 5, ond 6 Cateutus 7.1. Sets of measure zero 7.2. Definition of the Riemann integral 7.3. Existence of the Riemann integral 7.4. Properties of the Riemann integral 7.8. Derivatives 7.6. Rolle’s theorem 7.7. The law of the mean 7.8. Fundamentel theorems of caleulus 7.9. Improper integrals 7.10. Improper integrals (continved} The elementary functions. Taylor series at. 82. 83. 8.4, 8.5. 8.6. a7. Hyperbolic functions ‘The exponential function ‘The logarithmic function, Definition of x The trigonometric Functions Taylor's theorem ‘The binomial theorem LHospital’s rule Sequences and series of functions 91. 9.2, 93. 9.4, 9.5, 9.6. 97. Pointwise convergence of sequences of functions Uniform convergence of sequences of functions Consequences of uniform convergence Convergence and uniform convergence of series of functions Integration and differentiation of series of functions Abel summability differentiable function Three famous theorems 10.1 ‘The metric space Cla.) 163 166 167 in 179 179 180 196 188 193 205 2 218 224 224 26 28 230 235 242 252 252 255 Seee 38 MW. 12. CONTENTS. 10.2 103. 10.4. 105. ‘The Weierstrass approximation theorem Picard existence theorem for differential equations ‘The Arzela theorem on equicontinuous families Notes and addlonal exercises for Chapters 9 and 10 The Lebesgue integral ma 2. 11.3, ns. Ws. 6. nn. 118. ne. Length of open tets and closed sets Inner and outer measure. Meaturable sets Properties of measurable sels ‘Measurable functions Definition and existence of the Lebesgue integral for bounded functions Properties of the Lebesgue integral for bounded measurable functions The Lebesgue integral for unbounded functions ‘Some fundamental theorems ‘The metric space ©?[a,b} 11.10. The integral on (~ 00, 00) and in the plane Fourier series 120 12.2. 12.3, 12.4, 1258. 126. 127, Definition of Fourier 5 Formulation of convergence problems The (C, 1) summability of Fourier series The £? theory of Four Convergence of Fourier series Orthonormal expansions in £7{a,6] Notes and addtional exercises for Chapters 11 and 12 Appendix: The algebraic and order axioms for the real number Special Symbols Index 283 287 302 an 315 321 328 336 341 348, 355 955 358 362 364 369 373 380 388 395 INTRODUCTION ASSUMPTIONS AND NOTATIONS ‘A. The book does not begin with an extended development of the real numbers However, the reader who wishes to proceed in strictly logical order should first digest the basic definitions and theorems about sets and functions in Sections 1.1 through 1.3, and then turn to the Appendix for the algebra and order axioms of the reals and the theorems on arithmetic and inequalities that are derived from those axioms. After the Appendix the reader should go to Section 1.7 where the least upper bound axiom is presented. At this point the reader will have seen a careful treatment of all the basic assumptions about the reals. Anyone choosing this approach may skip now to paragraph ci B. There are some who feel, however, that it is preferable at first to be less formal about the real numbers so that the reader can get to the meat of the book more quickly From this point of view it is better to delay reading the Appendix and simply proceed directly through the main body of the text. For those who wish to take this approach we ‘mention briefly some facts about the reals. ‘An integer is @ “whole number.” Thus 6,0, ~ 3 are integers. A rational number is a real number that can be expressed as a quotient of intexers. Thus 3/2 and 9/276 are rational numbers. Any integer k is thus @ rational number since we may write k= /1 An irrational number is a real number that is not a rational number. For example, & solution to the equation x?=2 must be an irrational number. The reader should have some facility in handling inequalities. He should know that if xvand y are real nuinbers and x—y. Also, if 00 we define |x| to be x. For x<0 we define |x| to be ~ x. Finally we define {01=0. Thus for any real numbers x, |x| is the “numerical value” of x. We call |x| the absolute value of x. By considering various cases according to the sign of x and of y, the reader should have no difficulty in proving the immensely important results Ix+yl0 and for rational values of x and y. In Chapter 8 we define a* for any real number x and then prove the familiar laws of exponents for arbitrary exponents. The notations a'/? and Va both mean the positive square root of a. (The existence of a positive square root for any positive number is presented in exercise 8 of Section 6.2.) D. Here are some notations. If a and 6 are real numbers with aa. By (—20,a) we mean the set ofall real x with x |x>0, y >0}—the set of all points (x,y) such that x is nonnegative and y is nonnegative Similarly, (0,1]=(x10< <1) DEFINITION. If b is an element of the set 4, we write bEA. If b is not an element of A. we write be A. Thus a& (a,b,c) but d= {a,b,c). As another illustration, suppose we define a base- ball team as the set of its players and define the American League to be the set of its twelve member teams. Then, in the notations we have just introduced, ‘American League= (A's, Tigers,..., Rangers}, A’s= (Jackson, Bando, ....Campaneris}, A’s@ American League, Jackson € As. 4 | SETS ANO FUNCTIONS ‘Note that the elements of the American League are themselves sets, which illustrates the fact that a set can be an element of another set. Note also that although Jackson plays in the American League, he is not an element of the American League as we have defined it, Hence Jackson @ American League. Exercises 1.1 1, Describe the following sets of real numbers geometrically: A={x]x<7}, B= (xl|x|>2}, C= (x|lx|=1}. 2. Describe the following sets of points in the plane geometrically: A= (Cay btty7= 1}, B= { signs in arithmetic, We now define them. ouRE 2. L2E. DEFINITION. If every element of the set 4 is an element of the set B, we write ACB (read “A is contained in B” or “A is included in B”) or B>A (read “B contains A”). EAC B, we say that 4 is a subset of B. A proper subset of B is a subset ACB such that A % B. (See Figure 3.) URE 3. Thus if A=(16,7), B=(1,3,6,7,8). C=(2,3,4,5,...,100), o then ACB but BZC (even though C has 99 elements and B has only 5). Also OCD and DCD for any set D. 1.2F, DEFINITION. We say that ‘wo sets are equal if they contain precisely the same elements, Thus 4=B if and only if 4c Band BCA (verity Note that for & and C in (1) of 1.26, none of the relations BCC,C.c B,C=B hold. 1.2G. It is often the case that all sets 4,B,C,... in a given discussion are subsets of a “big” set S. Then S—A jis called the complement of A (relative to S), the phrase in parentheses sometimes being omitted. For example, the set of rational numbers is the complement of the set of irrational numbers (relative to the reals). When there is no ambiguity as to what S is we write S— =A’, Thus A” [meaning (4’)'] is equal to 4. Moreover, S=4 UA’. See Figure 4. We now prove our first theorem, 12H. THEOREM. If A,B are subsets of S, then (AUBY =A’ BB" a) FIOURE 4. 1.2 OPERATIONS ON sets | 7 and (An By= aus 2) ‘These equations are sometimes called De Morgan's laws. PROOF: If x€(4UB), then x= AUB. Thus x is an element of neither A nor B so that x@A’ and x€B". Thus x€A’q B', Hence (AU BY CA’N B’. Conversely, if ye A‘OB’, then yea" and yEB", so that yA and yB. Thus yZAUB, and so YE(AU BY, Hence 4’7 B'C(AU BY. This establishes (1). Equation (2) may be proved in the same manner or it can be deduced from (1) as follows: In (1) replace A,B by 4’,B" respectively, so that 4’, B’ are replaced by A”=A and B” = B. We obtain (4’U B'Y = An B. Now take the complement of both sides. Exercises 1.2 1. Let A be the set of letters in the word “trivial,” A=(a,i,l,r 1,0). Let B be the set of letters in the word “difficult.” Find AU BA‘) B,A— B, B— A. If S is the set of all 26 letters in the alphabet and 4’= S—4,B’=S—B, find A',B’,A’7B’. Then verify that A’) B=(AUBY. 2. For the sets A,B,C in Exercise | of Section 1.1, describe geometrically 4m B, Bn GANG. 3. Do the same for the sets 4, B,C of Exercise 2 of Section 1.1 4. For any sets A,B,C prove that (AuB)UC=AU(BUC). This is an associative law for union of sets and shows that AU BU C may be written without parentheses. 5. Prove; for any sets 4,B,C, that (An B)nC=An(BNC). This is an associative law for intersection of sets 6. Prove the distributive law Aa(BUC)=(An B)U(ANC). See Figure 5. FIGURES. An(BUC}=1AN BULAN) 1. Prove (AUB)-(AnB)= A~B)U(B-A) 8 | SETS AND FUNCTIONS 8. True or false (that is, prove true for all sets A,B,C, or give an example to show false): (a) (AUB)-C=AU(B-C). (b) (AUB)—A (©) (AN BUBNCUANC)=AN BAC. () (AUB)NC=AU(BNC). 9, True of false: (a) If ACB and BCC, then ACC. (b) If AcC and BCC, then AUBCC. © [0,1]3@,)). ) (slixi>4}9 rly > 4)=Cllel> 4), 1.3 FUNCTIONS: 1.3A, In the cruder calculus texts we see the following definition: “If to each x (in a set ‘S) there corresponds one and only one value of», then we say that y is a funetion of x.” This “definition,” although it embodies the essential idea of the function concept, does not conform to our purpose of keeping undefined terms to a minimum. (What does “correspond” mean?) In other places we see a function-defined as a graph. Again, this is not suitable for us since “graph” is as yet undefined. However, since a plane graph (intuitively) is a certain kind of set of points, and each point is (given by) a pair of numbers, this will lead us to an acceptable definition of function in Section 1.3C. 13B. DEFINITION. If A,B are sets, then the Cartesian product of A and B (denoted AX B) is the set of all ordered pairs* (a,b) where aE 4 and bE B. Thus the Cartesian product of the set of real numbers with itself gives the set of all ordered pairs of real numbers. We usually call this last set the plane (after we define the distance between pairs). See Figure 6. ‘The lateral surface of a right circular cylinder can be regarded as the Cartesian product of a line segment and a circle. (Why?) We are now in a position to define function. 1.3C. DEFINITION. Let A and B be any two sets. A function f from (or on) A into B is a subset of x B (and hence is a set of ordered pairs ) with the property that each GEA belongs to precisely one pair (a,b). Instead of (x,y) Ef we usually write y=f(2). Then y is called the image of x under J. The set 4 is called the domain of f. The range of f is the set (6 B|b=f(a) for some a). That is, the range of fis the subset of B consisting of all images of elements of 4. Such a function is sometimes called a mapping of A into B. If CCB, then f-\(C) is defined as (a€A|f(a)€C), the set of all points in the domain of f whose images are in C. If C has only one point in it, say C= (’), we usually write f~'(9) instead of f—'({1). The set f~'(C) is called the inverse image of C under. (Note that no definition has been given for the symbol f~' by itself.) IDA, then f(D) is defined as { f(x)]x=D)}. The set f(D) is called the image of D under J Consider Figure 7. The dots at the beginning of the arrows denote points in the domain of f. Thus the statement f(a)= is pictured by an arrow starting at a and ending +o keep the revord clear we had beiter define “ordered pais” What is needed is set with o and b ‘mentioned in an asymmetrical fashion, How about defining (2,2) to be ((a),(a,8})? FIGURE 7. Diogtom of 2 fonction {fom A inte B 10_| sets AND FuNcnons at b. According to the definition of function, no two distinct arrows may begin at any GEA, but two (or more) may end at some bE B. We should think of f as sending points of 4 (o points of B. Note that f(a)=f(c)=b so that f~"(b)=(a,c}. If a function has domain and range both consisting of real numbers, then we can use the familiar method of graphing the function in the x-y plane. It is often possible to infer a good deal of information from such a graph. However, for understanding basic concepts about functions such as limit and continuity, a diagram like that in Figure 7 often is more helpful than an x-y graph. For example, the set f= (B and if XC BY CB, then PUG =F IOUS) a In words, the inverse image of the union of two sets is the union of the inverse images. PROOF: Suppose a€/~'(X U Y). Then f(a)eX U ¥. Hence either f(a)EX or fae ¥ so that either ae f—\(X) or a€f~'(¥). But this says ae f—'OX)US-1(¥). Thus SUX UN CI COHUS-'). Conversely, if bEFOQUS-"(Y), then either bE S~1QD or 6Ef~'(¥). Hence either f(6)EX or f(6)EY so that f(b)EXU Y. Thus BES "(XUY), and 90 f"YUS-")CF~ "XU Y). This proves (1). ‘The next theorem can be proved in exactly the same way. LBP. THEOREM. If f:A—>B and if XC BY CB, then SKA Y= LX) INO). In words, the inverse image of the intersection of two sets in the intersection of the inverse images. PROOF: The proof is left as an exercise. The last two results concerned inverse images. Here is one about images. 13G. THEOREM. If f:d>B and XCA,YCA, then A(XUY)=fA)UAY)- In words, the image of the union of two sets is the union of the images. PRooF: If bE f(XUY), then b= f(a) for some aE XU Y. Either a€X or ae. ‘Thus either bE f(X) or BES(Y). Hence bE f(X)U f() which shows f(X U Y)C/OQU. J(). Conversely, if ¢& f(X)US(Y) then either ¢€ f(X) or C= /(Y). Then c is the image ‘of some point in X or ¢ is the image of some point in Y. Hence ¢ is the, image of some point in X UY, that is €€/(X U ¥). So(X)USMCMXU ¥). 13H. Conspicuously absent from this list of results is the relation FOXY X)G(Y) fork ca, ¥Ca Prove that this relation need nor hold. 1.31, DEFINITION (THE CoMPostTioN OF FUNCTIONS). If f:4—B and g:B-C, then we define the function g°f by ef(xy=a[f(s)] (EA). 12. | sets ANO FUNCTIONS ‘That is, the image of x under gof is defined to be the image of f(x) under g. The function gf is called the composition of f with g. [Some people write g(f) instead of sf Thus ges: AC. For example, if S(x)mltsiny (=< x<00), a(xj=x? 0K x< 0), then gef(x)=142sinxtsin’x — (—20B be defined by f(x)=B,g: B>C,h:C—D, prove that he(gef)= (hee) *f. 14 | Sets ANo FUNCTIONS 13. For which of the following pairs of functions f and g is g an extension of /? (0R and g:A—>R, we define f+ as the function whose value at xGA is equal to f(x)+ g(x). That is, (Fraxnfe)ta(x) (xe A). Tn set notation St e={Cuf(x) taxed) Itis clear that f+ g:A—R. Similarly, we define f—g and fg by (F-)()=I)-8(9) (we A), (O=M)a(0) (EA). Finally, if g(x)#0 for all x, we can define f/g by F) ye £2) (slo aay «ee ‘The sum, difference, product, and quotient of two real-valued functions with the same domain are again real-valued functions. What permits us to define the sum of two real-valued functions is the fact that addition of real numbers is defined. In general, if S:A->B,g:A—B, there is no way to define f+ g unless there is a “plus” operation in B, 14C. pernimion. If f:A-+R and ¢ is a real number (c€ R), the function of is defined by (A= e[f()] (xe A), ‘Thus the value of 3/ at x is 3 times the value of fat x. 14D. For a,b real numbers let max(a,6) denote the larger and min(a,b) denote the smaller of a and 6. [If a=, then max(a,b)=min(a,b)=a= 6] Then we can define max(J,g) and min(f.g) for real-valued functions f.g. DEFINITION, If f:A—>Rig:A—R, then max(/.g) is the function defined by man(Fig)x)=mas{/(se(s)] (XE A), 14 REALVALUED FUNCTIONS | 15 and min( fe) the funtion defined by min(f.g)(x)=min[f(x).g(x)]_ (*€A), Thus if f(x) msn (0x6 7/2)y6(8)~ cOss(06 x <-7/2) and mara) then Aaymeox (0¢x<2), a Mayesing (Zexed), DEFINITION. If f:4—>R, then || is the function defined by II@)=lf@) (eA). If a,b are real numbers, the formulae Ja-blta+d max(a, 6 min(a,b)= are easy to verify. (Do so.) From them follow immediately the formulae nal majg)= Ealtlts, nln ales rin(ig= ee, for real-valued functions fg. L.4E. In this section we consider sets which are all subsets of a “big” set S. If ACS, then A’=S—A (Section 1.2G). For each ACS we define a function x, as follows. DEFINITION. If ACS, then x, (called the characteristic function of A) is defined as x= (XE A), xa(xynO (x EA’), The reason for the name “characteristic function” is obvious—the set A is char- acterized (completely described) by x,. That is, d= B if and only if x4—Xp. The reader should verify the following useful equations for characteristic functions where A,B are subsets of 5. Xava=MAx(X4oXp)s Q) Xara = Min(K4Xe)™XaXer Xa-e™Xi-Xe (provided BCA), Xu tXet xeoh Xen 0+ For example, to establish (I), suppose x= AUB. Then x4, 9(2)= 1. But ether x4 + We are using | ere to denote the real-valued function whose value at each x65 i equal tthe number (that i, ere Lis the "function identealy I”), Thus the symbol 1 bas wo different mesnings—one & ‘number, the other a function. The reader will be able to tll fromthe context which meaning to assign. 4 The 0 denotes the function ieaticaly 0. 16 | Sets AND FuNcTONS or x€ B (or both), and so either x4(x)= 1 oF xp(x)= 1. Thus max(xq.x9)(2)= I. Hence Texaua(s)=martxexs\(x) (EAU B) @ If x@AUB, then x4, 9(x)=0. But x€A’7 B" by (1) of Section 12H and hence xeA and xB" so that xy(X)=0=X9(x). Thus max(x4,X_)(x)=0. Hence O=Xaual=man(Kxe(x) — (XBAUB). @) Equation (1) now follows from (2) and (3). Exercises 1.4 1, Let f(x) =2x(~ 29 B, then f is called one-to-one (denoted 1-1) if F(a Thus if fis 1-1 and 6= f(a,), then 6% f(a3) for any ay€A distinct from ay. Thus the function f defined by f( x)= x(— 90- [where the /(i) are all distinct from one another}. 11m is postive integer, then the statement "8 has m elements” means “B is equivalent to the set (den) 18 | Sets AND FUNCTIONS Hence, saying that 4 is countable means that its elements can be “counted” (arranged with “labels” 1,2,...). Instead of f(1),f@),.--. we usually write aa, For example, the set of all integers is countable. For by arranging the integers as 0,-1,+1,-2,42,..., we give a scheme by which they can be counted. [The last sentence is an imprecise but highly intuitive way of saying that the function f defined by Fyn th (nai. JOSE (n=2,4, is a [+1 correspondence between / and the set of all integers. For f(1),f(2),... is the same as 0,~1,1,~2,2,....] See Figure 9. This example shows that a set can be equivalent to a proper subset of itself. The same reasoning shows that if 4 and B are countable, then so is AU B. For A can be expressed as d= (a,,dp...} and similarly B= {y,b,...}- Thus @y,bj,d3sbydy.Bs,-.. 18 ‘a scheme for “counting” the elements of 4. B. (Of course, we must remove any b, which ‘occurs among the a's so that the same element in AU B is not counted twice.) 5 3 a 2| 2 1 1 7 G 1 2 6 FIGURE 9. Diagram of « 1-1 conespondence between the st of postive inegert nd the st ofl nagar | 1.5 EQUIVALENCE; COUNTABLITY | 19 The following theorem gives a much stronger result. LSP. THEOREM. If Ay,Az,... are countable sets, then* U2 ,4, is countable. In words, the countable union of countable sets is countable. PROOF: We may write 4y=(aj,al.as,...},42=(a}a}.a},...}y---4g™ (ay, af.af,...}, 80 that af is the kth element of the set 4,. Define the height of af to be j+k. ‘Then a} is the only element of height 2; likewise a} and a} are the only elements of height 3; and so on. Since for any positive integer m2 there are only m~ 1 elements of height m, we may arrange (count) the elements of UZ.4, according to their height as ataj,a},a3, being careful to remove any af that has already been counted. Pictorially, we are listing the elements of UZ. ,4, in the following array’ and counting them in the order indicated by the arrows: Wee vai Mm ‘The fact that this counting scheme eventually counts every aj proves that UZ.1d, is countable, We obtain the following important corollary. 1.5G. COROLLARY. The set of all rational numbers is countable. pRooF: The set of all rational numbers is the union U,£, where E, is the set of rationals which can be written with denominator n. That is, E,=(0/n,~1/n,1/n,~2/ 1n,2/ns..-). Now each E, is clearly equivalent to the set of all integers and is thus countable. (Why?) Hence the set of all rationals is the countable union of countable sets. Apply LSP. It seems clear that if we can count the elements of a set we can count the elements of any subset, We make this precise in the next theorem. 1.SH. THEOREM. If B is an infinite subset of the countable set A, then B is countable. PROOF: Let A=(dyay...}- Then each element of B is ang, Let n, be the smallest subscript for which a,, € B, let n, be the next smallest, and so on. Then B= (,dy---)+ The elements of B are thus labeled with 1,2,..., and So B is countable. 1.51. COROLLARY. The set of all rational numbers in (0, 1] is countable. PROOF: The proof follows directly from 1.5G and 1.5H. * We have not used the symbol U2. the Ay Aq before. It means, of couse the set ofall elements in at least one of 20 | sets AND FUNCTIONS Exercises 1.5 |, Which of the following define a 1-1 function? @ fi aer(-a so that }=0,10000--- (2), 4=0.01000' (2), =0.00010"-- (2), =}+4+4=0.1101000--- (2), where the (2) denotes binary expansion. Similarly, the ternary expansion of a real x uses the digits 0, 1,2. Thus x=Obpbyby 3) bby FP For example, 3), @). I+ (3), 2M 3), ‘The ternary expansion for a real number x is unique except for numbers such as } with ‘two expansions, one ending in a string of 2's, the other in a string of O's. 16D. The following set serves as a useful example later on. DEFINITION. The Cantor set K is the set of all numbers x in 0,1] which have a ternary expansion without the digit 1 ‘Thus the numbers $= 0.0222-++ (3) and }=0.20000-+- (3) are in K, but any x such that } L, then M cannot be a lub. for A since L is an v.b, and L b. (Thus no matter how small a is or how large is, there is an integral multiple of a that is greater than 6.) 26 | SEIS AND FUNCTIONS, PROOF: Let A=(nalnE). If the theorem were false, then b would be an upper bound for 4. Hence by 7D, the set A would have a least upper bound. Let B=|.u.b.4. Then, since B is the least upper bound, the number B—a is not an upper bound for A so that B—aR. Therefore N is a sequence. Roughly, then, a subsequence of {1}, is a sequence of integers whose terms get larger and larger. For example, the sequence of primes 2, 3, 5, 7, Hy... i8 a subsequence of (n). Other examples are 2, 4, 6 8... and 1, 3, 5,7, 21D. permrrion. If $=(s,)$., is a sequence of real numbers and N=(n,)%, is a subsequence of the sequence of positive integers, then the composite function S°WV is called a subsequence of S. Note that for € J we have =n, SeN(D=SIN(D]=S(r, and hence (adie ‘Thus our definition 2.1D conforms to the accepted notation $,,, sj»... for subsequences. In effect, N tells us which terms of $ to keep. For example, let us denote the sequence 1, 0, 1, 0,... by B, and define N= (n}i2, by 2i-1 (EI) so that m=1, another example, if 3, my=5,.... Then BeN is the subsequence 1, 1, I,... of B. For (G}fe1= (Va Py and N=(n)%,= (74), then (VFN CoN = (oy) Exercises 2.1 1. Let (s,) 224 be the sequence defined by sah Sper heat (MEBRA Ge Find s,. (The numbers s, are called the Fibonacci numbers.) 2, Write a formula or formulae for s, for each of the following sequences. [For example, 22 umroraseauence | 29 the sequence 2,1,4,3.6,5,87 1(n=2.4,6,...)4 (@) 1.0.1,0,1.0, (b) 1,3,6, 10, 15, (c) 1,-4.9, — 16,25, —36,. 8) AAN.21,3,1,4,1,5, 1600. 3. Which of the sequences (a), (b), (c), (d) in the previous exercise are subsequences of (nea? 4.1 S=[s)21=2n—1)f) and sequence of (KF. 5. Let § be a sequence. Prove that every subsequence of a subsequence of S is itself « subsequence of S. 6. If (5, }far is a subsequence of (s,)%. 1 prove that mek (kel). can be described by s,=m-+1 (n= 1,3,5, MYR i= (PVR y find s5,Sqa5qy 18 Na sub- 2.2 LIMIT OF A SEQUENCE ‘The concept of limit is one of the most important (and conceivably the most difficult) jn analysis. In this section we define the limit of a sequence (function on 1). Limits for other functions are discussed in the fourth chapter. Roughly speaking, the sequence {s,}%_, has the limit L if s,—L is “small” for all sufficiently large values of n. From this erude description, we would expect that the Sequence 1,1, 1,..-,has the limit 1, that the sequence 1,4, ,..., has the limit 0, and the sequence 1, ~2,3, ~4,..., does not have a limit. We shall see that our intuition in these cases is correct. In other cases, for example {sin(/n)}$. y, our intuition is not sharp enough to tell if 1 given sequence has a “limit” or to compute the “limit” if there is one. We need a precise definition of “limit of a sequence” and enough theorems about the definition to make computations easy 2.2A. pEFINITION. Let {5,}%2, be a sequence of real numbers. We say that 5, approaches the limit L (as approaches infinity),* if for every «> there is a positive integer NV such that Iy-Li) 0 1s, approaches the limit L we write digs Sok (n>0). Instead of “s, approaches the limit L” we often say that the sequence (5,] 2, has the limit L. We emphasize the fact that our definition requires that L be a real number. ‘Thus lim, ...8,= E means that for any €>0, the inequality |s,~L]<« must hold for * The phrase “atm approaches infinity” is part ofthe definition, We are not defining “infinity 30 | SEQUENCES OF REAL NUMBERS all values of n except at most a finite number—namely, = 1,2,....N—1. The value of NV will, in general, depend on the value of «. Thus for a given {5,}%_,, the proof that lim, ..25= L consists, upon being given an ¢>0, of finding a value of NV such that Is, El). ‘There is no need to find the smallest value of NV for which (1) holds. If, for each €>0, any N for which (1) is true has been found, this proves lim, 5, Consider Figure 11. All of the 5, except for at most a finite number of n, must be inside the parentheses. FIGURE 11, Digg of Jim = For example, consider the sequence 1,4,4,.... That is, consider (s,)%., where /n (n=1,2,...). We would naturally guess that this sequence has the limit L=0. Let us prove this. Given ¢>0 we must find WV so that (1) holds. In this case (1) reads |b-olce (>m, or dee (my @) Thus if we choose N so that 1/1V N. Now 1/N I/e. Hence if we take any NET such that N'>1/¢, then (1) will hold for this sequence (9), with L~0. This proves lim. 1/n=0. Note thatthe limit 0 is not equal to any tern ofthe sequence Tet"us now examine the sequence (5,)= where sj=1 (n=1,2,...), We have pre- viously guessed that this sequence has the limit L: To prove this we note that Sy L=1—1=0 0 that for any e>0, Iy-tl), Thus in this case, for any €> 0 we can make (1) hold by taking rare cases where N does not depend on ¢.) This proves lim,_.,, For a third example, consider (s,)2.) wheres, =n (n= 1,3...) that is, consider the sequence 1,2,3.... We shal prove this sequence does not have a limit. Assume the contrary. Then lim,.28,™ L for some L€ R. Then for any « there isan N for which (1) holds. In particular, for e= I there is an N for which (1) holds: I,-LI<1 (nN). (This is one of the This is equivalent to =1) L-l), 22 umToFA sequence | 31 The last statement says that all values of m greater than N lie between L—1 and L+ 1 This is clearly false and the contradiction shows that (5,)%7.,=(n}%-y does not have a limit. ‘The last example shows that a sequence whose terms get “too big" cannot have a limit This is not the only kind of sequence that does not have a limit. Consider the sequence (5) Par Where s,=(— 1" (n= 1,2,...). The terms of this sequence are ~ 1,1, —1,1, Suppose there were an LER for which lim, Then for e=} there would be an NE/ such that (1) holds. That is, K-p"-L1<} (>). 8) For n even (3) says [I L<45 4) While for m odd (3) says -LN m) meant Se (MPN) c Now the left side of (7) is less than Sn!”/n=8/nl/®, (Why?) Hence (1) will be true it Bce oN) a If we choose N so that 8/N'/?<¢, that is, choose N >64/e?, then (8) will certainly be true. (For 8/n'/?<8/N'/? if n> N.) We have thus shown that if NV is any positive integer greater than 64/¢2, then (8) and hence (7) and finally (6) will be true. This proves Timy see 522. Our intuition tells us that a sequence of nonnegative numbers cannot have a negative limit. This we now prove. 2.2B, THEOREM. If (5,)$2., is @ sequence of nonnegative numbers and if lim, then L>0. PROOF: Suppose the contrary, namely that L<0. Then for «=~ 1/2 there exists Ne/ such that L In-H< SE (ne), In particular which implies or But, by hypothesis, sy >0. This implies L>0, contradicting our supposition that L <0. Hence L> 0. This proof is a precise way of saying the following: If s, gets “arbitrarily close” to L when m is “large,” and L-<0, then 5, <0 for suficietly large n. Exercises 2.2 1. If (5,}82) is @ sequence of real numbers, if 5,0, prove L< M. 2 3. If (5,)2.) is @ sequence of real numbers and if, for every €>0, -L N. () Prove that lim, ..1/Vn-+T 6. If @ is a rational number, prove that the sequence (sinn!0r)%_, has a limit, 7. For each of the following sequences, prove either that the sequence has a limit or that the sequence does not have a limit (ial, wo {a}. (Ha), 8..(a) Prove that the sequence (107/n) has limit 0. (©) Prove that (n/10")= does not have a limit. (©) Note that the first” 10” terms of the sequence in (a) are greater than the corresponding terms of the sequence in (b). This emphasizes that the existence of 4 limit for a sequence does not depend on the first “few” (“few” = “any finite ‘number”) terms 9. Prove that (n~ 1/n)$., does not have a limit. 10. If s,=5"/n!, show that lim, 5, =0. (Hint: Prove that s, <(5°/5!)(5/) if n>5.) 11. If P is a polynomial function of third degree, (n>). P(xjeartbxttextd — (ajbied,xER), prove that P(nt) ne Bay 2.3 CONVERGENT SEQUENCES 2.3A, DEFINITION, If the sequence of real numbers {,}2., has the limit L, we say that {5,}a1 is convergent to L. If (s,)%., does not have a limit, we say that (5,)%, is divergent. From the examples of the last section we see that the sequences 1,1,1,... and are convergent (to the limits I and 0, respectively) and that the sequences and =1,+1,—1,41,... are divergent We now prove that a sequence cannot converge to more than one limit, 2.3B. tueonew. If the sequence of real numbers (s,}%21 is convergent to L, then (5,)221 cannot also converge to a limit distinet from” That is, if lim, .9%4=L and Tim, 5, M, then L= M. 34 | SEQUENCES OF REAL NUMBERS PROOF: Assume the contrary. Then LM so that [M~ L|>0. Let «= }[M~ LI. By the hypothesis lim, 5, L there exists ,€/ such that Iy-Li M). Similarly, since lim, ...5,=-M there exists Nj €/ such that by-Mi Ny). Let N=max(W,,N,). Then N >, and > N, So that both |sy—L| and |sy— AM] are less than ¢. Thus, [M~ L]=[(sy~ L) = (sy — ML < hse LL yy MI <2e=[M— Lf, which implies |M—L|<|M—L|. This contradiction shows M=L, which is what we wished to show. ‘The next result is almost obvious. In fact it is obvious, so the proof is left to the reader. 23C, THEOREM. If the sequence of real numbers {5,)%., is convergent to L, then any subsequence of (5,}%2y is also convergent to L. There is a useful corollary. 23D. COROLLARY. All subsequences of a convergent sequence of real numbers con- verge to the same limit, PROOF: If the sequence S converges to L then, by 2.3B, S converges to no other limit. By 2.3C, then, all subsequences of S converge to L (and to no other limit). This corollary yields an easy proof that S=((—I)")_, is divergent. For both 1,Ay1y--. and ~1,=1,—,... are subsequences of S and converge to different limits The example ((~1)"}Z. shows that a divergent sequence may have a convergent subsequence. The example’ (n)Z_, shows that a divergent sequence need have no convergent subsequence. ~ Here are some more examples. If @ is a rational number, 0<@<1, and S= {sinndir} 2. then S is divergent, For we can write @= a/b where a and b are integers and 6>2. The terms of $ for n=6,26,36,... are sinar,sin2ay,sin3az,.... Thus S contains the subsequence 0,0,0,.... But the terms of S for which n= 6+ 1,26+1,36+ I... are sin(ar+ ar/b),sinQar+ ax/b),sinGar + ax/b)..... or (~I}'sinen/b).(— 1)*sin(ax /b),(—I)"sin(an/b),.... These terms all have absolute value sin(ar/b) and hence do not approach 0. Thus S contains a subsequence which has the limit O and a subsequence which (may or may not converge but certainly) does not have the limit 0. By 23D, S is divergent For 0=0 or =I the sequence (sind). is clearly convergent to 0. It may be shown that if @ is irrational, then (sind) is divergent. This, however, is somewhat more difficult. Exercises 23 1. For any aE R show that [lat—{61] £ as n+00. (That is, if the subsequence of (5,} 2. of terms with even subscripts converges to L, as well as the subsequence with’ odd subscripts, then (5,) 21 converges to L also) 2.4 DIVERGENT SEQUENCES From the examples in Section 2.3 we see that the sequences (n}2, and ((—I)"}2.y are both divergent. As we have observed before, however, these sequences behave very differently. For (n)2_, diverges because its terms get “too big,” whereas ((—1)")24 diverges because its terms “oscillate too much.” In this section we make a classification of divergent sequences. 244. permrmion, Let (5,)$24 be a sequence of real numbers. We say that 5, approaches infinity as m approaches infinity if for any real number M >0 there is @ positive integer N such that 5,>M (n> N), In this case we write 5,00 as nao, Instead of “s, approaches infinity” we sometimes say (54) 24 diverges 10 infinity Just as we think of « in definition 2.2A as being “small,” we think of the M in this definition as being “large.” Thus if 5,20 as n-+20, then all but the first “few” of the s, are “large.” Itis obvious that ($2, diverges to infinity. For given M >0, just choose N EJ such that > M. Then certainly n>M (n> N). ‘The reader should be sure to verify that if 5,00 as m—s00, then 5, definitely does not approach a limit. (This justifies our use of the phrase ‘diverges to infinity.") We never refer to “infinity” as a limit of a sequence, A limit of « sequence must be a real number. 24B, DEFINITION. Let (s,)%., be a sequence of real numbers. We say that s, ap- proaches minus infinity as n approaches infinity if, for any real number M >, there is a positive integer N such that QS -M (n>). We then write 5,->~ 20 as noo and say (5,)%, diverges to minus infinity. ‘Again we think of M as “large” so that —M is “large negative.” For example, the sequence {log(I/n))., diverges to minus infinity. To prove this, 36 | SeQueNces OF REAL NUMBERS sven M 50 we mst find N.7 sucht lgbe-M (n>). 0 But this is equivalent to logn>M (nN), n>eM (nN), 2) Thus if we choose N > e™, then (2) and hence (1) will hold. The sequence 1,~2,3,~4,... does not approach either infinity or minus infinity. However, this sequence has the subsequence 1,3,5,... which approaches infinity and also has the subsequence -2,—4, ~6,... which approaches minus infinity. It is easy to show that if the sequence (s,)2_, diverges to infinity, then so does any subsequence of (5,}7.- (This is analogous to the result of 2.3C.) Some divergent sequences neither diverge to infinity nor diverge to minus infinity— they “oscillate.” 24C. DEFINITION. If the sequence {s,}%., of real numbers diverges but does not diverge to infinity and does not diverge to minus infinity, we say that (5,}®., oscillates. ‘An example of a sequence which oscillates is ((—1)")$.y. Another example is the sequence 1,2,1,3,1,4,1,5,.... For, by 23D, this sequence diverges since it has the divergent subsequence 1,2,3,4,.... Moreover, the sequence does not diverge to infinity since there is no NV €/ for which the statement 522 (n2N) is true, The sequence obviously does not diverge to minus infinity. Hence it oscillates We emphasize that “oscillate” does not mean “the terms go up and down.” The sequence 1, —4,+],—4,... converges to zero. Hence, by definition, it does not ascillate even though its terms “go up and down.” Oscillate is a term applied only to certain divergent sequences. Roughly speaking, a sequence oscillates if its terms “go up and down too much.” Exercises 2.4 1. Label each of the following sequences either (4) convergent, (1) divergent to infinity, (C) divergent to minus infinity, or (D) oscillating, (Use your intuition or information from your calculus course, Do not try to prove anything.) (a) {sin(ne/2)} 51 (&) {sinnr} (©) (e"}e- @) (el) (e) (nsin(a/n)}2.). O) (Yan(r/2-1/m} @ (l4he hee + U/mye, hy (— hay 2. Prove that (Vii )%., diverges to infinity, 3. Prove that (Vn+I — Vn )", is convergent. (Hint: Recall how to find dy /dx by the ‘Ax process when y= VX .) 4, Prove that if the sequence of real numbers (s,)*2, diverges to infinity, then (—5,).y diverges to minus infinity 5, Suppose {5,)$, converges to 0, Prove that ((—1)'5,)%.1 converges to 0. 6. Suppose [5,}', converges to L0. Prove that ((—1)'s,)%, oscillates. 7. Suppose {5,}# diverges to infinity. Prove that ((—1)'S,)2_, oscillates. 2.5 SOUNDED SEQUENCES | 37 2.5 BOUNDED SEQUENCES Recalling that a sequence of real numbers (s,}%_, is a function from / into R, we see that the range of (s,)%_, (namely (s,,53....}) is a subset of R. 2.5A. DEFINITION. We say that the sequence {5,}2., is bounded above if the range of (5,)Be1 is Bounded above (see 1.7A). Similarly, we say that (5,)=., is bounded below or bounded if the range of (s,}*. is respectively bounded below or bounded ‘Thus (s,}%2., is bounded if and only if there exists MER such that IgleM (nes), If a sequence diverges to infinity (or to minus infinity) the sequence is not bounded. (Werity.) A sequence that diverges to infinity must, however, be bounded below. (For such a sequence can have only a finite number of negative terms.) An oscillating sequence may or may not be bounded. The sequence 1, -2,3, —4, oscillates and is neither bounded above nor bounded below. The sequence ~1,1 =Iluse oscillates and is bounded. The sequence 1,2,1,3,1,4... oscillates and is bounded below but is not bounded above 2.5B, THEOREM. If the sequence of real numbers {5,}%=1 is convergent, then (5,1 is bounded. PROOF: Suppose L=lim,...- Then, given ¢= 1, there exists NE such that Iy-L). This implies Inl). @ (For |s,|=|L + (5,~ D)| <|L)+|5,—L1) If we let M=max(|s)|,|52),---5|8y— |}, then we have Igl0 there is a closed interval JCR of length such that s,€J for infinitely many values of 7, 38 | SeQUeNces OF REAL NUMBERS 2.6 MONOTONE SEQUENCES In the preceding section we saw that a sequence may be bounded and still not be convergent, In this section we consider a condition which, together with boundedness, will ensure that a sequence is convergent. 2.6A. DEFINITION. Let {s,)%., be a sequence of real numbers. If $, <<>> <5, Kya '7, then (5,)2.1 i8 called nondecreasing. Similarly, if s,> 5,2 °-- 25> 5.0; ><", then (5,)%. is called nonincreasing. A monotone sequence is a sequence Which is either nonincreasing or nondecreasing (or both). The sequence 1,14,13.1},... (that is, (2-1/2*-"}S.,) is nondecreasing (and bounded). The sequence {n}%_, is nondecreasing (and not bounded). These sequences ‘exemplify the results of the next two theorems, the first of which (2.6B) is of tremendous importance. 2.6B. THEOREM. A nondecreasing sequence which is bounded above* is convergent. PROOF: Suppose (1) s nondecreasing and bounded above. Then the set ) is nonempty subset of R which is bounded above. By .7D this set has a Laub. Let Metunb.(5)5p.-2) "lub ford We will prove that 5,>M as neo. Given €>0 the number M—e is not an u.b. for A Hence, for some N € 1,sy > M—. But, since {s,}%., is nondecreasing, this implies 52M (PN). a) On the other hand, since M is an w.b. for A, M>s, (nEl). A= (Si5x From (1) and (2) we conclude Iy-Mi0 we must find NE J such that 4>M (n>). a) Now, since M is not an upper bound for {5),5....} there must exist NEV such that sy >M, Then, for this N, (1) follows from the hypothesis that (s,}%.; is nondecreasing, This proves the theorem. ‘The proof of the following theorem follows the proofs of 2.6B and 2.6D exactly, with all upper bounds and least upper bounds replaced by lower bounds and greatest lower bounds. We leave the details to the reader. 2.6E, THEOREM. A nonincreasing sequence which is bounded below is convergent. A. nonincreasing sequence which is not bounded below diverges to minus infinity. 2.6F We close this section by showing that monotone subsequences always occur. 40 | SEQUENCES OF REAL NUMBERS THEOREM. Let S={s,}, be a sequence of real numbers. Then S has a monotone subsequence. PROOF: Let T, be the sequence s, Let T; denote the sequence 5 5, Indeed, for each n&7, let T, denote the sequence 5, Syays Syazs---- We divide the proof into two cases: case I. Suppose every sequence 7, has a greatest term. Let s,, be the greatest term in the sequence 7,, (If there is more than one greatest term, pick any one for s,.) Lets, be the greatest term in the sequence T,,,,. Then n>, and 5,,<5,,. Let s,, be the greatest term in the sequence T,,,. Then n>, and s,, <5,,. Continuing in this fashion we can construct (5,}/2—a nonincreasing subsequence of 5. case I. If case I does not hold, then for some n, € J, the sequence T,, has no greatest term, Since s,, is a term of T,,, there is a term 5,, of T,, that is greater than s,.Then there isa term s,, of T,, that is greater than s,,, Moreover, we may pick s,, with my >n, (why?) Continuing in this fashion, we can consiruct a nondecreasing subsequence (5, }7., of S. Hence in either case, Shas a monotonic subsequence, and the proof is complete, Exercises 2.6 1. Which of the following sequences are monotone? (a) (sinn) sa). (b) {tann} o {ale © Qn}, 2. If {5,}%1 is nondecreasing and bounded above, and L=lim, scbine Formulae the corresponding statement for noninreasng sequences, 4. If (s]ey and (4), are nondecreasing bounded sequences, and if 5,< 4 (WD), prove that lim, s,>4y, Prove that (s,)'2., is noninereasing. 8. Let elt tn Ges) Show that (s,}%., is monotone and bounded, and that lim,_..5, 2.7 OPERATIONS ON CONVERGENT SEQUENCES. | 41 9. Let {5,) 2 be @ Sequence of real numbers, and let stator +5, If {5,}22) is monotone and bounded, show that (1), is monotone and bounded. 10, For né7, let (nel). (a) Prove that (1,)$2., is nondecreasing. (b) Using only facts established in the proof of 2.6C, prove that (1,)%., is bounded above and then prove that lim, ..f, > im,_..(1+1/m)". 1, Let & denote the class of all sequences of real numbers. Let © denote the class ofall convergent sequences and ® the class of all divergent sequences. Further let 9p and ®y denote the classes of sequences that diverge to (plus) infinity and minus infinity, respectively. Let © denote the class of oscillating sequences. Finally, let denote the class of all bounded sequences and let R denote the class of all ‘monotone sequences. By citing the proper difinitions or theorems, verify the following statements @ &=cu9. () 9=9,U9,U0. : © cB. @ RNGce. © HXNB'c ©) 8n%,=0. 1p UDex. 2.7 OPERATIONS ON CONVERGENT SEQUENCES Since sequences of real numbers are real-valued functions, the definition of the sum, difference, product, and quotient of sequences follows from definition 1.4B. Thus if {5,Ja1 and {1,}%., are, sequences of real numbers, then (5,)2,+(t)z-1 is the sequence (5,+%)-1, and {5,}2.\°(} 2a is the sequence (5,°%,)¢-y. and 80 OA, Also, if CER, then e(s,}=, is the sequence (c5,) From the next theorem it follows that the sum of two convergent sequences is convergent. 2IA. THEOREM. If (5,} 21 and (4,)2 are sequences of real numbers, if lim, by and if lim, sty M, then lim,_.o(5,+ f,)= L-+ M. In words, the limit of the sum (of two convergent sequences) is the sum of the limits. PROOF: Given ¢>0 we must find 7 such that IGty)—“(Lt MI) N). ro) M|. Hence (1) will Now |(s,+ t)—(L+ M)|=[(5,— L)+(,— M)|S |5y— L]+ It certainly hold if = L+lg-MI), Q) We thus try to make both |s,~L| and |1,—M| less than ¢/2 by taking n sufficiently large. | 42 | sequences OF AeA. NUMBERS Since lim,_..8=L, there exists Ny such that Is, <5 (n> Ny). Also, since lim, oly M, there exists Ng such that I~ MI<$ (N,E1). Hence if we let N= max(N;,.N3), then the terms on the left of (2) are each less than «/2 when n> N. Thus for this N, @) and hence (1) hold and the proof is complete ‘The next theorem is easier to prove. 2.7B, THEOREM. If {5,) 2.1 is @ sequence of real numbers, if c€ R, and if lim, then lim,.,..¢5, = €L. PROOF: If e=0, the theorem is obvious. We therefore assume c0. Given €>0 we rust find N EJ such that Jes, eL|). Now, since lim, 05, But then leblsy-Li), which is equivalent to (1). ‘Theorem 2.7B is used in the proof of the following useful result, 2.C, THEOREM. (a) If O0 for nE/, {x")@_, is bounded below. By 2.6E, (x"}2_, is convergent. Let L=lim, ox". From 2.7B (with c= x) it follows that lim, .x-x"= xL. That is, (x VEY converges to xL. But (x"*!)2, is a subsequence of (x"}2.,. By 23D, L= xb and so L(1~x)=0. Since x1, this shows L=0, and part (a) if proved. (b) If x>1, then x"*!=.x:x">.x" so that (x"}#., is nondecreasing. We will show that {x"}., is not bounded above. For if (x}2_; were bounded above, then by 2.6B (x"}$., would converge to some LER. But the same reasoning as in (a) would show that = Lx, so that L=0=lim, x". But x*>1 and so (x"}2., obviously cannot converge to 0. This contradiction proves that (.x")_, is not bounded above. Conclusion (6) follows from 2.6D. We now treat the limit of the difference of two convergent sequences. 2.7D, THEOREM. If (5), and (1,)$2., are sequences of real numbers, if lim, 25, L, and if lim, fy M, then lim, (5, ~f)= L~ M. PROOF: Since lim, .../,=M, it follows from 2.7B (with c= ~1) that lim, ...(~1,) 27 OPERATIONS ON convenaent sequences | 43 =~ M. But then, using 2.7A (at the second equals sign), #0) gion y+ (— 0] = fim 5 in (—)= L+(—M)= tine which is what we wished to prove. A useful consequence of 2.7D and 2.28 is the following, 2.76. conoutany. If {5}, and (14). are convergent sequences of real numbers if Sq (MET), + lim, voty=M, then L< M, Proor: By 27D, M—L=lim,_.(,—5,). But (,-5,>0(nEJ). Hence by 2.28, ‘M~ L>0, which establishes our result. This corollary, of course, remains true even if s, > ¢, for a finite number of values ofr. We now show that the limit of the product of two convergent sequences is the product of their respective limits. We give two proofs of this result; each uses a technique useful in many other contexts. The first proof requires a lemma, 2.7F. veMMa. If (5,)$21 i8 a sequence of real numbers which converges to L, then (32)2.1 converges to £* s2=L?. That is, given €>0 we must find N GJ such pRooF: We must prove lim, that -L) or, equivalently, Ieg—Lflgg+Li). ” Now, by 2.5B (s,)., is bounded. Thus for some M>0 ) @) But then, using (2) and (3), Is. LI +f + Lay (M+) ‘Thus for this 1, (1) holds and the proof is complete. =e (ne) 2.7G. THEOREM. If (5,)%2., and (4,)$2.1 are sequences of real numbers, if lim,..2$," Ls and if lim, f= M, then lim, M, FIRST PROOF: We use the identity ab=3[(a+bf—(a—b)] (abe R) 0) 44 | SEQUENCES OF REAL NUMBERS Now, as 1-220, Sty oLtM — by27A, (s+) (L+M)? by 2.7F; Q) also Sy~tyoL-M — by2ID, (-t)(L- My by 27F. @) From (2), (@), and 27D, (51 Gat) Finally, using (1), (4), and 2.7B, Satu 4[ Gut 1)? (Gu 6)? ] + ALM ) = LM. Note that this proof uses no «. The technique of using the identity (1) to deal with the product is called polarization, (L+M)-(L~MP=4LM. @ SECOND PROOF: Given €>0 we must find EJ such that [sg LM|), a) The problem here is to do something algebraic [as we did in going from (1) to 2) of 2A] that will enable us to use our hypotheses lim, L and lity af, M. The trick of adding and subtracting the same quantity (in this case Li) will be used many times in this book. We have Sy ly EM = Sgly ~ Ly + Lly~ LM = f(g L) + L(t M). satu LMS Ital 15p— L1+1LI “16, M4 Hence (1) will certainly hold if ills L1+1LI-lq-Mi0. Then @) will certainly hold if Ols,—L+|Ll-|y-Mi). @) “Thus if we choose Ny €/ so that Qs-LI<$_ (n> Ny), and choose N3€/ such that WL MI<$ (032), then (3) will hold for N= max(N,,N3). Hence (2) and finally (1) will hold for this N, and we are done. ‘Now we turn our attention to the quotient of convergent sequences. 2.7H. ueMMa. If (1,}22y is a sequence of real numbers if lim, t= -M where M0, then* lim, 9(1/t))=1/M. * The hypothesis implies thats, cam be equal to zero for at most a finite number of». Thus 1/1, is defined forall bt at most finite numberof m 27. OPERATIONS ON CONVERGENT SEQUENCES | 45 PROOF: Either M>0 or M<0. We will prove the lemma in the case M >0. (The case M-<0 can be proved by applying the first case to (—1}%1) So we assume M >0. Given ¢>0 we must find N 7 such that 1 Tal N), MI Ta <6 PN) a Now there exists N, EJ such that |f,~M| N,). This implies (n>) In addition, there exists N37 such that In MicM (na ny, Thus if N= max(W;,N,), we have, for n>, Hence (1) holds for this V. This completes the proof. 2.71, THEOREM. If (5,)2., and (1,) $24 are sequences of real numbers, if lim, 5, = and if lim, of," -M Where M0, then lim, (5, /1,)=L/M. pRooF: Using 2.7H and 2.7G we have which is what we wished to show. 2.73. In Section 1.9 we proved directly from the definition of limit that We now do a similar problem illustrating the results of this section. PROBLEM: Prove First we write 3n2—6n _ 3—6/n Smt+4 544/02 We proved in Section 1.9 that lit, 4 (1/n)=0, Hence Jim £=6-020 by27B, 45 | SEQUENCES OF REAL NUMBERS. We also proved in Section 1.9 that lim,..,, 1=1. Hence fim3=3 by 2.78. Then tesa) eee by 27D. a Since we know lim,_,., (1/m)=0, we have by 216. Thus & Reasoning as before, we conclude dim (5+-4)=s+0=5 vy27A ® But then, from (1), @), and 2.71, . yn jim G-6/n) wie S¥4/n him (S447) 5” which is what we wished to show. Exercises 2.7 1. Prove (@) tim 2. Prove that if (s,)%., converges to 1, then (s}/7)2, converges to | Evaluate lim,_.,Va (Vat — Vn), . Suppose (5,]e1 is @ sequence of positive numbers and O 1, prove that lim, ,..c!/"= 1. [Hint: Write c'/*= 14-5, and take the mth power of both sides to show that (1s,)%., is bounded. Then conclude that s,->0 as n->20,] 9 Let )= V2 and let 5,4,=V2 Vs, for n> 1 (a) Prove, by induction, that s, <2'for all n. (b) Prove that s,.,>5, for all n. (6) Prove that (5,}22y is convergent. (@) Prove that lim,”...5)= 10. Suppose 5, > >0, and let 5,.1= J(6y +51) (> 2). Prove that () 5;3589,--- 18 noninereasing, (b) 83,5456.» is nondecreasing, (©) (5,)221 is convergent. I. Mfr, <5 <4, for all nEJ, and if both {r,}2., and (4,) 2, converge to s, prove that (s,}f1 converges to s. 2.8 OPERATIONS ON DIVERGENT SEQUENCES, In the preceding section we saw that the sum, difference, product, and quotient (if defined) of convergent sequences are again convergent. No such statement can be made in general about divergent sequences. Indeed, if (5,)$2., is a divergent sequence, then (—5,}221 is also divergent, and the sum of these two Sequences is clearly not divergent, Moreover, the product of the divergent sequence ((—1)"}2_, with itself is not divergent. For sequences that diverge to infinity, however, some positive results can be proved. 2BA. vutonEM, If (5,)21 and (Jq}2-1 are sequences of real numbers that diverge to infinity, then so do their sum and product. That is, (5, +4) and (54%) diverge to infinity Proor: Given M >0, choose N,€/ such that M (n>™), and choose Ny€/ such that 4>1 (n>). 1ax(N.N) (The above is possible since both 5,00 and (,->00 as n—+0.) Then, for N. we have . 5,44 >M+ID>M (n>), and Suh, M- M (n>N) Since M was an arbitrary positive number, this proves the theorem. 28B, THEOREM. If {s,)2., and {1,}2., are sequences of real numbers, if (5,}%.y diverges to infinity, and if (1,2, is bounded, then {5,+1,}_, diverges to infinity. 48 | SEQUENCES OF REAL NUMBERS. Proor: By hypothesis there exists Q >0 such that IWl0 choose N €/ such that 5.>M+Q (n>N). Then, for n>, Spt by > Se Ital >(M +O) That is, gy tt> (22), which shows that 5, + %,-»90 as n—00. 28C. COROLLARY. If (5,]%.1 diverges to infinity and if (14), converges, then (54+ 1,)a1 diverges to infinity PROOF: The proof follows directly from 2.5B and 2.8B, It is easy to show that 2.8A, 2.8B, and 2.8C remain true if “infinity” is replaced by “minus infinity.” 28D. Almost any kind of sequence can be formed from the sum of two properly chosen oscillating sequences. For example, the sum of the oscillating sequences 0,1,0,2,0,3,... and 1,0,2,0,3,0,... is the sequence 1,1,2,2,3,3,... which diverges 10 infinity. The sum of the oscillating sequences 1,0,1,0,1,0,... and 0,1,0,1.0.1,... is a convergent sequence. The sum of an oscillating sequence and itself is oscillating. Exercises 28 1. Give an example of sequences {5,}%_, and (t,) (8) 54920, yr = 00, Sy + by900, (B) 00, 300, 5, 2. Suppose that {5,}, is a divergent sequence of real numbers and c€ R, ¢%0. Prove that {¢5,} 2. diverges. 3. True or false? If (s,)%y is oscillating and not bounded, and (¢,)%., is bounded, then (5+ 6) is oscillating and not bounded. for which, as n—>00, 7. 2.9. LIMIT SUPERIOR AND LIMIT INFERIOR If (5,21 is @ convergent sequence, then lim, ...5, measures, roughly, “the size of 5, when 1 is large.” Of course, lim,....§, is a concept used only in connection with convergent sequences. In this section we introduce the related concepts of limit superior and limit inferior which can be applied to all sequences. Roughly, the limit superior of a sequence (s,}%, is a measure of “how big s, can be when m is large,” and the limit inferior or (5,}z-y is a measure of “how small s, can be when nn is large.” If lim, exists, it is then plausible that the limit, limit superior, and limit inferior of (5,)%2) are all equal, and this turns out to be the case. The real application of limit superior and limit inferior, however, is to sequences which are not known to be convergent. 2.9. WMT SUPERIOR AND LMT INFERIOR | 49 2.9A. First let us consider a sequence {5,}7., that is bounded above—say M,,., since M,4j=LUb. (5,4 jSpeae---) i8 the Lu.b. of a subset OF {5,.5,¢1nSyp42r--+)> Thus the sequence (M,)*. , is nonincreasing and thus either converges or diverges to minus infinity. DEFINITION. Let (5,)21 be a Sequence of real numbers that is bounded above, and let Wb. (Sp Sy 1rfpe2r (2) If (M,)2.4 converges, we define lim sup, oS, 10 be lim, (©) IF (4,22, diverges to minus infinity we write M,. lim sups,= — 2°. For example, let 5,=(~ 1)" (1). Then {5,}%., is bounded above. In this case M,=1 for every nEJ/and hence Tim .My= 1. Thus lim stp, ao Consider next the sequence 1, —I,1,~2,1, —3,1, — and so the limit superior of this sequence is 1 If ,=—n (WE 2), then M,=Lu.b.{—n, as n—vco, and s0 lim Sup, ..(—n)= — 2. 29B, DEFINITION. If (s,)_, is a sequence of real numbers that is not bounded above, we write lim SUP, 5, = 2. Obviously, lim Supy "= 22. At this point the Teader should verify the following statements. (I) If (5,) 2.) is bounded above and has @ subsequence that is bounded below by A, then lim sup, ...5, > Az (2) if (55) has no subsequence that is bounded below, then lira Sup, .05, =~ 20. We note that changing a finite number of terms of the sequence {5,). does not change lim sup, Thus the limit superior of the sequence 10! 1)—1,1,—1.1, lenis 1 2.9C, THEOREM. If (5,)$, is a convergent sequence of real numbers, then limsups,= fim 5 PROOF: Let L=lim,_,,.5,- Then given €>0 there exists N€/ such that In-L), L-e) Thus if n> N, then Leis an wb. for (syy5peiSpaz-e) and L—e is not an wb, Hence L=e0 there exists N, J such that HU. (SpSpenSseaJ—L). This implies SLE (n> Ny) o Similarly, since limint, ..8,=L, there exists Nj such that BLD. (5,.5:enseene-)—LI M4), which implies S>b—e (n> Ny) @) If N= max(¥,.N;), then from (1) and (2) we conclude Ig—Li). This proves lim, L, ‘There is a similar result on sequences diverging to infinity 291, THEOREM. If {5,}%.1 is a sequence of real numbers and if limsup,....5,=20 =limint, 5, then s, diverges to infinity PROOF: Since liminf,_ 95, BID. {5p Sp4nSpeaee}>M (n>). This implies that M is a lower bound (but not the gb.) or (syu5, ore)» 80 that 5,>M (n>), which establishes the required conclusion. =o, given M >0 there exists an N EJ such that ‘There is an obvious analogue of 2.91 for sequences diverging to minus infinity which the reader should formulate and prove. The converse of 2.1 is exercise 4 of this section. ‘We now prove a result for limit superior corresponding to 2.7E. 2.9). THEOREM. IT {5,]%-y are bounded sequences of real numbers, and if SS (NET), then limsup,. oS SlimSUp, of and liminf,.,..6, 6 liminf,.,.f 52 | SEQUENCES OF REAL NUMBERS PROOF: From the hypothesis s, <1, itis clear that Lub. (Sy5petsSpeaee+) SLB: { ole faa ae and BUD. (SyoSps Spe aseee } SBLD. (Syrtys totes arse?) (Can you prove this?) Taking the limit as n—>20 on both sides of these inequalities, and using 2.7E, we prove the theorem. Theorem 2.95, of course, remains true even if s,> 4, for a finite number of n It is not always true that lim sup (s,-+ 44) limsup s,+ limsup even for bounded sequences (5,2, and (14). For example, if s,=(— I)" (ne 1) and =I"! (ne), then 5, + 4,0 (nE 1). Here limsups, but limsup, .o(5,+%)=0. There are, however, important inequalities that can be proved, 29K. THEOREM. If (5,}$., and (1,)%.1 are bounded sequences of real numbers, then @ (b) limsup(s, +4) limi PROOF: (a) Let M,=LUb. (Sp 5pavSq030--)> P= 1U.b. {tyslyatvfaszres-}> Then KM, (>), Mm), and so the M,+P, — (k> 2), Thus M, +P, is an wb. OF (55+ lypSy 41 etiSena* feazree-)s 80 that Lub. (5,4 feSpe1t bea Seer frases) S My Py By 2.7E and 2.7A, lim Lub. (55-4 Searles weet feazeees}< fim (My + P,)= fim M+ fim, Py limsup (s, + 4,) 0, (a) s,<.M+¢ for all but a finite number of values of nm; (b) 5, > Me for infinitely many values of n. 2. If limint,_..5,= 7, then for any €>0, (c) 5,>m—e for all but a finite number of values of 7; (d) 5, 0, 3ve would have 5,0, 5, m+e (NPN), By 2.95, liminfs,> m+, which again contradicts the hypothesis. Thus (d) is true. Although we do not prove it, the converse of 2.9L is true, That is, if (5), is bounded sequence of real numbers and if M © R is such that (a) and (b) hold for every €>0, then limsup,_,..s,= M; similarly, if (€) and (d) hold for every ¢>0, then lim- ily Sy = Using 2.9L we can prove the following useful result. 29M. THEOREM. Any bounded sequence of real numbers has a. convergent sub- sequence. PROOF: Suppose (5,)%2., is a bounded sequence of real numbers and let M =limsup,_..5,. We shall construct a subsequence (5, )21 which converges to M. By (b) of 2.9L there are infinitely many values of m such that s,> MT. Let m, be one such value, That is, m € / and s,, > M~ 1, Similarly, since there are infinitely many values of n such that s,>M—+, we can find 1 =7 such that n>n, and s,,>M—$. Continuing then, for each integer k >I we can find n, €/ such that n, >m_, and 5, >M~ a Given €>0, by (a) of 2.9L we can find N 1 such that uSMte (NPN). @) Now, choose KEJ so that 1/KWN. Then, if k> K, we have 1/k N. Hence using (1) and 2), MneK), ‘54 | SEQUENCES OF REAL NUMBERS which implies -Ml K). This proves 5,,>M as k>20, which is what we wished to show. Exercises 2.9 1. Find the limit superior and the limit inferior for the following sequences. (@) 1,2,3.1.2,3123,.0 () (singra/2)) (©) ((+1/meosne ©) (U41/ my 2. Ifthe limsup of the sequence (s,)%., is equal to M, prove that the limsup of any subsequence of (5,}*, is limint 5 (Hint: Use 2.9L.) 2.10 CAUCHY SEQUENCES The most important criterion for proving that a sequence converges without knowing. its limit is called the Cauchy criterion, 210A. DEFINITION, Let (,}"2.1 be a sequence of real numbers. Then (s,}%.1 is ealled a Cauchy sequence if for any’€>0 there exists an NE such that Ise S1<€ (mn). Roughly, a sequence (s,}%., is Cauchy if s,, and s, are close together when m and n are large. First we show that a convergent sequence must be Cauchy. 210 cAUcHY SeQUENces | 55 2.10B. THEOREM. If the sequence of real numbers {5,}7_, converges, then (s,)%., is a Cauchy sequence. PROOF: Let L=lim,.,,,S- Then, given €>0, there exists an N EJ such that b-LI<$ (kN) ‘Thus if m,n > N, we have Is, 2 =KSn— L)+(L~5,)1 N). ‘Then tw Syl<1 (>). wo Hence, if m> N, we have 6a 50)+ 51S Iba Sl +5 and s0, using (1) [Snel < 1+ |syf (me). [sys then \sql< M1454) (MEL), so that (5), is bounded. If M=max(ls,). 2.10D. THEOREM. If {s,)$2.; is a Cauchy sequence of real numbers, then (5,)%., is convergent. Fist pRooF: By lemma 2.10C we know that limsup,_.§ and liminf, 5, are (inive) real numbers, By 2.9H, then, to prove the theorem itis sufficient to show that Jimsup s,= limint 5p Bur, by the theorem in 2.9G, we know that limsup,,.§, > liminf, 45: Thus all we need to prove is that limsup 5, < tim int, a 56 | SEQUENCES OF REAL NUMBERS Since (s,)., is Cauchy, given €>0 there exists €/ such that V5 SIS (m.n>N) and so. ‘ Inslc$ (n> N) Wi fotlwstat 4/2 andy —/ 2 ae rpetey, wpe and ewer Dono he Se (dyapivennn Hen if n> M syFe/2 and ay =e/2 ate upper and ower bounds Yor {ne wspsaene}e Ths implis, for n> N Sept laeasen ) GLub (sp spapSeaae) Sut § Since the left and right ends of this inequality differ by «, we must have Ueber en) BUM denen) Laub.{ 55,0 wSne21s2) SBLD(SySne wSgeaeee) FE Taking the limit on both sides and using 2.7E, we obtain limsups, 0! Since (5, )f, converges to 5, there exists J EJ such that Iy-sl<5 UY) a) Since (5,)%., is Cauchy, there exists K €/ such that Innsd<§ (mn> K), @ We may choose K so that K > J. Now suppose kE/ and k > K. Then kJ, so (1) implies Iansh<$ Also, nj > k > K, so (2) implies bem sal <5 Therefore In-sl0 there exists NjET such that |s,|<«/2(n> N,). If we let M s=max({4h]szh---lsy,- Dy then We have, for n> Ny, (Isl +++ +h5w al) + (swt os thal) 19 §—$—<—$<—<— IM +(n—N,+1)e/2 and hence MDM _» +$ (nem). a Now choose N31 so that eN>2(N;—1)M. Then (MDM Oe Se c$ (ne Ny. Q If N=max(N,,N;), then (1) and (2) imply Joul), and hence lim, .:!04|0. Thus lity ..0, 0, which is what we wished to show: This proves case I Case I, L#0. We have lim,_...5,=L so that lim, (5,— LYS, is (C1) summable to 0. That is, fim (4-L)=0 (C1), = L)=0. Hence by case I, neaSy which means Jig CTD HOt SHG) 4 a But Gir B+ (a- bt +6) ~Leo-L 60 | SeQUeNces OF REAL NUMBERS. Thus from (3) Jim, (,~L)=9, and so Jima b, This, by definition, implies lim,_..,5,= (C1), which proves case II. We have now seen that all convergent sequences are (C,1) summable (to their respective limits), and that the divergent sequence {(—1)")$_, is also (C, 1) summable. Not all divergent sequences are (C, 1) summable, For example, if 5,=n(n€), then {5,)%1 i8 not (C, 1) summable. For, in this case, Stayton ts, gF nL otte and so (0,)%.1 does not converge. In an exercise the reader is asked to show that no sequence that diverges to infinity can be (C, 1) summable. The sequence 1, ~1, 2, ~2, 3, ~3,... is an oscillating sequence. We shall show that it is not (C, 1) summable. [However, when we take up (C,2) summability, we shall see that this sequence is (C,2) summable.] For this sequence, ntl 2 (0= 1.3.5.0) (n Obviously, if m is even, then (5, +5,)+ (53+ 54)+ ++ +(5,-1+5,)= 0. Thus 20 (1=2,4,6,...). If n is odd, however, then n—I is even and 5,454 +++ +5445) stb ts 4,655.) Hence Sent 2n Since (n+1)/2n-+4 as n-00, the sequence (0,}., has the subsequence 4), a5, 05... converging to } and the subsequence 93, 04, 04... converging to 0. By 2.3, {0,) 2.1 is not convergent, and hence (5,}%-, is not (C, 1) summable. To keep the record even we give one more example of a divergent sequence that is (C,1) summable, In Section 2.3 we saw that if @ is a rational number in (0,1), then {sinner}, diverges. We shall show, however, that this sequence is (C1) summable to 0. For, from the identity 6, (n= 1,355, 08 Jx—cos(n+ J) which will be proved in Section 8.4, we see that 2On +--+ +sinnBe _ 608 }On—cos(n + 3)Or 7 OD nsin $0r o.= Sind ts and hence 1 \enl 20. Hence fim(-Y"=0 (2) In the last example, (C,2) summability gave the same “value” for ((~I)")2_, as did (C,1) summability but was harder to apply. We shall now show by example that a sequence may be (C,2) summable even if it is not (C,1) summable. Indeed, consider =1,2,—2,3,—3)..., Which we have already shown is not (C,1) summable. [Here in+1)/2 if m is odd and s,=—n/2 if n is even, Thus if n is even then n—1 is odd and @-)tt 2 62 | SEQUENCES OF REAL NUMBERS. and so rst (n= Vg + +255 Hy matty [n= (n=) + [(1=2)-2= (n= 3) 2] +2 #L2(/2)= (n/2)] aE Since n/2 is an integer if n is even, we have nn (af29(nf241) _ n(n+2) Lazo Z 2 snd hence at? PEA, rhb) If nis od, we have n=241 H-) ntt 2 Se 2 ry and so nsy# (n= Isto #352 # 251+ 5 n(n ty/2 (rcecne sof E54) (52)} 02H OnE (1424+ F912) + (n41)/2_ (w= D/SH(n41)/2 Gn a mtdnt3 (n+) (Wetify this for n=5, You should obtain 1+ 4s, 7 15 is * Thus t+} a8 2-100, and so 1, ~1,2,~2,3,—3,... is (C,2) summable to 4. Thus (C,2) can do some things that (C,1) cannot. The next theorem shows that anything (C,1) can do (C,2) can do also. 2.11E, THEOREM. If (C0), then lims=L (C2). 2.11 SUMMABLITY OF sequENCES | 63 cast I, L=0. We have lim, ,.,0,=0 where Stee ts We wish to prove that lim,_.r,=0, Now 154 (not iy ace ee (tate tay tetate ty, ttt eee +20,+ 0, +20,+ Tae Since a,-20 a8 120, given €>0 there exists jE such that £ n> N,). ni<§ (n> Mp) Let M=max(o;},Joah---[0y,-1)- Then, for n> Ny, [lol+2lo} +--+ +(1— Die, Inds 142+ M[1+2+ + (ND +(€/2)(N, + < Tsien and 0 MRM) ery (n> Np) ‘The remainder of the proof is exactly as in ease I of the proof of 2.11B. ASEH, L#0, We have lim, 8) £(C,1). Hence, by 2.11C, (C1). But then, by case I of this proof, lim,_,,.(s, — L)= 0 (C,2). That is, na n(s,~ L)+(a=1)(s.~ L) +--+ +(5,-L) J ko But, removing parentheses in the numerator of (1), we have lim (6 -D= 0. a nat (W—Nagt oo ("Saat -e ad or dim, (a 2)=0. Thus lim, L, which shows lim,_,,,8,= L(C,2), and the proof is complete. 64 | SEQUENCES OF REAL NUMBERS, 2.11F. coroLtary. If {5,)%., converges to L, then (s,)2, is (C,2) summable to L. PROOF: The proof follows direcily from 2.11B and 2.11E. 2.11G. Although we will not give details, we mention that the sequence 1, ~2,3, —4, 5,—6,... is not (C,2) summable but is (C,3) summable. For the record we give the definition of (C,k) summability for any kE 1. DEFINITION. Let (5,}%. be a sequence of real numbers, let k denote any fixed positive integer, and for nT let [leper a(n where {In summation notation, my rey 1 EC ‘Then {5,)@.; is said to be (C,k) summable to L if (A,)., converges to L. The reader should verify that the special cases k=1 and k=2 in this definition actually coincide with (C, 1) and (C,2) summability as previously defined. Tt may be shown that if k> I and (s,)., is (C,k~1) summable to L, then (5,) =, is (C,k) summable to L. Moreover, there will be a sequence which is (C,k) summable but not (C,k~1) summable. 2.11H. In general, the term “summability method” can be defined as a real-valued function T whose domain is a set of sequences. A point (sequence) is in the domain of T if T “sums” the sequence—that is, if the sequence is assigned a real number by T. ‘Thus the domain of (C,1) is a proper subset of the domain of (C,2). Since (C, 1) and (C,2) agree at a sequence where they are both defined, we may say that, from a function point of view, (C,2) is an extension of (C, 1). ‘The (C,k) summability methods are an important but very minute part of the class of summation methods. The various methods differ greatly in their ability to sum divergent sequences and in the ease with which they can be applied. However, we almost always insist on one minimum requirement for a summability method T, namely, that any convergent sequence be T summable to its limit DEFINITION. Let T denote a summability method for sequences [for example, (C.0,(C.2),---} Then T is said to be regular if, whenever (5,)%. converges to L, then {5,} 221 is also T summable to L. 2.1L THEOREM. (C,1) summability and (C,2) summability are regular. PROOF: The proof follows directly from 2.11B and 2.11F. 2.12 UMS FOR sequENcES OF sets | 65 Its interesting to note that if a method of summability sums too many sequences, it cannot be regular. Indeed, it may be shown that if every bounded sequence is T summable, then 7 is not regular. A famous. very general summability theorem will be presented in Section 3.12, Notes and Additional Exercises for Chapters 1-3. Exercises 2.11 1. Prove that the following sequences are (C, 1) summable. (@) 1,0,1,0,1,0,.. (b) 1,0,0, 1,0,0,1,0, (©) = 1.22, -1.2,2, ~1,2,2,. 2. IM sy.Syy--- is (C1) summable to s, and if 7 R, prove that f.5,.53S3--- is (C1) summable to s. 3. Prove that a sequence that diverges to infinity cannot be (C, 1) summable, 4, Let (5,)2, be a sequence of real numbers, ¢,=n~'(s,+5;++** +5,). Prove that if (5,) 2-1 is (Cy) summable, then lim,.,..(5,/n)=0. [Hint: Compute nd, ~(n~ ey] Deduce that 1,~1,2,~2,3,~3,... is not (C,1) summable. 5. Let (5) B.1 be & sequence of positive numbers with lim, Jim Woes, = 5, 5, where 5 >0. Prove (Hint: Take logarithms.) 6. If {5,}224 is a sequence of positive numbers and if limy .x(S/S,—1 Tim, geo! V5, = Le (Hints Let y= $4582" 52/ Spey exercise to (44) 21) 7. Without using (C1) summability, prove that 1,0, 1,0,1,0,... is (C,2) summable to }. 8. If {5,}%, is monotone, prove that (0,)_, is monotone where site +5 9, Prove theorem 2.11B by using the result of exercise 7 of Section 2.9. prove 54/51 ABpIY the preceding 2.12 LIMIT SUPERIOR AND LIMIT INFERIOR FOR SEQUENCES OF SETS 212A Suppose Ey,Ey,... are subsets of a set S. For each n€/ let xq denote the characteristic function of E,. Then, if x € S, the terms of the sequence (x,(x)) consist of O's and I's. Ivis then clear that either limsup, .,.4(x)=0 oF limsupy .4.X,(0)= 1, and similarly for liminf,, (x). We have the following theorem. THEOREM. Let (£,)%., be a sequence of subsets of a set S, and let x, be the characteristic function of B, (n€ 1). Let x be any point in S. Then (2) limsup, .0Xe(2)=1 if <= £, for infinitely many values of n, while tim SUP, .waXn(X)=0 if x E, for only a finite number of » Also (b) liminf, ,..X9(4)= 1 if x€£, for all but a finite number of values of m, while lim inf, ,..X(8) =0 if there are infinitely many values of n such that x@E,. 66 | SEQUENCES OF REAL NUMBERS PRooF: We shall prove (b). If x is in £, for all but a finite number of values of 7», then there exists NEJ such that xEE, (n>). Hence x,(x)=1 (n> N) and so Timi seXq (2) =Timy pq (2) Ie However, if there are infinitely many n such that £, does not contain x, then x,(x)=0 for infinitely many values of n. Hence =0 BUD. (XQ(X) Xue ooe for all n, and so liminf,,...X,(x)=0. This proves (b). The proof of (a) is left to the reader, tis then natural to make the following definition, 2.12B. periNimt0N. Let (£,}#., be a sequence of subsets of a set S. Then we define limsup, _..£, to be the set of all xS such that x is in £, for infinitely many values of n, We also define liminf,_.,.B, to be the set of all x€S such that x is in £, for all but a finite number of values of n From part (a) of the theorem it then follows that if x*(x) =limsup, .2x(x) (XS), and £*=limsup, ..£, then x* is the characteristic function of *. Similarly, part (b) of the theorem shows that if xa(x)=liminf, x(x) (ES), and E,=liminf, .E,, then 2a is the characteristic function of E,. ‘Briefly, then, the characteristic function of limsup,..E, is limsup, Xi and simie larly for limint. Exercises 2.12 1. Prove that, if (£,) 2. is a sequence of subsets of S, then Jiminf £,c limsup E, 2M E= = 8 and £,=E,= Ey =, compute limsup £, and timigt Ey 3. Let (6) fy be a sequence of subsets of . (@) Wr selimsupy cbs prove tht CUE. E; for every ne. (b) Prove that Timsup = 0) ( U 5) 4, (@) If xeliminf, .,E,, prove there exists mI such that xE M7. 4E x () Prove that mint Ex U ( A 5) 5. (@) I E\CE}CE,C ++, prove that 2.12 UMS FoR SEQUENCES OF sets | 67 (b) If £9. £3 £3 ++, prove that Ae, limsup E,= limint &, 6, Give a definition of lim, ...Ey 7. IF E, denotes the closed interval [7,2], find lim,.,,,£,. What can you say about the length of E, as n—+20? Do the answers to the first two parts of this question agree with, your intuition? SERIES OF REAL NUMBERS 3.1. CONVERGENCE AND DIVERGENCE We recall that the sum of the infinite series a,+a,+--++a,++- is defined as fim, (a+ *** +4,), provided that the limit exists. This, However, is the definition of the’swm of an infinite series and is not the definition of “infinite series” itself. Like the term “ordered pair,” the term “infinite series” is a highly intuitive one whose proper definition is not very illuminating. ION. The infinite series Ba, is an ordered pair ¢(a,)2.1-(%¢)2=1> where (4,}2., is a sequence of real numbers and Sarat ta, (nel) ‘The number a, is called the mth term of the series, The number s, is called the nth partial sum of the series. In addition to 32. ,a,, we sometimes denote a series by a,+a;++--ba,+-+> oF simply by a+a,+ "=". Thus the nth-partial sum of the series [—1+--- +(—1)"*! t+ is 1 if'm is odd and 0 if m is even. It is often convenient to index the terms of a series beginning with n=0. That is, we write some series as Edy (In this case we let 5,= dy a) +++ + dy) Thus the series T+ x+x74++++ can be written 3% gx". It is always trivial to verify that any definition or theorem about series written 3% ,a, has an exact analog for series written 3% ga, oF ¥Z.,a, for any integer p > 0. We shall not further belabor this point. ‘The ‘definition of convergence or divergence of the series 3%_,a, depends on the convergence or divergence of the sequence (5,}%., of partial sums. 3B. DEFINITION. Let EZ_,a, be a series of real numbers with partial sums s, ay ++++ 4a, (nEA). If the Sequence (5,)%, converses to A ER, we say-that the series BEL a, converges to A. If (5,) 2.5 diverges, we say that B22, diverges. “ 3.1 CONVERGENCE AND pivencence | 469 If D2 ,4, converges to A, we often write D2 a= A. Thus we use 3.4, not only to denote’a series, but also (in case the series converges) its sum. With this Warning we leave it to the reader to convince himself that no ambiguities arise. From theorems on convergent sequences follows the next result B.C. THEOREM. If ZZ,a, converges to A and 2. ,b, converges to B, then the series F(a, +2,) converges to A+ B. Also, if cER, then B2_,ca, converges to cA, PROOF: If symqy +++" +a, and t= b,+ +++ +b, then, by hypothesis, lim, 5,4 and lim, .of4= 8. But the mth partial sum of BF_,(a,+5,) is (a,b) +" $(a,+6,) =5,+ fj which, by 2.7A, approaches A+ B as n->a0. This proves B2.\(a,+0,)= 4+ B. ‘The second part of the theorem follows from 2.7B, An obvious consequence of 3.1C is that B22. (a,— b,)= A ~ B. ‘The following theorem gives a necessary (but not sufficient!) condition that a series be convergent. 31D. THEOREM. If Da, is a convergent series, then lim,_,..d, "0. PROOF: Suppose E_,a,=A. Then lim,.,=A where s,=a, +--+ +a,. But then lim, =A, Since a,=5,—5,-, we have, by 2.7D, lim,.a,=lim, ..5,— lim, A—A=0, which is what we wished to show. Thus we see immediately that the series D2 \(I—n)/(I+2n) must diverge. Here, 4,=(1~n)/(1+2n), and 0 lim,...4,= — $0. Thus by 3.1D, 32_,a, cannot be con- vergent. Similarly, the series 3% (—1)* must diverge since lim,_..(-~ 1)" does not even exist. We emphasize that the condition lim... 0 is not sufficient to ensure that 24 be convergent. In the next section we will see that 3% ,(1/n) is not convergent even though lim,_.a,=lim,__(1/n)=0. Exercises 3.1 1. Prove that if a, +4,+-+- converges to s, then a, +a, +--- converges to say. 2. Prove that the series B= {1/n(n-+1)] converges. (Hint: Write sien ntl) in tT and compute the partial sums of the series] 3. For what values of x does the series (I~ x) +(x~ x3) +(x?7—x°)+ ++ converge? 4, Prove that the series (a,~ a3) (a,~ a3) +(ay~a,)+ “++ converges if and only if the sequence {a,)7-y converges. 5. Does the series B=. log(1 + 1/n) converge or diverge? 6 Prove that for any a,b R the series a-+(a+b)+(a+26)+(a+36)+:++ diverges unless a= b=0, 1, Show that D7_a, converges if and only if given ¢>0 there exists V €1 such that za Ke (n>m>N), 70 | senes oF REAL NUMBERS 8. Prove that if a,+a,ba,+-+> converges to 4, then 4(a,+4,)+ (a+ 43)+ (ay aq)+ +++ converges. What is the sum of the second series? 9. Does B(n+1)/(n+2)] converse or diverge? Does > n+l S042) converge or diverge? 10, Show that if a,+a,+4,+--+ converges to L, then so does a,+0-+a,+0-+a,+ 04 -+-. More generally, show that any number of 0 terms may be inserted anywhere (or removed anywhere) in a convergent series without affecting its convergence or its 1, Prove that if 32,4, converges and B26, diverges, then D(a, + 6,) diverges. 12, Let Bf_,a, be & convergent series, Let (7,} 2, be any subsequence of the sequence of positive integers. Finally, let Bem dy verte tay (KET). Prove that 7b, converges and has the same sum as 2a, 13. Verify that the preceding exercise yields the following important result. If S2_,a, converges, then any series formed from ¥%_,a, by inserting parentheses [for ex- ample, (a, + 43) +(a, + -> + ay)+(--~)---] converges to the same sum. 14, Give an example of a series 3. ,a, stch that (a, + a3)-+ (a, +.a,) ++" converges but 4,44, +a,+a,t--~ diverges. (This shows that removing parentheses may cause difficulties.) ‘3.2. SERIES WITH NONNEGATIVE TERMS ‘The easiest series to deal with are those with nonnegative terms. For these series, all theory on convergence and divergence is embodied in the following theorem. 32A. THEOREM. If EZ. y4, is a series of nonnegative numbers with 5, =a,+--- +a, (nD), then (a) EF,a,, converges if the sequence (s,)%., is bounded; (b) B22, diverges if (5,}., is not bounded. PROOF: (a) Since a, >0 We have 541d) 42+ dy + ayy 1 Sy + dyes > Sy Thus (5,) 1 is nondecreasing and (by hypothesis) bounded, By 2.6B, {5,}"%.; is convergent, and thus 32. a, converges. (b) If (s,}e21 is not bounded then, by 2.5B, (s,}, diverges. Hence so does 3° ‘We now give two important’examples of series with nonnegative terms. The first is the geometric series I++ x24 3.2B. THEOREM. (a) If 0<.x<1, then BZ_ox" converges to 1/(1~x). (b) If x> 1, then 3 9x" diverges. 32 SERIES wind NONNEGATIVE TERMS | 71 PROOF: Conclusion (b) is an immediate consequence of theorem 3.1D since, if x> 1, then (x"}2., does not converge to 0. To prove (a) we have s,='I+x-++-- +x" and so is (net) x =e But if O0(n-+2)/2. Thus (5,)2-, contains a divergent subsequence and hence, by 2.3D, diverges. This proves the theorem. ‘We repeat that the divergence of the harmonic series shows that 27_,a, may diverge even if lim, ..2,=0. 32D. For series with nonnegative terms only we introduce the following notation, If Df. ,q, is a convergent series of nonnegative numbers, we sometimes write E2_\a, 2s,,. (This is possible since by bypass (fs ds wing) Now fy onde Hee [msi Sa) + (Sne2—Snet* + See SH) Euas ‘Thus for any mG/ there exists mJ such that eee ‘The partial sums of the series E7-(5,1~5)/Se4 thus do not form a Cauchy sequence and hence ST (See Exercise 7 of Section 3.1.) But 5,41 5¢=dp 44. Thus Let & = 1/5. Then ¢,~+0 as kse0 and Bf_.¢4,= 00. This completes the proof Exercises 3.2 1. If 3a, is a convergent series of positive numbers, and if (a,}*2, is a subsequence of (4,} 22, prove that 372 ja, converges. 2. Prove that converges, 1604, <1 (1>0) and if 0 x0 (m1), prove that there exists a series Ba, with a, 30 (kJ) and pti ta, (nel). 5, Prove that 1+}+]4+4+--+is divergent 6. For what values of x R does the series ae) (a) converge, and what is its sum? 3.3. ALTERNATING snes | 73 3.3 ALTERNATING SERIES An alternating series is an infinite series whose terms alternate in sign. For example, the series I= f4}—PHoo, [-243—44-0+, Loft f= dee are all alternating series. An alternating series may thus be written as Z7_,(—)'*"4, where each a, is positive for as E*_,(—1)%a, if the frst term in the series is negative]. We now dem- onstrate the fundamental result on alternating series. 33A. THEOREM. If (4,}2., is a sequence of positive numbers such that (8) a, > 4,2 ++ >a, 2.4412 *- (that is, (4,)%.; is nonincreasing), and (b) im, ad, 0, then the alternating series E., (—1)"*'4, is convergent. PRooF: Consider first the partial sums with odd index s,,5y,85,.... We have $5=5)— a,+ ay, Since, by, (a), asd, this implies sy <4). Indeed, for any mes we have Saget Sag ay agg y S Sap ye THUS $)> 5% °° F Sap % Saya 2 21780 that (in-i)Ge1 iS noninereasing, But sy,—,= (0) ~ 43) (dy~ a9) °° (day -9~ aq) + yyy. Since each quantity in parentheses is nonnegative and day >0, We RAVE S391 >O. Hence by 2.6E, (55,1) i8 convergent. Similarly, the sequence 53.54 .---,Sgyo-. 8 convergent. FOr S29 42 %2q day 41 aqe2> Sour ANd $0 {34} 224 is nondecreasing. But Also. $54 4, —(03— 04)" > ~CBya yyy 3) gg, THUS Syq 2, 80 that (8,)2ay 18 bounded above. Now, let M=lim, ‘Vand let L=lim, 18q. Then since a, 53q— Say—1 we have, by hypothesis (b), tim a, L-M. Thus L= M, and so both (55); and (‘9~4)F1 Converge to L. From this itis easy to show that {5,)%., converges tO £, and hence that Z5_,(— 1)"*"a, is convergent t0 L, which completes the proof. Note that the proof shows that s,.., > L and $3, < L. Thus 0€ S341 LSay-1— Say yyy And $0 [Sq — L] L, and so we ean conclude 0.6456< L < 0.7456.) Actually, it may be shown that L=log2=0.6932-- If 00 and lim, 3.3A apply?) 5. Show that 32_(— 1)'*"n/(Qn=1) diverges. Why doesn't theorem 3.4 CONDITIONAL CONVERGENCE AND ABSOLUTE CONVERGENCE We saw in the preceding section that the series ay and the series a ese te 2) 3.4 CONDITIONAL CONVERGENCE AND ABSOLUTE CONVERGENCE | 75 both converge. However, these two series differ in the following respect. If we take the absolute value of each term in (1), we obtain lttpeteen 8) which converges, whereas if we take the absolute value of each term in (2) we obtain Left pepee 4) which diverges. This leads us to the following definition, which divides convergent series into two classes. 34A. DEFINITION. Let B29, be a series of real numbers, (a) If BZ. ,|ay| converges, we say that D2 ya, converges absolutely. (b) If B24, converges but D5. |a,| diverges, we say that Da, converges condi- tionally. ‘Thus the series (1) converges absolutely while the series (2) converges conditionally. ‘We must justify the use of the word “converges” in the phrase “converges absolutely.” ‘This is done in the following theorem. 34B. THEOREM. If Df \q, converges absolutely, then E22, converges. PROOF: Let s,=a)-+ +++ ++a,. We wish to prove that {5,)2_, converges. By 2.10D, it is enough to show that {5,)%, is Cauchy. By hypothesis 22 \Ja,|). Inet ees + yl < ld il -** +I Ise SIS (mn >), This proves that (s,)., is Cauchy, which is what we wished to show. But (if m>, say), [Sq Hom fle Thus 34C._ If we separate a series 22,2, into the series of positive a, and the series of negative a,, we can show up an important distinction between absolutely convergent and conditionally convergent series. More precisely, if 3% ,a, is a series of real numbers, let Pr=a if 4,>0, 20 if 4,<0. [Thus for the series I=} +$—-++, p)= 1, s= $s Pag =1/@n—I), while p= =0 Similarly, let Ga if a, <0, qn0 if a,>0. The p, are thus the positive terms of EZ. ,4, (along with some 0's) while the 4, are the negative terms. It is easy t0 see that Pa=MAX(4y,0), 9, min(a,.0) 76 | SéRtES OF REAL NUMBERS, and hence by L4D, 2p," d,+|a,| and 24, 4,~\4]- o Also, = Pa es It is now not difficult to prove the following interesting result, THEOREM. (a) If Boyq, converges absolutely, then both B#.4p, and Bq, con verge. However, (b) If Ba, converges conditionally, then both Finally, (© 3217, and Z_.g, both converge, then BZ. ,a, converges absolutely. aap, and B19, diverge. PROOF: (a) If E=_.d, and Ela) both converge, then, by 3.1C, so does 3 (a, + la). Thus from (+), 3" ,2p, converges. By 3.1C again, this implies the convergence of DF. py, The series B29, may be proved convergent by similar reasoning. (b) We now assume that Ba, converges but that 22. ,|a,| diverges. From (+) we have |a,|=2p,~d,, If B5.4p, converged, then, by 3.1C, s0 would 3 er. -ay= Sta contradicting our assumption, Hence 3 ,p, diverges. Again, 32_1g, may be handled in the same way. (©) Since Pp= (ap laq))/2 and q,=(a,~lag))/2 we have la,|=P,~ dq. Hence if ZEsiip, and ZF_,g, both converge, then so does ZP_,la,), which shows that E2.ya, converges absolutely. ‘Thus since 1-}+}-J+-+» is conditionally convergent series, it follows that 1404}+0+}+0+4--~ diverges, and hence that 1+}+4+--- diverges. The last theorem tells us, roughly, that an absolutely convergent series converges because its terms are “small” while a conditionally convergent series converges because cof “cancellation” between its positive and negative terms. Exercises 3.4 1. Classify as to divergent, conditionally convergent, or absolutely convergent: rod 1 @intep-fee. (b) =f tb —fte> © f-}Hi- dee (@) I-1+}-d4$-d4e, a aca Ol ati pti ta yt 2. Can a series of nonnegative numbers converge conditionally? 3, Prove that if 32. la,|<00, then [S%. a) < 22a 4. IF 3%. ,a, converges absolutely, and if <= #1 for every nEJ, prove that D2. ,eay converges, 5. IEP. ea, converges for every sequence (¢,) 2, such that ¢,= +1 (ne), prove that Zea, converges absolutely. 3.5 REARRANGEMENT OF seRiES | 77 3.5 REARRANGEMENTS OF SERIES 35. Roughly speaking, a rearrangement of a series B2_,a, is a series 22,6, whose terms are the same as those of D%_4, but occur in different order. (A precise definition of rearrangement is given later in this section.) We shall see that rearranging an absolutely convergent series has no effect on its sum but that rearranging a conditionally convergent series can have drastic effect. ‘We have seen that the series B2_(—1)"*!/n converges conditionally to some LER (where we have stated but not proved that L=log2). In addition, we know 0.6< L<08 so that L0, We have ay By 3c, HLedndtinbtbokte, and so, certainly, 4L=0+4}-0-4404+4-0- Q) If we then add (2) to (1) we obtain, again by 3.1C, y= (1+0)+(S$+5)+($-0)+(t 1 1) (4 +(540)+(St+5)+(5+0)+( Balt }-$tbe bo dtpednde ° ‘The series on the right of (3) is a rearrangement of the series on the right of (1), but they converge to different sums! 35B. Indeed, we can find a rearrangement of ZZ_,(—1)"*!/n that will converge 0 ‘any preassigned real number—say, for example, 512, From 34C we know that 1+$4}4--> diverges, By 3.2A the partial sums of this series must be unbounded. ‘Thus 1+ $+ $+--- +1/N will be greater than 512 for all sufficiently large odd integers LN. Let Ny be the smallest odd integer such that 1 1 styl. lege bee >512. Then et an lege gt ty Sz (why?). Now let N, be the smallest odd integer greater than N, such that 1 1 Ayty., lepedeet Then ea 1 lege gee tae Continuing in this fashion we may construct a rearrangement of B2_,(—1)"*!/n that converges to 512. You supply the details. Now let us define “rearrangement” precisely 78 | semes oF neat NUMBERS 3.5C. DEFINITION. Let N= {n,}s%; be a sequence of positive integers where each posi- tive integer occurs exactly once among the 7,. (That is, N is a 1-1 function from I onto 1.) If E2. ,g, is a series of real numbers and if baa, (FED, then B72), is called a rearrangement of 22.4, If A=(a)@, and B=(b,)7%,, then, in the definition 35C, we have A°N= B. If (mJ) is the inverse function for N, then (by Exercise 12 of Section 1.5) BeN~"'s0 that a, by. This shows that 37,4, is also a rearrangement of D-yb, if EP, is @ rearrangement of D4 Tn the exercises the reader is asked to supply the proof of the following theorem, which consists of an imitation of the method in 3.5B. N 3.5D THEOREM. Let 27_,a, be a conditionally convergent series of real numbers. Then for any x€ R there is a rearrangement of D2 ya, which converges tox. For absolutely convergent series the story is entirely different. We first treat the case of a series of nonnegative terms. 3.5E 12a. If BZ_,a, is a series of nonnegative numbers which converges to AE R, and B7_,b, is a rearrangement of Z_,a,, then 27. ,b, converges and 32 b,=A. PRooF: For each NE/, let sy (mJ we have + by. Since 6,=4, for some sequence Let M=max(n,...smy). Then, certainly, sy +1ar<( 3 1a} Sl) ‘The sequence of partial sums of Efagley| is thus bounded hove, and hence 4 ee 320, 4 That is, without regard t gor 80 | sees OF REAL NUMBERS, ‘The foregoing inequalities also show the absolute convergence of the series gba + doh, + a,by+ dgbs + a,b, + aby + aghs +> a (whose sum is 27.96). By 3.5F we may rearrange the terms in (1) to obtain® © = [dabo] + {401+ abo 4151] + [agbs+ axby+ ayb,+ a,b, + a,b.) +> @) Inside the nth bracket (2=0,1,2,...) on the right of (2) are all products a,b, where either jor k is equal to n and neither j nor & is greater than m. Let us examine the sum of the terms in each bracket, If Aya agtayt- +a, and B, , fot byte thy we have dgho= AoBor Aaqby + bo 4b, = (dg 4,)(bo* by) ~ agbg™ A By ~ AoBoy gb + dghy+ aby dyby + ayb,™ (ap a, + 4;)(Dg+ B,+D;) — (0+ 44) b0 + 6,)= A2By— ABs, and in general, for n>1 the quantity in the nth bracket on the right of (2) is equal to A,B,~ Ay By The Sum of the first m brackets on the right of (2) is therefore [AB +[4,B)— ApBol+ «++ +14,B,— Ay By-i]= 4,8, which approaches AB as n-s0a. The right side of (2) is thus equal fo AB and the proof is complete. 35H corottary. If for some x= R the power series D2. absolutely convergent, then x" and 32 (SerSee}- So 0 =o where 6, = 2% not by pRoor: Let 4 <4 Then, by 356 (S[Se)3< % Ge SAB Equation (1) thus follows from (2). where 443" ax" XS aby Exercises 3.5 1. Prove that if |x|<1, then De tt St bE IE * See Exercise 13 of Section 3. 3.6 TESTS FOR ABSOLUTE CONVERGENCE | 81 2. Prove that if a+, 0y4+++~ is absolutely convergent, then a +a;ay4=-- =(ay+ yay») (ay + 04:+ag+ **) Ts this tru for all conditionally convergent seis? 3. What, if anything, is wrong with the following? Inde f-d edge sIH(L-I)Fd+(G- HEH (EH +> MULES ELE E HR) Leda do a(t bebe )-(l4 dtd) =0. 4, Show that there exists a rearrangement _,b, of |-}+4—444—++-such tha, if ta by toe + By, then limsup‘,= 100, timint 1,= — 100 Show that any conditionally convergent series has a rearrangement that diverges. Prove theorem 35D. Write the following products in the form 3%. 9¢,x": (2) Bfaots"EF.0%"), (©) Sex" BzG— 1%") 8. If 0<.4<1, prove that (ra}=Speone = 9. Let L= 32. {(~1)"*]/n. Show that each of the following seties converges to the indicated sum: ( I-f-d4}-d- bt bt (b) bebe }—g-d-daedage 3.6 TESTS FOR ABSOLUTE CONVERGENCE In the last section we discussed the behavior of absolutely convergent and condition- ally convergent series in general, In this section we take up methods (tests) used to decide whether or not a specific series converges absolutely. 3.6A, DEFINITION, Let 37,2, and E216, be two series of real numbers. We shall say that E%. 4, is dominated by* D2. 6, if there exists NE/ such that lal < [by ("> N). (That is, Jag] <|0,] except for a finite number of values of »,) In this case we write Dac D by For example, 2. (—1)'/m?-<22_,1/Qn+1) since |(—)/n"|<1/Qn+1) for n>3. Also 100+ 4+ 4+" +is dominated by 1+ 4+4 +--+. (This shows that 22. ,ay-€22_ by does not necessarily imply 32_,a, N. Hence if s,=]a,|+ +++ +a, we have, for n> N, Se Slal tes + Lay] + lard tos + byl < lay] ++ #layl+ Me The sequence of partial sums of 3'7_,|a,| is thus bounded above and the theorem follows from 3.24, ‘Theorem 3.6B is called the comparison test for absolute convergence since it involves term-by-term comparison of E,| and E72 |é,- It is the basis for the other tests in this section 3.6C. From 3.6B, it follows immediately that for any x€(—1,1) the geometric series ES 1x" converges absolutely, For Dx"« Z|" m0 mmo and the series on the right converges (absolutely) by 3.2B, We emphasize that theorem 3.6B deals only with absolute convergence. Note that the series 22_,1/n is dominated by the conditionally convergent series 2_(—1)"/n but the former series does not converge at all. The concept of “dominated” is usable only in connection with absolute values. ‘The following result may be deduced from part (b) of 3.2A. We omit the proof. 36D tHeorEM. If E7\a, is dominated by E21, and BP sla,|=o0, then DP. byl <0. (That is, if 27. 4,€25_ 6, and BP. jfay|= 00, then Be |B,|= 00.) For example, consider D2, =D2_,1/(2n+5). This series dominates E2_,1/3n, which diverges. Hence 2¢_,1/(@n+5) diverges. We now give the first important consequence of 3.6B. 3.6E, THEOREM. (a) If Z_,b, converges absolutely and if lim, EP.1a, converges absolutely (b) It Be. jlayl= oo and if tim, ...]ay|/Iby| exists, then 32 )b,|= 0. la,1/1q| exists, then PROOF: (a) By 2.5B, {|a,/by|)%.y is bounded. Thus for some M >0, lal Mlb (net), ‘This shows that Zsa, is dominated by the absolutely convergent series E2_,Mb,, By 3.6B, BF al <0. (b) Avin the proof of (a) we have Ja,|1 then 32a, diverges; (©) Hac 0 so, by 29L, there exists N EJ such that N) Then [ay 41/ay1 < B, [ay 42/4y 41 0 we have similarly wen |_| Aven vst e pe re ay Thus layeel, then by 2.9L, |ay.1/,|>1 for all n> N (for some NEI). But then layl1, while if L=1, we can conclude nothing. Here are some examples to illustrate 3.6F. Consider first 32 n"/n!. Here a,=n"/n! so that lash tM gt OHI" ay Tal ae na 7 2G But by 26C, lim, (1+ 1/ny"=e>2 and so . In particular, @>2 so that ‘BEL in"/n! diverges. These computations show that for the series E2_yn!/n" we have A=1/e<} <1 50 that B_yn!/n" converges. Consider next the series 2 9x"/n! for some xR, Here we have Va, |" Geri Isl nat Thus lim,_..4y.e1/4|=0, which shows that the series in question converges absolutely for any real x. (Recall from calculus that the sum of the series is e*.) Finally, let us try to determine the values of x for which Z.,x*/n converges absolutely. For this series we have lim, ..[a,..,/a,/=|x|- Thus the series is absolutely ‘convergent for [x|<1 and is divergent for |x| >. The ratio test fails if |x| = 1— that is, if x= 1 or x= ~, But for these values of x the series becomes 22 ,1/n and 2_,(—1)"/ zn, neither of which is absolutely convergent. Thus 3%. x"/n is absolutely convergent only for ~1. (This includes the case limsupy.. Wia,| <0) IEA=1, the test fail PROOF: If A). This implies |a,|< B"(n > N). Thus D2. [a4] is dominated by E¥_,B", which is (abso- lutely) convergent. By 3.6B, 32 ,|a,| <0. This proves (a). If limsup Wfay] > 1 then, by 29L, W/{a,| >1 for infinitely many values of m. But this implies that |a,|>1 for infinitely many n, and so (4,}-1 does not converge 0 0. By 3.1D, E24, diverges. This proves (b). 6 TESTS FOR ABSOLUTE CoNveceNce | as Note that for the divergent series 3:2 ,1/m and the series U_y1/n? (which will be shown to converge) we have lim,_,,, W/ja,| =1. [Remembering from calculus that lim, ..(logn/n)=0, we have lim Vi7m im e(-Han/, Thus also, Aa V7? ~ km (70 Here is a corollary on power series Ba” 3.6H. THEOREM. Let (4,) 2.9 be a sequence of real numbers. Then (a) If limsup, .., W/la,| () If limsup,.,.,W lay] =L>0, then E%o9,x" converges absolutely for |x| <1/L and diverges for |x|>1/L3 ©) H limsup, .-, Wia,] =20, then EF.gagx” converges only for x=0 and diverges for all other x. proor: We have Wja,x"] =[x|Waj] Thus if limsup,_., Wla,| =0, then, for any xlimsup, ig Vla,x"| =[x|-0=0 and by (@) of 366, Bf9¢,x" is absolutely convergent. This proves (a). To prove (b) we have limsup, .., Wla,x"] =|Lx|. Again by 3.66, D% 9a,x” will converge absolutely if |Lx|<1 and diverge if |Lx|>1. This proves (b). The proof of (c) is left to the reader. the series EP ga,” converges absolutely for all real x3 In any of the three cases in 3.6H the following is true. 3.61. COROLLARY. If the power series 37a," converges for x= xp then it converges absolutely for all x such that [x|<[xol- ‘The following theorem shows that if the ratio test works, so will the root test. 3.6). THEOREM. Let (a,}. be a sequence of nonzero real numbers. Then limsup Wag] timint |=" @ Hence if the ratio test implies 3 |a,|-<00, so does the root test, and, if the ratio test implies E2. |4,|= 20, so,does the root test. PROOF: We will prove (1) If Timsup,..1dq.3/(a)|™ 20, then (1) is obvious. Suppose limsup,... where 4 ER. Then by 29L, if €>0, there exists NE such that la,41/(4,)= A KAte (n>N), 86 | SERIES OF REAL NUMBERS Thus lay 411 <(4 + Olay), lay42l $(4 + Olay ni] $(4 +6) ay. Indeed, if n> N, we have lay ain=ny) (A +6)" “Jag Thus ule B(Ate" (n>) where B= |ay|/(A +6)", and so Wa, 0 prove that the series 1~ 37-447 =... and x37 +3 —.., converge absolutely. Using theorem 3.5G find the first few terms in their product. Deduce sin2x=2sinx cos. 5. (a) Does the ratio test give any information about the series QP +G)' +O +GF HG) +2 (6) Does the series converge? 13.7 SERIES WHOSE TERMS FORM A NONINCREASING sequence | 67 1 © 2 teen 2 (1+1/n)* 69 see 12, Show that EB x"/n" converges for all xR. 3.7 SERIES WHOSE TERMS FORM A NONINCREASING SEQUENCE ‘The tests of the previous section fail to give any information about the important series 2_,1/n?. This series has the special property that its terms form a nonincreasing sequence. Such series are often treated by the integral test familiar from calculus. However, since we have not talked about integrals as yet, we use another very interesting test called the Cauchy condensation test. RIA. THEOREM. If (4,)%., is @ monincreasing sequence of positive numbers and if EPL gay converses, then 32a, converges. Roof: We have a, 2a, ast agt a; ag > Say, and in general Aygo Hage > Zane = (2 ages), so that a The remainder of the proof is left as an exercise. Note the similarity to the proof of 3.2C. 3.7C. conoitary. The series B%.,1/n? converges. /n? we have 3 0 PRooF: For ay, Hence by 3.7A, Note that for 32. 1/n we have a,=1/n, and so 32.2%; the divergence of 52. 1/n follows from 3.7B, ‘The series 37. 41/(logm) diverges. For here a,=1/(ilogn) and so Br | 5 2 ge Sle} which diverges. By 3.7B, E7_41/(nlogn)= so. The series 32 ,1/In(logn}'] converges. For DP 2(1/2")= 00. Thus < = Moye 1 EP ae I he terms ofa convergent seis orm anoninresing sequence, they must approach zeto "fase" than I/n Ths teal caled Prinsheins theorem (litugh was erisinaly cscovered by Abel, we now present 3.7D, THEOREM. If (4,}%.) is a nonincreasing sequence of positive numbers and if Za, converses, then lim, m4, =0. PROOE: Let sy ay++++ tay IPERS aye, then * has 4m di sn = 0. Now Ope t gga t May 2 dag tag 1 dag, Thus limy sa Sa9~ 3.8 SUMMATION By rams | a9 and 50 0 ty, < 535 Sy. Thus lim, dp, =0 and 80 Tim 2na,,=0. « But any¢1 dq: Thus Qn Daye <( 252") ena) By) ‘lim, n+ 1), 4120 @ ‘The conclusion of the theorem follows from (1) and (2). Theorem 3.7D is no longer true if we drop the hypothesis that (4,)%-, is noninereas- ing; consider the series 2_,a, where (n= 1,4,9,16,...), aon, itnis nota petet square Then ie teat fer sate dteetadee ade es The partial sums of B2_,a, are thus bounded above by 1,iyt wal (sepebe= bobo) which, in turn, is less than 2 BZ 1/n?, Thus Zeya, converges. But na, does not approach zero as n—+c0, since na,=1 whenever n is'a perfect square, Note also that the converse of 3.7D is not true. That is, there exists a nonincreasing sequence of positive numbers (a,)%., such that lim,_,..22,=0 but such that 32a, diverges. Indeed, let a,=2 and let a, =1/(nlogn) for n>2. Exercises 3.7 1. For what values of x does 31/1" converge? 2. For what values of x does 2_,1/(n(logn)*] converge? 3. Prove that for any real x the series 31 /(logn)* diverges. 4, (a) If the terms of the convergent series 2 ,a, are positive and form a nonincreasing, sequence, use 3.7B to prove that lim, 2" (b) Deduce another proof of 3.7D. Use 3.7D to give another proof of 3.2C. 3.8 SUMMATION BY PARTS 3.8A. THEOREM, Let (a,)%., and (b,)%, be two sequences of real numbers and let S,=ay+ s+ +a,, Then, for each ne I, = & slbevi- be) a 90 | semes OF REAL NUMBERS, ——r— B eyty= 3 balsa. b 59) 662 5)) 4 +p Beer Se-a BGS = (by= by) +90, D 5e(be21~ O4) + S:Pa~ SP est Spur » by) 55 1p =P) Say == Silber Wt bese which proves (1). If we introduce the notation Aa,= a, ,,~ 4; for any sequence (4,) f=, then +1) resembles a “derivative of a, with respect to k.” The formula (1) becomes Dd FAS 1 Sider D Ab, Q) which resembles the formula [bds= so] - f%sao of integration by parts. The formula (1) is thus sometimes called summation by parts. ‘A consequence of 3.8A is the following result called Abel’s lemma. It, in turn, yields a new test for convergence and, in a later section, a theorem on summation of series, 3.8B, avet's LEMMA. If (a, is a sequence of real numbers whose partial sums 5, =D] 1a satisly més0 we must find N’€/ such that > ab m>N). Now, by hypothesis, there exists MM >0 such that s,|<.M (n€/). Hence for any m,n, 24 15+ B= and so. Salem (mnenmen). By 3.8B (applied to (4,)Z and (4,7?) this implies > aby But, by hypothesis, there exists N'EJ such that 6, N),. Hence 2Mb, N) so that <2Mb, — (mnEl:men), > ab m>N), which is what we wished to show. 38D. For example, from the identity cee ea wwe see that if sinx/2#0, then the partial sums 5, = 23. simkx satsify [sy] <1/Jsinx/2l Thus the series > one > both converge for all real x. (For (1/n)2_, and (1/logn}%.y are nonincreasing and converge to zero.) A somewhat different test, called Abel's test, can be obtained by assuming more about ZF ia, but ess about (6,}72 2sin% (sinx-+sin2x +++ +sinmx) = cos Togn 3.8E, THroREM. If Df,a, is a convergent series of real numbers and if the sequence (4,) #21 is monotone and convergent, then ¥#._,a,b, also converges. PROOF: Let us suppose that (b,}%., is nondecreasing and let c,=b—, where My bye (IF (b,}2.4 were noninereasing, we would let c=5,—b.) Then ¢>0, 92 | sens oF neat NUMBERS lim, soGq=0, and (c,}2u1 is nonincreasing. By 3.8C, then the series E%_,a,c, converges. By BIC, the series S2_,ba, .converges. But aqb,=ba,~dgcq. Thus D2 a,b, (6a, — a,¢,) converges (again by 3.1C), which is what we wished to show. For example, the series I-1ep-gap-peee converges, whereas the sequence 0,4.1,3,3.4.4 ‘Therefore, by 3.8E, the series is monotonic and convergent. 21,3 21,3 poate ‘must converge. Note that theorem 3.3A does not apply to this last series. (Why?) Exercises 3.8 1. Prove that if E%_,na, converges, then so does 2 2. Given that osx +eos3x-+e0s5x-+ +++ +e0s(2n=I)x= if sin.x40, prove that B2_,cos(2n~ I)x/@n~ 1) converges if x is not a multiple of 7. Prove that 3¢_,(1/n)log(I+1/n) converges. (Hint: Use (2) of 3.8A with &=logh) Show that the series 1-}-44+4+4—4—44 +——-++ is convergent. Deduce theorem 3.3 as a special case of Dirichlet’s test (3.80). 3.9. (C,1) SUMMABILITY OF SERIES Just as the convergence of the series 3a, is defined to mean the convergence of the sequence (5,) 2, of partial sums, the (C,1) summability of 7,4, will now be defined to mean the (C1) summability of (5,) Ber. (See 2-11A.) 3.9. periiTion. Let EZ_\a, be a series of real numbers with partial sums 5, = a,++++ +a, We shall say that 2_,a, is (C,1) summable to A if eo Jimse4 (C0, In this case we write Sana (G. = (C1). For the sequence of partial sums for this series is and this sequence is (C,1) summable to ~ 3 The series 1=2+3-4+5~6+-++ has the sequence of partial sums 1, ~1,2,—2,3, 3s. AS we saw after theorem 2.11B, this sequence is not (C, 1) summable. Hence the seties 1-2+3-445—6+-++ is not (C, 1) summable. 3.9B. DEFINITION. A method T of summability for series is called regular if every convergent series is T summable to its sum, (That is, if B_,a, is T summable to whenever 22. a, converges to 4.) 3.9C, THEOREM, The (C,1) summability method for series is regular. 39 (C.1) SUMMABIITY OF seus | 99 PROOF: If EZ.,a, converges to A, then the sequence of partial sums (5, Je converges to A. Hence, by 2.11B, lim, ..5,=A(C,1). This in turn implies 32.) =A(C,1), which proves the theorem. 39D. We know that a series may be (C, 1) summable even if it does not converge. We now prove a theorem giving a simple condition on a series which, together with (C, 1) summability of the series, will ensure that the series does converge. We first need a Jemma, teva. Let B2_,a, be (C,1) summable and let 4, 1 t2a; +--+ tna, Then if (a) the series 3a, converges. proor: We first show that 12 (nt1)s,—n, (NEL) @ where o, =n Ns, 45,4 ++ +5,). We have f |= 9, So that (2) is certainly true for in= I, We proceed by induction. Suppose (2) is true for some value of n. We then have b+ (8t Dey y= (0418, 20,4 (24 Dig HFS + apt) 204 = (MED Spar) (1 +5) = (MAM Sos) (B14 5+ Spt) teat and so Ines = (OF2)5 p41 — (0+ Op Thus (2) is true for n-+1 which completes the induction. Now suppose that 25° ,4,=4(C,1). Then lim,....9,4. From (2) and (1) we then have oe on gh(S+6) tinon tn, chy (S+a)-torarea In particular, {5,)%. converges, which proves the lemma. THEOREM. If Z_,a, is (C,1) summable and if tim,....na, then 22,4, converses. PROOF: The sequence {na,)2 (C,1) summable to 0. That is, converges to 0 by hypothesis. By 2.11B, {na,}2., is But this says precisely that lim, ,,.f,/=0 where ¢, is as in the lemma. The hypotheses of the lemma thus hold and the theorem follows. 3.9E. The only method of summability for series that we have discussed in this section is the (C,1) method. However, from our development it should be clear that any regular method of summability for sequences defines a corresponding regular method for series. 954 | SERIES OF REAL NUMBERS Thus we would say 2229, A(C,2) if the sequence (s,)7_ of partial sums of 22,4, is (C,2) summable to A. From 2.11D it then follows, for example, that 1-243-445-64-- =} (€,2), (For the sequence of partial sums is, in this case, 1, ~ 1,2, ~2,3,~3,....) This (C,2) method for series is regular since (C,2) summability for sequences is regular. In a later section we consider methods of summability that do not deal with the sequence of partial sums, See Section 9.6 Exercises 3.9 1. Examine whether the following series are (C, 1) summable. (@) 1-34 1-341-34- () I-$4}-14 (©) 1+0-1-0+1+0-1-0+ 2. Does the series (c) of the preceding exercise have the same (C,1) sum as I-1HI-1HI- Tet 3. Prove that Leite terete inne is not (C,1) summable. 4. Show that a divergent series of positive terms cannot be (C, 1) summable. 5. Prove that if D%.,4; is (C.1) summable, then the sequence (2,7, is (C,1) summable to 0. (Compare this result with 3.1D.) 6. Prove that if Z%,a, is (C,1) summable, then lim, _..s,/n=0 where s, tate tay 3.10 THE CLASS & Most of the important concepts we have so far introduced (Set, sequence, function, series) are involved in the definition of the class 3.104. pertrioN. Be sh 2)'( > a)" ® yand ¢=(1,)_, are in ‘Otten called the Cauchy ofthe Buniakovsk inequality ano meciass® | 95 pRoor: We may assume that there is at least one 5,—say sy—not equal to 0. Otherwise the theorem is trivial. For fixed > NV and any xe R we have S (is, +4)?2 0, Expanding the parentheses on the left we have #3 f42x Dut D pro. This can be written Ax?+ Bx + C >0 where A=Ssi>0, Be Sam co ba [From calculus we know that the minimum value of Ax?+ Bx-+C (A >0) occurs when B/2A (verify'). This is motivation for the next step in the proof.) If we set x= —B/2A, we have A(— B/2A)'+ B(~R/2A)+C 20, or B?<44C. But this says If we replace 5, % by |sch él in (2), we obtain Spel)" (Sel “A 8o)"(S0) Pa The sequence of partial sums of Ef :|s.4| is thus bounded, and so EP. yIs,fy|<09. In particular, 374% converges by 3.48. If we now let n approach infinity in (2) and use 2.7E, we obtain (1) 3.10, THEOREM. (THE MINKOWSKI INEQUALITY.) If s=(5,)@21 and (= (1,)2., are in P, then s+0= (5, +4) 201 i8 in @ and 2 pe A +34] [§ (+4)? i 7 [32 Proor: By hypothesis, the series B2.ys3 and DZ 12 converge. Also, the series EZ jsf, converges by 3.10B. Since (5, +1,)"=53+25,0, +13, it follows from 3.1C that DiLiGe+f)? converges and © +4) Sse2F aur Be Applying the Schwarz inequality to the second sum on the right, we obtain Serwte Sar $ 2)" and so 5 crite $2)" 5 "| : ‘Taking the square root on both sides completes the proof. 96 | SERIES OF REAL NUMBERS The class is used as an example of a metric space in the next chapter. For this purpose it is useful to introduce the notion of the norm of an element in F, 3.10D. DEFINITION. If s={5,}¥-, is an element of (*, we define isi}, called the norm of, . ist=( 2)" ‘The norm of a sequence of / is thus a nonnegative number. (Actually then, the norm is a function with domain ® and range (0, 20).) The sequence 0,0,0,... has norm 0. Any other sequence in has positive norm. If eR and s=8, then |jes|;=[e| sls (verify). 3.108. rnizoneM. The norm for sequences in 1 has the following properties: Is|,20 (SEP), () Isl,=0 if andonlyit s=(0)* @ llesty=lel-lis, (CER SEP), GB) lis+elle<[lsiia*iltla (EP). (a) PRooF: Only (4) has not been verified. But (4) is just a restatement of the Minkowski inequality 3.10F. Do not read this section unless you know a little bit about vector spaces. First of all, since s+1€F if s€# and s€F, and since es P if c€ Rs E07 it is clear that F is a vector space over the real numbers. In n-dimensional Euclidean vector space, a vector (Sj..-y5,) has length (92) The dot (scalar) product of two vectors ($y...) and fjromrfp) i8 equal to 3] and is in absolute value less than the product of the lengths of the two vectors. That is. Ta sul 34)"(34) © Thus the norm (S183) of a sequence in (is analogous to the length (Bf .,5f)" of a vector in n-space. The Schwarz inequality corresponds to (*), The Minkowski inequality states that the “length” (norm) of the sum of two vectors (sequences) in |? is less than or equal to the sum of their lengths. In n-space the corresponding fact is that a straight line joining two points is no longer than any broken line joining them, Exercises 3.10 1. Which of the following sequences are in © (isa )”, © (en, © {ant 2. Give an example of a sequence (5,)¢., in @ such that 2. [s,|=00 {3.11 REAL NUMBERS AND DECIMAL EXPANSIONS | 97 Prove that if (5,) 2-1, then lim, 4. Show that equality can hold in (1) of 3.108 if and only if (14), is “proportional” to (5) 2a1~ that is, if and only if (NED) for some NER. If (a,)2.,€®, prove that D2_,a,/m converges absolutely. Prove that if 32. Ja,|<20, then (a,)%. is in P. 1 For each KEI’ let ¢ be the sequence 0,0,...,0,1,0,0,...,all of whose terms are 0 except the kth term, which is equal to 1. Show that |je,—ga= V2 if ket). What is Weal? (in n-space can you find infinitely many vectors of length | such that any two of the vectors are distance V2 from each other? Answer: No. In n-space how many such vectors can you find? Answer: 1.) 8. Ifa, > 01) and if B%_, a, converges, prove that converges. 3.11, REAL NUMBERS AND DECIMAL EXPANSIONS The theory of infinite series enables us to treat more carefully the connection between real numbers and “decimal expansions” first mentioned in Section 1.1. We consider only expansions for numbers in (0, ]. Here is the definition of decimal expansion. 3.11A, DEFINITION. A decimal expansion is an infinite series B2_,a,/10" where each a, is an integer, 0< a, <9. Its, of course, customary to write the decimal expansion E..a,/10" as .a,a,03° ‘The number a, is called the nth digit of the decimal expansion. We how show that every decimal expansion “represents” a number in (0 1) 3.11B tHsonem. Every decimal expansion converges to a number in {0,1} PRooF: The decimal expansion Z%.,a,/10" is clearly dominated by the series Zz_,9/10", But D2,9/10" is (absolutely) convergent and thus, by 3.6B, so is D2_14,/ 9 ‘Sa S(a)-( Sa) Thus 0< 32. 4,/10"< Bz_,9/10"< 1 and the proof is complete. 3.11C. DEFINITION. If the decimal expansion B2_,2,/10"=.a,0,a,"- converges to xe (0,1), we say that .aya,ay°++is a decimal expansion for x (or that x has the decimal expansion .ay4,a5°--). 90 | SeREs OF REAL NUMBERS ‘Thus 3.11B tells us that every decimal expansion is the decimal expansion for some 6 [0,1]. We next show that each x [0,1] has at least one decimal expansion. 3.11. THEOREM. If x€[0, 1], then there is a decimal ‘expansion converging to x. PROOF: Suppose first that x can be written as x=k/10" for some k, nEI where <1, Then k=a,-10"!+a,-10'24--- +4, for some integers a,. Thus x=a,/10 “+ 4,/ 10+ +++ +a,/10" so that .ayd,:- 4,000» converges to x. If x cannot be written as x= 4/10" for any n€ J, then x lies in one of the ten open intervals (0, ), (qb. f)oo--s (Be HB)- If x(0m/10, (m+ 1)/10), let ay=m. Then xa, <+h- Now x lies in one of the ten open intervals MH HAV, A My 2) (m, 9 my 10 (ta 10 * was) T6* w00° 10° To6)+~~(T0* 705-70 * To} i fi aie eeg ee P (155-16) Let a,=p. Then x~.a,a;<75. Continuing in this fashion we can define a decimal expansion .a,ayd,° + that clearly converges to x. This completes the proof. Unfortunately, some numbers in (0, 1] have more than one decimal expansion. For example, }=0.5000:- + and |=0.4999:-+ . Indeed, since 1=0.999-+- we have 1/10"= (000: «-0999-- (where 1 0's precede the string of 9's). From this and the proof of 3.11B, the reader should have no difficulty in proving the following result. 3.1IE, THEOREM. Every decimal expansion that ends in a string of 9's converges to a number of the form k/10". Conversely, every number in (0, 1] of the form k/10" has a decimal expansion ending in a string of 9's. Thus every number in (0,1) of the form /10" has (at least) two decimal expansions, one ending in a string of 9's and another in a string of O's. ‘The next theorem enables us to prove that any number in [0,1] not of the form k/10" has precisely one decimal expansion. 3.11F. THEOREM. If the two decimal expansions .a,0,a4--- and .b,b,b,+ are distinet (that is, if a, #6, for some k 1), and if neither expansion ends in a string of 9's, then they converge to different sums. PROOF: Let m be the largest integer sugh that aj=o, (k2qey- Then dey) > byait land so {yn 000° + >, 000+ ++ 0.1000: =.b,..,000-+ + +0.0999- + by 1999- + > bya Pagabna 3” "+ since .byb;*-+does not end in a string of 9's, Certainly, then, .d,4 \4y42e3°** >. bes ibna2bn-es*** And SO .0\4;4y:*~ >.b,bybs.... This proves the theorem. 3.11G. conotary. If x (0,1) is not of the form k/10", then x has one and only one decimal expansion. If x is of the form k/10", then x has precisely two decimal expansions. 3.12 NOTES AND EXERCISES FOR CHAPTERS 1, 2,AND3 | 99, PRooF: The proof follows directly from 3.11D, 3.11E, and 3.11F. (Verify!) Exercises 3.11 1. A repeating decimal expansion is one of the form 1g" ** yb by" * Bygb By" ~~ by yba* ** Bag * Show that the decimal expansion of a rational number is repeating. (Hint: If x=p/4, calculate the decimal expansion for x by repeated short or long division and consider the number of possible remainders.) 2. Conversely, show that every repeating decimal expansion converges to a rational number. 3. Prove that between any two distinct real numbers there is a rational number, 4, Prove that the open interval (0,1) is equivalent to the open square xy pl0< xy <1). 3.12 NOTES AND ADDITIONAL EXERCISES FOR CHAPTERS 1, 2, AND 3. |. Some results from set theory. 312A. The following result is known as the Schréder-Bernstein theorem, TuroREM: Let A and & be disjoint nonempty sets. Suppose that 4 is equivalent to a subset of B and that B is equivalent to a subset of 4. Then A and B are equivalent. PRoor: We begin the proof as follows: By hypothesis, there is a 1-1 function J from A onto a subset B, of B, and a 1-1 function g from B onto a subset 4, of 4. Fix x4. Ifx@A—A,, then x is not the image under g of any point in B. In this case we say that x has no ancestor. However, if x=A,, then there is a (unique) y€B such that x= (y). In this case we say that y is an ancestor of x. Next, consider this y. If YE By, then there is a unique EA such that f(z)=y. We then say that zis an ancestor ‘of y and an ancestor of x. On the other hand, if y © B~ B,, then y has no ancestor and y is the only ancestor of x. Note that y, if it exist, is equal to g”"(x), and 2, if it exists, is equal to f~'(y)=f~'[g~'(a)}- Thus for any xe, the ancestors of x are the elements of those sets in the sequence oe Letoo)e © that are ot empty (and hence consist ofa singe point of 8 ot 4), OF couse, if any term of (+) is empty, then so are all the subsequent terms. So any #€.4 has 01,2." or possibly infinitely many ancestors. (I the same ancestor oeeurs more than once in («it shouldbe counted ws many times ait occurs. Can the tame ancestor occur exally tie?) Let A, be the set ofall points in A which have infinitely many ancestor, and et dg, A, be respectively the ets of all pons in that have a (finite) odd or even number of ancestors. Let B,,Bg,By be the corresponding subsets of 2 EXERCISE: Finish the proof, 100. | SERIES OF REAL NUMBERS 3.128, By a partially ordered set we mean a set A together with a relation < between some but not necessarily all elements of 4 satisfying Xex (xe), aw xE < we write x < y. For example, let X be any nonempty set and let 4 be the family of all subsets of X. If x,y EA, define x < y to mean xCy (remember that x and y are subsets of X). Then the relation x < y holds for some of the x,y @A and itis clear that the conditions (1)-(3) are satisfied. Hence 4, the family of subsets of X, in “partially ordered by inclusion.” For another example let 4=/=(1,2,-+-) and define x< y to mean that x divides y (that is, that y is an integral multiple of x), Thus 4<24, $< 24. Note that 5<24 does not imply 24<5. EXERCISE: Show that (1)-@) are satisfied in this example, 3.12. Suppose A is a set partially ordered by the relation <. If BCA, xEA, we say that x is an upper bound for B if bex (bEB) Thus in the last example, if B= {3,6,10}, then 120 is an upper bound for B. Next we define a chain in A as a subset B of A such that any two elements of B are comparable, That is, we say B is a chain if, whenever x,y € B, either x< y ory 0 such that © lel #2 ‘Then there exists n,>m, such that 2, bul Ie folows that Saleh Now there exists m,>m, such that Sicagr—2. This is a contradiction (why?) and the proof is complete 3.121. Now we prove that conditions 4,B,C are sufficient for T summability to be regul ‘Assume A,B,C. Suppose (s,} converges to L. We must prove that (1,,} also converges to L. We have ty LoS lgaty Le = Seate-uiee[ St] EXERCISE: Finish the proof. Consult the book Divergent Series, by G. H. Hardy (Oxford, 1949), for more material of this nature, IV. More about the product of series. 3.125. exeRcise: The assumption in 3.5G of absolute convergence for Bos, and ZFLgb, cannot be weakened simply to convergence. Demonstrate this by taking Ss ae (K=0,1,2,...) VEFI and showing that 22.96, diverges. However, a stronger theorem than 3.5G can be proved 3.12K, MERTEN’s THEOREM Suppose BP g2, converges absolutely to A, and DPob, converges (not necessarily absolutely) to'B. Then E.9¢, converges to C= AB, where Sara (n=O, ) SKETCH OF PROOF: case I; It must be shown that C,+0 as n-o20, where C,=Z4o¢y. Let B= Ze = 2, la! 0 there exists NJ such that IGS la Bot s+ tay yByltea (nN). 106 | SERIES OF REAL NUMBERS case I; BHO. Let B b=b, (nel). Then E7.9b,=0. Apply Case I to Boa, and BP oby, EXERCISE: Fill in the details and finish the proof. (MMISCELLANEOUS EXERCISES. 1, Suppose f:XY. (@) IEACX, show that 4c f~'Lf(A).. (b) If BCY, show that fIf" (BIC B. 2. Suppose f: + ¥. Show that fis onto if and only if e=fse)] for every ECY. 3, Suppose f:X—> ¥. Show that fis 1-1 if and only if for every AC. 4, Let s;=1and tere 245, (n>2) Show that {s,}2., is bounded and nondecreasing. Then compute lim, Prove that mx"), converges to 0 for every x such that 0<.x<1 . Let (5,}., be a Sequence of real numbers. Prove that (5,) 2, converges to L if and only if every subsequence of (s,)*., has a subsequence that converges to L. 7. Let (5,) 21 be a bounded sequence of real numbers. Let A be the set of all numbers L such that {5,]2, has a subsequence that converges to L. Prove that bd. limsup sy 8, Let 4 be a nonempty set and let X,CAY,CA (nT). (a) Prove that limint X,U liminf Y,Climinf(X,U ¥,) and that equality need not occur, (b) Prove that imsup X,U limsup ¥, 9. Prove lim, .91"/*= 1, (Begin by setting s,=n'/"—1 and use the binomial theorem.) 10. Prove that S¢_,1/n('*!/” diverges. 11, Suppose that a,>0 (nEJ) and that E2.4a, converges. Prove that B2., Vad.) converses. 12, Prove that 4 belongs to the Cantor set but is not an end point of any of the open intervals removed. 3.12 NOTES AND EXERCISES FOR CHAPTERS 1, 2,AND3 | 107 1. The following game was invented by S. Mazur. Player A owns the irrationals in {0,1}; player B owns the rationals in [0,1], One player (either A or B) starts by choosing @ closed interval in {0,1] of length <4. The second player now chooses @ closed interval of length <} inside the interval already chosen. The first player now chooses an interval of length <4 inside the preceding interval, and so on. By ‘Theorem 2.10E there is a unique point x in all of these intervals. If x is irrational, then A wins. If x is rational, then B wins Prove that A can always win, independent of B's strategy LIMITS AND METRIC SPACES 4.1 LIMIT OF A FUNCTION ON THE REAL LINE In Chapter 2 we define limit of a sequence. We now recall from calculus the definition of limit of a “function of a real variable” on which are based the definitions of continuous function and derivative. Later in the chapter we generalize to a wide class of spaces (called metric spaces) which includes the real line R as a very special case. Let a R, and let f be a real-valued function whose domain includes all points in some ‘open interval (a— iia h) except possibly the point a itself. 4.1A. DEFINITION. We say that f(x) approaches L (where LER) as x approaches a if given €>0, there exists § >0 such that Weg-La saye4 FIGURE 15. Example 3 of Jim f(a)= L no [om 4.1 LMT OF A FUNCTION ON THE REAL LUNE fiext1 <2) seat 2) FIOURE 16. Example of «function f such tht f(x) dows not approach olin as x approaches @ Fora final pictorial example, consider Figure 17. Ths shows the graph off where soy=sink (a0) Here, «gets close to a=0, the vale f(s) oils rapidly. Even if we look on only one side of a, it is clear that there is no number L toward which the value f(x) tends. Hence f has no limit at 0. Nei ta cor ezine ee First, let us prove lim,_,,(x?+2x)= 15. Here f(x)=x?+2x, L= 15, a=3. Given e>0 we must find 6 >0 such that Wat42x)-15]0 we have found a 6 {namely 8=min(1.¢/9)] for which (1) holds, and this proves lim,_(x?+2x)= 15. Note that in this example a=3 and | f(x) — L|0 we must find 6 >0 such that IVEF3 -2]V3 + 2. Thus if |x— 1] <3 <1, then [et Ives +2] V3+2 If we pick 5=min(1.e(V3 +2), then 6/(V3 +2)0 such that ; <1 (O<|x| <6). (4) Now 2), (ents sn(2m+ £) sin OSD" for any n€/. Thus sin(1/:x)=1 for x=2/r(4n+1) and hence for some x €(0.8). since lim ..2/#(4n+ 1)=0. For this (4) implies I-4/<1 6) Similarly, sinQnm+3z/2)=-1 for mEJ. There will thas be an x(0.6) for which sin(I/x)= ~1. By (4) again, [-I-Z\<1. ©) ‘The reader should be able to deduce a contradiction from (5) and (6). Hence lim, 9 sin(1/-x) does not exist, 4.1 UMIT OF A FUNCTION ON THE REAL UNE | 113 4.1. We wish to emphasize the strong analogy between definition 4.1A and definition 22A. Indeed, consider the “table of analogues.” TABLE OF ANALOGUES 22a a S=()e1 g % Se) L L x 3 n>W O<|x~alcd If we substitute each entry in the right-hand column for the corresponding entry on the left, we change definition 2.2A into definition 4.14. However, more than a mechanical process is involved here. Corresponding entries in the table actually “have the same meaning.” For example, S=(s,}2_, is the funetion (sequence) involved in definition 2.24, while fis the function involved in 4.1. Also s, is, the value of Sat n, while f(x) is the value of f at x. Finally, n>. means “n is, sufficiently close to infinity” (but not, of course, equal to infinity), while 0<|x—a|<8 means “x is sufficiently close to a but not equal to a.” We will now prove a theorem corresponding to 2.7A. The reader should first study the proof and then see how it could be obtained from the proof of 2.7A by mechanical substitution from our table. 41C, suttoneM, “Hf tim, f(x) =L and lim,_gg(s) = M, then f(s)+ g(x) has limit as xa and, in fact, lim, (f(x) + a())= L+ M. PRooF: Given €>0 we must find 6 >0 such that IF) t+g(]-(L4 MI 0 such that I)-LI<$ O<|x~a] <8). Similarly, there exists 8,>0 such that la@y—M1 <5 (0<|x-a]<8,), Thus if 8 =min(6,,6;) and if 0<|x—a|<6, then UG)-LI<§. lae)-MI<§, and so L9G) +26] (L+ M)I=ISC) L] +L) MY SIPC) Lt |a(2)- M5 +5 §re Thus (1) holds for 6=min(6;,8;) and the proof is completed. “In this chapter, whenever we write a hypothesis such as lim, .4/(2)= its understood tha isa funetion whose domain contains all points whose distance from ais leis than some h(h>-O) except perhaps the point 2 itsel 114 | units AND MeTRIc SPACES Using the proofs of 2.7D, 2.7G, and 2,71 as models, the reader should now be able to prove the following theorem, 41D, THeorem. If lim, ,f(s)=L and lim,_,g(x) = M, then (2) lim, of f(@)~ (|= LM, () lim, ZC) 404) and if M#0, (6) lim, 6 )/2091= L/M. We occasionally need to handle limits of the form lim,_.. f(2). 4.1E, permion. We say that f(x) approaches L as x approaches infinity, if given €>0, there exists ME R such that Lf@)-LM), In this case we write lim, (x)= L, oF f(x)>L as x20, Definition 4,1E requires, of course, that the domain of the (real-valued) function f contain some interval of the form (c, 20). Note the very strong resemblance of 4.1E to 224. For example, let us prove that lim,_,.,(1/27)=0. Given ¢>0 we must find M © such that ke (x>M), a Since (1) is equivalent to Aeve (x>M), itis clear that (1) will hold if we take M=1/Vé Its also useful to consider “one-sided” limits 4.1, pestsirion, We say that f(x) approaches L as x approaches a from the right, if given €>0, there exists 8 >0 such that We)-L]0, there exists 8 >0 such that W)-MI0 the number M ~¢ is thus not an upper bound for the range of f. Hence there exists y €(a,b) such that f(y) > M~e. Let = 5—y. Then H(b—8)=f(y)>M—e Since f is nondecreasing, this implies S(2)>M—e (b-60 such that (c~d,c+8) (bounded open interval) is contained in (a,6). Then the values of fon the open interval (c= 8,0) are bounded above by /(e) and hence by 4.1H, lim... /(x) exists. Similarly, the values of fon the open interval (c,¢-+8) are bounded below by /(c). Hence again by 41H, lim, se f(3) exists. If f in nonincreasing we use 4.11 instead of 4.1H. This completes the proof. Note that we did not assume that (a,0) was bounded, As long as we are on this subject we may as well define strictly increasing function and strictly decreasing function. 4.1K, permvmion. The real-valued function f on the interval J CR is said to be strictly increasing if JENS) (x5: HYVES). Similarly, fis said to be strictly decreasing if J)>I0) (&< yi BVED) Thus if fis nonincreasing on J, then f is strictly increasing on J if and only if fis 1-1 ond. Exercises 4.1 1, (@) If |x=2] <1, prove that |x?7—4]<5. (b) If [x~3]< 4, prove that |x?—x—6|<0.51 (©) If [x41] Ofor |x~a|0. If lim,_4f(s)= L>0, show that there exists 6 >0 such that S(x)>0 (0<|x=al0 (ae R,a¥0), Q) lal=|-a| (ae R), QB) Jar s|O (xy) (6) (The distance between two distinct points is strictly positive.) 0,9) = 0( 9%). o (The distance from x toy is equal to the distance from y to x.) phy) 0 (XYEMXAY), elsy)=e(y.x) (SEM), ply) [0, 20) by d(xx)=0 (XER), d(xy)=1 (xyER: xHy). That is, the “distance” d(x,») between any two distinct points x,y ER is 1. The reader should verify that d is a metric for R. The metric d is called the diserete ‘metric. We will henceforth denote the metric space (R,d) by R,. The examples | and 2 show that a given set may have more than one metric. 3. Fix MEL. I x= Cx joes) ANd Y= CYjyeredq) BFE WO OFdered n-tuples of real numbers, define , va otsvo=| 3, ton [For m=2, p(x») is thus the usual distance formula for points in the Cartesian plane.] We will show that p satisfies the triangle inequality. Thus, if We must show pxyy)< p(x.2)+ 02,9). For k= Lom Iet a= x4 Zerby Then ol. wo-(Sai) (Sa) oe. and (x.y): [ > toe] We must thus show that . 12 fn te [ = ar| <( = a) + = ) But this follows from 3.10C. Tt is trivial to verify that p satisfies the other requirements for a metric. We denote by R" the metric space formed by the set of all n-tuples of real numbers with this metric p. The metric space K” is called Euclidean n-space. (Note that for m= 1, R" becomes the R! of example I since Zon] “ior 4, Let © denote the set of all bounded sequences of real numbers. If x= (x,}2_ and Y= (04) 2-1 are points in &, define aCey)™ Lub, [xh For example, f= (141/n}2n, y= (2~1/m)2u then otso~ jp, +t) (2-2) tw, [14 2 for Again, it is easy to see that p satisfies the first three requirements of a metric. To demonstrate the triangle inequality, let z= {2,)%., also be a point in , For any 120 | urs ANo Mera seaces ET we have Decal le = 2a Ae Yad <1 Zul #12 Ii jUeusby |e zal LUD. Len Yale and so In- xl 0, there exists a point xEA such that O0>0) (P,Q. show that r defines a metric for the same set. 9. Let 0 denote the point (0,0) in R®, For o,7 as in Exercise 4, sketch the following subsets of R?: = max(|x;— 2) 191 Pal) 4 {PER%0(0,P)<1}, {PER %7(0,P)<1). Compare with C=(PER%0.P)<1), where p is the metric for R?, 10. If P, Q, R are points in R? and p(P, R)=p(P,Q)+ p(Q,R), what can you say about the relative position of P, Q, R? Answer the same question with Ry in place of R°. 11, Let A denote the open interval (0,1). Show that the set of cluster points of A in R'is (0,1) 12, IFA =(0,0), find the set of cluster points of 4 in Ry 4.3 LIMITS IN METRIC SPACES If we examine definition 4.1A we see that lim,..f(x)= L means that given ¢>0 there ‘exists 6 >0 such that the distance from f(x) to L is less than ¢ provided that the distance from x to a is less than 8 (but greater than 0). Now that we have stated this definition in terms of distances, itis not difficult to see how to formulate the corresponding definition for arbitrary metric spaces. Suppose that and 0) except possibly x=a. We also assume that a is a cluster point of the domain of f. 43A. DEFINITION. We say that f(x) approaches L (where LE M,) as x approaches a if given €>0, there exists 6 >0 such that PAML<€ — (OL as x-va. If (My.oy then pa(f(x),L)=|fC)~ Ll. py(x,a)=|x—al, and 4.3 reduces to 4.14. (MP2) = RI In later chapters we very often consider functions f on the metric space M=[a,b] (closed bounded interval with absolute-value metric). For this space the statement fim foe L o involves only points x to the right of a (since points in R' to the left of a are not in M). In 4.1F, L is referred to as the “right-hand limit of f.” However, there is no need for us to use this terminology as long as we remember on what space f is defined. A similar remark applies to fim f(a) =. These remarks are also relevant when we define the derivative of a real-valued function on (a,D} Here is an example illustrating 4.34. Let f:+R! be defined as follows: If x =(5,)2.1€ 8, let f(2)=x;. That is, the image under f of any sequence in isthe first term of the sequence, Now let a= (4,)., be any fixed element of 2. We will prove that Jim, .f(8)= 4). Given ¢>0 we must find 8 >0 suchrthat the distance from f(x) toa, (in the metric for R?) is less than € whenever the distance from x to a (in the metric for #) is less than 8 but greater than 0, That is, we must find 6 >0 such that U@)-al<¢ — O0, le(x)-NI<1_ (00 there exists 8,>0 such that QU()-U<$ OXe(x.0)0 such that Helse—MI<5 0, there exists NET such that PssL) be @ metric space and let (5,} 7. be a sequence of points in M. We say that (5,)-y is a Cauchy sequence if given ¢°>0, there exists N= such that Asp )<€ (mn >N), 124 | UNITS AND METRIC SPACES The proof of the following theorem is identical to that of 2.10B. 43E, THeoREM. Let (M,p) be a metric space. If (¢,)., is a convergent sequence of points of M, then (5,}"2., is Cauchy. 4.3F, Now comes a very important point, For some metric spaces there are Cauchy sequences which are not convergent. That is, theorem 2.10D cannot be extended to all metric spaces. For example, let M be the set of all points 1 such that Fale) < PCa) 0 such that [x(x)~ LI0 we must find 6 >0 such that Is) -s(@li0 such that Ia)-a(l0, prove that there exists h>0 such that S(x)>0 (a-hex0, prove that there exists h>0 such that [(al>0 (an h0 there exists 6 >0 such that [/(x)— f(a)|0 there exists 5 >0 such that F(y-f@i0, there exists 8 >0 such that, if the distance from x to a is less than 6, then the distance from f(x) to f(a) will be less than ¢, [This is sometimes put roughly as “if x is close to a, then f(x) is close to f(a)."] Theorem 5.2A shows that the definition of continuity is based on the metric in R'. 190 | CONTINUOUS FUNCTIONS ON METRIC SPACES Consider Figure 20. In order for fto be continuous at a, given any € parentheses about Sa) there must be 6 parentheses about a such that an arrow which begins inside the 6 parentheses must end inside the e parentheses fi) +6 fe) fla)~e oURE 20. In order to give another reformulation of the definition of continuity we introduce the following definition. 52B. DEFINITION. If GER! and r>0, we define Bla;r] to be the set of all xeR! whose distance to a is less than r. That is, Blas] =(xeR"||x—al0 there exists 60 such that f(x) BL f(ayie] if x= B[a;5}” That is, the entire open ball Bla:8] is mapped by f into the open B[f(a);«]. Thus f is continuous at a if and only if, for any open ball # about f(a), there is an open ball about a which f maps entirely into B. Tt turns out to be moze useful to state + this definition in terms of inverse images. 52 REFORMATION | 131 5.2C, THEOREM. The real-valued function f is continuous at @€ R! if and only if the inverse image* under f of any open ball B[f(a);«] about f(a) contains an open ball Bla;6] about a. (That is, given ¢>0 there exists 6 >0 such that LBL S(a)se))> BL a8},) Itis of the utmost importance that the reader fully understand 5.2C before going on. Our final reformulation of the continuity concept will be in terms of sequences. Observe first that the sequence {.x,)%_, converges to a if and only if given >0 there exists N €/ such that x EBlae] (n> N), ‘That is, given any open ball # about a, all but a finite number of the x, are in B. $.2D. THEoREM. The real-valued function f is continuous at a€R! if and only if, whenever (x,)%., is a sequence of real numbers converging to a, then the sequence (Fu) Ea converges to f(a). That is, fis continuous at a if and only if Jim x, implies lim f(s)=f(a) 0 PROOF: Let us first assume that f is continuous at a and prove that (*) holds. Let {4} 2.4 be any sequence of real numbers converging to a. (Then f(x,) will be defined for m sufficiently large.] We must show that lim,_,.. f(x,)=f(a)—that is, given €>0 we ‘must find N €J such that S(%)EBLS(@)ie]_— (nN). a But since f is continuous at a there exists 6 >0 such that A(XEB[S(a):e] (x= B[a38}). Q Furthermore, since lim,_....x,"4, there exists N EJ such that XE B[a:3] (n>). @) For this NY, (1) follows from (2) and (3). Conversely, suppose (*) holds, We must prove that f is continuous at a. Assume the contrary. Then, by 5.2C for some €>0 the inverse image under f of B= BLf(a)se] contains no open ball about a. In particular, f~'(B) does not contain B{a: 1/n] for any n€ I. Thus for each n€ J, there is a point x, = B(a;1/n] such that f(x,) = B. That is, ben al 6 This clearly contradicts (*), so f must be continuous at a. 52E. We use 5.2D to give an easy proof of 5.1C. Suppose, then, that the hypotheses of 5.1C hold. By 5.1D, all we need show is that dim sf f(s) ]=2[ 4) a where (x,}2 is any sequence of real numbers such that Jim x, @ Since f is continuous at a, (2) and 5.2D imply Jim, £5) =S(@) @ But since g is continuous at f(a), (3) and 5.2D imply (1) and the proof is complete +See 136. 122 | conTINUOUS FUNCTIONS ON METRIC SPACES Exercises 5.2 1. Use $.2D to prove 5.18. 2. Use 5.2C to prove 5.1C. 3. Use 5.2D to prove that if f is continuous at a€ R', then || is also continuous at a, 5.3 FUNCTIONS CONTINUOUS ON A METRIC SPACE All the formulations of the continuity of a real-valued function at a point in R! were based on the metric for R'. It is therefore easy to extend the concept of continuity to functions from any metric space into another. We first define “open ball” for a metric space 53A. perinmmion, Let 0, then Blas] is defined to be the set of all points in M whose distance to a is less than r. That is Blazr]={x=Mlo(x,a) and (M,,p2) be metric spaces, let a M,, and. let f be any function whose range is contained in M, and whose domain contains some open ball Bfa;h] (4>0). 5B. DEFINITION. The function f is continuous at aM, if lim, f(x limit is defined in 4.3A). (a) (where The proof of the following theorem consists merely of translating the proofs of 5.2A, 5.2C, and 5.2D into metric-space notation. Suppose f is as in the paragraph preceding 538. 53C, THEOREM. The function f is continuous at a€M, if and only if any one (and hence all) of the following conditions hold. (2) Given €>0 there exists 6 >0 such that pal Fela] M g:M,>M,,* If f is continuous at a@M, and g is continuous at f(a)€M;, then gef is continuous at a PRooF: By (c) of 5.3C all we need show is that Jim ef) ]= 2 /)] whenever (x,}., is a sequence in M, such that lim x, ‘The proof then proceeds exactly like the proof of 5.2E. For real-valued functions on metric spaces there is @ generalization of 5.1B. The following theorem may be easily deduced from 4.3B (in the same way that 5.1B was deduced from 4.1C and 4.1D). 53K. THroREM. Let M be metric space, and let f and g be real-valued functions which are continuous at a€M. Then f+g, f—g, and fg are al8o continuous at a, Furthermore, if g(a) #0, then f/g is continuous at a. We emphasize that, so far, only continuity at a point has been defined. The continuity of a function f at a point a is a local property—that is, continuity of f at a depends only on “what goes on near a.” ‘Now we will define what we mean by a function continuous on a whole metric space. 53. DEFINITION. Let M, and M, be metric spaces and let f:M,—>M;, We say that fis ‘continuous function from M, into M, (or, more simply, is continuous on M,) if fis continuous at each point in M,. 5.3G, THEOREM. If f and g are real-valued continuous functions on a metric space M, then so are fg, fg, and fg. Furthermore, if g(x)#0 (xEM), then f/g is also continuous on M. PROOF: The proof follows directly from 5.3F and 5.3E. “Any polynomial function f (that is, f(x)=agx"-bayx"-"-b +++ ++ aj} is thus a con tinuous function on R'. For constant functions are continuous on R! and so is the function g(x)=.x. The function f can be written as a sum of products of these kinds of functions and is thus, by 5.3G, continuous. The function h defined by h(x)=(1+x°)/(1+x) can be written f/g where f and g are polynomials, Since g(x) is never zero, it follows that h is continuous on R! Here is more curious illustration. Let fbe any function from the metric space Ry into a metric space M. We have already observed that for any a R,, the open ball B[a; 1} contains only the point a. Thus for any ¢>0, the inverse iniage under f of BL/(a):e] + For simplicity of statement we are assuming thatthe domains of fad g ate all of My and Af, eespetively. 124 | CONTINUOUS FUNCTIONS ON METRIC SPACES certainly contains Bq; 1]- By (b) of 5.3C, this shows that fis continuous at a, Since a was an arbitrary point in R, we have 53H. curtostry, Every function from R, (into a metric space) is continuous on Ry, Exercises 5.3 1. Give an example of a function which is continuous on R! and whose range is, (a) (0,20) (b) (0, 0) © 1 @) (0.1) You may assume that e%, logx, sinx, and so on, are continuous where they look continuous. 2. Let f be the function from R? onto R! defined by S(ayr=x (Kaye R?), Show that f is continuous on R* 3. I f:R?>R? is defined by FCar)= Ore (Kav ER?)s show that fis continuous on R?, 4, If:R'R!, g:R'OR' if f and g are both continuous on R', and if Mae BOY — (x~PER?). prove that J is continuous on R* 5. Define f: #+F as follows. If sis the sequence 5),5..--. let f(s) be the sequence 0,sy.5z.01+- Show that fis continuous on F, 6. Let M be a metric space. Suppose f:M—>R, and that f is 1-1. Show that if f is continuous at a M, then (a) is an open bal in M. 7. True or false: If fis a I-1 continuous function from a metric space M, into a metric space M,, and if B is an open ball in M,, then f(B) is an open ball in Mf, 8. Let A be @ nonempty set. Find a metric p for A for which there exist r,r3R with O<4r,< rz and such that Alain, ]= Alain] for every aA. 9, Lev M, be met spaces and suppose f:My-»M, Prove that fis continuous i tnd only fj sends convergent sequence in fo convergent sequence in My ee, oC ——r———— =F ——— numbers x in (0,1). Thus f:(0,1)->{0, 1} (a) Prove that f is nor continuous at any rational. (b) Prove that f is continuous at each irrational. (Hint: Show that for any €>0 there are only a finite number of rational numbers p/q in (0,1) such that 1/q>«.) (©) Show that f can be extended to a function g on R? such that g is continuous at tach irrtional buts ot continuous at any rational 5.4 OPEN SETS In order to formulate properties of continuous functions on a metric space M we need to attach names to various kinds of subsets of M such as open, closed, bounded, totally bounded, compact, and so on. We begin by defining open set. 54 ovensers | 138 SAA. DEFINITION. Let M be a metric space. We say that the subset G of M is an open subset of M (or, more simply, that G is open) if for every xG, there exists a number >0 such that the entire open ball B{x;r] is contained in G. ‘As an intuitive example consider the set 4 of all points in the plane R? inside an ellipse. (Draw a picture.) If P A, we can draw a circle with center P which lies entirely inside the ellipse. The set B of points inside this circle is then an open ball (in R?) which lies entirely in 4. This shows that 4 is open in R?. Next, let us prove that for an arbitrary metric space (M.p), any open ball © = Blas] is itself an open set. (This will justify the use of the word “open” in “open ball”) If XG, we must find r>0 such that B[xjr]C B . Let r= p(x,a) and let r be any positive number less than 5~f. (Why is s—t positive?) If yEB[xsr}, then p(a.y) M,. Then f is continuous on M, if and only if f-1(G) is open in M, whenever G is open in M. Briefly, fis continuous if and only if the inverse image of every open set is open.) PROOF: Suppose first that fis continuous on M,. We wish to show that if G is open in M,, then f-'(G) is open in M,. Thus if xef~1(G), we must find an open ball B[x;r] contained in f~'(G). Now since x€f~'(G) then y= f(x) © G. Hence there is an open ball BLy;s] contained in G (since G is open in M,). By (b) of 5.3C, f~'(BLy:s)) contains some B[xsr]. Hence f~"(G)3f~'(BLy:s)3 B[xsr] which is what we wished to show. Now suppose f~'(G) is open in M, whenever G is open in M,, To show that f is continuous on M, it is sufficient to show that f is continuous at an arbitrary point GEM, Let % = B[f(a);e] be any ball about f(a). Then % is open in M, and so, by assumption, f~'(@ ) is open in M,. Since a€f~(B) and f~'(B) is open, there is an open ball Bja;5] contained in f~!( ). But then, by (b) of 5.3C, fis continuous at a ‘This completes the proof. Exercises 54 1. This concerns the proof of 5.4F. Show that if G is an open subset of R! and if xG, then there is such a thing as the largest open interval /, containing x such that LG. 2. Use your intuition to decide which of the following subsets of R? are open. (a) (Cop lxty=), (b) {Cxy>]x ty > 1). (©) {]x and y rational). (@) R?~ (<0,0>}. (That is, R? with the origin removed.) 3. Let xj,x, be distinct points in a metric space M. Find disjoint open sets G, and G, such thgt x,€G, and x,€G,. 4, Let E be the set of positive real numbers. Find f-"(E) for each of the following functions f. (@ f@)=sinx (-2). 5. Prove that if fis any continuous real-valued function on R', then f~'(E) is open in R® (where E is as in the preceding exercise). 6. Let f and g be continuous real-valued functions on the metric space M. Let A be the set of all x EM such that f(x) 1s, clearly, not a limit point of. ‘We now define a closed subset of as a subset that contains all its limit points 5.5C, DEFINITION, Let E be a subset of the metric space M, We say that E is a closed subset of M if B= E. In view of 5.5B, to show that a subset E of M is closed it is enough to show that Eck. Before proceeding to examples we give another formulation of the concept of limit point. 55D. THEOREM. Let E be a subset of the metric space M. Then the point xEM is a limit point of £ if and only if every open ball B[x;r] about x contains at least one point of E. PROOF: Suppose x is a limit point of E. Then there is a sequence (x,)P-, of points of E that converges to x. If B[x;7] is any open ball about x, then B[x;7] contains x, for any n such that p(x,,)Mz, Then f is continuous on M, if atid only if f~ '(F) is a closed subset of M, whenever Fis a closed subset of M;, PRooF: Suppose first that f is continuous on M,. If FCM, is a closed set, then, by 5.51, F’ is open. By 5.4G, f-'(F’) is open in M,. But since FU F’= M, we have, by 1.3E, SP)US- F=f (Mt). That is, f-'(P)UF~'(F’)= M,. Hence f~'(F) is the com- plement (relative to M,) of f~'(F'). Since f~\(F’) is open, then f~'(F) is closed, which is What we wished to show. The converse part of the proof is left to the reader. If the reader understands the equivalence of the various formulations of continuity, he should have no difficulty in proving the following theorem, SSK. THEOREM. Let f be a I-I function from a metric space M, onto a metric space Mg, (That is, fis a 1-1 correspondence between M, and M,,) Then if f has any one of the following properties, it has them all. (a) Both f and /~! are continuous (on M, and M,, respectively), (b) The set GCM; is open if and only if its image f(G)C.M, is open. (©) The set FCM, is closed if and only if its image /(F) is closed. SSL. perinirion. If f has any one (and hence all) of the properties in 5.5K, we call fa homeomorphism from M, onto M,. If a homeomorphism from M, onto M, exists, we say that M, and M, are homeomorphic ‘The metric spaces [0,1] and {0,2} (with absolute value metric) are thus homeomorphi For if f(x)= 2x, then-f is a homeomorphism of [0,1] onto {0,2}. If f(x)=logx, then fis a homeomorphism of (0,0) onto R (verify). See if you can prove that (0,1) and (0, 1] are not homeomorphic. ‘The final concept in this section is that of “dense subset.” 5.5M_perinimion, Let M be a metric space. The subset 4 of M is said to be dense in ‘M if T= M. (That is, A is dense in M if every point in M is a limit point of 4.) 42 | CONTINUOUS FUNCTIONS ON METRIC SPACES. For example, the set A of rationals in dense in R'. For, by 3.11, every irrational is the limit of a sequence of rationals. On the other hand, R, has no dense subset (except R, itself). For if AC Ry then T= (by 54D and 5.51). Hence, if 4 Ry then Ay Ry, and s0 A is not dense in Ry. Exercises 5.5 1. Use your intuition to decide which of the sets of exercise 2, Section 5.4, are closed in Re 2. For each of fis open nor closed. Prove that any finite subset of a metric space M is closed, Let A and B be subsets of a metric space M. If ACB, prove that Ac. 5. (a) True or false? If A and B are subsets of R! and if ACB, then ACB. (b) The same question with R' replaced by Ry. 6. If ae R', prove that [a,c0) is a closed subset of R' 7. Let F, G'be subsets of a metric space M such that F is closed and G is open in M. Show that F— Gis closed and G~ F is open in M. 8. If00. Prove that 4 is closed, 11. Let fe a continuous real-valued function on the metric space M. Let B be the set of all x€M such that f(x)=0. Prove that B is closed. 12. If A and B are closed subsets of R', prove that 4 x B is a closed subset of R. 13, Give an example of a sequence A,,A3.... of nonempty closed subsets of R! such that both of the following conditions hold: (@) A\DA29.433-* ) NZ A,= 14, Let Prove that fis a homeomorphism of! onto (1,1). 15. Show that R! and R, are not homeomorphic. 16. Let My,Mz,My be metric spaces. If M, and M, are homeomorphic, and if M, and M, are homeomorphic, prove that M, and M, are homeomorphic. 17, Prove that (0,20) (with absolute-value metric) is homeomorphic to (0,1). 18. Give an example of a countable subset of R® which is dense in R?. 19. Give an example of a countable subset of f which is dense in ®. (This isa difficult one.) 5.8 DISCONTINUOUS FUNCTIONS ON R? | 143 20. Let M be a metric space and let AC.B.CM. If A is dense in B and if B is dense in 1M, prove that A is dense in M. 21. Give an example of a set E such that both E and its complement are dense in R' Can E be closed? 5.6 DISCONTINUOUS FUNCTIONS ON R! As an interesting digression from our discussion of metric spaces, we are going to investigate the set of points at which a given real-valued function on R'is discontinuous (discontinuous = not continuous). After the proof of 5.5G we noted that a countable union of closed subsets of R' need not be closed. 5.6A. DEFINITION. The subset D of R! is said to be of type F, if D=Us.,F, where each F, is a closed subset of R! ‘Thus if F is closed, then F is of type F, since we can write F= UF, where Fi=F and Fy=Fy= +++ =3. ‘Any open interval (a,b) is also of type F, since (a,b) is the union of the (countably many) closed intervals [a+ 1/7,b~1/n] for nT with 2/nR', and D is the set of points of R! at which fis nor continuous, then D is of type F,. This, however, requires a little machinery, 5.6B. DEFINITION. Let f:R'>R'. If J is any bounded open interval in R?, we define oLf;J] (called the oscillation of f over J) ast ePID tab £(2)~ hb. f0). Then if aE R', we define uf fsa] (called the oscillation of f at a) to be olf oft] where the g.l.b. is taken over all bounded open intervals J containing a. Clearly wf f;J]>0 for any interval J, and hence o[f;a]>0 for any point a. The number «J /;/] measures, roughly, the distance from the “lowest part” to the “highest part” of the graph of f on J. It is intuitively clear that if fis continuous at a, and J is a “small” interval containing a, then of f;J] must be “small.” We leave the proof of the next theorem to the reader. S.6C, THeoREM. If f:R'—+R! and aR‘, then the following statements hold: (1) IF fis continuous at a, then a f;a]=0, Q) I Fis not continuous at a, then af f;a}>0. Now we start using closed sets. 56D. rHvoeM. Let f:R'->R!, For any r>0 let E, be the set of all a€R! such that olfia]> I/r. Then E, is closed. * For simplicity, we assume throughout this section that fs bounded. Tha is, we assume thatthe range off is bounded subset of R". This isto avoid infinite values for off: All the resus ofthis setion hold also for unbounded f. 144 | CONTINUOUS FUNCTIONS ON METRIC SPACES PRooF: Let x be any limit point of E,. We must show that x £,. That is, we must show that of f:x]> 1/r. To do this, itis sufficient to show that if J is a bounded open interval containing x, then af f:J]> I/r (since of f; x] is the gb. of such af fF). But by 5.5D the open interval (open bail) J must contain a point y of &, (since J contains the limit point x of £,). But then of f:J]> ol fiy]> 1/r and the proof is complete. ‘The result we have been looking for now follows. 5.6E. THEOREM, Let f:R'>R'! and let D be the set of points in R! at which fis not continuous. Then D is of type Fy. PRoor: If x€D, then, by 5.6C, af f:x]>0. For some mEJ, then, we must have oLfix]> 1/n. This proves that D CUS. ;E;, where Ey), is as in 5.6D. Conversely, if XE UP Lym then of f:x]>0 and 50 XE D. Thus D= UE. But by 5.6D each Ey, is closed. Thas D is'@ countable union of closed sets, whichis what we wished to show In exercise 10 of Section 5.1 we gave an example of a function which was continuous at each irrational but discontinuous at each rational, We are going to show that there is no function which is continuous at each rational but discontinuous at each irrational. To do this itis enough, by 5.6E, to show that the set of all irrationals is not of type F,. This involves introducing the notion of category, which hhas great importance in higher analysis. 5.6F. DEFINITION. The subset A of R! is said to be nowhere dense (in R'!) if 4 contains zo (nonempty) open interval. ‘Thus the closed set F is nowhere dense if F itself contains no open interval. For ‘example, the set J of positive integers is nowhere dense. The Cantor set K is another ‘example of a closed set that is nowhere dense. (For K is closed since its complement is the union of open intervals. Furthermore, K is nowhere dense since, according to 1.6D, a chunk of every open interval will be removed in the geometric construction of K.) 5.6G. DEFINITION. The subset D of R' is said to be of the first category if D= USE, where each E, is nowhere dense in R'. If D is not of the first category, we say that D is Of the second category. It follows immediately that any countable set D is of the first category since D is the countable union of one-point sets, and any one-point set is (closed and) nowhere dense. In particular, the set of rationals is of the first category. Furthermore 56H. THroREM. If A and B are sets of the first category, then A\U B is also of the first category, proor: If A=UZ.,H, and B=UZ,\E, where each E, and each H, is nowhere dense, then AU B is'the union of all the E,'s and H,’s (of which there are a countable number). Hence A U B is the first category. ‘On the other hand, the whole space A! is not of the first category. This important result is known as the Baire category theorem (for R"), 5.61. THEOREM. The set R’ is of the second category. 5.8 DISCONTINUOUS FUNCTIONS ON R! | 145 PROOF: Suppose the contrary. Then R'= U%.,F, where each F, is nowhere dense. We may assume that the F, are closed. Otherwise we could consider F, in place of F, since R'= USF, and the F, are clased and nowhere dense. Take any x, not in Fe Since F, is closed there is an open interval 1, about x, which does not intersect F), Let J, be a closed interval with O4). Then E, is closed (why?) (b) For each NES let Fy= Ey A Evai0 Evers =A wBy Then Fy is closed (why?) (©) If x is irrational, then lim,.,.,f,(x)=x(x)= 1 and so there exists NEV such that XE Fy (why?) (@) If xe Fy for some N, then x is irrational (why?). (©) Thus Uj Fiy is precisely the set of irrationals, But this implies that the set of irrationals is of type F, (why?). This contradicts 5.6K, and the contradiction proves the theorem. 5.7 THE DISTANCE FROM A POINT TO A SET Itis fruitful to extend the notion of distance between points to the notion of distance from a point to a set. 5.7A. DEFINITION. Let (M,p> be a metric space. Let A be a nonempty subset of M and let x be a point of M. We define p(x,A), the distance from x to A, as (A) =elb.{o(xy)]¥ EA). For example, if M= R!, A=(0,1), and x=2, then o(x.4)= gid. 2-1 1 Note that although p(x,4)=1, there is no yA such that p(x,y Thus if there is a closest point z in 4 to « (or more than one closest point), then b(x,A) = p(x,2). However, as in the example, there may be no closest point. Here is a connection with closed sets. 5:7B. ruieonem. Let 4 be a nonempty subset of the metric space M and let x be a point of M. Then P(x,d)=0 if and only if xe. PRoor: First suppose p(x.A)=0. Then 0 is the greatest lower bound of the set (o(%)]y€A). Thus if €>0, then ¢ is not a lower bound for the set so there exists y © A with p(x.») 0. Viewed as a function, the distance p(x,A) has interesting properties and applications. (See the exercises.) 5.7D. THEOREM. Let A be # nonempty subset of the metric space M. Define SO)=o(%A) (EM). Then fis continuous on M. PROOF: Let x be a point of M. Given €>0 let x, be any point of M such that Ax,x) <8=€/2. Since I) 00) = Bh pl) there exists yA such that f(x) +¢/2> a(xy).Then A 1P1) < play) + OYA) 0, show that the set of xeM “such that p(x,4) > € is closed. 2/ Show that every open subset of a metric space is the union of countably many closed sets, 3. Let F,,F; be disjoint closed subsets of a metric space M. Prove that there exists disjoint open sets G,,G, such that G,> F,,G,2 F CONNECTEDNESS, COMPLETENESS, AND COMPACTNESS 6.1. MORE ABOUT OPEN SETS 6.1A. As we noted in the previous chapter, if we denote the metric space (0,1) (with absolute value metric) by A, then the interval (0,4) is an open subset of A even though (0, 4) is not open subset of !. Thus whether a given set is open or not depends on the metric space of which it is considered a subset. Indeed, if (M,p) is any metric space and ACM, then, by 5B, A is always an open set in the metric space (4,p) even though A may not be open in . Before we take up new concepts in this chapter, we wish to investigate this phenomenon more closely. Let (A,p> be any metric space and let A be any nonempty subset of M. Then <4,p) is also a metric space. Now if a€ A we must distinguish between open balls in A about a and open balls in M about a. For example, if M,p)=R" and <4,p)=[0,1}, then the ‘open ball B(0; 4] in R! is the interval (~ 4, + 4), while the open ball B{0; $] in A=(0, 1] is the interval (0.4). (That is, in A=[0, 1] the set of all points of A whose distance to 0 is less than 4 is the interval [0,}).) Consequently, let us introduce the following notations: Irae, let Blair] = {xEAlp(a.x) 0 such that B,{a;7,]CG,. Define Gy as Gy= YU By [airy ‘Then Gy, is open in M since G is the union of open balls of M (5.4C). Also, from (*) of G.1A it follows that GyA=G,. Conversely, suppose Gy is open in M and let G,= A") Gy. We wish to prove that Gy is open in A. If a€ G,, then a€ Gy. Since Gy is open in M there is an open ball By{a;7] contained in Gy. But then Byla;rINAcGyNA, which says that Byla;rlc Gy. For each aG,, then, we have shown that there is an open ball B,{a;r] contained in Gy This proves that G, is open in A. For example, if M=R' and A=(0,1], the set G,=(0,4) is open in 4. But G,=4n, (~eo,4), and (~20,4) is open in M. Thus, the Gy of 6.1B can be taken to be (~ 26, }). The importance of 6.1B is that it enables us to give more than one formulation of connectedness. Exercises 6.1 1. Give an example of a metric space M and a nonempty proper subset ACM with the property that every open subset of 4 is also an open subset of M. 2. Let A =[0,1]. Which of the following subsets of A are open subsets of A? @ Gb ©) GD, © HD. Which of (a), (b), and (e) are open subsets of R" Which are open subsets of R*? (Regard R! as a subset of R2) 3. Is there any subset of R! that is an open subset of R72 4, Let A be an open subset of the metric space M. Prove that if BCA, then B is open in A if and only if B is open in M. 6.2 CONNECTED SETS Our intuition tells us that, whatever the definition of “connected set” turns out to be, the interval (0, 1} should be called a connected subset of R! while the union [0,1}U[2,3] should not. The usual way to define connectedness, however, does not immediately appeal to the intuition. It is usually not trivial to prove that a given set is connected even though it looks “connected.” We will list two equivalent properties and then define a set to be connected if it has cither (and hence both) of these\properties. We recall first that in a metric space the sets M and @ are both open and closed. If these are the only subsets of M that are ‘open and closed, we will (ultimately) call Mf connected. 6.2A. THEOREM. Let (M,p) be a metric space and let A be a subset of M. Then if 4 has either one of the following properties, it has the other. (2) Ttis impossible to find nonempty subsets 4,,4z of M such that A= 4,U4,4,0 A,= 0,4, Ay=@. (Here 4, means the closure of 4, in (Mp) 150. | CONNECTEDNESS, COMPLETENESS, AND COMPACTNESS (b) When (4,p) is itself regarded as a metric space, then there is no set except 4 and @ which is both open and closed in ¢4,p). PROOF: We will prove that (a) implies (b). Suppose, then, that (a) holds. If (b) were false, then there would be a nonempty proper subset 4, of 4 such that 4, is both open and closed in A. But then, by 5.51, 4;= 4A, would also be both open and closed in A ‘We now show that 4,7\4,=@. Suppose x M is a point in ay. Then x is the limit of ‘a sequence of points in A,. If x were in Ay, then x would be in 4. Since x isthe limit of a sequence of points in A,, this implies x4, since A, is closed in A. But_then A\NA,#@ contradicting A,=A —A,. Hence if x 6A,, then x Ay, so that A,4,=0. Similarly 4,7.4,= ©. Clearly = 4,UA,. But this contradicts (a), and so (b) must be true if (a) is, ‘Now we will prove that (b) implies (a). Suppose that (b) is true. If (a) were false, then there would exist nonempty subsets 4,,A, of M such that A=4,UA; and 4\0.4;=© = A\(1Ay, Let G=M— A, Then G is an open subset of M (by 5.31). Since A, is disjoint from A, we have A, CG. This proves that G74 =.4,. But then, by 6.1B, 4, is an open subset of 4. Similar reasoning will show that 4, is an open subset of A. Hence j= A~ Az is a closed subset of 4. We have thus produced a nonempty proper subset of A (namely 4,) that is both open and closed in A. This contradicts (b). Thus (a) must be true if (b) is. 6.28, DEFINITION. Let ay, Thus a,'to <0,1>. Surprisingly enough, it may be shown that B is a connected set in R?, We will not give details. This example shows that untrained intuition may not always be helpful in connectedness problems. From the following theorem we can deduce a result useful in’caleulus. 62D. THEOREM. Let f be a continuous function from a metric space M, into a metric space Mz, If M, (the domain of f) is connected, then the range of f is also connected, PROOF: Let A=/(M,) so that f:M,=>4, If A were not connected, then there would exist a nonempty proper subset B of A such that B is both open and closed in 4. But then, by 5.4G and 5.5J, f~'(B) would be a nonempty proper subset of M, that is both open and closed in M,. This would contradict the hypothesis that M, is connected Hence 4 is connected and the proof is complete. ‘The special case of 6.2D in which M, is a closed bounded interval [4,6] and M,, yields the corollary we have been looking for. 62E. conoLLaKy. If f is a continuous real-valued function on the closed bounded interval [a,b], then f takes on every value between f(a) and f(b). PROOF: By 6.2C the interval [a,b] is connected. By 6.2D the range of f is thus connected. The corollary then follows from 6.2C. Here is another application of connectedness. We will show that [0,1) and (0,1) are not homeomorphic. Assume the contrary. Then there exists a 1-1 function f from {0, 1) ‘onto (0, 1) such that both f and f~' are continuous. Let a= f(0). Then the restriction of f 10 (0,1) is a continuous function from (0,1) onto (0,a)\(a, 1). But this contradicts 6.2D since (0, 1) is connected while (0,4) (a, 1) is not. See Figure 23, ‘We next give an interesting reformulation of connectedness, 6.2F. THEoREM. Let M be a metric space. Then M is connected if and only if every continuous characteristic function on M is constant. That is, M is connected if and only if the function identically 0 and the function identically 1 are the only characteristic functions on M that are continuous on M. PROOF: Let M be any metric space and let x be the characteristic function of ACM. Then A =x" (1). Thus if x is continuous, then 4 is closed (by 5.5) since is the inverse image under x of the closed subset (1) of R'. Similarly, if x is continuous, then M—A =x~"(0) is closed, and so A is also open. Thus, if x is a continuous characteristic funetion on M, then 4 is both open and closed in M. But if M is connected, then either A=M or AO, and in either case x is constant ‘The proof of the converse is let t0 the reader. Theorem 6.2F can be used to give a snappy proof of 6.2D. Indeed, suppose f is a continuous function from the connected metric space M, into a metric space My. To 152 | CONNECTEDNESS, COMPLETENESS, AND COMPACTNESS, o 0 r £ 1 connected 2 De tot ° 0 FIGURE 23. show that f(M,) is connected iit is sufficient, by 6.2F, to show that any continuous characteristic function x on f(M,) is constant. But since x and f are both continuous, then xf is a continuous characteristic function on M, (use 5.3D), and hence x°f is constant by 6.2F. It follows that x is constant. Another interesting application of 6.2F is the following. 62G, THeoREM. If A, and A, are connected subsets of a metric space M, and if A\1Aa#@, then A,U Ay is also connected, 6.3, BOUNDED SETS AND TOTALLY BOUNDED sets | 153 PROOF: Let x be a continuous characteristic function on 4,U Az. If x94, Az, then by 6.2F X()=X(%) (EAI) since A, is connected, and x(X)=x(%) (XE AQ) since A, is connected. Hence x is identically equal to x(x) and is thus constant, By 6.2F, A\U4z is connected and the proof is complete. Exercises 6.2 1. Prove that if fis a nonconstant real-valued continuous function on R', then the range of fis not countable. 2. Prove that there is no continuous real-valued function fon R! such that (3) is irrational if x is rational and J (s) is rational if x is irrational 3, Prove that the interval [0,1] is not a connected subset of Ry. 4. True or false? If A and C are connected subsets of the metric space M, and if ACBCC, then B is connected. 5. IA is a connected subset of the metric space M, prove that Z is connected. (Hint Use 62F.) 6. If A is a connected subset of the metric space M, and if AC BCA, prove that B is connected 7. Prove that the set of all points on a minus sign (considered as a subset of R?) is not homeomorphic to the set of all points on a plus sign. (Hint: First decide whether or not the center point of the plus sign can be the image under a homeomorphism of an fend point of the minus sign.) 8. Fill in the details in the following proof that every number ¢>0 has a square root. (a) Let f(x)= x? (0< x <0). Then f is continuous on [0,¢), and 0 is in the range of i (b) If c>0, then ¢<(1-+c)? and (1+) isin the range of f. (6) Therefore, c is in the range of f (why). Hence f(x] umber x, is the square root of ¢. ¢ for some x,. The 6.3. BOUNDED SETS AND TOTALLY BOUNDED SETS 63A. DEFINITION. Let (IM,p) be a metric space. We say that the subset A of M is bounded if there exists a positive number L such that Plxy)0, there exist a finite number of subsets Ay,Apy.nvdy of M such that diam4y 0, 4 can be covered by a finite number of subsets of M whose diameters are all less than «. ‘Some authors use the term “precompact” instead of “totally bounded”. “Totally bounded” is a stronger restriction than “bounded.” 6.3D. THEOREM. If the subset A of the metric space (M,p} is totally bounded, then A is bounded. PROOF: If 4 is totally bounded, then there exist nonempty subsets Ay,Aa.-.sd, of M such that diam, <1 (k=1,....g) and A CUf)4g. For each k="I,...uf, let ay be any point in Ay. Then let D=p(44,4;)+ p(dy.a3)+"-= +p(dy- 24). Now for any points xy’ EA, we have xEA,,y EA, for some i and j (since the A, cover A). We may assume that i0, Given €>0, 4 is certainly covered by 16.3 SOUNDED SETS AND TOTALLY BOUNDED sers | 155 where n is any positive integer such that m(c/2)>2L. Thus, in R!, a set is totally bounded if and only if it is bounded ‘The same result holds true for R". The proof (for n=2) is reserved for an exercise, On the other hand, in R, “bounded” and “totally bounded” are not at all equivalent. For we have seen that every subset of Ry is bounded. However, if BC R, and diam B<4, then B can contain at most one point. A finite number of subsets of R, each of which has diameter <} can therefore cover only a finite subset of R,. Thus 63E. COROLLARY. A subset 4 of R, is totally bounded if and only if 4 contains only a finite number of points We will soon show that the subset £ of defined in 6.3B is bounded but is not totally bounded. In summary then, in any metric space a totally bounded set is bounded. However, in some metric spaces there are bounded sets which are not totally bounded. We now give two important reformulations of “totally bounded.” 63F. DEFINITION. Let A be a subset of the metric space M. The subset B of A is said to be e-dense in A (where €>0) if for every xGA there exists y EB such that p(x,»)is totally bounded if and only if, for every €>0, A contains a finite subset (x),.-..%,} Which is e-dense in A. PRooF: Fix €>0. If A is totally bounded, then A= U7_.4, where diam, 0 such that A contains no finite e-dense subset. Thus if x,€, then the set (x,} is not edense in A, so there exists x64 such that p(j9)> € But then (x,t) is not e-dense in A and so there exists x€A such that plxyx5)> € and p(x,.x,)> €. Continuing in this fashion we may construct a sequence (2) 221 of points of 4 such that p(x,,x,)> € for any j,kEZ,(jxk). But then (x,)21 has no Cauchy subsequence, which contradicts our hypothesis. This contradiction shows that A must be totally bounded and the proof is complete. From 6.3H it follows immediately that the subset E of # defined in 6.3B is not totally bounded. For, since p(g.e,)=V2 if jak, the sequence een... has no Cauchy subsequence. Exercises 6.3 Prove that every bounded subset of R? is totally bounded. Give an example of a bounded subset of {* which is not totally bounded. Give an example of an infinite subset of f which is totally bounded. Prove that every finite subset of a metric space M is totally bounded. . If 0, 7. Let A be an infinite bounded subset of R'. Prove that there is at least one cluster point of A in R'. (Hint: Suppose ACJ, where J; is a closed bounded interval. If J, is, divided in half, then at least one of the halves must contain infinitely many points of A. Call this half Jy, Continue this process.) This result is known as the Bolzano- Weierstrass theorem. 8. Give another proof of the Bolzano-Weierstrass theorem, beginning as follows: Let A be an infinite bounded subset of R'. Let (a,)$., be a sequence of distinct points of A. Then (4,}2.4 contains a Cauchy subsequence (why?). Now finish the proof. 6.4 COMPLETE METRIC SPACES In 2.10D we saw that in the metric space R! every Cauchy sequence of points in R! converges {0 a point in R'. In 4.3F we noted that there are metric spaces 0 there exists V E/ such that Als! 3) N. That is, [s— since (mma), a which implies I= ce (WEN). Thus, if n>, Ks ila= ifs” Thus, for some A >0, 09] +5 fy ). ® Now, for any kE J, we have from (1) [nfs] < Js—s Ice (mm>N), Hence (for fixed k) the sequence {31}, is a Cauchy sequence in R? and so, by 2.10D, converges to a number 5, € R'!. Let s denote the sequence (5,)%.,. First we will show thats @, From (2) we have Ts N). Hence for LEI, Bloprsnyrce (nm>N) Letting m-s00 (and using lim 298{"= 5p. 2.7A, and 2.7E), we have Zor-sree (n> N:LED, and so SoP-yise a0), 158 | CONNECTEDNESS, COMPLETENESS, AND COMPACTNESS, But this says that p(s”,s)=|s—sll,<¢ if n>, which proves that s,s... con- verges in © to the point 5. This completes the proof. GAC. THEOREM. If 0 there exists, by (b), an integer NET such that diam Fy N we then have p(dya,,) diam F, for n sufficiently large. Hence 5 cannot’ be in nZ.,F,. This completes the proof. 64E, We are now going to discuss a class of functions called contractions. Although their usefulness will not be immediately apparent, they tum out to have important applications. In a later chapter we use a result on contractions to prove an existence theorem for differential equations. To simplify notation in this discussion, if T:M—M and if xe M, we will write Tx instead of T(x). We will also write 7? instead of T°T, 7? instead of TeT?, ete DEFINITION, Let (M,p) be a metric space. If T: MM, we say that T is a contraction on M if there exists a R with 0n, say, $0 that m= n+p) we have PS Xm) Pur Xap) < Puen) + PKs Xnz2) ++ +P tp— ate) = ol Tq, Tey) + (TH eg, THR) += + (TA beg, THe) Thus by (1) Pp Xn) $ @"OL% x4) +a" pl xpyy) #22 Hat? Ip( x9.) Kat A(xyxi)(Ibatatt and hence @"P%%1) o Pn Sin) $a Since lim, ,,a"=0, it follows easily that p(x,.%_) can be made arbitrarily small by taking n (and hence m) sufficiently large. Thus (x,)_1 is Cauchy, Since (by hypothesis) M is complete, there exists xEM such that lim,_...*,=.x. Therefore lim, ...7x,= Tx (why?) But Tx,= 2,4 and so { Tx,)y, being a subsequence of (x,). must converge 10 lim, 4.0%. It follows that Txr™ x, 50 that x is a fixed point, All that remains to be shown is that ify €.M,y%x, then y cannot be a fixed point. For suppose the contrary. Then Ty=y and so (since Tx=x),9(x,») =0(Tx, Ty) < aplx,y). Since plx,y)#0 this implies 1< a which is a contradiction. Hence Ty #y and the proof is complete. 140 | CONNECTEDNESS, COMPLETENESS, AND COMPACTNESS Exercises 6.4 1, Prove that R, is complete. 2. Prove that the interval (0, 1) with absolute value metric is not a complete metric space, Prove that (0,1) with the metric of R, is a complete metric space. 3, Prove that R? is complete. Prove that € is complete. (Model your proof after the proof that ? is complete.) SI Taja8 (0 is said to be compact if (M,p> is both complete and totally bounded. For example, the metric space {a.6] (with absolute value metric) is totally bounded and, by 6.4C, is complete. Hence [a6] is compact. The space R? is complete but not totally bounded. Hence R' is not compact. The metric space (0,1) (with absolute value metric) is totally bounded but not complete and hence is not compact. From 6.3E we see that an infinite subset of Ry cannot be compact. We leave it to the reader to show that every finite subset of R, actually is compact. (See Exercise 2.) ‘A very useful reformulation of compactness is continued in the following. 6.5B. THEOREM. ‘The metric space (M,p> is compact if and only if every sequence of, points in M has a subsequence converging to a point in M. PROOF: Suppose first that M is compact and that (x,)%_, is any sequence of points in M. Then, by 6.3H, since M is totally bounded, the sequence (x,)% has a Cauchy subsequence (x,,}f-1- But (&,,}f.5 converges to a point in M since M is complete. ‘Thus if M is compact, then every sequence in M contains a convergent subsequence. 65 compact meric spaces | 161 Conversely, suppose every sequence in M has a convergent subsequence. Then, by 63H, M js totally bounded. To show that M is complete we must show that every Cauchy sequence {x4}. in M converges to a point of M. By assumption, {x,), has a subsequence (x,,)j., which converges to a point x in M. Since (x,)‘_, is Cauchy, it is then not difficult to show that (x). itself converges to x. Hence M is complete and so M is compact. This completes the proof. Here is a useful corollary, 65C. COROLLARY. If A is a closed subset of the compact metric space (M,p) then the metric space (A,p) is also compact. PROOF: Any sequence (x,)®.1 of points of A is a sequence of points of M and hence, by 6.5B, has a subsequence converging to a point x in M. But then x is a limit point of A and so x€A (since A is closed). Thus any sequence in A has a subsequence converging to a point in A. By 6.5B, A is compact. In the other direction we have the following, 65D. THEOREM. Let 4 be a subset of the metric space (M,p>. If (4,p) is compact, then A is a closed subset of (Mp). PROOF: Let xEM be any limit point of A. Then there is a sequence (1x4). in A converging to x. But (x,)%, is a Cauchy sequence in A and so, since 4 is complete, (,} Fey converges to a point in A. This point must be x and so x. Thus 4 contains all'its limit points and so A is closed. (Note that the proof uses only the completeness of A. However, itis in the context of compactness, rather than completeness, that the result, will be useful) 65E. If M is any set, the family $ of subsets A of M is said to form a covering of M if MCU,4eqA. We will be primarily interested in open coverings—that is, coverings F of 4 metric space M where each 4 € is an open subset of M. For example, the family of open intervals (1/n,1~1/n) for n=3,4,5,... is an open covering of the metric space (0,1) (with absolute value metric). Note that there are infinitely many open sets in this covering of (0,1) and that no finite number of these sets form a covering! On the other hand, the reader should experiment with some open coverings of {0,1} We will find that if ¥ is any open covering of [0, I, then some finite number of sets in S will also form a covering, A proof of this will be given shortly. 6.F. DEFINITION. A metric space M is said to have the Heine-Borel property if, whenever is an open covering of M, then there exist a finite number of sets GyoeeG,€F such that (G),...,G,) is still a covering of M. ‘The Heine-Borel property is sometimes stated as “every open covering of M admits a finite subcovering,” From the first example in 6.5E we see that the metric space (0,1) does not have the Heine-Borel property. The next two theorems show the reason the Heine-Borel property is discussed in this section. 162 | CONNECTEDNESS, COMPLETENESS, AND COMPACINESS. 6.5G. THEOREM. If M is a compact metric space, then M has the Heine-Borel proper- yt PROOF: Suppose the contrary. Then, for some open covering F, no finite number of sets in § form a covering of M. Now M is totally bounded and hence may be wiitten as the union of a finite number of bounded subsets each of whose diameter is less than | But then one of these subsets, call it 4,, cannot be covered by a finite number of sets in . (Otherwise, all of M could be covered by a finite number of sets in F.) But, since diam =diam., (verify) 4; isa closed subset of M (5.5E) whose diameter is less than | and which cannot be covered by a finite number of sets in. Since 4, is itself totally bounded, the same reasoning shows the existence of a subset 4, of A, with diam4,-<} and such that 4, cannot be covered by a finite number of sets in 5. Thus 4,c.4y, diam, <}, and Z, cannot be covered by a finite number of sets in $. Continuing in this fashion we can show, for any n€ I, the existence of 4, cM such that 4,3 4,> 34,3 +++,diamA, <1/n, and such that no finite number of sets in $ form a covering of any 4,. By 64D, there is precisely one point x in 1. ,4y. Now, since is a covering of M, there isa set'G in $ such that x € G. But G is open (Since & is an open covering) and so there exists an open ball B[x:r]C G for some r>0. But if NEI satisfies 1/N n, such that x, Bx: 4]; indeed, for any k I there exists m,> mg_ stich that a © B LX: 1/4]. This subsequence (x, }f-1 of (x) a1 elearly converges t0 the point x i'M, and the proof is complete. We list one more condition equivalent to compactness. 6.51. DEFINITION. A family 5 of subsets of a set M is said to have the finite intersection property if the intersection of any finite number of sets in is never empty. (That is, if whenever Fy, Fy... FES then FN FN +++ Fy #S.) * The Heine-Borel property is often taken a the definition of compactness, since the Heine-orel property «does not depend directly on the metric but rather on the notion of open sei which is available in a class of Spaces more general than meric spaces. 64 CONTINUOUS FUNCTIONS ON COMPACT METRIC SPACES | 163 Thus the family of all closed intervals [~1/n, 1/n] has the finite interseétion property So does the family of open intervals (0,1/n) 65). THEOREM. The metric space M is compact if and only if, whenever is a family of closed subsets of M with the finite intersection property, then N peg FO. PROOF: Suppose first that M is a compact metric space and that $ is a family of closed subsets of M with the finite intersection property. We must show that O Fro, 0) For each F EG let G= F'=M— F, and let 9 be the family of all such open sets G. If Fy FaevosF, ate any finite number of sets in F, then FAN 0B, =M=(G\U GU" UG) @ by (1) of 12H. Similarly, A F=M= UG. @) cas cae By hypothesis, the left side of (2) is not empty, and so M~(G,U +++ UG,)#@. Hence no finite number of sets in § form a covering of M. Since M is compact, it follows from 6.5G that § itself is not a covering of M. Hence the right side of (3) is not empty. This proves (1), which is what we wished to show. We leave the proof of the converse to the reader. Exercises 6.5 1. Use 6.58 to prove 6.5C. 2. Prove that every finite subset of any metric space is compact. 3. Prove that a subset 4 of R? is compact if and only if A is closed and bounded. 4. If A and B are compact subsets of R'!, prove that Ax B is a compact subset of R*. 5. Let f be @ continuous real-valued function on (a,b). Prove that the graph of fis a compact subset of R? 6. For each x in (0,1) let Z, denote the open interval (x/2,(x+1)/2). Show that the family $ of all such J, is an open covering of (0,1) which admits no finite subcovering of (0,1). 7. Add two appropriate sets to the family $ of the preceding exercise to form an open covering X of {0,1}. Show that ‘ does admit a finite subcovering of (0, 1} 8, Give an example of a connected subset of R! that is not compact. 9. Prove that a connected subset of R, is compact. 10. Give an example of a closed bounded subset of which is not compact. 11. Prove that the metric space M is compact if and only if every infinite subset of M has a cluster point in M. 66 CONTINUOUS FUNCTIONS ON COMPACT METRIC SPACES Because of the tremendous importance of the following theorem we give two proofs, cach based on a different criterion for compactness. 66A. THEOREM. Let f be a continuous function from the compact metric space M, into the metric space M,. Then the range f(.M,) of f is also compact. 164 | CONNECTEDNESS, COMPLETENESS, AND COMPACTNESS PROF I: Let @ be any open covering of f(M,). For each GE@ the set f~"(G) is, by 54G, an open subset of Mj. The family of ail such sets f~'(G) for G € 8 is therefore an open covering of M,. Since M, is compact, a finite number of these sets—say I7'(Gy)--.f” “G,)—also form a covering of M,. But then G,,...,G, form a covering of ‘J(M,). Thus there are a finite number of sets in'. which form a covering of f(M,), and 80 f(M;) is compact. PROOF 2: Let {4}, be any sequence of points in f(M,). For each n€J choose SEM, such that /(%)=y_. By 65B, (x,}%., has a subsequence (x) {1 which converges to a point x in M,. Since fis continuous on M,, it then follows from 5.3C that Tim, sof %q)= fC). That is, ()4,)P, converges to f(XEf(M,). Thus any sequence (Je: in CM) has a convergent subsequence and so, by 6.5B, f(M,) is compact. Since compact spaces are bounded the following corollary is immediate. 6.6B. COROLLARY. Let f be a continuous function from the compact metric space M, into the metric space M,. Then the range f(M,) of fis a bounded subset of M,. AL this time it is convenient to introduce the notion of “bounded function.” 6.6C. perINITION, Let {be a function from a set A into a metric space M. We say that the function fis bounded if its range f(A) is a bounded subset of M. ‘Thus 6.6B states that a continuous function on a compact metric space M, (into a metric space M;) must be bounded. When M,= 2! and M, is the closed bounded interval {a,6] we thus have corollary 6 6D. 66D. conoLLary. If the real-valued function f is continuous on a closed bounded interval in R!, then f must be bounded The reader should construct examples to show that 6.6D is no longer true if either of the words “closed” or “bounded” is removed from the hypothesis. 6.6E. If f is a real-valued function on a set A, we say (quite naturally) that f attains @ maximum value at aGA if S(a)a f(x) (xeA), For example, if fet (-10 and all xeM. Hence f(x) < L~1/N for all x€M. But this says that L—1/N is an upper bound for f(M), which is a contradiction, since L was supposed to bee the least upper bound. Thus we must have L € f(M), and the conclusion follows as in proof 1 R! and f attains a maximum value at a€ A, then Sla)= hab, f(x) 2. Show that if 1 Lex? then f attains a maximum value but does not attain a minimum value 3. Give an example ofa continuous bounded function on (~ oo, 0) that attains neither maximum nor 2 minimum value 4. Give an example of a real-valued continuous function on (0,1) that attains a minimum value but does not attain a maximum. 5. If fis continuous real-valued function on the compact connected metric space M. prove that f takes on every value between its minimum value and its maximum value. SQy= (-@0, there exists 8 >0 such that V@)-fai2. ‘Thus (even though g is continuous at the point 10 as well as at the point 1) the number 8=} is usable at a=I but not at a= 10 as a 5 corresponding to «=2 tis not difficult to show, in fact, that there is no one 8 >0 such that the statement |g(x)-g(al<2_ (Ix a]<8) Q) is true for all a& R!, For g(x)— g(a)=(x~a)(x +a). Suppose there were a 6 for which Q) held for alla. We would then have for a>0 and x=a+6/2, 0, é é x)-sla)l=bx~al-[xtal= $a B]< |a(s)~slala|s—al-e+al= $-[2a+ 5] <2 This would imply ob<2 forall a>0, and this is clearly false. Hence, for this function g, corresponding to «=2 there is no 8 that will “work” for all a simultaneously. (Nevertheless, g is continuous at each aE R') If a continuous function is such that, given ¢, we can always choose 5 so that 5 depends only on « but nor on a, then we say thet the function is uniformly continuous, ‘We now make this precise for functions on metric spaces, 6.8B, DEFMSITION. Let (M,,p,) and (M,,9,) be metric spaces. If f:M,>M,, we say that fis uniformly continuous on M, if given e>0, there exists § >0 such that pl Sf(@)]}0 there exists 6 >0 such that [f(x)-flall0 there exists r>0 (depending on a) such that elf] <5 [oi(xa)0 choose 8 >0 such that mL] <€ [nice <é:xyed] « 170 | CONNECTEDNESS, COMPLETENESS, AND COMPACINESS (We can find such a 6 since f is uniformly continuous on A.) Now, since (x,}2.1 is Cauchy in A there exists N€ J such that PAS) <8 (mn > N) Q) From (1) and (2) it follows that PLS mS (%)]0 choose 6, such that LFS] From (3) we then have {since f(x)= F(x) and f(y)= F(»)] pol FOF] SS () We conclude from (4), (5), and (6) that pil F(a), F(b)]0 find 6 >0 such that fsinx—sinal can be discrete even though p is not the discrete 176 | CONNECTEDNESS, COMPLETENESS, AND CONPACTNESS metric (that i, the metric taking only the values 0 and 1), Indeed the space (udbder)s with absolute value metric, is discrete. In each of the following exercises, M is a metric space. EXERCISE I. Suppose ) G is open whenever © is a family of open subsets of M. Prove that M is discrete. EXERCISE 2. Suppose that every subset of M is either open or closed. Prove that M hhas at most one cluster point. Exencise 3. If M is infinite (that is, M has infinitely many points), prove that M contains an open subset G such that G and G’ (the complement of @) are both infinite. EXERCISE 4. If M is infinite, show that M has an infinite subset A which, with respect to the metric for M, is a discrete space. EXERCISE 5, Suppose that for every open subset G of M, the closure G of G is also open . Prove that M is discrete. ‘MISCELLANEOUS EXERCISES. |. Let 0,0EM. (©) ANG#O for every nonempty open subset G of M. 6. Let M be a metric space. We have defined the open ball B[a;r] as Blasr] = (xe M|o(a.x)) (a) Show that B*[a;,] is a closed set (b) Show by example that B[.a;r] is not necessarily equal to B*[a:,}. That is, the closure of the open ball may not be equal to the closed ball. 6.9 NOTES AND EXERCISES FOR CHAPTERS 4, 5, AND 6 | 177 (©) Give an example of a metric space in which there is an open ball that is closed set but is not a closed ball. (Use a subset of R? which contains 1,0), <= 1,0), and an appropriate interval on the y axis.) 7. Let M be a metric space which has a countable dense subset. If ACM, prove that A has a countable dense subset. 8. Suppose f:R'R', Prove that f is continuous at aE R! if and only if whenever {%,)a1 Converges to a, the sequence (f(x,)) 1 has a subsequence converging to F(a). 9. Let M,Mz be metric spaces and suppose f:M,—»M,, Prove that f is continuous if and only if f(A)c FAY for every ACM. 10. Give an example of a continuous function f from a metric space M into a metric space N such that SAA PA) for some ACN. IL, Let 4 be a subset of the metric space M. Define F)=0(%.A) (EM). Prove that fis uniformly continuous on M. 12, Let A be a connected metric space with at least two points. Show that there exists a continuous real-valued function on A that is not constant. Use this to prove that A is uncountable. 13. A map of the state of Iowa falls on the ground (flat) somewhere within the state. Prove that there is a point on the map that is directly above the corresponding point on the ground. 14, Let f be a continuous function from (0, 1] into (0,1). Prove that f has a fixed point 15. Let'A be an infinite subset of the compact metric space M. If the space A is discrete, prove that A is not closed in M. 16. Let M be a compact metric space and f:M—>M. Assume PLFOSOI] <0) Cr EMx#y) . (Start by showing that p{x.f(x)] Prove that there exists x= M such that f(x)= attains a minimum on M.) 17. Let M be a compact metric space in which the closure of every open ball B[a;7] is the closed ball B*(a;7], Prove that every open ball is connected SKETCH OF PROOF: Prove by contradiction. Suppose B= B{a;r] is not connected Then B= CUD where C and D are nonempty and open in B. We may assume aeC. Let He)=0(ax) (x= B-C). If gb. f(x)=5, there is a point dD such that f(4)=s. The point d is a limit seme point of B[a;s]CC. Fill in the details and finish the proof. 18, Let M be a compact metric space. Suppose f: M—M is such that ale) (x.y). Choose « such that pl /(x),/Q)I> (x.y) +2«. Deduce @ contradiction. Thus fff) =p(x.y) for all x,y ©. Finally, show that f is onto M. CALCULUS 7.1 SETS OF MEASURE ZERO In the next section we define the Riemann integral—the integral considered in elementary calculus courses. We will see that a bounded function f has a Riemann integral provided f is continuous, at “almost every” point, The precise meaning of “almost every” will be defined in terms of the following concept of set of measure zero. If J is an interval of real numbers, we denote the length of J by |J| TIA. DEFINITION. The subset E of R! is said to be of measure zero if for each ¢>0, there exists a finite or countable number of open intervals /,,f,,... such that ECU yl, and El, 0, E can be covered by a union of open intervals whose lengths add up to be less than ¢. It is obvious, then, that a set consisting. of one point has measure zero. ‘The following result is very useful, 7.1B. THEOREM. If each of the subsets £),E,,... of R! is of measure zero, then UE, is also of measure zero, ROOF: Fix €>0, Since E, has measure zero, for each EJ there exists a finite or countable number of open intervals that cover E, and whose lengths add up to less than ¢/2%. The union of all such open intervals (for all n€1) then covers U®.,Eq, and the lengths of all these (countably many*) intervals add up to <¢/2+«/2+""> =e. Hence ULE, has measure zero. Since one-point sets are of measure zero we deduce the following corollary. 7.1C. conottary. Every countable subset of R! has measure zero, + We are using 5. cd va | cacas In fact, there are even uncountable subsets of R! that have measure zero. In Chapter 11 we will be able to show that the Cantor set of I.6D (which is uncountable) is of measure zero. On the other hand, a nonempty open interval (no matter how small) is never of measure zero. (See Exercise 3.) ‘We now make “almost every” precise. 7.1D. DEFINITION. A statement is said to hold at almost every point of [a,b] (or almost everywhere in [a,b)) if the set of points of {a,6] at which the statement does not hold is of measure zero. Thus “fis continuous at almost every point of {a,6]” means the same as “if E is the set of points of [a,b] at which f is not continuous, then E is of measure zero.” We could also say “fis continuous almost everywhere in [4,b].” Exercises 7.1 1. IFA is not of measure zero, if B CA, and if B is of measure zero, prove that A—B is not of measure zero. 2. If a<6, prove that [a,b] cannot be covered by a finite number of open intervals whose lengths add up to less than b—a. Use the Heine-Borel property of [a,b] to deduce that {a,6] is not of measure zero. If. a<_, prove that (a,) is not of measure zero. 4, (a) Show that the set of all rational numbers is of measure zero. (b) Show that the set of all irrational numbers is not of measure zero. 5. True or false? If f is continuous on {0,1}, and if g(x)= f(x) at almost every x€[0,1}, then g is continuous almost everywhere in [0, 1}. 7.2 DEFINITION OF THE RIEMANN INTEGRAL ‘Throughout the remainder of this chapter we consider only real-valued functions. 7.2A, pEFINITION. Let § be any bounded interval of real numbers, and let f be a bounded (real-valued) function on $. We define MLf; 4}, mLfi$] and of f: $1 as MRS] = bub SC). m[ PG ]= eb. SC. [FE] = ML (Thus of f;$] is exactly the same as in definition 5.6B save that we now do not require that 4 be open.) If a is a point of §, we define w{ fra] as of fa]=allb.of fl] where the gi.b. is taken over all open subintervals J of $ such that a€J.* (This is also consistent with 5.6B.) 7.2B, periniTion. By a subdivision of the closed bounded interval [4,6] we mean a finite subset {x.x;,..-,%,) Of [4,6] such that a=xy Lfrol- Note that if fis continuous and nonnegative valued on {a,b}, then U[j:o] is the sum of the areas of n rectangles each of which has one of the Jy as base and whose height is equal 0 max.e,, f(x). That is, U[fra] is the sum of the areas of the “circumscribed rectangles” as pictured in claculus texts, Similarly, L{ fsa] is the sum of the areas of the “inscribed rectangles.” This geometric interpretation makes the following result quite plausible (atleast for continuous functions) 72D. LEMMA. Let f be a bounded function on [a,b]. Then every upper sum for f is greater than or equal to every lower sum for f. That is, if ¢ and r are any two subdivisions of {a,b}, then Ufo] > L[ fir}. PRoor: We will show that if o* is any refinement of o, then UL fio] > UL fio"). Q It is enough to prove this in the case where o* is obtained from 0 by adding only one point of subdivision. (For we can then apply induction.) Thus we may suppose that o has component intervals Dj.-yfyvovf, and a* has component intervals Iyeelf. I" oly where K=J2Ulg* and |hl=ligl+lig*|. Since Ich we have MUfif]< ML felgh Similarly, MUfs1f*]< MUf:1,) Thus ULfor]= MLAS HI ML fle Pel + ML Slt] Ue) irk <> MAG) ML fede Ul Ue) = UL fe] Sok which proves (1). Similarly, if +* is any refinement of +, it may be shown that Lift] Ul four] > LE four] > LL fer]. ‘This proves the lemma. * For any bounded interval J\J| denotes the length of J vee | carcuws 7.2E. From the preceding lemma it follows that eld. UL fio] >lub.L[ fo], a) where the giLb, and Lu.b. are both taken over all subdivisions o of [a,b]. For, i is any subdivision of [a,6), then the lemma shows that L[:1] is a lower bound for the set ofall upper sums UL fsa}. Hence Liz] 0, there exists a subdivision o of [a,b] such that UL fio] 0 there exists @ such that (1) holds. Then, since fs ULfie] and f's> LL Ko}, we have fits fire Since € was arbitrary, it follows that ie Ne I and hence, by (2) of 7.26, that This proves fE 9.[a.6}. Conversely, suppose f€ {a,b}. Then % = ghb.UL fro] rea | carcuws Given €>0 we may (by definition of gi.b.) choose a subdivision o such that oie [s+ §> UL ho}. Similarly, we may choose a subdivision + such that S- $< ups, Hence, LLfir]+ $> Ufo] By (1) and (2) of 7.2D we then have Lfour]+$>U[ four $. This is equivalent to (1) (with 64+ in place of 0). Exercises 7.2 1, Let f(xy=x (00 there exists 8 >0 such that WO) -SO0 there exists a subdivision 0 of {4,6} such that, if J, is any component interval of @, then MUS mln] < ba @ For this 0, UL fie] -L[ fro] 0. Hence, E= U%_\Eq where E,, is the set of all x€[a,b] such that wf f:x]> 1/m. To prove that E is of measure zero it is sufficient, by 7.1B, to show that each E,, is of measure zero. This we will now do. Fix m. Since FE 5 [a,b], given €>0 there exists by 7.2 a subdivision o of (a,b) such that U[f;o]— L[f;o]<¢/2m. Thus if J,,...,/, are the (closed) component intervals of 0, wwe have Dole 3 MspI B ml i] Male UL fe] ~ Lf and hence, Delia alsa: ® Now £,,=EgU Ex* where £3 is the set of points of Ey that are points of the subs «, and x= E~Es, Obviously, ECJ,U~-* Ud, where the J, are open subintervals such that Ji|++-"'+[,l<¢/2 (since there are ‘only a finite number of points of subdivsion). But if x £3*, then x is an interior point of some J. Hence, of 4] Deol fex]> 1/m, If we denote by Tageorrdy, those of the component intervals of o that contain a point of E3* (in their interior), we have Male FU) 0 choose me J such that (b—a)/m<¢/2. If E,, is defined as in the first part of the proof, then, by hypothesis, £,, is of measure’ zero, Hence, Eq USL dt, where each J, is an open subinterval of [a,6] and Susser? LF [ab But £, is closed in R! by 5.6D. Hence, E,, is a closed subset of (4,] and is thus compact. Therefore a finite number of the J,~-Say J, -cover Ey. Now [ab}-(h,0-+ Ula) That is, isa union of closed intervals J),..-.Jp [ab] =1,,U-+ Uh, Ud,Us* Udy Since no interval J,(i=1,...,p) contains a point of Eq, there exists by the lemma a subdivision +, of J; such that UL fix) LLJfs7,]<|J)]/m. Now define the subdivision a of [2,6] as o=7,U ++ Ut. Then the component intervals of 0 are the component intervals Of theses together with T,,...s/q- Hence, UY fe]~ [fia] & (YLsin] p+ DMs [sla] —m[ 0,] ah Sur Be of Feta] Hal <4 eof f(a] Ml <$+o[fi[ab]] IF [ad] By 7.2G we have © 91{a,6], and the proof is complete. Exercises 73 1. Which of the following funetions f are in @0, 1]? (a) The characteristic function of the set (0, , Bs dys---s1) (b) f(v=sin(l/x)— O0 res | carcuwus 7.4 PROPERTIES OF THE RIEMANN INTEGRAL Al the results in this section are used in proving fundamental calculus theorems as well as in the solution of standard calculus problems. The reader should interpret » geometrically each of these results by thinking of f f (for continuous nonnegative- valued f) as the area under the curve y= (x) from x= to x=, 74A. THEOREM. If FE Rla,6] and a f° s so that ULfa}+ user] fr Taking greatest lower Bounds asin the fist part of the proof we obtain fe fofes @ ‘The theorem follows from (1) and (2). * More precisely, the restriction off to [a] is in faye} 7.4 PROPERTIES OF THE REMANN INTEGRAL | 189) Any feshinan student wilinstinstively wete f[sea as aff'd Here is he justification, 74B, TueoREM. If f= %[a,6] and A is any real number, then ME @ [a,b] and BAD [yorfs PRoor: If X=0, the theorem is obvious. Suppose A>0. Since Af is continuous at ‘every point where fis continuous, it is clear that Af R{a,b]. Since A>0, if J is any interval contained in [a,6}, then M[MJ]=AML Aid] (verify), and so, for any subdivision o of (2,6), UPVio] =AU[ fro}. It follows easily, on taking the g.L.b. of both sides (over all 9) that Pere) f Meal f (>). a) Hence, the theorem is proved for A>0. Now for any J we also have M[-fd]=—m[ AJ] Henee, fi -f)= ghb.U[ fio} ehb. (-L[fio]} =~ bt[io]=— fs That is, Lon=- L's 2 It w<0, then A= ~w>0 and so, by (2) and (1) (since we & a,b) fol This completes the proof. Oya ['a-afirau fs From theorem 7.4C it follows that [0itee $x?)dr= [Vie dxt [Pra 7AC. THEOREM. If fE Rfa,b],g€ A[a,b], then f+ gE A[a,b] and [loro fr fie wo | carcuus Proor: By 7.3A, the sets E, and E, of points at which f and g, respectively, are not continuous are both of measure zero, Fence, by 7.1B, the set £/U E, is of measure zero. But if x€[a,6]—(E)U &,), then f, g, and hence fg are continuous at x. Thus fg is continuous at almost every point in [a,6), and s0 f+ 2 fa,6}, If J is any interval contained in (a,b), and if yEJ, we have f(»)+ (9) < MLLJI+ M{g:J|, Hence, M[f+g:J]0 there is, by 7.2G, a subdivision a, of [a,b] such that UL fii] 0 almost everywhere, we have, by 74D, 7.4C, and 7.4B, o< fe-=[ter-ni= f'et [n= fie fr This proves the corollary. almost everywhere (a< x <6), then TAF, corourary. If f= ®[a,6), then | f|E Rfa,b] and [f's|sfin PROOF: Since [| will be continuous at every point where fis continuous (exercise 4 of Section 5.1), it is clear by 73A that [J € 6.[a,6}. Now, since f(x) <|f(2)|=|/(3) for all x€[,6}, 74 implies flr fin 0 Since — f(x) <|f(x)| for all x€[a,6}, 74E implies ~ fis fiin 2 The corollary follows from (1) and (2). 14G. If b0 for some c€[a,6}, prove that f[serac>o 8. IF fis continuous on (a,b) if F()20 (ae x0, the curve (1) is “ascending” at ¢; if f'(c) <0, the curve is “descending” at c: while if f"(c)=0, the curve has a horizontal tangent at c. Also remember that if fc) exists, then the curve (1) is “smooth” at x=. 7.5G. Suppose f'(x) exists for every x in some interval J. If e€J and if tim LO-FO a) ‘exists, we say that f has a second derivative at c. We then denote the limit (1) by S"(e}—that is, F9)-F(O iO ie ae —coee Similarly, the nth derivative of f at ¢ is defined as P(0)= Hn Z ee) provided that f(x) exists for all x in an interval containing ¢ and provided that the limit exists. From 7.5B it then follows that if f((c) exists, then f™) is continuous at c. We assume that the reader is familiar with the geometric significance of the second derivative f°—namely, that the graph vasa) is concave up at those points where f"(x)>0, and concave down at those points where F(X) <0. 7 DERVATIVES | 199 Exercises 7.5 1. Prove that the derivative of a constant function on [a,6] is the identically zero function on (a,b) 2. IF f(y has a derivative at ¢, if bE R!, and if g(x)=of(2) for all x in an interval containing c, show that &(c)™ Of'(c). 3. Find /'(x) by the chain rule if (@ f(x)=sine? — (-m0 where af(0). 8. Let J(xy=x_ if xis rational, #(x)=sinx if x is rational Prove that /"(0)= 1. 9. True or false? If fis a function on [a,b if cE[a,b], and if f"(c) >0, then f is strictly increasing on some open subinterval of [a,6] containing c. 10. If fis @ real-valued function on [a,b] and if f has a right-hand derivative at c€[a.b}, prove that fis continuous on the right at c 200 | catcuwus 7.6 ROLLE'S THEOREM Maximum and minimum problems are a very important part of elementary calculus. For our purposes, however, we need only the following fraction of maximum-minimum theory. 7.6A. THEOREM. Let f be a continuous real-valued function on the closed bounded interval (4,6). If the maximum value for f is attained at ¢ where a0. then IQ)-IO), . LE-F() S lim 0 and so 0 for 0<|x~0 and hence, f(2)—f(c) >0. This contradicts the hypothesis that f attains a maximum at c. If f(6)<0, then Sle fo 1 ) co for 0<|x-c|< a. If x€(e~8,.c), then x—e<0 and hence, f(x)~ f(e)>0. which is again a contradiction. Hence, f'(c)=0. An obvious modification of the proof will establish the following. 7.6B. THEOREM. Theorem 7.6A remains true with “maximum value" replaced by “minimum value.” If the curve x) (asx0 for some x€(a,b), the maximum value of f on {a,b] (remember f attains a maximum value at some point of [a,b] by 6.6G) will not be attained at a or b since f(a)= f(b)=0. Hence. f will attain its maximum value at some c€(a,6) and the theorem follows from 7.6A, If J§()<0 for some x in (a,b), the theorem follows from 7.6B. This completes the proof. ‘We wish to emphasize the fact that the proof of Rolle’s theorem depends on the theorem that a continuous function on a closed bounded interval attains a maximum. and a minimum value. 1 Soe exercise Id of Section 4. 78 ROWES THEOREM | 201 Note that in 7.6C we do not require that /” exist on [a,6] but only on (a,6). Thus if f(=Vize (-10, g cannot attain its minimum value at b. Thus a0 for all x in J except possibly the end points of J (if there are any), then fis strictly increasing on J (and hence, f is 1-1). PRooF: If a,beJ with a0 by hypothesis, and hence, f()~f(a)>0. That is, if a), then fand fg satisfy the hypotheses of 7.7C. 4. If f()=0 for all x in (a,b), prove that fis constant on (a,6). 5. If fis continuous on (2,6) if f(x) exists for a0 (asx6d) If g is the inverse function for f, show that g is continuous on [/(a).f(B)} 7.8 FUNDAMENTAL THEOREMS OF CALCULUS We begin by asking the following question. What must we know about a function fon an interval [4,6] in order to be sure that fis the derivative of some function F on (4,6)? We see from 7.6E that even such « relatively “nice” function as that given by (-1O let J, denote the interval [xo,o-+]. Then, with of f:/,] as in 7.24 we have Fo) Hence, by 7.4E, AM) oF] ] sf" Vastra) +o £4] [Fin] < M9 < Ms) +e[ Sih] (CEM) Dividing by h we have Foot) FO) $659) ~6[ ft] SS FFL Ft] “ {Equation (1) ean be established for 4<0 in identical fashion] But, since by hypothesis f is continuous at xp, we have limel ft )=0 @ The theorem then follows from (1) and (2) It should be clear that 18A is a consequence of 7.88. We now head toward the second fundamental theorem of calculus that justifies cvaluating an integral ° fr by “anti-differentiating the integrand and plugging in a and b." We first need wo theorems that are of vast importance. 7.8C THEoREM. If /’(x)=0 fpr every x in the closed bounded interval (a,b), then f is constant on [a,b}—that is, S(x)=C (aexeb) for some CER, PRoor: For any x with a0), find °Q), 2. State all theorems on integration used in the following computation: i 3. If fis continuous on (— 00, 00) and if f[oe-sac= (5-9) (2-0). F(x) frou (-a0 (a&x0 (a<¢0 write F()-F(a)= 3B [FoF] for a subdivision o= (.Xj,.+.%,) a8 in 7.2G, Then apply 7.7A.) 7.9 IMPROPER INTEGRALS The definition ofthe integral fost does not follow from Section 7.2 since the interval [a,20) is not bounded. Such an integral is called an improper integral. The theory of this type of integral resembles to & reat extent the theory of infinite series, For this reason we do not give as many details as usual, We can define [sede 22 | carcuus as follows: If f €@[a,s] for every s>a, then [seas is defined to be the ordered pair ¢f,F> where F(s) ffeyar (adx=2(Vs -1 ) Ue (V5 = diverges since 1 fo and fs both converge, then s0 does “(Je e).and [usa [r= fs This may be proved by the method of 3.1C. Similarly, if fr converges and c€ R, then Jf converges and [ane fs ere ene If fe Gas] for every s>a and if f|f(x)}de converges, we say that f/(ayde converges absolutely. If F(s)= fs eoias and if F is bounded (above) on (a, 0), then lim,,..,F(s) exists by exercise 21 of Seetion 4.1 and hence f“f(x)dv converges absolutely. Wf “peep converges absolutely, if ¢€ 9[a,] fr every #> a, and if IsQ<[fO] (asx0 this means simply that [Ue )+iFEo)) de converges. But, since f "| f(x)|dx converges, it follows (by subtraction) that i * F(xpde ite converges, which is what we wished to show. 1 [fone converses but does aot converge absolutely, we say that [“f(ayde converges conditionally. A classical example of a conditionally convergent improper integral is To show that f“(sina)/-sde converges, we have for any +> (using integration by parts) 0} ay eam Now (700 all terms on the right of (1) approach limits. Thus tim f Zax exists, which proves that “(sinx)/.xdx converges. [This also shows that [Bae [aed even though th second integrals absolutely convergent while the ist integral (as we wilnow show) nt] Now we show that fcins)/xds does nt converge absoitely. For any W/€1 we nave We isinx| SY pert de fsinx! |S paste, [ee Oe cage (ta Te yo ALS AL isintes mi Now Sn( ach) = sin coner + costsin ne msinucosnn Since conne= =1 this shows that in(u+ nn) =fsine. Hence, if O< ux, then inn na)|=sinu. Thus r in.x| Q) tS nat By 3.2C the right side of (2) can be made as large as we please by taking N’ sufficiently large. This and (2) show that tim [Seal ax cannot exist, Hence, f~(sinx)/xdv does not converge absolutely. ‘The same method will prove the following important result. 7.9A. THEOREM. The improper integral fle diverges. peer pRoor: For any integer N we have v Poesy mI nt s [x87 f Dal Ted Dat ze w Again, since as N20 the right side of (1) diverges to infinity, we see that lig fhe does not exist. This proves the theorem. We have just used the divergence of an infinite series to establish the divergence of an improper integral. It is more usual to use an integral in the investigation of a series. This is known as the integral test for series 7.9B. THEOREM. Let f be a nonincreasing function on [1,00) such that f(x) #0 (IS(sx)> fol) (nexenth) since f is nonincreasing. Integrating from n to n+1 we then have Sf poyace fs oyax> "feat ds or Say> [Fader Sl). ‘Thus for N & J we have Not » wot * ES)? [Mare ZS sort = D $0). oO) It J * fla)de converges to A, then, by (I). PG) sf ‘Sxde 2 we have 1 lim, fig does nt exis Hence, Hin, f"1/(1~ sd doesnot exist wich proves hat f/ (1—x)dx does not converge. In this problem the divergence of 2 1 f T=” was deduced trom the divergence of f1/(-+und Integrals of the type discussed in this section are sometimes called improper integrals of the first kind, in contrast to a second kind of improper integral, which we discuss in the next section. Exercises 7.9 1. Which of the following integrals are convergent? @ fsa. ©) [ras 79° weRorer inreceas | 217 ee Bet 4 oS © Tatas (©) [xcosxax. © [°—sae f ro - ) f° — pe. © oe Show that x 1 1 Cape reel \. Show that st ds (+x) True or false? If fis continuous on {1,00) and it f"/(xpde converges, then lim,_..f()=0. (Answer: false) : 7 Sf feoax converges and if lim,_,,.f(x) = L, prove that L=0. Give an example of a continuous function f such that J(a)20 (le x<00), and such that Sse) cones but f “f(xjde diverges, . Give an example of a continuous function f such that f(2)>0 (I< x 0 (1 0. This suggests treating 1 1 1 ae asthe im f sex Uh ae J Ve (which turns out to be equa 102). In general, if f€ R[a+¢,b} for all « such that Ot nV does not fall into any one of the categories we have thus far described since it is an integral over (0,20) and 1/(x?-+ Vx) is not bounded for x near 0. However, we will agree to call f"1/(x2+ VE Jd a convergent 7.10C, The integral proper integral since we can break it up imo de and ym (haa P4VE Now J, is a convergent integral of the second kind [since 1/(x?+ VE )<1/V for Jy P+ VE 7.10 IMPROPER INTEGRALS, (CONTINUED) | 221 0<.x§ I] and J; is a convergent improper integral of the first kind [since 1/(x?+ Vi ) <1/x? for 1<.x< 00) In general if an integral J can be broken up in this way into two or more improper integrals Jj,...rJy of the first or second kinds, and if each J, (k= 1,....2) is convergent, ‘we will say that J is a convergent improper integral. However, if one or more of the J, is divergent, we will say that J is a divergent improper integral. Thus if (1/x2)dx is a divergent improper integral since 1 2 f ‘de and f Lae are improper integrals (of the second and first kinds, respectively) one of which is divergent. Similarly, f ° At ay |g let is a divergent improper integral since both f 14% ae and f Ite ay J, +a? ag lea? are divergent improper integrals of the second kind. (Note (I+.x)/(I+x4)>1/x for 1x60) TOD. As we have just observed, the integral a lex yy jag Lx? diverges since f 148 pp oy J, 14x does not approach a limit as s-+29. Similarly, ° Lex dX ax Q) Lass ® does not approach a limit as s-»20, However, the sum of (1) and (2) does approach a Timit as s-noe, For the sum of (1) and (2) + [esa few J et ace f ltx bee J, ex? and lim 2 zm | cavcuus does exist, The sum of (1) and (2) may also be written "(1+ x)/(1+.x?)dx. Hence, we have shown that exists, even though f" (1+.x)/(1-+.°)ds diverges. We call ne fee dx +x? the Cauchy principal value of f * Let xde. In general, the Cauchy principal value of f f(y i defined to be tin f pone i is imi exits. We denote Cavey principal value by CR. Thas fxd diverge but cy. fxd jim f’ sd jim 0=0, Iman exercise you will be asked to show that if f(x)de converges to 4, then CPV.” f(x)de=A also. As we have seen, however, the Cauchy principal value of J ses) sony xis even if the interval verges Exercises 7.10 1. Which of the following improper integrals are convergent? 1 © [ane © f——« b en ° pa 0 [Ba VE 3/2 ® [eta Tee 2. Prove that if s<1, then [ora de= Prove thet if 5> 1, then the integral diverges. 7.10 WMPROPER INTEGRALS, (CONTINUED) | 223 Prove that is convergent if and only if 00 if x>0, we have U(x)<0 if x<0 and, for example, U(— = Ud). Next, since Io Viet Vie Vie 24 ‘8.1 Hrpersouc Functions | 225 for 1<¢<2o, it follows from 79A that [71/VTEF di is divergent. Hence, f “1/Vi-+2 dt diverges, which shows that U is not bounded above. Equation (1) then Shows that U is not bounded below, either. By 7.8A, L U(x): mcxcen). Q) View ¢ ) Q) Hence, by 7.5B, U is continuous on (~ 20,00). Since U is bounded neither above nor below, theorem 6.2D implies that the range of U must be (20,00). Also, (2) and 7.7B show that U is I-1 on (~ 0,00). Thus U is a 1-1 continuous function from R' onto R'. If we denote the inverse function of U by S, then Sis a 1-1 function from R' onto R!. Moreover, since by 6.7 the restriction of U to any compact interval is a homeomorphism on that interval, it follows that $ is continuous at each point of R! (see exercise 4 of Section 6.7). If S(a)=6, then U(b)=a. By 7.5E and (2), S(a= View ae vO Since = 5(a) we thus have S(=Vit[S(P (-» 1 Since $= U~" and © vor f d h Viet ‘+ Remember y/ means nonnegative square 100, 226 | THE ELEMENTARY FUNCTIONS, TAYLOR SERIES we have $(0)=0. That is, sinkhO=0. From (6) we see then that cosh O= 1. Also, from (1) it follows that if U(x)=y, then U(~x)=—y. Hence, S(y)= =(-2)= = S(=y). That is, sinh(—y)=—sinhy — (~co0_ (-a 1 for all x we see from (11) that E(x)> $x) for all x. Thus since S is not bounded above, neither is E. That is, E takes on arbitrarily large positive values. By (13) E must then take on arbitrarily small postive values. (For, by (13), if E(x)= M (large), then E(—x)=1/M (small). Theorem 62D and (14) then show that the range of E is precisely (0,0) From (11) we have E"(x)= C'(x) + S'(x)= S(x)+ C(x). Thus. EQ)=EQ) (-20 for all x and hence, by 7.7B, that E is 1-1 on * In these sections only we continue the numbering of equations from one Setion into the net 82 THEEXPONENTIAL FUNCTION | 227 (~ 20, 00). It follows (exercise 4 of Section 6.7) that E is a homeomorphism of (20,0) conto (0.0). Finally, let us prove that E(x+a)=E(x)E(a) — [a,x€(~20,0)]- (16) Fix a and let F(x)= E(x +4)E(~x). Using 7.5C and 7.5D we have, on differentiation with respect to x, F(x) =E(x+a)[-E(-s)]FE(xt aE (-x) (